]> git.proxmox.com Git - ovs.git/blame - lib/dpif.c
vswitchd: skip right number of arguments in dpdk_init()
[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;
2240af25
DDP
275 error = (dpif_class->enumerate
276 ? dpif_class->enumerate(names, dpif_class)
277 : 0);
999401aa
JG
278 if (error) {
279 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
10a89ef0 280 ovs_strerror(error));
d3d22744 281 }
5703b15f 282 dp_class_unref(registered_class);
1a6f1e2a 283
999401aa 284 return error;
1a6f1e2a
JG
285}
286
54ed8a5d
BP
287/* Parses 'datapath_name_', which is of the form [type@]name into its
288 * component pieces. 'name' and 'type' must be freed by the caller.
289 *
290 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
1a6f1e2a
JG
291void
292dp_parse_name(const char *datapath_name_, char **name, char **type)
293{
294 char *datapath_name = xstrdup(datapath_name_);
295 char *separator;
296
297 separator = strchr(datapath_name, '@');
298 if (separator) {
299 *separator = '\0';
300 *type = datapath_name;
54ed8a5d 301 *name = xstrdup(dpif_normalize_type(separator + 1));
1a6f1e2a
JG
302 } else {
303 *name = datapath_name;
54ed8a5d 304 *type = xstrdup(dpif_normalize_type(NULL));
1a6f1e2a 305 }
d3d22744
BP
306}
307
96fba48f 308static int
1a6f1e2a 309do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
064af421 310{
96fba48f 311 struct dpif *dpif = NULL;
064af421 312 int error;
999401aa
JG
313 struct registered_dpif_class *registered_class;
314
315 dp_initialize();
064af421 316
3a225db7 317 type = dpif_normalize_type(type);
5703b15f 318 registered_class = dp_class_lookup(type);
999401aa
JG
319 if (!registered_class) {
320 VLOG_WARN("could not create datapath %s of unknown type %s", name,
321 type);
322 error = EAFNOSUPPORT;
323 goto exit;
324 }
325
4a387741
BP
326 error = registered_class->dpif_class->open(registered_class->dpif_class,
327 name, create, &dpif);
999401aa 328 if (!error) {
cb22974d 329 ovs_assert(dpif->dpif_class == registered_class->dpif_class);
5703b15f
BP
330 } else {
331 dp_class_unref(registered_class);
064af421 332 }
064af421 333
96fba48f
BP
334exit:
335 *dpifp = error ? NULL : dpif;
336 return error;
064af421
BP
337}
338
1a6f1e2a
JG
339/* Tries to open an existing datapath named 'name' and type 'type'. Will fail
340 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
341 * the empty string to specify the default system type. Returns 0 if
342 * successful, otherwise a positive errno value. On success stores a pointer
343 * to the datapath in '*dpifp', otherwise a null pointer. */
96fba48f 344int
1a6f1e2a 345dpif_open(const char *name, const char *type, struct dpif **dpifp)
064af421 346{
1a6f1e2a 347 return do_open(name, type, false, dpifp);
064af421
BP
348}
349
1a6f1e2a
JG
350/* Tries to create and open a new datapath with the given 'name' and 'type'.
351 * 'type' may be either NULL or the empty string to specify the default system
352 * type. Will fail if a datapath with 'name' and 'type' already exists.
353 * Returns 0 if successful, otherwise a positive errno value. On success
354 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
064af421 355int
1a6f1e2a 356dpif_create(const char *name, const char *type, struct dpif **dpifp)
064af421 357{
1a6f1e2a 358 return do_open(name, type, true, dpifp);
96fba48f 359}
064af421 360
1a6f1e2a
JG
361/* Tries to open a datapath with the given 'name' and 'type', creating it if it
362 * does not exist. 'type' may be either NULL or the empty string to specify
363 * the default system type. Returns 0 if successful, otherwise a positive
364 * errno value. On success stores a pointer to the datapath in '*dpifp',
365 * otherwise a null pointer. */
efacbce6 366int
1a6f1e2a 367dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
efacbce6
BP
368{
369 int error;
370
1a6f1e2a 371 error = dpif_create(name, type, dpifp);
efacbce6 372 if (error == EEXIST || error == EBUSY) {
1a6f1e2a 373 error = dpif_open(name, type, dpifp);
efacbce6
BP
374 if (error) {
375 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
10a89ef0 376 name, ovs_strerror(error));
efacbce6
BP
377 }
378 } else if (error) {
10a89ef0
BP
379 VLOG_WARN("failed to create datapath %s: %s",
380 name, ovs_strerror(error));
efacbce6
BP
381 }
382 return error;
383}
384
96fba48f
BP
385/* Closes and frees the connection to 'dpif'. Does not destroy the datapath
386 * itself; call dpif_delete() first, instead, if that is desirable. */
387void
388dpif_close(struct dpif *dpif)
389{
390 if (dpif) {
5703b15f 391 struct registered_dpif_class *rc;
999401aa 392
5703b15f 393 rc = shash_find_data(&dpif_classes, dpif->dpif_class->type);
999401aa 394 dpif_uninit(dpif, true);
5703b15f 395 dp_class_unref(rc);
064af421
BP
396 }
397}
398
640e1b20
BP
399/* Performs periodic work needed by 'dpif'. */
400void
401dpif_run(struct dpif *dpif)
402{
403 if (dpif->dpif_class->run) {
404 dpif->dpif_class->run(dpif);
405 }
406}
407
408/* Arranges for poll_block() to wake up when dp_run() needs to be called for
409 * 'dpif'. */
410void
411dpif_wait(struct dpif *dpif)
412{
413 if (dpif->dpif_class->wait) {
414 dpif->dpif_class->wait(dpif);
415 }
416}
417
1a6f1e2a
JG
418/* Returns the name of datapath 'dpif' prefixed with the type
419 * (for use in log messages). */
b29ba128
BP
420const char *
421dpif_name(const struct dpif *dpif)
422{
1a6f1e2a
JG
423 return dpif->full_name;
424}
425
426/* Returns the name of datapath 'dpif' without the type
427 * (for use in device names). */
428const char *
429dpif_base_name(const struct dpif *dpif)
430{
431 return dpif->base_name;
b29ba128
BP
432}
433
c7a26215
JP
434/* Returns the type of datapath 'dpif'. */
435const char *
436dpif_type(const struct dpif *dpif)
437{
438 return dpif->dpif_class->type;
439}
440
3a225db7
BP
441/* Returns the fully spelled out name for the given datapath 'type'.
