]>
Commit | Line | Data |
---|---|---|
abe529af BP |
1 | /* |
2 | * Copyright (c) 2009, 2010, 2011 Nicira Networks. | |
3 | * | |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
5bee6e26 | 19 | #include "ofproto/ofproto-provider.h" |
abe529af BP |
20 | |
21 | #include <errno.h> | |
22 | ||
23 | #include "autopath.h" | |
24 | #include "bond.h" | |
daff3353 | 25 | #include "bundle.h" |
abe529af BP |
26 | #include "byte-order.h" |
27 | #include "connmgr.h" | |
28 | #include "coverage.h" | |
29 | #include "cfm.h" | |
30 | #include "dpif.h" | |
31 | #include "dynamic-string.h" | |
32 | #include "fail-open.h" | |
33 | #include "hmapx.h" | |
34 | #include "lacp.h" | |
35 | #include "mac-learning.h" | |
36 | #include "multipath.h" | |
37 | #include "netdev.h" | |
38 | #include "netlink.h" | |
39 | #include "nx-match.h" | |
40 | #include "odp-util.h" | |
41 | #include "ofp-util.h" | |
42 | #include "ofpbuf.h" | |
43 | #include "ofp-print.h" | |
bae473fe | 44 | #include "ofproto-dpif-sflow.h" |
abe529af BP |
45 | #include "poll-loop.h" |
46 | #include "timer.h" | |
6c1491fb | 47 | #include "unaligned.h" |
abe529af BP |
48 | #include "unixctl.h" |
49 | #include "vlan-bitmap.h" | |
50 | #include "vlog.h" | |
51 | ||
52 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
53 | ||
54 | COVERAGE_DEFINE(ofproto_dpif_ctlr_action); | |
55 | COVERAGE_DEFINE(ofproto_dpif_expired); | |
56 | COVERAGE_DEFINE(ofproto_dpif_no_packet_in); | |
57 | COVERAGE_DEFINE(ofproto_dpif_xlate); | |
58 | COVERAGE_DEFINE(facet_changed_rule); | |
59 | COVERAGE_DEFINE(facet_invalidated); | |
60 | COVERAGE_DEFINE(facet_revalidate); | |
61 | COVERAGE_DEFINE(facet_unexpected); | |
62 | ||
63 | /* Maximum depth of flow table recursion (due to NXAST_RESUBMIT actions) in a | |
64 | * flow translation. */ | |
65 | #define MAX_RESUBMIT_RECURSION 16 | |
66 | ||
67 | struct ofport_dpif; | |
68 | struct ofproto_dpif; | |
69 | ||
70 | struct rule_dpif { | |
71 | struct rule up; | |
72 | ||
73 | long long int used; /* Time last used; time created if not used. */ | |
74 | ||
75 | /* These statistics: | |
76 | * | |
77 | * - Do include packets and bytes from facets that have been deleted or | |
78 | * whose own statistics have been folded into the rule. | |
79 | * | |
80 | * - Do include packets and bytes sent "by hand" that were accounted to | |
81 | * the rule without any facet being involved (this is a rare corner | |
82 | * case in rule_execute()). | |
83 | * | |
84 | * - Do not include packet or bytes that can be obtained from any facet's | |
85 | * packet_count or byte_count member or that can be obtained from the | |
86 | * datapath by, e.g., dpif_flow_get() for any facet. | |
87 | */ | |
88 | uint64_t packet_count; /* Number of packets received. */ | |
89 | uint64_t byte_count; /* Number of bytes received. */ | |
90 | ||
91 | struct list facets; /* List of "struct facet"s. */ | |
92 | }; | |
93 | ||
94 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
95 | { | |
96 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
97 | } | |
98 | ||
99 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *ofproto, | |
100 | const struct flow *flow); | |
101 | ||
102 | #define MAX_MIRRORS 32 | |
103 | typedef uint32_t mirror_mask_t; | |
104 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
105 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
106 | struct ofmirror { | |
107 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
108 | size_t idx; /* In ofproto's "mirrors" array. */ | |
109 | void *aux; /* Key supplied by ofproto's client. */ | |
110 | char *name; /* Identifier for log messages. */ | |
111 | ||
112 | /* Selection criteria. */ | |
113 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
114 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
115 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
116 | ||
117 | /* Output (mutually exclusive). */ | |
118 | struct ofbundle *out; /* Output port or NULL. */ | |
119 | int out_vlan; /* Output VLAN or -1. */ | |
120 | }; | |
121 | ||
122 | static void mirror_destroy(struct ofmirror *); | |
123 | ||
124 | /* A group of one or more OpenFlow ports. */ | |
125 | #define OFBUNDLE_FLOOD ((struct ofbundle *) 1) | |
126 | struct ofbundle { | |
127 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
128 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
129 | void *aux; /* Key supplied by ofproto's client. */ | |
130 | char *name; /* Identifier for log messages. */ | |
131 | ||
132 | /* Configuration. */ | |
133 | struct list ports; /* Contains "struct ofport"s. */ | |
134 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ | |
135 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
136 | * NULL if all VLANs are trunked. */ | |
137 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
138 | struct bond *bond; /* Nonnull iff more than one port. */ | |
139 | ||
140 | /* Status. */ | |
141 | bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */ | |
142 | ||
143 | /* Port mirroring info. */ | |
144 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
145 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
146 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
147 | }; | |
148 | ||
149 | static void bundle_remove(struct ofport *); | |
150 | static void bundle_destroy(struct ofbundle *); | |
151 | static void bundle_del_port(struct ofport_dpif *); | |
152 | static void bundle_run(struct ofbundle *); | |
153 | static void bundle_wait(struct ofbundle *); | |
154 | ||
155 | struct action_xlate_ctx { | |
156 | /* action_xlate_ctx_init() initializes these members. */ | |
157 | ||
158 | /* The ofproto. */ | |
159 | struct ofproto_dpif *ofproto; | |
160 | ||
161 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
162 | * this flow when actions change header fields. */ | |
163 | struct flow flow; | |
164 | ||
165 | /* The packet corresponding to 'flow', or a null pointer if we are | |
166 | * revalidating without a packet to refer to. */ | |
167 | const struct ofpbuf *packet; | |
168 | ||
169 | /* If nonnull, called just before executing a resubmit action. | |
170 | * | |
171 | * This is normally null so the client has to set it manually after | |
172 | * calling action_xlate_ctx_init(). */ | |
173 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *); | |
174 | ||
abe529af BP |
175 | /* xlate_actions() initializes and uses these members. The client might want |
176 | * to look at them after it returns. */ | |
177 | ||
178 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
179 | tag_type tags; /* Tags associated with OFPP_NORMAL actions. */ | |
180 | bool may_set_up_flow; /* True ordinarily; false if the actions must | |
181 | * be reassessed for every packet. */ | |
182 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ | |
183 | ||
184 | /* xlate_actions() initializes and uses these members, but the client has no | |
185 | * reason to look at them. */ | |
186 | ||
187 | int recurse; /* Recursion level, via xlate_table_action. */ | |
afabef2b | 188 | uint32_t priority; /* Current flow priority. 0 if none. */ |
b3e9b2ed EJ |
189 | struct flow base_flow; /* Flow at the last commit. */ |
190 | uint32_t base_priority; /* Priority at the last commit. */ | |
abe529af BP |
191 | }; |
192 | ||
193 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
194 | struct ofproto_dpif *, const struct flow *, | |
195 | const struct ofpbuf *); | |
196 | static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, | |
197 | const union ofp_action *in, size_t n_in); | |
198 | ||
199 | /* An exact-match instantiation of an OpenFlow flow. */ | |
200 | struct facet { | |
201 | long long int used; /* Time last used; time created if not used. */ | |
202 | ||
203 | /* These statistics: | |
204 | * | |
205 | * - Do include packets and bytes sent "by hand", e.g. with | |
206 | * dpif_execute(). | |
207 | * | |
208 | * - Do include packets and bytes that were obtained from the datapath | |
209 | * when a flow was deleted (e.g. dpif_flow_del()) or when its | |
210 | * statistics were reset (e.g. dpif_flow_put() with | |
211 | * DPIF_FP_ZERO_STATS). | |
212 | * | |
213 | * - Do not include any packets or bytes that can currently be obtained | |
214 | * from the datapath by, e.g., dpif_flow_get(). | |
215 | */ | |
216 | uint64_t packet_count; /* Number of packets received. */ | |
217 | uint64_t byte_count; /* Number of bytes received. */ | |
218 | ||
219 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
220 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
221 | ||
222 | uint64_t rs_packet_count; /* Packets pushed to resubmit children. */ | |
223 | uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */ | |
224 | long long int rs_used; /* Used time pushed to resubmit children. */ | |
225 | ||
226 | /* Number of bytes passed to account_cb. This may include bytes that can | |
227 | * currently obtained from the datapath (thus, it can be greater than | |
228 | * byte_count). */ | |
229 | uint64_t accounted_bytes; | |
230 | ||
231 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
232 | struct list list_node; /* In owning rule's 'facets' list. */ | |
233 | struct rule_dpif *rule; /* Owning rule. */ | |
234 | struct flow flow; /* Exact-match flow. */ | |
235 | bool installed; /* Installed in datapath? */ | |
236 | bool may_install; /* True ordinarily; false if actions must | |
237 | * be reassessed for every packet. */ | |
238 | size_t actions_len; /* Number of bytes in actions[]. */ | |
239 | struct nlattr *actions; /* Datapath actions. */ | |
240 | tag_type tags; /* Tags. */ | |
241 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
242 | }; | |
243 | ||
244 | static struct facet *facet_create(struct rule_dpif *, const struct flow *, | |
245 | const struct ofpbuf *packet); | |
246 | static void facet_remove(struct ofproto_dpif *, struct facet *); | |
247 | static void facet_free(struct facet *); | |
248 | ||
249 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); | |
250 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, | |
251 | const struct flow *); | |
252 | static bool facet_revalidate(struct ofproto_dpif *, struct facet *); | |
253 | ||
254 | static void facet_execute(struct ofproto_dpif *, struct facet *, | |
255 | struct ofpbuf *packet); | |
256 | ||
257 | static int facet_put__(struct ofproto_dpif *, struct facet *, | |
258 | const struct nlattr *actions, size_t actions_len, | |
259 | struct dpif_flow_stats *); | |
260 | static void facet_install(struct ofproto_dpif *, struct facet *, | |
261 | bool zero_stats); | |
262 | static void facet_uninstall(struct ofproto_dpif *, struct facet *); | |
263 | static void facet_flush_stats(struct ofproto_dpif *, struct facet *); | |
264 | ||
265 | static void facet_make_actions(struct ofproto_dpif *, struct facet *, | |
266 | const struct ofpbuf *packet); | |
267 | static void facet_update_time(struct ofproto_dpif *, struct facet *, | |
268 | long long int used); | |
269 | static void facet_update_stats(struct ofproto_dpif *, struct facet *, | |
270 | const struct dpif_flow_stats *); | |
3a88e544 | 271 | static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *); |
abe529af BP |
272 | static void facet_push_stats(struct facet *); |
273 | static void facet_account(struct ofproto_dpif *, struct facet *, | |
274 | uint64_t extra_bytes); | |
275 | ||
276 | static bool facet_is_controller_flow(struct facet *); | |
277 | ||
278 | static void flow_push_stats(const struct rule_dpif *, | |
279 | struct flow *, uint64_t packets, uint64_t bytes, | |
280 | long long int used); | |
281 | ||
282 | struct ofport_dpif { | |
283 | struct ofport up; | |
284 | ||
285 | uint32_t odp_port; | |
286 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
287 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
288 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
289 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 290 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 291 | bool may_enable; /* May be enabled in bonds. */ |
abe529af BP |
292 | }; |
293 | ||
294 | static struct ofport_dpif * | |
295 | ofport_dpif_cast(const struct ofport *ofport) | |
296 | { | |
297 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
298 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
299 | } | |
300 | ||
301 | static void port_run(struct ofport_dpif *); | |
302 | static void port_wait(struct ofport_dpif *); | |
a5610457 | 303 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
abe529af | 304 | |
7ee20df1 BP |
305 | struct dpif_completion { |
306 | struct list list_node; | |
307 | struct ofoperation *op; | |
308 | }; | |
309 | ||
abe529af BP |
310 | struct ofproto_dpif { |
311 | struct ofproto up; | |
312 | struct dpif *dpif; | |
313 | int max_ports; | |
314 | ||
6c1491fb BP |
315 | /* Statistics. */ |
316 | uint64_t n_matches; | |
317 | ||
abe529af BP |
318 | /* Bridging. */ |
319 | struct netflow *netflow; | |
bae473fe | 320 | struct dpif_sflow *sflow; |
abe529af BP |
321 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
322 | struct mac_learning *ml; | |
323 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
324 | bool has_bonded_bundles; | |
325 | ||
326 | /* Expiration. */ | |
327 | struct timer next_expiration; | |
328 | ||
329 | /* Facets. */ | |
330 | struct hmap facets; | |
331 | bool need_revalidate; | |
332 | struct tag_set revalidate_set; | |
7ee20df1 BP |
333 | |
334 | /* Support for debugging async flow mods. */ | |
335 | struct list completions; | |
daff3353 EJ |
336 | |
337 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
abe529af BP |
338 | }; |
339 | ||
7ee20df1 BP |
340 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
341 | * for debugging the asynchronous flow_mod implementation.) */ | |
342 | static bool clogged; | |
343 | ||
abe529af BP |
344 | static void ofproto_dpif_unixctl_init(void); |
345 | ||
346 | static struct ofproto_dpif * | |
347 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
348 | { | |
349 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
350 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
351 | } | |
352 | ||
353 | static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, | |
354 | uint16_t ofp_port); | |
355 | static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, | |
356 | uint32_t odp_port); | |
357 | ||
358 | /* Packet processing. */ | |
359 | static void update_learning_table(struct ofproto_dpif *, | |
360 | const struct flow *, int vlan, | |
361 | struct ofbundle *); | |
362 | static bool is_admissible(struct ofproto_dpif *, const struct flow *, | |
363 | bool have_packet, tag_type *, int *vlanp, | |
364 | struct ofbundle **in_bundlep); | |
365 | static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *); | |
366 | ||
367 | /* Flow expiration. */ | |
368 | static int expire(struct ofproto_dpif *); | |
369 | ||
370 | /* Utilities. */ | |
b2fda3ef | 371 | static int send_packet(struct ofproto_dpif *, uint32_t odp_port, |
abe529af BP |
372 | const struct ofpbuf *packet); |
373 | ||
374 | /* Global variables. */ | |
375 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
376 | \f | |
377 | /* Factory functions. */ | |
378 | ||
379 | static void | |
380 | enumerate_types(struct sset *types) | |
381 | { | |
382 | dp_enumerate_types(types); | |
383 | } | |
384 | ||
385 | static int | |
386 | enumerate_names(const char *type, struct sset *names) | |
387 | { | |
388 | return dp_enumerate_names(type, names); | |
389 | } | |
390 | ||
391 | static int | |
392 | del(const char *type, const char *name) | |
393 | { | |
394 | struct dpif *dpif; | |
395 | int error; | |
396 | ||
397 | error = dpif_open(name, type, &dpif); | |
398 | if (!error) { | |
399 | error = dpif_delete(dpif); | |
400 | dpif_close(dpif); | |
401 | } | |
402 | return error; | |
403 | } | |
404 | \f | |
405 | /* Basic life-cycle. */ | |
406 | ||
407 | static struct ofproto * | |
408 | alloc(void) | |
409 | { | |
410 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
411 | return &ofproto->up; | |
412 | } | |
413 | ||
414 | static void | |
415 | dealloc(struct ofproto *ofproto_) | |
416 | { | |
417 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
418 | free(ofproto); | |
419 | } | |
420 | ||
421 | static int | |
422 | construct(struct ofproto *ofproto_) | |
423 | { | |
424 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
425 | const char *name = ofproto->up.name; | |
426 | int error; | |
427 | int i; | |
428 | ||
429 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
430 | if (error) { | |
431 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
432 | return error; | |
433 | } | |
434 | ||
435 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 436 | ofproto->n_matches = 0; |
abe529af BP |
437 | |
438 | error = dpif_recv_set_mask(ofproto->dpif, | |
439 | ((1u << DPIF_UC_MISS) | | |
440 | (1u << DPIF_UC_ACTION) | | |
441 | (1u << DPIF_UC_SAMPLE))); | |
442 | if (error) { | |
443 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
444 | dpif_close(ofproto->dpif); | |
445 | return error; | |
446 | } | |
447 | dpif_flow_flush(ofproto->dpif); | |
448 | dpif_recv_purge(ofproto->dpif); | |
449 | ||
450 | ofproto->netflow = NULL; | |
451 | ofproto->sflow = NULL; | |
452 | hmap_init(&ofproto->bundles); | |
453 | ofproto->ml = mac_learning_create(); | |
454 | for (i = 0; i < MAX_MIRRORS; i++) { | |
455 | ofproto->mirrors[i] = NULL; | |
456 | } | |
457 | ofproto->has_bonded_bundles = false; | |
458 | ||
459 | timer_set_duration(&ofproto->next_expiration, 1000); | |
460 | ||
461 | hmap_init(&ofproto->facets); | |
462 | ofproto->need_revalidate = false; | |
463 | tag_set_init(&ofproto->revalidate_set); | |
464 | ||
7ee20df1 BP |
465 | list_init(&ofproto->completions); |
466 | ||
6c1491fb BP |
467 | ofproto->up.tables = xmalloc(sizeof *ofproto->up.tables); |
468 | classifier_init(&ofproto->up.tables[0]); | |
469 | ofproto->up.n_tables = 1; | |
470 | ||
abe529af BP |
471 | ofproto_dpif_unixctl_init(); |
472 | ||
daff3353 EJ |
473 | ofproto->has_bundle_action = false; |
474 | ||
abe529af BP |
475 | return 0; |
476 | } | |
477 | ||
7ee20df1 BP |
478 | static void |
479 | complete_operations(struct ofproto_dpif *ofproto) | |
480 | { | |
481 | struct dpif_completion *c, *next; | |
482 | ||
483 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
484 | ofoperation_complete(c->op, 0); | |
485 | list_remove(&c->list_node); | |
486 | free(c); | |
487 | } | |
488 | } | |
489 | ||
abe529af BP |
490 | static void |
