2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014, 2015, 2016 Nicira, Inc.
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:
8 * http://www.apache.org/licenses/LICENSE-2.0
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.
18 #include "dpif-provider.h"
28 #include "dp-packet.h"
29 #include "dpif-netdev.h"
30 #include "openvswitch/dynamic-string.h"
34 #include "odp-execute.h"
36 #include "openvswitch/ofp-print.h"
37 #include "openvswitch/ofp-util.h"
38 #include "openvswitch/ofpbuf.h"
40 #include "poll-loop.h"
41 #include "route-table.h"
43 #include "openvswitch/shash.h"
46 #include "tnl-neigh-cache.h"
47 #include "tnl-ports.h"
51 #include "openvswitch/ofp-errors.h"
52 #include "openvswitch/vlog.h"
54 VLOG_DEFINE_THIS_MODULE(dpif
);
56 COVERAGE_DEFINE(dpif_destroy
);
57 COVERAGE_DEFINE(dpif_port_add
);
58 COVERAGE_DEFINE(dpif_port_del
);
59 COVERAGE_DEFINE(dpif_flow_flush
);
60 COVERAGE_DEFINE(dpif_flow_get
);
61 COVERAGE_DEFINE(dpif_flow_put
);
62 COVERAGE_DEFINE(dpif_flow_del
);
63 COVERAGE_DEFINE(dpif_execute
);
64 COVERAGE_DEFINE(dpif_purge
);
65 COVERAGE_DEFINE(dpif_execute_with_help
);
67 static const struct dpif_class
*base_dpif_classes
[] = {
68 #if defined(__linux__) || defined(_WIN32)
74 struct registered_dpif_class
{
75 const struct dpif_class
*dpif_class
;
78 static struct shash dpif_classes
= SHASH_INITIALIZER(&dpif_classes
);
79 static struct sset dpif_blacklist
= SSET_INITIALIZER(&dpif_blacklist
);
81 /* Protects 'dpif_classes', including the refcount, and 'dpif_blacklist'. */
82 static struct ovs_mutex dpif_mutex
= OVS_MUTEX_INITIALIZER
;
84 /* Rate limit for individual messages going to or from the datapath, output at
85 * DBG level. This is very high because, if these are enabled, it is because
86 * we really need to see them. */
87 static struct vlog_rate_limit dpmsg_rl
= VLOG_RATE_LIMIT_INIT(600, 600);
89 /* Not really much point in logging many dpif errors. */
90 static struct vlog_rate_limit error_rl
= VLOG_RATE_LIMIT_INIT(60, 5);
92 static void log_flow_message(const struct dpif
*dpif
, int error
,
93 const char *operation
,
94 const struct nlattr
*key
, size_t key_len
,
95 const struct nlattr
*mask
, size_t mask_len
,
97 const struct dpif_flow_stats
*stats
,
98 const struct nlattr
*actions
, size_t actions_len
);
99 static void log_operation(const struct dpif
*, const char *operation
,
101 static bool should_log_flow_message(int error
);
102 static void log_flow_put_message(struct dpif
*, const struct dpif_flow_put
*,
104 static void log_flow_del_message(struct dpif
*, const struct dpif_flow_del
*,
106 static void log_execute_message(struct dpif
*, const struct dpif_execute
*,
107 bool subexecute
, int error
);
108 static void log_flow_get_message(const struct dpif
*,
109 const struct dpif_flow_get
*, int error
);
111 /* Incremented whenever tnl route, arp, etc changes. */
112 struct seq
*tnl_conf_seq
;
117 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
119 if (ovsthread_once_start(&once
)) {
122 tnl_conf_seq
= seq_create();
123 dpctl_unixctl_register();
125 tnl_neigh_cache_init();
128 for (i
= 0; i
< ARRAY_SIZE(base_dpif_classes
); i
++) {
129 dp_register_provider(base_dpif_classes
[i
]);
132 ovsthread_once_done(&once
);
137 dp_register_provider__(const struct dpif_class
*new_class
)
139 struct registered_dpif_class
*registered_class
;
142 if (sset_contains(&dpif_blacklist
, new_class
->type
)) {
143 VLOG_DBG("attempted to register blacklisted provider: %s",
148 if (shash_find(&dpif_classes
, new_class
->type
)) {
149 VLOG_WARN("attempted to register duplicate datapath provider: %s",
154 error
= new_class
->init
? new_class
->init() : 0;
156 VLOG_WARN("failed to initialize %s datapath class: %s",
157 new_class
->type
, ovs_strerror(error
));
161 registered_class
= xmalloc(sizeof *registered_class
);
162 registered_class
->dpif_class
= new_class
;
163 registered_class
->refcount
= 0;
165 shash_add(&dpif_classes
, new_class
->type
, registered_class
);
170 /* Registers a new datapath provider. After successful registration, new
171 * datapaths of that type can be opened using dpif_open(). */
173 dp_register_provider(const struct dpif_class
*new_class
)
177 ovs_mutex_lock(&dpif_mutex
);
178 error
= dp_register_provider__(new_class
);
179 ovs_mutex_unlock(&dpif_mutex
);
184 /* Unregisters a datapath provider. 'type' must have been previously
185 * registered and not currently be in use by any dpifs. After unregistration
186 * new datapaths of that type cannot be opened using dpif_open(). */
188 dp_unregister_provider__(const char *type
)
190 struct shash_node
*node
;
191 struct registered_dpif_class
*registered_class
;
193 node
= shash_find(&dpif_classes
, type
);
198 registered_class
= node
->data
;
199 if (registered_class
->refcount
) {
200 VLOG_WARN("attempted to unregister in use datapath provider: %s", type
);
204 shash_delete(&dpif_classes
, node
);
205 free(registered_class
);
210 /* Unregisters a datapath provider. 'type' must have been previously
211 * registered and not currently be in use by any dpifs. After unregistration
212 * new datapaths of that type cannot be opened using dpif_open(). */
214 dp_unregister_provider(const char *type
)
220 ovs_mutex_lock(&dpif_mutex
);
221 error
= dp_unregister_provider__(type
);
222 ovs_mutex_unlock(&dpif_mutex
);
227 /* Blacklists a provider. Causes future calls of dp_register_provider() with
228 * a dpif_class which implements 'type' to fail. */
230 dp_blacklist_provider(const char *type
)
232 ovs_mutex_lock(&dpif_mutex
);
233 sset_add(&dpif_blacklist
, type
);
234 ovs_mutex_unlock(&dpif_mutex
);
237 /* Adds the types of all currently registered datapath providers to 'types'.