442 *
443 * Normalized type string can be compared with strcmp(). Unnormalized type
444 * string might be the same even if they have different spellings. */
445const char *
446dpif_normalize_type(const char *type)
447{
448 return type && type[0] ? type : "system";
449}
450
96fba48f
BP
451/* Destroys the datapath that 'dpif' is connected to, first removing all of its
452 * ports. After calling this function, it does not make sense to pass 'dpif'
453 * to any functions other than dpif_name() or dpif_close(). */
064af421
BP
454int
455dpif_delete(struct dpif *dpif)
456{
96fba48f
BP
457 int error;
458
064af421 459 COVERAGE_INC(dpif_destroy);
96fba48f 460
1acb6baa 461 error = dpif->dpif_class->destroy(dpif);
96fba48f
BP
462 log_operation(dpif, "delete", error);
463 return error;
064af421
BP
464}
465
96fba48f
BP
466/* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
467 * otherwise a positive errno value. */
064af421 468int
a8d9304d 469dpif_get_dp_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
064af421 470{
1acb6baa 471 int error = dpif->dpif_class->get_stats(dpif, stats);
96fba48f
BP
472 if (error) {
473 memset(stats, 0, sizeof *stats);
474 }
475 log_operation(dpif, "get_stats", error);
476 return error;
064af421
BP
477}
478
0aeaabc8
JP
479const char *
480dpif_port_open_type(const char *datapath_type, const char *port_type)
481{
5703b15f 482 struct registered_dpif_class *rc;
0aeaabc8
JP
483
484 datapath_type = dpif_normalize_type(datapath_type);
485
97be1538 486 ovs_mutex_lock(&dpif_mutex);
5703b15f
BP
487 rc = shash_find_data(&dpif_classes, datapath_type);
488 if (rc && rc->dpif_class->port_open_type) {
489 port_type = rc->dpif_class->port_open_type(rc->dpif_class, port_type);
0aeaabc8 490 }
97be1538 491 ovs_mutex_unlock(&dpif_mutex);
0aeaabc8 492
5703b15f 493 return port_type;
0aeaabc8
JP
494}
495
232dfa4a 496/* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
4e022ec0 497 * non-null and its value is not ODPP_NONE, then attempts to use the
232dfa4a
JP
498 * value as the port number.
499 *
500 * If successful, returns 0 and sets '*port_nop' to the new port's port
501 * number (if 'port_nop' is non-null). On failure, returns a positive
4e022ec0 502 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
232dfa4a 503 * non-null). */
064af421 504int
4e022ec0 505dpif_port_add(struct dpif *dpif, struct netdev *netdev, odp_port_t *port_nop)
064af421 506{
c3827f61 507 const char *netdev_name = netdev_get_name(netdev);
4e022ec0 508 odp_port_t port_no = ODPP_NONE;
9ee3ae3e 509 int error;
064af421
BP
510
511 COVERAGE_INC(dpif_port_add);
9ee3ae3e 512
232dfa4a
JP
513 if (port_nop) {
514 port_no = *port_nop;
515 }
516
c3827f61 517 error = dpif->dpif_class->port_add(dpif, netdev, &port_no);
9ee3ae3e 518 if (!error) {
9b56fe13 519 VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu32,
c3827f61 520 dpif_name(dpif), netdev_name, port_no);
064af421 521 } else {
9ee3ae3e 522 VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
10a89ef0 523 dpif_name(dpif), netdev_name, ovs_strerror(error));
4e022ec0 524 port_no = ODPP_NONE;
9ee3ae3e
BP
525 }
526 if (port_nop) {
527 *port_nop = port_no;
064af421 528 }
9ee3ae3e 529 return error;
064af421
BP
530}
531
96fba48f
BP
532/* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
533 * otherwise a positive errno value. */
064af421 534int
4e022ec0 535dpif_port_del(struct dpif *dpif, odp_port_t port_no)
064af421 536{
96fba48f
BP
537 int error;
538
064af421 539 COVERAGE_INC(dpif_port_del);
96fba48f 540
1acb6baa 541 error = dpif->dpif_class->port_del(dpif, port_no);
a1811296 542 if (!error) {
9b56fe13 543 VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu32")",
a1811296
BP
544 dpif_name(dpif), port_no);
545 } else {
546 log_operation(dpif, "port_del", error);
547 }
96fba48f 548 return error;
064af421
BP
549}
550
4c738a8d
BP
551/* Makes a deep copy of 'src' into 'dst'. */
552void
553dpif_port_clone(struct dpif_port *dst, const struct dpif_port *src)
554{
555 dst->name = xstrdup(src->name);
556 dst->type = xstrdup(src->type);
557 dst->port_no = src->port_no;
558}
559
560/* Frees memory allocated to members of 'dpif_port'.
561 *
562 * Do not call this function on a dpif_port obtained from
563 * dpif_port_dump_next(): that function retains ownership of the data in the
564 * dpif_port. */
565void
566dpif_port_destroy(struct dpif_port *dpif_port)
567{
568 free(dpif_port->name);
569 free(dpif_port->type);
570}
571
4afba28d
JP
572/* Checks if port named 'devname' exists in 'dpif'. If so, returns
573 * true; otherwise, returns false. */
574bool
575dpif_port_exists(const struct dpif *dpif, const char *devname)
576{
577 int error = dpif->dpif_class->port_query_by_name(dpif, devname, NULL);
bee6b8bc 578 if (error != 0 && error != ENOENT && error != ENODEV) {
4afba28d 579 VLOG_WARN_RL(&error_rl, "%s: failed to query port %s: %s",
10a89ef0 580 dpif_name(dpif), devname, ovs_strerror(error));
4afba28d
JP
581 }
582
583 return !error;
584}
585
96fba48f
BP
586/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
587 * initializes '*port' appropriately; on failure, returns a positive errno
4c738a8d
BP
588 * value.
589 *
590 * The caller owns the data in 'port' and must free it with
591 * dpif_port_destroy() when it is no longer needed. */
064af421 592int
4e022ec0 593dpif_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
4c738a8d 594 struct dpif_port *port)
064af421 595{
1acb6baa 596 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
96fba48f 597 if (!error) {
9b56fe13 598 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu32" is device %s",
4c738a8d 599 dpif_name(dpif), port_no, port->name);
064af421 600 } else {
96fba48f 601 memset(port, 0, sizeof *port);
9b56fe13 602 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu32": %s",
10a89ef0 603 dpif_name(dpif), port_no, ovs_strerror(error));
064af421 604 }
96fba48f 605 return error;
064af421
BP
606}
607
96fba48f
BP
608/* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
609 * initializes '*port' appropriately; on failure, returns a positive errno
4c738a8d
BP
610 * value.
