]> git.proxmox.com Git - ovs.git/blame - lib/dpif.c
tests: Use set_field instead of load_reg in OF1.2 test for metadata.
[ovs.git] / lib / dpif.c
CommitLineData
064af421 1/*
2f51a7eb 2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 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"
27#include "dynamic-string.h"
28#include "flow.h"
c3827f61 29#include "netdev.h"
064af421 30#include "netlink.h"
7fd91025 31#include "odp-execute.h"
064af421 32#include "odp-util.h"
90bf1e07 33#include "ofp-errors.h"
064af421 34#include "ofp-print.h"
71ce9235 35#include "ofp-util.h"
064af421
BP
36#include "ofpbuf.h"
37#include "packets.h"
38#include "poll-loop.h"
999401aa 39#include "shash.h"
d0c23a1a 40#include "sset.h"
c97fb132 41#include "timeval.h"
064af421
BP
42#include "util.h"
43#include "valgrind.h"
064af421 44#include "vlog.h"
5136ce49 45
d98e6007 46VLOG_DEFINE_THIS_MODULE(dpif);
064af421 47
d76f09ea
BP
48COVERAGE_DEFINE(dpif_destroy);
49COVERAGE_DEFINE(dpif_port_add);
50COVERAGE_DEFINE(dpif_port_del);
51COVERAGE_DEFINE(dpif_flow_flush);
52COVERAGE_DEFINE(dpif_flow_get);
53COVERAGE_DEFINE(dpif_flow_put);
54COVERAGE_DEFINE(dpif_flow_del);
d76f09ea
BP
55COVERAGE_DEFINE(dpif_execute);
56COVERAGE_DEFINE(dpif_purge);
7fd91025 57COVERAGE_DEFINE(dpif_execute_with_help);
d76f09ea 58
999401aa 59static const struct dpif_class *base_dpif_classes[] = {
2f51a7eb 60#ifdef __linux__
96fba48f 61 &dpif_linux_class,
c83cdd30 62#endif
72865317 63 &dpif_netdev_class,
c228a364 64};
999401aa
JG
65
66struct registered_dpif_class {
d2d8fbeb 67 const struct dpif_class *dpif_class;
999401aa
JG
68 int refcount;
69};
70static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes);
579a77e0 71static struct sset dpif_blacklist = SSET_INITIALIZER(&dpif_blacklist);
c228a364 72
5703b15f 73/* Protects 'dpif_classes', including the refcount, and 'dpif_blacklist'. */
97be1538 74static struct ovs_mutex dpif_mutex = OVS_MUTEX_INITIALIZER;
5703b15f 75
064af421
BP
76/* Rate limit for individual messages going to or from the datapath, output at
77 * DBG level. This is very high because, if these are enabled, it is because
78 * we really need to see them. */
79static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
80
81/* Not really much point in logging many dpif errors. */
e2781405 82static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
064af421 83
feebdea2
BP
84static void log_flow_message(const struct dpif *dpif, int error,
85 const char *operation,
86 const struct nlattr *key, size_t key_len,
61fb711d 87 const struct nlattr *mask, size_t mask_len,
c97fb132 88 const struct dpif_flow_stats *stats,
feebdea2 89 const struct nlattr *actions, size_t actions_len);
96fba48f
BP
90static void log_operation(const struct dpif *, const char *operation,
91 int error);
96fba48f 92static bool should_log_flow_message(int error);
89625d1e
BP
93static void log_flow_put_message(struct dpif *, const struct dpif_flow_put *,
94 int error);
b99d3cee
BP
95static void log_flow_del_message(struct dpif *, const struct dpif_flow_del *,
96 int error);
89625d1e
BP
97static void log_execute_message(struct dpif *, const struct dpif_execute *,
98 int error);
064af421 99
999401aa
JG
100static void
101dp_initialize(void)
102{
eb8ed438 103 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
999401aa 104
eb8ed438 105 if (ovsthread_once_start(&once)) {
999401aa
JG
106 int i;
107
999401aa
JG
108 for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
109 dp_register_provider(base_dpif_classes[i]);
110 }
eb8ed438 111 ovsthread_once_done(&once);
999401aa
JG
112 }
113}
114
5703b15f
BP
115static int
116dp_register_provider__(const struct dpif_class *new_class)
999401aa
JG
117{
118 struct registered_dpif_class *registered_class;
119
579a77e0
EJ
120 if (sset_contains(&dpif_blacklist, new_class->type)) {
121 VLOG_DBG("attempted to register blacklisted provider: %s",
122 new_class->type);
123 return EINVAL;
124 }
125
999401aa
JG
126 if (shash_find(&dpif_classes, new_class->type)) {
127 VLOG_WARN("attempted to register duplicate datapath provider: %s",
128 new_class->type);
129 return EEXIST;
130 }
1a6f1e2a 131
999401aa 132 registered_class = xmalloc(sizeof *registered_class);
d2d8fbeb 133 registered_class->dpif_class = new_class;
999401aa
JG
134 registered_class->refcount = 0;
135
136 shash_add(&dpif_classes, new_class->type, registered_class);
137
138 return 0;
139}
140
5703b15f
BP
141/* Registers a new datapath provider. After successful registration, new
142 * datapaths of that type can be opened using dpif_open(). */
143int
144dp_register_provider(const struct dpif_class *new_class)
145{
146 int error;
147
97be1538 148 ovs_mutex_lock(&dpif_mutex);
5703b15f 149 error = dp_register_provider__(new_class);
97be1538 150 ovs_mutex_unlock(&dpif_mutex);
5703b15f
BP
151
152 return error;
153}
154
999401aa
JG
155/* Unregisters a datapath provider. 'type' must have been previously
156 * registered and not currently be in use by any dpifs. After unregistration
157 * new datapaths of that type cannot be opened using dpif_open(). */
5703b15f
BP
158static int
159dp_unregister_provider__(const char *type)
999401aa
JG
160{
161 struct shash_node *node;
162 struct registered_dpif_class *registered_class;
163
164 node = shash_find(&dpif_classes, type);
165 if (!node) {
166 VLOG_WARN("attempted to unregister a datapath provider that is not "
167 "registered: %s", type);
168 return EAFNOSUPPORT;
169 }
170
171 registered_class = node->data;
172 if (registered_class->refcount) {
173 VLOG_WARN("attempted to unregister in use datapath provider: %s", type);
174 return EBUSY;
175 }
176
177 shash_delete(&dpif_classes, node);
178 free(registered_class);
179
180 return 0;
181}
182
5703b15f
BP
183/* Unregisters a datapath provider. 'type' must have been previously
184 * registered and not currently be in use by any dpifs. After unregistration
185 * new datapaths of that type cannot be opened using dpif_open(). */
186int
187dp_unregister_provider(const char *type)
188{
189 int error;
190
191 dp_initialize();
192
97be1538 193 ovs_mutex_lock(&dpif_mutex);
5703b15f 194 error = dp_unregister_provider__(type);
97be1538 195 ovs_mutex_unlock(&dpif_mutex);
5703b15f
BP
196
197 return error;
198}
199
579a77e0
EJ
200/* Blacklists a provider. Causes future calls of dp_register_provider() with
201 * a dpif_class which implements 'type' to fail. */
202void
203dp_blacklist_provider(const char *type)
204{
97be1538 205 ovs_mutex_lock(&dpif_mutex);
579a77e0 206 sset_add(&dpif_blacklist, type);
97be1538 207 ovs_mutex_unlock(&dpif_mutex);
579a77e0
EJ
208}
209
999401aa 210/* Clears 'types' and enumerates the types of all currently registered datapath
d0c23a1a 211 * providers into it. The caller must first initialize the sset. */
1a6f1e2a 212void
d0c23a1a 213dp_enumerate_types(struct sset *types)
1a6f1e2a 214{
999401aa 215 struct shash_node *node;
1a6f1e2a 216
999401aa 217 dp_initialize();
d0c23a1a 218 sset_clear(types);
1a6f1e2a 219
97be1538 220 ovs_mutex_lock(&dpif_mutex);
999401aa
JG
221 SHASH_FOR_EACH(node, &dpif_classes) {
222 const struct registered_dpif_class *registered_class = node->data;
d0c23a1a 223 sset_add(types, registered_class->dpif_class->type);
1a6f1e2a 224 }
97be1538 225 ovs_mutex_unlock(&dpif_mutex);
5703b15f
BP
226}
227
228static void
229dp_class_unref(struct registered_dpif_class *rc)
230{
97be1538 231 ovs_mutex_lock(&dpif_mutex);
5703b15f
BP
232 ovs_assert(rc->refcount);
233 rc->refcount--;
97be1538 234 ovs_mutex_unlock(&dpif_mutex);
5703b15f
BP
235}
236
237static struct registered_dpif_class *
238dp_class_lookup(const char *type)
239{
240 struct registered_dpif_class *rc;
241
97be1538 242 ovs_mutex_lock(&dpif_mutex);
5703b15f
BP
243 rc = shash_find_data(&dpif_classes, type);
244 if (rc) {
245 rc->refcount++;
246 }
97be1538 247 ovs_mutex_unlock(&dpif_mutex);
5703b15f
BP
248
249 return rc;
1a6f1e2a
JG
250}
251
252/* Clears 'names' and enumerates the names of all known created datapaths with
d0c23a1a 253 * the given 'type'. The caller must first initialize the sset. Returns 0 if
1a6f1e2a 254 * successful, otherwise a positive errno value.