238 * The caller must first initialize the sset. */
240 dp_enumerate_types(struct sset
*types
)
242 struct shash_node
*node
;
246 ovs_mutex_lock(&dpif_mutex
);
247 SHASH_FOR_EACH(node
, &dpif_classes
) {
248 const struct registered_dpif_class
*registered_class
= node
->data
;
249 sset_add(types
, registered_class
->dpif_class
->type
);
251 ovs_mutex_unlock(&dpif_mutex
);
255 dp_class_unref(struct registered_dpif_class
*rc
)
257 ovs_mutex_lock(&dpif_mutex
);
258 ovs_assert(rc
->refcount
);
260 ovs_mutex_unlock(&dpif_mutex
);
263 static struct registered_dpif_class
*
264 dp_class_lookup(const char *type
)
266 struct registered_dpif_class
*rc
;
268 ovs_mutex_lock(&dpif_mutex
);
269 rc
= shash_find_data(&dpif_classes
, type
);
273 ovs_mutex_unlock(&dpif_mutex
);
278 /* Clears 'names' and enumerates the names of all known created datapaths with
279 * the given 'type'. The caller must first initialize the sset. Returns 0 if
280 * successful, otherwise a positive errno value.
282 * Some kinds of datapaths might not be practically enumerable. This is not
283 * considered an error. */
285 dp_enumerate_names(const char *type
, struct sset
*names
)
287 struct registered_dpif_class
*registered_class
;
288 const struct dpif_class
*dpif_class
;
294 registered_class
= dp_class_lookup(type
);
295 if (!registered_class
) {
296 VLOG_WARN("could not enumerate unknown type: %s", type
);
300 dpif_class
= registered_class
->dpif_class
;
301 error
= (dpif_class
->enumerate
302 ? dpif_class
->enumerate(names
, dpif_class
)
305 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class
->type
,
306 ovs_strerror(error
));
308 dp_class_unref(registered_class
);
313 /* Parses 'datapath_name_', which is of the form [type@]name into its
314 * component pieces. 'name' and 'type' must be freed by the caller.
316 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
318 dp_parse_name(const char *datapath_name_
, char **name
, char **type
)
320 char *datapath_name
= xstrdup(datapath_name_
);
323 separator
= strchr(datapath_name
, '@');
326 *type
= datapath_name
;
327 *name
= xstrdup(dpif_normalize_type(separator
+ 1));
329 *name
= datapath_name
;
330 *type
= xstrdup(dpif_normalize_type(NULL
));
335 do_open(const char *name
, const char *type
, bool create
, struct dpif
**dpifp
)
337 struct dpif
*dpif
= NULL
;
339 struct registered_dpif_class
*registered_class
;
343 type
= dpif_normalize_type(type
);
344 registered_class
= dp_class_lookup(type
);
345 if (!registered_class
) {
346 VLOG_WARN("could not create datapath %s of unknown type %s", name
,
348 error
= EAFNOSUPPORT
;
352 error
= registered_class
->dpif_class
->open(registered_class
->dpif_class
,
353 name
, create
, &dpif
);
355 ovs_assert(dpif
->dpif_class
== registered_class
->dpif_class
);
357 dp_class_unref(registered_class
);
361 *dpifp
= error
? NULL
: dpif
;
365 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
366 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
367 * the empty string to specify the default system type. Returns 0 if
368 * successful, otherwise a positive errno value. On success stores a pointer
369 * to the datapath in '*dpifp', otherwise a null pointer. */
371 dpif_open(const char *name
, const char *type
, struct dpif
**dpifp
)
373 return do_open(name
, type
, false, dpifp
);
376 /* Tries to create and open a new datapath with the given 'name' and 'type'.
377 * 'type' may be either NULL or the empty string to specify the default system
378 * type. Will fail if a datapath with 'name' and 'type' already exists.
379 * Returns 0 if successful, otherwise a positive errno value. On success
380 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
382 dpif_create(const char *name
, const char *type
, struct dpif
**dpifp
)
384 return do_open(name
, type
, true, dpifp
);
387 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
388 * does not exist. 'type' may be either NULL or the empty string to specify
389 * the default system type. Returns 0 if successful, otherwise a positive
390 * errno value. On success stores a pointer to the datapath in '*dpifp',
391 * otherwise a null pointer. */
393 dpif_create_and_open(const char *name
, const char *type
, struct dpif
**dpifp
)
397 error
= dpif_create(name
, type
, dpifp
);
398 if (error
== EEXIST
|| error
== EBUSY
) {
399 error
= dpif_open(name
, type
, dpifp
);
401 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
402 name
, ovs_strerror(error
));
405 VLOG_WARN("failed to create datapath %s: %s",
406 name
, ovs_strerror(error
));
411 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
412 * itself; call dpif_delete() first, instead, if that is desirable. */
414 dpif_close(struct dpif
*dpif
)
417 struct registered_dpif_class
*rc
;
419 rc
= shash_find_data(&dpif_classes
, dpif
->dpif_class
->type
);
420 dpif_uninit(dpif
, true);
425 /* Performs periodic work needed by 'dpif'. */
427 dpif_run(struct dpif
*dpif
)
429 if (dpif
->dpif_class
->run
) {
430 return dpif
->dpif_class
->run(dpif
);
435 /* Arranges for poll_block() to wake up when dp_run() needs to be called for
438 dpif_wait(struct dpif
*dpif
)
440 if (dpif
->dpif_class
->wait
) {
441 dpif
->dpif_class
->wait(dpif
);
445 /* Returns the name of datapath 'dpif' prefixed with the type
446 * (for use in log messages). */
448 dpif_name(const struct dpif
*dpif
)
450 return dpif
->full_name
;
453 /* Returns the name of datapath 'dpif' without the type
454 * (for use in device names). */
456 dpif_base_name(const struct dpif
*dpif
)
458 return dpif
->base_name
;
461 /* Returns the type of datapath 'dpif'. */
463 dpif_type(const struct dpif
*dpif
)
465 return dpif
->dpif_class
->type
;
468 /* Returns the fully spelled out name for the given datapath 'type'.
470 * Normalized type string can be compared with strcmp(). Unnormalized type
471 * string might be the same even if they have different spellings. */
473 dpif_normalize_type(const char *type
)
475 return type
&& type
[0] ? type
: "system";
478 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
479 * ports. After calling this function, it does not make sense to pass 'dpif'
480 * to any functions other than dpif_name() or dpif_close(). */
482 dpif_delete(struct dpif
*dpif
)
486 COVERAGE_INC(dpif_destroy
);
488 error
= dpif
->dpif_class
->destroy(dpif
);
489 log_operation(dpif
, "delete", error
);
493 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
494 * otherwise a positive errno value. */
496 dpif_get_dp_stats(const struct dpif
*dpif
, struct dpif_dp_stats
*stats
)
498 int error
= dpif
->dpif_class
->get_stats(dpif
, stats
);
500 memset(stats
, 0, sizeof *stats
);
502 log_operation(dpif
, "get_stats", error
);
507 dpif_port_open_type(const char *datapath_type
, const char *port_type
)
509 struct registered_dpif_class
*rc
;
511 datapath_type
= dpif_normalize_type(datapath_type
);
513 ovs_mutex_lock(&dpif_mutex
);
514 rc
= shash_find_data(&dpif_classes
, datapath_type
);
515 if (rc
&& rc
->dpif_class
->port_open_type
) {
516 port_type
= rc
->dpif_class
->port_open_type(rc
->dpif_class
, port_type
);
518 ovs_mutex_unlock(&dpif_mutex
);
523 /* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
524 * non-null and its value is not ODPP_NONE, then attempts to use the
525 * value as the port number.
