]>
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 | ||
9ba15e2a | 124 | /* Output (exactly one of out == NULL and out_vlan == -1 is true). */ |
abe529af BP |
125 | struct ofbundle *out; /* Output port or NULL. */ |
126 | int out_vlan; /* Output VLAN or -1. */ | |
9ba15e2a | 127 | mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */ |
abe529af BP |
128 | }; |
129 | ||
130 | static void mirror_destroy(struct ofmirror *); | |
131 | ||
132 | /* A group of one or more OpenFlow ports. */ | |
133 | #define OFBUNDLE_FLOOD ((struct ofbundle *) 1) | |
134 | struct ofbundle { | |
135 | struct ofproto_dpif *ofproto; /* Owning ofproto. */ | |
136 | struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */ | |
137 | void *aux; /* Key supplied by ofproto's client. */ | |
138 | char *name; /* Identifier for log messages. */ | |
139 | ||
140 | /* Configuration. */ | |
141 | struct list ports; /* Contains "struct ofport"s. */ | |
ecac4ebf | 142 | enum port_vlan_mode vlan_mode; /* VLAN mode */ |
abe529af BP |
143 | int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */ |
144 | unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1. | |
145 | * NULL if all VLANs are trunked. */ | |
146 | struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */ | |
147 | struct bond *bond; /* Nonnull iff more than one port. */ | |
5e9ceccd | 148 | bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */ |
abe529af BP |
149 | |
150 | /* Status. */ | |
151 | bool floodable; /* True if no port has OFPPC_NO_FLOOD set. */ | |
152 | ||
153 | /* Port mirroring info. */ | |
154 | mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */ | |
155 | mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */ | |
156 | mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */ | |
157 | }; | |
158 | ||
159 | static void bundle_remove(struct ofport *); | |
7bde8dd8 | 160 | static void bundle_update(struct ofbundle *); |
abe529af BP |
161 | static void bundle_destroy(struct ofbundle *); |
162 | static void bundle_del_port(struct ofport_dpif *); | |
163 | static void bundle_run(struct ofbundle *); | |
164 | static void bundle_wait(struct ofbundle *); | |
165 | ||
21f7563c JP |
166 | static void stp_run(struct ofproto_dpif *ofproto); |
167 | static void stp_wait(struct ofproto_dpif *ofproto); | |
168 | ||
5da5ec37 BP |
169 | static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan); |
170 | ||
abe529af BP |
171 | struct action_xlate_ctx { |
172 | /* action_xlate_ctx_init() initializes these members. */ | |
173 | ||
174 | /* The ofproto. */ | |
175 | struct ofproto_dpif *ofproto; | |
176 | ||
177 | /* Flow to which the OpenFlow actions apply. xlate_actions() will modify | |
178 | * this flow when actions change header fields. */ | |
179 | struct flow flow; | |
180 | ||
181 | /* The packet corresponding to 'flow', or a null pointer if we are | |
182 | * revalidating without a packet to refer to. */ | |
183 | const struct ofpbuf *packet; | |
184 | ||
75a75043 BP |
185 | /* Should OFPP_NORMAL MAC learning and NXAST_LEARN actions execute? We |
186 | * want to execute them if we are actually processing a packet, or if we | |
187 | * are accounting for packets that the datapath has processed, but not if | |
188 | * we are just revalidating. */ | |
189 | bool may_learn; | |
190 | ||
abe529af BP |
191 | /* If nonnull, called just before executing a resubmit action. |
192 | * | |
193 | * This is normally null so the client has to set it manually after | |
194 | * calling action_xlate_ctx_init(). */ | |
195 | void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *); | |
196 | ||
abe529af BP |
197 | /* xlate_actions() initializes and uses these members. The client might want |
198 | * to look at them after it returns. */ | |
199 | ||
200 | struct ofpbuf *odp_actions; /* Datapath actions. */ | |
75a75043 | 201 | tag_type tags; /* Tags associated with actions. */ |
abe529af BP |
202 | bool may_set_up_flow; /* True ordinarily; false if the actions must |
203 | * be reassessed for every packet. */ | |
75a75043 BP |
204 | bool has_learn; /* Actions include NXAST_LEARN? */ |
205 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
abe529af BP |
206 | uint16_t nf_output_iface; /* Output interface index for NetFlow. */ |
207 | ||
208 | /* xlate_actions() initializes and uses these members, but the client has no | |
209 | * reason to look at them. */ | |
210 | ||
211 | int recurse; /* Recursion level, via xlate_table_action. */ | |
b3e9b2ed | 212 | struct flow base_flow; /* Flow at the last commit. */ |
abff858b | 213 | uint32_t original_priority; /* Priority when packet arrived. */ |
29901626 | 214 | uint8_t table_id; /* OpenFlow table ID where flow was found. */ |
6ff686f2 PS |
215 | uint32_t sflow_n_outputs; /* Number of output ports. */ |
216 | uint16_t sflow_odp_port; /* Output port for composing sFlow action. */ | |
217 | uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */ | |
848e8809 | 218 | bool exit; /* No further actions should be processed. */ |
abe529af BP |
219 | }; |
220 | ||
221 | static void action_xlate_ctx_init(struct action_xlate_ctx *, | |
222 | struct ofproto_dpif *, const struct flow *, | |
223 | const struct ofpbuf *); | |
224 | static struct ofpbuf *xlate_actions(struct action_xlate_ctx *, | |
225 | const union ofp_action *in, size_t n_in); | |
226 | ||
227 | /* An exact-match instantiation of an OpenFlow flow. */ | |
228 | struct facet { | |
229 | long long int used; /* Time last used; time created if not used. */ | |
230 | ||
231 | /* These statistics: | |
232 | * | |
233 | * - Do include packets and bytes sent "by hand", e.g. with | |
234 | * dpif_execute(). | |
235 | * | |
236 | * - Do include packets and bytes that were obtained from the datapath | |
907a4c5e | 237 | * when its statistics were reset (e.g. dpif_flow_put() with |
abe529af | 238 | * DPIF_FP_ZERO_STATS). |
abe529af BP |
239 | */ |
240 | uint64_t packet_count; /* Number of packets received. */ | |
241 | uint64_t byte_count; /* Number of bytes received. */ | |
242 | ||
243 | uint64_t dp_packet_count; /* Last known packet count in the datapath. */ | |
244 | uint64_t dp_byte_count; /* Last known byte count in the datapath. */ | |
245 | ||
246 | uint64_t rs_packet_count; /* Packets pushed to resubmit children. */ | |
247 | uint64_t rs_byte_count; /* Bytes pushed to resubmit children. */ | |
248 | long long int rs_used; /* Used time pushed to resubmit children. */ | |
249 | ||
907a4c5e | 250 | uint64_t accounted_bytes; /* Bytes processed by facet_account(). */ |
abe529af BP |
251 | |
252 | struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */ | |
253 | struct list list_node; /* In owning rule's 'facets' list. */ | |
254 | struct rule_dpif *rule; /* Owning rule. */ | |
255 | struct flow flow; /* Exact-match flow. */ | |
256 | bool installed; /* Installed in datapath? */ | |
257 | bool may_install; /* True ordinarily; false if actions must | |
258 | * be reassessed for every packet. */ | |
75a75043 BP |
259 | bool has_learn; /* Actions include NXAST_LEARN? */ |
260 | bool has_normal; /* Actions output to OFPP_NORMAL? */ | |
abe529af BP |
261 | size_t actions_len; /* Number of bytes in actions[]. */ |
262 | struct nlattr *actions; /* Datapath actions. */ | |
263 | tag_type tags; /* Tags. */ | |
264 | struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */ | |
265 | }; | |
266 | ||
f3827897 | 267 | static struct facet *facet_create(struct rule_dpif *, const struct flow *); |
abe529af BP |
268 | static void facet_remove(struct ofproto_dpif *, struct facet *); |
269 | static void facet_free(struct facet *); | |
270 | ||
271 | static struct facet *facet_find(struct ofproto_dpif *, const struct flow *); | |
272 | static struct facet *facet_lookup_valid(struct ofproto_dpif *, | |
273 | const struct flow *); | |
274 | static bool facet_revalidate(struct ofproto_dpif *, struct facet *); | |
275 | ||
3d9e05f8 BP |
276 | static bool execute_controller_action(struct ofproto_dpif *, |
277 | const struct flow *, | |
278 | const struct nlattr *odp_actions, | |
279 | size_t actions_len, | |
280 | struct ofpbuf *packet); | |
abe529af BP |
281 | static void facet_execute(struct ofproto_dpif *, struct facet *, |
282 | struct ofpbuf *packet); | |
283 | ||
284 | static int facet_put__(struct ofproto_dpif *, struct facet *, | |
285 | const struct nlattr *actions, size_t actions_len, | |
286 | struct dpif_flow_stats *); | |
287 | static void facet_install(struct ofproto_dpif *, struct facet *, | |
288 | bool zero_stats); | |
289 | static void facet_uninstall(struct ofproto_dpif *, struct facet *); | |
290 | static void facet_flush_stats(struct ofproto_dpif *, struct facet *); | |
291 | ||
292 | static void facet_make_actions(struct ofproto_dpif *, struct facet *, | |
293 | const struct ofpbuf *packet); | |
294 | static void facet_update_time(struct ofproto_dpif *, struct facet *, | |
295 | long long int used); | |
296 | static void facet_update_stats(struct ofproto_dpif *, struct facet *, | |
297 | const struct dpif_flow_stats *); | |
bbb5d219 | 298 | static void facet_reset_counters(struct facet *); |
3a88e544 | 299 | static void facet_reset_dp_stats(struct facet *, struct dpif_flow_stats *); |
abe529af | 300 | static void facet_push_stats(struct facet *); |
55af77bb | 301 | static void facet_account(struct ofproto_dpif *, struct facet *); |
abe529af BP |
302 | |
303 | static bool facet_is_controller_flow(struct facet *); | |
304 | ||
305 | static void flow_push_stats(const struct rule_dpif *, | |
306 | struct flow *, uint64_t packets, uint64_t bytes, | |
307 | long long int used); | |
308 | ||
54a9cbc9 BP |
309 | static uint32_t rule_calculate_tag(const struct flow *, |
310 | const struct flow_wildcards *, | |
311 | uint32_t basis); | |
312 | static void rule_invalidate(const struct rule_dpif *); | |
313 | ||
abe529af BP |
314 | struct ofport_dpif { |
315 | struct ofport up; | |
316 | ||
317 | uint32_t odp_port; | |
318 | struct ofbundle *bundle; /* Bundle that contains this port, if any. */ | |
319 | struct list bundle_node; /* In struct ofbundle's "ports" list. */ | |
320 | struct cfm *cfm; /* Connectivity Fault Management, if any. */ | |
321 | tag_type tag; /* Tag associated with this port. */ | |
00794817 | 322 | uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */ |
015e08bc | 323 | bool may_enable; /* May be enabled in bonds. */ |
21f7563c JP |
324 | |
325 | struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */ | |
326 | enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */ | |
327 | long long int stp_state_entered; | |
abe529af BP |
328 | }; |
329 | ||
330 | static struct ofport_dpif * | |
331 | ofport_dpif_cast(const struct ofport *ofport) | |
332 | { | |
333 | assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class); | |
334 | return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL; | |
335 | } | |
336 | ||
337 | static void port_run(struct ofport_dpif *); | |
338 | static void port_wait(struct ofport_dpif *); | |
a5610457 | 339 | static int set_cfm(struct ofport *, const struct cfm_settings *); |
abe529af | 340 | |
7ee20df1 BP |
341 | struct dpif_completion { |
342 | struct list list_node; | |
343 | struct ofoperation *op; | |
344 | }; | |
345 | ||
54a9cbc9 BP |
346 | /* Extra information about a classifier table. |
347 | * Currently used just for optimized flow revalidation. */ | |
348 | struct table_dpif { | |
349 | /* If either of these is nonnull, then this table has a form that allows | |
350 | * flows to be tagged to avoid revalidating most flows for the most common | |
351 | * kinds of flow table changes. */ | |
352 | struct cls_table *catchall_table; /* Table that wildcards all fields. */ | |
353 | struct cls_table *other_table; /* Table with any other wildcard set. */ | |
354 | uint32_t basis; /* Keeps each table's tags separate. */ | |
355 | }; | |
356 | ||
abe529af BP |
357 | struct ofproto_dpif { |
358 | struct ofproto up; | |
359 | struct dpif *dpif; | |
360 | int max_ports; | |
361 | ||
6c1491fb BP |
362 | /* Statistics. */ |
363 | uint64_t n_matches; | |
364 | ||
abe529af BP |
365 | /* Bridging. */ |
366 | struct netflow *netflow; | |
bae473fe | 367 | struct dpif_sflow *sflow; |
abe529af BP |
368 | struct hmap bundles; /* Contains "struct ofbundle"s. */ |
369 | struct mac_learning *ml; | |
370 | struct ofmirror *mirrors[MAX_MIRRORS]; | |
371 | bool has_bonded_bundles; | |
372 | ||
373 | /* Expiration. */ | |
374 | struct timer next_expiration; | |
375 | ||
376 | /* Facets. */ | |
377 | struct hmap facets; | |
54a9cbc9 BP |
378 | |
379 | /* Revalidation. */ | |
380 | struct table_dpif tables[N_TABLES]; | |
abe529af BP |
381 | bool need_revalidate; |
382 | struct tag_set revalidate_set; | |
7ee20df1 BP |
383 | |
384 | /* Support for debugging async flow mods. */ | |
385 | struct list completions; | |
daff3353 EJ |
386 | |
387 | bool has_bundle_action; /* True when the first bundle action appears. */ | |
21f7563c JP |
388 | |
389 | /* Spanning tree. */ | |
390 | struct stp *stp; | |
391 | long long int stp_last_tick; | |
abe529af BP |
392 | }; |
393 | ||
7ee20df1 BP |
394 | /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only |
395 | * for debugging the asynchronous flow_mod implementation.) */ | |
396 | static bool clogged; | |
397 | ||
abe529af BP |
398 | static void ofproto_dpif_unixctl_init(void); |
399 | ||
400 | static struct ofproto_dpif * | |
401 | ofproto_dpif_cast(const struct ofproto *ofproto) | |
402 | { | |
403 | assert(ofproto->ofproto_class == &ofproto_dpif_class); | |
404 | return CONTAINER_OF(ofproto, struct ofproto_dpif, up); | |
405 | } | |
406 | ||
407 | static struct ofport_dpif *get_ofp_port(struct ofproto_dpif *, | |
408 | uint16_t ofp_port); | |
409 | static struct ofport_dpif *get_odp_port(struct ofproto_dpif *, | |
410 | uint32_t odp_port); | |
411 | ||
412 | /* Packet processing. */ | |
413 | static void update_learning_table(struct ofproto_dpif *, | |
414 | const struct flow *, int vlan, | |
415 | struct ofbundle *); | |
501f8d1f BP |
416 | /* Upcalls. */ |
417 | #define FLOW_MISS_MAX_BATCH 50 | |
395e68ce | 418 | |
abe529af | 419 | static void handle_upcall(struct ofproto_dpif *, struct dpif_upcall *); |
501f8d1f BP |
420 | static void handle_miss_upcalls(struct ofproto_dpif *, |
421 | struct dpif_upcall *, size_t n); | |
abe529af BP |
422 | |
423 | /* Flow expiration. */ | |
424 | static int expire(struct ofproto_dpif *); | |
425 | ||
426 | /* Utilities. */ | |
b2fda3ef | 427 | static int send_packet(struct ofproto_dpif *, uint32_t odp_port, |
abe529af | 428 | const struct ofpbuf *packet); |
6ff686f2 PS |
429 | static size_t |
430 | compose_sflow_action(const struct ofproto_dpif *, struct ofpbuf *odp_actions, | |
431 | const struct flow *, uint32_t odp_port); | |
abe529af BP |
432 | /* Global variables. */ |
433 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
434 | \f | |
435 | /* Factory functions. */ | |
436 | ||
437 | static void | |
438 | enumerate_types(struct sset *types) | |
439 | { | |
440 | dp_enumerate_types(types); | |
441 | } | |
442 | ||
443 | static int | |
444 | enumerate_names(const char *type, struct sset *names) | |
445 | { | |
446 | return dp_enumerate_names(type, names); | |
447 | } | |
448 | ||
449 | static int | |
450 | del(const char *type, const char *name) | |
451 | { | |
452 | struct dpif *dpif; | |
453 | int error; | |
454 | ||
455 | error = dpif_open(name, type, &dpif); | |
456 | if (!error) { | |
457 | error = dpif_delete(dpif); | |
458 | dpif_close(dpif); | |
459 | } | |
460 | return error; | |
461 | } | |
462 | \f | |
463 | /* Basic life-cycle. */ | |
464 | ||
465 | static struct ofproto * | |
466 | alloc(void) | |
467 | { | |
468 | struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto); | |
469 | return &ofproto->up; | |
470 | } | |
471 | ||
472 | static void | |
473 | dealloc(struct ofproto *ofproto_) | |
474 | { | |
475 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
476 | free(ofproto); | |
477 | } | |
478 | ||
479 | static int | |
073e2a6f | 480 | construct(struct ofproto *ofproto_, int *n_tablesp) |
abe529af BP |
481 | { |
482 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
483 | const char *name = ofproto->up.name; | |
484 | int error; | |
485 | int i; | |
486 | ||
487 | error = dpif_create_and_open(name, ofproto->up.type, &ofproto->dpif); | |
488 | if (error) { | |
489 | VLOG_ERR("failed to open datapath %s: %s", name, strerror(error)); | |
490 | return error; | |
491 | } | |
492 | ||
493 | ofproto->max_ports = dpif_get_max_ports(ofproto->dpif); | |
6c1491fb | 494 | ofproto->n_matches = 0; |
abe529af | 495 | |
be8194bb JG |
496 | dpif_flow_flush(ofproto->dpif); |
497 | dpif_recv_purge(ofproto->dpif); | |
498 | ||
abe529af BP |
499 | error = dpif_recv_set_mask(ofproto->dpif, |
500 | ((1u << DPIF_UC_MISS) | | |
6ff686f2 | 501 | (1u << DPIF_UC_ACTION))); |
abe529af BP |
502 | if (error) { |
503 | VLOG_ERR("failed to listen on datapath %s: %s", name, strerror(error)); | |
504 | dpif_close(ofproto->dpif); | |
505 | return error; | |
506 | } | |
abe529af BP |
507 | |
508 | ofproto->netflow = NULL; | |
509 | ofproto->sflow = NULL; | |
21f7563c | 510 | ofproto->stp = NULL; |
abe529af BP |
511 | hmap_init(&ofproto->bundles); |
512 | ofproto->ml = mac_learning_create(); | |
513 | for (i = 0; i < MAX_MIRRORS; i++) { | |
514 | ofproto->mirrors[i] = NULL; | |
515 | } | |
516 | ofproto->has_bonded_bundles = false; | |
517 | ||
518 | timer_set_duration(&ofproto->next_expiration, 1000); | |
519 | ||
520 | hmap_init(&ofproto->facets); | |
54a9cbc9 BP |
521 | |
522 | for (i = 0; i < N_TABLES; i++) { | |
523 | struct table_dpif *table = &ofproto->tables[i]; | |
524 | ||
525 | table->catchall_table = NULL; | |
526 | table->other_table = NULL; | |
527 | table->basis = random_uint32(); | |
528 | } | |
abe529af BP |
529 | ofproto->need_revalidate = false; |
530 | tag_set_init(&ofproto->revalidate_set); | |
531 | ||
7ee20df1 BP |
532 | list_init(&ofproto->completions); |
533 | ||
abe529af BP |
534 | ofproto_dpif_unixctl_init(); |
535 | ||
daff3353 EJ |
536 | ofproto->has_bundle_action = false; |
537 | ||
9cdaaebe | 538 | *n_tablesp = N_TABLES; |
abe529af BP |
539 | return 0; |
540 | } | |
541 | ||
7ee20df1 BP |
542 | static void |
543 | complete_operations(struct ofproto_dpif *ofproto) | |
544 | { | |
545 | struct dpif_completion *c, *next; | |
546 | ||
547 | LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) { | |
548 | ofoperation_complete(c->op, 0); | |
549 | list_remove(&c->list_node); | |
550 | free(c); | |
551 | } | |
552 | } | |
553 | ||
abe529af BP |
554 | static void |
555 | destruct(struct ofproto *ofproto_) | |
556 | { | |
557 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
7ee20df1 | 558 | struct rule_dpif *rule, *next_rule; |
0697b5c3 | 559 | struct classifier *table; |
abe529af BP |
560 | int i; |
561 | ||
7ee20df1 BP |
562 | complete_operations(ofproto); |
563 | ||
0697b5c3 BP |
564 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
565 | struct cls_cursor cursor; | |
566 | ||
567 | cls_cursor_init(&cursor, table, NULL); | |
568 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
569 | ofproto_rule_destroy(&rule->up); | |
570 | } | |
7ee20df1 BP |
571 | } |
572 | ||
abe529af BP |
573 | for (i = 0; i < MAX_MIRRORS; i++) { |
574 | mirror_destroy(ofproto->mirrors[i]); | |
575 | } | |
576 | ||
577 | netflow_destroy(ofproto->netflow); | |
bae473fe | 578 | dpif_sflow_destroy(ofproto->sflow); |
abe529af BP |
579 | hmap_destroy(&ofproto->bundles); |
580 | mac_learning_destroy(ofproto->ml); | |
581 | ||
582 | hmap_destroy(&ofproto->facets); | |
583 | ||
584 | dpif_close(ofproto->dpif); | |
585 | } | |
586 | ||
587 | static int | |
588 | run(struct ofproto *ofproto_) | |
589 | { | |
590 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
501f8d1f | 591 | struct dpif_upcall misses[FLOW_MISS_MAX_BATCH]; |
abe529af BP |
592 | struct ofport_dpif *ofport; |
593 | struct ofbundle *bundle; | |
501f8d1f | 594 | size_t n_misses; |
abe529af BP |
595 | int i; |
596 | ||
7ee20df1 BP |
597 | if (!clogged) { |
598 | complete_operations(ofproto); | |
599 | } | |
abe529af BP |
600 | dpif_run(ofproto->dpif); |
601 | ||
501f8d1f BP |
602 | n_misses = 0; |
603 | for (i = 0; i < FLOW_MISS_MAX_BATCH; i++) { | |
604 | struct dpif_upcall *upcall = &misses[n_misses]; | |
abe529af BP |
605 | int error; |
606 | ||
501f8d1f | 607 | error = dpif_recv(ofproto->dpif, upcall); |
abe529af | 608 | if (error) { |
501f8d1f | 609 | if (error == ENODEV && n_misses == 0) { |
abe529af BP |
610 | return error; |
611 | } | |
612 | break; | |
613 | } | |
614 | ||
501f8d1f BP |
615 | if (upcall->type == DPIF_UC_MISS) { |
616 | /* Handle it later. */ | |
617 | n_misses++; | |
618 | } else { | |
619 | handle_upcall(ofproto, upcall); | |
620 | } | |
abe529af BP |
621 | } |
622 | ||
501f8d1f BP |
623 | handle_miss_upcalls(ofproto, misses, n_misses); |
624 | ||
abe529af BP |
625 | if (timer_expired(&ofproto->next_expiration)) { |
626 | int delay = expire(ofproto); | |
627 | timer_set_duration(&ofproto->next_expiration, delay); | |
628 | } | |
629 | ||
630 | if (ofproto->netflow) { | |
631 | netflow_run(ofproto->netflow); | |
632 | } | |
633 | if (ofproto->sflow) { | |
bae473fe | 634 | dpif_sflow_run(ofproto->sflow); |
abe529af BP |
635 | } |
636 | ||
637 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
638 | port_run(ofport); | |
639 | } | |
640 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
641 | bundle_run(bundle); | |
642 | } | |
643 | ||
21f7563c | 644 | stp_run(ofproto); |
1c313b88 BP |
645 | mac_learning_run(ofproto->ml, &ofproto->revalidate_set); |
646 | ||
abe529af BP |
647 | /* Now revalidate if there's anything to do. */ |
648 | if (ofproto->need_revalidate | |
649 | || !tag_set_is_empty(&ofproto->revalidate_set)) { | |
650 | struct tag_set revalidate_set = ofproto->revalidate_set; | |
651 | bool revalidate_all = ofproto->need_revalidate; | |
652 | struct facet *facet, *next; | |
653 | ||
654 | /* Clear the revalidation flags. */ | |
655 | tag_set_init(&ofproto->revalidate_set); | |
656 | ofproto->need_revalidate = false; | |
657 | ||
658 | HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) { | |
659 | if (revalidate_all | |
660 | || tag_set_intersects(&revalidate_set, facet->tags)) { | |
661 | facet_revalidate(ofproto, facet); | |
662 | } | |
663 | } | |
664 | } | |
665 | ||
666 | return 0; | |
667 | } | |
668 | ||
669 | static void | |
670 | wait(struct ofproto *ofproto_) | |
671 | { | |
672 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
673 | struct ofport_dpif *ofport; | |
674 | struct ofbundle *bundle; | |
675 | ||
7ee20df1 BP |
676 | if (!clogged && !list_is_empty(&ofproto->completions)) { |
677 | poll_immediate_wake(); | |
678 | } | |
679 | ||
abe529af BP |
680 | dpif_wait(ofproto->dpif); |
681 | dpif_recv_wait(ofproto->dpif); | |
682 | if (ofproto->sflow) { | |
bae473fe | 683 | dpif_sflow_wait(ofproto->sflow); |
abe529af BP |
684 | } |
685 | if (!tag_set_is_empty(&ofproto->revalidate_set)) { | |
686 | poll_immediate_wake(); | |
687 | } | |
688 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { | |
689 | port_wait(ofport); | |
690 | } | |
691 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
692 | bundle_wait(bundle); | |
693 | } | |
1c313b88 | 694 | mac_learning_wait(ofproto->ml); |
21f7563c | 695 | stp_wait(ofproto); |
abe529af BP |
696 | if (ofproto->need_revalidate) { |
697 | /* Shouldn't happen, but if it does just go around again. */ | |
698 | VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()"); | |
699 | poll_immediate_wake(); | |
700 | } else { | |
701 | timer_wait(&ofproto->next_expiration); | |
702 | } | |
703 | } | |
704 | ||
705 | static void | |
706 | flush(struct ofproto *ofproto_) | |
707 | { | |
708 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
709 | struct facet *facet, *next_facet; | |
710 | ||
711 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
712 | /* Mark the facet as not installed so that facet_remove() doesn't | |
713 | * bother trying to uninstall it. There is no point in uninstalling it | |
714 | * individually since we are about to blow away all the facets with | |
715 | * dpif_flow_flush(). */ | |
716 | facet->installed = false; | |
717 | facet->dp_packet_count = 0; | |
718 | facet->dp_byte_count = 0; | |
719 | facet_remove(ofproto, facet); | |
720 | } | |
721 | dpif_flow_flush(ofproto->dpif); | |
722 | } | |
723 | ||
6c1491fb BP |
724 | static void |
725 | get_features(struct ofproto *ofproto_ OVS_UNUSED, | |
726 | bool *arp_match_ip, uint32_t *actions) | |
727 | { | |
728 | *arp_match_ip = true; | |
729 | *actions = ((1u << OFPAT_OUTPUT) | | |
730 | (1u << OFPAT_SET_VLAN_VID) | | |
731 | (1u << OFPAT_SET_VLAN_PCP) | | |
732 | (1u << OFPAT_STRIP_VLAN) | | |
733 | (1u << OFPAT_SET_DL_SRC) | | |
734 | (1u << OFPAT_SET_DL_DST) | | |
735 | (1u << OFPAT_SET_NW_SRC) | | |
736 | (1u << OFPAT_SET_NW_DST) | | |
737 | (1u << OFPAT_SET_NW_TOS) | | |
738 | (1u << OFPAT_SET_TP_SRC) | | |
739 | (1u << OFPAT_SET_TP_DST) | | |
