]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
922fed06 | 2 | * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc. |
064af421 | 3 | * |
a14bc59f BP |
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: | |
064af421 | 7 | * |
a14bc59f BP |
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. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
96fba48f | 18 | #include "dpif-provider.h" |
064af421 | 19 | |
064af421 BP |
20 | #include <ctype.h> |
21 | #include <errno.h> | |
064af421 | 22 | #include <inttypes.h> |
064af421 BP |
23 | #include <stdlib.h> |
24 | #include <string.h> | |
064af421 BP |
25 | |
26 | #include "coverage.h" | |
fceef209 | 27 | #include "dpctl.h" |
e14deea0 | 28 | #include "dp-packet.h" |
c4ea7529 | 29 | #include "dpif-netdev.h" |
3e8a2ad1 | 30 | #include "openvswitch/dynamic-string.h" |
064af421 | 31 | #include "flow.h" |
c3827f61 | 32 | #include "netdev.h" |
064af421 | 33 | #include "netlink.h" |
7fd91025 | 34 | #include "odp-execute.h" |
064af421 | 35 | #include "odp-util.h" |
25d436fb | 36 | #include "openvswitch/ofp-print.h" |
f4248336 | 37 | #include "openvswitch/ofp-util.h" |
64c96779 | 38 | #include "openvswitch/ofpbuf.h" |
064af421 BP |
39 | #include "packets.h" |
40 | #include "poll-loop.h" | |
1bc50ef3 | 41 | #include "route-table.h" |
36f29fb1 | 42 | #include "seq.h" |
ee89ea7b | 43 | #include "openvswitch/shash.h" |
d0c23a1a | 44 | #include "sset.h" |
c97fb132 | 45 | #include "timeval.h" |
53902038 | 46 | #include "tnl-neigh-cache.h" |
a36de779 | 47 | #include "tnl-ports.h" |
064af421 | 48 | #include "util.h" |
78145f6e | 49 | #include "uuid.h" |
064af421 | 50 | #include "valgrind.h" |
e03c096d | 51 | #include "openvswitch/ofp-errors.h" |
e6211adc | 52 | #include "openvswitch/vlog.h" |
5136ce49 | 53 | |
d98e6007 | 54 | VLOG_DEFINE_THIS_MODULE(dpif); |
064af421 | 55 | |
d76f09ea BP |
56 | COVERAGE_DEFINE(dpif_destroy); |
57 | COVERAGE_DEFINE(dpif_port_add); | |
58 | COVERAGE_DEFINE(dpif_port_del); | |
59 | COVERAGE_DEFINE(dpif_flow_flush); | |
60 | COVERAGE_DEFINE(dpif_flow_get); | |
61 | COVERAGE_DEFINE(dpif_flow_put); | |
62 | COVERAGE_DEFINE(dpif_flow_del); | |
d76f09ea BP |
63 | COVERAGE_DEFINE(dpif_execute); |
64 | COVERAGE_DEFINE(dpif_purge); | |
7fd91025 | 65 | COVERAGE_DEFINE(dpif_execute_with_help); |
d76f09ea | 66 | |
999401aa | 67 | static const struct dpif_class *base_dpif_classes[] = { |
93451a0a AS |
68 | #if defined(__linux__) || defined(_WIN32) |
69 | &dpif_netlink_class, | |
c83cdd30 | 70 | #endif |
72865317 | 71 | &dpif_netdev_class, |
c228a364 | 72 | }; |
999401aa JG |
73 | |
74 | struct registered_dpif_class { | |
d2d8fbeb | 75 | const struct dpif_class *dpif_class; |
999401aa JG |
76 | int refcount; |
77 | }; | |
78 | static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes); | |
579a77e0 | 79 | static struct sset dpif_blacklist = SSET_INITIALIZER(&dpif_blacklist); |
c228a364 | 80 | |
5703b15f | 81 | /* Protects 'dpif_classes', including the refcount, and 'dpif_blacklist'. */ |
97be1538 | 82 | static struct ovs_mutex dpif_mutex = OVS_MUTEX_INITIALIZER; |
5703b15f | 83 | |
064af421 BP |
84 | /* Rate limit for individual messages going to or from the datapath, output at |
85 | * DBG level. This is very high because, if these are enabled, it is because | |
86 | * we really need to see them. */ | |
87 | static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600); | |
88 | ||
89 | /* Not really much point in logging many dpif errors. */ | |
e2781405 | 90 | static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5); |
064af421 | 91 | |
feebdea2 BP |
92 | static void log_flow_message(const struct dpif *dpif, int error, |
93 | const char *operation, | |
94 | const struct nlattr *key, size_t key_len, | |
61fb711d | 95 | const struct nlattr *mask, size_t mask_len, |
70e5ed6f | 96 | const ovs_u128 *ufid, |
c97fb132 | 97 | const struct dpif_flow_stats *stats, |
feebdea2 | 98 | const struct nlattr *actions, size_t actions_len); |
96fba48f BP |
99 | static void log_operation(const struct dpif *, const char *operation, |
100 | int error); | |
96fba48f | 101 | static bool should_log_flow_message(int error); |
89625d1e BP |
102 | static void log_flow_put_message(struct dpif *, const struct dpif_flow_put *, |
103 | int error); | |
b99d3cee BP |
104 | static void log_flow_del_message(struct dpif *, const struct dpif_flow_del *, |
105 | int error); | |
89625d1e | 106 | static void log_execute_message(struct dpif *, const struct dpif_execute *, |
fc65bafc | 107 | bool subexecute, int error); |
6fe09f8c JS |
108 | static void log_flow_get_message(const struct dpif *, |
109 | const struct dpif_flow_get *, int error); | |
064af421 | 110 | |
36f29fb1 PS |
111 | /* Incremented whenever tnl route, arp, etc changes. */ |
112 | struct seq *tnl_conf_seq; | |
113 | ||
999401aa JG |
114 | static void |
115 | dp_initialize(void) | |
116 | { | |
eb8ed438 | 117 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
999401aa | 118 | |
eb8ed438 | 119 | if (ovsthread_once_start(&once)) { |
999401aa JG |
120 | int i; |
121 | ||
36f29fb1 | 122 | tnl_conf_seq = seq_create(); |
fceef209 | 123 | dpctl_unixctl_register(); |
a36de779 | 124 | tnl_port_map_init(); |
53902038 | 125 | tnl_neigh_cache_init(); |
b772066f | 126 | route_table_init(); |
1bc50ef3 | 127 | |
36f29fb1 PS |
128 | for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) { |
129 | dp_register_provider(base_dpif_classes[i]); | |
130 | } | |
131 | ||
1bc50ef3 | 132 | ovsthread_once_done(&once); |
999401aa JG |
133 | } |
134 | } | |
135 | ||
5703b15f BP |
136 | static int |
137 | dp_register_provider__(const struct dpif_class *new_class) | |
999401aa JG |
138 | { |
139 | struct registered_dpif_class *registered_class; | |
c8973eb6 | 140 | int error; |
999401aa | 141 | |
579a77e0 EJ |
142 | if (sset_contains(&dpif_blacklist, new_class->type)) { |
143 | VLOG_DBG("attempted to register blacklisted provider: %s", | |
144 | new_class->type); | |
145 | return EINVAL; | |
146 | } | |
147 | ||
999401aa JG |
148 | if (shash_find(&dpif_classes, new_class->type)) { |
149 | VLOG_WARN("attempted to register duplicate datapath provider: %s", | |
150 | new_class->type); | |
151 | return EEXIST; | |
152 | } | |
1a6f1e2a | 153 | |
c8973eb6 DDP |
154 | error = new_class->init ? new_class->init() : 0; |
155 | if (error) { | |
156 | VLOG_WARN("failed to initialize %s datapath class: %s", | |
157 | new_class->type, ovs_strerror(error)); | |
158 | return error; | |
159 | } | |
160 | ||
999401aa | 161 | registered_class = xmalloc(sizeof *registered_class); |
d2d8fbeb | 162 | registered_class->dpif_class = new_class; |
999401aa JG |
163 | registered_class->refcount = 0; |
164 | ||
165 | shash_add(&dpif_classes, new_class->type, registered_class); | |
166 | ||
167 | return 0; | |
168 | } | |
169 | ||
5703b15f BP |
170 | /* Registers a new datapath provider. After successful registration, new |
171 | * datapaths of that type can be opened using dpif_open(). */ | |
172 | int | |
173 | dp_register_provider(const struct dpif_class *new_class) | |
174 | { | |
175 | int error; | |
176 | ||
97be1538 | 177 | ovs_mutex_lock(&dpif_mutex); |
5703b15f | 178 | error = dp_register_provider__(new_class); |
97be1538 | 179 | ovs_mutex_unlock(&dpif_mutex); |
5703b15f BP |
180 | |
181 | return error; | |
182 | } | |
183 | ||
999401aa JG |
184 | /* Unregisters a datapath provider. 'type' must have been previously |
185 | * registered and not currently be in use by any dpifs. After unregistration | |
186 | * new datapaths of that type cannot be opened using dpif_open(). */ | |
5703b15f BP |
187 | static int |
188 | dp_unregister_provider__(const char *type) | |
999401aa JG |
189 | { |
190 | struct shash_node *node; | |
191 | struct registered_dpif_class *registered_class; | |
192 | ||
193 | node = shash_find(&dpif_classes, type); | |
194 | if (!node) { | |
999401aa JG |
195 | return EAFNOSUPPORT; |
196 | } | |
197 | ||
198 | registered_class = node->data; | |
199 | if (registered_class->refcount) { | |
200 | VLOG_WARN("attempted to unregister in use datapath provider: %s", type); | |
201 | return EBUSY; | |
202 | } | |
203 | ||
204 | shash_delete(&dpif_classes, node); | |
205 | free(registered_class); | |
206 | ||
207 | return 0; | |
208 | } | |
209 | ||
5703b15f BP |
210 | /* Unregisters a datapath provider. 'type' must have been previously |
211 | * registered and not currently be in use by any dpifs. After unregistration | |
212 | * new datapaths of that type cannot be opened using dpif_open(). */ | |
213 | int | |
214 | dp_unregister_provider(const char *type) | |
215 | { | |
216 | int error; | |
217 | ||
218 | dp_initialize(); | |
219 | ||
97be1538 | 220 | ovs_mutex_lock(&dpif_mutex); |
5703b15f | 221 | error = dp_unregister_provider__(type); |
97be1538 | 222 | ovs_mutex_unlock(&dpif_mutex); |
5703b15f BP |
223 | |
224 | return error; | |
225 | } | |
226 | ||
579a77e0 EJ |
227 | /* Blacklists a provider. Causes future calls of dp_register_provider() with |
228 | * a dpif_class which implements 'type' to fail. */ | |
229 | void | |
230 | dp_blacklist_provider(const char *type) | |
231 | { | |
97be1538 | 232 | ovs_mutex_lock(&dpif_mutex); |
579a77e0 | 233 | sset_add(&dpif_blacklist, type); |
