]> git.proxmox.com Git - ovs.git/blob - ofproto/ofproto.c
fbaf7dd914caaa8e2d17584abcfedf704447f1f0
[ovs.git] / ofproto / ofproto.c
1 /*
2 * Copyright (c) 2009-2016 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 *,
301 const struct ofputil_group_mod *);
302 static void handle_openflow(struct ofconn *, const struct ofpbuf *);
303 static enum ofperr ofproto_flow_mod_start(struct ofproto *,
304 struct ofproto_flow_mod *)
305 OVS_REQUIRES(ofproto_mutex);
306 static void ofproto_flow_mod_finish(struct ofproto *,
307 struct ofproto_flow_mod *,
308 const struct flow_mod_requester *)
309 OVS_REQUIRES(ofproto_mutex);
310 static enum ofperr handle_flow_mod__(struct ofproto *,
311 struct ofproto_flow_mod *,
312 const struct flow_mod_requester *)
313 OVS_EXCLUDED(ofproto_mutex);
314 static void calc_duration(long long int start, long long int now,
315 uint32_t *sec, uint32_t *nsec);
316
317 /* ofproto. */
318 static uint64_t pick_datapath_id(const struct ofproto *);
319 static uint64_t pick_fallback_dpid(void);
320 static void ofproto_destroy__(struct ofproto *);
321 static void update_mtu(struct ofproto *, struct ofport *);
322 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
323 static void meter_insert_rule(struct rule *);
324
325 /* unixctl. */
326 static void ofproto_unixctl_init(void);
327
328 /* All registered ofproto classes, in probe order. */
329 static const struct ofproto_class **ofproto_classes;
330 static size_t n_ofproto_classes;
331 static size_t allocated_ofproto_classes;
332
333 /* Global lock that protects all flow table operations. */
334 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
335
336 unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
337 unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT;
338
339 size_t n_handlers, n_revalidators;
340 size_t n_dpdk_rxqs;
341 char *pmd_cpu_mask;
342
343 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
344 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
345
346 /* Initial mappings of port to OpenFlow number mappings. */
347 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
348
349 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
350
351 /* The default value of true waits for flow restore. */
352 static bool flow_restore_wait = true;
353
354 /* Must be called to initialize the ofproto library.
355 *
356 * The caller may pass in 'iface_hints', which contains an shash of
357 * "iface_hint" elements indexed by the interface's name. The provider
358 * may use these hints to describe the startup configuration in order to
359 * reinitialize its state. The caller owns the provided data, so a
360 * provider will make copies of anything required. An ofproto provider
361 * will remove any existing state that is not described by the hint, and
362 * may choose to remove it all. */
363 void
364 ofproto_init(const struct shash *iface_hints)
365 {
366 struct shash_node *node;
367 size_t i;
368
369 ofproto_class_register(&ofproto_dpif_class);
370
371 /* Make a local copy, since we don't own 'iface_hints' elements. */
372 SHASH_FOR_EACH(node, iface_hints) {
373 const struct iface_hint *orig_hint = node->data;
374 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
375 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
376
377 new_hint->br_name = xstrdup(orig_hint->br_name);
378 new_hint->br_type = xstrdup(br_type);
379 new_hint->ofp_port = orig_hint->ofp_port;
380
381 shash_add(&init_ofp_ports, node->name, new_hint);
382 }
383
384 for (i = 0; i < n_ofproto_classes; i++) {
385 ofproto_classes[i]->init(&init_ofp_ports);
386 }
387
388 ofproto_unixctl_init();
389 }
390
391 /* 'type' should be a normalized datapath type, as returned by
392 * ofproto_normalize_type(). Returns the corresponding ofproto_class
393 * structure, or a null pointer if there is none registered for 'type'. */
394 static const struct ofproto_class *
395 ofproto_class_find__(const char *type)
396 {
397 size_t i;
398
399 for (i = 0; i < n_ofproto_classes; i++) {
400 const struct ofproto_class *class = ofproto_classes[i];
401 struct sset types;
402 bool found;
403
404 sset_init(&types);
405 class->enumerate_types(&types);
406 found = sset_contains(&types, type);
407 sset_destroy(&types);
408
409 if (found) {
410 return class;
411 }
412 }
413 VLOG_WARN("unknown datapath type %s", type);
414 return NULL;
415 }
416
417 /* Registers a new ofproto class. After successful registration, new ofprotos
418 * of that type can be created using ofproto_create(). */
419 int
420 ofproto_class_register(const struct ofproto_class *new_class)
421 {
422 size_t i;
423
424 for (i = 0; i < n_ofproto_classes; i++) {
425 if (ofproto_classes[i] == new_class) {
426 return EEXIST;
427 }
428 }
429
430 if (n_ofproto_classes >= allocated_ofproto_classes) {
431 ofproto_classes = x2nrealloc(ofproto_classes,
432 &allocated_ofproto_classes,
433 sizeof *ofproto_classes);
434 }
435 ofproto_classes[n_ofproto_classes++] = new_class;
436 return 0;
437 }
438
439 /* Unregisters a datapath provider. 'type' must have been previously
440 * registered and not currently be in use by any ofprotos. After
441 * unregistration new datapaths of that type cannot be opened using
442 * ofproto_create(). */
443 int
444 ofproto_class_unregister(const struct ofproto_class *class)
445 {
446 size_t i;
447
448 for (i = 0; i < n_ofproto_classes; i++) {
449 if (ofproto_classes[i] == class) {
450 for (i++; i < n_ofproto_classes; i++) {
451 ofproto_classes[i - 1] = ofproto_classes[i];
452 }
453 n_ofproto_classes--;
454 return 0;
455 }
456 }
457 VLOG_WARN("attempted to unregister an ofproto class that is not "
458 "registered");
459 return EAFNOSUPPORT;
460 }
461
462 /* Clears 'types' and enumerates all registered ofproto types into it. The
463 * caller must first initialize the sset. */
464 void
465 ofproto_enumerate_types(struct sset *types)
466 {
467 size_t i;
468
469 sset_clear(types);
470 for (i = 0; i < n_ofproto_classes; i++) {
471 ofproto_classes[i]->enumerate_types(types);
472 }
473 }
474
475 /* Returns the fully spelled out name for the given ofproto 'type'.
476 *
477 * Normalized type string can be compared with strcmp(). Unnormalized type
478 * string might be the same even if they have different spellings. */
479 const char *
480 ofproto_normalize_type(const char *type)
481 {
482 return type && type[0] ? type : "system";
483 }
484
485 /* Clears 'names' and enumerates the names of all known created ofprotos with
486 * the given 'type'. The caller must first initialize the sset. Returns 0 if
487 * successful, otherwise a positive errno value.
488 *
489 * Some kinds of datapaths might not be practically enumerable. This is not
490 * considered an error. */
491 int
492 ofproto_enumerate_names(const char *type, struct sset *names)
493 {
494 const struct ofproto_class *class = ofproto_class_find__(type);
495 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
496 }
497
498 static void
499 ofproto_bump_tables_version(struct ofproto *ofproto)
500 {
501 ++ofproto->tables_version;
502 ofproto->ofproto_class->set_tables_version(ofproto,
503 ofproto->tables_version);
504 }
505
506 int
507 ofproto_create(const char *datapath_name, const char *datapath_type,
508 struct ofproto **ofprotop)
509 {
510 const struct ofproto_class *class;
511 struct ofproto *ofproto;
512 int error;
513 int i;
514
515 *ofprotop = NULL;
516
517 datapath_type = ofproto_normalize_type(datapath_type);
518 class = ofproto_class_find__(datapath_type);
519 if (!class) {
520 VLOG_WARN("could not create datapath %s of unknown type %s",
521 datapath_name, datapath_type);
522 return EAFNOSUPPORT;
523 }
524
525 ofproto = class->alloc();
526 if (!ofproto) {
527 VLOG_ERR("failed to allocate datapath %s of type %s",
528 datapath_name, datapath_type);
529 return ENOMEM;
530 }
531
532 /* Initialize. */
533 ovs_mutex_lock(&ofproto_mutex);
534 memset(ofproto, 0, sizeof *ofproto);
535 ofproto->ofproto_class = class;
536 ofproto->name = xstrdup(datapath_name);
537 ofproto->type = xstrdup(datapath_type);
538 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
539 hash_string(ofproto->name, 0));
540 ofproto->datapath_id = 0;
541 ofproto->forward_bpdu = false;
542 ofproto->fallback_dpid = pick_fallback_dpid();
543 ofproto->mfr_desc = NULL;
544 ofproto->hw_desc = NULL;
545 ofproto->sw_desc = NULL;
546 ofproto->serial_desc = NULL;
547 ofproto->dp_desc = NULL;
548 ofproto->frag_handling = OFPUTIL_FRAG_NORMAL;
549 hmap_init(&ofproto->ports);
550 hmap_init(&ofproto->ofport_usage);
551 shash_init(&ofproto->port_by_name);
552 simap_init(&ofproto->ofp_requests);
553 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
554 ofproto->eviction_group_timer = LLONG_MIN;
555 ofproto->tables = NULL;
556 ofproto->n_tables = 0;
557 ofproto->tables_version = CLS_MIN_VERSION;
558 hindex_init(&ofproto->cookies);
559 hmap_init(&ofproto->learned_cookies);
560 list_init(&ofproto->expirable);
561 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
562 guarded_list_init(&ofproto->rule_executes);
563 ofproto->vlan_bitmap = NULL;
564 ofproto->vlans_changed = false;
565 ofproto->min_mtu = INT_MAX;
566 ovs_rwlock_init(&ofproto->groups_rwlock);
567 hmap_init(&ofproto->groups);
568 ovs_mutex_unlock(&ofproto_mutex);
569 ofproto->ogf.types = 0xf;
570 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
571 OFPGFC_SELECT_WEIGHT;
572 for (i = 0; i < 4; i++) {
573 ofproto->ogf.max_groups[i] = OFPG_MAX;
574 ofproto->ogf.ofpacts[i] = (UINT64_C(1) << N_OFPACTS) - 1;
575 }
576 tun_metadata_init();
577
578 error = ofproto->ofproto_class->construct(ofproto);
579 if (error) {
580 VLOG_ERR("failed to open datapath %s: %s",
581 datapath_name, ovs_strerror(error));
582 connmgr_destroy(ofproto->connmgr);
583 ofproto_destroy__(ofproto);
584 return error;
585 }
586
587 /* Check that hidden tables, if any, are at the end. */
588 ovs_assert(ofproto->n_tables);
589 for (i = 0; i + 1 < ofproto->n_tables; i++) {
590 enum oftable_flags flags = ofproto->tables[i].flags;
591 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
592
593 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
594 }
595
596 ofproto->datapath_id = pick_datapath_id(ofproto);
597 init_ports(ofproto);
598
599 /* Initialize meters table. */
600 if (ofproto->ofproto_class->meter_get_features) {
601 ofproto->ofproto_class->meter_get_features(ofproto,
602 &ofproto->meter_features);
603 } else {
604 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
605 }
606 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
607 * sizeof(struct meter *));
608
609 /* Set the initial tables version. */
610 ofproto_bump_tables_version(ofproto);
611
612 *ofprotop = ofproto;
613 return 0;
614 }
615
616 /* Must be called (only) by an ofproto implementation in its constructor
617 * function. See the large comment on 'construct' in struct ofproto_class for
618 * details. */
619 void
620 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
621 {
622 struct oftable *table;
623
624 ovs_assert(!ofproto->n_tables);
625 ovs_assert(n_tables >= 1 && n_tables <= 255);
626
627 ofproto->n_tables = n_tables;
628 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
629 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
630 oftable_init(table);
631 }
632 }
633
634 /* To be optionally called (only) by an ofproto implementation in its
635 * constructor function. See the large comment on 'construct' in struct
636 * ofproto_class for details.
637 *
638 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
639 * will then ensure that actions passed into the ofproto implementation will
640 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
641 * function is not called, there will be no such restriction.
642 *
643 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
644 * the 'max_ports' restriction. */
645 void
646 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
647 {
648 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
649 ofproto->max_ports = max_ports;
650 }
651
652 uint64_t
653 ofproto_get_datapath_id(const struct ofproto *ofproto)
654 {
655 return ofproto->datapath_id;
656 }
657
658 void
659 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
660 {
661 uint64_t old_dpid = p->datapath_id;
662 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
663 if (p->datapath_id != old_dpid) {
664 /* Force all active connections to reconnect, since there is no way to
665 * notify a controller that the datapath ID has changed. */
666 ofproto_reconnect_controllers(p);
667 }
668 }
669
670 void
671 ofproto_set_controllers(struct ofproto *p,
672 const struct ofproto_controller *controllers,
673 size_t n_controllers, uint32_t allowed_versions)
674 {
675 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
676 allowed_versions);
677 }
678
679 void
680 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
681 {
682 connmgr_set_fail_mode(p->connmgr, fail_mode);
683 }
684
685 /* Drops the connections between 'ofproto' and all of its controllers, forcing
686 * them to reconnect. */
687 void
688 ofproto_reconnect_controllers(struct ofproto *ofproto)
689 {
690 connmgr_reconnect(ofproto->connmgr);
691 }
692
693 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
694 * in-band control should guarantee access, in the same way that in-band
695 * control guarantees access to OpenFlow controllers. */
696 void
697 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
698 const struct sockaddr_in *extras, size_t n)
699 {
700 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
701 }
702
703 /* Sets the OpenFlow queue used by flows set up by in-band control on
704 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
705 * flows will use the default queue. */
706 void
707 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
708 {
709 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
710 }
711
712 /* Sets the number of flows at which eviction from the kernel flow table
713 * will occur. */
714 void
715 ofproto_set_flow_limit(unsigned limit)
716 {
717 ofproto_flow_limit = limit;
718 }
719
720 /* Sets the maximum idle time for flows in the datapath before they are
721 * expired. */
722 void
723 ofproto_set_max_idle(unsigned max_idle)
724 {
725 ofproto_max_idle = max_idle;
726 }
727
728 /* If forward_bpdu is true, the NORMAL action will forward frames with
729 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
730 * the NORMAL action will drop these frames. */
731 void
732 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
733 {
734 bool old_val = ofproto->forward_bpdu;
735 ofproto->forward_bpdu = forward_bpdu;
736 if (old_val != ofproto->forward_bpdu) {
737 if (ofproto->ofproto_class->forward_bpdu_changed) {
738 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
739 }
740 }
741 }
742
743 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
744 * 'idle_time', in seconds, and the maximum number of MAC table entries to
745 * 'max_entries'. */
746 void
747 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
748 size_t max_entries)
749 {
750 if (ofproto->ofproto_class->set_mac_table_config) {
751 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
752 max_entries);
753 }
754 }
755
756 /* Multicast snooping configuration. */
757
758 /* Configures multicast snooping on 'ofproto' using the settings
759 * defined in 's'. If 's' is NULL, disables multicast snooping.
760 *
761 * Returns 0 if successful, otherwise a positive errno value. */
762 int
763 ofproto_set_mcast_snooping(struct ofproto *ofproto,
764 const struct ofproto_mcast_snooping_settings *s)
765 {
766 return (ofproto->ofproto_class->set_mcast_snooping
767 ? ofproto->ofproto_class->set_mcast_snooping(ofproto, s)
768 : EOPNOTSUPP);
769 }
770
771 /* Configures multicast snooping flood settings on 'ofp_port' of 'ofproto'.
772 *
773 * Returns 0 if successful, otherwise a positive errno value.*/
774 int
775 ofproto_port_set_mcast_snooping(struct ofproto *ofproto, void *aux,
776 const struct ofproto_mcast_snooping_port_settings *s)
777 {
778 return (ofproto->ofproto_class->set_mcast_snooping_port
779 ? ofproto->ofproto_class->set_mcast_snooping_port(ofproto, aux, s)
780 : EOPNOTSUPP);
781 }
782
783 void
784 ofproto_set_n_dpdk_rxqs(int n_rxqs)
785 {
786 n_dpdk_rxqs = MAX(n_rxqs, 0);
787 }
788
789 void
790 ofproto_set_cpu_mask(const char *cmask)
791 {
792 free(pmd_cpu_mask);
793
794 pmd_cpu_mask = cmask ? xstrdup(cmask) : NULL;
795 }
796
797 void
798 ofproto_set_threads(int n_handlers_, int n_revalidators_)
799 {
800 int threads = MAX(count_cpu_cores(), 2);
801
802 n_revalidators = MAX(n_revalidators_, 0);
803 n_handlers = MAX(n_handlers_, 0);
804
805 if (!n_revalidators) {
806 n_revalidators = n_handlers
807 ? MAX(threads - (int) n_handlers, 1)
808 : threads / 4 + 1;
809 }
810
811 if (!n_handlers) {
812 n_handlers = MAX(threads - (int) n_revalidators, 1);
813 }
814 }
815
816 void
817 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
818 {
819 free(p->dp_desc);
820 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
821 }
822
823 int
824 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
825 {
826 return connmgr_set_snoops(ofproto->connmgr, snoops);
827 }
828
829 int
830 ofproto_set_netflow(struct ofproto *ofproto,
831 const struct netflow_options *nf_options)
832 {
833 if (nf_options && sset_is_empty(&nf_options->collectors)) {
834 nf_options = NULL;
835 }
836
837 if (ofproto->ofproto_class->set_netflow) {
838 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
839 } else {
840 return nf_options ? EOPNOTSUPP : 0;
841 }
842 }
843
844 int
845 ofproto_set_sflow(struct ofproto *ofproto,
846 const struct ofproto_sflow_options *oso)
847 {
848 if (oso && sset_is_empty(&oso->targets)) {
849 oso = NULL;
850 }
851
852 if (ofproto->ofproto_class->set_sflow) {
853 return ofproto->ofproto_class->set_sflow(ofproto, oso);
854 } else {
855 return oso ? EOPNOTSUPP : 0;
856 }
857 }
858
859 int
860 ofproto_set_ipfix(struct ofproto *ofproto,
861 const struct ofproto_ipfix_bridge_exporter_options *bo,
862 const struct ofproto_ipfix_flow_exporter_options *fo,
863 size_t n_fo)
864 {
865 if (ofproto->ofproto_class->set_ipfix) {
866 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
867 } else {
868 return (bo || fo) ? EOPNOTSUPP : 0;
869 }
870 }
871
872 void
873 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
874 {
875 flow_restore_wait = flow_restore_wait_db;
876 }
877
878 bool
879 ofproto_get_flow_restore_wait(void)
880 {
881 return flow_restore_wait;
882 }
883
884 \f
885 /* Spanning Tree Protocol (STP) configuration. */
886
887 /* Configures STP on 'ofproto' using the settings defined in 's'. If
888 * 's' is NULL, disables STP.
889 *
890 * Returns 0 if successful, otherwise a positive errno value. */
891 int
892 ofproto_set_stp(struct ofproto *ofproto,
893 const struct ofproto_stp_settings *s)
894 {
895 return (ofproto->ofproto_class->set_stp
896 ? ofproto->ofproto_class->set_stp(ofproto, s)
897 : EOPNOTSUPP);
898 }
899
900 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
901 * 'enabled' member of 's' is false, then the other members are not
902 * meaningful.
903 *
904 * Returns 0 if successful, otherwise a positive errno value. */
905 int
906 ofproto_get_stp_status(struct ofproto *ofproto,
907 struct ofproto_stp_status *s)
908 {
909 return (ofproto->ofproto_class->get_stp_status
910 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
911 : EOPNOTSUPP);
912 }
913
914 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
915 * in 's'. The caller is responsible for assigning STP port numbers
916 * (using the 'port_num' member in the range of 1 through 255, inclusive)
917 * and ensuring there are no duplicates. If the 's' is NULL, then STP
918 * is disabled on the port.
919 *
920 * Returns 0 if successful, otherwise a positive errno value.*/
921 int
922 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
923 const struct ofproto_port_stp_settings *s)
924 {
925 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
926 if (!ofport) {
927 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
928 ofproto->name, ofp_port);
929 return ENODEV;
930 }
931
932 return (ofproto->ofproto_class->set_stp_port
933 ? ofproto->ofproto_class->set_stp_port(ofport, s)
934 : EOPNOTSUPP);
935 }
936
937 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
938 * 's'. If the 'enabled' member in 's' is false, then the other members
939 * are not meaningful.
940 *
941 * Returns 0 if successful, otherwise a positive errno value.*/
942 int
943 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
944 struct ofproto_port_stp_status *s)
945 {
946 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
947 if (!ofport) {
948 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
949 "port %"PRIu16, ofproto->name, ofp_port);
950 return ENODEV;
951 }
952
953 return (ofproto->ofproto_class->get_stp_port_status
954 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
955 : EOPNOTSUPP);
956 }
957
958 /* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
959 * 's'. If the 'enabled' member in 's' is false, then the other members
960 * are not meaningful.
961 *
962 * Returns 0 if successful, otherwise a positive errno value.*/
963 int
964 ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
965 struct ofproto_port_stp_stats *s)
966 {
967 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
968 if (!ofport) {
969 VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
970 "port %"PRIu16, ofproto->name, ofp_port);
971 return ENODEV;
972 }
973
974 return (ofproto->ofproto_class->get_stp_port_stats
975 ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
976 : EOPNOTSUPP);
977 }
978
979 /* Rapid Spanning Tree Protocol (RSTP) configuration. */
980
981 /* Configures RSTP on 'ofproto' using the settings defined in 's'. If
982 * 's' is NULL, disables RSTP.
983 *
984 * Returns 0 if successful, otherwise a positive errno value. */
985 int
986 ofproto_set_rstp(struct ofproto *ofproto,
987 const struct ofproto_rstp_settings *s)
988 {
989 if (!ofproto->ofproto_class->set_rstp) {
990 return EOPNOTSUPP;
991 }
992 ofproto->ofproto_class->set_rstp(ofproto, s);
993 return 0;
994 }
995
996 /* Retrieves RSTP status of 'ofproto' and stores it in 's'. If the
997 * 'enabled' member of 's' is false, then the other members are not
998 * meaningful.
999 *
1000 * Returns 0 if successful, otherwise a positive errno value. */
1001 int
1002 ofproto_get_rstp_status(struct ofproto *ofproto,
1003 struct ofproto_rstp_status *s)
1004 {
1005 if (!ofproto->ofproto_class->get_rstp_status) {
1006 return EOPNOTSUPP;
1007 }
1008 ofproto->ofproto_class->get_rstp_status(ofproto, s);
1009 return 0;
1010 }
1011
1012 /* Configures RSTP on 'ofp_port' of 'ofproto' using the settings defined
1013 * in 's'. The caller is responsible for assigning RSTP port numbers
1014 * (using the 'port_num' member in the range of 1 through 255, inclusive)
1015 * and ensuring there are no duplicates. If the 's' is NULL, then RSTP
1016 * is disabled on the port.
1017 *
1018 * Returns 0 if successful, otherwise a positive errno value.*/
1019 int
1020 ofproto_port_set_rstp(struct ofproto *ofproto, ofp_port_t ofp_port,
1021 const struct ofproto_port_rstp_settings *s)
1022 {
1023 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1024 if (!ofport) {
1025 VLOG_WARN("%s: cannot configure RSTP on nonexistent port %"PRIu16,
1026 ofproto->name, ofp_port);
1027 return ENODEV;
1028 }
1029
1030 if (!ofproto->ofproto_class->set_rstp_port) {
1031 return EOPNOTSUPP;
1032 }
1033 ofproto->ofproto_class->set_rstp_port(ofport, s);
1034 return 0;
1035 }
1036
1037 /* Retrieves RSTP port status of 'ofp_port' on 'ofproto' and stores it in
1038 * 's'. If the 'enabled' member in 's' is false, then the other members
1039 * are not meaningful.
1040 *
1041 * Returns 0 if successful, otherwise a positive errno value.*/
1042 int
1043 ofproto_port_get_rstp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1044 struct ofproto_port_rstp_status *s)
1045 {
1046 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1047 if (!ofport) {
1048 VLOG_WARN_RL(&rl, "%s: cannot get RSTP status on nonexistent "
1049 "port %"PRIu16, ofproto->name, ofp_port);
1050 return ENODEV;
1051 }
1052
1053 if (!ofproto->ofproto_class->get_rstp_port_status) {
1054 return EOPNOTSUPP;
1055 }
1056 ofproto->ofproto_class->get_rstp_port_status(ofport, s);
1057 return 0;
1058 }
1059 \f
1060 /* Queue DSCP configuration. */
1061
1062 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1063 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
1064 * to implement QoS. Instead, it is used to implement features which require
1065 * knowledge of what queues exist on a port, and some basic information about
1066 * them.
1067 *
1068 * Returns 0 if successful, otherwise a positive errno value. */
1069 int
1070 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
1071 const struct ofproto_port_queue *queues,
1072 size_t n_queues)
1073 {
1074 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1075
1076 if (!ofport) {
1077 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
1078 ofproto->name, ofp_port);
1079 return ENODEV;
1080 }
1081
1082 return (ofproto->ofproto_class->set_queues
1083 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
1084 : EOPNOTSUPP);
1085 }
1086 \f
1087 /* LLDP configuration. */
1088 void
1089 ofproto_port_set_lldp(struct ofproto *ofproto,
1090 ofp_port_t ofp_port,
1091 const struct smap *cfg)
1092 {
1093 struct ofport *ofport;
1094 int error;
1095
1096 ofport = ofproto_get_port(ofproto, ofp_port);
1097 if (!ofport) {
1098 VLOG_WARN("%s: cannot configure LLDP on nonexistent port %"PRIu16,
1099 ofproto->name, ofp_port);
1100 return;
1101 }
1102 error = (ofproto->ofproto_class->set_lldp
1103 ? ofproto->ofproto_class->set_lldp(ofport, cfg)
1104 : EOPNOTSUPP);
1105 if (error) {
1106 VLOG_WARN("%s: lldp configuration on port %"PRIu16" (%s) failed (%s)",
1107 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1108 ovs_strerror(error));
1109 }
1110 }
1111
1112 int
1113 ofproto_set_aa(struct ofproto *ofproto, void *aux OVS_UNUSED,
1114 const struct aa_settings *s)
1115 {
1116 if (!ofproto->ofproto_class->set_aa) {
1117 return EOPNOTSUPP;
1118 }
1119 ofproto->ofproto_class->set_aa(ofproto, s);
1120 return 0;
1121 }
1122
1123 int
1124 ofproto_aa_mapping_register(struct ofproto *ofproto, void *aux,
1125 const struct aa_mapping_settings *s)
1126 {
1127 if (!ofproto->ofproto_class->aa_mapping_set) {
1128 return EOPNOTSUPP;
1129 }
1130 ofproto->ofproto_class->aa_mapping_set(ofproto, aux, s);
1131 return 0;
1132 }
1133
1134 int
1135 ofproto_aa_mapping_unregister(struct ofproto *ofproto, void *aux)
1136 {
1137 if (!ofproto->ofproto_class->aa_mapping_unset) {
1138 return EOPNOTSUPP;
1139 }
1140 ofproto->ofproto_class->aa_mapping_unset(ofproto, aux);
1141 return 0;
1142 }
1143
1144 int
1145 ofproto_aa_vlan_get_queued(struct ofproto *ofproto,
1146 struct ovs_list *list)
1147 {
1148 if (!ofproto->ofproto_class->aa_vlan_get_queued) {
1149 return EOPNOTSUPP;
1150 }
1151 ofproto->ofproto_class->aa_vlan_get_queued(ofproto, list);
1152 return 0;
1153 }
1154
1155 unsigned int
1156 ofproto_aa_vlan_get_queue_size(struct ofproto *ofproto)
1157 {
1158 if (!ofproto->ofproto_class->aa_vlan_get_queue_size) {
1159 return EOPNOTSUPP;
1160 }
1161 return ofproto->ofproto_class->aa_vlan_get_queue_size(ofproto);
1162 }
1163
1164 /* Connectivity Fault Management configuration. */
1165
1166 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
1167 void
1168 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
1169 {
1170 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1171 if (ofport && ofproto->ofproto_class->set_cfm) {
1172 ofproto->ofproto_class->set_cfm(ofport, NULL);
1173 }
1174 }
1175
1176 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
1177 * basic configuration from the configuration members in 'cfm', and the remote
1178 * maintenance point ID from remote_mpid. Ignores the statistics members of
1179 * 'cfm'.
1180 *
1181 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
1182 void
1183 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
1184 const struct cfm_settings *s)
1185 {
1186 struct ofport *ofport;
1187 int error;
1188
1189 ofport = ofproto_get_port(ofproto, ofp_port);
1190 if (!ofport) {
1191 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
1192 ofproto->name, ofp_port);
1193 return;
1194 }
1195
1196 /* XXX: For configuration simplicity, we only support one remote_mpid
1197 * outside of the CFM module. It's not clear if this is the correct long
1198 * term solution or not. */
1199 error = (ofproto->ofproto_class->set_cfm
1200 ? ofproto->ofproto_class->set_cfm(ofport, s)
1201 : EOPNOTSUPP);
1202 if (error) {
1203 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
1204 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1205 ovs_strerror(error));
1206 }
1207 }
1208
1209 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
1210 * 'ofproto' does not have a port 'ofp_port'. */
1211 void
1212 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
1213 const struct smap *cfg)
1214 {
1215 struct ofport *ofport;
1216 int error;
1217
1218 ofport = ofproto_get_port(ofproto, ofp_port);
1219 if (!ofport) {
1220 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
1221 ofproto->name, ofp_port);
1222 return;
1223 }
1224
1225 error = (ofproto->ofproto_class->set_bfd
1226 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
1227 : EOPNOTSUPP);
1228 if (error) {
1229 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
1230 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1231 ovs_strerror(error));
1232 }
1233 }
1234
1235 /* Checks the status change of BFD on 'ofport'.
1236 *
1237 * Returns true if 'ofproto_class' does not support 'bfd_status_changed'. */
1238 bool
1239 ofproto_port_bfd_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
1240 {
1241 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1242 return (ofport && ofproto->ofproto_class->bfd_status_changed
1243 ? ofproto->ofproto_class->bfd_status_changed(ofport)
1244 : true);
1245 }
1246
1247 /* Populates 'status' with the status of BFD on 'ofport'. Returns 0 on
1248 * success. Returns a positive errno otherwise. Has no effect if 'ofp_port'
1249 * is not an OpenFlow port in 'ofproto'.
1250 *
1251 * The caller must provide and own '*status'. */
1252 int
1253 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1254 struct smap *status)
1255 {
1256 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1257 return (ofport && ofproto->ofproto_class->get_bfd_status
1258 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1259 : EOPNOTSUPP);
1260 }
1261
1262 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1263 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1264 * 0 if LACP partner information is not current (generally indicating a
1265 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1266 int
1267 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1268 {
1269 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1270 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1271 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1272 : -1);
1273 }
1274
1275 int
1276 ofproto_port_get_lacp_stats(const struct ofport *port, struct lacp_slave_stats *stats)
1277 {
1278 struct ofproto *ofproto = port->ofproto;
1279 int error;
1280
1281 if (ofproto->ofproto_class->port_get_lacp_stats) {
1282 error = ofproto->ofproto_class->port_get_lacp_stats(port, stats);
1283 } else {
1284 error = EOPNOTSUPP;
1285 }
1286
1287 return error;
1288 }
1289 \f
1290 /* Bundles. */
1291
1292 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1293 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1294 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1295 * configuration plus, if there is more than one slave, a bonding
1296 * configuration.