740 | (1u << OFPAT_ENQUEUE)); | |
741 | } | |
742 | ||
743 | static void | |
744 | get_tables(struct ofproto *ofproto_, struct ofp_table_stats *ots) | |
745 | { | |
746 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
a8d9304d | 747 | struct dpif_dp_stats s; |
6c1491fb BP |
748 | |
749 | strcpy(ots->name, "classifier"); | |
750 | ||
751 | dpif_get_dp_stats(ofproto->dpif, &s); | |
752 | put_32aligned_be64(&ots->lookup_count, htonll(s.n_hit + s.n_missed)); | |
753 | put_32aligned_be64(&ots->matched_count, | |
754 | htonll(s.n_hit + ofproto->n_matches)); | |
755 | } | |
756 | ||
abe529af BP |
757 | static int |
758 | set_netflow(struct ofproto *ofproto_, | |
759 | const struct netflow_options *netflow_options) | |
760 | { | |
761 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
762 | ||
763 | if (netflow_options) { | |
764 | if (!ofproto->netflow) { | |
765 | ofproto->netflow = netflow_create(); | |
766 | } | |
767 | return netflow_set_options(ofproto->netflow, netflow_options); | |
768 | } else { | |
769 | netflow_destroy(ofproto->netflow); | |
770 | ofproto->netflow = NULL; | |
771 | return 0; | |
772 | } | |
773 | } | |
774 | ||
775 | static struct ofport * | |
776 | port_alloc(void) | |
777 | { | |
778 | struct ofport_dpif *port = xmalloc(sizeof *port); | |
779 | return &port->up; | |
780 | } | |
781 | ||
782 | static void | |
783 | port_dealloc(struct ofport *port_) | |
784 | { | |
785 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
786 | free(port); | |
787 | } | |
788 | ||
789 | static int | |
790 | port_construct(struct ofport *port_) | |
791 | { | |
792 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
793 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
794 | ||
f11c28c4 | 795 | ofproto->need_revalidate = true; |
abe529af BP |
796 | port->odp_port = ofp_port_to_odp_port(port->up.ofp_port); |
797 | port->bundle = NULL; | |
798 | port->cfm = NULL; | |
799 | port->tag = tag_create_random(); | |
d5ffa7f2 | 800 | port->may_enable = true; |
21f7563c JP |
801 | port->stp_port = NULL; |
802 | port->stp_state = STP_DISABLED; | |
abe529af BP |
803 | |
804 | if (ofproto->sflow) { | |
bae473fe JP |
805 | dpif_sflow_add_port(ofproto->sflow, port->odp_port, |
806 | netdev_get_name(port->up.netdev)); | |
abe529af BP |
807 | } |
808 | ||
809 | return 0; | |
810 | } | |
811 | ||
812 | static void | |
813 | port_destruct(struct ofport *port_) | |
814 | { | |
815 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
816 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
817 | ||
f11c28c4 | 818 | ofproto->need_revalidate = true; |
abe529af | 819 | bundle_remove(port_); |
a5610457 | 820 | set_cfm(port_, NULL); |
abe529af | 821 | if (ofproto->sflow) { |
bae473fe | 822 | dpif_sflow_del_port(ofproto->sflow, port->odp_port); |
abe529af BP |
823 | } |
824 | } | |
825 | ||
826 | static void | |
827 | port_modified(struct ofport *port_) | |
828 | { | |
829 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
830 | ||
831 | if (port->bundle && port->bundle->bond) { | |
832 | bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev); | |
833 | } | |
834 | } | |
835 | ||
836 | static void | |
837 | port_reconfigured(struct ofport *port_, ovs_be32 old_config) | |
838 | { | |
839 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
840 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto); | |
841 | ovs_be32 changed = old_config ^ port->up.opp.config; | |
842 | ||
843 | if (changed & htonl(OFPPC_NO_RECV | OFPPC_NO_RECV_STP | | |
844 | OFPPC_NO_FWD | OFPPC_NO_FLOOD)) { | |
845 | ofproto->need_revalidate = true; | |
7bde8dd8 JP |
846 | |
847 | if (changed & htonl(OFPPC_NO_FLOOD) && port->bundle) { | |
848 | bundle_update(port->bundle); | |
849 | } | |
abe529af BP |
850 | } |
851 | } | |
852 | ||
853 | static int | |
854 | set_sflow(struct ofproto *ofproto_, | |
855 | const struct ofproto_sflow_options *sflow_options) | |
856 | { | |
857 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
bae473fe | 858 | struct dpif_sflow *ds = ofproto->sflow; |
6ff686f2 | 859 | |
abe529af | 860 | if (sflow_options) { |
bae473fe | 861 | if (!ds) { |
abe529af BP |
862 | struct ofport_dpif *ofport; |
863 | ||
bae473fe | 864 | ds = ofproto->sflow = dpif_sflow_create(ofproto->dpif); |
abe529af | 865 | HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) { |
bae473fe JP |
866 | dpif_sflow_add_port(ds, ofport->odp_port, |
867 | netdev_get_name(ofport->up.netdev)); | |
abe529af | 868 | } |
6ff686f2 | 869 | ofproto->need_revalidate = true; |
abe529af | 870 | } |
bae473fe | 871 | dpif_sflow_set_options(ds, sflow_options); |
abe529af | 872 | } else { |
6ff686f2 PS |
873 | if (ds) { |
874 | dpif_sflow_destroy(ds); | |
875 | ofproto->need_revalidate = true; | |
876 | ofproto->sflow = NULL; | |
877 | } | |
abe529af BP |
878 | } |
879 | return 0; | |
880 | } | |
881 | ||
882 | static int | |
a5610457 | 883 | set_cfm(struct ofport *ofport_, const struct cfm_settings *s) |
abe529af BP |
884 | { |
885 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
886 | int error; | |
887 | ||
a5610457 | 888 | if (!s) { |
abe529af BP |
889 | error = 0; |
890 | } else { | |
891 | if (!ofport->cfm) { | |
8c977421 EJ |
892 | struct ofproto_dpif *ofproto; |
893 | ||
894 | ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
895 | ofproto->need_revalidate = true; | |
6f629657 | 896 | ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev)); |
abe529af BP |
897 | } |
898 | ||
a5610457 | 899 | if (cfm_configure(ofport->cfm, s)) { |
abe529af BP |
900 | return 0; |
901 | } | |
902 | ||
903 | error = EINVAL; | |
904 | } | |
905 | cfm_destroy(ofport->cfm); | |
906 | ofport->cfm = NULL; | |
907 | return error; | |
908 | } | |
909 | ||
910 | static int | |
a5610457 | 911 | get_cfm_fault(const struct ofport *ofport_) |
abe529af BP |
912 | { |
913 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
a5610457 EJ |
914 | |
915 | return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1; | |
abe529af | 916 | } |
1de11730 EJ |
917 | |
918 | static int | |
919 | get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps, | |
920 | size_t *n_rmps) | |
921 | { | |
922 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
923 | ||
924 | if (ofport->cfm) { | |
925 | cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps); | |
926 | return 0; | |
927 | } else { | |
928 | return -1; | |
929 | } | |
930 | } | |
abe529af | 931 | \f |
21f7563c JP |
932 | /* Spanning Tree. */ |
933 | ||
934 | static void | |
935 | send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_) | |
936 | { | |
937 | struct ofproto_dpif *ofproto = ofproto_; | |
938 | struct stp_port *sp = stp_get_port(ofproto->stp, port_num); | |
939 | struct ofport_dpif *ofport; | |
940 | ||
941 | ofport = stp_port_get_aux(sp); | |
942 | if (!ofport) { | |
943 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d", | |
944 | ofproto->up.name, port_num); | |
945 | } else { | |
946 | struct eth_header *eth = pkt->l2; | |
947 | ||
948 | netdev_get_etheraddr(ofport->up.netdev, eth->eth_src); | |
949 | if (eth_addr_is_zero(eth->eth_src)) { | |
950 | VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d " | |
951 | "with unknown MAC", ofproto->up.name, port_num); | |
952 | } else { | |
ea131871 JG |
953 | send_packet(ofproto_dpif_cast(ofport->up.ofproto), |
954 | ofport->odp_port, pkt); | |
21f7563c JP |
955 | } |
956 | } | |
957 | ofpbuf_delete(pkt); | |
958 | } | |
959 | ||
960 | /* Configures STP on 'ofproto_' using the settings defined in 's'. */ | |
961 | static int | |
962 | set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s) | |
963 | { | |
964 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
965 | ||
966 | /* Only revalidate flows if the configuration changed. */ | |
967 | if (!s != !ofproto->stp) { | |
968 | ofproto->need_revalidate = true; | |
969 | } | |
970 | ||
971 | if (s) { | |
972 | if (!ofproto->stp) { | |
973 | ofproto->stp = stp_create(ofproto_->name, s->system_id, | |
974 | send_bpdu_cb, ofproto); | |
975 | ofproto->stp_last_tick = time_msec(); | |
976 | } | |
977 | ||
978 | stp_set_bridge_id(ofproto->stp, s->system_id); | |
979 | stp_set_bridge_priority(ofproto->stp, s->priority); | |
980 | stp_set_hello_time(ofproto->stp, s->hello_time); | |
981 | stp_set_max_age(ofproto->stp, s->max_age); | |
982 | stp_set_forward_delay(ofproto->stp, s->fwd_delay); | |
983 | } else { | |
984 | stp_destroy(ofproto->stp); | |
985 | ofproto->stp = NULL; | |
986 | } | |
987 | ||
988 | return 0; | |
989 | } | |
990 | ||
991 | static int | |
992 | get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s) | |
993 | { | |
994 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
995 | ||
996 | if (ofproto->stp) { | |
997 | s->enabled = true; | |
998 | s->bridge_id = stp_get_bridge_id(ofproto->stp); | |
999 | s->designated_root = stp_get_designated_root(ofproto->stp); | |
1000 | s->root_path_cost = stp_get_root_path_cost(ofproto->stp); | |
1001 | } else { | |
1002 | s->enabled = false; | |
1003 | } | |
1004 | ||
1005 | return 0; | |
1006 | } | |
1007 | ||
1008 | static void | |
1009 | update_stp_port_state(struct ofport_dpif *ofport) | |
1010 | { | |
1011 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1012 | enum stp_state state; | |
1013 | ||
1014 | /* Figure out new state. */ | |
1015 | state = ofport->stp_port ? stp_port_get_state(ofport->stp_port) | |
1016 | : STP_DISABLED; | |
1017 | ||
1018 | /* Update state. */ | |
1019 | if (ofport->stp_state != state) { | |
1020 | ovs_be32 of_state; | |
1021 | bool fwd_change; | |
1022 | ||
1023 | VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s", | |
1024 | netdev_get_name(ofport->up.netdev), | |
1025 | stp_state_name(ofport->stp_state), | |
1026 | stp_state_name(state)); | |
1027 | if (stp_learn_in_state(ofport->stp_state) | |
1028 | != stp_learn_in_state(state)) { | |
1029 | /* xxx Learning action flows should also be flushed. */ | |
1030 | mac_learning_flush(ofproto->ml); | |
1031 | } | |
1032 | fwd_change = stp_forward_in_state(ofport->stp_state) | |
1033 | != stp_forward_in_state(state); | |
1034 | ||
1035 | ofproto->need_revalidate = true; | |
1036 | ofport->stp_state = state; | |
1037 | ofport->stp_state_entered = time_msec(); | |
1038 | ||
b308140a | 1039 | if (fwd_change && ofport->bundle) { |
21f7563c JP |
1040 | bundle_update(ofport->bundle); |
1041 | } | |
1042 | ||
1043 | /* Update the STP state bits in the OpenFlow port description. */ | |
1044 | of_state = (ofport->up.opp.state & htonl(~OFPPS_STP_MASK)) | |
1045 | | htonl(state == STP_LISTENING ? OFPPS_STP_LISTEN | |
1046 | : state == STP_LEARNING ? OFPPS_STP_LEARN | |
1047 | : state == STP_FORWARDING ? OFPPS_STP_FORWARD | |
1048 | : state == STP_BLOCKING ? OFPPS_STP_BLOCK | |
1049 | : 0); | |
1050 | ofproto_port_set_state(&ofport->up, of_state); | |
1051 | } | |
1052 | } | |
1053 | ||
1054 | /* Configures STP on 'ofport_' using the settings defined in 's'. The | |
1055 | * caller is responsible for assigning STP port numbers and ensuring | |
1056 | * there are no duplicates. */ | |
1057 | static int | |
1058 | set_stp_port(struct ofport *ofport_, | |
1059 | const struct ofproto_port_stp_settings *s) | |
1060 | { | |
1061 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1062 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1063 | struct stp_port *sp = ofport->stp_port; | |
1064 | ||
1065 | if (!s || !s->enable) { | |
1066 | if (sp) { | |
1067 | ofport->stp_port = NULL; | |
1068 | stp_port_disable(sp); | |
ecd12731 | 1069 | update_stp_port_state(ofport); |
21f7563c JP |
1070 | } |
1071 | return 0; | |
1072 | } else if (sp && stp_port_no(sp) != s->port_num | |
1073 | && ofport == stp_port_get_aux(sp)) { | |
1074 | /* The port-id changed, so disable the old one if it's not | |
1075 | * already in use by another port. */ | |
1076 | stp_port_disable(sp); | |
1077 | } | |
1078 | ||
1079 | sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num); | |
1080 | stp_port_enable(sp); | |
1081 | ||
1082 | stp_port_set_aux(sp, ofport); | |
1083 | stp_port_set_priority(sp, s->priority); | |
1084 | stp_port_set_path_cost(sp, s->path_cost); | |
1085 | ||
1086 | update_stp_port_state(ofport); | |
1087 | ||
1088 | return 0; | |
1089 | } | |
1090 | ||
1091 | static int | |
1092 | get_stp_port_status(struct ofport *ofport_, | |
1093 | struct ofproto_port_stp_status *s) | |
1094 | { | |
1095 | struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
1096 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1097 | struct stp_port *sp = ofport->stp_port; | |
1098 | ||
1099 | if (!ofproto->stp || !sp) { | |
1100 | s->enabled = false; | |
1101 | return 0; | |
1102 | } | |
1103 | ||
1104 | s->enabled = true; | |
1105 | s->port_id = stp_port_get_id(sp); | |
1106 | s->state = stp_port_get_state(sp); | |
1107 | s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000; | |
1108 | s->role = stp_port_get_role(sp); | |
80740385 | 1109 | stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count); |
21f7563c JP |
1110 | |
1111 | return 0; | |
1112 | } | |
1113 | ||
1114 | static void | |
1115 | stp_run(struct ofproto_dpif *ofproto) | |
1116 | { | |
1117 | if (ofproto->stp) { | |
1118 | long long int now = time_msec(); | |
1119 | long long int elapsed = now - ofproto->stp_last_tick; | |
1120 | struct stp_port *sp; | |
1121 | ||
1122 | if (elapsed > 0) { | |
1123 | stp_tick(ofproto->stp, MIN(INT_MAX, elapsed)); | |
1124 | ofproto->stp_last_tick = now; | |
1125 | } | |
1126 | while (stp_get_changed_port(ofproto->stp, &sp)) { | |
1127 | struct ofport_dpif *ofport = stp_port_get_aux(sp); | |
1128 | ||
1129 | if (ofport) { | |
1130 | update_stp_port_state(ofport); | |
1131 | } | |
1132 | } | |
1133 | } | |
1134 | } | |
1135 | ||
1136 | static void | |
1137 | stp_wait(struct ofproto_dpif *ofproto) | |
1138 | { | |
1139 | if (ofproto->stp) { | |
1140 | poll_timer_wait(1000); | |
1141 | } | |
1142 | } | |
1143 | ||
1144 | /* Returns true if STP should process 'flow'. */ | |
1145 | static bool | |
1146 | stp_should_process_flow(const struct flow *flow) | |
1147 | { | |
1148 | return eth_addr_equals(flow->dl_dst, eth_addr_stp); | |
1149 | } | |
1150 | ||
1151 | static void | |
1152 | stp_process_packet(const struct ofport_dpif *ofport, | |
1153 | const struct ofpbuf *packet) | |
1154 | { | |
1155 | struct ofpbuf payload = *packet; | |
1156 | struct eth_header *eth = payload.data; | |
1157 | struct stp_port *sp = ofport->stp_port; | |
1158 | ||
1159 | /* Sink packets on ports that have STP disabled when the bridge has | |
1160 | * STP enabled. */ | |
1161 | if (!sp || stp_port_get_state(sp) == STP_DISABLED) { | |
1162 | return; | |
1163 | } | |
1164 | ||
1165 | /* Trim off padding on payload. */ | |
c573540b BP |
1166 | if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) { |
1167 | payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN; | |
21f7563c JP |
1168 | } |
1169 | ||
1170 | if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) { | |
1171 | stp_received_bpdu(sp, payload.data, payload.size); | |
1172 | } | |
1173 | } | |
1174 | \f | |
abe529af BP |
1175 | /* Bundles. */ |
1176 | ||
1177 | /* Expires all MAC learning entries associated with 'port' and forces ofproto | |
1178 | * to revalidate every flow. */ | |
1179 | static void | |
1180 | bundle_flush_macs(struct ofbundle *bundle) | |
1181 | { | |
1182 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1183 | struct mac_learning *ml = ofproto->ml; | |
1184 | struct mac_entry *mac, *next_mac; | |
1185 | ||
1186 | ofproto->need_revalidate = true; | |
1187 | LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) { | |
1188 | if (mac->port.p == bundle) { | |
1189 | mac_learning_expire(ml, mac); | |
1190 | } | |
1191 | } | |
1192 | } | |
1193 | ||
1194 | static struct ofbundle * | |
1195 | bundle_lookup(const struct ofproto_dpif *ofproto, void *aux) | |
1196 | { | |
1197 | struct ofbundle *bundle; | |
1198 | ||
1199 | HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0), | |
1200 | &ofproto->bundles) { | |
1201 | if (bundle->aux == aux) { | |
1202 | return bundle; | |
1203 | } | |
1204 | } | |
1205 | return NULL; | |
1206 | } | |
1207 | ||
1208 | /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the | |
1209 | * ones that are found to 'bundles'. */ | |
1210 | static void | |
1211 | bundle_lookup_multiple(struct ofproto_dpif *ofproto, | |
1212 | void **auxes, size_t n_auxes, | |
1213 | struct hmapx *bundles) | |
1214 | { | |
1215 | size_t i; | |
1216 | ||
1217 | hmapx_init(bundles); | |
1218 | for (i = 0; i < n_auxes; i++) { | |
1219 | struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]); | |
1220 | if (bundle) { | |
1221 | hmapx_add(bundles, bundle); | |
1222 | } | |
1223 | } | |
1224 | } | |
1225 | ||
7bde8dd8 JP |
1226 | static void |
1227 | bundle_update(struct ofbundle *bundle) | |
1228 | { | |
1229 | struct ofport_dpif *port; | |
1230 | ||
1231 | bundle->floodable = true; | |
1232 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
21f7563c JP |
1233 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD) |
1234 | || !stp_forward_in_state(port->stp_state)) { | |
7bde8dd8 JP |
1235 | bundle->floodable = false; |
1236 | break; | |
1237 | } | |
1238 | } | |
1239 | } | |
1240 | ||
abe529af BP |
1241 | static void |
1242 | bundle_del_port(struct ofport_dpif *port) | |
1243 | { | |
1244 | struct ofbundle *bundle = port->bundle; | |
1245 | ||
6f77f4ae BP |
1246 | bundle->ofproto->need_revalidate = true; |
1247 | ||
abe529af BP |
1248 | list_remove(&port->bundle_node); |
1249 | port->bundle = NULL; | |
1250 | ||
1251 | if (bundle->lacp) { | |
1252 | lacp_slave_unregister(bundle->lacp, port); | |
1253 | } | |
1254 | if (bundle->bond) { | |
1255 | bond_slave_unregister(bundle->bond, port); | |
1256 | } | |
1257 | ||
7bde8dd8 | 1258 | bundle_update(bundle); |
abe529af BP |
1259 | } |
1260 | ||
1261 | static bool | |
1262 | bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port, | |
00794817 BP |
1263 | struct lacp_slave_settings *lacp, |
1264 | uint32_t bond_stable_id) | |
abe529af BP |
1265 | { |
1266 | struct ofport_dpif *port; | |
1267 | ||
1268 | port = get_ofp_port(bundle->ofproto, ofp_port); | |
1269 | if (!port) { | |
1270 | return false; | |
1271 | } | |
1272 | ||
1273 | if (port->bundle != bundle) { | |
6f77f4ae | 1274 | bundle->ofproto->need_revalidate = true; |
abe529af BP |
1275 | if (port->bundle) { |
1276 | bundle_del_port(port); | |
1277 | } | |
1278 | ||
1279 | port->bundle = bundle; | |
1280 | list_push_back(&bundle->ports, &port->bundle_node); | |
21f7563c JP |
1281 | if (port->up.opp.config & htonl(OFPPC_NO_FLOOD) |
1282 | || !stp_forward_in_state(port->stp_state)) { | |
abe529af BP |
1283 | bundle->floodable = false; |
1284 | } | |
1285 | } | |
1286 | if (lacp) { | |
4a86aece | 1287 | port->bundle->ofproto->need_revalidate = true; |
abe529af BP |
1288 | lacp_slave_register(bundle->lacp, port, lacp); |
1289 | } | |
1290 | ||
00794817 BP |
1291 | port->bond_stable_id = bond_stable_id; |
1292 | ||
abe529af BP |
1293 | return true; |
1294 | } | |
1295 | ||
1296 | static void | |
1297 | bundle_destroy(struct ofbundle *bundle) | |
1298 | { | |
1299 | struct ofproto_dpif *ofproto; | |
1300 | struct ofport_dpif *port, *next_port; | |
1301 | int i; | |
1302 | ||
1303 | if (!bundle) { | |
1304 | return; | |
1305 | } | |
1306 | ||
1307 | ofproto = bundle->ofproto; | |
1308 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1309 | struct ofmirror *m = ofproto->mirrors[i]; | |
1310 | if (m) { | |
1311 | if (m->out == bundle) { | |
1312 | mirror_destroy(m); | |
1313 | } else if (hmapx_find_and_delete(&m->srcs, bundle) | |
1314 | || hmapx_find_and_delete(&m->dsts, bundle)) { | |
1315 | ofproto->need_revalidate = true; | |
1316 | } | |
1317 | } | |
1318 | } | |
1319 | ||
1320 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1321 | bundle_del_port(port); | |
1322 | } | |
1323 | ||
1324 | bundle_flush_macs(bundle); | |
1325 | hmap_remove(&ofproto->bundles, &bundle->hmap_node); | |
1326 | free(bundle->name); | |
1327 | free(bundle->trunks); | |
1328 | lacp_destroy(bundle->lacp); | |
1329 | bond_destroy(bundle->bond); | |
1330 | free(bundle); | |
1331 | } | |
1332 | ||
1333 | static int | |
1334 | bundle_set(struct ofproto *ofproto_, void *aux, | |
1335 | const struct ofproto_bundle_settings *s) | |
1336 | { | |
1337 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1338 | bool need_flush = false; | |
abe529af BP |
1339 | struct ofport_dpif *port; |
1340 | struct ofbundle *bundle; | |
ecac4ebf BP |
1341 | unsigned long *trunks; |
1342 | int vlan; | |
abe529af BP |
1343 | size_t i; |
1344 | bool ok; | |
1345 | ||
1346 | if (!s) { | |
1347 | bundle_destroy(bundle_lookup(ofproto, aux)); | |
1348 | return 0; | |
1349 | } | |
1350 | ||
1351 | assert(s->n_slaves == 1 || s->bond != NULL); | |
1352 | assert((s->lacp != NULL) == (s->lacp_slaves != NULL)); | |
1353 | ||
1354 | bundle = bundle_lookup(ofproto, aux); | |
1355 | if (!bundle) { | |
1356 | bundle = xmalloc(sizeof *bundle); | |
1357 | ||
1358 | bundle->ofproto = ofproto; | |
1359 | hmap_insert(&ofproto->bundles, &bundle->hmap_node, | |
1360 | hash_pointer(aux, 0)); | |
1361 | bundle->aux = aux; | |
1362 | bundle->name = NULL; | |
1363 | ||
1364 | list_init(&bundle->ports); | |
ecac4ebf | 1365 | bundle->vlan_mode = PORT_VLAN_TRUNK; |
abe529af BP |
1366 | bundle->vlan = -1; |
1367 | bundle->trunks = NULL; | |
5e9ceccd | 1368 | bundle->use_priority_tags = s->use_priority_tags; |
abe529af BP |
1369 | bundle->lacp = NULL; |
1370 | bundle->bond = NULL; | |
1371 | ||
1372 | bundle->floodable = true; | |
1373 | ||
1374 | bundle->src_mirrors = 0; | |
1375 | bundle->dst_mirrors = 0; | |
1376 | bundle->mirror_out = 0; | |
1377 | } | |
1378 | ||
1379 | if (!bundle->name || strcmp(s->name, bundle->name)) { | |
1380 | free(bundle->name); | |
1381 | bundle->name = xstrdup(s->name); | |
1382 | } | |
1383 | ||
1384 | /* LACP. */ | |
1385 | if (s->lacp) { | |
1386 | if (!bundle->lacp) { | |
8c977421 | 1387 | ofproto->need_revalidate = true; |
abe529af BP |
1388 | bundle->lacp = lacp_create(); |
1389 | } | |
1390 | lacp_configure(bundle->lacp, s->lacp); | |
1391 | } else { | |
1392 | lacp_destroy(bundle->lacp); | |
1393 | bundle->lacp = NULL; | |
1394 | } | |
1395 | ||
1396 | /* Update set of ports. */ | |
1397 | ok = true; | |
1398 | for (i = 0; i < s->n_slaves; i++) { | |
1399 | if (!bundle_add_port(bundle, s->slaves[i], | |
00794817 BP |
1400 | s->lacp ? &s->lacp_slaves[i] : NULL, |
1401 | s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) { | |
abe529af BP |
1402 | ok = false; |
1403 | } | |
1404 | } | |
1405 | if (!ok || list_size(&bundle->ports) != s->n_slaves) { | |
1406 | struct ofport_dpif *next_port; | |
1407 | ||
1408 | LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) { | |
1409 | for (i = 0; i < s->n_slaves; i++) { | |
56c769ab | 1410 | if (s->slaves[i] == port->up.ofp_port) { |
abe529af BP |
1411 | goto found; |
1412 | } | |
1413 | } | |
1414 | ||
1415 | bundle_del_port(port); | |
1416 | found: ; | |
1417 | } | |
1418 | } | |
1419 | assert(list_size(&bundle->ports) <= s->n_slaves); | |
1420 | ||
1421 | if (list_is_empty(&bundle->ports)) { | |
1422 | bundle_destroy(bundle); | |
1423 | return EINVAL; | |
1424 | } | |
1425 | ||
ecac4ebf | 1426 | /* Set VLAN tagging mode */ |
5e9ceccd BP |
1427 | if (s->vlan_mode != bundle->vlan_mode |
1428 | || s->use_priority_tags != bundle->use_priority_tags) { | |
ecac4ebf | 1429 | bundle->vlan_mode = s->vlan_mode; |
5e9ceccd | 1430 | bundle->use_priority_tags = s->use_priority_tags; |
ecac4ebf BP |
1431 | need_flush = true; |
1432 | } | |
1433 | ||
abe529af | 1434 | /* Set VLAN tag. */ |
ecac4ebf BP |
1435 | vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1 |
1436 | : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan | |
1437 | : 0); | |
1438 | if (vlan != bundle->vlan) { | |
1439 | bundle->vlan = vlan; | |
abe529af BP |
1440 | need_flush = true; |
1441 | } | |
1442 | ||
1443 | /* Get trunked VLANs. */ | |
ecac4ebf BP |
1444 | switch (s->vlan_mode) { |
1445 | case PORT_VLAN_ACCESS: | |
1446 | trunks = NULL; | |
1447 | break; | |
1448 | ||
1449 | case PORT_VLAN_TRUNK: | |
1450 | trunks = (unsigned long *) s->trunks; | |
1451 | break; | |
1452 | ||
1453 | case PORT_VLAN_NATIVE_UNTAGGED: | |
1454 | case PORT_VLAN_NATIVE_TAGGED: | |
1455 | if (vlan != 0 && (!s->trunks | |
1456 | || !bitmap_is_set(s->trunks, vlan) | |
1457 | || bitmap_is_set(s->trunks, 0))) { | |
1458 | /* Force trunking the native VLAN and prohibit trunking VLAN 0. */ | |
1459 | if (s->trunks) { | |
1460 | trunks = bitmap_clone(s->trunks, 4096); | |
1461 | } else { | |
1462 | trunks = bitmap_allocate1(4096); | |
1463 | } | |
1464 | bitmap_set1(trunks, vlan); | |
1465 | bitmap_set0(trunks, 0); | |
1466 | } else { | |
1467 | trunks = (unsigned long *) s->trunks; | |
1468 | } | |
1469 | break; | |
1470 | ||
1471 | default: | |
1472 | NOT_REACHED(); | |
1473 | } | |
abe529af BP |
1474 | if (!vlan_bitmap_equal(trunks, bundle->trunks)) { |
1475 | free(bundle->trunks); | |
ecac4ebf BP |
1476 | if (trunks == s->trunks) { |
