]> git.proxmox.com Git - ovs.git/blob - ofproto/ofproto.c
treewide: Change doubled ";;" to single ";" in C code.
[ovs.git] / ofproto / ofproto.c
1 /*
2 * Copyright (c) 2009-2015 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 "dp-packet.h"
48 #include "packets.h"
49 #include "pinsched.h"
50 #include "pktbuf.h"
51 #include "poll-loop.h"
52 #include "random.h"
53 #include "seq.h"
54 #include "shash.h"
55 #include "simap.h"
56 #include "smap.h"
57 #include "sset.h"
58 #include "timeval.h"
59 #include "tun-metadata.h"
60 #include "unaligned.h"
61 #include "unixctl.h"
62 #include "openvswitch/vlog.h"
63 #include "bundles.h"
64
65 VLOG_DEFINE_THIS_MODULE(ofproto);
66
67 COVERAGE_DEFINE(ofproto_flush);
68 COVERAGE_DEFINE(ofproto_packet_out);
69 COVERAGE_DEFINE(ofproto_queue_req);
70 COVERAGE_DEFINE(ofproto_recv_openflow);
71 COVERAGE_DEFINE(ofproto_reinit_ports);
72 COVERAGE_DEFINE(ofproto_update_port);
73
74 /* Default fields to use for prefix tries in each flow table, unless something
75 * else is configured. */
76 const enum mf_field_id default_prefix_fields[2] =
77 { MFF_IPV4_DST, MFF_IPV4_SRC };
78
79 /* oftable. */
80 static void oftable_init(struct oftable *);
81 static void oftable_destroy(struct oftable *);
82
83 static void oftable_set_name(struct oftable *, const char *name);
84
85 static enum ofperr evict_rules_from_table(struct oftable *)
86 OVS_REQUIRES(ofproto_mutex);
87 static void oftable_configure_eviction(struct oftable *,
88 unsigned int eviction,
89 const struct mf_subfield *fields,
90 size_t n_fields)
91 OVS_REQUIRES(ofproto_mutex);
92
93 /* This is the only combination of OpenFlow eviction flags that OVS supports: a
94 * combination of OF1.4+ importance, the remaining lifetime of the flow, and
95 * fairness based on user-specified fields. */
96 #define OFPROTO_EVICTION_FLAGS \
97 (OFPTMPEF14_OTHER | OFPTMPEF14_IMPORTANCE | OFPTMPEF14_LIFETIME)
98
99 /* A set of rules within a single OpenFlow table (oftable) that have the same
100 * values for the oftable's eviction_fields. A rule to be evicted, when one is
101 * needed, is taken from the eviction group that contains the greatest number
102 * of rules.
103 *
104 * An oftable owns any number of eviction groups, each of which contains any
105 * number of rules.
106 *
107 * Membership in an eviction group is imprecise, based on the hash of the
108 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
109 * That is, if two rules have different eviction_fields, but those
110 * eviction_fields hash to the same value, then they will belong to the same
111 * eviction_group anyway.
112 *
113 * (When eviction is not enabled on an oftable, we don't track any eviction
114 * groups, to save time and space.) */
115 struct eviction_group {
116 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
117 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
118 struct heap rules; /* Contains "struct rule"s. */
119 };
120
121 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
122 OVS_REQUIRES(ofproto_mutex);
123 static uint64_t rule_eviction_priority(struct ofproto *ofproto, struct rule *)
124 OVS_REQUIRES(ofproto_mutex);
125 static void eviction_group_add_rule(struct rule *)
126 OVS_REQUIRES(ofproto_mutex);
127 static void eviction_group_remove_rule(struct rule *)
128 OVS_REQUIRES(ofproto_mutex);
129
130 /* Criteria that flow_mod and other operations use for selecting rules on
131 * which to operate. */
132 struct rule_criteria {
133 /* An OpenFlow table or 255 for all tables. */
134 uint8_t table_id;
135
136 /* OpenFlow matching criteria. Interpreted different in "loose" way by
137 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
138 * defined in the OpenFlow spec. */
139 struct cls_rule cr;
140 cls_version_t version;
141
142 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
143 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
144 * The rule will not be selected if M is 1 and B != C. */
145 ovs_be64 cookie;
146 ovs_be64 cookie_mask;
147
148 /* Selection based on actions within a rule:
149 *
150 * If out_port != OFPP_ANY, selects only rules that output to out_port.
151 * If out_group != OFPG_ALL, select only rules that output to out_group. */
152 ofp_port_t out_port;
153 uint32_t out_group;
154
155 /* If true, collects only rules that are modifiable. */
156 bool include_hidden;
157 bool include_readonly;
158 };
159
160 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
161 const struct match *match, int priority,
162 cls_version_t version,
163 ovs_be64 cookie, ovs_be64 cookie_mask,
164 ofp_port_t out_port, uint32_t out_group);
165 static void rule_criteria_require_rw(struct rule_criteria *,
166 bool can_write_readonly);
167 static void rule_criteria_destroy(struct rule_criteria *);
168
169 static enum ofperr collect_rules_loose(struct ofproto *,
170 const struct rule_criteria *,
171 struct rule_collection *);
172
173 /* A packet that needs to be passed to rule_execute().
174 *
175 * (We can't do this immediately from ofopgroup_complete() because that holds
176 * ofproto_mutex, which rule_execute() needs released.) */
177 struct rule_execute {
178 struct ovs_list list_node; /* In struct ofproto's "rule_executes" list. */
179 struct rule *rule; /* Owns a reference to the rule. */
180 ofp_port_t in_port;
181 struct dp_packet *packet; /* Owns the packet. */
182 };
183
184 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
185 static void destroy_rule_executes(struct ofproto *);
186
187 struct learned_cookie {
188 union {
189 /* In struct ofproto's 'learned_cookies' hmap. */
190 struct hmap_node hmap_node OVS_GUARDED_BY(ofproto_mutex);
191
192 /* In 'dead_cookies' list when removed from hmap. */
193 struct ovs_list list_node;
194 } u;
195
196 /* Key. */
197 ovs_be64 cookie OVS_GUARDED_BY(ofproto_mutex);
198 uint8_t table_id OVS_GUARDED_BY(ofproto_mutex);
199
200 /* Number of references from "learn" actions.
201 *
202 * When this drops to 0, all of the flows in 'table_id' with the specified
203 * 'cookie' are deleted. */
204 int n OVS_GUARDED_BY(ofproto_mutex);
205 };
206
207 static const struct ofpact_learn *next_learn_with_delete(
208 const struct rule_actions *, const struct ofpact_learn *start);
209
210 static void learned_cookies_inc(struct ofproto *, const struct rule_actions *)
211 OVS_REQUIRES(ofproto_mutex);
212 static void learned_cookies_dec(struct ofproto *, const struct rule_actions *,
213 struct ovs_list *dead_cookies)
214 OVS_REQUIRES(ofproto_mutex);
215 static void learned_cookies_flush(struct ofproto *, struct ovs_list *dead_cookies)
216 OVS_REQUIRES(ofproto_mutex);
217
218 /* ofport. */
219 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
220 static void ofport_destroy(struct ofport *);
221
222 static void update_port(struct ofproto *, const char *devname);
223 static int init_ports(struct ofproto *);
224 static void reinit_ports(struct ofproto *);
225
226 static long long int ofport_get_usage(const struct ofproto *,
227 ofp_port_t ofp_port);
228 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
229 long long int last_used);
230 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
231
232 /* Ofport usage.
233 *
234 * Keeps track of the currently used and recently used ofport values and is
235 * used to prevent immediate recycling of ofport values. */
236 struct ofport_usage {
237 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
238 ofp_port_t ofp_port; /* OpenFlow port number. */
239 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
240 represents in-use ofports. */
241 };
242
243 /* rule. */
244 static void ofproto_rule_send_removed(struct rule *)
245 OVS_EXCLUDED(ofproto_mutex);
246 static bool rule_is_readonly(const struct rule *);
247 static void ofproto_rule_insert__(struct ofproto *, struct rule *)
248 OVS_REQUIRES(ofproto_mutex);
249 static void ofproto_rule_remove__(struct ofproto *, struct rule *)
250 OVS_REQUIRES(ofproto_mutex);
251
252 /* The source of a flow_mod request, in the code that processes flow_mods.
253 *
254 * A flow table modification request can be generated externally, via OpenFlow,
255 * or internally through a function call. This structure indicates the source
256 * of an OpenFlow-generated flow_mod. For an internal flow_mod, it isn't
257 * meaningful and thus supplied as NULL. */
258 struct flow_mod_requester {
259 struct ofconn *ofconn; /* Connection on which flow_mod arrived. */
260 const struct ofp_header *request;
261 };
262
263 /* OpenFlow. */
264 static enum ofperr replace_rule_create(struct ofproto *,
265 struct ofputil_flow_mod *,
266 struct cls_rule *cr, uint8_t table_id,
267 struct rule *old_rule,
268 struct rule **new_rule)
269 OVS_REQUIRES(ofproto_mutex);
270
271 static void replace_rule_start(struct ofproto *, cls_version_t version,
272 struct rule *old_rule, struct rule *new_rule,
273 struct cls_conjunction *, size_t n_conjs)
274 OVS_REQUIRES(ofproto_mutex);
275
276 static void replace_rule_revert(struct ofproto *, struct rule *old_rule,
277 struct rule *new_rule)
278 OVS_REQUIRES(ofproto_mutex);
279
280 static void replace_rule_finish(struct ofproto *, struct ofputil_flow_mod *,
281 const struct flow_mod_requester *,
282 struct rule *old_rule, struct rule *new_rule,
283 struct ovs_list *dead_cookies)
284 OVS_REQUIRES(ofproto_mutex);
285 static void delete_flows__(struct rule_collection *,
286 enum ofp_flow_removed_reason,
287 const struct flow_mod_requester *)
288 OVS_REQUIRES(ofproto_mutex);
289
290 static void send_buffered_packet(const struct flow_mod_requester *,
291 uint32_t buffer_id, struct rule *)
292 OVS_REQUIRES(ofproto_mutex);
293
294 static bool ofproto_group_exists__(const struct ofproto *ofproto,
295 uint32_t group_id)
296 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
297 static bool ofproto_group_exists(const struct ofproto *ofproto,
298 uint32_t group_id)
299 OVS_EXCLUDED(ofproto->groups_rwlock);
300 static enum ofperr add_group(struct ofproto *, struct ofputil_group_mod *);
301 static void handle_openflow(struct ofconn *, const struct ofpbuf *);
302 static enum ofperr ofproto_flow_mod_start(struct ofproto *,
303 struct ofproto_flow_mod *)
304 OVS_REQUIRES(ofproto_mutex);
305 static void ofproto_flow_mod_finish(struct ofproto *,
306 struct ofproto_flow_mod *,
307 const struct flow_mod_requester *)
308 OVS_REQUIRES(ofproto_mutex);
309 static enum ofperr handle_flow_mod__(struct ofproto *,
310 struct ofproto_flow_mod *,
311 const struct flow_mod_requester *)
312 OVS_EXCLUDED(ofproto_mutex);
313 static void calc_duration(long long int start, long long int now,
314 uint32_t *sec, uint32_t *nsec);
315
316 /* ofproto. */
317 static uint64_t pick_datapath_id(const struct ofproto *);
318 static uint64_t pick_fallback_dpid(void);
319 static void ofproto_destroy__(struct ofproto *);
320 static void update_mtu(struct ofproto *, struct ofport *);
321 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
322 static void meter_insert_rule(struct rule *);
323
324 /* unixctl. */
325 static void ofproto_unixctl_init(void);
326
327 /* All registered ofproto classes, in probe order. */
328 static const struct ofproto_class **ofproto_classes;
329 static size_t n_ofproto_classes;
330 static size_t allocated_ofproto_classes;
331
332 /* Global lock that protects all flow table operations. */
333 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
334
335 unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
336 unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT;
337
338 size_t n_handlers, n_revalidators;
339 size_t n_dpdk_rxqs;
340 char *pmd_cpu_mask;
341
342 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
343 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
344
345 /* Initial mappings of port to OpenFlow number mappings. */
346 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
347
348 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
349
350 /* The default value of true waits for flow restore. */
351 static bool flow_restore_wait = true;
352
353 /* Must be called to initialize the ofproto library.
354 *
355 * The caller may pass in 'iface_hints', which contains an shash of
356 * "iface_hint" elements indexed by the interface's name. The provider
357 * may use these hints to describe the startup configuration in order to
358 * reinitialize its state. The caller owns the provided data, so a
359 * provider will make copies of anything required. An ofproto provider
360 * will remove any existing state that is not described by the hint, and
361 * may choose to remove it all. */
362 void
363 ofproto_init(const struct shash *iface_hints)
364 {
365 struct shash_node *node;
366 size_t i;
367
368 ofproto_class_register(&ofproto_dpif_class);
369
370 /* Make a local copy, since we don't own 'iface_hints' elements. */
371 SHASH_FOR_EACH(node, iface_hints) {
372 const struct iface_hint *orig_hint = node->data;
373 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
374 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
375
376 new_hint->br_name = xstrdup(orig_hint->br_name);
377 new_hint->br_type = xstrdup(br_type);
378 new_hint->ofp_port = orig_hint->ofp_port;
379
380 shash_add(&init_ofp_ports, node->name, new_hint);
381 }
382
383 for (i = 0; i < n_ofproto_classes; i++) {
384 ofproto_classes[i]->init(&init_ofp_ports);
385 }
386
387 ofproto_unixctl_init();
388 }
389
390 /* 'type' should be a normalized datapath type, as returned by
391 * ofproto_normalize_type(). Returns the corresponding ofproto_class
392 * structure, or a null pointer if there is none registered for 'type'. */
393 static const struct ofproto_class *
394 ofproto_class_find__(const char *type)
395 {
396 size_t i;
397
398 for (i = 0; i < n_ofproto_classes; i++) {
399 const struct ofproto_class *class = ofproto_classes[i];
400 struct sset types;
401 bool found;
402
403 sset_init(&types);
404 class->enumerate_types(&types);
405 found = sset_contains(&types, type);
406 sset_destroy(&types);
407
408 if (found) {
409 return class;
410 }
411 }
412 VLOG_WARN("unknown datapath type %s", type);
413 return NULL;
414 }
415
416 /* Registers a new ofproto class. After successful registration, new ofprotos
417 * of that type can be created using ofproto_create(). */
418 int
419 ofproto_class_register(const struct ofproto_class *new_class)
420 {
421 size_t i;
422
423 for (i = 0; i < n_ofproto_classes; i++) {
424 if (ofproto_classes[i] == new_class) {
425 return EEXIST;
426 }
427 }
428
429 if (n_ofproto_classes >= allocated_ofproto_classes) {
430 ofproto_classes = x2nrealloc(ofproto_classes,
431 &allocated_ofproto_classes,
432 sizeof *ofproto_classes);
433 }
434 ofproto_classes[n_ofproto_classes++] = new_class;
435 return 0;
436 }
437
438 /* Unregisters a datapath provider. 'type' must have been previously
439 * registered and not currently be in use by any ofprotos. After
440 * unregistration new datapaths of that type cannot be opened using
441 * ofproto_create(). */
442 int
443 ofproto_class_unregister(const struct ofproto_class *class)
444 {
445 size_t i;
446
447 for (i = 0; i < n_ofproto_classes; i++) {
448 if (ofproto_classes[i] == class) {
449 for (i++; i < n_ofproto_classes; i++) {
450 ofproto_classes[i - 1] = ofproto_classes[i];
451 }
452 n_ofproto_classes--;
453 return 0;
454 }
455 }
456 VLOG_WARN("attempted to unregister an ofproto class that is not "
457 "registered");
458 return EAFNOSUPPORT;
459 }
460
461 /* Clears 'types' and enumerates all registered ofproto types into it. The
462 * caller must first initialize the sset. */
463 void
464 ofproto_enumerate_types(struct sset *types)
465 {
466 size_t i;
467
468 sset_clear(types);
469 for (i = 0; i < n_ofproto_classes; i++) {
470 ofproto_classes[i]->enumerate_types(types);
471 }
472 }
473
474 /* Returns the fully spelled out name for the given ofproto 'type'.
475 *
476 * Normalized type string can be compared with strcmp(). Unnormalized type
477 * string might be the same even if they have different spellings. */
478 const char *
479 ofproto_normalize_type(const char *type)
480 {
481 return type && type[0] ? type : "system";
482 }
483
484 /* Clears 'names' and enumerates the names of all known created ofprotos with
485 * the given 'type'. The caller must first initialize the sset. Returns 0 if
486 * successful, otherwise a positive errno value.
487 *
488 * Some kinds of datapaths might not be practically enumerable. This is not
489 * considered an error. */
490 int
491 ofproto_enumerate_names(const char *type, struct sset *names)
492 {
493 const struct ofproto_class *class = ofproto_class_find__(type);
494 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
495 }
496
497 static void
498 ofproto_bump_tables_version(struct ofproto *ofproto)
499 {
500 ++ofproto->tables_version;
501 ofproto->ofproto_class->set_tables_version(ofproto,
502 ofproto->tables_version);
503 }
504
505 int
506 ofproto_create(const char *datapath_name, const char *datapath_type,
507 struct ofproto **ofprotop)
508 {
509 const struct ofproto_class *class;
510 struct ofproto *ofproto;
511 int error;
512 int i;
513
514 *ofprotop = NULL;
515
516 datapath_type = ofproto_normalize_type(datapath_type);
517 class = ofproto_class_find__(datapath_type);
518 if (!class) {
519 VLOG_WARN("could not create datapath %s of unknown type %s",
520 datapath_name, datapath_type);
521 return EAFNOSUPPORT;
522 }
523
524 ofproto = class->alloc();
525 if (!ofproto) {
526 VLOG_ERR("failed to allocate datapath %s of type %s",
527 datapath_name, datapath_type);
528 return ENOMEM;
529 }
530
531 /* Initialize. */
532 ovs_mutex_lock(&ofproto_mutex);
533 memset(ofproto, 0, sizeof *ofproto);
534 ofproto->ofproto_class = class;
535 ofproto->name = xstrdup(datapath_name);
536 ofproto->type = xstrdup(datapath_type);
537 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
538 hash_string(ofproto->name, 0));
539 ofproto->datapath_id = 0;
540 ofproto->forward_bpdu = false;
541 ofproto->fallback_dpid = pick_fallback_dpid();
542 ofproto->mfr_desc = NULL;
543 ofproto->hw_desc = NULL;
544 ofproto->sw_desc = NULL;
545 ofproto->serial_desc = NULL;
546 ofproto->dp_desc = NULL;
547 ofproto->frag_handling = OFPC_FRAG_NORMAL;
548 hmap_init(&ofproto->ports);
549 hmap_init(&ofproto->ofport_usage);
550 shash_init(&ofproto->port_by_name);
551 simap_init(&ofproto->ofp_requests);
552 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
553 ofproto->eviction_group_timer = LLONG_MIN;
554 ofproto->tables = NULL;
555 ofproto->n_tables = 0;
556 ofproto->tables_version = CLS_MIN_VERSION;
557 hindex_init(&ofproto->cookies);
558 hmap_init(&ofproto->learned_cookies);
559 list_init(&ofproto->expirable);
560 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
561 guarded_list_init(&ofproto->rule_executes);
562 ofproto->vlan_bitmap = NULL;
563 ofproto->vlans_changed = false;
564 ofproto->min_mtu = INT_MAX;
565 ovs_rwlock_init(&ofproto->groups_rwlock);
566 hmap_init(&ofproto->groups);
567 ovs_mutex_unlock(&ofproto_mutex);
568 ofproto->ogf.types = 0xf;
569 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
570 OFPGFC_SELECT_WEIGHT;
571 for (i = 0; i < 4; i++) {
572 ofproto->ogf.max_groups[i] = OFPG_MAX;
573 ofproto->ogf.ofpacts[i] = (UINT64_C(1) << N_OFPACTS) - 1;
574 }
575 tun_metadata_init();
576
577 error = ofproto->ofproto_class->construct(ofproto);
578 if (error) {
579 VLOG_ERR("failed to open datapath %s: %s",
580 datapath_name, ovs_strerror(error));
581 connmgr_destroy(ofproto->connmgr);
582 ofproto_destroy__(ofproto);
583 return error;
584 }
585
586 /* Check that hidden tables, if any, are at the end. */
587 ovs_assert(ofproto->n_tables);
588 for (i = 0; i + 1 < ofproto->n_tables; i++) {
589 enum oftable_flags flags = ofproto->tables[i].flags;
590 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
591
592 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
593 }
594
595 ofproto->datapath_id = pick_datapath_id(ofproto);
596 init_ports(ofproto);
597
598 /* Initialize meters table. */
599 if (ofproto->ofproto_class->meter_get_features) {
600 ofproto->ofproto_class->meter_get_features(ofproto,
601 &ofproto->meter_features);
602 } else {
603 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
604 }
605 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
606 * sizeof(struct meter *));
607
608 /* Set the initial tables version. */
609 ofproto_bump_tables_version(ofproto);
610
611 *ofprotop = ofproto;
612 return 0;
613 }
614
615 /* Must be called (only) by an ofproto implementation in its constructor
616 * function. See the large comment on 'construct' in struct ofproto_class for
617 * details. */
618 void
619 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
620 {
621 struct oftable *table;
622
623 ovs_assert(!ofproto->n_tables);
624 ovs_assert(n_tables >= 1 && n_tables <= 255);
625
626 ofproto->n_tables = n_tables;
627 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
628 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
629 oftable_init(table);
630 }
631 }
632
633 /* To be optionally called (only) by an ofproto implementation in its
634 * constructor function. See the large comment on 'construct' in struct
635 * ofproto_class for details.
636 *
637 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
638 * will then ensure that actions passed into the ofproto implementation will
639 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
640 * function is not called, there will be no such restriction.
641 *
642 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
643 * the 'max_ports' restriction. */
644 void
645 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
646 {
647 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
648 ofproto->max_ports = max_ports;
649 }
650
651 uint64_t
652 ofproto_get_datapath_id(const struct ofproto *ofproto)
653 {
654 return ofproto->datapath_id;
655 }
656
657 void
658 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
659 {
660 uint64_t old_dpid = p->datapath_id;
661 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
662 if (p->datapath_id != old_dpid) {
663 /* Force all active connections to reconnect, since there is no way to
664 * notify a controller that the datapath ID has changed. */
665 ofproto_reconnect_controllers(p);
666 }
667 }
668
669 void
670 ofproto_set_controllers(struct ofproto *p,
671 const struct ofproto_controller *controllers,
672 size_t n_controllers, uint32_t allowed_versions)
673 {
674 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
675 allowed_versions);
676 }
677
678 void
679 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
680 {
681 connmgr_set_fail_mode(p->connmgr, fail_mode);
682 }
683
684 /* Drops the connections between 'ofproto' and all of its controllers, forcing
685 * them to reconnect. */
686 void
687 ofproto_reconnect_controllers(struct ofproto *ofproto)
688 {
689 connmgr_reconnect(ofproto->connmgr);
690 }
691
692 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
693 * in-band control should guarantee access, in the same way that in-band
694 * control guarantees access to OpenFlow controllers. */
695 void
696 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
697 const struct sockaddr_in *extras, size_t n)
698 {
699 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
700 }
701
702 /* Sets the OpenFlow queue used by flows set up by in-band control on
703 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
704 * flows will use the default queue. */
705 void
706 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
707 {
708 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
709 }
710
711 /* Sets the number of flows at which eviction from the kernel flow table
712 * will occur. */
713 void
714 ofproto_set_flow_limit(unsigned limit)
715 {
716 ofproto_flow_limit = limit;
717 }
718
719 /* Sets the maximum idle time for flows in the datapath before they are
720 * expired. */
721 void
722 ofproto_set_max_idle(unsigned max_idle)
723 {
724 ofproto_max_idle = max_idle;
725 }
726
727 /* If forward_bpdu is true, the NORMAL action will forward frames with
728 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
729 * the NORMAL action will drop these frames. */
730 void
731 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
732 {
733 bool old_val = ofproto->forward_bpdu;
734 ofproto->forward_bpdu = forward_bpdu;
735 if (old_val != ofproto->forward_bpdu) {
736 if (ofproto->ofproto_class->forward_bpdu_changed) {
737 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
738 }
739 }
740 }
741
742 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
743 * 'idle_time', in seconds, and the maximum number of MAC table entries to
744 * 'max_entries'. */
745 void
746 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
747 size_t max_entries)
748 {
749 if (ofproto->ofproto_class->set_mac_table_config) {
750 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
751 max_entries);
752 }
753 }
754
755 /* Multicast snooping configuration. */
756
757 /* Configures multicast snooping on 'ofproto' using the settings
758 * defined in 's'. If 's' is NULL, disables multicast snooping.
759 *
760 * Returns 0 if successful, otherwise a positive errno value. */
761 int
762 ofproto_set_mcast_snooping(struct ofproto *ofproto,
763 const struct ofproto_mcast_snooping_settings *s)
764 {
765 return (ofproto->ofproto_class->set_mcast_snooping
766 ? ofproto->ofproto_class->set_mcast_snooping(ofproto, s)
767 : EOPNOTSUPP);
768 }
769
770 /* Configures multicast snooping flood settings on 'ofp_port' of 'ofproto'.
771 *
772 * Returns 0 if successful, otherwise a positive errno value.*/
773 int
774 ofproto_port_set_mcast_snooping(struct ofproto *ofproto, void *aux,
775 const struct ofproto_mcast_snooping_port_settings *s)
776 {
777 return (ofproto->ofproto_class->set_mcast_snooping_port
778 ? ofproto->ofproto_class->set_mcast_snooping_port(ofproto, aux, s)
779 : EOPNOTSUPP);
780 }
781
782 void
783 ofproto_set_n_dpdk_rxqs(int n_rxqs)
784 {
785 n_dpdk_rxqs = MAX(n_rxqs, 0);
786 }
787
788 void
789 ofproto_set_cpu_mask(const char *cmask)
790 {
791 free(pmd_cpu_mask);
792
793 pmd_cpu_mask = cmask ? xstrdup(cmask) : NULL;
794 }
795
796 void
797 ofproto_set_threads(int n_handlers_, int n_revalidators_)
798 {
799 int threads = MAX(count_cpu_cores(), 2);
800
801 n_revalidators = MAX(n_revalidators_, 0);
802 n_handlers = MAX(n_handlers_, 0);
803
804 if (!n_revalidators) {
805 n_revalidators = n_handlers
806 ? MAX(threads - (int) n_handlers, 1)
807 : threads / 4 + 1;
808 }
809
810 if (!n_handlers) {
811 n_handlers = MAX(threads - (int) n_revalidators, 1);
812 }
813 }
814
815 void
816 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
817 {
818 free(p->dp_desc);
819 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
820 }
821
822 int
823 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
824 {
825 return connmgr_set_snoops(ofproto->connmgr, snoops);
826 }
827
828 int
829 ofproto_set_netflow(struct ofproto *ofproto,
830 const struct netflow_options *nf_options)
831 {
832 if (nf_options && sset_is_empty(&nf_options->collectors)) {
833 nf_options = NULL;
834 }
835
836 if (ofproto->ofproto_class->set_netflow) {
837 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
838 } else {
839 return nf_options ? EOPNOTSUPP : 0;
840 }
841 }
842
843 int
844 ofproto_set_sflow(struct ofproto *ofproto,
845 const struct ofproto_sflow_options *oso)
846 {
847 if (oso && sset_is_empty(&oso->targets)) {
848 oso = NULL;
849 }
850
851 if (ofproto->ofproto_class->set_sflow) {
852 return ofproto->ofproto_class->set_sflow(ofproto, oso);
853 } else {
854 return oso ? EOPNOTSUPP : 0;
855 }
856 }
857
858 int
859 ofproto_set_ipfix(struct ofproto *ofproto,
860 const struct ofproto_ipfix_bridge_exporter_options *bo,
861 const struct ofproto_ipfix_flow_exporter_options *fo,
862 size_t n_fo)
863 {
864 if (ofproto->ofproto_class->set_ipfix) {
865 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
866 } else {
867 return (bo || fo) ? EOPNOTSUPP : 0;
868 }
869 }
870
871 void
872 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
873 {
874 flow_restore_wait = flow_restore_wait_db;
875 }
876
877 bool
878 ofproto_get_flow_restore_wait(void)
879 {
880 return flow_restore_wait;
881 }
882
883 \f
884 /* Spanning Tree Protocol (STP) configuration. */
885
886 /* Configures STP on 'ofproto' using the settings defined in 's'. If
887 * 's' is NULL, disables STP.
888 *
889 * Returns 0 if successful, otherwise a positive errno value. */
890 int
891 ofproto_set_stp(struct ofproto *ofproto,
892 const struct ofproto_stp_settings *s)
893 {
894 return (ofproto->ofproto_class->set_stp
895 ? ofproto->ofproto_class->set_stp(ofproto, s)
896 : EOPNOTSUPP);
897 }
898
899 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
900 * 'enabled' member of 's' is false, then the other members are not
901 * meaningful.
902 *
903 * Returns 0 if successful, otherwise a positive errno value. */
904 int
905 ofproto_get_stp_status(struct ofproto *ofproto,
906 struct ofproto_stp_status *s)
907 {
908 return (ofproto->ofproto_class->get_stp_status
909 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
910 : EOPNOTSUPP);
911 }
912
913 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
914 * in 's'. The caller is responsible for assigning STP port numbers
915 * (using the 'port_num' member in the range of 1 through 255, inclusive)
916 * and ensuring there are no duplicates. If the 's' is NULL, then STP
917 * is disabled on the port.
918 *
919 * Returns 0 if successful, otherwise a positive errno value.*/
920 int
921 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
922 const struct ofproto_port_stp_settings *s)
923 {
924 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
925 if (!ofport) {
926 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
927 ofproto->name, ofp_port);
928 return ENODEV;
929 }
930
931 return (ofproto->ofproto_class->set_stp_port
932 ? ofproto->ofproto_class->set_stp_port(ofport, s)
933 : EOPNOTSUPP);
934 }
935
936 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
937 * 's'. If the 'enabled' member in 's' is false, then the other members
938 * are not meaningful.
939 *
940 * Returns 0 if successful, otherwise a positive errno value.*/
941 int
942 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
943 struct ofproto_port_stp_status *s)
944 {
945 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
946 if (!ofport) {
947 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
948 "port %"PRIu16, ofproto->name, ofp_port);
949 return ENODEV;
950 }
951
952 return (ofproto->ofproto_class->get_stp_port_status
953 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
954 : EOPNOTSUPP);
955 }
956
957 /* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
958 * 's'. If the 'enabled' member in 's' is false, then the other members
959 * are not meaningful.
960 *
961 * Returns 0 if successful, otherwise a positive errno value.*/
962 int
963 ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
964 struct ofproto_port_stp_stats *s)
965 {
966 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
967 if (!ofport) {
968 VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
969 "port %"PRIu16, ofproto->name, ofp_port);
970 return ENODEV;
971 }
972
973 return (ofproto->ofproto_class->get_stp_port_stats
974 ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
975 : EOPNOTSUPP);
976 }
977
978 /* Rapid Spanning Tree Protocol (RSTP) configuration. */
979
980 /* Configures RSTP on 'ofproto' using the settings defined in 's'. If
981 * 's' is NULL, disables RSTP.
982 *
983 * Returns 0 if successful, otherwise a positive errno value. */
984 int
985 ofproto_set_rstp(struct ofproto *ofproto,
986 const struct ofproto_rstp_settings *s)
987 {
988 if (!ofproto->ofproto_class->set_rstp) {
989 return EOPNOTSUPP;
990 }
991 ofproto->ofproto_class->set_rstp(ofproto, s);
992 return 0;
993 }
994
995 /* Retrieves RSTP status of 'ofproto' and stores it in 's'. If the
996 * 'enabled' member of 's' is false, then the other members are not
997 * meaningful.
998 *
999 * Returns 0 if successful, otherwise a positive errno value. */
1000 int
1001 ofproto_get_rstp_status(struct ofproto *ofproto,
1002 struct ofproto_rstp_status *s)
1003 {
1004 if (!ofproto->ofproto_class->get_rstp_status) {
1005 return EOPNOTSUPP;
1006 }
1007 ofproto->ofproto_class->get_rstp_status(ofproto, s);
1008 return 0;
1009 }
1010
1011 /* Configures RSTP on 'ofp_port' of 'ofproto' using the settings defined
1012 * in 's'. The caller is responsible for assigning RSTP port numbers
1013 * (using the 'port_num' member in the range of 1 through 255, inclusive)
1014 * and ensuring there are no duplicates. If the 's' is NULL, then RSTP
1015 * is disabled on the port.
