]>
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; | |
8b28d864 | 1270 | mirror->aux = aux; |
abe529af BP |
1271 | mirror->out_vlan = -1; |
1272 | mirror->name = NULL; | |
1273 | } | |
1274 | ||
1275 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
1276 | free(mirror->name); | |
1277 | mirror->name = xstrdup(s->name); | |
1278 | } | |
1279 | ||
1280 | /* Get the new configuration. */ | |
1281 | if (s->out_bundle) { | |
1282 | out = bundle_lookup(ofproto, s->out_bundle); | |
1283 | if (!out) { | |
1284 | mirror_destroy(mirror); | |
1285 | return EINVAL; | |
1286 | } | |
1287 | out_vlan = -1; | |
1288 | } else { | |
1289 | out = NULL; | |
1290 | out_vlan = s->out_vlan; | |
1291 | } | |
1292 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
1293 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
1294 | ||
1295 | /* If the configuration has not changed, do nothing. */ | |
1296 | if (hmapx_equals(&srcs, &mirror->srcs) | |
1297 | && hmapx_equals(&dsts, &mirror->dsts) | |
1298 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
1299 | && mirror->out == out | |
1300 | && mirror->out_vlan == out_vlan) | |
1301 | { | |
1302 | hmapx_destroy(&srcs); | |
1303 | hmapx_destroy(&dsts); | |
1304 | return 0; | |
1305 | } | |
1306 | ||
1307 | hmapx_swap(&srcs, &mirror->srcs); | |
1308 | hmapx_destroy(&srcs); | |
1309 | ||
1310 | hmapx_swap(&dsts, &mirror->dsts); | |
1311 | hmapx_destroy(&dsts); | |
1312 | ||
1313 | free(mirror->vlans); | |
1314 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
1315 | ||
1316 | mirror->out = out; | |
1317 | mirror->out_vlan = out_vlan; | |
1318 | ||
1319 | /* Update bundles. */ | |
1320 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1321 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
1322 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
1323 | bundle->src_mirrors |= mirror_bit; | |
1324 | } else { | |
1325 | bundle->src_mirrors &= ~mirror_bit; | |
1326 | } | |
1327 | ||
1328 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
1329 | bundle->dst_mirrors |= mirror_bit; | |
1330 | } else { | |
1331 | bundle->dst_mirrors &= ~mirror_bit; | |
1332 | } | |
1333 | ||
1334 | if (mirror->out == bundle) { | |
1335 | bundle->mirror_out |= mirror_bit; | |
1336 | } else { | |
1337 | bundle->mirror_out &= ~mirror_bit; | |
1338 | } | |
1339 | } | |
1340 | ||
1341 | ofproto->need_revalidate = true; | |
1342 | mac_learning_flush(ofproto->ml); | |
1343 | ||
1344 | return 0; | |
1345 | } | |
1346 | ||
1347 | static void | |
1348 | mirror_destroy(struct ofmirror *mirror) | |
1349 | { | |
1350 | struct ofproto_dpif *ofproto; | |
1351 | mirror_mask_t mirror_bit; | |
1352 | struct ofbundle *bundle; | |
1353 | ||
1354 | if (!mirror) { | |
1355 | return; | |
1356 | } | |
1357 | ||
1358 | ofproto = mirror->ofproto; | |
1359 | ofproto->need_revalidate = true; | |
1360 | mac_learning_flush(ofproto->ml); | |
1361 | ||
1362 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1363 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1364 | bundle->src_mirrors &= ~mirror_bit; | |
1365 | bundle->dst_mirrors &= ~mirror_bit; | |
1366 | bundle->mirror_out &= ~mirror_bit; | |
1367 | } | |
1368 | ||
1369 | hmapx_destroy(&mirror->srcs); | |
1370 | hmapx_destroy(&mirror->dsts); | |
1371 | free(mirror->vlans); | |
1372 | ||
1373 | ofproto->mirrors[mirror->idx] = NULL; | |
1374 | free(mirror->name); | |
1375 | free(mirror); | |
1376 | } | |
1377 | ||
1378 | static int | |
1379 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
1380 | { | |
1381 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1382 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
1383 | ofproto->need_revalidate = true; | |
1384 | mac_learning_flush(ofproto->ml); | |
1385 | } | |
1386 | return 0; | |
1387 | } | |
1388 | ||
1389 | static bool | |
1390 | is_mirror_output_bundle(struct ofproto *ofproto_, void *aux) | |
1391 | { | |
1392 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1393 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
1394 | return bundle && bundle->mirror_out != 0; | |
1395 | } | |
1396 | \f | |
1397 | /* Ports. */ | |
1398 | ||
1399 | static struct ofport_dpif * | |
1400 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
1401 | { | |
7df6a8bd BP |
1402 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
1403 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
1404 | } |
1405 | ||
1406 | static struct ofport_dpif * | |
1407 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
1408 | { | |
1409 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
1410 | } | |
1411 | ||
1412 | static void | |
1413 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
1414 | struct dpif_port *dpif_port) | |
1415 | { | |
1416 | ofproto_port->name = dpif_port->name; | |
1417 | ofproto_port->type = dpif_port->type; | |
1418 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
1419 | } | |
1420 | ||
1421 | static void | |
1422 | port_run(struct ofport_dpif *ofport) | |
1423 | { | |
015e08bc EJ |
1424 | bool enable = netdev_get_carrier(ofport->up.netdev); |
1425 | ||
abe529af BP |
1426 | if (ofport->cfm) { |
1427 | cfm_run(ofport->cfm); | |
1428 | ||
1429 | if (cfm_should_send_ccm(ofport->cfm)) { | |
1430 | struct ofpbuf packet; | |
abe529af BP |
1431 | |
1432 | ofpbuf_init(&packet, 0); | |
c0a2e71d | 1433 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr); |
abe529af | 1434 | send_packet(ofproto_dpif_cast(ofport->up.ofproto), |
b2fda3ef | 1435 | ofport->odp_port, &packet); |
abe529af BP |
1436 | ofpbuf_uninit(&packet); |
1437 | } | |
015e08bc EJ |
1438 | |
1439 | enable = enable && !cfm_get_fault(ofport->cfm); | |
abe529af | 1440 | } |
015e08bc EJ |
1441 | |
1442 | if (ofport->bundle) { | |
1443 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
1444 | } | |
1445 | ||
daff3353 EJ |
1446 | if (ofport->may_enable != enable) { |
1447 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1448 | ||
1449 | if (ofproto->has_bundle_action) { | |
1450 | ofproto->need_revalidate = true; | |
1451 | } | |
1452 | } | |
1453 | ||
015e08bc | 1454 | ofport->may_enable = enable; |
abe529af BP |
1455 | } |
1456 | ||
1457 | static void | |
1458 | port_wait(struct ofport_dpif *ofport) | |
1459 | { | |
1460 | if (ofport->cfm) { | |
1461 | cfm_wait(ofport->cfm); | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | static int | |
1466 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
1467 | struct ofproto_port *ofproto_port) | |
1468 | { | |
1469 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1470 | struct dpif_port dpif_port; | |
1471 | int error; | |
1472 | ||
1473 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
1474 | if (!error) { | |
1475 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
1476 | } | |
1477 | return error; | |
1478 | } | |
1479 | ||
1480 | static int | |
1481 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
1482 | { | |
1483 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1484 | uint16_t odp_port; | |
1485 | int error; | |
1486 | ||
1487 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
1488 | if (!error) { | |
1489 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
1490 | } | |
1491 | return error; | |
1492 | } | |
1493 | ||
1494 | static int | |
1495 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
1496 | { | |
1497 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1498 | int error; | |
1499 | ||
1500 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
1501 | if (!error) { | |
1502 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
1503 | if (ofport) { | |
1504 | /* The caller is going to close ofport->up.netdev. If this is a | |
1505 | * bonded port, then the bond is using that netdev, so remove it | |
1506 | * from the bond. The client will need to reconfigure everything | |
1507 | * after deleting ports, so then the slave will get re-added. */ | |
1508 | bundle_remove(&ofport->up); | |
1509 | } | |
1510 | } | |
1511 | return error; | |
1512 | } | |
1513 | ||
1514 | struct port_dump_state { | |
1515 | struct dpif_port_dump dump; | |
1516 | bool done; | |
1517 | }; | |
1518 | ||
1519 | static int | |
1520 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
1521 | { | |
1522 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1523 | struct port_dump_state *state; | |
1524 | ||
1525 | *statep = state = xmalloc(sizeof *state); | |
1526 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
1527 | state->done = false; | |
1528 | return 0; | |
1529 | } | |
1530 | ||
1531 | static int | |
1532 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
1533 | struct ofproto_port *port) | |
1534 | { | |
1535 | struct port_dump_state *state = state_; | |
1536 | struct dpif_port dpif_port; | |
1537 | ||
1538 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
1539 | ofproto_port_from_dpif_port(port, &dpif_port); | |
1540 | return 0; | |
1541 | } else { | |
1542 | int error = dpif_port_dump_done(&state->dump); | |
1543 | state->done = true; | |
1544 | return error ? error : EOF; | |
1545 | } | |
1546 | } | |
1547 | ||
1548 | static int | |
1549 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
1550 | { | |
1551 | struct port_dump_state *state = state_; | |
1552 | ||
1553 | if (!state->done) { | |
1554 | dpif_port_dump_done(&state->dump); | |
1555 | } | |
1556 | free(state); | |
1557 | return 0; | |
1558 | } | |
1559 | ||
1560 | static int | |
1561 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
1562 | { | |
1563 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1564 | return dpif_port_poll(ofproto->dpif, devnamep); | |
1565 | } | |
1566 | ||
1567 | static void | |
1568 | port_poll_wait(const struct ofproto *ofproto_) | |
1569 | { | |
1570 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1571 | dpif_port_poll_wait(ofproto->dpif); | |
1572 | } | |
1573 | ||
1574 | static int | |
1575 | port_is_lacp_current(const struct ofport *ofport_) | |
1576 | { | |
1577 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1578 | return (ofport->bundle && ofport->bundle->lacp | |
1579 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
1580 | : -1); | |
1581 | } | |
1582 | \f | |
1583 | /* Upcall handling. */ | |
1584 | ||
1585 | /* Given 'upcall', of type DPIF_UC_ACTION or DPIF_UC_MISS, sends an | |
1586 | * OFPT_PACKET_IN message to each OpenFlow controller as necessary according to | |
1587 | * their individual configurations. | |
1588 | * | |
1589 | * If 'clone' is true, the caller retains ownership of 'upcall->packet'. | |
1590 | * Otherwise, ownership is transferred to this function. */ | |
1591 | static void | |
1592 | send_packet_in(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall, | |
1593 | const struct flow *flow, bool clone) | |
1594 | { | |
1595 | struct ofputil_packet_in pin; | |
1596 | ||
1597 | pin.packet = upcall->packet; | |
1598 | pin.in_port = flow->in_port; | |
1599 | pin.reason = upcall->type == DPIF_UC_MISS ? OFPR_NO_MATCH : OFPR_ACTION; | |
1600 | pin.buffer_id = 0; /* not yet known */ | |
1601 | pin.send_len = upcall->userdata; | |
78bd1cd0 | 1602 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, |
abe529af BP |
1603 | clone ? NULL : upcall->packet); |
1604 | } | |
1605 | ||
1606 | static bool | |
1607 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
1608 | const struct ofpbuf *packet) | |
1609 | { | |
1610 | if (cfm_should_process_flow(flow)) { | |
1611 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); | |
da37ebac | 1612 | if (packet && ofport && ofport->cfm) { |
abe529af BP |
1613 | cfm_process_heartbeat(ofport->cfm, packet); |
1614 | } | |
1615 | return true; | |
1616 | } else if (flow->dl_type == htons(ETH_TYPE_LACP)) { | |
1617 | struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); | |
da37ebac | 1618 | if (packet && port && port->bundle && port->bundle->lacp) { |
abe529af BP |
1619 | const struct lacp_pdu *pdu = parse_lacp_packet(packet); |
1620 | if (pdu) { | |
1621 | lacp_process_pdu(port->bundle->lacp, port, pdu); | |
1622 | } | |
abe529af | 1623 | } |
da37ebac | 1624 | return true; |
abe529af BP |
1625 | } |
1626 | return false; | |
1627 | } | |
1628 | ||
1629 | static void | |
1630 | handle_miss_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) | |
1631 | { | |
1632 | struct facet *facet; | |
1633 | struct flow flow; | |
1634 | ||
1635 | /* Obtain in_port and tun_id, at least. */ | |
1636 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
1637 | ||
1638 | /* Set header pointers in 'flow'. */ | |
1639 | flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); | |
1640 | ||
1641 | /* Handle 802.1ag and LACP. */ | |
1642 | if (process_special(ofproto, &flow, upcall->packet)) { | |
1643 | ofpbuf_delete(upcall->packet); | |
6c1491fb | 1644 | ofproto->n_matches++; |
abe529af BP |
1645 | return; |
1646 | } | |
1647 | ||
1648 | /* Check with in-band control to see if this packet should be sent | |
1649 | * to the local port regardless of the flow table. */ | |
1650 | if (connmgr_msg_in_hook(ofproto->up.connmgr, &flow, upcall->packet)) { | |
f7f2ec05 | 1651 | send_packet(ofproto, ODPP_LOCAL, upcall->packet); |
abe529af BP |
1652 | } |