491 | destruct(struct ofproto *ofproto_) | |
492 | { | |
493 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 BP |
494 | struct rule_dpif *rule, *next_rule; |
495 | struct cls_cursor cursor; | |
abe529af BP |
496 | int i; |
497 | ||
7ee20df1 BP |
498 | complete_operations(ofproto); |
499 | ||
500 | cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL); | |
501 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
502 | ofproto_rule_destroy(&rule->up); | |
503 | } | |
504 | ||
abe529af BP |
505 | for (i = 0; i < MAX_MIRRORS; i++) { |
506 | mirror_destroy(ofproto->mirrors[i]); | |
507 | } | |
508 | ||
509 | netflow_destroy(ofproto->netflow); | |
bae473fe | 510 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
511 | hmap_destroy(&ofproto->bundles); |
512 | mac_learning_destroy(ofproto->ml); | |
513 | ||
514 | hmap_destroy(&ofproto->facets); | |
515 | ||
516 | dpif_close(ofproto->dpif); | |
517 | } | |
518 | ||
519 | static int | |
520 | run(struct ofproto *ofproto_) | |
521 | { | |
522 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
523 | struct ofport_dpif *ofport; | |
524 | struct ofbundle *bundle; | |
525 | int i; | |
526 | ||
7ee20df1 BP |
527 | if (!clogged) { |
528 | complete_operations(ofproto); | |
529 | } | |
abe529af BP |
530 | dpif_run(ofproto->dpif); |
531 | ||
532 | for (i = 0; i < 50; i++) { | |
533 | struct dpif_upcall packet; | |
534 | int error; | |
535 | ||
536 | error = dpif_recv(ofproto->dpif, &packet); | |
537 | if (error) { | |
538 | if (error == ENODEV) { | |
539 | /* Datapath destroyed. */ | |
540 | return error; | |
541 | } | |
542 | break; | |
543 | } | |
544 | ||
545 | handle_upcall(ofproto, &packet); | |
546 | } | |
547 | ||
548 | if (timer_expired(&ofproto->next_expiration)) { | |
549 | int delay = expire(ofproto); | |
550 | timer_set_duration(&ofproto->next_expiration, delay); | |
551 | } | |
552 | ||
553 | if (ofproto->netflow) { | |
554 | netflow_run(ofproto->netflow); | |
555 | } | |
556 | if (ofproto->sflow) { | |
bae473fe | 557 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
558 | } |
559 | ||
560 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
561 | port_run(ofport); | |
562 | } | |
563 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
564 | bundle_run(bundle); | |
565 | } | |
566 | ||
567 | /* Now revalidate if there's anything to do. */ | |
568 | if (ofproto->need_revalidate | |
569 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
570 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
571 | bool revalidate_all = ofproto->need_revalidate; | |
572 | struct facet *facet, *next; | |
573 | ||
574 | /* Clear the revalidation flags. */ | |
575 | tag_set_init(&ofproto->revalidate_set); | |
576 | ofproto->need_revalidate = false; | |
577 | ||
578 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
579 | if (revalidate_all | |
580 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
581 | facet_revalidate(ofproto, facet); | |
582 | } | |
583 | } | |
584 | } | |
585 | ||
586 | return 0; | |
587 | } | |
588 | ||
589 | static void | |
590 | wait(struct ofproto *ofproto_) | |
591 | { | |
592 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
593 | struct ofport_dpif *ofport; | |
594 | struct ofbundle *bundle; | |
595 | ||
7ee20df1 BP |
596 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
597 | poll_immediate_wake(); | |
598 | } | |
599 | ||
abe529af BP |
600 | dpif_wait(ofproto->dpif); |
601 | dpif_recv_wait(ofproto->dpif); | |
602 | if (ofproto->sflow) { | |
bae473fe | 603 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
604 | } |
605 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
606 | poll_immediate_wake(); | |
607 | } | |
608 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
609 | port_wait(ofport); | |
610 | } | |
611 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
612 | bundle_wait(bundle); | |
613 | } | |
614 | if (ofproto->need_revalidate) { | |
615 | /* Shouldn't happen, but if it does just go around again. */ | |
616 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
617 | poll_immediate_wake(); | |
618 | } else { | |
619 | timer_wait(&ofproto->next_expiration); | |
620 | } | |
621 | } | |
622 | ||
623 | static void | |
624 | flush(struct ofproto *ofproto_) | |
625 | { | |
626 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
627 | struct facet *facet, *next_facet; | |
628 | ||
629 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
630 | /* Mark the facet as not installed so that facet_remove() doesn't | |
631 | * bother trying to uninstall it. There is no point in uninstalling it | |
632 | * individually since we are about to blow away all the facets with | |
633 | * dpif_flow_flush(). */ | |
634 | facet->installed = false; | |
635 | facet->dp_packet_count = 0; | |
636 | facet->dp_byte_count = 0; | |
637 | facet_remove(ofproto, facet); | |
638 | } | |
639 | dpif_flow_flush(ofproto->dpif); | |
640 | } | |
641 | ||
6c1491fb BP |
642 | static void |
643 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
644 | bool *arp_match_ip, uint32_t *actions) | |
645 | { | |
646 | *arp_match_ip = true; | |
647 | *actions = ((1u << OFPAT_OUTPUT) | | |
648 | (1u << OFPAT_SET_VLAN_VID) | | |
649 | (1u << OFPAT_SET_VLAN_PCP) | | |
650 | (1u << OFPAT_STRIP_VLAN) | | |
651 | (1u << OFPAT_SET_DL_SRC) | | |
652 | (1u << OFPAT_SET_DL_DST) | | |
653 | (1u << OFPAT_SET_NW_SRC) | | |
654 | (1u << OFPAT_SET_NW_DST) | | |
655 | (1u << OFPAT_SET_NW_TOS) | | |
656 | (1u << OFPAT_SET_TP_SRC) | | |
657 | (1u << OFPAT_SET_TP_DST) | | |
658 | (1u << OFPAT_ENQUEUE)); | |
659 | } | |
660 | ||
661 | static void | |
662 | get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) | |
663 | { | |
664 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
665 | struct odp_stats s; | |
666 | ||
667 | strcpy(ots->name, "classifier"); | |
668 | ||
669 | dpif_get_dp_stats(ofproto->dpif, &s); | |
670 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
671 | put_32aligned_be64(&ots->matched_count, | |
672 | htonll(s.n_hit + ofproto->n_matches)); | |
673 | } | |
674 | ||
abe529af BP |
675 | static int |
676 | set_netflow(struct ofproto *ofproto_, | |
677 | const struct netflow_options *netflow_options) | |
678 | { | |
679 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
680 | ||
681 | if (netflow_options) { | |
682 | if (!ofproto->netflow) { | |
683 | ofproto->netflow = netflow_create(); | |
684 | } | |
685 | return netflow_set_options(ofproto->netflow, netflow_options); | |
686 | } else { | |
687 | netflow_destroy(ofproto->netflow); | |
688 | ofproto->netflow = NULL; | |
689 | return 0; | |
690 | } | |
691 | } | |
692 | ||
693 | static struct ofport * | |
694 | port_alloc(void) | |
695 | { | |
696 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
697 | return &port->up; | |
698 | } | |
699 | ||
700 | static void | |
701 | port_dealloc(struct ofport *port_) | |
702 | { | |
703 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
704 | free(port); | |
705 | } | |
706 | ||
707 | static int | |
708 | port_construct(struct ofport *port_) | |
709 | { | |
710 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
711 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
712 | ||
713 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); | |
714 | port->bundle = NULL; | |
715 | port->cfm = NULL; | |
716 | port->tag = tag_create_random(); | |
d5ffa7f2 | 717 | port->may_enable = true; |
abe529af BP |
718 | |
719 | if (ofproto->sflow) { | |
bae473fe JP |
720 | dpif_sflow_add_port(ofproto->sflow, port->odp_port, |
721 | netdev_get_name(port->up.netdev)); | |
abe529af BP |
722 | } |
723 | ||
724 | return 0; | |
725 | } | |
726 | ||
727 | static void | |
728 | port_destruct(struct ofport *port_) | |
729 | { | |
730 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
731 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
732 | ||
733 | bundle_remove(port_); | |
a5610457 | 734 | set_cfm(port_, NULL); |
abe529af | 735 | if (ofproto->sflow) { |
bae473fe | 736 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af BP |
737 | } |
738 | } | |
739 | ||
740 | static void | |
741 | port_modified(struct ofport *port_) | |
742 | { | |
743 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
744 | ||
745 | if (port->bundle && port->bundle->bond) { | |
746 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
747 | } | |
748 | } | |
749 | ||
750 | static void | |
751 | port_reconfigured(struct ofport *port_, ovs_be32 old_config) | |
752 | { | |
753 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
754 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
755 | ovs_be32 changed = old_config ^ port->up.opp.config; | |
756 | ||
757 | if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP | | |
758 | OFPPC_NO_FWD | OFPPC_NO_FLOOD)) { | |
759 | ofproto->need_revalidate = true; | |
760 | } | |
761 | } | |
762 | ||
763 | static int | |
764 | set_sflow(struct ofproto *ofproto_, | |
765 | const struct ofproto_sflow_options *sflow_options) | |
766 | { | |
767 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 768 | struct dpif_sflow *ds = ofproto->sflow; |
abe529af | 769 | if (sflow_options) { |
bae473fe | 770 | if (!ds) { |
abe529af BP |
771 | struct ofport_dpif *ofport; |
772 | ||
bae473fe | 773 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 774 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
bae473fe JP |
775 | dpif_sflow_add_port(ds, ofport->odp_port, |
776 | netdev_get_name(ofport->up.netdev)); | |
abe529af BP |
777 | } |
778 | } | |
bae473fe | 779 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 780 | } else { |
bae473fe | 781 | dpif_sflow_destroy(ds); |
abe529af BP |
782 | ofproto->sflow = NULL; |
783 | } | |
784 | return 0; | |
785 | } | |
786 | ||
787 | static int | |
a5610457 | 788 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
789 | { |
790 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
791 | int error; | |
792 | ||
a5610457 | 793 | if (!s) { |
abe529af BP |
794 | error = 0; |
795 | } else { | |
796 | if (!ofport->cfm) { | |
6f629657 | 797 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
798 | } |
799 | ||
a5610457 | 800 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
801 | return 0; |
802 | } | |
803 | ||
804 | error = EINVAL; | |
805 | } | |
806 | cfm_destroy(ofport->cfm); | |
807 | ofport->cfm = NULL; | |
808 | return error; | |
809 | } | |
810 | ||
811 | static int | |
a5610457 | 812 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
813 | { |
814 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
815 | |
816 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af BP |
817 | } |
818 | \f | |
819 | /* Bundles. */ | |
820 | ||
821 | /* Expires all MAC learning entries associated with 'port' and forces ofproto | |
822 | * to revalidate every flow. */ | |
823 | static void | |
824 | bundle_flush_macs(struct ofbundle *bundle) | |
825 | { | |
826 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
827 | struct mac_learning *ml = ofproto->ml; | |
828 | struct mac_entry *mac, *next_mac; | |
829 | ||
830 | ofproto->need_revalidate = true; | |
831 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
832 | if (mac->port.p == bundle) { | |
833 | mac_learning_expire(ml, mac); | |
834 | } | |
835 | } | |
836 | } | |
837 | ||
838 | static struct ofbundle * | |
839 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
840 | { | |
841 | struct ofbundle *bundle; | |
842 | ||
843 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
844 | &ofproto->bundles) { | |
845 | if (bundle->aux == aux) { | |
846 | return bundle; | |
847 | } | |
848 | } | |
849 | return NULL; | |
850 | } | |
851 | ||
852 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
853 | * ones that are found to 'bundles'. */ | |
854 | static void | |
855 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
856 | void **auxes, size_t n_auxes, | |
857 | struct hmapx *bundles) | |
858 | { | |
859 | size_t i; | |
860 | ||
861 | hmapx_init(bundles); | |
862 | for (i = 0; i < n_auxes; i++) { | |
863 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
864 | if (bundle) { | |
865 | hmapx_add(bundles, bundle); | |
866 | } | |
867 | } | |
868 | } | |
869 | ||
870 | static void | |
871 | bundle_del_port(struct ofport_dpif *port) | |
872 | { | |
873 | struct ofbundle *bundle = port->bundle; | |
874 | ||
6f77f4ae BP |
875 | bundle->ofproto->need_revalidate = true; |
876 | ||
abe529af BP |
877 | list_remove(&port->bundle_node); |
878 | port->bundle = NULL; | |
879 | ||
880 | if (bundle->lacp) { | |
881 | lacp_slave_unregister(bundle->lacp, port); | |
882 | } | |
883 | if (bundle->bond) { | |
884 | bond_slave_unregister(bundle->bond, port); | |
885 | } | |
886 | ||
887 | bundle->floodable = true; | |
888 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
889 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { | |
890 | bundle->floodable = false; | |
891 | } | |
892 | } | |
893 | } | |
894 | ||
895 | static bool | |
896 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
897 | struct lacp_slave_settings *lacp, |
898 | uint32_t bond_stable_id) | |
abe529af BP |
899 | { |
900 | struct ofport_dpif *port; | |
901 | ||
902 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
903 | if (!port) { | |
904 | return false; | |
905 | } | |
906 | ||
907 | if (port->bundle != bundle) { | |
6f77f4ae | 908 | bundle->ofproto->need_revalidate = true; |
abe529af BP |
909 | if (port->bundle) { |
910 | bundle_del_port(port); | |
911 | } | |
912 | ||
913 | port->bundle = bundle; | |
914 | list_push_back(&bundle->ports, &port->bundle_node); | |
915 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { | |
916 | bundle->floodable = false; | |
917 | } | |
918 | } | |
919 | if (lacp) { | |
920 | lacp_slave_register(bundle->lacp, port, lacp); | |
921 | } | |
922 | ||
00794817 BP |
923 | port->bond_stable_id = bond_stable_id; |
924 | ||
abe529af BP |
925 | return true; |
926 | } | |
927 | ||
928 | static void | |
929 | bundle_destroy(struct ofbundle *bundle) | |
930 | { | |
931 | struct ofproto_dpif *ofproto; | |
932 | struct ofport_dpif *port, *next_port; | |
933 | int i; | |
934 | ||
935 | if (!bundle) { | |
936 | return; | |
937 | } | |
938 | ||
939 | ofproto = bundle->ofproto; | |
940 | for (i = 0; i < MAX_MIRRORS; i++) { | |
941 | struct ofmirror *m = ofproto->mirrors[i]; | |
942 | if (m) { | |
943 | if (m->out == bundle) { | |
944 | mirror_destroy(m); | |
945 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
946 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
947 | ofproto->need_revalidate = true; | |
948 | } | |
949 | } | |
950 | } | |
951 | ||
952 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
953 | bundle_del_port(port); | |
954 | } | |
955 | ||
956 | bundle_flush_macs(bundle); | |
957 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); | |
958 | free(bundle->name); | |
959 | free(bundle->trunks); | |
960 | lacp_destroy(bundle->lacp); | |
961 | bond_destroy(bundle->bond); | |
962 | free(bundle); | |
963 | } | |
964 | ||
965 | static int | |
966 | bundle_set(struct ofproto *ofproto_, void *aux, | |
967 | const struct ofproto_bundle_settings *s) | |
968 | { | |
969 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
970 | bool need_flush = false; | |
971 | const unsigned long *trunks; | |
972 | struct ofport_dpif *port; | |
973 | struct ofbundle *bundle; | |
974 | size_t i; | |
975 | bool ok; | |
976 | ||
977 | if (!s) { | |
978 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
979 | return 0; | |
980 | } | |
981 | ||
982 | assert(s->n_slaves == 1 || s->bond != NULL); | |
983 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
984 | ||
985 | bundle = bundle_lookup(ofproto, aux); | |
986 | if (!bundle) { | |
987 | bundle = xmalloc(sizeof *bundle); | |
988 | ||
989 | bundle->ofproto = ofproto; | |
990 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
991 | hash_pointer(aux, 0)); | |
992 | bundle->aux = aux; | |
993 | bundle->name = NULL; | |
994 | ||
995 | list_init(&bundle->ports); | |
996 | bundle->vlan = -1; | |
997 | bundle->trunks = NULL; | |
998 | bundle->lacp = NULL; | |
999 | bundle->bond = NULL; | |
1000 | ||
1001 | bundle->floodable = true; | |
1002 | ||
1003 | bundle->src_mirrors = 0; | |
1004 | bundle->dst_mirrors = 0; | |
1005 | bundle->mirror_out = 0; | |
1006 | } | |
1007 | ||
1008 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1009 | free(bundle->name); | |
1010 | bundle->name = xstrdup(s->name); | |
1011 | } | |
1012 | ||
1013 | /* LACP. */ | |
1014 | if (s->lacp) { | |
1015 | if (!bundle->lacp) { | |
1016 | bundle->lacp = lacp_create(); | |
1017 | } | |
1018 | lacp_configure(bundle->lacp, s->lacp); | |
1019 | } else { | |
1020 | lacp_destroy(bundle->lacp); | |
1021 | bundle->lacp = NULL; | |
1022 | } | |
1023 | ||
1024 | /* Update set of ports. */ | |
1025 | ok = true; | |
1026 | for (i = 0; i < s->n_slaves; i++) { | |
1027 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1028 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1029 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1030 | ok = false; |
1031 | } | |
1032 | } | |
1033 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1034 | struct ofport_dpif *next_port; | |
1035 | ||
1036 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1037 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1038 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1039 | goto found; |
1040 | } | |
1041 | } | |
1042 | ||
1043 | bundle_del_port(port); | |
1044 | found: ; | |
1045 | } | |
1046 | } | |
1047 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1048 | ||
1049 | if (list_is_empty(&bundle->ports)) { | |
1050 | bundle_destroy(bundle); | |
1051 | return EINVAL; | |
1052 | } | |
1053 | ||
1054 | /* Set VLAN tag. */ | |
1055 | if (s->vlan != bundle->vlan) { | |
1056 | bundle->vlan = s->vlan; | |
1057 | need_flush = true; | |
1058 | } | |
1059 | ||
1060 | /* Get trunked VLANs. */ | |
1061 | trunks = s->vlan == -1 ? NULL : s->trunks; | |
1062 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { | |
1063 | free(bundle->trunks); | |
1064 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1065 | need_flush = true; | |
1066 | } | |
1067 | ||
1068 | /* Bonding. */ | |
1069 | if (!list_is_short(&bundle->ports)) { | |
1070 | bundle->ofproto->has_bonded_bundles = true; | |
1071 | if (bundle->bond) { | |
1072 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
1073 | ofproto->need_revalidate = true; | |
1074 | } | |
1075 | } else { | |