97be1538 | 234 | ovs_mutex_unlock(&dpif_mutex); |
579a77e0 EJ |
235 | } |
236 | ||
5b5b11ea BP |
237 | /* Adds the types of all currently registered datapath providers to 'types'. |
238 | * The caller must first initialize the sset. */ | |
1a6f1e2a | 239 | void |
d0c23a1a | 240 | dp_enumerate_types(struct sset *types) |
1a6f1e2a | 241 | { |
999401aa | 242 | struct shash_node *node; |
1a6f1e2a | 243 | |
999401aa | 244 | dp_initialize(); |
1a6f1e2a | 245 | |
97be1538 | 246 | ovs_mutex_lock(&dpif_mutex); |
999401aa JG |
247 | SHASH_FOR_EACH(node, &dpif_classes) { |
248 | const struct registered_dpif_class *registered_class = node->data; | |
d0c23a1a | 249 | sset_add(types, registered_class->dpif_class->type); |
1a6f1e2a | 250 | } |
97be1538 | 251 | ovs_mutex_unlock(&dpif_mutex); |
5703b15f BP |
252 | } |
253 | ||
254 | static void | |
255 | dp_class_unref(struct registered_dpif_class *rc) | |
256 | { | |
97be1538 | 257 | ovs_mutex_lock(&dpif_mutex); |
5703b15f BP |
258 | ovs_assert(rc->refcount); |
259 | rc->refcount--; | |
97be1538 | 260 | ovs_mutex_unlock(&dpif_mutex); |
5703b15f BP |
261 | } |
262 | ||
263 | static struct registered_dpif_class * | |
264 | dp_class_lookup(const char *type) | |
265 | { | |
266 | struct registered_dpif_class *rc; | |
267 | ||
97be1538 | 268 | ovs_mutex_lock(&dpif_mutex); |
5703b15f BP |
269 | rc = shash_find_data(&dpif_classes, type); |
270 | if (rc) { | |
271 | rc->refcount++; | |
272 | } | |
97be1538 | 273 | ovs_mutex_unlock(&dpif_mutex); |
5703b15f BP |
274 | |
275 | return rc; | |
1a6f1e2a JG |
276 | } |
277 | ||
278 | /* Clears 'names' and enumerates the names of all known created datapaths with | |
d0c23a1a | 279 | * the given 'type'. The caller must first initialize the sset. Returns 0 if |
1a6f1e2a | 280 | * successful, otherwise a positive errno value. |
d3d22744 BP |
281 | * |
282 | * Some kinds of datapaths might not be practically enumerable. This is not | |
283 | * considered an error. */ | |
284 | int | |
d0c23a1a | 285 | dp_enumerate_names(const char *type, struct sset *names) |
d3d22744 | 286 | { |
5703b15f | 287 | struct registered_dpif_class *registered_class; |
999401aa JG |
288 | const struct dpif_class *dpif_class; |
289 | int error; | |
d3d22744 | 290 | |
999401aa | 291 | dp_initialize(); |
d0c23a1a | 292 | sset_clear(names); |
1a6f1e2a | 293 | |
5703b15f | 294 | registered_class = dp_class_lookup(type); |
999401aa JG |
295 | if (!registered_class) { |
296 | VLOG_WARN("could not enumerate unknown type: %s", type); | |
297 | return EAFNOSUPPORT; | |
298 | } | |
1a6f1e2a | 299 | |
d2d8fbeb | 300 | dpif_class = registered_class->dpif_class; |
2240af25 DDP |
301 | error = (dpif_class->enumerate |
302 | ? dpif_class->enumerate(names, dpif_class) | |
303 | : 0); | |
999401aa JG |
304 | if (error) { |
305 | VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type, | |
10a89ef0 | 306 | ovs_strerror(error)); |
d3d22744 | 307 | } |
5703b15f | 308 | dp_class_unref(registered_class); |
1a6f1e2a | 309 | |
999401aa | 310 | return error; |
1a6f1e2a JG |
311 | } |
312 | ||
54ed8a5d BP |
313 | /* Parses 'datapath_name_', which is of the form [type@]name into its |
314 | * component pieces. 'name' and 'type' must be freed by the caller. | |
315 | * | |
316 | * The returned 'type' is normalized, as if by dpif_normalize_type(). */ | |
1a6f1e2a JG |
317 | void |
318 | dp_parse_name(const char *datapath_name_, char **name, char **type) | |
319 | { | |
320 | char *datapath_name = xstrdup(datapath_name_); | |
321 | char *separator; | |
322 | ||
323 | separator = strchr(datapath_name, '@'); | |
324 | if (separator) { | |
325 | *separator = '\0'; | |
326 | *type = datapath_name; | |
54ed8a5d | 327 | *name = xstrdup(dpif_normalize_type(separator + 1)); |
1a6f1e2a JG |
328 | } else { |
329 | *name = datapath_name; | |
54ed8a5d | 330 | *type = xstrdup(dpif_normalize_type(NULL)); |
1a6f1e2a | 331 | } |
d3d22744 BP |
332 | } |
333 | ||
96fba48f | 334 | static int |
1a6f1e2a | 335 | do_open(const char *name, const char *type, bool create, struct dpif **dpifp) |
064af421 | 336 | { |
96fba48f | 337 | struct dpif *dpif = NULL; |
064af421 | 338 | int error; |
999401aa JG |
339 | struct registered_dpif_class *registered_class; |
340 | ||
341 | dp_initialize(); | |
064af421 | 342 | |
3a225db7 | 343 | type = dpif_normalize_type(type); |
5703b15f | 344 | registered_class = dp_class_lookup(type); |
999401aa JG |
345 | if (!registered_class) { |
346 | VLOG_WARN("could not create datapath %s of unknown type %s", name, | |
347 | type); | |
348 | error = EAFNOSUPPORT; | |
349 | goto exit; | |
350 | } | |
351 | ||
4a387741 BP |
352 | error = registered_class->dpif_class->open(registered_class->dpif_class, |
353 | name, create, &dpif); | |
999401aa | 354 | if (!error) { |
cb22974d | 355 | ovs_assert(dpif->dpif_class == registered_class->dpif_class); |
5703b15f BP |
356 | } else { |
357 | dp_class_unref(registered_class); | |
064af421 | 358 | } |
064af421 | 359 | |
96fba48f BP |
360 | exit: |
361 | *dpifp = error ? NULL : dpif; | |
362 | return error; | |
064af421 BP |
363 | } |
364 | ||
1a6f1e2a JG |
365 | /* Tries to open an existing datapath named 'name' and type 'type'. Will fail |
366 | * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or | |
367 | * the empty string to specify the default system type. Returns 0 if | |
368 | * successful, otherwise a positive errno value. On success stores a pointer | |
369 | * to the datapath in '*dpifp', otherwise a null pointer. */ | |
96fba48f | 370 | int |
1a6f1e2a | 371 | dpif_open(const char *name, const char *type, struct dpif **dpifp) |
064af421 | 372 | { |
1a6f1e2a | 373 | return do_open(name, type, false, dpifp); |
064af421 BP |
374 | } |
375 | ||
1a6f1e2a JG |
376 | /* Tries to create and open a new datapath with the given 'name' and 'type'. |
377 | * 'type' may be either NULL or the empty string to specify the default system | |
378 | * type. Will fail if a datapath with 'name' and 'type' already exists. | |
379 | * Returns 0 if successful, otherwise a positive errno value. On success | |
380 | * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */ | |
064af421 | 381 | int |
1a6f1e2a | 382 | dpif_create(const char *name, const char *type, struct dpif **dpifp) |
064af421 | 383 | { |
1a6f1e2a | 384 | return do_open(name, type, true, dpifp); |
96fba48f | 385 | } |
064af421 | 386 | |
1a6f1e2a JG |
387 | /* Tries to open a datapath with the given 'name' and 'type', creating it if it |
388 | * does not exist. 'type' may be either NULL or the empty string to specify | |
389 | * the default system type. Returns 0 if successful, otherwise a positive | |
390 | * errno value. On success stores a pointer to the datapath in '*dpifp', | |
391 | * otherwise a null pointer. */ | |
efacbce6 | 392 | int |
1a6f1e2a | 393 | dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp) |
efacbce6 BP |
394 | { |
395 | int error; | |
396 | ||
1a6f1e2a | 397 | error = dpif_create(name, type, dpifp); |
efacbce6 | 398 | if (error == EEXIST || error == EBUSY) { |
1a6f1e2a | 399 | error = dpif_open(name, type, dpifp); |
efacbce6 BP |
400 | if (error) { |
401 | VLOG_WARN("datapath %s already exists but cannot be opened: %s", | |
10a89ef0 | 402 | name, ovs_strerror(error)); |
efacbce6 BP |
403 | } |
404 | } else if (error) { | |
10a89ef0 BP |
405 | VLOG_WARN("failed to create datapath %s: %s", |
406 | name, ovs_strerror(error)); | |
efacbce6 BP |
407 | } |
408 | return error; | |
409 | } | |
410 | ||
96fba48f BP |
411 | /* Closes and frees the connection to 'dpif'. Does not destroy the datapath |
412 | * itself; call dpif_delete() first, instead, if that is desirable. */ | |
413 | void | |
414 | dpif_close(struct dpif *dpif) | |
415 | { | |
416 | if (dpif) { | |
5703b15f | 417 | struct registered_dpif_class *rc; |
999401aa | 418 | |
5703b15f | 419 | rc = shash_find_data(&dpif_classes, dpif->dpif_class->type); |
999401aa | 420 | dpif_uninit(dpif, true); |
5703b15f | 421 | dp_class_unref(rc); |
064af421 BP |
422 | } |
423 | } | |
424 | ||
640e1b20 | 425 | /* Performs periodic work needed by 'dpif'. */ |
a36de779 | 426 | bool |
640e1b20 BP |
427 | dpif_run(struct dpif *dpif) |
428 | { | |
429 | if (dpif->dpif_class->run) { | |
a36de779 | 430 | return dpif->dpif_class->run(dpif); |
640e1b20 | 431 | } |
a36de779 | 432 | return false; |
640e1b20 BP |
433 | } |
434 | ||
435 | /* Arranges for poll_block() to wake up when dp_run() needs to be called for | |
436 | * 'dpif'. */ | |
437 | void | |
438 | dpif_wait(struct dpif *dpif) | |
439 | { | |
440 | if (dpif->dpif_class->wait) { | |
441 | dpif->dpif_class->wait(dpif); | |
442 | } | |
443 | } | |
444 | ||
1a6f1e2a JG |
445 | /* Returns the name of datapath 'dpif' prefixed with the type |
446 | * (for use in log messages). */ | |
b29ba128 BP |
447 | const char * |
448 | dpif_name(const struct dpif *dpif) | |
449 | { | |
1a6f1e2a JG |
450 | return dpif->full_name; |
451 | } | |
452 | ||
453 | /* Returns the name of datapath 'dpif' without the type | |
454 | * (for use in device names). */ | |
455 | const char * | |
456 | dpif_base_name(const struct dpif *dpif) | |
457 | { | |
458 | return dpif->base_name; | |
b29ba128 BP |
459 | } |
460 | ||
c7a26215 JP |
461 | /* Returns the type of datapath 'dpif'. */ |