1653 | ||
1654 | facet = facet_lookup_valid(ofproto, &flow); | |
1655 | if (!facet) { | |
1656 | struct rule_dpif *rule = rule_dpif_lookup(ofproto, &flow); | |
1657 | if (!rule) { | |
1658 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
1659 | struct ofport_dpif *port = get_ofp_port(ofproto, flow.in_port); | |
1660 | if (port) { | |
1661 | if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) { | |
1662 | COVERAGE_INC(ofproto_dpif_no_packet_in); | |
1663 | /* XXX install 'drop' flow entry */ | |
1664 | ofpbuf_delete(upcall->packet); | |
1665 | return; | |
1666 | } | |
1667 | } else { | |
1668 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
1669 | flow.in_port); | |
1670 | } | |
1671 | ||
1672 | send_packet_in(ofproto, upcall, &flow, false); | |
1673 | return; | |
1674 | } | |
1675 | ||
1676 | facet = facet_create(rule, &flow, upcall->packet); | |
1677 | } else if (!facet->may_install) { | |
1678 | /* The facet is not installable, that is, we need to process every | |
1679 | * packet, so process the current packet's actions into 'facet'. */ | |
1680 | facet_make_actions(ofproto, facet, upcall->packet); | |
1681 | } | |
1682 | ||
1683 | if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
1684 | /* | |
1685 | * Extra-special case for fail-open mode. | |
1686 | * | |
1687 | * We are in fail-open mode and the packet matched the fail-open rule, | |
1688 | * but we are connected to a controller too. We should send the packet | |
1689 | * up to the controller in the hope that it will try to set up a flow | |
1690 | * and thereby allow us to exit fail-open. | |
1691 | * | |
1692 | * See the top-level comment in fail-open.c for more information. | |
1693 | */ | |
1694 | send_packet_in(ofproto, upcall, &flow, true); | |
1695 | } | |
1696 | ||
1697 | facet_execute(ofproto, facet, upcall->packet); | |
1698 | facet_install(ofproto, facet, false); | |
6c1491fb | 1699 | ofproto->n_matches++; |
abe529af BP |
1700 | } |
1701 | ||
1702 | static void | |
1703 | handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) | |
1704 | { | |
1705 | struct flow flow; | |
1706 | ||
1707 | switch (upcall->type) { | |
1708 | case DPIF_UC_ACTION: | |
1709 | COVERAGE_INC(ofproto_dpif_ctlr_action); | |
1710 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
1711 | send_packet_in(ofproto, upcall, &flow, false); | |
1712 | break; | |
1713 | ||
1714 | case DPIF_UC_SAMPLE: | |
1715 | if (ofproto->sflow) { | |
1716 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
bae473fe | 1717 | dpif_sflow_received(ofproto->sflow, upcall, &flow); |
abe529af BP |
1718 | } |
1719 | ofpbuf_delete(upcall->packet); | |
1720 | break; | |
1721 | ||
1722 | case DPIF_UC_MISS: | |
1723 | handle_miss_upcall(ofproto, upcall); | |
1724 | break; | |
1725 | ||
1726 | case DPIF_N_UC_TYPES: | |
1727 | default: | |
1728 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
1729 | break; | |
1730 | } | |
1731 | } | |
1732 | \f | |
1733 | /* Flow expiration. */ | |
1734 | ||
1735 | static int facet_max_idle(const struct ofproto_dpif *); | |
1736 | static void update_stats(struct ofproto_dpif *); | |
1737 | static void rule_expire(struct rule_dpif *); | |
1738 | static void expire_facets(struct ofproto_dpif *, int dp_max_idle); | |
1739 | ||
1740 | /* This function is called periodically by run(). Its job is to collect | |
1741 | * updates for the flows that have been installed into the datapath, most | |
1742 | * importantly when they last were used, and then use that information to | |
1743 | * expire flows that have not been used recently. | |
1744 | * | |
1745 | * Returns the number of milliseconds after which it should be called again. */ | |
1746 | static int | |
1747 | expire(struct ofproto_dpif *ofproto) | |
1748 | { | |
1749 | struct rule_dpif *rule, *next_rule; | |
1750 | struct cls_cursor cursor; | |
1751 | int dp_max_idle; | |
1752 | ||
1753 | /* Update stats for each flow in the datapath. */ | |
1754 | update_stats(ofproto); | |
1755 | ||
1756 | /* Expire facets that have been idle too long. */ | |
1757 | dp_max_idle = facet_max_idle(ofproto); | |
1758 | expire_facets(ofproto, dp_max_idle); | |
1759 | ||
1760 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
6c1491fb | 1761 | cls_cursor_init(&cursor, &ofproto->up.tables[0], NULL); |
abe529af BP |
1762 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { |
1763 | rule_expire(rule); | |
1764 | } | |
1765 | ||
1766 | /* All outstanding data in existing flows has been accounted, so it's a | |
1767 | * good time to do bond rebalancing. */ | |
1768 | if (ofproto->has_bonded_bundles) { | |
1769 | struct ofbundle *bundle; | |
1770 | ||
1771 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1772 | if (bundle->bond) { | |
1773 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
1774 | } | |
1775 | } | |
1776 | } | |
1777 | ||
1778 | return MIN(dp_max_idle, 1000); | |
1779 | } | |
1780 | ||
1781 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
1782 | * | |
1783 | * This function also pushes statistics updates to rules which each facet | |
1784 | * resubmits into. Generally these statistics will be accurate. However, if a | |
1785 | * facet changes the rule it resubmits into at some time in between | |
1786 | * update_stats() runs, it is possible that statistics accrued to the | |
1787 | * old rule will be incorrectly attributed to the new rule. This could be | |
1788 | * avoided by calling update_stats() whenever rules are created or | |
1789 | * deleted. However, the performance impact of making so many calls to the | |
1790 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
1791 | */ | |
1792 | static void | |
1793 | update_stats(struct ofproto_dpif *p) | |
1794 | { | |
1795 | const struct dpif_flow_stats *stats; | |
1796 | struct dpif_flow_dump dump; | |
1797 | const struct nlattr *key; | |
1798 | size_t key_len; | |
1799 | ||
1800 | dpif_flow_dump_start(&dump, p->dpif); | |
1801 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
1802 | struct facet *facet; | |
1803 | struct flow flow; | |
1804 | ||
1805 | if (odp_flow_key_to_flow(key, key_len, &flow)) { | |
1806 | struct ds s; | |
1807 | ||
1808 | ds_init(&s); | |
1809 | odp_flow_key_format(key, key_len, &s); | |
1810 | VLOG_WARN_RL(&rl, "failed to convert ODP flow key to flow: %s", | |
1811 | ds_cstr(&s)); | |
1812 | ds_destroy(&s); | |
1813 | ||
1814 | continue; | |
1815 | } | |
1816 | facet = facet_find(p, &flow); | |
1817 | ||
1818 | if (facet && facet->installed) { | |
1819 | ||
1820 | if (stats->n_packets >= facet->dp_packet_count) { | |
1821 | uint64_t extra = stats->n_packets - facet->dp_packet_count; | |
1822 | facet->packet_count += extra; | |
1823 | } else { | |
1824 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
1825 | } | |
1826 | ||
1827 | if (stats->n_bytes >= facet->dp_byte_count) { | |
1828 | facet->byte_count += stats->n_bytes - facet->dp_byte_count; | |
1829 | } else { | |
1830 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
1831 | } | |
1832 | ||
1833 | facet->dp_packet_count = stats->n_packets; | |
1834 | facet->dp_byte_count = stats->n_bytes; | |
1835 | ||
1836 | facet_update_time(p, facet, stats->used); | |
1837 | facet_account(p, facet, stats->n_bytes); | |
1838 | facet_push_stats(facet); | |
1839 | } else { | |
1840 | /* There's a flow in the datapath that we know nothing about. | |
1841 | * Delete it. */ | |
1842 | COVERAGE_INC(facet_unexpected); | |
1843 | dpif_flow_del(p->dpif, key, key_len, NULL); | |
1844 | } | |
1845 | } | |
1846 | dpif_flow_dump_done(&dump); | |
1847 | } | |
1848 | ||
1849 | /* Calculates and returns the number of milliseconds of idle time after which | |
1850 | * facets should expire from the datapath and we should fold their statistics | |
1851 | * into their parent rules in userspace. */ | |
1852 | static int | |
1853 | facet_max_idle(const struct ofproto_dpif *ofproto) | |
1854 | { | |
1855 | /* | |
1856 | * Idle time histogram. | |
1857 | * | |
1858 | * Most of the time a switch has a relatively small number of facets. When | |
1859 | * this is the case we might as well keep statistics for all of them in | |
1860 | * userspace and to cache them in the kernel datapath for performance as | |
1861 | * well. | |
1862 | * | |
1863 | * As the number of facets increases, the memory required to maintain | |
1864 | * statistics about them in userspace and in the kernel becomes | |
1865 | * significant. However, with a large number of facets it is likely that | |
1866 | * only a few of them are "heavy hitters" that consume a large amount of | |
1867 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
1868 | * kernel and maintaining in userspaces; other facets we can discard. | |
1869 | * | |
1870 | * The technique used to compute the idle time is to build a histogram with | |
1871 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet | |
1872 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
1873 | * After the histogram has been built, we compute the cutoff so that only | |
1874 | * the most-recently-used 1% of facets (but at least 1000 flows) are kept | |
1875 | * cached. At least the most-recently-used bucket of facets is kept, so | |
1876 | * actually an arbitrary number of facets can be kept in any given | |
1877 | * expiration run (though the next run will delete most of those unless | |
1878 | * they receive additional data). | |
1879 | * | |
1880 | * This requires a second pass through the facets, in addition to the pass | |
1881 | * made by update_stats(), because the former function never looks | |
1882 | * at uninstallable facets. | |
1883 | */ | |
1884 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
1885 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
1886 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 1887 | int total, subtotal, bucket; |
abe529af | 1888 | struct facet *facet; |
abe529af BP |
1889 | long long int now; |
1890 | int i; | |
1891 | ||
1892 | total = hmap_count(&ofproto->facets); | |
1893 | if (total <= 1000) { | |
1894 | return N_BUCKETS * BUCKET_WIDTH; | |
1895 | } | |
1896 | ||
1897 | /* Build histogram. */ | |
1898 | now = time_msec(); | |
1899 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
1900 | long long int idle = now - facet->used; | |
1901 | int bucket = (idle <= 0 ? 0 | |
1902 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
1903 | : (unsigned int) idle / BUCKET_WIDTH); | |
1904 | buckets[bucket]++; | |
1905 | } | |
1906 | ||
1907 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
1908 | subtotal = bucket = 0; |
1909 | do { | |
1910 | subtotal += buckets[bucket++]; | |
1911 | } while (bucket < N_BUCKETS && subtotal < MAX(1000, total / 100)); | |
abe529af BP |
1912 | |
1913 | if (VLOG_IS_DBG_ENABLED()) { | |
1914 | struct ds s; | |
1915 | ||
1916 | ds_init(&s); | |
1917 | ds_put_cstr(&s, "keep"); | |
1918 | for (i = 0; i < N_BUCKETS; i++) { | |
1919 | if (i == bucket) { | |
1920 | ds_put_cstr(&s, ", drop"); | |
1921 | } | |
1922 | if (buckets[i]) { | |
1923 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
1924 | } | |
1925 | } | |
1926 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
1927 | ds_destroy(&s); | |
1928 | } | |
1929 | ||
1930 | return bucket * BUCKET_WIDTH; | |
1931 | } | |
1932 | ||
1933 | static void | |
1934 | facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
1935 | { | |
1936 | if (ofproto->netflow && !facet_is_controller_flow(facet) && | |
1937 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
1938 | struct ofexpired expired; | |
1939 | ||
1940 | if (facet->installed) { | |
1941 | struct dpif_flow_stats stats; | |
1942 | ||
1943 | facet_put__(ofproto, facet, facet->actions, facet->actions_len, | |
1944 | &stats); | |
1945 | facet_update_stats(ofproto, facet, &stats); | |
1946 | } | |
1947 | ||
1948 | expired.flow = facet->flow; | |
1949 | expired.packet_count = facet->packet_count; | |
1950 | expired.byte_count = facet->byte_count; | |
1951 | expired.used = facet->used; | |
1952 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
1953 | } | |
1954 | } | |
1955 | ||
1956 | static void | |
1957 | expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle) | |
1958 | { | |
1959 | long long int cutoff = time_msec() - dp_max_idle; | |
1960 | struct facet *facet, *next_facet; | |
1961 | ||
1962 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
1963 | facet_active_timeout(ofproto, facet); | |
1964 | if (facet->used < cutoff) { | |
1965 | facet_remove(ofproto, facet); | |
1966 | } | |
1967 | } | |
1968 | } | |
1969 | ||
1970 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
1971 | * then delete it entirely. */ | |
1972 | static void | |
1973 | rule_expire(struct rule_dpif *rule) | |
1974 | { | |
1975 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
1976 | struct facet *facet, *next_facet; | |
1977 | long long int now; | |
1978 | uint8_t reason; | |
1979 | ||
1980 | /* Has 'rule' expired? */ | |
1981 | now = time_msec(); | |
1982 | if (rule->up.hard_timeout | |
1983 | && now > rule->up.created + rule->up.hard_timeout * 1000) { | |
1984 | reason = OFPRR_HARD_TIMEOUT; | |
