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
);
66 COVERAGE_DEFINE(dpif_meter_set
);
67 COVERAGE_DEFINE(dpif_meter_get
);
68 COVERAGE_DEFINE(dpif_meter_del
);
70 static const struct dpif_class
*base_dpif_classes
[] = {
71 #if defined(__linux__) || defined(_WIN32)
77 struct registered_dpif_class
{
78 const struct dpif_class
*dpif_class
;
81 static struct shash dpif_classes
= SHASH_INITIALIZER(&dpif_classes
);
82 static struct sset dpif_blacklist
= SSET_INITIALIZER(&dpif_blacklist
);
84 /* Protects 'dpif_classes', including the refcount, and 'dpif_blacklist'. */
85 static struct ovs_mutex dpif_mutex
= OVS_MUTEX_INITIALIZER
;
87 /* Rate limit for individual messages going to or from the datapath, output at
88 * DBG level. This is very high because, if these are enabled, it is because
89 * we really need to see them. */
90 static struct vlog_rate_limit dpmsg_rl
= VLOG_RATE_LIMIT_INIT(600, 600);
92 /* Not really much point in logging many dpif errors. */
93 static struct vlog_rate_limit error_rl
= VLOG_RATE_LIMIT_INIT(60, 5);
95 static void log_flow_message(const struct dpif
*dpif
, int error
,
96 const char *operation
,
97 const struct nlattr
*key
, size_t key_len
,
98 const struct nlattr
*mask
, size_t mask_len
,
100 const struct dpif_flow_stats
*stats
,
101 const struct nlattr
*actions
, size_t actions_len
);
102 static void log_operation(const struct dpif
*, const char *operation
,
104 static bool should_log_flow_message(int error
);
105 static void log_flow_put_message(struct dpif
*, const struct dpif_flow_put
*,
107 static void log_flow_del_message(struct dpif
*, const struct dpif_flow_del
*,
109 static void log_execute_message(struct dpif
*, const struct dpif_execute
*,
110 bool subexecute
, int error
);
111 static void log_flow_get_message(const struct dpif
*,
112 const struct dpif_flow_get
*, int error
);
114 /* Incremented whenever tnl route, arp, etc changes. */
115 struct seq
*tnl_conf_seq
;
120 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
122 if (ovsthread_once_start(&once
)) {
125 tnl_conf_seq
= seq_create();
126 dpctl_unixctl_register();
128 tnl_neigh_cache_init();
131 for (i
= 0; i
< ARRAY_SIZE(base_dpif_classes
); i
++) {
132 dp_register_provider(base_dpif_classes
[i
]);
135 ovsthread_once_done(&once
);
140 dp_register_provider__(const struct dpif_class
*new_class
)
142 struct registered_dpif_class
*registered_class
;
145 if (sset_contains(&dpif_blacklist
, new_class
->type
)) {
146 VLOG_DBG("attempted to register blacklisted provider: %s",
151 if (shash_find(&dpif_classes
, new_class
->type
)) {
152 VLOG_WARN("attempted to register duplicate datapath provider: %s",
157 error
= new_class
->init
? new_class
->init() : 0;
159 VLOG_WARN("failed to initialize %s datapath class: %s",
160 new_class
->type
, ovs_strerror(error
));
164 registered_class
= xmalloc(sizeof *registered_class
);
165 registered_class
->dpif_class
= new_class
;
166 registered_class
->refcount
= 0;
168 shash_add(&dpif_classes
, new_class
->type
, registered_class
);
173 /* Registers a new datapath provider. After successful registration, new
174 * datapaths of that type can be opened using dpif_open(). */
176 dp_register_provider(const struct dpif_class
*new_class
)
180 ovs_mutex_lock(&dpif_mutex
);
181 error
= dp_register_provider__(new_class
);
182 ovs_mutex_unlock(&dpif_mutex
);
187 /* Unregisters a datapath provider. 'type' must have been previously
188 * registered and not currently be in use by any dpifs. After unregistration
189 * new datapaths of that type cannot be opened using dpif_open(). */
191 dp_unregister_provider__(const char *type
)
193 struct shash_node
*node
;
194 struct registered_dpif_class
*registered_class
;
196 node
= shash_find(&dpif_classes
, type
);
201 registered_class
= node
->data
;
202 if (registered_class
->refcount
) {
203 VLOG_WARN("attempted to unregister in use datapath provider: %s", type
);
207 shash_delete(&dpif_classes
, node
);
208 free(registered_class
);
213 /* Unregisters a datapath provider. 'type' must have been previously
214 * registered and not currently be in use by any dpifs. After unregistration
215 * new datapaths of that type cannot be opened using dpif_open(). */
217 dp_unregister_provider(const char *type
)
223 ovs_mutex_lock(&dpif_mutex
);
224 error
= dp_unregister_provider__(type
);
225 ovs_mutex_unlock(&dpif_mutex
);
230 /* Blacklists a provider. Causes future calls of dp_register_provider() with
231 * a dpif_class which implements 'type' to fail. */
233 dp_blacklist_provider(const char *type
)
235 ovs_mutex_lock(&dpif_mutex
);
236 sset_add(&dpif_blacklist
, type
);
237 ovs_mutex_unlock(&dpif_mutex
);
240 /* Adds the types of all currently registered datapath providers to 'types'.
241 * The caller must first initialize the sset. */
243 dp_enumerate_types(struct sset
*types
)
245 struct shash_node
*node
;
249 ovs_mutex_lock(&dpif_mutex
);
250 SHASH_FOR_EACH(node
, &dpif_classes
) {
251 const struct registered_dpif_class
*registered_class
= node
->data
;
252 sset_add(types
, registered_class
->dpif_class
->type
);
254 ovs_mutex_unlock(&dpif_mutex
);
258 dp_class_unref(struct registered_dpif_class
*rc
)
260 ovs_mutex_lock(&dpif_mutex
);
261 ovs_assert(rc
->refcount
);
263 ovs_mutex_unlock(&dpif_mutex
);
266 static struct registered_dpif_class
*
267 dp_class_lookup(const char *type
)
269 struct registered_dpif_class
*rc
;
271 ovs_mutex_lock(&dpif_mutex
);
272 rc
= shash_find_data(&dpif_classes
, type
);
276 ovs_mutex_unlock(&dpif_mutex
);
281 /* Clears 'names' and enumerates the names of all known created datapaths with
282 * the given 'type'. The caller must first initialize the sset. Returns 0 if
283 * successful, otherwise a positive errno value.
285 * Some kinds of datapaths might not be practically enumerable. This is not
286 * considered an error. */
288 dp_enumerate_names(const char *type
, struct sset
*names
)
290 struct registered_dpif_class
*registered_class
;
291 const struct dpif_class
*dpif_class
;
297 registered_class
= dp_class_lookup(type
);
298 if (!registered_class
) {
299 VLOG_WARN("could not enumerate unknown type: %s", type
);
303 dpif_class
= registered_class
->dpif_class
;
304 error
= (dpif_class
->enumerate
305 ? dpif_class
->enumerate(names
, dpif_class
)
308 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class
->type
,
309 ovs_strerror(error
));
311 dp_class_unref(registered_class
);
316 /* Parses 'datapath_name_', which is of the form [type@]name into its
317 * component pieces. 'name' and 'type' must be freed by the caller.
