]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/ofproto.c
ofproto: Make flow handling more symmetric.
[mirror_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 <errno.h>
20 #include <inttypes.h>
21 #include <stdbool.h>
22 #include <stdlib.h>
23 #include <unistd.h>
24
25 #include "bitmap.h"
26 #include "bundles.h"
27 #include "byte-order.h"
28 #include "classifier.h"
29 #include "connectivity.h"
30 #include "connmgr.h"
31 #include "coverage.h"
32 #include "dp-packet.h"
33 #include "hash.h"
34 #include "openvswitch/hmap.h"
35 #include "netdev.h"
36 #include "nx-match.h"
37 #include "ofproto.h"
38 #include "ofproto-provider.h"
39 #include "openflow/nicira-ext.h"
40 #include "openflow/openflow.h"
41 #include "openvswitch/dynamic-string.h"
42 #include "openvswitch/meta-flow.h"
43 #include "openvswitch/ofp-actions.h"
44 #include "openvswitch/ofp-errors.h"
45 #include "openvswitch/ofp-msgs.h"
46 #include "openvswitch/ofp-print.h"
47 #include "openvswitch/ofp-util.h"
48 #include "openvswitch/ofpbuf.h"
49 #include "openvswitch/vlog.h"
50 #include "ovs-rcu.h"
51 #include "packets.h"
52 #include "pinsched.h"
53 #include "pktbuf.h"
54 #include "poll-loop.h"
55 #include "random.h"
56 #include "seq.h"
57 #include "openvswitch/shash.h"
58 #include "simap.h"
59 #include "smap.h"
60 #include "sset.h"
61 #include "timeval.h"
62 #include "tun-metadata.h"
63 #include "unaligned.h"
64 #include "unixctl.h"
65 #include "util.h"
66
67 VLOG_DEFINE_THIS_MODULE(ofproto);
68
69 COVERAGE_DEFINE(ofproto_flush);
70 COVERAGE_DEFINE(ofproto_packet_out);
71 COVERAGE_DEFINE(ofproto_queue_req);
72 COVERAGE_DEFINE(ofproto_recv_openflow);
73 COVERAGE_DEFINE(ofproto_reinit_ports);
74 COVERAGE_DEFINE(ofproto_update_port);
75
76 /* Default fields to use for prefix tries in each flow table, unless something
77 * else is configured. */
78 const enum mf_field_id default_prefix_fields[2] =
79 { MFF_IPV4_DST, MFF_IPV4_SRC };
80
81 /* oftable. */
82 static void oftable_init(struct oftable *);
83 static void oftable_destroy(struct oftable *);
84
85 static void oftable_set_name(struct oftable *, const char *name);
86
87 static enum ofperr evict_rules_from_table(struct oftable *)
88 OVS_REQUIRES(ofproto_mutex);
89 static void oftable_configure_eviction(struct oftable *,
90 unsigned int eviction,
91 const struct mf_subfield *fields,
92 size_t n_fields)
93 OVS_REQUIRES(ofproto_mutex);
94
95 /* This is the only combination of OpenFlow eviction flags that OVS supports: a
96 * combination of OF1.4+ importance, the remaining lifetime of the flow, and
97 * fairness based on user-specified fields. */
98 #define OFPROTO_EVICTION_FLAGS \
99 (OFPTMPEF14_OTHER | OFPTMPEF14_IMPORTANCE | OFPTMPEF14_LIFETIME)
100
101 /* A set of rules within a single OpenFlow table (oftable) that have the same
102 * values for the oftable's eviction_fields. A rule to be evicted, when one is
103 * needed, is taken from the eviction group that contains the greatest number
104 * of rules.
105 *
106 * An oftable owns any number of eviction groups, each of which contains any
107 * number of rules.
108 *
109 * Membership in an eviction group is imprecise, based on the hash of the
110 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
111 * That is, if two rules have different eviction_fields, but those
112 * eviction_fields hash to the same value, then they will belong to the same
113 * eviction_group anyway.
114 *
115 * (When eviction is not enabled on an oftable, we don't track any eviction
116 * groups, to save time and space.) */
117 struct eviction_group {
118 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
119 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
120 struct heap rules; /* Contains "struct rule"s. */
121 };
122
123 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
124 OVS_REQUIRES(ofproto_mutex);
125 static uint64_t rule_eviction_priority(struct ofproto *ofproto, struct rule *)
126 OVS_REQUIRES(ofproto_mutex);
127 static void eviction_group_add_rule(struct rule *)
128 OVS_REQUIRES(ofproto_mutex);
129 static void eviction_group_remove_rule(struct rule *)
130 OVS_REQUIRES(ofproto_mutex);
131
132 /* Criteria that flow_mod and other operations use for selecting rules on
133 * which to operate. */
134 struct rule_criteria {
135 /* An OpenFlow table or 255 for all tables. */
136 uint8_t table_id;
137
138 /* OpenFlow matching criteria. Interpreted different in "loose" way by
139 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
140 * defined in the OpenFlow spec. */
141 struct cls_rule cr;
142 ovs_version_t version;
143
144 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
145 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
146 * The rule will not be selected if M is 1 and B != C. */
147 ovs_be64 cookie;
148 ovs_be64 cookie_mask;
149
150 /* Selection based on actions within a rule:
151 *
152 * If out_port != OFPP_ANY, selects only rules that output to out_port.
153 * If out_group != OFPG_ALL, select only rules that output to out_group. */
154 ofp_port_t out_port;
155 uint32_t out_group;
156
157 /* If true, collects only rules that are modifiable. */
158 bool include_hidden;
159 bool include_readonly;
160 };
161
162 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
163 const struct match *match, int priority,
164 ovs_version_t version,
165 ovs_be64 cookie, ovs_be64 cookie_mask,
166 ofp_port_t out_port, uint32_t out_group);
167 static void rule_criteria_require_rw(struct rule_criteria *,
168 bool can_write_readonly);
169 static void rule_criteria_destroy(struct rule_criteria *);
170
171 static enum ofperr collect_rules_loose(struct ofproto *,
172 const struct rule_criteria *,
173 struct rule_collection *);
174
175 /* A packet that needs to be passed to rule_execute().
176 *
177 * (We can't do this immediately from ofopgroup_complete() because that holds
178 * ofproto_mutex, which rule_execute() needs released.) */
179 struct rule_execute {
180 struct ovs_list list_node; /* In struct ofproto's "rule_executes" list. */
181 struct rule *rule; /* Owns a reference to the rule. */
182 ofp_port_t in_port;
183 struct dp_packet *packet; /* Owns the packet. */
184 };
185
186 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
187 static void destroy_rule_executes(struct ofproto *);
188
189 struct learned_cookie {
190 union {
191 /* In struct ofproto's 'learned_cookies' hmap. */
192 struct hmap_node hmap_node OVS_GUARDED_BY(ofproto_mutex);
193
194 /* In 'dead_cookies' list when removed from hmap. */
195 struct ovs_list list_node;
196 } u;
197
198 /* Key. */
199 ovs_be64 cookie OVS_GUARDED_BY(ofproto_mutex);
200 uint8_t table_id OVS_GUARDED_BY(ofproto_mutex);
201
202 /* Number of references from "learn" actions.
203 *
204 * When this drops to 0, all of the flows in 'table_id' with the specified
205 * 'cookie' are deleted. */
206 int n OVS_GUARDED_BY(ofproto_mutex);
207 };
208
209 static const struct ofpact_learn *next_learn_with_delete(
210 const struct rule_actions *, const struct ofpact_learn *start);
211
212 static void learned_cookies_inc(struct ofproto *, const struct rule_actions *)
213 OVS_REQUIRES(ofproto_mutex);
214 static void learned_cookies_dec(struct ofproto *, const struct rule_actions *,
215 struct ovs_list *dead_cookies)
216 OVS_REQUIRES(ofproto_mutex);
217 static void learned_cookies_flush(struct ofproto *, struct ovs_list *dead_cookies)
218 OVS_REQUIRES(ofproto_mutex);
219
220 /* ofport. */
221 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
222 static void ofport_destroy(struct ofport *, bool del);
223 static inline bool ofport_is_internal(const struct ofport *);
224
225 static int update_port(struct ofproto *, const char *devname);
226 static int init_ports(struct ofproto *);
227 static void reinit_ports(struct ofproto *);
228
229 static long long int ofport_get_usage(const struct ofproto *,
230 ofp_port_t ofp_port);
231 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
232 long long int last_used);
233 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
234
235 /* Ofport usage.
236 *
237 * Keeps track of the currently used and recently used ofport values and is
238 * used to prevent immediate recycling of ofport values. */
239 struct ofport_usage {
240 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
241 ofp_port_t ofp_port; /* OpenFlow port number. */
242 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
243 represents in-use ofports. */
244 };
245
246 /* rule. */
247 static void ofproto_rule_send_removed(struct rule *)
248 OVS_EXCLUDED(ofproto_mutex);
249 static bool rule_is_readonly(const struct rule *);
250 static void ofproto_rule_insert__(struct ofproto *, struct rule *)
251 OVS_REQUIRES(ofproto_mutex);
252 static void ofproto_rule_remove__(struct ofproto *, struct rule *)
253 OVS_REQUIRES(ofproto_mutex);
254
255 /* The source of a flow_mod request, in the code that processes flow_mods.
256 *
257 * A flow table modification request can be generated externally, via OpenFlow,
258 * or internally through a function call. This structure indicates the source
259 * of an OpenFlow-generated flow_mod. For an internal flow_mod, it isn't
260 * meaningful and thus supplied as NULL. */
261 struct flow_mod_requester {
262 struct ofconn *ofconn; /* Connection on which flow_mod arrived. */
263 const struct ofp_header *request;
264 };
265
266 /* OpenFlow. */
267 static enum ofperr replace_rule_create(struct ofproto *,
268 struct ofputil_flow_mod *,
269 struct cls_rule *cr, uint8_t table_id,
270 struct rule *old_rule,
271 struct rule **new_rule)
272 OVS_REQUIRES(ofproto_mutex);
273
274 static void replace_rule_start(struct ofproto *, ovs_version_t version,
275 struct rule *old_rule, struct rule *new_rule,
276 struct cls_conjunction *, size_t n_conjs)
277 OVS_REQUIRES(ofproto_mutex);
278
279 static void replace_rule_revert(struct ofproto *, struct rule *old_rule,
280 struct rule *new_rule)
281 OVS_REQUIRES(ofproto_mutex);
282
283 static void replace_rule_finish(struct ofproto *, struct ofputil_flow_mod *,
284 const struct flow_mod_requester *,
285 struct rule *old_rule, struct rule *new_rule,
286 struct ovs_list *dead_cookies)
287 OVS_REQUIRES(ofproto_mutex);
288 static void delete_flows__(struct rule_collection *,
289 enum ofp_flow_removed_reason,
290 const struct flow_mod_requester *)
291 OVS_REQUIRES(ofproto_mutex);
292
293 static void send_buffered_packet(const struct flow_mod_requester *,
294 uint32_t buffer_id, struct rule *)
295 OVS_REQUIRES(ofproto_mutex);
296
297 static bool ofproto_group_exists(const struct ofproto *, uint32_t group_id);
298 static enum ofperr add_group(struct ofproto *,
299 const struct ofputil_group_mod *);
300 static void handle_openflow(struct ofconn *, const struct ofpbuf *);
301 static enum ofperr ofproto_flow_mod_start(struct ofproto *,
302 struct ofproto_flow_mod *)
303 OVS_REQUIRES(ofproto_mutex);
304 static void ofproto_flow_mod_finish(struct ofproto *,
305 struct ofproto_flow_mod *,
306 const struct flow_mod_requester *)
307 OVS_REQUIRES(ofproto_mutex);
308 static enum ofperr handle_flow_mod__(struct ofproto *,
309 struct ofproto_flow_mod *,
310 const struct flow_mod_requester *)
311 OVS_EXCLUDED(ofproto_mutex);
312 static void calc_duration(long long int start, long long int now,
313 uint32_t *sec, uint32_t *nsec);
314
315 /* ofproto. */
316 static uint64_t pick_datapath_id(const struct ofproto *);
317 static uint64_t pick_fallback_dpid(void);
318 static void ofproto_destroy__(struct ofproto *);
319 static void update_mtu(struct ofproto *, struct ofport *);
320 static void update_mtu_ofproto(struct ofproto *);
321 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
322 static void meter_insert_rule(struct rule *);
323
324 /* unixctl. */
325 static void ofproto_unixctl_init(void);
326
327 /* All registered ofproto classes, in probe order. */
328 static const struct ofproto_class **ofproto_classes;
329 static size_t n_ofproto_classes;
330 static size_t allocated_ofproto_classes;
331
332 /* Global lock that protects all flow table operations. */
333 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
334
335 unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
336 unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT;
337
338 size_t n_handlers, n_revalidators;
339 char *pmd_cpu_mask;
340
341 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
342 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
343
344 /* Initial mappings of port to OpenFlow number mappings. */
345 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
346
347 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
348
349 /* The default value of true waits for flow restore. */
350 static bool flow_restore_wait = true;
351
352 /* Must be called to initialize the ofproto library.
353 *
354 * The caller may pass in 'iface_hints', which contains an shash of
355 * "iface_hint" elements indexed by the interface's name. The provider
356 * may use these hints to describe the startup configuration in order to
357 * reinitialize its state. The caller owns the provided data, so a
358 * provider will make copies of anything required. An ofproto provider
359 * will remove any existing state that is not described by the hint, and
360 * may choose to remove it all. */
361 void
362 ofproto_init(const struct shash *iface_hints)
363 {
364 struct shash_node *node;
365 size_t i;
366
367 ofproto_class_register(&ofproto_dpif_class);
368
369 /* Make a local copy, since we don't own 'iface_hints' elements. */
370 SHASH_FOR_EACH(node, iface_hints) {
371 const struct iface_hint *orig_hint = node->data;
372 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
373 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
374
375 new_hint->br_name = xstrdup(orig_hint->br_name);
376 new_hint->br_type = xstrdup(br_type);
377 new_hint->ofp_port = orig_hint->ofp_port;
378
379 shash_add(&init_ofp_ports, node->name, new_hint);
380 }
381
382 for (i = 0; i < n_ofproto_classes; i++) {
383 ofproto_classes[i]->init(&init_ofp_ports);
384 }
385
386 ofproto_unixctl_init();
387 }
388
389 /* 'type' should be a normalized datapath type, as returned by
390 * ofproto_normalize_type(). Returns the corresponding ofproto_class
391 * structure, or a null pointer if there is none registered for 'type'. */
392 static const struct ofproto_class *
393 ofproto_class_find__(const char *type)
394 {
395 size_t i;
396
397 for (i = 0; i < n_ofproto_classes; i++) {
398 const struct ofproto_class *class = ofproto_classes[i];
399 struct sset types;
400 bool found;
401
402 sset_init(&types);
403 class->enumerate_types(&types);
404 found = sset_contains(&types, type);
405 sset_destroy(&types);
406
407 if (found) {
408 return class;
409 }
410 }
411 VLOG_WARN("unknown datapath type %s", type);
412 return NULL;
413 }
414
415 /* Registers a new ofproto class. After successful registration, new ofprotos
416 * of that type can be created using ofproto_create(). */
417 int
418 ofproto_class_register(const struct ofproto_class *new_class)
419 {
420 size_t i;
421
422 for (i = 0; i < n_ofproto_classes; i++) {
423 if (ofproto_classes[i] == new_class) {
424 return EEXIST;
425 }
426 }
427
428 if (n_ofproto_classes >= allocated_ofproto_classes) {
429 ofproto_classes = x2nrealloc(ofproto_classes,
430 &allocated_ofproto_classes,
431 sizeof *ofproto_classes);
432 }
433 ofproto_classes[n_ofproto_classes++] = new_class;
434 return 0;
435 }
436
437 /* Unregisters a datapath provider. 'type' must have been previously
438 * registered and not currently be in use by any ofprotos. After
439 * unregistration new datapaths of that type cannot be opened using
440 * ofproto_create(). */
441 int
442 ofproto_class_unregister(const struct ofproto_class *class)
443 {
444 size_t i;
445
446 for (i = 0; i < n_ofproto_classes; i++) {
447 if (ofproto_classes[i] == class) {
448 for (i++; i < n_ofproto_classes; i++) {
449 ofproto_classes[i - 1] = ofproto_classes[i];
450 }
451 n_ofproto_classes--;
452 return 0;
453 }
454 }
455 VLOG_WARN("attempted to unregister an ofproto class that is not "
456 "registered");
457 return EAFNOSUPPORT;
458 }
459
460 /* Clears 'types' and enumerates all registered ofproto types into it. The
461 * caller must first initialize the sset. */
462 void
463 ofproto_enumerate_types(struct sset *types)
464 {
465 size_t i;
466
467 sset_clear(types);
468 for (i = 0; i < n_ofproto_classes; i++) {
469 ofproto_classes[i]->enumerate_types(types);
470 }
471 }
472
473 /* Returns the fully spelled out name for the given ofproto 'type'.
474 *
475 * Normalized type string can be compared with strcmp(). Unnormalized type
476 * string might be the same even if they have different spellings. */
477 const char *
478 ofproto_normalize_type(const char *type)
479 {
480 return type && type[0] ? type : "system";
481 }
482
483 /* Clears 'names' and enumerates the names of all known created ofprotos with
484 * the given 'type'. The caller must first initialize the sset. Returns 0 if
485 * successful, otherwise a positive errno value.
486 *
487 * Some kinds of datapaths might not be practically enumerable. This is not
488 * considered an error. */
489 int
490 ofproto_enumerate_names(const char *type, struct sset *names)
491 {
492 const struct ofproto_class *class = ofproto_class_find__(type);
493 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
494 }
495
496 static void
497 ofproto_bump_tables_version(struct ofproto *ofproto)
498 {
499 ++ofproto->tables_version;
500 ofproto->ofproto_class->set_tables_version(ofproto,
501 ofproto->tables_version);
502 }
503
504 int
505 ofproto_create(const char *datapath_name, const char *datapath_type,
506 struct ofproto **ofprotop)
507 {
508 const struct ofproto_class *class;
509 struct ofproto *ofproto;
510 int error;
511 int i;
512
513 *ofprotop = NULL;
514
515 datapath_type = ofproto_normalize_type(datapath_type);
516 class = ofproto_class_find__(datapath_type);
517 if (!class) {
518 VLOG_WARN("could not create datapath %s of unknown type %s",
519 datapath_name, datapath_type);
520 return EAFNOSUPPORT;
521 }
522
523 ofproto = class->alloc();
524 if (!ofproto) {
525 VLOG_ERR("failed to allocate datapath %s of type %s",
526 datapath_name, datapath_type);
527 return ENOMEM;
528 }
529
530 /* Initialize. */
531 ovs_mutex_lock(&ofproto_mutex);
532 memset(ofproto, 0, sizeof *ofproto);
533 ofproto->ofproto_class = class;
534 ofproto->name = xstrdup(datapath_name);
535 ofproto->type = xstrdup(datapath_type);
536 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
537 hash_string(ofproto->name, 0));
538 ofproto->datapath_id = 0;
539 ofproto->forward_bpdu = false;
540 ofproto->fallback_dpid = pick_fallback_dpid();
541 ofproto->mfr_desc = NULL;
542 ofproto->hw_desc = NULL;
543 ofproto->sw_desc = NULL;
544 ofproto->serial_desc = NULL;
545 ofproto->dp_desc = NULL;
546 ofproto->frag_handling = OFPUTIL_FRAG_NORMAL;
547 hmap_init(&ofproto->ports);
548 hmap_init(&ofproto->ofport_usage);
549 shash_init(&ofproto->port_by_name);
550 simap_init(&ofproto->ofp_requests);
551 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
552 ofproto->eviction_group_timer = LLONG_MIN;
553 ofproto->tables = NULL;
554 ofproto->n_tables = 0;
555 ofproto->tables_version = OVS_VERSION_MIN;
556 hindex_init(&ofproto->cookies);
557 hmap_init(&ofproto->learned_cookies);
558 ovs_list_init(&ofproto->expirable);
559 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
560 guarded_list_init(&ofproto->rule_executes);
561 ofproto->min_mtu = INT_MAX;
562 ovs_rwlock_init(&ofproto->groups_rwlock);
563 cmap_init(&ofproto->groups);
564 ovs_mutex_unlock(&ofproto_mutex);
565 ofproto->ogf.types = 0xf;
566 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
567 OFPGFC_SELECT_WEIGHT;
568 for (i = 0; i < 4; i++) {
569 ofproto->ogf.max_groups[i] = OFPG_MAX;
570 ofproto->ogf.ofpacts[i] = (UINT64_C(1) << N_OFPACTS) - 1;
571 }
572 tun_metadata_init();
573
574 error = ofproto->ofproto_class->construct(ofproto);
575 if (error) {
576 VLOG_ERR("failed to open datapath %s: %s",
577 datapath_name, ovs_strerror(error));
578 connmgr_destroy(ofproto->connmgr);
579 ofproto_destroy__(ofproto);
580 return error;
581 }
582
583 /* Check that hidden tables, if any, are at the end. */
584 ovs_assert(ofproto->n_tables);
585 for (i = 0; i + 1 < ofproto->n_tables; i++) {
586 enum oftable_flags flags = ofproto->tables[i].flags;
587 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
588
589 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
590 }
591
592 ofproto->datapath_id = pick_datapath_id(ofproto);
593 init_ports(ofproto);
594
595 /* Initialize meters table. */
596 if (ofproto->ofproto_class->meter_get_features) {
597 ofproto->ofproto_class->meter_get_features(ofproto,
598 &ofproto->meter_features);
599 } else {
600 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
601 }
602 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
603 * sizeof(struct meter *));
604
605 /* Set the initial tables version. */
606 ofproto_bump_tables_version(ofproto);
607
608 *ofprotop = ofproto;
609 return 0;
610 }
611
612 /* Must be called (only) by an ofproto implementation in its constructor
613 * function. See the large comment on 'construct' in struct ofproto_class for
614 * details. */
615 void
616 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
617 {
618 struct oftable *table;
619
620 ovs_assert(!ofproto->n_tables);
621 ovs_assert(n_tables >= 1 && n_tables <= 255);
622
623 ofproto->n_tables = n_tables;
624 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
625 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
626 oftable_init(table);
627 }
628 }
629
630 /* To be optionally called (only) by an ofproto implementation in its
631 * constructor function. See the large comment on 'construct' in struct
632 * ofproto_class for details.
633 *
634 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
635 * will then ensure that actions passed into the ofproto implementation will
636 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
637 * function is not called, there will be no such restriction.
638 *
639 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
640 * the 'max_ports' restriction. */
641 void
642 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
643 {
644 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
645 ofproto->max_ports = max_ports;
646 }
647
648 uint64_t
649 ofproto_get_datapath_id(const struct ofproto *ofproto)
650 {
651 return ofproto->datapath_id;
652 }
653
654 void
655 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
656 {
657 uint64_t old_dpid = p->datapath_id;
658 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
659 if (p->datapath_id != old_dpid) {
660 /* Force all active connections to reconnect, since there is no way to
661 * notify a controller that the datapath ID has changed. */
662 ofproto_reconnect_controllers(p);
663 }
664 }
665
666 void
667 ofproto_set_controllers(struct ofproto *p,
668 const struct ofproto_controller *controllers,
669 size_t n_controllers, uint32_t allowed_versions)
670 {
671 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
672 allowed_versions);
673 }
674
675 void
676 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
677 {
678 connmgr_set_fail_mode(p->connmgr, fail_mode);
679 }
680
681 /* Drops the connections between 'ofproto' and all of its controllers, forcing
682 * them to reconnect. */
683 void
684 ofproto_reconnect_controllers(struct ofproto *ofproto)
685 {
686 connmgr_reconnect(ofproto->connmgr);
687 }
688
689 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
690 * in-band control should guarantee access, in the same way that in-band
691 * control guarantees access to OpenFlow controllers. */
692 void
693 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
694 const struct sockaddr_in *extras, size_t n)
695 {
696 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
697 }
698
699 /* Sets the OpenFlow queue used by flows set up by in-band control on
700 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
701 * flows will use the default queue. */
702 void
703 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
704 {
705 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
706 }
707
708 /* Sets the number of flows at which eviction from the kernel flow table
709 * will occur. */
710 void
711 ofproto_set_flow_limit(unsigned limit)
712 {
713 ofproto_flow_limit = limit;
714 }
715
716 /* Sets the maximum idle time for flows in the datapath before they are
717 * expired. */
718 void
719 ofproto_set_max_idle(unsigned max_idle)
720 {
721 ofproto_max_idle = max_idle;
722 }
723
724 /* If forward_bpdu is true, the NORMAL action will forward frames with
725 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
726 * the NORMAL action will drop these frames. */
727 void
728 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
729 {
730 bool old_val = ofproto->forward_bpdu;
731 ofproto->forward_bpdu = forward_bpdu;
732 if (old_val != ofproto->forward_bpdu) {
733 if (ofproto->ofproto_class->forward_bpdu_changed) {
734 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
735 }
736 }
737 }
738
739 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
740 * 'idle_time', in seconds, and the maximum number of MAC table entries to
741 * 'max_entries'. */
742 void
743 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
744 size_t max_entries)
745 {
746 if (ofproto->ofproto_class->set_mac_table_config) {
747 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
748 max_entries);
749 }
750 }
751
752 /* Multicast snooping configuration. */
753
754 /* Configures multicast snooping on 'ofproto' using the settings
755 * defined in 's'. If 's' is NULL, disables multicast snooping.
756 *
757 * Returns 0 if successful, otherwise a positive errno value. */
758 int
759 ofproto_set_mcast_snooping(struct ofproto *ofproto,
760 const struct ofproto_mcast_snooping_settings *s)
761 {
762 return (ofproto->ofproto_class->set_mcast_snooping
763 ? ofproto->ofproto_class->set_mcast_snooping(ofproto, s)
764 : EOPNOTSUPP);
765 }
766
767 /* Configures multicast snooping flood settings on 'ofp_port' of 'ofproto'.
768 *
769 * Returns 0 if successful, otherwise a positive errno value.*/
770 int
771 ofproto_port_set_mcast_snooping(struct ofproto *ofproto, void *aux,
772 const struct ofproto_mcast_snooping_port_settings *s)
773 {
774 return (ofproto->ofproto_class->set_mcast_snooping_port
775 ? ofproto->ofproto_class->set_mcast_snooping_port(ofproto, aux, s)
776 : EOPNOTSUPP);
777 }
778
779 void
780 ofproto_set_cpu_mask(const char *cmask)
781 {
782 free(pmd_cpu_mask);
783 pmd_cpu_mask = nullable_xstrdup(cmask);
784 }
785
786 void
787 ofproto_set_threads(int n_handlers_, int n_revalidators_)
788 {
789 int threads = MAX(count_cpu_cores(), 2);
790
791 n_revalidators = MAX(n_revalidators_, 0);
792 n_handlers = MAX(n_handlers_, 0);
793
794 if (!n_revalidators) {
795 n_revalidators = n_handlers
796 ? MAX(threads - (int) n_handlers, 1)
797 : threads / 4 + 1;
798 }
799
800 if (!n_handlers) {
801 n_handlers = MAX(threads - (int) n_revalidators, 1);
802 }
803 }
804
805 void
806 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
807 {
808 free(p->dp_desc);
809 p->dp_desc = nullable_xstrdup(dp_desc);
810 }
811
812 int
813 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
814 {
815 return connmgr_set_snoops(ofproto->connmgr, snoops);
816 }
817
818 int
819 ofproto_set_netflow(struct ofproto *ofproto,
820 const struct netflow_options *nf_options)
821 {
822 if (nf_options && sset_is_empty(&nf_options->collectors)) {
823 nf_options = NULL;
824 }
825
826 if (ofproto->ofproto_class->set_netflow) {
827 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
828 } else {
829 return nf_options ? EOPNOTSUPP : 0;
830 }
831 }
832
833 int
834 ofproto_set_sflow(struct ofproto *ofproto,
835 const struct ofproto_sflow_options *oso)
836 {
837 if (oso && sset_is_empty(&oso->targets)) {
838 oso = NULL;
839 }
840
841 if (ofproto->ofproto_class->set_sflow) {
842 return ofproto->ofproto_class->set_sflow(ofproto, oso);
843 } else {
844 return oso ? EOPNOTSUPP : 0;
845 }
846 }
847
848 int
849 ofproto_set_ipfix(struct ofproto *ofproto,
850 const struct ofproto_ipfix_bridge_exporter_options *bo,
851 const struct ofproto_ipfix_flow_exporter_options *fo,
852 size_t n_fo)
853 {
854 if (ofproto->ofproto_class->set_ipfix) {
855 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
856 } else {
857 return (bo || fo) ? EOPNOTSUPP : 0;
858 }
859 }
860
861 static int
862 ofproto_get_ipfix_stats(struct ofproto *ofproto,
863 bool bridge_ipfix,
864 struct ovs_list *replies)
865 {
866 int error;
867
868 if (ofproto->ofproto_class->get_ipfix_stats) {
869 error = ofproto->ofproto_class->get_ipfix_stats(ofproto,
870 bridge_ipfix,
871 replies);
872 } else {
873 error = EOPNOTSUPP;
874 }
875
876 return error;
877 }
878
879 static enum ofperr
880 handle_ipfix_bridge_stats_request(struct ofconn *ofconn,
881 const struct ofp_header *request)
882 {
883 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
884 struct ovs_list replies;
885 enum ofperr error;
886
887 ofpmp_init(&replies, request);
888 error = ofproto_get_ipfix_stats(ofproto, true, &replies);
889
890 if (!error) {
891 ofconn_send_replies(ofconn, &replies);
892 } else {
893 ofpbuf_list_delete(&replies);
894 }
895
896 return error;
897 }
898
899 static enum ofperr
900 handle_ipfix_flow_stats_request(struct ofconn *ofconn,
901 const struct ofp_header *request)
902 {
903 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
904 struct ovs_list replies;
905 enum ofperr error;
906
907 ofpmp_init(&replies, request);
908 error = ofproto_get_ipfix_stats(ofproto, false, &replies);
909
910 if (!error) {
911 ofconn_send_replies(ofconn, &replies);
912 } else {
913 ofpbuf_list_delete(&replies);
914 }
915
916 return error;
917 }
918
919 void
920 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
921 {
922 flow_restore_wait = flow_restore_wait_db;
923 }
924
925 bool
926 ofproto_get_flow_restore_wait(void)
927 {
928 return flow_restore_wait;
929 }
930
931 \f
932 /* Spanning Tree Protocol (STP) configuration. */
933
934 /* Configures STP on 'ofproto' using the settings defined in 's'. If
935 * 's' is NULL, disables STP.
936 *
937 * Returns 0 if successful, otherwise a positive errno value. */
938 int
939 ofproto_set_stp(struct ofproto *ofproto,
940 const struct ofproto_stp_settings *s)
941 {
942 return (ofproto->ofproto_class->set_stp
943 ? ofproto->ofproto_class->set_stp(ofproto, s)
944 : EOPNOTSUPP);
945 }
946
947 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
948 * 'enabled' member of 's' is false, then the other members are not
949 * meaningful.
950 *
951 * Returns 0 if successful, otherwise a positive errno value. */
952 int
953 ofproto_get_stp_status(struct ofproto *ofproto,
954 struct ofproto_stp_status *s)
955 {
956 return (ofproto->ofproto_class->get_stp_status
957 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
958 : EOPNOTSUPP);
959 }
960
961 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
962 * in 's'. The caller is responsible for assigning STP port numbers
963 * (using the 'port_num' member in the range of 1 through 255, inclusive)
964 * and ensuring there are no duplicates. If the 's' is NULL, then STP
965 * is disabled on the port.
966 *
967 * Returns 0 if successful, otherwise a positive errno value.*/
968 int
969 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
970 const struct ofproto_port_stp_settings *s)
971 {
972 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
973 if (!ofport) {
974 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
975 ofproto->name, ofp_port);
976 return ENODEV;
977 }
978
979 return (ofproto->ofproto_class->set_stp_port
980 ? ofproto->ofproto_class->set_stp_port(ofport, s)
981 : EOPNOTSUPP);
982 }
983
984 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
985 * 's'. If the 'enabled' member in 's' is false, then the other members
986 * are not meaningful.
987 *
988 * Returns 0 if successful, otherwise a positive errno value.*/
989 int
990 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
991 struct ofproto_port_stp_status *s)
992 {
993 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
994 if (!ofport) {
995 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
996 "port %"PRIu16, ofproto->name, ofp_port);
997 return ENODEV;
998 }
999
1000 return (ofproto->ofproto_class->get_stp_port_status
1001 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
1002 : EOPNOTSUPP);
1003 }
1004
1005 /* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
1006 * 's'. If the 'enabled' member in 's' is false, then the other members
1007 * are not meaningful.
1008 *
1009 * Returns 0 if successful, otherwise a positive errno value.*/
1010 int
1011 ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
1012 struct ofproto_port_stp_stats *s)
1013 {
1014 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1015 if (!ofport) {
1016 VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
1017 "port %"PRIu16, ofproto->name, ofp_port);
1018 return ENODEV;
1019 }
1020
1021 return (ofproto->ofproto_class->get_stp_port_stats
1022 ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
1023 : EOPNOTSUPP);
1024 }
1025
1026 /* Rapid Spanning Tree Protocol (RSTP) configuration. */
1027
1028 /* Configures RSTP on 'ofproto' using the settings defined in 's'. If
1029 * 's' is NULL, disables RSTP.