527 * If successful, returns 0 and sets '*port_nop' to the new port's port
528 * number (if 'port_nop' is non-null). On failure, returns a positive
529 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
532 dpif_port_add(struct dpif
*dpif
, struct netdev
*netdev
, odp_port_t
*port_nop
)
534 const char *netdev_name
= netdev_get_name(netdev
);
535 odp_port_t port_no
= ODPP_NONE
;
538 COVERAGE_INC(dpif_port_add
);
544 error
= dpif
->dpif_class
->port_add(dpif
, netdev
, &port_no
);
546 VLOG_DBG_RL(&dpmsg_rl
, "%s: added %s as port %"PRIu32
,
547 dpif_name(dpif
), netdev_name
, port_no
);
549 VLOG_WARN_RL(&error_rl
, "%s: failed to add %s as port: %s",
550 dpif_name(dpif
), netdev_name
, ovs_strerror(error
));
559 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
560 * otherwise a positive errno value. */
562 dpif_port_del(struct dpif
*dpif
, odp_port_t port_no
)
566 COVERAGE_INC(dpif_port_del
);
568 error
= dpif
->dpif_class
->port_del(dpif
, port_no
);
570 VLOG_DBG_RL(&dpmsg_rl
, "%s: port_del(%"PRIu32
")",
571 dpif_name(dpif
), port_no
);
573 log_operation(dpif
, "port_del", error
);
578 /* Makes a deep copy of 'src' into 'dst'. */
580 dpif_port_clone(struct dpif_port
*dst
, const struct dpif_port
*src
)
582 dst
->name
= xstrdup(src
->name
);
583 dst
->type
= xstrdup(src
->type
);
584 dst
->port_no
= src
->port_no
;
587 /* Frees memory allocated to members of 'dpif_port'.
589 * Do not call this function on a dpif_port obtained from
590 * dpif_port_dump_next(): that function retains ownership of the data in the
593 dpif_port_destroy(struct dpif_port
*dpif_port
)
595 free(dpif_port
->name
);
596 free(dpif_port
->type
);
599 /* Checks if port named 'devname' exists in 'dpif'. If so, returns
600 * true; otherwise, returns false. */
602 dpif_port_exists(const struct dpif
*dpif
, const char *devname
)
604 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, NULL
);
605 if (error
!= 0 && error
!= ENODEV
) {
606 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %s: %s",
607 dpif_name(dpif
), devname
, ovs_strerror(error
));
613 /* Refreshes configuration of 'dpif's port. */
615 dpif_port_set_config(struct dpif
*dpif
, odp_port_t port_no
,
616 const struct smap
*cfg
)
620 if (dpif
->dpif_class
->port_set_config
) {
621 error
= dpif
->dpif_class
->port_set_config(dpif
, port_no
, cfg
);
623 log_operation(dpif
, "port_set_config", error
);
630 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
631 * initializes '*port' appropriately; on failure, returns a positive errno
634 * Retuns ENODEV if the port doesn't exist.
636 * The caller owns the data in 'port' and must free it with
637 * dpif_port_destroy() when it is no longer needed. */
639 dpif_port_query_by_number(const struct dpif
*dpif
, odp_port_t port_no
,
640 struct dpif_port
*port
)
642 int error
= dpif
->dpif_class
->port_query_by_number(dpif
, port_no
, port
);
644 VLOG_DBG_RL(&dpmsg_rl
, "%s: port %"PRIu32
" is device %s",
645 dpif_name(dpif
), port_no
, port
->name
);
647 memset(port
, 0, sizeof *port
);
648 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %"PRIu32
": %s",
649 dpif_name(dpif
), port_no
, ovs_strerror(error
));
654 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
655 * initializes '*port' appropriately; on failure, returns a positive errno
658 * Retuns ENODEV if the port doesn't exist.
660 * The caller owns the data in 'port' and must free it with
661 * dpif_port_destroy() when it is no longer needed. */
663 dpif_port_query_by_name(const struct dpif
*dpif
, const char *devname
,
664 struct dpif_port
*port
)
666 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, port
);
668 VLOG_DBG_RL(&dpmsg_rl
, "%s: device %s is on port %"PRIu32
,
669 dpif_name(dpif
), devname
, port
->port_no
);
671 memset(port
, 0, sizeof *port
);
673 /* For ENODEV we use DBG level because the caller is probably
674 * interested in whether 'dpif' actually has a port 'devname', so that
675 * it's not an issue worth logging if it doesn't. Other errors are
676 * uncommon and more likely to indicate a real problem. */
677 VLOG_RL(&error_rl
, error
== ENODEV
? VLL_DBG
: VLL_WARN
,
678 "%s: failed to query port %s: %s",
679 dpif_name(dpif
), devname
, ovs_strerror(error
));
684 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
685 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
686 * flows whose packets arrived on port 'port_no'. In the case where the
687 * provider allocates multiple Netlink PIDs to a single port, it may use
688 * 'hash' to spread load among them. The caller need not use a particular
689 * hash function; a 5-tuple hash is suitable.
691 * (The datapath implementation might use some different hash function for
692 * distributing packets received via flow misses among PIDs. This means
693 * that packets received via flow misses might be reordered relative to
694 * packets received via userspace actions. This is not ordinarily a
697 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
698 * allocated to any port, that the client may use for special purposes.
700 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
701 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
702 * disabled and then re-enabled, so a client that does that must be prepared to
703 * update all of the flows that it installed that contain
704 * OVS_ACTION_ATTR_USERSPACE actions. */
706 dpif_port_get_pid(const struct dpif
*dpif
, odp_port_t port_no
, uint32_t hash
)
708 return (dpif
->dpif_class
->port_get_pid
709 ? (dpif
->dpif_class
->port_get_pid
)(dpif
, port_no
, hash
)
713 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
714 * the port's name into the 'name_size' bytes in 'name', ensuring that the
715 * result is null-terminated. On failure, returns a positive errno value and
716 * makes 'name' the empty string. */
718 dpif_port_get_name(struct dpif
*dpif
, odp_port_t port_no
,
719 char *name
, size_t name_size
)
721 struct dpif_port port
;
724 ovs_assert(name_size
> 0);
726 error
= dpif_port_query_by_number(dpif
, port_no
, &port
);
728 ovs_strlcpy(name
, port
.name
, name_size
);
729 dpif_port_destroy(&port
);
736 /* Initializes 'dump' to begin dumping the ports in a dpif.
738 * This function provides no status indication. An error status for the entire
739 * dump operation is provided when it is completed by calling
740 * dpif_port_dump_done().
