]>
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" | |
75a75043 | 35 | #include "learn.h" |
abe529af BP |
36 | #include "mac-learning.h" |
37 | #include "multipath.h" | |
38 | #include "netdev.h" | |
39 | #include "netlink.h" | |
40 | #include "nx-match.h" | |
41 | #include "odp-util.h" | |
42 | #include "ofp-util.h" | |
43 | #include "ofpbuf.h" | |
44 | #include "ofp-print.h" | |
bae473fe | 45 | #include "ofproto-dpif-sflow.h" |
abe529af BP |
46 | #include "poll-loop.h" |
47 | #include "timer.h" | |
6c1491fb | 48 | #include "unaligned.h" |
abe529af BP |
49 | #include "unixctl.h" |
50 | #include "vlan-bitmap.h" | |
51 | #include "vlog.h" | |
52 | ||
53 | VLOG_DEFINE_THIS_MODULE(ofproto_dpif); | |
54 | ||
55 | COVERAGE_DEFINE(ofproto_dpif_ctlr_action); | |
56 | COVERAGE_DEFINE(ofproto_dpif_expired); | |
57 | COVERAGE_DEFINE(ofproto_dpif_no_packet_in); | |
58 | COVERAGE_DEFINE(ofproto_dpif_xlate); | |
59 | COVERAGE_DEFINE(facet_changed_rule); | |
60 | COVERAGE_DEFINE(facet_invalidated); | |
61 | COVERAGE_DEFINE(facet_revalidate); | |
62 | COVERAGE_DEFINE(facet_unexpected); | |
63 | ||
29901626 | 64 | /* Maximum depth of flow table recursion (due to resubmit actions) in a |
abe529af BP |
65 | * flow translation. */ |
66 | #define MAX_RESUBMIT_RECURSION 16 | |
67 | ||
9cdaaebe BP |
68 | /* Number of implemented OpenFlow tables. */ |
69 | enum { N_TABLES = 255 }; | |
70 | BUILD_ASSERT_DECL(N_TABLES >= 1 && N_TABLES <= 255); | |
71 | ||
abe529af BP |
72 | struct ofport_dpif; |
73 | struct ofproto_dpif; | |
74 | ||
75 | struct rule_dpif { | |
76 | struct rule up; | |
77 | ||
78 | long long int used; /* Time last used; time created if not used. */ | |
79 | ||
80 | /* These statistics: | |
81 | * | |
82 | * - Do include packets and bytes from facets that have been deleted or | |
83 | * whose own statistics have been folded into the rule. | |
84 | * | |
85 | * - Do include packets and bytes sent "by hand" that were accounted to | |
86 | * the rule without any facet being involved (this is a rare corner | |
87 | * case in rule_execute()). | |
88 | * | |
89 | * - Do not include packet or bytes that can be obtained from any facet's | |
90 | * packet_count or byte_count member or that can be obtained from the | |
91 | * datapath by, e.g., dpif_flow_get() for any facet. | |
92 | */ | |
93 | uint64_t packet_count; /* Number of packets received. */ | |
94 | uint64_t byte_count; /* Number of bytes received. */ | |
95 | ||
54a9cbc9 BP |
96 | tag_type tag; /* Caches rule_calculate_tag() result. */ |
97 | ||
abe529af BP |
98 | struct list facets; /* List of "struct facet"s. */ |
99 | }; | |
100 | ||
101 | static struct rule_dpif *rule_dpif_cast(const struct rule *rule) | |
102 | { | |
103 | return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL; | |
104 | } | |
105 | ||
29901626 BP |
106 | static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *, |
107 | const struct flow *, uint8_t table); | |
abe529af BP |
108 | |
109 | #define MAX_MIRRORS 32 | |
110 | typedef uint32_t mirror_mask_t; | |
111 | #define MIRROR_MASK_C(X) UINT32_C(X) | |
112 | BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS); | |
113 | struct ofmirror { | |
114 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
115 | size_t idx; /* In ofproto's "mirrors" array. */ | |
116 | void *aux; /* Key supplied by ofproto's client. */ | |
117 | char *name; /* Identifier for log messages. */ | |
118 | ||
119 | /* Selection criteria. */ | |
120 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
121 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
122 | unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */ | |
123 | ||
124 | /* Output (mutually exclusive). */ | |
125 | struct ofbundle *out; /* Output port or NULL. */ | |
126 | int out_vlan; /* Output VLAN or -1. */ | |
127 | }; | |
128 | ||
129 | static void mirror_destroy(struct ofmirror *); | |
130 | ||
131 | /* A group of one or more OpenFlow ports. */ | |
132 | #define OFBUNDLE_FLOOD ((struct ofbundle *) 1) | |
133 | struct ofbundle { | |
134 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
135 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
136 | void *aux; /* Key supplied by ofproto's client. */ | |
137 | char *name; /* Identifier for log messages. */ | |
138 | ||
139 | /* Configuration. */ | |
140 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 141 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
142 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
143 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
144 | * NULL if all VLANs are trunked. */ | |
145 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
146 | struct bond *bond; /* Nonnull iff more than one port. */ | |
147 | ||
148 | /* Status. */ | |
149 | bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */ | |
150 | ||
151 | /* Port mirroring info. */ | |
152 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
153 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
154 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
155 | }; | |
156 | ||
157 | static void bundle_remove(struct ofport *); | |
158 | static void bundle_destroy(struct ofbundle *); | |
159 | static void bundle_del_port(struct ofport_dpif *); | |
160 | static void bundle_run(struct ofbundle *); | |
161 | static void bundle_wait(struct ofbundle *); | |
162 | ||
163 | struct action_xlate_ctx { | |
164 | /* action_xlate_ctx_init() initializes these members. */ | |
165 | ||
166 | /* The ofproto. */ | |
167 | struct ofproto_dpif *ofproto; | |
168 | ||
169 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
170 | * this flow when actions change header fields. */ | |
171 | struct flow flow; | |
172 | ||
173 | /* The packet corresponding to 'flow', or a null pointer if we are | |
174 | * revalidating without a packet to refer to. */ | |
175 | const struct ofpbuf *packet; | |
176 | ||
75a75043 BP |
177 | /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We |
178 | * want to execute them if we are actually processing a packet, or if we | |
179 | * are accounting for packets that the datapath has processed, but not if | |
180 | * we are just revalidating. */ | |
181 | bool may_learn; | |
182 | ||
abe529af BP |
183 | /* If nonnull, called just before executing a resubmit action. |
184 | * | |
185 | * This is normally null so the client has to set it manually after | |
186 | * calling action_xlate_ctx_init(). */ | |
187 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *); | |
188 | ||
abe529af BP |
189 | /* xlate_actions() initializes and uses these members. The client might want |
190 | * to look at them after it returns. */ | |
191 | ||
192 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 193 | tag_type tags; /* Tags associated with actions. */ |
abe529af BP |
194 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
195 | * be reassessed for every packet. */ | |
75a75043 BP |
196 | bool has_learn; /* Actions include NXAST_LEARN? */ |
197 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
abe529af BP |
198 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
199 | ||
200 | /* xlate_actions() initializes and uses these members, but the client has no | |
201 | * reason to look at them. */ | |
202 | ||
203 | int recurse; /* Recursion level, via xlate_table_action. */ | |
afabef2b | 204 | uint32_t priority; /* Current flow priority. 0 if none. */ |
b3e9b2ed EJ |
205 | struct flow base_flow; /* Flow at the last commit. */ |
206 | uint32_t base_priority; /* Priority at the last commit. */ | |
29901626 | 207 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
208 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
209 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
210 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
abe529af BP |
211 | }; |
212 | ||
213 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
214 | struct ofproto_dpif *, const struct flow *, | |
215 | const struct ofpbuf *); | |
216 | static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, | |
217 | const union ofp_action *in, size_t n_in); | |
218 | ||
219 | /* An exact-match instantiation of an OpenFlow flow. */ | |
220 | struct facet { | |
221 | long long int used; /* Time last used; time created if not used. */ | |
222 | ||
223 | /* These statistics: | |
224 | * | |
225 | * - Do include packets and bytes sent "by hand", e.g. with | |
226 | * dpif_execute(). | |
227 | * | |
228 | * - Do include packets and bytes that were obtained from the datapath | |
907a4c5e | 229 | * when its statistics were reset (e.g. dpif_flow_put() with |
abe529af | 230 | * DPIF_FP_ZERO_STATS). |
abe529af BP |
231 | */ |
232 | uint64_t packet_count; /* Number of packets received. */ | |
233 | uint64_t byte_count; /* Number of bytes received. */ | |
234 | ||
235 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
236 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
237 | ||
238 | uint64_t rs_packet_count; /* Packets pushed to resubmit children. */ | |
239 | uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */ | |
240 | long long int rs_used; /* Used time pushed to resubmit children. */ | |
241 | ||
907a4c5e | 242 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
abe529af BP |
243 | |
244 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
245 | struct list list_node; /* In owning rule's 'facets' list. */ | |
246 | struct rule_dpif *rule; /* Owning rule. */ | |
247 | struct flow flow; /* Exact-match flow. */ | |
248 | bool installed; /* Installed in datapath? */ | |
249 | bool may_install; /* True ordinarily; false if actions must | |
250 | * be reassessed for every packet. */ | |
75a75043 BP |
251 | bool has_learn; /* Actions include NXAST_LEARN? */ |
252 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
abe529af BP |
253 | size_t actions_len; /* Number of bytes in actions[]. */ |
254 | struct nlattr *actions; /* Datapath actions. */ | |
255 | tag_type tags; /* Tags. */ | |
256 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
257 | }; | |
258 | ||
f3827897 | 259 | static struct facet *facet_create(struct rule_dpif *, const struct flow *); |
abe529af BP |
260 | static void facet_remove(struct ofproto_dpif *, struct facet *); |
261 | static void facet_free(struct facet *); | |
262 | ||
263 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); | |
264 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, | |
265 | const struct flow *); | |
266 | static bool facet_revalidate(struct ofproto_dpif *, struct facet *); | |
267 | ||
3d9e05f8 BP |
268 | static bool execute_controller_action(struct ofproto_dpif *, |
269 | const struct flow *, | |
270 | const struct nlattr *odp_actions, | |
271 | size_t actions_len, | |
272 | struct ofpbuf *packet); | |
abe529af BP |
273 | static void facet_execute(struct ofproto_dpif *, struct facet *, |
274 | struct ofpbuf *packet); | |
275 | ||
276 | static int facet_put__(struct ofproto_dpif *, struct facet *, | |
277 | const struct nlattr *actions, size_t actions_len, | |
278 | struct dpif_flow_stats *); | |
279 | static void facet_install(struct ofproto_dpif *, struct facet *, | |
280 | bool zero_stats); | |
281 | static void facet_uninstall(struct ofproto_dpif *, struct facet *); | |
282 | static void facet_flush_stats(struct ofproto_dpif *, struct facet *); | |
283 | ||
284 | static void facet_make_actions(struct ofproto_dpif *, struct facet *, | |
285 | const struct ofpbuf *packet); | |
286 | static void facet_update_time(struct ofproto_dpif *, struct facet *, | |
287 | long long int used); | |
288 | static void facet_update_stats(struct ofproto_dpif *, struct facet *, | |
289 | const struct dpif_flow_stats *); | |
bbb5d219 | 290 | static void facet_reset_counters(struct facet *); |
3a88e544 | 291 | static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *); |
abe529af | 292 | static void facet_push_stats(struct facet *); |
55af77bb | 293 | static void facet_account(struct ofproto_dpif *, struct facet *); |
abe529af BP |
294 | |
295 | static bool facet_is_controller_flow(struct facet *); | |
296 | ||
297 | static void flow_push_stats(const struct rule_dpif *, | |
298 | struct flow *, uint64_t packets, uint64_t bytes, | |
299 | long long int used); | |
300 | ||
54a9cbc9 BP |
301 | static uint32_t rule_calculate_tag(const struct flow *, |
302 | const struct flow_wildcards *, | |
303 | uint32_t basis); | |
304 | static void rule_invalidate(const struct rule_dpif *); | |
305 | ||
abe529af BP |
306 | struct ofport_dpif { |
307 | struct ofport up; | |
308 | ||
309 | uint32_t odp_port; | |
310 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
311 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
312 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
313 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 314 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 315 | bool may_enable; /* May be enabled in bonds. */ |
abe529af BP |
316 | }; |
317 | ||
318 | static struct ofport_dpif * | |
319 | ofport_dpif_cast(const struct ofport *ofport) | |
320 | { | |
321 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
322 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
323 | } | |
324 | ||
325 | static void port_run(struct ofport_dpif *); | |
326 | static void port_wait(struct ofport_dpif *); | |
a5610457 | 327 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
abe529af | 328 | |
7ee20df1 BP |
329 | struct dpif_completion { |
330 | struct list list_node; | |
331 | struct ofoperation *op; | |
332 | }; | |
333 | ||
54a9cbc9 BP |
334 | /* Extra information about a classifier table. |
335 | * Currently used just for optimized flow revalidation. */ | |
336 | struct table_dpif { | |
337 | /* If either of these is nonnull, then this table has a form that allows | |
338 | * flows to be tagged to avoid revalidating most flows for the most common | |
339 | * kinds of flow table changes. */ | |
340 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
341 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
342 | uint32_t basis; /* Keeps each table's tags separate. */ | |
343 | }; | |
344 | ||
abe529af BP |
345 | struct ofproto_dpif { |
346 | struct ofproto up; | |
347 | struct dpif *dpif; | |
348 | int max_ports; | |
349 | ||
6c1491fb BP |
350 | /* Statistics. */ |
351 | uint64_t n_matches; | |
352 | ||
abe529af BP |
353 | /* Bridging. */ |
354 | struct netflow *netflow; | |
bae473fe | 355 | struct dpif_sflow *sflow; |
abe529af BP |
356 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
357 | struct mac_learning *ml; | |
358 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
359 | bool has_bonded_bundles; | |
360 | ||
361 | /* Expiration. */ | |
362 | struct timer next_expiration; | |
363 | ||
364 | /* Facets. */ | |
365 | struct hmap facets; | |
54a9cbc9 BP |
366 | |
367 | /* Revalidation. */ | |
368 | struct table_dpif tables[N_TABLES]; | |
abe529af BP |
369 | bool need_revalidate; |
370 | struct tag_set revalidate_set; | |
7ee20df1 BP |
371 | |
372 | /* Support for debugging async flow mods. */ | |
373 | struct list completions; | |
daff3353 EJ |
374 | |
375 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
abe529af BP |
376 | }; |
377 | ||
7ee20df1 BP |
378 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
379 | * for debugging the asynchronous flow_mod implementation.) */ | |
380 | static bool clogged; | |
381 | ||
abe529af BP |
382 | static void ofproto_dpif_unixctl_init(void); |
383 | ||
384 | static struct ofproto_dpif * | |
385 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
386 | { | |
387 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
388 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
389 | } | |
390 | ||
391 | static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, | |
392 | uint16_t ofp_port); | |
393 | static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, | |
394 | uint32_t odp_port); | |
395 | ||
396 | /* Packet processing. */ | |
397 | static void update_learning_table(struct ofproto_dpif *, | |
398 | const struct flow *, int vlan, | |
399 | struct ofbundle *); | |
400 | static bool is_admissible(struct ofproto_dpif *, const struct flow *, | |
401 | bool have_packet, tag_type *, int *vlanp, | |
402 | struct ofbundle **in_bundlep); | |
501f8d1f BP |
403 | |
404 | /* Upcalls. */ | |
405 | #define FLOW_MISS_MAX_BATCH 50 | |
abe529af | 406 | static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *); |
501f8d1f BP |
407 | static void handle_miss_upcalls(struct ofproto_dpif *, |
408 | struct dpif_upcall *, size_t n); | |
abe529af BP |
409 | |
410 | /* Flow expiration. */ | |
411 | static int expire(struct ofproto_dpif *); | |
412 | ||
413 | /* Utilities. */ | |
b2fda3ef | 414 | static int send_packet(struct ofproto_dpif *, uint32_t odp_port, |
abe529af | 415 | const struct ofpbuf *packet); |
6ff686f2 PS |
416 | static size_t |
417 | compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, | |
418 | const struct flow *, uint32_t odp_port); | |
abe529af BP |
419 | /* Global variables. */ |
420 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
421 | \f | |
422 | /* Factory functions. */ | |
423 | ||
424 | static void | |
425 | enumerate_types(struct sset *types) | |
426 | { | |
427 | dp_enumerate_types(types); | |
428 | } | |
429 | ||
430 | static int | |
431 | enumerate_names(const char *type, struct sset *names) | |
432 | { | |
433 | return dp_enumerate_names(type, names); | |
434 | } | |
435 | ||
436 | static int | |
437 | del(const char *type, const char *name) | |
438 | { | |
439 | struct dpif *dpif; | |
440 | int error; | |
441 | ||
442 | error = dpif_open(name, type, &dpif); | |
443 | if (!error) { | |
444 | error = dpif_delete(dpif); | |
445 | dpif_close(dpif); | |
446 | } | |
447 | return error; | |
448 | } | |
449 | \f | |
450 | /* Basic life-cycle. */ | |
451 | ||
452 | static struct ofproto * | |
453 | alloc(void) | |
454 | { | |
455 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
456 | return &ofproto->up; | |
457 | } | |
458 | ||
459 | static void | |
460 | dealloc(struct ofproto *ofproto_) | |
461 | { | |
462 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
463 | free(ofproto); | |
464 | } | |
465 | ||
466 | static int | |
073e2a6f | 467 | construct(struct ofproto *ofproto_, int *n_tablesp) |
abe529af BP |
468 | { |
469 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
470 | const char *name = ofproto->up.name; | |
471 | int error; | |
472 | int i; | |
473 | ||
474 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
475 | if (error) { | |
476 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
477 | return error; | |
478 | } | |
479 | ||
480 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 481 | ofproto->n_matches = 0; |
abe529af | 482 | |
be8194bb JG |
483 | dpif_flow_flush(ofproto->dpif); |
484 | dpif_recv_purge(ofproto->dpif); | |
485 | ||
abe529af BP |
486 | error = dpif_recv_set_mask(ofproto->dpif, |
487 | ((1u << DPIF_UC_MISS) | | |
6ff686f2 | 488 | (1u << DPIF_UC_ACTION))); |
abe529af BP |
489 | if (error) { |
490 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
491 | dpif_close(ofproto->dpif); | |
492 | return error; | |
493 | } | |
abe529af BP |
494 | |
495 | ofproto->netflow = NULL; | |
496 | ofproto->sflow = NULL; | |
497 | hmap_init(&ofproto->bundles); | |
498 | ofproto->ml = mac_learning_create(); | |
499 | for (i = 0; i < MAX_MIRRORS; i++) { | |
500 | ofproto->mirrors[i] = NULL; | |
501 | } | |
502 | ofproto->has_bonded_bundles = false; | |
503 | ||
504 | timer_set_duration(&ofproto->next_expiration, 1000); | |
505 | ||
506 | hmap_init(&ofproto->facets); | |
54a9cbc9 BP |
507 | |
508 | for (i = 0; i < N_TABLES; i++) { | |
509 | struct table_dpif *table = &ofproto->tables[i]; | |
510 | ||
511 | table->catchall_table = NULL; | |
512 | table->other_table = NULL; | |
513 | table->basis = random_uint32(); | |
514 | } | |
abe529af BP |
515 | ofproto->need_revalidate = false; |
516 | tag_set_init(&ofproto->revalidate_set); | |
517 | ||
7ee20df1 BP |
518 | list_init(&ofproto->completions); |
519 | ||
abe529af BP |
520 | ofproto_dpif_unixctl_init(); |
521 | ||
daff3353 EJ |
522 | ofproto->has_bundle_action = false; |
523 | ||
9cdaaebe | 524 | *n_tablesp = N_TABLES; |
abe529af BP |
525 | return 0; |
526 | } | |
527 | ||
7ee20df1 BP |
528 | static void |
529 | complete_operations(struct ofproto_dpif *ofproto) | |
530 | { | |
531 | struct dpif_completion *c, *next; | |
532 | ||
533 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
534 | ofoperation_complete(c->op, 0); | |
535 | list_remove(&c->list_node); | |
536 | free(c); | |
537 | } | |
538 | } | |
539 | ||
abe529af BP |
540 | static void |
541 | destruct(struct ofproto *ofproto_) | |
542 | { | |
543 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 544 | struct rule_dpif *rule, *next_rule; |
0697b5c3 | 545 | struct classifier *table; |
abe529af BP |
546 | int i; |
547 | ||
7ee20df1 BP |
548 | complete_operations(ofproto); |
549 | ||
0697b5c3 BP |
550 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
551 | struct cls_cursor cursor; | |
552 | ||
553 | cls_cursor_init(&cursor, table, NULL); | |
554 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
555 | ofproto_rule_destroy(&rule->up); | |
556 | } | |
7ee20df1 BP |
557 | } |
558 | ||
abe529af BP |
559 | for (i = 0; i < MAX_MIRRORS; i++) { |
560 | mirror_destroy(ofproto->mirrors[i]); | |
561 | } | |
562 | ||
563 | netflow_destroy(ofproto->netflow); | |
bae473fe | 564 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
565 | hmap_destroy(&ofproto->bundles); |
566 | mac_learning_destroy(ofproto->ml); | |
567 | ||
568 | hmap_destroy(&ofproto->facets); | |
569 | ||
570 | dpif_close(ofproto->dpif); | |
571 | } | |
572 | ||
573 | static int | |
574 | run(struct ofproto *ofproto_) | |
575 | { | |
576 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
501f8d1f | 577 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
abe529af BP |
578 | struct ofport_dpif *ofport; |
579 | struct ofbundle *bundle; | |
501f8d1f | 580 | size_t n_misses; |
abe529af BP |
581 | int i; |
582 | ||
7ee20df1 BP |
583 | if (!clogged) { |
584 | complete_operations(ofproto); | |
585 | } | |
abe529af BP |
586 | dpif_run(ofproto->dpif); |
587 | ||
501f8d1f BP |
588 | n_misses = 0; |
589 | for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) { | |
590 | struct dpif_upcall *upcall = &misses[n_misses]; | |
abe529af BP |
591 | int error; |
592 | ||
501f8d1f | 593 | error = dpif_recv(ofproto->dpif, upcall); |
abe529af | 594 | if (error) { |
501f8d1f | 595 | if (error == ENODEV && n_misses == 0) { |
abe529af BP |
596 | return error; |
597 | } | |
598 | break; | |
599 | } | |
600 | ||
501f8d1f BP |
601 | if (upcall->type == DPIF_UC_MISS) { |
602 | /* Handle it later. */ | |
603 | n_misses++; | |
604 | } else { | |