462 | const char * | |
463 | dpif_type(const struct dpif *dpif) | |
464 | { | |
465 | return dpif->dpif_class->type; | |
466 | } | |
467 | ||
3a225db7 BP |
468 | /* Returns the fully spelled out name for the given datapath 'type'. |
469 | * | |
470 | * Normalized type string can be compared with strcmp(). Unnormalized type | |
471 | * string might be the same even if they have different spellings. */ | |
472 | const char * | |
473 | dpif_normalize_type(const char *type) | |
474 | { | |
475 | return type && type[0] ? type : "system"; | |
476 | } | |
477 | ||
96fba48f BP |
478 | /* Destroys the datapath that 'dpif' is connected to, first removing all of its |
479 | * ports. After calling this function, it does not make sense to pass 'dpif' | |
480 | * to any functions other than dpif_name() or dpif_close(). */ | |
064af421 BP |
481 | int |
482 | dpif_delete(struct dpif *dpif) | |
483 | { | |
96fba48f BP |
484 | int error; |
485 | ||
064af421 | 486 | COVERAGE_INC(dpif_destroy); |
96fba48f | 487 | |
1acb6baa | 488 | error = dpif->dpif_class->destroy(dpif); |
96fba48f BP |
489 | log_operation(dpif, "delete", error); |
490 | return error; | |
064af421 BP |
491 | } |
492 | ||
96fba48f BP |
493 | /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful, |
494 | * otherwise a positive errno value. */ | |
064af421 | 495 | int |
a8d9304d | 496 | dpif_get_dp_stats(const struct dpif *dpif, struct dpif_dp_stats *stats) |
064af421 | 497 | { |
1acb6baa | 498 | int error = dpif->dpif_class->get_stats(dpif, stats); |
96fba48f BP |
499 | if (error) { |
500 | memset(stats, 0, sizeof *stats); | |
501 | } | |
502 | log_operation(dpif, "get_stats", error); | |
503 | return error; | |
064af421 BP |
504 | } |
505 | ||
0aeaabc8 JP |
506 | const char * |
507 | dpif_port_open_type(const char *datapath_type, const char *port_type) | |
508 | { | |
5703b15f | 509 | struct registered_dpif_class *rc; |
0aeaabc8 JP |
510 | |
511 | datapath_type = dpif_normalize_type(datapath_type); | |
512 | ||
97be1538 | 513 | ovs_mutex_lock(&dpif_mutex); |
5703b15f BP |
514 | rc = shash_find_data(&dpif_classes, datapath_type); |
515 | if (rc && rc->dpif_class->port_open_type) { | |
516 | port_type = rc->dpif_class->port_open_type(rc->dpif_class, port_type); | |
0aeaabc8 | 517 | } |
97be1538 | 518 | ovs_mutex_unlock(&dpif_mutex); |
0aeaabc8 | 519 | |
5703b15f | 520 | return port_type; |
0aeaabc8 JP |
521 | } |
522 | ||
232dfa4a | 523 | /* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is |
4e022ec0 | 524 | * non-null and its value is not ODPP_NONE, then attempts to use the |
232dfa4a JP |
525 | * value as the port number. |
526 | * | |
527 | * If successful, returns 0 and sets '*port_nop' to the new port's port | |
528 | * number (if 'port_nop' is non-null). On failure, returns a positive | |
4e022ec0 | 529 | * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is |
232dfa4a | 530 | * non-null). */ |
064af421 | 531 | int |
4e022ec0 | 532 | dpif_port_add(struct dpif *dpif, struct netdev *netdev, odp_port_t *port_nop) |
064af421 | 533 | { |
c3827f61 | 534 | const char *netdev_name = netdev_get_name(netdev); |
4e022ec0 | 535 | odp_port_t port_no = ODPP_NONE; |
9ee3ae3e | 536 | int error; |
064af421 BP |
537 | |
538 | COVERAGE_INC(dpif_port_add); | |
9ee3ae3e | 539 | |
232dfa4a JP |
540 | if (port_nop) { |
541 | port_no = *port_nop; | |
542 | } | |
543 | ||
c3827f61 | 544 | error = dpif->dpif_class->port_add(dpif, netdev, &port_no); |
9ee3ae3e | 545 | if (!error) { |
9b56fe13 | 546 | VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu32, |
c3827f61 | 547 | dpif_name(dpif), netdev_name, port_no); |
064af421 | 548 | } else { |
9ee3ae3e | 549 | VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s", |
10a89ef0 | 550 | dpif_name(dpif), netdev_name, ovs_strerror(error)); |
4e022ec0 | 551 | port_no = ODPP_NONE; |
9ee3ae3e BP |
552 | } |
553 | if (port_nop) { | |
554 | *port_nop = port_no; | |
064af421 | 555 | } |
9ee3ae3e | 556 | return error; |
064af421 BP |
557 | } |
558 | ||
96fba48f BP |
559 | /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful, |
560 | * otherwise a positive errno value. */ | |
064af421 | 561 | int |
4e022ec0 | 562 | dpif_port_del(struct dpif *dpif, odp_port_t port_no) |
064af421 | 563 | { |
96fba48f BP |
564 | int error; |
565 | ||
064af421 | 566 | COVERAGE_INC(dpif_port_del); |
96fba48f | 567 | |
1acb6baa | 568 | error = dpif->dpif_class->port_del(dpif, port_no); |
a1811296 | 569 | if (!error) { |
9b56fe13 | 570 | VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu32")", |
a1811296 BP |
571 | dpif_name(dpif), port_no); |
572 | } else { | |
573 | log_operation(dpif, "port_del", error); | |
574 | } | |
96fba48f | 575 | return error; |
064af421 BP |
576 | } |
577 | ||
4c738a8d BP |
578 | /* Makes a deep copy of 'src' into 'dst'. */ |
579 | void | |
580 | dpif_port_clone(struct dpif_port *dst, const struct dpif_port *src) | |
581 | { | |
582 | dst->name = xstrdup(src->name); | |
583 | dst->type = xstrdup(src->type); | |
584 | dst->port_no = src->port_no; | |
585 | } | |
586 | ||
587 | /* Frees memory allocated to members of 'dpif_port'. | |
588 | * | |
589 | * Do not call this function on a dpif_port obtained from | |
590 | * dpif_port_dump_next(): that function retains ownership of the data in the | |
591 | * dpif_port. */ | |
592 | void | |
593 | dpif_port_destroy(struct dpif_port *dpif_port) | |
594 | { | |
595 | free(dpif_port->name); | |
596 | free(dpif_port->type); | |
597 | } | |
598 | ||
4afba28d JP |
599 | /* Checks if port named 'devname' exists in 'dpif'. If so, returns |
600 | * true; otherwise, returns false. */ | |
601 | bool | |
602 | dpif_port_exists(const struct dpif *dpif, const char *devname) | |
603 | { | |
604 | int error = dpif->dpif_class->port_query_by_name(dpif, devname, NULL); | |
0f6a066f | 605 | if (error != 0 && error != ENODEV) { |
4afba28d | 606 | VLOG_WARN_RL(&error_rl, "%s: failed to query port %s: %s", |
10a89ef0 | 607 | dpif_name(dpif), devname, ovs_strerror(error)); |
4afba28d JP |
608 | } |
609 | ||
610 | return !error; | |
611 | } | |
612 | ||
91364d18 IM |
613 | /* Refreshes configuration of 'dpif's port. */ |
614 | int | |
615 | dpif_port_set_config(struct dpif *dpif, odp_port_t port_no, | |
616 | const struct smap *cfg) | |
617 | { | |
618 | int error = 0; | |
619 | ||
620 | if (dpif->dpif_class->port_set_config) { | |
621 | error = dpif->dpif_class->port_set_config(dpif, port_no, cfg); | |
622 | if (error) { | |
623 | log_operation(dpif, "port_set_config", error); | |
624 | } | |
625 | } | |
626 | ||
627 | return error; | |
628 | } | |
629 | ||
96fba48f BP |
630 | /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and |
631 | * initializes '*port' appropriately; on failure, returns a positive errno | |
4c738a8d BP |
632 | * value. |
633 | * | |
0f6a066f DDP |
634 | * Retuns ENODEV if the port doesn't exist. |
635 | * | |
4c738a8d BP |
636 | * The caller owns the data in 'port' and must free it with |
637 | * dpif_port_destroy() when it is no longer needed. */ | |
064af421 | 638 | int |
4e022ec0 | 639 | dpif_port_query_by_number(const struct dpif *dpif, odp_port_t port_no, |
4c738a8d | 640 | struct dpif_port *port) |
064af421 | 641 | { |
1acb6baa | 642 | int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port); |
96fba48f | 643 | if (!error) { |
9b56fe13 | 644 | VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu32" is device %s", |
4c738a8d | 645 | dpif_name(dpif), port_no, port->name); |
064af421 | 646 | } else { |
96fba48f | 647 | memset(port, 0, sizeof *port); |
9b56fe13 | 648 | VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu32": %s", |
10a89ef0 | 649 | dpif_name(dpif), port_no, ovs_strerror(error)); |
064af421 | 650 | } |
96fba48f | 651 | return error; |
064af421 BP |
652 | } |
653 | ||
96fba48f BP |
654 | /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and |
655 | * initializes '*port' appropriately; on failure, returns a positive errno | |
4c738a8d BP |
656 | * value. |
657 | * | |
0f6a066f DDP |
658 | * Retuns ENODEV if the port doesn't exist. |
659 | * | |
4c738a8d BP |
660 | * The caller owns the data in 'port' and must free it with |
661 | * dpif_port_destroy() when it is no longer needed. */ | |
064af421 BP |
662 | int |
663 | dpif_port_query_by_name(const struct dpif *dpif, const char *devname, | |
4c738a8d | 664 | struct dpif_port *port) |
064af421 | 665 | { |
1acb6baa | 666 | int error = dpif->dpif_class->port_query_by_name(dpif, devname, port); |
96fba48f | 667 | if (!error) { |
9b56fe13 | 668 | VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu32, |
4c738a8d | 669 | dpif_name(dpif), devname, port->port_no); |
064af421 | 670 | } else { |
96fba48f BP |
671 | memset(port, 0, sizeof *port); |
672 | ||
0f6a066f | 673 | /* For ENODEV we use DBG level because the caller is probably |
d647f0a7 BP |
674 | * interested in whether 'dpif' actually has a port 'devname', so that |
675 | * it's not an issue worth logging if it doesn't. Other errors are | |
676 | * uncommon and more likely to indicate a real problem. */ | |
0f6a066f | 677 | VLOG_RL(&error_rl, error == ENODEV ? VLL_DBG : VLL_WARN, |
d647f0a7 | 678 | "%s: failed to query port %s: %s", |
10a89ef0 | 679 | dpif_name(dpif), devname, ovs_strerror(error)); |