743 dpif_port_dump_start(struct dpif_port_dump
*dump
, const struct dpif
*dpif
)
746 dump
->error
= dpif
->dpif_class
->port_dump_start(dpif
, &dump
->state
);
747 log_operation(dpif
, "port_dump_start", dump
->error
);
750 /* Attempts to retrieve another port from 'dump', which must have been
751 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
752 * into 'port' and returns true. On failure, returns false.
754 * Failure might indicate an actual error or merely that the last port has been
755 * dumped. An error status for the entire dump operation is provided when it
756 * is completed by calling dpif_port_dump_done().
758 * The dpif owns the data stored in 'port'. It will remain valid until at
759 * least the next time 'dump' is passed to dpif_port_dump_next() or
760 * dpif_port_dump_done(). */
762 dpif_port_dump_next(struct dpif_port_dump
*dump
, struct dpif_port
*port
)
764 const struct dpif
*dpif
= dump
->dpif
;
770 dump
->error
= dpif
->dpif_class
->port_dump_next(dpif
, dump
->state
, port
);
771 if (dump
->error
== EOF
) {
772 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all ports", dpif_name(dpif
));
774 log_operation(dpif
, "port_dump_next", dump
->error
);
778 dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
784 /* Completes port table dump operation 'dump', which must have been initialized
785 * with dpif_port_dump_start(). Returns 0 if the dump operation was
786 * error-free, otherwise a positive errno value describing the problem. */
788 dpif_port_dump_done(struct dpif_port_dump
*dump
)
790 const struct dpif
*dpif
= dump
->dpif
;
792 dump
->error
= dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
793 log_operation(dpif
, "port_dump_done", dump
->error
);
795 return dump
->error
== EOF
? 0 : dump
->error
;
798 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
799 * 'dpif' has changed, this function does one of the following:
801 * - Stores the name of the device that was added to or deleted from 'dpif' in
802 * '*devnamep' and returns 0. The caller is responsible for freeing
803 * '*devnamep' (with free()) when it no longer needs it.
805 * - Returns ENOBUFS and sets '*devnamep' to NULL.
807 * This function may also return 'false positives', where it returns 0 and
808 * '*devnamep' names a device that was not actually added or deleted or it
809 * returns ENOBUFS without any change.
811 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
812 * return other positive errno values to indicate that something has gone
815 dpif_port_poll(const struct dpif
*dpif
, char **devnamep
)
817 int error
= dpif
->dpif_class
->port_poll(dpif
, devnamep
);
824 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
825 * value other than EAGAIN. */
827 dpif_port_poll_wait(const struct dpif
*dpif
)
829 dpif
->dpif_class
->port_poll_wait(dpif
);
832 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
833 * arguments must have been initialized through a call to flow_extract().
834 * 'used' is stored into stats->used. */
836 dpif_flow_stats_extract(const struct flow
*flow
, const struct dp_packet
*packet
,
837 long long int used
, struct dpif_flow_stats
*stats
)
839 stats
->tcp_flags
= ntohs(flow
->tcp_flags
);
840 stats
->n_bytes
= dp_packet_size(packet
);
841 stats
->n_packets
= 1;
845 /* Appends a human-readable representation of 'stats' to 's'. */
847 dpif_flow_stats_format(const struct dpif_flow_stats
*stats
, struct ds
*s
)
849 ds_put_format(s
, "packets:%"PRIu64
", bytes:%"PRIu64
", used:",
850 stats
->n_packets
, stats
->n_bytes
);
852 ds_put_format(s
, "%.3fs", (time_msec() - stats
->used
) / 1000.0);
854 ds_put_format(s
, "never");
856 if (stats
->tcp_flags
) {
857 ds_put_cstr(s
, ", flags:");
858 packet_format_tcp_flags(s
, stats
->tcp_flags
);
862 /* Places the hash of the 'key_len' bytes starting at 'key' into '*hash'. */
864 dpif_flow_hash(const struct dpif
*dpif OVS_UNUSED
,
865 const void *key
, size_t key_len
, ovs_u128
*hash
)
867 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
868 static uint32_t secret
;
870 if (ovsthread_once_start(&once
)) {
871 secret
= random_uint32();
872 ovsthread_once_done(&once
);
874 hash_bytes128(key
, key_len
, secret
, hash
);
875 uuid_set_bits_v4((struct uuid
*)hash
);
878 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
879 * positive errno value. */
881 dpif_flow_flush(struct dpif
*dpif
)
885 COVERAGE_INC(dpif_flow_flush
);
887 error
= dpif
->dpif_class
->flow_flush(dpif
);
888 log_operation(dpif
, "flow_flush", error
);
892 /* Attempts to install 'key' into the datapath, fetches it, then deletes it.
893 * Returns true if the datapath supported installing 'flow', false otherwise.
896 dpif_probe_feature(struct dpif
*dpif
, const char *name
,
897 const struct ofpbuf
*key
, const ovs_u128
*ufid
)
899 struct dpif_flow flow
;
901 uint64_t stub
[DPIF_FLOW_BUFSIZE
/ 8];
902 bool enable_feature
= false;
905 /* Use DPIF_FP_MODIFY to cover the case where ovs-vswitchd is killed (and
906 * restarted) at just the right time such that feature probes from the
907 * previous run are still present in the datapath. */
908 error
= dpif_flow_put(dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
| DPIF_FP_PROBE
,
909 key
->data
, key
->size
, NULL
, 0, NULL
, 0,
910 ufid
, NON_PMD_CORE_ID
, NULL
);
912 if (error
!= EINVAL
) {
913 VLOG_WARN("%s: %s flow probe failed (%s)",
914 dpif_name(dpif
), name
, ovs_strerror(error
));
919 ofpbuf_use_stack(&reply
, &stub
, sizeof stub
);
920 error
= dpif_flow_get(dpif
, key
->data
, key
->size
, ufid
,
921 NON_PMD_CORE_ID
, &reply
, &flow
);
923 && (!ufid
|| (flow
.ufid_present
924 && ovs_u128_equals(*ufid
, flow
.ufid
)))) {
925 enable_feature
= true;
928 error
= dpif_flow_del(dpif
, key
->data
, key
->size
, ufid
,
929 NON_PMD_CORE_ID
, NULL
);
931 VLOG_WARN("%s: failed to delete %s feature probe flow",
