]>
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 | 65 | * flow translation. */ |
642a5c05 | 66 | #define MAX_RESUBMIT_RECURSION 32 |
abe529af | 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 *); | |
7bde8dd8 | 158 | static void bundle_update(struct ofbundle *); |
abe529af BP |
159 | static void bundle_destroy(struct ofbundle *); |
160 | static void bundle_del_port(struct ofport_dpif *); | |
161 | static void bundle_run(struct ofbundle *); | |
162 | static void bundle_wait(struct ofbundle *); | |
163 | ||
21f7563c JP |
164 | static void stp_run(struct ofproto_dpif *ofproto); |
165 | static void stp_wait(struct ofproto_dpif *ofproto); | |
166 | ||
abe529af BP |
167 | struct action_xlate_ctx { |
168 | /* action_xlate_ctx_init() initializes these members. */ | |
169 | ||
170 | /* The ofproto. */ | |
171 | struct ofproto_dpif *ofproto; | |
172 | ||
173 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
174 | * this flow when actions change header fields. */ | |
175 | struct flow flow; | |
176 | ||
177 | /* The packet corresponding to 'flow', or a null pointer if we are | |
178 | * revalidating without a packet to refer to. */ | |
179 | const struct ofpbuf *packet; | |
180 | ||
75a75043 BP |
181 | /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We |
182 | * want to execute them if we are actually processing a packet, or if we | |
183 | * are accounting for packets that the datapath has processed, but not if | |
184 | * we are just revalidating. */ | |
185 | bool may_learn; | |
186 | ||
abe529af BP |
187 | /* If nonnull, called just before executing a resubmit action. |
188 | * | |
189 | * This is normally null so the client has to set it manually after | |
190 | * calling action_xlate_ctx_init(). */ | |
191 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *); | |
192 | ||
abe529af BP |
193 | /* xlate_actions() initializes and uses these members. The client might want |
194 | * to look at them after it returns. */ | |
195 | ||
196 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 197 | tag_type tags; /* Tags associated with actions. */ |
abe529af BP |
198 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
199 | * be reassessed for every packet. */ | |
75a75043 BP |
200 | bool has_learn; /* Actions include NXAST_LEARN? */ |
201 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
abe529af BP |
202 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
203 | ||
204 | /* xlate_actions() initializes and uses these members, but the client has no | |
205 | * reason to look at them. */ | |
206 | ||
207 | int recurse; /* Recursion level, via xlate_table_action. */ | |
b3e9b2ed | 208 | struct flow base_flow; /* Flow at the last commit. */ |
abff858b | 209 | uint32_t original_priority; /* Priority when packet arrived. */ |
29901626 | 210 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
211 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
212 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
213 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
848e8809 | 214 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
215 | }; |
216 | ||
217 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
218 | struct ofproto_dpif *, const struct flow *, | |
219 | const struct ofpbuf *); | |
220 | static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, | |
221 | const union ofp_action *in, size_t n_in); | |
222 | ||
223 | /* An exact-match instantiation of an OpenFlow flow. */ | |
224 | struct facet { | |
225 | long long int used; /* Time last used; time created if not used. */ | |
226 | ||
227 | /* These statistics: | |
228 | * | |
229 | * - Do include packets and bytes sent "by hand", e.g. with | |
230 | * dpif_execute(). | |
231 | * | |
232 | * - Do include packets and bytes that were obtained from the datapath | |
907a4c5e | 233 | * when its statistics were reset (e.g. dpif_flow_put() with |
abe529af | 234 | * DPIF_FP_ZERO_STATS). |
abe529af BP |
235 | */ |
236 | uint64_t packet_count; /* Number of packets received. */ | |
237 | uint64_t byte_count; /* Number of bytes received. */ | |
238 | ||
239 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
240 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
241 | ||
242 | uint64_t rs_packet_count; /* Packets pushed to resubmit children. */ | |
243 | uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */ | |
244 | long long int rs_used; /* Used time pushed to resubmit children. */ | |
245 | ||
907a4c5e | 246 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
abe529af BP |
247 | |
248 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
249 | struct list list_node; /* In owning rule's 'facets' list. */ | |
250 | struct rule_dpif *rule; /* Owning rule. */ | |
251 | struct flow flow; /* Exact-match flow. */ | |
252 | bool installed; /* Installed in datapath? */ | |
253 | bool may_install; /* True ordinarily; false if actions must | |
254 | * be reassessed for every packet. */ | |
75a75043 BP |
255 | bool has_learn; /* Actions include NXAST_LEARN? */ |
256 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
abe529af BP |
257 | size_t actions_len; /* Number of bytes in actions[]. */ |
258 | struct nlattr *actions; /* Datapath actions. */ | |
259 | tag_type tags; /* Tags. */ | |
260 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
261 | }; | |
262 | ||
f3827897 | 263 | static struct facet *facet_create(struct rule_dpif *, const struct flow *); |
abe529af BP |
264 | static void facet_remove(struct ofproto_dpif *, struct facet *); |
265 | static void facet_free(struct facet *); | |
266 | ||
267 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); | |
268 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, | |
269 | const struct flow *); | |
270 | static bool facet_revalidate(struct ofproto_dpif *, struct facet *); | |
271 | ||
3d9e05f8 BP |
272 | static bool execute_controller_action(struct ofproto_dpif *, |
273 | const struct flow *, | |
274 | const struct nlattr *odp_actions, | |
275 | size_t actions_len, | |
276 | struct ofpbuf *packet); | |
abe529af BP |
277 | static void facet_execute(struct ofproto_dpif *, struct facet *, |
278 | struct ofpbuf *packet); | |
279 | ||
280 | static int facet_put__(struct ofproto_dpif *, struct facet *, | |
281 | const struct nlattr *actions, size_t actions_len, | |
282 | struct dpif_flow_stats *); | |
283 | static void facet_install(struct ofproto_dpif *, struct facet *, | |
284 | bool zero_stats); | |
285 | static void facet_uninstall(struct ofproto_dpif *, struct facet *); | |
286 | static void facet_flush_stats(struct ofproto_dpif *, struct facet *); | |
287 | ||
288 | static void facet_make_actions(struct ofproto_dpif *, struct facet *, | |
289 | const struct ofpbuf *packet); | |
290 | static void facet_update_time(struct ofproto_dpif *, struct facet *, | |
291 | long long int used); | |
292 | static void facet_update_stats(struct ofproto_dpif *, struct facet *, | |
293 | const struct dpif_flow_stats *); | |
bbb5d219 | 294 | static void facet_reset_counters(struct facet *); |
3a88e544 | 295 | static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *); |
abe529af | 296 | static void facet_push_stats(struct facet *); |
55af77bb | 297 | static void facet_account(struct ofproto_dpif *, struct facet *); |
abe529af BP |
298 | |
299 | static bool facet_is_controller_flow(struct facet *); | |
300 | ||
301 | static void flow_push_stats(const struct rule_dpif *, | |
302 | struct flow *, uint64_t packets, uint64_t bytes, | |
303 | long long int used); | |
304 | ||
54a9cbc9 BP |
305 | static uint32_t rule_calculate_tag(const struct flow *, |
306 | const struct flow_wildcards *, | |
307 | uint32_t basis); | |
308 | static void rule_invalidate(const struct rule_dpif *); | |
309 | ||
abe529af BP |
310 | struct ofport_dpif { |
311 | struct ofport up; | |
312 | ||
313 | uint32_t odp_port; | |
314 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
315 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
316 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
317 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 318 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 319 | bool may_enable; /* May be enabled in bonds. */ |
21f7563c JP |
320 | |
321 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ | |
322 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
323 | long long int stp_state_entered; | |
abe529af BP |
324 | }; |
325 | ||
326 | static struct ofport_dpif * | |
327 | ofport_dpif_cast(const struct ofport *ofport) | |
328 | { | |
329 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
330 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
331 | } | |
332 | ||
333 | static void port_run(struct ofport_dpif *); | |
334 | static void port_wait(struct ofport_dpif *); | |
a5610457 | 335 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
abe529af | 336 | |
7ee20df1 BP |
337 | struct dpif_completion { |
338 | struct list list_node; | |
339 | struct ofoperation *op; | |
340 | }; | |
341 | ||
54a9cbc9 BP |
342 | /* Extra information about a classifier table. |
343 | * Currently used just for optimized flow revalidation. */ | |
344 | struct table_dpif { | |
345 | /* If either of these is nonnull, then this table has a form that allows | |
346 | * flows to be tagged to avoid revalidating most flows for the most common | |
347 | * kinds of flow table changes. */ | |
348 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
349 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
350 | uint32_t basis; /* Keeps each table's tags separate. */ | |
351 | }; | |
352 | ||
abe529af BP |
353 | struct ofproto_dpif { |
354 | struct ofproto up; | |
355 | struct dpif *dpif; | |
356 | int max_ports; | |
357 | ||
6c1491fb BP |
358 | /* Statistics. */ |
359 | uint64_t n_matches; | |
360 | ||
abe529af BP |
361 | /* Bridging. */ |
362 | struct netflow *netflow; | |
bae473fe | 363 | struct dpif_sflow *sflow; |
abe529af BP |
364 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
365 | struct mac_learning *ml; | |
366 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
367 | bool has_bonded_bundles; | |
368 | ||
369 | /* Expiration. */ | |
370 | struct timer next_expiration; | |
371 | ||
372 | /* Facets. */ | |
373 | struct hmap facets; | |
54a9cbc9 BP |
374 | |
375 | /* Revalidation. */ | |
376 | struct table_dpif tables[N_TABLES]; | |
abe529af BP |
377 | bool need_revalidate; |
378 | struct tag_set revalidate_set; | |
7ee20df1 BP |
379 | |
380 | /* Support for debugging async flow mods. */ | |
381 | struct list completions; | |
daff3353 EJ |
382 | |
383 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
21f7563c JP |
384 | |
385 | /* Spanning tree. */ | |
386 | struct stp *stp; | |
387 | long long int stp_last_tick; | |
abe529af BP |
388 | }; |
389 | ||
7ee20df1 BP |
390 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
391 | * for debugging the asynchronous flow_mod implementation.) */ | |
392 | static bool clogged; | |
393 | ||
abe529af BP |
394 | static void ofproto_dpif_unixctl_init(void); |
395 | ||
396 | static struct ofproto_dpif * | |
397 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
398 | { | |
399 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
400 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
401 | } | |
402 | ||
403 | static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, | |
404 | uint16_t ofp_port); | |
405 | static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, | |
406 | uint32_t odp_port); | |
407 | ||
408 | /* Packet processing. */ | |
409 | static void update_learning_table(struct ofproto_dpif *, | |
410 | const struct flow *, int vlan, | |
411 | struct ofbundle *); | |
412 | static bool is_admissible(struct ofproto_dpif *, const struct flow *, | |
413 | bool have_packet, tag_type *, int *vlanp, | |
414 | struct ofbundle **in_bundlep); | |
501f8d1f BP |
415 | |
416 | /* Upcalls. */ | |
417 | #define FLOW_MISS_MAX_BATCH 50 | |
abe529af | 418 | static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *); |
501f8d1f BP |
419 | static void handle_miss_upcalls(struct ofproto_dpif *, |
420 | struct dpif_upcall *, size_t n); | |
abe529af BP |
421 | |
422 | /* Flow expiration. */ | |
423 | static int expire(struct ofproto_dpif *); | |
424 | ||
425 | /* Utilities. */ | |
b2fda3ef | 426 | static int send_packet(struct ofproto_dpif *, uint32_t odp_port, |
abe529af | 427 | const struct ofpbuf *packet); |
6ff686f2 PS |
428 | static size_t |
429 | compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, | |
430 | const struct flow *, uint32_t odp_port); | |
abe529af BP |
431 | /* Global variables. */ |
432 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
433 | \f | |
434 | /* Factory functions. */ | |
435 | ||
436 | static void | |
437 | enumerate_types(struct sset *types) | |
438 | { | |
439 | dp_enumerate_types(types); | |
440 | } | |
441 | ||
442 | static int | |
443 | enumerate_names(const char *type, struct sset *names) | |
444 | { | |
445 | return dp_enumerate_names(type, names); | |
446 | } | |
447 | ||
448 | static int | |
449 | del(const char *type, const char *name) | |
450 | { | |
451 | struct dpif *dpif; | |
452 | int error; | |
453 | ||
454 | error = dpif_open(name, type, &dpif); | |
455 | if (!error) { | |
456 | error = dpif_delete(dpif); | |
457 | dpif_close(dpif); | |
458 | } | |
459 | return error; | |
460 | } | |
461 | \f | |
462 | /* Basic life-cycle. */ | |
463 | ||
464 | static struct ofproto * | |
465 | alloc(void) | |
466 | { | |
467 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
468 | return &ofproto->up; | |
469 | } | |
470 | ||
471 | static void | |
472 | dealloc(struct ofproto *ofproto_) | |
473 | { | |
474 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
475 | free(ofproto); | |
476 | } | |
477 | ||
478 | static int | |
073e2a6f | 479 | construct(struct ofproto *ofproto_, int *n_tablesp) |
abe529af BP |
480 | { |
481 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
482 | const char *name = ofproto->up.name; | |
483 | int error; | |
484 | int i; | |
485 | ||
486 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
487 | if (error) { | |
488 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
489 | return error; | |
490 | } | |
491 | ||
492 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 493 | ofproto->n_matches = 0; |
abe529af | 494 | |
be8194bb JG |
495 | dpif_flow_flush(ofproto->dpif); |
496 | dpif_recv_purge(ofproto->dpif); | |
497 | ||
abe529af BP |
498 | error = dpif_recv_set_mask(ofproto->dpif, |
499 | ((1u << DPIF_UC_MISS) | | |
6ff686f2 | 500 | (1u << DPIF_UC_ACTION))); |
abe529af BP |
501 | if (error) { |
502 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
503 | dpif_close(ofproto->dpif); | |
504 | return error; | |
505 | } | |
abe529af BP |
506 | |
507 | ofproto->netflow = NULL; | |
508 | ofproto->sflow = NULL; | |
21f7563c | 509 | ofproto->stp = NULL; |
abe529af BP |
510 | hmap_init(&ofproto->bundles); |
511 | ofproto->ml = mac_learning_create(); | |
512 | for (i = 0; i < MAX_MIRRORS; i++) { | |
513 | ofproto->mirrors[i] = NULL; | |
514 | } | |
515 | ofproto->has_bonded_bundles = false; | |
516 | ||
517 | timer_set_duration(&ofproto->next_expiration, 1000); | |
518 | ||
519 | hmap_init(&ofproto->facets); | |
54a9cbc9 BP |
520 | |
521 | for (i = 0; i < N_TABLES; i++) { | |
522 | struct table_dpif *table = &ofproto->tables[i]; | |
523 | ||
524 | table->catchall_table = NULL; | |
525 | table->other_table = NULL; | |
526 | table->basis = random_uint32(); | |
527 | } | |
abe529af BP |
528 | ofproto->need_revalidate = false; |
529 | tag_set_init(&ofproto->revalidate_set); | |
530 | ||
7ee20df1 BP |
531 | list_init(&ofproto->completions); |
532 | ||
abe529af BP |
533 | ofproto_dpif_unixctl_init(); |
534 | ||
daff3353 EJ |
535 | ofproto->has_bundle_action = false; |
536 | ||
9cdaaebe | 537 | *n_tablesp = N_TABLES; |
abe529af BP |
538 | return 0; |
539 | } | |
540 | ||
7ee20df1 BP |
541 | static void |
542 | complete_operations(struct ofproto_dpif *ofproto) | |
543 | { | |
544 | struct dpif_completion *c, *next; | |
545 | ||
546 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
547 | ofoperation_complete(c->op, 0); | |
548 | list_remove(&c->list_node); | |
549 | free(c); | |
550 | } | |
551 | } | |
552 | ||
abe529af BP |
553 | static void |
554 | destruct(struct ofproto *ofproto_) | |
555 | { | |
556 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 557 | struct rule_dpif *rule, *next_rule; |
0697b5c3 | 558 | struct classifier *table; |
abe529af BP |
559 | int i; |
560 | ||
7ee20df1 BP |
561 | complete_operations(ofproto); |
562 | ||
0697b5c3 BP |
563 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
564 | struct cls_cursor cursor; | |
565 | ||
566 | cls_cursor_init(&cursor, table, NULL); | |
567 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
568 | ofproto_rule_destroy(&rule->up); | |
569 | } | |
7ee20df1 BP |
570 | } |
571 | ||
abe529af BP |
572 | for (i = 0; i < MAX_MIRRORS; i++) { |
573 | mirror_destroy(ofproto->mirrors[i]); | |
574 | } | |
575 | ||
576 | netflow_destroy(ofproto->netflow); | |
bae473fe | 577 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
578 | hmap_destroy(&ofproto->bundles); |
579 | mac_learning_destroy(ofproto->ml); | |
580 | ||
581 | hmap_destroy(&ofproto->facets); | |
582 | ||
583 | dpif_close(ofproto->dpif); | |
584 | } | |
585 | ||
586 | static int | |
587 | run(struct ofproto *ofproto_) | |
588 | { | |
589 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
501f8d1f | 590 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
abe529af BP |
591 | struct ofport_dpif *ofport; |
592 | struct ofbundle *bundle; | |
501f8d1f | 593 | size_t n_misses; |
abe529af BP |
594 | int i; |
595 | ||
7ee20df1 BP |
596 | if (!clogged) { |
597 | complete_operations(ofproto); | |
598 | } | |
abe529af BP |
599 | dpif_run(ofproto->dpif); |
600 | ||
501f8d1f BP |
601 | n_misses = 0; |
602 | for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) { | |
603 | struct dpif_upcall *upcall = &misses[n_misses]; | |
abe529af BP |
604 | int error; |
605 | ||
501f8d1f | 606 | error = dpif_recv(ofproto->dpif, upcall); |
abe529af | 607 | if (error) { |
501f8d1f | 608 | if (error == ENODEV && n_misses == 0) { |
abe529af BP |
609 | return error; |
610 | } | |
611 | break; | |
612 | } | |
613 | ||
501f8d1f BP |
614 | if (upcall->type == DPIF_UC_MISS) { |
615 | /* Handle it later. */ | |
616 | n_misses++; | |
617 | } else { | |
618 | handle_upcall(ofproto, upcall); | |
619 | } | |
abe529af BP |
620 | } |
621 | ||
501f8d1f BP |
622 | handle_miss_upcalls(ofproto, misses, n_misses); |
623 | ||
abe529af BP |
624 | if (timer_expired(&ofproto->next_expiration)) { |
625 | int delay = expire(ofproto); | |
626 | timer_set_duration(&ofproto->next_expiration, delay); | |
627 | } | |
628 | ||
629 | if (ofproto->netflow) { | |
630 | netflow_run(ofproto->netflow); | |
631 | } | |
632 | if (ofproto->sflow) { | |
bae473fe | 633 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
634 | } |
635 | ||
636 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
637 | port_run(ofport); | |
638 | } | |
639 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
640 | bundle_run(bundle); | |
641 | } | |
642 | ||
21f7563c | 643 | stp_run(ofproto); |
1c313b88 BP |
644 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
645 | ||
abe529af BP |
646 | /* Now revalidate if there's anything to do. */ |
647 | if (ofproto->need_revalidate | |
648 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
649 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
650 | bool revalidate_all = ofproto->need_revalidate; | |
651 | struct facet *facet, *next; | |
652 | ||
653 | /* Clear the revalidation flags. */ | |
654 | tag_set_init(&ofproto->revalidate_set); | |
655 | ofproto->need_revalidate = false; | |
656 | ||
657 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
658 | if (revalidate_all | |
659 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
660 | facet_revalidate(ofproto, facet); | |
661 | } | |
662 | } | |
663 | } | |
664 | ||
665 | return 0; | |
666 | } | |
667 | ||
668 | static void | |
669 | wait(struct ofproto *ofproto_) | |
670 | { | |
671 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
672 | struct ofport_dpif *ofport; | |
673 | struct ofbundle *bundle; | |
674 | ||
7ee20df1 BP |
675 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
676 | poll_immediate_wake(); | |
677 | } | |
678 | ||
abe529af BP |
679 | dpif_wait(ofproto->dpif); |
680 | dpif_recv_wait(ofproto->dpif); | |
681 | if (ofproto->sflow) { | |
bae473fe | 682 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
683 | } |
684 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
685 | poll_immediate_wake(); | |
686 | } | |
687 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
688 | port_wait(ofport); | |
689 | } | |
690 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
691 | bundle_wait(bundle); | |
692 | } | |
1c313b88 | 693 | mac_learning_wait(ofproto->ml); |
21f7563c | 694 | stp_wait(ofproto); |
abe529af BP |
695 | if (ofproto->need_revalidate) { |
696 | /* Shouldn't happen, but if it does just go around again. */ | |
697 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
698 | poll_immediate_wake(); | |
699 | } else { | |
700 | timer_wait(&ofproto->next_expiration); | |
701 | } | |
702 | } | |
703 | ||
704 | static void | |
705 | flush(struct ofproto *ofproto_) | |
706 | { | |
707 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
708 | struct facet *facet, *next_facet; | |
709 | ||
710 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
711 | /* Mark the facet as not installed so that facet_remove() doesn't | |
712 | * bother trying to uninstall it. There is no point in uninstalling it | |
713 | * individually since we are about to blow away all the facets with | |
714 | * dpif_flow_flush(). */ | |
715 | facet->installed = false; | |
716 | facet->dp_packet_count = 0; | |
717 | facet->dp_byte_count = 0; | |
718 | facet_remove(ofproto, facet); | |
719 | } | |
720 | dpif_flow_flush(ofproto->dpif); | |
721 | } | |
722 | ||
6c1491fb BP |
723 | static void |
724 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
725 | bool *arp_match_ip, uint32_t *actions) | |
726 | { | |
727 | *arp_match_ip = true; | |
728 | *actions = ((1u << OFPAT_OUTPUT) | | |
729 | (1u << OFPAT_SET_VLAN_VID) | | |
730 | (1u << OFPAT_SET_VLAN_PCP) | | |
731 | (1u << OFPAT_STRIP_VLAN) | | |
732 | (1u << OFPAT_SET_DL_SRC) | | |
733 | (1u << OFPAT_SET_DL_DST) | | |
734 | (1u << OFPAT_SET_NW_SRC) | | |
735 | (1u << OFPAT_SET_NW_DST) | | |
736 | (1u << OFPAT_SET_NW_TOS) | | |
737 | (1u << OFPAT_SET_TP_SRC) | | |
738 | (1u << OFPAT_SET_TP_DST) | | |
739 | (1u << OFPAT_ENQUEUE)); | |
740 | } | |
741 | ||
742 | static void | |
743 | get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) | |
744 | { | |
745 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 746 | struct dpif_dp_stats s; |
6c1491fb BP |
747 | |
748 | strcpy(ots->name, "classifier"); | |
749 | ||
750 | dpif_get_dp_stats(ofproto->dpif, &s); | |
751 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
752 | put_32aligned_be64(&ots->matched_count, | |
753 | htonll(s.n_hit + ofproto->n_matches)); | |
754 | } | |
755 | ||
abe529af BP |
756 | static int |
757 | set_netflow(struct ofproto *ofproto_, | |
758 | const struct netflow_options *netflow_options) | |
759 | { | |
760 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
761 | ||
762 | if (netflow_options) { | |
763 | if (!ofproto->netflow) { | |
764 | ofproto->netflow = netflow_create(); | |
765 | } | |
766 | return netflow_set_options(ofproto->netflow, netflow_options); | |
767 | } else { | |
768 | netflow_destroy(ofproto->netflow); | |
769 | ofproto->netflow = NULL; | |
770 | return 0; | |
771 | } | |
772 | } | |
773 | ||
774 | static struct ofport * | |
775 | port_alloc(void) | |
776 | { | |
777 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
778 | return &port->up; | |
779 | } | |
780 | ||
781 | static void | |
782 | port_dealloc(struct ofport *port_) | |
783 | { | |
784 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
785 | free(port); | |
786 | } | |
787 | ||
788 | static int | |
789 | port_construct(struct ofport *port_) | |
790 | { | |
791 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