1076 | bundle->bond = bond_create(s->bond); | |
6f77f4ae | 1077 | ofproto->need_revalidate = true; |
abe529af BP |
1078 | } |
1079 | ||
1080 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 1081 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
1082 | port->up.netdev); |
1083 | } | |
1084 | } else { | |
1085 | bond_destroy(bundle->bond); | |
1086 | bundle->bond = NULL; | |
1087 | } | |
1088 | ||
1089 | /* If we changed something that would affect MAC learning, un-learn | |
1090 | * everything on this port and force flow revalidation. */ | |
1091 | if (need_flush) { | |
1092 | bundle_flush_macs(bundle); | |
1093 | } | |
1094 | ||
1095 | return 0; | |
1096 | } | |
1097 | ||
1098 | static void | |
1099 | bundle_remove(struct ofport *port_) | |
1100 | { | |
1101 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1102 | struct ofbundle *bundle = port->bundle; | |
1103 | ||
1104 | if (bundle) { | |
1105 | bundle_del_port(port); | |
1106 | if (list_is_empty(&bundle->ports)) { | |
1107 | bundle_destroy(bundle); | |
1108 | } else if (list_is_short(&bundle->ports)) { | |
1109 | bond_destroy(bundle->bond); | |
1110 | bundle->bond = NULL; | |
1111 | } | |
1112 | } | |
1113 | } | |
1114 | ||
1115 | static void | |
1116 | send_pdu_cb(void *port_, const struct lacp_pdu *pdu) | |
1117 | { | |
1118 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
1119 | struct ofport_dpif *port = port_; | |
1120 | uint8_t ea[ETH_ADDR_LEN]; | |
1121 | int error; | |
1122 | ||
1123 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
1124 | if (!error) { | |
1125 | struct lacp_pdu *packet_pdu; | |
1126 | struct ofpbuf packet; | |
1127 | ||
1128 | ofpbuf_init(&packet, 0); | |
1129 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
1130 | sizeof *packet_pdu); | |
1131 | *packet_pdu = *pdu; | |
1132 | error = netdev_send(port->up.netdev, &packet); | |
1133 | if (error) { | |
1134 | VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed " | |
1135 | "(%s)", port->bundle->name, | |
1136 | netdev_get_name(port->up.netdev), strerror(error)); | |
1137 | } | |
1138 | ofpbuf_uninit(&packet); | |
1139 | } else { | |
1140 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
1141 | "%s (%s)", port->bundle->name, | |
1142 | netdev_get_name(port->up.netdev), strerror(error)); | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | static void | |
1147 | bundle_send_learning_packets(struct ofbundle *bundle) | |
1148 | { | |
1149 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1150 | int error, n_packets, n_errors; | |
1151 | struct mac_entry *e; | |
1152 | ||
1153 | error = n_packets = n_errors = 0; | |
1154 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
1155 | if (e->port.p != bundle) { | |
1156 | int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan); | |
1157 | if (ret) { | |
1158 | error = ret; | |
1159 | n_errors++; | |
1160 | } | |
1161 | n_packets++; | |
1162 | } | |
1163 | } | |
1164 | ||
1165 | if (n_errors) { | |
1166 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1167 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
1168 | "packets, last error was: %s", | |
1169 | bundle->name, n_errors, n_packets, strerror(error)); | |
1170 | } else { | |
1171 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
1172 | bundle->name, n_packets); | |
1173 | } | |
1174 | } | |
1175 | ||
1176 | static void | |
1177 | bundle_run(struct ofbundle *bundle) | |
1178 | { | |
1179 | if (bundle->lacp) { | |
1180 | lacp_run(bundle->lacp, send_pdu_cb); | |
1181 | } | |
1182 | if (bundle->bond) { | |
1183 | struct ofport_dpif *port; | |
1184 | ||
1185 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 1186 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
1187 | } |
1188 | ||
1189 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
1190 | lacp_negotiated(bundle->lacp)); | |
1191 | if (bond_should_send_learning_packets(bundle->bond)) { | |
1192 | bundle_send_learning_packets(bundle); | |
1193 | } | |
1194 | } | |
1195 | } | |
1196 | ||
1197 | static void | |
1198 | bundle_wait(struct ofbundle *bundle) | |
1199 | { | |
1200 | if (bundle->lacp) { | |
1201 | lacp_wait(bundle->lacp); | |
1202 | } | |
1203 | if (bundle->bond) { | |
1204 | bond_wait(bundle->bond); | |
1205 | } | |
1206 | } | |
1207 | \f | |
1208 | /* Mirrors. */ | |
1209 | ||
1210 | static int | |
1211 | mirror_scan(struct ofproto_dpif *ofproto) | |
1212 | { | |
1213 | int idx; | |
1214 | ||
1215 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
1216 | if (!ofproto->mirrors[idx]) { | |
1217 | return idx; | |
1218 | } | |
1219 | } | |
1220 | return -1; | |
1221 | } | |
1222 | ||
1223 | static struct ofmirror * | |
1224 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
1225 | { | |
1226 | int i; | |
1227 | ||
1228 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1229 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
1230 | if (mirror && mirror->aux == aux) { | |
1231 | return mirror; | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | return NULL; | |
1236 | } | |
1237 | ||
1238 | static int | |
1239 | mirror_set(struct ofproto *ofproto_, void *aux, | |
1240 | const struct ofproto_mirror_settings *s) | |
1241 | { | |
1242 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1243 | mirror_mask_t mirror_bit; | |
1244 | struct ofbundle *bundle; | |
1245 | struct ofmirror *mirror; | |
1246 | struct ofbundle *out; | |
1247 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
1248 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
1249 | int out_vlan; | |
1250 | ||
1251 | mirror = mirror_lookup(ofproto, aux); | |
1252 | if (!s) { | |
1253 | mirror_destroy(mirror); | |
1254 | return 0; | |
1255 | } | |
1256 | if (!mirror) { | |
1257 | int idx; | |
1258 | ||
1259 | idx = mirror_scan(ofproto); | |
1260 | if (idx < 0) { | |
1261 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
1262 | "cannot create %s", | |
1263 | ofproto->up.name, MAX_MIRRORS, s->name); | |
1264 | return EFBIG; | |
1265 | } | |
1266 | ||
1267 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
1268 | mirror->ofproto = ofproto; | |
1269 | mirror->idx = idx; | |
1270 | mirror->out_vlan = -1; | |
1271 | mirror->name = NULL; | |
1272 | } | |
1273 | ||
1274 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
1275 | free(mirror->name); | |
1276 | mirror->name = xstrdup(s->name); | |
1277 | } | |
1278 | ||
1279 | /* Get the new configuration. */ | |
1280 | if (s->out_bundle) { | |
1281 | out = bundle_lookup(ofproto, s->out_bundle); | |
1282 | if (!out) { | |
1283 | mirror_destroy(mirror); | |
1284 | return EINVAL; | |
1285 | } | |
1286 | out_vlan = -1; | |
1287 | } else { | |
1288 | out = NULL; | |
1289 | out_vlan = s->out_vlan; | |
1290 | } | |
1291 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
1292 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
1293 | ||
1294 | /* If the configuration has not changed, do nothing. */ | |
1295 | if (hmapx_equals(&srcs, &mirror->srcs) | |
1296 | && hmapx_equals(&dsts, &mirror->dsts) | |
1297 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
1298 | && mirror->out == out | |
1299 | && mirror->out_vlan == out_vlan) | |
1300 | { | |
1301 | hmapx_destroy(&srcs); | |
1302 | hmapx_destroy(&dsts); | |
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | hmapx_swap(&srcs, &mirror->srcs); | |
1307 | hmapx_destroy(&srcs); | |
1308 | ||
1309 | hmapx_swap(&dsts, &mirror->dsts); | |
1310 | hmapx_destroy(&dsts); | |
1311 | ||
1312 | free(mirror->vlans); | |
1313 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
1314 | ||
1315 | mirror->out = out; | |
1316 | mirror->out_vlan = out_vlan; | |
1317 | ||
1318 | /* Update bundles. */ | |
1319 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1320 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
1321 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
1322 | bundle->src_mirrors |= mirror_bit; | |
1323 | } else { | |
1324 | bundle->src_mirrors &= ~mirror_bit; | |
1325 | } | |
1326 | ||
1327 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
1328 | bundle->dst_mirrors |= mirror_bit; | |
1329 | } else { | |
1330 | bundle->dst_mirrors &= ~mirror_bit; | |
1331 | } | |
1332 | ||
1333 | if (mirror->out == bundle) { | |
1334 | bundle->mirror_out |= mirror_bit; | |
1335 | } else { | |
1336 | bundle->mirror_out &= ~mirror_bit; | |
1337 | } | |
1338 | } | |
1339 | ||
1340 | ofproto->need_revalidate = true; | |
1341 | mac_learning_flush(ofproto->ml); | |
1342 | ||
1343 | return 0; | |
1344 | } | |
1345 | ||
1346 | static void | |
1347 | mirror_destroy(struct ofmirror *mirror) | |
1348 | { | |
1349 | struct ofproto_dpif *ofproto; | |
1350 | mirror_mask_t mirror_bit; | |
1351 | struct ofbundle *bundle; | |
1352 | ||
1353 | if (!mirror) { | |
1354 | return; | |
1355 | } | |
1356 | ||
1357 | ofproto = mirror->ofproto; | |
1358 | ofproto->need_revalidate = true; | |
1359 | mac_learning_flush(ofproto->ml); | |
1360 | ||
1361 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1362 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1363 | bundle->src_mirrors &= ~mirror_bit; | |
1364 | bundle->dst_mirrors &= ~mirror_bit; | |
1365 | bundle->mirror_out &= ~mirror_bit; | |
1366 | } | |
1367 | ||
1368 | hmapx_destroy(&mirror->srcs); | |
1369 | hmapx_destroy(&mirror->dsts); | |
1370 | free(mirror->vlans); | |
1371 | ||
1372 | ofproto->mirrors[mirror->idx] = NULL; | |
1373 | free(mirror->name); | |
1374 | free(mirror); | |
1375 | } | |
1376 | ||
1377 | static int | |
1378 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
1379 | { | |
1380 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1381 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
1382 | ofproto->need_revalidate = true; | |
1383 | mac_learning_flush(ofproto->ml); | |
1384 | } | |
1385 | return 0; | |
1386 | } | |
1387 | ||
1388 | static bool | |
1389 | is_mirror_output_bundle(struct ofproto *ofproto_, void *aux) | |
1390 | { | |
1391 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1392 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
1393 | return bundle && bundle->mirror_out != 0; | |
1394 | } | |
1395 | \f | |
1396 | /* Ports. */ | |
1397 | ||
1398 | static struct ofport_dpif * | |
1399 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
1400 | { | |
7df6a8bd BP |
1401 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
1402 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
1403 | } |
1404 | ||
1405 | static struct ofport_dpif * | |
1406 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
1407 | { | |
1408 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
1409 | } | |
1410 | ||
1411 | static void | |
1412 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
1413 | struct dpif_port *dpif_port) | |
1414 | { | |
1415 | ofproto_port->name = dpif_port->name; | |
1416 | ofproto_port->type = dpif_port->type; | |
1417 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
1418 | } | |
1419 | ||
1420 | static void | |
1421 | port_run(struct ofport_dpif *ofport) | |
1422 | { | |
015e08bc EJ |
1423 | bool enable = netdev_get_carrier(ofport->up.netdev); |
1424 | ||
abe529af BP |
1425 | if (ofport->cfm) { |
1426 | cfm_run(ofport->cfm); | |
1427 | ||
1428 | if (cfm_should_send_ccm(ofport->cfm)) { | |
1429 | struct ofpbuf packet; | |
abe529af BP |
1430 | |
1431 | ofpbuf_init(&packet, 0); | |
c0a2e71d | 1432 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr); |
abe529af | 1433 | send_packet(ofproto_dpif_cast(ofport->up.ofproto), |
b2fda3ef | 1434 | ofport->odp_port, &packet); |
abe529af BP |
1435 | ofpbuf_uninit(&packet); |
1436 | } | |
015e08bc EJ |
1437 | |
1438 | enable = enable && !cfm_get_fault(ofport->cfm); | |
abe529af | 1439 | } |
015e08bc EJ |
1440 | |
1441 | if (ofport->bundle) { | |
1442 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
1443 | } | |
1444 | ||
daff3353 EJ |
1445 | if (ofport->may_enable != enable) { |
1446 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1447 | ||
1448 | if (ofproto->has_bundle_action) { | |
1449 | ofproto->need_revalidate = true; | |
1450 | } | |
1451 | } | |
1452 | ||
015e08bc | 1453 | ofport->may_enable = enable; |
abe529af BP |
1454 | } |
1455 | ||
1456 | static void | |
1457 | port_wait(struct ofport_dpif *ofport) | |
1458 | { | |
1459 | if (ofport->cfm) { | |
1460 | cfm_wait(ofport->cfm); | |
1461 | } | |
1462 | } | |
1463 | ||
1464 | static int | |
1465 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
1466 | struct ofproto_port *ofproto_port) | |
1467 | { | |
1468 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1469 | struct dpif_port dpif_port; | |
1470 | int error; | |
1471 | ||
1472 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
1473 | if (!error) { | |
1474 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
1475 | } | |
1476 | return error; | |
1477 | } | |
1478 | ||
1479 | static int | |
1480 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
1481 | { | |
1482 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1483 | uint16_t odp_port; | |
1484 | int error; | |
1485 | ||
1486 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
1487 | if (!error) { | |
1488 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
1489 | } | |
1490 | return error; | |
1491 | } | |
1492 | ||
1493 | static int | |
1494 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
1495 | { | |
1496 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1497 | int error; | |
1498 | ||
1499 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
1500 | if (!error) { | |
1501 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
1502 | if (ofport) { | |
1503 | /* The caller is going to close ofport->up.netdev. If this is a | |
1504 | * bonded port, then the bond is using that netdev, so remove it | |
1505 | * from the bond. The client will need to reconfigure everything | |
1506 | * after deleting ports, so then the slave will get re-added. */ | |
1507 | bundle_remove(&ofport->up); | |
1508 | } | |
1509 | } | |
1510 | return error; | |
1511 | } | |
1512 | ||
1513 | struct port_dump_state { | |
1514 | struct dpif_port_dump dump; | |
1515 | bool done; | |
1516 | }; | |
1517 | ||
1518 | static int | |
1519 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
1520 | { | |
1521 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1522 | struct port_dump_state *state; | |
1523 | ||
1524 | *statep = state = xmalloc(sizeof *state); | |
1525 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
1526 | state->done = false; | |
1527 | return 0; | |
1528 | } | |
1529 | ||
1530 | static int | |
1531 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
1532 | struct ofproto_port *port) | |
1533 | { | |
1534 | struct port_dump_state *state = state_; | |
1535 | struct dpif_port dpif_port; | |
1536 | ||
1537 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
1538 | ofproto_port_from_dpif_port(port, &dpif_port); | |
1539 | return 0; | |
1540 | } else { | |
1541 | int error = dpif_port_dump_done(&state->dump); | |
1542 | state->done = true; | |
1543 | return error ? error : EOF; | |
1544 | } | |
1545 | } | |
1546 | ||
1547 | static int | |
1548 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
1549 | { | |
1550 | struct port_dump_state *state = state_; | |
1551 | ||
1552 | if (!state->done) { | |
1553 | dpif_port_dump_done(&state->dump); | |
1554 | } | |
1555 | free(state); | |
1556 | return 0; | |
1557 | } | |
1558 | ||
1559 | static int | |
1560 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
1561 | { | |
1562 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1563 | return dpif_port_poll(ofproto->dpif, devnamep); | |
1564 | } | |
1565 | ||
1566 | static void | |
1567 | port_poll_wait(const struct ofproto *ofproto_) | |
1568 | { | |
1569 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1570 | dpif_port_poll_wait(ofproto->dpif); | |
1571 | } | |
1572 | ||
1573 | static int | |
1574 | port_is_lacp_current(const struct ofport *ofport_) | |
1575 | { | |
1576 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1577 | return (ofport->bundle && ofport->bundle->lacp | |
1578 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
1579 | : -1); | |
1580 | } | |
1581 | \f | |
1582 | /* Upcall handling. */ | |
1583 | ||
1584 | /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an | |
1585 | * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to | |
1586 | * their individual configurations. | |
1587 | * | |
1588 | * If 'clone' is true, the caller retains ownership of 'upcall->packet'. | |
1589 | * Otherwise, ownership is transferred to this function. */ | |
1590 | static void | |
1591 | send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall, | |
1592 | const struct flow *flow, bool clone) | |
1593 | { | |
1594 | struct ofputil_packet_in pin; | |
1595 | ||
1596 | pin.packet = upcall->packet; | |
1597 | pin.in_port = flow->in_port; | |
1598 | pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION; | |
1599 | pin.buffer_id = 0; /* not yet known */ | |
1600 | pin.send_len = upcall->userdata; | |
78bd1cd0 | 1601 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, |
abe529af BP |
1602 | clone ? NULL : upcall->packet); |
1603 | } | |
1604 | ||
1605 | static bool | |
1606 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
1607 | const struct ofpbuf *packet) | |
1608 | { | |
1609 | if (cfm_should_process_flow(flow)) { | |
1610 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); | |
da37ebac | 1611 | if (packet && ofport && ofport->cfm) { |
abe529af BP |
1612 | cfm_process_heartbeat(ofport->cfm, packet); |
1613 | } | |
1614 | return true; | |
1615 | } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { | |
1616 | struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); | |
da37ebac | 1617 | if (packet && port && port->bundle && port->bundle->lacp) { |
abe529af BP |
1618 | const struct lacp_pdu *pdu = parse_lacp_packet(packet); |
1619 | if (pdu) { | |
1620 | lacp_process_pdu(port->bundle->lacp, port, pdu); | |
1621 | } | |
abe529af | 1622 | } |
da37ebac | 1623 | return true; |
abe529af BP |
1624 | } |
1625 | return false; | |
1626 | } | |
1627 | ||
1628 | static void | |
1629 | handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) | |
1630 | { | |
1631 | struct facet *facet; | |
1632 | struct flow flow; | |
1633 | ||
1634 | /* Obtain in_port and tun_id, at least. */ | |
1635 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
1636 | ||
1637 | /* Set header pointers in 'flow'. */ | |
1638 | flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); | |