1297 *
1298 * If 'aux' is already registered then this function updates its configuration
1299 * to 's'. Otherwise, this function registers a new bundle.
1300 *
1301 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1302 * port. */
1303 int
1304 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1305 const struct ofproto_bundle_settings *s)
1306 {
1307 return (ofproto->ofproto_class->bundle_set
1308 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1309 : EOPNOTSUPP);
1310 }
1311
1312 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1313 * If no such bundle has been registered, this has no effect. */
1314 int
1315 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1316 {
1317 return ofproto_bundle_register(ofproto, aux, NULL);
1318 }
1319
1320 \f
1321 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1322 * If 'aux' is already registered then this function updates its configuration
1323 * to 's'. Otherwise, this function registers a new mirror. */
1324 int
1325 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1326 const struct ofproto_mirror_settings *s)
1327 {
1328 return (ofproto->ofproto_class->mirror_set
1329 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1330 : EOPNOTSUPP);
1331 }
1332
1333 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1334 * If no mirror has been registered, this has no effect. */
1335 int
1336 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1337 {
1338 return ofproto_mirror_register(ofproto, aux, NULL);
1339 }
1340
1341 /* Retrieves statistics from mirror associated with client data pointer
1342 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1343 * 'bytes', respectively. If a particular counters is not supported,
1344 * the appropriate argument is set to UINT64_MAX.
1345 */
1346 int
1347 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1348 uint64_t *packets, uint64_t *bytes)
1349 {
1350 if (!ofproto->ofproto_class->mirror_get_stats) {
1351 *packets = *bytes = UINT64_MAX;
1352 return EOPNOTSUPP;
1353 }
1354
1355 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1356 packets, bytes);
1357 }
1358
1359 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1360 * which all packets are flooded, instead of using MAC learning. If
1361 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1362 *
1363 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1364 * port. */
1365 int
1366 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1367 {
1368 return (ofproto->ofproto_class->set_flood_vlans
1369 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1370 : EOPNOTSUPP);
1371 }
1372
1373 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1374 * output for a mirror. */
1375 bool
1376 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1377 {
1378 return (ofproto->ofproto_class->is_mirror_output_bundle
1379 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1380 : false);
1381 }
1382 \f
1383 /* Configuration of OpenFlow tables. */
1384
1385 /* Returns the number of OpenFlow tables in 'ofproto'. */
1386 int
1387 ofproto_get_n_tables(const struct ofproto *ofproto)
1388 {
1389 return ofproto->n_tables;
1390 }
1391
1392 /* Returns the number of Controller visible OpenFlow tables
1393 * in 'ofproto'. This number will exclude Hidden tables.
1394 * This funtion's return value should be less or equal to that of
1395 * ofproto_get_n_tables() . */
1396 uint8_t
1397 ofproto_get_n_visible_tables(const struct ofproto *ofproto)
1398 {
1399 uint8_t n = ofproto->n_tables;
1400
1401 /* Count only non-hidden tables in the number of tables. (Hidden tables,
1402 * if present, are always at the end.) */
1403 while(n && (ofproto->tables[n - 1].flags & OFTABLE_HIDDEN)) {
1404 n--;
1405 }
1406
1407 return n;
1408 }
1409
1410 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1411 * settings from 's'. 'table_id' must be in the range 0 through the number of
1412 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1413 *
1414 * For read-only tables, only the name may be configured. */
1415 void
1416 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1417 const struct ofproto_table_settings *s)
1418 {
1419 struct oftable *table;
1420
1421 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1422 table = &ofproto->tables[table_id];
1423
1424 oftable_set_name(table, s->name);
1425
1426 if (table->flags & OFTABLE_READONLY) {
1427 return;
1428 }
1429
1430 if (classifier_set_prefix_fields(&table->cls,
1431 s->prefix_fields, s->n_prefix_fields)) {
1432 /* XXX: Trigger revalidation. */
1433 }
1434
1435 ovs_mutex_lock(&ofproto_mutex);
1436 unsigned int new_eviction = (s->enable_eviction
1437 ? table->eviction | EVICTION_CLIENT
1438 : table->eviction & ~EVICTION_CLIENT);
1439 oftable_configure_eviction(table, new_eviction, s->groups, s->n_groups);
1440 table->max_flows = s->max_flows;
1441 evict_rules_from_table(table);
1442 ovs_mutex_unlock(&ofproto_mutex);
1443 }
1444 \f
1445 bool
1446 ofproto_has_snoops(const struct ofproto *ofproto)
1447 {
1448 return connmgr_has_snoops(ofproto->connmgr);
1449 }
1450
1451 void
1452 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1453 {
1454 connmgr_get_snoops(ofproto->connmgr, snoops);
1455 }
1456
1457 /* Deletes 'rule' from 'ofproto'.
1458 *
1459 * Within an ofproto implementation, this function allows an ofproto
1460 * implementation to destroy any rules that remain when its ->destruct()
1461 * function is called. This function is not suitable for use elsewhere in an
1462 * ofproto implementation.
1463 *
1464 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1465 * Cycle" in ofproto-provider.h. */
1466 void
1467 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1468 OVS_EXCLUDED(ofproto_mutex)
1469 {
1470 /* This skips the ofmonitor and flow-removed notifications because the
1471 * switch is being deleted and any OpenFlow channels have been or soon will
1472 * be killed. */
1473 ovs_mutex_lock(&ofproto_mutex);
1474
1475 if (!rule->removed) {
1476 /* Make sure there is no postponed removal of the rule. */
1477 ovs_assert(cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
1478
1479 if (!classifier_remove(&rule->ofproto->tables[rule->table_id].cls,
1480 &rule->cr)) {
1481 OVS_NOT_REACHED();
1482 }
1483 ofproto_rule_remove__(rule->ofproto, rule);
1484 ofproto->ofproto_class->rule_delete(rule);
1485 ofproto_rule_unref(rule);
1486 }
1487 ovs_mutex_unlock(&ofproto_mutex);
1488 }
1489
1490 static void
1491 ofproto_flush__(struct ofproto *ofproto)
1492 OVS_EXCLUDED(ofproto_mutex)
1493 {
1494 struct oftable *table;
1495
1496 /* This will flush all datapath flows. */
1497 if (ofproto->ofproto_class->flush) {
1498 ofproto->ofproto_class->flush(ofproto);
1499 }
1500
1501 /* XXX: There is a small race window here, where new datapath flows can be
1502 * created by upcall handlers based on the existing flow table. We can not
1503 * call ofproto class flush while holding 'ofproto_mutex' to prevent this,
1504 * as then we could deadlock on syncing with the handler threads waiting on
1505 * the same mutex. */
1506
1507 ovs_mutex_lock(&ofproto_mutex);
1508 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1509 struct rule_collection rules;
1510 struct rule *rule;
1511
1512 if (table->flags & OFTABLE_HIDDEN) {
1513 continue;
1514 }
1515
1516 rule_collection_init(&rules);
1517
1518 CLS_FOR_EACH (rule, cr, &table->cls) {
1519 rule_collection_add(&rules, rule);
1520 }
1521 delete_flows__(&rules, OFPRR_DELETE, NULL);
1522 }
1523 /* XXX: Concurrent handler threads may insert new learned flows based on
1524 * learn actions of the now deleted flows right after we release
1525 * 'ofproto_mutex'. */
1526 ovs_mutex_unlock(&ofproto_mutex);
1527 }
1528
1529 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1530
1531 static void
1532 ofproto_destroy__(struct ofproto *ofproto)
1533 OVS_EXCLUDED(ofproto_mutex)
1534 {
1535 struct oftable *table;
1536
1537 destroy_rule_executes(ofproto);
1538
1539 guarded_list_destroy(&ofproto->rule_executes);
1540 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1541 hmap_destroy(&ofproto->groups);
1542
1543 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1544 free(ofproto->name);
1545 free(ofproto->type);
1546 free(ofproto->mfr_desc);
1547 free(ofproto->hw_desc);
1548 free(ofproto->sw_desc);
1549 free(ofproto->serial_desc);
1550 free(ofproto->dp_desc);
1551 hmap_destroy(&ofproto->ports);
1552 hmap_destroy(&ofproto->ofport_usage);
1553 shash_destroy(&ofproto->port_by_name);
1554 simap_destroy(&ofproto->ofp_requests);
1555
1556 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1557 oftable_destroy(table);
1558 }
1559 free(ofproto->tables);
1560
1561 ovs_assert(hindex_is_empty(&ofproto->cookies));
1562 hindex_destroy(&ofproto->cookies);
1563
1564 ovs_assert(hmap_is_empty(&ofproto->learned_cookies));
1565 hmap_destroy(&ofproto->learned_cookies);
1566
1567 free(ofproto->vlan_bitmap);
1568
1569 ofproto->ofproto_class->dealloc(ofproto);
1570 }
1571
1572 /* Destroying rules is doubly deferred, must have 'ofproto' around for them.
1573 * - 1st we defer the removal of the rules from the classifier
1574 * - 2nd we defer the actual destruction of the rules. */
1575 static void
1576 ofproto_destroy_defer__(struct ofproto *ofproto)
1577 OVS_EXCLUDED(ofproto_mutex)
1578 {
1579 ovsrcu_postpone(ofproto_destroy__, ofproto);
1580 }
1581
1582 void
1583 ofproto_destroy(struct ofproto *p)
1584 OVS_EXCLUDED(ofproto_mutex)
1585 {
1586 struct ofport *ofport, *next_ofport;
1587 struct ofport_usage *usage, *next_usage;
1588
1589 if (!p) {
1590 return;
1591 }
1592
1593 if (p->meters) {
1594 meter_delete(p, 1, p->meter_features.max_meters);
1595 p->meter_features.max_meters = 0;
1596 free(p->meters);
1597 p->meters = NULL;
1598 }
1599
1600 ofproto_flush__(p);
1601 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1602 ofport_destroy(ofport);
1603 }
1604
1605 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1606 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1607 free(usage);
1608 }
1609
1610 p->ofproto_class->destruct(p);
1611
1612 /* We should not postpone this because it involves deleting a listening
1613 * socket which we may want to reopen soon. 'connmgr' should not be used
1614 * by other threads */
1615 connmgr_destroy(p->connmgr);
1616
1617 /* Destroying rules is deferred, must have 'ofproto' around for them. */
1618 ovsrcu_postpone(ofproto_destroy_defer__, p);
1619 }
1620
1621 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1622 * kernel datapath, for example, this destroys the datapath in the kernel, and
1623 * with the netdev-based datapath, it tears down the data structures that
1624 * represent the datapath.
1625 *
1626 * The datapath should not be currently open as an ofproto. */
1627 int
1628 ofproto_delete(const char *name, const char *type)
1629 {
1630 const struct ofproto_class *class = ofproto_class_find__(type);
1631 return (!class ? EAFNOSUPPORT
1632 : !class->del ? EACCES
1633 : class->del(type, name));
1634 }
1635
1636 static void
1637 process_port_change(struct ofproto *ofproto, int error, char *devname)
1638 {
1639 if (error == ENOBUFS) {
1640 reinit_ports(ofproto);
1641 } else if (!error) {
1642 update_port(ofproto, devname);
1643 free(devname);
1644 }
1645 }
1646
1647 int
1648 ofproto_type_run(const char *datapath_type)
1649 {
1650 const struct ofproto_class *class;
1651 int error;
1652
1653 datapath_type = ofproto_normalize_type(datapath_type);
1654 class = ofproto_class_find__(datapath_type);
1655
1656 error = class->type_run ? class->type_run(datapath_type) : 0;
1657 if (error && error != EAGAIN) {
1658 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1659 datapath_type, ovs_strerror(error));
1660 }
1661 return error;
1662 }
1663
1664 void
1665 ofproto_type_wait(const char *datapath_type)
1666 {
1667 const struct ofproto_class *class;
1668
1669 datapath_type = ofproto_normalize_type(datapath_type);
1670 class = ofproto_class_find__(datapath_type);
1671
1672 if (class->type_wait) {
1673 class->type_wait(datapath_type);
1674 }
1675 }
1676
1677 int
1678 ofproto_run(struct ofproto *p)
1679 {
1680 int error;
1681 uint64_t new_seq;
1682
1683 error = p->ofproto_class->run(p);
1684 if (error && error != EAGAIN) {
1685 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1686 }
1687
1688 run_rule_executes(p);
1689
1690 /* Restore the eviction group heap invariant occasionally. */
1691 if (p->eviction_group_timer < time_msec()) {
1692 size_t i;
1693
1694 p->eviction_group_timer = time_msec() + 1000;
1695
1696 for (i = 0; i < p->n_tables; i++) {
1697 struct oftable *table = &p->tables[i];
1698 struct eviction_group *evg;
1699 struct rule *rule;
1700
1701 if (!table->eviction) {
1702 continue;
1703 }
1704
1705 if (table->n_flows > 100000) {
1706 static struct vlog_rate_limit count_rl =
1707 VLOG_RATE_LIMIT_INIT(1, 1);
1708 VLOG_WARN_RL(&count_rl, "Table %"PRIuSIZE" has an excessive"
1709 " number of rules: %d", i, table->n_flows);
1710 }
1711
1712 ovs_mutex_lock(&ofproto_mutex);
1713 CLS_FOR_EACH (rule, cr, &table->cls) {
1714 if (rule->idle_timeout || rule->hard_timeout) {
1715 if (!rule->eviction_group) {
1716 eviction_group_add_rule(rule);
1717 } else {
1718 heap_raw_change(&rule->evg_node,
1719 rule_eviction_priority(p, rule));
1720 }
1721 }
1722 }
1723
1724 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1725 heap_rebuild(&evg->rules);
1726 }
1727 ovs_mutex_unlock(&ofproto_mutex);
1728 }
1729 }
1730
1731 if (p->ofproto_class->port_poll) {
1732 char *devname;
1733
1734 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1735 process_port_change(p, error, devname);
1736 }
1737 }
1738
1739 new_seq = seq_read(connectivity_seq_get());
1740 if (new_seq != p->change_seq) {
1741 struct sset devnames;
1742 const char *devname;
1743 struct ofport *ofport;
1744
1745 /* Update OpenFlow port status for any port whose netdev has changed.
1746 *
1747 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1748 * destroyed, so it's not safe to update ports directly from the
1749 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1750 * need this two-phase approach. */
1751 sset_init(&devnames);
1752 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1753 uint64_t port_change_seq;
1754
1755 port_change_seq = netdev_get_change_seq(ofport->netdev);
1756 if (ofport->change_seq != port_change_seq) {
1757 ofport->change_seq = port_change_seq;
1758 sset_add(&devnames, netdev_get_name(ofport->netdev));
1759 }
1760 }
1761 SSET_FOR_EACH (devname, &devnames) {
1762 update_port(p, devname);
1763 }
1764 sset_destroy(&devnames);
1765
1766 p->change_seq = new_seq;
1767 }
1768
1769 connmgr_run(p->connmgr, handle_openflow);
1770
1771 return error;
1772 }
1773
1774 void
1775 ofproto_wait(struct ofproto *p)
1776 {
1777 p->ofproto_class->wait(p);
1778 if (p->ofproto_class->port_poll_wait) {
1779 p->ofproto_class->port_poll_wait(p);
1780 }
1781 seq_wait(connectivity_seq_get(), p->change_seq);
1782 connmgr_wait(p->connmgr);
1783 }
1784
1785 bool
1786 ofproto_is_alive(const struct ofproto *p)
1787 {
1788 return connmgr_has_controllers(p->connmgr);
1789 }
1790
1791 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1792 * memory_report(). */
1793 void
1794 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1795 {
1796 const struct oftable *table;
1797 unsigned int n_rules;
1798
1799 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1800
1801 n_rules = 0;
1802 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1803 n_rules += table->n_flows;
1804 }
1805 simap_increase(usage, "rules", n_rules);
1806
1807 if (ofproto->ofproto_class->get_memory_usage) {
1808 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1809 }
1810
1811 connmgr_get_memory_usage(ofproto->connmgr, usage);
1812 }
1813
1814 void
1815 ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
1816 {
1817 const struct ofproto_class *class;
1818
1819 datapath_type = ofproto_normalize_type(datapath_type);
1820 class = ofproto_class_find__(datapath_type);
1821
1822 if (class && class->type_get_memory_usage) {
1823 class->type_get_memory_usage(datapath_type, usage);
1824 }
1825 }
1826
1827 void
1828 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1829 struct shash *info)
1830 {
1831 connmgr_get_controller_info(ofproto->connmgr, info);
1832 }
1833
1834 void
1835 ofproto_free_ofproto_controller_info(struct shash *info)
1836 {
1837 connmgr_free_controller_info(info);
1838 }
1839
1840 /* Makes a deep copy of 'old' into 'port'. */
1841 void
1842 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1843 {
1844 port->name = xstrdup(old->name);
1845 port->type = xstrdup(old->type);
1846 port->ofp_port = old->ofp_port;
1847 }
1848
1849 /* Frees memory allocated to members of 'ofproto_port'.
1850 *
1851 * Do not call this function on an ofproto_port obtained from
1852 * ofproto_port_dump_next(): that function retains ownership of the data in the
1853 * ofproto_port. */
1854 void
1855 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1856 {
1857 free(ofproto_port->name);
1858 free(ofproto_port->type);
1859 }
1860
1861 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1862 *
1863 * This function provides no status indication. An error status for the entire
1864 * dump operation is provided when it is completed by calling
1865 * ofproto_port_dump_done().
1866 */
1867 void
1868 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1869 const struct ofproto *ofproto)
1870 {
1871 dump->ofproto = ofproto;
1872 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1873 &dump->state);
1874 }
1875
1876 /* Attempts to retrieve another port from 'dump', which must have been created
1877 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1878 * 'port' and returns true. On failure, returns false.
1879 *
1880 * Failure might indicate an actual error or merely that the last port has been
1881 * dumped. An error status for the entire dump operation is provided when it
1882 * is completed by calling ofproto_port_dump_done().
1883 *
1884 * The ofproto owns the data stored in 'port'. It will remain valid until at
1885 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1886 * ofproto_port_dump_done(). */
1887 bool
1888 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1889 struct ofproto_port *port)
1890 {
1891 const struct ofproto *ofproto = dump->ofproto;
1892
1893 if (dump->error) {
1894 return false;
1895 }
1896
1897 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1898 port);
1899 if (dump->error) {
1900 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1901 return false;
1902 }
1903 return true;
1904 }
1905
1906 /* Completes port table dump operation 'dump', which must have been created
1907 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1908 * error-free, otherwise a positive errno value describing the problem. */
1909 int
1910 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1911 {
1912 const struct ofproto *ofproto = dump->ofproto;
1913 if (!dump->error) {
1914 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1915 dump->state);
1916 }
1917 return dump->error == EOF ? 0 : dump->error;
1918 }
1919
1920 /* Returns the type to pass to netdev_open() when a datapath of type
1921 * 'datapath_type' has a port of type 'port_type', for a few special
1922 * cases when a netdev type differs from a port type. For example, when
1923 * using the userspace datapath, a port of type "internal" needs to be
1924 * opened as "tap".
1925 *
1926 * Returns either 'type' itself or a string literal, which must not be
1927 * freed. */
1928 const char *
1929 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1930 {
1931 const struct ofproto_class *class;
1932
1933 datapath_type = ofproto_normalize_type(datapath_type);
1934 class = ofproto_class_find__(datapath_type);
1935 if (!class) {
1936 return port_type;
1937 }
1938
1939 return (class->port_open_type
1940 ? class->port_open_type(datapath_type, port_type)
1941 : port_type);
1942 }
1943
1944 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1945 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1946 * the port's OpenFlow port number.
1947 *
1948 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1949 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1950 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1951 * 'ofp_portp' is non-null). */
1952 int
1953 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1954 ofp_port_t *ofp_portp)
1955 {
1956 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1957 int error;
1958
1959 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1960 if (!error) {
1961 const char *netdev_name = netdev_get_name(netdev);
1962
1963 simap_put(&ofproto->ofp_requests, netdev_name,
1964 ofp_to_u16(ofp_port));
1965 update_port(ofproto, netdev_name);
1966 }
1967 if (ofp_portp) {
1968 *ofp_portp = OFPP_NONE;
1969 if (!error) {
1970 struct ofproto_port ofproto_port;
1971
1972 error = ofproto_port_query_by_name(ofproto,
1973 netdev_get_name(netdev),
1974 &ofproto_port);
1975 if (!error) {
1976 *ofp_portp = ofproto_port.ofp_port;
1977 ofproto_port_destroy(&ofproto_port);
1978 }
1979 }
1980 }
1981 return error;
1982 }
1983
1984 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1985 * initializes '*port' appropriately; on failure, returns a positive errno
1986 * value.
1987 *
1988 * The caller owns the data in 'ofproto_port' and must free it with
1989 * ofproto_port_destroy() when it is no longer needed. */
1990 int
1991 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1992 struct ofproto_port *port)
1993 {
1994 int error;
1995
1996 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1997 if (error) {
1998 memset(port, 0, sizeof *port);
1999 }
2000 return error;
2001 }
2002
2003 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
2004 * Returns 0 if successful, otherwise a positive errno. */
2005 int
2006 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
2007 {
2008 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2009 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
2010 struct simap_node *ofp_request_node;
2011 int error;
2012
2013 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
2014 if (ofp_request_node) {
2015 simap_delete(&ofproto->ofp_requests, ofp_request_node);
2016 }
2017
2018 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
2019 if (!error && ofport) {
2020 /* 'name' is the netdev's name and update_port() is going to close the
2021 * netdev. Just in case update_port() refers to 'name' after it
2022 * destroys 'ofport', make a copy of it around the update_port()
2023 * call. */
2024 char *devname = xstrdup(name);
2025 update_port(ofproto, devname);
2026 free(devname);
2027 }
2028 return error;
2029 }
2030
2031 static void
2032 flow_mod_init(struct ofputil_flow_mod *fm,
2033 const struct match *match, int priority,
2034 const struct ofpact *ofpacts, size_t ofpacts_len,
2035 enum ofp_flow_mod_command command)
2036 {
2037 *fm = (struct ofputil_flow_mod) {
2038 .match = *match,
2039 .priority = priority,
2040 .table_id = 0,
2041 .command = command,
2042 .buffer_id = UINT32_MAX,
2043 .out_port = OFPP_ANY,
2044 .out_group = OFPG_ANY,
2045 .ofpacts = CONST_CAST(struct ofpact *, ofpacts),
2046 .ofpacts_len = ofpacts_len,
2047 .delete_reason = OFPRR_DELETE,
2048 };
2049 }
2050
2051 static int
2052 simple_flow_mod(struct ofproto *ofproto,
2053 const struct match *match, int priority,
2054 const struct ofpact *ofpacts, size_t ofpacts_len,
2055 enum ofp_flow_mod_command command)
2056 {
2057 struct ofproto_flow_mod ofm;
2058
2059 flow_mod_init(&ofm.fm, match, priority, ofpacts, ofpacts_len, command);
2060
2061 return handle_flow_mod__(ofproto, &ofm, NULL);
2062 }
2063
2064 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
2065 * performs the 'n_actions' actions in 'actions'. The new flow will not
2066 * timeout.
2067 *
2068 * If cls_rule->priority is in the range of priorities supported by OpenFlow
2069 * (0...65535, inclusive) then the flow will be visible to OpenFlow
2070 * controllers; otherwise, it will be hidden.
2071 *
2072 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
2073 *
2074 * This is a helper function for in-band control and fail-open. */
2075 void
2076 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
2077 int priority,
2078 const struct ofpact *ofpacts, size_t ofpacts_len)
2079 OVS_EXCLUDED(ofproto_mutex)
2080 {
2081 const struct rule *rule;
2082 bool must_add;
2083
2084 /* First do a cheap check whether the rule we're looking for already exists
2085 * with the actions that we want. If it does, then we're done. */
2086 rule = rule_from_cls_rule(classifier_find_match_exactly(
2087 &ofproto->tables[0].cls, match, priority,
2088 CLS_MAX_VERSION));
2089 if (rule) {
2090 const struct rule_actions *actions = rule_get_actions(rule);
2091 must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len,
2092 ofpacts, ofpacts_len);
2093 } else {
2094 must_add = true;
2095 }
2096
2097 /* If there's no such rule or the rule doesn't have the actions we want,
2098 * fall back to a executing a full flow mod. We can't optimize this at
2099 * all because we didn't take enough locks above to ensure that the flow
2100 * table didn't already change beneath us. */
2101 if (must_add) {
2102 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
2103 OFPFC_MODIFY_STRICT);
2104 }
2105 }
2106
2107 /* Executes the flow modification specified in 'fm'. Returns 0 on success, or
2108 * an OFPERR_* OpenFlow error code on failure.
2109 *
2110 * This is a helper function for in-band control and fail-open and the "learn"
2111 * action. */
2112 enum ofperr
2113 ofproto_flow_mod(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
2114 OVS_EXCLUDED(ofproto_mutex)
2115 {
2116 struct ofputil_flow_mod *fm = &ofm->fm;
2117
2118 /* Optimize for the most common case of a repeated learn action.
2119 * If an identical flow already exists we only need to update its
2120 * 'modified' time. */
2121 if (fm->command == OFPFC_MODIFY_STRICT && fm->table_id != OFPTT_ALL
2122 && !(fm->flags & OFPUTIL_FF_RESET_COUNTS)) {
2123 struct oftable *table = &ofproto->tables[fm->table_id];
2124 struct rule *rule;
2125 bool done = false;
2126
2127 rule = rule_from_cls_rule(classifier_find_match_exactly(
2128 &table->cls, &fm->match, fm->priority,
2129 CLS_MAX_VERSION));
2130 if (rule) {
2131 /* Reading many of the rule fields and writing on 'modified'
2132 * requires the rule->mutex. Also, rule->actions may change
2133 * if rule->mutex is not held. */
2134 const struct rule_actions *actions;
2135
2136 ovs_mutex_lock(&rule->mutex);
2137 actions = rule_get_actions(rule);
2138 if (rule->idle_timeout == fm->idle_timeout
2139 && rule->hard_timeout == fm->hard_timeout
2140 && rule->importance == fm->importance
2141 && rule->flags == (fm->flags & OFPUTIL_FF_STATE)
2142 && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie))
2143 && ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2144 actions->ofpacts, actions->ofpacts_len)) {
2145 /* Rule already exists and need not change, only update the
2146 modified timestamp. */
2147 rule->modified = time_msec();
2148 done = true;
2149 }
2150 ovs_mutex_unlock(&rule->mutex);
2151 }
2152
2153 if (done) {
2154 return 0;
2155 }
2156 }
2157
2158 return handle_flow_mod__(ofproto, ofm, NULL);
2159 }
2160
2161 /* Searches for a rule with matching criteria exactly equal to 'target' in
2162 * ofproto's table 0 and, if it finds one, deletes it.
2163 *
2164 * This is a helper function for in-band control and fail-open. */
2165 void
2166 ofproto_delete_flow(struct ofproto *ofproto,
2167 const struct match *target, int priority)
2168 OVS_EXCLUDED(ofproto_mutex)
2169 {
2170 struct classifier *cls = &ofproto->tables[0].cls;
2171 struct rule *rule;
2172
2173 /* First do a cheap check whether the rule we're looking for has already
2174 * been deleted. If so, then we're done. */
2175 rule = rule_from_cls_rule(classifier_find_match_exactly(
2176 cls, target, priority, CLS_MAX_VERSION));
2177 if (!rule) {
2178 return;
2179 }
2180
2181 /* Execute a flow mod. We can't optimize this at all because we didn't
2182 * take enough locks above to ensure that the flow table didn't already
2183 * change beneath us. */
2184 simple_flow_mod(ofproto, target, priority, NULL, 0, OFPFC_DELETE_STRICT);
2185 }
2186
2187 /* Delete all of the flows from all of ofproto's flow tables, then reintroduce
2188 * the flows required by in-band control and fail-open. */
2189 void
2190 ofproto_flush_flows(struct ofproto *ofproto)
2191 {
2192 COVERAGE_INC(ofproto_flush);
2193 ofproto_flush__(ofproto);
2194 connmgr_flushed(ofproto->connmgr);
2195 }
2196 \f
2197 static void
2198 reinit_ports(struct ofproto *p)
2199 {
2200 struct ofproto_port_dump dump;
2201 struct sset devnames;
2202 struct ofport *ofport;
2203 struct ofproto_port ofproto_port;
2204 const char *devname;
2205
2206 COVERAGE_INC(ofproto_reinit_ports);
2207
2208 sset_init(&devnames);
2209 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2210 sset_add(&devnames, netdev_get_name(ofport->netdev));
2211 }
2212 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2213 sset_add(&devnames, ofproto_port.name);
2214 }
2215
2216 SSET_FOR_EACH (devname, &devnames) {
2217 update_port(p, devname);
2218 }
2219 sset_destroy(&devnames);
2220 }
2221
2222 static ofp_port_t
2223 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
2224 {
2225 uint16_t port_idx;
2226
2227 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
2228 port_idx = port_idx ? port_idx : UINT16_MAX;
2229
2230 if (port_idx >= ofproto->max_ports
2231 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2232 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2233 long long int last_used_at, lru = LLONG_MAX;
2234
2235 /* Search for a free OpenFlow port number. We try not to
2236 * immediately reuse them to prevent problems due to old
2237 * flows.