1016 *
1017 * Returns 0 if successful, otherwise a positive errno value.*/
1018 int
1019 ofproto_port_set_rstp(struct ofproto *ofproto, ofp_port_t ofp_port,
1020 const struct ofproto_port_rstp_settings *s)
1021 {
1022 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1023 if (!ofport) {
1024 VLOG_WARN("%s: cannot configure RSTP on nonexistent port %"PRIu16,
1025 ofproto->name, ofp_port);
1026 return ENODEV;
1027 }
1028
1029 if (!ofproto->ofproto_class->set_rstp_port) {
1030 return EOPNOTSUPP;
1031 }
1032 ofproto->ofproto_class->set_rstp_port(ofport, s);
1033 return 0;
1034 }
1035
1036 /* Retrieves RSTP port status of 'ofp_port' on 'ofproto' and stores it in
1037 * 's'. If the 'enabled' member in 's' is false, then the other members
1038 * are not meaningful.
1039 *
1040 * Returns 0 if successful, otherwise a positive errno value.*/
1041 int
1042 ofproto_port_get_rstp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1043 struct ofproto_port_rstp_status *s)
1044 {
1045 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1046 if (!ofport) {
1047 VLOG_WARN_RL(&rl, "%s: cannot get RSTP status on nonexistent "
1048 "port %"PRIu16, ofproto->name, ofp_port);
1049 return ENODEV;
1050 }
1051
1052 if (!ofproto->ofproto_class->get_rstp_port_status) {
1053 return EOPNOTSUPP;
1054 }
1055 ofproto->ofproto_class->get_rstp_port_status(ofport, s);
1056 return 0;
1057 }
1058 \f
1059 /* Queue DSCP configuration. */
1060
1061 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1062 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
1063 * to implement QoS. Instead, it is used to implement features which require
1064 * knowledge of what queues exist on a port, and some basic information about
1065 * them.
1066 *
1067 * Returns 0 if successful, otherwise a positive errno value. */
1068 int
1069 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
1070 const struct ofproto_port_queue *queues,
1071 size_t n_queues)
1072 {
1073 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1074
1075 if (!ofport) {
1076 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
1077 ofproto->name, ofp_port);
1078 return ENODEV;
1079 }
1080
1081 return (ofproto->ofproto_class->set_queues
1082 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
1083 : EOPNOTSUPP);
1084 }
1085 \f
1086 /* LLDP configuration. */
1087 void
1088 ofproto_port_set_lldp(struct ofproto *ofproto,
1089 ofp_port_t ofp_port,
1090 const struct smap *cfg)
1091 {
1092 struct ofport *ofport;
1093 int error;
1094
1095 ofport = ofproto_get_port(ofproto, ofp_port);
1096 if (!ofport) {
1097 VLOG_WARN("%s: cannot configure LLDP on nonexistent port %"PRIu16,
1098 ofproto->name, ofp_port);
1099 return;
1100 }
1101 error = (ofproto->ofproto_class->set_lldp
1102 ? ofproto->ofproto_class->set_lldp(ofport, cfg)
1103 : EOPNOTSUPP);
1104 if (error) {
1105 VLOG_WARN("%s: lldp configuration on port %"PRIu16" (%s) failed (%s)",
1106 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1107 ovs_strerror(error));
1108 }
1109 }
1110
1111 int
1112 ofproto_set_aa(struct ofproto *ofproto, void *aux OVS_UNUSED,
1113 const struct aa_settings *s)
1114 {
1115 if (!ofproto->ofproto_class->set_aa) {
1116 return EOPNOTSUPP;
1117 }
1118 ofproto->ofproto_class->set_aa(ofproto, s);
1119 return 0;
1120 }
1121
1122 int
1123 ofproto_aa_mapping_register(struct ofproto *ofproto, void *aux,
1124 const struct aa_mapping_settings *s)
1125 {
1126 if (!ofproto->ofproto_class->aa_mapping_set) {
1127 return EOPNOTSUPP;
1128 }
1129 ofproto->ofproto_class->aa_mapping_set(ofproto, aux, s);
1130 return 0;
1131 }
1132
1133 int
1134 ofproto_aa_mapping_unregister(struct ofproto *ofproto, void *aux)
1135 {
1136 if (!ofproto->ofproto_class->aa_mapping_unset) {
1137 return EOPNOTSUPP;
1138 }
1139 ofproto->ofproto_class->aa_mapping_unset(ofproto, aux);
1140 return 0;
1141 }
1142
1143 int
1144 ofproto_aa_vlan_get_queued(struct ofproto *ofproto,
1145 struct ovs_list *list)
1146 {
1147 if (!ofproto->ofproto_class->aa_vlan_get_queued) {
1148 return EOPNOTSUPP;
1149 }
1150 ofproto->ofproto_class->aa_vlan_get_queued(ofproto, list);
1151 return 0;
1152 }
1153
1154 unsigned int
1155 ofproto_aa_vlan_get_queue_size(struct ofproto *ofproto)
1156 {
1157 if (!ofproto->ofproto_class->aa_vlan_get_queue_size) {
1158 return EOPNOTSUPP;
1159 }
1160 return ofproto->ofproto_class->aa_vlan_get_queue_size(ofproto);
1161 }
1162
1163 /* Connectivity Fault Management configuration. */
1164
1165 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
1166 void
1167 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
1168 {
1169 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1170 if (ofport && ofproto->ofproto_class->set_cfm) {
1171 ofproto->ofproto_class->set_cfm(ofport, NULL);
1172 }
1173 }
1174
1175 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
1176 * basic configuration from the configuration members in 'cfm', and the remote
1177 * maintenance point ID from remote_mpid. Ignores the statistics members of
1178 * 'cfm'.
1179 *
1180 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
1181 void
1182 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
1183 const struct cfm_settings *s)
1184 {
1185 struct ofport *ofport;
1186 int error;
1187
1188 ofport = ofproto_get_port(ofproto, ofp_port);
1189 if (!ofport) {
1190 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
1191 ofproto->name, ofp_port);
1192 return;
1193 }
1194
1195 /* XXX: For configuration simplicity, we only support one remote_mpid
1196 * outside of the CFM module. It's not clear if this is the correct long
1197 * term solution or not. */
1198 error = (ofproto->ofproto_class->set_cfm
1199 ? ofproto->ofproto_class->set_cfm(ofport, s)
1200 : EOPNOTSUPP);
1201 if (error) {
1202 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
1203 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1204 ovs_strerror(error));
1205 }
1206 }
1207
1208 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
1209 * 'ofproto' does not have a port 'ofp_port'. */
1210 void
1211 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
1212 const struct smap *cfg)
1213 {
1214 struct ofport *ofport;
1215 int error;
1216
1217 ofport = ofproto_get_port(ofproto, ofp_port);
1218 if (!ofport) {
1219 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
1220 ofproto->name, ofp_port);
1221 return;
1222 }
1223
1224 error = (ofproto->ofproto_class->set_bfd
1225 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
1226 : EOPNOTSUPP);
1227 if (error) {
1228 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
1229 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1230 ovs_strerror(error));
1231 }
1232 }
1233
1234 /* Checks the status change of BFD on 'ofport'.
1235 *
1236 * Returns true if 'ofproto_class' does not support 'bfd_status_changed'. */
1237 bool
1238 ofproto_port_bfd_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
1239 {
1240 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1241 return (ofport && ofproto->ofproto_class->bfd_status_changed
1242 ? ofproto->ofproto_class->bfd_status_changed(ofport)
1243 : true);
1244 }
1245
1246 /* Populates 'status' with the status of BFD on 'ofport'. Returns 0 on
1247 * success. Returns a positive errno otherwise. Has no effect if 'ofp_port'
1248 * is not an OpenFlow port in 'ofproto'.
1249 *
1250 * The caller must provide and own '*status'. */
1251 int
1252 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1253 struct smap *status)
1254 {
1255 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1256 return (ofport && ofproto->ofproto_class->get_bfd_status
1257 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1258 : EOPNOTSUPP);
1259 }
1260
1261 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1262 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1263 * 0 if LACP partner information is not current (generally indicating a
1264 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1265 int
1266 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1267 {
1268 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1269 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1270 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1271 : -1);
1272 }
1273
1274 int
1275 ofproto_port_get_lacp_stats(const struct ofport *port, struct lacp_slave_stats *stats)
1276 {
1277 struct ofproto *ofproto = port->ofproto;
1278 int error;
1279
1280 if (ofproto->ofproto_class->port_get_lacp_stats) {
1281 error = ofproto->ofproto_class->port_get_lacp_stats(port, stats);
1282 } else {
1283 error = EOPNOTSUPP;
1284 }
1285
1286 return error;
1287 }
1288 \f
1289 /* Bundles. */
1290
1291 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1292 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1293 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1294 * configuration plus, if there is more than one slave, a bonding
1295 * configuration.
1296 *
1297 * If 'aux' is already registered then this function updates its configuration
1298 * to 's'. Otherwise, this function registers a new bundle.
1299 *
1300 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1301 * port. */
1302 int
1303 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1304 const struct ofproto_bundle_settings *s)
1305 {
1306 return (ofproto->ofproto_class->bundle_set
1307 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1308 : EOPNOTSUPP);
1309 }
1310
1311 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1312 * If no such bundle has been registered, this has no effect. */
1313 int
1314 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1315 {
1316 return ofproto_bundle_register(ofproto, aux, NULL);
1317 }
1318
1319 \f
1320 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1321 * If 'aux' is already registered then this function updates its configuration
1322 * to 's'. Otherwise, this function registers a new mirror. */
1323 int
1324 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1325 const struct ofproto_mirror_settings *s)
1326 {
1327 return (ofproto->ofproto_class->mirror_set
1328 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1329 : EOPNOTSUPP);
1330 }
1331
1332 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1333 * If no mirror has been registered, this has no effect. */
1334 int
1335 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1336 {
1337 return ofproto_mirror_register(ofproto, aux, NULL);
1338 }
1339
1340 /* Retrieves statistics from mirror associated with client data pointer
1341 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1342 * 'bytes', respectively. If a particular counters is not supported,
1343 * the appropriate argument is set to UINT64_MAX.
1344 */
1345 int
1346 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1347 uint64_t *packets, uint64_t *bytes)
1348 {
1349 if (!ofproto->ofproto_class->mirror_get_stats) {
1350 *packets = *bytes = UINT64_MAX;
1351 return EOPNOTSUPP;
1352 }
1353
1354 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1355 packets, bytes);
1356 }
1357
1358 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1359 * which all packets are flooded, instead of using MAC learning. If
1360 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1361 *
1362 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1363 * port. */
1364 int
1365 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1366 {
1367 return (ofproto->ofproto_class->set_flood_vlans
1368 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1369 : EOPNOTSUPP);
1370 }
1371
1372 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1373 * output for a mirror. */
1374 bool
1375 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1376 {
1377 return (ofproto->ofproto_class->is_mirror_output_bundle
1378 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1379 : false);
1380 }
1381 \f
1382 /* Configuration of OpenFlow tables. */
1383
1384 /* Returns the number of OpenFlow tables in 'ofproto'. */
1385 int
1386 ofproto_get_n_tables(const struct ofproto *ofproto)
1387 {
1388 return ofproto->n_tables;
1389 }
1390
1391 /* Returns the number of Controller visible OpenFlow tables
1392 * in 'ofproto'. This number will exclude Hidden tables.
1393 * This funtion's return value should be less or equal to that of
1394 * ofproto_get_n_tables() . */
1395 uint8_t
1396 ofproto_get_n_visible_tables(const struct ofproto *ofproto)
1397 {
1398 uint8_t n = ofproto->n_tables;
1399
1400 /* Count only non-hidden tables in the number of tables. (Hidden tables,
1401 * if present, are always at the end.) */
1402 while(n && (ofproto->tables[n - 1].flags & OFTABLE_HIDDEN)) {
1403 n--;
1404 }
1405
1406 return n;
1407 }
1408
1409 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1410 * settings from 's'. 'table_id' must be in the range 0 through the number of
1411 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1412 *
1413 * For read-only tables, only the name may be configured. */
1414 void
1415 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1416 const struct ofproto_table_settings *s)
1417 {
1418 struct oftable *table;
1419
1420 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1421 table = &ofproto->tables[table_id];
1422
1423 oftable_set_name(table, s->name);
1424
1425 if (table->flags & OFTABLE_READONLY) {
1426 return;
1427 }
1428
1429 if (classifier_set_prefix_fields(&table->cls,
1430 s->prefix_fields, s->n_prefix_fields)) {
1431 /* XXX: Trigger revalidation. */
1432 }
1433
1434 ovs_mutex_lock(&ofproto_mutex);
1435 unsigned int new_eviction = (s->enable_eviction
1436 ? table->eviction | EVICTION_CLIENT
1437 : table->eviction & ~EVICTION_CLIENT);
1438 oftable_configure_eviction(table, new_eviction, s->groups, s->n_groups);
1439 table->max_flows = s->max_flows;
1440 evict_rules_from_table(table);
1441 ovs_mutex_unlock(&ofproto_mutex);
1442 }
1443 \f
1444 bool
1445 ofproto_has_snoops(const struct ofproto *ofproto)
1446 {
1447 return connmgr_has_snoops(ofproto->connmgr);
1448 }
1449
1450 void
1451 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1452 {
1453 connmgr_get_snoops(ofproto->connmgr, snoops);
1454 }
1455
1456 /* Deletes 'rule' from 'ofproto'.
1457 *
1458 * Within an ofproto implementation, this function allows an ofproto
1459 * implementation to destroy any rules that remain when its ->destruct()
1460 * function is called. This function is not suitable for use elsewhere in an
1461 * ofproto implementation.
1462 *
1463 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1464 * Cycle" in ofproto-provider.h. */
1465 void
1466 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1467 OVS_EXCLUDED(ofproto_mutex)
1468 {
1469 /* This skips the ofmonitor and flow-removed notifications because the
1470 * switch is being deleted and any OpenFlow channels have been or soon will
1471 * be killed. */
1472 ovs_mutex_lock(&ofproto_mutex);
1473
1474 if (!rule->removed) {
1475 /* Make sure there is no postponed removal of the rule. */
1476 ovs_assert(cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
1477
1478 if (!classifier_remove(&rule->ofproto->tables[rule->table_id].cls,
1479 &rule->cr)) {
1480 OVS_NOT_REACHED();
1481 }
1482 ofproto_rule_remove__(rule->ofproto, rule);
1483 ofproto->ofproto_class->rule_delete(rule);
1484 ofproto_rule_unref(rule);
1485 }
1486 ovs_mutex_unlock(&ofproto_mutex);
1487 }
1488
1489 static void
1490 ofproto_flush__(struct ofproto *ofproto)
1491 OVS_EXCLUDED(ofproto_mutex)
1492 {
1493 struct oftable *table;
1494
1495 /* This will flush all datapath flows. */
1496 if (ofproto->ofproto_class->flush) {
1497 ofproto->ofproto_class->flush(ofproto);
1498 }
1499
1500 /* XXX: There is a small race window here, where new datapath flows can be
1501 * created by upcall handlers based on the existing flow table. We can not
1502 * call ofproto class flush while holding 'ofproto_mutex' to prevent this,
1503 * as then we could deadlock on syncing with the handler threads waiting on
1504 * the same mutex. */
1505
1506 ovs_mutex_lock(&ofproto_mutex);
1507 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1508 struct rule_collection rules;
1509 struct rule *rule;
1510
1511 if (table->flags & OFTABLE_HIDDEN) {
1512 continue;
1513 }
1514
1515 rule_collection_init(&rules);
1516
1517 CLS_FOR_EACH (rule, cr, &table->cls) {
1518 rule_collection_add(&rules, rule);
1519 }
1520 delete_flows__(&rules, OFPRR_DELETE, NULL);
1521 }
1522 /* XXX: Concurrent handler threads may insert new learned flows based on
1523 * learn actions of the now deleted flows right after we release
1524 * 'ofproto_mutex'. */
1525 ovs_mutex_unlock(&ofproto_mutex);
1526 }
1527
1528 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1529
1530 static void
1531 ofproto_destroy__(struct ofproto *ofproto)
1532 OVS_EXCLUDED(ofproto_mutex)
1533 {
1534 struct oftable *table;
1535
1536 destroy_rule_executes(ofproto);
1537
1538 guarded_list_destroy(&ofproto->rule_executes);
1539 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1540 hmap_destroy(&ofproto->groups);
1541
1542 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1543 free(ofproto->name);
1544 free(ofproto->type);
1545 free(ofproto->mfr_desc);
1546 free(ofproto->hw_desc);
1547 free(ofproto->sw_desc);
1548 free(ofproto->serial_desc);
1549 free(ofproto->dp_desc);
1550 hmap_destroy(&ofproto->ports);
1551 hmap_destroy(&ofproto->ofport_usage);
1552 shash_destroy(&ofproto->port_by_name);
1553 simap_destroy(&ofproto->ofp_requests);
1554
1555 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1556 oftable_destroy(table);
1557 }
1558 free(ofproto->tables);
1559
1560 ovs_assert(hindex_is_empty(&ofproto->cookies));
1561 hindex_destroy(&ofproto->cookies);
1562
1563 ovs_assert(hmap_is_empty(&ofproto->learned_cookies));
1564 hmap_destroy(&ofproto->learned_cookies);
1565
1566 free(ofproto->vlan_bitmap);
1567
1568 ofproto->ofproto_class->dealloc(ofproto);
1569 }
1570
1571 /* Destroying rules is doubly deferred, must have 'ofproto' around for them.
1572 * - 1st we defer the removal of the rules from the classifier
1573 * - 2nd we defer the actual destruction of the rules. */
1574 static void
1575 ofproto_destroy_defer__(struct ofproto *ofproto)
1576 OVS_EXCLUDED(ofproto_mutex)
1577 {
1578 ovsrcu_postpone(ofproto_destroy__, ofproto);
1579 }
1580
1581 void
1582 ofproto_destroy(struct ofproto *p)
1583 OVS_EXCLUDED(ofproto_mutex)
1584 {
1585 struct ofport *ofport, *next_ofport;
1586 struct ofport_usage *usage, *next_usage;
1587
1588 if (!p) {
1589 return;
1590 }
1591
1592 if (p->meters) {
1593 meter_delete(p, 1, p->meter_features.max_meters);
1594 p->meter_features.max_meters = 0;
1595 free(p->meters);
1596 p->meters = NULL;
1597 }
1598
1599 ofproto_flush__(p);
1600 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1601 ofport_destroy(ofport);
1602 }
1603
1604 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1605 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1606 free(usage);
1607 }
1608
1609 p->ofproto_class->destruct(p);
1610
1611 /* We should not postpone this because it involves deleting a listening
1612 * socket which we may want to reopen soon. 'connmgr' should not be used
1613 * by other threads */
1614 connmgr_destroy(p->connmgr);
1615
1616 /* Destroying rules is deferred, must have 'ofproto' around for them. */
1617 ovsrcu_postpone(ofproto_destroy_defer__, p);
1618 }
1619
1620 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1621 * kernel datapath, for example, this destroys the datapath in the kernel, and
1622 * with the netdev-based datapath, it tears down the data structures that
1623 * represent the datapath.
1624 *
1625 * The datapath should not be currently open as an ofproto. */
1626 int
1627 ofproto_delete(const char *name, const char *type)
1628 {
1629 const struct ofproto_class *class = ofproto_class_find__(type);
1630 return (!class ? EAFNOSUPPORT
1631 : !class->del ? EACCES
1632 : class->del(type, name));
1633 }
1634
1635 static void
1636 process_port_change(struct ofproto *ofproto, int error, char *devname)
1637 {
1638 if (error == ENOBUFS) {
1639 reinit_ports(ofproto);
1640 } else if (!error) {
1641 update_port(ofproto, devname);
1642 free(devname);
1643 }
1644 }
1645
1646 int
1647 ofproto_type_run(const char *datapath_type)
1648 {
1649 const struct ofproto_class *class;
1650 int error;
1651
1652 datapath_type = ofproto_normalize_type(datapath_type);
1653 class = ofproto_class_find__(datapath_type);
1654
1655 error = class->type_run ? class->type_run(datapath_type) : 0;
1656 if (error && error != EAGAIN) {
1657 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1658 datapath_type, ovs_strerror(error));
1659 }
1660 return error;
1661 }
1662
1663 void
1664 ofproto_type_wait(const char *datapath_type)
1665 {
1666 const struct ofproto_class *class;
1667
1668 datapath_type = ofproto_normalize_type(datapath_type);
1669 class = ofproto_class_find__(datapath_type);
1670
1671 if (class->type_wait) {
1672 class->type_wait(datapath_type);
1673 }
1674 }
1675
1676 int
1677 ofproto_run(struct ofproto *p)
1678 {
1679 int error;
1680 uint64_t new_seq;
1681
1682 error = p->ofproto_class->run(p);
1683 if (error && error != EAGAIN) {
1684 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1685 }
1686
1687 run_rule_executes(p);
1688
1689 /* Restore the eviction group heap invariant occasionally. */
1690 if (p->eviction_group_timer < time_msec()) {
1691 size_t i;
1692
1693 p->eviction_group_timer = time_msec() + 1000;
1694
1695 for (i = 0; i < p->n_tables; i++) {
1696 struct oftable *table = &p->tables[i];
1697 struct eviction_group *evg;
1698 struct rule *rule;
1699
1700 if (!table->eviction) {
1701 continue;
1702 }
1703
1704 if (table->n_flows > 100000) {
1705 static struct vlog_rate_limit count_rl =
1706 VLOG_RATE_LIMIT_INIT(1, 1);
1707 VLOG_WARN_RL(&count_rl, "Table %"PRIuSIZE" has an excessive"
1708 " number of rules: %d", i, table->n_flows);
1709 }
1710
1711 ovs_mutex_lock(&ofproto_mutex);
1712 CLS_FOR_EACH (rule, cr, &table->cls) {
1713 if (rule->idle_timeout || rule->hard_timeout) {
1714 if (!rule->eviction_group) {
1715 eviction_group_add_rule(rule);
1716 } else {
1717 heap_raw_change(&rule->evg_node,
1718 rule_eviction_priority(p, rule));
1719 }
1720 }
1721 }
1722
1723 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1724 heap_rebuild(&evg->rules);
1725 }
1726 ovs_mutex_unlock(&ofproto_mutex);
1727 }
1728 }
1729
1730 if (p->ofproto_class->port_poll) {
1731 char *devname;
1732
1733 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1734 process_port_change(p, error, devname);
1735 }
1736 }
1737
1738 new_seq = seq_read(connectivity_seq_get());
1739 if (new_seq != p->change_seq) {
1740 struct sset devnames;
1741 const char *devname;
1742 struct ofport *ofport;
1743
1744 /* Update OpenFlow port status for any port whose netdev has changed.
1745 *
1746 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1747 * destroyed, so it's not safe to update ports directly from the
1748 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1749 * need this two-phase approach. */
1750 sset_init(&devnames);
1751 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1752 uint64_t port_change_seq;
1753
1754 port_change_seq = netdev_get_change_seq(ofport->netdev);
1755 if (ofport->change_seq != port_change_seq) {
1756 ofport->change_seq = port_change_seq;
1757 sset_add(&devnames, netdev_get_name(ofport->netdev));
1758 }
1759 }
1760 SSET_FOR_EACH (devname, &devnames) {
1761 update_port(p, devname);
1762 }
1763 sset_destroy(&devnames);
1764
1765 p->change_seq = new_seq;
1766 }
1767
1768 connmgr_run(p->connmgr, handle_openflow);
1769
1770 return error;
1771 }
1772
1773 void
1774 ofproto_wait(struct ofproto *p)
1775 {
1776 p->ofproto_class->wait(p);
1777 if (p->ofproto_class->port_poll_wait) {
1778 p->ofproto_class->port_poll_wait(p);
1779 }
1780 seq_wait(connectivity_seq_get(), p->change_seq);
1781 connmgr_wait(p->connmgr);
1782 }
1783
1784 bool
1785 ofproto_is_alive(const struct ofproto *p)
1786 {
1787 return connmgr_has_controllers(p->connmgr);
1788 }
1789
1790 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1791 * memory_report(). */
1792 void
1793 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1794 {
1795 const struct oftable *table;
1796 unsigned int n_rules;
1797
1798 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1799
1800 n_rules = 0;
1801 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1802 n_rules += table->n_flows;
1803 }
1804 simap_increase(usage, "rules", n_rules);
1805
1806 if (ofproto->ofproto_class->get_memory_usage) {
1807 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1808 }
1809
1810 connmgr_get_memory_usage(ofproto->connmgr, usage);
1811 }
1812
1813 void
1814 ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
1815 {
1816 const struct ofproto_class *class;
1817
1818 datapath_type = ofproto_normalize_type(datapath_type);
1819 class = ofproto_class_find__(datapath_type);
1820
1821 if (class && class->type_get_memory_usage) {
1822 class->type_get_memory_usage(datapath_type, usage);
1823 }
1824 }
1825
1826 void
1827 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1828 struct shash *info)
1829 {
1830 connmgr_get_controller_info(ofproto->connmgr, info);
1831 }
1832
1833 void
1834 ofproto_free_ofproto_controller_info(struct shash *info)
1835 {
1836 connmgr_free_controller_info(info);
1837 }
1838
1839 /* Makes a deep copy of 'old' into 'port'. */
1840 void
1841 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1842 {
1843 port->name = xstrdup(old->name);
1844 port->type = xstrdup(old->type);
1845 port->ofp_port = old->ofp_port;
1846 }
1847
1848 /* Frees memory allocated to members of 'ofproto_port'.
1849 *
1850 * Do not call this function on an ofproto_port obtained from
1851 * ofproto_port_dump_next(): that function retains ownership of the data in the
1852 * ofproto_port. */
1853 void
1854 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1855 {
1856 free(ofproto_port->name);
1857 free(ofproto_port->type);
1858 }
1859
1860 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1861 *
1862 * This function provides no status indication. An error status for the entire
1863 * dump operation is provided when it is completed by calling
1864 * ofproto_port_dump_done().
1865 */
1866 void
1867 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1868 const struct ofproto *ofproto)
1869 {
1870 dump->ofproto = ofproto;
1871 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1872 &dump->state);
1873 }
1874
1875 /* Attempts to retrieve another port from 'dump', which must have been created
1876 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1877 * 'port' and returns true. On failure, returns false.
1878 *
1879 * Failure might indicate an actual error or merely that the last port has been
1880 * dumped. An error status for the entire dump operation is provided when it
1881 * is completed by calling ofproto_port_dump_done().
1882 *
1883 * The ofproto owns the data stored in 'port'. It will remain valid until at
1884 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1885 * ofproto_port_dump_done(). */
1886 bool
1887 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1888 struct ofproto_port *port)
1889 {
1890 const struct ofproto *ofproto = dump->ofproto;
1891
1892 if (dump->error) {
1893 return false;
1894 }
1895
1896 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1897 port);
1898 if (dump->error) {
1899 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1900 return false;
1901 }
1902 return true;
1903 }
1904
1905 /* Completes port table dump operation 'dump', which must have been created
1906 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1907 * error-free, otherwise a positive errno value describing the problem. */
1908 int
1909 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1910 {
1911 const struct ofproto *ofproto = dump->ofproto;
1912 if (!dump->error) {
1913 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1914 dump->state);
1915 }
1916 return dump->error == EOF ? 0 : dump->error;
1917 }
1918
1919 /* Returns the type to pass to netdev_open() when a datapath of type
1920 * 'datapath_type' has a port of type 'port_type', for a few special
1921 * cases when a netdev type differs from a port type. For example, when
1922 * using the userspace datapath, a port of type "internal" needs to be
1923 * opened as "tap".
1924 *
1925 * Returns either 'type' itself or a string literal, which must not be
1926 * freed. */
1927 const char *
1928 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1929 {
1930 const struct ofproto_class *class;
1931
1932 datapath_type = ofproto_normalize_type(datapath_type);
1933 class = ofproto_class_find__(datapath_type);
1934 if (!class) {
1935 return port_type;
1936 }
1937
1938 return (class->port_open_type
1939 ? class->port_open_type(datapath_type, port_type)
1940 : port_type);
1941 }
1942
1943 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1944 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1945 * the port's OpenFlow port number.
1946 *
1947 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1948 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1949 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1950 * 'ofp_portp' is non-null). */
1951 int
1952 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1953 ofp_port_t *ofp_portp)
1954 {
1955 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1956 int error;
1957
1958 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1959 if (!error) {
1960 const char *netdev_name = netdev_get_name(netdev);
1961
1962 simap_put(&ofproto->ofp_requests, netdev_name,
1963 ofp_to_u16(ofp_port));
1964 update_port(ofproto, netdev_name);
1965 }
1966 if (ofp_portp) {
1967 *ofp_portp = OFPP_NONE;
1968 if (!error) {
1969 struct ofproto_port ofproto_port;
1970
1971 error = ofproto_port_query_by_name(ofproto,
1972 netdev_get_name(netdev),
1973 &ofproto_port);
1974 if (!error) {
1975 *ofp_portp = ofproto_port.ofp_port;
1976 ofproto_port_destroy(&ofproto_port);
1977 }
1978 }
1979 }
1980 return error;
1981 }
1982
1983 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1984 * initializes '*port' appropriately; on failure, returns a positive errno
1985 * value.
1986 *
1987 * The caller owns the data in 'ofproto_port' and must free it with
1988 * ofproto_port_destroy() when it is no longer needed. */
1989 int
1990 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1991 struct ofproto_port *port)
1992 {
1993 int error;
1994
1995 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1996 if (error) {
1997 memset(port, 0, sizeof *port);
1998 }
1999 return error;
2000 }
2001
2002 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
2003 * Returns 0 if successful, otherwise a positive errno. */
2004 int
2005 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
2006 {
2007 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2008 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
2009 struct simap_node *ofp_request_node;
2010 int error;
2011
2012 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
2013 if (ofp_request_node) {
2014 simap_delete(&ofproto->ofp_requests, ofp_request_node);
2015 }
2016
2017 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
2018 if (!error && ofport) {
2019 /* 'name' is the netdev's name and update_port() is going to close the
2020 * netdev. Just in case update_port() refers to 'name' after it
2021 * destroys 'ofport', make a copy of it around the update_port()
2022 * call. */
2023 char *devname = xstrdup(name);
2024 update_port(ofproto, devname);
2025 free(devname);
2026 }
2027 return error;
2028 }
2029
2030 static void
2031 flow_mod_init(struct ofputil_flow_mod *fm,
2032 const struct match *match, int priority,
2033 const struct ofpact *ofpacts, size_t ofpacts_len,
2034 enum ofp_flow_mod_command command)
2035 {
2036 memset(fm, 0, sizeof *fm);
2037 fm->match = *match;
2038 fm->priority = priority;
2039 fm->cookie = 0;
2040 fm->new_cookie = 0;
2041 fm->modify_cookie = false;
2042 fm->table_id = 0;
2043 fm->command = command;
2044 fm->idle_timeout = 0;
2045 fm->hard_timeout = 0;
2046 fm->importance = 0;
2047 fm->buffer_id = UINT32_MAX;
2048 fm->out_port = OFPP_ANY;
2049 fm->out_group = OFPG_ANY;
2050 fm->flags = 0;
2051 fm->ofpacts = CONST_CAST(struct ofpact *, ofpacts);
2052 fm->ofpacts_len = ofpacts_len;
2053 fm->delete_reason = OFPRR_DELETE;
2054 }
2055
2056 static int
2057 simple_flow_mod(struct ofproto *ofproto,
2058 const struct match *match, int priority,
2059 const struct ofpact *ofpacts, size_t ofpacts_len,
2060 enum ofp_flow_mod_command command)
2061 {
2062 struct ofproto_flow_mod ofm;
2063
2064 flow_mod_init(&ofm.fm, match, priority, ofpacts, ofpacts_len, command);
2065
2066 return handle_flow_mod__(ofproto, &ofm, NULL);
2067 }
2068
2069 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
2070 * performs the 'n_actions' actions in 'actions'. The new flow will not
2071 * timeout.