932 dpif_name(dpif
), name
);
935 return enable_feature
;
938 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_GET. */
940 dpif_flow_get(struct dpif
*dpif
,
941 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
942 const unsigned pmd_id
, struct ofpbuf
*buf
, struct dpif_flow
*flow
)
947 op
.type
= DPIF_OP_FLOW_GET
;
948 op
.u
.flow_get
.key
= key
;
949 op
.u
.flow_get
.key_len
= key_len
;
950 op
.u
.flow_get
.ufid
= ufid
;
951 op
.u
.flow_get
.pmd_id
= pmd_id
;
952 op
.u
.flow_get
.buffer
= buf
;
954 memset(flow
, 0, sizeof *flow
);
955 op
.u
.flow_get
.flow
= flow
;
956 op
.u
.flow_get
.flow
->key
= key
;
957 op
.u
.flow_get
.flow
->key_len
= key_len
;
960 dpif_operate(dpif
, &opp
, 1);
965 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_PUT. */
967 dpif_flow_put(struct dpif
*dpif
, enum dpif_flow_put_flags flags
,
968 const struct nlattr
*key
, size_t key_len
,
969 const struct nlattr
*mask
, size_t mask_len
,
970 const struct nlattr
*actions
, size_t actions_len
,
971 const ovs_u128
*ufid
, const unsigned pmd_id
,
972 struct dpif_flow_stats
*stats
)
977 op
.type
= DPIF_OP_FLOW_PUT
;
978 op
.u
.flow_put
.flags
= flags
;
979 op
.u
.flow_put
.key
= key
;
980 op
.u
.flow_put
.key_len
= key_len
;
981 op
.u
.flow_put
.mask
= mask
;
982 op
.u
.flow_put
.mask_len
= mask_len
;
983 op
.u
.flow_put
.actions
= actions
;
984 op
.u
.flow_put
.actions_len
= actions_len
;
985 op
.u
.flow_put
.ufid
= ufid
;
986 op
.u
.flow_put
.pmd_id
= pmd_id
;
987 op
.u
.flow_put
.stats
= stats
;
990 dpif_operate(dpif
, &opp
, 1);
995 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_DEL. */
997 dpif_flow_del(struct dpif
*dpif
,
998 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
999 const unsigned pmd_id
, struct dpif_flow_stats
*stats
)
1001 struct dpif_op
*opp
;
1004 op
.type
= DPIF_OP_FLOW_DEL
;
1005 op
.u
.flow_del
.key
= key
;
1006 op
.u
.flow_del
.key_len
= key_len
;
1007 op
.u
.flow_del
.ufid
= ufid
;
1008 op
.u
.flow_del
.pmd_id
= pmd_id
;
1009 op
.u
.flow_del
.stats
= stats
;
1010 op
.u
.flow_del
.terse
= false;
1013 dpif_operate(dpif
, &opp
, 1);
1018 /* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
1019 * flows in 'dpif'. If 'terse' is true, then only UFID and statistics will
1020 * be returned in the dump. Otherwise, all fields will be returned.
1022 * This function always successfully returns a dpif_flow_dump. Error
1023 * reporting is deferred to dpif_flow_dump_destroy(). */
1024 struct dpif_flow_dump
*
1025 dpif_flow_dump_create(const struct dpif
*dpif
, bool terse
)
1027 return dpif
->dpif_class
->flow_dump_create(dpif
, terse
);
1030 /* Destroys 'dump', which must have been created with dpif_flow_dump_create().
1031 * All dpif_flow_dump_thread structures previously created for 'dump' must
1032 * previously have been destroyed.
1034 * Returns 0 if the dump operation was error-free, otherwise a positive errno
1035 * value describing the problem. */
1037 dpif_flow_dump_destroy(struct dpif_flow_dump
*dump
)
1039 const struct dpif
*dpif
= dump
->dpif
;
1040 int error
= dpif
->dpif_class
->flow_dump_destroy(dump
);
1041 log_operation(dpif
, "flow_dump_destroy", error
);
1042 return error
== EOF
? 0 : error
;
1045 /* Returns new thread-local state for use with dpif_flow_dump_next(). */
1046 struct dpif_flow_dump_thread
*
1047 dpif_flow_dump_thread_create(struct dpif_flow_dump
*dump
)
1049 return dump
->dpif
->dpif_class
->flow_dump_thread_create(dump
);
1052 /* Releases 'thread'. */
1054 dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread
*thread
)
1056 thread
->dpif
->dpif_class
->flow_dump_thread_destroy(thread
);
1059 /* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0
1060 * if and only if no flows remained to be retrieved, otherwise a positive
1061 * number reflecting the number of elements in 'flows[]' that were updated.
1062 * The number of flows returned might be less than 'max_flows' because
1063 * fewer than 'max_flows' remained, because this particular datapath does not
1064 * benefit from batching, or because an error occurred partway through
1065 * retrieval. Thus, the caller should continue calling until a 0 return value,
1066 * even if intermediate return values are less than 'max_flows'.
1068 * No error status is immediately provided. An error status for the entire
1069 * dump operation is provided when it is completed by calling
1070 * dpif_flow_dump_destroy().
1072 * All of the data stored into 'flows' is owned by the datapath, not by the
1073 * caller, and the caller must not modify or free it. The datapath guarantees
1074 * that it remains accessible and unchanged until the first of:
1075 * - The next call to dpif_flow_dump_next() for 'thread', or
1076 * - The next rcu quiescent period. */
1078 dpif_flow_dump_next(struct dpif_flow_dump_thread
*thread
,
1079 struct dpif_flow
*flows
, int max_flows
)
1081 struct dpif
*dpif
= thread
->dpif
;
1084 ovs_assert(max_flows
> 0);
1085 n
= dpif
->dpif_class
->flow_dump_next(thread
, flows
, max_flows
);
1087 struct dpif_flow
*f
;
1089 for (f
= flows
; f
< &flows
[n
] && should_log_flow_message(0); f
++) {
1090 log_flow_message(dpif
, 0, "flow_dump",
1091 f
->key
, f
->key_len
, f
->mask
, f
->mask_len
,
1092 &f
->ufid
, &f
->stats
, f
->actions
, f
->actions_len
);
1095 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all flows", dpif_name(dpif
));
1100 struct dpif_execute_helper_aux
{
1102 const struct flow
*flow
;
1106 /* This is called for actions that need the context of the datapath to be
1109 dpif_execute_helper_cb(void *aux_
, struct dp_packet_batch
*packets_
,
1110 const struct nlattr
*action
, bool may_steal
)
1112 struct dpif_execute_helper_aux
*aux
= aux_
;
1113 int type
= nl_attr_type(action
);
1114 struct dp_packet
*packet
= packets_
->packets
[0];
1116 ovs_assert(packets_