d3d22744
BP
255 *
256 * Some kinds of datapaths might not be practically enumerable. This is not
257 * considered an error. */
258int
d0c23a1a 259dp_enumerate_names(const char *type, struct sset *names)
d3d22744 260{
5703b15f 261 struct registered_dpif_class *registered_class;
999401aa
JG
262 const struct dpif_class *dpif_class;
263 int error;
d3d22744 264
999401aa 265 dp_initialize();
d0c23a1a 266 sset_clear(names);
1a6f1e2a 267
5703b15f 268 registered_class = dp_class_lookup(type);
999401aa
JG
269 if (!registered_class) {
270 VLOG_WARN("could not enumerate unknown type: %s", type);
271 return EAFNOSUPPORT;
272 }
1a6f1e2a 273
d2d8fbeb 274 dpif_class = registered_class->dpif_class;
999401aa 275 error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
999401aa
JG
276 if (error) {
277 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
10a89ef0 278 ovs_strerror(error));
d3d22744 279 }
5703b15f 280 dp_class_unref(registered_class);
1a6f1e2a 281
999401aa 282 return error;
1a6f1e2a
JG
283}
284
54ed8a5d
BP
285/* Parses 'datapath_name_', which is of the form [type@]name into its
286 * component pieces. 'name' and 'type' must be freed by the caller.
287 *
288 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
1a6f1e2a
JG
289void
290dp_parse_name(const char *datapath_name_, char **name, char **type)
291{
292 char *datapath_name = xstrdup(datapath_name_);
293 char *separator;
294
295 separator = strchr(datapath_name, '@');
296 if (separator) {
297 *separator = '\0';
298 *type = datapath_name;
54ed8a5d 299 *name = xstrdup(dpif_normalize_type(separator + 1));
1a6f1e2a
JG
300 } else {
301 *name = datapath_name;
54ed8a5d 302 *type = xstrdup(dpif_normalize_type(NULL));
1a6f1e2a 303 }
d3d22744
BP
304}
305
96fba48f 306static int
1a6f1e2a 307do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
064af421 308{
96fba48f 309 struct dpif *dpif = NULL;
064af421 310 int error;
999401aa
JG
311 struct registered_dpif_class *registered_class;
312
313 dp_initialize();
064af421 314
3a225db7 315 type = dpif_normalize_type(type);
5703b15f 316 registered_class = dp_class_lookup(type);
999401aa
JG
317 if (!registered_class) {
318 VLOG_WARN("could not create datapath %s of unknown type %s", name,
319 type);
320 error = EAFNOSUPPORT;
321 goto exit;
322 }
323
4a387741
BP
324 error = registered_class->dpif_class->open(registered_class->dpif_class,
325 name, create, &dpif);
999401aa 326 if (!error) {
cb22974d 327 ovs_assert(dpif->dpif_class == registered_class->dpif_class);
5703b15f
BP
328 } else {
329 dp_class_unref(registered_class);
064af421 330 }
064af421 331
96fba48f
BP
332exit:
333 *dpifp = error ? NULL : dpif;
334 return error;
064af421
BP
335}
336
1a6f1e2a
JG
337/* Tries to open an existing datapath named 'name' and type 'type'. Will fail
338 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
339 * the empty string to specify the default system type. Returns 0 if
340 * successful, otherwise a positive errno value. On success stores a pointer
341 * to the datapath in '*dpifp', otherwise a null pointer. */
96fba48f 342int
1a6f1e2a 343dpif_open(const char *name, const char *type, struct dpif **dpifp)
064af421 344{
1a6f1e2a 345 return do_open(name, type, false, dpifp);
064af421
BP
346}
347
1a6f1e2a
JG
348/* Tries to create and open a new datapath with the given 'name' and 'type'.
349 * 'type' may be either NULL or the empty string to specify the default system
350 * type. Will fail if a datapath with 'name' and 'type' already exists.
351 * Returns 0 if successful, otherwise a positive errno value. On success
352 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
064af421 353int
1a6f1e2a 354dpif_create(const char *name, const char *type, struct dpif **dpifp)
064af421 355{
1a6f1e2a 356 return do_open(name, type, true, dpifp);
96fba48f 357}
064af421 358
1a6f1e2a
JG
359/* Tries to open a datapath with the given 'name' and 'type', creating it if it
360 * does not exist. 'type' may be either NULL or the empty string to specify
361 * the default system type. Returns 0 if successful, otherwise a positive
362 * errno value. On success stores a pointer to the datapath in '*dpifp',
363 * otherwise a null pointer. */
efacbce6 364int
1a6f1e2a 365dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
efacbce6
BP
366{
367 int error;
368
1a6f1e2a 369 error = dpif_create(name, type, dpifp);
efacbce6 370 if (error == EEXIST || error == EBUSY) {
1a6f1e2a 371 error = dpif_open(name, type, dpifp);
efacbce6
BP
372 if (error) {
373 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
10a89ef0 374 name, ovs_strerror(error));
efacbce6
BP
375 }
376 } else if (error) {
10a89ef0
BP
377 VLOG_WARN("failed to create datapath %s: %s",
378 name, ovs_strerror(error));
efacbce6
BP
379 }
380 return error;
381}
382
96fba48f
BP
383/* Closes and frees the connection to 'dpif'. Does not destroy the datapath
384 * itself; call dpif_delete() first, instead, if that is desirable. */
385void
386dpif_close(struct dpif *dpif)
387{
388 if (dpif) {
5703b15f 389 struct registered_dpif_class *rc;
999401aa 390
5703b15f 391 rc = shash_find_data(&dpif_classes, dpif->dpif_class->type);
999401aa 392 dpif_uninit(dpif, true);
5703b15f 393 dp_class_unref(rc);
064af421
BP
394 }
395}
396
640e1b20
BP
397/* Performs periodic work needed by 'dpif'. */
398void
399dpif_run(struct dpif *dpif)
400{
401 if (dpif->dpif_class->run) {
402 dpif->dpif_class->run(dpif);
403 }
404}
405
406/* Arranges for poll_block() to wake up when dp_run() needs to be called for
407 * 'dpif'. */
408void
409dpif_wait(struct dpif *dpif)
410{
411 if (dpif->dpif_class->wait) {
412 dpif->dpif_class->wait(dpif);
413 }
414}
415
1a6f1e2a
JG
416/* Returns the name of datapath 'dpif' prefixed with the type
417 * (for use in log messages). */
b29ba128
BP
418const char *
419dpif_name(const struct dpif *dpif)
420{
1a6f1e2a
JG
421 return dpif->full_name;
422}
423
424/* Returns the name of datapath 'dpif' without the type
425 * (for use in device names). */
426const char *
427dpif_base_name(const struct dpif *dpif)
428{
429 return dpif->base_name;
b29ba128
BP
430}
431
c7a26215
JP
432/* Returns the type of datapath 'dpif'. */
433const char *
434dpif_type(const struct dpif *dpif)
435{
436 return dpif->dpif_class->type;
437}
438
3a225db7
BP
439/* Returns the fully spelled out name for the given datapath 'type'.