1639 | ||
1640 | /* Handle 802.1ag and LACP. */ | |
1641 | if (process_special(ofproto, &flow, upcall->packet)) { | |
1642 | ofpbuf_delete(upcall->packet); | |
6c1491fb | 1643 | ofproto->n_matches++; |
abe529af BP |
1644 | return; |
1645 | } | |
1646 | ||
1647 | /* Check with in-band control to see if this packet should be sent | |
1648 | * to the local port regardless of the flow table. */ | |
1649 | if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) { | |
f7f2ec05 | 1650 | send_packet(ofproto, ODPP_LOCAL, upcall->packet); |
abe529af BP |
1651 | } |
1652 | ||
1653 | facet = facet_lookup_valid(ofproto, &flow); | |
1654 | if (!facet) { | |
1655 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow); | |
1656 | if (!rule) { | |
1657 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
1658 | struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port); | |
1659 | if (port) { | |
1660 | if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) { | |
1661 | COVERAGE_INC(ofproto_dpif_no_packet_in); | |
1662 | /* XXX install 'drop' flow entry */ | |
1663 | ofpbuf_delete(upcall->packet); | |
1664 | return; | |
1665 | } | |
1666 | } else { | |
1667 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
1668 | flow.in_port); | |
1669 | } | |
1670 | ||
1671 | send_packet_in(ofproto, upcall, &flow, false); | |
1672 | return; | |
1673 | } | |
1674 | ||
1675 | facet = facet_create(rule, &flow, upcall->packet); | |
1676 | } else if (!facet->may_install) { | |
1677 | /* The facet is not installable, that is, we need to process every | |
1678 | * packet, so process the current packet's actions into 'facet'. */ | |
1679 | facet_make_actions(ofproto, facet, upcall->packet); | |
1680 | } | |
1681 | ||
1682 | if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
1683 | /* | |
1684 | * Extra-special case for fail-open mode. | |
1685 | * | |
1686 | * We are in fail-open mode and the packet matched the fail-open rule, | |
1687 | * but we are connected to a controller too. We should send the packet | |
1688 | * up to the controller in the hope that it will try to set up a flow | |
1689 | * and thereby allow us to exit fail-open. | |
1690 | * | |
1691 | * See the top-level comment in fail-open.c for more information. | |
1692 | */ | |
1693 | send_packet_in(ofproto, upcall, &flow, true); | |
1694 | } | |
1695 | ||
1696 | facet_execute(ofproto, facet, upcall->packet); | |
1697 | facet_install(ofproto, facet, false); | |
6c1491fb | 1698 | ofproto->n_matches++; |
abe529af BP |
1699 | } |
1700 | ||
1701 | static void | |
1702 | handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) | |
1703 | { | |
1704 | struct flow flow; | |
1705 | ||
1706 | switch (upcall->type) { | |
1707 | case DPIF_UC_ACTION: | |
1708 | COVERAGE_INC(ofproto_dpif_ctlr_action); | |
1709 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
1710 | send_packet_in(ofproto, upcall, &flow, false); | |
1711 | break; | |
1712 | ||
1713 | case DPIF_UC_SAMPLE: | |
1714 | if (ofproto->sflow) { | |
1715 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
bae473fe | 1716 | dpif_sflow_received(ofproto->sflow, upcall, &flow); |
abe529af BP |
1717 | } |
1718 | ofpbuf_delete(upcall->packet); | |
1719 | break; | |
1720 | ||
1721 | case DPIF_UC_MISS: | |
1722 | handle_miss_upcall(ofproto, upcall); | |
1723 | break; | |
1724 | ||
1725 | case DPIF_N_UC_TYPES: | |
1726 | default: | |
1727 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
1728 | break; | |
1729 | } | |
1730 | } | |
1731 | \f | |
1732 | /* Flow expiration. */ | |
1733 | ||
1734 | static int facet_max_idle(const struct ofproto_dpif *); | |
1735 | static void update_stats(struct ofproto_dpif *); | |
1736 | static void rule_expire(struct rule_dpif *); | |
1737 | static void expire_facets(struct ofproto_dpif *, int dp_max_idle); | |
1738 | ||
1739 | /* This function is called periodically by run(). Its job is to collect | |
1740 | * updates for the flows that have been installed into the datapath, most | |
1741 | * importantly when they last were used, and then use that information to | |
1742 | * expire flows that have not been used recently. | |
1743 | * | |
1744 | * Returns the number of milliseconds after which it should be called again. */ | |
1745 | static int | |
1746 | expire(struct ofproto_dpif *ofproto) | |
1747 | { | |
1748 | struct rule_dpif *rule, *next_rule; | |
1749 | struct cls_cursor cursor; | |
1750 | int dp_max_idle; | |
1751 | ||
1752 | /* Update stats for each flow in the datapath. */ | |
1753 | update_stats(ofproto); | |
1754 | ||
1755 | /* Expire facets that have been idle too long. */ | |
1756 | dp_max_idle = facet_max_idle(ofproto); | |
1757 | expire_facets(ofproto, dp_max_idle); | |
1758 | ||
1759 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
6c1491fb | 1760 | cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL); |
abe529af BP |
1761 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
1762 | rule_expire(rule); | |
1763 | } | |
1764 | ||
1765 | /* All outstanding data in existing flows has been accounted, so it's a | |
1766 | * good time to do bond rebalancing. */ | |
1767 | if (ofproto->has_bonded_bundles) { | |
1768 | struct ofbundle *bundle; | |
1769 | ||
1770 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1771 | if (bundle->bond) { | |
1772 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
1773 | } | |
1774 | } | |
1775 | } | |
1776 | ||
1777 | return MIN(dp_max_idle, 1000); | |
1778 | } | |
1779 | ||
1780 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
1781 | * | |
1782 | * This function also pushes statistics updates to rules which each facet | |
1783 | * resubmits into. Generally these statistics will be accurate. However, if a | |
1784 | * facet changes the rule it resubmits into at some time in between | |
1785 | * update_stats() runs, it is possible that statistics accrued to the | |
1786 | * old rule will be incorrectly attributed to the new rule. This could be | |
1787 | * avoided by calling update_stats() whenever rules are created or | |
1788 | * deleted. However, the performance impact of making so many calls to the | |
1789 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
1790 | */ | |
1791 | static void | |
1792 | update_stats(struct ofproto_dpif *p) | |
1793 | { | |
1794 | const struct dpif_flow_stats *stats; | |
1795 | struct dpif_flow_dump dump; | |
1796 | const struct nlattr *key; | |
1797 | size_t key_len; | |
1798 | ||
1799 | dpif_flow_dump_start(&dump, p->dpif); | |
1800 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
1801 | struct facet *facet; | |
1802 | struct flow flow; | |
1803 | ||
1804 | if (odp_flow_key_to_flow(key, key_len, &flow)) { | |
1805 | struct ds s; | |
1806 | ||
1807 | ds_init(&s); | |
1808 | odp_flow_key_format(key, key_len, &s); | |
1809 | VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s", | |
1810 | ds_cstr(&s)); | |
1811 | ds_destroy(&s); | |
1812 | ||
1813 | continue; | |
1814 | } | |
1815 | facet = facet_find(p, &flow); | |
1816 | ||
1817 | if (facet && facet->installed) { | |
1818 | ||
1819 | if (stats->n_packets >= facet->dp_packet_count) { | |
1820 | uint64_t extra = stats->n_packets - facet->dp_packet_count; | |
1821 | facet->packet_count += extra; | |
1822 | } else { | |
1823 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
1824 | } | |
1825 | ||
1826 | if (stats->n_bytes >= facet->dp_byte_count) { | |
1827 | facet->byte_count += stats->n_bytes - facet->dp_byte_count; | |
1828 | } else { | |
1829 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
1830 | } | |
1831 | ||
1832 | facet->dp_packet_count = stats->n_packets; | |
1833 | facet->dp_byte_count = stats->n_bytes; | |
1834 | ||
1835 | facet_update_time(p, facet, stats->used); | |
1836 | facet_account(p, facet, stats->n_bytes); | |
1837 | facet_push_stats(facet); | |
1838 | } else { | |
1839 | /* There's a flow in the datapath that we know nothing about. | |
1840 | * Delete it. */ | |
1841 | COVERAGE_INC(facet_unexpected); | |
1842 | dpif_flow_del(p->dpif, key, key_len, NULL); | |
1843 | } | |
1844 | } | |
1845 | dpif_flow_dump_done(&dump); | |
1846 | } | |
1847 | ||
1848 | /* Calculates and returns the number of milliseconds of idle time after which | |
1849 | * facets should expire from the datapath and we should fold their statistics | |
1850 | * into their parent rules in userspace. */ | |
1851 | static int | |
1852 | facet_max_idle(const struct ofproto_dpif *ofproto) | |
1853 | { | |
1854 | /* | |
1855 | * Idle time histogram. | |
1856 | * | |
1857 | * Most of the time a switch has a relatively small number of facets. When | |
1858 | * this is the case we might as well keep statistics for all of them in | |
1859 | * userspace and to cache them in the kernel datapath for performance as | |
1860 | * well. | |
1861 | * | |
1862 | * As the number of facets increases, the memory required to maintain | |
1863 | * statistics about them in userspace and in the kernel becomes | |
1864 | * significant. However, with a large number of facets it is likely that | |
1865 | * only a few of them are "heavy hitters" that consume a large amount of | |
1866 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
1867 | * kernel and maintaining in userspaces; other facets we can discard. | |
1868 | * | |
1869 | * The technique used to compute the idle time is to build a histogram with | |
1870 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet | |
1871 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
1872 | * After the histogram has been built, we compute the cutoff so that only | |
1873 | * the most-recently-used 1% of facets (but at least 1000 flows) are kept | |
1874 | * cached. At least the most-recently-used bucket of facets is kept, so | |
1875 | * actually an arbitrary number of facets can be kept in any given | |
1876 | * expiration run (though the next run will delete most of those unless | |
1877 | * they receive additional data). | |
1878 | * | |
1879 | * This requires a second pass through the facets, in addition to the pass | |
1880 | * made by update_stats(), because the former function never looks | |
1881 | * at uninstallable facets. | |
1882 | */ | |
1883 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
1884 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
1885 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 1886 | int total, subtotal, bucket; |
abe529af | 1887 | struct facet *facet; |
abe529af BP |
1888 | long long int now; |
1889 | int i; | |
1890 | ||
1891 | total = hmap_count(&ofproto->facets); | |
1892 | if (total <= 1000) { | |
1893 | return N_BUCKETS * BUCKET_WIDTH; | |
1894 | } | |
1895 | ||
1896 | /* Build histogram. */ | |
1897 | now = time_msec(); | |
1898 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
1899 | long long int idle = now - facet->used; | |
1900 | int bucket = (idle <= 0 ? 0 | |
1901 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
1902 | : (unsigned int) idle / BUCKET_WIDTH); | |
1903 | buckets[bucket]++; | |
1904 | } | |
1905 | ||
1906 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
1907 | subtotal = bucket = 0; |
1908 | do { | |
1909 | subtotal += buckets[bucket++]; | |
1910 | } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100)); | |
abe529af BP |
1911 | |
1912 | if (VLOG_IS_DBG_ENABLED()) { | |
1913 | struct ds s; | |
1914 | ||
1915 | ds_init(&s); | |
1916 | ds_put_cstr(&s, "keep"); | |
1917 | for (i = 0; i < N_BUCKETS; i++) { | |
1918 | if (i == bucket) { | |
1919 | ds_put_cstr(&s, ", drop"); | |
1920 | } | |
1921 | if (buckets[i]) { | |
1922 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
1923 | } | |
1924 | } | |
1925 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
1926 | ds_destroy(&s); | |
1927 | } | |
1928 | ||
1929 | return bucket * BUCKET_WIDTH; | |
1930 | } | |
1931 | ||
1932 | static void | |
1933 | facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
1934 | { | |
1935 | if (ofproto->netflow && !facet_is_controller_flow(facet) && | |
1936 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
1937 | struct ofexpired expired; | |
1938 | ||
1939 | if (facet->installed) { | |
1940 | struct dpif_flow_stats stats; | |
1941 | ||
1942 | facet_put__(ofproto, facet, facet->actions, facet->actions_len, | |
1943 | &stats); | |
1944 | facet_update_stats(ofproto, facet, &stats); | |
1945 | } | |
1946 | ||
1947 | expired.flow = facet->flow; | |
1948 | expired.packet_count = facet->packet_count; | |
1949 | expired.byte_count = facet->byte_count; | |
1950 | expired.used = facet->used; | |
1951 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
1952 | } | |
1953 | } | |
1954 | ||
1955 | static void | |
1956 | expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle) | |
1957 | { | |
1958 | long long int cutoff = time_msec() - dp_max_idle; | |
1959 | struct facet *facet, *next_facet; | |
1960 | ||
1961 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1962 | facet_active_timeout(ofproto, facet); | |
1963 | if (facet->used < cutoff) { | |
1964 | facet_remove(ofproto, facet); | |
1965 | } | |
1966 | } | |
1967 | } | |
1968 | ||
1969 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
1970 | * then delete it entirely. */ | |
1971 | static void | |
1972 | rule_expire(struct rule_dpif *rule) | |
1973 | { | |
1974 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
1975 | struct facet *facet, *next_facet; | |
1976 | long long int now; | |
1977 | uint8_t reason; | |
1978 | ||
1979 | /* Has 'rule' expired? */ | |
1980 | now = time_msec(); | |
1981 | if (rule->up.hard_timeout | |
1982 | && now > rule->up.created + rule->up.hard_timeout * 1000) { | |
1983 | reason = OFPRR_HARD_TIMEOUT; | |
1984 | } else if (rule->up.idle_timeout && list_is_empty(&rule->facets) | |
1985 | && now > rule->used + rule->up.idle_timeout * 1000) { | |
1986 | reason = OFPRR_IDLE_TIMEOUT; | |
1987 | } else { | |
1988 | return; | |
1989 | } | |
1990 | ||
1991 | COVERAGE_INC(ofproto_dpif_expired); | |
1992 | ||
1993 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
1994 | * timeout, because that only happens when the rule has no facets left.) */ | |
1995 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
1996 | facet_remove(ofproto, facet); | |
1997 | } | |
1998 | ||
1999 | /* Get rid of the rule. */ | |
2000 | ofproto_rule_expire(&rule->up, reason); | |
2001 | } | |
2002 | \f | |
2003 | /* Facets. */ | |
2004 | ||
2005 | /* Creates and returns a new facet owned by 'rule', given a 'flow' and an | |
2006 | * example 'packet' within that flow. | |
2007 | * | |
2008 | * The caller must already have determined that no facet with an identical | |
2009 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
2010 | * the ofproto's classifier table. */ | |
2011 | static struct facet * | |
2012 | facet_create(struct rule_dpif *rule, const struct flow *flow, | |
2013 | const struct ofpbuf *packet) | |
2014 | { | |
2015 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2016 | struct facet *facet; | |
2017 | ||
2018 | facet = xzalloc(sizeof *facet); | |
2019 | facet->used = time_msec(); | |
2020 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2021 | list_push_back(&rule->facets, &facet->list_node); | |
2022 | facet->rule = rule; | |
2023 | facet->flow = *flow; | |
2024 | netflow_flow_init(&facet->nf_flow); | |
2025 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2026 | ||
2027 | facet_make_actions(ofproto, facet, packet); | |
2028 | ||
2029 | return facet; | |
2030 | } | |
2031 | ||
2032 | static void | |
2033 | facet_free(struct facet *facet) | |
2034 | { | |
2035 | free(facet->actions); | |
2036 | free(facet); | |
2037 | } | |
2038 | ||
2039 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on | |
2040 | * 'packet', which arrived on 'in_port'. | |
2041 | * | |
2042 | * Takes ownership of 'packet'. */ | |
2043 | static bool | |
2044 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2045 | const struct nlattr *odp_actions, size_t actions_len, | |
2046 | struct ofpbuf *packet) | |
2047 | { | |
2048 | if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) | |
2049 | && odp_actions->nla_type == ODP_ACTION_ATTR_CONTROLLER) { | |
2050 | /* As an optimization, avoid a round-trip from userspace to kernel to | |
2051 | * userspace. This also avoids possibly filling up kernel packet | |
2052 | * buffers along the way. */ | |
2053 | struct dpif_upcall upcall; | |
2054 | ||
2055 | upcall.type = DPIF_UC_ACTION; | |
2056 | upcall.packet = packet; | |
2057 | upcall.key = NULL; | |
2058 | upcall.key_len = 0; | |
2059 | upcall.userdata = nl_attr_get_u64(odp_actions); | |
2060 | upcall.sample_pool = 0; | |
2061 | upcall.actions = NULL; | |
2062 | upcall.actions_len = 0; | |
2063 | ||
2064 | send_packet_in(ofproto, &upcall, flow, false); | |
2065 | ||
2066 | return true; | |
2067 | } else { | |
80e5eed9 BP |
2068 | struct odputil_keybuf keybuf; |
2069 | struct ofpbuf key; | |
abe529af BP |
2070 | int error; |
2071 | ||
80e5eed9 BP |
2072 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
2073 | odp_flow_key_from_flow(&key, flow); | |
2074 | ||
2075 | error = dpif_execute(ofproto->dpif, key.data, key.size, | |
2076 | odp_actions, actions_len, packet); | |
2077 | ||
abe529af BP |
2078 | ofpbuf_delete(packet); |
2079 | return !error; | |
2080 | } | |
2081 | } | |
2082 | ||
2083 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s | |
2084 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
2085 | * ofp_packet_in) bytes of headroom. | |
2086 | * | |
2087 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, | |
2088 | * applying flow_extract() to 'packet' would yield the same flow as | |
2089 | * 'facet->flow'. | |
2090 | * | |
2091 | * 'facet' must have accurately composed ODP actions; that is, it must not be | |
2092 | * in need of revalidation. | |
2093 | * | |
2094 | * Takes ownership of 'packet'. */ | |
2095 | static void | |
2096 | facet_execute(struct ofproto_dpif *ofproto, struct facet *facet, | |
2097 | struct ofpbuf *packet) | |
2098 | { | |
2099 | struct dpif_flow_stats stats; | |
2100 | ||
2101 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2102 | ||
2103 | flow_extract_stats(&facet->flow, packet, &stats); | |
2104 | stats.used = time_msec(); | |
2105 | if (execute_odp_actions(ofproto, &facet->flow, | |
2106 | facet->actions, facet->actions_len, packet)) { | |
2107 | facet_update_stats(ofproto, facet, &stats); | |
2108 | } | |
2109 | } | |
2110 | ||
2111 | /* Remove 'facet' from 'ofproto' and free up the associated memory: | |