792 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
793 | ||
f11c28c4 | 794 | ofproto->need_revalidate = true; |
abe529af BP |
795 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
796 | port->bundle = NULL; | |
797 | port->cfm = NULL; | |
798 | port->tag = tag_create_random(); | |
d5ffa7f2 | 799 | port->may_enable = true; |
21f7563c JP |
800 | port->stp_port = NULL; |
801 | port->stp_state = STP_DISABLED; | |
abe529af BP |
802 | |
803 | if (ofproto->sflow) { | |
bae473fe JP |
804 | dpif_sflow_add_port(ofproto->sflow, port->odp_port, |
805 | netdev_get_name(port->up.netdev)); | |
abe529af BP |
806 | } |
807 | ||
808 | return 0; | |
809 | } | |
810 | ||
811 | static void | |
812 | port_destruct(struct ofport *port_) | |
813 | { | |
814 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
815 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
816 | ||
f11c28c4 | 817 | ofproto->need_revalidate = true; |
abe529af | 818 | bundle_remove(port_); |
a5610457 | 819 | set_cfm(port_, NULL); |
abe529af | 820 | if (ofproto->sflow) { |
bae473fe | 821 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af BP |
822 | } |
823 | } | |
824 | ||
825 | static void | |
826 | port_modified(struct ofport *port_) | |
827 | { | |
828 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
829 | ||
830 | if (port->bundle && port->bundle->bond) { | |
831 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
832 | } | |
833 | } | |
834 | ||
835 | static void | |
836 | port_reconfigured(struct ofport *port_, ovs_be32 old_config) | |
837 | { | |
838 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
839 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
840 | ovs_be32 changed = old_config ^ port->up.opp.config; | |
841 | ||
842 | if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP | | |
843 | OFPPC_NO_FWD | OFPPC_NO_FLOOD)) { | |
844 | ofproto->need_revalidate = true; | |
7bde8dd8 JP |
845 | |
846 | if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) { | |
847 | bundle_update(port->bundle); | |
848 | } | |
abe529af BP |
849 | } |
850 | } | |
851 | ||
852 | static int | |
853 | set_sflow(struct ofproto *ofproto_, | |
854 | const struct ofproto_sflow_options *sflow_options) | |
855 | { | |
856 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 857 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 858 | |
abe529af | 859 | if (sflow_options) { |
bae473fe | 860 | if (!ds) { |
abe529af BP |
861 | struct ofport_dpif *ofport; |
862 | ||
bae473fe | 863 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 864 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
bae473fe JP |
865 | dpif_sflow_add_port(ds, ofport->odp_port, |
866 | netdev_get_name(ofport->up.netdev)); | |
abe529af | 867 | } |
6ff686f2 | 868 | ofproto->need_revalidate = true; |
abe529af | 869 | } |
bae473fe | 870 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 871 | } else { |
6ff686f2 PS |
872 | if (ds) { |
873 | dpif_sflow_destroy(ds); | |
874 | ofproto->need_revalidate = true; | |
875 | ofproto->sflow = NULL; | |
876 | } | |
abe529af BP |
877 | } |
878 | return 0; | |
879 | } | |
880 | ||
881 | static int | |
a5610457 | 882 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
883 | { |
884 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
885 | int error; | |
886 | ||
a5610457 | 887 | if (!s) { |
abe529af BP |
888 | error = 0; |
889 | } else { | |
890 | if (!ofport->cfm) { | |
8c977421 EJ |
891 | struct ofproto_dpif *ofproto; |
892 | ||
893 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
894 | ofproto->need_revalidate = true; | |
6f629657 | 895 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
896 | } |
897 | ||
a5610457 | 898 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
899 | return 0; |
900 | } | |
901 | ||
902 | error = EINVAL; | |
903 | } | |
904 | cfm_destroy(ofport->cfm); | |
905 | ofport->cfm = NULL; | |
906 | return error; | |
907 | } | |
908 | ||
909 | static int | |
a5610457 | 910 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
911 | { |
912 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
913 | |
914 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 915 | } |
1de11730 EJ |
916 | |
917 | static int | |
918 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
919 | size_t *n_rmps) | |
920 | { | |
921 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
922 | ||
923 | if (ofport->cfm) { | |
924 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
925 | return 0; | |
926 | } else { | |
927 | return -1; | |
928 | } | |
929 | } | |
abe529af | 930 | \f |
21f7563c JP |
931 | /* Spanning Tree. */ |
932 | ||
933 | static void | |
934 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
935 | { | |
936 | struct ofproto_dpif *ofproto = ofproto_; | |
937 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
938 | struct ofport_dpif *ofport; | |
939 | ||
940 | ofport = stp_port_get_aux(sp); | |
941 | if (!ofport) { | |
942 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
943 | ofproto->up.name, port_num); | |
944 | } else { | |
945 | struct eth_header *eth = pkt->l2; | |
946 | ||
947 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
948 | if (eth_addr_is_zero(eth->eth_src)) { | |
949 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
950 | "with unknown MAC", ofproto->up.name, port_num); | |
951 | } else { | |
952 | int error = netdev_send(ofport->up.netdev, pkt); | |
953 | if (error) { | |
954 | VLOG_WARN_RL(&rl, "%s: sending BPDU on port %s failed (%s)", | |
955 | ofproto->up.name, | |
956 | netdev_get_name(ofport->up.netdev), | |
957 | strerror(error)); | |
958 | } | |
959 | } | |
960 | } | |
961 | ofpbuf_delete(pkt); | |
962 | } | |
963 | ||
964 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
965 | static int | |
966 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
967 | { | |
968 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
969 | ||
970 | /* Only revalidate flows if the configuration changed. */ | |
971 | if (!s != !ofproto->stp) { | |
972 | ofproto->need_revalidate = true; | |
973 | } | |
974 | ||
975 | if (s) { | |
976 | if (!ofproto->stp) { | |
977 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
978 | send_bpdu_cb, ofproto); | |
979 | ofproto->stp_last_tick = time_msec(); | |
980 | } | |
981 | ||
982 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
983 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
984 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
985 | stp_set_max_age(ofproto->stp, s->max_age); | |
986 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
987 | } else { | |
988 | stp_destroy(ofproto->stp); | |
989 | ofproto->stp = NULL; | |
990 | } | |
991 | ||
992 | return 0; | |
993 | } | |
994 | ||
995 | static int | |
996 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
997 | { | |
998 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
999 | ||
1000 | if (ofproto->stp) { | |
1001 | s->enabled = true; | |
1002 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
1003 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1004 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1005 | } else { | |
1006 | s->enabled = false; | |
1007 | } | |
1008 | ||
1009 | return 0; | |
1010 | } | |
1011 | ||
1012 | static void | |
1013 | update_stp_port_state(struct ofport_dpif *ofport) | |
1014 | { | |
1015 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1016 | enum stp_state state; | |
1017 | ||
1018 | /* Figure out new state. */ | |
1019 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1020 | : STP_DISABLED; | |
1021 | ||
1022 | /* Update state. */ | |
1023 | if (ofport->stp_state != state) { | |
1024 | ovs_be32 of_state; | |
1025 | bool fwd_change; | |
1026 | ||
1027 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1028 | netdev_get_name(ofport->up.netdev), | |
1029 | stp_state_name(ofport->stp_state), | |
1030 | stp_state_name(state)); | |
1031 | if (stp_learn_in_state(ofport->stp_state) | |
1032 | != stp_learn_in_state(state)) { | |
1033 | /* xxx Learning action flows should also be flushed. */ | |
1034 | mac_learning_flush(ofproto->ml); | |
1035 | } | |
1036 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1037 | != stp_forward_in_state(state); | |
1038 | ||
1039 | ofproto->need_revalidate = true; | |
1040 | ofport->stp_state = state; | |
1041 | ofport->stp_state_entered = time_msec(); | |
1042 | ||
1043 | if (fwd_change) { | |
1044 | bundle_update(ofport->bundle); | |
1045 | } | |
1046 | ||
1047 | /* Update the STP state bits in the OpenFlow port description. */ | |
1048 | of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK)) | |
1049 | | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN | |
1050 | : state == STP_LEARNING ? OFPPS_STP_LEARN | |
1051 | : state == STP_FORWARDING ? OFPPS_STP_FORWARD | |
1052 | : state == STP_BLOCKING ? OFPPS_STP_BLOCK | |
1053 | : 0); | |
1054 | ofproto_port_set_state(&ofport->up, of_state); | |
1055 | } | |
1056 | } | |
1057 | ||
1058 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1059 | * caller is responsible for assigning STP port numbers and ensuring | |
1060 | * there are no duplicates. */ | |
1061 | static int | |
1062 | set_stp_port(struct ofport *ofport_, | |
1063 | const struct ofproto_port_stp_settings *s) | |
1064 | { | |
1065 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1066 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1067 | struct stp_port *sp = ofport->stp_port; | |
1068 | ||
1069 | if (!s || !s->enable) { | |
1070 | if (sp) { | |
1071 | ofport->stp_port = NULL; | |
1072 | stp_port_disable(sp); | |
1073 | } | |
1074 | return 0; | |
1075 | } else if (sp && stp_port_no(sp) != s->port_num | |
1076 | && ofport == stp_port_get_aux(sp)) { | |
1077 | /* The port-id changed, so disable the old one if it's not | |
1078 | * already in use by another port. */ | |
1079 | stp_port_disable(sp); | |
1080 | } | |
1081 | ||
1082 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1083 | stp_port_enable(sp); | |
1084 | ||
1085 | stp_port_set_aux(sp, ofport); | |
1086 | stp_port_set_priority(sp, s->priority); | |
1087 | stp_port_set_path_cost(sp, s->path_cost); | |
1088 | ||
1089 | update_stp_port_state(ofport); | |
1090 | ||
1091 | return 0; | |
1092 | } | |
1093 | ||
1094 | static int | |
1095 | get_stp_port_status(struct ofport *ofport_, | |
1096 | struct ofproto_port_stp_status *s) | |
1097 | { | |
1098 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1099 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1100 | struct stp_port *sp = ofport->stp_port; | |
1101 | ||
1102 | if (!ofproto->stp || !sp) { | |
1103 | s->enabled = false; | |
1104 | return 0; | |
1105 | } | |
1106 | ||
1107 | s->enabled = true; | |
1108 | s->port_id = stp_port_get_id(sp); | |
1109 | s->state = stp_port_get_state(sp); | |
1110 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1111 | s->role = stp_port_get_role(sp); | |
1112 | ||
1113 | return 0; | |
1114 | } | |
1115 | ||
1116 | static void | |
1117 | stp_run(struct ofproto_dpif *ofproto) | |
1118 | { | |
1119 | if (ofproto->stp) { | |
1120 | long long int now = time_msec(); | |
1121 | long long int elapsed = now - ofproto->stp_last_tick; | |
1122 | struct stp_port *sp; | |
1123 | ||
1124 | if (elapsed > 0) { | |
1125 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1126 | ofproto->stp_last_tick = now; | |
1127 | } | |
1128 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1129 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1130 | ||
1131 | if (ofport) { | |
1132 | update_stp_port_state(ofport); | |
1133 | } | |
1134 | } | |
1135 | } | |
1136 | } | |
1137 | ||
1138 | static void | |
1139 | stp_wait(struct ofproto_dpif *ofproto) | |
1140 | { | |
1141 | if (ofproto->stp) { | |
1142 | poll_timer_wait(1000); | |
1143 | } | |
1144 | } | |
1145 | ||
1146 | /* Returns true if STP should process 'flow'. */ | |
1147 | static bool | |
1148 | stp_should_process_flow(const struct flow *flow) | |
1149 | { | |
1150 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1151 | } | |
1152 | ||
1153 | static void | |
1154 | stp_process_packet(const struct ofport_dpif *ofport, | |
1155 | const struct ofpbuf *packet) | |
1156 | { | |
1157 | struct ofpbuf payload = *packet; | |
1158 | struct eth_header *eth = payload.data; | |
1159 | struct stp_port *sp = ofport->stp_port; | |
1160 | ||
1161 | /* Sink packets on ports that have STP disabled when the bridge has | |
1162 | * STP enabled. */ | |
1163 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1164 | return; | |
1165 | } | |
1166 | ||
1167 | /* Trim off padding on payload. */ | |
c573540b BP |
1168 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
1169 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
1170 | } |
1171 | ||
1172 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1173 | stp_received_bpdu(sp, payload.data, payload.size); | |
1174 | } | |
1175 | } | |
1176 | \f | |
abe529af BP |
1177 | /* Bundles. */ |
1178 | ||
1179 | /* Expires all MAC learning entries associated with 'port' and forces ofproto | |
1180 | * to revalidate every flow. */ | |
1181 | static void | |
1182 | bundle_flush_macs(struct ofbundle *bundle) | |
1183 | { | |
1184 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1185 | struct mac_learning *ml = ofproto->ml; | |
1186 | struct mac_entry *mac, *next_mac; | |
1187 | ||
1188 | ofproto->need_revalidate = true; | |
1189 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
1190 | if (mac->port.p == bundle) { | |
1191 | mac_learning_expire(ml, mac); | |
1192 | } | |
1193 | } | |
1194 | } | |
1195 | ||
1196 | static struct ofbundle * | |
1197 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1198 | { | |
1199 | struct ofbundle *bundle; | |
1200 | ||
1201 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1202 | &ofproto->bundles) { | |
1203 | if (bundle->aux == aux) { | |
1204 | return bundle; | |
1205 | } | |
1206 | } | |
1207 | return NULL; | |
1208 | } | |
1209 | ||
1210 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1211 | * ones that are found to 'bundles'. */ | |
1212 | static void | |
1213 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1214 | void **auxes, size_t n_auxes, | |
1215 | struct hmapx *bundles) | |
1216 | { | |
1217 | size_t i; | |
1218 | ||
1219 | hmapx_init(bundles); | |
1220 | for (i = 0; i < n_auxes; i++) { | |
1221 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1222 | if (bundle) { | |
1223 | hmapx_add(bundles, bundle); | |
1224 | } | |
1225 | } | |
1226 | } | |
1227 | ||
7bde8dd8 JP |
1228 | static void |
1229 | bundle_update(struct ofbundle *bundle) | |
1230 | { | |
1231 | struct ofport_dpif *port; | |
1232 | ||
1233 | bundle->floodable = true; | |
1234 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
21f7563c JP |
1235 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD) |
1236 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
1237 | bundle->floodable = false; |
1238 | break; | |
1239 | } | |
1240 | } | |
1241 | } | |
1242 | ||
abe529af BP |
1243 | static void |
1244 | bundle_del_port(struct ofport_dpif *port) | |
1245 | { | |
1246 | struct ofbundle *bundle = port->bundle; | |
1247 | ||
6f77f4ae BP |
1248 | bundle->ofproto->need_revalidate = true; |
1249 | ||
abe529af BP |
1250 | list_remove(&port->bundle_node); |
1251 | port->bundle = NULL; | |
1252 | ||
1253 | if (bundle->lacp) { | |
1254 | lacp_slave_unregister(bundle->lacp, port); | |
1255 | } | |
1256 | if (bundle->bond) { | |
1257 | bond_slave_unregister(bundle->bond, port); | |
1258 | } | |
1259 | ||
7bde8dd8 | 1260 | bundle_update(bundle); |
abe529af BP |
1261 | } |
1262 | ||
1263 | static bool | |
1264 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
1265 | struct lacp_slave_settings *lacp, |
1266 | uint32_t bond_stable_id) | |
abe529af BP |
1267 | { |
1268 | struct ofport_dpif *port; | |
1269 | ||
1270 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1271 | if (!port) { | |
1272 | return false; | |
1273 | } | |
1274 | ||
1275 | if (port->bundle != bundle) { | |
6f77f4ae | 1276 | bundle->ofproto->need_revalidate = true; |
abe529af BP |
1277 | if (port->bundle) { |
1278 | bundle_del_port(port); | |
1279 | } | |
1280 | ||
1281 | port->bundle = bundle; | |
1282 | list_push_back(&bundle->ports, &port->bundle_node); | |
21f7563c JP |
1283 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD) |
1284 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
1285 | bundle->floodable = false; |
1286 | } | |
1287 | } | |
1288 | if (lacp) { | |
4a86aece | 1289 | port->bundle->ofproto->need_revalidate = true; |
abe529af BP |
1290 | lacp_slave_register(bundle->lacp, port, lacp); |
1291 | } | |
1292 | ||
00794817 BP |
1293 | port->bond_stable_id = bond_stable_id; |
1294 | ||
abe529af BP |
1295 | return true; |
1296 | } | |
1297 | ||
1298 | static void | |
1299 | bundle_destroy(struct ofbundle *bundle) | |
1300 | { | |
1301 | struct ofproto_dpif *ofproto; | |
1302 | struct ofport_dpif *port, *next_port; | |
1303 | int i; | |
1304 | ||
1305 | if (!bundle) { | |
1306 | return; | |
1307 | } | |
1308 | ||
1309 | ofproto = bundle->ofproto; | |
1310 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1311 | struct ofmirror *m = ofproto->mirrors[i]; | |
1312 | if (m) { | |
1313 | if (m->out == bundle) { | |
1314 | mirror_destroy(m); | |
1315 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1316 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
1317 | ofproto->need_revalidate = true; | |
1318 | } | |
1319 | } | |
1320 | } | |
1321 | ||
1322 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1323 | bundle_del_port(port); | |
1324 | } | |
1325 | ||
1326 | bundle_flush_macs(bundle); | |
1327 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); | |
1328 | free(bundle->name); | |
1329 | free(bundle->trunks); | |
1330 | lacp_destroy(bundle->lacp); | |
1331 | bond_destroy(bundle->bond); | |
1332 | free(bundle); | |
1333 | } | |
1334 | ||
1335 | static int | |
1336 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1337 | const struct ofproto_bundle_settings *s) | |
1338 | { | |
1339 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1340 | bool need_flush = false; | |
abe529af BP |
1341 | struct ofport_dpif *port; |
1342 | struct ofbundle *bundle; | |
ecac4ebf BP |
1343 | unsigned long *trunks; |
1344 | int vlan; | |
abe529af BP |
1345 | size_t i; |
1346 | bool ok; | |
1347 | ||
1348 | if (!s) { | |
1349 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1350 | return 0; | |
1351 | } | |
1352 | ||
1353 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1354 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1355 | ||
1356 | bundle = bundle_lookup(ofproto, aux); | |
1357 | if (!bundle) { | |
1358 | bundle = xmalloc(sizeof *bundle); | |
1359 | ||
1360 | bundle->ofproto = ofproto; | |
1361 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1362 | hash_pointer(aux, 0)); | |
1363 | bundle->aux = aux; | |
1364 | bundle->name = NULL; | |
1365 | ||
1366 | list_init(&bundle->ports); | |
ecac4ebf | 1367 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1368 | bundle->vlan = -1; |
1369 | bundle->trunks = NULL; | |
1370 | bundle->lacp = NULL; | |
1371 | bundle->bond = NULL; | |
1372 | ||
1373 | bundle->floodable = true; | |
1374 | ||
1375 | bundle->src_mirrors = 0; | |
1376 | bundle->dst_mirrors = 0; | |
1377 | bundle->mirror_out = 0; | |
1378 | } | |
1379 | ||
1380 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1381 | free(bundle->name); | |
1382 | bundle->name = xstrdup(s->name); | |
1383 | } | |
1384 | ||
1385 | /* LACP. */ | |
1386 | if (s->lacp) { | |
1387 | if (!bundle->lacp) { | |
8c977421 | 1388 | ofproto->need_revalidate = true; |
abe529af BP |
1389 | bundle->lacp = lacp_create(); |
1390 | } | |
1391 | lacp_configure(bundle->lacp, s->lacp); | |
1392 | } else { | |
1393 | lacp_destroy(bundle->lacp); | |
1394 | bundle->lacp = NULL; | |
1395 | } | |
1396 | ||
1397 | /* Update set of ports. */ | |
1398 | ok = true; | |
1399 | for (i = 0; i < s->n_slaves; i++) { | |
1400 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1401 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1402 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1403 | ok = false; |
1404 | } | |
1405 | } | |
1406 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1407 | struct ofport_dpif *next_port; | |
1408 | ||
1409 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1410 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1411 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1412 | goto found; |
1413 | } | |
1414 | } | |
1415 | ||
1416 | bundle_del_port(port); | |
1417 | found: ; | |
1418 | } | |
1419 | } | |
1420 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1421 | ||
1422 | if (list_is_empty(&bundle->ports)) { | |
1423 | bundle_destroy(bundle); | |
1424 | return EINVAL; | |
1425 | } | |
1426 | ||
ecac4ebf BP |
1427 | /* Set VLAN tagging mode */ |
1428 | if (s->vlan_mode != bundle->vlan_mode) { | |
1429 | bundle->vlan_mode = s->vlan_mode; | |
1430 | need_flush = true; | |
1431 | } | |
1432 | ||
abe529af | 1433 | /* Set VLAN tag. */ |
ecac4ebf BP |
1434 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1435 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1436 | : 0); | |
1437 | if (vlan != bundle->vlan) { | |
1438 | bundle->vlan = vlan; | |
abe529af BP |
1439 | need_flush = true; |
1440 | } | |
1441 | ||
1442 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1443 | switch (s->vlan_mode) { |
1444 | case PORT_VLAN_ACCESS: | |
1445 | trunks = NULL; | |
1446 | break; | |
1447 | ||
1448 | case PORT_VLAN_TRUNK: | |
1449 | trunks = (unsigned long *) s->trunks; | |
1450 | break; | |
1451 | ||
1452 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1453 | case PORT_VLAN_NATIVE_TAGGED: | |
1454 | if (vlan != 0 && (!s->trunks | |
1455 | || !bitmap_is_set(s->trunks, vlan) | |
1456 | || bitmap_is_set(s->trunks, 0))) { | |
1457 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1458 | if (s->trunks) { | |
1459 | trunks = bitmap_clone(s->trunks, 4096); | |
1460 | } else { | |
1461 | trunks = bitmap_allocate1(4096); | |
1462 | } | |
1463 | bitmap_set1(trunks, vlan); | |
1464 | bitmap_set0(trunks, 0); | |
1465 | } else { | |
1466 | trunks = (unsigned long *) s->trunks; | |
1467 | } | |
1468 | break; | |
1469 | ||
1470 | default: | |
1471 | NOT_REACHED(); | |
1472 | } | |
abe529af BP |
1473 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
1474 | free(bundle->trunks); | |
ecac4ebf BP |
1475 | if (trunks == s->trunks) { |
1476 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1477 | } else { | |
1478 | bundle->trunks = trunks; | |
1479 | trunks = NULL; | |
1480 | } | |
abe529af BP |
1481 | need_flush = true; |
1482 | } | |
ecac4ebf BP |
1483 | if (trunks != s->trunks) { |
1484 | free(trunks); | |
1485 | } | |
abe529af BP |
1486 | |
1487 | /* Bonding. */ | |
1488 | if (!list_is_short(&bundle->ports)) { | |
1489 | bundle->ofproto->has_bonded_bundles = true; | |
1490 | if (bundle->bond) { | |
1491 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
1492 | ofproto->need_revalidate = true; | |
1493 | } | |
1494 | } else { | |
1495 | bundle->bond = bond_create(s->bond); | |
6f77f4ae | 1496 | ofproto->need_revalidate = true; |
abe529af BP |
1497 | } |
1498 | ||
1499 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 1500 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
1501 | port->up.netdev); |
1502 | } | |
1503 | } else { | |
1504 | bond_destroy(bundle->bond); | |
1505 | bundle->bond = NULL; | |
1506 | } | |
1507 | ||
1508 | /* If we changed something that would affect MAC learning, un-learn | |
1509 | * everything on this port and force flow revalidation. */ | |
1510 | if (need_flush) { | |
1511 | bundle_flush_macs(bundle); | |
1512 | } | |
1513 | ||
1514 | return 0; | |
1515 | } | |
1516 | ||
1517 | static void | |
1518 | bundle_remove(struct ofport *port_) | |
1519 | { | |
1520 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1521 | struct ofbundle *bundle = port->bundle; | |
1522 | ||
1523 | if (bundle) { | |
1524 | bundle_del_port(port); | |
1525 | if (list_is_empty(&bundle->ports)) { | |
1526 | bundle_destroy(bundle); | |
1527 | } else if (list_is_short(&bundle->ports)) { | |
1528 | bond_destroy(bundle->bond); | |
1529 | bundle->bond = NULL; | |
1530 | } | |
1531 | } | |
1532 | } | |
1533 | ||
1534 | static void | |