440 *
441 * Normalized type string can be compared with strcmp(). Unnormalized type
442 * string might be the same even if they have different spellings. */
443const char *
444dpif_normalize_type(const char *type)
445{
446 return type && type[0] ? type : "system";
447}
448
96fba48f
BP
449/* Destroys the datapath that 'dpif' is connected to, first removing all of its
450 * ports. After calling this function, it does not make sense to pass 'dpif'
451 * to any functions other than dpif_name() or dpif_close(). */
064af421
BP
452int
453dpif_delete(struct dpif *dpif)
454{
96fba48f
BP
455 int error;
456
064af421 457 COVERAGE_INC(dpif_destroy);
96fba48f 458
1acb6baa 459 error = dpif->dpif_class->destroy(dpif);
96fba48f
BP
460 log_operation(dpif, "delete", error);
461 return error;
064af421
BP
462}
463
96fba48f
BP
464/* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
465 * otherwise a positive errno value. */
064af421 466int
a8d9304d 467dpif_get_dp_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
064af421 468{
1acb6baa 469 int error = dpif->dpif_class->get_stats(dpif, stats);
96fba48f
BP
470 if (error) {
471 memset(stats, 0, sizeof *stats);
472 }
473 log_operation(dpif, "get_stats", error);
474 return error;
064af421
BP
475}
476
0aeaabc8
JP
477const char *
478dpif_port_open_type(const char *datapath_type, const char *port_type)
479{
5703b15f 480 struct registered_dpif_class *rc;
0aeaabc8
JP
481
482 datapath_type = dpif_normalize_type(datapath_type);
483
97be1538 484 ovs_mutex_lock(&dpif_mutex);
5703b15f
BP
485 rc = shash_find_data(&dpif_classes, datapath_type);
486 if (rc && rc->dpif_class->port_open_type) {
487 port_type = rc->dpif_class->port_open_type(rc->dpif_class, port_type);
0aeaabc8 488 }
97be1538 489 ovs_mutex_unlock(&dpif_mutex);
0aeaabc8 490
5703b15f 491 return port_type;
0aeaabc8
JP
492}
493
232dfa4a 494/* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
4e022ec0 495 * non-null and its value is not ODPP_NONE, then attempts to use the
232dfa4a
JP
496 * value as the port number.
497 *
498 * If successful, returns 0 and sets '*port_nop' to the new port's port
499 * number (if 'port_nop' is non-null). On failure, returns a positive
4e022ec0 500 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
232dfa4a 501 * non-null). */
064af421 502int
4e022ec0 503dpif_port_add(struct dpif *dpif, struct netdev *netdev, odp_port_t *port_nop)
064af421 504{
c3827f61 505 const char *netdev_name = netdev_get_name(netdev);
4e022ec0 506 odp_port_t port_no = ODPP_NONE;
9ee3ae3e 507 int error;
064af421
BP
508
509 COVERAGE_INC(dpif_port_add);
9ee3ae3e 510
232dfa4a
JP
511 if (port_nop) {
512 port_no = *port_nop;
513 }
514
c3827f61 515 error = dpif->dpif_class->port_add(dpif, netdev, &port_no);
9ee3ae3e 516 if (!error) {
9b56fe13 517 VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu32,
c3827f61 518 dpif_name(dpif), netdev_name, port_no);
064af421 519 } else {
9ee3ae3e 520 VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
10a89ef0 521 dpif_name(dpif), netdev_name, ovs_strerror(error));
4e022ec0 522 port_no = ODPP_NONE;
9ee3ae3e
BP
523 }
524 if (port_nop) {
525 *port_nop = port_no;
064af421 526 }
9ee3ae3e 527 return error;
064af421
BP
528}
529
96fba48f
BP
530/* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
531 * otherwise a positive errno value. */
064af421 532int
4e022ec0 533dpif_port_del(struct dpif *dpif, odp_port_t port_no)
064af421 534{
96fba48f
BP
535 int error;
536
064af421 537 COVERAGE_INC(dpif_port_del);
96fba48f 538
1acb6baa 539 error = dpif->dpif_class->port_del(dpif, port_no);
a1811296 540 if (!error) {
9b56fe13 541 VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu32")",
a1811296
BP
542 dpif_name(dpif), port_no);
543 } else {
544 log_operation(dpif, "port_del", error);
545 }
96fba48f 546 return error;
064af421
BP
547}
548
4c738a8d
BP
549/* Makes a deep copy of 'src' into 'dst'. */
550void
551dpif_port_clone(struct dpif_port *dst, const struct dpif_port *src)
552{
553 dst->name = xstrdup(src->name);
554 dst->type = xstrdup(src->type);
555 dst->port_no = src->port_no;
556}
557
558/* Frees memory allocated to members of 'dpif_port'.
559 *
560 * Do not call this function on a dpif_port obtained from
561 * dpif_port_dump_next(): that function retains ownership of the data in the
562 * dpif_port. */
563void
564dpif_port_destroy(struct dpif_port *dpif_port)
565{
566 free(dpif_port->name);
567 free(dpif_port->type);
568}
569
4afba28d
JP
570/* Checks if port named 'devname' exists in 'dpif'. If so, returns
571 * true; otherwise, returns false. */
572bool
573dpif_port_exists(const struct dpif *dpif, const char *devname)
574{
575 int error = dpif->dpif_class->port_query_by_name(dpif, devname, NULL);
bee6b8bc 576 if (error != 0 && error != ENOENT && error != ENODEV) {
4afba28d 577 VLOG_WARN_RL(&error_rl, "%s: failed to query port %s: %s",
10a89ef0 578 dpif_name(dpif), devname, ovs_strerror(error));
4afba28d
JP
579 }
580
581 return !error;
582}
583
96fba48f
BP
584/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
585 * initializes '*port' appropriately; on failure, returns a positive errno
4c738a8d
BP
586 * value.
587 *
588 * The caller owns the data in 'port' and must free it with
589 * dpif_port_destroy() when it is no longer needed. */
064af421 590int
4e022ec0 591dpif_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
4c738a8d 592 struct dpif_port *port)
064af421 593{
1acb6baa 594 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
96fba48f 595 if (!error) {
9b56fe13 596 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu32" is device %s",
4c738a8d 597 dpif_name(dpif), port_no, port->name);
064af421 598 } else {
96fba48f 599 memset(port, 0, sizeof *port);
9b56fe13 600 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu32": %s",
10a89ef0 601 dpif_name(dpif), port_no, ovs_strerror(error));
064af421 602 }
96fba48f 603 return error;
064af421
BP
604}
605
96fba48f
BP
606/* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
607 * initializes '*port' appropriately; on failure, returns a positive errno
4c738a8d
BP
608 * value.