2112 | * | |
2113 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2114 | * rule's statistics, via facet_uninstall(). | |
2115 | * | |
2116 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2117 | */ | |
2118 | static void | |
2119 | facet_remove(struct ofproto_dpif *ofproto, struct facet *facet) | |
2120 | { | |
2121 | facet_uninstall(ofproto, facet); | |
2122 | facet_flush_stats(ofproto, facet); | |
2123 | hmap_remove(&ofproto->facets, &facet->hmap_node); | |
2124 | list_remove(&facet->list_node); | |
2125 | facet_free(facet); | |
2126 | } | |
2127 | ||
2128 | /* Composes the ODP actions for 'facet' based on its rule's actions. */ | |
2129 | static void | |
2130 | facet_make_actions(struct ofproto_dpif *p, struct facet *facet, | |
2131 | const struct ofpbuf *packet) | |
2132 | { | |
2133 | const struct rule_dpif *rule = facet->rule; | |
2134 | struct ofpbuf *odp_actions; | |
2135 | struct action_xlate_ctx ctx; | |
2136 | ||
2137 | action_xlate_ctx_init(&ctx, p, &facet->flow, packet); | |
2138 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
2139 | facet->tags = ctx.tags; | |
2140 | facet->may_install = ctx.may_set_up_flow; | |
2141 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
2142 | ||
2143 | if (facet->actions_len != odp_actions->size | |
2144 | || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { | |
2145 | free(facet->actions); | |
2146 | facet->actions_len = odp_actions->size; | |
2147 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
2148 | } | |
2149 | ||
2150 | ofpbuf_delete(odp_actions); | |
2151 | } | |
2152 | ||
3a88e544 BP |
2153 | /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len' |
2154 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
2155 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
2156 | * since 'facet' was last updated. | |
2157 | * | |
2158 | * Returns 0 if successful, otherwise a positive errno value.*/ | |
abe529af BP |
2159 | static int |
2160 | facet_put__(struct ofproto_dpif *ofproto, struct facet *facet, | |
2161 | const struct nlattr *actions, size_t actions_len, | |
2162 | struct dpif_flow_stats *stats) | |
2163 | { | |
2164 | struct odputil_keybuf keybuf; | |
2165 | enum dpif_flow_put_flags flags; | |
2166 | struct ofpbuf key; | |
3a88e544 | 2167 | int ret; |
abe529af BP |
2168 | |
2169 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
2170 | if (stats) { | |
2171 | flags |= DPIF_FP_ZERO_STATS; | |
abe529af BP |
2172 | } |
2173 | ||
2174 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2175 | odp_flow_key_from_flow(&key, &facet->flow); | |
2176 | ||
3a88e544 BP |
2177 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, |
2178 | actions, actions_len, stats); | |
2179 | ||
2180 | if (stats) { | |
2181 | facet_reset_dp_stats(facet, stats); | |
2182 | } | |
2183 | ||
2184 | return ret; | |
abe529af BP |
2185 | } |
2186 | ||
2187 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If | |
2188 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2189 | * 'facet'. */ | |
2190 | static void | |
2191 | facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats) | |
2192 | { | |
2193 | struct dpif_flow_stats stats; | |
2194 | ||
2195 | if (facet->may_install | |
2196 | && !facet_put__(p, facet, facet->actions, facet->actions_len, | |
2197 | zero_stats ? &stats : NULL)) { | |
2198 | facet->installed = true; | |
2199 | } | |
2200 | } | |
2201 | ||
d78be13b BP |
2202 | static int |
2203 | vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci) | |
2204 | { | |
2205 | return vlan_tci != htons(0) ? vlan_tci_to_vid(vlan_tci) : OFP_VLAN_NONE; | |
2206 | } | |
2207 | ||
abe529af BP |
2208 | static void |
2209 | facet_account(struct ofproto_dpif *ofproto, | |
2210 | struct facet *facet, uint64_t extra_bytes) | |
2211 | { | |
2212 | uint64_t total_bytes, n_bytes; | |
2213 | struct ofbundle *in_bundle; | |
2214 | const struct nlattr *a; | |
2215 | tag_type dummy = 0; | |
2216 | unsigned int left; | |
d78be13b | 2217 | ovs_be16 vlan_tci; |
abe529af BP |
2218 | int vlan; |
2219 | ||
2220 | total_bytes = facet->byte_count + extra_bytes; | |
2221 | if (total_bytes <= facet->accounted_bytes) { | |
2222 | return; | |
2223 | } | |
2224 | n_bytes = total_bytes - facet->accounted_bytes; | |
2225 | facet->accounted_bytes = total_bytes; | |
2226 | ||
2227 | /* Test that 'tags' is nonzero to ensure that only flows that include an | |
2228 | * OFPP_NORMAL action are used for learning and bond slave rebalancing. | |
2229 | * This works because OFPP_NORMAL always sets a nonzero tag value. | |
2230 | * | |
2231 | * Feed information from the active flows back into the learning table to | |
2232 | * ensure that table is always in sync with what is actually flowing | |
2233 | * through the datapath. */ | |
2234 | if (!facet->tags | |
2235 | || !is_admissible(ofproto, &facet->flow, false, &dummy, | |
2236 | &vlan, &in_bundle)) { | |
2237 | return; | |
2238 | } | |
2239 | ||
2240 | update_learning_table(ofproto, &facet->flow, vlan, in_bundle); | |
2241 | ||
2242 | if (!ofproto->has_bonded_bundles) { | |
2243 | return; | |
2244 | } | |
d78be13b BP |
2245 | |
2246 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
2247 | * as a basis. We also need to track the actual VLAN on which the packet | |
2248 | * is going to be sent to ensure that it matches the one passed to | |
2249 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
2250 | * hash bucket.) */ | |
2251 | vlan_tci = facet->flow.vlan_tci; | |
abe529af | 2252 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) { |
d78be13b | 2253 | struct ofport_dpif *port; |
abe529af | 2254 | |
d78be13b BP |
2255 | switch (nl_attr_type(a)) { |
2256 | case ODP_ACTION_ATTR_OUTPUT: | |
abe529af BP |
2257 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
2258 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b BP |
2259 | bond_account(port->bundle->bond, &facet->flow, |
2260 | vlan_tci_to_openflow_vlan(vlan_tci), n_bytes); | |
abe529af | 2261 | } |
d78be13b BP |
2262 | break; |
2263 | ||
2264 | case ODP_ACTION_ATTR_STRIP_VLAN: | |
2265 | vlan_tci = htons(0); | |
2266 | break; | |
2267 | ||
2268 | case ODP_ACTION_ATTR_SET_DL_TCI: | |
2269 | vlan_tci = nl_attr_get_be16(a); | |
2270 | break; | |
abe529af BP |
2271 | } |
2272 | } | |
2273 | } | |
2274 | ||
2275 | /* If 'rule' is installed in the datapath, uninstalls it. */ | |
2276 | static void | |
2277 | facet_uninstall(struct ofproto_dpif *p, struct facet *facet) | |
2278 | { | |
2279 | if (facet->installed) { | |
2280 | struct odputil_keybuf keybuf; | |
2281 | struct dpif_flow_stats stats; | |
2282 | struct ofpbuf key; | |
3a88e544 | 2283 | int error; |
abe529af BP |
2284 | |
2285 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2286 | odp_flow_key_from_flow(&key, &facet->flow); | |
2287 | ||
3a88e544 BP |
2288 | error = dpif_flow_del(p->dpif, key.data, key.size, &stats); |
2289 | facet_reset_dp_stats(facet, &stats); | |
2290 | if (!error) { | |
abe529af BP |
2291 | facet_update_stats(p, facet, &stats); |
2292 | } | |
2293 | facet->installed = false; | |
abe529af BP |
2294 | } else { |
2295 | assert(facet->dp_packet_count == 0); | |
2296 | assert(facet->dp_byte_count == 0); | |
2297 | } | |
2298 | } | |
2299 | ||
2300 | /* Returns true if the only action for 'facet' is to send to the controller. | |
2301 | * (We don't report NetFlow expiration messages for such facets because they | |
2302 | * are just part of the control logic for the network, not real traffic). */ | |
2303 | static bool | |
2304 | facet_is_controller_flow(struct facet *facet) | |
2305 | { | |
2306 | return (facet | |
2307 | && facet->rule->up.n_actions == 1 | |
2308 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
2309 | htons(OFPP_CONTROLLER))); | |
2310 | } | |
2311 | ||
3a88e544 BP |
2312 | /* Resets 'facet''s datapath statistics counters. This should be called when |
2313 | * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null, | |
2314 | * it should contain the statistics returned by dpif when 'facet' was reset in | |
2315 | * the datapath. 'stats' will be modified to only included statistics new | |
2316 | * since 'facet' was last updated. */ | |
2317 | static void | |
2318 | facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats) | |
2319 | { | |
2320 | if (stats && facet->dp_packet_count <= stats->n_packets | |
2321 | && facet->dp_byte_count <= stats->n_bytes) { | |
2322 | stats->n_packets -= facet->dp_packet_count; | |
2323 | stats->n_bytes -= facet->dp_byte_count; | |
2324 | } | |
2325 | ||
2326 | facet->dp_packet_count = 0; | |
2327 | facet->dp_byte_count = 0; | |
2328 | } | |
2329 | ||
abe529af BP |
2330 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
2331 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
2332 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
2333 | * its packet and byte counts before this function is called. */ | |
2334 | static void | |
2335 | facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet) | |
2336 | { | |
2337 | assert(!facet->dp_byte_count); | |
2338 | assert(!facet->dp_packet_count); | |
2339 | ||
2340 | facet_push_stats(facet); | |
2341 | facet_account(ofproto, facet, 0); | |
2342 | ||
2343 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
2344 | struct ofexpired expired; | |
2345 | expired.flow = facet->flow; | |
2346 | expired.packet_count = facet->packet_count; | |
2347 | expired.byte_count = facet->byte_count; | |
2348 | expired.used = facet->used; | |
2349 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2350 | } | |
2351 | ||
2352 | facet->rule->packet_count += facet->packet_count; | |
2353 | facet->rule->byte_count += facet->byte_count; | |
2354 | ||
2355 | /* Reset counters to prevent double counting if 'facet' ever gets | |
2356 | * reinstalled. */ | |
2357 | facet->packet_count = 0; | |
2358 | facet->byte_count = 0; | |
2359 | facet->rs_packet_count = 0; | |
2360 | facet->rs_byte_count = 0; | |
2361 | facet->accounted_bytes = 0; | |
2362 | ||
2363 | netflow_flow_clear(&facet->nf_flow); | |
2364 | } | |
2365 | ||
2366 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2367 | * Returns it if found, otherwise a null pointer. | |
2368 | * | |
2369 | * The returned facet might need revalidation; use facet_lookup_valid() | |
2370 | * instead if that is important. */ | |
2371 | static struct facet * | |
2372 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) | |
2373 | { | |
2374 | struct facet *facet; | |
2375 | ||
2376 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
2377 | &ofproto->facets) { | |
2378 | if (flow_equal(flow, &facet->flow)) { | |
2379 | return facet; | |
2380 | } | |
2381 | } | |
2382 | ||
2383 | return NULL; | |
2384 | } | |
2385 | ||
2386 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2387 | * Returns it if found, otherwise a null pointer. | |
2388 | * | |
2389 | * The returned facet is guaranteed to be valid. */ | |
2390 | static struct facet * | |
2391 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) | |
2392 | { | |
2393 | struct facet *facet = facet_find(ofproto, flow); | |
2394 | ||
2395 | /* The facet we found might not be valid, since we could be in need of | |
2396 | * revalidation. If it is not valid, don't return it. */ | |
2397 | if (facet | |
2398 | && ofproto->need_revalidate | |
2399 | && !facet_revalidate(ofproto, facet)) { | |
2400 | COVERAGE_INC(facet_invalidated); | |
2401 | return NULL; | |
2402 | } | |
2403 | ||
2404 | return facet; | |
2405 | } | |
2406 | ||
2407 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': | |
2408 | * | |
2409 | * - If the rule found is different from 'facet''s current rule, moves | |
2410 | * 'facet' to the new rule and recompiles its actions. | |
2411 | * | |
2412 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
2413 | * where it is and recompiles its actions anyway. | |
2414 | * | |
2415 | * - If there is none, destroys 'facet'. | |
2416 | * | |
2417 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
2418 | static bool | |
2419 | facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet) | |
2420 | { | |
2421 | struct action_xlate_ctx ctx; | |
2422 | struct ofpbuf *odp_actions; | |
2423 | struct rule_dpif *new_rule; | |
2424 | bool actions_changed; | |
2425 | ||
2426 | COVERAGE_INC(facet_revalidate); | |
2427 | ||
2428 | /* Determine the new rule. */ | |
2429 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); | |
2430 | if (!new_rule) { | |
2431 | /* No new rule, so delete the facet. */ | |
2432 | facet_remove(ofproto, facet); | |
2433 | return false; | |
2434 | } | |
2435 | ||
2436 | /* Calculate new ODP actions. | |
2437 | * | |
2438 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
2439 | * emit a NetFlow expiration and, if so, we need to have the old state | |
2440 | * around to properly compose it. */ | |
2441 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); | |
2442 | odp_actions = xlate_actions(&ctx, | |
2443 | new_rule->up.actions, new_rule->up.n_actions); | |
2444 | actions_changed = (facet->actions_len != odp_actions->size | |
2445 | || memcmp(facet->actions, odp_actions->data, | |
2446 | facet->actions_len)); | |
2447 | ||
2448 | /* If the ODP actions changed or the installability changed, then we need | |
2449 | * to talk to the datapath. */ | |
2450 | if (actions_changed || ctx.may_set_up_flow != facet->installed) { | |
2451 | if (ctx.may_set_up_flow) { | |
2452 | struct dpif_flow_stats stats; | |
2453 | ||
2454 | facet_put__(ofproto, facet, | |
2455 | odp_actions->data, odp_actions->size, &stats); | |
2456 | facet_update_stats(ofproto, facet, &stats); | |
2457 | } else { | |
2458 | facet_uninstall(ofproto, facet); | |
2459 | } | |
2460 | ||
2461 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
2462 | * 'facet' into 'rule'. */ | |
2463 | facet_flush_stats(ofproto, facet); | |
2464 | } | |
2465 | ||
2466 | /* Update 'facet' now that we've taken care of all the old state. */ | |
2467 | facet->tags = ctx.tags; | |
2468 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
2469 | facet->may_install = ctx.may_set_up_flow; | |
2470 | if (actions_changed) { | |
2471 | free(facet->actions); | |
2472 | facet->actions_len = odp_actions->size; | |
2473 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
2474 | } | |
2475 | if (facet->rule != new_rule) { | |
2476 | COVERAGE_INC(facet_changed_rule); | |
2477 | list_remove(&facet->list_node); | |
2478 | list_push_back(&new_rule->facets, &facet->list_node); | |
2479 | facet->rule = new_rule; | |
2480 | facet->used = new_rule->up.created; | |
2481 | facet->rs_used = facet->used; | |
2482 | } | |
2483 | ||
2484 | ofpbuf_delete(odp_actions); | |
2485 | ||
2486 | return true; | |
2487 | } | |
2488 | ||
2489 | /* Updates 'facet''s used time. Caller is responsible for calling | |
2490 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
2491 | static void | |
2492 | facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet, | |
2493 | long long int used) | |
2494 | { | |
2495 | if (used > facet->used) { | |
2496 | facet->used = used; | |
2497 | if (used > facet->rule->used) { | |
2498 | facet->rule->used = used; | |
2499 | } | |
2500 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
2501 | } | |
2502 | } | |
2503 | ||
2504 | /* Folds the statistics from 'stats' into the counters in 'facet'. | |
2505 | * | |
2506 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
2507 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
2508 | * packet that was sent by hand or if it represents statistics that have been | |
2509 | * cleared out of the datapath. */ | |
2510 | static void | |
2511 | facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet, | |
2512 | const struct dpif_flow_stats *stats) | |
2513 | { | |
2514 | if (stats->n_packets || stats->used > facet->used) { | |
2515 | facet_update_time(ofproto, facet, stats->used); | |
2516 | facet->packet_count += stats->n_packets; | |
2517 | facet->byte_count += stats->n_bytes; | |
2518 | facet_push_stats(facet); | |
2519 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
2520 | } | |
2521 | } | |
2522 | ||
2523 | static void | |
2524 | facet_push_stats(struct facet *facet) | |
2525 | { | |
2526 | uint64_t rs_packets, rs_bytes; | |
2527 | ||
2528 | assert(facet->packet_count >= facet->rs_packet_count); | |
2529 | assert(facet->byte_count >= facet->rs_byte_count); | |
2530 | assert(facet->used >= facet->rs_used); | |
2531 | ||
2532 | rs_packets = facet->packet_count - facet->rs_packet_count; | |
2533 | rs_bytes = facet->byte_count - facet->rs_byte_count; | |
2534 | ||
2535 | if (rs_packets || rs_bytes || facet->used > facet->rs_used) { | |
2536 | facet->rs_packet_count = facet->packet_count; | |
2537 | facet->rs_byte_count = facet->byte_count; | |
2538 | facet->rs_used = facet->used; | |
2539 | ||
2540 | flow_push_stats(facet->rule, &facet->flow, | |
2541 | rs_packets, rs_bytes, facet->used); | |
2542 | } | |
2543 | } | |
2544 | ||
2545 | struct ofproto_push { | |
2546 | struct action_xlate_ctx ctx; | |
2547 | uint64_t packets; | |
2548 | uint64_t bytes; | |
2549 | long long int used; | |
2550 | }; | |
2551 | ||
2552 | static void | |
2553 | push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
2554 | { | |
2555 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
2556 | ||
2557 | if (rule) { | |
2558 | rule->packet_count += push->packets; | |
2559 | rule->byte_count += push->bytes; | |
2560 | rule->used = MAX(push->used, rule->used); | |
2561 | } | |
2562 | } | |
2563 | ||
2564 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
2565 | * 'rule''s actions. */ | |
2566 | static void | |
2567 | flow_push_stats(const struct rule_dpif *rule, | |
2568 | struct flow *flow, uint64_t packets, uint64_t bytes, | |
2569 | long long int used) | |
2570 | { | |
2571 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2572 | struct ofproto_push push; | |
2573 | ||
2574 | push.packets = packets; | |
2575 | push.bytes = bytes; | |
2576 | push.used = used; | |
2577 | ||
2578 | action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); | |
2579 | push.ctx.resubmit_hook = push_resubmit; | |
2580 | ofpbuf_delete(xlate_actions(&push.ctx, | |
2581 | rule->up.actions, rule->up.n_actions)); | |
2582 | } | |
2583 | \f | |
2584 | /* Rules. */ | |
2585 | ||
2586 | static struct rule_dpif * | |
2587 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) | |
2588 | { | |
154896e3 | 2589 | return rule_dpif_cast(rule_from_cls_rule( |
6c1491fb BP |
2590 | classifier_lookup(&ofproto->up.tables[0], |