319 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
321 dp_parse_name(const char *datapath_name_
, char **name
, char **type
)
323 char *datapath_name
= xstrdup(datapath_name_
);
326 separator
= strchr(datapath_name
, '@');
329 *type
= datapath_name
;
330 *name
= xstrdup(dpif_normalize_type(separator
+ 1));
332 *name
= datapath_name
;
333 *type
= xstrdup(dpif_normalize_type(NULL
));
338 do_open(const char *name
, const char *type
, bool create
, struct dpif
**dpifp
)
340 struct dpif
*dpif
= NULL
;
342 struct registered_dpif_class
*registered_class
;
346 type
= dpif_normalize_type(type
);
347 registered_class
= dp_class_lookup(type
);
348 if (!registered_class
) {
349 VLOG_WARN("could not create datapath %s of unknown type %s", name
,
351 error
= EAFNOSUPPORT
;
355 error
= registered_class
->dpif_class
->open(registered_class
->dpif_class
,
356 name
, create
, &dpif
);
358 ovs_assert(dpif
->dpif_class
== registered_class
->dpif_class
);
360 dp_class_unref(registered_class
);
364 *dpifp
= error
? NULL
: dpif
;
368 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
369 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
370 * the empty string to specify the default system type. Returns 0 if
371 * successful, otherwise a positive errno value. On success stores a pointer
372 * to the datapath in '*dpifp', otherwise a null pointer. */
374 dpif_open(const char *name
, const char *type
, struct dpif
**dpifp
)
376 return do_open(name
, type
, false, dpifp
);
379 /* Tries to create and open a new datapath with the given 'name' and 'type'.
380 * 'type' may be either NULL or the empty string to specify the default system
381 * type. Will fail if a datapath with 'name' and 'type' already exists.
382 * Returns 0 if successful, otherwise a positive errno value. On success
383 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
385 dpif_create(const char *name
, const char *type
, struct dpif
**dpifp
)
387 return do_open(name
, type
, true, dpifp
);
390 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
391 * does not exist. 'type' may be either NULL or the empty string to specify
392 * the default system type. Returns 0 if successful, otherwise a positive
393 * errno value. On success stores a pointer to the datapath in '*dpifp',
394 * otherwise a null pointer. */
396 dpif_create_and_open(const char *name
, const char *type
, struct dpif
**dpifp
)
400 error
= dpif_create(name
, type
, dpifp
);
401 if (error
== EEXIST
|| error
== EBUSY
) {
402 error
= dpif_open(name
, type
, dpifp
);
404 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
405 name
, ovs_strerror(error
));
408 VLOG_WARN("failed to create datapath %s: %s",
409 name
, ovs_strerror(error
));
414 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
415 * itself; call dpif_delete() first, instead, if that is desirable. */
417 dpif_close(struct dpif
*dpif
)
420 struct registered_dpif_class
*rc
;
422 rc
= shash_find_data(&dpif_classes
, dpif
->dpif_class
->type
);
423 dpif_uninit(dpif
, true);
428 /* Performs periodic work needed by 'dpif'. */
430 dpif_run(struct dpif
*dpif
)
432 if (dpif
->dpif_class
->run
) {
433 return dpif
->dpif_class
->run(dpif
);
438 /* Arranges for poll_block() to wake up when dp_run() needs to be called for
441 dpif_wait(struct dpif
*dpif
)
443 if (dpif
->dpif_class
->wait
) {
444 dpif
->dpif_class
->wait(dpif
);
448 /* Returns the name of datapath 'dpif' prefixed with the type
449 * (for use in log messages). */
451 dpif_name(const struct dpif
*dpif
)
453 return dpif
->full_name
;
456 /* Returns the name of datapath 'dpif' without the type
457 * (for use in device names). */
459 dpif_base_name(const struct dpif
*dpif
)
461 return dpif
->base_name
;
464 /* Returns the type of datapath 'dpif'. */
466 dpif_type(const struct dpif
*dpif
)
468 return dpif
->dpif_class
->type
;
471 /* Returns the fully spelled out name for the given datapath 'type'.
473 * Normalized type string can be compared with strcmp(). Unnormalized type
474 * string might be the same even if they have different spellings. */
476 dpif_normalize_type(const char *type
)
478 return type
&& type
[0] ? type
: "system";
481 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
482 * ports. After calling this function, it does not make sense to pass 'dpif'
483 * to any functions other than dpif_name() or dpif_close(). */
485 dpif_delete(struct dpif
*dpif
)
489 COVERAGE_INC(dpif_destroy
);
491 error
= dpif
->dpif_class
->destroy(dpif
);
492 log_operation(dpif
, "delete", error
);
496 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
497 * otherwise a positive errno value. */
499 dpif_get_dp_stats(const struct dpif
*dpif
, struct dpif_dp_stats
*stats
)
501 int error
= dpif
->dpif_class
->get_stats(dpif
, stats
);
503 memset(stats
, 0, sizeof *stats
);
505 log_operation(dpif
, "get_stats", error
);
510 dpif_port_open_type(const char *datapath_type
, const char *port_type
)
512 struct registered_dpif_class
*rc
;
514 datapath_type
= dpif_normalize_type(datapath_type
);
516 ovs_mutex_lock(&dpif_mutex
);
517 rc
= shash_find_data(&dpif_classes
, datapath_type
);
518 if (rc
&& rc
->dpif_class
->port_open_type
) {
519 port_type
= rc
->dpif_class
->port_open_type(rc
->dpif_class
, port_type
);
521 ovs_mutex_unlock(&dpif_mutex
);
526 /* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
527 * non-null and its value is not ODPP_NONE, then attempts to use the
528 * value as the port number.
530 * If successful, returns 0 and sets '*port_nop' to the new port's port
531 * number (if 'port_nop' is non-null). On failure, returns a positive
532 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
535 dpif_port_add(struct dpif
*dpif
, struct netdev
*netdev
, odp_port_t
*port_nop
)
537 const char *netdev_name
= netdev_get_name(netdev
);
538 odp_port_t port_no
= ODPP_NONE
;
541 COVERAGE_INC(dpif_port_add
);
547 error
= dpif
->dpif_class
->port_add(dpif
, netdev
, &port_no
);
549 VLOG_DBG_RL(&dpmsg_rl
, "%s: added %s as port %"PRIu32
,
550 dpif_name(dpif
), netdev_name
, port_no
);
552 VLOG_WARN_RL(&error_rl
, "%s: failed to add %s as port: %s",
553 dpif_name(dpif
), netdev_name
, ovs_strerror(error
));
562 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
563 * otherwise a positive errno value. */
565 dpif_port_del(struct dpif
*dpif
, odp_port_t port_no
)
569 COVERAGE_INC(dpif_port_del
);
571 error
= dpif
->dpif_class
->port_del(dpif
, port_no
);
573 VLOG_DBG_RL(&dpmsg_rl
, "%s: port_del(%"PRIu32
")",
574 dpif_name(dpif
), port_no
);
576 log_operation(dpif
, "port_del", error
);
581 /* Makes a deep copy of 'src' into 'dst'. */
583 dpif_port_clone(struct dpif_port
*dst
, const struct dpif_port
*src
)
585 dst
->name
= xstrdup(src
->name
);
586 dst
->type
= xstrdup(src
->type
);
587 dst
->port_no
= src
->port_no
;
590 /* Frees memory allocated to members of 'dpif_port'.