605 | handle_upcall(ofproto, upcall); | |
606 | } | |
abe529af BP |
607 | } |
608 | ||
501f8d1f BP |
609 | handle_miss_upcalls(ofproto, misses, n_misses); |
610 | ||
abe529af BP |
611 | if (timer_expired(&ofproto->next_expiration)) { |
612 | int delay = expire(ofproto); | |
613 | timer_set_duration(&ofproto->next_expiration, delay); | |
614 | } | |
615 | ||
616 | if (ofproto->netflow) { | |
617 | netflow_run(ofproto->netflow); | |
618 | } | |
619 | if (ofproto->sflow) { | |
bae473fe | 620 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
621 | } |
622 | ||
623 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
624 | port_run(ofport); | |
625 | } | |
626 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
627 | bundle_run(bundle); | |
628 | } | |
629 | ||
1c313b88 BP |
630 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
631 | ||
abe529af BP |
632 | /* Now revalidate if there's anything to do. */ |
633 | if (ofproto->need_revalidate | |
634 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
635 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
636 | bool revalidate_all = ofproto->need_revalidate; | |
637 | struct facet *facet, *next; | |
638 | ||
639 | /* Clear the revalidation flags. */ | |
640 | tag_set_init(&ofproto->revalidate_set); | |
641 | ofproto->need_revalidate = false; | |
642 | ||
643 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
644 | if (revalidate_all | |
645 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
646 | facet_revalidate(ofproto, facet); | |
647 | } | |
648 | } | |
649 | } | |
650 | ||
651 | return 0; | |
652 | } | |
653 | ||
654 | static void | |
655 | wait(struct ofproto *ofproto_) | |
656 | { | |
657 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
658 | struct ofport_dpif *ofport; | |
659 | struct ofbundle *bundle; | |
660 | ||
7ee20df1 BP |
661 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
662 | poll_immediate_wake(); | |
663 | } | |
664 | ||
abe529af BP |
665 | dpif_wait(ofproto->dpif); |
666 | dpif_recv_wait(ofproto->dpif); | |
667 | if (ofproto->sflow) { | |
bae473fe | 668 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
669 | } |
670 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
671 | poll_immediate_wake(); | |
672 | } | |
673 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
674 | port_wait(ofport); | |
675 | } | |
676 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
677 | bundle_wait(bundle); | |
678 | } | |
1c313b88 | 679 | mac_learning_wait(ofproto->ml); |
abe529af BP |
680 | if (ofproto->need_revalidate) { |
681 | /* Shouldn't happen, but if it does just go around again. */ | |
682 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
683 | poll_immediate_wake(); | |
684 | } else { | |
685 | timer_wait(&ofproto->next_expiration); | |
686 | } | |
687 | } | |
688 | ||
689 | static void | |
690 | flush(struct ofproto *ofproto_) | |
691 | { | |
692 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
693 | struct facet *facet, *next_facet; | |
694 | ||
695 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
696 | /* Mark the facet as not installed so that facet_remove() doesn't | |
697 | * bother trying to uninstall it. There is no point in uninstalling it | |
698 | * individually since we are about to blow away all the facets with | |
699 | * dpif_flow_flush(). */ | |
700 | facet->installed = false; | |
701 | facet->dp_packet_count = 0; | |
702 | facet->dp_byte_count = 0; | |
703 | facet_remove(ofproto, facet); | |
704 | } | |
705 | dpif_flow_flush(ofproto->dpif); | |
706 | } | |
707 | ||
6c1491fb BP |
708 | static void |
709 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
710 | bool *arp_match_ip, uint32_t *actions) | |
711 | { | |
712 | *arp_match_ip = true; | |
713 | *actions = ((1u << OFPAT_OUTPUT) | | |
714 | (1u << OFPAT_SET_VLAN_VID) | | |
715 | (1u << OFPAT_SET_VLAN_PCP) | | |
716 | (1u << OFPAT_STRIP_VLAN) | | |
717 | (1u << OFPAT_SET_DL_SRC) | | |
718 | (1u << OFPAT_SET_DL_DST) | | |
719 | (1u << OFPAT_SET_NW_SRC) | | |
720 | (1u << OFPAT_SET_NW_DST) | | |
721 | (1u << OFPAT_SET_NW_TOS) | | |
722 | (1u << OFPAT_SET_TP_SRC) | | |
723 | (1u << OFPAT_SET_TP_DST) | | |
724 | (1u << OFPAT_ENQUEUE)); | |
725 | } | |
726 | ||
727 | static void | |
728 | get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) | |
729 | { | |
730 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 731 | struct dpif_dp_stats s; |
6c1491fb BP |
732 | |
733 | strcpy(ots->name, "classifier"); | |
734 | ||
735 | dpif_get_dp_stats(ofproto->dpif, &s); | |
736 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
737 | put_32aligned_be64(&ots->matched_count, | |
738 | htonll(s.n_hit + ofproto->n_matches)); | |
739 | } | |
740 | ||
abe529af BP |
741 | static int |
742 | set_netflow(struct ofproto *ofproto_, | |
743 | const struct netflow_options *netflow_options) | |
744 | { | |
745 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
746 | ||
747 | if (netflow_options) { | |
748 | if (!ofproto->netflow) { | |
749 | ofproto->netflow = netflow_create(); | |
750 | } | |
751 | return netflow_set_options(ofproto->netflow, netflow_options); | |
752 | } else { | |
753 | netflow_destroy(ofproto->netflow); | |
754 | ofproto->netflow = NULL; | |
755 | return 0; | |
756 | } | |
757 | } | |
758 | ||
759 | static struct ofport * | |
760 | port_alloc(void) | |
761 | { | |
762 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
763 | return &port->up; | |
764 | } | |
765 | ||
766 | static void | |
767 | port_dealloc(struct ofport *port_) | |
768 | { | |
769 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
770 | free(port); | |
771 | } | |
772 | ||
773 | static int | |
774 | port_construct(struct ofport *port_) | |
775 | { | |
776 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
777 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
778 | ||
f11c28c4 | 779 | ofproto->need_revalidate = true; |
abe529af BP |
780 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
781 | port->bundle = NULL; | |
782 | port->cfm = NULL; | |
783 | port->tag = tag_create_random(); | |
d5ffa7f2 | 784 | port->may_enable = true; |
abe529af BP |
785 | |
786 | if (ofproto->sflow) { | |
bae473fe JP |
787 | dpif_sflow_add_port(ofproto->sflow, port->odp_port, |
788 | netdev_get_name(port->up.netdev)); | |
abe529af BP |
789 | } |
790 | ||
791 | return 0; | |
792 | } | |
793 | ||
794 | static void | |
795 | port_destruct(struct ofport *port_) | |
796 | { | |
797 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
798 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
799 | ||
f11c28c4 | 800 | ofproto->need_revalidate = true; |
abe529af | 801 | bundle_remove(port_); |
a5610457 | 802 | set_cfm(port_, NULL); |
abe529af | 803 | if (ofproto->sflow) { |
bae473fe | 804 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af BP |
805 | } |
806 | } | |
807 | ||
808 | static void | |
809 | port_modified(struct ofport *port_) | |
810 | { | |
811 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
812 | ||
813 | if (port->bundle && port->bundle->bond) { | |
814 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
815 | } | |
816 | } | |
817 | ||
818 | static void | |
819 | port_reconfigured(struct ofport *port_, ovs_be32 old_config) | |
820 | { | |
821 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
822 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
823 | ovs_be32 changed = old_config ^ port->up.opp.config; | |
824 | ||
825 | if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP | | |
826 | OFPPC_NO_FWD | OFPPC_NO_FLOOD)) { | |
827 | ofproto->need_revalidate = true; | |
828 | } | |
829 | } | |
830 | ||
831 | static int | |
832 | set_sflow(struct ofproto *ofproto_, | |
833 | const struct ofproto_sflow_options *sflow_options) | |
834 | { | |
835 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 836 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 837 | |
abe529af | 838 | if (sflow_options) { |
bae473fe | 839 | if (!ds) { |
abe529af BP |
840 | struct ofport_dpif *ofport; |
841 | ||
bae473fe | 842 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 843 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
bae473fe JP |
844 | dpif_sflow_add_port(ds, ofport->odp_port, |
845 | netdev_get_name(ofport->up.netdev)); | |
abe529af | 846 | } |
6ff686f2 | 847 | ofproto->need_revalidate = true; |
abe529af | 848 | } |
bae473fe | 849 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 850 | } else { |
6ff686f2 PS |
851 | if (ds) { |
852 | dpif_sflow_destroy(ds); | |
853 | ofproto->need_revalidate = true; | |
854 | ofproto->sflow = NULL; | |
855 | } | |
abe529af BP |
856 | } |
857 | return 0; | |
858 | } | |
859 | ||
860 | static int | |
a5610457 | 861 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
862 | { |
863 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
864 | int error; | |
865 | ||
a5610457 | 866 | if (!s) { |
abe529af BP |
867 | error = 0; |
868 | } else { | |
869 | if (!ofport->cfm) { | |
8c977421 EJ |
870 | struct ofproto_dpif *ofproto; |
871 | ||
872 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
873 | ofproto->need_revalidate = true; | |
6f629657 | 874 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
875 | } |
876 | ||
a5610457 | 877 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
878 | return 0; |
879 | } | |
880 | ||
881 | error = EINVAL; | |
882 | } | |
883 | cfm_destroy(ofport->cfm); | |
884 | ofport->cfm = NULL; | |
885 | return error; | |
886 | } | |
887 | ||
888 | static int | |
a5610457 | 889 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
890 | { |
891 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
892 | |
893 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 894 | } |
1de11730 EJ |
895 | |
896 | static int | |
897 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
898 | size_t *n_rmps) | |
899 | { | |
900 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
901 | ||
902 | if (ofport->cfm) { | |
903 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
904 | return 0; | |
905 | } else { | |
906 | return -1; | |
907 | } | |
908 | } | |
abe529af BP |
909 | \f |
910 | /* Bundles. */ | |
911 | ||
912 | /* Expires all MAC learning entries associated with 'port' and forces ofproto | |
913 | * to revalidate every flow. */ | |
914 | static void | |
915 | bundle_flush_macs(struct ofbundle *bundle) | |
916 | { | |
917 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
918 | struct mac_learning *ml = ofproto->ml; | |
919 | struct mac_entry *mac, *next_mac; | |
920 | ||
921 | ofproto->need_revalidate = true; | |
922 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
923 | if (mac->port.p == bundle) { | |
924 | mac_learning_expire(ml, mac); | |
925 | } | |
926 | } | |
927 | } | |
928 | ||
929 | static struct ofbundle * | |
930 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
931 | { | |
932 | struct ofbundle *bundle; | |
933 | ||
934 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
935 | &ofproto->bundles) { | |
936 | if (bundle->aux == aux) { | |
937 | return bundle; | |
938 | } | |
939 | } | |
940 | return NULL; | |
941 | } | |
942 | ||
943 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
944 | * ones that are found to 'bundles'. */ | |
945 | static void | |
946 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
947 | void **auxes, size_t n_auxes, | |
948 | struct hmapx *bundles) | |
949 | { | |
950 | size_t i; | |
951 | ||
952 | hmapx_init(bundles); | |
953 | for (i = 0; i < n_auxes; i++) { | |
954 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
955 | if (bundle) { | |
956 | hmapx_add(bundles, bundle); | |
957 | } | |
958 | } | |
959 | } | |
960 | ||
961 | static void | |
962 | bundle_del_port(struct ofport_dpif *port) | |
963 | { | |
964 | struct ofbundle *bundle = port->bundle; | |
965 | ||
6f77f4ae BP |
966 | bundle->ofproto->need_revalidate = true; |
967 | ||
abe529af BP |
968 | list_remove(&port->bundle_node); |
969 | port->bundle = NULL; | |
970 | ||
971 | if (bundle->lacp) { | |
972 | lacp_slave_unregister(bundle->lacp, port); | |
973 | } | |
974 | if (bundle->bond) { | |
975 | bond_slave_unregister(bundle->bond, port); | |
976 | } | |
977 | ||
978 | bundle->floodable = true; | |
979 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
980 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { | |
981 | bundle->floodable = false; | |
982 | } | |
983 | } | |
984 | } | |
985 | ||
986 | static bool | |
987 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
988 | struct lacp_slave_settings *lacp, |
989 | uint32_t bond_stable_id) | |
abe529af BP |
990 | { |
991 | struct ofport_dpif *port; | |
992 | ||
993 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
994 | if (!port) { | |
995 | return false; | |
996 | } | |
997 | ||
998 | if (port->bundle != bundle) { | |
6f77f4ae | 999 | bundle->ofproto->need_revalidate = true; |
abe529af BP |
1000 | if (port->bundle) { |
1001 | bundle_del_port(port); | |
1002 | } | |
1003 | ||
1004 | port->bundle = bundle; | |
1005 | list_push_back(&bundle->ports, &port->bundle_node); | |
1006 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD)) { | |
1007 | bundle->floodable = false; | |
1008 | } | |
1009 | } | |
1010 | if (lacp) { | |
4a86aece | 1011 | port->bundle->ofproto->need_revalidate = true; |
abe529af BP |
1012 | lacp_slave_register(bundle->lacp, port, lacp); |
1013 | } | |
1014 | ||
00794817 BP |
1015 | port->bond_stable_id = bond_stable_id; |
1016 | ||
abe529af BP |
1017 | return true; |
1018 | } | |
1019 | ||
1020 | static void | |
1021 | bundle_destroy(struct ofbundle *bundle) | |
1022 | { | |
1023 | struct ofproto_dpif *ofproto; | |
1024 | struct ofport_dpif *port, *next_port; | |
1025 | int i; | |
1026 | ||
1027 | if (!bundle) { | |
1028 | return; | |
1029 | } | |
1030 | ||
1031 | ofproto = bundle->ofproto; | |
1032 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1033 | struct ofmirror *m = ofproto->mirrors[i]; | |
1034 | if (m) { | |
1035 | if (m->out == bundle) { | |
1036 | mirror_destroy(m); | |
1037 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1038 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
1039 | ofproto->need_revalidate = true; | |
1040 | } | |
1041 | } | |
1042 | } | |
1043 | ||
1044 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1045 | bundle_del_port(port); | |
1046 | } | |
1047 | ||
1048 | bundle_flush_macs(bundle); | |
1049 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); | |
1050 | free(bundle->name); | |
1051 | free(bundle->trunks); | |
1052 | lacp_destroy(bundle->lacp); | |
1053 | bond_destroy(bundle->bond); | |
1054 | free(bundle); | |
1055 | } | |
1056 | ||
1057 | static int | |
1058 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1059 | const struct ofproto_bundle_settings *s) | |
1060 | { | |
1061 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1062 | bool need_flush = false; | |
abe529af BP |
1063 | struct ofport_dpif *port; |
1064 | struct ofbundle *bundle; | |
ecac4ebf BP |
1065 | unsigned long *trunks; |
1066 | int vlan; | |
abe529af BP |
1067 | size_t i; |
1068 | bool ok; | |
1069 | ||
1070 | if (!s) { | |
1071 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1076 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1077 | ||
1078 | bundle = bundle_lookup(ofproto, aux); | |
1079 | if (!bundle) { | |
1080 | bundle = xmalloc(sizeof *bundle); | |
1081 | ||
1082 | bundle->ofproto = ofproto; | |
1083 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1084 | hash_pointer(aux, 0)); | |
1085 | bundle->aux = aux; | |
1086 | bundle->name = NULL; | |
1087 | ||
1088 | list_init(&bundle->ports); | |
ecac4ebf | 1089 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1090 | bundle->vlan = -1; |
1091 | bundle->trunks = NULL; | |
1092 | bundle->lacp = NULL; | |
1093 | bundle->bond = NULL; | |
1094 | ||
1095 | bundle->floodable = true; | |
1096 | ||
1097 | bundle->src_mirrors = 0; | |
1098 | bundle->dst_mirrors = 0; | |
1099 | bundle->mirror_out = 0; | |
1100 | } | |
1101 | ||
1102 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1103 | free(bundle->name); | |
1104 | bundle->name = xstrdup(s->name); | |
1105 | } | |
1106 | ||
1107 | /* LACP. */ | |
1108 | if (s->lacp) { | |
1109 | if (!bundle->lacp) { | |
8c977421 | 1110 | ofproto->need_revalidate = true; |
abe529af BP |
1111 | bundle->lacp = lacp_create(); |
1112 | } | |
1113 | lacp_configure(bundle->lacp, s->lacp); | |
1114 | } else { | |
1115 | lacp_destroy(bundle->lacp); | |
1116 | bundle->lacp = NULL; | |
1117 | } | |
1118 | ||
1119 | /* Update set of ports. */ | |
1120 | ok = true; | |
1121 | for (i = 0; i < s->n_slaves; i++) { | |
1122 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1123 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1124 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1125 | ok = false; |
1126 | } | |
1127 | } | |
1128 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1129 | struct ofport_dpif *next_port; | |
1130 | ||
1131 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1132 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1133 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1134 | goto found; |
1135 | } | |
1136 | } | |
1137 | ||
1138 | bundle_del_port(port); | |
1139 | found: ; | |
1140 | } | |
1141 | } | |
1142 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1143 | ||
1144 | if (list_is_empty(&bundle->ports)) { | |
1145 | bundle_destroy(bundle); | |
1146 | return EINVAL; | |
1147 | } | |
1148 | ||
ecac4ebf BP |
1149 | /* Set VLAN tagging mode */ |
1150 | if (s->vlan_mode != bundle->vlan_mode) { | |
1151 | bundle->vlan_mode = s->vlan_mode; | |
1152 | need_flush = true; | |
1153 | } | |
1154 | ||
abe529af | 1155 | /* Set VLAN tag. */ |
ecac4ebf BP |
1156 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1157 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1158 | : 0); | |
1159 | if (vlan != bundle->vlan) { | |
1160 | bundle->vlan = vlan; | |
abe529af BP |
1161 | need_flush = true; |
1162 | } | |
1163 | ||
1164 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1165 | switch (s->vlan_mode) { |
1166 | case PORT_VLAN_ACCESS: | |
1167 | trunks = NULL; | |
1168 | break; | |
1169 | ||
1170 | case PORT_VLAN_TRUNK: | |
1171 | trunks = (unsigned long *) s->trunks; | |
1172 | break; | |
1173 | ||
1174 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1175 | case PORT_VLAN_NATIVE_TAGGED: | |
1176 | if (vlan != 0 && (!s->trunks | |
1177 | || !bitmap_is_set(s->trunks, vlan) | |
1178 | || bitmap_is_set(s->trunks, 0))) { | |
1179 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1180 | if (s->trunks) { | |
1181 | trunks = bitmap_clone(s->trunks, 4096); | |
1182 | } else { | |
1183 | trunks = bitmap_allocate1(4096); | |
1184 | } | |
1185 | bitmap_set1(trunks, vlan); | |
1186 | bitmap_set0(trunks, 0); | |
1187 | } else { | |
1188 | trunks = (unsigned long *) s->trunks; | |
1189 | } | |
1190 | break; | |
1191 | ||
1192 | default: | |
1193 | NOT_REACHED(); | |
1194 | } | |
abe529af BP |
1195 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
1196 | free(bundle->trunks); | |
ecac4ebf BP |
1197 | if (trunks == s->trunks) { |
1198 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1199 | } else { | |
1200 | bundle->trunks = trunks; | |
1201 | trunks = NULL; | |
1202 | } | |
abe529af BP |
1203 | need_flush = true; |
1204 | } | |
ecac4ebf BP |
1205 | if (trunks != s->trunks) { |
1206 | free(trunks); | |
1207 | } | |
abe529af BP |
1208 | |
1209 | /* Bonding. */ | |
1210 | if (!list_is_short(&bundle->ports)) { | |
1211 | bundle->ofproto->has_bonded_bundles = true; | |
1212 | if (bundle->bond) { | |
1213 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
1214 | ofproto->need_revalidate = true; | |
1215 | } | |
1216 | } else { | |
1217 | bundle->bond = bond_create(s->bond); | |
6f77f4ae | 1218 | ofproto->need_revalidate = true; |
abe529af BP |
1219 | } |
1220 | ||
1221 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 1222 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
1223 | port->up.netdev); |
1224 | } | |
1225 | } else { | |
1226 | bond_destroy(bundle->bond); | |
1227 | bundle->bond = NULL; | |
1228 | } | |
1229 | ||
1230 | /* If we changed something that would affect MAC learning, un-learn | |
1231 | * everything on this port and force flow revalidation. */ | |
1232 | if (need_flush) { | |
1233 | bundle_flush_macs(bundle); | |
1234 | } | |
1235 | ||
1236 | return 0; | |
1237 | } | |
1238 | ||
1239 | static void | |
1240 | bundle_remove(struct ofport *port_) | |
1241 | { | |
1242 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1243 | struct ofbundle *bundle = port->bundle; | |
1244 | ||
1245 | if (bundle) { | |
1246 | bundle_del_port(port); | |
1247 | if (list_is_empty(&bundle->ports)) { | |
1248 | bundle_destroy(bundle); | |
1249 | } else if (list_is_short(&bundle->ports)) { | |
1250 | bond_destroy(bundle->bond); | |
1251 | bundle->bond = NULL; | |
1252 | } | |
1253 | } | |
1254 | } | |
1255 | ||
1256 | static void | |
5f877369 | 1257 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
1258 | { |
1259 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
1260 | struct ofport_dpif *port = port_; | |
1261 | uint8_t ea[ETH_ADDR_LEN]; | |
1262 | int error; | |
1263 | ||
1264 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
1265 | if (!error) { | |
abe529af | 1266 | struct ofpbuf packet; |
5f877369 | 1267 | void *packet_pdu; |
abe529af BP |
1268 | |
1269 | ofpbuf_init(&packet, 0); | |
1270 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
1271 | pdu_size); |
1272 | memcpy(packet_pdu, pdu, pdu_size); | |
1273 | ||
abe529af BP |
1274 | error = netdev_send(port->up.netdev, &packet); |
1275 | if (error) { | |
1276 | VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed " | |
1277 | "(%s)", port->bundle->name, | |
1278 | netdev_get_name(port->up.netdev), strerror(error)); | |
1279 | } | |
1280 | ofpbuf_uninit(&packet); | |
1281 | } else { | |
1282 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
1283 | "%s (%s)", port->bundle->name, | |
1284 | netdev_get_name(port->up.netdev), strerror(error)); | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | static void | |
1289 | bundle_send_learning_packets(struct ofbundle *bundle) | |
1290 | { | |
1291 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1292 | int error, n_packets, n_errors; | |
1293 | struct mac_entry *e; | |
1294 | ||
1295 | error = n_packets = n_errors = 0; | |
1296 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
1297 | if (e->port.p != bundle) { | |
1298 | int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan); | |
1299 | if (ret) { | |
1300 | error = ret; | |
1301 | n_errors++; | |
1302 | } | |
1303 | n_packets++; | |
1304 | } | |
1305 | } | |
1306 | ||
1307 | if (n_errors) { | |