2591 | flow))); | |
abe529af BP |
2592 | } |
2593 | ||
7ee20df1 BP |
2594 | static void |
2595 | complete_operation(struct rule_dpif *rule) | |
2596 | { | |
2597 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2598 | ||
2599 | ofproto->need_revalidate = true; | |
2600 | if (clogged) { | |
2601 | struct dpif_completion *c = xmalloc(sizeof *c); | |
2602 | c->op = rule->up.pending; | |
2603 | list_push_back(&ofproto->completions, &c->list_node); | |
2604 | } else { | |
2605 | ofoperation_complete(rule->up.pending, 0); | |
2606 | } | |
2607 | } | |
2608 | ||
abe529af BP |
2609 | static struct rule * |
2610 | rule_alloc(void) | |
2611 | { | |
2612 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
2613 | return &rule->up; | |
2614 | } | |
2615 | ||
2616 | static void | |
2617 | rule_dealloc(struct rule *rule_) | |
2618 | { | |
2619 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2620 | free(rule); | |
2621 | } | |
2622 | ||
2623 | static int | |
2624 | rule_construct(struct rule *rule_) | |
2625 | { | |
2626 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2627 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 2628 | struct rule_dpif *victim; |
5bf0e941 BP |
2629 | int error; |
2630 | ||
2631 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
2632 | &rule->up.cr.flow, ofproto->max_ports); | |
2633 | if (error) { | |
2634 | return error; | |
2635 | } | |
abe529af BP |
2636 | |
2637 | rule->used = rule->up.created; | |
2638 | rule->packet_count = 0; | |
2639 | rule->byte_count = 0; | |
abe529af | 2640 | |
7ee20df1 BP |
2641 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
2642 | if (victim && !list_is_empty(&victim->facets)) { | |
2643 | struct facet *facet; | |
2644 | ||
2645 | rule->facets = victim->facets; | |
2646 | list_moved(&rule->facets); | |
2647 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
2648 | facet->rule = rule; | |
2649 | } | |
2650 | } else { | |
2651 | /* Must avoid list_moved() in this case. */ | |
2652 | list_init(&rule->facets); | |
2653 | } | |
abe529af | 2654 | |
7ee20df1 | 2655 | complete_operation(rule); |
abe529af BP |
2656 | return 0; |
2657 | } | |
2658 | ||
2659 | static void | |
2660 | rule_destruct(struct rule *rule_) | |
2661 | { | |
2662 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2663 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2664 | struct facet *facet, *next_facet; | |
2665 | ||
abe529af BP |
2666 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
2667 | facet_revalidate(ofproto, facet); | |
2668 | } | |
7ee20df1 BP |
2669 | |
2670 | complete_operation(rule); | |
abe529af BP |
2671 | } |
2672 | ||
2673 | static void | |
2674 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
2675 | { | |
2676 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2677 | struct facet *facet; | |
2678 | ||
2679 | /* Start from historical data for 'rule' itself that are no longer tracked | |
2680 | * in facets. This counts, for example, facets that have expired. */ | |
2681 | *packets = rule->packet_count; | |
2682 | *bytes = rule->byte_count; | |
2683 | ||
2684 | /* Add any statistics that are tracked by facets. This includes | |
2685 | * statistical data recently updated by ofproto_update_stats() as well as | |
2686 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
2687 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
2688 | *packets += facet->packet_count; | |
2689 | *bytes += facet->byte_count; | |
2690 | } | |
2691 | } | |
2692 | ||
5bf0e941 | 2693 | static int |
abe529af BP |
2694 | rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet) |
2695 | { | |
2696 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2697 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2698 | struct action_xlate_ctx ctx; | |
2699 | struct ofpbuf *odp_actions; | |
2700 | struct facet *facet; | |
2701 | size_t size; | |
2702 | ||
2703 | /* First look for a related facet. If we find one, account it to that. */ | |
2704 | facet = facet_lookup_valid(ofproto, flow); | |
2705 | if (facet && facet->rule == rule) { | |
2706 | facet_execute(ofproto, facet, packet); | |
5bf0e941 | 2707 | return 0; |
abe529af BP |
2708 | } |
2709 | ||
2710 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then | |
2711 | * create a new facet for it and use that. */ | |
2712 | if (rule_dpif_lookup(ofproto, flow) == rule) { | |
2713 | facet = facet_create(rule, flow, packet); | |
2714 | facet_execute(ofproto, facet, packet); | |
2715 | facet_install(ofproto, facet, true); | |
5bf0e941 | 2716 | return 0; |
abe529af BP |
2717 | } |
2718 | ||
2719 | /* We can't account anything to a facet. If we were to try, then that | |
2720 | * facet would have a non-matching rule, busting our invariants. */ | |
2721 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
2722 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
2723 | size = packet->size; | |
2724 | if (execute_odp_actions(ofproto, flow, odp_actions->data, | |
2725 | odp_actions->size, packet)) { | |
2726 | rule->used = time_msec(); | |
2727 | rule->packet_count++; | |
2728 | rule->byte_count += size; | |
2729 | flow_push_stats(rule, flow, 1, size, rule->used); | |
2730 | } | |
2731 | ofpbuf_delete(odp_actions); | |
5bf0e941 BP |
2732 | |
2733 | return 0; | |
abe529af BP |
2734 | } |
2735 | ||
7ee20df1 BP |
2736 | static void |
2737 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
2738 | { |
2739 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2740 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2741 | int error; | |
2742 | ||
7ee20df1 BP |
2743 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
2744 | &rule->up.cr.flow, ofproto->max_ports); | |
2745 | if (error) { | |
2746 | ofoperation_complete(rule->up.pending, error); | |
2747 | return; | |
abe529af | 2748 | } |
7ee20df1 BP |
2749 | |
2750 | complete_operation(rule); | |
abe529af BP |
2751 | } |
2752 | \f | |
b2fda3ef | 2753 | /* Sends 'packet' out of port 'odp_port' within 'p'. |
abe529af BP |
2754 | * Returns 0 if successful, otherwise a positive errno value. */ |
2755 | static int | |
b2fda3ef | 2756 | send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, |
abe529af BP |
2757 | const struct ofpbuf *packet) |
2758 | { | |
80e5eed9 BP |
2759 | struct ofpbuf key, odp_actions; |
2760 | struct odputil_keybuf keybuf; | |
2761 | struct flow flow; | |
abe529af BP |
2762 | int error; |
2763 | ||
80e5eed9 BP |
2764 | flow_extract((struct ofpbuf *) packet, 0, 0, &flow); |
2765 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2766 | odp_flow_key_from_flow(&key, &flow); | |
2767 | ||
abe529af | 2768 | ofpbuf_init(&odp_actions, 32); |
abe529af | 2769 | nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
2770 | error = dpif_execute(ofproto->dpif, |
2771 | key.data, key.size, | |
2772 | odp_actions.data, odp_actions.size, | |
abe529af BP |
2773 | packet); |
2774 | ofpbuf_uninit(&odp_actions); | |
2775 | ||
2776 | if (error) { | |
2777 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
2778 | ofproto->up.name, odp_port, strerror(error)); | |
2779 | } | |
2780 | return error; | |
2781 | } | |
2782 | \f | |
2783 | /* OpenFlow to ODP action translation. */ | |
2784 | ||
2785 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2786 | struct action_xlate_ctx *ctx); | |
2787 | static bool xlate_normal(struct action_xlate_ctx *); | |
2788 | ||
b3e9b2ed EJ |
2789 | static void |
2790 | commit_odp_actions(struct action_xlate_ctx *ctx) | |
2791 | { | |
2792 | const struct flow *flow = &ctx->flow; | |
2793 | struct flow *base = &ctx->base_flow; | |
2794 | struct ofpbuf *odp_actions = ctx->odp_actions; | |
2795 | ||
2796 | if (base->tun_id != flow->tun_id) { | |
2797 | nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id); | |
2798 | base->tun_id = flow->tun_id; | |
2799 | } | |
2800 | ||
2801 | if (base->nw_src != flow->nw_src) { | |
2802 | nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src); | |
2803 | base->nw_src = flow->nw_src; | |
2804 | } | |
2805 | ||
2806 | if (base->nw_dst != flow->nw_dst) { | |
2807 | nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst); | |
2808 | base->nw_dst = flow->nw_dst; | |
2809 | } | |
2810 | ||
150a9f15 BP |
2811 | if (base->nw_tos != flow->nw_tos) { |
2812 | nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos); | |
2813 | base->nw_tos = flow->nw_tos; | |
2814 | } | |
2815 | ||
b3e9b2ed EJ |
2816 | if (base->vlan_tci != flow->vlan_tci) { |
2817 | if (!(flow->vlan_tci & htons(VLAN_CFI))) { | |
2818 | nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); | |
2819 | } else { | |
2820 | nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, | |
2821 | flow->vlan_tci & ~htons(VLAN_CFI)); | |
2822 | } | |
2823 | base->vlan_tci = flow->vlan_tci; | |
2824 | } | |
2825 | ||
2826 | if (base->tp_src != flow->tp_src) { | |
2827 | nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src); | |
2828 | base->tp_src = flow->tp_src; | |
2829 | } | |
2830 | ||
2831 | if (base->tp_dst != flow->tp_dst) { | |
2832 | nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst); | |
2833 | base->tp_dst = flow->tp_dst; | |
2834 | } | |
2835 | ||
2836 | if (!eth_addr_equals(base->dl_src, flow->dl_src)) { | |
2837 | nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC, | |
2838 | flow->dl_src, ETH_ADDR_LEN); | |
2839 | memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN); | |
2840 | } | |
2841 | ||
2842 | if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) { | |
2843 | nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST, | |
2844 | flow->dl_dst, ETH_ADDR_LEN); | |
2845 | memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN); | |
2846 | } | |
2847 | ||
2848 | if (ctx->base_priority != ctx->priority) { | |
2849 | if (ctx->priority) { | |
2850 | nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, | |
2851 | ctx->priority); | |
2852 | } else { | |
2853 | nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY); | |
2854 | } | |
2855 | ctx->base_priority = ctx->priority; | |
2856 | } | |
2857 | } | |
2858 | ||
abe529af BP |
2859 | static void |
2860 | add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) | |
2861 | { | |
2862 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); | |
2863 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); | |
2864 | ||
2865 | if (ofport) { | |
2866 | if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) { | |
2867 | /* Forwarding disabled on port. */ | |
2868 | return; | |
2869 | } | |
2870 | } else { | |
2871 | /* | |
2872 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2873 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2874 | * and we're pre-populating the flow table. | |
2875 | */ | |
2876 | } | |
2877 | ||
b3e9b2ed | 2878 | commit_odp_actions(ctx); |
abe529af BP |
2879 | nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); |
2880 | ctx->nf_output_iface = ofp_port; | |
2881 | } | |
2882 | ||
2883 | static void | |
2884 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2885 | { | |
2886 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
2887 | struct rule_dpif *rule; | |
2888 | uint16_t old_in_port; | |
2889 | ||
2890 | /* Look up a flow with 'in_port' as the input port. Then restore the | |
2891 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2892 | * have surprising behavior). */ | |
2893 | old_in_port = ctx->flow.in_port; | |
2894 | ctx->flow.in_port = in_port; | |
2895 | rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow); | |
2896 | ctx->flow.in_port = old_in_port; | |
2897 | ||
2898 | if (ctx->resubmit_hook) { | |
2899 | ctx->resubmit_hook(ctx, rule); | |
2900 | } | |
2901 | ||
2902 | if (rule) { | |
2903 | ctx->recurse++; | |
2904 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); | |
2905 | ctx->recurse--; | |
2906 | } | |
2907 | } else { | |
2908 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
2909 | ||
2910 | VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", | |
2911 | MAX_RESUBMIT_RECURSION); | |
2912 | } | |
2913 | } | |
2914 | ||
2915 | static void | |
b3e9b2ed | 2916 | flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask) |
abe529af BP |
2917 | { |
2918 | struct ofport_dpif *ofport; | |
2919 | ||
b3e9b2ed EJ |
2920 | commit_odp_actions(ctx); |
2921 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { | |
abe529af | 2922 | uint16_t ofp_port = ofport->up.ofp_port; |
b3e9b2ed EJ |
2923 | if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) { |
2924 | nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, | |
abe529af BP |
2925 | ofport->odp_port); |
2926 | } | |
2927 | } | |
b3e9b2ed EJ |
2928 | |
2929 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
2930 | } |
2931 | ||
2932 | static void | |
2933 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
2934 | uint16_t port, uint16_t max_len) | |
2935 | { | |
2936 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
2937 | ||
2938 | ctx->nf_output_iface = NF_OUT_DROP; | |
2939 | ||
2940 | switch (port) { | |
2941 | case OFPP_IN_PORT: | |
2942 | add_output_action(ctx, ctx->flow.in_port); | |
2943 | break; | |
2944 | case OFPP_TABLE: | |
2945 | xlate_table_action(ctx, ctx->flow.in_port); | |
2946 | break; | |
2947 | case OFPP_NORMAL: | |
2948 | xlate_normal(ctx); | |
2949 | break; | |
2950 | case OFPP_FLOOD: | |
b3e9b2ed | 2951 | flood_packets(ctx, htonl(OFPPC_NO_FLOOD)); |
abe529af BP |
2952 | break; |
2953 | case OFPP_ALL: | |
b3e9b2ed | 2954 | flood_packets(ctx, htonl(0)); |
abe529af BP |
2955 | break; |
2956 | case OFPP_CONTROLLER: | |
b3e9b2ed | 2957 | commit_odp_actions(ctx); |
abe529af BP |
2958 | nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_CONTROLLER, max_len); |
2959 | break; | |
2960 | case OFPP_LOCAL: | |
2961 | add_output_action(ctx, OFPP_LOCAL); | |
2962 | break; | |
e81d2933 EJ |
2963 | case OFPP_NONE: |
2964 | break; | |
abe529af BP |
2965 | default: |
2966 | if (port != ctx->flow.in_port) { | |
2967 | add_output_action(ctx, port); | |
2968 | } | |
2969 | break; | |
2970 | } | |
2971 | ||
2972 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2973 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2974 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2975 | ctx->nf_output_iface = prev_nf_output_iface; | |
2976 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2977 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2978 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2979 | } | |
2980 | } | |
2981 | ||
2982 | static void | |
2983 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2984 | const struct ofp_action_output *oao) | |
2985 | { | |
2986 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2987 | } | |
2988 | ||
abe529af BP |
2989 | static void |
2990 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2991 | const struct ofp_action_enqueue *oae) | |
2992 | { | |
2993 | uint16_t ofp_port, odp_port; | |
b3e9b2ed | 2994 | uint32_t ctx_priority, priority; |
abe529af BP |
2995 | int error; |
2996 | ||
2997 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2998 | &priority); | |
2999 | if (error) { | |
3000 | /* Fall back to ordinary output action. */ | |
3001 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
3002 | return; | |
3003 | } | |
3004 | ||
3005 | /* Figure out ODP output port. */ | |
3006 | ofp_port = ntohs(oae->port); | |
3007 | if (ofp_port == OFPP_IN_PORT) { | |
3008 | ofp_port = ctx->flow.in_port; | |
3009 | } | |
3010 | odp_port = ofp_port_to_odp_port(ofp_port); | |
3011 | ||
3012 | /* Add ODP actions. */ | |
b3e9b2ed EJ |
3013 | ctx_priority = ctx->priority; |
3014 | ctx->priority = priority; | |
abe529af | 3015 | add_output_action(ctx, odp_port); |
b3e9b2ed | 3016 | ctx->priority = ctx_priority; |
abe529af BP |
3017 | |
3018 | /* Update NetFlow output port. */ | |
3019 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
3020 | ctx->nf_output_iface = odp_port; | |
3021 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
3022 | ctx->nf_output_iface = NF_OUT_MULTI; | |
3023 | } | |
3024 | } | |
3025 | ||
3026 | static void | |
3027 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
3028 | const struct nx_action_set_queue *nasq) | |
3029 | { | |
3030 | uint32_t priority; | |
3031 | int error; | |
3032 | ||
3033 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
3034 | &priority); | |
3035 | if (error) { | |
3036 | /* Couldn't translate queue to a priority, so ignore. A warning | |
3037 | * has already been logged. */ | |
3038 | return; | |
3039 | } | |
3040 | ||
afabef2b | 3041 | ctx->priority = priority; |
abe529af BP |
3042 | } |
3043 | ||
3044 | struct xlate_reg_state { | |
3045 | ovs_be16 vlan_tci; | |
3046 | ovs_be64 tun_id; | |
3047 | }; | |
3048 | ||
abe529af BP |
3049 | static void |
3050 | xlate_autopath(struct action_xlate_ctx *ctx, | |
3051 | const struct nx_action_autopath *naa) | |
3052 | { | |
3053 | uint16_t ofp_port = ntohl(naa->id); | |
3054 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
3055 | ||
3056 | if (!port || !port->bundle) { | |
3057 | ofp_port = OFPP_NONE; | |
3058 | } else if (port->bundle->bond) { | |
3059 | /* Autopath does not support VLAN hashing. */ | |
3060 | struct ofport_dpif *slave = bond_choose_output_slave( | |
3061 | port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags); | |
3062 | if (slave) { | |
3063 | ofp_port = slave->up.ofp_port; | |
3064 | } | |
3065 | } | |
3066 | autopath_execute(naa, &ctx->flow, ofp_port); | |
3067 | } | |
3068 | ||
daff3353 EJ |
3069 | static bool |
3070 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
3071 | { | |
3072 | struct ofproto_dpif *ofproto = ofproto_; | |
3073 | struct ofport_dpif *port; | |
3074 | ||
3075 | switch (ofp_port) { | |
3076 | case OFPP_IN_PORT: | |
3077 | case OFPP_TABLE: | |
3078 | case OFPP_NORMAL: | |
3079 | case OFPP_FLOOD: | |
3080 | case OFPP_ALL: | |
3081 | case OFPP_LOCAL: | |
3082 | return true; | |
3083 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
3084 | return false; | |
3085 | default: | |
3086 | port = get_ofp_port(ofproto, ofp_port); | |
3087 | return port ? port->may_enable : false; | |
3088 | } | |
3089 | } | |
3090 | ||
abe529af BP |
3091 | static void |
3092 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
3093 | struct action_xlate_ctx *ctx) | |
3094 | { | |
3095 | const struct ofport_dpif *port; | |
abe529af | 3096 | const union ofp_action *ia; |
b4b8c781 | 3097 | size_t left; |
abe529af BP |
3098 | |
3099 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
3100 | if (port | |
