]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/ofproto.c
Remove unused variables and functions.
[mirror_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_flush);
61 COVERAGE_DEFINE(ofproto_packet_out);
62 COVERAGE_DEFINE(ofproto_queue_req);
63 COVERAGE_DEFINE(ofproto_recv_openflow);
64 COVERAGE_DEFINE(ofproto_reinit_ports);
65 COVERAGE_DEFINE(ofproto_update_port);
66
67 enum ofproto_state {
68 S_OPENFLOW, /* Processing OpenFlow commands. */
69 S_EVICT, /* Evicting flows from over-limit tables. */
70 S_FLUSH, /* Deleting all flow table rules. */
71 };
72
73 enum ofoperation_type {
74 OFOPERATION_ADD,
75 OFOPERATION_DELETE,
76 OFOPERATION_MODIFY,
77 OFOPERATION_REPLACE
78 };
79
80 /* A single OpenFlow request can execute any number of operations. The
81 * ofopgroup maintain OpenFlow state common to all of the operations, e.g. the
82 * ofconn to which an error reply should be sent if necessary.
83 *
84 * ofproto initiates some operations internally. These operations are still
85 * assigned to groups but will not have an associated ofconn. */
86 struct ofopgroup {
87 struct ofproto *ofproto; /* Owning ofproto. */
88 struct list ofproto_node; /* In ofproto's "pending" list. */
89 struct list ops; /* List of "struct ofoperation"s. */
90 int n_running; /* Number of ops still pending. */
91
92 /* Data needed to send OpenFlow reply on failure or to send a buffered
93 * packet on success.
94 *
95 * If list_is_empty(ofconn_node) then this ofopgroup never had an
96 * associated ofconn or its ofconn's connection dropped after it initiated
97 * the operation. In the latter case 'ofconn' is a wild pointer that
98 * refers to freed memory, so the 'ofconn' member must be used only if
99 * !list_is_empty(ofconn_node).
100 */
101 struct list ofconn_node; /* In ofconn's list of pending opgroups. */
102 struct ofconn *ofconn; /* ofconn for reply (but see note above). */
103 struct ofp_header *request; /* Original request (truncated at 64 bytes). */
104 uint32_t buffer_id; /* Buffer id from original request. */
105 };
106
107 static struct ofopgroup *ofopgroup_create_unattached(struct ofproto *);
108 static struct ofopgroup *ofopgroup_create(struct ofproto *, struct ofconn *,
109 const struct ofp_header *,
110 uint32_t buffer_id);
111 static void ofopgroup_submit(struct ofopgroup *);
112 static void ofopgroup_complete(struct ofopgroup *);
113
114 /* A single flow table operation. */
115 struct ofoperation {
116 struct ofopgroup *group; /* Owning group. */
117 struct list group_node; /* In ofopgroup's "ops" list. */
118 struct hmap_node hmap_node; /* In ofproto's "deletions" hmap. */
119 struct rule *rule; /* Rule being operated upon. */
120 enum ofoperation_type type; /* Type of operation. */
121
122 /* OFOPERATION_MODIFY, OFOPERATION_REPLACE: The old actions, if the actions
123 * are changing. */
124 struct rule_actions *actions;
125
126 /* OFOPERATION_DELETE. */
127 enum ofp_flow_removed_reason reason; /* Reason flow was removed. */
128
129 ovs_be64 flow_cookie; /* Rule's old flow cookie. */
130 uint16_t idle_timeout; /* Rule's old idle timeout. */
131 uint16_t hard_timeout; /* Rule's old hard timeout. */
132 enum ofputil_flow_mod_flags flags; /* Rule's old flags. */
133 enum ofperr error; /* 0 if no error. */
134 };
135
136 static struct ofoperation *ofoperation_create(struct ofopgroup *,
137 struct rule *,
138 enum ofoperation_type,
139 enum ofp_flow_removed_reason);
140 static void ofoperation_destroy(struct ofoperation *);
141
142 /* oftable. */
143 static void oftable_init(struct oftable *);
144 static void oftable_destroy(struct oftable *);
145
146 static void oftable_set_name(struct oftable *, const char *name);
147
148 static void oftable_disable_eviction(struct oftable *);
149 static void oftable_enable_eviction(struct oftable *,
150 const struct mf_subfield *fields,
151 size_t n_fields);
152
153 static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex);
154 static void oftable_remove_rule__(struct ofproto *, struct rule *)
155 OVS_REQUIRES(ofproto_mutex);
156 static void oftable_insert_rule(struct rule *);
157
158 /* A set of rules within a single OpenFlow table (oftable) that have the same
159 * values for the oftable's eviction_fields. A rule to be evicted, when one is
160 * needed, is taken from the eviction group that contains the greatest number
161 * of rules.
162 *
163 * An oftable owns any number of eviction groups, each of which contains any
164 * number of rules.
165 *
166 * Membership in an eviction group is imprecise, based on the hash of the
167 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
168 * That is, if two rules have different eviction_fields, but those
169 * eviction_fields hash to the same value, then they will belong to the same
170 * eviction_group anyway.
171 *
172 * (When eviction is not enabled on an oftable, we don't track any eviction
173 * groups, to save time and space.) */
174 struct eviction_group {
175 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
176 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
177 struct heap rules; /* Contains "struct rule"s. */
178 };
179
180 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep);
181 static void ofproto_evict(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
182 static uint32_t rule_eviction_priority(struct rule *);
183 static void eviction_group_add_rule(struct rule *);
184 static void eviction_group_remove_rule(struct rule *);
185
186 /* Criteria that flow_mod and other operations use for selecting rules on
187 * which to operate. */
188 struct rule_criteria {
189 /* An OpenFlow table or 255 for all tables. */
190 uint8_t table_id;
191
192 /* OpenFlow matching criteria. Interpreted different in "loose" way by
193 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
194 * defined in the OpenFlow spec. */
195 struct cls_rule cr;
196
197 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
198 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
199 * The rule will not be selected if M is 1 and B != C. */
200 ovs_be64 cookie;
201 ovs_be64 cookie_mask;
202
203 /* Selection based on actions within a rule:
204 *
205 * If out_port != OFPP_ANY, selects only rules that output to out_port.
206 * If out_group != OFPG_ALL, select only rules that output to out_group. */
207 ofp_port_t out_port;
208 uint32_t out_group;
209 };
210
211 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
212 const struct match *match,
213 unsigned int priority,
214 ovs_be64 cookie, ovs_be64 cookie_mask,
215 ofp_port_t out_port, uint32_t out_group);
216 static void rule_criteria_destroy(struct rule_criteria *);
217
218 /* A packet that needs to be passed to rule_execute().
219 *
220 * (We can't do this immediately from ofopgroup_complete() because that holds
221 * ofproto_mutex, which rule_execute() needs released.) */
222 struct rule_execute {
223 struct list list_node; /* In struct ofproto's "rule_executes" list. */
224 struct rule *rule; /* Owns a reference to the rule. */
225 ofp_port_t in_port;
226 struct ofpbuf *packet; /* Owns the packet. */
227 };
228
229 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
230 static void destroy_rule_executes(struct ofproto *);
231
232 /* ofport. */
233 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
234 static void ofport_destroy(struct ofport *);
235
236 static void update_port(struct ofproto *, const char *devname);
237 static int init_ports(struct ofproto *);
238 static void reinit_ports(struct ofproto *);
239
240 static long long int ofport_get_usage(const struct ofproto *,
241 ofp_port_t ofp_port);
242 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
243 long long int last_used);
244
245 /* Ofport usage.
246 *
247 * Keeps track of the currently used and recently used ofport values and is
248 * used to prevent immediate recycling of ofport values. */
249 struct ofport_usage {
250 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
251 ofp_port_t ofp_port; /* OpenFlow port number. */
252 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
253 represents in-use ofports. */
254 };
255
256 /* rule. */
257 static void ofproto_rule_destroy__(struct rule *);
258 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
259 static bool rule_is_modifiable(const struct rule *);
260
261 /* OpenFlow. */
262 static enum ofperr add_flow(struct ofproto *, struct ofconn *,
263 struct ofputil_flow_mod *,
264 const struct ofp_header *);
265 static enum ofperr modify_flows__(struct ofproto *, struct ofconn *,
266 struct ofputil_flow_mod *,
267 const struct ofp_header *,
268 const struct rule_collection *);
269 static void delete_flow__(struct rule *rule, struct ofopgroup *,
270 enum ofp_flow_removed_reason)
271 OVS_REQUIRES(ofproto_mutex);
272 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
273 static bool handle_openflow(struct ofconn *, const struct ofpbuf *);
274 static enum ofperr handle_flow_mod__(struct ofproto *, struct ofconn *,
275 struct ofputil_flow_mod *,
276 const struct ofp_header *)
277 OVS_EXCLUDED(ofproto_mutex);
278 static void calc_duration(long long int start, long long int now,
279 uint32_t *sec, uint32_t *nsec);
280
281 /* ofproto. */
282 static uint64_t pick_datapath_id(const struct ofproto *);
283 static uint64_t pick_fallback_dpid(void);
284 static void ofproto_destroy__(struct ofproto *);
285 static void update_mtu(struct ofproto *, struct ofport *);
286 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
287
288 /* unixctl. */
289 static void ofproto_unixctl_init(void);
290
291 /* All registered ofproto classes, in probe order. */
292 static const struct ofproto_class **ofproto_classes;
293 static size_t n_ofproto_classes;
294 static size_t allocated_ofproto_classes;
295
296 /* Global lock that protects all flow table operations. */
297 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
298
299 unsigned flow_eviction_threshold = OFPROTO_FLOW_EVICTION_THRESHOLD_DEFAULT;
300 unsigned n_handler_threads;
301 enum ofproto_flow_miss_model flow_miss_model = OFPROTO_HANDLE_MISS_AUTO;
302
303 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
304 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
305
306 /* Initial mappings of port to OpenFlow number mappings. */
307 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
308
309 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
310
311 /* The default value of true waits for flow restore. */
312 static bool flow_restore_wait = true;
313
314 /* Must be called to initialize the ofproto library.
315 *
316 * The caller may pass in 'iface_hints', which contains an shash of
317 * "iface_hint" elements indexed by the interface's name. The provider
318 * may use these hints to describe the startup configuration in order to
319 * reinitialize its state. The caller owns the provided data, so a
320 * provider will make copies of anything required. An ofproto provider
321 * will remove any existing state that is not described by the hint, and
322 * may choose to remove it all. */
323 void
324 ofproto_init(const struct shash *iface_hints)
325 {
326 struct shash_node *node;
327 size_t i;
328
329 ofproto_class_register(&ofproto_dpif_class);
330
331 /* Make a local copy, since we don't own 'iface_hints' elements. */
332 SHASH_FOR_EACH(node, iface_hints) {
333 const struct iface_hint *orig_hint = node->data;
334 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
335 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
336
337 new_hint->br_name = xstrdup(orig_hint->br_name);
338 new_hint->br_type = xstrdup(br_type);
339 new_hint->ofp_port = orig_hint->ofp_port;
340
341 shash_add(&init_ofp_ports, node->name, new_hint);
342 }
343
344 for (i = 0; i < n_ofproto_classes; i++) {
345 ofproto_classes[i]->init(&init_ofp_ports);
346 }
347 }
348
349 /* 'type' should be a normalized datapath type, as returned by
350 * ofproto_normalize_type(). Returns the corresponding ofproto_class
351 * structure, or a null pointer if there is none registered for 'type'. */
352 static const struct ofproto_class *
353 ofproto_class_find__(const char *type)
354 {
355 size_t i;
356
357 for (i = 0; i < n_ofproto_classes; i++) {
358 const struct ofproto_class *class = ofproto_classes[i];
359 struct sset types;
360 bool found;
361
362 sset_init(&types);
363 class->enumerate_types(&types);
364 found = sset_contains(&types, type);
365 sset_destroy(&types);
366
367 if (found) {
368 return class;
369 }
370 }
371 VLOG_WARN("unknown datapath type %s", type);
372 return NULL;
373 }
374
375 /* Registers a new ofproto class. After successful registration, new ofprotos
376 * of that type can be created using ofproto_create(). */
377 int
378 ofproto_class_register(const struct ofproto_class *new_class)
379 {
380 size_t i;
381
382 for (i = 0; i < n_ofproto_classes; i++) {
383 if (ofproto_classes[i] == new_class) {
384 return EEXIST;
385 }
386 }
387
388 if (n_ofproto_classes >= allocated_ofproto_classes) {
389 ofproto_classes = x2nrealloc(ofproto_classes,
390 &allocated_ofproto_classes,
391 sizeof *ofproto_classes);
392 }
393 ofproto_classes[n_ofproto_classes++] = new_class;
394 return 0;
395 }
396
397 /* Unregisters a datapath provider. 'type' must have been previously
398 * registered and not currently be in use by any ofprotos. After
399 * unregistration new datapaths of that type cannot be opened using
400 * ofproto_create(). */
401 int
402 ofproto_class_unregister(const struct ofproto_class *class)
403 {
404 size_t i;
405
406 for (i = 0; i < n_ofproto_classes; i++) {
407 if (ofproto_classes[i] == class) {
408 for (i++; i < n_ofproto_classes; i++) {
409 ofproto_classes[i - 1] = ofproto_classes[i];
410 }
411 n_ofproto_classes--;
412 return 0;
413 }
414 }
415 VLOG_WARN("attempted to unregister an ofproto class that is not "
416 "registered");
417 return EAFNOSUPPORT;
418 }
419
420 /* Clears 'types' and enumerates all registered ofproto types into it. The
421 * caller must first initialize the sset. */
422 void
423 ofproto_enumerate_types(struct sset *types)
424 {
425 size_t i;
426
427 for (i = 0; i < n_ofproto_classes; i++) {
428 ofproto_classes[i]->enumerate_types(types);
429 }
430 }
431
432 /* Returns the fully spelled out name for the given ofproto 'type'.
433 *
434 * Normalized type string can be compared with strcmp(). Unnormalized type
435 * string might be the same even if they have different spellings. */
436 const char *
437 ofproto_normalize_type(const char *type)
438 {
439 return type && type[0] ? type : "system";
440 }
441
442 /* Clears 'names' and enumerates the names of all known created ofprotos with
443 * the given 'type'. The caller must first initialize the sset. Returns 0 if
444 * successful, otherwise a positive errno value.
445 *
446 * Some kinds of datapaths might not be practically enumerable. This is not
447 * considered an error. */
448 int
449 ofproto_enumerate_names(const char *type, struct sset *names)
450 {
451 const struct ofproto_class *class = ofproto_class_find__(type);
452 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
453 }
454
455 int
456 ofproto_create(const char *datapath_name, const char *datapath_type,
457 struct ofproto **ofprotop)
458 {
459 const struct ofproto_class *class;
460 struct ofproto *ofproto;
461 int error;
462 int i;
463
464 *ofprotop = NULL;
465
466 ofproto_unixctl_init();
467
468 datapath_type = ofproto_normalize_type(datapath_type);
469 class = ofproto_class_find__(datapath_type);
470 if (!class) {
471 VLOG_WARN("could not create datapath %s of unknown type %s",
472 datapath_name, datapath_type);
473 return EAFNOSUPPORT;
474 }
475
476 ofproto = class->alloc();
477 if (!ofproto) {
478 VLOG_ERR("failed to allocate datapath %s of type %s",
479 datapath_name, datapath_type);
480 return ENOMEM;
481 }
482
483 /* Initialize. */
484 ovs_mutex_lock(&ofproto_mutex);
485 memset(ofproto, 0, sizeof *ofproto);
486 ofproto->ofproto_class = class;
487 ofproto->name = xstrdup(datapath_name);
488 ofproto->type = xstrdup(datapath_type);
489 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
490 hash_string(ofproto->name, 0));
491 ofproto->datapath_id = 0;
492 ofproto->forward_bpdu = false;
493 ofproto->fallback_dpid = pick_fallback_dpid();
494 ofproto->mfr_desc = NULL;
495 ofproto->hw_desc = NULL;
496 ofproto->sw_desc = NULL;
497 ofproto->serial_desc = NULL;
498 ofproto->dp_desc = NULL;
499 ofproto->frag_handling = OFPC_FRAG_NORMAL;
500 hmap_init(&ofproto->ports);
501 hmap_init(&ofproto->ofport_usage);
502 shash_init(&ofproto->port_by_name);
503 simap_init(&ofproto->ofp_requests);
504 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
505 ofproto->eviction_group_timer = LLONG_MIN;
506 ofproto->tables = NULL;
507 ofproto->n_tables = 0;
508 hindex_init(&ofproto->cookies);
509 list_init(&ofproto->expirable);
510 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
511 ofproto->state = S_OPENFLOW;
512 list_init(&ofproto->pending);
513 ofproto->n_pending = 0;
514 hmap_init(&ofproto->deletions);
515 guarded_list_init(&ofproto->rule_executes);
516 ofproto->n_add = ofproto->n_delete = ofproto->n_modify = 0;
517 ofproto->first_op = ofproto->last_op = LLONG_MIN;
518 ofproto->next_op_report = LLONG_MAX;
519 ofproto->op_backoff = LLONG_MIN;
520 ofproto->vlan_bitmap = NULL;
521 ofproto->vlans_changed = false;
522 ofproto->min_mtu = INT_MAX;
523 ovs_rwlock_init(&ofproto->groups_rwlock);
524 hmap_init(&ofproto->groups);
525 ovs_mutex_unlock(&ofproto_mutex);
526
527 error = ofproto->ofproto_class->construct(ofproto);
528 if (error) {
529 VLOG_ERR("failed to open datapath %s: %s",
530 datapath_name, ovs_strerror(error));
531 ofproto_destroy__(ofproto);
532 return error;
533 }
534
535 /* Check that hidden tables, if any, are at the end. */
536 ovs_assert(ofproto->n_tables);
537 for (i = 0; i + 1 < ofproto->n_tables; i++) {
538 enum oftable_flags flags = ofproto->tables[i].flags;
539 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
540
541 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
542 }
543
544 ofproto->datapath_id = pick_datapath_id(ofproto);
545 init_ports(ofproto);
546
547 /* Initialize meters table. */
548 if (ofproto->ofproto_class->meter_get_features) {
549 ofproto->ofproto_class->meter_get_features(ofproto,
550 &ofproto->meter_features);
551 } else {
552 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
553 }
554 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
555 * sizeof(struct meter *));
556
557 *ofprotop = ofproto;
558 return 0;
559 }
560
561 /* Must be called (only) by an ofproto implementation in its constructor
562 * function. See the large comment on 'construct' in struct ofproto_class for
563 * details. */
564 void
565 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
566 {
567 struct oftable *table;
568
569 ovs_assert(!ofproto->n_tables);
570 ovs_assert(n_tables >= 1 && n_tables <= 255);
571
572 ofproto->n_tables = n_tables;
573 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
574 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
575 oftable_init(table);
576 }
577 }
578
579 /* To be optionally called (only) by an ofproto implementation in its
580 * constructor function. See the large comment on 'construct' in struct
581 * ofproto_class for details.
582 *
583 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
584 * will then ensure that actions passed into the ofproto implementation will
585 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
586 * function is not called, there will be no such restriction.
587 *
588 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
589 * the 'max_ports' restriction. */
590 void
591 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
592 {
593 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
594 ofproto->max_ports = max_ports;
595 }
596
597 uint64_t
598 ofproto_get_datapath_id(const struct ofproto *ofproto)
599 {
600 return ofproto->datapath_id;
601 }
602
603 void
604 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
605 {
606 uint64_t old_dpid = p->datapath_id;
607 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
608 if (p->datapath_id != old_dpid) {
609 /* Force all active connections to reconnect, since there is no way to
610 * notify a controller that the datapath ID has changed. */
611 ofproto_reconnect_controllers(p);
612 }
613 }
614
615 void
616 ofproto_set_controllers(struct ofproto *p,
617 const struct ofproto_controller *controllers,
618 size_t n_controllers, uint32_t allowed_versions)
619 {
620 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
621 allowed_versions);
622 }
623
624 void
625 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
626 {
627 connmgr_set_fail_mode(p->connmgr, fail_mode);
628 }
629
630 /* Drops the connections between 'ofproto' and all of its controllers, forcing
631 * them to reconnect. */
632 void
633 ofproto_reconnect_controllers(struct ofproto *ofproto)
634 {
635 connmgr_reconnect(ofproto->connmgr);
636 }
637
638 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
639 * in-band control should guarantee access, in the same way that in-band
640 * control guarantees access to OpenFlow controllers. */
641 void
642 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
643 const struct sockaddr_in *extras, size_t n)
644 {
645 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
646 }
647
648 /* Sets the OpenFlow queue used by flows set up by in-band control on
649 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
650 * flows will use the default queue. */
651 void
652 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
653 {
654 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
655 }
656
657 /* Sets the number of flows at which eviction from the kernel flow table
658 * will occur. */
659 void
660 ofproto_set_flow_eviction_threshold(unsigned threshold)
661 {
662 flow_eviction_threshold = MAX(OFPROTO_FLOW_EVICTION_THRESHOLD_MIN,
663 threshold);
664 }
665
666 /* Sets the path for handling flow misses. */
667 void
668 ofproto_set_flow_miss_model(unsigned model)
669 {
670 flow_miss_model = model;
671 }
672
673 /* If forward_bpdu is true, the NORMAL action will forward frames with
674 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
675 * the NORMAL action will drop these frames. */
676 void
677 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
678 {
679 bool old_val = ofproto->forward_bpdu;
680 ofproto->forward_bpdu = forward_bpdu;
681 if (old_val != ofproto->forward_bpdu) {
682 if (ofproto->ofproto_class->forward_bpdu_changed) {
683 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
684 }
685 }
686 }
687
688 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
689 * 'idle_time', in seconds, and the maximum number of MAC table entries to
690 * 'max_entries'. */
691 void
692 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
693 size_t max_entries)
694 {
695 if (ofproto->ofproto_class->set_mac_table_config) {
696 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
697 max_entries);
698 }
699 }
700
701 /* Sets number of upcall handler threads. The default is
702 * (number of online cores - 2). */
703 void
704 ofproto_set_n_handler_threads(unsigned limit)
705 {
706 if (limit) {
707 n_handler_threads = limit;
708 } else {
709 int n_proc = sysconf(_SC_NPROCESSORS_ONLN);
710 n_handler_threads = n_proc > 2 ? n_proc - 2 : 1;
711 }
712 }
713
714 void
715 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
716 {
717 free(p->dp_desc);
718 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
719 }
720
721 int
722 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
723 {
724 return connmgr_set_snoops(ofproto->connmgr, snoops);
725 }
726
727 int
728 ofproto_set_netflow(struct ofproto *ofproto,
729 const struct netflow_options *nf_options)
730 {
731 if (nf_options && sset_is_empty(&nf_options->collectors)) {
732 nf_options = NULL;
733 }
734
735 if (ofproto->ofproto_class->set_netflow) {
736 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
737 } else {
738 return nf_options ? EOPNOTSUPP : 0;
739 }
740 }
741
742 int
743 ofproto_set_sflow(struct ofproto *ofproto,
744 const struct ofproto_sflow_options *oso)
745 {
746 if (oso && sset_is_empty(&oso->targets)) {
747 oso = NULL;
748 }
749
750 if (ofproto->ofproto_class->set_sflow) {
751 return ofproto->ofproto_class->set_sflow(ofproto, oso);
752 } else {
753 return oso ? EOPNOTSUPP : 0;
754 }
755 }
756
757 int
758 ofproto_set_ipfix(struct ofproto *ofproto,
759 const struct ofproto_ipfix_bridge_exporter_options *bo,
760 const struct ofproto_ipfix_flow_exporter_options *fo,
761 size_t n_fo)
762 {
763 if (ofproto->ofproto_class->set_ipfix) {
764 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
765 } else {
766 return (bo || fo) ? EOPNOTSUPP : 0;
767 }
768 }
769
770 void
771 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
772 {
773 flow_restore_wait = flow_restore_wait_db;
774 }
775
776 bool
777 ofproto_get_flow_restore_wait(void)
778 {
779 return flow_restore_wait;
780 }
781
782 \f
783 /* Spanning Tree Protocol (STP) configuration. */
784
785 /* Configures STP on 'ofproto' using the settings defined in 's'. If
786 * 's' is NULL, disables STP.
787 *
788 * Returns 0 if successful, otherwise a positive errno value. */
789 int
790 ofproto_set_stp(struct ofproto *ofproto,
791 const struct ofproto_stp_settings *s)
792 {
793 return (ofproto->ofproto_class->set_stp
794 ? ofproto->ofproto_class->set_stp(ofproto, s)
795 : EOPNOTSUPP);
796 }
797
798 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
799 * 'enabled' member of 's' is false, then the other members are not
800 * meaningful.
801 *
802 * Returns 0 if successful, otherwise a positive errno value. */
803 int
804 ofproto_get_stp_status(struct ofproto *ofproto,
805 struct ofproto_stp_status *s)
806 {
807 return (ofproto->ofproto_class->get_stp_status
808 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
809 : EOPNOTSUPP);
810 }
811
812 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
813 * in 's'. The caller is responsible for assigning STP port numbers
814 * (using the 'port_num' member in the range of 1 through 255, inclusive)
815 * and ensuring there are no duplicates. If the 's' is NULL, then STP
816 * is disabled on the port.
817 *
818 * Returns 0 if successful, otherwise a positive errno value.*/
819 int
820 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
821 const struct ofproto_port_stp_settings *s)
822 {
823 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
824 if (!ofport) {
825 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
826 ofproto->name, ofp_port);
827 return ENODEV;
828 }
829
830 return (ofproto->ofproto_class->set_stp_port
831 ? ofproto->ofproto_class->set_stp_port(ofport, s)
832 : EOPNOTSUPP);
833 }
834
835 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
836 * 's'. If the 'enabled' member in 's' is false, then the other members
837 * are not meaningful.
