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/ofpbuf.h"
39 #include "openvswitch/poll-loop.h"
40 #include "route-table.h"
42 #include "openvswitch/shash.h"
45 #include "tnl-neigh-cache.h"
46 #include "tnl-ports.h"
50 #include "openvswitch/ofp-errors.h"
51 #include "openvswitch/vlog.h"
52 #include "lib/netdev-provider.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_operation(const struct dpif
*, const char *operation
,
97 static bool should_log_flow_message(const struct vlog_module
*module
,
100 /* Incremented whenever tnl route, arp, etc changes. */
101 struct seq
*tnl_conf_seq
;
104 dpif_is_internal_port(const char *type
)
106 /* For userspace datapath, tap devices are the equivalent
107 * of internal devices in the kernel datapath, so both
108 * these types are 'internal' devices. */
109 return !strcmp(type
, "internal") || !strcmp(type
, "tap");
115 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
117 if (ovsthread_once_start(&once
)) {
120 tnl_conf_seq
= seq_create();
121 dpctl_unixctl_register();
123 tnl_neigh_cache_init();
126 for (i
= 0; i
< ARRAY_SIZE(base_dpif_classes
); i
++) {
127 dp_register_provider(base_dpif_classes
[i
]);
130 ovsthread_once_done(&once
);
135 dp_register_provider__(const struct dpif_class
*new_class
)
137 struct registered_dpif_class
*registered_class
;
140 if (sset_contains(&dpif_blacklist
, new_class
->type
)) {
141 VLOG_DBG("attempted to register blacklisted provider: %s",
146 if (shash_find(&dpif_classes
, new_class
->type
)) {
147 VLOG_WARN("attempted to register duplicate datapath provider: %s",
152 error
= new_class
->init
? new_class
->init() : 0;
154 VLOG_WARN("failed to initialize %s datapath class: %s",
155 new_class
->type
, ovs_strerror(error
));
159 registered_class
= xmalloc(sizeof *registered_class
);
160 registered_class
->dpif_class
= new_class
;
161 registered_class
->refcount
= 0;
163 shash_add(&dpif_classes
, new_class
->type
, registered_class
);
168 /* Registers a new datapath provider. After successful registration, new
169 * datapaths of that type can be opened using dpif_open(). */
171 dp_register_provider(const struct dpif_class
*new_class
)
175 ovs_mutex_lock(&dpif_mutex
);
176 error
= dp_register_provider__(new_class
);
177 ovs_mutex_unlock(&dpif_mutex
);
182 /* Unregisters a datapath provider. 'type' must have been previously
183 * registered and not currently be in use by any dpifs. After unregistration
184 * new datapaths of that type cannot be opened using dpif_open(). */
186 dp_unregister_provider__(const char *type
)
188 struct shash_node
*node
;
189 struct registered_dpif_class
*registered_class
;
191 node
= shash_find(&dpif_classes
, type
);
196 registered_class
= node
->data
;
197 if (registered_class
->refcount
) {
198 VLOG_WARN("attempted to unregister in use datapath provider: %s", type
);
202 shash_delete(&dpif_classes
, node
);
203 free(registered_class
);
208 /* Unregisters a datapath provider. 'type' must have been previously
209 * registered and not currently be in use by any dpifs. After unregistration
210 * new datapaths of that type cannot be opened using dpif_open(). */
212 dp_unregister_provider(const char *type
)
218 ovs_mutex_lock(&dpif_mutex
);
219 error
= dp_unregister_provider__(type
);
220 ovs_mutex_unlock(&dpif_mutex
);
225 /* Blacklists a provider. Causes future calls of dp_register_provider() with
226 * a dpif_class which implements 'type' to fail. */
228 dp_blacklist_provider(const char *type
)
230 ovs_mutex_lock(&dpif_mutex
);
231 sset_add(&dpif_blacklist
, type
);
232 ovs_mutex_unlock(&dpif_mutex
);
235 /* Adds the types of all currently registered datapath providers to 'types'.
236 * The caller must first initialize the sset. */
238 dp_enumerate_types(struct sset
*types
)
240 struct shash_node
*node
;
244 ovs_mutex_lock(&dpif_mutex
);
245 SHASH_FOR_EACH(node
, &dpif_classes
) {
246 const struct registered_dpif_class
*registered_class
= node
->data
;
247 sset_add(types
, registered_class
->dpif_class
->type
);
249 ovs_mutex_unlock(&dpif_mutex
);
253 dp_class_unref(struct registered_dpif_class
*rc
)
255 ovs_mutex_lock(&dpif_mutex
);
256 ovs_assert(rc
->refcount
);
258 ovs_mutex_unlock(&dpif_mutex
);
261 static struct registered_dpif_class
*
262 dp_class_lookup(const char *type
)
264 struct registered_dpif_class
*rc
;
266 ovs_mutex_lock(&dpif_mutex
);
267 rc
= shash_find_data(&dpif_classes
, type
);
271 ovs_mutex_unlock(&dpif_mutex
);
276 /* Clears 'names' and enumerates the names of all known created datapaths with
277 * the given 'type'. The caller must first initialize the sset. Returns 0 if
278 * successful, otherwise a positive errno value.