064af421 | 680 | } |
96fba48f | 681 | return error; |
064af421 BP |
682 | } |
683 | ||
1954e6bb AW |
684 | /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE |
685 | * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in | |
686 | * flows whose packets arrived on port 'port_no'. In the case where the | |
687 | * provider allocates multiple Netlink PIDs to a single port, it may use | |
688 | * 'hash' to spread load among them. The caller need not use a particular | |
689 | * hash function; a 5-tuple hash is suitable. | |
690 | * | |
691 | * (The datapath implementation might use some different hash function for | |
692 | * distributing packets received via flow misses among PIDs. This means | |
693 | * that packets received via flow misses might be reordered relative to | |
694 | * packets received via userspace actions. This is not ordinarily a | |
695 | * problem.) | |
98403001 | 696 | * |
4e022ec0 | 697 | * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not |
625b0720 BP |
698 | * allocated to any port, that the client may use for special purposes. |
699 | * | |
98403001 BP |
700 | * The return value is only meaningful when DPIF_UC_ACTION has been enabled in |
701 | * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is | |
702 | * disabled and then re-enabled, so a client that does that must be prepared to | |
703 | * update all of the flows that it installed that contain | |
704 | * OVS_ACTION_ATTR_USERSPACE actions. */ | |
705 | uint32_t | |
1954e6bb | 706 | dpif_port_get_pid(const struct dpif *dpif, odp_port_t port_no, uint32_t hash) |
98403001 BP |
707 | { |
708 | return (dpif->dpif_class->port_get_pid | |
1954e6bb | 709 | ? (dpif->dpif_class->port_get_pid)(dpif, port_no, hash) |
98403001 BP |
710 | : 0); |
711 | } | |
712 | ||
96fba48f BP |
713 | /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies |
714 | * the port's name into the 'name_size' bytes in 'name', ensuring that the | |
715 | * result is null-terminated. On failure, returns a positive errno value and | |
716 | * makes 'name' the empty string. */ | |
335562c0 | 717 | int |
4e022ec0 | 718 | dpif_port_get_name(struct dpif *dpif, odp_port_t port_no, |
335562c0 BP |
719 | char *name, size_t name_size) |
720 | { | |
4c738a8d | 721 | struct dpif_port port; |
335562c0 BP |
722 | int error; |
723 | ||
cb22974d | 724 | ovs_assert(name_size > 0); |
335562c0 BP |
725 | |
726 | error = dpif_port_query_by_number(dpif, port_no, &port); | |
727 | if (!error) { | |
4c738a8d BP |
728 | ovs_strlcpy(name, port.name, name_size); |
729 | dpif_port_destroy(&port); | |
335562c0 BP |
730 | } else { |
731 | *name = '\0'; | |
732 | } | |
733 | return error; | |
734 | } | |
735 | ||
b0ec0f27 | 736 | /* Initializes 'dump' to begin dumping the ports in a dpif. |
96fba48f | 737 | * |
b0ec0f27 BP |
738 | * This function provides no status indication. An error status for the entire |
739 | * dump operation is provided when it is completed by calling | |
740 | * dpif_port_dump_done(). | |
741 | */ | |
742 | void | |
743 | dpif_port_dump_start(struct dpif_port_dump *dump, const struct dpif *dpif) | |
744 | { | |
745 | dump->dpif = dpif; | |
746 | dump->error = dpif->dpif_class->port_dump_start(dpif, &dump->state); | |
747 | log_operation(dpif, "port_dump_start", dump->error); | |
748 | } | |
749 | ||
750 | /* Attempts to retrieve another port from 'dump', which must have been | |
4c738a8d | 751 | * initialized with dpif_port_dump_start(). On success, stores a new dpif_port |
b0ec0f27 | 752 | * into 'port' and returns true. On failure, returns false. |
96fba48f | 753 | * |
b0ec0f27 BP |
754 | * Failure might indicate an actual error or merely that the last port has been |
755 | * dumped. An error status for the entire dump operation is provided when it | |
4c738a8d BP |
756 | * is completed by calling dpif_port_dump_done(). |
757 | * | |
758 | * The dpif owns the data stored in 'port'. It will remain valid until at | |
759 | * least the next time 'dump' is passed to dpif_port_dump_next() or | |
760 | * dpif_port_dump_done(). */ | |
b0ec0f27 | 761 | bool |
4c738a8d | 762 | dpif_port_dump_next(struct dpif_port_dump *dump, struct dpif_port *port) |
064af421 | 763 | { |
b0ec0f27 | 764 | const struct dpif *dpif = dump->dpif; |
064af421 | 765 | |
b0ec0f27 BP |
766 | if (dump->error) { |
767 | return false; | |
768 | } | |
f4ba4c4f | 769 | |
b0ec0f27 BP |
770 | dump->error = dpif->dpif_class->port_dump_next(dpif, dump->state, port); |
771 | if (dump->error == EOF) { | |
772 | VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all ports", dpif_name(dpif)); | |
773 | } else { | |
774 | log_operation(dpif, "port_dump_next", dump->error); | |
775 | } | |
064af421 | 776 | |
b0ec0f27 BP |
777 | if (dump->error) { |
778 | dpif->dpif_class->port_dump_done(dpif, dump->state); | |
779 | return false; | |
f4ba4c4f | 780 | } |
b0ec0f27 BP |
781 | return true; |
782 | } | |
064af421 | 783 | |
b0ec0f27 BP |
784 | /* Completes port table dump operation 'dump', which must have been initialized |
785 | * with dpif_port_dump_start(). Returns 0 if the dump operation was | |
786 | * error-free, otherwise a positive errno value describing the problem. */ | |
787 | int | |
788 | dpif_port_dump_done(struct dpif_port_dump *dump) | |
789 | { | |
790 | const struct dpif *dpif = dump->dpif; | |
791 | if (!dump->error) { | |
792 | dump->error = dpif->dpif_class->port_dump_done(dpif, dump->state); | |
793 | log_operation(dpif, "port_dump_done", dump->error); | |
f4ba4c4f | 794 | } |
b0ec0f27 | 795 | return dump->error == EOF ? 0 : dump->error; |
064af421 BP |
796 | } |
797 | ||
e9e28be3 BP |
798 | /* Polls for changes in the set of ports in 'dpif'. If the set of ports in |
799 | * 'dpif' has changed, this function does one of the following: | |
800 | * | |
801 | * - Stores the name of the device that was added to or deleted from 'dpif' in | |
802 | * '*devnamep' and returns 0. The caller is responsible for freeing | |
803 | * '*devnamep' (with free()) when it no longer needs it. | |
804 | * | |
805 | * - Returns ENOBUFS and sets '*devnamep' to NULL. | |
806 | * | |
807 | * This function may also return 'false positives', where it returns 0 and | |
808 | * '*devnamep' names a device that was not actually added or deleted or it | |
809 | * returns ENOBUFS without any change. | |
810 | * | |
811 | * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also | |
812 | * return other positive errno values to indicate that something has gone | |
813 | * wrong. */ | |
814 | int | |
815 | dpif_port_poll(const struct dpif *dpif, char **devnamep) | |
816 | { | |
1acb6baa | 817 | int error = dpif->dpif_class->port_poll(dpif, devnamep); |
e9e28be3 BP |
818 | if (error) { |
819 | *devnamep = NULL; | |
820 | } | |
821 | return error; | |
822 | } | |
823 | ||
824 | /* Arranges for the poll loop to wake up when port_poll(dpif) will return a | |
825 | * value other than EAGAIN. */ | |
826 | void | |
827 | dpif_port_poll_wait(const struct dpif *dpif) | |
828 | { | |
1acb6baa | 829 | dpif->dpif_class->port_poll_wait(dpif); |
e9e28be3 BP |
830 | } |
831 | ||
572b7068 | 832 | /* Extracts the flow stats for a packet. The 'flow' and 'packet' |
a7752d4a BP |
833 | * arguments must have been initialized through a call to flow_extract(). |
834 | * 'used' is stored into stats->used. */ | |
572b7068 | 835 | void |
cf62fa4c | 836 | dpif_flow_stats_extract(const struct flow *flow, const struct dp_packet *packet, |
a7752d4a | 837 | long long int used, struct dpif_flow_stats *stats) |
572b7068 | 838 | { |
e0eecb1c | 839 | stats->tcp_flags = ntohs(flow->tcp_flags); |
cf62fa4c | 840 | stats->n_bytes = dp_packet_size(packet); |
572b7068 | 841 | stats->n_packets = 1; |
a7752d4a | 842 | stats->used = used; |
572b7068 BP |
843 | } |
844 | ||
c97fb132 BP |
845 | /* Appends a human-readable representation of 'stats' to 's'. */ |
846 | void | |
847 | dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s) | |
848 | { | |
849 | ds_put_format(s, "packets:%"PRIu64", bytes:%"PRIu64", used:", | |
850 | stats->n_packets, stats->n_bytes); | |
851 | if (stats->used) { | |
852 | ds_put_format(s, "%.3fs", (time_msec() - stats->used) / 1000.0); | |
853 | } else { | |
854 | ds_put_format(s, "never"); | |
855 | } | |
7393104d BP |
856 | if (stats->tcp_flags) { |
857 | ds_put_cstr(s, ", flags:"); | |
858 | packet_format_tcp_flags(s, stats->tcp_flags); | |
859 | } | |
c97fb132 BP |
860 | } |
861 | ||
7af12bd7 JS |
862 | /* Places the hash of the 'key_len' bytes starting at 'key' into '*hash'. */ |
863 | void | |
864 | dpif_flow_hash(const struct dpif *dpif OVS_UNUSED, | |
865 | const void *key, size_t key_len, ovs_u128 *hash) | |
866 | { | |
867 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
868 | static uint32_t secret; | |
869 | ||
870 | if (ovsthread_once_start(&once)) { | |
871 | secret = random_uint32(); | |
872 | ovsthread_once_done(&once); | |
873 | } | |
874 | hash_bytes128(key, key_len, secret, hash); | |
78145f6e | 875 | uuid_set_bits_v4((struct uuid *)hash); |
7af12bd7 JS |
876 | } |
877 | ||
96fba48f BP |
878 | /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a |
879 | * positive errno value. */ | |
880 | int | |
881 | dpif_flow_flush(struct dpif *dpif) | |
064af421 | 882 | { |