2238 *
2239 * We limit the automatically assigned port numbers to the lower half
2240 * of the port range, to reserve the upper half for assignment by
2241 * controllers. */
2242 for (;;) {
2243 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
2244 ofproto->alloc_port_no = 1;
2245 }
2246 last_used_at = ofport_get_usage(ofproto,
2247 u16_to_ofp(ofproto->alloc_port_no));
2248 if (!last_used_at) {
2249 port_idx = ofproto->alloc_port_no;
2250 break;
2251 } else if ( last_used_at < time_msec() - 60*60*1000) {
2252 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2253 * more than an hour ago, consider it usable. */
2254 ofport_remove_usage(ofproto,
2255 u16_to_ofp(ofproto->alloc_port_no));
2256 port_idx = ofproto->alloc_port_no;
2257 break;
2258 } else if (last_used_at < lru) {
2259 lru = last_used_at;
2260 lru_ofport = ofproto->alloc_port_no;
2261 }
2262
2263 if (ofproto->alloc_port_no == end_port_no) {
2264 if (lru_ofport) {
2265 port_idx = lru_ofport;
2266 break;
2267 }
2268 return OFPP_NONE;
2269 }
2270 }
2271 }
2272 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2273 return u16_to_ofp(port_idx);
2274 }
2275
2276 static void
2277 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2278 {
2279 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2280 ofport_set_usage(ofproto, ofp_port, time_msec());
2281 }
2282 }
2283
2284 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2285 * pointer if the netdev cannot be opened. On success, also fills in
2286 * '*pp'. */
2287 static struct netdev *
2288 ofport_open(struct ofproto *ofproto,
2289 struct ofproto_port *ofproto_port,
2290 struct ofputil_phy_port *pp)
2291 {
2292 enum netdev_flags flags;
2293 struct netdev *netdev;
2294 int error;
2295
2296 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2297 if (error) {
2298 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2299 "cannot be opened (%s)",
2300 ofproto->name,
2301 ofproto_port->name, ofproto_port->ofp_port,
2302 ofproto_port->name, ovs_strerror(error));
2303 return NULL;
2304 }
2305
2306 if (ofproto_port->ofp_port == OFPP_NONE) {
2307 if (!strcmp(ofproto->name, ofproto_port->name)) {
2308 ofproto_port->ofp_port = OFPP_LOCAL;
2309 } else {
2310 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2311 ofproto_port->name);
2312 }
2313 }
2314 pp->port_no = ofproto_port->ofp_port;
2315 netdev_get_etheraddr(netdev, &pp->hw_addr);
2316 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2317 netdev_get_flags(netdev, &flags);
2318 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2319 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2320 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2321 &pp->supported, &pp->peer);
2322 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2323 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2324
2325 return netdev;
2326 }
2327
2328 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2329 * port number, and 'config' bits other than OFPUTIL_PC_PORT_DOWN are
2330 * disregarded. */
2331 static bool
2332 ofport_equal(const struct ofputil_phy_port *a,
2333 const struct ofputil_phy_port *b)
2334 {
2335 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2336 && a->state == b->state
2337 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2338 && a->curr == b->curr
2339 && a->advertised == b->advertised
2340 && a->supported == b->supported
2341 && a->peer == b->peer
2342 && a->curr_speed == b->curr_speed
2343 && a->max_speed == b->max_speed);
2344 }
2345
2346 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2347 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2348 * one with the same name or port number). */
2349 static void
2350 ofport_install(struct ofproto *p,
2351 struct netdev *netdev, const struct ofputil_phy_port *pp)
2352 {
2353 const char *netdev_name = netdev_get_name(netdev);
2354 struct ofport *ofport;
2355 int error;
2356
2357 /* Create ofport. */
2358 ofport = p->ofproto_class->port_alloc();
2359 if (!ofport) {
2360 error = ENOMEM;
2361 goto error;
2362 }
2363 ofport->ofproto = p;
2364 ofport->netdev = netdev;
2365 ofport->change_seq = netdev_get_change_seq(netdev);
2366 ofport->pp = *pp;
2367 ofport->ofp_port = pp->port_no;
2368 ofport->created = time_msec();
2369
2370 /* Add port to 'p'. */
2371 hmap_insert(&p->ports, &ofport->hmap_node,
2372 hash_ofp_port(ofport->ofp_port));
2373 shash_add(&p->port_by_name, netdev_name, ofport);
2374
2375 update_mtu(p, ofport);
2376
2377 /* Let the ofproto_class initialize its private data. */
2378 error = p->ofproto_class->port_construct(ofport);
2379 if (error) {
2380 goto error;
2381 }
2382 connmgr_send_port_status(p->connmgr, NULL, pp, OFPPR_ADD);
2383 return;
2384
2385 error:
2386 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2387 p->name, netdev_name, ovs_strerror(error));
2388 if (ofport) {
2389 ofport_destroy__(ofport);
2390 } else {
2391 netdev_close(netdev);
2392 }
2393 }
2394
2395 /* Removes 'ofport' from 'p' and destroys it. */
2396 static void
2397 ofport_remove(struct ofport *ofport)
2398 {
2399 connmgr_send_port_status(ofport->ofproto->connmgr, NULL, &ofport->pp,
2400 OFPPR_DELETE);
2401 ofport_destroy(ofport);
2402 }
2403
2404 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2405 * destroys it. */
2406 static void
2407 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2408 {
2409 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2410 if (port) {
2411 ofport_remove(port);
2412 }
2413 }
2414
2415 /* Updates 'port' with new 'pp' description.
2416 *
2417 * Does not handle a name or port number change. The caller must implement
2418 * such a change as a delete followed by an add. */
2419 static void
2420 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2421 {
2422 port->pp.hw_addr = pp->hw_addr;
2423 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2424 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2425 port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN)
2426 | (pp->state & OFPUTIL_PS_LINK_DOWN));
2427 port->pp.curr = pp->curr;
2428 port->pp.advertised = pp->advertised;
2429 port->pp.supported = pp->supported;
2430 port->pp.peer = pp->peer;
2431 port->pp.curr_speed = pp->curr_speed;
2432 port->pp.max_speed = pp->max_speed;
2433
2434 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2435 &port->pp, OFPPR_MODIFY);
2436 }
2437
2438 /* Update OpenFlow 'state' in 'port' and notify controller. */
2439 void
2440 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2441 {
2442 if (port->pp.state != state) {
2443 port->pp.state = state;
2444 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2445 &port->pp, OFPPR_MODIFY);
2446 }
2447 }
2448
2449 void
2450 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2451 {
2452 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2453 if (port) {
2454 if (port->ofproto->ofproto_class->set_realdev) {
2455 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2456 }
2457 if (port->ofproto->ofproto_class->set_stp_port) {
2458 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2459 }
2460 if (port->ofproto->ofproto_class->set_rstp_port) {
2461 port->ofproto->ofproto_class->set_rstp_port(port, NULL);
2462 }
2463 if (port->ofproto->ofproto_class->set_cfm) {
2464 port->ofproto->ofproto_class->set_cfm(port, NULL);
2465 }
2466 if (port->ofproto->ofproto_class->bundle_remove) {
2467 port->ofproto->ofproto_class->bundle_remove(port);
2468 }
2469 }
2470 }
2471
2472 static void
2473 ofport_destroy__(struct ofport *port)
2474 {
2475 struct ofproto *ofproto = port->ofproto;
2476 const char *name = netdev_get_name(port->netdev);
2477
2478 hmap_remove(&ofproto->ports, &port->hmap_node);
2479 shash_delete(&ofproto->port_by_name,
2480 shash_find(&ofproto->port_by_name, name));
2481
2482 netdev_close(port->netdev);
2483 ofproto->ofproto_class->port_dealloc(port);
2484 }
2485
2486 static void
2487 ofport_destroy(struct ofport *port)
2488 {
2489 if (port) {
2490 dealloc_ofp_port(port->ofproto, port->ofp_port);
2491 port->ofproto->ofproto_class->port_destruct(port);
2492 ofport_destroy__(port);
2493 }
2494 }
2495
2496 struct ofport *
2497 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2498 {
2499 struct ofport *port;
2500
2501 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2502 &ofproto->ports) {
2503 if (port->ofp_port == ofp_port) {
2504 return port;
2505 }
2506 }
2507 return NULL;
2508 }
2509
2510 static long long int
2511 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2512 {
2513 struct ofport_usage *usage;
2514
2515 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2516 &ofproto->ofport_usage) {
2517 if (usage->ofp_port == ofp_port) {
2518 return usage->last_used;
2519 }
2520 }
2521 return 0;
2522 }
2523
2524 static void
2525 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2526 long long int last_used)
2527 {
2528 struct ofport_usage *usage;
2529 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2530 &ofproto->ofport_usage) {
2531 if (usage->ofp_port == ofp_port) {
2532 usage->last_used = last_used;
2533 return;
2534 }
2535 }
2536 ovs_assert(last_used == LLONG_MAX);
2537
2538 usage = xmalloc(sizeof *usage);
2539 usage->ofp_port = ofp_port;
2540 usage->last_used = last_used;
2541 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2542 hash_ofp_port(ofp_port));
2543 }
2544
2545 static void
2546 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2547 {
2548 struct ofport_usage *usage;
2549 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2550 &ofproto->ofport_usage) {
2551 if (usage->ofp_port == ofp_port) {
2552 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2553 free(usage);
2554 break;
2555 }
2556 }
2557 }
2558
2559 int
2560 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2561 {
2562 struct ofproto *ofproto = port->ofproto;
2563 int error;
2564
2565 if (ofproto->ofproto_class->port_get_stats) {
2566 error = ofproto->ofproto_class->port_get_stats(port, stats);
2567 } else {
2568 error = EOPNOTSUPP;
2569 }
2570
2571 return error;
2572 }
2573
2574 static void
2575 update_port(struct ofproto *ofproto, const char *name)
2576 {
2577 struct ofproto_port ofproto_port;
2578 struct ofputil_phy_port pp;
2579 struct netdev *netdev;
2580 struct ofport *port;
2581
2582 COVERAGE_INC(ofproto_update_port);
2583
2584 /* Fetch 'name''s location and properties from the datapath. */
2585 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2586 ? ofport_open(ofproto, &ofproto_port, &pp)
2587 : NULL);
2588
2589 if (netdev) {
2590 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2591 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2592 struct netdev *old_netdev = port->netdev;
2593
2594 /* 'name' hasn't changed location. Any properties changed? */
2595 if (!ofport_equal(&port->pp, &pp)) {
2596 ofport_modified(port, &pp);
2597 }
2598
2599 update_mtu(ofproto, port);
2600
2601 /* Install the newly opened netdev in case it has changed.
2602 * Don't close the old netdev yet in case port_modified has to
2603 * remove a retained reference to it.*/
2604 port->netdev = netdev;
2605 port->change_seq = netdev_get_change_seq(netdev);
2606
2607 if (port->ofproto->ofproto_class->port_modified) {
2608 port->ofproto->ofproto_class->port_modified(port);
2609 }
2610
2611 netdev_close(old_netdev);
2612 } else {
2613 /* If 'port' is nonnull then its name differs from 'name' and thus
2614 * we should delete it. If we think there's a port named 'name'
2615 * then its port number must be wrong now so delete it too. */
2616 if (port) {
2617 ofport_remove(port);
2618 }
2619 ofport_remove_with_name(ofproto, name);
2620 ofport_install(ofproto, netdev, &pp);
2621 }
2622 } else {
2623 /* Any port named 'name' is gone now. */
2624 ofport_remove_with_name(ofproto, name);
2625 }
2626 ofproto_port_destroy(&ofproto_port);
2627 }
2628
2629 static int
2630 init_ports(struct ofproto *p)
2631 {
2632 struct ofproto_port_dump dump;
2633 struct ofproto_port ofproto_port;
2634 struct shash_node *node, *next;
2635
2636 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2637 const char *name = ofproto_port.name;
2638
2639 if (shash_find(&p->port_by_name, name)) {
2640 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2641 p->name, name);
2642 } else {
2643 struct ofputil_phy_port pp;
2644 struct netdev *netdev;
2645
2646 /* Check if an OpenFlow port number had been requested. */
2647 node = shash_find(&init_ofp_ports, name);
2648 if (node) {
2649 const struct iface_hint *iface_hint = node->data;
2650 simap_put(&p->ofp_requests, name,
2651 ofp_to_u16(iface_hint->ofp_port));
2652 }
2653
2654 netdev = ofport_open(p, &ofproto_port, &pp);
2655 if (netdev) {
2656 ofport_install(p, netdev, &pp);
2657 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2658 p->alloc_port_no = MAX(p->alloc_port_no,
2659 ofp_to_u16(ofproto_port.ofp_port));
2660 }
2661 }
2662 }
2663 }
2664
2665 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2666 struct iface_hint *iface_hint = node->data;
2667
2668 if (!strcmp(iface_hint->br_name, p->name)) {
2669 free(iface_hint->br_name);
2670 free(iface_hint->br_type);
2671 free(iface_hint);
2672 shash_delete(&init_ofp_ports, node);
2673 }
2674 }
2675
2676 return 0;
2677 }
2678
2679 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2680 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2681 static int
2682 find_min_mtu(struct ofproto *p)
2683 {
2684 struct ofport *ofport;
2685 int mtu = 0;
2686
2687 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2688 struct netdev *netdev = ofport->netdev;
2689 int dev_mtu;
2690
2691 /* Skip any internal ports, since that's what we're trying to
2692 * set. */
2693 if (!strcmp(netdev_get_type(netdev), "internal")) {
2694 continue;
2695 }
2696
2697 if (netdev_get_mtu(netdev, &dev_mtu)) {
2698 continue;
2699 }
2700 if (!mtu || dev_mtu < mtu) {
2701 mtu = dev_mtu;
2702 }
2703 }
2704
2705 return mtu ? mtu: ETH_PAYLOAD_MAX;
2706 }
2707
2708 /* Update MTU of all datapath devices on 'p' to the minimum of the
2709 * non-datapath ports in event of 'port' added or changed. */
2710 static void
2711 update_mtu(struct ofproto *p, struct ofport *port)
2712 {
2713 struct ofport *ofport;
2714 struct netdev *netdev = port->netdev;
2715 int dev_mtu, old_min;
2716
2717 if (netdev_get_mtu(netdev, &dev_mtu)) {
2718 port->mtu = 0;
2719 return;
2720 }
2721 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2722 if (dev_mtu > p->min_mtu) {
2723 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2724 dev_mtu = p->min_mtu;
2725 }
2726 }
2727 port->mtu = dev_mtu;
2728 return;
2729 }
2730
2731 /* For non-internal port find new min mtu. */
2732 old_min = p->min_mtu;
2733 port->mtu = dev_mtu;
2734 p->min_mtu = find_min_mtu(p);
2735 if (p->min_mtu == old_min) {
2736 return;
2737 }
2738
2739 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2740 struct netdev *netdev = ofport->netdev;
2741
2742 if (!strcmp(netdev_get_type(netdev), "internal")) {
2743 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2744 ofport->mtu = p->min_mtu;
2745 }
2746 }
2747 }
2748 }
2749 \f
2750 static void
2751 ofproto_rule_destroy__(struct rule *rule)
2752 OVS_NO_THREAD_SAFETY_ANALYSIS
2753 {
2754 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2755 rule_actions_destroy(rule_get_actions(rule));
2756 ovs_mutex_destroy(&rule->mutex);
2757 rule->ofproto->ofproto_class->rule_dealloc(rule);
2758 }
2759
2760 static void
2761 rule_destroy_cb(struct rule *rule)
2762 OVS_NO_THREAD_SAFETY_ANALYSIS
2763 {
2764 /* Send rule removed if needed. */
2765 if (rule->flags & OFPUTIL_FF_SEND_FLOW_REM
2766 && rule->removed_reason != OVS_OFPRR_NONE
2767 && !rule_is_hidden(rule)) {
2768 ofproto_rule_send_removed(rule);
2769 }
2770 rule->ofproto->ofproto_class->rule_destruct(rule);
2771 ofproto_rule_destroy__(rule);
2772 }
2773
2774 void
2775 ofproto_rule_ref(struct rule *rule)
2776 {
2777 if (rule) {
2778 ovs_refcount_ref(&rule->ref_count);
2779 }
2780 }
2781
2782 bool
2783 ofproto_rule_try_ref(struct rule *rule)
2784 {
2785 if (rule) {
2786 return ovs_refcount_try_ref_rcu(&rule->ref_count);
2787 }
2788 return false;
2789 }
2790
2791 /* Decrements 'rule''s ref_count and schedules 'rule' to be destroyed if the
2792 * ref_count reaches 0.
2793 *
2794 * Use of RCU allows short term use (between RCU quiescent periods) without
2795 * keeping a reference. A reference must be taken if the rule needs to
2796 * stay around accross the RCU quiescent periods. */
2797 void
2798 ofproto_rule_unref(struct rule *rule)
2799 {
2800 if (rule && ovs_refcount_unref_relaxed(&rule->ref_count) == 1) {
2801 ovsrcu_postpone(rule_destroy_cb, rule);
2802 }
2803 }
2804
2805 static void
2806 remove_rule_rcu__(struct rule *rule)
2807 OVS_REQUIRES(ofproto_mutex)
2808 {
2809 struct ofproto *ofproto = rule->ofproto;
2810 struct oftable *table = &ofproto->tables[rule->table_id];
2811
2812 ovs_assert(!cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
2813 if (!classifier_remove(&table->cls, &rule->cr)) {
2814 OVS_NOT_REACHED();
2815 }
2816 ofproto->ofproto_class->rule_delete(rule);
2817 ofproto_rule_unref(rule);
2818 }
2819
2820 static void
2821 remove_rule_rcu(struct rule *rule)
2822 OVS_EXCLUDED(ofproto_mutex)
2823 {
2824 ovs_mutex_lock(&ofproto_mutex);
2825 remove_rule_rcu__(rule);
2826 ovs_mutex_unlock(&ofproto_mutex);
2827 }
2828
2829 /* Removes and deletes rules from a NULL-terminated array of rule pointers. */
2830 static void
2831 remove_rules_rcu(struct rule **rules)
2832 OVS_EXCLUDED(ofproto_mutex)
2833 {
2834 struct rule **orig_rules = rules;
2835
2836 if (*rules) {
2837 struct ofproto *ofproto = rules[0]->ofproto;
2838 unsigned long tables[BITMAP_N_LONGS(256)];
2839 struct rule *rule;
2840 size_t table_id;
2841
2842 memset(tables, 0, sizeof tables);
2843
2844 ovs_mutex_lock(&ofproto_mutex);
2845 while ((rule = *rules++)) {
2846 /* Defer once for each new table. This defers the subtable cleanup
2847 * until later, so that when removing large number of flows the
2848 * operation is faster. */
2849 if (!bitmap_is_set(tables, rule->table_id)) {
2850 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
2851
2852 bitmap_set1(tables, rule->table_id);
2853 classifier_defer(cls);
2854 }
2855 remove_rule_rcu__(rule);
2856 }
2857
2858 BITMAP_FOR_EACH_1(table_id, 256, tables) {
2859 struct classifier *cls = &ofproto->tables[table_id].cls;
2860
2861 classifier_publish(cls);
2862 }
2863 ovs_mutex_unlock(&ofproto_mutex);
2864 }
2865
2866 free(orig_rules);
2867 }
2868
2869 void
2870 ofproto_group_ref(struct ofgroup *group)
2871 {
2872 if (group) {
2873 ovs_refcount_ref(&group->ref_count);
2874 }
2875 }
2876
2877 void
2878 ofproto_group_unref(struct ofgroup *group)
2879 {
2880 if (group && ovs_refcount_unref(&group->ref_count) == 1) {
2881 group->ofproto->ofproto_class->group_destruct(group);
2882 ofputil_bucket_list_destroy(&group->buckets);
2883 group->ofproto->ofproto_class->group_dealloc(group);
2884 }
2885 }
2886
2887 static uint32_t get_provider_meter_id(const struct ofproto *,
2888 uint32_t of_meter_id);
2889
2890 /* Creates and returns a new 'struct rule_actions', whose actions are a copy
2891 * of from the 'ofpacts_len' bytes of 'ofpacts'. */
2892 const struct rule_actions *
2893 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2894 {
2895 struct rule_actions *actions;
2896
2897 actions = xmalloc(sizeof *actions + ofpacts_len);
2898 actions->ofpacts_len = ofpacts_len;
2899 actions->has_meter = ofpacts_get_meter(ofpacts, ofpacts_len) != 0;
2900 memcpy(actions->ofpacts, ofpacts, ofpacts_len);
2901
2902 actions->has_learn_with_delete = (next_learn_with_delete(actions, NULL)
2903 != NULL);
2904
2905 return actions;
2906 }
2907
2908 /* Free the actions after the RCU quiescent period is reached. */
2909 void
2910 rule_actions_destroy(const struct rule_actions *actions)
2911 {
2912 if (actions) {
2913 ovsrcu_postpone(free, CONST_CAST(struct rule_actions *, actions));
2914 }
2915 }
2916
2917 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2918 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2919 bool
2920 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2921 OVS_REQUIRES(ofproto_mutex)
2922 {
2923 if (port == OFPP_ANY) {
2924 return true;
2925 } else {
2926 const struct rule_actions *actions = rule_get_actions(rule);
2927 return ofpacts_output_to_port(actions->ofpacts,
2928 actions->ofpacts_len, port);
2929 }
2930 }
2931
2932 /* Returns true if 'rule' has group and equals group_id. */
2933 static bool
2934 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2935 OVS_REQUIRES(ofproto_mutex)
2936 {
2937 if (group_id == OFPG_ANY) {
2938 return true;
2939 } else {
2940 const struct rule_actions *actions = rule_get_actions(rule);
2941 return ofpacts_output_to_group(actions->ofpacts,
2942 actions->ofpacts_len, group_id);
2943 }
2944 }
2945
2946 static void
2947 rule_execute_destroy(struct rule_execute *e)
2948 {
2949 ofproto_rule_unref(e->rule);
2950 list_remove(&e->list_node);
2951 free(e);
2952 }
2953
2954 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2955 * by passing them to the ofproto provider. */
2956 static void
2957 run_rule_executes(struct ofproto *ofproto)
2958 OVS_EXCLUDED(ofproto_mutex)
2959 {
2960 struct rule_execute *e, *next;
2961 struct ovs_list executes;
2962
2963 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2964 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2965 struct flow flow;
2966
2967 flow_extract(e->packet, &flow);
2968 flow.in_port.ofp_port = e->in_port;
2969 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2970
2971 rule_execute_destroy(e);
2972 }
2973 }
2974
2975 /* Destroys and discards all "rule_execute" operations queued up in
2976 * ofproto->rule_executes. */
2977 static void
2978 destroy_rule_executes(struct ofproto *ofproto)
2979 {
2980 struct rule_execute *e, *next;
2981 struct ovs_list executes;
2982
2983 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2984 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2985 dp_packet_delete(e->packet);
2986 rule_execute_destroy(e);
2987 }
2988 }
2989
2990 static bool
2991 rule_is_readonly(const struct rule *rule)
2992 {
2993 const struct oftable *table = &rule->ofproto->tables[rule->table_id];
2994 return (table->flags & OFTABLE_READONLY) != 0;
2995 }
2996 \f
2997 static uint32_t
2998 hash_learned_cookie(ovs_be64 cookie_, uint8_t table_id)
2999 {
3000 uint64_t cookie = (OVS_FORCE uint64_t) cookie_;
3001 return hash_3words(cookie, cookie >> 32, table_id);
3002 }
3003
3004 static void
3005 learned_cookies_update_one__(struct ofproto *ofproto,
3006 const struct ofpact_learn *learn,
3007 int delta, struct ovs_list *dead_cookies)
3008 OVS_REQUIRES(ofproto_mutex)
3009 {
3010 uint32_t hash = hash_learned_cookie(learn->cookie, learn->table_id);
3011 struct learned_cookie *c;
3012
3013 HMAP_FOR_EACH_WITH_HASH (c, u.hmap_node, hash, &ofproto->learned_cookies) {
3014 if (c->cookie == learn->cookie && c->table_id == learn->table_id) {
3015 c->n += delta;
3016 ovs_assert(c->n >= 0);
3017
3018 if (!c->n) {
3019 hmap_remove(&ofproto->learned_cookies, &c->u.hmap_node);
3020 list_push_back(dead_cookies, &c->u.list_node);
3021 }
3022
3023 return;
3024 }
3025 }
3026
3027 ovs_assert(delta > 0);
3028 c = xmalloc(sizeof *c);
3029 hmap_insert(&ofproto->learned_cookies, &c->u.hmap_node, hash);
3030 c->cookie = learn->cookie;
3031 c->table_id = learn->table_id;
3032 c->n = delta;
3033 }
3034
3035 static const struct ofpact_learn *
3036 next_learn_with_delete(const struct rule_actions *actions,
3037 const struct ofpact_learn *start)
3038 {
3039 const struct ofpact *pos;
3040
3041 for (pos = start ? ofpact_next(&start->ofpact) : actions->ofpacts;
3042 pos < ofpact_end(actions->ofpacts, actions->ofpacts_len);
3043 pos = ofpact_next(pos)) {
3044 if (pos->type == OFPACT_LEARN) {
3045 const struct ofpact_learn *learn = ofpact_get_LEARN(pos);
3046 if (learn->flags & NX_LEARN_F_DELETE_LEARNED) {
3047 return learn;
3048 }
3049 }
3050 }
3051
3052 return NULL;
3053 }
3054
3055 static void
3056 learned_cookies_update__(struct ofproto *ofproto,
3057 const struct rule_actions *actions,
3058 int delta, struct ovs_list *dead_cookies)
3059 OVS_REQUIRES(ofproto_mutex)
3060 {
3061 if (actions->has_learn_with_delete) {
3062 const struct ofpact_learn *learn;
3063
3064 for (learn = next_learn_with_delete(actions, NULL); learn;
3065 learn = next_learn_with_delete(actions, learn)) {
3066 learned_cookies_update_one__(ofproto, learn, delta, dead_cookies);
3067 }
3068 }
3069 }
3070
3071 static void
3072 learned_cookies_inc(struct ofproto *ofproto,
3073 const struct rule_actions *actions)
3074 OVS_REQUIRES(ofproto_mutex)
3075 {
3076 learned_cookies_update__(ofproto, actions, +1, NULL);
3077 }
3078
3079 static void
3080 learned_cookies_dec(struct ofproto *ofproto,
3081 const struct rule_actions *actions,
3082 struct ovs_list *dead_cookies)
3083 OVS_REQUIRES(ofproto_mutex)
3084 {
3085 learned_cookies_update__(ofproto, actions, -1, dead_cookies);
3086 }
3087
3088 static void
3089 learned_cookies_flush(struct ofproto *ofproto, struct ovs_list *dead_cookies)
3090 OVS_REQUIRES(ofproto_mutex)
3091 {
3092 struct learned_cookie *c;
3093
3094 LIST_FOR_EACH_POP (c, u.list_node, dead_cookies) {
3095 struct rule_criteria criteria;
3096 struct rule_collection rules;
3097 struct match match;
3098
3099 match_init_catchall(&match);
3100 rule_criteria_init(&criteria, c->table_id, &match, 0, CLS_MAX_VERSION,
3101 c->cookie, OVS_BE64_MAX, OFPP_ANY, OFPG_ANY);
3102 rule_criteria_require_rw(&criteria, false);
3103 collect_rules_loose(ofproto, &criteria, &rules);
3104 rule_criteria_destroy(&criteria);
3105 delete_flows__(&rules, OFPRR_DELETE, NULL);
3106
3107 free(c);
3108 }
3109 }
3110 \f
3111 static enum ofperr
3112 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
3113 {
3114 ofconn_send_reply(ofconn, make_echo_reply(oh));
3115 return 0;
3116 }
3117
3118 static void
3119 query_tables(struct ofproto *ofproto,
3120 struct ofputil_table_features **featuresp,
3121 struct ofputil_table_stats **statsp)
3122 {
3123 struct mf_bitmap rw_fields = oxm_writable_fields();
3124 struct mf_bitmap match = oxm_matchable_fields();
3125 struct mf_bitmap mask = oxm_maskable_fields();
3126
3127 struct ofputil_table_features *features;
3128 struct ofputil_table_stats *stats;
3129 int i;
3130
3131 features = *featuresp = xcalloc(ofproto->n_tables, sizeof *features);
3132 for (i = 0; i < ofproto->n_tables; i++) {
3133 struct ofputil_table_features *f = &features[i];
3134
3135 f->table_id = i;
3136 sprintf(f->name, "table%d", i);
3137 f->metadata_match = OVS_BE64_MAX;
3138 f->metadata_write = OVS_BE64_MAX;
3139 atomic_read_relaxed(&ofproto->tables[i].miss_config, &f->miss_config);
3140 f->max_entries = 1000000;
3141
3142 bool more_tables = false;
3143 for (int j = i + 1; j < ofproto->n_tables; j++) {
3144 if (!(ofproto->tables[j].flags & OFTABLE_HIDDEN)) {
3145 bitmap_set1(f->nonmiss.next, j);
3146 more_tables = true;
3147 }
3148 }
3149 f->nonmiss.instructions = (1u << N_OVS_INSTRUCTIONS) - 1;
3150 if (!more_tables) {
3151 f->nonmiss.instructions &= ~(1u << OVSINST_OFPIT11_GOTO_TABLE);
3152 }
3153 f->nonmiss.write.ofpacts = (UINT64_C(1) << N_OFPACTS) - 1;
3154 f->nonmiss.write.set_fields = rw_fields;
3155 f->nonmiss.apply = f->nonmiss.write;
3156 f->miss = f->nonmiss;
3157
3158 f->match = match;
3159 f->mask = mask;
3160 f->wildcard = match;
3161 }
3162
3163 if (statsp) {
3164 stats = *statsp = xcalloc(ofproto->n_tables, sizeof *stats);
3165 for (i = 0; i < ofproto->n_tables; i++) {
3166 struct ofputil_table_stats *s = &stats[i];
3167
3168 s->table_id = i;
3169 s->active_count = ofproto->tables[i].n_flows;
3170 if (i == 0) {
3171 s->active_count -= connmgr_count_hidden_rules(
3172 ofproto->connmgr);
3173 }
3174 }
3175 } else {
3176 stats = NULL;
3177 }
3178
3179 ofproto->ofproto_class->query_tables(ofproto, features, stats);
3180
3181 for (i = 0; i < ofproto->n_tables; i++) {
3182 const struct oftable *table = &ofproto->tables[i];
3183 struct ofputil_table_features *f = &features[i];
3184
3185 if (table->name) {
3186 ovs_strzcpy(f->name, table->name, sizeof f->name);
3187 }
3188
3189 if (table->max_flows < f->max_entries) {
3190 f->max_entries = table->max_flows;
3191 }
3192 }
3193 }
3194
3195 static void
3196 query_switch_features(struct ofproto *ofproto,
3197 bool *arp_match_ip, uint64_t *ofpacts)
3198 {
3199 struct ofputil_table_features *features, *f;
3200
3201 *arp_match_ip = false;
3202 *ofpacts = 0;
3203
3204 query_tables(ofproto, &features, NULL);
3205 for (f = features; f < &features[ofproto->n_tables]; f++) {
3206 *ofpacts |= f->nonmiss.apply.ofpacts | f->miss.apply.ofpacts;
3207 if (bitmap_is_set(f->match.bm, MFF_ARP_SPA) ||
3208 bitmap_is_set(f->match.bm, MFF_ARP_TPA)) {
3209 *arp_match_ip = true;
3210 }
3211 }
3212 free(features);
3213
3214 /* Sanity check. */
3215 ovs_assert(*ofpacts & (UINT64_C(1) << OFPACT_OUTPUT));
3216 }
3217
3218 static enum ofperr
3219 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
3220 {
3221 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3222 struct ofputil_switch_features features;
3223 struct ofport *port;
3224 bool arp_match_ip;
3225 struct ofpbuf *b;
3226
3227 query_switch_features(ofproto, &arp_match_ip, &features.ofpacts);
3228
3229 features.datapath_id = ofproto->datapath_id;
3230 features.n_buffers = pktbuf_capacity();
3231 features.n_tables = ofproto_get_n_visible_tables(ofproto);
3232 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
3233 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS |
3234 OFPUTIL_C_GROUP_STATS);
3235 if (arp_match_ip) {
3236 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
3237 }
3238 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
3239 features.auxiliary_id = 0;
3240 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
3241 oh->xid);
3242 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3243 ofputil_put_switch_features_port(&port->pp, b);
3244 }
3245
3246 ofconn_send_reply(ofconn, b);
3247 return 0;
3248 }
3249
3250 static enum ofperr
3251 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
3252 {
3253 struct ofputil_switch_config config;
3254 config.frag = ofconn_get_ofproto(ofconn)->frag_handling;
3255 config.invalid_ttl_to_controller
3256 = ofconn_get_invalid_ttl_to_controller(ofconn);
3257 config.miss_send_len = ofconn_get_miss_send_len(ofconn);
3258
3259 ofconn_send_reply(ofconn, ofputil_encode_get_config_reply(oh, &config));
3260
3261 return 0;
3262 }
3263
3264 static enum ofperr
3265 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
3266 {
3267 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3268 struct ofputil_switch_config config;
3269 enum ofperr error;
3270
3271 error = ofputil_decode_set_config(oh, &config);
3272 if (error) {
3273 return error;
3274 }
3275
3276 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
3277 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
3278 enum ofputil_frag_handling cur = ofproto->frag_handling;
3279 enum ofputil_frag_handling next = config.frag;
3280
3281 if (cur != next) {
3282 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
3283 ofproto->frag_handling = next;
3284 } else {
3285 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
3286 ofproto->name,
3287 ofputil_frag_handling_to_string(next));
3288 }
3289 }
3290 }
3291
3292 if (config.invalid_ttl_to_controller >= 0) {
3293 ofconn_set_invalid_ttl_to_controller(ofconn,
3294 config.invalid_ttl_to_controller);
3295 }
3296
3297 ofconn_set_miss_send_len(ofconn, config.miss_send_len);
3298
3299 return 0;
3300 }
3301
3302 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3303 * error message code for the caller to propagate upward. Otherwise, returns
3304 * 0.