2072 *
2073 * If cls_rule->priority is in the range of priorities supported by OpenFlow
2074 * (0...65535, inclusive) then the flow will be visible to OpenFlow
2075 * controllers; otherwise, it will be hidden.
2076 *
2077 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
2078 *
2079 * This is a helper function for in-band control and fail-open. */
2080 void
2081 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
2082 int priority,
2083 const struct ofpact *ofpacts, size_t ofpacts_len)
2084 OVS_EXCLUDED(ofproto_mutex)
2085 {
2086 const struct rule *rule;
2087 bool must_add;
2088
2089 /* First do a cheap check whether the rule we're looking for already exists
2090 * with the actions that we want. If it does, then we're done. */
2091 rule = rule_from_cls_rule(classifier_find_match_exactly(
2092 &ofproto->tables[0].cls, match, priority,
2093 CLS_MAX_VERSION));
2094 if (rule) {
2095 const struct rule_actions *actions = rule_get_actions(rule);
2096 must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len,
2097 ofpacts, ofpacts_len);
2098 } else {
2099 must_add = true;
2100 }
2101
2102 /* If there's no such rule or the rule doesn't have the actions we want,
2103 * fall back to a executing a full flow mod. We can't optimize this at
2104 * all because we didn't take enough locks above to ensure that the flow
2105 * table didn't already change beneath us. */
2106 if (must_add) {
2107 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
2108 OFPFC_MODIFY_STRICT);
2109 }
2110 }
2111
2112 /* Executes the flow modification specified in 'fm'. Returns 0 on success, or
2113 * an OFPERR_* OpenFlow error code on failure.
2114 *
2115 * This is a helper function for in-band control and fail-open and the "learn"
2116 * action. */
2117 enum ofperr
2118 ofproto_flow_mod(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
2119 OVS_EXCLUDED(ofproto_mutex)
2120 {
2121 struct ofputil_flow_mod *fm = &ofm->fm;
2122
2123 /* Optimize for the most common case of a repeated learn action.
2124 * If an identical flow already exists we only need to update its
2125 * 'modified' time. */
2126 if (fm->command == OFPFC_MODIFY_STRICT && fm->table_id != OFPTT_ALL
2127 && !(fm->flags & OFPUTIL_FF_RESET_COUNTS)) {
2128 struct oftable *table = &ofproto->tables[fm->table_id];
2129 struct rule *rule;
2130 bool done = false;
2131
2132 rule = rule_from_cls_rule(classifier_find_match_exactly(
2133 &table->cls, &fm->match, fm->priority,
2134 CLS_MAX_VERSION));
2135 if (rule) {
2136 /* Reading many of the rule fields and writing on 'modified'
2137 * requires the rule->mutex. Also, rule->actions may change
2138 * if rule->mutex is not held. */
2139 const struct rule_actions *actions;
2140
2141 ovs_mutex_lock(&rule->mutex);
2142 actions = rule_get_actions(rule);
2143 if (rule->idle_timeout == fm->idle_timeout
2144 && rule->hard_timeout == fm->hard_timeout
2145 && rule->importance == fm->importance
2146 && rule->flags == (fm->flags & OFPUTIL_FF_STATE)
2147 && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie))
2148 && ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2149 actions->ofpacts, actions->ofpacts_len)) {
2150 /* Rule already exists and need not change, only update the
2151 modified timestamp. */
2152 rule->modified = time_msec();
2153 done = true;
2154 }
2155 ovs_mutex_unlock(&rule->mutex);
2156 }
2157
2158 if (done) {
2159 return 0;
2160 }
2161 }
2162
2163 return handle_flow_mod__(ofproto, ofm, NULL);
2164 }
2165
2166 /* Searches for a rule with matching criteria exactly equal to 'target' in
2167 * ofproto's table 0 and, if it finds one, deletes it.
2168 *
2169 * This is a helper function for in-band control and fail-open. */
2170 void
2171 ofproto_delete_flow(struct ofproto *ofproto,
2172 const struct match *target, int priority)
2173 OVS_EXCLUDED(ofproto_mutex)
2174 {
2175 struct classifier *cls = &ofproto->tables[0].cls;
2176 struct rule *rule;
2177
2178 /* First do a cheap check whether the rule we're looking for has already
2179 * been deleted. If so, then we're done. */
2180 rule = rule_from_cls_rule(classifier_find_match_exactly(
2181 cls, target, priority, CLS_MAX_VERSION));
2182 if (!rule) {
2183 return;
2184 }
2185
2186 /* Execute a flow mod. We can't optimize this at all because we didn't
2187 * take enough locks above to ensure that the flow table didn't already
2188 * change beneath us. */
2189 simple_flow_mod(ofproto, target, priority, NULL, 0, OFPFC_DELETE_STRICT);
2190 }
2191
2192 /* Delete all of the flows from all of ofproto's flow tables, then reintroduce
2193 * the flows required by in-band control and fail-open. */
2194 void
2195 ofproto_flush_flows(struct ofproto *ofproto)
2196 {
2197 COVERAGE_INC(ofproto_flush);
2198 ofproto_flush__(ofproto);
2199 connmgr_flushed(ofproto->connmgr);
2200 }
2201 \f
2202 static void
2203 reinit_ports(struct ofproto *p)
2204 {
2205 struct ofproto_port_dump dump;
2206 struct sset devnames;
2207 struct ofport *ofport;
2208 struct ofproto_port ofproto_port;
2209 const char *devname;
2210
2211 COVERAGE_INC(ofproto_reinit_ports);
2212
2213 sset_init(&devnames);
2214 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2215 sset_add(&devnames, netdev_get_name(ofport->netdev));
2216 }
2217 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2218 sset_add(&devnames, ofproto_port.name);
2219 }
2220
2221 SSET_FOR_EACH (devname, &devnames) {
2222 update_port(p, devname);
2223 }
2224 sset_destroy(&devnames);
2225 }
2226
2227 static ofp_port_t
2228 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
2229 {
2230 uint16_t port_idx;
2231
2232 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
2233 port_idx = port_idx ? port_idx : UINT16_MAX;
2234
2235 if (port_idx >= ofproto->max_ports
2236 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2237 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2238 long long int last_used_at, lru = LLONG_MAX;
2239
2240 /* Search for a free OpenFlow port number. We try not to
2241 * immediately reuse them to prevent problems due to old
2242 * flows.
2243 *
2244 * We limit the automatically assigned port numbers to the lower half
2245 * of the port range, to reserve the upper half for assignment by
2246 * controllers. */
2247 for (;;) {
2248 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
2249 ofproto->alloc_port_no = 1;
2250 }
2251 last_used_at = ofport_get_usage(ofproto,
2252 u16_to_ofp(ofproto->alloc_port_no));
2253 if (!last_used_at) {
2254 port_idx = ofproto->alloc_port_no;
2255 break;
2256 } else if ( last_used_at < time_msec() - 60*60*1000) {
2257 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2258 * more than an hour ago, consider it usable. */
2259 ofport_remove_usage(ofproto,
2260 u16_to_ofp(ofproto->alloc_port_no));
2261 port_idx = ofproto->alloc_port_no;
2262 break;
2263 } else if (last_used_at < lru) {
2264 lru = last_used_at;
2265 lru_ofport = ofproto->alloc_port_no;
2266 }
2267
2268 if (ofproto->alloc_port_no == end_port_no) {
2269 if (lru_ofport) {
2270 port_idx = lru_ofport;
2271 break;
2272 }
2273 return OFPP_NONE;
2274 }
2275 }
2276 }
2277 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2278 return u16_to_ofp(port_idx);
2279 }
2280
2281 static void
2282 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2283 {
2284 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2285 ofport_set_usage(ofproto, ofp_port, time_msec());
2286 }
2287 }
2288
2289 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2290 * pointer if the netdev cannot be opened. On success, also fills in
2291 * '*pp'. */
2292 static struct netdev *
2293 ofport_open(struct ofproto *ofproto,
2294 struct ofproto_port *ofproto_port,
2295 struct ofputil_phy_port *pp)
2296 {
2297 enum netdev_flags flags;
2298 struct netdev *netdev;
2299 int error;
2300
2301 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2302 if (error) {
2303 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2304 "cannot be opened (%s)",
2305 ofproto->name,
2306 ofproto_port->name, ofproto_port->ofp_port,
2307 ofproto_port->name, ovs_strerror(error));
2308 return NULL;
2309 }
2310
2311 if (ofproto_port->ofp_port == OFPP_NONE) {
2312 if (!strcmp(ofproto->name, ofproto_port->name)) {
2313 ofproto_port->ofp_port = OFPP_LOCAL;
2314 } else {
2315 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2316 ofproto_port->name);
2317 }
2318 }
2319 pp->port_no = ofproto_port->ofp_port;
2320 netdev_get_etheraddr(netdev, pp->hw_addr);
2321 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2322 netdev_get_flags(netdev, &flags);
2323 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2324 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2325 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2326 &pp->supported, &pp->peer);
2327 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2328 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2329
2330 return netdev;
2331 }
2332
2333 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2334 * port number, and 'config' bits other than OFPUTIL_PC_PORT_DOWN are
2335 * disregarded. */
2336 static bool
2337 ofport_equal(const struct ofputil_phy_port *a,
2338 const struct ofputil_phy_port *b)
2339 {
2340 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2341 && a->state == b->state
2342 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2343 && a->curr == b->curr
2344 && a->advertised == b->advertised
2345 && a->supported == b->supported
2346 && a->peer == b->peer
2347 && a->curr_speed == b->curr_speed
2348 && a->max_speed == b->max_speed);
2349 }
2350
2351 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2352 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2353 * one with the same name or port number). */
2354 static void
2355 ofport_install(struct ofproto *p,
2356 struct netdev *netdev, const struct ofputil_phy_port *pp)
2357 {
2358 const char *netdev_name = netdev_get_name(netdev);
2359 struct ofport *ofport;
2360 int error;
2361
2362 /* Create ofport. */
2363 ofport = p->ofproto_class->port_alloc();
2364 if (!ofport) {
2365 error = ENOMEM;
2366 goto error;
2367 }
2368 ofport->ofproto = p;
2369 ofport->netdev = netdev;
2370 ofport->change_seq = netdev_get_change_seq(netdev);
2371 ofport->pp = *pp;
2372 ofport->ofp_port = pp->port_no;
2373 ofport->created = time_msec();
2374
2375 /* Add port to 'p'. */
2376 hmap_insert(&p->ports, &ofport->hmap_node,
2377 hash_ofp_port(ofport->ofp_port));
2378 shash_add(&p->port_by_name, netdev_name, ofport);
2379
2380 update_mtu(p, ofport);
2381
2382 /* Let the ofproto_class initialize its private data. */
2383 error = p->ofproto_class->port_construct(ofport);
2384 if (error) {
2385 goto error;
2386 }
2387 connmgr_send_port_status(p->connmgr, NULL, pp, OFPPR_ADD);
2388 return;
2389
2390 error:
2391 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2392 p->name, netdev_name, ovs_strerror(error));
2393 if (ofport) {
2394 ofport_destroy__(ofport);
2395 } else {
2396 netdev_close(netdev);
2397 }
2398 }
2399
2400 /* Removes 'ofport' from 'p' and destroys it. */
2401 static void
2402 ofport_remove(struct ofport *ofport)
2403 {
2404 connmgr_send_port_status(ofport->ofproto->connmgr, NULL, &ofport->pp,
2405 OFPPR_DELETE);
2406 ofport_destroy(ofport);
2407 }
2408
2409 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2410 * destroys it. */
2411 static void
2412 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2413 {
2414 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2415 if (port) {
2416 ofport_remove(port);
2417 }
2418 }
2419
2420 /* Updates 'port' with new 'pp' description.
2421 *
2422 * Does not handle a name or port number change. The caller must implement
2423 * such a change as a delete followed by an add. */
2424 static void
2425 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2426 {
2427 memcpy(port->pp.hw_addr, pp->hw_addr, ETH_ADDR_LEN);
2428 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2429 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2430 port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN)
2431 | (pp->state & OFPUTIL_PS_LINK_DOWN));
2432 port->pp.curr = pp->curr;
2433 port->pp.advertised = pp->advertised;
2434 port->pp.supported = pp->supported;
2435 port->pp.peer = pp->peer;
2436 port->pp.curr_speed = pp->curr_speed;
2437 port->pp.max_speed = pp->max_speed;
2438
2439 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2440 &port->pp, OFPPR_MODIFY);
2441 }
2442
2443 /* Update OpenFlow 'state' in 'port' and notify controller. */
2444 void
2445 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2446 {
2447 if (port->pp.state != state) {
2448 port->pp.state = state;
2449 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2450 &port->pp, OFPPR_MODIFY);
2451 }
2452 }
2453
2454 void
2455 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2456 {
2457 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2458 if (port) {
2459 if (port->ofproto->ofproto_class->set_realdev) {
2460 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2461 }
2462 if (port->ofproto->ofproto_class->set_stp_port) {
2463 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2464 }
2465 if (port->ofproto->ofproto_class->set_rstp_port) {
2466 port->ofproto->ofproto_class->set_rstp_port(port, NULL);
2467 }
2468 if (port->ofproto->ofproto_class->set_cfm) {
2469 port->ofproto->ofproto_class->set_cfm(port, NULL);
2470 }
2471 if (port->ofproto->ofproto_class->bundle_remove) {
2472 port->ofproto->ofproto_class->bundle_remove(port);
2473 }
2474 }
2475 }
2476
2477 static void
2478 ofport_destroy__(struct ofport *port)
2479 {
2480 struct ofproto *ofproto = port->ofproto;
2481 const char *name = netdev_get_name(port->netdev);
2482
2483 hmap_remove(&ofproto->ports, &port->hmap_node);
2484 shash_delete(&ofproto->port_by_name,
2485 shash_find(&ofproto->port_by_name, name));
2486
2487 netdev_close(port->netdev);
2488 ofproto->ofproto_class->port_dealloc(port);
2489 }
2490
2491 static void
2492 ofport_destroy(struct ofport *port)
2493 {
2494 if (port) {
2495 dealloc_ofp_port(port->ofproto, port->ofp_port);
2496 port->ofproto->ofproto_class->port_destruct(port);
2497 ofport_destroy__(port);
2498 }
2499 }
2500
2501 struct ofport *
2502 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2503 {
2504 struct ofport *port;
2505
2506 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2507 &ofproto->ports) {
2508 if (port->ofp_port == ofp_port) {
2509 return port;
2510 }
2511 }
2512 return NULL;
2513 }
2514
2515 static long long int
2516 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2517 {
2518 struct ofport_usage *usage;
2519
2520 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2521 &ofproto->ofport_usage) {
2522 if (usage->ofp_port == ofp_port) {
2523 return usage->last_used;
2524 }
2525 }
2526 return 0;
2527 }
2528
2529 static void
2530 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2531 long long int last_used)
2532 {
2533 struct ofport_usage *usage;
2534 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2535 &ofproto->ofport_usage) {
2536 if (usage->ofp_port == ofp_port) {
2537 usage->last_used = last_used;
2538 return;
2539 }
2540 }
2541 ovs_assert(last_used == LLONG_MAX);
2542
2543 usage = xmalloc(sizeof *usage);
2544 usage->ofp_port = ofp_port;
2545 usage->last_used = last_used;
2546 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2547 hash_ofp_port(ofp_port));
2548 }
2549
2550 static void
2551 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2552 {
2553 struct ofport_usage *usage;
2554 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2555 &ofproto->ofport_usage) {
2556 if (usage->ofp_port == ofp_port) {
2557 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2558 free(usage);
2559 break;
2560 }
2561 }
2562 }
2563
2564 int
2565 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2566 {
2567 struct ofproto *ofproto = port->ofproto;
2568 int error;
2569
2570 if (ofproto->ofproto_class->port_get_stats) {
2571 error = ofproto->ofproto_class->port_get_stats(port, stats);
2572 } else {
2573 error = EOPNOTSUPP;
2574 }
2575
2576 return error;
2577 }
2578
2579 static void
2580 update_port(struct ofproto *ofproto, const char *name)
2581 {
2582 struct ofproto_port ofproto_port;
2583 struct ofputil_phy_port pp;
2584 struct netdev *netdev;
2585 struct ofport *port;
2586
2587 COVERAGE_INC(ofproto_update_port);
2588
2589 /* Fetch 'name''s location and properties from the datapath. */
2590 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2591 ? ofport_open(ofproto, &ofproto_port, &pp)
2592 : NULL);
2593
2594 if (netdev) {
2595 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2596 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2597 struct netdev *old_netdev = port->netdev;
2598
2599 /* 'name' hasn't changed location. Any properties changed? */
2600 if (!ofport_equal(&port->pp, &pp)) {
2601 ofport_modified(port, &pp);
2602 }
2603
2604 update_mtu(ofproto, port);
2605
2606 /* Install the newly opened netdev in case it has changed.
2607 * Don't close the old netdev yet in case port_modified has to
2608 * remove a retained reference to it.*/
2609 port->netdev = netdev;
2610 port->change_seq = netdev_get_change_seq(netdev);
2611
2612 if (port->ofproto->ofproto_class->port_modified) {
2613 port->ofproto->ofproto_class->port_modified(port);
2614 }
2615
2616 netdev_close(old_netdev);
2617 } else {
2618 /* If 'port' is nonnull then its name differs from 'name' and thus
2619 * we should delete it. If we think there's a port named 'name'
2620 * then its port number must be wrong now so delete it too. */
2621 if (port) {
2622 ofport_remove(port);
2623 }
2624 ofport_remove_with_name(ofproto, name);
2625 ofport_install(ofproto, netdev, &pp);
2626 }
2627 } else {
2628 /* Any port named 'name' is gone now. */
2629 ofport_remove_with_name(ofproto, name);
2630 }
2631 ofproto_port_destroy(&ofproto_port);
2632 }
2633
2634 static int
2635 init_ports(struct ofproto *p)
2636 {
2637 struct ofproto_port_dump dump;
2638 struct ofproto_port ofproto_port;
2639 struct shash_node *node, *next;
2640
2641 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2642 const char *name = ofproto_port.name;
2643
2644 if (shash_find(&p->port_by_name, name)) {
2645 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2646 p->name, name);
2647 } else {
2648 struct ofputil_phy_port pp;
2649 struct netdev *netdev;
2650
2651 /* Check if an OpenFlow port number had been requested. */
2652 node = shash_find(&init_ofp_ports, name);
2653 if (node) {
2654 const struct iface_hint *iface_hint = node->data;
2655 simap_put(&p->ofp_requests, name,
2656 ofp_to_u16(iface_hint->ofp_port));
2657 }
2658
2659 netdev = ofport_open(p, &ofproto_port, &pp);
2660 if (netdev) {
2661 ofport_install(p, netdev, &pp);
2662 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2663 p->alloc_port_no = MAX(p->alloc_port_no,
2664 ofp_to_u16(ofproto_port.ofp_port));
2665 }
2666 }
2667 }
2668 }
2669
2670 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2671 struct iface_hint *iface_hint = node->data;
2672
2673 if (!strcmp(iface_hint->br_name, p->name)) {
2674 free(iface_hint->br_name);
2675 free(iface_hint->br_type);
2676 free(iface_hint);
2677 shash_delete(&init_ofp_ports, node);
2678 }
2679 }
2680
2681 return 0;
2682 }
2683
2684 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2685 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2686 static int
2687 find_min_mtu(struct ofproto *p)
2688 {
2689 struct ofport *ofport;
2690 int mtu = 0;
2691
2692 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2693 struct netdev *netdev = ofport->netdev;
2694 int dev_mtu;
2695
2696 /* Skip any internal ports, since that's what we're trying to
2697 * set. */
2698 if (!strcmp(netdev_get_type(netdev), "internal")) {
2699 continue;
2700 }
2701
2702 if (netdev_get_mtu(netdev, &dev_mtu)) {
2703 continue;
2704 }
2705 if (!mtu || dev_mtu < mtu) {
2706 mtu = dev_mtu;
2707 }
2708 }
2709
2710 return mtu ? mtu: ETH_PAYLOAD_MAX;
2711 }
2712
2713 /* Update MTU of all datapath devices on 'p' to the minimum of the
2714 * non-datapath ports in event of 'port' added or changed. */
2715 static void
2716 update_mtu(struct ofproto *p, struct ofport *port)
2717 {
2718 struct ofport *ofport;
2719 struct netdev *netdev = port->netdev;
2720 int dev_mtu, old_min;
2721
2722 if (netdev_get_mtu(netdev, &dev_mtu)) {
2723 port->mtu = 0;
2724 return;
2725 }
2726 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2727 if (dev_mtu > p->min_mtu) {
2728 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2729 dev_mtu = p->min_mtu;
2730 }
2731 }
2732 port->mtu = dev_mtu;
2733 return;
2734 }
2735
2736 /* For non-internal port find new min mtu. */
2737 old_min = p->min_mtu;
2738 port->mtu = dev_mtu;
2739 p->min_mtu = find_min_mtu(p);
2740 if (p->min_mtu == old_min) {
2741 return;
2742 }
2743
2744 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2745 struct netdev *netdev = ofport->netdev;
2746
2747 if (!strcmp(netdev_get_type(netdev), "internal")) {
2748 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2749 ofport->mtu = p->min_mtu;
2750 }
2751 }
2752 }
2753 }
2754 \f
2755 static void
2756 ofproto_rule_destroy__(struct rule *rule)
2757 OVS_NO_THREAD_SAFETY_ANALYSIS
2758 {
2759 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2760 rule_actions_destroy(rule_get_actions(rule));
2761 ovs_mutex_destroy(&rule->mutex);
2762 rule->ofproto->ofproto_class->rule_dealloc(rule);
2763 }
2764
2765 static void
2766 rule_destroy_cb(struct rule *rule)
2767 OVS_NO_THREAD_SAFETY_ANALYSIS
2768 {
2769 /* Send rule removed if needed. */
2770 if (rule->flags & OFPUTIL_FF_SEND_FLOW_REM
2771 && rule->removed_reason != OVS_OFPRR_NONE
2772 && !rule_is_hidden(rule)) {
2773 ofproto_rule_send_removed(rule);
2774 }
2775 rule->ofproto->ofproto_class->rule_destruct(rule);
2776 ofproto_rule_destroy__(rule);
2777 }
2778
2779 void
2780 ofproto_rule_ref(struct rule *rule)
2781 {
2782 if (rule) {
2783 ovs_refcount_ref(&rule->ref_count);
2784 }
2785 }
2786
2787 bool
2788 ofproto_rule_try_ref(struct rule *rule)
2789 {
2790 if (rule) {
2791 return ovs_refcount_try_ref_rcu(&rule->ref_count);
2792 }
2793 return false;
2794 }
2795
2796 /* Decrements 'rule''s ref_count and schedules 'rule' to be destroyed if the
2797 * ref_count reaches 0.
2798 *
2799 * Use of RCU allows short term use (between RCU quiescent periods) without
2800 * keeping a reference. A reference must be taken if the rule needs to
2801 * stay around accross the RCU quiescent periods. */
2802 void
2803 ofproto_rule_unref(struct rule *rule)
2804 {
2805 if (rule && ovs_refcount_unref_relaxed(&rule->ref_count) == 1) {
2806 ovsrcu_postpone(rule_destroy_cb, rule);
2807 }
2808 }
2809
2810 static void
2811 remove_rule_rcu__(struct rule *rule)
2812 OVS_REQUIRES(ofproto_mutex)
2813 {
2814 struct ofproto *ofproto = rule->ofproto;
2815 struct oftable *table = &ofproto->tables[rule->table_id];
2816
2817 ovs_assert(!cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
2818 if (!classifier_remove(&table->cls, &rule->cr)) {
2819 OVS_NOT_REACHED();
2820 }
2821 ofproto->ofproto_class->rule_delete(rule);
2822 ofproto_rule_unref(rule);
2823 }
2824
2825 static void
2826 remove_rule_rcu(struct rule *rule)
2827 OVS_EXCLUDED(ofproto_mutex)
2828 {
2829 ovs_mutex_lock(&ofproto_mutex);
2830 remove_rule_rcu__(rule);
2831 ovs_mutex_unlock(&ofproto_mutex);
2832 }
2833
2834 /* Removes and deletes rules from a NULL-terminated array of rule pointers. */
2835 static void
2836 remove_rules_rcu(struct rule **rules)
2837 OVS_EXCLUDED(ofproto_mutex)
2838 {
2839 struct rule **orig_rules = rules;
2840
2841 if (*rules) {
2842 struct ofproto *ofproto = rules[0]->ofproto;
2843 unsigned long tables[BITMAP_N_LONGS(256)];
2844 struct rule *rule;
2845 size_t table_id;
2846
2847 memset(tables, 0, sizeof tables);
2848
2849 ovs_mutex_lock(&ofproto_mutex);
2850 while ((rule = *rules++)) {
2851 /* Defer once for each new table. This defers the subtable cleanup
2852 * until later, so that when removing large number of flows the
2853 * operation is faster. */
2854 if (!bitmap_is_set(tables, rule->table_id)) {
2855 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
2856
2857 bitmap_set1(tables, rule->table_id);
2858 classifier_defer(cls);
2859 }
2860 remove_rule_rcu__(rule);
2861 }
2862
2863 BITMAP_FOR_EACH_1(table_id, 256, tables) {
2864 struct classifier *cls = &ofproto->tables[table_id].cls;
2865
2866 classifier_publish(cls);
2867 }
2868 ovs_mutex_unlock(&ofproto_mutex);
2869 }
2870
2871 free(orig_rules);
2872 }
2873
2874 void
2875 ofproto_group_ref(struct ofgroup *group)
2876 {
2877 if (group) {
2878 ovs_refcount_ref(&group->ref_count);
2879 }
2880 }
2881
2882 void
2883 ofproto_group_unref(struct ofgroup *group)
2884 {
2885 if (group && ovs_refcount_unref(&group->ref_count) == 1) {
2886 group->ofproto->ofproto_class->group_destruct(group);
2887 ofputil_bucket_list_destroy(&group->buckets);
2888 group->ofproto->ofproto_class->group_dealloc(group);
2889 }
2890 }
2891
2892 static uint32_t get_provider_meter_id(const struct ofproto *,
2893 uint32_t of_meter_id);
2894
2895 /* Creates and returns a new 'struct rule_actions', whose actions are a copy
2896 * of from the 'ofpacts_len' bytes of 'ofpacts'. */
2897 const struct rule_actions *
2898 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2899 {
2900 struct rule_actions *actions;
2901
2902 actions = xmalloc(sizeof *actions + ofpacts_len);
2903 actions->ofpacts_len = ofpacts_len;
2904 actions->has_meter = ofpacts_get_meter(ofpacts, ofpacts_len) != 0;
2905 memcpy(actions->ofpacts, ofpacts, ofpacts_len);
2906
2907 actions->has_learn_with_delete = (next_learn_with_delete(actions, NULL)
2908 != NULL);
2909
2910 return actions;
2911 }
2912
2913 /* Free the actions after the RCU quiescent period is reached. */
2914 void
2915 rule_actions_destroy(const struct rule_actions *actions)
2916 {
2917 if (actions) {
2918 ovsrcu_postpone(free, CONST_CAST(struct rule_actions *, actions));
2919 }
2920 }
2921
2922 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2923 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2924 bool
2925 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2926 OVS_REQUIRES(ofproto_mutex)
2927 {
2928 if (port == OFPP_ANY) {
2929 return true;
2930 } else {
2931 const struct rule_actions *actions = rule_get_actions(rule);
2932 return ofpacts_output_to_port(actions->ofpacts,
2933 actions->ofpacts_len, port);
2934 }
2935 }
2936
2937 /* Returns true if 'rule' has group and equals group_id. */
2938 static bool
2939 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2940 OVS_REQUIRES(ofproto_mutex)
2941 {
2942 if (group_id == OFPG_ANY) {
2943 return true;
2944 } else {
2945 const struct rule_actions *actions = rule_get_actions(rule);
2946 return ofpacts_output_to_group(actions->ofpacts,
2947 actions->ofpacts_len, group_id);
2948 }
2949 }
2950
2951 static void
2952 rule_execute_destroy(struct rule_execute *e)
2953 {
2954 ofproto_rule_unref(e->rule);
2955 list_remove(&e->list_node);
2956 free(e);
2957 }
2958
2959 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2960 * by passing them to the ofproto provider. */
2961 static void
2962 run_rule_executes(struct ofproto *ofproto)
2963 OVS_EXCLUDED(ofproto_mutex)
2964 {
2965 struct rule_execute *e, *next;
2966 struct ovs_list executes;
2967
2968 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2969 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2970 struct flow flow;
2971
2972 flow_extract(e->packet, &flow);
2973 flow.in_port.ofp_port = e->in_port;
2974 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2975
2976 rule_execute_destroy(e);
2977 }
2978 }
2979
2980 /* Destroys and discards all "rule_execute" operations queued up in
2981 * ofproto->rule_executes. */
2982 static void
2983 destroy_rule_executes(struct ofproto *ofproto)
2984 {
2985 struct rule_execute *e, *next;
2986 struct ovs_list executes;
2987
2988 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2989 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2990 dp_packet_delete(e->packet);
2991 rule_execute_destroy(e);
2992 }
2993 }
2994
2995 static bool
2996 rule_is_readonly(const struct rule *rule)
2997 {
2998 const struct oftable *table = &rule->ofproto->tables[rule->table_id];
2999 return (table->flags & OFTABLE_READONLY) != 0;
3000 }
3001 \f
3002 static uint32_t
3003 hash_learned_cookie(ovs_be64 cookie_, uint8_t table_id)
3004 {
3005 uint64_t cookie = (OVS_FORCE uint64_t) cookie_;
3006 return hash_3words(cookie, cookie >> 32, table_id);
3007 }
3008
3009 static void
3010 learned_cookies_update_one__(struct ofproto *ofproto,
3011 const struct ofpact_learn *learn,
3012 int delta, struct ovs_list *dead_cookies)
3013 OVS_REQUIRES(ofproto_mutex)
3014 {
3015 uint32_t hash = hash_learned_cookie(learn->cookie, learn->table_id);
3016 struct learned_cookie *c;
3017
3018 HMAP_FOR_EACH_WITH_HASH (c, u.hmap_node, hash, &ofproto->learned_cookies) {
3019 if (c->cookie == learn->cookie && c->table_id == learn->table_id) {
3020 c->n += delta;
3021 ovs_assert(c->n >= 0);
3022
3023 if (!c->n) {
3024 hmap_remove(&ofproto->learned_cookies, &c->u.hmap_node);
3025 list_push_back(dead_cookies, &c->u.list_node);
3026 }
3027
3028 return;
3029 }
3030 }
3031
3032 ovs_assert(delta > 0);
3033 c = xmalloc(sizeof *c);
3034 hmap_insert(&ofproto->learned_cookies, &c->u.hmap_node, hash);
3035 c->cookie = learn->cookie;
3036 c->table_id = learn->table_id;
3037 c->n = delta;
3038 }
3039
3040 static const struct ofpact_learn *
3041 next_learn_with_delete(const struct rule_actions *actions,
3042 const struct ofpact_learn *start)
3043 {
3044 const struct ofpact *pos;
3045
3046 for (pos = start ? ofpact_next(&start->ofpact) : actions->ofpacts;
3047 pos < ofpact_end(actions->ofpacts, actions->ofpacts_len);
3048 pos = ofpact_next(pos)) {
3049 if (pos->type == OFPACT_LEARN) {
3050 const struct ofpact_learn *learn = ofpact_get_LEARN(pos);
3051 if (learn->flags & NX_LEARN_F_DELETE_LEARNED) {
3052 return learn;
3053 }
3054 }
3055 }
3056
3057 return NULL;
3058 }
3059
3060 static void
3061 learned_cookies_update__(struct ofproto *ofproto,
3062 const struct rule_actions *actions,
3063 int delta, struct ovs_list *dead_cookies)
3064 OVS_REQUIRES(ofproto_mutex)
3065 {
3066 if (actions->has_learn_with_delete) {
3067 const struct ofpact_learn *learn;
3068
3069 for (learn = next_learn_with_delete(actions, NULL); learn;
3070 learn = next_learn_with_delete(actions, learn)) {
3071 learned_cookies_update_one__(ofproto, learn, delta, dead_cookies);
3072 }
3073 }
3074 }
3075
3076 static void
3077 learned_cookies_inc(struct ofproto *ofproto,
3078 const struct rule_actions *actions)
3079 OVS_REQUIRES(ofproto_mutex)
3080 {
3081 learned_cookies_update__(ofproto, actions, +1, NULL);
3082 }
3083
3084 static void
3085 learned_cookies_dec(struct ofproto *ofproto,
3086 const struct rule_actions *actions,
3087 struct ovs_list *dead_cookies)
3088 OVS_REQUIRES(ofproto_mutex)
3089 {
3090 learned_cookies_update__(ofproto, actions, -1, dead_cookies);
3091 }
3092
3093 static void
3094 learned_cookies_flush(struct ofproto *ofproto, struct ovs_list *dead_cookies)
3095 OVS_REQUIRES(ofproto_mutex)
3096 {
3097 struct learned_cookie *c;
3098
3099 LIST_FOR_EACH_POP (c, u.list_node, dead_cookies) {
3100 struct rule_criteria criteria;
3101 struct rule_collection rules;
3102 struct match match;
3103
3104 match_init_catchall(&match);
3105 rule_criteria_init(&criteria, c->table_id, &match, 0, CLS_MAX_VERSION,
3106 c->cookie, OVS_BE64_MAX, OFPP_ANY, OFPG_ANY);
3107 rule_criteria_require_rw(&criteria, false);
3108 collect_rules_loose(ofproto, &criteria, &rules);
3109 rule_criteria_destroy(&criteria);
3110 delete_flows__(&rules, OFPRR_DELETE, NULL);
3111
3112 free(c);
3113 }
3114 }
3115 \f
3116 static enum ofperr
3117 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
3118 {
3119 ofconn_send_reply(ofconn, make_echo_reply(oh));
3120 return 0;
3121 }
3122
3123 static void
3124 query_tables(struct ofproto *ofproto,
3125 struct ofputil_table_features **featuresp,
3126 struct ofputil_table_stats **statsp)
3127 {
3128 struct mf_bitmap rw_fields = oxm_writable_fields();
3129 struct mf_bitmap match = oxm_matchable_fields();
3130 struct mf_bitmap mask = oxm_maskable_fields();
3131
3132 struct ofputil_table_features *features;
3133 struct ofputil_table_stats *stats;
3134 int i;
3135
3136 features = *featuresp = xcalloc(ofproto->n_tables, sizeof *features);
3137 for (i = 0; i < ofproto->n_tables; i++) {
3138 struct ofputil_table_features *f = &features[i];
3139
3140 f->table_id = i;
3141 sprintf(f->name, "table%d", i);
3142 f->metadata_match = OVS_BE64_MAX;
3143 f->metadata_write = OVS_BE64_MAX;
3144 atomic_read_relaxed(&ofproto->tables[i].miss_config, &f->miss_config);
3145 f->max_entries = 1000000;
3146
3147 bool more_tables = false;
3148 for (int j = i + 1; j < ofproto->n_tables; j++) {
3149 if (!(ofproto->tables[j].flags & OFTABLE_HIDDEN)) {
3150 bitmap_set1(f->nonmiss.next, j);
3151 more_tables = true;
3152 }
3153 }
3154 f->nonmiss.instructions = (1u << N_OVS_INSTRUCTIONS) - 1;
3155 if (!more_tables) {
3156 f->nonmiss.instructions &= ~(1u << OVSINST_OFPIT11_GOTO_TABLE);
3157 }
3158 f->nonmiss.write.ofpacts = (UINT64_C(1) << N_OFPACTS) - 1;
3159 f->nonmiss.write.set_fields = rw_fields;
3160 f->nonmiss.apply = f->nonmiss.write;
3161 f->miss = f->nonmiss;
3162
3163 f->match = match;
3164 f->mask = mask;
3165 f->wildcard = match;
3166 }
3167
3168 if (statsp) {
3169 stats = *statsp = xcalloc(ofproto->n_tables, sizeof *stats);
3170 for (i = 0; i < ofproto->n_tables; i++) {
3171 struct ofputil_table_stats *s = &stats[i];
3172
3173 s->table_id = i;
3174 s->active_count = ofproto->tables[i].n_flows;
3175 if (i == 0) {
3176 s->active_count -= connmgr_count_hidden_rules(
3177 ofproto->connmgr);
3178 }
3179 }
3180 } else {
3181 stats = NULL;
3182 }
3183
3184 ofproto->ofproto_class->query_tables(ofproto, features, stats);
3185
3186 for (i = 0; i < ofproto->n_tables; i++) {
3187 const struct oftable *table = &ofproto->tables[i];
3188 struct ofputil_table_features *f = &features[i];
3189
3190 if (table->name) {
3191 ovs_strzcpy(f->name, table->name, sizeof f->name);
3192 }
3193
3194 if (table->max_flows < f->max_entries) {
3195 f->max_entries = table->max_flows;
3196 }
3197 }
3198 }
3199
3200 static void
3201 query_switch_features(struct ofproto *ofproto,
3202 bool *arp_match_ip, uint64_t *ofpacts)
3203 {
3204 struct ofputil_table_features *features, *f;
3205
3206 *arp_match_ip = false;
3207 *ofpacts = 0;
3208
3209 query_tables(ofproto, &features, NULL);
3210 for (f = features; f < &features[ofproto->n_tables]; f++) {
3211 *ofpacts |= f->nonmiss.apply.ofpacts | f->miss.apply.ofpacts;
3212 if (bitmap_is_set(f->match.bm, MFF_ARP_SPA) ||
3213 bitmap_is_set(f->match.bm, MFF_ARP_TPA)) {
3214 *arp_match_ip = true;
3215 }
3216 }
3217 free(features);
3218
3219 /* Sanity check. */
3220 ovs_assert(*ofpacts & (UINT64_C(1) << OFPACT_OUTPUT));
3221 }
3222
3223 static enum ofperr
3224 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
3225 {
3226 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3227 struct ofputil_switch_features features;
3228 struct ofport *port;
3229 bool arp_match_ip;
3230 struct ofpbuf *b;
3231
3232 query_switch_features(ofproto, &arp_match_ip, &features.ofpacts);
3233
3234 features.datapath_id = ofproto->datapath_id;
3235 features.n_buffers = pktbuf_capacity();
3236 features.n_tables = ofproto_get_n_visible_tables(ofproto);
3237 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
3238 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS |
3239 OFPUTIL_C_GROUP_STATS);
3240 if (arp_match_ip) {
3241 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
3242 }
3243 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
3244 features.auxiliary_id = 0;
3245 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
3246 oh->xid);
3247 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3248 ofputil_put_switch_features_port(&port->pp, b);
3249 }
3250
3251 ofconn_send_reply(ofconn, b);
3252 return 0;
3253 }
3254
3255 static enum ofperr
3256 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
3257 {
3258 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3259 struct ofp_switch_config *osc;
3260 enum ofp_config_flags flags;
3261 struct ofpbuf *buf;
3262
3263 /* Send reply. */
3264 buf = ofpraw_alloc_reply(OFPRAW_OFPT_GET_CONFIG_REPLY, oh, 0);
3265 osc = ofpbuf_put_uninit(buf, sizeof *osc);
3266 flags = ofproto->frag_handling;
3267 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
3268 if (oh->version < OFP13_VERSION
3269 && ofconn_get_invalid_ttl_to_controller(ofconn)) {
3270 flags |= OFPC_INVALID_TTL_TO_CONTROLLER;
3271 }
3272 osc->flags = htons(flags);
3273 osc->miss_send_len = htons(ofconn_get_miss_send_len(ofconn));
3274 ofconn_send_reply(ofconn, buf);
3275
3276 return 0;
3277 }
3278
3279 static enum ofperr
3280 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
3281 {
3282 const struct ofp_switch_config *osc = ofpmsg_body(oh);
3283 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3284 uint16_t flags = ntohs(osc->flags);
3285
3286 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
3287 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
3288 enum ofp_config_flags cur = ofproto->frag_handling;
3289 enum ofp_config_flags next = flags & OFPC_FRAG_MASK;
3290
3291 ovs_assert((cur & OFPC_FRAG_MASK) == cur);
3292 if (cur != next) {
3293 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
3294 ofproto->frag_handling = next;
3295 } else {
3296 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
3297 ofproto->name,
3298 ofputil_frag_handling_to_string(next));
3299 }
3300 }
3301 }
3302 /* OFPC_INVALID_TTL_TO_CONTROLLER is deprecated in OF 1.3 */
3303 ofconn_set_invalid_ttl_to_controller(ofconn,
3304 (oh->version < OFP13_VERSION
3305 && flags & OFPC_INVALID_TTL_TO_CONTROLLER));
3306
3307 ofconn_set_miss_send_len(ofconn, ntohs(osc->miss_send_len));
3308
3309 return 0;
3310 }
3311
3312 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3313 * error message code for the caller to propagate upward. Otherwise, returns
3314 * 0.