1985 | } else if (rule->up.idle_timeout && list_is_empty(&rule->facets) | |
1986 | && now > rule->used + rule->up.idle_timeout * 1000) { | |
1987 | reason = OFPRR_IDLE_TIMEOUT; | |
1988 | } else { | |
1989 | return; | |
1990 | } | |
1991 | ||
1992 | COVERAGE_INC(ofproto_dpif_expired); | |
1993 | ||
1994 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
1995 | * timeout, because that only happens when the rule has no facets left.) */ | |
1996 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
1997 | facet_remove(ofproto, facet); | |
1998 | } | |
1999 | ||
2000 | /* Get rid of the rule. */ | |
2001 | ofproto_rule_expire(&rule->up, reason); | |
2002 | } | |
2003 | \f | |
2004 | /* Facets. */ | |
2005 | ||
2006 | /* Creates and returns a new facet owned by 'rule', given a 'flow' and an | |
2007 | * example 'packet' within that flow. | |
2008 | * | |
2009 | * The caller must already have determined that no facet with an identical | |
2010 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
2011 | * the ofproto's classifier table. */ | |
2012 | static struct facet * | |
2013 | facet_create(struct rule_dpif *rule, const struct flow *flow, | |
2014 | const struct ofpbuf *packet) | |
2015 | { | |
2016 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2017 | struct facet *facet; | |
2018 | ||
2019 | facet = xzalloc(sizeof *facet); | |
2020 | facet->used = time_msec(); | |
2021 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2022 | list_push_back(&rule->facets, &facet->list_node); | |
2023 | facet->rule = rule; | |
2024 | facet->flow = *flow; | |
2025 | netflow_flow_init(&facet->nf_flow); | |
2026 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2027 | ||
2028 | facet_make_actions(ofproto, facet, packet); | |
2029 | ||
2030 | return facet; | |
2031 | } | |
2032 | ||
2033 | static void | |
2034 | facet_free(struct facet *facet) | |
2035 | { | |
2036 | free(facet->actions); | |
2037 | free(facet); | |
2038 | } | |
2039 | ||
2040 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on | |
2041 | * 'packet', which arrived on 'in_port'. | |
2042 | * | |
2043 | * Takes ownership of 'packet'. */ | |
2044 | static bool | |
2045 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2046 | const struct nlattr *odp_actions, size_t actions_len, | |
2047 | struct ofpbuf *packet) | |
2048 | { | |
2049 | if (actions_len == NLA_ALIGN(NLA_HDRLEN + sizeof(uint64_t)) | |
b85d8d61 | 2050 | && odp_actions->nla_type == ODP_ACTION_ATTR_USERSPACE) { |
abe529af BP |
2051 | /* As an optimization, avoid a round-trip from userspace to kernel to |
2052 | * userspace. This also avoids possibly filling up kernel packet | |
2053 | * buffers along the way. */ | |
2054 | struct dpif_upcall upcall; | |
2055 | ||
2056 | upcall.type = DPIF_UC_ACTION; | |
2057 | upcall.packet = packet; | |
2058 | upcall.key = NULL; | |
2059 | upcall.key_len = 0; | |
2060 | upcall.userdata = nl_attr_get_u64(odp_actions); | |
2061 | upcall.sample_pool = 0; | |
2062 | upcall.actions = NULL; | |
2063 | upcall.actions_len = 0; | |
2064 | ||
2065 | send_packet_in(ofproto, &upcall, flow, false); | |
2066 | ||
2067 | return true; | |
2068 | } else { | |
80e5eed9 BP |
2069 | struct odputil_keybuf keybuf; |
2070 | struct ofpbuf key; | |
abe529af BP |
2071 | int error; |
2072 | ||
80e5eed9 BP |
2073 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
2074 | odp_flow_key_from_flow(&key, flow); | |
2075 | ||
2076 | error = dpif_execute(ofproto->dpif, key.data, key.size, | |
2077 | odp_actions, actions_len, packet); | |
2078 | ||
abe529af BP |
2079 | ofpbuf_delete(packet); |
2080 | return !error; | |
2081 | } | |
2082 | } | |
2083 | ||
2084 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s | |
2085 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
2086 | * ofp_packet_in) bytes of headroom. | |
2087 | * | |
2088 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, | |
2089 | * applying flow_extract() to 'packet' would yield the same flow as | |
2090 | * 'facet->flow'. | |
2091 | * | |
2092 | * 'facet' must have accurately composed ODP actions; that is, it must not be | |
2093 | * in need of revalidation. | |
2094 | * | |
2095 | * Takes ownership of 'packet'. */ | |
2096 | static void | |
2097 | facet_execute(struct ofproto_dpif *ofproto, struct facet *facet, | |
2098 | struct ofpbuf *packet) | |
2099 | { | |
2100 | struct dpif_flow_stats stats; | |
2101 | ||
2102 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2103 | ||
2104 | flow_extract_stats(&facet->flow, packet, &stats); | |
2105 | stats.used = time_msec(); | |
2106 | if (execute_odp_actions(ofproto, &facet->flow, | |
2107 | facet->actions, facet->actions_len, packet)) { | |
2108 | facet_update_stats(ofproto, facet, &stats); | |
2109 | } | |
2110 | } | |
2111 | ||
2112 | /* Remove 'facet' from 'ofproto' and free up the associated memory: | |
2113 | * | |
2114 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2115 | * rule's statistics, via facet_uninstall(). | |
2116 | * | |
2117 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2118 | */ | |
2119 | static void | |
2120 | facet_remove(struct ofproto_dpif *ofproto, struct facet *facet) | |
2121 | { | |
2122 | facet_uninstall(ofproto, facet); | |
2123 | facet_flush_stats(ofproto, facet); | |
2124 | hmap_remove(&ofproto->facets, &facet->hmap_node); | |
2125 | list_remove(&facet->list_node); | |
2126 | facet_free(facet); | |
2127 | } | |
2128 | ||
2129 | /* Composes the ODP actions for 'facet' based on its rule's actions. */ | |
2130 | static void | |
2131 | facet_make_actions(struct ofproto_dpif *p, struct facet *facet, | |
2132 | const struct ofpbuf *packet) | |
2133 | { | |
2134 | const struct rule_dpif *rule = facet->rule; | |
2135 | struct ofpbuf *odp_actions; | |
2136 | struct action_xlate_ctx ctx; | |
2137 | ||
2138 | action_xlate_ctx_init(&ctx, p, &facet->flow, packet); | |
2139 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
2140 | facet->tags = ctx.tags; | |
2141 | facet->may_install = ctx.may_set_up_flow; | |
2142 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
2143 | ||
2144 | if (facet->actions_len != odp_actions->size | |
2145 | || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { | |
2146 | free(facet->actions); | |
2147 | facet->actions_len = odp_actions->size; | |
2148 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
2149 | } | |
2150 | ||
2151 | ofpbuf_delete(odp_actions); | |
2152 | } | |
2153 | ||
3a88e544 BP |
2154 | /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len' |
2155 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
2156 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
2157 | * since 'facet' was last updated. | |
2158 | * | |
2159 | * Returns 0 if successful, otherwise a positive errno value.*/ | |
abe529af BP |
2160 | static int |
2161 | facet_put__(struct ofproto_dpif *ofproto, struct facet *facet, | |
2162 | const struct nlattr *actions, size_t actions_len, | |
2163 | struct dpif_flow_stats *stats) | |
2164 | { | |
2165 | struct odputil_keybuf keybuf; | |
2166 | enum dpif_flow_put_flags flags; | |
2167 | struct ofpbuf key; | |
3a88e544 | 2168 | int ret; |
abe529af BP |
2169 | |
2170 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
2171 | if (stats) { | |
2172 | flags |= DPIF_FP_ZERO_STATS; | |
abe529af BP |
2173 | } |
2174 | ||
2175 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2176 | odp_flow_key_from_flow(&key, &facet->flow); | |
2177 | ||
3a88e544 BP |
2178 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, |
2179 | actions, actions_len, stats); | |
2180 | ||
2181 | if (stats) { | |
2182 | facet_reset_dp_stats(facet, stats); | |
2183 | } | |
2184 | ||
2185 | return ret; | |
abe529af BP |
2186 | } |
2187 | ||
2188 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If | |
2189 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2190 | * 'facet'. */ | |
2191 | static void | |
2192 | facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats) | |
2193 | { | |
2194 | struct dpif_flow_stats stats; | |
2195 | ||
2196 | if (facet->may_install | |
2197 | && !facet_put__(p, facet, facet->actions, facet->actions_len, | |
2198 | zero_stats ? &stats : NULL)) { | |
2199 | facet->installed = true; | |
2200 | } | |
2201 | } | |
2202 | ||
d78be13b BP |
2203 | static int |
2204 | vlan_tci_to_openflow_vlan(ovs_be16 vlan_tci) | |
2205 | { | |
2206 | return vlan_tci != htons(0) ? vlan_tci_to_vid(vlan_tci) : OFP_VLAN_NONE; | |
2207 | } | |
2208 | ||
abe529af BP |
2209 | static void |
2210 | facet_account(struct ofproto_dpif *ofproto, | |
2211 | struct facet *facet, uint64_t extra_bytes) | |
2212 | { | |
2213 | uint64_t total_bytes, n_bytes; | |
2214 | struct ofbundle *in_bundle; | |
2215 | const struct nlattr *a; | |
2216 | tag_type dummy = 0; | |
2217 | unsigned int left; | |
d78be13b | 2218 | ovs_be16 vlan_tci; |
abe529af BP |
2219 | int vlan; |
2220 | ||
2221 | total_bytes = facet->byte_count + extra_bytes; | |
2222 | if (total_bytes <= facet->accounted_bytes) { | |
2223 | return; | |
2224 | } | |
2225 | n_bytes = total_bytes - facet->accounted_bytes; | |
2226 | facet->accounted_bytes = total_bytes; | |
2227 | ||
2228 | /* Test that 'tags' is nonzero to ensure that only flows that include an | |
2229 | * OFPP_NORMAL action are used for learning and bond slave rebalancing. | |
2230 | * This works because OFPP_NORMAL always sets a nonzero tag value. | |
2231 | * | |
2232 | * Feed information from the active flows back into the learning table to | |
2233 | * ensure that table is always in sync with what is actually flowing | |
2234 | * through the datapath. */ | |
2235 | if (!facet->tags | |
2236 | || !is_admissible(ofproto, &facet->flow, false, &dummy, | |
2237 | &vlan, &in_bundle)) { | |
2238 | return; | |
2239 | } | |
2240 | ||
2241 | update_learning_table(ofproto, &facet->flow, vlan, in_bundle); | |
2242 | ||
2243 | if (!ofproto->has_bonded_bundles) { | |
2244 | return; | |
2245 | } | |
d78be13b BP |
2246 | |
2247 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
2248 | * as a basis. We also need to track the actual VLAN on which the packet | |
2249 | * is going to be sent to ensure that it matches the one passed to | |
2250 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
2251 | * hash bucket.) */ | |
2252 | vlan_tci = facet->flow.vlan_tci; | |
abe529af | 2253 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) { |
d78be13b | 2254 | struct ofport_dpif *port; |
abe529af | 2255 | |
d78be13b BP |
2256 | switch (nl_attr_type(a)) { |
2257 | case ODP_ACTION_ATTR_OUTPUT: | |
abe529af BP |
2258 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
2259 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b BP |
2260 | bond_account(port->bundle->bond, &facet->flow, |
2261 | vlan_tci_to_openflow_vlan(vlan_tci), n_bytes); | |
abe529af | 2262 | } |
d78be13b BP |
2263 | break; |
2264 | ||
2265 | case ODP_ACTION_ATTR_STRIP_VLAN: | |
2266 | vlan_tci = htons(0); | |
2267 | break; | |
2268 | ||
2269 | case ODP_ACTION_ATTR_SET_DL_TCI: | |
2270 | vlan_tci = nl_attr_get_be16(a); | |
2271 | break; | |
abe529af BP |
2272 | } |
2273 | } | |
2274 | } | |
2275 | ||
2276 | /* If 'rule' is installed in the datapath, uninstalls it. */ | |
2277 | static void | |
2278 | facet_uninstall(struct ofproto_dpif *p, struct facet *facet) | |
2279 | { | |
2280 | if (facet->installed) { | |
2281 | struct odputil_keybuf keybuf; | |
2282 | struct dpif_flow_stats stats; | |
2283 | struct ofpbuf key; | |
3a88e544 | 2284 | int error; |
abe529af BP |
2285 | |
2286 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2287 | odp_flow_key_from_flow(&key, &facet->flow); | |
2288 | ||
3a88e544 BP |
2289 | error = dpif_flow_del(p->dpif, key.data, key.size, &stats); |
2290 | facet_reset_dp_stats(facet, &stats); | |
2291 | if (!error) { | |
abe529af BP |
2292 | facet_update_stats(p, facet, &stats); |
2293 | } | |
2294 | facet->installed = false; | |
abe529af BP |
2295 | } else { |
2296 | assert(facet->dp_packet_count == 0); | |
2297 | assert(facet->dp_byte_count == 0); | |
2298 | } | |
2299 | } | |
2300 | ||
2301 | /* Returns true if the only action for 'facet' is to send to the controller. | |
2302 | * (We don't report NetFlow expiration messages for such facets because they | |
2303 | * are just part of the control logic for the network, not real traffic). */ | |
2304 | static bool | |
2305 | facet_is_controller_flow(struct facet *facet) | |
2306 | { | |
2307 | return (facet | |
2308 | && facet->rule->up.n_actions == 1 | |
2309 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
2310 | htons(OFPP_CONTROLLER))); | |
2311 | } | |
2312 | ||
3a88e544 BP |
2313 | /* Resets 'facet''s datapath statistics counters. This should be called when |
2314 | * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null, | |
2315 | * it should contain the statistics returned by dpif when 'facet' was reset in | |
2316 | * the datapath. 'stats' will be modified to only included statistics new | |
2317 | * since 'facet' was last updated. */ | |
2318 | static void | |
2319 | facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats) | |
2320 | { | |
2321 | if (stats && facet->dp_packet_count <= stats->n_packets | |