280 * Some kinds of datapaths might not be practically enumerable. This is not
281 * considered an error. */
283 dp_enumerate_names(const char *type
, struct sset
*names
)
285 struct registered_dpif_class
*registered_class
;
286 const struct dpif_class
*dpif_class
;
292 registered_class
= dp_class_lookup(type
);
293 if (!registered_class
) {
294 VLOG_WARN("could not enumerate unknown type: %s", type
);
298 dpif_class
= registered_class
->dpif_class
;
299 error
= (dpif_class
->enumerate
300 ? dpif_class
->enumerate(names
, dpif_class
)
303 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class
->type
,
304 ovs_strerror(error
));
306 dp_class_unref(registered_class
);
311 /* Parses 'datapath_name_', which is of the form [type@]name into its
312 * component pieces. 'name' and 'type' must be freed by the caller.
314 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
316 dp_parse_name(const char *datapath_name_
, char **name
, char **type
)
318 char *datapath_name
= xstrdup(datapath_name_
);
321 separator
= strchr(datapath_name
, '@');
324 *type
= datapath_name
;
325 *name
= xstrdup(dpif_normalize_type(separator
+ 1));
327 *name
= datapath_name
;
328 *type
= xstrdup(dpif_normalize_type(NULL
));
333 do_open(const char *name
, const char *type
, bool create
, struct dpif
**dpifp
)
335 struct dpif
*dpif
= NULL
;
337 struct registered_dpif_class
*registered_class
;
341 type
= dpif_normalize_type(type
);
342 registered_class
= dp_class_lookup(type
);
343 if (!registered_class
) {
344 VLOG_WARN("could not create datapath %s of unknown type %s", name
,
346 error
= EAFNOSUPPORT
;
350 error
= registered_class
->dpif_class
->open(registered_class
->dpif_class
,
351 name
, create
, &dpif
);
353 struct dpif_port_dump port_dump
;
354 struct dpif_port dpif_port
;
356 ovs_assert(dpif
->dpif_class
== registered_class
->dpif_class
);
358 DPIF_PORT_FOR_EACH(&dpif_port
, &port_dump
, dpif
) {
359 struct netdev
*netdev
;
362 if (dpif_is_internal_port(dpif_port
.type
)) {
366 err
= netdev_open(dpif_port
.name
, dpif_port
.type
, &netdev
);
369 netdev_ports_insert(netdev
, dpif
->dpif_class
, &dpif_port
);
370 netdev_close(netdev
);
372 VLOG_WARN("could not open netdev %s type %s: %s",
373 dpif_port
.name
, dpif_port
.type
, ovs_strerror(err
));
377 dp_class_unref(registered_class
);
381 *dpifp
= error
? NULL
: dpif
;
385 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
386 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
387 * the empty string to specify the default system type. Returns 0 if
388 * successful, otherwise a positive errno value. On success stores a pointer
389 * to the datapath in '*dpifp', otherwise a null pointer. */
391 dpif_open(const char *name
, const char *type
, struct dpif
**dpifp
)
393 return do_open(name
, type
, false, dpifp
);
396 /* Tries to create and open a new datapath with the given 'name' and 'type'.
397 * 'type' may be either NULL or the empty string to specify the default system
398 * type. Will fail if a datapath with 'name' and 'type' already exists.
399 * Returns 0 if successful, otherwise a positive errno value. On success
400 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
402 dpif_create(const char *name
, const char *type
, struct dpif
**dpifp
)
404 return do_open(name
, type
, true, dpifp
);
407 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
408 * does not exist. 'type' may be either NULL or the empty string to specify
409 * the default system type. Returns 0 if successful, otherwise a positive
410 * errno value. On success stores a pointer to the datapath in '*dpifp',
411 * otherwise a null pointer. */
413 dpif_create_and_open(const char *name
, const char *type
, struct dpif
**dpifp
)
417 error
= dpif_create(name
, type
, dpifp
);
418 if (error
== EEXIST
|| error
== EBUSY
) {
419 error
= dpif_open(name
, type
, dpifp
);
421 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
422 name
, ovs_strerror(error
));
425 VLOG_WARN("failed to create datapath %s: %s",
426 name
, ovs_strerror(error
));
432 dpif_remove_netdev_ports(struct dpif
*dpif
) {
433 struct dpif_port_dump port_dump
;
434 struct dpif_port dpif_port
;
436 DPIF_PORT_FOR_EACH (&dpif_port
, &port_dump
, dpif
) {
437 if (!dpif_is_internal_port(dpif_port
.type
)) {
438 netdev_ports_remove(dpif_port
.port_no
, dpif
->dpif_class
);
443 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
444 * itself; call dpif_delete() first, instead, if that is desirable. */
446 dpif_close(struct dpif
*dpif
)
449 struct registered_dpif_class
*rc
;
451 rc
= shash_find_data(&dpif_classes
, dpif
->dpif_class
->type
);
453 if (rc
->refcount
== 1) {
454 dpif_remove_netdev_ports(dpif
);
456 dpif_uninit(dpif
, true);
461 /* Performs periodic work needed by 'dpif'. */
463 dpif_run(struct dpif
*dpif
)
465 if (dpif
->dpif_class
->run
) {
466 return dpif
->dpif_class
->run(dpif
);
471 /* Arranges for poll_block() to wake up when dp_run() needs to be called for
474 dpif_wait(struct dpif
*dpif
)
476 if (dpif
->dpif_class
->wait
) {
477 dpif
->dpif_class
->wait(dpif
);
481 /* Returns the name of datapath 'dpif' prefixed with the type
482 * (for use in log messages). */
484 dpif_name(const struct dpif
*dpif
)
486 return dpif
->full_name
;
489 /* Returns the name of datapath 'dpif' without the type
490 * (for use in device names). */
492 dpif_base_name(const struct dpif
*dpif
)
494 return dpif
->base_name
;
497 /* Returns the type of datapath 'dpif'. */
499 dpif_type(const struct dpif
*dpif
)
501 return dpif
->dpif_class
->type
;
504 /* Returns the fully spelled out name for the given datapath 'type'.