96fba48f BP |
883 | int error; |
884 | ||
885 | COVERAGE_INC(dpif_flow_flush); | |
886 | ||
1acb6baa | 887 | error = dpif->dpif_class->flow_flush(dpif); |
96fba48f BP |
888 | log_operation(dpif, "flow_flush", error); |
889 | return error; | |
064af421 BP |
890 | } |
891 | ||
2c85851f JS |
892 | /* Attempts to install 'key' into the datapath, fetches it, then deletes it. |
893 | * Returns true if the datapath supported installing 'flow', false otherwise. | |
894 | */ | |
895 | bool | |
896 | dpif_probe_feature(struct dpif *dpif, const char *name, | |
897 | const struct ofpbuf *key, const ovs_u128 *ufid) | |
898 | { | |
899 | struct dpif_flow flow; | |
900 | struct ofpbuf reply; | |
901 | uint64_t stub[DPIF_FLOW_BUFSIZE / 8]; | |
902 | bool enable_feature = false; | |
903 | int error; | |
904 | ||
9ab0fce1 JS |
905 | /* Use DPIF_FP_MODIFY to cover the case where ovs-vswitchd is killed (and |
906 | * restarted) at just the right time such that feature probes from the | |
907 | * previous run are still present in the datapath. */ | |
908 | error = dpif_flow_put(dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY | DPIF_FP_PROBE, | |
6fd6ed71 | 909 | key->data, key->size, NULL, 0, NULL, 0, |
1c1e46ed | 910 | ufid, PMD_ID_NULL, NULL); |
9ab0fce1 | 911 | if (error) { |
2c85851f JS |
912 | if (error != EINVAL) { |
913 | VLOG_WARN("%s: %s flow probe failed (%s)", | |
914 | dpif_name(dpif), name, ovs_strerror(error)); | |
915 | } | |
916 | return false; | |
917 | } | |
918 | ||
919 | ofpbuf_use_stack(&reply, &stub, sizeof stub); | |
6fd6ed71 | 920 | error = dpif_flow_get(dpif, key->data, key->size, ufid, |
1c1e46ed | 921 | PMD_ID_NULL, &reply, &flow); |
2c85851f | 922 | if (!error |
bdd7ecf5 | 923 | && (!ufid || (flow.ufid_present |
2ff8484b | 924 | && ovs_u128_equals(*ufid, flow.ufid)))) { |
2c85851f JS |
925 | enable_feature = true; |
926 | } | |
927 | ||
6fd6ed71 | 928 | error = dpif_flow_del(dpif, key->data, key->size, ufid, |
1c1e46ed | 929 | PMD_ID_NULL, NULL); |
2c85851f JS |
930 | if (error) { |
931 | VLOG_WARN("%s: failed to delete %s feature probe flow", | |
932 | dpif_name(dpif), name); | |
933 | } | |
934 | ||
935 | return enable_feature; | |
936 | } | |
937 | ||
6fe09f8c | 938 | /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_GET. */ |
96fba48f | 939 | int |
6fe09f8c | 940 | dpif_flow_get(struct dpif *dpif, |
70e5ed6f | 941 | const struct nlattr *key, size_t key_len, const ovs_u128 *ufid, |
bd5131ba | 942 | const unsigned pmd_id, struct ofpbuf *buf, struct dpif_flow *flow) |
064af421 | 943 | { |
6fe09f8c JS |
944 | struct dpif_op *opp; |
945 | struct dpif_op op; | |
96fba48f | 946 | |
6fe09f8c JS |
947 | op.type = DPIF_OP_FLOW_GET; |
948 | op.u.flow_get.key = key; | |
949 | op.u.flow_get.key_len = key_len; | |
70e5ed6f | 950 | op.u.flow_get.ufid = ufid; |
1c1e46ed | 951 | op.u.flow_get.pmd_id = pmd_id; |
6fe09f8c | 952 | op.u.flow_get.buffer = buf; |
70e5ed6f JS |
953 | |
954 | memset(flow, 0, sizeof *flow); | |
6fe09f8c JS |
955 | op.u.flow_get.flow = flow; |
956 | op.u.flow_get.flow->key = key; | |
957 | op.u.flow_get.flow->key_len = key_len; | |
96fba48f | 958 | |
6fe09f8c JS |
959 | opp = &op; |
960 | dpif_operate(dpif, &opp, 1); | |
961 | ||
962 | return op.error; | |
064af421 BP |
963 | } |
964 | ||
1a0c894a | 965 | /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_PUT. */ |
064af421 | 966 | int |
ba25b8f4 | 967 | dpif_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags, |
feebdea2 | 968 | const struct nlattr *key, size_t key_len, |
e6cc0bab | 969 | const struct nlattr *mask, size_t mask_len, |
feebdea2 | 970 | const struct nlattr *actions, size_t actions_len, |
bd5131ba | 971 | const ovs_u128 *ufid, const unsigned pmd_id, |
1c1e46ed | 972 | struct dpif_flow_stats *stats) |
064af421 | 973 | { |
1a0c894a BP |
974 | struct dpif_op *opp; |
975 | struct dpif_op op; | |
976 | ||
977 | op.type = DPIF_OP_FLOW_PUT; | |
978 | op.u.flow_put.flags = flags; | |
979 | op.u.flow_put.key = key; | |
980 | op.u.flow_put.key_len = key_len; | |
981 | op.u.flow_put.mask = mask; | |
982 | op.u.flow_put.mask_len = mask_len; | |
983 | op.u.flow_put.actions = actions; | |
984 | op.u.flow_put.actions_len = actions_len; | |
70e5ed6f | 985 | op.u.flow_put.ufid = ufid; |
1c1e46ed | 986 | op.u.flow_put.pmd_id = pmd_id; |
1a0c894a BP |
987 | op.u.flow_put.stats = stats; |
988 | ||
989 | opp = &op; | |
990 | dpif_operate(dpif, &opp, 1); | |
991 | ||
992 | return op.error; | |
064af421 BP |
993 | } |
994 | ||
1a0c894a | 995 | /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_DEL. */ |
064af421 | 996 | int |
feebdea2 | 997 | dpif_flow_del(struct dpif *dpif, |
70e5ed6f | 998 | const struct nlattr *key, size_t key_len, const ovs_u128 *ufid, |
bd5131ba | 999 | const unsigned pmd_id, struct dpif_flow_stats *stats) |
064af421 | 1000 | { |
1a0c894a BP |
1001 | struct dpif_op *opp; |
1002 | struct dpif_op op; | |
1003 | ||
1004 | op.type = DPIF_OP_FLOW_DEL; | |
1005 | op.u.flow_del.key = key; | |
1006 | op.u.flow_del.key_len = key_len; | |
70e5ed6f | 1007 | op.u.flow_del.ufid = ufid; |
1c1e46ed | 1008 | op.u.flow_del.pmd_id = pmd_id; |
1a0c894a | 1009 | op.u.flow_del.stats = stats; |
8e1ffd75 | 1010 | op.u.flow_del.terse = false; |
f1aa2072 | 1011 | |
1a0c894a BP |
1012 | opp = &op; |
1013 | dpif_operate(dpif, &opp, 1); | |
1014 | ||
1015 | return op.error; | |
064af421 BP |
1016 | } |
1017 | ||
ac64794a | 1018 | /* Creates and returns a new 'struct dpif_flow_dump' for iterating through the |
64bb477f JS |
1019 | * flows in 'dpif'. If 'terse' is true, then only UFID and statistics will |
1020 | * be returned in the dump. Otherwise, all fields will be returned. | |
ac64794a BP |
1021 | * |
1022 | * This function always successfully returns a dpif_flow_dump. Error | |
1023 | * reporting is deferred to dpif_flow_dump_destroy(). */ | |
1024 | struct dpif_flow_dump * | |
64bb477f | 1025 | dpif_flow_dump_create(const struct dpif *dpif, bool terse) |
e723fd32 | 1026 | { |
64bb477f | 1027 | return dpif->dpif_class->flow_dump_create(dpif, terse); |
e723fd32 JS |
1028 | } |
1029 | ||
ac64794a BP |
1030 | /* Destroys 'dump', which must have been created with dpif_flow_dump_create(). |
1031 | * All dpif_flow_dump_thread structures previously created for 'dump' must | |
1032 | * previously have been destroyed. | |
1033 | * | |
1034 | * Returns 0 if the dump operation was error-free, otherwise a positive errno | |
1035 | * value describing the problem. */ | |
1036 | int | |
1037 | dpif_flow_dump_destroy(struct dpif_flow_dump *dump) | |
e723fd32 | 1038 | { |
ac64794a BP |
1039 | const struct dpif *dpif = dump->dpif; |
1040 | int error = dpif->dpif_class->flow_dump_destroy(dump); | |
1041 | log_operation(dpif, "flow_dump_destroy", error); | |
1042 | return error == EOF ? 0 : error; | |
e723fd32 JS |
1043 | } |
1044 | ||
ac64794a BP |
1045 | /* Returns new thread-local state for use with dpif_flow_dump_next(). */ |
1046 | struct dpif_flow_dump_thread * | |
1047 | dpif_flow_dump_thread_create(struct dpif_flow_dump *dump) | |
064af421 | 1048 | { |
ac64794a | 1049 | return dump->dpif->dpif_class->flow_dump_thread_create(dump); |
064af421 BP |
1050 | } |
1051 | ||
ac64794a BP |
1052 | /* Releases 'thread'. */ |
1053 | void | |
1054 | dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread) | |
704a1e09 | 1055 | { |
ac64794a | 1056 | thread->dpif->dpif_class->flow_dump_thread_destroy(thread); |
704a1e09 BP |
1057 | } |
1058 | ||
ac64794a BP |
1059 | /* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0 |
1060 | * if and only if no flows remained to be retrieved, otherwise a positive | |
1061 | * number reflecting the number of elements in 'flows[]' that were updated. | |
1062 | * The number of flows returned might be less than 'max_flows' because | |
1063 | * fewer than 'max_flows' remained, because this particular datapath does not | |
1064 | * benefit from batching, or because an error occurred partway through | |
1065 | * retrieval. Thus, the caller should continue calling until a 0 return value, | |
1066 | * even if intermediate return values are less than 'max_flows'. | |
bdeadfdd | 1067 | * |
ac64794a BP |
1068 | * No error status is immediately provided. An error status for the entire |
1069 | * dump operation is provided when it is completed by calling | |
1070 | * dpif_flow_dump_destroy(). | |
bdeadfdd | 1071 | * |
ac64794a BP |
1072 | * All of the data stored into 'flows' is owned by the datapath, not by the |
1073 | * caller, and the caller must not modify or free it. The datapath guarantees | |
58df55ce JS |
1074 | * that it remains accessible and unchanged until the first of: |
1075 | * - The next call to dpif_flow_dump_next() for 'thread', or | |
1076 | * - The next rcu quiescent period. */ | |
704a1e09 | 1077 | int |
ac64794a BP |
1078 | dpif_flow_dump_next(struct dpif_flow_dump_thread *thread, |
1079 | struct dpif_flow *flows, int max_flows) | |
704a1e09 | 1080 | { |
ac64794a BP |
1081 | struct dpif *dpif = thread->dpif; |
1082 | int n; | |
1083 | ||
1084 | ovs_assert(max_flows > 0); | |
1085 | n = dpif->dpif_class->flow_dump_next(thread, flows, max_flows); | |
1086 | if (n > 0) { | |
1087 | struct dpif_flow *f; | |
1088 | ||
1089 | for (f = flows; f < &flows[n] && should_log_flow_message(0); f++) { | |