1477 | bundle->trunks = vlan_bitmap_clone(trunks); | |
1478 | } else { | |
1479 | bundle->trunks = trunks; | |
1480 | trunks = NULL; | |
1481 | } | |
abe529af BP |
1482 | need_flush = true; |
1483 | } | |
ecac4ebf BP |
1484 | if (trunks != s->trunks) { |
1485 | free(trunks); | |
1486 | } | |
abe529af BP |
1487 | |
1488 | /* Bonding. */ | |
1489 | if (!list_is_short(&bundle->ports)) { | |
1490 | bundle->ofproto->has_bonded_bundles = true; | |
1491 | if (bundle->bond) { | |
1492 | if (bond_reconfigure(bundle->bond, s->bond)) { | |
1493 | ofproto->need_revalidate = true; | |
1494 | } | |
1495 | } else { | |
1496 | bundle->bond = bond_create(s->bond); | |
6f77f4ae | 1497 | ofproto->need_revalidate = true; |
abe529af BP |
1498 | } |
1499 | ||
1500 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
00794817 | 1501 | bond_slave_register(bundle->bond, port, port->bond_stable_id, |
abe529af BP |
1502 | port->up.netdev); |
1503 | } | |
1504 | } else { | |
1505 | bond_destroy(bundle->bond); | |
1506 | bundle->bond = NULL; | |
1507 | } | |
1508 | ||
1509 | /* If we changed something that would affect MAC learning, un-learn | |
1510 | * everything on this port and force flow revalidation. */ | |
1511 | if (need_flush) { | |
1512 | bundle_flush_macs(bundle); | |
1513 | } | |
1514 | ||
1515 | return 0; | |
1516 | } | |
1517 | ||
1518 | static void | |
1519 | bundle_remove(struct ofport *port_) | |
1520 | { | |
1521 | struct ofport_dpif *port = ofport_dpif_cast(port_); | |
1522 | struct ofbundle *bundle = port->bundle; | |
1523 | ||
1524 | if (bundle) { | |
1525 | bundle_del_port(port); | |
1526 | if (list_is_empty(&bundle->ports)) { | |
1527 | bundle_destroy(bundle); | |
1528 | } else if (list_is_short(&bundle->ports)) { | |
1529 | bond_destroy(bundle->bond); | |
1530 | bundle->bond = NULL; | |
1531 | } | |
1532 | } | |
1533 | } | |
1534 | ||
1535 | static void | |
5f877369 | 1536 | send_pdu_cb(void *port_, const void *pdu, size_t pdu_size) |
abe529af BP |
1537 | { |
1538 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10); | |
1539 | struct ofport_dpif *port = port_; | |
1540 | uint8_t ea[ETH_ADDR_LEN]; | |
1541 | int error; | |
1542 | ||
1543 | error = netdev_get_etheraddr(port->up.netdev, ea); | |
1544 | if (!error) { | |
abe529af | 1545 | struct ofpbuf packet; |
5f877369 | 1546 | void *packet_pdu; |
abe529af BP |
1547 | |
1548 | ofpbuf_init(&packet, 0); | |
1549 | packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP, | |
5f877369 EJ |
1550 | pdu_size); |
1551 | memcpy(packet_pdu, pdu, pdu_size); | |
1552 | ||
ea131871 JG |
1553 | send_packet(ofproto_dpif_cast(port->up.ofproto), port->odp_port, |
1554 | &packet); | |
abe529af BP |
1555 | ofpbuf_uninit(&packet); |
1556 | } else { | |
1557 | VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface " | |
1558 | "%s (%s)", port->bundle->name, | |
1559 | netdev_get_name(port->up.netdev), strerror(error)); | |
1560 | } | |
1561 | } | |
1562 | ||
1563 | static void | |
1564 | bundle_send_learning_packets(struct ofbundle *bundle) | |
1565 | { | |
1566 | struct ofproto_dpif *ofproto = bundle->ofproto; | |
1567 | int error, n_packets, n_errors; | |
1568 | struct mac_entry *e; | |
1569 | ||
1570 | error = n_packets = n_errors = 0; | |
1571 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
1572 | if (e->port.p != bundle) { | |
ea131871 JG |
1573 | struct ofpbuf *learning_packet; |
1574 | struct ofport_dpif *port; | |
1575 | int ret; | |
1576 | ||
1577 | learning_packet = bond_compose_learning_packet(bundle->bond, e->mac, | |
1578 | e->vlan, | |
1579 | (void **)&port); | |
1580 | ret = send_packet(ofproto_dpif_cast(port->up.ofproto), | |
1581 | port->odp_port, learning_packet); | |
1582 | ofpbuf_delete(learning_packet); | |
abe529af BP |
1583 | if (ret) { |
1584 | error = ret; | |
1585 | n_errors++; | |
1586 | } | |
1587 | n_packets++; | |
1588 | } | |
1589 | } | |
1590 | ||
1591 | if (n_errors) { | |
1592 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1593 | VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning " | |
1594 | "packets, last error was: %s", | |
1595 | bundle->name, n_errors, n_packets, strerror(error)); | |
1596 | } else { | |
1597 | VLOG_DBG("bond %s: sent %d gratuitous learning packets", | |
1598 | bundle->name, n_packets); | |
1599 | } | |
1600 | } | |
1601 | ||
1602 | static void | |
1603 | bundle_run(struct ofbundle *bundle) | |
1604 | { | |
1605 | if (bundle->lacp) { | |
1606 | lacp_run(bundle->lacp, send_pdu_cb); | |
1607 | } | |
1608 | if (bundle->bond) { | |
1609 | struct ofport_dpif *port; | |
1610 | ||
1611 | LIST_FOR_EACH (port, bundle_node, &bundle->ports) { | |
015e08bc | 1612 | bond_slave_set_may_enable(bundle->bond, port, port->may_enable); |
abe529af BP |
1613 | } |
1614 | ||
1615 | bond_run(bundle->bond, &bundle->ofproto->revalidate_set, | |
1616 | lacp_negotiated(bundle->lacp)); | |
1617 | if (bond_should_send_learning_packets(bundle->bond)) { | |
1618 | bundle_send_learning_packets(bundle); | |
1619 | } | |
1620 | } | |
1621 | } | |
1622 | ||
1623 | static void | |
1624 | bundle_wait(struct ofbundle *bundle) | |
1625 | { | |
1626 | if (bundle->lacp) { | |
1627 | lacp_wait(bundle->lacp); | |
1628 | } | |
1629 | if (bundle->bond) { | |
1630 | bond_wait(bundle->bond); | |
1631 | } | |
1632 | } | |
1633 | \f | |
1634 | /* Mirrors. */ | |
1635 | ||
1636 | static int | |
1637 | mirror_scan(struct ofproto_dpif *ofproto) | |
1638 | { | |
1639 | int idx; | |
1640 | ||
1641 | for (idx = 0; idx < MAX_MIRRORS; idx++) { | |
1642 | if (!ofproto->mirrors[idx]) { | |
1643 | return idx; | |
1644 | } | |
1645 | } | |
1646 | return -1; | |
1647 | } | |
1648 | ||
1649 | static struct ofmirror * | |
1650 | mirror_lookup(struct ofproto_dpif *ofproto, void *aux) | |
1651 | { | |
1652 | int i; | |
1653 | ||
1654 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1655 | struct ofmirror *mirror = ofproto->mirrors[i]; | |
1656 | if (mirror && mirror->aux == aux) { | |
1657 | return mirror; | |
1658 | } | |
1659 | } | |
1660 | ||
1661 | return NULL; | |
1662 | } | |
1663 | ||
9ba15e2a BP |
1664 | /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */ |
1665 | static void | |
1666 | mirror_update_dups(struct ofproto_dpif *ofproto) | |
1667 | { | |
1668 | int i; | |
1669 | ||
1670 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1671 | struct ofmirror *m = ofproto->mirrors[i]; | |
1672 | ||
1673 | if (m) { | |
1674 | m->dup_mirrors = MIRROR_MASK_C(1) << i; | |
1675 | } | |
1676 | } | |
1677 | ||
1678 | for (i = 0; i < MAX_MIRRORS; i++) { | |
1679 | struct ofmirror *m1 = ofproto->mirrors[i]; | |
1680 | int j; | |
1681 | ||
1682 | if (!m1) { | |
1683 | continue; | |
1684 | } | |
1685 | ||
1686 | for (j = i + 1; j < MAX_MIRRORS; j++) { | |
1687 | struct ofmirror *m2 = ofproto->mirrors[j]; | |
1688 | ||
edb0540b | 1689 | if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) { |
9ba15e2a BP |
1690 | m1->dup_mirrors |= MIRROR_MASK_C(1) << j; |
1691 | m2->dup_mirrors |= m1->dup_mirrors; | |
1692 | } | |
1693 | } | |
1694 | } | |
1695 | } | |
1696 | ||
abe529af BP |
1697 | static int |
1698 | mirror_set(struct ofproto *ofproto_, void *aux, | |
1699 | const struct ofproto_mirror_settings *s) | |
1700 | { | |
1701 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1702 | mirror_mask_t mirror_bit; | |
1703 | struct ofbundle *bundle; | |
1704 | struct ofmirror *mirror; | |
1705 | struct ofbundle *out; | |
1706 | struct hmapx srcs; /* Contains "struct ofbundle *"s. */ | |
1707 | struct hmapx dsts; /* Contains "struct ofbundle *"s. */ | |
1708 | int out_vlan; | |
1709 | ||
1710 | mirror = mirror_lookup(ofproto, aux); | |
1711 | if (!s) { | |
1712 | mirror_destroy(mirror); | |
1713 | return 0; | |
1714 | } | |
1715 | if (!mirror) { | |
1716 | int idx; | |
1717 | ||
1718 | idx = mirror_scan(ofproto); | |
1719 | if (idx < 0) { | |
1720 | VLOG_WARN("bridge %s: maximum of %d port mirrors reached, " | |
1721 | "cannot create %s", | |
1722 | ofproto->up.name, MAX_MIRRORS, s->name); | |
1723 | return EFBIG; | |
1724 | } | |
1725 | ||
1726 | mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror); | |
1727 | mirror->ofproto = ofproto; | |
1728 | mirror->idx = idx; | |
8b28d864 | 1729 | mirror->aux = aux; |
abe529af BP |
1730 | mirror->out_vlan = -1; |
1731 | mirror->name = NULL; | |
1732 | } | |
1733 | ||
1734 | if (!mirror->name || strcmp(s->name, mirror->name)) { | |
1735 | free(mirror->name); | |
1736 | mirror->name = xstrdup(s->name); | |
1737 | } | |
1738 | ||
1739 | /* Get the new configuration. */ | |
1740 | if (s->out_bundle) { | |
1741 | out = bundle_lookup(ofproto, s->out_bundle); | |
1742 | if (!out) { | |
1743 | mirror_destroy(mirror); | |
1744 | return EINVAL; | |
1745 | } | |
1746 | out_vlan = -1; | |
1747 | } else { | |
1748 | out = NULL; | |
1749 | out_vlan = s->out_vlan; | |
1750 | } | |
1751 | bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs); | |
1752 | bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts); | |
1753 | ||
1754 | /* If the configuration has not changed, do nothing. */ | |
1755 | if (hmapx_equals(&srcs, &mirror->srcs) | |
1756 | && hmapx_equals(&dsts, &mirror->dsts) | |
1757 | && vlan_bitmap_equal(mirror->vlans, s->src_vlans) | |
1758 | && mirror->out == out | |
1759 | && mirror->out_vlan == out_vlan) | |
1760 | { | |
1761 | hmapx_destroy(&srcs); | |
1762 | hmapx_destroy(&dsts); | |
1763 | return 0; | |
1764 | } | |
1765 | ||
1766 | hmapx_swap(&srcs, &mirror->srcs); | |
1767 | hmapx_destroy(&srcs); | |
1768 | ||
1769 | hmapx_swap(&dsts, &mirror->dsts); | |
1770 | hmapx_destroy(&dsts); | |
1771 | ||
1772 | free(mirror->vlans); | |
1773 | mirror->vlans = vlan_bitmap_clone(s->src_vlans); | |
1774 | ||
1775 | mirror->out = out; | |
1776 | mirror->out_vlan = out_vlan; | |
1777 | ||
1778 | /* Update bundles. */ | |
1779 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1780 | HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) { | |
1781 | if (hmapx_contains(&mirror->srcs, bundle)) { | |
1782 | bundle->src_mirrors |= mirror_bit; | |
1783 | } else { | |
1784 | bundle->src_mirrors &= ~mirror_bit; | |
1785 | } | |
1786 | ||
1787 | if (hmapx_contains(&mirror->dsts, bundle)) { | |
1788 | bundle->dst_mirrors |= mirror_bit; | |
1789 | } else { | |
1790 | bundle->dst_mirrors &= ~mirror_bit; | |
1791 | } | |
1792 | ||
1793 | if (mirror->out == bundle) { | |
1794 | bundle->mirror_out |= mirror_bit; | |
1795 | } else { | |
1796 | bundle->mirror_out &= ~mirror_bit; | |
1797 | } | |
1798 | } | |
1799 | ||
1800 | ofproto->need_revalidate = true; | |
1801 | mac_learning_flush(ofproto->ml); | |
9ba15e2a | 1802 | mirror_update_dups(ofproto); |
abe529af BP |
1803 | |
1804 | return 0; | |
1805 | } | |
1806 | ||
1807 | static void | |
1808 | mirror_destroy(struct ofmirror *mirror) | |
1809 | { | |
1810 | struct ofproto_dpif *ofproto; | |
1811 | mirror_mask_t mirror_bit; | |
1812 | struct ofbundle *bundle; | |
1813 | ||
1814 | if (!mirror) { | |
1815 | return; | |
1816 | } | |
1817 | ||
1818 | ofproto = mirror->ofproto; | |
1819 | ofproto->need_revalidate = true; | |
1820 | mac_learning_flush(ofproto->ml); | |
1821 | ||
1822 | mirror_bit = MIRROR_MASK_C(1) << mirror->idx; | |
1823 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
1824 | bundle->src_mirrors &= ~mirror_bit; | |
1825 | bundle->dst_mirrors &= ~mirror_bit; | |
1826 | bundle->mirror_out &= ~mirror_bit; | |
1827 | } | |
1828 | ||
1829 | hmapx_destroy(&mirror->srcs); | |
1830 | hmapx_destroy(&mirror->dsts); | |
1831 | free(mirror->vlans); | |
1832 | ||
1833 | ofproto->mirrors[mirror->idx] = NULL; | |
1834 | free(mirror->name); | |
1835 | free(mirror); | |
9ba15e2a BP |
1836 | |
1837 | mirror_update_dups(ofproto); | |
abe529af BP |
1838 | } |
1839 | ||
1840 | static int | |
1841 | set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans) | |
1842 | { | |
1843 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1844 | if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) { | |
1845 | ofproto->need_revalidate = true; | |
1846 | mac_learning_flush(ofproto->ml); | |
1847 | } | |
1848 | return 0; | |
1849 | } | |
1850 | ||
1851 | static bool | |
b4affc74 | 1852 | is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux) |
abe529af BP |
1853 | { |
1854 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1855 | struct ofbundle *bundle = bundle_lookup(ofproto, aux); | |
1856 | return bundle && bundle->mirror_out != 0; | |
1857 | } | |
8402c74b SS |
1858 | |
1859 | static void | |
b53055f4 | 1860 | forward_bpdu_changed(struct ofproto *ofproto_) |
8402c74b SS |
1861 | { |
1862 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1863 | /* Revalidate cached flows whenever forward_bpdu option changes. */ | |
1864 | ofproto->need_revalidate = true; | |
1865 | } | |
abe529af BP |
1866 | \f |
1867 | /* Ports. */ | |
1868 | ||
1869 | static struct ofport_dpif * | |
1870 | get_ofp_port(struct ofproto_dpif *ofproto, uint16_t ofp_port) | |
1871 | { | |
7df6a8bd BP |
1872 | struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port); |
1873 | return ofport ? ofport_dpif_cast(ofport) : NULL; | |
abe529af BP |
1874 | } |
1875 | ||
1876 | static struct ofport_dpif * | |
1877 | get_odp_port(struct ofproto_dpif *ofproto, uint32_t odp_port) | |
1878 | { | |
1879 | return get_ofp_port(ofproto, odp_port_to_ofp_port(odp_port)); | |
1880 | } | |
1881 | ||
1882 | static void | |
1883 | ofproto_port_from_dpif_port(struct ofproto_port *ofproto_port, | |
1884 | struct dpif_port *dpif_port) | |
1885 | { | |
1886 | ofproto_port->name = dpif_port->name; | |
1887 | ofproto_port->type = dpif_port->type; | |
1888 | ofproto_port->ofp_port = odp_port_to_ofp_port(dpif_port->port_no); | |
1889 | } | |
1890 | ||
1891 | static void | |
1892 | port_run(struct ofport_dpif *ofport) | |
1893 | { | |
015e08bc EJ |
1894 | bool enable = netdev_get_carrier(ofport->up.netdev); |
1895 | ||
abe529af BP |
1896 | if (ofport->cfm) { |
1897 | cfm_run(ofport->cfm); | |
1898 | ||
1899 | if (cfm_should_send_ccm(ofport->cfm)) { | |
1900 | struct ofpbuf packet; | |
abe529af BP |
1901 | |
1902 | ofpbuf_init(&packet, 0); | |
c0a2e71d | 1903 | cfm_compose_ccm(ofport->cfm, &packet, ofport->up.opp.hw_addr); |
abe529af | 1904 | send_packet(ofproto_dpif_cast(ofport->up.ofproto), |
b2fda3ef | 1905 | ofport->odp_port, &packet); |
abe529af BP |
1906 | ofpbuf_uninit(&packet); |
1907 | } | |
015e08bc | 1908 | |
86dc6501 EJ |
1909 | enable = enable && !cfm_get_fault(ofport->cfm) |
1910 | && cfm_get_opup(ofport->cfm); | |
abe529af | 1911 | } |
015e08bc EJ |
1912 | |
1913 | if (ofport->bundle) { | |
1914 | enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport); | |
1915 | } | |
1916 | ||
daff3353 EJ |
1917 | if (ofport->may_enable != enable) { |
1918 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto); | |
1919 | ||
1920 | if (ofproto->has_bundle_action) { | |
1921 | ofproto->need_revalidate = true; | |
1922 | } | |
1923 | } | |
1924 | ||
015e08bc | 1925 | ofport->may_enable = enable; |
abe529af BP |
1926 | } |
1927 | ||
1928 | static void | |
1929 | port_wait(struct ofport_dpif *ofport) | |
1930 | { | |
1931 | if (ofport->cfm) { | |
1932 | cfm_wait(ofport->cfm); | |
1933 | } | |
1934 | } | |
1935 | ||
1936 | static int | |
1937 | port_query_by_name(const struct ofproto *ofproto_, const char *devname, | |
1938 | struct ofproto_port *ofproto_port) | |
1939 | { | |
1940 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1941 | struct dpif_port dpif_port; | |
1942 | int error; | |
1943 | ||
1944 | error = dpif_port_query_by_name(ofproto->dpif, devname, &dpif_port); | |
1945 | if (!error) { | |
1946 | ofproto_port_from_dpif_port(ofproto_port, &dpif_port); | |
1947 | } | |
1948 | return error; | |
1949 | } | |
1950 | ||
1951 | static int | |
1952 | port_add(struct ofproto *ofproto_, struct netdev *netdev, uint16_t *ofp_portp) | |
1953 | { | |
1954 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1955 | uint16_t odp_port; | |
1956 | int error; | |
1957 | ||
1958 | error = dpif_port_add(ofproto->dpif, netdev, &odp_port); | |
1959 | if (!error) { | |
1960 | *ofp_portp = odp_port_to_ofp_port(odp_port); | |
1961 | } | |
1962 | return error; | |
1963 | } | |
1964 | ||
1965 | static int | |
1966 | port_del(struct ofproto *ofproto_, uint16_t ofp_port) | |
1967 | { | |
1968 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1969 | int error; | |
1970 | ||
1971 | error = dpif_port_del(ofproto->dpif, ofp_port_to_odp_port(ofp_port)); | |
1972 | if (!error) { | |
1973 | struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port); | |
1974 | if (ofport) { | |
1975 | /* The caller is going to close ofport->up.netdev. If this is a | |
1976 | * bonded port, then the bond is using that netdev, so remove it | |
1977 | * from the bond. The client will need to reconfigure everything | |
1978 | * after deleting ports, so then the slave will get re-added. */ | |
1979 | bundle_remove(&ofport->up); | |
1980 | } | |
1981 | } | |
1982 | return error; | |
1983 | } | |
1984 | ||
1985 | struct port_dump_state { | |
1986 | struct dpif_port_dump dump; | |
1987 | bool done; | |
1988 | }; | |
1989 | ||
1990 | static int | |
1991 | port_dump_start(const struct ofproto *ofproto_, void **statep) | |
1992 | { | |
1993 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
1994 | struct port_dump_state *state; | |
1995 | ||
1996 | *statep = state = xmalloc(sizeof *state); | |
1997 | dpif_port_dump_start(&state->dump, ofproto->dpif); | |
1998 | state->done = false; | |
1999 | return 0; | |
2000 | } | |
2001 | ||
2002 | static int | |
2003 | port_dump_next(const struct ofproto *ofproto_ OVS_UNUSED, void *state_, | |
2004 | struct ofproto_port *port) | |
2005 | { | |
2006 | struct port_dump_state *state = state_; | |
2007 | struct dpif_port dpif_port; | |
2008 | ||
2009 | if (dpif_port_dump_next(&state->dump, &dpif_port)) { | |
2010 | ofproto_port_from_dpif_port(port, &dpif_port); | |
2011 | return 0; | |
2012 | } else { | |
2013 | int error = dpif_port_dump_done(&state->dump); | |
2014 | state->done = true; | |
2015 | return error ? error : EOF; | |
2016 | } | |
2017 | } | |
2018 | ||
2019 | static int | |
2020 | port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_) | |
2021 | { | |
2022 | struct port_dump_state *state = state_; | |
2023 | ||
2024 | if (!state->done) { | |
2025 | dpif_port_dump_done(&state->dump); | |
2026 | } | |
2027 | free(state); | |
2028 | return 0; | |
2029 | } | |
2030 | ||
2031 | static int | |
2032 | port_poll(const struct ofproto *ofproto_, char **devnamep) | |
2033 | { | |
2034 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2035 | return dpif_port_poll(ofproto->dpif, devnamep); | |
2036 | } | |
2037 | ||
2038 | static void | |
2039 | port_poll_wait(const struct ofproto *ofproto_) | |
2040 | { | |
2041 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
2042 | dpif_port_poll_wait(ofproto->dpif); | |
2043 | } | |
2044 | ||
2045 | static int | |
2046 | port_is_lacp_current(const struct ofport *ofport_) | |
2047 | { | |
2048 | const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_); | |
2049 | return (ofport->bundle && ofport->bundle->lacp | |
2050 | ? lacp_slave_is_current(ofport->bundle->lacp, ofport) | |
2051 | : -1); | |
2052 | } | |
2053 | \f | |
2054 | /* Upcall handling. */ | |
2055 | ||
501f8d1f BP |
2056 | /* Flow miss batching. |
2057 | * | |
2058 | * Some dpifs implement operations faster when you hand them off in a batch. | |
2059 | * To allow batching, "struct flow_miss" queues the dpif-related work needed | |
2060 | * for a given flow. Each "struct flow_miss" corresponds to sending one or | |
2061 | * more packets, plus possibly installing the flow in the dpif. | |
2062 | * | |
2063 | * So far we only batch the operations that affect flow setup time the most. | |
2064 | * It's possible to batch more than that, but the benefit might be minimal. */ | |
2065 | struct flow_miss { | |
2066 | struct hmap_node hmap_node; | |
2067 | struct flow flow; | |
2068 | const struct nlattr *key; | |
2069 | size_t key_len; | |
2070 | struct list packets; | |
2071 | }; | |
2072 | ||
2073 | struct flow_miss_op { | |
2074 | union dpif_op dpif_op; | |
2075 | struct facet *facet; | |
2076 | }; | |
2077 | ||
62cd7072 BP |
2078 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each |
2079 | * OpenFlow controller as necessary according to their individual | |
2080 | * configurations. | |
2081 | * | |
2082 | * If 'clone' is true, the caller retains ownership of 'packet'. Otherwise, | |
2083 | * ownership is transferred to this function. */ | |
2084 | static void | |
2085 | send_packet_in_miss(struct ofproto_dpif *ofproto, struct ofpbuf *packet, | |
2086 | const struct flow *flow, bool clone) | |
2087 | { | |
2088 | struct ofputil_packet_in pin; | |
2089 | ||
2090 | pin.packet = packet; | |
2091 | pin.in_port = flow->in_port; | |
2092 | pin.reason = OFPR_NO_MATCH; | |
2093 | pin.buffer_id = 0; /* not yet known */ | |
2094 | pin.send_len = 0; /* not used for flow table misses */ | |
2095 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, | |
2096 | clone ? NULL : packet); | |
2097 | } | |
2098 | ||
2099 | /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_ACTION to each | |
2100 | * OpenFlow controller as necessary according to their individual | |
2101 | * configurations. | |
2102 | * | |
2103 | * 'send_len' should be the number of bytes of 'packet' to send to the | |
2104 | * controller, as specified in the action that caused the packet to be sent. | |
abe529af BP |
2105 | * |
2106 | * If 'clone' is true, the caller retains ownership of 'upcall->packet'. | |
2107 | * Otherwise, ownership is transferred to this function. */ | |
2108 | static void | |
62cd7072 BP |
2109 | send_packet_in_action(struct ofproto_dpif *ofproto, struct ofpbuf *packet, |
2110 | uint64_t userdata, const struct flow *flow, bool clone) | |
abe529af BP |
2111 | { |
2112 | struct ofputil_packet_in pin; | |
6ff686f2 | 2113 | struct user_action_cookie cookie; |
abe529af | 2114 | |
62cd7072 BP |
2115 | memcpy(&cookie, &userdata, sizeof(cookie)); |
2116 | ||
2117 | pin.packet = packet; | |
abe529af | 2118 | pin.in_port = flow->in_port; |
62cd7072 | 2119 | pin.reason = OFPR_ACTION; |
abe529af | 2120 | pin.buffer_id = 0; /* not yet known */ |
6ff686f2 | 2121 | pin.send_len = cookie.data; |
78bd1cd0 | 2122 | connmgr_send_packet_in(ofproto->up.connmgr, &pin, flow, |
62cd7072 | 2123 | clone ? NULL : packet); |
abe529af BP |
2124 | } |
2125 | ||
2126 | static bool | |
2127 | process_special(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2128 | const struct ofpbuf *packet) | |
2129 | { | |
b6e001b6 EJ |
2130 | struct ofport_dpif *ofport = get_ofp_port(ofproto, flow->in_port); |
2131 | ||
2132 | if (!ofport) { | |
2133 | return false; | |
2134 | } | |
2135 | ||
ef9819b5 | 2136 | if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) { |
b6e001b6 | 2137 | if (packet) { |
abe529af BP |
2138 | cfm_process_heartbeat(ofport->cfm, packet); |
2139 | } | |
2140 | return true; | |
b6e001b6 EJ |
2141 | } else if (ofport->bundle && ofport->bundle->lacp |
2142 | && flow->dl_type == htons(ETH_TYPE_LACP)) { | |
2143 | if (packet) { | |
2144 | lacp_process_packet(ofport->bundle->lacp, ofport, packet); | |
abe529af | 2145 | } |
da37ebac | 2146 | return true; |
21f7563c JP |
2147 | } else if (ofproto->stp && stp_should_process_flow(flow)) { |
2148 | if (packet) { | |
2149 | stp_process_packet(ofport, packet); | |
2150 | } | |
2151 | return true; | |
abe529af BP |
2152 | } |
2153 | return false; | |
2154 | } | |
2155 | ||
501f8d1f BP |
2156 | static struct flow_miss * |
2157 | flow_miss_create(struct hmap *todo, const struct flow *flow, | |
2158 | const struct nlattr *key, size_t key_len) | |
abe529af | 2159 | { |
501f8d1f BP |
2160 | uint32_t hash = flow_hash(flow, 0); |
2161 | struct flow_miss *miss; | |
abe529af | 2162 | |
501f8d1f BP |
2163 | HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) { |
2164 | if (flow_equal(&miss->flow, flow)) { | |
2165 | return miss; | |
2166 | } | |
2167 | } | |
abe529af | 2168 | |
501f8d1f BP |
2169 | miss = xmalloc(sizeof *miss); |
2170 | hmap_insert(todo, &miss->hmap_node, hash); | |
2171 | miss->flow = *flow; | |
2172 | miss->key = key; | |
2173 | miss->key_len = key_len; | |
2174 | list_init(&miss->packets); | |
2175 | return miss; | |
2176 | } | |
abe529af | 2177 | |
501f8d1f BP |
2178 | static void |
2179 | handle_flow_miss(struct ofproto_dpif *ofproto, struct flow_miss *miss, | |
2180 | struct flow_miss_op *ops, size_t *n_ops) | |
2181 | { | |
2182 | const struct flow *flow = &miss->flow; | |
2183 | struct ofpbuf *packet, *next_packet; | |
2184 | struct facet *facet; | |
abe529af | 2185 | |
501f8d1f | 2186 | facet = facet_lookup_valid(ofproto, flow); |
abe529af | 2187 | if (!facet) { |
501f8d1f BP |
2188 | struct rule_dpif *rule; |
2189 | ||
2190 | rule = rule_dpif_lookup(ofproto, flow, 0); | |
abe529af BP |
2191 | if (!rule) { |
2192 | /* Don't send a packet-in if OFPPC_NO_PACKET_IN asserted. */ | |
501f8d1f | 2193 | struct ofport_dpif *port = get_ofp_port(ofproto, flow->in_port); |