3315 *
3316 * The log message mentions 'msg_type'. */
3317 static enum ofperr
3318 reject_slave_controller(struct ofconn *ofconn)
3319 {
3320 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
3321 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
3322 return OFPERR_OFPBRC_IS_SLAVE;
3323 } else {
3324 return 0;
3325 }
3326 }
3327
3328 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
3329 * for 'ofproto':
3330 *
3331 * - If they use a meter, then 'ofproto' has that meter configured.
3332 *
3333 * - If they use any groups, then 'ofproto' has that group configured.
3334 *
3335 * Returns 0 if successful, otherwise an OpenFlow error. */
3336 static enum ofperr
3337 ofproto_check_ofpacts(struct ofproto *ofproto,
3338 const struct ofpact ofpacts[], size_t ofpacts_len)
3339 {
3340 const struct ofpact *a;
3341 uint32_t mid;
3342
3343 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
3344 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
3345 return OFPERR_OFPMMFC_INVALID_METER;
3346 }
3347
3348 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
3349 if (a->type == OFPACT_GROUP
3350 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
3351 return OFPERR_OFPBAC_BAD_OUT_GROUP;
3352 }
3353 }
3354
3355 return 0;
3356 }
3357
3358 static enum ofperr
3359 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3360 {
3361 struct ofproto *p = ofconn_get_ofproto(ofconn);
3362 struct ofputil_packet_out po;
3363 struct dp_packet *payload;
3364 uint64_t ofpacts_stub[1024 / 8];
3365 struct ofpbuf ofpacts;
3366 struct flow flow;
3367 enum ofperr error;
3368
3369 COVERAGE_INC(ofproto_packet_out);
3370
3371 error = reject_slave_controller(ofconn);
3372 if (error) {
3373 goto exit;
3374 }
3375
3376 /* Decode message. */
3377 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3378 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
3379 if (error) {
3380 goto exit_free_ofpacts;
3381 }
3382 if (ofp_to_u16(po.in_port) >= p->max_ports
3383 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
3384 error = OFPERR_OFPBRC_BAD_PORT;
3385 goto exit_free_ofpacts;
3386 }
3387
3388 /* Get payload. */
3389 if (po.buffer_id != UINT32_MAX) {
3390 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
3391 if (error || !payload) {
3392 goto exit_free_ofpacts;
3393 }
3394 } else {
3395 /* Ensure that the L3 header is 32-bit aligned. */
3396 payload = dp_packet_clone_data_with_headroom(po.packet, po.packet_len, 2);
3397 }
3398
3399 /* Verify actions against packet, then send packet if successful. */
3400 flow_extract(payload, &flow);
3401 flow.in_port.ofp_port = po.in_port;
3402 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
3403 if (!error) {
3404 error = p->ofproto_class->packet_out(p, payload, &flow,
3405 po.ofpacts, po.ofpacts_len);
3406 }
3407 dp_packet_delete(payload);
3408
3409 exit_free_ofpacts:
3410 ofpbuf_uninit(&ofpacts);
3411 exit:
3412 return error;
3413 }
3414
3415 static void
3416 update_port_config(struct ofconn *ofconn, struct ofport *port,
3417 enum ofputil_port_config config,
3418 enum ofputil_port_config mask)
3419 {
3420 enum ofputil_port_config toggle = (config ^ port->pp.config) & mask;
3421
3422 if (toggle & OFPUTIL_PC_PORT_DOWN
3423 && (config & OFPUTIL_PC_PORT_DOWN
3424 ? netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL)
3425 : netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL))) {
3426 /* We tried to bring the port up or down, but it failed, so don't
3427 * update the "down" bit. */
3428 toggle &= ~OFPUTIL_PC_PORT_DOWN;
3429 }
3430
3431 if (toggle) {
3432 enum ofputil_port_config old_config = port->pp.config;
3433 port->pp.config ^= toggle;
3434 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
3435 connmgr_send_port_status(port->ofproto->connmgr, ofconn, &port->pp,
3436 OFPPR_MODIFY);
3437 }
3438 }
3439
3440 static enum ofperr
3441 port_mod_start(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3442 struct ofport **port)
3443 {
3444 struct ofproto *p = ofconn_get_ofproto(ofconn);
3445
3446 *port = ofproto_get_port(p, pm->port_no);
3447 if (!*port) {
3448 return OFPERR_OFPPMFC_BAD_PORT;
3449 }
3450 if (!eth_addr_equals((*port)->pp.hw_addr, pm->hw_addr)) {
3451 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3452 }
3453 return 0;
3454 }
3455
3456 static void
3457 port_mod_finish(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3458 struct ofport *port)
3459 {
3460 update_port_config(ofconn, port, pm->config, pm->mask);
3461 if (pm->advertise) {
3462 netdev_set_advertisements(port->netdev, pm->advertise);
3463 }
3464 }
3465
3466 static enum ofperr
3467 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3468 {
3469 struct ofputil_port_mod pm;
3470 struct ofport *port;
3471 enum ofperr error;
3472
3473 error = reject_slave_controller(ofconn);
3474 if (error) {
3475 return error;
3476 }
3477
3478 error = ofputil_decode_port_mod(oh, &pm, false);
3479 if (error) {
3480 return error;
3481 }
3482
3483 error = port_mod_start(ofconn, &pm, &port);
3484 if (!error) {
3485 port_mod_finish(ofconn, &pm, port);
3486 }
3487 return error;
3488 }
3489
3490 static enum ofperr
3491 handle_desc_stats_request(struct ofconn *ofconn,
3492 const struct ofp_header *request)
3493 {
3494 static const char *default_mfr_desc = "Nicira, Inc.";
3495 static const char *default_hw_desc = "Open vSwitch";
3496 static const char *default_sw_desc = VERSION;
3497 static const char *default_serial_desc = "None";
3498 static const char *default_dp_desc = "None";
3499
3500 struct ofproto *p = ofconn_get_ofproto(ofconn);
3501 struct ofp_desc_stats *ods;
3502 struct ofpbuf *msg;
3503
3504 msg = ofpraw_alloc_stats_reply(request, 0);
3505 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3506 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3507 sizeof ods->mfr_desc);
3508 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3509 sizeof ods->hw_desc);
3510 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3511 sizeof ods->sw_desc);
3512 ovs_strlcpy(ods->serial_num,
3513 p->serial_desc ? p->serial_desc : default_serial_desc,
3514 sizeof ods->serial_num);
3515 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3516 sizeof ods->dp_desc);
3517 ofconn_send_reply(ofconn, msg);
3518
3519 return 0;
3520 }
3521
3522 static enum ofperr
3523 handle_table_stats_request(struct ofconn *ofconn,
3524 const struct ofp_header *request)
3525 {
3526 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3527 struct ofputil_table_features *features;
3528 struct ofputil_table_stats *stats;
3529 struct ofpbuf *reply;
3530 size_t i;
3531
3532 query_tables(ofproto, &features, &stats);
3533
3534 reply = ofputil_encode_table_stats_reply(request);
3535 for (i = 0; i < ofproto->n_tables; i++) {
3536 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3537 ofputil_append_table_stats_reply(reply, &stats[i], &features[i]);
3538 }
3539 }
3540 ofconn_send_reply(ofconn, reply);
3541
3542 free(features);
3543 free(stats);
3544
3545 return 0;
3546 }
3547
3548 static enum ofperr
3549 handle_table_features_request(struct ofconn *ofconn,
3550 const struct ofp_header *request)
3551 {
3552 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3553 struct ofputil_table_features *features;
3554 struct ovs_list replies;
3555 struct ofpbuf msg;
3556 size_t i;
3557
3558 ofpbuf_use_const(&msg, request, ntohs(request->length));
3559 ofpraw_pull_assert(&msg);
3560 if (msg.size || ofpmp_more(request)) {
3561 return OFPERR_OFPTFFC_EPERM;
3562 }
3563
3564 query_tables(ofproto, &features, NULL);
3565
3566 ofpmp_init(&replies, request);
3567 for (i = 0; i < ofproto->n_tables; i++) {
3568 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3569 ofputil_append_table_features_reply(&features[i], &replies);
3570 }
3571 }
3572 ofconn_send_replies(ofconn, &replies);
3573
3574 free(features);
3575
3576 return 0;
3577 }
3578
3579 /* This function queries the database for dumping table-desc. */
3580 static void
3581 query_tables_desc(struct ofproto *ofproto, struct ofputil_table_desc **descp)
3582 {
3583 struct ofputil_table_desc *table_desc;
3584 size_t i;
3585
3586 table_desc = *descp = xcalloc(ofproto->n_tables, sizeof *table_desc);
3587 for (i = 0; i < ofproto->n_tables; i++) {
3588 struct ofputil_table_desc *td = &table_desc[i];
3589 td->table_id = i;
3590 td->eviction = (ofproto->tables[i].eviction & EVICTION_OPENFLOW
3591 ? OFPUTIL_TABLE_EVICTION_ON
3592 : OFPUTIL_TABLE_EVICTION_OFF);
3593 td->eviction_flags = OFPROTO_EVICTION_FLAGS;
3594 }
3595 }
3596
3597 /* Function to handle dump-table-desc request. */
3598 static enum ofperr
3599 handle_table_desc_request(struct ofconn *ofconn,
3600 const struct ofp_header *request)
3601 {
3602 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3603 struct ofputil_table_desc *table_desc;
3604 struct ovs_list replies;
3605 size_t i;
3606
3607 query_tables_desc(ofproto, &table_desc);
3608 ofpmp_init(&replies, request);
3609 for (i = 0; i < ofproto->n_tables; i++) {
3610 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3611 ofputil_append_table_desc_reply(&table_desc[i], &replies,
3612 request->version);
3613 }
3614 }
3615 ofconn_send_replies(ofconn, &replies);
3616 free(table_desc);
3617 return 0;
3618 }
3619
3620 static void
3621 append_port_stat(struct ofport *port, struct ovs_list *replies)
3622 {
3623 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3624
3625 calc_duration(port->created, time_msec(),
3626 &ops.duration_sec, &ops.duration_nsec);
3627
3628 /* Intentionally ignore return value, since errors will set
3629 * 'stats' to all-1s, which is correct for OpenFlow, and
3630 * netdev_get_stats() will log errors. */
3631 ofproto_port_get_stats(port, &ops.stats);
3632
3633 ofputil_append_port_stat(replies, &ops);
3634 }
3635
3636 static void
3637 handle_port_request(struct ofconn *ofconn,
3638 const struct ofp_header *request, ofp_port_t port_no,
3639 void (*cb)(struct ofport *, struct ovs_list *replies))
3640 {
3641 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3642 struct ofport *port;
3643 struct ovs_list replies;
3644
3645 ofpmp_init(&replies, request);
3646 if (port_no != OFPP_ANY) {
3647 port = ofproto_get_port(ofproto, port_no);
3648 if (port) {
3649 cb(port, &replies);
3650 }
3651 } else {
3652 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3653 cb(port, &replies);
3654 }
3655 }
3656
3657 ofconn_send_replies(ofconn, &replies);
3658 }
3659
3660 static enum ofperr
3661 handle_port_stats_request(struct ofconn *ofconn,
3662 const struct ofp_header *request)
3663 {
3664 ofp_port_t port_no;
3665 enum ofperr error;
3666
3667 error = ofputil_decode_port_stats_request(request, &port_no);
3668 if (!error) {
3669 handle_port_request(ofconn, request, port_no, append_port_stat);
3670 }
3671 return error;
3672 }
3673
3674 static void
3675 append_port_desc(struct ofport *port, struct ovs_list *replies)
3676 {
3677 ofputil_append_port_desc_stats_reply(&port->pp, replies);
3678 }
3679
3680 static enum ofperr
3681 handle_port_desc_stats_request(struct ofconn *ofconn,
3682 const struct ofp_header *request)
3683 {
3684 ofp_port_t port_no;
3685 enum ofperr error;
3686
3687 error = ofputil_decode_port_desc_stats_request(request, &port_no);
3688 if (!error) {
3689 handle_port_request(ofconn, request, port_no, append_port_desc);
3690 }
3691 return error;
3692 }
3693
3694 static uint32_t
3695 hash_cookie(ovs_be64 cookie)
3696 {
3697 return hash_uint64((OVS_FORCE uint64_t)cookie);
3698 }
3699
3700 static void
3701 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3702 OVS_REQUIRES(ofproto_mutex)
3703 {
3704 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3705 hash_cookie(rule->flow_cookie));
3706 }
3707
3708 static void
3709 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3710 OVS_REQUIRES(ofproto_mutex)
3711 {
3712 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3713 }
3714
3715 static void
3716 calc_duration(long long int start, long long int now,
3717 uint32_t *sec, uint32_t *nsec)
3718 {
3719 long long int msecs = now - start;
3720 *sec = msecs / 1000;
3721 *nsec = (msecs % 1000) * (1000 * 1000);
3722 }
3723
3724 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3725 * true if 'table_id' is OK, false otherwise. */
3726 static bool
3727 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3728 {
3729 return table_id == OFPTT_ALL || table_id < ofproto->n_tables;
3730 }
3731
3732 static struct oftable *
3733 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3734 {
3735 struct oftable *table;
3736
3737 for (table = &ofproto->tables[table_id];
3738 table < &ofproto->tables[ofproto->n_tables];
3739 table++) {
3740 if (!(table->flags & OFTABLE_HIDDEN)) {
3741 return table;
3742 }
3743 }
3744
3745 return NULL;
3746 }
3747
3748 static struct oftable *
3749 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3750 {
3751 if (table_id == 0xff) {
3752 return next_visible_table(ofproto, 0);
3753 } else if (table_id < ofproto->n_tables) {
3754 return &ofproto->tables[table_id];
3755 } else {
3756 return NULL;
3757 }
3758 }
3759
3760 static struct oftable *
3761 next_matching_table(const struct ofproto *ofproto,
3762 const struct oftable *table, uint8_t table_id)
3763 {
3764 return (table_id == 0xff
3765 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3766 : NULL);
3767 }
3768
3769 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3770 *
3771 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3772 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3773 *
3774 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3775 * only once, for that table. (This can be used to access tables marked
3776 * OFTABLE_HIDDEN.)
3777 *
3778 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3779 * entered at all. (Perhaps you should have validated TABLE_ID with
3780 * check_table_id().)
3781 *
3782 * All parameters are evaluated multiple times.
3783 */
3784 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3785 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3786 (TABLE) != NULL; \
3787 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3788
3789 /* Initializes 'criteria' in a straightforward way based on the other
3790 * parameters.
3791 *
3792 * By default, the criteria include flows that are read-only, on the assumption
3793 * that the collected flows won't be modified. Call rule_criteria_require_rw()
3794 * if flows will be modified.
3795 *
3796 * For "loose" matching, the 'priority' parameter is unimportant and may be
3797 * supplied as 0. */
3798 static void
3799 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3800 const struct match *match, int priority,
3801 cls_version_t version, ovs_be64 cookie,
3802 ovs_be64 cookie_mask, ofp_port_t out_port,
3803 uint32_t out_group)
3804 {
3805 criteria->table_id = table_id;
3806 cls_rule_init(&criteria->cr, match, priority);
3807 criteria->version = version;
3808 criteria->cookie = cookie;
3809 criteria->cookie_mask = cookie_mask;
3810 criteria->out_port = out_port;
3811 criteria->out_group = out_group;
3812
3813 /* We ordinarily want to skip hidden rules, but there has to be a way for
3814 * code internal to OVS to modify and delete them, so if the criteria
3815 * specify a priority that can only be for a hidden flow, then allow hidden
3816 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3817 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3818 * priority.) */
3819 criteria->include_hidden = priority > UINT16_MAX;
3820
3821 /* We assume that the criteria are being used to collect flows for reading
3822 * but not modification. Thus, we should collect read-only flows. */
3823 criteria->include_readonly = true;
3824 }
3825
3826 /* By default, criteria initialized by rule_criteria_init() will match flows
3827 * that are read-only, on the assumption that the collected flows won't be
3828 * modified. Call this function to match only flows that are be modifiable.
3829 *
3830 * Specify 'can_write_readonly' as false in ordinary circumstances, true if the
3831 * caller has special privileges that allow it to modify even "read-only"
3832 * flows. */
3833 static void
3834 rule_criteria_require_rw(struct rule_criteria *criteria,
3835 bool can_write_readonly)
3836 {
3837 criteria->include_readonly = can_write_readonly;
3838 }
3839
3840 static void
3841 rule_criteria_destroy(struct rule_criteria *criteria)
3842 {
3843 cls_rule_destroy(&criteria->cr);
3844 }
3845
3846 void
3847 rule_collection_init(struct rule_collection *rules)
3848 {
3849 rules->rules = rules->stub;
3850 rules->n = 0;
3851 rules->capacity = ARRAY_SIZE(rules->stub);
3852 }
3853
3854 void
3855 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3856 {
3857 if (rules->n >= rules->capacity) {
3858 size_t old_size, new_size;
3859
3860 old_size = rules->capacity * sizeof *rules->rules;
3861 rules->capacity *= 2;
3862 new_size = rules->capacity * sizeof *rules->rules;
3863
3864 if (rules->rules == rules->stub) {
3865 rules->rules = xmalloc(new_size);
3866 memcpy(rules->rules, rules->stub, old_size);
3867 } else {
3868 rules->rules = xrealloc(rules->rules, new_size);
3869 }
3870 }
3871
3872 rules->rules[rules->n++] = rule;
3873 }
3874
3875 void
3876 rule_collection_ref(struct rule_collection *rules)
3877 OVS_REQUIRES(ofproto_mutex)
3878 {
3879 size_t i;
3880
3881 for (i = 0; i < rules->n; i++) {
3882 ofproto_rule_ref(rules->rules[i]);
3883 }
3884 }
3885
3886 void
3887 rule_collection_unref(struct rule_collection *rules)
3888 {
3889 size_t i;
3890
3891 for (i = 0; i < rules->n; i++) {
3892 ofproto_rule_unref(rules->rules[i]);
3893 }
3894 }
3895
3896 /* Returns a NULL-terminated array of rule pointers,
3897 * destroys 'rules'. */
3898 static struct rule **
3899 rule_collection_detach(struct rule_collection *rules)
3900 {
3901 struct rule **rule_array;
3902
3903 rule_collection_add(rules, NULL);
3904
3905 if (rules->rules == rules->stub) {
3906 rules->rules = xmemdup(rules->rules, rules->n * sizeof *rules->rules);
3907 }
3908
3909 rule_array = rules->rules;
3910 rule_collection_init(rules);
3911
3912 return rule_array;
3913 }
3914
3915 void
3916 rule_collection_destroy(struct rule_collection *rules)
3917 {
3918 if (rules->rules != rules->stub) {
3919 free(rules->rules);
3920 }
3921
3922 /* Make repeated destruction harmless. */
3923 rule_collection_init(rules);
3924 }
3925
3926 /* Schedules postponed removal of rules, destroys 'rules'. */
3927 static void
3928 rule_collection_remove_postponed(struct rule_collection *rules)
3929 OVS_REQUIRES(ofproto_mutex)
3930 {
3931 if (rules->n > 0) {
3932 if (rules->n == 1) {
3933 ovsrcu_postpone(remove_rule_rcu, rules->rules[0]);
3934 } else {
3935 ovsrcu_postpone(remove_rules_rcu, rule_collection_detach(rules));
3936 }
3937 }
3938 }
3939
3940 /* Checks whether 'rule' matches 'c' and, if so, adds it to 'rules'. This
3941 * function verifies most of the criteria in 'c' itself, but the caller must
3942 * check 'c->cr' itself.
3943 *
3944 * Rules that have already been marked for removal are not collected.
3945 *
3946 * Increments '*n_readonly' if 'rule' wasn't added because it's read-only (and
3947 * 'c' only includes modifiable rules). */
3948 static void
3949 collect_rule(struct rule *rule, const struct rule_criteria *c,
3950 struct rule_collection *rules, size_t *n_readonly)
3951 OVS_REQUIRES(ofproto_mutex)
3952 {
3953 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3954 && ofproto_rule_has_out_port(rule, c->out_port)
3955 && ofproto_rule_has_out_group(rule, c->out_group)
3956 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)
3957 && (!rule_is_hidden(rule) || c->include_hidden)
3958 && cls_rule_visible_in_version(&rule->cr, c->version)) {
3959 /* Rule matches all the criteria... */
3960 if (!rule_is_readonly(rule) || c->include_readonly) {
3961 /* ...add it. */
3962 rule_collection_add(rules, rule);
3963 } else {
3964 /* ...except it's read-only. */
3965 ++*n_readonly;
3966 }
3967 }
3968 }
3969
3970 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3971 * on classifiers rules are done in the "loose" way required for OpenFlow
3972 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3973 * 'rules'.
3974 *
3975 * Returns 0 on success, otherwise an OpenFlow error code. */
3976 static enum ofperr
3977 collect_rules_loose(struct ofproto *ofproto,
3978 const struct rule_criteria *criteria,
3979 struct rule_collection *rules)
3980 OVS_REQUIRES(ofproto_mutex)
3981 {
3982 struct oftable *table;
3983 enum ofperr error = 0;
3984 size_t n_readonly = 0;
3985
3986 rule_collection_init(rules);
3987
3988 if (!check_table_id(ofproto, criteria->table_id)) {
3989 error = OFPERR_OFPBRC_BAD_TABLE_ID;
3990 goto exit;
3991 }
3992
3993 if (criteria->cookie_mask == OVS_BE64_MAX) {
3994 struct rule *rule;
3995
3996 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
3997 hash_cookie(criteria->cookie),
3998 &ofproto->cookies) {
3999 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
4000 collect_rule(rule, criteria, rules, &n_readonly);
4001 }
4002 }
4003 } else {
4004 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4005 struct rule *rule;
4006
4007 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &criteria->cr,
4008 criteria->version) {
4009 collect_rule(rule, criteria, rules, &n_readonly);
4010 }
4011 }
4012 }
4013
4014 exit:
4015 if (!error && !rules->n && n_readonly) {
4016 /* We didn't find any rules to modify. We did find some read-only
4017 * rules that we're not allowed to modify, so report that. */
4018 error = OFPERR_OFPBRC_EPERM;
4019 }
4020 if (error) {
4021 rule_collection_destroy(rules);
4022 }
4023 return error;
4024 }
4025
4026 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4027 * on classifiers rules are done in the "strict" way required for OpenFlow
4028 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
4029 * rules on list 'rules'.