1030 *
1031 * Returns 0 if successful, otherwise a positive errno value. */
1032 int
1033 ofproto_set_rstp(struct ofproto *ofproto,
1034 const struct ofproto_rstp_settings *s)
1035 {
1036 if (!ofproto->ofproto_class->set_rstp) {
1037 return EOPNOTSUPP;
1038 }
1039 ofproto->ofproto_class->set_rstp(ofproto, s);
1040 return 0;
1041 }
1042
1043 /* Retrieves RSTP status of 'ofproto' and stores it in 's'. If the
1044 * 'enabled' member of 's' is false, then the other members are not
1045 * meaningful.
1046 *
1047 * Returns 0 if successful, otherwise a positive errno value. */
1048 int
1049 ofproto_get_rstp_status(struct ofproto *ofproto,
1050 struct ofproto_rstp_status *s)
1051 {
1052 if (!ofproto->ofproto_class->get_rstp_status) {
1053 return EOPNOTSUPP;
1054 }
1055 ofproto->ofproto_class->get_rstp_status(ofproto, s);
1056 return 0;
1057 }
1058
1059 /* Configures RSTP on 'ofp_port' of 'ofproto' using the settings defined
1060 * in 's'. The caller is responsible for assigning RSTP port numbers
1061 * (using the 'port_num' member in the range of 1 through 255, inclusive)
1062 * and ensuring there are no duplicates. If the 's' is NULL, then RSTP
1063 * is disabled on the port.
1064 *
1065 * Returns 0 if successful, otherwise a positive errno value.*/
1066 int
1067 ofproto_port_set_rstp(struct ofproto *ofproto, ofp_port_t ofp_port,
1068 const struct ofproto_port_rstp_settings *s)
1069 {
1070 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1071 if (!ofport) {
1072 VLOG_WARN("%s: cannot configure RSTP on nonexistent port %"PRIu16,
1073 ofproto->name, ofp_port);
1074 return ENODEV;
1075 }
1076
1077 if (!ofproto->ofproto_class->set_rstp_port) {
1078 return EOPNOTSUPP;
1079 }
1080 ofproto->ofproto_class->set_rstp_port(ofport, s);
1081 return 0;
1082 }
1083
1084 /* Retrieves RSTP port status of 'ofp_port' on 'ofproto' and stores it in
1085 * 's'. If the 'enabled' member in 's' is false, then the other members
1086 * are not meaningful.
1087 *
1088 * Returns 0 if successful, otherwise a positive errno value.*/
1089 int
1090 ofproto_port_get_rstp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1091 struct ofproto_port_rstp_status *s)
1092 {
1093 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1094 if (!ofport) {
1095 VLOG_WARN_RL(&rl, "%s: cannot get RSTP status on nonexistent "
1096 "port %"PRIu16, ofproto->name, ofp_port);
1097 return ENODEV;
1098 }
1099
1100 if (!ofproto->ofproto_class->get_rstp_port_status) {
1101 return EOPNOTSUPP;
1102 }
1103 ofproto->ofproto_class->get_rstp_port_status(ofport, s);
1104 return 0;
1105 }
1106 \f
1107 /* Queue DSCP configuration. */
1108
1109 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1110 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
1111 * to implement QoS. Instead, it is used to implement features which require
1112 * knowledge of what queues exist on a port, and some basic information about
1113 * them.
1114 *
1115 * Returns 0 if successful, otherwise a positive errno value. */
1116 int
1117 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
1118 const struct ofproto_port_queue *queues,
1119 size_t n_queues)
1120 {
1121 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1122
1123 if (!ofport) {
1124 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
1125 ofproto->name, ofp_port);
1126 return ENODEV;
1127 }
1128
1129 return (ofproto->ofproto_class->set_queues
1130 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
1131 : EOPNOTSUPP);
1132 }
1133 \f
1134 /* LLDP configuration. */
1135 void
1136 ofproto_port_set_lldp(struct ofproto *ofproto,
1137 ofp_port_t ofp_port,
1138 const struct smap *cfg)
1139 {
1140 struct ofport *ofport;
1141 int error;
1142
1143 ofport = ofproto_get_port(ofproto, ofp_port);
1144 if (!ofport) {
1145 VLOG_WARN("%s: cannot configure LLDP on nonexistent port %"PRIu16,
1146 ofproto->name, ofp_port);
1147 return;
1148 }
1149 error = (ofproto->ofproto_class->set_lldp
1150 ? ofproto->ofproto_class->set_lldp(ofport, cfg)
1151 : EOPNOTSUPP);
1152 if (error) {
1153 VLOG_WARN("%s: lldp configuration on port %"PRIu16" (%s) failed (%s)",
1154 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1155 ovs_strerror(error));
1156 }
1157 }
1158
1159 int
1160 ofproto_set_aa(struct ofproto *ofproto, void *aux OVS_UNUSED,
1161 const struct aa_settings *s)
1162 {
1163 if (!ofproto->ofproto_class->set_aa) {
1164 return EOPNOTSUPP;
1165 }
1166 ofproto->ofproto_class->set_aa(ofproto, s);
1167 return 0;
1168 }
1169
1170 int
1171 ofproto_aa_mapping_register(struct ofproto *ofproto, void *aux,
1172 const struct aa_mapping_settings *s)
1173 {
1174 if (!ofproto->ofproto_class->aa_mapping_set) {
1175 return EOPNOTSUPP;
1176 }
1177 ofproto->ofproto_class->aa_mapping_set(ofproto, aux, s);
1178 return 0;
1179 }
1180
1181 int
1182 ofproto_aa_mapping_unregister(struct ofproto *ofproto, void *aux)
1183 {
1184 if (!ofproto->ofproto_class->aa_mapping_unset) {
1185 return EOPNOTSUPP;
1186 }
1187 ofproto->ofproto_class->aa_mapping_unset(ofproto, aux);
1188 return 0;
1189 }
1190
1191 int
1192 ofproto_aa_vlan_get_queued(struct ofproto *ofproto,
1193 struct ovs_list *list)
1194 {
1195 if (!ofproto->ofproto_class->aa_vlan_get_queued) {
1196 return EOPNOTSUPP;
1197 }
1198 ofproto->ofproto_class->aa_vlan_get_queued(ofproto, list);
1199 return 0;
1200 }
1201
1202 unsigned int
1203 ofproto_aa_vlan_get_queue_size(struct ofproto *ofproto)
1204 {
1205 if (!ofproto->ofproto_class->aa_vlan_get_queue_size) {
1206 return EOPNOTSUPP;
1207 }
1208 return ofproto->ofproto_class->aa_vlan_get_queue_size(ofproto);
1209 }
1210
1211 /* Connectivity Fault Management configuration. */
1212
1213 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
1214 void
1215 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
1216 {
1217 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1218 if (ofport && ofproto->ofproto_class->set_cfm) {
1219 ofproto->ofproto_class->set_cfm(ofport, NULL);
1220 }
1221 }
1222
1223 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
1224 * basic configuration from the configuration members in 'cfm', and the remote
1225 * maintenance point ID from remote_mpid. Ignores the statistics members of
1226 * 'cfm'.
1227 *
1228 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
1229 void
1230 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
1231 const struct cfm_settings *s)
1232 {
1233 struct ofport *ofport;
1234 int error;
1235
1236 ofport = ofproto_get_port(ofproto, ofp_port);
1237 if (!ofport) {
1238 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
1239 ofproto->name, ofp_port);
1240 return;
1241 }
1242
1243 /* XXX: For configuration simplicity, we only support one remote_mpid
1244 * outside of the CFM module. It's not clear if this is the correct long
1245 * term solution or not. */
1246 error = (ofproto->ofproto_class->set_cfm
1247 ? ofproto->ofproto_class->set_cfm(ofport, s)
1248 : EOPNOTSUPP);
1249 if (error) {
1250 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
1251 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1252 ovs_strerror(error));
1253 }
1254 }
1255
1256 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
1257 * 'ofproto' does not have a port 'ofp_port'. */
1258 void
1259 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
1260 const struct smap *cfg)
1261 {
1262 struct ofport *ofport;
1263 int error;
1264
1265 ofport = ofproto_get_port(ofproto, ofp_port);
1266 if (!ofport) {
1267 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
1268 ofproto->name, ofp_port);
1269 return;
1270 }
1271
1272 error = (ofproto->ofproto_class->set_bfd
1273 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
1274 : EOPNOTSUPP);
1275 if (error) {
1276 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
1277 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1278 ovs_strerror(error));
1279 }
1280 }
1281
1282 /* Checks the status change of BFD on 'ofport'.
1283 *
1284 * Returns true if 'ofproto_class' does not support 'bfd_status_changed'. */
1285 bool
1286 ofproto_port_bfd_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
1287 {
1288 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1289 return (ofport && ofproto->ofproto_class->bfd_status_changed
1290 ? ofproto->ofproto_class->bfd_status_changed(ofport)
1291 : true);
1292 }
1293
1294 /* Populates 'status' with the status of BFD on 'ofport'. Returns 0 on
1295 * success. Returns a positive errno otherwise. Has no effect if 'ofp_port'
1296 * is not an OpenFlow port in 'ofproto'.
1297 *
1298 * The caller must provide and own '*status'. */
1299 int
1300 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1301 struct smap *status)
1302 {
1303 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1304 return (ofport && ofproto->ofproto_class->get_bfd_status
1305 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1306 : EOPNOTSUPP);
1307 }
1308
1309 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1310 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1311 * 0 if LACP partner information is not current (generally indicating a
1312 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1313 int
1314 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1315 {
1316 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1317 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1318 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1319 : -1);
1320 }
1321
1322 int
1323 ofproto_port_get_lacp_stats(const struct ofport *port, struct lacp_slave_stats *stats)
1324 {
1325 struct ofproto *ofproto = port->ofproto;
1326 int error;
1327
1328 if (ofproto->ofproto_class->port_get_lacp_stats) {
1329 error = ofproto->ofproto_class->port_get_lacp_stats(port, stats);
1330 } else {
1331 error = EOPNOTSUPP;
1332 }
1333
1334 return error;
1335 }
1336 \f
1337 /* Bundles. */
1338
1339 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1340 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1341 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1342 * configuration plus, if there is more than one slave, a bonding
1343 * configuration.
1344 *
1345 * If 'aux' is already registered then this function updates its configuration
1346 * to 's'. Otherwise, this function registers a new bundle.
1347 *
1348 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1349 * port. */
1350 int
1351 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1352 const struct ofproto_bundle_settings *s)
1353 {
1354 return (ofproto->ofproto_class->bundle_set
1355 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1356 : EOPNOTSUPP);
1357 }
1358
1359 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1360 * If no such bundle has been registered, this has no effect. */
1361 int
1362 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1363 {
1364 return ofproto_bundle_register(ofproto, aux, NULL);
1365 }
1366
1367 \f
1368 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1369 * If 'aux' is already registered then this function updates its configuration
1370 * to 's'. Otherwise, this function registers a new mirror. */
1371 int
1372 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1373 const struct ofproto_mirror_settings *s)
1374 {
1375 return (ofproto->ofproto_class->mirror_set
1376 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1377 : EOPNOTSUPP);
1378 }
1379
1380 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1381 * If no mirror has been registered, this has no effect. */
1382 int
1383 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1384 {
1385 return ofproto_mirror_register(ofproto, aux, NULL);
1386 }
1387
1388 /* Retrieves statistics from mirror associated with client data pointer
1389 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1390 * 'bytes', respectively. If a particular counters is not supported,
1391 * the appropriate argument is set to UINT64_MAX.
1392 */
1393 int
1394 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1395 uint64_t *packets, uint64_t *bytes)
1396 {
1397 if (!ofproto->ofproto_class->mirror_get_stats) {
1398 *packets = *bytes = UINT64_MAX;
1399 return EOPNOTSUPP;
1400 }
1401
1402 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1403 packets, bytes);
1404 }
1405
1406 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1407 * which all packets are flooded, instead of using MAC learning. If
1408 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1409 *
1410 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1411 * port. */
1412 int
1413 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1414 {
1415 return (ofproto->ofproto_class->set_flood_vlans
1416 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1417 : EOPNOTSUPP);
1418 }
1419
1420 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1421 * output for a mirror. */
1422 bool
1423 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1424 {
1425 return (ofproto->ofproto_class->is_mirror_output_bundle
1426 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1427 : false);
1428 }
1429 \f
1430 /* Configuration of OpenFlow tables. */
1431
1432 /* Returns the number of OpenFlow tables in 'ofproto'. */
1433 int
1434 ofproto_get_n_tables(const struct ofproto *ofproto)
1435 {
1436 return ofproto->n_tables;
1437 }
1438
1439 /* Returns the number of Controller visible OpenFlow tables
1440 * in 'ofproto'. This number will exclude Hidden tables.
1441 * This funtion's return value should be less or equal to that of
1442 * ofproto_get_n_tables() . */
1443 uint8_t
1444 ofproto_get_n_visible_tables(const struct ofproto *ofproto)
1445 {
1446 uint8_t n = ofproto->n_tables;
1447
1448 /* Count only non-hidden tables in the number of tables. (Hidden tables,
1449 * if present, are always at the end.) */
1450 while(n && (ofproto->tables[n - 1].flags & OFTABLE_HIDDEN)) {
1451 n--;
1452 }
1453
1454 return n;
1455 }
1456
1457 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1458 * settings from 's'. 'table_id' must be in the range 0 through the number of
1459 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1460 *
1461 * For read-only tables, only the name may be configured. */
1462 void
1463 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1464 const struct ofproto_table_settings *s)
1465 {
1466 struct oftable *table;
1467
1468 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1469 table = &ofproto->tables[table_id];
1470
1471 oftable_set_name(table, s->name);
1472
1473 if (table->flags & OFTABLE_READONLY) {
1474 return;
1475 }
1476
1477 if (classifier_set_prefix_fields(&table->cls,
1478 s->prefix_fields, s->n_prefix_fields)) {
1479 /* XXX: Trigger revalidation. */
1480 }
1481
1482 ovs_mutex_lock(&ofproto_mutex);
1483 unsigned int new_eviction = (s->enable_eviction
1484 ? table->eviction | EVICTION_CLIENT
1485 : table->eviction & ~EVICTION_CLIENT);
1486 oftable_configure_eviction(table, new_eviction, s->groups, s->n_groups);
1487 table->max_flows = s->max_flows;
1488 evict_rules_from_table(table);
1489 ovs_mutex_unlock(&ofproto_mutex);
1490 }
1491 \f
1492 bool
1493 ofproto_has_snoops(const struct ofproto *ofproto)
1494 {
1495 return connmgr_has_snoops(ofproto->connmgr);
1496 }
1497
1498 void
1499 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1500 {
1501 connmgr_get_snoops(ofproto->connmgr, snoops);
1502 }
1503
1504 /* Deletes 'rule' from 'ofproto'.
1505 *
1506 * Within an ofproto implementation, this function allows an ofproto
1507 * implementation to destroy any rules that remain when its ->destruct()
1508 * function is called. This function is not suitable for use elsewhere in an
1509 * ofproto implementation.
1510 *
1511 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1512 * Cycle" in ofproto-provider.h. */
1513 void
1514 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1515 OVS_EXCLUDED(ofproto_mutex)
1516 {
1517 /* This skips the ofmonitor and flow-removed notifications because the
1518 * switch is being deleted and any OpenFlow channels have been or soon will
1519 * be killed. */
1520 ovs_mutex_lock(&ofproto_mutex);
1521
1522 if (!rule->removed) {
1523 /* Make sure there is no postponed removal of the rule. */
1524 ovs_assert(cls_rule_visible_in_version(&rule->cr, OVS_VERSION_MAX));
1525
1526 if (!classifier_remove(&rule->ofproto->tables[rule->table_id].cls,
1527 &rule->cr)) {
1528 OVS_NOT_REACHED();
1529 }
1530 ofproto_rule_remove__(rule->ofproto, rule);
1531 if (ofproto->ofproto_class->rule_delete) {
1532 ofproto->ofproto_class->rule_delete(rule);
1533 }
1534 ofproto_rule_unref(rule);
1535 }
1536 ovs_mutex_unlock(&ofproto_mutex);
1537 }
1538
1539 static void
1540 ofproto_flush__(struct ofproto *ofproto)
1541 OVS_EXCLUDED(ofproto_mutex)
1542 {
1543 struct oftable *table;
1544
1545 /* This will flush all datapath flows. */
1546 if (ofproto->ofproto_class->flush) {
1547 ofproto->ofproto_class->flush(ofproto);
1548 }
1549
1550 /* XXX: There is a small race window here, where new datapath flows can be
1551 * created by upcall handlers based on the existing flow table. We can not
1552 * call ofproto class flush while holding 'ofproto_mutex' to prevent this,
1553 * as then we could deadlock on syncing with the handler threads waiting on
1554 * the same mutex. */
1555
1556 ovs_mutex_lock(&ofproto_mutex);
1557 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1558 struct rule_collection rules;
1559 struct rule *rule;
1560
1561 if (table->flags & OFTABLE_HIDDEN) {
1562 continue;
1563 }
1564
1565 rule_collection_init(&rules);
1566
1567 CLS_FOR_EACH (rule, cr, &table->cls) {
1568 rule_collection_add(&rules, rule);
1569 }
1570 delete_flows__(&rules, OFPRR_DELETE, NULL);
1571 }
1572 /* XXX: Concurrent handler threads may insert new learned flows based on
1573 * learn actions of the now deleted flows right after we release
1574 * 'ofproto_mutex'. */
1575 ovs_mutex_unlock(&ofproto_mutex);
1576 }
1577
1578 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1579
1580 static void
1581 ofproto_destroy__(struct ofproto *ofproto)
1582 OVS_EXCLUDED(ofproto_mutex)
1583 {
1584 struct oftable *table;
1585
1586 destroy_rule_executes(ofproto);
1587
1588 guarded_list_destroy(&ofproto->rule_executes);
1589 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1590 cmap_destroy(&ofproto->groups);
1591
1592 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1593 free(ofproto->name);
1594 free(ofproto->type);
1595 free(ofproto->mfr_desc);
1596 free(ofproto->hw_desc);
1597 free(ofproto->sw_desc);
1598 free(ofproto->serial_desc);
1599 free(ofproto->dp_desc);
1600 hmap_destroy(&ofproto->ports);
1601 hmap_destroy(&ofproto->ofport_usage);
1602 shash_destroy(&ofproto->port_by_name);
1603 simap_destroy(&ofproto->ofp_requests);
1604
1605 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1606 oftable_destroy(table);
1607 }
1608 free(ofproto->tables);
1609
1610 ovs_assert(hindex_is_empty(&ofproto->cookies));
1611 hindex_destroy(&ofproto->cookies);
1612
1613 ovs_assert(hmap_is_empty(&ofproto->learned_cookies));
1614 hmap_destroy(&ofproto->learned_cookies);
1615
1616 ofproto->ofproto_class->dealloc(ofproto);
1617 }
1618
1619 /* Destroying rules is doubly deferred, must have 'ofproto' around for them.
1620 * - 1st we defer the removal of the rules from the classifier
1621 * - 2nd we defer the actual destruction of the rules. */
1622 static void
1623 ofproto_destroy_defer__(struct ofproto *ofproto)
1624 OVS_EXCLUDED(ofproto_mutex)
1625 {
1626 ovsrcu_postpone(ofproto_destroy__, ofproto);
1627 }
1628
1629 void
1630 ofproto_destroy(struct ofproto *p, bool del)
1631 OVS_EXCLUDED(ofproto_mutex)
1632 {
1633 struct ofport *ofport, *next_ofport;
1634 struct ofport_usage *usage;
1635
1636 if (!p) {
1637 return;
1638 }
1639
1640 if (p->meters) {
1641 meter_delete(p, 1, p->meter_features.max_meters);
1642 p->meter_features.max_meters = 0;
1643 free(p->meters);
1644 p->meters = NULL;
1645 }
1646
1647 ofproto_flush__(p);
1648 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1649 ofport_destroy(ofport, del);
1650 }
1651
1652 HMAP_FOR_EACH_POP (usage, hmap_node, &p->ofport_usage) {
1653 free(usage);
1654 }
1655
1656 p->ofproto_class->destruct(p);
1657
1658 /* We should not postpone this because it involves deleting a listening
1659 * socket which we may want to reopen soon. 'connmgr' should not be used
1660 * by other threads */
1661 connmgr_destroy(p->connmgr);
1662
1663 /* Destroying rules is deferred, must have 'ofproto' around for them. */
1664 ovsrcu_postpone(ofproto_destroy_defer__, p);
1665 }
1666
1667 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1668 * kernel datapath, for example, this destroys the datapath in the kernel, and
1669 * with the netdev-based datapath, it tears down the data structures that
1670 * represent the datapath.
1671 *
1672 * The datapath should not be currently open as an ofproto. */
1673 int
1674 ofproto_delete(const char *name, const char *type)
1675 {
1676 const struct ofproto_class *class = ofproto_class_find__(type);
1677 return (!class ? EAFNOSUPPORT
1678 : !class->del ? EACCES
1679 : class->del(type, name));
1680 }
1681
1682 static void
1683 process_port_change(struct ofproto *ofproto, int error, char *devname)
1684 {
1685 if (error == ENOBUFS) {
1686 reinit_ports(ofproto);
1687 } else if (!error) {
1688 update_port(ofproto, devname);
1689 free(devname);
1690 }
1691 }
1692
1693 int
1694 ofproto_type_run(const char *datapath_type)
1695 {
1696 const struct ofproto_class *class;
1697 int error;
1698
1699 datapath_type = ofproto_normalize_type(datapath_type);
1700 class = ofproto_class_find__(datapath_type);
1701
1702 error = class->type_run ? class->type_run(datapath_type) : 0;
1703 if (error && error != EAGAIN) {
1704 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1705 datapath_type, ovs_strerror(error));
1706 }
1707 return error;
1708 }
1709
1710 void
1711 ofproto_type_wait(const char *datapath_type)
1712 {
1713 const struct ofproto_class *class;
1714
1715 datapath_type = ofproto_normalize_type(datapath_type);
1716 class = ofproto_class_find__(datapath_type);
1717
1718 if (class->type_wait) {
1719 class->type_wait(datapath_type);
1720 }
1721 }
1722
1723 int
1724 ofproto_run(struct ofproto *p)
1725 {
1726 int error;
1727 uint64_t new_seq;
1728
1729 error = p->ofproto_class->run(p);
1730 if (error && error != EAGAIN) {
1731 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1732 }
1733
1734 run_rule_executes(p);
1735
1736 /* Restore the eviction group heap invariant occasionally. */
1737 if (p->eviction_group_timer < time_msec()) {
1738 size_t i;
1739
1740 p->eviction_group_timer = time_msec() + 1000;
1741
1742 for (i = 0; i < p->n_tables; i++) {
1743 struct oftable *table = &p->tables[i];
1744 struct eviction_group *evg;
1745 struct rule *rule;
1746
1747 if (!table->eviction) {
1748 continue;
1749 }
1750
1751 if (table->n_flows > 100000) {
1752 static struct vlog_rate_limit count_rl =
1753 VLOG_RATE_LIMIT_INIT(1, 1);
1754 VLOG_WARN_RL(&count_rl, "Table %"PRIuSIZE" has an excessive"
1755 " number of rules: %d", i, table->n_flows);
1756 }
1757
1758 ovs_mutex_lock(&ofproto_mutex);
1759 CLS_FOR_EACH (rule, cr, &table->cls) {
1760 if (rule->idle_timeout || rule->hard_timeout) {
1761 if (!rule->eviction_group) {
1762 eviction_group_add_rule(rule);
1763 } else {
1764 heap_raw_change(&rule->evg_node,
1765 rule_eviction_priority(p, rule));
1766 }
1767 }
1768 }
1769
1770 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1771 heap_rebuild(&evg->rules);
1772 }
1773 ovs_mutex_unlock(&ofproto_mutex);
1774 }
1775 }
1776
1777 if (p->ofproto_class->port_poll) {
1778 char *devname;
1779
1780 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1781 process_port_change(p, error, devname);
1782 }
1783 }
1784
1785 new_seq = seq_read(connectivity_seq_get());
1786 if (new_seq != p->change_seq) {
1787 struct sset devnames;
1788 const char *devname;
1789 struct ofport *ofport;
1790
1791 /* Update OpenFlow port status for any port whose netdev has changed.
1792 *
1793 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1794 * destroyed, so it's not safe to update ports directly from the
1795 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1796 * need this two-phase approach. */
1797 sset_init(&devnames);
1798 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1799 uint64_t port_change_seq;
1800
1801 port_change_seq = netdev_get_change_seq(ofport->netdev);
1802 if (ofport->change_seq != port_change_seq) {
1803 ofport->change_seq = port_change_seq;
1804 sset_add(&devnames, netdev_get_name(ofport->netdev));
1805 }
1806 }
1807 SSET_FOR_EACH (devname, &devnames) {
1808 update_port(p, devname);
1809 }
1810 sset_destroy(&devnames);
1811
1812 p->change_seq = new_seq;
1813 }
1814
1815 connmgr_run(p->connmgr, handle_openflow);
1816
1817 return error;
1818 }
1819
1820 void
1821 ofproto_wait(struct ofproto *p)
1822 {
1823 p->ofproto_class->wait(p);
1824 if (p->ofproto_class->port_poll_wait) {
1825 p->ofproto_class->port_poll_wait(p);
1826 }
1827 seq_wait(connectivity_seq_get(), p->change_seq);
1828 connmgr_wait(p->connmgr);
1829 }
1830
1831 bool
1832 ofproto_is_alive(const struct ofproto *p)
1833 {
1834 return connmgr_has_controllers(p->connmgr);
1835 }
1836
1837 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1838 * memory_report(). */
1839 void
1840 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1841 {
1842 const struct oftable *table;
1843 unsigned int n_rules;
1844
1845 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1846
1847 n_rules = 0;
1848 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1849 n_rules += table->n_flows;
1850 }
1851 simap_increase(usage, "rules", n_rules);
1852
1853 if (ofproto->ofproto_class->get_memory_usage) {
1854 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1855 }
1856
1857 connmgr_get_memory_usage(ofproto->connmgr, usage);
1858 }
1859
1860 void
1861 ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
1862 {
1863 const struct ofproto_class *class;
1864
1865 datapath_type = ofproto_normalize_type(datapath_type);
1866 class = ofproto_class_find__(datapath_type);
1867
1868 if (class && class->type_get_memory_usage) {
1869 class->type_get_memory_usage(datapath_type, usage);
1870 }
1871 }
1872
1873 void
1874 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1875 struct shash *info)
1876 {
1877 connmgr_get_controller_info(ofproto->connmgr, info);
1878 }
1879
1880 void
1881 ofproto_free_ofproto_controller_info(struct shash *info)
1882 {
1883 connmgr_free_controller_info(info);
1884 }
1885
1886 /* Makes a deep copy of 'old' into 'port'. */
1887 void
1888 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1889 {
1890 port->name = xstrdup(old->name);
1891 port->type = xstrdup(old->type);
1892 port->ofp_port = old->ofp_port;
1893 }
1894
1895 /* Frees memory allocated to members of 'ofproto_port'.
1896 *
1897 * Do not call this function on an ofproto_port obtained from
1898 * ofproto_port_dump_next(): that function retains ownership of the data in the
1899 * ofproto_port. */
1900 void
1901 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1902 {
1903 free(ofproto_port->name);
1904 free(ofproto_port->type);
1905 }
1906
1907 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1908 *
1909 * This function provides no status indication. An error status for the entire
1910 * dump operation is provided when it is completed by calling
1911 * ofproto_port_dump_done().
1912 */
1913 void
1914 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1915 const struct ofproto *ofproto)
1916 {
1917 dump->ofproto = ofproto;
1918 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1919 &dump->state);
1920 }
1921
1922 /* Attempts to retrieve another port from 'dump', which must have been created
1923 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1924 * 'port' and returns true. On failure, returns false.
1925 *
1926 * Failure might indicate an actual error or merely that the last port has been
1927 * dumped. An error status for the entire dump operation is provided when it
1928 * is completed by calling ofproto_port_dump_done().
1929 *
1930 * The ofproto owns the data stored in 'port'. It will remain valid until at
1931 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1932 * ofproto_port_dump_done(). */
1933 bool
1934 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1935 struct ofproto_port *port)
1936 {
1937 const struct ofproto *ofproto = dump->ofproto;
1938
1939 if (dump->error) {
1940 return false;
1941 }
1942
1943 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1944 port);
1945 if (dump->error) {
1946 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1947 return false;
1948 }
1949 return true;
1950 }
1951
1952 /* Completes port table dump operation 'dump', which must have been created
1953 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1954 * error-free, otherwise a positive errno value describing the problem. */
1955 int
1956 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1957 {
1958 const struct ofproto *ofproto = dump->ofproto;
1959 if (!dump->error) {
1960 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1961 dump->state);
1962 }
1963 return dump->error == EOF ? 0 : dump->error;
1964 }
1965
1966 /* Returns the type to pass to netdev_open() when a datapath of type
1967 * 'datapath_type' has a port of type 'port_type', for a few special
1968 * cases when a netdev type differs from a port type. For example, when
1969 * using the userspace datapath, a port of type "internal" needs to be
1970 * opened as "tap".
1971 *
1972 * Returns either 'type' itself or a string literal, which must not be
1973 * freed. */
1974 const char *
1975 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1976 {
1977 const struct ofproto_class *class;
1978
1979 datapath_type = ofproto_normalize_type(datapath_type);
1980 class = ofproto_class_find__(datapath_type);
1981 if (!class) {
1982 return port_type;
1983 }
1984
1985 return (class->port_open_type
1986 ? class->port_open_type(datapath_type, port_type)
1987 : port_type);
1988 }
1989
1990 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1991 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1992 * the port's OpenFlow port number.
1993 *
1994 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1995 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1996 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1997 * 'ofp_portp' is non-null). */
1998 int
1999 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
2000 ofp_port_t *ofp_portp)
2001 {
2002 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
2003 int error;
2004
2005 error = ofproto->ofproto_class->port_add(ofproto, netdev);
2006 if (!error) {
2007 const char *netdev_name = netdev_get_name(netdev);
2008
2009 simap_put(&ofproto->ofp_requests, netdev_name,
2010 ofp_to_u16(ofp_port));
2011 error = update_port(ofproto, netdev_name);
2012 }
2013 if (ofp_portp) {
2014 *ofp_portp = OFPP_NONE;
2015 if (!error) {
2016 struct ofproto_port ofproto_port;
2017
2018 error = ofproto_port_query_by_name(ofproto,
2019 netdev_get_name(netdev),
2020 &ofproto_port);
2021 if (!error) {
2022 *ofp_portp = ofproto_port.ofp_port;
2023 ofproto_port_destroy(&ofproto_port);
2024 }
2025 }
2026 }
2027 return error;
2028 }
2029
2030 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
2031 * initializes '*port' appropriately; on failure, returns a positive errno
2032 * value.
2033 *
2034 * The caller owns the data in 'ofproto_port' and must free it with
2035 * ofproto_port_destroy() when it is no longer needed. */
2036 int
2037 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
2038 struct ofproto_port *port)
2039 {
2040 int error;
2041
2042 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
2043 if (error) {
2044 memset(port, 0, sizeof *port);
2045 }
2046 return error;
2047 }
2048
2049 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
2050 * Returns 0 if successful, otherwise a positive errno. */
2051 int
2052 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
2053 {
2054 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2055 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
2056 struct simap_node *ofp_request_node;
2057 int error;
2058
2059 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
2060 if (ofp_request_node) {
2061 simap_delete(&ofproto->ofp_requests, ofp_request_node);
2062 }
2063
2064 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
2065 if (!error && ofport) {
2066 /* 'name' is the netdev's name and update_port() is going to close the
2067 * netdev. Just in case update_port() refers to 'name' after it
2068 * destroys 'ofport', make a copy of it around the update_port()
2069 * call. */
2070 char *devname = xstrdup(name);
2071 update_port(ofproto, devname);
2072 free(devname);
2073 }
2074 return error;
2075 }
2076
2077 /* Refreshes datapath configuration of port number 'ofp_port' in 'ofproto'.
2078 *
2079 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
2080 void
2081 ofproto_port_set_config(struct ofproto *ofproto, ofp_port_t ofp_port,
2082 const struct smap *cfg)
2083 {
2084 struct ofport *ofport;
2085 int error;
2086
2087 ofport = ofproto_get_port(ofproto, ofp_port);
2088 if (!ofport) {
2089 VLOG_WARN("%s: cannot configure datapath on nonexistent port %"PRIu16,
2090 ofproto->name, ofp_port);
2091 return;
2092 }
2093
2094 error = (ofproto->ofproto_class->port_set_config
2095 ? ofproto->ofproto_class->port_set_config(ofport, cfg)
2096 : EOPNOTSUPP);
2097 if (error) {
2098 VLOG_WARN("%s: datapath configuration on port %"PRIu16
2099 " (%s) failed (%s)",
2100 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
2101 ovs_strerror(error));
2102 }
2103 }
2104
2105
2106 static void
2107 flow_mod_init(struct ofputil_flow_mod *fm,
2108 const struct match *match, int priority,
2109 const struct ofpact *ofpacts, size_t ofpacts_len,
2110 enum ofp_flow_mod_command command)
2111 {
2112 *fm = (struct ofputil_flow_mod) {
2113 .match = *match,
2114 .priority = priority,
2115 .table_id = 0,
2116 .command = command,
2117 .buffer_id = UINT32_MAX,
2118 .out_port = OFPP_ANY,
2119 .out_group = OFPG_ANY,
2120 .ofpacts = CONST_CAST(struct ofpact *, ofpacts),
2121 .ofpacts_len = ofpacts_len,
2122 .delete_reason = OFPRR_DELETE,
2123 };
2124 }
2125
2126 static int
2127 simple_flow_mod(struct ofproto *ofproto,
2128 const struct match *match, int priority,
2129 const struct ofpact *ofpacts, size_t ofpacts_len,
2130 enum ofp_flow_mod_command command)
2131 {
2132 struct ofproto_flow_mod ofm;
2133
2134 flow_mod_init(&ofm.fm, match, priority, ofpacts, ofpacts_len, command);
2135
2136 return handle_flow_mod__(ofproto, &ofm, NULL);
2137 }
2138
2139 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
2140 * performs the 'n_actions' actions in 'actions'. The new flow will not
2141 * timeout.