->count
== 1);
1118 switch ((enum ovs_action_attr
)type
) {
1119 case OVS_ACTION_ATTR_CT
:
1120 case OVS_ACTION_ATTR_OUTPUT
:
1121 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
1122 case OVS_ACTION_ATTR_TUNNEL_POP
:
1123 case OVS_ACTION_ATTR_USERSPACE
:
1124 case OVS_ACTION_ATTR_RECIRC
: {
1125 struct dpif_execute execute
;
1126 struct ofpbuf execute_actions
;
1127 uint64_t stub
[256 / 8];
1128 struct pkt_metadata
*md
= &packet
->md
;
1131 dst_set
= flow_tnl_dst_is_set(&md
->tunnel
);
1133 /* The Linux kernel datapath throws away the tunnel information
1134 * that we supply as metadata. We have to use a "set" action to
1136 ofpbuf_use_stub(&execute_actions
, stub
, sizeof stub
);
1137 odp_put_tunnel_action(&md
->tunnel
, &execute_actions
);
1138 ofpbuf_put(&execute_actions
, action
, NLA_ALIGN(action
->nla_len
));
1140 execute
.actions
= execute_actions
.data
;
1141 execute
.actions_len
= execute_actions
.size
;
1143 execute
.actions
= action
;
1144 execute
.actions_len
= NLA_ALIGN(action
->nla_len
);
1147 struct dp_packet
*clone
= NULL
;
1148 uint32_t cutlen
= dp_packet_get_cutlen(packet
);
1149 if (cutlen
&& (type
== OVS_ACTION_ATTR_OUTPUT
1150 || type
== OVS_ACTION_ATTR_TUNNEL_PUSH
1151 || type
== OVS_ACTION_ATTR_TUNNEL_POP
1152 || type
== OVS_ACTION_ATTR_USERSPACE
)) {
1153 dp_packet_reset_cutlen(packet
);
1155 packet
= clone
= dp_packet_clone(packet
);
1157 dp_packet_set_size(packet
, dp_packet_size(packet
) - cutlen
);
1160 execute
.packet
= packet
;
1161 execute
.flow
= aux
->flow
;
1162 execute
.needs_help
= false;
1163 execute
.probe
= false;
1165 aux
->error
= dpif_execute(aux
->dpif
, &execute
);
1166 log_execute_message(aux
->dpif
, &execute
, true, aux
->error
);
1168 dp_packet_delete(clone
);
1171 ofpbuf_uninit(&execute_actions
);
1176 case OVS_ACTION_ATTR_HASH
:
1177 case OVS_ACTION_ATTR_PUSH_VLAN
:
1178 case OVS_ACTION_ATTR_POP_VLAN
:
1179 case OVS_ACTION_ATTR_PUSH_MPLS
:
1180 case OVS_ACTION_ATTR_POP_MPLS
:
1181 case OVS_ACTION_ATTR_SET
:
1182 case OVS_ACTION_ATTR_SET_MASKED
:
1183 case OVS_ACTION_ATTR_SAMPLE
:
1184 case OVS_ACTION_ATTR_TRUNC
:
1185 case OVS_ACTION_ATTR_CLONE
:
1186 case OVS_ACTION_ATTR_UNSPEC
:
1187 case __OVS_ACTION_ATTR_MAX
:
1192 /* Executes 'execute' by performing most of the actions in userspace and
1193 * passing the fully constructed packets to 'dpif' for output and userspace
1196 * This helps with actions that a given 'dpif' doesn't implement directly. */
1198 dpif_execute_with_help(struct dpif
*dpif
, struct dpif_execute
*execute
)
1200 struct dpif_execute_helper_aux aux
= {dpif
, execute
->flow
, 0};
1201 struct dp_packet_batch pb
;
1203 COVERAGE_INC(dpif_execute_with_help
);
1205 dp_packet_batch_init_packet(&pb
, execute
->packet
);
1206 odp_execute_actions(&aux
, &pb
, false, execute
->actions
,
1207 execute
->actions_len
, dpif_execute_helper_cb
);
1211 /* Returns true if the datapath needs help executing 'execute'. */
1213 dpif_execute_needs_help(const struct dpif_execute
*execute
)
1215 return execute
->needs_help
|| nl_attr_oversized(execute
->actions_len
);
1218 /* A dpif_operate() wrapper for performing a single DPIF_OP_EXECUTE. */
1220 dpif_execute(struct dpif
*dpif
, struct dpif_execute
*execute
)
1222 if (execute
->actions_len
) {
1223 struct dpif_op
*opp
;
1226 op
.type
= DPIF_OP_EXECUTE
;
1227 op
.u
.execute
= *execute
;
1230 dpif_operate(dpif
, &opp
, 1);
1238 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1239 * which they are specified. Places each operation's results in the "output"
1240 * members documented in comments, and 0 in the 'error' member on success or a
1241 * positive errno on failure. */
1243 dpif_operate(struct dpif
*dpif
, struct dpif_op
**ops
, size_t n_ops
)
1248 /* Count 'chunk', the number of ops that can be executed without
1249 * needing any help. Ops that need help should be rare, so we
1250 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1251 for (chunk
= 0; chunk
< n_ops
; chunk
++) {
1252 struct dpif_op
*op
= ops
[chunk
];
1254 if (op
->type
== DPIF_OP_EXECUTE
1255 && dpif_execute_needs_help(&op
->u
.execute
)) {
1261 /* Execute a chunk full of ops that the dpif provider can
1262 * handle itself, without help. */
1265 dpif
->dpif_class
->operate(dpif
, ops
, chunk
);
1267 for (i
= 0; i
< chunk
; i
++) {
1268 struct dpif_op
*op
= ops
[i
];
1269 int error
= op
->error
;
1272 case DPIF_OP_FLOW_PUT
: {
1273 struct dpif_flow_put
*put
= &op
->u
.flow_put
;
1275 COVERAGE_INC(dpif_flow_put
);
1276 log_flow_put_message(dpif
, put
, error
);
1277 if (error
&& put
->stats
) {
1278 memset(put
->stats
, 0, sizeof *put
->stats
);
1283 case DPIF_OP_FLOW_GET
: {
1284 struct dpif_flow_get
*get
= &op
->u
.flow_get
;
1286 COVERAGE_INC(dpif_flow_get
);
1288 memset(get
->flow
, 0, sizeof *get
->flow
);
1290 log_flow_get_message(dpif
, get
, error
);
1295 case DPIF_OP_FLOW_DEL
: {
1296 struct dpif_flow_del
*del
= &op
->u
.flow_del
;
1298 COVERAGE_INC(dpif_flow_del
);
1299 log_flow_del_message(dpif
, del
, error
);
1300 if (error
&& del
->stats
) {
1301 memset(del
->stats
, 0, sizeof *del
->stats
);
1306 case DPIF_OP_EXECUTE
:
1307 COVERAGE_INC(dpif_execute
);
1308 log_execute_message(dpif
, &op
->u
.execute
, false, error
);
1316 /* Help the dpif provider to execute one op. */
1317 struct dpif_op
*op
= ops
[0];
1319 COVERAGE_INC(dpif_execute
);
1320 op
->error
= dpif_execute_with_help(dpif
, &op
->u
.execute
);
1327 /* Returns a string that represents 'type', for use in log messages. */
1329 dpif_upcall_type_to_string(enum dpif_upcall_type type
)
1332 case DPIF_UC_MISS
: return "miss";
1333 case DPIF_UC_ACTION
: return "action";
1334 case DPIF_N_UC_TYPES
: default: return "<unknown>";
1338 /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1339 * if successful, otherwise a positive errno value.