4030 *
4031 * Returns 0 on success, otherwise an OpenFlow error code. */
4032 static enum ofperr
4033 collect_rules_strict(struct ofproto *ofproto,
4034 const struct rule_criteria *criteria,
4035 struct rule_collection *rules)
4036 OVS_REQUIRES(ofproto_mutex)
4037 {
4038 struct oftable *table;
4039 size_t n_readonly = 0;
4040 enum ofperr error = 0;
4041
4042 rule_collection_init(rules);
4043
4044 if (!check_table_id(ofproto, criteria->table_id)) {
4045 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4046 goto exit;
4047 }
4048
4049 if (criteria->cookie_mask == OVS_BE64_MAX) {
4050 struct rule *rule;
4051
4052 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4053 hash_cookie(criteria->cookie),
4054 &ofproto->cookies) {
4055 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
4056 collect_rule(rule, criteria, rules, &n_readonly);
4057 }
4058 }
4059 } else {
4060 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4061 struct rule *rule;
4062
4063 rule = rule_from_cls_rule(classifier_find_rule_exactly(
4064 &table->cls, &criteria->cr,
4065 criteria->version));
4066 if (rule) {
4067 collect_rule(rule, criteria, rules, &n_readonly);
4068 }
4069 }
4070 }
4071
4072 exit:
4073 if (!error && !rules->n && n_readonly) {
4074 /* We didn't find any rules to modify. We did find some read-only
4075 * rules that we're not allowed to modify, so report that. */
4076 error = OFPERR_OFPBRC_EPERM;
4077 }
4078 if (error) {
4079 rule_collection_destroy(rules);
4080 }
4081 return error;
4082 }
4083
4084 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
4085 * forced into the range of a uint16_t. */
4086 static int
4087 age_secs(long long int age_ms)
4088 {
4089 return (age_ms < 0 ? 0
4090 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
4091 : (unsigned int) age_ms / 1000);
4092 }
4093
4094 static enum ofperr
4095 handle_flow_stats_request(struct ofconn *ofconn,
4096 const struct ofp_header *request)
4097 OVS_EXCLUDED(ofproto_mutex)
4098 {
4099 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4100 struct ofputil_flow_stats_request fsr;
4101 struct rule_criteria criteria;
4102 struct rule_collection rules;
4103 struct ovs_list replies;
4104 enum ofperr error;
4105 size_t i;
4106
4107 error = ofputil_decode_flow_stats_request(&fsr, request);
4108 if (error) {
4109 return error;
4110 }
4111
4112 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, CLS_MAX_VERSION,
4113 fsr.cookie, fsr.cookie_mask, fsr.out_port,
4114 fsr.out_group);
4115
4116 ovs_mutex_lock(&ofproto_mutex);
4117 error = collect_rules_loose(ofproto, &criteria, &rules);
4118 rule_criteria_destroy(&criteria);
4119 if (!error) {
4120 rule_collection_ref(&rules);
4121 }
4122 ovs_mutex_unlock(&ofproto_mutex);
4123
4124 if (error) {
4125 return error;
4126 }
4127
4128 ofpmp_init(&replies, request);
4129 for (i = 0; i < rules.n; i++) {
4130 struct rule *rule = rules.rules[i];
4131 long long int now = time_msec();
4132 struct ofputil_flow_stats fs;
4133 long long int created, used, modified;
4134 const struct rule_actions *actions;
4135 enum ofputil_flow_mod_flags flags;
4136
4137 ovs_mutex_lock(&rule->mutex);
4138 fs.cookie = rule->flow_cookie;
4139 fs.idle_timeout = rule->idle_timeout;
4140 fs.hard_timeout = rule->hard_timeout;
4141 fs.importance = rule->importance;
4142 created = rule->created;
4143 modified = rule->modified;
4144 actions = rule_get_actions(rule);
4145 flags = rule->flags;
4146 ovs_mutex_unlock(&rule->mutex);
4147
4148 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
4149 &fs.byte_count, &used);
4150
4151 minimatch_expand(&rule->cr.match, &fs.match);
4152 fs.table_id = rule->table_id;
4153 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
4154 fs.priority = rule->cr.priority;
4155 fs.idle_age = age_secs(now - used);
4156 fs.hard_age = age_secs(now - modified);
4157 fs.ofpacts = actions->ofpacts;
4158 fs.ofpacts_len = actions->ofpacts_len;
4159
4160 fs.flags = flags;
4161 ofputil_append_flow_stats_reply(&fs, &replies);
4162 }
4163
4164 rule_collection_unref(&rules);
4165 rule_collection_destroy(&rules);
4166
4167 ofconn_send_replies(ofconn, &replies);
4168
4169 return 0;
4170 }
4171
4172 static void
4173 flow_stats_ds(struct rule *rule, struct ds *results)
4174 {
4175 uint64_t packet_count, byte_count;
4176 const struct rule_actions *actions;
4177 long long int created, used;
4178
4179 rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4180 &byte_count, &used);
4181
4182 ovs_mutex_lock(&rule->mutex);
4183 actions = rule_get_actions(rule);
4184 created = rule->created;
4185 ovs_mutex_unlock(&rule->mutex);
4186
4187 if (rule->table_id != 0) {
4188 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
4189 }
4190 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
4191 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
4192 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
4193 cls_rule_format(&rule->cr, results);
4194 ds_put_char(results, ',');
4195
4196 ds_put_cstr(results, "actions=");
4197 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
4198
4199 ds_put_cstr(results, "\n");
4200 }
4201
4202 /* Adds a pretty-printed description of all flows to 'results', including
4203 * hidden flows (e.g., set up by in-band control). */
4204 void
4205 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
4206 {
4207 struct oftable *table;
4208
4209 OFPROTO_FOR_EACH_TABLE (table, p) {
4210 struct rule *rule;
4211
4212 CLS_FOR_EACH (rule, cr, &table->cls) {
4213 flow_stats_ds(rule, results);
4214 }
4215 }
4216 }
4217
4218 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
4219 * '*engine_type' and '*engine_id', respectively. */
4220 void
4221 ofproto_get_netflow_ids(const struct ofproto *ofproto,
4222 uint8_t *engine_type, uint8_t *engine_id)
4223 {
4224 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
4225 }
4226
4227 /* Checks the status change of CFM on 'ofport'.
4228 *
4229 * Returns true if 'ofproto_class' does not support 'cfm_status_changed'. */
4230 bool
4231 ofproto_port_cfm_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
4232 {
4233 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4234 return (ofport && ofproto->ofproto_class->cfm_status_changed
4235 ? ofproto->ofproto_class->cfm_status_changed(ofport)
4236 : true);
4237 }
4238
4239 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.
4240 * Returns 0 if the port's CFM status was successfully stored into
4241 * '*status'. Returns positive errno if the port did not have CFM
4242 * configured.
4243 *
4244 * The caller must provide and own '*status', and must free 'status->rmps'.
4245 * '*status' is indeterminate if the return value is non-zero. */
4246 int
4247 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
4248 struct cfm_status *status)
4249 {
4250 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4251 return (ofport && ofproto->ofproto_class->get_cfm_status
4252 ? ofproto->ofproto_class->get_cfm_status(ofport, status)
4253 : EOPNOTSUPP);
4254 }
4255
4256 static enum ofperr
4257 handle_aggregate_stats_request(struct ofconn *ofconn,
4258 const struct ofp_header *oh)
4259 OVS_EXCLUDED(ofproto_mutex)
4260 {
4261 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4262 struct ofputil_flow_stats_request request;
4263 struct ofputil_aggregate_stats stats;
4264 bool unknown_packets, unknown_bytes;
4265 struct rule_criteria criteria;
4266 struct rule_collection rules;
4267 struct ofpbuf *reply;
4268 enum ofperr error;
4269 size_t i;
4270
4271 error = ofputil_decode_flow_stats_request(&request, oh);
4272 if (error) {
4273 return error;
4274 }
4275
4276 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
4277 CLS_MAX_VERSION, request.cookie, request.cookie_mask,
4278 request.out_port, request.out_group);
4279
4280 ovs_mutex_lock(&ofproto_mutex);
4281 error = collect_rules_loose(ofproto, &criteria, &rules);
4282 rule_criteria_destroy(&criteria);
4283 if (!error) {
4284 rule_collection_ref(&rules);
4285 }
4286 ovs_mutex_unlock(&ofproto_mutex);
4287
4288 if (error) {
4289 return error;
4290 }
4291
4292 memset(&stats, 0, sizeof stats);
4293 unknown_packets = unknown_bytes = false;
4294 for (i = 0; i < rules.n; i++) {
4295 struct rule *rule = rules.rules[i];
4296 uint64_t packet_count;
4297 uint64_t byte_count;
4298 long long int used;
4299
4300 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4301 &byte_count, &used);
4302
4303 if (packet_count == UINT64_MAX) {
4304 unknown_packets = true;
4305 } else {
4306 stats.packet_count += packet_count;
4307 }
4308
4309 if (byte_count == UINT64_MAX) {
4310 unknown_bytes = true;
4311 } else {
4312 stats.byte_count += byte_count;
4313 }
4314
4315 stats.flow_count++;
4316 }
4317 if (unknown_packets) {
4318 stats.packet_count = UINT64_MAX;
4319 }
4320 if (unknown_bytes) {
4321 stats.byte_count = UINT64_MAX;
4322 }
4323
4324 rule_collection_unref(&rules);
4325 rule_collection_destroy(&rules);
4326
4327 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
4328 ofconn_send_reply(ofconn, reply);
4329
4330 return 0;
4331 }
4332
4333 struct queue_stats_cbdata {
4334 struct ofport *ofport;
4335 struct ovs_list replies;
4336 long long int now;
4337 };
4338
4339 static void
4340 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
4341 const struct netdev_queue_stats *stats)
4342 {
4343 struct ofputil_queue_stats oqs;
4344
4345 oqs.port_no = cbdata->ofport->pp.port_no;
4346 oqs.queue_id = queue_id;
4347 oqs.tx_bytes = stats->tx_bytes;
4348 oqs.tx_packets = stats->tx_packets;
4349 oqs.tx_errors = stats->tx_errors;
4350 if (stats->created != LLONG_MIN) {
4351 calc_duration(stats->created, cbdata->now,
4352 &oqs.duration_sec, &oqs.duration_nsec);
4353 } else {
4354 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
4355 }
4356 ofputil_append_queue_stat(&cbdata->replies, &oqs);
4357 }
4358
4359 static void
4360 handle_queue_stats_dump_cb(uint32_t queue_id,
4361 struct netdev_queue_stats *stats,
4362 void *cbdata_)
4363 {
4364 struct queue_stats_cbdata *cbdata = cbdata_;
4365
4366 put_queue_stats(cbdata, queue_id, stats);
4367 }
4368
4369 static enum ofperr
4370 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
4371 struct queue_stats_cbdata *cbdata)
4372 {
4373 cbdata->ofport = port;
4374 if (queue_id == OFPQ_ALL) {
4375 netdev_dump_queue_stats(port->netdev,
4376 handle_queue_stats_dump_cb, cbdata);
4377 } else {
4378 struct netdev_queue_stats stats;
4379
4380 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
4381 put_queue_stats(cbdata, queue_id, &stats);
4382 } else {
4383 return OFPERR_OFPQOFC_BAD_QUEUE;
4384 }
4385 }
4386 return 0;
4387 }
4388
4389 static enum ofperr
4390 handle_queue_stats_request(struct ofconn *ofconn,
4391 const struct ofp_header *rq)
4392 {
4393 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4394 struct queue_stats_cbdata cbdata;
4395 struct ofport *port;
4396 enum ofperr error;
4397 struct ofputil_queue_stats_request oqsr;
4398
4399 COVERAGE_INC(ofproto_queue_req);
4400
4401 ofpmp_init(&cbdata.replies, rq);
4402 cbdata.now = time_msec();
4403
4404 error = ofputil_decode_queue_stats_request(rq, &oqsr);
4405 if (error) {
4406 return error;
4407 }
4408
4409 if (oqsr.port_no == OFPP_ANY) {
4410 error = OFPERR_OFPQOFC_BAD_QUEUE;
4411 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
4412 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
4413 error = 0;
4414 }
4415 }
4416 } else {
4417 port = ofproto_get_port(ofproto, oqsr.port_no);
4418 error = (port
4419 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
4420 : OFPERR_OFPQOFC_BAD_PORT);
4421 }
4422 if (!error) {
4423 ofconn_send_replies(ofconn, &cbdata.replies);
4424 } else {
4425 ofpbuf_list_delete(&cbdata.replies);
4426 }
4427
4428 return error;
4429 }
4430
4431 static enum ofperr
4432 evict_rules_from_table(struct oftable *table)
4433 OVS_REQUIRES(ofproto_mutex)
4434 {
4435 enum ofperr error = 0;
4436 struct rule_collection rules;
4437 unsigned int count = table->n_flows;
4438 unsigned int max_flows = table->max_flows;
4439
4440 rule_collection_init(&rules);
4441
4442 while (count-- > max_flows) {
4443 struct rule *rule;
4444
4445 if (!choose_rule_to_evict(table, &rule)) {
4446 error = OFPERR_OFPFMFC_TABLE_FULL;
4447 break;
4448 } else {
4449 eviction_group_remove_rule(rule);
4450 rule_collection_add(&rules, rule);
4451 }
4452 }
4453 delete_flows__(&rules, OFPRR_EVICTION, NULL);
4454
4455 return error;
4456 }
4457
4458 static void
4459 get_conjunctions(const struct ofputil_flow_mod *fm,
4460 struct cls_conjunction **conjsp, size_t *n_conjsp)
4461 OVS_REQUIRES(ofproto_mutex)
4462 {
4463 struct cls_conjunction *conjs = NULL;
4464 int n_conjs = 0;
4465
4466 const struct ofpact *ofpact;
4467 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4468 if (ofpact->type == OFPACT_CONJUNCTION) {
4469 n_conjs++;
4470 } else if (ofpact->type != OFPACT_NOTE) {
4471 /* "conjunction" may appear with "note" actions but not with any
4472 * other type of actions. */
4473 ovs_assert(!n_conjs);
4474 break;
4475 }
4476 }
4477 if (n_conjs) {
4478 int i = 0;
4479
4480 conjs = xzalloc(n_conjs * sizeof *conjs);
4481 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4482 if (ofpact->type == OFPACT_CONJUNCTION) {
4483 struct ofpact_conjunction *oc = ofpact_get_CONJUNCTION(ofpact);
4484 conjs[i].clause = oc->clause;
4485 conjs[i].n_clauses = oc->n_clauses;
4486 conjs[i].id = oc->id;
4487 i++;
4488 }
4489 }
4490 }
4491
4492 *conjsp = conjs;
4493 *n_conjsp = n_conjs;
4494 }
4495
4496 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4497 * in which no matching flow already exists in the flow table.
4498 *
4499 * Adds the flow specified by 'fm', to the ofproto's flow table. Returns 0 on
4500 * success, or an OpenFlow error code on failure.
4501 *
4502 * On successful return the caller must complete the operation either by
4503 * calling add_flow_finish(), or add_flow_revert() if the operation needs to
4504 * be reverted.
4505 *
4506 * The caller retains ownership of 'fm->ofpacts'. */
4507 static enum ofperr
4508 add_flow_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4509 OVS_REQUIRES(ofproto_mutex)
4510 {
4511 struct ofputil_flow_mod *fm = &ofm->fm;
4512 struct rule **old_rule = &ofm->old_rules.stub[0];
4513 struct rule **new_rule = &ofm->new_rules.stub[0];
4514 struct oftable *table;
4515 struct cls_rule cr;
4516 struct rule *rule;
4517 uint8_t table_id;
4518 struct cls_conjunction *conjs;
4519 size_t n_conjs;
4520 enum ofperr error;
4521
4522 if (!check_table_id(ofproto, fm->table_id)) {
4523 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4524 return error;
4525 }
4526
4527 /* Pick table. */
4528 if (fm->table_id == 0xff) {
4529 if (ofproto->ofproto_class->rule_choose_table) {
4530 error = ofproto->ofproto_class->rule_choose_table(ofproto,
4531 &fm->match,
4532 &table_id);
4533 if (error) {
4534 return error;
4535 }
4536 ovs_assert(table_id < ofproto->n_tables);
4537 } else {
4538 table_id = 0;
4539 }
4540 } else if (fm->table_id < ofproto->n_tables) {
4541 table_id = fm->table_id;
4542 } else {
4543 return OFPERR_OFPBRC_BAD_TABLE_ID;
4544 }
4545
4546 table = &ofproto->tables[table_id];
4547 if (table->flags & OFTABLE_READONLY
4548 && !(fm->flags & OFPUTIL_FF_NO_READONLY)) {
4549 return OFPERR_OFPBRC_EPERM;
4550 }
4551
4552 if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)
4553 && !match_has_default_hidden_fields(&fm->match)) {
4554 VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set "
4555 "non-default values to hidden fields", ofproto->name);
4556 return OFPERR_OFPBRC_EPERM;
4557 }
4558
4559 cls_rule_init(&cr, &fm->match, fm->priority);
4560
4561 /* Check for the existence of an identical rule.
4562 * This will not return rules earlier marked for removal. */
4563 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr,
4564 ofm->version));
4565 *old_rule = rule;
4566 if (!rule) {
4567 /* Check for overlap, if requested. */
4568 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP
4569 && classifier_rule_overlaps(&table->cls, &cr, ofm->version)) {
4570 cls_rule_destroy(&cr);
4571 return OFPERR_OFPFMFC_OVERLAP;
4572 }
4573
4574 /* If necessary, evict an existing rule to clear out space. */
4575 if (table->n_flows >= table->max_flows) {
4576 if (!choose_rule_to_evict(table, &rule)) {
4577 error = OFPERR_OFPFMFC_TABLE_FULL;
4578 cls_rule_destroy(&cr);
4579 return error;
4580 }
4581 eviction_group_remove_rule(rule);
4582 /* Marks '*old_rule' as an evicted rule rather than replaced rule.
4583 */
4584 fm->delete_reason = OFPRR_EVICTION;
4585 *old_rule = rule;
4586 }
4587 } else {
4588 fm->modify_cookie = true;
4589 }
4590
4591 /* Allocate new rule. */
4592 error = replace_rule_create(ofproto, fm, &cr, table - ofproto->tables,
4593 rule, new_rule);
4594 if (error) {
4595 return error;
4596 }
4597
4598 get_conjunctions(fm, &conjs, &n_conjs);
4599 replace_rule_start(ofproto, ofm->version, rule, *new_rule, conjs, n_conjs);
4600 free(conjs);
4601
4602 return 0;
4603 }
4604
4605 /* Revert the effects of add_flow_start(). */
4606 static void
4607 add_flow_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4608 OVS_REQUIRES(ofproto_mutex)
4609 {
4610 struct ofputil_flow_mod *fm = &ofm->fm;
4611 struct rule *old_rule = ofm->old_rules.stub[0];
4612 struct rule *new_rule = ofm->new_rules.stub[0];
4613
4614 if (old_rule && fm->delete_reason == OFPRR_EVICTION) {
4615 /* Revert the eviction. */
4616 eviction_group_add_rule(old_rule);
4617 }
4618
4619 replace_rule_revert(ofproto, old_rule, new_rule);
4620 }
4621
4622 /* To be called after version bump. */
4623 static void
4624 add_flow_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
4625 const struct flow_mod_requester *req)
4626 OVS_REQUIRES(ofproto_mutex)
4627 {
4628 struct ofputil_flow_mod *fm = &ofm->fm;
4629 struct rule *old_rule = ofm->old_rules.stub[0];
4630 struct rule *new_rule = ofm->new_rules.stub[0];
4631 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
4632
4633 replace_rule_finish(ofproto, fm, req, old_rule, new_rule, &dead_cookies);
4634 learned_cookies_flush(ofproto, &dead_cookies);
4635
4636 if (old_rule) {
4637 ovsrcu_postpone(remove_rule_rcu, old_rule);
4638 } else {
4639 if (minimask_get_vid_mask(new_rule->cr.match.mask) == VLAN_VID_MASK) {
4640 if (ofproto->vlan_bitmap) {
4641 uint16_t vid = miniflow_get_vid(new_rule->cr.match.flow);
4642
4643 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4644 bitmap_set1(ofproto->vlan_bitmap, vid);
4645 ofproto->vlans_changed = true;
4646 }
4647 } else {
4648 ofproto->vlans_changed = true;
4649 }
4650 }
4651
4652 ofmonitor_report(ofproto->connmgr, new_rule, NXFME_ADDED, 0,
4653 req ? req->ofconn : NULL,
4654 req ? req->request->xid : 0, NULL);
4655 }
4656
4657 send_buffered_packet(req, fm->buffer_id, new_rule);
4658 }
4659 \f
4660 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4661
4662 /* Create a new rule based on attributes in 'fm', match in 'cr', 'table_id',
4663 * and 'old_rule'. Note that the rule is NOT inserted into a any data
4664 * structures yet. Takes ownership of 'cr'. */
4665 static enum ofperr
4666 replace_rule_create(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4667 struct cls_rule *cr, uint8_t table_id,
4668 struct rule *old_rule, struct rule **new_rule)
4669 {
4670 struct rule *rule;
4671 enum ofperr error;
4672
4673 /* Allocate new rule. */
4674 rule = ofproto->ofproto_class->rule_alloc();
4675 if (!rule) {
4676 cls_rule_destroy(cr);
4677 VLOG_WARN_RL(&rl, "%s: failed to allocate a rule.", ofproto->name);
4678 return OFPERR_OFPFMFC_UNKNOWN;
4679 }
4680
4681 /* Initialize base state. */
4682 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4683 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), cr);
4684 ovs_refcount_init(&rule->ref_count);
4685 rule->flow_cookie = fm->new_cookie;
4686 rule->created = rule->modified = time_msec();
4687
4688 ovs_mutex_init(&rule->mutex);
4689 ovs_mutex_lock(&rule->mutex);
4690 rule->idle_timeout = fm->idle_timeout;
4691 rule->hard_timeout = fm->hard_timeout;
4692 *CONST_CAST(uint16_t *, &rule->importance) = fm->importance;
4693 rule->removed_reason = OVS_OFPRR_NONE;
4694
4695 *CONST_CAST(uint8_t *, &rule->table_id) = table_id;
4696 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4697 *CONST_CAST(const struct rule_actions **, &rule->actions)
4698 = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
4699 list_init(&rule->meter_list_node);
4700 rule->eviction_group = NULL;
4701 list_init(&rule->expirable);
4702 rule->monitor_flags = 0;
4703 rule->add_seqno = 0;
4704 rule->modify_seqno = 0;
4705
4706 /* Copy values from old rule for modify semantics. */
4707 if (old_rule && fm->delete_reason != OFPRR_EVICTION) {
4708 bool change_cookie = (fm->modify_cookie
4709 && fm->new_cookie != OVS_BE64_MAX
4710 && fm->new_cookie != old_rule->flow_cookie);
4711
4712 ovs_mutex_lock(&old_rule->mutex);
4713 if (fm->command != OFPFC_ADD) {
4714 rule->idle_timeout = old_rule->idle_timeout;
4715 rule->hard_timeout = old_rule->hard_timeout;
4716 *CONST_CAST(uint16_t *, &rule->importance) = old_rule->importance;
4717 rule->flags = old_rule->flags;
4718 rule->created = old_rule->created;
4719 }
4720 if (!change_cookie) {
4721 rule->flow_cookie = old_rule->flow_cookie;
4722 }
4723 ovs_mutex_unlock(&old_rule->mutex);
4724 }
4725 ovs_mutex_unlock(&rule->mutex);
4726
4727 /* Construct rule, initializing derived state. */
4728 error = ofproto->ofproto_class->rule_construct(rule);
4729 if (error) {
4730 ofproto_rule_destroy__(rule);
4731 return error;
4732 }
4733
4734 rule->removed = true; /* Not yet in ofproto data structures. */
4735
4736 *new_rule = rule;
4737 return 0;
4738 }
4739
4740 static void
4741 replace_rule_start(struct ofproto *ofproto, cls_version_t version,
4742 struct rule *old_rule, struct rule *new_rule,
4743 struct cls_conjunction *conjs, size_t n_conjs)
4744 {
4745 struct oftable *table = &ofproto->tables[new_rule->table_id];
4746
4747 /* 'old_rule' may be either an evicted rule or replaced rule. */
4748 if (old_rule) {
4749 /* Mark the old rule for removal in the next version. */
4750 cls_rule_make_invisible_in_version(&old_rule->cr, version);
4751 } else {
4752 table->n_flows++;
4753 }
4754 /* Insert flow to the classifier, so that later flow_mods may relate
4755 * to it. This is reversible, in case later errors require this to
4756 * be reverted. */
4757 ofproto_rule_insert__(ofproto, new_rule);
4758 /* Make the new rule visible for classifier lookups only from the next
4759 * version. */
4760 classifier_insert(&table->cls, &new_rule->cr, version, conjs, n_conjs);
4761 }
4762
4763 static void replace_rule_revert(struct ofproto *ofproto,
4764 struct rule *old_rule, struct rule *new_rule)
4765 {
4766 struct oftable *table = &ofproto->tables[new_rule->table_id];
4767
4768 if (old_rule) {
4769 /* Restore the original visibility of the old rule. */
4770 cls_rule_restore_visibility(&old_rule->cr);
4771 } else {
4772 /* Restore table's rule count. */
4773 table->n_flows--;
4774 }
4775
4776 /* Remove the new rule immediately. It was never visible to lookups. */
4777 if (!classifier_remove(&table->cls, &new_rule->cr)) {
4778 OVS_NOT_REACHED();
4779 }
4780 ofproto_rule_remove__(ofproto, new_rule);
4781 /* The rule was not inserted to the ofproto provider, so we can
4782 * release it without deleting it from the ofproto provider. */
4783 ofproto_rule_unref(new_rule);
4784 }
4785
4786 /* Adds the 'new_rule', replacing the 'old_rule'. */
4787 static void
4788 replace_rule_finish(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4789 const struct flow_mod_requester *req,
4790 struct rule *old_rule, struct rule *new_rule,
4791 struct ovs_list *dead_cookies)
4792 OVS_REQUIRES(ofproto_mutex)
4793 {
4794 bool forward_stats = !(fm->flags & OFPUTIL_FF_RESET_COUNTS);
4795 struct rule *replaced_rule;
4796
4797 replaced_rule = fm->delete_reason != OFPRR_EVICTION ? old_rule : NULL;
4798
4799 /* Insert the new flow to the ofproto provider. A non-NULL 'replaced_rule'
4800 * is a duplicate rule the 'new_rule' is replacing. The provider should
4801 * link the stats from the old rule to the new one if 'forward_stats' is
4802 * 'true'. The 'replaced_rule' will be deleted right after this call. */
4803 ofproto->ofproto_class->rule_insert(new_rule, replaced_rule,
4804 forward_stats);
4805 learned_cookies_inc(ofproto, rule_get_actions(new_rule));
4806
4807 if (old_rule) {
4808 const struct rule_actions *old_actions = rule_get_actions(old_rule);
4809
4810 /* Remove the old rule from data structures. Removal from the
4811 * classifier and the deletion of the rule is RCU postponed by the
4812 * caller. */
4813 ofproto_rule_remove__(ofproto, old_rule);
4814 learned_cookies_dec(ofproto, old_actions, dead_cookies);
4815
4816 if (replaced_rule) {
4817 enum nx_flow_update_event event = fm->command == OFPFC_ADD
4818 ? NXFME_ADDED : NXFME_MODIFIED;
4819
4820 bool change_cookie = (fm->modify_cookie
4821 && fm->new_cookie != OVS_BE64_MAX
4822 && fm->new_cookie != old_rule->flow_cookie);
4823
4824 bool change_actions = !ofpacts_equal(fm->ofpacts,
4825 fm->ofpacts_len,
4826 old_actions->ofpacts,
4827 old_actions->ofpacts_len);
4828
4829 if (event != NXFME_MODIFIED || change_actions || change_cookie) {
4830 ofmonitor_report(ofproto->connmgr, new_rule, event, 0,
4831 req ? req->ofconn : NULL,
4832 req ? req->request->xid : 0,
4833 change_actions ? old_actions : NULL);
4834 }
4835 } else {
4836 /* XXX: This is slight duplication with delete_flows_finish__() */
4837
4838 old_rule->removed_reason = OFPRR_EVICTION;
4839
4840 ofmonitor_report(ofproto->connmgr, old_rule, NXFME_DELETED,
4841 OFPRR_EVICTION,
4842 req ? req->ofconn : NULL,
4843 req ? req->request->xid : 0, NULL);
4844 }
4845 }
4846 }
4847
4848 static enum ofperr
4849 modify_flows_start__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4850 OVS_REQUIRES(ofproto_mutex)
4851 {
4852 struct ofputil_flow_mod *fm = &ofm->fm;
4853 struct rule_collection *old_rules = &ofm->old_rules;
4854 struct rule_collection *new_rules = &ofm->new_rules;
4855 enum ofperr error;
4856
4857 rule_collection_init(new_rules);
4858
4859 if (old_rules->n > 0) {
4860 struct cls_conjunction *conjs;
4861 size_t n_conjs;
4862 size_t i;
4863
4864 /* Create a new 'modified' rule for each old rule. */
4865 for (i = 0; i < old_rules->n; i++) {
4866 struct rule *old_rule = old_rules->rules[i];
4867 struct rule *new_rule;
4868 struct cls_rule cr;
4869
4870 cls_rule_clone(&cr, &old_rule->cr);
4871 error = replace_rule_create(ofproto, fm, &cr, old_rule->table_id,
4872 old_rule, &new_rule);
4873 if (!error) {
4874 rule_collection_add(new_rules, new_rule);
4875 } else {
4876 rule_collection_unref(new_rules);
4877 rule_collection_destroy(new_rules);
4878 return error;
4879 }
4880 }
4881 ovs_assert(new_rules->n == old_rules->n);
4882
4883 get_conjunctions(fm, &conjs, &n_conjs);
4884 for (i = 0; i < old_rules->n; i++) {
4885 replace_rule_start(ofproto, ofm->version, old_rules->rules[i],
4886 new_rules->rules[i], conjs, n_conjs);
4887 }
4888 free(conjs);
4889 } else if (!(fm->cookie_mask != htonll(0)
4890 || fm->new_cookie == OVS_BE64_MAX)) {
4891 /* No match, add a new flow. */
4892 error = add_flow_start(ofproto, ofm);
4893 if (!error) {
4894 ovs_assert(fm->delete_reason == OFPRR_EVICTION
4895 || !old_rules->rules[0]);
4896 }
4897 new_rules->n = 1;
4898 } else {
4899 error = 0;
4900 }
4901
4902 return error;
4903 }
4904
4905 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4906 * failure.