1090 | log_flow_message(dpif, 0, "flow_dump", | |
1091 | f->key, f->key_len, f->mask, f->mask_len, | |
70e5ed6f | 1092 | &f->ufid, &f->stats, f->actions, f->actions_len); |
ac64794a BP |
1093 | } |
1094 | } else { | |
1095 | VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif)); | |
1096 | } | |
1097 | return n; | |
064af421 BP |
1098 | } |
1099 | ||
7fd91025 BP |
1100 | struct dpif_execute_helper_aux { |
1101 | struct dpif *dpif; | |
1cceb31b | 1102 | const struct flow *flow; |
7fd91025 BP |
1103 | int error; |
1104 | }; | |
1105 | ||
09f9da0b JR |
1106 | /* This is called for actions that need the context of the datapath to be |
1107 | * meaningful. */ | |
7fd91025 | 1108 | static void |
1895cc8d | 1109 | dpif_execute_helper_cb(void *aux_, struct dp_packet_batch *packets_, |
18b58592 | 1110 | const struct nlattr *action, bool may_steal) |
7fd91025 | 1111 | { |
758c456d | 1112 | struct dpif_execute_helper_aux *aux = aux_; |
09f9da0b | 1113 | int type = nl_attr_type(action); |
1895cc8d | 1114 | struct dp_packet *packet = packets_->packets[0]; |
8cbf4f47 | 1115 | |
1895cc8d | 1116 | ovs_assert(packets_->count == 1); |
758c456d | 1117 | |
09f9da0b | 1118 | switch ((enum ovs_action_attr)type) { |
07659514 | 1119 | case OVS_ACTION_ATTR_CT: |
09f9da0b | 1120 | case OVS_ACTION_ATTR_OUTPUT: |
a36de779 PS |
1121 | case OVS_ACTION_ATTR_TUNNEL_PUSH: |
1122 | case OVS_ACTION_ATTR_TUNNEL_POP: | |
09f9da0b | 1123 | case OVS_ACTION_ATTR_USERSPACE: |
2b651e44 BP |
1124 | case OVS_ACTION_ATTR_RECIRC: { |
1125 | struct dpif_execute execute; | |
1126 | struct ofpbuf execute_actions; | |
1127 | uint64_t stub[256 / 8]; | |
cf62fa4c | 1128 | struct pkt_metadata *md = &packet->md; |
18b58592 | 1129 | bool dst_set; |
2b651e44 | 1130 | |
ffe4c74f JB |
1131 | dst_set = flow_tnl_dst_is_set(&md->tunnel); |
1132 | if (dst_set) { | |
2b651e44 BP |
1133 | /* The Linux kernel datapath throws away the tunnel information |
1134 | * that we supply as metadata. We have to use a "set" action to | |
1135 | * supply it. */ | |
1136 | ofpbuf_use_stub(&execute_actions, stub, sizeof stub); | |
1137 | odp_put_tunnel_action(&md->tunnel, &execute_actions); | |
1138 | ofpbuf_put(&execute_actions, action, NLA_ALIGN(action->nla_len)); | |
1139 | ||
6fd6ed71 PS |
1140 | execute.actions = execute_actions.data; |
1141 | execute.actions_len = execute_actions.size; | |
2b651e44 BP |
1142 | } else { |
1143 | execute.actions = action; | |
1144 | execute.actions_len = NLA_ALIGN(action->nla_len); | |
1145 | } | |
1146 | ||
18b58592 AZ |
1147 | struct dp_packet *clone = NULL; |
1148 | uint32_t cutlen = dp_packet_get_cutlen(packet); | |
1149 | if (cutlen && (type == OVS_ACTION_ATTR_OUTPUT | |
1150 | || type == OVS_ACTION_ATTR_TUNNEL_PUSH | |
1151 | || type == OVS_ACTION_ATTR_TUNNEL_POP | |
1152 | || type == OVS_ACTION_ATTR_USERSPACE)) { | |
1153 | dp_packet_reset_cutlen(packet); | |
aaca4fe0 | 1154 | if (!may_steal) { |
18b58592 | 1155 | packet = clone = dp_packet_clone(packet); |
aaca4fe0 | 1156 | } |
aaca4fe0 | 1157 | dp_packet_set_size(packet, dp_packet_size(packet) - cutlen); |
aaca4fe0 WT |
1158 | } |
1159 | ||
8cbf4f47 | 1160 | execute.packet = packet; |
1cceb31b | 1161 | execute.flow = aux->flow; |
758c456d | 1162 | execute.needs_help = false; |
43f9ac0a | 1163 | execute.probe = false; |
27130224 | 1164 | execute.mtu = 0; |
1a0c894a | 1165 | aux->error = dpif_execute(aux->dpif, &execute); |
fc65bafc BP |
1166 | log_execute_message(aux->dpif, &execute, true, aux->error); |
1167 | ||
18b58592 AZ |
1168 | dp_packet_delete(clone); |
1169 | ||
ffe4c74f | 1170 | if (dst_set) { |
2b651e44 BP |
1171 | ofpbuf_uninit(&execute_actions); |
1172 | } | |
09f9da0b | 1173 | break; |
2b651e44 | 1174 | } |
758c456d | 1175 | |
c6bf49f3 | 1176 | case OVS_ACTION_ATTR_HASH: |
09f9da0b JR |
1177 | case OVS_ACTION_ATTR_PUSH_VLAN: |
1178 | case OVS_ACTION_ATTR_POP_VLAN: | |
1179 | case OVS_ACTION_ATTR_PUSH_MPLS: | |
1180 | case OVS_ACTION_ATTR_POP_MPLS: | |
1181 | case OVS_ACTION_ATTR_SET: | |
6d670e7f | 1182 | case OVS_ACTION_ATTR_SET_MASKED: |
09f9da0b | 1183 | case OVS_ACTION_ATTR_SAMPLE: |
aaca4fe0 | 1184 | case OVS_ACTION_ATTR_TRUNC: |
09f9da0b JR |
1185 | case OVS_ACTION_ATTR_UNSPEC: |
1186 | case __OVS_ACTION_ATTR_MAX: | |
1187 | OVS_NOT_REACHED(); | |
1188 | } | |
7fd91025 BP |
1189 | } |
1190 | ||
1191 | /* Executes 'execute' by performing most of the actions in userspace and | |
1192 | * passing the fully constructed packets to 'dpif' for output and userspace | |
1193 | * actions. | |
1194 | * | |
1195 | * This helps with actions that a given 'dpif' doesn't implement directly. */ | |
1196 | static int | |
758c456d | 1197 | dpif_execute_with_help(struct dpif *dpif, struct dpif_execute *execute) |
7fd91025 | 1198 | { |
1cceb31b | 1199 | struct dpif_execute_helper_aux aux = {dpif, execute->flow, 0}; |
1895cc8d | 1200 | struct dp_packet_batch pb; |
7fd91025 BP |
1201 | |
1202 | COVERAGE_INC(dpif_execute_with_help); | |
1203 | ||
1895cc8d PS |
1204 | packet_batch_init_packet(&pb, execute->packet); |
1205 | odp_execute_actions(&aux, &pb, false, execute->actions, | |
91088554 | 1206 | execute->actions_len, dpif_execute_helper_cb); |
7fd91025 BP |
1207 | return aux.error; |
1208 | } | |
1209 | ||
87e5119b BP |
1210 | /* Returns true if the datapath needs help executing 'execute'. */ |
1211 | static bool | |
1212 | dpif_execute_needs_help(const struct dpif_execute *execute) | |
1213 | { | |
1214 | return execute->needs_help || nl_attr_oversized(execute->actions_len); | |
1215 | } | |
1216 | ||
1a0c894a | 1217 | /* A dpif_operate() wrapper for performing a single DPIF_OP_EXECUTE. */ |
758c456d JR |
1218 | int |
1219 | dpif_execute(struct dpif *dpif, struct dpif_execute *execute) | |
89625d1e | 1220 | { |
1a0c894a BP |
1221 | if (execute->actions_len) { |
1222 | struct dpif_op *opp; | |
1223 | struct dpif_op op; | |
89625d1e | 1224 | |
1a0c894a BP |
1225 | op.type = DPIF_OP_EXECUTE; |
1226 | op.u.execute = *execute; | |
89625d1e | 1227 | |
1a0c894a BP |
1228 | opp = &op; |
1229 | dpif_operate(dpif, &opp, 1); | |
89625d1e | 1230 | |
1a0c894a BP |
1231 | return op.error; |
1232 | } else { | |
1233 | return 0; | |
1234 | } | |
89625d1e BP |
1235 | } |
1236 | ||
6bc60024 | 1237 | /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in |
1a0c894a BP |
1238 | * which they are specified. Places each operation's results in the "output" |
1239 | * members documented in comments, and 0 in the 'error' member on success or a | |
1240 | * positive errno on failure. */ | |
6bc60024 | 1241 | void |
c2b565b5 | 1242 | dpif_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) |
6bc60024 | 1243 | { |
1a0c894a BP |
1244 | while (n_ops > 0) { |
1245 | size_t chunk; | |
7fd91025 | 1246 | |
1a0c894a BP |
1247 | /* Count 'chunk', the number of ops that can be executed without |
1248 | * needing any help. Ops that need help should be rare, so we | |
1249 | * expect this to ordinarily be 'n_ops', that is, all the ops. */ | |
1250 | for (chunk = 0; chunk < n_ops; chunk++) { | |
1251 | struct dpif_op *op = ops[chunk]; | |
1252 | ||
1253 | if (op->type == DPIF_OP_EXECUTE | |
1254 | && dpif_execute_needs_help(&op->u.execute)) { | |
1255 | break; | |
1256 | } | |
1257 | } | |
7fd91025 | 1258 | |
1a0c894a BP |
1259 | if (chunk) { |
1260 | /* Execute a chunk full of ops that the dpif provider can | |
1261 | * handle itself, without help. */ | |
1262 | size_t i; | |
7fd91025 | 1263 | |
1a0c894a | 1264 | dpif->dpif_class->operate(dpif, ops, chunk); |
7fd91025 | 1265 | |
1a0c894a BP |
1266 | for (i = 0; i < chunk; i++) { |
1267 | struct dpif_op *op = ops[i]; | |
1268 | int error = op->error; | |
7fd91025 | 1269 | |
1a0c894a BP |
1270 | switch (op->type) { |
1271 | case DPIF_OP_FLOW_PUT: { | |
1272 | struct dpif_flow_put *put = &op->u.flow_put; | |
7fd91025 | 1273 | |
1a0c894a BP |
1274 | COVERAGE_INC(dpif_flow_put); |
1275 | log_flow_put_message(dpif, put, error); | |
1276 | if (error && put->stats) { | |
1277 | memset(put->stats, 0, sizeof *put->stats); | |
7fd91025 | 1278 | } |
1a0c894a | 1279 | break; |
7fd91025 BP |
1280 | } |
1281 | ||
6fe09f8c JS |
1282 | case DPIF_OP_FLOW_GET: { |
1283 | struct dpif_flow_get *get = &op->u.flow_get; | |
1284 | ||
1285 | COVERAGE_INC(dpif_flow_get); | |
6fe09f8c JS |
1286 | if (error) { |
1287 | memset(get->flow, 0, sizeof *get->flow); | |
1288 | } | |
72d52166 MC |
1289 | log_flow_get_message(dpif, get, error); |
1290 | ||
6fe09f8c JS |
1291 | break; |
1292 | } | |
1293 | ||
1a0c894a BP |
1294 | case DPIF_OP_FLOW_DEL: { |
1295 | struct dpif_flow_del *del = &op->u.flow_del; | |
7fd91025 | 1296 | |
1a0c894a BP |
1297 | COVERAGE_INC(dpif_flow_del); |
1298 | log_flow_del_message(dpif, del, error); | |
1299 | if (error && del->stats) { | |
1300 | memset(del->stats, 0, sizeof *del->stats); | |
1301 | } | |
1302 | break; | |
1303 | } | |
f23d2845 | 1304 | |
1a0c894a BP |
1305 | case DPIF_OP_EXECUTE: |
1306 | COVERAGE_INC(dpif_execute); | |
1307 | log_execute_message(dpif, &op->u.execute, false, error); | |
1308 | break; | |
1309 | } | |
1310 | } | |
b99d3cee | 1311 | |
1a0c894a BP |
1312 | ops += chunk; |
1313 | n_ops -= chunk; | |
1314 | } else { | |
1315 | /* Help the dpif provider to execute one op. */ | |