abe529af BP |
2194 | if (port) { |
2195 | if (port->up.opp.config & htonl(OFPPC_NO_PACKET_IN)) { | |
2196 | COVERAGE_INC(ofproto_dpif_no_packet_in); | |
2197 | /* XXX install 'drop' flow entry */ | |
abe529af BP |
2198 | return; |
2199 | } | |
2200 | } else { | |
2201 | VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, | |
501f8d1f BP |
2202 | flow->in_port); |
2203 | } | |
2204 | ||
2205 | LIST_FOR_EACH_SAFE (packet, next_packet, list_node, | |
2206 | &miss->packets) { | |
2207 | list_remove(&packet->list_node); | |
2208 | send_packet_in_miss(ofproto, packet, flow, false); | |
abe529af BP |
2209 | } |
2210 | ||
abe529af BP |
2211 | return; |
2212 | } | |
2213 | ||
501f8d1f | 2214 | facet = facet_create(rule, flow); |
abe529af BP |
2215 | } |
2216 | ||
501f8d1f BP |
2217 | LIST_FOR_EACH_SAFE (packet, next_packet, list_node, &miss->packets) { |
2218 | list_remove(&packet->list_node); | |
2219 | ofproto->n_matches++; | |
2220 | ||
2221 | if (facet->rule->up.cr.priority == FAIL_OPEN_PRIORITY) { | |
2222 | /* | |
2223 | * Extra-special case for fail-open mode. | |
2224 | * | |
2225 | * We are in fail-open mode and the packet matched the fail-open | |
2226 | * rule, but we are connected to a controller too. We should send | |
2227 | * the packet up to the controller in the hope that it will try to | |
2228 | * set up a flow and thereby allow us to exit fail-open. | |
2229 | * | |
2230 | * See the top-level comment in fail-open.c for more information. | |
2231 | */ | |
2232 | send_packet_in_miss(ofproto, packet, flow, true); | |
2233 | } | |
2234 | ||
2235 | if (!facet->may_install) { | |
2236 | facet_make_actions(ofproto, facet, packet); | |
2237 | } | |
2238 | if (!execute_controller_action(ofproto, &facet->flow, | |
2239 | facet->actions, facet->actions_len, | |
2240 | packet)) { | |
2241 | struct flow_miss_op *op = &ops[(*n_ops)++]; | |
2242 | struct dpif_execute *execute = &op->dpif_op.execute; | |
2243 | ||
2244 | op->facet = facet; | |
2245 | execute->type = DPIF_OP_EXECUTE; | |
2246 | execute->key = miss->key; | |
2247 | execute->key_len = miss->key_len; | |
2248 | execute->actions | |
2249 | = (facet->may_install | |
2250 | ? facet->actions | |
2251 | : xmemdup(facet->actions, facet->actions_len)); | |
2252 | execute->actions_len = facet->actions_len; | |
2253 | execute->packet = packet; | |
2254 | } | |
2255 | } | |
2256 | ||
2257 | if (facet->may_install) { | |
2258 | struct flow_miss_op *op = &ops[(*n_ops)++]; | |
2259 | struct dpif_flow_put *put = &op->dpif_op.flow_put; | |
2260 | ||
2261 | op->facet = facet; | |
2262 | put->type = DPIF_OP_FLOW_PUT; | |
2263 | put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
2264 | put->key = miss->key; | |
2265 | put->key_len = miss->key_len; | |
2266 | put->actions = facet->actions; | |
2267 | put->actions_len = facet->actions_len; | |
2268 | put->stats = NULL; | |
2269 | } | |
2270 | } | |
2271 | ||
2272 | static void | |
2273 | handle_miss_upcalls(struct ofproto_dpif *ofproto, struct dpif_upcall *upcalls, | |
2274 | size_t n_upcalls) | |
2275 | { | |
2276 | struct dpif_upcall *upcall; | |
2277 | struct flow_miss *miss, *next_miss; | |
2278 | struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2]; | |
2279 | union dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2]; | |
2280 | struct hmap todo; | |
2281 | size_t n_ops; | |
2282 | size_t i; | |
2283 | ||
2284 | if (!n_upcalls) { | |
2285 | return; | |
2286 | } | |
2287 | ||
2288 | /* Construct the to-do list. | |
2289 | * | |
2290 | * This just amounts to extracting the flow from each packet and sticking | |
2291 | * the packets that have the same flow in the same "flow_miss" structure so | |
2292 | * that we can process them together. */ | |
2293 | hmap_init(&todo); | |
2294 | for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) { | |
2295 | struct flow_miss *miss; | |
2296 | struct flow flow; | |
2297 | ||
2298 | /* Obtain in_port and tun_id, at least, then set 'flow''s header | |
2299 | * pointers. */ | |
2300 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
abff858b PS |
2301 | flow_extract(upcall->packet, flow.priority, flow.tun_id, |
2302 | flow.in_port, &flow); | |
501f8d1f | 2303 | |
21f7563c | 2304 | /* Handle 802.1ag, LACP, and STP specially. */ |
501f8d1f BP |
2305 | if (process_special(ofproto, &flow, upcall->packet)) { |
2306 | ofpbuf_delete(upcall->packet); | |
2307 | ofproto->n_matches++; | |
2308 | continue; | |
2309 | } | |
2310 | ||
2311 | /* Add other packets to a to-do list. */ | |
2312 | miss = flow_miss_create(&todo, &flow, upcall->key, upcall->key_len); | |
2313 | list_push_back(&miss->packets, &upcall->packet->list_node); | |
2314 | } | |
2315 | ||
2316 | /* Process each element in the to-do list, constructing the set of | |
2317 | * operations to batch. */ | |
2318 | n_ops = 0; | |
2319 | HMAP_FOR_EACH_SAFE (miss, next_miss, hmap_node, &todo) { | |
2320 | handle_flow_miss(ofproto, miss, flow_miss_ops, &n_ops); | |
2321 | ofpbuf_list_delete(&miss->packets); | |
2322 | hmap_remove(&todo, &miss->hmap_node); | |
2323 | free(miss); | |
abe529af | 2324 | } |
501f8d1f BP |
2325 | assert(n_ops <= ARRAY_SIZE(flow_miss_ops)); |
2326 | hmap_destroy(&todo); | |
2327 | ||
2328 | /* Execute batch. */ | |
2329 | for (i = 0; i < n_ops; i++) { | |
2330 | dpif_ops[i] = &flow_miss_ops[i].dpif_op; | |
2331 | } | |
2332 | dpif_operate(ofproto->dpif, dpif_ops, n_ops); | |
2333 | ||
2334 | /* Free memory and update facets. */ | |
2335 | for (i = 0; i < n_ops; i++) { | |
2336 | struct flow_miss_op *op = &flow_miss_ops[i]; | |
2337 | struct dpif_execute *execute; | |
2338 | struct dpif_flow_put *put; | |
2339 | ||
2340 | switch (op->dpif_op.type) { | |
2341 | case DPIF_OP_EXECUTE: | |
2342 | execute = &op->dpif_op.execute; | |
2343 | if (op->facet->actions != execute->actions) { | |
2344 | free((struct nlattr *) execute->actions); | |
2345 | } | |
2346 | ofpbuf_delete((struct ofpbuf *) execute->packet); | |
2347 | break; | |
abe529af | 2348 | |
501f8d1f BP |
2349 | case DPIF_OP_FLOW_PUT: |
2350 | put = &op->dpif_op.flow_put; | |
2351 | if (!put->error) { | |
2352 | op->facet->installed = true; | |
2353 | } | |
2354 | break; | |
2355 | } | |
2356 | } | |
abe529af BP |
2357 | } |
2358 | ||
2359 | static void | |
6ff686f2 PS |
2360 | handle_userspace_upcall(struct ofproto_dpif *ofproto, |
2361 | struct dpif_upcall *upcall) | |
abe529af BP |
2362 | { |
2363 | struct flow flow; | |
6ff686f2 | 2364 | struct user_action_cookie cookie; |
abe529af | 2365 | |
6ff686f2 | 2366 | memcpy(&cookie, &upcall->userdata, sizeof(cookie)); |
abe529af | 2367 | |
6ff686f2 | 2368 | if (cookie.type == USER_ACTION_COOKIE_SFLOW) { |
abe529af BP |
2369 | if (ofproto->sflow) { |
2370 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
6ff686f2 | 2371 | dpif_sflow_received(ofproto->sflow, upcall->packet, &flow, &cookie); |
abe529af BP |
2372 | } |
2373 | ofpbuf_delete(upcall->packet); | |
6ff686f2 PS |
2374 | |
2375 | } else if (cookie.type == USER_ACTION_COOKIE_CONTROLLER) { | |
2376 | COVERAGE_INC(ofproto_dpif_ctlr_action); | |
2377 | odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); | |
62cd7072 BP |
2378 | send_packet_in_action(ofproto, upcall->packet, upcall->userdata, |
2379 | &flow, false); | |
6ff686f2 PS |
2380 | } else { |
2381 | VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata); | |
2382 | } | |
2383 | } | |
2384 | ||
2385 | static void | |
2386 | handle_upcall(struct ofproto_dpif *ofproto, struct dpif_upcall *upcall) | |
2387 | { | |
2388 | switch (upcall->type) { | |
2389 | case DPIF_UC_ACTION: | |
2390 | handle_userspace_upcall(ofproto, upcall); | |
abe529af BP |
2391 | break; |
2392 | ||
2393 | case DPIF_UC_MISS: | |
501f8d1f BP |
2394 | /* The caller handles these. */ |
2395 | NOT_REACHED(); | |
abe529af BP |
2396 | |
2397 | case DPIF_N_UC_TYPES: | |
2398 | default: | |
2399 | VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type); | |
2400 | break; | |
2401 | } | |
2402 | } | |
2403 | \f | |
2404 | /* Flow expiration. */ | |
2405 | ||
2406 | static int facet_max_idle(const struct ofproto_dpif *); | |
2407 | static void update_stats(struct ofproto_dpif *); | |
2408 | static void rule_expire(struct rule_dpif *); | |
2409 | static void expire_facets(struct ofproto_dpif *, int dp_max_idle); | |
2410 | ||
2411 | /* This function is called periodically by run(). Its job is to collect | |
2412 | * updates for the flows that have been installed into the datapath, most | |
2413 | * importantly when they last were used, and then use that information to | |
2414 | * expire flows that have not been used recently. | |
2415 | * | |
2416 | * Returns the number of milliseconds after which it should be called again. */ | |
2417 | static int | |
2418 | expire(struct ofproto_dpif *ofproto) | |
2419 | { | |
2420 | struct rule_dpif *rule, *next_rule; | |
0697b5c3 | 2421 | struct classifier *table; |
abe529af BP |
2422 | int dp_max_idle; |
2423 | ||
2424 | /* Update stats for each flow in the datapath. */ | |
2425 | update_stats(ofproto); | |
2426 | ||
2427 | /* Expire facets that have been idle too long. */ | |
2428 | dp_max_idle = facet_max_idle(ofproto); | |
2429 | expire_facets(ofproto, dp_max_idle); | |
2430 | ||
2431 | /* Expire OpenFlow flows whose idle_timeout or hard_timeout has passed. */ | |
0697b5c3 BP |
2432 | OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) { |
2433 | struct cls_cursor cursor; | |
2434 | ||
2435 | cls_cursor_init(&cursor, table, NULL); | |
2436 | CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) { | |
2437 | rule_expire(rule); | |
2438 | } | |
abe529af BP |
2439 | } |
2440 | ||
2441 | /* All outstanding data in existing flows has been accounted, so it's a | |
2442 | * good time to do bond rebalancing. */ | |
2443 | if (ofproto->has_bonded_bundles) { | |
2444 | struct ofbundle *bundle; | |
2445 | ||
2446 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
2447 | if (bundle->bond) { | |
2448 | bond_rebalance(bundle->bond, &ofproto->revalidate_set); | |
2449 | } | |
2450 | } | |
2451 | } | |
2452 | ||
2453 | return MIN(dp_max_idle, 1000); | |
2454 | } | |
2455 | ||
2456 | /* Update 'packet_count', 'byte_count', and 'used' members of installed facets. | |
2457 | * | |
2458 | * This function also pushes statistics updates to rules which each facet | |
2459 | * resubmits into. Generally these statistics will be accurate. However, if a | |
2460 | * facet changes the rule it resubmits into at some time in between | |
2461 | * update_stats() runs, it is possible that statistics accrued to the | |
2462 | * old rule will be incorrectly attributed to the new rule. This could be | |
2463 | * avoided by calling update_stats() whenever rules are created or | |
2464 | * deleted. However, the performance impact of making so many calls to the | |
2465 | * datapath do not justify the benefit of having perfectly accurate statistics. | |
2466 | */ | |
2467 | static void | |
2468 | update_stats(struct ofproto_dpif *p) | |
2469 | { | |
2470 | const struct dpif_flow_stats *stats; | |
2471 | struct dpif_flow_dump dump; | |
2472 | const struct nlattr *key; | |
2473 | size_t key_len; | |
2474 | ||
2475 | dpif_flow_dump_start(&dump, p->dpif); | |
2476 | while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) { | |
2477 | struct facet *facet; | |
2478 | struct flow flow; | |
2479 | ||
2480 | if (odp_flow_key_to_flow(key, key_len, &flow)) { | |
2481 | struct ds s; | |
2482 | ||
2483 | ds_init(&s); | |
2484 | odp_flow_key_format(key, key_len, &s); | |
df2c07f4 | 2485 | VLOG_WARN_RL(&rl, "failed to convert datapath flow key to flow: %s", |
abe529af BP |
2486 | ds_cstr(&s)); |
2487 | ds_destroy(&s); | |
2488 | ||
2489 | continue; | |
2490 | } | |
2491 | facet = facet_find(p, &flow); | |
2492 | ||
2493 | if (facet && facet->installed) { | |
2494 | ||
2495 | if (stats->n_packets >= facet->dp_packet_count) { | |
2496 | uint64_t extra = stats->n_packets - facet->dp_packet_count; | |
2497 | facet->packet_count += extra; | |
2498 | } else { | |
2499 | VLOG_WARN_RL(&rl, "unexpected packet count from the datapath"); | |
2500 | } | |
2501 | ||
2502 | if (stats->n_bytes >= facet->dp_byte_count) { | |
2503 | facet->byte_count += stats->n_bytes - facet->dp_byte_count; | |
2504 | } else { | |
2505 | VLOG_WARN_RL(&rl, "unexpected byte count from datapath"); | |
2506 | } | |
2507 | ||
2508 | facet->dp_packet_count = stats->n_packets; | |
2509 | facet->dp_byte_count = stats->n_bytes; | |
2510 | ||
2511 | facet_update_time(p, facet, stats->used); | |
55af77bb | 2512 | facet_account(p, facet); |
abe529af BP |
2513 | facet_push_stats(facet); |
2514 | } else { | |
2515 | /* There's a flow in the datapath that we know nothing about. | |
2516 | * Delete it. */ | |
2517 | COVERAGE_INC(facet_unexpected); | |
2518 | dpif_flow_del(p->dpif, key, key_len, NULL); | |
2519 | } | |
2520 | } | |
2521 | dpif_flow_dump_done(&dump); | |
2522 | } | |
2523 | ||
2524 | /* Calculates and returns the number of milliseconds of idle time after which | |
2525 | * facets should expire from the datapath and we should fold their statistics | |
2526 | * into their parent rules in userspace. */ | |
2527 | static int | |
2528 | facet_max_idle(const struct ofproto_dpif *ofproto) | |
2529 | { | |
2530 | /* | |
2531 | * Idle time histogram. | |
2532 | * | |
2533 | * Most of the time a switch has a relatively small number of facets. When | |
2534 | * this is the case we might as well keep statistics for all of them in | |
2535 | * userspace and to cache them in the kernel datapath for performance as | |
2536 | * well. | |
2537 | * | |
2538 | * As the number of facets increases, the memory required to maintain | |
2539 | * statistics about them in userspace and in the kernel becomes | |
2540 | * significant. However, with a large number of facets it is likely that | |
2541 | * only a few of them are "heavy hitters" that consume a large amount of | |
2542 | * bandwidth. At this point, only heavy hitters are worth caching in the | |
2543 | * kernel and maintaining in userspaces; other facets we can discard. | |
2544 | * | |
2545 | * The technique used to compute the idle time is to build a histogram with | |
2546 | * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each facet | |
2547 | * that is installed in the kernel gets dropped in the appropriate bucket. | |
2548 | * After the histogram has been built, we compute the cutoff so that only | |
084f5290 SH |
2549 | * the most-recently-used 1% of facets (but at least |
2550 | * ofproto->up.flow_eviction_threshold flows) are kept cached. At least | |
2551 | * the most-recently-used bucket of facets is kept, so actually an | |
2552 | * arbitrary number of facets can be kept in any given expiration run | |
2553 | * (though the next run will delete most of those unless they receive | |
2554 | * additional data). | |
abe529af BP |
2555 | * |
2556 | * This requires a second pass through the facets, in addition to the pass | |
2557 | * made by update_stats(), because the former function never looks | |
2558 | * at uninstallable facets. | |
2559 | */ | |
2560 | enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) }; | |
2561 | enum { N_BUCKETS = 5000 / BUCKET_WIDTH }; | |
2562 | int buckets[N_BUCKETS] = { 0 }; | |
f11c1ef4 | 2563 | int total, subtotal, bucket; |
abe529af | 2564 | struct facet *facet; |
abe529af BP |
2565 | long long int now; |
2566 | int i; | |
2567 | ||
2568 | total = hmap_count(&ofproto->facets); | |
084f5290 | 2569 | if (total <= ofproto->up.flow_eviction_threshold) { |
abe529af BP |
2570 | return N_BUCKETS * BUCKET_WIDTH; |
2571 | } | |
2572 | ||
2573 | /* Build histogram. */ | |
2574 | now = time_msec(); | |
2575 | HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) { | |
2576 | long long int idle = now - facet->used; | |
2577 | int bucket = (idle <= 0 ? 0 | |
2578 | : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1 | |
2579 | : (unsigned int) idle / BUCKET_WIDTH); | |
2580 | buckets[bucket]++; | |
2581 | } | |
2582 | ||
2583 | /* Find the first bucket whose flows should be expired. */ | |
f11c1ef4 SH |
2584 | subtotal = bucket = 0; |
2585 | do { | |
2586 | subtotal += buckets[bucket++]; | |
084f5290 SH |
2587 | } while (bucket < N_BUCKETS && |
2588 | subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100)); | |
abe529af BP |
2589 | |
2590 | if (VLOG_IS_DBG_ENABLED()) { | |
2591 | struct ds s; | |
2592 | ||
2593 | ds_init(&s); | |
2594 | ds_put_cstr(&s, "keep"); | |
2595 | for (i = 0; i < N_BUCKETS; i++) { | |
2596 | if (i == bucket) { | |
2597 | ds_put_cstr(&s, ", drop"); | |
2598 | } | |
2599 | if (buckets[i]) { | |
2600 | ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]); | |
2601 | } | |
2602 | } | |
2603 | VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s)); | |
2604 | ds_destroy(&s); | |
2605 | } | |
2606 | ||
2607 | return bucket * BUCKET_WIDTH; | |
2608 | } | |
2609 | ||
2610 | static void | |
2611 | facet_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet) | |
2612 | { | |
2613 | if (ofproto->netflow && !facet_is_controller_flow(facet) && | |
2614 | netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) { | |
2615 | struct ofexpired expired; | |
2616 | ||
2617 | if (facet->installed) { | |
2618 | struct dpif_flow_stats stats; | |
2619 | ||
2620 | facet_put__(ofproto, facet, facet->actions, facet->actions_len, | |
2621 | &stats); | |
2622 | facet_update_stats(ofproto, facet, &stats); | |
2623 | } | |
2624 | ||
2625 | expired.flow = facet->flow; | |
2626 | expired.packet_count = facet->packet_count; | |
2627 | expired.byte_count = facet->byte_count; | |
2628 | expired.used = facet->used; | |
2629 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
2630 | } | |
2631 | } | |
2632 | ||
2633 | static void | |
2634 | expire_facets(struct ofproto_dpif *ofproto, int dp_max_idle) | |
2635 | { | |
2636 | long long int cutoff = time_msec() - dp_max_idle; | |
2637 | struct facet *facet, *next_facet; | |
2638 | ||
2639 | HMAP_FOR_EACH_SAFE (facet, next_facet, hmap_node, &ofproto->facets) { | |
2640 | facet_active_timeout(ofproto, facet); | |
2641 | if (facet->used < cutoff) { | |
2642 | facet_remove(ofproto, facet); | |
2643 | } | |
2644 | } | |
2645 | } | |
2646 | ||
2647 | /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules, | |
2648 | * then delete it entirely. */ | |
2649 | static void | |
2650 | rule_expire(struct rule_dpif *rule) | |
2651 | { | |
2652 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2653 | struct facet *facet, *next_facet; | |
2654 | long long int now; | |
2655 | uint8_t reason; | |
2656 | ||
2657 | /* Has 'rule' expired? */ | |
2658 | now = time_msec(); | |
2659 | if (rule->up.hard_timeout | |
308881af | 2660 | && now > rule->up.modified + rule->up.hard_timeout * 1000) { |
abe529af BP |
2661 | reason = OFPRR_HARD_TIMEOUT; |
2662 | } else if (rule->up.idle_timeout && list_is_empty(&rule->facets) | |
2663 | && now > rule->used + rule->up.idle_timeout * 1000) { | |
2664 | reason = OFPRR_IDLE_TIMEOUT; | |
2665 | } else { | |
2666 | return; | |
2667 | } | |
2668 | ||
2669 | COVERAGE_INC(ofproto_dpif_expired); | |
2670 | ||
2671 | /* Update stats. (This is a no-op if the rule expired due to an idle | |
2672 | * timeout, because that only happens when the rule has no facets left.) */ | |
2673 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { | |
2674 | facet_remove(ofproto, facet); | |
2675 | } | |
2676 | ||
2677 | /* Get rid of the rule. */ | |
2678 | ofproto_rule_expire(&rule->up, reason); | |
2679 | } | |
2680 | \f | |
2681 | /* Facets. */ | |
2682 | ||
f3827897 | 2683 | /* Creates and returns a new facet owned by 'rule', given a 'flow'. |
abe529af BP |
2684 | * |
2685 | * The caller must already have determined that no facet with an identical | |
2686 | * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in | |
f3827897 BP |
2687 | * the ofproto's classifier table. |
2688 | * | |
2689 | * The facet will initially have no ODP actions. The caller should fix that | |
2690 | * by calling facet_make_actions(). */ | |
abe529af | 2691 | static struct facet * |
f3827897 | 2692 | facet_create(struct rule_dpif *rule, const struct flow *flow) |
abe529af BP |
2693 | { |
2694 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
2695 | struct facet *facet; | |
2696 | ||
2697 | facet = xzalloc(sizeof *facet); | |
2698 | facet->used = time_msec(); | |
2699 | hmap_insert(&ofproto->facets, &facet->hmap_node, flow_hash(flow, 0)); | |
2700 | list_push_back(&rule->facets, &facet->list_node); | |
2701 | facet->rule = rule; | |
2702 | facet->flow = *flow; | |
2703 | netflow_flow_init(&facet->nf_flow); | |
2704 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used); | |
2705 | ||
abe529af BP |
2706 | return facet; |
2707 | } | |
2708 | ||
2709 | static void | |
2710 | facet_free(struct facet *facet) | |
2711 | { | |
2712 | free(facet->actions); | |
2713 | free(facet); | |
2714 | } | |
2715 | ||
abe529af | 2716 | static bool |
3d9e05f8 BP |
2717 | execute_controller_action(struct ofproto_dpif *ofproto, |
2718 | const struct flow *flow, | |
2719 | const struct nlattr *odp_actions, size_t actions_len, | |
2720 | struct ofpbuf *packet) | |
2721 | { | |
2722 | if (actions_len | |
2723 | && odp_actions->nla_type == OVS_ACTION_ATTR_USERSPACE | |
2724 | && NLA_ALIGN(odp_actions->nla_len) == actions_len) { | |
62cd7072 BP |
2725 | /* As an optimization, avoid a round-trip from userspace to kernel to |
2726 | * userspace. This also avoids possibly filling up kernel packet | |
2727 | * buffers along the way. | |
2728 | * | |
2729 | * This optimization will not accidentally catch sFlow | |
2730 | * OVS_ACTION_ATTR_USERSPACE actions, since those are encapsulated | |
2731 | * inside OVS_ACTION_ATTR_SAMPLE. */ | |
2732 | const struct nlattr *nla; | |
2733 | ||
2734 | nla = nl_attr_find_nested(odp_actions, OVS_USERSPACE_ATTR_USERDATA); | |
2735 | send_packet_in_action(ofproto, packet, nl_attr_get_u64(nla), flow, | |
2736 | false); | |
2737 | return true; | |
3d9e05f8 BP |
2738 | } else { |
2739 | return false; | |
2740 | } | |
2741 | } | |
2742 | ||
2743 | /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on | |
2744 | * 'packet', which arrived on 'in_port'. | |
2745 | * | |
2746 | * Takes ownership of 'packet'. */ | |
2747 | static bool | |
2748 | execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow, | |
2749 | const struct nlattr *odp_actions, size_t actions_len, | |
2750 | struct ofpbuf *packet) | |
2751 | { | |
2752 | struct odputil_keybuf keybuf; | |
2753 | struct ofpbuf key; | |
2754 | int error; | |
2755 | ||
2756 | if (execute_controller_action(ofproto, flow, odp_actions, actions_len, | |
2757 | packet)) { | |
2758 | return true; | |
6ff686f2 | 2759 | } |
abe529af | 2760 | |
6ff686f2 PS |
2761 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
2762 | odp_flow_key_from_flow(&key, flow); | |
80e5eed9 | 2763 | |
6ff686f2 PS |
2764 | error = dpif_execute(ofproto->dpif, key.data, key.size, |
2765 | odp_actions, actions_len, packet); | |
80e5eed9 | 2766 | |
6ff686f2 PS |
2767 | ofpbuf_delete(packet); |
2768 | return !error; | |
abe529af BP |
2769 | } |
2770 | ||
2771 | /* Executes the actions indicated by 'facet' on 'packet' and credits 'facet''s | |
2772 | * statistics appropriately. 'packet' must have at least sizeof(struct | |
2773 | * ofp_packet_in) bytes of headroom. | |
2774 | * | |
2775 | * For correct results, 'packet' must actually be in 'facet''s flow; that is, | |
2776 | * applying flow_extract() to 'packet' would yield the same flow as | |
2777 | * 'facet->flow'. | |
2778 | * | |
df2c07f4 JP |
2779 | * 'facet' must have accurately composed datapath actions; that is, it must |
2780 | * not be in need of revalidation. | |
abe529af BP |
2781 | * |
2782 | * Takes ownership of 'packet'. */ | |
2783 | static void | |
2784 | facet_execute(struct ofproto_dpif *ofproto, struct facet *facet, | |
2785 | struct ofpbuf *packet) | |
2786 | { | |
2787 | struct dpif_flow_stats stats; | |
2788 | ||
2789 | assert(ofpbuf_headroom(packet) >= sizeof(struct ofp_packet_in)); | |
2790 | ||
572b7068 | 2791 | dpif_flow_stats_extract(&facet->flow, packet, &stats); |
abe529af BP |
2792 | stats.used = time_msec(); |
2793 | if (execute_odp_actions(ofproto, &facet->flow, | |
2794 | facet->actions, facet->actions_len, packet)) { | |
2795 | facet_update_stats(ofproto, facet, &stats); | |
2796 | } | |
2797 | } | |
2798 | ||
2799 | /* Remove 'facet' from 'ofproto' and free up the associated memory: | |
2800 | * | |
2801 | * - If 'facet' was installed in the datapath, uninstalls it and updates its | |
2802 | * rule's statistics, via facet_uninstall(). | |
2803 | * | |
2804 | * - Removes 'facet' from its rule and from ofproto->facets. | |
2805 | */ | |
2806 | static void | |
2807 | facet_remove(struct ofproto_dpif *ofproto, struct facet *facet) | |
2808 | { | |
2809 | facet_uninstall(ofproto, facet); | |
2810 | facet_flush_stats(ofproto, facet); | |
2811 | hmap_remove(&ofproto->facets, &facet->hmap_node); | |
2812 | list_remove(&facet->list_node); | |