4907 *
4908 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4909 * if any. */
4910 static enum ofperr
4911 modify_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4912 OVS_REQUIRES(ofproto_mutex)
4913 {
4914 struct ofputil_flow_mod *fm = &ofm->fm;
4915 struct rule_collection *old_rules = &ofm->old_rules;
4916 struct rule_criteria criteria;
4917 enum ofperr error;
4918
4919 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
4920 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG11_ANY);
4921 rule_criteria_require_rw(&criteria,
4922 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
4923 error = collect_rules_loose(ofproto, &criteria, old_rules);
4924 rule_criteria_destroy(&criteria);
4925
4926 if (!error) {
4927 error = modify_flows_start__(ofproto, ofm);
4928 }
4929
4930 if (error) {
4931 rule_collection_destroy(old_rules);
4932 }
4933 return error;
4934 }
4935
4936 static void
4937 modify_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4938 OVS_REQUIRES(ofproto_mutex)
4939 {
4940 struct rule_collection *old_rules = &ofm->old_rules;
4941 struct rule_collection *new_rules = &ofm->new_rules;
4942
4943 /* Old rules were not changed yet, only need to revert new rules. */
4944 if (old_rules->n == 0 && new_rules->n == 1) {
4945 add_flow_revert(ofproto, ofm);
4946 } else if (old_rules->n > 0) {
4947 for (size_t i = 0; i < old_rules->n; i++) {
4948 replace_rule_revert(ofproto, old_rules->rules[i],
4949 new_rules->rules[i]);
4950 }
4951 rule_collection_destroy(new_rules);
4952 rule_collection_destroy(old_rules);
4953 }
4954 }
4955
4956 static void
4957 modify_flows_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
4958 const struct flow_mod_requester *req)
4959 OVS_REQUIRES(ofproto_mutex)
4960 {
4961 struct ofputil_flow_mod *fm = &ofm->fm;
4962 struct rule_collection *old_rules = &ofm->old_rules;
4963 struct rule_collection *new_rules = &ofm->new_rules;
4964
4965 if (old_rules->n == 0 && new_rules->n == 1) {
4966 add_flow_finish(ofproto, ofm, req);
4967 } else if (old_rules->n > 0) {
4968 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
4969
4970 ovs_assert(new_rules->n == old_rules->n);
4971
4972 for (size_t i = 0; i < old_rules->n; i++) {
4973 replace_rule_finish(ofproto, fm, req, old_rules->rules[i],
4974 new_rules->rules[i], &dead_cookies);
4975 }
4976 learned_cookies_flush(ofproto, &dead_cookies);
4977 rule_collection_remove_postponed(old_rules);
4978
4979 send_buffered_packet(req, fm->buffer_id, new_rules->rules[0]);
4980 rule_collection_destroy(new_rules);
4981 }
4982 }
4983
4984 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
4985 * code on failure. */
4986 static enum ofperr
4987 modify_flow_start_strict(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4988 OVS_REQUIRES(ofproto_mutex)
4989 {
4990 struct ofputil_flow_mod *fm = &ofm->fm;
4991 struct rule_collection *old_rules = &ofm->old_rules;
4992 struct rule_criteria criteria;
4993 enum ofperr error;
4994
4995 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
4996 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask, OFPP_ANY,
4997 OFPG11_ANY);
4998 rule_criteria_require_rw(&criteria,
4999 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5000 error = collect_rules_strict(ofproto, &criteria, old_rules);
5001 rule_criteria_destroy(&criteria);
5002
5003 if (!error) {
5004 /* collect_rules_strict() can return max 1 rule. */
5005 error = modify_flows_start__(ofproto, ofm);
5006 }
5007
5008 if (error) {
5009 rule_collection_destroy(old_rules);
5010 }
5011 return error;
5012 }
5013 \f
5014 /* OFPFC_DELETE implementation. */
5015
5016 static void
5017 delete_flows_start__(struct ofproto *ofproto, cls_version_t version,
5018 const struct rule_collection *rules)
5019 OVS_REQUIRES(ofproto_mutex)
5020 {
5021 for (size_t i = 0; i < rules->n; i++) {
5022 struct rule *rule = rules->rules[i];
5023 struct oftable *table = &ofproto->tables[rule->table_id];
5024
5025 table->n_flows--;
5026 cls_rule_make_invisible_in_version(&rule->cr, version);
5027 }
5028 }
5029
5030 static void
5031 delete_flows_finish__(struct ofproto *ofproto,
5032 struct rule_collection *rules,
5033 enum ofp_flow_removed_reason reason,
5034 const struct flow_mod_requester *req)
5035 OVS_REQUIRES(ofproto_mutex)
5036 {
5037 if (rules->n) {
5038 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5039
5040 for (size_t i = 0; i < rules->n; i++) {
5041 struct rule *rule = rules->rules[i];
5042
5043 /* This value will be used to send the flow removed message right
5044 * before the rule is actually destroyed. */
5045 rule->removed_reason = reason;
5046
5047 ofmonitor_report(ofproto->connmgr, rule, NXFME_DELETED, reason,
5048 req ? req->ofconn : NULL,
5049 req ? req->request->xid : 0, NULL);
5050 ofproto_rule_remove__(ofproto, rule);
5051 learned_cookies_dec(ofproto, rule_get_actions(rule),
5052 &dead_cookies);
5053 }
5054 rule_collection_remove_postponed(rules);
5055
5056 learned_cookies_flush(ofproto, &dead_cookies);
5057 }
5058 }
5059
5060 /* Deletes the rules listed in 'rules'.
5061 * The deleted rules will become invisible to the lookups in the next version.
5062 * Destroys 'rules'. */
5063 static void
5064 delete_flows__(struct rule_collection *rules,
5065 enum ofp_flow_removed_reason reason,
5066 const struct flow_mod_requester *req)
5067 OVS_REQUIRES(ofproto_mutex)
5068 {
5069 if (rules->n) {
5070 struct ofproto *ofproto = rules->rules[0]->ofproto;
5071
5072 delete_flows_start__(ofproto, ofproto->tables_version + 1, rules);
5073 ofproto_bump_tables_version(ofproto);
5074 delete_flows_finish__(ofproto, rules, reason, req);
5075 ofmonitor_flush(ofproto->connmgr);
5076 }
5077 }
5078
5079 /* Implements OFPFC_DELETE. */
5080 static enum ofperr
5081 delete_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5082 OVS_REQUIRES(ofproto_mutex)
5083 {
5084 const struct ofputil_flow_mod *fm = &ofm->fm;
5085 struct rule_collection *rules = &ofm->old_rules;
5086 struct rule_criteria criteria;
5087 enum ofperr error;
5088
5089 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
5090 fm->cookie, fm->cookie_mask, fm->out_port,
5091 fm->out_group);
5092 rule_criteria_require_rw(&criteria,
5093 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5094 error = collect_rules_loose(ofproto, &criteria, rules);
5095 rule_criteria_destroy(&criteria);
5096
5097 if (!error) {
5098 delete_flows_start__(ofproto, ofm->version, rules);
5099 }
5100
5101 return error;
5102 }
5103
5104 static void
5105 delete_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5106 OVS_REQUIRES(ofproto_mutex)
5107 {
5108 struct rule_collection *rules = &ofm->old_rules;
5109
5110 for (size_t i = 0; i < rules->n; i++) {
5111 struct rule *rule = rules->rules[i];
5112 struct oftable *table = &ofproto->tables[rule->table_id];
5113
5114 /* Restore table's rule count. */
5115 table->n_flows++;
5116
5117 /* Restore the original visibility of the rule. */
5118 cls_rule_restore_visibility(&rule->cr);
5119 }
5120 rule_collection_destroy(rules);
5121 }
5122
5123 static void
5124 delete_flows_finish(struct ofproto *ofproto,
5125 struct ofproto_flow_mod *ofm,
5126 const struct flow_mod_requester *req)
5127 OVS_REQUIRES(ofproto_mutex)
5128 {
5129 delete_flows_finish__(ofproto, &ofm->old_rules, ofm->fm.delete_reason,
5130 req);
5131 }
5132
5133 /* Implements OFPFC_DELETE_STRICT. */
5134 static enum ofperr
5135 delete_flow_start_strict(struct ofproto *ofproto,
5136 struct ofproto_flow_mod *ofm)
5137 OVS_REQUIRES(ofproto_mutex)
5138 {
5139 const struct ofputil_flow_mod *fm = &ofm->fm;
5140 struct rule_collection *rules = &ofm->old_rules;
5141 struct rule_criteria criteria;
5142 enum ofperr error;
5143
5144 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5145 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask,
5146 fm->out_port, fm->out_group);
5147 rule_criteria_require_rw(&criteria,
5148 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5149 error = collect_rules_strict(ofproto, &criteria, rules);
5150 rule_criteria_destroy(&criteria);
5151
5152 if (!error) {
5153 delete_flows_start__(ofproto, ofm->version, rules);
5154 }
5155
5156 return error;
5157 }
5158
5159 /* This may only be called by rule_destroy_cb()! */
5160 static void
5161 ofproto_rule_send_removed(struct rule *rule)
5162 OVS_EXCLUDED(ofproto_mutex)
5163 {
5164 struct ofputil_flow_removed fr;
5165 long long int used;
5166
5167 minimatch_expand(&rule->cr.match, &fr.match);
5168 fr.priority = rule->cr.priority;
5169
5170 ovs_mutex_lock(&ofproto_mutex);
5171 fr.cookie = rule->flow_cookie;
5172 fr.reason = rule->removed_reason;
5173 fr.table_id = rule->table_id;
5174 calc_duration(rule->created, time_msec(),
5175 &fr.duration_sec, &fr.duration_nsec);
5176 ovs_mutex_lock(&rule->mutex);
5177 fr.idle_timeout = rule->idle_timeout;
5178 fr.hard_timeout = rule->hard_timeout;
5179 ovs_mutex_unlock(&rule->mutex);
5180 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
5181 &fr.byte_count, &used);
5182 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
5183 ovs_mutex_unlock(&ofproto_mutex);
5184 }
5185
5186 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
5187 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
5188 * ofproto.
5189 *
5190 * ofproto implementation ->run() functions should use this function to expire
5191 * OpenFlow flows. */
5192 void
5193 ofproto_rule_expire(struct rule *rule, uint8_t reason)
5194 OVS_REQUIRES(ofproto_mutex)
5195 {
5196 struct rule_collection rules;
5197
5198 rules.rules = rules.stub;
5199 rules.n = 1;
5200 rules.stub[0] = rule;
5201 delete_flows__(&rules, reason, NULL);
5202 }
5203
5204 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5205 * means "infinite". */
5206 static void
5207 reduce_timeout(uint16_t max, uint16_t *timeout)
5208 {
5209 if (max && (!*timeout || *timeout > max)) {
5210 *timeout = max;
5211 }
5212 }
5213
5214 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
5215 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
5216 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
5217 *
5218 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
5219 void
5220 ofproto_rule_reduce_timeouts(struct rule *rule,
5221 uint16_t idle_timeout, uint16_t hard_timeout)
5222 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
5223 {
5224 if (!idle_timeout && !hard_timeout) {
5225 return;
5226 }
5227
5228 ovs_mutex_lock(&ofproto_mutex);
5229 if (list_is_empty(&rule->expirable)) {
5230 list_insert(&rule->ofproto->expirable, &rule->expirable);
5231 }
5232 ovs_mutex_unlock(&ofproto_mutex);
5233
5234 ovs_mutex_lock(&rule->mutex);
5235 reduce_timeout(idle_timeout, &rule->idle_timeout);
5236 reduce_timeout(hard_timeout, &rule->hard_timeout);
5237 ovs_mutex_unlock(&rule->mutex);
5238 }
5239 \f
5240 static enum ofperr
5241 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5242 OVS_EXCLUDED(ofproto_mutex)
5243 {
5244 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5245 struct ofproto_flow_mod ofm;
5246 uint64_t ofpacts_stub[1024 / 8];
5247 struct ofpbuf ofpacts;
5248 enum ofperr error;
5249
5250 error = reject_slave_controller(ofconn);
5251 if (error) {
5252 goto exit;
5253 }
5254
5255 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5256 error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn),
5257 &ofpacts,
5258 u16_to_ofp(ofproto->max_ports),
5259 ofproto->n_tables);
5260 if (!error) {
5261 error = ofproto_check_ofpacts(ofproto, ofm.fm.ofpacts,
5262 ofm.fm.ofpacts_len);
5263 }
5264 if (!error) {
5265 struct flow_mod_requester req;
5266
5267 req.ofconn = ofconn;
5268 req.request = oh;
5269 error = handle_flow_mod__(ofproto, &ofm, &req);
5270 }
5271 if (error) {
5272 goto exit_free_ofpacts;
5273 }
5274
5275 ofconn_report_flow_mod(ofconn, ofm.fm.command);
5276
5277 exit_free_ofpacts:
5278 ofpbuf_uninit(&ofpacts);
5279 exit:
5280 return error;
5281 }
5282
5283 static enum ofperr
5284 handle_flow_mod__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5285 const struct flow_mod_requester *req)
5286 OVS_EXCLUDED(ofproto_mutex)
5287 {
5288 enum ofperr error;
5289
5290 ovs_mutex_lock(&ofproto_mutex);
5291 ofm->version = ofproto->tables_version + 1;
5292 error = ofproto_flow_mod_start(ofproto, ofm);
5293 if (!error) {
5294 ofproto_bump_tables_version(ofproto);
5295 ofproto_flow_mod_finish(ofproto, ofm, req);
5296 }
5297 ofmonitor_flush(ofproto->connmgr);
5298 ovs_mutex_unlock(&ofproto_mutex);
5299
5300 run_rule_executes(ofproto);
5301 return error;
5302 }
5303
5304 static enum ofperr
5305 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
5306 {
5307 struct ofputil_role_request request;
5308 struct ofputil_role_request reply;
5309 struct ofpbuf *buf;
5310 enum ofperr error;
5311
5312 error = ofputil_decode_role_message(oh, &request);
5313 if (error) {
5314 return error;
5315 }
5316
5317 if (request.role != OFPCR12_ROLE_NOCHANGE) {
5318 if (request.have_generation_id
5319 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
5320 return OFPERR_OFPRRFC_STALE;
5321 }
5322
5323 ofconn_set_role(ofconn, request.role);
5324 }
5325
5326 reply.role = ofconn_get_role(ofconn);
5327 reply.have_generation_id = ofconn_get_master_election_id(
5328 ofconn, &reply.generation_id);
5329 buf = ofputil_encode_role_reply(oh, &reply);
5330 ofconn_send_reply(ofconn, buf);
5331
5332 return 0;
5333 }
5334
5335 static enum ofperr
5336 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
5337 const struct ofp_header *oh)
5338 {
5339 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
5340 enum ofputil_protocol cur, next;
5341
5342 cur = ofconn_get_protocol(ofconn);
5343 next = ofputil_protocol_set_tid(cur, msg->set != 0);
5344 ofconn_set_protocol(ofconn, next);
5345
5346 return 0;
5347 }
5348
5349 static enum ofperr
5350 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
5351 {
5352 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
5353 enum ofputil_protocol cur, next;
5354 enum ofputil_protocol next_base;
5355
5356 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
5357 if (!next_base) {
5358 return OFPERR_OFPBRC_EPERM;
5359 }
5360
5361 cur = ofconn_get_protocol(ofconn);
5362 next = ofputil_protocol_set_base(cur, next_base);
5363 ofconn_set_protocol(ofconn, next);
5364
5365 return 0;
5366 }
5367
5368 static enum ofperr
5369 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
5370 const struct ofp_header *oh)
5371 {
5372 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
5373 uint32_t format;
5374
5375 format = ntohl(msg->format);
5376 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
5377 return OFPERR_OFPBRC_EPERM;
5378 }
5379
5380 ofconn_set_packet_in_format(ofconn, format);
5381 return 0;
5382 }
5383
5384 static enum ofperr
5385 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
5386 {
5387 const struct nx_async_config *msg = ofpmsg_body(oh);
5388 uint32_t master[OAM_N_TYPES];
5389 uint32_t slave[OAM_N_TYPES];
5390
5391 master[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[0]);
5392 master[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[0]);
5393 master[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[0]);
5394
5395 slave[OAM_PACKET_IN] = ntohl(msg->packet_in_mask[1]);
5396 slave[OAM_PORT_STATUS] = ntohl(msg->port_status_mask[1]);
5397 slave[OAM_FLOW_REMOVED] = ntohl(msg->flow_removed_mask[1]);
5398
5399 ofconn_set_async_config(ofconn, master, slave);
5400 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
5401 !ofconn_get_miss_send_len(ofconn)) {
5402 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
5403 }
5404
5405 return 0;
5406 }
5407
5408 static enum ofperr
5409 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
5410 {
5411 struct ofpbuf *buf;
5412 uint32_t master[OAM_N_TYPES];
5413 uint32_t slave[OAM_N_TYPES];
5414 struct nx_async_config *msg;
5415
5416 ofconn_get_async_config(ofconn, master, slave);
5417 buf = ofpraw_alloc_reply(OFPRAW_OFPT13_GET_ASYNC_REPLY, oh, 0);
5418 msg = ofpbuf_put_zeros(buf, sizeof *msg);
5419
5420 msg->packet_in_mask[0] = htonl(master[OAM_PACKET_IN]);
5421 msg->port_status_mask[0] = htonl(master[OAM_PORT_STATUS]);
5422 msg->flow_removed_mask[0] = htonl(master[OAM_FLOW_REMOVED]);
5423
5424 msg->packet_in_mask[1] = htonl(slave[OAM_PACKET_IN]);
5425 msg->port_status_mask[1] = htonl(slave[OAM_PORT_STATUS]);
5426 msg->flow_removed_mask[1] = htonl(slave[OAM_FLOW_REMOVED]);
5427
5428 ofconn_send_reply(ofconn, buf);
5429
5430 return 0;
5431 }
5432
5433 static enum ofperr
5434 handle_nxt_set_controller_id(struct ofconn *ofconn,
5435 const struct ofp_header *oh)
5436 {
5437 const struct nx_controller_id *nci = ofpmsg_body(oh);
5438
5439 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
5440 return OFPERR_NXBRC_MUST_BE_ZERO;
5441 }
5442
5443 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
5444 return 0;
5445 }
5446
5447 static enum ofperr
5448 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
5449 {
5450 struct ofpbuf *buf;
5451
5452 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
5453 ? OFPRAW_OFPT10_BARRIER_REPLY
5454 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
5455 ofconn_send_reply(ofconn, buf);
5456 return 0;
5457 }
5458
5459 static void
5460 ofproto_compose_flow_refresh_update(const struct rule *rule,
5461 enum nx_flow_monitor_flags flags,
5462 struct ovs_list *msgs)
5463 OVS_REQUIRES(ofproto_mutex)
5464 {
5465 const struct rule_actions *actions;
5466 struct ofputil_flow_update fu;
5467 struct match match;
5468
5469 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
5470 ? NXFME_ADDED : NXFME_MODIFIED);
5471 fu.reason = 0;
5472 ovs_mutex_lock(&rule->mutex);
5473 fu.idle_timeout = rule->idle_timeout;
5474 fu.hard_timeout = rule->hard_timeout;
5475 ovs_mutex_unlock(&rule->mutex);
5476 fu.table_id = rule->table_id;
5477 fu.cookie = rule->flow_cookie;
5478 minimatch_expand(&rule->cr.match, &match);
5479 fu.match = &match;
5480 fu.priority = rule->cr.priority;
5481
5482 actions = flags & NXFMF_ACTIONS ? rule_get_actions(rule) : NULL;
5483 fu.ofpacts = actions ? actions->ofpacts : NULL;
5484 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
5485
5486 if (list_is_empty(msgs)) {
5487 ofputil_start_flow_update(msgs);
5488 }
5489 ofputil_append_flow_update(&fu, msgs);
5490 }
5491
5492 void
5493 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
5494 struct ovs_list *msgs)
5495 OVS_REQUIRES(ofproto_mutex)
5496 {
5497 size_t i;
5498
5499 for (i = 0; i < rules->n; i++) {
5500 struct rule *rule = rules->rules[i];
5501 enum nx_flow_monitor_flags flags = rule->monitor_flags;
5502 rule->monitor_flags = 0;
5503
5504 ofproto_compose_flow_refresh_update(rule, flags, msgs);
5505 }
5506 }
5507
5508 static void
5509 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
5510 struct rule *rule, uint64_t seqno,
5511 struct rule_collection *rules)
5512 OVS_REQUIRES(ofproto_mutex)
5513 {
5514 enum nx_flow_monitor_flags update;
5515
5516 if (rule_is_hidden(rule)) {
5517 return;
5518 }
5519
5520 if (!ofproto_rule_has_out_port(rule, m->out_port)) {
5521 return;
5522 }
5523
5524 if (seqno) {
5525 if (rule->add_seqno > seqno) {
5526 update = NXFMF_ADD | NXFMF_MODIFY;
5527 } else if (rule->modify_seqno > seqno) {
5528 update = NXFMF_MODIFY;
5529 } else {
5530 return;
5531 }
5532
5533 if (!(m->flags & update)) {
5534 return;
5535 }
5536 } else {
5537 update = NXFMF_INITIAL;
5538 }
5539
5540 if (!rule->monitor_flags) {
5541 rule_collection_add(rules, rule);
5542 }
5543 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
5544 }
5545
5546 static void
5547 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
5548 uint64_t seqno,
5549 struct rule_collection *rules)
5550 OVS_REQUIRES(ofproto_mutex)
5551 {
5552 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
5553 const struct oftable *table;
5554 struct cls_rule target;
5555
5556 cls_rule_init_from_minimatch(&target, &m->match, 0);
5557 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
5558 struct rule *rule;
5559
5560 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &target, CLS_MAX_VERSION) {
5561 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
5562 }
5563 }
5564 cls_rule_destroy(&target);
5565 }
5566
5567 static void
5568 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
5569 struct rule_collection *rules)
5570 OVS_REQUIRES(ofproto_mutex)
5571 {
5572 if (m->flags & NXFMF_INITIAL) {
5573 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
5574 }
5575 }
5576
5577 void
5578 ofmonitor_collect_resume_rules(struct ofmonitor *m,
5579 uint64_t seqno, struct rule_collection *rules)
5580 OVS_REQUIRES(ofproto_mutex)
5581 {
5582 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
5583 }
5584
5585 static enum ofperr
5586 flow_monitor_delete(struct ofconn *ofconn, uint32_t id)
5587 OVS_REQUIRES(ofproto_mutex)
5588 {
5589 struct ofmonitor *m;
5590 enum ofperr error;
5591
5592 m = ofmonitor_lookup(ofconn, id);
5593 if (m) {
5594 ofmonitor_destroy(m);
5595 error = 0;
5596 } else {
5597 error = OFPERR_OFPMOFC_UNKNOWN_MONITOR;
5598 }
5599
5600 return error;
5601 }
5602
5603 static enum ofperr
5604 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
5605 OVS_EXCLUDED(ofproto_mutex)
5606 {
5607 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5608 struct ofmonitor **monitors;
5609 size_t n_monitors, allocated_monitors;
5610 struct rule_collection rules;
5611 struct ovs_list replies;
5612 enum ofperr error;
5613 struct ofpbuf b;
5614 size_t i;
5615
5616 ofpbuf_use_const(&b, oh, ntohs(oh->length));
5617 monitors = NULL;
5618 n_monitors = allocated_monitors = 0;
5619
5620 ovs_mutex_lock(&ofproto_mutex);
5621 for (;;) {
5622 struct ofputil_flow_monitor_request request;
5623 struct ofmonitor *m;
5624 int retval;
5625
5626 retval = ofputil_decode_flow_monitor_request(&request, &b);
5627 if (retval == EOF) {
5628 break;
5629 } else if (retval) {
5630 error = retval;
5631 goto error;
5632 }
5633
5634 if (request.table_id != 0xff
5635 && request.table_id >= ofproto->n_tables) {
5636 error = OFPERR_OFPBRC_BAD_TABLE_ID;
5637 goto error;
5638 }
5639
5640 error = ofmonitor_create(&request, ofconn, &m);
5641 if (error) {
5642 goto error;
5643 }
5644
5645 if (n_monitors >= allocated_monitors) {
5646 monitors = x2nrealloc(monitors, &allocated_monitors,
5647 sizeof *monitors);
5648 }
5649 monitors[n_monitors++] = m;
5650 }
5651
5652 rule_collection_init(&rules);
5653 for (i = 0; i < n_monitors; i++) {
5654 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
5655 }
5656
5657 ofpmp_init(&replies, oh);
5658 ofmonitor_compose_refresh_updates(&rules, &replies);
5659 ovs_mutex_unlock(&ofproto_mutex);
5660
5661 rule_collection_destroy(&rules);
5662
5663 ofconn_send_replies(ofconn, &replies);
5664 free(monitors);
5665
5666 return 0;
5667
5668 error:
5669 for (i = 0; i < n_monitors; i++) {
5670 ofmonitor_destroy(monitors[i]);
5671 }
5672 free(monitors);
5673 ovs_mutex_unlock(&ofproto_mutex);
5674
5675 return error;
5676 }
5677
5678 static enum ofperr
5679 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
5680 OVS_EXCLUDED(ofproto_mutex)
5681 {
5682 enum ofperr error;
5683 uint32_t id;
5684
5685 id = ofputil_decode_flow_monitor_cancel(oh);
5686
5687 ovs_mutex_lock(&ofproto_mutex);
5688 error = flow_monitor_delete(ofconn, id);
5689 ovs_mutex_unlock(&ofproto_mutex);
5690
5691 return error;
5692 }
5693
5694 /* Meters implementation.
5695 *
5696 * Meter table entry, indexed by the OpenFlow meter_id.
5697 * 'created' is used to compute the duration for meter stats.
5698 * 'list rules' is needed so that we can delete the dependent rules when the
5699 * meter table entry is deleted.
5700 * 'provider_meter_id' is for the provider's private use.
5701 */
5702 struct meter {
5703 long long int created; /* Time created. */
5704 struct ovs_list rules; /* List of "struct rule_dpif"s. */
5705 ofproto_meter_id provider_meter_id;
5706 uint16_t flags; /* Meter flags. */
5707 uint16_t n_bands; /* Number of meter bands. */
5708 struct ofputil_meter_band *bands;
5709 };
5710
5711 /*
5712 * This is used in instruction validation at flow set-up time,
5713 * as flows may not use non-existing meters.
5714 * Return value of UINT32_MAX signifies an invalid meter.
5715 */
5716 static uint32_t
5717 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
5718 {
5719 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
5720 const struct meter *meter = ofproto->meters[of_meter_id];
5721 if (meter) {
5722 return meter->provider_meter_id.uint32;
5723 }
5724 }
5725 return UINT32_MAX;
5726 }
5727
5728 /* Finds the meter invoked by 'rule''s actions and adds 'rule' to the meter's
5729 * list of rules. */
5730 static void
5731 meter_insert_rule(struct rule *rule)
5732 {
5733 const struct rule_actions *a = rule_get_actions(rule);
5734 uint32_t meter_id = ofpacts_get_meter(a->ofpacts, a->ofpacts_len);
5735 struct meter *meter = rule->ofproto->meters[meter_id];
5736
5737 list_insert(&meter->rules, &rule->meter_list_node);
5738 }
5739
5740 static void
5741 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
5742 {
5743 free(meter->bands);
5744
5745 meter->flags = config->flags;
5746 meter->n_bands = config->n_bands;
5747 meter->bands = xmemdup(config->bands,
5748 config->n_bands * sizeof *meter->bands);
5749 }
5750
5751 static struct meter *
5752 meter_create(const struct ofputil_meter_config *config,
5753 ofproto_meter_id provider_meter_id)
5754 {
5755 struct meter *meter;
5756
5757 meter = xzalloc(sizeof *meter);
5758 meter->provider_meter_id = provider_meter_id;
5759 meter->created = time_msec();
5760 list_init(&meter->rules);
5761
5762 meter_update(meter, config);
5763
5764 return meter;
5765 }
5766
5767 static void
5768 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5769 OVS_REQUIRES(ofproto_mutex)
5770 {
5771 uint32_t mid;
5772 for (mid = first; mid <= last; ++mid) {
5773 struct meter *meter = ofproto->meters[mid];
5774 if (meter) {
5775 ofproto->meters[mid] = NULL;
5776 ofproto->ofproto_class->meter_del(ofproto,
5777 meter->provider_meter_id);
5778 free(meter->bands);
5779 free(meter);
5780 }
5781 }
5782 }
5783
5784 static enum ofperr
5785 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5786 {
5787 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5788 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5789 enum ofperr error;
5790
5791 if (*meterp) {
5792 return OFPERR_OFPMMFC_METER_EXISTS;
5793 }
5794
5795 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5796 &mm->meter);
5797 if (!error) {
5798 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5799 *meterp = meter_create(&mm->meter, provider_meter_id);
5800 }
5801 return error;
5802 }
5803
5804 static enum ofperr
5805 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5806 {
5807 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5808 enum ofperr error;
5809 uint32_t provider_meter_id;
5810
5811 if (!meter) {
5812 return OFPERR_OFPMMFC_UNKNOWN_METER;
5813 }
5814
5815 provider_meter_id = meter->provider_meter_id.uint32;
5816 error = ofproto->ofproto_class->meter_set(ofproto,
5817 &meter->provider_meter_id,
5818 &mm->meter);
5819 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5820 if (!error) {
5821 meter_update(meter, &mm->meter);
5822 }
5823 return error;
5824 }
5825
5826 static enum ofperr
5827 handle_delete_meter(struct ofconn *ofconn, struct ofputil_meter_mod *mm)
5828 OVS_EXCLUDED(ofproto_mutex)
5829 {
5830 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5831 uint32_t meter_id = mm->meter.meter_id;
5832 struct rule_collection rules;
5833 enum ofperr error = 0;
5834 uint32_t first, last;
5835
5836 if (meter_id == OFPM13_ALL) {
5837 first = 1;
5838 last = ofproto->meter_features.max_meters;
5839 } else {
5840 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5841 return 0;
5842 }
5843 first = last = meter_id;
5844 }
5845
5846 /* First delete the rules that use this meter. If any of those rules are
5847 * currently being modified, postpone the whole operation until later. */
5848 rule_collection_init(&rules);
5849 ovs_mutex_lock(&ofproto_mutex);
5850 for (meter_id = first; meter_id <= last; ++meter_id) {
5851 struct meter *meter = ofproto->meters[meter_id];
5852 if (meter && !list_is_empty(&meter->rules)) {
5853 struct rule *rule;
5854
5855 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5856 rule_collection_add(&rules, rule);
5857 }
5858 }
5859 }
5860 delete_flows__(&rules, OFPRR_METER_DELETE, NULL);
5861
5862 /* Delete the meters. */
5863 meter_delete(ofproto, first, last);
5864
5865 ovs_mutex_unlock(&ofproto_mutex);
5866
5867 return error;
5868 }
5869
5870 static enum ofperr
5871 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5872 {
5873 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5874 struct ofputil_meter_mod mm;
5875 uint64_t bands_stub[256 / 8];
5876 struct ofpbuf bands;
5877 uint32_t meter_id;
5878 enum ofperr error;
5879
5880 error = reject_slave_controller(ofconn);
5881 if (error) {
5882 return error;
5883 }
5884
5885 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5886
5887 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5888 if (error) {
5889 goto exit_free_bands;
5890 }
5891
5892 meter_id = mm.meter.meter_id;
5893
5894 if (mm.command != OFPMC13_DELETE) {
5895 /* Fails also when meters are not implemented by the provider. */
5896 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5897 error = OFPERR_OFPMMFC_INVALID_METER;
5898 goto exit_free_bands;
5899 } else if (meter_id > ofproto->meter_features.max_meters) {
5900 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5901 goto exit_free_bands;
5902 }
5903 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5904 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5905 goto exit_free_bands;
5906 }
5907 }
5908
5909 switch (mm.command) {
5910 case OFPMC13_ADD:
5911 error = handle_add_meter(ofproto, &mm);
5912 break;
5913
5914 case OFPMC13_MODIFY:
5915 error = handle_modify_meter(ofproto, &mm);
5916 break;
5917
5918 case OFPMC13_DELETE:
5919 error = handle_delete_meter(ofconn, &mm);
5920 break;
5921
5922 default:
5923 error = OFPERR_OFPMMFC_BAD_COMMAND;
5924 break;
5925 }
5926
5927 exit_free_bands:
5928 ofpbuf_uninit(&bands);
5929 return error;
5930 }
5931
5932 static enum ofperr
5933 handle_meter_features_request(struct ofconn *ofconn,
5934 const struct ofp_header *request)
5935 {
5936 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5937 struct ofputil_meter_features features;
5938 struct ofpbuf *b;
5939
5940 if (ofproto->ofproto_class->meter_get_features) {
5941 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5942 } else {
5943 memset(&features, 0, sizeof features);
5944 }
5945 b = ofputil_encode_meter_features_reply(&features, request);
5946
5947 ofconn_send_reply(ofconn, b);
5948 return 0;
5949 }
5950
5951 static enum ofperr
5952 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5953 enum ofptype type)
5954 {
5955 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5956 struct ovs_list replies;
5957 uint64_t bands_stub[256 / 8];
5958 struct ofpbuf bands;
5959 uint32_t meter_id, first, last;
5960
5961 ofputil_decode_meter_request(request, &meter_id);
5962
5963 if (meter_id == OFPM13_ALL) {
5964 first = 1;
5965 last = ofproto->meter_features.max_meters;
5966 } else {
5967 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5968 !ofproto->meters[meter_id]) {
5969 return OFPERR_OFPMMFC_UNKNOWN_METER;
5970 }
5971 first = last = meter_id;
5972 }
5973
5974 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5975 ofpmp_init(&replies, request);
5976
5977 for (meter_id = first; meter_id <= last; ++meter_id) {
5978 struct meter *meter = ofproto->meters[meter_id];
5979 if (!meter) {
5980 continue; /* Skip non-existing meters. */
5981 }
5982 if (type == OFPTYPE_METER_STATS_REQUEST) {
5983 struct ofputil_meter_stats stats;
5984
5985 stats.meter_id = meter_id;
5986
5987 /* Provider sets the packet and byte counts, we do the rest. */
5988 stats.flow_count = list_size(&meter->rules);
5989 calc_duration(meter->created, time_msec(),
5990 &stats.duration_sec, &stats.duration_nsec);
5991 stats.n_bands = meter->n_bands;
5992 ofpbuf_clear(&bands);
5993 stats.bands
5994 = ofpbuf_put_uninit(&bands,
5995 meter->n_bands * sizeof *stats.bands);
5996
5997 if (!ofproto->ofproto_class->meter_get(ofproto,
5998 meter->provider_meter_id,
5999 &stats)) {
6000 ofputil_append_meter_stats(&replies, &stats);
6001 }
6002 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
6003 struct ofputil_meter_config config;
6004
6005 config.meter_id = meter_id;
6006 config.flags = meter->flags;
6007 config.n_bands = meter->n_bands;
6008 config.bands = meter->bands;
6009 ofputil_append_meter_config(&replies, &config);
6010 }
6011 }
6012
6013 ofconn_send_replies(ofconn, &replies);
6014 ofpbuf_uninit(&bands);
6015 return 0;
6016 }
6017
6018 static bool
6019 ofproto_group_lookup__(const struct ofproto *ofproto, uint32_t group_id,
6020 struct ofgroup **group)
6021 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6022 {
6023 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
6024 hash_int(group_id, 0), &ofproto->groups) {
6025 if ((*group)->group_id == group_id) {
6026 return true;
6027 }
6028 }
6029
6030 return false;
6031 }
6032
6033 /* If the group exists, this function increments the groups's reference count.