1308 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1309 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
1310 | "packets, last error was: %s", | |
1311 | bundle->name, n_errors, n_packets, strerror(error)); | |
1312 | } else { | |
1313 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
1314 | bundle->name, n_packets); | |
1315 | } | |
1316 | } | |
1317 | ||
1318 | static void | |
1319 | bundle_run(struct ofbundle *bundle) | |
1320 | { | |
1321 | if (bundle->lacp) { | |
1322 | lacp_run(bundle->lacp, send_pdu_cb); | |
1323 | } | |
1324 | if (bundle->bond) { | |
1325 | struct ofport_dpif *port; | |
1326 | ||
1327 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 1328 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
1329 | } |
1330 | ||
1331 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
1332 | lacp_negotiated(bundle->lacp)); | |
1333 | if (bond_should_send_learning_packets(bundle->bond)) { | |
1334 | bundle_send_learning_packets(bundle); | |
1335 | } | |
1336 | } | |
1337 | } | |
1338 | ||
1339 | static void | |
1340 | bundle_wait(struct ofbundle *bundle) | |
1341 | { | |
1342 | if (bundle->lacp) { | |
1343 | lacp_wait(bundle->lacp); | |
1344 | } | |
1345 | if (bundle->bond) { | |
1346 | bond_wait(bundle->bond); | |
1347 | } | |
1348 | } | |
1349 | \f | |
1350 | /* Mirrors. */ | |
1351 | ||
1352 | static int | |
1353 | mirror_scan(struct ofproto_dpif *ofproto) | |
1354 | { | |
1355 | int idx; | |
1356 | ||
1357 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
1358 | if (!ofproto->mirrors[idx]) { | |
1359 | return idx; | |
1360 | } | |
1361 | } | |
1362 | return -1; | |
1363 | } | |
1364 | ||
1365 | static struct ofmirror * | |
1366 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
1367 | { | |
1368 | int i; | |
1369 | ||
1370 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1371 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
1372 | if (mirror && mirror->aux == aux) { | |
1373 | return mirror; | |
1374 | } | |
1375 | } | |
1376 | ||
1377 | return NULL; | |
1378 | } | |
1379 | ||
1380 | static int | |
1381 | mirror_set(struct ofproto *ofproto_, void *aux, | |
1382 | const struct ofproto_mirror_settings *s) | |
1383 | { | |
1384 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1385 | mirror_mask_t mirror_bit; | |
1386 | struct ofbundle *bundle; | |
1387 | struct ofmirror *mirror; | |
1388 | struct ofbundle *out; | |
1389 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
1390 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
1391 | int out_vlan; | |
1392 | ||
1393 | mirror = mirror_lookup(ofproto, aux); | |
1394 | if (!s) { | |
1395 | mirror_destroy(mirror); | |
1396 | return 0; | |
1397 | } | |
1398 | if (!mirror) { | |
1399 | int idx; | |
1400 | ||
1401 | idx = mirror_scan(ofproto); | |
1402 | if (idx < 0) { | |
1403 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
1404 | "cannot create %s", | |
1405 | ofproto->up.name, MAX_MIRRORS, s->name); | |
1406 | return EFBIG; | |
1407 | } | |
1408 | ||
1409 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
1410 | mirror->ofproto = ofproto; | |
1411 | mirror->idx = idx; | |
8b28d864 | 1412 | mirror->aux = aux; |
abe529af BP |
1413 | mirror->out_vlan = -1; |
1414 | mirror->name = NULL; | |
1415 | } | |
1416 | ||
1417 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
1418 | free(mirror->name); | |
1419 | mirror->name = xstrdup(s->name); | |
1420 | } | |
1421 | ||
1422 | /* Get the new configuration. */ | |
1423 | if (s->out_bundle) { | |
1424 | out = bundle_lookup(ofproto, s->out_bundle); | |
1425 | if (!out) { | |
1426 | mirror_destroy(mirror); | |
1427 | return EINVAL; | |
1428 | } | |
1429 | out_vlan = -1; | |
1430 | } else { | |
1431 | out = NULL; | |
1432 | out_vlan = s->out_vlan; | |
1433 | } | |
1434 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
1435 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
1436 | ||
1437 | /* If the configuration has not changed, do nothing. */ | |
1438 | if (hmapx_equals(&srcs, &mirror->srcs) | |
1439 | && hmapx_equals(&dsts, &mirror->dsts) | |
1440 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
1441 | && mirror->out == out | |
1442 | && mirror->out_vlan == out_vlan) | |
1443 | { | |
1444 | hmapx_destroy(&srcs); | |
1445 | hmapx_destroy(&dsts); | |
1446 | return 0; | |
1447 | } | |
1448 | ||
1449 | hmapx_swap(&srcs, &mirror->srcs); | |
1450 | hmapx_destroy(&srcs); | |
1451 | ||
1452 | hmapx_swap(&dsts, &mirror->dsts); | |
1453 | hmapx_destroy(&dsts); | |
1454 | ||
1455 | free(mirror->vlans); | |
1456 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
1457 | ||
1458 | mirror->out = out; | |
1459 | mirror->out_vlan = out_vlan; | |
1460 | ||
1461 | /* Update bundles. */ | |
1462 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1463 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
1464 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
1465 | bundle->src_mirrors |= mirror_bit; | |
1466 | } else { | |
1467 | bundle->src_mirrors &= ~mirror_bit; | |
1468 | } | |
1469 | ||
1470 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
1471 | bundle->dst_mirrors |= mirror_bit; | |
1472 | } else { | |
1473 | bundle->dst_mirrors &= ~mirror_bit; | |
1474 | } | |
1475 | ||
1476 | if (mirror->out == bundle) { | |
1477 | bundle->mirror_out |= mirror_bit; | |
1478 | } else { | |
1479 | bundle->mirror_out &= ~mirror_bit; | |
1480 | } | |
1481 | } | |
1482 | ||
1483 | ofproto->need_revalidate = true; | |
1484 | mac_learning_flush(ofproto->ml); | |
1485 | ||
1486 | return 0; | |
1487 | } | |
1488 | ||
1489 | static void | |
1490 | mirror_destroy(struct ofmirror *mirror) | |
1491 | { | |
1492 | struct ofproto_dpif *ofproto; | |
1493 | mirror_mask_t mirror_bit; | |
1494 | struct ofbundle *bundle; | |
1495 | ||
1496 | if (!mirror) { | |
1497 | return; | |
1498 | } | |
1499 | ||
1500 | ofproto = mirror->ofproto; | |
1501 | ofproto->need_revalidate = true; | |
1502 | mac_learning_flush(ofproto->ml); | |
1503 | ||
1504 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1505 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1506 | bundle->src_mirrors &= ~mirror_bit; | |
1507 | bundle->dst_mirrors &= ~mirror_bit; | |
1508 | bundle->mirror_out &= ~mirror_bit; | |
1509 | } | |
1510 | ||
1511 | hmapx_destroy(&mirror->srcs); | |
1512 | hmapx_destroy(&mirror->dsts); | |
1513 | free(mirror->vlans); | |
1514 | ||
1515 | ofproto->mirrors[mirror->idx] = NULL; | |
1516 | free(mirror->name); | |
1517 | free(mirror); | |
1518 | } | |
1519 | ||
1520 | static int | |
1521 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
1522 | { | |
1523 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1524 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
1525 | ofproto->need_revalidate = true; | |
1526 | mac_learning_flush(ofproto->ml); | |
1527 | } | |
1528 | return 0; | |
1529 | } | |
1530 | ||
1531 | static bool | |
1532 | is_mirror_output_bundle(struct ofproto *ofproto_, void *aux) | |
1533 | { | |
1534 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1535 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
1536 | return bundle && bundle->mirror_out != 0; | |
1537 | } | |
8402c74b SS |
1538 | |
1539 | static void | |
b53055f4 | 1540 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
1541 | { |
1542 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1543 | /* Revalidate cached flows whenever forward_bpdu option changes. */ | |
1544 | ofproto->need_revalidate = true; | |
1545 | } | |
abe529af BP |
1546 | \f |
1547 | /* Ports. */ | |
1548 | ||
1549 | static struct ofport_dpif * | |
1550 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
1551 | { | |
7df6a8bd BP |
1552 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
1553 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
1554 | } |
1555 | ||
1556 | static struct ofport_dpif * | |
1557 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
1558 | { | |
1559 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
1560 | } | |
1561 | ||
1562 | static void | |
1563 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
1564 | struct dpif_port *dpif_port) | |
1565 | { | |
1566 | ofproto_port->name = dpif_port->name; | |
1567 | ofproto_port->type = dpif_port->type; | |
1568 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
1569 | } | |
1570 | ||
1571 | static void | |
1572 | port_run(struct ofport_dpif *ofport) | |
1573 | { | |
015e08bc EJ |
1574 | bool enable = netdev_get_carrier(ofport->up.netdev); |
1575 | ||
abe529af BP |
1576 | if (ofport->cfm) { |
1577 | cfm_run(ofport->cfm); | |
1578 | ||
1579 | if (cfm_should_send_ccm(ofport->cfm)) { | |
1580 | struct ofpbuf packet; | |
abe529af BP |
1581 | |
1582 | ofpbuf_init(&packet, 0); | |
c0a2e71d | 1583 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr); |
abe529af | 1584 | send_packet(ofproto_dpif_cast(ofport->up.ofproto), |
b2fda3ef | 1585 | ofport->odp_port, &packet); |
abe529af BP |
1586 | ofpbuf_uninit(&packet); |
1587 | } | |
015e08bc | 1588 | |
86dc6501 EJ |
1589 | enable = enable && !cfm_get_fault(ofport->cfm) |
1590 | && cfm_get_opup(ofport->cfm); | |
abe529af | 1591 | } |
015e08bc EJ |
1592 | |
1593 | if (ofport->bundle) { | |
1594 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
1595 | } | |
1596 | ||
daff3353 EJ |
1597 | if (ofport->may_enable != enable) { |
1598 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1599 | ||
1600 | if (ofproto->has_bundle_action) { | |
1601 | ofproto->need_revalidate = true; | |
1602 | } | |
1603 | } | |
1604 | ||
015e08bc | 1605 | ofport->may_enable = enable; |
abe529af BP |
1606 | } |
1607 | ||
1608 | static void | |
1609 | port_wait(struct ofport_dpif *ofport) | |
1610 | { | |
1611 | if (ofport->cfm) { | |
1612 | cfm_wait(ofport->cfm); | |
1613 | } | |
1614 | } | |
1615 | ||
1616 | static int | |
1617 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
1618 | struct ofproto_port *ofproto_port) | |
1619 | { | |
1620 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1621 | struct dpif_port dpif_port; | |
1622 | int error; | |
1623 | ||
1624 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
1625 | if (!error) { | |
1626 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
1627 | } | |
1628 | return error; | |
1629 | } | |
1630 | ||
1631 | static int | |
1632 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
1633 | { | |
1634 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1635 | uint16_t odp_port; | |
1636 | int error; | |
1637 | ||
1638 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
1639 | if (!error) { | |
1640 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
1641 | } | |
1642 | return error; | |
1643 | } | |
1644 | ||
1645 | static int | |
1646 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
1647 | { | |
1648 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1649 | int error; | |
1650 | ||
1651 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
1652 | if (!error) { | |
1653 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
1654 | if (ofport) { | |
1655 | /* The caller is going to close ofport->up.netdev. If this is a | |
1656 | * bonded port, then the bond is using that netdev, so remove it | |
1657 | * from the bond. The client will need to reconfigure everything | |
1658 | * after deleting ports, so then the slave will get re-added. */ | |
1659 | bundle_remove(&ofport->up); | |
1660 | } | |
1661 | } | |
1662 | return error; | |
1663 | } | |
1664 | ||
1665 | struct port_dump_state { | |
1666 | struct dpif_port_dump dump; | |
1667 | bool done; | |
1668 | }; | |
1669 | ||
1670 | static int | |
1671 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
1672 | { | |
1673 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1674 | struct port_dump_state *state; | |
1675 | ||
1676 | *statep = state = xmalloc(sizeof *state); | |
1677 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
1678 | state->done = false; | |
1679 | return 0; | |
1680 | } | |
1681 | ||
1682 | static int | |
1683 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
1684 | struct ofproto_port *port) | |
1685 | { | |
1686 | struct port_dump_state *state = state_; | |
1687 | struct dpif_port dpif_port; | |
1688 | ||
1689 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
1690 | ofproto_port_from_dpif_port(port, &dpif_port); | |
1691 | return 0; | |
1692 | } else { | |
1693 | int error = dpif_port_dump_done(&state->dump); | |
1694 | state->done = true; | |
1695 | return error ? error : EOF; | |
1696 | } | |
1697 | } | |
1698 | ||
1699 | static int | |
1700 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
1701 | { | |
1702 | struct port_dump_state *state = state_; | |
1703 | ||
1704 | if (!state->done) { | |
1705 | dpif_port_dump_done(&state->dump); | |
1706 | } | |
1707 | free(state); | |
1708 | return 0; | |
1709 | } | |
1710 | ||
1711 | static int | |
1712 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
1713 | { | |
1714 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1715 | return dpif_port_poll(ofproto->dpif, devnamep); | |
1716 | } | |
1717 | ||
1718 | static void | |
1719 | port_poll_wait(const struct ofproto *ofproto_) | |
1720 | { | |
1721 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1722 | dpif_port_poll_wait(ofproto->dpif); | |
1723 | } | |
1724 | ||
1725 | static int | |
1726 | port_is_lacp_current(const struct ofport *ofport_) | |
1727 | { | |
1728 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1729 | return (ofport->bundle && ofport->bundle->lacp | |
1730 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
1731 | : -1); | |
1732 | } | |
1733 | \f | |
1734 | /* Upcall handling. */ | |
1735 | ||
501f8d1f BP |
1736 | /* Flow miss batching. |
1737 | * | |
1738 | * Some dpifs implement operations faster when you hand them off in a batch. | |
1739 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
1740 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
1741 | * more packets, plus possibly installing the flow in the dpif. | |
1742 | * | |
1743 | * So far we only batch the operations that affect flow setup time the most. | |
1744 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
1745 | struct flow_miss { | |
1746 | struct hmap_node hmap_node; | |
1747 | struct flow flow; | |
1748 | const struct nlattr *key; | |
1749 | size_t key_len; | |
1750 | struct list packets; | |
1751 | }; | |
1752 | ||
1753 | struct flow_miss_op { | |
1754 | union dpif_op dpif_op; | |
1755 | struct facet *facet; | |
1756 | }; | |
1757 | ||
62cd7072 BP |
1758 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
1759 | * OpenFlow controller as necessary according to their individual | |
1760 | * configurations. | |
1761 | * | |
1762 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, | |
1763 | * ownership is transferred to this function. */ | |
1764 | static void | |
1765 | send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet, | |
1766 | const struct flow *flow, bool clone) | |
1767 | { | |
1768 | struct ofputil_packet_in pin; | |
1769 | ||
1770 | pin.packet = packet; | |
1771 | pin.in_port = flow->in_port; | |
1772 | pin.reason = OFPR_NO_MATCH; | |
1773 | pin.buffer_id = 0; /* not yet known */ | |
1774 | pin.send_len = 0; /* not used for flow table misses */ | |
1775 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, | |
1776 | clone ? NULL : packet); | |
1777 | } | |
1778 | ||
1779 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each | |
1780 | * OpenFlow controller as necessary according to their individual | |
1781 | * configurations. | |
1782 | * | |
1783 | * 'send_len' should be the number of bytes of 'packet' to send to the | |
1784 | * controller, as specified in the action that caused the packet to be sent. | |
abe529af BP |
1785 | * |
1786 | * If 'clone' is true, the caller retains ownership of 'upcall->packet'. | |
1787 | * Otherwise, ownership is transferred to this function. */ | |
1788 | static void | |
62cd7072 BP |
1789 | send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet, |
1790 | uint64_t userdata, const struct flow *flow, bool clone) | |
abe529af BP |
1791 | { |
1792 | struct ofputil_packet_in pin; | |
6ff686f2 | 1793 | struct user_action_cookie cookie; |
abe529af | 1794 | |
62cd7072 BP |
1795 | memcpy(&cookie, &userdata, sizeof(cookie)); |
1796 | ||
1797 | pin.packet = packet; | |
abe529af | 1798 | pin.in_port = flow->in_port; |
62cd7072 | 1799 | pin.reason = OFPR_ACTION; |
abe529af | 1800 | pin.buffer_id = 0; /* not yet known */ |
6ff686f2 | 1801 | pin.send_len = cookie.data; |
78bd1cd0 | 1802 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, |
62cd7072 | 1803 | clone ? NULL : packet); |
abe529af BP |
1804 | } |
1805 | ||
1806 | static bool | |
1807 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
1808 | const struct ofpbuf *packet) | |
1809 | { | |
b6e001b6 EJ |
1810 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
1811 | ||
1812 | if (!ofport) { | |
1813 | return false; | |
1814 | } | |
1815 | ||
ef9819b5 | 1816 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 1817 | if (packet) { |
abe529af BP |
1818 | cfm_process_heartbeat(ofport->cfm, packet); |
1819 | } | |
1820 | return true; | |
b6e001b6 EJ |
1821 | } else if (ofport->bundle && ofport->bundle->lacp |
1822 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
1823 | if (packet) { | |
1824 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 1825 | } |
da37ebac | 1826 | return true; |
abe529af BP |
1827 | } |
1828 | return false; | |
1829 | } | |
1830 | ||
501f8d1f BP |
1831 | static struct flow_miss * |
1832 | flow_miss_create(struct hmap *todo, const struct flow *flow, | |
1833 | const struct nlattr *key, size_t key_len) | |
abe529af | 1834 | { |
501f8d1f BP |
1835 | uint32_t hash = flow_hash(flow, 0); |
1836 | struct flow_miss *miss; | |
abe529af | 1837 | |
501f8d1f BP |
1838 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
1839 | if (flow_equal(&miss->flow, flow)) { | |
1840 | return miss; | |
1841 | } | |
1842 | } | |
abe529af | 1843 | |
501f8d1f BP |
1844 | miss = xmalloc(sizeof *miss); |
1845 | hmap_insert(todo, &miss->hmap_node, hash); | |
1846 | miss->flow = *flow; | |
1847 | miss->key = key; | |
1848 | miss->key_len = key_len; | |
1849 | list_init(&miss->packets); | |
1850 | return miss; | |
1851 | } | |
abe529af | 1852 | |
501f8d1f BP |
1853 | static void |
1854 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
1855 | struct flow_miss_op *ops, size_t *n_ops) | |
1856 | { | |
1857 | const struct flow *flow = &miss->flow; | |
1858 | struct ofpbuf *packet, *next_packet; | |
1859 | struct facet *facet; | |
abe529af | 1860 | |
501f8d1f | 1861 | facet = facet_lookup_valid(ofproto, flow); |
abe529af | 1862 | if (!facet) { |
501f8d1f BP |
1863 | struct rule_dpif *rule; |
1864 | ||
1865 | rule = rule_dpif_lookup(ofproto, flow, 0); | |
abe529af BP |
1866 | if (!rule) { |
1867 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
501f8d1f | 1868 | struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); |
abe529af BP |
1869 | if (port) { |
1870 | if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) { | |
1871 | COVERAGE_INC(ofproto_dpif_no_packet_in); | |
1872 | /* XXX install 'drop' flow entry */ | |
abe529af BP |
1873 | return; |
1874 | } | |
1875 | } else { | |
1876 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
501f8d1f BP |
1877 | flow->in_port); |
1878 | } | |
1879 | ||
1880 | LIST_FOR_EACH_SAFE (packet, next_packet, list_node, | |
1881 | &miss->packets) { | |
1882 | list_remove(&packet->list_node); | |
1883 | send_packet_in_miss(ofproto, packet, flow, false); | |
abe529af BP |
1884 | } |
1885 | ||
abe529af BP |
1886 | return; |
1887 | } | |
1888 | ||
501f8d1f | 1889 | facet = facet_create(rule, flow); |
abe529af BP |
1890 | } |
1891 | ||
501f8d1f BP |
1892 | LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) { |
1893 | list_remove(&packet->list_node); | |
1894 | ofproto->n_matches++; | |
1895 | ||
1896 | if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
1897 | /* | |
1898 | * Extra-special case for fail-open mode. | |
1899 | * | |
1900 | * We are in fail-open mode and the packet matched the fail-open | |
1901 | * rule, but we are connected to a controller too. We should send | |
1902 | * the packet up to the controller in the hope that it will try to | |
1903 | * set up a flow and thereby allow us to exit fail-open. | |
1904 | * | |
1905 | * See the top-level comment in fail-open.c for more information. | |
1906 | */ | |
1907 | send_packet_in_miss(ofproto, packet, flow, true); | |
1908 | } | |
1909 | ||
1910 | if (!facet->may_install) { | |
1911 | facet_make_actions(ofproto, facet, packet); | |
1912 | } | |
1913 | if (!execute_controller_action(ofproto, &facet->flow, | |
1914 | facet->actions, facet->actions_len, | |
1915 | packet)) { | |
1916 | struct flow_miss_op *op = &ops[(*n_ops)++]; | |
1917 | struct dpif_execute *execute = &op->dpif_op.execute; | |
1918 | ||
1919 | op->facet = facet; | |
1920 | execute->type = DPIF_OP_EXECUTE; | |
1921 | execute->key = miss->key; | |
1922 | execute->key_len = miss->key_len; | |
1923 | execute->actions | |
1924 | = (facet->may_install | |
1925 | ? facet->actions | |
1926 | : xmemdup(facet->actions, facet->actions_len)); | |
1927 | execute->actions_len = facet->actions_len; | |
1928 | execute->packet = packet; | |
1929 | } | |
1930 | } | |
1931 | ||
1932 | if (facet->may_install) { | |
1933 | struct flow_miss_op *op = &ops[(*n_ops)++]; | |
1934 | struct dpif_flow_put *put = &op->dpif_op.flow_put; | |
1935 | ||
1936 | op->facet = facet; | |
1937 | put->type = DPIF_OP_FLOW_PUT; | |
1938 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
1939 | put->key = miss->key; | |
1940 | put->key_len = miss->key_len; | |
1941 | put->actions = facet->actions; | |
1942 | put->actions_len = facet->actions_len; | |
1943 | put->stats = NULL; | |
1944 | } | |
1945 | } | |
1946 | ||
1947 | static void | |
1948 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
1949 | size_t n_upcalls) | |
1950 | { | |
1951 | struct dpif_upcall *upcall; | |
1952 | struct flow_miss *miss, *next_miss; | |
1953 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; | |
1954 | union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; | |
1955 | struct hmap todo; | |
1956 | size_t n_ops; | |
1957 | size_t i; | |
1958 | ||
1959 | if (!n_upcalls) { | |
1960 | return; | |
1961 | } | |
1962 | ||
1963 | /* Construct the to-do list. | |
1964 | * | |
1965 | * This just amounts to extracting the flow from each packet and sticking | |
1966 | * the packets that have the same flow in the same "flow_miss" structure so | |
1967 | * that we can process them together. */ | |
1968 | hmap_init(&todo); | |
1969 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { | |
1970 | struct flow_miss *miss; | |
1971 | struct flow flow; | |
1972 | ||