592 * Do not call this function on a dpif_port obtained from
593 * dpif_port_dump_next(): that function retains ownership of the data in the
596 dpif_port_destroy(struct dpif_port
*dpif_port
)
598 free(dpif_port
->name
);
599 free(dpif_port
->type
);
602 /* Checks if port named 'devname' exists in 'dpif'. If so, returns
603 * true; otherwise, returns false. */
605 dpif_port_exists(const struct dpif
*dpif
, const char *devname
)
607 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, NULL
);
608 if (error
!= 0 && error
!= ENODEV
) {
609 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %s: %s",
610 dpif_name(dpif
), devname
, ovs_strerror(error
));
616 /* Refreshes configuration of 'dpif's port. */
618 dpif_port_set_config(struct dpif
*dpif
, odp_port_t port_no
,
619 const struct smap
*cfg
)
623 if (dpif
->dpif_class
->port_set_config
) {
624 error
= dpif
->dpif_class
->port_set_config(dpif
, port_no
, cfg
);
626 log_operation(dpif
, "port_set_config", error
);
633 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
634 * initializes '*port' appropriately; on failure, returns a positive errno
637 * Retuns ENODEV if the port doesn't exist.
639 * The caller owns the data in 'port' and must free it with
640 * dpif_port_destroy() when it is no longer needed. */
642 dpif_port_query_by_number(const struct dpif
*dpif
, odp_port_t port_no
,
643 struct dpif_port
*port
)
645 int error
= dpif
->dpif_class
->port_query_by_number(dpif
, port_no
, port
);
647 VLOG_DBG_RL(&dpmsg_rl
, "%s: port %"PRIu32
" is device %s",
648 dpif_name(dpif
), port_no
, port
->name
);
650 memset(port
, 0, sizeof *port
);
651 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %"PRIu32
": %s",
652 dpif_name(dpif
), port_no
, ovs_strerror(error
));
657 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
658 * initializes '*port' appropriately; on failure, returns a positive errno
661 * Retuns ENODEV if the port doesn't exist.
663 * The caller owns the data in 'port' and must free it with
664 * dpif_port_destroy() when it is no longer needed. */
666 dpif_port_query_by_name(const struct dpif
*dpif
, const char *devname
,
667 struct dpif_port
*port
)
669 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, port
);
671 VLOG_DBG_RL(&dpmsg_rl
, "%s: device %s is on port %"PRIu32
,
672 dpif_name(dpif
), devname
, port
->port_no
);
674 memset(port
, 0, sizeof *port
);
676 /* For ENODEV we use DBG level because the caller is probably
677 * interested in whether 'dpif' actually has a port 'devname', so that
678 * it's not an issue worth logging if it doesn't. Other errors are
679 * uncommon and more likely to indicate a real problem. */
680 VLOG_RL(&error_rl
, error
== ENODEV
? VLL_DBG
: VLL_WARN
,
681 "%s: failed to query port %s: %s",
682 dpif_name(dpif
), devname
, ovs_strerror(error
));
687 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
688 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
689 * flows whose packets arrived on port 'port_no'. In the case where the
690 * provider allocates multiple Netlink PIDs to a single port, it may use
691 * 'hash' to spread load among them. The caller need not use a particular
692 * hash function; a 5-tuple hash is suitable.
694 * (The datapath implementation might use some different hash function for
695 * distributing packets received via flow misses among PIDs. This means
696 * that packets received via flow misses might be reordered relative to
697 * packets received via userspace actions. This is not ordinarily a
700 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
701 * allocated to any port, that the client may use for special purposes.
703 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
704 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
705 * disabled and then re-enabled, so a client that does that must be prepared to
706 * update all of the flows that it installed that contain
707 * OVS_ACTION_ATTR_USERSPACE actions. */
709 dpif_port_get_pid(const struct dpif
*dpif
, odp_port_t port_no
, uint32_t hash
)
711 return (dpif
->dpif_class
->port_get_pid
712 ? (dpif
->dpif_class
->port_get_pid
)(dpif
, port_no
, hash
)
716 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
717 * the port's name into the 'name_size' bytes in 'name', ensuring that the
718 * result is null-terminated. On failure, returns a positive errno value and
719 * makes 'name' the empty string. */
721 dpif_port_get_name(struct dpif
*dpif
, odp_port_t port_no
,
722 char *name
, size_t name_size
)
724 struct dpif_port port
;
727 ovs_assert(name_size
> 0);
729 error
= dpif_port_query_by_number(dpif
, port_no
, &port
);
731 ovs_strlcpy(name
, port
.name
, name_size
);
732 dpif_port_destroy(&port
);
739 /* Initializes 'dump' to begin dumping the ports in a dpif.
741 * This function provides no status indication. An error status for the entire
742 * dump operation is provided when it is completed by calling
743 * dpif_port_dump_done().
746 dpif_port_dump_start(struct dpif_port_dump
*dump
, const struct dpif
*dpif
)
749 dump
->error
= dpif
->dpif_class
->port_dump_start(dpif
, &dump
->state
);
750 log_operation(dpif
, "port_dump_start", dump
->error
);
753 /* Attempts to retrieve another port from 'dump', which must have been
754 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
755 * into 'port' and returns true. On failure, returns false.
757 * Failure might indicate an actual error or merely that the last port has been
758 * dumped. An error status for the entire dump operation is provided when it
759 * is completed by calling dpif_port_dump_done().
761 * The dpif owns the data stored in 'port'. It will remain valid until at
762 * least the next time 'dump' is passed to dpif_port_dump_next() or
763 * dpif_port_dump_done(). */
765 dpif_port_dump_next(struct dpif_port_dump
*dump
, struct dpif_port
*port
)
767 const struct dpif
*dpif
= dump
->dpif
;
773 dump
->error
= dpif
->dpif_class
->port_dump_next(dpif
, dump
->state
, port
);
774 if (dump
->error
== EOF
) {
775 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all ports", dpif_name(dpif
));
777 log_operation(dpif
, "port_dump_next", dump
->error
);
781 dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
787 /* Completes port table dump operation 'dump', which must have been initialized
788 * with dpif_port_dump_start(). Returns 0 if the dump operation was
789 * error-free, otherwise a positive errno value describing the problem. */
791 dpif_port_dump_done(struct dpif_port_dump
*dump
)
793 const struct dpif
*dpif
= dump
->dpif
;
795 dump
->error
= dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
796 log_operation(dpif
, "port_dump_done", dump
->error
);
798 return dump
->error
== EOF
? 0 : dump
->error
;
801 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
802 * 'dpif' has changed, this function does one of the following:
804 * - Stores the name of the device that was added to or deleted from 'dpif' in
805 * '*devnamep' and returns 0. The caller is responsible for freeing
806 * '*devnamep' (with free()) when it no longer needs it.
808 * - Returns ENOBUFS and sets '*devnamep' to NULL.
810 * This function may also return 'false positives', where it returns 0 and
811 * '*devnamep' names a device that was not actually added or deleted or it
812 * returns ENOBUFS without any change.
814 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
815 * return other positive errno values to indicate that something has gone
818 dpif_port_poll(const struct dpif
*dpif
, char **devnamep
)
820 int error
= dpif
->dpif_class
->port_poll(dpif
, devnamep
);
827 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
828 * value other than EAGAIN. */
830 dpif_port_poll_wait(const struct dpif
*dpif
)
832 dpif
->dpif_class
->port_poll_wait(dpif
);
835 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
836 * arguments must have been initialized through a call to flow_extract().