6034 *
6035 * Make sure to call ofproto_group_unref() after no longer needing to maintain
6036 * a reference to the group. */
6037 bool
6038 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
6039 struct ofgroup **group)
6040 {
6041 bool found;
6042
6043 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6044 found = ofproto_group_lookup__(ofproto, group_id, group);
6045 if (found) {
6046 ofproto_group_ref(*group);
6047 }
6048 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6049 return found;
6050 }
6051
6052 static bool
6053 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
6054 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6055 {
6056 struct ofgroup *grp;
6057
6058 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
6059 hash_int(group_id, 0), &ofproto->groups) {
6060 if (grp->group_id == group_id) {
6061 return true;
6062 }
6063 }
6064 return false;
6065 }
6066
6067 static bool
6068 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
6069 OVS_EXCLUDED(ofproto->groups_rwlock)
6070 {
6071 bool exists;
6072
6073 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6074 exists = ofproto_group_exists__(ofproto, group_id);
6075 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6076
6077 return exists;
6078 }
6079
6080 static uint32_t
6081 group_get_ref_count(struct ofgroup *group)
6082 OVS_EXCLUDED(ofproto_mutex)
6083 {
6084 struct ofproto *ofproto = CONST_CAST(struct ofproto *, group->ofproto);
6085 struct rule_criteria criteria;
6086 struct rule_collection rules;
6087 struct match match;
6088 enum ofperr error;
6089 uint32_t count;
6090
6091 match_init_catchall(&match);
6092 rule_criteria_init(&criteria, 0xff, &match, 0, CLS_MAX_VERSION, htonll(0),
6093 htonll(0), OFPP_ANY, group->group_id);
6094 ovs_mutex_lock(&ofproto_mutex);
6095 error = collect_rules_loose(ofproto, &criteria, &rules);
6096 ovs_mutex_unlock(&ofproto_mutex);
6097 rule_criteria_destroy(&criteria);
6098
6099 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
6100
6101 rule_collection_destroy(&rules);
6102 return count;
6103 }
6104
6105 static void
6106 append_group_stats(struct ofgroup *group, struct ovs_list *replies)
6107 {
6108 struct ofputil_group_stats ogs;
6109 const struct ofproto *ofproto = group->ofproto;
6110 long long int now = time_msec();
6111 int error;
6112
6113 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
6114
6115 /* Provider sets the packet and byte counts, we do the rest. */
6116 ogs.ref_count = group_get_ref_count(group);
6117 ogs.n_buckets = group->n_buckets;
6118
6119 error = (ofproto->ofproto_class->group_get_stats
6120 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
6121 : EOPNOTSUPP);
6122 if (error) {
6123 ogs.packet_count = UINT64_MAX;
6124 ogs.byte_count = UINT64_MAX;
6125 memset(ogs.bucket_stats, 0xff,
6126 ogs.n_buckets * sizeof *ogs.bucket_stats);
6127 }
6128
6129 ogs.group_id = group->group_id;
6130 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
6131
6132 ofputil_append_group_stats(replies, &ogs);
6133
6134 free(ogs.bucket_stats);
6135 }
6136
6137 static void
6138 handle_group_request(struct ofconn *ofconn,
6139 const struct ofp_header *request, uint32_t group_id,
6140 void (*cb)(struct ofgroup *, struct ovs_list *replies))
6141 {
6142 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6143 struct ofgroup *group;
6144 struct ovs_list replies;
6145
6146 ofpmp_init(&replies, request);
6147 if (group_id == OFPG_ALL) {
6148 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6149 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
6150 cb(group, &replies);
6151 }
6152 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6153 } else {
6154 if (ofproto_group_lookup(ofproto, group_id, &group)) {
6155 cb(group, &replies);
6156 ofproto_group_unref(group);
6157 }
6158 }
6159 ofconn_send_replies(ofconn, &replies);
6160 }
6161
6162 static enum ofperr
6163 handle_group_stats_request(struct ofconn *ofconn,
6164 const struct ofp_header *request)
6165 {
6166 uint32_t group_id;
6167 enum ofperr error;
6168
6169 error = ofputil_decode_group_stats_request(request, &group_id);
6170 if (error) {
6171 return error;
6172 }
6173
6174 handle_group_request(ofconn, request, group_id, append_group_stats);
6175 return 0;
6176 }
6177
6178 static void
6179 append_group_desc(struct ofgroup *group, struct ovs_list *replies)
6180 {
6181 struct ofputil_group_desc gds;
6182
6183 gds.group_id = group->group_id;
6184 gds.type = group->type;
6185 gds.props = group->props;
6186
6187 ofputil_append_group_desc_reply(&gds, &group->buckets, replies);
6188 }
6189
6190 static enum ofperr
6191 handle_group_desc_stats_request(struct ofconn *ofconn,
6192 const struct ofp_header *request)
6193 {
6194 handle_group_request(ofconn, request,
6195 ofputil_decode_group_desc_request(request),
6196 append_group_desc);
6197 return 0;
6198 }
6199
6200 static enum ofperr
6201 handle_group_features_stats_request(struct ofconn *ofconn,
6202 const struct ofp_header *request)
6203 {
6204 struct ofproto *p = ofconn_get_ofproto(ofconn);
6205 struct ofpbuf *msg;
6206
6207 msg = ofputil_encode_group_features_reply(&p->ogf, request);
6208 if (msg) {
6209 ofconn_send_reply(ofconn, msg);
6210 }
6211
6212 return 0;
6213 }
6214
6215 static enum ofperr
6216 handle_queue_get_config_request(struct ofconn *ofconn,
6217 const struct ofp_header *oh)
6218 {
6219 struct ofproto *p = ofconn_get_ofproto(ofconn);
6220 struct netdev_queue_dump queue_dump;
6221 struct ofport *ofport;
6222 unsigned int queue_id;
6223 struct ofpbuf *reply;
6224 struct smap details;
6225 ofp_port_t request;
6226 enum ofperr error;
6227
6228 error = ofputil_decode_queue_get_config_request(oh, &request);
6229 if (error) {
6230 return error;
6231 }
6232
6233 ofport = ofproto_get_port(p, request);
6234 if (!ofport) {
6235 return OFPERR_OFPQOFC_BAD_PORT;
6236 }
6237
6238 reply = ofputil_encode_queue_get_config_reply(oh);
6239
6240 smap_init(&details);
6241 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump, ofport->netdev) {
6242 struct ofputil_queue_config queue;
6243
6244 /* None of the existing queues have compatible properties, so we
6245 * hard-code omitting min_rate and max_rate. */
6246 queue.queue_id = queue_id;
6247 queue.min_rate = UINT16_MAX;
6248 queue.max_rate = UINT16_MAX;
6249 ofputil_append_queue_get_config_reply(reply, &queue);
6250 }
6251 smap_destroy(&details);
6252
6253 ofconn_send_reply(ofconn, reply);
6254
6255 return 0;
6256 }
6257
6258 static enum ofperr
6259 init_group(struct ofproto *ofproto, struct ofputil_group_mod *gm,
6260 struct ofgroup **ofgroup)
6261 {
6262 enum ofperr error;
6263 const long long int now = time_msec();
6264
6265 if (gm->group_id > OFPG_MAX) {
6266 return OFPERR_OFPGMFC_INVALID_GROUP;
6267 }
6268 if (gm->type > OFPGT11_FF) {
6269 return OFPERR_OFPGMFC_BAD_TYPE;
6270 }
6271
6272 *ofgroup = ofproto->ofproto_class->group_alloc();
6273 if (!*ofgroup) {
6274 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
6275 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
6276 }
6277
6278 (*ofgroup)->ofproto = ofproto;
6279 *CONST_CAST(uint32_t *, &((*ofgroup)->group_id)) = gm->group_id;
6280 *CONST_CAST(enum ofp11_group_type *, &(*ofgroup)->type) = gm->type;
6281 *CONST_CAST(long long int *, &((*ofgroup)->created)) = now;
6282 *CONST_CAST(long long int *, &((*ofgroup)->modified)) = now;
6283 ovs_refcount_init(&(*ofgroup)->ref_count);
6284
6285 list_move(&(*ofgroup)->buckets, &gm->buckets);
6286 *CONST_CAST(uint32_t *, &(*ofgroup)->n_buckets) =
6287 list_size(&(*ofgroup)->buckets);
6288
6289 memcpy(CONST_CAST(struct ofputil_group_props *, &(*ofgroup)->props),
6290 &gm->props, sizeof (struct ofputil_group_props));
6291
6292 /* Construct called BEFORE any locks are held. */
6293 error = ofproto->ofproto_class->group_construct(*ofgroup);
6294 if (error) {
6295 ofputil_bucket_list_destroy(&(*ofgroup)->buckets);
6296 ofproto->ofproto_class->group_dealloc(*ofgroup);
6297 }
6298 return error;
6299 }
6300
6301 /* Implements the OFPGC11_ADD operation specified by 'gm', adding a group to
6302 * 'ofproto''s group table. Returns 0 on success or an OpenFlow error code on
6303 * failure. */
6304 static enum ofperr
6305 add_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
6306 {
6307 struct ofgroup *ofgroup;
6308 enum ofperr error;
6309
6310 /* Allocate new group and initialize it. */
6311 error = init_group(ofproto, gm, &ofgroup);
6312 if (error) {
6313 return error;
6314 }
6315
6316 /* We wrlock as late as possible to minimize the time we jam any other
6317 * threads: No visible state changes before acquiring the lock. */
6318 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6319
6320 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6321 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6322 goto unlock_out;
6323 }
6324
6325 if (ofproto_group_exists__(ofproto, gm->group_id)) {
6326 error = OFPERR_OFPGMFC_GROUP_EXISTS;
6327 goto unlock_out;
6328 }
6329
6330 if (!error) {
6331 /* Insert new group. */
6332 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
6333 hash_int(ofgroup->group_id, 0));
6334 ofproto->n_groups[ofgroup->type]++;
6335
6336 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6337 return error;
6338 }
6339
6340 unlock_out:
6341 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6342 ofproto->ofproto_class->group_destruct(ofgroup);
6343 ofputil_bucket_list_destroy(&ofgroup->buckets);
6344 ofproto->ofproto_class->group_dealloc(ofgroup);
6345
6346 return error;
6347 }
6348
6349 /* Adds all of the buckets from 'ofgroup' to 'new_ofgroup'. The buckets
6350 * already in 'new_ofgroup' will be placed just after the (copy of the) bucket
6351 * in 'ofgroup' with bucket ID 'command_bucket_id'. Special
6352 * 'command_bucket_id' values OFPG15_BUCKET_FIRST and OFPG15_BUCKET_LAST are
6353 * also honored. */
6354 static enum ofperr
6355 copy_buckets_for_insert_bucket(const struct ofgroup *ofgroup,
6356 struct ofgroup *new_ofgroup,
6357 uint32_t command_bucket_id)
6358 {
6359 struct ofputil_bucket *last = NULL;
6360
6361 if (command_bucket_id <= OFPG15_BUCKET_MAX) {
6362 /* Check here to ensure that a bucket corresponding to
6363 * command_bucket_id exists in the old bucket list.
6364 *
6365 * The subsequent search of below of new_ofgroup covers
6366 * both buckets in the old bucket list and buckets added
6367 * by the insert buckets group mod message this function processes. */
6368 if (!ofputil_bucket_find(&ofgroup->buckets, command_bucket_id)) {
6369 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6370 }
6371
6372 if (!list_is_empty(&new_ofgroup->buckets)) {
6373 last = ofputil_bucket_list_back(&new_ofgroup->buckets);
6374 }
6375 }
6376
6377 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, NULL);
6378
6379 if (ofputil_bucket_check_duplicate_id(&ofgroup->buckets)) {
6380 VLOG_WARN_RL(&rl, "Duplicate bucket id");
6381 return OFPERR_OFPGMFC_BUCKET_EXISTS;
6382 }
6383
6384 /* Rearrange list according to command_bucket_id */
6385 if (command_bucket_id == OFPG15_BUCKET_LAST) {
6386 struct ofputil_bucket *new_first;
6387 const struct ofputil_bucket *first;
6388
6389 first = ofputil_bucket_list_front(&ofgroup->buckets);
6390 new_first = ofputil_bucket_find(&new_ofgroup->buckets,
6391 first->bucket_id);
6392
6393 list_splice(new_ofgroup->buckets.next, &new_first->list_node,
6394 &new_ofgroup->buckets);
6395 } else if (command_bucket_id <= OFPG15_BUCKET_MAX && last) {
6396 struct ofputil_bucket *after;
6397
6398 /* Presence of bucket is checked above so after should never be NULL */
6399 after = ofputil_bucket_find(&new_ofgroup->buckets, command_bucket_id);
6400
6401 list_splice(after->list_node.next, new_ofgroup->buckets.next,
6402 last->list_node.next);
6403 }
6404
6405 return 0;
6406 }
6407
6408 /* Appends all of the a copy of all the buckets from 'ofgroup' to 'new_ofgroup'
6409 * with the exception of the bucket whose bucket id is 'command_bucket_id'.
6410 * Special 'command_bucket_id' values OFPG15_BUCKET_FIRST, OFPG15_BUCKET_LAST
6411 * and OFPG15_BUCKET_ALL are also honored. */
6412 static enum ofperr
6413 copy_buckets_for_remove_bucket(const struct ofgroup *ofgroup,
6414 struct ofgroup *new_ofgroup,
6415 uint32_t command_bucket_id)
6416 {
6417 const struct ofputil_bucket *skip = NULL;
6418
6419 if (command_bucket_id == OFPG15_BUCKET_ALL) {
6420 return 0;
6421 }
6422
6423 if (command_bucket_id == OFPG15_BUCKET_FIRST) {
6424 if (!list_is_empty(&ofgroup->buckets)) {
6425 skip = ofputil_bucket_list_front(&ofgroup->buckets);
6426 }
6427 } else if (command_bucket_id == OFPG15_BUCKET_LAST) {
6428 if (!list_is_empty(&ofgroup->buckets)) {
6429 skip = ofputil_bucket_list_back(&ofgroup->buckets);
6430 }
6431 } else {
6432 skip = ofputil_bucket_find(&ofgroup->buckets, command_bucket_id);
6433 if (!skip) {
6434 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6435 }
6436 }
6437
6438 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, skip);
6439
6440 return 0;
6441 }
6442
6443 /* Implements OFPGC11_MODIFY, OFPGC15_INSERT_BUCKET and
6444 * OFPGC15_REMOVE_BUCKET. Returns 0 on success or an OpenFlow error code
6445 * on failure.
6446 *
6447 * Note that the group is re-created and then replaces the old group in
6448 * ofproto's ofgroup hash map. Thus, the group is never altered while users of
6449 * the xlate module hold a pointer to the group. */
6450 static enum ofperr
6451 modify_group(struct ofproto *ofproto, struct ofputil_group_mod *gm)
6452 {
6453 struct ofgroup *ofgroup, *new_ofgroup, *retiring;
6454 enum ofperr error;
6455
6456 error = init_group(ofproto, gm, &new_ofgroup);
6457 if (error) {
6458 return error;
6459 }
6460
6461 retiring = new_ofgroup;
6462
6463 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6464 if (!ofproto_group_lookup__(ofproto, gm->group_id, &ofgroup)) {
6465 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
6466 goto out;
6467 }
6468
6469 /* Ofproto's group write lock is held now. */
6470 if (ofgroup->type != gm->type
6471 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6472 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6473 goto out;
6474 }
6475
6476 /* Manipulate bucket list for bucket commands */
6477 if (gm->command == OFPGC15_INSERT_BUCKET) {
6478 error = copy_buckets_for_insert_bucket(ofgroup, new_ofgroup,
6479 gm->command_bucket_id);
6480 } else if (gm->command == OFPGC15_REMOVE_BUCKET) {
6481 error = copy_buckets_for_remove_bucket(ofgroup, new_ofgroup,
6482 gm->command_bucket_id);
6483 }
6484 if (error) {
6485 goto out;
6486 }
6487
6488 /* The group creation time does not change during modification. */
6489 *CONST_CAST(long long int *, &(new_ofgroup->created)) = ofgroup->created;
6490 *CONST_CAST(long long int *, &(new_ofgroup->modified)) = time_msec();
6491
6492 error = ofproto->ofproto_class->group_modify(new_ofgroup);
6493 if (error) {
6494 goto out;
6495 }
6496
6497 retiring = ofgroup;
6498 /* Replace ofgroup in ofproto's groups hash map with new_ofgroup. */
6499 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6500 hmap_insert(&ofproto->groups, &new_ofgroup->hmap_node,
6501 hash_int(new_ofgroup->group_id, 0));
6502 if (ofgroup->type != new_ofgroup->type) {
6503 ofproto->n_groups[ofgroup->type]--;
6504 ofproto->n_groups[new_ofgroup->type]++;
6505 }
6506
6507 out:
6508 ofproto_group_unref(retiring);
6509 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6510 return error;
6511 }
6512
6513 static void
6514 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
6515 OVS_RELEASES(ofproto->groups_rwlock)
6516 {
6517 struct match match;
6518 struct ofproto_flow_mod ofm;
6519
6520 /* Delete all flow entries containing this group in a group action */
6521 match_init_catchall(&match);
6522 flow_mod_init(&ofm.fm, &match, 0, NULL, 0, OFPFC_DELETE);
6523 ofm.fm.delete_reason = OFPRR_GROUP_DELETE;
6524 ofm.fm.out_group = ofgroup->group_id;
6525 handle_flow_mod__(ofproto, &ofm, NULL);
6526
6527 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6528 /* No-one can find this group any more. */
6529 ofproto->n_groups[ofgroup->type]--;
6530 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6531 ofproto_group_unref(ofgroup);
6532 }
6533
6534 /* Implements OFPGC11_DELETE. */
6535 static void
6536 delete_group(struct ofproto *ofproto, uint32_t group_id)
6537 {
6538 struct ofgroup *ofgroup;
6539
6540 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6541 if (group_id == OFPG_ALL) {
6542 for (;;) {
6543 struct hmap_node *node = hmap_first(&ofproto->groups);
6544 if (!node) {
6545 break;
6546 }
6547 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
6548 delete_group__(ofproto, ofgroup);
6549 /* Lock for each node separately, so that we will not jam the
6550 * other threads for too long time. */
6551 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6552 }
6553 } else {
6554 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
6555 hash_int(group_id, 0), &ofproto->groups) {
6556 if (ofgroup->group_id == group_id) {
6557 delete_group__(ofproto, ofgroup);
6558 return;
6559 }
6560 }
6561 }
6562 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6563 }
6564
6565 /* Delete all groups from 'ofproto'.
6566 *
6567 * This is intended for use within an ofproto provider's 'destruct'
6568 * function. */
6569 void
6570 ofproto_group_delete_all(struct ofproto *ofproto)
6571 {
6572 delete_group(ofproto, OFPG_ALL);
6573 }
6574
6575 static enum ofperr
6576 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6577 {
6578 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6579 struct ofputil_group_mod gm;
6580 enum ofperr error;
6581
6582 error = reject_slave_controller(ofconn);
6583 if (error) {
6584 return error;
6585 }
6586
6587 error = ofputil_decode_group_mod(oh, &gm);
6588 if (error) {
6589 return error;
6590 }
6591
6592 switch (gm.command) {
6593 case OFPGC11_ADD:
6594 return add_group(ofproto, &gm);
6595
6596 case OFPGC11_MODIFY:
6597 return modify_group(ofproto, &gm);
6598
6599 case OFPGC11_DELETE:
6600 delete_group(ofproto, gm.group_id);
6601 return 0;
6602
6603 case OFPGC15_INSERT_BUCKET:
6604 return modify_group(ofproto, &gm);
6605
6606 case OFPGC15_REMOVE_BUCKET:
6607 return modify_group(ofproto, &gm);
6608
6609 default:
6610 if (gm.command > OFPGC11_DELETE) {
6611 VLOG_WARN_RL(&rl, "%s: Invalid group_mod command type %d",
6612 ofproto->name, gm.command);
6613 }
6614 return OFPERR_OFPGMFC_BAD_COMMAND;
6615 }
6616 }
6617
6618 enum ofputil_table_miss
6619 ofproto_table_get_miss_config(const struct ofproto *ofproto, uint8_t table_id)
6620 {
6621 enum ofputil_table_miss miss;
6622
6623 atomic_read_relaxed(&ofproto->tables[table_id].miss_config, &miss);
6624 return miss;
6625 }
6626
6627 static void
6628 table_mod__(struct oftable *oftable,
6629 enum ofputil_table_miss miss, enum ofputil_table_eviction eviction)
6630 {
6631 if (miss == OFPUTIL_TABLE_MISS_DEFAULT) {
6632 /* This is how an OFPT_TABLE_MOD decodes if it doesn't specify any
6633 * table-miss configuration (because the protocol used doesn't have
6634 * such a concept), so there's nothing to do. */
6635 } else {
6636 atomic_store_relaxed(&oftable->miss_config, miss);
6637 }
6638
6639 unsigned int new_eviction = oftable->eviction;
6640 if (eviction == OFPUTIL_TABLE_EVICTION_ON) {
6641 new_eviction |= EVICTION_OPENFLOW;
6642 } else if (eviction == OFPUTIL_TABLE_EVICTION_OFF) {
6643 new_eviction &= ~EVICTION_OPENFLOW;
6644 }
6645
6646 if (new_eviction != oftable->eviction) {
6647 ovs_mutex_lock(&ofproto_mutex);
6648 oftable_configure_eviction(oftable, new_eviction,
6649 oftable->eviction_fields,
6650 oftable->n_eviction_fields);
6651 ovs_mutex_unlock(&ofproto_mutex);
6652 }
6653 }
6654
6655 static enum ofperr
6656 table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
6657 {
6658 if (!check_table_id(ofproto, tm->table_id)) {
6659 return OFPERR_OFPTMFC_BAD_TABLE;
6660 }
6661
6662 /* Don't allow the eviction flags to be changed (except to the only fixed
6663 * value that OVS supports). OF1.4 says this is normal: "The
6664 * OFPTMPT_EVICTION property usually cannot be modified using a
6665 * OFP_TABLE_MOD request, because the eviction mechanism is switch
6666 * defined". */
6667 if (tm->eviction_flags != UINT32_MAX
6668 && tm->eviction_flags != OFPROTO_EVICTION_FLAGS) {
6669 return OFPERR_OFPTMFC_BAD_CONFIG;
6670 }
6671
6672 if (tm->table_id == OFPTT_ALL) {
6673 struct oftable *oftable;
6674 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6675 if (!(oftable->flags & (OFTABLE_HIDDEN | OFTABLE_READONLY))) {
6676 table_mod__(oftable, tm->miss, tm->eviction);
6677 }
6678 }
6679 } else {
6680 struct oftable *oftable = &ofproto->tables[tm->table_id];
6681 if (oftable->flags & OFTABLE_READONLY) {
6682 return OFPERR_OFPTMFC_EPERM;
6683 }
6684 table_mod__(oftable, tm->miss, tm->eviction);
6685 }
6686
6687 return 0;
6688 }
6689
6690 static enum ofperr
6691 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6692 {
6693 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6694 struct ofputil_table_mod tm;
6695 enum ofperr error;
6696
6697 error = reject_slave_controller(ofconn);
6698 if (error) {
6699 return error;
6700 }
6701
6702 error = ofputil_decode_table_mod(oh, &tm);
6703 if (error) {
6704 return error;
6705 }
6706
6707 return table_mod(ofproto, &tm);
6708 }
6709
6710 static enum ofperr
6711 ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6712 OVS_REQUIRES(ofproto_mutex)
6713 {
6714 switch (ofm->fm.command) {
6715 case OFPFC_ADD:
6716 return add_flow_start(ofproto, ofm);
6717 /* , &be->old_rules.stub[0],
6718 &be->new_rules.stub[0]); */
6719 case OFPFC_MODIFY:
6720 return modify_flows_start_loose(ofproto, ofm);
6721 case OFPFC_MODIFY_STRICT:
6722 return modify_flow_start_strict(ofproto, ofm);
6723 case OFPFC_DELETE:
6724 return delete_flows_start_loose(ofproto, ofm);
6725
6726 case OFPFC_DELETE_STRICT:
6727 return delete_flow_start_strict(ofproto, ofm);
6728 }
6729
6730 return OFPERR_OFPFMFC_BAD_COMMAND;
6731 }
6732
6733 static void
6734 ofproto_flow_mod_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6735 OVS_REQUIRES(ofproto_mutex)
6736 {
6737 switch (ofm->fm.command) {
6738 case OFPFC_ADD:
6739 add_flow_revert(ofproto, ofm);
6740 break;
6741
6742 case OFPFC_MODIFY:
6743 case OFPFC_MODIFY_STRICT:
6744 modify_flows_revert(ofproto, ofm);
6745 break;
6746
6747 case OFPFC_DELETE:
6748 case OFPFC_DELETE_STRICT:
6749 delete_flows_revert(ofproto, ofm);
6750 break;
6751
6752 default:
6753 break;
6754 }
6755 }
6756
6757 static void
6758 ofproto_flow_mod_finish(struct ofproto *ofproto,
6759 struct ofproto_flow_mod *ofm,
6760 const struct flow_mod_requester *req)
6761 OVS_REQUIRES(ofproto_mutex)
6762 {
6763 switch (ofm->fm.command) {
6764 case OFPFC_ADD:
6765 add_flow_finish(ofproto, ofm, req);
6766 break;
6767
6768 case OFPFC_MODIFY:
6769 case OFPFC_MODIFY_STRICT:
6770 modify_flows_finish(ofproto, ofm, req);
6771 break;
6772
6773 case OFPFC_DELETE:
6774 case OFPFC_DELETE_STRICT:
6775 delete_flows_finish(ofproto, ofm, req);
6776 break;
6777
6778 default:
6779 break;
6780 }
6781 }
6782
6783 /* Commit phases (all while locking ofproto_mutex):
6784 *
6785 * 1. Begin: Gather resources and make changes visible in the next version.
6786 * - Mark affected rules for removal in the next version.
6787 * - Create new replacement rules, make visible in the next
6788 * version.
6789 * - Do not send any events or notifications.
6790 *
6791 * 2. Revert: Fail if any errors are found. After this point no errors are
6792 * possible. No visible changes were made, so rollback is minimal (remove
6793 * added invisible rules, restore visibility of rules marked for removal).
6794 *
6795 * 3. Finish: Make the changes visible for lookups. Insert replacement rules to
6796 * the ofproto provider. Remove replaced and deleted rules from ofproto data
6797 * structures, and Schedule postponed removal of deleted rules from the
6798 * classifier. Send notifications, buffered packets, etc.
6799 */
6800 static enum ofperr
6801 do_bundle_commit(struct ofconn *ofconn, uint32_t id, uint16_t flags)
6802 {
6803 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6804 cls_version_t version = ofproto->tables_version + 1;
6805 struct ofp_bundle *bundle;
6806 struct ofp_bundle_entry *be;
6807 enum ofperr error;
6808
6809 bundle = ofconn_get_bundle(ofconn, id);
6810
6811 if (!bundle) {
6812 return OFPERR_OFPBFC_BAD_ID;
6813 }
6814 if (bundle->flags != flags) {
6815 error = OFPERR_OFPBFC_BAD_FLAGS;
6816 } else {
6817 bool prev_is_port_mod = false;
6818
6819 error = 0;
6820 ovs_mutex_lock(&ofproto_mutex);
6821
6822 /* 1. Begin. */
6823 LIST_FOR_EACH (be, node, &bundle->msg_list) {
6824 if (be->type == OFPTYPE_PORT_MOD) {
6825 /* Our port mods are not atomic. */
6826 if (flags & OFPBF_ATOMIC) {
6827 error = OFPERR_OFPBFC_MSG_FAILED;
6828 } else {
6829 prev_is_port_mod = true;
6830 error = port_mod_start(ofconn, &be->opm.pm, &be->opm.port);
6831 }
6832 } else if (be->type == OFPTYPE_FLOW_MOD) {
6833 /* Flow mods between port mods are applied as a single
6834 * version, but the versions are published only after
6835 * we know the commit is successful. */
6836 if (prev_is_port_mod) {
6837 ++version;
6838 }
6839 prev_is_port_mod = false;
6840 /* Store the version in which the changes should take
6841 * effect. */
6842 be->ofm.version = version;
6843 error = ofproto_flow_mod_start(ofproto, &be->ofm);
6844 } else {
6845 OVS_NOT_REACHED();
6846 }
6847 if (error) {
6848 break;
6849 }
6850 }
6851
6852 if (error) {
6853 /* Send error referring to the original message. */
6854 if (error) {
6855 ofconn_send_error(ofconn, be->ofp_msg, error);
6856 error = OFPERR_OFPBFC_MSG_FAILED;
6857 }
6858
6859 /* 2. Revert. Undo all the changes made above. */
6860 LIST_FOR_EACH_REVERSE_CONTINUE(be, node, &bundle->msg_list) {
6861 if (be->type == OFPTYPE_FLOW_MOD) {
6862 ofproto_flow_mod_revert(ofproto, &be->ofm);
6863 }
6864 /* Nothing needs to be reverted for a port mod. */
6865 }
6866 } else {
6867 /* 4. Finish. */
6868 LIST_FOR_EACH (be, node, &bundle->msg_list) {
6869 if (be->type == OFPTYPE_FLOW_MOD) {
6870 struct flow_mod_requester req = { ofconn, be->ofp_msg };
6871
6872 /* Bump the lookup version to the one of the current
6873 * message. This makes all the changes in the bundle at
6874 * this version visible to lookups at once. */
6875 if (ofproto->tables_version < be->ofm.version) {
6876 ofproto->tables_version = be->ofm.version;
6877 ofproto->ofproto_class->set_tables_version(
6878 ofproto, ofproto->tables_version);
6879 }
6880
6881 ofproto_flow_mod_finish(ofproto, &be->ofm, &req);
6882 } else if (be->type == OFPTYPE_PORT_MOD) {
6883 /* Perform the actual port mod. This is not atomic, i.e.,
6884 * the effects will be immediately seen by upcall
6885 * processing regardless of the lookup version. It should
6886 * be noted that port configuration changes can originate
6887 * also from OVSDB changes asynchronously to all upcall
6888 * processing. */
6889 port_mod_finish(ofconn, &be->opm.pm, be->opm.port);
6890 }
6891 }
6892 }
6893
6894 ofmonitor_flush(ofproto->connmgr);
6895 ovs_mutex_unlock(&ofproto_mutex);
6896
6897 run_rule_executes(ofproto);
6898 }
6899
6900 /* The bundle is discarded regardless the outcome. */
6901 ofp_bundle_remove__(ofconn, bundle, !error);
6902 return error;
6903 }
6904
6905 static enum ofperr
6906 handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh)
6907 {
6908 struct ofputil_bundle_ctrl_msg bctrl;
6909 struct ofputil_bundle_ctrl_msg reply;
6910 struct ofpbuf *buf;
6911 enum ofperr error;
6912
6913 error = reject_slave_controller(ofconn);
6914 if (error) {
6915 return error;
6916 }
6917
6918 error = ofputil_decode_bundle_ctrl(oh, &bctrl);
6919 if (error) {
6920 return error;
6921 }
6922 reply.flags = 0;
6923 reply.bundle_id = bctrl.bundle_id;
6924
6925 switch (bctrl.type) {
6926 case OFPBCT_OPEN_REQUEST:
6927 error = ofp_bundle_open(ofconn, bctrl.bundle_id, bctrl.flags);
6928 reply.type = OFPBCT_OPEN_REPLY;
6929 break;
6930 case OFPBCT_CLOSE_REQUEST:
6931 error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags);
6932 reply.type = OFPBCT_CLOSE_REPLY;
6933 break;
6934 case OFPBCT_COMMIT_REQUEST:
6935 error = do_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags);
6936 reply.type = OFPBCT_COMMIT_REPLY;
6937 break;
6938 case OFPBCT_DISCARD_REQUEST:
6939 error = ofp_bundle_discard(ofconn, bctrl.bundle_id);
6940 reply.type = OFPBCT_DISCARD_REPLY;
6941 break;
6942
6943 case OFPBCT_OPEN_REPLY:
6944 case OFPBCT_CLOSE_REPLY:
6945 case OFPBCT_COMMIT_REPLY:
6946 case OFPBCT_DISCARD_REPLY:
6947 return OFPERR_OFPBFC_BAD_TYPE;
6948 break;
6949 }
6950
6951 if (!error) {
6952 buf = ofputil_encode_bundle_ctrl_reply(oh, &reply);
6953 ofconn_send_reply(ofconn, buf);
6954 }
6955 return error;
6956 }
6957
6958 static enum ofperr
6959 handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh)
6960 {
6961 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6962 enum ofperr error;
6963 struct ofputil_bundle_add_msg badd;
6964 struct ofp_bundle_entry *bmsg;
6965 enum ofptype type;
6966
6967 error = reject_slave_controller(ofconn);
6968 if (error) {
6969 return error;
6970 }
6971
6972 error = ofputil_decode_bundle_add(oh, &badd, &type);
6973 if (error) {
6974 return error;
6975 }
6976
6977 bmsg = ofp_bundle_entry_alloc(type, badd.msg);
6978
6979 if (type == OFPTYPE_PORT_MOD) {
6980 error = ofputil_decode_port_mod(badd.msg, &bmsg->opm.pm, false);
6981 } else if (type == OFPTYPE_FLOW_MOD) {
6982 struct ofpbuf ofpacts;
6983 uint64_t ofpacts_stub[1024 / 8];
6984
6985 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6986 error = ofputil_decode_flow_mod(&bmsg->ofm.fm, badd.msg,
6987 ofconn_get_protocol(ofconn),
6988 &ofpacts,
6989 u16_to_ofp(ofproto->max_ports),
6990 ofproto->n_tables);
6991 /* Move actions to heap. */
6992 bmsg->ofm.fm.ofpacts = ofpbuf_steal_data(&ofpacts);
6993
6994 if (!error && bmsg->ofm.fm.ofpacts_len) {
6995 error = ofproto_check_ofpacts(ofproto, bmsg->ofm.fm.ofpacts,
6996 bmsg->ofm.fm.ofpacts_len);
6997 }
6998 } else {
6999 OVS_NOT_REACHED();
7000 }
7001
7002 if (!error) {
7003 error = ofp_bundle_add_message(ofconn, badd.bundle_id, badd.flags,
7004 bmsg);
7005 }
7006
7007 if (error) {
7008 ofp_bundle_entry_free(bmsg);
7009 }
7010
7011 return error;
7012 }
7013
7014 static enum ofperr
7015 handle_geneve_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
7016 {
7017 struct ofputil_geneve_table_mod gtm;
7018 enum ofperr error;
7019
7020 error = reject_slave_controller(ofconn);
7021 if (error) {
7022 return error;
7023 }
7024
7025 error = ofputil_decode_geneve_table_mod(oh, &gtm);
7026 if (error) {
7027 return error;
7028 }
7029
7030 error = tun_metadata_table_mod(&gtm);
7031
7032 ofputil_uninit_geneve_table(&gtm.mappings);
7033 return error;
7034 }
7035
7036 static enum ofperr
7037 handle_geneve_table_request(struct ofconn *ofconn, const struct ofp_header *oh)
7038 {
7039 struct ofputil_geneve_table_reply gtr;
7040 struct ofpbuf *b;
7041
7042 tun_metadata_table_request(&gtr);
7043 b = ofputil_encode_geneve_table_reply(oh, &gtr);
7044 ofputil_uninit_geneve_table(&gtr.mappings);
7045
7046 ofconn_send_reply(ofconn, b);
7047 return 0;
7048 }
7049
7050 static enum ofperr
7051 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
7052 OVS_EXCLUDED(ofproto_mutex)
7053 {
7054 const struct ofp_header *oh = msg->data;
7055 enum ofptype type;
7056 enum ofperr error;
7057
7058 error = ofptype_decode(&type, oh);
7059 if (error) {
7060 return error;
7061 }
7062 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
7063 && ofpmp_more(oh)) {
7064 /* We have no buffer implementation for multipart requests.