2142 *
2143 * If cls_rule->priority is in the range of priorities supported by OpenFlow
2144 * (0...65535, inclusive) then the flow will be visible to OpenFlow
2145 * controllers; otherwise, it will be hidden.
2146 *
2147 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
2148 *
2149 * This is a helper function for in-band control and fail-open. */
2150 void
2151 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
2152 int priority,
2153 const struct ofpact *ofpacts, size_t ofpacts_len)
2154 OVS_EXCLUDED(ofproto_mutex)
2155 {
2156 const struct rule *rule;
2157 bool must_add;
2158
2159 /* First do a cheap check whether the rule we're looking for already exists
2160 * with the actions that we want. If it does, then we're done. */
2161 rule = rule_from_cls_rule(classifier_find_match_exactly(
2162 &ofproto->tables[0].cls, match, priority,
2163 OVS_VERSION_MAX));
2164 if (rule) {
2165 const struct rule_actions *actions = rule_get_actions(rule);
2166 must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len,
2167 ofpacts, ofpacts_len);
2168 } else {
2169 must_add = true;
2170 }
2171
2172 /* If there's no such rule or the rule doesn't have the actions we want,
2173 * fall back to a executing a full flow mod. We can't optimize this at
2174 * all because we didn't take enough locks above to ensure that the flow
2175 * table didn't already change beneath us. */
2176 if (must_add) {
2177 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
2178 OFPFC_MODIFY_STRICT);
2179 }
2180 }
2181
2182 /* Executes the flow modification specified in 'fm'. Returns 0 on success, or
2183 * an OFPERR_* OpenFlow error code on failure.
2184 *
2185 * This is a helper function for in-band control and fail-open and the "learn"
2186 * action. */
2187 enum ofperr
2188 ofproto_flow_mod(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
2189 OVS_EXCLUDED(ofproto_mutex)
2190 {
2191 struct ofputil_flow_mod *fm = &ofm->fm;
2192
2193 /* Optimize for the most common case of a repeated learn action.
2194 * If an identical flow already exists we only need to update its
2195 * 'modified' time. */
2196 if (fm->command == OFPFC_MODIFY_STRICT && fm->table_id != OFPTT_ALL
2197 && !(fm->flags & OFPUTIL_FF_RESET_COUNTS)) {
2198 struct oftable *table = &ofproto->tables[fm->table_id];
2199 struct rule *rule;
2200 bool done = false;
2201
2202 rule = rule_from_cls_rule(classifier_find_match_exactly(
2203 &table->cls, &fm->match, fm->priority,
2204 OVS_VERSION_MAX));
2205 if (rule) {
2206 /* Reading many of the rule fields and writing on 'modified'
2207 * requires the rule->mutex. Also, rule->actions may change
2208 * if rule->mutex is not held. */
2209 const struct rule_actions *actions;
2210
2211 ovs_mutex_lock(&rule->mutex);
2212 actions = rule_get_actions(rule);
2213 if (rule->idle_timeout == fm->idle_timeout
2214 && rule->hard_timeout == fm->hard_timeout
2215 && rule->importance == fm->importance
2216 && rule->flags == (fm->flags & OFPUTIL_FF_STATE)
2217 && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie))
2218 && ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2219 actions->ofpacts, actions->ofpacts_len)) {
2220 /* Rule already exists and need not change, only update the
2221 modified timestamp. */
2222 rule->modified = time_msec();
2223 done = true;
2224 }
2225 ovs_mutex_unlock(&rule->mutex);
2226 }
2227
2228 if (done) {
2229 return 0;
2230 }
2231 }
2232
2233 return handle_flow_mod__(ofproto, ofm, NULL);
2234 }
2235
2236 /* Searches for a rule with matching criteria exactly equal to 'target' in
2237 * ofproto's table 0 and, if it finds one, deletes it.
2238 *
2239 * This is a helper function for in-band control and fail-open. */
2240 void
2241 ofproto_delete_flow(struct ofproto *ofproto,
2242 const struct match *target, int priority)
2243 OVS_EXCLUDED(ofproto_mutex)
2244 {
2245 struct classifier *cls = &ofproto->tables[0].cls;
2246 struct rule *rule;
2247
2248 /* First do a cheap check whether the rule we're looking for has already
2249 * been deleted. If so, then we're done. */
2250 rule = rule_from_cls_rule(classifier_find_match_exactly(
2251 cls, target, priority, OVS_VERSION_MAX));
2252 if (!rule) {
2253 return;
2254 }
2255
2256 /* Execute a flow mod. We can't optimize this at all because we didn't
2257 * take enough locks above to ensure that the flow table didn't already
2258 * change beneath us. */
2259 simple_flow_mod(ofproto, target, priority, NULL, 0, OFPFC_DELETE_STRICT);
2260 }
2261
2262 /* Delete all of the flows from all of ofproto's flow tables, then reintroduce
2263 * the flows required by in-band control and fail-open. */
2264 void
2265 ofproto_flush_flows(struct ofproto *ofproto)
2266 {
2267 COVERAGE_INC(ofproto_flush);
2268 ofproto_flush__(ofproto);
2269 connmgr_flushed(ofproto->connmgr);
2270 }
2271 \f
2272 static void
2273 reinit_ports(struct ofproto *p)
2274 {
2275 struct ofproto_port_dump dump;
2276 struct sset devnames;
2277 struct ofport *ofport;
2278 struct ofproto_port ofproto_port;
2279 const char *devname;
2280
2281 COVERAGE_INC(ofproto_reinit_ports);
2282
2283 sset_init(&devnames);
2284 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2285 sset_add(&devnames, netdev_get_name(ofport->netdev));
2286 }
2287 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2288 sset_add(&devnames, ofproto_port.name);
2289 }
2290
2291 SSET_FOR_EACH (devname, &devnames) {
2292 update_port(p, devname);
2293 }
2294 sset_destroy(&devnames);
2295 }
2296
2297 static ofp_port_t
2298 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
2299 {
2300 uint16_t port_idx;
2301
2302 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
2303 port_idx = port_idx ? port_idx : UINT16_MAX;
2304
2305 if (port_idx >= ofproto->max_ports
2306 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2307 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2308 long long int last_used_at, lru = LLONG_MAX;
2309
2310 /* Search for a free OpenFlow port number. We try not to
2311 * immediately reuse them to prevent problems due to old
2312 * flows.
2313 *
2314 * We limit the automatically assigned port numbers to the lower half
2315 * of the port range, to reserve the upper half for assignment by
2316 * controllers. */
2317 for (;;) {
2318 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
2319 ofproto->alloc_port_no = 1;
2320 }
2321 last_used_at = ofport_get_usage(ofproto,
2322 u16_to_ofp(ofproto->alloc_port_no));
2323 if (!last_used_at) {
2324 port_idx = ofproto->alloc_port_no;
2325 break;
2326 } else if ( last_used_at < time_msec() - 60*60*1000) {
2327 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2328 * more than an hour ago, consider it usable. */
2329 ofport_remove_usage(ofproto,
2330 u16_to_ofp(ofproto->alloc_port_no));
2331 port_idx = ofproto->alloc_port_no;
2332 break;
2333 } else if (last_used_at < lru) {
2334 lru = last_used_at;
2335 lru_ofport = ofproto->alloc_port_no;
2336 }
2337
2338 if (ofproto->alloc_port_no == end_port_no) {
2339 if (lru_ofport) {
2340 port_idx = lru_ofport;
2341 break;
2342 }
2343 return OFPP_NONE;
2344 }
2345 }
2346 }
2347 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2348 return u16_to_ofp(port_idx);
2349 }
2350
2351 static void
2352 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2353 {
2354 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2355 ofport_set_usage(ofproto, ofp_port, time_msec());
2356 }
2357 }
2358
2359 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2360 * pointer if the netdev cannot be opened. On success, also fills in
2361 * '*pp'. */
2362 static struct netdev *
2363 ofport_open(struct ofproto *ofproto,
2364 struct ofproto_port *ofproto_port,
2365 struct ofputil_phy_port *pp)
2366 {
2367 enum netdev_flags flags;
2368 struct netdev *netdev;
2369 int error;
2370
2371 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2372 if (error) {
2373 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2374 "cannot be opened (%s)",
2375 ofproto->name,
2376 ofproto_port->name, ofproto_port->ofp_port,
2377 ofproto_port->name, ovs_strerror(error));
2378 return NULL;
2379 }
2380
2381 if (ofproto_port->ofp_port == OFPP_NONE) {
2382 if (!strcmp(ofproto->name, ofproto_port->name)) {
2383 ofproto_port->ofp_port = OFPP_LOCAL;
2384 } else {
2385 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2386 ofproto_port->name);
2387 }
2388 }
2389 pp->port_no = ofproto_port->ofp_port;
2390 netdev_get_etheraddr(netdev, &pp->hw_addr);
2391 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2392 netdev_get_flags(netdev, &flags);
2393 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2394 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2395 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2396 &pp->supported, &pp->peer);
2397 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2398 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2399
2400 return netdev;
2401 }
2402
2403 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2404 * port number, and 'config' bits other than OFPUTIL_PC_PORT_DOWN are
2405 * disregarded. */
2406 static bool
2407 ofport_equal(const struct ofputil_phy_port *a,
2408 const struct ofputil_phy_port *b)
2409 {
2410 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2411 && a->state == b->state
2412 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2413 && a->curr == b->curr
2414 && a->advertised == b->advertised
2415 && a->supported == b->supported
2416 && a->peer == b->peer
2417 && a->curr_speed == b->curr_speed
2418 && a->max_speed == b->max_speed);
2419 }
2420
2421 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2422 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2423 * one with the same name or port number). */
2424 static int
2425 ofport_install(struct ofproto *p,
2426 struct netdev *netdev, const struct ofputil_phy_port *pp)
2427 {
2428 const char *netdev_name = netdev_get_name(netdev);
2429 struct ofport *ofport;
2430 int error;
2431
2432 /* Create ofport. */
2433 ofport = p->ofproto_class->port_alloc();
2434 if (!ofport) {
2435 error = ENOMEM;
2436 goto error;
2437 }
2438 ofport->ofproto = p;
2439 ofport->netdev = netdev;
2440 ofport->change_seq = netdev_get_change_seq(netdev);
2441 ofport->pp = *pp;
2442 ofport->ofp_port = pp->port_no;
2443 ofport->created = time_msec();
2444
2445 /* Add port to 'p'. */
2446 hmap_insert(&p->ports, &ofport->hmap_node,
2447 hash_ofp_port(ofport->ofp_port));
2448 shash_add(&p->port_by_name, netdev_name, ofport);
2449
2450 update_mtu(p, ofport);
2451
2452 /* Let the ofproto_class initialize its private data. */
2453 error = p->ofproto_class->port_construct(ofport);
2454 if (error) {
2455 goto error;
2456 }
2457 connmgr_send_port_status(p->connmgr, NULL, pp, OFPPR_ADD);
2458 return 0;
2459
2460 error:
2461 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2462 p->name, netdev_name, ovs_strerror(error));
2463 if (ofport) {
2464 ofport_destroy__(ofport);
2465 } else {
2466 netdev_close(netdev);
2467 }
2468 return error;
2469 }
2470
2471 /* Removes 'ofport' from 'p' and destroys it. */
2472 static void
2473 ofport_remove(struct ofport *ofport)
2474 {
2475 struct ofproto *p = ofport->ofproto;
2476 bool is_internal = ofport_is_internal(ofport);
2477
2478 connmgr_send_port_status(ofport->ofproto->connmgr, NULL, &ofport->pp,
2479 OFPPR_DELETE);
2480 ofport_destroy(ofport, true);
2481 if (!is_internal) {
2482 update_mtu_ofproto(p);
2483 }
2484 }
2485
2486 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2487 * destroys it. */
2488 static void
2489 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2490 {
2491 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2492 if (port) {
2493 ofport_remove(port);
2494 }
2495 }
2496
2497 /* Updates 'port' with new 'pp' description.
2498 *
2499 * Does not handle a name or port number change. The caller must implement
2500 * such a change as a delete followed by an add. */
2501 static void
2502 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2503 {
2504 port->pp.hw_addr = pp->hw_addr;
2505 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2506 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2507 port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN)
2508 | (pp->state & OFPUTIL_PS_LINK_DOWN));
2509 port->pp.curr = pp->curr;
2510 port->pp.advertised = pp->advertised;
2511 port->pp.supported = pp->supported;
2512 port->pp.peer = pp->peer;
2513 port->pp.curr_speed = pp->curr_speed;
2514 port->pp.max_speed = pp->max_speed;
2515
2516 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2517 &port->pp, OFPPR_MODIFY);
2518 }
2519
2520 /* Update OpenFlow 'state' in 'port' and notify controller. */
2521 void
2522 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2523 {
2524 if (port->pp.state != state) {
2525 port->pp.state = state;
2526 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2527 &port->pp, OFPPR_MODIFY);
2528 }
2529 }
2530
2531 void
2532 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2533 {
2534 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2535 if (port) {
2536 if (port->ofproto->ofproto_class->set_stp_port) {
2537 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2538 }
2539 if (port->ofproto->ofproto_class->set_rstp_port) {
2540 port->ofproto->ofproto_class->set_rstp_port(port, NULL);
2541 }
2542 if (port->ofproto->ofproto_class->set_cfm) {
2543 port->ofproto->ofproto_class->set_cfm(port, NULL);
2544 }
2545 if (port->ofproto->ofproto_class->bundle_remove) {
2546 port->ofproto->ofproto_class->bundle_remove(port);
2547 }
2548 }
2549 }
2550
2551 static void
2552 ofport_destroy__(struct ofport *port)
2553 {
2554 struct ofproto *ofproto = port->ofproto;
2555 const char *name = netdev_get_name(port->netdev);
2556
2557 hmap_remove(&ofproto->ports, &port->hmap_node);
2558 shash_delete(&ofproto->port_by_name,
2559 shash_find(&ofproto->port_by_name, name));
2560
2561 netdev_close(port->netdev);
2562 ofproto->ofproto_class->port_dealloc(port);
2563 }
2564
2565 static void
2566 ofport_destroy(struct ofport *port, bool del)
2567 {
2568 if (port) {
2569 dealloc_ofp_port(port->ofproto, port->ofp_port);
2570 port->ofproto->ofproto_class->port_destruct(port, del);
2571 ofport_destroy__(port);
2572 }
2573 }
2574
2575 struct ofport *
2576 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2577 {
2578 struct ofport *port;
2579
2580 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2581 &ofproto->ports) {
2582 if (port->ofp_port == ofp_port) {
2583 return port;
2584 }
2585 }
2586 return NULL;
2587 }
2588
2589 static long long int
2590 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2591 {
2592 struct ofport_usage *usage;
2593
2594 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2595 &ofproto->ofport_usage) {
2596 if (usage->ofp_port == ofp_port) {
2597 return usage->last_used;
2598 }
2599 }
2600 return 0;
2601 }
2602
2603 static void
2604 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2605 long long int last_used)
2606 {
2607 struct ofport_usage *usage;
2608 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2609 &ofproto->ofport_usage) {
2610 if (usage->ofp_port == ofp_port) {
2611 usage->last_used = last_used;
2612 return;
2613 }
2614 }
2615 ovs_assert(last_used == LLONG_MAX);
2616
2617 usage = xmalloc(sizeof *usage);
2618 usage->ofp_port = ofp_port;
2619 usage->last_used = last_used;
2620 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2621 hash_ofp_port(ofp_port));
2622 }
2623
2624 static void
2625 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2626 {
2627 struct ofport_usage *usage;
2628 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2629 &ofproto->ofport_usage) {
2630 if (usage->ofp_port == ofp_port) {
2631 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2632 free(usage);
2633 break;
2634 }
2635 }
2636 }
2637
2638 int
2639 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2640 {
2641 struct ofproto *ofproto = port->ofproto;
2642 int error;
2643
2644 if (ofproto->ofproto_class->port_get_stats) {
2645 error = ofproto->ofproto_class->port_get_stats(port, stats);
2646 } else {
2647 error = EOPNOTSUPP;
2648 }
2649
2650 return error;
2651 }
2652
2653 static int
2654 update_port(struct ofproto *ofproto, const char *name)
2655 {
2656 struct ofproto_port ofproto_port;
2657 struct ofputil_phy_port pp;
2658 struct netdev *netdev;
2659 struct ofport *port;
2660 int error = 0;
2661
2662 COVERAGE_INC(ofproto_update_port);
2663
2664 /* Fetch 'name''s location and properties from the datapath. */
2665 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2666 ? ofport_open(ofproto, &ofproto_port, &pp)
2667 : NULL);
2668
2669 if (netdev) {
2670 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2671 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2672 struct netdev *old_netdev = port->netdev;
2673
2674 /* 'name' hasn't changed location. Any properties changed? */
2675 if (!ofport_equal(&port->pp, &pp)) {
2676 ofport_modified(port, &pp);
2677 }
2678
2679 update_mtu(ofproto, port);
2680
2681 /* Install the newly opened netdev in case it has changed.
2682 * Don't close the old netdev yet in case port_modified has to
2683 * remove a retained reference to it.*/
2684 port->netdev = netdev;
2685 port->change_seq = netdev_get_change_seq(netdev);
2686
2687 if (port->ofproto->ofproto_class->port_modified) {
2688 port->ofproto->ofproto_class->port_modified(port);
2689 }
2690
2691 netdev_close(old_netdev);
2692 } else {
2693 /* If 'port' is nonnull then its name differs from 'name' and thus
2694 * we should delete it. If we think there's a port named 'name'
2695 * then its port number must be wrong now so delete it too. */
2696 if (port) {
2697 ofport_remove(port);
2698 }
2699 ofport_remove_with_name(ofproto, name);
2700 error = ofport_install(ofproto, netdev, &pp);
2701 }
2702 } else {
2703 /* Any port named 'name' is gone now. */
2704 ofport_remove_with_name(ofproto, name);
2705 }
2706 ofproto_port_destroy(&ofproto_port);
2707
2708 return error;
2709 }
2710
2711 static int
2712 init_ports(struct ofproto *p)
2713 {
2714 struct ofproto_port_dump dump;
2715 struct ofproto_port ofproto_port;
2716 struct shash_node *node, *next;
2717
2718 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2719 const char *name = ofproto_port.name;
2720
2721 if (shash_find(&p->port_by_name, name)) {
2722 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2723 p->name, name);
2724 } else {
2725 struct ofputil_phy_port pp;
2726 struct netdev *netdev;
2727
2728 /* Check if an OpenFlow port number had been requested. */
2729 node = shash_find(&init_ofp_ports, name);
2730 if (node) {
2731 const struct iface_hint *iface_hint = node->data;
2732 simap_put(&p->ofp_requests, name,
2733 ofp_to_u16(iface_hint->ofp_port));
2734 }
2735
2736 netdev = ofport_open(p, &ofproto_port, &pp);
2737 if (netdev) {
2738 ofport_install(p, netdev, &pp);
2739 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2740 p->alloc_port_no = MAX(p->alloc_port_no,
2741 ofp_to_u16(ofproto_port.ofp_port));
2742 }
2743 }
2744 }
2745 }
2746
2747 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2748 struct iface_hint *iface_hint = node->data;
2749
2750 if (!strcmp(iface_hint->br_name, p->name)) {
2751 free(iface_hint->br_name);
2752 free(iface_hint->br_type);
2753 free(iface_hint);
2754 shash_delete(&init_ofp_ports, node);
2755 }
2756 }
2757
2758 return 0;
2759 }
2760
2761 static inline bool
2762 ofport_is_internal(const struct ofport *port)
2763 {
2764 return !strcmp(netdev_get_type(port->netdev), "internal");
2765 }
2766
2767 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2768 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2769 static int
2770 find_min_mtu(struct ofproto *p)
2771 {
2772 struct ofport *ofport;
2773 int mtu = 0;
2774
2775 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2776 struct netdev *netdev = ofport->netdev;
2777 int dev_mtu;
2778
2779 /* Skip any internal ports, since that's what we're trying to
2780 * set. */
2781 if (ofport_is_internal(ofport)) {
2782 continue;
2783 }
2784
2785 if (netdev_get_mtu(netdev, &dev_mtu)) {
2786 continue;
2787 }
2788 if (!mtu || dev_mtu < mtu) {
2789 mtu = dev_mtu;
2790 }
2791 }
2792
2793 return mtu ? mtu: ETH_PAYLOAD_MAX;
2794 }
2795
2796 /* Update MTU of all datapath devices on 'p' to the minimum of the
2797 * non-datapath ports in event of 'port' added or changed. */
2798 static void
2799 update_mtu(struct ofproto *p, struct ofport *port)
2800 {
2801 struct netdev *netdev = port->netdev;
2802 int dev_mtu;
2803
2804 if (netdev_get_mtu(netdev, &dev_mtu)) {
2805 port->mtu = 0;
2806 return;
2807 }
2808 if (ofport_is_internal(port)) {
2809 if (dev_mtu > p->min_mtu) {
2810 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2811 dev_mtu = p->min_mtu;
2812 }
2813 }
2814 port->mtu = dev_mtu;
2815 return;
2816 }
2817
2818 port->mtu = dev_mtu;
2819 /* For non-internal port find new min mtu. */
2820 update_mtu_ofproto(p);
2821 }
2822
2823 static void
2824 update_mtu_ofproto(struct ofproto *p)
2825 {
2826 /* For non-internal port find new min mtu. */
2827 struct ofport *ofport;
2828 int old_min = p->min_mtu;
2829
2830 p->min_mtu = find_min_mtu(p);
2831 if (p->min_mtu == old_min) {
2832 return;
2833 }
2834
2835 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2836 struct netdev *netdev = ofport->netdev;
2837
2838 if (ofport_is_internal(ofport)) {
2839 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2840 ofport->mtu = p->min_mtu;
2841 }
2842 }
2843 }
2844 }
2845 \f
2846 static void
2847 ofproto_rule_destroy__(struct rule *rule)
2848 OVS_NO_THREAD_SAFETY_ANALYSIS
2849 {
2850 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2851 rule_actions_destroy(rule_get_actions(rule));
2852 ovs_mutex_destroy(&rule->mutex);
2853 rule->ofproto->ofproto_class->rule_dealloc(rule);
2854 }
2855
2856 static void
2857 rule_destroy_cb(struct rule *rule)
2858 OVS_NO_THREAD_SAFETY_ANALYSIS
2859 {
2860 /* Send rule removed if needed. */
2861 if (rule->flags & OFPUTIL_FF_SEND_FLOW_REM
2862 && rule->removed_reason != OVS_OFPRR_NONE
2863 && !rule_is_hidden(rule)) {
2864 ofproto_rule_send_removed(rule);
2865 }
2866 rule->ofproto->ofproto_class->rule_destruct(rule);
2867 ofproto_rule_destroy__(rule);
2868 }
2869
2870 void
2871 ofproto_rule_ref(struct rule *rule)
2872 {
2873 if (rule) {
2874 ovs_refcount_ref(&rule->ref_count);
2875 }
2876 }
2877
2878 bool
2879 ofproto_rule_try_ref(struct rule *rule)
2880 {
2881 if (rule) {
2882 return ovs_refcount_try_ref_rcu(&rule->ref_count);
2883 }
2884 return false;
2885 }
2886
2887 /* Decrements 'rule''s ref_count and schedules 'rule' to be destroyed if the
2888 * ref_count reaches 0.
2889 *
2890 * Use of RCU allows short term use (between RCU quiescent periods) without
2891 * keeping a reference. A reference must be taken if the rule needs to
2892 * stay around accross the RCU quiescent periods. */
2893 void
2894 ofproto_rule_unref(struct rule *rule)
2895 {
2896 if (rule && ovs_refcount_unref_relaxed(&rule->ref_count) == 1) {
2897 ovsrcu_postpone(rule_destroy_cb, rule);
2898 }
2899 }
2900
2901 static void
2902 remove_rule_rcu__(struct rule *rule)
2903 OVS_REQUIRES(ofproto_mutex)
2904 {
2905 struct ofproto *ofproto = rule->ofproto;
2906 struct oftable *table = &ofproto->tables[rule->table_id];
2907
2908 ovs_assert(!cls_rule_visible_in_version(&rule->cr, OVS_VERSION_MAX));
2909 if (!classifier_remove(&table->cls, &rule->cr)) {
2910 OVS_NOT_REACHED();
2911 }
2912 if (ofproto->ofproto_class->rule_delete) {
2913 ofproto->ofproto_class->rule_delete(rule);
2914 }
2915 ofproto_rule_unref(rule);
2916 }
2917
2918 static void
2919 remove_rule_rcu(struct rule *rule)
2920 OVS_EXCLUDED(ofproto_mutex)
2921 {
2922 ovs_mutex_lock(&ofproto_mutex);
2923 remove_rule_rcu__(rule);
2924 ovs_mutex_unlock(&ofproto_mutex);
2925 }
2926
2927 /* Removes and deletes rules from a NULL-terminated array of rule pointers. */
2928 static void
2929 remove_rules_rcu(struct rule **rules)
2930 OVS_EXCLUDED(ofproto_mutex)
2931 {
2932 struct rule **orig_rules = rules;
2933
2934 if (*rules) {
2935 struct ofproto *ofproto = rules[0]->ofproto;
2936 unsigned long tables[BITMAP_N_LONGS(256)];
2937 struct rule *rule;
2938 size_t table_id;
2939
2940 memset(tables, 0, sizeof tables);
2941
2942 ovs_mutex_lock(&ofproto_mutex);
2943 while ((rule = *rules++)) {
2944 /* Defer once for each new table. This defers the subtable cleanup
2945 * until later, so that when removing large number of flows the
2946 * operation is faster. */
2947 if (!bitmap_is_set(tables, rule->table_id)) {
2948 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
2949
2950 bitmap_set1(tables, rule->table_id);
2951 classifier_defer(cls);
2952 }
2953 remove_rule_rcu__(rule);
2954 }
2955
2956 BITMAP_FOR_EACH_1(table_id, 256, tables) {
2957 struct classifier *cls = &ofproto->tables[table_id].cls;
2958
2959 classifier_publish(cls);
2960 }
2961 ovs_mutex_unlock(&ofproto_mutex);
2962 }
2963
2964 free(orig_rules);
2965 }
2966
2967 void
2968 ofproto_group_ref(struct ofgroup *group)
2969 {
2970 if (group) {
2971 ovs_refcount_ref(&group->ref_count);
2972 }
2973 }
2974
2975 bool
2976 ofproto_group_try_ref(struct ofgroup *group)
2977 {
2978 if (group) {
2979 return ovs_refcount_try_ref_rcu(&group->ref_count);
2980 }
2981 return false;
2982 }
2983
2984 static void
2985 group_destroy_cb(struct ofgroup *group)
2986 {
2987 group->ofproto->ofproto_class->group_destruct(group);
2988 ofputil_bucket_list_destroy(CONST_CAST(struct ovs_list *,
2989 &group->buckets));
2990 group->ofproto->ofproto_class->group_dealloc(group);
2991 }
2992
2993 void
2994 ofproto_group_unref(struct ofgroup *group)
2995 {
2996 if (group && ovs_refcount_unref_relaxed(&group->ref_count) == 1) {
2997 ovsrcu_postpone(group_destroy_cb, group);
2998 }
2999 }
3000
3001 static uint32_t get_provider_meter_id(const struct ofproto *,
3002 uint32_t of_meter_id);
3003
3004 /* Creates and returns a new 'struct rule_actions', whose actions are a copy
3005 * of from the 'ofpacts_len' bytes of 'ofpacts'. */
3006 const struct rule_actions *
3007 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
3008 {
3009 struct rule_actions *actions;
3010
3011 actions = xmalloc(sizeof *actions + ofpacts_len);
3012 actions->ofpacts_len = ofpacts_len;
3013 actions->has_meter = ofpacts_get_meter(ofpacts, ofpacts_len) != 0;
3014 memcpy(actions->ofpacts, ofpacts, ofpacts_len);
3015
3016 actions->has_learn_with_delete = (next_learn_with_delete(actions, NULL)
3017 != NULL);
3018
3019 return actions;
3020 }
3021
3022 /* Free the actions after the RCU quiescent period is reached. */
3023 void
3024 rule_actions_destroy(const struct rule_actions *actions)
3025 {
3026 if (actions) {
3027 ovsrcu_postpone(free, CONST_CAST(struct rule_actions *, actions));
3028 }
3029 }
3030
3031 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
3032 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
3033 bool
3034 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
3035 OVS_REQUIRES(ofproto_mutex)
3036 {
3037 if (port == OFPP_ANY) {
3038 return true;
3039 } else {
3040 const struct rule_actions *actions = rule_get_actions(rule);
3041 return ofpacts_output_to_port(actions->ofpacts,
3042 actions->ofpacts_len, port);
3043 }
3044 }
3045
3046 /* Returns true if 'rule' has group and equals group_id. */
3047 static bool
3048 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
3049 OVS_REQUIRES(ofproto_mutex)
3050 {
3051 if (group_id == OFPG_ANY) {
3052 return true;
3053 } else {
3054 const struct rule_actions *actions = rule_get_actions(rule);
3055 return ofpacts_output_to_group(actions->ofpacts,
3056 actions->ofpacts_len, group_id);
3057 }
3058 }
3059
3060 static void
3061 rule_execute_destroy(struct rule_execute *e)
3062 {
3063 ofproto_rule_unref(e->rule);
3064 ovs_list_remove(&e->list_node);
3065 free(e);
3066 }
3067
3068 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
3069 * by passing them to the ofproto provider. */
3070 static void
3071 run_rule_executes(struct ofproto *ofproto)
3072 OVS_EXCLUDED(ofproto_mutex)
3073 {
3074 struct rule_execute *e, *next;
3075 struct ovs_list executes;
3076
3077 guarded_list_pop_all(&ofproto->rule_executes, &executes);
3078 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
3079 struct flow flow;
3080
3081 flow_extract(e->packet, &flow);
3082 flow.in_port.ofp_port = e->in_port;
3083 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
3084
3085 rule_execute_destroy(e);
3086 }
3087 }
3088
3089 /* Destroys and discards all "rule_execute" operations queued up in
3090 * ofproto->rule_executes. */
3091 static void
3092 destroy_rule_executes(struct ofproto *ofproto)
3093 {
3094 struct rule_execute *e, *next;
3095 struct ovs_list executes;
3096
3097 guarded_list_pop_all(&ofproto->rule_executes, &executes);
3098 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
3099 dp_packet_delete(e->packet);
3100 rule_execute_destroy(e);
3101 }
3102 }
3103
3104 static bool
3105 rule_is_readonly(const struct rule *rule)
3106 {
3107 const struct oftable *table = &rule->ofproto->tables[rule->table_id];
3108 return (table->flags & OFTABLE_READONLY) != 0;
3109 }
3110 \f
3111 static uint32_t
3112 hash_learned_cookie(ovs_be64 cookie_, uint8_t table_id)
3113 {
3114 uint64_t cookie = (OVS_FORCE uint64_t) cookie_;
3115 return hash_3words(cookie, cookie >> 32, table_id);
3116 }
3117
3118 static void
3119 learned_cookies_update_one__(struct ofproto *ofproto,
3120 const struct ofpact_learn *learn,
3121 int delta, struct ovs_list *dead_cookies)
3122 OVS_REQUIRES(ofproto_mutex)
3123 {
3124 uint32_t hash = hash_learned_cookie(learn->cookie, learn->table_id);
3125 struct learned_cookie *c;
3126
3127 HMAP_FOR_EACH_WITH_HASH (c, u.hmap_node, hash, &ofproto->learned_cookies) {
3128 if (c->cookie == learn->cookie && c->table_id == learn->table_id) {
3129 c->n += delta;
3130 ovs_assert(c->n >= 0);
3131
3132 if (!c->n) {
3133 hmap_remove(&ofproto->learned_cookies, &c->u.hmap_node);
3134 ovs_list_push_back(dead_cookies, &c->u.list_node);
3135 }
3136
3137 return;
3138 }
3139 }
3140
3141 ovs_assert(delta > 0);
3142 c = xmalloc(sizeof *c);
3143 hmap_insert(&ofproto->learned_cookies, &c->u.hmap_node, hash);
3144 c->cookie = learn->cookie;
3145 c->table_id = learn->table_id;
3146 c->n = delta;
3147 }
3148
3149 static const struct ofpact_learn *
3150 next_learn_with_delete(const struct rule_actions *actions,
3151 const struct ofpact_learn *start)
3152 {
3153 const struct ofpact *pos;
3154
3155 for (pos = start ? ofpact_next(&start->ofpact) : actions->ofpacts;
3156 pos < ofpact_end(actions->ofpacts, actions->ofpacts_len);
3157 pos = ofpact_next(pos)) {
3158 if (pos->type == OFPACT_LEARN) {
3159 const struct ofpact_learn *learn = ofpact_get_LEARN(pos);
3160 if (learn->flags & NX_LEARN_F_DELETE_LEARNED) {
3161 return learn;
3162 }
3163 }
3164 }
3165
3166 return NULL;
3167 }
3168
3169 static void
3170 learned_cookies_update__(struct ofproto *ofproto,
3171 const struct rule_actions *actions,
3172 int delta, struct ovs_list *dead_cookies)
3173 OVS_REQUIRES(ofproto_mutex)
3174 {
3175 if (actions->has_learn_with_delete) {
3176 const struct ofpact_learn *learn;
3177
3178 for (learn = next_learn_with_delete(actions, NULL); learn;
3179 learn = next_learn_with_delete(actions, learn)) {
3180 learned_cookies_update_one__(ofproto, learn, delta, dead_cookies);
3181 }
3182 }
3183 }
3184
3185 static void
3186 learned_cookies_inc(struct ofproto *ofproto,
3187 const struct rule_actions *actions)
3188 OVS_REQUIRES(ofproto_mutex)
3189 {
3190 learned_cookies_update__(ofproto, actions, +1, NULL);
3191 }
3192
3193 static void
3194 learned_cookies_dec(struct ofproto *ofproto,
3195 const struct rule_actions *actions,
3196 struct ovs_list *dead_cookies)
3197 OVS_REQUIRES(ofproto_mutex)
3198 {
3199 learned_cookies_update__(ofproto, actions, -1, dead_cookies);
3200 }
3201
3202 static void
3203 learned_cookies_flush(struct ofproto *ofproto, struct ovs_list *dead_cookies)
3204 OVS_REQUIRES(ofproto_mutex)
3205 {
3206 struct learned_cookie *c;
3207
3208 LIST_FOR_EACH_POP (c, u.list_node, dead_cookies) {
3209 struct rule_criteria criteria;
3210 struct rule_collection rules;
3211 struct match match;
3212
3213 match_init_catchall(&match);
3214 rule_criteria_init(&criteria, c->table_id, &match, 0, OVS_VERSION_MAX,
3215 c->cookie, OVS_BE64_MAX, OFPP_ANY, OFPG_ANY);
3216 rule_criteria_require_rw(&criteria, false);
3217 collect_rules_loose(ofproto, &criteria, &rules);
3218 rule_criteria_destroy(&criteria);
3219 delete_flows__(&rules, OFPRR_DELETE, NULL);
3220
3221 free(c);
3222 }
3223 }
3224 \f
3225 static enum ofperr
3226 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
3227 {
3228 ofconn_send_reply(ofconn, make_echo_reply(oh));
3229 return 0;
3230 }
3231
3232 static void
3233 query_tables(struct ofproto *ofproto,
3234 struct ofputil_table_features **featuresp,
3235 struct ofputil_table_stats **statsp)
3236 {
3237 struct mf_bitmap rw_fields = oxm_writable_fields();
3238 struct mf_bitmap match = oxm_matchable_fields();
3239 struct mf_bitmap mask = oxm_maskable_fields();
3240
3241 struct ofputil_table_features *features;
3242 struct ofputil_table_stats *stats;
3243 int i;
3244
3245 features = *featuresp = xcalloc(ofproto->n_tables, sizeof *features);
3246 for (i = 0; i < ofproto->n_tables; i++) {
3247 struct ofputil_table_features *f = &features[i];
3248
3249 f->table_id = i;
3250 sprintf(f->name, "table%d", i);
3251 f->metadata_match = OVS_BE64_MAX;
3252 f->metadata_write = OVS_BE64_MAX;
3253 atomic_read_relaxed(&ofproto->tables[i].miss_config, &f->miss_config);
3254 f->max_entries = 1000000;
3255
3256 bool more_tables = false;
3257 for (int j = i + 1; j < ofproto->n_tables; j++) {
3258 if (!(ofproto->tables[j].flags & OFTABLE_HIDDEN)) {
3259 bitmap_set1(f->nonmiss.next, j);
3260 more_tables = true;
3261 }
3262 }
3263 f->nonmiss.instructions = (1u << N_OVS_INSTRUCTIONS) - 1;
3264 if (!more_tables) {
3265 f->nonmiss.instructions &= ~(1u << OVSINST_OFPIT11_GOTO_TABLE);
3266 }
3267 f->nonmiss.write.ofpacts = (UINT64_C(1) << N_OFPACTS) - 1;
3268 f->nonmiss.write.set_fields = rw_fields;
3269 f->nonmiss.apply = f->nonmiss.write;
3270 f->miss = f->nonmiss;
3271
3272 f->match = match;
3273 f->mask = mask;
3274 f->wildcard = match;
3275 }
3276
3277 if (statsp) {
3278 stats = *statsp = xcalloc(ofproto->n_tables, sizeof *stats);
3279 for (i = 0; i < ofproto->n_tables; i++) {
3280 struct ofputil_table_stats *s = &stats[i];
3281
3282 s->table_id = i;
3283 s->active_count = ofproto->tables[i].n_flows;
3284 if (i == 0) {
3285 s->active_count -= connmgr_count_hidden_rules(
3286 ofproto->connmgr);
3287 }
3288 }
3289 } else {
3290 stats = NULL;
3291 }
3292
3293 ofproto->ofproto_class->query_tables(ofproto, features, stats);
3294
3295 for (i = 0; i < ofproto->n_tables; i++) {
3296 const struct oftable *table = &ofproto->tables[i];
3297 struct ofputil_table_features *f = &features[i];
3298
3299 if (table->name) {
3300 ovs_strzcpy(f->name, table->name, sizeof f->name);
3301 }
3302
3303 if (table->max_flows < f->max_entries) {
3304 f->max_entries = table->max_flows;
3305 }
3306 }
3307 }
3308
3309 static void
3310 query_switch_features(struct ofproto *ofproto,
3311 bool *arp_match_ip, uint64_t *ofpacts)
3312 {
3313 struct ofputil_table_features *features, *f;
3314
3315 *arp_match_ip = false;
3316 *ofpacts = 0;
3317
3318 query_tables(ofproto, &features, NULL);
3319 for (f = features; f < &features[ofproto->n_tables]; f++) {
3320 *ofpacts |= f->nonmiss.apply.ofpacts | f->miss.apply.ofpacts;
3321 if (bitmap_is_set(f->match.bm, MFF_ARP_SPA) ||
3322 bitmap_is_set(f->match.bm, MFF_ARP_TPA)) {
3323 *arp_match_ip = true;
3324 }
3325 }
3326 free(features);
3327
3328 /* Sanity check. */
3329 ovs_assert(*ofpacts & (UINT64_C(1) << OFPACT_OUTPUT));
3330 }
3331
3332 static enum ofperr
3333 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
3334 {
3335 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3336 struct ofputil_switch_features features;
3337 struct ofport *port;
3338 bool arp_match_ip;
3339 struct ofpbuf *b;
3340
3341 query_switch_features(ofproto, &arp_match_ip, &features.ofpacts);
3342
3343 features.datapath_id = ofproto->datapath_id;
3344 features.n_buffers = pktbuf_capacity();
3345 features.n_tables = ofproto_get_n_visible_tables(ofproto);
3346 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
3347 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS |
3348 OFPUTIL_C_GROUP_STATS);
3349 if (arp_match_ip) {
3350 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
3351 }
3352 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
3353 features.auxiliary_id = 0;
3354 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
3355 oh->xid);
3356 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3357 ofputil_put_switch_features_port(&port->pp, b);
3358 }
3359
3360 ofconn_send_reply(ofconn, b);
3361 return 0;
3362 }
3363
3364 static enum ofperr
3365 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
3366 {
3367 struct ofputil_switch_config config;
3368 config.frag = ofconn_get_ofproto(ofconn)->frag_handling;
3369 config.invalid_ttl_to_controller
3370 = ofconn_get_invalid_ttl_to_controller(ofconn);
3371 config.miss_send_len = ofconn_get_miss_send_len(ofconn);
3372
3373 ofconn_send_reply(ofconn, ofputil_encode_get_config_reply(oh, &config));
3374
3375 return 0;
3376 }
3377
3378 static enum ofperr
3379 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
3380 {
3381 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3382 struct ofputil_switch_config config;
3383 enum ofperr error;
3384
3385 error = ofputil_decode_set_config(oh, &config);
3386 if (error) {
3387 return error;
3388 }
3389
3390 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
3391 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
3392 enum ofputil_frag_handling cur = ofproto->frag_handling;
3393 enum ofputil_frag_handling next = config.frag;
3394
3395 if (cur != next) {
3396 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
3397 ofproto->frag_handling = next;
3398 } else {
3399 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
3400 ofproto->name,
3401 ofputil_frag_handling_to_string(next));
3402 }
3403 }
3404 }
3405
3406 if (config.invalid_ttl_to_controller >= 0) {
3407 ofconn_set_invalid_ttl_to_controller(ofconn,
3408 config.invalid_ttl_to_controller);
3409 }
3410
3411 ofconn_set_miss_send_len(ofconn, config.miss_send_len);
3412
3413 return 0;
3414 }
3415
3416 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3417 * error message code for the caller to propagate upward. Otherwise, returns
3418 * 0.