837 * 'used' is stored into stats->used. */
839 dpif_flow_stats_extract(const struct flow
*flow
, const struct dp_packet
*packet
,
840 long long int used
, struct dpif_flow_stats
*stats
)
842 stats
->tcp_flags
= ntohs(flow
->tcp_flags
);
843 stats
->n_bytes
= dp_packet_size(packet
);
844 stats
->n_packets
= 1;
848 /* Appends a human-readable representation of 'stats' to 's'. */
850 dpif_flow_stats_format(const struct dpif_flow_stats
*stats
, struct ds
*s
)
852 ds_put_format(s
, "packets:%"PRIu64
", bytes:%"PRIu64
", used:",
853 stats
->n_packets
, stats
->n_bytes
);
855 ds_put_format(s
, "%.3fs", (time_msec() - stats
->used
) / 1000.0);
857 ds_put_format(s
, "never");
859 if (stats
->tcp_flags
) {
860 ds_put_cstr(s
, ", flags:");
861 packet_format_tcp_flags(s
, stats
->tcp_flags
);
865 /* Places the hash of the 'key_len' bytes starting at 'key' into '*hash'. */
867 dpif_flow_hash(const struct dpif
*dpif OVS_UNUSED
,
868 const void *key
, size_t key_len
, ovs_u128
*hash
)
870 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
871 static uint32_t secret
;
873 if (ovsthread_once_start(&once
)) {
874 secret
= random_uint32();
875 ovsthread_once_done(&once
);
877 hash_bytes128(key
, key_len
, secret
, hash
);
878 uuid_set_bits_v4((struct uuid
*)hash
);
881 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
882 * positive errno value. */
884 dpif_flow_flush(struct dpif
*dpif
)
888 COVERAGE_INC(dpif_flow_flush
);
890 error
= dpif
->dpif_class
->flow_flush(dpif
);
891 log_operation(dpif
, "flow_flush", error
);
895 /* Attempts to install 'key' into the datapath, fetches it, then deletes it.
896 * Returns true if the datapath supported installing 'flow', false otherwise.
899 dpif_probe_feature(struct dpif
*dpif
, const char *name
,
900 const struct ofpbuf
*key
, const struct ofpbuf
*actions
,
901 const ovs_u128
*ufid
)
903 struct dpif_flow flow
;
905 uint64_t stub
[DPIF_FLOW_BUFSIZE
/ 8];
906 bool enable_feature
= false;
908 const struct nlattr
*nl_actions
= actions
? actions
->data
: NULL
;
909 const size_t nl_actions_size
= actions
? actions
->size
: 0;
911 /* Use DPIF_FP_MODIFY to cover the case where ovs-vswitchd is killed (and
912 * restarted) at just the right time such that feature probes from the
913 * previous run are still present in the datapath. */
914 error
= dpif_flow_put(dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
| DPIF_FP_PROBE
,
915 key
->data
, key
->size
, NULL
, 0,
916 nl_actions
, nl_actions_size
,
917 ufid
, NON_PMD_CORE_ID
, NULL
);
919 if (error
!= EINVAL
&& error
!= EOVERFLOW
) {
920 VLOG_WARN("%s: %s flow probe failed (%s)",
921 dpif_name(dpif
), name
, ovs_strerror(error
));
926 ofpbuf_use_stack(&reply
, &stub
, sizeof stub
);
927 error
= dpif_flow_get(dpif
, key
->data
, key
->size
, ufid
,
928 NON_PMD_CORE_ID
, &reply
, &flow
);
930 && (!ufid
|| (flow
.ufid_present
931 && ovs_u128_equals(*ufid
, flow
.ufid
)))) {
932 enable_feature
= true;
935 error
= dpif_flow_del(dpif
, key
->data
, key
->size
, ufid
,
936 NON_PMD_CORE_ID
, NULL
);
938 VLOG_WARN("%s: failed to delete %s feature probe flow",
939 dpif_name(dpif
), name
);
942 return enable_feature
;
945 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_GET. */
947 dpif_flow_get(struct dpif
*dpif
,
948 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
949 const unsigned pmd_id
, struct ofpbuf
*buf
, struct dpif_flow
*flow
)
954 op
.type
= DPIF_OP_FLOW_GET
;
955 op
.u
.flow_get
.key
= key
;
956 op
.u
.flow_get
.key_len
= key_len
;
957 op
.u
.flow_get
.ufid
= ufid
;
958 op
.u
.flow_get
.pmd_id
= pmd_id
;
959 op
.u
.flow_get
.buffer
= buf
;
961 memset(flow
, 0, sizeof *flow
);
962 op
.u
.flow_get
.flow
= flow
;
963 op
.u
.flow_get
.flow
->key
= key
;
964 op
.u
.flow_get
.flow
->key_len
= key_len
;
967 dpif_operate(dpif
, &opp
, 1);
972 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_PUT. */
974 dpif_flow_put(struct dpif
*dpif
, enum dpif_flow_put_flags flags
,
975 const struct nlattr
*key
, size_t key_len
,
976 const struct nlattr
*mask
, size_t mask_len
,
977 const struct nlattr
*actions
, size_t actions_len
,
978 const ovs_u128
*ufid
, const unsigned pmd_id
,
979 struct dpif_flow_stats
*stats
)
984 op
.type
= DPIF_OP_FLOW_PUT
;
985 op
.u
.flow_put
.flags
= flags
;
986 op
.u
.flow_put
.key
= key
;
987 op
.u
.flow_put
.key_len
= key_len
;
988 op
.u
.flow_put
.mask
= mask
;
989 op
.u
.flow_put
.mask_len
= mask_len
;
990 op
.u
.flow_put
.actions
= actions
;
991 op
.u
.flow_put
.actions_len
= actions_len
;
992 op
.u
.flow_put
.ufid
= ufid
;
993 op
.u
.flow_put
.pmd_id
= pmd_id
;
994 op
.u
.flow_put
.stats
= stats
;
997 dpif_operate(dpif
, &opp
, 1);
1002 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_DEL. */
1004 dpif_flow_del(struct dpif
*dpif
,
1005 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
1006 const unsigned pmd_id
, struct dpif_flow_stats
*stats
)
1008 struct dpif_op
*opp
;
1011 op
.type
= DPIF_OP_FLOW_DEL
;
1012 op
.u
.flow_del
.key
= key
;
1013 op
.u
.flow_del
.key_len
= key_len
;
1014 op
.u
.flow_del
.ufid
= ufid
;
1015 op
.u
.flow_del
.pmd_id
= pmd_id
;
1016 op
.u
.flow_del
.stats
= stats
;
1017 op
.u
.flow_del
.terse
= false;
1020 dpif_operate(dpif
, &opp
, 1);
1025 /* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
1026 * flows in 'dpif'. If 'terse' is true, then only UFID and statistics will
1027 * be returned in the dump. Otherwise, all fields will be returned.