7065 * Report overflow for requests which consists of multiple
7066 * messages. */
7067 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
7068 }
7069
7070 switch (type) {
7071 /* OpenFlow requests. */
7072 case OFPTYPE_ECHO_REQUEST:
7073 return handle_echo_request(ofconn, oh);
7074
7075 case OFPTYPE_FEATURES_REQUEST:
7076 return handle_features_request(ofconn, oh);
7077
7078 case OFPTYPE_GET_CONFIG_REQUEST:
7079 return handle_get_config_request(ofconn, oh);
7080
7081 case OFPTYPE_SET_CONFIG:
7082 return handle_set_config(ofconn, oh);
7083
7084 case OFPTYPE_PACKET_OUT:
7085 return handle_packet_out(ofconn, oh);
7086
7087 case OFPTYPE_PORT_MOD:
7088 return handle_port_mod(ofconn, oh);
7089
7090 case OFPTYPE_FLOW_MOD:
7091 return handle_flow_mod(ofconn, oh);
7092
7093 case OFPTYPE_GROUP_MOD:
7094 return handle_group_mod(ofconn, oh);
7095
7096 case OFPTYPE_TABLE_MOD:
7097 return handle_table_mod(ofconn, oh);
7098
7099 case OFPTYPE_METER_MOD:
7100 return handle_meter_mod(ofconn, oh);
7101
7102 case OFPTYPE_BARRIER_REQUEST:
7103 return handle_barrier_request(ofconn, oh);
7104
7105 case OFPTYPE_ROLE_REQUEST:
7106 return handle_role_request(ofconn, oh);
7107
7108 /* OpenFlow replies. */
7109 case OFPTYPE_ECHO_REPLY:
7110 return 0;
7111
7112 /* Nicira extension requests. */
7113 case OFPTYPE_FLOW_MOD_TABLE_ID:
7114 return handle_nxt_flow_mod_table_id(ofconn, oh);
7115
7116 case OFPTYPE_SET_FLOW_FORMAT:
7117 return handle_nxt_set_flow_format(ofconn, oh);
7118
7119 case OFPTYPE_SET_PACKET_IN_FORMAT:
7120 return handle_nxt_set_packet_in_format(ofconn, oh);
7121
7122 case OFPTYPE_SET_CONTROLLER_ID:
7123 return handle_nxt_set_controller_id(ofconn, oh);
7124
7125 case OFPTYPE_FLOW_AGE:
7126 /* Nothing to do. */
7127 return 0;
7128
7129 case OFPTYPE_FLOW_MONITOR_CANCEL:
7130 return handle_flow_monitor_cancel(ofconn, oh);
7131
7132 case OFPTYPE_SET_ASYNC_CONFIG:
7133 return handle_nxt_set_async_config(ofconn, oh);
7134
7135 case OFPTYPE_GET_ASYNC_REQUEST:
7136 return handle_nxt_get_async_request(ofconn, oh);
7137
7138 /* Statistics requests. */
7139 case OFPTYPE_DESC_STATS_REQUEST:
7140 return handle_desc_stats_request(ofconn, oh);
7141
7142 case OFPTYPE_FLOW_STATS_REQUEST:
7143 return handle_flow_stats_request(ofconn, oh);
7144
7145 case OFPTYPE_AGGREGATE_STATS_REQUEST:
7146 return handle_aggregate_stats_request(ofconn, oh);
7147
7148 case OFPTYPE_TABLE_STATS_REQUEST:
7149 return handle_table_stats_request(ofconn, oh);
7150
7151 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
7152 return handle_table_features_request(ofconn, oh);
7153
7154 case OFPTYPE_TABLE_DESC_REQUEST:
7155 return handle_table_desc_request(ofconn, oh);
7156
7157 case OFPTYPE_PORT_STATS_REQUEST:
7158 return handle_port_stats_request(ofconn, oh);
7159
7160 case OFPTYPE_QUEUE_STATS_REQUEST:
7161 return handle_queue_stats_request(ofconn, oh);
7162
7163 case OFPTYPE_PORT_DESC_STATS_REQUEST:
7164 return handle_port_desc_stats_request(ofconn, oh);
7165
7166 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
7167 return handle_flow_monitor_request(ofconn, oh);
7168
7169 case OFPTYPE_METER_STATS_REQUEST:
7170 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
7171 return handle_meter_request(ofconn, oh, type);
7172
7173 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
7174 return handle_meter_features_request(ofconn, oh);
7175
7176 case OFPTYPE_GROUP_STATS_REQUEST:
7177 return handle_group_stats_request(ofconn, oh);
7178
7179 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
7180 return handle_group_desc_stats_request(ofconn, oh);
7181
7182 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
7183 return handle_group_features_stats_request(ofconn, oh);
7184
7185 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
7186 return handle_queue_get_config_request(ofconn, oh);
7187
7188 case OFPTYPE_BUNDLE_CONTROL:
7189 return handle_bundle_control(ofconn, oh);
7190
7191 case OFPTYPE_BUNDLE_ADD_MESSAGE:
7192 return handle_bundle_add(ofconn, oh);
7193
7194 case OFPTYPE_NXT_GENEVE_TABLE_MOD:
7195 return handle_geneve_table_mod(ofconn, oh);
7196
7197 case OFPTYPE_NXT_GENEVE_TABLE_REQUEST:
7198 return handle_geneve_table_request(ofconn, oh);
7199
7200 case OFPTYPE_HELLO:
7201 case OFPTYPE_ERROR:
7202 case OFPTYPE_FEATURES_REPLY:
7203 case OFPTYPE_GET_CONFIG_REPLY:
7204 case OFPTYPE_PACKET_IN:
7205 case OFPTYPE_FLOW_REMOVED:
7206 case OFPTYPE_PORT_STATUS:
7207 case OFPTYPE_BARRIER_REPLY:
7208 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
7209 case OFPTYPE_DESC_STATS_REPLY:
7210 case OFPTYPE_FLOW_STATS_REPLY:
7211 case OFPTYPE_QUEUE_STATS_REPLY:
7212 case OFPTYPE_PORT_STATS_REPLY:
7213 case OFPTYPE_TABLE_STATS_REPLY:
7214 case OFPTYPE_AGGREGATE_STATS_REPLY:
7215 case OFPTYPE_PORT_DESC_STATS_REPLY:
7216 case OFPTYPE_ROLE_REPLY:
7217 case OFPTYPE_FLOW_MONITOR_PAUSED:
7218 case OFPTYPE_FLOW_MONITOR_RESUMED:
7219 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
7220 case OFPTYPE_GET_ASYNC_REPLY:
7221 case OFPTYPE_GROUP_STATS_REPLY:
7222 case OFPTYPE_GROUP_DESC_STATS_REPLY:
7223 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
7224 case OFPTYPE_METER_STATS_REPLY:
7225 case OFPTYPE_METER_CONFIG_STATS_REPLY:
7226 case OFPTYPE_METER_FEATURES_STATS_REPLY:
7227 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
7228 case OFPTYPE_TABLE_DESC_REPLY:
7229 case OFPTYPE_ROLE_STATUS:
7230 case OFPTYPE_NXT_GENEVE_TABLE_REPLY:
7231 default:
7232 if (ofpmsg_is_stat_request(oh)) {
7233 return OFPERR_OFPBRC_BAD_STAT;
7234 } else {
7235 return OFPERR_OFPBRC_BAD_TYPE;
7236 }
7237 }
7238 }
7239
7240 static void
7241 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
7242 OVS_EXCLUDED(ofproto_mutex)
7243 {
7244 enum ofperr error = handle_openflow__(ofconn, ofp_msg);
7245
7246 if (error) {
7247 ofconn_send_error(ofconn, ofp_msg->data, error);
7248 }
7249 COVERAGE_INC(ofproto_recv_openflow);
7250 }
7251 \f
7252 /* Asynchronous operations. */
7253
7254 static void
7255 send_buffered_packet(const struct flow_mod_requester *req, uint32_t buffer_id,
7256 struct rule *rule)
7257 OVS_REQUIRES(ofproto_mutex)
7258 {
7259 if (req && req->ofconn && buffer_id != UINT32_MAX) {
7260 struct ofproto *ofproto = ofconn_get_ofproto(req->ofconn);
7261 struct dp_packet *packet;
7262 ofp_port_t in_port;
7263 enum ofperr error;
7264
7265 error = ofconn_pktbuf_retrieve(req->ofconn, buffer_id, &packet,
7266 &in_port);
7267 if (packet) {
7268 struct rule_execute *re;
7269
7270 ofproto_rule_ref(rule);
7271
7272 re = xmalloc(sizeof *re);
7273 re->rule = rule;
7274 re->in_port = in_port;
7275 re->packet = packet;
7276
7277 if (!guarded_list_push_back(&ofproto->rule_executes,
7278 &re->list_node, 1024)) {
7279 ofproto_rule_unref(rule);
7280 dp_packet_delete(re->packet);
7281 free(re);
7282 }
7283 } else {
7284 ofconn_send_error(req->ofconn, req->request, error);
7285 }
7286 }
7287 }
7288 \f
7289 static uint64_t
7290 pick_datapath_id(const struct ofproto *ofproto)
7291 {
7292 const struct ofport *port;
7293
7294 port = ofproto_get_port(ofproto, OFPP_LOCAL);
7295 if (port) {
7296 uint8_t ea[ETH_ADDR_LEN];
7297 int error;
7298
7299 error = netdev_get_etheraddr(port->netdev, ea);
7300 if (!error) {
7301 return eth_addr_to_uint64(ea);
7302 }
7303 VLOG_WARN("%s: could not get MAC address for %s (%s)",
7304 ofproto->name, netdev_get_name(port->netdev),
7305 ovs_strerror(error));
7306 }
7307 return ofproto->fallback_dpid;
7308 }
7309
7310 static uint64_t
7311 pick_fallback_dpid(void)
7312 {
7313 uint8_t ea[ETH_ADDR_LEN];
7314 eth_addr_nicira_random(ea);
7315 return eth_addr_to_uint64(ea);
7316 }
7317 \f
7318 /* Table overflow policy. */
7319
7320 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
7321 * to NULL if the table is not configured to evict rules or if the table
7322 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
7323 * or with no timeouts are not evictable.) */
7324 static bool
7325 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
7326 OVS_REQUIRES(ofproto_mutex)
7327 {
7328 struct eviction_group *evg;
7329
7330 *rulep = NULL;
7331 if (!table->eviction) {
7332 return false;
7333 }
7334
7335 /* In the common case, the outer and inner loops here will each be entered
7336 * exactly once:
7337 *
7338 * - The inner loop normally "return"s in its first iteration. If the
7339 * eviction group has any evictable rules, then it always returns in
7340 * some iteration.
7341 *
7342 * - The outer loop only iterates more than once if the largest eviction
7343 * group has no evictable rules.
7344 *
7345 * - The outer loop can exit only if table's 'max_flows' is all filled up
7346 * by unevictable rules. */
7347 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
7348 struct rule *rule;
7349
7350 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
7351 *rulep = rule;
7352 return true;
7353 }
7354 }
7355
7356 return false;
7357 }
7358 \f
7359 /* Eviction groups. */
7360
7361 /* Returns the priority to use for an eviction_group that contains 'n_rules'
7362 * rules. The priority contains low-order random bits to ensure that eviction
7363 * groups with the same number of rules are prioritized randomly. */
7364 static uint32_t
7365 eviction_group_priority(size_t n_rules)
7366 {
7367 uint16_t size = MIN(UINT16_MAX, n_rules);
7368 return (size << 16) | random_uint16();
7369 }
7370
7371 /* Updates 'evg', an eviction_group within 'table', following a change that
7372 * adds or removes rules in 'evg'. */
7373 static void
7374 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
7375 OVS_REQUIRES(ofproto_mutex)
7376 {
7377 heap_change(&table->eviction_groups_by_size, &evg->size_node,
7378 eviction_group_priority(heap_count(&evg->rules)));
7379 }
7380
7381 /* Destroys 'evg', an eviction_group within 'table':
7382 *
7383 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
7384 * rules themselves, just removes them from the eviction group.)
7385 *
7386 * - Removes 'evg' from 'table'.
7387 *
7388 * - Frees 'evg'. */
7389 static void
7390 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
7391 OVS_REQUIRES(ofproto_mutex)
7392 {
7393 while (!heap_is_empty(&evg->rules)) {
7394 struct rule *rule;
7395
7396 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
7397 rule->eviction_group = NULL;
7398 }
7399 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
7400 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
7401 heap_destroy(&evg->rules);
7402 free(evg);
7403 }
7404
7405 /* Removes 'rule' from its eviction group, if any. */
7406 static void
7407 eviction_group_remove_rule(struct rule *rule)
7408 OVS_REQUIRES(ofproto_mutex)
7409 {
7410 if (rule->eviction_group) {
7411 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7412 struct eviction_group *evg = rule->eviction_group;
7413
7414 rule->eviction_group = NULL;
7415 heap_remove(&evg->rules, &rule->evg_node);
7416 if (heap_is_empty(&evg->rules)) {
7417 eviction_group_destroy(table, evg);
7418 } else {
7419 eviction_group_resized(table, evg);
7420 }
7421 }
7422 }
7423
7424 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
7425 * returns the hash value. */
7426 static uint32_t
7427 eviction_group_hash_rule(struct rule *rule)
7428 OVS_REQUIRES(ofproto_mutex)
7429 {
7430 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7431 const struct mf_subfield *sf;
7432 struct flow flow;
7433 uint32_t hash;
7434
7435 hash = table->eviction_group_id_basis;
7436 miniflow_expand(rule->cr.match.flow, &flow);
7437 for (sf = table->eviction_fields;
7438 sf < &table->eviction_fields[table->n_eviction_fields];
7439 sf++)
7440 {
7441 if (mf_are_prereqs_ok(sf->field, &flow)) {
7442 union mf_value value;
7443
7444 mf_get_value(sf->field, &flow, &value);
7445 if (sf->ofs) {
7446 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
7447 }
7448 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
7449 unsigned int start = sf->ofs + sf->n_bits;
7450 bitwise_zero(&value, sf->field->n_bytes, start,
7451 sf->field->n_bytes * 8 - start);
7452 }
7453 hash = hash_bytes(&value, sf->field->n_bytes, hash);
7454 } else {
7455 hash = hash_int(hash, 0);
7456 }
7457 }
7458
7459 return hash;
7460 }
7461
7462 /* Returns an eviction group within 'table' with the given 'id', creating one
7463 * if necessary. */
7464 static struct eviction_group *
7465 eviction_group_find(struct oftable *table, uint32_t id)
7466 OVS_REQUIRES(ofproto_mutex)
7467 {
7468 struct eviction_group *evg;
7469
7470 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
7471 return evg;
7472 }
7473
7474 evg = xmalloc(sizeof *evg);
7475 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
7476 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
7477 eviction_group_priority(0));
7478 heap_init(&evg->rules);
7479
7480 return evg;
7481 }
7482
7483 /* Returns an eviction priority for 'rule'. The return value should be
7484 * interpreted so that higher priorities make a rule a more attractive
7485 * candidate for eviction. */
7486 static uint64_t
7487 rule_eviction_priority(struct ofproto *ofproto, struct rule *rule)
7488 OVS_REQUIRES(ofproto_mutex)
7489 {
7490 /* Calculate absolute time when this flow will expire. If it will never
7491 * expire, then return 0 to make it unevictable. */
7492 long long int expiration = LLONG_MAX;
7493 if (rule->hard_timeout) {
7494 /* 'modified' needs protection even when we hold 'ofproto_mutex'. */
7495 ovs_mutex_lock(&rule->mutex);
7496 long long int modified = rule->modified;
7497 ovs_mutex_unlock(&rule->mutex);
7498
7499 expiration = modified + rule->hard_timeout * 1000;
7500 }
7501 if (rule->idle_timeout) {
7502 uint64_t packets, bytes;
7503 long long int used;
7504 long long int idle_expiration;
7505
7506 ofproto->ofproto_class->rule_get_stats(rule, &packets, &bytes, &used);
7507 idle_expiration = used + rule->idle_timeout * 1000;
7508 expiration = MIN(expiration, idle_expiration);
7509 }
7510 if (expiration == LLONG_MAX) {
7511 return 0;
7512 }
7513
7514 /* Calculate the time of expiration as a number of (approximate) seconds
7515 * after program startup.
7516 *
7517 * This should work OK for program runs that last UINT32_MAX seconds or
7518 * less. Therefore, please restart OVS at least once every 136 years. */
7519 uint32_t expiration_ofs = (expiration >> 10) - (time_boot_msec() >> 10);
7520
7521 /* Combine expiration time with OpenFlow "importance" to form a single
7522 * priority value. We want flows with relatively low "importance" to be
7523 * evicted before even considering expiration time, so put "importance" in
7524 * the most significant bits and expiration time in the least significant
7525 * bits.
7526 *
7527 * Small 'priority' should be evicted before those with large 'priority'.
7528 * The caller expects the opposite convention (a large return value being
7529 * more attractive for eviction) so we invert it before returning. */
7530 uint64_t priority = ((uint64_t) rule->importance << 32) + expiration_ofs;
7531 return UINT64_MAX - priority;
7532 }
7533
7534 /* Adds 'rule' to an appropriate eviction group for its oftable's
7535 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
7536 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
7537 * own).
7538 *
7539 * The caller must ensure that 'rule' is not already in an eviction group. */
7540 static void
7541 eviction_group_add_rule(struct rule *rule)
7542 OVS_REQUIRES(ofproto_mutex)
7543 {
7544 struct ofproto *ofproto = rule->ofproto;
7545 struct oftable *table = &ofproto->tables[rule->table_id];
7546 bool has_timeout;
7547
7548 /* Timeouts may be modified only when holding 'ofproto_mutex'. We have it
7549 * so no additional protection is needed. */
7550 has_timeout = rule->hard_timeout || rule->idle_timeout;
7551
7552 if (table->eviction && has_timeout) {
7553 struct eviction_group *evg;
7554
7555 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
7556
7557 rule->eviction_group = evg;
7558 heap_insert(&evg->rules, &rule->evg_node,
7559 rule_eviction_priority(ofproto, rule));
7560 eviction_group_resized(table, evg);
7561 }
7562 }
7563 \f
7564 /* oftables. */
7565
7566 /* Initializes 'table'. */
7567 static void
7568 oftable_init(struct oftable *table)
7569 {
7570 memset(table, 0, sizeof *table);
7571 classifier_init(&table->cls, flow_segment_u64s);
7572 table->max_flows = UINT_MAX;
7573 table->n_flows = 0;
7574 hmap_init(&table->eviction_groups_by_id);
7575 heap_init(&table->eviction_groups_by_size);
7576 atomic_init(&table->miss_config, OFPUTIL_TABLE_MISS_DEFAULT);
7577
7578 classifier_set_prefix_fields(&table->cls, default_prefix_fields,
7579 ARRAY_SIZE(default_prefix_fields));
7580
7581 atomic_init(&table->n_matched, 0);
7582 atomic_init(&table->n_missed, 0);
7583 }
7584
7585 /* Destroys 'table', including its classifier and eviction groups.
7586 *
7587 * The caller is responsible for freeing 'table' itself. */
7588 static void
7589 oftable_destroy(struct oftable *table)
7590 {
7591 ovs_assert(classifier_is_empty(&table->cls));
7592
7593 ovs_mutex_lock(&ofproto_mutex);
7594 oftable_configure_eviction(table, 0, NULL, 0);
7595 ovs_mutex_unlock(&ofproto_mutex);
7596
7597 hmap_destroy(&table->eviction_groups_by_id);
7598 heap_destroy(&table->eviction_groups_by_size);
7599 classifier_destroy(&table->cls);
7600 free(table->name);
7601 }
7602
7603 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
7604 * string, then 'table' will use its default name.
7605 *
7606 * This only affects the name exposed for a table exposed through the OpenFlow
7607 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
7608 static void
7609 oftable_set_name(struct oftable *table, const char *name)
7610 {
7611 if (name && name[0]) {
7612 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
7613 if (!table->name || strncmp(name, table->name, len)) {
7614 free(table->name);
7615 table->name = xmemdup0(name, len);
7616 }
7617 } else {
7618 free(table->name);
7619 table->name = NULL;
7620 }
7621 }
7622
7623 /* oftables support a choice of two policies when adding a rule would cause the
7624 * number of flows in the table to exceed the configured maximum number: either
7625 * they can refuse to add the new flow or they can evict some existing flow.
7626 * This function configures the latter policy on 'table', with fairness based
7627 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
7628 * 'n_fields' as 0 disables fairness.) */
7629 static void
7630 oftable_configure_eviction(struct oftable *table, unsigned int eviction,
7631 const struct mf_subfield *fields, size_t n_fields)
7632 OVS_REQUIRES(ofproto_mutex)
7633 {
7634 struct rule *rule;
7635
7636 if ((table->eviction != 0) == (eviction != 0)
7637 && n_fields == table->n_eviction_fields
7638 && (!n_fields
7639 || !memcmp(fields, table->eviction_fields,
7640 n_fields * sizeof *fields))) {
7641 /* The set of eviction fields did not change. If 'eviction' changed,
7642 * it remains nonzero, so that we can just update table->eviction
7643 * without fussing with the eviction groups. */
7644 table->eviction = eviction;
7645 return;
7646 }
7647
7648 /* Destroy existing eviction groups, then destroy and recreate data
7649 * structures to recover memory. */
7650 struct eviction_group *evg, *next;
7651 HMAP_FOR_EACH_SAFE (evg, next, id_node, &table->eviction_groups_by_id) {
7652 eviction_group_destroy(table, evg);
7653 }
7654 hmap_destroy(&table->eviction_groups_by_id);
7655 hmap_init(&table->eviction_groups_by_id);
7656 heap_destroy(&table->eviction_groups_by_size);
7657 heap_init(&table->eviction_groups_by_size);
7658
7659 /* Replace eviction groups by the new ones, if there is a change. Free the
7660 * old fields only after allocating the new ones, because 'fields ==
7661 * table->eviction_fields' is possible. */
7662 struct mf_subfield *old_fields = table->eviction_fields;
7663 table->n_eviction_fields = n_fields;
7664 table->eviction_fields = (fields
7665 ? xmemdup(fields, n_fields * sizeof *fields)
7666 : NULL);
7667 free(old_fields);
7668
7669 /* Add the new eviction groups, if enabled. */
7670 table->eviction = eviction;
7671 if (table->eviction) {
7672 table->eviction_group_id_basis = random_uint32();
7673 CLS_FOR_EACH (rule, cr, &table->cls) {
7674 eviction_group_add_rule(rule);
7675 }
7676 }
7677 }
7678
7679 /* Inserts 'rule' from the ofproto data structures BEFORE caller has inserted
7680 * it to the classifier. */
7681 static void
7682 ofproto_rule_insert__(struct ofproto *ofproto, struct rule *rule)
7683 OVS_REQUIRES(ofproto_mutex)
7684 {
7685 const struct rule_actions *actions = rule_get_actions(rule);
7686
7687 ovs_assert(rule->removed);
7688
7689 if (rule->hard_timeout || rule->idle_timeout) {
7690 list_insert(&ofproto->expirable, &rule->expirable);
7691 }
7692 cookies_insert(ofproto, rule);
7693 eviction_group_add_rule(rule);
7694 if (actions->has_meter) {
7695 meter_insert_rule(rule);
7696 }
7697 rule->removed = false;
7698 }
7699
7700 /* Removes 'rule' from the ofproto data structures. Caller may have deferred
7701 * the removal from the classifier. */
7702 static void
7703 ofproto_rule_remove__(struct ofproto *ofproto, struct rule *rule)
7704 OVS_REQUIRES(ofproto_mutex)
7705 {
7706 ovs_assert(!rule->removed);
7707
7708 cookies_remove(ofproto, rule);
7709
7710 eviction_group_remove_rule(rule);
7711 if (!list_is_empty(&rule->expirable)) {
7712 list_remove(&rule->expirable);
7713 }
7714 if (!list_is_empty(&rule->meter_list_node)) {
7715 list_remove(&rule->meter_list_node);
7716 list_init(&rule->meter_list_node);
7717 }
7718
7719 rule->removed = true;
7720 }
7721 \f
7722 /* unixctl commands. */
7723
7724 struct ofproto *
7725 ofproto_lookup(const char *name)
7726 {
7727 struct ofproto *ofproto;
7728
7729 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
7730 &all_ofprotos) {
7731 if (!strcmp(ofproto->name, name)) {
7732 return ofproto;
7733 }
7734 }
7735 return NULL;
7736 }
7737
7738 static void
7739 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
7740 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7741 {
7742 struct ofproto *ofproto;
7743 struct ds results;
7744
7745 ds_init(&results);
7746 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
7747 ds_put_format(&results, "%s\n", ofproto->name);
7748 }
7749 unixctl_command_reply(conn, ds_cstr(&results));
7750 ds_destroy(&results);
7751 }
7752
7753 static void
7754 ofproto_unixctl_init(void)
7755 {
7756 static bool registered;
7757 if (registered) {
7758 return;
7759 }
7760 registered = true;
7761
7762 unixctl_command_register("ofproto/list", "", 0, 0,
7763 ofproto_unixctl_list, NULL);
7764 }
7765 \f
7766 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7767 *
7768 * This is deprecated. It is only for compatibility with broken device drivers
7769 * in old versions of Linux that do not properly support VLANs when VLAN
7770 * devices are not used. When broken device drivers are no longer in
7771 * widespread use, we will delete these interfaces. */
7772
7773 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
7774 * (exactly) by an OpenFlow rule in 'ofproto'. */
7775 void
7776 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
7777 {
7778 struct match match;
7779 struct cls_rule target;
7780 const struct oftable *oftable;
7781
7782 match_init_catchall(&match);
7783 match_set_vlan_vid_masked(&match, htons(VLAN_CFI), htons(VLAN_CFI));
7784 cls_rule_init(&target, &match, 0);
7785
7786 free(ofproto->vlan_bitmap);
7787 ofproto->vlan_bitmap = bitmap_allocate(4096);
7788 ofproto->vlans_changed = false;
7789
7790 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
7791 struct rule *rule;
7792
7793 CLS_FOR_EACH_TARGET (rule, cr, &oftable->cls, &target,
7794 CLS_MAX_VERSION) {
7795 if (minimask_get_vid_mask(rule->cr.match.mask) == VLAN_VID_MASK) {
7796 uint16_t vid = miniflow_get_vid(rule->cr.match.flow);
7797
7798 bitmap_set1(vlan_bitmap, vid);
7799 bitmap_set1(ofproto->vlan_bitmap, vid);
7800 }
7801 }
7802 }
7803
7804 cls_rule_destroy(&target);
7805 }
7806
7807 /* Returns true if new VLANs have come into use by the flow table since the
7808 * last call to ofproto_get_vlan_usage().
7809 *
7810 * We don't track when old VLANs stop being used. */
7811 bool
7812 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
7813 {
7814 return ofproto->vlans_changed;
7815 }
7816
7817 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
7818 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
7819 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
7820 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
7821 * then the VLAN device is un-enslaved. */
7822 int
7823 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
7824 ofp_port_t realdev_ofp_port, int vid)
7825 {
7826 struct ofport *ofport;
7827 int error;
7828
7829 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
7830
7831 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
7832 if (!ofport) {
7833 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
7834 ofproto->name, vlandev_ofp_port);
7835 return EINVAL;
7836 }
7837
7838 if (!ofproto->ofproto_class->set_realdev) {
7839 if (!vlandev_ofp_port) {
7840 return 0;
7841 }
7842 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
7843 return EOPNOTSUPP;
7844 }
7845
7846 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
7847 if (error) {
7848 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
7849 ofproto->name, vlandev_ofp_port,
7850 netdev_get_name(ofport->netdev), ovs_strerror(error));
7851 }
7852 return error;
7853 }