3305 *
3306 * The log message mentions 'msg_type'. */
3307 static enum ofperr
3308 reject_slave_controller(struct ofconn *ofconn)
3309 {
3310 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
3311 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
3312 return OFPERR_OFPBRC_IS_SLAVE;
3313 } else {
3314 return 0;
3315 }
3316 }
3317
3318 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
3319 * for 'ofproto':
3320 *
3321 * - If they use a meter, then 'ofproto' has that meter configured.
3322 *
3323 * - If they use any groups, then 'ofproto' has that group configured.
3324 *
3325 * Returns 0 if successful, otherwise an OpenFlow error. */
3326 enum ofperr
3327 ofproto_check_ofpacts(struct ofproto *ofproto,
3328 const struct ofpact ofpacts[], size_t ofpacts_len)
3329 {
3330 const struct ofpact *a;
3331 uint32_t mid;
3332
3333 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
3334 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
3335 return OFPERR_OFPMMFC_INVALID_METER;
3336 }
3337
3338 OFPACT_FOR_EACH_FLATTENED (a, ofpacts, ofpacts_len) {
3339 if (a->type == OFPACT_GROUP
3340 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
3341 return OFPERR_OFPBAC_BAD_OUT_GROUP;
3342 }
3343 }
3344
3345 return 0;
3346 }
3347
3348 static enum ofperr
3349 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3350 {
3351 struct ofproto *p = ofconn_get_ofproto(ofconn);
3352 struct ofputil_packet_out po;
3353 struct dp_packet *payload;
3354 uint64_t ofpacts_stub[1024 / 8];
3355 struct ofpbuf ofpacts;
3356 struct flow flow;
3357 enum ofperr error;
3358
3359 COVERAGE_INC(ofproto_packet_out);
3360
3361 error = reject_slave_controller(ofconn);
3362 if (error) {
3363 goto exit;
3364 }
3365
3366 /* Decode message. */
3367 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3368 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
3369 if (error) {
3370 goto exit_free_ofpacts;
3371 }
3372 if (ofp_to_u16(po.in_port) >= p->max_ports
3373 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
3374 error = OFPERR_OFPBRC_BAD_PORT;
3375 goto exit_free_ofpacts;
3376 }
3377
3378 /* Get payload. */
3379 if (po.buffer_id != UINT32_MAX) {
3380 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
3381 if (error || !payload) {
3382 goto exit_free_ofpacts;
3383 }
3384 } else {
3385 /* Ensure that the L3 header is 32-bit aligned. */
3386 payload = dp_packet_clone_data_with_headroom(po.packet, po.packet_len, 2);
3387 }
3388
3389 /* Verify actions against packet, then send packet if successful. */
3390 flow_extract(payload, &flow);
3391 flow.in_port.ofp_port = po.in_port;
3392
3393 /* Check actions like for flow mods. We pass a 'table_id' of 0 to
3394 * ofproto_check_consistency(), which isn't strictly correct because these
3395 * actions aren't in any table. This is OK as 'table_id' is only used to
3396 * check instructions (e.g., goto-table), which can't appear on the action
3397 * list of a packet-out. */
3398 error = ofpacts_check_consistency(po.ofpacts, po.ofpacts_len,
3399 &flow, u16_to_ofp(p->max_ports),
3400 0, p->n_tables,
3401 ofconn_get_protocol(ofconn));
3402 if (!error) {
3403 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
3404 if (!error) {
3405 error = p->ofproto_class->packet_out(p, payload, &flow,
3406 po.ofpacts, po.ofpacts_len);
3407 }
3408 }
3409 dp_packet_delete(payload);
3410
3411 exit_free_ofpacts:
3412 ofpbuf_uninit(&ofpacts);
3413 exit:
3414 return error;
3415 }
3416
3417 static void
3418 update_port_config(struct ofconn *ofconn, struct ofport *port,
3419 enum ofputil_port_config config,
3420 enum ofputil_port_config mask)
3421 {
3422 enum ofputil_port_config toggle = (config ^ port->pp.config) & mask;
3423
3424 if (toggle & OFPUTIL_PC_PORT_DOWN
3425 && (config & OFPUTIL_PC_PORT_DOWN
3426 ? netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL)
3427 : netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL))) {
3428 /* We tried to bring the port up or down, but it failed, so don't
3429 * update the "down" bit. */
3430 toggle &= ~OFPUTIL_PC_PORT_DOWN;
3431 }
3432
3433 if (toggle) {
3434 enum ofputil_port_config old_config = port->pp.config;
3435 port->pp.config ^= toggle;
3436 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
3437 connmgr_send_port_status(port->ofproto->connmgr, ofconn, &port->pp,
3438 OFPPR_MODIFY);
3439 }
3440 }
3441
3442 static enum ofperr
3443 port_mod_start(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3444 struct ofport **port)
3445 {
3446 struct ofproto *p = ofconn_get_ofproto(ofconn);
3447
3448 *port = ofproto_get_port(p, pm->port_no);
3449 if (!*port) {
3450 return OFPERR_OFPPMFC_BAD_PORT;
3451 }
3452 if (!eth_addr_equals((*port)->pp.hw_addr, pm->hw_addr)) {
3453 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3454 }
3455 return 0;
3456 }
3457
3458 static void
3459 port_mod_finish(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3460 struct ofport *port)
3461 {
3462 update_port_config(ofconn, port, pm->config, pm->mask);
3463 if (pm->advertise) {
3464 netdev_set_advertisements(port->netdev, pm->advertise);
3465 }
3466 }
3467
3468 static enum ofperr
3469 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3470 {
3471 struct ofputil_port_mod pm;
3472 struct ofport *port;
3473 enum ofperr error;
3474
3475 error = reject_slave_controller(ofconn);
3476 if (error) {
3477 return error;
3478 }
3479
3480 error = ofputil_decode_port_mod(oh, &pm, false);
3481 if (error) {
3482 return error;
3483 }
3484
3485 error = port_mod_start(ofconn, &pm, &port);
3486 if (!error) {
3487 port_mod_finish(ofconn, &pm, port);
3488 }
3489 return error;
3490 }
3491
3492 static enum ofperr
3493 handle_desc_stats_request(struct ofconn *ofconn,
3494 const struct ofp_header *request)
3495 {
3496 static const char *default_mfr_desc = "Nicira, Inc.";
3497 static const char *default_hw_desc = "Open vSwitch";
3498 static const char *default_sw_desc = VERSION;
3499 static const char *default_serial_desc = "None";
3500 static const char *default_dp_desc = "None";
3501
3502 struct ofproto *p = ofconn_get_ofproto(ofconn);
3503 struct ofp_desc_stats *ods;
3504 struct ofpbuf *msg;
3505
3506 msg = ofpraw_alloc_stats_reply(request, 0);
3507 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3508 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3509 sizeof ods->mfr_desc);
3510 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3511 sizeof ods->hw_desc);
3512 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3513 sizeof ods->sw_desc);
3514 ovs_strlcpy(ods->serial_num,
3515 p->serial_desc ? p->serial_desc : default_serial_desc,
3516 sizeof ods->serial_num);
3517 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3518 sizeof ods->dp_desc);
3519 ofconn_send_reply(ofconn, msg);
3520
3521 return 0;
3522 }
3523
3524 static enum ofperr
3525 handle_table_stats_request(struct ofconn *ofconn,
3526 const struct ofp_header *request)
3527 {
3528 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3529 struct ofputil_table_features *features;
3530 struct ofputil_table_stats *stats;
3531 struct ofpbuf *reply;
3532 size_t i;
3533
3534 query_tables(ofproto, &features, &stats);
3535
3536 reply = ofputil_encode_table_stats_reply(request);
3537 for (i = 0; i < ofproto->n_tables; i++) {
3538 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3539 ofputil_append_table_stats_reply(reply, &stats[i], &features[i]);
3540 }
3541 }
3542 ofconn_send_reply(ofconn, reply);
3543
3544 free(features);
3545 free(stats);
3546
3547 return 0;
3548 }
3549
3550 static enum ofperr
3551 handle_table_features_request(struct ofconn *ofconn,
3552 const struct ofp_header *request)
3553 {
3554 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3555 struct ofputil_table_features *features;
3556 struct ovs_list replies;
3557 struct ofpbuf msg;
3558 size_t i;
3559
3560 ofpbuf_use_const(&msg, request, ntohs(request->length));
3561 ofpraw_pull_assert(&msg);
3562 if (msg.size || ofpmp_more(request)) {
3563 return OFPERR_OFPTFFC_EPERM;
3564 }
3565
3566 query_tables(ofproto, &features, NULL);
3567
3568 ofpmp_init(&replies, request);
3569 for (i = 0; i < ofproto->n_tables; i++) {
3570 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3571 ofputil_append_table_features_reply(&features[i], &replies);
3572 }
3573 }
3574 ofconn_send_replies(ofconn, &replies);
3575
3576 free(features);
3577
3578 return 0;
3579 }
3580
3581 static void
3582 query_table_desc__(struct ofputil_table_desc *td,
3583 struct ofproto *ofproto, uint8_t table_id)
3584 {
3585 unsigned int count = ofproto->tables[table_id].n_flows;
3586 unsigned int max_flows = ofproto->tables[table_id].max_flows;
3587
3588 td->table_id = table_id;
3589 td->eviction = (ofproto->tables[table_id].eviction & EVICTION_OPENFLOW
3590 ? OFPUTIL_TABLE_EVICTION_ON
3591 : OFPUTIL_TABLE_EVICTION_OFF);
3592 td->eviction_flags = OFPROTO_EVICTION_FLAGS;
3593 td->vacancy = (ofproto->tables[table_id].vacancy_enabled
3594 ? OFPUTIL_TABLE_VACANCY_ON
3595 : OFPUTIL_TABLE_VACANCY_OFF);
3596 td->table_vacancy.vacancy_down = ofproto->tables[table_id].vacancy_down;
3597 td->table_vacancy.vacancy_up = ofproto->tables[table_id].vacancy_up;
3598 td->table_vacancy.vacancy = max_flows ? (count * 100) / max_flows : 0;
3599 }
3600
3601 /* This function queries the database for dumping table-desc. */
3602 static void
3603 query_tables_desc(struct ofproto *ofproto, struct ofputil_table_desc **descp)
3604 {
3605 struct ofputil_table_desc *table_desc;
3606 size_t i;
3607
3608 table_desc = *descp = xcalloc(ofproto->n_tables, sizeof *table_desc);
3609 for (i = 0; i < ofproto->n_tables; i++) {
3610 struct ofputil_table_desc *td = &table_desc[i];
3611 query_table_desc__(td, ofproto, i);
3612 }
3613 }
3614
3615 /* Function to handle dump-table-desc request. */
3616 static enum ofperr
3617 handle_table_desc_request(struct ofconn *ofconn,
3618 const struct ofp_header *request)
3619 {
3620 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3621 struct ofputil_table_desc *table_desc;
3622 struct ovs_list replies;
3623 size_t i;
3624
3625 query_tables_desc(ofproto, &table_desc);
3626 ofpmp_init(&replies, request);
3627 for (i = 0; i < ofproto->n_tables; i++) {
3628 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3629 ofputil_append_table_desc_reply(&table_desc[i], &replies,
3630 request->version);
3631 }
3632 }
3633 ofconn_send_replies(ofconn, &replies);
3634 free(table_desc);
3635 return 0;
3636 }
3637
3638 static void
3639 append_port_stat(struct ofport *port, struct ovs_list *replies)
3640 {
3641 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3642
3643 calc_duration(port->created, time_msec(),
3644 &ops.duration_sec, &ops.duration_nsec);
3645
3646 /* Intentionally ignore return value, since errors will set
3647 * 'stats' to all-1s, which is correct for OpenFlow, and
3648 * netdev_get_stats() will log errors. */
3649 ofproto_port_get_stats(port, &ops.stats);
3650
3651 ofputil_append_port_stat(replies, &ops);
3652 }
3653
3654 static void
3655 handle_port_request(struct ofconn *ofconn,
3656 const struct ofp_header *request, ofp_port_t port_no,
3657 void (*cb)(struct ofport *, struct ovs_list *replies))
3658 {
3659 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3660 struct ofport *port;
3661 struct ovs_list replies;
3662
3663 ofpmp_init(&replies, request);
3664 if (port_no != OFPP_ANY) {
3665 port = ofproto_get_port(ofproto, port_no);
3666 if (port) {
3667 cb(port, &replies);
3668 }
3669 } else {
3670 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3671 cb(port, &replies);
3672 }
3673 }
3674
3675 ofconn_send_replies(ofconn, &replies);
3676 }
3677
3678 static enum ofperr
3679 handle_port_stats_request(struct ofconn *ofconn,
3680 const struct ofp_header *request)
3681 {
3682 ofp_port_t port_no;
3683 enum ofperr error;
3684
3685 error = ofputil_decode_port_stats_request(request, &port_no);
3686 if (!error) {
3687 handle_port_request(ofconn, request, port_no, append_port_stat);
3688 }
3689 return error;
3690 }
3691
3692 static void
3693 append_port_desc(struct ofport *port, struct ovs_list *replies)
3694 {
3695 ofputil_append_port_desc_stats_reply(&port->pp, replies);
3696 }
3697
3698 static enum ofperr
3699 handle_port_desc_stats_request(struct ofconn *ofconn,
3700 const struct ofp_header *request)
3701 {
3702 ofp_port_t port_no;
3703 enum ofperr error;
3704
3705 error = ofputil_decode_port_desc_stats_request(request, &port_no);
3706 if (!error) {
3707 handle_port_request(ofconn, request, port_no, append_port_desc);
3708 }
3709 return error;
3710 }
3711
3712 static uint32_t
3713 hash_cookie(ovs_be64 cookie)
3714 {
3715 return hash_uint64((OVS_FORCE uint64_t)cookie);
3716 }
3717
3718 static void
3719 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3720 OVS_REQUIRES(ofproto_mutex)
3721 {
3722 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3723 hash_cookie(rule->flow_cookie));
3724 }
3725
3726 static void
3727 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3728 OVS_REQUIRES(ofproto_mutex)
3729 {
3730 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3731 }
3732
3733 static void
3734 calc_duration(long long int start, long long int now,
3735 uint32_t *sec, uint32_t *nsec)
3736 {
3737 long long int msecs = now - start;
3738 *sec = msecs / 1000;
3739 *nsec = (msecs % 1000) * (1000 * 1000);
3740 }
3741
3742 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3743 * true if 'table_id' is OK, false otherwise. */
3744 static bool
3745 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3746 {
3747 return table_id == OFPTT_ALL || table_id < ofproto->n_tables;
3748 }
3749
3750 static struct oftable *
3751 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3752 {
3753 struct oftable *table;
3754
3755 for (table = &ofproto->tables[table_id];
3756 table < &ofproto->tables[ofproto->n_tables];
3757 table++) {
3758 if (!(table->flags & OFTABLE_HIDDEN)) {
3759 return table;
3760 }
3761 }
3762
3763 return NULL;
3764 }
3765
3766 static struct oftable *
3767 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3768 {
3769 if (table_id == 0xff) {
3770 return next_visible_table(ofproto, 0);
3771 } else if (table_id < ofproto->n_tables) {
3772 return &ofproto->tables[table_id];
3773 } else {
3774 return NULL;
3775 }
3776 }
3777
3778 static struct oftable *
3779 next_matching_table(const struct ofproto *ofproto,
3780 const struct oftable *table, uint8_t table_id)
3781 {
3782 return (table_id == 0xff
3783 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3784 : NULL);
3785 }
3786
3787 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3788 *
3789 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3790 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3791 *
3792 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3793 * only once, for that table. (This can be used to access tables marked
3794 * OFTABLE_HIDDEN.)
3795 *
3796 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3797 * entered at all. (Perhaps you should have validated TABLE_ID with
3798 * check_table_id().)
3799 *
3800 * All parameters are evaluated multiple times.
3801 */
3802 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3803 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3804 (TABLE) != NULL; \
3805 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3806
3807 /* Initializes 'criteria' in a straightforward way based on the other
3808 * parameters.
3809 *
3810 * By default, the criteria include flows that are read-only, on the assumption
3811 * that the collected flows won't be modified. Call rule_criteria_require_rw()
3812 * if flows will be modified.
3813 *
3814 * For "loose" matching, the 'priority' parameter is unimportant and may be
3815 * supplied as 0. */
3816 static void
3817 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3818 const struct match *match, int priority,
3819 cls_version_t version, ovs_be64 cookie,
3820 ovs_be64 cookie_mask, ofp_port_t out_port,
3821 uint32_t out_group)
3822 {
3823 criteria->table_id = table_id;
3824 cls_rule_init(&criteria->cr, match, priority);
3825 criteria->version = version;
3826 criteria->cookie = cookie;
3827 criteria->cookie_mask = cookie_mask;
3828 criteria->out_port = out_port;
3829 criteria->out_group = out_group;
3830
3831 /* We ordinarily want to skip hidden rules, but there has to be a way for
3832 * code internal to OVS to modify and delete them, so if the criteria
3833 * specify a priority that can only be for a hidden flow, then allow hidden
3834 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3835 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3836 * priority.) */
3837 criteria->include_hidden = priority > UINT16_MAX;
3838
3839 /* We assume that the criteria are being used to collect flows for reading
3840 * but not modification. Thus, we should collect read-only flows. */
3841 criteria->include_readonly = true;
3842 }
3843
3844 /* By default, criteria initialized by rule_criteria_init() will match flows
3845 * that are read-only, on the assumption that the collected flows won't be
3846 * modified. Call this function to match only flows that are be modifiable.
3847 *
3848 * Specify 'can_write_readonly' as false in ordinary circumstances, true if the
3849 * caller has special privileges that allow it to modify even "read-only"
3850 * flows. */
3851 static void
3852 rule_criteria_require_rw(struct rule_criteria *criteria,
3853 bool can_write_readonly)
3854 {
3855 criteria->include_readonly = can_write_readonly;
3856 }
3857
3858 static void
3859 rule_criteria_destroy(struct rule_criteria *criteria)
3860 {
3861 cls_rule_destroy(&criteria->cr);
3862 }
3863
3864 void
3865 rule_collection_init(struct rule_collection *rules)
3866 {
3867 rules->rules = rules->stub;
3868 rules->n = 0;
3869 rules->capacity = ARRAY_SIZE(rules->stub);
3870 }
3871
3872 void
3873 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3874 {
3875 if (rules->n >= rules->capacity) {
3876 size_t old_size, new_size;
3877
3878 old_size = rules->capacity * sizeof *rules->rules;
3879 rules->capacity *= 2;
3880 new_size = rules->capacity * sizeof *rules->rules;
3881
3882 if (rules->rules == rules->stub) {
3883 rules->rules = xmalloc(new_size);
3884 memcpy(rules->rules, rules->stub, old_size);
3885 } else {
3886 rules->rules = xrealloc(rules->rules, new_size);
3887 }
3888 }
3889
3890 rules->rules[rules->n++] = rule;
3891 }
3892
3893 void
3894 rule_collection_ref(struct rule_collection *rules)
3895 OVS_REQUIRES(ofproto_mutex)
3896 {
3897 size_t i;
3898
3899 for (i = 0; i < rules->n; i++) {
3900 ofproto_rule_ref(rules->rules[i]);
3901 }
3902 }
3903
3904 void
3905 rule_collection_unref(struct rule_collection *rules)
3906 {
3907 size_t i;
3908
3909 for (i = 0; i < rules->n; i++) {
3910 ofproto_rule_unref(rules->rules[i]);
3911 }
3912 }
3913
3914 /* Returns a NULL-terminated array of rule pointers,
3915 * destroys 'rules'. */
3916 static struct rule **
3917 rule_collection_detach(struct rule_collection *rules)
3918 {
3919 struct rule **rule_array;
3920
3921 rule_collection_add(rules, NULL);
3922
3923 if (rules->rules == rules->stub) {
3924 rules->rules = xmemdup(rules->rules, rules->n * sizeof *rules->rules);
3925 }
3926
3927 rule_array = rules->rules;
3928 rule_collection_init(rules);
3929
3930 return rule_array;
3931 }
3932
3933 void
3934 rule_collection_destroy(struct rule_collection *rules)
3935 {
3936 if (rules->rules != rules->stub) {
3937 free(rules->rules);
3938 }
3939
3940 /* Make repeated destruction harmless. */
3941 rule_collection_init(rules);
3942 }
3943
3944 /* Schedules postponed removal of rules, destroys 'rules'. */
3945 static void
3946 rule_collection_remove_postponed(struct rule_collection *rules)
3947 OVS_REQUIRES(ofproto_mutex)
3948 {
3949 if (rules->n > 0) {
3950 if (rules->n == 1) {
3951 ovsrcu_postpone(remove_rule_rcu, rules->rules[0]);
3952 } else {
3953 ovsrcu_postpone(remove_rules_rcu, rule_collection_detach(rules));
3954 }
3955 }
3956 }
3957
3958 /* Checks whether 'rule' matches 'c' and, if so, adds it to 'rules'. This
3959 * function verifies most of the criteria in 'c' itself, but the caller must
3960 * check 'c->cr' itself.
3961 *
3962 * Rules that have already been marked for removal are not collected.
3963 *
3964 * Increments '*n_readonly' if 'rule' wasn't added because it's read-only (and
3965 * 'c' only includes modifiable rules). */
3966 static void
3967 collect_rule(struct rule *rule, const struct rule_criteria *c,
3968 struct rule_collection *rules, size_t *n_readonly)
3969 OVS_REQUIRES(ofproto_mutex)
3970 {
3971 if ((c->table_id == rule->table_id || c->table_id == 0xff)
3972 && ofproto_rule_has_out_port(rule, c->out_port)
3973 && ofproto_rule_has_out_group(rule, c->out_group)
3974 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)
3975 && (!rule_is_hidden(rule) || c->include_hidden)
3976 && cls_rule_visible_in_version(&rule->cr, c->version)) {
3977 /* Rule matches all the criteria... */
3978 if (!rule_is_readonly(rule) || c->include_readonly) {
3979 /* ...add it. */
3980 rule_collection_add(rules, rule);
3981 } else {
3982 /* ...except it's read-only. */
3983 ++*n_readonly;
3984 }
3985 }
3986 }
3987
3988 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
3989 * on classifiers rules are done in the "loose" way required for OpenFlow
3990 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
3991 * 'rules'.
3992 *
3993 * Returns 0 on success, otherwise an OpenFlow error code. */
3994 static enum ofperr
3995 collect_rules_loose(struct ofproto *ofproto,
3996 const struct rule_criteria *criteria,
3997 struct rule_collection *rules)
3998 OVS_REQUIRES(ofproto_mutex)
3999 {
4000 struct oftable *table;
4001 enum ofperr error = 0;
4002 size_t n_readonly = 0;
4003
4004 rule_collection_init(rules);
4005
4006 if (!check_table_id(ofproto, criteria->table_id)) {
4007 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4008 goto exit;
4009 }
4010
4011 if (criteria->cookie_mask == OVS_BE64_MAX) {
4012 struct rule *rule;
4013
4014 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4015 hash_cookie(criteria->cookie),
4016 &ofproto->cookies) {
4017 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
4018 collect_rule(rule, criteria, rules, &n_readonly);
4019 }
4020 }
4021 } else {
4022 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4023 struct rule *rule;
4024
4025 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &criteria->cr,
4026 criteria->version) {
4027 collect_rule(rule, criteria, rules, &n_readonly);
4028 }
4029 }
4030 }
4031
4032 exit:
4033 if (!error && !rules->n && n_readonly) {
4034 /* We didn't find any rules to modify. We did find some read-only
4035 * rules that we're not allowed to modify, so report that. */
4036 error = OFPERR_OFPBRC_EPERM;
4037 }
4038 if (error) {
4039 rule_collection_destroy(rules);
4040 }
4041 return error;
4042 }
4043
4044 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4045 * on classifiers rules are done in the "strict" way required for OpenFlow
4046 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
4047 * rules on list 'rules'.
4048 *
4049 * Returns 0 on success, otherwise an OpenFlow error code. */
4050 static enum ofperr
4051 collect_rules_strict(struct ofproto *ofproto,
4052 const struct rule_criteria *criteria,
4053 struct rule_collection *rules)
4054 OVS_REQUIRES(ofproto_mutex)
4055 {
4056 struct oftable *table;
4057 size_t n_readonly = 0;
4058 enum ofperr error = 0;
4059
4060 rule_collection_init(rules);
4061
4062 if (!check_table_id(ofproto, criteria->table_id)) {
4063 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4064 goto exit;
4065 }
4066
4067 if (criteria->cookie_mask == OVS_BE64_MAX) {
4068 struct rule *rule;
4069
4070 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4071 hash_cookie(criteria->cookie),
4072 &ofproto->cookies) {
4073 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
4074 collect_rule(rule, criteria, rules, &n_readonly);
4075 }
4076 }
4077 } else {
4078 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4079 struct rule *rule;
4080
4081 rule = rule_from_cls_rule(classifier_find_rule_exactly(
4082 &table->cls, &criteria->cr,
4083 criteria->version));
4084 if (rule) {
4085 collect_rule(rule, criteria, rules, &n_readonly);
4086 }
4087 }
4088 }
4089
4090 exit:
4091 if (!error && !rules->n && n_readonly) {
4092 /* We didn't find any rules to modify. We did find some read-only
4093 * rules that we're not allowed to modify, so report that. */
4094 error = OFPERR_OFPBRC_EPERM;
4095 }
4096 if (error) {
4097 rule_collection_destroy(rules);
4098 }
4099 return error;
4100 }
4101
4102 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
4103 * forced into the range of a uint16_t. */
4104 static int
4105 age_secs(long long int age_ms)
4106 {
4107 return (age_ms < 0 ? 0
4108 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
4109 : (unsigned int) age_ms / 1000);
4110 }
4111
4112 static enum ofperr
4113 handle_flow_stats_request(struct ofconn *ofconn,
4114 const struct ofp_header *request)
4115 OVS_EXCLUDED(ofproto_mutex)
4116 {
4117 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4118 struct ofputil_flow_stats_request fsr;
4119 struct rule_criteria criteria;
4120 struct rule_collection rules;
4121 struct ovs_list replies;
4122 enum ofperr error;
4123 size_t i;
4124
4125 error = ofputil_decode_flow_stats_request(&fsr, request);
4126 if (error) {
4127 return error;
4128 }
4129
4130 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, CLS_MAX_VERSION,
4131 fsr.cookie, fsr.cookie_mask, fsr.out_port,
4132 fsr.out_group);
4133
4134 ovs_mutex_lock(&ofproto_mutex);
4135 error = collect_rules_loose(ofproto, &criteria, &rules);
4136 rule_criteria_destroy(&criteria);
4137 if (!error) {
4138 rule_collection_ref(&rules);
4139 }
4140 ovs_mutex_unlock(&ofproto_mutex);
4141
4142 if (error) {
4143 return error;
4144 }
4145
4146 ofpmp_init(&replies, request);
4147 for (i = 0; i < rules.n; i++) {
4148 struct rule *rule = rules.rules[i];
4149 long long int now = time_msec();
4150 struct ofputil_flow_stats fs;
4151 long long int created, used, modified;
4152 const struct rule_actions *actions;
4153 enum ofputil_flow_mod_flags flags;
4154
4155 ovs_mutex_lock(&rule->mutex);
4156 fs.cookie = rule->flow_cookie;
4157 fs.idle_timeout = rule->idle_timeout;
4158 fs.hard_timeout = rule->hard_timeout;
4159 fs.importance = rule->importance;
4160 created = rule->created;
4161 modified = rule->modified;
4162 actions = rule_get_actions(rule);
4163 flags = rule->flags;
4164 ovs_mutex_unlock(&rule->mutex);
4165
4166 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
4167 &fs.byte_count, &used);
4168
4169 minimatch_expand(&rule->cr.match, &fs.match);
4170 fs.table_id = rule->table_id;
4171 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
4172 fs.priority = rule->cr.priority;
4173 fs.idle_age = age_secs(now - used);
4174 fs.hard_age = age_secs(now - modified);
4175 fs.ofpacts = actions->ofpacts;
4176 fs.ofpacts_len = actions->ofpacts_len;
4177
4178 fs.flags = flags;
4179 ofputil_append_flow_stats_reply(&fs, &replies);
4180 }
4181
4182 rule_collection_unref(&rules);
4183 rule_collection_destroy(&rules);
4184
4185 ofconn_send_replies(ofconn, &replies);
4186
4187 return 0;
4188 }
4189
4190 static void
4191 flow_stats_ds(struct rule *rule, struct ds *results)
4192 {
4193 uint64_t packet_count, byte_count;
4194 const struct rule_actions *actions;
4195 long long int created, used;
4196
4197 rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4198 &byte_count, &used);
4199
4200 ovs_mutex_lock(&rule->mutex);
4201 actions = rule_get_actions(rule);
4202 created = rule->created;
4203 ovs_mutex_unlock(&rule->mutex);
4204
4205 if (rule->table_id != 0) {
4206 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
4207 }
4208 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
4209 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
4210 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
4211 cls_rule_format(&rule->cr, results);
4212 ds_put_char(results, ',');
4213
4214 ds_put_cstr(results, "actions=");
4215 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
4216
4217 ds_put_cstr(results, "\n");
4218 }
4219
4220 /* Adds a pretty-printed description of all flows to 'results', including
4221 * hidden flows (e.g., set up by in-band control). */
4222 void
4223 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
4224 {
4225 struct oftable *table;
4226
4227 OFPROTO_FOR_EACH_TABLE (table, p) {
4228 struct rule *rule;
4229
4230 CLS_FOR_EACH (rule, cr, &table->cls) {
4231 flow_stats_ds(rule, results);
4232 }
4233 }
4234 }
4235
4236 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
4237 * '*engine_type' and '*engine_id', respectively. */
4238 void
4239 ofproto_get_netflow_ids(const struct ofproto *ofproto,
4240 uint8_t *engine_type, uint8_t *engine_id)
4241 {
4242 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
4243 }
4244
4245 /* Checks the status change of CFM on 'ofport'.
4246 *
4247 * Returns true if 'ofproto_class' does not support 'cfm_status_changed'. */
4248 bool
4249 ofproto_port_cfm_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
4250 {
4251 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4252 return (ofport && ofproto->ofproto_class->cfm_status_changed
4253 ? ofproto->ofproto_class->cfm_status_changed(ofport)
4254 : true);
4255 }
4256
4257 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.