3101 | && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && | |
3102 | port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
3103 | ? htonl(OFPPC_NO_RECV_STP) | |
3104 | : htonl(OFPPC_NO_RECV))) { | |
3105 | /* Drop this flow. */ | |
3106 | return; | |
3107 | } | |
3108 | ||
b4b8c781 | 3109 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 3110 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
3111 | const struct nx_action_resubmit *nar; |
3112 | const struct nx_action_set_tunnel *nast; | |
3113 | const struct nx_action_set_queue *nasq; | |
3114 | const struct nx_action_multipath *nam; | |
3115 | const struct nx_action_autopath *naa; | |
daff3353 | 3116 | const struct nx_action_bundle *nab; |
38f2e360 BP |
3117 | enum ofputil_action_code code; |
3118 | ovs_be64 tun_id; | |
3119 | ||
3120 | code = ofputil_decode_action_unsafe(ia); | |
3121 | switch (code) { | |
3122 | case OFPUTIL_OFPAT_OUTPUT: | |
abe529af BP |
3123 | xlate_output_action(ctx, &ia->output); |
3124 | break; | |
3125 | ||
38f2e360 | 3126 | case OFPUTIL_OFPAT_SET_VLAN_VID: |
abe529af BP |
3127 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
3128 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
3129 | break; |
3130 | ||
38f2e360 | 3131 | case OFPUTIL_OFPAT_SET_VLAN_PCP: |
abe529af BP |
3132 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
3133 | ctx->flow.vlan_tci |= htons( | |
3134 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
3135 | break; |
3136 | ||
38f2e360 | 3137 | case OFPUTIL_OFPAT_STRIP_VLAN: |
abe529af | 3138 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
3139 | break; |
3140 | ||
38f2e360 | 3141 | case OFPUTIL_OFPAT_SET_DL_SRC: |
abe529af | 3142 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
3143 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
3144 | break; | |
3145 | ||
38f2e360 | 3146 | case OFPUTIL_OFPAT_SET_DL_DST: |
abe529af | 3147 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
3148 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
3149 | break; | |
3150 | ||
38f2e360 | 3151 | case OFPUTIL_OFPAT_SET_NW_SRC: |
abe529af BP |
3152 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
3153 | break; | |
3154 | ||
38f2e360 | 3155 | case OFPUTIL_OFPAT_SET_NW_DST: |
abe529af BP |
3156 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
3157 | break; | |
3158 | ||
38f2e360 | 3159 | case OFPUTIL_OFPAT_SET_NW_TOS: |
1fa96cf4 | 3160 | ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK; |
abe529af BP |
3161 | break; |
3162 | ||
38f2e360 | 3163 | case OFPUTIL_OFPAT_SET_TP_SRC: |
abe529af BP |
3164 | ctx->flow.tp_src = ia->tp_port.tp_port; |
3165 | break; | |
3166 | ||
38f2e360 | 3167 | case OFPUTIL_OFPAT_SET_TP_DST: |
abe529af BP |
3168 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
3169 | break; | |
3170 | ||
38f2e360 BP |
3171 | case OFPUTIL_OFPAT_ENQUEUE: |
3172 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
3173 | break; | |
3174 | ||
3175 | case OFPUTIL_NXAST_RESUBMIT: | |
3176 | nar = (const struct nx_action_resubmit *) ia; | |
3177 | xlate_table_action(ctx, ntohs(nar->in_port)); | |
abe529af BP |
3178 | break; |
3179 | ||
38f2e360 BP |
3180 | case OFPUTIL_NXAST_SET_TUNNEL: |
3181 | nast = (const struct nx_action_set_tunnel *) ia; | |
3182 | tun_id = htonll(ntohl(nast->tun_id)); | |
3183 | ctx->flow.tun_id = tun_id; | |
3184 | break; | |
3185 | ||
3186 | case OFPUTIL_NXAST_SET_QUEUE: | |
3187 | nasq = (const struct nx_action_set_queue *) ia; | |
3188 | xlate_set_queue_action(ctx, nasq); | |
3189 | break; | |
3190 | ||
3191 | case OFPUTIL_NXAST_POP_QUEUE: | |
3192 | ctx->priority = 0; | |
3193 | break; | |
3194 | ||
3195 | case OFPUTIL_NXAST_REG_MOVE: | |
3196 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
3197 | &ctx->flow); | |
3198 | break; | |
3199 | ||
3200 | case OFPUTIL_NXAST_REG_LOAD: | |
3201 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
3202 | &ctx->flow); | |
3203 | break; | |
3204 | ||
3205 | case OFPUTIL_NXAST_NOTE: | |
3206 | /* Nothing to do. */ | |
3207 | break; | |
3208 | ||
3209 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
3210 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
3211 | ctx->flow.tun_id = tun_id; | |
3212 | break; | |
3213 | ||
3214 | case OFPUTIL_NXAST_MULTIPATH: | |
3215 | nam = (const struct nx_action_multipath *) ia; | |
3216 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
3217 | break; |
3218 | ||
38f2e360 BP |
3219 | case OFPUTIL_NXAST_AUTOPATH: |
3220 | naa = (const struct nx_action_autopath *) ia; | |
3221 | xlate_autopath(ctx, naa); | |
abe529af | 3222 | break; |
daff3353 EJ |
3223 | |
3224 | case OFPUTIL_NXAST_BUNDLE: | |
3225 | ctx->ofproto->has_bundle_action = true; | |
3226 | nab = (const struct nx_action_bundle *) ia; | |
3227 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
3228 | slave_enabled_cb, | |
3229 | ctx->ofproto), 0); | |
3230 | break; | |
a368bb53 EJ |
3231 | |
3232 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
3233 | ctx->ofproto->has_bundle_action = true; | |
3234 | nab = (const struct nx_action_bundle *) ia; | |
3235 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
3236 | ctx->ofproto); | |
3237 | break; | |
abe529af BP |
3238 | } |
3239 | } | |
3240 | } | |
3241 | ||
3242 | static void | |
3243 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
3244 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
3245 | const struct ofpbuf *packet) | |
3246 | { | |
3247 | ctx->ofproto = ofproto; | |
3248 | ctx->flow = *flow; | |
3249 | ctx->packet = packet; | |
3250 | ctx->resubmit_hook = NULL; | |
abe529af BP |
3251 | } |
3252 | ||
3253 | static struct ofpbuf * | |
3254 | xlate_actions(struct action_xlate_ctx *ctx, | |
3255 | const union ofp_action *in, size_t n_in) | |
3256 | { | |
3257 | COVERAGE_INC(ofproto_dpif_xlate); | |
3258 | ||
3259 | ctx->odp_actions = ofpbuf_new(512); | |
3260 | ctx->tags = 0; | |
3261 | ctx->may_set_up_flow = true; | |
3262 | ctx->nf_output_iface = NF_OUT_DROP; | |
3263 | ctx->recurse = 0; | |
afabef2b | 3264 | ctx->priority = 0; |
b3e9b2ed EJ |
3265 | ctx->base_priority = 0; |
3266 | ctx->base_flow = ctx->flow; | |
abe529af | 3267 | |
fc08b7a2 | 3268 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af BP |
3269 | ctx->may_set_up_flow = false; |
3270 | } else { | |
3271 | do_xlate_actions(in, n_in, ctx); | |
3272 | } | |
3273 | ||
abe529af BP |
3274 | /* Check with in-band control to see if we're allowed to set up this |
3275 | * flow. */ | |
3276 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, | |
3277 | ctx->odp_actions->data, | |
3278 | ctx->odp_actions->size)) { | |
3279 | ctx->may_set_up_flow = false; | |
3280 | } | |
3281 | ||
3282 | return ctx->odp_actions; | |
3283 | } | |
3284 | \f | |
3285 | /* OFPP_NORMAL implementation. */ | |
3286 | ||
3287 | struct dst { | |
3288 | struct ofport_dpif *port; | |
3289 | uint16_t vlan; | |
3290 | }; | |
3291 | ||
3292 | struct dst_set { | |
3293 | struct dst builtin[32]; | |
3294 | struct dst *dsts; | |
3295 | size_t n, allocated; | |
3296 | }; | |
3297 | ||
3298 | static void dst_set_init(struct dst_set *); | |
3299 | static void dst_set_add(struct dst_set *, const struct dst *); | |
3300 | static void dst_set_free(struct dst_set *); | |
3301 | ||
3302 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); | |
3303 | ||
3304 | static bool | |
3305 | set_dst(struct action_xlate_ctx *ctx, struct dst *dst, | |
3306 | const struct ofbundle *in_bundle, const struct ofbundle *out_bundle) | |
3307 | { | |
3308 | dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE | |
3309 | : in_bundle->vlan >= 0 ? in_bundle->vlan | |
3310 | : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE | |
3311 | : vlan_tci_to_vid(ctx->flow.vlan_tci)); | |
3312 | ||
3313 | dst->port = (!out_bundle->bond | |
3314 | ? ofbundle_get_a_port(out_bundle) | |
3315 | : bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
3316 | dst->vlan, &ctx->tags)); | |
3317 | ||
3318 | return dst->port != NULL; | |
3319 | } | |
3320 | ||
3321 | static int | |
3322 | mirror_mask_ffs(mirror_mask_t mask) | |
3323 | { | |
3324 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
3325 | return ffs(mask); | |
3326 | } | |
3327 | ||
3328 | static void | |
3329 | dst_set_init(struct dst_set *set) | |
3330 | { | |
3331 | set->dsts = set->builtin; | |
3332 | set->n = 0; | |
3333 | set->allocated = ARRAY_SIZE(set->builtin); | |
3334 | } | |
3335 | ||
3336 | static void | |
3337 | dst_set_add(struct dst_set *set, const struct dst *dst) | |
3338 | { | |
3339 | if (set->n >= set->allocated) { | |
3340 | size_t new_allocated; | |
3341 | struct dst *new_dsts; | |
3342 | ||
3343 | new_allocated = set->allocated * 2; | |
3344 | new_dsts = xmalloc(new_allocated * sizeof *new_dsts); | |
3345 | memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); | |
3346 | ||
3347 | dst_set_free(set); | |
3348 | ||
3349 | set->dsts = new_dsts; | |
3350 | set->allocated = new_allocated; | |
3351 | } | |
3352 | set->dsts[set->n++] = *dst; | |
3353 | } | |
3354 | ||
3355 | static void | |
3356 | dst_set_free(struct dst_set *set) | |
3357 | { | |
3358 | if (set->dsts != set->builtin) { | |
3359 | free(set->dsts); | |
3360 | } | |
3361 | } | |
3362 | ||
3363 | static bool | |
3364 | dst_is_duplicate(const struct dst_set *set, const struct dst *test) | |
3365 | { | |
3366 | size_t i; | |
3367 | for (i = 0; i < set->n; i++) { | |
3368 | if (set->dsts[i].vlan == test->vlan | |
3369 | && set->dsts[i].port == test->port) { | |
3370 | return true; | |
3371 | } | |
3372 | } | |
3373 | return false; | |
3374 | } | |
3375 | ||
3376 | static bool | |
3377 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
3378 | { | |
fc3d7408 BP |
3379 | return (bundle->vlan < 0 |
3380 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); | |
abe529af BP |
3381 | } |
3382 | ||
3383 | static bool | |
3384 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
3385 | { | |
3386 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
3387 | } | |
3388 | ||
3389 | /* Returns an arbitrary interface within 'bundle'. */ | |
3390 | static struct ofport_dpif * | |
3391 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
3392 | { | |
3393 | return CONTAINER_OF(list_front(&bundle->ports), | |
3394 | struct ofport_dpif, bundle_node); | |
3395 | } | |
3396 | ||
3397 | static void | |
3398 | compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan, | |
3399 | const struct ofbundle *in_bundle, | |
3400 | const struct ofbundle *out_bundle, struct dst_set *set) | |
3401 | { | |
3402 | struct dst dst; | |
3403 | ||
3404 | if (out_bundle == OFBUNDLE_FLOOD) { | |
3405 | struct ofbundle *bundle; | |
3406 | ||
3407 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { | |
3408 | if (bundle != in_bundle | |
3409 | && ofbundle_includes_vlan(bundle, vlan) | |
3410 | && bundle->floodable | |
3411 | && !bundle->mirror_out | |
3412 | && set_dst(ctx, &dst, in_bundle, bundle)) { | |
3413 | dst_set_add(set, &dst); | |
3414 | } | |
3415 | } | |
3416 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
3417 | } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) { | |
3418 | dst_set_add(set, &dst); | |
3419 | ctx->nf_output_iface = dst.port->odp_port; | |
3420 | } | |
3421 | } | |
3422 | ||
3423 | static bool | |
3424 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
3425 | { | |
fc3d7408 | 3426 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
3427 | } |
3428 | ||
07817dfe BP |
3429 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
3430 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
3431 | * protocols that may confuse switches. */ | |
3432 | static bool | |
3433 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
3434 | { | |
3435 | /* If you change this function's behavior, please update corresponding | |
3436 | * documentation in vswitch.xml at the same time. */ | |
3437 | if (dst[0] != 0x01) { | |
3438 | /* All the currently banned MACs happen to start with 01 currently, so | |
3439 | * this is a quick way to eliminate most of the good ones. */ | |
3440 | } else { | |
3441 | if (eth_addr_is_reserved(dst)) { | |
3442 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
3443 | * (01-80-c2-00-00-0x). */ | |
3444 | return false; | |
3445 | } | |
3446 | ||
3447 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
3448 | /* Cisco OUI. */ | |
3449 | if ((dst[3] & 0xfe) == 0xcc && | |
3450 | (dst[4] & 0xfe) == 0xcc && | |
3451 | (dst[5] & 0xfe) == 0xcc) { | |
3452 | /* Drop the following protocols plus others following the same | |
3453 | pattern: | |
3454 | ||
3455 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
3456 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
3457 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
3458 | return false; | |
3459 | } | |
3460 | ||
3461 | if (!(dst[3] | dst[4] | dst[5])) { | |
3462 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
3463 | return false; | |
3464 | } | |
3465 | } | |
3466 | } | |
3467 | return true; | |
3468 | } | |
3469 | ||
abe529af BP |
3470 | static void |
3471 | compose_mirror_dsts(struct action_xlate_ctx *ctx, | |
3472 | uint16_t vlan, const struct ofbundle *in_bundle, | |
3473 | struct dst_set *set) | |
3474 | { | |
3475 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
3476 | mirror_mask_t mirrors; | |
3477 | int flow_vlan; | |
3478 | size_t i; | |
3479 | ||
3480 | mirrors = in_bundle->src_mirrors; | |
3481 | for (i = 0; i < set->n; i++) { | |
3482 | mirrors |= set->dsts[i].port->bundle->dst_mirrors; | |
3483 | } | |
3484 | ||
3485 | if (!mirrors) { | |
3486 | return; | |
3487 | } | |
3488 | ||
3489 | flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
3490 | if (flow_vlan == 0) { | |
3491 | flow_vlan = OFP_VLAN_NONE; | |
3492 | } | |
3493 | ||
3494 | while (mirrors) { | |
3495 | struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
3496 | if (vlan_is_mirrored(m, vlan)) { | |
3497 | struct dst dst; | |
3498 | ||
3499 | if (m->out) { | |
3500 | if (set_dst(ctx, &dst, in_bundle, m->out) | |
3501 | && !dst_is_duplicate(set, &dst)) { | |
3502 | dst_set_add(set, &dst); | |
3503 | } | |
07817dfe | 3504 | } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) { |
abe529af BP |
3505 | struct ofbundle *bundle; |
3506 | ||
3507 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3508 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
3509 | && set_dst(ctx, &dst, in_bundle, bundle)) | |
3510 | { | |
3511 | if (bundle->vlan < 0) { | |
3512 | dst.vlan = m->out_vlan; | |
3513 | } | |
3514 | if (dst_is_duplicate(set, &dst)) { | |
3515 | continue; | |
3516 | } | |
3517 | ||
3518 | /* Use the vlan tag on the original flow instead of | |
3519 | * the one passed in the vlan parameter. This ensures | |
3520 | * that we compare the vlan from before any implicit | |
3521 | * tagging tags place. This is necessary because | |
3522 | * dst->vlan is the final vlan, after removing implicit | |
3523 | * tags. */ | |
3524 | if (bundle == in_bundle && dst.vlan == flow_vlan) { | |
3525 | /* Don't send out input port on same VLAN. */ | |
3526 | continue; | |
3527 | } | |
3528 | dst_set_add(set, &dst); | |
3529 | } | |
3530 | } | |
3531 | } | |
3532 | } | |
3533 | mirrors &= mirrors - 1; | |
3534 | } | |
3535 | } | |
3536 | ||
3537 | static void | |
3538 | compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan, | |
3539 | const struct ofbundle *in_bundle, | |
3540 | const struct ofbundle *out_bundle) | |
3541 | { | |
3542 | uint16_t initial_vlan, cur_vlan; | |
3543 | const struct dst *dst; | |
3544 | struct dst_set set; | |
3545 | ||
3546 | dst_set_init(&set); | |
3547 | compose_dsts(ctx, vlan, in_bundle, out_bundle, &set); | |
3548 | compose_mirror_dsts(ctx, vlan, in_bundle, &set); | |
3549 | ||
3550 | /* Output all the packets we can without having to change the VLAN. */ | |
3551 | initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
3552 | if (initial_vlan == 0) { | |
3553 | initial_vlan = OFP_VLAN_NONE; | |
3554 | } | |
3555 | for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { | |
3556 | if (dst->vlan != initial_vlan) { | |
3557 | continue; | |
3558 | } | |
3559 | nl_msg_put_u32(ctx->odp_actions, | |
3560 | ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); | |
3561 | } | |
3562 | ||
3563 | /* Then output the rest. */ | |
3564 | cur_vlan = initial_vlan; | |
3565 | for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { | |
3566 | if (dst->vlan == initial_vlan) { | |
3567 | continue; | |
3568 | } | |
3569 | if (dst->vlan != cur_vlan) { | |
3570 | if (dst->vlan == OFP_VLAN_NONE) { | |
3571 | nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); | |
3572 | } else { | |
3573 | ovs_be16 tci; | |
3574 | tci = htons(dst->vlan & VLAN_VID_MASK); | |
3575 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
3576 | nl_msg_put_be16(ctx->odp_actions, | |
3577 | ODP_ACTION_ATTR_SET_DL_TCI, tci); | |
3578 | } | |
3579 | cur_vlan = dst->vlan; | |
3580 | } | |
3581 | nl_msg_put_u32(ctx->odp_actions, | |
3582 | ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); | |
3583 | } | |
3584 | ||
3585 | dst_set_free(&set); | |
3586 | } | |
3587 | ||
3588 | /* Returns the effective vlan of a packet, taking into account both the | |
3589 | * 802.1Q header and implicitly tagged ports. A value of 0 indicates that | |
3590 | * the packet is untagged and -1 indicates it has an invalid header and | |
3591 | * should be dropped. */ | |
3592 | static int | |
3593 | flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3594 | struct ofbundle *in_bundle, bool have_packet) | |
3595 | { | |
3596 | int vlan = vlan_tci_to_vid(flow->vlan_tci); | |
3597 | if (in_bundle->vlan >= 0) { | |
3598 | if (vlan) { | |
3599 | if (have_packet) { | |
3600 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3601 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
3602 | "packet received on port %s configured with " | |
3603 | "implicit VLAN %"PRIu16, | |
3604 | ofproto->up.name, vlan, | |
3605 | in_bundle->name, in_bundle->vlan); | |
3606 | } | |
3607 | return -1; | |
3608 | } | |
3609 | vlan = in_bundle->vlan; | |
3610 | } else { | |
3611 | if (!ofbundle_includes_vlan(in_bundle, vlan)) { | |
3612 | if (have_packet) { | |
3613 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3614 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
3615 | "packet received on port %s not configured for " | |
3616 | "trunking VLAN %d", | |
3617 | ofproto->up.name, vlan, in_bundle->name, vlan); | |
3618 | } | |