1029 * This function always successfully returns a dpif_flow_dump. Error
1030 * reporting is deferred to dpif_flow_dump_destroy(). */
1031 struct dpif_flow_dump
*
1032 dpif_flow_dump_create(const struct dpif
*dpif
, bool terse
)
1034 return dpif
->dpif_class
->flow_dump_create(dpif
, terse
);
1037 /* Destroys 'dump', which must have been created with dpif_flow_dump_create().
1038 * All dpif_flow_dump_thread structures previously created for 'dump' must
1039 * previously have been destroyed.
1041 * Returns 0 if the dump operation was error-free, otherwise a positive errno
1042 * value describing the problem. */
1044 dpif_flow_dump_destroy(struct dpif_flow_dump
*dump
)
1046 const struct dpif
*dpif
= dump
->dpif
;
1047 int error
= dpif
->dpif_class
->flow_dump_destroy(dump
);
1048 log_operation(dpif
, "flow_dump_destroy", error
);
1049 return error
== EOF
? 0 : error
;
1052 /* Returns new thread-local state for use with dpif_flow_dump_next(). */
1053 struct dpif_flow_dump_thread
*
1054 dpif_flow_dump_thread_create(struct dpif_flow_dump
*dump
)
1056 return dump
->dpif
->dpif_class
->flow_dump_thread_create(dump
);
1059 /* Releases 'thread'. */
1061 dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread
*thread
)
1063 thread
->dpif
->dpif_class
->flow_dump_thread_destroy(thread
);
1066 /* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0
1067 * if and only if no flows remained to be retrieved, otherwise a positive
1068 * number reflecting the number of elements in 'flows[]' that were updated.
1069 * The number of flows returned might be less than 'max_flows' because
1070 * fewer than 'max_flows' remained, because this particular datapath does not
1071 * benefit from batching, or because an error occurred partway through
1072 * retrieval. Thus, the caller should continue calling until a 0 return value,
1073 * even if intermediate return values are less than 'max_flows'.
1075 * No error status is immediately provided. An error status for the entire
1076 * dump operation is provided when it is completed by calling
1077 * dpif_flow_dump_destroy().
1079 * All of the data stored into 'flows' is owned by the datapath, not by the
1080 * caller, and the caller must not modify or free it. The datapath guarantees
1081 * that it remains accessible and unchanged until the first of:
1082 * - The next call to dpif_flow_dump_next() for 'thread', or
1083 * - The next rcu quiescent period. */
1085 dpif_flow_dump_next(struct dpif_flow_dump_thread
*thread
,
1086 struct dpif_flow
*flows
, int max_flows
)
1088 struct dpif
*dpif
= thread
->dpif
;
1091 ovs_assert(max_flows
> 0);
1092 n
= dpif
->dpif_class
->flow_dump_next(thread
, flows
, max_flows
);
1094 struct dpif_flow
*f
;
1096 for (f
= flows
; f
< &flows
[n
] && should_log_flow_message(0); f
++) {
1097 log_flow_message(dpif
, 0, "flow_dump",
1098 f
->key
, f
->key_len
, f
->mask
, f
->mask_len
,
1099 &f
->ufid
, &f
->stats
, f
->actions
, f
->actions_len
);
1102 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all flows", dpif_name(dpif
));
1107 struct dpif_execute_helper_aux
{
1109 const struct flow
*flow
;
1111 const struct nlattr
*meter_action
; /* Non-NULL, if have a meter action. */
1114 /* This is called for actions that need the context of the datapath to be
1117 dpif_execute_helper_cb(void *aux_
, struct dp_packet_batch
*packets_
,
1118 const struct nlattr
*action
, bool may_steal
)
1120 struct dpif_execute_helper_aux
*aux
= aux_
;
1121 int type
= nl_attr_type(action
);
1122 struct dp_packet
*packet
= packets_
->packets
[0];
1124 ovs_assert(packets_
->count
== 1);
1126 switch ((enum ovs_action_attr
)type
) {
1127 case OVS_ACTION_ATTR_METER
:
1128 /* Maintain a pointer to the first meter action seen. */
1129 if (!aux
->meter_action
) {
1130 aux
->meter_action
= action
;
1134 case OVS_ACTION_ATTR_CT
:
1135 case OVS_ACTION_ATTR_OUTPUT
:
1136 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
1137 case OVS_ACTION_ATTR_TUNNEL_POP
:
1138 case OVS_ACTION_ATTR_USERSPACE
:
1139 case OVS_ACTION_ATTR_RECIRC
: {
1140 struct dpif_execute execute
;
1141 struct ofpbuf execute_actions
;
1142 uint64_t stub
[256 / 8];
1143 struct pkt_metadata
*md
= &packet
->md
;
1145 if (flow_tnl_dst_is_set(&md
->tunnel
) || aux
->meter_action
) {
1146 ofpbuf_use_stub(&execute_actions
, stub
, sizeof stub
);
1148 if (aux
->meter_action
) {
1149 const struct nlattr
*a
= aux
->meter_action
;
1151 /* XXX: This code collects meter actions since the last action
1152 * execution via the datapath to be executed right before the
1153 * current action that needs to be executed by the datapath.
1154 * This is only an approximation, but better than nothing.
1155 * Fundamentally, we should have a mechanism by which the
1156 * datapath could return the result of the meter action so that
1157 * we could execute them at the right order. */
1159 ofpbuf_put(&execute_actions
, a
, NLA_ALIGN(a
->nla_len
));
1160 /* Find next meter action before 'action', if any. */
1162 a
= nl_attr_next(a
);
1163 } while (a
!= action
&&
1164 nl_attr_type(a
) != OVS_ACTION_ATTR_METER
);
1165 } while (a
!= action
);
1168 /* The Linux kernel datapath throws away the tunnel information
1169 * that we supply as metadata. We have to use a "set" action to
1171 if (md
->tunnel
.ip_dst
) {
1172 odp_put_tunnel_action(&md
->tunnel
, &execute_actions
);
1174 ofpbuf_put(&execute_actions
, action
, NLA_ALIGN(action
->nla_len
));
1176 execute
.actions
= execute_actions
.data
;
1177 execute
.actions_len
= execute_actions
.size
;
1179 execute
.actions
= action
;
1180 execute
.actions_len
= NLA_ALIGN(action
->nla_len
);
1183 struct dp_packet
*clone
= NULL
;
1184 uint32_t cutlen
= dp_packet_get_cutlen(packet
);
1185 if (cutlen
&& (type
== OVS_ACTION_ATTR_OUTPUT
1186 || type
== OVS_ACTION_ATTR_TUNNEL_PUSH
1187 || type
== OVS_ACTION_ATTR_TUNNEL_POP
1188 || type
== OVS_ACTION_ATTR_USERSPACE
)) {
1189 dp_packet_reset_cutlen(packet
);
1191 packet