611 *
612 * The caller owns the data in 'port' and must free it with
613 * dpif_port_destroy() when it is no longer needed. */
064af421
BP
614int
615dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
4c738a8d 616 struct dpif_port *port)
064af421 617{
1acb6baa 618 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
96fba48f 619 if (!error) {
9b56fe13 620 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu32,
4c738a8d 621 dpif_name(dpif), devname, port->port_no);
064af421 622 } else {
96fba48f
BP
623 memset(port, 0, sizeof *port);
624
d647f0a7
BP
625 /* For ENOENT or ENODEV we use DBG level because the caller is probably
626 * interested in whether 'dpif' actually has a port 'devname', so that
627 * it's not an issue worth logging if it doesn't. Other errors are
628 * uncommon and more likely to indicate a real problem. */
629 VLOG_RL(&error_rl,
630 error == ENOENT || error == ENODEV ? VLL_DBG : VLL_WARN,
631 "%s: failed to query port %s: %s",
10a89ef0 632 dpif_name(dpif), devname, ovs_strerror(error));
064af421 633 }
96fba48f 634 return error;
064af421
BP
635}
636
1954e6bb
AW
637/* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
638 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
639 * flows whose packets arrived on port 'port_no'. In the case where the
640 * provider allocates multiple Netlink PIDs to a single port, it may use
641 * 'hash' to spread load among them. The caller need not use a particular
642 * hash function; a 5-tuple hash is suitable.
643 *
644 * (The datapath implementation might use some different hash function for
645 * distributing packets received via flow misses among PIDs. This means
646 * that packets received via flow misses might be reordered relative to
647 * packets received via userspace actions. This is not ordinarily a
648 * problem.)
98403001 649 *
4e022ec0 650 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
625b0720
BP
651 * allocated to any port, that the client may use for special purposes.
652 *
98403001
BP
653 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
654 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
655 * disabled and then re-enabled, so a client that does that must be prepared to
656 * update all of the flows that it installed that contain
657 * OVS_ACTION_ATTR_USERSPACE actions. */
658uint32_t
1954e6bb 659dpif_port_get_pid(const struct dpif *dpif, odp_port_t port_no, uint32_t hash)
98403001
BP
660{
661 return (dpif->dpif_class->port_get_pid
1954e6bb 662 ? (dpif->dpif_class->port_get_pid)(dpif, port_no, hash)
98403001
BP
663 : 0);
664}
665
96fba48f
BP
666/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
667 * the port's name into the 'name_size' bytes in 'name', ensuring that the
668 * result is null-terminated. On failure, returns a positive errno value and
669 * makes 'name' the empty string. */
335562c0 670int
4e022ec0 671dpif_port_get_name(struct dpif *dpif, odp_port_t port_no,
335562c0
BP
672 char *name, size_t name_size)
673{
4c738a8d 674 struct dpif_port port;
335562c0
BP
675 int error;
676
cb22974d 677 ovs_assert(name_size > 0);
335562c0
BP
678
679 error = dpif_port_query_by_number(dpif, port_no, &port);
680 if (!error) {
4c738a8d
BP
681 ovs_strlcpy(name, port.name, name_size);
682 dpif_port_destroy(&port);
335562c0
BP
683 } else {
684 *name = '\0';
685 }
686 return error;
687}
688
b0ec0f27 689/* Initializes 'dump' to begin dumping the ports in a dpif.
96fba48f 690 *
b0ec0f27
BP
691 * This function provides no status indication. An error status for the entire
692 * dump operation is provided when it is completed by calling
693 * dpif_port_dump_done().
694 */
695void
696dpif_port_dump_start(struct dpif_port_dump *dump, const struct dpif *dpif)
697{
698 dump->dpif = dpif;
699 dump->error = dpif->dpif_class->port_dump_start(dpif, &dump->state);
700 log_operation(dpif, "port_dump_start", dump->error);
701}
702
703/* Attempts to retrieve another port from 'dump', which must have been
4c738a8d 704 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
b0ec0f27 705 * into 'port' and returns true. On failure, returns false.
96fba48f 706 *
b0ec0f27
BP
707 * Failure might indicate an actual error or merely that the last port has been
708 * dumped. An error status for the entire dump operation is provided when it
4c738a8d
BP
709 * is completed by calling dpif_port_dump_done().
710 *
711 * The dpif owns the data stored in 'port'. It will remain valid until at
712 * least the next time 'dump' is passed to dpif_port_dump_next() or
713 * dpif_port_dump_done(). */
b0ec0f27 714bool
4c738a8d 715dpif_port_dump_next(struct dpif_port_dump *dump, struct dpif_port *port)
064af421 716{
b0ec0f27 717 const struct dpif *dpif = dump->dpif;
064af421 718
b0ec0f27
BP
719 if (dump->error) {
720 return false;
721 }
f4ba4c4f 722
b0ec0f27
BP
723 dump->error = dpif->dpif_class->port_dump_next(dpif, dump->state, port);
724 if (dump->error == EOF) {
725 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all ports", dpif_name(dpif));
726 } else {
727 log_operation(dpif, "port_dump_next", dump->error);
728 }
064af421 729
b0ec0f27
BP
730 if (dump->error) {
731 dpif->dpif_class->port_dump_done(dpif, dump->state);
732 return false;
f4ba4c4f 733 }
b0ec0f27
BP
734 return true;
735}
064af421 736
b0ec0f27
BP
737/* Completes port table dump operation 'dump', which must have been initialized
738 * with dpif_port_dump_start(). Returns 0 if the dump operation was
739 * error-free, otherwise a positive errno value describing the problem. */
740int
741dpif_port_dump_done(struct dpif_port_dump *dump)
742{
743 const struct dpif *dpif = dump->dpif;
744 if (!dump->error) {
745 dump->error = dpif->dpif_class->port_dump_done(dpif, dump->state);
746 log_operation(dpif, "port_dump_done", dump->error);
f4ba4c4f 747 }
b0ec0f27 748 return dump->error == EOF ? 0 : dump->error;
064af421
BP
749}
750
e9e28be3
BP
751/* Polls for changes in the set of ports in 'dpif'. If the set of ports in
752 * 'dpif' has changed, this function does one of the following:
753 *
754 * - Stores the name of the device that was added to or deleted from 'dpif' in
755 * '*devnamep' and returns 0. The caller is responsible for freeing
756 * '*devnamep' (with free()) when it no longer needs it.
757 *
758 * - Returns ENOBUFS and sets '*devnamep' to NULL.