838 *
839 * Returns 0 if successful, otherwise a positive errno value.*/
840 int
841 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
842 struct ofproto_port_stp_status *s)
843 {
844 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
845 if (!ofport) {
846 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
847 "port %"PRIu16, ofproto->name, ofp_port);
848 return ENODEV;
849 }
850
851 return (ofproto->ofproto_class->get_stp_port_status
852 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
853 : EOPNOTSUPP);
854 }
855 \f
856 /* Queue DSCP configuration. */
857
858 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
859 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
860 * to implement QoS. Instead, it is used to implement features which require
861 * knowledge of what queues exist on a port, and some basic information about
862 * them.
863 *
864 * Returns 0 if successful, otherwise a positive errno value. */
865 int
866 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
867 const struct ofproto_port_queue *queues,
868 size_t n_queues)
869 {
870 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
871
872 if (!ofport) {
873 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
874 ofproto->name, ofp_port);
875 return ENODEV;
876 }
877
878 return (ofproto->ofproto_class->set_queues
879 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
880 : EOPNOTSUPP);
881 }
882 \f
883 /* Connectivity Fault Management configuration. */
884
885 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
886 void
887 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
888 {
889 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
890 if (ofport && ofproto->ofproto_class->set_cfm) {
891 ofproto->ofproto_class->set_cfm(ofport, NULL);
892 }
893 }
894
895 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
896 * basic configuration from the configuration members in 'cfm', and the remote
897 * maintenance point ID from remote_mpid. Ignores the statistics members of
898 * 'cfm'.
899 *
900 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
901 void
902 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
903 const struct cfm_settings *s)
904 {
905 struct ofport *ofport;
906 int error;
907
908 ofport = ofproto_get_port(ofproto, ofp_port);
909 if (!ofport) {
910 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
911 ofproto->name, ofp_port);
912 return;
913 }
914
915 /* XXX: For configuration simplicity, we only support one remote_mpid
916 * outside of the CFM module. It's not clear if this is the correct long
917 * term solution or not. */
918 error = (ofproto->ofproto_class->set_cfm
919 ? ofproto->ofproto_class->set_cfm(ofport, s)
920 : EOPNOTSUPP);
921 if (error) {
922 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
923 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
924 ovs_strerror(error));
925 }
926 }
927
928 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
929 * 'ofproto' does not have a port 'ofp_port'. */
930 void
931 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
932 const struct smap *cfg)
933 {
934 struct ofport *ofport;
935 int error;
936
937 ofport = ofproto_get_port(ofproto, ofp_port);
938 if (!ofport) {
939 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
940 ofproto->name, ofp_port);
941 return;
942 }
943
944 error = (ofproto->ofproto_class->set_bfd
945 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
946 : EOPNOTSUPP);
947 if (error) {
948 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
949 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
950 ovs_strerror(error));
951 }
952 }
953
954 /* Populates 'status' with key value pairs indicating the status of the BFD
955 * session on 'ofp_port'. This information is intended to be populated in the
956 * OVS database. Has no effect if 'ofp_port' is not na OpenFlow port in
957 * 'ofproto'. */
958 int
959 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
960 struct smap *status)
961 {
962 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
963 return (ofport && ofproto->ofproto_class->get_bfd_status
964 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
965 : EOPNOTSUPP);
966 }
967
968 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
969 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
970 * 0 if LACP partner information is not current (generally indicating a
971 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
972 int
973 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
974 {
975 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
976 return (ofport && ofproto->ofproto_class->port_is_lacp_current
977 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
978 : -1);
979 }
980 \f
981 /* Bundles. */
982
983 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
984 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
985 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
986 * configuration plus, if there is more than one slave, a bonding
987 * configuration.
988 *
989 * If 'aux' is already registered then this function updates its configuration
990 * to 's'. Otherwise, this function registers a new bundle.
991 *
992 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
993 * port. */
994 int
995 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
996 const struct ofproto_bundle_settings *s)
997 {
998 return (ofproto->ofproto_class->bundle_set
999 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1000 : EOPNOTSUPP);
1001 }
1002
1003 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1004 * If no such bundle has been registered, this has no effect. */
1005 int
1006 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1007 {
1008 return ofproto_bundle_register(ofproto, aux, NULL);
1009 }
1010
1011 \f
1012 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1013 * If 'aux' is already registered then this function updates its configuration
1014 * to 's'. Otherwise, this function registers a new mirror. */
1015 int
1016 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1017 const struct ofproto_mirror_settings *s)
1018 {
1019 return (ofproto->ofproto_class->mirror_set
1020 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1021 : EOPNOTSUPP);
1022 }
1023
1024 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1025 * If no mirror has been registered, this has no effect. */
1026 int
1027 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1028 {
1029 return ofproto_mirror_register(ofproto, aux, NULL);
1030 }
1031
1032 /* Retrieves statistics from mirror associated with client data pointer
1033 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1034 * 'bytes', respectively. If a particular counters is not supported,
1035 * the appropriate argument is set to UINT64_MAX. */
1036 int
1037 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1038 uint64_t *packets, uint64_t *bytes)
1039 {
1040 if (!ofproto->ofproto_class->mirror_get_stats) {
1041 *packets = *bytes = UINT64_MAX;
1042 return EOPNOTSUPP;
1043 }
1044
1045 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1046 packets, bytes);
1047 }
1048
1049 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1050 * which all packets are flooded, instead of using MAC learning. If
1051 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1052 *
1053 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1054 * port. */
1055 int
1056 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1057 {
1058 return (ofproto->ofproto_class->set_flood_vlans
1059 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1060 : EOPNOTSUPP);
1061 }
1062
1063 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1064 * output for a mirror. */
1065 bool
1066 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1067 {
1068 return (ofproto->ofproto_class->is_mirror_output_bundle
1069 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1070 : false);
1071 }
1072 \f
1073 /* Configuration of OpenFlow tables. */
1074
1075 /* Returns the number of OpenFlow tables in 'ofproto'. */
1076 int
1077 ofproto_get_n_tables(const struct ofproto *ofproto)
1078 {
1079 return ofproto->n_tables;
1080 }
1081
1082 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1083 * settings from 's'. 'table_id' must be in the range 0 through the number of
1084 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1085 *
1086 * For read-only tables, only the name may be configured. */
1087 void
1088 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1089 const struct ofproto_table_settings *s)
1090 {
1091 struct oftable *table;
1092
1093 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1094 table = &ofproto->tables[table_id];
1095
1096 oftable_set_name(table, s->name);
1097
1098 if (table->flags & OFTABLE_READONLY) {
1099 return;
1100 }
1101
1102 if (s->groups) {
1103 oftable_enable_eviction(table, s->groups, s->n_groups);
1104 } else {
1105 oftable_disable_eviction(table);
1106 }
1107
1108 table->max_flows = s->max_flows;
1109 ovs_rwlock_rdlock(&table->cls.rwlock);
1110 if (classifier_count(&table->cls) > table->max_flows
1111 && table->eviction_fields) {
1112 /* 'table' contains more flows than allowed. We might not be able to
1113 * evict them right away because of the asynchronous nature of flow
1114 * table changes. Schedule eviction for later. */
1115 switch (ofproto->state) {
1116 case S_OPENFLOW:
1117 ofproto->state = S_EVICT;
1118 break;
1119 case S_EVICT:
1120 case S_FLUSH:
1121 /* We're already deleting flows, nothing more to do. */
1122 break;
1123 }
1124 }
1125 ovs_rwlock_unlock(&table->cls.rwlock);
1126 }
1127 \f
1128 bool
1129 ofproto_has_snoops(const struct ofproto *ofproto)
1130 {
1131 return connmgr_has_snoops(ofproto->connmgr);
1132 }
1133
1134 void
1135 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1136 {
1137 connmgr_get_snoops(ofproto->connmgr, snoops);
1138 }
1139
1140 static void
1141 ofproto_rule_delete__(struct ofproto *ofproto, struct rule *rule)
1142 OVS_REQUIRES(ofproto_mutex)
1143 {
1144 struct ofopgroup *group;
1145
1146 ovs_assert(!rule->pending);
1147
1148 group = ofopgroup_create_unattached(ofproto);
1149 delete_flow__(rule, group, OFPRR_DELETE);
1150 ofopgroup_submit(group);
1151 }
1152
1153 /* Deletes 'rule' from 'cls' within 'ofproto'.
1154 *
1155 * Within an ofproto implementation, this function allows an ofproto
1156 * implementation to destroy any rules that remain when its ->destruct()
1157 * function is called. This function is not suitable for use elsewhere in an
1158 * ofproto implementation.
1159 *
1160 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1161 * Cycle" in ofproto-provider.h. */
1162 void
1163 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1164 OVS_EXCLUDED(ofproto_mutex)
1165 {
1166 struct ofopgroup *group;
1167
1168 ovs_mutex_lock(&ofproto_mutex);
1169 ovs_assert(!rule->pending);
1170
1171 group = ofopgroup_create_unattached(ofproto);
1172 ofoperation_create(group, rule, OFOPERATION_DELETE, OFPRR_DELETE);
1173 oftable_remove_rule__(ofproto, rule);
1174 ofproto->ofproto_class->rule_delete(rule);
1175 ofopgroup_submit(group);
1176
1177 ovs_mutex_unlock(&ofproto_mutex);
1178 }
1179
1180 static void
1181 ofproto_flush__(struct ofproto *ofproto)
1182 OVS_EXCLUDED(ofproto_mutex)
1183 {
1184 struct oftable *table;
1185
1186 if (ofproto->ofproto_class->flush) {
1187 ofproto->ofproto_class->flush(ofproto);
1188 }
1189
1190 ovs_mutex_lock(&ofproto_mutex);
1191 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1192 struct rule *rule, *next_rule;
1193 struct cls_cursor cursor;
1194
1195 if (table->flags & OFTABLE_HIDDEN) {
1196 continue;
1197 }
1198
1199 ovs_rwlock_rdlock(&table->cls.rwlock);
1200 cls_cursor_init(&cursor, &table->cls, NULL);
1201 ovs_rwlock_unlock(&table->cls.rwlock);
1202 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1203 if (!rule->pending) {
1204 ofproto_rule_delete__(ofproto, rule);
1205 }
1206 }
1207 }
1208 ovs_mutex_unlock(&ofproto_mutex);
1209 }
1210
1211 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1212
1213 static void
1214 ofproto_destroy__(struct ofproto *ofproto)
1215 OVS_EXCLUDED(ofproto_mutex)
1216 {
1217 struct oftable *table;
1218
1219 ovs_assert(list_is_empty(&ofproto->pending));
1220
1221 destroy_rule_executes(ofproto);
1222 guarded_list_destroy(&ofproto->rule_executes);
1223
1224 delete_group(ofproto, OFPG_ALL);
1225 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1226 hmap_destroy(&ofproto->groups);
1227
1228 connmgr_destroy(ofproto->connmgr);
1229
1230 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1231 free(ofproto->name);
1232 free(ofproto->type);
1233 free(ofproto->mfr_desc);
1234 free(ofproto->hw_desc);
1235 free(ofproto->sw_desc);
1236 free(ofproto->serial_desc);
1237 free(ofproto->dp_desc);
1238 hmap_destroy(&ofproto->ports);
1239 hmap_destroy(&ofproto->ofport_usage);
1240 shash_destroy(&ofproto->port_by_name);
1241 simap_destroy(&ofproto->ofp_requests);
1242
1243 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1244 oftable_destroy(table);
1245 }
1246 free(ofproto->tables);
1247
1248 hmap_destroy(&ofproto->deletions);
1249
1250 free(ofproto->vlan_bitmap);
1251
1252 ofproto->ofproto_class->dealloc(ofproto);
1253 }
1254
1255 void
1256 ofproto_destroy(struct ofproto *p)
1257 OVS_EXCLUDED(ofproto_mutex)
1258 {
1259 struct ofport *ofport, *next_ofport;
1260 struct ofport_usage *usage, *next_usage;
1261
1262 if (!p) {
1263 return;
1264 }
1265
1266 if (p->meters) {
1267 meter_delete(p, 1, p->meter_features.max_meters);
1268 p->meter_features.max_meters = 0;
1269 free(p->meters);
1270 p->meters = NULL;
1271 }
1272
1273 ofproto_flush__(p);
1274 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1275 ofport_destroy(ofport);
1276 }
1277
1278 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1279 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1280 free(usage);
1281 }
1282
1283 p->ofproto_class->destruct(p);
1284 ofproto_destroy__(p);
1285 }
1286
1287 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1288 * kernel datapath, for example, this destroys the datapath in the kernel, and
1289 * with the netdev-based datapath, it tears down the data structures that
1290 * represent the datapath.
1291 *
1292 * The datapath should not be currently open as an ofproto. */
1293 int
1294 ofproto_delete(const char *name, const char *type)
1295 {
1296 const struct ofproto_class *class = ofproto_class_find__(type);
1297 return (!class ? EAFNOSUPPORT
1298 : !class->del ? EACCES
1299 : class->del(type, name));
1300 }
1301
1302 static void
1303 process_port_change(struct ofproto *ofproto, int error, char *devname)
1304 {
1305 if (error == ENOBUFS) {
1306 reinit_ports(ofproto);
1307 } else if (!error) {
1308 update_port(ofproto, devname);
1309 free(devname);
1310 }
1311 }
1312
1313 int
1314 ofproto_type_run(const char *datapath_type)
1315 {
1316 const struct ofproto_class *class;
1317 int error;
1318
1319 datapath_type = ofproto_normalize_type(datapath_type);
1320 class = ofproto_class_find__(datapath_type);
1321
1322 error = class->type_run ? class->type_run(datapath_type) : 0;
1323 if (error && error != EAGAIN) {
1324 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1325 datapath_type, ovs_strerror(error));
1326 }
1327 return error;
1328 }
1329
1330 int
1331 ofproto_type_run_fast(const char *datapath_type)
1332 {
1333 const struct ofproto_class *class;
1334 int error;
1335
1336 datapath_type = ofproto_normalize_type(datapath_type);
1337 class = ofproto_class_find__(datapath_type);
1338
1339 error = class->type_run_fast ? class->type_run_fast(datapath_type) : 0;
1340 if (error && error != EAGAIN) {
1341 VLOG_ERR_RL(&rl, "%s: type_run_fast failed (%s)",
1342 datapath_type, ovs_strerror(error));
1343 }
1344 return error;
1345 }
1346
1347 void
1348 ofproto_type_wait(const char *datapath_type)
1349 {
1350 const struct ofproto_class *class;
1351
1352 datapath_type = ofproto_normalize_type(datapath_type);
1353 class = ofproto_class_find__(datapath_type);
1354
1355 if (class->type_wait) {
1356 class->type_wait(datapath_type);
1357 }
1358 }
1359
1360 static bool
1361 any_pending_ops(const struct ofproto *p)
1362 OVS_EXCLUDED(ofproto_mutex)
1363 {
1364 bool b;
1365
1366 ovs_mutex_lock(&ofproto_mutex);
1367 b = !list_is_empty(&p->pending);
1368 ovs_mutex_unlock(&ofproto_mutex);
1369
1370 return b;
1371 }
1372
1373 int
1374 ofproto_run(struct ofproto *p)
1375 {
1376 struct sset changed_netdevs;
1377 const char *changed_netdev;
1378 struct ofport *ofport;
1379 int error;
1380
1381 error = p->ofproto_class->run(p);
1382 if (error && error != EAGAIN) {
1383 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1384 }
1385
1386 run_rule_executes(p);
1387
1388 /* Restore the eviction group heap invariant occasionally. */
1389 if (p->eviction_group_timer < time_msec()) {
1390 size_t i;
1391
1392 p->eviction_group_timer = time_msec() + 1000;
1393
1394 for (i = 0; i < p->n_tables; i++) {
1395 struct oftable *table = &p->tables[i];
1396 struct eviction_group *evg;
1397 struct cls_cursor cursor;
1398 struct rule *rule;
1399
1400 if (!table->eviction_fields) {
1401 continue;
1402 }
1403
1404 ovs_mutex_lock(&ofproto_mutex);
1405 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1406 heap_rebuild(&evg->rules);
1407 }
1408
1409 ovs_rwlock_rdlock(&table->cls.rwlock);
1410 cls_cursor_init(&cursor, &table->cls, NULL);
1411 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1412 if (!rule->eviction_group
1413 && (rule->idle_timeout || rule->hard_timeout)) {
1414 eviction_group_add_rule(rule);
1415 }
1416 }
1417 ovs_rwlock_unlock(&table->cls.rwlock);
1418 ovs_mutex_unlock(&ofproto_mutex);
1419 }
1420 }
1421
1422 if (p->ofproto_class->port_poll) {
1423 char *devname;
1424
1425 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1426 process_port_change(p, error, devname);
1427 }
1428 }
1429
1430 /* Update OpenFlow port status for any port whose netdev has changed.
1431 *
1432 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1433 * destroyed, so it's not safe to update ports directly from the
1434 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1435 * need this two-phase approach. */
1436 sset_init(&changed_netdevs);
1437 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1438 unsigned int change_seq = netdev_change_seq(ofport->netdev);
1439 if (ofport->change_seq != change_seq) {
1440 ofport->change_seq = change_seq;
1441 sset_add(&changed_netdevs, netdev_get_name(ofport->netdev));
1442 }
1443 }
1444 SSET_FOR_EACH (changed_netdev, &changed_netdevs) {
1445 update_port(p, changed_netdev);
1446 }
1447 sset_destroy(&changed_netdevs);
1448
1449 switch (p->state) {
1450 case S_OPENFLOW:
1451 connmgr_run(p->connmgr, handle_openflow);
1452 break;
1453
1454 case S_EVICT:
1455 connmgr_run(p->connmgr, NULL);
1456 ofproto_evict(p);
1457 if (!any_pending_ops(p)) {
1458 p->state = S_OPENFLOW;
1459 }
1460 break;
1461
1462 case S_FLUSH:
1463 connmgr_run(p->connmgr, NULL);
1464 ofproto_flush__(p);
1465 if (!any_pending_ops(p)) {
1466 connmgr_flushed(p->connmgr);
1467 p->state = S_OPENFLOW;
1468 }
1469 break;
1470
1471 default:
1472 NOT_REACHED();
1473 }
1474
1475 if (time_msec() >= p->next_op_report) {
1476 long long int ago = (time_msec() - p->first_op) / 1000;
1477 long long int interval = (p->last_op - p->first_op) / 1000;
1478 struct ds s;
1479
1480 ds_init(&s);
1481 ds_put_format(&s, "%d flow_mods ",
1482 p->n_add + p->n_delete + p->n_modify);
1483 if (interval == ago) {
1484 ds_put_format(&s, "in the last %lld s", ago);
1485 } else if (interval) {
1486 ds_put_format(&s, "in the %lld s starting %lld s ago",
1487 interval, ago);
1488 } else {
1489 ds_put_format(&s, "%lld s ago", ago);
1490 }
1491
1492 ds_put_cstr(&s, " (");
1493 if (p->n_add) {
1494 ds_put_format(&s, "%d adds, ", p->n_add);
1495 }
1496 if (p->n_delete) {
1497 ds_put_format(&s, "%d deletes, ", p->n_delete);
1498 }
1499 if (p->n_modify) {
1500 ds_put_format(&s, "%d modifications, ", p->n_modify);
1501 }
1502 s.length -= 2;
1503 ds_put_char(&s, ')');
1504
1505 VLOG_INFO("%s: %s", p->name, ds_cstr(&s));
1506 ds_destroy(&s);
1507
1508 p->n_add = p->n_delete = p->n_modify = 0;
1509 p->next_op_report = LLONG_MAX;
1510 }
1511
1512 return error;
1513 }
1514
1515 /* Performs periodic activity required by 'ofproto' that needs to be done
1516 * with the least possible latency.
1517 *
1518 * It makes sense to call this function a couple of times per poll loop, to
1519 * provide a significant performance boost on some benchmarks with the
1520 * ofproto-dpif implementation. */
1521 int
1522 ofproto_run_fast(struct ofproto *p)
1523 {
1524 int error;
1525
1526 error = p->ofproto_class->run_fast ? p->ofproto_class->run_fast(p) : 0;
1527 if (error && error != EAGAIN) {
1528 VLOG_ERR_RL(&rl, "%s: fastpath run failed (%s)",
1529 p->name, ovs_strerror(error));
1530 }
1531 return error;
1532 }
1533
1534 void
1535 ofproto_wait(struct ofproto *p)
1536 {
1537 struct ofport *ofport;
1538
1539 p->ofproto_class->wait(p);
1540 if (p->ofproto_class->port_poll_wait) {
1541 p->ofproto_class->port_poll_wait(p);
1542 }
1543
1544 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1545 if (ofport->change_seq != netdev_change_seq(ofport->netdev)) {
1546 poll_immediate_wake();
1547 }
1548 }
1549
1550 switch (p->state) {
1551 case S_OPENFLOW:
1552 connmgr_wait(p->connmgr, true);
1553 break;
1554
1555 case S_EVICT:
1556 case S_FLUSH:
1557 connmgr_wait(p->connmgr, false);
1558 if (!any_pending_ops(p)) {
1559 poll_immediate_wake();
1560 }
1561 break;
1562 }
1563 }
1564
1565 bool
1566 ofproto_is_alive(const struct ofproto *p)
1567 {
1568 return connmgr_has_controllers(p->connmgr);
1569 }
1570
1571 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1572 * memory_report(). */
1573 void
1574 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1575 {
1576 const struct oftable *table;
1577 unsigned int n_rules;
1578
1579 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1580
1581 ovs_mutex_lock(&ofproto_mutex);
1582 simap_increase(usage, "ops",
1583 ofproto->n_pending + hmap_count(&ofproto->deletions));
1584 ovs_mutex_unlock(&ofproto_mutex);
1585
1586 n_rules = 0;
1587 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1588 ovs_rwlock_rdlock(&table->cls.rwlock);
1589 n_rules += classifier_count(&table->cls);
1590 ovs_rwlock_unlock(&table->cls.rwlock);
1591 }
1592 simap_increase(usage, "rules", n_rules);
1593
1594 if (ofproto->ofproto_class->get_memory_usage) {
1595 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1596 }
1597
1598 connmgr_get_memory_usage(ofproto->connmgr, usage);
1599 }
1600
1601 void
1602 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1603 struct shash *info)
1604 {
1605 connmgr_get_controller_info(ofproto->connmgr, info);
1606 }
1607
1608 void
1609 ofproto_free_ofproto_controller_info(struct shash *info)
1610 {
1611 connmgr_free_controller_info(info);
1612 }
1613
1614 /* Makes a deep copy of 'old' into 'port'. */
1615 void
1616 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1617 {
1618 port->name = xstrdup(old->name);
1619 port->type = xstrdup(old->type);
1620 port->ofp_port = old->ofp_port;
1621 }
1622
1623 /* Frees memory allocated to members of 'ofproto_port'.
1624 *
1625 * Do not call this function on an ofproto_port obtained from
1626 * ofproto_port_dump_next(): that function retains ownership of the data in the
1627 * ofproto_port. */
1628 void
1629 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1630 {
1631 free(ofproto_port->name);
1632 free(ofproto_port->type);
1633 }
1634
1635 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1636 *
1637 * This function provides no status indication. An error status for the entire
1638 * dump operation is provided when it is completed by calling
1639 * ofproto_port_dump_done().
1640 */
1641 void
1642 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1643 const struct ofproto *ofproto)
1644 {
1645 dump->ofproto = ofproto;
1646 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1647 &dump->state);
1648 }
1649
1650 /* Attempts to retrieve another port from 'dump', which must have been created
1651 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1652 * 'port' and returns true. On failure, returns false.
1653 *
1654 * Failure might indicate an actual error or merely that the last port has been
1655 * dumped. An error status for the entire dump operation is provided when it
1656 * is completed by calling ofproto_port_dump_done().
1657 *
1658 * The ofproto owns the data stored in 'port'. It will remain valid until at
1659 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1660 * ofproto_port_dump_done(). */
1661 bool
1662 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1663 struct ofproto_port *port)
1664 {
1665 const struct ofproto *ofproto = dump->ofproto;
1666
1667 if (dump->error) {
1668 return false;
1669 }
1670
1671 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1672 port);
1673 if (dump->error) {
1674 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1675 return false;
1676 }
1677 return true;
1678 }
1679
1680 /* Completes port table dump operation 'dump', which must have been created
1681 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1682 * error-free, otherwise a positive errno value describing the problem. */
1683 int
1684 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1685 {
1686 const struct ofproto *ofproto = dump->ofproto;
1687 if (!dump->error) {
1688 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1689 dump->state);
1690 }
1691 return dump->error == EOF ? 0 : dump->error;
1692 }
1693
1694 /* Returns the type to pass to netdev_open() when a datapath of type
1695 * 'datapath_type' has a port of type 'port_type', for a few special
1696 * cases when a netdev type differs from a port type. For example, when
1697 * using the userspace datapath, a port of type "internal" needs to be
1698 * opened as "tap".
1699 *
1700 * Returns either 'type' itself or a string literal, which must not be
1701 * freed. */
1702 const char *
1703 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1704 {
1705 const struct ofproto_class *class;
1706
1707 datapath_type = ofproto_normalize_type(datapath_type);
1708 class = ofproto_class_find__(datapath_type);
1709 if (!class) {
1710 return port_type;
1711 }
1712
1713 return (class->port_open_type
1714 ? class->port_open_type(datapath_type, port_type)
1715 : port_type);
1716 }
1717
1718 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1719 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1720 * the port's OpenFlow port number.