609 *
610 * The caller owns the data in 'port' and must free it with
611 * dpif_port_destroy() when it is no longer needed. */
064af421
BP
612int
613dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
4c738a8d 614 struct dpif_port *port)
064af421 615{
1acb6baa 616 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
96fba48f 617 if (!error) {
9b56fe13 618 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu32,
4c738a8d 619 dpif_name(dpif), devname, port->port_no);
064af421 620 } else {
96fba48f
BP
621 memset(port, 0, sizeof *port);
622
d647f0a7
BP
623 /* For ENOENT or ENODEV we use DBG level because the caller is probably
624 * interested in whether 'dpif' actually has a port 'devname', so that
625 * it's not an issue worth logging if it doesn't. Other errors are
626 * uncommon and more likely to indicate a real problem. */
627 VLOG_RL(&error_rl,
628 error == ENOENT || error == ENODEV ? VLL_DBG : VLL_WARN,
629 "%s: failed to query port %s: %s",
10a89ef0 630 dpif_name(dpif), devname, ovs_strerror(error));
064af421 631 }
96fba48f 632 return error;
064af421
BP
633}
634
1954e6bb
AW
635/* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
636 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
637 * flows whose packets arrived on port 'port_no'. In the case where the
638 * provider allocates multiple Netlink PIDs to a single port, it may use
639 * 'hash' to spread load among them. The caller need not use a particular
640 * hash function; a 5-tuple hash is suitable.
641 *
642 * (The datapath implementation might use some different hash function for
643 * distributing packets received via flow misses among PIDs. This means
644 * that packets received via flow misses might be reordered relative to
645 * packets received via userspace actions. This is not ordinarily a
646 * problem.)
98403001 647 *
4e022ec0 648 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
625b0720
BP
649 * allocated to any port, that the client may use for special purposes.
650 *
98403001
BP
651 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
652 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
653 * disabled and then re-enabled, so a client that does that must be prepared to
654 * update all of the flows that it installed that contain
655 * OVS_ACTION_ATTR_USERSPACE actions. */
656uint32_t
1954e6bb 657dpif_port_get_pid(const struct dpif *dpif, odp_port_t port_no, uint32_t hash)
98403001
BP
658{
659 return (dpif->dpif_class->port_get_pid
1954e6bb 660 ? (dpif->dpif_class->port_get_pid)(dpif, port_no, hash)
98403001
BP
661 : 0);
662}
663
96fba48f
BP
664/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
665 * the port's name into the 'name_size' bytes in 'name', ensuring that the
666 * result is null-terminated. On failure, returns a positive errno value and
667 * makes 'name' the empty string. */
335562c0 668int
4e022ec0 669dpif_port_get_name(struct dpif *dpif, odp_port_t port_no,
335562c0
BP
670 char *name, size_t name_size)
671{
4c738a8d 672 struct dpif_port port;
335562c0
BP
673 int error;
674
cb22974d 675 ovs_assert(name_size > 0);
335562c0
BP
676
677 error = dpif_port_query_by_number(dpif, port_no, &port);
678 if (!error) {
4c738a8d
BP
679 ovs_strlcpy(name, port.name, name_size);
680 dpif_port_destroy(&port);
335562c0
BP
681 } else {
682 *name = '\0';
683 }
684 return error;
685}
686
b0ec0f27 687/* Initializes 'dump' to begin dumping the ports in a dpif.
96fba48f 688 *
b0ec0f27
BP
689 * This function provides no status indication. An error status for the entire
690 * dump operation is provided when it is completed by calling
691 * dpif_port_dump_done().
692 */
693void
694dpif_port_dump_start(struct dpif_port_dump *dump, const struct dpif *dpif)
695{
696 dump->dpif = dpif;
697 dump->error = dpif->dpif_class->port_dump_start(dpif, &dump->state);
698 log_operation(dpif, "port_dump_start", dump->error);
699}
700
701/* Attempts to retrieve another port from 'dump', which must have been
4c738a8d 702 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
b0ec0f27 703 * into 'port' and returns true. On failure, returns false.
96fba48f 704 *
b0ec0f27
BP
705 * Failure might indicate an actual error or merely that the last port has been
706 * dumped. An error status for the entire dump operation is provided when it
4c738a8d
BP
707 * is completed by calling dpif_port_dump_done().
708 *
709 * The dpif owns the data stored in 'port'. It will remain valid until at
710 * least the next time 'dump' is passed to dpif_port_dump_next() or
711 * dpif_port_dump_done(). */
b0ec0f27 712bool
4c738a8d 713dpif_port_dump_next(struct dpif_port_dump *dump, struct dpif_port *port)
064af421 714{
b0ec0f27 715 const struct dpif *dpif = dump->dpif;
064af421 716
b0ec0f27
BP
717 if (dump->error) {
718 return false;
719 }
f4ba4c4f 720
b0ec0f27
BP
721 dump->error = dpif->dpif_class->port_dump_next(dpif, dump->state, port);
722 if (dump->error == EOF) {
723 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all ports", dpif_name(dpif));
724 } else {
725 log_operation(dpif, "port_dump_next", dump->error);
726 }
064af421 727
b0ec0f27
BP
728 if (dump->error) {
729 dpif->dpif_class->port_dump_done(dpif, dump->state);
730 return false;
f4ba4c4f 731 }
b0ec0f27
BP
732 return true;
733}
064af421 734
b0ec0f27
BP
735/* Completes port table dump operation 'dump', which must have been initialized
736 * with dpif_port_dump_start(). Returns 0 if the dump operation was
737 * error-free, otherwise a positive errno value describing the problem. */
738int
739dpif_port_dump_done(struct dpif_port_dump *dump)
740{
741 const struct dpif *dpif = dump->dpif;
742 if (!dump->error) {
743 dump->error = dpif->dpif_class->port_dump_done(dpif, dump->state);
744 log_operation(dpif, "port_dump_done", dump->error);
f4ba4c4f 745 }
b0ec0f27 746 return dump->error == EOF ? 0 : dump->error;
064af421
BP
747}
748
e9e28be3
BP
749/* Polls for changes in the set of ports in 'dpif'. If the set of ports in
750 * 'dpif' has changed, this function does one of the following:
751 *
752 * - Stores the name of the device that was added to or deleted from 'dpif' in
753 * '*devnamep' and returns 0. The caller is responsible for freeing
754 * '*devnamep' (with free()) when it no longer needs it.
755 *
756 * - Returns ENOBUFS and sets '*devnamep' to NULL.
757 *
758 * This function may also return 'false positives', where it returns 0 and
759 * '*devnamep' names a device that was not actually added or deleted or it
760 * returns ENOBUFS without any change.
761 *
762 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
763 * return other positive errno values to indicate that something has gone
764 * wrong. */
765int
766dpif_port_poll(const struct dpif *dpif, char **devnamep)
767{
1acb6baa 768 int error = dpif->dpif_class->port_poll(dpif, devnamep);
e9e28be3
BP
769 if (error) {
770 *devnamep = NULL;
771 }
772 return error;
773}
774
775/* Arranges for the poll loop to wake up when port_poll(dpif) will return a
776 * value other than EAGAIN. */
777void
778dpif_port_poll_wait(const struct dpif *dpif)
779{
1acb6baa 780 dpif->dpif_class->port_poll_wait(dpif);
e9e28be3
BP
781}
782
572b7068 783/* Extracts the flow stats for a packet. The 'flow' and 'packet'
a7752d4a
BP
784 * arguments must have been initialized through a call to flow_extract().