2813 | facet_free(facet); | |
2814 | } | |
2815 | ||
df2c07f4 | 2816 | /* Composes the datapath actions for 'facet' based on its rule's actions. */ |
abe529af BP |
2817 | static void |
2818 | facet_make_actions(struct ofproto_dpif *p, struct facet *facet, | |
2819 | const struct ofpbuf *packet) | |
2820 | { | |
2821 | const struct rule_dpif *rule = facet->rule; | |
2822 | struct ofpbuf *odp_actions; | |
2823 | struct action_xlate_ctx ctx; | |
2824 | ||
2825 | action_xlate_ctx_init(&ctx, p, &facet->flow, packet); | |
2826 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
2827 | facet->tags = ctx.tags; | |
2828 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
2829 | facet->has_learn = ctx.has_learn; |
2830 | facet->has_normal = ctx.has_normal; | |
abe529af BP |
2831 | facet->nf_flow.output_iface = ctx.nf_output_iface; |
2832 | ||
2833 | if (facet->actions_len != odp_actions->size | |
2834 | || memcmp(facet->actions, odp_actions->data, odp_actions->size)) { | |
2835 | free(facet->actions); | |
2836 | facet->actions_len = odp_actions->size; | |
2837 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
2838 | } | |
2839 | ||
2840 | ofpbuf_delete(odp_actions); | |
2841 | } | |
2842 | ||
3a88e544 BP |
2843 | /* Updates 'facet''s flow in the datapath setting its actions to 'actions_len' |
2844 | * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters | |
2845 | * in the datapath will be zeroed and 'stats' will be updated with traffic new | |
2846 | * since 'facet' was last updated. | |
2847 | * | |
2848 | * Returns 0 if successful, otherwise a positive errno value.*/ | |
abe529af BP |
2849 | static int |
2850 | facet_put__(struct ofproto_dpif *ofproto, struct facet *facet, | |
2851 | const struct nlattr *actions, size_t actions_len, | |
2852 | struct dpif_flow_stats *stats) | |
2853 | { | |
2854 | struct odputil_keybuf keybuf; | |
2855 | enum dpif_flow_put_flags flags; | |
2856 | struct ofpbuf key; | |
3a88e544 | 2857 | int ret; |
abe529af BP |
2858 | |
2859 | flags = DPIF_FP_CREATE | DPIF_FP_MODIFY; | |
2860 | if (stats) { | |
2861 | flags |= DPIF_FP_ZERO_STATS; | |
abe529af BP |
2862 | } |
2863 | ||
2864 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2865 | odp_flow_key_from_flow(&key, &facet->flow); | |
2866 | ||
3a88e544 BP |
2867 | ret = dpif_flow_put(ofproto->dpif, flags, key.data, key.size, |
2868 | actions, actions_len, stats); | |
2869 | ||
2870 | if (stats) { | |
2871 | facet_reset_dp_stats(facet, stats); | |
2872 | } | |
2873 | ||
2874 | return ret; | |
abe529af BP |
2875 | } |
2876 | ||
2877 | /* If 'facet' is installable, inserts or re-inserts it into 'p''s datapath. If | |
2878 | * 'zero_stats' is true, clears any existing statistics from the datapath for | |
2879 | * 'facet'. */ | |
2880 | static void | |
2881 | facet_install(struct ofproto_dpif *p, struct facet *facet, bool zero_stats) | |
2882 | { | |
2883 | struct dpif_flow_stats stats; | |
2884 | ||
2885 | if (facet->may_install | |
2886 | && !facet_put__(p, facet, facet->actions, facet->actions_len, | |
2887 | zero_stats ? &stats : NULL)) { | |
2888 | facet->installed = true; | |
2889 | } | |
2890 | } | |
2891 | ||
2892 | static void | |
55af77bb | 2893 | facet_account(struct ofproto_dpif *ofproto, struct facet *facet) |
abe529af | 2894 | { |
55af77bb | 2895 | uint64_t n_bytes; |
abe529af | 2896 | const struct nlattr *a; |
abe529af | 2897 | unsigned int left; |
d78be13b | 2898 | ovs_be16 vlan_tci; |
abe529af | 2899 | |
55af77bb | 2900 | if (facet->byte_count <= facet->accounted_bytes) { |
abe529af BP |
2901 | return; |
2902 | } | |
55af77bb EJ |
2903 | n_bytes = facet->byte_count - facet->accounted_bytes; |
2904 | facet->accounted_bytes = facet->byte_count; | |
abe529af | 2905 | |
75a75043 | 2906 | /* Feed information from the active flows back into the learning table to |
abe529af BP |
2907 | * ensure that table is always in sync with what is actually flowing |
2908 | * through the datapath. */ | |
75a75043 BP |
2909 | if (facet->has_learn || facet->has_normal) { |
2910 | struct action_xlate_ctx ctx; | |
abe529af | 2911 | |
75a75043 BP |
2912 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); |
2913 | ctx.may_learn = true; | |
2914 | ofpbuf_delete(xlate_actions(&ctx, facet->rule->up.actions, | |
2915 | facet->rule->up.n_actions)); | |
2916 | } | |
abe529af | 2917 | |
75a75043 | 2918 | if (!facet->has_normal || !ofproto->has_bonded_bundles) { |
abe529af BP |
2919 | return; |
2920 | } | |
d78be13b BP |
2921 | |
2922 | /* This loop feeds byte counters to bond_account() for rebalancing to use | |
2923 | * as a basis. We also need to track the actual VLAN on which the packet | |
2924 | * is going to be sent to ensure that it matches the one passed to | |
2925 | * bond_choose_output_slave(). (Otherwise, we will account to the wrong | |
2926 | * hash bucket.) */ | |
2927 | vlan_tci = facet->flow.vlan_tci; | |
abe529af | 2928 | NL_ATTR_FOR_EACH_UNSAFE (a, left, facet->actions, facet->actions_len) { |
fea393b1 | 2929 | const struct ovs_action_push_vlan *vlan; |
d78be13b | 2930 | struct ofport_dpif *port; |
abe529af | 2931 | |
d78be13b | 2932 | switch (nl_attr_type(a)) { |
df2c07f4 | 2933 | case OVS_ACTION_ATTR_OUTPUT: |
abe529af BP |
2934 | port = get_odp_port(ofproto, nl_attr_get_u32(a)); |
2935 | if (port && port->bundle && port->bundle->bond) { | |
d78be13b | 2936 | bond_account(port->bundle->bond, &facet->flow, |
dc155bff | 2937 | vlan_tci_to_vid(vlan_tci), n_bytes); |
abe529af | 2938 | } |
d78be13b BP |
2939 | break; |
2940 | ||
fea393b1 BP |
2941 | case OVS_ACTION_ATTR_POP_VLAN: |
2942 | vlan_tci = htons(0); | |
d78be13b BP |
2943 | break; |
2944 | ||
fea393b1 BP |
2945 | case OVS_ACTION_ATTR_PUSH_VLAN: |
2946 | vlan = nl_attr_get(a); | |
2947 | vlan_tci = vlan->vlan_tci; | |
d78be13b | 2948 | break; |
abe529af BP |
2949 | } |
2950 | } | |
2951 | } | |
2952 | ||
2953 | /* If 'rule' is installed in the datapath, uninstalls it. */ | |
2954 | static void | |
2955 | facet_uninstall(struct ofproto_dpif *p, struct facet *facet) | |
2956 | { | |
2957 | if (facet->installed) { | |
2958 | struct odputil_keybuf keybuf; | |
2959 | struct dpif_flow_stats stats; | |
2960 | struct ofpbuf key; | |
3a88e544 | 2961 | int error; |
abe529af BP |
2962 | |
2963 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
2964 | odp_flow_key_from_flow(&key, &facet->flow); | |
2965 | ||
3a88e544 BP |
2966 | error = dpif_flow_del(p->dpif, key.data, key.size, &stats); |
2967 | facet_reset_dp_stats(facet, &stats); | |
2968 | if (!error) { | |
abe529af BP |
2969 | facet_update_stats(p, facet, &stats); |
2970 | } | |
2971 | facet->installed = false; | |
abe529af BP |
2972 | } else { |
2973 | assert(facet->dp_packet_count == 0); | |
2974 | assert(facet->dp_byte_count == 0); | |
2975 | } | |
2976 | } | |
2977 | ||
2978 | /* Returns true if the only action for 'facet' is to send to the controller. | |
2979 | * (We don't report NetFlow expiration messages for such facets because they | |
2980 | * are just part of the control logic for the network, not real traffic). */ | |
2981 | static bool | |
2982 | facet_is_controller_flow(struct facet *facet) | |
2983 | { | |
2984 | return (facet | |
2985 | && facet->rule->up.n_actions == 1 | |
2986 | && action_outputs_to_port(&facet->rule->up.actions[0], | |
2987 | htons(OFPP_CONTROLLER))); | |
2988 | } | |
2989 | ||
3a88e544 BP |
2990 | /* Resets 'facet''s datapath statistics counters. This should be called when |
2991 | * 'facet''s statistics are cleared in the datapath. If 'stats' is non-null, | |
2992 | * it should contain the statistics returned by dpif when 'facet' was reset in | |
2993 | * the datapath. 'stats' will be modified to only included statistics new | |
2994 | * since 'facet' was last updated. */ | |
2995 | static void | |
2996 | facet_reset_dp_stats(struct facet *facet, struct dpif_flow_stats *stats) | |
2997 | { | |
2998 | if (stats && facet->dp_packet_count <= stats->n_packets | |
2999 | && facet->dp_byte_count <= stats->n_bytes) { | |
3000 | stats->n_packets -= facet->dp_packet_count; | |
3001 | stats->n_bytes -= facet->dp_byte_count; | |
3002 | } | |
3003 | ||
3004 | facet->dp_packet_count = 0; | |
3005 | facet->dp_byte_count = 0; | |
3006 | } | |
3007 | ||
abe529af BP |
3008 | /* Folds all of 'facet''s statistics into its rule. Also updates the |
3009 | * accounting ofhook and emits a NetFlow expiration if appropriate. All of | |
3010 | * 'facet''s statistics in the datapath should have been zeroed and folded into | |
3011 | * its packet and byte counts before this function is called. */ | |
3012 | static void | |
3013 | facet_flush_stats(struct ofproto_dpif *ofproto, struct facet *facet) | |
3014 | { | |
3015 | assert(!facet->dp_byte_count); | |
3016 | assert(!facet->dp_packet_count); | |
3017 | ||
3018 | facet_push_stats(facet); | |
55af77bb | 3019 | facet_account(ofproto, facet); |
abe529af BP |
3020 | |
3021 | if (ofproto->netflow && !facet_is_controller_flow(facet)) { | |
3022 | struct ofexpired expired; | |
3023 | expired.flow = facet->flow; | |
3024 | expired.packet_count = facet->packet_count; | |
3025 | expired.byte_count = facet->byte_count; | |
3026 | expired.used = facet->used; | |
3027 | netflow_expire(ofproto->netflow, &facet->nf_flow, &expired); | |
3028 | } | |
3029 | ||
3030 | facet->rule->packet_count += facet->packet_count; | |
3031 | facet->rule->byte_count += facet->byte_count; | |
3032 | ||
3033 | /* Reset counters to prevent double counting if 'facet' ever gets | |
3034 | * reinstalled. */ | |
bbb5d219 | 3035 | facet_reset_counters(facet); |
abe529af BP |
3036 | |
3037 | netflow_flow_clear(&facet->nf_flow); | |
3038 | } | |
3039 | ||
3040 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3041 | * Returns it if found, otherwise a null pointer. | |
3042 | * | |
3043 | * The returned facet might need revalidation; use facet_lookup_valid() | |
3044 | * instead if that is important. */ | |
3045 | static struct facet * | |
3046 | facet_find(struct ofproto_dpif *ofproto, const struct flow *flow) | |
3047 | { | |
3048 | struct facet *facet; | |
3049 | ||
3050 | HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, flow_hash(flow, 0), | |
3051 | &ofproto->facets) { | |
3052 | if (flow_equal(flow, &facet->flow)) { | |
3053 | return facet; | |
3054 | } | |
3055 | } | |
3056 | ||
3057 | return NULL; | |
3058 | } | |
3059 | ||
3060 | /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'. | |
3061 | * Returns it if found, otherwise a null pointer. | |
3062 | * | |
3063 | * The returned facet is guaranteed to be valid. */ | |
3064 | static struct facet * | |
3065 | facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow) | |
3066 | { | |
3067 | struct facet *facet = facet_find(ofproto, flow); | |
3068 | ||
3069 | /* The facet we found might not be valid, since we could be in need of | |
3070 | * revalidation. If it is not valid, don't return it. */ | |
3071 | if (facet | |
0e4b3771 BP |
3072 | && (ofproto->need_revalidate |
3073 | || tag_set_intersects(&ofproto->revalidate_set, facet->tags)) | |
abe529af BP |
3074 | && !facet_revalidate(ofproto, facet)) { |
3075 | COVERAGE_INC(facet_invalidated); | |
3076 | return NULL; | |
3077 | } | |
3078 | ||
3079 | return facet; | |
3080 | } | |
3081 | ||
3082 | /* Re-searches 'ofproto''s classifier for a rule matching 'facet': | |
3083 | * | |
3084 | * - If the rule found is different from 'facet''s current rule, moves | |
3085 | * 'facet' to the new rule and recompiles its actions. | |
3086 | * | |
3087 | * - If the rule found is the same as 'facet''s current rule, leaves 'facet' | |
3088 | * where it is and recompiles its actions anyway. | |
3089 | * | |
3090 | * - If there is none, destroys 'facet'. | |
3091 | * | |
3092 | * Returns true if 'facet' still exists, false if it has been destroyed. */ | |
3093 | static bool | |
3094 | facet_revalidate(struct ofproto_dpif *ofproto, struct facet *facet) | |
3095 | { | |
3096 | struct action_xlate_ctx ctx; | |
3097 | struct ofpbuf *odp_actions; | |
3098 | struct rule_dpif *new_rule; | |
3099 | bool actions_changed; | |
3100 | ||
3101 | COVERAGE_INC(facet_revalidate); | |
3102 | ||
3103 | /* Determine the new rule. */ | |
29901626 | 3104 | new_rule = rule_dpif_lookup(ofproto, &facet->flow, 0); |
abe529af BP |
3105 | if (!new_rule) { |
3106 | /* No new rule, so delete the facet. */ | |
3107 | facet_remove(ofproto, facet); | |
3108 | return false; | |
3109 | } | |
3110 | ||
df2c07f4 | 3111 | /* Calculate new datapath actions. |
abe529af BP |
3112 | * |
3113 | * We do not modify any 'facet' state yet, because we might need to, e.g., | |
3114 | * emit a NetFlow expiration and, if so, we need to have the old state | |
3115 | * around to properly compose it. */ | |
3116 | action_xlate_ctx_init(&ctx, ofproto, &facet->flow, NULL); | |
3117 | odp_actions = xlate_actions(&ctx, | |
3118 | new_rule->up.actions, new_rule->up.n_actions); | |
3119 | actions_changed = (facet->actions_len != odp_actions->size | |
3120 | || memcmp(facet->actions, odp_actions->data, | |
3121 | facet->actions_len)); | |
3122 | ||
df2c07f4 JP |
3123 | /* If the datapath actions changed or the installability changed, |
3124 | * then we need to talk to the datapath. */ | |
abe529af BP |
3125 | if (actions_changed || ctx.may_set_up_flow != facet->installed) { |
3126 | if (ctx.may_set_up_flow) { | |
3127 | struct dpif_flow_stats stats; | |
3128 | ||
3129 | facet_put__(ofproto, facet, | |
3130 | odp_actions->data, odp_actions->size, &stats); | |
3131 | facet_update_stats(ofproto, facet, &stats); | |
3132 | } else { | |
3133 | facet_uninstall(ofproto, facet); | |
3134 | } | |
3135 | ||
3136 | /* The datapath flow is gone or has zeroed stats, so push stats out of | |
3137 | * 'facet' into 'rule'. */ | |
3138 | facet_flush_stats(ofproto, facet); | |
3139 | } | |
3140 | ||
3141 | /* Update 'facet' now that we've taken care of all the old state. */ | |
3142 | facet->tags = ctx.tags; | |
3143 | facet->nf_flow.output_iface = ctx.nf_output_iface; | |
3144 | facet->may_install = ctx.may_set_up_flow; | |
75a75043 BP |
3145 | facet->has_learn = ctx.has_learn; |
3146 | facet->has_normal = ctx.has_normal; | |
abe529af BP |
3147 | if (actions_changed) { |
3148 | free(facet->actions); | |
3149 | facet->actions_len = odp_actions->size; | |
3150 | facet->actions = xmemdup(odp_actions->data, odp_actions->size); | |
3151 | } | |
3152 | if (facet->rule != new_rule) { | |
3153 | COVERAGE_INC(facet_changed_rule); | |
3154 | list_remove(&facet->list_node); | |
3155 | list_push_back(&new_rule->facets, &facet->list_node); | |
3156 | facet->rule = new_rule; | |
3157 | facet->used = new_rule->up.created; | |
3158 | facet->rs_used = facet->used; | |
3159 | } | |
3160 | ||
3161 | ofpbuf_delete(odp_actions); | |
3162 | ||
3163 | return true; | |
3164 | } | |
3165 | ||
3166 | /* Updates 'facet''s used time. Caller is responsible for calling | |
3167 | * facet_push_stats() to update the flows which 'facet' resubmits into. */ | |
3168 | static void | |
3169 | facet_update_time(struct ofproto_dpif *ofproto, struct facet *facet, | |
3170 | long long int used) | |
3171 | { | |
3172 | if (used > facet->used) { | |
3173 | facet->used = used; | |
3174 | if (used > facet->rule->used) { | |
3175 | facet->rule->used = used; | |
3176 | } | |
3177 | netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used); | |
3178 | } | |
3179 | } | |
3180 | ||
3181 | /* Folds the statistics from 'stats' into the counters in 'facet'. | |
3182 | * | |
3183 | * Because of the meaning of a facet's counters, it only makes sense to do this | |
3184 | * if 'stats' are not tracked in the datapath, that is, if 'stats' represents a | |
3185 | * packet that was sent by hand or if it represents statistics that have been | |
3186 | * cleared out of the datapath. */ | |
3187 | static void | |
3188 | facet_update_stats(struct ofproto_dpif *ofproto, struct facet *facet, | |
3189 | const struct dpif_flow_stats *stats) | |
3190 | { | |
3191 | if (stats->n_packets || stats->used > facet->used) { | |
3192 | facet_update_time(ofproto, facet, stats->used); | |
3193 | facet->packet_count += stats->n_packets; | |
3194 | facet->byte_count += stats->n_bytes; | |
3195 | facet_push_stats(facet); | |
3196 | netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags); | |
3197 | } | |
3198 | } | |
3199 | ||
bbb5d219 EJ |
3200 | static void |
3201 | facet_reset_counters(struct facet *facet) | |
3202 | { | |
3203 | facet->packet_count = 0; | |
3204 | facet->byte_count = 0; | |
3205 | facet->rs_packet_count = 0; | |
3206 | facet->rs_byte_count = 0; | |
3207 | facet->accounted_bytes = 0; | |
3208 | } | |
3209 | ||
abe529af BP |
3210 | static void |
3211 | facet_push_stats(struct facet *facet) | |
3212 | { | |
3213 | uint64_t rs_packets, rs_bytes; | |
3214 | ||
3215 | assert(facet->packet_count >= facet->rs_packet_count); | |
3216 | assert(facet->byte_count >= facet->rs_byte_count); | |
3217 | assert(facet->used >= facet->rs_used); | |
3218 | ||
3219 | rs_packets = facet->packet_count - facet->rs_packet_count; | |
3220 | rs_bytes = facet->byte_count - facet->rs_byte_count; | |
3221 | ||
3222 | if (rs_packets || rs_bytes || facet->used > facet->rs_used) { | |
3223 | facet->rs_packet_count = facet->packet_count; | |
3224 | facet->rs_byte_count = facet->byte_count; | |
3225 | facet->rs_used = facet->used; | |
3226 | ||
3227 | flow_push_stats(facet->rule, &facet->flow, | |
3228 | rs_packets, rs_bytes, facet->used); | |
3229 | } | |
3230 | } | |
3231 | ||
3232 | struct ofproto_push { | |
3233 | struct action_xlate_ctx ctx; | |
3234 | uint64_t packets; | |
3235 | uint64_t bytes; | |
3236 | long long int used; | |
3237 | }; | |
3238 | ||
3239 | static void | |
3240 | push_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
3241 | { | |
3242 | struct ofproto_push *push = CONTAINER_OF(ctx, struct ofproto_push, ctx); | |
3243 | ||
3244 | if (rule) { | |
3245 | rule->packet_count += push->packets; | |
3246 | rule->byte_count += push->bytes; | |
3247 | rule->used = MAX(push->used, rule->used); | |
3248 | } | |
3249 | } | |
3250 | ||
3251 | /* Pushes flow statistics to the rules which 'flow' resubmits into given | |
3252 | * 'rule''s actions. */ | |
3253 | static void | |
3254 | flow_push_stats(const struct rule_dpif *rule, | |
3255 | struct flow *flow, uint64_t packets, uint64_t bytes, | |
3256 | long long int used) | |
3257 | { | |
3258 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3259 | struct ofproto_push push; | |
3260 | ||
3261 | push.packets = packets; | |
3262 | push.bytes = bytes; | |
3263 | push.used = used; | |
3264 | ||
3265 | action_xlate_ctx_init(&push.ctx, ofproto, flow, NULL); | |
3266 | push.ctx.resubmit_hook = push_resubmit; | |
3267 | ofpbuf_delete(xlate_actions(&push.ctx, | |
3268 | rule->up.actions, rule->up.n_actions)); | |
3269 | } | |
3270 | \f | |
3271 | /* Rules. */ | |
3272 | ||
3273 | static struct rule_dpif * | |
29901626 BP |
3274 | rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow, |
3275 | uint8_t table_id) | |
abe529af | 3276 | { |
7257b535 BP |
3277 | struct cls_rule *cls_rule; |
3278 | struct classifier *cls; | |
3279 | ||
9cdaaebe BP |
3280 | if (table_id >= N_TABLES) { |
3281 | return NULL; | |
3282 | } | |
3283 | ||
7257b535 | 3284 | cls = &ofproto->up.tables[table_id]; |
eadef313 | 3285 | if (flow->nw_frag & FLOW_NW_FRAG_ANY |
7257b535 BP |
3286 | && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) { |
3287 | /* For OFPC_NORMAL frag_handling, we must pretend that transport ports | |
3288 | * are unavailable. */ | |
3289 | struct flow ofpc_normal_flow = *flow; | |
3290 | ofpc_normal_flow.tp_src = htons(0); | |
3291 | ofpc_normal_flow.tp_dst = htons(0); | |
3292 | cls_rule = classifier_lookup(cls, &ofpc_normal_flow); | |
3293 | } else { | |
3294 | cls_rule = classifier_lookup(cls, flow); | |
3295 | } | |
3296 | return rule_dpif_cast(rule_from_cls_rule(cls_rule)); | |
abe529af BP |
3297 | } |
3298 | ||
7ee20df1 BP |
3299 | static void |
3300 | complete_operation(struct rule_dpif *rule) | |
3301 | { | |
3302 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3303 | ||
54a9cbc9 | 3304 | rule_invalidate(rule); |
7ee20df1 BP |
3305 | if (clogged) { |
3306 | struct dpif_completion *c = xmalloc(sizeof *c); | |
3307 | c->op = rule->up.pending; | |
3308 | list_push_back(&ofproto->completions, &c->list_node); | |
3309 | } else { | |
3310 | ofoperation_complete(rule->up.pending, 0); | |
3311 | } | |
3312 | } | |
3313 | ||
abe529af BP |
3314 | static struct rule * |
3315 | rule_alloc(void) | |
3316 | { | |
3317 | struct rule_dpif *rule = xmalloc(sizeof *rule); | |
3318 | return &rule->up; | |
3319 | } | |
3320 | ||
3321 | static void | |
3322 | rule_dealloc(struct rule *rule_) | |
3323 | { | |
3324 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3325 | free(rule); | |
3326 | } | |
3327 | ||
3328 | static int | |
3329 | rule_construct(struct rule *rule_) | |
3330 | { | |
3331 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3332 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
7ee20df1 | 3333 | struct rule_dpif *victim; |
54a9cbc9 | 3334 | uint8_t table_id; |
5bf0e941 BP |
3335 | int error; |
3336 | ||
3337 | error = validate_actions(rule->up.actions, rule->up.n_actions, | |
3338 | &rule->up.cr.flow, ofproto->max_ports); | |
3339 | if (error) { | |
3340 | return error; | |
3341 | } | |
abe529af BP |
3342 | |
3343 | rule->used = rule->up.created; | |
3344 | rule->packet_count = 0; | |
3345 | rule->byte_count = 0; | |
abe529af | 3346 | |
7ee20df1 BP |
3347 | victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending)); |
3348 | if (victim && !list_is_empty(&victim->facets)) { | |
3349 | struct facet *facet; | |
3350 | ||
3351 | rule->facets = victim->facets; | |
3352 | list_moved(&rule->facets); | |
3353 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
bbb5d219 EJ |
3354 | /* XXX: We're only clearing our local counters here. It's possible |
3355 | * that quite a few packets are unaccounted for in the datapath | |
3356 | * statistics. These will be accounted to the new rule instead of | |
3357 | * cleared as required. This could be fixed by clearing out the | |
3358 | * datapath statistics for this facet, but currently it doesn't | |
3359 | * seem worth it. */ | |
3360 | facet_reset_counters(facet); | |
7ee20df1 BP |
3361 | facet->rule = rule; |
3362 | } | |
3363 | } else { | |
3364 | /* Must avoid list_moved() in this case. */ | |
3365 | list_init(&rule->facets); | |
3366 | } | |
abe529af | 3367 | |
54a9cbc9 BP |
3368 | table_id = rule->up.table_id; |
3369 | rule->tag = (victim ? victim->tag | |
3370 | : table_id == 0 ? 0 | |
3371 | : rule_calculate_tag(&rule->up.cr.flow, &rule->up.cr.wc, | |
3372 | ofproto->tables[table_id].basis)); | |
3373 | ||
7ee20df1 | 3374 | complete_operation(rule); |
abe529af BP |
3375 | return 0; |
3376 | } | |
3377 | ||
3378 | static void | |
3379 | rule_destruct(struct rule *rule_) | |
3380 | { | |
3381 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3382 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3383 | struct facet *facet, *next_facet; | |
3384 | ||
abe529af BP |
3385 | LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) { |
3386 | facet_revalidate(ofproto, facet); | |
3387 | } | |
7ee20df1 BP |
3388 | |
3389 | complete_operation(rule); | |
abe529af BP |
3390 | } |
3391 | ||
3392 | static void | |
3393 | rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes) | |
3394 | { | |
3395 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3396 | struct facet *facet; | |
3397 | ||
3398 | /* Start from historical data for 'rule' itself that are no longer tracked | |
3399 | * in facets. This counts, for example, facets that have expired. */ | |
3400 | *packets = rule->packet_count; | |
3401 | *bytes = rule->byte_count; | |
3402 | ||
3403 | /* Add any statistics that are tracked by facets. This includes | |
3404 | * statistical data recently updated by ofproto_update_stats() as well as | |
3405 | * stats for packets that were executed "by hand" via dpif_execute(). */ | |
3406 | LIST_FOR_EACH (facet, list_node, &rule->facets) { | |
3407 | *packets += facet->packet_count; | |
3408 | *bytes += facet->byte_count; | |
3409 | } | |
3410 | } | |
3411 | ||
5bf0e941 | 3412 | static int |
abe529af BP |
3413 | rule_execute(struct rule *rule_, struct flow *flow, struct ofpbuf *packet) |
3414 | { | |
3415 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3416 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3417 | struct action_xlate_ctx ctx; | |
3418 | struct ofpbuf *odp_actions; | |
3419 | struct facet *facet; | |
3420 | size_t size; | |
3421 | ||
3422 | /* First look for a related facet. If we find one, account it to that. */ | |
3423 | facet = facet_lookup_valid(ofproto, flow); | |
3424 | if (facet && facet->rule == rule) { | |
eea109bb BP |
3425 | if (!facet->may_install) { |
3426 | facet_make_actions(ofproto, facet, packet); | |
3427 | } | |
abe529af | 3428 | facet_execute(ofproto, facet, packet); |
5bf0e941 | 3429 | return 0; |
abe529af BP |
3430 | } |
3431 | ||