1721 *
1722 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1723 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1724 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1725 * 'ofp_portp' is non-null). */
1726 int
1727 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1728 ofp_port_t *ofp_portp)
1729 {
1730 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1731 int error;
1732
1733 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1734 if (!error) {
1735 const char *netdev_name = netdev_get_name(netdev);
1736
1737 simap_put(&ofproto->ofp_requests, netdev_name,
1738 ofp_to_u16(ofp_port));
1739 update_port(ofproto, netdev_name);
1740 }
1741 if (ofp_portp) {
1742 struct ofproto_port ofproto_port;
1743
1744 ofproto_port_query_by_name(ofproto, netdev_get_name(netdev),
1745 &ofproto_port);
1746 *ofp_portp = error ? OFPP_NONE : ofproto_port.ofp_port;
1747 ofproto_port_destroy(&ofproto_port);
1748 }
1749 return error;
1750 }
1751
1752 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1753 * initializes '*port' appropriately; on failure, returns a positive errno
1754 * value.
1755 *
1756 * The caller owns the data in 'ofproto_port' and must free it with
1757 * ofproto_port_destroy() when it is no longer needed. */
1758 int
1759 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1760 struct ofproto_port *port)
1761 {
1762 int error;
1763
1764 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1765 if (error) {
1766 memset(port, 0, sizeof *port);
1767 }
1768 return error;
1769 }
1770
1771 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1772 * Returns 0 if successful, otherwise a positive errno. */
1773 int
1774 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1775 {
1776 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1777 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1778 struct simap_node *ofp_request_node;
1779 int error;
1780
1781 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1782 if (ofp_request_node) {
1783 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1784 }
1785
1786 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1787 if (!error && ofport) {
1788 /* 'name' is the netdev's name and update_port() is going to close the
1789 * netdev. Just in case update_port() refers to 'name' after it
1790 * destroys 'ofport', make a copy of it around the update_port()
1791 * call. */
1792 char *devname = xstrdup(name);
1793 update_port(ofproto, devname);
1794 free(devname);
1795 }
1796 return error;
1797 }
1798
1799 static int
1800 simple_flow_mod(struct ofproto *ofproto,
1801 const struct match *match, unsigned int priority,
1802 const struct ofpact *ofpacts, size_t ofpacts_len,
1803 enum ofp_flow_mod_command command)
1804 {
1805 struct ofputil_flow_mod fm;
1806
1807 memset(&fm, 0, sizeof fm);
1808 fm.match = *match;
1809 fm.priority = priority;
1810 fm.cookie = 0;
1811 fm.new_cookie = 0;
1812 fm.modify_cookie = false;
1813 fm.table_id = 0;
1814 fm.command = command;
1815 fm.idle_timeout = 0;
1816 fm.hard_timeout = 0;
1817 fm.buffer_id = UINT32_MAX;
1818 fm.out_port = OFPP_ANY;
1819 fm.out_group = OFPG_ANY;
1820 fm.flags = 0;
1821 fm.ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1822 fm.ofpacts_len = ofpacts_len;
1823
1824 return handle_flow_mod__(ofproto, NULL, &fm, NULL);
1825 }
1826
1827 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1828 * performs the 'n_actions' actions in 'actions'. The new flow will not
1829 * timeout.
1830 *
1831 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1832 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1833 * controllers; otherwise, it will be hidden.
1834 *
1835 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1836 *
1837 * This is a helper function for in-band control and fail-open. */
1838 void
1839 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1840 unsigned int priority,
1841 const struct ofpact *ofpacts, size_t ofpacts_len)
1842 OVS_EXCLUDED(ofproto_mutex)
1843 {
1844 const struct rule *rule;
1845 bool must_add;
1846
1847 /* First do a cheap check whether the rule we're looking for already exists
1848 * with the actions that we want. If it does, then we're done. */
1849 ovs_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1850 rule = rule_from_cls_rule(classifier_find_match_exactly(
1851 &ofproto->tables[0].cls, match, priority));
1852 if (rule) {
1853 ovs_mutex_lock(&rule->mutex);
1854 must_add = !ofpacts_equal(rule->actions->ofpacts,
1855 rule->actions->ofpacts_len,
1856 ofpacts, ofpacts_len);
1857 ovs_mutex_unlock(&rule->mutex);
1858 } else {
1859 must_add = true;
1860 }
1861 ovs_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1862
1863 /* If there's no such rule or the rule doesn't have the actions we want,
1864 * fall back to a executing a full flow mod. We can't optimize this at
1865 * all because we didn't take enough locks above to ensure that the flow
1866 * table didn't already change beneath us. */
1867 if (must_add) {
1868 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1869 OFPFC_MODIFY_STRICT);
1870 }
1871 }
1872
1873 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1874 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1875 * operation cannot be initiated now but may be retried later.
1876 *
1877 * This is a helper function for in-band control and fail-open and the "learn"
1878 * action. */
1879 int
1880 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1881 OVS_EXCLUDED(ofproto_mutex)
1882 {
1883 return handle_flow_mod__(ofproto, NULL, fm, NULL);
1884 }
1885
1886 /* Searches for a rule with matching criteria exactly equal to 'target' in
1887 * ofproto's table 0 and, if it finds one, deletes it.
1888 *
1889 * This is a helper function for in-band control and fail-open. */
1890 bool
1891 ofproto_delete_flow(struct ofproto *ofproto,
1892 const struct match *target, unsigned int priority)
1893 OVS_EXCLUDED(ofproto_mutex)
1894 {
1895 struct classifier *cls = &ofproto->tables[0].cls;
1896 struct rule *rule;
1897
1898 /* First do a cheap check whether the rule we're looking for has already
1899 * been deleted. If so, then we're done. */
1900 ovs_rwlock_rdlock(&cls->rwlock);
1901 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1902 priority));
1903 ovs_rwlock_unlock(&cls->rwlock);
1904 if (!rule) {
1905 return true;
1906 }
1907
1908 /* Fall back to a executing a full flow mod. We can't optimize this at all
1909 * because we didn't take enough locks above to ensure that the flow table
1910 * didn't already change beneath us. */
1911 return simple_flow_mod(ofproto, target, priority, NULL, 0,
1912 OFPFC_DELETE_STRICT) != OFPROTO_POSTPONE;
1913 }
1914
1915 /* Starts the process of deleting all of the flows from all of ofproto's flow
1916 * tables and then reintroducing the flows required by in-band control and
1917 * fail-open. The process will complete in a later call to ofproto_run(). */
1918 void
1919 ofproto_flush_flows(struct ofproto *ofproto)
1920 {
1921 COVERAGE_INC(ofproto_flush);
1922 ofproto->state = S_FLUSH;
1923 }
1924 \f
1925 static void
1926 reinit_ports(struct ofproto *p)
1927 {
1928 struct ofproto_port_dump dump;
1929 struct sset devnames;
1930 struct ofport *ofport;
1931 struct ofproto_port ofproto_port;
1932 const char *devname;
1933
1934 COVERAGE_INC(ofproto_reinit_ports);
1935
1936 sset_init(&devnames);
1937 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1938 sset_add(&devnames, netdev_get_name(ofport->netdev));
1939 }
1940 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1941 sset_add(&devnames, ofproto_port.name);
1942 }
1943
1944 SSET_FOR_EACH (devname, &devnames) {
1945 update_port(p, devname);
1946 }
1947 sset_destroy(&devnames);
1948 }
1949
1950 static ofp_port_t
1951 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1952 {
1953 uint16_t port_idx;
1954
1955 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1956 port_idx = port_idx ? port_idx : UINT16_MAX;
1957
1958 if (port_idx >= ofproto->max_ports
1959 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
1960 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
1961 long long int last_used_at, lru = LLONG_MAX;
1962
1963 /* Search for a free OpenFlow port number. We try not to
1964 * immediately reuse them to prevent problems due to old
1965 * flows. */
1966 for (;;) {
1967 if (++ofproto->alloc_port_no >= ofproto->max_ports) {
1968 ofproto->alloc_port_no = 1;
1969 }
1970 last_used_at = ofport_get_usage(ofproto,
1971 u16_to_ofp(ofproto->alloc_port_no));
1972 if (!last_used_at) {
1973 port_idx = ofproto->alloc_port_no;
1974 break;
1975 } else if (last_used_at < lru) {
1976 lru = last_used_at;
1977 lru_ofport = ofproto->alloc_port_no;
1978 }
1979
1980 if (ofproto->alloc_port_no == end_port_no) {
1981 if (lru_ofport) {
1982 port_idx = lru_ofport;
1983 break;
1984 }
1985 return OFPP_NONE;
1986 }
1987 }
1988 }
1989 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
1990 return u16_to_ofp(port_idx);
1991 }
1992
1993 static void
1994 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
1995 {
1996 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
1997 ofport_set_usage(ofproto, ofp_port, time_msec());
1998 }
1999 }
2000
2001 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2002 * pointer if the netdev cannot be opened. On success, also fills in
2003 * 'opp'. */
2004 static struct netdev *
2005 ofport_open(struct ofproto *ofproto,
2006 struct ofproto_port *ofproto_port,
2007 struct ofputil_phy_port *pp)
2008 {
2009 enum netdev_flags flags;
2010 struct netdev *netdev;
2011 int error;
2012
2013 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2014 if (error) {
2015 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2016 "cannot be opened (%s)",
2017 ofproto->name,
2018 ofproto_port->name, ofproto_port->ofp_port,
2019 ofproto_port->name, ovs_strerror(error));
2020 return NULL;
2021 }
2022
2023 if (ofproto_port->ofp_port == OFPP_NONE) {
2024 if (!strcmp(ofproto->name, ofproto_port->name)) {
2025 ofproto_port->ofp_port = OFPP_LOCAL;
2026 } else {
2027 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2028 ofproto_port->name);
2029 }
2030 }
2031 pp->port_no = ofproto_port->ofp_port;
2032 netdev_get_etheraddr(netdev, pp->hw_addr);
2033 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2034 netdev_get_flags(netdev, &flags);
2035 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2036 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2037 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2038 &pp->supported, &pp->peer);
2039 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2040 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2041
2042 return netdev;
2043 }
2044
2045 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2046 * port number, and 'config' bits other than OFPUTIL_PS_LINK_DOWN are
2047 * disregarded. */
2048 static bool
2049 ofport_equal(const struct ofputil_phy_port *a,
2050 const struct ofputil_phy_port *b)
2051 {
2052 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2053 && a->state == b->state
2054 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2055 && a->curr == b->curr
2056 && a->advertised == b->advertised
2057 && a->supported == b->supported
2058 && a->peer == b->peer
2059 && a->curr_speed == b->curr_speed
2060 && a->max_speed == b->max_speed);
2061 }
2062
2063 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2064 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2065 * one with the same name or port number). */
2066 static void
2067 ofport_install(struct ofproto *p,
2068 struct netdev *netdev, const struct ofputil_phy_port *pp)
2069 {
2070 const char *netdev_name = netdev_get_name(netdev);
2071 struct ofport *ofport;
2072 int error;
2073
2074 /* Create ofport. */
2075 ofport = p->ofproto_class->port_alloc();
2076 if (!ofport) {
2077 error = ENOMEM;
2078 goto error;
2079 }
2080 ofport->ofproto = p;
2081 ofport->netdev = netdev;
2082 ofport->change_seq = netdev_change_seq(netdev);
2083 ofport->pp = *pp;
2084 ofport->ofp_port = pp->port_no;
2085 ofport->created = time_msec();
2086
2087 /* Add port to 'p'. */
2088 hmap_insert(&p->ports, &ofport->hmap_node,
2089 hash_ofp_port(ofport->ofp_port));
2090 shash_add(&p->port_by_name, netdev_name, ofport);
2091
2092 update_mtu(p, ofport);
2093
2094 /* Let the ofproto_class initialize its private data. */
2095 error = p->ofproto_class->port_construct(ofport);
2096 if (error) {
2097 goto error;
2098 }
2099 connmgr_send_port_status(p->connmgr, pp, OFPPR_ADD);
2100 return;
2101
2102 error:
2103 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2104 p->name, netdev_name, ovs_strerror(error));
2105 if (ofport) {
2106 ofport_destroy__(ofport);
2107 } else {
2108 netdev_close(netdev);
2109 }
2110 }
2111
2112 /* Removes 'ofport' from 'p' and destroys it. */
2113 static void
2114 ofport_remove(struct ofport *ofport)
2115 {
2116 connmgr_send_port_status(ofport->ofproto->connmgr, &ofport->pp,
2117 OFPPR_DELETE);
2118 ofport_destroy(ofport);
2119 }
2120
2121 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2122 * destroys it. */
2123 static void
2124 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2125 {
2126 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2127 if (port) {
2128 ofport_remove(port);
2129 }
2130 }
2131
2132 /* Updates 'port' with new 'pp' description.
2133 *
2134 * Does not handle a name or port number change. The caller must implement
2135 * such a change as a delete followed by an add. */
2136 static void
2137 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2138 {
2139 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2140 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2141 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2142 port->pp.state = pp->state;
2143 port->pp.curr = pp->curr;
2144 port->pp.advertised = pp->advertised;
2145 port->pp.supported = pp->supported;
2146 port->pp.peer = pp->peer;
2147 port->pp.curr_speed = pp->curr_speed;
2148 port->pp.max_speed = pp->max_speed;
2149
2150 connmgr_send_port_status(port->ofproto->connmgr, &port->pp, OFPPR_MODIFY);
2151 }
2152
2153 /* Update OpenFlow 'state' in 'port' and notify controller. */
2154 void
2155 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2156 {
2157 if (port->pp.state != state) {
2158 port->pp.state = state;
2159 connmgr_send_port_status(port->ofproto->connmgr, &port->pp,
2160 OFPPR_MODIFY);
2161 }
2162 }
2163
2164 void
2165 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2166 {
2167 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2168 if (port) {
2169 if (port->ofproto->ofproto_class->set_realdev) {
2170 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2171 }
2172 if (port->ofproto->ofproto_class->set_stp_port) {
2173 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2174 }
2175 if (port->ofproto->ofproto_class->set_cfm) {
2176 port->ofproto->ofproto_class->set_cfm(port, NULL);
2177 }
2178 if (port->ofproto->ofproto_class->bundle_remove) {
2179 port->ofproto->ofproto_class->bundle_remove(port);
2180 }
2181 }
2182 }
2183
2184 static void
2185 ofport_destroy__(struct ofport *port)
2186 {
2187 struct ofproto *ofproto = port->ofproto;
2188 const char *name = netdev_get_name(port->netdev);
2189
2190 hmap_remove(&ofproto->ports, &port->hmap_node);
2191 shash_delete(&ofproto->port_by_name,
2192 shash_find(&ofproto->port_by_name, name));
2193
2194 netdev_close(port->netdev);
2195 ofproto->ofproto_class->port_dealloc(port);
2196 }
2197
2198 static void
2199 ofport_destroy(struct ofport *port)
2200 {
2201 if (port) {
2202 dealloc_ofp_port(port->ofproto, port->ofp_port);
2203 port->ofproto->ofproto_class->port_destruct(port);
2204 ofport_destroy__(port);
2205 }
2206 }
2207
2208 struct ofport *
2209 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2210 {
2211 struct ofport *port;
2212
2213 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2214 &ofproto->ports) {
2215 if (port->ofp_port == ofp_port) {
2216 return port;
2217 }
2218 }
2219 return NULL;
2220 }
2221
2222 static long long int
2223 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2224 {
2225 struct ofport_usage *usage;
2226
2227 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2228 &ofproto->ofport_usage) {
2229 if (usage->ofp_port == ofp_port) {
2230 return usage->last_used;
2231 }
2232 }
2233 return 0;
2234 }
2235
2236 static void
2237 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2238 long long int last_used)
2239 {
2240 struct ofport_usage *usage;
2241 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2242 &ofproto->ofport_usage) {
2243 if (usage->ofp_port == ofp_port) {
2244 usage->last_used = last_used;
2245 return;
2246 }
2247 }
2248 ovs_assert(last_used == LLONG_MAX);
2249
2250 usage = xmalloc(sizeof *usage);
2251 usage->ofp_port = ofp_port;
2252 usage->last_used = last_used;
2253 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2254 hash_ofp_port(ofp_port));
2255 }
2256
2257 int
2258 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2259 {
2260 struct ofproto *ofproto = port->ofproto;
2261 int error;
2262
2263 if (ofproto->ofproto_class->port_get_stats) {
2264 error = ofproto->ofproto_class->port_get_stats(port, stats);
2265 } else {
2266 error = EOPNOTSUPP;
2267 }
2268
2269 return error;
2270 }
2271
2272 static void
2273 update_port(struct ofproto *ofproto, const char *name)
2274 {
2275 struct ofproto_port ofproto_port;
2276 struct ofputil_phy_port pp;
2277 struct netdev *netdev;
2278 struct ofport *port;
2279
2280 COVERAGE_INC(ofproto_update_port);
2281
2282 /* Fetch 'name''s location and properties from the datapath. */
2283 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2284 ? ofport_open(ofproto, &ofproto_port, &pp)
2285 : NULL);
2286
2287 if (netdev) {
2288 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2289 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2290 struct netdev *old_netdev = port->netdev;
2291
2292 /* 'name' hasn't changed location. Any properties changed? */
2293 if (!ofport_equal(&port->pp, &pp)) {
2294 ofport_modified(port, &pp);
2295 }
2296
2297 update_mtu(ofproto, port);
2298
2299 /* Install the newly opened netdev in case it has changed.
2300 * Don't close the old netdev yet in case port_modified has to
2301 * remove a retained reference to it.*/
2302 port->netdev = netdev;
2303 port->change_seq = netdev_change_seq(netdev);
2304
2305 if (port->ofproto->ofproto_class->port_modified) {
2306 port->ofproto->ofproto_class->port_modified(port);
2307 }
2308
2309 netdev_close(old_netdev);
2310 } else {
2311 /* If 'port' is nonnull then its name differs from 'name' and thus
2312 * we should delete it. If we think there's a port named 'name'
2313 * then its port number must be wrong now so delete it too. */
2314 if (port) {
2315 ofport_remove(port);
2316 }
2317 ofport_remove_with_name(ofproto, name);
2318 ofport_install(ofproto, netdev, &pp);
2319 }
2320 } else {
2321 /* Any port named 'name' is gone now. */
2322 ofport_remove_with_name(ofproto, name);
2323 }
2324 ofproto_port_destroy(&ofproto_port);
2325 }
2326
2327 static int
2328 init_ports(struct ofproto *p)
2329 {
2330 struct ofproto_port_dump dump;
2331 struct ofproto_port ofproto_port;
2332 struct shash_node *node, *next;
2333
2334 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2335 const char *name = ofproto_port.name;
2336
2337 if (shash_find(&p->port_by_name, name)) {
2338 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2339 p->name, name);
2340 } else {
2341 struct ofputil_phy_port pp;
2342 struct netdev *netdev;
2343
2344 /* Check if an OpenFlow port number had been requested. */
2345 node = shash_find(&init_ofp_ports, name);
2346 if (node) {
2347 const struct iface_hint *iface_hint = node->data;
2348 simap_put(&p->ofp_requests, name,
2349 ofp_to_u16(iface_hint->ofp_port));
2350 }
2351
2352 netdev = ofport_open(p, &ofproto_port, &pp);
2353 if (netdev) {
2354 ofport_install(p, netdev, &pp);
2355 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2356 p->alloc_port_no = MAX(p->alloc_port_no,
2357 ofp_to_u16(ofproto_port.ofp_port));
2358 }
2359 }
2360 }
2361 }
2362
2363 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2364 struct iface_hint *iface_hint = node->data;
2365
2366 if (!strcmp(iface_hint->br_name, p->name)) {
2367 free(iface_hint->br_name);
2368 free(iface_hint->br_type);
2369 free(iface_hint);
2370 shash_delete(&init_ofp_ports, node);
2371 }
2372 }
2373
2374 return 0;
2375 }
2376
2377 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2378 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2379 static int
2380 find_min_mtu(struct ofproto *p)
2381 {
2382 struct ofport *ofport;
2383 int mtu = 0;
2384
2385 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2386 struct netdev *netdev = ofport->netdev;
2387 int dev_mtu;
2388
2389 /* Skip any internal ports, since that's what we're trying to
2390 * set. */
2391 if (!strcmp(netdev_get_type(netdev), "internal")) {
2392 continue;
2393 }
2394
2395 if (netdev_get_mtu(netdev, &dev_mtu)) {
2396 continue;
2397 }
2398 if (!mtu || dev_mtu < mtu) {
2399 mtu = dev_mtu;
2400 }
2401 }
2402
2403 return mtu ? mtu: ETH_PAYLOAD_MAX;
2404 }
2405
2406 /* Update MTU of all datapath devices on 'p' to the minimum of the
2407 * non-datapath ports in event of 'port' added or changed. */
2408 static void
2409 update_mtu(struct ofproto *p, struct ofport *port)
2410 {
2411 struct ofport *ofport;
2412 struct netdev *netdev = port->netdev;
2413 int dev_mtu, old_min;
2414
2415 if (netdev_get_mtu(netdev, &dev_mtu)) {
2416 port->mtu = 0;
2417 return;
2418 }
2419 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2420 if (dev_mtu > p->min_mtu) {
2421 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2422 dev_mtu = p->min_mtu;
2423 }
2424 }
2425 port->mtu = dev_mtu;
2426 return;
2427 }
2428
2429 /* For non-internal port find new min mtu. */
2430 old_min = p->min_mtu;
2431 port->mtu = dev_mtu;
2432 p->min_mtu = find_min_mtu(p);
2433 if (p->min_mtu == old_min) {
2434 return;
2435 }
2436
2437 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2438 struct netdev *netdev = ofport->netdev;
2439
2440 if (!strcmp(netdev_get_type(netdev), "internal")) {
2441 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2442 ofport->mtu = p->min_mtu;
2443 }
2444 }
2445 }
2446 }
2447 \f
2448 void
2449 ofproto_rule_ref(struct rule *rule)
2450 {
2451 if (rule) {
2452 unsigned int orig;
2453
2454 atomic_add(&rule->ref_count, 1, &orig);
2455 ovs_assert(orig != 0);
2456 }
2457 }
2458
2459 void
2460 ofproto_rule_unref(struct rule *rule)
2461 {
2462 if (rule) {
2463 unsigned int orig;
2464
2465 atomic_sub(&rule->ref_count, 1, &orig);
2466 if (orig == 1) {
2467 rule->ofproto->ofproto_class->rule_destruct(rule);
2468 ofproto_rule_destroy__(rule);
2469 } else {
2470 ovs_assert(orig != 0);
2471 }
2472 }
2473 }
2474
2475 struct rule_actions *
2476 rule_get_actions(const struct rule *rule)
2477 OVS_EXCLUDED(rule->mutex)
2478 {
2479 struct rule_actions *actions;
2480
2481 ovs_mutex_lock(&rule->mutex);
2482 actions = rule_get_actions__(rule);
2483 ovs_mutex_unlock(&rule->mutex);
2484
2485 return actions;
2486 }
2487
2488 struct rule_actions *
2489 rule_get_actions__(const struct rule *rule)
2490 OVS_REQUIRES(rule->mutex)
2491 {
2492 rule_actions_ref(rule->actions);
2493 return rule->actions;
2494 }
2495
2496 static void
2497 ofproto_rule_destroy__(struct rule *rule)
2498 OVS_NO_THREAD_SAFETY_ANALYSIS
2499 {
2500 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2501 rule_actions_unref(rule->actions);
2502 ovs_mutex_destroy(&rule->mutex);
2503 rule->ofproto->ofproto_class->rule_dealloc(rule);
2504 }
2505
2506 static uint32_t get_provider_meter_id(const struct ofproto *,
2507 uint32_t of_meter_id);
2508
2509 /* Creates and returns a new 'struct rule_actions', with a ref_count of 1,
2510 * whose actions are a copy of from the 'ofpacts_len' bytes of 'ofpacts'. */
2511 struct rule_actions *
2512 rule_actions_create(const struct ofproto *ofproto,
2513 const struct ofpact *ofpacts, size_t ofpacts_len)
2514 {
2515 struct rule_actions *actions;
2516
2517 actions = xmalloc(sizeof *actions);
2518 atomic_init(&actions->ref_count, 1);
2519 actions->ofpacts = xmemdup(ofpacts, ofpacts_len);
2520 actions->ofpacts_len = ofpacts_len;
2521 actions->provider_meter_id
2522 = get_provider_meter_id(ofproto,
2523 ofpacts_get_meter(ofpacts, ofpacts_len));
2524
2525 return actions;
2526 }
2527
2528 /* Increments 'actions''s ref_count. */
2529 void
2530 rule_actions_ref(struct rule_actions *actions)
2531 {
2532 if (actions) {
2533 unsigned int orig;
2534
2535 atomic_add(&actions->ref_count, 1, &orig);
2536 ovs_assert(orig != 0);
2537 }
2538 }
2539
2540 /* Decrements 'actions''s ref_count and frees 'actions' if the ref_count
2541 * reaches 0. */
2542 void
2543 rule_actions_unref(struct rule_actions *actions)
2544 {
2545 if (actions) {
2546 unsigned int orig;
2547
2548 atomic_sub(&actions->ref_count, 1, &orig);
2549 if (orig == 1) {
2550 free(actions->ofpacts);
2551 free(actions);
2552 } else {
2553 ovs_assert(orig != 0);
2554 }
2555 }
2556 }
2557
2558 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2559 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2560 static bool
2561 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2562 OVS_REQUIRES(ofproto_mutex)
2563 {
2564 return (port == OFPP_ANY
2565 || ofpacts_output_to_port(rule->actions->ofpacts,
2566 rule->actions->ofpacts_len, port));
2567 }
2568
2569 /* Returns true if 'rule' has group and equals group_id. */
2570 static bool
2571 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2572 OVS_REQUIRES(ofproto_mutex)
2573 {
2574 return (group_id == OFPG11_ANY
2575 || ofpacts_output_to_group(rule->actions->ofpacts,
2576 rule->actions->ofpacts_len, group_id));
2577 }
2578
2579 /* Returns true if a rule related to 'op' has an OpenFlow OFPAT_OUTPUT or
2580 * OFPAT_ENQUEUE action that outputs to 'out_port'. */
2581 bool
2582 ofoperation_has_out_port(const struct ofoperation *op, ofp_port_t out_port)
2583 OVS_REQUIRES(ofproto_mutex)
2584 {
2585 if (ofproto_rule_has_out_port(op->rule, out_port)) {
2586 return true;
2587 }
2588
2589 switch (op->type) {
2590 case OFOPERATION_ADD:
2591 case OFOPERATION_DELETE:
2592 return false;
2593
2594 case OFOPERATION_MODIFY:
2595 case OFOPERATION_REPLACE:
2596 return ofpacts_output_to_port(op->actions->ofpacts,
2597 op->actions->ofpacts_len, out_port);
2598 }
2599
2600 NOT_REACHED();
2601 }
2602
2603 static void
2604 rule_execute_destroy(struct rule_execute *e)
2605 {
2606 ofproto_rule_unref(e->rule);
2607 list_remove(&e->list_node);
2608 free(e);
2609 }
2610
2611 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2612 * by passing them to the ofproto provider. */
2613 static void
2614 run_rule_executes(struct ofproto *ofproto)
2615 OVS_EXCLUDED(ofproto_mutex)
2616 {
2617 struct rule_execute *e, *next;
2618 struct list executes;
2619
2620 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2621 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2622 union flow_in_port in_port_;
2623 struct flow flow;
2624
2625 in_port_.ofp_port = e->in_port;
2626 flow_extract(e->packet, 0, 0, NULL, &in_port_, &flow);
2627 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2628
2629 rule_execute_destroy(e);
2630 }
2631 }
2632
2633 /* Destroys and discards all "rule_execute" operations queued up in
2634 * ofproto->rule_executes. */
2635 static void
2636 destroy_rule_executes(struct ofproto *ofproto)
2637 {
2638 struct rule_execute *e, *next;
2639 struct list executes;
2640
2641 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2642 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2643 ofpbuf_delete(e->packet);
2644 rule_execute_destroy(e);
2645 }
2646 }
2647
2648 /* Returns true if 'rule' should be hidden from the controller.