= clone
= dp_packet_clone(packet
);
1193 dp_packet_set_size(packet
, dp_packet_size(packet
) - cutlen
);
1196 execute
.packet
= packet
;
1197 execute
.flow
= aux
->flow
;
1198 execute
.needs_help
= false;
1199 execute
.probe
= false;
1201 aux
->error
= dpif_execute(aux
->dpif
, &execute
);
1202 log_execute_message(aux
->dpif
, &execute
, true, aux
->error
);
1204 dp_packet_delete(clone
);
1206 if (flow_tnl_dst_is_set(&md
->tunnel
) || aux
->meter_action
) {
1207 ofpbuf_uninit(&execute_actions
);
1209 /* Do not re-use the same meters for later output actions. */
1210 aux
->meter_action
= NULL
;
1215 case OVS_ACTION_ATTR_HASH
:
1216 case OVS_ACTION_ATTR_PUSH_VLAN
:
1217 case OVS_ACTION_ATTR_POP_VLAN
:
1218 case OVS_ACTION_ATTR_PUSH_MPLS
:
1219 case OVS_ACTION_ATTR_POP_MPLS
:
1220 case OVS_ACTION_ATTR_SET
:
1221 case OVS_ACTION_ATTR_SET_MASKED
:
1222 case OVS_ACTION_ATTR_SAMPLE
:
1223 case OVS_ACTION_ATTR_TRUNC
:
1224 case OVS_ACTION_ATTR_PUSH_ETH
:
1225 case OVS_ACTION_ATTR_POP_ETH
:
1226 case OVS_ACTION_ATTR_CLONE
:
1227 case OVS_ACTION_ATTR_UNSPEC
:
1228 case __OVS_ACTION_ATTR_MAX
:
1233 /* Executes 'execute' by performing most of the actions in userspace and
1234 * passing the fully constructed packets to 'dpif' for output and userspace
1237 * This helps with actions that a given 'dpif' doesn't implement directly. */
1239 dpif_execute_with_help(struct dpif
*dpif
, struct dpif_execute
*execute
)
1241 struct dpif_execute_helper_aux aux
= {dpif
, execute
->flow
, 0, NULL
};
1242 struct dp_packet_batch pb
;
1244 COVERAGE_INC(dpif_execute_with_help
);
1246 dp_packet_batch_init_packet(&pb
, execute
->packet
);
1247 odp_execute_actions(&aux
, &pb
, false, execute
->actions
,
1248 execute
->actions_len
, dpif_execute_helper_cb
);
1252 /* Returns true if the datapath needs help executing 'execute'. */
1254 dpif_execute_needs_help(const struct dpif_execute
*execute
)
1256 return execute
->needs_help
|| nl_attr_oversized(execute
->actions_len
);
1259 /* A dpif_operate() wrapper for performing a single DPIF_OP_EXECUTE. */
1261 dpif_execute(struct dpif
*dpif
, struct dpif_execute
*execute
)
1263 if (execute
->actions_len
) {
1264 struct dpif_op
*opp
;
1267 op
.type
= DPIF_OP_EXECUTE
;
1268 op
.u
.execute
= *execute
;
1271 dpif_operate(dpif
, &opp
, 1);
1279 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1280 * which they are specified. Places each operation's results in the "output"
1281 * members documented in comments, and 0 in the 'error' member on success or a
1282 * positive errno on failure. */
1284 dpif_operate(struct dpif
*dpif
, struct dpif_op
**ops
, size_t n_ops
)
1289 /* Count 'chunk', the number of ops that can be executed without
1290 * needing any help. Ops that need help should be rare, so we
1291 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1292 for (chunk
= 0; chunk
< n_ops
; chunk
++) {
1293 struct dpif_op
*op
= ops
[chunk
];
1295 if (op
->type
== DPIF_OP_EXECUTE
1296 && dpif_execute_needs_help(&op
->u
.execute
)) {
1302 /* Execute a chunk full of ops that the dpif provider can
1303 * handle itself, without help. */
1306 dpif
->dpif_class
->operate(dpif
, ops
, chunk
);
1308 for (i
= 0; i
< chunk
; i
++) {
1309 struct dpif_op
*op
= ops
[i
];
1310 int error
= op
->error
;
1313 case DPIF_OP_FLOW_PUT
: {
1314 struct dpif_flow_put
*put
= &op
->u
.flow_put
;
1316 COVERAGE_INC(dpif_flow_put
);
1317 log_flow_put_message(dpif
, put
, error
);
1318 if (error
&& put
->stats
) {
1319 memset(put
->stats
, 0, sizeof *put
->stats
);
1324 case DPIF_OP_FLOW_GET
: {
1325 struct dpif_flow_get
*get
= &op
->u
.flow_get
;
1327 COVERAGE_INC(dpif_flow_get
);
1329 memset(get
->flow
, 0, sizeof *get
->flow
);
1331 log_flow_get_message(dpif
, get
, error
);
1336 case DPIF_OP_FLOW_DEL
: {
1337 struct dpif_flow_del
*del
= &op
->u
.flow_del
;
1339 COVERAGE_INC(dpif_flow_del
);
1340 log_flow_del_message(dpif
, del
, error
);
1341 if (error
&& del
->stats
) {
1342 memset(del
->stats
, 0, sizeof *del
->stats
);
1347 case DPIF_OP_EXECUTE
:
1348 COVERAGE_INC(dpif_execute
);
1349 log_execute_message(dpif
, &op
->u
.execute
, false, error
);
1357 /* Help the dpif provider to execute one op. */
1358 struct dpif_op
*op
= ops
[0];
1360 COVERAGE_INC(dpif_execute
);
1361 op
->error
= dpif_execute_with_help(dpif
, &op
->u
.execute
);
1368 /* Returns a string that represents 'type', for use in log messages. */
1370 dpif_upcall_type_to_string(enum dpif_upcall_type type
)
1373 case DPIF_UC_MISS
: return "miss";
1374 case DPIF_UC_ACTION
: return "action";
1375 case DPIF_N_UC_TYPES
: default: return "<unknown>";
1379 /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1380 * if successful, otherwise a positive errno value.
1382 * Turning packet receive off and then back on may change the Netlink PID
1383 * assignments returned by dpif_port_get_pid(). If the client does this, it
1384 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1385 * using the new PID assignment. */
1387 dpif_recv_set(struct dpif
*dpif
, bool enable
)
1391 if (dpif
->dpif_class
->recv_set
) {
1392 error
= dpif
->dpif_class
->recv_set(dpif
, enable
);
1393 log_operation(dpif
, "recv_set", error
);
1398 /* Refreshes the poll loops and Netlink sockets associated to each port,
1399 * when the number of upcall handlers (upcall receiving thread) is changed
1400 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
1403 * Since multiple upcall handlers can read upcalls simultaneously from
1404 * 'dpif', each port can have multiple Netlink sockets, one per upcall
1405 * handler. So, handlers_set() is responsible for the following tasks:
1407 * When receiving upcall is enabled, extends or creates the
1408 * configuration to support:
1410 * - 'n_handlers' Netlink sockets for each port.
1412 * - 'n_handlers' poll loops, one for each upcall handler.
1414 * - registering the Netlink sockets for the same upcall handler to
1415 * the corresponding poll loop.