3432 | /* Otherwise, if 'rule' is in fact the correct rule for 'packet', then | |
3433 | * create a new facet for it and use that. */ | |
29901626 | 3434 | if (rule_dpif_lookup(ofproto, flow, 0) == rule) { |
f3827897 BP |
3435 | facet = facet_create(rule, flow); |
3436 | facet_make_actions(ofproto, facet, packet); | |
abe529af BP |
3437 | facet_execute(ofproto, facet, packet); |
3438 | facet_install(ofproto, facet, true); | |
5bf0e941 | 3439 | return 0; |
abe529af BP |
3440 | } |
3441 | ||
3442 | /* We can't account anything to a facet. If we were to try, then that | |
3443 | * facet would have a non-matching rule, busting our invariants. */ | |
3444 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
3445 | odp_actions = xlate_actions(&ctx, rule->up.actions, rule->up.n_actions); | |
3446 | size = packet->size; | |
3447 | if (execute_odp_actions(ofproto, flow, odp_actions->data, | |
3448 | odp_actions->size, packet)) { | |
3449 | rule->used = time_msec(); | |
3450 | rule->packet_count++; | |
3451 | rule->byte_count += size; | |
3452 | flow_push_stats(rule, flow, 1, size, rule->used); | |
3453 | } | |
3454 | ofpbuf_delete(odp_actions); | |
5bf0e941 BP |
3455 | |
3456 | return 0; | |
abe529af BP |
3457 | } |
3458 | ||
7ee20df1 BP |
3459 | static void |
3460 | rule_modify_actions(struct rule *rule_) | |
abe529af BP |
3461 | { |
3462 | struct rule_dpif *rule = rule_dpif_cast(rule_); | |
3463 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
3464 | int error; | |
3465 | ||
7ee20df1 BP |
3466 | error = validate_actions(rule->up.actions, rule->up.n_actions, |
3467 | &rule->up.cr.flow, ofproto->max_ports); | |
3468 | if (error) { | |
3469 | ofoperation_complete(rule->up.pending, error); | |
3470 | return; | |
abe529af | 3471 | } |
7ee20df1 BP |
3472 | |
3473 | complete_operation(rule); | |
abe529af BP |
3474 | } |
3475 | \f | |
b47e2a82 | 3476 | /* Sends 'packet' out of port 'odp_port' within 'ofproto'. |
abe529af BP |
3477 | * Returns 0 if successful, otherwise a positive errno value. */ |
3478 | static int | |
b2fda3ef | 3479 | send_packet(struct ofproto_dpif *ofproto, uint32_t odp_port, |
abe529af BP |
3480 | const struct ofpbuf *packet) |
3481 | { | |
80e5eed9 BP |
3482 | struct ofpbuf key, odp_actions; |
3483 | struct odputil_keybuf keybuf; | |
3484 | struct flow flow; | |
abe529af BP |
3485 | int error; |
3486 | ||
abff858b | 3487 | flow_extract((struct ofpbuf *) packet, 0, 0, 0, &flow); |
80e5eed9 BP |
3488 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); |
3489 | odp_flow_key_from_flow(&key, &flow); | |
3490 | ||
abe529af | 3491 | ofpbuf_init(&odp_actions, 32); |
6ff686f2 PS |
3492 | compose_sflow_action(ofproto, &odp_actions, &flow, odp_port); |
3493 | ||
df2c07f4 | 3494 | nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); |
80e5eed9 BP |
3495 | error = dpif_execute(ofproto->dpif, |
3496 | key.data, key.size, | |
3497 | odp_actions.data, odp_actions.size, | |
abe529af BP |
3498 | packet); |
3499 | ofpbuf_uninit(&odp_actions); | |
3500 | ||
3501 | if (error) { | |
3502 | VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)", | |
3503 | ofproto->up.name, odp_port, strerror(error)); | |
3504 | } | |
3505 | return error; | |
3506 | } | |
3507 | \f | |
df2c07f4 | 3508 | /* OpenFlow to datapath action translation. */ |
abe529af BP |
3509 | |
3510 | static void do_xlate_actions(const union ofp_action *in, size_t n_in, | |
3511 | struct action_xlate_ctx *ctx); | |
4cd78906 | 3512 | static void xlate_normal(struct action_xlate_ctx *); |
abe529af | 3513 | |
98403001 BP |
3514 | static size_t |
3515 | put_userspace_action(const struct ofproto_dpif *ofproto, | |
3516 | struct ofpbuf *odp_actions, | |
3517 | const struct flow *flow, | |
3518 | const struct user_action_cookie *cookie) | |
3519 | { | |
98403001 BP |
3520 | uint32_t pid; |
3521 | ||
3522 | pid = dpif_port_get_pid(ofproto->dpif, | |
3523 | ofp_port_to_odp_port(flow->in_port)); | |
3524 | ||
39db78a0 | 3525 | return odp_put_userspace_action(pid, cookie, odp_actions); |
98403001 BP |
3526 | } |
3527 | ||
6ff686f2 PS |
3528 | /* Compose SAMPLE action for sFlow. */ |
3529 | static size_t | |
3530 | compose_sflow_action(const struct ofproto_dpif *ofproto, | |
3531 | struct ofpbuf *odp_actions, | |
3532 | const struct flow *flow, | |
3533 | uint32_t odp_port) | |
3534 | { | |
3535 | uint32_t port_ifindex; | |
3536 | uint32_t probability; | |
98403001 | 3537 | struct user_action_cookie cookie; |
6ff686f2 | 3538 | size_t sample_offset, actions_offset; |
98403001 | 3539 | int cookie_offset, n_output; |
6ff686f2 PS |
3540 | |
3541 | if (!ofproto->sflow || flow->in_port == OFPP_NONE) { | |
3542 | return 0; | |
3543 | } | |
3544 | ||
3545 | if (odp_port == OVSP_NONE) { | |
3546 | port_ifindex = 0; | |
3547 | n_output = 0; | |
3548 | } else { | |
3549 | port_ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port); | |
3550 | n_output = 1; | |
3551 | } | |
3552 | ||
3553 | sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE); | |
3554 | ||
3555 | /* Number of packets out of UINT_MAX to sample. */ | |
3556 | probability = dpif_sflow_get_probability(ofproto->sflow); | |
3557 | nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability); | |
3558 | ||
3559 | actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS); | |
3560 | ||
98403001 BP |
3561 | cookie.type = USER_ACTION_COOKIE_SFLOW; |
3562 | cookie.data = port_ifindex; | |
3563 | cookie.n_output = n_output; | |
3564 | cookie.vlan_tci = 0; | |
3565 | cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie); | |
6ff686f2 PS |
3566 | |
3567 | nl_msg_end_nested(odp_actions, actions_offset); | |
3568 | nl_msg_end_nested(odp_actions, sample_offset); | |
98403001 | 3569 | return cookie_offset; |
6ff686f2 PS |
3570 | } |
3571 | ||
3572 | /* SAMPLE action must be first action in any given list of actions. | |
3573 | * At this point we do not have all information required to build it. So try to | |
3574 | * build sample action as complete as possible. */ | |
3575 | static void | |
3576 | add_sflow_action(struct action_xlate_ctx *ctx) | |
3577 | { | |
3578 | ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto, | |
3579 | ctx->odp_actions, | |
3580 | &ctx->flow, OVSP_NONE); | |
3581 | ctx->sflow_odp_port = 0; | |
3582 | ctx->sflow_n_outputs = 0; | |
3583 | } | |
3584 | ||
3585 | /* Fix SAMPLE action according to data collected while composing ODP actions. | |
3586 | * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested | |
3587 | * USERSPACE action's user-cookie which is required for sflow. */ | |
3588 | static void | |
3589 | fix_sflow_action(struct action_xlate_ctx *ctx) | |
3590 | { | |
3591 | const struct flow *base = &ctx->base_flow; | |
3592 | struct user_action_cookie *cookie; | |
3593 | ||
3594 | if (!ctx->user_cookie_offset) { | |
3595 | return; | |
3596 | } | |
3597 | ||
3598 | cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset, | |
3599 | sizeof(*cookie)); | |
3600 | assert(cookie != NULL); | |
3601 | assert(cookie->type == USER_ACTION_COOKIE_SFLOW); | |
3602 | ||
3603 | if (ctx->sflow_n_outputs) { | |
3604 | cookie->data = dpif_sflow_odp_port_to_ifindex(ctx->ofproto->sflow, | |
3605 | ctx->sflow_odp_port); | |
3606 | } | |
3607 | if (ctx->sflow_n_outputs >= 255) { | |
3608 | cookie->n_output = 255; | |
3609 | } else { | |
3610 | cookie->n_output = ctx->sflow_n_outputs; | |
3611 | } | |
3612 | cookie->vlan_tci = base->vlan_tci; | |
3613 | } | |
3614 | ||
823518f1 | 3615 | static void |
fea393b1 BP |
3616 | commit_set_action(struct ofpbuf *odp_actions, enum ovs_key_attr key_type, |
3617 | const void *key, size_t key_size) | |
823518f1 | 3618 | { |
fea393b1 | 3619 | size_t offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SET); |
4edb9ae9 PS |
3620 | nl_msg_put_unspec(odp_actions, key_type, key, key_size); |
3621 | nl_msg_end_nested(odp_actions, offset); | |
3622 | } | |
3623 | ||
3624 | static void | |
3625 | commit_set_tun_id_action(const struct flow *flow, struct flow *base, | |
3626 | struct ofpbuf *odp_actions) | |
3627 | { | |
3628 | if (base->tun_id == flow->tun_id) { | |
3629 | return; | |
823518f1 | 3630 | } |
4edb9ae9 PS |
3631 | base->tun_id = flow->tun_id; |
3632 | ||
fea393b1 BP |
3633 | commit_set_action(odp_actions, OVS_KEY_ATTR_TUN_ID, |
3634 | &base->tun_id, sizeof(base->tun_id)); | |
823518f1 BP |
3635 | } |
3636 | ||
b3e9b2ed | 3637 | static void |
4edb9ae9 PS |
3638 | commit_set_ether_addr_action(const struct flow *flow, struct flow *base, |
3639 | struct ofpbuf *odp_actions) | |
3640 | { | |
3641 | struct ovs_key_ethernet eth_key; | |
3642 | ||
3643 | if (eth_addr_equals(base->dl_src, flow->dl_src) && | |
3644 | eth_addr_equals(base->dl_dst, flow->dl_dst)) { | |
3645 | return; | |
3646 | } | |
3647 | ||
3648 | memcpy(base->dl_src, flow->dl_src, ETH_ADDR_LEN); | |
3649 | memcpy(base->dl_dst, flow->dl_dst, ETH_ADDR_LEN); | |
3650 | ||
3651 | memcpy(eth_key.eth_src, base->dl_src, ETH_ADDR_LEN); | |
3652 | memcpy(eth_key.eth_dst, base->dl_dst, ETH_ADDR_LEN); | |
3653 | ||
fea393b1 BP |
3654 | commit_set_action(odp_actions, OVS_KEY_ATTR_ETHERNET, |
3655 | ð_key, sizeof(eth_key)); | |
4edb9ae9 PS |
3656 | } |
3657 | ||
3658 | static void | |
3659 | commit_vlan_action(struct action_xlate_ctx *ctx, ovs_be16 new_tci) | |
b3e9b2ed | 3660 | { |
b3e9b2ed | 3661 | struct flow *base = &ctx->base_flow; |
b3e9b2ed | 3662 | |
4edb9ae9 PS |
3663 | if (base->vlan_tci == new_tci) { |
3664 | return; | |
b3e9b2ed EJ |
3665 | } |
3666 | ||
4edb9ae9 | 3667 | if (base->vlan_tci & htons(VLAN_CFI)) { |
fea393b1 | 3668 | nl_msg_put_flag(ctx->odp_actions, OVS_ACTION_ATTR_POP_VLAN); |
b3e9b2ed EJ |
3669 | } |
3670 | ||
4edb9ae9 | 3671 | if (new_tci & htons(VLAN_CFI)) { |
fea393b1 | 3672 | struct ovs_action_push_vlan vlan; |
4edb9ae9 | 3673 | |
fea393b1 | 3674 | vlan.vlan_tpid = htons(ETH_TYPE_VLAN); |
8ddc056d | 3675 | vlan.vlan_tci = new_tci; |
fea393b1 BP |
3676 | nl_msg_put_unspec(ctx->odp_actions, OVS_ACTION_ATTR_PUSH_VLAN, |
3677 | &vlan, sizeof vlan); | |
b3e9b2ed | 3678 | } |
4edb9ae9 PS |
3679 | base->vlan_tci = new_tci; |
3680 | } | |
b3e9b2ed | 3681 | |
4edb9ae9 PS |
3682 | static void |
3683 | commit_set_nw_action(const struct flow *flow, struct flow *base, | |
3684 | struct ofpbuf *odp_actions) | |
3685 | { | |
3686 | struct ovs_key_ipv4 ipv4_key; | |
3687 | ||
3688 | if (base->dl_type != htons(ETH_TYPE_IP) || | |
3689 | !base->nw_src || !base->nw_dst) { | |
3690 | return; | |
3691 | } | |
3692 | ||
3693 | if (base->nw_src == flow->nw_src && | |
3694 | base->nw_dst == flow->nw_dst && | |
eadef313 | 3695 | base->nw_tos == flow->nw_tos && |
a61680c6 | 3696 | base->nw_ttl == flow->nw_ttl && |
eadef313 | 3697 | base->nw_frag == flow->nw_frag) { |
4edb9ae9 | 3698 | return; |
150a9f15 BP |
3699 | } |
3700 | ||
4edb9ae9 PS |
3701 | ipv4_key.ipv4_src = base->nw_src = flow->nw_src; |
3702 | ipv4_key.ipv4_dst = base->nw_dst = flow->nw_dst; | |
4edb9ae9 | 3703 | ipv4_key.ipv4_proto = base->nw_proto; |
eadef313 | 3704 | ipv4_key.ipv4_tos = flow->nw_tos; |
a61680c6 | 3705 | ipv4_key.ipv4_ttl = flow->nw_ttl; |
eadef313 JP |
3706 | ipv4_key.ipv4_frag = (base->nw_frag == 0 ? OVS_FRAG_TYPE_NONE |
3707 | : base->nw_frag == FLOW_NW_FRAG_ANY | |
3708 | ? OVS_FRAG_TYPE_FIRST : OVS_FRAG_TYPE_LATER); | |
4edb9ae9 | 3709 | |
fea393b1 BP |
3710 | commit_set_action(odp_actions, OVS_KEY_ATTR_IPV4, |
3711 | &ipv4_key, sizeof(ipv4_key)); | |
4edb9ae9 PS |
3712 | } |
3713 | ||
3714 | static void | |
3715 | commit_set_port_action(const struct flow *flow, struct flow *base, | |
3716 | struct ofpbuf *odp_actions) | |
3717 | { | |
3718 | if (!base->tp_src || !base->tp_dst) { | |
3719 | return; | |
b3e9b2ed EJ |
3720 | } |
3721 | ||
4edb9ae9 PS |
3722 | if (base->tp_src == flow->tp_src && |
3723 | base->tp_dst == flow->tp_dst) { | |
3724 | return; | |
b3e9b2ed EJ |
3725 | } |
3726 | ||
4edb9ae9 PS |
3727 | if (flow->nw_proto == IPPROTO_TCP) { |
3728 | struct ovs_key_tcp port_key; | |
3729 | ||
3730 | port_key.tcp_src = base->tp_src = flow->tp_src; | |
3731 | port_key.tcp_dst = base->tp_dst = flow->tp_dst; | |
3732 | ||
fea393b1 BP |
3733 | commit_set_action(odp_actions, OVS_KEY_ATTR_TCP, |
3734 | &port_key, sizeof(port_key)); | |
4edb9ae9 PS |
3735 | |
3736 | } else if (flow->nw_proto == IPPROTO_UDP) { | |
3737 | struct ovs_key_udp port_key; | |
3738 | ||
3739 | port_key.udp_src = base->tp_src = flow->tp_src; | |
3740 | port_key.udp_dst = base->tp_dst = flow->tp_dst; | |
3741 | ||
fea393b1 BP |
3742 | commit_set_action(odp_actions, OVS_KEY_ATTR_UDP, |
3743 | &port_key, sizeof(port_key)); | |
b3e9b2ed | 3744 | } |
4edb9ae9 | 3745 | } |
b3e9b2ed | 3746 | |
4edb9ae9 | 3747 | static void |
abff858b PS |
3748 | commit_set_priority_action(const struct flow *flow, struct flow *base, |
3749 | struct ofpbuf *odp_actions) | |
4edb9ae9 | 3750 | { |
abff858b | 3751 | if (base->priority == flow->priority) { |
4edb9ae9 | 3752 | return; |
b3e9b2ed | 3753 | } |
abff858b | 3754 | base->priority = flow->priority; |
b3e9b2ed | 3755 | |
fea393b1 BP |
3756 | commit_set_action(odp_actions, OVS_KEY_ATTR_PRIORITY, |
3757 | &base->priority, sizeof(base->priority)); | |
4edb9ae9 PS |
3758 | } |
3759 | ||
3760 | static void | |
3761 | commit_odp_actions(struct action_xlate_ctx *ctx) | |
3762 | { | |
3763 | const struct flow *flow = &ctx->flow; | |
3764 | struct flow *base = &ctx->base_flow; | |
3765 | struct ofpbuf *odp_actions = ctx->odp_actions; | |
3766 | ||
3767 | commit_set_tun_id_action(flow, base, odp_actions); | |
3768 | commit_set_ether_addr_action(flow, base, odp_actions); | |
3769 | commit_vlan_action(ctx, flow->vlan_tci); | |
3770 | commit_set_nw_action(flow, base, odp_actions); | |
3771 | commit_set_port_action(flow, base, odp_actions); | |
abff858b | 3772 | commit_set_priority_action(flow, base, odp_actions); |
b3e9b2ed EJ |
3773 | } |
3774 | ||
6ff686f2 PS |
3775 | static void |
3776 | compose_output_action(struct action_xlate_ctx *ctx, uint16_t odp_port) | |
3777 | { | |
3778 | nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port); | |
3779 | ctx->sflow_odp_port = odp_port; | |
3780 | ctx->sflow_n_outputs++; | |
3781 | } | |
3782 | ||
abe529af BP |
3783 | static void |
3784 | add_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port) | |
3785 | { | |
3786 | const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port); | |
3787 | uint16_t odp_port = ofp_port_to_odp_port(ofp_port); | |
3788 | ||
3789 | if (ofport) { | |
21f7563c JP |
3790 | if (ofport->up.opp.config & htonl(OFPPC_NO_FWD) |
3791 | || !stp_forward_in_state(ofport->stp_state)) { | |
abe529af BP |
3792 | /* Forwarding disabled on port. */ |
3793 | return; | |
3794 | } | |
3795 | } else { | |
3796 | /* | |
3797 | * We don't have an ofport record for this port, but it doesn't hurt to | |
3798 | * allow forwarding to it anyhow. Maybe such a port will appear later | |
3799 | * and we're pre-populating the flow table. | |
3800 | */ | |
3801 | } | |
3802 | ||
b3e9b2ed | 3803 | commit_odp_actions(ctx); |
6ff686f2 | 3804 | compose_output_action(ctx, odp_port); |
abe529af BP |
3805 | ctx->nf_output_iface = ofp_port; |
3806 | } | |
3807 | ||
3808 | static void | |
29901626 BP |
3809 | xlate_table_action(struct action_xlate_ctx *ctx, |
3810 | uint16_t in_port, uint8_t table_id) | |
abe529af BP |
3811 | { |
3812 | if (ctx->recurse < MAX_RESUBMIT_RECURSION) { | |
54a9cbc9 | 3813 | struct ofproto_dpif *ofproto = ctx->ofproto; |
abe529af BP |
3814 | struct rule_dpif *rule; |
3815 | uint16_t old_in_port; | |
29901626 BP |
3816 | uint8_t old_table_id; |
3817 | ||
3818 | old_table_id = ctx->table_id; | |
3819 | ctx->table_id = table_id; | |
abe529af | 3820 | |
54a9cbc9 | 3821 | /* Look up a flow with 'in_port' as the input port. */ |
abe529af BP |
3822 | old_in_port = ctx->flow.in_port; |
3823 | ctx->flow.in_port = in_port; | |
54a9cbc9 BP |
3824 | rule = rule_dpif_lookup(ofproto, &ctx->flow, table_id); |
3825 | ||
3826 | /* Tag the flow. */ | |
3827 | if (table_id > 0 && table_id < N_TABLES) { | |
3828 | struct table_dpif *table = &ofproto->tables[table_id]; | |
3829 | if (table->other_table) { | |
3830 | ctx->tags |= (rule | |
3831 | ? rule->tag | |
3832 | : rule_calculate_tag(&ctx->flow, | |
3833 | &table->other_table->wc, | |
3834 | table->basis)); | |
3835 | } | |
3836 | } | |
3837 | ||
3838 | /* Restore the original input port. Otherwise OFPP_NORMAL and | |
3839 | * OFPP_IN_PORT will have surprising behavior. */ | |
abe529af BP |
3840 | ctx->flow.in_port = old_in_port; |
3841 | ||
3842 | if (ctx->resubmit_hook) { | |
3843 | ctx->resubmit_hook(ctx, rule); | |
3844 | } | |
3845 | ||
3846 | if (rule) { | |
3847 | ctx->recurse++; | |
3848 | do_xlate_actions(rule->up.actions, rule->up.n_actions, ctx); | |
3849 | ctx->recurse--; | |
3850 | } | |
29901626 BP |
3851 | |
3852 | ctx->table_id = old_table_id; | |
abe529af BP |
3853 | } else { |
3854 | static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
3855 | ||
29901626 | 3856 | VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times", |
abe529af BP |
3857 | MAX_RESUBMIT_RECURSION); |
3858 | } | |
3859 | } | |
3860 | ||
29901626 BP |
3861 | static void |
3862 | xlate_resubmit_table(struct action_xlate_ctx *ctx, | |
3863 | const struct nx_action_resubmit *nar) | |
3864 | { | |
3865 | uint16_t in_port; | |
3866 | uint8_t table_id; | |
3867 | ||
3868 | in_port = (nar->in_port == htons(OFPP_IN_PORT) | |
3869 | ? ctx->flow.in_port | |
3870 | : ntohs(nar->in_port)); | |
3871 | table_id = nar->table == 255 ? ctx->table_id : nar->table; | |
3872 | ||
3873 | xlate_table_action(ctx, in_port, table_id); | |
3874 | } | |
3875 | ||
abe529af | 3876 | static void |
b3e9b2ed | 3877 | flood_packets(struct action_xlate_ctx *ctx, ovs_be32 mask) |
abe529af BP |
3878 | { |
3879 | struct ofport_dpif *ofport; | |
3880 | ||
b3e9b2ed EJ |
3881 | commit_odp_actions(ctx); |
3882 | HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) { | |
abe529af | 3883 | uint16_t ofp_port = ofport->up.ofp_port; |
21f7563c JP |
3884 | if (ofp_port != ctx->flow.in_port |
3885 | && !(ofport->up.opp.config & mask) | |
3886 | && stp_forward_in_state(ofport->stp_state)) { | |
6ff686f2 | 3887 | compose_output_action(ctx, ofport->odp_port); |
abe529af BP |
3888 | } |
3889 | } | |
b3e9b2ed EJ |
3890 | |
3891 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
abe529af BP |
3892 | } |
3893 | ||
6ff686f2 | 3894 | static void |
98403001 | 3895 | compose_controller_action(struct action_xlate_ctx *ctx, int len) |
6ff686f2 PS |
3896 | { |
3897 | struct user_action_cookie cookie; | |
3898 | ||
3899 | cookie.type = USER_ACTION_COOKIE_CONTROLLER; | |
3900 | cookie.data = len; | |
3901 | cookie.n_output = 0; | |
3902 | cookie.vlan_tci = 0; | |
98403001 | 3903 | put_userspace_action(ctx->ofproto, ctx->odp_actions, &ctx->flow, &cookie); |
6ff686f2 PS |
3904 | } |
3905 | ||
abe529af BP |
3906 | static void |
3907 | xlate_output_action__(struct action_xlate_ctx *ctx, | |
3908 | uint16_t port, uint16_t max_len) | |
3909 | { | |
3910 | uint16_t prev_nf_output_iface = ctx->nf_output_iface; | |
3911 | ||
3912 | ctx->nf_output_iface = NF_OUT_DROP; | |
3913 | ||
3914 | switch (port) { | |
3915 | case OFPP_IN_PORT: | |
3916 | add_output_action(ctx, ctx->flow.in_port); | |
3917 | break; | |
3918 | case OFPP_TABLE: | |
29901626 | 3919 | xlate_table_action(ctx, ctx->flow.in_port, ctx->table_id); |
abe529af BP |
3920 | break; |
3921 | case OFPP_NORMAL: | |
3922 | xlate_normal(ctx); | |
3923 | break; | |
3924 | case OFPP_FLOOD: | |
b3e9b2ed | 3925 | flood_packets(ctx, htonl(OFPPC_NO_FLOOD)); |
abe529af BP |
3926 | break; |
3927 | case OFPP_ALL: | |
b3e9b2ed | 3928 | flood_packets(ctx, htonl(0)); |
abe529af BP |
3929 | break; |
3930 | case OFPP_CONTROLLER: | |
b3e9b2ed | 3931 | commit_odp_actions(ctx); |
98403001 | 3932 | compose_controller_action(ctx, max_len); |
abe529af BP |
3933 | break; |
3934 | case OFPP_LOCAL: | |
3935 | add_output_action(ctx, OFPP_LOCAL); | |
3936 | break; | |
e81d2933 EJ |
3937 | case OFPP_NONE: |
3938 | break; | |
abe529af BP |
3939 | default: |
3940 | if (port != ctx->flow.in_port) { | |
3941 | add_output_action(ctx, port); | |
3942 | } | |
3943 | break; | |
3944 | } | |
3945 | ||
3946 | if (prev_nf_output_iface == NF_OUT_FLOOD) { | |
3947 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
3948 | } else if (ctx->nf_output_iface == NF_OUT_DROP) { | |
3949 | ctx->nf_output_iface = prev_nf_output_iface; | |
3950 | } else if (prev_nf_output_iface != NF_OUT_DROP && | |
3951 | ctx->nf_output_iface != NF_OUT_FLOOD) { | |
3952 | ctx->nf_output_iface = NF_OUT_MULTI; | |
3953 | } | |
3954 | } | |
3955 | ||
f694937d EJ |
3956 | static void |
3957 | xlate_output_reg_action(struct action_xlate_ctx *ctx, | |
3958 | const struct nx_action_output_reg *naor) | |
3959 | { | |
3960 | uint64_t ofp_port; | |
3961 | ||
3962 | ofp_port = nxm_read_field_bits(naor->src, naor->ofs_nbits, &ctx->flow); | |
3963 | ||
3964 | if (ofp_port <= UINT16_MAX) { | |
3965 | xlate_output_action__(ctx, ofp_port, ntohs(naor->max_len)); | |
3966 | } | |
3967 | } | |
3968 | ||
abe529af BP |
3969 | static void |
3970 | xlate_output_action(struct action_xlate_ctx *ctx, | |
3971 | const struct ofp_action_output *oao) | |
3972 | { | |
3973 | xlate_output_action__(ctx, ntohs(oao->port), ntohs(oao->max_len)); | |
3974 | } | |
3975 | ||
abe529af BP |
3976 | static void |
3977 | xlate_enqueue_action(struct action_xlate_ctx *ctx, | |
3978 | const struct ofp_action_enqueue *oae) | |
3979 | { | |
3980 | uint16_t ofp_port, odp_port; | |
abff858b | 3981 | uint32_t flow_priority, priority; |
abe529af BP |
3982 | int error; |
3983 | ||
3984 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(oae->queue_id), | |
3985 | &priority); | |
3986 | if (error) { | |
3987 | /* Fall back to ordinary output action. */ | |
3988 | xlate_output_action__(ctx, ntohs(oae->port), 0); | |
3989 | return; | |
3990 | } | |
3991 | ||
df2c07f4 | 3992 | /* Figure out datapath output port. */ |
abe529af BP |
3993 | ofp_port = ntohs(oae->port); |
3994 | if (ofp_port == OFPP_IN_PORT) { | |
3995 | ofp_port = ctx->flow.in_port; | |
8ba855c1 BP |
3996 | } else if (ofp_port == ctx->flow.in_port) { |
3997 | return; | |
abe529af BP |
3998 | } |
3999 | odp_port = ofp_port_to_odp_port(ofp_port); | |
4000 | ||
df2c07f4 | 4001 | /* Add datapath actions. */ |
abff858b PS |
4002 | flow_priority = ctx->flow.priority; |
4003 | ctx->flow.priority = priority; | |
abe529af | 4004 | add_output_action(ctx, odp_port); |
abff858b | 4005 | ctx->flow.priority = flow_priority; |
abe529af BP |
4006 | |
4007 | /* Update NetFlow output port. */ | |
4008 | if (ctx->nf_output_iface == NF_OUT_DROP) { | |
4009 | ctx->nf_output_iface = odp_port; | |
4010 | } else if (ctx->nf_output_iface != NF_OUT_FLOOD) { | |
4011 | ctx->nf_output_iface = NF_OUT_MULTI; | |
4012 | } | |
4013 | } | |
4014 | ||
4015 | static void | |
4016 | xlate_set_queue_action(struct action_xlate_ctx *ctx, | |
4017 | const struct nx_action_set_queue *nasq) | |
4018 | { | |
4019 | uint32_t priority; | |
4020 | int error; | |
4021 | ||
4022 | error = dpif_queue_to_priority(ctx->ofproto->dpif, ntohl(nasq->queue_id), | |
4023 | &priority); | |
4024 | if (error) { | |
4025 | /* Couldn't translate queue to a priority, so ignore. A warning | |
4026 | * has already been logged. */ | |
4027 | return; | |
4028 | } | |
4029 | ||
abff858b | 4030 | ctx->flow.priority = priority; |
abe529af BP |
4031 | } |
4032 | ||
4033 | struct xlate_reg_state { | |
4034 | ovs_be16 vlan_tci; | |
4035 | ovs_be64 tun_id; | |
4036 | }; | |
4037 | ||
abe529af BP |
4038 | static void |
4039 | xlate_autopath(struct action_xlate_ctx *ctx, | |
4040 | const struct nx_action_autopath *naa) | |
4041 | { | |
4042 | uint16_t ofp_port = ntohl(naa->id); | |
4043 | struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port); | |
4044 | ||
4045 | if (!port || !port->bundle) { | |
4046 | ofp_port = OFPP_NONE; | |
4047 | } else if (port->bundle->bond) { | |
4048 | /* Autopath does not support VLAN hashing. */ | |
4049 | struct ofport_dpif *slave = bond_choose_output_slave( | |
dc155bff | 4050 | port->bundle->bond, &ctx->flow, 0, &ctx->tags); |
abe529af BP |
4051 | if (slave) { |
4052 | ofp_port = slave->up.ofp_port; | |
4053 | } | |
4054 | } | |
4055 | autopath_execute(naa, &ctx->flow, ofp_port); | |
4056 | } | |
4057 | ||
daff3353 EJ |
4058 | static bool |
4059 | slave_enabled_cb(uint16_t ofp_port, void *ofproto_) | |
4060 | { | |
4061 | struct ofproto_dpif *ofproto = ofproto_; | |
4062 | struct ofport_dpif *port; | |
4063 | ||
4064 | switch (ofp_port) { | |
4065 | case OFPP_IN_PORT: | |
4066 | case OFPP_TABLE: | |
4067 | case OFPP_NORMAL: | |
4068 | case OFPP_FLOOD: | |
4069 | case OFPP_ALL: | |
439e4d8c | 4070 | case OFPP_NONE: |
daff3353 EJ |
4071 | return true; |
4072 | case OFPP_CONTROLLER: /* Not supported by the bundle action. */ | |
4073 | return false; | |
4074 | default: | |
4075 | port = get_ofp_port(ofproto, ofp_port); | |
4076 | return port ? port->may_enable : false; | |
4077 | } | |
4078 | } | |
4079 | ||
75a75043 BP |
4080 | static void |