759 *
760 * This function may also return 'false positives', where it returns 0 and
761 * '*devnamep' names a device that was not actually added or deleted or it
762 * returns ENOBUFS without any change.
763 *
764 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
765 * return other positive errno values to indicate that something has gone
766 * wrong. */
767int
768dpif_port_poll(const struct dpif *dpif, char **devnamep)
769{
1acb6baa 770 int error = dpif->dpif_class->port_poll(dpif, devnamep);
e9e28be3
BP
771 if (error) {
772 *devnamep = NULL;
773 }
774 return error;
775}
776
777/* Arranges for the poll loop to wake up when port_poll(dpif) will return a
778 * value other than EAGAIN. */
779void
780dpif_port_poll_wait(const struct dpif *dpif)
781{
1acb6baa 782 dpif->dpif_class->port_poll_wait(dpif);
e9e28be3
BP
783}
784
572b7068 785/* Extracts the flow stats for a packet. The 'flow' and 'packet'
a7752d4a
BP
786 * arguments must have been initialized through a call to flow_extract().
787 * 'used' is stored into stats->used. */
572b7068 788void
a39edbd4 789dpif_flow_stats_extract(const struct flow *flow, const struct ofpbuf *packet,
a7752d4a 790 long long int used, struct dpif_flow_stats *stats)
572b7068 791{
e0eecb1c 792 stats->tcp_flags = ntohs(flow->tcp_flags);
1f317cb5 793 stats->n_bytes = ofpbuf_size(packet);
572b7068 794 stats->n_packets = 1;
a7752d4a 795 stats->used = used;
572b7068
BP
796}
797
c97fb132
BP
798/* Appends a human-readable representation of 'stats' to 's'. */
799void
800dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s)
801{
802 ds_put_format(s, "packets:%"PRIu64", bytes:%"PRIu64", used:",
803 stats->n_packets, stats->n_bytes);
804 if (stats->used) {
805 ds_put_format(s, "%.3fs", (time_msec() - stats->used) / 1000.0);
806 } else {
807 ds_put_format(s, "never");
808 }
7393104d
BP
809 if (stats->tcp_flags) {
810 ds_put_cstr(s, ", flags:");
811 packet_format_tcp_flags(s, stats->tcp_flags);
812 }
c97fb132
BP
813}
814
96fba48f
BP
815/* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
816 * positive errno value. */
817int
818dpif_flow_flush(struct dpif *dpif)
064af421 819{
96fba48f
BP
820 int error;
821
822 COVERAGE_INC(dpif_flow_flush);
823
1acb6baa 824 error = dpif->dpif_class->flow_flush(dpif);
96fba48f
BP
825 log_operation(dpif, "flow_flush", error);
826 return error;
064af421
BP
827}
828
feebdea2 829/* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
df2c07f4 830 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
feebdea2 831 * 'key'.
96fba48f 832 *
feebdea2
BP
833 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
834 * failure, returns a positive errno value.
96fba48f 835 *
feebdea2
BP
836 * If 'actionsp' is nonnull, then on success '*actionsp' will be set to an
837 * ofpbuf owned by the caller that contains the Netlink attributes for the
838 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete()) when
839 * it is no longer needed.
840 *
841 * If 'stats' is nonnull, then on success it will be updated with the flow's
842 * statistics. */
96fba48f 843int
693c4a01 844dpif_flow_get(const struct dpif *dpif,
feebdea2 845 const struct nlattr *key, size_t key_len,
c97fb132 846 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
064af421 847{
96fba48f
BP
848 int error;
849
850 COVERAGE_INC(dpif_flow_get);
851
693c4a01 852 error = dpif->dpif_class->flow_get(dpif, key, key_len, actionsp, stats);
b843fa1b 853 if (error) {
feebdea2
BP
854 if (actionsp) {
855 *actionsp = NULL;
856 }
857 if (stats) {
858 memset(stats, 0, sizeof *stats);
859 }
b843fa1b 860 }
96fba48f 861 if (should_log_flow_message(error)) {
feebdea2
BP
862 const struct nlattr *actions;
863 size_t actions_len;
864
865 if (!error && actionsp) {
1f317cb5
PS
866 actions = ofpbuf_data(*actionsp);
867 actions_len = ofpbuf_size(*actionsp);
feebdea2
BP
868 } else {
869 actions = NULL;
870 actions_len = 0;
871 }
61fb711d
JP
872 log_flow_message(dpif, error, "flow_get", key, key_len,
873 NULL, 0, stats, actions, actions_len);
064af421 874 }
96fba48f 875 return error;
064af421
BP
876}
877
89625d1e
BP
878static int
879dpif_flow_put__(struct dpif *dpif, const struct dpif_flow_put *put)
880{
881 int error;
882
883 COVERAGE_INC(dpif_flow_put);
cb22974d
BP
884 ovs_assert(!(put->flags & ~(DPIF_FP_CREATE | DPIF_FP_MODIFY
885 | DPIF_FP_ZERO_STATS)));
89625d1e
BP
886
887 error = dpif->dpif_class->flow_put(dpif, put);
888 if (error && put->stats) {
889 memset(put->stats, 0, sizeof *put->stats);
890 }
891 log_flow_put_message(dpif, put, error);
892 return error;
893}
894
feebdea2 895/* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
e6cc0bab 896 * attribute OVS_FLOW_ATTR_KEY with types OVS_KEY_ATTR_* in the 'key_len' bytes
ee75c546
BP
897 * starting at 'key', and OVS_FLOW_ATTR_MASK with types of OVS_KEY_ATTR_* in
898 * the 'mask_len' bytes starting at 'mask'. The associated actions are
899 * specified by the Netlink attributes with types OVS_ACTION_ATTR_* in the
900 * 'actions_len' bytes starting at 'actions'.
96fba48f 901 *
feebdea2 902 * - If the flow's key does not exist in 'dpif', then the flow will be added if
ba25b8f4 903 * 'flags' includes DPIF_FP_CREATE. Otherwise the operation will fail with
96fba48f
BP
904 * ENOENT.
905 *
ee75c546
BP
906 * The datapath may reject attempts to insert overlapping flows with EINVAL
907 * or EEXIST, but clients should not rely on this: avoiding overlapping flows
908 * is primarily the client's responsibility.
909 *
feebdea2 910 * If the operation succeeds, then 'stats', if nonnull, will be zeroed.