785 * 'used' is stored into stats->used. */
572b7068 786void
a39edbd4 787dpif_flow_stats_extract(const struct flow *flow, const struct ofpbuf *packet,
a7752d4a 788 long long int used, struct dpif_flow_stats *stats)
572b7068 789{
e0eecb1c 790 stats->tcp_flags = ntohs(flow->tcp_flags);
1f317cb5 791 stats->n_bytes = ofpbuf_size(packet);
572b7068 792 stats->n_packets = 1;
a7752d4a 793 stats->used = used;
572b7068
BP
794}
795
c97fb132
BP
796/* Appends a human-readable representation of 'stats' to 's'. */
797void
798dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s)
799{
800 ds_put_format(s, "packets:%"PRIu64", bytes:%"PRIu64", used:",
801 stats->n_packets, stats->n_bytes);
802 if (stats->used) {
803 ds_put_format(s, "%.3fs", (time_msec() - stats->used) / 1000.0);
804 } else {
805 ds_put_format(s, "never");
806 }
7393104d
BP
807 if (stats->tcp_flags) {
808 ds_put_cstr(s, ", flags:");
809 packet_format_tcp_flags(s, stats->tcp_flags);
810 }
c97fb132
BP
811}
812
96fba48f
BP
813/* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
814 * positive errno value. */
815int
816dpif_flow_flush(struct dpif *dpif)
064af421 817{
96fba48f
BP
818 int error;
819
820 COVERAGE_INC(dpif_flow_flush);
821
1acb6baa 822 error = dpif->dpif_class->flow_flush(dpif);
96fba48f
BP
823 log_operation(dpif, "flow_flush", error);
824 return error;
064af421
BP
825}
826
feebdea2 827/* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
df2c07f4 828 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
feebdea2 829 * 'key'.
96fba48f 830 *
feebdea2
BP
831 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
832 * failure, returns a positive errno value.
96fba48f 833 *
feebdea2
BP
834 * If 'actionsp' is nonnull, then on success '*actionsp' will be set to an
835 * ofpbuf owned by the caller that contains the Netlink attributes for the
836 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete()) when
837 * it is no longer needed.
838 *
839 * If 'stats' is nonnull, then on success it will be updated with the flow's
840 * statistics. */
96fba48f 841int
693c4a01 842dpif_flow_get(const struct dpif *dpif,
feebdea2 843 const struct nlattr *key, size_t key_len,
c97fb132 844 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
064af421 845{
96fba48f
BP
846 int error;
847
848 COVERAGE_INC(dpif_flow_get);
849
693c4a01 850 error = dpif->dpif_class->flow_get(dpif, key, key_len, actionsp, stats);
b843fa1b 851 if (error) {
feebdea2
BP
852 if (actionsp) {
853 *actionsp = NULL;
854 }
855 if (stats) {
856 memset(stats, 0, sizeof *stats);
857 }
b843fa1b 858 }
96fba48f 859 if (should_log_flow_message(error)) {
feebdea2
BP
860 const struct nlattr *actions;
861 size_t actions_len;
862
863 if (!error && actionsp) {
1f317cb5
PS
864 actions = ofpbuf_data(*actionsp);
865 actions_len = ofpbuf_size(*actionsp);
feebdea2
BP
866 } else {
867 actions = NULL;
868 actions_len = 0;
869 }
61fb711d
JP
870 log_flow_message(dpif, error, "flow_get", key, key_len,
871 NULL, 0, stats, actions, actions_len);
064af421 872 }
96fba48f 873 return error;
064af421
BP
874}
875
89625d1e
BP
876static int
877dpif_flow_put__(struct dpif *dpif, const struct dpif_flow_put *put)
878{
879 int error;
880
881 COVERAGE_INC(dpif_flow_put);
cb22974d
BP
882 ovs_assert(!(put->flags & ~(DPIF_FP_CREATE | DPIF_FP_MODIFY
883 | DPIF_FP_ZERO_STATS)));
89625d1e
BP
884
885 error = dpif->dpif_class->flow_put(dpif, put);
886 if (error && put->stats) {
887 memset(put->stats, 0, sizeof *put->stats);
888 }
889 log_flow_put_message(dpif, put, error);
890 return error;
891}
892
feebdea2 893/* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
e6cc0bab 894 * attribute OVS_FLOW_ATTR_KEY with types OVS_KEY_ATTR_* in the 'key_len' bytes
ee75c546
BP
895 * starting at 'key', and OVS_FLOW_ATTR_MASK with types of OVS_KEY_ATTR_* in
896 * the 'mask_len' bytes starting at 'mask'. The associated actions are
897 * specified by the Netlink attributes with types OVS_ACTION_ATTR_* in the
898 * 'actions_len' bytes starting at 'actions'.
96fba48f 899 *
feebdea2 900 * - If the flow's key does not exist in 'dpif', then the flow will be added if
ba25b8f4 901 * 'flags' includes DPIF_FP_CREATE. Otherwise the operation will fail with
96fba48f
BP
902 * ENOENT.
903 *
ee75c546
BP
904 * The datapath may reject attempts to insert overlapping flows with EINVAL
905 * or EEXIST, but clients should not rely on this: avoiding overlapping flows
906 * is primarily the client's responsibility.
907 *
feebdea2 908 * If the operation succeeds, then 'stats', if nonnull, will be zeroed.
96fba48f 909 *
feebdea2 910 * - If the flow's key does exist in 'dpif', then the flow's actions will be
ba25b8f4 911 * updated if 'flags' includes DPIF_FP_MODIFY. Otherwise the operation will
feebdea2 912 * fail with EEXIST. If the flow's actions are updated, then its statistics
ba25b8f4 913 * will be zeroed if 'flags' includes DPIF_FP_ZERO_STATS, and left as-is
feebdea2
BP
914 * otherwise.
915 *
916 * If the operation succeeds, then 'stats', if nonnull, will be set to the
917 * flow's statistics before the update.
96fba48f 918 */
064af421 919int
ba25b8f4 920dpif_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
feebdea2 921 const struct nlattr *key, size_t key_len,
e6cc0bab 922 const struct nlattr *mask, size_t mask_len,
feebdea2 923 const struct nlattr *actions, size_t actions_len,
c97fb132 924 struct dpif_flow_stats *stats)
064af421 925{
89625d1e
BP
926 struct dpif_flow_put put;
927
928 put.flags = flags;
929 put.key = key;
930 put.key_len = key_len;
e6cc0bab
AZ
931 put.mask = mask;
932 put.mask_len = mask_len;
89625d1e
BP
933 put.actions = actions;
934 put.actions_len = actions_len;
935 put.stats = stats;
936 return dpif_flow_put__(dpif, &put);
064af421
BP
937}
938
b99d3cee
BP
939static int
940dpif_flow_del__(struct dpif *dpif, struct dpif_flow_del *del)
941{
942 int error;
943
944 COVERAGE_INC(dpif_flow_del);
945
946 error = dpif->dpif_class->flow_del(dpif, del);
947 if (error && del->stats) {
948 memset(del->stats, 0, sizeof *del->stats);
949 }
950 log_flow_del_message(dpif, del, error);
951 return error;
952}
953
feebdea2
BP
954/* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif' does
955 * not contain such a flow. The flow is specified by the Netlink attributes
df2c07f4 956 * with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at 'key'.
96fba48f 957 *
feebdea2
BP
958 * If the operation succeeds, then 'stats', if nonnull, will be set to the
959 * flow's statistics before its deletion. */
064af421 960int
feebdea2
BP
961dpif_flow_del(struct dpif *dpif,
962 const struct nlattr *key, size_t key_len,
c97fb132 963 struct dpif_flow_stats *stats)
064af421 964{
b99d3cee 965 struct dpif_flow_del del;
f1aa2072 966
b99d3cee
BP
967 del.key = key;
968 del.key_len = key_len;
969 del.stats = stats;
970 return dpif_flow_del__(dpif, &del);
064af421
BP
971}
972
ac64794a
BP
973/* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
974 * flows in 'dpif'.