4258 * Returns 0 if the port's CFM status was successfully stored into
4259 * '*status'. Returns positive errno if the port did not have CFM
4260 * configured.
4261 *
4262 * The caller must provide and own '*status', and must free 'status->rmps'.
4263 * '*status' is indeterminate if the return value is non-zero. */
4264 int
4265 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
4266 struct cfm_status *status)
4267 {
4268 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4269 return (ofport && ofproto->ofproto_class->get_cfm_status
4270 ? ofproto->ofproto_class->get_cfm_status(ofport, status)
4271 : EOPNOTSUPP);
4272 }
4273
4274 static enum ofperr
4275 handle_aggregate_stats_request(struct ofconn *ofconn,
4276 const struct ofp_header *oh)
4277 OVS_EXCLUDED(ofproto_mutex)
4278 {
4279 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4280 struct ofputil_flow_stats_request request;
4281 struct ofputil_aggregate_stats stats;
4282 bool unknown_packets, unknown_bytes;
4283 struct rule_criteria criteria;
4284 struct rule_collection rules;
4285 struct ofpbuf *reply;
4286 enum ofperr error;
4287 size_t i;
4288
4289 error = ofputil_decode_flow_stats_request(&request, oh);
4290 if (error) {
4291 return error;
4292 }
4293
4294 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
4295 CLS_MAX_VERSION, request.cookie, request.cookie_mask,
4296 request.out_port, request.out_group);
4297
4298 ovs_mutex_lock(&ofproto_mutex);
4299 error = collect_rules_loose(ofproto, &criteria, &rules);
4300 rule_criteria_destroy(&criteria);
4301 if (!error) {
4302 rule_collection_ref(&rules);
4303 }
4304 ovs_mutex_unlock(&ofproto_mutex);
4305
4306 if (error) {
4307 return error;
4308 }
4309
4310 memset(&stats, 0, sizeof stats);
4311 unknown_packets = unknown_bytes = false;
4312 for (i = 0; i < rules.n; i++) {
4313 struct rule *rule = rules.rules[i];
4314 uint64_t packet_count;
4315 uint64_t byte_count;
4316 long long int used;
4317
4318 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4319 &byte_count, &used);
4320
4321 if (packet_count == UINT64_MAX) {
4322 unknown_packets = true;
4323 } else {
4324 stats.packet_count += packet_count;
4325 }
4326
4327 if (byte_count == UINT64_MAX) {
4328 unknown_bytes = true;
4329 } else {
4330 stats.byte_count += byte_count;
4331 }
4332
4333 stats.flow_count++;
4334 }
4335 if (unknown_packets) {
4336 stats.packet_count = UINT64_MAX;
4337 }
4338 if (unknown_bytes) {
4339 stats.byte_count = UINT64_MAX;
4340 }
4341
4342 rule_collection_unref(&rules);
4343 rule_collection_destroy(&rules);
4344
4345 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
4346 ofconn_send_reply(ofconn, reply);
4347
4348 return 0;
4349 }
4350
4351 struct queue_stats_cbdata {
4352 struct ofport *ofport;
4353 struct ovs_list replies;
4354 long long int now;
4355 };
4356
4357 static void
4358 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
4359 const struct netdev_queue_stats *stats)
4360 {
4361 struct ofputil_queue_stats oqs;
4362
4363 oqs.port_no = cbdata->ofport->pp.port_no;
4364 oqs.queue_id = queue_id;
4365 oqs.tx_bytes = stats->tx_bytes;
4366 oqs.tx_packets = stats->tx_packets;
4367 oqs.tx_errors = stats->tx_errors;
4368 if (stats->created != LLONG_MIN) {
4369 calc_duration(stats->created, cbdata->now,
4370 &oqs.duration_sec, &oqs.duration_nsec);
4371 } else {
4372 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
4373 }
4374 ofputil_append_queue_stat(&cbdata->replies, &oqs);
4375 }
4376
4377 static void
4378 handle_queue_stats_dump_cb(uint32_t queue_id,
4379 struct netdev_queue_stats *stats,
4380 void *cbdata_)
4381 {
4382 struct queue_stats_cbdata *cbdata = cbdata_;
4383
4384 put_queue_stats(cbdata, queue_id, stats);
4385 }
4386
4387 static enum ofperr
4388 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
4389 struct queue_stats_cbdata *cbdata)
4390 {
4391 cbdata->ofport = port;
4392 if (queue_id == OFPQ_ALL) {
4393 netdev_dump_queue_stats(port->netdev,
4394 handle_queue_stats_dump_cb, cbdata);
4395 } else {
4396 struct netdev_queue_stats stats;
4397
4398 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
4399 put_queue_stats(cbdata, queue_id, &stats);
4400 } else {
4401 return OFPERR_OFPQOFC_BAD_QUEUE;
4402 }
4403 }
4404 return 0;
4405 }
4406
4407 static enum ofperr
4408 handle_queue_stats_request(struct ofconn *ofconn,
4409 const struct ofp_header *rq)
4410 {
4411 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4412 struct queue_stats_cbdata cbdata;
4413 struct ofport *port;
4414 enum ofperr error;
4415 struct ofputil_queue_stats_request oqsr;
4416
4417 COVERAGE_INC(ofproto_queue_req);
4418
4419 ofpmp_init(&cbdata.replies, rq);
4420 cbdata.now = time_msec();
4421
4422 error = ofputil_decode_queue_stats_request(rq, &oqsr);
4423 if (error) {
4424 return error;
4425 }
4426
4427 if (oqsr.port_no == OFPP_ANY) {
4428 error = OFPERR_OFPQOFC_BAD_QUEUE;
4429 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
4430 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
4431 error = 0;
4432 }
4433 }
4434 } else {
4435 port = ofproto_get_port(ofproto, oqsr.port_no);
4436 error = (port
4437 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
4438 : OFPERR_OFPQOFC_BAD_PORT);
4439 }
4440 if (!error) {
4441 ofconn_send_replies(ofconn, &cbdata.replies);
4442 } else {
4443 ofpbuf_list_delete(&cbdata.replies);
4444 }
4445
4446 return error;
4447 }
4448
4449 static enum ofperr
4450 evict_rules_from_table(struct oftable *table)
4451 OVS_REQUIRES(ofproto_mutex)
4452 {
4453 enum ofperr error = 0;
4454 struct rule_collection rules;
4455 unsigned int count = table->n_flows;
4456 unsigned int max_flows = table->max_flows;
4457
4458 rule_collection_init(&rules);
4459
4460 while (count-- > max_flows) {
4461 struct rule *rule;
4462
4463 if (!choose_rule_to_evict(table, &rule)) {
4464 error = OFPERR_OFPFMFC_TABLE_FULL;
4465 break;
4466 } else {
4467 eviction_group_remove_rule(rule);
4468 rule_collection_add(&rules, rule);
4469 }
4470 }
4471 delete_flows__(&rules, OFPRR_EVICTION, NULL);
4472
4473 return error;
4474 }
4475
4476 static void
4477 get_conjunctions(const struct ofputil_flow_mod *fm,
4478 struct cls_conjunction **conjsp, size_t *n_conjsp)
4479 OVS_REQUIRES(ofproto_mutex)
4480 {
4481 struct cls_conjunction *conjs = NULL;
4482 int n_conjs = 0;
4483
4484 const struct ofpact *ofpact;
4485 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4486 if (ofpact->type == OFPACT_CONJUNCTION) {
4487 n_conjs++;
4488 } else if (ofpact->type != OFPACT_NOTE) {
4489 /* "conjunction" may appear with "note" actions but not with any
4490 * other type of actions. */
4491 ovs_assert(!n_conjs);
4492 break;
4493 }
4494 }
4495 if (n_conjs) {
4496 int i = 0;
4497
4498 conjs = xzalloc(n_conjs * sizeof *conjs);
4499 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4500 if (ofpact->type == OFPACT_CONJUNCTION) {
4501 struct ofpact_conjunction *oc = ofpact_get_CONJUNCTION(ofpact);
4502 conjs[i].clause = oc->clause;
4503 conjs[i].n_clauses = oc->n_clauses;
4504 conjs[i].id = oc->id;
4505 i++;
4506 }
4507 }
4508 }
4509
4510 *conjsp = conjs;
4511 *n_conjsp = n_conjs;
4512 }
4513
4514 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4515 * in which no matching flow already exists in the flow table.
4516 *
4517 * Adds the flow specified by 'fm', to the ofproto's flow table. Returns 0 on
4518 * success, or an OpenFlow error code on failure.
4519 *
4520 * On successful return the caller must complete the operation either by
4521 * calling add_flow_finish(), or add_flow_revert() if the operation needs to
4522 * be reverted.
4523 *
4524 * The caller retains ownership of 'fm->ofpacts'. */
4525 static enum ofperr
4526 add_flow_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4527 OVS_REQUIRES(ofproto_mutex)
4528 {
4529 struct ofputil_flow_mod *fm = &ofm->fm;
4530 struct rule **old_rule = &ofm->old_rules.stub[0];
4531 struct rule **new_rule = &ofm->new_rules.stub[0];
4532 struct oftable *table;
4533 struct cls_rule cr;
4534 struct rule *rule;
4535 uint8_t table_id;
4536 struct cls_conjunction *conjs;
4537 size_t n_conjs;
4538 enum ofperr error;
4539
4540 if (!check_table_id(ofproto, fm->table_id)) {
4541 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4542 return error;
4543 }
4544
4545 /* Pick table. */
4546 if (fm->table_id == 0xff) {
4547 if (ofproto->ofproto_class->rule_choose_table) {
4548 error = ofproto->ofproto_class->rule_choose_table(ofproto,
4549 &fm->match,
4550 &table_id);
4551 if (error) {
4552 return error;
4553 }
4554 ovs_assert(table_id < ofproto->n_tables);
4555 } else {
4556 table_id = 0;
4557 }
4558 } else if (fm->table_id < ofproto->n_tables) {
4559 table_id = fm->table_id;
4560 } else {
4561 return OFPERR_OFPBRC_BAD_TABLE_ID;
4562 }
4563
4564 table = &ofproto->tables[table_id];
4565 if (table->flags & OFTABLE_READONLY
4566 && !(fm->flags & OFPUTIL_FF_NO_READONLY)) {
4567 return OFPERR_OFPBRC_EPERM;
4568 }
4569
4570 if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)
4571 && !match_has_default_hidden_fields(&fm->match)) {
4572 VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set "
4573 "non-default values to hidden fields", ofproto->name);
4574 return OFPERR_OFPBRC_EPERM;
4575 }
4576
4577 cls_rule_init(&cr, &fm->match, fm->priority);
4578
4579 /* Check for the existence of an identical rule.
4580 * This will not return rules earlier marked for removal. */
4581 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr,
4582 ofm->version));
4583 *old_rule = rule;
4584 if (!rule) {
4585 /* Check for overlap, if requested. */
4586 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP
4587 && classifier_rule_overlaps(&table->cls, &cr, ofm->version)) {
4588 cls_rule_destroy(&cr);
4589 return OFPERR_OFPFMFC_OVERLAP;
4590 }
4591
4592 /* If necessary, evict an existing rule to clear out space. */
4593 if (table->n_flows >= table->max_flows) {
4594 if (!choose_rule_to_evict(table, &rule)) {
4595 error = OFPERR_OFPFMFC_TABLE_FULL;
4596 cls_rule_destroy(&cr);
4597 return error;
4598 }
4599 eviction_group_remove_rule(rule);
4600 /* Marks '*old_rule' as an evicted rule rather than replaced rule.
4601 */
4602 fm->delete_reason = OFPRR_EVICTION;
4603 *old_rule = rule;
4604 }
4605 } else {
4606 fm->modify_cookie = true;
4607 }
4608
4609 /* Allocate new rule. */
4610 error = replace_rule_create(ofproto, fm, &cr, table - ofproto->tables,
4611 rule, new_rule);
4612 if (error) {
4613 return error;
4614 }
4615
4616 get_conjunctions(fm, &conjs, &n_conjs);
4617 replace_rule_start(ofproto, ofm->version, rule, *new_rule, conjs, n_conjs);
4618 free(conjs);
4619
4620 return 0;
4621 }
4622
4623 /* Revert the effects of add_flow_start(). */
4624 static void
4625 add_flow_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
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
4632 if (old_rule && fm->delete_reason == OFPRR_EVICTION) {
4633 /* Revert the eviction. */
4634 eviction_group_add_rule(old_rule);
4635 }
4636
4637 replace_rule_revert(ofproto, old_rule, new_rule);
4638 }
4639
4640 /* To be called after version bump. */
4641 static void
4642 add_flow_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
4643 const struct flow_mod_requester *req)
4644 OVS_REQUIRES(ofproto_mutex)
4645 {
4646 struct ofputil_flow_mod *fm = &ofm->fm;
4647 struct rule *old_rule = ofm->old_rules.stub[0];
4648 struct rule *new_rule = ofm->new_rules.stub[0];
4649 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
4650
4651 replace_rule_finish(ofproto, fm, req, old_rule, new_rule, &dead_cookies);
4652 learned_cookies_flush(ofproto, &dead_cookies);
4653
4654 if (old_rule) {
4655 ovsrcu_postpone(remove_rule_rcu, old_rule);
4656 } else {
4657 if (minimask_get_vid_mask(new_rule->cr.match.mask) == VLAN_VID_MASK) {
4658 if (ofproto->vlan_bitmap) {
4659 uint16_t vid = miniflow_get_vid(new_rule->cr.match.flow);
4660
4661 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4662 bitmap_set1(ofproto->vlan_bitmap, vid);
4663 ofproto->vlans_changed = true;
4664 }
4665 } else {
4666 ofproto->vlans_changed = true;
4667 }
4668 }
4669
4670 ofmonitor_report(ofproto->connmgr, new_rule, NXFME_ADDED, 0,
4671 req ? req->ofconn : NULL,
4672 req ? req->request->xid : 0, NULL);
4673 }
4674
4675 send_buffered_packet(req, fm->buffer_id, new_rule);
4676 }
4677 \f
4678 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4679
4680 /* Create a new rule based on attributes in 'fm', match in 'cr', 'table_id',
4681 * and 'old_rule'. Note that the rule is NOT inserted into a any data
4682 * structures yet. Takes ownership of 'cr'. */
4683 static enum ofperr
4684 replace_rule_create(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4685 struct cls_rule *cr, uint8_t table_id,
4686 struct rule *old_rule, struct rule **new_rule)
4687 {
4688 struct rule *rule;
4689 enum ofperr error;
4690
4691 /* Allocate new rule. */
4692 rule = ofproto->ofproto_class->rule_alloc();
4693 if (!rule) {
4694 cls_rule_destroy(cr);
4695 VLOG_WARN_RL(&rl, "%s: failed to allocate a rule.", ofproto->name);
4696 return OFPERR_OFPFMFC_UNKNOWN;
4697 }
4698
4699 /* Initialize base state. */
4700 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4701 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), cr);
4702 ovs_refcount_init(&rule->ref_count);
4703 rule->flow_cookie = fm->new_cookie;
4704 rule->created = rule->modified = time_msec();
4705
4706 ovs_mutex_init(&rule->mutex);
4707 ovs_mutex_lock(&rule->mutex);
4708 rule->idle_timeout = fm->idle_timeout;
4709 rule->hard_timeout = fm->hard_timeout;
4710 *CONST_CAST(uint16_t *, &rule->importance) = fm->importance;
4711 rule->removed_reason = OVS_OFPRR_NONE;
4712
4713 *CONST_CAST(uint8_t *, &rule->table_id) = table_id;
4714 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4715 *CONST_CAST(const struct rule_actions **, &rule->actions)
4716 = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
4717 list_init(&rule->meter_list_node);
4718 rule->eviction_group = NULL;
4719 list_init(&rule->expirable);
4720 rule->monitor_flags = 0;
4721 rule->add_seqno = 0;
4722 rule->modify_seqno = 0;
4723
4724 /* Copy values from old rule for modify semantics. */
4725 if (old_rule && fm->delete_reason != OFPRR_EVICTION) {
4726 bool change_cookie = (fm->modify_cookie
4727 && fm->new_cookie != OVS_BE64_MAX
4728 && fm->new_cookie != old_rule->flow_cookie);
4729
4730 ovs_mutex_lock(&old_rule->mutex);
4731 if (fm->command != OFPFC_ADD) {
4732 rule->idle_timeout = old_rule->idle_timeout;
4733 rule->hard_timeout = old_rule->hard_timeout;
4734 *CONST_CAST(uint16_t *, &rule->importance) = old_rule->importance;
4735 rule->flags = old_rule->flags;
4736 rule->created = old_rule->created;
4737 }
4738 if (!change_cookie) {
4739 rule->flow_cookie = old_rule->flow_cookie;
4740 }
4741 ovs_mutex_unlock(&old_rule->mutex);
4742 }
4743 ovs_mutex_unlock(&rule->mutex);
4744
4745 /* Construct rule, initializing derived state. */
4746 error = ofproto->ofproto_class->rule_construct(rule);
4747 if (error) {
4748 ofproto_rule_destroy__(rule);
4749 return error;
4750 }
4751
4752 rule->removed = true; /* Not yet in ofproto data structures. */
4753
4754 *new_rule = rule;
4755 return 0;
4756 }
4757
4758 static void
4759 replace_rule_start(struct ofproto *ofproto, cls_version_t version,
4760 struct rule *old_rule, struct rule *new_rule,
4761 struct cls_conjunction *conjs, size_t n_conjs)
4762 {
4763 struct oftable *table = &ofproto->tables[new_rule->table_id];
4764
4765 /* 'old_rule' may be either an evicted rule or replaced rule. */
4766 if (old_rule) {
4767 /* Mark the old rule for removal in the next version. */
4768 cls_rule_make_invisible_in_version(&old_rule->cr, version);
4769 } else {
4770 table->n_flows++;
4771 }
4772 /* Insert flow to the classifier, so that later flow_mods may relate
4773 * to it. This is reversible, in case later errors require this to
4774 * be reverted. */
4775 ofproto_rule_insert__(ofproto, new_rule);
4776 /* Make the new rule visible for classifier lookups only from the next
4777 * version. */
4778 classifier_insert(&table->cls, &new_rule->cr, version, conjs, n_conjs);
4779 }
4780
4781 static void replace_rule_revert(struct ofproto *ofproto,
4782 struct rule *old_rule, struct rule *new_rule)
4783 {
4784 struct oftable *table = &ofproto->tables[new_rule->table_id];
4785
4786 if (old_rule) {
4787 /* Restore the original visibility of the old rule. */
4788 cls_rule_restore_visibility(&old_rule->cr);
4789 } else {
4790 /* Restore table's rule count. */
4791 table->n_flows--;
4792 }
4793
4794 /* Remove the new rule immediately. It was never visible to lookups. */
4795 if (!classifier_remove(&table->cls, &new_rule->cr)) {
4796 OVS_NOT_REACHED();
4797 }
4798 ofproto_rule_remove__(ofproto, new_rule);
4799 /* The rule was not inserted to the ofproto provider, so we can
4800 * release it without deleting it from the ofproto provider. */
4801 ofproto_rule_unref(new_rule);
4802 }
4803
4804 /* Adds the 'new_rule', replacing the 'old_rule'. */
4805 static void
4806 replace_rule_finish(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4807 const struct flow_mod_requester *req,
4808 struct rule *old_rule, struct rule *new_rule,
4809 struct ovs_list *dead_cookies)
4810 OVS_REQUIRES(ofproto_mutex)
4811 {
4812 bool forward_stats = !(fm->flags & OFPUTIL_FF_RESET_COUNTS);
4813 struct rule *replaced_rule;
4814
4815 replaced_rule = fm->delete_reason != OFPRR_EVICTION ? old_rule : NULL;
4816
4817 /* Insert the new flow to the ofproto provider. A non-NULL 'replaced_rule'
4818 * is a duplicate rule the 'new_rule' is replacing. The provider should
4819 * link the stats from the old rule to the new one if 'forward_stats' is
4820 * 'true'. The 'replaced_rule' will be deleted right after this call. */
4821 ofproto->ofproto_class->rule_insert(new_rule, replaced_rule,
4822 forward_stats);
4823 learned_cookies_inc(ofproto, rule_get_actions(new_rule));
4824
4825 if (old_rule) {
4826 const struct rule_actions *old_actions = rule_get_actions(old_rule);
4827
4828 /* Remove the old rule from data structures. Removal from the
4829 * classifier and the deletion of the rule is RCU postponed by the
4830 * caller. */
4831 ofproto_rule_remove__(ofproto, old_rule);
4832 learned_cookies_dec(ofproto, old_actions, dead_cookies);
4833
4834 if (replaced_rule) {
4835 enum nx_flow_update_event event = fm->command == OFPFC_ADD
4836 ? NXFME_ADDED : NXFME_MODIFIED;
4837
4838 bool change_cookie = (fm->modify_cookie
4839 && fm->new_cookie != OVS_BE64_MAX
4840 && fm->new_cookie != old_rule->flow_cookie);
4841
4842 bool change_actions = !ofpacts_equal(fm->ofpacts,
4843 fm->ofpacts_len,
4844 old_actions->ofpacts,
4845 old_actions->ofpacts_len);
4846
4847 if (event != NXFME_MODIFIED || change_actions || change_cookie) {
4848 ofmonitor_report(ofproto->connmgr, new_rule, event, 0,
4849 req ? req->ofconn : NULL,
4850 req ? req->request->xid : 0,
4851 change_actions ? old_actions : NULL);
4852 }
4853 } else {
4854 /* XXX: This is slight duplication with delete_flows_finish__() */
4855
4856 old_rule->removed_reason = OFPRR_EVICTION;
4857
4858 ofmonitor_report(ofproto->connmgr, old_rule, NXFME_DELETED,
4859 OFPRR_EVICTION,
4860 req ? req->ofconn : NULL,
4861 req ? req->request->xid : 0, NULL);
4862 }
4863 }
4864 }
4865
4866 static enum ofperr
4867 modify_flows_start__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4868 OVS_REQUIRES(ofproto_mutex)
4869 {
4870 struct ofputil_flow_mod *fm = &ofm->fm;
4871 struct rule_collection *old_rules = &ofm->old_rules;
4872 struct rule_collection *new_rules = &ofm->new_rules;
4873 enum ofperr error;
4874
4875 rule_collection_init(new_rules);
4876
4877 if (old_rules->n > 0) {
4878 struct cls_conjunction *conjs;
4879 size_t n_conjs;
4880 size_t i;
4881
4882 /* Create a new 'modified' rule for each old rule. */
4883 for (i = 0; i < old_rules->n; i++) {
4884 struct rule *old_rule = old_rules->rules[i];
4885 struct rule *new_rule;
4886 struct cls_rule cr;
4887
4888 cls_rule_clone(&cr, &old_rule->cr);
4889 error = replace_rule_create(ofproto, fm, &cr, old_rule->table_id,
4890 old_rule, &new_rule);
4891 if (!error) {
4892 rule_collection_add(new_rules, new_rule);
4893 } else {
4894 rule_collection_unref(new_rules);
4895 rule_collection_destroy(new_rules);
4896 return error;
4897 }
4898 }
4899 ovs_assert(new_rules->n == old_rules->n);
4900
4901 get_conjunctions(fm, &conjs, &n_conjs);
4902 for (i = 0; i < old_rules->n; i++) {
4903 replace_rule_start(ofproto, ofm->version, old_rules->rules[i],
4904 new_rules->rules[i], conjs, n_conjs);
4905 }
4906 free(conjs);
4907 } else if (!(fm->cookie_mask != htonll(0)
4908 || fm->new_cookie == OVS_BE64_MAX)) {
4909 /* No match, add a new flow. */
4910 error = add_flow_start(ofproto, ofm);
4911 if (!error) {
4912 ovs_assert(fm->delete_reason == OFPRR_EVICTION
4913 || !old_rules->rules[0]);
4914 }
4915 new_rules->n = 1;
4916 } else {
4917 error = 0;
4918 }
4919
4920 return error;
4921 }
4922
4923 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4924 * failure.
4925 *
4926 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4927 * if any. */
4928 static enum ofperr
4929 modify_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4930 OVS_REQUIRES(ofproto_mutex)
4931 {
4932 struct ofputil_flow_mod *fm = &ofm->fm;
4933 struct rule_collection *old_rules = &ofm->old_rules;
4934 struct rule_criteria criteria;
4935 enum ofperr error;
4936
4937 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
4938 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG_ANY);
4939 rule_criteria_require_rw(&criteria,
4940 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
4941 error = collect_rules_loose(ofproto, &criteria, old_rules);
4942 rule_criteria_destroy(&criteria);
4943
4944 if (!error) {
4945 error = modify_flows_start__(ofproto, ofm);
4946 }
4947
4948 if (error) {
4949 rule_collection_destroy(old_rules);
4950 }
4951 return error;
4952 }
4953
4954 static void
4955 modify_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4956 OVS_REQUIRES(ofproto_mutex)
4957 {
4958 struct rule_collection *old_rules = &ofm->old_rules;
4959 struct rule_collection *new_rules = &ofm->new_rules;
4960
4961 /* Old rules were not changed yet, only need to revert new rules. */
4962 if (old_rules->n == 0 && new_rules->n == 1) {
4963 add_flow_revert(ofproto, ofm);
4964 } else if (old_rules->n > 0) {
4965 for (size_t i = 0; i < old_rules->n; i++) {
4966 replace_rule_revert(ofproto, old_rules->rules[i],
4967 new_rules->rules[i]);
4968 }
4969 rule_collection_destroy(new_rules);
4970 rule_collection_destroy(old_rules);
4971 }
4972 }
4973
4974 static void
4975 modify_flows_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
4976 const struct flow_mod_requester *req)
4977 OVS_REQUIRES(ofproto_mutex)
4978 {
4979 struct ofputil_flow_mod *fm = &ofm->fm;
4980 struct rule_collection *old_rules = &ofm->old_rules;
4981 struct rule_collection *new_rules = &ofm->new_rules;
4982
4983 if (old_rules->n == 0 && new_rules->n == 1) {
4984 add_flow_finish(ofproto, ofm, req);
4985 } else if (old_rules->n > 0) {
4986 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
4987
4988 ovs_assert(new_rules->n == old_rules->n);
4989
4990 for (size_t i = 0; i < old_rules->n; i++) {
4991 replace_rule_finish(ofproto, fm, req, old_rules->rules[i],
4992 new_rules->rules[i], &dead_cookies);
4993 }
4994 learned_cookies_flush(ofproto, &dead_cookies);
4995 rule_collection_remove_postponed(old_rules);
4996
4997 send_buffered_packet(req, fm->buffer_id, new_rules->rules[0]);
4998 rule_collection_destroy(new_rules);
4999 }
5000 }
5001
5002 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
5003 * code on failure. */
5004 static enum ofperr
5005 modify_flow_start_strict(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5006 OVS_REQUIRES(ofproto_mutex)
5007 {
5008 struct ofputil_flow_mod *fm = &ofm->fm;
5009 struct rule_collection *old_rules = &ofm->old_rules;
5010 struct rule_criteria criteria;
5011 enum ofperr error;
5012
5013 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5014 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask, OFPP_ANY,
5015 OFPG_ANY);
5016 rule_criteria_require_rw(&criteria,
5017 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5018 error = collect_rules_strict(ofproto, &criteria, old_rules);
5019 rule_criteria_destroy(&criteria);
5020
5021 if (!error) {
5022 /* collect_rules_strict() can return max 1 rule. */
5023 error = modify_flows_start__(ofproto, ofm);
5024 }
5025
5026 if (error) {
5027 rule_collection_destroy(old_rules);
5028 }
5029 return error;
5030 }
5031 \f
5032 /* OFPFC_DELETE implementation. */
5033
5034 static void
5035 delete_flows_start__(struct ofproto *ofproto, cls_version_t version,
5036 const struct rule_collection *rules)
5037 OVS_REQUIRES(ofproto_mutex)
5038 {
5039 for (size_t i = 0; i < rules->n; i++) {
5040 struct rule *rule = rules->rules[i];
5041 struct oftable *table = &ofproto->tables[rule->table_id];
5042
5043 table->n_flows--;
5044 cls_rule_make_invisible_in_version(&rule->cr, version);
5045 }
5046 }
5047
5048 static void
5049 delete_flows_finish__(struct ofproto *ofproto,
5050 struct rule_collection *rules,
5051 enum ofp_flow_removed_reason reason,
5052 const struct flow_mod_requester *req)
5053 OVS_REQUIRES(ofproto_mutex)
5054 {
5055 if (rules->n) {
5056 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5057
5058 for (size_t i = 0; i < rules->n; i++) {
5059 struct rule *rule = rules->rules[i];
5060
5061 /* This value will be used to send the flow removed message right
5062 * before the rule is actually destroyed. */
5063 rule->removed_reason = reason;
5064
5065 ofmonitor_report(ofproto->connmgr, rule, NXFME_DELETED, reason,
5066 req ? req->ofconn : NULL,
5067 req ? req->request->xid : 0, NULL);
5068 ofproto_rule_remove__(ofproto, rule);
5069 learned_cookies_dec(ofproto, rule_get_actions(rule),
5070 &dead_cookies);
5071 }
5072 rule_collection_remove_postponed(rules);
5073
5074 learned_cookies_flush(ofproto, &dead_cookies);
5075 }
5076 }
5077
5078 /* Deletes the rules listed in 'rules'.
5079 * The deleted rules will become invisible to the lookups in the next version.