2322 | && facet->dp_byte_count <= stats->n_bytes) { | |
2323 | stats->n_packets -= facet->dp_packet_count; | |
2324 | stats->n_bytes -= facet->dp_byte_count; | |
2325 | } | |
2326 | ||
2327 | facet->dp_packet_count = 0; | |
2328 | facet->dp_byte_count = 0; | |
2329 | } | |
2330 | ||
abe529af BP |
2331 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
2332 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
2333 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
2334 | * its packet and byte counts before this function is called. */ | |
2335 | static void | |
2336 | facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet) | |
2337 | { | |
2338 | assert(!facet->dp_byte_count); | |
2339 | assert(!facet->dp_packet_count); | |
2340 | ||
2341 | facet_push_stats(facet); | |
2342 | facet_account(ofproto, facet, 0); | |
2343 | ||
2344 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
2345 | struct ofexpired expired; | |
2346 | expired.flow = facet->flow; | |
2347 | expired.packet_count = facet->packet_count; | |
2348 | expired.byte_count = facet->byte_count; | |
2349 | expired.used = facet->used; | |
2350 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2351 | } | |
2352 | ||
2353 | facet->rule->packet_count += facet->packet_count; | |
2354 | facet->rule->byte_count += facet->byte_count; | |
2355 | ||
2356 | /* Reset counters to prevent double counting if 'facet' ever gets | |
2357 | * reinstalled. */ | |
2358 | facet->packet_count = 0; | |
2359 | facet->byte_count = 0; | |
2360 | facet->rs_packet_count = 0; | |
2361 | facet->rs_byte_count = 0; | |
2362 | facet->accounted_bytes = 0; | |
2363 | ||
2364 | netflow_flow_clear(&facet->nf_flow); | |
2365 | } | |
2366 | ||
2367 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2368 | * Returns it if found, otherwise a null pointer. | |
2369 | * | |
2370 | * The returned facet might need revalidation; use facet_lookup_valid() | |
2371 | * instead if that is important. */ | |
2372 | static struct facet * | |
2373 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) | |
2374 | { | |
2375 | struct facet *facet; | |
2376 | ||
2377 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
2378 | &ofproto->facets) { | |
2379 | if (flow_equal(flow, &facet->flow)) { | |
2380 | return facet; | |
2381 | } | |
2382 | } | |
2383 | ||
2384 | return NULL; | |
2385 | } | |
2386 | ||
2387 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2388 | * Returns it if found, otherwise a null pointer. | |
2389 | * | |
2390 | * The returned facet is guaranteed to be valid. */ | |
2391 | static struct facet * | |
2392 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) | |
2393 | { | |
2394 | struct facet *facet = facet_find(ofproto, flow); | |
2395 | ||
2396 | /* The facet we found might not be valid, since we could be in need of | |
2397 | * revalidation. If it is not valid, don't return it. */ | |
2398 | if (facet | |
2399 | && ofproto->need_revalidate | |
2400 | && !facet_revalidate(ofproto, facet)) { | |
2401 | COVERAGE_INC(facet_invalidated); | |
2402 | return NULL; | |
2403 | } | |
2404 | ||
2405 | return facet; | |
2406 | } | |
2407 | ||
2408 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': | |
2409 | * | |
2410 | * - If the rule found is different from 'facet''s current rule, moves | |
2411 | * 'facet' to the new rule and recompiles its actions. | |
2412 | * | |
2413 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
2414 | * where it is and recompiles its actions anyway. | |
2415 | * | |
2416 | * - If there is none, destroys 'facet'. | |
2417 | * | |
2418 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
2419 | static bool | |
2420 | facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet) | |
2421 | { | |
2422 | struct action_xlate_ctx ctx; | |
2423 | struct ofpbuf *odp_actions; | |
2424 | struct rule_dpif *new_rule; | |
2425 | bool actions_changed; | |
2426 | ||
2427 | COVERAGE_INC(facet_revalidate); | |
2428 | ||
2429 | /* Determine the new rule. */ | |
2430 | new_rule = rule_dpif_lookup(ofproto, &facet->flow); | |
2431 | if (!new_rule) { | |
2432 | /* No new rule, so delete the facet. */ | |
2433 | facet_remove(ofproto, facet); | |
2434 | return false; | |
2435 | } | |
2436 | ||
2437 | /* Calculate new ODP actions. | |
2438 | * | |
2439 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
2440 | * emit a NetFlow expiration and, if so, we need to have the old state | |
2441 | * around to properly compose it. */ | |
2442 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); | |
2443 | odp_actions = xlate_actions(&ctx, | |
2444 | new_rule->up.actions, new_rule->up.n_actions); | |
2445 | actions_changed = (facet->actions_len != odp_actions->size | |
2446 | || memcmp(facet->actions, odp_actions->data, | |
2447 | facet->actions_len)); | |
2448 | ||
2449 | /* If the ODP actions changed or the installability changed, then we need | |
2450 | * to talk to the datapath. */ | |
2451 | if (actions_changed || ctx.may_set_up_flow != facet->installed) { | |
2452 | if (ctx.may_set_up_flow) { | |
2453 | struct dpif_flow_stats stats; | |
2454 | ||
2455 | facet_put__(ofproto, facet, | |
2456 | odp_actions->data, odp_actions->size, &stats); | |
2457 | facet_update_stats(ofproto, facet, &stats); | |
2458 | } else { | |
2459 | facet_uninstall(ofproto, facet); | |
2460 | } | |
2461 | ||
2462 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
2463 | * 'facet' into 'rule'. */ | |
2464 | facet_flush_stats(ofproto, facet); | |
2465 | } | |
2466 | ||
2467 | /* Update 'facet' now that we've taken care of all the old state. */ | |
2468 | facet->tags = ctx.tags; | |
2469 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
2470 | facet->may_install = ctx.may_set_up_flow; | |
2471 | if (actions_changed) { | |
2472 | free(facet->actions); | |
2473 | facet->actions_len = odp_actions->size; | |
2474 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
2475 | } | |
2476 | if (facet->rule != new_rule) { | |
2477 | COVERAGE_INC(facet_changed_rule); | |
2478 | list_remove(&facet->list_node); | |
2479 | list_push_back(&new_rule->facets, &facet->list_node); | |
2480 | facet->rule = new_rule; | |
2481 | facet->used = new_rule->up.created; | |
2482 | facet->rs_used = facet->used; | |
2483 | } | |
2484 | ||
2485 | ofpbuf_delete(odp_actions); | |
2486 | ||
2487 | return true; | |
2488 | } | |
2489 | ||
2490 | /* Updates 'facet''s used time. Caller is responsible for calling | |
2491 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
2492 | static void | |
2493 | facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet, | |
2494 | long long int used) | |
2495 | { | |
2496 | if (used > facet->used) { | |
2497 | facet->used = used; | |
2498 | if (used > facet->rule->used) { | |
2499 | facet->rule->used = used; | |
2500 | } | |
2501 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
2502 | } | |
2503 | } | |
2504 | ||
2505 | /* Folds the statistics from 'stats' into the counters in 'facet'. | |
2506 | * | |
2507 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
2508 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
2509 | * packet that was sent by hand or if it represents statistics that have been | |
2510 | * cleared out of the datapath. */ | |
2511 | static void | |
2512 | facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet, | |
2513 | const struct dpif_flow_stats *stats) | |
2514 | { | |
2515 | if (stats->n_packets || stats->used > facet->used) { | |
2516 | facet_update_time(ofproto, facet, stats->used); | |
2517 | facet->packet_count += stats->n_packets; | |
2518 | facet->byte_count += stats->n_bytes; | |
2519 | facet_push_stats(facet); | |
2520 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
2521 | } | |
2522 | } | |
2523 | ||
2524 | static void | |
2525 | facet_push_stats(struct facet *facet) | |
2526 | { | |
2527 | uint64_t rs_packets, rs_bytes; | |
2528 | ||
2529 | assert(facet->packet_count >= facet->rs_packet_count); | |
2530 | assert(facet->byte_count >= facet->rs_byte_count); | |
2531 | assert(facet->used >= facet->rs_used); | |
2532 | ||
2533 | rs_packets = facet->packet_count - facet->rs_packet_count; | |
2534 | rs_bytes = facet->byte_count - facet->rs_byte_count; | |
2535 | ||
2536 | if (rs_packets || rs_bytes || facet->used > facet->rs_used) { | |
2537 | facet->rs_packet_count = facet->packet_count; | |
2538 | facet->rs_byte_count = facet->byte_count; | |
2539 | facet->rs_used = facet->used; | |
2540 | ||
2541 | flow_push_stats(facet->rule, &facet->flow, | |
2542 | rs_packets, rs_bytes, facet->used); | |
2543 | } | |
2544 | } | |
2545 | ||
2546 | struct ofproto_push { | |
2547 | struct action_xlate_ctx ctx; | |
2548 | uint64_t packets; | |
2549 | uint64_t bytes; | |
2550 | long long int used; | |
2551 | }; | |
2552 | ||
2553 | static void | |
2554 | push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
2555 | { | |
2556 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
2557 | ||
2558 | if (rule) { | |
2559 | rule->packet_count += push->packets; | |
2560 | rule->byte_count += push->bytes; | |
2561 | rule->used = MAX(push->used, rule->used); | |
2562 | } | |
2563 | } | |
2564 | ||
2565 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
2566 | * 'rule''s actions. */ | |
2567 | static void | |
2568 | flow_push_stats(const struct rule_dpif *rule, | |
2569 | struct flow *flow, uint64_t packets, uint64_t bytes, | |
2570 | long long int used) | |
2571 | { | |
2572 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2573 | struct ofproto_push push; | |
2574 | ||
2575 | push.packets = packets; | |
2576 | push.bytes = bytes; | |
2577 | push.used = used; | |
2578 | ||
2579 | action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); | |
2580 | push.ctx.resubmit_hook = push_resubmit; | |
2581 | ofpbuf_delete(xlate_actions(&push.ctx, | |
2582 | rule->up.actions, rule->up.n_actions)); | |
2583 | } | |
2584 | \f | |
2585 | /* Rules. */ | |
2586 | ||
2587 | static struct rule_dpif * | |
2588 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow) | |
2589 | { | |
154896e3 | 2590 | return rule_dpif_cast(rule_from_cls_rule( |
6c1491fb BP |
2591 | classifier_lookup(&ofproto->up.tables[0], |
2592 | flow))); | |
abe529af BP |
2593 | } |
2594 | ||
7ee20df1 BP |
2595 | static void |
2596 | complete_operation(struct rule_dpif *rule) | |
2597 | { | |
2598 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2599 | ||
2600 | ofproto->need_revalidate = true; | |
2601 | if (clogged) { | |
2602 | struct dpif_completion *c = xmalloc(sizeof *c); | |
2603 | c->op = rule->up.pending; | |
2604 | list_push_back(&ofproto->completions, &c->list_node); | |
2605 | } else { | |
2606 | ofoperation_complete(rule->up.pending, 0); | |
2607 | } | |
2608 | } | |
2609 | ||
abe529af BP |
2610 | static struct rule * |
2611 | rule_alloc(void) | |
2612 | { | |
2613 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
2614 | return &rule->up; | |
2615 | } | |
2616 | ||
2617 | static void | |
2618 | rule_dealloc(struct rule *rule_) | |
2619 | { | |
2620 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2621 | free(rule); | |
2622 | } | |
2623 | ||
2624 | static int | |
2625 | rule_construct(struct rule *rule_) | |
2626 | { | |
2627 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2628 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 2629 | struct rule_dpif *victim; |
5bf0e941 BP |
2630 | int error; |
2631 | ||
2632 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
2633 | &rule->up.cr.flow, ofproto->max_ports); | |
2634 | if (error) { | |
2635 | return error; | |
2636 | } | |
abe529af BP |
2637 | |
2638 | rule->used = rule->up.created; | |
2639 | rule->packet_count = 0; | |
2640 | rule->byte_count = 0; | |
abe529af | 2641 | |
7ee20df1 BP |
2642 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
2643 | if (victim && !list_is_empty(&victim->facets)) { | |
2644 | struct facet *facet; | |
2645 | ||
2646 | rule->facets = victim->facets; | |
2647 | list_moved(&rule->facets); | |
2648 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
2649 | facet->rule = rule; | |
2650 | } | |
2651 | } else { | |
2652 | /* Must avoid list_moved() in this case. */ | |
2653 | list_init(&rule->facets); | |
2654 | } | |
abe529af | 2655 | |
7ee20df1 | 2656 | complete_operation(rule); |
abe529af BP |
2657 | return 0; |
2658 | } | |
2659 | ||
2660 | static void | |
2661 | rule_destruct(struct rule *rule_) | |
2662 | { | |
2663 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2664 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2665 | struct facet *facet, *next_facet; | |
2666 | ||
abe529af BP |
2667 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
2668 | facet_revalidate(ofproto, facet); | |
2669 | } | |
7ee20df1 BP |
2670 | |
2671 | complete_operation(rule); | |
abe529af BP |
2672 | } |
2673 | ||
2674 | static void | |
2675 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
2676 | { | |
2677 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2678 | struct facet *facet; | |
2679 | ||
2680 | /* Start from historical data for 'rule' itself that are no longer tracked | |
2681 | * in facets. This counts, for example, facets that have expired. */ | |
