]> git.proxmox.com Git - mirror_ovs.git/blob - ofproto/ofproto.c
Move lib/ofp-msgs.h to include/openvswitch directory
[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 "ofproto.h"
20 #include <errno.h>
21 #include <inttypes.h>
22 #include <stdbool.h>
23 #include <stdlib.h>
24 #include <unistd.h>
25 #include "bitmap.h"
26 #include "byte-order.h"
27 #include "classifier.h"
28 #include "connectivity.h"
29 #include "connmgr.h"
30 #include "coverage.h"
31 #include "hash.h"
32 #include "hmap.h"
33 #include "netdev.h"
34 #include "nx-match.h"
35 #include "ofp-actions.h"
36 #include "ofp-print.h"
37 #include "ofproto-provider.h"
38 #include "openflow/nicira-ext.h"
39 #include "openflow/openflow.h"
40 #include "openvswitch/dynamic-string.h"
41 #include "openvswitch/meta-flow.h"
42 #include "openvswitch/ofp-errors.h"
43 #include "openvswitch/ofp-msgs.h"
44 #include "openvswitch/ofp-util.h"
45 #include "openvswitch/ofpbuf.h"
46 #include "openvswitch/vlog.h"
47 #include "ovs-rcu.h"
48 #include "dp-packet.h"
49 #include "packets.h"
50 #include "pinsched.h"
51 #include "pktbuf.h"
52 #include "poll-loop.h"
53 #include "random.h"
54 #include "seq.h"
55 #include "shash.h"
56 #include "simap.h"
57 #include "smap.h"
58 #include "sset.h"
59 #include "timeval.h"
60 #include "tun-metadata.h"
61 #include "unaligned.h"
62 #include "unixctl.h"
63 #include "bundles.h"
64
65 VLOG_DEFINE_THIS_MODULE(ofproto);
66
67 COVERAGE_DEFINE(ofproto_flush);
68 COVERAGE_DEFINE(ofproto_packet_out);
69 COVERAGE_DEFINE(ofproto_queue_req);
70 COVERAGE_DEFINE(ofproto_recv_openflow);
71 COVERAGE_DEFINE(ofproto_reinit_ports);
72 COVERAGE_DEFINE(ofproto_update_port);
73
74 /* Default fields to use for prefix tries in each flow table, unless something
75 * else is configured. */
76 const enum mf_field_id default_prefix_fields[2] =
77 { MFF_IPV4_DST, MFF_IPV4_SRC };
78
79 /* oftable. */
80 static void oftable_init(struct oftable *);
81 static void oftable_destroy(struct oftable *);
82
83 static void oftable_set_name(struct oftable *, const char *name);
84
85 static enum ofperr evict_rules_from_table(struct oftable *)
86 OVS_REQUIRES(ofproto_mutex);
87 static void oftable_configure_eviction(struct oftable *,
88 unsigned int eviction,
89 const struct mf_subfield *fields,
90 size_t n_fields)
91 OVS_REQUIRES(ofproto_mutex);
92
93 /* This is the only combination of OpenFlow eviction flags that OVS supports: a
94 * combination of OF1.4+ importance, the remaining lifetime of the flow, and
95 * fairness based on user-specified fields. */
96 #define OFPROTO_EVICTION_FLAGS \
97 (OFPTMPEF14_OTHER | OFPTMPEF14_IMPORTANCE | OFPTMPEF14_LIFETIME)
98
99 /* A set of rules within a single OpenFlow table (oftable) that have the same
100 * values for the oftable's eviction_fields. A rule to be evicted, when one is
101 * needed, is taken from the eviction group that contains the greatest number
102 * of rules.
103 *
104 * An oftable owns any number of eviction groups, each of which contains any
105 * number of rules.
106 *
107 * Membership in an eviction group is imprecise, based on the hash of the
108 * oftable's eviction_fields (in the eviction_group's id_node.hash member).
109 * That is, if two rules have different eviction_fields, but those
110 * eviction_fields hash to the same value, then they will belong to the same
111 * eviction_group anyway.
112 *
113 * (When eviction is not enabled on an oftable, we don't track any eviction
114 * groups, to save time and space.) */
115 struct eviction_group {
116 struct hmap_node id_node; /* In oftable's "eviction_groups_by_id". */
117 struct heap_node size_node; /* In oftable's "eviction_groups_by_size". */
118 struct heap rules; /* Contains "struct rule"s. */
119 };
120
121 static bool choose_rule_to_evict(struct oftable *table, struct rule **rulep)
122 OVS_REQUIRES(ofproto_mutex);
123 static uint64_t rule_eviction_priority(struct ofproto *ofproto, struct rule *)
124 OVS_REQUIRES(ofproto_mutex);
125 static void eviction_group_add_rule(struct rule *)
126 OVS_REQUIRES(ofproto_mutex);
127 static void eviction_group_remove_rule(struct rule *)
128 OVS_REQUIRES(ofproto_mutex);
129
130 /* Criteria that flow_mod and other operations use for selecting rules on
131 * which to operate. */
132 struct rule_criteria {
133 /* An OpenFlow table or 255 for all tables. */
134 uint8_t table_id;
135
136 /* OpenFlow matching criteria. Interpreted different in "loose" way by
137 * collect_rules_loose() and "strict" way by collect_rules_strict(), as
138 * defined in the OpenFlow spec. */
139 struct cls_rule cr;
140 cls_version_t version;
141
142 /* Matching criteria for the OpenFlow cookie. Consider a bit B in a rule's
143 * cookie and the corresponding bits C in 'cookie' and M in 'cookie_mask'.
144 * The rule will not be selected if M is 1 and B != C. */
145 ovs_be64 cookie;
146 ovs_be64 cookie_mask;
147
148 /* Selection based on actions within a rule:
149 *
150 * If out_port != OFPP_ANY, selects only rules that output to out_port.
151 * If out_group != OFPG_ALL, select only rules that output to out_group. */
152 ofp_port_t out_port;
153 uint32_t out_group;
154
155 /* If true, collects only rules that are modifiable. */
156 bool include_hidden;
157 bool include_readonly;
158 };
159
160 static void rule_criteria_init(struct rule_criteria *, uint8_t table_id,
161 const struct match *match, int priority,
162 cls_version_t version,
163 ovs_be64 cookie, ovs_be64 cookie_mask,
164 ofp_port_t out_port, uint32_t out_group);
165 static void rule_criteria_require_rw(struct rule_criteria *,
166 bool can_write_readonly);
167 static void rule_criteria_destroy(struct rule_criteria *);
168
169 static enum ofperr collect_rules_loose(struct ofproto *,
170 const struct rule_criteria *,
171 struct rule_collection *);
172
173 /* A packet that needs to be passed to rule_execute().
174 *
175 * (We can't do this immediately from ofopgroup_complete() because that holds
176 * ofproto_mutex, which rule_execute() needs released.) */
177 struct rule_execute {
178 struct ovs_list list_node; /* In struct ofproto's "rule_executes" list. */
179 struct rule *rule; /* Owns a reference to the rule. */
180 ofp_port_t in_port;
181 struct dp_packet *packet; /* Owns the packet. */
182 };
183
184 static void run_rule_executes(struct ofproto *) OVS_EXCLUDED(ofproto_mutex);
185 static void destroy_rule_executes(struct ofproto *);
186
187 struct learned_cookie {
188 union {
189 /* In struct ofproto's 'learned_cookies' hmap. */
190 struct hmap_node hmap_node OVS_GUARDED_BY(ofproto_mutex);
191
192 /* In 'dead_cookies' list when removed from hmap. */
193 struct ovs_list list_node;
194 } u;
195
196 /* Key. */
197 ovs_be64 cookie OVS_GUARDED_BY(ofproto_mutex);
198 uint8_t table_id OVS_GUARDED_BY(ofproto_mutex);
199
200 /* Number of references from "learn" actions.
201 *
202 * When this drops to 0, all of the flows in 'table_id' with the specified
203 * 'cookie' are deleted. */
204 int n OVS_GUARDED_BY(ofproto_mutex);
205 };
206
207 static const struct ofpact_learn *next_learn_with_delete(
208 const struct rule_actions *, const struct ofpact_learn *start);
209
210 static void learned_cookies_inc(struct ofproto *, const struct rule_actions *)
211 OVS_REQUIRES(ofproto_mutex);
212 static void learned_cookies_dec(struct ofproto *, const struct rule_actions *,
213 struct ovs_list *dead_cookies)
214 OVS_REQUIRES(ofproto_mutex);
215 static void learned_cookies_flush(struct ofproto *, struct ovs_list *dead_cookies)
216 OVS_REQUIRES(ofproto_mutex);
217
218 /* ofport. */
219 static void ofport_destroy__(struct ofport *) OVS_EXCLUDED(ofproto_mutex);
220 static void ofport_destroy(struct ofport *, bool del);
221
222 static int update_port(struct ofproto *, const char *devname);
223 static int init_ports(struct ofproto *);
224 static void reinit_ports(struct ofproto *);
225
226 static long long int ofport_get_usage(const struct ofproto *,
227 ofp_port_t ofp_port);
228 static void ofport_set_usage(struct ofproto *, ofp_port_t ofp_port,
229 long long int last_used);
230 static void ofport_remove_usage(struct ofproto *, ofp_port_t ofp_port);
231
232 /* Ofport usage.
233 *
234 * Keeps track of the currently used and recently used ofport values and is
235 * used to prevent immediate recycling of ofport values. */
236 struct ofport_usage {
237 struct hmap_node hmap_node; /* In struct ofproto's "ofport_usage" hmap. */
238 ofp_port_t ofp_port; /* OpenFlow port number. */
239 long long int last_used; /* Last time the 'ofp_port' was used. LLONG_MAX
240 represents in-use ofports. */
241 };
242
243 /* rule. */
244 static void ofproto_rule_send_removed(struct rule *)
245 OVS_EXCLUDED(ofproto_mutex);
246 static bool rule_is_readonly(const struct rule *);
247 static void ofproto_rule_insert__(struct ofproto *, struct rule *)
248 OVS_REQUIRES(ofproto_mutex);
249 static void ofproto_rule_remove__(struct ofproto *, struct rule *)
250 OVS_REQUIRES(ofproto_mutex);
251
252 /* The source of a flow_mod request, in the code that processes flow_mods.
253 *
254 * A flow table modification request can be generated externally, via OpenFlow,
255 * or internally through a function call. This structure indicates the source
256 * of an OpenFlow-generated flow_mod. For an internal flow_mod, it isn't
257 * meaningful and thus supplied as NULL. */
258 struct flow_mod_requester {
259 struct ofconn *ofconn; /* Connection on which flow_mod arrived. */
260 const struct ofp_header *request;
261 };
262
263 /* OpenFlow. */
264 static enum ofperr replace_rule_create(struct ofproto *,
265 struct ofputil_flow_mod *,
266 struct cls_rule *cr, uint8_t table_id,
267 struct rule *old_rule,
268 struct rule **new_rule)
269 OVS_REQUIRES(ofproto_mutex);
270
271 static void replace_rule_start(struct ofproto *, cls_version_t version,
272 struct rule *old_rule, struct rule *new_rule,
273 struct cls_conjunction *, size_t n_conjs)
274 OVS_REQUIRES(ofproto_mutex);
275
276 static void replace_rule_revert(struct ofproto *, struct rule *old_rule,
277 struct rule *new_rule)
278 OVS_REQUIRES(ofproto_mutex);
279
280 static void replace_rule_finish(struct ofproto *, struct ofputil_flow_mod *,
281 const struct flow_mod_requester *,
282 struct rule *old_rule, struct rule *new_rule,
283 struct ovs_list *dead_cookies)
284 OVS_REQUIRES(ofproto_mutex);
285 static void delete_flows__(struct rule_collection *,
286 enum ofp_flow_removed_reason,
287 const struct flow_mod_requester *)
288 OVS_REQUIRES(ofproto_mutex);
289
290 static void send_buffered_packet(const struct flow_mod_requester *,
291 uint32_t buffer_id, struct rule *)
292 OVS_REQUIRES(ofproto_mutex);
293
294 static bool ofproto_group_exists__(const struct ofproto *ofproto,
295 uint32_t group_id)
296 OVS_REQ_RDLOCK(ofproto->groups_rwlock);
297 static bool ofproto_group_exists(const struct ofproto *ofproto,
298 uint32_t group_id)
299 OVS_EXCLUDED(ofproto->groups_rwlock);
300 static enum ofperr add_group(struct ofproto *,
301 const struct ofputil_group_mod *);
302 static void handle_openflow(struct ofconn *, const struct ofpbuf *);
303 static enum ofperr ofproto_flow_mod_start(struct ofproto *,
304 struct ofproto_flow_mod *)
305 OVS_REQUIRES(ofproto_mutex);
306 static void ofproto_flow_mod_finish(struct ofproto *,
307 struct ofproto_flow_mod *,
308 const struct flow_mod_requester *)
309 OVS_REQUIRES(ofproto_mutex);
310 static enum ofperr handle_flow_mod__(struct ofproto *,
311 struct ofproto_flow_mod *,
312 const struct flow_mod_requester *)
313 OVS_EXCLUDED(ofproto_mutex);
314 static void calc_duration(long long int start, long long int now,
315 uint32_t *sec, uint32_t *nsec);
316
317 /* ofproto. */
318 static uint64_t pick_datapath_id(const struct ofproto *);
319 static uint64_t pick_fallback_dpid(void);
320 static void ofproto_destroy__(struct ofproto *);
321 static void update_mtu(struct ofproto *, struct ofport *);
322 static void meter_delete(struct ofproto *, uint32_t first, uint32_t last);
323 static void meter_insert_rule(struct rule *);
324
325 /* unixctl. */
326 static void ofproto_unixctl_init(void);
327
328 /* All registered ofproto classes, in probe order. */
329 static const struct ofproto_class **ofproto_classes;
330 static size_t n_ofproto_classes;
331 static size_t allocated_ofproto_classes;
332
333 /* Global lock that protects all flow table operations. */
334 struct ovs_mutex ofproto_mutex = OVS_MUTEX_INITIALIZER;
335
336 unsigned ofproto_flow_limit = OFPROTO_FLOW_LIMIT_DEFAULT;
337 unsigned ofproto_max_idle = OFPROTO_MAX_IDLE_DEFAULT;
338
339 size_t n_handlers, n_revalidators;
340 char *pmd_cpu_mask;
341
342 /* Map from datapath name to struct ofproto, for use by unixctl commands. */
343 static struct hmap all_ofprotos = HMAP_INITIALIZER(&all_ofprotos);
344
345 /* Initial mappings of port to OpenFlow number mappings. */
346 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
347
348 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
349
350 /* The default value of true waits for flow restore. */
351 static bool flow_restore_wait = true;
352
353 /* Must be called to initialize the ofproto library.
354 *
355 * The caller may pass in 'iface_hints', which contains an shash of
356 * "iface_hint" elements indexed by the interface's name. The provider
357 * may use these hints to describe the startup configuration in order to
358 * reinitialize its state. The caller owns the provided data, so a
359 * provider will make copies of anything required. An ofproto provider
360 * will remove any existing state that is not described by the hint, and
361 * may choose to remove it all. */
362 void
363 ofproto_init(const struct shash *iface_hints)
364 {
365 struct shash_node *node;
366 size_t i;
367
368 ofproto_class_register(&ofproto_dpif_class);
369
370 /* Make a local copy, since we don't own 'iface_hints' elements. */
371 SHASH_FOR_EACH(node, iface_hints) {
372 const struct iface_hint *orig_hint = node->data;
373 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
374 const char *br_type = ofproto_normalize_type(orig_hint->br_type);
375
376 new_hint->br_name = xstrdup(orig_hint->br_name);
377 new_hint->br_type = xstrdup(br_type);
378 new_hint->ofp_port = orig_hint->ofp_port;
379
380 shash_add(&init_ofp_ports, node->name, new_hint);
381 }
382
383 for (i = 0; i < n_ofproto_classes; i++) {
384 ofproto_classes[i]->init(&init_ofp_ports);
385 }
386
387 ofproto_unixctl_init();
388 }
389
390 /* 'type' should be a normalized datapath type, as returned by
391 * ofproto_normalize_type(). Returns the corresponding ofproto_class
392 * structure, or a null pointer if there is none registered for 'type'. */
393 static const struct ofproto_class *
394 ofproto_class_find__(const char *type)
395 {
396 size_t i;
397
398 for (i = 0; i < n_ofproto_classes; i++) {
399 const struct ofproto_class *class = ofproto_classes[i];
400 struct sset types;
401 bool found;
402
403 sset_init(&types);
404 class->enumerate_types(&types);
405 found = sset_contains(&types, type);
406 sset_destroy(&types);
407
408 if (found) {
409 return class;
410 }
411 }
412 VLOG_WARN("unknown datapath type %s", type);
413 return NULL;
414 }
415
416 /* Registers a new ofproto class. After successful registration, new ofprotos
417 * of that type can be created using ofproto_create(). */
418 int
419 ofproto_class_register(const struct ofproto_class *new_class)
420 {
421 size_t i;
422
423 for (i = 0; i < n_ofproto_classes; i++) {
424 if (ofproto_classes[i] == new_class) {
425 return EEXIST;
426 }
427 }
428
429 if (n_ofproto_classes >= allocated_ofproto_classes) {
430 ofproto_classes = x2nrealloc(ofproto_classes,
431 &allocated_ofproto_classes,
432 sizeof *ofproto_classes);
433 }
434 ofproto_classes[n_ofproto_classes++] = new_class;
435 return 0;
436 }
437
438 /* Unregisters a datapath provider. 'type' must have been previously
439 * registered and not currently be in use by any ofprotos. After
440 * unregistration new datapaths of that type cannot be opened using
441 * ofproto_create(). */
442 int
443 ofproto_class_unregister(const struct ofproto_class *class)
444 {
445 size_t i;
446
447 for (i = 0; i < n_ofproto_classes; i++) {
448 if (ofproto_classes[i] == class) {
449 for (i++; i < n_ofproto_classes; i++) {
450 ofproto_classes[i - 1] = ofproto_classes[i];
451 }
452 n_ofproto_classes--;
453 return 0;
454 }
455 }
456 VLOG_WARN("attempted to unregister an ofproto class that is not "
457 "registered");
458 return EAFNOSUPPORT;
459 }
460
461 /* Clears 'types' and enumerates all registered ofproto types into it. The
462 * caller must first initialize the sset. */
463 void
464 ofproto_enumerate_types(struct sset *types)
465 {
466 size_t i;
467
468 sset_clear(types);
469 for (i = 0; i < n_ofproto_classes; i++) {
470 ofproto_classes[i]->enumerate_types(types);
471 }
472 }
473
474 /* Returns the fully spelled out name for the given ofproto 'type'.
475 *
476 * Normalized type string can be compared with strcmp(). Unnormalized type
477 * string might be the same even if they have different spellings. */
478 const char *
479 ofproto_normalize_type(const char *type)
480 {
481 return type && type[0] ? type : "system";
482 }
483
484 /* Clears 'names' and enumerates the names of all known created ofprotos with
485 * the given 'type'. The caller must first initialize the sset. Returns 0 if
486 * successful, otherwise a positive errno value.
487 *
488 * Some kinds of datapaths might not be practically enumerable. This is not
489 * considered an error. */
490 int
491 ofproto_enumerate_names(const char *type, struct sset *names)
492 {
493 const struct ofproto_class *class = ofproto_class_find__(type);
494 return class ? class->enumerate_names(type, names) : EAFNOSUPPORT;
495 }
496
497 static void
498 ofproto_bump_tables_version(struct ofproto *ofproto)
499 {
500 ++ofproto->tables_version;
501 ofproto->ofproto_class->set_tables_version(ofproto,
502 ofproto->tables_version);
503 }
504
505 int
506 ofproto_create(const char *datapath_name, const char *datapath_type,
507 struct ofproto **ofprotop)
508 {
509 const struct ofproto_class *class;
510 struct ofproto *ofproto;
511 int error;
512 int i;
513
514 *ofprotop = NULL;
515
516 datapath_type = ofproto_normalize_type(datapath_type);
517 class = ofproto_class_find__(datapath_type);
518 if (!class) {
519 VLOG_WARN("could not create datapath %s of unknown type %s",
520 datapath_name, datapath_type);
521 return EAFNOSUPPORT;
522 }
523
524 ofproto = class->alloc();
525 if (!ofproto) {
526 VLOG_ERR("failed to allocate datapath %s of type %s",
527 datapath_name, datapath_type);
528 return ENOMEM;
529 }
530
531 /* Initialize. */
532 ovs_mutex_lock(&ofproto_mutex);
533 memset(ofproto, 0, sizeof *ofproto);
534 ofproto->ofproto_class = class;
535 ofproto->name = xstrdup(datapath_name);
536 ofproto->type = xstrdup(datapath_type);
537 hmap_insert(&all_ofprotos, &ofproto->hmap_node,
538 hash_string(ofproto->name, 0));
539 ofproto->datapath_id = 0;
540 ofproto->forward_bpdu = false;
541 ofproto->fallback_dpid = pick_fallback_dpid();
542 ofproto->mfr_desc = NULL;
543 ofproto->hw_desc = NULL;
544 ofproto->sw_desc = NULL;
545 ofproto->serial_desc = NULL;
546 ofproto->dp_desc = NULL;
547 ofproto->frag_handling = OFPUTIL_FRAG_NORMAL;
548 hmap_init(&ofproto->ports);
549 hmap_init(&ofproto->ofport_usage);
550 shash_init(&ofproto->port_by_name);
551 simap_init(&ofproto->ofp_requests);
552 ofproto->max_ports = ofp_to_u16(OFPP_MAX);
553 ofproto->eviction_group_timer = LLONG_MIN;
554 ofproto->tables = NULL;
555 ofproto->n_tables = 0;
556 ofproto->tables_version = CLS_MIN_VERSION;
557 hindex_init(&ofproto->cookies);
558 hmap_init(&ofproto->learned_cookies);
559 ovs_list_init(&ofproto->expirable);
560 ofproto->connmgr = connmgr_create(ofproto, datapath_name, datapath_name);
561 guarded_list_init(&ofproto->rule_executes);
562 ofproto->vlan_bitmap = NULL;
563 ofproto->vlans_changed = false;
564 ofproto->min_mtu = INT_MAX;
565 ovs_rwlock_init(&ofproto->groups_rwlock);
566 hmap_init(&ofproto->groups);
567 ovs_mutex_unlock(&ofproto_mutex);
568 ofproto->ogf.types = 0xf;
569 ofproto->ogf.capabilities = OFPGFC_CHAINING | OFPGFC_SELECT_LIVENESS |
570 OFPGFC_SELECT_WEIGHT;
571 for (i = 0; i < 4; i++) {
572 ofproto->ogf.max_groups[i] = OFPG_MAX;
573 ofproto->ogf.ofpacts[i] = (UINT64_C(1) << N_OFPACTS) - 1;
574 }
575 tun_metadata_init();
576
577 error = ofproto->ofproto_class->construct(ofproto);
578 if (error) {
579 VLOG_ERR("failed to open datapath %s: %s",
580 datapath_name, ovs_strerror(error));
581 connmgr_destroy(ofproto->connmgr);
582 ofproto_destroy__(ofproto);
583 return error;
584 }
585
586 /* Check that hidden tables, if any, are at the end. */
587 ovs_assert(ofproto->n_tables);
588 for (i = 0; i + 1 < ofproto->n_tables; i++) {
589 enum oftable_flags flags = ofproto->tables[i].flags;
590 enum oftable_flags next_flags = ofproto->tables[i + 1].flags;
591
592 ovs_assert(!(flags & OFTABLE_HIDDEN) || next_flags & OFTABLE_HIDDEN);
593 }
594
595 ofproto->datapath_id = pick_datapath_id(ofproto);
596 init_ports(ofproto);
597
598 /* Initialize meters table. */
599 if (ofproto->ofproto_class->meter_get_features) {
600 ofproto->ofproto_class->meter_get_features(ofproto,
601 &ofproto->meter_features);
602 } else {
603 memset(&ofproto->meter_features, 0, sizeof ofproto->meter_features);
604 }
605 ofproto->meters = xzalloc((ofproto->meter_features.max_meters + 1)
606 * sizeof(struct meter *));
607
608 /* Set the initial tables version. */
609 ofproto_bump_tables_version(ofproto);
610
611 *ofprotop = ofproto;
612 return 0;
613 }
614
615 /* Must be called (only) by an ofproto implementation in its constructor
616 * function. See the large comment on 'construct' in struct ofproto_class for
617 * details. */
618 void
619 ofproto_init_tables(struct ofproto *ofproto, int n_tables)
620 {
621 struct oftable *table;
622
623 ovs_assert(!ofproto->n_tables);
624 ovs_assert(n_tables >= 1 && n_tables <= 255);
625
626 ofproto->n_tables = n_tables;
627 ofproto->tables = xmalloc(n_tables * sizeof *ofproto->tables);
628 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
629 oftable_init(table);
630 }
631 }
632
633 /* To be optionally called (only) by an ofproto implementation in its
634 * constructor function. See the large comment on 'construct' in struct
635 * ofproto_class for details.
636 *
637 * Sets the maximum number of ports to 'max_ports'. The ofproto generic layer
638 * will then ensure that actions passed into the ofproto implementation will
639 * not refer to OpenFlow ports numbered 'max_ports' or higher. If this
640 * function is not called, there will be no such restriction.
641 *
642 * Reserved ports numbered OFPP_MAX and higher are special and not subject to
643 * the 'max_ports' restriction. */
644 void
645 ofproto_init_max_ports(struct ofproto *ofproto, uint16_t max_ports)
646 {
647 ovs_assert(max_ports <= ofp_to_u16(OFPP_MAX));
648 ofproto->max_ports = max_ports;
649 }
650
651 uint64_t
652 ofproto_get_datapath_id(const struct ofproto *ofproto)
653 {
654 return ofproto->datapath_id;
655 }
656
657 void
658 ofproto_set_datapath_id(struct ofproto *p, uint64_t datapath_id)
659 {
660 uint64_t old_dpid = p->datapath_id;
661 p->datapath_id = datapath_id ? datapath_id : pick_datapath_id(p);
662 if (p->datapath_id != old_dpid) {
663 /* Force all active connections to reconnect, since there is no way to
664 * notify a controller that the datapath ID has changed. */
665 ofproto_reconnect_controllers(p);
666 }
667 }
668
669 void
670 ofproto_set_controllers(struct ofproto *p,
671 const struct ofproto_controller *controllers,
672 size_t n_controllers, uint32_t allowed_versions)
673 {
674 connmgr_set_controllers(p->connmgr, controllers, n_controllers,
675 allowed_versions);
676 }
677
678 void
679 ofproto_set_fail_mode(struct ofproto *p, enum ofproto_fail_mode fail_mode)
680 {
681 connmgr_set_fail_mode(p->connmgr, fail_mode);
682 }
683
684 /* Drops the connections between 'ofproto' and all of its controllers, forcing
685 * them to reconnect. */
686 void
687 ofproto_reconnect_controllers(struct ofproto *ofproto)
688 {
689 connmgr_reconnect(ofproto->connmgr);
690 }
691
692 /* Sets the 'n' TCP port addresses in 'extras' as ones to which 'ofproto''s
693 * in-band control should guarantee access, in the same way that in-band
694 * control guarantees access to OpenFlow controllers. */
695 void
696 ofproto_set_extra_in_band_remotes(struct ofproto *ofproto,
697 const struct sockaddr_in *extras, size_t n)
698 {
699 connmgr_set_extra_in_band_remotes(ofproto->connmgr, extras, n);
700 }
701
702 /* Sets the OpenFlow queue used by flows set up by in-band control on
703 * 'ofproto' to 'queue_id'. If 'queue_id' is negative, then in-band control
704 * flows will use the default queue. */
705 void
706 ofproto_set_in_band_queue(struct ofproto *ofproto, int queue_id)
707 {
708 connmgr_set_in_band_queue(ofproto->connmgr, queue_id);
709 }
710
711 /* Sets the number of flows at which eviction from the kernel flow table
712 * will occur. */
713 void
714 ofproto_set_flow_limit(unsigned limit)
715 {
716 ofproto_flow_limit = limit;
717 }
718
719 /* Sets the maximum idle time for flows in the datapath before they are
720 * expired. */
721 void
722 ofproto_set_max_idle(unsigned max_idle)
723 {
724 ofproto_max_idle = max_idle;
725 }
726
727 /* If forward_bpdu is true, the NORMAL action will forward frames with
728 * reserved (e.g. STP) destination Ethernet addresses. if forward_bpdu is false,
729 * the NORMAL action will drop these frames. */
730 void
731 ofproto_set_forward_bpdu(struct ofproto *ofproto, bool forward_bpdu)
732 {
733 bool old_val = ofproto->forward_bpdu;
734 ofproto->forward_bpdu = forward_bpdu;
735 if (old_val != ofproto->forward_bpdu) {
736 if (ofproto->ofproto_class->forward_bpdu_changed) {
737 ofproto->ofproto_class->forward_bpdu_changed(ofproto);
738 }
739 }
740 }
741
742 /* Sets the MAC aging timeout for the OFPP_NORMAL action on 'ofproto' to
743 * 'idle_time', in seconds, and the maximum number of MAC table entries to
744 * 'max_entries'. */
745 void
746 ofproto_set_mac_table_config(struct ofproto *ofproto, unsigned idle_time,
747 size_t max_entries)
748 {
749 if (ofproto->ofproto_class->set_mac_table_config) {
750 ofproto->ofproto_class->set_mac_table_config(ofproto, idle_time,
751 max_entries);
752 }
753 }
754
755 /* Multicast snooping configuration. */
756
757 /* Configures multicast snooping on 'ofproto' using the settings
758 * defined in 's'. If 's' is NULL, disables multicast snooping.
759 *
760 * Returns 0 if successful, otherwise a positive errno value. */
761 int
762 ofproto_set_mcast_snooping(struct ofproto *ofproto,
763 const struct ofproto_mcast_snooping_settings *s)
764 {
765 return (ofproto->ofproto_class->set_mcast_snooping
766 ? ofproto->ofproto_class->set_mcast_snooping(ofproto, s)
767 : EOPNOTSUPP);
768 }
769
770 /* Configures multicast snooping flood settings on 'ofp_port' of 'ofproto'.
771 *
772 * Returns 0 if successful, otherwise a positive errno value.*/
773 int
774 ofproto_port_set_mcast_snooping(struct ofproto *ofproto, void *aux,
775 const struct ofproto_mcast_snooping_port_settings *s)
776 {
777 return (ofproto->ofproto_class->set_mcast_snooping_port
778 ? ofproto->ofproto_class->set_mcast_snooping_port(ofproto, aux, s)
779 : EOPNOTSUPP);
780 }
781
782 void
783 ofproto_set_cpu_mask(const char *cmask)
784 {
785 free(pmd_cpu_mask);
786
787 pmd_cpu_mask = cmask ? xstrdup(cmask) : NULL;
788 }
789
790 void
791 ofproto_set_threads(int n_handlers_, int n_revalidators_)
792 {
793 int threads = MAX(count_cpu_cores(), 2);
794
795 n_revalidators = MAX(n_revalidators_, 0);
796 n_handlers = MAX(n_handlers_, 0);
797
798 if (!n_revalidators) {
799 n_revalidators = n_handlers
800 ? MAX(threads - (int) n_handlers, 1)
801 : threads / 4 + 1;
802 }
803
804 if (!n_handlers) {
805 n_handlers = MAX(threads - (int) n_revalidators, 1);
806 }
807 }
808
809 void
810 ofproto_set_dp_desc(struct ofproto *p, const char *dp_desc)
811 {
812 free(p->dp_desc);
813 p->dp_desc = dp_desc ? xstrdup(dp_desc) : NULL;
814 }
815
816 int
817 ofproto_set_snoops(struct ofproto *ofproto, const struct sset *snoops)
818 {
819 return connmgr_set_snoops(ofproto->connmgr, snoops);
820 }
821
822 int
823 ofproto_set_netflow(struct ofproto *ofproto,
824 const struct netflow_options *nf_options)
825 {
826 if (nf_options && sset_is_empty(&nf_options->collectors)) {
827 nf_options = NULL;
828 }
829
830 if (ofproto->ofproto_class->set_netflow) {
831 return ofproto->ofproto_class->set_netflow(ofproto, nf_options);
832 } else {
833 return nf_options ? EOPNOTSUPP : 0;
834 }
835 }
836
837 int
838 ofproto_set_sflow(struct ofproto *ofproto,
839 const struct ofproto_sflow_options *oso)
840 {
841 if (oso && sset_is_empty(&oso->targets)) {
842 oso = NULL;
843 }
844
845 if (ofproto->ofproto_class->set_sflow) {
846 return ofproto->ofproto_class->set_sflow(ofproto, oso);
847 } else {
848 return oso ? EOPNOTSUPP : 0;
849 }
850 }
851
852 int
853 ofproto_set_ipfix(struct ofproto *ofproto,
854 const struct ofproto_ipfix_bridge_exporter_options *bo,
855 const struct ofproto_ipfix_flow_exporter_options *fo,
856 size_t n_fo)
857 {
858 if (ofproto->ofproto_class->set_ipfix) {
859 return ofproto->ofproto_class->set_ipfix(ofproto, bo, fo, n_fo);
860 } else {
861 return (bo || fo) ? EOPNOTSUPP : 0;
862 }
863 }
864
865 void
866 ofproto_set_flow_restore_wait(bool flow_restore_wait_db)
867 {
868 flow_restore_wait = flow_restore_wait_db;
869 }
870
871 bool
872 ofproto_get_flow_restore_wait(void)
873 {
874 return flow_restore_wait;
875 }
876
877 \f
878 /* Spanning Tree Protocol (STP) configuration. */
879
880 /* Configures STP on 'ofproto' using the settings defined in 's'. If
881 * 's' is NULL, disables STP.
882 *
883 * Returns 0 if successful, otherwise a positive errno value. */
884 int
885 ofproto_set_stp(struct ofproto *ofproto,
886 const struct ofproto_stp_settings *s)
887 {
888 return (ofproto->ofproto_class->set_stp
889 ? ofproto->ofproto_class->set_stp(ofproto, s)
890 : EOPNOTSUPP);
891 }
892
893 /* Retrieves STP status of 'ofproto' and stores it in 's'. If the
894 * 'enabled' member of 's' is false, then the other members are not
895 * meaningful.
896 *
897 * Returns 0 if successful, otherwise a positive errno value. */
898 int
899 ofproto_get_stp_status(struct ofproto *ofproto,
900 struct ofproto_stp_status *s)
901 {
902 return (ofproto->ofproto_class->get_stp_status
903 ? ofproto->ofproto_class->get_stp_status(ofproto, s)
904 : EOPNOTSUPP);
905 }
906
907 /* Configures STP on 'ofp_port' of 'ofproto' using the settings defined
908 * in 's'. The caller is responsible for assigning STP port numbers
909 * (using the 'port_num' member in the range of 1 through 255, inclusive)
910 * and ensuring there are no duplicates. If the 's' is NULL, then STP
911 * is disabled on the port.
912 *
913 * Returns 0 if successful, otherwise a positive errno value.*/
914 int
915 ofproto_port_set_stp(struct ofproto *ofproto, ofp_port_t ofp_port,
916 const struct ofproto_port_stp_settings *s)
917 {
918 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
919 if (!ofport) {
920 VLOG_WARN("%s: cannot configure STP on nonexistent port %"PRIu16,
921 ofproto->name, ofp_port);
922 return ENODEV;
923 }
924
925 return (ofproto->ofproto_class->set_stp_port
926 ? ofproto->ofproto_class->set_stp_port(ofport, s)
927 : EOPNOTSUPP);
928 }
929
930 /* Retrieves STP port status of 'ofp_port' on 'ofproto' and stores it in
931 * 's'. If the 'enabled' member in 's' is false, then the other members
932 * are not meaningful.
933 *
934 * Returns 0 if successful, otherwise a positive errno value.*/
935 int
936 ofproto_port_get_stp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
937 struct ofproto_port_stp_status *s)
938 {
939 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
940 if (!ofport) {
941 VLOG_WARN_RL(&rl, "%s: cannot get STP status on nonexistent "
942 "port %"PRIu16, ofproto->name, ofp_port);
943 return ENODEV;
944 }
945
946 return (ofproto->ofproto_class->get_stp_port_status
947 ? ofproto->ofproto_class->get_stp_port_status(ofport, s)
948 : EOPNOTSUPP);
949 }
950
951 /* Retrieves STP port statistics of 'ofp_port' on 'ofproto' and stores it in
952 * 's'. If the 'enabled' member in 's' is false, then the other members
953 * are not meaningful.
954 *
955 * Returns 0 if successful, otherwise a positive errno value.*/
956 int
957 ofproto_port_get_stp_stats(struct ofproto *ofproto, ofp_port_t ofp_port,
958 struct ofproto_port_stp_stats *s)
959 {
960 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
961 if (!ofport) {
962 VLOG_WARN_RL(&rl, "%s: cannot get STP stats on nonexistent "
963 "port %"PRIu16, ofproto->name, ofp_port);
964 return ENODEV;
965 }
966
967 return (ofproto->ofproto_class->get_stp_port_stats
968 ? ofproto->ofproto_class->get_stp_port_stats(ofport, s)
969 : EOPNOTSUPP);
970 }
971
972 /* Rapid Spanning Tree Protocol (RSTP) configuration. */
973
974 /* Configures RSTP on 'ofproto' using the settings defined in 's'. If
975 * 's' is NULL, disables RSTP.
976 *
977 * Returns 0 if successful, otherwise a positive errno value. */
978 int
979 ofproto_set_rstp(struct ofproto *ofproto,
980 const struct ofproto_rstp_settings *s)
981 {
982 if (!ofproto->ofproto_class->set_rstp) {
983 return EOPNOTSUPP;
984 }
985 ofproto->ofproto_class->set_rstp(ofproto, s);
986 return 0;
987 }
988
989 /* Retrieves RSTP status of 'ofproto' and stores it in 's'. If the
990 * 'enabled' member of 's' is false, then the other members are not
991 * meaningful.