5080 * Destroys 'rules'. */
5081 static void
5082 delete_flows__(struct rule_collection *rules,
5083 enum ofp_flow_removed_reason reason,
5084 const struct flow_mod_requester *req)
5085 OVS_REQUIRES(ofproto_mutex)
5086 {
5087 if (rules->n) {
5088 struct ofproto *ofproto = rules->rules[0]->ofproto;
5089
5090 delete_flows_start__(ofproto, ofproto->tables_version + 1, rules);
5091 ofproto_bump_tables_version(ofproto);
5092 delete_flows_finish__(ofproto, rules, reason, req);
5093 ofmonitor_flush(ofproto->connmgr);
5094 }
5095 }
5096
5097 /* Implements OFPFC_DELETE. */
5098 static enum ofperr
5099 delete_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5100 OVS_REQUIRES(ofproto_mutex)
5101 {
5102 const struct ofputil_flow_mod *fm = &ofm->fm;
5103 struct rule_collection *rules = &ofm->old_rules;
5104 struct rule_criteria criteria;
5105 enum ofperr error;
5106
5107 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
5108 fm->cookie, fm->cookie_mask, fm->out_port,
5109 fm->out_group);
5110 rule_criteria_require_rw(&criteria,
5111 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5112 error = collect_rules_loose(ofproto, &criteria, rules);
5113 rule_criteria_destroy(&criteria);
5114
5115 if (!error) {
5116 delete_flows_start__(ofproto, ofm->version, rules);
5117 }
5118
5119 return error;
5120 }
5121
5122 static void
5123 delete_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5124 OVS_REQUIRES(ofproto_mutex)
5125 {
5126 struct rule_collection *rules = &ofm->old_rules;
5127
5128 for (size_t i = 0; i < rules->n; i++) {
5129 struct rule *rule = rules->rules[i];
5130 struct oftable *table = &ofproto->tables[rule->table_id];
5131
5132 /* Restore table's rule count. */
5133 table->n_flows++;
5134
5135 /* Restore the original visibility of the rule. */
5136 cls_rule_restore_visibility(&rule->cr);
5137 }
5138 rule_collection_destroy(rules);
5139 }
5140
5141 static void
5142 delete_flows_finish(struct ofproto *ofproto,
5143 struct ofproto_flow_mod *ofm,
5144 const struct flow_mod_requester *req)
5145 OVS_REQUIRES(ofproto_mutex)
5146 {
5147 delete_flows_finish__(ofproto, &ofm->old_rules, ofm->fm.delete_reason,
5148 req);
5149 }
5150
5151 /* Implements OFPFC_DELETE_STRICT. */
5152 static enum ofperr
5153 delete_flow_start_strict(struct ofproto *ofproto,
5154 struct ofproto_flow_mod *ofm)
5155 OVS_REQUIRES(ofproto_mutex)
5156 {
5157 const struct ofputil_flow_mod *fm = &ofm->fm;
5158 struct rule_collection *rules = &ofm->old_rules;
5159 struct rule_criteria criteria;
5160 enum ofperr error;
5161
5162 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5163 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask,
5164 fm->out_port, fm->out_group);
5165 rule_criteria_require_rw(&criteria,
5166 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5167 error = collect_rules_strict(ofproto, &criteria, rules);
5168 rule_criteria_destroy(&criteria);
5169
5170 if (!error) {
5171 delete_flows_start__(ofproto, ofm->version, rules);
5172 }
5173
5174 return error;
5175 }
5176
5177 /* This may only be called by rule_destroy_cb()! */
5178 static void
5179 ofproto_rule_send_removed(struct rule *rule)
5180 OVS_EXCLUDED(ofproto_mutex)
5181 {
5182 struct ofputil_flow_removed fr;
5183 long long int used;
5184
5185 minimatch_expand(&rule->cr.match, &fr.match);
5186 fr.priority = rule->cr.priority;
5187
5188 ovs_mutex_lock(&ofproto_mutex);
5189 fr.cookie = rule->flow_cookie;
5190 fr.reason = rule->removed_reason;
5191 fr.table_id = rule->table_id;
5192 calc_duration(rule->created, time_msec(),
5193 &fr.duration_sec, &fr.duration_nsec);
5194 ovs_mutex_lock(&rule->mutex);
5195 fr.idle_timeout = rule->idle_timeout;
5196 fr.hard_timeout = rule->hard_timeout;
5197 ovs_mutex_unlock(&rule->mutex);
5198 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
5199 &fr.byte_count, &used);
5200 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
5201 ovs_mutex_unlock(&ofproto_mutex);
5202 }
5203
5204 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
5205 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
5206 * ofproto.
5207 *
5208 * ofproto implementation ->run() functions should use this function to expire
5209 * OpenFlow flows. */
5210 void
5211 ofproto_rule_expire(struct rule *rule, uint8_t reason)
5212 OVS_REQUIRES(ofproto_mutex)
5213 {
5214 struct rule_collection rules;
5215
5216 rules.rules = rules.stub;
5217 rules.n = 1;
5218 rules.stub[0] = rule;
5219 delete_flows__(&rules, reason, NULL);
5220 }
5221
5222 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5223 * means "infinite". */
5224 static void
5225 reduce_timeout(uint16_t max, uint16_t *timeout)
5226 {
5227 if (max && (!*timeout || *timeout > max)) {
5228 *timeout = max;
5229 }
5230 }
5231
5232 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
5233 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
5234 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
5235 *
5236 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
5237 void
5238 ofproto_rule_reduce_timeouts(struct rule *rule,
5239 uint16_t idle_timeout, uint16_t hard_timeout)
5240 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
5241 {
5242 if (!idle_timeout && !hard_timeout) {
5243 return;
5244 }
5245
5246 ovs_mutex_lock(&ofproto_mutex);
5247 if (list_is_empty(&rule->expirable)) {
5248 list_insert(&rule->ofproto->expirable, &rule->expirable);
5249 }
5250 ovs_mutex_unlock(&ofproto_mutex);
5251
5252 ovs_mutex_lock(&rule->mutex);
5253 reduce_timeout(idle_timeout, &rule->idle_timeout);
5254 reduce_timeout(hard_timeout, &rule->hard_timeout);
5255 ovs_mutex_unlock(&rule->mutex);
5256 }
5257 \f
5258 static enum ofperr
5259 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5260 OVS_EXCLUDED(ofproto_mutex)
5261 {
5262 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5263 struct ofproto_flow_mod ofm;
5264 uint64_t ofpacts_stub[1024 / 8];
5265 struct ofpbuf ofpacts;
5266 enum ofperr error;
5267
5268 error = reject_slave_controller(ofconn);
5269 if (error) {
5270 goto exit;
5271 }
5272
5273 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5274 error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn),
5275 &ofpacts,
5276 u16_to_ofp(ofproto->max_ports),
5277 ofproto->n_tables);
5278 if (!error) {
5279 error = ofproto_check_ofpacts(ofproto, ofm.fm.ofpacts,
5280 ofm.fm.ofpacts_len);
5281 }
5282 if (!error) {
5283 struct flow_mod_requester req;
5284
5285 req.ofconn = ofconn;
5286 req.request = oh;
5287 error = handle_flow_mod__(ofproto, &ofm, &req);
5288 }
5289 if (error) {
5290 goto exit_free_ofpacts;
5291 }
5292
5293 ofconn_report_flow_mod(ofconn, ofm.fm.command);
5294
5295 exit_free_ofpacts:
5296 ofpbuf_uninit(&ofpacts);
5297 exit:
5298 return error;
5299 }
5300
5301 static enum ofperr
5302 handle_flow_mod__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5303 const struct flow_mod_requester *req)
5304 OVS_EXCLUDED(ofproto_mutex)
5305 {
5306 enum ofperr error;
5307
5308 ovs_mutex_lock(&ofproto_mutex);
5309 ofm->version = ofproto->tables_version + 1;
5310 error = ofproto_flow_mod_start(ofproto, ofm);
5311 if (!error) {
5312 ofproto_bump_tables_version(ofproto);
5313 ofproto_flow_mod_finish(ofproto, ofm, req);
5314 }
5315 ofmonitor_flush(ofproto->connmgr);
5316 ovs_mutex_unlock(&ofproto_mutex);
5317
5318 run_rule_executes(ofproto);
5319 return error;
5320 }
5321
5322 static enum ofperr
5323 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
5324 {
5325 struct ofputil_role_request request;
5326 struct ofputil_role_request reply;
5327 struct ofpbuf *buf;
5328 enum ofperr error;
5329
5330 error = ofputil_decode_role_message(oh, &request);
5331 if (error) {
5332 return error;
5333 }
5334
5335 if (request.role != OFPCR12_ROLE_NOCHANGE) {
5336 if (request.role != OFPCR12_ROLE_EQUAL
5337 && request.have_generation_id
5338 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
5339 return OFPERR_OFPRRFC_STALE;
5340 }
5341
5342 ofconn_set_role(ofconn, request.role);
5343 }
5344
5345 reply.role = ofconn_get_role(ofconn);
5346 reply.have_generation_id = ofconn_get_master_election_id(
5347 ofconn, &reply.generation_id);
5348 buf = ofputil_encode_role_reply(oh, &reply);
5349 ofconn_send_reply(ofconn, buf);
5350
5351 return 0;
5352 }
5353
5354 static enum ofperr
5355 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
5356 const struct ofp_header *oh)
5357 {
5358 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
5359 enum ofputil_protocol cur, next;
5360
5361 cur = ofconn_get_protocol(ofconn);
5362 next = ofputil_protocol_set_tid(cur, msg->set != 0);
5363 ofconn_set_protocol(ofconn, next);
5364
5365 return 0;
5366 }
5367
5368 static enum ofperr
5369 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
5370 {
5371 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
5372 enum ofputil_protocol cur, next;
5373 enum ofputil_protocol next_base;
5374
5375 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
5376 if (!next_base) {
5377 return OFPERR_OFPBRC_EPERM;
5378 }
5379
5380 cur = ofconn_get_protocol(ofconn);
5381 next = ofputil_protocol_set_base(cur, next_base);
5382 ofconn_set_protocol(ofconn, next);
5383
5384 return 0;
5385 }
5386
5387 static enum ofperr
5388 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
5389 const struct ofp_header *oh)
5390 {
5391 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
5392 uint32_t format;
5393
5394 format = ntohl(msg->format);
5395 if (format != NXPIF_OPENFLOW10 && format != NXPIF_NXM) {
5396 return OFPERR_OFPBRC_EPERM;
5397 }
5398
5399 ofconn_set_packet_in_format(ofconn, format);
5400 return 0;
5401 }
5402
5403 static enum ofperr
5404 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
5405 {
5406 struct ofputil_async_cfg ac = OFPUTIL_ASYNC_CFG_INIT;
5407 enum ofperr error;
5408
5409 error = ofputil_decode_set_async_config(oh, false, &ac);
5410 if (error) {
5411 return error;
5412 }
5413
5414 ofconn_set_async_config(ofconn, &ac);
5415 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
5416 !ofconn_get_miss_send_len(ofconn)) {
5417 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
5418 }
5419
5420 return 0;
5421 }
5422
5423 static enum ofperr
5424 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
5425 {
5426 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
5427 ofconn_send_reply(ofconn, ofputil_encode_get_async_config(oh, &ac));
5428
5429 return 0;
5430 }
5431
5432 static enum ofperr
5433 handle_nxt_set_controller_id(struct ofconn *ofconn,
5434 const struct ofp_header *oh)
5435 {
5436 const struct nx_controller_id *nci = ofpmsg_body(oh);
5437
5438 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
5439 return OFPERR_NXBRC_MUST_BE_ZERO;
5440 }
5441
5442 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
5443 return 0;
5444 }
5445
5446 static enum ofperr
5447 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
5448 {
5449 struct ofpbuf *buf;
5450
5451 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
5452 ? OFPRAW_OFPT10_BARRIER_REPLY
5453 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
5454 ofconn_send_reply(ofconn, buf);
5455 return 0;
5456 }
5457
5458 static void
5459 ofproto_compose_flow_refresh_update(const struct rule *rule,
5460 enum nx_flow_monitor_flags flags,
5461 struct ovs_list *msgs)
5462 OVS_REQUIRES(ofproto_mutex)
5463 {
5464 const struct rule_actions *actions;
5465 struct ofputil_flow_update fu;
5466 struct match match;
5467
5468 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
5469 ? NXFME_ADDED : NXFME_MODIFIED);
5470 fu.reason = 0;
5471 ovs_mutex_lock(&rule->mutex);
5472 fu.idle_timeout = rule->idle_timeout;
5473 fu.hard_timeout = rule->hard_timeout;
5474 ovs_mutex_unlock(&rule->mutex);
5475 fu.table_id = rule->table_id;
5476 fu.cookie = rule->flow_cookie;
5477 minimatch_expand(&rule->cr.match, &match);
5478 fu.match = &match;
5479 fu.priority = rule->cr.priority;
5480
5481 actions = flags & NXFMF_ACTIONS ? rule_get_actions(rule) : NULL;
5482 fu.ofpacts = actions ? actions->ofpacts : NULL;
5483 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
5484
5485 if (list_is_empty(msgs)) {
5486 ofputil_start_flow_update(msgs);
5487 }
5488 ofputil_append_flow_update(&fu, msgs);
5489 }
5490
5491 void
5492 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
5493 struct ovs_list *msgs)
5494 OVS_REQUIRES(ofproto_mutex)
5495 {
5496 size_t i;
5497
5498 for (i = 0; i < rules->n; i++) {
5499 struct rule *rule = rules->rules[i];
5500 enum nx_flow_monitor_flags flags = rule->monitor_flags;
5501 rule->monitor_flags = 0;
5502
5503 ofproto_compose_flow_refresh_update(rule, flags, msgs);
5504 }
5505 }
5506
5507 static void
5508 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
5509 struct rule *rule, uint64_t seqno,
5510 struct rule_collection *rules)
5511 OVS_REQUIRES(ofproto_mutex)
5512 {
5513 enum nx_flow_monitor_flags update;
5514
5515 if (rule_is_hidden(rule)) {
5516 return;
5517 }
5518
5519 if (!ofproto_rule_has_out_port(rule, m->out_port)) {
5520 return;
5521 }
5522
5523 if (seqno) {
5524 if (rule->add_seqno > seqno) {
5525 update = NXFMF_ADD | NXFMF_MODIFY;
5526 } else if (rule->modify_seqno > seqno) {
5527 update = NXFMF_MODIFY;
5528 } else {
5529 return;
5530 }
5531
5532 if (!(m->flags & update)) {
5533 return;
5534 }
5535 } else {
5536 update = NXFMF_INITIAL;
5537 }
5538
5539 if (!rule->monitor_flags) {
5540 rule_collection_add(rules, rule);
5541 }
5542 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
5543 }
5544
5545 static void
5546 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
5547 uint64_t seqno,
5548 struct rule_collection *rules)
5549 OVS_REQUIRES(ofproto_mutex)
5550 {
5551 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
5552 const struct oftable *table;
5553 struct cls_rule target;
5554
5555 cls_rule_init_from_minimatch(&target, &m->match, 0);
5556 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
5557 struct rule *rule;
5558
5559 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &target, CLS_MAX_VERSION) {
5560 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
5561 }
5562 }
5563 cls_rule_destroy(&target);
5564 }
5565
5566 static void
5567 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
5568 struct rule_collection *rules)
5569 OVS_REQUIRES(ofproto_mutex)
5570 {
5571 if (m->flags & NXFMF_INITIAL) {
5572 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
5573 }
5574 }
5575
5576 void
5577 ofmonitor_collect_resume_rules(struct ofmonitor *m,
5578 uint64_t seqno, struct rule_collection *rules)
5579 OVS_REQUIRES(ofproto_mutex)
5580 {
5581 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
5582 }
5583
5584 static enum ofperr
5585 flow_monitor_delete(struct ofconn *ofconn, uint32_t id)
5586 OVS_REQUIRES(ofproto_mutex)
5587 {
5588 struct ofmonitor *m;
5589 enum ofperr error;
5590
5591 m = ofmonitor_lookup(ofconn, id);
5592 if (m) {
5593 ofmonitor_destroy(m);
5594 error = 0;
5595 } else {
5596 error = OFPERR_OFPMOFC_UNKNOWN_MONITOR;
5597 }
5598
5599 return error;
5600 }
5601
5602 static enum ofperr
5603 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
5604 OVS_EXCLUDED(ofproto_mutex)
5605 {
5606 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5607 struct ofmonitor **monitors;
5608 size_t n_monitors, allocated_monitors;
5609 struct rule_collection rules;
5610 struct ovs_list replies;
5611 enum ofperr error;
5612 struct ofpbuf b;
5613 size_t i;
5614
5615 ofpbuf_use_const(&b, oh, ntohs(oh->length));
5616 monitors = NULL;
5617 n_monitors = allocated_monitors = 0;
5618
5619 ovs_mutex_lock(&ofproto_mutex);
5620 for (;;) {
5621 struct ofputil_flow_monitor_request request;
5622 struct ofmonitor *m;
5623 int retval;
5624
5625 retval = ofputil_decode_flow_monitor_request(&request, &b);
5626 if (retval == EOF) {
5627 break;
5628 } else if (retval) {
5629 error = retval;
5630 goto error;
5631 }
5632
5633 if (request.table_id != 0xff
5634 && request.table_id >= ofproto->n_tables) {
5635 error = OFPERR_OFPBRC_BAD_TABLE_ID;
5636 goto error;
5637 }
5638
5639 error = ofmonitor_create(&request, ofconn, &m);
5640 if (error) {
5641 goto error;
5642 }
5643
5644 if (n_monitors >= allocated_monitors) {
5645 monitors = x2nrealloc(monitors, &allocated_monitors,
5646 sizeof *monitors);
5647 }
5648 monitors[n_monitors++] = m;
5649 }
5650
5651 rule_collection_init(&rules);
5652 for (i = 0; i < n_monitors; i++) {
5653 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
5654 }
5655
5656 ofpmp_init(&replies, oh);
5657 ofmonitor_compose_refresh_updates(&rules, &replies);
5658 ovs_mutex_unlock(&ofproto_mutex);
5659
5660 rule_collection_destroy(&rules);
5661
5662 ofconn_send_replies(ofconn, &replies);
5663 free(monitors);
5664
5665 return 0;
5666
5667 error:
5668 for (i = 0; i < n_monitors; i++) {
5669 ofmonitor_destroy(monitors[i]);
5670 }
5671 free(monitors);
5672 ovs_mutex_unlock(&ofproto_mutex);
5673
5674 return error;
5675 }
5676
5677 static enum ofperr
5678 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
5679 OVS_EXCLUDED(ofproto_mutex)
5680 {
5681 enum ofperr error;
5682 uint32_t id;
5683
5684 id = ofputil_decode_flow_monitor_cancel(oh);
5685
5686 ovs_mutex_lock(&ofproto_mutex);
5687 error = flow_monitor_delete(ofconn, id);
5688 ovs_mutex_unlock(&ofproto_mutex);
5689
5690 return error;
5691 }
5692
5693 /* Meters implementation.
5694 *
5695 * Meter table entry, indexed by the OpenFlow meter_id.
5696 * 'created' is used to compute the duration for meter stats.
5697 * 'list rules' is needed so that we can delete the dependent rules when the
5698 * meter table entry is deleted.
5699 * 'provider_meter_id' is for the provider's private use.
5700 */
5701 struct meter {
5702 long long int created; /* Time created. */
5703 struct ovs_list rules; /* List of "struct rule_dpif"s. */
5704 ofproto_meter_id provider_meter_id;
5705 uint16_t flags; /* Meter flags. */
5706 uint16_t n_bands; /* Number of meter bands. */
5707 struct ofputil_meter_band *bands;
5708 };
5709
5710 /*
5711 * This is used in instruction validation at flow set-up time,
5712 * as flows may not use non-existing meters.
5713 * Return value of UINT32_MAX signifies an invalid meter.
5714 */
5715 static uint32_t
5716 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
5717 {
5718 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
5719 const struct meter *meter = ofproto->meters[of_meter_id];
5720 if (meter) {
5721 return meter->provider_meter_id.uint32;
5722 }
5723 }
5724 return UINT32_MAX;
5725 }
5726
5727 /* Finds the meter invoked by 'rule''s actions and adds 'rule' to the meter's
5728 * list of rules. */
5729 static void
5730 meter_insert_rule(struct rule *rule)
5731 {
5732 const struct rule_actions *a = rule_get_actions(rule);
5733 uint32_t meter_id = ofpacts_get_meter(a->ofpacts, a->ofpacts_len);
5734 struct meter *meter = rule->ofproto->meters[meter_id];
5735
5736 list_insert(&meter->rules, &rule->meter_list_node);
5737 }
5738
5739 static void
5740 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
5741 {
5742 free(meter->bands);
5743
5744 meter->flags = config->flags;
5745 meter->n_bands = config->n_bands;
5746 meter->bands = xmemdup(config->bands,
5747 config->n_bands * sizeof *meter->bands);
5748 }
5749
5750 static struct meter *
5751 meter_create(const struct ofputil_meter_config *config,
5752 ofproto_meter_id provider_meter_id)
5753 {
5754 struct meter *meter;
5755
5756 meter = xzalloc(sizeof *meter);
5757 meter->provider_meter_id = provider_meter_id;
5758 meter->created = time_msec();
5759 list_init(&meter->rules);
5760
5761 meter_update(meter, config);
5762
5763 return meter;
5764 }
5765
5766 static void
5767 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5768 OVS_REQUIRES(ofproto_mutex)
5769 {
5770 uint32_t mid;
5771 for (mid = first; mid <= last; ++mid) {
5772 struct meter *meter = ofproto->meters[mid];
5773 if (meter) {
5774 ofproto->meters[mid] = NULL;
5775 ofproto->ofproto_class->meter_del(ofproto,
5776 meter->provider_meter_id);
5777 free(meter->bands);
5778 free(meter);
5779 }
5780 }
5781 }
5782
5783 static enum ofperr
5784 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5785 {
5786 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5787 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5788 enum ofperr error;
5789
5790 if (*meterp) {
5791 return OFPERR_OFPMMFC_METER_EXISTS;
5792 }
5793
5794 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5795 &mm->meter);
5796 if (!error) {
5797 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5798 *meterp = meter_create(&mm->meter, provider_meter_id);
5799 }
5800 return error;
5801 }
5802
5803 static enum ofperr
5804 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5805 {
5806 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5807 enum ofperr error;
5808 uint32_t provider_meter_id;
5809
5810 if (!meter) {
5811 return OFPERR_OFPMMFC_UNKNOWN_METER;
5812 }
5813
5814 provider_meter_id = meter->provider_meter_id.uint32;
5815 error = ofproto->ofproto_class->meter_set(ofproto,
5816 &meter->provider_meter_id,
5817 &mm->meter);
5818 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5819 if (!error) {
5820 meter_update(meter, &mm->meter);
5821 }
5822 return error;
5823 }
5824
5825 static enum ofperr
5826 handle_delete_meter(struct ofconn *ofconn, struct ofputil_meter_mod *mm)
5827 OVS_EXCLUDED(ofproto_mutex)
5828 {
5829 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5830 uint32_t meter_id = mm->meter.meter_id;
5831 struct rule_collection rules;
5832 enum ofperr error = 0;
5833 uint32_t first, last;
5834
5835 if (meter_id == OFPM13_ALL) {
5836 first = 1;
5837 last = ofproto->meter_features.max_meters;
5838 } else {
5839 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5840 return 0;
5841 }
5842 first = last = meter_id;
5843 }
5844
5845 /* First delete the rules that use this meter. If any of those rules are
5846 * currently being modified, postpone the whole operation until later. */
5847 rule_collection_init(&rules);
5848 ovs_mutex_lock(&ofproto_mutex);
5849 for (meter_id = first; meter_id <= last; ++meter_id) {
5850 struct meter *meter = ofproto->meters[meter_id];
5851 if (meter && !list_is_empty(&meter->rules)) {
5852 struct rule *rule;
5853
5854 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5855 rule_collection_add(&rules, rule);
5856 }
5857 }
5858 }
5859 delete_flows__(&rules, OFPRR_METER_DELETE, NULL);
5860
5861 /* Delete the meters. */
5862 meter_delete(ofproto, first, last);
5863
5864 ovs_mutex_unlock(&ofproto_mutex);
5865
5866 return error;
5867 }
5868
5869 static enum ofperr
5870 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5871 {
5872 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5873 struct ofputil_meter_mod mm;
5874 uint64_t bands_stub[256 / 8];
5875 struct ofpbuf bands;
5876 uint32_t meter_id;
5877 enum ofperr error;
5878
5879 error = reject_slave_controller(ofconn);
5880 if (error) {
5881 return error;
5882 }
5883
5884 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5885
5886 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5887 if (error) {
5888 goto exit_free_bands;
5889 }
5890
5891 meter_id = mm.meter.meter_id;
5892
5893 if (mm.command != OFPMC13_DELETE) {
5894 /* Fails also when meters are not implemented by the provider. */
5895 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5896 error = OFPERR_OFPMMFC_INVALID_METER;
5897 goto exit_free_bands;
5898 } else if (meter_id > ofproto->meter_features.max_meters) {
5899 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5900 goto exit_free_bands;
5901 }
5902 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5903 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5904 goto exit_free_bands;
5905 }
5906 }
5907
5908 switch (mm.command) {
5909 case OFPMC13_ADD:
5910 error = handle_add_meter(ofproto, &mm);
5911 break;
5912
5913 case OFPMC13_MODIFY:
5914 error = handle_modify_meter(ofproto, &mm);
5915 break;
5916
5917 case OFPMC13_DELETE:
5918 error = handle_delete_meter(ofconn, &mm);
5919 break;
5920
5921 default:
5922 error = OFPERR_OFPMMFC_BAD_COMMAND;
5923 break;
5924 }
5925
5926 if (!error) {
5927 struct ofputil_requestforward rf;
5928 rf.xid = oh->xid;
5929 rf.reason = OFPRFR_METER_MOD;
5930 rf.meter_mod = &mm;
5931 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
5932 }
5933
5934 exit_free_bands:
5935 ofpbuf_uninit(&bands);
5936 return error;
5937 }
5938
5939 static enum ofperr
5940 handle_meter_features_request(struct ofconn *ofconn,
5941 const struct ofp_header *request)
5942 {
5943 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5944 struct ofputil_meter_features features;
5945 struct ofpbuf *b;
5946
5947 if (ofproto->ofproto_class->meter_get_features) {
5948 ofproto->ofproto_class->meter_get_features(ofproto, &features);
5949 } else {
5950 memset(&features, 0, sizeof features);
5951 }
5952 b = ofputil_encode_meter_features_reply(&features, request);
5953
5954 ofconn_send_reply(ofconn, b);
5955 return 0;
5956 }
5957
5958 static enum ofperr
5959 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
5960 enum ofptype type)
5961 {
5962 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5963 struct ovs_list replies;
5964 uint64_t bands_stub[256 / 8];
5965 struct ofpbuf bands;
5966 uint32_t meter_id, first, last;
5967
5968 ofputil_decode_meter_request(request, &meter_id);
5969
5970 if (meter_id == OFPM13_ALL) {
5971 first = 1;
5972 last = ofproto->meter_features.max_meters;
5973 } else {
5974 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
5975 !ofproto->meters[meter_id]) {
5976 return OFPERR_OFPMMFC_UNKNOWN_METER;
5977 }
5978 first = last = meter_id;
5979 }
5980
5981 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5982 ofpmp_init(&replies, request);
5983
5984 for (meter_id = first; meter_id <= last; ++meter_id) {
5985 struct meter *meter = ofproto->meters[meter_id];
5986 if (!meter) {
5987 continue; /* Skip non-existing meters. */
5988 }
5989 if (type == OFPTYPE_METER_STATS_REQUEST) {
5990 struct ofputil_meter_stats stats;
5991
5992 stats.meter_id = meter_id;
5993
5994 /* Provider sets the packet and byte counts, we do the rest. */
5995 stats.flow_count = list_size(&meter->rules);
5996 calc_duration(meter->created, time_msec(),
5997 &stats.duration_sec, &stats.duration_nsec);
5998 stats.n_bands = meter->n_bands;
5999 ofpbuf_clear(&bands);
6000 stats.bands
6001 = ofpbuf_put_uninit(&bands,
6002 meter->n_bands * sizeof *stats.bands);
6003
6004 if (!ofproto->ofproto_class->meter_get(ofproto,
6005 meter->provider_meter_id,
6006 &stats)) {
6007 ofputil_append_meter_stats(&replies, &stats);
6008 }
6009 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
6010 struct ofputil_meter_config config;
6011
6012 config.meter_id = meter_id;
6013 config.flags = meter->flags;
6014 config.n_bands = meter->n_bands;
6015 config.bands = meter->bands;
6016 ofputil_append_meter_config(&replies, &config);
6017 }
6018 }
6019
6020 ofconn_send_replies(ofconn, &replies);
6021 ofpbuf_uninit(&bands);
6022 return 0;
6023 }
6024
6025 static bool
6026 ofproto_group_lookup__(const struct ofproto *ofproto, uint32_t group_id,
6027 struct ofgroup **group)
6028 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6029 {
6030 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
6031 hash_int(group_id, 0), &ofproto->groups) {
6032 if ((*group)->group_id == group_id) {
6033 return true;
6034 }
6035 }
6036
6037 return false;
6038 }
6039
6040 /* If the group exists, this function increments the groups's reference count.