2649 *
2650 * Rules with priority higher than UINT16_MAX are set up by ofproto itself
2651 * (e.g. by in-band control) and are intentionally hidden from the
2652 * controller. */
2653 static bool
2654 ofproto_rule_is_hidden(const struct rule *rule)
2655 {
2656 return rule->cr.priority > UINT16_MAX;
2657 }
2658
2659 static enum oftable_flags
2660 rule_get_flags(const struct rule *rule)
2661 {
2662 return rule->ofproto->tables[rule->table_id].flags;
2663 }
2664
2665 static bool
2666 rule_is_modifiable(const struct rule *rule)
2667 {
2668 return !(rule_get_flags(rule) & OFTABLE_READONLY);
2669 }
2670 \f
2671 static enum ofperr
2672 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2673 {
2674 ofconn_send_reply(ofconn, make_echo_reply(oh));
2675 return 0;
2676 }
2677
2678 static enum ofperr
2679 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2680 {
2681 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2682 struct ofputil_switch_features features;
2683 struct ofport *port;
2684 bool arp_match_ip;
2685 struct ofpbuf *b;
2686 int n_tables;
2687 int i;
2688
2689 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2690 &features.actions);
2691 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2692
2693 /* Count only non-hidden tables in the number of tables. (Hidden tables,
2694 * if present, are always at the end.) */
2695 n_tables = ofproto->n_tables;
2696 for (i = 0; i < ofproto->n_tables; i++) {
2697 if (ofproto->tables[i].flags & OFTABLE_HIDDEN) {
2698 n_tables = i;
2699 break;
2700 }
2701 }
2702
2703 features.datapath_id = ofproto->datapath_id;
2704 features.n_buffers = pktbuf_capacity();
2705 features.n_tables = n_tables;
2706 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2707 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2708 if (arp_match_ip) {
2709 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2710 }
2711 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2712 features.auxiliary_id = 0;
2713 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2714 oh->xid);
2715 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2716 ofputil_put_switch_features_port(&port->pp, b);
2717 }
2718
2719 ofconn_send_reply(ofconn, b);
2720 return 0;
2721 }
2722
2723 static enum ofperr
2724 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2725 {
2726 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2727 struct ofp_switch_config *osc;
2728 enum ofp_config_flags flags;
2729 struct ofpbuf *buf;
2730
2731 /* Send reply. */
2732 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2733 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2734 flags = ofproto->frag_handling;
2735 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2736 if (oh->version < OFP13_VERSION
2737 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2738 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2739 }
2740 osc->flags = htons(flags);
2741 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2742 ofconn_send_reply(ofconn, buf);
2743
2744 return 0;
2745 }
2746
2747 static enum ofperr
2748 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2749 {
2750 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2751 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2752 uint16_t flags = ntohs(osc->flags);
2753
2754 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2755 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2756 enum ofp_config_flags cur = ofproto->frag_handling;
2757 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2758
2759 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2760 if (cur != next) {
2761 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2762 ofproto->frag_handling = next;
2763 } else {
2764 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2765 ofproto->name,
2766 ofputil_frag_handling_to_string(next));
2767 }
2768 }
2769 }
2770 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2771 ofconn_set_invalid_ttl_to_controller(ofconn,
2772 (oh->version < OFP13_VERSION
2773 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2774
2775 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2776
2777 return 0;
2778 }
2779
2780 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2781 * error message code for the caller to propagate upward. Otherwise, returns
2782 * 0.
2783 *
2784 * The log message mentions 'msg_type'. */
2785 static enum ofperr
2786 reject_slave_controller(struct ofconn *ofconn)
2787 {
2788 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2789 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2790 return OFPERR_OFPBRC_EPERM;
2791 } else {
2792 return 0;
2793 }
2794 }
2795
2796 /* Checks that the 'ofpacts_len' bytes of actions in 'ofpacts' are appropriate
2797 * for a packet with the prerequisites satisfied by 'flow' in table 'table_id'.
2798 * 'flow' may be temporarily modified, but is restored at return.
2799 */
2800 static enum ofperr
2801 ofproto_check_ofpacts(struct ofproto *ofproto,
2802 const struct ofpact ofpacts[], size_t ofpacts_len,
2803 struct flow *flow, uint8_t table_id)
2804 {
2805 enum ofperr error;
2806 uint32_t mid;
2807
2808 error = ofpacts_check(ofpacts, ofpacts_len, flow,
2809 u16_to_ofp(ofproto->max_ports), table_id);
2810 if (error) {
2811 return error;
2812 }
2813
2814 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2815 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2816 return OFPERR_OFPMMFC_INVALID_METER;
2817 }
2818 return 0;
2819 }
2820
2821 static enum ofperr
2822 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2823 {
2824 struct ofproto *p = ofconn_get_ofproto(ofconn);
2825 struct ofputil_packet_out po;
2826 struct ofpbuf *payload;
2827 uint64_t ofpacts_stub[1024 / 8];
2828 struct ofpbuf ofpacts;
2829 struct flow flow;
2830 union flow_in_port in_port_;
2831 enum ofperr error;
2832
2833 COVERAGE_INC(ofproto_packet_out);
2834
2835 error = reject_slave_controller(ofconn);
2836 if (error) {
2837 goto exit;
2838 }
2839
2840 /* Decode message. */
2841 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2842 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2843 if (error) {
2844 goto exit_free_ofpacts;
2845 }
2846 if (ofp_to_u16(po.in_port) >= p->max_ports
2847 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2848 error = OFPERR_OFPBRC_BAD_PORT;
2849 goto exit_free_ofpacts;
2850 }
2851
2852
2853 /* Get payload. */
2854 if (po.buffer_id != UINT32_MAX) {
2855 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2856 if (error || !payload) {
2857 goto exit_free_ofpacts;
2858 }
2859 } else {
2860 /* Ensure that the L3 header is 32-bit aligned. */
2861 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2862 }
2863
2864 /* Verify actions against packet, then send packet if successful. */
2865 in_port_.ofp_port = po.in_port;
2866 flow_extract(payload, 0, 0, NULL, &in_port_, &flow);
2867 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len, &flow, 0);
2868 if (!error) {
2869 error = p->ofproto_class->packet_out(p, payload, &flow,
2870 po.ofpacts, po.ofpacts_len);
2871 }
2872 ofpbuf_delete(payload);
2873
2874 exit_free_ofpacts:
2875 ofpbuf_uninit(&ofpacts);
2876 exit:
2877 return error;
2878 }
2879
2880 static void
2881 update_port_config(struct ofport *port,
2882 enum ofputil_port_config config,
2883 enum ofputil_port_config mask)
2884 {
2885 enum ofputil_port_config old_config = port->pp.config;
2886 enum ofputil_port_config toggle;
2887
2888 toggle = (config ^ port->pp.config) & mask;
2889 if (toggle & OFPUTIL_PC_PORT_DOWN) {
2890 if (config & OFPUTIL_PC_PORT_DOWN) {
2891 netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL);
2892 } else {
2893 netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL);
2894 }
2895 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2896 }
2897
2898 port->pp.config ^= toggle;
2899 if (port->pp.config != old_config) {
2900 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2901 }
2902 }
2903
2904 static enum ofperr
2905 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2906 {
2907 struct ofproto *p = ofconn_get_ofproto(ofconn);
2908 struct ofputil_port_mod pm;
2909 struct ofport *port;
2910 enum ofperr error;
2911
2912 error = reject_slave_controller(ofconn);
2913 if (error) {
2914 return error;
2915 }
2916
2917 error = ofputil_decode_port_mod(oh, &pm);
2918 if (error) {
2919 return error;
2920 }
2921
2922 port = ofproto_get_port(p, pm.port_no);
2923 if (!port) {
2924 return OFPERR_OFPPMFC_BAD_PORT;
2925 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
2926 return OFPERR_OFPPMFC_BAD_HW_ADDR;
2927 } else {
2928 update_port_config(port, pm.config, pm.mask);
2929 if (pm.advertise) {
2930 netdev_set_advertisements(port->netdev, pm.advertise);
2931 }
2932 }
2933 return 0;
2934 }
2935
2936 static enum ofperr
2937 handle_desc_stats_request(struct ofconn *ofconn,
2938 const struct ofp_header *request)
2939 {
2940 static const char *default_mfr_desc = "Nicira, Inc.";
2941 static const char *default_hw_desc = "Open vSwitch";
2942 static const char *default_sw_desc = VERSION;
2943 static const char *default_serial_desc = "None";
2944 static const char *default_dp_desc = "None";
2945
2946 struct ofproto *p = ofconn_get_ofproto(ofconn);
2947 struct ofp_desc_stats *ods;
2948 struct ofpbuf *msg;
2949
2950 msg = ofpraw_alloc_stats_reply(request, 0);
2951 ods = ofpbuf_put_zeros(msg, sizeof *ods);
2952 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
2953 sizeof ods->mfr_desc);
2954 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
2955 sizeof ods->hw_desc);
2956 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
2957 sizeof ods->sw_desc);
2958 ovs_strlcpy(ods->serial_num,
2959 p->serial_desc ? p->serial_desc : default_serial_desc,
2960 sizeof ods->serial_num);
2961 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
2962 sizeof ods->dp_desc);
2963 ofconn_send_reply(ofconn, msg);
2964
2965 return 0;
2966 }
2967
2968 static enum ofperr
2969 handle_table_stats_request(struct ofconn *ofconn,
2970 const struct ofp_header *request)
2971 {
2972 struct ofproto *p = ofconn_get_ofproto(ofconn);
2973 struct ofp12_table_stats *ots;
2974 struct ofpbuf *msg;
2975 int n_tables;
2976 size_t i;
2977
2978 /* Set up default values.
2979 *
2980 * ofp12_table_stats is used as a generic structure as
2981 * it is able to hold all the fields for ofp10_table_stats
2982 * and ofp11_table_stats (and of course itself).
2983 */
2984 ots = xcalloc(p->n_tables, sizeof *ots);
2985 for (i = 0; i < p->n_tables; i++) {
2986 ots[i].table_id = i;
2987 sprintf(ots[i].name, "table%zu", i);
2988 ots[i].match = htonll(OFPXMT13_MASK);
2989 ots[i].wildcards = htonll(OFPXMT13_MASK);
2990 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
2991 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
2992 ots[i].write_setfields = htonll(OFPXMT13_MASK);
2993 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
2994 ots[i].metadata_match = OVS_BE64_MAX;
2995 ots[i].metadata_write = OVS_BE64_MAX;
2996 ots[i].instructions = htonl(OFPIT11_ALL);
2997 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
2998 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
2999 ovs_rwlock_rdlock(&p->tables[i].cls.rwlock);
3000 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
3001 ovs_rwlock_unlock(&p->tables[i].cls.rwlock);
3002 }
3003
3004 p->ofproto_class->get_tables(p, ots);
3005
3006 /* Post-process the tables, dropping hidden tables. */
3007 n_tables = p->n_tables;
3008 for (i = 0; i < p->n_tables; i++) {
3009 const struct oftable *table = &p->tables[i];
3010
3011 if (table->flags & OFTABLE_HIDDEN) {
3012 n_tables = i;
3013 break;
3014 }
3015
3016 if (table->name) {
3017 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
3018 }
3019
3020 if (table->max_flows < ntohl(ots[i].max_entries)) {
3021 ots[i].max_entries = htonl(table->max_flows);
3022 }
3023 }
3024
3025 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
3026 ofconn_send_reply(ofconn, msg);
3027
3028 free(ots);
3029
3030 return 0;
3031 }
3032
3033 static void
3034 append_port_stat(struct ofport *port, struct list *replies)
3035 {
3036 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3037
3038 calc_duration(port->created, time_msec(),
3039 &ops.duration_sec, &ops.duration_nsec);
3040
3041 /* Intentionally ignore return value, since errors will set
3042 * 'stats' to all-1s, which is correct for OpenFlow, and
3043 * netdev_get_stats() will log errors. */
3044 ofproto_port_get_stats(port, &ops.stats);
3045
3046 ofputil_append_port_stat(replies, &ops);
3047 }
3048
3049 static enum ofperr
3050 handle_port_stats_request(struct ofconn *ofconn,
3051 const struct ofp_header *request)
3052 {
3053 struct ofproto *p = ofconn_get_ofproto(ofconn);
3054 struct ofport *port;
3055 struct list replies;
3056 ofp_port_t port_no;
3057 enum ofperr error;
3058
3059 error = ofputil_decode_port_stats_request(request, &port_no);
3060 if (error) {
3061 return error;
3062 }
3063
3064 ofpmp_init(&replies, request);
3065 if (port_no != OFPP_ANY) {
3066 port = ofproto_get_port(p, port_no);
3067 if (port) {
3068 append_port_stat(port, &replies);
3069 }
3070 } else {
3071 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3072 append_port_stat(port, &replies);
3073 }
3074 }
3075
3076 ofconn_send_replies(ofconn, &replies);
3077 return 0;
3078 }
3079
3080 static enum ofperr
3081 handle_port_desc_stats_request(struct ofconn *ofconn,
3082 const struct ofp_header *request)
3083 {
3084 struct ofproto *p = ofconn_get_ofproto(ofconn);
3085 enum ofp_version version;
3086 struct ofport *port;
3087 struct list replies;
3088
3089 ofpmp_init(&replies, request);
3090
3091 version = ofputil_protocol_to_ofp_version(ofconn_get_protocol(ofconn));
3092 HMAP_FOR_EACH (port, hmap_node, &p->ports) {
3093 ofputil_append_port_desc_stats_reply(version, &port->pp, &replies);
3094 }
3095
3096 ofconn_send_replies(ofconn, &replies);
3097 return 0;
3098 }
3099
3100 static uint32_t
3101 hash_cookie(ovs_be64 cookie)
3102 {
3103 return hash_2words((OVS_FORCE uint64_t)cookie >> 32,
3104 (OVS_FORCE uint64_t)cookie);
3105 }
3106
3107 static void
3108 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3109 OVS_REQUIRES(ofproto_mutex)
3110 {
3111 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3112 hash_cookie(rule->flow_cookie));
3113 }
3114
3115 static void
3116 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3117 OVS_REQUIRES(ofproto_mutex)
3118 {
3119 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3120 }
3121
3122 static void
3123 ofproto_rule_change_cookie(struct ofproto *ofproto, struct rule *rule,
3124 ovs_be64 new_cookie)
3125 OVS_REQUIRES(ofproto_mutex)
3126 {
3127 if (new_cookie != rule->flow_cookie) {
3128 cookies_remove(ofproto, rule);
3129
3130 ovs_mutex_lock(&rule->mutex);
3131 rule->flow_cookie = new_cookie;
3132 ovs_mutex_unlock(&rule->mutex);
3133
3134 cookies_insert(ofproto, rule);
3135 }
3136 }
3137
3138 static void
3139 calc_duration(long long int start, long long int now,
3140 uint32_t *sec, uint32_t *nsec)
3141 {
3142 long long int msecs = now - start;
3143 *sec = msecs / 1000;
3144 *nsec = (msecs % 1000) * (1000 * 1000);
3145 }
3146
3147 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3148 * 0 if 'table_id' is OK, otherwise an OpenFlow error code. */
3149 static enum ofperr
3150 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3151 {
3152 return (table_id == 0xff || table_id < ofproto->n_tables
3153 ? 0
3154 : OFPERR_OFPBRC_BAD_TABLE_ID);
3155
3156 }
3157
3158 static struct oftable *
3159 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3160 {
3161 struct oftable *table;
3162
3163 for (table = &ofproto->tables[table_id];
3164 table < &ofproto->tables[ofproto->n_tables];
3165 table++) {
3166 if (!(table->flags & OFTABLE_HIDDEN)) {
3167 return table;
3168 }
3169 }
3170
3171 return NULL;
3172 }
3173
3174 static struct oftable *
3175 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3176 {
3177 if (table_id == 0xff) {
3178 return next_visible_table(ofproto, 0);
3179 } else if (table_id < ofproto->n_tables) {
3180 return &ofproto->tables[table_id];
3181 } else {
3182 return NULL;
3183 }
3184 }
3185
3186 static struct oftable *
3187 next_matching_table(const struct ofproto *ofproto,
3188 const struct oftable *table, uint8_t table_id)
3189 {
3190 return (table_id == 0xff
3191 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3192 : NULL);
3193 }
3194
3195 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3196 *
3197 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3198 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3199 *
3200 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3201 * only once, for that table. (This can be used to access tables marked
3202 * OFTABLE_HIDDEN.)
3203 *
3204 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3205 * entered at all. (Perhaps you should have validated TABLE_ID with
3206 * check_table_id().)
3207 *
3208 * All parameters are evaluated multiple times.
3209 */
3210 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3211 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3212 (TABLE) != NULL; \
3213 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3214
3215 /* Initializes 'criteria' in a straightforward way based on the other
3216 * parameters.
3217 *
3218 * For "loose" matching, the 'priority' parameter is unimportant and may be
3219 * supplied as 0. */
3220 static void
3221 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3222 const struct match *match, unsigned int priority,
3223 ovs_be64 cookie, ovs_be64 cookie_mask,
3224 ofp_port_t out_port, uint32_t out_group)
3225 {
3226 criteria->table_id = table_id;
3227 cls_rule_init(&criteria->cr, match, priority);
3228 criteria->cookie = cookie;
3229 criteria->cookie_mask = cookie_mask;
3230 criteria->out_port = out_port;
3231 criteria->out_group = out_group;
3232 }
3233
3234 static void
3235 rule_criteria_destroy(struct rule_criteria *criteria)
3236 {
3237 cls_rule_destroy(&criteria->cr);
3238 }
3239
3240 void
3241 rule_collection_init(struct rule_collection *rules)
3242 {
3243 rules->rules = rules->stub;
3244 rules->n = 0;
3245 rules->capacity = ARRAY_SIZE(rules->stub);
3246 }
3247
3248 void
3249 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3250 {
3251 if (rules->n >= rules->capacity) {
3252 size_t old_size, new_size;
3253
3254 old_size = rules->capacity * sizeof *rules->rules;
3255 rules->capacity *= 2;
3256 new_size = rules->capacity * sizeof *rules->rules;
3257
3258 if (rules->rules == rules->stub) {
3259 rules->rules = xmalloc(new_size);
3260 memcpy(rules->rules, rules->stub, old_size);
3261 } else {
3262 rules->rules = xrealloc(rules->rules, new_size);
3263 }
3264 }
3265
3266 rules->rules[rules->n++] = rule;
3267 }
3268
3269 void
3270 rule_collection_ref(struct rule_collection *rules)
3271 OVS_REQUIRES(ofproto_mutex)
3272 {
3273 size_t i;
3274
3275 for (i = 0; i < rules->n; i++) {
3276 ofproto_rule_ref(rules->rules[i]);
3277 }
3278 }
3279
3280 void
3281 rule_collection_unref(struct rule_collection *rules)
3282 {
3283 size_t i;
3284
3285 for (i = 0; i < rules->n; i++) {
3286 ofproto_rule_unref(rules->rules[i]);
3287 }
3288 }
3289
3290 void
3291 rule_collection_destroy(struct rule_collection *rules)
3292 {
3293 if (rules->rules != rules->stub) {
3294 free(rules->rules);
3295 }
3296 }
3297
3298 static enum ofperr
3299 collect_rule(struct rule *rule, const struct rule_criteria *c,
3300 struct rule_collection *rules)
3301 OVS_REQUIRES(ofproto_mutex)
3302 {
3303 /* We ordinarily want to skip hidden rules, but there has to be a way for
3304 * code internal to OVS to modify and delete them, so if the criteria
3305 * specify a priority that can only be for a hidden flow, then allow hidden
3306 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3307 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3308 * priority.) */
3309 if (ofproto_rule_is_hidden(rule) && c->cr.priority <= UINT16_MAX) {
3310 return 0;
3311 } else if (rule->pending) {
3312 return OFPROTO_POSTPONE;
3313 } else {
3314 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3315 && ofproto_rule_has_out_port(rule, c->out_port)
3316 && ofproto_rule_has_out_group(rule, c->out_group)
3317 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)) {
3318 rule_collection_add(rules, rule);
3319 }
3320 return 0;
3321 }
3322 }
3323
3324 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3325 * on classifiers rules are done in the "loose" way required for OpenFlow
3326 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3327 * 'rules'.
3328 *
3329 * Hidden rules are always omitted.
3330 *
3331 * Returns 0 on success, otherwise an OpenFlow error code. */
3332 static enum ofperr
3333 collect_rules_loose(struct ofproto *ofproto,
3334 const struct rule_criteria *criteria,
3335 struct rule_collection *rules)
3336 OVS_REQUIRES(ofproto_mutex)
3337 {
3338 struct oftable *table;
3339 enum ofperr error;
3340
3341 rule_collection_init(rules);
3342
3343 error = check_table_id(ofproto, criteria->table_id);
3344 if (error) {
3345 goto exit;
3346 }
3347
3348 if (criteria->cookie_mask == OVS_BE64_MAX) {
3349 struct rule *rule;
3350
3351 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3352 hash_cookie(criteria->cookie),
3353 &ofproto->cookies) {
3354 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3355 error = collect_rule(rule, criteria, rules);
3356 if (error) {
3357 break;
3358 }
3359 }
3360 }
3361 } else {
3362 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3363 struct cls_cursor cursor;
3364 struct rule *rule;
3365
3366 ovs_rwlock_rdlock(&table->cls.rwlock);
3367 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3368 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3369 error = collect_rule(rule, criteria, rules);
3370 if (error) {
3371 break;
3372 }
3373 }
3374 ovs_rwlock_unlock(&table->cls.rwlock);
3375 }
3376 }
3377
3378 exit:
3379 if (error) {
3380 rule_collection_destroy(rules);
3381 }
3382 return error;
3383 }
3384
3385 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3386 * on classifiers rules are done in the "strict" way required for OpenFlow
3387 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3388 * rules on list 'rules'.
3389 *
3390 * Hidden rules are always omitted.