992 *
993 * Returns 0 if successful, otherwise a positive errno value. */
994 int
995 ofproto_get_rstp_status(struct ofproto *ofproto,
996 struct ofproto_rstp_status *s)
997 {
998 if (!ofproto->ofproto_class->get_rstp_status) {
999 return EOPNOTSUPP;
1000 }
1001 ofproto->ofproto_class->get_rstp_status(ofproto, s);
1002 return 0;
1003 }
1004
1005 /* Configures RSTP on 'ofp_port' of 'ofproto' using the settings defined
1006 * in 's'. The caller is responsible for assigning RSTP port numbers
1007 * (using the 'port_num' member in the range of 1 through 255, inclusive)
1008 * and ensuring there are no duplicates. If the 's' is NULL, then RSTP
1009 * is disabled on the port.
1010 *
1011 * Returns 0 if successful, otherwise a positive errno value.*/
1012 int
1013 ofproto_port_set_rstp(struct ofproto *ofproto, ofp_port_t ofp_port,
1014 const struct ofproto_port_rstp_settings *s)
1015 {
1016 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1017 if (!ofport) {
1018 VLOG_WARN("%s: cannot configure RSTP on nonexistent port %"PRIu16,
1019 ofproto->name, ofp_port);
1020 return ENODEV;
1021 }
1022
1023 if (!ofproto->ofproto_class->set_rstp_port) {
1024 return EOPNOTSUPP;
1025 }
1026 ofproto->ofproto_class->set_rstp_port(ofport, s);
1027 return 0;
1028 }
1029
1030 /* Retrieves RSTP port status of 'ofp_port' on 'ofproto' and stores it in
1031 * 's'. If the 'enabled' member in 's' is false, then the other members
1032 * are not meaningful.
1033 *
1034 * Returns 0 if successful, otherwise a positive errno value.*/
1035 int
1036 ofproto_port_get_rstp_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1037 struct ofproto_port_rstp_status *s)
1038 {
1039 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1040 if (!ofport) {
1041 VLOG_WARN_RL(&rl, "%s: cannot get RSTP status on nonexistent "
1042 "port %"PRIu16, ofproto->name, ofp_port);
1043 return ENODEV;
1044 }
1045
1046 if (!ofproto->ofproto_class->get_rstp_port_status) {
1047 return EOPNOTSUPP;
1048 }
1049 ofproto->ofproto_class->get_rstp_port_status(ofport, s);
1050 return 0;
1051 }
1052 \f
1053 /* Queue DSCP configuration. */
1054
1055 /* Registers meta-data associated with the 'n_qdscp' Qualities of Service
1056 * 'queues' attached to 'ofport'. This data is not intended to be sufficient
1057 * to implement QoS. Instead, it is used to implement features which require
1058 * knowledge of what queues exist on a port, and some basic information about
1059 * them.
1060 *
1061 * Returns 0 if successful, otherwise a positive errno value. */
1062 int
1063 ofproto_port_set_queues(struct ofproto *ofproto, ofp_port_t ofp_port,
1064 const struct ofproto_port_queue *queues,
1065 size_t n_queues)
1066 {
1067 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1068
1069 if (!ofport) {
1070 VLOG_WARN("%s: cannot set queues on nonexistent port %"PRIu16,
1071 ofproto->name, ofp_port);
1072 return ENODEV;
1073 }
1074
1075 return (ofproto->ofproto_class->set_queues
1076 ? ofproto->ofproto_class->set_queues(ofport, queues, n_queues)
1077 : EOPNOTSUPP);
1078 }
1079 \f
1080 /* LLDP configuration. */
1081 void
1082 ofproto_port_set_lldp(struct ofproto *ofproto,
1083 ofp_port_t ofp_port,
1084 const struct smap *cfg)
1085 {
1086 struct ofport *ofport;
1087 int error;
1088
1089 ofport = ofproto_get_port(ofproto, ofp_port);
1090 if (!ofport) {
1091 VLOG_WARN("%s: cannot configure LLDP on nonexistent port %"PRIu16,
1092 ofproto->name, ofp_port);
1093 return;
1094 }
1095 error = (ofproto->ofproto_class->set_lldp
1096 ? ofproto->ofproto_class->set_lldp(ofport, cfg)
1097 : EOPNOTSUPP);
1098 if (error) {
1099 VLOG_WARN("%s: lldp configuration on port %"PRIu16" (%s) failed (%s)",
1100 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1101 ovs_strerror(error));
1102 }
1103 }
1104
1105 int
1106 ofproto_set_aa(struct ofproto *ofproto, void *aux OVS_UNUSED,
1107 const struct aa_settings *s)
1108 {
1109 if (!ofproto->ofproto_class->set_aa) {
1110 return EOPNOTSUPP;
1111 }
1112 ofproto->ofproto_class->set_aa(ofproto, s);
1113 return 0;
1114 }
1115
1116 int
1117 ofproto_aa_mapping_register(struct ofproto *ofproto, void *aux,
1118 const struct aa_mapping_settings *s)
1119 {
1120 if (!ofproto->ofproto_class->aa_mapping_set) {
1121 return EOPNOTSUPP;
1122 }
1123 ofproto->ofproto_class->aa_mapping_set(ofproto, aux, s);
1124 return 0;
1125 }
1126
1127 int
1128 ofproto_aa_mapping_unregister(struct ofproto *ofproto, void *aux)
1129 {
1130 if (!ofproto->ofproto_class->aa_mapping_unset) {
1131 return EOPNOTSUPP;
1132 }
1133 ofproto->ofproto_class->aa_mapping_unset(ofproto, aux);
1134 return 0;
1135 }
1136
1137 int
1138 ofproto_aa_vlan_get_queued(struct ofproto *ofproto,
1139 struct ovs_list *list)
1140 {
1141 if (!ofproto->ofproto_class->aa_vlan_get_queued) {
1142 return EOPNOTSUPP;
1143 }
1144 ofproto->ofproto_class->aa_vlan_get_queued(ofproto, list);
1145 return 0;
1146 }
1147
1148 unsigned int
1149 ofproto_aa_vlan_get_queue_size(struct ofproto *ofproto)
1150 {
1151 if (!ofproto->ofproto_class->aa_vlan_get_queue_size) {
1152 return EOPNOTSUPP;
1153 }
1154 return ofproto->ofproto_class->aa_vlan_get_queue_size(ofproto);
1155 }
1156
1157 /* Connectivity Fault Management configuration. */
1158
1159 /* Clears the CFM configuration from 'ofp_port' on 'ofproto'. */
1160 void
1161 ofproto_port_clear_cfm(struct ofproto *ofproto, ofp_port_t ofp_port)
1162 {
1163 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1164 if (ofport && ofproto->ofproto_class->set_cfm) {
1165 ofproto->ofproto_class->set_cfm(ofport, NULL);
1166 }
1167 }
1168
1169 /* Configures connectivity fault management on 'ofp_port' in 'ofproto'. Takes
1170 * basic configuration from the configuration members in 'cfm', and the remote
1171 * maintenance point ID from remote_mpid. Ignores the statistics members of
1172 * 'cfm'.
1173 *
1174 * This function has no effect if 'ofproto' does not have a port 'ofp_port'. */
1175 void
1176 ofproto_port_set_cfm(struct ofproto *ofproto, ofp_port_t ofp_port,
1177 const struct cfm_settings *s)
1178 {
1179 struct ofport *ofport;
1180 int error;
1181
1182 ofport = ofproto_get_port(ofproto, ofp_port);
1183 if (!ofport) {
1184 VLOG_WARN("%s: cannot configure CFM on nonexistent port %"PRIu16,
1185 ofproto->name, ofp_port);
1186 return;
1187 }
1188
1189 /* XXX: For configuration simplicity, we only support one remote_mpid
1190 * outside of the CFM module. It's not clear if this is the correct long
1191 * term solution or not. */
1192 error = (ofproto->ofproto_class->set_cfm
1193 ? ofproto->ofproto_class->set_cfm(ofport, s)
1194 : EOPNOTSUPP);
1195 if (error) {
1196 VLOG_WARN("%s: CFM configuration on port %"PRIu16" (%s) failed (%s)",
1197 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1198 ovs_strerror(error));
1199 }
1200 }
1201
1202 /* Configures BFD on 'ofp_port' in 'ofproto'. This function has no effect if
1203 * 'ofproto' does not have a port 'ofp_port'. */
1204 void
1205 ofproto_port_set_bfd(struct ofproto *ofproto, ofp_port_t ofp_port,
1206 const struct smap *cfg)
1207 {
1208 struct ofport *ofport;
1209 int error;
1210
1211 ofport = ofproto_get_port(ofproto, ofp_port);
1212 if (!ofport) {
1213 VLOG_WARN("%s: cannot configure bfd on nonexistent port %"PRIu16,
1214 ofproto->name, ofp_port);
1215 return;
1216 }
1217
1218 error = (ofproto->ofproto_class->set_bfd
1219 ? ofproto->ofproto_class->set_bfd(ofport, cfg)
1220 : EOPNOTSUPP);
1221 if (error) {
1222 VLOG_WARN("%s: bfd configuration on port %"PRIu16" (%s) failed (%s)",
1223 ofproto->name, ofp_port, netdev_get_name(ofport->netdev),
1224 ovs_strerror(error));
1225 }
1226 }
1227
1228 /* Checks the status change of BFD on 'ofport'.
1229 *
1230 * Returns true if 'ofproto_class' does not support 'bfd_status_changed'. */
1231 bool
1232 ofproto_port_bfd_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
1233 {
1234 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1235 return (ofport && ofproto->ofproto_class->bfd_status_changed
1236 ? ofproto->ofproto_class->bfd_status_changed(ofport)
1237 : true);
1238 }
1239
1240 /* Populates 'status' with the status of BFD on 'ofport'. Returns 0 on
1241 * success. Returns a positive errno otherwise. Has no effect if 'ofp_port'
1242 * is not an OpenFlow port in 'ofproto'.
1243 *
1244 * The caller must provide and own '*status'. */
1245 int
1246 ofproto_port_get_bfd_status(struct ofproto *ofproto, ofp_port_t ofp_port,
1247 struct smap *status)
1248 {
1249 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1250 return (ofport && ofproto->ofproto_class->get_bfd_status
1251 ? ofproto->ofproto_class->get_bfd_status(ofport, status)
1252 : EOPNOTSUPP);
1253 }
1254
1255 /* Checks the status of LACP negotiation for 'ofp_port' within ofproto.
1256 * Returns 1 if LACP partner information for 'ofp_port' is up-to-date,
1257 * 0 if LACP partner information is not current (generally indicating a
1258 * connectivity problem), or -1 if LACP is not enabled on 'ofp_port'. */
1259 int
1260 ofproto_port_is_lacp_current(struct ofproto *ofproto, ofp_port_t ofp_port)
1261 {
1262 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
1263 return (ofport && ofproto->ofproto_class->port_is_lacp_current
1264 ? ofproto->ofproto_class->port_is_lacp_current(ofport)
1265 : -1);
1266 }
1267
1268 int
1269 ofproto_port_get_lacp_stats(const struct ofport *port, struct lacp_slave_stats *stats)
1270 {
1271 struct ofproto *ofproto = port->ofproto;
1272 int error;
1273
1274 if (ofproto->ofproto_class->port_get_lacp_stats) {
1275 error = ofproto->ofproto_class->port_get_lacp_stats(port, stats);
1276 } else {
1277 error = EOPNOTSUPP;
1278 }
1279
1280 return error;
1281 }
1282 \f
1283 /* Bundles. */
1284
1285 /* Registers a "bundle" associated with client data pointer 'aux' in 'ofproto'.
1286 * A bundle is the same concept as a Port in OVSDB, that is, it consists of one
1287 * or more "slave" devices (Interfaces, in OVSDB) along with a VLAN
1288 * configuration plus, if there is more than one slave, a bonding
1289 * configuration.
1290 *
1291 * If 'aux' is already registered then this function updates its configuration
1292 * to 's'. Otherwise, this function registers a new bundle.
1293 *
1294 * Bundles only affect the NXAST_AUTOPATH action and output to the OFPP_NORMAL
1295 * port. */
1296 int
1297 ofproto_bundle_register(struct ofproto *ofproto, void *aux,
1298 const struct ofproto_bundle_settings *s)
1299 {
1300 return (ofproto->ofproto_class->bundle_set
1301 ? ofproto->ofproto_class->bundle_set(ofproto, aux, s)
1302 : EOPNOTSUPP);
1303 }
1304
1305 /* Unregisters the bundle registered on 'ofproto' with auxiliary data 'aux'.
1306 * If no such bundle has been registered, this has no effect. */
1307 int
1308 ofproto_bundle_unregister(struct ofproto *ofproto, void *aux)
1309 {
1310 return ofproto_bundle_register(ofproto, aux, NULL);
1311 }
1312
1313 \f
1314 /* Registers a mirror associated with client data pointer 'aux' in 'ofproto'.
1315 * If 'aux' is already registered then this function updates its configuration
1316 * to 's'. Otherwise, this function registers a new mirror. */
1317 int
1318 ofproto_mirror_register(struct ofproto *ofproto, void *aux,
1319 const struct ofproto_mirror_settings *s)
1320 {
1321 return (ofproto->ofproto_class->mirror_set
1322 ? ofproto->ofproto_class->mirror_set(ofproto, aux, s)
1323 : EOPNOTSUPP);
1324 }
1325
1326 /* Unregisters the mirror registered on 'ofproto' with auxiliary data 'aux'.
1327 * If no mirror has been registered, this has no effect. */
1328 int
1329 ofproto_mirror_unregister(struct ofproto *ofproto, void *aux)
1330 {
1331 return ofproto_mirror_register(ofproto, aux, NULL);
1332 }
1333
1334 /* Retrieves statistics from mirror associated with client data pointer
1335 * 'aux' in 'ofproto'. Stores packet and byte counts in 'packets' and
1336 * 'bytes', respectively. If a particular counters is not supported,
1337 * the appropriate argument is set to UINT64_MAX.
1338 */
1339 int
1340 ofproto_mirror_get_stats(struct ofproto *ofproto, void *aux,
1341 uint64_t *packets, uint64_t *bytes)
1342 {
1343 if (!ofproto->ofproto_class->mirror_get_stats) {
1344 *packets = *bytes = UINT64_MAX;
1345 return EOPNOTSUPP;
1346 }
1347
1348 return ofproto->ofproto_class->mirror_get_stats(ofproto, aux,
1349 packets, bytes);
1350 }
1351
1352 /* Configures the VLANs whose bits are set to 1 in 'flood_vlans' as VLANs on
1353 * which all packets are flooded, instead of using MAC learning. If
1354 * 'flood_vlans' is NULL, then MAC learning applies to all VLANs.
1355 *
1356 * Flood VLANs affect only the treatment of packets output to the OFPP_NORMAL
1357 * port. */
1358 int
1359 ofproto_set_flood_vlans(struct ofproto *ofproto, unsigned long *flood_vlans)
1360 {
1361 return (ofproto->ofproto_class->set_flood_vlans
1362 ? ofproto->ofproto_class->set_flood_vlans(ofproto, flood_vlans)
1363 : EOPNOTSUPP);
1364 }
1365
1366 /* Returns true if 'aux' is a registered bundle that is currently in use as the
1367 * output for a mirror. */
1368 bool
1369 ofproto_is_mirror_output_bundle(const struct ofproto *ofproto, void *aux)
1370 {
1371 return (ofproto->ofproto_class->is_mirror_output_bundle
1372 ? ofproto->ofproto_class->is_mirror_output_bundle(ofproto, aux)
1373 : false);
1374 }
1375 \f
1376 /* Configuration of OpenFlow tables. */
1377
1378 /* Returns the number of OpenFlow tables in 'ofproto'. */
1379 int
1380 ofproto_get_n_tables(const struct ofproto *ofproto)
1381 {
1382 return ofproto->n_tables;
1383 }
1384
1385 /* Returns the number of Controller visible OpenFlow tables
1386 * in 'ofproto'. This number will exclude Hidden tables.
1387 * This funtion's return value should be less or equal to that of
1388 * ofproto_get_n_tables() . */
1389 uint8_t
1390 ofproto_get_n_visible_tables(const struct ofproto *ofproto)
1391 {
1392 uint8_t n = ofproto->n_tables;
1393
1394 /* Count only non-hidden tables in the number of tables. (Hidden tables,
1395 * if present, are always at the end.) */
1396 while(n && (ofproto->tables[n - 1].flags & OFTABLE_HIDDEN)) {
1397 n--;
1398 }
1399
1400 return n;
1401 }
1402
1403 /* Configures the OpenFlow table in 'ofproto' with id 'table_id' with the
1404 * settings from 's'. 'table_id' must be in the range 0 through the number of
1405 * OpenFlow tables in 'ofproto' minus 1, inclusive.
1406 *
1407 * For read-only tables, only the name may be configured. */
1408 void
1409 ofproto_configure_table(struct ofproto *ofproto, int table_id,
1410 const struct ofproto_table_settings *s)
1411 {
1412 struct oftable *table;
1413
1414 ovs_assert(table_id >= 0 && table_id < ofproto->n_tables);
1415 table = &ofproto->tables[table_id];
1416
1417 oftable_set_name(table, s->name);
1418
1419 if (table->flags & OFTABLE_READONLY) {
1420 return;
1421 }
1422
1423 if (classifier_set_prefix_fields(&table->cls,
1424 s->prefix_fields, s->n_prefix_fields)) {
1425 /* XXX: Trigger revalidation. */
1426 }
1427
1428 ovs_mutex_lock(&ofproto_mutex);
1429 unsigned int new_eviction = (s->enable_eviction
1430 ? table->eviction | EVICTION_CLIENT
1431 : table->eviction & ~EVICTION_CLIENT);
1432 oftable_configure_eviction(table, new_eviction, s->groups, s->n_groups);
1433 table->max_flows = s->max_flows;
1434 evict_rules_from_table(table);
1435 ovs_mutex_unlock(&ofproto_mutex);
1436 }
1437 \f
1438 bool
1439 ofproto_has_snoops(const struct ofproto *ofproto)
1440 {
1441 return connmgr_has_snoops(ofproto->connmgr);
1442 }
1443
1444 void
1445 ofproto_get_snoops(const struct ofproto *ofproto, struct sset *snoops)
1446 {
1447 connmgr_get_snoops(ofproto->connmgr, snoops);
1448 }
1449
1450 /* Deletes 'rule' from 'ofproto'.
1451 *
1452 * Within an ofproto implementation, this function allows an ofproto
1453 * implementation to destroy any rules that remain when its ->destruct()
1454 * function is called. This function is not suitable for use elsewhere in an
1455 * ofproto implementation.
1456 *
1457 * This function implements steps 4.4 and 4.5 in the section titled "Rule Life
1458 * Cycle" in ofproto-provider.h. */
1459 void
1460 ofproto_rule_delete(struct ofproto *ofproto, struct rule *rule)
1461 OVS_EXCLUDED(ofproto_mutex)
1462 {
1463 /* This skips the ofmonitor and flow-removed notifications because the
1464 * switch is being deleted and any OpenFlow channels have been or soon will
1465 * be killed. */
1466 ovs_mutex_lock(&ofproto_mutex);
1467
1468 if (!rule->removed) {
1469 /* Make sure there is no postponed removal of the rule. */
1470 ovs_assert(cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
1471
1472 if (!classifier_remove(&rule->ofproto->tables[rule->table_id].cls,
1473 &rule->cr)) {
1474 OVS_NOT_REACHED();
1475 }
1476 ofproto_rule_remove__(rule->ofproto, rule);
1477 ofproto->ofproto_class->rule_delete(rule);
1478 ofproto_rule_unref(rule);
1479 }
1480 ovs_mutex_unlock(&ofproto_mutex);
1481 }
1482
1483 static void
1484 ofproto_flush__(struct ofproto *ofproto)
1485 OVS_EXCLUDED(ofproto_mutex)
1486 {
1487 struct oftable *table;
1488
1489 /* This will flush all datapath flows. */
1490 if (ofproto->ofproto_class->flush) {
1491 ofproto->ofproto_class->flush(ofproto);
1492 }
1493
1494 /* XXX: There is a small race window here, where new datapath flows can be
1495 * created by upcall handlers based on the existing flow table. We can not
1496 * call ofproto class flush while holding 'ofproto_mutex' to prevent this,
1497 * as then we could deadlock on syncing with the handler threads waiting on
1498 * the same mutex. */
1499
1500 ovs_mutex_lock(&ofproto_mutex);
1501 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1502 struct rule_collection rules;
1503 struct rule *rule;
1504
1505 if (table->flags & OFTABLE_HIDDEN) {
1506 continue;
1507 }
1508
1509 rule_collection_init(&rules);
1510
1511 CLS_FOR_EACH (rule, cr, &table->cls) {
1512 rule_collection_add(&rules, rule);
1513 }
1514 delete_flows__(&rules, OFPRR_DELETE, NULL);
1515 }
1516 /* XXX: Concurrent handler threads may insert new learned flows based on
1517 * learn actions of the now deleted flows right after we release
1518 * 'ofproto_mutex'. */
1519 ovs_mutex_unlock(&ofproto_mutex);
1520 }
1521
1522 static void delete_group(struct ofproto *ofproto, uint32_t group_id);
1523
1524 static void
1525 ofproto_destroy__(struct ofproto *ofproto)
1526 OVS_EXCLUDED(ofproto_mutex)
1527 {
1528 struct oftable *table;
1529
1530 destroy_rule_executes(ofproto);
1531
1532 guarded_list_destroy(&ofproto->rule_executes);
1533 ovs_rwlock_destroy(&ofproto->groups_rwlock);
1534 hmap_destroy(&ofproto->groups);
1535
1536 hmap_remove(&all_ofprotos, &ofproto->hmap_node);
1537 free(ofproto->name);
1538 free(ofproto->type);
1539 free(ofproto->mfr_desc);
1540 free(ofproto->hw_desc);
1541 free(ofproto->sw_desc);
1542 free(ofproto->serial_desc);
1543 free(ofproto->dp_desc);
1544 hmap_destroy(&ofproto->ports);
1545 hmap_destroy(&ofproto->ofport_usage);
1546 shash_destroy(&ofproto->port_by_name);
1547 simap_destroy(&ofproto->ofp_requests);
1548
1549 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1550 oftable_destroy(table);
1551 }
1552 free(ofproto->tables);
1553
1554 ovs_assert(hindex_is_empty(&ofproto->cookies));
1555 hindex_destroy(&ofproto->cookies);
1556
1557 ovs_assert(hmap_is_empty(&ofproto->learned_cookies));
1558 hmap_destroy(&ofproto->learned_cookies);
1559
1560 free(ofproto->vlan_bitmap);
1561
1562 ofproto->ofproto_class->dealloc(ofproto);
1563 }
1564
1565 /* Destroying rules is doubly deferred, must have 'ofproto' around for them.
1566 * - 1st we defer the removal of the rules from the classifier
1567 * - 2nd we defer the actual destruction of the rules. */
1568 static void
1569 ofproto_destroy_defer__(struct ofproto *ofproto)
1570 OVS_EXCLUDED(ofproto_mutex)
1571 {
1572 ovsrcu_postpone(ofproto_destroy__, ofproto);
1573 }
1574
1575 void
1576 ofproto_destroy(struct ofproto *p, bool del)
1577 OVS_EXCLUDED(ofproto_mutex)
1578 {
1579 struct ofport *ofport, *next_ofport;
1580 struct ofport_usage *usage, *next_usage;
1581
1582 if (!p) {
1583 return;
1584 }
1585
1586 if (p->meters) {
1587 meter_delete(p, 1, p->meter_features.max_meters);
1588 p->meter_features.max_meters = 0;
1589 free(p->meters);
1590 p->meters = NULL;
1591 }
1592
1593 ofproto_flush__(p);
1594 HMAP_FOR_EACH_SAFE (ofport, next_ofport, hmap_node, &p->ports) {
1595 ofport_destroy(ofport, del);
1596 }
1597
1598 HMAP_FOR_EACH_SAFE (usage, next_usage, hmap_node, &p->ofport_usage) {
1599 hmap_remove(&p->ofport_usage, &usage->hmap_node);
1600 free(usage);
1601 }
1602
1603 p->ofproto_class->destruct(p);
1604
1605 /* We should not postpone this because it involves deleting a listening
1606 * socket which we may want to reopen soon. 'connmgr' should not be used
1607 * by other threads */
1608 connmgr_destroy(p->connmgr);
1609
1610 /* Destroying rules is deferred, must have 'ofproto' around for them. */
1611 ovsrcu_postpone(ofproto_destroy_defer__, p);
1612 }
1613
1614 /* Destroys the datapath with the respective 'name' and 'type'. With the Linux
1615 * kernel datapath, for example, this destroys the datapath in the kernel, and
1616 * with the netdev-based datapath, it tears down the data structures that
1617 * represent the datapath.
1618 *
1619 * The datapath should not be currently open as an ofproto. */
1620 int
1621 ofproto_delete(const char *name, const char *type)
1622 {
1623 const struct ofproto_class *class = ofproto_class_find__(type);
1624 return (!class ? EAFNOSUPPORT
1625 : !class->del ? EACCES
1626 : class->del(type, name));
1627 }
1628
1629 static void
1630 process_port_change(struct ofproto *ofproto, int error, char *devname)
1631 {
1632 if (error == ENOBUFS) {
1633 reinit_ports(ofproto);
1634 } else if (!error) {
1635 update_port(ofproto, devname);
1636 free(devname);
1637 }
1638 }
1639
1640 int
1641 ofproto_type_run(const char *datapath_type)
1642 {
1643 const struct ofproto_class *class;
1644 int error;
1645
1646 datapath_type = ofproto_normalize_type(datapath_type);
1647 class = ofproto_class_find__(datapath_type);
1648
1649 error = class->type_run ? class->type_run(datapath_type) : 0;
1650 if (error && error != EAGAIN) {
1651 VLOG_ERR_RL(&rl, "%s: type_run failed (%s)",
1652 datapath_type, ovs_strerror(error));
1653 }
1654 return error;
1655 }
1656
1657 void
1658 ofproto_type_wait(const char *datapath_type)
1659 {
1660 const struct ofproto_class *class;
1661
1662 datapath_type = ofproto_normalize_type(datapath_type);
1663 class = ofproto_class_find__(datapath_type);
1664
1665 if (class->type_wait) {
1666 class->type_wait(datapath_type);
1667 }
1668 }
1669
1670 int
1671 ofproto_run(struct ofproto *p)
1672 {
1673 int error;
1674 uint64_t new_seq;
1675
1676 error = p->ofproto_class->run(p);
1677 if (error && error != EAGAIN) {
1678 VLOG_ERR_RL(&rl, "%s: run failed (%s)", p->name, ovs_strerror(error));
1679 }
1680
1681 run_rule_executes(p);
1682
1683 /* Restore the eviction group heap invariant occasionally. */
1684 if (p->eviction_group_timer < time_msec()) {
1685 size_t i;
1686
1687 p->eviction_group_timer = time_msec() + 1000;
1688
1689 for (i = 0; i < p->n_tables; i++) {
1690 struct oftable *table = &p->tables[i];
1691 struct eviction_group *evg;
1692 struct rule *rule;
1693
1694 if (!table->eviction) {
1695 continue;
1696 }
1697
1698 if (table->n_flows > 100000) {
1699 static struct vlog_rate_limit count_rl =
1700 VLOG_RATE_LIMIT_INIT(1, 1);
1701 VLOG_WARN_RL(&count_rl, "Table %"PRIuSIZE" has an excessive"
1702 " number of rules: %d", i, table->n_flows);
1703 }
1704
1705 ovs_mutex_lock(&ofproto_mutex);
1706 CLS_FOR_EACH (rule, cr, &table->cls) {
1707 if (rule->idle_timeout || rule->hard_timeout) {
1708 if (!rule->eviction_group) {
1709 eviction_group_add_rule(rule);
1710 } else {
1711 heap_raw_change(&rule->evg_node,
1712 rule_eviction_priority(p, rule));
1713 }
1714 }
1715 }
1716
1717 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
1718 heap_rebuild(&evg->rules);
1719 }
1720 ovs_mutex_unlock(&ofproto_mutex);
1721 }
1722 }
1723
1724 if (p->ofproto_class->port_poll) {
1725 char *devname;
1726
1727 while ((error = p->ofproto_class->port_poll(p, &devname)) != EAGAIN) {
1728 process_port_change(p, error, devname);
1729 }
1730 }
1731
1732 new_seq = seq_read(connectivity_seq_get());
1733 if (new_seq != p->change_seq) {
1734 struct sset devnames;
1735 const char *devname;
1736 struct ofport *ofport;
1737
1738 /* Update OpenFlow port status for any port whose netdev has changed.
1739 *
1740 * Refreshing a given 'ofport' can cause an arbitrary ofport to be
1741 * destroyed, so it's not safe to update ports directly from the
1742 * HMAP_FOR_EACH loop, or even to use HMAP_FOR_EACH_SAFE. Instead, we
1743 * need this two-phase approach. */
1744 sset_init(&devnames);
1745 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
1746 uint64_t port_change_seq;
1747
1748 port_change_seq = netdev_get_change_seq(ofport->netdev);
1749 if (ofport->change_seq != port_change_seq) {
1750 ofport->change_seq = port_change_seq;
1751 sset_add(&devnames, netdev_get_name(ofport->netdev));
1752 }
1753 }
1754 SSET_FOR_EACH (devname, &devnames) {
1755 update_port(p, devname);
1756 }
1757 sset_destroy(&devnames);
1758
1759 p->change_seq = new_seq;
1760 }
1761
1762 connmgr_run(p->connmgr, handle_openflow);
1763
1764 return error;
1765 }
1766
1767 void
1768 ofproto_wait(struct ofproto *p)
1769 {
1770 p->ofproto_class->wait(p);
1771 if (p->ofproto_class->port_poll_wait) {
1772 p->ofproto_class->port_poll_wait(p);
1773 }
1774 seq_wait(connectivity_seq_get(), p->change_seq);
1775 connmgr_wait(p->connmgr);
1776 }
1777
1778 bool
1779 ofproto_is_alive(const struct ofproto *p)
1780 {
1781 return connmgr_has_controllers(p->connmgr);
1782 }
1783
1784 /* Adds some memory usage statistics for 'ofproto' into 'usage', for use with
1785 * memory_report(). */
1786 void
1787 ofproto_get_memory_usage(const struct ofproto *ofproto, struct simap *usage)
1788 {
1789 const struct oftable *table;
1790 unsigned int n_rules;
1791
1792 simap_increase(usage, "ports", hmap_count(&ofproto->ports));
1793
1794 n_rules = 0;
1795 OFPROTO_FOR_EACH_TABLE (table, ofproto) {
1796 n_rules += table->n_flows;
1797 }
1798 simap_increase(usage, "rules", n_rules);
1799
1800 if (ofproto->ofproto_class->get_memory_usage) {
1801 ofproto->ofproto_class->get_memory_usage(ofproto, usage);
1802 }
1803
1804 connmgr_get_memory_usage(ofproto->connmgr, usage);
1805 }
1806
1807 void
1808 ofproto_type_get_memory_usage(const char *datapath_type, struct simap *usage)
1809 {
1810 const struct ofproto_class *class;
1811
1812 datapath_type = ofproto_normalize_type(datapath_type);
1813 class = ofproto_class_find__(datapath_type);
1814
1815 if (class && class->type_get_memory_usage) {
1816 class->type_get_memory_usage(datapath_type, usage);
1817 }
1818 }
1819
1820 void
1821 ofproto_get_ofproto_controller_info(const struct ofproto *ofproto,
1822 struct shash *info)
1823 {
1824 connmgr_get_controller_info(ofproto->connmgr, info);
1825 }
1826
1827 void
1828 ofproto_free_ofproto_controller_info(struct shash *info)
1829 {
1830 connmgr_free_controller_info(info);
1831 }
1832
1833 /* Makes a deep copy of 'old' into 'port'. */
1834 void
1835 ofproto_port_clone(struct ofproto_port *port, const struct ofproto_port *old)
1836 {
1837 port->name = xstrdup(old->name);
1838 port->type = xstrdup(old->type);
1839 port->ofp_port = old->ofp_port;
1840 }
1841
1842 /* Frees memory allocated to members of 'ofproto_port'.
1843 *
1844 * Do not call this function on an ofproto_port obtained from
1845 * ofproto_port_dump_next(): that function retains ownership of the data in the
1846 * ofproto_port. */
1847 void
1848 ofproto_port_destroy(struct ofproto_port *ofproto_port)
1849 {
1850 free(ofproto_port->name);
1851 free(ofproto_port->type);
1852 }
1853
1854 /* Initializes 'dump' to begin dumping the ports in an ofproto.
1855 *
1856 * This function provides no status indication. An error status for the entire
1857 * dump operation is provided when it is completed by calling
1858 * ofproto_port_dump_done().
1859 */
1860 void
1861 ofproto_port_dump_start(struct ofproto_port_dump *dump,
1862 const struct ofproto *ofproto)
1863 {
1864 dump->ofproto = ofproto;
1865 dump->error = ofproto->ofproto_class->port_dump_start(ofproto,
1866 &dump->state);
1867 }
1868
1869 /* Attempts to retrieve another port from 'dump', which must have been created
1870 * with ofproto_port_dump_start(). On success, stores a new ofproto_port into
1871 * 'port' and returns true. On failure, returns false.
1872 *
1873 * Failure might indicate an actual error or merely that the last port has been
1874 * dumped. An error status for the entire dump operation is provided when it
1875 * is completed by calling ofproto_port_dump_done().
1876 *
1877 * The ofproto owns the data stored in 'port'. It will remain valid until at
1878 * least the next time 'dump' is passed to ofproto_port_dump_next() or
1879 * ofproto_port_dump_done(). */
1880 bool
1881 ofproto_port_dump_next(struct ofproto_port_dump *dump,
1882 struct ofproto_port *port)
1883 {
1884 const struct ofproto *ofproto = dump->ofproto;
1885
1886 if (dump->error) {
1887 return false;
1888 }
1889
1890 dump->error = ofproto->ofproto_class->port_dump_next(ofproto, dump->state,
1891 port);
1892 if (dump->error) {
1893 ofproto->ofproto_class->port_dump_done(ofproto, dump->state);
1894 return false;
1895 }
1896 return true;
1897 }
1898
1899 /* Completes port table dump operation 'dump', which must have been created
1900 * with ofproto_port_dump_start(). Returns 0 if the dump operation was
1901 * error-free, otherwise a positive errno value describing the problem. */
1902 int
1903 ofproto_port_dump_done(struct ofproto_port_dump *dump)
1904 {
1905 const struct ofproto *ofproto = dump->ofproto;
1906 if (!dump->error) {
1907 dump->error = ofproto->ofproto_class->port_dump_done(ofproto,
1908 dump->state);
1909 }
1910 return dump->error == EOF ? 0 : dump->error;
1911 }
1912
1913 /* Returns the type to pass to netdev_open() when a datapath of type
1914 * 'datapath_type' has a port of type 'port_type', for a few special
1915 * cases when a netdev type differs from a port type. For example, when
1916 * using the userspace datapath, a port of type "internal" needs to be
1917 * opened as "tap".
1918 *
1919 * Returns either 'type' itself or a string literal, which must not be
1920 * freed. */
1921 const char *
1922 ofproto_port_open_type(const char *datapath_type, const char *port_type)
1923 {
1924 const struct ofproto_class *class;
1925
1926 datapath_type = ofproto_normalize_type(datapath_type);
1927 class = ofproto_class_find__(datapath_type);
1928 if (!class) {
1929 return port_type;
1930 }
1931
1932 return (class->port_open_type
1933 ? class->port_open_type(datapath_type, port_type)
1934 : port_type);
1935 }
1936
1937 /* Attempts to add 'netdev' as a port on 'ofproto'. If 'ofp_portp' is
1938 * non-null and '*ofp_portp' is not OFPP_NONE, attempts to use that as
1939 * the port's OpenFlow port number.
1940 *
1941 * If successful, returns 0 and sets '*ofp_portp' to the new port's
1942 * OpenFlow port number (if 'ofp_portp' is non-null). On failure,
1943 * returns a positive errno value and sets '*ofp_portp' to OFPP_NONE (if
1944 * 'ofp_portp' is non-null). */
1945 int
1946 ofproto_port_add(struct ofproto *ofproto, struct netdev *netdev,
1947 ofp_port_t *ofp_portp)
1948 {
1949 ofp_port_t ofp_port = ofp_portp ? *ofp_portp : OFPP_NONE;
1950 int error;
1951
1952 error = ofproto->ofproto_class->port_add(ofproto, netdev);
1953 if (!error) {
1954 const char *netdev_name = netdev_get_name(netdev);
1955
1956 simap_put(&ofproto->ofp_requests, netdev_name,
1957 ofp_to_u16(ofp_port));
1958 error = update_port(ofproto, netdev_name);
1959 }
1960 if (ofp_portp) {
1961 *ofp_portp = OFPP_NONE;
1962 if (!error) {
1963 struct ofproto_port ofproto_port;
1964
1965 error = ofproto_port_query_by_name(ofproto,
1966 netdev_get_name(netdev),
1967 &ofproto_port);
1968 if (!error) {
1969 *ofp_portp = ofproto_port.ofp_port;
1970 ofproto_port_destroy(&ofproto_port);
1971 }
1972 }
1973 }
1974 return error;
1975 }
1976
1977 /* Looks up a port named 'devname' in 'ofproto'. On success, returns 0 and
1978 * initializes '*port' appropriately; on failure, returns a positive errno
1979 * value.
1980 *
1981 * The caller owns the data in 'ofproto_port' and must free it with
1982 * ofproto_port_destroy() when it is no longer needed. */
1983 int
1984 ofproto_port_query_by_name(const struct ofproto *ofproto, const char *devname,
1985 struct ofproto_port *port)
1986 {
1987 int error;
1988
1989 error = ofproto->ofproto_class->port_query_by_name(ofproto, devname, port);
1990 if (error) {
1991 memset(port, 0, sizeof *port);
1992 }
1993 return error;
1994 }
1995
1996 /* Deletes port number 'ofp_port' from the datapath for 'ofproto'.