96fba48f 911 *
feebdea2 912 * - If the flow's key does exist in 'dpif', then the flow's actions will be
ba25b8f4 913 * updated if 'flags' includes DPIF_FP_MODIFY. Otherwise the operation will
feebdea2 914 * fail with EEXIST. If the flow's actions are updated, then its statistics
ba25b8f4 915 * will be zeroed if 'flags' includes DPIF_FP_ZERO_STATS, and left as-is
feebdea2
BP
916 * otherwise.
917 *
918 * If the operation succeeds, then 'stats', if nonnull, will be set to the
919 * flow's statistics before the update.
96fba48f 920 */
064af421 921int
ba25b8f4 922dpif_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
feebdea2 923 const struct nlattr *key, size_t key_len,
e6cc0bab 924 const struct nlattr *mask, size_t mask_len,
feebdea2 925 const struct nlattr *actions, size_t actions_len,
c97fb132 926 struct dpif_flow_stats *stats)
064af421 927{
89625d1e
BP
928 struct dpif_flow_put put;
929
930 put.flags = flags;
931 put.key = key;
932 put.key_len = key_len;
e6cc0bab
AZ
933 put.mask = mask;
934 put.mask_len = mask_len;
89625d1e
BP
935 put.actions = actions;
936 put.actions_len = actions_len;
937 put.stats = stats;
938 return dpif_flow_put__(dpif, &put);
064af421
BP
939}
940
b99d3cee
BP
941static int
942dpif_flow_del__(struct dpif *dpif, struct dpif_flow_del *del)
943{
944 int error;
945
946 COVERAGE_INC(dpif_flow_del);
947
948 error = dpif->dpif_class->flow_del(dpif, del);
949 if (error && del->stats) {
950 memset(del->stats, 0, sizeof *del->stats);
951 }
952 log_flow_del_message(dpif, del, error);
953 return error;
954}
955
feebdea2
BP
956/* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif' does
957 * not contain such a flow. The flow is specified by the Netlink attributes
df2c07f4 958 * with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at 'key'.
96fba48f 959 *
feebdea2
BP
960 * If the operation succeeds, then 'stats', if nonnull, will be set to the
961 * flow's statistics before its deletion. */
064af421 962int
feebdea2
BP
963dpif_flow_del(struct dpif *dpif,
964 const struct nlattr *key, size_t key_len,
c97fb132 965 struct dpif_flow_stats *stats)
064af421 966{
b99d3cee 967 struct dpif_flow_del del;
f1aa2072 968
b99d3cee
BP
969 del.key = key;
970 del.key_len = key_len;
971 del.stats = stats;
972 return dpif_flow_del__(dpif, &del);
064af421
BP
973}
974
ac64794a
BP
975/* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
976 * flows in 'dpif'.
977 *
978 * This function always successfully returns a dpif_flow_dump. Error
979 * reporting is deferred to dpif_flow_dump_destroy(). */
980struct dpif_flow_dump *
981dpif_flow_dump_create(const struct dpif *dpif)
e723fd32 982{
ac64794a 983 return dpif->dpif_class->flow_dump_create(dpif);
e723fd32
JS
984}
985
ac64794a
BP
986/* Destroys 'dump', which must have been created with dpif_flow_dump_create().
987 * All dpif_flow_dump_thread structures previously created for 'dump' must
988 * previously have been destroyed.
989 *
990 * Returns 0 if the dump operation was error-free, otherwise a positive errno
991 * value describing the problem. */
992int
993dpif_flow_dump_destroy(struct dpif_flow_dump *dump)
e723fd32 994{
ac64794a
BP
995 const struct dpif *dpif = dump->dpif;
996 int error = dpif->dpif_class->flow_dump_destroy(dump);
997 log_operation(dpif, "flow_dump_destroy", error);
998 return error == EOF ? 0 : error;
e723fd32
JS
999}
1000
ac64794a
BP
1001/* Returns new thread-local state for use with dpif_flow_dump_next(). */
1002struct dpif_flow_dump_thread *
1003dpif_flow_dump_thread_create(struct dpif_flow_dump *dump)
064af421 1004{
ac64794a 1005 return dump->dpif->dpif_class->flow_dump_thread_create(dump);
064af421
BP
1006}
1007
ac64794a
BP
1008/* Releases 'thread'. */
1009void
1010dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread *thread)
704a1e09 1011{
ac64794a 1012 thread->dpif->dpif_class->flow_dump_thread_destroy(thread);
704a1e09
BP
1013}
1014
ac64794a
BP
1015/* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0
1016 * if and only if no flows remained to be retrieved, otherwise a positive
1017 * number reflecting the number of elements in 'flows[]' that were updated.
1018 * The number of flows returned might be less than 'max_flows' because
1019 * fewer than 'max_flows' remained, because this particular datapath does not
1020 * benefit from batching, or because an error occurred partway through
1021 * retrieval. Thus, the caller should continue calling until a 0 return value,
1022 * even if intermediate return values are less than 'max_flows'.
bdeadfdd 1023 *
ac64794a
BP
1024 * No error status is immediately provided. An error status for the entire
1025 * dump operation is provided when it is completed by calling
1026 * dpif_flow_dump_destroy().