975 *
976 * This function always successfully returns a dpif_flow_dump. Error
977 * reporting is deferred to dpif_flow_dump_destroy(). */
978struct dpif_flow_dump *
979dpif_flow_dump_create(const struct dpif *dpif)
e723fd32 980{
ac64794a 981 return dpif->dpif_class->flow_dump_create(dpif);
e723fd32
JS
982}
983
ac64794a
BP
984/* Destroys 'dump', which must have been created with dpif_flow_dump_create().
985 * All dpif_flow_dump_thread structures previously created for 'dump' must
986 * previously have been destroyed.
987 *
988 * Returns 0 if the dump operation was error-free, otherwise a positive errno
989 * value describing the problem. */
990int
991dpif_flow_dump_destroy(struct dpif_flow_dump *dump)
e723fd32 992{
ac64794a
BP
993 const struct dpif *dpif = dump->dpif;
994 int error = dpif->dpif_class->flow_dump_destroy(dump);
995 log_operation(dpif, "flow_dump_destroy", error);
996 return error == EOF ? 0 : error;
e723fd32
JS
997}
998
ac64794a
BP
999/* Returns new thread-local state for use with dpif_flow_dump_next(). */
1000struct dpif_flow_dump_thread *
1001dpif_flow_dump_thread_create(struct dpif_flow_dump *dump)
064af421 1002{
ac64794a 1003 return dump->dpif->dpif_class->flow_dump_thread_create(dump);
064af421
BP
1004}
1005
ac64794a
BP
1006/* Releases 'thread'. */
1007void
1008dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread)
704a1e09 1009{
ac64794a 1010 thread->dpif->dpif_class->flow_dump_thread_destroy(thread);
704a1e09
BP
1011}
1012
ac64794a
BP
1013/* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0
1014 * if and only if no flows remained to be retrieved, otherwise a positive
1015 * number reflecting the number of elements in 'flows[]' that were updated.
1016 * The number of flows returned might be less than 'max_flows' because
1017 * fewer than 'max_flows' remained, because this particular datapath does not
1018 * benefit from batching, or because an error occurred partway through
1019 * retrieval. Thus, the caller should continue calling until a 0 return value,
1020 * even if intermediate return values are less than 'max_flows'.
bdeadfdd 1021 *
ac64794a
BP
1022 * No error status is immediately provided. An error status for the entire
1023 * dump operation is provided when it is completed by calling
1024 * dpif_flow_dump_destroy().
bdeadfdd 1025 *
ac64794a
BP
1026 * All of the data stored into 'flows' is owned by the datapath, not by the
1027 * caller, and the caller must not modify or free it. The datapath guarantees
1028 * that it remains accessible and unchanged until at least the next call to
1029 * dpif_flow_dump_next() for 'thread'. */
704a1e09 1030int
ac64794a
BP
1031dpif_flow_dump_next(struct dpif_flow_dump_thread *thread,
1032 struct dpif_flow *flows, int max_flows)
704a1e09 1033{
ac64794a
BP
1034 struct dpif *dpif = thread->dpif;
1035 int n;
1036
1037 ovs_assert(max_flows > 0);
1038 n = dpif->dpif_class->flow_dump_next(thread, flows, max_flows);
1039 if (n > 0) {
1040 struct dpif_flow *f;
1041
1042 for (f = flows; f < &flows[n] && should_log_flow_message(0); f++) {
1043 log_flow_message(dpif, 0, "flow_dump",
1044 f->key, f->key_len, f->mask, f->mask_len,
1045 &f->stats, f->actions, f->actions_len);
1046 }
1047 } else {
1048 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif));
1049 }
1050 return n;
064af421
BP
1051}
1052
7fd91025
BP
1053struct dpif_execute_helper_aux {
1054 struct dpif *dpif;
1055 int error;
1056};
1057
09f9da0b
JR
1058/* This is called for actions that need the context of the datapath to be
1059 * meaningful. */
7fd91025 1060static void
758c456d 1061dpif_execute_helper_cb(void *aux_, struct ofpbuf *packet,
572f732a 1062 struct pkt_metadata *md,
09f9da0b 1063 const struct nlattr *action, bool may_steal OVS_UNUSED)
7fd91025 1064{
758c456d
JR
1065 struct dpif_execute_helper_aux *aux = aux_;
1066 struct dpif_execute execute;
09f9da0b 1067 int type = nl_attr_type(action);
758c456d 1068
09f9da0b
JR
1069 switch ((enum ovs_action_attr)type) {
1070 case OVS_ACTION_ATTR_OUTPUT:
1071 case OVS_ACTION_ATTR_USERSPACE:
c6bf49f3 1072 case OVS_ACTION_ATTR_RECIRC:
758c456d
JR
1073 execute.actions = action;
1074 execute.actions_len = NLA_ALIGN(action->nla_len);
1075 execute.packet = packet;
1076 execute.md = *md;
1077 execute.needs_help = false;
1078 aux->error = aux->dpif->dpif_class->execute(aux->dpif, &execute);
09f9da0b 1079 break;
758c456d 1080
c6bf49f3 1081 case OVS_ACTION_ATTR_HASH:
09f9da0b
JR
1082 case OVS_ACTION_ATTR_PUSH_VLAN:
1083 case OVS_ACTION_ATTR_POP_VLAN:
1084 case OVS_ACTION_ATTR_PUSH_MPLS:
1085 case OVS_ACTION_ATTR_POP_MPLS:
1086 case OVS_ACTION_ATTR_SET:
1087 case OVS_ACTION_ATTR_SAMPLE:
1088 case OVS_ACTION_ATTR_UNSPEC:
1089 case __OVS_ACTION_ATTR_MAX:
1090 OVS_NOT_REACHED();
1091 }
7fd91025
BP
1092}
1093
1094/* Executes 'execute' by performing most of the actions in userspace and
1095 * passing the fully constructed packets to 'dpif' for output and userspace
1096 * actions.
1097 *
1098 * This helps with actions that a given 'dpif' doesn't implement directly. */
1099static int
758c456d 1100dpif_execute_with_help(struct dpif *dpif, struct dpif_execute *execute)
7fd91025 1101{
758c456d 1102 struct dpif_execute_helper_aux aux = {dpif, 0};
7fd91025
BP
1103
1104 COVERAGE_INC(dpif_execute_with_help);
1105
df1e5a3b 1106 odp_execute_actions(&aux, execute->packet, false, &execute->md,
7fd91025 1107 execute->actions, execute->actions_len,
09f9da0b 1108 dpif_execute_helper_cb);
7fd91025
BP
1109 return aux.error;
1110}
1111
758c456d
JR
1112/* Causes 'dpif' to perform the 'execute->actions_len' bytes of actions in
1113 * 'execute->actions' on the Ethernet frame in 'execute->packet' and on packet
1114 * metadata in 'execute->md'. The implementation is allowed to modify both the
1115 * '*execute->packet' and 'execute->md'.
1116 *
1117 * Some dpif providers do not implement every action. The Linux kernel
1118 * datapath, in particular, does not implement ARP field modification. If
1119 * 'needs_help' is true, the dpif layer executes in userspace all of the
1120 * actions that it can, and for OVS_ACTION_ATTR_OUTPUT and
1121 * OVS_ACTION_ATTR_USERSPACE actions it passes the packet through to the dpif
1122 * implementation.
1123 *
1124 * This works even if 'execute->actions_len' is too long for a Netlink
1125 * attribute.
1126 *
1127 * Returns 0 if successful, otherwise a positive errno value. */
1128int
1129dpif_execute(struct dpif *dpif, struct dpif_execute *execute)
89625d1e
BP
1130{
1131 int error;
1132
1133 COVERAGE_INC(dpif_execute);
1134 if (execute->actions_len > 0) {
758c456d 1135 error = (execute->needs_help || nl_attr_oversized(execute->actions_len)
7fd91025
BP
1136 ? dpif_execute_with_help(dpif, execute)
1137 : dpif->dpif_class->execute(dpif, execute));
89625d1e
BP
1138 } else {
1139 error = 0;
1140 }
1141
1142 log_execute_message(dpif, execute, error);
1143
1144 return error;
1145}
1146
6bc60024
BP
1147/* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1148 * which they are specified, placing each operation's results in the "output"
1149 * members documented in comments.