5f877369 | 1535 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
1536 | { |
1537 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
1538 | struct ofport_dpif *port = port_; | |
1539 | uint8_t ea[ETH_ADDR_LEN]; | |
1540 | int error; | |
1541 | ||
1542 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
1543 | if (!error) { | |
abe529af | 1544 | struct ofpbuf packet; |
5f877369 | 1545 | void *packet_pdu; |
abe529af BP |
1546 | |
1547 | ofpbuf_init(&packet, 0); | |
1548 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
1549 | pdu_size); |
1550 | memcpy(packet_pdu, pdu, pdu_size); | |
1551 | ||
abe529af BP |
1552 | error = netdev_send(port->up.netdev, &packet); |
1553 | if (error) { | |
1554 | VLOG_WARN_RL(&rl, "port %s: sending LACP PDU on iface %s failed " | |
1555 | "(%s)", port->bundle->name, | |
1556 | netdev_get_name(port->up.netdev), strerror(error)); | |
1557 | } | |
1558 | ofpbuf_uninit(&packet); | |
1559 | } else { | |
1560 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
1561 | "%s (%s)", port->bundle->name, | |
1562 | netdev_get_name(port->up.netdev), strerror(error)); | |
1563 | } | |
1564 | } | |
1565 | ||
1566 | static void | |
1567 | bundle_send_learning_packets(struct ofbundle *bundle) | |
1568 | { | |
1569 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1570 | int error, n_packets, n_errors; | |
1571 | struct mac_entry *e; | |
1572 | ||
1573 | error = n_packets = n_errors = 0; | |
1574 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
1575 | if (e->port.p != bundle) { | |
1576 | int ret = bond_send_learning_packet(bundle->bond, e->mac, e->vlan); | |
1577 | if (ret) { | |
1578 | error = ret; | |
1579 | n_errors++; | |
1580 | } | |
1581 | n_packets++; | |
1582 | } | |
1583 | } | |
1584 | ||
1585 | if (n_errors) { | |
1586 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1587 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
1588 | "packets, last error was: %s", | |
1589 | bundle->name, n_errors, n_packets, strerror(error)); | |
1590 | } else { | |
1591 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
1592 | bundle->name, n_packets); | |
1593 | } | |
1594 | } | |
1595 | ||
1596 | static void | |
1597 | bundle_run(struct ofbundle *bundle) | |
1598 | { | |
1599 | if (bundle->lacp) { | |
1600 | lacp_run(bundle->lacp, send_pdu_cb); | |
1601 | } | |
1602 | if (bundle->bond) { | |
1603 | struct ofport_dpif *port; | |
1604 | ||
1605 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 1606 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
1607 | } |
1608 | ||
1609 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
1610 | lacp_negotiated(bundle->lacp)); | |
1611 | if (bond_should_send_learning_packets(bundle->bond)) { | |
1612 | bundle_send_learning_packets(bundle); | |
1613 | } | |
1614 | } | |
1615 | } | |
1616 | ||
1617 | static void | |
1618 | bundle_wait(struct ofbundle *bundle) | |
1619 | { | |
1620 | if (bundle->lacp) { | |
1621 | lacp_wait(bundle->lacp); | |
1622 | } | |
1623 | if (bundle->bond) { | |
1624 | bond_wait(bundle->bond); | |
1625 | } | |
1626 | } | |
1627 | \f | |
1628 | /* Mirrors. */ | |
1629 | ||
1630 | static int | |
1631 | mirror_scan(struct ofproto_dpif *ofproto) | |
1632 | { | |
1633 | int idx; | |
1634 | ||
1635 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
1636 | if (!ofproto->mirrors[idx]) { | |
1637 | return idx; | |
1638 | } | |
1639 | } | |
1640 | return -1; | |
1641 | } | |
1642 | ||
1643 | static struct ofmirror * | |
1644 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
1645 | { | |
1646 | int i; | |
1647 | ||
1648 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1649 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
1650 | if (mirror && mirror->aux == aux) { | |
1651 | return mirror; | |
1652 | } | |
1653 | } | |
1654 | ||
1655 | return NULL; | |
1656 | } | |
1657 | ||
1658 | static int | |
1659 | mirror_set(struct ofproto *ofproto_, void *aux, | |
1660 | const struct ofproto_mirror_settings *s) | |
1661 | { | |
1662 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1663 | mirror_mask_t mirror_bit; | |
1664 | struct ofbundle *bundle; | |
1665 | struct ofmirror *mirror; | |
1666 | struct ofbundle *out; | |
1667 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
1668 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
1669 | int out_vlan; | |
1670 | ||
1671 | mirror = mirror_lookup(ofproto, aux); | |
1672 | if (!s) { | |
1673 | mirror_destroy(mirror); | |
1674 | return 0; | |
1675 | } | |
1676 | if (!mirror) { | |
1677 | int idx; | |
1678 | ||
1679 | idx = mirror_scan(ofproto); | |
1680 | if (idx < 0) { | |
1681 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
1682 | "cannot create %s", | |
1683 | ofproto->up.name, MAX_MIRRORS, s->name); | |
1684 | return EFBIG; | |
1685 | } | |
1686 | ||
1687 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
1688 | mirror->ofproto = ofproto; | |
1689 | mirror->idx = idx; | |
8b28d864 | 1690 | mirror->aux = aux; |
abe529af BP |
1691 | mirror->out_vlan = -1; |
1692 | mirror->name = NULL; | |
1693 | } | |
1694 | ||
1695 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
1696 | free(mirror->name); | |
1697 | mirror->name = xstrdup(s->name); | |
1698 | } | |
1699 | ||
1700 | /* Get the new configuration. */ | |
1701 | if (s->out_bundle) { | |
1702 | out = bundle_lookup(ofproto, s->out_bundle); | |
1703 | if (!out) { | |
1704 | mirror_destroy(mirror); | |
1705 | return EINVAL; | |
1706 | } | |
1707 | out_vlan = -1; | |
1708 | } else { | |
1709 | out = NULL; | |
1710 | out_vlan = s->out_vlan; | |
1711 | } | |
1712 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
1713 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
1714 | ||
1715 | /* If the configuration has not changed, do nothing. */ | |
1716 | if (hmapx_equals(&srcs, &mirror->srcs) | |
1717 | && hmapx_equals(&dsts, &mirror->dsts) | |
1718 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
1719 | && mirror->out == out | |
1720 | && mirror->out_vlan == out_vlan) | |
1721 | { | |
1722 | hmapx_destroy(&srcs); | |
1723 | hmapx_destroy(&dsts); | |
1724 | return 0; | |
1725 | } | |
1726 | ||
1727 | hmapx_swap(&srcs, &mirror->srcs); | |
1728 | hmapx_destroy(&srcs); | |
1729 | ||
1730 | hmapx_swap(&dsts, &mirror->dsts); | |
1731 | hmapx_destroy(&dsts); | |
1732 | ||
1733 | free(mirror->vlans); | |
1734 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
1735 | ||
1736 | mirror->out = out; | |
1737 | mirror->out_vlan = out_vlan; | |
1738 | ||
1739 | /* Update bundles. */ | |
1740 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1741 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
1742 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
1743 | bundle->src_mirrors |= mirror_bit; | |
1744 | } else { | |
1745 | bundle->src_mirrors &= ~mirror_bit; | |
1746 | } | |
1747 | ||
1748 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
1749 | bundle->dst_mirrors |= mirror_bit; | |
1750 | } else { | |
1751 | bundle->dst_mirrors &= ~mirror_bit; | |
1752 | } | |
1753 | ||
1754 | if (mirror->out == bundle) { | |
1755 | bundle->mirror_out |= mirror_bit; | |
1756 | } else { | |
1757 | bundle->mirror_out &= ~mirror_bit; | |
1758 | } | |
1759 | } | |
1760 | ||
1761 | ofproto->need_revalidate = true; | |
1762 | mac_learning_flush(ofproto->ml); | |
1763 | ||
1764 | return 0; | |
1765 | } | |
1766 | ||
1767 | static void | |
1768 | mirror_destroy(struct ofmirror *mirror) | |
1769 | { | |
1770 | struct ofproto_dpif *ofproto; | |
1771 | mirror_mask_t mirror_bit; | |
1772 | struct ofbundle *bundle; | |
1773 | ||
1774 | if (!mirror) { | |
1775 | return; | |
1776 | } | |
1777 | ||
1778 | ofproto = mirror->ofproto; | |
1779 | ofproto->need_revalidate = true; | |
1780 | mac_learning_flush(ofproto->ml); | |
1781 | ||
1782 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1783 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1784 | bundle->src_mirrors &= ~mirror_bit; | |
1785 | bundle->dst_mirrors &= ~mirror_bit; | |
1786 | bundle->mirror_out &= ~mirror_bit; | |
1787 | } | |
1788 | ||
1789 | hmapx_destroy(&mirror->srcs); | |
1790 | hmapx_destroy(&mirror->dsts); | |
1791 | free(mirror->vlans); | |
1792 | ||
1793 | ofproto->mirrors[mirror->idx] = NULL; | |
1794 | free(mirror->name); | |
1795 | free(mirror); | |
1796 | } | |
1797 | ||
1798 | static int | |
1799 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
1800 | { | |
1801 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1802 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
1803 | ofproto->need_revalidate = true; | |
1804 | mac_learning_flush(ofproto->ml); | |
1805 | } | |
1806 | return 0; | |
1807 | } | |
1808 | ||
1809 | static bool | |
b4affc74 | 1810 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
1811 | { |
1812 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1813 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
1814 | return bundle && bundle->mirror_out != 0; | |
1815 | } | |
8402c74b SS |
1816 | |
1817 | static void | |
b53055f4 | 1818 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
1819 | { |
1820 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1821 | /* Revalidate cached flows whenever forward_bpdu option changes. */ | |
1822 | ofproto->need_revalidate = true; | |
1823 | } | |
abe529af BP |
1824 | \f |
1825 | /* Ports. */ | |
1826 | ||
1827 | static struct ofport_dpif * | |
1828 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
1829 | { | |
7df6a8bd BP |
1830 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
1831 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
1832 | } |
1833 | ||
1834 | static struct ofport_dpif * | |
1835 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
1836 | { | |
1837 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
1838 | } | |
1839 | ||
1840 | static void | |
1841 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
1842 | struct dpif_port *dpif_port) | |
1843 | { | |
1844 | ofproto_port->name = dpif_port->name; | |
1845 | ofproto_port->type = dpif_port->type; | |
1846 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
1847 | } | |
1848 | ||
1849 | static void | |
1850 | port_run(struct ofport_dpif *ofport) | |
1851 | { | |
015e08bc EJ |
1852 | bool enable = netdev_get_carrier(ofport->up.netdev); |
1853 | ||
abe529af BP |
1854 | if (ofport->cfm) { |
1855 | cfm_run(ofport->cfm); | |
1856 | ||
1857 | if (cfm_should_send_ccm(ofport->cfm)) { | |
1858 | struct ofpbuf packet; | |
abe529af BP |
1859 | |
1860 | ofpbuf_init(&packet, 0); | |
c0a2e71d | 1861 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr); |
abe529af | 1862 | send_packet(ofproto_dpif_cast(ofport->up.ofproto), |
b2fda3ef | 1863 | ofport->odp_port, &packet); |
abe529af BP |
1864 | ofpbuf_uninit(&packet); |
1865 | } | |
015e08bc | 1866 | |
86dc6501 EJ |
1867 | enable = enable && !cfm_get_fault(ofport->cfm) |
1868 | && cfm_get_opup(ofport->cfm); | |
abe529af | 1869 | } |
015e08bc EJ |
1870 | |
1871 | if (ofport->bundle) { | |
1872 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
1873 | } | |
1874 | ||
daff3353 EJ |
1875 | if (ofport->may_enable != enable) { |
1876 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1877 | ||
1878 | if (ofproto->has_bundle_action) { | |
1879 | ofproto->need_revalidate = true; | |
1880 | } | |
1881 | } | |
1882 | ||
015e08bc | 1883 | ofport->may_enable = enable; |
abe529af BP |
1884 | } |
1885 | ||
1886 | static void | |
1887 | port_wait(struct ofport_dpif *ofport) | |
1888 | { | |
1889 | if (ofport->cfm) { | |
1890 | cfm_wait(ofport->cfm); | |
1891 | } | |
1892 | } | |
1893 | ||
1894 | static int | |
1895 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
1896 | struct ofproto_port *ofproto_port) | |
1897 | { | |
1898 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1899 | struct dpif_port dpif_port; | |
1900 | int error; | |
1901 | ||
1902 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
1903 | if (!error) { | |
1904 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
1905 | } | |
1906 | return error; | |
1907 | } | |
1908 | ||
1909 | static int | |
1910 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
1911 | { | |
1912 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1913 | uint16_t odp_port; | |
1914 | int error; | |
1915 | ||
1916 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
1917 | if (!error) { | |
1918 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
1919 | } | |
1920 | return error; | |
1921 | } | |
1922 | ||
1923 | static int | |
1924 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
1925 | { | |
1926 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1927 | int error; | |
1928 | ||
1929 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
1930 | if (!error) { | |
1931 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
1932 | if (ofport) { | |
1933 | /* The caller is going to close ofport->up.netdev. If this is a | |
1934 | * bonded port, then the bond is using that netdev, so remove it | |
1935 | * from the bond. The client will need to reconfigure everything | |
1936 | * after deleting ports, so then the slave will get re-added. */ | |
1937 | bundle_remove(&ofport->up); | |
1938 | } | |
1939 | } | |
1940 | return error; | |
1941 | } | |
1942 | ||
1943 | struct port_dump_state { | |
1944 | struct dpif_port_dump dump; | |
1945 | bool done; | |
1946 | }; | |
1947 | ||
1948 | static int | |
1949 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
1950 | { | |
1951 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1952 | struct port_dump_state *state; | |
1953 | ||
1954 | *statep = state = xmalloc(sizeof *state); | |
1955 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
1956 | state->done = false; | |
1957 | return 0; | |
1958 | } | |
1959 | ||
1960 | static int | |
1961 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
1962 | struct ofproto_port *port) | |
1963 | { | |
1964 | struct port_dump_state *state = state_; | |
1965 | struct dpif_port dpif_port; | |
1966 | ||
1967 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
1968 | ofproto_port_from_dpif_port(port, &dpif_port); | |
1969 | return 0; | |
1970 | } else { | |
1971 | int error = dpif_port_dump_done(&state->dump); | |
1972 | state->done = true; | |
1973 | return error ? error : EOF; | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | static int | |
1978 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
1979 | { | |
1980 | struct port_dump_state *state = state_; | |
1981 | ||
1982 | if (!state->done) { | |
1983 | dpif_port_dump_done(&state->dump); | |
1984 | } | |
1985 | free(state); | |
1986 | return 0; | |
1987 | } | |
1988 | ||
1989 | static int | |
1990 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
1991 | { | |
1992 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1993 | return dpif_port_poll(ofproto->dpif, devnamep); | |
1994 | } | |
1995 | ||
1996 | static void | |
1997 | port_poll_wait(const struct ofproto *ofproto_) | |
1998 | { | |
1999 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2000 | dpif_port_poll_wait(ofproto->dpif); | |
2001 | } | |
2002 | ||
2003 | static int | |
2004 | port_is_lacp_current(const struct ofport *ofport_) | |
2005 | { | |
2006 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2007 | return (ofport->bundle && ofport->bundle->lacp | |
2008 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2009 | : -1); | |
2010 | } | |
2011 | \f | |
2012 | /* Upcall handling. */ | |
2013 | ||
501f8d1f BP |
2014 | /* Flow miss batching. |
2015 | * | |
2016 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2017 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2018 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2019 | * more packets, plus possibly installing the flow in the dpif. | |
2020 | * | |
2021 | * So far we only batch the operations that affect flow setup time the most. | |
2022 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2023 | struct flow_miss { | |
2024 | struct hmap_node hmap_node; | |
2025 | struct flow flow; | |
2026 | const struct nlattr *key; | |
2027 | size_t key_len; | |
2028 | struct list packets; | |
2029 | }; | |
2030 | ||
2031 | struct flow_miss_op { | |
2032 | union dpif_op dpif_op; | |
2033 | struct facet *facet; | |
2034 | }; | |
2035 | ||
62cd7072 BP |
2036 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
2037 | * OpenFlow controller as necessary according to their individual | |
2038 | * configurations. | |
2039 | * | |
2040 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, | |
2041 | * ownership is transferred to this function. */ | |
2042 | static void | |
2043 | send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet, | |
2044 | const struct flow *flow, bool clone) | |
2045 | { | |
2046 | struct ofputil_packet_in pin; | |
2047 | ||
2048 | pin.packet = packet; | |
2049 | pin.in_port = flow->in_port; | |
2050 | pin.reason = OFPR_NO_MATCH; | |
2051 | pin.buffer_id = 0; /* not yet known */ | |
2052 | pin.send_len = 0; /* not used for flow table misses */ | |
2053 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, | |
2054 | clone ? NULL : packet); | |
2055 | } | |
2056 | ||
2057 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each | |
2058 | * OpenFlow controller as necessary according to their individual | |
2059 | * configurations. | |
2060 | * | |
2061 | * 'send_len' should be the number of bytes of 'packet' to send to the | |
2062 | * controller, as specified in the action that caused the packet to be sent. | |
abe529af BP |
2063 | * |
2064 | * If 'clone' is true, the caller retains ownership of 'upcall->packet'. | |
2065 | * Otherwise, ownership is transferred to this function. */ | |
2066 | static void | |
62cd7072 BP |
2067 | send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet, |
2068 | uint64_t userdata, const struct flow *flow, bool clone) | |
abe529af BP |
2069 | { |
2070 | struct ofputil_packet_in pin; | |
6ff686f2 | 2071 | struct user_action_cookie cookie; |
abe529af | 2072 | |
62cd7072 BP |
2073 | memcpy(&cookie, &userdata, sizeof(cookie)); |
2074 | ||
2075 | pin.packet = packet; | |
abe529af | 2076 | pin.in_port = flow->in_port; |
62cd7072 | 2077 | pin.reason = OFPR_ACTION; |
abe529af | 2078 | pin.buffer_id = 0; /* not yet known */ |
6ff686f2 | 2079 | pin.send_len = cookie.data; |
78bd1cd0 | 2080 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, |
62cd7072 | 2081 | clone ? NULL : packet); |
abe529af BP |
2082 | } |
2083 | ||
2084 | static bool | |
2085 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2086 | const struct ofpbuf *packet) | |
2087 | { | |
b6e001b6 EJ |
2088 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
2089 | ||
2090 | if (!ofport) { | |
2091 | return false; | |
2092 | } | |
2093 | ||
ef9819b5 | 2094 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 2095 | if (packet) { |
abe529af BP |
2096 | cfm_process_heartbeat(ofport->cfm, packet); |
2097 | } | |
2098 | return true; | |
b6e001b6 EJ |
2099 | } else if (ofport->bundle && ofport->bundle->lacp |
2100 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2101 | if (packet) { | |
2102 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 2103 | } |
da37ebac | 2104 | return true; |
21f7563c JP |
2105 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
2106 | if (packet) { | |
2107 | stp_process_packet(ofport, packet); | |
2108 | } | |
2109 | return true; | |
abe529af BP |
2110 | } |
2111 | return false; | |
2112 | } | |
2113 | ||
501f8d1f BP |
2114 | static struct flow_miss * |
2115 | flow_miss_create(struct hmap *todo, const struct flow *flow, | |
2116 | const struct nlattr *key, size_t key_len) | |
abe529af | 2117 | { |
501f8d1f BP |
2118 | uint32_t hash = flow_hash(flow, 0); |
2119 | struct flow_miss *miss; | |
abe529af | 2120 | |
501f8d1f BP |
2121 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
2122 | if (flow_equal(&miss->flow, flow)) { | |
2123 | return miss; | |
2124 | } | |
2125 | } | |
abe529af | 2126 | |
501f8d1f BP |
2127 | miss = xmalloc(sizeof *miss); |
2128 | hmap_insert(todo, &miss->hmap_node, hash); | |
2129 | miss->flow = *flow; | |
2130 | miss->key = key; | |
2131 | miss->key_len = key_len; | |
2132 | list_init(&miss->packets); | |
2133 | return miss; | |
2134 | } | |
abe529af | 2135 | |
501f8d1f BP |
2136 | static void |
2137 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2138 | struct flow_miss_op *ops, size_t *n_ops) | |
2139 | { | |
2140 | const struct flow *flow = &miss->flow; | |
2141 | struct ofpbuf *packet, *next_packet; | |
2142 | struct facet *facet; | |
abe529af | 2143 | |
501f8d1f | 2144 | facet = facet_lookup_valid(ofproto, flow); |
abe529af | 2145 | if (!facet) { |
501f8d1f BP |
2146 | struct rule_dpif *rule; |
2147 | ||
2148 | rule = rule_dpif_lookup(ofproto, flow, 0); | |
abe529af BP |
2149 | if (!rule) { |
2150 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
501f8d1f | 2151 | struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); |
abe529af BP |
2152 | if (port) { |
2153 | if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) { | |
2154 | COVERAGE_INC(ofproto_dpif_no_packet_in); | |
2155 | /* XXX install 'drop' flow entry */ | |
abe529af BP |
2156 | return; |
2157 | } | |
2158 | } else { | |
2159 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
501f8d1f BP |
2160 | flow->in_port); |
2161 | } | |
2162 | ||
2163 | LIST_FOR_EACH_SAFE (packet, next_packet, list_node, | |
2164 | &miss->packets) { | |
2165 | list_remove(&packet->list_node); | |
2166 | send_packet_in_miss(ofproto, packet, flow, false); | |
abe529af BP |
2167 | } |
2168 | ||
abe529af BP |
2169 | return; |
2170 | } | |
2171 | ||
501f8d1f | 2172 | facet = facet_create(rule, flow); |
abe529af BP |
2173 | } |
2174 | ||
501f8d1f BP |
2175 | LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) { |
2176 | list_remove(&packet->list_node); | |
2177 | ofproto->n_matches++; | |
2178 | ||
2179 | if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2180 | /* | |
2181 | * Extra-special case for fail-open mode. | |
2182 | * | |
2183 | * We are in fail-open mode and the packet matched the fail-open | |
2184 | * rule, but we are connected to a controller too. We should send | |
2185 | * the packet up to the controller in the hope that it will try to | |
2186 | * set up a flow and thereby allow us to exit fail-open. | |
2187 | * | |
2188 | * See the top-level comment in fail-open.c for more information. | |
2189 | */ | |
2190 | send_packet_in_miss(ofproto, packet, flow, true); | |
2191 | } | |
2192 | ||
2193 | if (!facet->may_install) { | |
2194 | facet_make_actions(ofproto, facet, packet); | |
2195 | } | |
2196 | if (!execute_controller_action(ofproto, &facet->flow, | |
2197 | facet->actions, facet->actions_len, | |
2198 | packet)) { | |
2199 | struct flow_miss_op *op = &ops[(*n_ops)++]; | |
2200 | struct dpif_execute *execute = &op->dpif_op.execute; | |
2201 | ||
2202 | op->facet = facet; | |
2203 | execute->type = DPIF_OP_EXECUTE; | |
2204 | execute->key = miss->key; | |
2205 | execute->key_len = miss->key_len; | |
2206 | execute->actions | |
2207 | = (facet->may_install | |
2208 | ? facet->actions | |
2209 | : xmemdup(facet->actions, facet->actions_len)); | |
2210 | execute->actions_len = facet->actions_len; | |
2211 | execute->packet = packet; | |
2212 | } | |
2213 | } | |
2214 | ||
2215 | if (facet->may_install) { | |
2216 | struct flow_miss_op *op = &ops[(*n_ops)++]; | |
2217 | struct dpif_flow_put *put = &op->dpif_op.flow_put; | |
2218 | ||
2219 | op->facet = facet; | |
2220 | put->type = DPIF_OP_FLOW_PUT; | |
2221 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
2222 | put->key = miss->key; | |
2223 | put->key_len = miss->key_len; | |
2224 | put->actions = facet->actions; | |
2225 | put->actions_len = facet->actions_len; | |
2226 | put->stats = NULL; | |
2227 | } | |
2228 | } | |
2229 | ||
2230 | static void | |
2231 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
2232 | size_t n_upcalls) | |
2233 | { | |
2234 | struct dpif_upcall *upcall; | |
2235 | struct flow_miss *miss, *next_miss; | |
2236 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; | |
2237 | union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; | |
2238 | struct hmap todo; | |
2239 | size_t n_ops; | |
2240 | size_t i; | |
2241 | ||
2242 | if (!n_upcalls) { | |
2243 | return; | |
2244 | } | |
2245 | ||