1316 | struct dpif_op *op = ops[0]; | |
b99d3cee | 1317 | |
1a0c894a BP |
1318 | COVERAGE_INC(dpif_execute); |
1319 | op->error = dpif_execute_with_help(dpif, &op->u.execute); | |
1320 | ops++; | |
1321 | n_ops--; | |
6bc60024 BP |
1322 | } |
1323 | } | |
1324 | } | |
1325 | ||
01545c1a BP |
1326 | /* Returns a string that represents 'type', for use in log messages. */ |
1327 | const char * | |
1328 | dpif_upcall_type_to_string(enum dpif_upcall_type type) | |
1329 | { | |
1330 | switch (type) { | |
1331 | case DPIF_UC_MISS: return "miss"; | |
1332 | case DPIF_UC_ACTION: return "action"; | |
01545c1a BP |
1333 | case DPIF_N_UC_TYPES: default: return "<unknown>"; |
1334 | } | |
1335 | } | |
1336 | ||
a12b3ead BP |
1337 | /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0 |
1338 | * if successful, otherwise a positive errno value. | |
98403001 | 1339 | * |
a12b3ead | 1340 | * Turning packet receive off and then back on may change the Netlink PID |
98403001 BP |
1341 | * assignments returned by dpif_port_get_pid(). If the client does this, it |
1342 | * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions | |
1343 | * using the new PID assignment. */ | |
8f24562a | 1344 | int |
a12b3ead | 1345 | dpif_recv_set(struct dpif *dpif, bool enable) |
8f24562a | 1346 | { |
6b31e073 RW |
1347 | int error = 0; |
1348 | ||
1349 | if (dpif->dpif_class->recv_set) { | |
1350 | error = dpif->dpif_class->recv_set(dpif, enable); | |
1351 | log_operation(dpif, "recv_set", error); | |
1352 | } | |
96fba48f | 1353 | return error; |
8f24562a BP |
1354 | } |
1355 | ||
1954e6bb AW |
1356 | /* Refreshes the poll loops and Netlink sockets associated to each port, |
1357 | * when the number of upcall handlers (upcall receiving thread) is changed | |
1358 | * to 'n_handlers' and receiving packets for 'dpif' is enabled by | |
1359 | * recv_set(). | |
1360 | * | |
1361 | * Since multiple upcall handlers can read upcalls simultaneously from | |
1362 | * 'dpif', each port can have multiple Netlink sockets, one per upcall | |
1363 | * handler. So, handlers_set() is responsible for the following tasks: | |
1364 | * | |
1365 | * When receiving upcall is enabled, extends or creates the | |
1366 | * configuration to support: | |
1367 | * | |
1368 | * - 'n_handlers' Netlink sockets for each port. | |
1369 | * | |
1370 | * - 'n_handlers' poll loops, one for each upcall handler. | |
1371 | * | |
1372 | * - registering the Netlink sockets for the same upcall handler to | |
1373 | * the corresponding poll loop. | |
1374 | * | |
1375 | * Returns 0 if successful, otherwise a positive errno value. */ | |
1376 | int | |
1377 | dpif_handlers_set(struct dpif *dpif, uint32_t n_handlers) | |
1378 | { | |
6b31e073 RW |
1379 | int error = 0; |
1380 | ||
1381 | if (dpif->dpif_class->handlers_set) { | |
1382 | error = dpif->dpif_class->handlers_set(dpif, n_handlers); | |
1383 | log_operation(dpif, "handlers_set", error); | |
1384 | } | |
1954e6bb AW |
1385 | return error; |
1386 | } | |
1387 | ||
e4e74c3a AW |
1388 | void |
1389 | dpif_register_dp_purge_cb(struct dpif *dpif, dp_purge_callback *cb, void *aux) | |
1390 | { | |
1391 | if (dpif->dpif_class->register_dp_purge_cb) { | |
1392 | dpif->dpif_class->register_dp_purge_cb(dpif, cb, aux); | |
1393 | } | |
1394 | } | |
1395 | ||
6b31e073 | 1396 | void |
623540e4 | 1397 | dpif_register_upcall_cb(struct dpif *dpif, upcall_callback *cb, void *aux) |
6b31e073 RW |
1398 | { |
1399 | if (dpif->dpif_class->register_upcall_cb) { | |
623540e4 | 1400 | dpif->dpif_class->register_upcall_cb(dpif, cb, aux); |
6b31e073 RW |
1401 | } |
1402 | } | |
1403 | ||
1404 | void | |
1405 | dpif_enable_upcall(struct dpif *dpif) | |
1406 | { | |
1407 | if (dpif->dpif_class->enable_upcall) { | |
1408 | dpif->dpif_class->enable_upcall(dpif); | |
1409 | } | |
1410 | } | |
1411 | ||
1412 | void | |
1413 | dpif_disable_upcall(struct dpif *dpif) | |
1414 | { | |
1415 | if (dpif->dpif_class->disable_upcall) { | |
1416 | dpif->dpif_class->disable_upcall(dpif); | |
1417 | } | |
1418 | } | |
1419 | ||
1420 | void | |
1421 | dpif_print_packet(struct dpif *dpif, struct dpif_upcall *upcall) | |
1422 | { | |
1423 | if (!VLOG_DROP_DBG(&dpmsg_rl)) { | |
1424 | struct ds flow; | |
1425 | char *packet; | |
1426 | ||
cf62fa4c PS |
1427 | packet = ofp_packet_to_string(dp_packet_data(&upcall->packet), |
1428 | dp_packet_size(&upcall->packet)); | |
6b31e073 RW |
1429 | |
1430 | ds_init(&flow); | |
1431 | odp_flow_key_format(upcall->key, upcall->key_len, &flow); | |
1432 | ||
1433 | VLOG_DBG("%s: %s upcall:\n%s\n%s", | |
1434 | dpif_name(dpif), dpif_upcall_type_to_string(upcall->type), | |
1435 | ds_cstr(&flow), packet); | |
1436 | ||
1437 | ds_destroy(&flow); | |
1438 | free(packet); | |
1439 | } | |
1440 | } | |
1441 | ||
f2eee189 AW |
1442 | /* If 'dpif' creates its own I/O polling threads, refreshes poll threads |
1443 | * configuration. */ | |
1444 | int | |
a14b8947 | 1445 | dpif_poll_threads_set(struct dpif *dpif, const char *cmask) |
f2eee189 AW |
1446 | { |
1447 | int error = 0; | |
1448 | ||
1449 | if (dpif->dpif_class->poll_threads_set) { | |
a14b8947 | 1450 | error = dpif->dpif_class->poll_threads_set(dpif, cmask); |
f2eee189 AW |
1451 | if (error) { |
1452 | log_operation(dpif, "poll_threads_set", error); | |
1453 | } | |
1454 | } | |
1455 | ||
1456 | return error; | |
1457 | } | |
1458 | ||
1954e6bb AW |
1459 | /* Polls for an upcall from 'dpif' for an upcall handler. Since there |
1460 | * there can be multiple poll loops, 'handler_id' is needed as index to | |
1461 | * identify the corresponding poll loop. If successful, stores the upcall | |
1462 | * into '*upcall', using 'buf' for storage. Should only be called if | |
1463 | * 'recv_set' has been used to enable receiving packets from 'dpif'. | |
96fba48f | 1464 | * |
da546e07 JR |
1465 | * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided |
1466 | * 'buf', so their memory cannot be freed separately from 'buf'. | |
856081f6 | 1467 | * |
837a88dc JR |
1468 | * The caller owns the data of 'upcall->packet' and may modify it. If |
1469 | * packet's headroom is exhausted as it is manipulated, 'upcall->packet' | |
1470 | * will be reallocated. This requires the data of 'upcall->packet' to be | |
1471 | * released with ofpbuf_uninit() before 'upcall' is destroyed. However, | |
1472 | * when an error is returned, the 'upcall->packet' may be uninitialized | |
1473 | * and should not be released. | |
1474 | * | |
96fba48f | 1475 | * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN |
856081f6 | 1476 | * if no upcall is immediately available. */ |
064af421 | 1477 | int |
1954e6bb AW |
1478 | dpif_recv(struct dpif *dpif, uint32_t handler_id, struct dpif_upcall *upcall, |
1479 | struct ofpbuf *buf) | |
064af421 | 1480 | { |
6b31e073 | 1481 | int error = EAGAIN; |
01545c1a | 1482 | |
6b31e073 RW |
1483 | if (dpif->dpif_class->recv) { |
1484 | error = dpif->dpif_class->recv(dpif, handler_id, upcall, buf); | |
1485 | if (!error) { | |
1486 | dpif_print_packet(dpif, upcall); | |
1487 | } else if (error != EAGAIN) { | |
1488 | log_operation(dpif, "recv", error); | |
1489 | } | |
064af421 | 1490 | } |
064af421 BP |
1491 | return error; |
1492 | } | |
1493 | ||
96fba48f | 1494 | /* Discards all messages that would otherwise be received by dpif_recv() on |
1ba530f4 BP |
1495 | * 'dpif'. */ |
1496 | void | |
96fba48f BP |
1497 | dpif_recv_purge(struct dpif *dpif) |
1498 | { | |
96fba48f | 1499 | COVERAGE_INC(dpif_purge); |
1ba530f4 BP |
1500 | if (dpif->dpif_class->recv_purge) { |
1501 | dpif->dpif_class->recv_purge(dpif); | |
96fba48f | 1502 | } |
96fba48f BP |
1503 | } |
1504 | ||
1954e6bb AW |
1505 | /* Arranges for the poll loop for an upcall handler to wake up when 'dpif' |
1506 | * 'dpif' has a message queued to be received with the recv member | |
1507 | * function. Since there can be multiple poll loops, 'handler_id' is | |
1508 | * needed as index to identify the corresponding poll loop. */ | |
064af421 | 1509 | void |
1954e6bb | 1510 | dpif_recv_wait(struct dpif *dpif, uint32_t handler_id) |
064af421 | 1511 | { |
6b31e073 RW |
1512 | if (dpif->dpif_class->recv_wait) { |
1513 | dpif->dpif_class->recv_wait(dpif, handler_id); | |
1514 | } | |
064af421 | 1515 | } |
53a4218d | 1516 | |
b5cbbcf6 AZ |
1517 | /* |
1518 | * Return the datapath version. Caller is responsible for freeing | |
1519 | * the string. | |
1520 | */ | |
1521 | char * | |
1522 | dpif_get_dp_version(const struct dpif *dpif) | |
1523 | { | |
1524 | char *version = NULL; | |
1525 | ||
1526 | if (dpif->dpif_class->get_datapath_version) { | |
1527 | version = dpif->dpif_class->get_datapath_version(); | |
1528 | } | |
1529 | ||
1530 | return version; | |
1531 | } | |
1532 | ||
96fba48f BP |
1533 | /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type' |
1534 | * and '*engine_id', respectively. */ | |
53a4218d BP |
1535 | void |
1536 | dpif_get_netflow_ids(const struct dpif *dpif, | |
1537 | uint8_t *engine_type, uint8_t *engine_id) | |
1538 | { | |
96fba48f BP |
1539 | *engine_type = dpif->netflow_engine_type; |
1540 | *engine_id = dpif->netflow_engine_id; | |
1541 | } | |
aae51f53 BP |
1542 | |
1543 | /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority | |
abff858b PS |