1973 | /* Obtain in_port and tun_id, at least, then set 'flow''s header | |
1974 | * pointers. */ | |
1975 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
1976 | flow_extract(upcall->packet, flow.tun_id, flow.in_port, &flow); | |
1977 | ||
1978 | /* Handle 802.1ag and LACP specially. */ | |
1979 | if (process_special(ofproto, &flow, upcall->packet)) { | |
1980 | ofpbuf_delete(upcall->packet); | |
1981 | ofproto->n_matches++; | |
1982 | continue; | |
1983 | } | |
1984 | ||
1985 | /* Add other packets to a to-do list. */ | |
1986 | miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len); | |
1987 | list_push_back(&miss->packets, &upcall->packet->list_node); | |
1988 | } | |
1989 | ||
1990 | /* Process each element in the to-do list, constructing the set of | |
1991 | * operations to batch. */ | |
1992 | n_ops = 0; | |
1993 | HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) { | |
1994 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); | |
1995 | ofpbuf_list_delete(&miss->packets); | |
1996 | hmap_remove(&todo, &miss->hmap_node); | |
1997 | free(miss); | |
abe529af | 1998 | } |
501f8d1f BP |
1999 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
2000 | hmap_destroy(&todo); | |
2001 | ||
2002 | /* Execute batch. */ | |
2003 | for (i = 0; i < n_ops; i++) { | |
2004 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
2005 | } | |
2006 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
2007 | ||
2008 | /* Free memory and update facets. */ | |
2009 | for (i = 0; i < n_ops; i++) { | |
2010 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
2011 | struct dpif_execute *execute; | |
2012 | struct dpif_flow_put *put; | |
2013 | ||
2014 | switch (op->dpif_op.type) { | |
2015 | case DPIF_OP_EXECUTE: | |
2016 | execute = &op->dpif_op.execute; | |
2017 | if (op->facet->actions != execute->actions) { | |
2018 | free((struct nlattr *) execute->actions); | |
2019 | } | |
2020 | ofpbuf_delete((struct ofpbuf *) execute->packet); | |
2021 | break; | |
abe529af | 2022 | |
501f8d1f BP |
2023 | case DPIF_OP_FLOW_PUT: |
2024 | put = &op->dpif_op.flow_put; | |
2025 | if (!put->error) { | |
2026 | op->facet->installed = true; | |
2027 | } | |
2028 | break; | |
2029 | } | |
2030 | } | |
abe529af BP |
2031 | } |
2032 | ||
2033 | static void | |
6ff686f2 PS |
2034 | handle_userspace_upcall(struct ofproto_dpif *ofproto, |
2035 | struct dpif_upcall *upcall) | |
abe529af BP |
2036 | { |
2037 | struct flow flow; | |
6ff686f2 | 2038 | struct user_action_cookie cookie; |
abe529af | 2039 | |
6ff686f2 | 2040 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
abe529af | 2041 | |
6ff686f2 | 2042 | if (cookie.type == USER_ACTION_COOKIE_SFLOW) { |
abe529af BP |
2043 | if (ofproto->sflow) { |
2044 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
6ff686f2 | 2045 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie); |
abe529af BP |
2046 | } |
2047 | ofpbuf_delete(upcall->packet); | |
6ff686f2 PS |
2048 | |
2049 | } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) { | |
2050 | COVERAGE_INC(ofproto_dpif_ctlr_action); | |
2051 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
62cd7072 BP |
2052 | send_packet_in_action(ofproto, upcall->packet, upcall->userdata, |
2053 | &flow, false); | |
6ff686f2 PS |
2054 | } else { |
2055 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
2056 | } | |
2057 | } | |
2058 | ||
2059 | static void | |
2060 | handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) | |
2061 | { | |
2062 | switch (upcall->type) { | |
2063 | case DPIF_UC_ACTION: | |
2064 | handle_userspace_upcall(ofproto, upcall); | |
abe529af BP |
2065 | break; |
2066 | ||
2067 | case DPIF_UC_MISS: | |
501f8d1f BP |
2068 | /* The caller handles these. */ |
2069 | NOT_REACHED(); | |
abe529af BP |
2070 | |
2071 | case DPIF_N_UC_TYPES: | |
2072 | default: | |
2073 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
2074 | break; | |
2075 | } | |
2076 | } | |
2077 | \f | |
2078 | /* Flow expiration. */ | |
2079 | ||
2080 | static int facet_max_idle(const struct ofproto_dpif *); | |
2081 | static void update_stats(struct ofproto_dpif *); | |
2082 | static void rule_expire(struct rule_dpif *); | |
2083 | static void expire_facets(struct ofproto_dpif *, int dp_max_idle); | |
2084 | ||
2085 | /* This function is called periodically by run(). Its job is to collect | |
2086 | * updates for the flows that have been installed into the datapath, most | |
2087 | * importantly when they last were used, and then use that information to | |
2088 | * expire flows that have not been used recently. | |
2089 | * | |
2090 | * Returns the number of milliseconds after which it should be called again. */ | |
2091 | static int | |
2092 | expire(struct ofproto_dpif *ofproto) | |
2093 | { | |
2094 | struct rule_dpif *rule, *next_rule; | |
0697b5c3 | 2095 | struct classifier *table; |
abe529af BP |
2096 | int dp_max_idle; |
2097 | ||
2098 | /* Update stats for each flow in the datapath. */ | |
2099 | update_stats(ofproto); | |
2100 | ||
2101 | /* Expire facets that have been idle too long. */ | |
2102 | dp_max_idle = facet_max_idle(ofproto); | |
2103 | expire_facets(ofproto, dp_max_idle); | |
2104 | ||
2105 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
2106 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
2107 | struct cls_cursor cursor; | |
2108 | ||
2109 | cls_cursor_init(&cursor, table, NULL); | |
2110 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
2111 | rule_expire(rule); | |
2112 | } | |
abe529af BP |
2113 | } |
2114 | ||
2115 | /* All outstanding data in existing flows has been accounted, so it's a | |
2116 | * good time to do bond rebalancing. */ | |
2117 | if (ofproto->has_bonded_bundles) { | |
2118 | struct ofbundle *bundle; | |
2119 | ||
2120 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2121 | if (bundle->bond) { | |
2122 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
2123 | } | |
2124 | } | |
2125 | } | |
2126 | ||
2127 | return MIN(dp_max_idle, 1000); | |
2128 | } | |
2129 | ||
2130 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
2131 | * | |
2132 | * This function also pushes statistics updates to rules which each facet | |
2133 | * resubmits into. Generally these statistics will be accurate. However, if a | |
2134 | * facet changes the rule it resubmits into at some time in between | |
2135 | * update_stats() runs, it is possible that statistics accrued to the | |
2136 | * old rule will be incorrectly attributed to the new rule. This could be | |
2137 | * avoided by calling update_stats() whenever rules are created or | |
2138 | * deleted. However, the performance impact of making so many calls to the | |
2139 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
2140 | */ | |
2141 | static void | |
2142 | update_stats(struct ofproto_dpif *p) | |
2143 | { | |
2144 | const struct dpif_flow_stats *stats; | |
2145 | struct dpif_flow_dump dump; | |
2146 | const struct nlattr *key; | |
2147 | size_t key_len; | |
2148 | ||
2149 | dpif_flow_dump_start(&dump, p->dpif); | |
2150 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
2151 | struct facet *facet; | |
2152 | struct flow flow; | |
2153 | ||
2154 | if (odp_flow_key_to_flow(key, key_len, &flow)) { | |
2155 | struct ds s; | |
2156 | ||
2157 | ds_init(&s); | |
2158 | odp_flow_key_format(key, key_len, &s); | |
df2c07f4 | 2159 | VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s", |
abe529af BP |
2160 | ds_cstr(&s)); |
2161 | ds_destroy(&s); | |
2162 | ||
2163 | continue; | |
2164 | } | |
2165 | facet = facet_find(p, &flow); | |
2166 | ||
2167 | if (facet && facet->installed) { | |
2168 | ||
2169 | if (stats->n_packets >= facet->dp_packet_count) { | |
2170 | uint64_t extra = stats->n_packets - facet->dp_packet_count; | |
2171 | facet->packet_count += extra; | |
2172 | } else { | |
2173 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
2174 | } | |
2175 | ||
2176 | if (stats->n_bytes >= facet->dp_byte_count) { | |
2177 | facet->byte_count += stats->n_bytes - facet->dp_byte_count; | |
2178 | } else { | |
2179 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
2180 | } | |
2181 | ||
2182 | facet->dp_packet_count = stats->n_packets; | |
2183 | facet->dp_byte_count = stats->n_bytes; | |
2184 | ||
2185 | facet_update_time(p, facet, stats->used); | |
55af77bb | 2186 | facet_account(p, facet); |
abe529af BP |
2187 | facet_push_stats(facet); |
2188 | } else { | |
2189 | /* There's a flow in the datapath that we know nothing about. | |
2190 | * Delete it. */ | |
2191 | COVERAGE_INC(facet_unexpected); | |
2192 | dpif_flow_del(p->dpif, key, key_len, NULL); | |
2193 | } | |
2194 | } | |
2195 | dpif_flow_dump_done(&dump); | |
2196 | } | |
2197 | ||
2198 | /* Calculates and returns the number of milliseconds of idle time after which | |
2199 | * facets should expire from the datapath and we should fold their statistics | |
2200 | * into their parent rules in userspace. */ | |
2201 | static int | |
2202 | facet_max_idle(const struct ofproto_dpif *ofproto) | |
2203 | { | |
2204 | /* | |
2205 | * Idle time histogram. | |
2206 | * | |
2207 | * Most of the time a switch has a relatively small number of facets. When | |
2208 | * this is the case we might as well keep statistics for all of them in | |
2209 | * userspace and to cache them in the kernel datapath for performance as | |
2210 | * well. | |
2211 | * | |
2212 | * As the number of facets increases, the memory required to maintain | |
2213 | * statistics about them in userspace and in the kernel becomes | |
2214 | * significant. However, with a large number of facets it is likely that | |
2215 | * only a few of them are "heavy hitters" that consume a large amount of | |
2216 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
2217 | * kernel and maintaining in userspaces; other facets we can discard. | |
2218 | * | |
2219 | * The technique used to compute the idle time is to build a histogram with | |
2220 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet | |
2221 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
2222 | * After the histogram has been built, we compute the cutoff so that only | |
084f5290 SH |
2223 | * the most-recently-used 1% of facets (but at least |
2224 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least | |
2225 | * the most-recently-used bucket of facets is kept, so actually an | |
2226 | * arbitrary number of facets can be kept in any given expiration run | |
2227 | * (though the next run will delete most of those unless they receive | |
2228 | * additional data). | |
abe529af BP |
2229 | * |
2230 | * This requires a second pass through the facets, in addition to the pass | |
2231 | * made by update_stats(), because the former function never looks | |
2232 | * at uninstallable facets. | |
2233 | */ | |
2234 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
2235 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
2236 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 2237 | int total, subtotal, bucket; |
abe529af | 2238 | struct facet *facet; |
abe529af BP |
2239 | long long int now; |
2240 | int i; | |
2241 | ||
2242 | total = hmap_count(&ofproto->facets); | |
084f5290 | 2243 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
2244 | return N_BUCKETS * BUCKET_WIDTH; |
2245 | } | |
2246 | ||
2247 | /* Build histogram. */ | |
2248 | now = time_msec(); | |
2249 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
2250 | long long int idle = now - facet->used; | |
2251 | int bucket = (idle <= 0 ? 0 | |
2252 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
2253 | : (unsigned int) idle / BUCKET_WIDTH); | |
2254 | buckets[bucket]++; | |
2255 | } | |
2256 | ||
2257 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
2258 | subtotal = bucket = 0; |
2259 | do { | |
2260 | subtotal += buckets[bucket++]; | |
084f5290 SH |
2261 | } while (bucket < N_BUCKETS && |
2262 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
2263 | |
2264 | if (VLOG_IS_DBG_ENABLED()) { | |
2265 | struct ds s; | |
2266 | ||
2267 | ds_init(&s); | |
2268 | ds_put_cstr(&s, "keep"); | |
2269 | for (i = 0; i < N_BUCKETS; i++) { | |
2270 | if (i == bucket) { | |
2271 | ds_put_cstr(&s, ", drop"); | |
2272 | } | |
2273 | if (buckets[i]) { | |
2274 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
2275 | } | |
2276 | } | |
2277 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
2278 | ds_destroy(&s); | |
2279 | } | |
2280 | ||
2281 | return bucket * BUCKET_WIDTH; | |
2282 | } | |
2283 | ||
2284 | static void | |
2285 | facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
2286 | { | |
2287 | if (ofproto->netflow && !facet_is_controller_flow(facet) && | |
2288 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
2289 | struct ofexpired expired; | |
2290 | ||
2291 | if (facet->installed) { | |
2292 | struct dpif_flow_stats stats; | |
2293 | ||
2294 | facet_put__(ofproto, facet, facet->actions, facet->actions_len, | |
2295 | &stats); | |
2296 | facet_update_stats(ofproto, facet, &stats); | |
2297 | } | |
2298 | ||
2299 | expired.flow = facet->flow; | |
2300 | expired.packet_count = facet->packet_count; | |
2301 | expired.byte_count = facet->byte_count; | |
2302 | expired.used = facet->used; | |
2303 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2304 | } | |
2305 | } | |
2306 | ||
2307 | static void | |
2308 | expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle) | |
2309 | { | |
2310 | long long int cutoff = time_msec() - dp_max_idle; | |
2311 | struct facet *facet, *next_facet; | |
2312 | ||
2313 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
2314 | facet_active_timeout(ofproto, facet); | |
2315 | if (facet->used < cutoff) { | |
2316 | facet_remove(ofproto, facet); | |
2317 | } | |
2318 | } | |
2319 | } | |
2320 | ||
2321 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
2322 | * then delete it entirely. */ | |
2323 | static void | |
2324 | rule_expire(struct rule_dpif *rule) | |
2325 | { | |
2326 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2327 | struct facet *facet, *next_facet; | |
2328 | long long int now; | |
2329 | uint8_t reason; | |
2330 | ||
2331 | /* Has 'rule' expired? */ | |
2332 | now = time_msec(); | |
2333 | if (rule->up.hard_timeout | |
308881af | 2334 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af BP |
2335 | reason = OFPRR_HARD_TIMEOUT; |
2336 | } else if (rule->up.idle_timeout && list_is_empty(&rule->facets) | |
2337 | && now > rule->used + rule->up.idle_timeout * 1000) { | |
2338 | reason = OFPRR_IDLE_TIMEOUT; | |
2339 | } else { | |
2340 | return; | |
2341 | } | |
2342 | ||
2343 | COVERAGE_INC(ofproto_dpif_expired); | |
2344 | ||
2345 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
2346 | * timeout, because that only happens when the rule has no facets left.) */ | |
2347 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
2348 | facet_remove(ofproto, facet); | |
2349 | } | |
2350 | ||
2351 | /* Get rid of the rule. */ | |
2352 | ofproto_rule_expire(&rule->up, reason); | |
2353 | } | |
2354 | \f | |
2355 | /* Facets. */ | |
2356 | ||
f3827897 | 2357 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
2358 | * |
2359 | * The caller must already have determined that no facet with an identical | |
2360 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
2361 | * the ofproto's classifier table. |
2362 | * | |
2363 | * The facet will initially have no ODP actions. The caller should fix that | |
2364 | * by calling facet_make_actions(). */ | |
abe529af | 2365 | static struct facet * |
f3827897 | 2366 | facet_create(struct rule_dpif *rule, const struct flow *flow) |
abe529af BP |
2367 | { |
2368 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2369 | struct facet *facet; | |
2370 | ||
2371 | facet = xzalloc(sizeof *facet); | |
2372 | facet->used = time_msec(); | |
2373 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2374 | list_push_back(&rule->facets, &facet->list_node); | |
2375 | facet->rule = rule; | |
2376 | facet->flow = *flow; | |
2377 | netflow_flow_init(&facet->nf_flow); | |
2378 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2379 | ||
abe529af BP |
2380 | return facet; |
2381 | } | |
2382 | ||
2383 | static void | |
2384 | facet_free(struct facet *facet) | |
2385 | { | |
2386 | free(facet->actions); | |
2387 | free(facet); | |
2388 | } | |
2389 | ||
abe529af | 2390 | static bool |
3d9e05f8 BP |
2391 | execute_controller_action(struct ofproto_dpif *ofproto, |
2392 | const struct flow *flow, | |
2393 | const struct nlattr *odp_actions, size_t actions_len, | |
2394 | struct ofpbuf *packet) | |
2395 | { | |
2396 | if (actions_len | |
2397 | && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE | |
2398 | && NLA_ALIGN(odp_actions->nla_len) == actions_len) { | |
62cd7072 BP |
2399 | /* As an optimization, avoid a round-trip from userspace to kernel to |
2400 | * userspace. This also avoids possibly filling up kernel packet | |
2401 | * buffers along the way. | |
2402 | * | |
2403 | * This optimization will not accidentally catch sFlow | |
2404 | * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated | |
2405 | * inside OVS_ACTION_ATTR_SAMPLE. */ | |
2406 | const struct nlattr *nla; | |
2407 | ||
2408 | nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA); | |
2409 | send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow, | |
2410 | false); | |
2411 | return true; | |
3d9e05f8 BP |
2412 | } else { |
2413 | return false; | |
2414 | } | |
2415 | } | |
2416 | ||
2417 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on | |
2418 | * 'packet', which arrived on 'in_port'. | |
2419 | * | |
2420 | * Takes ownership of 'packet'. */ | |
2421 | static bool | |
2422 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2423 | const struct nlattr *odp_actions, size_t actions_len, | |
2424 | struct ofpbuf *packet) | |
2425 | { | |
2426 | struct odputil_keybuf keybuf; | |
2427 | struct ofpbuf key; | |
2428 | int error; | |
2429 | ||
2430 | if (execute_controller_action(ofproto, flow, odp_actions, actions_len, | |
2431 | packet)) { | |
2432 | return true; | |
6ff686f2 | 2433 | } |
abe529af | 2434 | |
6ff686f2 PS |
2435 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
2436 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 2437 | |
6ff686f2 PS |
2438 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
2439 | odp_actions, actions_len, packet); | |
80e5eed9 | 2440 | |
6ff686f2 PS |
2441 | ofpbuf_delete(packet); |
2442 | return !error; | |
abe529af BP |
2443 | } |
2444 | ||
2445 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s | |
2446 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
2447 | * ofp_packet_in) bytes of headroom. | |
2448 | * | |
2449 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, | |
2450 | * applying flow_extract() to 'packet' would yield the same flow as | |
2451 | * 'facet->flow'. | |
2452 | * | |
df2c07f4 JP |
2453 | * 'facet' must have accurately composed datapath actions; that is, it must |
2454 | * not be in need of revalidation. | |
abe529af BP |
2455 | * |
2456 | * Takes ownership of 'packet'. */ | |
2457 | static void | |
2458 | facet_execute(struct ofproto_dpif *ofproto, struct facet *facet, | |
2459 | struct ofpbuf *packet) | |
2460 | { | |
2461 | struct dpif_flow_stats stats; | |
2462 | ||
2463 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2464 | ||
572b7068 | 2465 | dpif_flow_stats_extract(&facet->flow, packet, &stats); |
abe529af BP |
2466 | stats.used = time_msec(); |
2467 | if (execute_odp_actions(ofproto, &facet->flow, | |
2468 | facet->actions, facet->actions_len, packet)) { | |
2469 | facet_update_stats(ofproto, facet, &stats); | |
2470 | } | |
2471 | } | |
2472 | ||
2473 | /* Remove 'facet' from 'ofproto' and free up the associated memory: | |
2474 | * | |
2475 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2476 | * rule's statistics, via facet_uninstall(). | |
2477 | * | |
2478 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2479 | */ | |
2480 | static void | |
2481 | facet_remove(struct ofproto_dpif *ofproto, struct facet *facet) | |
2482 | { | |
2483 | facet_uninstall(ofproto, facet); | |
2484 | facet_flush_stats(ofproto, facet); | |
2485 | hmap_remove(&ofproto->facets, &facet->hmap_node); | |
2486 | list_remove(&facet->list_node); | |
2487 | facet_free(facet); | |
2488 | } | |
2489 | ||
df2c07f4 | 2490 | /* Composes the datapath actions for 'facet' based on its rule's actions. */ |
abe529af BP |
2491 | static void |
2492 | facet_make_actions(struct ofproto_dpif *p, struct facet *facet, | |
2493 | const struct ofpbuf *packet) | |
2494 | { | |
2495 | const struct rule_dpif *rule = facet->rule; | |
2496 | struct ofpbuf *odp_actions; | |
2497 | struct action_xlate_ctx ctx; | |
2498 | ||
2499 | action_xlate_ctx_init(&ctx, p, &facet->flow, packet); | |
2500 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
2501 | facet->tags = ctx.tags; | |
2502 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
2503 | facet->has_learn = ctx.has_learn; |
2504 | facet->has_normal = ctx.has_normal; | |
abe529af BP |
2505 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
2506 | ||
2507 | if (facet->actions_len != odp_actions->size | |
2508 | || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { | |
2509 | free(facet->actions); | |
2510 | facet->actions_len = odp_actions->size; | |
2511 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
2512 | } | |
2513 | ||
2514 | ofpbuf_delete(odp_actions); | |
2515 | } | |
2516 | ||
3a88e544 BP |
2517 | /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len' |
2518 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
2519 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
2520 | * since 'facet' was last updated. | |
2521 | * | |
2522 | * Returns 0 if successful, otherwise a positive errno value.*/ | |
abe529af BP |
2523 | static int |
2524 | facet_put__(struct ofproto_dpif *ofproto, struct facet *facet, | |
2525 | const struct nlattr *actions, size_t actions_len, | |
2526 | struct dpif_flow_stats *stats) | |
2527 | { | |
2528 | struct odputil_keybuf keybuf; | |
2529 | enum dpif_flow_put_flags flags; | |
2530 | struct ofpbuf key; | |
3a88e544 | 2531 | int ret; |
abe529af BP |
2532 | |
2533 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
2534 | if (stats) { | |
2535 | flags |= DPIF_FP_ZERO_STATS; | |
abe529af BP |
2536 | } |
2537 | ||
2538 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2539 | odp_flow_key_from_flow(&key, &facet->flow); | |
2540 | ||
3a88e544 BP |
2541 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, |
2542 | actions, actions_len, stats); | |
2543 | ||
2544 | if (stats) { | |
2545 | facet_reset_dp_stats(facet, stats); | |
2546 | } | |
2547 | ||
2548 | return ret; | |
abe529af BP |
2549 | } |
2550 | ||
2551 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If | |