1997 * Returns 0 if successful, otherwise a positive errno. */
1998 int
1999 ofproto_port_del(struct ofproto *ofproto, ofp_port_t ofp_port)
2000 {
2001 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
2002 const char *name = ofport ? netdev_get_name(ofport->netdev) : "<unknown>";
2003 struct simap_node *ofp_request_node;
2004 int error;
2005
2006 ofp_request_node = simap_find(&ofproto->ofp_requests, name);
2007 if (ofp_request_node) {
2008 simap_delete(&ofproto->ofp_requests, ofp_request_node);
2009 }
2010
2011 error = ofproto->ofproto_class->port_del(ofproto, ofp_port);
2012 if (!error && ofport) {
2013 /* 'name' is the netdev's name and update_port() is going to close the
2014 * netdev. Just in case update_port() refers to 'name' after it
2015 * destroys 'ofport', make a copy of it around the update_port()
2016 * call. */
2017 char *devname = xstrdup(name);
2018 update_port(ofproto, devname);
2019 free(devname);
2020 }
2021 return error;
2022 }
2023
2024 static void
2025 flow_mod_init(struct ofputil_flow_mod *fm,
2026 const struct match *match, int priority,
2027 const struct ofpact *ofpacts, size_t ofpacts_len,
2028 enum ofp_flow_mod_command command)
2029 {
2030 *fm = (struct ofputil_flow_mod) {
2031 .match = *match,
2032 .priority = priority,
2033 .table_id = 0,
2034 .command = command,
2035 .buffer_id = UINT32_MAX,
2036 .out_port = OFPP_ANY,
2037 .out_group = OFPG_ANY,
2038 .ofpacts = CONST_CAST(struct ofpact *, ofpacts),
2039 .ofpacts_len = ofpacts_len,
2040 .delete_reason = OFPRR_DELETE,
2041 };
2042 }
2043
2044 static int
2045 simple_flow_mod(struct ofproto *ofproto,
2046 const struct match *match, int priority,
2047 const struct ofpact *ofpacts, size_t ofpacts_len,
2048 enum ofp_flow_mod_command command)
2049 {
2050 struct ofproto_flow_mod ofm;
2051
2052 flow_mod_init(&ofm.fm, match, priority, ofpacts, ofpacts_len, command);
2053
2054 return handle_flow_mod__(ofproto, &ofm, NULL);
2055 }
2056
2057 /* Adds a flow to OpenFlow flow table 0 in 'p' that matches 'cls_rule' and
2058 * performs the 'n_actions' actions in 'actions'. The new flow will not
2059 * timeout.
2060 *
2061 * If cls_rule->priority is in the range of priorities supported by OpenFlow
2062 * (0...65535, inclusive) then the flow will be visible to OpenFlow
2063 * controllers; otherwise, it will be hidden.
2064 *
2065 * The caller retains ownership of 'cls_rule' and 'ofpacts'.
2066 *
2067 * This is a helper function for in-band control and fail-open. */
2068 void
2069 ofproto_add_flow(struct ofproto *ofproto, const struct match *match,
2070 int priority,
2071 const struct ofpact *ofpacts, size_t ofpacts_len)
2072 OVS_EXCLUDED(ofproto_mutex)
2073 {
2074 const struct rule *rule;
2075 bool must_add;
2076
2077 /* First do a cheap check whether the rule we're looking for already exists
2078 * with the actions that we want. If it does, then we're done. */
2079 rule = rule_from_cls_rule(classifier_find_match_exactly(
2080 &ofproto->tables[0].cls, match, priority,
2081 CLS_MAX_VERSION));
2082 if (rule) {
2083 const struct rule_actions *actions = rule_get_actions(rule);
2084 must_add = !ofpacts_equal(actions->ofpacts, actions->ofpacts_len,
2085 ofpacts, ofpacts_len);
2086 } else {
2087 must_add = true;
2088 }
2089
2090 /* If there's no such rule or the rule doesn't have the actions we want,
2091 * fall back to a executing a full flow mod. We can't optimize this at
2092 * all because we didn't take enough locks above to ensure that the flow
2093 * table didn't already change beneath us. */
2094 if (must_add) {
2095 simple_flow_mod(ofproto, match, priority, ofpacts, ofpacts_len,
2096 OFPFC_MODIFY_STRICT);
2097 }
2098 }
2099
2100 /* Executes the flow modification specified in 'fm'. Returns 0 on success, or
2101 * an OFPERR_* OpenFlow error code on failure.
2102 *
2103 * This is a helper function for in-band control and fail-open and the "learn"
2104 * action. */
2105 enum ofperr
2106 ofproto_flow_mod(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
2107 OVS_EXCLUDED(ofproto_mutex)
2108 {
2109 struct ofputil_flow_mod *fm = &ofm->fm;
2110
2111 /* Optimize for the most common case of a repeated learn action.
2112 * If an identical flow already exists we only need to update its
2113 * 'modified' time. */
2114 if (fm->command == OFPFC_MODIFY_STRICT && fm->table_id != OFPTT_ALL
2115 && !(fm->flags & OFPUTIL_FF_RESET_COUNTS)) {
2116 struct oftable *table = &ofproto->tables[fm->table_id];
2117 struct rule *rule;
2118 bool done = false;
2119
2120 rule = rule_from_cls_rule(classifier_find_match_exactly(
2121 &table->cls, &fm->match, fm->priority,
2122 CLS_MAX_VERSION));
2123 if (rule) {
2124 /* Reading many of the rule fields and writing on 'modified'
2125 * requires the rule->mutex. Also, rule->actions may change
2126 * if rule->mutex is not held. */
2127 const struct rule_actions *actions;
2128
2129 ovs_mutex_lock(&rule->mutex);
2130 actions = rule_get_actions(rule);
2131 if (rule->idle_timeout == fm->idle_timeout
2132 && rule->hard_timeout == fm->hard_timeout
2133 && rule->importance == fm->importance
2134 && rule->flags == (fm->flags & OFPUTIL_FF_STATE)
2135 && (!fm->modify_cookie || (fm->new_cookie == rule->flow_cookie))
2136 && ofpacts_equal(fm->ofpacts, fm->ofpacts_len,
2137 actions->ofpacts, actions->ofpacts_len)) {
2138 /* Rule already exists and need not change, only update the
2139 modified timestamp. */
2140 rule->modified = time_msec();
2141 done = true;
2142 }
2143 ovs_mutex_unlock(&rule->mutex);
2144 }
2145
2146 if (done) {
2147 return 0;
2148 }
2149 }
2150
2151 return handle_flow_mod__(ofproto, ofm, NULL);
2152 }
2153
2154 /* Searches for a rule with matching criteria exactly equal to 'target' in
2155 * ofproto's table 0 and, if it finds one, deletes it.
2156 *
2157 * This is a helper function for in-band control and fail-open. */
2158 void
2159 ofproto_delete_flow(struct ofproto *ofproto,
2160 const struct match *target, int priority)
2161 OVS_EXCLUDED(ofproto_mutex)
2162 {
2163 struct classifier *cls = &ofproto->tables[0].cls;
2164 struct rule *rule;
2165
2166 /* First do a cheap check whether the rule we're looking for has already
2167 * been deleted. If so, then we're done. */
2168 rule = rule_from_cls_rule(classifier_find_match_exactly(
2169 cls, target, priority, CLS_MAX_VERSION));
2170 if (!rule) {
2171 return;
2172 }
2173
2174 /* Execute a flow mod. We can't optimize this at all because we didn't
2175 * take enough locks above to ensure that the flow table didn't already
2176 * change beneath us. */
2177 simple_flow_mod(ofproto, target, priority, NULL, 0, OFPFC_DELETE_STRICT);
2178 }
2179
2180 /* Delete all of the flows from all of ofproto's flow tables, then reintroduce
2181 * the flows required by in-band control and fail-open. */
2182 void
2183 ofproto_flush_flows(struct ofproto *ofproto)
2184 {
2185 COVERAGE_INC(ofproto_flush);
2186 ofproto_flush__(ofproto);
2187 connmgr_flushed(ofproto->connmgr);
2188 }
2189 \f
2190 static void
2191 reinit_ports(struct ofproto *p)
2192 {
2193 struct ofproto_port_dump dump;
2194 struct sset devnames;
2195 struct ofport *ofport;
2196 struct ofproto_port ofproto_port;
2197 const char *devname;
2198
2199 COVERAGE_INC(ofproto_reinit_ports);
2200
2201 sset_init(&devnames);
2202 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2203 sset_add(&devnames, netdev_get_name(ofport->netdev));
2204 }
2205 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2206 sset_add(&devnames, ofproto_port.name);
2207 }
2208
2209 SSET_FOR_EACH (devname, &devnames) {
2210 update_port(p, devname);
2211 }
2212 sset_destroy(&devnames);
2213 }
2214
2215 static ofp_port_t
2216 alloc_ofp_port(struct ofproto *ofproto, const char *netdev_name)
2217 {
2218 uint16_t port_idx;
2219
2220 port_idx = simap_get(&ofproto->ofp_requests, netdev_name);
2221 port_idx = port_idx ? port_idx : UINT16_MAX;
2222
2223 if (port_idx >= ofproto->max_ports
2224 || ofport_get_usage(ofproto, u16_to_ofp(port_idx)) == LLONG_MAX) {
2225 uint16_t lru_ofport = 0, end_port_no = ofproto->alloc_port_no;
2226 long long int last_used_at, lru = LLONG_MAX;
2227
2228 /* Search for a free OpenFlow port number. We try not to
2229 * immediately reuse them to prevent problems due to old
2230 * flows.
2231 *
2232 * We limit the automatically assigned port numbers to the lower half
2233 * of the port range, to reserve the upper half for assignment by
2234 * controllers. */
2235 for (;;) {
2236 if (++ofproto->alloc_port_no >= MIN(ofproto->max_ports, 32768)) {
2237 ofproto->alloc_port_no = 1;
2238 }
2239 last_used_at = ofport_get_usage(ofproto,
2240 u16_to_ofp(ofproto->alloc_port_no));
2241 if (!last_used_at) {
2242 port_idx = ofproto->alloc_port_no;
2243 break;
2244 } else if ( last_used_at < time_msec() - 60*60*1000) {
2245 /* If the port with ofport 'ofproto->alloc_port_no' was deleted
2246 * more than an hour ago, consider it usable. */
2247 ofport_remove_usage(ofproto,
2248 u16_to_ofp(ofproto->alloc_port_no));
2249 port_idx = ofproto->alloc_port_no;
2250 break;
2251 } else if (last_used_at < lru) {
2252 lru = last_used_at;
2253 lru_ofport = ofproto->alloc_port_no;
2254 }
2255
2256 if (ofproto->alloc_port_no == end_port_no) {
2257 if (lru_ofport) {
2258 port_idx = lru_ofport;
2259 break;
2260 }
2261 return OFPP_NONE;
2262 }
2263 }
2264 }
2265 ofport_set_usage(ofproto, u16_to_ofp(port_idx), LLONG_MAX);
2266 return u16_to_ofp(port_idx);
2267 }
2268
2269 static void
2270 dealloc_ofp_port(struct ofproto *ofproto, ofp_port_t ofp_port)
2271 {
2272 if (ofp_to_u16(ofp_port) < ofproto->max_ports) {
2273 ofport_set_usage(ofproto, ofp_port, time_msec());
2274 }
2275 }
2276
2277 /* Opens and returns a netdev for 'ofproto_port' in 'ofproto', or a null
2278 * pointer if the netdev cannot be opened. On success, also fills in
2279 * '*pp'. */
2280 static struct netdev *
2281 ofport_open(struct ofproto *ofproto,
2282 struct ofproto_port *ofproto_port,
2283 struct ofputil_phy_port *pp)
2284 {
2285 enum netdev_flags flags;
2286 struct netdev *netdev;
2287 int error;
2288
2289 error = netdev_open(ofproto_port->name, ofproto_port->type, &netdev);
2290 if (error) {
2291 VLOG_WARN_RL(&rl, "%s: ignoring port %s (%"PRIu16") because netdev %s "
2292 "cannot be opened (%s)",
2293 ofproto->name,
2294 ofproto_port->name, ofproto_port->ofp_port,
2295 ofproto_port->name, ovs_strerror(error));
2296 return NULL;
2297 }
2298
2299 if (ofproto_port->ofp_port == OFPP_NONE) {
2300 if (!strcmp(ofproto->name, ofproto_port->name)) {
2301 ofproto_port->ofp_port = OFPP_LOCAL;
2302 } else {
2303 ofproto_port->ofp_port = alloc_ofp_port(ofproto,
2304 ofproto_port->name);
2305 }
2306 }
2307 pp->port_no = ofproto_port->ofp_port;
2308 netdev_get_etheraddr(netdev, &pp->hw_addr);
2309 ovs_strlcpy(pp->name, ofproto_port->name, sizeof pp->name);
2310 netdev_get_flags(netdev, &flags);
2311 pp->config = flags & NETDEV_UP ? 0 : OFPUTIL_PC_PORT_DOWN;
2312 pp->state = netdev_get_carrier(netdev) ? 0 : OFPUTIL_PS_LINK_DOWN;
2313 netdev_get_features(netdev, &pp->curr, &pp->advertised,
2314 &pp->supported, &pp->peer);
2315 pp->curr_speed = netdev_features_to_bps(pp->curr, 0) / 1000;
2316 pp->max_speed = netdev_features_to_bps(pp->supported, 0) / 1000;
2317
2318 return netdev;
2319 }
2320
2321 /* Returns true if most fields of 'a' and 'b' are equal. Differences in name,
2322 * port number, and 'config' bits other than OFPUTIL_PC_PORT_DOWN are
2323 * disregarded. */
2324 static bool
2325 ofport_equal(const struct ofputil_phy_port *a,
2326 const struct ofputil_phy_port *b)
2327 {
2328 return (eth_addr_equals(a->hw_addr, b->hw_addr)
2329 && a->state == b->state
2330 && !((a->config ^ b->config) & OFPUTIL_PC_PORT_DOWN)
2331 && a->curr == b->curr
2332 && a->advertised == b->advertised
2333 && a->supported == b->supported
2334 && a->peer == b->peer
2335 && a->curr_speed == b->curr_speed
2336 && a->max_speed == b->max_speed);
2337 }
2338
2339 /* Adds an ofport to 'p' initialized based on the given 'netdev' and 'opp'.
2340 * The caller must ensure that 'p' does not have a conflicting ofport (that is,
2341 * one with the same name or port number). */
2342 static int
2343 ofport_install(struct ofproto *p,
2344 struct netdev *netdev, const struct ofputil_phy_port *pp)
2345 {
2346 const char *netdev_name = netdev_get_name(netdev);
2347 struct ofport *ofport;
2348 int error;
2349
2350 /* Create ofport. */
2351 ofport = p->ofproto_class->port_alloc();
2352 if (!ofport) {
2353 error = ENOMEM;
2354 goto error;
2355 }
2356 ofport->ofproto = p;
2357 ofport->netdev = netdev;
2358 ofport->change_seq = netdev_get_change_seq(netdev);
2359 ofport->pp = *pp;
2360 ofport->ofp_port = pp->port_no;
2361 ofport->created = time_msec();
2362
2363 /* Add port to 'p'. */
2364 hmap_insert(&p->ports, &ofport->hmap_node,
2365 hash_ofp_port(ofport->ofp_port));
2366 shash_add(&p->port_by_name, netdev_name, ofport);
2367
2368 update_mtu(p, ofport);
2369
2370 /* Let the ofproto_class initialize its private data. */
2371 error = p->ofproto_class->port_construct(ofport);
2372 if (error) {
2373 goto error;
2374 }
2375 connmgr_send_port_status(p->connmgr, NULL, pp, OFPPR_ADD);
2376 return 0;
2377
2378 error:
2379 VLOG_WARN_RL(&rl, "%s: could not add port %s (%s)",
2380 p->name, netdev_name, ovs_strerror(error));
2381 if (ofport) {
2382 ofport_destroy__(ofport);
2383 } else {
2384 netdev_close(netdev);
2385 }
2386 return error;
2387 }
2388
2389 /* Removes 'ofport' from 'p' and destroys it. */
2390 static void
2391 ofport_remove(struct ofport *ofport)
2392 {
2393 connmgr_send_port_status(ofport->ofproto->connmgr, NULL, &ofport->pp,
2394 OFPPR_DELETE);
2395 ofport_destroy(ofport, true);
2396 }
2397
2398 /* If 'ofproto' contains an ofport named 'name', removes it from 'ofproto' and
2399 * destroys it. */
2400 static void
2401 ofport_remove_with_name(struct ofproto *ofproto, const char *name)
2402 {
2403 struct ofport *port = shash_find_data(&ofproto->port_by_name, name);
2404 if (port) {
2405 ofport_remove(port);
2406 }
2407 }
2408
2409 /* Updates 'port' with new 'pp' description.
2410 *
2411 * Does not handle a name or port number change. The caller must implement
2412 * such a change as a delete followed by an add. */
2413 static void
2414 ofport_modified(struct ofport *port, struct ofputil_phy_port *pp)
2415 {
2416 port->pp.hw_addr = pp->hw_addr;
2417 port->pp.config = ((port->pp.config & ~OFPUTIL_PC_PORT_DOWN)
2418 | (pp->config & OFPUTIL_PC_PORT_DOWN));
2419 port->pp.state = ((port->pp.state & ~OFPUTIL_PS_LINK_DOWN)
2420 | (pp->state & OFPUTIL_PS_LINK_DOWN));
2421 port->pp.curr = pp->curr;
2422 port->pp.advertised = pp->advertised;
2423 port->pp.supported = pp->supported;
2424 port->pp.peer = pp->peer;
2425 port->pp.curr_speed = pp->curr_speed;
2426 port->pp.max_speed = pp->max_speed;
2427
2428 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2429 &port->pp, OFPPR_MODIFY);
2430 }
2431
2432 /* Update OpenFlow 'state' in 'port' and notify controller. */
2433 void
2434 ofproto_port_set_state(struct ofport *port, enum ofputil_port_state state)
2435 {
2436 if (port->pp.state != state) {
2437 port->pp.state = state;
2438 connmgr_send_port_status(port->ofproto->connmgr, NULL,
2439 &port->pp, OFPPR_MODIFY);
2440 }
2441 }
2442
2443 void
2444 ofproto_port_unregister(struct ofproto *ofproto, ofp_port_t ofp_port)
2445 {
2446 struct ofport *port = ofproto_get_port(ofproto, ofp_port);
2447 if (port) {
2448 if (port->ofproto->ofproto_class->set_realdev) {
2449 port->ofproto->ofproto_class->set_realdev(port, 0, 0);
2450 }
2451 if (port->ofproto->ofproto_class->set_stp_port) {
2452 port->ofproto->ofproto_class->set_stp_port(port, NULL);
2453 }
2454 if (port->ofproto->ofproto_class->set_rstp_port) {
2455 port->ofproto->ofproto_class->set_rstp_port(port, NULL);
2456 }
2457 if (port->ofproto->ofproto_class->set_cfm) {
2458 port->ofproto->ofproto_class->set_cfm(port, NULL);
2459 }
2460 if (port->ofproto->ofproto_class->bundle_remove) {
2461 port->ofproto->ofproto_class->bundle_remove(port);
2462 }
2463 }
2464 }
2465
2466 static void
2467 ofport_destroy__(struct ofport *port)
2468 {
2469 struct ofproto *ofproto = port->ofproto;
2470 const char *name = netdev_get_name(port->netdev);
2471
2472 hmap_remove(&ofproto->ports, &port->hmap_node);
2473 shash_delete(&ofproto->port_by_name,
2474 shash_find(&ofproto->port_by_name, name));
2475
2476 netdev_close(port->netdev);
2477 ofproto->ofproto_class->port_dealloc(port);
2478 }
2479
2480 static void
2481 ofport_destroy(struct ofport *port, bool del)
2482 {
2483 if (port) {
2484 dealloc_ofp_port(port->ofproto, port->ofp_port);
2485 port->ofproto->ofproto_class->port_destruct(port, del);
2486 ofport_destroy__(port);
2487 }
2488 }
2489
2490 struct ofport *
2491 ofproto_get_port(const struct ofproto *ofproto, ofp_port_t ofp_port)
2492 {
2493 struct ofport *port;
2494
2495 HMAP_FOR_EACH_IN_BUCKET (port, hmap_node, hash_ofp_port(ofp_port),
2496 &ofproto->ports) {
2497 if (port->ofp_port == ofp_port) {
2498 return port;
2499 }
2500 }
2501 return NULL;
2502 }
2503
2504 static long long int
2505 ofport_get_usage(const struct ofproto *ofproto, ofp_port_t ofp_port)
2506 {
2507 struct ofport_usage *usage;
2508
2509 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2510 &ofproto->ofport_usage) {
2511 if (usage->ofp_port == ofp_port) {
2512 return usage->last_used;
2513 }
2514 }
2515 return 0;
2516 }
2517
2518 static void
2519 ofport_set_usage(struct ofproto *ofproto, ofp_port_t ofp_port,
2520 long long int last_used)
2521 {
2522 struct ofport_usage *usage;
2523 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2524 &ofproto->ofport_usage) {
2525 if (usage->ofp_port == ofp_port) {
2526 usage->last_used = last_used;
2527 return;
2528 }
2529 }
2530 ovs_assert(last_used == LLONG_MAX);
2531
2532 usage = xmalloc(sizeof *usage);
2533 usage->ofp_port = ofp_port;
2534 usage->last_used = last_used;
2535 hmap_insert(&ofproto->ofport_usage, &usage->hmap_node,
2536 hash_ofp_port(ofp_port));
2537 }
2538
2539 static void
2540 ofport_remove_usage(struct ofproto *ofproto, ofp_port_t ofp_port)
2541 {
2542 struct ofport_usage *usage;
2543 HMAP_FOR_EACH_IN_BUCKET (usage, hmap_node, hash_ofp_port(ofp_port),
2544 &ofproto->ofport_usage) {
2545 if (usage->ofp_port == ofp_port) {
2546 hmap_remove(&ofproto->ofport_usage, &usage->hmap_node);
2547 free(usage);
2548 break;
2549 }
2550 }
2551 }
2552
2553 int
2554 ofproto_port_get_stats(const struct ofport *port, struct netdev_stats *stats)
2555 {
2556 struct ofproto *ofproto = port->ofproto;
2557 int error;
2558
2559 if (ofproto->ofproto_class->port_get_stats) {
2560 error = ofproto->ofproto_class->port_get_stats(port, stats);
2561 } else {
2562 error = EOPNOTSUPP;
2563 }
2564
2565 return error;
2566 }
2567
2568 static int
2569 update_port(struct ofproto *ofproto, const char *name)
2570 {
2571 struct ofproto_port ofproto_port;
2572 struct ofputil_phy_port pp;
2573 struct netdev *netdev;
2574 struct ofport *port;
2575 int error = 0;
2576
2577 COVERAGE_INC(ofproto_update_port);
2578
2579 /* Fetch 'name''s location and properties from the datapath. */
2580 netdev = (!ofproto_port_query_by_name(ofproto, name, &ofproto_port)
2581 ? ofport_open(ofproto, &ofproto_port, &pp)
2582 : NULL);
2583
2584 if (netdev) {
2585 port = ofproto_get_port(ofproto, ofproto_port.ofp_port);
2586 if (port && !strcmp(netdev_get_name(port->netdev), name)) {
2587 struct netdev *old_netdev = port->netdev;
2588
2589 /* 'name' hasn't changed location. Any properties changed? */
2590 if (!ofport_equal(&port->pp, &pp)) {
2591 ofport_modified(port, &pp);
2592 }
2593
2594 update_mtu(ofproto, port);
2595
2596 /* Install the newly opened netdev in case it has changed.
2597 * Don't close the old netdev yet in case port_modified has to
2598 * remove a retained reference to it.*/
2599 port->netdev = netdev;
2600 port->change_seq = netdev_get_change_seq(netdev);
2601
2602 if (port->ofproto->ofproto_class->port_modified) {
2603 port->ofproto->ofproto_class->port_modified(port);
2604 }
2605
2606 netdev_close(old_netdev);
2607 } else {
2608 /* If 'port' is nonnull then its name differs from 'name' and thus
2609 * we should delete it. If we think there's a port named 'name'
2610 * then its port number must be wrong now so delete it too. */
2611 if (port) {
2612 ofport_remove(port);
2613 }
2614 ofport_remove_with_name(ofproto, name);
2615 error = ofport_install(ofproto, netdev, &pp);
2616 }
2617 } else {
2618 /* Any port named 'name' is gone now. */
2619 ofport_remove_with_name(ofproto, name);
2620 }
2621 ofproto_port_destroy(&ofproto_port);
2622
2623 return error;
2624 }
2625
2626 static int
2627 init_ports(struct ofproto *p)
2628 {
2629 struct ofproto_port_dump dump;
2630 struct ofproto_port ofproto_port;
2631 struct shash_node *node, *next;
2632
2633 OFPROTO_PORT_FOR_EACH (&ofproto_port, &dump, p) {
2634 const char *name = ofproto_port.name;
2635
2636 if (shash_find(&p->port_by_name, name)) {
2637 VLOG_WARN_RL(&rl, "%s: ignoring duplicate device %s in datapath",
2638 p->name, name);
2639 } else {
2640 struct ofputil_phy_port pp;
2641 struct netdev *netdev;
2642
2643 /* Check if an OpenFlow port number had been requested. */
2644 node = shash_find(&init_ofp_ports, name);
2645 if (node) {
2646 const struct iface_hint *iface_hint = node->data;
2647 simap_put(&p->ofp_requests, name,
2648 ofp_to_u16(iface_hint->ofp_port));
2649 }
2650
2651 netdev = ofport_open(p, &ofproto_port, &pp);
2652 if (netdev) {
2653 ofport_install(p, netdev, &pp);
2654 if (ofp_to_u16(ofproto_port.ofp_port) < p->max_ports) {
2655 p->alloc_port_no = MAX(p->alloc_port_no,
2656 ofp_to_u16(ofproto_port.ofp_port));
2657 }
2658 }
2659 }
2660 }
2661
2662 SHASH_FOR_EACH_SAFE(node, next, &init_ofp_ports) {
2663 struct iface_hint *iface_hint = node->data;
2664
2665 if (!strcmp(iface_hint->br_name, p->name)) {
2666 free(iface_hint->br_name);
2667 free(iface_hint->br_type);
2668 free(iface_hint);
2669 shash_delete(&init_ofp_ports, node);
2670 }
2671 }
2672
2673 return 0;
2674 }
2675
2676 /* Find the minimum MTU of all non-datapath devices attached to 'p'.
2677 * Returns ETH_PAYLOAD_MAX or the minimum of the ports. */
2678 static int
2679 find_min_mtu(struct ofproto *p)
2680 {
2681 struct ofport *ofport;
2682 int mtu = 0;
2683
2684 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2685 struct netdev *netdev = ofport->netdev;
2686 int dev_mtu;
2687
2688 /* Skip any internal ports, since that's what we're trying to
2689 * set. */
2690 if (!strcmp(netdev_get_type(netdev), "internal")) {
2691 continue;
2692 }
2693
2694 if (netdev_get_mtu(netdev, &dev_mtu)) {
2695 continue;
2696 }
2697 if (!mtu || dev_mtu < mtu) {
2698 mtu = dev_mtu;
2699 }
2700 }
2701
2702 return mtu ? mtu: ETH_PAYLOAD_MAX;
2703 }
2704
2705 /* Update MTU of all datapath devices on 'p' to the minimum of the
2706 * non-datapath ports in event of 'port' added or changed. */
2707 static void
2708 update_mtu(struct ofproto *p, struct ofport *port)
2709 {
2710 struct ofport *ofport;
2711 struct netdev *netdev = port->netdev;
2712 int dev_mtu, old_min;
2713
2714 if (netdev_get_mtu(netdev, &dev_mtu)) {
2715 port->mtu = 0;
2716 return;
2717 }
2718 if (!strcmp(netdev_get_type(port->netdev), "internal")) {
2719 if (dev_mtu > p->min_mtu) {
2720 if (!netdev_set_mtu(port->netdev, p->min_mtu)) {
2721 dev_mtu = p->min_mtu;
2722 }
2723 }
2724 port->mtu = dev_mtu;
2725 return;
2726 }
2727
2728 /* For non-internal port find new min mtu. */
2729 old_min = p->min_mtu;
2730 port->mtu = dev_mtu;
2731 p->min_mtu = find_min_mtu(p);
2732 if (p->min_mtu == old_min) {
2733 return;
2734 }
2735
2736 HMAP_FOR_EACH (ofport, hmap_node, &p->ports) {
2737 struct netdev *netdev = ofport->netdev;
2738
2739 if (!strcmp(netdev_get_type(netdev), "internal")) {
2740 if (!netdev_set_mtu(netdev, p->min_mtu)) {
2741 ofport->mtu = p->min_mtu;
2742 }
2743 }
2744 }
2745 }
2746 \f
2747 static void
2748 ofproto_rule_destroy__(struct rule *rule)
2749 OVS_NO_THREAD_SAFETY_ANALYSIS
2750 {
2751 cls_rule_destroy(CONST_CAST(struct cls_rule *, &rule->cr));
2752 rule_actions_destroy(rule_get_actions(rule));
2753 ovs_mutex_destroy(&rule->mutex);
2754 rule->ofproto->ofproto_class->rule_dealloc(rule);
2755 }
2756
2757 static void
2758 rule_destroy_cb(struct rule *rule)
2759 OVS_NO_THREAD_SAFETY_ANALYSIS
2760 {
2761 /* Send rule removed if needed. */
2762 if (rule->flags & OFPUTIL_FF_SEND_FLOW_REM
2763 && rule->removed_reason != OVS_OFPRR_NONE
2764 && !rule_is_hidden(rule)) {
2765 ofproto_rule_send_removed(rule);
2766 }
2767 rule->ofproto->ofproto_class->rule_destruct(rule);
2768 ofproto_rule_destroy__(rule);
2769 }
2770
2771 void
2772 ofproto_rule_ref(struct rule *rule)
2773 {
2774 if (rule) {
2775 ovs_refcount_ref(&rule->ref_count);
2776 }
2777 }
2778
2779 bool
2780 ofproto_rule_try_ref(struct rule *rule)
2781 {
2782 if (rule) {
2783 return ovs_refcount_try_ref_rcu(&rule->ref_count);
2784 }
2785 return false;
2786 }
2787
2788 /* Decrements 'rule''s ref_count and schedules 'rule' to be destroyed if the
2789 * ref_count reaches 0.