3391 *
3392 * Returns 0 on success, otherwise an OpenFlow error code. */
3393 static enum ofperr
3394 collect_rules_strict(struct ofproto *ofproto,
3395 const struct rule_criteria *criteria,
3396 struct rule_collection *rules)
3397 OVS_REQUIRES(ofproto_mutex)
3398 {
3399 struct oftable *table;
3400 int error;
3401
3402 rule_collection_init(rules);
3403
3404 error = check_table_id(ofproto, criteria->table_id);
3405 if (error) {
3406 goto exit;
3407 }
3408
3409 if (criteria->cookie_mask == OVS_BE64_MAX) {
3410 struct rule *rule;
3411
3412 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3413 hash_cookie(criteria->cookie),
3414 &ofproto->cookies) {
3415 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3416 error = collect_rule(rule, criteria, rules);
3417 if (error) {
3418 break;
3419 }
3420 }
3421 }
3422 } else {
3423 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3424 struct rule *rule;
3425
3426 ovs_rwlock_rdlock(&table->cls.rwlock);
3427 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3428 &table->cls, &criteria->cr));
3429 ovs_rwlock_unlock(&table->cls.rwlock);
3430 if (rule) {
3431 error = collect_rule(rule, criteria, rules);
3432 if (error) {
3433 break;
3434 }
3435 }
3436 }
3437 }
3438
3439 exit:
3440 if (error) {
3441 rule_collection_destroy(rules);
3442 }
3443 return error;
3444 }
3445
3446 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3447 * forced into the range of a uint16_t. */
3448 static int
3449 age_secs(long long int age_ms)
3450 {
3451 return (age_ms < 0 ? 0
3452 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3453 : (unsigned int) age_ms / 1000);
3454 }
3455
3456 static enum ofperr
3457 handle_flow_stats_request(struct ofconn *ofconn,
3458 const struct ofp_header *request)
3459 OVS_EXCLUDED(ofproto_mutex)
3460 {
3461 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3462 struct ofputil_flow_stats_request fsr;
3463 struct rule_criteria criteria;
3464 struct rule_collection rules;
3465 struct list replies;
3466 enum ofperr error;
3467 size_t i;
3468
3469 error = ofputil_decode_flow_stats_request(&fsr, request);
3470 if (error) {
3471 return error;
3472 }
3473
3474 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3475 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3476
3477 ovs_mutex_lock(&ofproto_mutex);
3478 error = collect_rules_loose(ofproto, &criteria, &rules);
3479 rule_criteria_destroy(&criteria);
3480 if (!error) {
3481 rule_collection_ref(&rules);
3482 }
3483 ovs_mutex_unlock(&ofproto_mutex);
3484
3485 if (error) {
3486 return error;
3487 }
3488
3489 ofpmp_init(&replies, request);
3490 for (i = 0; i < rules.n; i++) {
3491 struct rule *rule = rules.rules[i];
3492 long long int now = time_msec();
3493 struct ofputil_flow_stats fs;
3494 long long int created, used, modified;
3495 struct rule_actions *actions;
3496 enum ofputil_flow_mod_flags flags;
3497
3498 ovs_mutex_lock(&rule->mutex);
3499 fs.cookie = rule->flow_cookie;
3500 fs.idle_timeout = rule->idle_timeout;
3501 fs.hard_timeout = rule->hard_timeout;
3502 created = rule->created;
3503 used = rule->used;
3504 modified = rule->modified;
3505 actions = rule_get_actions__(rule);
3506 flags = rule->flags;
3507 ovs_mutex_unlock(&rule->mutex);
3508
3509 minimatch_expand(&rule->cr.match, &fs.match);
3510 fs.table_id = rule->table_id;
3511 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3512 fs.priority = rule->cr.priority;
3513 fs.idle_age = age_secs(now - used);
3514 fs.hard_age = age_secs(now - modified);
3515 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3516 &fs.byte_count);
3517 fs.ofpacts = actions->ofpacts;
3518 fs.ofpacts_len = actions->ofpacts_len;
3519
3520 fs.flags = flags;
3521 ofputil_append_flow_stats_reply(&fs, &replies);
3522
3523 rule_actions_unref(actions);
3524 }
3525
3526 rule_collection_unref(&rules);
3527 rule_collection_destroy(&rules);
3528
3529 ofconn_send_replies(ofconn, &replies);
3530
3531 return 0;
3532 }
3533
3534 static void
3535 flow_stats_ds(struct rule *rule, struct ds *results)
3536 {
3537 uint64_t packet_count, byte_count;
3538 struct rule_actions *actions;
3539 long long int created;
3540
3541 rule->ofproto->ofproto_class->rule_get_stats(rule,
3542 &packet_count, &byte_count);
3543
3544 ovs_mutex_lock(&rule->mutex);
3545 actions = rule_get_actions__(rule);
3546 created = rule->created;
3547 ovs_mutex_unlock(&rule->mutex);
3548
3549 if (rule->table_id != 0) {
3550 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3551 }
3552 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3553 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3554 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3555 cls_rule_format(&rule->cr, results);
3556 ds_put_char(results, ',');
3557
3558 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3559
3560 ds_put_cstr(results, "\n");
3561
3562 rule_actions_unref(actions);
3563 }
3564
3565 /* Adds a pretty-printed description of all flows to 'results', including
3566 * hidden flows (e.g., set up by in-band control). */
3567 void
3568 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3569 {
3570 struct oftable *table;
3571
3572 OFPROTO_FOR_EACH_TABLE (table, p) {
3573 struct cls_cursor cursor;
3574 struct rule *rule;
3575
3576 ovs_rwlock_rdlock(&table->cls.rwlock);
3577 cls_cursor_init(&cursor, &table->cls, NULL);
3578 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3579 flow_stats_ds(rule, results);
3580 }
3581 ovs_rwlock_unlock(&table->cls.rwlock);
3582 }
3583 }
3584
3585 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3586 * '*engine_type' and '*engine_id', respectively. */
3587 void
3588 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3589 uint8_t *engine_type, uint8_t *engine_id)
3590 {
3591 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3592 }
3593
3594 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'. Returns
3595 * true if the port's CFM status was successfully stored into '*status'.
3596 * Returns false if the port did not have CFM configured, in which case
3597 * '*status' is indeterminate.
3598 *
3599 * The caller must provide and owns '*status', and must free 'status->rmps'. */
3600 bool
3601 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3602 struct ofproto_cfm_status *status)
3603 {
3604 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3605 return (ofport
3606 && ofproto->ofproto_class->get_cfm_status
3607 && ofproto->ofproto_class->get_cfm_status(ofport, status));
3608 }
3609
3610 static enum ofperr
3611 handle_aggregate_stats_request(struct ofconn *ofconn,
3612 const struct ofp_header *oh)
3613 OVS_EXCLUDED(ofproto_mutex)
3614 {
3615 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3616 struct ofputil_flow_stats_request request;
3617 struct ofputil_aggregate_stats stats;
3618 bool unknown_packets, unknown_bytes;
3619 struct rule_criteria criteria;
3620 struct rule_collection rules;
3621 struct ofpbuf *reply;
3622 enum ofperr error;
3623 size_t i;
3624
3625 error = ofputil_decode_flow_stats_request(&request, oh);
3626 if (error) {
3627 return error;
3628 }
3629
3630 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3631 request.cookie, request.cookie_mask,
3632 request.out_port, request.out_group);
3633
3634 ovs_mutex_lock(&ofproto_mutex);
3635 error = collect_rules_loose(ofproto, &criteria, &rules);
3636 rule_criteria_destroy(&criteria);
3637 if (!error) {
3638 rule_collection_ref(&rules);
3639 }
3640 ovs_mutex_unlock(&ofproto_mutex);
3641
3642 if (error) {
3643 return error;
3644 }
3645
3646 memset(&stats, 0, sizeof stats);
3647 unknown_packets = unknown_bytes = false;
3648 for (i = 0; i < rules.n; i++) {
3649 struct rule *rule = rules.rules[i];
3650 uint64_t packet_count;
3651 uint64_t byte_count;
3652
3653 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3654 &byte_count);
3655
3656 if (packet_count == UINT64_MAX) {
3657 unknown_packets = true;
3658 } else {
3659 stats.packet_count += packet_count;
3660 }
3661
3662 if (byte_count == UINT64_MAX) {
3663 unknown_bytes = true;
3664 } else {
3665 stats.byte_count += byte_count;
3666 }
3667
3668 stats.flow_count++;
3669 }
3670 if (unknown_packets) {
3671 stats.packet_count = UINT64_MAX;
3672 }
3673 if (unknown_bytes) {
3674 stats.byte_count = UINT64_MAX;
3675 }
3676
3677 rule_collection_unref(&rules);
3678 rule_collection_destroy(&rules);
3679
3680 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3681 ofconn_send_reply(ofconn, reply);
3682
3683 return 0;
3684 }
3685
3686 struct queue_stats_cbdata {
3687 struct ofport *ofport;
3688 struct list replies;
3689 long long int now;
3690 };
3691
3692 static void
3693 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3694 const struct netdev_queue_stats *stats)
3695 {
3696 struct ofputil_queue_stats oqs;
3697
3698 oqs.port_no = cbdata->ofport->pp.port_no;
3699 oqs.queue_id = queue_id;
3700 oqs.tx_bytes = stats->tx_bytes;
3701 oqs.tx_packets = stats->tx_packets;
3702 oqs.tx_errors = stats->tx_errors;
3703 if (stats->created != LLONG_MIN) {
3704 calc_duration(stats->created, cbdata->now,
3705 &oqs.duration_sec, &oqs.duration_nsec);
3706 } else {
3707 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3708 }
3709 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3710 }
3711
3712 static void
3713 handle_queue_stats_dump_cb(uint32_t queue_id,
3714 struct netdev_queue_stats *stats,
3715 void *cbdata_)
3716 {
3717 struct queue_stats_cbdata *cbdata = cbdata_;
3718
3719 put_queue_stats(cbdata, queue_id, stats);
3720 }
3721
3722 static enum ofperr
3723 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3724 struct queue_stats_cbdata *cbdata)
3725 {
3726 cbdata->ofport = port;
3727 if (queue_id == OFPQ_ALL) {
3728 netdev_dump_queue_stats(port->netdev,
3729 handle_queue_stats_dump_cb, cbdata);
3730 } else {
3731 struct netdev_queue_stats stats;
3732
3733 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3734 put_queue_stats(cbdata, queue_id, &stats);
3735 } else {
3736 return OFPERR_OFPQOFC_BAD_QUEUE;
3737 }
3738 }
3739 return 0;
3740 }
3741
3742 static enum ofperr
3743 handle_queue_stats_request(struct ofconn *ofconn,
3744 const struct ofp_header *rq)
3745 {
3746 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3747 struct queue_stats_cbdata cbdata;
3748 struct ofport *port;
3749 enum ofperr error;
3750 struct ofputil_queue_stats_request oqsr;
3751
3752 COVERAGE_INC(ofproto_queue_req);
3753
3754 ofpmp_init(&cbdata.replies, rq);
3755 cbdata.now = time_msec();
3756
3757 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3758 if (error) {
3759 return error;
3760 }
3761
3762 if (oqsr.port_no == OFPP_ANY) {
3763 error = OFPERR_OFPQOFC_BAD_QUEUE;
3764 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3765 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3766 error = 0;
3767 }
3768 }
3769 } else {
3770 port = ofproto_get_port(ofproto, oqsr.port_no);
3771 error = (port
3772 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3773 : OFPERR_OFPQOFC_BAD_PORT);
3774 }
3775 if (!error) {
3776 ofconn_send_replies(ofconn, &cbdata.replies);
3777 } else {
3778 ofpbuf_list_delete(&cbdata.replies);
3779 }
3780
3781 return error;
3782 }
3783
3784 static bool
3785 is_flow_deletion_pending(const struct ofproto *ofproto,
3786 const struct cls_rule *cls_rule,
3787 uint8_t table_id)
3788 OVS_REQUIRES(ofproto_mutex)
3789 {
3790 if (!hmap_is_empty(&ofproto->deletions)) {
3791 struct ofoperation *op;
3792
3793 HMAP_FOR_EACH_WITH_HASH (op, hmap_node,
3794 cls_rule_hash(cls_rule, table_id),
3795 &ofproto->deletions) {
3796 if (cls_rule_equal(cls_rule, &op->rule->cr)) {
3797 return true;
3798 }
3799 }
3800 }
3801
3802 return false;
3803 }
3804
3805 static bool
3806 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3807 OVS_REQUIRES(ofproto_mutex)
3808 OVS_NO_THREAD_SAFETY_ANALYSIS
3809 {
3810 return classifier_count(&table->cls) + extra_space > table->max_flows;
3811 }
3812
3813 static enum ofperr
3814 evict_rules_from_table(struct ofproto *ofproto, struct oftable *table,
3815 unsigned int extra_space)
3816 OVS_REQUIRES(ofproto_mutex)
3817 {
3818 while (should_evict_a_rule(table, extra_space)) {
3819 struct rule *rule;
3820
3821 if (!choose_rule_to_evict(table, &rule)) {
3822 return OFPERR_OFPFMFC_TABLE_FULL;
3823 } else if (rule->pending) {
3824 return OFPROTO_POSTPONE;
3825 } else {
3826 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
3827 delete_flow__(rule, group, OFPRR_EVICTION);
3828 ofopgroup_submit(group);
3829 }
3830 }
3831
3832 return 0;
3833 }
3834
3835 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3836 * in which no matching flow already exists in the flow table.
3837 *
3838 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3839 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3840 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3841 * initiated now but may be retried later.
3842 *
3843 * The caller retains ownership of 'fm->ofpacts'.
3844 *
3845 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3846 * if any. */
3847 static enum ofperr
3848 add_flow(struct ofproto *ofproto, struct ofconn *ofconn,
3849 struct ofputil_flow_mod *fm, const struct ofp_header *request)
3850 OVS_REQUIRES(ofproto_mutex)
3851 {
3852 struct oftable *table;
3853 struct ofopgroup *group;
3854 struct cls_rule cr;
3855 struct rule *rule;
3856 uint8_t table_id;
3857 int error;
3858
3859 error = check_table_id(ofproto, fm->table_id);
3860 if (error) {
3861 return error;
3862 }
3863
3864 /* Pick table. */
3865 if (fm->table_id == 0xff) {
3866 if (ofproto->ofproto_class->rule_choose_table) {
3867 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3868 &fm->match,
3869 &table_id);
3870 if (error) {
3871 return error;
3872 }
3873 ovs_assert(table_id < ofproto->n_tables);
3874 } else {
3875 table_id = 0;
3876 }
3877 } else if (fm->table_id < ofproto->n_tables) {
3878 table_id = fm->table_id;
3879 } else {
3880 return OFPERR_OFPBRC_BAD_TABLE_ID;
3881 }
3882
3883 table = &ofproto->tables[table_id];
3884
3885 if (table->flags & OFTABLE_READONLY) {
3886 return OFPERR_OFPBRC_EPERM;
3887 }
3888
3889 cls_rule_init(&cr, &fm->match, fm->priority);
3890
3891 /* Transform "add" into "modify" if there's an existing identical flow. */
3892 ovs_rwlock_rdlock(&table->cls.rwlock);
3893 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3894 ovs_rwlock_unlock(&table->cls.rwlock);
3895 if (rule) {
3896 cls_rule_destroy(&cr);
3897 if (!rule_is_modifiable(rule)) {
3898 return OFPERR_OFPBRC_EPERM;
3899 } else if (rule->pending) {
3900 return OFPROTO_POSTPONE;
3901 } else {
3902 struct rule_collection rules;
3903
3904 rule_collection_init(&rules);
3905 rule_collection_add(&rules, rule);
3906 fm->modify_cookie = true;
3907 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
3908 rule_collection_destroy(&rules);
3909
3910 return error;
3911 }
3912 }
3913
3914 /* Verify actions. */
3915 error = ofproto_check_ofpacts(ofproto, fm->ofpacts, fm->ofpacts_len,
3916 &fm->match.flow, table_id);
3917 if (error) {
3918 cls_rule_destroy(&cr);
3919 return error;
3920 }
3921
3922 /* Serialize against pending deletion. */
3923 if (is_flow_deletion_pending(ofproto, &cr, table_id)) {
3924 cls_rule_destroy(&cr);
3925 return OFPROTO_POSTPONE;
3926 }
3927
3928 /* Check for overlap, if requested. */
3929 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3930 bool overlaps;
3931
3932 ovs_rwlock_rdlock(&table->cls.rwlock);
3933 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3934 ovs_rwlock_unlock(&table->cls.rwlock);
3935
3936 if (overlaps) {
3937 cls_rule_destroy(&cr);
3938 return OFPERR_OFPFMFC_OVERLAP;
3939 }
3940 }
3941
3942 /* If necessary, evict an existing rule to clear out space. */
3943 error = evict_rules_from_table(ofproto, table, 1);
3944 if (error) {
3945 cls_rule_destroy(&cr);
3946 return error;
3947 }
3948
3949 /* Allocate new rule. */
3950 rule = ofproto->ofproto_class->rule_alloc();
3951 if (!rule) {
3952 cls_rule_destroy(&cr);
3953 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
3954 ofproto->name, ovs_strerror(error));
3955 return ENOMEM;
3956 }
3957
3958 /* Initialize base state. */
3959 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
3960 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
3961 atomic_init(&rule->ref_count, 1);
3962 rule->pending = NULL;
3963 rule->flow_cookie = fm->new_cookie;
3964 rule->created = rule->modified = rule->used = time_msec();
3965
3966 ovs_mutex_init(&rule->mutex);
3967 ovs_mutex_lock(&rule->mutex);
3968 rule->idle_timeout = fm->idle_timeout;
3969 rule->hard_timeout = fm->hard_timeout;
3970 ovs_mutex_unlock(&rule->mutex);
3971
3972 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
3973 rule->flags = fm->flags & OFPUTIL_FF_STATE;
3974 rule->actions = rule_actions_create(ofproto, fm->ofpacts, fm->ofpacts_len);
3975 list_init(&rule->meter_list_node);
3976 rule->eviction_group = NULL;
3977 list_init(&rule->expirable);
3978 rule->monitor_flags = 0;
3979 rule->add_seqno = 0;
3980 rule->modify_seqno = 0;
3981
3982 /* Construct rule, initializing derived state. */
3983 error = ofproto->ofproto_class->rule_construct(rule);
3984 if (error) {
3985 ofproto_rule_destroy__(rule);
3986 return error;
3987 }
3988
3989 /* Insert rule. */
3990 oftable_insert_rule(rule);
3991
3992 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
3993 ofoperation_create(group, rule, OFOPERATION_ADD, 0);
3994 ofproto->ofproto_class->rule_insert(rule);
3995 ofopgroup_submit(group);
3996
3997 return error;
3998 }
3999 \f
4000 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4001
4002 /* Modifies the rules listed in 'rules', changing their actions to match those
4003 * in 'fm'.
4004 *
4005 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4006 * if any.
4007 *
4008 * Returns 0 on success, otherwise an OpenFlow error code. */
4009 static enum ofperr
4010 modify_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4011 struct ofputil_flow_mod *fm, const struct ofp_header *request,
4012 const struct rule_collection *rules)
4013 OVS_REQUIRES(ofproto_mutex)
4014 {
4015 enum ofoperation_type type;
4016 struct ofopgroup *group;
4017 enum ofperr error;
4018 size_t i;
4019
4020 type = fm->command == OFPFC_ADD ? OFOPERATION_REPLACE : OFOPERATION_MODIFY;
4021 group = ofopgroup_create(ofproto, ofconn, request, fm->buffer_id);
4022 error = OFPERR_OFPBRC_EPERM;
4023 for (i = 0; i < rules->n; i++) {
4024 struct rule *rule = rules->rules[i];
4025 struct ofoperation *op;
4026 bool actions_changed;
4027 bool reset_counters;
4028
4029 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
4030
4031 if (rule_is_modifiable(rule)) {
4032 /* At least one rule is modifiable, don't report EPERM error. */
4033 error = 0;
4034 } else {
4035 continue;
4036 }
4037
4038 /* Verify actions. */
4039 error = ofpacts_check(fm->ofpacts, fm->ofpacts_len, &fm->match.flow,
4040 u16_to_ofp(ofproto->max_ports), rule->table_id);
4041 if (error) {
4042 return error;
4043 }
4044
4045 actions_changed = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
4046 rule->actions->ofpacts,
4047 rule->actions->ofpacts_len);
4048
4049 op = ofoperation_create(group, rule, type, 0);
4050
4051 if (fm->modify_cookie && fm->new_cookie != OVS_BE64_MAX) {
4052 ofproto_rule_change_cookie(ofproto, rule, fm->new_cookie);
4053 }
4054 if (type == OFOPERATION_REPLACE) {
4055 ovs_mutex_lock(&rule->mutex);
4056 rule->idle_timeout = fm->idle_timeout;
4057 rule->hard_timeout = fm->hard_timeout;
4058 ovs_mutex_unlock(&rule->mutex);
4059
4060 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4061 if (fm->idle_timeout || fm->hard_timeout) {
4062 if (!rule->eviction_group) {
4063 eviction_group_add_rule(rule);
4064 }
4065 } else {
4066 eviction_group_remove_rule(rule);
4067 }
4068 }
4069
4070 reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4071 if (actions_changed || reset_counters) {
4072 struct rule_actions *new_actions;
4073
4074 op->actions = rule->actions;
4075 new_actions = rule_actions_create(ofproto,
4076 fm->ofpacts, fm->ofpacts_len);
4077
4078 ovs_mutex_lock(&rule->mutex);
4079 rule->actions = new_actions;
4080 ovs_mutex_unlock(&rule->mutex);
4081
4082 rule->ofproto->ofproto_class->rule_modify_actions(rule,
4083 reset_counters);
4084 } else {
4085 ofoperation_complete(op, 0);
4086 }
4087 }
4088 ofopgroup_submit(group);
4089
4090 return error;
4091 }
4092
4093 static enum ofperr
4094 modify_flows_add(struct ofproto *ofproto, struct ofconn *ofconn,
4095 struct ofputil_flow_mod *fm, const struct ofp_header *request)
4096 OVS_REQUIRES(ofproto_mutex)
4097 {
4098 if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) {
4099 return 0;
4100 }
4101 return add_flow(ofproto, ofconn, fm, request);
4102 }
4103
4104 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4105 * failure.
4106 *
4107 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4108 * if any. */
4109 static enum ofperr
4110 modify_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4111 struct ofputil_flow_mod *fm,
4112 const struct ofp_header *request)
4113 OVS_REQUIRES(ofproto_mutex)
4114 {
4115 struct rule_criteria criteria;
4116 struct rule_collection rules;
4117 int error;
4118
4119 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4120 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4121 error = collect_rules_loose(ofproto, &criteria, &rules);
4122 rule_criteria_destroy(&criteria);
4123
4124 if (!error) {
4125 error = (rules.n > 0
4126 ? modify_flows__(ofproto, ofconn, fm, request, &rules)
4127 : modify_flows_add(ofproto, ofconn, fm, request));
4128 }
4129
4130 rule_collection_destroy(&rules);
4131
4132 return error;
4133 }
4134
4135 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4136 * code on failure.
4137 *
4138 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4139 * if any. */
4140 static enum ofperr
4141 modify_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4142 struct ofputil_flow_mod *fm,
4143 const struct ofp_header *request)
4144 OVS_REQUIRES(ofproto_mutex)
4145 {
4146 struct rule_criteria criteria;
4147 struct rule_collection rules;
4148 int error;
4149
4150 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4151 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4152 error = collect_rules_strict(ofproto, &criteria, &rules);
4153 rule_criteria_destroy(&criteria);
4154
4155 if (!error) {
4156 if (rules.n == 0) {
4157 error = modify_flows_add(ofproto, ofconn, fm, request);
4158 } else if (rules.n == 1) {
4159 error = modify_flows__(ofproto, ofconn, fm, request, &rules);
4160 }
4161 }
4162
4163 rule_collection_destroy(&rules);
4164
4165 return error;
4166 }
4167 \f
4168 /* OFPFC_DELETE implementation. */
4169
4170 static void
4171 delete_flow__(struct rule *rule, struct ofopgroup *group,
4172 enum ofp_flow_removed_reason reason)
4173 OVS_REQUIRES(ofproto_mutex)
4174 {
4175 struct ofproto *ofproto = rule->ofproto;
4176
4177 ofproto_rule_send_removed(rule, reason);
4178
4179 ofoperation_create(group, rule, OFOPERATION_DELETE, reason);
4180 oftable_remove_rule(rule);
4181 ofproto->ofproto_class->rule_delete(rule);
4182 }
4183
4184 /* Deletes the rules listed in 'rules'.
4185 *
4186 * Returns 0 on success, otherwise an OpenFlow error code. */
4187 static enum ofperr
4188 delete_flows__(struct ofproto *ofproto, struct ofconn *ofconn,
4189 const struct ofp_header *request,
4190 const struct rule_collection *rules,
4191 enum ofp_flow_removed_reason reason)
4192 OVS_REQUIRES(ofproto_mutex)
4193 {
4194 struct ofopgroup *group;
4195 size_t i;
4196
4197 group = ofopgroup_create(ofproto, ofconn, request, UINT32_MAX);
4198 for (i = 0; i < rules->n; i++) {
4199 delete_flow__(rules->rules[i], group, reason);
4200 }
4201 ofopgroup_submit(group);
4202
4203 return 0;
4204 }
4205
4206 /* Implements OFPFC_DELETE. */
4207 static enum ofperr
4208 delete_flows_loose(struct ofproto *ofproto, struct ofconn *ofconn,
4209 const struct ofputil_flow_mod *fm,
4210 const struct ofp_header *request)
4211 OVS_REQUIRES(ofproto_mutex)
4212 {
4213 struct rule_criteria criteria;
4214 struct rule_collection rules;
4215 enum ofperr error;
4216
4217 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4218 fm->cookie, fm->cookie_mask,
4219 fm->out_port, fm->out_group);
4220 error = collect_rules_loose(ofproto, &criteria, &rules);
4221 rule_criteria_destroy(&criteria);
4222
4223 if (!error && rules.n > 0) {
4224 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4225 }
4226 rule_collection_destroy(&rules);
4227
4228 return error;
4229 }
4230
4231 /* Implements OFPFC_DELETE_STRICT. */
4232 static enum ofperr
4233 delete_flow_strict(struct ofproto *ofproto, struct ofconn *ofconn,
4234 const struct ofputil_flow_mod *fm,
4235 const struct ofp_header *request)
4236 OVS_REQUIRES(ofproto_mutex)
4237 {
4238 struct rule_criteria criteria;
4239 struct rule_collection rules;
4240 enum ofperr error;
4241
4242 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4243 fm->cookie, fm->cookie_mask,
4244 fm->out_port, fm->out_group);
4245 error = collect_rules_strict(ofproto, &criteria, &rules);
4246 rule_criteria_destroy(&criteria);
4247
4248 if (!error && rules.n > 0) {
4249 error = delete_flows__(ofproto, ofconn, request, &rules, OFPRR_DELETE);
4250 }
4251 rule_collection_destroy(&rules);
4252
4253 return error;
4254 }
4255
4256 static void
4257 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4258 OVS_REQUIRES(ofproto_mutex)
4259 {
4260 struct ofputil_flow_removed fr;
4261
4262 if (ofproto_rule_is_hidden(rule) ||
4263 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4264 return;
4265 }
4266
4267 minimatch_expand(&rule->cr.match, &fr.match);
4268 fr.priority = rule->cr.priority;
4269 fr.cookie = rule->flow_cookie;
4270 fr.reason = reason;
4271 fr.table_id = rule->table_id;
4272 calc_duration(rule->created, time_msec(),
4273 &fr.duration_sec, &fr.duration_nsec);
4274 ovs_mutex_lock(&rule->mutex);
4275 fr.idle_timeout = rule->idle_timeout;
4276 fr.hard_timeout = rule->hard_timeout;
4277 ovs_mutex_unlock(&rule->mutex);
4278 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4279 &fr.byte_count);
4280
4281 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4282 }
4283
4284 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4285 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4286 * ofproto.