1417 * Returns 0 if successful, otherwise a positive errno value. */
1419 dpif_handlers_set(struct dpif
*dpif
, uint32_t n_handlers
)
1423 if (dpif
->dpif_class
->handlers_set
) {
1424 error
= dpif
->dpif_class
->handlers_set(dpif
, n_handlers
);
1425 log_operation(dpif
, "handlers_set", error
);
1431 dpif_register_dp_purge_cb(struct dpif
*dpif
, dp_purge_callback
*cb
, void *aux
)
1433 if (dpif
->dpif_class
->register_dp_purge_cb
) {
1434 dpif
->dpif_class
->register_dp_purge_cb(dpif
, cb
, aux
);
1439 dpif_register_upcall_cb(struct dpif
*dpif
, upcall_callback
*cb
, void *aux
)
1441 if (dpif
->dpif_class
->register_upcall_cb
) {
1442 dpif
->dpif_class
->register_upcall_cb(dpif
, cb
, aux
);
1447 dpif_enable_upcall(struct dpif
*dpif
)
1449 if (dpif
->dpif_class
->enable_upcall
) {
1450 dpif
->dpif_class
->enable_upcall(dpif
);
1455 dpif_disable_upcall(struct dpif
*dpif
)
1457 if (dpif
->dpif_class
->disable_upcall
) {
1458 dpif
->dpif_class
->disable_upcall(dpif
);
1463 dpif_print_packet(struct dpif
*dpif
, struct dpif_upcall
*upcall
)
1465 if (!VLOG_DROP_DBG(&dpmsg_rl
)) {
1469 packet
= ofp_packet_to_string(dp_packet_data(&upcall
->packet
),
1470 dp_packet_size(&upcall
->packet
));
1473 odp_flow_key_format(upcall
->key
, upcall
->key_len
, &flow
);
1475 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1476 dpif_name(dpif
), dpif_upcall_type_to_string(upcall
->type
),
1477 ds_cstr(&flow
), packet
);
1484 /* Pass custom configuration to the datapath implementation. Some of the
1485 * changes can be postponed until dpif_run() is called. */
1487 dpif_set_config(struct dpif
*dpif
, const struct smap
*cfg
)
1491 if (dpif
->dpif_class
->set_config
) {
1492 error
= dpif
->dpif_class
->set_config(dpif
, cfg
);
1494 log_operation(dpif
, "set_config", error
);
1501 /* Polls for an upcall from 'dpif' for an upcall handler. Since there
1502 * there can be multiple poll loops, 'handler_id' is needed as index to
1503 * identify the corresponding poll loop. If successful, stores the upcall
1504 * into '*upcall', using 'buf' for storage. Should only be called if
1505 * 'recv_set' has been used to enable receiving packets from 'dpif'.
1507 * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
1508 * 'buf', so their memory cannot be freed separately from 'buf'.
1510 * The caller owns the data of 'upcall->packet' and may modify it. If
1511 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
1512 * will be reallocated. This requires the data of 'upcall->packet' to be
1513 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
1514 * when an error is returned, the 'upcall->packet' may be uninitialized
1515 * and should not be released.
1517 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1518 * if no upcall is immediately available. */
1520 dpif_recv(struct dpif
*dpif
, uint32_t handler_id
, struct dpif_upcall
*upcall
,
1525 if (dpif
->dpif_class
->recv
) {
1526 error
= dpif
->dpif_class
->recv(dpif
, handler_id
, upcall
, buf
);
1528 dpif_print_packet(dpif
, upcall
);
1529 } else if (error
!= EAGAIN
) {
1530 log_operation(dpif
, "recv", error
);
1536 /* Discards all messages that would otherwise be received by dpif_recv() on
1539 dpif_recv_purge(struct dpif
*dpif
)
1541 COVERAGE_INC(dpif_purge
);
1542 if (dpif
->dpif_class
->recv_purge
) {
1543 dpif
->dpif_class
->recv_purge(dpif
);
1547 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
1548 * 'dpif' has a message queued to be received with the recv member
1549 * function. Since there can be multiple poll loops, 'handler_id' is
1550 * needed as index to identify the corresponding poll loop. */
1552 dpif_recv_wait(struct dpif
*dpif
, uint32_t handler_id
)
1554 if (dpif
->dpif_class
->recv_wait
) {
1555 dpif
->dpif_class
->recv_wait(dpif
, handler_id
);
1560 * Return the datapath version. Caller is responsible for freeing
1564 dpif_get_dp_version(const struct dpif
*dpif
)
1566 char *version
= NULL
;
1568 if (dpif
->dpif_class
->get_datapath_version
) {
1569 version
= dpif
->dpif_class
->get_datapath_version();
1575 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1576 * and '*engine_id', respectively. */
1578 dpif_get_netflow_ids(const struct dpif
*dpif
,
1579 uint8_t *engine_type
, uint8_t *engine_id
)
1581 *engine_type
= dpif
->netflow_engine_type
;
1582 *engine_id
= dpif
->netflow_engine_id
;
1585 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1586 * value used for setting packet priority.
1587 * On success, returns 0 and stores the priority into '*priority'.
1588 * On failure, returns a positive errno value and stores 0 into '*priority'. */
1590 dpif_queue_to_priority(const struct dpif
*dpif
, uint32_t queue_id
,
1593 int error
= (dpif
->dpif_class
->queue_to_priority
1594 ? dpif
->dpif_class
->queue_to_priority(dpif
, queue_id
,
1600 log_operation(dpif
, "queue_to_priority", error
);
1605 dpif_init(struct dpif
*dpif
, const struct dpif_class
*dpif_class
,
1607 uint8_t netflow_engine_type
, uint8_t netflow_engine_id
)
1609 dpif
->dpif_class
= dpif_class
;
1610 dpif
->base_name
= xstrdup(name
);
1611 dpif
->full_name
= xasprintf("%s@%s", dpif_class
->type
, name
);
1612 dpif
->netflow_engine_type
= netflow_engine_type
;
1613 dpif
->netflow_engine_id
= netflow_engine_id
;
1616 /* Undoes the results of initialization.
1618 * Normally this function only needs to be called from dpif_close().
1619 * However, it may be called by providers due to an error on opening
1620 * that occurs after initialization. It this case dpif_close() would
1621 * never be called. */
1623 dpif_uninit(struct dpif
*dpif
, bool close
)
1625 char *base_name
= dpif
->base_name
;
1626 char *full_name
= dpif
->full_name
;
1629 dpif
->dpif_class
->close(dpif
);
1637 log_operation(const struct dpif
*dpif
, const char *operation
, int error
)
1640 VLOG_DBG_RL(&dpmsg_rl
, "%s: %s success", dpif_name(dpif
), operation
);
1641 } else if (ofperr_is_valid(error
)) {
1642 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1643 dpif_name(dpif
), operation
, ofperr_get_name(error
));
1645 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1646 dpif_name(dpif
), operation
, ovs_strerror(error
));
1650 static enum vlog_level
1651 flow_message_log_level(int error
)
1653 /* If flows arrive in a batch, userspace may push down multiple
1654 * unique flow definitions that overlap when wildcards are applied.