2682 | *packets = rule->packet_count; | |
2683 | *bytes = rule->byte_count; | |
2684 | ||
2685 | /* Add any statistics that are tracked by facets. This includes | |
2686 | * statistical data recently updated by ofproto_update_stats() as well as | |
2687 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
2688 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
2689 | *packets += facet->packet_count; | |
2690 | *bytes += facet->byte_count; | |
2691 | } | |
2692 | } | |
2693 | ||
5bf0e941 | 2694 | static int |
abe529af BP |
2695 | rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet) |
2696 | { | |
2697 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2698 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2699 | struct action_xlate_ctx ctx; | |
2700 | struct ofpbuf *odp_actions; | |
2701 | struct facet *facet; | |
2702 | size_t size; | |
2703 | ||
2704 | /* First look for a related facet. If we find one, account it to that. */ | |
2705 | facet = facet_lookup_valid(ofproto, flow); | |
2706 | if (facet && facet->rule == rule) { | |
2707 | facet_execute(ofproto, facet, packet); | |
5bf0e941 | 2708 | return 0; |
abe529af BP |
2709 | } |
2710 | ||
2711 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then | |
2712 | * create a new facet for it and use that. */ | |
2713 | if (rule_dpif_lookup(ofproto, flow) == rule) { | |
2714 | facet = facet_create(rule, flow, packet); | |
2715 | facet_execute(ofproto, facet, packet); | |
2716 | facet_install(ofproto, facet, true); | |
5bf0e941 | 2717 | return 0; |
abe529af BP |
2718 | } |
2719 | ||
2720 | /* We can't account anything to a facet. If we were to try, then that | |
2721 | * facet would have a non-matching rule, busting our invariants. */ | |
2722 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
2723 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
2724 | size = packet->size; | |
2725 | if (execute_odp_actions(ofproto, flow, odp_actions->data, | |
2726 | odp_actions->size, packet)) { | |
2727 | rule->used = time_msec(); | |
2728 | rule->packet_count++; | |
2729 | rule->byte_count += size; | |
2730 | flow_push_stats(rule, flow, 1, size, rule->used); | |
2731 | } | |
2732 | ofpbuf_delete(odp_actions); | |
5bf0e941 BP |
2733 | |
2734 | return 0; | |
abe529af BP |
2735 | } |
2736 | ||
7ee20df1 BP |
2737 | static void |
2738 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
2739 | { |
2740 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2741 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2742 | int error; | |
2743 | ||
7ee20df1 BP |
2744 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
2745 | &rule->up.cr.flow, ofproto->max_ports); | |
2746 | if (error) { | |
2747 | ofoperation_complete(rule->up.pending, error); | |
2748 | return; | |
abe529af | 2749 | } |
7ee20df1 BP |
2750 | |
2751 | complete_operation(rule); | |
abe529af BP |
2752 | } |
2753 | \f | |
b2fda3ef | 2754 | /* Sends 'packet' out of port 'odp_port' within 'p'. |
abe529af BP |
2755 | * Returns 0 if successful, otherwise a positive errno value. */ |
2756 | static int | |
b2fda3ef | 2757 | send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, |
abe529af BP |
2758 | const struct ofpbuf *packet) |
2759 | { | |
80e5eed9 BP |
2760 | struct ofpbuf key, odp_actions; |
2761 | struct odputil_keybuf keybuf; | |
2762 | struct flow flow; | |
abe529af BP |
2763 | int error; |
2764 | ||
80e5eed9 BP |
2765 | flow_extract((struct ofpbuf *) packet, 0, 0, &flow); |
2766 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2767 | odp_flow_key_from_flow(&key, &flow); | |
2768 | ||
abe529af | 2769 | ofpbuf_init(&odp_actions, 32); |
abe529af | 2770 | nl_msg_put_u32(&odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
2771 | error = dpif_execute(ofproto->dpif, |
2772 | key.data, key.size, | |
2773 | odp_actions.data, odp_actions.size, | |
abe529af BP |
2774 | packet); |
2775 | ofpbuf_uninit(&odp_actions); | |
2776 | ||
2777 | if (error) { | |
2778 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
2779 | ofproto->up.name, odp_port, strerror(error)); | |
2780 | } | |
2781 | return error; | |
2782 | } | |
2783 | \f | |
2784 | /* OpenFlow to ODP action translation. */ | |
2785 | ||
2786 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
2787 | struct action_xlate_ctx *ctx); | |
4cd78906 | 2788 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 2789 | |
b3e9b2ed EJ |
2790 | static void |
2791 | commit_odp_actions(struct action_xlate_ctx *ctx) | |
2792 | { | |
2793 | const struct flow *flow = &ctx->flow; | |
2794 | struct flow *base = &ctx->base_flow; | |
2795 | struct ofpbuf *odp_actions = ctx->odp_actions; | |
2796 | ||
2797 | if (base->tun_id != flow->tun_id) { | |
2798 | nl_msg_put_be64(odp_actions, ODP_ACTION_ATTR_SET_TUNNEL, flow->tun_id); | |
2799 | base->tun_id = flow->tun_id; | |
2800 | } | |
2801 | ||
2802 | if (base->nw_src != flow->nw_src) { | |
2803 | nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_SRC, flow->nw_src); | |
2804 | base->nw_src = flow->nw_src; | |
2805 | } | |
2806 | ||
2807 | if (base->nw_dst != flow->nw_dst) { | |
2808 | nl_msg_put_be32(odp_actions, ODP_ACTION_ATTR_SET_NW_DST, flow->nw_dst); | |
2809 | base->nw_dst = flow->nw_dst; | |
2810 | } | |
2811 | ||
150a9f15 BP |
2812 | if (base->nw_tos != flow->nw_tos) { |
2813 | nl_msg_put_u8(odp_actions, ODP_ACTION_ATTR_SET_NW_TOS, flow->nw_tos); | |
2814 | base->nw_tos = flow->nw_tos; | |
2815 | } | |
2816 | ||
b3e9b2ed EJ |
2817 | if (base->vlan_tci != flow->vlan_tci) { |
2818 | if (!(flow->vlan_tci & htons(VLAN_CFI))) { | |
2819 | nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); | |
2820 | } else { | |
2821 | nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_DL_TCI, | |
2822 | flow->vlan_tci & ~htons(VLAN_CFI)); | |
2823 | } | |
2824 | base->vlan_tci = flow->vlan_tci; | |
2825 | } | |
2826 | ||
2827 | if (base->tp_src != flow->tp_src) { | |
2828 | nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_SRC, flow->tp_src); | |
2829 | base->tp_src = flow->tp_src; | |
2830 | } | |
2831 | ||
2832 | if (base->tp_dst != flow->tp_dst) { | |
2833 | nl_msg_put_be16(odp_actions, ODP_ACTION_ATTR_SET_TP_DST, flow->tp_dst); | |
2834 | base->tp_dst = flow->tp_dst; | |
2835 | } | |
2836 | ||
2837 | if (!eth_addr_equals(base->dl_src, flow->dl_src)) { | |
2838 | nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_SRC, | |
2839 | flow->dl_src, ETH_ADDR_LEN); | |
2840 | memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN); | |
2841 | } | |
2842 | ||
2843 | if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) { | |
2844 | nl_msg_put_unspec(odp_actions, ODP_ACTION_ATTR_SET_DL_DST, | |
2845 | flow->dl_dst, ETH_ADDR_LEN); | |
2846 | memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN); | |
2847 | } | |
2848 | ||
2849 | if (ctx->base_priority != ctx->priority) { | |
2850 | if (ctx->priority) { | |
2851 | nl_msg_put_u32(odp_actions, ODP_ACTION_ATTR_SET_PRIORITY, | |
2852 | ctx->priority); | |
2853 | } else { | |
2854 | nl_msg_put_flag(odp_actions, ODP_ACTION_ATTR_POP_PRIORITY); | |
2855 | } | |
2856 | ctx->base_priority = ctx->priority; | |
2857 | } | |
2858 | } | |
2859 | ||
abe529af BP |
2860 | static void |
2861 | add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) | |
2862 | { | |
2863 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); | |
2864 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); | |
2865 | ||
2866 | if (ofport) { | |
2867 | if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) { | |
2868 | /* Forwarding disabled on port. */ | |
2869 | return; | |
2870 | } | |
2871 | } else { | |
2872 | /* | |
2873 | * We don't have an ofport record for this port, but it doesn't hurt to | |
2874 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
2875 | * and we're pre-populating the flow table. | |
2876 | */ | |
2877 | } | |
2878 | ||
b3e9b2ed | 2879 | commit_odp_actions(ctx); |
abe529af BP |
2880 | nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, odp_port); |
2881 | ctx->nf_output_iface = ofp_port; | |
2882 | } | |
2883 | ||
2884 | static void | |
2885 | xlate_table_action(struct action_xlate_ctx *ctx, uint16_t in_port) | |
2886 | { | |
2887 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
2888 | struct rule_dpif *rule; | |
2889 | uint16_t old_in_port; | |
2890 | ||
2891 | /* Look up a flow with 'in_port' as the input port. Then restore the | |
2892 | * original input port (otherwise OFPP_NORMAL and OFPP_IN_PORT will | |
2893 | * have surprising behavior). */ | |
2894 | old_in_port = ctx->flow.in_port; | |
2895 | ctx->flow.in_port = in_port; | |
2896 | rule = rule_dpif_lookup(ctx->ofproto, &ctx->flow); | |
2897 | ctx->flow.in_port = old_in_port; | |
2898 | ||
2899 | if (ctx->resubmit_hook) { | |
2900 | ctx->resubmit_hook(ctx, rule); | |
2901 | } | |
2902 | ||
2903 | if (rule) { | |
2904 | ctx->recurse++; | |
2905 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); | |
2906 | ctx->recurse--; | |
2907 | } | |
2908 | } else { | |
2909 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
2910 | ||
2911 | VLOG_ERR_RL(&recurse_rl, "NXAST_RESUBMIT recursed over %d times", | |
2912 | MAX_RESUBMIT_RECURSION); | |
2913 | } | |
2914 | } | |
2915 | ||
2916 | static void | |
b3e9b2ed | 2917 | flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask) |
abe529af BP |
2918 | { |
2919 | struct ofport_dpif *ofport; | |
2920 | ||
b3e9b2ed EJ |
2921 | commit_odp_actions(ctx); |
2922 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { | |
abe529af | 2923 | uint16_t ofp_port = ofport->up.ofp_port; |
b3e9b2ed EJ |
2924 | if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) { |
2925 | nl_msg_put_u32(ctx->odp_actions, ODP_ACTION_ATTR_OUTPUT, | |
abe529af BP |
2926 | ofport->odp_port); |
2927 | } | |
2928 | } | |
b3e9b2ed EJ |
2929 | |
2930 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
2931 | } |
2932 | ||
2933 | static void | |
2934 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
2935 | uint16_t port, uint16_t max_len) | |
2936 | { | |
2937 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
2938 | ||
2939 | ctx->nf_output_iface = NF_OUT_DROP; | |
2940 | ||
2941 | switch (port) { | |
2942 | case OFPP_IN_PORT: | |
2943 | add_output_action(ctx, ctx->flow.in_port); | |
2944 | break; | |
2945 | case OFPP_TABLE: | |
2946 | xlate_table_action(ctx, ctx->flow.in_port); | |
2947 | break; | |
2948 | case OFPP_NORMAL: | |
2949 | xlate_normal(ctx); | |
2950 | break; | |
2951 | case OFPP_FLOOD: | |
b3e9b2ed | 2952 | flood_packets(ctx, htonl(OFPPC_NO_FLOOD)); |
abe529af BP |
2953 | break; |
2954 | case OFPP_ALL: | |
b3e9b2ed | 2955 | flood_packets(ctx, htonl(0)); |
abe529af BP |
2956 | break; |
2957 | case OFPP_CONTROLLER: | |
b3e9b2ed | 2958 | commit_odp_actions(ctx); |
b85d8d61 | 2959 | nl_msg_put_u64(ctx->odp_actions, ODP_ACTION_ATTR_USERSPACE, max_len); |
abe529af BP |
2960 | break; |
2961 | case OFPP_LOCAL: | |
2962 | add_output_action(ctx, OFPP_LOCAL); | |
2963 | break; | |
e81d2933 EJ |
2964 | case OFPP_NONE: |
2965 | break; | |
abe529af BP |
2966 | default: |
2967 | if (port != ctx->flow.in_port) { | |
2968 | add_output_action(ctx, port); | |
2969 | } | |
2970 | break; | |
2971 | } | |
2972 | ||
2973 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
2974 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
2975 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
2976 | ctx->nf_output_iface = prev_nf_output_iface; | |
2977 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
2978 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
2979 | ctx->nf_output_iface = NF_OUT_MULTI; | |
2980 | } | |
2981 | } | |
2982 | ||
2983 | static void | |
2984 | xlate_output_action(struct action_xlate_ctx *ctx, | |
2985 | const struct ofp_action_output *oao) | |
2986 | { | |
2987 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
2988 | } | |
2989 | ||
abe529af BP |
2990 | static void |
2991 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
2992 | const struct ofp_action_enqueue *oae) | |
2993 | { | |
2994 | uint16_t ofp_port, odp_port; | |
b3e9b2ed | 2995 | uint32_t ctx_priority, priority; |
abe529af BP |
2996 | int error; |
2997 | ||
2998 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
2999 | &priority); | |
3000 | if (error) { | |
3001 | /* Fall back to ordinary output action. */ | |
3002 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
3003 | return; | |
3004 | } | |
3005 | ||
3006 | /* Figure out ODP output port. */ | |
3007 | ofp_port = ntohs(oae->port); | |
3008 | if (ofp_port == OFPP_IN_PORT) { | |
3009 | ofp_port = ctx->flow.in_port; | |
3010 | } | |
3011 | odp_port = ofp_port_to_odp_port(ofp_port); | |
3012 | ||
3013 | /* Add ODP actions. */ | |
b3e9b2ed EJ |
3014 | ctx_priority = ctx->priority; |
3015 | ctx->priority = priority; | |
abe529af | 3016 | add_output_action(ctx, odp_port); |
b3e9b2ed | 3017 | ctx->priority = ctx_priority; |
abe529af BP |
3018 | |
3019 | /* Update NetFlow output port. */ | |
3020 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
3021 | ctx->nf_output_iface = odp_port; | |
3022 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