4287 *
4288 * 'rule' must not have a pending operation (that is, 'rule->pending' must be
4289 * NULL).
4290 *
4291 * ofproto implementation ->run() functions should use this function to expire
4292 * OpenFlow flows. */
4293 void
4294 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4295 OVS_REQUIRES(ofproto_mutex)
4296 {
4297 struct ofproto *ofproto = rule->ofproto;
4298
4299 ovs_assert(reason == OFPRR_HARD_TIMEOUT || reason == OFPRR_IDLE_TIMEOUT
4300 || reason == OFPRR_DELETE || reason == OFPRR_GROUP_DELETE);
4301
4302 ofproto_rule_send_removed(rule, reason);
4303 ofproto_rule_delete__(ofproto, rule);
4304 }
4305
4306 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4307 * means "infinite". */
4308 static void
4309 reduce_timeout(uint16_t max, uint16_t *timeout)
4310 {
4311 if (max && (!*timeout || *timeout > max)) {
4312 *timeout = max;
4313 }
4314 }
4315
4316 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4317 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4318 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4319 *
4320 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4321 void
4322 ofproto_rule_reduce_timeouts(struct rule *rule,
4323 uint16_t idle_timeout, uint16_t hard_timeout)
4324 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4325 {
4326 if (!idle_timeout && !hard_timeout) {
4327 return;
4328 }
4329
4330 ovs_mutex_lock(&ofproto_mutex);
4331 if (list_is_empty(&rule->expirable)) {
4332 list_insert(&rule->ofproto->expirable, &rule->expirable);
4333 }
4334 ovs_mutex_unlock(&ofproto_mutex);
4335
4336 ovs_mutex_lock(&rule->mutex);
4337 reduce_timeout(idle_timeout, &rule->idle_timeout);
4338 reduce_timeout(hard_timeout, &rule->hard_timeout);
4339 ovs_mutex_unlock(&rule->mutex);
4340 }
4341 \f
4342 static enum ofperr
4343 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4344 OVS_EXCLUDED(ofproto_mutex)
4345 {
4346 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4347 struct ofputil_flow_mod fm;
4348 uint64_t ofpacts_stub[1024 / 8];
4349 struct ofpbuf ofpacts;
4350 enum ofperr error;
4351 long long int now;
4352
4353 error = reject_slave_controller(ofconn);
4354 if (error) {
4355 goto exit;
4356 }
4357
4358 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4359 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4360 &ofpacts);
4361 if (!error) {
4362 error = handle_flow_mod__(ofproto, ofconn, &fm, oh);
4363 }
4364 if (error) {
4365 goto exit_free_ofpacts;
4366 }
4367
4368 /* Record the operation for logging a summary report. */
4369 switch (fm.command) {
4370 case OFPFC_ADD:
4371 ofproto->n_add++;
4372 break;
4373
4374 case OFPFC_MODIFY:
4375 case OFPFC_MODIFY_STRICT:
4376 ofproto->n_modify++;
4377 break;
4378
4379 case OFPFC_DELETE:
4380 case OFPFC_DELETE_STRICT:
4381 ofproto->n_delete++;
4382 break;
4383 }
4384
4385 now = time_msec();
4386 if (ofproto->next_op_report == LLONG_MAX) {
4387 ofproto->first_op = now;
4388 ofproto->next_op_report = MAX(now + 10 * 1000,
4389 ofproto->op_backoff);
4390 ofproto->op_backoff = ofproto->next_op_report + 60 * 1000;
4391 }
4392 ofproto->last_op = now;
4393
4394 exit_free_ofpacts:
4395 ofpbuf_uninit(&ofpacts);
4396 exit:
4397 return error;
4398 }
4399
4400 static enum ofperr
4401 handle_flow_mod__(struct ofproto *ofproto, struct ofconn *ofconn,
4402 struct ofputil_flow_mod *fm, const struct ofp_header *oh)
4403 OVS_EXCLUDED(ofproto_mutex)
4404 {
4405 enum ofperr error;
4406
4407 ovs_mutex_lock(&ofproto_mutex);
4408 if (ofproto->n_pending < 50) {
4409 switch (fm->command) {
4410 case OFPFC_ADD:
4411 error = add_flow(ofproto, ofconn, fm, oh);
4412 break;
4413
4414 case OFPFC_MODIFY:
4415 error = modify_flows_loose(ofproto, ofconn, fm, oh);
4416 break;
4417
4418 case OFPFC_MODIFY_STRICT:
4419 error = modify_flow_strict(ofproto, ofconn, fm, oh);
4420 break;
4421
4422 case OFPFC_DELETE:
4423 error = delete_flows_loose(ofproto, ofconn, fm, oh);
4424 break;
4425
4426 case OFPFC_DELETE_STRICT:
4427 error = delete_flow_strict(ofproto, ofconn, fm, oh);
4428 break;
4429
4430 default:
4431 if (fm->command > 0xff) {
4432 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4433 "flow_mod_table_id extension is not enabled",
4434 ofproto->name);
4435 }
4436 error = OFPERR_OFPFMFC_BAD_COMMAND;
4437 break;
4438 }
4439 } else {
4440 ovs_assert(!list_is_empty(&ofproto->pending));
4441 error = OFPROTO_POSTPONE;
4442 }
4443 ovs_mutex_unlock(&ofproto_mutex);
4444
4445 run_rule_executes(ofproto);
4446 return error;
4447 }
4448
4449 static enum ofperr
4450 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4451 {
4452 struct ofputil_role_request request;
4453 struct ofputil_role_request reply;
4454 struct ofpbuf *buf;
4455 enum ofperr error;
4456
4457 error = ofputil_decode_role_message(oh, &request);
4458 if (error) {
4459 return error;
4460 }
4461
4462 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4463 if (ofconn_get_role(ofconn) != request.role
4464 && ofconn_has_pending_opgroups(ofconn)) {
4465 return OFPROTO_POSTPONE;
4466 }
4467
4468 if (request.have_generation_id
4469 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4470 return OFPERR_OFPRRFC_STALE;
4471 }
4472
4473 ofconn_set_role(ofconn, request.role);
4474 }
4475
4476 reply.role = ofconn_get_role(ofconn);
4477 reply.have_generation_id = ofconn_get_master_election_id(
4478 ofconn, &reply.generation_id);
4479 buf = ofputil_encode_role_reply(oh, &reply);
4480 ofconn_send_reply(ofconn, buf);
4481
4482 return 0;
4483 }
4484
4485 static enum ofperr
4486 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4487 const struct ofp_header *oh)
4488 {
4489 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4490 enum ofputil_protocol cur, next;
4491
4492 cur = ofconn_get_protocol(ofconn);
4493 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4494 ofconn_set_protocol(ofconn, next);
4495
4496 return 0;
4497 }
4498
4499 static enum ofperr
4500 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4501 {
4502 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4503 enum ofputil_protocol cur, next;
4504 enum ofputil_protocol next_base;
4505
4506 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4507 if (!next_base) {
4508 return OFPERR_OFPBRC_EPERM;
4509 }
4510
4511 cur = ofconn_get_protocol(ofconn);
4512 next = ofputil_protocol_set_base(cur, next_base);
4513 if (cur != next && ofconn_has_pending_opgroups(ofconn)) {
4514 /* Avoid sending async messages in surprising protocol. */
4515 return OFPROTO_POSTPONE;
4516 }
4517
4518 ofconn_set_protocol(ofconn, next);
4519 return 0;
4520 }
4521
4522 static enum ofperr
4523 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4524 const struct ofp_header *oh)
4525 {
4526 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4527 uint32_t format;
4528
4529 format = ntohl(msg->format);
4530 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4531 return OFPERR_OFPBRC_EPERM;
4532 }
4533
4534 if (format != ofconn_get_packet_in_format(ofconn)
4535 && ofconn_has_pending_opgroups(ofconn)) {
4536 /* Avoid sending async message in surprsing packet in format. */
4537 return OFPROTO_POSTPONE;
4538 }
4539
4540 ofconn_set_packet_in_format(ofconn, format);
4541 return 0;
4542 }
4543
4544 static enum ofperr
4545 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4546 {
4547 const struct nx_async_config *msg = ofpmsg_body(oh);
4548 uint32_t master[OAM_N_TYPES];
4549 uint32_t slave[OAM_N_TYPES];
4550
4551 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4552 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4553 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4554
4555 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4556 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4557 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4558
4559 ofconn_set_async_config(ofconn, master, slave);
4560 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4561 !ofconn_get_miss_send_len(ofconn)) {
4562 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4563 }
4564
4565 return 0;
4566 }
4567
4568 static enum ofperr
4569 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4570 {
4571 struct ofpbuf *buf;
4572 uint32_t master[OAM_N_TYPES];
4573 uint32_t slave[OAM_N_TYPES];
4574 struct nx_async_config *msg;
4575
4576 ofconn_get_async_config(ofconn, master, slave);
4577 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4578 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4579
4580 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4581 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4582 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4583
4584 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4585 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4586 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4587
4588 ofconn_send_reply(ofconn, buf);
4589
4590 return 0;
4591 }
4592
4593 static enum ofperr
4594 handle_nxt_set_controller_id(struct ofconn *ofconn,
4595 const struct ofp_header *oh)
4596 {
4597 const struct nx_controller_id *nci = ofpmsg_body(oh);
4598
4599 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4600 return OFPERR_NXBRC_MUST_BE_ZERO;
4601 }
4602
4603 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4604 return 0;
4605 }
4606
4607 static enum ofperr
4608 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4609 {
4610 struct ofpbuf *buf;
4611
4612 if (ofconn_has_pending_opgroups(ofconn)) {
4613 return OFPROTO_POSTPONE;
4614 }
4615
4616 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4617 ? OFPRAW_OFPT10_BARRIER_REPLY
4618 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4619 ofconn_send_reply(ofconn, buf);
4620 return 0;
4621 }
4622
4623 static void
4624 ofproto_compose_flow_refresh_update(const struct rule *rule,
4625 enum nx_flow_monitor_flags flags,
4626 struct list *msgs)
4627 OVS_REQUIRES(ofproto_mutex)
4628 {
4629 struct ofoperation *op = rule->pending;
4630 const struct rule_actions *actions;
4631 struct ofputil_flow_update fu;
4632 struct match match;
4633
4634 if (op && op->type == OFOPERATION_ADD) {
4635 /* We'll report the final flow when the operation completes. Reporting
4636 * it now would cause a duplicate report later. */
4637 return;
4638 }
4639
4640 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4641 ? NXFME_ADDED : NXFME_MODIFIED);
4642 fu.reason = 0;
4643 ovs_mutex_lock(&rule->mutex);
4644 fu.idle_timeout = rule->idle_timeout;
4645 fu.hard_timeout = rule->hard_timeout;
4646 ovs_mutex_unlock(&rule->mutex);
4647 fu.table_id = rule->table_id;
4648 fu.cookie = rule->flow_cookie;
4649 minimatch_expand(&rule->cr.match, &match);
4650 fu.match = &match;
4651 fu.priority = rule->cr.priority;
4652
4653 if (!(flags & NXFMF_ACTIONS)) {
4654 actions = NULL;
4655 } else if (!op) {
4656 actions = rule->actions;
4657 } else {
4658 /* An operation is in progress. Use the previous version of the flow's
4659 * actions, so that when the operation commits we report the change. */
4660 switch (op->type) {
4661 case OFOPERATION_ADD:
4662 NOT_REACHED();
4663
4664 case OFOPERATION_MODIFY:
4665 case OFOPERATION_REPLACE:
4666 actions = op->actions ? op->actions : rule->actions;
4667 break;
4668
4669 case OFOPERATION_DELETE:
4670 actions = rule->actions;
4671 break;
4672
4673 default:
4674 NOT_REACHED();
4675 }
4676 }
4677 fu.ofpacts = actions ? actions->ofpacts : NULL;
4678 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4679
4680 if (list_is_empty(msgs)) {
4681 ofputil_start_flow_update(msgs);
4682 }
4683 ofputil_append_flow_update(&fu, msgs);
4684 }
4685
4686 void
4687 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4688 struct list *msgs)
4689 OVS_REQUIRES(ofproto_mutex)
4690 {
4691 size_t i;
4692
4693 for (i = 0; i < rules->n; i++) {
4694 struct rule *rule = rules->rules[i];
4695 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4696 rule->monitor_flags = 0;
4697
4698 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4699 }
4700 }
4701
4702 static void
4703 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4704 struct rule *rule, uint64_t seqno,
4705 struct rule_collection *rules)
4706 OVS_REQUIRES(ofproto_mutex)
4707 {
4708 enum nx_flow_monitor_flags update;
4709
4710 if (ofproto_rule_is_hidden(rule)) {
4711 return;
4712 }
4713
4714 if (!(rule->pending
4715 ? ofoperation_has_out_port(rule->pending, m->out_port)
4716 : ofproto_rule_has_out_port(rule, m->out_port))) {
4717 return;
4718 }
4719
4720 if (seqno) {
4721 if (rule->add_seqno > seqno) {
4722 update = NXFMF_ADD | NXFMF_MODIFY;
4723 } else if (rule->modify_seqno > seqno) {
4724 update = NXFMF_MODIFY;
4725 } else {
4726 return;
4727 }
4728
4729 if (!(m->flags & update)) {
4730 return;
4731 }
4732 } else {
4733 update = NXFMF_INITIAL;
4734 }
4735
4736 if (!rule->monitor_flags) {
4737 rule_collection_add(rules, rule);
4738 }
4739 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4740 }
4741
4742 static void
4743 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4744 uint64_t seqno,
4745 struct rule_collection *rules)
4746 OVS_REQUIRES(ofproto_mutex)
4747 {
4748 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4749 const struct ofoperation *op;
4750 const struct oftable *table;
4751 struct cls_rule target;
4752
4753 cls_rule_init_from_minimatch(&target, &m->match, 0);
4754 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4755 struct cls_cursor cursor;
4756 struct rule *rule;
4757
4758 ovs_rwlock_rdlock(&table->cls.rwlock);
4759 cls_cursor_init(&cursor, &table->cls, &target);
4760 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4761 ovs_assert(!rule->pending); /* XXX */
4762 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4763 }
4764 ovs_rwlock_unlock(&table->cls.rwlock);
4765 }
4766
4767 HMAP_FOR_EACH (op, hmap_node, &ofproto->deletions) {
4768 struct rule *rule = op->rule;
4769
4770 if (((m->table_id == 0xff
4771 ? !(ofproto->tables[rule->table_id].flags & OFTABLE_HIDDEN)
4772 : m->table_id == rule->table_id))
4773 && cls_rule_is_loose_match(&rule->cr, &target.match)) {
4774 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4775 }
4776 }
4777 cls_rule_destroy(&target);
4778 }
4779
4780 static void
4781 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4782 struct rule_collection *rules)
4783 OVS_REQUIRES(ofproto_mutex)
4784 {
4785 if (m->flags & NXFMF_INITIAL) {
4786 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4787 }
4788 }
4789
4790 void
4791 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4792 uint64_t seqno, struct rule_collection *rules)
4793 OVS_REQUIRES(ofproto_mutex)
4794 {
4795 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4796 }
4797
4798 static enum ofperr
4799 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4800 OVS_EXCLUDED(ofproto_mutex)
4801 {
4802 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4803 struct ofmonitor **monitors;
4804 size_t n_monitors, allocated_monitors;
4805 struct rule_collection rules;
4806 struct list replies;
4807 enum ofperr error;
4808 struct ofpbuf b;
4809 size_t i;
4810
4811 error = 0;
4812 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4813 monitors = NULL;
4814 n_monitors = allocated_monitors = 0;
4815
4816 ovs_mutex_lock(&ofproto_mutex);
4817 for (;;) {
4818 struct ofputil_flow_monitor_request request;
4819 struct ofmonitor *m;
4820 int retval;
4821
4822 retval = ofputil_decode_flow_monitor_request(&request, &b);
4823 if (retval == EOF) {
4824 break;
4825 } else if (retval) {
4826 error = retval;
4827 goto error;
4828 }
4829
4830 if (request.table_id != 0xff
4831 && request.table_id >= ofproto->n_tables) {
4832 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4833 goto error;
4834 }
4835
4836 error = ofmonitor_create(&request, ofconn, &m);
4837 if (error) {
4838 goto error;
4839 }
4840
4841 if (n_monitors >= allocated_monitors) {
4842 monitors = x2nrealloc(monitors, &allocated_monitors,
4843 sizeof *monitors);
4844 }
4845 monitors[n_monitors++] = m;
4846 }
4847
4848 rule_collection_init(&rules);
4849 for (i = 0; i < n_monitors; i++) {
4850 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4851 }
4852
4853 ofpmp_init(&replies, oh);
4854 ofmonitor_compose_refresh_updates(&rules, &replies);
4855 ovs_mutex_unlock(&ofproto_mutex);
4856
4857 rule_collection_destroy(&rules);
4858
4859 ofconn_send_replies(ofconn, &replies);
4860 free(monitors);
4861
4862 return 0;
4863
4864 error:
4865 ovs_mutex_unlock(&ofproto_mutex);
4866
4867 for (i = 0; i < n_monitors; i++) {
4868 ofmonitor_destroy(monitors[i]);
4869 }
4870 free(monitors);
4871 return error;
4872 }
4873
4874 static enum ofperr
4875 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4876 OVS_EXCLUDED(ofproto_mutex)
4877 {
4878 struct ofmonitor *m;
4879 enum ofperr error;
4880 uint32_t id;
4881
4882 id = ofputil_decode_flow_monitor_cancel(oh);
4883
4884 ovs_mutex_lock(&ofproto_mutex);
4885 m = ofmonitor_lookup(ofconn, id);
4886 if (m) {
4887 ofmonitor_destroy(m);
4888 error = 0;
4889 } else {
4890 error = OFPERR_NXBRC_FM_BAD_ID;
4891 }
4892 ovs_mutex_unlock(&ofproto_mutex);
4893
4894 return error;
4895 }
4896
4897 /* Meters implementation.
4898 *
4899 * Meter table entry, indexed by the OpenFlow meter_id.
4900 * These are always dynamically allocated to allocate enough space for
4901 * the bands.
4902 * 'created' is used to compute the duration for meter stats.
4903 * 'list rules' is needed so that we can delete the dependent rules when the
4904 * meter table entry is deleted.
4905 * 'provider_meter_id' is for the provider's private use.
4906 */
4907 struct meter {
4908 long long int created; /* Time created. */
4909 struct list rules; /* List of "struct rule_dpif"s. */
4910 ofproto_meter_id provider_meter_id;
4911 uint16_t flags; /* Meter flags. */
4912 uint16_t n_bands; /* Number of meter bands. */
4913 struct ofputil_meter_band *bands;
4914 };
4915
4916 /*
4917 * This is used in instruction validation at flow set-up time,
4918 * as flows may not use non-existing meters.
4919 * Return value of UINT32_MAX signifies an invalid meter.
4920 */
4921 static uint32_t
4922 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
4923 {
4924 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4925 const struct meter *meter = ofproto->meters[of_meter_id];
4926 if (meter) {
4927 return meter->provider_meter_id.uint32;
4928 }
4929 }
4930 return UINT32_MAX;
4931 }
4932
4933 static void
4934 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4935 {
4936 free(meter->bands);
4937
4938 meter->flags = config->flags;
4939 meter->n_bands = config->n_bands;
4940 meter->bands = xmemdup(config->bands,
4941 config->n_bands * sizeof *meter->bands);
4942 }
4943
4944 static struct meter *
4945 meter_create(const struct ofputil_meter_config *config,
4946 ofproto_meter_id provider_meter_id)
4947 {
4948 struct meter *meter;
4949
4950 meter = xzalloc(sizeof *meter);
4951 meter->provider_meter_id = provider_meter_id;
4952 meter->created = time_msec();
4953 list_init(&meter->rules);
4954
4955 meter_update(meter, config);
4956
4957 return meter;
4958 }
4959
4960 static void
4961 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4962 OVS_REQUIRES(ofproto_mutex)
4963 {
4964 uint32_t mid;
4965 for (mid = first; mid <= last; ++mid) {
4966 struct meter *meter = ofproto->meters[mid];
4967 if (meter) {
4968 ofproto->meters[mid] = NULL;
4969 ofproto->ofproto_class->meter_del(ofproto,
4970 meter->provider_meter_id);
4971 free(meter->bands);
4972 free(meter);
4973 }
4974 }
4975 }
4976
4977 static enum ofperr
4978 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4979 {
4980 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4981 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
4982 enum ofperr error;
4983
4984 if (*meterp) {
4985 return OFPERR_OFPMMFC_METER_EXISTS;
4986 }
4987
4988 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
4989 &mm->meter);
4990 if (!error) {
4991 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
4992 *meterp = meter_create(&mm->meter, provider_meter_id);
4993 }
4994 return error;
4995 }
4996
4997 static enum ofperr
4998 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4999 {
5000 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5001 enum ofperr error;
5002 uint32_t provider_meter_id;
5003
5004 if (!meter) {
5005 return OFPERR_OFPMMFC_UNKNOWN_METER;
5006 }
5007
5008 provider_meter_id = meter->provider_meter_id.uint32;
5009 error = ofproto->ofproto_class->meter_set(ofproto,
5010 &meter->provider_meter_id,
5011 &mm->meter);
5012 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5013 if (!error) {
5014 meter_update(meter, &mm->meter);
5015 }
5016 return error;
5017 }
5018
5019 static enum ofperr
5020 handle_delete_meter(struct ofconn *ofconn, const struct ofp_header *oh,
5021 struct ofputil_meter_mod *mm)
5022 OVS_EXCLUDED(ofproto_mutex)
5023 {
5024 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5025 uint32_t meter_id = mm->meter.meter_id;
5026 struct rule_collection rules;
5027 enum ofperr error = 0;
5028 uint32_t first, last;
5029
5030 if (meter_id == OFPM13_ALL) {
5031 first = 1;
5032 last = ofproto->meter_features.max_meters;
5033 } else {
5034 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5035 return 0;
5036 }
5037 first = last = meter_id;
5038 }
5039
5040 /* First delete the rules that use this meter. If any of those rules are
5041 * currently being modified, postpone the whole operation until later. */
5042 rule_collection_init(&rules);
5043 ovs_mutex_lock(&ofproto_mutex);
5044 for (meter_id = first; meter_id <= last; ++meter_id) {
5045 struct meter *meter = ofproto->meters[meter_id];
5046 if (meter && !list_is_empty(&meter->rules)) {
5047 struct rule *rule;
5048
5049 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5050 if (rule->pending) {
5051 error = OFPROTO_POSTPONE;
5052 goto exit;
5053 }
5054 rule_collection_add(&rules, rule);
5055 }
5056 }
5057 }
5058 if (rules.n > 0) {
5059 delete_flows__(ofproto, ofconn, oh, &rules, OFPRR_METER_DELETE);
5060 }
5061
5062 /* Delete the meters. */
5063 meter_delete(ofproto, first, last);
5064
5065 exit:
5066 ovs_mutex_unlock(&ofproto_mutex);
5067 rule_collection_destroy(&rules);
5068
5069 return error;
5070 }
5071
5072 static enum ofperr
5073 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5074 {
5075 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5076 struct ofputil_meter_mod mm;
5077 uint64_t bands_stub[256 / 8];
5078 struct ofpbuf bands;
5079 uint32_t meter_id;
5080 enum ofperr error;
5081
5082 error = reject_slave_controller(ofconn);
5083 if (error) {
5084 return error;
5085 }
5086
5087 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5088
5089 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5090 if (error) {
5091 goto exit_free_bands;
5092 }
5093
5094 meter_id = mm.meter.meter_id;
5095
5096 if (mm.command != OFPMC13_DELETE) {
5097 /* Fails also when meters are not implemented by the provider. */
5098 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5099 error = OFPERR_OFPMMFC_INVALID_METER;
5100 goto exit_free_bands;
5101 } else if (meter_id > ofproto->meter_features.max_meters) {
5102 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5103 goto exit_free_bands;
5104 }
5105 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5106 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5107 goto exit_free_bands;
5108 }
5109 }
5110
5111 switch (mm.command) {
5112 case OFPMC13_ADD:
5113 error = handle_add_meter(ofproto, &mm);
5114 break;
5115
5116 case OFPMC13_MODIFY:
5117 error = handle_modify_meter(ofproto, &mm);
5118 break;
5119
5120 case OFPMC13_DELETE:
5121 error = handle_delete_meter(ofconn, oh, &mm);
5122 break;
5123
5124 default:
5125 error = OFPERR_OFPMMFC_BAD_COMMAND;
5126 break;
5127 }
5128
5129 exit_free_bands:
5130 ofpbuf_uninit(&bands);
5131 return error;
5132 }
5133
5134 static enum ofperr
5135 handle_meter_features_request(struct ofconn *ofconn,
5136 const struct ofp_header *request)
5137 {
5138 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5139 struct ofputil_meter_features features;
5140 struct ofpbuf *b;
5141
5142 if (ofproto->ofproto_class->meter_get_features) {
5143 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5144 } else {
5145 memset(&features, 0, sizeof features);
5146 }
5147 b = ofputil_encode_meter_features_reply(&features, request);
5148
5149 ofconn_send_reply(ofconn, b);
5150 return 0;
5151 }
5152
5153 static enum ofperr
5154 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5155 enum ofptype type)
5156 {
5157 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5158 struct list replies;
5159 uint64_t bands_stub[256 / 8];
5160 struct ofpbuf bands;
5161 uint32_t meter_id, first, last;
5162
5163 ofputil_decode_meter_request(request, &meter_id);
5164
5165 if (meter_id == OFPM13_ALL) {
5166 first = 1;
5167 last = ofproto->meter_features.max_meters;
5168 } else {
5169 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5170 !ofproto->meters[meter_id]) {
5171 return OFPERR_OFPMMFC_UNKNOWN_METER;
5172 }
5173 first = last = meter_id;
5174 }
5175
5176 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5177 ofpmp_init(&replies, request);
5178
5179 for (meter_id = first; meter_id <= last; ++meter_id) {
5180 struct meter *meter = ofproto->meters[meter_id];
5181 if (!meter) {
5182 continue; /* Skip non-existing meters. */
5183 }
5184 if (type == OFPTYPE_METER_STATS_REQUEST) {
5185 struct ofputil_meter_stats stats;
5186
5187 stats.meter_id = meter_id;
5188
5189 /* Provider sets the packet and byte counts, we do the rest. */
5190 stats.flow_count = list_size(&meter->rules);
5191 calc_duration(meter->created, time_msec(),
5192 &stats.duration_sec, &stats.duration_nsec);
5193 stats.n_bands = meter->n_bands;
5194 ofpbuf_clear(&bands);
5195 stats.bands
5196 = ofpbuf_put_uninit(&bands,
5197 meter->n_bands * sizeof *stats.bands);
5198
5199 if (!ofproto->ofproto_class->meter_get(ofproto,
5200 meter->provider_meter_id,
5201 &stats)) {
5202 ofputil_append_meter_stats(&replies, &stats);
5203 }
5204 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5205 struct ofputil_meter_config config;
5206
5207 config.meter_id = meter_id;
5208 config.flags = meter->flags;
5209 config.n_bands = meter->n_bands;
5210 config.bands = meter->bands;
5211 ofputil_append_meter_config(&replies, &config);
5212 }
5213 }
5214
5215 ofconn_send_replies(ofconn, &replies);
5216 ofpbuf_uninit(&bands);
5217 return 0;
5218 }
5219
5220 bool
5221 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5222 struct ofgroup **group)
5223 OVS_TRY_RDLOCK(true, (*group)->rwlock)
5224 {
5225 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5226 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5227 hash_int(group_id, 0), &ofproto->groups) {
5228 if ((*group)->group_id == group_id) {
5229 ovs_rwlock_rdlock(&(*group)->rwlock);
5230 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5231 return true;
5232 }
5233 }
5234 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5235 return false;
5236 }
5237
5238 void
5239 ofproto_group_release(struct ofgroup *group)
5240 OVS_RELEASES(group->rwlock)
5241 {
5242 ovs_rwlock_unlock(&group->rwlock);
5243 }
5244
5245 static bool
5246 ofproto_group_write_lookup(const struct ofproto *ofproto, uint32_t group_id,
5247 struct ofgroup **group)
5248 OVS_TRY_WRLOCK(true, ofproto->groups_rwlock)
5249 OVS_TRY_WRLOCK(true, (*group)->rwlock)
5250 {
5251 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5252 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5253 hash_int(group_id, 0), &ofproto->groups) {
5254 if ((*group)->group_id == group_id) {
5255 ovs_rwlock_wrlock(&(*group)->rwlock);
5256 return true;
5257 }
5258 }
5259 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5260 return false;
5261 }
5262
5263 static bool
5264 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5265 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5266 {
5267 struct ofgroup *grp;
5268
5269 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5270 hash_int(group_id, 0), &ofproto->groups) {
5271 if (grp->group_id == group_id) {
5272 return true;
5273 }
5274 }
5275 return false;
5276 }
5277
5278 static void
5279 append_group_stats(struct ofgroup *group, struct list *replies)
5280 OVS_REQ_RDLOCK(group->rwlock)
5281 {
5282 struct ofputil_group_stats ogs;
5283 struct ofproto *ofproto = group->ofproto;
5284 long long int now = time_msec();
5285 int error;
5286
5287 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5288
5289 error = (ofproto->ofproto_class->group_get_stats
5290 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5291 : EOPNOTSUPP);
5292 if (error) {
5293 ogs.ref_count = UINT32_MAX;
5294 ogs.packet_count = UINT64_MAX;
5295 ogs.byte_count = UINT64_MAX;
5296 ogs.n_buckets = group->n_buckets;
5297 memset(ogs.bucket_stats, 0xff,
5298 ogs.n_buckets * sizeof *ogs.bucket_stats);
5299 }
5300
5301 ogs.group_id = group->group_id;
5302 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5303
5304 ofputil_append_group_stats(replies, &ogs);
5305
5306 free(ogs.bucket_stats);
5307 }
5308
5309 static enum ofperr
5310 handle_group_stats_request(struct ofconn *ofconn,
5311 const struct ofp_header *request)
5312 {
5313 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5314 struct list replies;
5315 enum ofperr error;
5316 struct ofgroup *group;
5317 uint32_t group_id;
5318
5319 error = ofputil_decode_group_stats_request(request, &group_id);
5320 if (error) {
5321 return error;
5322 }
5323
5324 ofpmp_init(&replies, request);
5325
5326 if (group_id == OFPG_ALL) {
5327 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5328 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5329 ovs_rwlock_rdlock(&group->rwlock);
5330 append_group_stats(group, &replies);
5331 ovs_rwlock_unlock(&group->rwlock);
5332 }
5333 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5334 } else {
5335 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5336 append_group_stats(group, &replies);
5337 ofproto_group_release(group);
5338 }
5339 }
5340
5341 ofconn_send_replies(ofconn, &replies);
5342
5343 return 0;
5344 }
5345
5346 static enum ofperr
5347 handle_group_desc_stats_request(struct ofconn *ofconn,
5348 const struct ofp_header *request)
5349 {
5350 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5351 struct list replies;
5352 struct ofputil_group_desc gds;
5353 struct ofgroup *group;
5354
5355 ofpmp_init(&replies, request);
5356
5357 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5358 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5359 gds.group_id = group->group_id;
5360 gds.type = group->type;
5361 ofputil_append_group_desc_reply(&gds, &group->buckets, &replies);
5362 }
5363 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5364
5365 ofconn_send_replies(ofconn, &replies);
5366
5367 return 0;
5368 }
5369
5370 static enum ofperr
5371 handle_group_features_stats_request(struct ofconn *ofconn,
5372 const struct ofp_header *request)
5373 {
5374 struct ofproto *p = ofconn_get_ofproto(ofconn);
5375 struct ofpbuf *msg;
5376
5377 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5378 if (msg) {
5379 ofconn_send_reply(ofconn, msg);
5380 }
5381
5382 return 0;
5383 }
5384
5385 /* Implements OFPGC11_ADD
5386 * in which no matching flow already exists in the flow table.