1150 *
1151 * This function exists because some datapaths can perform batched operations
1152 * faster than individual operations. */
1153void
c2b565b5 1154dpif_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
6bc60024 1155{
6bc60024 1156 if (dpif->dpif_class->operate) {
7fd91025
BP
1157 while (n_ops > 0) {
1158 size_t chunk;
1159
1160 /* Count 'chunk', the number of ops that can be executed without
1161 * needing any help. Ops that need help should be rare, so we
1162 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1163 for (chunk = 0; chunk < n_ops; chunk++) {
1164 struct dpif_op *op = ops[chunk];
1165
1166 if (op->type == DPIF_OP_EXECUTE && op->u.execute.needs_help) {
1167 break;
1168 }
1169 }
1170
1171 if (chunk) {
1172 /* Execute a chunk full of ops that the dpif provider can
1173 * handle itself, without help. */
1174 size_t i;
1175
1176 dpif->dpif_class->operate(dpif, ops, chunk);
1177
1178 for (i = 0; i < chunk; i++) {
1179 struct dpif_op *op = ops[i];
1180
1181 switch (op->type) {
1182 case DPIF_OP_FLOW_PUT:
1183 log_flow_put_message(dpif, &op->u.flow_put, op->error);
1184 break;
1185
1186 case DPIF_OP_FLOW_DEL:
1187 log_flow_del_message(dpif, &op->u.flow_del, op->error);
1188 break;
1189
1190 case DPIF_OP_EXECUTE:
1191 log_execute_message(dpif, &op->u.execute, op->error);
1192 break;
1193 }
1194 }
1195
1196 ops += chunk;
1197 n_ops -= chunk;
1198 } else {
1199 /* Help the dpif provider to execute one op. */
1200 struct dpif_op *op = ops[0];
1201
758c456d 1202 op->error = dpif_execute(dpif, &op->u.execute);
7fd91025
BP
1203 ops++;
1204 n_ops--;
1205 }
1206 }
1207 } else {
1208 size_t i;
f23d2845
BP
1209
1210 for (i = 0; i < n_ops; i++) {
1211 struct dpif_op *op = ops[i];
1212
1213 switch (op->type) {
1214 case DPIF_OP_FLOW_PUT:
7fd91025 1215 op->error = dpif_flow_put__(dpif, &op->u.flow_put);
f23d2845
BP
1216 break;
1217
b99d3cee 1218 case DPIF_OP_FLOW_DEL:
7fd91025 1219 op->error = dpif_flow_del__(dpif, &op->u.flow_del);
b99d3cee
BP
1220 break;
1221
f23d2845 1222 case DPIF_OP_EXECUTE:
758c456d 1223 op->error = dpif_execute(dpif, &op->u.execute);
f23d2845 1224 break;
b99d3cee 1225
7fd91025 1226 default:
428b2edd 1227 OVS_NOT_REACHED();
7fd91025 1228 }
6bc60024
BP
1229 }
1230 }
1231}
1232
01545c1a
BP
1233/* Returns a string that represents 'type', for use in log messages. */
1234const char *
1235dpif_upcall_type_to_string(enum dpif_upcall_type type)
1236{
1237 switch (type) {
1238 case DPIF_UC_MISS: return "miss";
1239 case DPIF_UC_ACTION: return "action";
01545c1a
BP
1240 case DPIF_N_UC_TYPES: default: return "<unknown>";
1241 }
1242}
1243
a12b3ead
BP
1244/* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1245 * if successful, otherwise a positive errno value.
98403001 1246 *
a12b3ead 1247 * Turning packet receive off and then back on may change the Netlink PID
98403001
BP
1248 * assignments returned by dpif_port_get_pid(). If the client does this, it
1249 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1250 * using the new PID assignment. */
8f24562a 1251int
a12b3ead 1252dpif_recv_set(struct dpif *dpif, bool enable)
8f24562a 1253{
a12b3ead
BP
1254 int error = dpif->dpif_class->recv_set(dpif, enable);
1255 log_operation(dpif, "recv_set", error);
96fba48f 1256 return error;
8f24562a
BP
1257}
1258
1954e6bb
AW
1259/* Refreshes the poll loops and Netlink sockets associated to each port,
1260 * when the number of upcall handlers (upcall receiving thread) is changed
1261 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
1262 * recv_set().
1263 *
1264 * Since multiple upcall handlers can read upcalls simultaneously from
1265 * 'dpif', each port can have multiple Netlink sockets, one per upcall
1266 * handler. So, handlers_set() is responsible for the following tasks:
1267 *
1268 * When receiving upcall is enabled, extends or creates the
1269 * configuration to support:
1270 *
1271 * - 'n_handlers' Netlink sockets for each port.
1272 *
1273 * - 'n_handlers' poll loops, one for each upcall handler.
1274 *
1275 * - registering the Netlink sockets for the same upcall handler to
1276 * the corresponding poll loop.
1277 *
1278 * Returns 0 if successful, otherwise a positive errno value. */
1279int
1280dpif_handlers_set(struct dpif *dpif, uint32_t n_handlers)
1281{
1282 int error = dpif->dpif_class->handlers_set(dpif, n_handlers);
1283 log_operation(dpif, "handlers_set", error);
1284 return error;
1285}
1286
1287/* Polls for an upcall from 'dpif' for an upcall handler. Since there
1288 * there can be multiple poll loops, 'handler_id' is needed as index to
1289 * identify the corresponding poll loop. If successful, stores the upcall
1290 * into '*upcall', using 'buf' for storage. Should only be called if
1291 * 'recv_set' has been used to enable receiving packets from 'dpif'.
96fba48f 1292 *
da546e07
JR
1293 * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
1294 * 'buf', so their memory cannot be freed separately from 'buf'.
856081f6 1295 *
837a88dc
JR
1296 * The caller owns the data of 'upcall->packet' and may modify it. If
1297 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
1298 * will be reallocated. This requires the data of 'upcall->packet' to be
1299 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
1300 * when an error is returned, the 'upcall->packet' may be uninitialized
1301 * and should not be released.
1302 *
96fba48f 1303 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
856081f6 1304 * if no upcall is immediately available. */
064af421 1305int
1954e6bb
AW
1306dpif_recv(struct dpif *dpif, uint32_t handler_id, struct dpif_upcall *upcall,
1307 struct ofpbuf *buf)
064af421 1308{
1954e6bb 1309 int error = dpif->dpif_class->recv(dpif, handler_id, upcall, buf);
856081f6 1310 if (!error && !VLOG_DROP_DBG(&dpmsg_rl)) {
01545c1a
BP
1311 struct ds flow;
1312 char *packet;
1313
1f317cb5
PS
1314 packet = ofp_packet_to_string(ofpbuf_data(&upcall->packet),
1315 ofpbuf_size(&upcall->packet));
01545c1a
BP
1316
1317 ds_init(&flow);
1318 odp_flow_key_format(upcall->key, upcall->key_len, &flow);
1319
1320 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1321 dpif_name(dpif), dpif_upcall_type_to_string(upcall->type),
1322 ds_cstr(&flow), packet);
1323
1324 ds_destroy(&flow);
1325 free(packet);
5fcc0d00
BP
1326 } else if (error && error != EAGAIN) {
1327 log_operation(dpif, "recv", error);
064af421 1328 }
064af421
BP
1329 return error;
1330}
1331
96fba48f 1332/* Discards all messages that would otherwise be received by dpif_recv() on
1ba530f4
BP
1333 * 'dpif'. */
1334void
96fba48f
BP
1335dpif_recv_purge(struct dpif *dpif)
1336{
96fba48f 1337 COVERAGE_INC(dpif_purge);
1ba530f4
BP
1338 if (dpif->dpif_class->recv_purge) {
1339 dpif->dpif_class->recv_purge(dpif);
96fba48f 1340 }
96fba48f
BP
1341}
1342
1954e6bb
AW
1343/* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
1344 * 'dpif' has a message queued to be received with the recv member
1345 * function. Since there can be multiple poll loops, 'handler_id' is
1346 * needed as index to identify the corresponding poll loop. */
064af421 1347void
1954e6bb 1348dpif_recv_wait(struct dpif *dpif, uint32_t handler_id)
064af421 1349{
1954e6bb 1350 dpif->dpif_class->recv_wait(dpif, handler_id);
064af421 1351}
53a4218d 1352
96fba48f
BP
1353/* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1354 * and '*engine_id', respectively. */
53a4218d
BP
1355void
1356dpif_get_netflow_ids(const struct dpif *dpif,
1357 uint8_t *engine_type, uint8_t *engine_id)
1358{
96fba48f
BP
1359 *engine_type = dpif->netflow_engine_type;
1360 *engine_id = dpif->netflow_engine_id;
1361}
aae51f53
BP
1362
1363/* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
abff858b
PS
1364 * value used for setting packet priority.