2790 *
2791 * Use of RCU allows short term use (between RCU quiescent periods) without
2792 * keeping a reference. A reference must be taken if the rule needs to
2793 * stay around accross the RCU quiescent periods. */
2794 void
2795 ofproto_rule_unref(struct rule *rule)
2796 {
2797 if (rule && ovs_refcount_unref_relaxed(&rule->ref_count) == 1) {
2798 ovsrcu_postpone(rule_destroy_cb, rule);
2799 }
2800 }
2801
2802 static void
2803 remove_rule_rcu__(struct rule *rule)
2804 OVS_REQUIRES(ofproto_mutex)
2805 {
2806 struct ofproto *ofproto = rule->ofproto;
2807 struct oftable *table = &ofproto->tables[rule->table_id];
2808
2809 ovs_assert(!cls_rule_visible_in_version(&rule->cr, CLS_MAX_VERSION));
2810 if (!classifier_remove(&table->cls, &rule->cr)) {
2811 OVS_NOT_REACHED();
2812 }
2813 ofproto->ofproto_class->rule_delete(rule);
2814 ofproto_rule_unref(rule);
2815 }
2816
2817 static void
2818 remove_rule_rcu(struct rule *rule)
2819 OVS_EXCLUDED(ofproto_mutex)
2820 {
2821 ovs_mutex_lock(&ofproto_mutex);
2822 remove_rule_rcu__(rule);
2823 ovs_mutex_unlock(&ofproto_mutex);
2824 }
2825
2826 /* Removes and deletes rules from a NULL-terminated array of rule pointers. */
2827 static void
2828 remove_rules_rcu(struct rule **rules)
2829 OVS_EXCLUDED(ofproto_mutex)
2830 {
2831 struct rule **orig_rules = rules;
2832
2833 if (*rules) {
2834 struct ofproto *ofproto = rules[0]->ofproto;
2835 unsigned long tables[BITMAP_N_LONGS(256)];
2836 struct rule *rule;
2837 size_t table_id;
2838
2839 memset(tables, 0, sizeof tables);
2840
2841 ovs_mutex_lock(&ofproto_mutex);
2842 while ((rule = *rules++)) {
2843 /* Defer once for each new table. This defers the subtable cleanup
2844 * until later, so that when removing large number of flows the
2845 * operation is faster. */
2846 if (!bitmap_is_set(tables, rule->table_id)) {
2847 struct classifier *cls = &ofproto->tables[rule->table_id].cls;
2848
2849 bitmap_set1(tables, rule->table_id);
2850 classifier_defer(cls);
2851 }
2852 remove_rule_rcu__(rule);
2853 }
2854
2855 BITMAP_FOR_EACH_1(table_id, 256, tables) {
2856 struct classifier *cls = &ofproto->tables[table_id].cls;
2857
2858 classifier_publish(cls);
2859 }
2860 ovs_mutex_unlock(&ofproto_mutex);
2861 }
2862
2863 free(orig_rules);
2864 }
2865
2866 void
2867 ofproto_group_ref(struct ofgroup *group)
2868 {
2869 if (group) {
2870 ovs_refcount_ref(&group->ref_count);
2871 }
2872 }
2873
2874 void
2875 ofproto_group_unref(struct ofgroup *group)
2876 {
2877 if (group && ovs_refcount_unref(&group->ref_count) == 1) {
2878 group->ofproto->ofproto_class->group_destruct(group);
2879 ofputil_bucket_list_destroy(&group->buckets);
2880 group->ofproto->ofproto_class->group_dealloc(group);
2881 }
2882 }
2883
2884 static uint32_t get_provider_meter_id(const struct ofproto *,
2885 uint32_t of_meter_id);
2886
2887 /* Creates and returns a new 'struct rule_actions', whose actions are a copy
2888 * of from the 'ofpacts_len' bytes of 'ofpacts'. */
2889 const struct rule_actions *
2890 rule_actions_create(const struct ofpact *ofpacts, size_t ofpacts_len)
2891 {
2892 struct rule_actions *actions;
2893
2894 actions = xmalloc(sizeof *actions + ofpacts_len);
2895 actions->ofpacts_len = ofpacts_len;
2896 actions->has_meter = ofpacts_get_meter(ofpacts, ofpacts_len) != 0;
2897 memcpy(actions->ofpacts, ofpacts, ofpacts_len);
2898
2899 actions->has_learn_with_delete = (next_learn_with_delete(actions, NULL)
2900 != NULL);
2901
2902 return actions;
2903 }
2904
2905 /* Free the actions after the RCU quiescent period is reached. */
2906 void
2907 rule_actions_destroy(const struct rule_actions *actions)
2908 {
2909 if (actions) {
2910 ovsrcu_postpone(free, CONST_CAST(struct rule_actions *, actions));
2911 }
2912 }
2913
2914 /* Returns true if 'rule' has an OpenFlow OFPAT_OUTPUT or OFPAT_ENQUEUE action
2915 * that outputs to 'port' (output to OFPP_FLOOD and OFPP_ALL doesn't count). */
2916 bool
2917 ofproto_rule_has_out_port(const struct rule *rule, ofp_port_t port)
2918 OVS_REQUIRES(ofproto_mutex)
2919 {
2920 if (port == OFPP_ANY) {
2921 return true;
2922 } else {
2923 const struct rule_actions *actions = rule_get_actions(rule);
2924 return ofpacts_output_to_port(actions->ofpacts,
2925 actions->ofpacts_len, port);
2926 }
2927 }
2928
2929 /* Returns true if 'rule' has group and equals group_id. */
2930 static bool
2931 ofproto_rule_has_out_group(const struct rule *rule, uint32_t group_id)
2932 OVS_REQUIRES(ofproto_mutex)
2933 {
2934 if (group_id == OFPG_ANY) {
2935 return true;
2936 } else {
2937 const struct rule_actions *actions = rule_get_actions(rule);
2938 return ofpacts_output_to_group(actions->ofpacts,
2939 actions->ofpacts_len, group_id);
2940 }
2941 }
2942
2943 static void
2944 rule_execute_destroy(struct rule_execute *e)
2945 {
2946 ofproto_rule_unref(e->rule);
2947 ovs_list_remove(&e->list_node);
2948 free(e);
2949 }
2950
2951 /* Executes all "rule_execute" operations queued up in ofproto->rule_executes,
2952 * by passing them to the ofproto provider. */
2953 static void
2954 run_rule_executes(struct ofproto *ofproto)
2955 OVS_EXCLUDED(ofproto_mutex)
2956 {
2957 struct rule_execute *e, *next;
2958 struct ovs_list executes;
2959
2960 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2961 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2962 struct flow flow;
2963
2964 flow_extract(e->packet, &flow);
2965 flow.in_port.ofp_port = e->in_port;
2966 ofproto->ofproto_class->rule_execute(e->rule, &flow, e->packet);
2967
2968 rule_execute_destroy(e);
2969 }
2970 }
2971
2972 /* Destroys and discards all "rule_execute" operations queued up in
2973 * ofproto->rule_executes. */
2974 static void
2975 destroy_rule_executes(struct ofproto *ofproto)
2976 {
2977 struct rule_execute *e, *next;
2978 struct ovs_list executes;
2979
2980 guarded_list_pop_all(&ofproto->rule_executes, &executes);
2981 LIST_FOR_EACH_SAFE (e, next, list_node, &executes) {
2982 dp_packet_delete(e->packet);
2983 rule_execute_destroy(e);
2984 }
2985 }
2986
2987 static bool
2988 rule_is_readonly(const struct rule *rule)
2989 {
2990 const struct oftable *table = &rule->ofproto->tables[rule->table_id];
2991 return (table->flags & OFTABLE_READONLY) != 0;
2992 }
2993 \f
2994 static uint32_t
2995 hash_learned_cookie(ovs_be64 cookie_, uint8_t table_id)
2996 {
2997 uint64_t cookie = (OVS_FORCE uint64_t) cookie_;
2998 return hash_3words(cookie, cookie >> 32, table_id);
2999 }
3000
3001 static void
3002 learned_cookies_update_one__(struct ofproto *ofproto,
3003 const struct ofpact_learn *learn,
3004 int delta, struct ovs_list *dead_cookies)
3005 OVS_REQUIRES(ofproto_mutex)
3006 {
3007 uint32_t hash = hash_learned_cookie(learn->cookie, learn->table_id);
3008 struct learned_cookie *c;
3009
3010 HMAP_FOR_EACH_WITH_HASH (c, u.hmap_node, hash, &ofproto->learned_cookies) {
3011 if (c->cookie == learn->cookie && c->table_id == learn->table_id) {
3012 c->n += delta;
3013 ovs_assert(c->n >= 0);
3014
3015 if (!c->n) {
3016 hmap_remove(&ofproto->learned_cookies, &c->u.hmap_node);
3017 ovs_list_push_back(dead_cookies, &c->u.list_node);
3018 }
3019
3020 return;
3021 }
3022 }
3023
3024 ovs_assert(delta > 0);
3025 c = xmalloc(sizeof *c);
3026 hmap_insert(&ofproto->learned_cookies, &c->u.hmap_node, hash);
3027 c->cookie = learn->cookie;
3028 c->table_id = learn->table_id;
3029 c->n = delta;
3030 }
3031
3032 static const struct ofpact_learn *
3033 next_learn_with_delete(const struct rule_actions *actions,
3034 const struct ofpact_learn *start)
3035 {
3036 const struct ofpact *pos;
3037
3038 for (pos = start ? ofpact_next(&start->ofpact) : actions->ofpacts;
3039 pos < ofpact_end(actions->ofpacts, actions->ofpacts_len);
3040 pos = ofpact_next(pos)) {
3041 if (pos->type == OFPACT_LEARN) {
3042 const struct ofpact_learn *learn = ofpact_get_LEARN(pos);
3043 if (learn->flags & NX_LEARN_F_DELETE_LEARNED) {
3044 return learn;
3045 }
3046 }
3047 }
3048
3049 return NULL;
3050 }
3051
3052 static void
3053 learned_cookies_update__(struct ofproto *ofproto,
3054 const struct rule_actions *actions,
3055 int delta, struct ovs_list *dead_cookies)
3056 OVS_REQUIRES(ofproto_mutex)
3057 {
3058 if (actions->has_learn_with_delete) {
3059 const struct ofpact_learn *learn;
3060
3061 for (learn = next_learn_with_delete(actions, NULL); learn;
3062 learn = next_learn_with_delete(actions, learn)) {
3063 learned_cookies_update_one__(ofproto, learn, delta, dead_cookies);
3064 }
3065 }
3066 }
3067
3068 static void
3069 learned_cookies_inc(struct ofproto *ofproto,
3070 const struct rule_actions *actions)
3071 OVS_REQUIRES(ofproto_mutex)
3072 {
3073 learned_cookies_update__(ofproto, actions, +1, NULL);
3074 }
3075
3076 static void
3077 learned_cookies_dec(struct ofproto *ofproto,
3078 const struct rule_actions *actions,
3079 struct ovs_list *dead_cookies)
3080 OVS_REQUIRES(ofproto_mutex)
3081 {
3082 learned_cookies_update__(ofproto, actions, -1, dead_cookies);
3083 }
3084
3085 static void
3086 learned_cookies_flush(struct ofproto *ofproto, struct ovs_list *dead_cookies)
3087 OVS_REQUIRES(ofproto_mutex)
3088 {
3089 struct learned_cookie *c;
3090
3091 LIST_FOR_EACH_POP (c, u.list_node, dead_cookies) {
3092 struct rule_criteria criteria;
3093 struct rule_collection rules;
3094 struct match match;
3095
3096 match_init_catchall(&match);
3097 rule_criteria_init(&criteria, c->table_id, &match, 0, CLS_MAX_VERSION,
3098 c->cookie, OVS_BE64_MAX, OFPP_ANY, OFPG_ANY);
3099 rule_criteria_require_rw(&criteria, false);
3100 collect_rules_loose(ofproto, &criteria, &rules);
3101 rule_criteria_destroy(&criteria);
3102 delete_flows__(&rules, OFPRR_DELETE, NULL);
3103
3104 free(c);
3105 }
3106 }
3107 \f
3108 static enum ofperr
3109 handle_echo_request(struct ofconn *ofconn, const struct ofp_header *oh)
3110 {
3111 ofconn_send_reply(ofconn, make_echo_reply(oh));
3112 return 0;
3113 }
3114
3115 static void
3116 query_tables(struct ofproto *ofproto,
3117 struct ofputil_table_features **featuresp,
3118 struct ofputil_table_stats **statsp)
3119 {
3120 struct mf_bitmap rw_fields = oxm_writable_fields();
3121 struct mf_bitmap match = oxm_matchable_fields();
3122 struct mf_bitmap mask = oxm_maskable_fields();
3123
3124 struct ofputil_table_features *features;
3125 struct ofputil_table_stats *stats;
3126 int i;
3127
3128 features = *featuresp = xcalloc(ofproto->n_tables, sizeof *features);
3129 for (i = 0; i < ofproto->n_tables; i++) {
3130 struct ofputil_table_features *f = &features[i];
3131
3132 f->table_id = i;
3133 sprintf(f->name, "table%d", i);
3134 f->metadata_match = OVS_BE64_MAX;
3135 f->metadata_write = OVS_BE64_MAX;
3136 atomic_read_relaxed(&ofproto->tables[i].miss_config, &f->miss_config);
3137 f->max_entries = 1000000;
3138
3139 bool more_tables = false;
3140 for (int j = i + 1; j < ofproto->n_tables; j++) {
3141 if (!(ofproto->tables[j].flags & OFTABLE_HIDDEN)) {
3142 bitmap_set1(f->nonmiss.next, j);
3143 more_tables = true;
3144 }
3145 }
3146 f->nonmiss.instructions = (1u << N_OVS_INSTRUCTIONS) - 1;
3147 if (!more_tables) {
3148 f->nonmiss.instructions &= ~(1u << OVSINST_OFPIT11_GOTO_TABLE);
3149 }
3150 f->nonmiss.write.ofpacts = (UINT64_C(1) << N_OFPACTS) - 1;
3151 f->nonmiss.write.set_fields = rw_fields;
3152 f->nonmiss.apply = f->nonmiss.write;
3153 f->miss = f->nonmiss;
3154
3155 f->match = match;
3156 f->mask = mask;
3157 f->wildcard = match;
3158 }
3159
3160 if (statsp) {
3161 stats = *statsp = xcalloc(ofproto->n_tables, sizeof *stats);
3162 for (i = 0; i < ofproto->n_tables; i++) {
3163 struct ofputil_table_stats *s = &stats[i];
3164
3165 s->table_id = i;
3166 s->active_count = ofproto->tables[i].n_flows;
3167 if (i == 0) {
3168 s->active_count -= connmgr_count_hidden_rules(
3169 ofproto->connmgr);
3170 }
3171 }
3172 } else {
3173 stats = NULL;
3174 }
3175
3176 ofproto->ofproto_class->query_tables(ofproto, features, stats);
3177
3178 for (i = 0; i < ofproto->n_tables; i++) {
3179 const struct oftable *table = &ofproto->tables[i];
3180 struct ofputil_table_features *f = &features[i];
3181
3182 if (table->name) {
3183 ovs_strzcpy(f->name, table->name, sizeof f->name);
3184 }
3185
3186 if (table->max_flows < f->max_entries) {
3187 f->max_entries = table->max_flows;
3188 }
3189 }
3190 }
3191
3192 static void
3193 query_switch_features(struct ofproto *ofproto,
3194 bool *arp_match_ip, uint64_t *ofpacts)
3195 {
3196 struct ofputil_table_features *features, *f;
3197
3198 *arp_match_ip = false;
3199 *ofpacts = 0;
3200
3201 query_tables(ofproto, &features, NULL);
3202 for (f = features; f < &features[ofproto->n_tables]; f++) {
3203 *ofpacts |= f->nonmiss.apply.ofpacts | f->miss.apply.ofpacts;
3204 if (bitmap_is_set(f->match.bm, MFF_ARP_SPA) ||
3205 bitmap_is_set(f->match.bm, MFF_ARP_TPA)) {
3206 *arp_match_ip = true;
3207 }
3208 }
3209 free(features);
3210
3211 /* Sanity check. */
3212 ovs_assert(*ofpacts & (UINT64_C(1) << OFPACT_OUTPUT));
3213 }
3214
3215 static enum ofperr
3216 handle_features_request(struct ofconn *ofconn, const struct ofp_header *oh)
3217 {
3218 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3219 struct ofputil_switch_features features;
3220 struct ofport *port;
3221 bool arp_match_ip;
3222 struct ofpbuf *b;
3223
3224 query_switch_features(ofproto, &arp_match_ip, &features.ofpacts);
3225
3226 features.datapath_id = ofproto->datapath_id;
3227 features.n_buffers = pktbuf_capacity();
3228 features.n_tables = ofproto_get_n_visible_tables(ofproto);
3229 features.capabilities = (OFPUTIL_C_FLOW_STATS | OFPUTIL_C_TABLE_STATS |
3230 OFPUTIL_C_PORT_STATS | OFPUTIL_C_QUEUE_STATS |
3231 OFPUTIL_C_GROUP_STATS);
3232 if (arp_match_ip) {
3233 features.capabilities |= OFPUTIL_C_ARP_MATCH_IP;
3234 }
3235 /* FIXME: Fill in proper features.auxiliary_id for auxiliary connections */
3236 features.auxiliary_id = 0;
3237 b = ofputil_encode_switch_features(&features, ofconn_get_protocol(ofconn),
3238 oh->xid);
3239 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3240 ofputil_put_switch_features_port(&port->pp, b);
3241 }
3242
3243 ofconn_send_reply(ofconn, b);
3244 return 0;
3245 }
3246
3247 static enum ofperr
3248 handle_get_config_request(struct ofconn *ofconn, const struct ofp_header *oh)
3249 {
3250 struct ofputil_switch_config config;
3251 config.frag = ofconn_get_ofproto(ofconn)->frag_handling;
3252 config.invalid_ttl_to_controller
3253 = ofconn_get_invalid_ttl_to_controller(ofconn);
3254 config.miss_send_len = ofconn_get_miss_send_len(ofconn);
3255
3256 ofconn_send_reply(ofconn, ofputil_encode_get_config_reply(oh, &config));
3257
3258 return 0;
3259 }
3260
3261 static enum ofperr
3262 handle_set_config(struct ofconn *ofconn, const struct ofp_header *oh)
3263 {
3264 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3265 struct ofputil_switch_config config;
3266 enum ofperr error;
3267
3268 error = ofputil_decode_set_config(oh, &config);
3269 if (error) {
3270 return error;
3271 }
3272
3273 if (ofconn_get_type(ofconn) != OFCONN_PRIMARY
3274 || ofconn_get_role(ofconn) != OFPCR12_ROLE_SLAVE) {
3275 enum ofputil_frag_handling cur = ofproto->frag_handling;
3276 enum ofputil_frag_handling next = config.frag;
3277
3278 if (cur != next) {
3279 if (ofproto->ofproto_class->set_frag_handling(ofproto, next)) {
3280 ofproto->frag_handling = next;
3281 } else {
3282 VLOG_WARN_RL(&rl, "%s: unsupported fragment handling mode %s",
3283 ofproto->name,
3284 ofputil_frag_handling_to_string(next));
3285 }
3286 }
3287 }
3288
3289 if (config.invalid_ttl_to_controller >= 0) {
3290 ofconn_set_invalid_ttl_to_controller(ofconn,
3291 config.invalid_ttl_to_controller);
3292 }
3293
3294 ofconn_set_miss_send_len(ofconn, config.miss_send_len);
3295
3296 return 0;
3297 }
3298
3299 /* Checks whether 'ofconn' is a slave controller. If so, returns an OpenFlow
3300 * error message code for the caller to propagate upward. Otherwise, returns
3301 * 0.
3302 *
3303 * The log message mentions 'msg_type'. */
3304 static enum ofperr
3305 reject_slave_controller(struct ofconn *ofconn)
3306 {
3307 if (ofconn_get_type(ofconn) == OFCONN_PRIMARY
3308 && ofconn_get_role(ofconn) == OFPCR12_ROLE_SLAVE) {
3309 return OFPERR_OFPBRC_IS_SLAVE;
3310 } else {
3311 return 0;
3312 }
3313 }
3314
3315 /* Checks that the 'ofpacts_len' bytes of action in 'ofpacts' are appropriate
3316 * for 'ofproto':
3317 *
3318 * - If they use a meter, then 'ofproto' has that meter configured.
3319 *
3320 * - If they use any groups, then 'ofproto' has that group configured.
3321 *
3322 * Returns 0 if successful, otherwise an OpenFlow error. */
3323 enum ofperr
3324 ofproto_check_ofpacts(struct ofproto *ofproto,
3325 const struct ofpact ofpacts[], size_t ofpacts_len)
3326 {
3327 const struct ofpact *a;
3328 uint32_t mid;
3329
3330 mid = ofpacts_get_meter(ofpacts, ofpacts_len);
3331 if (mid && get_provider_meter_id(ofproto, mid) == UINT32_MAX) {
3332 return OFPERR_OFPMMFC_INVALID_METER;
3333 }
3334
3335 OFPACT_FOR_EACH_FLATTENED (a, ofpacts, ofpacts_len) {
3336 if (a->type == OFPACT_GROUP
3337 && !ofproto_group_exists(ofproto, ofpact_get_GROUP(a)->group_id)) {
3338 return OFPERR_OFPBAC_BAD_OUT_GROUP;
3339 }
3340 }
3341
3342 return 0;
3343 }
3344
3345 static enum ofperr
3346 handle_packet_out(struct ofconn *ofconn, const struct ofp_header *oh)
3347 {
3348 struct ofproto *p = ofconn_get_ofproto(ofconn);
3349 struct ofputil_packet_out po;
3350 struct dp_packet *payload;
3351 uint64_t ofpacts_stub[1024 / 8];
3352 struct ofpbuf ofpacts;
3353 struct flow flow;
3354 enum ofperr error;
3355
3356 COVERAGE_INC(ofproto_packet_out);
3357
3358 error = reject_slave_controller(ofconn);
3359 if (error) {
3360 goto exit;
3361 }
3362
3363 /* Decode message. */
3364 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
3365 error = ofputil_decode_packet_out(&po, oh, &ofpacts);
3366 if (error) {
3367 goto exit_free_ofpacts;
3368 }
3369 if (ofp_to_u16(po.in_port) >= p->max_ports
3370 && ofp_to_u16(po.in_port) < ofp_to_u16(OFPP_MAX)) {
3371 error = OFPERR_OFPBRC_BAD_PORT;
3372 goto exit_free_ofpacts;
3373 }
3374
3375 /* Get payload. */
3376 if (po.buffer_id != UINT32_MAX) {
3377 error = ofconn_pktbuf_retrieve(ofconn, po.buffer_id, &payload, NULL);
3378 if (error) {
3379 goto exit_free_ofpacts;
3380 }
3381 } else {
3382 /* Ensure that the L3 header is 32-bit aligned. */
3383 payload = dp_packet_clone_data_with_headroom(po.packet, po.packet_len, 2);
3384 }
3385
3386 /* Verify actions against packet, then send packet if successful. */
3387 flow_extract(payload, &flow);
3388 flow.in_port.ofp_port = po.in_port;
3389
3390 /* Check actions like for flow mods. We pass a 'table_id' of 0 to
3391 * ofproto_check_consistency(), which isn't strictly correct because these
3392 * actions aren't in any table. This is OK as 'table_id' is only used to
3393 * check instructions (e.g., goto-table), which can't appear on the action
3394 * list of a packet-out. */
3395 error = ofpacts_check_consistency(po.ofpacts, po.ofpacts_len,
3396 &flow, u16_to_ofp(p->max_ports),
3397 0, p->n_tables,
3398 ofconn_get_protocol(ofconn));
3399 if (!error) {
3400 error = ofproto_check_ofpacts(p, po.ofpacts, po.ofpacts_len);
3401 if (!error) {
3402 error = p->ofproto_class->packet_out(p, payload, &flow,
3403 po.ofpacts, po.ofpacts_len);
3404 }
3405 }
3406 dp_packet_delete(payload);
3407
3408 exit_free_ofpacts:
3409 ofpbuf_uninit(&ofpacts);
3410 exit:
3411 return error;
3412 }
3413
3414 static enum ofperr
3415 handle_nxt_resume(struct ofconn *ofconn, const struct ofp_header *oh)
3416 {
3417 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3418 struct ofputil_packet_in_private pin;
3419 enum ofperr error;
3420
3421 error = ofputil_decode_packet_in_private(oh, false, &pin, NULL, NULL);
3422 if (error) {
3423 return error;
3424 }
3425
3426 error = (ofproto->ofproto_class->nxt_resume
3427 ? ofproto->ofproto_class->nxt_resume(ofproto, &pin)
3428 : OFPERR_NXR_NOT_SUPPORTED);
3429
3430 ofputil_packet_in_private_destroy(&pin);
3431
3432 return error;
3433 }
3434
3435 static void
3436 update_port_config(struct ofconn *ofconn, struct ofport *port,
3437 enum ofputil_port_config config,
3438 enum ofputil_port_config mask)
3439 {
3440 enum ofputil_port_config toggle = (config ^ port->pp.config) & mask;
3441
3442 if (toggle & OFPUTIL_PC_PORT_DOWN
3443 && (config & OFPUTIL_PC_PORT_DOWN
3444 ? netdev_turn_flags_off(port->netdev, NETDEV_UP, NULL)
3445 : netdev_turn_flags_on(port->netdev, NETDEV_UP, NULL))) {
3446 /* We tried to bring the port up or down, but it failed, so don't
3447 * update the "down" bit. */
3448 toggle &= ~OFPUTIL_PC_PORT_DOWN;
3449 }
3450
3451 if (toggle) {
3452 enum ofputil_port_config old_config = port->pp.config;
3453 port->pp.config ^= toggle;
3454 port->ofproto->ofproto_class->port_reconfigured(port, old_config);
3455 connmgr_send_port_status(port->ofproto->connmgr, ofconn, &port->pp,
3456 OFPPR_MODIFY);
3457 }
3458 }
3459
3460 static enum ofperr
3461 port_mod_start(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3462 struct ofport **port)
3463 {
3464 struct ofproto *p = ofconn_get_ofproto(ofconn);
3465
3466 *port = ofproto_get_port(p, pm->port_no);
3467 if (!*port) {
3468 return OFPERR_OFPPMFC_BAD_PORT;
3469 }
3470 if (!eth_addr_equals((*port)->pp.hw_addr, pm->hw_addr)) {
3471 return OFPERR_OFPPMFC_BAD_HW_ADDR;
3472 }
3473 return 0;
3474 }
3475
3476 static void
3477 port_mod_finish(struct ofconn *ofconn, struct ofputil_port_mod *pm,
3478 struct ofport *port)
3479 {
3480 update_port_config(ofconn, port, pm->config, pm->mask);
3481 if (pm->advertise) {
3482 netdev_set_advertisements(port->netdev, pm->advertise);
3483 }
3484 }
3485
3486 static enum ofperr
3487 handle_port_mod(struct ofconn *ofconn, const struct ofp_header *oh)
3488 {
3489 struct ofputil_port_mod pm;
3490 struct ofport *port;
3491 enum ofperr error;
3492
3493 error = reject_slave_controller(ofconn);
3494 if (error) {
3495 return error;
3496 }
3497
3498 error = ofputil_decode_port_mod(oh, &pm, false);
3499 if (error) {
3500 return error;
3501 }
3502
3503 error = port_mod_start(ofconn, &pm, &port);
3504 if (!error) {
3505 port_mod_finish(ofconn, &pm, port);
3506 }
3507 return error;
3508 }
3509
3510 static enum ofperr
3511 handle_desc_stats_request(struct ofconn *ofconn,
3512 const struct ofp_header *request)
3513 {
3514 static const char *default_mfr_desc = "Nicira, Inc.";
3515 static const char *default_hw_desc = "Open vSwitch";
3516 static const char *default_sw_desc = VERSION;
3517 static const char *default_serial_desc = "None";
3518 static const char *default_dp_desc = "None";
3519
3520 struct ofproto *p = ofconn_get_ofproto(ofconn);
3521 struct ofp_desc_stats *ods;
3522 struct ofpbuf *msg;
3523
3524 msg = ofpraw_alloc_stats_reply(request, 0);
3525 ods = ofpbuf_put_zeros(msg, sizeof *ods);
3526 ovs_strlcpy(ods->mfr_desc, p->mfr_desc ? p->mfr_desc : default_mfr_desc,
3527 sizeof ods->mfr_desc);
3528 ovs_strlcpy(ods->hw_desc, p->hw_desc ? p->hw_desc : default_hw_desc,
3529 sizeof ods->hw_desc);
3530 ovs_strlcpy(ods->sw_desc, p->sw_desc ? p->sw_desc : default_sw_desc,
3531 sizeof ods->sw_desc);
3532 ovs_strlcpy(ods->serial_num,
3533 p->serial_desc ? p->serial_desc : default_serial_desc,
3534 sizeof ods->serial_num);
3535 ovs_strlcpy(ods->dp_desc, p->dp_desc ? p->dp_desc : default_dp_desc,
3536 sizeof ods->dp_desc);
3537 ofconn_send_reply(ofconn, msg);
3538
3539 return 0;
3540 }
3541
3542 static enum ofperr
3543 handle_table_stats_request(struct ofconn *ofconn,
3544 const struct ofp_header *request)
3545 {
3546 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3547 struct ofputil_table_features *features;
3548 struct ofputil_table_stats *stats;
3549 struct ofpbuf *reply;
3550 size_t i;
3551
3552 query_tables(ofproto, &features, &stats);
3553
3554 reply = ofputil_encode_table_stats_reply(request);
3555 for (i = 0; i < ofproto->n_tables; i++) {
3556 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3557 ofputil_append_table_stats_reply(reply, &stats[i], &features[i]);
3558 }
3559 }
3560 ofconn_send_reply(ofconn, reply);
3561
3562 free(features);
3563 free(stats);
3564
3565 return 0;
3566 }
3567
3568 static enum ofperr
3569 handle_table_features_request(struct ofconn *ofconn,
3570 const struct ofp_header *request)
3571 {
3572 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3573 struct ofpbuf msg = ofpbuf_const_initializer(request,
3574 ntohs(request->length));
3575 ofpraw_pull_assert(&msg);
3576 if (msg.size || ofpmp_more(request)) {
3577 return OFPERR_OFPTFFC_EPERM;
3578 }
3579
3580 struct ofputil_table_features *features;
3581 query_tables(ofproto, &features, NULL);
3582
3583 struct ovs_list replies;
3584 ofpmp_init(&replies, request);
3585 for (size_t i = 0; i < ofproto->n_tables; i++) {
3586 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3587 ofputil_append_table_features_reply(&features[i], &replies);
3588 }
3589 }
3590 ofconn_send_replies(ofconn, &replies);
3591
3592 free(features);
3593
3594 return 0;
3595 }
3596
3597 /* Returns the vacancy of 'oftable', a number that ranges from 0 (if the table
3598 * is full) to 100 (if the table is empty).
3599 *
3600 * A table without a limit on flows is considered to be empty. */
3601 static uint8_t
3602 oftable_vacancy(const struct oftable *t)
3603 {
3604 return (!t->max_flows ? 100
3605 : t->n_flows >= t->max_flows ? 0
3606 : (t->max_flows - t->n_flows) * 100.0 / t->max_flows);
3607 }
3608
3609 static void
3610 query_table_desc__(struct ofputil_table_desc *td,
3611 struct ofproto *ofproto, uint8_t table_id)
3612 {
3613 const struct oftable *t = &ofproto->tables[table_id];
3614
3615 td->table_id = table_id;
3616 td->eviction = (t->eviction & EVICTION_OPENFLOW
3617 ? OFPUTIL_TABLE_EVICTION_ON
3618 : OFPUTIL_TABLE_EVICTION_OFF);
3619 td->eviction_flags = OFPROTO_EVICTION_FLAGS;
3620 td->vacancy = (t->vacancy_event
3621 ? OFPUTIL_TABLE_VACANCY_ON
3622 : OFPUTIL_TABLE_VACANCY_OFF);
3623 td->table_vacancy.vacancy_down = t->vacancy_down;
3624 td->table_vacancy.vacancy_up = t->vacancy_up;
3625 td->table_vacancy.vacancy = oftable_vacancy(t);
3626 }
3627
3628 /* This function queries the database for dumping table-desc. */
3629 static void
3630 query_tables_desc(struct ofproto *ofproto, struct ofputil_table_desc **descp)
3631 {
3632 struct ofputil_table_desc *table_desc;
3633 size_t i;
3634
3635 table_desc = *descp = xcalloc(ofproto->n_tables, sizeof *table_desc);
3636 for (i = 0; i < ofproto->n_tables; i++) {
3637 struct ofputil_table_desc *td = &table_desc[i];
3638 query_table_desc__(td, ofproto, i);
3639 }
3640 }
3641
3642 /* Function to handle dump-table-desc request. */
3643 static enum ofperr
3644 handle_table_desc_request(struct ofconn *ofconn,
3645 const struct ofp_header *request)
3646 {
3647 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3648 struct ofputil_table_desc *table_desc;
3649 struct ovs_list replies;
3650 size_t i;
3651
3652 query_tables_desc(ofproto, &table_desc);
3653 ofpmp_init(&replies, request);
3654 for (i = 0; i < ofproto->n_tables; i++) {
3655 if (!(ofproto->tables[i].flags & OFTABLE_HIDDEN)) {
3656 ofputil_append_table_desc_reply(&table_desc[i], &replies,
3657 request->version);
3658 }
3659 }
3660 ofconn_send_replies(ofconn, &replies);
3661 free(table_desc);
3662 return 0;
3663 }
3664
3665 /* This function determines and sends the vacancy event, based on the value
3666 * of current vacancy and threshold vacancy. If the current vacancy is less
3667 * than or equal to vacancy_down, vacancy up events must be enabled, and when
3668 * the current vacancy is greater or equal to vacancy_up, vacancy down events
3669 * must be enabled. */
3670 static void
3671 send_table_status(struct ofproto *ofproto, uint8_t table_id)
3672 {
3673 struct oftable *t = &ofproto->tables[table_id];
3674 if (!t->vacancy_event) {
3675 return;
3676 }
3677
3678 uint8_t vacancy = oftable_vacancy(t);
3679 enum ofp14_table_reason event;
3680 if (vacancy < t->vacancy_down) {
3681 event = OFPTR_VACANCY_DOWN;
3682 } else if (vacancy > t->vacancy_up) {
3683 event = OFPTR_VACANCY_UP;
3684 } else {
3685 return;
3686 }
3687
3688 if (event == t->vacancy_event) {
3689 struct ofputil_table_desc td;
3690 query_table_desc__(&td, ofproto, table_id);
3691 connmgr_send_table_status(ofproto->connmgr, &td, event);
3692
3693 t->vacancy_event = (event == OFPTR_VACANCY_DOWN
3694 ? OFPTR_VACANCY_UP
3695 : OFPTR_VACANCY_DOWN);
3696 }
3697 }
3698
3699 static void
3700 append_port_stat(struct ofport *port, struct ovs_list *replies)
3701 {
3702 struct ofputil_port_stats ops = { .port_no = port->pp.port_no };
3703
3704 calc_duration(port->created, time_msec(),
3705 &ops.duration_sec, &ops.duration_nsec);
3706
3707 /* Intentionally ignore return value, since errors will set
3708 * 'stats' to all-1s, which is correct for OpenFlow, and
3709 * netdev_get_stats() will log errors. */
3710 ofproto_port_get_stats(port, &ops.stats);
3711
3712 ofputil_append_port_stat(replies, &ops);
3713 }
3714
3715 static void
3716 handle_port_request(struct ofconn *ofconn,
3717 const struct ofp_header *request, ofp_port_t port_no,
3718 void (*cb)(struct ofport *, struct ovs_list *replies))
3719 {
3720 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
3721 struct ofport *port;
3722 struct ovs_list replies;
3723
3724 ofpmp_init(&replies, request);
3725 if (port_no != OFPP_ANY) {
3726 port = ofproto_get_port(ofproto, port_no);
3727 if (port) {
3728 cb(port, &replies);
3729 }
3730 } else {
3731 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
3732 cb(port, &replies);
3733 }
3734 }
3735
3736 ofconn_send_replies(ofconn, &replies);
3737 }
3738
3739 static enum ofperr
3740 handle_port_stats_request(struct ofconn *ofconn,
3741 const struct ofp_header *request)
3742 {
3743 ofp_port_t port_no;
3744 enum ofperr error;
3745
3746 error = ofputil_decode_port_stats_request(request, &port_no);
3747 if (!error) {
3748 handle_port_request(ofconn, request, port_no, append_port_stat);
3749 }
3750 return error;
3751 }
3752
3753 static void
3754 append_port_desc(struct ofport *port, struct ovs_list *replies)
3755 {
3756 ofputil_append_port_desc_stats_reply(&port->pp, replies);
3757 }
3758
3759 static enum ofperr
3760 handle_port_desc_stats_request(struct ofconn *ofconn,
3761 const struct ofp_header *request)
3762 {
3763 ofp_port_t port_no;
3764 enum ofperr error;
3765
3766 error = ofputil_decode_port_desc_stats_request(request, &port_no);
3767 if (!error) {
3768 handle_port_request(ofconn, request, port_no, append_port_desc);
3769 }
3770 return error;
3771 }
3772
3773 static uint32_t
3774 hash_cookie(ovs_be64 cookie)
3775 {
3776 return hash_uint64((OVS_FORCE uint64_t)cookie);
3777 }
3778
3779 static void
3780 cookies_insert(struct ofproto *ofproto, struct rule *rule)
3781 OVS_REQUIRES(ofproto_mutex)
3782 {
3783 hindex_insert(&ofproto->cookies, &rule->cookie_node,
3784 hash_cookie(rule->flow_cookie));
3785 }
3786
3787 static void
3788 cookies_remove(struct ofproto *ofproto, struct rule *rule)
3789 OVS_REQUIRES(ofproto_mutex)
3790 {
3791 hindex_remove(&ofproto->cookies, &rule->cookie_node);
3792 }
3793
3794 static void
3795 calc_duration(long long int start, long long int now,
3796 uint32_t *sec, uint32_t *nsec)
3797 {
3798 long long int msecs = now - start;
3799 *sec = msecs / 1000;
3800 *nsec = (msecs % 1000) * (1000 * 1000);
3801 }
3802
3803 /* Checks whether 'table_id' is 0xff or a valid table ID in 'ofproto'. Returns
3804 * true if 'table_id' is OK, false otherwise. */
3805 static bool
3806 check_table_id(const struct ofproto *ofproto, uint8_t table_id)
3807 {
3808 return table_id == OFPTT_ALL || table_id < ofproto->n_tables;
3809 }
3810
3811 static struct oftable *
3812 next_visible_table(const struct ofproto *ofproto, uint8_t table_id)
3813 {
3814 struct oftable *table;
3815
3816 for (table = &ofproto->tables[table_id];
3817 table < &ofproto->tables[ofproto->n_tables];
3818 table++) {
3819 if (!(table->flags & OFTABLE_HIDDEN)) {
3820 return table;
3821 }
3822 }
3823
3824 return NULL;
3825 }
3826
3827 static struct oftable *
3828 first_matching_table(const struct ofproto *ofproto, uint8_t table_id)
3829 {
3830 if (table_id == 0xff) {
3831 return next_visible_table(ofproto, 0);
3832 } else if (table_id < ofproto->n_tables) {
3833 return &ofproto->tables[table_id];
3834 } else {
3835 return NULL;
3836 }
3837 }
3838
3839 static struct oftable *
3840 next_matching_table(const struct ofproto *ofproto,
3841 const struct oftable *table, uint8_t table_id)
3842 {
3843 return (table_id == 0xff
3844 ? next_visible_table(ofproto, (table - ofproto->tables) + 1)
3845 : NULL);
3846 }
3847
3848 /* Assigns TABLE to each oftable, in turn, that matches TABLE_ID in OFPROTO:
3849 *
3850 * - If TABLE_ID is 0xff, this iterates over every classifier table in
3851 * OFPROTO, skipping tables marked OFTABLE_HIDDEN.
3852 *
3853 * - If TABLE_ID is the number of a table in OFPROTO, then the loop iterates
3854 * only once, for that table. (This can be used to access tables marked
3855 * OFTABLE_HIDDEN.)
3856 *
3857 * - Otherwise, TABLE_ID isn't valid for OFPROTO, so the loop won't be
3858 * entered at all. (Perhaps you should have validated TABLE_ID with
3859 * check_table_id().)
3860 *
3861 * All parameters are evaluated multiple times.
3862 */
3863 #define FOR_EACH_MATCHING_TABLE(TABLE, TABLE_ID, OFPROTO) \
3864 for ((TABLE) = first_matching_table(OFPROTO, TABLE_ID); \
3865 (TABLE) != NULL; \
3866 (TABLE) = next_matching_table(OFPROTO, TABLE, TABLE_ID))
3867
3868 /* Initializes 'criteria' in a straightforward way based on the other
3869 * parameters.
3870 *
3871 * By default, the criteria include flows that are read-only, on the assumption
3872 * that the collected flows won't be modified. Call rule_criteria_require_rw()
3873 * if flows will be modified.
3874 *
3875 * For "loose" matching, the 'priority' parameter is unimportant and may be
3876 * supplied as 0. */
3877 static void
3878 rule_criteria_init(struct rule_criteria *criteria, uint8_t table_id,
3879 const struct match *match, int priority,
3880 cls_version_t version, ovs_be64 cookie,
3881 ovs_be64 cookie_mask, ofp_port_t out_port,
3882 uint32_t out_group)
3883 {
3884 criteria->table_id = table_id;
3885 cls_rule_init(&criteria->cr, match, priority);
3886 criteria->version = version;
3887 criteria->cookie = cookie;
3888 criteria->cookie_mask = cookie_mask;
3889 criteria->out_port = out_port;
3890 criteria->out_group = out_group;
3891
3892 /* We ordinarily want to skip hidden rules, but there has to be a way for
3893 * code internal to OVS to modify and delete them, so if the criteria
3894 * specify a priority that can only be for a hidden flow, then allow hidden
3895 * rules to be selected. (This doesn't allow OpenFlow clients to meddle
3896 * with hidden flows because OpenFlow uses only a 16-bit field to specify
3897 * priority.) */
3898 criteria->include_hidden = priority > UINT16_MAX;
3899
3900 /* We assume that the criteria are being used to collect flows for reading
3901 * but not modification. Thus, we should collect read-only flows. */
3902 criteria->include_readonly = true;
3903 }
3904
3905 /* By default, criteria initialized by rule_criteria_init() will match flows
3906 * that are read-only, on the assumption that the collected flows won't be
3907 * modified. Call this function to match only flows that are be modifiable.
3908 *
3909 * Specify 'can_write_readonly' as false in ordinary circumstances, true if the
3910 * caller has special privileges that allow it to modify even "read-only"
3911 * flows. */
3912 static void
3913 rule_criteria_require_rw(struct rule_criteria *criteria,
3914 bool can_write_readonly)
3915 {
3916 criteria->include_readonly = can_write_readonly;
3917 }
3918
3919 static void
3920 rule_criteria_destroy(struct rule_criteria *criteria)
3921 {
3922 cls_rule_destroy(&criteria->cr);
3923 }
3924
3925 void
3926 rule_collection_init(struct rule_collection *rules)
3927 {
3928 rules->rules = rules->stub;
3929 rules->n = 0;
3930 rules->capacity = ARRAY_SIZE(rules->stub);
3931 }
3932
3933 void
3934 rule_collection_add(struct rule_collection *rules, struct rule *rule)
3935 {
3936 if (rules->n >= rules->capacity) {
3937 size_t old_size, new_size;
3938
3939 old_size = rules->capacity * sizeof *rules->rules;
3940 rules->capacity *= 2;
3941 new_size = rules->capacity * sizeof *rules->rules;
3942
3943 if (rules->rules == rules->stub) {
3944 rules->rules = xmalloc(new_size);
3945 memcpy(rules->rules, rules->stub, old_size);
3946 } else {
3947 rules->rules = xrealloc(rules->rules, new_size);
3948 }
3949 }
3950
3951 rules->rules[rules->n++] = rule;
3952 }
3953
3954 void
3955 rule_collection_ref(struct rule_collection *rules)
3956 OVS_REQUIRES(ofproto_mutex)
3957 {
3958 size_t i;
3959
3960 for (i = 0; i < rules->n; i++) {
3961 ofproto_rule_ref(rules->rules[i]);
3962 }
3963 }
3964
3965 void
3966 rule_collection_unref(struct rule_collection *rules)
3967 {
3968 size_t i;
3969
3970 for (i = 0; i < rules->n; i++) {
3971 ofproto_rule_unref(rules->rules[i]);
3972 }
3973 }
3974
3975 /* Returns a NULL-terminated array of rule pointers,
3976 * destroys 'rules'. */
3977 static struct rule **
3978 rule_collection_detach(struct rule_collection *rules)
3979 {
3980 struct rule **rule_array;
3981
3982 rule_collection_add(rules, NULL);
3983
3984 if (rules->rules == rules->stub) {
3985 rules->rules = xmemdup(rules->rules, rules->n * sizeof *rules->rules);
3986 }
3987
3988 rule_array = rules->rules;
3989 rule_collection_init(rules);
3990
3991 return rule_array;
3992 }
3993
3994 void
3995 rule_collection_destroy(struct rule_collection *rules)
3996 {
3997 if (rules->rules != rules->stub) {
3998 free(rules->rules);
3999 }
4000
4001 /* Make repeated destruction harmless. */
4002 rule_collection_init(rules);
4003 }
4004
4005 /* Schedules postponed removal of rules, destroys 'rules'. */
4006 static void
4007 rule_collection_remove_postponed(struct rule_collection *rules)
4008 OVS_REQUIRES(ofproto_mutex)
4009 {
4010 if (rules->n > 0) {
4011 if (rules->n == 1) {
4012 ovsrcu_postpone(remove_rule_rcu, rules->rules[0]);
4013 } else {
4014 ovsrcu_postpone(remove_rules_rcu, rule_collection_detach(rules));
4015 }
4016 }
4017 }
4018
4019 /* Checks whether 'rule' matches 'c' and, if so, adds it to 'rules'. This
4020 * function verifies most of the criteria in 'c' itself, but the caller must
4021 * check 'c->cr' itself.