1341 * Turning packet receive off and then back on may change the Netlink PID
1342 * assignments returned by dpif_port_get_pid(). If the client does this, it
1343 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1344 * using the new PID assignment. */
1346 dpif_recv_set(struct dpif
*dpif
, bool enable
)
1350 if (dpif
->dpif_class
->recv_set
) {
1351 error
= dpif
->dpif_class
->recv_set(dpif
, enable
);
1352 log_operation(dpif
, "recv_set", error
);
1357 /* Refreshes the poll loops and Netlink sockets associated to each port,
1358 * when the number of upcall handlers (upcall receiving thread) is changed
1359 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
1362 * Since multiple upcall handlers can read upcalls simultaneously from
1363 * 'dpif', each port can have multiple Netlink sockets, one per upcall
1364 * handler. So, handlers_set() is responsible for the following tasks:
1366 * When receiving upcall is enabled, extends or creates the
1367 * configuration to support:
1369 * - 'n_handlers' Netlink sockets for each port.
1371 * - 'n_handlers' poll loops, one for each upcall handler.
1373 * - registering the Netlink sockets for the same upcall handler to
1374 * the corresponding poll loop.
1376 * Returns 0 if successful, otherwise a positive errno value. */
1378 dpif_handlers_set(struct dpif
*dpif
, uint32_t n_handlers
)
1382 if (dpif
->dpif_class
->handlers_set
) {
1383 error
= dpif
->dpif_class
->handlers_set(dpif
, n_handlers
);
1384 log_operation(dpif
, "handlers_set", error
);
1390 dpif_register_dp_purge_cb(struct dpif
*dpif
, dp_purge_callback
*cb
, void *aux
)
1392 if (dpif
->dpif_class
->register_dp_purge_cb
) {
1393 dpif
->dpif_class
->register_dp_purge_cb(dpif
, cb
, aux
);
1398 dpif_register_upcall_cb(struct dpif
*dpif
, upcall_callback
*cb
, void *aux
)
1400 if (dpif
->dpif_class
->register_upcall_cb
) {
1401 dpif
->dpif_class
->register_upcall_cb(dpif
, cb
, aux
);
1406 dpif_enable_upcall(struct dpif
*dpif
)
1408 if (dpif
->dpif_class
->enable_upcall
) {
1409 dpif
->dpif_class
->enable_upcall(dpif
);
1414 dpif_disable_upcall(struct dpif
*dpif
)
1416 if (dpif
->dpif_class
->disable_upcall
) {
1417 dpif
->dpif_class
->disable_upcall(dpif
);
1422 dpif_print_packet(struct dpif
*dpif
, struct dpif_upcall
*upcall
)
1424 if (!VLOG_DROP_DBG(&dpmsg_rl
)) {
1428 packet
= ofp_packet_to_string(dp_packet_data(&upcall
->packet
),
1429 dp_packet_size(&upcall
->packet
));
1432 odp_flow_key_format(upcall
->key
, upcall
->key_len
, &flow
);
1434 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1435 dpif_name(dpif
), dpif_upcall_type_to_string(upcall
->type
),
1436 ds_cstr(&flow
), packet
);
1443 /* Pass custom configuration to the datapath implementation. Some of the
1444 * changes can be postponed until dpif_run() is called. */
1446 dpif_set_config(struct dpif
*dpif
, const struct smap
*cfg
)
1450 if (dpif
->dpif_class
->set_config
) {
1451 error
= dpif
->dpif_class
->set_config(dpif
, cfg
);
1453 log_operation(dpif
, "set_config", error
);
1460 /* Polls for an upcall from 'dpif' for an upcall handler. Since there
1461 * there can be multiple poll loops, 'handler_id' is needed as index to
1462 * identify the corresponding poll loop. If successful, stores the upcall
1463 * into '*upcall', using 'buf' for storage. Should only be called if
1464 * 'recv_set' has been used to enable receiving packets from 'dpif'.
1466 * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
1467 * 'buf', so their memory cannot be freed separately from 'buf'.
1469 * The caller owns the data of 'upcall->packet' and may modify it. If
1470 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
1471 * will be reallocated. This requires the data of 'upcall->packet' to be
1472 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
1473 * when an error is returned, the 'upcall->packet' may be uninitialized
1474 * and should not be released.
1476 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1477 * if no upcall is immediately available. */
1479 dpif_recv(struct dpif
*dpif
, uint32_t handler_id
, struct dpif_upcall
*upcall
,
1484 if (dpif
->dpif_class
->recv
) {
1485 error
= dpif
->dpif_class
->recv(dpif
, handler_id
, upcall
, buf
);
1487 dpif_print_packet(dpif
, upcall
);
1488 } else if (error
!= EAGAIN
) {
1489 log_operation(dpif
, "recv", error
);
1495 /* Discards all messages that would otherwise be received by dpif_recv() on
1498 dpif_recv_purge(struct dpif
*dpif
)
1500 COVERAGE_INC(dpif_purge
);
1501 if (dpif
->dpif_class
->recv_purge
) {
1502 dpif
->dpif_class
->recv_purge(dpif
);
1506 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
1507 * 'dpif' has a message queued to be received with the recv member
1508 * function. Since there can be multiple poll loops, 'handler_id' is
1509 * needed as index to identify the corresponding poll loop. */
1511 dpif_recv_wait(struct dpif
*dpif
, uint32_t handler_id
)
1513 if (dpif
->dpif_class
->recv_wait
) {
1514 dpif
->dpif_class
->recv_wait(dpif
, handler_id
);
1519 * Return the datapath version. Caller is responsible for freeing
1523 dpif_get_dp_version(const struct dpif
*dpif
)
1525 char *version
= NULL
;
1527 if (dpif
->dpif_class
->get_datapath_version
) {
1528 version
= dpif
->dpif_class
->get_datapath_version();
1534 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1535 * and '*engine_id', respectively. */
1537 dpif_get_netflow_ids(const struct dpif
*dpif
,
1538 uint8_t *engine_type
, uint8_t *engine_id
)
1540 *engine_type
= dpif
->netflow_engine_type
;
1541 *engine_id
= dpif
->netflow_engine_id
;
1544 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1545 * value used for setting packet priority.
1546 * On success, returns 0 and stores the priority into '*priority'.
1547 * On failure, returns a positive errno value and stores 0 into '*priority'. */
1549 dpif_queue_to_priority(const struct dpif
*dpif
, uint32_t queue_id
,
1552 int error
= (dpif
->dpif_class
->queue_to_priority
1553 ? dpif
->dpif_class
->queue_to_priority(dpif
, queue_id
,
1559 log_operation(dpif
, "queue_to_priority", error
);
1564 dpif_init(struct dpif
*dpif
, const struct dpif_class
*dpif_class
,
1566 uint8_t netflow_engine_type
, uint8_t netflow_engine_id
)
1568 dpif
->dpif_class
= dpif_class
;
1569 dpif
->base_name
= xstrdup(name
);
1570 dpif
->full_name
= xasprintf("%s@%s", dpif_class
->type
, name
);
1571 dpif
->netflow_engine_type
= netflow_engine_type
;
1572 dpif
->netflow_engine_id
= netflow_engine_id
;
1575 /* Undoes the results of initialization.