2552 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2553 | * 'facet'. */ | |
2554 | static void | |
2555 | facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats) | |
2556 | { | |
2557 | struct dpif_flow_stats stats; | |
2558 | ||
2559 | if (facet->may_install | |
2560 | && !facet_put__(p, facet, facet->actions, facet->actions_len, | |
2561 | zero_stats ? &stats : NULL)) { | |
2562 | facet->installed = true; | |
2563 | } | |
2564 | } | |
2565 | ||
2566 | static void | |
55af77bb | 2567 | facet_account(struct ofproto_dpif *ofproto, struct facet *facet) |
abe529af | 2568 | { |
55af77bb | 2569 | uint64_t n_bytes; |
abe529af | 2570 | const struct nlattr *a; |
abe529af | 2571 | unsigned int left; |
d78be13b | 2572 | ovs_be16 vlan_tci; |
abe529af | 2573 | |
55af77bb | 2574 | if (facet->byte_count <= facet->accounted_bytes) { |
abe529af BP |
2575 | return; |
2576 | } | |
55af77bb EJ |
2577 | n_bytes = facet->byte_count - facet->accounted_bytes; |
2578 | facet->accounted_bytes = facet->byte_count; | |
abe529af | 2579 | |
75a75043 | 2580 | /* Feed information from the active flows back into the learning table to |
abe529af BP |
2581 | * ensure that table is always in sync with what is actually flowing |
2582 | * through the datapath. */ | |
75a75043 BP |
2583 | if (facet->has_learn || facet->has_normal) { |
2584 | struct action_xlate_ctx ctx; | |
abe529af | 2585 | |
75a75043 BP |
2586 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); |
2587 | ctx.may_learn = true; | |
2588 | ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions, | |
2589 | facet->rule->up.n_actions)); | |
2590 | } | |
abe529af | 2591 | |
75a75043 | 2592 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
2593 | return; |
2594 | } | |
d78be13b BP |
2595 | |
2596 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
2597 | * as a basis. We also need to track the actual VLAN on which the packet | |
2598 | * is going to be sent to ensure that it matches the one passed to | |
2599 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
2600 | * hash bucket.) */ | |
2601 | vlan_tci = facet->flow.vlan_tci; | |
abe529af | 2602 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) { |
d78be13b | 2603 | struct ofport_dpif *port; |
abe529af | 2604 | |
d78be13b | 2605 | switch (nl_attr_type(a)) { |
df2c07f4 | 2606 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
2607 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
2608 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 2609 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 2610 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 2611 | } |
d78be13b BP |
2612 | break; |
2613 | ||
d9065a90 | 2614 | case OVS_ACTION_ATTR_POP_VLAN: |
d78be13b BP |
2615 | vlan_tci = htons(0); |
2616 | break; | |
2617 | ||
d9065a90 | 2618 | case OVS_ACTION_ATTR_PUSH_VLAN: |
d78be13b BP |
2619 | vlan_tci = nl_attr_get_be16(a); |
2620 | break; | |
abe529af BP |
2621 | } |
2622 | } | |
2623 | } | |
2624 | ||
2625 | /* If 'rule' is installed in the datapath, uninstalls it. */ | |
2626 | static void | |
2627 | facet_uninstall(struct ofproto_dpif *p, struct facet *facet) | |
2628 | { | |
2629 | if (facet->installed) { | |
2630 | struct odputil_keybuf keybuf; | |
2631 | struct dpif_flow_stats stats; | |
2632 | struct ofpbuf key; | |
3a88e544 | 2633 | int error; |
abe529af BP |
2634 | |
2635 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2636 | odp_flow_key_from_flow(&key, &facet->flow); | |
2637 | ||
3a88e544 BP |
2638 | error = dpif_flow_del(p->dpif, key.data, key.size, &stats); |
2639 | facet_reset_dp_stats(facet, &stats); | |
2640 | if (!error) { | |
abe529af BP |
2641 | facet_update_stats(p, facet, &stats); |
2642 | } | |
2643 | facet->installed = false; | |
abe529af BP |
2644 | } else { |
2645 | assert(facet->dp_packet_count == 0); | |
2646 | assert(facet->dp_byte_count == 0); | |
2647 | } | |
2648 | } | |
2649 | ||
2650 | /* Returns true if the only action for 'facet' is to send to the controller. | |
2651 | * (We don't report NetFlow expiration messages for such facets because they | |
2652 | * are just part of the control logic for the network, not real traffic). */ | |
2653 | static bool | |
2654 | facet_is_controller_flow(struct facet *facet) | |
2655 | { | |
2656 | return (facet | |
2657 | && facet->rule->up.n_actions == 1 | |
2658 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
2659 | htons(OFPP_CONTROLLER))); | |
2660 | } | |
2661 | ||
3a88e544 BP |
2662 | /* Resets 'facet''s datapath statistics counters. This should be called when |
2663 | * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null, | |
2664 | * it should contain the statistics returned by dpif when 'facet' was reset in | |
2665 | * the datapath. 'stats' will be modified to only included statistics new | |
2666 | * since 'facet' was last updated. */ | |
2667 | static void | |
2668 | facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats) | |
2669 | { | |
2670 | if (stats && facet->dp_packet_count <= stats->n_packets | |
2671 | && facet->dp_byte_count <= stats->n_bytes) { | |
2672 | stats->n_packets -= facet->dp_packet_count; | |
2673 | stats->n_bytes -= facet->dp_byte_count; | |
2674 | } | |
2675 | ||
2676 | facet->dp_packet_count = 0; | |
2677 | facet->dp_byte_count = 0; | |
2678 | } | |
2679 | ||
abe529af BP |
2680 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
2681 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
2682 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
2683 | * its packet and byte counts before this function is called. */ | |
2684 | static void | |
2685 | facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet) | |
2686 | { | |
2687 | assert(!facet->dp_byte_count); | |
2688 | assert(!facet->dp_packet_count); | |
2689 | ||
2690 | facet_push_stats(facet); | |
55af77bb | 2691 | facet_account(ofproto, facet); |
abe529af BP |
2692 | |
2693 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
2694 | struct ofexpired expired; | |
2695 | expired.flow = facet->flow; | |
2696 | expired.packet_count = facet->packet_count; | |
2697 | expired.byte_count = facet->byte_count; | |
2698 | expired.used = facet->used; | |
2699 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2700 | } | |
2701 | ||
2702 | facet->rule->packet_count += facet->packet_count; | |
2703 | facet->rule->byte_count += facet->byte_count; | |
2704 | ||
2705 | /* Reset counters to prevent double counting if 'facet' ever gets | |
2706 | * reinstalled. */ | |
bbb5d219 | 2707 | facet_reset_counters(facet); |
abe529af BP |
2708 | |
2709 | netflow_flow_clear(&facet->nf_flow); | |
2710 | } | |
2711 | ||
2712 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2713 | * Returns it if found, otherwise a null pointer. | |
2714 | * | |
2715 | * The returned facet might need revalidation; use facet_lookup_valid() | |
2716 | * instead if that is important. */ | |
2717 | static struct facet * | |
2718 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) | |
2719 | { | |
2720 | struct facet *facet; | |
2721 | ||
2722 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
2723 | &ofproto->facets) { | |
2724 | if (flow_equal(flow, &facet->flow)) { | |
2725 | return facet; | |
2726 | } | |
2727 | } | |
2728 | ||
2729 | return NULL; | |
2730 | } | |
2731 | ||
2732 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
2733 | * Returns it if found, otherwise a null pointer. | |
2734 | * | |
2735 | * The returned facet is guaranteed to be valid. */ | |
2736 | static struct facet * | |
2737 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) | |
2738 | { | |
2739 | struct facet *facet = facet_find(ofproto, flow); | |
2740 | ||
2741 | /* The facet we found might not be valid, since we could be in need of | |
2742 | * revalidation. If it is not valid, don't return it. */ | |
2743 | if (facet | |
0e4b3771 BP |
2744 | && (ofproto->need_revalidate |
2745 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags)) | |
abe529af BP |
2746 | && !facet_revalidate(ofproto, facet)) { |
2747 | COVERAGE_INC(facet_invalidated); | |
2748 | return NULL; | |
2749 | } | |
2750 | ||
2751 | return facet; | |
2752 | } | |
2753 | ||
2754 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': | |
2755 | * | |
2756 | * - If the rule found is different from 'facet''s current rule, moves | |
2757 | * 'facet' to the new rule and recompiles its actions. | |
2758 | * | |
2759 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
2760 | * where it is and recompiles its actions anyway. | |
2761 | * | |
2762 | * - If there is none, destroys 'facet'. | |
2763 | * | |
2764 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
2765 | static bool | |
2766 | facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet) | |
2767 | { | |
2768 | struct action_xlate_ctx ctx; | |
2769 | struct ofpbuf *odp_actions; | |
2770 | struct rule_dpif *new_rule; | |
2771 | bool actions_changed; | |
2772 | ||
2773 | COVERAGE_INC(facet_revalidate); | |
2774 | ||
2775 | /* Determine the new rule. */ | |
29901626 | 2776 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); |
abe529af BP |
2777 | if (!new_rule) { |
2778 | /* No new rule, so delete the facet. */ | |
2779 | facet_remove(ofproto, facet); | |
2780 | return false; | |
2781 | } | |
2782 | ||
df2c07f4 | 2783 | /* Calculate new datapath actions. |
abe529af BP |
2784 | * |
2785 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
2786 | * emit a NetFlow expiration and, if so, we need to have the old state | |
2787 | * around to properly compose it. */ | |
2788 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); | |
2789 | odp_actions = xlate_actions(&ctx, | |
2790 | new_rule->up.actions, new_rule->up.n_actions); | |
2791 | actions_changed = (facet->actions_len != odp_actions->size | |
2792 | || memcmp(facet->actions, odp_actions->data, | |
2793 | facet->actions_len)); | |
2794 | ||
df2c07f4 JP |
2795 | /* If the datapath actions changed or the installability changed, |
2796 | * then we need to talk to the datapath. */ | |
abe529af BP |
2797 | if (actions_changed || ctx.may_set_up_flow != facet->installed) { |
2798 | if (ctx.may_set_up_flow) { | |
2799 | struct dpif_flow_stats stats; | |
2800 | ||
2801 | facet_put__(ofproto, facet, | |
2802 | odp_actions->data, odp_actions->size, &stats); | |
2803 | facet_update_stats(ofproto, facet, &stats); | |
2804 | } else { | |
2805 | facet_uninstall(ofproto, facet); | |
2806 | } | |
2807 | ||
2808 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
2809 | * 'facet' into 'rule'. */ | |
2810 | facet_flush_stats(ofproto, facet); | |
2811 | } | |
2812 | ||
2813 | /* Update 'facet' now that we've taken care of all the old state. */ | |
2814 | facet->tags = ctx.tags; | |
2815 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
2816 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
2817 | facet->has_learn = ctx.has_learn; |
2818 | facet->has_normal = ctx.has_normal; | |
abe529af BP |
2819 | if (actions_changed) { |
2820 | free(facet->actions); | |
2821 | facet->actions_len = odp_actions->size; | |
2822 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
2823 | } | |
2824 | if (facet->rule != new_rule) { | |
2825 | COVERAGE_INC(facet_changed_rule); | |
2826 | list_remove(&facet->list_node); | |
2827 | list_push_back(&new_rule->facets, &facet->list_node); | |
2828 | facet->rule = new_rule; | |
2829 | facet->used = new_rule->up.created; | |
2830 | facet->rs_used = facet->used; | |
2831 | } | |
2832 | ||
2833 | ofpbuf_delete(odp_actions); | |
2834 | ||
2835 | return true; | |
2836 | } | |
2837 | ||
2838 | /* Updates 'facet''s used time. Caller is responsible for calling | |
2839 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
2840 | static void | |
2841 | facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet, | |
2842 | long long int used) | |
2843 | { | |
2844 | if (used > facet->used) { | |
2845 | facet->used = used; | |
2846 | if (used > facet->rule->used) { | |
2847 | facet->rule->used = used; | |
2848 | } | |
2849 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
2850 | } | |
2851 | } | |
2852 | ||
2853 | /* Folds the statistics from 'stats' into the counters in 'facet'. | |
2854 | * | |
2855 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
2856 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
2857 | * packet that was sent by hand or if it represents statistics that have been | |
2858 | * cleared out of the datapath. */ | |
2859 | static void | |
2860 | facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet, | |
2861 | const struct dpif_flow_stats *stats) | |
2862 | { | |
2863 | if (stats->n_packets || stats->used > facet->used) { | |
2864 | facet_update_time(ofproto, facet, stats->used); | |
2865 | facet->packet_count += stats->n_packets; | |
2866 | facet->byte_count += stats->n_bytes; | |
2867 | facet_push_stats(facet); | |
2868 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
2869 | } | |
2870 | } | |
2871 | ||
bbb5d219 EJ |
2872 | static void |
2873 | facet_reset_counters(struct facet *facet) | |
2874 | { | |
2875 | facet->packet_count = 0; | |
2876 | facet->byte_count = 0; | |
2877 | facet->rs_packet_count = 0; | |
2878 | facet->rs_byte_count = 0; | |
2879 | facet->accounted_bytes = 0; | |
2880 | } | |
2881 | ||
abe529af BP |
2882 | static void |
2883 | facet_push_stats(struct facet *facet) | |
2884 | { | |
2885 | uint64_t rs_packets, rs_bytes; | |
2886 | ||
2887 | assert(facet->packet_count >= facet->rs_packet_count); | |
2888 | assert(facet->byte_count >= facet->rs_byte_count); | |
2889 | assert(facet->used >= facet->rs_used); | |
2890 | ||
2891 | rs_packets = facet->packet_count - facet->rs_packet_count; | |
2892 | rs_bytes = facet->byte_count - facet->rs_byte_count; | |
2893 | ||
2894 | if (rs_packets || rs_bytes || facet->used > facet->rs_used) { | |
2895 | facet->rs_packet_count = facet->packet_count; | |
2896 | facet->rs_byte_count = facet->byte_count; | |
2897 | facet->rs_used = facet->used; | |
2898 | ||
2899 | flow_push_stats(facet->rule, &facet->flow, | |
2900 | rs_packets, rs_bytes, facet->used); | |
2901 | } | |
2902 | } | |
2903 | ||
2904 | struct ofproto_push { | |
2905 | struct action_xlate_ctx ctx; | |
2906 | uint64_t packets; | |
2907 | uint64_t bytes; | |
2908 | long long int used; | |
2909 | }; | |
2910 | ||
2911 | static void | |
2912 | push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
2913 | { | |
2914 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
2915 | ||
2916 | if (rule) { | |
2917 | rule->packet_count += push->packets; | |
2918 | rule->byte_count += push->bytes; | |
2919 | rule->used = MAX(push->used, rule->used); | |
2920 | } | |
2921 | } | |
2922 | ||
2923 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
2924 | * 'rule''s actions. */ | |
2925 | static void | |
2926 | flow_push_stats(const struct rule_dpif *rule, | |
2927 | struct flow *flow, uint64_t packets, uint64_t bytes, | |
2928 | long long int used) | |
2929 | { | |
2930 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2931 | struct ofproto_push push; | |
2932 | ||
2933 | push.packets = packets; | |
2934 | push.bytes = bytes; | |
2935 | push.used = used; | |
2936 | ||
2937 | action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); | |
2938 | push.ctx.resubmit_hook = push_resubmit; | |
2939 | ofpbuf_delete(xlate_actions(&push.ctx, | |
2940 | rule->up.actions, rule->up.n_actions)); | |
2941 | } | |
2942 | \f | |
2943 | /* Rules. */ | |
2944 | ||
2945 | static struct rule_dpif * | |
29901626 BP |
2946 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
2947 | uint8_t table_id) | |
abe529af | 2948 | { |
9cdaaebe BP |
2949 | if (table_id >= N_TABLES) { |
2950 | return NULL; | |
2951 | } | |
2952 | ||
154896e3 | 2953 | return rule_dpif_cast(rule_from_cls_rule( |
29901626 | 2954 | classifier_lookup(&ofproto->up.tables[table_id], |
6c1491fb | 2955 | flow))); |
abe529af BP |
2956 | } |
2957 | ||
7ee20df1 BP |
2958 | static void |
2959 | complete_operation(struct rule_dpif *rule) | |
2960 | { | |
2961 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2962 | ||
54a9cbc9 | 2963 | rule_invalidate(rule); |
7ee20df1 BP |
2964 | if (clogged) { |
2965 | struct dpif_completion *c = xmalloc(sizeof *c); | |
2966 | c->op = rule->up.pending; | |
2967 | list_push_back(&ofproto->completions, &c->list_node); | |
2968 | } else { | |
2969 | ofoperation_complete(rule->up.pending, 0); | |
2970 | } | |
2971 | } | |
2972 | ||
abe529af BP |
2973 | static struct rule * |
2974 | rule_alloc(void) | |
2975 | { | |
2976 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
2977 | return &rule->up; | |
2978 | } | |
2979 | ||
2980 | static void | |
2981 | rule_dealloc(struct rule *rule_) | |
2982 | { | |
2983 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2984 | free(rule); | |
2985 | } | |
2986 | ||
2987 | static int | |
2988 | rule_construct(struct rule *rule_) | |
2989 | { | |
2990 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
2991 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 2992 | struct rule_dpif *victim; |
54a9cbc9 | 2993 | uint8_t table_id; |
5bf0e941 BP |
2994 | int error; |
2995 | ||
2996 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
2997 | &rule->up.cr.flow, ofproto->max_ports); | |
2998 | if (error) { | |
2999 | return error; | |
3000 | } | |
abe529af BP |
3001 | |
3002 | rule->used = rule->up.created; | |
3003 | rule->packet_count = 0; | |
3004 | rule->byte_count = 0; | |
abe529af | 3005 | |
7ee20df1 BP |
3006 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
3007 | if (victim && !list_is_empty(&victim->facets)) { | |
3008 | struct facet *facet; | |
3009 | ||
3010 | rule->facets = victim->facets; | |
3011 | list_moved(&rule->facets); | |
3012 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
3013 | /* XXX: We're only clearing our local counters here. It's possible |
3014 | * that quite a few packets are unaccounted for in the datapath | |
3015 | * statistics. These will be accounted to the new rule instead of | |
3016 | * cleared as required. This could be fixed by clearing out the | |
3017 | * datapath statistics for this facet, but currently it doesn't | |
3018 | * seem worth it. */ | |
3019 | facet_reset_counters(facet); | |
7ee20df1 BP |
3020 | facet->rule = rule; |
3021 | } | |
3022 | } else { | |
3023 | /* Must avoid list_moved() in this case. */ | |
3024 | list_init(&rule->facets); | |
3025 | } | |
abe529af | 3026 | |
54a9cbc9 BP |
3027 | table_id = rule->up.table_id; |
3028 | rule->tag = (victim ? victim->tag | |
3029 | : table_id == 0 ? 0 | |
3030 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
3031 | ofproto->tables[table_id].basis)); | |
3032 | ||
7ee20df1 | 3033 | complete_operation(rule); |
abe529af BP |
3034 | return 0; |
3035 | } | |
3036 | ||
3037 | static void | |
3038 | rule_destruct(struct rule *rule_) | |
3039 | { | |
3040 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3041 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3042 | struct facet *facet, *next_facet; | |
3043 | ||
abe529af BP |
3044 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
3045 | facet_revalidate(ofproto, facet); | |
3046 | } | |
7ee20df1 BP |
3047 | |
3048 | complete_operation(rule); | |
abe529af BP |
3049 | } |
3050 | ||
3051 | static void | |
3052 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
3053 | { | |
3054 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3055 | struct facet *facet; | |
3056 | ||
3057 | /* Start from historical data for 'rule' itself that are no longer tracked | |
3058 | * in facets. This counts, for example, facets that have expired. */ | |
3059 | *packets = rule->packet_count; | |
3060 | *bytes = rule->byte_count; | |
3061 | ||
3062 | /* Add any statistics that are tracked by facets. This includes | |
3063 | * statistical data recently updated by ofproto_update_stats() as well as | |
3064 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
3065 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
3066 | *packets += facet->packet_count; | |
3067 | *bytes += facet->byte_count; | |
3068 | } | |
3069 | } | |
3070 | ||
5bf0e941 | 3071 | static int |
abe529af BP |
3072 | rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet) |
3073 | { | |
3074 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3075 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3076 | struct action_xlate_ctx ctx; | |
3077 | struct ofpbuf *odp_actions; | |
3078 | struct facet *facet; | |
3079 | size_t size; | |
3080 | ||
3081 | /* First look for a related facet. If we find one, account it to that. */ | |
3082 | facet = facet_lookup_valid(ofproto, flow); | |
3083 | if (facet && facet->rule == rule) { | |
eea109bb BP |
3084 | if (!facet->may_install) { |
3085 | facet_make_actions(ofproto, facet, packet); | |
3086 | } | |
abe529af | 3087 | facet_execute(ofproto, facet, packet); |
5bf0e941 | 3088 | return 0; |
abe529af BP |
3089 | } |
3090 | ||
3091 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then | |
3092 | * create a new facet for it and use that. */ | |
29901626 | 3093 | if (rule_dpif_lookup(ofproto, flow, 0) == rule) { |
f3827897 BP |
3094 | facet = facet_create(rule, flow); |
3095 | facet_make_actions(ofproto, facet, packet); | |
abe529af BP |
3096 | facet_execute(ofproto, facet, packet); |
3097 | facet_install(ofproto, facet, true); | |
5bf0e941 | 3098 | return 0; |
abe529af BP |
3099 | } |
3100 | ||
3101 | /* We can't account anything to a facet. If we were to try, then that | |
3102 | * facet would have a non-matching rule, busting our invariants. */ | |
3103 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
3104 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
3105 | size = packet->size; | |
3106 | if (execute_odp_actions(ofproto, flow, odp_actions->data, | |
3107 | odp_actions->size, packet)) { | |
3108 | rule->used = time_msec(); | |
3109 | rule->packet_count++; | |
3110 | rule->byte_count += size; | |
3111 | flow_push_stats(rule, flow, 1, size, rule->used); | |
3112 | } | |
3113 | ofpbuf_delete(odp_actions); | |
5bf0e941 BP |
3114 | |
3115 | return 0; | |
abe529af BP |
3116 | } |
3117 | ||
7ee20df1 BP |
3118 | static void |
3119 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
3120 | { |
3121 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3122 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3123 | int error; | |
3124 | ||
7ee20df1 BP |
3125 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
3126 | &rule->up.cr.flow, ofproto->max_ports); | |
3127 | if (error) { | |
3128 | ofoperation_complete(rule->up.pending, error); | |
3129 | return; | |
abe529af | 3130 | } |
7ee20df1 BP |
3131 | |
3132 | complete_operation(rule); | |
abe529af BP |
3133 | } |
3134 | \f | |
b47e2a82 | 3135 | /* Sends 'packet' out of port 'odp_port' within 'ofproto'. |