3419 *
3420 * The log message mentions 'msg_type'. */
3421 static enum ofperr
3422 reject_slave_controller(struct ofconn *ofconn)
3423 {
3424 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
3425 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
3426 return OFPERR_OFPBRC_IS_SLAVE;
3427 } else {
3428 return 0;
3429 }
3430 }
3431
3432 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
3433 * for 'ofproto':
3434 *
3435 * - If they use a meter, then 'ofproto' has that meter configured.
3436 *
3437 * - If they use any groups, then 'ofproto' has that group configured.
3438 *
3439 * Returns 0 if successful, otherwise an OpenFlow error. */
3440 enum ofperr
3441 ofproto_check_ofpacts(struct ofproto *ofproto,
3442 const struct ofpact ofpacts[], size_t ofpacts_len)
3443 {
3444 const struct ofpact *a;
3445 uint32_t mid;
3446
3447 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
3448 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
3449 return OFPERR_OFPMMFC_INVALID_METER;
3450 }
3451
3452 OFPACT_FOR_EACH_FLATTENED (a, ofpacts, ofpacts_len) {
3453 if (a->type == OFPACT_GROUP
3454 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
3455 return OFPERR_OFPBAC_BAD_OUT_GROUP;
3456 }
3457 }
3458
3459 return 0;
3460 }
3461
3462 static enum ofperr
3463 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3464 {
3465 struct ofproto *p = ofconn_get_ofproto(ofconn);
3466 struct ofputil_packet_out po;
3467 struct dp_packet *payload;
3468 uint64_t ofpacts_stub[1024 / 8];
3469 struct ofpbuf ofpacts;
3470 struct flow flow;
3471 enum ofperr error;
3472
3473 COVERAGE_INC(ofproto_packet_out);
3474
3475 error = reject_slave_controller(ofconn);
3476 if (error) {
3477 goto exit;
3478 }
3479
3480 /* Decode message. */
3481 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3482 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
3483 if (error) {
3484 goto exit_free_ofpacts;
3485 }
3486 if (ofp_to_u16(po.in_port) >= p->max_ports
3487 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
3488 error = OFPERR_OFPBRC_BAD_PORT;
3489 goto exit_free_ofpacts;
3490 }
3491
3492 /* Get payload. */
3493 if (po.buffer_id != UINT32_MAX) {
3494 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
3495 if (error) {
3496 goto exit_free_ofpacts;
3497 }
3498 } else {
3499 /* Ensure that the L3 header is 32-bit aligned. */
3500 payload = dp_packet_clone_data_with_headroom(po.packet, po.packet_len, 2);
3501 }
3502
3503 /* Verify actions against packet, then send packet if successful. */
3504 flow_extract(payload, &flow);
3505 flow.in_port.ofp_port = po.in_port;
3506
3507 /* Check actions like for flow mods. We pass a 'table_id' of 0 to
3508 * ofproto_check_consistency(), which isn't strictly correct because these
3509 * actions aren't in any table. This is OK as 'table_id' is only used to
3510 * check instructions (e.g., goto-table), which can't appear on the action
3511 * list of a packet-out. */
3512 error = ofpacts_check_consistency(po.ofpacts, po.ofpacts_len,
3513 &flow, u16_to_ofp(p->max_ports),
3514 0, p->n_tables,
3515 ofconn_get_protocol(ofconn));
3516 if (!error) {
3517 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
3518 if (!error) {
3519 error = p->ofproto_class->packet_out(p, payload, &flow,
3520 po.ofpacts, po.ofpacts_len);
3521 }
3522 }
3523 dp_packet_delete(payload);
3524
3525 exit_free_ofpacts:
3526 ofpbuf_uninit(&ofpacts);
3527 exit:
3528 return error;
3529 }
3530
3531 static enum ofperr
3532 handle_nxt_resume(struct ofconn *ofconn, const struct ofp_header *oh)
3533 {
3534 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3535 struct ofputil_packet_in_private pin;
3536 enum ofperr error;
3537
3538 error = ofputil_decode_packet_in_private(oh, false, &pin, NULL, NULL);
3539 if (error) {
3540 return error;
3541 }
3542
3543 error = (ofproto->ofproto_class->nxt_resume
3544 ? ofproto->ofproto_class->nxt_resume(ofproto, &pin)
3545 : OFPERR_NXR_NOT_SUPPORTED);
3546
3547 ofputil_packet_in_private_destroy(&pin);
3548
3549 return error;
3550 }
3551
3552 static void
3553 update_port_config(struct ofconn *ofconn, struct ofport *port,
3554 enum ofputil_port_config config,
3555 enum ofputil_port_config mask)
3556 {
3557 enum ofputil_port_config toggle = (config ^ port->pp.config) & mask;
3558
3559 if (toggle & OFPUTIL_PC_PORT_DOWN
3560 && (config & OFPUTIL_PC_PORT_DOWN
3561 ? netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL)
3562 : netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL))) {
3563 /* We tried to bring the port up or down, but it failed, so don't
3564 * update the "down" bit. */
3565 toggle &= ~OFPUTIL_PC_PORT_DOWN;
3566 }
3567
3568 if (toggle) {
3569 enum ofputil_port_config old_config = port->pp.config;
3570 port->pp.config ^= toggle;
3571 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
3572 connmgr_send_port_status(port->ofproto->connmgr, ofconn, &port->pp,
3573 OFPPR_MODIFY);
3574 }
3575 }
3576
3577 static enum ofperr
3578 port_mod_start(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3579 struct ofport **port)
3580 {
3581 struct ofproto *p = ofconn_get_ofproto(ofconn);
3582
3583 *port = ofproto_get_port(p, pm->port_no);
3584 if (!*port) {
3585 return OFPERR_OFPPMFC_BAD_PORT;
3586 }
3587 if (!eth_addr_equals((*port)->pp.hw_addr, pm->hw_addr)) {
3588 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3589 }
3590 return 0;
3591 }
3592
3593 static void
3594 port_mod_finish(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3595 struct ofport *port)
3596 {
3597 update_port_config(ofconn, port, pm->config, pm->mask);
3598 if (pm->advertise) {
3599 netdev_set_advertisements(port->netdev, pm->advertise);
3600 }
3601 }
3602
3603 static enum ofperr
3604 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3605 {
3606 struct ofputil_port_mod pm;
3607 struct ofport *port;
3608 enum ofperr error;
3609
3610 error = reject_slave_controller(ofconn);
3611 if (error) {
3612 return error;
3613 }
3614
3615 error = ofputil_decode_port_mod(oh, &pm, false);
3616 if (error) {
3617 return error;
3618 }
3619
3620 error = port_mod_start(ofconn, &pm, &port);
3621 if (!error) {
3622 port_mod_finish(ofconn, &pm, port);
3623 }
3624 return error;
3625 }
3626
3627 static enum ofperr
3628 handle_desc_stats_request(struct ofconn *ofconn,
3629 const struct ofp_header *request)
3630 {
3631 static const char *default_mfr_desc = "Nicira, Inc.";
3632 static const char *default_hw_desc = "Open vSwitch";
3633 static const char *default_sw_desc = VERSION;
3634 static const char *default_serial_desc = "None";
3635 static const char *default_dp_desc = "None";
3636
3637 struct ofproto *p = ofconn_get_ofproto(ofconn);
3638 struct ofp_desc_stats *ods;
3639 struct ofpbuf *msg;
3640
3641 msg = ofpraw_alloc_stats_reply(request, 0);
3642 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3643 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3644 sizeof ods->mfr_desc);
3645 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3646 sizeof ods->hw_desc);
3647 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3648 sizeof ods->sw_desc);
3649 ovs_strlcpy(ods->serial_num,
3650 p->serial_desc ? p->serial_desc : default_serial_desc,
3651 sizeof ods->serial_num);
3652 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3653 sizeof ods->dp_desc);
3654 ofconn_send_reply(ofconn, msg);
3655
3656 return 0;
3657 }
3658
3659 static enum ofperr
3660 handle_table_stats_request(struct ofconn *ofconn,
3661 const struct ofp_header *request)
3662 {
3663 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3664 struct ofputil_table_features *features;
3665 struct ofputil_table_stats *stats;
3666 struct ofpbuf *reply;
3667 size_t i;
3668
3669 query_tables(ofproto, &features, &stats);
3670
3671 reply = ofputil_encode_table_stats_reply(request);
3672 for (i = 0; i < ofproto->n_tables; i++) {
3673 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3674 ofputil_append_table_stats_reply(reply, &stats[i], &features[i]);
3675 }
3676 }
3677 ofconn_send_reply(ofconn, reply);
3678
3679 free(features);
3680 free(stats);
3681
3682 return 0;
3683 }
3684
3685 static enum ofperr
3686 handle_table_features_request(struct ofconn *ofconn,
3687 const struct ofp_header *request)
3688 {
3689 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3690 struct ofpbuf msg = ofpbuf_const_initializer(request,
3691 ntohs(request->length));
3692 ofpraw_pull_assert(&msg);
3693 if (msg.size || ofpmp_more(request)) {
3694 return OFPERR_OFPTFFC_EPERM;
3695 }
3696
3697 struct ofputil_table_features *features;
3698 query_tables(ofproto, &features, NULL);
3699
3700 struct ovs_list replies;
3701 ofpmp_init(&replies, request);
3702 for (size_t i = 0; i < ofproto->n_tables; i++) {
3703 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3704 ofputil_append_table_features_reply(&features[i], &replies);
3705 }
3706 }
3707 ofconn_send_replies(ofconn, &replies);
3708
3709 free(features);
3710
3711 return 0;
3712 }
3713
3714 /* Returns the vacancy of 'oftable', a number that ranges from 0 (if the table
3715 * is full) to 100 (if the table is empty).
3716 *
3717 * A table without a limit on flows is considered to be empty. */
3718 static uint8_t
3719 oftable_vacancy(const struct oftable *t)
3720 {
3721 return (!t->max_flows ? 100
3722 : t->n_flows >= t->max_flows ? 0
3723 : (t->max_flows - t->n_flows) * 100.0 / t->max_flows);
3724 }
3725
3726 static void
3727 query_table_desc__(struct ofputil_table_desc *td,
3728 struct ofproto *ofproto, uint8_t table_id)
3729 {
3730 const struct oftable *t = &ofproto->tables[table_id];
3731
3732 td->table_id = table_id;
3733 td->eviction = (t->eviction & EVICTION_OPENFLOW
3734 ? OFPUTIL_TABLE_EVICTION_ON
3735 : OFPUTIL_TABLE_EVICTION_OFF);
3736 td->eviction_flags = OFPROTO_EVICTION_FLAGS;
3737 td->vacancy = (t->vacancy_event
3738 ? OFPUTIL_TABLE_VACANCY_ON
3739 : OFPUTIL_TABLE_VACANCY_OFF);
3740 td->table_vacancy.vacancy_down = t->vacancy_down;
3741 td->table_vacancy.vacancy_up = t->vacancy_up;
3742 td->table_vacancy.vacancy = oftable_vacancy(t);
3743 }
3744
3745 /* This function queries the database for dumping table-desc. */
3746 static void
3747 query_tables_desc(struct ofproto *ofproto, struct ofputil_table_desc **descp)
3748 {
3749 struct ofputil_table_desc *table_desc;
3750 size_t i;
3751
3752 table_desc = *descp = xcalloc(ofproto->n_tables, sizeof *table_desc);
3753 for (i = 0; i < ofproto->n_tables; i++) {
3754 struct ofputil_table_desc *td = &table_desc[i];
3755 query_table_desc__(td, ofproto, i);
3756 }
3757 }
3758
3759 /* Function to handle dump-table-desc request. */
3760 static enum ofperr
3761 handle_table_desc_request(struct ofconn *ofconn,
3762 const struct ofp_header *request)
3763 {
3764 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3765 struct ofputil_table_desc *table_desc;
3766 struct ovs_list replies;
3767 size_t i;
3768
3769 query_tables_desc(ofproto, &table_desc);
3770 ofpmp_init(&replies, request);
3771 for (i = 0; i < ofproto->n_tables; i++) {
3772 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3773 ofputil_append_table_desc_reply(&table_desc[i], &replies,
3774 request->version);
3775 }
3776 }
3777 ofconn_send_replies(ofconn, &replies);
3778 free(table_desc);
3779 return 0;
3780 }
3781
3782 /* This function determines and sends the vacancy event, based on the value
3783 * of current vacancy and threshold vacancy. If the current vacancy is less
3784 * than or equal to vacancy_down, vacancy up events must be enabled, and when
3785 * the current vacancy is greater or equal to vacancy_up, vacancy down events
3786 * must be enabled. */
3787 static void
3788 send_table_status(struct ofproto *ofproto, uint8_t table_id)
3789 {
3790 struct oftable *t = &ofproto->tables[table_id];
3791 if (!t->vacancy_event) {
3792 return;
3793 }
3794
3795 uint8_t vacancy = oftable_vacancy(t);
3796 enum ofp14_table_reason event;
3797 if (vacancy < t->vacancy_down) {
3798 event = OFPTR_VACANCY_DOWN;
3799 } else if (vacancy > t->vacancy_up) {
3800 event = OFPTR_VACANCY_UP;
3801 } else {
3802 return;
3803 }
3804
3805 if (event == t->vacancy_event) {
3806 struct ofputil_table_desc td;
3807 query_table_desc__(&td, ofproto, table_id);
3808 connmgr_send_table_status(ofproto->connmgr, &td, event);
3809
3810 t->vacancy_event = (event == OFPTR_VACANCY_DOWN
3811 ? OFPTR_VACANCY_UP
3812 : OFPTR_VACANCY_DOWN);
3813 }
3814 }
3815
3816 static void
3817 append_port_stat(struct ofport *port, struct ovs_list *replies)
3818 {
3819 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3820
3821 calc_duration(port->created, time_msec(),
3822 &ops.duration_sec, &ops.duration_nsec);
3823
3824 /* Intentionally ignore return value, since errors will set
3825 * 'stats' to all-1s, which is correct for OpenFlow, and
3826 * netdev_get_stats() will log errors. */
3827 ofproto_port_get_stats(port, &ops.stats);
3828
3829 ofputil_append_port_stat(replies, &ops);
3830 }
3831
3832 static void
3833 handle_port_request(struct ofconn *ofconn,
3834 const struct ofp_header *request, ofp_port_t port_no,
3835 void (*cb)(struct ofport *, struct ovs_list *replies))
3836 {
3837 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3838 struct ofport *port;
3839 struct ovs_list replies;
3840
3841 ofpmp_init(&replies, request);
3842 if (port_no != OFPP_ANY) {
3843 port = ofproto_get_port(ofproto, port_no);
3844 if (port) {
3845 cb(port, &replies);
3846 }
3847 } else {
3848 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3849 cb(port, &replies);
3850 }
3851 }
3852
3853 ofconn_send_replies(ofconn, &replies);
3854 }
3855
3856 static enum ofperr
3857 handle_port_stats_request(struct ofconn *ofconn,
3858 const struct ofp_header *request)
3859 {
3860 ofp_port_t port_no;
3861 enum ofperr error;
3862
3863 error = ofputil_decode_port_stats_request(request, &port_no);
3864 if (!error) {
3865 handle_port_request(ofconn, request, port_no, append_port_stat);
3866 }
3867 return error;
3868 }
3869
3870 static void
3871 append_port_desc(struct ofport *port, struct ovs_list *replies)
3872 {
3873 ofputil_append_port_desc_stats_reply(&port->pp, replies);
3874 }
3875
3876 static enum ofperr
3877 handle_port_desc_stats_request(struct ofconn *ofconn,
3878 const struct ofp_header *request)
3879 {
3880 ofp_port_t port_no;
3881 enum ofperr error;
3882
3883 error = ofputil_decode_port_desc_stats_request(request, &port_no);
3884 if (!error) {
3885 handle_port_request(ofconn, request, port_no, append_port_desc);
3886 }
3887 return error;
3888 }
3889
3890 static uint32_t
3891 hash_cookie(ovs_be64 cookie)
3892 {
3893 return hash_uint64((OVS_FORCE uint64_t)cookie);
3894 }
3895
3896 static void
3897 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3898 OVS_REQUIRES(ofproto_mutex)
3899 {
3900 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3901 hash_cookie(rule->flow_cookie));
3902 }
3903
3904 static void
3905 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3906 OVS_REQUIRES(ofproto_mutex)
3907 {
3908 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3909 }
3910
3911 static void
3912 calc_duration(long long int start, long long int now,
3913 uint32_t *sec, uint32_t *nsec)
3914 {
3915 long long int msecs = now - start;
3916 *sec = msecs / 1000;
3917 *nsec = (msecs % 1000) * (1000 * 1000);
3918 }
3919
3920 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3921 * true if 'table_id' is OK, false otherwise. */
3922 static bool
3923 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3924 {
3925 return table_id == OFPTT_ALL || table_id < ofproto->n_tables;
3926 }
3927
3928 static struct oftable *
3929 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3930 {
3931 struct oftable *table;
3932
3933 for (table = &ofproto->tables[table_id];
3934 table < &ofproto->tables[ofproto->n_tables];
3935 table++) {
3936 if (!(table->flags & OFTABLE_HIDDEN)) {
3937 return table;
3938 }
3939 }
3940
3941 return NULL;
3942 }
3943
3944 static struct oftable *
3945 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3946 {
3947 if (table_id == 0xff) {
3948 return next_visible_table(ofproto, 0);
3949 } else if (table_id < ofproto->n_tables) {
3950 return &ofproto->tables[table_id];
3951 } else {
3952 return NULL;
3953 }
3954 }
3955
3956 static struct oftable *
3957 next_matching_table(const struct ofproto *ofproto,
3958 const struct oftable *table, uint8_t table_id)
3959 {
3960 return (table_id == 0xff
3961 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3962 : NULL);
3963 }
3964
3965 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3966 *
3967 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3968 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3969 *
3970 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3971 * only once, for that table. (This can be used to access tables marked
3972 * OFTABLE_HIDDEN.)
3973 *
3974 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3975 * entered at all. (Perhaps you should have validated TABLE_ID with
3976 * check_table_id().)
3977 *
3978 * All parameters are evaluated multiple times.
3979 */
3980 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3981 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3982 (TABLE) != NULL; \
3983 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3984
3985 /* Initializes 'criteria' in a straightforward way based on the other
3986 * parameters.
3987 *
3988 * By default, the criteria include flows that are read-only, on the assumption
3989 * that the collected flows won't be modified. Call rule_criteria_require_rw()
3990 * if flows will be modified.
3991 *
3992 * For "loose" matching, the 'priority' parameter is unimportant and may be
3993 * supplied as 0. */
3994 static void
3995 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3996 const struct match *match, int priority,
3997 ovs_version_t version, ovs_be64 cookie,
3998 ovs_be64 cookie_mask, ofp_port_t out_port,
3999 uint32_t out_group)
4000 {
4001 criteria->table_id = table_id;
4002 cls_rule_init(&criteria->cr, match, priority);
4003 criteria->version = version;
4004 criteria->cookie = cookie;
4005 criteria->cookie_mask = cookie_mask;
4006 criteria->out_port = out_port;
4007 criteria->out_group = out_group;
4008
4009 /* We ordinarily want to skip hidden rules, but there has to be a way for
4010 * code internal to OVS to modify and delete them, so if the criteria
4011 * specify a priority that can only be for a hidden flow, then allow hidden
4012 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
4013 * with hidden flows because OpenFlow uses only a 16-bit field to specify
4014 * priority.) */
4015 criteria->include_hidden = priority > UINT16_MAX;
4016
4017 /* We assume that the criteria are being used to collect flows for reading
4018 * but not modification. Thus, we should collect read-only flows. */
4019 criteria->include_readonly = true;
4020 }
4021
4022 /* By default, criteria initialized by rule_criteria_init() will match flows
4023 * that are read-only, on the assumption that the collected flows won't be
4024 * modified. Call this function to match only flows that are be modifiable.
4025 *
4026 * Specify 'can_write_readonly' as false in ordinary circumstances, true if the
4027 * caller has special privileges that allow it to modify even "read-only"
4028 * flows. */
4029 static void
4030 rule_criteria_require_rw(struct rule_criteria *criteria,
4031 bool can_write_readonly)
4032 {
4033 criteria->include_readonly = can_write_readonly;
4034 }
4035
4036 static void
4037 rule_criteria_destroy(struct rule_criteria *criteria)
4038 {
4039 cls_rule_destroy(&criteria->cr);
4040 }
4041
4042 void
4043 rule_collection_init(struct rule_collection *rules)
4044 {
4045 rules->rules = rules->stub;
4046 rules->n = 0;
4047 rules->capacity = ARRAY_SIZE(rules->stub);
4048 }
4049
4050 void
4051 rule_collection_add(struct rule_collection *rules, struct rule *rule)
4052 {
4053 if (rules->n >= rules->capacity) {
4054 size_t old_size, new_size;
4055
4056 old_size = rules->capacity * sizeof *rules->rules;
4057 rules->capacity *= 2;
4058 new_size = rules->capacity * sizeof *rules->rules;
4059
4060 if (rules->rules == rules->stub) {
4061 rules->rules = xmalloc(new_size);
4062 memcpy(rules->rules, rules->stub, old_size);
4063 } else {
4064 rules->rules = xrealloc(rules->rules, new_size);
4065 }
4066 }
4067
4068 rules->rules[rules->n++] = rule;
4069 }
4070
4071 void
4072 rule_collection_ref(struct rule_collection *rules)
4073 OVS_REQUIRES(ofproto_mutex)
4074 {
4075 size_t i;
4076
4077 for (i = 0; i < rules->n; i++) {
4078 ofproto_rule_ref(rules->rules[i]);
4079 }
4080 }
4081
4082 void
4083 rule_collection_unref(struct rule_collection *rules)
4084 {
4085 size_t i;
4086
4087 for (i = 0; i < rules->n; i++) {
4088 ofproto_rule_unref(rules->rules[i]);
4089 }
4090 }
4091
4092 /* Returns a NULL-terminated array of rule pointers,
4093 * destroys 'rules'. */
4094 static struct rule **
4095 rule_collection_detach(struct rule_collection *rules)
4096 {
4097 struct rule **rule_array;
4098
4099 rule_collection_add(rules, NULL);
4100
4101 if (rules->rules == rules->stub) {
4102 rules->rules = xmemdup(rules->rules, rules->n * sizeof *rules->rules);
4103 }
4104
4105 rule_array = rules->rules;
4106 rule_collection_init(rules);
4107
4108 return rule_array;
4109 }
4110
4111 void
4112 rule_collection_destroy(struct rule_collection *rules)
4113 {
4114 if (rules->rules != rules->stub) {
4115 free(rules->rules);
4116 }
4117
4118 /* Make repeated destruction harmless. */
4119 rule_collection_init(rules);
4120 }
4121
4122 /* Schedules postponed removal of rules, destroys 'rules'. */
4123 static void
4124 rule_collection_remove_postponed(struct rule_collection *rules)
4125 OVS_REQUIRES(ofproto_mutex)
4126 {
4127 if (rules->n > 0) {
4128 if (rules->n == 1) {
4129 ovsrcu_postpone(remove_rule_rcu, rules->rules[0]);
4130 } else {
4131 ovsrcu_postpone(remove_rules_rcu, rule_collection_detach(rules));
4132 }
4133 }
4134 }
4135
4136 /* Checks whether 'rule' matches 'c' and, if so, adds it to 'rules'. This
4137 * function verifies most of the criteria in 'c' itself, but the caller must
4138 * check 'c->cr' itself.
4139 *
4140 * Rules that have already been marked for removal are not collected.
4141 *
4142 * Increments '*n_readonly' if 'rule' wasn't added because it's read-only (and
4143 * 'c' only includes modifiable rules). */
4144 static void
4145 collect_rule(struct rule *rule, const struct rule_criteria *c,
4146 struct rule_collection *rules, size_t *n_readonly)
4147 OVS_REQUIRES(ofproto_mutex)
4148 {
4149 if ((c->table_id == rule->table_id || c->table_id == 0xff)
4150 && ofproto_rule_has_out_port(rule, c->out_port)
4151 && ofproto_rule_has_out_group(rule, c->out_group)
4152 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)
4153 && (!rule_is_hidden(rule) || c->include_hidden)
4154 && cls_rule_visible_in_version(&rule->cr, c->version)) {
4155 /* Rule matches all the criteria... */
4156 if (!rule_is_readonly(rule) || c->include_readonly) {
4157 /* ...add it. */
4158 rule_collection_add(rules, rule);
4159 } else {
4160 /* ...except it's read-only. */
4161 ++*n_readonly;
4162 }
4163 }
4164 }
4165
4166 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4167 * on classifiers rules are done in the "loose" way required for OpenFlow
4168 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
4169 * 'rules'.