1544 | * value used for setting packet priority. |
1545 | * On success, returns 0 and stores the priority into '*priority'. | |
1546 | * On failure, returns a positive errno value and stores 0 into '*priority'. */ | |
aae51f53 BP |
1547 | int |
1548 | dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id, | |
1549 | uint32_t *priority) | |
1550 | { | |
1551 | int error = (dpif->dpif_class->queue_to_priority | |
1552 | ? dpif->dpif_class->queue_to_priority(dpif, queue_id, | |
1553 | priority) | |
1554 | : EOPNOTSUPP); | |
1555 | if (error) { | |
1556 | *priority = 0; | |
1557 | } | |
1558 | log_operation(dpif, "queue_to_priority", error); | |
1559 | return error; | |
1560 | } | |
96fba48f BP |
1561 | \f |
1562 | void | |
1acb6baa BP |
1563 | dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class, |
1564 | const char *name, | |
96fba48f BP |
1565 | uint8_t netflow_engine_type, uint8_t netflow_engine_id) |
1566 | { | |
1acb6baa | 1567 | dpif->dpif_class = dpif_class; |
1a6f1e2a | 1568 | dpif->base_name = xstrdup(name); |
a4af0040 | 1569 | dpif->full_name = xasprintf("%s@%s", dpif_class->type, name); |
96fba48f BP |
1570 | dpif->netflow_engine_type = netflow_engine_type; |
1571 | dpif->netflow_engine_id = netflow_engine_id; | |
1572 | } | |
999401aa JG |
1573 | |
1574 | /* Undoes the results of initialization. | |
1575 | * | |
1576 | * Normally this function only needs to be called from dpif_close(). | |
1577 | * However, it may be called by providers due to an error on opening | |
1578 | * that occurs after initialization. It this case dpif_close() would | |
1579 | * never be called. */ | |
1580 | void | |
1581 | dpif_uninit(struct dpif *dpif, bool close) | |
1582 | { | |
1583 | char *base_name = dpif->base_name; | |
1584 | char *full_name = dpif->full_name; | |
1585 | ||
1586 | if (close) { | |
a4af0040 | 1587 | dpif->dpif_class->close(dpif); |
999401aa JG |
1588 | } |
1589 | ||
1590 | free(base_name); | |
1591 | free(full_name); | |
1592 | } | |
96fba48f BP |
1593 | \f |
1594 | static void | |
1595 | log_operation(const struct dpif *dpif, const char *operation, int error) | |
1596 | { | |
1597 | if (!error) { | |
1598 | VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation); | |
90bf1e07 | 1599 | } else if (ofperr_is_valid(error)) { |
96fba48f | 1600 | VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)", |
90bf1e07 | 1601 | dpif_name(dpif), operation, ofperr_get_name(error)); |
71ce9235 | 1602 | } else { |
90bf1e07 | 1603 | VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)", |
10a89ef0 | 1604 | dpif_name(dpif), operation, ovs_strerror(error)); |
96fba48f BP |
1605 | } |
1606 | } | |
1607 | ||
1608 | static enum vlog_level | |
1609 | flow_message_log_level(int error) | |
1610 | { | |
9b1a48c2 JP |
1611 | /* If flows arrive in a batch, userspace may push down multiple |
1612 | * unique flow definitions that overlap when wildcards are applied. | |
1613 | * Kernels that support flow wildcarding will reject these flows as | |
1614 | * duplicates (EEXIST), so lower the log level to debug for these | |
1615 | * types of messages. */ | |
1616 | return (error && error != EEXIST) ? VLL_WARN : VLL_DBG; | |
96fba48f BP |
1617 | } |
1618 | ||
1619 | static bool | |
1620 | should_log_flow_message(int error) | |
1621 | { | |
922fed06 | 1622 | return !vlog_should_drop(&this_module, flow_message_log_level(error), |
96fba48f BP |
1623 | error ? &error_rl : &dpmsg_rl); |
1624 | } | |
1625 | ||
1626 | static void | |
1627 | log_flow_message(const struct dpif *dpif, int error, const char *operation, | |
36956a7d | 1628 | const struct nlattr *key, size_t key_len, |
61fb711d | 1629 | const struct nlattr *mask, size_t mask_len, |
70e5ed6f | 1630 | const ovs_u128 *ufid, const struct dpif_flow_stats *stats, |
cf22f8cb | 1631 | const struct nlattr *actions, size_t actions_len) |
96fba48f BP |
1632 | { |
1633 | struct ds ds = DS_EMPTY_INITIALIZER; | |
1634 | ds_put_format(&ds, "%s: ", dpif_name(dpif)); | |
1635 | if (error) { | |
1636 | ds_put_cstr(&ds, "failed to "); | |
1637 | } | |
1638 | ds_put_format(&ds, "%s ", operation); | |
1639 | if (error) { | |
10a89ef0 | 1640 | ds_put_format(&ds, "(%s) ", ovs_strerror(error)); |
96fba48f | 1641 | } |
70e5ed6f JS |
1642 | if (ufid) { |
1643 | odp_format_ufid(ufid, &ds); | |
1644 | ds_put_cstr(&ds, " "); | |
1645 | } | |
0a37839c | 1646 | odp_flow_format(key, key_len, mask, mask_len, NULL, &ds, true); |
96fba48f BP |
1647 | if (stats) { |
1648 | ds_put_cstr(&ds, ", "); | |
c97fb132 | 1649 | dpif_flow_stats_format(stats, &ds); |
96fba48f | 1650 | } |
cdee00fd | 1651 | if (actions || actions_len) { |
96fba48f | 1652 | ds_put_cstr(&ds, ", actions:"); |
cdee00fd | 1653 | format_odp_actions(&ds, actions, actions_len); |
96fba48f | 1654 | } |
922fed06 | 1655 | vlog(&this_module, flow_message_log_level(error), "%s", ds_cstr(&ds)); |
96fba48f BP |
1656 | ds_destroy(&ds); |
1657 | } | |
89625d1e BP |
1658 | |
1659 | static void | |
1660 | log_flow_put_message(struct dpif *dpif, const struct dpif_flow_put *put, | |
1661 | int error) | |
1662 | { | |
43f9ac0a | 1663 | if (should_log_flow_message(error) && !(put->flags & DPIF_FP_PROBE)) { |
89625d1e BP |
1664 | struct ds s; |
1665 | ||
1666 | ds_init(&s); | |
1667 | ds_put_cstr(&s, "put"); | |
1668 | if (put->flags & DPIF_FP_CREATE) { | |
1669 | ds_put_cstr(&s, "[create]"); | |
1670 | } | |
1671 | if (put->flags & DPIF_FP_MODIFY) { | |
1672 | ds_put_cstr(&s, "[modify]"); | |
1673 | } | |
1674 | if (put->flags & DPIF_FP_ZERO_STATS) { | |
1675 | ds_put_cstr(&s, "[zero]"); | |
1676 | } | |
1677 | log_flow_message(dpif, error, ds_cstr(&s), | |
61fb711d | 1678 | put->key, put->key_len, put->mask, put->mask_len, |
70e5ed6f JS |
1679 | put->ufid, put->stats, put->actions, |
1680 | put->actions_len); | |
89625d1e BP |
1681 | ds_destroy(&s); |
1682 | } | |
1683 | } | |
1684 | ||
b99d3cee BP |
1685 | static void |
1686 | log_flow_del_message(struct dpif *dpif, const struct dpif_flow_del *del, | |
1687 | int error) | |
1688 | { | |
1689 | if (should_log_flow_message(error)) { | |
1690 | log_flow_message(dpif, error, "flow_del", del->key, del->key_len, | |
70e5ed6f JS |
1691 | NULL, 0, del->ufid, !error ? del->stats : NULL, |
1692 | NULL, 0); | |
b99d3cee BP |
1693 | } |
1694 | } | |
1695 | ||
f0fe12fc BP |
1696 | /* Logs that 'execute' was executed on 'dpif' and completed with errno 'error' |
1697 | * (0 for success). 'subexecute' should be true if the execution is a result | |
1698 | * of breaking down a larger execution that needed help, false otherwise. | |
1699 | * | |
1700 | * | |
1701 | * XXX In theory, the log message could be deceptive because this function is | |
1702 | * called after the dpif_provider's '->execute' function, which is allowed to | |
1703 | * modify execute->packet and execute->md. In practice, though: | |
1704 | * | |
93451a0a | 1705 | * - dpif-netlink doesn't modify execute->packet or execute->md. |
f0fe12fc BP |
1706 | * |
1707 | * - dpif-netdev does modify them but it is less likely to have problems | |
1708 | * because it is built into ovs-vswitchd and cannot have version skew, | |
1709 | * etc. | |
1710 | * | |
1711 | * It would still be better to avoid the potential problem. I don't know of a | |
1712 | * good way to do that, though, that isn't expensive. */ | |
89625d1e BP |
1713 | static void |
1714 | log_execute_message(struct dpif *dpif, const struct dpif_execute *execute, | |
fc65bafc | 1715 | bool subexecute, int error) |
89625d1e | 1716 | { |
43f9ac0a JR |
1717 | if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl)) |
1718 | && !execute->probe) { | |
89625d1e BP |
1719 | struct ds ds = DS_EMPTY_INITIALIZER; |
1720 | char *packet; | |
1721 | ||
cf62fa4c PS |
1722 | packet = ofp_packet_to_string(dp_packet_data(execute->packet), |
1723 | dp_packet_size(execute->packet)); | |
fc65bafc BP |
1724 | ds_put_format(&ds, "%s: %sexecute ", |
1725 | dpif_name(dpif), | |
1726 | (subexecute ? "sub-" | |
1727 | : dpif_execute_needs_help(execute) ? "super-" | |
1728 | : "")); | |
89625d1e BP |
1729 | format_odp_actions(&ds, execute->actions, execute->actions_len); |
1730 | if (error) { | |
10a89ef0 | 1731 | ds_put_format(&ds, " failed (%s)", ovs_strerror(error)); |
89625d1e BP |
1732 | } |
1733 | ds_put_format(&ds, " on packet %s", packet); | |
27130224 | 1734 | ds_put_format(&ds, " mtu %d", execute->mtu); |
922fed06 | 1735 | vlog(&this_module, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds)); |
89625d1e BP |
1736 | ds_destroy(&ds); |
1737 | free(packet); | |
1738 | } | |
1739 | } | |
6fe09f8c JS |
1740 | |
1741 | static void | |
1742 | log_flow_get_message(const struct dpif *dpif, const struct dpif_flow_get *get, | |
1743 | int error) | |
1744 | { | |
1745 | if (should_log_flow_message(error)) { | |
1746 | log_flow_message(dpif, error, "flow_get", | |
1747 | get->key, get->key_len, | |
1748 | get->flow->mask, get->flow->mask_len, | |
70e5ed6f | 1749 | get->ufid, &get->flow->stats, |
6fe09f8c JS |
1750 | get->flow->actions, get->flow->actions_len); |
1751 | } | |
1752 | } | |
a36de779 PS |
1753 | |
1754 | bool | |
1755 | dpif_supports_tnl_push_pop(const struct dpif *dpif) | |
1756 | { | |
c4ea7529 | 1757 | return dpif_is_netdev(dpif); |
a36de779 | 1758 | } |