2246 | /* Construct the to-do list. | |
2247 | * | |
2248 | * This just amounts to extracting the flow from each packet and sticking | |
2249 | * the packets that have the same flow in the same "flow_miss" structure so | |
2250 | * that we can process them together. */ | |
2251 | hmap_init(&todo); | |
2252 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { | |
2253 | struct flow_miss *miss; | |
2254 | struct flow flow; | |
2255 | ||
2256 | /* Obtain in_port and tun_id, at least, then set 'flow''s header | |
2257 | * pointers. */ | |
2258 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
abff858b PS |
2259 | flow_extract(upcall->packet, flow.priority, flow.tun_id, |
2260 | flow.in_port, &flow); | |
501f8d1f | 2261 | |
21f7563c | 2262 | /* Handle 802.1ag, LACP, and STP specially. */ |
501f8d1f BP |
2263 | if (process_special(ofproto, &flow, upcall->packet)) { |
2264 | ofpbuf_delete(upcall->packet); | |
2265 | ofproto->n_matches++; | |
2266 | continue; | |
2267 | } | |
2268 | ||
2269 | /* Add other packets to a to-do list. */ | |
2270 | miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len); | |
2271 | list_push_back(&miss->packets, &upcall->packet->list_node); | |
2272 | } | |
2273 | ||
2274 | /* Process each element in the to-do list, constructing the set of | |
2275 | * operations to batch. */ | |
2276 | n_ops = 0; | |
2277 | HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) { | |
2278 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); | |
2279 | ofpbuf_list_delete(&miss->packets); | |
2280 | hmap_remove(&todo, &miss->hmap_node); | |
2281 | free(miss); | |
abe529af | 2282 | } |
501f8d1f BP |
2283 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
2284 | hmap_destroy(&todo); | |
2285 | ||
2286 | /* Execute batch. */ | |
2287 | for (i = 0; i < n_ops; i++) { | |
2288 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
2289 | } | |
2290 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
2291 | ||
2292 | /* Free memory and update facets. */ | |
2293 | for (i = 0; i < n_ops; i++) { | |
2294 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
2295 | struct dpif_execute *execute; | |
2296 | struct dpif_flow_put *put; | |
2297 | ||
2298 | switch (op->dpif_op.type) { | |
2299 | case DPIF_OP_EXECUTE: | |
2300 | execute = &op->dpif_op.execute; | |
2301 | if (op->facet->actions != execute->actions) { | |
2302 | free((struct nlattr *) execute->actions); | |
2303 | } | |
2304 | ofpbuf_delete((struct ofpbuf *) execute->packet); | |
2305 | break; | |
abe529af | 2306 | |
501f8d1f BP |
2307 | case DPIF_OP_FLOW_PUT: |
2308 | put = &op->dpif_op.flow_put; | |
2309 | if (!put->error) { | |
2310 | op->facet->installed = true; | |
2311 | } | |
2312 | break; | |
2313 | } | |
2314 | } | |
abe529af BP |
2315 | } |
2316 | ||
2317 | static void | |
6ff686f2 PS |
2318 | handle_userspace_upcall(struct ofproto_dpif *ofproto, |
2319 | struct dpif_upcall *upcall) | |
abe529af BP |
2320 | { |
2321 | struct flow flow; | |
6ff686f2 | 2322 | struct user_action_cookie cookie; |
abe529af | 2323 | |
6ff686f2 | 2324 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
abe529af | 2325 | |
6ff686f2 | 2326 | if (cookie.type == USER_ACTION_COOKIE_SFLOW) { |
abe529af BP |
2327 | if (ofproto->sflow) { |
2328 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
6ff686f2 | 2329 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie); |
abe529af BP |
2330 | } |
2331 | ofpbuf_delete(upcall->packet); | |
6ff686f2 PS |
2332 | |
2333 | } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) { | |
2334 | COVERAGE_INC(ofproto_dpif_ctlr_action); | |
2335 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
62cd7072 BP |
2336 | send_packet_in_action(ofproto, upcall->packet, upcall->userdata, |
2337 | &flow, false); | |
6ff686f2 PS |
2338 | } else { |
2339 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
2340 | } | |
2341 | } | |
2342 | ||
2343 | static void | |
2344 | handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) | |
2345 | { | |
2346 | switch (upcall->type) { | |
2347 | case DPIF_UC_ACTION: | |
2348 | handle_userspace_upcall(ofproto, upcall); | |
abe529af BP |
2349 | break; |
2350 | ||
2351 | case DPIF_UC_MISS: | |
501f8d1f BP |
2352 | /* The caller handles these. */ |
2353 | NOT_REACHED(); | |
abe529af BP |
2354 | |
2355 | case DPIF_N_UC_TYPES: | |
2356 | default: | |
2357 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
2358 | break; | |
2359 | } | |
2360 | } | |
2361 | \f | |
2362 | /* Flow expiration. */ | |
2363 | ||
2364 | static int facet_max_idle(const struct ofproto_dpif *); | |
2365 | static void update_stats(struct ofproto_dpif *); | |
2366 | static void rule_expire(struct rule_dpif *); | |
2367 | static void expire_facets(struct ofproto_dpif *, int dp_max_idle); | |
2368 | ||
2369 | /* This function is called periodically by run(). Its job is to collect | |
2370 | * updates for the flows that have been installed into the datapath, most | |
2371 | * importantly when they last were used, and then use that information to | |
2372 | * expire flows that have not been used recently. | |
2373 | * | |
2374 | * Returns the number of milliseconds after which it should be called again. */ | |
2375 | static int | |
2376 | expire(struct ofproto_dpif *ofproto) | |
2377 | { | |
2378 | struct rule_dpif *rule, *next_rule; | |
0697b5c3 | 2379 | struct classifier *table; |
abe529af BP |
2380 | int dp_max_idle; |
2381 | ||
2382 | /* Update stats for each flow in the datapath. */ | |
2383 | update_stats(ofproto); | |
2384 | ||
2385 | /* Expire facets that have been idle too long. */ | |
2386 | dp_max_idle = facet_max_idle(ofproto); | |
2387 | expire_facets(ofproto, dp_max_idle); | |
2388 | ||
2389 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
2390 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
2391 | struct cls_cursor cursor; | |
2392 | ||
2393 | cls_cursor_init(&cursor, table, NULL); | |
2394 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
2395 | rule_expire(rule); | |
2396 | } | |
abe529af BP |
2397 | } |
2398 | ||
2399 | /* All outstanding data in existing flows has been accounted, so it's a | |
2400 | * good time to do bond rebalancing. */ | |
2401 | if (ofproto->has_bonded_bundles) { | |
2402 | struct ofbundle *bundle; | |
2403 | ||
2404 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2405 | if (bundle->bond) { | |
2406 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
2407 | } | |
2408 | } | |
2409 | } | |
2410 | ||
2411 | return MIN(dp_max_idle, 1000); | |
2412 | } | |
2413 | ||
2414 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
2415 | * | |
2416 | * This function also pushes statistics updates to rules which each facet | |
2417 | * resubmits into. Generally these statistics will be accurate. However, if a | |
2418 | * facet changes the rule it resubmits into at some time in between | |
2419 | * update_stats() runs, it is possible that statistics accrued to the | |
2420 | * old rule will be incorrectly attributed to the new rule. This could be | |
2421 | * avoided by calling update_stats() whenever rules are created or | |
2422 | * deleted. However, the performance impact of making so many calls to the | |
2423 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
2424 | */ | |
2425 | static void | |
2426 | update_stats(struct ofproto_dpif *p) | |
2427 | { | |
2428 | const struct dpif_flow_stats *stats; | |
2429 | struct dpif_flow_dump dump; | |
2430 | const struct nlattr *key; | |
2431 | size_t key_len; | |
2432 | ||
2433 | dpif_flow_dump_start(&dump, p->dpif); | |
2434 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
2435 | struct facet *facet; | |
2436 | struct flow flow; | |
2437 | ||
2438 | if (odp_flow_key_to_flow(key, key_len, &flow)) { | |
2439 | struct ds s; | |
2440 | ||
2441 | ds_init(&s); | |
2442 | odp_flow_key_format(key, key_len, &s); | |
df2c07f4 | 2443 | VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s", |
abe529af BP |
2444 | ds_cstr(&s)); |
2445 | ds_destroy(&s); | |
2446 | ||
2447 | continue; | |
2448 | } | |
2449 | facet = facet_find(p, &flow); | |
2450 | ||
2451 | if (facet && facet->installed) { | |
2452 | ||
2453 | if (stats->n_packets >= facet->dp_packet_count) { | |
2454 | uint64_t extra = stats->n_packets - facet->dp_packet_count; | |
2455 | facet->packet_count += extra; | |
2456 | } else { | |
2457 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
2458 | } | |
2459 | ||
2460 | if (stats->n_bytes >= facet->dp_byte_count) { | |
2461 | facet->byte_count += stats->n_bytes - facet->dp_byte_count; | |
2462 | } else { | |
2463 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
2464 | } | |
2465 | ||
2466 | facet->dp_packet_count = stats->n_packets; | |
2467 | facet->dp_byte_count = stats->n_bytes; | |
2468 | ||
2469 | facet_update_time(p, facet, stats->used); | |
55af77bb | 2470 | facet_account(p, facet); |
abe529af BP |
2471 | facet_push_stats(facet); |
2472 | } else { | |
2473 | /* There's a flow in the datapath that we know nothing about. | |
2474 | * Delete it. */ | |
2475 | COVERAGE_INC(facet_unexpected); | |
2476 | dpif_flow_del(p->dpif, key, key_len, NULL); | |
2477 | } | |
2478 | } | |
2479 | dpif_flow_dump_done(&dump); | |
2480 | } | |
2481 | ||
2482 | /* Calculates and returns the number of milliseconds of idle time after which | |
2483 | * facets should expire from the datapath and we should fold their statistics | |
2484 | * into their parent rules in userspace. */ | |
2485 | static int | |
2486 | facet_max_idle(const struct ofproto_dpif *ofproto) | |
2487 | { | |
2488 | /* | |
2489 | * Idle time histogram. | |
2490 | * | |
2491 | * Most of the time a switch has a relatively small number of facets. When | |
2492 | * this is the case we might as well keep statistics for all of them in | |
2493 | * userspace and to cache them in the kernel datapath for performance as | |
2494 | * well. | |
2495 | * | |
2496 | * As the number of facets increases, the memory required to maintain | |
2497 | * statistics about them in userspace and in the kernel becomes | |
2498 | * significant. However, with a large number of facets it is likely that | |
2499 | * only a few of them are "heavy hitters" that consume a large amount of | |
2500 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
2501 | * kernel and maintaining in userspaces; other facets we can discard. | |
2502 | * | |
2503 | * The technique used to compute the idle time is to build a histogram with | |
2504 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet | |
2505 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
2506 | * After the histogram has been built, we compute the cutoff so that only | |
084f5290 SH |
2507 | * the most-recently-used 1% of facets (but at least |
2508 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least | |
2509 | * the most-recently-used bucket of facets is kept, so actually an | |
2510 | * arbitrary number of facets can be kept in any given expiration run | |
2511 | * (though the next run will delete most of those unless they receive | |
2512 | * additional data). | |
abe529af BP |
2513 | * |
2514 | * This requires a second pass through the facets, in addition to the pass | |
2515 | * made by update_stats(), because the former function never looks | |
2516 | * at uninstallable facets. | |
2517 | */ | |
2518 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
2519 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
2520 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 2521 | int total, subtotal, bucket; |
abe529af | 2522 | struct facet *facet; |
abe529af BP |
2523 | long long int now; |
2524 | int i; | |
2525 | ||
2526 | total = hmap_count(&ofproto->facets); | |
084f5290 | 2527 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
2528 | return N_BUCKETS * BUCKET_WIDTH; |
2529 | } | |
2530 | ||
2531 | /* Build histogram. */ | |
2532 | now = time_msec(); | |
2533 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
2534 | long long int idle = now - facet->used; | |
2535 | int bucket = (idle <= 0 ? 0 | |
2536 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
2537 | : (unsigned int) idle / BUCKET_WIDTH); | |
2538 | buckets[bucket]++; | |
2539 | } | |
2540 | ||
2541 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
2542 | subtotal = bucket = 0; |
2543 | do { | |
2544 | subtotal += buckets[bucket++]; | |
084f5290 SH |
2545 | } while (bucket < N_BUCKETS && |
2546 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
2547 | |
2548 | if (VLOG_IS_DBG_ENABLED()) { | |
2549 | struct ds s; | |
2550 | ||
2551 | ds_init(&s); | |
2552 | ds_put_cstr(&s, "keep"); | |
2553 | for (i = 0; i < N_BUCKETS; i++) { | |
2554 | if (i == bucket) { | |
2555 | ds_put_cstr(&s, ", drop"); | |
2556 | } | |
2557 | if (buckets[i]) { | |
2558 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
2559 | } | |
2560 | } | |
2561 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
2562 | ds_destroy(&s); | |
2563 | } | |
2564 | ||
2565 | return bucket * BUCKET_WIDTH; | |
2566 | } | |
2567 | ||
2568 | static void | |
2569 | facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
2570 | { | |
2571 | if (ofproto->netflow && !facet_is_controller_flow(facet) && | |
2572 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
2573 | struct ofexpired expired; | |
2574 | ||
2575 | if (facet->installed) { | |
2576 | struct dpif_flow_stats stats; | |
2577 | ||
2578 | facet_put__(ofproto, facet, facet->actions, facet->actions_len, | |
2579 | &stats); | |
2580 | facet_update_stats(ofproto, facet, &stats); | |
2581 | } | |
2582 | ||
2583 | expired.flow = facet->flow; | |
2584 | expired.packet_count = facet->packet_count; | |
2585 | expired.byte_count = facet->byte_count; | |
2586 | expired.used = facet->used; | |
2587 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2588 | } | |
2589 | } | |
2590 | ||
2591 | static void | |
2592 | expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle) | |
2593 | { | |
2594 | long long int cutoff = time_msec() - dp_max_idle; | |
2595 | struct facet *facet, *next_facet; | |
2596 | ||
2597 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
2598 | facet_active_timeout(ofproto, facet); | |
2599 | if (facet->used < cutoff) { | |
2600 | facet_remove(ofproto, facet); | |
2601 | } | |
2602 | } | |
2603 | } | |
2604 | ||
2605 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
2606 | * then delete it entirely. */ | |
2607 | static void | |
2608 | rule_expire(struct rule_dpif *rule) | |
2609 | { | |
2610 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2611 | struct facet *facet, *next_facet; | |
2612 | long long int now; | |
2613 | uint8_t reason; | |
2614 | ||
2615 | /* Has 'rule' expired? */ | |
2616 | now = time_msec(); | |
2617 | if (rule->up.hard_timeout | |
308881af | 2618 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af BP |
2619 | reason = OFPRR_HARD_TIMEOUT; |
2620 | } else if (rule->up.idle_timeout && list_is_empty(&rule->facets) | |
2621 | && now > rule->used + rule->up.idle_timeout * 1000) { | |
2622 | reason = OFPRR_IDLE_TIMEOUT; | |
2623 | } else { | |
2624 | return; | |
2625 | } | |
2626 | ||
2627 | COVERAGE_INC(ofproto_dpif_expired); | |
2628 | ||
2629 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
2630 | * timeout, because that only happens when the rule has no facets left.) */ | |
2631 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
2632 | facet_remove(ofproto, facet); | |
2633 | } | |
2634 | ||
2635 | /* Get rid of the rule. */ | |
2636 | ofproto_rule_expire(&rule->up, reason); | |
2637 | } | |
2638 | \f | |
2639 | /* Facets. */ | |
2640 | ||
f3827897 | 2641 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
2642 | * |
2643 | * The caller must already have determined that no facet with an identical | |
2644 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
2645 | * the ofproto's classifier table. |
2646 | * | |
2647 | * The facet will initially have no ODP actions. The caller should fix that | |
2648 | * by calling facet_make_actions(). */ | |
abe529af | 2649 | static struct facet * |
f3827897 | 2650 | facet_create(struct rule_dpif *rule, const struct flow *flow) |
abe529af BP |
2651 | { |
2652 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2653 | struct facet *facet; | |
2654 | ||
2655 | facet = xzalloc(sizeof *facet); | |
2656 | facet->used = time_msec(); | |
2657 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2658 | list_push_back(&rule->facets, &facet->list_node); | |
2659 | facet->rule = rule; | |
2660 | facet->flow = *flow; | |
2661 | netflow_flow_init(&facet->nf_flow); | |
2662 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2663 | ||
abe529af BP |
2664 | return facet; |
2665 | } | |
2666 | ||
2667 | static void | |
2668 | facet_free(struct facet *facet) | |
2669 | { | |
2670 | free(facet->actions); | |
2671 | free(facet); | |
2672 | } | |
2673 | ||
abe529af | 2674 | static bool |
3d9e05f8 BP |
2675 | execute_controller_action(struct ofproto_dpif *ofproto, |
2676 | const struct flow *flow, | |
2677 | const struct nlattr *odp_actions, size_t actions_len, | |
2678 | struct ofpbuf *packet) | |
2679 | { | |
2680 | if (actions_len | |
2681 | && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE | |
2682 | && NLA_ALIGN(odp_actions->nla_len) == actions_len) { | |
62cd7072 BP |
2683 | /* As an optimization, avoid a round-trip from userspace to kernel to |
2684 | * userspace. This also avoids possibly filling up kernel packet | |
2685 | * buffers along the way. | |
2686 | * | |
2687 | * This optimization will not accidentally catch sFlow | |
2688 | * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated | |
2689 | * inside OVS_ACTION_ATTR_SAMPLE. */ | |
2690 | const struct nlattr *nla; | |
2691 | ||
2692 | nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA); | |
2693 | send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow, | |
2694 | false); | |
2695 | return true; | |
3d9e05f8 BP |
2696 | } else { |
2697 | return false; | |
2698 | } | |
2699 | } | |
2700 | ||
2701 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on | |
2702 | * 'packet', which arrived on 'in_port'. | |
2703 | * | |
2704 | * Takes ownership of 'packet'. */ | |
2705 | static bool | |
2706 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2707 | const struct nlattr *odp_actions, size_t actions_len, | |
2708 | struct ofpbuf *packet) | |
2709 | { | |
2710 | struct odputil_keybuf keybuf; | |
2711 | struct ofpbuf key; | |
2712 | int error; | |
2713 | ||
2714 | if (execute_controller_action(ofproto, flow, odp_actions, actions_len, | |
2715 | packet)) { | |
2716 | return true; | |
6ff686f2 | 2717 | } |
abe529af | 2718 | |
6ff686f2 PS |
2719 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
2720 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 2721 | |
6ff686f2 PS |
2722 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
2723 | odp_actions, actions_len, packet); | |
80e5eed9 | 2724 | |
6ff686f2 PS |
2725 | ofpbuf_delete(packet); |
2726 | return !error; | |
abe529af BP |
2727 | } |
2728 | ||
2729 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s | |
2730 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
2731 | * ofp_packet_in) bytes of headroom. | |
2732 | * | |
2733 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, | |
2734 | * applying flow_extract() to 'packet' would yield the same flow as | |
2735 | * 'facet->flow'. | |
2736 | * | |
df2c07f4 JP |
2737 | * 'facet' must have accurately composed datapath actions; that is, it must |
2738 | * not be in need of revalidation. | |
abe529af BP |
2739 | * |
2740 | * Takes ownership of 'packet'. */ | |
2741 | static void | |
2742 | facet_execute(struct ofproto_dpif *ofproto, struct facet *facet, | |
2743 | struct ofpbuf *packet) | |
2744 | { | |
2745 | struct dpif_flow_stats stats; | |
2746 | ||
2747 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2748 | ||
572b7068 | 2749 | dpif_flow_stats_extract(&facet->flow, packet, &stats); |
abe529af BP |
2750 | stats.used = time_msec(); |
2751 | if (execute_odp_actions(ofproto, &facet->flow, | |
2752 | facet->actions, facet->actions_len, packet)) { | |
2753 | facet_update_stats(ofproto, facet, &stats); | |
2754 | } | |
2755 | } | |
2756 | ||
2757 | /* Remove 'facet' from 'ofproto' and free up the associated memory: | |
2758 | * | |
2759 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2760 | * rule's statistics, via facet_uninstall(). | |
2761 | * | |
2762 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2763 | */ | |
2764 | static void | |
2765 | facet_remove(struct ofproto_dpif *ofproto, struct facet *facet) | |
2766 | { | |
2767 | facet_uninstall(ofproto, facet); | |
2768 | facet_flush_stats(ofproto, facet); | |
2769 | hmap_remove(&ofproto->facets, &facet->hmap_node); | |
2770 | list_remove(&facet->list_node); | |
2771 | facet_free(facet); | |
2772 | } | |
2773 | ||
df2c07f4 | 2774 | /* Composes the datapath actions for 'facet' based on its rule's actions. */ |
abe529af BP |
2775 | static void |
2776 | facet_make_actions(struct ofproto_dpif *p, struct facet *facet, | |
2777 | const struct ofpbuf *packet) | |
2778 | { | |
2779 | const struct rule_dpif *rule = facet->rule; | |
2780 | struct ofpbuf *odp_actions; | |
2781 | struct action_xlate_ctx ctx; | |
2782 | ||
2783 | action_xlate_ctx_init(&ctx, p, &facet->flow, packet); | |
2784 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
2785 | facet->tags = ctx.tags; | |
2786 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
2787 | facet->has_learn = ctx.has_learn; |
2788 | facet->has_normal = ctx.has_normal; | |
abe529af BP |
2789 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
2790 | ||
2791 | if (facet->actions_len != odp_actions->size | |
2792 | || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { | |
2793 | free(facet->actions); | |
2794 | facet->actions_len = odp_actions->size; | |
2795 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
2796 | } | |
2797 | ||
2798 | ofpbuf_delete(odp_actions); | |
2799 | } | |
2800 | ||
3a88e544 BP |
2801 | /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len' |
2802 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
2803 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
2804 | * since 'facet' was last updated. | |
2805 | * | |
2806 | * Returns 0 if successful, otherwise a positive errno value.*/ | |
abe529af BP |
2807 | static int |
2808 | facet_put__(struct ofproto_dpif *ofproto, struct facet *facet, | |
2809 | const struct nlattr *actions, size_t actions_len, | |
2810 | struct dpif_flow_stats *stats) | |
2811 | { | |
2812 | struct odputil_keybuf keybuf; | |
2813 | enum dpif_flow_put_flags flags; | |
2814 | struct ofpbuf key; | |
3a88e544 | 2815 | int ret; |
abe529af BP |
2816 | |
2817 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
2818 | if (stats) { | |
2819 | flags |= DPIF_FP_ZERO_STATS; | |
abe529af BP |
2820 | } |
2821 | ||
2822 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2823 | odp_flow_key_from_flow(&key, &facet->flow); | |
2824 | ||
3a88e544 BP |
2825 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, |
2826 | actions, actions_len, stats); | |
2827 | ||
2828 | if (stats) { | |
2829 | facet_reset_dp_stats(facet, stats); | |
2830 | } | |
2831 | ||
2832 | return ret; | |
abe529af BP |
2833 | } |
2834 | ||
2835 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If | |
2836 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2837 | * 'facet'. */ | |
2838 | static void | |
2839 | facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats) | |
2840 | { | |
2841 | struct dpif_flow_stats stats; | |
2842 | ||
2843 | if (facet->may_install | |
2844 | && !facet_put__(p, facet, facet->actions, facet->actions_len, | |
2845 | zero_stats ? &stats : NULL)) { | |
2846 | facet->installed = true; | |
2847 | } | |
2848 | } | |
2849 | ||
2850 | static void | |
55af77bb | 2851 | facet_account(struct ofproto_dpif *ofproto, struct facet *facet) |
abe529af | 2852 | { |
55af77bb | 2853 | uint64_t n_bytes; |
abe529af | 2854 | const struct nlattr *a; |
abe529af | 2855 | unsigned int left; |
d78be13b | 2856 | ovs_be16 vlan_tci; |
abe529af | 2857 | |
55af77bb | 2858 | if (facet->byte_count <= facet->accounted_bytes) { |
abe529af BP |
2859 | return; |
2860 | } | |
55af77bb EJ |
2861 | n_bytes = facet->byte_count - facet->accounted_bytes; |