4170 *
4171 * Returns 0 on success, otherwise an OpenFlow error code. */
4172 static enum ofperr
4173 collect_rules_loose(struct ofproto *ofproto,
4174 const struct rule_criteria *criteria,
4175 struct rule_collection *rules)
4176 OVS_REQUIRES(ofproto_mutex)
4177 {
4178 struct oftable *table;
4179 enum ofperr error = 0;
4180 size_t n_readonly = 0;
4181
4182 rule_collection_init(rules);
4183
4184 if (!check_table_id(ofproto, criteria->table_id)) {
4185 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4186 goto exit;
4187 }
4188
4189 if (criteria->cookie_mask == OVS_BE64_MAX) {
4190 struct rule *rule;
4191
4192 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4193 hash_cookie(criteria->cookie),
4194 &ofproto->cookies) {
4195 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
4196 collect_rule(rule, criteria, rules, &n_readonly);
4197 }
4198 }
4199 } else {
4200 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4201 struct rule *rule;
4202
4203 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &criteria->cr,
4204 criteria->version) {
4205 collect_rule(rule, criteria, rules, &n_readonly);
4206 }
4207 }
4208 }
4209
4210 exit:
4211 if (!error && !rules->n && n_readonly) {
4212 /* We didn't find any rules to modify. We did find some read-only
4213 * rules that we're not allowed to modify, so report that. */
4214 error = OFPERR_OFPBRC_EPERM;
4215 }
4216 if (error) {
4217 rule_collection_destroy(rules);
4218 }
4219 return error;
4220 }
4221
4222 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4223 * on classifiers rules are done in the "strict" way required for OpenFlow
4224 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
4225 * rules on list 'rules'.
4226 *
4227 * Returns 0 on success, otherwise an OpenFlow error code. */
4228 static enum ofperr
4229 collect_rules_strict(struct ofproto *ofproto,
4230 const struct rule_criteria *criteria,
4231 struct rule_collection *rules)
4232 OVS_REQUIRES(ofproto_mutex)
4233 {
4234 struct oftable *table;
4235 size_t n_readonly = 0;
4236 enum ofperr error = 0;
4237
4238 rule_collection_init(rules);
4239
4240 if (!check_table_id(ofproto, criteria->table_id)) {
4241 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4242 goto exit;
4243 }
4244
4245 if (criteria->cookie_mask == OVS_BE64_MAX) {
4246 struct rule *rule;
4247
4248 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4249 hash_cookie(criteria->cookie),
4250 &ofproto->cookies) {
4251 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
4252 collect_rule(rule, criteria, rules, &n_readonly);
4253 }
4254 }
4255 } else {
4256 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4257 struct rule *rule;
4258
4259 rule = rule_from_cls_rule(classifier_find_rule_exactly(
4260 &table->cls, &criteria->cr,
4261 criteria->version));
4262 if (rule) {
4263 collect_rule(rule, criteria, rules, &n_readonly);
4264 }
4265 }
4266 }
4267
4268 exit:
4269 if (!error && !rules->n && n_readonly) {
4270 /* We didn't find any rules to modify. We did find some read-only
4271 * rules that we're not allowed to modify, so report that. */
4272 error = OFPERR_OFPBRC_EPERM;
4273 }
4274 if (error) {
4275 rule_collection_destroy(rules);
4276 }
4277 return error;
4278 }
4279
4280 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
4281 * forced into the range of a uint16_t. */
4282 static int
4283 age_secs(long long int age_ms)
4284 {
4285 return (age_ms < 0 ? 0
4286 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
4287 : (unsigned int) age_ms / 1000);
4288 }
4289
4290 static enum ofperr
4291 handle_flow_stats_request(struct ofconn *ofconn,
4292 const struct ofp_header *request)
4293 OVS_EXCLUDED(ofproto_mutex)
4294 {
4295 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4296 struct ofputil_flow_stats_request fsr;
4297 struct rule_criteria criteria;
4298 struct rule_collection rules;
4299 struct ovs_list replies;
4300 enum ofperr error;
4301 size_t i;
4302
4303 error = ofputil_decode_flow_stats_request(&fsr, request);
4304 if (error) {
4305 return error;
4306 }
4307
4308 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, OVS_VERSION_MAX,
4309 fsr.cookie, fsr.cookie_mask, fsr.out_port,
4310 fsr.out_group);
4311
4312 ovs_mutex_lock(&ofproto_mutex);
4313 error = collect_rules_loose(ofproto, &criteria, &rules);
4314 rule_criteria_destroy(&criteria);
4315 if (!error) {
4316 rule_collection_ref(&rules);
4317 }
4318 ovs_mutex_unlock(&ofproto_mutex);
4319
4320 if (error) {
4321 return error;
4322 }
4323
4324 ofpmp_init(&replies, request);
4325 for (i = 0; i < rules.n; i++) {
4326 struct rule *rule = rules.rules[i];
4327 long long int now = time_msec();
4328 struct ofputil_flow_stats fs;
4329 long long int created, used, modified;
4330 const struct rule_actions *actions;
4331 enum ofputil_flow_mod_flags flags;
4332
4333 ovs_mutex_lock(&rule->mutex);
4334 fs.cookie = rule->flow_cookie;
4335 fs.idle_timeout = rule->idle_timeout;
4336 fs.hard_timeout = rule->hard_timeout;
4337 fs.importance = rule->importance;
4338 created = rule->created;
4339 modified = rule->modified;
4340 actions = rule_get_actions(rule);
4341 flags = rule->flags;
4342 ovs_mutex_unlock(&rule->mutex);
4343
4344 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
4345 &fs.byte_count, &used);
4346
4347 minimatch_expand(&rule->cr.match, &fs.match);
4348 fs.table_id = rule->table_id;
4349 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
4350 fs.priority = rule->cr.priority;
4351 fs.idle_age = age_secs(now - used);
4352 fs.hard_age = age_secs(now - modified);
4353 fs.ofpacts = actions->ofpacts;
4354 fs.ofpacts_len = actions->ofpacts_len;
4355
4356 fs.flags = flags;
4357 ofputil_append_flow_stats_reply(&fs, &replies);
4358 }
4359
4360 rule_collection_unref(&rules);
4361 rule_collection_destroy(&rules);
4362
4363 ofconn_send_replies(ofconn, &replies);
4364
4365 return 0;
4366 }
4367
4368 static void
4369 flow_stats_ds(struct rule *rule, struct ds *results)
4370 {
4371 uint64_t packet_count, byte_count;
4372 const struct rule_actions *actions;
4373 long long int created, used;
4374
4375 rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4376 &byte_count, &used);
4377
4378 ovs_mutex_lock(&rule->mutex);
4379 actions = rule_get_actions(rule);
4380 created = rule->created;
4381 ovs_mutex_unlock(&rule->mutex);
4382
4383 if (rule->table_id != 0) {
4384 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
4385 }
4386 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
4387 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
4388 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
4389 cls_rule_format(&rule->cr, results);
4390 ds_put_char(results, ',');
4391
4392 ds_put_cstr(results, "actions=");
4393 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
4394
4395 ds_put_cstr(results, "\n");
4396 }
4397
4398 /* Adds a pretty-printed description of all flows to 'results', including
4399 * hidden flows (e.g., set up by in-band control). */
4400 void
4401 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
4402 {
4403 struct oftable *table;
4404
4405 OFPROTO_FOR_EACH_TABLE (table, p) {
4406 struct rule *rule;
4407
4408 CLS_FOR_EACH (rule, cr, &table->cls) {
4409 flow_stats_ds(rule, results);
4410 }
4411 }
4412 }
4413
4414 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
4415 * '*engine_type' and '*engine_id', respectively. */
4416 void
4417 ofproto_get_netflow_ids(const struct ofproto *ofproto,
4418 uint8_t *engine_type, uint8_t *engine_id)
4419 {
4420 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
4421 }
4422
4423 /* Checks the status change of CFM on 'ofport'.
4424 *
4425 * Returns true if 'ofproto_class' does not support 'cfm_status_changed'. */
4426 bool
4427 ofproto_port_cfm_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
4428 {
4429 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4430 return (ofport && ofproto->ofproto_class->cfm_status_changed
4431 ? ofproto->ofproto_class->cfm_status_changed(ofport)
4432 : true);
4433 }
4434
4435 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.
4436 * Returns 0 if the port's CFM status was successfully stored into
4437 * '*status'. Returns positive errno if the port did not have CFM
4438 * configured.
4439 *
4440 * The caller must provide and own '*status', and must free 'status->rmps'.
4441 * '*status' is indeterminate if the return value is non-zero. */
4442 int
4443 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
4444 struct cfm_status *status)
4445 {
4446 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4447 return (ofport && ofproto->ofproto_class->get_cfm_status
4448 ? ofproto->ofproto_class->get_cfm_status(ofport, status)
4449 : EOPNOTSUPP);
4450 }
4451
4452 static enum ofperr
4453 handle_aggregate_stats_request(struct ofconn *ofconn,
4454 const struct ofp_header *oh)
4455 OVS_EXCLUDED(ofproto_mutex)
4456 {
4457 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4458 struct ofputil_flow_stats_request request;
4459 struct ofputil_aggregate_stats stats;
4460 bool unknown_packets, unknown_bytes;
4461 struct rule_criteria criteria;
4462 struct rule_collection rules;
4463 struct ofpbuf *reply;
4464 enum ofperr error;
4465 size_t i;
4466
4467 error = ofputil_decode_flow_stats_request(&request, oh);
4468 if (error) {
4469 return error;
4470 }
4471
4472 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
4473 OVS_VERSION_MAX, request.cookie, request.cookie_mask,
4474 request.out_port, request.out_group);
4475
4476 ovs_mutex_lock(&ofproto_mutex);
4477 error = collect_rules_loose(ofproto, &criteria, &rules);
4478 rule_criteria_destroy(&criteria);
4479 if (!error) {
4480 rule_collection_ref(&rules);
4481 }
4482 ovs_mutex_unlock(&ofproto_mutex);
4483
4484 if (error) {
4485 return error;
4486 }
4487
4488 memset(&stats, 0, sizeof stats);
4489 unknown_packets = unknown_bytes = false;
4490 for (i = 0; i < rules.n; i++) {
4491 struct rule *rule = rules.rules[i];
4492 uint64_t packet_count;
4493 uint64_t byte_count;
4494 long long int used;
4495
4496 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4497 &byte_count, &used);
4498
4499 if (packet_count == UINT64_MAX) {
4500 unknown_packets = true;
4501 } else {
4502 stats.packet_count += packet_count;
4503 }
4504
4505 if (byte_count == UINT64_MAX) {
4506 unknown_bytes = true;
4507 } else {
4508 stats.byte_count += byte_count;
4509 }
4510
4511 stats.flow_count++;
4512 }
4513 if (unknown_packets) {
4514 stats.packet_count = UINT64_MAX;
4515 }
4516 if (unknown_bytes) {
4517 stats.byte_count = UINT64_MAX;
4518 }
4519
4520 rule_collection_unref(&rules);
4521 rule_collection_destroy(&rules);
4522
4523 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
4524 ofconn_send_reply(ofconn, reply);
4525
4526 return 0;
4527 }
4528
4529 struct queue_stats_cbdata {
4530 struct ofport *ofport;
4531 struct ovs_list replies;
4532 long long int now;
4533 };
4534
4535 static void
4536 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
4537 const struct netdev_queue_stats *stats)
4538 {
4539 struct ofputil_queue_stats oqs;
4540
4541 oqs.port_no = cbdata->ofport->pp.port_no;
4542 oqs.queue_id = queue_id;
4543 oqs.tx_bytes = stats->tx_bytes;
4544 oqs.tx_packets = stats->tx_packets;
4545 oqs.tx_errors = stats->tx_errors;
4546 if (stats->created != LLONG_MIN) {
4547 calc_duration(stats->created, cbdata->now,
4548 &oqs.duration_sec, &oqs.duration_nsec);
4549 } else {
4550 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
4551 }
4552 ofputil_append_queue_stat(&cbdata->replies, &oqs);
4553 }
4554
4555 static void
4556 handle_queue_stats_dump_cb(uint32_t queue_id,
4557 struct netdev_queue_stats *stats,
4558 void *cbdata_)
4559 {
4560 struct queue_stats_cbdata *cbdata = cbdata_;
4561
4562 put_queue_stats(cbdata, queue_id, stats);
4563 }
4564
4565 static enum ofperr
4566 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
4567 struct queue_stats_cbdata *cbdata)
4568 {
4569 cbdata->ofport = port;
4570 if (queue_id == OFPQ_ALL) {
4571 netdev_dump_queue_stats(port->netdev,
4572 handle_queue_stats_dump_cb, cbdata);
4573 } else {
4574 struct netdev_queue_stats stats;
4575
4576 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
4577 put_queue_stats(cbdata, queue_id, &stats);
4578 } else {
4579 return OFPERR_OFPQOFC_BAD_QUEUE;
4580 }
4581 }
4582 return 0;
4583 }
4584
4585 static enum ofperr
4586 handle_queue_stats_request(struct ofconn *ofconn,
4587 const struct ofp_header *rq)
4588 {
4589 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4590 struct queue_stats_cbdata cbdata;
4591 struct ofport *port;
4592 enum ofperr error;
4593 struct ofputil_queue_stats_request oqsr;
4594
4595 COVERAGE_INC(ofproto_queue_req);
4596
4597 ofpmp_init(&cbdata.replies, rq);
4598 cbdata.now = time_msec();
4599
4600 error = ofputil_decode_queue_stats_request(rq, &oqsr);
4601 if (error) {
4602 return error;
4603 }
4604
4605 if (oqsr.port_no == OFPP_ANY) {
4606 error = OFPERR_OFPQOFC_BAD_QUEUE;
4607 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
4608 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
4609 error = 0;
4610 }
4611 }
4612 } else {
4613 port = ofproto_get_port(ofproto, oqsr.port_no);
4614 error = (port
4615 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
4616 : OFPERR_OFPQOFC_BAD_PORT);
4617 }
4618 if (!error) {
4619 ofconn_send_replies(ofconn, &cbdata.replies);
4620 } else {
4621 ofpbuf_list_delete(&cbdata.replies);
4622 }
4623
4624 return error;
4625 }
4626
4627 static enum ofperr
4628 evict_rules_from_table(struct oftable *table)
4629 OVS_REQUIRES(ofproto_mutex)
4630 {
4631 enum ofperr error = 0;
4632 struct rule_collection rules;
4633 unsigned int count = table->n_flows;
4634 unsigned int max_flows = table->max_flows;
4635
4636 rule_collection_init(&rules);
4637
4638 while (count-- > max_flows) {
4639 struct rule *rule;
4640
4641 if (!choose_rule_to_evict(table, &rule)) {
4642 error = OFPERR_OFPFMFC_TABLE_FULL;
4643 break;
4644 } else {
4645 eviction_group_remove_rule(rule);
4646 rule_collection_add(&rules, rule);
4647 }
4648 }
4649 delete_flows__(&rules, OFPRR_EVICTION, NULL);
4650
4651 return error;
4652 }
4653
4654 static void
4655 get_conjunctions(const struct ofputil_flow_mod *fm,
4656 struct cls_conjunction **conjsp, size_t *n_conjsp)
4657 OVS_REQUIRES(ofproto_mutex)
4658 {
4659 struct cls_conjunction *conjs = NULL;
4660 int n_conjs = 0;
4661
4662 const struct ofpact *ofpact;
4663 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4664 if (ofpact->type == OFPACT_CONJUNCTION) {
4665 n_conjs++;
4666 } else if (ofpact->type != OFPACT_NOTE) {
4667 /* "conjunction" may appear with "note" actions but not with any
4668 * other type of actions. */
4669 ovs_assert(!n_conjs);
4670 break;
4671 }
4672 }
4673 if (n_conjs) {
4674 int i = 0;
4675
4676 conjs = xzalloc(n_conjs * sizeof *conjs);
4677 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4678 if (ofpact->type == OFPACT_CONJUNCTION) {
4679 struct ofpact_conjunction *oc = ofpact_get_CONJUNCTION(ofpact);
4680 conjs[i].clause = oc->clause;
4681 conjs[i].n_clauses = oc->n_clauses;
4682 conjs[i].id = oc->id;
4683 i++;
4684 }
4685 }
4686 }
4687
4688 *conjsp = conjs;
4689 *n_conjsp = n_conjs;
4690 }
4691
4692 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4693 * in which no matching flow already exists in the flow table.
4694 *
4695 * Adds the flow specified by 'fm', to the ofproto's flow table. Returns 0 on
4696 * success, or an OpenFlow error code on failure.
4697 *
4698 * On successful return the caller must complete the operation either by
4699 * calling add_flow_finish(), or add_flow_revert() if the operation needs to
4700 * be reverted.
4701 *
4702 * The caller retains ownership of 'fm->ofpacts'. */
4703 static enum ofperr
4704 add_flow_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4705 OVS_REQUIRES(ofproto_mutex)
4706 {
4707 struct ofputil_flow_mod *fm = &ofm->fm;
4708 struct rule **old_rule = &ofm->old_rules.stub[0];
4709 struct rule **new_rule = &ofm->new_rules.stub[0];
4710 struct oftable *table;
4711 struct cls_rule cr;
4712 struct rule *rule;
4713 uint8_t table_id;
4714 struct cls_conjunction *conjs;
4715 size_t n_conjs;
4716 enum ofperr error;
4717
4718 if (!check_table_id(ofproto, fm->table_id)) {
4719 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4720 return error;
4721 }
4722
4723 /* Pick table. */
4724 if (fm->table_id == 0xff) {
4725 if (ofproto->ofproto_class->rule_choose_table) {
4726 error = ofproto->ofproto_class->rule_choose_table(ofproto,
4727 &fm->match,
4728 &table_id);
4729 if (error) {
4730 return error;
4731 }
4732 ovs_assert(table_id < ofproto->n_tables);
4733 } else {
4734 table_id = 0;
4735 }
4736 } else if (fm->table_id < ofproto->n_tables) {
4737 table_id = fm->table_id;
4738 } else {
4739 return OFPERR_OFPBRC_BAD_TABLE_ID;
4740 }
4741
4742 table = &ofproto->tables[table_id];
4743 if (table->flags & OFTABLE_READONLY
4744 && !(fm->flags & OFPUTIL_FF_NO_READONLY)) {
4745 return OFPERR_OFPBRC_EPERM;
4746 }
4747
4748 if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)
4749 && !match_has_default_hidden_fields(&fm->match)) {
4750 VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set "
4751 "non-default values to hidden fields", ofproto->name);
4752 return OFPERR_OFPBRC_EPERM;
4753 }
4754
4755 cls_rule_init(&cr, &fm->match, fm->priority);
4756
4757 /* Check for the existence of an identical rule.
4758 * This will not return rules earlier marked for removal. */
4759 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr,
4760 ofm->version));
4761 *old_rule = rule;
4762 if (!rule) {
4763 /* Check for overlap, if requested. */
4764 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP
4765 && classifier_rule_overlaps(&table->cls, &cr, ofm->version)) {
4766 cls_rule_destroy(&cr);
4767 return OFPERR_OFPFMFC_OVERLAP;
4768 }
4769
4770 /* If necessary, evict an existing rule to clear out space. */
4771 if (table->n_flows >= table->max_flows) {
4772 if (!choose_rule_to_evict(table, &rule)) {
4773 error = OFPERR_OFPFMFC_TABLE_FULL;
4774 cls_rule_destroy(&cr);
4775 return error;
4776 }
4777 eviction_group_remove_rule(rule);
4778 /* Marks '*old_rule' as an evicted rule rather than replaced rule.
4779 */
4780 fm->delete_reason = OFPRR_EVICTION;
4781 *old_rule = rule;
4782 }
4783 } else {
4784 fm->modify_cookie = true;
4785 }
4786
4787 /* Allocate new rule. */
4788 error = replace_rule_create(ofproto, fm, &cr, table - ofproto->tables,
4789 rule, new_rule);
4790 if (error) {
4791 return error;
4792 }
4793
4794 get_conjunctions(fm, &conjs, &n_conjs);
4795 replace_rule_start(ofproto, ofm->version, rule, *new_rule, conjs, n_conjs);
4796 free(conjs);
4797
4798 return 0;
4799 }
4800
4801 /* Revert the effects of add_flow_start(). */
4802 static void
4803 add_flow_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4804 OVS_REQUIRES(ofproto_mutex)
4805 {
4806 struct ofputil_flow_mod *fm = &ofm->fm;
4807 struct rule *old_rule = ofm->old_rules.stub[0];
4808 struct rule *new_rule = ofm->new_rules.stub[0];
4809
4810 if (old_rule && fm->delete_reason == OFPRR_EVICTION) {
4811 /* Revert the eviction. */
4812 eviction_group_add_rule(old_rule);
4813 }
4814
4815 replace_rule_revert(ofproto, old_rule, new_rule);
4816 }
4817
4818 /* To be called after version bump. */
4819 static void
4820 add_flow_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
4821 const struct flow_mod_requester *req)
4822 OVS_REQUIRES(ofproto_mutex)
4823 {
4824 struct ofputil_flow_mod *fm = &ofm->fm;
4825 struct rule *old_rule = ofm->old_rules.stub[0];
4826 struct rule *new_rule = ofm->new_rules.stub[0];
4827 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
4828
4829 replace_rule_finish(ofproto, fm, req, old_rule, new_rule, &dead_cookies);
4830 learned_cookies_flush(ofproto, &dead_cookies);
4831
4832 if (old_rule) {
4833 ovsrcu_postpone(remove_rule_rcu, old_rule);
4834 } else {
4835 ofmonitor_report(ofproto->connmgr, new_rule, NXFME_ADDED, 0,
4836 req ? req->ofconn : NULL,
4837 req ? req->request->xid : 0, NULL);
4838
4839 /* Send Vacancy Events for OF1.4+. */
4840 send_table_status(ofproto, new_rule->table_id);
4841 }
4842
4843 send_buffered_packet(req, fm->buffer_id, new_rule);
4844 }
4845 \f
4846 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4847
4848 /* Create a new rule based on attributes in 'fm', match in 'cr', 'table_id',
4849 * and 'old_rule'. Note that the rule is NOT inserted into a any data
4850 * structures yet. Takes ownership of 'cr'. */
4851 static enum ofperr
4852 replace_rule_create(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4853 struct cls_rule *cr, uint8_t table_id,
4854 struct rule *old_rule, struct rule **new_rule)
4855 {
4856 struct rule *rule;
4857 enum ofperr error;
4858
4859 /* Allocate new rule. */
4860 rule = ofproto->ofproto_class->rule_alloc();
4861 if (!rule) {
4862 cls_rule_destroy(cr);
4863 VLOG_WARN_RL(&rl, "%s: failed to allocate a rule.", ofproto->name);
4864 return OFPERR_OFPFMFC_UNKNOWN;
4865 }
4866
4867 /* Initialize base state. */
4868 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4869 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), cr);
4870 ovs_refcount_init(&rule->ref_count);
4871 rule->flow_cookie = fm->new_cookie;
4872 rule->created = rule->modified = time_msec();
4873
4874 ovs_mutex_init(&rule->mutex);
4875 ovs_mutex_lock(&rule->mutex);
4876 rule->idle_timeout = fm->idle_timeout;
4877 rule->hard_timeout = fm->hard_timeout;
4878 *CONST_CAST(uint16_t *, &rule->importance) = fm->importance;
4879 rule->removed_reason = OVS_OFPRR_NONE;
4880
4881 *CONST_CAST(uint8_t *, &rule->table_id) = table_id;
4882 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4883 *CONST_CAST(const struct rule_actions **, &rule->actions)
4884 = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
4885 ovs_list_init(&rule->meter_list_node);
4886 rule->eviction_group = NULL;
4887 ovs_list_init(&rule->expirable);
4888 rule->monitor_flags = 0;
4889 rule->add_seqno = 0;
4890 rule->modify_seqno = 0;
4891
4892 /* Copy values from old rule for modify semantics. */
4893 if (old_rule && fm->delete_reason != OFPRR_EVICTION) {
4894 bool change_cookie = (fm->modify_cookie
4895 && fm->new_cookie != OVS_BE64_MAX
4896 && fm->new_cookie != old_rule->flow_cookie);
4897
4898 ovs_mutex_lock(&old_rule->mutex);
4899 if (fm->command != OFPFC_ADD) {
4900 rule->idle_timeout = old_rule->idle_timeout;
4901 rule->hard_timeout = old_rule->hard_timeout;
4902 *CONST_CAST(uint16_t *, &rule->importance) = old_rule->importance;
4903 rule->flags = old_rule->flags;
4904 rule->created = old_rule->created;
4905 }
4906 if (!change_cookie) {
4907 rule->flow_cookie = old_rule->flow_cookie;
4908 }
4909 ovs_mutex_unlock(&old_rule->mutex);
4910 }
4911 ovs_mutex_unlock(&rule->mutex);
4912
4913 /* Construct rule, initializing derived state. */
4914 error = ofproto->ofproto_class->rule_construct(rule);
4915 if (error) {
4916 ofproto_rule_destroy__(rule);
4917 return error;
4918 }
4919
4920 rule->removed = true; /* Not yet in ofproto data structures. */
4921
4922 *new_rule = rule;
4923 return 0;
4924 }
4925
4926 static void
4927 replace_rule_start(struct ofproto *ofproto, ovs_version_t version,
4928 struct rule *old_rule, struct rule *new_rule,
4929 struct cls_conjunction *conjs, size_t n_conjs)
4930 {
4931 struct oftable *table = &ofproto->tables[new_rule->table_id];
4932
4933 /* 'old_rule' may be either an evicted rule or replaced rule. */
4934 if (old_rule) {
4935 /* Mark the old rule for removal in the next version. */
4936 cls_rule_make_invisible_in_version(&old_rule->cr, version);
4937
4938 /* Remove the old rule from data structures. */
4939 ofproto_rule_remove__(ofproto, old_rule);
4940 } else {
4941 table->n_flows++;
4942 }
4943 /* Insert flow to the classifier, so that later flow_mods may relate
4944 * to it. This is reversible, in case later errors require this to
4945 * be reverted. */
4946 ofproto_rule_insert__(ofproto, new_rule);
4947 /* Make the new rule visible for classifier lookups only from the next
4948 * version. */
4949 classifier_insert(&table->cls, &new_rule->cr, version, conjs, n_conjs);
4950 }
4951
4952 static void replace_rule_revert(struct ofproto *ofproto,
4953 struct rule *old_rule, struct rule *new_rule)
4954 {
4955 struct oftable *table = &ofproto->tables[new_rule->table_id];
4956
4957 if (old_rule) {
4958 /* Restore the old rule to data structures. */
4959 ofproto_rule_insert__(ofproto, old_rule);
4960
4961 /* Restore the original visibility of the old rule. */
4962 cls_rule_restore_visibility(&old_rule->cr);
4963 } else {
4964 /* Restore table's rule count. */
4965 table->n_flows--;
4966 }
4967
4968 /* Remove the new rule immediately. It was never visible to lookups. */
4969 if (!classifier_remove(&table->cls, &new_rule->cr)) {
4970 OVS_NOT_REACHED();
4971 }
4972 ofproto_rule_remove__(ofproto, new_rule);
4973 /* The rule was not inserted to the ofproto provider, so we can
4974 * release it without deleting it from the ofproto provider. */
4975 ofproto_rule_unref(new_rule);
4976 }
4977
4978 /* Adds the 'new_rule', replacing the 'old_rule'. */
4979 static void
4980 replace_rule_finish(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4981 const struct flow_mod_requester *req,
4982 struct rule *old_rule, struct rule *new_rule,
4983 struct ovs_list *dead_cookies)
4984 OVS_REQUIRES(ofproto_mutex)
4985 {
4986 bool forward_stats = !(fm->flags & OFPUTIL_FF_RESET_COUNTS);
4987 struct rule *replaced_rule;
4988
4989 replaced_rule = fm->delete_reason != OFPRR_EVICTION ? old_rule : NULL;
4990
4991 /* Insert the new flow to the ofproto provider. A non-NULL 'replaced_rule'
4992 * is a duplicate rule the 'new_rule' is replacing. The provider should
4993 * link the stats from the old rule to the new one if 'forward_stats' is
4994 * 'true'. The 'replaced_rule' will be deleted right after this call. */
4995 ofproto->ofproto_class->rule_insert(new_rule, replaced_rule,
4996 forward_stats);
4997 learned_cookies_inc(ofproto, rule_get_actions(new_rule));
4998
4999 if (old_rule) {
5000 const struct rule_actions *old_actions = rule_get_actions(old_rule);
5001
5002 learned_cookies_dec(ofproto, old_actions, dead_cookies);
5003
5004 if (replaced_rule) {
5005 enum nx_flow_update_event event = fm->command == OFPFC_ADD
5006 ? NXFME_ADDED : NXFME_MODIFIED;
5007
5008 bool change_cookie = (fm->modify_cookie
5009 && fm->new_cookie != OVS_BE64_MAX
5010 && fm->new_cookie != old_rule->flow_cookie);
5011
5012 bool change_actions = !ofpacts_equal(fm->ofpacts,
5013 fm->ofpacts_len,
5014 old_actions->ofpacts,
5015 old_actions->ofpacts_len);
5016
5017 if (event != NXFME_MODIFIED || change_actions || change_cookie) {
5018 ofmonitor_report(ofproto->connmgr, new_rule, event, 0,
5019 req ? req->ofconn : NULL,
5020 req ? req->request->xid : 0,
5021 change_actions ? old_actions : NULL);
5022 }
5023 } else {
5024 /* XXX: This is slight duplication with delete_flows_finish__() */
5025
5026 old_rule->removed_reason = OFPRR_EVICTION;
5027
5028 ofmonitor_report(ofproto->connmgr, old_rule, NXFME_DELETED,
5029 OFPRR_EVICTION,
5030 req ? req->ofconn : NULL,
5031 req ? req->request->xid : 0, NULL);
5032 }
5033 }
5034 }
5035
5036 static enum ofperr
5037 modify_flows_start__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5038 OVS_REQUIRES(ofproto_mutex)
5039 {
5040 struct ofputil_flow_mod *fm = &ofm->fm;
5041 struct rule_collection *old_rules = &ofm->old_rules;
5042 struct rule_collection *new_rules = &ofm->new_rules;
5043 enum ofperr error;
5044
5045 rule_collection_init(new_rules);
5046
5047 if (old_rules->n > 0) {
5048 struct cls_conjunction *conjs;
5049 size_t n_conjs;
5050 size_t i;
5051
5052 /* Create a new 'modified' rule for each old rule. */
5053 for (i = 0; i < old_rules->n; i++) {
5054 struct rule *old_rule = old_rules->rules[i];
5055 struct rule *new_rule;
5056 struct cls_rule cr;
5057
5058 cls_rule_clone(&cr, &old_rule->cr);
5059 error = replace_rule_create(ofproto, fm, &cr, old_rule->table_id,
5060 old_rule, &new_rule);
5061 if (!error) {
5062 rule_collection_add(new_rules, new_rule);
5063 } else {
5064 rule_collection_unref(new_rules);
5065 rule_collection_destroy(new_rules);
5066 return error;
5067 }
5068 }
5069 ovs_assert(new_rules->n == old_rules->n);
5070
5071 get_conjunctions(fm, &conjs, &n_conjs);
5072 for (i = 0; i < old_rules->n; i++) {
5073 replace_rule_start(ofproto, ofm->version, old_rules->rules[i],
5074 new_rules->rules[i], conjs, n_conjs);
5075 }
5076 free(conjs);
5077 } else if (!(fm->cookie_mask != htonll(0)
5078 || fm->new_cookie == OVS_BE64_MAX)) {
5079 /* No match, add a new flow. */
5080 error = add_flow_start(ofproto, ofm);
5081 if (!error) {
5082 ovs_assert(fm->delete_reason == OFPRR_EVICTION
5083 || !old_rules->rules[0]);
5084 }
5085 new_rules->n = 1;
5086 } else {
5087 error = 0;
5088 }
5089
5090 return error;
5091 }
5092
5093 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
5094 * failure.