3023 | ctx->nf_output_iface = NF_OUT_MULTI; | |
3024 | } | |
3025 | } | |
3026 | ||
3027 | static void | |
3028 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
3029 | const struct nx_action_set_queue *nasq) | |
3030 | { | |
3031 | uint32_t priority; | |
3032 | int error; | |
3033 | ||
3034 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
3035 | &priority); | |
3036 | if (error) { | |
3037 | /* Couldn't translate queue to a priority, so ignore. A warning | |
3038 | * has already been logged. */ | |
3039 | return; | |
3040 | } | |
3041 | ||
afabef2b | 3042 | ctx->priority = priority; |
abe529af BP |
3043 | } |
3044 | ||
3045 | struct xlate_reg_state { | |
3046 | ovs_be16 vlan_tci; | |
3047 | ovs_be64 tun_id; | |
3048 | }; | |
3049 | ||
abe529af BP |
3050 | static void |
3051 | xlate_autopath(struct action_xlate_ctx *ctx, | |
3052 | const struct nx_action_autopath *naa) | |
3053 | { | |
3054 | uint16_t ofp_port = ntohl(naa->id); | |
3055 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
3056 | ||
3057 | if (!port || !port->bundle) { | |
3058 | ofp_port = OFPP_NONE; | |
3059 | } else if (port->bundle->bond) { | |
3060 | /* Autopath does not support VLAN hashing. */ | |
3061 | struct ofport_dpif *slave = bond_choose_output_slave( | |
3062 | port->bundle->bond, &ctx->flow, OFP_VLAN_NONE, &ctx->tags); | |
3063 | if (slave) { | |
3064 | ofp_port = slave->up.ofp_port; | |
3065 | } | |
3066 | } | |
3067 | autopath_execute(naa, &ctx->flow, ofp_port); | |
3068 | } | |
3069 | ||
daff3353 EJ |
3070 | static bool |
3071 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
3072 | { | |
3073 | struct ofproto_dpif *ofproto = ofproto_; | |
3074 | struct ofport_dpif *port; | |
3075 | ||
3076 | switch (ofp_port) { | |
3077 | case OFPP_IN_PORT: | |
3078 | case OFPP_TABLE: | |
3079 | case OFPP_NORMAL: | |
3080 | case OFPP_FLOOD: | |
3081 | case OFPP_ALL: | |
3082 | case OFPP_LOCAL: | |
3083 | return true; | |
3084 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
3085 | return false; | |
3086 | default: | |
3087 | port = get_ofp_port(ofproto, ofp_port); | |
3088 | return port ? port->may_enable : false; | |
3089 | } | |
3090 | } | |
3091 | ||
abe529af BP |
3092 | static void |
3093 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
3094 | struct action_xlate_ctx *ctx) | |
3095 | { | |
3096 | const struct ofport_dpif *port; | |
abe529af | 3097 | const union ofp_action *ia; |
b4b8c781 | 3098 | size_t left; |
abe529af BP |
3099 | |
3100 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
3101 | if (port | |
3102 | && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && | |
3103 | port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
3104 | ? htonl(OFPPC_NO_RECV_STP) | |
3105 | : htonl(OFPPC_NO_RECV))) { | |
3106 | /* Drop this flow. */ | |
3107 | return; | |
3108 | } | |
3109 | ||
b4b8c781 | 3110 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 3111 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
3112 | const struct nx_action_resubmit *nar; |
3113 | const struct nx_action_set_tunnel *nast; | |
3114 | const struct nx_action_set_queue *nasq; | |
3115 | const struct nx_action_multipath *nam; | |
3116 | const struct nx_action_autopath *naa; | |
daff3353 | 3117 | const struct nx_action_bundle *nab; |
38f2e360 BP |
3118 | enum ofputil_action_code code; |
3119 | ovs_be64 tun_id; | |
3120 | ||
3121 | code = ofputil_decode_action_unsafe(ia); | |
3122 | switch (code) { | |
3123 | case OFPUTIL_OFPAT_OUTPUT: | |
abe529af BP |
3124 | xlate_output_action(ctx, &ia->output); |
3125 | break; | |
3126 | ||
38f2e360 | 3127 | case OFPUTIL_OFPAT_SET_VLAN_VID: |
abe529af BP |
3128 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
3129 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
3130 | break; |
3131 | ||
38f2e360 | 3132 | case OFPUTIL_OFPAT_SET_VLAN_PCP: |
abe529af BP |
3133 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
3134 | ctx->flow.vlan_tci |= htons( | |
3135 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
3136 | break; |
3137 | ||
38f2e360 | 3138 | case OFPUTIL_OFPAT_STRIP_VLAN: |
abe529af | 3139 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
3140 | break; |
3141 | ||
38f2e360 | 3142 | case OFPUTIL_OFPAT_SET_DL_SRC: |
abe529af | 3143 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
3144 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
3145 | break; | |
3146 | ||
38f2e360 | 3147 | case OFPUTIL_OFPAT_SET_DL_DST: |
abe529af | 3148 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
3149 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
3150 | break; | |
3151 | ||
38f2e360 | 3152 | case OFPUTIL_OFPAT_SET_NW_SRC: |
abe529af BP |
3153 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
3154 | break; | |
3155 | ||
38f2e360 | 3156 | case OFPUTIL_OFPAT_SET_NW_DST: |
abe529af BP |
3157 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
3158 | break; | |
3159 | ||
38f2e360 | 3160 | case OFPUTIL_OFPAT_SET_NW_TOS: |
1fa96cf4 | 3161 | ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK; |
abe529af BP |
3162 | break; |
3163 | ||
38f2e360 | 3164 | case OFPUTIL_OFPAT_SET_TP_SRC: |
abe529af BP |
3165 | ctx->flow.tp_src = ia->tp_port.tp_port; |
3166 | break; | |
3167 | ||
38f2e360 | 3168 | case OFPUTIL_OFPAT_SET_TP_DST: |
abe529af BP |
3169 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
3170 | break; | |
3171 | ||
38f2e360 BP |
3172 | case OFPUTIL_OFPAT_ENQUEUE: |
3173 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
3174 | break; | |
3175 | ||
3176 | case OFPUTIL_NXAST_RESUBMIT: | |
3177 | nar = (const struct nx_action_resubmit *) ia; | |
3178 | xlate_table_action(ctx, ntohs(nar->in_port)); | |
abe529af BP |
3179 | break; |
3180 | ||
38f2e360 BP |
3181 | case OFPUTIL_NXAST_SET_TUNNEL: |
3182 | nast = (const struct nx_action_set_tunnel *) ia; | |
3183 | tun_id = htonll(ntohl(nast->tun_id)); | |
3184 | ctx->flow.tun_id = tun_id; | |
3185 | break; | |
3186 | ||
3187 | case OFPUTIL_NXAST_SET_QUEUE: | |
3188 | nasq = (const struct nx_action_set_queue *) ia; | |
3189 | xlate_set_queue_action(ctx, nasq); | |
3190 | break; | |
3191 | ||
3192 | case OFPUTIL_NXAST_POP_QUEUE: | |
3193 | ctx->priority = 0; | |
3194 | break; | |
3195 | ||
3196 | case OFPUTIL_NXAST_REG_MOVE: | |
3197 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
3198 | &ctx->flow); | |
3199 | break; | |
3200 | ||
3201 | case OFPUTIL_NXAST_REG_LOAD: | |
3202 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
3203 | &ctx->flow); | |
3204 | break; | |
3205 | ||
3206 | case OFPUTIL_NXAST_NOTE: | |
3207 | /* Nothing to do. */ | |
3208 | break; | |
3209 | ||
3210 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
3211 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
3212 | ctx->flow.tun_id = tun_id; | |
3213 | break; | |
3214 | ||
3215 | case OFPUTIL_NXAST_MULTIPATH: | |
3216 | nam = (const struct nx_action_multipath *) ia; | |
3217 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
3218 | break; |
3219 | ||
38f2e360 BP |
3220 | case OFPUTIL_NXAST_AUTOPATH: |
3221 | naa = (const struct nx_action_autopath *) ia; | |
3222 | xlate_autopath(ctx, naa); | |
abe529af | 3223 | break; |
daff3353 EJ |
3224 | |
3225 | case OFPUTIL_NXAST_BUNDLE: | |
3226 | ctx->ofproto->has_bundle_action = true; | |
3227 | nab = (const struct nx_action_bundle *) ia; | |
3228 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
3229 | slave_enabled_cb, | |
3230 | ctx->ofproto), 0); | |
3231 | break; | |
a368bb53 EJ |
3232 | |
3233 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
3234 | ctx->ofproto->has_bundle_action = true; | |
3235 | nab = (const struct nx_action_bundle *) ia; | |
3236 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
3237 | ctx->ofproto); | |
3238 | break; | |
abe529af BP |
3239 | } |
3240 | } | |
3241 | } | |
3242 | ||
3243 | static void | |
3244 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
3245 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
3246 | const struct ofpbuf *packet) | |
3247 | { | |
3248 | ctx->ofproto = ofproto; | |
3249 | ctx->flow = *flow; | |
3250 | ctx->packet = packet; | |
3251 | ctx->resubmit_hook = NULL; | |
abe529af BP |
3252 | } |
3253 | ||
3254 | static struct ofpbuf * | |
3255 | xlate_actions(struct action_xlate_ctx *ctx, | |
3256 | const union ofp_action *in, size_t n_in) | |
3257 | { | |
3258 | COVERAGE_INC(ofproto_dpif_xlate); | |
3259 | ||
3260 | ctx->odp_actions = ofpbuf_new(512); | |
3261 | ctx->tags = 0; | |
3262 | ctx->may_set_up_flow = true; | |
3263 | ctx->nf_output_iface = NF_OUT_DROP; | |
3264 | ctx->recurse = 0; | |
afabef2b | 3265 | ctx->priority = 0; |
b3e9b2ed EJ |
3266 | ctx->base_priority = 0; |
3267 | ctx->base_flow = ctx->flow; | |
abe529af | 3268 | |
fc08b7a2 | 3269 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af BP |
3270 | ctx->may_set_up_flow = false; |
3271 | } else { | |
3272 | do_xlate_actions(in, n_in, ctx); | |
3273 | } | |
3274 | ||
abe529af BP |
3275 | /* Check with in-band control to see if we're allowed to set up this |
3276 | * flow. */ | |
3277 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, | |
3278 | ctx->odp_actions->data, | |
3279 | ctx->odp_actions->size)) { | |
3280 | ctx->may_set_up_flow = false; | |
3281 | } | |
3282 | ||
3283 | return ctx->odp_actions; | |
3284 | } | |
3285 | \f | |
3286 | /* OFPP_NORMAL implementation. */ | |
3287 | ||
3288 | struct dst { | |
3289 | struct ofport_dpif *port; | |
3290 | uint16_t vlan; | |
3291 | }; | |
3292 | ||
3293 | struct dst_set { | |
3294 | struct dst builtin[32]; | |
3295 | struct dst *dsts; | |
3296 | size_t n, allocated; | |
3297 | }; | |
3298 | ||
3299 | static void dst_set_init(struct dst_set *); | |
3300 | static void dst_set_add(struct dst_set *, const struct dst *); | |
3301 | static void dst_set_free(struct dst_set *); | |
3302 | ||
3303 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); | |
3304 | ||
3305 | static bool | |
3306 | set_dst(struct action_xlate_ctx *ctx, struct dst *dst, | |
3307 | const struct ofbundle *in_bundle, const struct ofbundle *out_bundle) | |
3308 | { | |
3309 | dst->vlan = (out_bundle->vlan >= 0 ? OFP_VLAN_NONE | |
3310 | : in_bundle->vlan >= 0 ? in_bundle->vlan | |
3311 | : ctx->flow.vlan_tci == 0 ? OFP_VLAN_NONE | |
3312 | : vlan_tci_to_vid(ctx->flow.vlan_tci)); | |
3313 | ||
3314 | dst->port = (!out_bundle->bond | |
3315 | ? ofbundle_get_a_port(out_bundle) | |
3316 | : bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
3317 | dst->vlan, &ctx->tags)); | |
3318 | ||
3319 | return dst->port != NULL; | |
3320 | } | |
3321 | ||
3322 | static int | |
3323 | mirror_mask_ffs(mirror_mask_t mask) | |
3324 | { | |
3325 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
3326 | return ffs(mask); | |
3327 | } | |
3328 | ||
3329 | static void | |
3330 | dst_set_init(struct dst_set *set) | |
3331 | { | |
3332 | set->dsts = set->builtin; | |
3333 | set->n = 0; | |
3334 | set->allocated = ARRAY_SIZE(set->builtin); | |
3335 | } | |
3336 | ||
3337 | static void | |
3338 | dst_set_add(struct dst_set *set, const struct dst *dst) | |
3339 | { | |
3340 | if (set->n >= set->allocated) { | |
3341 | size_t new_allocated; | |
3342 | struct dst *new_dsts; | |
3343 | ||
3344 | new_allocated = set->allocated * 2; | |
3345 | new_dsts = xmalloc(new_allocated * sizeof *new_dsts); | |
3346 | memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); | |
3347 | ||
3348 | dst_set_free(set); | |
3349 | ||
3350 | set->dsts = new_dsts; | |
3351 | set->allocated = new_allocated; | |
3352 | } | |
3353 | set->dsts[set->n++] = *dst; | |
3354 | } | |
3355 | ||
3356 | static void | |
3357 | dst_set_free(struct dst_set *set) | |
3358 | { | |
3359 | if (set->dsts != set->builtin) { | |
3360 | free(set->dsts); | |
3361 | } | |
3362 | } | |
3363 | ||
3364 | static bool | |
3365 | dst_is_duplicate(const struct dst_set *set, const struct dst *test) | |
3366 | { | |
3367 | size_t i; | |
3368 | for (i = 0; i < set->n; i++) { | |
3369 | if (set->dsts[i].vlan == test->vlan | |
3370 | && set->dsts[i].port == test->port) { | |
3371 | return true; | |
3372 | } | |
3373 | } | |
3374 | return false; | |
3375 | } | |
3376 | ||
3377 | static bool | |
3378 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
3379 | { | |
fc3d7408 BP |
3380 | return (bundle->vlan < 0 |
3381 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); | |
abe529af BP |
3382 | } |
3383 | ||
3384 | static bool | |
3385 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
3386 | { | |
3387 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
3388 | } | |
3389 | ||
3390 | /* Returns an arbitrary interface within 'bundle'. */ | |
3391 | static struct ofport_dpif * | |
3392 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
3393 | { | |
3394 | return CONTAINER_OF(list_front(&bundle->ports), | |
3395 | struct ofport_dpif, bundle_node); | |
3396 | } | |
3397 | ||
3398 | static void | |
3399 | compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan, | |
3400 | const struct ofbundle *in_bundle, | |
3401 | const struct ofbundle *out_bundle, struct dst_set *set) | |
3402 | { | |
3403 | struct dst dst; | |
3404 | ||