3619 | return -1; | |
3620 | } | |
3621 | } | |
3622 | ||
3623 | return vlan; | |
3624 | } | |
3625 | ||
3626 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after | |
3627 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
3628 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
3629 | static bool | |
3630 | is_gratuitous_arp(const struct flow *flow) | |
3631 | { | |
3632 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
3633 | && eth_addr_is_broadcast(flow->dl_dst) | |
3634 | && (flow->nw_proto == ARP_OP_REPLY | |
3635 | || (flow->nw_proto == ARP_OP_REQUEST | |
3636 | && flow->nw_src == flow->nw_dst))); | |
3637 | } | |
3638 | ||
3639 | static void | |
3640 | update_learning_table(struct ofproto_dpif *ofproto, | |
3641 | const struct flow *flow, int vlan, | |
3642 | struct ofbundle *in_bundle) | |
3643 | { | |
3644 | struct mac_entry *mac; | |
3645 | ||
3646 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { | |
3647 | return; | |
3648 | } | |
3649 | ||
3650 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
3651 | if (is_gratuitous_arp(flow)) { | |
3652 | /* We don't want to learn from gratuitous ARP packets that are | |
3653 | * reflected back over bond slaves so we lock the learning table. */ | |
3654 | if (!in_bundle->bond) { | |
3655 | mac_entry_set_grat_arp_lock(mac); | |
3656 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
3657 | return; | |
3658 | } | |
3659 | } | |
3660 | ||
3661 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
3662 | /* The log messages here could actually be useful in debugging, | |
3663 | * so keep the rate limit relatively high. */ | |
3664 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
3665 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
3666 | "on port %s in VLAN %d", | |
3667 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
3668 | in_bundle->name, vlan); | |
3669 | ||
3670 | mac->port.p = in_bundle; | |
3671 | tag_set_add(&ofproto->revalidate_set, | |
3672 | mac_learning_changed(ofproto->ml, mac)); | |
3673 | } | |
3674 | } | |
3675 | ||
3676 | /* Determines whether packets in 'flow' within 'br' should be forwarded or | |
3677 | * dropped. Returns true if they may be forwarded, false if they should be | |
3678 | * dropped. | |
3679 | * | |
3680 | * If 'have_packet' is true, it indicates that the caller is processing a | |
3681 | * received packet. If 'have_packet' is false, then the caller is just | |
3682 | * revalidating an existing flow because configuration has changed. Either | |
3683 | * way, 'have_packet' only affects logging (there is no point in logging errors | |
3684 | * during revalidation). | |
3685 | * | |
3686 | * Sets '*in_portp' to the input port. This will be a null pointer if | |
3687 | * flow->in_port does not designate a known input port (in which case | |
3688 | * is_admissible() returns false). | |
3689 | * | |
3690 | * When returning true, sets '*vlanp' to the effective VLAN of the input | |
3691 | * packet, as returned by flow_get_vlan(). | |
3692 | * | |
3693 | * May also add tags to '*tags', although the current implementation only does | |
3694 | * so in one special case. | |
3695 | */ | |
3696 | static bool | |
3697 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3698 | bool have_packet, | |
3699 | tag_type *tags, int *vlanp, struct ofbundle **in_bundlep) | |
3700 | { | |
3701 | struct ofport_dpif *in_port; | |
3702 | struct ofbundle *in_bundle; | |
3703 | int vlan; | |
3704 | ||
3705 | /* Find the port and bundle for the received packet. */ | |
3706 | in_port = get_ofp_port(ofproto, flow->in_port); | |
23adee42 | 3707 | *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL; |
abe529af BP |
3708 | if (!in_port || !in_bundle) { |
3709 | /* No interface? Something fishy... */ | |
3710 | if (have_packet) { | |
3711 | /* Odd. A few possible reasons here: | |
3712 | * | |
3713 | * - We deleted a port but there are still a few packets queued up | |
3714 | * from it. | |
3715 | * | |
3716 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
3717 | * we don't know about. | |
3718 | * | |
3719 | * - Packet arrived on the local port but the local port is not | |
3720 | * part of a bundle. | |
3721 | */ | |
3722 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3723 | ||
3724 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
3725 | "port %"PRIu16, | |
3726 | ofproto->up.name, flow->in_port); | |
3727 | } | |
3728 | return false; | |
3729 | } | |
3730 | *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet); | |
3731 | if (vlan < 0) { | |
3732 | return false; | |
3733 | } | |
3734 | ||
3735 | /* Drop frames for reserved multicast addresses. */ | |
3736 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
3737 | return false; | |
3738 | } | |
3739 | ||
3740 | /* Drop frames on bundles reserved for mirroring. */ | |
3741 | if (in_bundle->mirror_out) { | |
3742 | if (have_packet) { | |
3743 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3744 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
3745 | "%s, which is reserved exclusively for mirroring", | |
3746 | ofproto->up.name, in_bundle->name); | |
3747 | } | |
3748 | return false; | |
3749 | } | |
3750 | ||
3751 | if (in_bundle->bond) { | |
3752 | struct mac_entry *mac; | |
3753 | ||
3754 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
3755 | flow->dl_dst, tags)) { | |
3756 | case BV_ACCEPT: | |
3757 | break; | |
3758 | ||
3759 | case BV_DROP: | |
3760 | return false; | |
3761 | ||
3762 | case BV_DROP_IF_MOVED: | |
3763 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
3764 | if (mac && mac->port.p != in_bundle && | |
3765 | (!is_gratuitous_arp(flow) | |
3766 | || mac_entry_is_grat_arp_locked(mac))) { | |
3767 | return false; | |
3768 | } | |
3769 | break; | |
3770 | } | |
3771 | } | |
3772 | ||
3773 | return true; | |
3774 | } | |
3775 | ||
3776 | /* If the composed actions may be applied to any packet in the given 'flow', | |
3777 | * returns true. Otherwise, the actions should only be applied to 'packet', or | |
3778 | * not at all, if 'packet' was NULL. */ | |
3779 | static bool | |
3780 | xlate_normal(struct action_xlate_ctx *ctx) | |
3781 | { | |
3782 | struct ofbundle *in_bundle; | |
3783 | struct ofbundle *out_bundle; | |
3784 | struct mac_entry *mac; | |
3785 | int vlan; | |
3786 | ||
3787 | /* Check whether we should drop packets in this flow. */ | |
3788 | if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL, | |
3789 | &ctx->tags, &vlan, &in_bundle)) { | |
3790 | out_bundle = NULL; | |
3791 | goto done; | |
3792 | } | |
3793 | ||
3794 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
3795 | if (ctx->packet) { | |
3796 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); | |
3797 | } | |
3798 | ||
3799 | /* Determine output bundle. */ | |
3800 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
3801 | &ctx->tags); | |
3802 | if (mac) { | |
3803 | out_bundle = mac->port.p; | |
3804 | } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) { | |
3805 | /* If we are revalidating but don't have a learning entry then eject | |
3806 | * the flow. Installing a flow that floods packets opens up a window | |
3807 | * of time where we could learn from a packet reflected on a bond and | |
3808 | * blackhole packets before the learning table is updated to reflect | |
3809 | * the correct port. */ | |
3810 | return false; | |
3811 | } else { | |
3812 | out_bundle = OFBUNDLE_FLOOD; | |
3813 | } | |
3814 | ||
3815 | /* Don't send packets out their input bundles. */ | |
3816 | if (in_bundle == out_bundle) { | |
3817 | out_bundle = NULL; | |
3818 | } | |
3819 | ||
3820 | done: | |
3821 | if (in_bundle) { | |
3822 | compose_actions(ctx, vlan, in_bundle, out_bundle); | |
3823 | } | |
3824 | ||
3825 | return true; | |
3826 | } | |
3827 | \f | |
3828 | static bool | |
3829 | get_drop_frags(struct ofproto *ofproto_) | |
3830 | { | |
3831 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3832 | bool drop_frags; | |
3833 | ||
3834 | dpif_get_drop_frags(ofproto->dpif, &drop_frags); | |
3835 | return drop_frags; | |
3836 | } | |
3837 | ||
3838 | static void | |
3839 | set_drop_frags(struct ofproto *ofproto_, bool drop_frags) | |
3840 | { | |
3841 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3842 | ||
3843 | dpif_set_drop_frags(ofproto->dpif, drop_frags); | |
3844 | } | |
3845 | ||
3846 | static int | |
3847 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, | |
3848 | const struct flow *flow, | |
3849 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
3850 | { | |
3851 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3852 | int error; | |
3853 | ||
3854 | error = validate_actions(ofp_actions, n_ofp_actions, flow, | |
3855 | ofproto->max_ports); | |
3856 | if (!error) { | |
80e5eed9 | 3857 | struct odputil_keybuf keybuf; |
abe529af BP |
3858 | struct action_xlate_ctx ctx; |
3859 | struct ofpbuf *odp_actions; | |
80e5eed9 BP |
3860 | struct ofpbuf key; |
3861 | ||
3862 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
3863 | odp_flow_key_from_flow(&key, flow); | |
abe529af BP |
3864 | |
3865 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
3866 | odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); | |
80e5eed9 BP |
3867 | dpif_execute(ofproto->dpif, key.data, key.size, |
3868 | odp_actions->data, odp_actions->size, packet); | |
abe529af BP |
3869 | ofpbuf_delete(odp_actions); |
3870 | } | |
3871 | return error; | |
3872 | } | |
3873 | ||
3874 | static void | |
3875 | get_netflow_ids(const struct ofproto *ofproto_, | |
3876 | uint8_t *engine_type, uint8_t *engine_id) | |
3877 | { | |
3878 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3879 | ||
3880 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
3881 | } | |
3882 | \f | |
3883 | static struct ofproto_dpif * | |
3884 | ofproto_dpif_lookup(const char *name) | |
3885 | { | |
3886 | struct ofproto *ofproto = ofproto_lookup(name); | |
3887 | return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class | |
3888 | ? ofproto_dpif_cast(ofproto) | |
3889 | : NULL); | |
3890 | } | |
3891 | ||
3892 | static void | |
3893 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, | |
3894 | const char *args, void *aux OVS_UNUSED) | |
3895 | { | |
3896 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3897 | const struct ofproto_dpif *ofproto; | |
3898 | const struct mac_entry *e; | |
3899 | ||
3900 | ofproto = ofproto_dpif_lookup(args); | |
3901 | if (!ofproto) { | |
3902 | unixctl_command_reply(conn, 501, "no such bridge"); | |
3903 | return; | |
3904 | } | |
3905 | ||
3906 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
3907 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
3908 | struct ofbundle *bundle = e->port.p; | |
3909 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
3910 | ofbundle_get_a_port(bundle)->odp_port, | |
3911 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
3912 | } | |
3913 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
3914 | ds_destroy(&ds); | |
3915 | } | |
3916 | ||
3917 | struct ofproto_trace { | |
3918 | struct action_xlate_ctx ctx; | |
3919 | struct flow flow; | |
3920 | struct ds *result; | |
3921 | }; | |
3922 | ||
3923 | static void | |
3924 | trace_format_rule(struct ds *result, int level, const struct rule *rule) | |
3925 | { | |
3926 | ds_put_char_multiple(result, '\t', level); | |
3927 | if (!rule) { | |
3928 | ds_put_cstr(result, "No match\n"); | |
3929 | return; | |
3930 | } | |
3931 | ||
3932 | ds_put_format(result, "Rule: cookie=%#"PRIx64" ", | |
3933 | ntohll(rule->flow_cookie)); | |
3934 | cls_rule_format(&rule->cr, result); | |
3935 | ds_put_char(result, '\n'); | |
3936 | ||
3937 | ds_put_char_multiple(result, '\t', level); | |
3938 | ds_put_cstr(result, "OpenFlow "); | |
b4b8c781 | 3939 | ofp_print_actions(result, rule->actions, rule->n_actions); |
abe529af BP |
3940 | ds_put_char(result, '\n'); |
3941 | } | |
3942 | ||
3943 | static void | |
3944 | trace_format_flow(struct ds *result, int level, const char *title, | |
3945 | struct ofproto_trace *trace) | |
3946 | { | |
3947 | ds_put_char_multiple(result, '\t', level); | |
3948 | ds_put_format(result, "%s: ", title); | |
3949 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
3950 | ds_put_cstr(result, "unchanged"); | |
3951 | } else { | |
3952 | flow_format(result, &trace->ctx.flow); | |
3953 | trace->flow = trace->ctx.flow; | |
3954 | } | |
3955 | ds_put_char(result, '\n'); | |
3956 | } | |
3957 | ||
3958 | static void | |
3959 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
3960 | { | |
3961 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
3962 | struct ds *result = trace->result; | |
3963 | ||
3964 | ds_put_char(result, '\n'); | |
3965 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
3966 | trace_format_rule(result, ctx->recurse + 1, &rule->up); | |
3967 | } | |
3968 | ||
3969 | static void | |
3970 | ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, | |
3971 | void *aux OVS_UNUSED) | |
3972 | { | |
3973 | char *dpname, *in_port_s, *tun_id_s, *packet_s; | |
3974 | char *args = xstrdup(args_); | |
3975 | char *save_ptr = NULL; | |
3976 | struct ofproto_dpif *ofproto; | |
3977 | struct ofpbuf packet; | |
3978 | struct rule_dpif *rule; | |
3979 | struct ds result; | |
3980 | struct flow flow; | |
3981 | uint16_t in_port; | |
3982 | ovs_be64 tun_id; | |
3983 | char *s; | |
3984 | ||
3985 | ofpbuf_init(&packet, strlen(args) / 2); | |
3986 | ds_init(&result); | |
3987 | ||
3988 | dpname = strtok_r(args, " ", &save_ptr); | |
3989 | tun_id_s = strtok_r(NULL, " ", &save_ptr); | |
3990 | in_port_s = strtok_r(NULL, " ", &save_ptr); | |
3991 | packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ | |
3992 | if (!dpname || !in_port_s || !packet_s) { | |
3993 | unixctl_command_reply(conn, 501, "Bad command syntax"); | |
3994 | goto exit; | |
3995 | } | |
3996 | ||
3997 | ofproto = ofproto_dpif_lookup(dpname); | |
3998 | if (!ofproto) { | |
3999 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
4000 | "for help)"); | |
4001 | goto exit; | |
4002 | } | |
4003 | ||
4004 | tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
4005 | in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
4006 | ||
4007 | packet_s = ofpbuf_put_hex(&packet, packet_s, NULL); | |
4008 | packet_s += strspn(packet_s, " "); | |
4009 | if (*packet_s != '\0') { | |
4010 | unixctl_command_reply(conn, 501, "Trailing garbage in command"); | |
4011 | goto exit; | |
4012 | } | |
4013 | if (packet.size < ETH_HEADER_LEN) { | |
4014 | unixctl_command_reply(conn, 501, "Packet data too short for Ethernet"); | |
4015 | goto exit; | |
4016 | } | |
4017 | ||
4018 | ds_put_cstr(&result, "Packet: "); | |
4019 | s = ofp_packet_to_string(packet.data, packet.size, packet.size); | |
4020 | ds_put_cstr(&result, s); | |
4021 | free(s); | |
4022 | ||
4023 | flow_extract(&packet, tun_id, in_port, &flow); | |
4024 | ds_put_cstr(&result, "Flow: "); | |
4025 | flow_format(&result, &flow); | |
4026 | ds_put_char(&result, '\n'); | |
4027 | ||
4028 | rule = rule_dpif_lookup(ofproto, &flow); | |
4029 | trace_format_rule(&result, 0, &rule->up); | |
4030 | if (rule) { | |
4031 | struct ofproto_trace trace; | |
4032 | struct ofpbuf *odp_actions; | |
4033 | ||
4034 | trace.result = &result; | |
4035 | trace.flow = flow; | |
4036 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet); | |
4037 | trace.ctx.resubmit_hook = trace_resubmit; | |
4038 | odp_actions = xlate_actions(&trace.ctx, | |
4039 | rule->up.actions, rule->up.n_actions); | |
4040 | ||
4041 | ds_put_char(&result, '\n'); | |
4042 | trace_format_flow(&result, 0, "Final flow", &trace); | |
4043 | ds_put_cstr(&result, "Datapath actions: "); | |
4044 | format_odp_actions(&result, odp_actions->data, odp_actions->size); | |
4045 | ofpbuf_delete(odp_actions); | |
4046 | } | |
4047 | ||
4048 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
4049 | ||
4050 | exit: | |
4051 | ds_destroy(&result); | |
4052 | ofpbuf_uninit(&packet); | |
4053 | free(args); | |
4054 | } | |
4055 | ||
7ee20df1 BP |
4056 | static void |
4057 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, | |
4058 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
4059 | { | |
4060 | clogged = true; | |
4061 | unixctl_command_reply(conn, 200, NULL); | |
4062 | } | |
4063 | ||
4064 | static void | |
4065 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, | |
4066 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
4067 | { | |
4068 | clogged = false; | |
4069 | unixctl_command_reply(conn, 200, NULL); | |
4070 | } | |
4071 | ||
abe529af BP |
4072 | static void |
4073 | ofproto_dpif_unixctl_init(void) | |
4074 | { | |
4075 | static bool registered; | |
4076 | if (registered) { | |
4077 | return; | |
4078 | } | |
4079 | registered = true; | |
4080 | ||
4081 | unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL); | |
4082 | unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL); | |
7ee20df1 BP |
4083 | |
4084 | unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL); | |
4085 | unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL); | |
abe529af BP |
4086 | } |
4087 | \f | |
4088 | const struct ofproto_class ofproto_dpif_class = { | |
4089 | enumerate_types, | |
4090 | enumerate_names, | |
4091 | del, | |
4092 | alloc, | |
4093 | construct, | |
4094 | destruct, | |
4095 | dealloc, | |
4096 | run, | |
4097 | wait, | |
4098 | flush, | |
6c1491fb BP |
4099 | get_features, |
4100 | get_tables, | |
abe529af BP |
4101 | port_alloc, |
4102 | port_construct, | |
4103 | port_destruct, | |
4104 | port_dealloc, | |
4105 | port_modified, | |
4106 | port_reconfigured, | |
4107 | port_query_by_name, | |
4108 | port_add, | |
4109 | port_del, | |
4110 | port_dump_start, | |
4111 | port_dump_next, | |
4112 | port_dump_done, | |
4113 | port_poll, | |
4114 | port_poll_wait, | |
4115 | port_is_lacp_current, | |
0ab6decf | 4116 | NULL, /* rule_choose_table */ |
abe529af BP |
4117 | rule_alloc, |
4118 | rule_construct, | |
4119 | rule_destruct, | |
4120 | rule_dealloc, | |
abe529af BP |
4121 | rule_get_stats, |
4122 | rule_execute, | |
4123 | rule_modify_actions, | |
4124 | get_drop_frags, | |
4125 | set_drop_frags, | |
4126 | packet_out, | |
4127 | set_netflow, | |
4128 | get_netflow_ids, | |
4129 | set_sflow, | |
4130 | set_cfm, | |
a5610457 | 4131 | get_cfm_fault, |
abe529af BP |
4132 | bundle_set, |
4133 | bundle_remove, | |
4134 | mirror_set, | |
4135 | set_flood_vlans, | |
4136 | is_mirror_output_bundle, | |
4137 | }; |