6041 *
6042 * Make sure to call ofproto_group_unref() after no longer needing to maintain
6043 * a reference to the group. */
6044 bool
6045 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
6046 struct ofgroup **group)
6047 {
6048 bool found;
6049
6050 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6051 found = ofproto_group_lookup__(ofproto, group_id, group);
6052 if (found) {
6053 ofproto_group_ref(*group);
6054 }
6055 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6056 return found;
6057 }
6058
6059 static bool
6060 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
6061 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6062 {
6063 struct ofgroup *grp;
6064
6065 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
6066 hash_int(group_id, 0), &ofproto->groups) {
6067 if (grp->group_id == group_id) {
6068 return true;
6069 }
6070 }
6071 return false;
6072 }
6073
6074 static bool
6075 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
6076 OVS_EXCLUDED(ofproto->groups_rwlock)
6077 {
6078 bool exists;
6079
6080 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6081 exists = ofproto_group_exists__(ofproto, group_id);
6082 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6083
6084 return exists;
6085 }
6086
6087 static uint32_t
6088 group_get_ref_count(struct ofgroup *group)
6089 OVS_EXCLUDED(ofproto_mutex)
6090 {
6091 struct ofproto *ofproto = CONST_CAST(struct ofproto *, group->ofproto);
6092 struct rule_criteria criteria;
6093 struct rule_collection rules;
6094 struct match match;
6095 enum ofperr error;
6096 uint32_t count;
6097
6098 match_init_catchall(&match);
6099 rule_criteria_init(&criteria, 0xff, &match, 0, CLS_MAX_VERSION, htonll(0),
6100 htonll(0), OFPP_ANY, group->group_id);
6101 ovs_mutex_lock(&ofproto_mutex);
6102 error = collect_rules_loose(ofproto, &criteria, &rules);
6103 ovs_mutex_unlock(&ofproto_mutex);
6104 rule_criteria_destroy(&criteria);
6105
6106 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
6107
6108 rule_collection_destroy(&rules);
6109 return count;
6110 }
6111
6112 static void
6113 append_group_stats(struct ofgroup *group, struct ovs_list *replies)
6114 {
6115 struct ofputil_group_stats ogs;
6116 const struct ofproto *ofproto = group->ofproto;
6117 long long int now = time_msec();
6118 int error;
6119
6120 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
6121
6122 /* Provider sets the packet and byte counts, we do the rest. */
6123 ogs.ref_count = group_get_ref_count(group);
6124 ogs.n_buckets = group->n_buckets;
6125
6126 error = (ofproto->ofproto_class->group_get_stats
6127 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
6128 : EOPNOTSUPP);
6129 if (error) {
6130 ogs.packet_count = UINT64_MAX;
6131 ogs.byte_count = UINT64_MAX;
6132 memset(ogs.bucket_stats, 0xff,
6133 ogs.n_buckets * sizeof *ogs.bucket_stats);
6134 }
6135
6136 ogs.group_id = group->group_id;
6137 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
6138
6139 ofputil_append_group_stats(replies, &ogs);
6140
6141 free(ogs.bucket_stats);
6142 }
6143
6144 static void
6145 handle_group_request(struct ofconn *ofconn,
6146 const struct ofp_header *request, uint32_t group_id,
6147 void (*cb)(struct ofgroup *, struct ovs_list *replies))
6148 {
6149 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6150 struct ofgroup *group;
6151 struct ovs_list replies;
6152
6153 ofpmp_init(&replies, request);
6154 if (group_id == OFPG_ALL) {
6155 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6156 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
6157 cb(group, &replies);
6158 }
6159 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6160 } else {
6161 if (ofproto_group_lookup(ofproto, group_id, &group)) {
6162 cb(group, &replies);
6163 ofproto_group_unref(group);
6164 }
6165 }
6166 ofconn_send_replies(ofconn, &replies);
6167 }
6168
6169 static enum ofperr
6170 handle_group_stats_request(struct ofconn *ofconn,
6171 const struct ofp_header *request)
6172 {
6173 uint32_t group_id;
6174 enum ofperr error;
6175
6176 error = ofputil_decode_group_stats_request(request, &group_id);
6177 if (error) {
6178 return error;
6179 }
6180
6181 handle_group_request(ofconn, request, group_id, append_group_stats);
6182 return 0;
6183 }
6184
6185 static void
6186 append_group_desc(struct ofgroup *group, struct ovs_list *replies)
6187 {
6188 struct ofputil_group_desc gds;
6189
6190 gds.group_id = group->group_id;
6191 gds.type = group->type;
6192 gds.props = group->props;
6193
6194 ofputil_append_group_desc_reply(&gds, &group->buckets, replies);
6195 }
6196
6197 static enum ofperr
6198 handle_group_desc_stats_request(struct ofconn *ofconn,
6199 const struct ofp_header *request)
6200 {
6201 handle_group_request(ofconn, request,
6202 ofputil_decode_group_desc_request(request),
6203 append_group_desc);
6204 return 0;
6205 }
6206
6207 static enum ofperr
6208 handle_group_features_stats_request(struct ofconn *ofconn,
6209 const struct ofp_header *request)
6210 {
6211 struct ofproto *p = ofconn_get_ofproto(ofconn);
6212 struct ofpbuf *msg;
6213
6214 msg = ofputil_encode_group_features_reply(&p->ogf, request);
6215 if (msg) {
6216 ofconn_send_reply(ofconn, msg);
6217 }
6218
6219 return 0;
6220 }
6221
6222 static void
6223 put_queue_get_config_reply(struct ofport *port, uint32_t queue,
6224 struct ovs_list *replies)
6225 {
6226 struct ofputil_queue_config qc;
6227
6228 /* None of the existing queues have compatible properties, so we hard-code
6229 * omitting min_rate and max_rate. */
6230 qc.port = port->ofp_port;
6231 qc.queue = queue;
6232 qc.min_rate = UINT16_MAX;
6233 qc.max_rate = UINT16_MAX;
6234 ofputil_append_queue_get_config_reply(&qc, replies);
6235 }
6236
6237 static int
6238 handle_queue_get_config_request_for_port(struct ofport *port, uint32_t queue,
6239 struct ovs_list *replies)
6240 {
6241 struct smap details = SMAP_INITIALIZER(&details);
6242 if (queue != OFPQ_ALL) {
6243 int error = netdev_get_queue(port->netdev, queue, &details);
6244 switch (error) {
6245 case 0:
6246 put_queue_get_config_reply(port, queue, replies);
6247 break;
6248 case EOPNOTSUPP:
6249 case EINVAL:
6250 return OFPERR_OFPQOFC_BAD_QUEUE;
6251 default:
6252 return OFPERR_NXQOFC_QUEUE_ERROR;
6253 }
6254 } else {
6255 struct netdev_queue_dump queue_dump;
6256 uint32_t queue_id;
6257
6258 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump,
6259 port->netdev) {
6260 put_queue_get_config_reply(port, queue_id, replies);
6261 }
6262 }
6263 smap_destroy(&details);
6264 return 0;
6265 }
6266
6267 static enum ofperr
6268 handle_queue_get_config_request(struct ofconn *ofconn,
6269 const struct ofp_header *oh)
6270 {
6271 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6272 struct ovs_list replies;
6273 struct ofport *port;
6274 ofp_port_t req_port;
6275 uint32_t req_queue;
6276 enum ofperr error;
6277
6278 error = ofputil_decode_queue_get_config_request(oh, &req_port, &req_queue);
6279 if (error) {
6280 return error;
6281 }
6282
6283 ofputil_start_queue_get_config_reply(oh, &replies);
6284 if (req_port == OFPP_ANY) {
6285 error = OFPERR_OFPQOFC_BAD_QUEUE;
6286 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
6287 if (!handle_queue_get_config_request_for_port(port, req_queue,
6288 &replies)) {
6289 error = 0;
6290 }
6291 }
6292 } else {
6293 port = ofproto_get_port(ofproto, req_port);
6294 error = (port
6295 ? handle_queue_get_config_request_for_port(port, req_queue,
6296 &replies)
6297 : OFPERR_OFPQOFC_BAD_PORT);
6298 }
6299 if (!error) {
6300 ofconn_send_replies(ofconn, &replies);
6301 } else {
6302 ofpbuf_list_delete(&replies);
6303 }
6304
6305 return error;
6306 }
6307
6308 static enum ofperr
6309 init_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm,
6310 struct ofgroup **ofgroup)
6311 {
6312 enum ofperr error;
6313 const long long int now = time_msec();
6314
6315 if (gm->group_id > OFPG_MAX) {
6316 return OFPERR_OFPGMFC_INVALID_GROUP;
6317 }
6318 if (gm->type > OFPGT11_FF) {
6319 return OFPERR_OFPGMFC_BAD_TYPE;
6320 }
6321
6322 *ofgroup = ofproto->ofproto_class->group_alloc();
6323 if (!*ofgroup) {
6324 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
6325 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
6326 }
6327
6328 (*ofgroup)->ofproto = ofproto;
6329 *CONST_CAST(uint32_t *, &((*ofgroup)->group_id)) = gm->group_id;
6330 *CONST_CAST(enum ofp11_group_type *, &(*ofgroup)->type) = gm->type;
6331 *CONST_CAST(long long int *, &((*ofgroup)->created)) = now;
6332 *CONST_CAST(long long int *, &((*ofgroup)->modified)) = now;
6333 ovs_refcount_init(&(*ofgroup)->ref_count);
6334
6335 list_init(&(*ofgroup)->buckets);
6336 ofputil_bucket_clone_list(&(*ofgroup)->buckets, &gm->buckets, NULL);
6337
6338 *CONST_CAST(uint32_t *, &(*ofgroup)->n_buckets) =
6339 list_size(&(*ofgroup)->buckets);
6340
6341 memcpy(CONST_CAST(struct ofputil_group_props *, &(*ofgroup)->props),
6342 &gm->props, sizeof (struct ofputil_group_props));
6343
6344 /* Construct called BEFORE any locks are held. */
6345 error = ofproto->ofproto_class->group_construct(*ofgroup);
6346 if (error) {
6347 ofputil_bucket_list_destroy(&(*ofgroup)->buckets);
6348 ofproto->ofproto_class->group_dealloc(*ofgroup);
6349 }
6350 return error;
6351 }
6352
6353 /* Implements the OFPGC11_ADD operation specified by 'gm', adding a group to
6354 * 'ofproto''s group table. Returns 0 on success or an OpenFlow error code on
6355 * failure. */
6356 static enum ofperr
6357 add_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6358 {
6359 struct ofgroup *ofgroup;
6360 enum ofperr error;
6361
6362 /* Allocate new group and initialize it. */
6363 error = init_group(ofproto, gm, &ofgroup);
6364 if (error) {
6365 return error;
6366 }
6367
6368 /* We wrlock as late as possible to minimize the time we jam any other
6369 * threads: No visible state changes before acquiring the lock. */
6370 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6371
6372 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6373 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6374 goto unlock_out;
6375 }
6376
6377 if (ofproto_group_exists__(ofproto, gm->group_id)) {
6378 error = OFPERR_OFPGMFC_GROUP_EXISTS;
6379 goto unlock_out;
6380 }
6381
6382 if (!error) {
6383 /* Insert new group. */
6384 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
6385 hash_int(ofgroup->group_id, 0));
6386 ofproto->n_groups[ofgroup->type]++;
6387
6388 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6389 return error;
6390 }
6391
6392 unlock_out:
6393 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6394 ofproto->ofproto_class->group_destruct(ofgroup);
6395 ofputil_bucket_list_destroy(&ofgroup->buckets);
6396 ofproto->ofproto_class->group_dealloc(ofgroup);
6397
6398 return error;
6399 }
6400
6401 /* Adds all of the buckets from 'ofgroup' to 'new_ofgroup'. The buckets
6402 * already in 'new_ofgroup' will be placed just after the (copy of the) bucket
6403 * in 'ofgroup' with bucket ID 'command_bucket_id'. Special
6404 * 'command_bucket_id' values OFPG15_BUCKET_FIRST and OFPG15_BUCKET_LAST are
6405 * also honored. */
6406 static enum ofperr
6407 copy_buckets_for_insert_bucket(const struct ofgroup *ofgroup,
6408 struct ofgroup *new_ofgroup,
6409 uint32_t command_bucket_id)
6410 {
6411 struct ofputil_bucket *last = NULL;
6412
6413 if (command_bucket_id <= OFPG15_BUCKET_MAX) {
6414 /* Check here to ensure that a bucket corresponding to
6415 * command_bucket_id exists in the old bucket list.
6416 *
6417 * The subsequent search of below of new_ofgroup covers
6418 * both buckets in the old bucket list and buckets added
6419 * by the insert buckets group mod message this function processes. */
6420 if (!ofputil_bucket_find(&ofgroup->buckets, command_bucket_id)) {
6421 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6422 }
6423
6424 if (!list_is_empty(&new_ofgroup->buckets)) {
6425 last = ofputil_bucket_list_back(&new_ofgroup->buckets);
6426 }
6427 }
6428
6429 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, NULL);
6430
6431 if (ofputil_bucket_check_duplicate_id(&new_ofgroup->buckets)) {
6432 VLOG_INFO_RL(&rl, "Duplicate bucket id");
6433 return OFPERR_OFPGMFC_BUCKET_EXISTS;
6434 }
6435
6436 /* Rearrange list according to command_bucket_id */
6437 if (command_bucket_id == OFPG15_BUCKET_LAST) {
6438 if (!list_is_empty(&ofgroup->buckets)) {
6439 struct ofputil_bucket *new_first;
6440 const struct ofputil_bucket *first;
6441
6442 first = ofputil_bucket_list_front(&ofgroup->buckets);
6443 new_first = ofputil_bucket_find(&new_ofgroup->buckets,
6444 first->bucket_id);
6445
6446 list_splice(new_ofgroup->buckets.next, &new_first->list_node,
6447 &new_ofgroup->buckets);
6448 }
6449 } else if (command_bucket_id <= OFPG15_BUCKET_MAX && last) {
6450 struct ofputil_bucket *after;
6451
6452 /* Presence of bucket is checked above so after should never be NULL */
6453 after = ofputil_bucket_find(&new_ofgroup->buckets, command_bucket_id);
6454
6455 list_splice(after->list_node.next, new_ofgroup->buckets.next,
6456 last->list_node.next);
6457 }
6458
6459 return 0;
6460 }
6461
6462 /* Appends all of the a copy of all the buckets from 'ofgroup' to 'new_ofgroup'
6463 * with the exception of the bucket whose bucket id is 'command_bucket_id'.
6464 * Special 'command_bucket_id' values OFPG15_BUCKET_FIRST, OFPG15_BUCKET_LAST
6465 * and OFPG15_BUCKET_ALL are also honored. */
6466 static enum ofperr
6467 copy_buckets_for_remove_bucket(const struct ofgroup *ofgroup,
6468 struct ofgroup *new_ofgroup,
6469 uint32_t command_bucket_id)
6470 {
6471 const struct ofputil_bucket *skip = NULL;
6472
6473 if (command_bucket_id == OFPG15_BUCKET_ALL) {
6474 return 0;
6475 }
6476
6477 if (command_bucket_id == OFPG15_BUCKET_FIRST) {
6478 if (!list_is_empty(&ofgroup->buckets)) {
6479 skip = ofputil_bucket_list_front(&ofgroup->buckets);
6480 }
6481 } else if (command_bucket_id == OFPG15_BUCKET_LAST) {
6482 if (!list_is_empty(&ofgroup->buckets)) {
6483 skip = ofputil_bucket_list_back(&ofgroup->buckets);
6484 }
6485 } else {
6486 skip = ofputil_bucket_find(&ofgroup->buckets, command_bucket_id);
6487 if (!skip) {
6488 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6489 }
6490 }
6491
6492 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, skip);
6493
6494 return 0;
6495 }
6496
6497 /* Implements OFPGC11_MODIFY, OFPGC15_INSERT_BUCKET and
6498 * OFPGC15_REMOVE_BUCKET. Returns 0 on success or an OpenFlow error code
6499 * on failure.
6500 *
6501 * Note that the group is re-created and then replaces the old group in
6502 * ofproto's ofgroup hash map. Thus, the group is never altered while users of
6503 * the xlate module hold a pointer to the group. */
6504 static enum ofperr
6505 modify_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6506 {
6507 struct ofgroup *ofgroup, *new_ofgroup, *retiring;
6508 enum ofperr error;
6509
6510 error = init_group(ofproto, gm, &new_ofgroup);
6511 if (error) {
6512 return error;
6513 }
6514
6515 retiring = new_ofgroup;
6516
6517 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6518 if (!ofproto_group_lookup__(ofproto, gm->group_id, &ofgroup)) {
6519 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
6520 goto out;
6521 }
6522
6523 /* Ofproto's group write lock is held now. */
6524 if (ofgroup->type != gm->type
6525 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6526 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6527 goto out;
6528 }
6529
6530 /* Manipulate bucket list for bucket commands */
6531 if (gm->command == OFPGC15_INSERT_BUCKET) {
6532 error = copy_buckets_for_insert_bucket(ofgroup, new_ofgroup,
6533 gm->command_bucket_id);
6534 } else if (gm->command == OFPGC15_REMOVE_BUCKET) {
6535 error = copy_buckets_for_remove_bucket(ofgroup, new_ofgroup,
6536 gm->command_bucket_id);
6537 }
6538 if (error) {
6539 goto out;
6540 }
6541
6542 /* The group creation time does not change during modification. */
6543 *CONST_CAST(long long int *, &(new_ofgroup->created)) = ofgroup->created;
6544 *CONST_CAST(long long int *, &(new_ofgroup->modified)) = time_msec();
6545
6546 error = ofproto->ofproto_class->group_modify(new_ofgroup);
6547 if (error) {
6548 goto out;
6549 }
6550
6551 retiring = ofgroup;
6552 /* Replace ofgroup in ofproto's groups hash map with new_ofgroup. */
6553 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6554 hmap_insert(&ofproto->groups, &new_ofgroup->hmap_node,
6555 hash_int(new_ofgroup->group_id, 0));
6556 if (ofgroup->type != new_ofgroup->type) {
6557 ofproto->n_groups[ofgroup->type]--;
6558 ofproto->n_groups[new_ofgroup->type]++;
6559 }
6560
6561 out:
6562 ofproto_group_unref(retiring);
6563 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6564 return error;
6565 }
6566
6567 static void
6568 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
6569 OVS_RELEASES(ofproto->groups_rwlock)
6570 {
6571 struct match match;
6572 struct ofproto_flow_mod ofm;
6573
6574 /* Delete all flow entries containing this group in a group action */
6575 match_init_catchall(&match);
6576 flow_mod_init(&ofm.fm, &match, 0, NULL, 0, OFPFC_DELETE);
6577 ofm.fm.delete_reason = OFPRR_GROUP_DELETE;
6578 ofm.fm.out_group = ofgroup->group_id;
6579 ofm.fm.table_id = OFPTT_ALL;
6580 handle_flow_mod__(ofproto, &ofm, NULL);
6581
6582 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6583 /* No-one can find this group any more. */
6584 ofproto->n_groups[ofgroup->type]--;
6585 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6586 ofproto_group_unref(ofgroup);
6587 }
6588
6589 /* Implements OFPGC11_DELETE. */
6590 static void
6591 delete_group(struct ofproto *ofproto, uint32_t group_id)
6592 {
6593 struct ofgroup *ofgroup;
6594
6595 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6596 if (group_id == OFPG_ALL) {
6597 for (;;) {
6598 struct hmap_node *node = hmap_first(&ofproto->groups);
6599 if (!node) {
6600 break;
6601 }
6602 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
6603 delete_group__(ofproto, ofgroup);
6604 /* Lock for each node separately, so that we will not jam the
6605 * other threads for too long time. */
6606 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6607 }
6608 } else {
6609 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
6610 hash_int(group_id, 0), &ofproto->groups) {
6611 if (ofgroup->group_id == group_id) {
6612 delete_group__(ofproto, ofgroup);
6613 return;
6614 }
6615 }
6616 }
6617 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6618 }
6619
6620 /* Delete all groups from 'ofproto'.
6621 *
6622 * This is intended for use within an ofproto provider's 'destruct'
6623 * function. */
6624 void
6625 ofproto_group_delete_all(struct ofproto *ofproto)
6626 {
6627 delete_group(ofproto, OFPG_ALL);
6628 }
6629
6630 static enum ofperr
6631 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6632 {
6633 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6634 struct ofputil_group_mod gm;
6635 enum ofperr error;
6636
6637 error = reject_slave_controller(ofconn);
6638 if (error) {
6639 return error;
6640 }
6641
6642 error = ofputil_decode_group_mod(oh, &gm);
6643 if (error) {
6644 return error;
6645 }
6646
6647 switch (gm.command) {
6648 case OFPGC11_ADD:
6649 error = add_group(ofproto, &gm);
6650 break;
6651
6652 case OFPGC11_MODIFY:
6653 error = modify_group(ofproto, &gm);
6654 break;
6655
6656 case OFPGC11_DELETE:
6657 delete_group(ofproto, gm.group_id);
6658 error = 0;
6659 break;
6660
6661 case OFPGC15_INSERT_BUCKET:
6662 error = modify_group(ofproto, &gm);
6663 break;
6664
6665 case OFPGC15_REMOVE_BUCKET:
6666 error = modify_group(ofproto, &gm);
6667 break;
6668
6669 default:
6670 if (gm.command > OFPGC11_DELETE) {
6671 VLOG_INFO_RL(&rl, "%s: Invalid group_mod command type %d",
6672 ofproto->name, gm.command);
6673 }
6674 error = OFPERR_OFPGMFC_BAD_COMMAND;
6675 }
6676
6677 if (!error) {
6678 struct ofputil_requestforward rf;
6679 rf.xid = oh->xid;
6680 rf.reason = OFPRFR_GROUP_MOD;
6681 rf.group_mod = &gm;
6682 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6683 }
6684 ofputil_bucket_list_destroy(&gm.buckets);
6685
6686 return error;
6687 }
6688
6689 enum ofputil_table_miss
6690 ofproto_table_get_miss_config(const struct ofproto *ofproto, uint8_t table_id)
6691 {
6692 enum ofputil_table_miss miss;
6693
6694 atomic_read_relaxed(&ofproto->tables[table_id].miss_config, &miss);
6695 return miss;
6696 }
6697
6698 static void
6699 table_mod__(struct oftable *oftable,
6700 const struct ofputil_table_mod *tm)
6701 {
6702 if (tm->miss == OFPUTIL_TABLE_MISS_DEFAULT) {
6703 /* This is how an OFPT_TABLE_MOD decodes if it doesn't specify any
6704 * table-miss configuration (because the protocol used doesn't have
6705 * such a concept), so there's nothing to do. */
6706 } else {
6707 atomic_store_relaxed(&oftable->miss_config, tm->miss);
6708 }
6709
6710 unsigned int new_eviction = oftable->eviction;
6711 if (tm->eviction == OFPUTIL_TABLE_EVICTION_ON) {
6712 new_eviction |= EVICTION_OPENFLOW;
6713 } else if (tm->eviction == OFPUTIL_TABLE_EVICTION_OFF) {
6714 new_eviction &= ~EVICTION_OPENFLOW;
6715 }
6716
6717 if (new_eviction != oftable->eviction) {
6718 ovs_mutex_lock(&ofproto_mutex);
6719 oftable_configure_eviction(oftable, new_eviction,
6720 oftable->eviction_fields,
6721 oftable->n_eviction_fields);
6722 ovs_mutex_unlock(&ofproto_mutex);
6723 }
6724
6725 if (tm->vacancy != OFPUTIL_TABLE_VACANCY_DEFAULT) {
6726 ovs_mutex_lock(&ofproto_mutex);
6727 oftable->vacancy_enabled = (tm->vacancy == OFPUTIL_TABLE_VACANCY_ON
6728 ? OFPTC14_VACANCY_EVENTS
6729 : 0);
6730 oftable->vacancy_down = tm->table_vacancy.vacancy_down;
6731 oftable->vacancy_up = tm->table_vacancy.vacancy_up;
6732 ovs_mutex_unlock(&ofproto_mutex);
6733 }
6734 }
6735
6736 static enum ofperr
6737 table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
6738 {
6739 if (!check_table_id(ofproto, tm->table_id)) {
6740 return OFPERR_OFPTMFC_BAD_TABLE;
6741 }
6742
6743 /* Don't allow the eviction flags to be changed (except to the only fixed
6744 * value that OVS supports). OF1.4 says this is normal: "The
6745 * OFPTMPT_EVICTION property usually cannot be modified using a
6746 * OFP_TABLE_MOD request, because the eviction mechanism is switch
6747 * defined". */
6748 if (tm->eviction_flags != UINT32_MAX
6749 && tm->eviction_flags != OFPROTO_EVICTION_FLAGS) {
6750 return OFPERR_OFPTMFC_BAD_CONFIG;
6751 }
6752
6753 if (tm->table_id == OFPTT_ALL) {
6754 struct oftable *oftable;
6755 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6756 if (!(oftable->flags & (OFTABLE_HIDDEN | OFTABLE_READONLY))) {
6757 table_mod__(oftable, tm);
6758 }
6759 }
6760 } else {
6761 struct oftable *oftable = &ofproto->tables[tm->table_id];
6762 if (oftable->flags & OFTABLE_READONLY) {
6763 return OFPERR_OFPTMFC_EPERM;
6764 }
6765 table_mod__(oftable, tm);
6766 }
6767
6768 return 0;
6769 }
6770
6771 static enum ofperr
6772 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6773 {
6774 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6775 struct ofputil_table_mod tm;
6776 enum ofperr error;
6777
6778 error = reject_slave_controller(ofconn);
6779 if (error) {
6780 return error;
6781 }
6782
6783 error = ofputil_decode_table_mod(oh, &tm);
6784 if (error) {
6785 return error;
6786 }
6787
6788 return table_mod(ofproto, &tm);
6789 }
6790
6791 static enum ofperr
6792 ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6793 OVS_REQUIRES(ofproto_mutex)
6794 {
6795 switch (ofm->fm.command) {
6796 case OFPFC_ADD:
6797 return add_flow_start(ofproto, ofm);
6798 /* , &be->old_rules.stub[0],
6799 &be->new_rules.stub[0]); */
6800 case OFPFC_MODIFY:
6801 return modify_flows_start_loose(ofproto, ofm);
6802 case OFPFC_MODIFY_STRICT:
6803 return modify_flow_start_strict(ofproto, ofm);
6804 case OFPFC_DELETE:
6805 return delete_flows_start_loose(ofproto, ofm);
6806
6807 case OFPFC_DELETE_STRICT:
6808 return delete_flow_start_strict(ofproto, ofm);
6809 }
6810
6811 return OFPERR_OFPFMFC_BAD_COMMAND;
6812 }
6813
6814 static void
6815 ofproto_flow_mod_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6816 OVS_REQUIRES(ofproto_mutex)
6817 {
6818 switch (ofm->fm.command) {
6819 case OFPFC_ADD:
6820 add_flow_revert(ofproto, ofm);
6821 break;
6822
6823 case OFPFC_MODIFY:
6824 case OFPFC_MODIFY_STRICT:
6825 modify_flows_revert(ofproto, ofm);
6826 break;
6827
6828 case OFPFC_DELETE:
6829 case OFPFC_DELETE_STRICT:
6830 delete_flows_revert(ofproto, ofm);
6831 break;
6832
6833 default:
6834 break;
6835 }
6836 }
6837
6838 static void
6839 ofproto_flow_mod_finish(struct ofproto *ofproto,
6840 struct ofproto_flow_mod *ofm,
6841 const struct flow_mod_requester *req)
6842 OVS_REQUIRES(ofproto_mutex)
6843 {
6844 switch (ofm->fm.command) {
6845 case OFPFC_ADD:
6846 add_flow_finish(ofproto, ofm, req);
6847 break;
6848
6849 case OFPFC_MODIFY:
6850 case OFPFC_MODIFY_STRICT:
6851 modify_flows_finish(ofproto, ofm, req);
6852 break;
6853
6854 case OFPFC_DELETE:
6855 case OFPFC_DELETE_STRICT:
6856 delete_flows_finish(ofproto, ofm, req);
6857 break;
6858
6859 default:
6860 break;
6861 }
6862 }
6863
6864 /* Commit phases (all while locking ofproto_mutex):
6865 *
6866 * 1. Begin: Gather resources and make changes visible in the next version.
6867 * - Mark affected rules for removal in the next version.
6868 * - Create new replacement rules, make visible in the next
6869 * version.
6870 * - Do not send any events or notifications.
6871 *
6872 * 2. Revert: Fail if any errors are found. After this point no errors are
6873 * possible. No visible changes were made, so rollback is minimal (remove
6874 * added invisible rules, restore visibility of rules marked for removal).
6875 *
6876 * 3. Finish: Make the changes visible for lookups. Insert replacement rules to
6877 * the ofproto provider. Remove replaced and deleted rules from ofproto data
6878 * structures, and Schedule postponed removal of deleted rules from the
6879 * classifier. Send notifications, buffered packets, etc.
6880 */
6881 static enum ofperr
6882 do_bundle_commit(struct ofconn *ofconn, uint32_t id, uint16_t flags)
6883 {
6884 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6885 cls_version_t version = ofproto->tables_version + 1;
6886 struct ofp_bundle *bundle;
6887 struct ofp_bundle_entry *be;
6888 enum ofperr error;
6889
6890 bundle = ofconn_get_bundle(ofconn, id);
6891
6892 if (!bundle) {
6893 return OFPERR_OFPBFC_BAD_ID;
6894 }
6895 if (bundle->flags != flags) {
6896 error = OFPERR_OFPBFC_BAD_FLAGS;
6897 } else {
6898 bool prev_is_port_mod = false;
6899
6900 error = 0;
6901 ovs_mutex_lock(&ofproto_mutex);
6902
6903 /* 1. Begin. */
6904 LIST_FOR_EACH (be, node, &bundle->msg_list) {
6905 if (be->type == OFPTYPE_PORT_MOD) {
6906 /* Our port mods are not atomic. */
6907 if (flags & OFPBF_ATOMIC) {
6908 error = OFPERR_OFPBFC_MSG_FAILED;
6909 } else {
6910 prev_is_port_mod = true;
6911 error = port_mod_start(ofconn, &be->opm.pm, &be->opm.port);
6912 }
6913 } else if (be->type == OFPTYPE_FLOW_MOD) {
6914 /* Flow mods between port mods are applied as a single
6915 * version, but the versions are published only after
6916 * we know the commit is successful. */
6917 if (prev_is_port_mod) {
6918 ++version;
6919 }
6920 prev_is_port_mod = false;
6921 /* Store the version in which the changes should take
6922 * effect. */
6923 be->ofm.version = version;
6924 error = ofproto_flow_mod_start(ofproto, &be->ofm);
6925 } else {
6926 OVS_NOT_REACHED();
6927 }
6928 if (error) {
6929 break;
6930 }
6931 }
6932
6933 if (error) {
6934 /* Send error referring to the original message. */
6935 if (error) {
6936 ofconn_send_error(ofconn, be->ofp_msg, error);
6937 error = OFPERR_OFPBFC_MSG_FAILED;
6938 }
6939
6940 /* 2. Revert. Undo all the changes made above. */
6941 LIST_FOR_EACH_REVERSE_CONTINUE(be, node, &bundle->msg_list) {
6942 if (be->type == OFPTYPE_FLOW_MOD) {
6943 ofproto_flow_mod_revert(ofproto, &be->ofm);
6944 }
6945 /* Nothing needs to be reverted for a port mod. */
6946 }
6947 } else {
6948 /* 4. Finish. */
6949 LIST_FOR_EACH (be, node, &bundle->msg_list) {
6950 if (be->type == OFPTYPE_FLOW_MOD) {
6951 struct flow_mod_requester req = { ofconn, be->ofp_msg };
6952
6953 /* Bump the lookup version to the one of the current
6954 * message. This makes all the changes in the bundle at
6955 * this version visible to lookups at once. */
6956 if (ofproto->tables_version < be->ofm.version) {
6957 ofproto->tables_version = be->ofm.version;
6958 ofproto->ofproto_class->set_tables_version(
6959 ofproto, ofproto->tables_version);
6960 }
6961
6962 ofproto_flow_mod_finish(ofproto, &be->ofm, &req);
6963 } else if (be->type == OFPTYPE_PORT_MOD) {
6964 /* Perform the actual port mod. This is not atomic, i.e.,
6965 * the effects will be immediately seen by upcall
6966 * processing regardless of the lookup version. It should
6967 * be noted that port configuration changes can originate
6968 * also from OVSDB changes asynchronously to all upcall
6969 * processing. */
6970 port_mod_finish(ofconn, &be->opm.pm, be->opm.port);
6971 }
6972 }
6973 }
6974
6975 ofmonitor_flush(ofproto->connmgr);
6976 ovs_mutex_unlock(&ofproto_mutex);
6977
6978 run_rule_executes(ofproto);
6979 }
6980
6981 /* The bundle is discarded regardless the outcome. */
6982 ofp_bundle_remove__(ofconn, bundle, !error);
6983 return error;
6984 }
6985
6986 static enum ofperr
6987 handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh)
6988 {
6989 struct ofputil_bundle_ctrl_msg bctrl;
6990 struct ofputil_bundle_ctrl_msg reply;
6991 struct ofpbuf *buf;
6992 enum ofperr error;
6993
6994 error = reject_slave_controller(ofconn);
6995 if (error) {
6996 return error;
6997 }
6998
6999 error = ofputil_decode_bundle_ctrl(oh, &bctrl);
7000 if (error) {
7001 return error;
7002 }
7003 reply.flags = 0;
7004 reply.bundle_id = bctrl.bundle_id;
7005
7006 switch (bctrl.type) {
7007 case OFPBCT_OPEN_REQUEST:
7008 error = ofp_bundle_open(ofconn, bctrl.bundle_id, bctrl.flags);
7009 reply.type = OFPBCT_OPEN_REPLY;
7010 break;
7011 case OFPBCT_CLOSE_REQUEST:
7012 error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags);
7013 reply.type = OFPBCT_CLOSE_REPLY;
7014 break;
7015 case OFPBCT_COMMIT_REQUEST:
7016 error = do_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags);
7017 reply.type = OFPBCT_COMMIT_REPLY;
7018 break;
7019 case OFPBCT_DISCARD_REQUEST:
7020 error = ofp_bundle_discard(ofconn, bctrl.bundle_id);
7021 reply.type = OFPBCT_DISCARD_REPLY;
7022 break;
7023
7024 case OFPBCT_OPEN_REPLY:
7025 case OFPBCT_CLOSE_REPLY:
7026 case OFPBCT_COMMIT_REPLY:
7027 case OFPBCT_DISCARD_REPLY:
7028 return OFPERR_OFPBFC_BAD_TYPE;
7029 break;
7030 }
7031
7032 if (!error) {
7033 buf = ofputil_encode_bundle_ctrl_reply(oh, &reply);
7034 ofconn_send_reply(ofconn, buf);
7035 }
7036 return error;
7037 }
7038
7039 static enum ofperr
7040 handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh)
7041 {
7042 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
7043 enum ofperr error;
7044 struct ofputil_bundle_add_msg badd;
7045 struct ofp_bundle_entry *bmsg;
7046 enum ofptype type;
7047
7048 error = reject_slave_controller(ofconn);
7049 if (error) {
7050 return error;
7051 }
7052
7053 error = ofputil_decode_bundle_add(oh, &badd, &type);
7054 if (error) {
7055 return error;
7056 }
7057
7058 bmsg = ofp_bundle_entry_alloc(type, badd.msg);
7059
7060 if (type == OFPTYPE_PORT_MOD) {
7061 error = ofputil_decode_port_mod(badd.msg, &bmsg->opm.pm, false);
7062 } else if (type == OFPTYPE_FLOW_MOD) {
7063 struct ofpbuf ofpacts;
7064 uint64_t ofpacts_stub[1024 / 8];
7065
7066 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
7067 error = ofputil_decode_flow_mod(&bmsg->ofm.fm, badd.msg,
7068 ofconn_get_protocol(ofconn),
7069 &ofpacts,
7070 u16_to_ofp(ofproto->max_ports),
7071 ofproto->n_tables);
7072 /* Move actions to heap. */
7073 bmsg->ofm.fm.ofpacts = ofpbuf_steal_data(&ofpacts);
7074
7075 if (!error && bmsg->ofm.fm.ofpacts_len) {
7076 error = ofproto_check_ofpacts(ofproto, bmsg->ofm.fm.ofpacts,
7077 bmsg->ofm.fm.ofpacts_len);
7078 }
7079 } else {
7080 OVS_NOT_REACHED();
7081 }
7082
7083 if (!error) {
7084 error = ofp_bundle_add_message(ofconn, badd.bundle_id, badd.flags,
7085 bmsg);
7086 }
7087
7088 if (error) {
7089 ofp_bundle_entry_free(bmsg);
7090 }
7091
7092 return error;
7093 }
7094
7095 static enum ofperr
7096 handle_tlv_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
7097 {
7098 struct ofputil_tlv_table_mod ttm;
7099 enum ofperr error;
7100
7101 error = reject_slave_controller(ofconn);
7102 if (error) {
7103 return error;
7104 }
7105
7106 error = ofputil_decode_tlv_table_mod(oh, &ttm);
7107 if (error) {
7108 return error;
7109 }
7110
7111 error = tun_metadata_table_mod(&ttm);
7112
7113 ofputil_uninit_tlv_table(&ttm.mappings);
7114 return error;
7115 }
7116
7117 static enum ofperr
7118 handle_tlv_table_request(struct ofconn *ofconn, const struct ofp_header *oh)
7119 {
7120 struct ofputil_tlv_table_reply ttr;
7121 struct ofpbuf *b;
7122
7123 tun_metadata_table_request(&ttr);
7124 b = ofputil_encode_tlv_table_reply(oh, &ttr);
7125 ofputil_uninit_tlv_table(&ttr.mappings);
7126
7127 ofconn_send_reply(ofconn, b);
7128 return 0;
7129 }
7130
7131 static enum ofperr
7132 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
7133 OVS_EXCLUDED(ofproto_mutex)
7134 {
7135 const struct ofp_header *oh = msg->data;
7136 enum ofptype type;
7137 enum ofperr error;
7138
7139 error = ofptype_decode(&type, oh);
7140 if (error) {
7141 return error;
7142 }
7143 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
7144 && ofpmp_more(oh)) {
7145 /* We have no buffer implementation for multipart requests.