4022 *
4023 * Rules that have already been marked for removal are not collected.
4024 *
4025 * Increments '*n_readonly' if 'rule' wasn't added because it's read-only (and
4026 * 'c' only includes modifiable rules). */
4027 static void
4028 collect_rule(struct rule *rule, const struct rule_criteria *c,
4029 struct rule_collection *rules, size_t *n_readonly)
4030 OVS_REQUIRES(ofproto_mutex)
4031 {
4032 if ((c->table_id == rule->table_id || c->table_id == 0xff)
4033 && ofproto_rule_has_out_port(rule, c->out_port)
4034 && ofproto_rule_has_out_group(rule, c->out_group)
4035 && !((rule->flow_cookie ^ c->cookie) & c->cookie_mask)
4036 && (!rule_is_hidden(rule) || c->include_hidden)
4037 && cls_rule_visible_in_version(&rule->cr, c->version)) {
4038 /* Rule matches all the criteria... */
4039 if (!rule_is_readonly(rule) || c->include_readonly) {
4040 /* ...add it. */
4041 rule_collection_add(rules, rule);
4042 } else {
4043 /* ...except it's read-only. */
4044 ++*n_readonly;
4045 }
4046 }
4047 }
4048
4049 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4050 * on classifiers rules are done in the "loose" way required for OpenFlow
4051 * OFPFC_MODIFY and OFPFC_DELETE requests. Puts the selected rules on list
4052 * 'rules'.
4053 *
4054 * Returns 0 on success, otherwise an OpenFlow error code. */
4055 static enum ofperr
4056 collect_rules_loose(struct ofproto *ofproto,
4057 const struct rule_criteria *criteria,
4058 struct rule_collection *rules)
4059 OVS_REQUIRES(ofproto_mutex)
4060 {
4061 struct oftable *table;
4062 enum ofperr error = 0;
4063 size_t n_readonly = 0;
4064
4065 rule_collection_init(rules);
4066
4067 if (!check_table_id(ofproto, criteria->table_id)) {
4068 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4069 goto exit;
4070 }
4071
4072 if (criteria->cookie_mask == OVS_BE64_MAX) {
4073 struct rule *rule;
4074
4075 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4076 hash_cookie(criteria->cookie),
4077 &ofproto->cookies) {
4078 if (cls_rule_is_loose_match(&rule->cr, &criteria->cr.match)) {
4079 collect_rule(rule, criteria, rules, &n_readonly);
4080 }
4081 }
4082 } else {
4083 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4084 struct rule *rule;
4085
4086 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &criteria->cr,
4087 criteria->version) {
4088 collect_rule(rule, criteria, rules, &n_readonly);
4089 }
4090 }
4091 }
4092
4093 exit:
4094 if (!error && !rules->n && n_readonly) {
4095 /* We didn't find any rules to modify. We did find some read-only
4096 * rules that we're not allowed to modify, so report that. */
4097 error = OFPERR_OFPBRC_EPERM;
4098 }
4099 if (error) {
4100 rule_collection_destroy(rules);
4101 }
4102 return error;
4103 }
4104
4105 /* Searches 'ofproto' for rules that match the criteria in 'criteria'. Matches
4106 * on classifiers rules are done in the "strict" way required for OpenFlow
4107 * OFPFC_MODIFY_STRICT and OFPFC_DELETE_STRICT requests. Puts the selected
4108 * rules on list 'rules'.
4109 *
4110 * Returns 0 on success, otherwise an OpenFlow error code. */
4111 static enum ofperr
4112 collect_rules_strict(struct ofproto *ofproto,
4113 const struct rule_criteria *criteria,
4114 struct rule_collection *rules)
4115 OVS_REQUIRES(ofproto_mutex)
4116 {
4117 struct oftable *table;
4118 size_t n_readonly = 0;
4119 enum ofperr error = 0;
4120
4121 rule_collection_init(rules);
4122
4123 if (!check_table_id(ofproto, criteria->table_id)) {
4124 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4125 goto exit;
4126 }
4127
4128 if (criteria->cookie_mask == OVS_BE64_MAX) {
4129 struct rule *rule;
4130
4131 HINDEX_FOR_EACH_WITH_HASH (rule, cookie_node,
4132 hash_cookie(criteria->cookie),
4133 &ofproto->cookies) {
4134 if (cls_rule_equal(&rule->cr, &criteria->cr)) {
4135 collect_rule(rule, criteria, rules, &n_readonly);
4136 }
4137 }
4138 } else {
4139 FOR_EACH_MATCHING_TABLE (table, criteria->table_id, ofproto) {
4140 struct rule *rule;
4141
4142 rule = rule_from_cls_rule(classifier_find_rule_exactly(
4143 &table->cls, &criteria->cr,
4144 criteria->version));
4145 if (rule) {
4146 collect_rule(rule, criteria, rules, &n_readonly);
4147 }
4148 }
4149 }
4150
4151 exit:
4152 if (!error && !rules->n && n_readonly) {
4153 /* We didn't find any rules to modify. We did find some read-only
4154 * rules that we're not allowed to modify, so report that. */
4155 error = OFPERR_OFPBRC_EPERM;
4156 }
4157 if (error) {
4158 rule_collection_destroy(rules);
4159 }
4160 return error;
4161 }
4162
4163 /* Returns 'age_ms' (a duration in milliseconds), converted to seconds and
4164 * forced into the range of a uint16_t. */
4165 static int
4166 age_secs(long long int age_ms)
4167 {
4168 return (age_ms < 0 ? 0
4169 : age_ms >= UINT16_MAX * 1000 ? UINT16_MAX
4170 : (unsigned int) age_ms / 1000);
4171 }
4172
4173 static enum ofperr
4174 handle_flow_stats_request(struct ofconn *ofconn,
4175 const struct ofp_header *request)
4176 OVS_EXCLUDED(ofproto_mutex)
4177 {
4178 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4179 struct ofputil_flow_stats_request fsr;
4180 struct rule_criteria criteria;
4181 struct rule_collection rules;
4182 struct ovs_list replies;
4183 enum ofperr error;
4184 size_t i;
4185
4186 error = ofputil_decode_flow_stats_request(&fsr, request);
4187 if (error) {
4188 return error;
4189 }
4190
4191 rule_criteria_init(&criteria, fsr.table_id, &fsr.match, 0, CLS_MAX_VERSION,
4192 fsr.cookie, fsr.cookie_mask, fsr.out_port,
4193 fsr.out_group);
4194
4195 ovs_mutex_lock(&ofproto_mutex);
4196 error = collect_rules_loose(ofproto, &criteria, &rules);
4197 rule_criteria_destroy(&criteria);
4198 if (!error) {
4199 rule_collection_ref(&rules);
4200 }
4201 ovs_mutex_unlock(&ofproto_mutex);
4202
4203 if (error) {
4204 return error;
4205 }
4206
4207 ofpmp_init(&replies, request);
4208 for (i = 0; i < rules.n; i++) {
4209 struct rule *rule = rules.rules[i];
4210 long long int now = time_msec();
4211 struct ofputil_flow_stats fs;
4212 long long int created, used, modified;
4213 const struct rule_actions *actions;
4214 enum ofputil_flow_mod_flags flags;
4215
4216 ovs_mutex_lock(&rule->mutex);
4217 fs.cookie = rule->flow_cookie;
4218 fs.idle_timeout = rule->idle_timeout;
4219 fs.hard_timeout = rule->hard_timeout;
4220 fs.importance = rule->importance;
4221 created = rule->created;
4222 modified = rule->modified;
4223 actions = rule_get_actions(rule);
4224 flags = rule->flags;
4225 ovs_mutex_unlock(&rule->mutex);
4226
4227 ofproto->ofproto_class->rule_get_stats(rule, &fs.packet_count,
4228 &fs.byte_count, &used);
4229
4230 minimatch_expand(&rule->cr.match, &fs.match);
4231 fs.table_id = rule->table_id;
4232 calc_duration(created, now, &fs.duration_sec, &fs.duration_nsec);
4233 fs.priority = rule->cr.priority;
4234 fs.idle_age = age_secs(now - used);
4235 fs.hard_age = age_secs(now - modified);
4236 fs.ofpacts = actions->ofpacts;
4237 fs.ofpacts_len = actions->ofpacts_len;
4238
4239 fs.flags = flags;
4240 ofputil_append_flow_stats_reply(&fs, &replies);
4241 }
4242
4243 rule_collection_unref(&rules);
4244 rule_collection_destroy(&rules);
4245
4246 ofconn_send_replies(ofconn, &replies);
4247
4248 return 0;
4249 }
4250
4251 static void
4252 flow_stats_ds(struct rule *rule, struct ds *results)
4253 {
4254 uint64_t packet_count, byte_count;
4255 const struct rule_actions *actions;
4256 long long int created, used;
4257
4258 rule->ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4259 &byte_count, &used);
4260
4261 ovs_mutex_lock(&rule->mutex);
4262 actions = rule_get_actions(rule);
4263 created = rule->created;
4264 ovs_mutex_unlock(&rule->mutex);
4265
4266 if (rule->table_id != 0) {
4267 ds_put_format(results, "table_id=%"PRIu8", ", rule->table_id);
4268 }
4269 ds_put_format(results, "duration=%llds, ", (time_msec() - created) / 1000);
4270 ds_put_format(results, "n_packets=%"PRIu64", ", packet_count);
4271 ds_put_format(results, "n_bytes=%"PRIu64", ", byte_count);
4272 cls_rule_format(&rule->cr, results);
4273 ds_put_char(results, ',');
4274
4275 ds_put_cstr(results, "actions=");
4276 ofpacts_format(actions->ofpacts, actions->ofpacts_len, results);
4277
4278 ds_put_cstr(results, "\n");
4279 }
4280
4281 /* Adds a pretty-printed description of all flows to 'results', including
4282 * hidden flows (e.g., set up by in-band control). */
4283 void
4284 ofproto_get_all_flows(struct ofproto *p, struct ds *results)
4285 {
4286 struct oftable *table;
4287
4288 OFPROTO_FOR_EACH_TABLE (table, p) {
4289 struct rule *rule;
4290
4291 CLS_FOR_EACH (rule, cr, &table->cls) {
4292 flow_stats_ds(rule, results);
4293 }
4294 }
4295 }
4296
4297 /* Obtains the NetFlow engine type and engine ID for 'ofproto' into
4298 * '*engine_type' and '*engine_id', respectively. */
4299 void
4300 ofproto_get_netflow_ids(const struct ofproto *ofproto,
4301 uint8_t *engine_type, uint8_t *engine_id)
4302 {
4303 ofproto->ofproto_class->get_netflow_ids(ofproto, engine_type, engine_id);
4304 }
4305
4306 /* Checks the status change of CFM on 'ofport'.
4307 *
4308 * Returns true if 'ofproto_class' does not support 'cfm_status_changed'. */
4309 bool
4310 ofproto_port_cfm_status_changed(struct ofproto *ofproto, ofp_port_t ofp_port)
4311 {
4312 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4313 return (ofport && ofproto->ofproto_class->cfm_status_changed
4314 ? ofproto->ofproto_class->cfm_status_changed(ofport)
4315 : true);
4316 }
4317
4318 /* Checks the status of CFM configured on 'ofp_port' within 'ofproto'.
4319 * Returns 0 if the port's CFM status was successfully stored into
4320 * '*status'. Returns positive errno if the port did not have CFM
4321 * configured.
4322 *
4323 * The caller must provide and own '*status', and must free 'status->rmps'.
4324 * '*status' is indeterminate if the return value is non-zero. */
4325 int
4326 ofproto_port_get_cfm_status(const struct ofproto *ofproto, ofp_port_t ofp_port,
4327 struct cfm_status *status)
4328 {
4329 struct ofport *ofport = ofproto_get_port(ofproto, ofp_port);
4330 return (ofport && ofproto->ofproto_class->get_cfm_status
4331 ? ofproto->ofproto_class->get_cfm_status(ofport, status)
4332 : EOPNOTSUPP);
4333 }
4334
4335 static enum ofperr
4336 handle_aggregate_stats_request(struct ofconn *ofconn,
4337 const struct ofp_header *oh)
4338 OVS_EXCLUDED(ofproto_mutex)
4339 {
4340 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4341 struct ofputil_flow_stats_request request;
4342 struct ofputil_aggregate_stats stats;
4343 bool unknown_packets, unknown_bytes;
4344 struct rule_criteria criteria;
4345 struct rule_collection rules;
4346 struct ofpbuf *reply;
4347 enum ofperr error;
4348 size_t i;
4349
4350 error = ofputil_decode_flow_stats_request(&request, oh);
4351 if (error) {
4352 return error;
4353 }
4354
4355 rule_criteria_init(&criteria, request.table_id, &request.match, 0,
4356 CLS_MAX_VERSION, request.cookie, request.cookie_mask,
4357 request.out_port, request.out_group);
4358
4359 ovs_mutex_lock(&ofproto_mutex);
4360 error = collect_rules_loose(ofproto, &criteria, &rules);
4361 rule_criteria_destroy(&criteria);
4362 if (!error) {
4363 rule_collection_ref(&rules);
4364 }
4365 ovs_mutex_unlock(&ofproto_mutex);
4366
4367 if (error) {
4368 return error;
4369 }
4370
4371 memset(&stats, 0, sizeof stats);
4372 unknown_packets = unknown_bytes = false;
4373 for (i = 0; i < rules.n; i++) {
4374 struct rule *rule = rules.rules[i];
4375 uint64_t packet_count;
4376 uint64_t byte_count;
4377 long long int used;
4378
4379 ofproto->ofproto_class->rule_get_stats(rule, &packet_count,
4380 &byte_count, &used);
4381
4382 if (packet_count == UINT64_MAX) {
4383 unknown_packets = true;
4384 } else {
4385 stats.packet_count += packet_count;
4386 }
4387
4388 if (byte_count == UINT64_MAX) {
4389 unknown_bytes = true;
4390 } else {
4391 stats.byte_count += byte_count;
4392 }
4393
4394 stats.flow_count++;
4395 }
4396 if (unknown_packets) {
4397 stats.packet_count = UINT64_MAX;
4398 }
4399 if (unknown_bytes) {
4400 stats.byte_count = UINT64_MAX;
4401 }
4402
4403 rule_collection_unref(&rules);
4404 rule_collection_destroy(&rules);
4405
4406 reply = ofputil_encode_aggregate_stats_reply(&stats, oh);
4407 ofconn_send_reply(ofconn, reply);
4408
4409 return 0;
4410 }
4411
4412 struct queue_stats_cbdata {
4413 struct ofport *ofport;
4414 struct ovs_list replies;
4415 long long int now;
4416 };
4417
4418 static void
4419 put_queue_stats(struct queue_stats_cbdata *cbdata, uint32_t queue_id,
4420 const struct netdev_queue_stats *stats)
4421 {
4422 struct ofputil_queue_stats oqs;
4423
4424 oqs.port_no = cbdata->ofport->pp.port_no;
4425 oqs.queue_id = queue_id;
4426 oqs.tx_bytes = stats->tx_bytes;
4427 oqs.tx_packets = stats->tx_packets;
4428 oqs.tx_errors = stats->tx_errors;
4429 if (stats->created != LLONG_MIN) {
4430 calc_duration(stats->created, cbdata->now,
4431 &oqs.duration_sec, &oqs.duration_nsec);
4432 } else {
4433 oqs.duration_sec = oqs.duration_nsec = UINT32_MAX;
4434 }
4435 ofputil_append_queue_stat(&cbdata->replies, &oqs);
4436 }
4437
4438 static void
4439 handle_queue_stats_dump_cb(uint32_t queue_id,
4440 struct netdev_queue_stats *stats,
4441 void *cbdata_)
4442 {
4443 struct queue_stats_cbdata *cbdata = cbdata_;
4444
4445 put_queue_stats(cbdata, queue_id, stats);
4446 }
4447
4448 static enum ofperr
4449 handle_queue_stats_for_port(struct ofport *port, uint32_t queue_id,
4450 struct queue_stats_cbdata *cbdata)
4451 {
4452 cbdata->ofport = port;
4453 if (queue_id == OFPQ_ALL) {
4454 netdev_dump_queue_stats(port->netdev,
4455 handle_queue_stats_dump_cb, cbdata);
4456 } else {
4457 struct netdev_queue_stats stats;
4458
4459 if (!netdev_get_queue_stats(port->netdev, queue_id, &stats)) {
4460 put_queue_stats(cbdata, queue_id, &stats);
4461 } else {
4462 return OFPERR_OFPQOFC_BAD_QUEUE;
4463 }
4464 }
4465 return 0;
4466 }
4467
4468 static enum ofperr
4469 handle_queue_stats_request(struct ofconn *ofconn,
4470 const struct ofp_header *rq)
4471 {
4472 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
4473 struct queue_stats_cbdata cbdata;
4474 struct ofport *port;
4475 enum ofperr error;
4476 struct ofputil_queue_stats_request oqsr;
4477
4478 COVERAGE_INC(ofproto_queue_req);
4479
4480 ofpmp_init(&cbdata.replies, rq);
4481 cbdata.now = time_msec();
4482
4483 error = ofputil_decode_queue_stats_request(rq, &oqsr);
4484 if (error) {
4485 return error;
4486 }
4487
4488 if (oqsr.port_no == OFPP_ANY) {
4489 error = OFPERR_OFPQOFC_BAD_QUEUE;
4490 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
4491 if (!handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)) {
4492 error = 0;
4493 }
4494 }
4495 } else {
4496 port = ofproto_get_port(ofproto, oqsr.port_no);
4497 error = (port
4498 ? handle_queue_stats_for_port(port, oqsr.queue_id, &cbdata)
4499 : OFPERR_OFPQOFC_BAD_PORT);
4500 }
4501 if (!error) {
4502 ofconn_send_replies(ofconn, &cbdata.replies);
4503 } else {
4504 ofpbuf_list_delete(&cbdata.replies);
4505 }
4506
4507 return error;
4508 }
4509
4510 static enum ofperr
4511 evict_rules_from_table(struct oftable *table)
4512 OVS_REQUIRES(ofproto_mutex)
4513 {
4514 enum ofperr error = 0;
4515 struct rule_collection rules;
4516 unsigned int count = table->n_flows;
4517 unsigned int max_flows = table->max_flows;
4518
4519 rule_collection_init(&rules);
4520
4521 while (count-- > max_flows) {
4522 struct rule *rule;
4523
4524 if (!choose_rule_to_evict(table, &rule)) {
4525 error = OFPERR_OFPFMFC_TABLE_FULL;
4526 break;
4527 } else {
4528 eviction_group_remove_rule(rule);
4529 rule_collection_add(&rules, rule);
4530 }
4531 }
4532 delete_flows__(&rules, OFPRR_EVICTION, NULL);
4533
4534 return error;
4535 }
4536
4537 static void
4538 get_conjunctions(const struct ofputil_flow_mod *fm,
4539 struct cls_conjunction **conjsp, size_t *n_conjsp)
4540 OVS_REQUIRES(ofproto_mutex)
4541 {
4542 struct cls_conjunction *conjs = NULL;
4543 int n_conjs = 0;
4544
4545 const struct ofpact *ofpact;
4546 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4547 if (ofpact->type == OFPACT_CONJUNCTION) {
4548 n_conjs++;
4549 } else if (ofpact->type != OFPACT_NOTE) {
4550 /* "conjunction" may appear with "note" actions but not with any
4551 * other type of actions. */
4552 ovs_assert(!n_conjs);
4553 break;
4554 }
4555 }
4556 if (n_conjs) {
4557 int i = 0;
4558
4559 conjs = xzalloc(n_conjs * sizeof *conjs);
4560 OFPACT_FOR_EACH (ofpact, fm->ofpacts, fm->ofpacts_len) {
4561 if (ofpact->type == OFPACT_CONJUNCTION) {
4562 struct ofpact_conjunction *oc = ofpact_get_CONJUNCTION(ofpact);
4563 conjs[i].clause = oc->clause;
4564 conjs[i].n_clauses = oc->n_clauses;
4565 conjs[i].id = oc->id;
4566 i++;
4567 }
4568 }
4569 }
4570
4571 *conjsp = conjs;
4572 *n_conjsp = n_conjs;
4573 }
4574
4575 /* Implements OFPFC_ADD and the cases for OFPFC_MODIFY and OFPFC_MODIFY_STRICT
4576 * in which no matching flow already exists in the flow table.
4577 *
4578 * Adds the flow specified by 'fm', to the ofproto's flow table. Returns 0 on
4579 * success, or an OpenFlow error code on failure.
4580 *
4581 * On successful return the caller must complete the operation either by
4582 * calling add_flow_finish(), or add_flow_revert() if the operation needs to
4583 * be reverted.
4584 *
4585 * The caller retains ownership of 'fm->ofpacts'. */
4586 static enum ofperr
4587 add_flow_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4588 OVS_REQUIRES(ofproto_mutex)
4589 {
4590 struct ofputil_flow_mod *fm = &ofm->fm;
4591 struct rule **old_rule = &ofm->old_rules.stub[0];
4592 struct rule **new_rule = &ofm->new_rules.stub[0];
4593 struct oftable *table;
4594 struct cls_rule cr;
4595 struct rule *rule;
4596 uint8_t table_id;
4597 struct cls_conjunction *conjs;
4598 size_t n_conjs;
4599 enum ofperr error;
4600
4601 if (!check_table_id(ofproto, fm->table_id)) {
4602 error = OFPERR_OFPBRC_BAD_TABLE_ID;
4603 return error;
4604 }
4605
4606 /* Pick table. */
4607 if (fm->table_id == 0xff) {
4608 if (ofproto->ofproto_class->rule_choose_table) {
4609 error = ofproto->ofproto_class->rule_choose_table(ofproto,
4610 &fm->match,
4611 &table_id);
4612 if (error) {
4613 return error;
4614 }
4615 ovs_assert(table_id < ofproto->n_tables);
4616 } else {
4617 table_id = 0;
4618 }
4619 } else if (fm->table_id < ofproto->n_tables) {
4620 table_id = fm->table_id;
4621 } else {
4622 return OFPERR_OFPBRC_BAD_TABLE_ID;
4623 }
4624
4625 table = &ofproto->tables[table_id];
4626 if (table->flags & OFTABLE_READONLY
4627 && !(fm->flags & OFPUTIL_FF_NO_READONLY)) {
4628 return OFPERR_OFPBRC_EPERM;
4629 }
4630
4631 if (!(fm->flags & OFPUTIL_FF_HIDDEN_FIELDS)
4632 && !match_has_default_hidden_fields(&fm->match)) {
4633 VLOG_WARN_RL(&rl, "%s: (add_flow) only internal flows can set "
4634 "non-default values to hidden fields", ofproto->name);
4635 return OFPERR_OFPBRC_EPERM;
4636 }
4637
4638 cls_rule_init(&cr, &fm->match, fm->priority);
4639
4640 /* Check for the existence of an identical rule.
4641 * This will not return rules earlier marked for removal. */
4642 rule = rule_from_cls_rule(classifier_find_rule_exactly(&table->cls, &cr,
4643 ofm->version));
4644 *old_rule = rule;
4645 if (!rule) {
4646 /* Check for overlap, if requested. */
4647 if (fm->flags & OFPUTIL_FF_CHECK_OVERLAP
4648 && classifier_rule_overlaps(&table->cls, &cr, ofm->version)) {
4649 cls_rule_destroy(&cr);
4650 return OFPERR_OFPFMFC_OVERLAP;
4651 }
4652
4653 /* If necessary, evict an existing rule to clear out space. */
4654 if (table->n_flows >= table->max_flows) {
4655 if (!choose_rule_to_evict(table, &rule)) {
4656 error = OFPERR_OFPFMFC_TABLE_FULL;
4657 cls_rule_destroy(&cr);
4658 return error;
4659 }
4660 eviction_group_remove_rule(rule);
4661 /* Marks '*old_rule' as an evicted rule rather than replaced rule.
4662 */
4663 fm->delete_reason = OFPRR_EVICTION;
4664 *old_rule = rule;
4665 }
4666 } else {
4667 fm->modify_cookie = true;
4668 }
4669
4670 /* Allocate new rule. */
4671 error = replace_rule_create(ofproto, fm, &cr, table - ofproto->tables,
4672 rule, new_rule);
4673 if (error) {
4674 return error;
4675 }
4676
4677 get_conjunctions(fm, &conjs, &n_conjs);
4678 replace_rule_start(ofproto, ofm->version, rule, *new_rule, conjs, n_conjs);
4679 free(conjs);
4680
4681 return 0;
4682 }
4683
4684 /* Revert the effects of add_flow_start(). */
4685 static void
4686 add_flow_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4687 OVS_REQUIRES(ofproto_mutex)
4688 {
4689 struct ofputil_flow_mod *fm = &ofm->fm;
4690 struct rule *old_rule = ofm->old_rules.stub[0];
4691 struct rule *new_rule = ofm->new_rules.stub[0];
4692
4693 if (old_rule && fm->delete_reason == OFPRR_EVICTION) {
4694 /* Revert the eviction. */
4695 eviction_group_add_rule(old_rule);
4696 }
4697
4698 replace_rule_revert(ofproto, old_rule, new_rule);
4699 }
4700
4701 /* To be called after version bump. */
4702 static void
4703 add_flow_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
4704 const struct flow_mod_requester *req)
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 ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
4711
4712 replace_rule_finish(ofproto, fm, req, old_rule, new_rule, &dead_cookies);
4713 learned_cookies_flush(ofproto, &dead_cookies);
4714
4715 if (old_rule) {
4716 ovsrcu_postpone(remove_rule_rcu, old_rule);
4717 } else {
4718 if (minimask_get_vid_mask(new_rule->cr.match.mask) == VLAN_VID_MASK) {
4719 if (ofproto->vlan_bitmap) {
4720 uint16_t vid = miniflow_get_vid(new_rule->cr.match.flow);
4721
4722 if (!bitmap_is_set(ofproto->vlan_bitmap, vid)) {
4723 bitmap_set1(ofproto->vlan_bitmap, vid);
4724 ofproto->vlans_changed = true;
4725 }
4726 } else {
4727 ofproto->vlans_changed = true;
4728 }
4729 }
4730
4731 ofmonitor_report(ofproto->connmgr, new_rule, NXFME_ADDED, 0,
4732 req ? req->ofconn : NULL,
4733 req ? req->request->xid : 0, NULL);
4734
4735 /* Send Vacancy Events for OF1.4+. */
4736 send_table_status(ofproto, new_rule->table_id);
4737 }
4738
4739 send_buffered_packet(req, fm->buffer_id, new_rule);
4740 }
4741 \f
4742 /* OFPFC_MODIFY and OFPFC_MODIFY_STRICT. */
4743
4744 /* Create a new rule based on attributes in 'fm', match in 'cr', 'table_id',
4745 * and 'old_rule'. Note that the rule is NOT inserted into a any data
4746 * structures yet. Takes ownership of 'cr'. */
4747 static enum ofperr
4748 replace_rule_create(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4749 struct cls_rule *cr, uint8_t table_id,
4750 struct rule *old_rule, struct rule **new_rule)
4751 {
4752 struct rule *rule;
4753 enum ofperr error;
4754
4755 /* Allocate new rule. */
4756 rule = ofproto->ofproto_class->rule_alloc();
4757 if (!rule) {
4758 cls_rule_destroy(cr);
4759 VLOG_WARN_RL(&rl, "%s: failed to allocate a rule.", ofproto->name);
4760 return OFPERR_OFPFMFC_UNKNOWN;
4761 }
4762
4763 /* Initialize base state. */
4764 *CONST_CAST(struct ofproto **, &rule->ofproto) = ofproto;
4765 cls_rule_move(CONST_CAST(struct cls_rule *, &rule->cr), cr);
4766 ovs_refcount_init(&rule->ref_count);
4767 rule->flow_cookie = fm->new_cookie;
4768 rule->created = rule->modified = time_msec();
4769
4770 ovs_mutex_init(&rule->mutex);
4771 ovs_mutex_lock(&rule->mutex);
4772 rule->idle_timeout = fm->idle_timeout;
4773 rule->hard_timeout = fm->hard_timeout;
4774 *CONST_CAST(uint16_t *, &rule->importance) = fm->importance;
4775 rule->removed_reason = OVS_OFPRR_NONE;
4776
4777 *CONST_CAST(uint8_t *, &rule->table_id) = table_id;
4778 rule->flags = fm->flags & OFPUTIL_FF_STATE;
4779 *CONST_CAST(const struct rule_actions **, &rule->actions)
4780 = rule_actions_create(fm->ofpacts, fm->ofpacts_len);
4781 ovs_list_init(&rule->meter_list_node);
4782 rule->eviction_group = NULL;
4783 ovs_list_init(&rule->expirable);
4784 rule->monitor_flags = 0;
4785 rule->add_seqno = 0;
4786 rule->modify_seqno = 0;
4787
4788 /* Copy values from old rule for modify semantics. */
4789 if (old_rule && fm->delete_reason != OFPRR_EVICTION) {
4790 bool change_cookie = (fm->modify_cookie
4791 && fm->new_cookie != OVS_BE64_MAX
4792 && fm->new_cookie != old_rule->flow_cookie);
4793
4794 ovs_mutex_lock(&old_rule->mutex);
4795 if (fm->command != OFPFC_ADD) {
4796 rule->idle_timeout = old_rule->idle_timeout;
4797 rule->hard_timeout = old_rule->hard_timeout;
4798 *CONST_CAST(uint16_t *, &rule->importance) = old_rule->importance;
4799 rule->flags = old_rule->flags;
4800 rule->created = old_rule->created;
4801 }
4802 if (!change_cookie) {
4803 rule->flow_cookie = old_rule->flow_cookie;
4804 }
4805 ovs_mutex_unlock(&old_rule->mutex);
4806 }
4807 ovs_mutex_unlock(&rule->mutex);
4808
4809 /* Construct rule, initializing derived state. */
4810 error = ofproto->ofproto_class->rule_construct(rule);
4811 if (error) {
4812 ofproto_rule_destroy__(rule);
4813 return error;
4814 }
4815
4816 rule->removed = true; /* Not yet in ofproto data structures. */
4817
4818 *new_rule = rule;
4819 return 0;
4820 }
4821
4822 static void
4823 replace_rule_start(struct ofproto *ofproto, cls_version_t version,
4824 struct rule *old_rule, struct rule *new_rule,
4825 struct cls_conjunction *conjs, size_t n_conjs)
4826 {
4827 struct oftable *table = &ofproto->tables[new_rule->table_id];
4828
4829 /* 'old_rule' may be either an evicted rule or replaced rule. */
4830 if (old_rule) {
4831 /* Mark the old rule for removal in the next version. */
4832 cls_rule_make_invisible_in_version(&old_rule->cr, version);
4833 } else {
4834 table->n_flows++;
4835 }
4836 /* Insert flow to the classifier, so that later flow_mods may relate
4837 * to it. This is reversible, in case later errors require this to
4838 * be reverted. */
4839 ofproto_rule_insert__(ofproto, new_rule);
4840 /* Make the new rule visible for classifier lookups only from the next
4841 * version. */
4842 classifier_insert(&table->cls, &new_rule->cr, version, conjs, n_conjs);
4843 }
4844
4845 static void replace_rule_revert(struct ofproto *ofproto,
4846 struct rule *old_rule, struct rule *new_rule)
4847 {
4848 struct oftable *table = &ofproto->tables[new_rule->table_id];
4849
4850 if (old_rule) {
4851 /* Restore the original visibility of the old rule. */
4852 cls_rule_restore_visibility(&old_rule->cr);
4853 } else {
4854 /* Restore table's rule count. */
4855 table->n_flows--;
4856 }
4857
4858 /* Remove the new rule immediately. It was never visible to lookups. */
4859 if (!classifier_remove(&table->cls, &new_rule->cr)) {
4860 OVS_NOT_REACHED();
4861 }
4862 ofproto_rule_remove__(ofproto, new_rule);
4863 /* The rule was not inserted to the ofproto provider, so we can
4864 * release it without deleting it from the ofproto provider. */
4865 ofproto_rule_unref(new_rule);
4866 }
4867
4868 /* Adds the 'new_rule', replacing the 'old_rule'. */
4869 static void
4870 replace_rule_finish(struct ofproto *ofproto, struct ofputil_flow_mod *fm,
4871 const struct flow_mod_requester *req,
4872 struct rule *old_rule, struct rule *new_rule,
4873 struct ovs_list *dead_cookies)
4874 OVS_REQUIRES(ofproto_mutex)
4875 {
4876 bool forward_stats = !(fm->flags & OFPUTIL_FF_RESET_COUNTS);
4877 struct rule *replaced_rule;
4878
4879 replaced_rule = fm->delete_reason != OFPRR_EVICTION ? old_rule : NULL;
4880
4881 /* Insert the new flow to the ofproto provider. A non-NULL 'replaced_rule'
4882 * is a duplicate rule the 'new_rule' is replacing. The provider should
4883 * link the stats from the old rule to the new one if 'forward_stats' is
4884 * 'true'. The 'replaced_rule' will be deleted right after this call. */
4885 ofproto->ofproto_class->rule_insert(new_rule, replaced_rule,
4886 forward_stats);
4887 learned_cookies_inc(ofproto, rule_get_actions(new_rule));
4888
4889 if (old_rule) {
4890 const struct rule_actions *old_actions = rule_get_actions(old_rule);
4891
4892 /* Remove the old rule from data structures. Removal from the
4893 * classifier and the deletion of the rule is RCU postponed by the
4894 * caller. */
4895 ofproto_rule_remove__(ofproto, old_rule);
4896 learned_cookies_dec(ofproto, old_actions, dead_cookies);
4897
4898 if (replaced_rule) {
4899 enum nx_flow_update_event event = fm->command == OFPFC_ADD
4900 ? NXFME_ADDED : NXFME_MODIFIED;
4901
4902 bool change_cookie = (fm->modify_cookie
4903 && fm->new_cookie != OVS_BE64_MAX
4904 && fm->new_cookie != old_rule->flow_cookie);
4905
4906 bool change_actions = !ofpacts_equal(fm->ofpacts,
4907 fm->ofpacts_len,
4908 old_actions->ofpacts,
4909 old_actions->ofpacts_len);
4910
4911 if (event != NXFME_MODIFIED || change_actions || change_cookie) {
4912 ofmonitor_report(ofproto->connmgr, new_rule, event, 0,
4913 req ? req->ofconn : NULL,
4914 req ? req->request->xid : 0,
4915 change_actions ? old_actions : NULL);
4916 }
4917 } else {
4918 /* XXX: This is slight duplication with delete_flows_finish__() */
4919
4920 old_rule->removed_reason = OFPRR_EVICTION;
4921
4922 ofmonitor_report(ofproto->connmgr, old_rule, NXFME_DELETED,
4923 OFPRR_EVICTION,
4924 req ? req->ofconn : NULL,
4925 req ? req->request->xid : 0, NULL);
4926 }
4927 }
4928 }
4929
4930 static enum ofperr
4931 modify_flows_start__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4932 OVS_REQUIRES(ofproto_mutex)
4933 {
4934 struct ofputil_flow_mod *fm = &ofm->fm;
4935 struct rule_collection *old_rules = &ofm->old_rules;
4936 struct rule_collection *new_rules = &ofm->new_rules;
4937 enum ofperr error;
4938
4939 rule_collection_init(new_rules);
4940
4941 if (old_rules->n > 0) {
4942 struct cls_conjunction *conjs;
4943 size_t n_conjs;
4944 size_t i;
4945
4946 /* Create a new 'modified' rule for each old rule. */
4947 for (i = 0; i < old_rules->n; i++) {
4948 struct rule *old_rule = old_rules->rules[i];
4949 struct rule *new_rule;
4950 struct cls_rule cr;
4951
4952 cls_rule_clone(&cr, &old_rule->cr);
4953 error = replace_rule_create(ofproto, fm, &cr, old_rule->table_id,
4954 old_rule, &new_rule);
4955 if (!error) {
4956 rule_collection_add(new_rules, new_rule);
4957 } else {
4958 rule_collection_unref(new_rules);
4959 rule_collection_destroy(new_rules);
4960 return error;
4961 }
4962 }
4963 ovs_assert(new_rules->n == old_rules->n);
4964
4965 get_conjunctions(fm, &conjs, &n_conjs);
4966 for (i = 0; i < old_rules->n; i++) {
4967 replace_rule_start(ofproto, ofm->version, old_rules->rules[i],
4968 new_rules->rules[i], conjs, n_conjs);
4969 }
4970 free(conjs);
4971 } else if (!(fm->cookie_mask != htonll(0)
4972 || fm->new_cookie == OVS_BE64_MAX)) {
4973 /* No match, add a new flow. */
4974 error = add_flow_start(ofproto, ofm);
4975 if (!error) {
4976 ovs_assert(fm->delete_reason == OFPRR_EVICTION
4977 || !old_rules->rules[0]);
4978 }
4979 new_rules->n = 1;
4980 } else {
4981 error = 0;
4982 }
4983
4984 return error;
4985 }
4986
4987 /* Implements OFPFC_MODIFY. Returns 0 on success or an OpenFlow error code on
4988 * failure.