1365 * On success, returns 0 and stores the priority into '*priority'.
1366 * On failure, returns a positive errno value and stores 0 into '*priority'. */
aae51f53
BP
1367int
1368dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1369 uint32_t *priority)
1370{
1371 int error = (dpif->dpif_class->queue_to_priority
1372 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1373 priority)
1374 : EOPNOTSUPP);
1375 if (error) {
1376 *priority = 0;
1377 }
1378 log_operation(dpif, "queue_to_priority", error);
1379 return error;
1380}
96fba48f
BP
1381\f
1382void
1acb6baa
BP
1383dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1384 const char *name,
96fba48f
BP
1385 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1386{
1acb6baa 1387 dpif->dpif_class = dpif_class;
1a6f1e2a 1388 dpif->base_name = xstrdup(name);
a4af0040 1389 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
96fba48f
BP
1390 dpif->netflow_engine_type = netflow_engine_type;
1391 dpif->netflow_engine_id = netflow_engine_id;
1392}
999401aa
JG
1393
1394/* Undoes the results of initialization.
1395 *
1396 * Normally this function only needs to be called from dpif_close().
1397 * However, it may be called by providers due to an error on opening
1398 * that occurs after initialization. It this case dpif_close() would
1399 * never be called. */
1400void
1401dpif_uninit(struct dpif *dpif, bool close)
1402{
1403 char *base_name = dpif->base_name;
1404 char *full_name = dpif->full_name;
1405
1406 if (close) {
a4af0040 1407 dpif->dpif_class->close(dpif);
999401aa
JG
1408 }
1409
1410 free(base_name);
1411 free(full_name);
1412}
96fba48f
BP
1413\f
1414static void
1415log_operation(const struct dpif *dpif, const char *operation, int error)
1416{
1417 if (!error) {
1418 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
90bf1e07 1419 } else if (ofperr_is_valid(error)) {
96fba48f 1420 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
90bf1e07 1421 dpif_name(dpif), operation, ofperr_get_name(error));
71ce9235 1422 } else {
90bf1e07 1423 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
10a89ef0 1424 dpif_name(dpif), operation, ovs_strerror(error));
96fba48f
BP
1425 }
1426}
1427
1428static enum vlog_level
1429flow_message_log_level(int error)
1430{
9b1a48c2
JP
1431 /* If flows arrive in a batch, userspace may push down multiple
1432 * unique flow definitions that overlap when wildcards are applied.
1433 * Kernels that support flow wildcarding will reject these flows as
1434 * duplicates (EEXIST), so lower the log level to debug for these
1435 * types of messages. */
1436 return (error && error != EEXIST) ? VLL_WARN : VLL_DBG;
96fba48f
BP
1437}
1438
1439static bool
1440should_log_flow_message(int error)
1441{
1442 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1443 error ? &error_rl : &dpmsg_rl);
1444}
1445
1446static void
1447log_flow_message(const struct dpif *dpif, int error, const char *operation,
36956a7d 1448 const struct nlattr *key, size_t key_len,
61fb711d 1449 const struct nlattr *mask, size_t mask_len,
c97fb132 1450 const struct dpif_flow_stats *stats,
cf22f8cb 1451 const struct nlattr *actions, size_t actions_len)
96fba48f
BP
1452{
1453 struct ds ds = DS_EMPTY_INITIALIZER;
1454 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1455 if (error) {
1456 ds_put_cstr(&ds, "failed to ");
1457 }
1458 ds_put_format(&ds, "%s ", operation);
1459 if (error) {
10a89ef0 1460 ds_put_format(&ds, "(%s) ", ovs_strerror(error));
96fba48f 1461 }
0a37839c 1462 odp_flow_format(key, key_len, mask, mask_len, NULL, &ds, true);
96fba48f
BP
1463 if (stats) {
1464 ds_put_cstr(&ds, ", ");
c97fb132 1465 dpif_flow_stats_format(stats, &ds);
96fba48f 1466 }
cdee00fd 1467 if (actions || actions_len) {
96fba48f 1468 ds_put_cstr(&ds, ", actions:");
cdee00fd 1469 format_odp_actions(&ds, actions, actions_len);
96fba48f
BP
1470 }
1471 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1472 ds_destroy(&ds);
1473}
89625d1e
BP
1474
1475static void
1476log_flow_put_message(struct dpif *dpif, const struct dpif_flow_put *put,
1477 int error)
1478{
1479 if (should_log_flow_message(error)) {
1480 struct ds s;
1481
1482 ds_init(&s);
1483 ds_put_cstr(&s, "put");
1484 if (put->flags & DPIF_FP_CREATE) {
1485 ds_put_cstr(&s, "[create]");
1486 }
1487 if (put->flags & DPIF_FP_MODIFY) {
1488 ds_put_cstr(&s, "[modify]");
1489 }
1490 if (put->flags & DPIF_FP_ZERO_STATS) {
1491 ds_put_cstr(&s, "[zero]");
1492 }
1493 log_flow_message(dpif, error, ds_cstr(&s),
61fb711d
JP
1494 put->key, put->key_len, put->mask, put->mask_len,
1495 put->stats, put->actions, put->actions_len);
89625d1e
BP
1496 ds_destroy(&s);
1497 }
1498}
1499
b99d3cee
BP
1500static void
1501log_flow_del_message(struct dpif *dpif, const struct dpif_flow_del *del,
1502 int error)
1503{
1504 if (should_log_flow_message(error)) {
1505 log_flow_message(dpif, error, "flow_del", del->key, del->key_len,
61fb711d 1506 NULL, 0, !error ? del->stats : NULL, NULL, 0);
b99d3cee
BP
1507 }
1508}
1509
89625d1e
BP
1510static void
1511log_execute_message(struct dpif *dpif, const struct dpif_execute *execute,
1512 int error)
1513{
1514 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
1515 struct ds ds = DS_EMPTY_INITIALIZER;
1516 char *packet;
1517
1f317cb5
PS
1518 packet = ofp_packet_to_string(ofpbuf_data(execute->packet),
1519 ofpbuf_size(execute->packet));
89625d1e
BP
1520 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
1521 format_odp_actions(&ds, execute->actions, execute->actions_len);
1522 if (error) {
10a89ef0 1523 ds_put_format(&ds, " failed (%s)", ovs_strerror(error));
89625d1e
BP
1524 }
1525 ds_put_format(&ds, " on packet %s", packet);
1526 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
1527 ds_destroy(&ds);
1528 free(packet);
1529 }
1530}