1655 * Kernels that support flow wildcarding will reject these flows as
1656 * duplicates (EEXIST), so lower the log level to debug for these
1657 * types of messages. */
1658 return (error
&& error
!= EEXIST
) ? VLL_WARN
: VLL_DBG
;
1662 should_log_flow_message(int error
)
1664 return !vlog_should_drop(&this_module
, flow_message_log_level(error
),
1665 error
? &error_rl
: &dpmsg_rl
);
1669 log_flow_message(const struct dpif
*dpif
, int error
, const char *operation
,
1670 const struct nlattr
*key
, size_t key_len
,
1671 const struct nlattr
*mask
, size_t mask_len
,
1672 const ovs_u128
*ufid
, const struct dpif_flow_stats
*stats
,
1673 const struct nlattr
*actions
, size_t actions_len
)
1675 struct ds ds
= DS_EMPTY_INITIALIZER
;
1676 ds_put_format(&ds
, "%s: ", dpif_name(dpif
));
1678 ds_put_cstr(&ds
, "failed to ");
1680 ds_put_format(&ds
, "%s ", operation
);
1682 ds_put_format(&ds
, "(%s) ", ovs_strerror(error
));
1685 odp_format_ufid(ufid
, &ds
);
1686 ds_put_cstr(&ds
, " ");
1688 odp_flow_format(key
, key_len
, mask
, mask_len
, NULL
, &ds
, true);
1690 ds_put_cstr(&ds
, ", ");
1691 dpif_flow_stats_format(stats
, &ds
);
1693 if (actions
|| actions_len
) {
1694 ds_put_cstr(&ds
, ", actions:");
1695 format_odp_actions(&ds
, actions
, actions_len
);
1697 vlog(&this_module
, flow_message_log_level(error
), "%s", ds_cstr(&ds
));
1702 log_flow_put_message(struct dpif
*dpif
, const struct dpif_flow_put
*put
,
1705 if (should_log_flow_message(error
) && !(put
->flags
& DPIF_FP_PROBE
)) {
1709 ds_put_cstr(&s
, "put");
1710 if (put
->flags
& DPIF_FP_CREATE
) {
1711 ds_put_cstr(&s
, "[create]");
1713 if (put
->flags
& DPIF_FP_MODIFY
) {
1714 ds_put_cstr(&s
, "[modify]");
1716 if (put
->flags
& DPIF_FP_ZERO_STATS
) {
1717 ds_put_cstr(&s
, "[zero]");
1719 log_flow_message(dpif
, error
, ds_cstr(&s
),
1720 put
->key
, put
->key_len
, put
->mask
, put
->mask_len
,
1721 put
->ufid
, put
->stats
, put
->actions
,
1728 log_flow_del_message(struct dpif
*dpif
, const struct dpif_flow_del
*del
,
1731 if (should_log_flow_message(error
)) {
1732 log_flow_message(dpif
, error
, "flow_del", del
->key
, del
->key_len
,
1733 NULL
, 0, del
->ufid
, !error
? del
->stats
: NULL
,
1738 /* Logs that 'execute' was executed on 'dpif' and completed with errno 'error'
1739 * (0 for success). 'subexecute' should be true if the execution is a result
1740 * of breaking down a larger execution that needed help, false otherwise.
1743 * XXX In theory, the log message could be deceptive because this function is
1744 * called after the dpif_provider's '->execute' function, which is allowed to
1745 * modify execute->packet and execute->md. In practice, though:
1747 * - dpif-netlink doesn't modify execute->packet or execute->md.
1749 * - dpif-netdev does modify them but it is less likely to have problems
1750 * because it is built into ovs-vswitchd and cannot have version skew,
1753 * It would still be better to avoid the potential problem. I don't know of a
1754 * good way to do that, though, that isn't expensive. */
1756 log_execute_message(struct dpif
*dpif
, const struct dpif_execute
*execute
,
1757 bool subexecute
, int error
)
1759 if (!(error
? VLOG_DROP_WARN(&error_rl
) : VLOG_DROP_DBG(&dpmsg_rl
))
1760 && !execute
->probe
) {
1761 struct ds ds
= DS_EMPTY_INITIALIZER
;
1764 packet
= ofp_packet_to_string(dp_packet_data(execute
->packet
),
1765 dp_packet_size(execute
->packet
));
1766 ds_put_format(&ds
, "%s: %sexecute ",
1768 (subexecute
? "sub-"
1769 : dpif_execute_needs_help(execute
) ? "super-"
1771 format_odp_actions(&ds
, execute
->actions
, execute
->actions_len
);
1773 ds_put_format(&ds
, " failed (%s)", ovs_strerror(error
));
1775 ds_put_format(&ds
, " on packet %s", packet
);
1776 ds_put_format(&ds
, " mtu %d", execute
->mtu
);
1777 vlog(&this_module
, error
? VLL_WARN
: VLL_DBG
, "%s", ds_cstr(&ds
));
1784 log_flow_get_message(const struct dpif
*dpif
, const struct dpif_flow_get
*get
,
1787 if (should_log_flow_message(error
)) {
1788 log_flow_message(dpif
, error
, "flow_get",
1789 get
->key
, get
->key_len
,
1790 get
->flow
->mask
, get
->flow
->mask_len
,
1791 get
->ufid
, &get
->flow
->stats
,
1792 get
->flow
->actions
, get
->flow
->actions_len
);
1797 dpif_supports_tnl_push_pop(const struct dpif
*dpif
)
1799 return dpif_is_netdev(dpif
);
1804 dpif_meter_get_features(const struct dpif
*dpif
,
1805 struct ofputil_meter_features
*features
)
1807 memset(features
, 0, sizeof *features
);
1808 if (dpif
->dpif_class
->meter_get_features
) {
1809 dpif
->dpif_class
->meter_get_features(dpif
, features
);
1813 /* Adds or modifies meter identified by 'meter_id' in 'dpif'. If '*meter_id'
1814 * is UINT32_MAX, adds a new meter, otherwise modifies an existing meter.
1816 * If meter is successfully added, sets '*meter_id' to the new meter's
1819 dpif_meter_set(struct dpif
*dpif
, ofproto_meter_id
*meter_id
,
1820 struct ofputil_meter_config
*config
)
1824 COVERAGE_INC(dpif_meter_set
);
1826 error
= dpif
->dpif_class
->meter_set(dpif
, meter_id
, config
);
1828 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" set",
1829 dpif_name(dpif
), meter_id
->uint32
);
1831 VLOG_WARN_RL(&error_rl
, "%s: failed to set DPIF meter %"PRIu32
": %s",
1832 dpif_name(dpif
), meter_id
->uint32
, ovs_strerror(error
));
1833 meter_id
->uint32
= UINT32_MAX
;
1839 dpif_meter_get(const struct dpif
*dpif
, ofproto_meter_id meter_id
,
1840 struct ofputil_meter_stats
*stats
, uint16_t n_bands
)
1844 COVERAGE_INC(dpif_meter_get
);
1846 error
= dpif
->dpif_class
->meter_get(dpif
, meter_id
, stats
, n_bands
);
1848 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" get stats",
1849 dpif_name(dpif
), meter_id
.uint32
);
1851 VLOG_WARN_RL(&error_rl
,
1852 "%s: failed to get DPIF meter %"PRIu32
" stats: %s",
1853 dpif_name(dpif
), meter_id
.uint32
, ovs_strerror(error
));
1854 stats
->packet_in_count
= ~0;
1855 stats
->byte_in_count
= ~0;
1862 dpif_meter_del(struct dpif
*dpif
, ofproto_meter_id meter_id
,
1863 struct ofputil_meter_stats
*stats
, uint16_t n_bands
)
1867 COVERAGE_INC(dpif_meter_del
);
1869 error
= dpif
->dpif_class
->meter_del(dpif
, meter_id
, stats
, n_bands
);
1871 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" deleted",
1872 dpif_name(dpif
), meter_id
.uint32
);
1874 VLOG_WARN_RL(&error_rl
,
1875 "%s: failed to delete DPIF meter %"PRIu32
": %s",
1876 dpif_name(dpif
), meter_id
.uint32
, ovs_strerror(error
));
1878 stats
->packet_in_count
= ~0;
1879 stats
->byte_in_count
= ~0;