4989 *
4990 * 'ofconn' is used to retrieve the packet buffer specified in fm->buffer_id,
4991 * if any. */
4992 static enum ofperr
4993 modify_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
4994 OVS_REQUIRES(ofproto_mutex)
4995 {
4996 struct ofputil_flow_mod *fm = &ofm->fm;
4997 struct rule_collection *old_rules = &ofm->old_rules;
4998 struct rule_criteria criteria;
4999 enum ofperr error;
5000
5001 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
5002 fm->cookie, fm->cookie_mask, OFPP_ANY, OFPG_ANY);
5003 rule_criteria_require_rw(&criteria,
5004 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5005 error = collect_rules_loose(ofproto, &criteria, old_rules);
5006 rule_criteria_destroy(&criteria);
5007
5008 if (!error) {
5009 error = modify_flows_start__(ofproto, ofm);
5010 }
5011
5012 if (error) {
5013 rule_collection_destroy(old_rules);
5014 }
5015 return error;
5016 }
5017
5018 static void
5019 modify_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5020 OVS_REQUIRES(ofproto_mutex)
5021 {
5022 struct rule_collection *old_rules = &ofm->old_rules;
5023 struct rule_collection *new_rules = &ofm->new_rules;
5024
5025 /* Old rules were not changed yet, only need to revert new rules. */
5026 if (old_rules->n == 0 && new_rules->n == 1) {
5027 add_flow_revert(ofproto, ofm);
5028 } else if (old_rules->n > 0) {
5029 for (size_t i = 0; i < old_rules->n; i++) {
5030 replace_rule_revert(ofproto, old_rules->rules[i],
5031 new_rules->rules[i]);
5032 }
5033 rule_collection_destroy(new_rules);
5034 rule_collection_destroy(old_rules);
5035 }
5036 }
5037
5038 static void
5039 modify_flows_finish(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5040 const struct flow_mod_requester *req)
5041 OVS_REQUIRES(ofproto_mutex)
5042 {
5043 struct ofputil_flow_mod *fm = &ofm->fm;
5044 struct rule_collection *old_rules = &ofm->old_rules;
5045 struct rule_collection *new_rules = &ofm->new_rules;
5046
5047 if (old_rules->n == 0 && new_rules->n == 1) {
5048 add_flow_finish(ofproto, ofm, req);
5049 } else if (old_rules->n > 0) {
5050 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5051
5052 ovs_assert(new_rules->n == old_rules->n);
5053
5054 for (size_t i = 0; i < old_rules->n; i++) {
5055 replace_rule_finish(ofproto, fm, req, old_rules->rules[i],
5056 new_rules->rules[i], &dead_cookies);
5057 }
5058 learned_cookies_flush(ofproto, &dead_cookies);
5059 rule_collection_remove_postponed(old_rules);
5060
5061 send_buffered_packet(req, fm->buffer_id, new_rules->rules[0]);
5062 rule_collection_destroy(new_rules);
5063 }
5064 }
5065
5066 /* Implements OFPFC_MODIFY_STRICT. Returns 0 on success or an OpenFlow error
5067 * code on failure. */
5068 static enum ofperr
5069 modify_flow_start_strict(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5070 OVS_REQUIRES(ofproto_mutex)
5071 {
5072 struct ofputil_flow_mod *fm = &ofm->fm;
5073 struct rule_collection *old_rules = &ofm->old_rules;
5074 struct rule_criteria criteria;
5075 enum ofperr error;
5076
5077 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5078 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask, OFPP_ANY,
5079 OFPG_ANY);
5080 rule_criteria_require_rw(&criteria,
5081 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5082 error = collect_rules_strict(ofproto, &criteria, old_rules);
5083 rule_criteria_destroy(&criteria);
5084
5085 if (!error) {
5086 /* collect_rules_strict() can return max 1 rule. */
5087 error = modify_flows_start__(ofproto, ofm);
5088 }
5089
5090 if (error) {
5091 rule_collection_destroy(old_rules);
5092 }
5093 return error;
5094 }
5095 \f
5096 /* OFPFC_DELETE implementation. */
5097
5098 static void
5099 delete_flows_start__(struct ofproto *ofproto, cls_version_t version,
5100 const struct rule_collection *rules)
5101 OVS_REQUIRES(ofproto_mutex)
5102 {
5103 for (size_t i = 0; i < rules->n; i++) {
5104 struct rule *rule = rules->rules[i];
5105 struct oftable *table = &ofproto->tables[rule->table_id];
5106
5107 table->n_flows--;
5108 cls_rule_make_invisible_in_version(&rule->cr, version);
5109 }
5110 }
5111
5112 static void
5113 delete_flows_finish__(struct ofproto *ofproto,
5114 struct rule_collection *rules,
5115 enum ofp_flow_removed_reason reason,
5116 const struct flow_mod_requester *req)
5117 OVS_REQUIRES(ofproto_mutex)
5118 {
5119 if (rules->n) {
5120 struct ovs_list dead_cookies = OVS_LIST_INITIALIZER(&dead_cookies);
5121
5122 for (size_t i = 0; i < rules->n; i++) {
5123 struct rule *rule = rules->rules[i];
5124
5125 /* This value will be used to send the flow removed message right
5126 * before the rule is actually destroyed. */
5127 rule->removed_reason = reason;
5128
5129 ofmonitor_report(ofproto->connmgr, rule, NXFME_DELETED, reason,
5130 req ? req->ofconn : NULL,
5131 req ? req->request->xid : 0, NULL);
5132
5133 /* Send Vacancy Event for OF1.4+. */
5134 send_table_status(ofproto, rule->table_id);
5135
5136 ofproto_rule_remove__(ofproto, rule);
5137 learned_cookies_dec(ofproto, rule_get_actions(rule),
5138 &dead_cookies);
5139 }
5140 rule_collection_remove_postponed(rules);
5141
5142 learned_cookies_flush(ofproto, &dead_cookies);
5143 }
5144 }
5145
5146 /* Deletes the rules listed in 'rules'.
5147 * The deleted rules will become invisible to the lookups in the next version.
5148 * Destroys 'rules'. */
5149 static void
5150 delete_flows__(struct rule_collection *rules,
5151 enum ofp_flow_removed_reason reason,
5152 const struct flow_mod_requester *req)
5153 OVS_REQUIRES(ofproto_mutex)
5154 {
5155 if (rules->n) {
5156 struct ofproto *ofproto = rules->rules[0]->ofproto;
5157
5158 delete_flows_start__(ofproto, ofproto->tables_version + 1, rules);
5159 ofproto_bump_tables_version(ofproto);
5160 delete_flows_finish__(ofproto, rules, reason, req);
5161 ofmonitor_flush(ofproto->connmgr);
5162 }
5163 }
5164
5165 /* Implements OFPFC_DELETE. */
5166 static enum ofperr
5167 delete_flows_start_loose(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5168 OVS_REQUIRES(ofproto_mutex)
5169 {
5170 const struct ofputil_flow_mod *fm = &ofm->fm;
5171 struct rule_collection *rules = &ofm->old_rules;
5172 struct rule_criteria criteria;
5173 enum ofperr error;
5174
5175 rule_criteria_init(&criteria, fm->table_id, &fm->match, 0, CLS_MAX_VERSION,
5176 fm->cookie, fm->cookie_mask, fm->out_port,
5177 fm->out_group);
5178 rule_criteria_require_rw(&criteria,
5179 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5180 error = collect_rules_loose(ofproto, &criteria, rules);
5181 rule_criteria_destroy(&criteria);
5182
5183 if (!error) {
5184 delete_flows_start__(ofproto, ofm->version, rules);
5185 }
5186
5187 return error;
5188 }
5189
5190 static void
5191 delete_flows_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
5192 OVS_REQUIRES(ofproto_mutex)
5193 {
5194 struct rule_collection *rules = &ofm->old_rules;
5195
5196 for (size_t i = 0; i < rules->n; i++) {
5197 struct rule *rule = rules->rules[i];
5198 struct oftable *table = &ofproto->tables[rule->table_id];
5199
5200 /* Restore table's rule count. */
5201 table->n_flows++;
5202
5203 /* Restore the original visibility of the rule. */
5204 cls_rule_restore_visibility(&rule->cr);
5205 }
5206 rule_collection_destroy(rules);
5207 }
5208
5209 static void
5210 delete_flows_finish(struct ofproto *ofproto,
5211 struct ofproto_flow_mod *ofm,
5212 const struct flow_mod_requester *req)
5213 OVS_REQUIRES(ofproto_mutex)
5214 {
5215 delete_flows_finish__(ofproto, &ofm->old_rules, ofm->fm.delete_reason,
5216 req);
5217 }
5218
5219 /* Implements OFPFC_DELETE_STRICT. */
5220 static enum ofperr
5221 delete_flow_start_strict(struct ofproto *ofproto,
5222 struct ofproto_flow_mod *ofm)
5223 OVS_REQUIRES(ofproto_mutex)
5224 {
5225 const struct ofputil_flow_mod *fm = &ofm->fm;
5226 struct rule_collection *rules = &ofm->old_rules;
5227 struct rule_criteria criteria;
5228 enum ofperr error;
5229
5230 rule_criteria_init(&criteria, fm->table_id, &fm->match, fm->priority,
5231 CLS_MAX_VERSION, fm->cookie, fm->cookie_mask,
5232 fm->out_port, fm->out_group);
5233 rule_criteria_require_rw(&criteria,
5234 (fm->flags & OFPUTIL_FF_NO_READONLY) != 0);
5235 error = collect_rules_strict(ofproto, &criteria, rules);
5236 rule_criteria_destroy(&criteria);
5237
5238 if (!error) {
5239 delete_flows_start__(ofproto, ofm->version, rules);
5240 }
5241
5242 return error;
5243 }
5244
5245 /* This may only be called by rule_destroy_cb()! */
5246 static void
5247 ofproto_rule_send_removed(struct rule *rule)
5248 OVS_EXCLUDED(ofproto_mutex)
5249 {
5250 struct ofputil_flow_removed fr;
5251 long long int used;
5252
5253 minimatch_expand(&rule->cr.match, &fr.match);
5254 fr.priority = rule->cr.priority;
5255
5256 ovs_mutex_lock(&ofproto_mutex);
5257 fr.cookie = rule->flow_cookie;
5258 fr.reason = rule->removed_reason;
5259 fr.table_id = rule->table_id;
5260 calc_duration(rule->created, time_msec(),
5261 &fr.duration_sec, &fr.duration_nsec);
5262 ovs_mutex_lock(&rule->mutex);
5263 fr.idle_timeout = rule->idle_timeout;
5264 fr.hard_timeout = rule->hard_timeout;
5265 ovs_mutex_unlock(&rule->mutex);
5266 rule->ofproto->ofproto_class->rule_get_stats(rule, &fr.packet_count,
5267 &fr.byte_count, &used);
5268 connmgr_send_flow_removed(rule->ofproto->connmgr, &fr);
5269 ovs_mutex_unlock(&ofproto_mutex);
5270 }
5271
5272 /* Sends an OpenFlow "flow removed" message with the given 'reason' (either
5273 * OFPRR_HARD_TIMEOUT or OFPRR_IDLE_TIMEOUT), and then removes 'rule' from its
5274 * ofproto.
5275 *
5276 * ofproto implementation ->run() functions should use this function to expire
5277 * OpenFlow flows. */
5278 void
5279 ofproto_rule_expire(struct rule *rule, uint8_t reason)
5280 OVS_REQUIRES(ofproto_mutex)
5281 {
5282 struct rule_collection rules;
5283
5284 rules.rules = rules.stub;
5285 rules.n = 1;
5286 rules.stub[0] = rule;
5287 delete_flows__(&rules, reason, NULL);
5288 }
5289
5290 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
5291 * means "infinite". */
5292 static void
5293 reduce_timeout(uint16_t max, uint16_t *timeout)
5294 {
5295 if (max && (!*timeout || *timeout > max)) {
5296 *timeout = max;
5297 }
5298 }
5299
5300 /* If 'idle_timeout' is nonzero, and 'rule' has no idle timeout or an idle
5301 * timeout greater than 'idle_timeout', lowers 'rule''s idle timeout to
5302 * 'idle_timeout' seconds. Similarly for 'hard_timeout'.
5303 *
5304 * Suitable for implementing OFPACT_FIN_TIMEOUT. */
5305 void
5306 ofproto_rule_reduce_timeouts(struct rule *rule,
5307 uint16_t idle_timeout, uint16_t hard_timeout)
5308 OVS_EXCLUDED(ofproto_mutex, rule->mutex)
5309 {
5310 if (!idle_timeout && !hard_timeout) {
5311 return;
5312 }
5313
5314 ovs_mutex_lock(&ofproto_mutex);
5315 if (ovs_list_is_empty(&rule->expirable)) {
5316 ovs_list_insert(&rule->ofproto->expirable, &rule->expirable);
5317 }
5318 ovs_mutex_unlock(&ofproto_mutex);
5319
5320 ovs_mutex_lock(&rule->mutex);
5321 reduce_timeout(idle_timeout, &rule->idle_timeout);
5322 reduce_timeout(hard_timeout, &rule->hard_timeout);
5323 ovs_mutex_unlock(&rule->mutex);
5324 }
5325 \f
5326 static enum ofperr
5327 handle_flow_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5328 OVS_EXCLUDED(ofproto_mutex)
5329 {
5330 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5331 struct ofproto_flow_mod ofm;
5332 uint64_t ofpacts_stub[1024 / 8];
5333 struct ofpbuf ofpacts;
5334 enum ofperr error;
5335
5336 error = reject_slave_controller(ofconn);
5337 if (error) {
5338 goto exit;
5339 }
5340
5341 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
5342 error = ofputil_decode_flow_mod(&ofm.fm, oh, ofconn_get_protocol(ofconn),
5343 &ofpacts,
5344 u16_to_ofp(ofproto->max_ports),
5345 ofproto->n_tables);
5346 if (!error) {
5347 error = ofproto_check_ofpacts(ofproto, ofm.fm.ofpacts,
5348 ofm.fm.ofpacts_len);
5349 }
5350 if (!error) {
5351 struct flow_mod_requester req;
5352
5353 req.ofconn = ofconn;
5354 req.request = oh;
5355 error = handle_flow_mod__(ofproto, &ofm, &req);
5356 }
5357 if (error) {
5358 goto exit_free_ofpacts;
5359 }
5360
5361 ofconn_report_flow_mod(ofconn, ofm.fm.command);
5362
5363 exit_free_ofpacts:
5364 ofpbuf_uninit(&ofpacts);
5365 exit:
5366 return error;
5367 }
5368
5369 static enum ofperr
5370 handle_flow_mod__(struct ofproto *ofproto, struct ofproto_flow_mod *ofm,
5371 const struct flow_mod_requester *req)
5372 OVS_EXCLUDED(ofproto_mutex)
5373 {
5374 enum ofperr error;
5375
5376 ovs_mutex_lock(&ofproto_mutex);
5377 ofm->version = ofproto->tables_version + 1;
5378 error = ofproto_flow_mod_start(ofproto, ofm);
5379 if (!error) {
5380 ofproto_bump_tables_version(ofproto);
5381 ofproto_flow_mod_finish(ofproto, ofm, req);
5382 }
5383 ofmonitor_flush(ofproto->connmgr);
5384 ovs_mutex_unlock(&ofproto_mutex);
5385
5386 run_rule_executes(ofproto);
5387 return error;
5388 }
5389
5390 static enum ofperr
5391 handle_role_request(struct ofconn *ofconn, const struct ofp_header *oh)
5392 {
5393 struct ofputil_role_request request;
5394 struct ofputil_role_request reply;
5395 struct ofpbuf *buf;
5396 enum ofperr error;
5397
5398 error = ofputil_decode_role_message(oh, &request);
5399 if (error) {
5400 return error;
5401 }
5402
5403 if (request.role != OFPCR12_ROLE_NOCHANGE) {
5404 if (request.role != OFPCR12_ROLE_EQUAL
5405 && request.have_generation_id
5406 && !ofconn_set_master_election_id(ofconn, request.generation_id)) {
5407 return OFPERR_OFPRRFC_STALE;
5408 }
5409
5410 ofconn_set_role(ofconn, request.role);
5411 }
5412
5413 reply.role = ofconn_get_role(ofconn);
5414 reply.have_generation_id = ofconn_get_master_election_id(
5415 ofconn, &reply.generation_id);
5416 buf = ofputil_encode_role_reply(oh, &reply);
5417 ofconn_send_reply(ofconn, buf);
5418
5419 return 0;
5420 }
5421
5422 static enum ofperr
5423 handle_nxt_flow_mod_table_id(struct ofconn *ofconn,
5424 const struct ofp_header *oh)
5425 {
5426 const struct nx_flow_mod_table_id *msg = ofpmsg_body(oh);
5427 enum ofputil_protocol cur, next;
5428
5429 cur = ofconn_get_protocol(ofconn);
5430 next = ofputil_protocol_set_tid(cur, msg->set != 0);
5431 ofconn_set_protocol(ofconn, next);
5432
5433 return 0;
5434 }
5435
5436 static enum ofperr
5437 handle_nxt_set_flow_format(struct ofconn *ofconn, const struct ofp_header *oh)
5438 {
5439 const struct nx_set_flow_format *msg = ofpmsg_body(oh);
5440 enum ofputil_protocol cur, next;
5441 enum ofputil_protocol next_base;
5442
5443 next_base = ofputil_nx_flow_format_to_protocol(ntohl(msg->format));
5444 if (!next_base) {
5445 return OFPERR_OFPBRC_EPERM;
5446 }
5447
5448 cur = ofconn_get_protocol(ofconn);
5449 next = ofputil_protocol_set_base(cur, next_base);
5450 ofconn_set_protocol(ofconn, next);
5451
5452 return 0;
5453 }
5454
5455 static enum ofperr
5456 handle_nxt_set_packet_in_format(struct ofconn *ofconn,
5457 const struct ofp_header *oh)
5458 {
5459 const struct nx_set_packet_in_format *msg = ofpmsg_body(oh);
5460 uint32_t format;
5461
5462 format = ntohl(msg->format);
5463 if (!ofputil_packet_in_format_is_valid(format)) {
5464 return OFPERR_OFPBRC_EPERM;
5465 }
5466
5467 ofconn_set_packet_in_format(ofconn, format);
5468 return 0;
5469 }
5470
5471 static enum ofperr
5472 handle_nxt_set_async_config(struct ofconn *ofconn, const struct ofp_header *oh)
5473 {
5474 struct ofputil_async_cfg basis = ofconn_get_async_config(ofconn);
5475 struct ofputil_async_cfg ac;
5476 enum ofperr error;
5477
5478 error = ofputil_decode_set_async_config(oh, false, &basis, &ac);
5479 if (error) {
5480 return error;
5481 }
5482
5483 ofconn_set_async_config(ofconn, &ac);
5484 if (ofconn_get_type(ofconn) == OFCONN_SERVICE &&
5485 !ofconn_get_miss_send_len(ofconn)) {
5486 ofconn_set_miss_send_len(ofconn, OFP_DEFAULT_MISS_SEND_LEN);
5487 }
5488
5489 return 0;
5490 }
5491
5492 static enum ofperr
5493 handle_nxt_get_async_request(struct ofconn *ofconn, const struct ofp_header *oh)
5494 {
5495 struct ofputil_async_cfg ac = ofconn_get_async_config(ofconn);
5496 ofconn_send_reply(ofconn, ofputil_encode_get_async_reply(oh, &ac));
5497
5498 return 0;
5499 }
5500
5501 static enum ofperr
5502 handle_nxt_set_controller_id(struct ofconn *ofconn,
5503 const struct ofp_header *oh)
5504 {
5505 const struct nx_controller_id *nci = ofpmsg_body(oh);
5506
5507 if (!is_all_zeros(nci->zero, sizeof nci->zero)) {
5508 return OFPERR_NXBRC_MUST_BE_ZERO;
5509 }
5510
5511 ofconn_set_controller_id(ofconn, ntohs(nci->controller_id));
5512 return 0;
5513 }
5514
5515 static enum ofperr
5516 handle_barrier_request(struct ofconn *ofconn, const struct ofp_header *oh)
5517 {
5518 struct ofpbuf *buf;
5519
5520 buf = ofpraw_alloc_reply((oh->version == OFP10_VERSION
5521 ? OFPRAW_OFPT10_BARRIER_REPLY
5522 : OFPRAW_OFPT11_BARRIER_REPLY), oh, 0);
5523 ofconn_send_reply(ofconn, buf);
5524 return 0;
5525 }
5526
5527 static void
5528 ofproto_compose_flow_refresh_update(const struct rule *rule,
5529 enum nx_flow_monitor_flags flags,
5530 struct ovs_list *msgs)
5531 OVS_REQUIRES(ofproto_mutex)
5532 {
5533 const struct rule_actions *actions;
5534 struct ofputil_flow_update fu;
5535 struct match match;
5536
5537 fu.event = (flags & (NXFMF_INITIAL | NXFMF_ADD)
5538 ? NXFME_ADDED : NXFME_MODIFIED);
5539 fu.reason = 0;
5540 ovs_mutex_lock(&rule->mutex);
5541 fu.idle_timeout = rule->idle_timeout;
5542 fu.hard_timeout = rule->hard_timeout;
5543 ovs_mutex_unlock(&rule->mutex);
5544 fu.table_id = rule->table_id;
5545 fu.cookie = rule->flow_cookie;
5546 minimatch_expand(&rule->cr.match, &match);
5547 fu.match = &match;
5548 fu.priority = rule->cr.priority;
5549
5550 actions = flags & NXFMF_ACTIONS ? rule_get_actions(rule) : NULL;
5551 fu.ofpacts = actions ? actions->ofpacts : NULL;
5552 fu.ofpacts_len = actions ? actions->ofpacts_len : 0;
5553
5554 if (ovs_list_is_empty(msgs)) {
5555 ofputil_start_flow_update(msgs);
5556 }
5557 ofputil_append_flow_update(&fu, msgs);
5558 }
5559
5560 void
5561 ofmonitor_compose_refresh_updates(struct rule_collection *rules,
5562 struct ovs_list *msgs)
5563 OVS_REQUIRES(ofproto_mutex)
5564 {
5565 size_t i;
5566
5567 for (i = 0; i < rules->n; i++) {
5568 struct rule *rule = rules->rules[i];
5569 enum nx_flow_monitor_flags flags = rule->monitor_flags;
5570 rule->monitor_flags = 0;
5571
5572 ofproto_compose_flow_refresh_update(rule, flags, msgs);
5573 }
5574 }
5575
5576 static void
5577 ofproto_collect_ofmonitor_refresh_rule(const struct ofmonitor *m,
5578 struct rule *rule, uint64_t seqno,
5579 struct rule_collection *rules)
5580 OVS_REQUIRES(ofproto_mutex)
5581 {
5582 enum nx_flow_monitor_flags update;
5583
5584 if (rule_is_hidden(rule)) {
5585 return;
5586 }
5587
5588 if (!ofproto_rule_has_out_port(rule, m->out_port)) {
5589 return;
5590 }
5591
5592 if (seqno) {
5593 if (rule->add_seqno > seqno) {
5594 update = NXFMF_ADD | NXFMF_MODIFY;
5595 } else if (rule->modify_seqno > seqno) {
5596 update = NXFMF_MODIFY;
5597 } else {
5598 return;
5599 }
5600
5601 if (!(m->flags & update)) {
5602 return;
5603 }
5604 } else {
5605 update = NXFMF_INITIAL;
5606 }
5607
5608 if (!rule->monitor_flags) {
5609 rule_collection_add(rules, rule);
5610 }
5611 rule->monitor_flags |= update | (m->flags & NXFMF_ACTIONS);
5612 }
5613
5614 static void
5615 ofproto_collect_ofmonitor_refresh_rules(const struct ofmonitor *m,
5616 uint64_t seqno,
5617 struct rule_collection *rules)
5618 OVS_REQUIRES(ofproto_mutex)
5619 {
5620 const struct ofproto *ofproto = ofconn_get_ofproto(m->ofconn);
5621 const struct oftable *table;
5622 struct cls_rule target;
5623
5624 cls_rule_init_from_minimatch(&target, &m->match, 0);
5625 FOR_EACH_MATCHING_TABLE (table, m->table_id, ofproto) {
5626 struct rule *rule;
5627
5628 CLS_FOR_EACH_TARGET (rule, cr, &table->cls, &target, CLS_MAX_VERSION) {
5629 ofproto_collect_ofmonitor_refresh_rule(m, rule, seqno, rules);
5630 }
5631 }
5632 cls_rule_destroy(&target);
5633 }
5634
5635 static void
5636 ofproto_collect_ofmonitor_initial_rules(struct ofmonitor *m,
5637 struct rule_collection *rules)
5638 OVS_REQUIRES(ofproto_mutex)
5639 {
5640 if (m->flags & NXFMF_INITIAL) {
5641 ofproto_collect_ofmonitor_refresh_rules(m, 0, rules);
5642 }
5643 }
5644
5645 void
5646 ofmonitor_collect_resume_rules(struct ofmonitor *m,
5647 uint64_t seqno, struct rule_collection *rules)
5648 OVS_REQUIRES(ofproto_mutex)
5649 {
5650 ofproto_collect_ofmonitor_refresh_rules(m, seqno, rules);
5651 }
5652
5653 static enum ofperr
5654 flow_monitor_delete(struct ofconn *ofconn, uint32_t id)
5655 OVS_REQUIRES(ofproto_mutex)
5656 {
5657 struct ofmonitor *m;
5658 enum ofperr error;
5659
5660 m = ofmonitor_lookup(ofconn, id);
5661 if (m) {
5662 ofmonitor_destroy(m);
5663 error = 0;
5664 } else {
5665 error = OFPERR_OFPMOFC_UNKNOWN_MONITOR;
5666 }
5667
5668 return error;
5669 }
5670
5671 static enum ofperr
5672 handle_flow_monitor_request(struct ofconn *ofconn, const struct ofp_header *oh)
5673 OVS_EXCLUDED(ofproto_mutex)
5674 {
5675 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5676
5677 struct ofpbuf b = ofpbuf_const_initializer(oh, ntohs(oh->length));
5678
5679 struct ofmonitor **monitors = NULL;
5680 size_t allocated_monitors = 0;
5681 size_t n_monitors = 0;
5682
5683 enum ofperr error;
5684
5685 ovs_mutex_lock(&ofproto_mutex);
5686 for (;;) {
5687 struct ofputil_flow_monitor_request request;
5688 struct ofmonitor *m;
5689 int retval;
5690
5691 retval = ofputil_decode_flow_monitor_request(&request, &b);
5692 if (retval == EOF) {
5693 break;
5694 } else if (retval) {
5695 error = retval;
5696 goto error;
5697 }
5698
5699 if (request.table_id != 0xff
5700 && request.table_id >= ofproto->n_tables) {
5701 error = OFPERR_OFPBRC_BAD_TABLE_ID;
5702 goto error;
5703 }
5704
5705 error = ofmonitor_create(&request, ofconn, &m);
5706 if (error) {
5707 goto error;
5708 }
5709
5710 if (n_monitors >= allocated_monitors) {
5711 monitors = x2nrealloc(monitors, &allocated_monitors,
5712 sizeof *monitors);
5713 }
5714 monitors[n_monitors++] = m;
5715 }
5716
5717 struct rule_collection rules;
5718 rule_collection_init(&rules);
5719 for (size_t i = 0; i < n_monitors; i++) {
5720 ofproto_collect_ofmonitor_initial_rules(monitors[i], &rules);
5721 }
5722
5723 struct ovs_list replies;
5724 ofpmp_init(&replies, oh);
5725 ofmonitor_compose_refresh_updates(&rules, &replies);
5726 ovs_mutex_unlock(&ofproto_mutex);
5727
5728 rule_collection_destroy(&rules);
5729
5730 ofconn_send_replies(ofconn, &replies);
5731 free(monitors);
5732
5733 return 0;
5734
5735 error:
5736 for (size_t i = 0; i < n_monitors; i++) {
5737 ofmonitor_destroy(monitors[i]);
5738 }
5739 free(monitors);
5740 ovs_mutex_unlock(&ofproto_mutex);
5741
5742 return error;
5743 }
5744
5745 static enum ofperr
5746 handle_flow_monitor_cancel(struct ofconn *ofconn, const struct ofp_header *oh)
5747 OVS_EXCLUDED(ofproto_mutex)
5748 {
5749 enum ofperr error;
5750 uint32_t id;
5751
5752 id = ofputil_decode_flow_monitor_cancel(oh);
5753
5754 ovs_mutex_lock(&ofproto_mutex);
5755 error = flow_monitor_delete(ofconn, id);
5756 ovs_mutex_unlock(&ofproto_mutex);
5757
5758 return error;
5759 }
5760
5761 /* Meters implementation.
5762 *
5763 * Meter table entry, indexed by the OpenFlow meter_id.
5764 * 'created' is used to compute the duration for meter stats.
5765 * 'list rules' is needed so that we can delete the dependent rules when the
5766 * meter table entry is deleted.
5767 * 'provider_meter_id' is for the provider's private use.
5768 */
5769 struct meter {
5770 long long int created; /* Time created. */
5771 struct ovs_list rules; /* List of "struct rule_dpif"s. */
5772 ofproto_meter_id provider_meter_id;
5773 uint16_t flags; /* Meter flags. */
5774 uint16_t n_bands; /* Number of meter bands. */
5775 struct ofputil_meter_band *bands;
5776 };
5777
5778 /*
5779 * This is used in instruction validation at flow set-up time,
5780 * as flows may not use non-existing meters.
5781 * Return value of UINT32_MAX signifies an invalid meter.
5782 */
5783 static uint32_t
5784 get_provider_meter_id(const struct ofproto *ofproto, uint32_t of_meter_id)
5785 {
5786 if (of_meter_id && of_meter_id <= ofproto->meter_features.max_meters) {
5787 const struct meter *meter = ofproto->meters[of_meter_id];
5788 if (meter) {
5789 return meter->provider_meter_id.uint32;
5790 }
5791 }
5792 return UINT32_MAX;
5793 }
5794
5795 /* Finds the meter invoked by 'rule''s actions and adds 'rule' to the meter's
5796 * list of rules. */
5797 static void
5798 meter_insert_rule(struct rule *rule)
5799 {
5800 const struct rule_actions *a = rule_get_actions(rule);
5801 uint32_t meter_id = ofpacts_get_meter(a->ofpacts, a->ofpacts_len);
5802 struct meter *meter = rule->ofproto->meters[meter_id];
5803
5804 ovs_list_insert(&meter->rules, &rule->meter_list_node);
5805 }
5806
5807 static void
5808 meter_update(struct meter *meter, const struct ofputil_meter_config *config)
5809 {
5810 free(meter->bands);
5811
5812 meter->flags = config->flags;
5813 meter->n_bands = config->n_bands;
5814 meter->bands = xmemdup(config->bands,
5815 config->n_bands * sizeof *meter->bands);
5816 }
5817
5818 static struct meter *
5819 meter_create(const struct ofputil_meter_config *config,
5820 ofproto_meter_id provider_meter_id)
5821 {
5822 struct meter *meter;
5823
5824 meter = xzalloc(sizeof *meter);
5825 meter->provider_meter_id = provider_meter_id;
5826 meter->created = time_msec();
5827 ovs_list_init(&meter->rules);
5828
5829 meter_update(meter, config);
5830
5831 return meter;
5832 }
5833
5834 static void
5835 meter_delete(struct ofproto *ofproto, uint32_t first, uint32_t last)
5836 OVS_REQUIRES(ofproto_mutex)
5837 {
5838 uint32_t mid;
5839 for (mid = first; mid <= last; ++mid) {
5840 struct meter *meter = ofproto->meters[mid];
5841 if (meter) {
5842 ofproto->meters[mid] = NULL;
5843 ofproto->ofproto_class->meter_del(ofproto,
5844 meter->provider_meter_id);
5845 free(meter->bands);
5846 free(meter);
5847 }
5848 }
5849 }
5850
5851 static enum ofperr
5852 handle_add_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5853 {
5854 ofproto_meter_id provider_meter_id = { UINT32_MAX };
5855 struct meter **meterp = &ofproto->meters[mm->meter.meter_id];
5856 enum ofperr error;
5857
5858 if (*meterp) {
5859 return OFPERR_OFPMMFC_METER_EXISTS;
5860 }
5861
5862 error = ofproto->ofproto_class->meter_set(ofproto, &provider_meter_id,
5863 &mm->meter);
5864 if (!error) {
5865 ovs_assert(provider_meter_id.uint32 != UINT32_MAX);
5866 *meterp = meter_create(&mm->meter, provider_meter_id);
5867 }
5868 return error;
5869 }
5870
5871 static enum ofperr
5872 handle_modify_meter(struct ofproto *ofproto, struct ofputil_meter_mod *mm)
5873 {
5874 struct meter *meter = ofproto->meters[mm->meter.meter_id];
5875 enum ofperr error;
5876 uint32_t provider_meter_id;
5877
5878 if (!meter) {
5879 return OFPERR_OFPMMFC_UNKNOWN_METER;
5880 }
5881
5882 provider_meter_id = meter->provider_meter_id.uint32;
5883 error = ofproto->ofproto_class->meter_set(ofproto,
5884 &meter->provider_meter_id,
5885 &mm->meter);
5886 ovs_assert(meter->provider_meter_id.uint32 == provider_meter_id);
5887 if (!error) {
5888 meter_update(meter, &mm->meter);
5889 }
5890 return error;
5891 }
5892
5893 static enum ofperr
5894 handle_delete_meter(struct ofconn *ofconn, struct ofputil_meter_mod *mm)
5895 OVS_EXCLUDED(ofproto_mutex)
5896 {
5897 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5898 uint32_t meter_id = mm->meter.meter_id;
5899 struct rule_collection rules;
5900 enum ofperr error = 0;
5901 uint32_t first, last;
5902
5903 if (meter_id == OFPM13_ALL) {
5904 first = 1;
5905 last = ofproto->meter_features.max_meters;
5906 } else {
5907 if (!meter_id || meter_id > ofproto->meter_features.max_meters) {
5908 return 0;
5909 }
5910 first = last = meter_id;
5911 }
5912
5913 /* First delete the rules that use this meter. If any of those rules are
5914 * currently being modified, postpone the whole operation until later. */
5915 rule_collection_init(&rules);
5916 ovs_mutex_lock(&ofproto_mutex);
5917 for (meter_id = first; meter_id <= last; ++meter_id) {
5918 struct meter *meter = ofproto->meters[meter_id];
5919 if (meter && !ovs_list_is_empty(&meter->rules)) {
5920 struct rule *rule;
5921
5922 LIST_FOR_EACH (rule, meter_list_node, &meter->rules) {
5923 rule_collection_add(&rules, rule);
5924 }
5925 }
5926 }
5927 delete_flows__(&rules, OFPRR_METER_DELETE, NULL);
5928
5929 /* Delete the meters. */
5930 meter_delete(ofproto, first, last);
5931
5932 ovs_mutex_unlock(&ofproto_mutex);
5933
5934 return error;
5935 }
5936
5937 static enum ofperr
5938 handle_meter_mod(struct ofconn *ofconn, const struct ofp_header *oh)
5939 {
5940 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
5941 struct ofputil_meter_mod mm;
5942 uint64_t bands_stub[256 / 8];
5943 struct ofpbuf bands;
5944 uint32_t meter_id;
5945 enum ofperr error;
5946
5947 error = reject_slave_controller(ofconn);
5948 if (error) {
5949 return error;
5950 }
5951
5952 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
5953
5954 error = ofputil_decode_meter_mod(oh, &mm, &bands);
5955 if (error) {
5956 goto exit_free_bands;
5957 }
5958
5959 meter_id = mm.meter.meter_id;
5960
5961 if (mm.command != OFPMC13_DELETE) {
5962 /* Fails also when meters are not implemented by the provider. */
5963 if (meter_id == 0 || meter_id > OFPM13_MAX) {
5964 error = OFPERR_OFPMMFC_INVALID_METER;
5965 goto exit_free_bands;
5966 } else if (meter_id > ofproto->meter_features.max_meters) {
5967 error = OFPERR_OFPMMFC_OUT_OF_METERS;
5968 goto exit_free_bands;
5969 }
5970 if (mm.meter.n_bands > ofproto->meter_features.max_bands) {
5971 error = OFPERR_OFPMMFC_OUT_OF_BANDS;
5972 goto exit_free_bands;
5973 }
5974 }
5975
5976 switch (mm.command) {
5977 case OFPMC13_ADD:
5978 error = handle_add_meter(ofproto, &mm);
5979 break;
5980
5981 case OFPMC13_MODIFY:
5982 error = handle_modify_meter(ofproto, &mm);
5983 break;
5984
5985 case OFPMC13_DELETE:
5986 error = handle_delete_meter(ofconn, &mm);
5987 break;
5988
5989 default:
5990 error = OFPERR_OFPMMFC_BAD_COMMAND;
5991 break;
5992 }
5993
5994 if (!error) {
5995 struct ofputil_requestforward rf;
5996 rf.xid = oh->xid;
5997 rf.reason = OFPRFR_METER_MOD;
5998 rf.meter_mod = &mm;
5999 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6000 }
6001
6002 exit_free_bands:
6003 ofpbuf_uninit(&bands);
6004 return error;
6005 }
6006
6007 static enum ofperr
6008 handle_meter_features_request(struct ofconn *ofconn,
6009 const struct ofp_header *request)
6010 {
6011 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6012 struct ofputil_meter_features features;
6013 struct ofpbuf *b;
6014
6015 if (ofproto->ofproto_class->meter_get_features) {
6016 ofproto->ofproto_class->meter_get_features(ofproto, &features);
6017 } else {
6018 memset(&features, 0, sizeof features);
6019 }
6020 b = ofputil_encode_meter_features_reply(&features, request);
6021
6022 ofconn_send_reply(ofconn, b);
6023 return 0;
6024 }
6025
6026 static enum ofperr
6027 handle_meter_request(struct ofconn *ofconn, const struct ofp_header *request,
6028 enum ofptype type)
6029 {
6030 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6031 struct ovs_list replies;
6032 uint64_t bands_stub[256 / 8];
6033 struct ofpbuf bands;
6034 uint32_t meter_id, first, last;
6035
6036 ofputil_decode_meter_request(request, &meter_id);
6037
6038 if (meter_id == OFPM13_ALL) {
6039 first = 1;
6040 last = ofproto->meter_features.max_meters;
6041 } else {
6042 if (!meter_id || meter_id > ofproto->meter_features.max_meters ||
6043 !ofproto->meters[meter_id]) {
6044 return OFPERR_OFPMMFC_UNKNOWN_METER;
6045 }
6046 first = last = meter_id;
6047 }
6048
6049 ofpbuf_use_stub(&bands, bands_stub, sizeof bands_stub);
6050 ofpmp_init(&replies, request);
6051
6052 for (meter_id = first; meter_id <= last; ++meter_id) {
6053 struct meter *meter = ofproto->meters[meter_id];
6054 if (!meter) {
6055 continue; /* Skip non-existing meters. */
6056 }
6057 if (type == OFPTYPE_METER_STATS_REQUEST) {
6058 struct ofputil_meter_stats stats;
6059
6060 stats.meter_id = meter_id;
6061
6062 /* Provider sets the packet and byte counts, we do the rest. */
6063 stats.flow_count = ovs_list_size(&meter->rules);
6064 calc_duration(meter->created, time_msec(),
6065 &stats.duration_sec, &stats.duration_nsec);
6066 stats.n_bands = meter->n_bands;
6067 ofpbuf_clear(&bands);
6068 stats.bands
6069 = ofpbuf_put_uninit(&bands,
6070 meter->n_bands * sizeof *stats.bands);
6071
6072 if (!ofproto->ofproto_class->meter_get(ofproto,
6073 meter->provider_meter_id,
6074 &stats)) {
6075 ofputil_append_meter_stats(&replies, &stats);
6076 }
6077 } else { /* type == OFPTYPE_METER_CONFIG_REQUEST */
6078 struct ofputil_meter_config config;
6079
6080 config.meter_id = meter_id;
6081 config.flags = meter->flags;
6082 config.n_bands = meter->n_bands;
6083 config.bands = meter->bands;
6084 ofputil_append_meter_config(&replies, &config);
6085 }
6086 }
6087
6088 ofconn_send_replies(ofconn, &replies);
6089 ofpbuf_uninit(&bands);
6090 return 0;
6091 }
6092
6093 static bool
6094 ofproto_group_lookup__(const struct ofproto *ofproto, uint32_t group_id,
6095 struct ofgroup **group)
6096 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6097 {
6098 HMAP_FOR_EACH_IN_BUCKET (*group, hmap_node,
6099 hash_int(group_id, 0), &ofproto->groups) {
6100 if ((*group)->group_id == group_id) {
6101 return true;
6102 }
6103 }
6104
6105 return false;
6106 }
6107
6108 /* If the group exists, this function increments the groups's reference count.