2862 | facet->accounted_bytes = facet->byte_count; | |
abe529af | 2863 | |
75a75043 | 2864 | /* Feed information from the active flows back into the learning table to |
abe529af BP |
2865 | * ensure that table is always in sync with what is actually flowing |
2866 | * through the datapath. */ | |
75a75043 BP |
2867 | if (facet->has_learn || facet->has_normal) { |
2868 | struct action_xlate_ctx ctx; | |
abe529af | 2869 | |
75a75043 BP |
2870 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); |
2871 | ctx.may_learn = true; | |
2872 | ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions, | |
2873 | facet->rule->up.n_actions)); | |
2874 | } | |
abe529af | 2875 | |
75a75043 | 2876 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
2877 | return; |
2878 | } | |
d78be13b BP |
2879 | |
2880 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
2881 | * as a basis. We also need to track the actual VLAN on which the packet | |
2882 | * is going to be sent to ensure that it matches the one passed to | |
2883 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
2884 | * hash bucket.) */ | |
2885 | vlan_tci = facet->flow.vlan_tci; | |
abe529af | 2886 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) { |
d78be13b | 2887 | struct ofport_dpif *port; |
abe529af | 2888 | |
d78be13b | 2889 | switch (nl_attr_type(a)) { |
4edb9ae9 | 2890 | const struct nlattr *nested; |
df2c07f4 | 2891 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
2892 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
2893 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 2894 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 2895 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 2896 | } |
d78be13b BP |
2897 | break; |
2898 | ||
4edb9ae9 PS |
2899 | case OVS_ACTION_ATTR_POP: |
2900 | if (nl_attr_get_u16(a) == OVS_KEY_ATTR_8021Q) { | |
2901 | vlan_tci = htons(0); | |
2902 | } | |
d78be13b BP |
2903 | break; |
2904 | ||
4edb9ae9 PS |
2905 | case OVS_ACTION_ATTR_PUSH: |
2906 | nested = nl_attr_get(a); | |
2907 | if (nl_attr_type(nested) == OVS_KEY_ATTR_8021Q) { | |
2908 | const struct ovs_key_8021q *q_key; | |
2909 | ||
2910 | q_key = nl_attr_get_unspec(nested, sizeof(*q_key)); | |
2911 | vlan_tci = q_key->q_tci; | |
2912 | } | |
d78be13b | 2913 | break; |
abe529af BP |
2914 | } |
2915 | } | |
2916 | } | |
2917 | ||
2918 | /* If 'rule' is installed in the datapath, uninstalls it. */ | |
2919 | static void | |
2920 | facet_uninstall(struct ofproto_dpif *p, struct facet *facet) | |
2921 | { | |
2922 | if (facet->installed) { | |
2923 | struct odputil_keybuf keybuf; | |
2924 | struct dpif_flow_stats stats; | |
2925 | struct ofpbuf key; | |
3a88e544 | 2926 | int error; |
abe529af BP |
2927 | |
2928 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2929 | odp_flow_key_from_flow(&key, &facet->flow); | |
2930 | ||
3a88e544 BP |
2931 | error = dpif_flow_del(p->dpif, key.data, key.size, &stats); |
2932 | facet_reset_dp_stats(facet, &stats); | |
2933 | if (!error) { | |
abe529af BP |
2934 | facet_update_stats(p, facet, &stats); |
2935 | } | |
2936 | facet->installed = false; | |
abe529af BP |
2937 | } else { |
2938 | assert(facet->dp_packet_count == 0); | |
2939 | assert(facet->dp_byte_count == 0); | |
2940 | } | |
2941 | } | |
2942 | ||
2943 | /* Returns true if the only action for 'facet' is to send to the controller. | |
2944 | * (We don't report NetFlow expiration messages for such facets because they | |
2945 | * are just part of the control logic for the network, not real traffic). */ | |
2946 | static bool | |
2947 | facet_is_controller_flow(struct facet *facet) | |
2948 | { | |
2949 | return (facet | |
2950 | && facet->rule->up.n_actions == 1 | |
2951 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
2952 | htons(OFPP_CONTROLLER))); | |
2953 | } | |
2954 | ||
3a88e544 BP |
2955 | /* Resets 'facet''s datapath statistics counters. This should be called when |
2956 | * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null, | |
2957 | * it should contain the statistics returned by dpif when 'facet' was reset in | |
2958 | * the datapath. 'stats' will be modified to only included statistics new | |
2959 | * since 'facet' was last updated. */ | |
2960 | static void | |
2961 | facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats) | |
2962 | { | |
2963 | if (stats && facet->dp_packet_count <= stats->n_packets | |
2964 | && facet->dp_byte_count <= stats->n_bytes) { | |
2965 | stats->n_packets -= facet->dp_packet_count; | |
2966 | stats->n_bytes -= facet->dp_byte_count; | |
2967 | } | |
2968 | ||
2969 | facet->dp_packet_count = 0; | |
2970 | facet->dp_byte_count = 0; | |
2971 | } | |
2972 | ||
abe529af BP |
2973 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
2974 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
2975 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
2976 | * its packet and byte counts before this function is called. */ | |
2977 | static void | |
2978 | facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet) | |
2979 | { | |
2980 | assert(!facet->dp_byte_count); | |
2981 | assert(!facet->dp_packet_count); | |
2982 | ||
2983 | facet_push_stats(facet); | |
55af77bb | 2984 | facet_account(ofproto, facet); |
abe529af BP |
2985 | |
2986 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
2987 | struct ofexpired expired; | |
2988 | expired.flow = facet->flow; | |
2989 | expired.packet_count = facet->packet_count; | |
2990 | expired.byte_count = facet->byte_count; | |
2991 | expired.used = facet->used; | |
2992 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2993 | } | |
2994 | ||
2995 | facet->rule->packet_count += facet->packet_count; | |
2996 | facet->rule->byte_count += facet->byte_count; | |
2997 | ||
2998 | /* Reset counters to prevent double counting if 'facet' ever gets | |
2999 | * reinstalled. */ | |
bbb5d219 | 3000 | facet_reset_counters(facet); |
abe529af BP |
3001 | |
3002 | netflow_flow_clear(&facet->nf_flow); | |
3003 | } | |
3004 | ||
3005 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3006 | * Returns it if found, otherwise a null pointer. | |
3007 | * | |
3008 | * The returned facet might need revalidation; use facet_lookup_valid() | |
3009 | * instead if that is important. */ | |
3010 | static struct facet * | |
3011 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) | |
3012 | { | |
3013 | struct facet *facet; | |
3014 | ||
3015 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
3016 | &ofproto->facets) { | |
3017 | if (flow_equal(flow, &facet->flow)) { | |
3018 | return facet; | |
3019 | } | |
3020 | } | |
3021 | ||
3022 | return NULL; | |
3023 | } | |
3024 | ||
3025 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3026 | * Returns it if found, otherwise a null pointer. | |
3027 | * | |
3028 | * The returned facet is guaranteed to be valid. */ | |
3029 | static struct facet * | |
3030 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) | |
3031 | { | |
3032 | struct facet *facet = facet_find(ofproto, flow); | |
3033 | ||
3034 | /* The facet we found might not be valid, since we could be in need of | |
3035 | * revalidation. If it is not valid, don't return it. */ | |
3036 | if (facet | |
0e4b3771 BP |
3037 | && (ofproto->need_revalidate |
3038 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags)) | |
abe529af BP |
3039 | && !facet_revalidate(ofproto, facet)) { |
3040 | COVERAGE_INC(facet_invalidated); | |
3041 | return NULL; | |
3042 | } | |
3043 | ||
3044 | return facet; | |
3045 | } | |
3046 | ||
3047 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': | |
3048 | * | |
3049 | * - If the rule found is different from 'facet''s current rule, moves | |
3050 | * 'facet' to the new rule and recompiles its actions. | |
3051 | * | |
3052 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
3053 | * where it is and recompiles its actions anyway. | |
3054 | * | |
3055 | * - If there is none, destroys 'facet'. | |
3056 | * | |
3057 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
3058 | static bool | |
3059 | facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet) | |
3060 | { | |
3061 | struct action_xlate_ctx ctx; | |
3062 | struct ofpbuf *odp_actions; | |
3063 | struct rule_dpif *new_rule; | |
3064 | bool actions_changed; | |
3065 | ||
3066 | COVERAGE_INC(facet_revalidate); | |
3067 | ||
3068 | /* Determine the new rule. */ | |
29901626 | 3069 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); |
abe529af BP |
3070 | if (!new_rule) { |
3071 | /* No new rule, so delete the facet. */ | |
3072 | facet_remove(ofproto, facet); | |
3073 | return false; | |
3074 | } | |
3075 | ||
df2c07f4 | 3076 | /* Calculate new datapath actions. |
abe529af BP |
3077 | * |
3078 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
3079 | * emit a NetFlow expiration and, if so, we need to have the old state | |
3080 | * around to properly compose it. */ | |
3081 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); | |
3082 | odp_actions = xlate_actions(&ctx, | |
3083 | new_rule->up.actions, new_rule->up.n_actions); | |
3084 | actions_changed = (facet->actions_len != odp_actions->size | |
3085 | || memcmp(facet->actions, odp_actions->data, | |
3086 | facet->actions_len)); | |
3087 | ||
df2c07f4 JP |
3088 | /* If the datapath actions changed or the installability changed, |
3089 | * then we need to talk to the datapath. */ | |
abe529af BP |
3090 | if (actions_changed || ctx.may_set_up_flow != facet->installed) { |
3091 | if (ctx.may_set_up_flow) { | |
3092 | struct dpif_flow_stats stats; | |
3093 | ||
3094 | facet_put__(ofproto, facet, | |
3095 | odp_actions->data, odp_actions->size, &stats); | |
3096 | facet_update_stats(ofproto, facet, &stats); | |
3097 | } else { | |
3098 | facet_uninstall(ofproto, facet); | |
3099 | } | |
3100 | ||
3101 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
3102 | * 'facet' into 'rule'. */ | |
3103 | facet_flush_stats(ofproto, facet); | |
3104 | } | |
3105 | ||
3106 | /* Update 'facet' now that we've taken care of all the old state. */ | |
3107 | facet->tags = ctx.tags; | |
3108 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
3109 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
3110 | facet->has_learn = ctx.has_learn; |
3111 | facet->has_normal = ctx.has_normal; | |
abe529af BP |
3112 | if (actions_changed) { |
3113 | free(facet->actions); | |
3114 | facet->actions_len = odp_actions->size; | |
3115 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
3116 | } | |
3117 | if (facet->rule != new_rule) { | |
3118 | COVERAGE_INC(facet_changed_rule); | |
3119 | list_remove(&facet->list_node); | |
3120 | list_push_back(&new_rule->facets, &facet->list_node); | |
3121 | facet->rule = new_rule; | |
3122 | facet->used = new_rule->up.created; | |
3123 | facet->rs_used = facet->used; | |
3124 | } | |
3125 | ||
3126 | ofpbuf_delete(odp_actions); | |
3127 | ||
3128 | return true; | |
3129 | } | |
3130 | ||
3131 | /* Updates 'facet''s used time. Caller is responsible for calling | |
3132 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
3133 | static void | |
3134 | facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet, | |
3135 | long long int used) | |
3136 | { | |
3137 | if (used > facet->used) { | |
3138 | facet->used = used; | |
3139 | if (used > facet->rule->used) { | |
3140 | facet->rule->used = used; | |
3141 | } | |
3142 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
3143 | } | |
3144 | } | |
3145 | ||
3146 | /* Folds the statistics from 'stats' into the counters in 'facet'. | |
3147 | * | |
3148 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
3149 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
3150 | * packet that was sent by hand or if it represents statistics that have been | |
3151 | * cleared out of the datapath. */ | |
3152 | static void | |
3153 | facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet, | |
3154 | const struct dpif_flow_stats *stats) | |
3155 | { | |
3156 | if (stats->n_packets || stats->used > facet->used) { | |
3157 | facet_update_time(ofproto, facet, stats->used); | |
3158 | facet->packet_count += stats->n_packets; | |
3159 | facet->byte_count += stats->n_bytes; | |
3160 | facet_push_stats(facet); | |
3161 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
3162 | } | |
3163 | } | |
3164 | ||
bbb5d219 EJ |
3165 | static void |
3166 | facet_reset_counters(struct facet *facet) | |
3167 | { | |
3168 | facet->packet_count = 0; | |
3169 | facet->byte_count = 0; | |
3170 | facet->rs_packet_count = 0; | |
3171 | facet->rs_byte_count = 0; | |
3172 | facet->accounted_bytes = 0; | |
3173 | } | |
3174 | ||
abe529af BP |
3175 | static void |
3176 | facet_push_stats(struct facet *facet) | |
3177 | { | |
3178 | uint64_t rs_packets, rs_bytes; | |
3179 | ||
3180 | assert(facet->packet_count >= facet->rs_packet_count); | |
3181 | assert(facet->byte_count >= facet->rs_byte_count); | |
3182 | assert(facet->used >= facet->rs_used); | |
3183 | ||
3184 | rs_packets = facet->packet_count - facet->rs_packet_count; | |
3185 | rs_bytes = facet->byte_count - facet->rs_byte_count; | |
3186 | ||
3187 | if (rs_packets || rs_bytes || facet->used > facet->rs_used) { | |
3188 | facet->rs_packet_count = facet->packet_count; | |
3189 | facet->rs_byte_count = facet->byte_count; | |
3190 | facet->rs_used = facet->used; | |
3191 | ||
3192 | flow_push_stats(facet->rule, &facet->flow, | |
3193 | rs_packets, rs_bytes, facet->used); | |
3194 | } | |
3195 | } | |
3196 | ||
3197 | struct ofproto_push { | |
3198 | struct action_xlate_ctx ctx; | |
3199 | uint64_t packets; | |
3200 | uint64_t bytes; | |
3201 | long long int used; | |
3202 | }; | |
3203 | ||
3204 | static void | |
3205 | push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
3206 | { | |
3207 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
3208 | ||
3209 | if (rule) { | |
3210 | rule->packet_count += push->packets; | |
3211 | rule->byte_count += push->bytes; | |
3212 | rule->used = MAX(push->used, rule->used); | |
3213 | } | |
3214 | } | |
3215 | ||
3216 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
3217 | * 'rule''s actions. */ | |
3218 | static void | |
3219 | flow_push_stats(const struct rule_dpif *rule, | |
3220 | struct flow *flow, uint64_t packets, uint64_t bytes, | |
3221 | long long int used) | |
3222 | { | |
3223 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3224 | struct ofproto_push push; | |
3225 | ||
3226 | push.packets = packets; | |
3227 | push.bytes = bytes; | |
3228 | push.used = used; | |
3229 | ||
3230 | action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); | |
3231 | push.ctx.resubmit_hook = push_resubmit; | |
3232 | ofpbuf_delete(xlate_actions(&push.ctx, | |
3233 | rule->up.actions, rule->up.n_actions)); | |
3234 | } | |
3235 | \f | |
3236 | /* Rules. */ | |
3237 | ||
3238 | static struct rule_dpif * | |
29901626 BP |
3239 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
3240 | uint8_t table_id) | |
abe529af | 3241 | { |
7257b535 BP |
3242 | struct cls_rule *cls_rule; |
3243 | struct classifier *cls; | |
3244 | ||
9cdaaebe BP |
3245 | if (table_id >= N_TABLES) { |
3246 | return NULL; | |
3247 | } | |
3248 | ||
7257b535 BP |
3249 | cls = &ofproto->up.tables[table_id]; |
3250 | if (flow->tos_frag & FLOW_FRAG_ANY | |
3251 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { | |
3252 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
3253 | * are unavailable. */ | |
3254 | struct flow ofpc_normal_flow = *flow; | |
3255 | ofpc_normal_flow.tp_src = htons(0); | |
3256 | ofpc_normal_flow.tp_dst = htons(0); | |
3257 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
3258 | } else { | |
3259 | cls_rule = classifier_lookup(cls, flow); | |
3260 | } | |
3261 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
3262 | } |
3263 | ||
7ee20df1 BP |
3264 | static void |
3265 | complete_operation(struct rule_dpif *rule) | |
3266 | { | |
3267 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3268 | ||
54a9cbc9 | 3269 | rule_invalidate(rule); |
7ee20df1 BP |
3270 | if (clogged) { |
3271 | struct dpif_completion *c = xmalloc(sizeof *c); | |
3272 | c->op = rule->up.pending; | |
3273 | list_push_back(&ofproto->completions, &c->list_node); | |
3274 | } else { | |
3275 | ofoperation_complete(rule->up.pending, 0); | |
3276 | } | |
3277 | } | |
3278 | ||
abe529af BP |
3279 | static struct rule * |
3280 | rule_alloc(void) | |
3281 | { | |
3282 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
3283 | return &rule->up; | |
3284 | } | |
3285 | ||
3286 | static void | |
3287 | rule_dealloc(struct rule *rule_) | |
3288 | { | |
3289 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3290 | free(rule); | |
3291 | } | |
3292 | ||
3293 | static int | |
3294 | rule_construct(struct rule *rule_) | |
3295 | { | |
3296 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3297 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 3298 | struct rule_dpif *victim; |
54a9cbc9 | 3299 | uint8_t table_id; |
5bf0e941 BP |
3300 | int error; |
3301 | ||
3302 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
3303 | &rule->up.cr.flow, ofproto->max_ports); | |
3304 | if (error) { | |
3305 | return error; | |
3306 | } | |
abe529af BP |
3307 | |
3308 | rule->used = rule->up.created; | |
3309 | rule->packet_count = 0; | |
3310 | rule->byte_count = 0; | |
abe529af | 3311 | |
7ee20df1 BP |
3312 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
3313 | if (victim && !list_is_empty(&victim->facets)) { | |
3314 | struct facet *facet; | |
3315 | ||
3316 | rule->facets = victim->facets; | |
3317 | list_moved(&rule->facets); | |
3318 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
3319 | /* XXX: We're only clearing our local counters here. It's possible |
3320 | * that quite a few packets are unaccounted for in the datapath | |
3321 | * statistics. These will be accounted to the new rule instead of | |
3322 | * cleared as required. This could be fixed by clearing out the | |
3323 | * datapath statistics for this facet, but currently it doesn't | |
3324 | * seem worth it. */ | |
3325 | facet_reset_counters(facet); | |
7ee20df1 BP |
3326 | facet->rule = rule; |
3327 | } | |
3328 | } else { | |
3329 | /* Must avoid list_moved() in this case. */ | |
3330 | list_init(&rule->facets); | |
3331 | } | |
abe529af | 3332 | |
54a9cbc9 BP |
3333 | table_id = rule->up.table_id; |
3334 | rule->tag = (victim ? victim->tag | |
3335 | : table_id == 0 ? 0 | |
3336 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
3337 | ofproto->tables[table_id].basis)); | |
3338 | ||
7ee20df1 | 3339 | complete_operation(rule); |
abe529af BP |
3340 | return 0; |
3341 | } | |
3342 | ||
3343 | static void | |
3344 | rule_destruct(struct rule *rule_) | |
3345 | { | |
3346 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3347 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3348 | struct facet *facet, *next_facet; | |
3349 | ||
abe529af BP |
3350 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
3351 | facet_revalidate(ofproto, facet); | |
3352 | } | |
7ee20df1 BP |
3353 | |
3354 | complete_operation(rule); | |
abe529af BP |
3355 | } |
3356 | ||
3357 | static void | |
3358 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
3359 | { | |
3360 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3361 | struct facet *facet; | |
3362 | ||
3363 | /* Start from historical data for 'rule' itself that are no longer tracked | |
3364 | * in facets. This counts, for example, facets that have expired. */ | |
3365 | *packets = rule->packet_count; | |
3366 | *bytes = rule->byte_count; | |
3367 | ||
3368 | /* Add any statistics that are tracked by facets. This includes | |
3369 | * statistical data recently updated by ofproto_update_stats() as well as | |
3370 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
3371 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
3372 | *packets += facet->packet_count; | |
3373 | *bytes += facet->byte_count; | |
3374 | } | |
3375 | } | |
3376 | ||
5bf0e941 | 3377 | static int |
abe529af BP |
3378 | rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet) |
3379 | { | |
3380 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3381 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3382 | struct action_xlate_ctx ctx; | |
3383 | struct ofpbuf *odp_actions; | |
3384 | struct facet *facet; | |
3385 | size_t size; | |
3386 | ||
3387 | /* First look for a related facet. If we find one, account it to that. */ | |
3388 | facet = facet_lookup_valid(ofproto, flow); | |
3389 | if (facet && facet->rule == rule) { | |
eea109bb BP |
3390 | if (!facet->may_install) { |
3391 | facet_make_actions(ofproto, facet, packet); | |
3392 | } | |
abe529af | 3393 | facet_execute(ofproto, facet, packet); |
5bf0e941 | 3394 | return 0; |
abe529af BP |
3395 | } |
3396 | ||
3397 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then | |
3398 | * create a new facet for it and use that. */ | |
29901626 | 3399 | if (rule_dpif_lookup(ofproto, flow, 0) == rule) { |
f3827897 BP |
3400 | facet = facet_create(rule, flow); |
3401 | facet_make_actions(ofproto, facet, packet); | |
abe529af BP |
3402 | facet_execute(ofproto, facet, packet); |
3403 | facet_install(ofproto, facet, true); | |
5bf0e941 | 3404 | return 0; |
abe529af BP |
3405 | } |
3406 | ||
3407 | /* We can't account anything to a facet. If we were to try, then that | |
3408 | * facet would have a non-matching rule, busting our invariants. */ | |
3409 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
3410 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
3411 | size = packet->size; | |
3412 | if (execute_odp_actions(ofproto, flow, odp_actions->data, | |
3413 | odp_actions->size, packet)) { | |
3414 | rule->used = time_msec(); | |
3415 | rule->packet_count++; | |
3416 | rule->byte_count += size; | |
3417 | flow_push_stats(rule, flow, 1, size, rule->used); | |
3418 | } | |
3419 | ofpbuf_delete(odp_actions); | |
5bf0e941 BP |
3420 | |
3421 | return 0; | |
abe529af BP |
3422 | } |
3423 | ||
7ee20df1 BP |
3424 | static void |
3425 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
3426 | { |
3427 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3428 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3429 | int error; | |
3430 | ||
7ee20df1 BP |
3431 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
3432 | &rule->up.cr.flow, ofproto->max_ports); | |
3433 | if (error) { | |
3434 | ofoperation_complete(rule->up.pending, error); | |
3435 | return; | |
abe529af | 3436 | } |
7ee20df1 BP |
3437 | |
3438 | complete_operation(rule); | |
abe529af BP |
3439 | } |
3440 | \f | |
b47e2a82 | 3441 | /* Sends 'packet' out of port 'odp_port' within 'ofproto'. |
abe529af BP |
3442 | * Returns 0 if successful, otherwise a positive errno value. */ |
3443 | static int | |
b2fda3ef | 3444 | send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, |
abe529af BP |
3445 | const struct ofpbuf *packet) |
3446 | { | |
80e5eed9 BP |
3447 | struct ofpbuf key, odp_actions; |
3448 | struct odputil_keybuf keybuf; | |
3449 | struct flow flow; | |
abe529af BP |
3450 | int error; |
3451 | ||
abff858b | 3452 | flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow); |
80e5eed9 BP |
3453 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
3454 | odp_flow_key_from_flow(&key, &flow); | |
3455 | ||
abe529af | 3456 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
3457 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
3458 | ||
df2c07f4 | 3459 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
3460 | error = dpif_execute(ofproto->dpif, |
3461 | key.data, key.size, | |
3462 | odp_actions.data, odp_actions.size, | |
abe529af BP |
3463 | packet); |
3464 | ofpbuf_uninit(&odp_actions); | |
3465 | ||
3466 | if (error) { | |
3467 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
3468 | ofproto->up.name, odp_port, strerror(error)); | |
3469 | } | |
3470 | return error; | |
3471 | } | |
3472 | \f | |
df2c07f4 | 3473 | /* OpenFlow to datapath action translation. */ |
abe529af BP |
3474 | |
3475 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
3476 | struct action_xlate_ctx *ctx); | |
4cd78906 | 3477 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 3478 | |
98403001 BP |
3479 | static size_t |
3480 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
3481 | struct ofpbuf *odp_actions, | |