4081 | xlate_learn_action(struct action_xlate_ctx *ctx, | |
4082 | const struct nx_action_learn *learn) | |
4083 | { | |
4084 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1); | |
4085 | struct ofputil_flow_mod fm; | |
4086 | int error; | |
4087 | ||
4088 | learn_execute(learn, &ctx->flow, &fm); | |
4089 | ||
4090 | error = ofproto_flow_mod(&ctx->ofproto->up, &fm); | |
4091 | if (error && !VLOG_DROP_WARN(&rl)) { | |
4092 | char *msg = ofputil_error_to_string(error); | |
4093 | VLOG_WARN("learning action failed to modify flow table (%s)", msg); | |
4094 | free(msg); | |
4095 | } | |
4096 | ||
4097 | free(fm.actions); | |
4098 | } | |
4099 | ||
21f7563c JP |
4100 | static bool |
4101 | may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx) | |
4102 | { | |
4103 | if (port->up.opp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp) | |
4104 | ? htonl(OFPPC_NO_RECV_STP) | |
4105 | : htonl(OFPPC_NO_RECV))) { | |
4106 | return false; | |
4107 | } | |
4108 | ||
4109 | /* Only drop packets here if both forwarding and learning are | |
4110 | * disabled. If just learning is enabled, we need to have | |
4111 | * OFPP_NORMAL and the learning action have a look at the packet | |
4112 | * before we can drop it. */ | |
4113 | if (!stp_forward_in_state(port->stp_state) | |
4114 | && !stp_learn_in_state(port->stp_state)) { | |
4115 | return false; | |
4116 | } | |
4117 | ||
4118 | return true; | |
4119 | } | |
4120 | ||
abe529af BP |
4121 | static void |
4122 | do_xlate_actions(const union ofp_action *in, size_t n_in, | |
4123 | struct action_xlate_ctx *ctx) | |
4124 | { | |
4125 | const struct ofport_dpif *port; | |
abe529af | 4126 | const union ofp_action *ia; |
b4b8c781 | 4127 | size_t left; |
abe529af BP |
4128 | |
4129 | port = get_ofp_port(ctx->ofproto, ctx->flow.in_port); | |
21f7563c | 4130 | if (port && !may_receive(port, ctx)) { |
abe529af BP |
4131 | /* Drop this flow. */ |
4132 | return; | |
4133 | } | |
4134 | ||
b4b8c781 | 4135 | OFPUTIL_ACTION_FOR_EACH_UNSAFE (ia, left, in, n_in) { |
abe529af | 4136 | const struct ofp_action_dl_addr *oada; |
38f2e360 BP |
4137 | const struct nx_action_resubmit *nar; |
4138 | const struct nx_action_set_tunnel *nast; | |
4139 | const struct nx_action_set_queue *nasq; | |
4140 | const struct nx_action_multipath *nam; | |
4141 | const struct nx_action_autopath *naa; | |
daff3353 | 4142 | const struct nx_action_bundle *nab; |
f694937d | 4143 | const struct nx_action_output_reg *naor; |
38f2e360 BP |
4144 | enum ofputil_action_code code; |
4145 | ovs_be64 tun_id; | |
4146 | ||
848e8809 EJ |
4147 | if (ctx->exit) { |
4148 | break; | |
4149 | } | |
4150 | ||
38f2e360 BP |
4151 | code = ofputil_decode_action_unsafe(ia); |
4152 | switch (code) { | |
4153 | case OFPUTIL_OFPAT_OUTPUT: | |
abe529af BP |
4154 | xlate_output_action(ctx, &ia->output); |
4155 | break; | |
4156 | ||
38f2e360 | 4157 | case OFPUTIL_OFPAT_SET_VLAN_VID: |
abe529af BP |
4158 | ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK); |
4159 | ctx->flow.vlan_tci |= ia->vlan_vid.vlan_vid | htons(VLAN_CFI); | |
abe529af BP |
4160 | break; |
4161 | ||
38f2e360 | 4162 | case OFPUTIL_OFPAT_SET_VLAN_PCP: |
abe529af BP |
4163 | ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK); |
4164 | ctx->flow.vlan_tci |= htons( | |
4165 | (ia->vlan_pcp.vlan_pcp << VLAN_PCP_SHIFT) | VLAN_CFI); | |
abe529af BP |
4166 | break; |
4167 | ||
38f2e360 | 4168 | case OFPUTIL_OFPAT_STRIP_VLAN: |
abe529af | 4169 | ctx->flow.vlan_tci = htons(0); |
abe529af BP |
4170 | break; |
4171 | ||
38f2e360 | 4172 | case OFPUTIL_OFPAT_SET_DL_SRC: |
abe529af | 4173 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4174 | memcpy(ctx->flow.dl_src, oada->dl_addr, ETH_ADDR_LEN); |
4175 | break; | |
4176 | ||
38f2e360 | 4177 | case OFPUTIL_OFPAT_SET_DL_DST: |
abe529af | 4178 | oada = ((struct ofp_action_dl_addr *) ia); |
abe529af BP |
4179 | memcpy(ctx->flow.dl_dst, oada->dl_addr, ETH_ADDR_LEN); |
4180 | break; | |
4181 | ||
38f2e360 | 4182 | case OFPUTIL_OFPAT_SET_NW_SRC: |
abe529af BP |
4183 | ctx->flow.nw_src = ia->nw_addr.nw_addr; |
4184 | break; | |
4185 | ||
38f2e360 | 4186 | case OFPUTIL_OFPAT_SET_NW_DST: |
abe529af BP |
4187 | ctx->flow.nw_dst = ia->nw_addr.nw_addr; |
4188 | break; | |
4189 | ||
38f2e360 | 4190 | case OFPUTIL_OFPAT_SET_NW_TOS: |
eadef313 JP |
4191 | ctx->flow.nw_tos &= ~IP_DSCP_MASK; |
4192 | ctx->flow.nw_tos |= ia->nw_tos.nw_tos & IP_DSCP_MASK; | |
abe529af BP |
4193 | break; |
4194 | ||
38f2e360 | 4195 | case OFPUTIL_OFPAT_SET_TP_SRC: |
abe529af BP |
4196 | ctx->flow.tp_src = ia->tp_port.tp_port; |
4197 | break; | |
4198 | ||
38f2e360 | 4199 | case OFPUTIL_OFPAT_SET_TP_DST: |
abe529af BP |
4200 | ctx->flow.tp_dst = ia->tp_port.tp_port; |
4201 | break; | |
4202 | ||
38f2e360 BP |
4203 | case OFPUTIL_OFPAT_ENQUEUE: |
4204 | xlate_enqueue_action(ctx, (const struct ofp_action_enqueue *) ia); | |
4205 | break; | |
4206 | ||
4207 | case OFPUTIL_NXAST_RESUBMIT: | |
4208 | nar = (const struct nx_action_resubmit *) ia; | |
29901626 BP |
4209 | xlate_table_action(ctx, ntohs(nar->in_port), ctx->table_id); |
4210 | break; | |
4211 | ||
4212 | case OFPUTIL_NXAST_RESUBMIT_TABLE: | |
4213 | xlate_resubmit_table(ctx, (const struct nx_action_resubmit *) ia); | |
abe529af BP |
4214 | break; |
4215 | ||
38f2e360 BP |
4216 | case OFPUTIL_NXAST_SET_TUNNEL: |
4217 | nast = (const struct nx_action_set_tunnel *) ia; | |
4218 | tun_id = htonll(ntohl(nast->tun_id)); | |
4219 | ctx->flow.tun_id = tun_id; | |
4220 | break; | |
4221 | ||
4222 | case OFPUTIL_NXAST_SET_QUEUE: | |
4223 | nasq = (const struct nx_action_set_queue *) ia; | |
4224 | xlate_set_queue_action(ctx, nasq); | |
4225 | break; | |
4226 | ||
4227 | case OFPUTIL_NXAST_POP_QUEUE: | |
abff858b | 4228 | ctx->flow.priority = ctx->original_priority; |
38f2e360 BP |
4229 | break; |
4230 | ||
4231 | case OFPUTIL_NXAST_REG_MOVE: | |
4232 | nxm_execute_reg_move((const struct nx_action_reg_move *) ia, | |
4233 | &ctx->flow); | |
4234 | break; | |
4235 | ||
4236 | case OFPUTIL_NXAST_REG_LOAD: | |
4237 | nxm_execute_reg_load((const struct nx_action_reg_load *) ia, | |
4238 | &ctx->flow); | |
4239 | break; | |
4240 | ||
4241 | case OFPUTIL_NXAST_NOTE: | |
4242 | /* Nothing to do. */ | |
4243 | break; | |
4244 | ||
4245 | case OFPUTIL_NXAST_SET_TUNNEL64: | |
4246 | tun_id = ((const struct nx_action_set_tunnel64 *) ia)->tun_id; | |
4247 | ctx->flow.tun_id = tun_id; | |
4248 | break; | |
4249 | ||
4250 | case OFPUTIL_NXAST_MULTIPATH: | |
4251 | nam = (const struct nx_action_multipath *) ia; | |
4252 | multipath_execute(nam, &ctx->flow); | |
abe529af BP |
4253 | break; |
4254 | ||
38f2e360 BP |
4255 | case OFPUTIL_NXAST_AUTOPATH: |
4256 | naa = (const struct nx_action_autopath *) ia; | |
4257 | xlate_autopath(ctx, naa); | |
abe529af | 4258 | break; |
daff3353 EJ |
4259 | |
4260 | case OFPUTIL_NXAST_BUNDLE: | |
4261 | ctx->ofproto->has_bundle_action = true; | |
4262 | nab = (const struct nx_action_bundle *) ia; | |
4263 | xlate_output_action__(ctx, bundle_execute(nab, &ctx->flow, | |
4264 | slave_enabled_cb, | |
4265 | ctx->ofproto), 0); | |
4266 | break; | |
a368bb53 EJ |
4267 | |
4268 | case OFPUTIL_NXAST_BUNDLE_LOAD: | |
4269 | ctx->ofproto->has_bundle_action = true; | |
4270 | nab = (const struct nx_action_bundle *) ia; | |
4271 | bundle_execute_load(nab, &ctx->flow, slave_enabled_cb, | |
4272 | ctx->ofproto); | |
4273 | break; | |
f694937d EJ |
4274 | |
4275 | case OFPUTIL_NXAST_OUTPUT_REG: | |
4276 | naor = (const struct nx_action_output_reg *) ia; | |
4277 | xlate_output_reg_action(ctx, naor); | |
4278 | break; | |
75a75043 BP |
4279 | |
4280 | case OFPUTIL_NXAST_LEARN: | |
4281 | ctx->has_learn = true; | |
4282 | if (ctx->may_learn) { | |
4283 | xlate_learn_action(ctx, (const struct nx_action_learn *) ia); | |
4284 | } | |
4285 | break; | |
848e8809 EJ |
4286 | |
4287 | case OFPUTIL_NXAST_EXIT: | |
4288 | ctx->exit = true; | |
4289 | break; | |
abe529af BP |
4290 | } |
4291 | } | |
21f7563c JP |
4292 | |
4293 | /* We've let OFPP_NORMAL and the learning action look at the packet, | |
4294 | * so drop it now if forwarding is disabled. */ | |
4295 | if (port && !stp_forward_in_state(port->stp_state)) { | |
4296 | ofpbuf_clear(ctx->odp_actions); | |
4297 | add_sflow_action(ctx); | |
4298 | } | |
abe529af BP |
4299 | } |
4300 | ||
4301 | static void | |
4302 | action_xlate_ctx_init(struct action_xlate_ctx *ctx, | |
4303 | struct ofproto_dpif *ofproto, const struct flow *flow, | |
4304 | const struct ofpbuf *packet) | |
4305 | { | |
4306 | ctx->ofproto = ofproto; | |
4307 | ctx->flow = *flow; | |
4308 | ctx->packet = packet; | |
75a75043 | 4309 | ctx->may_learn = packet != NULL; |
abe529af | 4310 | ctx->resubmit_hook = NULL; |
abe529af BP |
4311 | } |
4312 | ||
4313 | static struct ofpbuf * | |
4314 | xlate_actions(struct action_xlate_ctx *ctx, | |
4315 | const union ofp_action *in, size_t n_in) | |
4316 | { | |
4317 | COVERAGE_INC(ofproto_dpif_xlate); | |
4318 | ||
4319 | ctx->odp_actions = ofpbuf_new(512); | |
b6848f13 | 4320 | ofpbuf_reserve(ctx->odp_actions, NL_A_U32_SIZE); |
97e42c92 BP |
4321 | ctx->tags = 0; |
4322 | ctx->may_set_up_flow = true; | |
4323 | ctx->has_learn = false; | |
4324 | ctx->has_normal = false; | |
4325 | ctx->nf_output_iface = NF_OUT_DROP; | |
4326 | ctx->recurse = 0; | |
abff858b | 4327 | ctx->original_priority = ctx->flow.priority; |
97e42c92 BP |
4328 | ctx->base_flow = ctx->flow; |
4329 | ctx->base_flow.tun_id = 0; | |
4330 | ctx->table_id = 0; | |
848e8809 | 4331 | ctx->exit = false; |
7257b535 | 4332 | |
eadef313 | 4333 | if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) { |
7257b535 BP |
4334 | switch (ctx->ofproto->up.frag_handling) { |
4335 | case OFPC_FRAG_NORMAL: | |
4336 | /* We must pretend that transport ports are unavailable. */ | |
97e42c92 BP |
4337 | ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0); |
4338 | ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0); | |
7257b535 BP |
4339 | break; |
4340 | ||
4341 | case OFPC_FRAG_DROP: | |
4342 | return ctx->odp_actions; | |
4343 | ||
4344 | case OFPC_FRAG_REASM: | |
4345 | NOT_REACHED(); | |
4346 | ||
4347 | case OFPC_FRAG_NX_MATCH: | |
4348 | /* Nothing to do. */ | |
4349 | break; | |
4350 | } | |
4351 | } | |
4352 | ||
fc08b7a2 | 4353 | if (process_special(ctx->ofproto, &ctx->flow, ctx->packet)) { |
abe529af | 4354 | ctx->may_set_up_flow = false; |
b6848f13 | 4355 | return ctx->odp_actions; |
abe529af | 4356 | } else { |
6ff686f2 | 4357 | add_sflow_action(ctx); |
abe529af | 4358 | do_xlate_actions(in, n_in, ctx); |
abe529af | 4359 | |
b6848f13 BP |
4360 | if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow, |
4361 | ctx->odp_actions->data, | |
4362 | ctx->odp_actions->size)) { | |
4363 | ctx->may_set_up_flow = false; | |
4364 | if (ctx->packet | |
4365 | && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow, | |
4366 | ctx->packet)) { | |
a7c4eaf6 | 4367 | compose_output_action(ctx, OVSP_LOCAL); |
b6848f13 BP |
4368 | } |
4369 | } | |
a7c4eaf6 | 4370 | fix_sflow_action(ctx); |
abe529af BP |
4371 | } |
4372 | ||
4373 | return ctx->odp_actions; | |
4374 | } | |
4375 | \f | |
4376 | /* OFPP_NORMAL implementation. */ | |
4377 | ||
abe529af BP |
4378 | static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *); |
4379 | ||
ecac4ebf BP |
4380 | /* Given 'vid', the VID obtained from the 802.1Q header that was received as |
4381 | * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle', | |
4382 | * the bundle on which the packet was received, returns the VLAN to which the | |
4383 | * packet belongs. | |
4384 | * | |
4385 | * Both 'vid' and the return value are in the range 0...4095. */ | |
4386 | static uint16_t | |
4387 | input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid) | |
4388 | { | |
4389 | switch (in_bundle->vlan_mode) { | |
4390 | case PORT_VLAN_ACCESS: | |
4391 | return in_bundle->vlan; | |
4392 | break; | |
4393 | ||
4394 | case PORT_VLAN_TRUNK: | |
4395 | return vid; | |
4396 | ||
4397 | case PORT_VLAN_NATIVE_UNTAGGED: | |
4398 | case PORT_VLAN_NATIVE_TAGGED: | |
4399 | return vid ? vid : in_bundle->vlan; | |
4400 | ||
4401 | default: | |
4402 | NOT_REACHED(); | |
4403 | } | |
4404 | } | |
4405 | ||
5da5ec37 BP |
4406 | /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'. |
4407 | * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs | |
4408 | * a warning. | |
4409 | * | |
4410 | * 'vid' should be the VID obtained from the 802.1Q header that was received as | |
4411 | * part of a packet (specify 0 if there was no 802.1Q header), in the range | |
4412 | * 0...4095. */ | |
4413 | static bool | |
4414 | input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn) | |
4415 | { | |
4416 | switch (in_bundle->vlan_mode) { | |
4417 | case PORT_VLAN_ACCESS: | |
4418 | if (vid) { | |
4419 | if (warn) { | |
4420 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4421 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged " | |
4422 | "packet received on port %s configured as VLAN " | |
4423 | "%"PRIu16" access port", | |
4424 | in_bundle->ofproto->up.name, vid, | |
4425 | in_bundle->name, in_bundle->vlan); | |
4426 | } | |
4427 | return false; | |
4428 | } | |
4429 | return true; | |
4430 | ||
4431 | case PORT_VLAN_NATIVE_UNTAGGED: | |
4432 | case PORT_VLAN_NATIVE_TAGGED: | |
4433 | if (!vid) { | |
4434 | /* Port must always carry its native VLAN. */ | |
4435 | return true; | |
4436 | } | |
4437 | /* Fall through. */ | |
4438 | case PORT_VLAN_TRUNK: | |
4439 | if (!ofbundle_includes_vlan(in_bundle, vid)) { | |
4440 | if (warn) { | |
4441 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4442 | VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet " | |
4443 | "received on port %s not configured for trunking " | |
4444 | "VLAN %"PRIu16, | |
4445 | in_bundle->ofproto->up.name, vid, | |
4446 | in_bundle->name, vid); | |
4447 | } | |
4448 | return false; | |
4449 | } | |
4450 | return true; | |
4451 | ||
4452 | default: | |
4453 | NOT_REACHED(); | |
4454 | } | |
4455 | ||
4456 | } | |
4457 | ||
ecac4ebf BP |
4458 | /* Given 'vlan', the VLAN that a packet belongs to, and |
4459 | * 'out_bundle', a bundle on which the packet is to be output, returns the VID | |
4460 | * that should be included in the 802.1Q header. (If the return value is 0, | |
4461 | * then the 802.1Q header should only be included in the packet if there is a | |
4462 | * nonzero PCP.) | |
4463 | * | |
4464 | * Both 'vlan' and the return value are in the range 0...4095. */ | |
4465 | static uint16_t | |
4466 | output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan) | |
4467 | { | |
4468 | switch (out_bundle->vlan_mode) { | |
4469 | case PORT_VLAN_ACCESS: | |
4470 | return 0; | |
4471 | ||
4472 | case PORT_VLAN_TRUNK: | |
4473 | case PORT_VLAN_NATIVE_TAGGED: | |
4474 | return vlan; | |
4475 | ||
4476 | case PORT_VLAN_NATIVE_UNTAGGED: | |
4477 | return vlan == out_bundle->vlan ? 0 : vlan; | |
4478 | ||
4479 | default: | |
4480 | NOT_REACHED(); | |
4481 | } | |
4482 | } | |
4483 | ||
395e68ce BP |
4484 | static void |
4485 | output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle, | |
4486 | uint16_t vlan) | |
abe529af | 4487 | { |
395e68ce BP |
4488 | struct ofport_dpif *port; |
4489 | uint16_t vid; | |
4490 | ovs_be16 tci; | |
ecac4ebf | 4491 | |
395e68ce BP |
4492 | vid = output_vlan_to_vid(out_bundle, vlan); |
4493 | if (!out_bundle->bond) { | |
4494 | port = ofbundle_get_a_port(out_bundle); | |
4495 | } else { | |
4496 | port = bond_choose_output_slave(out_bundle->bond, &ctx->flow, | |
4497 | vid, &ctx->tags); | |
4498 | if (!port) { | |
4499 | /* No slaves enabled, so drop packet. */ | |
4500 | return; | |
4501 | } | |
4502 | } | |
abe529af | 4503 | |
5e9ceccd BP |
4504 | tci = htons(vid); |
4505 | if (tci || out_bundle->use_priority_tags) { | |
4506 | tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK); | |
4507 | if (tci) { | |
4508 | tci |= htons(VLAN_CFI); | |
4509 | } | |
395e68ce BP |
4510 | } |
4511 | commit_vlan_action(ctx, tci); | |
4512 | ||
4513 | compose_output_action(ctx, port->odp_port); | |
4514 | ctx->nf_output_iface = port->odp_port; | |
abe529af BP |
4515 | } |
4516 | ||
4517 | static int | |
4518 | mirror_mask_ffs(mirror_mask_t mask) | |
4519 | { | |
4520 | BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask)); | |
4521 | return ffs(mask); | |
4522 | } | |
4523 | ||
abe529af BP |
4524 | static bool |
4525 | ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
4526 | { | |
ecac4ebf | 4527 | return (bundle->vlan_mode != PORT_VLAN_ACCESS |
fc3d7408 | 4528 | && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan))); |
abe529af BP |
4529 | } |
4530 | ||
4531 | static bool | |
4532 | ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan) | |
4533 | { | |
4534 | return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan); | |
4535 | } | |
4536 | ||
4537 | /* Returns an arbitrary interface within 'bundle'. */ | |
4538 | static struct ofport_dpif * | |
4539 | ofbundle_get_a_port(const struct ofbundle *bundle) | |
4540 | { | |
4541 | return CONTAINER_OF(list_front(&bundle->ports), | |
4542 | struct ofport_dpif, bundle_node); | |
4543 | } | |
4544 | ||
395e68ce | 4545 | static mirror_mask_t |
abe529af BP |
4546 | compose_dsts(struct action_xlate_ctx *ctx, uint16_t vlan, |
4547 | const struct ofbundle *in_bundle, | |
395e68ce | 4548 | const struct ofbundle *out_bundle) |
abe529af | 4549 | { |
395e68ce | 4550 | mirror_mask_t dst_mirrors = 0; |
abe529af BP |
4551 | |
4552 | if (out_bundle == OFBUNDLE_FLOOD) { | |
4553 | struct ofbundle *bundle; | |
4554 | ||
4555 | HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) { | |
4556 | if (bundle != in_bundle | |
4557 | && ofbundle_includes_vlan(bundle, vlan) | |
4558 | && bundle->floodable | |
395e68ce BP |
4559 | && !bundle->mirror_out) { |
4560 | output_normal(ctx, bundle, vlan); | |
4561 | dst_mirrors |= bundle->dst_mirrors; | |
abe529af BP |
4562 | } |
4563 | } | |
4564 | ctx->nf_output_iface = NF_OUT_FLOOD; | |
395e68ce BP |
4565 | } else if (out_bundle) { |
4566 | output_normal(ctx, out_bundle, vlan); | |
4567 | dst_mirrors = out_bundle->dst_mirrors; | |
abe529af | 4568 | } |
395e68ce BP |
4569 | |
4570 | return dst_mirrors; | |
abe529af BP |
4571 | } |
4572 | ||
4573 | static bool | |
4574 | vlan_is_mirrored(const struct ofmirror *m, int vlan) | |
4575 | { | |
fc3d7408 | 4576 | return !m->vlans || bitmap_is_set(m->vlans, vlan); |
abe529af BP |
4577 | } |
4578 | ||
07817dfe BP |
4579 | /* Returns true if a packet with Ethernet destination MAC 'dst' may be mirrored |
4580 | * to a VLAN. In general most packets may be mirrored but we want to drop | |
4581 | * protocols that may confuse switches. */ | |
4582 | static bool | |
4583 | eth_dst_may_rspan(const uint8_t dst[ETH_ADDR_LEN]) | |
4584 | { | |
4585 | /* If you change this function's behavior, please update corresponding | |
4586 | * documentation in vswitch.xml at the same time. */ | |
4587 | if (dst[0] != 0x01) { | |
4588 | /* All the currently banned MACs happen to start with 01 currently, so | |
4589 | * this is a quick way to eliminate most of the good ones. */ | |
4590 | } else { | |
4591 | if (eth_addr_is_reserved(dst)) { | |
4592 | /* Drop STP, IEEE pause frames, and other reserved protocols | |
4593 | * (01-80-c2-00-00-0x). */ | |
4594 | return false; | |
4595 | } | |
4596 | ||
4597 | if (dst[0] == 0x01 && dst[1] == 0x00 && dst[2] == 0x0c) { | |
4598 | /* Cisco OUI. */ | |
4599 | if ((dst[3] & 0xfe) == 0xcc && | |
4600 | (dst[4] & 0xfe) == 0xcc && | |
4601 | (dst[5] & 0xfe) == 0xcc) { | |
4602 | /* Drop the following protocols plus others following the same | |
4603 | pattern: | |
4604 | ||
4605 | CDP, VTP, DTP, PAgP (01-00-0c-cc-cc-cc) | |
4606 | Spanning Tree PVSTP+ (01-00-0c-cc-cc-cd) | |
4607 | STP Uplink Fast (01-00-0c-cd-cd-cd) */ | |
4608 | return false; | |
4609 | } | |
4610 | ||
4611 | if (!(dst[3] | dst[4] | dst[5])) { | |
4612 | /* Drop Inter Switch Link packets (01-00-0c-00-00-00). */ | |
4613 | return false; | |
4614 | } | |
4615 | } | |
4616 | } | |
4617 | return true; | |
4618 | } | |
4619 | ||
abe529af | 4620 | static void |
395e68ce BP |
4621 | output_mirrors(struct action_xlate_ctx *ctx, |
4622 | uint16_t vlan, const struct ofbundle *in_bundle, | |
4623 | mirror_mask_t dst_mirrors) | |
abe529af BP |
4624 | { |
4625 | struct ofproto_dpif *ofproto = ctx->ofproto; | |
4626 | mirror_mask_t mirrors; | |
abe529af | 4627 | |
395e68ce | 4628 | mirrors = in_bundle->src_mirrors | dst_mirrors; |
abe529af BP |
4629 | if (!mirrors) { |
4630 | return; | |
4631 | } | |
4632 | ||
9ba15e2a BP |
4633 | while (mirrors) { |
4634 | struct ofmirror *m; | |
9ba15e2a BP |
4635 | |
4636 | m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1]; | |
4637 | ||
4638 | if (!vlan_is_mirrored(m, vlan)) { | |
4639 | mirrors &= mirrors - 1; | |
4640 | continue; | |
4641 | } | |
4642 | ||
4643 | mirrors &= ~m->dup_mirrors; | |
4644 | if (m->out) { | |
395e68ce | 4645 | output_normal(ctx, m->out, vlan); |
9ba15e2a BP |
4646 | } else if (eth_dst_may_rspan(ctx->flow.dl_dst) |
4647 | && vlan != m->out_vlan) { | |
4648 | struct ofbundle *bundle; | |
4649 | ||
4650 | HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) { | |
4651 | if (ofbundle_includes_vlan(bundle, m->out_vlan) | |
395e68ce BP |
4652 | && !bundle->mirror_out) { |
4653 | output_normal(ctx, bundle, m->out_vlan); | |
abe529af BP |
4654 | } |
4655 | } | |
4656 | } | |
abe529af BP |
4657 | } |
4658 | } | |
4659 | ||
abe529af BP |
4660 | /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after |
4661 | * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to | |
4662 | * indicate this; newer upstream kernels use gratuitous ARP requests. */ | |
4663 | static bool | |
4664 | is_gratuitous_arp(const struct flow *flow) | |
4665 | { | |
4666 | return (flow->dl_type == htons(ETH_TYPE_ARP) | |
4667 | && eth_addr_is_broadcast(flow->dl_dst) | |
4668 | && (flow->nw_proto == ARP_OP_REPLY | |
4669 | || (flow->nw_proto == ARP_OP_REQUEST | |
4670 | && flow->nw_src == flow->nw_dst))); | |
4671 | } | |
4672 | ||
4673 | static void | |
4674 | update_learning_table(struct ofproto_dpif *ofproto, | |
4675 | const struct flow *flow, int vlan, | |
4676 | struct ofbundle *in_bundle) | |
4677 | { | |
4678 | struct mac_entry *mac; | |
4679 | ||
4680 | if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) { | |
4681 | return; | |
4682 | } | |
4683 | ||
4684 | mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan); | |
4685 | if (is_gratuitous_arp(flow)) { | |
4686 | /* We don't want to learn from gratuitous ARP packets that are | |
4687 | * reflected back over bond slaves so we lock the learning table. */ | |
4688 | if (!in_bundle->bond) { | |
4689 | mac_entry_set_grat_arp_lock(mac); | |
4690 | } else if (mac_entry_is_grat_arp_locked(mac)) { | |
4691 | return; | |
4692 | } | |
4693 | } | |
4694 | ||
4695 | if (mac_entry_is_new(mac) || mac->port.p != in_bundle) { | |
4696 | /* The log messages here could actually be useful in debugging, | |
4697 | * so keep the rate limit relatively high. */ | |
4698 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300); | |
4699 | VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is " | |
4700 | "on port %s in VLAN %d", | |
4701 | ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src), | |
4702 | in_bundle->name, vlan); | |
4703 | ||
4704 | mac->port.p = in_bundle; | |
4705 | tag_set_add(&ofproto->revalidate_set, | |
4706 | mac_learning_changed(ofproto->ml, mac)); | |
4707 | } | |
4708 | } | |
4709 | ||
395e68ce BP |
4710 | static struct ofport_dpif * |
4711 | lookup_input_bundle(struct ofproto_dpif *ofproto, uint16_t in_port, bool warn) | |
4712 | { | |
4713 | struct ofport_dpif *ofport; | |
4714 | ||
4715 | /* Find the port and bundle for the received packet. */ | |
4716 | ofport = get_ofp_port(ofproto, in_port); | |
4717 | if (ofport && ofport->bundle) { | |
4718 | return ofport; | |
4719 | } | |
4720 | ||
4721 | /* Odd. A few possible reasons here: | |
4722 | * | |
4723 | * - We deleted a port but there are still a few packets queued up | |
4724 | * from it. | |
4725 | * | |
4726 | * - Someone externally added a port (e.g. "ovs-dpctl add-if") that | |
4727 | * we don't know about. | |
4728 | * | |
4729 | * - The ofproto client didn't configure the port as part of a bundle. | |
4730 | */ | |
4731 | if (warn) { | |
4732 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4733 | ||
4734 | VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown " | |
4735 | "port %"PRIu16, ofproto->up.name, in_port); | |
4736 | } | |
4737 | return NULL; | |
4738 | } | |
4739 | ||
5da5ec37 | 4740 | /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or |
abe529af BP |
4741 | * dropped. Returns true if they may be forwarded, false if they should be |
4742 | * dropped. | |
4743 | * | |
395e68ce BP |
4744 | * 'in_port' must be the ofport_dpif that corresponds to flow->in_port. |
4745 | * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull). | |
abe529af | 4746 | * |
395e68ce BP |
4747 | * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as |