6109 *
6110 * Make sure to call ofproto_group_unref() after no longer needing to maintain
6111 * a reference to the group. */
6112 bool
6113 ofproto_group_lookup(const struct ofproto *ofproto, uint32_t group_id,
6114 struct ofgroup **group)
6115 {
6116 bool found;
6117
6118 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6119 found = ofproto_group_lookup__(ofproto, group_id, group);
6120 if (found) {
6121 ofproto_group_ref(*group);
6122 }
6123 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6124 return found;
6125 }
6126
6127 static bool
6128 ofproto_group_exists__(const struct ofproto *ofproto, uint32_t group_id)
6129 OVS_REQ_RDLOCK(ofproto->groups_rwlock)
6130 {
6131 struct ofgroup *grp;
6132
6133 HMAP_FOR_EACH_IN_BUCKET (grp, hmap_node,
6134 hash_int(group_id, 0), &ofproto->groups) {
6135 if (grp->group_id == group_id) {
6136 return true;
6137 }
6138 }
6139 return false;
6140 }
6141
6142 static bool
6143 ofproto_group_exists(const struct ofproto *ofproto, uint32_t group_id)
6144 OVS_EXCLUDED(ofproto->groups_rwlock)
6145 {
6146 bool exists;
6147
6148 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6149 exists = ofproto_group_exists__(ofproto, group_id);
6150 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6151
6152 return exists;
6153 }
6154
6155 static uint32_t
6156 group_get_ref_count(struct ofgroup *group)
6157 OVS_EXCLUDED(ofproto_mutex)
6158 {
6159 struct ofproto *ofproto = CONST_CAST(struct ofproto *, group->ofproto);
6160 struct rule_criteria criteria;
6161 struct rule_collection rules;
6162 struct match match;
6163 enum ofperr error;
6164 uint32_t count;
6165
6166 match_init_catchall(&match);
6167 rule_criteria_init(&criteria, 0xff, &match, 0, CLS_MAX_VERSION, htonll(0),
6168 htonll(0), OFPP_ANY, group->group_id);
6169 ovs_mutex_lock(&ofproto_mutex);
6170 error = collect_rules_loose(ofproto, &criteria, &rules);
6171 ovs_mutex_unlock(&ofproto_mutex);
6172 rule_criteria_destroy(&criteria);
6173
6174 count = !error && rules.n < UINT32_MAX ? rules.n : UINT32_MAX;
6175
6176 rule_collection_destroy(&rules);
6177 return count;
6178 }
6179
6180 static void
6181 append_group_stats(struct ofgroup *group, struct ovs_list *replies)
6182 {
6183 struct ofputil_group_stats ogs;
6184 const struct ofproto *ofproto = group->ofproto;
6185 long long int now = time_msec();
6186 int error;
6187
6188 ogs.bucket_stats = xmalloc(group->n_buckets * sizeof *ogs.bucket_stats);
6189
6190 /* Provider sets the packet and byte counts, we do the rest. */
6191 ogs.ref_count = group_get_ref_count(group);
6192 ogs.n_buckets = group->n_buckets;
6193
6194 error = (ofproto->ofproto_class->group_get_stats
6195 ? ofproto->ofproto_class->group_get_stats(group, &ogs)
6196 : EOPNOTSUPP);
6197 if (error) {
6198 ogs.packet_count = UINT64_MAX;
6199 ogs.byte_count = UINT64_MAX;
6200 memset(ogs.bucket_stats, 0xff,
6201 ogs.n_buckets * sizeof *ogs.bucket_stats);
6202 }
6203
6204 ogs.group_id = group->group_id;
6205 calc_duration(group->created, now, &ogs.duration_sec, &ogs.duration_nsec);
6206
6207 ofputil_append_group_stats(replies, &ogs);
6208
6209 free(ogs.bucket_stats);
6210 }
6211
6212 static void
6213 handle_group_request(struct ofconn *ofconn,
6214 const struct ofp_header *request, uint32_t group_id,
6215 void (*cb)(struct ofgroup *, struct ovs_list *replies))
6216 {
6217 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6218 struct ofgroup *group;
6219 struct ovs_list replies;
6220
6221 ofpmp_init(&replies, request);
6222 if (group_id == OFPG_ALL) {
6223 ovs_rwlock_rdlock(&ofproto->groups_rwlock);
6224 HMAP_FOR_EACH (group, hmap_node, &ofproto->groups) {
6225 cb(group, &replies);
6226 }
6227 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6228 } else {
6229 if (ofproto_group_lookup(ofproto, group_id, &group)) {
6230 cb(group, &replies);
6231 ofproto_group_unref(group);
6232 }
6233 }
6234 ofconn_send_replies(ofconn, &replies);
6235 }
6236
6237 static enum ofperr
6238 handle_group_stats_request(struct ofconn *ofconn,
6239 const struct ofp_header *request)
6240 {
6241 uint32_t group_id;
6242 enum ofperr error;
6243
6244 error = ofputil_decode_group_stats_request(request, &group_id);
6245 if (error) {
6246 return error;
6247 }
6248
6249 handle_group_request(ofconn, request, group_id, append_group_stats);
6250 return 0;
6251 }
6252
6253 static void
6254 append_group_desc(struct ofgroup *group, struct ovs_list *replies)
6255 {
6256 struct ofputil_group_desc gds;
6257
6258 gds.group_id = group->group_id;
6259 gds.type = group->type;
6260 gds.props = group->props;
6261
6262 ofputil_append_group_desc_reply(&gds, &group->buckets, replies);
6263 }
6264
6265 static enum ofperr
6266 handle_group_desc_stats_request(struct ofconn *ofconn,
6267 const struct ofp_header *request)
6268 {
6269 handle_group_request(ofconn, request,
6270 ofputil_decode_group_desc_request(request),
6271 append_group_desc);
6272 return 0;
6273 }
6274
6275 static enum ofperr
6276 handle_group_features_stats_request(struct ofconn *ofconn,
6277 const struct ofp_header *request)
6278 {
6279 struct ofproto *p = ofconn_get_ofproto(ofconn);
6280 struct ofpbuf *msg;
6281
6282 msg = ofputil_encode_group_features_reply(&p->ogf, request);
6283 if (msg) {
6284 ofconn_send_reply(ofconn, msg);
6285 }
6286
6287 return 0;
6288 }
6289
6290 static void
6291 put_queue_get_config_reply(struct ofport *port, uint32_t queue,
6292 struct ovs_list *replies)
6293 {
6294 struct ofputil_queue_config qc;
6295
6296 /* None of the existing queues have compatible properties, so we hard-code
6297 * omitting min_rate and max_rate. */
6298 qc.port = port->ofp_port;
6299 qc.queue = queue;
6300 qc.min_rate = UINT16_MAX;
6301 qc.max_rate = UINT16_MAX;
6302 ofputil_append_queue_get_config_reply(&qc, replies);
6303 }
6304
6305 static int
6306 handle_queue_get_config_request_for_port(struct ofport *port, uint32_t queue,
6307 struct ovs_list *replies)
6308 {
6309 struct smap details = SMAP_INITIALIZER(&details);
6310 if (queue != OFPQ_ALL) {
6311 int error = netdev_get_queue(port->netdev, queue, &details);
6312 switch (error) {
6313 case 0:
6314 put_queue_get_config_reply(port, queue, replies);
6315 break;
6316 case EOPNOTSUPP:
6317 case EINVAL:
6318 return OFPERR_OFPQOFC_BAD_QUEUE;
6319 default:
6320 return OFPERR_NXQOFC_QUEUE_ERROR;
6321 }
6322 } else {
6323 struct netdev_queue_dump queue_dump;
6324 uint32_t queue_id;
6325
6326 NETDEV_QUEUE_FOR_EACH (&queue_id, &details, &queue_dump,
6327 port->netdev) {
6328 put_queue_get_config_reply(port, queue_id, replies);
6329 }
6330 }
6331 smap_destroy(&details);
6332 return 0;
6333 }
6334
6335 static enum ofperr
6336 handle_queue_get_config_request(struct ofconn *ofconn,
6337 const struct ofp_header *oh)
6338 {
6339 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6340 struct ovs_list replies;
6341 struct ofport *port;
6342 ofp_port_t req_port;
6343 uint32_t req_queue;
6344 enum ofperr error;
6345
6346 error = ofputil_decode_queue_get_config_request(oh, &req_port, &req_queue);
6347 if (error) {
6348 return error;
6349 }
6350
6351 ofputil_start_queue_get_config_reply(oh, &replies);
6352 if (req_port == OFPP_ANY) {
6353 error = OFPERR_OFPQOFC_BAD_QUEUE;
6354 HMAP_FOR_EACH (port, hmap_node, &ofproto->ports) {
6355 if (!handle_queue_get_config_request_for_port(port, req_queue,
6356 &replies)) {
6357 error = 0;
6358 }
6359 }
6360 } else {
6361 port = ofproto_get_port(ofproto, req_port);
6362 error = (port
6363 ? handle_queue_get_config_request_for_port(port, req_queue,
6364 &replies)
6365 : OFPERR_OFPQOFC_BAD_PORT);
6366 }
6367 if (!error) {
6368 ofconn_send_replies(ofconn, &replies);
6369 } else {
6370 ofpbuf_list_delete(&replies);
6371 }
6372
6373 return error;
6374 }
6375
6376 static enum ofperr
6377 init_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm,
6378 struct ofgroup **ofgroup)
6379 {
6380 enum ofperr error;
6381 const long long int now = time_msec();
6382
6383 if (gm->group_id > OFPG_MAX) {
6384 return OFPERR_OFPGMFC_INVALID_GROUP;
6385 }
6386 if (gm->type > OFPGT11_FF) {
6387 return OFPERR_OFPGMFC_BAD_TYPE;
6388 }
6389
6390 *ofgroup = ofproto->ofproto_class->group_alloc();
6391 if (!*ofgroup) {
6392 VLOG_WARN_RL(&rl, "%s: failed to allocate group", ofproto->name);
6393 return OFPERR_OFPGMFC_OUT_OF_GROUPS;
6394 }
6395
6396 (*ofgroup)->ofproto = ofproto;
6397 *CONST_CAST(uint32_t *, &((*ofgroup)->group_id)) = gm->group_id;
6398 *CONST_CAST(enum ofp11_group_type *, &(*ofgroup)->type) = gm->type;
6399 *CONST_CAST(long long int *, &((*ofgroup)->created)) = now;
6400 *CONST_CAST(long long int *, &((*ofgroup)->modified)) = now;
6401 ovs_refcount_init(&(*ofgroup)->ref_count);
6402
6403 ovs_list_init(&(*ofgroup)->buckets);
6404 ofputil_bucket_clone_list(&(*ofgroup)->buckets, &gm->buckets, NULL);
6405
6406 *CONST_CAST(uint32_t *, &(*ofgroup)->n_buckets) =
6407 ovs_list_size(&(*ofgroup)->buckets);
6408
6409 memcpy(CONST_CAST(struct ofputil_group_props *, &(*ofgroup)->props),
6410 &gm->props, sizeof (struct ofputil_group_props));
6411
6412 /* Construct called BEFORE any locks are held. */
6413 error = ofproto->ofproto_class->group_construct(*ofgroup);
6414 if (error) {
6415 ofputil_bucket_list_destroy(&(*ofgroup)->buckets);
6416 ofproto->ofproto_class->group_dealloc(*ofgroup);
6417 }
6418 return error;
6419 }
6420
6421 /* Implements the OFPGC11_ADD operation specified by 'gm', adding a group to
6422 * 'ofproto''s group table. Returns 0 on success or an OpenFlow error code on
6423 * failure. */
6424 static enum ofperr
6425 add_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6426 {
6427 struct ofgroup *ofgroup;
6428 enum ofperr error;
6429
6430 /* Allocate new group and initialize it. */
6431 error = init_group(ofproto, gm, &ofgroup);
6432 if (error) {
6433 return error;
6434 }
6435
6436 /* We wrlock as late as possible to minimize the time we jam any other
6437 * threads: No visible state changes before acquiring the lock. */
6438 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6439
6440 if (ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6441 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6442 goto unlock_out;
6443 }
6444
6445 if (ofproto_group_exists__(ofproto, gm->group_id)) {
6446 error = OFPERR_OFPGMFC_GROUP_EXISTS;
6447 goto unlock_out;
6448 }
6449
6450 if (!error) {
6451 /* Insert new group. */
6452 hmap_insert(&ofproto->groups, &ofgroup->hmap_node,
6453 hash_int(ofgroup->group_id, 0));
6454 ofproto->n_groups[ofgroup->type]++;
6455
6456 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6457 return error;
6458 }
6459
6460 unlock_out:
6461 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6462 ofproto->ofproto_class->group_destruct(ofgroup);
6463 ofputil_bucket_list_destroy(&ofgroup->buckets);
6464 ofproto->ofproto_class->group_dealloc(ofgroup);
6465
6466 return error;
6467 }
6468
6469 /* Adds all of the buckets from 'ofgroup' to 'new_ofgroup'. The buckets
6470 * already in 'new_ofgroup' will be placed just after the (copy of the) bucket
6471 * in 'ofgroup' with bucket ID 'command_bucket_id'. Special
6472 * 'command_bucket_id' values OFPG15_BUCKET_FIRST and OFPG15_BUCKET_LAST are
6473 * also honored. */
6474 static enum ofperr
6475 copy_buckets_for_insert_bucket(const struct ofgroup *ofgroup,
6476 struct ofgroup *new_ofgroup,
6477 uint32_t command_bucket_id)
6478 {
6479 struct ofputil_bucket *last = NULL;
6480
6481 if (command_bucket_id <= OFPG15_BUCKET_MAX) {
6482 /* Check here to ensure that a bucket corresponding to
6483 * command_bucket_id exists in the old bucket list.
6484 *
6485 * The subsequent search of below of new_ofgroup covers
6486 * both buckets in the old bucket list and buckets added
6487 * by the insert buckets group mod message this function processes. */
6488 if (!ofputil_bucket_find(&ofgroup->buckets, command_bucket_id)) {
6489 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6490 }
6491
6492 if (!ovs_list_is_empty(&new_ofgroup->buckets)) {
6493 last = ofputil_bucket_list_back(&new_ofgroup->buckets);
6494 }
6495 }
6496
6497 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, NULL);
6498
6499 if (ofputil_bucket_check_duplicate_id(&new_ofgroup->buckets)) {
6500 VLOG_INFO_RL(&rl, "Duplicate bucket id");
6501 return OFPERR_OFPGMFC_BUCKET_EXISTS;
6502 }
6503
6504 /* Rearrange list according to command_bucket_id */
6505 if (command_bucket_id == OFPG15_BUCKET_LAST) {
6506 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6507 struct ofputil_bucket *new_first;
6508 const struct ofputil_bucket *first;
6509
6510 first = ofputil_bucket_list_front(&ofgroup->buckets);
6511 new_first = ofputil_bucket_find(&new_ofgroup->buckets,
6512 first->bucket_id);
6513
6514 ovs_list_splice(new_ofgroup->buckets.next, &new_first->list_node,
6515 &new_ofgroup->buckets);
6516 }
6517 } else if (command_bucket_id <= OFPG15_BUCKET_MAX && last) {
6518 struct ofputil_bucket *after;
6519
6520 /* Presence of bucket is checked above so after should never be NULL */
6521 after = ofputil_bucket_find(&new_ofgroup->buckets, command_bucket_id);
6522
6523 ovs_list_splice(after->list_node.next, new_ofgroup->buckets.next,
6524 last->list_node.next);
6525 }
6526
6527 return 0;
6528 }
6529
6530 /* Appends all of the a copy of all the buckets from 'ofgroup' to 'new_ofgroup'
6531 * with the exception of the bucket whose bucket id is 'command_bucket_id'.
6532 * Special 'command_bucket_id' values OFPG15_BUCKET_FIRST, OFPG15_BUCKET_LAST
6533 * and OFPG15_BUCKET_ALL are also honored. */
6534 static enum ofperr
6535 copy_buckets_for_remove_bucket(const struct ofgroup *ofgroup,
6536 struct ofgroup *new_ofgroup,
6537 uint32_t command_bucket_id)
6538 {
6539 const struct ofputil_bucket *skip = NULL;
6540
6541 if (command_bucket_id == OFPG15_BUCKET_ALL) {
6542 return 0;
6543 }
6544
6545 if (command_bucket_id == OFPG15_BUCKET_FIRST) {
6546 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6547 skip = ofputil_bucket_list_front(&ofgroup->buckets);
6548 }
6549 } else if (command_bucket_id == OFPG15_BUCKET_LAST) {
6550 if (!ovs_list_is_empty(&ofgroup->buckets)) {
6551 skip = ofputil_bucket_list_back(&ofgroup->buckets);
6552 }
6553 } else {
6554 skip = ofputil_bucket_find(&ofgroup->buckets, command_bucket_id);
6555 if (!skip) {
6556 return OFPERR_OFPGMFC_UNKNOWN_BUCKET;
6557 }
6558 }
6559
6560 ofputil_bucket_clone_list(&new_ofgroup->buckets, &ofgroup->buckets, skip);
6561
6562 return 0;
6563 }
6564
6565 /* Implements OFPGC11_MODIFY, OFPGC15_INSERT_BUCKET and
6566 * OFPGC15_REMOVE_BUCKET. Returns 0 on success or an OpenFlow error code
6567 * on failure.
6568 *
6569 * Note that the group is re-created and then replaces the old group in
6570 * ofproto's ofgroup hash map. Thus, the group is never altered while users of
6571 * the xlate module hold a pointer to the group. */
6572 static enum ofperr
6573 modify_group(struct ofproto *ofproto, const struct ofputil_group_mod *gm)
6574 {
6575 struct ofgroup *ofgroup, *new_ofgroup, *retiring;
6576 enum ofperr error;
6577
6578 error = init_group(ofproto, gm, &new_ofgroup);
6579 if (error) {
6580 return error;
6581 }
6582
6583 retiring = new_ofgroup;
6584
6585 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6586 if (!ofproto_group_lookup__(ofproto, gm->group_id, &ofgroup)) {
6587 error = OFPERR_OFPGMFC_UNKNOWN_GROUP;
6588 goto out;
6589 }
6590
6591 /* Ofproto's group write lock is held now. */
6592 if (ofgroup->type != gm->type
6593 && ofproto->n_groups[gm->type] >= ofproto->ogf.max_groups[gm->type]) {
6594 error = OFPERR_OFPGMFC_OUT_OF_GROUPS;
6595 goto out;
6596 }
6597
6598 /* Manipulate bucket list for bucket commands */
6599 if (gm->command == OFPGC15_INSERT_BUCKET) {
6600 error = copy_buckets_for_insert_bucket(ofgroup, new_ofgroup,
6601 gm->command_bucket_id);
6602 } else if (gm->command == OFPGC15_REMOVE_BUCKET) {
6603 error = copy_buckets_for_remove_bucket(ofgroup, new_ofgroup,
6604 gm->command_bucket_id);
6605 }
6606 if (error) {
6607 goto out;
6608 }
6609
6610 /* The group creation time does not change during modification. */
6611 *CONST_CAST(long long int *, &(new_ofgroup->created)) = ofgroup->created;
6612 *CONST_CAST(long long int *, &(new_ofgroup->modified)) = time_msec();
6613
6614 error = ofproto->ofproto_class->group_modify(new_ofgroup);
6615 if (error) {
6616 goto out;
6617 }
6618
6619 retiring = ofgroup;
6620 /* Replace ofgroup in ofproto's groups hash map with new_ofgroup. */
6621 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6622 hmap_insert(&ofproto->groups, &new_ofgroup->hmap_node,
6623 hash_int(new_ofgroup->group_id, 0));
6624 if (ofgroup->type != new_ofgroup->type) {
6625 ofproto->n_groups[ofgroup->type]--;
6626 ofproto->n_groups[new_ofgroup->type]++;
6627 }
6628
6629 out:
6630 ofproto_group_unref(retiring);
6631 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6632 return error;
6633 }
6634
6635 static void
6636 delete_group__(struct ofproto *ofproto, struct ofgroup *ofgroup)
6637 OVS_RELEASES(ofproto->groups_rwlock)
6638 {
6639 struct match match;
6640 struct ofproto_flow_mod ofm;
6641
6642 /* Delete all flow entries containing this group in a group action */
6643 match_init_catchall(&match);
6644 flow_mod_init(&ofm.fm, &match, 0, NULL, 0, OFPFC_DELETE);
6645 ofm.fm.delete_reason = OFPRR_GROUP_DELETE;
6646 ofm.fm.out_group = ofgroup->group_id;
6647 ofm.fm.table_id = OFPTT_ALL;
6648 handle_flow_mod__(ofproto, &ofm, NULL);
6649
6650 hmap_remove(&ofproto->groups, &ofgroup->hmap_node);
6651 /* No-one can find this group any more. */
6652 ofproto->n_groups[ofgroup->type]--;
6653 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6654 ofproto_group_unref(ofgroup);
6655 }
6656
6657 /* Implements OFPGC11_DELETE. */
6658 static void
6659 delete_group(struct ofproto *ofproto, uint32_t group_id)
6660 {
6661 struct ofgroup *ofgroup;
6662
6663 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6664 if (group_id == OFPG_ALL) {
6665 for (;;) {
6666 struct hmap_node *node = hmap_first(&ofproto->groups);
6667 if (!node) {
6668 break;
6669 }
6670 ofgroup = CONTAINER_OF(node, struct ofgroup, hmap_node);
6671 delete_group__(ofproto, ofgroup);
6672 /* Lock for each node separately, so that we will not jam the
6673 * other threads for too long time. */
6674 ovs_rwlock_wrlock(&ofproto->groups_rwlock);
6675 }
6676 } else {
6677 HMAP_FOR_EACH_IN_BUCKET (ofgroup, hmap_node,
6678 hash_int(group_id, 0), &ofproto->groups) {
6679 if (ofgroup->group_id == group_id) {
6680 delete_group__(ofproto, ofgroup);
6681 return;
6682 }
6683 }
6684 }
6685 ovs_rwlock_unlock(&ofproto->groups_rwlock);
6686 }
6687
6688 /* Delete all groups from 'ofproto'.
6689 *
6690 * This is intended for use within an ofproto provider's 'destruct'
6691 * function. */
6692 void
6693 ofproto_group_delete_all(struct ofproto *ofproto)
6694 {
6695 delete_group(ofproto, OFPG_ALL);
6696 }
6697
6698 static enum ofperr
6699 handle_group_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6700 {
6701 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6702 struct ofputil_group_mod gm;
6703 enum ofperr error;
6704
6705 error = reject_slave_controller(ofconn);
6706 if (error) {
6707 return error;
6708 }
6709
6710 error = ofputil_decode_group_mod(oh, &gm);
6711 if (error) {
6712 return error;
6713 }
6714
6715 switch (gm.command) {
6716 case OFPGC11_ADD:
6717 error = add_group(ofproto, &gm);
6718 break;
6719
6720 case OFPGC11_MODIFY:
6721 error = modify_group(ofproto, &gm);
6722 break;
6723
6724 case OFPGC11_DELETE:
6725 delete_group(ofproto, gm.group_id);
6726 error = 0;
6727 break;
6728
6729 case OFPGC15_INSERT_BUCKET:
6730 error = modify_group(ofproto, &gm);
6731 break;
6732
6733 case OFPGC15_REMOVE_BUCKET:
6734 error = modify_group(ofproto, &gm);
6735 break;
6736
6737 default:
6738 if (gm.command > OFPGC11_DELETE) {
6739 VLOG_INFO_RL(&rl, "%s: Invalid group_mod command type %d",
6740 ofproto->name, gm.command);
6741 }
6742 error = OFPERR_OFPGMFC_BAD_COMMAND;
6743 }
6744
6745 if (!error) {
6746 struct ofputil_requestforward rf;
6747 rf.xid = oh->xid;
6748 rf.reason = OFPRFR_GROUP_MOD;
6749 rf.group_mod = &gm;
6750 connmgr_send_requestforward(ofproto->connmgr, ofconn, &rf);
6751 }
6752 ofputil_bucket_list_destroy(&gm.buckets);
6753
6754 return error;
6755 }
6756
6757 enum ofputil_table_miss
6758 ofproto_table_get_miss_config(const struct ofproto *ofproto, uint8_t table_id)
6759 {
6760 enum ofputil_table_miss miss;
6761
6762 atomic_read_relaxed(&ofproto->tables[table_id].miss_config, &miss);
6763 return miss;
6764 }
6765
6766 static void
6767 table_mod__(struct oftable *oftable,
6768 const struct ofputil_table_mod *tm)
6769 {
6770 if (tm->miss == OFPUTIL_TABLE_MISS_DEFAULT) {
6771 /* This is how an OFPT_TABLE_MOD decodes if it doesn't specify any
6772 * table-miss configuration (because the protocol used doesn't have
6773 * such a concept), so there's nothing to do. */
6774 } else {
6775 atomic_store_relaxed(&oftable->miss_config, tm->miss);
6776 }
6777
6778 unsigned int new_eviction = oftable->eviction;
6779 if (tm->eviction == OFPUTIL_TABLE_EVICTION_ON) {
6780 new_eviction |= EVICTION_OPENFLOW;
6781 } else if (tm->eviction == OFPUTIL_TABLE_EVICTION_OFF) {
6782 new_eviction &= ~EVICTION_OPENFLOW;
6783 }
6784
6785 if (new_eviction != oftable->eviction) {
6786 ovs_mutex_lock(&ofproto_mutex);
6787 oftable_configure_eviction(oftable, new_eviction,
6788 oftable->eviction_fields,
6789 oftable->n_eviction_fields);
6790 ovs_mutex_unlock(&ofproto_mutex);
6791 }
6792
6793 if (tm->vacancy != OFPUTIL_TABLE_VACANCY_DEFAULT) {
6794 ovs_mutex_lock(&ofproto_mutex);
6795 oftable->vacancy_down = tm->table_vacancy.vacancy_down;
6796 oftable->vacancy_up = tm->table_vacancy.vacancy_up;
6797 if (tm->vacancy == OFPUTIL_TABLE_VACANCY_OFF) {
6798 oftable->vacancy_event = 0;
6799 } else if (!oftable->vacancy_event) {
6800 uint8_t vacancy = oftable_vacancy(oftable);
6801 oftable->vacancy_event = (vacancy < oftable->vacancy_up
6802 ? OFPTR_VACANCY_UP
6803 : OFPTR_VACANCY_DOWN);
6804 }
6805 ovs_mutex_unlock(&ofproto_mutex);
6806 }
6807 }
6808
6809 static enum ofperr
6810 table_mod(struct ofproto *ofproto, const struct ofputil_table_mod *tm)
6811 {
6812 if (!check_table_id(ofproto, tm->table_id)) {
6813 return OFPERR_OFPTMFC_BAD_TABLE;
6814 }
6815
6816 /* Don't allow the eviction flags to be changed (except to the only fixed
6817 * value that OVS supports). OF1.4 says this is normal: "The
6818 * OFPTMPT_EVICTION property usually cannot be modified using a
6819 * OFP_TABLE_MOD request, because the eviction mechanism is switch
6820 * defined". */
6821 if (tm->eviction_flags != UINT32_MAX
6822 && tm->eviction_flags != OFPROTO_EVICTION_FLAGS) {
6823 return OFPERR_OFPTMFC_BAD_CONFIG;
6824 }
6825
6826 if (tm->table_id == OFPTT_ALL) {
6827 struct oftable *oftable;
6828 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
6829 if (!(oftable->flags & (OFTABLE_HIDDEN | OFTABLE_READONLY))) {
6830 table_mod__(oftable, tm);
6831 }
6832 }
6833 } else {
6834 struct oftable *oftable = &ofproto->tables[tm->table_id];
6835 if (oftable->flags & OFTABLE_READONLY) {
6836 return OFPERR_OFPTMFC_EPERM;
6837 }
6838 table_mod__(oftable, tm);
6839 }
6840
6841 return 0;
6842 }
6843
6844 static enum ofperr
6845 handle_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
6846 {
6847 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6848 struct ofputil_table_mod tm;
6849 enum ofperr error;
6850
6851 error = reject_slave_controller(ofconn);
6852 if (error) {
6853 return error;
6854 }
6855
6856 error = ofputil_decode_table_mod(oh, &tm);
6857 if (error) {
6858 return error;
6859 }
6860
6861 return table_mod(ofproto, &tm);
6862 }
6863
6864 static enum ofperr
6865 ofproto_flow_mod_start(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6866 OVS_REQUIRES(ofproto_mutex)
6867 {
6868 switch (ofm->fm.command) {
6869 case OFPFC_ADD:
6870 return add_flow_start(ofproto, ofm);
6871 /* , &be->old_rules.stub[0],
6872 &be->new_rules.stub[0]); */
6873 case OFPFC_MODIFY:
6874 return modify_flows_start_loose(ofproto, ofm);
6875 case OFPFC_MODIFY_STRICT:
6876 return modify_flow_start_strict(ofproto, ofm);
6877 case OFPFC_DELETE:
6878 return delete_flows_start_loose(ofproto, ofm);
6879
6880 case OFPFC_DELETE_STRICT:
6881 return delete_flow_start_strict(ofproto, ofm);
6882 }
6883
6884 return OFPERR_OFPFMFC_BAD_COMMAND;
6885 }
6886
6887 static void
6888 ofproto_flow_mod_revert(struct ofproto *ofproto, struct ofproto_flow_mod *ofm)
6889 OVS_REQUIRES(ofproto_mutex)
6890 {
6891 switch (ofm->fm.command) {
6892 case OFPFC_ADD:
6893 add_flow_revert(ofproto, ofm);
6894 break;
6895
6896 case OFPFC_MODIFY:
6897 case OFPFC_MODIFY_STRICT:
6898 modify_flows_revert(ofproto, ofm);
6899 break;
6900
6901 case OFPFC_DELETE:
6902 case OFPFC_DELETE_STRICT:
6903 delete_flows_revert(ofproto, ofm);
6904 break;
6905
6906 default:
6907 break;
6908 }
6909 }
6910
6911 static void
6912 ofproto_flow_mod_finish(struct ofproto *ofproto,
6913 struct ofproto_flow_mod *ofm,
6914 const struct flow_mod_requester *req)
6915 OVS_REQUIRES(ofproto_mutex)
6916 {
6917 switch (ofm->fm.command) {
6918 case OFPFC_ADD:
6919 add_flow_finish(ofproto, ofm, req);
6920 break;
6921
6922 case OFPFC_MODIFY:
6923 case OFPFC_MODIFY_STRICT:
6924 modify_flows_finish(ofproto, ofm, req);
6925 break;
6926
6927 case OFPFC_DELETE:
6928 case OFPFC_DELETE_STRICT:
6929 delete_flows_finish(ofproto, ofm, req);
6930 break;
6931
6932 default:
6933 break;
6934 }
6935 }
6936
6937 /* Commit phases (all while locking ofproto_mutex):
6938 *
6939 * 1. Begin: Gather resources and make changes visible in the next version.
6940 * - Mark affected rules for removal in the next version.
6941 * - Create new replacement rules, make visible in the next
6942 * version.
6943 * - Do not send any events or notifications.
6944 *
6945 * 2. Revert: Fail if any errors are found. After this point no errors are
6946 * possible. No visible changes were made, so rollback is minimal (remove
6947 * added invisible rules, restore visibility of rules marked for removal).
6948 *
6949 * 3. Finish: Make the changes visible for lookups. Insert replacement rules to
6950 * the ofproto provider. Remove replaced and deleted rules from ofproto data
6951 * structures, and Schedule postponed removal of deleted rules from the
6952 * classifier. Send notifications, buffered packets, etc.
