]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/ofproto.c
Implement learned flow deletion.
[mirror_ovs.git] / ofproto / ofproto.c
1 /*
2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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 "connectivity.h"
29 #include "connmgr.h"
30 #include "coverage.h"
31 #include "dynamic-string.h"
32 #include "hash.h"
33 #include "hmap.h"
34 #include "meta-flow.h"
35 #include "netdev.h"
36 #include "nx-match.h"
37 #include "ofp-actions.h"
38 #include "ofp-errors.h"
39 #include "ofp-msgs.h"
40 #include "ofp-print.h"
41 #include "ofp-util.h"
42 #include "ofpbuf.h"
43 #include "ofproto-provider.h"
44 #include "openflow/nicira-ext.h"
45 #include "openflow/openflow.h"
46 #include "ovs-rcu.h"
47 #include "packets.h"
48 #include "pinsched.h"
49 #include "pktbuf.h"
50 #include "poll-loop.h"
51 #include "random.h"
52 #include "seq.h"
53 #include "shash.h"
54 #include "simap.h"
55 #include "smap.h"
56 #include "sset.h"
57 #include "timeval.h"
58 #include "unaligned.h"
59 #include "unixctl.h"
60 #include "vlog.h"
61 #include "bundles.h"
62
63 VLOG_DEFINE_THIS_MODULE(ofproto);
64
65 COVERAGE_DEFINE(ofproto_flush);
66 COVERAGE_DEFINE(ofproto_packet_out);
67 COVERAGE_DEFINE(ofproto_queue_req);
68 COVERAGE_DEFINE(ofproto_recv_openflow);
69 COVERAGE_DEFINE(ofproto_reinit_ports);
70 COVERAGE_DEFINE(ofproto_update_port);
71
72 /* Default fields to use for prefix tries in each flow table, unless something
73 * else is configured. */
74 const enum mf_field_id default_prefix_fields[2] =
75 { MFF_IPV4_DST, MFF_IPV4_SRC };
76
77 /* oftable. */
78 static void oftable_init(struct oftable *);
79 static void oftable_destroy(struct oftable *);
80
81 static void oftable_set_name(struct oftable *, const char *name);
82
83 static enum ofperr evict_rules_from_table(struct oftable *,
84 unsigned int extra_space)
85 OVS_REQUIRES(ofproto_mutex);
86 static void oftable_disable_eviction(struct oftable *);
87 static void oftable_enable_eviction(struct oftable *,
88 const struct mf_subfield *fields,
89 size_t n_fields);
90
91 static void oftable_remove_rule(struct rule *rule) OVS_REQUIRES(ofproto_mutex);
92 static void oftable_remove_rule__(struct ofproto *, struct rule *)
93 OVS_REQUIRES(ofproto_mutex);
94
95 /* A set of rules within a single OpenFlow table (oftable) that have the same
96 * values for the oftable's eviction_fields. A rule to be evicted, when one is
97 * needed, is taken from the eviction group that contains the greatest number
98 * of rules.
99 *
100 * An oftable owns any number of eviction groups, each of which contains any
101 * number of rules.
102 *
103 * Membership in an eviction group is imprecise, based on the hash of the
104 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
105 * That is, if two rules have different eviction_fields, but those
106 * eviction_fields hash to the same value, then they will belong to the same
107 * eviction_group anyway.
108 *
109 * (When eviction is not enabled on an oftable, we don't track any eviction
110 * groups, to save time and space.) */
111 struct eviction_group {
112 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
113 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
114 struct heap rules; /* Contains "struct rule"s. */
115 };
116
117 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
118 OVS_REQUIRES(ofproto_mutex);
119 static uint32_t rule_eviction_priority(struct ofproto *ofproto, struct rule *)
120 OVS_REQUIRES(ofproto_mutex);;
121 static void eviction_group_add_rule(struct rule *)
122 OVS_REQUIRES(ofproto_mutex);
123 static void eviction_group_remove_rule(struct rule *)
124 OVS_REQUIRES(ofproto_mutex);
125
126 /* Criteria that flow_mod and other operations use for selecting rules on
127 * which to operate. */
128 struct rule_criteria {
129 /* An OpenFlow table or 255 for all tables. */
130 uint8_t table_id;
131
132 /* OpenFlow matching criteria. Interpreted different in "loose" way by
133 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
134 * defined in the OpenFlow spec. */
135 struct cls_rule cr;
136
137 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
138 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
139 * The rule will not be selected if M is 1 and B != C. */
140 ovs_be64 cookie;
141 ovs_be64 cookie_mask;
142
143 /* Selection based on actions within a rule:
144 *
145 * If out_port != OFPP_ANY, selects only rules that output to out_port.
146 * If out_group != OFPG_ALL, select only rules that output to out_group. */
147 ofp_port_t out_port;
148 uint32_t out_group;
149
150 /* If true, collects only rules that are modifiable. */
151 bool include_hidden;
152 bool include_readonly;
153 };
154
155 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
156 const struct match *match,
157 unsigned int priority,
158 ovs_be64 cookie, ovs_be64 cookie_mask,
159 ofp_port_t out_port, uint32_t out_group);
160 static void rule_criteria_require_rw(struct rule_criteria *,
161 bool can_write_readonly);
162 static void rule_criteria_destroy(struct rule_criteria *);
163
164 static enum ofperr collect_rules_loose(struct ofproto *,
165 const struct rule_criteria *,
166 struct rule_collection *);
167
168 /* A packet that needs to be passed to rule_execute().
169 *
170 * (We can't do this immediately from ofopgroup_complete() because that holds
171 * ofproto_mutex, which rule_execute() needs released.) */
172 struct rule_execute {
173 struct list list_node; /* In struct ofproto's "rule_executes" list. */
174 struct rule *rule; /* Owns a reference to the rule. */
175 ofp_port_t in_port;
176 struct ofpbuf *packet; /* Owns the packet. */
177 };
178
179 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
180 static void destroy_rule_executes(struct ofproto *);
181
182 struct learned_cookie {
183 union {
184 /* In struct ofproto's 'learned_cookies' hmap. */
185 struct hmap_node hmap_node OVS_GUARDED_BY(ofproto_mutex);
186
187 /* In 'dead_cookies' list when removed from hmap. */
188 struct list list_node;
189 } u;
190
191 /* Key. */
192 ovs_be64 cookie OVS_GUARDED_BY(ofproto_mutex);
193 uint8_t table_id OVS_GUARDED_BY(ofproto_mutex);
194
195 /* Number of references from "learn" actions.
196 *
197 * When this drops to 0, all of the flows in 'table_id' with the specified
198 * 'cookie' are deleted. */
199 int n OVS_GUARDED_BY(ofproto_mutex);
200 };
201
202 static const struct ofpact_learn *next_learn_with_delete(
203 const struct rule_actions *, const struct ofpact_learn *start);
204
205 static void learned_cookies_inc(struct ofproto *, const struct rule_actions *)
206 OVS_REQUIRES(ofproto_mutex);
207 static void learned_cookies_dec(struct ofproto *, const struct rule_actions *,
208 struct list *dead_cookies)
209 OVS_REQUIRES(ofproto_mutex);
210 static void learned_cookies_flush(struct ofproto *, struct list *dead_cookies)
211 OVS_REQUIRES(ofproto_mutex);
212
213 /* ofport. */
214 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
215 static void ofport_destroy(struct ofport *);
216
217 static void update_port(struct ofproto *, const char *devname);
218 static int init_ports(struct ofproto *);
219 static void reinit_ports(struct ofproto *);
220
221 static long long int ofport_get_usage(const struct ofproto *,
222 ofp_port_t ofp_port);
223 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
224 long long int last_used);
225 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
226
227 /* Ofport usage.
228 *
229 * Keeps track of the currently used and recently used ofport values and is
230 * used to prevent immediate recycling of ofport values. */
231 struct ofport_usage {
232 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
233 ofp_port_t ofp_port; /* OpenFlow port number. */
234 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
235 represents in-use ofports. */
236 };
237
238 /* rule. */
239 static void ofproto_rule_send_removed(struct rule *, uint8_t reason);
240 static bool rule_is_readonly(const struct rule *);
241
242 /* The source of a flow_mod request, in the code that processes flow_mods.
243 *
244 * A flow table modification request can be generated externally, via OpenFlow,
245 * or internally through a function call. This structure indicates the source
246 * of an OpenFlow-generated flow_mod. For an internal flow_mod, it isn't
247 * meaningful and thus supplied as NULL. */
248 struct flow_mod_requester {
249 struct ofconn *ofconn; /* Connection on which flow_mod arrived. */
250 ovs_be32 xid; /* OpenFlow xid of flow_mod request. */
251 };
252
253 /* OpenFlow. */
254 static enum ofperr add_flow(struct ofproto *, struct ofputil_flow_mod *,
255 const struct flow_mod_requester *);
256
257 static enum ofperr modify_flows__(struct ofproto *, struct ofputil_flow_mod *,
258 const struct rule_collection *,
259 const struct flow_mod_requester *);
260 static void delete_flows__(const struct rule_collection *,
261 enum ofp_flow_removed_reason,
262 const struct flow_mod_requester *)
263 OVS_REQUIRES(ofproto_mutex);
264
265 static enum ofperr send_buffered_packet(struct ofconn *, uint32_t buffer_id,
266 struct rule *)
267 OVS_REQUIRES(ofproto_mutex);
268
269 static bool ofproto_group_exists__(const struct ofproto *ofproto,
270 uint32_t group_id)
271 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
272 static bool ofproto_group_exists(const struct ofproto *ofproto,
273 uint32_t group_id)
274 OVS_EXCLUDED(ofproto->groups_rwlock);
275 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
276 static void handle_openflow(struct ofconn *, const struct ofpbuf *);
277 static enum ofperr handle_flow_mod__(struct ofproto *,
278 struct ofputil_flow_mod *,
279 const struct flow_mod_requester *)
280 OVS_EXCLUDED(ofproto_mutex);
281 static void calc_duration(long long int start, long long int now,
282 uint32_t *sec, uint32_t *nsec);
283
284 /* ofproto. */
285 static uint64_t pick_datapath_id(const struct ofproto *);
286 static uint64_t pick_fallback_dpid(void);
287 static void ofproto_destroy__(struct ofproto *);
288 static void update_mtu(struct ofproto *, struct ofport *);
289 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
290 static void meter_insert_rule(struct rule *);
291
292 /* unixctl. */
293 static void ofproto_unixctl_init(void);
294
295 /* All registered ofproto classes, in probe order. */
296 static const struct ofproto_class **ofproto_classes;
297 static size_t n_ofproto_classes;
298 static size_t allocated_ofproto_classes;
299
300 /* Global lock that protects all flow table operations. */
301 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
302
303 unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
304 unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT;
305
306 size_t n_handlers, n_revalidators;
307
308 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
309 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
310
311 /* Initial mappings of port to OpenFlow number mappings. */
312 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
313
314 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
315
316 /* The default value of true waits for flow restore. */
317 static bool flow_restore_wait = true;
318
319 /* Must be called to initialize the ofproto library.
320 *
321 * The caller may pass in 'iface_hints', which contains an shash of
322 * "iface_hint" elements indexed by the interface's name. The provider
323 * may use these hints to describe the startup configuration in order to
324 * reinitialize its state. The caller owns the provided data, so a
325 * provider will make copies of anything required. An ofproto provider
326 * will remove any existing state that is not described by the hint, and
327 * may choose to remove it all. */
328 void
329 ofproto_init(const struct shash *iface_hints)
330 {
331 struct shash_node *node;
332 size_t i;
333
334 ofproto_class_register(&ofproto_dpif_class);
335
336 /* Make a local copy, since we don't own 'iface_hints' elements. */
337 SHASH_FOR_EACH(node, iface_hints) {
338 const struct iface_hint *orig_hint = node->data;
339 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
340 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
341
342 new_hint->br_name = xstrdup(orig_hint->br_name);
343 new_hint->br_type = xstrdup(br_type);
344 new_hint->ofp_port = orig_hint->ofp_port;
345
346 shash_add(&init_ofp_ports, node->name, new_hint);
347 }
348
349 for (i = 0; i < n_ofproto_classes; i++) {
350 ofproto_classes[i]->init(&init_ofp_ports);
351 }
352 }
353
354 /* 'type' should be a normalized datapath type, as returned by
355 * ofproto_normalize_type(). Returns the corresponding ofproto_class
356 * structure, or a null pointer if there is none registered for 'type'. */
357 static const struct ofproto_class *
358 ofproto_class_find__(const char *type)
359 {
360 size_t i;
361
362 for (i = 0; i < n_ofproto_classes; i++) {
363 const struct ofproto_class *class = ofproto_classes[i];
364 struct sset types;
365 bool found;
366
367 sset_init(&types);
368 class->enumerate_types(&types);
369 found = sset_contains(&types, type);
370 sset_destroy(&types);
371
372 if (found) {
373 return class;
374 }
375 }
376 VLOG_WARN("unknown datapath type %s", type);
377 return NULL;
378 }
379
380 /* Registers a new ofproto class. After successful registration, new ofprotos
381 * of that type can be created using ofproto_create(). */
382 int
383 ofproto_class_register(const struct ofproto_class *new_class)
384 {
385 size_t i;
386
387 for (i = 0; i < n_ofproto_classes; i++) {
388 if (ofproto_classes[i] == new_class) {
389 return EEXIST;
390 }
391 }
392
393 if (n_ofproto_classes >= allocated_ofproto_classes) {
394 ofproto_classes = x2nrealloc(ofproto_classes,
395 &allocated_ofproto_classes,
396 sizeof *ofproto_classes);
397 }
398 ofproto_classes[n_ofproto_classes++] = new_class;
399 return 0;
400 }
401
402 /* Unregisters a datapath provider. 'type' must have been previously
403 * registered and not currently be in use by any ofprotos. After
404 * unregistration new datapaths of that type cannot be opened using
405 * ofproto_create(). */
406 int
407 ofproto_class_unregister(const struct ofproto_class *class)
408 {
409 size_t i;
410
411 for (i = 0; i < n_ofproto_classes; i++) {
412 if (ofproto_classes[i] == class) {
413 for (i++; i < n_ofproto_classes; i++) {
414 ofproto_classes[i - 1] = ofproto_classes[i];
415 }
416 n_ofproto_classes--;
417 return 0;
418 }
419 }
420 VLOG_WARN("attempted to unregister an ofproto class that is not "
421 "registered");
422 return EAFNOSUPPORT;
423 }
424
425 /* Clears 'types' and enumerates all registered ofproto types into it. The
426 * caller must first initialize the sset. */
427 void
428 ofproto_enumerate_types(struct sset *types)
429 {
430 size_t i;
431
432 sset_clear(types);
433 for (i = 0; i < n_ofproto_classes; i++) {
434 ofproto_classes[i]->enumerate_types(types);
435 }
436 }
437
438 /* Returns the fully spelled out name for the given ofproto 'type'.
439 *
440 * Normalized type string can be compared with strcmp(). Unnormalized type
441 * string might be the same even if they have different spellings. */
442 const char *
443 ofproto_normalize_type(const char *type)
444 {
445 return type && type[0] ? type : "system";
446 }
447
448 /* Clears 'names' and enumerates the names of all known created ofprotos with
449 * the given 'type'. The caller must first initialize the sset. Returns 0 if
450 * successful, otherwise a positive errno value.
451 *
452 * Some kinds of datapaths might not be practically enumerable. This is not
453 * considered an error. */
454 int
455 ofproto_enumerate_names(const char *type, struct sset *names)
456 {
457 const struct ofproto_class *class = ofproto_class_find__(type);
458 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
459 }
460
461 int
462 ofproto_create(const char *datapath_name, const char *datapath_type,
463 struct ofproto **ofprotop)
464 {
465 const struct ofproto_class *class;
466 struct ofproto *ofproto;
467 int error;
468 int i;
469
470 *ofprotop = NULL;
471
472 ofproto_unixctl_init();
473
474 datapath_type = ofproto_normalize_type(datapath_type);
475 class = ofproto_class_find__(datapath_type);
476 if (!class) {
477 VLOG_WARN("could not create datapath %s of unknown type %s",
478 datapath_name, datapath_type);
479 return EAFNOSUPPORT;
480 }
481
482 ofproto = class->alloc();
483 if (!ofproto) {
484 VLOG_ERR("failed to allocate datapath %s of type %s",
485 datapath_name, datapath_type);
486 return ENOMEM;
487 }
488
489 /* Initialize. */
490 ovs_mutex_lock(&ofproto_mutex);
491 memset(ofproto, 0, sizeof *ofproto);
492 ofproto->ofproto_class = class;
493 ofproto->name = xstrdup(datapath_name);
494 ofproto->type = xstrdup(datapath_type);
495 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
496 hash_string(ofproto->name, 0));
497 ofproto->datapath_id = 0;
498 ofproto->forward_bpdu = false;
499 ofproto->fallback_dpid = pick_fallback_dpid();
500 ofproto->mfr_desc = NULL;
501 ofproto->hw_desc = NULL;
502 ofproto->sw_desc = NULL;
503 ofproto->serial_desc = NULL;
504 ofproto->dp_desc = NULL;
505 ofproto->frag_handling = OFPC_FRAG_NORMAL;
506 hmap_init(&ofproto->ports);
507 hmap_init(&ofproto->ofport_usage);
508 shash_init(&ofproto->port_by_name);
509 simap_init(&ofproto->ofp_requests);
510 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
511 ofproto->eviction_group_timer = LLONG_MIN;
512 ofproto->tables = NULL;
513 ofproto->n_tables = 0;
514 hindex_init(&ofproto->cookies);
515 hmap_init(&ofproto->learned_cookies);
516 list_init(&ofproto->expirable);
517 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
518 guarded_list_init(&ofproto->rule_executes);
519 ofproto->vlan_bitmap = NULL;
520 ofproto->vlans_changed = false;
521 ofproto->min_mtu = INT_MAX;
522 ovs_rwlock_init(&ofproto->groups_rwlock);
523 hmap_init(&ofproto->groups);
524 ovs_mutex_unlock(&ofproto_mutex);
525 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
526 OFPGFC_SELECT_WEIGHT;
527 ofproto->ogf.max_groups[OFPGT11_ALL] = OFPG_MAX;
528 ofproto->ogf.max_groups[OFPGT11_SELECT] = OFPG_MAX;
529 ofproto->ogf.max_groups[OFPGT11_INDIRECT] = OFPG_MAX;
530 ofproto->ogf.max_groups[OFPGT11_FF] = OFPG_MAX;
531 ofproto->ogf.actions[0] =
532 (1 << OFPAT11_OUTPUT) |
533 (1 << OFPAT11_COPY_TTL_OUT) |
534 (1 << OFPAT11_COPY_TTL_IN) |
535 (1 << OFPAT11_SET_MPLS_TTL) |
536 (1 << OFPAT11_DEC_MPLS_TTL) |
537 (1 << OFPAT11_PUSH_VLAN) |
538 (1 << OFPAT11_POP_VLAN) |
539 (1 << OFPAT11_PUSH_MPLS) |
540 (1 << OFPAT11_POP_MPLS) |
541 (1 << OFPAT11_SET_QUEUE) |
542 (1 << OFPAT11_GROUP) |
543 (1 << OFPAT11_SET_NW_TTL) |
544 (1 << OFPAT11_DEC_NW_TTL) |
545 (1 << OFPAT12_SET_FIELD);
546 /* not supported:
547 * (1 << OFPAT13_PUSH_PBB) |
548 * (1 << OFPAT13_POP_PBB) */
549
550 error = ofproto->ofproto_class->construct(ofproto);
551 if (error) {
552 VLOG_ERR("failed to open datapath %s: %s",
553 datapath_name, ovs_strerror(error));
554 ofproto_destroy__(ofproto);
555 return error;
556 }
557
558 /* Check that hidden tables, if any, are at the end. */
559 ovs_assert(ofproto->n_tables);
560 for (i = 0; i + 1 < ofproto->n_tables; i++) {
561 enum oftable_flags flags = ofproto->tables[i].flags;
562 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
563
564 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
565 }
566
567 ofproto->datapath_id = pick_datapath_id(ofproto);
568 init_ports(ofproto);
569
570 /* Initialize meters table. */
571 if (ofproto->ofproto_class->meter_get_features) {
572 ofproto->ofproto_class->meter_get_features(ofproto,
573 &ofproto->meter_features);
574 } else {
575 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
576 }
577 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
578 * sizeof(struct meter *));
579
580 *ofprotop = ofproto;
581 return 0;
582 }
583
584 /* Must be called (only) by an ofproto implementation in its constructor
585 * function. See the large comment on 'construct' in struct ofproto_class for
586 * details. */
587 void
588 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
589 {
590 struct oftable *table;
591
592 ovs_assert(!ofproto->n_tables);
593 ovs_assert(n_tables >= 1 && n_tables <= 255);
594
595 ofproto->n_tables = n_tables;
596 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
597 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
598 oftable_init(table);
599 }
600 }
601
602 /* To be optionally called (only) by an ofproto implementation in its
603 * constructor function. See the large comment on 'construct' in struct
604 * ofproto_class for details.
605 *
606 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
607 * will then ensure that actions passed into the ofproto implementation will
608 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
609 * function is not called, there will be no such restriction.
610 *
611 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
612 * the 'max_ports' restriction. */
613 void
614 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
615 {
616 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
617 ofproto->max_ports = max_ports;
618 }
619
620 uint64_t
621 ofproto_get_datapath_id(const struct ofproto *ofproto)
622 {
623 return ofproto->datapath_id;
624 }
625
626 void
627 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
628 {
629 uint64_t old_dpid = p->datapath_id;
630 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
631 if (p->datapath_id != old_dpid) {
632 /* Force all active connections to reconnect, since there is no way to
633 * notify a controller that the datapath ID has changed. */
634 ofproto_reconnect_controllers(p);
635 }
636 }
637
638 void
639 ofproto_set_controllers(struct ofproto *p,
640 const struct ofproto_controller *controllers,
641 size_t n_controllers, uint32_t allowed_versions)
642 {
643 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
644 allowed_versions);
645 }
646
647 void
648 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
649 {
650 connmgr_set_fail_mode(p->connmgr, fail_mode);
651 }
652
653 /* Drops the connections between 'ofproto' and all of its controllers, forcing
654 * them to reconnect. */
655 void
656 ofproto_reconnect_controllers(struct ofproto *ofproto)
657 {
658 connmgr_reconnect(ofproto->connmgr);
659 }
660
661 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
662 * in-band control should guarantee access, in the same way that in-band
663 * control guarantees access to OpenFlow controllers. */
664 void
665 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
666 const struct sockaddr_in *extras, size_t n)
667 {
668 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
669 }
670
671 /* Sets the OpenFlow queue used by flows set up by in-band control on
672 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
673 * flows will use the default queue. */
674 void
675 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
676 {
677 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
678 }
679
680 /* Sets the number of flows at which eviction from the kernel flow table
681 * will occur. */
682 void
683 ofproto_set_flow_limit(unsigned limit)
684 {
685 ofproto_flow_limit = limit;
686 }
687
688 /* Sets the maximum idle time for flows in the datapath before they are
689 * expired. */
690 void
691 ofproto_set_max_idle(unsigned max_idle)
692 {
693 ofproto_max_idle = max_idle;
694 }
695
696 /* If forward_bpdu is true, the NORMAL action will forward frames with
697 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
698 * the NORMAL action will drop these frames. */
699 void
700 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
701 {
702 bool old_val = ofproto->forward_bpdu;
703 ofproto->forward_bpdu = forward_bpdu;
704 if (old_val != ofproto->forward_bpdu) {
705 if (ofproto->ofproto_class->forward_bpdu_changed) {
706 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
707 }
708 }
709 }
710
711 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
712 * 'idle_time', in seconds, and the maximum number of MAC table entries to
713 * 'max_entries'. */
714 void
715 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
716 size_t max_entries)
717 {
718 if (ofproto->ofproto_class->set_mac_table_config) {
719 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
720 max_entries);
721 }
722 }
723
724 void
725 ofproto_set_threads(int n_handlers_, int n_revalidators_)
726 {
727 int threads = MAX(count_cpu_cores(), 2);
728
729 n_revalidators = MAX(n_revalidators_, 0);
730 n_handlers = MAX(n_handlers_, 0);
731
732 if (!n_revalidators) {
733 n_revalidators = n_handlers
734 ? MAX(threads - (int) n_handlers, 1)
735 : threads / 4 + 1;
736 }
737
738 if (!n_handlers) {
739 n_handlers = MAX(threads - (int) n_revalidators, 1);
740 }
741 }
742
743 void
744 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
745 {
746 free(p->dp_desc);
747 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
748 }
749
750 int
751 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
752 {
753 return connmgr_set_snoops(ofproto->connmgr, snoops);
754 }
755
756 int
757 ofproto_set_netflow(struct ofproto *ofproto,
758 const struct netflow_options *nf_options)
759 {
760 if (nf_options && sset_is_empty(&nf_options->collectors)) {
761 nf_options = NULL;
762 }
763
764 if (ofproto->ofproto_class->set_netflow) {
765 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
766 } else {
767 return nf_options ? EOPNOTSUPP : 0;
768 }
769 }
770
771 int
772 ofproto_set_sflow(struct ofproto *ofproto,
773 const struct ofproto_sflow_options *oso)
774 {
775 if (oso && sset_is_empty(&oso->targets)) {
776 oso = NULL;
777 }
778
779 if (ofproto->ofproto_class->set_sflow) {
780 return ofproto->ofproto_class->set_sflow(ofproto, oso);
781 } else {
782 return oso ? EOPNOTSUPP : 0;
783 }
784 }
785
786 int
787 ofproto_set_ipfix(struct ofproto *ofproto,
788 const struct ofproto_ipfix_bridge_exporter_options *bo,
789 const struct ofproto_ipfix_flow_exporter_options *fo,
790 size_t n_fo)
791 {
792 if (ofproto->ofproto_class->set_ipfix) {
793 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
794 } else {
795 return (bo || fo) ? EOPNOTSUPP : 0;
796 }
797 }
798
799 void
800 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
801 {
802 flow_restore_wait = flow_restore_wait_db;
803 }
804
805 bool
806 ofproto_get_flow_restore_wait(void)
807 {
808 return flow_restore_wait;
809 }
810
811 \f
812 /* Spanning Tree Protocol (STP) configuration. */
813
814 /* Configures STP on 'ofproto' using the settings defined in 's'. If
815 * 's' is NULL, disables STP.