7146 * Report overflow for requests which consists of multiple
7147 * messages. */
7148 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
7149 }
7150
7151 switch (type) {
7152 /* OpenFlow requests. */
7153 case OFPTYPE_ECHO_REQUEST:
7154 return handle_echo_request(ofconn, oh);
7155
7156 case OFPTYPE_FEATURES_REQUEST:
7157 return handle_features_request(ofconn, oh);
7158
7159 case OFPTYPE_GET_CONFIG_REQUEST:
7160 return handle_get_config_request(ofconn, oh);
7161
7162 case OFPTYPE_SET_CONFIG:
7163 return handle_set_config(ofconn, oh);
7164
7165 case OFPTYPE_PACKET_OUT:
7166 return handle_packet_out(ofconn, oh);
7167
7168 case OFPTYPE_PORT_MOD:
7169 return handle_port_mod(ofconn, oh);
7170
7171 case OFPTYPE_FLOW_MOD:
7172 return handle_flow_mod(ofconn, oh);
7173
7174 case OFPTYPE_GROUP_MOD:
7175 return handle_group_mod(ofconn, oh);
7176
7177 case OFPTYPE_TABLE_MOD:
7178 return handle_table_mod(ofconn, oh);
7179
7180 case OFPTYPE_METER_MOD:
7181 return handle_meter_mod(ofconn, oh);
7182
7183 case OFPTYPE_BARRIER_REQUEST:
7184 return handle_barrier_request(ofconn, oh);
7185
7186 case OFPTYPE_ROLE_REQUEST:
7187 return handle_role_request(ofconn, oh);
7188
7189 /* OpenFlow replies. */
7190 case OFPTYPE_ECHO_REPLY:
7191 return 0;
7192
7193 /* Nicira extension requests. */
7194 case OFPTYPE_FLOW_MOD_TABLE_ID:
7195 return handle_nxt_flow_mod_table_id(ofconn, oh);
7196
7197 case OFPTYPE_SET_FLOW_FORMAT:
7198 return handle_nxt_set_flow_format(ofconn, oh);
7199
7200 case OFPTYPE_SET_PACKET_IN_FORMAT:
7201 return handle_nxt_set_packet_in_format(ofconn, oh);
7202
7203 case OFPTYPE_SET_CONTROLLER_ID:
7204 return handle_nxt_set_controller_id(ofconn, oh);
7205
7206 case OFPTYPE_FLOW_AGE:
7207 /* Nothing to do. */
7208 return 0;
7209
7210 case OFPTYPE_FLOW_MONITOR_CANCEL:
7211 return handle_flow_monitor_cancel(ofconn, oh);
7212
7213 case OFPTYPE_SET_ASYNC_CONFIG:
7214 return handle_nxt_set_async_config(ofconn, oh);
7215
7216 case OFPTYPE_GET_ASYNC_REQUEST:
7217 return handle_nxt_get_async_request(ofconn, oh);
7218
7219 /* Statistics requests. */
7220 case OFPTYPE_DESC_STATS_REQUEST:
7221 return handle_desc_stats_request(ofconn, oh);
7222
7223 case OFPTYPE_FLOW_STATS_REQUEST:
7224 return handle_flow_stats_request(ofconn, oh);
7225
7226 case OFPTYPE_AGGREGATE_STATS_REQUEST:
7227 return handle_aggregate_stats_request(ofconn, oh);
7228
7229 case OFPTYPE_TABLE_STATS_REQUEST:
7230 return handle_table_stats_request(ofconn, oh);
7231
7232 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
7233 return handle_table_features_request(ofconn, oh);
7234
7235 case OFPTYPE_TABLE_DESC_REQUEST:
7236 return handle_table_desc_request(ofconn, oh);
7237
7238 case OFPTYPE_PORT_STATS_REQUEST:
7239 return handle_port_stats_request(ofconn, oh);
7240
7241 case OFPTYPE_QUEUE_STATS_REQUEST:
7242 return handle_queue_stats_request(ofconn, oh);
7243
7244 case OFPTYPE_PORT_DESC_STATS_REQUEST:
7245 return handle_port_desc_stats_request(ofconn, oh);
7246
7247 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
7248 return handle_flow_monitor_request(ofconn, oh);
7249
7250 case OFPTYPE_METER_STATS_REQUEST:
7251 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
7252 return handle_meter_request(ofconn, oh, type);
7253
7254 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
7255 return handle_meter_features_request(ofconn, oh);
7256
7257 case OFPTYPE_GROUP_STATS_REQUEST:
7258 return handle_group_stats_request(ofconn, oh);
7259
7260 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
7261 return handle_group_desc_stats_request(ofconn, oh);
7262
7263 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
7264 return handle_group_features_stats_request(ofconn, oh);
7265
7266 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
7267 return handle_queue_get_config_request(ofconn, oh);
7268
7269 case OFPTYPE_BUNDLE_CONTROL:
7270 return handle_bundle_control(ofconn, oh);
7271
7272 case OFPTYPE_BUNDLE_ADD_MESSAGE:
7273 return handle_bundle_add(ofconn, oh);
7274
7275 case OFPTYPE_NXT_TLV_TABLE_MOD:
7276 return handle_tlv_table_mod(ofconn, oh);
7277
7278 case OFPTYPE_NXT_TLV_TABLE_REQUEST:
7279 return handle_tlv_table_request(ofconn, oh);
7280
7281 case OFPTYPE_HELLO:
7282 case OFPTYPE_ERROR:
7283 case OFPTYPE_FEATURES_REPLY:
7284 case OFPTYPE_GET_CONFIG_REPLY:
7285 case OFPTYPE_PACKET_IN:
7286 case OFPTYPE_FLOW_REMOVED:
7287 case OFPTYPE_PORT_STATUS:
7288 case OFPTYPE_BARRIER_REPLY:
7289 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
7290 case OFPTYPE_DESC_STATS_REPLY:
7291 case OFPTYPE_FLOW_STATS_REPLY:
7292 case OFPTYPE_QUEUE_STATS_REPLY:
7293 case OFPTYPE_PORT_STATS_REPLY:
7294 case OFPTYPE_TABLE_STATS_REPLY:
7295 case OFPTYPE_AGGREGATE_STATS_REPLY:
7296 case OFPTYPE_PORT_DESC_STATS_REPLY:
7297 case OFPTYPE_ROLE_REPLY:
7298 case OFPTYPE_FLOW_MONITOR_PAUSED:
7299 case OFPTYPE_FLOW_MONITOR_RESUMED:
7300 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
7301 case OFPTYPE_GET_ASYNC_REPLY:
7302 case OFPTYPE_GROUP_STATS_REPLY:
7303 case OFPTYPE_GROUP_DESC_STATS_REPLY:
7304 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
7305 case OFPTYPE_METER_STATS_REPLY:
7306 case OFPTYPE_METER_CONFIG_STATS_REPLY:
7307 case OFPTYPE_METER_FEATURES_STATS_REPLY:
7308 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
7309 case OFPTYPE_TABLE_DESC_REPLY:
7310 case OFPTYPE_ROLE_STATUS:
7311 case OFPTYPE_REQUESTFORWARD:
7312 case OFPTYPE_NXT_TLV_TABLE_REPLY:
7313 default:
7314 if (ofpmsg_is_stat_request(oh)) {
7315 return OFPERR_OFPBRC_BAD_STAT;
7316 } else {
7317 return OFPERR_OFPBRC_BAD_TYPE;
7318 }
7319 }
7320 }
7321
7322 static void
7323 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
7324 OVS_EXCLUDED(ofproto_mutex)
7325 {
7326 enum ofperr error = handle_openflow__(ofconn, ofp_msg);
7327
7328 if (error) {
7329 ofconn_send_error(ofconn, ofp_msg->data, error);
7330 }
7331 COVERAGE_INC(ofproto_recv_openflow);
7332 }
7333 \f
7334 /* Asynchronous operations. */
7335
7336 static void
7337 send_buffered_packet(const struct flow_mod_requester *req, uint32_t buffer_id,
7338 struct rule *rule)
7339 OVS_REQUIRES(ofproto_mutex)
7340 {
7341 if (req && req->ofconn && buffer_id != UINT32_MAX) {
7342 struct ofproto *ofproto = ofconn_get_ofproto(req->ofconn);
7343 struct dp_packet *packet;
7344 ofp_port_t in_port;
7345 enum ofperr error;
7346
7347 error = ofconn_pktbuf_retrieve(req->ofconn, buffer_id, &packet,
7348 &in_port);
7349 if (packet) {
7350 struct rule_execute *re;
7351
7352 ofproto_rule_ref(rule);
7353
7354 re = xmalloc(sizeof *re);
7355 re->rule = rule;
7356 re->in_port = in_port;
7357 re->packet = packet;
7358
7359 if (!guarded_list_push_back(&ofproto->rule_executes,
7360 &re->list_node, 1024)) {
7361 ofproto_rule_unref(rule);
7362 dp_packet_delete(re->packet);
7363 free(re);
7364 }
7365 } else {
7366 ofconn_send_error(req->ofconn, req->request, error);
7367 }
7368 }
7369 }
7370 \f
7371 static uint64_t
7372 pick_datapath_id(const struct ofproto *ofproto)
7373 {
7374 const struct ofport *port;
7375
7376 port = ofproto_get_port(ofproto, OFPP_LOCAL);
7377 if (port) {
7378 struct eth_addr ea;
7379 int error;
7380
7381 error = netdev_get_etheraddr(port->netdev, &ea);
7382 if (!error) {
7383 return eth_addr_to_uint64(ea);
7384 }
7385 VLOG_WARN("%s: could not get MAC address for %s (%s)",
7386 ofproto->name, netdev_get_name(port->netdev),
7387 ovs_strerror(error));
7388 }
7389 return ofproto->fallback_dpid;
7390 }
7391
7392 static uint64_t
7393 pick_fallback_dpid(void)
7394 {
7395 struct eth_addr ea;
7396 eth_addr_nicira_random(&ea);
7397 return eth_addr_to_uint64(ea);
7398 }
7399 \f
7400 /* Table overflow policy. */
7401
7402 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
7403 * to NULL if the table is not configured to evict rules or if the table
7404 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
7405 * or with no timeouts are not evictable.) */
7406 static bool
7407 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
7408 OVS_REQUIRES(ofproto_mutex)
7409 {
7410 struct eviction_group *evg;
7411
7412 *rulep = NULL;
7413 if (!table->eviction) {
7414 return false;
7415 }
7416
7417 /* In the common case, the outer and inner loops here will each be entered
7418 * exactly once:
7419 *
7420 * - The inner loop normally "return"s in its first iteration. If the
7421 * eviction group has any evictable rules, then it always returns in
7422 * some iteration.
7423 *
7424 * - The outer loop only iterates more than once if the largest eviction
7425 * group has no evictable rules.
7426 *
7427 * - The outer loop can exit only if table's 'max_flows' is all filled up
7428 * by unevictable rules. */
7429 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
7430 struct rule *rule;
7431
7432 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
7433 *rulep = rule;
7434 return true;
7435 }
7436 }
7437
7438 return false;
7439 }
7440 \f
7441 /* Eviction groups. */
7442
7443 /* Returns the priority to use for an eviction_group that contains 'n_rules'
7444 * rules. The priority contains low-order random bits to ensure that eviction
7445 * groups with the same number of rules are prioritized randomly. */
7446 static uint32_t
7447 eviction_group_priority(size_t n_rules)
7448 {
7449 uint16_t size = MIN(UINT16_MAX, n_rules);
7450 return (size << 16) | random_uint16();
7451 }
7452
7453 /* Updates 'evg', an eviction_group within 'table', following a change that
7454 * adds or removes rules in 'evg'. */
7455 static void
7456 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
7457 OVS_REQUIRES(ofproto_mutex)
7458 {
7459 heap_change(&table->eviction_groups_by_size, &evg->size_node,
7460 eviction_group_priority(heap_count(&evg->rules)));
7461 }
7462
7463 /* Destroys 'evg', an eviction_group within 'table':
7464 *
7465 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
7466 * rules themselves, just removes them from the eviction group.)
7467 *
7468 * - Removes 'evg' from 'table'.
7469 *
7470 * - Frees 'evg'. */
7471 static void
7472 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
7473 OVS_REQUIRES(ofproto_mutex)
7474 {
7475 while (!heap_is_empty(&evg->rules)) {
7476 struct rule *rule;
7477
7478 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
7479 rule->eviction_group = NULL;
7480 }
7481 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
7482 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
7483 heap_destroy(&evg->rules);
7484 free(evg);
7485 }
7486
7487 /* Removes 'rule' from its eviction group, if any. */
7488 static void
7489 eviction_group_remove_rule(struct rule *rule)
7490 OVS_REQUIRES(ofproto_mutex)
7491 {
7492 if (rule->eviction_group) {
7493 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7494 struct eviction_group *evg = rule->eviction_group;
7495
7496 rule->eviction_group = NULL;
7497 heap_remove(&evg->rules, &rule->evg_node);
7498 if (heap_is_empty(&evg->rules)) {
7499 eviction_group_destroy(table, evg);
7500 } else {
7501 eviction_group_resized(table, evg);
7502 }
7503 }
7504 }
7505
7506 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
7507 * returns the hash value. */
7508 static uint32_t
7509 eviction_group_hash_rule(struct rule *rule)
7510 OVS_REQUIRES(ofproto_mutex)
7511 {
7512 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7513 const struct mf_subfield *sf;
7514 struct flow flow;
7515 uint32_t hash;
7516
7517 hash = table->eviction_group_id_basis;
7518 miniflow_expand(rule->cr.match.flow, &flow);
7519 for (sf = table->eviction_fields;
7520 sf < &table->eviction_fields[table->n_eviction_fields];
7521 sf++)
7522 {
7523 if (mf_are_prereqs_ok(sf->field, &flow)) {
7524 union mf_value value;
7525
7526 mf_get_value(sf->field, &flow, &value);
7527 if (sf->ofs) {
7528 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
7529 }
7530 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
7531 unsigned int start = sf->ofs + sf->n_bits;
7532 bitwise_zero(&value, sf->field->n_bytes, start,
7533 sf->field->n_bytes * 8 - start);
7534 }
7535 hash = hash_bytes(&value, sf->field->n_bytes, hash);
7536 } else {
7537 hash = hash_int(hash, 0);
7538 }
7539 }
7540
7541 return hash;
7542 }
7543
7544 /* Returns an eviction group within 'table' with the given 'id', creating one
7545 * if necessary. */
7546 static struct eviction_group *
7547 eviction_group_find(struct oftable *table, uint32_t id)
7548 OVS_REQUIRES(ofproto_mutex)
7549 {
7550 struct eviction_group *evg;
7551
7552 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
7553 return evg;
7554 }
7555
7556 evg = xmalloc(sizeof *evg);
7557 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
7558 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
7559 eviction_group_priority(0));
7560 heap_init(&evg->rules);
7561
7562 return evg;
7563 }
7564
7565 /* Returns an eviction priority for 'rule'. The return value should be
7566 * interpreted so that higher priorities make a rule a more attractive
7567 * candidate for eviction. */
7568 static uint64_t
7569 rule_eviction_priority(struct ofproto *ofproto, struct rule *rule)
7570 OVS_REQUIRES(ofproto_mutex)
7571 {
7572 /* Calculate absolute time when this flow will expire. If it will never
7573 * expire, then return 0 to make it unevictable. */
7574 long long int expiration = LLONG_MAX;
7575 if (rule->hard_timeout) {
7576 /* 'modified' needs protection even when we hold 'ofproto_mutex'. */
7577 ovs_mutex_lock(&rule->mutex);
7578 long long int modified = rule->modified;
7579 ovs_mutex_unlock(&rule->mutex);
7580
7581 expiration = modified + rule->hard_timeout * 1000;
7582 }
7583 if (rule->idle_timeout) {
7584 uint64_t packets, bytes;
7585 long long int used;
7586 long long int idle_expiration;
7587
7588 ofproto->ofproto_class->rule_get_stats(rule, &packets, &bytes, &used);
7589 idle_expiration = used + rule->idle_timeout * 1000;
7590 expiration = MIN(expiration, idle_expiration);
7591 }
7592 if (expiration == LLONG_MAX) {
7593 return 0;
7594 }
7595
7596 /* Calculate the time of expiration as a number of (approximate) seconds
7597 * after program startup.
7598 *
7599 * This should work OK for program runs that last UINT32_MAX seconds or
7600 * less. Therefore, please restart OVS at least once every 136 years. */
7601 uint32_t expiration_ofs = (expiration >> 10) - (time_boot_msec() >> 10);
7602
7603 /* Combine expiration time with OpenFlow "importance" to form a single
7604 * priority value. We want flows with relatively low "importance" to be
7605 * evicted before even considering expiration time, so put "importance" in
7606 * the most significant bits and expiration time in the least significant
7607 * bits.
7608 *
7609 * Small 'priority' should be evicted before those with large 'priority'.
7610 * The caller expects the opposite convention (a large return value being
7611 * more attractive for eviction) so we invert it before returning. */
7612 uint64_t priority = ((uint64_t) rule->importance << 32) + expiration_ofs;
7613 return UINT64_MAX - priority;
7614 }
7615
7616 /* Adds 'rule' to an appropriate eviction group for its oftable's
7617 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
7618 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
7619 * own).
7620 *
7621 * The caller must ensure that 'rule' is not already in an eviction group. */
7622 static void
7623 eviction_group_add_rule(struct rule *rule)
7624 OVS_REQUIRES(ofproto_mutex)
7625 {
7626 struct ofproto *ofproto = rule->ofproto;
7627 struct oftable *table = &ofproto->tables[rule->table_id];
7628 bool has_timeout;
7629
7630 /* Timeouts may be modified only when holding 'ofproto_mutex'. We have it
7631 * so no additional protection is needed. */
7632 has_timeout = rule->hard_timeout || rule->idle_timeout;
7633
7634 if (table->eviction && has_timeout) {
7635 struct eviction_group *evg;
7636
7637 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
7638
7639 rule->eviction_group = evg;
7640 heap_insert(&evg->rules, &rule->evg_node,
7641 rule_eviction_priority(ofproto, rule));
7642 eviction_group_resized(table, evg);
7643 }
7644 }
7645 \f
7646 /* oftables. */
7647
7648 /* Initializes 'table'. */
7649 static void
7650 oftable_init(struct oftable *table)
7651 {
7652 memset(table, 0, sizeof *table);
7653 classifier_init(&table->cls, flow_segment_u64s);
7654 table->max_flows = UINT_MAX;
7655 table->n_flows = 0;
7656 hmap_init(&table->eviction_groups_by_id);
7657 heap_init(&table->eviction_groups_by_size);
7658 atomic_init(&table->miss_config, OFPUTIL_TABLE_MISS_DEFAULT);
7659
7660 classifier_set_prefix_fields(&table->cls, default_prefix_fields,
7661 ARRAY_SIZE(default_prefix_fields));
7662
7663 atomic_init(&table->n_matched, 0);
7664 atomic_init(&table->n_missed, 0);
7665 }
7666
7667 /* Destroys 'table', including its classifier and eviction groups.
7668 *
7669 * The caller is responsible for freeing 'table' itself. */
7670 static void
7671 oftable_destroy(struct oftable *table)
7672 {
7673 ovs_assert(classifier_is_empty(&table->cls));
7674
7675 ovs_mutex_lock(&ofproto_mutex);
7676 oftable_configure_eviction(table, 0, NULL, 0);
7677 ovs_mutex_unlock(&ofproto_mutex);
7678
7679 hmap_destroy(&table->eviction_groups_by_id);
7680 heap_destroy(&table->eviction_groups_by_size);
7681 classifier_destroy(&table->cls);
7682 free(table->name);
7683 }
7684
7685 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
7686 * string, then 'table' will use its default name.
7687 *
7688 * This only affects the name exposed for a table exposed through the OpenFlow
7689 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
7690 static void
7691 oftable_set_name(struct oftable *table, const char *name)
7692 {
7693 if (name && name[0]) {
7694 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
7695 if (!table->name || strncmp(name, table->name, len)) {
7696 free(table->name);
7697 table->name = xmemdup0(name, len);
7698 }
7699 } else {
7700 free(table->name);
7701 table->name = NULL;
7702 }
7703 }
7704
7705 /* oftables support a choice of two policies when adding a rule would cause the
7706 * number of flows in the table to exceed the configured maximum number: either
7707 * they can refuse to add the new flow or they can evict some existing flow.
7708 * This function configures the latter policy on 'table', with fairness based
7709 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
7710 * 'n_fields' as 0 disables fairness.) */
7711 static void
7712 oftable_configure_eviction(struct oftable *table, unsigned int eviction,
7713 const struct mf_subfield *fields, size_t n_fields)
7714 OVS_REQUIRES(ofproto_mutex)
7715 {
7716 struct rule *rule;
7717
7718 if ((table->eviction != 0) == (eviction != 0)
7719 && n_fields == table->n_eviction_fields
7720 && (!n_fields
7721 || !memcmp(fields, table->eviction_fields,
7722 n_fields * sizeof *fields))) {
7723 /* The set of eviction fields did not change. If 'eviction' changed,
7724 * it remains nonzero, so that we can just update table->eviction
7725 * without fussing with the eviction groups. */
7726 table->eviction = eviction;
7727 return;
7728 }
7729
7730 /* Destroy existing eviction groups, then destroy and recreate data
7731 * structures to recover memory. */
7732 struct eviction_group *evg, *next;
7733 HMAP_FOR_EACH_SAFE (evg, next, id_node, &table->eviction_groups_by_id) {
7734 eviction_group_destroy(table, evg);
7735 }
7736 hmap_destroy(&table->eviction_groups_by_id);
7737 hmap_init(&table->eviction_groups_by_id);
7738 heap_destroy(&table->eviction_groups_by_size);
7739 heap_init(&table->eviction_groups_by_size);
7740
7741 /* Replace eviction groups by the new ones, if there is a change. Free the
7742 * old fields only after allocating the new ones, because 'fields ==
7743 * table->eviction_fields' is possible. */
7744 struct mf_subfield *old_fields = table->eviction_fields;
7745 table->n_eviction_fields = n_fields;
7746 table->eviction_fields = (fields
7747 ? xmemdup(fields, n_fields * sizeof *fields)
7748 : NULL);
7749 free(old_fields);
7750
7751 /* Add the new eviction groups, if enabled. */
7752 table->eviction = eviction;
7753 if (table->eviction) {
7754 table->eviction_group_id_basis = random_uint32();
7755 CLS_FOR_EACH (rule, cr, &table->cls) {
7756 eviction_group_add_rule(rule);
7757 }
7758 }
7759 }
7760
7761 /* Inserts 'rule' from the ofproto data structures BEFORE caller has inserted
7762 * it to the classifier. */
7763 static void
7764 ofproto_rule_insert__(struct ofproto *ofproto, struct rule *rule)
7765 OVS_REQUIRES(ofproto_mutex)
7766 {
7767 const struct rule_actions *actions = rule_get_actions(rule);
7768
7769 ovs_assert(rule->removed);
7770
7771 if (rule->hard_timeout || rule->idle_timeout) {
7772 list_insert(&ofproto->expirable, &rule->expirable);
7773 }
7774 cookies_insert(ofproto, rule);
7775 eviction_group_add_rule(rule);
7776 if (actions->has_meter) {
7777 meter_insert_rule(rule);
7778 }
7779 rule->removed = false;
7780 }
7781
7782 /* Removes 'rule' from the ofproto data structures. Caller may have deferred
7783 * the removal from the classifier. */
7784 static void
7785 ofproto_rule_remove__(struct ofproto *ofproto, struct rule *rule)
7786 OVS_REQUIRES(ofproto_mutex)
7787 {
7788 ovs_assert(!rule->removed);
7789
7790 cookies_remove(ofproto, rule);
7791
7792 eviction_group_remove_rule(rule);
7793 if (!list_is_empty(&rule->expirable)) {
7794 list_remove(&rule->expirable);
7795 }
7796 if (!list_is_empty(&rule->meter_list_node)) {
7797 list_remove(&rule->meter_list_node);
7798 list_init(&rule->meter_list_node);
7799 }
7800
7801 rule->removed = true;
7802 }
7803 \f
7804 /* unixctl commands. */
7805
7806 struct ofproto *
7807 ofproto_lookup(const char *name)
7808 {
7809 struct ofproto *ofproto;
7810
7811 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
7812 &all_ofprotos) {
7813 if (!strcmp(ofproto->name, name)) {
7814 return ofproto;
7815 }
7816 }
7817 return NULL;
7818 }
7819
7820 static void
7821 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
7822 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7823 {
7824 struct ofproto *ofproto;
7825 struct ds results;
7826
7827 ds_init(&results);
7828 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
7829 ds_put_format(&results, "%s\n", ofproto->name);
7830 }
7831 unixctl_command_reply(conn, ds_cstr(&results));
7832 ds_destroy(&results);
7833 }
7834
7835 static void
7836 ofproto_unixctl_init(void)
7837 {
7838 static bool registered;
7839 if (registered) {
7840 return;
7841 }
7842 registered = true;
7843
7844 unixctl_command_register("ofproto/list", "", 0, 0,
7845 ofproto_unixctl_list, NULL);
7846 }
7847 \f
7848 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7849 *
7850 * This is deprecated. It is only for compatibility with broken device drivers
7851 * in old versions of Linux that do not properly support VLANs when VLAN
7852 * devices are not used. When broken device drivers are no longer in
7853 * widespread use, we will delete these interfaces. */
7854
7855 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
7856 * (exactly) by an OpenFlow rule in 'ofproto'. */
7857 void
7858 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
7859 {
7860 struct match match;
7861 struct cls_rule target;
7862 const struct oftable *oftable;
7863
7864 match_init_catchall(&match);
7865 match_set_vlan_vid_masked(&match, htons(VLAN_CFI), htons(VLAN_CFI));
7866 cls_rule_init(&target, &match, 0);
7867
7868 free(ofproto->vlan_bitmap);
7869 ofproto->vlan_bitmap = bitmap_allocate(4096);
7870 ofproto->vlans_changed = false;
7871
7872 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
7873 struct rule *rule;
7874
7875 CLS_FOR_EACH_TARGET (rule, cr, &oftable->cls, &target,
7876 CLS_MAX_VERSION) {
7877 if (minimask_get_vid_mask(rule->cr.match.mask) == VLAN_VID_MASK) {
7878 uint16_t vid = miniflow_get_vid(rule->cr.match.flow);
7879
7880 bitmap_set1(vlan_bitmap, vid);
7881 bitmap_set1(ofproto->vlan_bitmap, vid);
7882 }
7883 }
7884 }
7885
7886 cls_rule_destroy(&target);
7887 }
7888
7889 /* Returns true if new VLANs have come into use by the flow table since the
7890 * last call to ofproto_get_vlan_usage().
7891 *
7892 * We don't track when old VLANs stop being used. */
7893 bool
7894 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
7895 {
7896 return ofproto->vlans_changed;
7897 }
7898
7899 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
7900 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
7901 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
7902 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
7903 * then the VLAN device is un-enslaved. */
7904 int
7905 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
7906 ofp_port_t realdev_ofp_port, int vid)
7907 {
7908 struct ofport *ofport;
7909 int error;
7910
7911 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
7912
7913 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
7914 if (!ofport) {
7915 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
7916 ofproto->name, vlandev_ofp_port);
7917 return EINVAL;
7918 }
7919
7920 if (!ofproto->ofproto_class->set_realdev) {
7921 if (!vlandev_ofp_port) {
7922 return 0;
7923 }
7924 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
7925 return EOPNOTSUPP;
7926 }
7927
7928 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
7929 if (error) {
7930 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
7931 ofproto->name, vlandev_ofp_port,
7932 netdev_get_name(ofport->netdev), ovs_strerror(error));
7933 }
7934 return error;
7935 }