abe529af BP |
3136 | * Returns 0 if successful, otherwise a positive errno value. */ |
3137 | static int | |
b2fda3ef | 3138 | send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, |
abe529af BP |
3139 | const struct ofpbuf *packet) |
3140 | { | |
80e5eed9 BP |
3141 | struct ofpbuf key, odp_actions; |
3142 | struct odputil_keybuf keybuf; | |
3143 | struct flow flow; | |
abe529af BP |
3144 | int error; |
3145 | ||
80e5eed9 BP |
3146 | flow_extract((struct ofpbuf *) packet, 0, 0, &flow); |
3147 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
3148 | odp_flow_key_from_flow(&key, &flow); | |
3149 | ||
abe529af | 3150 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
3151 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
3152 | ||
df2c07f4 | 3153 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
3154 | error = dpif_execute(ofproto->dpif, |
3155 | key.data, key.size, | |
3156 | odp_actions.data, odp_actions.size, | |
abe529af BP |
3157 | packet); |
3158 | ofpbuf_uninit(&odp_actions); | |
3159 | ||
3160 | if (error) { | |
3161 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
3162 | ofproto->up.name, odp_port, strerror(error)); | |
3163 | } | |
3164 | return error; | |
3165 | } | |
3166 | \f | |
df2c07f4 | 3167 | /* OpenFlow to datapath action translation. */ |
abe529af BP |
3168 | |
3169 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
3170 | struct action_xlate_ctx *ctx); | |
4cd78906 | 3171 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 3172 | |
98403001 BP |
3173 | static size_t |
3174 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
3175 | struct ofpbuf *odp_actions, | |
3176 | const struct flow *flow, | |
3177 | const struct user_action_cookie *cookie) | |
3178 | { | |
3179 | size_t offset; | |
3180 | uint32_t pid; | |
3181 | ||
3182 | pid = dpif_port_get_pid(ofproto->dpif, | |
3183 | ofp_port_to_odp_port(flow->in_port)); | |
3184 | ||
3185 | offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE); | |
3186 | nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid); | |
3187 | nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA, | |
3188 | cookie, sizeof *cookie); | |
3189 | nl_msg_end_nested(odp_actions, offset); | |
3190 | ||
3191 | return odp_actions->size - NLA_ALIGN(sizeof *cookie); | |
3192 | } | |
3193 | ||
6ff686f2 PS |
3194 | /* Compose SAMPLE action for sFlow. */ |
3195 | static size_t | |
3196 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
3197 | struct ofpbuf *odp_actions, | |
3198 | const struct flow *flow, | |
3199 | uint32_t odp_port) | |
3200 | { | |
3201 | uint32_t port_ifindex; | |
3202 | uint32_t probability; | |
98403001 | 3203 | struct user_action_cookie cookie; |
6ff686f2 | 3204 | size_t sample_offset, actions_offset; |
98403001 | 3205 | int cookie_offset, n_output; |
6ff686f2 PS |
3206 | |
3207 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
3208 | return 0; | |
3209 | } | |
3210 | ||
3211 | if (odp_port == OVSP_NONE) { | |
3212 | port_ifindex = 0; | |
3213 | n_output = 0; | |
3214 | } else { | |
3215 | port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
3216 | n_output = 1; | |
3217 | } | |
3218 | ||
3219 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
3220 | ||
3221 | /* Number of packets out of UINT_MAX to sample. */ | |
3222 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
3223 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
3224 | ||
3225 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
3226 | ||
98403001 BP |
3227 | cookie.type = USER_ACTION_COOKIE_SFLOW; |
3228 | cookie.data = port_ifindex; | |
3229 | cookie.n_output = n_output; | |
3230 | cookie.vlan_tci = 0; | |
3231 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); | |
6ff686f2 PS |
3232 | |
3233 | nl_msg_end_nested(odp_actions, actions_offset); | |
3234 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 3235 | return cookie_offset; |
6ff686f2 PS |
3236 | } |
3237 | ||
3238 | /* SAMPLE action must be first action in any given list of actions. | |
3239 | * At this point we do not have all information required to build it. So try to | |
3240 | * build sample action as complete as possible. */ | |
3241 | static void | |
3242 | add_sflow_action(struct action_xlate_ctx *ctx) | |
3243 | { | |
3244 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
3245 | ctx->odp_actions, | |
3246 | &ctx->flow, OVSP_NONE); | |
3247 | ctx->sflow_odp_port = 0; | |
3248 | ctx->sflow_n_outputs = 0; | |
3249 | } | |
3250 | ||
3251 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
3252 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
3253 | * USERSPACE action's user-cookie which is required for sflow. */ | |
3254 | static void | |
3255 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
3256 | { | |
3257 | const struct flow *base = &ctx->base_flow; | |
3258 | struct user_action_cookie *cookie; | |
3259 | ||
3260 | if (!ctx->user_cookie_offset) { | |
3261 | return; | |
3262 | } | |
3263 | ||
3264 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
3265 | sizeof(*cookie)); | |
3266 | assert(cookie != NULL); | |
3267 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
3268 | ||
3269 | if (ctx->sflow_n_outputs) { | |
3270 | cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow, | |
3271 | ctx->sflow_odp_port); | |
3272 | } | |
3273 | if (ctx->sflow_n_outputs >= 255) { | |
3274 | cookie->n_output = 255; | |
3275 | } else { | |
3276 | cookie->n_output = ctx->sflow_n_outputs; | |
3277 | } | |
3278 | cookie->vlan_tci = base->vlan_tci; | |
3279 | } | |
3280 | ||
823518f1 BP |
3281 | static void |
3282 | commit_vlan_tci(struct action_xlate_ctx *ctx, ovs_be16 vlan_tci) | |
3283 | { | |
3284 | struct flow *base = &ctx->base_flow; | |
3285 | struct ofpbuf *odp_actions = ctx->odp_actions; | |
3286 | ||
3287 | if (base->vlan_tci != vlan_tci) { | |
3288 | if (!(vlan_tci & htons(VLAN_CFI))) { | |
3289 | nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN); | |
3290 | } else { | |
3291 | if (base->vlan_tci != htons(0)) { | |
3292 | nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_VLAN); | |
3293 | } | |
3294 | nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_PUSH_VLAN, | |
3295 | vlan_tci & ~htons(VLAN_CFI)); | |
3296 | } | |
3297 | base->vlan_tci = vlan_tci; | |
3298 | } | |
3299 | } | |
3300 | ||
b3e9b2ed EJ |
3301 | static void |
3302 | commit_odp_actions(struct action_xlate_ctx *ctx) | |
3303 | { | |
3304 | const struct flow *flow = &ctx->flow; | |
3305 | struct flow *base = &ctx->base_flow; | |
3306 | struct ofpbuf *odp_actions = ctx->odp_actions; | |
3307 | ||
3308 | if (base->tun_id != flow->tun_id) { | |
df2c07f4 | 3309 | nl_msg_put_be64(odp_actions, OVS_ACTION_ATTR_SET_TUNNEL, flow->tun_id); |
b3e9b2ed EJ |
3310 | base->tun_id = flow->tun_id; |
3311 | } | |
3312 | ||
3313 | if (base->nw_src != flow->nw_src) { | |
df2c07f4 | 3314 | nl_msg_put_be32(odp_actions, OVS_ACTION_ATTR_SET_NW_SRC, flow->nw_src); |
b3e9b2ed EJ |
3315 | base->nw_src = flow->nw_src; |
3316 | } | |
3317 | ||
3318 | if (base->nw_dst != flow->nw_dst) { | |
df2c07f4 | 3319 | nl_msg_put_be32(odp_actions, OVS_ACTION_ATTR_SET_NW_DST, flow->nw_dst); |
b3e9b2ed EJ |
3320 | base->nw_dst = flow->nw_dst; |
3321 | } | |
3322 | ||
150a9f15 | 3323 | if (base->nw_tos != flow->nw_tos) { |
df2c07f4 | 3324 | nl_msg_put_u8(odp_actions, OVS_ACTION_ATTR_SET_NW_TOS, flow->nw_tos); |
150a9f15 BP |
3325 | base->nw_tos = flow->nw_tos; |
3326 | } | |
3327 | ||
823518f1 | 3328 | commit_vlan_tci(ctx, flow->vlan_tci); |
b3e9b2ed EJ |
3329 | |
3330 | if (base->tp_src != flow->tp_src) { | |
df2c07f4 | 3331 | nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_SET_TP_SRC, flow->tp_src); |
b3e9b2ed EJ |
3332 | base->tp_src = flow->tp_src; |
3333 | } | |
3334 | ||
3335 | if (base->tp_dst != flow->tp_dst) { | |
df2c07f4 | 3336 | nl_msg_put_be16(odp_actions, OVS_ACTION_ATTR_SET_TP_DST, flow->tp_dst); |
b3e9b2ed EJ |
3337 | base->tp_dst = flow->tp_dst; |
3338 | } | |
3339 | ||
3340 | if (!eth_addr_equals(base->dl_src, flow->dl_src)) { | |
df2c07f4 | 3341 | nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_SET_DL_SRC, |
b3e9b2ed EJ |
3342 | flow->dl_src, ETH_ADDR_LEN); |
3343 | memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN); | |
3344 | } | |
3345 | ||
3346 | if (!eth_addr_equals(base->dl_dst, flow->dl_dst)) { | |
df2c07f4 | 3347 | nl_msg_put_unspec(odp_actions, OVS_ACTION_ATTR_SET_DL_DST, |
b3e9b2ed EJ |
3348 | flow->dl_dst, ETH_ADDR_LEN); |
3349 | memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN); | |
3350 | } | |
3351 | ||
3352 | if (ctx->base_priority != ctx->priority) { | |
3353 | if (ctx->priority) { | |
df2c07f4 | 3354 | nl_msg_put_u32(odp_actions, OVS_ACTION_ATTR_SET_PRIORITY, |
b3e9b2ed EJ |
3355 | ctx->priority); |
3356 | } else { | |
df2c07f4 | 3357 | nl_msg_put_flag(odp_actions, OVS_ACTION_ATTR_POP_PRIORITY); |
b3e9b2ed EJ |
3358 | } |
3359 | ctx->base_priority = ctx->priority; | |
3360 | } | |
3361 | } | |
3362 | ||
6ff686f2 PS |
3363 | static void |
3364 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port) | |
3365 | { | |
3366 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); | |
3367 | ctx->sflow_odp_port = odp_port; | |
3368 | ctx->sflow_n_outputs++; | |
3369 | } | |
3370 | ||
abe529af BP |
3371 | static void |
3372 | add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) | |
3373 | { | |
3374 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); | |
3375 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); | |
3376 | ||
3377 | if (ofport) { | |
3378 | if (ofport->up.opp.config & htonl(OFPPC_NO_FWD)) { | |
3379 | /* Forwarding disabled on port. */ | |
3380 | return; | |
3381 | } | |
3382 | } else { | |
3383 | /* | |
3384 | * We don't have an ofport record for this port, but it doesn't hurt to | |
3385 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
3386 | * and we're pre-populating the flow table. | |
3387 | */ | |
3388 | } | |
3389 | ||
b3e9b2ed | 3390 | commit_odp_actions(ctx); |
6ff686f2 | 3391 | compose_output_action(ctx, odp_port); |
abe529af BP |
3392 | ctx->nf_output_iface = ofp_port; |
3393 | } | |
3394 | ||
3395 | static void | |
29901626 BP |
3396 | xlate_table_action(struct action_xlate_ctx *ctx, |
3397 | uint16_t in_port, uint8_t table_id) | |
abe529af BP |
3398 | { |
3399 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 3400 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
3401 | struct rule_dpif *rule; |
3402 | uint16_t old_in_port; | |
29901626 BP |
3403 | uint8_t old_table_id; |
3404 | ||
3405 | old_table_id = ctx->table_id; | |
3406 | ctx->table_id = table_id; | |
abe529af | 3407 | |
54a9cbc9 | 3408 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
3409 | old_in_port = ctx->flow.in_port; |
3410 | ctx->flow.in_port = in_port; | |
54a9cbc9 BP |
3411 | rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id); |
3412 | ||
3413 | /* Tag the flow. */ | |
3414 | if (table_id > 0 && table_id < N_TABLES) { | |
3415 | struct table_dpif *table = &ofproto->tables[table_id]; | |
3416 | if (table->other_table) { | |
3417 | ctx->tags |= (rule | |
3418 | ? rule->tag | |
3419 | : rule_calculate_tag(&ctx->flow, | |
3420 | &table->other_table->wc, | |
3421 | table->basis)); | |
3422 | } | |
3423 | } | |
3424 | ||
3425 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
3426 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
3427 | ctx->flow.in_port = old_in_port; |
3428 | ||
3429 | if (ctx->resubmit_hook) { | |
3430 | ctx->resubmit_hook(ctx, rule); | |
3431 | } | |
3432 | ||
3433 | if (rule) { | |
3434 | ctx->recurse++; | |
3435 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); | |
3436 | ctx->recurse--; | |
3437 | } | |
29901626 BP |
3438 | |
3439 | ctx->table_id = old_table_id; | |
abe529af BP |
3440 | } else { |
3441 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
3442 | ||
29901626 | 3443 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af BP |
3444 | MAX_RESUBMIT_RECURSION); |
3445 | } | |
3446 | } | |
3447 | ||
29901626 BP |
3448 | static void |
3449 | xlate_resubmit_table(struct action_xlate_ctx *ctx, | |
3450 | const struct nx_action_resubmit *nar) | |
3451 | { | |
3452 | uint16_t in_port; | |
3453 | uint8_t table_id; | |
3454 | ||
3455 | in_port = (nar->in_port == htons(OFPP_IN_PORT) | |
3456 | ? ctx->flow.in_port | |
3457 | : ntohs(nar->in_port)); | |
3458 | table_id = nar->table == 255 ? ctx->table_id : nar->table; | |
3459 | ||
3460 | xlate_table_action(ctx, in_port, table_id); | |
3461 | } | |
3462 | ||
abe529af | 3463 | static void |
b3e9b2ed | 3464 | flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask) |
abe529af BP |
3465 | { |
3466 | struct ofport_dpif *ofport; | |
3467 | ||
b3e9b2ed EJ |
3468 | commit_odp_actions(ctx); |
3469 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { | |
abe529af | 3470 | uint16_t ofp_port = ofport->up.ofp_port; |
b3e9b2ed | 3471 | if (ofp_port != ctx->flow.in_port && !(ofport->up.opp.config & mask)) { |
6ff686f2 | 3472 | compose_output_action(ctx, ofport->odp_port); |
abe529af BP |
3473 | } |
3474 | } | |
b3e9b2ed EJ |
3475 | |
3476 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
3477 | } |
3478 | ||
6ff686f2 | 3479 | static void |
98403001 | 3480 | compose_controller_action(struct action_xlate_ctx *ctx, int len) |
6ff686f2 PS |
3481 | { |
3482 | struct user_action_cookie cookie; | |
3483 | ||
3484 | cookie.type = USER_ACTION_COOKIE_CONTROLLER; | |
3485 | cookie.data = len; | |
3486 | cookie.n_output = 0; | |
3487 | cookie.vlan_tci = 0; | |
98403001 | 3488 | put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie); |
6ff686f2 PS |
3489 | } |
3490 | ||
abe529af BP |
3491 | static void |
3492 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
3493 | uint16_t port, uint16_t max_len) | |
3494 | { | |
3495 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
3496 | ||
3497 | ctx->nf_output_iface = NF_OUT_DROP; | |
3498 | ||
3499 | switch (port) { | |
3500 | case OFPP_IN_PORT: | |
3501 | add_output_action(ctx, ctx->flow.in_port); | |
3502 | break; | |
3503 | case OFPP_TABLE: | |
29901626 | 3504 | xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); |
abe529af BP |
3505 | break; |
3506 | case OFPP_NORMAL: | |
3507 | xlate_normal(ctx); | |
3508 | break; | |
3509 | case OFPP_FLOOD: | |
b3e9b2ed | 3510 | flood_packets(ctx, htonl(OFPPC_NO_FLOOD)); |
abe529af BP |
3511 | break; |
3512 | case OFPP_ALL: | |
b3e9b2ed | 3513 | flood_packets(ctx, htonl(0)); |
abe529af BP |
3514 | break; |
3515 | case OFPP_CONTROLLER: | |
b3e9b2ed | 3516 | commit_odp_actions(ctx); |
98403001 | 3517 | compose_controller_action(ctx, max_len); |
abe529af BP |
3518 | break; |
3519 | case OFPP_LOCAL: | |
3520 | add_output_action(ctx, OFPP_LOCAL); | |
3521 | break; | |
e81d2933 EJ |
3522 | case OFPP_NONE: |
3523 | break; | |
abe529af BP |
3524 | default: |
3525 | if (port != ctx->flow.in_port) { | |
3526 | add_output_action(ctx, port); | |
3527 | } | |
3528 | break; | |
3529 | } | |
3530 | ||
3531 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
3532 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
3533 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
3534 | ctx->nf_output_iface = prev_nf_output_iface; | |
3535 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
3536 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
3537 | ctx->nf_output_iface = NF_OUT_MULTI; | |
3538 | } | |
3539 | } | |
3540 | ||
f694937d EJ |
3541 | static void |
3542 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
3543 | const struct nx_action_output_reg *naor) | |
3544 | { | |
3545 | uint64_t ofp_port; | |
3546 | ||
3547 | ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow); | |
3548 | ||
3549 | if (ofp_port <= UINT16_MAX) { | |
3550 | xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); | |
3551 | } | |
3552 | } | |
3553 | ||
abe529af BP |
3554 | static void |
3555 | xlate_output_action(struct action_xlate_ctx *ctx, | |
3556 | const struct ofp_action_output *oao) | |
3557 | { | |
3558 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
3559 | } | |
3560 | ||
abe529af BP |
3561 | static void |
3562 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
3563 | const struct ofp_action_enqueue *oae) | |
3564 | { | |
3565 | uint16_t ofp_port, odp_port; | |
b3e9b2ed | 3566 | uint32_t ctx_priority, priority; |
abe529af BP |
3567 | int error; |
3568 | ||
3569 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
3570 | &priority); | |
3571 | if (error) { | |
3572 | /* Fall back to ordinary output action. */ | |
3573 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
3574 | return; | |
3575 | } | |
3576 | ||
df2c07f4 | 3577 | /* Figure out datapath output port. */ |
abe529af BP |
3578 | ofp_port = ntohs(oae->port); |
3579 | if (ofp_port == OFPP_IN_PORT) { | |
3580 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
3581 | } else if (ofp_port == ctx->flow.in_port) { |
3582 | return; | |
abe529af BP |
3583 | } |
3584 | odp_port = ofp_port_to_odp_port(ofp_port); | |
3585 | ||
df2c07f4 | 3586 | /* Add datapath actions. */ |
b3e9b2ed EJ |
3587 | ctx_priority = ctx->priority; |
3588 | ctx->priority = priority; | |
abe529af | 3589 | add_output_action(ctx, odp_port); |
b3e9b2ed | 3590 | ctx->priority = ctx_priority; |
abe529af BP |
3591 | |
3592 | /* Update NetFlow output port. */ | |
3593 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
3594 | ctx->nf_output_iface = odp_port; | |
3595 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
3596 | ctx->nf_output_iface = NF_OUT_MULTI; | |
3597 | } | |
3598 | } | |
3599 | ||
3600 | static void | |
3601 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
3602 | const struct nx_action_set_queue *nasq) | |
3603 | { | |
3604 | uint32_t priority; | |
3605 | int error; | |
3606 | ||
3607 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
3608 | &priority); | |
3609 | if (error) { | |
3610 | /* Couldn't translate queue to a priority, so ignore. A warning | |
3611 | * has already been logged. */ | |
3612 | return; | |
3613 | } | |
3614 | ||
afabef2b | 3615 | ctx->priority = priority; |
abe529af BP |
3616 | } |
3617 | ||
3618 | struct xlate_reg_state { | |
3619 | ovs_be16 vlan_tci; | |
3620 | ovs_be64 tun_id; | |
3621 | }; | |
3622 | ||
abe529af BP |
3623 | static void |
3624 | xlate_autopath(struct action_xlate_ctx *ctx, | |
3625 | const struct nx_action_autopath *naa) | |
3626 | { | |
3627 | uint16_t ofp_port = ntohl(naa->id); | |
3628 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
3629 | ||
3630 | if (!port || !port->bundle) { | |
3631 | ofp_port = OFPP_NONE; | |
3632 | } else if (port->bundle->bond) { | |
3633 | /* Autopath does not support VLAN hashing. */ | |
3634 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 3635 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
3636 | if (slave) { |
3637 | ofp_port = slave->up.ofp_port; | |
3638 | } | |
3639 | } | |
3640 | autopath_execute(naa, &ctx->flow, ofp_port); | |
3641 | } | |
3642 | ||
daff3353 EJ |
3643 | static bool |
3644 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
3645 | { | |
3646 | struct ofproto_dpif *ofproto = ofproto_; | |
3647 | struct ofport_dpif *port; | |
3648 | ||
3649 | switch (ofp_port) { | |
3650 | case OFPP_IN_PORT: | |
3651 | case OFPP_TABLE: | |
3652 | case OFPP_NORMAL: | |
3653 | case OFPP_FLOOD: | |
3654 | case OFPP_ALL: | |
3655 | case OFPP_LOCAL: | |
3656 | return true; | |
3657 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
3658 | return false; | |
3659 | default: | |
3660 | port = get_ofp_port(ofproto, ofp_port); | |
3661 | return port ? port->may_enable : false; | |
3662 | } | |
3663 | } | |
3664 | ||
75a75043 BP |
3665 | static void |
3666 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
3667 | const struct nx_action_learn *learn) | |
3668 | { | |
3669 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
3670 | struct ofputil_flow_mod fm; | |
3671 | int error; | |
3672 | ||
3673 | learn_execute(learn, &ctx->flow, &fm); | |
3674 | ||
3675 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
3676 | if (error && !VLOG_DROP_WARN(&rl)) { | |
3677 | char *msg = ofputil_error_to_string(error); | |
3678 | VLOG_WARN("learning action failed to modify flow table (%s)", msg); | |
3679 | free(msg); | |
3680 | } | |
3681 | ||
3682 | free(fm.actions); | |
3683 | } | |
3684 | ||
abe529af BP |
3685 | static void |
3686 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
3687 | struct action_xlate_ctx *ctx) | |
3688 | { | |
3689 | const struct ofport_dpif *port; | |
abe529af | 3690 | const union ofp_action *ia; |
b4b8c781 | 3691 | size_t left; |
abe529af BP |
3692 | |
3693 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
3694 | if (port | |
3695 | && port->up.opp.config & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP) && | |
3696 | port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
3697 | ? htonl(OFPPC_NO_RECV_STP) | |
3698 | : htonl(OFPPC_NO_RECV))) { | |
3699 | /* Drop this flow. */ | |
3700 | return; | |
3701 | } | |
3702 | ||
b4b8c781 | 3703 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 3704 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
3705 | const struct nx_action_resubmit *nar; |
3706 | const struct nx_action_set_tunnel *nast; | |
3707 | const struct nx_action_set_queue *nasq; | |
3708 | const struct nx_action_multipath *nam; | |
3709 | const struct nx_action_autopath *naa; | |
daff3353 | 3710 | const struct nx_action_bundle *nab; |
f694937d | 3711 | const struct nx_action_output_reg *naor; |
38f2e360 BP |
3712 | enum ofputil_action_code code; |
3713 | ovs_be64 tun_id; | |
3714 | ||
3715 | code = ofputil_decode_action_unsafe(ia); | |
3716 | switch (code) { | |
3717 | case OFPUTIL_OFPAT_OUTPUT: | |
abe529af BP |
3718 | xlate_output_action(ctx, &ia->output); |
3719 | break; | |
3720 | ||
38f2e360 | 3721 | case OFPUTIL_OFPAT_SET_VLAN_VID: |
abe529af BP |
3722 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
3723 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
3724 | break; |
3725 | ||
38f2e360 | 3726 | case OFPUTIL_OFPAT_SET_VLAN_PCP: |
abe529af BP |
3727 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
3728 | ctx->flow.vlan_tci |= htons( | |
3729 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
3730 | break; |
3731 | ||
38f2e360 | 3732 | case OFPUTIL_OFPAT_STRIP_VLAN: |
abe529af | 3733 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
3734 | break; |
3735 | ||
38f2e360 | 3736 | case OFPUTIL_OFPAT_SET_DL_SRC: |
abe529af | 3737 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
3738 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
3739 | break; | |
3740 | ||
38f2e360 | 3741 | case OFPUTIL_OFPAT_SET_DL_DST: |
abe529af | 3742 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
3743 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
3744 | break; | |