bdeadfdd 1027 *
ac64794a
BP
1028 * All of the data stored into 'flows' is owned by the datapath, not by the
1029 * caller, and the caller must not modify or free it. The datapath guarantees
1030 * that it remains accessible and unchanged until at least the next call to
1031 * dpif_flow_dump_next() for 'thread'. */
704a1e09 1032int
ac64794a
BP
1033dpif_flow_dump_next(struct dpif_flow_dump_thread *thread,
1034 struct dpif_flow *flows, int max_flows)
704a1e09 1035{
ac64794a
BP
1036 struct dpif *dpif = thread->dpif;
1037 int n;
1038
1039 ovs_assert(max_flows > 0);
1040 n = dpif->dpif_class->flow_dump_next(thread, flows, max_flows);
1041 if (n > 0) {
1042 struct dpif_flow *f;
1043
1044 for (f = flows; f < &flows[n] && should_log_flow_message(0); f++) {
1045 log_flow_message(dpif, 0, "flow_dump",
1046 f->key, f->key_len, f->mask, f->mask_len,
1047 &f->stats, f->actions, f->actions_len);
1048 }
1049 } else {
1050 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif));
1051 }
1052 return n;
064af421
BP
1053}
1054
7fd91025
BP
1055struct dpif_execute_helper_aux {
1056 struct dpif *dpif;
1057 int error;
1058};
1059
09f9da0b
JR
1060/* This is called for actions that need the context of the datapath to be
1061 * meaningful. */
7fd91025 1062static void
758c456d 1063dpif_execute_helper_cb(void *aux_, struct ofpbuf *packet,
572f732a 1064 struct pkt_metadata *md,
09f9da0b 1065 const struct nlattr *action, bool may_steal OVS_UNUSED)
7fd91025 1066{
758c456d
JR
1067 struct dpif_execute_helper_aux *aux = aux_;
1068 struct dpif_execute execute;
09f9da0b 1069 int type = nl_attr_type(action);
758c456d 1070
09f9da0b
JR
1071 switch ((enum ovs_action_attr)type) {
1072 case OVS_ACTION_ATTR_OUTPUT:
1073 case OVS_ACTION_ATTR_USERSPACE:
c6bf49f3 1074 case OVS_ACTION_ATTR_RECIRC:
758c456d
JR
1075 execute.actions = action;
1076 execute.actions_len = NLA_ALIGN(action->nla_len);
1077 execute.packet = packet;
1078 execute.md = *md;
1079 execute.needs_help = false;
1080 aux->error = aux->dpif->dpif_class->execute(aux->dpif, &execute);
09f9da0b 1081 break;
758c456d 1082
c6bf49f3 1083 case OVS_ACTION_ATTR_HASH:
09f9da0b
JR
1084 case OVS_ACTION_ATTR_PUSH_VLAN:
1085 case OVS_ACTION_ATTR_POP_VLAN:
1086 case OVS_ACTION_ATTR_PUSH_MPLS:
1087 case OVS_ACTION_ATTR_POP_MPLS:
1088 case OVS_ACTION_ATTR_SET:
1089 case OVS_ACTION_ATTR_SAMPLE:
1090 case OVS_ACTION_ATTR_UNSPEC:
1091 case __OVS_ACTION_ATTR_MAX:
1092 OVS_NOT_REACHED();
1093 }
7fd91025
BP
1094}
1095
1096/* Executes 'execute' by performing most of the actions in userspace and
1097 * passing the fully constructed packets to 'dpif' for output and userspace
1098 * actions.
1099 *
1100 * This helps with actions that a given 'dpif' doesn't implement directly. */
1101static int
758c456d 1102dpif_execute_with_help(struct dpif *dpif, struct dpif_execute *execute)
7fd91025 1103{
758c456d 1104 struct dpif_execute_helper_aux aux = {dpif, 0};
7fd91025
BP
1105
1106 COVERAGE_INC(dpif_execute_with_help);
1107
df1e5a3b 1108 odp_execute_actions(&aux, execute->packet, false, &execute->md,
7fd91025 1109 execute->actions, execute->actions_len,
09f9da0b 1110 dpif_execute_helper_cb);
7fd91025
BP
1111 return aux.error;
1112}
1113
758c456d
JR
1114/* Causes 'dpif' to perform the 'execute->actions_len' bytes of actions in
1115 * 'execute->actions' on the Ethernet frame in 'execute->packet' and on packet
1116 * metadata in 'execute->md'. The implementation is allowed to modify both the
1117 * '*execute->packet' and 'execute->md'.
1118 *
1119 * Some dpif providers do not implement every action. The Linux kernel
1120 * datapath, in particular, does not implement ARP field modification. If
1121 * 'needs_help' is true, the dpif layer executes in userspace all of the
1122 * actions that it can, and for OVS_ACTION_ATTR_OUTPUT and
1123 * OVS_ACTION_ATTR_USERSPACE actions it passes the packet through to the dpif
1124 * implementation.
1125 *
1126 * This works even if 'execute->actions_len' is too long for a Netlink
1127 * attribute.
1128 *
1129 * Returns 0 if successful, otherwise a positive errno value. */
1130int
1131dpif_execute(struct dpif *dpif, struct dpif_execute *execute)
89625d1e
BP
1132{
1133 int error;
1134
1135 COVERAGE_INC(dpif_execute);
1136 if (execute->actions_len > 0) {
758c456d 1137 error = (execute->needs_help || nl_attr_oversized(execute->actions_len)
7fd91025
BP
1138 ? dpif_execute_with_help(dpif, execute)
1139 : dpif->dpif_class->execute(dpif, execute));
89625d1e
BP
1140 } else {
1141 error = 0;
1142 }
1143
1144 log_execute_message(dpif, execute, error);
1145
1146 return error;
1147}
1148
6bc60024
BP
1149/* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1150 * which they are specified, placing each operation's results in the "output"
1151 * members documented in comments.
1152 *
1153 * This function exists because some datapaths can perform batched operations
1154 * faster than individual operations. */
1155void
c2b565b5 1156dpif_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
6bc60024 1157{
6bc60024 1158 if (dpif->dpif_class->operate) {
7fd91025
BP
1159 while (n_ops > 0) {
1160 size_t chunk;
1161
1162 /* Count 'chunk', the number of ops that can be executed without
1163 * needing any help. Ops that need help should be rare, so we
1164 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1165 for (chunk = 0; chunk < n_ops; chunk++) {
1166 struct dpif_op *op = ops[chunk];
1167
1168 if (op->type == DPIF_OP_EXECUTE && op->u.execute.needs_help) {
1169 break;
1170 }
1171 }
1172
1173 if (chunk) {
1174 /* Execute a chunk full of ops that the dpif provider can
1175 * handle itself, without help. */
1176 size_t i;
1177
1178 dpif->dpif_class->operate(dpif, ops, chunk);
1179
1180 for (i = 0; i < chunk; i++) {
1181 struct dpif_op *op = ops[i];
1182
1183 switch (op->type) {
1184 case DPIF_OP_FLOW_PUT:
1185 log_flow_put_message(dpif, &op->u.flow_put, op->error);
1186 break;
1187
1188 case DPIF_OP_FLOW_DEL:
1189 log_flow_del_message(dpif, &op->u.flow_del, op->error);
1190 break;
1191
1192 case DPIF_OP_EXECUTE:
1193 log_execute_message(dpif, &op->u.execute, op->error);
1194 break;
1195 }
1196 }
1197
1198 ops += chunk;
1199 n_ops -= chunk;
1200 } else {
1201 /* Help the dpif provider to execute one op. */
1202 struct dpif_op *op = ops[0];
1203
758c456d 1204 op->error = dpif_execute(dpif, &op->u.execute);
7fd91025
BP
1205 ops++;
1206 n_ops--;
1207 }
1208 }
1209 } else {
1210 size_t i;
f23d2845
BP
1211
1212 for (i = 0; i < n_ops; i++) {
1213 struct dpif_op *op = ops[i];
1214
1215 switch (op->type) {
1216 case DPIF_OP_FLOW_PUT:
7fd91025 1217 op->error = dpif_flow_put__(dpif, &op->u.flow_put);
f23d2845
BP
1218 break;
1219
b99d3cee 1220 case DPIF_OP_FLOW_DEL:
7fd91025 1221 op->error = dpif_flow_del__(dpif, &op->u.flow_del);
b99d3cee
BP
1222 break;
1223
f23d2845 1224 case DPIF_OP_EXECUTE:
758c456d 1225 op->error = dpif_execute(dpif, &op->u.execute);
f23d2845 1226 break;
b99d3cee 1227
7fd91025 1228 default:
428b2edd 1229 OVS_NOT_REACHED();
7fd91025 1230 }
6bc60024
BP
1231 }
1232 }
1233}
1234
01545c1a
BP
1235/* Returns a string that represents 'type', for use in log messages. */
1236const char *
1237dpif_upcall_type_to_string(enum dpif_upcall_type type)
1238{
1239 switch (type) {
1240 case DPIF_UC_MISS: return "miss";
1241 case DPIF_UC_ACTION: return "action";
01545c1a
BP
1242 case DPIF_N_UC_TYPES: default: return "<unknown>";
1243 }
1244}
1245
a12b3ead
BP
1246/* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1247 * if successful, otherwise a positive errno value.