3405 | if (out_bundle == OFBUNDLE_FLOOD) { | |
3406 | struct ofbundle *bundle; | |
3407 | ||
3408 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { | |
3409 | if (bundle != in_bundle | |
3410 | && ofbundle_includes_vlan(bundle, vlan) | |
3411 | && bundle->floodable | |
3412 | && !bundle->mirror_out | |
3413 | && set_dst(ctx, &dst, in_bundle, bundle)) { | |
3414 | dst_set_add(set, &dst); | |
3415 | } | |
3416 | } | |
3417 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
3418 | } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) { | |
3419 | dst_set_add(set, &dst); | |
3420 | ctx->nf_output_iface = dst.port->odp_port; | |
3421 | } | |
3422 | } | |
3423 | ||
3424 | static bool | |
3425 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
3426 | { | |
fc3d7408 | 3427 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
3428 | } |
3429 | ||
07817dfe BP |
3430 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
3431 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
3432 | * protocols that may confuse switches. */ | |
3433 | static bool | |
3434 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
3435 | { | |
3436 | /* If you change this function's behavior, please update corresponding | |
3437 | * documentation in vswitch.xml at the same time. */ | |
3438 | if (dst[0] != 0x01) { | |
3439 | /* All the currently banned MACs happen to start with 01 currently, so | |
3440 | * this is a quick way to eliminate most of the good ones. */ | |
3441 | } else { | |
3442 | if (eth_addr_is_reserved(dst)) { | |
3443 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
3444 | * (01-80-c2-00-00-0x). */ | |
3445 | return false; | |
3446 | } | |
3447 | ||
3448 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
3449 | /* Cisco OUI. */ | |
3450 | if ((dst[3] & 0xfe) == 0xcc && | |
3451 | (dst[4] & 0xfe) == 0xcc && | |
3452 | (dst[5] & 0xfe) == 0xcc) { | |
3453 | /* Drop the following protocols plus others following the same | |
3454 | pattern: | |
3455 | ||
3456 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
3457 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
3458 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
3459 | return false; | |
3460 | } | |
3461 | ||
3462 | if (!(dst[3] | dst[4] | dst[5])) { | |
3463 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
3464 | return false; | |
3465 | } | |
3466 | } | |
3467 | } | |
3468 | return true; | |
3469 | } | |
3470 | ||
abe529af BP |
3471 | static void |
3472 | compose_mirror_dsts(struct action_xlate_ctx *ctx, | |
3473 | uint16_t vlan, const struct ofbundle *in_bundle, | |
3474 | struct dst_set *set) | |
3475 | { | |
3476 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
3477 | mirror_mask_t mirrors; | |
3478 | int flow_vlan; | |
3479 | size_t i; | |
3480 | ||
3481 | mirrors = in_bundle->src_mirrors; | |
3482 | for (i = 0; i < set->n; i++) { | |
3483 | mirrors |= set->dsts[i].port->bundle->dst_mirrors; | |
3484 | } | |
3485 | ||
3486 | if (!mirrors) { | |
3487 | return; | |
3488 | } | |
3489 | ||
3490 | flow_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
3491 | if (flow_vlan == 0) { | |
3492 | flow_vlan = OFP_VLAN_NONE; | |
3493 | } | |
3494 | ||
3495 | while (mirrors) { | |
3496 | struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
3497 | if (vlan_is_mirrored(m, vlan)) { | |
3498 | struct dst dst; | |
3499 | ||
3500 | if (m->out) { | |
3501 | if (set_dst(ctx, &dst, in_bundle, m->out) | |
3502 | && !dst_is_duplicate(set, &dst)) { | |
3503 | dst_set_add(set, &dst); | |
3504 | } | |
07817dfe | 3505 | } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) { |
abe529af BP |
3506 | struct ofbundle *bundle; |
3507 | ||
3508 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
3509 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
3510 | && set_dst(ctx, &dst, in_bundle, bundle)) | |
3511 | { | |
3512 | if (bundle->vlan < 0) { | |
3513 | dst.vlan = m->out_vlan; | |
3514 | } | |
3515 | if (dst_is_duplicate(set, &dst)) { | |
3516 | continue; | |
3517 | } | |
3518 | ||
3519 | /* Use the vlan tag on the original flow instead of | |
3520 | * the one passed in the vlan parameter. This ensures | |
3521 | * that we compare the vlan from before any implicit | |
3522 | * tagging tags place. This is necessary because | |
3523 | * dst->vlan is the final vlan, after removing implicit | |
3524 | * tags. */ | |
3525 | if (bundle == in_bundle && dst.vlan == flow_vlan) { | |
3526 | /* Don't send out input port on same VLAN. */ | |
3527 | continue; | |
3528 | } | |
3529 | dst_set_add(set, &dst); | |
3530 | } | |
3531 | } | |
3532 | } | |
3533 | } | |
3534 | mirrors &= mirrors - 1; | |
3535 | } | |
3536 | } | |
3537 | ||
3538 | static void | |
3539 | compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan, | |
3540 | const struct ofbundle *in_bundle, | |
3541 | const struct ofbundle *out_bundle) | |
3542 | { | |
3543 | uint16_t initial_vlan, cur_vlan; | |
3544 | const struct dst *dst; | |
3545 | struct dst_set set; | |
3546 | ||
3547 | dst_set_init(&set); | |
3548 | compose_dsts(ctx, vlan, in_bundle, out_bundle, &set); | |
3549 | compose_mirror_dsts(ctx, vlan, in_bundle, &set); | |
3550 | ||
3551 | /* Output all the packets we can without having to change the VLAN. */ | |
3552 | initial_vlan = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
3553 | if (initial_vlan == 0) { | |
3554 | initial_vlan = OFP_VLAN_NONE; | |
3555 | } | |
3556 | for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { | |
3557 | if (dst->vlan != initial_vlan) { | |
3558 | continue; | |
3559 | } | |
3560 | nl_msg_put_u32(ctx->odp_actions, | |
3561 | ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); | |
3562 | } | |
3563 | ||
3564 | /* Then output the rest. */ | |
3565 | cur_vlan = initial_vlan; | |
3566 | for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { | |
3567 | if (dst->vlan == initial_vlan) { | |
3568 | continue; | |
3569 | } | |
3570 | if (dst->vlan != cur_vlan) { | |
3571 | if (dst->vlan == OFP_VLAN_NONE) { | |
3572 | nl_msg_put_flag(ctx->odp_actions, ODP_ACTION_ATTR_STRIP_VLAN); | |
3573 | } else { | |
3574 | ovs_be16 tci; | |
3575 | tci = htons(dst->vlan & VLAN_VID_MASK); | |
3576 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
3577 | nl_msg_put_be16(ctx->odp_actions, | |
3578 | ODP_ACTION_ATTR_SET_DL_TCI, tci); | |
3579 | } | |
3580 | cur_vlan = dst->vlan; | |
3581 | } | |
3582 | nl_msg_put_u32(ctx->odp_actions, | |
3583 | ODP_ACTION_ATTR_OUTPUT, dst->port->odp_port); | |
3584 | } | |
3585 | ||
3586 | dst_set_free(&set); | |
3587 | } | |
3588 | ||
3589 | /* Returns the effective vlan of a packet, taking into account both the | |
3590 | * 802.1Q header and implicitly tagged ports. A value of 0 indicates that | |
3591 | * the packet is untagged and -1 indicates it has an invalid header and | |
3592 | * should be dropped. */ | |
3593 | static int | |
3594 | flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3595 | struct ofbundle *in_bundle, bool have_packet) | |
3596 | { | |
3597 | int vlan = vlan_tci_to_vid(flow->vlan_tci); | |
3598 | if (in_bundle->vlan >= 0) { | |
3599 | if (vlan) { | |
3600 | if (have_packet) { | |
3601 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3602 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
3603 | "packet received on port %s configured with " | |
3604 | "implicit VLAN %"PRIu16, | |
3605 | ofproto->up.name, vlan, | |
3606 | in_bundle->name, in_bundle->vlan); | |
3607 | } | |
3608 | return -1; | |
3609 | } | |
3610 | vlan = in_bundle->vlan; | |
3611 | } else { | |
3612 | if (!ofbundle_includes_vlan(in_bundle, vlan)) { | |
3613 | if (have_packet) { | |
3614 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3615 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
3616 | "packet received on port %s not configured for " | |
3617 | "trunking VLAN %d", | |
3618 | ofproto->up.name, vlan, in_bundle->name, vlan); | |
3619 | } | |
3620 | return -1; | |
3621 | } | |
3622 | } | |
3623 | ||
3624 | return vlan; | |
3625 | } | |
3626 | ||
3627 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after | |
3628 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
3629 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
3630 | static bool | |
3631 | is_gratuitous_arp(const struct flow *flow) | |
3632 | { | |
3633 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
3634 | && eth_addr_is_broadcast(flow->dl_dst) | |
3635 | && (flow->nw_proto == ARP_OP_REPLY | |
3636 | || (flow->nw_proto == ARP_OP_REQUEST | |
3637 | && flow->nw_src == flow->nw_dst))); | |
3638 | } | |
3639 | ||
3640 | static void | |
3641 | update_learning_table(struct ofproto_dpif *ofproto, | |
3642 | const struct flow *flow, int vlan, | |
3643 | struct ofbundle *in_bundle) | |
3644 | { | |
3645 | struct mac_entry *mac; | |
3646 | ||
3647 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { | |
3648 | return; | |
3649 | } | |
3650 | ||
3651 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
3652 | if (is_gratuitous_arp(flow)) { | |
3653 | /* We don't want to learn from gratuitous ARP packets that are | |
3654 | * reflected back over bond slaves so we lock the learning table. */ | |
3655 | if (!in_bundle->bond) { | |
3656 | mac_entry_set_grat_arp_lock(mac); | |
3657 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
3658 | return; | |
3659 | } | |
3660 | } | |
3661 | ||
3662 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
3663 | /* The log messages here could actually be useful in debugging, | |
3664 | * so keep the rate limit relatively high. */ | |
3665 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
3666 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
3667 | "on port %s in VLAN %d", | |
3668 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
3669 | in_bundle->name, vlan); | |
3670 | ||
3671 | mac->port.p = in_bundle; | |
3672 | tag_set_add(&ofproto->revalidate_set, | |
3673 | mac_learning_changed(ofproto->ml, mac)); | |
3674 | } | |
3675 | } | |
3676 | ||
3677 | /* Determines whether packets in 'flow' within 'br' should be forwarded or | |
3678 | * dropped. Returns true if they may be forwarded, false if they should be | |
3679 | * dropped. | |
3680 | * | |
3681 | * If 'have_packet' is true, it indicates that the caller is processing a | |
3682 | * received packet. If 'have_packet' is false, then the caller is just | |
3683 | * revalidating an existing flow because configuration has changed. Either | |
3684 | * way, 'have_packet' only affects logging (there is no point in logging errors | |
3685 | * during revalidation). | |
3686 | * | |
3687 | * Sets '*in_portp' to the input port. This will be a null pointer if | |
3688 | * flow->in_port does not designate a known input port (in which case | |
3689 | * is_admissible() returns false). | |
3690 | * | |
3691 | * When returning true, sets '*vlanp' to the effective VLAN of the input | |
3692 | * packet, as returned by flow_get_vlan(). | |
3693 | * | |
3694 | * May also add tags to '*tags', although the current implementation only does | |
3695 | * so in one special case. | |
3696 | */ | |
3697 | static bool | |
3698 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
3699 | bool have_packet, | |
3700 | tag_type *tags, int *vlanp, struct ofbundle **in_bundlep) | |
3701 | { | |
3702 | struct ofport_dpif *in_port; | |
3703 | struct ofbundle *in_bundle; | |
3704 | int vlan; | |
3705 | ||
3706 | /* Find the port and bundle for the received packet. */ | |
3707 | in_port = get_ofp_port(ofproto, flow->in_port); | |
23adee42 | 3708 | *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL; |
abe529af BP |
3709 | if (!in_port || !in_bundle) { |
3710 | /* No interface? Something fishy... */ | |
3711 | if (have_packet) { | |
3712 | /* Odd. A few possible reasons here: | |
3713 | * | |
3714 | * - We deleted a port but there are still a few packets queued up | |
3715 | * from it. | |
3716 | * | |
3717 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
3718 | * we don't know about. | |
3719 | * | |
3720 | * - Packet arrived on the local port but the local port is not | |
3721 | * part of a bundle. | |
3722 | */ | |
3723 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3724 | ||
3725 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
3726 | "port %"PRIu16, | |
3727 | ofproto->up.name, flow->in_port); | |
3728 | } | |
3729 | return false; | |
3730 | } | |
3731 | *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet); | |
3732 | if (vlan < 0) { | |
3733 | return false; | |
3734 | } | |
3735 | ||
3736 | /* Drop frames for reserved multicast addresses. */ | |
3737 | if (eth_addr_is_reserved(flow->dl_dst)) { | |
3738 | return false; | |
3739 | } | |
3740 | ||
3741 | /* Drop frames on bundles reserved for mirroring. */ | |
3742 | if (in_bundle->mirror_out) { | |
3743 | if (have_packet) { | |
3744 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
3745 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
3746 | "%s, which is reserved exclusively for mirroring", | |
3747 | ofproto->up.name, in_bundle->name); | |
3748 | } | |
3749 | return false; | |
3750 | } | |
3751 | ||
3752 | if (in_bundle->bond) { | |
3753 | struct mac_entry *mac; | |