3745 | ||
38f2e360 | 3746 | case OFPUTIL_OFPAT_SET_NW_SRC: |
abe529af BP |
3747 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
3748 | break; | |
3749 | ||
38f2e360 | 3750 | case OFPUTIL_OFPAT_SET_NW_DST: |
abe529af BP |
3751 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
3752 | break; | |
3753 | ||
38f2e360 | 3754 | case OFPUTIL_OFPAT_SET_NW_TOS: |
1fa96cf4 | 3755 | ctx->flow.nw_tos = ia->nw_tos.nw_tos & IP_DSCP_MASK; |
abe529af BP |
3756 | break; |
3757 | ||
38f2e360 | 3758 | case OFPUTIL_OFPAT_SET_TP_SRC: |
abe529af BP |
3759 | ctx->flow.tp_src = ia->tp_port.tp_port; |
3760 | break; | |
3761 | ||
38f2e360 | 3762 | case OFPUTIL_OFPAT_SET_TP_DST: |
abe529af BP |
3763 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
3764 | break; | |
3765 | ||
38f2e360 BP |
3766 | case OFPUTIL_OFPAT_ENQUEUE: |
3767 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
3768 | break; | |
3769 | ||
3770 | case OFPUTIL_NXAST_RESUBMIT: | |
3771 | nar = (const struct nx_action_resubmit *) ia; | |
29901626 BP |
3772 | xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); |
3773 | break; | |
3774 | ||
3775 | case OFPUTIL_NXAST_RESUBMIT_TABLE: | |
3776 | xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); | |
abe529af BP |
3777 | break; |
3778 | ||
38f2e360 BP |
3779 | case OFPUTIL_NXAST_SET_TUNNEL: |
3780 | nast = (const struct nx_action_set_tunnel *) ia; | |
3781 | tun_id = htonll(ntohl(nast->tun_id)); | |
3782 | ctx->flow.tun_id = tun_id; | |
3783 | break; | |
3784 | ||
3785 | case OFPUTIL_NXAST_SET_QUEUE: | |
3786 | nasq = (const struct nx_action_set_queue *) ia; | |
3787 | xlate_set_queue_action(ctx, nasq); | |
3788 | break; | |
3789 | ||
3790 | case OFPUTIL_NXAST_POP_QUEUE: | |
3791 | ctx->priority = 0; | |
3792 | break; | |
3793 | ||
3794 | case OFPUTIL_NXAST_REG_MOVE: | |
3795 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
3796 | &ctx->flow); | |
3797 | break; | |
3798 | ||
3799 | case OFPUTIL_NXAST_REG_LOAD: | |
3800 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
3801 | &ctx->flow); | |
3802 | break; | |
3803 | ||
3804 | case OFPUTIL_NXAST_NOTE: | |
3805 | /* Nothing to do. */ | |
3806 | break; | |
3807 | ||
3808 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
3809 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
3810 | ctx->flow.tun_id = tun_id; | |
3811 | break; | |
3812 | ||
3813 | case OFPUTIL_NXAST_MULTIPATH: | |
3814 | nam = (const struct nx_action_multipath *) ia; | |
3815 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
3816 | break; |
3817 | ||
38f2e360 BP |
3818 | case OFPUTIL_NXAST_AUTOPATH: |
3819 | naa = (const struct nx_action_autopath *) ia; | |
3820 | xlate_autopath(ctx, naa); | |
abe529af | 3821 | break; |
daff3353 EJ |
3822 | |
3823 | case OFPUTIL_NXAST_BUNDLE: | |
3824 | ctx->ofproto->has_bundle_action = true; | |
3825 | nab = (const struct nx_action_bundle *) ia; | |
3826 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
3827 | slave_enabled_cb, | |
3828 | ctx->ofproto), 0); | |
3829 | break; | |
a368bb53 EJ |
3830 | |
3831 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
3832 | ctx->ofproto->has_bundle_action = true; | |
3833 | nab = (const struct nx_action_bundle *) ia; | |
3834 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
3835 | ctx->ofproto); | |
3836 | break; | |
f694937d EJ |
3837 | |
3838 | case OFPUTIL_NXAST_OUTPUT_REG: | |
3839 | naor = (const struct nx_action_output_reg *) ia; | |
3840 | xlate_output_reg_action(ctx, naor); | |
3841 | break; | |
75a75043 BP |
3842 | |
3843 | case OFPUTIL_NXAST_LEARN: | |
3844 | ctx->has_learn = true; | |
3845 | if (ctx->may_learn) { | |
3846 | xlate_learn_action(ctx, (const struct nx_action_learn *) ia); | |
3847 | } | |
3848 | break; | |
abe529af BP |
3849 | } |
3850 | } | |
3851 | } | |
3852 | ||
3853 | static void | |
3854 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
3855 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
3856 | const struct ofpbuf *packet) | |
3857 | { | |
3858 | ctx->ofproto = ofproto; | |
3859 | ctx->flow = *flow; | |
3860 | ctx->packet = packet; | |
75a75043 | 3861 | ctx->may_learn = packet != NULL; |
abe529af | 3862 | ctx->resubmit_hook = NULL; |
abe529af BP |
3863 | } |
3864 | ||
3865 | static struct ofpbuf * | |
3866 | xlate_actions(struct action_xlate_ctx *ctx, | |
3867 | const union ofp_action *in, size_t n_in) | |
3868 | { | |
3869 | COVERAGE_INC(ofproto_dpif_xlate); | |
3870 | ||
3871 | ctx->odp_actions = ofpbuf_new(512); | |
b6848f13 | 3872 | ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE); |
abe529af BP |
3873 | ctx->tags = 0; |
3874 | ctx->may_set_up_flow = true; | |
75a75043 BP |
3875 | ctx->has_learn = false; |
3876 | ctx->has_normal = false; | |
abe529af BP |
3877 | ctx->nf_output_iface = NF_OUT_DROP; |
3878 | ctx->recurse = 0; | |
afabef2b | 3879 | ctx->priority = 0; |
b3e9b2ed EJ |
3880 | ctx->base_priority = 0; |
3881 | ctx->base_flow = ctx->flow; | |
bf1e0371 | 3882 | ctx->base_flow.tun_id = 0; |
29901626 | 3883 | ctx->table_id = 0; |
abe529af | 3884 | |
fc08b7a2 | 3885 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af | 3886 | ctx->may_set_up_flow = false; |
b6848f13 | 3887 | return ctx->odp_actions; |
abe529af | 3888 | } else { |
6ff686f2 | 3889 | add_sflow_action(ctx); |
abe529af | 3890 | do_xlate_actions(in, n_in, ctx); |
abe529af | 3891 | |
b6848f13 BP |
3892 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
3893 | ctx->odp_actions->data, | |
3894 | ctx->odp_actions->size)) { | |
3895 | ctx->may_set_up_flow = false; | |
3896 | if (ctx->packet | |
3897 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
3898 | ctx->packet)) { | |
a7c4eaf6 | 3899 | compose_output_action(ctx, OVSP_LOCAL); |
b6848f13 BP |
3900 | } |
3901 | } | |
a7c4eaf6 | 3902 | fix_sflow_action(ctx); |
abe529af BP |
3903 | } |
3904 | ||
3905 | return ctx->odp_actions; | |
3906 | } | |
3907 | \f | |
3908 | /* OFPP_NORMAL implementation. */ | |
3909 | ||
3910 | struct dst { | |
3911 | struct ofport_dpif *port; | |
dc155bff | 3912 | uint16_t vid; |
abe529af BP |
3913 | }; |
3914 | ||
3915 | struct dst_set { | |
3916 | struct dst builtin[32]; | |
3917 | struct dst *dsts; | |
3918 | size_t n, allocated; | |
3919 | }; | |
3920 | ||
3921 | static void dst_set_init(struct dst_set *); | |
3922 | static void dst_set_add(struct dst_set *, const struct dst *); | |
3923 | static void dst_set_free(struct dst_set *); | |
3924 | ||
3925 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); | |
3926 | ||
ecac4ebf BP |
3927 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
3928 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
3929 | * the bundle on which the packet was received, returns the VLAN to which the | |
3930 | * packet belongs. | |
3931 | * | |
3932 | * Both 'vid' and the return value are in the range 0...4095. */ | |
3933 | static uint16_t | |
3934 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
3935 | { | |
3936 | switch (in_bundle->vlan_mode) { | |
3937 | case PORT_VLAN_ACCESS: | |
3938 | return in_bundle->vlan; | |
3939 | break; | |
3940 | ||
3941 | case PORT_VLAN_TRUNK: | |
3942 | return vid; | |
3943 | ||
3944 | case PORT_VLAN_NATIVE_UNTAGGED: | |
3945 | case PORT_VLAN_NATIVE_TAGGED: | |
3946 | return vid ? vid : in_bundle->vlan; | |
3947 | ||
3948 | default: | |
3949 | NOT_REACHED(); | |
3950 | } | |
3951 | } | |
3952 | ||
3953 | /* Given 'vlan', the VLAN that a packet belongs to, and | |
3954 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
3955 | * that should be included in the 802.1Q header. (If the return value is 0, | |
3956 | * then the 802.1Q header should only be included in the packet if there is a | |
3957 | * nonzero PCP.) | |
3958 | * | |
3959 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
3960 | static uint16_t | |
3961 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
3962 | { | |
3963 | switch (out_bundle->vlan_mode) { | |
3964 | case PORT_VLAN_ACCESS: | |
3965 | return 0; | |
3966 | ||
3967 | case PORT_VLAN_TRUNK: | |
3968 | case PORT_VLAN_NATIVE_TAGGED: | |
3969 | return vlan; | |
3970 | ||
3971 | case PORT_VLAN_NATIVE_UNTAGGED: | |
3972 | return vlan == out_bundle->vlan ? 0 : vlan; | |
3973 | ||
3974 | default: | |
3975 | NOT_REACHED(); | |
3976 | } | |
3977 | } | |
3978 | ||
abe529af BP |
3979 | static bool |
3980 | set_dst(struct action_xlate_ctx *ctx, struct dst *dst, | |
3981 | const struct ofbundle *in_bundle, const struct ofbundle *out_bundle) | |
3982 | { | |
ecac4ebf BP |
3983 | uint16_t vlan; |
3984 | ||
3985 | vlan = input_vid_to_vlan(in_bundle, vlan_tci_to_vid(ctx->flow.vlan_tci)); | |
3986 | dst->vid = output_vlan_to_vid(out_bundle, vlan); | |
abe529af BP |
3987 | |
3988 | dst->port = (!out_bundle->bond | |
3989 | ? ofbundle_get_a_port(out_bundle) | |
3990 | : bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
dc155bff | 3991 | dst->vid, &ctx->tags)); |
abe529af BP |
3992 | return dst->port != NULL; |
3993 | } | |
3994 | ||
3995 | static int | |
3996 | mirror_mask_ffs(mirror_mask_t mask) | |
3997 | { | |
3998 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
3999 | return ffs(mask); | |
4000 | } | |
4001 | ||
4002 | static void | |
4003 | dst_set_init(struct dst_set *set) | |
4004 | { | |
4005 | set->dsts = set->builtin; | |
4006 | set->n = 0; | |
4007 | set->allocated = ARRAY_SIZE(set->builtin); | |
4008 | } | |
4009 | ||
4010 | static void | |
4011 | dst_set_add(struct dst_set *set, const struct dst *dst) | |
4012 | { | |
4013 | if (set->n >= set->allocated) { | |
4014 | size_t new_allocated; | |
4015 | struct dst *new_dsts; | |
4016 | ||
4017 | new_allocated = set->allocated * 2; | |
4018 | new_dsts = xmalloc(new_allocated * sizeof *new_dsts); | |
4019 | memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); | |
4020 | ||
4021 | dst_set_free(set); | |
4022 | ||
4023 | set->dsts = new_dsts; | |
4024 | set->allocated = new_allocated; | |
4025 | } | |
4026 | set->dsts[set->n++] = *dst; | |
4027 | } | |
4028 | ||
4029 | static void | |
4030 | dst_set_free(struct dst_set *set) | |
4031 | { | |
4032 | if (set->dsts != set->builtin) { | |
4033 | free(set->dsts); | |
4034 | } | |
4035 | } | |
4036 | ||
4037 | static bool | |
4038 | dst_is_duplicate(const struct dst_set *set, const struct dst *test) | |
4039 | { | |
4040 | size_t i; | |
4041 | for (i = 0; i < set->n; i++) { | |
dc155bff | 4042 | if (set->dsts[i].vid == test->vid |
abe529af BP |
4043 | && set->dsts[i].port == test->port) { |
4044 | return true; | |
4045 | } | |
4046 | } | |
4047 | return false; | |
4048 | } | |
4049 | ||
4050 | static bool | |
4051 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
4052 | { | |
ecac4ebf | 4053 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 4054 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
4055 | } |
4056 | ||
4057 | static bool | |
4058 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
4059 | { | |
4060 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
4061 | } | |
4062 | ||
4063 | /* Returns an arbitrary interface within 'bundle'. */ | |
4064 | static struct ofport_dpif * | |
4065 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
4066 | { | |
4067 | return CONTAINER_OF(list_front(&bundle->ports), | |
4068 | struct ofport_dpif, bundle_node); | |
4069 | } | |
4070 | ||
4071 | static void | |
4072 | compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan, | |
4073 | const struct ofbundle *in_bundle, | |
4074 | const struct ofbundle *out_bundle, struct dst_set *set) | |
4075 | { | |
4076 | struct dst dst; | |
4077 | ||
4078 | if (out_bundle == OFBUNDLE_FLOOD) { | |
4079 | struct ofbundle *bundle; | |
4080 | ||
4081 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { | |
4082 | if (bundle != in_bundle | |
4083 | && ofbundle_includes_vlan(bundle, vlan) | |
4084 | && bundle->floodable | |
4085 | && !bundle->mirror_out | |
4086 | && set_dst(ctx, &dst, in_bundle, bundle)) { | |
4087 | dst_set_add(set, &dst); | |
4088 | } | |
4089 | } | |
4090 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
4091 | } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) { | |
4092 | dst_set_add(set, &dst); | |
4093 | ctx->nf_output_iface = dst.port->odp_port; | |
4094 | } | |
4095 | } | |
4096 | ||
4097 | static bool | |
4098 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
4099 | { | |
fc3d7408 | 4100 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
4101 | } |
4102 | ||
07817dfe BP |
4103 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
4104 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
4105 | * protocols that may confuse switches. */ | |
4106 | static bool | |
4107 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
4108 | { | |
4109 | /* If you change this function's behavior, please update corresponding | |
4110 | * documentation in vswitch.xml at the same time. */ | |
4111 | if (dst[0] != 0x01) { | |
4112 | /* All the currently banned MACs happen to start with 01 currently, so | |
4113 | * this is a quick way to eliminate most of the good ones. */ | |
4114 | } else { | |
4115 | if (eth_addr_is_reserved(dst)) { | |
4116 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
4117 | * (01-80-c2-00-00-0x). */ | |
4118 | return false; | |
4119 | } | |
4120 | ||
4121 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
4122 | /* Cisco OUI. */ | |
4123 | if ((dst[3] & 0xfe) == 0xcc && | |
4124 | (dst[4] & 0xfe) == 0xcc && | |
4125 | (dst[5] & 0xfe) == 0xcc) { | |
4126 | /* Drop the following protocols plus others following the same | |
4127 | pattern: | |
4128 | ||
4129 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
4130 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
4131 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
4132 | return false; | |
4133 | } | |
4134 | ||
4135 | if (!(dst[3] | dst[4] | dst[5])) { | |
4136 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
4137 | return false; | |
4138 | } | |
4139 | } | |
4140 | } | |
4141 | return true; | |
4142 | } | |
4143 | ||
abe529af BP |
4144 | static void |
4145 | compose_mirror_dsts(struct action_xlate_ctx *ctx, | |
4146 | uint16_t vlan, const struct ofbundle *in_bundle, | |
4147 | struct dst_set *set) | |
4148 | { | |
4149 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
4150 | mirror_mask_t mirrors; | |
dc155bff | 4151 | uint16_t flow_vid; |
abe529af BP |
4152 | size_t i; |
4153 | ||
4154 | mirrors = in_bundle->src_mirrors; | |
4155 | for (i = 0; i < set->n; i++) { | |
4156 | mirrors |= set->dsts[i].port->bundle->dst_mirrors; | |
4157 | } | |
4158 | ||
4159 | if (!mirrors) { | |
4160 | return; | |
4161 | } | |
4162 | ||
dc155bff | 4163 | flow_vid = vlan_tci_to_vid(ctx->flow.vlan_tci); |
abe529af BP |
4164 | while (mirrors) { |
4165 | struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
4166 | if (vlan_is_mirrored(m, vlan)) { | |
4167 | struct dst dst; | |
4168 | ||
4169 | if (m->out) { | |
4170 | if (set_dst(ctx, &dst, in_bundle, m->out) | |
4171 | && !dst_is_duplicate(set, &dst)) { | |
4172 | dst_set_add(set, &dst); | |
4173 | } | |
07817dfe | 4174 | } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) { |
abe529af BP |
4175 | struct ofbundle *bundle; |
4176 | ||
4177 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
4178 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
4179 | && set_dst(ctx, &dst, in_bundle, bundle)) | |
4180 | { | |
ecac4ebf BP |
4181 | /* set_dst() got dst->vid from the input packet's VLAN, |
4182 | * not from m->out_vlan, so recompute it. */ | |
4183 | dst.vid = output_vlan_to_vid(bundle, m->out_vlan); | |
4184 | ||
abe529af BP |
4185 | if (dst_is_duplicate(set, &dst)) { |
4186 | continue; | |
4187 | } | |
4188 | ||
dc155bff | 4189 | if (bundle == in_bundle && dst.vid == flow_vid) { |
abe529af BP |
4190 | /* Don't send out input port on same VLAN. */ |
4191 | continue; | |
4192 | } | |
4193 | dst_set_add(set, &dst); | |
4194 | } | |
4195 | } | |
4196 | } | |
4197 | } | |
4198 | mirrors &= mirrors - 1; | |
4199 | } | |
4200 | } | |
4201 | ||
4202 | static void | |
4203 | compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan, | |
4204 | const struct ofbundle *in_bundle, | |
4205 | const struct ofbundle *out_bundle) | |
4206 | { | |
dc155bff | 4207 | uint16_t initial_vid, cur_vid; |
abe529af BP |
4208 | const struct dst *dst; |
4209 | struct dst_set set; | |
4210 | ||
4211 | dst_set_init(&set); | |
4212 | compose_dsts(ctx, vlan, in_bundle, out_bundle, &set); | |
4213 | compose_mirror_dsts(ctx, vlan, in_bundle, &set); | |
823518f1 BP |
4214 | if (!set.n) { |
4215 | dst_set_free(&set); | |
4216 | return; | |
4217 | } | |
abe529af BP |
4218 | |
4219 | /* Output all the packets we can without having to change the VLAN. */ | |
823518f1 | 4220 | commit_odp_actions(ctx); |
dc155bff | 4221 | initial_vid = vlan_tci_to_vid(ctx->flow.vlan_tci); |
abe529af | 4222 | for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { |
dc155bff | 4223 | if (dst->vid != initial_vid) { |
abe529af BP |
4224 | continue; |
4225 | } | |
6ff686f2 | 4226 | compose_output_action(ctx, dst->port->odp_port); |
abe529af BP |
4227 | } |
4228 | ||
4229 | /* Then output the rest. */ | |
dc155bff | 4230 | cur_vid = initial_vid; |
abe529af | 4231 | for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { |
dc155bff | 4232 | if (dst->vid == initial_vid) { |
abe529af BP |
4233 | continue; |
4234 | } | |
dc155bff | 4235 | if (dst->vid != cur_vid) { |
823518f1 | 4236 | ovs_be16 tci; |
d9065a90 | 4237 | |
dc155bff | 4238 | tci = htons(dst->vid); |
823518f1 BP |
4239 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); |
4240 | if (tci) { | |
4241 | tci |= htons(VLAN_CFI); | |
abe529af | 4242 | } |
823518f1 BP |
4243 | commit_vlan_tci(ctx, tci); |
4244 | ||
dc155bff | 4245 | cur_vid = dst->vid; |
abe529af | 4246 | } |
6ff686f2 | 4247 | compose_output_action(ctx, dst->port->odp_port); |
abe529af BP |
4248 | } |
4249 | ||
4250 | dst_set_free(&set); | |
4251 | } | |
4252 | ||
4253 | /* Returns the effective vlan of a packet, taking into account both the | |
4254 | * 802.1Q header and implicitly tagged ports. A value of 0 indicates that | |
4255 | * the packet is untagged and -1 indicates it has an invalid header and | |
4256 | * should be dropped. */ | |
4257 | static int | |
4258 | flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4259 | struct ofbundle *in_bundle, bool have_packet) | |
4260 | { | |
4261 | int vlan = vlan_tci_to_vid(flow->vlan_tci); | |
ecac4ebf BP |
4262 | if (vlan) { |
4263 | if (in_bundle->vlan_mode == PORT_VLAN_ACCESS) { | |
4264 | /* Drop tagged packet on access port */ | |
abe529af BP |
4265 | if (have_packet) { |
4266 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4267 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
4268 | "packet received on port %s configured with " | |
4269 | "implicit VLAN %"PRIu16, | |
4270 | ofproto->up.name, vlan, | |
4271 | in_bundle->name, in_bundle->vlan); | |
4272 | } | |
4273 | return -1; | |
ecac4ebf BP |
4274 | } else if (ofbundle_includes_vlan(in_bundle, vlan)) { |
4275 | return vlan; | |
4276 | } else { | |
4277 | /* Drop packets from a VLAN not member of the trunk */ | |
abe529af BP |
4278 | if (have_packet) { |
4279 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4280 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
4281 | "packet received on port %s not configured for " | |
4282 | "trunking VLAN %d", | |
4283 | ofproto->up.name, vlan, in_bundle->name, vlan); | |
4284 | } | |
4285 | return -1; | |
4286 | } | |
ecac4ebf BP |
4287 | } else { |
4288 | if (in_bundle->vlan_mode != PORT_VLAN_TRUNK) { | |
4289 | return in_bundle->vlan; | |
4290 | } else { | |
4291 | return ofbundle_includes_vlan(in_bundle, 0) ? 0 : -1; | |
4292 | } | |
abe529af | 4293 | } |
abe529af BP |
4294 | } |
4295 | ||
4296 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after | |
4297 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
4298 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
4299 | static bool | |
4300 | is_gratuitous_arp(const struct flow *flow) | |
4301 | { | |
4302 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
4303 | && eth_addr_is_broadcast(flow->dl_dst) | |
4304 | && (flow->nw_proto == ARP_OP_REPLY | |
4305 | || (flow->nw_proto == ARP_OP_REQUEST | |
4306 | && flow->nw_src == flow->nw_dst))); | |
4307 | } | |
4308 | ||
4309 | static void | |
4310 | update_learning_table(struct ofproto_dpif *ofproto, | |
4311 | const struct flow *flow, int vlan, | |
4312 | struct ofbundle *in_bundle) | |
4313 | { | |
4314 | struct mac_entry *mac; | |
4315 | ||
4316 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { | |
4317 | return; | |
4318 | } | |
4319 | ||
4320 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
4321 | if (is_gratuitous_arp(flow)) { | |
4322 | /* We don't want to learn from gratuitous ARP packets that are | |
4323 | * reflected back over bond slaves so we lock the learning table. */ | |
4324 | if (!in_bundle->bond) { | |
4325 | mac_entry_set_grat_arp_lock(mac); | |
4326 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
4327 | return; | |
4328 | } | |
4329 | } | |
4330 | ||
4331 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
4332 | /* The log messages here could actually be useful in debugging, | |
4333 | * so keep the rate limit relatively high. */ | |
4334 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
4335 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
4336 | "on port %s in VLAN %d", | |
4337 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
4338 | in_bundle->name, vlan); | |
4339 | ||
4340 | mac->port.p = in_bundle; | |
4341 | tag_set_add(&ofproto->revalidate_set, | |
4342 | mac_learning_changed(ofproto->ml, mac)); | |
4343 | } | |
4344 | } | |
4345 | ||
4346 | /* Determines whether packets in 'flow' within 'br' should be forwarded or | |
4347 | * dropped. Returns true if they may be forwarded, false if they should be | |
4348 | * dropped. | |
4349 | * | |
4350 | * If 'have_packet' is true, it indicates that the caller is processing a | |
4351 | * received packet. If 'have_packet' is false, then the caller is just | |
4352 | * revalidating an existing flow because configuration has changed. Either | |
4353 | * way, 'have_packet' only affects logging (there is no point in logging errors | |
4354 | * during revalidation). | |
4355 | * | |
4356 | * Sets '*in_portp' to the input port. This will be a null pointer if | |
4357 | * flow->in_port does not designate a known input port (in which case | |
4358 | * is_admissible() returns false). | |
4359 | * | |
4360 | * When returning true, sets '*vlanp' to the effective VLAN of the input | |
4361 | * packet, as returned by flow_get_vlan(). | |
4362 | * | |
4363 | * May also add tags to '*tags', although the current implementation only does | |
4364 | * so in one special case. | |
4365 | */ | |