98403001 1248 *
a12b3ead 1249 * Turning packet receive off and then back on may change the Netlink PID
98403001
BP
1250 * assignments returned by dpif_port_get_pid(). If the client does this, it
1251 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1252 * using the new PID assignment. */
8f24562a 1253int
a12b3ead 1254dpif_recv_set(struct dpif *dpif, bool enable)
8f24562a 1255{
a12b3ead
BP
1256 int error = dpif->dpif_class->recv_set(dpif, enable);
1257 log_operation(dpif, "recv_set", error);
96fba48f 1258 return error;
8f24562a
BP
1259}
1260
1954e6bb
AW
1261/* Refreshes the poll loops and Netlink sockets associated to each port,
1262 * when the number of upcall handlers (upcall receiving thread) is changed
1263 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
1264 * recv_set().
1265 *
1266 * Since multiple upcall handlers can read upcalls simultaneously from
1267 * 'dpif', each port can have multiple Netlink sockets, one per upcall
1268 * handler. So, handlers_set() is responsible for the following tasks:
1269 *
1270 * When receiving upcall is enabled, extends or creates the
1271 * configuration to support:
1272 *
1273 * - 'n_handlers' Netlink sockets for each port.
1274 *
1275 * - 'n_handlers' poll loops, one for each upcall handler.
1276 *
1277 * - registering the Netlink sockets for the same upcall handler to
1278 * the corresponding poll loop.
1279 *
1280 * Returns 0 if successful, otherwise a positive errno value. */
1281int
1282dpif_handlers_set(struct dpif *dpif, uint32_t n_handlers)
1283{
1284 int error = dpif->dpif_class->handlers_set(dpif, n_handlers);
1285 log_operation(dpif, "handlers_set", error);
1286 return error;
1287}
1288
1289/* Polls for an upcall from 'dpif' for an upcall handler. Since there
1290 * there can be multiple poll loops, 'handler_id' is needed as index to
1291 * identify the corresponding poll loop. If successful, stores the upcall
1292 * into '*upcall', using 'buf' for storage. Should only be called if
1293 * 'recv_set' has been used to enable receiving packets from 'dpif'.
96fba48f 1294 *
da546e07
JR
1295 * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
1296 * 'buf', so their memory cannot be freed separately from 'buf'.
856081f6 1297 *
837a88dc
JR
1298 * The caller owns the data of 'upcall->packet' and may modify it. If
1299 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
1300 * will be reallocated. This requires the data of 'upcall->packet' to be
1301 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
1302 * when an error is returned, the 'upcall->packet' may be uninitialized
1303 * and should not be released.
1304 *
96fba48f 1305 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
856081f6 1306 * if no upcall is immediately available. */
064af421 1307int
1954e6bb
AW
1308dpif_recv(struct dpif *dpif, uint32_t handler_id, struct dpif_upcall *upcall,
1309 struct ofpbuf *buf)
064af421 1310{
1954e6bb 1311 int error = dpif->dpif_class->recv(dpif, handler_id, upcall, buf);
856081f6 1312 if (!error && !VLOG_DROP_DBG(&dpmsg_rl)) {
01545c1a
BP
1313 struct ds flow;
1314 char *packet;
1315
1f317cb5
PS
1316 packet = ofp_packet_to_string(ofpbuf_data(&upcall->packet),
1317 ofpbuf_size(&upcall->packet));
01545c1a
BP
1318
1319 ds_init(&flow);
1320 odp_flow_key_format(upcall->key, upcall->key_len, &flow);
1321
1322 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1323 dpif_name(dpif), dpif_upcall_type_to_string(upcall->type),
1324 ds_cstr(&flow), packet);
1325
1326 ds_destroy(&flow);
1327 free(packet);
5fcc0d00
BP
1328 } else if (error && error != EAGAIN) {
1329 log_operation(dpif, "recv", error);
064af421 1330 }
064af421
BP
1331 return error;
1332}
1333
96fba48f 1334/* Discards all messages that would otherwise be received by dpif_recv() on
1ba530f4
BP
1335 * 'dpif'. */
1336void
96fba48f
BP
1337dpif_recv_purge(struct dpif *dpif)
1338{
96fba48f 1339 COVERAGE_INC(dpif_purge);
1ba530f4
BP
1340 if (dpif->dpif_class->recv_purge) {
1341 dpif->dpif_class->recv_purge(dpif);
96fba48f 1342 }
96fba48f
BP
1343}
1344
1954e6bb
AW
1345/* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
1346 * 'dpif' has a message queued to be received with the recv member
1347 * function. Since there can be multiple poll loops, 'handler_id' is
1348 * needed as index to identify the corresponding poll loop. */
064af421 1349void
1954e6bb 1350dpif_recv_wait(struct dpif *dpif, uint32_t handler_id)
064af421 1351{
1954e6bb 1352 dpif->dpif_class->recv_wait(dpif, handler_id);
064af421 1353}
53a4218d 1354
96fba48f
BP
1355/* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1356 * and '*engine_id', respectively. */
53a4218d
BP
1357void
1358dpif_get_netflow_ids(const struct dpif *dpif,
1359 uint8_t *engine_type, uint8_t *engine_id)
1360{
96fba48f
BP
1361 *engine_type = dpif->netflow_engine_type;
1362 *engine_id = dpif->netflow_engine_id;
1363}
aae51f53
BP
1364
1365/* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
abff858b
PS
1366 * value used for setting packet priority.