5387 *
5388 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
5389 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
5390 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
5391 * initiated now but may be retried later.
5392 *
5393 * Upon successful return, takes ownership of 'fm->ofpacts'. On failure,
5394 * ownership remains with the caller.
5395 *
5396 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
5397 * if any. */
5398 static enum ofperr
5399 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5400 {
5401 struct ofgroup *ofgroup;
5402 enum ofperr error;
5403
5404 if (gm->group_id > OFPG_MAX) {
5405 return OFPERR_OFPGMFC_INVALID_GROUP;
5406 }
5407 if (gm->type > OFPGT11_FF) {
5408 return OFPERR_OFPGMFC_BAD_TYPE;
5409 }
5410
5411 /* Allocate new group and initialize it. */
5412 ofgroup = ofproto->ofproto_class->group_alloc();
5413 if (!ofgroup) {
5414 VLOG_WARN_RL(&rl, "%s: failed to create group", ofproto->name);
5415 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5416 }
5417
5418 ovs_rwlock_init(&ofgroup->rwlock);
5419 ofgroup->ofproto = ofproto;
5420 ofgroup->group_id = gm->group_id;
5421 ofgroup->type = gm->type;
5422 ofgroup->created = ofgroup->modified = time_msec();
5423
5424 list_move(&ofgroup->buckets, &gm->buckets);
5425 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5426
5427 /* Construct called BEFORE any locks are held. */
5428 error = ofproto->ofproto_class->group_construct(ofgroup);
5429 if (error) {
5430 goto free_out;
5431 }
5432
5433 /* We wrlock as late as possible to minimize the time we jam any other
5434 * threads: No visible state changes before acquiring the lock. */
5435 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5436
5437 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5438 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5439 goto unlock_out;
5440 }
5441
5442 if (ofproto_group_exists(ofproto, gm->group_id)) {
5443 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5444 goto unlock_out;
5445 }
5446
5447 if (!error) {
5448 /* Insert new group. */
5449 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5450 hash_int(ofgroup->group_id, 0));
5451 ofproto->n_groups[ofgroup->type]++;
5452
5453 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5454 return error;
5455 }
5456
5457 unlock_out:
5458 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5459 ofproto->ofproto_class->group_destruct(ofgroup);
5460 free_out:
5461 ofputil_bucket_list_destroy(&ofgroup->buckets);
5462 ofproto->ofproto_class->group_dealloc(ofgroup);
5463
5464 return error;
5465 }
5466
5467 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5468 * failure.
5469 *
5470 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5471 * if any. */
5472 static enum ofperr
5473 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5474 {
5475 struct ofgroup *ofgroup;
5476 struct ofgroup *victim;
5477 enum ofperr error;
5478
5479 if (gm->group_id > OFPG_MAX) {
5480 return OFPERR_OFPGMFC_INVALID_GROUP;
5481 }
5482
5483 if (gm->type > OFPGT11_FF) {
5484 return OFPERR_OFPGMFC_BAD_TYPE;
5485 }
5486
5487 victim = ofproto->ofproto_class->group_alloc();
5488 if (!victim) {
5489 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5490 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5491 }
5492
5493 if (!ofproto_group_write_lookup(ofproto, gm->group_id, &ofgroup)) {
5494 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5495 goto free_out;
5496 }
5497 /* Both group's and its container's write locks held now.
5498 * Also, n_groups[] is protected by ofproto->groups_rwlock. */
5499 if (ofgroup->type != gm->type
5500 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5501 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5502 goto unlock_out;
5503 }
5504
5505 *victim = *ofgroup;
5506 list_move(&victim->buckets, &ofgroup->buckets);
5507
5508 ofgroup->type = gm->type;
5509 list_move(&ofgroup->buckets, &gm->buckets);
5510 ofgroup->n_buckets = list_size(&ofgroup->buckets);
5511
5512 error = ofproto->ofproto_class->group_modify(ofgroup, victim);
5513 if (!error) {
5514 ofputil_bucket_list_destroy(&victim->buckets);
5515 ofproto->n_groups[victim->type]--;
5516 ofproto->n_groups[ofgroup->type]++;
5517 ofgroup->modified = time_msec();
5518 } else {
5519 ofputil_bucket_list_destroy(&ofgroup->buckets);
5520
5521 *ofgroup = *victim;
5522 list_move(&ofgroup->buckets, &victim->buckets);
5523 }
5524
5525 unlock_out:
5526 ovs_rwlock_unlock(&ofgroup->rwlock);
5527 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5528 free_out:
5529 ofproto->ofproto_class->group_dealloc(victim);
5530 return error;
5531 }
5532
5533 static void
5534 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5535 OVS_RELEASES(ofproto->groups_rwlock)
5536 {
5537 /* Must wait until existing readers are done,
5538 * while holding the container's write lock at the same time. */
5539 ovs_rwlock_wrlock(&ofgroup->rwlock);
5540 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5541 /* No-one can find this group any more. */
5542 ofproto->n_groups[ofgroup->type]--;
5543 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5544
5545 ofproto->ofproto_class->group_destruct(ofgroup);
5546 ofputil_bucket_list_destroy(&ofgroup->buckets);
5547 ovs_rwlock_unlock(&ofgroup->rwlock);
5548 ovs_rwlock_destroy(&ofgroup->rwlock);
5549 ofproto->ofproto_class->group_dealloc(ofgroup);
5550 }
5551
5552 /* Implements OFPGC_DELETE. */
5553 static void
5554 delete_group(struct ofproto *ofproto, uint32_t group_id)
5555 {
5556 struct ofgroup *ofgroup;
5557
5558 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5559 if (group_id == OFPG_ALL) {
5560 for (;;) {
5561 struct hmap_node *node = hmap_first(&ofproto->groups);
5562 if (!node) {
5563 break;
5564 }
5565 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5566 delete_group__(ofproto, ofgroup);
5567 /* Lock for each node separately, so that we will not jam the
5568 * other threads for too long time. */
5569 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5570 }
5571 } else {
5572 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5573 hash_int(group_id, 0), &ofproto->groups) {
5574 if (ofgroup->group_id == group_id) {
5575 delete_group__(ofproto, ofgroup);
5576 return;
5577 }
5578 }
5579 }
5580 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5581 }
5582
5583 static enum ofperr
5584 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5585 {
5586 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5587 struct ofputil_group_mod gm;
5588 enum ofperr error;
5589
5590 error = reject_slave_controller(ofconn);
5591 if (error) {
5592 return error;
5593 }
5594
5595 error = ofputil_decode_group_mod(oh, &gm);
5596 if (error) {
5597 return error;
5598 }
5599
5600 switch (gm.command) {
5601 case OFPGC11_ADD:
5602 return add_group(ofproto, &gm);
5603
5604 case OFPGC11_MODIFY:
5605 return modify_group(ofproto, &gm);
5606
5607 case OFPGC11_DELETE:
5608 delete_group(ofproto, gm.group_id);
5609 return 0;
5610
5611 default:
5612 if (gm.command > OFPGC11_DELETE) {
5613 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5614 ofproto->name, gm.command);
5615 }
5616 return OFPERR_OFPGMFC_BAD_COMMAND;
5617 }
5618 }
5619
5620 static enum ofperr
5621 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5622 {
5623 struct ofputil_table_mod tm;
5624 enum ofperr error;
5625
5626 error = reject_slave_controller(ofconn);
5627 if (error) {
5628 return error;
5629 }
5630
5631 error = ofputil_decode_table_mod(oh, &tm);
5632 if (error) {
5633 return error;
5634 }
5635
5636 /* XXX Actual table mod support is not implemented yet. */
5637 return 0;
5638 }
5639
5640 static enum ofperr
5641 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5642 OVS_EXCLUDED(ofproto_mutex)
5643 {
5644 const struct ofp_header *oh = msg->data;
5645 enum ofptype type;
5646 enum ofperr error;
5647
5648 error = ofptype_decode(&type, oh);
5649 if (error) {
5650 return error;
5651 }
5652
5653 switch (type) {
5654 /* OpenFlow requests. */
5655 case OFPTYPE_ECHO_REQUEST:
5656 return handle_echo_request(ofconn, oh);
5657
5658 case OFPTYPE_FEATURES_REQUEST:
5659 return handle_features_request(ofconn, oh);
5660
5661 case OFPTYPE_GET_CONFIG_REQUEST:
5662 return handle_get_config_request(ofconn, oh);
5663
5664 case OFPTYPE_SET_CONFIG:
5665 return handle_set_config(ofconn, oh);
5666
5667 case OFPTYPE_PACKET_OUT:
5668 return handle_packet_out(ofconn, oh);
5669
5670 case OFPTYPE_PORT_MOD:
5671 return handle_port_mod(ofconn, oh);
5672
5673 case OFPTYPE_FLOW_MOD:
5674 return handle_flow_mod(ofconn, oh);
5675
5676 case OFPTYPE_GROUP_MOD:
5677 return handle_group_mod(ofconn, oh);
5678
5679 case OFPTYPE_TABLE_MOD:
5680 return handle_table_mod(ofconn, oh);
5681
5682 case OFPTYPE_METER_MOD:
5683 return handle_meter_mod(ofconn, oh);
5684
5685 case OFPTYPE_BARRIER_REQUEST:
5686 return handle_barrier_request(ofconn, oh);
5687
5688 case OFPTYPE_ROLE_REQUEST:
5689 return handle_role_request(ofconn, oh);
5690
5691 /* OpenFlow replies. */
5692 case OFPTYPE_ECHO_REPLY:
5693 return 0;
5694
5695 /* Nicira extension requests. */
5696 case OFPTYPE_FLOW_MOD_TABLE_ID:
5697 return handle_nxt_flow_mod_table_id(ofconn, oh);
5698
5699 case OFPTYPE_SET_FLOW_FORMAT:
5700 return handle_nxt_set_flow_format(ofconn, oh);
5701
5702 case OFPTYPE_SET_PACKET_IN_FORMAT:
5703 return handle_nxt_set_packet_in_format(ofconn, oh);
5704
5705 case OFPTYPE_SET_CONTROLLER_ID:
5706 return handle_nxt_set_controller_id(ofconn, oh);
5707
5708 case OFPTYPE_FLOW_AGE:
5709 /* Nothing to do. */
5710 return 0;
5711
5712 case OFPTYPE_FLOW_MONITOR_CANCEL:
5713 return handle_flow_monitor_cancel(ofconn, oh);
5714
5715 case OFPTYPE_SET_ASYNC_CONFIG:
5716 return handle_nxt_set_async_config(ofconn, oh);
5717
5718 case OFPTYPE_GET_ASYNC_REQUEST:
5719 return handle_nxt_get_async_request(ofconn, oh);
5720
5721 /* Statistics requests. */
5722 case OFPTYPE_DESC_STATS_REQUEST:
5723 return handle_desc_stats_request(ofconn, oh);
5724
5725 case OFPTYPE_FLOW_STATS_REQUEST:
5726 return handle_flow_stats_request(ofconn, oh);
5727
5728 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5729 return handle_aggregate_stats_request(ofconn, oh);
5730
5731 case OFPTYPE_TABLE_STATS_REQUEST:
5732 return handle_table_stats_request(ofconn, oh);
5733
5734 case OFPTYPE_PORT_STATS_REQUEST:
5735 return handle_port_stats_request(ofconn, oh);
5736
5737 case OFPTYPE_QUEUE_STATS_REQUEST:
5738 return handle_queue_stats_request(ofconn, oh);
5739
5740 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5741 return handle_port_desc_stats_request(ofconn, oh);
5742
5743 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5744 return handle_flow_monitor_request(ofconn, oh);
5745
5746 case OFPTYPE_METER_STATS_REQUEST:
5747 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5748 return handle_meter_request(ofconn, oh, type);
5749
5750 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5751 return handle_meter_features_request(ofconn, oh);
5752
5753 case OFPTYPE_GROUP_STATS_REQUEST:
5754 return handle_group_stats_request(ofconn, oh);
5755
5756 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5757 return handle_group_desc_stats_request(ofconn, oh);
5758
5759 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5760 return handle_group_features_stats_request(ofconn, oh);
5761
5762 /* FIXME: Change the following once they are implemented: */
5763 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5764 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5765 return OFPERR_OFPBRC_BAD_TYPE;
5766
5767 case OFPTYPE_HELLO:
5768 case OFPTYPE_ERROR:
5769 case OFPTYPE_FEATURES_REPLY:
5770 case OFPTYPE_GET_CONFIG_REPLY:
5771 case OFPTYPE_PACKET_IN:
5772 case OFPTYPE_FLOW_REMOVED:
5773 case OFPTYPE_PORT_STATUS:
5774 case OFPTYPE_BARRIER_REPLY:
5775 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5776 case OFPTYPE_DESC_STATS_REPLY:
5777 case OFPTYPE_FLOW_STATS_REPLY:
5778 case OFPTYPE_QUEUE_STATS_REPLY:
5779 case OFPTYPE_PORT_STATS_REPLY:
5780 case OFPTYPE_TABLE_STATS_REPLY:
5781 case OFPTYPE_AGGREGATE_STATS_REPLY:
5782 case OFPTYPE_PORT_DESC_STATS_REPLY:
5783 case OFPTYPE_ROLE_REPLY:
5784 case OFPTYPE_FLOW_MONITOR_PAUSED:
5785 case OFPTYPE_FLOW_MONITOR_RESUMED:
5786 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5787 case OFPTYPE_GET_ASYNC_REPLY:
5788 case OFPTYPE_GROUP_STATS_REPLY:
5789 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5790 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5791 case OFPTYPE_METER_STATS_REPLY:
5792 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5793 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5794 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5795 default:
5796 return OFPERR_OFPBRC_BAD_TYPE;
5797 }
5798 }
5799
5800 static bool
5801 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5802 OVS_EXCLUDED(ofproto_mutex)
5803 {
5804 int error = handle_openflow__(ofconn, ofp_msg);
5805 if (error && error != OFPROTO_POSTPONE) {
5806 ofconn_send_error(ofconn, ofp_msg->data, error);
5807 }
5808 COVERAGE_INC(ofproto_recv_openflow);
5809 return error != OFPROTO_POSTPONE;
5810 }
5811 \f
5812 /* Asynchronous operations. */
5813
5814 /* Creates and returns a new ofopgroup that is not associated with any
5815 * OpenFlow connection.
5816 *
5817 * The caller should add operations to the returned group with
5818 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5819 static struct ofopgroup *
5820 ofopgroup_create_unattached(struct ofproto *ofproto)
5821 OVS_REQUIRES(ofproto_mutex)
5822 {
5823 struct ofopgroup *group = xzalloc(sizeof *group);
5824 group->ofproto = ofproto;
5825 list_init(&group->ofproto_node);
5826 list_init(&group->ops);
5827 list_init(&group->ofconn_node);
5828 return group;
5829 }
5830
5831 /* Creates and returns a new ofopgroup for 'ofproto'.
5832 *
5833 * If 'ofconn' is NULL, the new ofopgroup is not associated with any OpenFlow
5834 * connection. The 'request' and 'buffer_id' arguments are ignored.
5835 *
5836 * If 'ofconn' is nonnull, then the new ofopgroup is associated with 'ofconn'.
5837 * If the ofopgroup eventually fails, then the error reply will include
5838 * 'request'. If the ofopgroup eventually succeeds, then the packet with
5839 * buffer id 'buffer_id' on 'ofconn' will be sent by 'ofconn''s ofproto.
5840 *
5841 * The caller should add operations to the returned group with
5842 * ofoperation_create() and then submit it with ofopgroup_submit(). */
5843 static struct ofopgroup *
5844 ofopgroup_create(struct ofproto *ofproto, struct ofconn *ofconn,
5845 const struct ofp_header *request, uint32_t buffer_id)
5846 OVS_REQUIRES(ofproto_mutex)
5847 {
5848 struct ofopgroup *group = ofopgroup_create_unattached(ofproto);
5849 if (ofconn) {
5850 size_t request_len = ntohs(request->length);
5851
5852 ovs_assert(ofconn_get_ofproto(ofconn) == ofproto);
5853
5854 ofconn_add_opgroup(ofconn, &group->ofconn_node);
5855 group->ofconn = ofconn;
5856 group->request = xmemdup(request, MIN(request_len, 64));
5857 group->buffer_id = buffer_id;
5858 }
5859 return group;
5860 }
5861
5862 /* Submits 'group' for processing.
5863 *
5864 * If 'group' contains no operations (e.g. none were ever added, or all of the
5865 * ones that were added completed synchronously), then it is destroyed
5866 * immediately. Otherwise it is added to the ofproto's list of pending
5867 * groups. */
5868 static void
5869 ofopgroup_submit(struct ofopgroup *group)
5870 OVS_REQUIRES(ofproto_mutex)
5871 {
5872 if (!group->n_running) {
5873 ofopgroup_complete(group);
5874 } else {
5875 list_push_back(&group->ofproto->pending, &group->ofproto_node);
5876 group->ofproto->n_pending++;
5877 }
5878 }
5879
5880 static void
5881 ofopgroup_complete(struct ofopgroup *group)
5882 OVS_REQUIRES(ofproto_mutex)
5883 {
5884 struct ofproto *ofproto = group->ofproto;
5885
5886 struct ofconn *abbrev_ofconn;
5887 ovs_be32 abbrev_xid;
5888
5889 struct ofoperation *op, *next_op;
5890 int error;
5891
5892 ovs_assert(!group->n_running);
5893
5894 error = 0;
5895 LIST_FOR_EACH (op, group_node, &group->ops) {
5896 if (op->error) {
5897 error = op->error;
5898 break;
5899 }
5900 }
5901
5902 if (!error && group->ofconn && group->buffer_id != UINT32_MAX) {
5903 LIST_FOR_EACH (op, group_node, &group->ops) {
5904 if (op->type != OFOPERATION_DELETE) {
5905 struct ofpbuf *packet;
5906 ofp_port_t in_port;
5907
5908 error = ofconn_pktbuf_retrieve(group->ofconn, group->buffer_id,
5909 &packet, &in_port);
5910 if (packet) {
5911 struct rule_execute *re;
5912
5913 ovs_assert(!error);
5914
5915 ofproto_rule_ref(op->rule);
5916
5917 re = xmalloc(sizeof *re);
5918 re->rule = op->rule;
5919 re->in_port = in_port;
5920 re->packet = packet;
5921
5922 if (!guarded_list_push_back(&ofproto->rule_executes,
5923 &re->list_node, 1024)) {
5924 ofproto_rule_unref(op->rule);
5925 ofpbuf_delete(re->packet);
5926 free(re);
5927 }
5928 }
5929 break;
5930 }
5931 }
5932 }
5933
5934 if (!error && !list_is_empty(&group->ofconn_node)) {
5935 abbrev_ofconn = group->ofconn;
5936 abbrev_xid = group->request->xid;
5937 } else {
5938 abbrev_ofconn = NULL;
5939 abbrev_xid = htonl(0);
5940 }
5941 LIST_FOR_EACH_SAFE (op, next_op, group_node, &group->ops) {
5942 struct rule *rule = op->rule;
5943
5944 /* We generally want to report the change to active OpenFlow flow
5945 monitors (e.g. NXST_FLOW_MONITOR). There are three exceptions:
5946
5947 - The operation failed.
5948
5949 - The affected rule is not visible to controllers.