816 *
817 * Returns 0 if successful, otherwise a positive errno value. */
818 int
819 ofproto_set_stp(struct ofproto *ofproto,
820 const struct ofproto_stp_settings *s)
821 {
822 return (ofproto->ofproto_class->set_stp
823 ? ofproto->ofproto_class->set_stp(ofproto, s)
824 : EOPNOTSUPP);
825 }
826
827 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
828 * 'enabled' member of 's' is false, then the other members are not
829 * meaningful.
830 *
831 * Returns 0 if successful, otherwise a positive errno value. */
832 int
833 ofproto_get_stp_status(struct ofproto *ofproto,
834 struct ofproto_stp_status *s)
835 {
836 return (ofproto->ofproto_class->get_stp_status
837 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
838 : EOPNOTSUPP);
839 }
840
841 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
842 * in 's'. The caller is responsible for assigning STP port numbers
843 * (using the 'port_num' member in the range of 1 through 255, inclusive)
844 * and ensuring there are no duplicates. If the 's' is NULL, then STP
845 * is disabled on the port.
846 *
847 * Returns 0 if successful, otherwise a positive errno value.*/
848 int
849 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
850 const struct ofproto_port_stp_settings *s)
851 {
852 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
853 if (!ofport) {
854 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
855 ofproto->name, ofp_port);
856 return ENODEV;
857 }
858
859 return (ofproto->ofproto_class->set_stp_port
860 ? ofproto->ofproto_class->set_stp_port(ofport, s)
861 : EOPNOTSUPP);
862 }
863
864 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
865 * 's'. If the 'enabled' member in 's' is false, then the other members
866 * are not meaningful.
867 *
868 * Returns 0 if successful, otherwise a positive errno value.*/
869 int
870 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
871 struct ofproto_port_stp_status *s)
872 {
873 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
874 if (!ofport) {
875 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
876 "port %"PRIu16, ofproto->name, ofp_port);
877 return ENODEV;
878 }
879
880 return (ofproto->ofproto_class->get_stp_port_status
881 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
882 : EOPNOTSUPP);
883 }
884
885 /* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
886 * 's'. If the 'enabled' member in 's' is false, then the other members
887 * are not meaningful.
888 *
889 * Returns 0 if successful, otherwise a positive errno value.*/
890 int
891 ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
892 struct ofproto_port_stp_stats *s)
893 {
894 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
895 if (!ofport) {
896 VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
897 "port %"PRIu16, ofproto->name, ofp_port);
898 return ENODEV;
899 }
900
901 return (ofproto->ofproto_class->get_stp_port_stats
902 ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
903 : EOPNOTSUPP);
904 }
905 \f
906 /* Queue DSCP configuration. */
907
908 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
909 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
910 * to implement QoS. Instead, it is used to implement features which require
911 * knowledge of what queues exist on a port, and some basic information about
912 * them.
913 *
914 * Returns 0 if successful, otherwise a positive errno value. */
915 int
916 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
917 const struct ofproto_port_queue *queues,
918 size_t n_queues)
919 {
920 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
921
922 if (!ofport) {
923 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
924 ofproto->name, ofp_port);
925 return ENODEV;
926 }
927
928 return (ofproto->ofproto_class->set_queues
929 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
930 : EOPNOTSUPP);
931 }
932 \f
933 /* Connectivity Fault Management configuration. */
934
935 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
936 void
937 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
938 {
939 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
940 if (ofport && ofproto->ofproto_class->set_cfm) {
941 ofproto->ofproto_class->set_cfm(ofport, NULL);
942 }
943 }
944
945 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
946 * basic configuration from the configuration members in 'cfm', and the remote
947 * maintenance point ID from remote_mpid. Ignores the statistics members of
948 * 'cfm'.
949 *
950 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
951 void
952 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
953 const struct cfm_settings *s)
954 {
955 struct ofport *ofport;
956 int error;
957
958 ofport = ofproto_get_port(ofproto, ofp_port);
959 if (!ofport) {
960 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
961 ofproto->name, ofp_port);
962 return;
963 }
964
965 /* XXX: For configuration simplicity, we only support one remote_mpid
966 * outside of the CFM module. It's not clear if this is the correct long
967 * term solution or not. */
968 error = (ofproto->ofproto_class->set_cfm
969 ? ofproto->ofproto_class->set_cfm(ofport, s)
970 : EOPNOTSUPP);
971 if (error) {
972 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
973 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
974 ovs_strerror(error));
975 }
976 }
977
978 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
979 * 'ofproto' does not have a port 'ofp_port'. */
980 void
981 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
982 const struct smap *cfg)
983 {
984 struct ofport *ofport;
985 int error;
986
987 ofport = ofproto_get_port(ofproto, ofp_port);
988 if (!ofport) {
989 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
990 ofproto->name, ofp_port);
991 return;
992 }
993
994 error = (ofproto->ofproto_class->set_bfd
995 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
996 : EOPNOTSUPP);
997 if (error) {
998 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
999 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1000 ovs_strerror(error));
1001 }
1002 }
1003
1004 /* Populates 'status' with the status of BFD on 'ofport'. Returns 0 on
1005 * success. Returns a negative number if there is no status change since
1006 * last update. Returns a positive errno otherwise. Has no effect if
1007 * 'ofp_port' is not an OpenFlow port in 'ofproto'.
1008 *
1009 * The caller must provide and own '*status'. */
1010 int
1011 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1012 struct smap *status)
1013 {
1014 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1015 return (ofport && ofproto->ofproto_class->get_bfd_status
1016 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1017 : EOPNOTSUPP);
1018 }
1019
1020 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1021 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1022 * 0 if LACP partner information is not current (generally indicating a
1023 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1024 int
1025 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1026 {
1027 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1028 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1029 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1030 : -1);
1031 }
1032 \f
1033 /* Bundles. */
1034
1035 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1036 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1037 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1038 * configuration plus, if there is more than one slave, a bonding
1039 * configuration.
1040 *
1041 * If 'aux' is already registered then this function updates its configuration
1042 * to 's'. Otherwise, this function registers a new bundle.
1043 *
1044 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1045 * port. */
1046 int
1047 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1048 const struct ofproto_bundle_settings *s)
1049 {
1050 return (ofproto->ofproto_class->bundle_set
1051 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1052 : EOPNOTSUPP);
1053 }
1054
1055 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1056 * If no such bundle has been registered, this has no effect. */
1057 int
1058 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1059 {
1060 return ofproto_bundle_register(ofproto, aux, NULL);
1061 }
1062
1063 \f
1064 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1065 * If 'aux' is already registered then this function updates its configuration
1066 * to 's'. Otherwise, this function registers a new mirror. */
1067 int
1068 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1069 const struct ofproto_mirror_settings *s)
1070 {
1071 return (ofproto->ofproto_class->mirror_set
1072 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1073 : EOPNOTSUPP);
1074 }
1075
1076 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1077 * If no mirror has been registered, this has no effect. */
1078 int
1079 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1080 {
1081 return ofproto_mirror_register(ofproto, aux, NULL);
1082 }
1083
1084 /* Retrieves statistics from mirror associated with client data pointer
1085 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1086 * 'bytes', respectively. If a particular counters is not supported,
1087 * the appropriate argument is set to UINT64_MAX. */
1088 int
1089 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1090 uint64_t *packets, uint64_t *bytes)
1091 {
1092 if (!ofproto->ofproto_class->mirror_get_stats) {
1093 *packets = *bytes = UINT64_MAX;
1094 return EOPNOTSUPP;
1095 }
1096
1097 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1098 packets, bytes);
1099 }
1100
1101 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1102 * which all packets are flooded, instead of using MAC learning. If
1103 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1104 *
1105 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1106 * port. */
1107 int
1108 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1109 {
1110 return (ofproto->ofproto_class->set_flood_vlans
1111 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1112 : EOPNOTSUPP);
1113 }
1114
1115 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1116 * output for a mirror. */
1117 bool
1118 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1119 {
1120 return (ofproto->ofproto_class->is_mirror_output_bundle
1121 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1122 : false);
1123 }
1124 \f
1125 /* Configuration of OpenFlow tables. */
1126
1127 /* Returns the number of OpenFlow tables in 'ofproto'. */
1128 int
1129 ofproto_get_n_tables(const struct ofproto *ofproto)
1130 {
1131 return ofproto->n_tables;
1132 }
1133
1134 /* Returns the number of Controller visible OpenFlow tables
1135 * in 'ofproto'. This number will exclude Hidden tables.
1136 * This funtion's return value should be less or equal to that of
1137 * ofproto_get_n_tables() . */
1138 uint8_t
1139 ofproto_get_n_visible_tables(const struct ofproto *ofproto)
1140 {
1141 uint8_t n = ofproto->n_tables;
1142
1143 /* Count only non-hidden tables in the number of tables. (Hidden tables,
1144 * if present, are always at the end.) */
1145 while(n && (ofproto->tables[n - 1].flags & OFTABLE_HIDDEN)) {
1146 n--;
1147 }
1148
1149 return n;
1150 }
1151
1152 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1153 * settings from 's'. 'table_id' must be in the range 0 through the number of
1154 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1155 *
1156 * For read-only tables, only the name may be configured. */
1157 void
1158 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1159 const struct ofproto_table_settings *s)
1160 {
1161 struct oftable *table;
1162
1163 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1164 table = &ofproto->tables[table_id];
1165
1166 oftable_set_name(table, s->name);
1167
1168 if (table->flags & OFTABLE_READONLY) {
1169 return;
1170 }
1171
1172 if (s->groups) {
1173 oftable_enable_eviction(table, s->groups, s->n_groups);
1174 } else {
1175 oftable_disable_eviction(table);
1176 }
1177
1178 table->max_flows = s->max_flows;
1179 fat_rwlock_wrlock(&table->cls.rwlock);
1180 classifier_set_prefix_fields(&table->cls,
1181 s->prefix_fields, s->n_prefix_fields);
1182 fat_rwlock_unlock(&table->cls.rwlock);
1183
1184 ovs_mutex_lock(&ofproto_mutex);
1185 evict_rules_from_table(table, 0);
1186 ovs_mutex_unlock(&ofproto_mutex);
1187 }
1188 \f
1189 bool
1190 ofproto_has_snoops(const struct ofproto *ofproto)
1191 {
1192 return connmgr_has_snoops(ofproto->connmgr);
1193 }
1194
1195 void
1196 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1197 {
1198 connmgr_get_snoops(ofproto->connmgr, snoops);
1199 }
1200
1201 static void
1202 ofproto_rule_delete__(struct rule *rule, uint8_t reason)
1203 OVS_REQUIRES(ofproto_mutex)
1204 {
1205 struct rule_collection rules;
1206
1207 rules.rules = rules.stub;
1208 rules.n = 1;
1209 rules.stub[0] = rule;
1210 delete_flows__(&rules, reason, NULL);
1211 }
1212
1213 /* Deletes 'rule' from 'ofproto'.
1214 *
1215 * Within an ofproto implementation, this function allows an ofproto
1216 * implementation to destroy any rules that remain when its ->destruct()
1217 * function is called. This function is not suitable for use elsewhere in an
1218 * ofproto implementation.
1219 *
1220 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1221 * Cycle" in ofproto-provider.h. */
1222 void
1223 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1224 OVS_EXCLUDED(ofproto_mutex)
1225 {
1226 /* This skips the ofmonitor and flow-removed notifications because the
1227 * switch is being deleted and any OpenFlow channels have been or soon will
1228 * be killed. */
1229 ovs_mutex_lock(&ofproto_mutex);
1230 oftable_remove_rule__(ofproto, rule);
1231 ofproto->ofproto_class->rule_delete(rule);
1232 ovs_mutex_unlock(&ofproto_mutex);
1233 }
1234
1235 static void
1236 ofproto_flush__(struct ofproto *ofproto)
1237 OVS_EXCLUDED(ofproto_mutex)
1238 {
1239 struct oftable *table;
1240
1241 if (ofproto->ofproto_class->flush) {
1242 ofproto->ofproto_class->flush(ofproto);
1243 }
1244
1245 ovs_mutex_lock(&ofproto_mutex);
1246 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1247 struct rule *rule, *next_rule;
1248 struct cls_cursor cursor;
1249
1250 if (table->flags & OFTABLE_HIDDEN) {
1251 continue;
1252 }
1253
1254 fat_rwlock_rdlock(&table->cls.rwlock);
1255 cls_cursor_init(&cursor, &table->cls, NULL);
1256 fat_rwlock_unlock(&table->cls.rwlock);
1257 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, cr, &cursor) {
1258 ofproto_rule_delete__(rule, OFPRR_DELETE);
1259 }
1260 }
1261 ovs_mutex_unlock(&ofproto_mutex);
1262 }
1263
1264 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1265
1266 static void
1267 ofproto_destroy__(struct ofproto *ofproto)
1268 OVS_EXCLUDED(ofproto_mutex)
1269 {
1270 struct oftable *table;
1271
1272 destroy_rule_executes(ofproto);
1273 delete_group(ofproto, OFPG_ALL);
1274
1275 guarded_list_destroy(&ofproto->rule_executes);
1276 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1277 hmap_destroy(&ofproto->groups);
1278
1279 connmgr_destroy(ofproto->connmgr);
1280
1281 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1282 free(ofproto->name);
1283 free(ofproto->type);
1284 free(ofproto->mfr_desc);
1285 free(ofproto->hw_desc);
1286 free(ofproto->sw_desc);
1287 free(ofproto->serial_desc);
1288 free(ofproto->dp_desc);
1289 hmap_destroy(&ofproto->ports);
1290 hmap_destroy(&ofproto->ofport_usage);
1291 shash_destroy(&ofproto->port_by_name);
1292 simap_destroy(&ofproto->ofp_requests);
1293
1294 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1295 oftable_destroy(table);
1296 }
1297 free(ofproto->tables);
1298
1299 ovs_assert(hindex_is_empty(&ofproto->cookies));
1300 hindex_destroy(&ofproto->cookies);
1301
1302 ovs_assert(hmap_is_empty(&ofproto->learned_cookies));
1303 hmap_destroy(&ofproto->learned_cookies);
1304
1305 free(ofproto->vlan_bitmap);
1306
1307 ofproto->ofproto_class->dealloc(ofproto);
1308 }
1309
1310 void
1311 ofproto_destroy(struct ofproto *p)
1312 OVS_EXCLUDED(ofproto_mutex)
1313 {
1314 struct ofport *ofport, *next_ofport;
1315 struct ofport_usage *usage, *next_usage;
1316
1317 if (!p) {
1318 return;
1319 }
1320
1321 if (p->meters) {
1322 meter_delete(p, 1, p->meter_features.max_meters);
1323 p->meter_features.max_meters = 0;
1324 free(p->meters);
1325 p->meters = NULL;
1326 }
1327
1328 ofproto_flush__(p);
1329 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1330 ofport_destroy(ofport);
1331 }
1332
1333 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1334 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1335 free(usage);
1336 }
1337
1338 p->ofproto_class->destruct(p);
1339 /* Destroying rules is deferred, must have 'ofproto' around for them. */
1340 ovsrcu_postpone(ofproto_destroy__, p);
1341 }
1342
1343 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1344 * kernel datapath, for example, this destroys the datapath in the kernel, and
1345 * with the netdev-based datapath, it tears down the data structures that
1346 * represent the datapath.
1347 *
1348 * The datapath should not be currently open as an ofproto. */
1349 int
1350 ofproto_delete(const char *name, const char *type)
1351 {
1352 const struct ofproto_class *class = ofproto_class_find__(type);
1353 return (!class ? EAFNOSUPPORT
1354 : !class->del ? EACCES
1355 : class->del(type, name));
1356 }
1357
1358 static void
1359 process_port_change(struct ofproto *ofproto, int error, char *devname)
1360 {
1361 if (error == ENOBUFS) {
1362 reinit_ports(ofproto);
1363 } else if (!error) {
1364 update_port(ofproto, devname);
1365 free(devname);
1366 }
1367 }
1368
1369 int
1370 ofproto_type_run(const char *datapath_type)
1371 {
1372 const struct ofproto_class *class;
1373 int error;
1374
1375 datapath_type = ofproto_normalize_type(datapath_type);
1376 class = ofproto_class_find__(datapath_type);
1377
1378 error = class->type_run ? class->type_run(datapath_type) : 0;
1379 if (error && error != EAGAIN) {
1380 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1381 datapath_type, ovs_strerror(error));
1382 }
1383 return error;
1384 }
1385
1386 void
1387 ofproto_type_wait(const char *datapath_type)
1388 {
1389 const struct ofproto_class *class;
1390
1391 datapath_type = ofproto_normalize_type(datapath_type);
1392 class = ofproto_class_find__(datapath_type);
1393
1394 if (class->type_wait) {
1395 class->type_wait(datapath_type);
1396 }
1397 }
1398
1399 int
1400 ofproto_run(struct ofproto *p)
1401 {
1402 int error;
1403 uint64_t new_seq;
1404
1405 error = p->ofproto_class->run(p);
1406 if (error && error != EAGAIN) {
1407 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1408 }
1409
1410 run_rule_executes(p);
1411
1412 /* Restore the eviction group heap invariant occasionally. */
1413 if (p->eviction_group_timer < time_msec()) {
1414 size_t i;
1415
1416 p->eviction_group_timer = time_msec() + 1000;
1417
1418 for (i = 0; i < p->n_tables; i++) {
1419 struct oftable *table = &p->tables[i];
1420 struct eviction_group *evg;
1421 struct cls_cursor cursor;
1422 struct rule *rule;
1423
1424 if (!table->eviction_fields) {
1425 continue;
1426 }
1427
1428 ovs_mutex_lock(&ofproto_mutex);
1429 fat_rwlock_rdlock(&table->cls.rwlock);
1430 cls_cursor_init(&cursor, &table->cls, NULL);
1431 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
1432 if (rule->idle_timeout || rule->hard_timeout) {
1433 if (!rule->eviction_group) {
1434 eviction_group_add_rule(rule);
1435 } else {
1436 heap_raw_change(&rule->evg_node,
1437 rule_eviction_priority(p, rule));
1438 }
1439 }
1440 }
1441 fat_rwlock_unlock(&table->cls.rwlock);
1442
1443 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1444 heap_rebuild(&evg->rules);
1445 }
1446 ovs_mutex_unlock(&ofproto_mutex);
1447 }
1448 }
1449
1450 if (p->ofproto_class->port_poll) {
1451 char *devname;
1452
1453 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1454 process_port_change(p, error, devname);
1455 }
1456 }
1457
1458 new_seq = seq_read(connectivity_seq_get());
1459 if (new_seq != p->change_seq) {
1460 struct sset devnames;
1461 const char *devname;
1462 struct ofport *ofport;
1463
1464 /* Update OpenFlow port status for any port whose netdev has changed.
1465 *
1466 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1467 * destroyed, so it's not safe to update ports directly from the
1468 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1469 * need this two-phase approach. */
1470 sset_init(&devnames);
1471 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1472 uint64_t port_change_seq;
1473
1474 port_change_seq = netdev_get_change_seq(ofport->netdev);
1475 if (ofport->change_seq != port_change_seq) {
1476 ofport->change_seq = port_change_seq;
1477 sset_add(&devnames, netdev_get_name(ofport->netdev));
1478 }
1479 }
1480 SSET_FOR_EACH (devname, &devnames) {
1481 update_port(p, devname);
1482 }
1483 sset_destroy(&devnames);
1484
1485 p->change_seq = new_seq;
1486 }
1487
1488 connmgr_run(p->connmgr, handle_openflow);
1489
1490 return error;
1491 }
1492
1493 void
1494 ofproto_wait(struct ofproto *p)
1495 {
1496 p->ofproto_class->wait(p);
1497 if (p->ofproto_class->port_poll_wait) {
1498 p->ofproto_class->port_poll_wait(p);
1499 }
1500 seq_wait(connectivity_seq_get(), p->change_seq);
1501 connmgr_wait(p->connmgr);
1502 }
1503
1504 bool
1505 ofproto_is_alive(const struct ofproto *p)
1506 {
1507 return connmgr_has_controllers(p->connmgr);
1508 }
1509
1510 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1511 * memory_report(). */
1512 void
1513 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1514 {
1515 const struct oftable *table;
1516 unsigned int n_rules;
1517
1518 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1519
1520 n_rules = 0;
1521 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1522 fat_rwlock_rdlock(&table->cls.rwlock);
1523 n_rules += classifier_count(&table->cls);
1524 fat_rwlock_unlock(&table->cls.rwlock);
1525 }
1526 simap_increase(usage, "rules", n_rules);
1527
1528 if (ofproto->ofproto_class->get_memory_usage) {
1529 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1530 }
1531
1532 connmgr_get_memory_usage(ofproto->connmgr, usage);
1533 }
1534
1535 void
1536 ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
1537 {
1538 const struct ofproto_class *class;
1539
1540 datapath_type = ofproto_normalize_type(datapath_type);
1541 class = ofproto_class_find__(datapath_type);
1542
1543 if (class && class->type_get_memory_usage) {
1544 class->type_get_memory_usage(datapath_type, usage);
1545 }
1546 }
1547
1548 void
1549 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1550 struct shash *info)
1551 {
1552 connmgr_get_controller_info(ofproto->connmgr, info);
1553 }
1554
1555 void
1556 ofproto_free_ofproto_controller_info(struct shash *info)
1557 {
1558 connmgr_free_controller_info(info);
1559 }
1560
1561 /* Makes a deep copy of 'old' into 'port'. */
1562 void
1563 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1564 {
1565 port->name = xstrdup(old->name);
1566 port->type = xstrdup(old->type);
1567 port->ofp_port = old->ofp_port;
1568 }
1569
1570 /* Frees memory allocated to members of 'ofproto_port'.
1571 *
1572 * Do not call this function on an ofproto_port obtained from
1573 * ofproto_port_dump_next(): that function retains ownership of the data in the
1574 * ofproto_port. */
1575 void
1576 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1577 {
1578 free(ofproto_port->name);
1579 free(ofproto_port->type);
1580 }
1581
1582 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1583 *
1584 * This function provides no status indication. An error status for the entire
1585 * dump operation is provided when it is completed by calling
1586 * ofproto_port_dump_done().
1587 */
1588 void
1589 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1590 const struct ofproto *ofproto)
1591 {
1592 dump->ofproto = ofproto;
1593 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1594 &dump->state);
1595 }
1596
1597 /* Attempts to retrieve another port from 'dump', which must have been created
1598 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1599 * 'port' and returns true. On failure, returns false.
1600 *
1601 * Failure might indicate an actual error or merely that the last port has been
1602 * dumped. An error status for the entire dump operation is provided when it
1603 * is completed by calling ofproto_port_dump_done().
1604 *
1605 * The ofproto owns the data stored in 'port'. It will remain valid until at
1606 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1607 * ofproto_port_dump_done(). */
1608 bool
1609 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1610 struct ofproto_port *port)
1611 {
1612 const struct ofproto *ofproto = dump->ofproto;
1613
1614 if (dump->error) {
1615 return false;
1616 }
1617
1618 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1619 port);
1620 if (dump->error) {
1621 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1622 return false;
1623 }
1624 return true;
1625 }
1626
1627 /* Completes port table dump operation 'dump', which must have been created
1628 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1629 * error-free, otherwise a positive errno value describing the problem. */
1630 int
1631 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1632 {
1633 const struct ofproto *ofproto = dump->ofproto;
1634 if (!dump->error) {
1635 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1636 dump->state);
1637 }
1638 return dump->error == EOF ? 0 : dump->error;
1639 }
1640
1641 /* Returns the type to pass to netdev_open() when a datapath of type
1642 * 'datapath_type' has a port of type 'port_type', for a few special
1643 * cases when a netdev type differs from a port type. For example, when
1644 * using the userspace datapath, a port of type "internal" needs to be
1645 * opened as "tap".
1646 *
1647 * Returns either 'type' itself or a string literal, which must not be
1648 * freed. */
1649 const char *
1650 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1651 {
1652 const struct ofproto_class *class;
1653
1654 datapath_type = ofproto_normalize_type(datapath_type);
1655 class = ofproto_class_find__(datapath_type);
1656 if (!class) {
1657 return port_type;
1658 }
1659
1660 return (class->port_open_type
1661 ? class->port_open_type(datapath_type, port_type)
1662 : port_type);
1663 }
1664
1665 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1666 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1667 * the port's OpenFlow port number.
1668 *
1669 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1670 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1671 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1672 * 'ofp_portp' is non-null). */
1673 int
1674 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1675 ofp_port_t *ofp_portp)
1676 {
1677 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1678 int error;
1679
1680 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1681 if (!error) {
1682 const char *netdev_name = netdev_get_name(netdev);
1683
1684 simap_put(&ofproto->ofp_requests, netdev_name,
1685 ofp_to_u16(ofp_port));
1686 update_port(ofproto, netdev_name);
1687 }
1688 if (ofp_portp) {
1689 *ofp_portp = OFPP_NONE;
1690 if (!error) {
1691 struct ofproto_port ofproto_port;
1692
1693 error = ofproto_port_query_by_name(ofproto,
1694 netdev_get_name(netdev),
1695 &ofproto_port);
1696 if (!error) {
1697 *ofp_portp = ofproto_port.ofp_port;
1698 ofproto_port_destroy(&ofproto_port);
1699 }
1700 }
1701 }
1702 return error;
1703 }
1704
1705 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1706 * initializes '*port' appropriately; on failure, returns a positive errno
1707 * value.