4748 | * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as | |
4749 | * checked by input_vid_is_valid(). | |
abe529af BP |
4750 | * |
4751 | * May also add tags to '*tags', although the current implementation only does | |
4752 | * so in one special case. | |
4753 | */ | |
4754 | static bool | |
4755 | is_admissible(struct ofproto_dpif *ofproto, const struct flow *flow, | |
395e68ce | 4756 | struct ofport_dpif *in_port, uint16_t vlan, tag_type *tags) |
abe529af | 4757 | { |
395e68ce | 4758 | struct ofbundle *in_bundle = in_port->bundle; |
abe529af | 4759 | |
395e68ce BP |
4760 | /* Drop frames for reserved multicast addresses |
4761 | * only if forward_bpdu option is absent. */ | |
21f7563c | 4762 | if (eth_addr_is_reserved(flow->dl_dst) && !ofproto->up.forward_bpdu) { |
abe529af BP |
4763 | return false; |
4764 | } | |
4765 | ||
abe529af BP |
4766 | if (in_bundle->bond) { |
4767 | struct mac_entry *mac; | |
4768 | ||
4769 | switch (bond_check_admissibility(in_bundle->bond, in_port, | |
4770 | flow->dl_dst, tags)) { | |
4771 | case BV_ACCEPT: | |
4772 | break; | |
4773 | ||
4774 | case BV_DROP: | |
4775 | return false; | |
4776 | ||
4777 | case BV_DROP_IF_MOVED: | |
4778 | mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL); | |
4779 | if (mac && mac->port.p != in_bundle && | |
4780 | (!is_gratuitous_arp(flow) | |
4781 | || mac_entry_is_grat_arp_locked(mac))) { | |
4782 | return false; | |
4783 | } | |
4784 | break; | |
4785 | } | |
4786 | } | |
4787 | ||
4788 | return true; | |
4789 | } | |
4790 | ||
4cd78906 | 4791 | static void |
abe529af BP |
4792 | xlate_normal(struct action_xlate_ctx *ctx) |
4793 | { | |
395e68ce BP |
4794 | mirror_mask_t dst_mirrors = 0; |
4795 | struct ofport_dpif *in_port; | |
abe529af BP |
4796 | struct ofbundle *in_bundle; |
4797 | struct ofbundle *out_bundle; | |
4798 | struct mac_entry *mac; | |
395e68ce BP |
4799 | uint16_t vlan; |
4800 | uint16_t vid; | |
abe529af | 4801 | |
75a75043 BP |
4802 | ctx->has_normal = true; |
4803 | ||
395e68ce BP |
4804 | /* Obtain in_port from ctx->flow.in_port. |
4805 | * | |
4806 | * lookup_input_bundle() also ensures that in_port belongs to a bundle. */ | |
4807 | in_port = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port, | |
4808 | ctx->packet != NULL); | |
4809 | if (!in_port) { | |
4810 | return; | |
4811 | } | |
4812 | in_bundle = in_port->bundle; | |
4813 | ||
4814 | /* Drop malformed frames. */ | |
4815 | if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) && | |
4816 | !(ctx->flow.vlan_tci & htons(VLAN_CFI))) { | |
4817 | if (ctx->packet != NULL) { | |
4818 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4819 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial " | |
4820 | "VLAN tag received on port %s", | |
4821 | ctx->ofproto->up.name, in_bundle->name); | |
4822 | } | |
4823 | return; | |
4824 | } | |
4825 | ||
4826 | /* Drop frames on bundles reserved for mirroring. */ | |
4827 | if (in_bundle->mirror_out) { | |
4828 | if (ctx->packet != NULL) { | |
4829 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
4830 | VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port " | |
4831 | "%s, which is reserved exclusively for mirroring", | |
4832 | ctx->ofproto->up.name, in_bundle->name); | |
4833 | } | |
4834 | return; | |
4835 | } | |
4836 | ||
4837 | /* Check VLAN. */ | |
4838 | vid = vlan_tci_to_vid(ctx->flow.vlan_tci); | |
4839 | if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) { | |
4840 | return; | |
4841 | } | |
4842 | vlan = input_vid_to_vlan(in_bundle, vid); | |
4843 | ||
4844 | /* Check other admissibility requirements. */ | |
4845 | if (!is_admissible(ctx->ofproto, &ctx->flow, in_port, vlan, &ctx->tags)) { | |
4846 | output_mirrors(ctx, vlan, in_bundle, 0); | |
4847 | return; | |
abe529af BP |
4848 | } |
4849 | ||
75a75043 BP |
4850 | /* Learn source MAC. */ |
4851 | if (ctx->may_learn) { | |
abe529af BP |
4852 | update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle); |
4853 | } | |
4854 | ||
4855 | /* Determine output bundle. */ | |
4856 | mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan, | |
4857 | &ctx->tags); | |
4858 | if (mac) { | |
4859 | out_bundle = mac->port.p; | |
4860 | } else if (!ctx->packet && !eth_addr_is_multicast(ctx->flow.dl_dst)) { | |
4861 | /* If we are revalidating but don't have a learning entry then eject | |
4862 | * the flow. Installing a flow that floods packets opens up a window | |
4863 | * of time where we could learn from a packet reflected on a bond and | |
4864 | * blackhole packets before the learning table is updated to reflect | |
4865 | * the correct port. */ | |
4cd78906 BP |
4866 | ctx->may_set_up_flow = false; |
4867 | return; | |
abe529af BP |
4868 | } else { |
4869 | out_bundle = OFBUNDLE_FLOOD; | |
4870 | } | |
4871 | ||
4872 | /* Don't send packets out their input bundles. */ | |
395e68ce BP |
4873 | if (in_bundle != out_bundle) { |
4874 | dst_mirrors = compose_dsts(ctx, vlan, in_bundle, out_bundle); | |
abe529af | 4875 | } |
395e68ce | 4876 | output_mirrors(ctx, vlan, in_bundle, dst_mirrors); |
abe529af BP |
4877 | } |
4878 | \f | |
54a9cbc9 BP |
4879 | /* Optimized flow revalidation. |
4880 | * | |
4881 | * It's a difficult problem, in general, to tell which facets need to have | |
4882 | * their actions recalculated whenever the OpenFlow flow table changes. We | |
4883 | * don't try to solve that general problem: for most kinds of OpenFlow flow | |
4884 | * table changes, we recalculate the actions for every facet. This is | |
4885 | * relatively expensive, but it's good enough if the OpenFlow flow table | |
4886 | * doesn't change very often. | |
4887 | * | |
4888 | * However, we can expect one particular kind of OpenFlow flow table change to | |
4889 | * happen frequently: changes caused by MAC learning. To avoid wasting a lot | |
4890 | * of CPU on revalidating every facet whenever MAC learning modifies the flow | |
4891 | * table, we add a special case that applies to flow tables in which every rule | |
4892 | * has the same form (that is, the same wildcards), except that the table is | |
4893 | * also allowed to have a single "catch-all" flow that matches all packets. We | |
4894 | * optimize this case by tagging all of the facets that resubmit into the table | |
4895 | * and invalidating the same tag whenever a flow changes in that table. The | |
4896 | * end result is that we revalidate just the facets that need it (and sometimes | |
4897 | * a few more, but not all of the facets or even all of the facets that | |
4898 | * resubmit to the table modified by MAC learning). */ | |
4899 | ||
4900 | /* Calculates the tag to use for 'flow' and wildcards 'wc' when it is inserted | |
4901 | * into an OpenFlow table with the given 'basis'. */ | |
4902 | static uint32_t | |
4903 | rule_calculate_tag(const struct flow *flow, const struct flow_wildcards *wc, | |
4904 | uint32_t secret) | |
4905 | { | |
4906 | if (flow_wildcards_is_catchall(wc)) { | |
4907 | return 0; | |
4908 | } else { | |
4909 | struct flow tag_flow = *flow; | |
4910 | flow_zero_wildcards(&tag_flow, wc); | |
4911 | return tag_create_deterministic(flow_hash(&tag_flow, secret)); | |
4912 | } | |
4913 | } | |
4914 | ||
4915 | /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the | |
4916 | * taggability of that table. | |
4917 | * | |
4918 | * This function must be called after *each* change to a flow table. If you | |
4919 | * skip calling it on some changes then the pointer comparisons at the end can | |
4920 | * be invalid if you get unlucky. For example, if a flow removal causes a | |
4921 | * cls_table to be destroyed and then a flow insertion causes a cls_table with | |
4922 | * different wildcards to be created with the same address, then this function | |
4923 | * will incorrectly skip revalidation. */ | |
4924 | static void | |
4925 | table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id) | |
4926 | { | |
4927 | struct table_dpif *table = &ofproto->tables[table_id]; | |
4928 | const struct classifier *cls = &ofproto->up.tables[table_id]; | |
4929 | struct cls_table *catchall, *other; | |
4930 | struct cls_table *t; | |
4931 | ||
4932 | catchall = other = NULL; | |
4933 | ||
4934 | switch (hmap_count(&cls->tables)) { | |
4935 | case 0: | |
4936 | /* We could tag this OpenFlow table but it would make the logic a | |
4937 | * little harder and it's a corner case that doesn't seem worth it | |
4938 | * yet. */ | |
4939 | break; | |
4940 | ||
4941 | case 1: | |
4942 | case 2: | |
4943 | HMAP_FOR_EACH (t, hmap_node, &cls->tables) { | |
4944 | if (cls_table_is_catchall(t)) { | |
4945 | catchall = t; | |
4946 | } else if (!other) { | |
4947 | other = t; | |
4948 | } else { | |
4949 | /* Indicate that we can't tag this by setting both tables to | |
4950 | * NULL. (We know that 'catchall' is already NULL.) */ | |
4951 | other = NULL; | |
4952 | } | |
4953 | } | |
4954 | break; | |
4955 | ||
4956 | default: | |
4957 | /* Can't tag this table. */ | |
4958 | break; | |
4959 | } | |
4960 | ||
4961 | if (table->catchall_table != catchall || table->other_table != other) { | |
4962 | table->catchall_table = catchall; | |
4963 | table->other_table = other; | |
4964 | ofproto->need_revalidate = true; | |
4965 | } | |
4966 | } | |
4967 | ||
4968 | /* Given 'rule' that has changed in some way (either it is a rule being | |
4969 | * inserted, a rule being deleted, or a rule whose actions are being | |
4970 | * modified), marks facets for revalidation to ensure that packets will be | |
4971 | * forwarded correctly according to the new state of the flow table. | |
4972 | * | |
4973 | * This function must be called after *each* change to a flow table. See | |
4974 | * the comment on table_update_taggable() for more information. */ | |
4975 | static void | |
4976 | rule_invalidate(const struct rule_dpif *rule) | |
4977 | { | |
4978 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto); | |
4979 | ||
4980 | table_update_taggable(ofproto, rule->up.table_id); | |
4981 | ||
4982 | if (!ofproto->need_revalidate) { | |
4983 | struct table_dpif *table = &ofproto->tables[rule->up.table_id]; | |
4984 | ||
4985 | if (table->other_table && rule->tag) { | |
4986 | tag_set_add(&ofproto->revalidate_set, rule->tag); | |
4987 | } else { | |
4988 | ofproto->need_revalidate = true; | |
4989 | } | |
4990 | } | |
4991 | } | |
4992 | \f | |
abe529af | 4993 | static bool |
7257b535 BP |
4994 | set_frag_handling(struct ofproto *ofproto_, |
4995 | enum ofp_config_flags frag_handling) | |
abe529af BP |
4996 | { |
4997 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
abe529af | 4998 | |
7257b535 BP |
4999 | if (frag_handling != OFPC_FRAG_REASM) { |
5000 | ofproto->need_revalidate = true; | |
5001 | return true; | |
5002 | } else { | |
5003 | return false; | |
5004 | } | |
abe529af BP |
5005 | } |
5006 | ||
5007 | static int | |
5008 | packet_out(struct ofproto *ofproto_, struct ofpbuf *packet, | |
5009 | const struct flow *flow, | |
5010 | const union ofp_action *ofp_actions, size_t n_ofp_actions) | |
5011 | { | |
5012 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5013 | int error; | |
5014 | ||
e1154f71 BP |
5015 | if (flow->in_port >= ofproto->max_ports && flow->in_port < OFPP_MAX) { |
5016 | return ofp_mkerr_nicira(OFPET_BAD_REQUEST, NXBRC_BAD_IN_PORT); | |
5017 | } | |
5018 | ||
abe529af BP |
5019 | error = validate_actions(ofp_actions, n_ofp_actions, flow, |
5020 | ofproto->max_ports); | |
5021 | if (!error) { | |
80e5eed9 | 5022 | struct odputil_keybuf keybuf; |
abe529af BP |
5023 | struct action_xlate_ctx ctx; |
5024 | struct ofpbuf *odp_actions; | |
80e5eed9 BP |
5025 | struct ofpbuf key; |
5026 | ||
5027 | ofpbuf_use_stack(&key, &keybuf, sizeof keybuf); | |
5028 | odp_flow_key_from_flow(&key, flow); | |
abe529af BP |
5029 | |
5030 | action_xlate_ctx_init(&ctx, ofproto, flow, packet); | |
5031 | odp_actions = xlate_actions(&ctx, ofp_actions, n_ofp_actions); | |
80e5eed9 BP |
5032 | dpif_execute(ofproto->dpif, key.data, key.size, |
5033 | odp_actions->data, odp_actions->size, packet); | |
abe529af BP |
5034 | ofpbuf_delete(odp_actions); |
5035 | } | |
5036 | return error; | |
5037 | } | |
5038 | ||
5039 | static void | |
5040 | get_netflow_ids(const struct ofproto *ofproto_, | |
5041 | uint8_t *engine_type, uint8_t *engine_id) | |
5042 | { | |
5043 | struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_); | |
5044 | ||
5045 | dpif_get_netflow_ids(ofproto->dpif, engine_type, engine_id); | |
5046 | } | |
5047 | \f | |
5048 | static struct ofproto_dpif * | |
5049 | ofproto_dpif_lookup(const char *name) | |
5050 | { | |
5051 | struct ofproto *ofproto = ofproto_lookup(name); | |
5052 | return (ofproto && ofproto->ofproto_class == &ofproto_dpif_class | |
5053 | ? ofproto_dpif_cast(ofproto) | |
5054 | : NULL); | |
5055 | } | |
5056 | ||
f0a3aa2e AA |
5057 | static void |
5058 | ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, | |
5059 | const char *args, void *aux OVS_UNUSED) | |
5060 | { | |
5061 | const struct ofproto_dpif *ofproto; | |
5062 | ||
5063 | ofproto = ofproto_dpif_lookup(args); | |
5064 | if (!ofproto) { | |
5065 | unixctl_command_reply(conn, 501, "no such bridge"); | |
5066 | return; | |
5067 | } | |
5068 | mac_learning_flush(ofproto->ml); | |
5069 | ||
5070 | unixctl_command_reply(conn, 200, "table successfully flushed"); | |
5071 | } | |
5072 | ||
abe529af BP |
5073 | static void |
5074 | ofproto_unixctl_fdb_show(struct unixctl_conn *conn, | |
5075 | const char *args, void *aux OVS_UNUSED) | |
5076 | { | |
5077 | struct ds ds = DS_EMPTY_INITIALIZER; | |
5078 | const struct ofproto_dpif *ofproto; | |
5079 | const struct mac_entry *e; | |
5080 | ||
5081 | ofproto = ofproto_dpif_lookup(args); | |
5082 | if (!ofproto) { | |
5083 | unixctl_command_reply(conn, 501, "no such bridge"); | |
5084 | return; | |
5085 | } | |
5086 | ||
5087 | ds_put_cstr(&ds, " port VLAN MAC Age\n"); | |
5088 | LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) { | |
5089 | struct ofbundle *bundle = e->port.p; | |
5090 | ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n", | |
5091 | ofbundle_get_a_port(bundle)->odp_port, | |
5092 | e->vlan, ETH_ADDR_ARGS(e->mac), mac_entry_age(e)); | |
5093 | } | |
5094 | unixctl_command_reply(conn, 200, ds_cstr(&ds)); | |
5095 | ds_destroy(&ds); | |
5096 | } | |
5097 | ||
5098 | struct ofproto_trace { | |
5099 | struct action_xlate_ctx ctx; | |
5100 | struct flow flow; | |
5101 | struct ds *result; | |
5102 | }; | |
5103 | ||
5104 | static void | |
29901626 BP |
5105 | trace_format_rule(struct ds *result, uint8_t table_id, int level, |
5106 | const struct rule_dpif *rule) | |
abe529af BP |
5107 | { |
5108 | ds_put_char_multiple(result, '\t', level); | |
5109 | if (!rule) { | |
5110 | ds_put_cstr(result, "No match\n"); | |
5111 | return; | |
5112 | } | |
5113 | ||
29901626 BP |
5114 | ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ", |
5115 | table_id, ntohll(rule->up.flow_cookie)); | |
79feb7df | 5116 | cls_rule_format(&rule->up.cr, result); |
abe529af BP |
5117 | ds_put_char(result, '\n'); |
5118 | ||
5119 | ds_put_char_multiple(result, '\t', level); | |
5120 | ds_put_cstr(result, "OpenFlow "); | |
79feb7df | 5121 | ofp_print_actions(result, rule->up.actions, rule->up.n_actions); |
abe529af BP |
5122 | ds_put_char(result, '\n'); |
5123 | } | |
5124 | ||
5125 | static void | |
5126 | trace_format_flow(struct ds *result, int level, const char *title, | |
5127 | struct ofproto_trace *trace) | |
5128 | { | |
5129 | ds_put_char_multiple(result, '\t', level); | |
5130 | ds_put_format(result, "%s: ", title); | |
5131 | if (flow_equal(&trace->ctx.flow, &trace->flow)) { | |
5132 | ds_put_cstr(result, "unchanged"); | |
5133 | } else { | |
5134 | flow_format(result, &trace->ctx.flow); | |
5135 | trace->flow = trace->ctx.flow; | |
5136 | } | |
5137 | ds_put_char(result, '\n'); | |
5138 | } | |
5139 | ||
eb9e1c26 EJ |
5140 | static void |
5141 | trace_format_regs(struct ds *result, int level, const char *title, | |
5142 | struct ofproto_trace *trace) | |
5143 | { | |
5144 | size_t i; | |
5145 | ||
5146 | ds_put_char_multiple(result, '\t', level); | |
5147 | ds_put_format(result, "%s:", title); | |
5148 | for (i = 0; i < FLOW_N_REGS; i++) { | |
5149 | ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]); | |
5150 | } | |
5151 | ds_put_char(result, '\n'); | |
5152 | } | |
5153 | ||
abe529af BP |
5154 | static void |
5155 | trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule) | |
5156 | { | |
5157 | struct ofproto_trace *trace = CONTAINER_OF(ctx, struct ofproto_trace, ctx); | |
5158 | struct ds *result = trace->result; | |
5159 | ||
5160 | ds_put_char(result, '\n'); | |
5161 | trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace); | |
eb9e1c26 | 5162 | trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace); |
29901626 | 5163 | trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule); |
abe529af BP |
5164 | } |
5165 | ||
5166 | static void | |
5167 | ofproto_unixctl_trace(struct unixctl_conn *conn, const char *args_, | |
5168 | void *aux OVS_UNUSED) | |
5169 | { | |
abff858b | 5170 | char *dpname, *arg1, *arg2, *arg3, *arg4; |
abe529af BP |
5171 | char *args = xstrdup(args_); |
5172 | char *save_ptr = NULL; | |
5173 | struct ofproto_dpif *ofproto; | |
876b0e1c BP |
5174 | struct ofpbuf odp_key; |
5175 | struct ofpbuf *packet; | |
abe529af BP |
5176 | struct rule_dpif *rule; |
5177 | struct ds result; | |
5178 | struct flow flow; | |
abe529af BP |
5179 | char *s; |
5180 | ||
876b0e1c BP |
5181 | packet = NULL; |
5182 | ofpbuf_init(&odp_key, 0); | |
abe529af BP |
5183 | ds_init(&result); |
5184 | ||
5185 | dpname = strtok_r(args, " ", &save_ptr); | |
876b0e1c BP |
5186 | arg1 = strtok_r(NULL, " ", &save_ptr); |
5187 | arg2 = strtok_r(NULL, " ", &save_ptr); | |
abff858b PS |
5188 | arg3 = strtok_r(NULL, " ", &save_ptr); |
5189 | arg4 = strtok_r(NULL, "", &save_ptr); /* Get entire rest of line. */ | |
8b3b8dd1 BP |
5190 | if (dpname && arg1 && (!arg2 || !strcmp(arg2, "-generate")) && !arg3) { |
5191 | /* ofproto/trace dpname flow [-generate] */ | |
876b0e1c BP |
5192 | int error; |
5193 | ||
df2c07f4 | 5194 | /* Convert string to datapath key. */ |
876b0e1c | 5195 | ofpbuf_init(&odp_key, 0); |
b2a60db8 | 5196 | error = odp_flow_key_from_string(arg1, NULL, &odp_key); |
876b0e1c BP |
5197 | if (error) { |
5198 | unixctl_command_reply(conn, 501, "Bad flow syntax"); | |
5199 | goto exit; | |
5200 | } | |
5201 | ||
5202 | /* Convert odp_key to flow. */ | |
5203 | error = odp_flow_key_to_flow(odp_key.data, odp_key.size, &flow); | |
5204 | if (error) { | |
5205 | unixctl_command_reply(conn, 501, "Invalid flow"); | |
5206 | goto exit; | |
5207 | } | |
8b3b8dd1 BP |
5208 | |
5209 | /* Generate a packet, if requested. */ | |
5210 | if (arg2) { | |
5211 | packet = ofpbuf_new(0); | |
5212 | flow_compose(packet, &flow); | |
5213 | } | |
abff858b PS |
5214 | } else if (dpname && arg1 && arg2 && arg3 && arg4) { |
5215 | /* ofproto/trace dpname priority tun_id in_port packet */ | |
876b0e1c BP |
5216 | uint16_t in_port; |
5217 | ovs_be64 tun_id; | |
abff858b | 5218 | uint32_t priority; |
876b0e1c | 5219 | |
abff858b PS |
5220 | priority = atoi(arg1); |
5221 | tun_id = htonll(strtoull(arg2, NULL, 0)); | |
5222 | in_port = ofp_port_to_odp_port(atoi(arg3)); | |
876b0e1c BP |
5223 | |
5224 | packet = ofpbuf_new(strlen(args) / 2); | |
abff858b PS |
5225 | arg4 = ofpbuf_put_hex(packet, arg4, NULL); |
5226 | arg4 += strspn(arg4, " "); | |
5227 | if (*arg4 != '\0') { | |
876b0e1c BP |
5228 | unixctl_command_reply(conn, 501, "Trailing garbage in command"); |
5229 | goto exit; | |
5230 | } | |
5231 | if (packet->size < ETH_HEADER_LEN) { | |
5232 | unixctl_command_reply(conn, 501, | |
5233 | "Packet data too short for Ethernet"); | |
5234 | goto exit; | |
5235 | } | |
5236 | ||
5237 | ds_put_cstr(&result, "Packet: "); | |
5238 | s = ofp_packet_to_string(packet->data, packet->size, packet->size); | |
5239 | ds_put_cstr(&result, s); | |
5240 | free(s); | |
5241 | ||
abff858b | 5242 | flow_extract(packet, priority, tun_id, in_port, &flow); |
876b0e1c | 5243 | } else { |
abe529af BP |
5244 | unixctl_command_reply(conn, 501, "Bad command syntax"); |
5245 | goto exit; | |
5246 | } | |
5247 | ||
5248 | ofproto = ofproto_dpif_lookup(dpname); | |
5249 | if (!ofproto) { | |
5250 | unixctl_command_reply(conn, 501, "Unknown ofproto (use ofproto/list " | |
5251 | "for help)"); | |
5252 | goto exit; | |
5253 | } | |
5254 | ||
abe529af BP |
5255 | ds_put_cstr(&result, "Flow: "); |
5256 | flow_format(&result, &flow); | |
5257 | ds_put_char(&result, '\n'); | |
5258 | ||
29901626 BP |
5259 | rule = rule_dpif_lookup(ofproto, &flow, 0); |
5260 | trace_format_rule(&result, 0, 0, rule); | |
abe529af BP |
5261 | if (rule) { |
5262 | struct ofproto_trace trace; | |
5263 | struct ofpbuf *odp_actions; | |
5264 | ||
5265 | trace.result = &result; | |
5266 | trace.flow = flow; | |
876b0e1c | 5267 | action_xlate_ctx_init(&trace.ctx, ofproto, &flow, packet); |
abe529af BP |
5268 | trace.ctx.resubmit_hook = trace_resubmit; |
5269 | odp_actions = xlate_actions(&trace.ctx, | |
5270 | rule->up.actions, rule->up.n_actions); | |
5271 | ||
5272 | ds_put_char(&result, '\n'); | |
5273 | trace_format_flow(&result, 0, "Final flow", &trace); | |
5274 | ds_put_cstr(&result, "Datapath actions: "); | |
5275 | format_odp_actions(&result, odp_actions->data, odp_actions->size); | |
5276 | ofpbuf_delete(odp_actions); | |
876b0e1c BP |
5277 | |
5278 | if (!trace.ctx.may_set_up_flow) { | |
5279 | if (packet) { | |
5280 | ds_put_cstr(&result, "\nThis flow is not cachable."); | |
5281 | } else { | |
5282 | ds_put_cstr(&result, "\nThe datapath actions are incomplete--" | |
5283 | "for complete actions, please supply a packet."); | |
5284 | } | |
5285 | } | |
abe529af BP |
5286 | } |
5287 | ||
5288 | unixctl_command_reply(conn, 200, ds_cstr(&result)); | |
5289 | ||
5290 | exit: | |
5291 | ds_destroy(&result); | |
876b0e1c BP |
5292 | ofpbuf_delete(packet); |
5293 | ofpbuf_uninit(&odp_key); | |
abe529af BP |
5294 | free(args); |
5295 | } | |
5296 | ||
7ee20df1 BP |
5297 | static void |
5298 | ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, | |
5299 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
5300 | { | |
5301 | clogged = true; | |
5302 | unixctl_command_reply(conn, 200, NULL); | |
5303 | } | |
5304 | ||
5305 | static void | |
5306 | ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, | |
5307 | const char *args_ OVS_UNUSED, void *aux OVS_UNUSED) | |
5308 | { | |
5309 | clogged = false; | |
5310 | unixctl_command_reply(conn, 200, NULL); | |
5311 | } | |
5312 | ||
abe529af BP |
5313 | static void |
5314 | ofproto_dpif_unixctl_init(void) | |
5315 | { | |
5316 | static bool registered; | |
5317 | if (registered) { | |
5318 | return; | |
5319 | } | |
5320 | registered = true; | |
5321 | ||
7ff2009a JP |
5322 | unixctl_command_register("ofproto/trace", |
5323 | "bridge {tun_id in_port packet | odp_flow [-generate]}", | |
5324 | ofproto_unixctl_trace, NULL); | |
f0a3aa2e AA |
5325 | unixctl_command_register("fdb/flush", "bridge", ofproto_unixctl_fdb_flush, |
5326 | NULL); | |
7ff2009a JP |
5327 | unixctl_command_register("fdb/show", "bridge", ofproto_unixctl_fdb_show, |
5328 | NULL); | |
5329 | unixctl_command_register("ofproto/clog", "", ofproto_dpif_clog, NULL); | |
5330 | unixctl_command_register("ofproto/unclog", "", ofproto_dpif_unclog, NULL); | |
abe529af BP |
5331 | } |
5332 | \f | |
5333 | const struct ofproto_class ofproto_dpif_class = { | |
5334 | enumerate_types, | |
5335 | enumerate_names, | |
5336 | del, | |
5337 | alloc, | |
5338 | construct, | |
5339 | destruct, | |
5340 | dealloc, | |
5341 | run, | |
5342 | wait, | |
5343 | flush, | |
6c1491fb BP |
5344 | get_features, |
5345 | get_tables, | |
abe529af BP |
5346 | port_alloc, |
5347 | port_construct, | |
5348 | port_destruct, | |
5349 | port_dealloc, | |
5350 | port_modified, | |
5351 | port_reconfigured, | |
5352 | port_query_by_name, | |
5353 | port_add, | |
5354 | port_del, | |
5355 | port_dump_start, | |
5356 | port_dump_next, | |
5357 | port_dump_done, | |
5358 | port_poll, | |
5359 | port_poll_wait, | |
5360 | port_is_lacp_current, | |
0ab6decf | 5361 | NULL, /* rule_choose_table */ |
abe529af BP |
5362 | rule_alloc, |
5363 | rule_construct, | |
5364 | rule_destruct, | |
5365 | rule_dealloc, | |
abe529af BP |
5366 | rule_get_stats, |
5367 | rule_execute, | |
5368 | rule_modify_actions, | |
7257b535 | 5369 | set_frag_handling, |
abe529af BP |
5370 | packet_out, |
5371 | set_netflow, | |
5372 | get_netflow_ids, | |
5373 | set_sflow, | |
5374 | set_cfm, | |
a5610457 | 5375 | get_cfm_fault, |
1de11730 | 5376 | get_cfm_remote_mpids, |
21f7563c JP |
5377 | set_stp, |
5378 | get_stp_status, | |
5379 | set_stp_port, | |
5380 | get_stp_port_status, | |
abe529af BP |
5381 | bundle_set, |
5382 | bundle_remove, | |
5383 | mirror_set, | |
5384 | set_flood_vlans, | |
5385 | is_mirror_output_bundle, | |
8402c74b | 5386 | forward_bpdu_changed, |
abe529af | 5387 | }; |