5950
5951 - The operation's only effect was to update rule->modified. */
5952 if (!(op->error
5953 || ofproto_rule_is_hidden(rule)
5954 || (op->type == OFOPERATION_MODIFY
5955 && op->actions
5956 && rule->flow_cookie == op->flow_cookie))) {
5957 /* Check that we can just cast from ofoperation_type to
5958 * nx_flow_update_event. */
5959 enum nx_flow_update_event event_type;
5960
5961 switch (op->type) {
5962 case OFOPERATION_ADD:
5963 case OFOPERATION_REPLACE:
5964 event_type = NXFME_ADDED;
5965 break;
5966
5967 case OFOPERATION_DELETE:
5968 event_type = NXFME_DELETED;
5969 break;
5970
5971 case OFOPERATION_MODIFY:
5972 event_type = NXFME_MODIFIED;
5973 break;
5974
5975 default:
5976 NOT_REACHED();
5977 }
5978
5979 ofmonitor_report(ofproto->connmgr, rule, event_type,
5980 op->reason, abbrev_ofconn, abbrev_xid);
5981 }
5982
5983 rule->pending = NULL;
5984
5985 switch (op->type) {
5986 case OFOPERATION_ADD:
5987 if (!op->error) {
5988 uint16_t vid_mask;
5989
5990 vid_mask = minimask_get_vid_mask(&rule->cr.match.mask);
5991 if (vid_mask == VLAN_VID_MASK) {
5992 if (ofproto->vlan_bitmap) {
5993 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
5994 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
5995 bitmap_set1(ofproto->vlan_bitmap, vid);
5996 ofproto->vlans_changed = true;
5997 }
5998 } else {
5999 ofproto->vlans_changed = true;
6000 }
6001 }
6002 } else {
6003 oftable_remove_rule(rule);
6004 ofproto_rule_unref(rule);
6005 }
6006 break;
6007
6008 case OFOPERATION_DELETE:
6009 ovs_assert(!op->error);
6010 ofproto_rule_unref(rule);
6011 op->rule = NULL;
6012 break;
6013
6014 case OFOPERATION_MODIFY:
6015 case OFOPERATION_REPLACE:
6016 if (!op->error) {
6017 long long int now = time_msec();
6018
6019 rule->modified = now;
6020 if (op->type == OFOPERATION_REPLACE) {
6021 rule->created = rule->used = now;
6022 }
6023 } else {
6024 ofproto_rule_change_cookie(ofproto, rule, op->flow_cookie);
6025 ovs_mutex_lock(&rule->mutex);
6026 rule->idle_timeout = op->idle_timeout;
6027 rule->hard_timeout = op->hard_timeout;
6028 ovs_mutex_unlock(&rule->mutex);
6029 if (op->actions) {
6030 struct rule_actions *old_actions;
6031
6032 ovs_mutex_lock(&rule->mutex);
6033 old_actions = rule->actions;
6034 rule->actions = op->actions;
6035 ovs_mutex_unlock(&rule->mutex);
6036
6037 op->actions = NULL;
6038 rule_actions_unref(old_actions);
6039 }
6040 rule->flags = op->flags;
6041 }
6042 break;
6043
6044 default:
6045 NOT_REACHED();
6046 }
6047
6048 ofoperation_destroy(op);
6049 }
6050
6051 ofmonitor_flush(ofproto->connmgr);
6052
6053 if (!list_is_empty(&group->ofproto_node)) {
6054 ovs_assert(ofproto->n_pending > 0);
6055 ofproto->n_pending--;
6056 list_remove(&group->ofproto_node);
6057 }
6058 if (!list_is_empty(&group->ofconn_node)) {
6059 list_remove(&group->ofconn_node);
6060 if (error) {
6061 ofconn_send_error(group->ofconn, group->request, error);
6062 }
6063 connmgr_retry(ofproto->connmgr);
6064 }
6065 free(group->request);
6066 free(group);
6067 }
6068
6069 /* Initiates a new operation on 'rule', of the specified 'type', within
6070 * 'group'. Prior to calling, 'rule' must not have any pending operation.
6071 *
6072 * For a 'type' of OFOPERATION_DELETE, 'reason' should specify the reason that
6073 * the flow is being deleted. For other 'type's, 'reason' is ignored (use 0).
6074 *
6075 * Returns the newly created ofoperation (which is also available as
6076 * rule->pending). */
6077 static struct ofoperation *
6078 ofoperation_create(struct ofopgroup *group, struct rule *rule,
6079 enum ofoperation_type type,
6080 enum ofp_flow_removed_reason reason)
6081 OVS_REQUIRES(ofproto_mutex)
6082 {
6083 struct ofproto *ofproto = group->ofproto;
6084 struct ofoperation *op;
6085
6086 ovs_assert(!rule->pending);
6087
6088 op = rule->pending = xzalloc(sizeof *op);
6089 op->group = group;
6090 list_push_back(&group->ops, &op->group_node);
6091 op->rule = rule;
6092 op->type = type;
6093 op->reason = reason;
6094 op->flow_cookie = rule->flow_cookie;
6095 ovs_mutex_lock(&rule->mutex);
6096 op->idle_timeout = rule->idle_timeout;
6097 op->hard_timeout = rule->hard_timeout;
6098 ovs_mutex_unlock(&rule->mutex);
6099 op->flags = rule->flags;
6100
6101 group->n_running++;
6102
6103 if (type == OFOPERATION_DELETE) {
6104 hmap_insert(&ofproto->deletions, &op->hmap_node,
6105 cls_rule_hash(&rule->cr, rule->table_id));
6106 }
6107
6108 return op;
6109 }
6110
6111 static void
6112 ofoperation_destroy(struct ofoperation *op)
6113 OVS_REQUIRES(ofproto_mutex)
6114 {
6115 struct ofopgroup *group = op->group;
6116
6117 if (op->rule) {
6118 op->rule->pending = NULL;
6119 }
6120 if (op->type == OFOPERATION_DELETE) {
6121 hmap_remove(&group->ofproto->deletions, &op->hmap_node);
6122 }
6123 list_remove(&op->group_node);
6124 rule_actions_unref(op->actions);
6125 free(op);
6126 }
6127
6128 /* Indicates that 'op' completed with status 'error', which is either 0 to
6129 * indicate success or an OpenFlow error code on failure.
6130 *
6131 * If 'error' is 0, indicating success, the operation will be committed
6132 * permanently to the flow table.
6133 *
6134 * If 'error' is nonzero, then generally the operation will be rolled back:
6135 *
6136 * - If 'op' is an "add flow" operation, ofproto removes the new rule or
6137 * restores the original rule. The caller must have uninitialized any
6138 * derived state in the new rule, as in step 5 of in the "Life Cycle" in
6139 * ofproto/ofproto-provider.h. ofoperation_complete() performs steps 6 and
6140 * and 7 for the new rule, calling its ->rule_dealloc() function.
6141 *
6142 * - If 'op' is a "modify flow" operation, ofproto restores the original
6143 * actions.
6144 *
6145 * - 'op' must not be a "delete flow" operation. Removing a rule is not
6146 * allowed to fail. It must always succeed.
6147 *
6148 * Please see the large comment in ofproto/ofproto-provider.h titled
6149 * "Asynchronous Operation Support" for more information. */
6150 void
6151 ofoperation_complete(struct ofoperation *op, enum ofperr error)
6152 {
6153 struct ofopgroup *group = op->group;
6154
6155 ovs_assert(group->n_running > 0);
6156 ovs_assert(!error || op->type != OFOPERATION_DELETE);
6157
6158 op->error = error;
6159 if (!--group->n_running && !list_is_empty(&group->ofproto_node)) {
6160 /* This function can be called from ->rule_construct(), in which case
6161 * ofproto_mutex is held, or it can be called from ->run(), in which
6162 * case ofproto_mutex is not held. But only in the latter case can we
6163 * arrive here, so we can safely take ofproto_mutex now. */
6164 ovs_mutex_lock(&ofproto_mutex);
6165 ovs_assert(op->rule->pending == op);
6166 ofopgroup_complete(group);
6167 ovs_mutex_unlock(&ofproto_mutex);
6168 }
6169 }
6170 \f
6171 static uint64_t
6172 pick_datapath_id(const struct ofproto *ofproto)
6173 {
6174 const struct ofport *port;
6175
6176 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6177 if (port) {
6178 uint8_t ea[ETH_ADDR_LEN];
6179 int error;
6180
6181 error = netdev_get_etheraddr(port->netdev, ea);
6182 if (!error) {
6183 return eth_addr_to_uint64(ea);
6184 }
6185 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6186 ofproto->name, netdev_get_name(port->netdev),
6187 ovs_strerror(error));
6188 }
6189 return ofproto->fallback_dpid;
6190 }
6191
6192 static uint64_t
6193 pick_fallback_dpid(void)
6194 {
6195 uint8_t ea[ETH_ADDR_LEN];
6196 eth_addr_nicira_random(ea);
6197 return eth_addr_to_uint64(ea);
6198 }
6199 \f
6200 /* Table overflow policy. */
6201
6202 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6203 * to NULL if the table is not configured to evict rules or if the table
6204 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6205 * or with no timeouts are not evictable.) */
6206 static bool
6207 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6208 OVS_REQUIRES(ofproto_mutex)
6209 {
6210 struct eviction_group *evg;
6211
6212 *rulep = NULL;
6213 if (!table->eviction_fields) {
6214 return false;
6215 }
6216
6217 /* In the common case, the outer and inner loops here will each be entered
6218 * exactly once:
6219 *
6220 * - The inner loop normally "return"s in its first iteration. If the
6221 * eviction group has any evictable rules, then it always returns in
6222 * some iteration.
6223 *
6224 * - The outer loop only iterates more than once if the largest eviction
6225 * group has no evictable rules.
6226 *
6227 * - The outer loop can exit only if table's 'max_flows' is all filled up
6228 * by unevictable rules. */
6229 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6230 struct rule *rule;
6231
6232 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6233 *rulep = rule;
6234 return true;
6235 }
6236 }
6237
6238 return false;
6239 }
6240
6241 /* Searches 'ofproto' for tables that have more flows than their configured
6242 * maximum and that have flow eviction enabled, and evicts as many flows as
6243 * necessary and currently feasible from them.
6244 *
6245 * This triggers only when an OpenFlow table has N flows in it and then the
6246 * client configures a maximum number of flows less than N. */
6247 static void
6248 ofproto_evict(struct ofproto *ofproto)
6249 {
6250 struct oftable *table;
6251
6252 ovs_mutex_lock(&ofproto_mutex);
6253 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
6254 evict_rules_from_table(ofproto, table, 0);
6255 }
6256 ovs_mutex_unlock(&ofproto_mutex);
6257 }
6258 \f
6259 /* Eviction groups. */
6260
6261 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6262 * rules. The priority contains low-order random bits to ensure that eviction
6263 * groups with the same number of rules are prioritized randomly. */
6264 static uint32_t
6265 eviction_group_priority(size_t n_rules)
6266 {
6267 uint16_t size = MIN(UINT16_MAX, n_rules);
6268 return (size << 16) | random_uint16();
6269 }
6270
6271 /* Updates 'evg', an eviction_group within 'table', following a change that
6272 * adds or removes rules in 'evg'. */
6273 static void
6274 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6275 OVS_REQUIRES(ofproto_mutex)
6276 {
6277 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6278 eviction_group_priority(heap_count(&evg->rules)));
6279 }
6280
6281 /* Destroys 'evg', an eviction_group within 'table':
6282 *
6283 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6284 * rules themselves, just removes them from the eviction group.)
6285 *
6286 * - Removes 'evg' from 'table'.
6287 *
6288 * - Frees 'evg'. */
6289 static void
6290 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6291 OVS_REQUIRES(ofproto_mutex)
6292 {
6293 while (!heap_is_empty(&evg->rules)) {
6294 struct rule *rule;
6295
6296 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6297 rule->eviction_group = NULL;
6298 }
6299 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6300 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6301 heap_destroy(&evg->rules);
6302 free(evg);
6303 }
6304
6305 /* Removes 'rule' from its eviction group, if any. */
6306 static void
6307 eviction_group_remove_rule(struct rule *rule)
6308 OVS_REQUIRES(ofproto_mutex)
6309 {
6310 if (rule->eviction_group) {
6311 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6312 struct eviction_group *evg = rule->eviction_group;
6313
6314 rule->eviction_group = NULL;
6315 heap_remove(&evg->rules, &rule->evg_node);
6316 if (heap_is_empty(&evg->rules)) {
6317 eviction_group_destroy(table, evg);
6318 } else {
6319 eviction_group_resized(table, evg);
6320 }
6321 }
6322 }
6323
6324 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6325 * returns the hash value. */
6326 static uint32_t
6327 eviction_group_hash_rule(struct rule *rule)
6328 OVS_REQUIRES(ofproto_mutex)
6329 {
6330 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6331 const struct mf_subfield *sf;
6332 struct flow flow;
6333 uint32_t hash;
6334
6335 hash = table->eviction_group_id_basis;
6336 miniflow_expand(&rule->cr.match.flow, &flow);
6337 for (sf = table->eviction_fields;
6338 sf < &table->eviction_fields[table->n_eviction_fields];
6339 sf++)
6340 {
6341 if (mf_are_prereqs_ok(sf->field, &flow)) {
6342 union mf_value value;
6343
6344 mf_get_value(sf->field, &flow, &value);
6345 if (sf->ofs) {
6346 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6347 }
6348 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6349 unsigned int start = sf->ofs + sf->n_bits;
6350 bitwise_zero(&value, sf->field->n_bytes, start,
6351 sf->field->n_bytes * 8 - start);
6352 }
6353 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6354 } else {
6355 hash = hash_int(hash, 0);
6356 }
6357 }
6358
6359 return hash;
6360 }
6361
6362 /* Returns an eviction group within 'table' with the given 'id', creating one
6363 * if necessary. */
6364 static struct eviction_group *
6365 eviction_group_find(struct oftable *table, uint32_t id)
6366 OVS_REQUIRES(ofproto_mutex)
6367 {
6368 struct eviction_group *evg;
6369
6370 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6371 return evg;
6372 }
6373
6374 evg = xmalloc(sizeof *evg);
6375 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6376 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6377 eviction_group_priority(0));
6378 heap_init(&evg->rules);
6379
6380 return evg;
6381 }
6382
6383 /* Returns an eviction priority for 'rule'. The return value should be
6384 * interpreted so that higher priorities make a rule more attractive candidates
6385 * for eviction. */
6386 static uint32_t
6387 rule_eviction_priority(struct rule *rule)
6388 OVS_REQUIRES(ofproto_mutex)
6389 {
6390 long long int hard_expiration;
6391 long long int idle_expiration;
6392 long long int expiration;
6393 uint32_t expiration_offset;
6394
6395 /* Calculate time of expiration. */
6396 ovs_mutex_lock(&rule->mutex);
6397 hard_expiration = (rule->hard_timeout
6398 ? rule->modified + rule->hard_timeout * 1000
6399 : LLONG_MAX);
6400 idle_expiration = (rule->idle_timeout
6401 ? rule->used + rule->idle_timeout * 1000
6402 : LLONG_MAX);
6403 expiration = MIN(hard_expiration, idle_expiration);
6404 ovs_mutex_unlock(&rule->mutex);
6405 if (expiration == LLONG_MAX) {
6406 return 0;
6407 }
6408
6409 /* Calculate the time of expiration as a number of (approximate) seconds
6410 * after program startup.
6411 *
6412 * This should work OK for program runs that last UINT32_MAX seconds or
6413 * less. Therefore, please restart OVS at least once every 136 years. */
6414 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6415
6416 /* Invert the expiration offset because we're using a max-heap. */
6417 return UINT32_MAX - expiration_offset;
6418 }
6419
6420 /* Adds 'rule' to an appropriate eviction group for its oftable's
6421 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6422 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6423 * own).
6424 *
6425 * The caller must ensure that 'rule' is not already in an eviction group. */
6426 static void
6427 eviction_group_add_rule(struct rule *rule)
6428 OVS_REQUIRES(ofproto_mutex)
6429 {
6430 struct ofproto *ofproto = rule->ofproto;
6431 struct oftable *table = &ofproto->tables[rule->table_id];
6432 bool has_timeout;
6433
6434 ovs_mutex_lock(&rule->mutex);
6435 has_timeout = rule->hard_timeout || rule->idle_timeout;
6436 ovs_mutex_unlock(&rule->mutex);
6437
6438 if (table->eviction_fields && has_timeout) {
6439 struct eviction_group *evg;
6440
6441 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6442
6443 rule->eviction_group = evg;
6444 heap_insert(&evg->rules, &rule->evg_node,
6445 rule_eviction_priority(rule));
6446 eviction_group_resized(table, evg);
6447 }
6448 }
6449 \f
6450 /* oftables. */
6451
6452 /* Initializes 'table'. */
6453 static void
6454 oftable_init(struct oftable *table)
6455 {
6456 memset(table, 0, sizeof *table);
6457 classifier_init(&table->cls);
6458 table->max_flows = UINT_MAX;
6459 }
6460
6461 /* Destroys 'table', including its classifier and eviction groups.
6462 *
6463 * The caller is responsible for freeing 'table' itself. */
6464 static void
6465 oftable_destroy(struct oftable *table)
6466 {
6467 ovs_rwlock_rdlock(&table->cls.rwlock);
6468 ovs_assert(classifier_is_empty(&table->cls));
6469 ovs_rwlock_unlock(&table->cls.rwlock);
6470 oftable_disable_eviction(table);
6471 classifier_destroy(&table->cls);
6472 free(table->name);
6473 }
6474
6475 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6476 * string, then 'table' will use its default name.
6477 *
6478 * This only affects the name exposed for a table exposed through the OpenFlow
6479 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6480 static void
6481 oftable_set_name(struct oftable *table, const char *name)
6482 {
6483 if (name && name[0]) {
6484 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6485 if (!table->name || strncmp(name, table->name, len)) {
6486 free(table->name);
6487 table->name = xmemdup0(name, len);
6488 }
6489 } else {
6490 free(table->name);
6491 table->name = NULL;
6492 }
6493 }
6494
6495 /* oftables support a choice of two policies when adding a rule would cause the
6496 * number of flows in the table to exceed the configured maximum number: either
6497 * they can refuse to add the new flow or they can evict some existing flow.
6498 * This function configures the former policy on 'table'. */
6499 static void
6500 oftable_disable_eviction(struct oftable *table)
6501 OVS_REQUIRES(ofproto_mutex)
6502 {
6503 if (table->eviction_fields) {
6504 struct eviction_group *evg, *next;
6505
6506 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6507 &table->eviction_groups_by_id) {
6508 eviction_group_destroy(table, evg);
6509 }
6510 hmap_destroy(&table->eviction_groups_by_id);
6511 heap_destroy(&table->eviction_groups_by_size);
6512
6513 free(table->eviction_fields);
6514 table->eviction_fields = NULL;
6515 table->n_eviction_fields = 0;
6516 }
6517 }
6518
6519 /* oftables support a choice of two policies when adding a rule would cause the
6520 * number of flows in the table to exceed the configured maximum number: either
6521 * they can refuse to add the new flow or they can evict some existing flow.
6522 * This function configures the latter policy on 'table', with fairness based
6523 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6524 * 'n_fields' as 0 disables fairness.) */
6525 static void
6526 oftable_enable_eviction(struct oftable *table,
6527 const struct mf_subfield *fields, size_t n_fields)
6528 OVS_REQUIRES(ofproto_mutex)
6529 {
6530 struct cls_cursor cursor;
6531 struct rule *rule;
6532
6533 if (table->eviction_fields
6534 && n_fields == table->n_eviction_fields
6535 && (!n_fields
6536 || !memcmp(fields, table->eviction_fields,
6537 n_fields * sizeof *fields))) {
6538 /* No change. */
6539 return;
6540 }
6541
6542 oftable_disable_eviction(table);
6543
6544 table->n_eviction_fields = n_fields;
6545 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6546
6547 table->eviction_group_id_basis = random_uint32();
6548 hmap_init(&table->eviction_groups_by_id);
6549 heap_init(&table->eviction_groups_by_size);
6550
6551 ovs_rwlock_rdlock(&table->cls.rwlock);
6552 cls_cursor_init(&cursor, &table->cls, NULL);
6553 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6554 eviction_group_add_rule(rule);
6555 }
6556 ovs_rwlock_unlock(&table->cls.rwlock);
6557 }
6558
6559 /* Removes 'rule' from the oftable that contains it. */
6560 static void
6561 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6562 OVS_REQUIRES(ofproto_mutex)
6563 {
6564 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6565
6566 ovs_rwlock_wrlock(&cls->rwlock);
6567 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6568 ovs_rwlock_unlock(&cls->rwlock);
6569
6570 cookies_remove(ofproto, rule);
6571
6572 eviction_group_remove_rule(rule);
6573 if (!list_is_empty(&rule->expirable)) {
6574 list_remove(&rule->expirable);
6575 }
6576 if (!list_is_empty(&rule->meter_list_node)) {
6577 list_remove(&rule->meter_list_node);
6578 list_init(&rule->meter_list_node);
6579 }
6580 }
6581
6582 static void
6583 oftable_remove_rule(struct rule *rule)
6584 OVS_REQUIRES(ofproto_mutex)
6585 {
6586 oftable_remove_rule__(rule->ofproto, rule);
6587 }
6588
6589 /* Inserts 'rule' into its oftable, which must not already contain any rule for
6590 * the same cls_rule. */
6591 static void
6592 oftable_insert_rule(struct rule *rule)
6593 OVS_REQUIRES(ofproto_mutex)
6594 {
6595 struct ofproto *ofproto = rule->ofproto;
6596 struct oftable *table = &ofproto->tables[rule->table_id];
6597 bool may_expire;
6598
6599 ovs_mutex_lock(&rule->mutex);
6600 may_expire = rule->hard_timeout || rule->idle_timeout;
6601 ovs_mutex_unlock(&rule->mutex);
6602
6603 if (may_expire) {
6604 list_insert(&ofproto->expirable, &rule->expirable);
6605 }
6606
6607 cookies_insert(ofproto, rule);
6608
6609 if (rule->actions->provider_meter_id != UINT32_MAX) {
6610 uint32_t meter_id = ofpacts_get_meter(rule->actions->ofpacts,
6611 rule->actions->ofpacts_len);
6612 struct meter *meter = ofproto->meters[meter_id];
6613 list_insert(&meter->rules, &rule->meter_list_node);
6614 }
6615 ovs_rwlock_wrlock(&table->cls.rwlock);
6616 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
6617 ovs_rwlock_unlock(&table->cls.rwlock);
6618 eviction_group_add_rule(rule);
6619 }
6620 \f
6621 /* unixctl commands. */
6622
6623 struct ofproto *
6624 ofproto_lookup(const char *name)
6625 {
6626 struct ofproto *ofproto;
6627
6628 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6629 &all_ofprotos) {
6630 if (!strcmp(ofproto->name, name)) {
6631 return ofproto;
6632 }
6633 }
6634 return NULL;
6635 }
6636
6637 static void
6638 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6639 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6640 {
6641 struct ofproto *ofproto;
6642 struct ds results;
6643
6644 ds_init(&results);
6645 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6646 ds_put_format(&results, "%s\n", ofproto->name);
6647 }
6648 unixctl_command_reply(conn, ds_cstr(&results));
6649 ds_destroy(&results);
6650 }
6651
6652 static void
6653 ofproto_unixctl_init(void)
6654 {
6655 static bool registered;
6656 if (registered) {
6657 return;
6658 }
6659 registered = true;
6660
6661 unixctl_command_register("ofproto/list", "", 0, 0,
6662 ofproto_unixctl_list, NULL);
6663 }
6664 \f
6665 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6666 *
6667 * This is deprecated. It is only for compatibility with broken device drivers
6668 * in old versions of Linux that do not properly support VLANs when VLAN
6669 * devices are not used. When broken device drivers are no longer in
6670 * widespread use, we will delete these interfaces. */
6671
6672 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6673 * (exactly) by an OpenFlow rule in 'ofproto'. */
6674 void
6675 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6676 {
6677 const struct oftable *oftable;
6678
6679 free(ofproto->vlan_bitmap);
6680 ofproto->vlan_bitmap = bitmap_allocate(4096);
6681 ofproto->vlans_changed = false;
6682
6683 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6684 const struct cls_table *table;
6685
6686 ovs_rwlock_rdlock(&oftable->cls.rwlock);
6687 HMAP_FOR_EACH (table, hmap_node, &oftable->cls.tables) {
6688 if (minimask_get_vid_mask(&table->mask) == VLAN_VID_MASK) {
6689 const struct cls_rule *rule;
6690
6691 HMAP_FOR_EACH (rule, hmap_node, &table->rules) {
6692 uint16_t vid = miniflow_get_vid(&rule->match.flow);
6693 bitmap_set1(vlan_bitmap, vid);
6694 bitmap_set1(ofproto->vlan_bitmap, vid);
6695 }
6696 }
6697 }
6698 ovs_rwlock_unlock(&oftable->cls.rwlock);
6699 }
6700 }
6701
6702 /* Returns true if new VLANs have come into use by the flow table since the
6703 * last call to ofproto_get_vlan_usage().
6704 *
6705 * We don't track when old VLANs stop being used. */
6706 bool
6707 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6708 {
6709 return ofproto->vlans_changed;
6710 }
6711
6712 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6713 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6714 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6715 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6716 * then the VLAN device is un-enslaved. */
6717 int
6718 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6719 ofp_port_t realdev_ofp_port, int vid)
6720 {
6721 struct ofport *ofport;
6722 int error;
6723
6724 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6725
6726 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6727 if (!ofport) {
6728 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6729 ofproto->name, vlandev_ofp_port);
6730 return EINVAL;
6731 }
6732
6733 if (!ofproto->ofproto_class->set_realdev) {
6734 if (!vlandev_ofp_port) {
6735 return 0;
6736 }
6737 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6738 return EOPNOTSUPP;
6739 }
6740
6741 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6742 if (error) {
6743 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6744 ofproto->name, vlandev_ofp_port,
6745 netdev_get_name(ofport->netdev), ovs_strerror(error));
6746 }
6747 return error;
6748 }