1708 *
1709 * The caller owns the data in 'ofproto_port' and must free it with
1710 * ofproto_port_destroy() when it is no longer needed. */
1711 int
1712 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1713 struct ofproto_port *port)
1714 {
1715 int error;
1716
1717 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1718 if (error) {
1719 memset(port, 0, sizeof *port);
1720 }
1721 return error;
1722 }
1723
1724 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1725 * Returns 0 if successful, otherwise a positive errno. */
1726 int
1727 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
1728 {
1729 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1730 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
1731 struct simap_node *ofp_request_node;
1732 int error;
1733
1734 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
1735 if (ofp_request_node) {
1736 simap_delete(&ofproto->ofp_requests, ofp_request_node);
1737 }
1738
1739 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
1740 if (!error && ofport) {
1741 /* 'name' is the netdev's name and update_port() is going to close the
1742 * netdev. Just in case update_port() refers to 'name' after it
1743 * destroys 'ofport', make a copy of it around the update_port()
1744 * call. */
1745 char *devname = xstrdup(name);
1746 update_port(ofproto, devname);
1747 free(devname);
1748 }
1749 return error;
1750 }
1751
1752 static void
1753 flow_mod_init(struct ofputil_flow_mod *fm,
1754 const struct match *match, unsigned int priority,
1755 const struct ofpact *ofpacts, size_t ofpacts_len,
1756 enum ofp_flow_mod_command command)
1757 {
1758 memset(fm, 0, sizeof *fm);
1759 fm->match = *match;
1760 fm->priority = priority;
1761 fm->cookie = 0;
1762 fm->new_cookie = 0;
1763 fm->modify_cookie = false;
1764 fm->table_id = 0;
1765 fm->command = command;
1766 fm->idle_timeout = 0;
1767 fm->hard_timeout = 0;
1768 fm->buffer_id = UINT32_MAX;
1769 fm->out_port = OFPP_ANY;
1770 fm->out_group = OFPG_ANY;
1771 fm->flags = 0;
1772 fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
1773 fm->ofpacts_len = ofpacts_len;
1774 fm->delete_reason = OFPRR_DELETE;
1775 }
1776
1777 static int
1778 simple_flow_mod(struct ofproto *ofproto,
1779 const struct match *match, unsigned int priority,
1780 const struct ofpact *ofpacts, size_t ofpacts_len,
1781 enum ofp_flow_mod_command command)
1782 {
1783 struct ofputil_flow_mod fm;
1784
1785 flow_mod_init(&fm, match, priority, ofpacts, ofpacts_len, command);
1786
1787 return handle_flow_mod__(ofproto, &fm, NULL);
1788 }
1789
1790 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
1791 * performs the 'n_actions' actions in 'actions'. The new flow will not
1792 * timeout.
1793 *
1794 * If cls_rule->priority is in the range of priorities supported by OpenFlow
1795 * (0...65535, inclusive) then the flow will be visible to OpenFlow
1796 * controllers; otherwise, it will be hidden.
1797 *
1798 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
1799 *
1800 * This is a helper function for in-band control and fail-open. */
1801 void
1802 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
1803 unsigned int priority,
1804 const struct ofpact *ofpacts, size_t ofpacts_len)
1805 OVS_EXCLUDED(ofproto_mutex)
1806 {
1807 const struct rule *rule;
1808 bool must_add;
1809
1810 /* First do a cheap check whether the rule we're looking for already exists
1811 * with the actions that we want. If it does, then we're done. */
1812 fat_rwlock_rdlock(&ofproto->tables[0].cls.rwlock);
1813 rule = rule_from_cls_rule(classifier_find_match_exactly(
1814 &ofproto->tables[0].cls, match, priority));
1815 if (rule) {
1816 const struct rule_actions *actions = rule_get_actions(rule);
1817 must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len,
1818 ofpacts, ofpacts_len);
1819 } else {
1820 must_add = true;
1821 }
1822 fat_rwlock_unlock(&ofproto->tables[0].cls.rwlock);
1823
1824 /* If there's no such rule or the rule doesn't have the actions we want,
1825 * fall back to a executing a full flow mod. We can't optimize this at
1826 * all because we didn't take enough locks above to ensure that the flow
1827 * table didn't already change beneath us. */
1828 if (must_add) {
1829 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
1830 OFPFC_MODIFY_STRICT);
1831 }
1832 }
1833
1834 /* Executes the flow modification specified in 'fm'. Returns 0 on success, an
1835 * OFPERR_* OpenFlow error code on failure, or OFPROTO_POSTPONE if the
1836 * operation cannot be initiated now but may be retried later.
1837 *
1838 * This is a helper function for in-band control and fail-open and the "learn"
1839 * action. */
1840 int
1841 ofproto_flow_mod(struct ofproto *ofproto, struct ofputil_flow_mod *fm)
1842 OVS_EXCLUDED(ofproto_mutex)
1843 {
1844 /* Optimize for the most common case of a repeated learn action.
1845 * If an identical flow already exists we only need to update its
1846 * 'modified' time. */
1847 if (fm->command == OFPFC_MODIFY_STRICT && fm->table_id != OFPTT_ALL
1848 && !(fm->flags & OFPUTIL_FF_RESET_COUNTS)) {
1849 struct oftable *table = &ofproto->tables[fm->table_id];
1850 struct rule *rule;
1851 bool done = false;
1852
1853 fat_rwlock_rdlock(&table->cls.rwlock);
1854 rule = rule_from_cls_rule(classifier_find_match_exactly(&table->cls,
1855 &fm->match,
1856 fm->priority));
1857 if (rule) {
1858 /* Reading many of the rule fields and writing on 'modified'
1859 * requires the rule->mutex. Also, rule->actions may change
1860 * if rule->mutex is not held. */
1861 const struct rule_actions *actions;
1862
1863 ovs_mutex_lock(&rule->mutex);
1864 actions = rule_get_actions(rule);
1865 if (rule->idle_timeout == fm->idle_timeout
1866 && rule->hard_timeout == fm->hard_timeout
1867 && rule->flags == (fm->flags & OFPUTIL_FF_STATE)
1868 && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie))
1869 && ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
1870 actions->ofpacts, actions->ofpacts_len)) {
1871 /* Rule already exists and need not change, only update the
1872 modified timestamp. */
1873 rule->modified = time_msec();
1874 done = true;
1875 }
1876 ovs_mutex_unlock(&rule->mutex);
1877 }
1878 fat_rwlock_unlock(&table->cls.rwlock);
1879
1880 if (done) {
1881 return 0;
1882 }
1883 }
1884
1885 return handle_flow_mod__(ofproto, fm, NULL);
1886 }
1887
1888 /* Searches for a rule with matching criteria exactly equal to 'target' in
1889 * ofproto's table 0 and, if it finds one, deletes it.
1890 *
1891 * This is a helper function for in-band control and fail-open. */
1892 void
1893 ofproto_delete_flow(struct ofproto *ofproto,
1894 const struct match *target, unsigned int priority)
1895 OVS_EXCLUDED(ofproto_mutex)
1896 {
1897 struct classifier *cls = &ofproto->tables[0].cls;
1898 struct rule *rule;
1899
1900 /* First do a cheap check whether the rule we're looking for has already
1901 * been deleted. If so, then we're done. */
1902 fat_rwlock_rdlock(&cls->rwlock);
1903 rule = rule_from_cls_rule(classifier_find_match_exactly(cls, target,
1904 priority));
1905 fat_rwlock_unlock(&cls->rwlock);
1906 if (!rule) {
1907 return;
1908 }
1909
1910 /* Execute a flow mod. We can't optimize this at all because we didn't
1911 * take enough locks above to ensure that the flow table didn't already
1912 * change beneath us. */
1913 simple_flow_mod(ofproto, target, priority, NULL, 0, OFPFC_DELETE_STRICT);
1914 }
1915
1916 /* Delete all of the flows from all of ofproto's flow tables, then reintroduce
1917 * the flows required by in-band control and fail-open. */
1918 void
1919 ofproto_flush_flows(struct ofproto *ofproto)
1920 {
1921 COVERAGE_INC(ofproto_flush);
1922 ofproto_flush__(ofproto);
1923 connmgr_flushed(ofproto->connmgr);
1924 }
1925 \f
1926 static void
1927 reinit_ports(struct ofproto *p)
1928 {
1929 struct ofproto_port_dump dump;
1930 struct sset devnames;
1931 struct ofport *ofport;
1932 struct ofproto_port ofproto_port;
1933 const char *devname;
1934
1935 COVERAGE_INC(ofproto_reinit_ports);
1936
1937 sset_init(&devnames);
1938 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1939 sset_add(&devnames, netdev_get_name(ofport->netdev));
1940 }
1941 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
1942 sset_add(&devnames, ofproto_port.name);
1943 }
1944
1945 SSET_FOR_EACH (devname, &devnames) {
1946 update_port(p, devname);
1947 }
1948 sset_destroy(&devnames);
1949 }
1950
1951 static ofp_port_t
1952 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
1953 {
1954 uint16_t port_idx;
1955
1956 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
1957 port_idx = port_idx ? port_idx : UINT16_MAX;
1958
1959 if (port_idx >= ofproto->max_ports
1960 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
1961 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
1962 long long int last_used_at, lru = LLONG_MAX;
1963
1964 /* Search for a free OpenFlow port number. We try not to
1965 * immediately reuse them to prevent problems due to old
1966 * flows.
1967 *
1968 * We limit the automatically assigned port numbers to the lower half
1969 * of the port range, to reserve the upper half for assignment by
1970 * controllers. */
1971 for (;;) {
1972 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
1973 ofproto->alloc_port_no = 1;
1974 }
1975 last_used_at = ofport_get_usage(ofproto,
1976 u16_to_ofp(ofproto->alloc_port_no));
1977 if (!last_used_at) {
1978 port_idx = ofproto->alloc_port_no;
1979 break;
1980 } else if ( last_used_at < time_msec() - 60*60*1000) {
1981 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
1982 * more than an hour ago, consider it usable. */
1983 ofport_remove_usage(ofproto,
1984 u16_to_ofp(ofproto->alloc_port_no));
1985 port_idx = ofproto->alloc_port_no;
1986 break;
1987 } else if (last_used_at < lru) {
1988 lru = last_used_at;
1989 lru_ofport = ofproto->alloc_port_no;
1990 }
1991
1992 if (ofproto->alloc_port_no == end_port_no) {
1993 if (lru_ofport) {
1994 port_idx = lru_ofport;
1995 break;
1996 }
1997 return OFPP_NONE;
1998 }
1999 }
2000 }
2001 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2002 return u16_to_ofp(port_idx);
2003 }
2004
2005 static void
2006 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2007 {
2008 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2009 ofport_set_usage(ofproto, ofp_port, time_msec());
2010 }
2011 }
2012
2013 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2014 * pointer if the netdev cannot be opened. On success, also fills in
2015 * '*pp'. */
2016 static struct netdev *
2017 ofport_open(struct ofproto *ofproto,
2018 struct ofproto_port *ofproto_port,
2019 struct ofputil_phy_port *pp)
2020 {
2021 enum netdev_flags flags;
2022 struct netdev *netdev;
2023 int error;
2024
2025 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2026 if (error) {
2027 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2028 "cannot be opened (%s)",
2029 ofproto->name,
2030 ofproto_port->name, ofproto_port->ofp_port,
2031 ofproto_port->name, ovs_strerror(error));
2032 return NULL;
2033 }
2034
2035 if (ofproto_port->ofp_port == OFPP_NONE) {
2036 if (!strcmp(ofproto->name, ofproto_port->name)) {
2037 ofproto_port->ofp_port = OFPP_LOCAL;
2038 } else {
2039 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2040 ofproto_port->name);
2041 }
2042 }
2043 pp->port_no = ofproto_port->ofp_port;
2044 netdev_get_etheraddr(netdev, pp->hw_addr);
2045 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2046 netdev_get_flags(netdev, &flags);
2047 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2048 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2049 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2050 &pp->supported, &pp->peer);
2051 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2052 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2053
2054 return netdev;
2055 }
2056
2057 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2058 * port number, and 'config' bits other than OFPUTIL_PC_PORT_DOWN are
2059 * disregarded. */
2060 static bool
2061 ofport_equal(const struct ofputil_phy_port *a,
2062 const struct ofputil_phy_port *b)
2063 {
2064 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2065 && a->state == b->state
2066 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2067 && a->curr == b->curr
2068 && a->advertised == b->advertised
2069 && a->supported == b->supported
2070 && a->peer == b->peer
2071 && a->curr_speed == b->curr_speed
2072 && a->max_speed == b->max_speed);
2073 }
2074
2075 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2076 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2077 * one with the same name or port number). */
2078 static void
2079 ofport_install(struct ofproto *p,
2080 struct netdev *netdev, const struct ofputil_phy_port *pp)
2081 {
2082 const char *netdev_name = netdev_get_name(netdev);
2083 struct ofport *ofport;
2084 int error;
2085
2086 /* Create ofport. */
2087 ofport = p->ofproto_class->port_alloc();
2088 if (!ofport) {
2089 error = ENOMEM;
2090 goto error;
2091 }
2092 ofport->ofproto = p;
2093 ofport->netdev = netdev;
2094 ofport->change_seq = netdev_get_change_seq(netdev);
2095 ofport->pp = *pp;
2096 ofport->ofp_port = pp->port_no;
2097 ofport->created = time_msec();
2098
2099 /* Add port to 'p'. */
2100 hmap_insert(&p->ports, &ofport->hmap_node,
2101 hash_ofp_port(ofport->ofp_port));
2102 shash_add(&p->port_by_name, netdev_name, ofport);
2103
2104 update_mtu(p, ofport);
2105
2106 /* Let the ofproto_class initialize its private data. */
2107 error = p->ofproto_class->port_construct(ofport);
2108 if (error) {
2109 goto error;
2110 }
2111 connmgr_send_port_status(p->connmgr, NULL, pp, OFPPR_ADD);
2112 return;
2113
2114 error:
2115 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2116 p->name, netdev_name, ovs_strerror(error));
2117 if (ofport) {
2118 ofport_destroy__(ofport);
2119 } else {
2120 netdev_close(netdev);
2121 }
2122 }
2123
2124 /* Removes 'ofport' from 'p' and destroys it. */
2125 static void
2126 ofport_remove(struct ofport *ofport)
2127 {
2128 connmgr_send_port_status(ofport->ofproto->connmgr, NULL, &ofport->pp,
2129 OFPPR_DELETE);
2130 ofport_destroy(ofport);
2131 }
2132
2133 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2134 * destroys it. */
2135 static void
2136 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2137 {
2138 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2139 if (port) {
2140 ofport_remove(port);
2141 }
2142 }
2143
2144 /* Updates 'port' with new 'pp' description.
2145 *
2146 * Does not handle a name or port number change. The caller must implement
2147 * such a change as a delete followed by an add. */
2148 static void
2149 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2150 {
2151 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2152 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2153 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2154 port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN)
2155 | (pp->state & OFPUTIL_PS_LINK_DOWN));
2156 port->pp.curr = pp->curr;
2157 port->pp.advertised = pp->advertised;
2158 port->pp.supported = pp->supported;
2159 port->pp.peer = pp->peer;
2160 port->pp.curr_speed = pp->curr_speed;
2161 port->pp.max_speed = pp->max_speed;
2162
2163 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2164 &port->pp, OFPPR_MODIFY);
2165 }
2166
2167 /* Update OpenFlow 'state' in 'port' and notify controller. */
2168 void
2169 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2170 {
2171 if (port->pp.state != state) {
2172 port->pp.state = state;
2173 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2174 &port->pp, OFPPR_MODIFY);
2175 }
2176 }
2177
2178 void
2179 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2180 {
2181 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2182 if (port) {
2183 if (port->ofproto->ofproto_class->set_realdev) {
2184 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2185 }
2186 if (port->ofproto->ofproto_class->set_stp_port) {
2187 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2188 }
2189 if (port->ofproto->ofproto_class->set_cfm) {
2190 port->ofproto->ofproto_class->set_cfm(port, NULL);
2191 }
2192 if (port->ofproto->ofproto_class->bundle_remove) {
2193 port->ofproto->ofproto_class->bundle_remove(port);
2194 }
2195 }
2196 }
2197
2198 static void
2199 ofport_destroy__(struct ofport *port)
2200 {
2201 struct ofproto *ofproto = port->ofproto;
2202 const char *name = netdev_get_name(port->netdev);
2203
2204 hmap_remove(&ofproto->ports, &port->hmap_node);
2205 shash_delete(&ofproto->port_by_name,
2206 shash_find(&ofproto->port_by_name, name));
2207
2208 netdev_close(port->netdev);
2209 ofproto->ofproto_class->port_dealloc(port);
2210 }
2211
2212 static void
2213 ofport_destroy(struct ofport *port)
2214 {
2215 if (port) {
2216 dealloc_ofp_port(port->ofproto, port->ofp_port);
2217 port->ofproto->ofproto_class->port_destruct(port);
2218 ofport_destroy__(port);
2219 }
2220 }
2221
2222 struct ofport *
2223 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2224 {
2225 struct ofport *port;
2226
2227 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2228 &ofproto->ports) {
2229 if (port->ofp_port == ofp_port) {
2230 return port;
2231 }
2232 }
2233 return NULL;
2234 }
2235
2236 static long long int
2237 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2238 {
2239 struct ofport_usage *usage;
2240
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 return usage->last_used;
2245 }
2246 }
2247 return 0;
2248 }
2249
2250 static void
2251 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2252 long long int last_used)
2253 {
2254 struct ofport_usage *usage;
2255 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2256 &ofproto->ofport_usage) {
2257 if (usage->ofp_port == ofp_port) {
2258 usage->last_used = last_used;
2259 return;
2260 }
2261 }
2262 ovs_assert(last_used == LLONG_MAX);
2263
2264 usage = xmalloc(sizeof *usage);
2265 usage->ofp_port = ofp_port;
2266 usage->last_used = last_used;
2267 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2268 hash_ofp_port(ofp_port));
2269 }
2270
2271 static void
2272 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2273 {
2274 struct ofport_usage *usage;
2275 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2276 &ofproto->ofport_usage) {
2277 if (usage->ofp_port == ofp_port) {
2278 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2279 free(usage);
2280 break;
2281 }
2282 }
2283 }
2284
2285 int
2286 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2287 {
2288 struct ofproto *ofproto = port->ofproto;
2289 int error;
2290
2291 if (ofproto->ofproto_class->port_get_stats) {
2292 error = ofproto->ofproto_class->port_get_stats(port, stats);
2293 } else {
2294 error = EOPNOTSUPP;
2295 }
2296
2297 return error;
2298 }
2299
2300 static void
2301 update_port(struct ofproto *ofproto, const char *name)
2302 {
2303 struct ofproto_port ofproto_port;
2304 struct ofputil_phy_port pp;
2305 struct netdev *netdev;
2306 struct ofport *port;
2307
2308 COVERAGE_INC(ofproto_update_port);
2309
2310 /* Fetch 'name''s location and properties from the datapath. */
2311 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2312 ? ofport_open(ofproto, &ofproto_port, &pp)
2313 : NULL);
2314
2315 if (netdev) {
2316 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2317 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2318 struct netdev *old_netdev = port->netdev;
2319
2320 /* 'name' hasn't changed location. Any properties changed? */
2321 if (!ofport_equal(&port->pp, &pp)) {
2322 ofport_modified(port, &pp);
2323 }
2324
2325 update_mtu(ofproto, port);
2326
2327 /* Install the newly opened netdev in case it has changed.
2328 * Don't close the old netdev yet in case port_modified has to
2329 * remove a retained reference to it.*/
2330 port->netdev = netdev;
2331 port->change_seq = netdev_get_change_seq(netdev);
2332
2333 if (port->ofproto->ofproto_class->port_modified) {
2334 port->ofproto->ofproto_class->port_modified(port);
2335 }
2336
2337 netdev_close(old_netdev);
2338 } else {
2339 /* If 'port' is nonnull then its name differs from 'name' and thus
2340 * we should delete it. If we think there's a port named 'name'
2341 * then its port number must be wrong now so delete it too. */
2342 if (port) {
2343 ofport_remove(port);
2344 }
2345 ofport_remove_with_name(ofproto, name);
2346 ofport_install(ofproto, netdev, &pp);
2347 }
2348 } else {
2349 /* Any port named 'name' is gone now. */
2350 ofport_remove_with_name(ofproto, name);
2351 }
2352 ofproto_port_destroy(&ofproto_port);
2353 }
2354
2355 static int
2356 init_ports(struct ofproto *p)
2357 {
2358 struct ofproto_port_dump dump;
2359 struct ofproto_port ofproto_port;
2360 struct shash_node *node, *next;
2361
2362 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2363 const char *name = ofproto_port.name;
2364
2365 if (shash_find(&p->port_by_name, name)) {
2366 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2367 p->name, name);
2368 } else {
2369 struct ofputil_phy_port pp;
2370 struct netdev *netdev;
2371
2372 /* Check if an OpenFlow port number had been requested. */
2373 node = shash_find(&init_ofp_ports, name);
2374 if (node) {
2375 const struct iface_hint *iface_hint = node->data;
2376 simap_put(&p->ofp_requests, name,
2377 ofp_to_u16(iface_hint->ofp_port));
2378 }
2379
2380 netdev = ofport_open(p, &ofproto_port, &pp);
2381 if (netdev) {
2382 ofport_install(p, netdev, &pp);
2383 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2384 p->alloc_port_no = MAX(p->alloc_port_no,
2385 ofp_to_u16(ofproto_port.ofp_port));
2386 }
2387 }
2388 }
2389 }
2390
2391 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2392 struct iface_hint *iface_hint = node->data;
2393
2394 if (!strcmp(iface_hint->br_name, p->name)) {
2395 free(iface_hint->br_name);
2396 free(iface_hint->br_type);
2397 free(iface_hint);
2398 shash_delete(&init_ofp_ports, node);
2399 }
2400 }
2401
2402 return 0;
2403 }
2404
2405 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2406 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2407 static int
2408 find_min_mtu(struct ofproto *p)
2409 {
2410 struct ofport *ofport;
2411 int mtu = 0;
2412
2413 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2414 struct netdev *netdev = ofport->netdev;
2415 int dev_mtu;
2416
2417 /* Skip any internal ports, since that's what we're trying to
2418 * set. */
2419 if (!strcmp(netdev_get_type(netdev), "internal")) {
2420 continue;
2421 }
2422
2423 if (netdev_get_mtu(netdev, &dev_mtu)) {
2424 continue;
2425 }
2426 if (!mtu || dev_mtu < mtu) {
2427 mtu = dev_mtu;
2428 }
2429 }
2430
2431 return mtu ? mtu: ETH_PAYLOAD_MAX;
2432 }
2433
2434 /* Update MTU of all datapath devices on 'p' to the minimum of the
2435 * non-datapath ports in event of 'port' added or changed. */
2436 static void
2437 update_mtu(struct ofproto *p, struct ofport *port)
2438 {
2439 struct ofport *ofport;
2440 struct netdev *netdev = port->netdev;
2441 int dev_mtu, old_min;
2442
2443 if (netdev_get_mtu(netdev, &dev_mtu)) {
2444 port->mtu = 0;
2445 return;
2446 }
2447 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2448 if (dev_mtu > p->min_mtu) {
2449 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2450 dev_mtu = p->min_mtu;
2451 }
2452 }
2453 port->mtu = dev_mtu;
2454 return;
2455 }
2456
2457 /* For non-internal port find new min mtu. */
2458 old_min = p->min_mtu;
2459 port->mtu = dev_mtu;
2460 p->min_mtu = find_min_mtu(p);
2461 if (p->min_mtu == old_min) {
2462 return;
2463 }
2464
2465 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2466 struct netdev *netdev = ofport->netdev;
2467
2468 if (!strcmp(netdev_get_type(netdev), "internal")) {
2469 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2470 ofport->mtu = p->min_mtu;
2471 }
2472 }
2473 }
2474 }
2475 \f
2476 static void
2477 ofproto_rule_destroy__(struct rule *rule)
2478 OVS_NO_THREAD_SAFETY_ANALYSIS
2479 {
2480 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2481 rule_actions_destroy(rule_get_actions(rule));
2482 ovs_mutex_destroy(&rule->mutex);
2483 rule->ofproto->ofproto_class->rule_dealloc(rule);
2484 }
2485
2486 static void
2487 rule_destroy_cb(struct rule *rule)
2488 {
2489 rule->ofproto->ofproto_class->rule_destruct(rule);
2490 ofproto_rule_destroy__(rule);
2491 }
2492
2493 void
2494 ofproto_rule_ref(struct rule *rule)
2495 {
2496 if (rule) {
2497 ovs_refcount_ref(&rule->ref_count);
2498 }
2499 }
2500
2501 /* Decrements 'rule''s ref_count and schedules 'rule' to be destroyed if the
2502 * ref_count reaches 0.