1577 * Normally this function only needs to be called from dpif_close().
1578 * However, it may be called by providers due to an error on opening
1579 * that occurs after initialization. It this case dpif_close() would
1580 * never be called. */
1582 dpif_uninit(struct dpif
*dpif
, bool close
)
1584 char *base_name
= dpif
->base_name
;
1585 char *full_name
= dpif
->full_name
;
1588 dpif
->dpif_class
->close(dpif
);
1596 log_operation(const struct dpif
*dpif
, const char *operation
, int error
)
1599 VLOG_DBG_RL(&dpmsg_rl
, "%s: %s success", dpif_name(dpif
), operation
);
1600 } else if (ofperr_is_valid(error
)) {
1601 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1602 dpif_name(dpif
), operation
, ofperr_get_name(error
));
1604 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1605 dpif_name(dpif
), operation
, ovs_strerror(error
));
1609 static enum vlog_level
1610 flow_message_log_level(int error
)
1612 /* If flows arrive in a batch, userspace may push down multiple
1613 * unique flow definitions that overlap when wildcards are applied.
1614 * Kernels that support flow wildcarding will reject these flows as
1615 * duplicates (EEXIST), so lower the log level to debug for these
1616 * types of messages. */
1617 return (error
&& error
!= EEXIST
) ? VLL_WARN
: VLL_DBG
;
1621 should_log_flow_message(int error
)
1623 return !vlog_should_drop(&this_module
, flow_message_log_level(error
),
1624 error
? &error_rl
: &dpmsg_rl
);
1628 log_flow_message(const struct dpif
*dpif
, int error
, const char *operation
,
1629 const struct nlattr
*key
, size_t key_len
,
1630 const struct nlattr
*mask
, size_t mask_len
,
1631 const ovs_u128
*ufid
, const struct dpif_flow_stats
*stats
,
1632 const struct nlattr
*actions
, size_t actions_len
)
1634 struct ds ds
= DS_EMPTY_INITIALIZER
;
1635 ds_put_format(&ds
, "%s: ", dpif_name(dpif
));
1637 ds_put_cstr(&ds
, "failed to ");
1639 ds_put_format(&ds
, "%s ", operation
);
1641 ds_put_format(&ds
, "(%s) ", ovs_strerror(error
));
1644 odp_format_ufid(ufid
, &ds
);
1645 ds_put_cstr(&ds
, " ");
1647 odp_flow_format(key
, key_len
, mask
, mask_len
, NULL
, &ds
, true);
1649 ds_put_cstr(&ds
, ", ");
1650 dpif_flow_stats_format(stats
, &ds
);
1652 if (actions
|| actions_len
) {
1653 ds_put_cstr(&ds
, ", actions:");
1654 format_odp_actions(&ds
, actions
, actions_len
);
1656 vlog(&this_module
, flow_message_log_level(error
), "%s", ds_cstr(&ds
));
1661 log_flow_put_message(struct dpif
*dpif
, const struct dpif_flow_put
*put
,
1664 if (should_log_flow_message(error
) && !(put
->flags
& DPIF_FP_PROBE
)) {
1668 ds_put_cstr(&s
, "put");
1669 if (put
->flags
& DPIF_FP_CREATE
) {
1670 ds_put_cstr(&s
, "[create]");
1672 if (put
->flags
& DPIF_FP_MODIFY
) {
1673 ds_put_cstr(&s
, "[modify]");
1675 if (put
->flags
& DPIF_FP_ZERO_STATS
) {
1676 ds_put_cstr(&s
, "[zero]");
1678 log_flow_message(dpif
, error
, ds_cstr(&s
),
1679 put
->key
, put
->key_len
, put
->mask
, put
->mask_len
,
1680 put
->ufid
, put
->stats
, put
->actions
,
1687 log_flow_del_message(struct dpif
*dpif
, const struct dpif_flow_del
*del
,
1690 if (should_log_flow_message(error
)) {
1691 log_flow_message(dpif
, error
, "flow_del", del
->key
, del
->key_len
,
1692 NULL
, 0, del
->ufid
, !error
? del
->stats
: NULL
,
1697 /* Logs that 'execute' was executed on 'dpif' and completed with errno 'error'
1698 * (0 for success). 'subexecute' should be true if the execution is a result
1699 * of breaking down a larger execution that needed help, false otherwise.
1702 * XXX In theory, the log message could be deceptive because this function is
1703 * called after the dpif_provider's '->execute' function, which is allowed to
1704 * modify execute->packet and execute->md. In practice, though:
1706 * - dpif-netlink doesn't modify execute->packet or execute->md.
1708 * - dpif-netdev does modify them but it is less likely to have problems
1709 * because it is built into ovs-vswitchd and cannot have version skew,
1712 * It would still be better to avoid the potential problem. I don't know of a
1713 * good way to do that, though, that isn't expensive. */
1715 log_execute_message(struct dpif
*dpif
, const struct dpif_execute
*execute
,
1716 bool subexecute
, int error
)
1718 if (!(error
? VLOG_DROP_WARN(&error_rl
) : VLOG_DROP_DBG(&dpmsg_rl
))
1719 && !execute
->probe
) {
1720 struct ds ds
= DS_EMPTY_INITIALIZER
;
1723 packet
= ofp_packet_to_string(dp_packet_data(execute
->packet
),
1724 dp_packet_size(execute
->packet
));
1725 ds_put_format(&ds
, "%s: %sexecute ",
1727 (subexecute
? "sub-"
1728 : dpif_execute_needs_help(execute
) ? "super-"
1730 format_odp_actions(&ds
, execute
->actions
, execute
->actions_len
);
1732 ds_put_format(&ds
, " failed (%s)", ovs_strerror(error
));
1734 ds_put_format(&ds
, " on packet %s", packet
);
1735 ds_put_format(&ds
, " mtu %d", execute
->mtu
);
1736 vlog(&this_module
, error
? VLL_WARN
: VLL_DBG
, "%s", ds_cstr(&ds
));
1743 log_flow_get_message(const struct dpif
*dpif
, const struct dpif_flow_get
*get
,
1746 if (should_log_flow_message(error
)) {
1747 log_flow_message(dpif
, error
, "flow_get",
1748 get
->key
, get
->key_len
,
1749 get
->flow
->mask
, get
->flow
->mask_len
,
1750 get
->ufid
, &get
->flow
->stats
,
1751 get
->flow
->actions
, get
->flow
->actions_len
);
1756 dpif_supports_tnl_push_pop(const struct dpif
*dpif
)
1758 return dpif_is_netdev(dpif
);