5095 *
5096 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
5097 * if any. */
5098 static enum ofperr
5099 modify_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5100 OVS_REQUIRES(ofproto_mutex)
5101 {
5102 struct ofputil_flow_mod *fm = &ofm->fm;
5103 struct rule_collection *old_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, OVS_VERSION_MAX,
5108 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG_ANY);
5109 rule_criteria_require_rw(&criteria,
5110 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5111 error = collect_rules_loose(ofproto, &criteria, old_rules);
5112 rule_criteria_destroy(&criteria);
5113
5114 if (!error) {
5115 error = modify_flows_start__(ofproto, ofm);
5116 }
5117
5118 if (error) {
5119 rule_collection_destroy(old_rules);
5120 }
5121 return error;
5122 }
5123
5124 static void
5125 modify_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5126 OVS_REQUIRES(ofproto_mutex)
5127 {
5128 struct rule_collection *old_rules = &ofm->old_rules;
5129 struct rule_collection *new_rules = &ofm->new_rules;
5130
5131 /* Old rules were not changed yet, only need to revert new rules. */
5132 if (old_rules->n == 0 && new_rules->n == 1) {
5133 add_flow_revert(ofproto, ofm);
5134 } else if (old_rules->n > 0) {
5135 for (size_t i = 0; i < old_rules->n; i++) {
5136 replace_rule_revert(ofproto, old_rules->rules[i],
5137 new_rules->rules[i]);
5138 }
5139 rule_collection_destroy(new_rules);
5140 rule_collection_destroy(old_rules);
5141 }
5142 }
5143
5144 static void
5145 modify_flows_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5146 const struct flow_mod_requester *req)
5147 OVS_REQUIRES(ofproto_mutex)
5148 {
5149 struct ofputil_flow_mod *fm = &ofm->fm;
5150 struct rule_collection *old_rules = &ofm->old_rules;
5151 struct rule_collection *new_rules = &ofm->new_rules;
5152
5153 if (old_rules->n == 0 && new_rules->n == 1) {
5154 add_flow_finish(ofproto, ofm, req);
5155 } else if (old_rules->n > 0) {
5156 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5157
5158 ovs_assert(new_rules->n == old_rules->n);
5159
5160 for (size_t i = 0; i < old_rules->n; i++) {
5161 replace_rule_finish(ofproto, fm, req, old_rules->rules[i],
5162 new_rules->rules[i], &dead_cookies);
5163 }
5164 learned_cookies_flush(ofproto, &dead_cookies);
5165 rule_collection_remove_postponed(old_rules);
5166
5167 send_buffered_packet(req, fm->buffer_id, new_rules->rules[0]);
5168 rule_collection_destroy(new_rules);
5169 }
5170 }
5171
5172 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
5173 * code on failure. */
5174 static enum ofperr
5175 modify_flow_start_strict(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5176 OVS_REQUIRES(ofproto_mutex)
5177 {
5178 struct ofputil_flow_mod *fm = &ofm->fm;
5179 struct rule_collection *old_rules = &ofm->old_rules;
5180 struct rule_criteria criteria;
5181 enum ofperr error;
5182
5183 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5184 OVS_VERSION_MAX, fm->cookie, fm->cookie_mask, OFPP_ANY,
5185 OFPG_ANY);
5186 rule_criteria_require_rw(&criteria,
5187 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5188 error = collect_rules_strict(ofproto, &criteria, old_rules);
5189 rule_criteria_destroy(&criteria);
5190
5191 if (!error) {
5192 /* collect_rules_strict() can return max 1 rule. */
5193 error = modify_flows_start__(ofproto, ofm);
5194 }
5195
5196 if (error) {
5197 rule_collection_destroy(old_rules);
5198 }
5199 return error;
5200 }
5201 \f
5202 /* OFPFC_DELETE implementation. */
5203
5204 static void
5205 delete_flows_start__(struct ofproto *ofproto, ovs_version_t version,
5206 const struct rule_collection *rules)
5207 OVS_REQUIRES(ofproto_mutex)
5208 {
5209 for (size_t i = 0; i < rules->n; i++) {
5210 struct rule *rule = rules->rules[i];
5211 struct oftable *table = &ofproto->tables[rule->table_id];
5212
5213 table->n_flows--;
5214 cls_rule_make_invisible_in_version(&rule->cr, version);
5215
5216 /* Remove rule from ofproto data structures. */
5217 ofproto_rule_remove__(ofproto, rule);
5218 }
5219 }
5220
5221 static void
5222 delete_flows_finish__(struct ofproto *ofproto,
5223 struct rule_collection *rules,
5224 enum ofp_flow_removed_reason reason,
5225 const struct flow_mod_requester *req)
5226 OVS_REQUIRES(ofproto_mutex)
5227 {
5228 if (rules->n) {
5229 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5230
5231 for (size_t i = 0; i < rules->n; i++) {
5232 struct rule *rule = rules->rules[i];
5233
5234 /* This value will be used to send the flow removed message right
5235 * before the rule is actually destroyed. */
5236 rule->removed_reason = reason;
5237
5238 ofmonitor_report(ofproto->connmgr, rule, NXFME_DELETED, reason,
5239 req ? req->ofconn : NULL,
5240 req ? req->request->xid : 0, NULL);
5241
5242 /* Send Vacancy Event for OF1.4+. */
5243 send_table_status(ofproto, rule->table_id);
5244
5245 learned_cookies_dec(ofproto, rule_get_actions(rule),
5246 &dead_cookies);
5247 }
5248 rule_collection_remove_postponed(rules);
5249
5250 learned_cookies_flush(ofproto, &dead_cookies);
5251 }
5252 }
5253
5254 /* Deletes the rules listed in 'rules'.
5255 * The deleted rules will become invisible to the lookups in the next version.
5256 * Destroys 'rules'. */
5257 static void
5258 delete_flows__(struct rule_collection *rules,
5259 enum ofp_flow_removed_reason reason,
5260 const struct flow_mod_requester *req)
5261 OVS_REQUIRES(ofproto_mutex)
5262 {
5263 if (rules->n) {
5264 struct ofproto *ofproto = rules->rules[0]->ofproto;
5265
5266 delete_flows_start__(ofproto, ofproto->tables_version + 1, rules);
5267 ofproto_bump_tables_version(ofproto);
5268 delete_flows_finish__(ofproto, rules, reason, req);
5269 ofmonitor_flush(ofproto->connmgr);
5270 }
5271 }
5272
5273 /* Implements OFPFC_DELETE. */
5274 static enum ofperr
5275 delete_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5276 OVS_REQUIRES(ofproto_mutex)
5277 {
5278 const struct ofputil_flow_mod *fm = &ofm->fm;
5279 struct rule_collection *rules = &ofm->old_rules;
5280 struct rule_criteria criteria;
5281 enum ofperr error;
5282
5283 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, OVS_VERSION_MAX,
5284 fm->cookie, fm->cookie_mask, fm->out_port,
5285 fm->out_group);
5286 rule_criteria_require_rw(&criteria,
5287 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5288 error = collect_rules_loose(ofproto, &criteria, rules);
5289 rule_criteria_destroy(&criteria);
5290
5291 if (!error) {
5292 delete_flows_start__(ofproto, ofm->version, rules);
5293 }
5294
5295 return error;
5296 }
5297
5298 static void
5299 delete_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5300 OVS_REQUIRES(ofproto_mutex)
5301 {
5302 struct rule_collection *rules = &ofm->old_rules;
5303
5304 for (size_t i = 0; i < rules->n; i++) {
5305 struct rule *rule = rules->rules[i];
5306 struct oftable *table = &ofproto->tables[rule->table_id];
5307
5308 /* Add rule back to ofproto data structures. */
5309 ofproto_rule_insert__(ofproto, rule);
5310
5311 /* Restore table's rule count. */
5312 table->n_flows++;
5313
5314 /* Restore the original visibility of the rule. */
5315 cls_rule_restore_visibility(&rule->cr);
5316 }
5317 rule_collection_destroy(rules);
5318 }
5319
5320 static void
5321 delete_flows_finish(struct ofproto *ofproto,
5322 struct ofproto_flow_mod *ofm,
5323 const struct flow_mod_requester *req)
5324 OVS_REQUIRES(ofproto_mutex)
5325 {
5326 delete_flows_finish__(ofproto, &ofm->old_rules, ofm->fm.delete_reason,
5327 req);
5328 }
5329
5330 /* Implements OFPFC_DELETE_STRICT. */
5331 static enum ofperr
5332 delete_flow_start_strict(struct ofproto *ofproto,
5333 struct ofproto_flow_mod *ofm)
5334 OVS_REQUIRES(ofproto_mutex)
5335 {
5336 const struct ofputil_flow_mod *fm = &ofm->fm;
5337 struct rule_collection *rules = &ofm->old_rules;
5338 struct rule_criteria criteria;
5339 enum ofperr error;
5340
5341 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5342 OVS_VERSION_MAX, fm->cookie, fm->cookie_mask,
5343 fm->out_port, fm->out_group);
5344 rule_criteria_require_rw(&criteria,
5345 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5346 error = collect_rules_strict(ofproto, &criteria, rules);
5347 rule_criteria_destroy(&criteria);
5348
5349 if (!error) {
5350 delete_flows_start__(ofproto, ofm->version, rules);
5351 }
5352
5353 return error;
5354 }
5355
5356 /* This may only be called by rule_destroy_cb()! */
5357 static void
5358 ofproto_rule_send_removed(struct rule *rule)
5359 OVS_EXCLUDED(ofproto_mutex)
5360 {
5361 struct ofputil_flow_removed fr;
5362 long long int used;
5363
5364 minimatch_expand(&rule->cr.match, &fr.match);
5365 fr.priority = rule->cr.priority;
5366
5367 ovs_mutex_lock(&ofproto_mutex);
5368 fr.cookie = rule->flow_cookie;
5369 fr.reason = rule->removed_reason;
5370 fr.table_id = rule->table_id;
5371 calc_duration(rule->created, time_msec(),
5372 &fr.duration_sec, &fr.duration_nsec);
5373 ovs_mutex_lock(&rule->mutex);
5374 fr.idle_timeout = rule->idle_timeout;
5375 fr.hard_timeout = rule->hard_timeout;
5376 ovs_mutex_unlock(&rule->mutex);
5377 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
5378 &fr.byte_count, &used);
5379 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
5380 ovs_mutex_unlock(&ofproto_mutex);
5381 }
5382
5383 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
5384 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
5385 * ofproto.
5386 *
5387 * ofproto implementation ->run() functions should use this function to expire
5388 * OpenFlow flows. */
5389 void
5390 ofproto_rule_expire(struct rule *rule, uint8_t reason)
5391 OVS_REQUIRES(ofproto_mutex)
5392 {
5393 struct rule_collection rules;
5394
5395 rules.rules = rules.stub;
5396 rules.n = 1;
5397 rules.stub[0] = rule;
5398 delete_flows__(&rules, reason, NULL);
5399 }
5400
5401 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5402 * means "infinite". */
5403 static void
5404 reduce_timeout(uint16_t max, uint16_t *timeout)
5405 {
5406 if (max && (!*timeout || *timeout > max)) {
5407 *timeout = max;
5408 }
5409 }
5410
5411 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
5412 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
5413 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
5414 *
5415 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
5416 void
5417 ofproto_rule_reduce_timeouts(struct rule *rule,
5418 uint16_t idle_timeout, uint16_t hard_timeout)
5419 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
5420 {
5421 if (!idle_timeout && !hard_timeout) {
5422 return;
5423 }
5424
5425 ovs_mutex_lock(&ofproto_mutex);
5426 if (ovs_list_is_empty(&rule->expirable)) {
5427 ovs_list_insert(&rule->ofproto->expirable, &rule->expirable);
5428 }
5429 ovs_mutex_unlock(&ofproto_mutex);
5430
5431 ovs_mutex_lock(&rule->mutex);
5432 reduce_timeout(idle_timeout, &rule->idle_timeout);
5433 reduce_timeout(hard_timeout, &rule->hard_timeout);
5434 ovs_mutex_unlock(&rule->mutex);
5435 }
5436 \f
5437 static enum ofperr
5438 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5439 OVS_EXCLUDED(ofproto_mutex)
5440 {
5441 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5442 struct ofproto_flow_mod ofm;
5443 uint64_t ofpacts_stub[1024 / 8];
5444 struct ofpbuf ofpacts;
5445 enum ofperr error;
5446
5447 error = reject_slave_controller(ofconn);
5448 if (error) {
5449 goto exit;
5450 }
5451
5452 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5453 error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn),
5454 &ofpacts,
5455 u16_to_ofp(ofproto->max_ports),
5456 ofproto->n_tables);
5457 if (!error) {
5458 error = ofproto_check_ofpacts(ofproto, ofm.fm.ofpacts,
5459 ofm.fm.ofpacts_len);
5460 }
5461 if (!error) {
5462 struct flow_mod_requester req;
5463
5464 req.ofconn = ofconn;
5465 req.request = oh;
5466 error = handle_flow_mod__(ofproto, &ofm, &req);
5467 }
5468 if (error) {
5469 goto exit_free_ofpacts;
5470 }
5471
5472 ofconn_report_flow_mod(ofconn, ofm.fm.command);
5473
5474 exit_free_ofpacts:
5475 ofpbuf_uninit(&ofpacts);
5476 exit:
5477 return error;
5478 }
5479
5480 static enum ofperr
5481 handle_flow_mod__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5482 const struct flow_mod_requester *req)
5483 OVS_EXCLUDED(ofproto_mutex)
5484 {
5485 enum ofperr error;
5486
5487 ovs_mutex_lock(&ofproto_mutex);
5488 ofm->version = ofproto->tables_version + 1;
5489 error = ofproto_flow_mod_start(ofproto, ofm);
5490 if (!error) {
5491 ofproto_bump_tables_version(ofproto);
5492 ofproto_flow_mod_finish(ofproto, ofm, req);
5493 }
5494 ofmonitor_flush(ofproto->connmgr);
5495 ovs_mutex_unlock(&ofproto_mutex);
5496
5497 run_rule_executes(ofproto);
5498 return error;
5499 }
5500
5501 static enum ofperr
5502 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
5503 {
5504 struct ofputil_role_request request;
5505 struct ofputil_role_request reply;
5506 struct ofpbuf *buf;
5507 enum ofperr error;
5508
5509 error = ofputil_decode_role_message(oh, &request);
5510 if (error) {
5511 return error;
5512 }
5513
5514 if (request.role != OFPCR12_ROLE_NOCHANGE) {
5515 if (request.role != OFPCR12_ROLE_EQUAL
5516 && request.have_generation_id
5517 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
5518 return OFPERR_OFPRRFC_STALE;
5519 }
5520
5521 ofconn_set_role(ofconn, request.role);
5522 }
5523
5524 reply.role = ofconn_get_role(ofconn);
5525 reply.have_generation_id = ofconn_get_master_election_id(
5526 ofconn, &reply.generation_id);
5527 buf = ofputil_encode_role_reply(oh, &reply);
5528 ofconn_send_reply(ofconn, buf);
5529
5530 return 0;
5531 }
5532
5533 static enum ofperr
5534 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
5535 const struct ofp_header *oh)
5536 {
5537 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
5538 enum ofputil_protocol cur, next;
5539
5540 cur = ofconn_get_protocol(ofconn);
5541 next = ofputil_protocol_set_tid(cur, msg->set != 0);
5542 ofconn_set_protocol(ofconn, next);
5543
5544 return 0;
5545 }
5546
5547 static enum ofperr
5548 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
5549 {
5550 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
5551 enum ofputil_protocol cur, next;
5552 enum ofputil_protocol next_base;
5553
5554 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
5555 if (!next_base) {
5556 return OFPERR_OFPBRC_EPERM;
5557 }
5558
5559 cur = ofconn_get_protocol(ofconn);
5560 next = ofputil_protocol_set_base(cur, next_base);
5561 ofconn_set_protocol(ofconn, next);
5562
5563 return 0;
5564 }
5565
5566 static enum ofperr
5567 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
5568 const struct ofp_header *oh)
5569 {
5570 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
5571 uint32_t format;
5572
5573 format = ntohl(msg->format);
5574 if (!ofputil_packet_in_format_is_valid(format)) {
5575 return OFPERR_OFPBRC_EPERM;
5576 }
5577
5578 ofconn_set_packet_in_format(ofconn, format);
5579 return 0;
5580 }
5581
5582 static enum ofperr
5583 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
5584 {
5585 struct ofputil_async_cfg basis = ofconn_get_async_config(ofconn);
5586 struct ofputil_async_cfg ac;
5587 enum ofperr error;
5588
5589 error = ofputil_decode_set_async_config(oh, false, &basis, &ac);
5590 if (error) {
5591 return error;
5592 }
5593
5594 ofconn_set_async_config(ofconn, &ac);
5595 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
5596 !ofconn_get_miss_send_len(ofconn)) {
5597 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
5598 }
5599
5600 return 0;
5601 }
5602
5603 static enum ofperr
5604 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
5605 {
5606 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
5607 ofconn_send_reply(ofconn, ofputil_encode_get_async_reply(oh, &ac));
5608
5609 return 0;
5610 }
5611
5612 static enum ofperr
5613 handle_nxt_set_controller_id(struct ofconn *ofconn,
5614 const struct ofp_header *oh)
5615 {
5616 const struct nx_controller_id *nci = ofpmsg_body(oh);
5617
5618 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
5619 return OFPERR_NXBRC_MUST_BE_ZERO;
5620 }
5621
5622 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
5623 return 0;
5624 }
5625
5626 static enum ofperr
5627 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
5628 {
5629 struct ofpbuf *buf;
5630
5631 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
5632 ? OFPRAW_OFPT10_BARRIER_REPLY
5633 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
5634 ofconn_send_reply(ofconn, buf);
5635 return 0;
5636 }
5637
5638 static void
5639 ofproto_compose_flow_refresh_update(const struct rule *rule,
5640 enum nx_flow_monitor_flags flags,
5641 struct ovs_list *msgs)
5642 OVS_REQUIRES(ofproto_mutex)
5643 {
5644 const struct rule_actions *actions;
5645 struct ofputil_flow_update fu;
5646 struct match match;
5647
5648 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
5649 ? NXFME_ADDED : NXFME_MODIFIED);
5650 fu.reason = 0;
5651 ovs_mutex_lock(&rule->mutex);
5652 fu.idle_timeout = rule->idle_timeout;
5653 fu.hard_timeout = rule->hard_timeout;
5654 ovs_mutex_unlock(&rule->mutex);
5655 fu.table_id = rule->table_id;
5656 fu.cookie = rule->flow_cookie;
5657 minimatch_expand(&rule->cr.match, &match);
5658 fu.match = &match;
5659 fu.priority = rule->cr.priority;
5660
5661 actions = flags & NXFMF_ACTIONS ? rule_get_actions(rule) : NULL;
5662 fu.ofpacts = actions ? actions->ofpacts : NULL;
5663 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
5664
5665 if (ovs_list_is_empty(msgs)) {
5666 ofputil_start_flow_update(msgs);
5667 }
5668 ofputil_append_flow_update(&fu, msgs);
5669 }
5670
5671 void
5672 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
5673 struct ovs_list *msgs)
5674 OVS_REQUIRES(ofproto_mutex)
5675 {
5676 size_t i;
5677
5678 for (i = 0; i < rules->n; i++) {
5679 struct rule *rule = rules->rules[i];
5680 enum nx_flow_monitor_flags flags = rule->monitor_flags;
5681 rule->monitor_flags = 0;
5682
5683 ofproto_compose_flow_refresh_update(rule, flags, msgs);
5684 }
5685 }
5686
5687 static void
5688 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
5689 struct rule *rule, uint64_t seqno,
5690 struct rule_collection *rules)
5691 OVS_REQUIRES(ofproto_mutex)
5692 {
5693 enum nx_flow_monitor_flags update;
5694
5695 if (rule_is_hidden(rule)) {
5696 return;
5697 }
5698
5699 if (!ofproto_rule_has_out_port(rule, m->out_port)) {
5700 return;
5701 }
5702
5703 if (seqno) {
5704 if (rule->add_seqno > seqno) {
5705 update = NXFMF_ADD | NXFMF_MODIFY;
5706 } else if (rule->modify_seqno > seqno) {
5707 update = NXFMF_MODIFY;
5708 } else {
5709 return;
5710 }
5711
5712 if (!(m->flags & update)) {
5713 return;
5714 }
5715 } else {
5716 update = NXFMF_INITIAL;
5717 }
5718
5719 if (!rule->monitor_flags) {
5720 rule_collection_add(rules, rule);
5721 }
5722 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
5723 }
5724
5725 static void
5726 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
5727 uint64_t seqno,
5728 struct rule_collection *rules)
5729 OVS_REQUIRES(ofproto_mutex)
5730 {
5731 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
5732 const struct oftable *table;
5733 struct cls_rule target;
5734
5735 cls_rule_init_from_minimatch(&target, &m->match, 0);
5736 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
5737 struct rule *rule;
5738
5739 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &target, OVS_VERSION_MAX) {
5740 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
5741 }
5742 }
5743 cls_rule_destroy(&target);
5744 }
5745
5746 static void
5747 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
5748 struct rule_collection *rules)
5749 OVS_REQUIRES(ofproto_mutex)
5750 {
5751 if (m->flags & NXFMF_INITIAL) {
5752 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
5753 }
5754 }
5755
5756 void
5757 ofmonitor_collect_resume_rules(struct ofmonitor *m,
5758 uint64_t seqno, struct rule_collection *rules)
5759 OVS_REQUIRES(ofproto_mutex)
5760 {
5761 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
5762 }
5763
5764 static enum ofperr
5765 flow_monitor_delete(struct ofconn *ofconn, uint32_t id)
5766 OVS_REQUIRES(ofproto_mutex)
5767 {
5768 struct ofmonitor *m;
5769 enum ofperr error;
5770
5771 m = ofmonitor_lookup(ofconn, id);
5772 if (m) {
5773 ofmonitor_destroy(m);
5774 error = 0;
5775 } else {
5776 error = OFPERR_OFPMOFC_UNKNOWN_MONITOR;
5777 }
5778
5779 return error;
5780 }
5781
5782 static enum ofperr
5783 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
5784 OVS_EXCLUDED(ofproto_mutex)
5785 {
5786 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5787
5788 struct ofpbuf b = ofpbuf_const_initializer(oh, ntohs(oh->length));
5789
5790 struct ofmonitor **monitors = NULL;
5791 size_t allocated_monitors = 0;
5792 size_t n_monitors = 0;
5793
5794 enum ofperr error;
5795
5796 ovs_mutex_lock(&ofproto_mutex);
5797 for (;;) {
5798 struct ofputil_flow_monitor_request request;
5799 struct ofmonitor *m;
5800 int retval;
5801
5802 retval = ofputil_decode_flow_monitor_request(&request, &b);
5803 if (retval == EOF) {
5804 break;
5805 } else if (retval) {
5806 error = retval;
5807 goto error;
5808 }
5809
5810 if (request.table_id != 0xff
5811 && request.table_id >= ofproto->n_tables) {
5812 error = OFPERR_OFPBRC_BAD_TABLE_ID;
5813 goto error;
5814 }
5815
5816 error = ofmonitor_create(&request, ofconn, &m);
5817 if (error) {
5818 goto error;
5819 }
5820
5821 if (n_monitors >= allocated_monitors) {
5822 monitors = x2nrealloc(monitors, &allocated_monitors,
5823 sizeof *monitors);
5824 }
5825 monitors[n_monitors++] = m;
5826 }
5827
5828 struct rule_collection rules;
5829 rule_collection_init(&rules);
5830 for (size_t i = 0; i < n_monitors; i++) {
5831 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
5832 }
5833
5834 struct ovs_list replies;
5835 ofpmp_init(&replies, oh);
5836 ofmonitor_compose_refresh_updates(&rules, &replies);
5837 ovs_mutex_unlock(&ofproto_mutex);
5838
5839 rule_collection_destroy(&rules);
5840
5841 ofconn_send_replies(ofconn, &replies);
5842 free(monitors);
5843
5844 return 0;
5845
5846 error:
5847 for (size_t i = 0; i < n_monitors; i++) {
5848 ofmonitor_destroy(monitors[i]);
5849 }
5850 free(monitors);
5851 ovs_mutex_unlock(&ofproto_mutex);
5852
5853 return error;
5854 }
5855
5856 static enum ofperr
5857 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
5858 OVS_EXCLUDED(ofproto_mutex)
5859 {
5860 enum ofperr error;
5861 uint32_t id;
5862
5863 id = ofputil_decode_flow_monitor_cancel(oh);
5864
5865 ovs_mutex_lock(&ofproto_mutex);
5866 error = flow_monitor_delete(ofconn, id);
5867 ovs_mutex_unlock(&ofproto_mutex);
5868
5869 return error;
5870 }
5871
5872 /* Meters implementation.
5873 *
5874 * Meter table entry, indexed by the OpenFlow meter_id.
5875 * 'created' is used to compute the duration for meter stats.
5876 * 'list rules' is needed so that we can delete the dependent rules when the
5877 * meter table entry is deleted.
5878 * 'provider_meter_id' is for the provider's private use.
5879 */
5880 struct meter {
5881 long long int created; /* Time created. */
5882 struct ovs_list rules; /* List of "struct rule_dpif"s. */
5883 ofproto_meter_id provider_meter_id;
5884 uint16_t flags; /* Meter flags. */
5885 uint16_t n_bands; /* Number of meter bands. */
5886 struct ofputil_meter_band *bands;
5887 };
5888
5889 /*
5890 * This is used in instruction validation at flow set-up time,
5891 * as flows may not use non-existing meters.
5892 * Return value of UINT32_MAX signifies an invalid meter.
5893 */
5894 static uint32_t
5895 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
5896 {
5897 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
5898 const struct meter *meter = ofproto->meters[of_meter_id];
5899 if (meter) {
5900 return meter->provider_meter_id.uint32;
5901 }
5902 }
5903 return UINT32_MAX;
5904 }
5905
5906 /* Finds the meter invoked by 'rule''s actions and adds 'rule' to the meter's
5907 * list of rules. */
5908 static void
5909 meter_insert_rule(struct rule *rule)
5910 {
5911 const struct rule_actions *a = rule_get_actions(rule);
5912 uint32_t meter_id = ofpacts_get_meter(a->ofpacts, a->ofpacts_len);
5913 struct meter *meter = rule->ofproto->meters[meter_id];
5914
5915 ovs_list_insert(&meter->rules, &rule->meter_list_node);
5916 }
5917
5918 static void
5919 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
5920 {
5921 free(meter->bands);
5922
5923 meter->flags = config->flags;
5924 meter->n_bands = config->n_bands;
5925 meter->bands = xmemdup(config->bands,
5926 config->n_bands * sizeof *meter->bands);
5927 }
5928
5929 static struct meter *
5930 meter_create(const struct ofputil_meter_config *config,
5931 ofproto_meter_id provider_meter_id)
5932 {
5933 struct meter *meter;
5934
5935 meter = xzalloc(sizeof *meter);
5936 meter->provider_meter_id = provider_meter_id;
5937 meter->created = time_msec();
5938 ovs_list_init(&meter->rules);
5939
5940 meter_update(meter, config);
5941
5942 return meter;
5943 }
5944
5945 static void
5946 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5947 OVS_REQUIRES(ofproto_mutex)
5948 {
5949 uint32_t mid;
5950 for (mid = first; mid <= last; ++mid) {
5951 struct meter *meter = ofproto->meters[mid];
5952 if (meter) {
5953 ofproto->meters[mid] = NULL;
5954 ofproto->ofproto_class->meter_del(ofproto,
5955 meter->provider_meter_id);
5956 free(meter->bands);
5957 free(meter);
5958 }
5959 }
5960 }
5961
5962 static enum ofperr
5963 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5964 {
5965 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5966 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5967 enum ofperr error;
5968
5969 if (*meterp) {
5970 return OFPERR_OFPMMFC_METER_EXISTS;
5971 }
5972
5973 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5974 &mm->meter);
5975 if (!error) {
5976 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5977 *meterp = meter_create(&mm->meter, provider_meter_id);
5978 }
5979 return error;
5980 }
5981
5982 static enum ofperr
5983 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5984 {
5985 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5986 enum ofperr error;
5987 uint32_t provider_meter_id;
5988
5989 if (!meter) {
5990 return OFPERR_OFPMMFC_UNKNOWN_METER;
5991 }
5992
5993 provider_meter_id = meter->provider_meter_id.uint32;
5994 error = ofproto->ofproto_class->meter_set(ofproto,
5995 &meter->provider_meter_id,
5996 &mm->meter);
5997 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5998 if (!error) {
5999 meter_update(meter, &mm->meter);
6000 }
6001 return error;
6002 }
6003
6004 static enum ofperr
6005 handle_delete_meter(struct ofconn *ofconn, struct ofputil_meter_mod *mm)
6006 OVS_EXCLUDED(ofproto_mutex)
6007 {
6008 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6009 uint32_t meter_id = mm->meter.meter_id;
6010 struct rule_collection rules;
6011 enum ofperr error = 0;
6012 uint32_t first, last;
6013
6014 if (meter_id == OFPM13_ALL) {
6015 first = 1;
6016 last = ofproto->meter_features.max_meters;
6017 } else {
6018 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
6019 return 0;
6020 }
6021 first = last = meter_id;
6022 }
6023
6024 /* First delete the rules that use this meter. If any of those rules are
6025 * currently being modified, postpone the whole operation until later. */
6026 rule_collection_init(&rules);
6027 ovs_mutex_lock(&ofproto_mutex);
6028 for (meter_id = first; meter_id <= last; ++meter_id) {
6029 struct meter *meter = ofproto->meters[meter_id];
6030 if (meter && !ovs_list_is_empty(&meter->rules)) {
6031 struct rule *rule;
6032
6033 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
6034 rule_collection_add(&rules, rule);
6035 }
6036 }
6037 }
6038 delete_flows__(&rules, OFPRR_METER_DELETE, NULL);
6039
6040 /* Delete the meters. */
6041 meter_delete(ofproto, first, last);
6042
6043 ovs_mutex_unlock(&ofproto_mutex);
6044
6045 return error;
6046 }
6047
6048 static enum ofperr
6049 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6050 {
6051 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6052 struct ofputil_meter_mod mm;
6053 uint64_t bands_stub[256 / 8];
6054 struct ofpbuf bands;
6055 uint32_t meter_id;
6056 enum ofperr error;
6057
6058 error = reject_slave_controller(ofconn);
6059 if (error) {
6060 return error;
6061 }
6062
6063 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
6064
6065 error = ofputil_decode_meter_mod(oh, &mm, &bands);
6066 if (error) {
6067 goto exit_free_bands;
6068 }
6069
6070 meter_id = mm.meter.meter_id;
6071
6072 if (mm.command != OFPMC13_DELETE) {
6073 /* Fails also when meters are not implemented by the provider. */
6074 if (meter_id == 0 || meter_id > OFPM13_MAX) {
6075 error = OFPERR_OFPMMFC_INVALID_METER;
6076 goto exit_free_bands;
6077 } else if (meter_id > ofproto->meter_features.max_meters) {
6078 error = OFPERR_OFPMMFC_OUT_OF_METERS;
6079 goto exit_free_bands;
6080 }
6081 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
6082 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
6083 goto exit_free_bands;
6084 }
6085 }
6086
6087 switch (mm.command) {
6088 case OFPMC13_ADD:
6089 error = handle_add_meter(ofproto, &mm);
6090 break;
6091
6092 case OFPMC13_MODIFY:
6093 error = handle_modify_meter(ofproto, &mm);
6094 break;
6095
6096 case OFPMC13_DELETE:
6097 error = handle_delete_meter(ofconn, &mm);
6098 break;
6099
6100 default:
6101 error = OFPERR_OFPMMFC_BAD_COMMAND;
6102 break;
6103 }
6104
6105 if (!error) {
6106 struct ofputil_requestforward rf;
6107 rf.xid = oh->xid;
6108 rf.reason = OFPRFR_METER_MOD;
6109 rf.meter_mod = &mm;
6110 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6111 }
6112
6113 exit_free_bands:
6114 ofpbuf_uninit(&bands);
6115 return error;
6116 }
6117
6118 static enum ofperr
6119 handle_meter_features_request(struct ofconn *ofconn,
6120 const struct ofp_header *request)
6121 {
6122 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6123 struct ofputil_meter_features features;
6124 struct ofpbuf *b;
6125
6126 if (ofproto->ofproto_class->meter_get_features) {
6127 ofproto->ofproto_class->meter_get_features(ofproto, &features);
6128 } else {
6129 memset(&features, 0, sizeof features);
6130 }
6131 b = ofputil_encode_meter_features_reply(&features, request);
6132
6133 ofconn_send_reply(ofconn, b);
6134 return 0;
6135 }
6136
6137 static enum ofperr
6138 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
6139 enum ofptype type)
6140 {
6141 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6142 struct ovs_list replies;
6143 uint64_t bands_stub[256 / 8];
6144 struct ofpbuf bands;
6145 uint32_t meter_id, first, last;
6146
6147 ofputil_decode_meter_request(request, &meter_id);
6148
6149 if (meter_id == OFPM13_ALL) {
6150 first = 1;
6151 last = ofproto->meter_features.max_meters;
6152 } else {
6153 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
6154 !ofproto->meters[meter_id]) {
6155 return OFPERR_OFPMMFC_UNKNOWN_METER;
6156 }
6157 first = last = meter_id;
6158 }
6159
6160 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
6161 ofpmp_init(&replies, request);
6162
6163 for (meter_id = first; meter_id <= last; ++meter_id) {
6164 struct meter *meter = ofproto->meters[meter_id];
6165 if (!meter) {
6166 continue; /* Skip non-existing meters. */
6167 }
6168 if (type == OFPTYPE_METER_STATS_REQUEST) {
6169 struct ofputil_meter_stats stats;
6170
6171 stats.meter_id = meter_id;
6172
6173 /* Provider sets the packet and byte counts, we do the rest. */
6174 stats.flow_count = ovs_list_size(&meter->rules);
6175 calc_duration(meter->created, time_msec(),
6176 &stats.duration_sec, &stats.duration_nsec);
6177 stats.n_bands = meter->n_bands;
6178 ofpbuf_clear(&bands);
6179 stats.bands
6180 = ofpbuf_put_uninit(&bands,
6181 meter->n_bands * sizeof *stats.bands);
6182
6183 if (!ofproto->ofproto_class->meter_get(ofproto,
6184 meter->provider_meter_id,
6185 &stats)) {
6186 ofputil_append_meter_stats(&replies, &stats);
6187 }
6188 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
6189 struct ofputil_meter_config config;
6190
6191 config.meter_id = meter_id;
6192 config.flags = meter->flags;
6193 config.n_bands = meter->n_bands;
6194 config.bands = meter->bands;
6195 ofputil_append_meter_config(&replies, &config);
6196 }
6197 }
6198
6199 ofconn_send_replies(ofconn, &replies);
6200 ofpbuf_uninit(&bands);
6201 return 0;
6202 }
6203
6204 /* Returned group is RCU protected. */
6205 static struct ofgroup *
6206 ofproto_group_lookup__(const struct ofproto *ofproto, uint32_t group_id)
6207 {
6208 struct ofgroup *group;
6209
6210 CMAP_FOR_EACH_WITH_HASH (group, cmap_node, hash_int(group_id, 0),
6211 &ofproto->groups) {
6212 if (group->group_id == group_id) {
6213 return group;
6214 }
6215 }
6216
6217 return NULL;
6218 }
6219
6220 /* If the group exists, this function increments the groups's reference count.