4366 | static bool | |
4367 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4368 | bool have_packet, | |
4369 | tag_type *tags, int *vlanp, struct ofbundle **in_bundlep) | |
4370 | { | |
4371 | struct ofport_dpif *in_port; | |
4372 | struct ofbundle *in_bundle; | |
4373 | int vlan; | |
4374 | ||
4375 | /* Find the port and bundle for the received packet. */ | |
4376 | in_port = get_ofp_port(ofproto, flow->in_port); | |
23adee42 | 4377 | *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL; |
abe529af BP |
4378 | if (!in_port || !in_bundle) { |
4379 | /* No interface? Something fishy... */ | |
4380 | if (have_packet) { | |
4381 | /* Odd. A few possible reasons here: | |
4382 | * | |
4383 | * - We deleted a port but there are still a few packets queued up | |
4384 | * from it. | |
4385 | * | |
4386 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
4387 | * we don't know about. | |
4388 | * | |
4389 | * - Packet arrived on the local port but the local port is not | |
4390 | * part of a bundle. | |
4391 | */ | |
4392 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4393 | ||
4394 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
4395 | "port %"PRIu16, | |
4396 | ofproto->up.name, flow->in_port); | |
4397 | } | |
75a75043 | 4398 | *vlanp = -1; |
abe529af BP |
4399 | return false; |
4400 | } | |
4401 | *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet); | |
4402 | if (vlan < 0) { | |
4403 | return false; | |
4404 | } | |
4405 | ||
b53055f4 | 4406 | /* Drop frames for reserved multicast addresses |
8402c74b | 4407 | * only if forward_bpdu option is absent. */ |
b53055f4 | 4408 | if (eth_addr_is_reserved(flow->dl_dst) && |
8402c74b | 4409 | !ofproto->up.forward_bpdu) { |
abe529af BP |
4410 | return false; |
4411 | } | |
4412 | ||
4413 | /* Drop frames on bundles reserved for mirroring. */ | |
4414 | if (in_bundle->mirror_out) { | |
4415 | if (have_packet) { | |
4416 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4417 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
4418 | "%s, which is reserved exclusively for mirroring", | |
4419 | ofproto->up.name, in_bundle->name); | |
4420 | } | |
4421 | return false; | |
4422 | } | |
4423 | ||
4424 | if (in_bundle->bond) { | |
4425 | struct mac_entry *mac; | |
4426 | ||
4427 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
4428 | flow->dl_dst, tags)) { | |
4429 | case BV_ACCEPT: | |
4430 | break; | |
4431 | ||
4432 | case BV_DROP: | |
4433 | return false; | |
4434 | ||
4435 | case BV_DROP_IF_MOVED: | |
4436 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
4437 | if (mac && mac->port.p != in_bundle && | |
4438 | (!is_gratuitous_arp(flow) | |
4439 | || mac_entry_is_grat_arp_locked(mac))) { | |
4440 | return false; | |
4441 | } | |
4442 | break; | |
4443 | } | |
4444 | } | |
4445 | ||
4446 | return true; | |
4447 | } | |
4448 | ||
4cd78906 | 4449 | static void |
abe529af BP |
4450 | xlate_normal(struct action_xlate_ctx *ctx) |
4451 | { | |
4452 | struct ofbundle *in_bundle; | |
4453 | struct ofbundle *out_bundle; | |
4454 | struct mac_entry *mac; | |
4455 | int vlan; | |
4456 | ||
75a75043 BP |
4457 | ctx->has_normal = true; |
4458 | ||
abe529af BP |
4459 | /* Check whether we should drop packets in this flow. */ |
4460 | if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL, | |
4461 | &ctx->tags, &vlan, &in_bundle)) { | |
4462 | out_bundle = NULL; | |
4463 | goto done; | |
4464 | } | |
4465 | ||
75a75043 BP |
4466 | /* Learn source MAC. */ |
4467 | if (ctx->may_learn) { | |
abe529af BP |
4468 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
4469 | } | |
4470 | ||
4471 | /* Determine output bundle. */ | |
4472 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
4473 | &ctx->tags); | |
4474 | if (mac) { | |
4475 | out_bundle = mac->port.p; | |
4476 | } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) { | |
4477 | /* If we are revalidating but don't have a learning entry then eject | |
4478 | * the flow. Installing a flow that floods packets opens up a window | |
4479 | * of time where we could learn from a packet reflected on a bond and | |
4480 | * blackhole packets before the learning table is updated to reflect | |
4481 | * the correct port. */ | |
4cd78906 BP |
4482 | ctx->may_set_up_flow = false; |
4483 | return; | |
abe529af BP |
4484 | } else { |
4485 | out_bundle = OFBUNDLE_FLOOD; | |
4486 | } | |
4487 | ||
4488 | /* Don't send packets out their input bundles. */ | |
4489 | if (in_bundle == out_bundle) { | |
4490 | out_bundle = NULL; | |
4491 | } | |
4492 | ||
4493 | done: | |
4494 | if (in_bundle) { | |
4495 | compose_actions(ctx, vlan, in_bundle, out_bundle); | |
4496 | } | |
abe529af BP |
4497 | } |
4498 | \f | |
54a9cbc9 BP |
4499 | /* Optimized flow revalidation. |
4500 | * | |
4501 | * It's a difficult problem, in general, to tell which facets need to have | |
4502 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
4503 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
4504 | * table changes, we recalculate the actions for every facet. This is | |
4505 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
4506 | * doesn't change very often. | |
4507 | * | |
4508 | * However, we can expect one particular kind of OpenFlow flow table change to | |
4509 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
4510 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
4511 | * table, we add a special case that applies to flow tables in which every rule | |
4512 | * has the same form (that is, the same wildcards), except that the table is | |
4513 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
4514 | * optimize this case by tagging all of the facets that resubmit into the table | |
4515 | * and invalidating the same tag whenever a flow changes in that table. The | |
4516 | * end result is that we revalidate just the facets that need it (and sometimes | |
4517 | * a few more, but not all of the facets or even all of the facets that | |
4518 | * resubmit to the table modified by MAC learning). */ | |
4519 | ||
4520 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
4521 | * into an OpenFlow table with the given 'basis'. */ | |
4522 | static uint32_t | |
4523 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, | |
4524 | uint32_t secret) | |
4525 | { | |
4526 | if (flow_wildcards_is_catchall(wc)) { | |
4527 | return 0; | |
4528 | } else { | |
4529 | struct flow tag_flow = *flow; | |
4530 | flow_zero_wildcards(&tag_flow, wc); | |
4531 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
4532 | } | |
4533 | } | |
4534 | ||
4535 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
4536 | * taggability of that table. | |
4537 | * | |
4538 | * This function must be called after *each* change to a flow table. If you | |
4539 | * skip calling it on some changes then the pointer comparisons at the end can | |
4540 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
4541 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
4542 | * different wildcards to be created with the same address, then this function | |
4543 | * will incorrectly skip revalidation. */ | |
4544 | static void | |
4545 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
4546 | { | |
4547 | struct table_dpif *table = &ofproto->tables[table_id]; | |
4548 | const struct classifier *cls = &ofproto->up.tables[table_id]; | |
4549 | struct cls_table *catchall, *other; | |
4550 | struct cls_table *t; | |
4551 | ||
4552 | catchall = other = NULL; | |
4553 | ||
4554 | switch (hmap_count(&cls->tables)) { | |
4555 | case 0: | |
4556 | /* We could tag this OpenFlow table but it would make the logic a | |
4557 | * little harder and it's a corner case that doesn't seem worth it | |
4558 | * yet. */ | |
4559 | break; | |
4560 | ||
4561 | case 1: | |
4562 | case 2: | |
4563 | HMAP_FOR_EACH (t, hmap_node, &cls->tables) { | |
4564 | if (cls_table_is_catchall(t)) { | |
4565 | catchall = t; | |
4566 | } else if (!other) { | |
4567 | other = t; | |
4568 | } else { | |
4569 | /* Indicate that we can't tag this by setting both tables to | |
4570 | * NULL. (We know that 'catchall' is already NULL.) */ | |
4571 | other = NULL; | |
4572 | } | |
4573 | } | |
4574 | break; | |
4575 | ||
4576 | default: | |
4577 | /* Can't tag this table. */ | |
4578 | break; | |
4579 | } | |
4580 | ||
4581 | if (table->catchall_table != catchall || table->other_table != other) { | |
4582 | table->catchall_table = catchall; | |
4583 | table->other_table = other; | |
4584 | ofproto->need_revalidate = true; | |
4585 | } | |
4586 | } | |
4587 | ||
4588 | /* Given 'rule' that has changed in some way (either it is a rule being | |
4589 | * inserted, a rule being deleted, or a rule whose actions are being | |
4590 | * modified), marks facets for revalidation to ensure that packets will be | |
4591 | * forwarded correctly according to the new state of the flow table. | |
4592 | * | |
4593 | * This function must be called after *each* change to a flow table. See | |
4594 | * the comment on table_update_taggable() for more information. */ | |
4595 | static void | |
4596 | rule_invalidate(const struct rule_dpif *rule) | |
4597 | { | |
4598 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4599 | ||
4600 | table_update_taggable(ofproto, rule->up.table_id); | |
4601 | ||
4602 | if (!ofproto->need_revalidate) { | |
4603 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
4604 | ||
4605 | if (table->other_table && rule->tag) { | |
4606 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
4607 | } else { | |
4608 | ofproto->need_revalidate = true; | |
4609 | } | |
4610 | } | |
4611 | } | |
4612 | \f | |
abe529af BP |
4613 | static bool |
4614 | get_drop_frags(struct ofproto *ofproto_) | |
4615 | { | |
4616 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
4617 | bool drop_frags; | |
4618 | ||
4619 | dpif_get_drop_frags(ofproto->dpif, &drop_frags); | |
4620 | return drop_frags; | |
4621 | } | |
4622 | ||
4623 | static void | |
4624 | set_drop_frags(struct ofproto *ofproto_, bool drop_frags) | |
4625 | { | |
4626 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
4627 | ||
4628 | dpif_set_drop_frags(ofproto->dpif, drop_frags); | |
4629 | } | |
4630 | ||
4631 | static int | |
4632 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, | |
4633 | const struct flow *flow, | |
4634 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
4635 | { | |
4636 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
4637 | int error; | |
4638 | ||
4639 | error = validate_actions(ofp_actions, n_ofp_actions, flow, | |
4640 | ofproto->max_ports); | |
4641 | if (!error) { | |
80e5eed9 | 4642 | struct odputil_keybuf keybuf; |
abe529af BP |
4643 | struct action_xlate_ctx ctx; |
4644 | struct ofpbuf *odp_actions; | |
80e5eed9 BP |
4645 | struct ofpbuf key; |
4646 | ||
4647 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
4648 | odp_flow_key_from_flow(&key, flow); | |
abe529af BP |
4649 | |
4650 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
4651 | odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); | |
80e5eed9 BP |
4652 | dpif_execute(ofproto->dpif, key.data, key.size, |
4653 | odp_actions->data, odp_actions->size, packet); | |
abe529af BP |
4654 | ofpbuf_delete(odp_actions); |
4655 | } | |
4656 | return error; | |
4657 | } | |
4658 | ||
4659 | static void | |
4660 | get_netflow_ids(const struct ofproto *ofproto_, | |
4661 | uint8_t *engine_type, uint8_t *engine_id) | |
4662 | { | |
4663 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
4664 | ||
4665 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
4666 | } | |
4667 | \f | |
4668 | static struct ofproto_dpif * | |
4669 | ofproto_dpif_lookup(const char *name) | |
4670 | { | |
4671 | struct ofproto *ofproto = ofproto_lookup(name); | |
4672 | return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class | |
4673 | ? ofproto_dpif_cast(ofproto) | |
4674 | : NULL); | |
4675 | } | |
4676 | ||
4677 | static void | |
4678 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, | |
4679 | const char *args, void *aux OVS_UNUSED) | |
4680 | { | |
4681 | struct ds ds = DS_EMPTY_INITIALIZER; | |
4682 | const struct ofproto_dpif *ofproto; | |
4683 | const struct mac_entry *e; | |
4684 | ||
4685 | ofproto = ofproto_dpif_lookup(args); | |
4686 | if (!ofproto) { | |
4687 | unixctl_command_reply(conn, 501, "no such bridge"); | |
4688 | return; | |
4689 | } | |
4690 | ||
4691 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
4692 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
4693 | struct ofbundle *bundle = e->port.p; | |
4694 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
4695 | ofbundle_get_a_port(bundle)->odp_port, | |
4696 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
4697 | } | |
4698 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
4699 | ds_destroy(&ds); | |
4700 | } | |
4701 | ||
4702 | struct ofproto_trace { | |
4703 | struct action_xlate_ctx ctx; | |
4704 | struct flow flow; | |
4705 | struct ds *result; | |
4706 | }; | |
4707 | ||
4708 | static void | |
29901626 BP |
4709 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
4710 | const struct rule_dpif *rule) | |
abe529af BP |
4711 | { |
4712 | ds_put_char_multiple(result, '\t', level); | |
4713 | if (!rule) { | |
4714 | ds_put_cstr(result, "No match\n"); | |
4715 | return; | |
4716 | } | |
4717 | ||
29901626 BP |
4718 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
4719 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 4720 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
4721 | ds_put_char(result, '\n'); |
4722 | ||
4723 | ds_put_char_multiple(result, '\t', level); | |
4724 | ds_put_cstr(result, "OpenFlow "); | |
79feb7df | 4725 | ofp_print_actions(result, rule->up.actions, rule->up.n_actions); |
abe529af BP |
4726 | ds_put_char(result, '\n'); |
4727 | } | |
4728 | ||
4729 | static void | |
4730 | trace_format_flow(struct ds *result, int level, const char *title, | |
4731 | struct ofproto_trace *trace) | |
4732 | { | |
4733 | ds_put_char_multiple(result, '\t', level); | |
4734 | ds_put_format(result, "%s: ", title); | |
4735 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
4736 | ds_put_cstr(result, "unchanged"); | |
4737 | } else { | |
4738 | flow_format(result, &trace->ctx.flow); | |
4739 | trace->flow = trace->ctx.flow; | |
4740 | } | |
4741 | ds_put_char(result, '\n'); | |
4742 | } | |
4743 | ||
eb9e1c26 EJ |
4744 | static void |
4745 | trace_format_regs(struct ds *result, int level, const char *title, | |
4746 | struct ofproto_trace *trace) | |
4747 | { | |
4748 | size_t i; | |
4749 | ||
4750 | ds_put_char_multiple(result, '\t', level); | |
4751 | ds_put_format(result, "%s:", title); | |
4752 | for (i = 0; i < FLOW_N_REGS; i++) { | |
4753 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
4754 | } | |
4755 | ds_put_char(result, '\n'); | |
4756 | } | |
4757 | ||
abe529af BP |
4758 | static void |
4759 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
4760 | { | |
4761 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
4762 | struct ds *result = trace->result; | |
4763 | ||
4764 | ds_put_char(result, '\n'); | |
4765 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 4766 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
29901626 | 4767 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
4768 | } |
4769 | ||
4770 | static void | |
4771 | ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, | |
4772 | void *aux OVS_UNUSED) | |
4773 | { | |
876b0e1c | 4774 | char *dpname, *arg1, *arg2, *arg3; |
abe529af BP |
4775 | char *args = xstrdup(args_); |
4776 | char *save_ptr = NULL; | |
4777 | struct ofproto_dpif *ofproto; | |
876b0e1c BP |
4778 | struct ofpbuf odp_key; |
4779 | struct ofpbuf *packet; | |
abe529af BP |
4780 | struct rule_dpif *rule; |
4781 | struct ds result; | |
4782 | struct flow flow; | |
abe529af BP |
4783 | char *s; |
4784 | ||
876b0e1c BP |
4785 | packet = NULL; |
4786 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
4787 | ds_init(&result); |
4788 | ||
4789 | dpname = strtok_r(args, " ", &save_ptr); | |
876b0e1c BP |
4790 | arg1 = strtok_r(NULL, " ", &save_ptr); |
4791 | arg2 = strtok_r(NULL, " ", &save_ptr); | |
4792 | arg3 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ | |
8b3b8dd1 BP |
4793 | if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) { |
4794 | /* ofproto/trace dpname flow [-generate] */ | |
876b0e1c BP |
4795 | int error; |
4796 | ||
df2c07f4 | 4797 | /* Convert string to datapath key. */ |
876b0e1c BP |
4798 | ofpbuf_init(&odp_key, 0); |
4799 | error = odp_flow_key_from_string(arg1, &odp_key); | |
4800 | if (error) { | |
4801 | unixctl_command_reply(conn, 501, "Bad flow syntax"); | |
4802 | goto exit; | |
4803 | } | |
4804 | ||
4805 | /* Convert odp_key to flow. */ | |
4806 | error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow); | |
4807 | if (error) { | |
4808 | unixctl_command_reply(conn, 501, "Invalid flow"); | |
4809 | goto exit; | |
4810 | } | |
8b3b8dd1 BP |
4811 | |
4812 | /* Generate a packet, if requested. */ | |
4813 | if (arg2) { | |
4814 | packet = ofpbuf_new(0); | |
4815 | flow_compose(packet, &flow); | |
4816 | } | |
876b0e1c BP |
4817 | } else if (dpname && arg1 && arg2 && arg3) { |
4818 | /* ofproto/trace dpname tun_id in_port packet */ | |
4819 | uint16_t in_port; | |
4820 | ovs_be64 tun_id; | |
4821 | ||
4822 | tun_id = htonll(strtoull(arg1, NULL, 0)); | |
4823 | in_port = ofp_port_to_odp_port(atoi(arg2)); | |
4824 | ||
4825 | packet = ofpbuf_new(strlen(args) / 2); | |
4826 | arg3 = ofpbuf_put_hex(packet, arg3, NULL); | |
4827 | arg3 += strspn(arg3, " "); | |
4828 | if (*arg3 != '\0') { | |
4829 | unixctl_command_reply(conn, 501, "Trailing garbage in command"); | |
4830 | goto exit; | |
4831 | } | |
4832 | if (packet->size < ETH_HEADER_LEN) { | |
4833 | unixctl_command_reply(conn, 501, | |
4834 | "Packet data too short for Ethernet"); | |
4835 | goto exit; | |
4836 | } | |
4837 | ||
4838 | ds_put_cstr(&result, "Packet: "); | |
4839 | s = ofp_packet_to_string(packet->data, packet->size, packet->size); | |
4840 | ds_put_cstr(&result, s); | |
4841 | free(s); | |
4842 | ||
4843 | flow_extract(packet, tun_id, in_port, &flow); | |
4844 | } else { | |
abe529af BP |
4845 | unixctl_command_reply(conn, 501, "Bad command syntax"); |
4846 | goto exit; | |
4847 | } | |
4848 | ||
4849 | ofproto = ofproto_dpif_lookup(dpname); | |
4850 | if (!ofproto) { | |
4851 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
4852 | "for help)"); | |
4853 | goto exit; | |
4854 | } | |
4855 | ||
abe529af BP |
4856 | ds_put_cstr(&result, "Flow: "); |
4857 | flow_format(&result, &flow); | |
4858 | ds_put_char(&result, '\n'); | |
4859 | ||
29901626 BP |
4860 | rule = rule_dpif_lookup(ofproto, &flow, 0); |
4861 | trace_format_rule(&result, 0, 0, rule); | |
abe529af BP |
4862 | if (rule) { |
4863 | struct ofproto_trace trace; | |
4864 | struct ofpbuf *odp_actions; | |
4865 | ||
4866 | trace.result = &result; | |
4867 | trace.flow = flow; | |
876b0e1c | 4868 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet); |
abe529af BP |
4869 | trace.ctx.resubmit_hook = trace_resubmit; |
4870 | odp_actions = xlate_actions(&trace.ctx, | |
4871 | rule->up.actions, rule->up.n_actions); | |
4872 | ||
4873 | ds_put_char(&result, '\n'); | |
4874 | trace_format_flow(&result, 0, "Final flow", &trace); | |
4875 | ds_put_cstr(&result, "Datapath actions: "); | |
4876 | format_odp_actions(&result, odp_actions->data, odp_actions->size); | |
4877 | ofpbuf_delete(odp_actions); | |
876b0e1c BP |
4878 | |
4879 | if (!trace.ctx.may_set_up_flow) { | |
4880 | if (packet) { | |
4881 | ds_put_cstr(&result, "\nThis flow is not cachable."); | |
4882 | } else { | |
4883 | ds_put_cstr(&result, "\nThe datapath actions are incomplete--" | |
4884 | "for complete actions, please supply a packet."); | |
4885 | } | |
4886 | } | |
abe529af BP |
4887 | } |
4888 | ||
4889 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
4890 | ||
4891 | exit: | |
4892 | ds_destroy(&result); | |
876b0e1c BP |
4893 | ofpbuf_delete(packet); |
4894 | ofpbuf_uninit(&odp_key); | |
abe529af BP |
4895 | free(args); |
4896 | } | |
4897 | ||
7ee20df1 BP |
4898 | static void |
4899 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, | |
4900 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
4901 | { | |
4902 | clogged = true; | |
4903 | unixctl_command_reply(conn, 200, NULL); | |
4904 | } | |
4905 | ||
4906 | static void | |
4907 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, | |
4908 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
4909 | { | |
4910 | clogged = false; | |
4911 | unixctl_command_reply(conn, 200, NULL); | |
4912 | } | |
4913 | ||
abe529af BP |
4914 | static void |
4915 | ofproto_dpif_unixctl_init(void) | |
4916 | { | |
4917 | static bool registered; | |
4918 | if (registered) { | |
4919 | return; | |
4920 | } | |
4921 | registered = true; | |
4922 | ||
7ff2009a JP |
4923 | unixctl_command_register("ofproto/trace", |
4924 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
4925 | ofproto_unixctl_trace, NULL); | |
4926 | unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show, | |
4927 | NULL); | |
4928 | unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL); | |
4929 | unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL); | |
abe529af BP |
4930 | } |
4931 | \f | |
4932 | const struct ofproto_class ofproto_dpif_class = { | |
4933 | enumerate_types, | |
4934 | enumerate_names, | |
4935 | del, | |
4936 | alloc, | |
4937 | construct, | |
4938 | destruct, | |
4939 | dealloc, | |
4940 | run, | |
4941 | wait, | |
4942 | flush, | |
6c1491fb BP |
4943 | get_features, |
4944 | get_tables, | |
abe529af BP |
4945 | port_alloc, |
4946 | port_construct, | |
4947 | port_destruct, | |
4948 | port_dealloc, | |
4949 | port_modified, | |
4950 | port_reconfigured, | |
4951 | port_query_by_name, | |
4952 | port_add, | |
4953 | port_del, | |
4954 | port_dump_start, | |
4955 | port_dump_next, | |
4956 | port_dump_done, | |
4957 | port_poll, | |
4958 | port_poll_wait, | |
4959 | port_is_lacp_current, | |
0ab6decf | 4960 | NULL, /* rule_choose_table */ |
abe529af BP |
4961 | rule_alloc, |
4962 | rule_construct, | |
4963 | rule_destruct, | |
4964 | rule_dealloc, | |
abe529af BP |
4965 | rule_get_stats, |
4966 | rule_execute, | |
4967 | rule_modify_actions, | |
4968 | get_drop_frags, | |
4969 | set_drop_frags, | |
4970 | packet_out, | |
4971 | set_netflow, | |
4972 | get_netflow_ids, | |
4973 | set_sflow, | |
4974 | set_cfm, | |
a5610457 | 4975 | get_cfm_fault, |
1de11730 | 4976 | get_cfm_remote_mpids, |
abe529af BP |
4977 | bundle_set, |
4978 | bundle_remove, | |
4979 | mirror_set, | |
4980 | set_flood_vlans, | |
4981 | is_mirror_output_bundle, | |
8402c74b | 4982 | forward_bpdu_changed, |
abe529af | 4983 | }; |