3482 | const struct flow *flow, | |
3483 | const struct user_action_cookie *cookie) | |
3484 | { | |
3485 | size_t offset; | |
3486 | uint32_t pid; | |
3487 | ||
3488 | pid = dpif_port_get_pid(ofproto->dpif, | |
3489 | ofp_port_to_odp_port(flow->in_port)); | |
3490 | ||
3491 | offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_USERSPACE); | |
3492 | nl_msg_put_u32(odp_actions, OVS_USERSPACE_ATTR_PID, pid); | |
3493 | nl_msg_put_unspec(odp_actions, OVS_USERSPACE_ATTR_USERDATA, | |
3494 | cookie, sizeof *cookie); | |
3495 | nl_msg_end_nested(odp_actions, offset); | |
3496 | ||
3497 | return odp_actions->size - NLA_ALIGN(sizeof *cookie); | |
3498 | } | |
3499 | ||
6ff686f2 PS |
3500 | /* Compose SAMPLE action for sFlow. */ |
3501 | static size_t | |
3502 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
3503 | struct ofpbuf *odp_actions, | |
3504 | const struct flow *flow, | |
3505 | uint32_t odp_port) | |
3506 | { | |
3507 | uint32_t port_ifindex; | |
3508 | uint32_t probability; | |
98403001 | 3509 | struct user_action_cookie cookie; |
6ff686f2 | 3510 | size_t sample_offset, actions_offset; |
98403001 | 3511 | int cookie_offset, n_output; |
6ff686f2 PS |
3512 | |
3513 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
3514 | return 0; | |
3515 | } | |
3516 | ||
3517 | if (odp_port == OVSP_NONE) { | |
3518 | port_ifindex = 0; | |
3519 | n_output = 0; | |
3520 | } else { | |
3521 | port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
3522 | n_output = 1; | |
3523 | } | |
3524 | ||
3525 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
3526 | ||
3527 | /* Number of packets out of UINT_MAX to sample. */ | |
3528 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
3529 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
3530 | ||
3531 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
3532 | ||
98403001 BP |
3533 | cookie.type = USER_ACTION_COOKIE_SFLOW; |
3534 | cookie.data = port_ifindex; | |
3535 | cookie.n_output = n_output; | |
3536 | cookie.vlan_tci = 0; | |
3537 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); | |
6ff686f2 PS |
3538 | |
3539 | nl_msg_end_nested(odp_actions, actions_offset); | |
3540 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 3541 | return cookie_offset; |
6ff686f2 PS |
3542 | } |
3543 | ||
3544 | /* SAMPLE action must be first action in any given list of actions. | |
3545 | * At this point we do not have all information required to build it. So try to | |
3546 | * build sample action as complete as possible. */ | |
3547 | static void | |
3548 | add_sflow_action(struct action_xlate_ctx *ctx) | |
3549 | { | |
3550 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
3551 | ctx->odp_actions, | |
3552 | &ctx->flow, OVSP_NONE); | |
3553 | ctx->sflow_odp_port = 0; | |
3554 | ctx->sflow_n_outputs = 0; | |
3555 | } | |
3556 | ||
3557 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
3558 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
3559 | * USERSPACE action's user-cookie which is required for sflow. */ | |
3560 | static void | |
3561 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
3562 | { | |
3563 | const struct flow *base = &ctx->base_flow; | |
3564 | struct user_action_cookie *cookie; | |
3565 | ||
3566 | if (!ctx->user_cookie_offset) { | |
3567 | return; | |
3568 | } | |
3569 | ||
3570 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
3571 | sizeof(*cookie)); | |
3572 | assert(cookie != NULL); | |
3573 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
3574 | ||
3575 | if (ctx->sflow_n_outputs) { | |
3576 | cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow, | |
3577 | ctx->sflow_odp_port); | |
3578 | } | |
3579 | if (ctx->sflow_n_outputs >= 255) { | |
3580 | cookie->n_output = 255; | |
3581 | } else { | |
3582 | cookie->n_output = ctx->sflow_n_outputs; | |
3583 | } | |
3584 | cookie->vlan_tci = base->vlan_tci; | |
3585 | } | |
3586 | ||
823518f1 | 3587 | static void |
4edb9ae9 PS |
3588 | commit_action__(struct ofpbuf *odp_actions, |
3589 | enum ovs_action_attr act_type, | |
3590 | enum ovs_key_attr key_type, | |
3591 | const void *key, size_t key_size) | |
823518f1 | 3592 | { |
4edb9ae9 | 3593 | size_t offset = nl_msg_start_nested(odp_actions, act_type); |
823518f1 | 3594 | |
4edb9ae9 PS |
3595 | nl_msg_put_unspec(odp_actions, key_type, key, key_size); |
3596 | nl_msg_end_nested(odp_actions, offset); | |
3597 | } | |
3598 | ||
3599 | static void | |
3600 | commit_set_tun_id_action(const struct flow *flow, struct flow *base, | |
3601 | struct ofpbuf *odp_actions) | |
3602 | { | |
3603 | if (base->tun_id == flow->tun_id) { | |
3604 | return; | |
823518f1 | 3605 | } |
4edb9ae9 PS |
3606 | base->tun_id = flow->tun_id; |
3607 | ||
3608 | commit_action__(odp_actions, OVS_ACTION_ATTR_SET, | |
3609 | OVS_KEY_ATTR_TUN_ID, &base->tun_id, sizeof(base->tun_id)); | |
823518f1 BP |
3610 | } |
3611 | ||
b3e9b2ed | 3612 | static void |
4edb9ae9 PS |
3613 | commit_set_ether_addr_action(const struct flow *flow, struct flow *base, |
3614 | struct ofpbuf *odp_actions) | |
3615 | { | |
3616 | struct ovs_key_ethernet eth_key; | |
3617 | ||
3618 | if (eth_addr_equals(base->dl_src, flow->dl_src) && | |
3619 | eth_addr_equals(base->dl_dst, flow->dl_dst)) { | |
3620 | return; | |
3621 | } | |
3622 | ||
3623 | memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN); | |
3624 | memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN); | |
3625 | ||
3626 | memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN); | |
3627 | memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN); | |
3628 | ||
3629 | commit_action__(odp_actions, OVS_ACTION_ATTR_SET, | |
3630 | OVS_KEY_ATTR_ETHERNET, ð_key, sizeof(eth_key)); | |
3631 | } | |
3632 | ||
3633 | static void | |
3634 | commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci) | |
b3e9b2ed | 3635 | { |
b3e9b2ed | 3636 | struct flow *base = &ctx->base_flow; |
b3e9b2ed | 3637 | |
4edb9ae9 PS |
3638 | if (base->vlan_tci == new_tci) { |
3639 | return; | |
b3e9b2ed EJ |
3640 | } |
3641 | ||
4edb9ae9 PS |
3642 | if (base->vlan_tci & htons(VLAN_CFI)) { |
3643 | nl_msg_put_u16(ctx->odp_actions, OVS_ACTION_ATTR_POP, | |
3644 | OVS_KEY_ATTR_8021Q); | |
b3e9b2ed EJ |
3645 | } |
3646 | ||
4edb9ae9 PS |
3647 | if (new_tci & htons(VLAN_CFI)) { |
3648 | struct ovs_key_8021q q_key; | |
3649 | ||
3650 | q_key.q_tpid = htons(ETH_TYPE_VLAN); | |
3651 | q_key.q_tci = new_tci & ~htons(VLAN_CFI); | |
3652 | ||
3653 | commit_action__(ctx->odp_actions, OVS_ACTION_ATTR_PUSH, | |
3654 | OVS_KEY_ATTR_8021Q, &q_key, sizeof(q_key)); | |
b3e9b2ed | 3655 | } |
4edb9ae9 PS |
3656 | base->vlan_tci = new_tci; |
3657 | } | |
b3e9b2ed | 3658 | |
4edb9ae9 PS |
3659 | static void |
3660 | commit_set_nw_action(const struct flow *flow, struct flow *base, | |
3661 | struct ofpbuf *odp_actions) | |
3662 | { | |
7257b535 | 3663 | int frag = base->tos_frag & FLOW_FRAG_MASK; |
4edb9ae9 PS |
3664 | struct ovs_key_ipv4 ipv4_key; |
3665 | ||
3666 | if (base->dl_type != htons(ETH_TYPE_IP) || | |
3667 | !base->nw_src || !base->nw_dst) { | |
3668 | return; | |
3669 | } | |
3670 | ||
3671 | if (base->nw_src == flow->nw_src && | |
3672 | base->nw_dst == flow->nw_dst && | |
7257b535 | 3673 | base->tos_frag == flow->tos_frag) { |
4edb9ae9 | 3674 | return; |
150a9f15 BP |
3675 | } |
3676 | ||
7257b535 | 3677 | |
4edb9ae9 PS |
3678 | memset(&ipv4_key, 0, sizeof(ipv4_key)); |
3679 | ipv4_key.ipv4_src = base->nw_src = flow->nw_src; | |
3680 | ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst; | |
4edb9ae9 | 3681 | ipv4_key.ipv4_proto = base->nw_proto; |
7257b535 BP |
3682 | ipv4_key.ipv4_tos = flow->tos_frag & IP_DSCP_MASK; |
3683 | ipv4_key.ipv4_frag = (frag == 0 ? OVS_FRAG_TYPE_NONE | |
3684 | : frag == FLOW_FRAG_ANY ? OVS_FRAG_TYPE_FIRST | |
3685 | : OVS_FRAG_TYPE_LATER); | |
4edb9ae9 PS |
3686 | |
3687 | commit_action__(odp_actions, OVS_ACTION_ATTR_SET, | |
3688 | OVS_KEY_ATTR_IPV4, &ipv4_key, sizeof(ipv4_key)); | |
3689 | } | |
3690 | ||
3691 | static void | |
3692 | commit_set_port_action(const struct flow *flow, struct flow *base, | |
3693 | struct ofpbuf *odp_actions) | |
3694 | { | |
3695 | if (!base->tp_src || !base->tp_dst) { | |
3696 | return; | |
b3e9b2ed EJ |
3697 | } |
3698 | ||
4edb9ae9 PS |
3699 | if (base->tp_src == flow->tp_src && |
3700 | base->tp_dst == flow->tp_dst) { | |
3701 | return; | |
b3e9b2ed EJ |
3702 | } |
3703 | ||
4edb9ae9 PS |
3704 | if (flow->nw_proto == IPPROTO_TCP) { |
3705 | struct ovs_key_tcp port_key; | |
3706 | ||
3707 | port_key.tcp_src = base->tp_src = flow->tp_src; | |
3708 | port_key.tcp_dst = base->tp_dst = flow->tp_dst; | |
3709 | ||
3710 | commit_action__(odp_actions, OVS_ACTION_ATTR_SET, | |
3711 | OVS_KEY_ATTR_TCP, &port_key, sizeof(port_key)); | |
3712 | ||
3713 | } else if (flow->nw_proto == IPPROTO_UDP) { | |
3714 | struct ovs_key_udp port_key; | |
3715 | ||
3716 | port_key.udp_src = base->tp_src = flow->tp_src; | |
3717 | port_key.udp_dst = base->tp_dst = flow->tp_dst; | |
3718 | ||
3719 | commit_action__(odp_actions, OVS_ACTION_ATTR_SET, | |
3720 | OVS_KEY_ATTR_UDP, &port_key, sizeof(port_key)); | |
b3e9b2ed | 3721 | } |
4edb9ae9 | 3722 | } |
b3e9b2ed | 3723 | |
4edb9ae9 | 3724 | static void |
abff858b PS |
3725 | commit_set_priority_action(const struct flow *flow, struct flow *base, |
3726 | struct ofpbuf *odp_actions) | |
4edb9ae9 | 3727 | { |
abff858b | 3728 | if (base->priority == flow->priority) { |
4edb9ae9 | 3729 | return; |
b3e9b2ed | 3730 | } |
abff858b | 3731 | base->priority = flow->priority; |
b3e9b2ed | 3732 | |
abff858b PS |
3733 | commit_action__(odp_actions, OVS_ACTION_ATTR_SET, |
3734 | OVS_KEY_ATTR_PRIORITY, &base->priority, | |
3735 | sizeof(base->priority)); | |
4edb9ae9 PS |
3736 | } |
3737 | ||
3738 | static void | |
3739 | commit_odp_actions(struct action_xlate_ctx *ctx) | |
3740 | { | |
3741 | const struct flow *flow = &ctx->flow; | |
3742 | struct flow *base = &ctx->base_flow; | |
3743 | struct ofpbuf *odp_actions = ctx->odp_actions; | |
3744 | ||
3745 | commit_set_tun_id_action(flow, base, odp_actions); | |
3746 | commit_set_ether_addr_action(flow, base, odp_actions); | |
3747 | commit_vlan_action(ctx, flow->vlan_tci); | |
3748 | commit_set_nw_action(flow, base, odp_actions); | |
3749 | commit_set_port_action(flow, base, odp_actions); | |
abff858b | 3750 | commit_set_priority_action(flow, base, odp_actions); |
b3e9b2ed EJ |
3751 | } |
3752 | ||
6ff686f2 PS |
3753 | static void |
3754 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port) | |
3755 | { | |
3756 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); | |
3757 | ctx->sflow_odp_port = odp_port; | |
3758 | ctx->sflow_n_outputs++; | |
3759 | } | |
3760 | ||
abe529af BP |
3761 | static void |
3762 | add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) | |
3763 | { | |
3764 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); | |
3765 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); | |
3766 | ||
3767 | if (ofport) { | |
21f7563c JP |
3768 | if (ofport->up.opp.config & htonl(OFPPC_NO_FWD) |
3769 | || !stp_forward_in_state(ofport->stp_state)) { | |
abe529af BP |
3770 | /* Forwarding disabled on port. */ |
3771 | return; | |
3772 | } | |
3773 | } else { | |
3774 | /* | |
3775 | * We don't have an ofport record for this port, but it doesn't hurt to | |
3776 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
3777 | * and we're pre-populating the flow table. | |
3778 | */ | |
3779 | } | |
3780 | ||
b3e9b2ed | 3781 | commit_odp_actions(ctx); |
6ff686f2 | 3782 | compose_output_action(ctx, odp_port); |
abe529af BP |
3783 | ctx->nf_output_iface = ofp_port; |
3784 | } | |
3785 | ||
3786 | static void | |
29901626 BP |
3787 | xlate_table_action(struct action_xlate_ctx *ctx, |
3788 | uint16_t in_port, uint8_t table_id) | |
abe529af BP |
3789 | { |
3790 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 3791 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
3792 | struct rule_dpif *rule; |
3793 | uint16_t old_in_port; | |
29901626 BP |
3794 | uint8_t old_table_id; |
3795 | ||
3796 | old_table_id = ctx->table_id; | |
3797 | ctx->table_id = table_id; | |
abe529af | 3798 | |
54a9cbc9 | 3799 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
3800 | old_in_port = ctx->flow.in_port; |
3801 | ctx->flow.in_port = in_port; | |
54a9cbc9 BP |
3802 | rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id); |
3803 | ||
3804 | /* Tag the flow. */ | |
3805 | if (table_id > 0 && table_id < N_TABLES) { | |
3806 | struct table_dpif *table = &ofproto->tables[table_id]; | |
3807 | if (table->other_table) { | |
3808 | ctx->tags |= (rule | |
3809 | ? rule->tag | |
3810 | : rule_calculate_tag(&ctx->flow, | |
3811 | &table->other_table->wc, | |
3812 | table->basis)); | |
3813 | } | |
3814 | } | |
3815 | ||
3816 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
3817 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
3818 | ctx->flow.in_port = old_in_port; |
3819 | ||
3820 | if (ctx->resubmit_hook) { | |
3821 | ctx->resubmit_hook(ctx, rule); | |
3822 | } | |
3823 | ||
3824 | if (rule) { | |
3825 | ctx->recurse++; | |
3826 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); | |
3827 | ctx->recurse--; | |
3828 | } | |
29901626 BP |
3829 | |
3830 | ctx->table_id = old_table_id; | |
abe529af BP |
3831 | } else { |
3832 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
3833 | ||
29901626 | 3834 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af BP |
3835 | MAX_RESUBMIT_RECURSION); |
3836 | } | |
3837 | } | |
3838 | ||
29901626 BP |
3839 | static void |
3840 | xlate_resubmit_table(struct action_xlate_ctx *ctx, | |
3841 | const struct nx_action_resubmit *nar) | |
3842 | { | |
3843 | uint16_t in_port; | |
3844 | uint8_t table_id; | |
3845 | ||
3846 | in_port = (nar->in_port == htons(OFPP_IN_PORT) | |
3847 | ? ctx->flow.in_port | |
3848 | : ntohs(nar->in_port)); | |
3849 | table_id = nar->table == 255 ? ctx->table_id : nar->table; | |
3850 | ||
3851 | xlate_table_action(ctx, in_port, table_id); | |
3852 | } | |
3853 | ||
abe529af | 3854 | static void |
b3e9b2ed | 3855 | flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask) |
abe529af BP |
3856 | { |
3857 | struct ofport_dpif *ofport; | |
3858 | ||
b3e9b2ed EJ |
3859 | commit_odp_actions(ctx); |
3860 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { | |
abe529af | 3861 | uint16_t ofp_port = ofport->up.ofp_port; |
21f7563c JP |
3862 | if (ofp_port != ctx->flow.in_port |
3863 | && !(ofport->up.opp.config & mask) | |
3864 | && stp_forward_in_state(ofport->stp_state)) { | |
6ff686f2 | 3865 | compose_output_action(ctx, ofport->odp_port); |
abe529af BP |
3866 | } |
3867 | } | |
b3e9b2ed EJ |
3868 | |
3869 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
3870 | } |
3871 | ||
6ff686f2 | 3872 | static void |
98403001 | 3873 | compose_controller_action(struct action_xlate_ctx *ctx, int len) |
6ff686f2 PS |
3874 | { |
3875 | struct user_action_cookie cookie; | |
3876 | ||
3877 | cookie.type = USER_ACTION_COOKIE_CONTROLLER; | |
3878 | cookie.data = len; | |
3879 | cookie.n_output = 0; | |
3880 | cookie.vlan_tci = 0; | |
98403001 | 3881 | put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie); |
6ff686f2 PS |
3882 | } |
3883 | ||
abe529af BP |
3884 | static void |
3885 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
3886 | uint16_t port, uint16_t max_len) | |
3887 | { | |
3888 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
3889 | ||
3890 | ctx->nf_output_iface = NF_OUT_DROP; | |
3891 | ||
3892 | switch (port) { | |
3893 | case OFPP_IN_PORT: | |
3894 | add_output_action(ctx, ctx->flow.in_port); | |
3895 | break; | |
3896 | case OFPP_TABLE: | |
29901626 | 3897 | xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); |
abe529af BP |
3898 | break; |
3899 | case OFPP_NORMAL: | |
3900 | xlate_normal(ctx); | |
3901 | break; | |
3902 | case OFPP_FLOOD: | |
b3e9b2ed | 3903 | flood_packets(ctx, htonl(OFPPC_NO_FLOOD)); |
abe529af BP |
3904 | break; |
3905 | case OFPP_ALL: | |
b3e9b2ed | 3906 | flood_packets(ctx, htonl(0)); |
abe529af BP |
3907 | break; |
3908 | case OFPP_CONTROLLER: | |
b3e9b2ed | 3909 | commit_odp_actions(ctx); |
98403001 | 3910 | compose_controller_action(ctx, max_len); |
abe529af BP |
3911 | break; |
3912 | case OFPP_LOCAL: | |
3913 | add_output_action(ctx, OFPP_LOCAL); | |
3914 | break; | |
e81d2933 EJ |
3915 | case OFPP_NONE: |
3916 | break; | |
abe529af BP |
3917 | default: |
3918 | if (port != ctx->flow.in_port) { | |
3919 | add_output_action(ctx, port); | |
3920 | } | |
3921 | break; | |
3922 | } | |
3923 | ||
3924 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
3925 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
3926 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
3927 | ctx->nf_output_iface = prev_nf_output_iface; | |
3928 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
3929 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
3930 | ctx->nf_output_iface = NF_OUT_MULTI; | |
3931 | } | |
3932 | } | |
3933 | ||
f694937d EJ |
3934 | static void |
3935 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
3936 | const struct nx_action_output_reg *naor) | |
3937 | { | |
3938 | uint64_t ofp_port; | |
3939 | ||
3940 | ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow); | |
3941 | ||
3942 | if (ofp_port <= UINT16_MAX) { | |
3943 | xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); | |
3944 | } | |
3945 | } | |
3946 | ||
abe529af BP |
3947 | static void |
3948 | xlate_output_action(struct action_xlate_ctx *ctx, | |
3949 | const struct ofp_action_output *oao) | |
3950 | { | |
3951 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
3952 | } | |
3953 | ||
abe529af BP |
3954 | static void |
3955 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
3956 | const struct ofp_action_enqueue *oae) | |
3957 | { | |
3958 | uint16_t ofp_port, odp_port; | |
abff858b | 3959 | uint32_t flow_priority, priority; |
abe529af BP |
3960 | int error; |
3961 | ||
3962 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
3963 | &priority); | |
3964 | if (error) { | |
3965 | /* Fall back to ordinary output action. */ | |
3966 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
3967 | return; | |
3968 | } | |
3969 | ||
df2c07f4 | 3970 | /* Figure out datapath output port. */ |
abe529af BP |
3971 | ofp_port = ntohs(oae->port); |
3972 | if (ofp_port == OFPP_IN_PORT) { | |
3973 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
3974 | } else if (ofp_port == ctx->flow.in_port) { |
3975 | return; | |
abe529af BP |
3976 | } |
3977 | odp_port = ofp_port_to_odp_port(ofp_port); | |
3978 | ||
df2c07f4 | 3979 | /* Add datapath actions. */ |
abff858b PS |
3980 | flow_priority = ctx->flow.priority; |
3981 | ctx->flow.priority = priority; | |
abe529af | 3982 | add_output_action(ctx, odp_port); |
abff858b | 3983 | ctx->flow.priority = flow_priority; |
abe529af BP |
3984 | |
3985 | /* Update NetFlow output port. */ | |
3986 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
3987 | ctx->nf_output_iface = odp_port; | |
3988 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
3989 | ctx->nf_output_iface = NF_OUT_MULTI; | |
3990 | } | |
3991 | } | |
3992 | ||
3993 | static void | |
3994 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
3995 | const struct nx_action_set_queue *nasq) | |
3996 | { | |
3997 | uint32_t priority; | |
3998 | int error; | |
3999 | ||
4000 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
4001 | &priority); | |
4002 | if (error) { | |
4003 | /* Couldn't translate queue to a priority, so ignore. A warning | |
4004 | * has already been logged. */ | |
4005 | return; | |
4006 | } | |
4007 | ||
abff858b | 4008 | ctx->flow.priority = priority; |
abe529af BP |
4009 | } |
4010 | ||
4011 | struct xlate_reg_state { | |
4012 | ovs_be16 vlan_tci; | |
4013 | ovs_be64 tun_id; | |
4014 | }; | |
4015 | ||
abe529af BP |
4016 | static void |
4017 | xlate_autopath(struct action_xlate_ctx *ctx, | |
4018 | const struct nx_action_autopath *naa) | |
4019 | { | |
4020 | uint16_t ofp_port = ntohl(naa->id); | |
4021 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
4022 | ||
4023 | if (!port || !port->bundle) { | |
4024 | ofp_port = OFPP_NONE; | |
4025 | } else if (port->bundle->bond) { | |
4026 | /* Autopath does not support VLAN hashing. */ | |
4027 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 4028 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
4029 | if (slave) { |
4030 | ofp_port = slave->up.ofp_port; | |
4031 | } | |
4032 | } | |
4033 | autopath_execute(naa, &ctx->flow, ofp_port); | |
4034 | } | |
4035 | ||
daff3353 EJ |
4036 | static bool |
4037 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
4038 | { | |
4039 | struct ofproto_dpif *ofproto = ofproto_; | |
4040 | struct ofport_dpif *port; | |
4041 | ||
4042 | switch (ofp_port) { | |
4043 | case OFPP_IN_PORT: | |
4044 | case OFPP_TABLE: | |
4045 | case OFPP_NORMAL: | |
4046 | case OFPP_FLOOD: | |
4047 | case OFPP_ALL: | |
439e4d8c | 4048 | case OFPP_NONE: |
daff3353 EJ |
4049 | return true; |
4050 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
4051 | return false; | |
4052 | default: | |
4053 | port = get_ofp_port(ofproto, ofp_port); | |
4054 | return port ? port->may_enable : false; | |
4055 | } | |
4056 | } | |
4057 | ||
75a75043 BP |
4058 | static void |
4059 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
4060 | const struct nx_action_learn *learn) | |
4061 | { | |
4062 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
4063 | struct ofputil_flow_mod fm; | |
4064 | int error; | |
4065 | ||
4066 | learn_execute(learn, &ctx->flow, &fm); | |
4067 | ||
4068 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
4069 | if (error && !VLOG_DROP_WARN(&rl)) { | |
4070 | char *msg = ofputil_error_to_string(error); | |
4071 | VLOG_WARN("learning action failed to modify flow table (%s)", msg); | |
4072 | free(msg); | |
4073 | } | |
4074 | ||
4075 | free(fm.actions); | |
4076 | } | |
4077 | ||
21f7563c JP |
4078 | static bool |
4079 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
4080 | { | |
4081 | if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
4082 | ? htonl(OFPPC_NO_RECV_STP) | |
4083 | : htonl(OFPPC_NO_RECV))) { | |
4084 | return false; | |
4085 | } | |
4086 | ||
4087 | /* Only drop packets here if both forwarding and learning are | |
4088 | * disabled. If just learning is enabled, we need to have | |
4089 | * OFPP_NORMAL and the learning action have a look at the packet | |
4090 | * before we can drop it. */ | |
4091 | if (!stp_forward_in_state(port->stp_state) | |
4092 | && !stp_learn_in_state(port->stp_state)) { | |
4093 | return false; | |
4094 | } | |
4095 | ||
4096 | return true; | |
4097 | } | |
4098 | ||
abe529af BP |
4099 | static void |
4100 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4101 | struct action_xlate_ctx *ctx) | |
4102 | { | |
4103 | const struct ofport_dpif *port; | |
abe529af | 4104 | const union ofp_action *ia; |
b4b8c781 | 4105 | size_t left; |
abe529af BP |
4106 | |
4107 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
21f7563c | 4108 | if (port && !may_receive(port, ctx)) { |
abe529af BP |
4109 | /* Drop this flow. */ |
4110 | return; | |
4111 | } | |
4112 | ||
b4b8c781 | 4113 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 4114 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
4115 | const struct nx_action_resubmit *nar; |
4116 | const struct nx_action_set_tunnel *nast; | |
4117 | const struct nx_action_set_queue *nasq; | |
4118 | const struct nx_action_multipath *nam; | |
4119 | const struct nx_action_autopath *naa; | |
daff3353 | 4120 | const struct nx_action_bundle *nab; |
f694937d | 4121 | const struct nx_action_output_reg *naor; |
38f2e360 BP |
4122 | enum ofputil_action_code code; |
4123 | ovs_be64 tun_id; | |
4124 | ||
848e8809 EJ |
4125 | if (ctx->exit) { |
4126 | break; | |
4127 | } | |
4128 | ||
38f2e360 BP |
4129 | code = ofputil_decode_action_unsafe(ia); |
4130 | switch (code) { | |
4131 | case OFPUTIL_OFPAT_OUTPUT: | |
abe529af BP |
4132 | xlate_output_action(ctx, &ia->output); |
4133 | break; | |
4134 | ||
38f2e360 | 4135 | case OFPUTIL_OFPAT_SET_VLAN_VID: |
abe529af BP |
4136 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
4137 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
4138 | break; |
4139 | ||
38f2e360 | 4140 | case OFPUTIL_OFPAT_SET_VLAN_PCP: |