6953 */
6954 static enum ofperr
6955 do_bundle_commit(struct ofconn *ofconn, uint32_t id, uint16_t flags)
6956 {
6957 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
6958 cls_version_t version = ofproto->tables_version + 1;
6959 struct ofp_bundle *bundle;
6960 struct ofp_bundle_entry *be;
6961 enum ofperr error;
6962
6963 bundle = ofconn_get_bundle(ofconn, id);
6964
6965 if (!bundle) {
6966 return OFPERR_OFPBFC_BAD_ID;
6967 }
6968 if (bundle->flags != flags) {
6969 error = OFPERR_OFPBFC_BAD_FLAGS;
6970 } else {
6971 bool prev_is_port_mod = false;
6972
6973 error = 0;
6974 ovs_mutex_lock(&ofproto_mutex);
6975
6976 /* 1. Begin. */
6977 LIST_FOR_EACH (be, node, &bundle->msg_list) {
6978 if (be->type == OFPTYPE_PORT_MOD) {
6979 /* Our port mods are not atomic. */
6980 if (flags & OFPBF_ATOMIC) {
6981 error = OFPERR_OFPBFC_MSG_FAILED;
6982 } else {
6983 prev_is_port_mod = true;
6984 error = port_mod_start(ofconn, &be->opm.pm, &be->opm.port);
6985 }
6986 } else if (be->type == OFPTYPE_FLOW_MOD) {
6987 /* Flow mods between port mods are applied as a single
6988 * version, but the versions are published only after
6989 * we know the commit is successful. */
6990 if (prev_is_port_mod) {
6991 ++version;
6992 }
6993 prev_is_port_mod = false;
6994 /* Store the version in which the changes should take
6995 * effect. */
6996 be->ofm.version = version;
6997 error = ofproto_flow_mod_start(ofproto, &be->ofm);
6998 } else {
6999 OVS_NOT_REACHED();
7000 }
7001 if (error) {
7002 break;
7003 }
7004 }
7005
7006 if (error) {
7007 /* Send error referring to the original message. */
7008 if (error) {
7009 ofconn_send_error(ofconn, be->ofp_msg, error);
7010 error = OFPERR_OFPBFC_MSG_FAILED;
7011 }
7012
7013 /* 2. Revert. Undo all the changes made above. */
7014 LIST_FOR_EACH_REVERSE_CONTINUE(be, node, &bundle->msg_list) {
7015 if (be->type == OFPTYPE_FLOW_MOD) {
7016 ofproto_flow_mod_revert(ofproto, &be->ofm);
7017 }
7018 /* Nothing needs to be reverted for a port mod. */
7019 }
7020 } else {
7021 /* 4. Finish. */
7022 LIST_FOR_EACH (be, node, &bundle->msg_list) {
7023 if (be->type == OFPTYPE_FLOW_MOD) {
7024 struct flow_mod_requester req = { ofconn, be->ofp_msg };
7025
7026 /* Bump the lookup version to the one of the current
7027 * message. This makes all the changes in the bundle at
7028 * this version visible to lookups at once. */
7029 if (ofproto->tables_version < be->ofm.version) {
7030 ofproto->tables_version = be->ofm.version;
7031 ofproto->ofproto_class->set_tables_version(
7032 ofproto, ofproto->tables_version);
7033 }
7034
7035 ofproto_flow_mod_finish(ofproto, &be->ofm, &req);
7036 } else if (be->type == OFPTYPE_PORT_MOD) {
7037 /* Perform the actual port mod. This is not atomic, i.e.,
7038 * the effects will be immediately seen by upcall
7039 * processing regardless of the lookup version. It should
7040 * be noted that port configuration changes can originate
7041 * also from OVSDB changes asynchronously to all upcall
7042 * processing. */
7043 port_mod_finish(ofconn, &be->opm.pm, be->opm.port);
7044 }
7045 }
7046 }
7047
7048 ofmonitor_flush(ofproto->connmgr);
7049 ovs_mutex_unlock(&ofproto_mutex);
7050
7051 run_rule_executes(ofproto);
7052 }
7053
7054 /* The bundle is discarded regardless the outcome. */
7055 ofp_bundle_remove__(ofconn, bundle, !error);
7056 return error;
7057 }
7058
7059 static enum ofperr
7060 handle_bundle_control(struct ofconn *ofconn, const struct ofp_header *oh)
7061 {
7062 struct ofputil_bundle_ctrl_msg bctrl;
7063 struct ofputil_bundle_ctrl_msg reply;
7064 struct ofpbuf *buf;
7065 enum ofperr error;
7066
7067 error = reject_slave_controller(ofconn);
7068 if (error) {
7069 return error;
7070 }
7071
7072 error = ofputil_decode_bundle_ctrl(oh, &bctrl);
7073 if (error) {
7074 return error;
7075 }
7076 reply.flags = 0;
7077 reply.bundle_id = bctrl.bundle_id;
7078
7079 switch (bctrl.type) {
7080 case OFPBCT_OPEN_REQUEST:
7081 error = ofp_bundle_open(ofconn, bctrl.bundle_id, bctrl.flags);
7082 reply.type = OFPBCT_OPEN_REPLY;
7083 break;
7084 case OFPBCT_CLOSE_REQUEST:
7085 error = ofp_bundle_close(ofconn, bctrl.bundle_id, bctrl.flags);
7086 reply.type = OFPBCT_CLOSE_REPLY;
7087 break;
7088 case OFPBCT_COMMIT_REQUEST:
7089 error = do_bundle_commit(ofconn, bctrl.bundle_id, bctrl.flags);
7090 reply.type = OFPBCT_COMMIT_REPLY;
7091 break;
7092 case OFPBCT_DISCARD_REQUEST:
7093 error = ofp_bundle_discard(ofconn, bctrl.bundle_id);
7094 reply.type = OFPBCT_DISCARD_REPLY;
7095 break;
7096
7097 case OFPBCT_OPEN_REPLY:
7098 case OFPBCT_CLOSE_REPLY:
7099 case OFPBCT_COMMIT_REPLY:
7100 case OFPBCT_DISCARD_REPLY:
7101 return OFPERR_OFPBFC_BAD_TYPE;
7102 break;
7103 }
7104
7105 if (!error) {
7106 buf = ofputil_encode_bundle_ctrl_reply(oh, &reply);
7107 ofconn_send_reply(ofconn, buf);
7108 }
7109 return error;
7110 }
7111
7112 static enum ofperr
7113 handle_bundle_add(struct ofconn *ofconn, const struct ofp_header *oh)
7114 {
7115 struct ofproto *ofproto = ofconn_get_ofproto(ofconn);
7116 enum ofperr error;
7117 struct ofputil_bundle_add_msg badd;
7118 struct ofp_bundle_entry *bmsg;
7119 enum ofptype type;
7120
7121 error = reject_slave_controller(ofconn);
7122 if (error) {
7123 return error;
7124 }
7125
7126 error = ofputil_decode_bundle_add(oh, &badd, &type);
7127 if (error) {
7128 return error;
7129 }
7130
7131 bmsg = ofp_bundle_entry_alloc(type, badd.msg);
7132
7133 if (type == OFPTYPE_PORT_MOD) {
7134 error = ofputil_decode_port_mod(badd.msg, &bmsg->opm.pm, false);
7135 } else if (type == OFPTYPE_FLOW_MOD) {
7136 struct ofpbuf ofpacts;
7137 uint64_t ofpacts_stub[1024 / 8];
7138
7139 ofpbuf_use_stub(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
7140 error = ofputil_decode_flow_mod(&bmsg->ofm.fm, badd.msg,
7141 ofconn_get_protocol(ofconn),
7142 &ofpacts,
7143 u16_to_ofp(ofproto->max_ports),
7144 ofproto->n_tables);
7145 /* Move actions to heap. */
7146 bmsg->ofm.fm.ofpacts = ofpbuf_steal_data(&ofpacts);
7147
7148 if (!error && bmsg->ofm.fm.ofpacts_len) {
7149 error = ofproto_check_ofpacts(ofproto, bmsg->ofm.fm.ofpacts,
7150 bmsg->ofm.fm.ofpacts_len);
7151 }
7152 } else {
7153 OVS_NOT_REACHED();
7154 }
7155
7156 if (!error) {
7157 error = ofp_bundle_add_message(ofconn, badd.bundle_id, badd.flags,
7158 bmsg);
7159 }
7160
7161 if (error) {
7162 ofp_bundle_entry_free(bmsg);
7163 }
7164
7165 return error;
7166 }
7167
7168 static enum ofperr
7169 handle_tlv_table_mod(struct ofconn *ofconn, const struct ofp_header *oh)
7170 {
7171 struct ofputil_tlv_table_mod ttm;
7172 enum ofperr error;
7173
7174 error = reject_slave_controller(ofconn);
7175 if (error) {
7176 return error;
7177 }
7178
7179 error = ofputil_decode_tlv_table_mod(oh, &ttm);
7180 if (error) {
7181 return error;
7182 }
7183
7184 error = tun_metadata_table_mod(&ttm);
7185
7186 ofputil_uninit_tlv_table(&ttm.mappings);
7187 return error;
7188 }
7189
7190 static enum ofperr
7191 handle_tlv_table_request(struct ofconn *ofconn, const struct ofp_header *oh)
7192 {
7193 struct ofputil_tlv_table_reply ttr;
7194 struct ofpbuf *b;
7195
7196 tun_metadata_table_request(&ttr);
7197 b = ofputil_encode_tlv_table_reply(oh, &ttr);
7198 ofputil_uninit_tlv_table(&ttr.mappings);
7199
7200 ofconn_send_reply(ofconn, b);
7201 return 0;
7202 }
7203
7204 static enum ofperr
7205 handle_openflow__(struct ofconn *ofconn, const struct ofpbuf *msg)
7206 OVS_EXCLUDED(ofproto_mutex)
7207 {
7208 const struct ofp_header *oh = msg->data;
7209 enum ofptype type;
7210 enum ofperr error;
7211
7212 error = ofptype_decode(&type, oh);
7213 if (error) {
7214 return error;
7215 }
7216 if (oh->version >= OFP13_VERSION && ofpmsg_is_stat_request(oh)
7217 && ofpmp_more(oh)) {
7218 /* We have no buffer implementation for multipart requests.
7219 * Report overflow for requests which consists of multiple
7220 * messages. */
7221 return OFPERR_OFPBRC_MULTIPART_BUFFER_OVERFLOW;
7222 }
7223
7224 switch (type) {
7225 /* OpenFlow requests. */
7226 case OFPTYPE_ECHO_REQUEST:
7227 return handle_echo_request(ofconn, oh);
7228
7229 case OFPTYPE_FEATURES_REQUEST:
7230 return handle_features_request(ofconn, oh);
7231
7232 case OFPTYPE_GET_CONFIG_REQUEST:
7233 return handle_get_config_request(ofconn, oh);
7234
7235 case OFPTYPE_SET_CONFIG:
7236 return handle_set_config(ofconn, oh);
7237
7238 case OFPTYPE_PACKET_OUT:
7239 return handle_packet_out(ofconn, oh);
7240
7241 case OFPTYPE_PORT_MOD:
7242 return handle_port_mod(ofconn, oh);
7243
7244 case OFPTYPE_FLOW_MOD:
7245 return handle_flow_mod(ofconn, oh);
7246
7247 case OFPTYPE_GROUP_MOD:
7248 return handle_group_mod(ofconn, oh);
7249
7250 case OFPTYPE_TABLE_MOD:
7251 return handle_table_mod(ofconn, oh);
7252
7253 case OFPTYPE_METER_MOD:
7254 return handle_meter_mod(ofconn, oh);
7255
7256 case OFPTYPE_BARRIER_REQUEST:
7257 return handle_barrier_request(ofconn, oh);
7258
7259 case OFPTYPE_ROLE_REQUEST:
7260 return handle_role_request(ofconn, oh);
7261
7262 /* OpenFlow replies. */
7263 case OFPTYPE_ECHO_REPLY:
7264 return 0;
7265
7266 /* Nicira extension requests. */
7267 case OFPTYPE_FLOW_MOD_TABLE_ID:
7268 return handle_nxt_flow_mod_table_id(ofconn, oh);
7269
7270 case OFPTYPE_SET_FLOW_FORMAT:
7271 return handle_nxt_set_flow_format(ofconn, oh);
7272
7273 case OFPTYPE_SET_PACKET_IN_FORMAT:
7274 return handle_nxt_set_packet_in_format(ofconn, oh);
7275
7276 case OFPTYPE_SET_CONTROLLER_ID:
7277 return handle_nxt_set_controller_id(ofconn, oh);
7278
7279 case OFPTYPE_FLOW_AGE:
7280 /* Nothing to do. */
7281 return 0;
7282
7283 case OFPTYPE_FLOW_MONITOR_CANCEL:
7284 return handle_flow_monitor_cancel(ofconn, oh);
7285
7286 case OFPTYPE_SET_ASYNC_CONFIG:
7287 return handle_nxt_set_async_config(ofconn, oh);
7288
7289 case OFPTYPE_GET_ASYNC_REQUEST:
7290 return handle_nxt_get_async_request(ofconn, oh);
7291
7292 case OFPTYPE_NXT_RESUME:
7293 return handle_nxt_resume(ofconn, oh);
7294
7295 /* Statistics requests. */
7296 case OFPTYPE_DESC_STATS_REQUEST:
7297 return handle_desc_stats_request(ofconn, oh);
7298
7299 case OFPTYPE_FLOW_STATS_REQUEST:
7300 return handle_flow_stats_request(ofconn, oh);
7301
7302 case OFPTYPE_AGGREGATE_STATS_REQUEST:
7303 return handle_aggregate_stats_request(ofconn, oh);
7304
7305 case OFPTYPE_TABLE_STATS_REQUEST:
7306 return handle_table_stats_request(ofconn, oh);
7307
7308 case OFPTYPE_TABLE_FEATURES_STATS_REQUEST:
7309 return handle_table_features_request(ofconn, oh);
7310
7311 case OFPTYPE_TABLE_DESC_REQUEST:
7312 return handle_table_desc_request(ofconn, oh);
7313
7314 case OFPTYPE_PORT_STATS_REQUEST:
7315 return handle_port_stats_request(ofconn, oh);
7316
7317 case OFPTYPE_QUEUE_STATS_REQUEST:
7318 return handle_queue_stats_request(ofconn, oh);
7319
7320 case OFPTYPE_PORT_DESC_STATS_REQUEST:
7321 return handle_port_desc_stats_request(ofconn, oh);
7322
7323 case OFPTYPE_FLOW_MONITOR_STATS_REQUEST:
7324 return handle_flow_monitor_request(ofconn, oh);
7325
7326 case OFPTYPE_METER_STATS_REQUEST:
7327 case OFPTYPE_METER_CONFIG_STATS_REQUEST:
7328 return handle_meter_request(ofconn, oh, type);
7329
7330 case OFPTYPE_METER_FEATURES_STATS_REQUEST:
7331 return handle_meter_features_request(ofconn, oh);
7332
7333 case OFPTYPE_GROUP_STATS_REQUEST:
7334 return handle_group_stats_request(ofconn, oh);
7335
7336 case OFPTYPE_GROUP_DESC_STATS_REQUEST:
7337 return handle_group_desc_stats_request(ofconn, oh);
7338
7339 case OFPTYPE_GROUP_FEATURES_STATS_REQUEST:
7340 return handle_group_features_stats_request(ofconn, oh);
7341
7342 case OFPTYPE_QUEUE_GET_CONFIG_REQUEST:
7343 return handle_queue_get_config_request(ofconn, oh);
7344
7345 case OFPTYPE_BUNDLE_CONTROL:
7346 return handle_bundle_control(ofconn, oh);
7347
7348 case OFPTYPE_BUNDLE_ADD_MESSAGE:
7349 return handle_bundle_add(ofconn, oh);
7350
7351 case OFPTYPE_NXT_TLV_TABLE_MOD:
7352 return handle_tlv_table_mod(ofconn, oh);
7353
7354 case OFPTYPE_NXT_TLV_TABLE_REQUEST:
7355 return handle_tlv_table_request(ofconn, oh);
7356
7357 case OFPTYPE_HELLO:
7358 case OFPTYPE_ERROR:
7359 case OFPTYPE_FEATURES_REPLY:
7360 case OFPTYPE_GET_CONFIG_REPLY:
7361 case OFPTYPE_PACKET_IN:
7362 case OFPTYPE_FLOW_REMOVED:
7363 case OFPTYPE_PORT_STATUS:
7364 case OFPTYPE_BARRIER_REPLY:
7365 case OFPTYPE_QUEUE_GET_CONFIG_REPLY:
7366 case OFPTYPE_DESC_STATS_REPLY:
7367 case OFPTYPE_FLOW_STATS_REPLY:
7368 case OFPTYPE_QUEUE_STATS_REPLY:
7369 case OFPTYPE_PORT_STATS_REPLY:
7370 case OFPTYPE_TABLE_STATS_REPLY:
7371 case OFPTYPE_AGGREGATE_STATS_REPLY:
7372 case OFPTYPE_PORT_DESC_STATS_REPLY:
7373 case OFPTYPE_ROLE_REPLY:
7374 case OFPTYPE_FLOW_MONITOR_PAUSED:
7375 case OFPTYPE_FLOW_MONITOR_RESUMED:
7376 case OFPTYPE_FLOW_MONITOR_STATS_REPLY:
7377 case OFPTYPE_GET_ASYNC_REPLY:
7378 case OFPTYPE_GROUP_STATS_REPLY:
7379 case OFPTYPE_GROUP_DESC_STATS_REPLY:
7380 case OFPTYPE_GROUP_FEATURES_STATS_REPLY:
7381 case OFPTYPE_METER_STATS_REPLY:
7382 case OFPTYPE_METER_CONFIG_STATS_REPLY:
7383 case OFPTYPE_METER_FEATURES_STATS_REPLY:
7384 case OFPTYPE_TABLE_FEATURES_STATS_REPLY:
7385 case OFPTYPE_TABLE_DESC_REPLY:
7386 case OFPTYPE_ROLE_STATUS:
7387 case OFPTYPE_REQUESTFORWARD:
7388 case OFPTYPE_TABLE_STATUS:
7389 case OFPTYPE_NXT_TLV_TABLE_REPLY:
7390 default:
7391 if (ofpmsg_is_stat_request(oh)) {
7392 return OFPERR_OFPBRC_BAD_STAT;
7393 } else {
7394 return OFPERR_OFPBRC_BAD_TYPE;
7395 }
7396 }
7397 }
7398
7399 static void
7400 handle_openflow(struct ofconn *ofconn, const struct ofpbuf *ofp_msg)
7401 OVS_EXCLUDED(ofproto_mutex)
7402 {
7403 enum ofperr error = handle_openflow__(ofconn, ofp_msg);
7404
7405 if (error) {
7406 ofconn_send_error(ofconn, ofp_msg->data, error);
7407 }
7408 COVERAGE_INC(ofproto_recv_openflow);
7409 }
7410 \f
7411 /* Asynchronous operations. */
7412
7413 static void
7414 send_buffered_packet(const struct flow_mod_requester *req, uint32_t buffer_id,
7415 struct rule *rule)
7416 OVS_REQUIRES(ofproto_mutex)
7417 {
7418 if (req && req->ofconn && buffer_id != UINT32_MAX) {
7419 struct ofproto *ofproto = ofconn_get_ofproto(req->ofconn);
7420 struct dp_packet *packet;
7421 ofp_port_t in_port;
7422 enum ofperr error;
7423
7424 error = ofconn_pktbuf_retrieve(req->ofconn, buffer_id, &packet,
7425 &in_port);
7426 if (packet) {
7427 struct rule_execute *re;
7428
7429 ofproto_rule_ref(rule);
7430
7431 re = xmalloc(sizeof *re);
7432 re->rule = rule;
7433 re->in_port = in_port;
7434 re->packet = packet;
7435
7436 if (!guarded_list_push_back(&ofproto->rule_executes,
7437 &re->list_node, 1024)) {
7438 ofproto_rule_unref(rule);
7439 dp_packet_delete(re->packet);
7440 free(re);
7441 }
7442 } else {
7443 ofconn_send_error(req->ofconn, req->request, error);
7444 }
7445 }
7446 }
7447 \f
7448 static uint64_t
7449 pick_datapath_id(const struct ofproto *ofproto)
7450 {
7451 const struct ofport *port;
7452
7453 port = ofproto_get_port(ofproto, OFPP_LOCAL);
7454 if (port) {
7455 struct eth_addr ea;
7456 int error;
7457
7458 error = netdev_get_etheraddr(port->netdev, &ea);
7459 if (!error) {
7460 return eth_addr_to_uint64(ea);
7461 }
7462 VLOG_WARN("%s: could not get MAC address for %s (%s)",
7463 ofproto->name, netdev_get_name(port->netdev),
7464 ovs_strerror(error));
7465 }
7466 return ofproto->fallback_dpid;
7467 }
7468
7469 static uint64_t
7470 pick_fallback_dpid(void)
7471 {
7472 struct eth_addr ea;
7473 eth_addr_nicira_random(&ea);
7474 return eth_addr_to_uint64(ea);
7475 }
7476 \f
7477 /* Table overflow policy. */
7478
7479 /* Chooses and updates 'rulep' with a rule to evict from 'table'. Sets 'rulep'
7480 * to NULL if the table is not configured to evict rules or if the table
7481 * contains no evictable rules. (Rules with a readlock on their evict rwlock,
7482 * or with no timeouts are not evictable.) */
7483 static bool
7484 choose_rule_to_evict(struct oftable *table, struct rule **rulep)
7485 OVS_REQUIRES(ofproto_mutex)
7486 {
7487 struct eviction_group *evg;
7488
7489 *rulep = NULL;
7490 if (!table->eviction) {
7491 return false;
7492 }
7493
7494 /* In the common case, the outer and inner loops here will each be entered
7495 * exactly once:
7496 *
7497 * - The inner loop normally "return"s in its first iteration. If the
7498 * eviction group has any evictable rules, then it always returns in
7499 * some iteration.
7500 *
7501 * - The outer loop only iterates more than once if the largest eviction
7502 * group has no evictable rules.
7503 *
7504 * - The outer loop can exit only if table's 'max_flows' is all filled up
7505 * by unevictable rules. */
7506 HEAP_FOR_EACH (evg, size_node, &table->eviction_groups_by_size) {
7507 struct rule *rule;
7508
7509 HEAP_FOR_EACH (rule, evg_node, &evg->rules) {
7510 *rulep = rule;
7511 return true;
7512 }
7513 }
7514
7515 return false;
7516 }
7517 \f
7518 /* Eviction groups. */
7519
7520 /* Returns the priority to use for an eviction_group that contains 'n_rules'
7521 * rules. The priority contains low-order random bits to ensure that eviction
7522 * groups with the same number of rules are prioritized randomly. */
7523 static uint32_t
7524 eviction_group_priority(size_t n_rules)
7525 {
7526 uint16_t size = MIN(UINT16_MAX, n_rules);
7527 return (size << 16) | random_uint16();
7528 }
7529
7530 /* Updates 'evg', an eviction_group within 'table', following a change that
7531 * adds or removes rules in 'evg'. */
7532 static void
7533 eviction_group_resized(struct oftable *table, struct eviction_group *evg)
7534 OVS_REQUIRES(ofproto_mutex)
7535 {
7536 heap_change(&table->eviction_groups_by_size, &evg->size_node,
7537 eviction_group_priority(heap_count(&evg->rules)));
7538 }
7539
7540 /* Destroys 'evg', an eviction_group within 'table':
7541 *
7542 * - Removes all the rules, if any, from 'evg'. (It doesn't destroy the
7543 * rules themselves, just removes them from the eviction group.)
7544 *
7545 * - Removes 'evg' from 'table'.
7546 *
7547 * - Frees 'evg'. */
7548 static void
7549 eviction_group_destroy(struct oftable *table, struct eviction_group *evg)
7550 OVS_REQUIRES(ofproto_mutex)
7551 {
7552 while (!heap_is_empty(&evg->rules)) {
7553 struct rule *rule;
7554
7555 rule = CONTAINER_OF(heap_pop(&evg->rules), struct rule, evg_node);
7556 rule->eviction_group = NULL;
7557 }
7558 hmap_remove(&table->eviction_groups_by_id, &evg->id_node);
7559 heap_remove(&table->eviction_groups_by_size, &evg->size_node);
7560 heap_destroy(&evg->rules);
7561 free(evg);
7562 }
7563
7564 /* Removes 'rule' from its eviction group, if any. */
7565 static void
7566 eviction_group_remove_rule(struct rule *rule)
7567 OVS_REQUIRES(ofproto_mutex)
7568 {
7569 if (rule->eviction_group) {
7570 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7571 struct eviction_group *evg = rule->eviction_group;
7572
7573 rule->eviction_group = NULL;
7574 heap_remove(&evg->rules, &rule->evg_node);
7575 if (heap_is_empty(&evg->rules)) {
7576 eviction_group_destroy(table, evg);
7577 } else {
7578 eviction_group_resized(table, evg);
7579 }
7580 }
7581 }
7582
7583 /* Hashes the 'rule''s values for the eviction_fields of 'rule''s table, and
7584 * returns the hash value. */
7585 static uint32_t
7586 eviction_group_hash_rule(struct rule *rule)
7587 OVS_REQUIRES(ofproto_mutex)
7588 {
7589 struct oftable *table = &rule->ofproto->tables[rule->table_id];
7590 const struct mf_subfield *sf;
7591 struct flow flow;
7592 uint32_t hash;
7593
7594 hash = table->eviction_group_id_basis;
7595 miniflow_expand(rule->cr.match.flow, &flow);
7596 for (sf = table->eviction_fields;
7597 sf < &table->eviction_fields[table->n_eviction_fields];
7598 sf++)
7599 {
7600 if (mf_are_prereqs_ok(sf->field, &flow)) {
7601 union mf_value value;
7602
7603 mf_get_value(sf->field, &flow, &value);
7604 if (sf->ofs) {
7605 bitwise_zero(&value, sf->field->n_bytes, 0, sf->ofs);
7606 }
7607 if (sf->ofs + sf->n_bits < sf->field->n_bytes * 8) {
7608 unsigned int start = sf->ofs + sf->n_bits;
7609 bitwise_zero(&value, sf->field->n_bytes, start,
7610 sf->field->n_bytes * 8 - start);
7611 }
7612 hash = hash_bytes(&value, sf->field->n_bytes, hash);
7613 } else {
7614 hash = hash_int(hash, 0);
7615 }
7616 }
7617
7618 return hash;
7619 }
7620
7621 /* Returns an eviction group within 'table' with the given 'id', creating one
7622 * if necessary. */
7623 static struct eviction_group *
7624 eviction_group_find(struct oftable *table, uint32_t id)
7625 OVS_REQUIRES(ofproto_mutex)
7626 {
7627 struct eviction_group *evg;
7628
7629 HMAP_FOR_EACH_WITH_HASH (evg, id_node, id, &table->eviction_groups_by_id) {
7630 return evg;
7631 }
7632
7633 evg = xmalloc(sizeof *evg);
7634 hmap_insert(&table->eviction_groups_by_id, &evg->id_node, id);
7635 heap_insert(&table->eviction_groups_by_size, &evg->size_node,
7636 eviction_group_priority(0));
7637 heap_init(&evg->rules);
7638
7639 return evg;
7640 }
7641
7642 /* Returns an eviction priority for 'rule'. The return value should be
7643 * interpreted so that higher priorities make a rule a more attractive
7644 * candidate for eviction. */
7645 static uint64_t
7646 rule_eviction_priority(struct ofproto *ofproto, struct rule *rule)
7647 OVS_REQUIRES(ofproto_mutex)
7648 {
7649 /* Calculate absolute time when this flow will expire. If it will never
7650 * expire, then return 0 to make it unevictable. */
7651 long long int expiration = LLONG_MAX;
7652 if (rule->hard_timeout) {
7653 /* 'modified' needs protection even when we hold 'ofproto_mutex'. */
7654 ovs_mutex_lock(&rule->mutex);
7655 long long int modified = rule->modified;
7656 ovs_mutex_unlock(&rule->mutex);
7657
7658 expiration = modified + rule->hard_timeout * 1000;
7659 }
7660 if (rule->idle_timeout) {
7661 uint64_t packets, bytes;
7662 long long int used;
7663 long long int idle_expiration;
7664
7665 ofproto->ofproto_class->rule_get_stats(rule, &packets, &bytes, &used);
7666 idle_expiration = used + rule->idle_timeout * 1000;
7667 expiration = MIN(expiration, idle_expiration);
7668 }
7669 if (expiration == LLONG_MAX) {
7670 return 0;
7671 }
7672
7673 /* Calculate the time of expiration as a number of (approximate) seconds
7674 * after program startup.
7675 *
7676 * This should work OK for program runs that last UINT32_MAX seconds or
7677 * less. Therefore, please restart OVS at least once every 136 years. */
7678 uint32_t expiration_ofs = (expiration >> 10) - (time_boot_msec() >> 10);
7679
7680 /* Combine expiration time with OpenFlow "importance" to form a single
7681 * priority value. We want flows with relatively low "importance" to be
7682 * evicted before even considering expiration time, so put "importance" in
7683 * the most significant bits and expiration time in the least significant
7684 * bits.
7685 *
7686 * Small 'priority' should be evicted before those with large 'priority'.
7687 * The caller expects the opposite convention (a large return value being
7688 * more attractive for eviction) so we invert it before returning. */
7689 uint64_t priority = ((uint64_t) rule->importance << 32) + expiration_ofs;
7690 return UINT64_MAX - priority;
7691 }
7692
7693 /* Adds 'rule' to an appropriate eviction group for its oftable's
7694 * configuration. Does nothing if 'rule''s oftable doesn't have eviction
7695 * enabled, or if 'rule' is a permanent rule (one that will never expire on its
7696 * own).
7697 *
7698 * The caller must ensure that 'rule' is not already in an eviction group. */
7699 static void
7700 eviction_group_add_rule(struct rule *rule)
7701 OVS_REQUIRES(ofproto_mutex)
7702 {
7703 struct ofproto *ofproto = rule->ofproto;
7704 struct oftable *table = &ofproto->tables[rule->table_id];
7705 bool has_timeout;
7706
7707 /* Timeouts may be modified only when holding 'ofproto_mutex'. We have it
7708 * so no additional protection is needed. */
7709 has_timeout = rule->hard_timeout || rule->idle_timeout;
7710
7711 if (table->eviction && has_timeout) {
7712 struct eviction_group *evg;
7713
7714 evg = eviction_group_find(table, eviction_group_hash_rule(rule));
7715
7716 rule->eviction_group = evg;
7717 heap_insert(&evg->rules, &rule->evg_node,
7718 rule_eviction_priority(ofproto, rule));
7719 eviction_group_resized(table, evg);
7720 }
7721 }
7722 \f
7723 /* oftables. */
7724
7725 /* Initializes 'table'. */
7726 static void
7727 oftable_init(struct oftable *table)
7728 {
7729 memset(table, 0, sizeof *table);
7730 classifier_init(&table->cls, flow_segment_u64s);
7731 table->max_flows = UINT_MAX;
7732 table->n_flows = 0;
7733 hmap_init(&table->eviction_groups_by_id);
7734 heap_init(&table->eviction_groups_by_size);
7735 atomic_init(&table->miss_config, OFPUTIL_TABLE_MISS_DEFAULT);
7736
7737 classifier_set_prefix_fields(&table->cls, default_prefix_fields,
7738 ARRAY_SIZE(default_prefix_fields));
7739
7740 atomic_init(&table->n_matched, 0);
7741 atomic_init(&table->n_missed, 0);
7742 }
7743
7744 /* Destroys 'table', including its classifier and eviction groups.
7745 *
7746 * The caller is responsible for freeing 'table' itself. */
7747 static void
7748 oftable_destroy(struct oftable *table)
7749 {
7750 ovs_assert(classifier_is_empty(&table->cls));
7751
7752 ovs_mutex_lock(&ofproto_mutex);
7753 oftable_configure_eviction(table, 0, NULL, 0);
7754 ovs_mutex_unlock(&ofproto_mutex);
7755
7756 hmap_destroy(&table->eviction_groups_by_id);
7757 heap_destroy(&table->eviction_groups_by_size);
7758 classifier_destroy(&table->cls);
7759 free(table->name);
7760 }
7761
7762 /* Changes the name of 'table' to 'name'. If 'name' is NULL or the empty
7763 * string, then 'table' will use its default name.
7764 *
7765 * This only affects the name exposed for a table exposed through the OpenFlow
7766 * OFPST_TABLE (as printed by "ovs-ofctl dump-tables"). */
7767 static void
7768 oftable_set_name(struct oftable *table, const char *name)
7769 {
7770 if (name && name[0]) {
7771 int len = strnlen(name, OFP_MAX_TABLE_NAME_LEN);
7772 if (!table->name || strncmp(name, table->name, len)) {
7773 free(table->name);
7774 table->name = xmemdup0(name, len);
7775 }
7776 } else {
7777 free(table->name);
7778 table->name = NULL;
7779 }
7780 }
7781
7782 /* oftables support a choice of two policies when adding a rule would cause the
7783 * number of flows in the table to exceed the configured maximum number: either
7784 * they can refuse to add the new flow or they can evict some existing flow.
7785 * This function configures the latter policy on 'table', with fairness based
7786 * on the values of the 'n_fields' fields specified in 'fields'. (Specifying
7787 * 'n_fields' as 0 disables fairness.) */
7788 static void
7789 oftable_configure_eviction(struct oftable *table, unsigned int eviction,
7790 const struct mf_subfield *fields, size_t n_fields)
7791 OVS_REQUIRES(ofproto_mutex)
7792 {
7793 struct rule *rule;
7794
7795 if ((table->eviction != 0) == (eviction != 0)
7796 && n_fields == table->n_eviction_fields
7797 && (!n_fields
7798 || !memcmp(fields, table->eviction_fields,
7799 n_fields * sizeof *fields))) {
7800 /* The set of eviction fields did not change. If 'eviction' changed,
7801 * it remains nonzero, so that we can just update table->eviction
7802 * without fussing with the eviction groups. */
7803 table->eviction = eviction;
7804 return;
7805 }
7806
7807 /* Destroy existing eviction groups, then destroy and recreate data
7808 * structures to recover memory. */
7809 struct eviction_group *evg, *next;
7810 HMAP_FOR_EACH_SAFE (evg, next, id_node, &table->eviction_groups_by_id) {
7811 eviction_group_destroy(table, evg);
7812 }
7813 hmap_destroy(&table->eviction_groups_by_id);
7814 hmap_init(&table->eviction_groups_by_id);
7815 heap_destroy(&table->eviction_groups_by_size);
7816 heap_init(&table->eviction_groups_by_size);
7817
7818 /* Replace eviction groups by the new ones, if there is a change. Free the
7819 * old fields only after allocating the new ones, because 'fields ==
7820 * table->eviction_fields' is possible. */
7821 struct mf_subfield *old_fields = table->eviction_fields;
7822 table->n_eviction_fields = n_fields;
7823 table->eviction_fields = (fields
7824 ? xmemdup(fields, n_fields * sizeof *fields)
7825 : NULL);
7826 free(old_fields);
7827
7828 /* Add the new eviction groups, if enabled. */
7829 table->eviction = eviction;
7830 if (table->eviction) {
7831 table->eviction_group_id_basis = random_uint32();
7832 CLS_FOR_EACH (rule, cr, &table->cls) {
7833 eviction_group_add_rule(rule);
7834 }
7835 }
7836 }
7837
7838 /* Inserts 'rule' from the ofproto data structures BEFORE caller has inserted
7839 * it to the classifier. */
7840 static void
7841 ofproto_rule_insert__(struct ofproto *ofproto, struct rule *rule)
7842 OVS_REQUIRES(ofproto_mutex)
7843 {
7844 const struct rule_actions *actions = rule_get_actions(rule);
7845
7846 ovs_assert(rule->removed);
7847
7848 if (rule->hard_timeout || rule->idle_timeout) {
7849 ovs_list_insert(&ofproto->expirable, &rule->expirable);
7850 }
7851 cookies_insert(ofproto, rule);
7852 eviction_group_add_rule(rule);
7853 if (actions->has_meter) {
7854 meter_insert_rule(rule);
7855 }
7856 rule->removed = false;
7857 }
7858
7859 /* Removes 'rule' from the ofproto data structures. Caller may have deferred
7860 * the removal from the classifier. */
7861 static void
7862 ofproto_rule_remove__(struct ofproto *ofproto, struct rule *rule)
7863 OVS_REQUIRES(ofproto_mutex)
7864 {
7865 ovs_assert(!rule->removed);
7866
7867 cookies_remove(ofproto, rule);
7868
7869 eviction_group_remove_rule(rule);
7870 if (!ovs_list_is_empty(&rule->expirable)) {
7871 ovs_list_remove(&rule->expirable);
7872 }
7873 if (!ovs_list_is_empty(&rule->meter_list_node)) {
7874 ovs_list_remove(&rule->meter_list_node);
7875 ovs_list_init(&rule->meter_list_node);
7876 }
7877
7878 rule->removed = true;
7879 }
7880 \f
7881 /* unixctl commands. */
7882
7883 struct ofproto *
7884 ofproto_lookup(const char *name)
7885 {
7886 struct ofproto *ofproto;
7887
7888 HMAP_FOR_EACH_WITH_HASH (ofproto, hmap_node, hash_string(name, 0),
7889 &all_ofprotos) {
7890 if (!strcmp(ofproto->name, name)) {
7891 return ofproto;
7892 }
7893 }
7894 return NULL;
7895 }
7896
7897 static void
7898 ofproto_unixctl_list(struct unixctl_conn *conn, int argc OVS_UNUSED,
7899 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7900 {
7901 struct ofproto *ofproto;
7902 struct ds results;
7903
7904 ds_init(&results);
7905 HMAP_FOR_EACH (ofproto, hmap_node, &all_ofprotos) {
7906 ds_put_format(&results, "%s\n", ofproto->name);
7907 }
7908 unixctl_command_reply(conn, ds_cstr(&results));
7909 ds_destroy(&results);
7910 }
7911
7912 static void
7913 ofproto_unixctl_init(void)
7914 {
7915 static bool registered;
7916 if (registered) {
7917 return;
7918 }
7919 registered = true;
7920
7921 unixctl_command_register("ofproto/list", "", 0, 0,
7922 ofproto_unixctl_list, NULL);
7923 }
7924 \f
7925 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
7926 *
7927 * This is deprecated. It is only for compatibility with broken device drivers
7928 * in old versions of Linux that do not properly support VLANs when VLAN
7929 * devices are not used. When broken device drivers are no longer in
7930 * widespread use, we will delete these interfaces. */
7931
7932 /* Sets a 1-bit in the 4096-bit 'vlan_bitmap' for each VLAN ID that is matched
7933 * (exactly) by an OpenFlow rule in 'ofproto'. */
7934 void
7935 ofproto_get_vlan_usage(struct ofproto *ofproto, unsigned long int *vlan_bitmap)
7936 {
7937 struct match match;
7938 struct cls_rule target;
7939 const struct oftable *oftable;
7940
7941 match_init_catchall(&match);
7942 match_set_vlan_vid_masked(&match, htons(VLAN_CFI), htons(VLAN_CFI));
7943 cls_rule_init(&target, &match, 0);
7944
7945 free(ofproto->vlan_bitmap);
7946 ofproto->vlan_bitmap = bitmap_allocate(4096);
7947 ofproto->vlans_changed = false;
7948
7949 OFPROTO_FOR_EACH_TABLE (oftable, ofproto) {
7950 struct rule *rule;
7951
7952 CLS_FOR_EACH_TARGET (rule, cr, &oftable->cls, &target,
7953 CLS_MAX_VERSION) {
7954 if (minimask_get_vid_mask(rule->cr.match.mask) == VLAN_VID_MASK) {
7955 uint16_t vid = miniflow_get_vid(rule->cr.match.flow);
7956
7957 bitmap_set1(vlan_bitmap, vid);
7958 bitmap_set1(ofproto->vlan_bitmap, vid);
7959 }
7960 }
7961 }
7962
7963 cls_rule_destroy(&target);
7964 }
7965
7966 /* Returns true if new VLANs have come into use by the flow table since the
7967 * last call to ofproto_get_vlan_usage().
7968 *
7969 * We don't track when old VLANs stop being used. */
7970 bool
7971 ofproto_has_vlan_usage_changed(const struct ofproto *ofproto)
7972 {
7973 return ofproto->vlans_changed;
7974 }
7975
7976 /* Configures a VLAN splinter binding between the ports identified by OpenFlow
7977 * port numbers 'vlandev_ofp_port' and 'realdev_ofp_port'. If
7978 * 'realdev_ofp_port' is nonzero, then the VLAN device is enslaved to the real
7979 * device as a VLAN splinter for VLAN ID 'vid'. If 'realdev_ofp_port' is zero,
7980 * then the VLAN device is un-enslaved. */
7981 int
7982 ofproto_port_set_realdev(struct ofproto *ofproto, ofp_port_t vlandev_ofp_port,
7983 ofp_port_t realdev_ofp_port, int vid)
7984 {
7985 struct ofport *ofport;
7986 int error;
7987
7988 ovs_assert(vlandev_ofp_port != realdev_ofp_port);
7989
7990 ofport = ofproto_get_port(ofproto, vlandev_ofp_port);
7991 if (!ofport) {
7992 VLOG_WARN("%s: cannot set realdev on nonexistent port %"PRIu16,
7993 ofproto->name, vlandev_ofp_port);
7994 return EINVAL;
7995 }
7996
7997 if (!ofproto->ofproto_class->set_realdev) {
7998 if (!vlandev_ofp_port) {
7999 return 0;
8000 }
8001 VLOG_WARN("%s: vlan splinters not supported", ofproto->name);
8002 return EOPNOTSUPP;
8003 }
8004
8005 error = ofproto->ofproto_class->set_realdev(ofport, realdev_ofp_port, vid);
8006 if (error) {
8007 VLOG_WARN("%s: setting realdev on port %"PRIu16" (%s) failed (%s)",
8008 ofproto->name, vlandev_ofp_port,
8009 netdev_get_name(ofport->netdev), ovs_strerror(error));
8010 }
8011 return error;
8012 }