2503 *
2504 * Use of RCU allows short term use (between RCU quiescent periods) without
2505 * keeping a reference. A reference must be taken if the rule needs to
2506 * stay around accross the RCU quiescent periods. */
2507 void
2508 ofproto_rule_unref(struct rule *rule)
2509 {
2510 if (rule && ovs_refcount_unref(&rule->ref_count) == 1) {
2511 ovsrcu_postpone(rule_destroy_cb, rule);
2512 }
2513 }
2514
2515 void
2516 ofproto_group_ref(struct ofgroup *group)
2517 {
2518 if (group) {
2519 ovs_refcount_ref(&group->ref_count);
2520 }
2521 }
2522
2523 void
2524 ofproto_group_unref(struct ofgroup *group)
2525 {
2526 if (group && ovs_refcount_unref(&group->ref_count) == 1) {
2527 group->ofproto->ofproto_class->group_destruct(group);
2528 ofputil_bucket_list_destroy(&group->buckets);
2529 group->ofproto->ofproto_class->group_dealloc(group);
2530 }
2531 }
2532
2533 static uint32_t get_provider_meter_id(const struct ofproto *,
2534 uint32_t of_meter_id);
2535
2536 /* Creates and returns a new 'struct rule_actions', whose actions are a copy
2537 * of from the 'ofpacts_len' bytes of 'ofpacts'. */
2538 const struct rule_actions *
2539 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2540 {
2541 struct rule_actions *actions;
2542
2543 actions = xmalloc(sizeof *actions + ofpacts_len);
2544 actions->ofpacts_len = ofpacts_len;
2545 actions->has_meter = ofpacts_get_meter(ofpacts, ofpacts_len) != 0;
2546 memcpy(actions->ofpacts, ofpacts, ofpacts_len);
2547
2548 actions->has_learn_with_delete = (next_learn_with_delete(actions, NULL)
2549 != NULL);
2550
2551 return actions;
2552 }
2553
2554 /* Free the actions after the RCU quiescent period is reached. */
2555 void
2556 rule_actions_destroy(const struct rule_actions *actions)
2557 {
2558 if (actions) {
2559 ovsrcu_postpone(free, CONST_CAST(struct rule_actions *, actions));
2560 }
2561 }
2562
2563 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2564 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2565 static bool
2566 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2567 OVS_REQUIRES(ofproto_mutex)
2568 {
2569 if (port == OFPP_ANY) {
2570 return true;
2571 } else {
2572 const struct rule_actions *actions = rule_get_actions(rule);
2573 return ofpacts_output_to_port(actions->ofpacts,
2574 actions->ofpacts_len, port);
2575 }
2576 }
2577
2578 /* Returns true if 'rule' has group and equals group_id. */
2579 static bool
2580 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2581 OVS_REQUIRES(ofproto_mutex)
2582 {
2583 if (group_id == OFPG_ANY) {
2584 return true;
2585 } else {
2586 const struct rule_actions *actions = rule_get_actions(rule);
2587 return ofpacts_output_to_group(actions->ofpacts,
2588 actions->ofpacts_len, group_id);
2589 }
2590 }
2591
2592 static void
2593 rule_execute_destroy(struct rule_execute *e)
2594 {
2595 ofproto_rule_unref(e->rule);
2596 list_remove(&e->list_node);
2597 free(e);
2598 }
2599
2600 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2601 * by passing them to the ofproto provider. */
2602 static void
2603 run_rule_executes(struct ofproto *ofproto)
2604 OVS_EXCLUDED(ofproto_mutex)
2605 {
2606 struct rule_execute *e, *next;
2607 struct list executes;
2608
2609 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2610 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2611 struct flow flow;
2612
2613 flow_extract(e->packet, NULL, &flow);
2614 flow.in_port.ofp_port = e->in_port;
2615 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2616
2617 rule_execute_destroy(e);
2618 }
2619 }
2620
2621 /* Destroys and discards all "rule_execute" operations queued up in
2622 * ofproto->rule_executes. */
2623 static void
2624 destroy_rule_executes(struct ofproto *ofproto)
2625 {
2626 struct rule_execute *e, *next;
2627 struct list executes;
2628
2629 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2630 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2631 ofpbuf_delete(e->packet);
2632 rule_execute_destroy(e);
2633 }
2634 }
2635
2636 static bool
2637 rule_is_readonly(const struct rule *rule)
2638 {
2639 const struct oftable *table = &rule->ofproto->tables[rule->table_id];
2640 return (table->flags & OFTABLE_READONLY) != 0;
2641 }
2642 \f
2643 static uint32_t
2644 hash_learned_cookie(ovs_be64 cookie_, uint8_t table_id)
2645 {
2646 uint64_t cookie = (OVS_FORCE uint64_t) cookie_;
2647 return hash_3words(cookie, cookie >> 32, table_id);
2648 }
2649
2650 static void
2651 learned_cookies_update_one__(struct ofproto *ofproto,
2652 const struct ofpact_learn *learn,
2653 int delta, struct list *dead_cookies)
2654 OVS_REQUIRES(ofproto_mutex)
2655 {
2656 uint32_t hash = hash_learned_cookie(learn->cookie, learn->table_id);
2657 struct learned_cookie *c;
2658
2659 HMAP_FOR_EACH_WITH_HASH (c, u.hmap_node, hash, &ofproto->learned_cookies) {
2660 if (c->cookie == learn->cookie && c->table_id == learn->table_id) {
2661 c->n += delta;
2662 ovs_assert(c->n >= 0);
2663
2664 if (!c->n) {
2665 hmap_remove(&ofproto->learned_cookies, &c->u.hmap_node);
2666 list_push_back(dead_cookies, &c->u.list_node);
2667 }
2668
2669 return;
2670 }
2671 }
2672
2673 ovs_assert(delta > 0);
2674 c = xmalloc(sizeof *c);
2675 hmap_insert(&ofproto->learned_cookies, &c->u.hmap_node, hash);
2676 c->cookie = learn->cookie;
2677 c->table_id = learn->table_id;
2678 c->n = delta;
2679 }
2680
2681 static const struct ofpact_learn *
2682 next_learn_with_delete(const struct rule_actions *actions,
2683 const struct ofpact_learn *start)
2684 {
2685 const struct ofpact *pos;
2686
2687 for (pos = start ? ofpact_next(&start->ofpact) : actions->ofpacts;
2688 pos < ofpact_end(actions->ofpacts, actions->ofpacts_len);
2689 pos = ofpact_next(pos)) {
2690 if (pos->type == OFPACT_LEARN) {
2691 const struct ofpact_learn *learn = ofpact_get_LEARN(pos);
2692 if (learn->flags & NX_LEARN_F_DELETE_LEARNED) {
2693 return learn;
2694 }
2695 }
2696 }
2697
2698 return NULL;
2699 }
2700
2701 static void
2702 learned_cookies_update__(struct ofproto *ofproto,
2703 const struct rule_actions *actions,
2704 int delta, struct list *dead_cookies)
2705 OVS_REQUIRES(ofproto_mutex)
2706 {
2707 if (actions->has_learn_with_delete) {
2708 const struct ofpact_learn *learn;
2709
2710 for (learn = next_learn_with_delete(actions, NULL); learn;
2711 learn = next_learn_with_delete(actions, learn)) {
2712 learned_cookies_update_one__(ofproto, learn, delta, dead_cookies);
2713 }
2714 }
2715 }
2716
2717 static void
2718 learned_cookies_inc(struct ofproto *ofproto,
2719 const struct rule_actions *actions)
2720 OVS_REQUIRES(ofproto_mutex)
2721 {
2722 learned_cookies_update__(ofproto, actions, +1, NULL);
2723 }
2724
2725 static void
2726 learned_cookies_dec(struct ofproto *ofproto,
2727 const struct rule_actions *actions,
2728 struct list *dead_cookies)
2729 OVS_REQUIRES(ofproto_mutex)
2730 {
2731 learned_cookies_update__(ofproto, actions, -1, dead_cookies);
2732 }
2733
2734 static void
2735 learned_cookies_flush(struct ofproto *ofproto, struct list *dead_cookies)
2736 OVS_REQUIRES(ofproto_mutex)
2737 {
2738 struct learned_cookie *c, *next;
2739
2740 LIST_FOR_EACH_SAFE (c, next, u.list_node, dead_cookies) {
2741 struct rule_criteria criteria;
2742 struct rule_collection rules;
2743 struct match match;
2744
2745 match_init_catchall(&match);
2746 rule_criteria_init(&criteria, c->table_id, &match, 0,
2747 c->cookie, OVS_BE64_MAX, OFPP_ANY, OFPG_ANY);
2748 rule_criteria_require_rw(&criteria, false);
2749 collect_rules_loose(ofproto, &criteria, &rules);
2750 delete_flows__(&rules, OFPRR_DELETE, NULL);
2751 rule_criteria_destroy(&criteria);
2752 rule_collection_destroy(&rules);
2753
2754 list_remove(&c->u.list_node);
2755 free(c);
2756 }
2757 }
2758 \f
2759 static enum ofperr
2760 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
2761 {
2762 ofconn_send_reply(ofconn, make_echo_reply(oh));
2763 return 0;
2764 }
2765
2766 static enum ofperr
2767 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
2768 {
2769 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2770 struct ofputil_switch_features features;
2771 struct ofport *port;
2772 bool arp_match_ip;
2773 struct ofpbuf *b;
2774
2775 ofproto->ofproto_class->get_features(ofproto, &arp_match_ip,
2776 &features.actions);
2777 ovs_assert(features.actions & OFPUTIL_A_OUTPUT); /* sanity check */
2778
2779 features.datapath_id = ofproto->datapath_id;
2780 features.n_buffers = pktbuf_capacity();
2781 features.n_tables = ofproto_get_n_visible_tables(ofproto);
2782 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
2783 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS);
2784 if (arp_match_ip) {
2785 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
2786 }
2787 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
2788 features.auxiliary_id = 0;
2789 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
2790 oh->xid);
2791 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
2792 ofputil_put_switch_features_port(&port->pp, b);
2793 }
2794
2795 ofconn_send_reply(ofconn, b);
2796 return 0;
2797 }
2798
2799 static enum ofperr
2800 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
2801 {
2802 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2803 struct ofp_switch_config *osc;
2804 enum ofp_config_flags flags;
2805 struct ofpbuf *buf;
2806
2807 /* Send reply. */
2808 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
2809 osc = ofpbuf_put_uninit(buf, sizeof *osc);
2810 flags = ofproto->frag_handling;
2811 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2812 if (oh->version < OFP13_VERSION
2813 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
2814 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
2815 }
2816 osc->flags = htons(flags);
2817 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
2818 ofconn_send_reply(ofconn, buf);
2819
2820 return 0;
2821 }
2822
2823 static enum ofperr
2824 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
2825 {
2826 const struct ofp_switch_config *osc = ofpmsg_body(oh);
2827 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
2828 uint16_t flags = ntohs(osc->flags);
2829
2830 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
2831 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
2832 enum ofp_config_flags cur = ofproto->frag_handling;
2833 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
2834
2835 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
2836 if (cur != next) {
2837 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
2838 ofproto->frag_handling = next;
2839 } else {
2840 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
2841 ofproto->name,
2842 ofputil_frag_handling_to_string(next));
2843 }
2844 }
2845 }
2846 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
2847 ofconn_set_invalid_ttl_to_controller(ofconn,
2848 (oh->version < OFP13_VERSION
2849 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
2850
2851 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
2852
2853 return 0;
2854 }
2855
2856 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
2857 * error message code for the caller to propagate upward. Otherwise, returns
2858 * 0.
2859 *
2860 * The log message mentions 'msg_type'. */
2861 static enum ofperr
2862 reject_slave_controller(struct ofconn *ofconn)
2863 {
2864 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
2865 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
2866 return OFPERR_OFPBRC_EPERM;
2867 } else {
2868 return 0;
2869 }
2870 }
2871
2872 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
2873 * for 'ofproto':
2874 *
2875 * - If they use a meter, then 'ofproto' has that meter configured.
2876 *
2877 * - If they use any groups, then 'ofproto' has that group configured.
2878 *
2879 * Returns 0 if successful, otherwise an OpenFlow error. */
2880 static enum ofperr
2881 ofproto_check_ofpacts(struct ofproto *ofproto,
2882 const struct ofpact ofpacts[], size_t ofpacts_len)
2883 {
2884 const struct ofpact *a;
2885 uint32_t mid;
2886
2887 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
2888 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
2889 return OFPERR_OFPMMFC_INVALID_METER;
2890 }
2891
2892 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
2893 if (a->type == OFPACT_GROUP
2894 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
2895 return OFPERR_OFPBAC_BAD_OUT_GROUP;
2896 }
2897 }
2898
2899 return 0;
2900 }
2901
2902 static enum ofperr
2903 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
2904 {
2905 struct ofproto *p = ofconn_get_ofproto(ofconn);
2906 struct ofputil_packet_out po;
2907 struct ofpbuf *payload;
2908 uint64_t ofpacts_stub[1024 / 8];
2909 struct ofpbuf ofpacts;
2910 struct flow flow;
2911 enum ofperr error;
2912
2913 COVERAGE_INC(ofproto_packet_out);
2914
2915 error = reject_slave_controller(ofconn);
2916 if (error) {
2917 goto exit;
2918 }
2919
2920 /* Decode message. */
2921 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
2922 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
2923 if (error) {
2924 goto exit_free_ofpacts;
2925 }
2926 if (ofp_to_u16(po.in_port) >= p->max_ports
2927 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
2928 error = OFPERR_OFPBRC_BAD_PORT;
2929 goto exit_free_ofpacts;
2930 }
2931
2932 /* Get payload. */
2933 if (po.buffer_id != UINT32_MAX) {
2934 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
2935 if (error || !payload) {
2936 goto exit_free_ofpacts;
2937 }
2938 } else {
2939 /* Ensure that the L3 header is 32-bit aligned. */
2940 payload = ofpbuf_clone_data_with_headroom(po.packet, po.packet_len, 2);
2941 }
2942
2943 /* Verify actions against packet, then send packet if successful. */
2944 flow_extract(payload, NULL, &flow);
2945 flow.in_port.ofp_port = po.in_port;
2946 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
2947 if (!error) {
2948 error = p->ofproto_class->packet_out(p, payload, &flow,
2949 po.ofpacts, po.ofpacts_len);
2950 }
2951 ofpbuf_delete(payload);
2952
2953 exit_free_ofpacts:
2954 ofpbuf_uninit(&ofpacts);
2955 exit:
2956 return error;
2957 }
2958
2959 static void
2960 update_port_config(struct ofconn *ofconn, struct ofport *port,
2961 enum ofputil_port_config config,
2962 enum ofputil_port_config mask)
2963 {
2964 enum ofputil_port_config toggle = (config ^ port->pp.config) & mask;
2965
2966 if (toggle & OFPUTIL_PC_PORT_DOWN
2967 && (config & OFPUTIL_PC_PORT_DOWN
2968 ? netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL)
2969 : netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL))) {
2970 /* We tried to bring the port up or down, but it failed, so don't
2971 * update the "down" bit. */
2972 toggle &= ~OFPUTIL_PC_PORT_DOWN;
2973 }
2974
2975 if (toggle) {
2976 enum ofputil_port_config old_config = port->pp.config;
2977 port->pp.config ^= toggle;
2978 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
2979 connmgr_send_port_status(port->ofproto->connmgr, ofconn, &port->pp,
2980 OFPPR_MODIFY);
2981 }
2982 }
2983
2984 static enum ofperr
2985 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
2986 {
2987 struct ofproto *p = ofconn_get_ofproto(ofconn);
2988 struct ofputil_port_mod pm;
2989 struct ofport *port;
2990 enum ofperr error;
2991
2992 error = reject_slave_controller(ofconn);
2993 if (error) {
2994 return error;
2995 }
2996
2997 error = ofputil_decode_port_mod(oh, &pm, false);
2998 if (error) {
2999 return error;
3000 }
3001
3002 port = ofproto_get_port(p, pm.port_no);
3003 if (!port) {
3004 return OFPERR_OFPPMFC_BAD_PORT;
3005 } else if (!eth_addr_equals(port->pp.hw_addr, pm.hw_addr)) {
3006 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3007 } else {
3008 update_port_config(ofconn, port, pm.config, pm.mask);
3009 if (pm.advertise) {
3010 netdev_set_advertisements(port->netdev, pm.advertise);
3011 }
3012 }
3013 return 0;
3014 }
3015
3016 static enum ofperr
3017 handle_desc_stats_request(struct ofconn *ofconn,
3018 const struct ofp_header *request)
3019 {
3020 static const char *default_mfr_desc = "Nicira, Inc.";
3021 static const char *default_hw_desc = "Open vSwitch";
3022 static const char *default_sw_desc = VERSION;
3023 static const char *default_serial_desc = "None";
3024 static const char *default_dp_desc = "None";
3025
3026 struct ofproto *p = ofconn_get_ofproto(ofconn);
3027 struct ofp_desc_stats *ods;
3028 struct ofpbuf *msg;
3029
3030 msg = ofpraw_alloc_stats_reply(request, 0);
3031 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3032 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3033 sizeof ods->mfr_desc);
3034 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3035 sizeof ods->hw_desc);
3036 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3037 sizeof ods->sw_desc);
3038 ovs_strlcpy(ods->serial_num,
3039 p->serial_desc ? p->serial_desc : default_serial_desc,
3040 sizeof ods->serial_num);
3041 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3042 sizeof ods->dp_desc);
3043 ofconn_send_reply(ofconn, msg);
3044
3045 return 0;
3046 }
3047
3048 static enum ofperr
3049 handle_table_stats_request(struct ofconn *ofconn,
3050 const struct ofp_header *request)
3051 {
3052 struct ofproto *p = ofconn_get_ofproto(ofconn);
3053 struct ofp12_table_stats *ots;
3054 struct ofpbuf *msg;
3055 int n_tables;
3056 size_t i;
3057
3058 /* Set up default values.
3059 *
3060 * ofp12_table_stats is used as a generic structure as
3061 * it is able to hold all the fields for ofp10_table_stats
3062 * and ofp11_table_stats (and of course itself).
3063 */
3064 ots = xcalloc(p->n_tables, sizeof *ots);
3065 for (i = 0; i < p->n_tables; i++) {
3066 ots[i].table_id = i;
3067 sprintf(ots[i].name, "table%"PRIuSIZE, i);
3068 ots[i].match = htonll(OFPXMT13_MASK);
3069 ots[i].wildcards = htonll(OFPXMT13_MASK);
3070 ots[i].write_actions = htonl(OFPAT11_OUTPUT);
3071 ots[i].apply_actions = htonl(OFPAT11_OUTPUT);
3072 ots[i].write_setfields = htonll(OFPXMT13_MASK);
3073 ots[i].apply_setfields = htonll(OFPXMT13_MASK);
3074 ots[i].metadata_match = OVS_BE64_MAX;
3075 ots[i].metadata_write = OVS_BE64_MAX;
3076 ots[i].instructions = htonl(OFPIT11_ALL);
3077 ots[i].config = htonl(OFPTC11_TABLE_MISS_MASK);
3078 ots[i].max_entries = htonl(1000000); /* An arbitrary big number. */
3079 fat_rwlock_rdlock(&p->tables[i].cls.rwlock);
3080 ots[i].active_count = htonl(classifier_count(&p->tables[i].cls));
3081 fat_rwlock_unlock(&p->tables[i].cls.rwlock);
3082 }
3083
3084 p->ofproto_class->get_tables(p, ots);
3085
3086 /* Post-process the tables, dropping hidden tables. */
3087 n_tables = p->n_tables;
3088 for (i = 0; i < p->n_tables; i++) {
3089 const struct oftable *table = &p->tables[i];
3090
3091 if (table->flags & OFTABLE_HIDDEN) {
3092 n_tables = i;
3093 break;
3094 }
3095
3096 if (table->name) {
3097 ovs_strzcpy(ots[i].name, table->name, sizeof ots[i].name);
3098 }
3099
3100 if (table->max_flows < ntohl(ots[i].max_entries)) {
3101 ots[i].max_entries = htonl(table->max_flows);
3102 }
3103 }
3104
3105 msg = ofputil_encode_table_stats_reply(ots, n_tables, request);
3106 ofconn_send_reply(ofconn, msg);
3107
3108 free(ots);
3109
3110 return 0;
3111 }
3112
3113 static void
3114 append_port_stat(struct ofport *port, struct list *replies)
3115 {
3116 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3117
3118 calc_duration(port->created, time_msec(),
3119 &ops.duration_sec, &ops.duration_nsec);
3120
3121 /* Intentionally ignore return value, since errors will set
3122 * 'stats' to all-1s, which is correct for OpenFlow, and
3123 * netdev_get_stats() will log errors. */
3124 ofproto_port_get_stats(port, &ops.stats);
3125
3126 ofputil_append_port_stat(replies, &ops);
3127 }
3128
3129 static void
3130 handle_port_request(struct ofconn *ofconn,
3131 const struct ofp_header *request, ofp_port_t port_no,
3132 void (*cb)(struct ofport *, struct list *replies))
3133 {
3134 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3135 struct ofport *port;
3136 struct list replies;
3137
3138 ofpmp_init(&replies, request);
3139 if (port_no != OFPP_ANY) {
3140 port = ofproto_get_port(ofproto, port_no);
3141 if (port) {
3142 cb(port, &replies);
3143 }
3144 } else {
3145 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3146 cb(port, &replies);
3147 }
3148 }
3149
3150 ofconn_send_replies(ofconn, &replies);
3151 }
3152
3153 static enum ofperr
3154 handle_port_stats_request(struct ofconn *ofconn,
3155 const struct ofp_header *request)
3156 {
3157 ofp_port_t port_no;
3158 enum ofperr error;
3159
3160 error = ofputil_decode_port_stats_request(request, &port_no);
3161 if (!error) {
3162 handle_port_request(ofconn, request, port_no, append_port_stat);
3163 }
3164 return error;
3165 }
3166
3167 static void
3168 append_port_desc(struct ofport *port, struct list *replies)
3169 {
3170 ofputil_append_port_desc_stats_reply(&port->pp, replies);
3171 }
3172
3173 static enum ofperr
3174 handle_port_desc_stats_request(struct ofconn *ofconn,
3175 const struct ofp_header *request)
3176 {
3177 ofp_port_t port_no;
3178 enum ofperr error;
3179
3180 error = ofputil_decode_port_desc_stats_request(request, &port_no);
3181 if (!error) {
3182 handle_port_request(ofconn, request, port_no, append_port_desc);
3183 }
3184 return error;
3185 }
3186
3187 static uint32_t
3188 hash_cookie(ovs_be64 cookie)
3189 {
3190 return hash_uint64((OVS_FORCE uint64_t)cookie);
3191 }
3192
3193 static void
3194 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3195 OVS_REQUIRES(ofproto_mutex)
3196 {
3197 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3198 hash_cookie(rule->flow_cookie));
3199 }
3200
3201 static void
3202 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3203 OVS_REQUIRES(ofproto_mutex)
3204 {
3205 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3206 }
3207
3208 static void
3209 calc_duration(long long int start, long long int now,
3210 uint32_t *sec, uint32_t *nsec)
3211 {
3212 long long int msecs = now - start;
3213 *sec = msecs / 1000;
3214 *nsec = (msecs % 1000) * (1000 * 1000);
3215 }
3216
3217 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3218 * true if 'table_id' is OK, false otherwise. */
3219 static bool
3220 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3221 {
3222 return table_id == OFPTT_ALL || table_id < ofproto->n_tables;
3223 }
3224
3225 static struct oftable *
3226 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3227 {
3228 struct oftable *table;
3229
3230 for (table = &ofproto->tables[table_id];
3231 table < &ofproto->tables[ofproto->n_tables];
3232 table++) {
3233 if (!(table->flags & OFTABLE_HIDDEN)) {
3234 return table;
3235 }
3236 }
3237
3238 return NULL;
3239 }
3240
3241 static struct oftable *
3242 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3243 {
3244 if (table_id == 0xff) {
3245 return next_visible_table(ofproto, 0);
3246 } else if (table_id < ofproto->n_tables) {
3247 return &ofproto->tables[table_id];
3248 } else {
3249 return NULL;
3250 }
3251 }
3252
3253 static struct oftable *
3254 next_matching_table(const struct ofproto *ofproto,
3255 const struct oftable *table, uint8_t table_id)
3256 {
3257 return (table_id == 0xff
3258 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3259 : NULL);
3260 }
3261
3262 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3263 *
3264 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3265 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3266 *
3267 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3268 * only once, for that table. (This can be used to access tables marked
3269 * OFTABLE_HIDDEN.)
3270 *
3271 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3272 * entered at all. (Perhaps you should have validated TABLE_ID with
3273 * check_table_id().)
3274 *
3275 * All parameters are evaluated multiple times.
3276 */
3277 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3278 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3279 (TABLE) != NULL; \
3280 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3281
3282 /* Initializes 'criteria' in a straightforward way based on the other
3283 * parameters.
3284 *
3285 * By default, the criteria include flows that are read-only, on the assumption
3286 * that the collected flows won't be modified. Call rule_criteria_require_rw()
3287 * if flows will be modified.
3288 *
3289 * For "loose" matching, the 'priority' parameter is unimportant and may be
3290 * supplied as 0. */
3291 static void
3292 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3293 const struct match *match, unsigned int priority,
3294 ovs_be64 cookie, ovs_be64 cookie_mask,
3295 ofp_port_t out_port, uint32_t out_group)
3296 {
3297 criteria->table_id = table_id;
3298 cls_rule_init(&criteria->cr, match, priority);
3299 criteria->cookie = cookie;
3300 criteria->cookie_mask = cookie_mask;
3301 criteria->out_port = out_port;
3302 criteria->out_group = out_group;
3303
3304 /* We ordinarily want to skip hidden rules, but there has to be a way for
3305 * code internal to OVS to modify and delete them, so if the criteria
3306 * specify a priority that can only be for a hidden flow, then allow hidden
3307 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3308 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3309 * priority.) */
3310 criteria->include_hidden = priority > UINT16_MAX;
3311
3312 /* We assume that the criteria are being used to collect flows for reading
3313 * but not modification. Thus, we should collect read-only flows. */
3314 criteria->include_readonly = true;
3315 }
3316
3317 /* By default, criteria initialized by rule_criteria_init() will match flows
3318 * that are read-only, on the assumption that the collected flows won't be
3319 * modified. Call this function to match only flows that are be modifiable.
3320 *
3321 * Specify 'can_write_readonly' as false in ordinary circumstances, true if the
3322 * caller has special privileges that allow it to modify even "read-only"
3323 * flows. */
3324 static void
3325 rule_criteria_require_rw(struct rule_criteria *criteria,
3326 bool can_write_readonly)
3327 {
3328 criteria->include_readonly = can_write_readonly;
3329 }
3330
3331 static void
3332 rule_criteria_destroy(struct rule_criteria *criteria)
3333 {
3334 cls_rule_destroy(&criteria->cr);
3335 }
3336
3337 void
3338 rule_collection_init(struct rule_collection *rules)
3339 {
3340 rules->rules = rules->stub;
3341 rules->n = 0;
3342 rules->capacity = ARRAY_SIZE(rules->stub);
3343 }
3344
3345 void
3346 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3347 {
3348 if (rules->n >= rules->capacity) {
3349 size_t old_size, new_size;
3350
3351 old_size = rules->capacity * sizeof *rules->rules;
3352 rules->capacity *= 2;
3353 new_size = rules->capacity * sizeof *rules->rules;
3354
3355 if (rules->rules == rules->stub) {
3356 rules->rules = xmalloc(new_size);
3357 memcpy(rules->rules, rules->stub, old_size);
3358 } else {
3359 rules->rules = xrealloc(rules->rules, new_size);
3360 }
3361 }
3362
3363 rules->rules[rules->n++] = rule;
3364 }
3365
3366 void
3367 rule_collection_ref(struct rule_collection *rules)
3368 OVS_REQUIRES(ofproto_mutex)
3369 {
3370 size_t i;
3371
3372 for (i = 0; i < rules->n; i++) {
3373 ofproto_rule_ref(rules->rules[i]);
3374 }
3375 }
3376
3377 void
3378 rule_collection_unref(struct rule_collection *rules)
3379 {
3380 size_t i;
3381
3382 for (i = 0; i < rules->n; i++) {
3383 ofproto_rule_unref(rules->rules[i]);
3384 }
3385 }
3386
3387 void
3388 rule_collection_destroy(struct rule_collection *rules)
3389 {
3390 if (rules->rules != rules->stub) {
3391 free(rules->rules);
3392 }
3393 }
3394
3395 /* Checks whether 'rule' matches 'c' and, if so, adds it to 'rules'. This
3396 * function verifies most of the criteria in 'c' itself, but the caller must
3397 * check 'c->cr' itself.