6221 *
6222 * Make sure to call ofproto_group_unref() after no longer needing to maintain
6223 * a reference to the group. */
6224 struct ofgroup *
6225 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
6226 bool take_ref)
6227 {
6228 struct ofgroup *group;
6229
6230 group = ofproto_group_lookup__(ofproto, group_id);
6231 if (group && take_ref) {
6232 /* Not holding a lock, so it is possible that another thread releases
6233 * the last reference just before we manage to get one. */
6234 return ofproto_group_try_ref(group) ? group : NULL;
6235 }
6236 return group;
6237 }
6238
6239 /* Caller should hold 'ofproto->groups_rwlock' if it is important that the
6240 * group is not removed by someone else. */
6241 static bool
6242 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
6243 {
6244 return ofproto_group_lookup__(ofproto, group_id) != NULL;
6245 }
6246
6247 /* XXX: This is potentially very expensive for large flow tables, and may be
6248 * called in a loop. Maybe explicitly maintain the count of rules referring to
6249 * the group instead?. */
6250 static uint32_t
6251 group_get_ref_count(struct ofgroup *group)
6252 OVS_EXCLUDED(ofproto_mutex)
6253 {
6254 struct ofproto *ofproto = CONST_CAST(struct ofproto *, group->ofproto);
6255 struct rule_criteria criteria;
6256 struct rule_collection rules;
6257 struct match match;
6258 enum ofperr error;
6259 uint32_t count;
6260
6261 match_init_catchall(&match);
6262 rule_criteria_init(&criteria, 0xff, &match, 0, OVS_VERSION_MAX, htonll(0),
6263 htonll(0), OFPP_ANY, group->group_id);
6264 ovs_mutex_lock(&ofproto_mutex);
6265 error = collect_rules_loose(ofproto, &criteria, &rules);
6266 ovs_mutex_unlock(&ofproto_mutex);
6267 rule_criteria_destroy(&criteria);
6268
6269 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
6270
6271 rule_collection_destroy(&rules);
6272 return count;
6273 }
6274
6275 static void
6276 append_group_stats(struct ofgroup *group, struct ovs_list *replies)
6277 {
6278 struct ofputil_group_stats ogs;
6279 const struct ofproto *ofproto = group->ofproto;
6280 long long int now = time_msec();
6281 int error;
6282
6283 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
6284
6285 /* Provider sets the packet and byte counts, we do the rest. */
6286 ogs.ref_count = group_get_ref_count(group);
6287 ogs.n_buckets = group->n_buckets;
6288
6289 error = (ofproto->ofproto_class->group_get_stats
6290 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
6291 : EOPNOTSUPP);
6292 if (error) {
6293 ogs.packet_count = UINT64_MAX;
6294 ogs.byte_count = UINT64_MAX;
6295 memset(ogs.bucket_stats, 0xff,
6296 ogs.n_buckets * sizeof *ogs.bucket_stats);
6297 }
6298
6299 ogs.group_id = group->group_id;
6300 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
6301
6302 ofputil_append_group_stats(replies, &ogs);
6303
6304 free(ogs.bucket_stats);
6305 }
6306
6307 static void
6308 handle_group_request(struct ofconn *ofconn,
6309 const struct ofp_header *request, uint32_t group_id,
6310 void (*cb)(struct ofgroup *, struct ovs_list *replies))
6311 {
6312 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6313 struct ofgroup *group;
6314 struct ovs_list replies;
6315
6316 ofpmp_init(&replies, request);
6317 if (group_id == OFPG_ALL) {
6318 /* Must exclude modifications to guarantee iterating all groups. */
6319 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6320 CMAP_FOR_EACH (group, cmap_node, &ofproto->groups) {
6321 cb(group, &replies);
6322 }
6323 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6324 } else {
6325 group = ofproto_group_lookup__(ofproto, group_id);
6326 if (group) {
6327 cb(group, &replies);
6328 }
6329 }
6330 ofconn_send_replies(ofconn, &replies);
6331 }
6332
6333 static enum ofperr
6334 handle_group_stats_request(struct ofconn *ofconn,
6335 const struct ofp_header *request)
6336 {
6337 uint32_t group_id;
6338 enum ofperr error;
6339
6340 error = ofputil_decode_group_stats_request(request, &group_id);
6341 if (error) {
6342 return error;
6343 }
6344
6345 handle_group_request(ofconn, request, group_id, append_group_stats);
6346 return 0;
6347 }
6348
6349 static void
6350 append_group_desc(struct ofgroup *group, struct ovs_list *replies)
6351 {
6352 struct ofputil_group_desc gds;
6353
6354 gds.group_id = group->group_id;
6355 gds.type = group->type;
6356 gds.props = group->props;
6357
6358 ofputil_append_group_desc_reply(&gds, &group->buckets, replies);
6359 }
6360
6361 static enum ofperr
6362 handle_group_desc_stats_request(struct ofconn *ofconn,
6363 const struct ofp_header *request)
6364 {
6365 handle_group_request(ofconn, request,
6366 ofputil_decode_group_desc_request(request),
6367 append_group_desc);
6368 return 0;
6369 }
6370
6371 static enum ofperr
6372 handle_group_features_stats_request(struct ofconn *ofconn,
6373 const struct ofp_header *request)
6374 {
6375 struct ofproto *p = ofconn_get_ofproto(ofconn);
6376 struct ofpbuf *msg;
6377
6378 msg = ofputil_encode_group_features_reply(&p->ogf, request);
6379 if (msg) {
6380 ofconn_send_reply(ofconn, msg);
6381 }
6382
6383 return 0;
6384 }
6385
6386 static void
6387 put_queue_get_config_reply(struct ofport *port, uint32_t queue,
6388 struct ovs_list *replies)
6389 {
6390 struct ofputil_queue_config qc;
6391
6392 /* None of the existing queues have compatible properties, so we hard-code
6393 * omitting min_rate and max_rate. */
6394 qc.port = port->ofp_port;
6395 qc.queue = queue;
6396 qc.min_rate = UINT16_MAX;
6397 qc.max_rate = UINT16_MAX;
6398 ofputil_append_queue_get_config_reply(&qc, replies);
6399 }
6400
6401 static int
6402 handle_queue_get_config_request_for_port(struct ofport *port, uint32_t queue,
6403 struct ovs_list *replies)
6404 {
6405 struct smap details = SMAP_INITIALIZER(&details);
6406 if (queue != OFPQ_ALL) {
6407 int error = netdev_get_queue(port->netdev, queue, &details);
6408 switch (error) {
6409 case 0:
6410 put_queue_get_config_reply(port, queue, replies);
6411 break;
6412 case EOPNOTSUPP:
6413 case EINVAL:
6414 return OFPERR_OFPQOFC_BAD_QUEUE;
6415 default:
6416 return OFPERR_NXQOFC_QUEUE_ERROR;
6417 }
6418 } else {
6419 struct netdev_queue_dump queue_dump;
6420 uint32_t queue_id;
6421
6422 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump,
6423 port->netdev) {
6424 put_queue_get_config_reply(port, queue_id, replies);
6425 }
6426 }
6427 smap_destroy(&details);
6428 return 0;
6429 }
6430
6431 static enum ofperr
6432 handle_queue_get_config_request(struct ofconn *ofconn,
6433 const struct ofp_header *oh)
6434 {
6435 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6436 struct ovs_list replies;
6437 struct ofport *port;
6438 ofp_port_t req_port;
6439 uint32_t req_queue;
6440 enum ofperr error;
6441
6442 error = ofputil_decode_queue_get_config_request(oh, &req_port, &req_queue);
6443 if (error) {
6444 return error;
6445 }
6446
6447 ofputil_start_queue_get_config_reply(oh, &replies);
6448 if (req_port == OFPP_ANY) {
6449 error = OFPERR_OFPQOFC_BAD_QUEUE;
6450 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
6451 if (!handle_queue_get_config_request_for_port(port, req_queue,
6452 &replies)) {
6453 error = 0;
6454 }
6455 }
6456 } else {
6457 port = ofproto_get_port(ofproto, req_port);
6458 error = (port
6459 ? handle_queue_get_config_request_for_port(port, req_queue,
6460 &replies)
6461 : OFPERR_OFPQOFC_BAD_PORT);
6462 }
6463 if (!error) {
6464 ofconn_send_replies(ofconn, &replies);
6465 } else {
6466 ofpbuf_list_delete(&replies);
6467 }
6468
6469 return error;
6470 }
6471
6472 static enum ofperr
6473 init_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm,
6474 struct ofgroup **ofgroup)
6475 {
6476 enum ofperr error;
6477 const long long int now = time_msec();
6478
6479 if (gm->group_id > OFPG_MAX) {
6480 return OFPERR_OFPGMFC_INVALID_GROUP;
6481 }
6482 if (gm->type > OFPGT11_FF) {
6483 return OFPERR_OFPGMFC_BAD_TYPE;
6484 }
6485
6486 *ofgroup = ofproto->ofproto_class->group_alloc();
6487 if (!*ofgroup) {
6488 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
6489 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
6490 }
6491
6492 (*ofgroup)->ofproto = ofproto;
6493 *CONST_CAST(uint32_t *, &((*ofgroup)->group_id)) = gm->group_id;
6494 *CONST_CAST(enum ofp11_group_type *, &(*ofgroup)->type) = gm->type;
6495 *CONST_CAST(long long int *, &((*ofgroup)->created)) = now;
6496 *CONST_CAST(long long int *, &((*ofgroup)->modified)) = now;
6497 ovs_refcount_init(&(*ofgroup)->ref_count);
6498
6499 ovs_list_init(CONST_CAST(struct ovs_list *, &(*ofgroup)->buckets));
6500 ofputil_bucket_clone_list(CONST_CAST(struct ovs_list *,
6501 &(*ofgroup)->buckets),
6502 &gm->buckets, NULL);
6503
6504 *CONST_CAST(uint32_t *, &(*ofgroup)->n_buckets) =
6505 ovs_list_size(&(*ofgroup)->buckets);
6506
6507 memcpy(CONST_CAST(struct ofputil_group_props *, &(*ofgroup)->props),
6508 &gm->props, sizeof (struct ofputil_group_props));
6509
6510 /* Construct called BEFORE any locks are held. */
6511 error = ofproto->ofproto_class->group_construct(*ofgroup);
6512 if (error) {
6513 ofputil_bucket_list_destroy(CONST_CAST(struct ovs_list *,
6514 &(*ofgroup)->buckets));
6515 ofproto->ofproto_class->group_dealloc(*ofgroup);
6516 }
6517 return error;
6518 }
6519
6520 /* Implements the OFPGC11_ADD operation specified by 'gm', adding a group to
6521 * 'ofproto''s group table. Returns 0 on success or an OpenFlow error code on
6522 * failure. */
6523 static enum ofperr
6524 add_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6525 {
6526 struct ofgroup *ofgroup;
6527 enum ofperr error;
6528
6529 /* Allocate new group and initialize it. */
6530 error = init_group(ofproto, gm, &ofgroup);
6531 if (error) {
6532 return error;
6533 }
6534
6535 /* We take the mutex as late as possible to minimize the time we jam any
6536 * other threads: No visible state changes before acquiring the lock. */
6537 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6538
6539 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6540 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6541 goto unlock_out;
6542 }
6543
6544 if (ofproto_group_exists(ofproto, gm->group_id)) {
6545 error = OFPERR_OFPGMFC_GROUP_EXISTS;
6546 goto unlock_out;
6547 }
6548
6549 /* Insert new group. */
6550 cmap_insert(&ofproto->groups, &ofgroup->cmap_node,
6551 hash_int(ofgroup->group_id, 0));
6552 ofproto->n_groups[ofgroup->type]++;
6553
6554 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6555 return 0;
6556
6557 unlock_out:
6558 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6559 group_destroy_cb(ofgroup);
6560 return error;
6561 }
6562
6563 /* Adds all of the buckets from 'ofgroup' to 'new_ofgroup'. The buckets
6564 * already in 'new_ofgroup' will be placed just after the (copy of the) bucket
6565 * in 'ofgroup' with bucket ID 'command_bucket_id'. Special
6566 * 'command_bucket_id' values OFPG15_BUCKET_FIRST and OFPG15_BUCKET_LAST are
6567 * also honored. */
6568 static enum ofperr
6569 copy_buckets_for_insert_bucket(const struct ofgroup *ofgroup,
6570 struct ofgroup *new_ofgroup,
6571 uint32_t command_bucket_id)
6572 {
6573 struct ofputil_bucket *last = NULL;
6574
6575 if (command_bucket_id <= OFPG15_BUCKET_MAX) {
6576 /* Check here to ensure that a bucket corresponding to
6577 * command_bucket_id exists in the old bucket list.
6578 *
6579 * The subsequent search of below of new_ofgroup covers
6580 * both buckets in the old bucket list and buckets added
6581 * by the insert buckets group mod message this function processes. */
6582 if (!ofputil_bucket_find(&ofgroup->buckets, command_bucket_id)) {
6583 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6584 }
6585
6586 if (!ovs_list_is_empty(&new_ofgroup->buckets)) {
6587 last = ofputil_bucket_list_back(&new_ofgroup->buckets);
6588 }
6589 }
6590
6591 ofputil_bucket_clone_list(CONST_CAST(struct ovs_list *,
6592 &new_ofgroup->buckets),
6593 &ofgroup->buckets, NULL);
6594
6595 if (ofputil_bucket_check_duplicate_id(&new_ofgroup->buckets)) {
6596 VLOG_INFO_RL(&rl, "Duplicate bucket id");
6597 return OFPERR_OFPGMFC_BUCKET_EXISTS;
6598 }
6599
6600 /* Rearrange list according to command_bucket_id */
6601 if (command_bucket_id == OFPG15_BUCKET_LAST) {
6602 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6603 struct ofputil_bucket *new_first;
6604 const struct ofputil_bucket *first;
6605
6606 first = ofputil_bucket_list_front(&ofgroup->buckets);
6607 new_first = ofputil_bucket_find(&new_ofgroup->buckets,
6608 first->bucket_id);
6609
6610 ovs_list_splice(new_ofgroup->buckets.next, &new_first->list_node,
6611 CONST_CAST(struct ovs_list *,
6612 &new_ofgroup->buckets));
6613 }
6614 } else if (command_bucket_id <= OFPG15_BUCKET_MAX && last) {
6615 struct ofputil_bucket *after;
6616
6617 /* Presence of bucket is checked above so after should never be NULL */
6618 after = ofputil_bucket_find(&new_ofgroup->buckets, command_bucket_id);
6619
6620 ovs_list_splice(after->list_node.next, new_ofgroup->buckets.next,
6621 last->list_node.next);
6622 }
6623
6624 return 0;
6625 }
6626
6627 /* Appends all of the a copy of all the buckets from 'ofgroup' to 'new_ofgroup'
6628 * with the exception of the bucket whose bucket id is 'command_bucket_id'.
6629 * Special 'command_bucket_id' values OFPG15_BUCKET_FIRST, OFPG15_BUCKET_LAST
6630 * and OFPG15_BUCKET_ALL are also honored. */
6631 static enum ofperr
6632 copy_buckets_for_remove_bucket(const struct ofgroup *ofgroup,
6633 struct ofgroup *new_ofgroup,
6634 uint32_t command_bucket_id)
6635 {
6636 const struct ofputil_bucket *skip = NULL;
6637
6638 if (command_bucket_id == OFPG15_BUCKET_ALL) {
6639 return 0;
6640 }
6641
6642 if (command_bucket_id == OFPG15_BUCKET_FIRST) {
6643 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6644 skip = ofputil_bucket_list_front(&ofgroup->buckets);
6645 }
6646 } else if (command_bucket_id == OFPG15_BUCKET_LAST) {
6647 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6648 skip = ofputil_bucket_list_back(&ofgroup->buckets);
6649 }
6650 } else {
6651 skip = ofputil_bucket_find(&ofgroup->buckets, command_bucket_id);
6652 if (!skip) {
6653 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6654 }
6655 }
6656
6657 ofputil_bucket_clone_list(CONST_CAST(struct ovs_list *,
6658 &new_ofgroup->buckets),
6659 &ofgroup->buckets, skip);
6660
6661 return 0;
6662 }
6663
6664 /* Implements OFPGC11_MODIFY, OFPGC15_INSERT_BUCKET and
6665 * OFPGC15_REMOVE_BUCKET. Returns 0 on success or an OpenFlow error code
6666 * on failure.
6667 *
6668 * Note that the group is re-created and then replaces the old group in
6669 * ofproto's ofgroup hash map. Thus, the group is never altered while users of
6670 * the xlate module hold a pointer to the group. */
6671 static enum ofperr
6672 modify_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6673 {
6674 struct ofgroup *ofgroup, *new_ofgroup, *retiring;
6675 enum ofperr error;
6676
6677 error = init_group(ofproto, gm, &new_ofgroup);
6678 if (error) {
6679 return error;
6680 }
6681
6682 retiring = new_ofgroup;
6683
6684 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6685 ofgroup = ofproto_group_lookup__(ofproto, gm->group_id);
6686 if (!ofgroup) {
6687 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
6688 goto out;
6689 }
6690
6691 /* Ofproto's group write lock is held now. */
6692 if (ofgroup->type != gm->type
6693 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6694 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6695 goto out;
6696 }
6697
6698 /* Manipulate bucket list for bucket commands */
6699 if (gm->command == OFPGC15_INSERT_BUCKET) {
6700 error = copy_buckets_for_insert_bucket(ofgroup, new_ofgroup,
6701 gm->command_bucket_id);
6702 } else if (gm->command == OFPGC15_REMOVE_BUCKET) {
6703 error = copy_buckets_for_remove_bucket(ofgroup, new_ofgroup,
6704 gm->command_bucket_id);
6705 }
6706 if (error) {
6707 goto out;
6708 }
6709
6710 /* The group creation time does not change during modification. */
6711 *CONST_CAST(long long int *, &(new_ofgroup->created)) = ofgroup->created;
6712 *CONST_CAST(long long int *, &(new_ofgroup->modified)) = time_msec();
6713
6714 /* XXX: OK to lose old group's stats? */
6715
6716 error = ofproto->ofproto_class->group_modify(new_ofgroup);
6717 if (error) {
6718 goto out;
6719 }
6720
6721 retiring = ofgroup;
6722 /* Replace ofgroup in ofproto's groups hash map with new_ofgroup. */
6723 cmap_replace(&ofproto->groups, &ofgroup->cmap_node,
6724 &new_ofgroup->cmap_node, hash_int(ofgroup->group_id, 0));
6725
6726 if (ofgroup->type != new_ofgroup->type) {
6727 ofproto->n_groups[ofgroup->type]--;
6728 ofproto->n_groups[new_ofgroup->type]++;
6729 }
6730
6731 out:
6732 ofproto_group_unref(retiring);
6733 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6734 return error;
6735 }
6736
6737 /* Implements the OFPGC11_ADD_OR_MOD command which creates the group when it does not
6738 * exist yet and modifies it otherwise */
6739 static enum ofperr
6740 add_or_modify_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6741 {
6742 enum ofperr error;
6743 bool exists;
6744
6745 exists = ofproto_group_exists(ofproto, gm->group_id);
6746
6747 /* XXX: Race: Should hold groups_rwlock here. */
6748
6749 if (!exists) {
6750 error = add_group(ofproto, gm);
6751 } else {
6752 error = modify_group(ofproto, gm);
6753 }
6754 return error;
6755 }
6756
6757 static void
6758 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
6759 OVS_RELEASES(ofproto->groups_rwlock)
6760 {
6761 struct match match;
6762 struct ofproto_flow_mod ofm;
6763
6764 /* Delete all flow entries containing this group in a group action */
6765 match_init_catchall(&match);
6766 flow_mod_init(&ofm.fm, &match, 0, NULL, 0, OFPFC_DELETE);
6767 ofm.fm.delete_reason = OFPRR_GROUP_DELETE;
6768 ofm.fm.out_group = ofgroup->group_id;
6769 ofm.fm.table_id = OFPTT_ALL;
6770 handle_flow_mod__(ofproto, &ofm, NULL);
6771
6772 cmap_remove(&ofproto->groups, &ofgroup->cmap_node,
6773 hash_int(ofgroup->group_id, 0));
6774 /* No-one can find this group any more, but current users may continue to
6775 * use it. */
6776 ofproto->n_groups[ofgroup->type]--;
6777 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6778 ofproto_group_unref(ofgroup);
6779 }
6780
6781 /* Implements OFPGC11_DELETE. */
6782 static void
6783 delete_group(struct ofproto *ofproto, uint32_t group_id)
6784 {
6785 struct ofgroup *ofgroup;
6786
6787 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6788 if (group_id == OFPG_ALL) {
6789 for (;;) {
6790 struct cmap_node *node = cmap_first(&ofproto->groups);
6791 if (!node) {
6792 break;
6793 }
6794 ofgroup = CONTAINER_OF(node, struct ofgroup, cmap_node);
6795 delete_group__(ofproto, ofgroup); /* Releases the mutex. */
6796 /* Lock for each node separately, so that we will not jam the
6797 * other threads for too long time. */
6798 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6799 }
6800 } else {
6801 CMAP_FOR_EACH_WITH_HASH (ofgroup, cmap_node,
6802 hash_int(group_id, 0), &ofproto->groups) {
6803 if (ofgroup->group_id == group_id) {
6804 delete_group__(ofproto, ofgroup);
6805 return;
6806 }
6807 }
6808 }
6809 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6810 }
6811
6812 /* Delete all groups from 'ofproto'.
6813 *
6814 * This is intended for use within an ofproto provider's 'destruct'
6815 * function. */
6816 void
6817 ofproto_group_delete_all(struct ofproto *ofproto)
6818 {
6819 delete_group(ofproto, OFPG_ALL);
6820 }
6821
6822 static enum ofperr
6823 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6824 {
6825 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6826 struct ofputil_group_mod gm;
6827 enum ofperr error;
6828
6829 error = reject_slave_controller(ofconn);
6830 if (error) {
6831 return error;
6832 }
6833
6834 error = ofputil_decode_group_mod(oh, &gm);
6835 if (error) {
6836 return error;
6837 }
6838
6839 switch (gm.command) {
6840 case OFPGC11_ADD:
6841 error = add_group(ofproto, &gm);
6842 break;
6843
6844 case OFPGC11_MODIFY:
6845 error = modify_group(ofproto, &gm);
6846 break;
6847
6848 case OFPGC11_ADD_OR_MOD:
6849 error = add_or_modify_group(ofproto, &gm);
6850 break;
6851
6852 case OFPGC11_DELETE:
6853 delete_group(ofproto, gm.group_id);
6854 error = 0;
6855 break;
6856
6857 case OFPGC15_INSERT_BUCKET:
6858 error = modify_group(ofproto, &gm);
6859 break;
6860
6861 case OFPGC15_REMOVE_BUCKET:
6862 error = modify_group(ofproto, &gm);
6863 break;
6864
6865 default:
6866 if (gm.command > OFPGC11_DELETE) {
6867 VLOG_INFO_RL(&rl, "%s: Invalid group_mod command type %d",
6868 ofproto->name, gm.command);
6869 }
6870 error = OFPERR_OFPGMFC_BAD_COMMAND;
6871 }
6872
6873 if (!error) {
6874 struct ofputil_requestforward rf;
6875 rf.xid = oh->xid;
6876 rf.reason = OFPRFR_GROUP_MOD;
6877 rf.group_mod = &gm;
6878 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6879 }
6880 ofputil_bucket_list_destroy(&gm.buckets);
6881
6882 return error;
6883 }
6884
6885 enum ofputil_table_miss
6886 ofproto_table_get_miss_config(const struct ofproto *ofproto, uint8_t table_id)
6887 {
6888 enum ofputil_table_miss miss;
6889
6890 atomic_read_relaxed(&ofproto->tables[table_id].miss_config, &miss);
6891 return miss;
6892 }
6893
6894 static void
6895 table_mod__(struct oftable *oftable,
6896 const struct ofputil_table_mod *tm)
6897 {
6898 if (tm->miss == OFPUTIL_TABLE_MISS_DEFAULT) {
6899 /* This is how an OFPT_TABLE_MOD decodes if it doesn't specify any
6900 * table-miss configuration (because the protocol used doesn't have
6901 * such a concept), so there's nothing to do. */
6902 } else {
6903 atomic_store_relaxed(&oftable->miss_config, tm->miss);
6904 }
6905
6906 unsigned int new_eviction = oftable->eviction;
6907 if (tm->eviction == OFPUTIL_TABLE_EVICTION_ON) {
6908 new_eviction |= EVICTION_OPENFLOW;
6909 } else if (tm->eviction == OFPUTIL_TABLE_EVICTION_OFF) {
6910 new_eviction &= ~EVICTION_OPENFLOW;
6911 }
6912
6913 if (new_eviction != oftable->eviction) {
6914 ovs_mutex_lock(&ofproto_mutex);
6915 oftable_configure_eviction(oftable, new_eviction,
6916 oftable->eviction_fields,
6917 oftable->n_eviction_fields);
6918 ovs_mutex_unlock(&ofproto_mutex);
6919 }
6920
6921 if (tm->vacancy != OFPUTIL_TABLE_VACANCY_DEFAULT) {
6922 ovs_mutex_lock(&ofproto_mutex);
6923 oftable->vacancy_down = tm->table_vacancy.vacancy_down;
6924 oftable->vacancy_up = tm->table_vacancy.vacancy_up;
6925 if (tm->vacancy == OFPUTIL_TABLE_VACANCY_OFF) {
6926 oftable->vacancy_event = 0;
6927 } else if (!oftable->vacancy_event) {
6928 uint8_t vacancy = oftable_vacancy(oftable);
6929 oftable->vacancy_event = (vacancy < oftable->vacancy_up
6930 ? OFPTR_VACANCY_UP
6931 : OFPTR_VACANCY_DOWN);
6932 }
6933 ovs_mutex_unlock(&ofproto_mutex);
6934 }
6935 }
6936
6937 static enum ofperr
6938 table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
6939 {
6940 if (!check_table_id(ofproto, tm->table_id)) {
6941 return OFPERR_OFPTMFC_BAD_TABLE;
6942 }
6943
6944 /* Don't allow the eviction flags to be changed (except to the only fixed
6945 * value that OVS supports). OF1.4 says this is normal: "The
6946 * OFPTMPT_EVICTION property usually cannot be modified using a
6947 * OFP_TABLE_MOD request, because the eviction mechanism is switch
6948 * defined". */
6949 if (tm->eviction_flags != UINT32_MAX
6950 && tm->eviction_flags != OFPROTO_EVICTION_FLAGS) {
6951 return OFPERR_OFPTMFC_BAD_CONFIG;
6952 }
6953
6954 if (tm->table_id == OFPTT_ALL) {
6955 struct oftable *oftable;
6956 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6957 if (!(oftable->flags & (OFTABLE_HIDDEN | OFTABLE_READONLY))) {
6958 table_mod__(oftable, tm);
6959 }
6960 }
6961 } else {
6962 struct oftable *oftable = &ofproto->tables[tm->table_id];
6963 if (oftable->flags & OFTABLE_READONLY) {
6964 return OFPERR_OFPTMFC_EPERM;
6965 }
6966 table_mod__(oftable, tm);
6967 }
6968
6969 return 0;
6970 }
6971
6972 static enum ofperr
6973 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6974 {
6975 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6976 struct ofputil_table_mod tm;
6977 enum ofperr error;
6978
6979 error = reject_slave_controller(ofconn);
6980 if (error) {
6981 return error;
6982 }
6983
6984 error = ofputil_decode_table_mod(oh, &tm);
6985 if (error) {
6986 return error;
6987 }
6988
6989 return table_mod(ofproto, &tm);
6990 }
6991
6992 static enum ofperr
6993 ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6994 OVS_REQUIRES(ofproto_mutex)
6995 {
6996 switch (ofm->fm.command) {
6997 case OFPFC_ADD:
6998 return add_flow_start(ofproto, ofm);
6999 /* , &be->old_rules.stub[0],
7000 &be->new_rules.stub[0]); */
7001 case OFPFC_MODIFY:
7002 return modify_flows_start_loose(ofproto, ofm);
7003 case OFPFC_MODIFY_STRICT:
7004 return modify_flow_start_strict(ofproto, ofm);
7005 case OFPFC_DELETE:
7006 return delete_flows_start_loose(ofproto, ofm);
7007
7008 case OFPFC_DELETE_STRICT:
7009 return delete_flow_start_strict(ofproto, ofm);
7010 }
7011
7012 return OFPERR_OFPFMFC_BAD_COMMAND;
7013 }
7014
7015 static void
7016 ofproto_flow_mod_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
7017 OVS_REQUIRES(ofproto_mutex)
7018 {
7019 switch (ofm->fm.command) {
7020 case OFPFC_ADD:
7021 add_flow_revert(ofproto, ofm);
7022 break;
7023
7024 case OFPFC_MODIFY:
7025 case OFPFC_MODIFY_STRICT:
7026 modify_flows_revert(ofproto, ofm);
7027 break;
7028
7029 case OFPFC_DELETE:
7030 case OFPFC_DELETE_STRICT:
7031 delete_flows_revert(ofproto, ofm);
7032 break;
7033
7034 default:
7035 break;
7036 }
7037 }
7038
7039 static void
7040 ofproto_flow_mod_finish(struct ofproto *ofproto,
7041 struct ofproto_flow_mod *ofm,
7042 const struct flow_mod_requester *req)
7043 OVS_REQUIRES(ofproto_mutex)
7044 {
7045 switch (ofm->fm.command) {
7046 case OFPFC_ADD:
7047 add_flow_finish(ofproto, ofm, req);
7048 break;
7049
7050 case OFPFC_MODIFY:
7051 case OFPFC_MODIFY_STRICT:
7052 modify_flows_finish(ofproto, ofm, req);
7053 break;
7054
7055 case OFPFC_DELETE:
7056 case OFPFC_DELETE_STRICT:
7057 delete_flows_finish(ofproto, ofm, req);
7058 break;
7059
7060 default:
7061 break;
7062 }
7063 }
7064
7065 /* Commit phases (all while locking ofproto_mutex):
7066 *
7067 * 1. Begin: Gather resources and make changes visible in the next version.
7068 * - Mark affected rules for removal in the next version.
7069 * - Create new replacement rules, make visible in the next
7070 * version.
7071 * - Do not send any events or notifications.
7072 *
7073 * 2. Revert: Fail if any errors are found. After this point no errors are
7074 * possible. No visible changes were made, so rollback is minimal (remove
7075 * added invisible rules, restore visibility of rules marked for removal).
7076 *
7077 * 3. Finish: Make the changes visible for lookups. Insert replacement rules to
7078 * the ofproto provider. Remove replaced and deleted rules from ofproto data
7079 * structures, and Schedule postponed removal of deleted rules from the
7080 * classifier. Send notifications, buffered packets, etc.