1367 * On success, returns 0 and stores the priority into '*priority'.
1368 * On failure, returns a positive errno value and stores 0 into '*priority'. */
aae51f53
BP
1369int
1370dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1371 uint32_t *priority)
1372{
1373 int error = (dpif->dpif_class->queue_to_priority
1374 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1375 priority)
1376 : EOPNOTSUPP);
1377 if (error) {
1378 *priority = 0;
1379 }
1380 log_operation(dpif, "queue_to_priority", error);
1381 return error;
1382}
96fba48f
BP
1383\f
1384void
1acb6baa
BP
1385dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1386 const char *name,
96fba48f
BP
1387 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1388{
1acb6baa 1389 dpif->dpif_class = dpif_class;
1a6f1e2a 1390 dpif->base_name = xstrdup(name);
a4af0040 1391 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
96fba48f
BP
1392 dpif->netflow_engine_type = netflow_engine_type;
1393 dpif->netflow_engine_id = netflow_engine_id;
1394}
999401aa
JG
1395
1396/* Undoes the results of initialization.
1397 *
1398 * Normally this function only needs to be called from dpif_close().
1399 * However, it may be called by providers due to an error on opening
1400 * that occurs after initialization. It this case dpif_close() would
1401 * never be called. */
1402void
1403dpif_uninit(struct dpif *dpif, bool close)
1404{
1405 char *base_name = dpif->base_name;
1406 char *full_name = dpif->full_name;
1407
1408 if (close) {
a4af0040 1409 dpif->dpif_class->close(dpif);
999401aa
JG
1410 }
1411
1412 free(base_name);
1413 free(full_name);
1414}
96fba48f
BP
1415\f
1416static void
1417log_operation(const struct dpif *dpif, const char *operation, int error)
1418{
1419 if (!error) {
1420 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
90bf1e07 1421 } else if (ofperr_is_valid(error)) {
96fba48f 1422 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
90bf1e07 1423 dpif_name(dpif), operation, ofperr_get_name(error));
71ce9235 1424 } else {
90bf1e07 1425 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
10a89ef0 1426 dpif_name(dpif), operation, ovs_strerror(error));
96fba48f
BP
1427 }
1428}
1429
1430static enum vlog_level
1431flow_message_log_level(int error)
1432{
9b1a48c2
JP
1433 /* If flows arrive in a batch, userspace may push down multiple
1434 * unique flow definitions that overlap when wildcards are applied.
1435 * Kernels that support flow wildcarding will reject these flows as
1436 * duplicates (EEXIST), so lower the log level to debug for these
1437 * types of messages. */
1438 return (error && error != EEXIST) ? VLL_WARN : VLL_DBG;
96fba48f
BP
1439}
1440
1441static bool
1442should_log_flow_message(int error)
1443{
1444 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1445 error ? &error_rl : &dpmsg_rl);
1446}
1447
1448static void
1449log_flow_message(const struct dpif *dpif, int error, const char *operation,
36956a7d 1450 const struct nlattr *key, size_t key_len,
61fb711d 1451 const struct nlattr *mask, size_t mask_len,
c97fb132 1452 const struct dpif_flow_stats *stats,
cf22f8cb 1453 const struct nlattr *actions, size_t actions_len)
96fba48f
BP
1454{
1455 struct ds ds = DS_EMPTY_INITIALIZER;
1456 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1457 if (error) {
1458 ds_put_cstr(&ds, "failed to ");
1459 }
1460 ds_put_format(&ds, "%s ", operation);
1461 if (error) {
10a89ef0 1462 ds_put_format(&ds, "(%s) ", ovs_strerror(error));
96fba48f 1463 }
0a37839c 1464 odp_flow_format(key, key_len, mask, mask_len, NULL, &ds, true);
96fba48f
BP
1465 if (stats) {
1466 ds_put_cstr(&ds, ", ");
c97fb132 1467 dpif_flow_stats_format(stats, &ds);
96fba48f 1468 }
cdee00fd 1469 if (actions || actions_len) {
96fba48f 1470 ds_put_cstr(&ds, ", actions:");
cdee00fd 1471 format_odp_actions(&ds, actions, actions_len);
96fba48f
BP
1472 }
1473 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1474 ds_destroy(&ds);
1475}
89625d1e
BP
1476
1477static void
1478log_flow_put_message(struct dpif *dpif, const struct dpif_flow_put *put,
1479 int error)
1480{
1481 if (should_log_flow_message(error)) {
1482 struct ds s;
1483
1484 ds_init(&s);
1485 ds_put_cstr(&s, "put");
1486 if (put->flags & DPIF_FP_CREATE) {
1487 ds_put_cstr(&s, "[create]");
1488 }
1489 if (put->flags & DPIF_FP_MODIFY) {
1490 ds_put_cstr(&s, "[modify]");
1491 }
1492 if (put->flags & DPIF_FP_ZERO_STATS) {
1493 ds_put_cstr(&s, "[zero]");
1494 }
1495 log_flow_message(dpif, error, ds_cstr(&s),
61fb711d
JP
1496 put->key, put->key_len, put->mask, put->mask_len,
1497 put->stats, put->actions, put->actions_len);
89625d1e
BP
1498 ds_destroy(&s);
1499 }
1500}
1501
b99d3cee
BP
1502static void
1503log_flow_del_message(struct dpif *dpif, const struct dpif_flow_del *del,
1504 int error)
1505{
1506 if (should_log_flow_message(error)) {
1507 log_flow_message(dpif, error, "flow_del", del->key, del->key_len,
61fb711d 1508 NULL, 0, !error ? del->stats : NULL, NULL, 0);
b99d3cee
BP
1509 }
1510}
1511
89625d1e
BP
1512static void
1513log_execute_message(struct dpif *dpif, const struct dpif_execute *execute,
1514 int error)
1515{
1516 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
1517 struct ds ds = DS_EMPTY_INITIALIZER;
1518 char *packet;
1519
1f317cb5
PS
1520 packet = ofp_packet_to_string(ofpbuf_data(execute->packet),
1521 ofpbuf_size(execute->packet));
89625d1e
BP
1522 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
1523 format_odp_actions(&ds, execute->actions, execute->actions_len);
1524 if (error) {
10a89ef0 1525 ds_put_format(&ds, " failed (%s)", ovs_strerror(error));
89625d1e
BP
1526 }
1527 ds_put_format(&ds, " on packet %s", packet);
1528 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
1529 ds_destroy(&ds);
1530 free(packet);
1531 }
1532}