3398 *
3399 * Increments '*n_readonly' if 'rule' wasn't added because it's read-only (and
3400 * 'c' only includes modifiable rules). */
3401 static void
3402 collect_rule(struct rule *rule, const struct rule_criteria *c,
3403 struct rule_collection *rules, size_t *n_readonly)
3404 OVS_REQUIRES(ofproto_mutex)
3405 {
3406 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3407 && ofproto_rule_has_out_port(rule, c->out_port)
3408 && ofproto_rule_has_out_group(rule, c->out_group)
3409 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)
3410 && (!rule_is_hidden(rule) || c->include_hidden)) {
3411 /* Rule matches all the criteria... */
3412 if (!rule_is_readonly(rule) || c->include_readonly) {
3413 /* ...add it. */
3414 rule_collection_add(rules, rule);
3415 } else {
3416 /* ...except it's read-only. */
3417 ++*n_readonly;
3418 }
3419 }
3420 }
3421
3422 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3423 * on classifiers rules are done in the "loose" way required for OpenFlow
3424 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3425 * 'rules'.
3426 *
3427 * Returns 0 on success, otherwise an OpenFlow error code. */
3428 static enum ofperr
3429 collect_rules_loose(struct ofproto *ofproto,
3430 const struct rule_criteria *criteria,
3431 struct rule_collection *rules)
3432 OVS_REQUIRES(ofproto_mutex)
3433 {
3434 struct oftable *table;
3435 enum ofperr error = 0;
3436 size_t n_readonly = 0;
3437
3438 rule_collection_init(rules);
3439
3440 if (!check_table_id(ofproto, criteria->table_id)) {
3441 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3442 goto exit;
3443 }
3444
3445 if (criteria->cookie_mask == OVS_BE64_MAX) {
3446 struct rule *rule;
3447
3448 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3449 hash_cookie(criteria->cookie),
3450 &ofproto->cookies) {
3451 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
3452 collect_rule(rule, criteria, rules, &n_readonly);
3453 }
3454 }
3455 } else {
3456 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3457 struct cls_cursor cursor;
3458 struct rule *rule;
3459
3460 fat_rwlock_rdlock(&table->cls.rwlock);
3461 cls_cursor_init(&cursor, &table->cls, &criteria->cr);
3462 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3463 collect_rule(rule, criteria, rules, &n_readonly);
3464 }
3465 fat_rwlock_unlock(&table->cls.rwlock);
3466 }
3467 }
3468
3469 exit:
3470 if (!error && !rules->n && n_readonly) {
3471 /* We didn't find any rules to modify. We did find some read-only
3472 * rules that we're not allowed to modify, so report that. */
3473 error = OFPERR_OFPBRC_EPERM;
3474 }
3475 if (error) {
3476 rule_collection_destroy(rules);
3477 }
3478 return error;
3479 }
3480
3481 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3482 * on classifiers rules are done in the "strict" way required for OpenFlow
3483 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
3484 * rules on list 'rules'.
3485 *
3486 * Returns 0 on success, otherwise an OpenFlow error code. */
3487 static enum ofperr
3488 collect_rules_strict(struct ofproto *ofproto,
3489 const struct rule_criteria *criteria,
3490 struct rule_collection *rules)
3491 OVS_REQUIRES(ofproto_mutex)
3492 {
3493 struct oftable *table;
3494 size_t n_readonly = 0;
3495 int error = 0;
3496
3497 rule_collection_init(rules);
3498
3499 if (!check_table_id(ofproto, criteria->table_id)) {
3500 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3501 goto exit;
3502 }
3503
3504 if (criteria->cookie_mask == OVS_BE64_MAX) {
3505 struct rule *rule;
3506
3507 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3508 hash_cookie(criteria->cookie),
3509 &ofproto->cookies) {
3510 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
3511 collect_rule(rule, criteria, rules, &n_readonly);
3512 }
3513 }
3514 } else {
3515 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
3516 struct rule *rule;
3517
3518 fat_rwlock_rdlock(&table->cls.rwlock);
3519 rule = rule_from_cls_rule(classifier_find_rule_exactly(
3520 &table->cls, &criteria->cr));
3521 fat_rwlock_unlock(&table->cls.rwlock);
3522 if (rule) {
3523 collect_rule(rule, criteria, rules, &n_readonly);
3524 }
3525 }
3526 }
3527
3528 exit:
3529 if (!error && !rules->n && n_readonly) {
3530 /* We didn't find any rules to modify. We did find some read-only
3531 * rules that we're not allowed to modify, so report that. */
3532 error = OFPERR_OFPBRC_EPERM;
3533 }
3534 if (error) {
3535 rule_collection_destroy(rules);
3536 }
3537 return error;
3538 }
3539
3540 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
3541 * forced into the range of a uint16_t. */
3542 static int
3543 age_secs(long long int age_ms)
3544 {
3545 return (age_ms < 0 ? 0
3546 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
3547 : (unsigned int) age_ms / 1000);
3548 }
3549
3550 static enum ofperr
3551 handle_flow_stats_request(struct ofconn *ofconn,
3552 const struct ofp_header *request)
3553 OVS_EXCLUDED(ofproto_mutex)
3554 {
3555 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3556 struct ofputil_flow_stats_request fsr;
3557 struct rule_criteria criteria;
3558 struct rule_collection rules;
3559 struct list replies;
3560 enum ofperr error;
3561 size_t i;
3562
3563 error = ofputil_decode_flow_stats_request(&fsr, request);
3564 if (error) {
3565 return error;
3566 }
3567
3568 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, fsr.cookie,
3569 fsr.cookie_mask, fsr.out_port, fsr.out_group);
3570
3571 ovs_mutex_lock(&ofproto_mutex);
3572 error = collect_rules_loose(ofproto, &criteria, &rules);
3573 rule_criteria_destroy(&criteria);
3574 if (!error) {
3575 rule_collection_ref(&rules);
3576 }
3577 ovs_mutex_unlock(&ofproto_mutex);
3578
3579 if (error) {
3580 return error;
3581 }
3582
3583 ofpmp_init(&replies, request);
3584 for (i = 0; i < rules.n; i++) {
3585 struct rule *rule = rules.rules[i];
3586 long long int now = time_msec();
3587 struct ofputil_flow_stats fs;
3588 long long int created, used, modified;
3589 const struct rule_actions *actions;
3590 enum ofputil_flow_mod_flags flags;
3591
3592 ovs_mutex_lock(&rule->mutex);
3593 fs.cookie = rule->flow_cookie;
3594 fs.idle_timeout = rule->idle_timeout;
3595 fs.hard_timeout = rule->hard_timeout;
3596 created = rule->created;
3597 modified = rule->modified;
3598 actions = rule_get_actions(rule);
3599 flags = rule->flags;
3600 ovs_mutex_unlock(&rule->mutex);
3601
3602 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
3603 &fs.byte_count, &used);
3604
3605 minimatch_expand(&rule->cr.match, &fs.match);
3606 fs.table_id = rule->table_id;
3607 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
3608 fs.priority = rule->cr.priority;
3609 fs.idle_age = age_secs(now - used);
3610 fs.hard_age = age_secs(now - modified);
3611 fs.ofpacts = actions->ofpacts;
3612 fs.ofpacts_len = actions->ofpacts_len;
3613
3614 fs.flags = flags;
3615 ofputil_append_flow_stats_reply(&fs, &replies);
3616 }
3617
3618 rule_collection_unref(&rules);
3619 rule_collection_destroy(&rules);
3620
3621 ofconn_send_replies(ofconn, &replies);
3622
3623 return 0;
3624 }
3625
3626 static void
3627 flow_stats_ds(struct rule *rule, struct ds *results)
3628 {
3629 uint64_t packet_count, byte_count;
3630 const struct rule_actions *actions;
3631 long long int created, used;
3632
3633 rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3634 &byte_count, &used);
3635
3636 ovs_mutex_lock(&rule->mutex);
3637 actions = rule_get_actions(rule);
3638 created = rule->created;
3639 ovs_mutex_unlock(&rule->mutex);
3640
3641 if (rule->table_id != 0) {
3642 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
3643 }
3644 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
3645 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
3646 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
3647 cls_rule_format(&rule->cr, results);
3648 ds_put_char(results, ',');
3649
3650 ds_put_cstr(results, "actions=");
3651 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
3652
3653 ds_put_cstr(results, "\n");
3654 }
3655
3656 /* Adds a pretty-printed description of all flows to 'results', including
3657 * hidden flows (e.g., set up by in-band control). */
3658 void
3659 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
3660 {
3661 struct oftable *table;
3662
3663 OFPROTO_FOR_EACH_TABLE (table, p) {
3664 struct cls_cursor cursor;
3665 struct rule *rule;
3666
3667 fat_rwlock_rdlock(&table->cls.rwlock);
3668 cls_cursor_init(&cursor, &table->cls, NULL);
3669 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
3670 flow_stats_ds(rule, results);
3671 }
3672 fat_rwlock_unlock(&table->cls.rwlock);
3673 }
3674 }
3675
3676 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
3677 * '*engine_type' and '*engine_id', respectively. */
3678 void
3679 ofproto_get_netflow_ids(const struct ofproto *ofproto,
3680 uint8_t *engine_type, uint8_t *engine_id)
3681 {
3682 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
3683 }
3684
3685 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.
3686 * Returns 0 if the port's CFM status was successfully stored into
3687 * '*status'. Returns positive errno if the port did not have CFM
3688 * configured. Returns negative number if there is no status change
3689 * since last update.
3690 *
3691 * The caller must provide and own '*status', and must free 'status->rmps'.
3692 * '*status' is indeterminate if the return value is non-zero. */
3693 int
3694 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
3695 struct ofproto_cfm_status *status)
3696 {
3697 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
3698 return (ofport && ofproto->ofproto_class->get_cfm_status
3699 ? ofproto->ofproto_class->get_cfm_status(ofport, status)
3700 : EOPNOTSUPP);
3701 }
3702
3703 static enum ofperr
3704 handle_aggregate_stats_request(struct ofconn *ofconn,
3705 const struct ofp_header *oh)
3706 OVS_EXCLUDED(ofproto_mutex)
3707 {
3708 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3709 struct ofputil_flow_stats_request request;
3710 struct ofputil_aggregate_stats stats;
3711 bool unknown_packets, unknown_bytes;
3712 struct rule_criteria criteria;
3713 struct rule_collection rules;
3714 struct ofpbuf *reply;
3715 enum ofperr error;
3716 size_t i;
3717
3718 error = ofputil_decode_flow_stats_request(&request, oh);
3719 if (error) {
3720 return error;
3721 }
3722
3723 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
3724 request.cookie, request.cookie_mask,
3725 request.out_port, request.out_group);
3726
3727 ovs_mutex_lock(&ofproto_mutex);
3728 error = collect_rules_loose(ofproto, &criteria, &rules);
3729 rule_criteria_destroy(&criteria);
3730 if (!error) {
3731 rule_collection_ref(&rules);
3732 }
3733 ovs_mutex_unlock(&ofproto_mutex);
3734
3735 if (error) {
3736 return error;
3737 }
3738
3739 memset(&stats, 0, sizeof stats);
3740 unknown_packets = unknown_bytes = false;
3741 for (i = 0; i < rules.n; i++) {
3742 struct rule *rule = rules.rules[i];
3743 uint64_t packet_count;
3744 uint64_t byte_count;
3745 long long int used;
3746
3747 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
3748 &byte_count, &used);
3749
3750 if (packet_count == UINT64_MAX) {
3751 unknown_packets = true;
3752 } else {
3753 stats.packet_count += packet_count;
3754 }
3755
3756 if (byte_count == UINT64_MAX) {
3757 unknown_bytes = true;
3758 } else {
3759 stats.byte_count += byte_count;
3760 }
3761
3762 stats.flow_count++;
3763 }
3764 if (unknown_packets) {
3765 stats.packet_count = UINT64_MAX;
3766 }
3767 if (unknown_bytes) {
3768 stats.byte_count = UINT64_MAX;
3769 }
3770
3771 rule_collection_unref(&rules);
3772 rule_collection_destroy(&rules);
3773
3774 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
3775 ofconn_send_reply(ofconn, reply);
3776
3777 return 0;
3778 }
3779
3780 struct queue_stats_cbdata {
3781 struct ofport *ofport;
3782 struct list replies;
3783 long long int now;
3784 };
3785
3786 static void
3787 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
3788 const struct netdev_queue_stats *stats)
3789 {
3790 struct ofputil_queue_stats oqs;
3791
3792 oqs.port_no = cbdata->ofport->pp.port_no;
3793 oqs.queue_id = queue_id;
3794 oqs.tx_bytes = stats->tx_bytes;
3795 oqs.tx_packets = stats->tx_packets;
3796 oqs.tx_errors = stats->tx_errors;
3797 if (stats->created != LLONG_MIN) {
3798 calc_duration(stats->created, cbdata->now,
3799 &oqs.duration_sec, &oqs.duration_nsec);
3800 } else {
3801 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
3802 }
3803 ofputil_append_queue_stat(&cbdata->replies, &oqs);
3804 }
3805
3806 static void
3807 handle_queue_stats_dump_cb(uint32_t queue_id,
3808 struct netdev_queue_stats *stats,
3809 void *cbdata_)
3810 {
3811 struct queue_stats_cbdata *cbdata = cbdata_;
3812
3813 put_queue_stats(cbdata, queue_id, stats);
3814 }
3815
3816 static enum ofperr
3817 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
3818 struct queue_stats_cbdata *cbdata)
3819 {
3820 cbdata->ofport = port;
3821 if (queue_id == OFPQ_ALL) {
3822 netdev_dump_queue_stats(port->netdev,
3823 handle_queue_stats_dump_cb, cbdata);
3824 } else {
3825 struct netdev_queue_stats stats;
3826
3827 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
3828 put_queue_stats(cbdata, queue_id, &stats);
3829 } else {
3830 return OFPERR_OFPQOFC_BAD_QUEUE;
3831 }
3832 }
3833 return 0;
3834 }
3835
3836 static enum ofperr
3837 handle_queue_stats_request(struct ofconn *ofconn,
3838 const struct ofp_header *rq)
3839 {
3840 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3841 struct queue_stats_cbdata cbdata;
3842 struct ofport *port;
3843 enum ofperr error;
3844 struct ofputil_queue_stats_request oqsr;
3845
3846 COVERAGE_INC(ofproto_queue_req);
3847
3848 ofpmp_init(&cbdata.replies, rq);
3849 cbdata.now = time_msec();
3850
3851 error = ofputil_decode_queue_stats_request(rq, &oqsr);
3852 if (error) {
3853 return error;
3854 }
3855
3856 if (oqsr.port_no == OFPP_ANY) {
3857 error = OFPERR_OFPQOFC_BAD_QUEUE;
3858 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3859 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
3860 error = 0;
3861 }
3862 }
3863 } else {
3864 port = ofproto_get_port(ofproto, oqsr.port_no);
3865 error = (port
3866 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
3867 : OFPERR_OFPQOFC_BAD_PORT);
3868 }
3869 if (!error) {
3870 ofconn_send_replies(ofconn, &cbdata.replies);
3871 } else {
3872 ofpbuf_list_delete(&cbdata.replies);
3873 }
3874
3875 return error;
3876 }
3877
3878 static bool
3879 should_evict_a_rule(struct oftable *table, unsigned int extra_space)
3880 OVS_REQUIRES(ofproto_mutex)
3881 OVS_NO_THREAD_SAFETY_ANALYSIS
3882 {
3883 return classifier_count(&table->cls) + extra_space > table->max_flows;
3884 }
3885
3886 static enum ofperr
3887 evict_rules_from_table(struct oftable *table, unsigned int extra_space)
3888 OVS_REQUIRES(ofproto_mutex)
3889 {
3890 while (should_evict_a_rule(table, extra_space)) {
3891 struct rule *rule;
3892
3893 if (!choose_rule_to_evict(table, &rule)) {
3894 return OFPERR_OFPFMFC_TABLE_FULL;
3895 } else {
3896 ofproto_rule_delete__(rule, OFPRR_EVICTION);
3897 }
3898 }
3899
3900 return 0;
3901 }
3902
3903 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
3904 * in which no matching flow already exists in the flow table.
3905 *
3906 * Adds the flow specified by 'ofm', which is followed by 'n_actions'
3907 * ofp_actions, to the ofproto's flow table. Returns 0 on success, an OpenFlow
3908 * error code on failure, or OFPROTO_POSTPONE if the operation cannot be
3909 * initiated now but may be retried later.
3910 *
3911 * The caller retains ownership of 'fm->ofpacts'.
3912 *
3913 * 'ofconn' is used to retrieve the packet buffer specified in ofm->buffer_id,
3914 * if any. */
3915 static enum ofperr
3916 add_flow(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
3917 const struct flow_mod_requester *req)
3918 OVS_REQUIRES(ofproto_mutex)
3919 {
3920 const struct rule_actions *actions;
3921 struct oftable *table;
3922 struct cls_rule cr;
3923 struct rule *rule;
3924 uint8_t table_id;
3925 int error = 0;
3926
3927 if (!check_table_id(ofproto, fm->table_id)) {
3928 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3929 return error;
3930 }
3931
3932 /* Pick table. */
3933 if (fm->table_id == 0xff) {
3934 if (ofproto->ofproto_class->rule_choose_table) {
3935 error = ofproto->ofproto_class->rule_choose_table(ofproto,
3936 &fm->match,
3937 &table_id);
3938 if (error) {
3939 return error;
3940 }
3941 ovs_assert(table_id < ofproto->n_tables);
3942 } else {
3943 table_id = 0;
3944 }
3945 } else if (fm->table_id < ofproto->n_tables) {
3946 table_id = fm->table_id;
3947 } else {
3948 return OFPERR_OFPBRC_BAD_TABLE_ID;
3949 }
3950
3951 table = &ofproto->tables[table_id];
3952 if (table->flags & OFTABLE_READONLY
3953 && !(fm->flags & OFPUTIL_FF_NO_READONLY)) {
3954 return OFPERR_OFPBRC_EPERM;
3955 }
3956
3957 if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)) {
3958 if (!match_has_default_hidden_fields(&fm->match)) {
3959 VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set "
3960 "non-default values to hidden fields", ofproto->name);
3961 return OFPERR_OFPBRC_EPERM;
3962 }
3963 }
3964
3965 cls_rule_init(&cr, &fm->match, fm->priority);
3966
3967 /* Transform "add" into "modify" if there's an existing identical flow. */
3968 fat_rwlock_rdlock(&table->cls.rwlock);
3969 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr));
3970 fat_rwlock_unlock(&table->cls.rwlock);
3971 if (rule) {
3972 struct rule_collection rules;
3973
3974 cls_rule_destroy(&cr);
3975
3976 rule_collection_init(&rules);
3977 rule_collection_add(&rules, rule);
3978 fm->modify_cookie = true;
3979 error = modify_flows__(ofproto, fm, &rules, req);
3980 rule_collection_destroy(&rules);
3981
3982 return error;
3983 }
3984
3985 /* Check for overlap, if requested. */
3986 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP) {
3987 bool overlaps;
3988
3989 fat_rwlock_rdlock(&table->cls.rwlock);
3990 overlaps = classifier_rule_overlaps(&table->cls, &cr);
3991 fat_rwlock_unlock(&table->cls.rwlock);
3992
3993 if (overlaps) {
3994 cls_rule_destroy(&cr);
3995 return OFPERR_OFPFMFC_OVERLAP;
3996 }
3997 }
3998
3999 /* If necessary, evict an existing rule to clear out space. */
4000 error = evict_rules_from_table(table, 1);
4001 if (error) {
4002 cls_rule_destroy(&cr);
4003 return error;
4004 }
4005
4006 /* Allocate new rule. */
4007 rule = ofproto->ofproto_class->rule_alloc();
4008 if (!rule) {
4009 cls_rule_destroy(&cr);
4010 VLOG_WARN_RL(&rl, "%s: failed to create rule (%s)",
4011 ofproto->name, ovs_strerror(error));
4012 return ENOMEM;
4013 }
4014
4015 /* Initialize base state. */
4016 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4017 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), &cr);
4018 ovs_refcount_init(&rule->ref_count);
4019 rule->flow_cookie = fm->new_cookie;
4020 rule->created = rule->modified = time_msec();
4021
4022 ovs_mutex_init(&rule->mutex);
4023 ovs_mutex_lock(&rule->mutex);
4024 rule->idle_timeout = fm->idle_timeout;
4025 rule->hard_timeout = fm->hard_timeout;
4026 ovs_mutex_unlock(&rule->mutex);
4027
4028 *CONST_CAST(uint8_t *, &rule->table_id) = table - ofproto->tables;
4029 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4030 actions = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
4031 ovsrcu_set(&rule->actions, actions);
4032 list_init(&rule->meter_list_node);
4033 rule->eviction_group = NULL;
4034 list_init(&rule->expirable);
4035 rule->monitor_flags = 0;
4036 rule->add_seqno = 0;
4037 rule->modify_seqno = 0;
4038
4039 /* Construct rule, initializing derived state. */
4040 error = ofproto->ofproto_class->rule_construct(rule);
4041 if (error) {
4042 ofproto_rule_destroy__(rule);
4043 return error;
4044 }
4045
4046 if (fm->hard_timeout || fm->idle_timeout) {
4047 list_insert(&ofproto->expirable, &rule->expirable);
4048 }
4049 cookies_insert(ofproto, rule);
4050 eviction_group_add_rule(rule);
4051 if (actions->has_meter) {
4052 meter_insert_rule(rule);
4053 }
4054
4055 fat_rwlock_wrlock(&table->cls.rwlock);
4056 classifier_insert(&table->cls, CONST_CAST(struct cls_rule *, &rule->cr));
4057 fat_rwlock_unlock(&table->cls.rwlock);
4058
4059 error = ofproto->ofproto_class->rule_insert(rule);
4060 if (error) {
4061 oftable_remove_rule(rule);
4062 ofproto_rule_unref(rule);
4063 return error;
4064 }
4065 learned_cookies_inc(ofproto, actions);
4066
4067 if (minimask_get_vid_mask(&rule->cr.match.mask) == VLAN_VID_MASK) {
4068 if (ofproto->vlan_bitmap) {
4069 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
4070 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4071 bitmap_set1(ofproto->vlan_bitmap, vid);
4072 ofproto->vlans_changed = true;
4073 }
4074 } else {
4075 ofproto->vlans_changed = true;
4076 }
4077 }
4078
4079 ofmonitor_report(ofproto->connmgr, rule, NXFME_ADDED, 0,
4080 req ? req->ofconn : NULL, req ? req->xid : 0);
4081
4082 return req ? send_buffered_packet(req->ofconn, fm->buffer_id, rule) : 0;
4083 }
4084 \f
4085 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4086
4087 /* Modifies the rules listed in 'rules', changing their actions to match those
4088 * in 'fm'.
4089 *
4090 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4091 * if any.
4092 *
4093 * Returns 0 on success, otherwise an OpenFlow error code. */
4094 static enum ofperr
4095 modify_flows__(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4096 const struct rule_collection *rules,
4097 const struct flow_mod_requester *req)
4098 OVS_REQUIRES(ofproto_mutex)
4099 {
4100 struct list dead_cookies = LIST_INITIALIZER(&dead_cookies);
4101 enum nx_flow_update_event event;
4102 enum ofperr error;
4103 size_t i;
4104
4105 if (ofproto->ofproto_class->rule_premodify_actions) {
4106 for (i = 0; i < rules->n; i++) {
4107 struct rule *rule = rules->rules[i];
4108
4109 error = ofproto->ofproto_class->rule_premodify_actions(
4110 rule, fm->ofpacts, fm->ofpacts_len);
4111 if (error) {
4112 return error;
4113 }
4114 }
4115 }
4116
4117 event = fm->command == OFPFC_ADD ? NXFME_ADDED : NXFME_MODIFIED;
4118 for (i = 0; i < rules->n; i++) {
4119 struct rule *rule = rules->rules[i];
4120
4121 /* 'fm' says that */
4122 bool change_cookie = (fm->modify_cookie
4123 && fm->new_cookie != OVS_BE64_MAX
4124 && fm->new_cookie != rule->flow_cookie);
4125
4126 const struct rule_actions *actions = rule_get_actions(rule);
4127 bool change_actions = !ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
4128 actions->ofpacts,
4129 actions->ofpacts_len);
4130
4131 bool reset_counters = (fm->flags & OFPUTIL_FF_RESET_COUNTS) != 0;
4132
4133 long long int now = time_msec();
4134
4135 /* FIXME: Implement OFPFUTIL_FF_RESET_COUNTS */
4136
4137 if (change_cookie) {
4138 cookies_remove(ofproto, rule);
4139 }
4140
4141 ovs_mutex_lock(&rule->mutex);
4142 if (fm->command == OFPFC_ADD) {
4143 rule->idle_timeout = fm->idle_timeout;
4144 rule->hard_timeout = fm->hard_timeout;
4145 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4146 rule->created = now;
4147 }
4148 if (change_cookie) {
4149 rule->flow_cookie = fm->new_cookie;
4150 }
4151 rule->modified = now;
4152 ovs_mutex_unlock(&rule->mutex);
4153
4154 if (change_cookie) {
4155 cookies_insert(ofproto, rule);
4156 }
4157 if (fm->command == OFPFC_ADD) {
4158 if (fm->idle_timeout || fm->hard_timeout) {
4159 if (!rule->eviction_group) {
4160 eviction_group_add_rule(rule);
4161 }
4162 } else {
4163 eviction_group_remove_rule(rule);
4164 }
4165 }
4166
4167 if (change_actions) {
4168 ovsrcu_set(&rule->actions, rule_actions_create(fm->ofpacts,
4169 fm->ofpacts_len));
4170 rule_actions_destroy(actions);
4171 }
4172
4173 if (change_actions || reset_counters) {
4174 ofproto->ofproto_class->rule_modify_actions(rule, reset_counters);
4175 }
4176
4177 if (event != NXFME_MODIFIED || change_actions || change_cookie) {
4178 ofmonitor_report(ofproto->connmgr, rule, event, 0,
4179 req ? req->ofconn : NULL, req ? req->xid : 0);
4180 }
4181
4182 if (change_actions) {
4183 learned_cookies_inc(ofproto, rule_get_actions(rule));
4184 learned_cookies_dec(ofproto, actions, &dead_cookies);
4185 }
4186 }
4187 learned_cookies_flush(ofproto, &dead_cookies);
4188
4189 if (fm->buffer_id != UINT32_MAX && req) {
4190 error = send_buffered_packet(req->ofconn, fm->buffer_id,
4191 rules->rules[0]);
4192 }
4193
4194 return error;
4195 }
4196
4197 static enum ofperr
4198 modify_flows_add(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4199 const struct flow_mod_requester *req)
4200 OVS_REQUIRES(ofproto_mutex)
4201 {
4202 if (fm->cookie_mask != htonll(0) || fm->new_cookie == OVS_BE64_MAX) {
4203 return 0;
4204 }
4205 return add_flow(ofproto, fm, req);
4206 }
4207
4208 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4209 * failure.