7081 */
7082 static enum ofperr
7083 do_bundle_commit(struct ofconn *ofconn, uint32_t id, uint16_t flags)
7084 {
7085 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
7086 ovs_version_t version = ofproto->tables_version + 1;
7087 struct ofp_bundle *bundle;
7088 struct ofp_bundle_entry *be;
7089 enum ofperr error;
7090
7091 bundle = ofconn_get_bundle(ofconn, id);
7092
7093 if (!bundle) {
7094 return OFPERR_OFPBFC_BAD_ID;
7095 }
7096 if (bundle->flags != flags) {
7097 error = OFPERR_OFPBFC_BAD_FLAGS;
7098 } else {
7099 bool prev_is_port_mod = false;
7100
7101 error = 0;
7102 ovs_mutex_lock(&ofproto_mutex);
7103
7104 /* 1. Begin. */
7105 LIST_FOR_EACH (be, node, &bundle->msg_list) {
7106 if (be->type == OFPTYPE_PORT_MOD) {
7107 /* Our port mods are not atomic. */
7108 if (flags & OFPBF_ATOMIC) {
7109 error = OFPERR_OFPBFC_MSG_FAILED;
7110 } else {
7111 prev_is_port_mod = true;
7112 error = port_mod_start(ofconn, &be->opm.pm, &be->opm.port);
7113 }
7114 } else if (be->type == OFPTYPE_FLOW_MOD) {
7115 /* Flow mods between port mods are applied as a single
7116 * version, but the versions are published only after
7117 * we know the commit is successful. */
7118 if (prev_is_port_mod) {
7119 ++version;
7120 }
7121 prev_is_port_mod = false;
7122 /* Store the version in which the changes should take
7123 * effect. */
7124 be->ofm.version = version;
7125 error = ofproto_flow_mod_start(ofproto, &be->ofm);
7126 } else {
7127 OVS_NOT_REACHED();
7128 }
7129 if (error) {
7130 break;
7131 }
7132 }
7133
7134 if (error) {
7135 /* Send error referring to the original message. */
7136 if (error) {
7137 ofconn_send_error(ofconn, be->ofp_msg, error);
7138 error = OFPERR_OFPBFC_MSG_FAILED;
7139 }
7140
7141 /* 2. Revert. Undo all the changes made above. */
7142 LIST_FOR_EACH_REVERSE_CONTINUE(be, node, &bundle->msg_list) {
7143 if (be->type == OFPTYPE_FLOW_MOD) {
7144 ofproto_flow_mod_revert(ofproto, &be->ofm);
7145 }
7146 /* Nothing needs to be reverted for a port mod. */
7147 }
7148 } else {
7149 /* 4. Finish. */
7150 LIST_FOR_EACH (be, node, &bundle->msg_list) {
7151 if (be->type == OFPTYPE_FLOW_MOD) {
7152 struct flow_mod_requester req = { ofconn, be->ofp_msg };
7153
7154 /* Bump the lookup version to the one of the current
7155 * message. This makes all the changes in the bundle at
7156 * this version visible to lookups at once. */
7157 if (ofproto->tables_version < be->ofm.version) {
7158 ofproto->tables_version = be->ofm.version;
7159 ofproto->ofproto_class->set_tables_version(
7160 ofproto, ofproto->tables_version);
7161 }
7162
7163 ofproto_flow_mod_finish(ofproto, &be->ofm, &req);
7164 } else if (be->type == OFPTYPE_PORT_MOD) {
7165 /* Perform the actual port mod. This is not atomic, i.e.,
7166 * the effects will be immediately seen by upcall
7167 * processing regardless of the lookup version. It should
7168 * be noted that port configuration changes can originate
7169 * also from OVSDB changes asynchronously to all upcall
7170 * processing. */
7171 port_mod_finish(ofconn, &be->opm.pm, be->opm.port);
7172 }
7173 }
7174 }
7175
7176 ofmonitor_flush(ofproto->connmgr);
7177 ovs_mutex_unlock(&ofproto_mutex);
7178
7179 run_rule_executes(ofproto);
7180 }
7181
7182 /* The bundle is discarded regardless the outcome. */
7183 ofp_bundle_remove__(ofconn, bundle, !error);
7184 return error;
7185 }
7186
7187 static enum ofperr
7188 handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh)
7189 {
7190 struct ofputil_bundle_ctrl_msg bctrl;
7191 struct ofputil_bundle_ctrl_msg reply;
7192 struct ofpbuf *buf;
7193 enum ofperr error;
7194
7195 error = reject_slave_controller(ofconn);
7196 if (error) {
7197 return error;
7198 }
7199
7200 error = ofputil_decode_bundle_ctrl(oh, &bctrl);
7201 if (error) {
7202 return error;
7203 }
7204 reply.flags = 0;
7205 reply.bundle_id = bctrl.bundle_id;
7206
7207 switch (bctrl.type) {
7208 case OFPBCT_OPEN_REQUEST:
7209 error = ofp_bundle_open(ofconn, bctrl.bundle_id, bctrl.flags);
7210 reply.type = OFPBCT_OPEN_REPLY;
7211 break;
7212 case OFPBCT_CLOSE_REQUEST:
7213 error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags);
7214 reply.type = OFPBCT_CLOSE_REPLY;
7215 break;
7216 case OFPBCT_COMMIT_REQUEST:
7217 error = do_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags);
7218 reply.type = OFPBCT_COMMIT_REPLY;
7219 break;
7220 case OFPBCT_DISCARD_REQUEST:
7221 error = ofp_bundle_discard(ofconn, bctrl.bundle_id);
7222 reply.type = OFPBCT_DISCARD_REPLY;
7223 break;
7224
7225 case OFPBCT_OPEN_REPLY:
7226 case OFPBCT_CLOSE_REPLY:
7227 case OFPBCT_COMMIT_REPLY:
7228 case OFPBCT_DISCARD_REPLY:
7229 return OFPERR_OFPBFC_BAD_TYPE;
7230 break;
7231 }
7232
7233 if (!error) {
7234 buf = ofputil_encode_bundle_ctrl_reply(oh, &reply);
7235 ofconn_send_reply(ofconn, buf);
7236 }
7237 return error;
7238 }
7239
7240 static enum ofperr
7241 handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh)
7242 {
7243 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
7244 enum ofperr error;
7245 struct ofputil_bundle_add_msg badd;
7246 struct ofp_bundle_entry *bmsg;
7247 enum ofptype type;
7248
7249 error = reject_slave_controller(ofconn);
7250 if (error) {
7251 return error;
7252 }
7253
7254 error = ofputil_decode_bundle_add(oh, &badd, &type);
7255 if (error) {
7256 return error;
7257 }
7258
7259 bmsg = ofp_bundle_entry_alloc(type, badd.msg);
7260
7261 if (type == OFPTYPE_PORT_MOD) {
7262 error = ofputil_decode_port_mod(badd.msg, &bmsg->opm.pm, false);
7263 } else if (type == OFPTYPE_FLOW_MOD) {
7264 struct ofpbuf ofpacts;
7265 uint64_t ofpacts_stub[1024 / 8];
7266
7267 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
7268 error = ofputil_decode_flow_mod(&bmsg->ofm.fm, badd.msg,
7269 ofconn_get_protocol(ofconn),
7270 &ofpacts,
7271 u16_to_ofp(ofproto->max_ports),
7272 ofproto->n_tables);
7273 /* Move actions to heap. */
7274 bmsg->ofm.fm.ofpacts = ofpbuf_steal_data(&ofpacts);
7275
7276 if (!error && bmsg->ofm.fm.ofpacts_len) {
7277 error = ofproto_check_ofpacts(ofproto, bmsg->ofm.fm.ofpacts,
7278 bmsg->ofm.fm.ofpacts_len);
7279 }
7280 } else {
7281 OVS_NOT_REACHED();
7282 }
7283
7284 if (!error) {
7285 error = ofp_bundle_add_message(ofconn, badd.bundle_id, badd.flags,
7286 bmsg);
7287 }
7288
7289 if (error) {
7290 ofp_bundle_entry_free(bmsg);
7291 }
7292
7293 return error;
7294 }
7295
7296 static enum ofperr
7297 handle_tlv_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
7298 {
7299 struct ofputil_tlv_table_mod ttm;
7300 enum ofperr error;
7301
7302 error = reject_slave_controller(ofconn);
7303 if (error) {
7304 return error;
7305 }
7306
7307 error = ofputil_decode_tlv_table_mod(oh, &ttm);
7308 if (error) {
7309 return error;
7310 }
7311
7312 error = tun_metadata_table_mod(&ttm);
7313
7314 ofputil_uninit_tlv_table(&ttm.mappings);
7315 return error;
7316 }
7317
7318 static enum ofperr
7319 handle_tlv_table_request(struct ofconn *ofconn, const struct ofp_header *oh)
7320 {
7321 struct ofputil_tlv_table_reply ttr;
7322 struct ofpbuf *b;
7323
7324 tun_metadata_table_request(&ttr);
7325 b = ofputil_encode_tlv_table_reply(oh, &ttr);
7326 ofputil_uninit_tlv_table(&ttr.mappings);
7327
7328 ofconn_send_reply(ofconn, b);
7329 return 0;
7330 }
7331
7332 static enum ofperr
7333 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
7334 OVS_EXCLUDED(ofproto_mutex)
7335 {
7336 const struct ofp_header *oh = msg->data;
7337 enum ofptype type;
7338 enum ofperr error;
7339
7340 error = ofptype_decode(&type, oh);
7341 if (error) {
7342 return error;
7343 }
7344 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
7345 && ofpmp_more(oh)) {
7346 /* We have no buffer implementation for multipart requests.
7347 * Report overflow for requests which consists of multiple
7348 * messages. */
7349 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
7350 }
7351
7352 switch (type) {
7353 /* OpenFlow requests. */
7354 case OFPTYPE_ECHO_REQUEST:
7355 return handle_echo_request(ofconn, oh);
7356
7357 case OFPTYPE_FEATURES_REQUEST:
7358 return handle_features_request(ofconn, oh);
7359
7360 case OFPTYPE_GET_CONFIG_REQUEST:
7361 return handle_get_config_request(ofconn, oh);
7362
7363 case OFPTYPE_SET_CONFIG:
7364 return handle_set_config(ofconn, oh);
7365
7366 case OFPTYPE_PACKET_OUT:
7367 return handle_packet_out(ofconn, oh);
7368
7369 case OFPTYPE_PORT_MOD:
7370 return handle_port_mod(ofconn, oh);
7371
7372 case OFPTYPE_FLOW_MOD:
7373 return handle_flow_mod(ofconn, oh);
7374
7375 case OFPTYPE_GROUP_MOD:
7376 return handle_group_mod(ofconn, oh);
7377
7378 case OFPTYPE_TABLE_MOD:
7379 return handle_table_mod(ofconn, oh);
7380
7381 case OFPTYPE_METER_MOD:
7382 return handle_meter_mod(ofconn, oh);
7383
7384 case OFPTYPE_BARRIER_REQUEST:
7385 return handle_barrier_request(ofconn, oh);
7386
7387 case OFPTYPE_ROLE_REQUEST:
7388 return handle_role_request(ofconn, oh);
7389
7390 /* OpenFlow replies. */
7391 case OFPTYPE_ECHO_REPLY:
7392 return 0;
7393
7394 /* Nicira extension requests. */
7395 case OFPTYPE_FLOW_MOD_TABLE_ID:
7396 return handle_nxt_flow_mod_table_id(ofconn, oh);
7397
7398 case OFPTYPE_SET_FLOW_FORMAT:
7399 return handle_nxt_set_flow_format(ofconn, oh);
7400
7401 case OFPTYPE_SET_PACKET_IN_FORMAT:
7402 return handle_nxt_set_packet_in_format(ofconn, oh);
7403
7404 case OFPTYPE_SET_CONTROLLER_ID:
7405 return handle_nxt_set_controller_id(ofconn, oh);
7406
7407 case OFPTYPE_FLOW_AGE:
7408 /* Nothing to do. */
7409 return 0;
7410
7411 case OFPTYPE_FLOW_MONITOR_CANCEL:
7412 return handle_flow_monitor_cancel(ofconn, oh);
7413
7414 case OFPTYPE_SET_ASYNC_CONFIG:
7415 return handle_nxt_set_async_config(ofconn, oh);
7416
7417 case OFPTYPE_GET_ASYNC_REQUEST:
7418 return handle_nxt_get_async_request(ofconn, oh);
7419
7420 case OFPTYPE_NXT_RESUME:
7421 return handle_nxt_resume(ofconn, oh);
7422
7423 /* Statistics requests. */
7424 case OFPTYPE_DESC_STATS_REQUEST:
7425 return handle_desc_stats_request(ofconn, oh);
7426
7427 case OFPTYPE_FLOW_STATS_REQUEST:
7428 return handle_flow_stats_request(ofconn, oh);
7429
7430 case OFPTYPE_AGGREGATE_STATS_REQUEST:
7431 return handle_aggregate_stats_request(ofconn, oh);
7432
7433 case OFPTYPE_TABLE_STATS_REQUEST:
7434 return handle_table_stats_request(ofconn, oh);
7435
7436 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
7437 return handle_table_features_request(ofconn, oh);
7438
7439 case OFPTYPE_TABLE_DESC_REQUEST:
7440 return handle_table_desc_request(ofconn, oh);
7441
7442 case OFPTYPE_PORT_STATS_REQUEST:
7443 return handle_port_stats_request(ofconn, oh);
7444
7445 case OFPTYPE_QUEUE_STATS_REQUEST:
7446 return handle_queue_stats_request(ofconn, oh);
7447
7448 case OFPTYPE_PORT_DESC_STATS_REQUEST:
7449 return handle_port_desc_stats_request(ofconn, oh);
7450
7451 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
7452 return handle_flow_monitor_request(ofconn, oh);
7453
7454 case OFPTYPE_METER_STATS_REQUEST:
7455 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
7456 return handle_meter_request(ofconn, oh, type);
7457
7458 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
7459 return handle_meter_features_request(ofconn, oh);
7460
7461 case OFPTYPE_GROUP_STATS_REQUEST:
7462 return handle_group_stats_request(ofconn, oh);
7463
7464 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
7465 return handle_group_desc_stats_request(ofconn, oh);
7466
7467 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
7468 return handle_group_features_stats_request(ofconn, oh);
7469
7470 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
7471 return handle_queue_get_config_request(ofconn, oh);
7472
7473 case OFPTYPE_BUNDLE_CONTROL:
7474 return handle_bundle_control(ofconn, oh);
7475
7476 case OFPTYPE_BUNDLE_ADD_MESSAGE:
7477 return handle_bundle_add(ofconn, oh);
7478
7479 case OFPTYPE_NXT_TLV_TABLE_MOD:
7480 return handle_tlv_table_mod(ofconn, oh);
7481
7482 case OFPTYPE_NXT_TLV_TABLE_REQUEST:
7483 return handle_tlv_table_request(ofconn, oh);
7484
7485 case OFPTYPE_IPFIX_BRIDGE_STATS_REQUEST:
7486 return handle_ipfix_bridge_stats_request(ofconn, oh);
7487
7488 case OFPTYPE_IPFIX_FLOW_STATS_REQUEST:
7489 return handle_ipfix_flow_stats_request(ofconn, oh);
7490
7491 case OFPTYPE_HELLO:
7492 case OFPTYPE_ERROR:
7493 case OFPTYPE_FEATURES_REPLY:
7494 case OFPTYPE_GET_CONFIG_REPLY:
7495 case OFPTYPE_PACKET_IN:
7496 case OFPTYPE_FLOW_REMOVED:
7497 case OFPTYPE_PORT_STATUS:
7498 case OFPTYPE_BARRIER_REPLY:
7499 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
7500 case OFPTYPE_DESC_STATS_REPLY:
7501 case OFPTYPE_FLOW_STATS_REPLY:
7502 case OFPTYPE_QUEUE_STATS_REPLY:
7503 case OFPTYPE_PORT_STATS_REPLY:
7504 case OFPTYPE_TABLE_STATS_REPLY:
7505 case OFPTYPE_AGGREGATE_STATS_REPLY:
7506 case OFPTYPE_PORT_DESC_STATS_REPLY:
7507 case OFPTYPE_ROLE_REPLY:
7508 case OFPTYPE_FLOW_MONITOR_PAUSED:
7509 case OFPTYPE_FLOW_MONITOR_RESUMED:
7510 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
7511 case OFPTYPE_GET_ASYNC_REPLY:
7512 case OFPTYPE_GROUP_STATS_REPLY:
7513 case OFPTYPE_GROUP_DESC_STATS_REPLY:
7514 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
7515 case OFPTYPE_METER_STATS_REPLY:
7516 case OFPTYPE_METER_CONFIG_STATS_REPLY:
7517 case OFPTYPE_METER_FEATURES_STATS_REPLY:
7518 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
7519 case OFPTYPE_TABLE_DESC_REPLY:
7520 case OFPTYPE_ROLE_STATUS:
7521 case OFPTYPE_REQUESTFORWARD:
7522 case OFPTYPE_TABLE_STATUS:
7523 case OFPTYPE_NXT_TLV_TABLE_REPLY:
7524 case OFPTYPE_IPFIX_BRIDGE_STATS_REPLY:
7525 case OFPTYPE_IPFIX_FLOW_STATS_REPLY:
7526 default:
7527 if (ofpmsg_is_stat_request(oh)) {
7528 return OFPERR_OFPBRC_BAD_STAT;
7529 } else {
7530 return OFPERR_OFPBRC_BAD_TYPE;
7531 }
7532 }
7533 }
7534
7535 static void
7536 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
7537 OVS_EXCLUDED(ofproto_mutex)
7538 {
7539 enum ofperr error = handle_openflow__(ofconn, ofp_msg);
7540
7541 if (error) {
7542 ofconn_send_error(ofconn, ofp_msg->data, error);
7543 }
7544 COVERAGE_INC(ofproto_recv_openflow);
7545 }
7546 \f
7547 /* Asynchronous operations. */
7548
7549 static void
7550 send_buffered_packet(const struct flow_mod_requester *req, uint32_t buffer_id,
7551 struct rule *rule)
7552 OVS_REQUIRES(ofproto_mutex)
7553 {
7554 if (req && req->ofconn && buffer_id != UINT32_MAX) {
7555 struct ofproto *ofproto = ofconn_get_ofproto(req->ofconn);
7556 struct dp_packet *packet;
7557 ofp_port_t in_port;
7558 enum ofperr error;
7559
7560 error = ofconn_pktbuf_retrieve(req->ofconn, buffer_id, &packet,
7561 &in_port);
7562 if (packet) {
7563 struct rule_execute *re;
7564
7565 ofproto_rule_ref(rule);
7566
7567 re = xmalloc(sizeof *re);
7568 re->rule = rule;
7569 re->in_port = in_port;
7570 re->packet = packet;
7571
7572 if (!guarded_list_push_back(&ofproto->rule_executes,
7573 &re->list_node, 1024)) {
7574 ofproto_rule_unref(rule);
7575 dp_packet_delete(re->packet);
7576 free(re);
7577 }
7578 } else {
7579 ofconn_send_error(req->ofconn, req->request, error);
7580 }
7581 }
7582 }
7583 \f
7584 static uint64_t
7585 pick_datapath_id(const struct ofproto *ofproto)
7586 {
7587 const struct ofport *port;
7588
7589 port = ofproto_get_port(ofproto, OFPP_LOCAL);
7590 if (port) {
7591 struct eth_addr ea;
7592 int error;
7593
7594 error = netdev_get_etheraddr(port->netdev, &ea);
7595 if (!error) {
7596 return eth_addr_to_uint64(ea);
7597 }
7598 VLOG_WARN("%s: could not get MAC address for %s (%s)",
7599 ofproto->name, netdev_get_name(port->netdev),
7600 ovs_strerror(error));
7601 }
7602 return ofproto->fallback_dpid;
7603 }
7604
7605 static uint64_t
7606 pick_fallback_dpid(void)
7607 {
7608 struct eth_addr ea;
7609 eth_addr_nicira_random(&ea);
7610 return eth_addr_to_uint64(ea);
7611 }
7612 \f
7613 /* Table overflow policy. */
7614
7615 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
7616 * to NULL if the table is not configured to evict rules or if the table
7617 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
7618 * or with no timeouts are not evictable.) */
7619 static bool
7620 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
7621 OVS_REQUIRES(ofproto_mutex)
7622 {
7623 struct eviction_group *evg;
7624
7625 *rulep = NULL;
7626 if (!table->eviction) {
7627 return false;
7628 }
7629
7630 /* In the common case, the outer and inner loops here will each be entered
7631 * exactly once:
7632 *
7633 * - The inner loop normally "return"s in its first iteration. If the
7634 * eviction group has any evictable rules, then it always returns in
7635 * some iteration.
7636 *
7637 * - The outer loop only iterates more than once if the largest eviction
7638 * group has no evictable rules.
7639 *
7640 * - The outer loop can exit only if table's 'max_flows' is all filled up
7641 * by unevictable rules. */
7642 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
7643 struct rule *rule;
7644
7645 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
7646 *rulep = rule;
7647 return true;
7648 }
7649 }
7650
7651 return false;
7652 }
7653 \f
7654 /* Eviction groups. */
7655
7656 /* Returns the priority to use for an eviction_group that contains 'n_rules'
7657 * rules. The priority contains low-order random bits to ensure that eviction
7658 * groups with the same number of rules are prioritized randomly. */
7659 static uint32_t
7660 eviction_group_priority(size_t n_rules)
7661 {
7662 uint16_t size = MIN(UINT16_MAX, n_rules);
7663 return (size << 16) | random_uint16();
7664 }
7665
7666 /* Updates 'evg', an eviction_group within 'table', following a change that
7667 * adds or removes rules in 'evg'. */
7668 static void
7669 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
7670 OVS_REQUIRES(ofproto_mutex)
7671 {
7672 heap_change(&table->eviction_groups_by_size, &evg->size_node,
7673 eviction_group_priority(heap_count(&evg->rules)));
7674 }
7675
7676 /* Destroys 'evg', an eviction_group within 'table':
7677 *
7678 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
7679 * rules themselves, just removes them from the eviction group.)
7680 *
7681 * - Removes 'evg' from 'table'.
7682 *
7683 * - Frees 'evg'. */
7684 static void
7685 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
7686 OVS_REQUIRES(ofproto_mutex)
7687 {
7688 while (!heap_is_empty(&evg->rules)) {
7689 struct rule *rule;
7690
7691 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
7692 rule->eviction_group = NULL;
7693 }
7694 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
7695 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
7696 heap_destroy(&evg->rules);
7697 free(evg);
7698 }
7699
7700 /* Removes 'rule' from its eviction group, if any. */
7701 static void
7702 eviction_group_remove_rule(struct rule *rule)
7703 OVS_REQUIRES(ofproto_mutex)
7704 {
7705 if (rule->eviction_group) {
7706 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7707 struct eviction_group *evg = rule->eviction_group;
7708
7709 rule->eviction_group = NULL;
7710 heap_remove(&evg->rules, &rule->evg_node);
7711 if (heap_is_empty(&evg->rules)) {
7712 eviction_group_destroy(table, evg);
7713 } else {
7714 eviction_group_resized(table, evg);
7715 }
7716 }
7717 }
7718
7719 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
7720 * returns the hash value. */
7721 static uint32_t
7722 eviction_group_hash_rule(struct rule *rule)
7723 OVS_REQUIRES(ofproto_mutex)
7724 {
7725 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7726 const struct mf_subfield *sf;
7727 struct flow flow;
7728 uint32_t hash;
7729
7730 hash = table->eviction_group_id_basis;
7731 miniflow_expand(rule->cr.match.flow, &flow);
7732 for (sf = table->eviction_fields;
7733 sf < &table->eviction_fields[table->n_eviction_fields];
7734 sf++)
7735 {
7736 if (mf_are_prereqs_ok(sf->field, &flow)) {
7737 union mf_value value;
7738
7739 mf_get_value(sf->field, &flow, &value);
7740 if (sf->ofs) {
7741 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
7742 }
7743 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
7744 unsigned int start = sf->ofs + sf->n_bits;
7745 bitwise_zero(&value, sf->field->n_bytes, start,
7746 sf->field->n_bytes * 8 - start);
7747 }
7748 hash = hash_bytes(&value, sf->field->n_bytes, hash);
7749 } else {
7750 hash = hash_int(hash, 0);
7751 }
7752 }
7753
7754 return hash;
7755 }
7756
7757 /* Returns an eviction group within 'table' with the given 'id', creating one
7758 * if necessary. */
7759 static struct eviction_group *
7760 eviction_group_find(struct oftable *table, uint32_t id)
7761 OVS_REQUIRES(ofproto_mutex)
7762 {
7763 struct eviction_group *evg;
7764
7765 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
7766 return evg;
7767 }
7768
7769 evg = xmalloc(sizeof *evg);
7770 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
7771 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
7772 eviction_group_priority(0));
7773 heap_init(&evg->rules);
7774
7775 return evg;
7776 }
7777
7778 /* Returns an eviction priority for 'rule'. The return value should be
7779 * interpreted so that higher priorities make a rule a more attractive
7780 * candidate for eviction. */
7781 static uint64_t
7782 rule_eviction_priority(struct ofproto *ofproto, struct rule *rule)
7783 OVS_REQUIRES(ofproto_mutex)
7784 {
7785 /* Calculate absolute time when this flow will expire. If it will never
7786 * expire, then return 0 to make it unevictable. */
7787 long long int expiration = LLONG_MAX;
7788 if (rule->hard_timeout) {
7789 /* 'modified' needs protection even when we hold 'ofproto_mutex'. */
7790 ovs_mutex_lock(&rule->mutex);
7791 long long int modified = rule->modified;
7792 ovs_mutex_unlock(&rule->mutex);
7793
7794 expiration = modified + rule->hard_timeout * 1000;
7795 }
7796 if (rule->idle_timeout) {
7797 uint64_t packets, bytes;
7798 long long int used;
7799 long long int idle_expiration;
7800
7801 ofproto->ofproto_class->rule_get_stats(rule, &packets, &bytes, &used);
7802 idle_expiration = used + rule->idle_timeout * 1000;
7803 expiration = MIN(expiration, idle_expiration);
7804 }
7805 if (expiration == LLONG_MAX) {
7806 return 0;
7807 }
7808
7809 /* Calculate the time of expiration as a number of (approximate) seconds
7810 * after program startup.
7811 *
7812 * This should work OK for program runs that last UINT32_MAX seconds or
7813 * less. Therefore, please restart OVS at least once every 136 years. */
7814 uint32_t expiration_ofs = (expiration >> 10) - (time_boot_msec() >> 10);
7815
7816 /* Combine expiration time with OpenFlow "importance" to form a single
7817 * priority value. We want flows with relatively low "importance" to be
7818 * evicted before even considering expiration time, so put "importance" in
7819 * the most significant bits and expiration time in the least significant
7820 * bits.
7821 *
7822 * Small 'priority' should be evicted before those with large 'priority'.
7823 * The caller expects the opposite convention (a large return value being
7824 * more attractive for eviction) so we invert it before returning. */
7825 uint64_t priority = ((uint64_t) rule->importance << 32) + expiration_ofs;
7826 return UINT64_MAX - priority;
7827 }
7828
7829 /* Adds 'rule' to an appropriate eviction group for its oftable's
7830 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
7831 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
7832 * own).
7833 *
7834 * The caller must ensure that 'rule' is not already in an eviction group. */
7835 static void
7836 eviction_group_add_rule(struct rule *rule)
7837 OVS_REQUIRES(ofproto_mutex)
7838 {
7839 struct ofproto *ofproto = rule->ofproto;
7840 struct oftable *table = &ofproto->tables[rule->table_id];
7841 bool has_timeout;
7842
7843 /* Timeouts may be modified only when holding 'ofproto_mutex'. We have it
7844 * so no additional protection is needed. */
7845 has_timeout = rule->hard_timeout || rule->idle_timeout;
7846
7847 if (table->eviction && has_timeout) {
7848 struct eviction_group *evg;
7849
7850 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
7851
7852 rule->eviction_group = evg;
7853 heap_insert(&evg->rules, &rule->evg_node,
7854 rule_eviction_priority(ofproto, rule));
7855 eviction_group_resized(table, evg);
7856 }
7857 }
7858 \f
7859 /* oftables. */
7860
7861 /* Initializes 'table'. */
7862 static void
7863 oftable_init(struct oftable *table)
7864 {
7865 memset(table, 0, sizeof *table);
7866 classifier_init(&table->cls, flow_segment_u64s);
7867 table->max_flows = UINT_MAX;
7868 table->n_flows = 0;
7869 hmap_init(&table->eviction_groups_by_id);
7870 heap_init(&table->eviction_groups_by_size);
7871 atomic_init(&table->miss_config, OFPUTIL_TABLE_MISS_DEFAULT);
7872
7873 classifier_set_prefix_fields(&table->cls, default_prefix_fields,
7874 ARRAY_SIZE(default_prefix_fields));
7875
7876 atomic_init(&table->n_matched, 0);
7877 atomic_init(&table->n_missed, 0);
7878 }
7879
7880 /* Destroys 'table', including its classifier and eviction groups.
7881 *
7882 * The caller is responsible for freeing 'table' itself. */
7883 static void
7884 oftable_destroy(struct oftable *table)
7885 {
7886 ovs_assert(classifier_is_empty(&table->cls));
7887
7888 ovs_mutex_lock(&ofproto_mutex);
7889 oftable_configure_eviction(table, 0, NULL, 0);
7890 ovs_mutex_unlock(&ofproto_mutex);
7891
7892 hmap_destroy(&table->eviction_groups_by_id);
7893 heap_destroy(&table->eviction_groups_by_size);
7894 classifier_destroy(&table->cls);
7895 free(table->name);
7896 }
7897
7898 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
7899 * string, then 'table' will use its default name.
7900 *
7901 * This only affects the name exposed for a table exposed through the OpenFlow
7902 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
7903 static void
7904 oftable_set_name(struct oftable *table, const char *name)
7905 {
7906 if (name && name[0]) {
7907 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
7908 if (!table->name || strncmp(name, table->name, len)) {
7909 free(table->name);
7910 table->name = xmemdup0(name, len);
7911 }
7912 } else {
7913 free(table->name);
7914 table->name = NULL;
7915 }
7916 }
7917
7918 /* oftables support a choice of two policies when adding a rule would cause the
7919 * number of flows in the table to exceed the configured maximum number: either
7920 * they can refuse to add the new flow or they can evict some existing flow.
7921 * This function configures the latter policy on 'table', with fairness based
7922 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
7923 * 'n_fields' as 0 disables fairness.) */
7924 static void
7925 oftable_configure_eviction(struct oftable *table, unsigned int eviction,
7926 const struct mf_subfield *fields, size_t n_fields)
7927 OVS_REQUIRES(ofproto_mutex)
7928 {
7929 struct rule *rule;
7930
7931 if ((table->eviction != 0) == (eviction != 0)
7932 && n_fields == table->n_eviction_fields
7933 && (!n_fields
7934 || !memcmp(fields, table->eviction_fields,
7935 n_fields * sizeof *fields))) {
7936 /* The set of eviction fields did not change. If 'eviction' changed,
7937 * it remains nonzero, so that we can just update table->eviction
7938 * without fussing with the eviction groups. */
7939 table->eviction = eviction;
7940 return;
7941 }
7942
7943 /* Destroy existing eviction groups, then destroy and recreate data
7944 * structures to recover memory. */
7945 struct eviction_group *evg, *next;
7946 HMAP_FOR_EACH_SAFE (evg, next, id_node, &table->eviction_groups_by_id) {
7947 eviction_group_destroy(table, evg);
7948 }
7949 hmap_destroy(&table->eviction_groups_by_id);
7950 hmap_init(&table->eviction_groups_by_id);
7951 heap_destroy(&table->eviction_groups_by_size);
7952 heap_init(&table->eviction_groups_by_size);
7953
7954 /* Replace eviction groups by the new ones, if there is a change. Free the
7955 * old fields only after allocating the new ones, because 'fields ==
7956 * table->eviction_fields' is possible. */
7957 struct mf_subfield *old_fields = table->eviction_fields;
7958 table->n_eviction_fields = n_fields;
7959 table->eviction_fields = (fields
7960 ? xmemdup(fields, n_fields * sizeof *fields)
7961 : NULL);
7962 free(old_fields);
7963
7964 /* Add the new eviction groups, if enabled. */
7965 table->eviction = eviction;
7966 if (table->eviction) {
7967 table->eviction_group_id_basis = random_uint32();
7968 CLS_FOR_EACH (rule, cr, &table->cls) {
7969 eviction_group_add_rule(rule);
7970 }
7971 }
7972 }
7973
7974 /* Inserts 'rule' from the ofproto data structures BEFORE caller has inserted
7975 * it to the classifier. */
7976 static void
7977 ofproto_rule_insert__(struct ofproto *ofproto, struct rule *rule)
7978 OVS_REQUIRES(ofproto_mutex)
7979 {
7980 const struct rule_actions *actions = rule_get_actions(rule);
7981
7982 ovs_assert(rule->removed);
7983
7984 if (rule->hard_timeout || rule->idle_timeout) {
7985 ovs_list_insert(&ofproto->expirable, &rule->expirable);
7986 }
7987 cookies_insert(ofproto, rule);
7988 eviction_group_add_rule(rule);
7989 if (actions->has_meter) {
7990 meter_insert_rule(rule);
7991 }
7992 rule->removed = false;
7993 }
7994
7995 /* Removes 'rule' from the ofproto data structures. Caller may have deferred
7996 * the removal from the classifier. */
7997 static void
7998 ofproto_rule_remove__(struct ofproto *ofproto, struct rule *rule)
7999 OVS_REQUIRES(ofproto_mutex)
8000 {
8001 ovs_assert(!rule->removed);
8002
8003 cookies_remove(ofproto, rule);
8004
8005 eviction_group_remove_rule(rule);
8006 if (!ovs_list_is_empty(&rule->expirable)) {
8007 ovs_list_remove(&rule->expirable);
8008 }
8009 if (!ovs_list_is_empty(&rule->meter_list_node)) {
8010 ovs_list_remove(&rule->meter_list_node);
8011 ovs_list_init(&rule->meter_list_node);
8012 }
8013
8014 rule->removed = true;
8015 }
8016 \f
8017 /* unixctl commands. */
8018
8019 struct ofproto *
8020 ofproto_lookup(const char *name)
8021 {
8022 struct ofproto *ofproto;
8023
8024 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
8025 &all_ofprotos) {
8026 if (!strcmp(ofproto->name, name)) {
8027 return ofproto;
8028 }
8029 }
8030 return NULL;
8031 }
8032
8033 static void
8034 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
8035 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8036 {
8037 struct ofproto *ofproto;
8038 struct ds results;
8039
8040 ds_init(&results);
8041 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
8042 ds_put_format(&results, "%s\n", ofproto->name);
8043 }
8044 unixctl_command_reply(conn, ds_cstr(&results));
8045 ds_destroy(&results);
8046 }
8047
8048 static void
8049 ofproto_unixctl_init(void)
8050 {
8051 static bool registered;
8052 if (registered) {
8053 return;
8054 }
8055 registered = true;
8056
8057 unixctl_command_register("ofproto/list", "", 0, 0,
8058 ofproto_unixctl_list, NULL);
8059 }