3754 | ||
3755 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
3756 | flow->dl_dst, tags)) { | |
3757 | case BV_ACCEPT: | |
3758 | break; | |
3759 | ||
3760 | case BV_DROP: | |
3761 | return false; | |
3762 | ||
3763 | case BV_DROP_IF_MOVED: | |
3764 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
3765 | if (mac && mac->port.p != in_bundle && | |
3766 | (!is_gratuitous_arp(flow) | |
3767 | || mac_entry_is_grat_arp_locked(mac))) { | |
3768 | return false; | |
3769 | } | |
3770 | break; | |
3771 | } | |
3772 | } | |
3773 | ||
3774 | return true; | |
3775 | } | |
3776 | ||
4cd78906 | 3777 | static void |
abe529af BP |
3778 | xlate_normal(struct action_xlate_ctx *ctx) |
3779 | { | |
3780 | struct ofbundle *in_bundle; | |
3781 | struct ofbundle *out_bundle; | |
3782 | struct mac_entry *mac; | |
3783 | int vlan; | |
3784 | ||
3785 | /* Check whether we should drop packets in this flow. */ | |
3786 | if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL, | |
3787 | &ctx->tags, &vlan, &in_bundle)) { | |
3788 | out_bundle = NULL; | |
3789 | goto done; | |
3790 | } | |
3791 | ||
3792 | /* Learn source MAC (but don't try to learn from revalidation). */ | |
3793 | if (ctx->packet) { | |
3794 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); | |
3795 | } | |
3796 | ||
3797 | /* Determine output bundle. */ | |
3798 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
3799 | &ctx->tags); | |
3800 | if (mac) { | |
3801 | out_bundle = mac->port.p; | |
3802 | } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) { | |
3803 | /* If we are revalidating but don't have a learning entry then eject | |
3804 | * the flow. Installing a flow that floods packets opens up a window | |
3805 | * of time where we could learn from a packet reflected on a bond and | |
3806 | * blackhole packets before the learning table is updated to reflect | |
3807 | * the correct port. */ | |
4cd78906 BP |
3808 | ctx->may_set_up_flow = false; |
3809 | return; | |
abe529af BP |
3810 | } else { |
3811 | out_bundle = OFBUNDLE_FLOOD; | |
3812 | } | |
3813 | ||
3814 | /* Don't send packets out their input bundles. */ | |
3815 | if (in_bundle == out_bundle) { | |
3816 | out_bundle = NULL; | |
3817 | } | |
3818 | ||
3819 | done: | |
3820 | if (in_bundle) { | |
3821 | compose_actions(ctx, vlan, in_bundle, out_bundle); | |
3822 | } | |
abe529af BP |
3823 | } |
3824 | \f | |
3825 | static bool | |
3826 | get_drop_frags(struct ofproto *ofproto_) | |
3827 | { | |
3828 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3829 | bool drop_frags; | |
3830 | ||
3831 | dpif_get_drop_frags(ofproto->dpif, &drop_frags); | |
3832 | return drop_frags; | |
3833 | } | |
3834 | ||
3835 | static void | |
3836 | set_drop_frags(struct ofproto *ofproto_, bool drop_frags) | |
3837 | { | |
3838 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3839 | ||
3840 | dpif_set_drop_frags(ofproto->dpif, drop_frags); | |
3841 | } | |
3842 | ||
3843 | static int | |
3844 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, | |
3845 | const struct flow *flow, | |
3846 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
3847 | { | |
3848 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3849 | int error; | |
3850 | ||
3851 | error = validate_actions(ofp_actions, n_ofp_actions, flow, | |
3852 | ofproto->max_ports); | |
3853 | if (!error) { | |
80e5eed9 | 3854 | struct odputil_keybuf keybuf; |
abe529af BP |
3855 | struct action_xlate_ctx ctx; |
3856 | struct ofpbuf *odp_actions; | |
80e5eed9 BP |
3857 | struct ofpbuf key; |
3858 | ||
3859 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
3860 | odp_flow_key_from_flow(&key, flow); | |
abe529af BP |
3861 | |
3862 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
3863 | odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); | |
80e5eed9 BP |
3864 | dpif_execute(ofproto->dpif, key.data, key.size, |
3865 | odp_actions->data, odp_actions->size, packet); | |
abe529af BP |
3866 | ofpbuf_delete(odp_actions); |
3867 | } | |
3868 | return error; | |
3869 | } | |
3870 | ||
3871 | static void | |
3872 | get_netflow_ids(const struct ofproto *ofproto_, | |
3873 | uint8_t *engine_type, uint8_t *engine_id) | |
3874 | { | |
3875 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
3876 | ||
3877 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
3878 | } | |
3879 | \f | |
3880 | static struct ofproto_dpif * | |
3881 | ofproto_dpif_lookup(const char *name) | |
3882 | { | |
3883 | struct ofproto *ofproto = ofproto_lookup(name); | |
3884 | return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class | |
3885 | ? ofproto_dpif_cast(ofproto) | |
3886 | : NULL); | |
3887 | } | |
3888 | ||
3889 | static void | |
3890 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, | |
3891 | const char *args, void *aux OVS_UNUSED) | |
3892 | { | |
3893 | struct ds ds = DS_EMPTY_INITIALIZER; | |
3894 | const struct ofproto_dpif *ofproto; | |
3895 | const struct mac_entry *e; | |
3896 | ||
3897 | ofproto = ofproto_dpif_lookup(args); | |
3898 | if (!ofproto) { | |
3899 | unixctl_command_reply(conn, 501, "no such bridge"); | |
3900 | return; | |
3901 | } | |
3902 | ||
3903 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
3904 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
3905 | struct ofbundle *bundle = e->port.p; | |
3906 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
3907 | ofbundle_get_a_port(bundle)->odp_port, | |
3908 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
3909 | } | |
3910 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
3911 | ds_destroy(&ds); | |
3912 | } | |
3913 | ||
3914 | struct ofproto_trace { | |
3915 | struct action_xlate_ctx ctx; | |
3916 | struct flow flow; | |
3917 | struct ds *result; | |
3918 | }; | |
3919 | ||
3920 | static void | |
3921 | trace_format_rule(struct ds *result, int level, const struct rule *rule) | |
3922 | { | |
3923 | ds_put_char_multiple(result, '\t', level); | |
3924 | if (!rule) { | |
3925 | ds_put_cstr(result, "No match\n"); | |
3926 | return; | |
3927 | } | |
3928 | ||
3929 | ds_put_format(result, "Rule: cookie=%#"PRIx64" ", | |
3930 | ntohll(rule->flow_cookie)); | |
3931 | cls_rule_format(&rule->cr, result); | |
3932 | ds_put_char(result, '\n'); | |
3933 | ||
3934 | ds_put_char_multiple(result, '\t', level); | |
3935 | ds_put_cstr(result, "OpenFlow "); | |
b4b8c781 | 3936 | ofp_print_actions(result, rule->actions, rule->n_actions); |
abe529af BP |
3937 | ds_put_char(result, '\n'); |
3938 | } | |
3939 | ||
3940 | static void | |
3941 | trace_format_flow(struct ds *result, int level, const char *title, | |
3942 | struct ofproto_trace *trace) | |
3943 | { | |
3944 | ds_put_char_multiple(result, '\t', level); | |
3945 | ds_put_format(result, "%s: ", title); | |
3946 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
3947 | ds_put_cstr(result, "unchanged"); | |
3948 | } else { | |
3949 | flow_format(result, &trace->ctx.flow); | |
3950 | trace->flow = trace->ctx.flow; | |
3951 | } | |
3952 | ds_put_char(result, '\n'); | |
3953 | } | |
3954 | ||
3955 | static void | |
3956 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
3957 | { | |
3958 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
3959 | struct ds *result = trace->result; | |
3960 | ||
3961 | ds_put_char(result, '\n'); | |
3962 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
3963 | trace_format_rule(result, ctx->recurse + 1, &rule->up); | |
3964 | } | |
3965 | ||
3966 | static void | |
3967 | ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, | |
3968 | void *aux OVS_UNUSED) | |
3969 | { | |
3970 | char *dpname, *in_port_s, *tun_id_s, *packet_s; | |
3971 | char *args = xstrdup(args_); | |
3972 | char *save_ptr = NULL; | |
3973 | struct ofproto_dpif *ofproto; | |
3974 | struct ofpbuf packet; | |
3975 | struct rule_dpif *rule; | |
3976 | struct ds result; | |
3977 | struct flow flow; | |
3978 | uint16_t in_port; | |
3979 | ovs_be64 tun_id; | |
3980 | char *s; | |
3981 | ||
3982 | ofpbuf_init(&packet, strlen(args) / 2); | |
3983 | ds_init(&result); | |
3984 | ||
3985 | dpname = strtok_r(args, " ", &save_ptr); | |
3986 | tun_id_s = strtok_r(NULL, " ", &save_ptr); | |
3987 | in_port_s = strtok_r(NULL, " ", &save_ptr); | |
3988 | packet_s = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ | |
3989 | if (!dpname || !in_port_s || !packet_s) { | |
3990 | unixctl_command_reply(conn, 501, "Bad command syntax"); | |
3991 | goto exit; | |
3992 | } | |
3993 | ||
3994 | ofproto = ofproto_dpif_lookup(dpname); | |
3995 | if (!ofproto) { | |
3996 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
3997 | "for help)"); | |
3998 | goto exit; | |
3999 | } | |
4000 | ||
4001 | tun_id = htonll(strtoull(tun_id_s, NULL, 0)); | |
4002 | in_port = ofp_port_to_odp_port(atoi(in_port_s)); | |
4003 | ||
4004 | packet_s = ofpbuf_put_hex(&packet, packet_s, NULL); | |
4005 | packet_s += strspn(packet_s, " "); | |
4006 | if (*packet_s != '\0') { | |
4007 | unixctl_command_reply(conn, 501, "Trailing garbage in command"); | |
4008 | goto exit; | |
4009 | } | |
4010 | if (packet.size < ETH_HEADER_LEN) { | |
4011 | unixctl_command_reply(conn, 501, "Packet data too short for Ethernet"); | |
4012 | goto exit; | |
4013 | } | |
4014 | ||
4015 | ds_put_cstr(&result, "Packet: "); | |
4016 | s = ofp_packet_to_string(packet.data, packet.size, packet.size); | |
4017 | ds_put_cstr(&result, s); | |
4018 | free(s); | |
4019 | ||
4020 | flow_extract(&packet, tun_id, in_port, &flow); | |
4021 | ds_put_cstr(&result, "Flow: "); | |
4022 | flow_format(&result, &flow); | |
4023 | ds_put_char(&result, '\n'); | |
4024 | ||
4025 | rule = rule_dpif_lookup(ofproto, &flow); | |
4026 | trace_format_rule(&result, 0, &rule->up); | |
4027 | if (rule) { | |
4028 | struct ofproto_trace trace; | |
4029 | struct ofpbuf *odp_actions; | |
4030 | ||
4031 | trace.result = &result; | |
4032 | trace.flow = flow; | |
4033 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, &packet); | |
4034 | trace.ctx.resubmit_hook = trace_resubmit; | |
4035 | odp_actions = xlate_actions(&trace.ctx, | |
4036 | rule->up.actions, rule->up.n_actions); | |
4037 | ||
4038 | ds_put_char(&result, '\n'); | |
4039 | trace_format_flow(&result, 0, "Final flow", &trace); | |
4040 | ds_put_cstr(&result, "Datapath actions: "); | |
4041 | format_odp_actions(&result, odp_actions->data, odp_actions->size); | |
4042 | ofpbuf_delete(odp_actions); | |
4043 | } | |
4044 | ||
4045 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
4046 | ||
4047 | exit: | |
4048 | ds_destroy(&result); | |
4049 | ofpbuf_uninit(&packet); | |
4050 | free(args); | |
4051 | } | |
4052 | ||
7ee20df1 BP |
4053 | static void |
4054 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, | |
4055 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
4056 | { | |
4057 | clogged = true; | |
4058 | unixctl_command_reply(conn, 200, NULL); | |
4059 | } | |
4060 | ||
4061 | static void | |
4062 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, | |
4063 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
4064 | { | |
4065 | clogged = false; | |
4066 | unixctl_command_reply(conn, 200, NULL); | |
4067 | } | |
4068 | ||
abe529af BP |
4069 | static void |
4070 | ofproto_dpif_unixctl_init(void) | |
4071 | { | |
4072 | static bool registered; | |
4073 | if (registered) { | |
4074 | return; | |
4075 | } | |
4076 | registered = true; | |
4077 | ||
4078 | unixctl_command_register("ofproto/trace", ofproto_unixctl_trace, NULL); | |
4079 | unixctl_command_register("fdb/show", ofproto_unixctl_fdb_show, NULL); | |
7ee20df1 BP |
4080 | |
4081 | unixctl_command_register("ofproto/clog", ofproto_dpif_clog, NULL); | |
4082 | unixctl_command_register("ofproto/unclog", ofproto_dpif_unclog, NULL); | |
abe529af BP |
4083 | } |
4084 | \f | |
4085 | const struct ofproto_class ofproto_dpif_class = { | |
4086 | enumerate_types, | |
4087 | enumerate_names, | |
4088 | del, | |
4089 | alloc, | |
4090 | construct, | |
4091 | destruct, | |
4092 | dealloc, | |
4093 | run, | |
4094 | wait, | |
4095 | flush, | |
6c1491fb BP |
4096 | get_features, |
4097 | get_tables, | |
abe529af BP |
4098 | port_alloc, |
4099 | port_construct, | |
4100 | port_destruct, | |
4101 | port_dealloc, | |
4102 | port_modified, | |
4103 | port_reconfigured, | |
4104 | port_query_by_name, | |
4105 | port_add, | |
4106 | port_del, | |
4107 | port_dump_start, | |
4108 | port_dump_next, | |
4109 | port_dump_done, | |
4110 | port_poll, | |
4111 | port_poll_wait, | |
4112 | port_is_lacp_current, | |
0ab6decf | 4113 | NULL, /* rule_choose_table */ |
abe529af BP |
4114 | rule_alloc, |
4115 | rule_construct, | |
4116 | rule_destruct, | |
4117 | rule_dealloc, | |
abe529af BP |
4118 | rule_get_stats, |
4119 | rule_execute, | |
4120 | rule_modify_actions, | |
4121 | get_drop_frags, | |
4122 | set_drop_frags, | |
4123 | packet_out, | |
4124 | set_netflow, | |
4125 | get_netflow_ids, | |
4126 | set_sflow, | |
4127 | set_cfm, | |
a5610457 | 4128 | get_cfm_fault, |
abe529af BP |
4129 | bundle_set, |
4130 | bundle_remove, | |
4131 | mirror_set, | |
4132 | set_flood_vlans, | |
4133 | is_mirror_output_bundle, | |
4134 | }; |