4210 *
4211 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4212 * if any. */
4213 static enum ofperr
4214 modify_flows_loose(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4215 const struct flow_mod_requester *req)
4216 OVS_REQUIRES(ofproto_mutex)
4217 {
4218 struct rule_criteria criteria;
4219 struct rule_collection rules;
4220 int error;
4221
4222 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4223 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4224 rule_criteria_require_rw(&criteria,
4225 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
4226 error = collect_rules_loose(ofproto, &criteria, &rules);
4227 rule_criteria_destroy(&criteria);
4228
4229 if (!error) {
4230 error = (rules.n > 0
4231 ? modify_flows__(ofproto, fm, &rules, req)
4232 : modify_flows_add(ofproto, fm, req));
4233 }
4234
4235 rule_collection_destroy(&rules);
4236
4237 return error;
4238 }
4239
4240 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4241 * code on failure. */
4242 static enum ofperr
4243 modify_flow_strict(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4244 const struct flow_mod_requester *req)
4245 OVS_REQUIRES(ofproto_mutex)
4246 {
4247 struct rule_criteria criteria;
4248 struct rule_collection rules;
4249 int error;
4250
4251 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4252 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4253 rule_criteria_require_rw(&criteria,
4254 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
4255 error = collect_rules_strict(ofproto, &criteria, &rules);
4256 rule_criteria_destroy(&criteria);
4257
4258 if (!error) {
4259 if (rules.n == 0) {
4260 error = modify_flows_add(ofproto, fm, req);
4261 } else if (rules.n == 1) {
4262 error = modify_flows__(ofproto, fm, &rules, req);
4263 }
4264 }
4265
4266 rule_collection_destroy(&rules);
4267
4268 return error;
4269 }
4270 \f
4271 /* OFPFC_DELETE implementation. */
4272
4273 /* Deletes the rules listed in 'rules'. */
4274 static void
4275 delete_flows__(const struct rule_collection *rules,
4276 enum ofp_flow_removed_reason reason,
4277 const struct flow_mod_requester *req)
4278 OVS_REQUIRES(ofproto_mutex)
4279 {
4280 if (rules->n) {
4281 struct list dead_cookies = LIST_INITIALIZER(&dead_cookies);
4282 struct ofproto *ofproto = rules->rules[0]->ofproto;
4283 size_t i;
4284
4285 for (i = 0; i < rules->n; i++) {
4286 struct rule *rule = rules->rules[i];
4287 const struct rule_actions *actions = rule_get_actions(rule);
4288
4289 ofproto_rule_send_removed(rule, reason);
4290
4291 ofmonitor_report(ofproto->connmgr, rule, NXFME_DELETED, reason,
4292 req ? req->ofconn : NULL, req ? req->xid : 0);
4293 oftable_remove_rule(rule);
4294 ofproto->ofproto_class->rule_delete(rule);
4295
4296 learned_cookies_dec(ofproto, actions, &dead_cookies);
4297 }
4298 learned_cookies_flush(ofproto, &dead_cookies);
4299 ofmonitor_flush(ofproto->connmgr);
4300 }
4301 }
4302
4303 /* Implements OFPFC_DELETE. */
4304 static enum ofperr
4305 delete_flows_loose(struct ofproto *ofproto,
4306 const struct ofputil_flow_mod *fm,
4307 const struct flow_mod_requester *req)
4308 OVS_REQUIRES(ofproto_mutex)
4309 {
4310 struct rule_criteria criteria;
4311 struct rule_collection rules;
4312 enum ofperr error;
4313
4314 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0,
4315 fm->cookie, fm->cookie_mask,
4316 fm->out_port, fm->out_group);
4317 rule_criteria_require_rw(&criteria,
4318 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
4319 error = collect_rules_loose(ofproto, &criteria, &rules);
4320 rule_criteria_destroy(&criteria);
4321
4322 if (!error && rules.n > 0) {
4323 delete_flows__(&rules, fm->delete_reason, req);
4324 }
4325 rule_collection_destroy(&rules);
4326
4327 return error;
4328 }
4329
4330 /* Implements OFPFC_DELETE_STRICT. */
4331 static enum ofperr
4332 delete_flow_strict(struct ofproto *ofproto, const struct ofputil_flow_mod *fm,
4333 const struct flow_mod_requester *req)
4334 OVS_REQUIRES(ofproto_mutex)
4335 {
4336 struct rule_criteria criteria;
4337 struct rule_collection rules;
4338 enum ofperr error;
4339
4340 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4341 fm->cookie, fm->cookie_mask,
4342 fm->out_port, fm->out_group);
4343 rule_criteria_require_rw(&criteria,
4344 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
4345 error = collect_rules_strict(ofproto, &criteria, &rules);
4346 rule_criteria_destroy(&criteria);
4347
4348 if (!error && rules.n > 0) {
4349 delete_flows__(&rules, fm->delete_reason, req);
4350 }
4351 rule_collection_destroy(&rules);
4352
4353 return error;
4354 }
4355
4356 static void
4357 ofproto_rule_send_removed(struct rule *rule, uint8_t reason)
4358 OVS_REQUIRES(ofproto_mutex)
4359 {
4360 struct ofputil_flow_removed fr;
4361 long long int used;
4362
4363 if (rule_is_hidden(rule) ||
4364 !(rule->flags & OFPUTIL_FF_SEND_FLOW_REM)) {
4365 return;
4366 }
4367
4368 minimatch_expand(&rule->cr.match, &fr.match);
4369 fr.priority = rule->cr.priority;
4370 fr.cookie = rule->flow_cookie;
4371 fr.reason = reason;
4372 fr.table_id = rule->table_id;
4373 calc_duration(rule->created, time_msec(),
4374 &fr.duration_sec, &fr.duration_nsec);
4375 ovs_mutex_lock(&rule->mutex);
4376 fr.idle_timeout = rule->idle_timeout;
4377 fr.hard_timeout = rule->hard_timeout;
4378 ovs_mutex_unlock(&rule->mutex);
4379 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
4380 &fr.byte_count, &used);
4381
4382 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
4383 }
4384
4385 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
4386 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
4387 * ofproto.
4388 *
4389 * ofproto implementation ->run() functions should use this function to expire
4390 * OpenFlow flows. */
4391 void
4392 ofproto_rule_expire(struct rule *rule, uint8_t reason)
4393 OVS_REQUIRES(ofproto_mutex)
4394 {
4395 ofproto_rule_delete__(rule, reason);
4396 }
4397
4398 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
4399 * means "infinite". */
4400 static void
4401 reduce_timeout(uint16_t max, uint16_t *timeout)
4402 {
4403 if (max && (!*timeout || *timeout > max)) {
4404 *timeout = max;
4405 }
4406 }
4407
4408 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
4409 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
4410 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
4411 *
4412 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
4413 void
4414 ofproto_rule_reduce_timeouts(struct rule *rule,
4415 uint16_t idle_timeout, uint16_t hard_timeout)
4416 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
4417 {
4418 if (!idle_timeout && !hard_timeout) {
4419 return;
4420 }
4421
4422 ovs_mutex_lock(&ofproto_mutex);
4423 if (list_is_empty(&rule->expirable)) {
4424 list_insert(&rule->ofproto->expirable, &rule->expirable);
4425 }
4426 ovs_mutex_unlock(&ofproto_mutex);
4427
4428 ovs_mutex_lock(&rule->mutex);
4429 reduce_timeout(idle_timeout, &rule->idle_timeout);
4430 reduce_timeout(hard_timeout, &rule->hard_timeout);
4431 ovs_mutex_unlock(&rule->mutex);
4432 }
4433 \f
4434 static enum ofperr
4435 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
4436 OVS_EXCLUDED(ofproto_mutex)
4437 {
4438 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4439 struct ofputil_flow_mod fm;
4440 uint64_t ofpacts_stub[1024 / 8];
4441 struct ofpbuf ofpacts;
4442 enum ofperr error;
4443
4444 error = reject_slave_controller(ofconn);
4445 if (error) {
4446 goto exit;
4447 }
4448
4449 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
4450 error = ofputil_decode_flow_mod(&fm, oh, ofconn_get_protocol(ofconn),
4451 &ofpacts,
4452 u16_to_ofp(ofproto->max_ports),
4453 ofproto->n_tables);
4454 if (!error) {
4455 error = ofproto_check_ofpacts(ofproto, fm.ofpacts, fm.ofpacts_len);
4456 }
4457 if (!error) {
4458 struct flow_mod_requester req;
4459
4460 req.ofconn = ofconn;
4461 req.xid = oh->xid;
4462 error = handle_flow_mod__(ofproto, &fm, &req);
4463 }
4464 if (error) {
4465 goto exit_free_ofpacts;
4466 }
4467
4468 ofconn_report_flow_mod(ofconn, fm.command);
4469
4470 exit_free_ofpacts:
4471 ofpbuf_uninit(&ofpacts);
4472 exit:
4473 return error;
4474 }
4475
4476 static enum ofperr
4477 handle_flow_mod__(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4478 const struct flow_mod_requester *req)
4479 OVS_EXCLUDED(ofproto_mutex)
4480 {
4481 enum ofperr error;
4482
4483 ovs_mutex_lock(&ofproto_mutex);
4484 switch (fm->command) {
4485 case OFPFC_ADD:
4486 error = add_flow(ofproto, fm, req);
4487 break;
4488
4489 case OFPFC_MODIFY:
4490 error = modify_flows_loose(ofproto, fm, req);
4491 break;
4492
4493 case OFPFC_MODIFY_STRICT:
4494 error = modify_flow_strict(ofproto, fm, req);
4495 break;
4496
4497 case OFPFC_DELETE:
4498 error = delete_flows_loose(ofproto, fm, req);
4499 break;
4500
4501 case OFPFC_DELETE_STRICT:
4502 error = delete_flow_strict(ofproto, fm, req);
4503 break;
4504
4505 default:
4506 if (fm->command > 0xff) {
4507 VLOG_WARN_RL(&rl, "%s: flow_mod has explicit table_id but "
4508 "flow_mod_table_id extension is not enabled",
4509 ofproto->name);
4510 }
4511 error = OFPERR_OFPFMFC_BAD_COMMAND;
4512 break;
4513 }
4514 ofmonitor_flush(ofproto->connmgr);
4515 ovs_mutex_unlock(&ofproto_mutex);
4516
4517 run_rule_executes(ofproto);
4518 return error;
4519 }
4520
4521 static enum ofperr
4522 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
4523 {
4524 struct ofputil_role_request request;
4525 struct ofputil_role_request reply;
4526 struct ofpbuf *buf;
4527 enum ofperr error;
4528
4529 error = ofputil_decode_role_message(oh, &request);
4530 if (error) {
4531 return error;
4532 }
4533
4534 if (request.role != OFPCR12_ROLE_NOCHANGE) {
4535 if (request.have_generation_id
4536 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
4537 return OFPERR_OFPRRFC_STALE;
4538 }
4539
4540 ofconn_set_role(ofconn, request.role);
4541 }
4542
4543 reply.role = ofconn_get_role(ofconn);
4544 reply.have_generation_id = ofconn_get_master_election_id(
4545 ofconn, &reply.generation_id);
4546 buf = ofputil_encode_role_reply(oh, &reply);
4547 ofconn_send_reply(ofconn, buf);
4548
4549 return 0;
4550 }
4551
4552 static enum ofperr
4553 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
4554 const struct ofp_header *oh)
4555 {
4556 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
4557 enum ofputil_protocol cur, next;
4558
4559 cur = ofconn_get_protocol(ofconn);
4560 next = ofputil_protocol_set_tid(cur, msg->set != 0);
4561 ofconn_set_protocol(ofconn, next);
4562
4563 return 0;
4564 }
4565
4566 static enum ofperr
4567 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
4568 {
4569 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
4570 enum ofputil_protocol cur, next;
4571 enum ofputil_protocol next_base;
4572
4573 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
4574 if (!next_base) {
4575 return OFPERR_OFPBRC_EPERM;
4576 }
4577
4578 cur = ofconn_get_protocol(ofconn);
4579 next = ofputil_protocol_set_base(cur, next_base);
4580 ofconn_set_protocol(ofconn, next);
4581
4582 return 0;
4583 }
4584
4585 static enum ofperr
4586 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
4587 const struct ofp_header *oh)
4588 {
4589 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
4590 uint32_t format;
4591
4592 format = ntohl(msg->format);
4593 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
4594 return OFPERR_OFPBRC_EPERM;
4595 }
4596
4597 ofconn_set_packet_in_format(ofconn, format);
4598 return 0;
4599 }
4600
4601 static enum ofperr
4602 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
4603 {
4604 const struct nx_async_config *msg = ofpmsg_body(oh);
4605 uint32_t master[OAM_N_TYPES];
4606 uint32_t slave[OAM_N_TYPES];
4607
4608 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
4609 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
4610 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
4611
4612 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
4613 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
4614 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
4615
4616 ofconn_set_async_config(ofconn, master, slave);
4617 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
4618 !ofconn_get_miss_send_len(ofconn)) {
4619 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
4620 }
4621
4622 return 0;
4623 }
4624
4625 static enum ofperr
4626 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
4627 {
4628 struct ofpbuf *buf;
4629 uint32_t master[OAM_N_TYPES];
4630 uint32_t slave[OAM_N_TYPES];
4631 struct nx_async_config *msg;
4632
4633 ofconn_get_async_config(ofconn, master, slave);
4634 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
4635 msg = ofpbuf_put_zeros(buf, sizeof *msg);
4636
4637 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
4638 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
4639 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
4640
4641 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
4642 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
4643 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
4644
4645 ofconn_send_reply(ofconn, buf);
4646
4647 return 0;
4648 }
4649
4650 static enum ofperr
4651 handle_nxt_set_controller_id(struct ofconn *ofconn,
4652 const struct ofp_header *oh)
4653 {
4654 const struct nx_controller_id *nci = ofpmsg_body(oh);
4655
4656 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
4657 return OFPERR_NXBRC_MUST_BE_ZERO;
4658 }
4659
4660 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
4661 return 0;
4662 }
4663
4664 static enum ofperr
4665 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
4666 {
4667 struct ofpbuf *buf;
4668
4669 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
4670 ? OFPRAW_OFPT10_BARRIER_REPLY
4671 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
4672 ofconn_send_reply(ofconn, buf);
4673 return 0;
4674 }
4675
4676 static void
4677 ofproto_compose_flow_refresh_update(const struct rule *rule,
4678 enum nx_flow_monitor_flags flags,
4679 struct list *msgs)
4680 OVS_REQUIRES(ofproto_mutex)
4681 {
4682 const struct rule_actions *actions;
4683 struct ofputil_flow_update fu;
4684 struct match match;
4685
4686 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
4687 ? NXFME_ADDED : NXFME_MODIFIED);
4688 fu.reason = 0;
4689 ovs_mutex_lock(&rule->mutex);
4690 fu.idle_timeout = rule->idle_timeout;
4691 fu.hard_timeout = rule->hard_timeout;
4692 ovs_mutex_unlock(&rule->mutex);
4693 fu.table_id = rule->table_id;
4694 fu.cookie = rule->flow_cookie;
4695 minimatch_expand(&rule->cr.match, &match);
4696 fu.match = &match;
4697 fu.priority = rule->cr.priority;
4698
4699 actions = flags & NXFMF_ACTIONS ? rule_get_actions(rule) : NULL;
4700 fu.ofpacts = actions ? actions->ofpacts : NULL;
4701 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
4702
4703 if (list_is_empty(msgs)) {
4704 ofputil_start_flow_update(msgs);
4705 }
4706 ofputil_append_flow_update(&fu, msgs);
4707 }
4708
4709 void
4710 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
4711 struct list *msgs)
4712 OVS_REQUIRES(ofproto_mutex)
4713 {
4714 size_t i;
4715
4716 for (i = 0; i < rules->n; i++) {
4717 struct rule *rule = rules->rules[i];
4718 enum nx_flow_monitor_flags flags = rule->monitor_flags;
4719 rule->monitor_flags = 0;
4720
4721 ofproto_compose_flow_refresh_update(rule, flags, msgs);
4722 }
4723 }
4724
4725 static void
4726 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
4727 struct rule *rule, uint64_t seqno,
4728 struct rule_collection *rules)
4729 OVS_REQUIRES(ofproto_mutex)
4730 {
4731 enum nx_flow_monitor_flags update;
4732
4733 if (rule_is_hidden(rule)) {
4734 return;
4735 }
4736
4737 if (!ofproto_rule_has_out_port(rule, m->out_port)) {
4738 return;
4739 }
4740
4741 if (seqno) {
4742 if (rule->add_seqno > seqno) {
4743 update = NXFMF_ADD | NXFMF_MODIFY;
4744 } else if (rule->modify_seqno > seqno) {
4745 update = NXFMF_MODIFY;
4746 } else {
4747 return;
4748 }
4749
4750 if (!(m->flags & update)) {
4751 return;
4752 }
4753 } else {
4754 update = NXFMF_INITIAL;
4755 }
4756
4757 if (!rule->monitor_flags) {
4758 rule_collection_add(rules, rule);
4759 }
4760 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
4761 }
4762
4763 static void
4764 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
4765 uint64_t seqno,
4766 struct rule_collection *rules)
4767 OVS_REQUIRES(ofproto_mutex)
4768 {
4769 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
4770 const struct oftable *table;
4771 struct cls_rule target;
4772
4773 cls_rule_init_from_minimatch(&target, &m->match, 0);
4774 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
4775 struct cls_cursor cursor;
4776 struct rule *rule;
4777
4778 fat_rwlock_rdlock(&table->cls.rwlock);
4779 cls_cursor_init(&cursor, &table->cls, &target);
4780 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
4781 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
4782 }
4783 fat_rwlock_unlock(&table->cls.rwlock);
4784 }
4785 cls_rule_destroy(&target);
4786 }
4787
4788 static void
4789 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
4790 struct rule_collection *rules)
4791 OVS_REQUIRES(ofproto_mutex)
4792 {
4793 if (m->flags & NXFMF_INITIAL) {
4794 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
4795 }
4796 }
4797
4798 void
4799 ofmonitor_collect_resume_rules(struct ofmonitor *m,
4800 uint64_t seqno, struct rule_collection *rules)
4801 OVS_REQUIRES(ofproto_mutex)
4802 {
4803 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
4804 }
4805
4806 static enum ofperr
4807 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
4808 OVS_EXCLUDED(ofproto_mutex)
4809 {
4810 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4811 struct ofmonitor **monitors;
4812 size_t n_monitors, allocated_monitors;
4813 struct rule_collection rules;
4814 struct list replies;
4815 enum ofperr error;
4816 struct ofpbuf b;
4817 size_t i;
4818
4819 error = 0;
4820 ofpbuf_use_const(&b, oh, ntohs(oh->length));
4821 monitors = NULL;
4822 n_monitors = allocated_monitors = 0;
4823
4824 ovs_mutex_lock(&ofproto_mutex);
4825 for (;;) {
4826 struct ofputil_flow_monitor_request request;
4827 struct ofmonitor *m;
4828 int retval;
4829
4830 retval = ofputil_decode_flow_monitor_request(&request, &b);
4831 if (retval == EOF) {
4832 break;
4833 } else if (retval) {
4834 error = retval;
4835 goto error;
4836 }
4837
4838 if (request.table_id != 0xff
4839 && request.table_id >= ofproto->n_tables) {
4840 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4841 goto error;
4842 }
4843
4844 error = ofmonitor_create(&request, ofconn, &m);
4845 if (error) {
4846 goto error;
4847 }
4848
4849 if (n_monitors >= allocated_monitors) {
4850 monitors = x2nrealloc(monitors, &allocated_monitors,
4851 sizeof *monitors);
4852 }
4853 monitors[n_monitors++] = m;
4854 }
4855
4856 rule_collection_init(&rules);
4857 for (i = 0; i < n_monitors; i++) {
4858 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
4859 }
4860
4861 ofpmp_init(&replies, oh);
4862 ofmonitor_compose_refresh_updates(&rules, &replies);
4863 ovs_mutex_unlock(&ofproto_mutex);
4864
4865 rule_collection_destroy(&rules);
4866
4867 ofconn_send_replies(ofconn, &replies);
4868 free(monitors);
4869
4870 return 0;
4871
4872 error:
4873 for (i = 0; i < n_monitors; i++) {
4874 ofmonitor_destroy(monitors[i]);
4875 }
4876 free(monitors);
4877 ovs_mutex_unlock(&ofproto_mutex);
4878
4879 return error;
4880 }
4881
4882 static enum ofperr
4883 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
4884 OVS_EXCLUDED(ofproto_mutex)
4885 {
4886 struct ofmonitor *m;
4887 enum ofperr error;
4888 uint32_t id;
4889
4890 id = ofputil_decode_flow_monitor_cancel(oh);
4891
4892 ovs_mutex_lock(&ofproto_mutex);
4893 m = ofmonitor_lookup(ofconn, id);
4894 if (m) {
4895 ofmonitor_destroy(m);
4896 error = 0;
4897 } else {
4898 error = OFPERR_NXBRC_FM_BAD_ID;
4899 }
4900 ovs_mutex_unlock(&ofproto_mutex);
4901
4902 return error;
4903 }
4904
4905 /* Meters implementation.
4906 *
4907 * Meter table entry, indexed by the OpenFlow meter_id.
4908 * 'created' is used to compute the duration for meter stats.
4909 * 'list rules' is needed so that we can delete the dependent rules when the
4910 * meter table entry is deleted.
4911 * 'provider_meter_id' is for the provider's private use.
4912 */
4913 struct meter {
4914 long long int created; /* Time created. */
4915 struct list rules; /* List of "struct rule_dpif"s. */
4916 ofproto_meter_id provider_meter_id;
4917 uint16_t flags; /* Meter flags. */
4918 uint16_t n_bands; /* Number of meter bands. */
4919 struct ofputil_meter_band *bands;
4920 };
4921
4922 /*
4923 * This is used in instruction validation at flow set-up time,
4924 * as flows may not use non-existing meters.
4925 * Return value of UINT32_MAX signifies an invalid meter.