abe529af BP |
4141 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
4142 | ctx->flow.vlan_tci |= htons( | |
4143 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
4144 | break; |
4145 | ||
38f2e360 | 4146 | case OFPUTIL_OFPAT_STRIP_VLAN: |
abe529af | 4147 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
4148 | break; |
4149 | ||
38f2e360 | 4150 | case OFPUTIL_OFPAT_SET_DL_SRC: |
abe529af | 4151 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4152 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
4153 | break; | |
4154 | ||
38f2e360 | 4155 | case OFPUTIL_OFPAT_SET_DL_DST: |
abe529af | 4156 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4157 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
4158 | break; | |
4159 | ||
38f2e360 | 4160 | case OFPUTIL_OFPAT_SET_NW_SRC: |
abe529af BP |
4161 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
4162 | break; | |
4163 | ||
38f2e360 | 4164 | case OFPUTIL_OFPAT_SET_NW_DST: |
abe529af BP |
4165 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
4166 | break; | |
4167 | ||
38f2e360 | 4168 | case OFPUTIL_OFPAT_SET_NW_TOS: |
7257b535 BP |
4169 | ctx->flow.tos_frag &= ~IP_DSCP_MASK; |
4170 | ctx->flow.tos_frag |= ia->nw_tos.nw_tos & IP_DSCP_MASK; | |
abe529af BP |
4171 | break; |
4172 | ||
38f2e360 | 4173 | case OFPUTIL_OFPAT_SET_TP_SRC: |
abe529af BP |
4174 | ctx->flow.tp_src = ia->tp_port.tp_port; |
4175 | break; | |
4176 | ||
38f2e360 | 4177 | case OFPUTIL_OFPAT_SET_TP_DST: |
abe529af BP |
4178 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
4179 | break; | |
4180 | ||
38f2e360 BP |
4181 | case OFPUTIL_OFPAT_ENQUEUE: |
4182 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
4183 | break; | |
4184 | ||
4185 | case OFPUTIL_NXAST_RESUBMIT: | |
4186 | nar = (const struct nx_action_resubmit *) ia; | |
29901626 BP |
4187 | xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); |
4188 | break; | |
4189 | ||
4190 | case OFPUTIL_NXAST_RESUBMIT_TABLE: | |
4191 | xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); | |
abe529af BP |
4192 | break; |
4193 | ||
38f2e360 BP |
4194 | case OFPUTIL_NXAST_SET_TUNNEL: |
4195 | nast = (const struct nx_action_set_tunnel *) ia; | |
4196 | tun_id = htonll(ntohl(nast->tun_id)); | |
4197 | ctx->flow.tun_id = tun_id; | |
4198 | break; | |
4199 | ||
4200 | case OFPUTIL_NXAST_SET_QUEUE: | |
4201 | nasq = (const struct nx_action_set_queue *) ia; | |
4202 | xlate_set_queue_action(ctx, nasq); | |
4203 | break; | |
4204 | ||
4205 | case OFPUTIL_NXAST_POP_QUEUE: | |
abff858b | 4206 | ctx->flow.priority = ctx->original_priority; |
38f2e360 BP |
4207 | break; |
4208 | ||
4209 | case OFPUTIL_NXAST_REG_MOVE: | |
4210 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
4211 | &ctx->flow); | |
4212 | break; | |
4213 | ||
4214 | case OFPUTIL_NXAST_REG_LOAD: | |
4215 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
4216 | &ctx->flow); | |
4217 | break; | |
4218 | ||
4219 | case OFPUTIL_NXAST_NOTE: | |
4220 | /* Nothing to do. */ | |
4221 | break; | |
4222 | ||
4223 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
4224 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
4225 | ctx->flow.tun_id = tun_id; | |
4226 | break; | |
4227 | ||
4228 | case OFPUTIL_NXAST_MULTIPATH: | |
4229 | nam = (const struct nx_action_multipath *) ia; | |
4230 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
4231 | break; |
4232 | ||
38f2e360 BP |
4233 | case OFPUTIL_NXAST_AUTOPATH: |
4234 | naa = (const struct nx_action_autopath *) ia; | |
4235 | xlate_autopath(ctx, naa); | |
abe529af | 4236 | break; |
daff3353 EJ |
4237 | |
4238 | case OFPUTIL_NXAST_BUNDLE: | |
4239 | ctx->ofproto->has_bundle_action = true; | |
4240 | nab = (const struct nx_action_bundle *) ia; | |
4241 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
4242 | slave_enabled_cb, | |
4243 | ctx->ofproto), 0); | |
4244 | break; | |
a368bb53 EJ |
4245 | |
4246 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
4247 | ctx->ofproto->has_bundle_action = true; | |
4248 | nab = (const struct nx_action_bundle *) ia; | |
4249 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
4250 | ctx->ofproto); | |
4251 | break; | |
f694937d EJ |
4252 | |
4253 | case OFPUTIL_NXAST_OUTPUT_REG: | |
4254 | naor = (const struct nx_action_output_reg *) ia; | |
4255 | xlate_output_reg_action(ctx, naor); | |
4256 | break; | |
75a75043 BP |
4257 | |
4258 | case OFPUTIL_NXAST_LEARN: | |
4259 | ctx->has_learn = true; | |
4260 | if (ctx->may_learn) { | |
4261 | xlate_learn_action(ctx, (const struct nx_action_learn *) ia); | |
4262 | } | |
4263 | break; | |
848e8809 EJ |
4264 | |
4265 | case OFPUTIL_NXAST_EXIT: | |
4266 | ctx->exit = true; | |
4267 | break; | |
abe529af BP |
4268 | } |
4269 | } | |
21f7563c JP |
4270 | |
4271 | /* We've let OFPP_NORMAL and the learning action look at the packet, | |
4272 | * so drop it now if forwarding is disabled. */ | |
4273 | if (port && !stp_forward_in_state(port->stp_state)) { | |
4274 | ofpbuf_clear(ctx->odp_actions); | |
4275 | add_sflow_action(ctx); | |
4276 | } | |
abe529af BP |
4277 | } |
4278 | ||
4279 | static void | |
4280 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
4281 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
4282 | const struct ofpbuf *packet) | |
4283 | { | |
4284 | ctx->ofproto = ofproto; | |
4285 | ctx->flow = *flow; | |
4286 | ctx->packet = packet; | |
75a75043 | 4287 | ctx->may_learn = packet != NULL; |
abe529af | 4288 | ctx->resubmit_hook = NULL; |
abe529af BP |
4289 | } |
4290 | ||
4291 | static struct ofpbuf * | |
4292 | xlate_actions(struct action_xlate_ctx *ctx, | |
4293 | const union ofp_action *in, size_t n_in) | |
4294 | { | |
4295 | COVERAGE_INC(ofproto_dpif_xlate); | |
4296 | ||
4297 | ctx->odp_actions = ofpbuf_new(512); | |
b6848f13 | 4298 | ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE); |
97e42c92 BP |
4299 | ctx->tags = 0; |
4300 | ctx->may_set_up_flow = true; | |
4301 | ctx->has_learn = false; | |
4302 | ctx->has_normal = false; | |
4303 | ctx->nf_output_iface = NF_OUT_DROP; | |
4304 | ctx->recurse = 0; | |
abff858b | 4305 | ctx->original_priority = ctx->flow.priority; |
97e42c92 BP |
4306 | ctx->base_flow = ctx->flow; |
4307 | ctx->base_flow.tun_id = 0; | |
4308 | ctx->table_id = 0; | |
848e8809 | 4309 | ctx->exit = false; |
7257b535 BP |
4310 | |
4311 | if (ctx->flow.tos_frag & FLOW_FRAG_ANY) { | |
4312 | switch (ctx->ofproto->up.frag_handling) { | |
4313 | case OFPC_FRAG_NORMAL: | |
4314 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
4315 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
4316 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
4317 | break; |
4318 | ||
4319 | case OFPC_FRAG_DROP: | |
4320 | return ctx->odp_actions; | |
4321 | ||
4322 | case OFPC_FRAG_REASM: | |
4323 | NOT_REACHED(); | |
4324 | ||
4325 | case OFPC_FRAG_NX_MATCH: | |
4326 | /* Nothing to do. */ | |
4327 | break; | |
4328 | } | |
4329 | } | |
4330 | ||
fc08b7a2 | 4331 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af | 4332 | ctx->may_set_up_flow = false; |
b6848f13 | 4333 | return ctx->odp_actions; |
abe529af | 4334 | } else { |
6ff686f2 | 4335 | add_sflow_action(ctx); |
abe529af | 4336 | do_xlate_actions(in, n_in, ctx); |
abe529af | 4337 | |
b6848f13 BP |
4338 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
4339 | ctx->odp_actions->data, | |
4340 | ctx->odp_actions->size)) { | |
4341 | ctx->may_set_up_flow = false; | |
4342 | if (ctx->packet | |
4343 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
4344 | ctx->packet)) { | |
a7c4eaf6 | 4345 | compose_output_action(ctx, OVSP_LOCAL); |
b6848f13 BP |
4346 | } |
4347 | } | |
a7c4eaf6 | 4348 | fix_sflow_action(ctx); |
abe529af BP |
4349 | } |
4350 | ||
4351 | return ctx->odp_actions; | |
4352 | } | |
4353 | \f | |
4354 | /* OFPP_NORMAL implementation. */ | |
4355 | ||
4356 | struct dst { | |
4357 | struct ofport_dpif *port; | |
dc155bff | 4358 | uint16_t vid; |
abe529af BP |
4359 | }; |
4360 | ||
4361 | struct dst_set { | |
4362 | struct dst builtin[32]; | |
4363 | struct dst *dsts; | |
4364 | size_t n, allocated; | |
4365 | }; | |
4366 | ||
4367 | static void dst_set_init(struct dst_set *); | |
4368 | static void dst_set_add(struct dst_set *, const struct dst *); | |
4369 | static void dst_set_free(struct dst_set *); | |
4370 | ||
4371 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); | |
4372 | ||
ecac4ebf BP |
4373 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
4374 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
4375 | * the bundle on which the packet was received, returns the VLAN to which the | |
4376 | * packet belongs. | |
4377 | * | |
4378 | * Both 'vid' and the return value are in the range 0...4095. */ | |
4379 | static uint16_t | |
4380 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
4381 | { | |
4382 | switch (in_bundle->vlan_mode) { | |
4383 | case PORT_VLAN_ACCESS: | |
4384 | return in_bundle->vlan; | |
4385 | break; | |
4386 | ||
4387 | case PORT_VLAN_TRUNK: | |
4388 | return vid; | |
4389 | ||
4390 | case PORT_VLAN_NATIVE_UNTAGGED: | |
4391 | case PORT_VLAN_NATIVE_TAGGED: | |
4392 | return vid ? vid : in_bundle->vlan; | |
4393 | ||
4394 | default: | |
4395 | NOT_REACHED(); | |
4396 | } | |
4397 | } | |
4398 | ||
4399 | /* Given 'vlan', the VLAN that a packet belongs to, and | |
4400 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
4401 | * that should be included in the 802.1Q header. (If the return value is 0, | |
4402 | * then the 802.1Q header should only be included in the packet if there is a | |
4403 | * nonzero PCP.) | |
4404 | * | |
4405 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
4406 | static uint16_t | |
4407 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
4408 | { | |
4409 | switch (out_bundle->vlan_mode) { | |
4410 | case PORT_VLAN_ACCESS: | |
4411 | return 0; | |
4412 | ||
4413 | case PORT_VLAN_TRUNK: | |
4414 | case PORT_VLAN_NATIVE_TAGGED: | |
4415 | return vlan; | |
4416 | ||
4417 | case PORT_VLAN_NATIVE_UNTAGGED: | |
4418 | return vlan == out_bundle->vlan ? 0 : vlan; | |
4419 | ||
4420 | default: | |
4421 | NOT_REACHED(); | |
4422 | } | |
4423 | } | |
4424 | ||
abe529af BP |
4425 | static bool |
4426 | set_dst(struct action_xlate_ctx *ctx, struct dst *dst, | |
4427 | const struct ofbundle *in_bundle, const struct ofbundle *out_bundle) | |
4428 | { | |
ecac4ebf BP |
4429 | uint16_t vlan; |
4430 | ||
4431 | vlan = input_vid_to_vlan(in_bundle, vlan_tci_to_vid(ctx->flow.vlan_tci)); | |
4432 | dst->vid = output_vlan_to_vid(out_bundle, vlan); | |
abe529af BP |
4433 | |
4434 | dst->port = (!out_bundle->bond | |
4435 | ? ofbundle_get_a_port(out_bundle) | |
4436 | : bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
dc155bff | 4437 | dst->vid, &ctx->tags)); |
abe529af BP |
4438 | return dst->port != NULL; |
4439 | } | |
4440 | ||
4441 | static int | |
4442 | mirror_mask_ffs(mirror_mask_t mask) | |
4443 | { | |
4444 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
4445 | return ffs(mask); | |
4446 | } | |
4447 | ||
4448 | static void | |
4449 | dst_set_init(struct dst_set *set) | |
4450 | { | |
4451 | set->dsts = set->builtin; | |
4452 | set->n = 0; | |
4453 | set->allocated = ARRAY_SIZE(set->builtin); | |
4454 | } | |
4455 | ||
4456 | static void | |
4457 | dst_set_add(struct dst_set *set, const struct dst *dst) | |
4458 | { | |
4459 | if (set->n >= set->allocated) { | |
4460 | size_t new_allocated; | |
4461 | struct dst *new_dsts; | |
4462 | ||
4463 | new_allocated = set->allocated * 2; | |
4464 | new_dsts = xmalloc(new_allocated * sizeof *new_dsts); | |
4465 | memcpy(new_dsts, set->dsts, set->n * sizeof *new_dsts); | |
4466 | ||
4467 | dst_set_free(set); | |
4468 | ||
4469 | set->dsts = new_dsts; | |
4470 | set->allocated = new_allocated; | |
4471 | } | |
4472 | set->dsts[set->n++] = *dst; | |
4473 | } | |
4474 | ||
4475 | static void | |
4476 | dst_set_free(struct dst_set *set) | |
4477 | { | |
4478 | if (set->dsts != set->builtin) { | |
4479 | free(set->dsts); | |
4480 | } | |
4481 | } | |
4482 | ||
4483 | static bool | |
4484 | dst_is_duplicate(const struct dst_set *set, const struct dst *test) | |
4485 | { | |
4486 | size_t i; | |
4487 | for (i = 0; i < set->n; i++) { | |
dc155bff | 4488 | if (set->dsts[i].vid == test->vid |
abe529af BP |
4489 | && set->dsts[i].port == test->port) { |
4490 | return true; | |
4491 | } | |
4492 | } | |
4493 | return false; | |
4494 | } | |
4495 | ||
4496 | static bool | |
4497 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
4498 | { | |
ecac4ebf | 4499 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 4500 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
4501 | } |
4502 | ||
4503 | static bool | |
4504 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
4505 | { | |
4506 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
4507 | } | |
4508 | ||
4509 | /* Returns an arbitrary interface within 'bundle'. */ | |
4510 | static struct ofport_dpif * | |
4511 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
4512 | { | |
4513 | return CONTAINER_OF(list_front(&bundle->ports), | |
4514 | struct ofport_dpif, bundle_node); | |
4515 | } | |
4516 | ||
4517 | static void | |
4518 | compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan, | |
4519 | const struct ofbundle *in_bundle, | |
4520 | const struct ofbundle *out_bundle, struct dst_set *set) | |
4521 | { | |
4522 | struct dst dst; | |
4523 | ||
4524 | if (out_bundle == OFBUNDLE_FLOOD) { | |
4525 | struct ofbundle *bundle; | |
4526 | ||
4527 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { | |
4528 | if (bundle != in_bundle | |
4529 | && ofbundle_includes_vlan(bundle, vlan) | |
4530 | && bundle->floodable | |
4531 | && !bundle->mirror_out | |
4532 | && set_dst(ctx, &dst, in_bundle, bundle)) { | |
4533 | dst_set_add(set, &dst); | |
4534 | } | |
4535 | } | |
4536 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
4537 | } else if (out_bundle && set_dst(ctx, &dst, in_bundle, out_bundle)) { | |
4538 | dst_set_add(set, &dst); | |
4539 | ctx->nf_output_iface = dst.port->odp_port; | |
4540 | } | |
4541 | } | |
4542 | ||
4543 | static bool | |
4544 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
4545 | { | |
fc3d7408 | 4546 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
4547 | } |
4548 | ||
07817dfe BP |
4549 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
4550 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
4551 | * protocols that may confuse switches. */ | |
4552 | static bool | |
4553 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
4554 | { | |
4555 | /* If you change this function's behavior, please update corresponding | |
4556 | * documentation in vswitch.xml at the same time. */ | |
4557 | if (dst[0] != 0x01) { | |
4558 | /* All the currently banned MACs happen to start with 01 currently, so | |
4559 | * this is a quick way to eliminate most of the good ones. */ | |
4560 | } else { | |
4561 | if (eth_addr_is_reserved(dst)) { | |
4562 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
4563 | * (01-80-c2-00-00-0x). */ | |
4564 | return false; | |
4565 | } | |
4566 | ||
4567 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
4568 | /* Cisco OUI. */ | |
4569 | if ((dst[3] & 0xfe) == 0xcc && | |
4570 | (dst[4] & 0xfe) == 0xcc && | |
4571 | (dst[5] & 0xfe) == 0xcc) { | |
4572 | /* Drop the following protocols plus others following the same | |
4573 | pattern: | |
4574 | ||
4575 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
4576 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
4577 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
4578 | return false; | |
4579 | } | |
4580 | ||
4581 | if (!(dst[3] | dst[4] | dst[5])) { | |
4582 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
4583 | return false; | |
4584 | } | |
4585 | } | |
4586 | } | |
4587 | return true; | |
4588 | } | |
4589 | ||
abe529af BP |
4590 | static void |
4591 | compose_mirror_dsts(struct action_xlate_ctx *ctx, | |
4592 | uint16_t vlan, const struct ofbundle *in_bundle, | |
4593 | struct dst_set *set) | |
4594 | { | |
4595 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
4596 | mirror_mask_t mirrors; | |
dc155bff | 4597 | uint16_t flow_vid; |
abe529af BP |
4598 | size_t i; |
4599 | ||
4600 | mirrors = in_bundle->src_mirrors; | |
4601 | for (i = 0; i < set->n; i++) { | |
4602 | mirrors |= set->dsts[i].port->bundle->dst_mirrors; | |
4603 | } | |
4604 | ||
4605 | if (!mirrors) { | |
4606 | return; | |
4607 | } | |
4608 | ||
dc155bff | 4609 | flow_vid = vlan_tci_to_vid(ctx->flow.vlan_tci); |
abe529af BP |
4610 | while (mirrors) { |
4611 | struct ofmirror *m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
4612 | if (vlan_is_mirrored(m, vlan)) { | |
4613 | struct dst dst; | |
4614 | ||
4615 | if (m->out) { | |
4616 | if (set_dst(ctx, &dst, in_bundle, m->out) | |
4617 | && !dst_is_duplicate(set, &dst)) { | |
4618 | dst_set_add(set, &dst); | |
4619 | } | |
07817dfe | 4620 | } else if (eth_dst_may_rspan(ctx->flow.dl_dst)) { |
abe529af BP |
4621 | struct ofbundle *bundle; |
4622 | ||
4623 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
4624 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
4625 | && set_dst(ctx, &dst, in_bundle, bundle)) | |
4626 | { | |
ecac4ebf BP |
4627 | /* set_dst() got dst->vid from the input packet's VLAN, |
4628 | * not from m->out_vlan, so recompute it. */ | |
4629 | dst.vid = output_vlan_to_vid(bundle, m->out_vlan); | |
4630 | ||
abe529af BP |
4631 | if (dst_is_duplicate(set, &dst)) { |
4632 | continue; | |
4633 | } | |
4634 | ||
dc155bff | 4635 | if (bundle == in_bundle && dst.vid == flow_vid) { |
abe529af BP |
4636 | /* Don't send out input port on same VLAN. */ |
4637 | continue; | |
4638 | } | |
4639 | dst_set_add(set, &dst); | |
4640 | } | |
4641 | } | |
4642 | } | |
4643 | } | |
4644 | mirrors &= mirrors - 1; | |
4645 | } | |
4646 | } | |
4647 | ||
4648 | static void | |
4649 | compose_actions(struct action_xlate_ctx *ctx, uint16_t vlan, | |
4650 | const struct ofbundle *in_bundle, | |
4651 | const struct ofbundle *out_bundle) | |
4652 | { | |
dc155bff | 4653 | uint16_t initial_vid, cur_vid; |
abe529af BP |
4654 | const struct dst *dst; |
4655 | struct dst_set set; | |
4656 | ||
4657 | dst_set_init(&set); | |
4658 | compose_dsts(ctx, vlan, in_bundle, out_bundle, &set); | |
4659 | compose_mirror_dsts(ctx, vlan, in_bundle, &set); | |
823518f1 BP |
4660 | if (!set.n) { |
4661 | dst_set_free(&set); | |
4662 | return; | |
4663 | } | |
abe529af BP |
4664 | |
4665 | /* Output all the packets we can without having to change the VLAN. */ | |
823518f1 | 4666 | commit_odp_actions(ctx); |
dc155bff | 4667 | initial_vid = vlan_tci_to_vid(ctx->flow.vlan_tci); |
abe529af | 4668 | for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { |
dc155bff | 4669 | if (dst->vid != initial_vid) { |
abe529af BP |
4670 | continue; |
4671 | } | |
6ff686f2 | 4672 | compose_output_action(ctx, dst->port->odp_port); |
abe529af BP |
4673 | } |
4674 | ||
4675 | /* Then output the rest. */ | |
dc155bff | 4676 | cur_vid = initial_vid; |
abe529af | 4677 | for (dst = set.dsts; dst < &set.dsts[set.n]; dst++) { |
dc155bff | 4678 | if (dst->vid == initial_vid) { |
abe529af BP |
4679 | continue; |
4680 | } | |
dc155bff | 4681 | if (dst->vid != cur_vid) { |
823518f1 | 4682 | ovs_be16 tci; |
d9065a90 | 4683 | |
dc155bff | 4684 | tci = htons(dst->vid); |
823518f1 BP |
4685 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); |
4686 | if (tci) { | |
4687 | tci |= htons(VLAN_CFI); | |
abe529af | 4688 | } |
4edb9ae9 | 4689 | commit_vlan_action(ctx, tci); |
823518f1 | 4690 | |
dc155bff | 4691 | cur_vid = dst->vid; |
abe529af | 4692 | } |
6ff686f2 | 4693 | compose_output_action(ctx, dst->port->odp_port); |
abe529af BP |
4694 | } |
4695 | ||
4696 | dst_set_free(&set); | |
4697 | } | |
4698 | ||
4699 | /* Returns the effective vlan of a packet, taking into account both the | |
4700 | * 802.1Q header and implicitly tagged ports. A value of 0 indicates that | |
4701 | * the packet is untagged and -1 indicates it has an invalid header and | |
4702 | * should be dropped. */ | |
4703 | static int | |
4704 | flow_get_vlan(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4705 | struct ofbundle *in_bundle, bool have_packet) | |
4706 | { | |
4707 | int vlan = vlan_tci_to_vid(flow->vlan_tci); | |
ecac4ebf BP |
4708 | if (vlan) { |
4709 | if (in_bundle->vlan_mode == PORT_VLAN_ACCESS) { | |
4710 | /* Drop tagged packet on access port */ | |
abe529af BP |
4711 | if (have_packet) { |
4712 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4713 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
4714 | "packet received on port %s configured with " | |
4715 | "implicit VLAN %"PRIu16, | |
4716 | ofproto->up.name, vlan, | |
4717 | in_bundle->name, in_bundle->vlan); | |
4718 | } | |
4719 | return -1; | |
ecac4ebf BP |
4720 | } else if (ofbundle_includes_vlan(in_bundle, vlan)) { |
4721 | return vlan; | |
4722 | } else { | |
4723 | /* Drop packets from a VLAN not member of the trunk */ | |
abe529af BP |
4724 | if (have_packet) { |
4725 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4726 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %d tagged " | |
4727 | "packet received on port %s not configured for " | |
4728 | "trunking VLAN %d", | |
4729 | ofproto->up.name, vlan, in_bundle->name, vlan); | |
4730 | } | |
4731 | return -1; | |
4732 | } | |
ecac4ebf BP |
4733 | } else { |
4734 | if (in_bundle->vlan_mode != PORT_VLAN_TRUNK) { | |
4735 | return in_bundle->vlan; | |
4736 | } else { | |
4737 | return ofbundle_includes_vlan(in_bundle, 0) ? 0 : -1; | |
4738 | } | |
abe529af | 4739 | } |
abe529af BP |
4740 | } |
4741 | ||
4742 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after | |
4743 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
4744 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
4745 | static bool | |
4746 | is_gratuitous_arp(const struct flow *flow) | |
4747 | { | |
4748 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
4749 | && eth_addr_is_broadcast(flow->dl_dst) | |
4750 | && (flow->nw_proto == ARP_OP_REPLY | |
4751 | || (flow->nw_proto == ARP_OP_REQUEST | |
4752 | && flow->nw_src == flow->nw_dst))); | |
4753 | } | |
4754 | ||
4755 | static void | |
4756 | update_learning_table(struct ofproto_dpif *ofproto, | |
4757 | const struct flow *flow, int vlan, | |
4758 | struct ofbundle *in_bundle) | |
4759 | { | |
4760 | struct mac_entry *mac; | |
4761 | ||
4762 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { | |
4763 | return; | |
4764 | } | |
4765 | ||
4766 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
4767 | if (is_gratuitous_arp(flow)) { | |
4768 | /* We don't want to learn from gratuitous ARP packets that are | |
4769 | * reflected back over bond slaves so we lock the learning table. */ | |
4770 | if (!in_bundle->bond) { | |
4771 | mac_entry_set_grat_arp_lock(mac); | |
4772 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
4773 | return; | |
4774 | } | |
4775 | } | |
4776 | ||
4777 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
4778 | /* The log messages here could actually be useful in debugging, | |
4779 | * so keep the rate limit relatively high. */ | |
4780 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
4781 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
4782 | "on port %s in VLAN %d", | |
4783 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
4784 | in_bundle->name, vlan); | |
4785 | ||
4786 | mac->port.p = in_bundle; | |
4787 | tag_set_add(&ofproto->revalidate_set, | |
4788 | mac_learning_changed(ofproto->ml, mac)); | |
4789 | } | |
4790 | } | |
4791 | ||
4792 | /* Determines whether packets in 'flow' within 'br' should be forwarded or | |
4793 | * dropped. Returns true if they may be forwarded, false if they should be | |
4794 | * dropped. | |
4795 | * | |
4796 | * If 'have_packet' is true, it indicates that the caller is processing a | |
4797 | * received packet. If 'have_packet' is false, then the caller is just | |
4798 | * revalidating an existing flow because configuration has changed. Either | |
4799 | * way, 'have_packet' only affects logging (there is no point in logging errors | |
4800 | * during revalidation). | |
4801 | * | |
4802 | * Sets '*in_portp' to the input port. This will be a null pointer if | |