4926 */
4927 static uint32_t
4928 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
4929 {
4930 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
4931 const struct meter *meter = ofproto->meters[of_meter_id];
4932 if (meter) {
4933 return meter->provider_meter_id.uint32;
4934 }
4935 }
4936 return UINT32_MAX;
4937 }
4938
4939 /* Finds the meter invoked by 'rule''s actions and adds 'rule' to the meter's
4940 * list of rules. */
4941 static void
4942 meter_insert_rule(struct rule *rule)
4943 {
4944 const struct rule_actions *a = rule_get_actions(rule);
4945 uint32_t meter_id = ofpacts_get_meter(a->ofpacts, a->ofpacts_len);
4946 struct meter *meter = rule->ofproto->meters[meter_id];
4947
4948 list_insert(&meter->rules, &rule->meter_list_node);
4949 }
4950
4951 static void
4952 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
4953 {
4954 free(meter->bands);
4955
4956 meter->flags = config->flags;
4957 meter->n_bands = config->n_bands;
4958 meter->bands = xmemdup(config->bands,
4959 config->n_bands * sizeof *meter->bands);
4960 }
4961
4962 static struct meter *
4963 meter_create(const struct ofputil_meter_config *config,
4964 ofproto_meter_id provider_meter_id)
4965 {
4966 struct meter *meter;
4967
4968 meter = xzalloc(sizeof *meter);
4969 meter->provider_meter_id = provider_meter_id;
4970 meter->created = time_msec();
4971 list_init(&meter->rules);
4972
4973 meter_update(meter, config);
4974
4975 return meter;
4976 }
4977
4978 static void
4979 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
4980 OVS_REQUIRES(ofproto_mutex)
4981 {
4982 uint32_t mid;
4983 for (mid = first; mid <= last; ++mid) {
4984 struct meter *meter = ofproto->meters[mid];
4985 if (meter) {
4986 ofproto->meters[mid] = NULL;
4987 ofproto->ofproto_class->meter_del(ofproto,
4988 meter->provider_meter_id);
4989 free(meter->bands);
4990 free(meter);
4991 }
4992 }
4993 }
4994
4995 static enum ofperr
4996 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
4997 {
4998 ofproto_meter_id provider_meter_id = { UINT32_MAX };
4999 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5000 enum ofperr error;
5001
5002 if (*meterp) {
5003 return OFPERR_OFPMMFC_METER_EXISTS;
5004 }
5005
5006 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5007 &mm->meter);
5008 if (!error) {
5009 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5010 *meterp = meter_create(&mm->meter, provider_meter_id);
5011 }
5012 return error;
5013 }
5014
5015 static enum ofperr
5016 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5017 {
5018 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5019 enum ofperr error;
5020 uint32_t provider_meter_id;
5021
5022 if (!meter) {
5023 return OFPERR_OFPMMFC_UNKNOWN_METER;
5024 }
5025
5026 provider_meter_id = meter->provider_meter_id.uint32;
5027 error = ofproto->ofproto_class->meter_set(ofproto,
5028 &meter->provider_meter_id,
5029 &mm->meter);
5030 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5031 if (!error) {
5032 meter_update(meter, &mm->meter);
5033 }
5034 return error;
5035 }
5036
5037 static enum ofperr
5038 handle_delete_meter(struct ofconn *ofconn, struct ofputil_meter_mod *mm)
5039 OVS_EXCLUDED(ofproto_mutex)
5040 {
5041 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5042 uint32_t meter_id = mm->meter.meter_id;
5043 struct rule_collection rules;
5044 enum ofperr error = 0;
5045 uint32_t first, last;
5046
5047 if (meter_id == OFPM13_ALL) {
5048 first = 1;
5049 last = ofproto->meter_features.max_meters;
5050 } else {
5051 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5052 return 0;
5053 }
5054 first = last = meter_id;
5055 }
5056
5057 /* First delete the rules that use this meter. If any of those rules are
5058 * currently being modified, postpone the whole operation until later. */
5059 rule_collection_init(&rules);
5060 ovs_mutex_lock(&ofproto_mutex);
5061 for (meter_id = first; meter_id <= last; ++meter_id) {
5062 struct meter *meter = ofproto->meters[meter_id];
5063 if (meter && !list_is_empty(&meter->rules)) {
5064 struct rule *rule;
5065
5066 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5067 rule_collection_add(&rules, rule);
5068 }
5069 }
5070 }
5071 if (rules.n > 0) {
5072 delete_flows__(&rules, OFPRR_METER_DELETE, NULL);
5073 }
5074
5075 /* Delete the meters. */
5076 meter_delete(ofproto, first, last);
5077
5078 ovs_mutex_unlock(&ofproto_mutex);
5079 rule_collection_destroy(&rules);
5080
5081 return error;
5082 }
5083
5084 static enum ofperr
5085 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5086 {
5087 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5088 struct ofputil_meter_mod mm;
5089 uint64_t bands_stub[256 / 8];
5090 struct ofpbuf bands;
5091 uint32_t meter_id;
5092 enum ofperr error;
5093
5094 error = reject_slave_controller(ofconn);
5095 if (error) {
5096 return error;
5097 }
5098
5099 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5100
5101 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5102 if (error) {
5103 goto exit_free_bands;
5104 }
5105
5106 meter_id = mm.meter.meter_id;
5107
5108 if (mm.command != OFPMC13_DELETE) {
5109 /* Fails also when meters are not implemented by the provider. */
5110 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5111 error = OFPERR_OFPMMFC_INVALID_METER;
5112 goto exit_free_bands;
5113 } else if (meter_id > ofproto->meter_features.max_meters) {
5114 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5115 goto exit_free_bands;
5116 }
5117 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5118 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5119 goto exit_free_bands;
5120 }
5121 }
5122
5123 switch (mm.command) {
5124 case OFPMC13_ADD:
5125 error = handle_add_meter(ofproto, &mm);
5126 break;
5127
5128 case OFPMC13_MODIFY:
5129 error = handle_modify_meter(ofproto, &mm);
5130 break;
5131
5132 case OFPMC13_DELETE:
5133 error = handle_delete_meter(ofconn, &mm);
5134 break;
5135
5136 default:
5137 error = OFPERR_OFPMMFC_BAD_COMMAND;
5138 break;
5139 }
5140
5141 exit_free_bands:
5142 ofpbuf_uninit(&bands);
5143 return error;
5144 }
5145
5146 static enum ofperr
5147 handle_meter_features_request(struct ofconn *ofconn,
5148 const struct ofp_header *request)
5149 {
5150 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5151 struct ofputil_meter_features features;
5152 struct ofpbuf *b;
5153
5154 if (ofproto->ofproto_class->meter_get_features) {
5155 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5156 } else {
5157 memset(&features, 0, sizeof features);
5158 }
5159 b = ofputil_encode_meter_features_reply(&features, request);
5160
5161 ofconn_send_reply(ofconn, b);
5162 return 0;
5163 }
5164
5165 static enum ofperr
5166 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5167 enum ofptype type)
5168 {
5169 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5170 struct list replies;
5171 uint64_t bands_stub[256 / 8];
5172 struct ofpbuf bands;
5173 uint32_t meter_id, first, last;
5174
5175 ofputil_decode_meter_request(request, &meter_id);
5176
5177 if (meter_id == OFPM13_ALL) {
5178 first = 1;
5179 last = ofproto->meter_features.max_meters;
5180 } else {
5181 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5182 !ofproto->meters[meter_id]) {
5183 return OFPERR_OFPMMFC_UNKNOWN_METER;
5184 }
5185 first = last = meter_id;
5186 }
5187
5188 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5189 ofpmp_init(&replies, request);
5190
5191 for (meter_id = first; meter_id <= last; ++meter_id) {
5192 struct meter *meter = ofproto->meters[meter_id];
5193 if (!meter) {
5194 continue; /* Skip non-existing meters. */
5195 }
5196 if (type == OFPTYPE_METER_STATS_REQUEST) {
5197 struct ofputil_meter_stats stats;
5198
5199 stats.meter_id = meter_id;
5200
5201 /* Provider sets the packet and byte counts, we do the rest. */
5202 stats.flow_count = list_size(&meter->rules);
5203 calc_duration(meter->created, time_msec(),
5204 &stats.duration_sec, &stats.duration_nsec);
5205 stats.n_bands = meter->n_bands;
5206 ofpbuf_clear(&bands);
5207 stats.bands
5208 = ofpbuf_put_uninit(&bands,
5209 meter->n_bands * sizeof *stats.bands);
5210
5211 if (!ofproto->ofproto_class->meter_get(ofproto,
5212 meter->provider_meter_id,
5213 &stats)) {
5214 ofputil_append_meter_stats(&replies, &stats);
5215 }
5216 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
5217 struct ofputil_meter_config config;
5218
5219 config.meter_id = meter_id;
5220 config.flags = meter->flags;
5221 config.n_bands = meter->n_bands;
5222 config.bands = meter->bands;
5223 ofputil_append_meter_config(&replies, &config);
5224 }
5225 }
5226
5227 ofconn_send_replies(ofconn, &replies);
5228 ofpbuf_uninit(&bands);
5229 return 0;
5230 }
5231
5232 static bool
5233 ofproto_group_lookup__(const struct ofproto *ofproto, uint32_t group_id,
5234 struct ofgroup **group)
5235 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5236 {
5237 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
5238 hash_int(group_id, 0), &ofproto->groups) {
5239 if ((*group)->group_id == group_id) {
5240 return true;
5241 }
5242 }
5243
5244 return false;
5245 }
5246
5247 /* If the group exists, this function increments the groups's reference count.
5248 *
5249 * Make sure to call ofproto_group_unref() after no longer needing to maintain
5250 * a reference to the group. */
5251 bool
5252 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
5253 struct ofgroup **group)
5254 {
5255 bool found;
5256
5257 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5258 found = ofproto_group_lookup__(ofproto, group_id, group);
5259 if (found) {
5260 ofproto_group_ref(*group);
5261 }
5262 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5263 return found;
5264 }
5265
5266 static bool
5267 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
5268 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
5269 {
5270 struct ofgroup *grp;
5271
5272 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
5273 hash_int(group_id, 0), &ofproto->groups) {
5274 if (grp->group_id == group_id) {
5275 return true;
5276 }
5277 }
5278 return false;
5279 }
5280
5281 static bool
5282 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
5283 OVS_EXCLUDED(ofproto->groups_rwlock)
5284 {
5285 bool exists;
5286
5287 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5288 exists = ofproto_group_exists__(ofproto, group_id);
5289 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5290
5291 return exists;
5292 }
5293
5294 static uint32_t
5295 group_get_ref_count(struct ofgroup *group)
5296 OVS_EXCLUDED(ofproto_mutex)
5297 {
5298 struct ofproto *ofproto = CONST_CAST(struct ofproto *, group->ofproto);
5299 struct rule_criteria criteria;
5300 struct rule_collection rules;
5301 struct match match;
5302 enum ofperr error;
5303 uint32_t count;
5304
5305 match_init_catchall(&match);
5306 rule_criteria_init(&criteria, 0xff, &match, 0, htonll(0), htonll(0),
5307 OFPP_ANY, group->group_id);
5308 ovs_mutex_lock(&ofproto_mutex);
5309 error = collect_rules_loose(ofproto, &criteria, &rules);
5310 ovs_mutex_unlock(&ofproto_mutex);
5311 rule_criteria_destroy(&criteria);
5312
5313 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
5314
5315 rule_collection_destroy(&rules);
5316 return count;
5317 }
5318
5319 static void
5320 append_group_stats(struct ofgroup *group, struct list *replies)
5321 {
5322 struct ofputil_group_stats ogs;
5323 const struct ofproto *ofproto = group->ofproto;
5324 long long int now = time_msec();
5325 int error;
5326
5327 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
5328
5329 /* Provider sets the packet and byte counts, we do the rest. */
5330 ogs.ref_count = group_get_ref_count(group);
5331 ogs.n_buckets = group->n_buckets;
5332
5333 error = (ofproto->ofproto_class->group_get_stats
5334 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
5335 : EOPNOTSUPP);
5336 if (error) {
5337 ogs.packet_count = UINT64_MAX;
5338 ogs.byte_count = UINT64_MAX;
5339 memset(ogs.bucket_stats, 0xff,
5340 ogs.n_buckets * sizeof *ogs.bucket_stats);
5341 }
5342
5343 ogs.group_id = group->group_id;
5344 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
5345
5346 ofputil_append_group_stats(replies, &ogs);
5347
5348 free(ogs.bucket_stats);
5349 }
5350
5351 static void
5352 handle_group_request(struct ofconn *ofconn,
5353 const struct ofp_header *request, uint32_t group_id,
5354 void (*cb)(struct ofgroup *, struct list *replies))
5355 {
5356 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5357 struct ofgroup *group;
5358 struct list replies;
5359
5360 ofpmp_init(&replies, request);
5361 if (group_id == OFPG_ALL) {
5362 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
5363 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
5364 cb(group, &replies);
5365 }
5366 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5367 } else {
5368 if (ofproto_group_lookup(ofproto, group_id, &group)) {
5369 cb(group, &replies);
5370 ofproto_group_unref(group);
5371 }
5372 }
5373 ofconn_send_replies(ofconn, &replies);
5374 }
5375
5376 static enum ofperr
5377 handle_group_stats_request(struct ofconn *ofconn,
5378 const struct ofp_header *request)
5379 {
5380 uint32_t group_id;
5381 enum ofperr error;
5382
5383 error = ofputil_decode_group_stats_request(request, &group_id);
5384 if (error) {
5385 return error;
5386 }
5387
5388 handle_group_request(ofconn, request, group_id, append_group_stats);
5389 return 0;
5390 }
5391
5392 static void
5393 append_group_desc(struct ofgroup *group, struct list *replies)
5394 {
5395 struct ofputil_group_desc gds;
5396
5397 gds.group_id = group->group_id;
5398 gds.type = group->type;
5399 ofputil_append_group_desc_reply(&gds, &group->buckets, replies);
5400 }
5401
5402 static enum ofperr
5403 handle_group_desc_stats_request(struct ofconn *ofconn,
5404 const struct ofp_header *request)
5405 {
5406 handle_group_request(ofconn, request,
5407 ofputil_decode_group_desc_request(request),
5408 append_group_desc);
5409 return 0;
5410 }
5411
5412 static enum ofperr
5413 handle_group_features_stats_request(struct ofconn *ofconn,
5414 const struct ofp_header *request)
5415 {
5416 struct ofproto *p = ofconn_get_ofproto(ofconn);
5417 struct ofpbuf *msg;
5418
5419 msg = ofputil_encode_group_features_reply(&p->ogf, request);
5420 if (msg) {
5421 ofconn_send_reply(ofconn, msg);
5422 }
5423
5424 return 0;
5425 }
5426
5427 static enum ofperr
5428 handle_queue_get_config_request(struct ofconn *ofconn,
5429 const struct ofp_header *oh)
5430 {
5431 struct ofproto *p = ofconn_get_ofproto(ofconn);
5432 struct netdev_queue_dump queue_dump;
5433 struct ofport *ofport;
5434 unsigned int queue_id;
5435 struct ofpbuf *reply;
5436 struct smap details;
5437 ofp_port_t request;
5438 enum ofperr error;
5439
5440 error = ofputil_decode_queue_get_config_request(oh, &request);
5441 if (error) {
5442 return error;
5443 }
5444
5445 ofport = ofproto_get_port(p, request);
5446 if (!ofport) {
5447 return OFPERR_OFPQOFC_BAD_PORT;
5448 }
5449
5450 reply = ofputil_encode_queue_get_config_reply(oh);
5451
5452 smap_init(&details);
5453 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
5454 struct ofputil_queue_config queue;
5455
5456 /* None of the existing queues have compatible properties, so we
5457 * hard-code omitting min_rate and max_rate. */
5458 queue.queue_id = queue_id;
5459 queue.min_rate = UINT16_MAX;
5460 queue.max_rate = UINT16_MAX;
5461 ofputil_append_queue_get_config_reply(reply, &queue);
5462 }
5463 smap_destroy(&details);
5464
5465 ofconn_send_reply(ofconn, reply);
5466
5467 return 0;
5468 }
5469
5470 static enum ofperr
5471 init_group(struct ofproto *ofproto, struct ofputil_group_mod *gm,
5472 struct ofgroup **ofgroup)
5473 {
5474 enum ofperr error;
5475 const long long int now = time_msec();
5476
5477 if (gm->group_id > OFPG_MAX) {
5478 return OFPERR_OFPGMFC_INVALID_GROUP;
5479 }
5480 if (gm->type > OFPGT11_FF) {
5481 return OFPERR_OFPGMFC_BAD_TYPE;
5482 }
5483
5484 *ofgroup = ofproto->ofproto_class->group_alloc();
5485 if (!*ofgroup) {
5486 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
5487 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
5488 }
5489
5490 (*ofgroup)->ofproto = ofproto;
5491 *CONST_CAST(uint32_t *, &((*ofgroup)->group_id)) = gm->group_id;
5492 *CONST_CAST(enum ofp11_group_type *, &(*ofgroup)->type) = gm->type;
5493 *CONST_CAST(long long int *, &((*ofgroup)->created)) = now;
5494 *CONST_CAST(long long int *, &((*ofgroup)->modified)) = now;
5495 ovs_refcount_init(&(*ofgroup)->ref_count);
5496
5497 list_move(&(*ofgroup)->buckets, &gm->buckets);
5498 *CONST_CAST(uint32_t *, &(*ofgroup)->n_buckets) =
5499 list_size(&(*ofgroup)->buckets);
5500
5501 /* Construct called BEFORE any locks are held. */
5502 error = ofproto->ofproto_class->group_construct(*ofgroup);
5503 if (error) {
5504 ofputil_bucket_list_destroy(&(*ofgroup)->buckets);
5505 ofproto->ofproto_class->group_dealloc(*ofgroup);
5506 }
5507 return error;
5508 }
5509
5510 /* Implements the OFPGC11_ADD operation specified by 'gm', adding a group to
5511 * 'ofproto''s group table. Returns 0 on success or an OpenFlow error code on
5512 * failure. */
5513 static enum ofperr
5514 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5515 {
5516 struct ofgroup *ofgroup;
5517 enum ofperr error;
5518
5519 /* Allocate new group and initialize it. */
5520 error = init_group(ofproto, gm, &ofgroup);
5521 if (error) {
5522 return error;
5523 }
5524
5525 /* We wrlock as late as possible to minimize the time we jam any other
5526 * threads: No visible state changes before acquiring the lock. */
5527 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5528
5529 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5530 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5531 goto unlock_out;
5532 }
5533
5534 if (ofproto_group_exists__(ofproto, gm->group_id)) {
5535 error = OFPERR_OFPGMFC_GROUP_EXISTS;
5536 goto unlock_out;
5537 }
5538
5539 if (!error) {
5540 /* Insert new group. */
5541 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
5542 hash_int(ofgroup->group_id, 0));
5543 ofproto->n_groups[ofgroup->type]++;
5544
5545 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5546 return error;
5547 }
5548
5549 unlock_out:
5550 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5551 ofproto->ofproto_class->group_destruct(ofgroup);
5552 ofputil_bucket_list_destroy(&ofgroup->buckets);
5553 ofproto->ofproto_class->group_dealloc(ofgroup);
5554
5555 return error;
5556 }
5557
5558 /* Implements OFPGC11_MODIFY. Returns 0 on success or an OpenFlow error code
5559 * on failure.
5560 *
5561 * Note that the group is re-created and then replaces the old group in
5562 * ofproto's ofgroup hash map. Thus, the group is never altered while users of
5563 * the xlate module hold a pointer to the group. */
5564 static enum ofperr
5565 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
5566 {
5567 struct ofgroup *ofgroup, *new_ofgroup, *retiring;
5568 enum ofperr error;
5569
5570 error = init_group(ofproto, gm, &new_ofgroup);
5571 if (error) {
5572 return error;
5573 }
5574
5575 retiring = new_ofgroup;
5576
5577 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5578 if (!ofproto_group_lookup__(ofproto, gm->group_id, &ofgroup)) {
5579 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
5580 goto out;
5581 }
5582
5583 /* Ofproto's group write lock is held now. */
5584 if (ofgroup->type != gm->type
5585 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
5586 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
5587 goto out;
5588 }
5589
5590 /* The group creation time does not change during modification. */
5591 *CONST_CAST(long long int *, &(new_ofgroup->created)) = ofgroup->created;
5592 *CONST_CAST(long long int *, &(new_ofgroup->modified)) = time_msec();
5593
5594 error = ofproto->ofproto_class->group_modify(new_ofgroup);
5595 if (error) {
5596 goto out;
5597 }
5598
5599 retiring = ofgroup;
5600 /* Replace ofgroup in ofproto's groups hash map with new_ofgroup. */
5601 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5602 hmap_insert(&ofproto->groups, &new_ofgroup->hmap_node,
5603 hash_int(new_ofgroup->group_id, 0));
5604 if (ofgroup->type != new_ofgroup->type) {
5605 ofproto->n_groups[ofgroup->type]--;
5606 ofproto->n_groups[new_ofgroup->type]++;
5607 }
5608
5609 out:
5610 ofproto_group_unref(retiring);
5611 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5612 return error;
5613 }
5614
5615 static void
5616 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
5617 OVS_RELEASES(ofproto->groups_rwlock)
5618 {
5619 struct match match;
5620 struct ofputil_flow_mod fm;
5621
5622 /* Delete all flow entries containing this group in a group action */
5623 match_init_catchall(&match);
5624 flow_mod_init(&fm, &match, 0, NULL, 0, OFPFC_DELETE);
5625 fm.delete_reason = OFPRR_GROUP_DELETE;
5626 fm.out_group = ofgroup->group_id;
5627 handle_flow_mod__(ofproto, &fm, NULL);
5628
5629 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
5630 /* No-one can find this group any more. */
5631 ofproto->n_groups[ofgroup->type]--;
5632 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5633 ofproto_group_unref(ofgroup);
5634 }
5635
5636 /* Implements OFPGC11_DELETE. */
5637 static void
5638 delete_group(struct ofproto *ofproto, uint32_t group_id)
5639 {
5640 struct ofgroup *ofgroup;
5641
5642 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5643 if (group_id == OFPG_ALL) {
5644 for (;;) {
5645 struct hmap_node *node = hmap_first(&ofproto->groups);
5646 if (!node) {
5647 break;
5648 }
5649 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
5650 delete_group__(ofproto, ofgroup);
5651 /* Lock for each node separately, so that we will not jam the
5652 * other threads for too long time. */
5653 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
5654 }
5655 } else {
5656 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
5657 hash_int(group_id, 0), &ofproto->groups) {
5658 if (ofgroup->group_id == group_id) {
5659 delete_group__(ofproto, ofgroup);
5660 return;
5661 }
5662 }
5663 }
5664 ovs_rwlock_unlock(&ofproto->groups_rwlock);
5665 }
5666
5667 static enum ofperr
5668 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5669 {
5670 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5671 struct ofputil_group_mod gm;
5672 enum ofperr error;
5673
5674 error = reject_slave_controller(ofconn);
5675 if (error) {
5676 return error;
5677 }
5678
5679 error = ofputil_decode_group_mod(oh, &gm);
5680 if (error) {
5681 return error;
5682 }
5683
5684 switch (gm.command) {
5685 case OFPGC11_ADD:
5686 return add_group(ofproto, &gm);
5687
5688 case OFPGC11_MODIFY:
5689 return modify_group(ofproto, &gm);
5690
5691 case OFPGC11_DELETE:
5692 delete_group(ofproto, gm.group_id);
5693 return 0;
5694
5695 default:
5696 if (gm.command > OFPGC11_DELETE) {
5697 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
5698 ofproto->name, gm.command);
5699 }
5700 return OFPERR_OFPGMFC_BAD_COMMAND;
5701 }
5702 }
5703
5704 enum ofproto_table_config
5705 ofproto_table_get_config(const struct ofproto *ofproto, uint8_t table_id)
5706 {
5707 unsigned int value;
5708 atomic_read(&ofproto->tables[table_id].config, &value);
5709 return (enum ofproto_table_config)value;
5710 }
5711
5712 static enum ofperr
5713 table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
5714 {
5715 /* Only accept currently supported configurations */
5716 if (tm->config & ~OFPTC11_TABLE_MISS_MASK) {
5717 return OFPERR_OFPTMFC_BAD_CONFIG;
5718 }
5719
5720 if (tm->table_id == OFPTT_ALL) {
5721 int i;
5722 for (i = 0; i < ofproto->n_tables; i++) {
5723 atomic_store(&ofproto->tables[i].config,
5724 (unsigned int)tm->config);
5725 }
5726 } else if (!check_table_id(ofproto, tm->table_id)) {
5727 return OFPERR_OFPTMFC_BAD_TABLE;
5728 } else {
5729 atomic_store(&ofproto->tables[tm->table_id].config,
5730 (unsigned int)tm->config);
5731 }
5732
5733 return 0;
5734 }
5735
5736 static enum ofperr
5737 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5738 {
5739 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5740 struct ofputil_table_mod tm;
5741 enum ofperr error;
5742
5743 error = reject_slave_controller(ofconn);
5744 if (error) {
5745 return error;
5746 }
5747
5748 error = ofputil_decode_table_mod(oh, &tm);
5749 if (error) {
5750 return error;
5751 }
5752
5753 return table_mod(ofproto, &tm);
5754 }
5755
5756 static enum ofperr
5757 handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh)
5758 {
5759 enum ofperr error;
5760 struct ofputil_bundle_ctrl_msg bctrl;
5761 struct ofpbuf *buf;
5762 struct ofputil_bundle_ctrl_msg reply;
5763
5764 error = ofputil_decode_bundle_ctrl(oh, &bctrl);
5765 if (error) {
5766 return error;
5767 }
5768 reply.flags = 0;
5769 reply.bundle_id = bctrl.bundle_id;
5770
5771 switch (bctrl.type) {
5772 case OFPBCT_OPEN_REQUEST:
5773 error = ofp_bundle_open(ofconn, bctrl.bundle_id, bctrl.flags);
5774 reply.type = OFPBCT_OPEN_REPLY;
5775 break;
5776 case OFPBCT_CLOSE_REQUEST:
5777 error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags);
5778 reply.type = OFPBCT_CLOSE_REPLY;;
5779 break;
5780 case OFPBCT_COMMIT_REQUEST:
5781 error = ofp_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags);
5782 reply.type = OFPBCT_COMMIT_REPLY;
5783 break;
5784 case OFPBCT_DISCARD_REQUEST:
5785 error = ofp_bundle_discard(ofconn, bctrl.bundle_id);
5786 reply.type = OFPBCT_DISCARD_REPLY;
5787 break;
5788
5789 case OFPBCT_OPEN_REPLY:
5790 case OFPBCT_CLOSE_REPLY:
5791 case OFPBCT_COMMIT_REPLY:
5792 case OFPBCT_DISCARD_REPLY:
5793 return OFPERR_OFPBFC_BAD_TYPE;
5794 break;
5795 }
5796
5797 if (!error) {
5798 buf = ofputil_encode_bundle_ctrl_reply(oh, &reply);
5799 ofconn_send_reply(ofconn, buf);
5800 }
5801 return error;
5802 }
5803
5804
5805 static enum ofperr
5806 handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh)
5807 {
5808 enum ofperr error;
5809 struct ofputil_bundle_add_msg badd;
5810
5811 error = ofputil_decode_bundle_add(oh, &badd);
5812 if (error) {
5813 return error;
5814 }
5815
5816 return ofp_bundle_add_message(ofconn, &badd);
5817 }
5818
5819 static enum ofperr
5820 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
5821 OVS_EXCLUDED(ofproto_mutex)
5822 {
5823 const struct ofp_header *oh = ofpbuf_data(msg);
5824 enum ofptype type;
5825 enum ofperr error;
5826
5827 error = ofptype_decode(&type, oh);
5828 if (error) {
5829 return error;
5830 }
5831 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
5832 && ofpmp_more(oh)) {
5833 /* We have no buffer implementation for multipart requests.
5834 * Report overflow for requests which consists of multiple
5835 * messages. */
5836 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
5837 }
5838
5839 switch (type) {
5840 /* OpenFlow requests. */
5841 case OFPTYPE_ECHO_REQUEST:
5842 return handle_echo_request(ofconn, oh);
5843
5844 case OFPTYPE_FEATURES_REQUEST:
5845 return handle_features_request(ofconn, oh);
5846
5847 case OFPTYPE_GET_CONFIG_REQUEST:
5848 return handle_get_config_request(ofconn, oh);
5849
5850 case OFPTYPE_SET_CONFIG:
5851 return handle_set_config(ofconn, oh);
5852
5853 case OFPTYPE_PACKET_OUT:
5854 return handle_packet_out(ofconn, oh);
5855
5856 case OFPTYPE_PORT_MOD:
5857 return handle_port_mod(ofconn, oh);
5858
5859 case OFPTYPE_FLOW_MOD:
5860 return handle_flow_mod(ofconn, oh);
5861
5862 case OFPTYPE_GROUP_MOD:
5863 return handle_group_mod(ofconn, oh);
5864
5865 case OFPTYPE_TABLE_MOD:
5866 return handle_table_mod(ofconn, oh);
5867
5868 case OFPTYPE_METER_MOD:
5869 return handle_meter_mod(ofconn, oh);
5870
5871 case OFPTYPE_BARRIER_REQUEST:
5872 return handle_barrier_request(ofconn, oh);
5873
5874 case OFPTYPE_ROLE_REQUEST:
5875 return handle_role_request(ofconn, oh);
5876
5877 /* OpenFlow replies. */
5878 case OFPTYPE_ECHO_REPLY:
5879 return 0;
5880
5881 /* Nicira extension requests. */
5882 case OFPTYPE_FLOW_MOD_TABLE_ID:
5883 return handle_nxt_flow_mod_table_id(ofconn, oh);
5884
5885 case OFPTYPE_SET_FLOW_FORMAT:
5886 return handle_nxt_set_flow_format(ofconn, oh);
5887
5888 case OFPTYPE_SET_PACKET_IN_FORMAT:
5889 return handle_nxt_set_packet_in_format(ofconn, oh);
5890
5891 case OFPTYPE_SET_CONTROLLER_ID:
5892 return handle_nxt_set_controller_id(ofconn, oh);
5893
5894 case OFPTYPE_FLOW_AGE:
5895 /* Nothing to do. */
5896 return 0;
5897
5898 case OFPTYPE_FLOW_MONITOR_CANCEL:
5899 return handle_flow_monitor_cancel(ofconn, oh);
5900
5901 case OFPTYPE_SET_ASYNC_CONFIG:
5902 return handle_nxt_set_async_config(ofconn, oh);
5903
5904 case OFPTYPE_GET_ASYNC_REQUEST:
5905 return handle_nxt_get_async_request(ofconn, oh);
5906
5907 /* Statistics requests. */
5908 case OFPTYPE_DESC_STATS_REQUEST:
5909 return handle_desc_stats_request(ofconn, oh);
5910
5911 case OFPTYPE_FLOW_STATS_REQUEST:
5912 return handle_flow_stats_request(ofconn, oh);
5913
5914 case OFPTYPE_AGGREGATE_STATS_REQUEST:
5915 return handle_aggregate_stats_request(ofconn, oh);
5916
5917 case OFPTYPE_TABLE_STATS_REQUEST:
5918 return handle_table_stats_request(ofconn, oh);
5919
5920 case OFPTYPE_PORT_STATS_REQUEST:
5921 return handle_port_stats_request(ofconn, oh);
5922
5923 case OFPTYPE_QUEUE_STATS_REQUEST:
5924 return handle_queue_stats_request(ofconn, oh);
5925
5926 case OFPTYPE_PORT_DESC_STATS_REQUEST:
5927 return handle_port_desc_stats_request(ofconn, oh);
5928
5929 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
5930 return handle_flow_monitor_request(ofconn, oh);
5931
5932 case OFPTYPE_METER_STATS_REQUEST:
5933 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
5934 return handle_meter_request(ofconn, oh, type);
5935
5936 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
5937 return handle_meter_features_request(ofconn, oh);
5938
5939 case OFPTYPE_GROUP_STATS_REQUEST:
5940 return handle_group_stats_request(ofconn, oh);
5941
5942 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
5943 return handle_group_desc_stats_request(ofconn, oh);
5944
5945 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
5946 return handle_group_features_stats_request(ofconn, oh);
5947
5948 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
5949 return handle_queue_get_config_request(ofconn, oh);
5950
5951 case OFPTYPE_BUNDLE_CONTROL:
5952 return handle_bundle_control(ofconn, oh);
5953
5954 case OFPTYPE_BUNDLE_ADD_MESSAGE:
5955 return handle_bundle_add(ofconn, oh);
5956
5957 case OFPTYPE_HELLO:
5958 case OFPTYPE_ERROR:
5959 case OFPTYPE_FEATURES_REPLY:
5960 case OFPTYPE_GET_CONFIG_REPLY:
5961 case OFPTYPE_PACKET_IN:
5962 case OFPTYPE_FLOW_REMOVED:
5963 case OFPTYPE_PORT_STATUS:
5964 case OFPTYPE_BARRIER_REPLY:
5965 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
5966 case OFPTYPE_DESC_STATS_REPLY:
5967 case OFPTYPE_FLOW_STATS_REPLY:
5968 case OFPTYPE_QUEUE_STATS_REPLY:
5969 case OFPTYPE_PORT_STATS_REPLY:
5970 case OFPTYPE_TABLE_STATS_REPLY:
5971 case OFPTYPE_AGGREGATE_STATS_REPLY:
5972 case OFPTYPE_PORT_DESC_STATS_REPLY:
5973 case OFPTYPE_ROLE_REPLY:
5974 case OFPTYPE_FLOW_MONITOR_PAUSED:
5975 case OFPTYPE_FLOW_MONITOR_RESUMED:
5976 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
5977 case OFPTYPE_GET_ASYNC_REPLY:
5978 case OFPTYPE_GROUP_STATS_REPLY:
5979 case OFPTYPE_GROUP_DESC_STATS_REPLY:
5980 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
5981 case OFPTYPE_METER_STATS_REPLY:
5982 case OFPTYPE_METER_CONFIG_STATS_REPLY:
5983 case OFPTYPE_METER_FEATURES_STATS_REPLY:
5984 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
5985 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
5986 case OFPTYPE_ROLE_STATUS:
5987 default:
5988 if (ofpmsg_is_stat_request(oh)) {
5989 return OFPERR_OFPBRC_BAD_STAT;
5990 } else {
5991 return OFPERR_OFPBRC_BAD_TYPE;
5992 }
5993 }
5994 }
5995
5996 static void
5997 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
5998 OVS_EXCLUDED(ofproto_mutex)
5999 {
6000 int error = handle_openflow__(ofconn, ofp_msg);
6001 if (error) {
6002 ofconn_send_error(ofconn, ofpbuf_data(ofp_msg), error);
6003 }
6004 COVERAGE_INC(ofproto_recv_openflow);
6005 }
6006 \f
6007 /* Asynchronous operations. */
6008
6009 static enum ofperr
6010 send_buffered_packet(struct ofconn *ofconn, uint32_t buffer_id,
6011 struct rule *rule)
6012 OVS_REQUIRES(ofproto_mutex)
6013 {
6014 enum ofperr error = 0;
6015 if (ofconn && buffer_id != UINT32_MAX) {
6016 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6017 struct ofpbuf *packet;
6018 ofp_port_t in_port;
6019
6020 error = ofconn_pktbuf_retrieve(ofconn, buffer_id, &packet, &in_port);
6021 if (packet) {
6022 struct rule_execute *re;
6023
6024 ofproto_rule_ref(rule);
6025
6026 re = xmalloc(sizeof *re);
6027 re->rule = rule;
6028 re->in_port = in_port;
6029 re->packet = packet;
6030
6031 if (!guarded_list_push_back(&ofproto->rule_executes,
6032 &re->list_node, 1024)) {
6033 ofproto_rule_unref(rule);
6034 ofpbuf_delete(re->packet);
6035 free(re);
6036 }
6037 }
6038 }
6039 return error;
6040 }
6041 \f
6042 static uint64_t
6043 pick_datapath_id(const struct ofproto *ofproto)
6044 {
6045 const struct ofport *port;
6046
6047 port = ofproto_get_port(ofproto, OFPP_LOCAL);
6048 if (port) {
6049 uint8_t ea[ETH_ADDR_LEN];
6050 int error;
6051
6052 error = netdev_get_etheraddr(port->netdev, ea);
6053 if (!error) {
6054 return eth_addr_to_uint64(ea);
6055 }
6056 VLOG_WARN("%s: could not get MAC address for %s (%s)",
6057 ofproto->name, netdev_get_name(port->netdev),
6058 ovs_strerror(error));
6059 }
6060 return ofproto->fallback_dpid;
6061 }
6062
6063 static uint64_t
6064 pick_fallback_dpid(void)
6065 {
6066 uint8_t ea[ETH_ADDR_LEN];
6067 eth_addr_nicira_random(ea);
6068 return eth_addr_to_uint64(ea);
6069 }
6070 \f
6071 /* Table overflow policy. */
6072
6073 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
6074 * to NULL if the table is not configured to evict rules or if the table
6075 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
6076 * or with no timeouts are not evictable.) */
6077 static bool
6078 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
6079 OVS_REQUIRES(ofproto_mutex)
6080 {
6081 struct eviction_group *evg;
6082
6083 *rulep = NULL;
6084 if (!table->eviction_fields) {
6085 return false;
6086 }
6087
6088 /* In the common case, the outer and inner loops here will each be entered
6089 * exactly once:
6090 *
6091 * - The inner loop normally "return"s in its first iteration. If the
6092 * eviction group has any evictable rules, then it always returns in
6093 * some iteration.
6094 *
6095 * - The outer loop only iterates more than once if the largest eviction
6096 * group has no evictable rules.
6097 *
6098 * - The outer loop can exit only if table's 'max_flows' is all filled up
6099 * by unevictable rules. */
6100 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
6101 struct rule *rule;
6102
6103 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
6104 *rulep = rule;
6105 return true;
6106 }
6107 }
6108
6109 return false;
6110 }
6111 \f
6112 /* Eviction groups. */
6113
6114 /* Returns the priority to use for an eviction_group that contains 'n_rules'
6115 * rules. The priority contains low-order random bits to ensure that eviction
6116 * groups with the same number of rules are prioritized randomly. */
6117 static uint32_t
6118 eviction_group_priority(size_t n_rules)
6119 {
6120 uint16_t size = MIN(UINT16_MAX, n_rules);
6121 return (size << 16) | random_uint16();
6122 }
6123
6124 /* Updates 'evg', an eviction_group within 'table', following a change that
6125 * adds or removes rules in 'evg'. */
6126 static void
6127 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
6128 OVS_REQUIRES(ofproto_mutex)
6129 {
6130 heap_change(&table->eviction_groups_by_size, &evg->size_node,
6131 eviction_group_priority(heap_count(&evg->rules)));
6132 }
6133
6134 /* Destroys 'evg', an eviction_group within 'table':
6135 *
6136 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
6137 * rules themselves, just removes them from the eviction group.)
6138 *
6139 * - Removes 'evg' from 'table'.
6140 *
6141 * - Frees 'evg'. */
6142 static void
6143 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
6144 OVS_REQUIRES(ofproto_mutex)
6145 {
6146 while (!heap_is_empty(&evg->rules)) {
6147 struct rule *rule;
6148
6149 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
6150 rule->eviction_group = NULL;
6151 }
6152 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
6153 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
6154 heap_destroy(&evg->rules);
6155 free(evg);
6156 }
6157
6158 /* Removes 'rule' from its eviction group, if any. */
6159 static void
6160 eviction_group_remove_rule(struct rule *rule)
6161 OVS_REQUIRES(ofproto_mutex)
6162 {
6163 if (rule->eviction_group) {
6164 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6165 struct eviction_group *evg = rule->eviction_group;
6166
6167 rule->eviction_group = NULL;
6168 heap_remove(&evg->rules, &rule->evg_node);
6169 if (heap_is_empty(&evg->rules)) {
6170 eviction_group_destroy(table, evg);
6171 } else {
6172 eviction_group_resized(table, evg);
6173 }
6174 }
6175 }
6176
6177 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
6178 * returns the hash value. */
6179 static uint32_t
6180 eviction_group_hash_rule(struct rule *rule)
6181 OVS_REQUIRES(ofproto_mutex)
6182 {
6183 struct oftable *table = &rule->ofproto->tables[rule->table_id];
6184 const struct mf_subfield *sf;
6185 struct flow flow;
6186 uint32_t hash;
6187
6188 hash = table->eviction_group_id_basis;
6189 miniflow_expand(&rule->cr.match.flow, &flow);
6190 for (sf = table->eviction_fields;
6191 sf < &table->eviction_fields[table->n_eviction_fields];
6192 sf++)
6193 {
6194 if (mf_are_prereqs_ok(sf->field, &flow)) {
6195 union mf_value value;
6196
6197 mf_get_value(sf->field, &flow, &value);
6198 if (sf->ofs) {
6199 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
6200 }
6201 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
6202 unsigned int start = sf->ofs + sf->n_bits;
6203 bitwise_zero(&value, sf->field->n_bytes, start,
6204 sf->field->n_bytes * 8 - start);
6205 }
6206 hash = hash_bytes(&value, sf->field->n_bytes, hash);
6207 } else {
6208 hash = hash_int(hash, 0);
6209 }
6210 }
6211
6212 return hash;
6213 }
6214
6215 /* Returns an eviction group within 'table' with the given 'id', creating one
6216 * if necessary. */
6217 static struct eviction_group *
6218 eviction_group_find(struct oftable *table, uint32_t id)
6219 OVS_REQUIRES(ofproto_mutex)
6220 {
6221 struct eviction_group *evg;
6222
6223 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
6224 return evg;
6225 }
6226
6227 evg = xmalloc(sizeof *evg);
6228 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
6229 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
6230 eviction_group_priority(0));
6231 heap_init(&evg->rules);
6232
6233 return evg;
6234 }
6235
6236 /* Returns an eviction priority for 'rule'. The return value should be
6237 * interpreted so that higher priorities make a rule more attractive candidates
6238 * for eviction.
6239 * Called only if have a timeout. */
6240 static uint32_t
6241 rule_eviction_priority(struct ofproto *ofproto, struct rule *rule)
6242 OVS_REQUIRES(ofproto_mutex)
6243 {
6244 long long int expiration = LLONG_MAX;
6245 long long int modified;
6246 uint32_t expiration_offset;
6247
6248 /* 'modified' needs protection even when we hold 'ofproto_mutex'. */
6249 ovs_mutex_lock(&rule->mutex);
6250 modified = rule->modified;
6251 ovs_mutex_unlock(&rule->mutex);
6252
6253 if (rule->hard_timeout) {
6254 expiration = modified + rule->hard_timeout * 1000;
6255 }
6256 if (rule->idle_timeout) {
6257 uint64_t packets, bytes;
6258 long long int used;
6259 long long int idle_expiration;
6260
6261 ofproto->ofproto_class->rule_get_stats(rule, &packets, &bytes, &used);
6262 idle_expiration = used + rule->idle_timeout * 1000;
6263 expiration = MIN(expiration, idle_expiration);
6264 }
6265
6266 if (expiration == LLONG_MAX) {
6267 return 0;
6268 }
6269
6270 /* Calculate the time of expiration as a number of (approximate) seconds
6271 * after program startup.
6272 *
6273 * This should work OK for program runs that last UINT32_MAX seconds or
6274 * less. Therefore, please restart OVS at least once every 136 years. */
6275 expiration_offset = (expiration >> 10) - (time_boot_msec() >> 10);
6276
6277 /* Invert the expiration offset because we're using a max-heap. */
6278 return UINT32_MAX - expiration_offset;
6279 }
6280
6281 /* Adds 'rule' to an appropriate eviction group for its oftable's
6282 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
6283 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
6284 * own).
6285 *
6286 * The caller must ensure that 'rule' is not already in an eviction group. */
6287 static void
6288 eviction_group_add_rule(struct rule *rule)
6289 OVS_REQUIRES(ofproto_mutex)
6290 {
6291 struct ofproto *ofproto = rule->ofproto;
6292 struct oftable *table = &ofproto->tables[rule->table_id];
6293 bool has_timeout;
6294
6295 /* Timeouts may be modified only when holding 'ofproto_mutex'. We have it
6296 * so no additional protection is needed. */
6297 has_timeout = rule->hard_timeout || rule->idle_timeout;
6298
6299 if (table->eviction_fields && has_timeout) {
6300 struct eviction_group *evg;
6301
6302 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
6303
6304 rule->eviction_group = evg;
6305 heap_insert(&evg->rules, &rule->evg_node,
6306 rule_eviction_priority(ofproto, rule));
6307 eviction_group_resized(table, evg);
6308 }
6309 }
6310 \f
6311 /* oftables. */
6312
6313 /* Initializes 'table'. */
6314 static void
6315 oftable_init(struct oftable *table)
6316 {
6317 memset(table, 0, sizeof *table);
6318 classifier_init(&table->cls, flow_segment_u32s);
6319 table->max_flows = UINT_MAX;
6320 atomic_init(&table->config, (unsigned int)OFPROTO_TABLE_MISS_DEFAULT);
6321
6322 fat_rwlock_wrlock(&table->cls.rwlock);
6323 classifier_set_prefix_fields(&table->cls, default_prefix_fields,
6324 ARRAY_SIZE(default_prefix_fields));
6325 fat_rwlock_unlock(&table->cls.rwlock);
6326
6327 atomic_init(&table->n_matched, 0);
6328 atomic_init(&table->n_missed, 0);
6329 }
6330
6331 /* Destroys 'table', including its classifier and eviction groups.
6332 *
6333 * The caller is responsible for freeing 'table' itself. */
6334 static void
6335 oftable_destroy(struct oftable *table)
6336 {
6337 fat_rwlock_rdlock(&table->cls.rwlock);
6338 ovs_assert(classifier_is_empty(&table->cls));
6339 fat_rwlock_unlock(&table->cls.rwlock);
6340 oftable_disable_eviction(table);
6341 classifier_destroy(&table->cls);
6342 free(table->name);
6343 }
6344
6345 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
6346 * string, then 'table' will use its default name.
6347 *
6348 * This only affects the name exposed for a table exposed through the OpenFlow
6349 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
6350 static void
6351 oftable_set_name(struct oftable *table, const char *name)
6352 {
6353 if (name && name[0]) {
6354 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
6355 if (!table->name || strncmp(name, table->name, len)) {
6356 free(table->name);
6357 table->name = xmemdup0(name, len);
6358 }
6359 } else {
6360 free(table->name);
6361 table->name = NULL;
6362 }
6363 }
6364
6365 /* oftables support a choice of two policies when adding a rule would cause the
6366 * number of flows in the table to exceed the configured maximum number: either
6367 * they can refuse to add the new flow or they can evict some existing flow.
6368 * This function configures the former policy on 'table'. */
6369 static void
6370 oftable_disable_eviction(struct oftable *table)
6371 OVS_REQUIRES(ofproto_mutex)
6372 {
6373 if (table->eviction_fields) {
6374 struct eviction_group *evg, *next;
6375
6376 HMAP_FOR_EACH_SAFE (evg, next, id_node,
6377 &table->eviction_groups_by_id) {
6378 eviction_group_destroy(table, evg);
6379 }
6380 hmap_destroy(&table->eviction_groups_by_id);
6381 heap_destroy(&table->eviction_groups_by_size);
6382
6383 free(table->eviction_fields);
6384 table->eviction_fields = NULL;
6385 table->n_eviction_fields = 0;
6386 }
6387 }
6388
6389 /* oftables support a choice of two policies when adding a rule would cause the
6390 * number of flows in the table to exceed the configured maximum number: either
6391 * they can refuse to add the new flow or they can evict some existing flow.
6392 * This function configures the latter policy on 'table', with fairness based
6393 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
6394 * 'n_fields' as 0 disables fairness.) */
6395 static void
6396 oftable_enable_eviction(struct oftable *table,
6397 const struct mf_subfield *fields, size_t n_fields)
6398 OVS_REQUIRES(ofproto_mutex)
6399 {
6400 struct cls_cursor cursor;
6401 struct rule *rule;
6402
6403 if (table->eviction_fields
6404 && n_fields == table->n_eviction_fields
6405 && (!n_fields
6406 || !memcmp(fields, table->eviction_fields,
6407 n_fields * sizeof *fields))) {
6408 /* No change. */
6409 return;
6410 }
6411
6412 oftable_disable_eviction(table);
6413
6414 table->n_eviction_fields = n_fields;
6415 table->eviction_fields = xmemdup(fields, n_fields * sizeof *fields);
6416
6417 table->eviction_group_id_basis = random_uint32();
6418 hmap_init(&table->eviction_groups_by_id);
6419 heap_init(&table->eviction_groups_by_size);
6420
6421 fat_rwlock_rdlock(&table->cls.rwlock);
6422 cls_cursor_init(&cursor, &table->cls, NULL);
6423 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6424 eviction_group_add_rule(rule);
6425 }
6426 fat_rwlock_unlock(&table->cls.rwlock);
6427 }
6428
6429 /* Removes 'rule' from the oftable that contains it. */
6430 static void
6431 oftable_remove_rule__(struct ofproto *ofproto, struct rule *rule)
6432 OVS_REQUIRES(ofproto_mutex)
6433 {
6434 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
6435
6436 fat_rwlock_wrlock(&cls->rwlock);
6437 classifier_remove(cls, CONST_CAST(struct cls_rule *, &rule->cr));
6438 fat_rwlock_unlock(&cls->rwlock);
6439
6440 cookies_remove(ofproto, rule);
6441
6442 eviction_group_remove_rule(rule);
6443 if (!list_is_empty(&rule->expirable)) {
6444 list_remove(&rule->expirable);
6445 }
6446 if (!list_is_empty(&rule->meter_list_node)) {
6447 list_remove(&rule->meter_list_node);
6448 list_init(&rule->meter_list_node);
6449 }
6450 }
6451
6452 static void
6453 oftable_remove_rule(struct rule *rule)
6454 OVS_REQUIRES(ofproto_mutex)
6455 {
6456 oftable_remove_rule__(rule->ofproto, rule);
6457 }
6458 \f
6459 /* unixctl commands. */
6460
6461 struct ofproto *
6462 ofproto_lookup(const char *name)
6463 {
6464 struct ofproto *ofproto;
6465
6466 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
6467 &all_ofprotos) {
6468 if (!strcmp(ofproto->name, name)) {
6469 return ofproto;
6470 }
6471 }
6472 return NULL;
6473 }
6474
6475 static void
6476 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
6477 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
6478 {
6479 struct ofproto *ofproto;
6480 struct ds results;
6481
6482 ds_init(&results);
6483 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
6484 ds_put_format(&results, "%s\n", ofproto->name);
6485 }
6486 unixctl_command_reply(conn, ds_cstr(&results));
6487 ds_destroy(&results);
6488 }
6489
6490 static void
6491 ofproto_unixctl_init(void)
6492 {
6493 static bool registered;
6494 if (registered) {
6495 return;
6496 }
6497 registered = true;
6498
6499 unixctl_command_register("ofproto/list", "", 0, 0,
6500 ofproto_unixctl_list, NULL);
6501 }
6502 \f
6503 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
6504 *
6505 * This is deprecated. It is only for compatibility with broken device drivers
6506 * in old versions of Linux that do not properly support VLANs when VLAN
6507 * devices are not used. When broken device drivers are no longer in
6508 * widespread use, we will delete these interfaces. */
6509
6510 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
6511 * (exactly) by an OpenFlow rule in 'ofproto'. */
6512 void
6513 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
6514 {
6515 struct match match;
6516 struct cls_rule target;
6517 const struct oftable *oftable;
6518
6519 match_init_catchall(&match);
6520 match_set_vlan_vid_masked(&match, htons(VLAN_CFI), htons(VLAN_CFI));
6521 cls_rule_init(&target, &match, 0);
6522
6523 free(ofproto->vlan_bitmap);
6524 ofproto->vlan_bitmap = bitmap_allocate(4096);
6525 ofproto->vlans_changed = false;
6526
6527 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6528 struct cls_cursor cursor;
6529 struct rule *rule;
6530
6531 fat_rwlock_rdlock(&oftable->cls.rwlock);
6532 cls_cursor_init(&cursor, &oftable->cls, &target);
6533 CLS_CURSOR_FOR_EACH (rule, cr, &cursor) {
6534 if (minimask_get_vid_mask(&rule->cr.match.mask) == VLAN_VID_MASK) {
6535 uint16_t vid = miniflow_get_vid(&rule->cr.match.flow);
6536
6537 bitmap_set1(vlan_bitmap, vid);
6538 bitmap_set1(ofproto->vlan_bitmap, vid);
6539 }
6540 }
6541 fat_rwlock_unlock(&oftable->cls.rwlock);
6542 }
6543
6544 cls_rule_destroy(&target);
6545 }
6546
6547 /* Returns true if new VLANs have come into use by the flow table since the
6548 * last call to ofproto_get_vlan_usage().
6549 *
6550 * We don't track when old VLANs stop being used. */
6551 bool
6552 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
6553 {
6554 return ofproto->vlans_changed;
6555 }
6556
6557 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
6558 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
6559 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
6560 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
6561 * then the VLAN device is un-enslaved. */
6562 int
6563 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
6564 ofp_port_t realdev_ofp_port, int vid)
6565 {
6566 struct ofport *ofport;
6567 int error;
6568
6569 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
6570
6571 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
6572 if (!ofport) {
6573 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
6574 ofproto->name, vlandev_ofp_port);
6575 return EINVAL;
6576 }
6577
6578 if (!ofproto->ofproto_class->set_realdev) {
6579 if (!vlandev_ofp_port) {
6580 return 0;
6581 }
6582 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
6583 return EOPNOTSUPP;
6584 }
6585
6586 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
6587 if (error) {
6588 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
6589 ofproto->name, vlandev_ofp_port,
6590 netdev_get_name(ofport->netdev), ovs_strerror(error));
6591 }
6592 return error;
6593 }