4803 | * flow->in_port does not designate a known input port (in which case | |
4804 | * is_admissible() returns false). | |
4805 | * | |
4806 | * When returning true, sets '*vlanp' to the effective VLAN of the input | |
4807 | * packet, as returned by flow_get_vlan(). | |
4808 | * | |
4809 | * May also add tags to '*tags', although the current implementation only does | |
4810 | * so in one special case. | |
4811 | */ | |
4812 | static bool | |
4813 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
4814 | bool have_packet, | |
4815 | tag_type *tags, int *vlanp, struct ofbundle **in_bundlep) | |
4816 | { | |
4817 | struct ofport_dpif *in_port; | |
4818 | struct ofbundle *in_bundle; | |
4819 | int vlan; | |
4820 | ||
4821 | /* Find the port and bundle for the received packet. */ | |
4822 | in_port = get_ofp_port(ofproto, flow->in_port); | |
23adee42 | 4823 | *in_bundlep = in_bundle = in_port ? in_port->bundle : NULL; |
abe529af BP |
4824 | if (!in_port || !in_bundle) { |
4825 | /* No interface? Something fishy... */ | |
4826 | if (have_packet) { | |
4827 | /* Odd. A few possible reasons here: | |
4828 | * | |
4829 | * - We deleted a port but there are still a few packets queued up | |
4830 | * from it. | |
4831 | * | |
4832 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
4833 | * we don't know about. | |
4834 | * | |
4835 | * - Packet arrived on the local port but the local port is not | |
4836 | * part of a bundle. | |
4837 | */ | |
4838 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4839 | ||
4840 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
4841 | "port %"PRIu16, | |
4842 | ofproto->up.name, flow->in_port); | |
4843 | } | |
75a75043 | 4844 | *vlanp = -1; |
abe529af BP |
4845 | return false; |
4846 | } | |
4847 | *vlanp = vlan = flow_get_vlan(ofproto, flow, in_bundle, have_packet); | |
4848 | if (vlan < 0) { | |
4849 | return false; | |
4850 | } | |
4851 | ||
21f7563c JP |
4852 | /* Drop frames for reserved multicast addresses only if forward_bpdu |
4853 | * option is absent. */ | |
4854 | if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) { | |
abe529af BP |
4855 | return false; |
4856 | } | |
4857 | ||
4858 | /* Drop frames on bundles reserved for mirroring. */ | |
4859 | if (in_bundle->mirror_out) { | |
4860 | if (have_packet) { | |
4861 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4862 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
4863 | "%s, which is reserved exclusively for mirroring", | |
4864 | ofproto->up.name, in_bundle->name); | |
4865 | } | |
4866 | return false; | |
4867 | } | |
4868 | ||
4869 | if (in_bundle->bond) { | |
4870 | struct mac_entry *mac; | |
4871 | ||
4872 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
4873 | flow->dl_dst, tags)) { | |
4874 | case BV_ACCEPT: | |
4875 | break; | |
4876 | ||
4877 | case BV_DROP: | |
4878 | return false; | |
4879 | ||
4880 | case BV_DROP_IF_MOVED: | |
4881 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
4882 | if (mac && mac->port.p != in_bundle && | |
4883 | (!is_gratuitous_arp(flow) | |
4884 | || mac_entry_is_grat_arp_locked(mac))) { | |
4885 | return false; | |
4886 | } | |
4887 | break; | |
4888 | } | |
4889 | } | |
4890 | ||
4891 | return true; | |
4892 | } | |
4893 | ||
4cd78906 | 4894 | static void |
abe529af BP |
4895 | xlate_normal(struct action_xlate_ctx *ctx) |
4896 | { | |
4897 | struct ofbundle *in_bundle; | |
4898 | struct ofbundle *out_bundle; | |
4899 | struct mac_entry *mac; | |
4900 | int vlan; | |
4901 | ||
75a75043 BP |
4902 | ctx->has_normal = true; |
4903 | ||
abe529af BP |
4904 | /* Check whether we should drop packets in this flow. */ |
4905 | if (!is_admissible(ctx->ofproto, &ctx->flow, ctx->packet != NULL, | |
4906 | &ctx->tags, &vlan, &in_bundle)) { | |
4907 | out_bundle = NULL; | |
4908 | goto done; | |
4909 | } | |
4910 | ||
75a75043 BP |
4911 | /* Learn source MAC. */ |
4912 | if (ctx->may_learn) { | |
abe529af BP |
4913 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
4914 | } | |
4915 | ||
4916 | /* Determine output bundle. */ | |
4917 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
4918 | &ctx->tags); | |
4919 | if (mac) { | |
4920 | out_bundle = mac->port.p; | |
4921 | } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) { | |
4922 | /* If we are revalidating but don't have a learning entry then eject | |
4923 | * the flow. Installing a flow that floods packets opens up a window | |
4924 | * of time where we could learn from a packet reflected on a bond and | |
4925 | * blackhole packets before the learning table is updated to reflect | |
4926 | * the correct port. */ | |
4cd78906 BP |
4927 | ctx->may_set_up_flow = false; |
4928 | return; | |
abe529af BP |
4929 | } else { |
4930 | out_bundle = OFBUNDLE_FLOOD; | |
4931 | } | |
4932 | ||
4933 | /* Don't send packets out their input bundles. */ | |
4934 | if (in_bundle == out_bundle) { | |
4935 | out_bundle = NULL; | |
4936 | } | |
4937 | ||
4938 | done: | |
4939 | if (in_bundle) { | |
4940 | compose_actions(ctx, vlan, in_bundle, out_bundle); | |
4941 | } | |
abe529af BP |
4942 | } |
4943 | \f | |
54a9cbc9 BP |
4944 | /* Optimized flow revalidation. |
4945 | * | |
4946 | * It's a difficult problem, in general, to tell which facets need to have | |
4947 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
4948 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
4949 | * table changes, we recalculate the actions for every facet. This is | |
4950 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
4951 | * doesn't change very often. | |
4952 | * | |
4953 | * However, we can expect one particular kind of OpenFlow flow table change to | |
4954 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
4955 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
4956 | * table, we add a special case that applies to flow tables in which every rule | |
4957 | * has the same form (that is, the same wildcards), except that the table is | |
4958 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
4959 | * optimize this case by tagging all of the facets that resubmit into the table | |
4960 | * and invalidating the same tag whenever a flow changes in that table. The | |
4961 | * end result is that we revalidate just the facets that need it (and sometimes | |
4962 | * a few more, but not all of the facets or even all of the facets that | |
4963 | * resubmit to the table modified by MAC learning). */ | |
4964 | ||
4965 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
4966 | * into an OpenFlow table with the given 'basis'. */ | |
4967 | static uint32_t | |
4968 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, | |
4969 | uint32_t secret) | |
4970 | { | |
4971 | if (flow_wildcards_is_catchall(wc)) { | |
4972 | return 0; | |
4973 | } else { | |
4974 | struct flow tag_flow = *flow; | |
4975 | flow_zero_wildcards(&tag_flow, wc); | |
4976 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
4977 | } | |
4978 | } | |
4979 | ||
4980 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
4981 | * taggability of that table. | |
4982 | * | |
4983 | * This function must be called after *each* change to a flow table. If you | |
4984 | * skip calling it on some changes then the pointer comparisons at the end can | |
4985 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
4986 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
4987 | * different wildcards to be created with the same address, then this function | |
4988 | * will incorrectly skip revalidation. */ | |
4989 | static void | |
4990 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
4991 | { | |
4992 | struct table_dpif *table = &ofproto->tables[table_id]; | |
4993 | const struct classifier *cls = &ofproto->up.tables[table_id]; | |
4994 | struct cls_table *catchall, *other; | |
4995 | struct cls_table *t; | |
4996 | ||
4997 | catchall = other = NULL; | |
4998 | ||
4999 | switch (hmap_count(&cls->tables)) { | |
5000 | case 0: | |
5001 | /* We could tag this OpenFlow table but it would make the logic a | |
5002 | * little harder and it's a corner case that doesn't seem worth it | |
5003 | * yet. */ | |
5004 | break; | |
5005 | ||
5006 | case 1: | |
5007 | case 2: | |
5008 | HMAP_FOR_EACH (t, hmap_node, &cls->tables) { | |
5009 | if (cls_table_is_catchall(t)) { | |
5010 | catchall = t; | |
5011 | } else if (!other) { | |
5012 | other = t; | |
5013 | } else { | |
5014 | /* Indicate that we can't tag this by setting both tables to | |
5015 | * NULL. (We know that 'catchall' is already NULL.) */ | |
5016 | other = NULL; | |
5017 | } | |
5018 | } | |
5019 | break; | |
5020 | ||
5021 | default: | |
5022 | /* Can't tag this table. */ | |
5023 | break; | |
5024 | } | |
5025 | ||
5026 | if (table->catchall_table != catchall || table->other_table != other) { | |
5027 | table->catchall_table = catchall; | |
5028 | table->other_table = other; | |
5029 | ofproto->need_revalidate = true; | |
5030 | } | |
5031 | } | |
5032 | ||
5033 | /* Given 'rule' that has changed in some way (either it is a rule being | |
5034 | * inserted, a rule being deleted, or a rule whose actions are being | |
5035 | * modified), marks facets for revalidation to ensure that packets will be | |
5036 | * forwarded correctly according to the new state of the flow table. | |
5037 | * | |
5038 | * This function must be called after *each* change to a flow table. See | |
5039 | * the comment on table_update_taggable() for more information. */ | |
5040 | static void | |
5041 | rule_invalidate(const struct rule_dpif *rule) | |
5042 | { | |
5043 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
5044 | ||
5045 | table_update_taggable(ofproto, rule->up.table_id); | |
5046 | ||
5047 | if (!ofproto->need_revalidate) { | |
5048 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
5049 | ||
5050 | if (table->other_table && rule->tag) { | |
5051 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
5052 | } else { | |
5053 | ofproto->need_revalidate = true; | |
5054 | } | |
5055 | } | |
5056 | } | |
5057 | \f | |
abe529af | 5058 | static bool |
7257b535 BP |
5059 | set_frag_handling(struct ofproto *ofproto_, |
5060 | enum ofp_config_flags frag_handling) | |
abe529af BP |
5061 | { |
5062 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af | 5063 | |
7257b535 BP |
5064 | if (frag_handling != OFPC_FRAG_REASM) { |
5065 | ofproto->need_revalidate = true; | |
5066 | return true; | |
5067 | } else { | |
5068 | return false; | |
5069 | } | |
abe529af BP |
5070 | } |
5071 | ||
5072 | static int | |
5073 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, | |
5074 | const struct flow *flow, | |
5075 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
5076 | { | |
5077 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5078 | int error; | |
5079 | ||
e1154f71 BP |
5080 | if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { |
5081 | return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT); | |
5082 | } | |
5083 | ||
abe529af BP |
5084 | error = validate_actions(ofp_actions, n_ofp_actions, flow, |
5085 | ofproto->max_ports); | |
5086 | if (!error) { | |
80e5eed9 | 5087 | struct odputil_keybuf keybuf; |
abe529af BP |
5088 | struct action_xlate_ctx ctx; |
5089 | struct ofpbuf *odp_actions; | |
80e5eed9 BP |
5090 | struct ofpbuf key; |
5091 | ||
5092 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
5093 | odp_flow_key_from_flow(&key, flow); | |
abe529af BP |
5094 | |
5095 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
5096 | odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); | |
80e5eed9 BP |
5097 | dpif_execute(ofproto->dpif, key.data, key.size, |
5098 | odp_actions->data, odp_actions->size, packet); | |
abe529af BP |
5099 | ofpbuf_delete(odp_actions); |
5100 | } | |
5101 | return error; | |
5102 | } | |
5103 | ||
5104 | static void | |
5105 | get_netflow_ids(const struct ofproto *ofproto_, | |
5106 | uint8_t *engine_type, uint8_t *engine_id) | |
5107 | { | |
5108 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5109 | ||
5110 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
5111 | } | |
5112 | \f | |
5113 | static struct ofproto_dpif * | |
5114 | ofproto_dpif_lookup(const char *name) | |
5115 | { | |
5116 | struct ofproto *ofproto = ofproto_lookup(name); | |
5117 | return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class | |
5118 | ? ofproto_dpif_cast(ofproto) | |
5119 | : NULL); | |
5120 | } | |
5121 | ||
f0a3aa2e AA |
5122 | static void |
5123 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, | |
5124 | const char *args, void *aux OVS_UNUSED) | |
5125 | { | |
5126 | const struct ofproto_dpif *ofproto; | |
5127 | ||
5128 | ofproto = ofproto_dpif_lookup(args); | |
5129 | if (!ofproto) { | |
5130 | unixctl_command_reply(conn, 501, "no such bridge"); | |
5131 | return; | |
5132 | } | |
5133 | mac_learning_flush(ofproto->ml); | |
5134 | ||
5135 | unixctl_command_reply(conn, 200, "table successfully flushed"); | |
5136 | } | |
5137 | ||
abe529af BP |
5138 | static void |
5139 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, | |
5140 | const char *args, void *aux OVS_UNUSED) | |
5141 | { | |
5142 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5143 | const struct ofproto_dpif *ofproto; | |
5144 | const struct mac_entry *e; | |
5145 | ||
5146 | ofproto = ofproto_dpif_lookup(args); | |
5147 | if (!ofproto) { | |
5148 | unixctl_command_reply(conn, 501, "no such bridge"); | |
5149 | return; | |
5150 | } | |
5151 | ||
5152 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
5153 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
5154 | struct ofbundle *bundle = e->port.p; | |
5155 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
5156 | ofbundle_get_a_port(bundle)->odp_port, | |
5157 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
5158 | } | |
5159 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
5160 | ds_destroy(&ds); | |
5161 | } | |
5162 | ||
5163 | struct ofproto_trace { | |
5164 | struct action_xlate_ctx ctx; | |
5165 | struct flow flow; | |
5166 | struct ds *result; | |
5167 | }; | |
5168 | ||
5169 | static void | |
29901626 BP |
5170 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
5171 | const struct rule_dpif *rule) | |
abe529af BP |
5172 | { |
5173 | ds_put_char_multiple(result, '\t', level); | |
5174 | if (!rule) { | |
5175 | ds_put_cstr(result, "No match\n"); | |
5176 | return; | |
5177 | } | |
5178 | ||
29901626 BP |
5179 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
5180 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 5181 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
5182 | ds_put_char(result, '\n'); |
5183 | ||
5184 | ds_put_char_multiple(result, '\t', level); | |
5185 | ds_put_cstr(result, "OpenFlow "); | |
79feb7df | 5186 | ofp_print_actions(result, rule->up.actions, rule->up.n_actions); |
abe529af BP |
5187 | ds_put_char(result, '\n'); |
5188 | } | |
5189 | ||
5190 | static void | |
5191 | trace_format_flow(struct ds *result, int level, const char *title, | |
5192 | struct ofproto_trace *trace) | |
5193 | { | |
5194 | ds_put_char_multiple(result, '\t', level); | |
5195 | ds_put_format(result, "%s: ", title); | |
5196 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
5197 | ds_put_cstr(result, "unchanged"); | |
5198 | } else { | |
5199 | flow_format(result, &trace->ctx.flow); | |
5200 | trace->flow = trace->ctx.flow; | |
5201 | } | |
5202 | ds_put_char(result, '\n'); | |
5203 | } | |
5204 | ||
eb9e1c26 EJ |
5205 | static void |
5206 | trace_format_regs(struct ds *result, int level, const char *title, | |
5207 | struct ofproto_trace *trace) | |
5208 | { | |
5209 | size_t i; | |
5210 | ||
5211 | ds_put_char_multiple(result, '\t', level); | |
5212 | ds_put_format(result, "%s:", title); | |
5213 | for (i = 0; i < FLOW_N_REGS; i++) { | |
5214 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
5215 | } | |
5216 | ds_put_char(result, '\n'); | |
5217 | } | |
5218 | ||
abe529af BP |
5219 | static void |
5220 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
5221 | { | |
5222 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
5223 | struct ds *result = trace->result; | |
5224 | ||
5225 | ds_put_char(result, '\n'); | |
5226 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 5227 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
29901626 | 5228 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
5229 | } |
5230 | ||
5231 | static void | |
5232 | ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, | |
5233 | void *aux OVS_UNUSED) | |
5234 | { | |
abff858b | 5235 | char *dpname, *arg1, *arg2, *arg3, *arg4; |
abe529af BP |
5236 | char *args = xstrdup(args_); |
5237 | char *save_ptr = NULL; | |
5238 | struct ofproto_dpif *ofproto; | |
876b0e1c BP |
5239 | struct ofpbuf odp_key; |
5240 | struct ofpbuf *packet; | |
abe529af BP |
5241 | struct rule_dpif *rule; |
5242 | struct ds result; | |
5243 | struct flow flow; | |
abe529af BP |
5244 | char *s; |
5245 | ||
876b0e1c BP |
5246 | packet = NULL; |
5247 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
5248 | ds_init(&result); |
5249 | ||
5250 | dpname = strtok_r(args, " ", &save_ptr); | |
876b0e1c BP |
5251 | arg1 = strtok_r(NULL, " ", &save_ptr); |
5252 | arg2 = strtok_r(NULL, " ", &save_ptr); | |
abff858b PS |
5253 | arg3 = strtok_r(NULL, " ", &save_ptr); |
5254 | arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ | |
8b3b8dd1 BP |
5255 | if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) { |
5256 | /* ofproto/trace dpname flow [-generate] */ | |
876b0e1c BP |
5257 | int error; |
5258 | ||
df2c07f4 | 5259 | /* Convert string to datapath key. */ |
876b0e1c BP |
5260 | ofpbuf_init(&odp_key, 0); |
5261 | error = odp_flow_key_from_string(arg1, &odp_key); | |
5262 | if (error) { | |
5263 | unixctl_command_reply(conn, 501, "Bad flow syntax"); | |
5264 | goto exit; | |
5265 | } | |
5266 | ||
5267 | /* Convert odp_key to flow. */ | |
5268 | error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow); | |
5269 | if (error) { | |
5270 | unixctl_command_reply(conn, 501, "Invalid flow"); | |
5271 | goto exit; | |
5272 | } | |
8b3b8dd1 BP |
5273 | |
5274 | /* Generate a packet, if requested. */ | |
5275 | if (arg2) { | |
5276 | packet = ofpbuf_new(0); | |
5277 | flow_compose(packet, &flow); | |
5278 | } | |
abff858b PS |
5279 | } else if (dpname && arg1 && arg2 && arg3 && arg4) { |
5280 | /* ofproto/trace dpname priority tun_id in_port packet */ | |
876b0e1c BP |
5281 | uint16_t in_port; |
5282 | ovs_be64 tun_id; | |
abff858b | 5283 | uint32_t priority; |
876b0e1c | 5284 | |
abff858b PS |
5285 | priority = atoi(arg1); |
5286 | tun_id = htonll(strtoull(arg2, NULL, 0)); | |
5287 | in_port = ofp_port_to_odp_port(atoi(arg3)); | |
876b0e1c BP |
5288 | |
5289 | packet = ofpbuf_new(strlen(args) / 2); | |
abff858b PS |
5290 | arg4 = ofpbuf_put_hex(packet, arg4, NULL); |
5291 | arg4 += strspn(arg4, " "); | |
5292 | if (*arg4 != '\0') { | |
876b0e1c BP |
5293 | unixctl_command_reply(conn, 501, "Trailing garbage in command"); |
5294 | goto exit; | |
5295 | } | |
5296 | if (packet->size < ETH_HEADER_LEN) { | |
5297 | unixctl_command_reply(conn, 501, | |
5298 | "Packet data too short for Ethernet"); | |
5299 | goto exit; | |
5300 | } | |
5301 | ||
5302 | ds_put_cstr(&result, "Packet: "); | |
5303 | s = ofp_packet_to_string(packet->data, packet->size, packet->size); | |
5304 | ds_put_cstr(&result, s); | |
5305 | free(s); | |
5306 | ||
abff858b | 5307 | flow_extract(packet, priority, tun_id, in_port, &flow); |
876b0e1c | 5308 | } else { |
abe529af BP |
5309 | unixctl_command_reply(conn, 501, "Bad command syntax"); |
5310 | goto exit; | |
5311 | } | |
5312 | ||
5313 | ofproto = ofproto_dpif_lookup(dpname); | |
5314 | if (!ofproto) { | |
5315 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
5316 | "for help)"); | |
5317 | goto exit; | |
5318 | } | |
5319 | ||
abe529af BP |
5320 | ds_put_cstr(&result, "Flow: "); |
5321 | flow_format(&result, &flow); | |
5322 | ds_put_char(&result, '\n'); | |
5323 | ||
29901626 BP |
5324 | rule = rule_dpif_lookup(ofproto, &flow, 0); |
5325 | trace_format_rule(&result, 0, 0, rule); | |
abe529af BP |
5326 | if (rule) { |
5327 | struct ofproto_trace trace; | |
5328 | struct ofpbuf *odp_actions; | |
5329 | ||
5330 | trace.result = &result; | |
5331 | trace.flow = flow; | |
876b0e1c | 5332 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet); |
abe529af BP |
5333 | trace.ctx.resubmit_hook = trace_resubmit; |
5334 | odp_actions = xlate_actions(&trace.ctx, | |
5335 | rule->up.actions, rule->up.n_actions); | |
5336 | ||
5337 | ds_put_char(&result, '\n'); | |
5338 | trace_format_flow(&result, 0, "Final flow", &trace); | |
5339 | ds_put_cstr(&result, "Datapath actions: "); | |
5340 | format_odp_actions(&result, odp_actions->data, odp_actions->size); | |
5341 | ofpbuf_delete(odp_actions); | |
876b0e1c BP |
5342 | |
5343 | if (!trace.ctx.may_set_up_flow) { | |
5344 | if (packet) { | |
5345 | ds_put_cstr(&result, "\nThis flow is not cachable."); | |
5346 | } else { | |
5347 | ds_put_cstr(&result, "\nThe datapath actions are incomplete--" | |
5348 | "for complete actions, please supply a packet."); | |
5349 | } | |
5350 | } | |
abe529af BP |
5351 | } |
5352 | ||
5353 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
5354 | ||
5355 | exit: | |
5356 | ds_destroy(&result); | |
876b0e1c BP |
5357 | ofpbuf_delete(packet); |
5358 | ofpbuf_uninit(&odp_key); | |
abe529af BP |
5359 | free(args); |
5360 | } | |
5361 | ||
7ee20df1 BP |
5362 | static void |
5363 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, | |
5364 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
5365 | { | |
5366 | clogged = true; | |
5367 | unixctl_command_reply(conn, 200, NULL); | |
5368 | } | |
5369 | ||
5370 | static void | |
5371 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, | |
5372 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
5373 | { | |
5374 | clogged = false; | |
5375 | unixctl_command_reply(conn, 200, NULL); | |
5376 | } | |
5377 | ||
abe529af BP |
5378 | static void |
5379 | ofproto_dpif_unixctl_init(void) | |
5380 | { | |
5381 | static bool registered; | |
5382 | if (registered) { | |
5383 | return; | |
5384 | } | |
5385 | registered = true; | |
5386 | ||
7ff2009a JP |
5387 | unixctl_command_register("ofproto/trace", |
5388 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
5389 | ofproto_unixctl_trace, NULL); | |
f0a3aa2e AA |
5390 | unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush, |
5391 | NULL); | |
7ff2009a JP |
5392 | unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show, |
5393 | NULL); | |
5394 | unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL); | |
5395 | unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL); | |
abe529af BP |
5396 | } |
5397 | \f | |
5398 | const struct ofproto_class ofproto_dpif_class = { | |
5399 | enumerate_types, | |
5400 | enumerate_names, | |
5401 | del, | |
5402 | alloc, | |
5403 | construct, | |
5404 | destruct, | |
5405 | dealloc, | |
5406 | run, | |
5407 | wait, | |
5408 | flush, | |
6c1491fb BP |
5409 | get_features, |
5410 | get_tables, | |
abe529af BP |
5411 | port_alloc, |
5412 | port_construct, | |
5413 | port_destruct, | |
5414 | port_dealloc, | |
5415 | port_modified, | |
5416 | port_reconfigured, | |
5417 | port_query_by_name, | |
5418 | port_add, | |
5419 | port_del, | |
5420 | port_dump_start, | |
5421 | port_dump_next, | |
5422 | port_dump_done, | |
5423 | port_poll, | |
5424 | port_poll_wait, | |
5425 | port_is_lacp_current, | |
0ab6decf | 5426 | NULL, /* rule_choose_table */ |
abe529af BP |
5427 | rule_alloc, |
5428 | rule_construct, | |
5429 | rule_destruct, | |
5430 | rule_dealloc, | |
abe529af BP |
5431 | rule_get_stats, |
5432 | rule_execute, | |
5433 | rule_modify_actions, | |
7257b535 | 5434 | set_frag_handling, |
abe529af BP |
5435 | packet_out, |
5436 | set_netflow, | |
5437 | get_netflow_ids, | |
5438 | set_sflow, | |
5439 | set_cfm, | |
a5610457 | 5440 | get_cfm_fault, |
1de11730 | 5441 | get_cfm_remote_mpids, |
21f7563c JP |
5442 | set_stp, |
5443 | get_stp_status, | |
5444 | set_stp_port, | |
5445 | get_stp_port_status, | |
abe529af BP |
5446 | bundle_set, |
5447 | bundle_remove, | |
5448 | mirror_set, | |
5449 | set_flood_vlans, | |
5450 | is_mirror_output_bundle, | |
8402c74b | 5451 | forward_bpdu_changed, |
abe529af | 5452 | }; |