506 * Normalized type string can be compared with strcmp(). Unnormalized type
507 * string might be the same even if they have different spellings. */
509 dpif_normalize_type(const char *type
)
511 return type
&& type
[0] ? type
: "system";
514 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
515 * ports. After calling this function, it does not make sense to pass 'dpif'
516 * to any functions other than dpif_name() or dpif_close(). */
518 dpif_delete(struct dpif
*dpif
)
522 COVERAGE_INC(dpif_destroy
);
524 error
= dpif
->dpif_class
->destroy(dpif
);
525 log_operation(dpif
, "delete", error
);
529 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
530 * otherwise a positive errno value. */
532 dpif_get_dp_stats(const struct dpif
*dpif
, struct dpif_dp_stats
*stats
)
534 int error
= dpif
->dpif_class
->get_stats(dpif
, stats
);
536 memset(stats
, 0, sizeof *stats
);
538 log_operation(dpif
, "get_stats", error
);
543 dpif_port_open_type(const char *datapath_type
, const char *port_type
)
545 struct registered_dpif_class
*rc
;
547 datapath_type
= dpif_normalize_type(datapath_type
);
549 ovs_mutex_lock(&dpif_mutex
);
550 rc
= shash_find_data(&dpif_classes
, datapath_type
);
551 if (rc
&& rc
->dpif_class
->port_open_type
) {
552 port_type
= rc
->dpif_class
->port_open_type(rc
->dpif_class
, port_type
);
554 ovs_mutex_unlock(&dpif_mutex
);
559 /* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
560 * non-null and its value is not ODPP_NONE, then attempts to use the
561 * value as the port number.
563 * If successful, returns 0 and sets '*port_nop' to the new port's port
564 * number (if 'port_nop' is non-null). On failure, returns a positive
565 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
568 dpif_port_add(struct dpif
*dpif
, struct netdev
*netdev
, odp_port_t
*port_nop
)
570 const char *netdev_name
= netdev_get_name(netdev
);
571 odp_port_t port_no
= ODPP_NONE
;
574 COVERAGE_INC(dpif_port_add
);
580 error
= dpif
->dpif_class
->port_add(dpif
, netdev
, &port_no
);
582 VLOG_DBG_RL(&dpmsg_rl
, "%s: added %s as port %"PRIu32
,
583 dpif_name(dpif
), netdev_name
, port_no
);
585 if (!dpif_is_internal_port(netdev_get_type(netdev
))) {
587 struct dpif_port dpif_port
;
589 dpif_port
.type
= CONST_CAST(char *, netdev_get_type(netdev
));
590 dpif_port
.name
= CONST_CAST(char *, netdev_name
);
591 dpif_port
.port_no
= port_no
;
592 netdev_ports_insert(netdev
, dpif
->dpif_class
, &dpif_port
);
595 if (error
!= EEXIST
) {
596 VLOG_WARN_RL(&error_rl
, "%s: failed to add %s as port: %s",
597 dpif_name(dpif
), netdev_name
, ovs_strerror(error
));
599 /* It's fairly common for upper layers to try to add a duplicate
600 * port, and they know how to handle it properly. */
610 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
611 * otherwise a positive errno value. */
613 dpif_port_del(struct dpif
*dpif
, odp_port_t port_no
, bool local_delete
)
617 COVERAGE_INC(dpif_port_del
);
620 error
= dpif
->dpif_class
->port_del(dpif
, port_no
);
622 VLOG_DBG_RL(&dpmsg_rl
, "%s: port_del(%"PRIu32
")",
623 dpif_name(dpif
), port_no
);
625 log_operation(dpif
, "port_del", error
);
629 netdev_ports_remove(port_no
, dpif
->dpif_class
);
633 /* Makes a deep copy of 'src' into 'dst'. */
635 dpif_port_clone(struct dpif_port
*dst
, const struct dpif_port
*src
)
637 dst
->name
= xstrdup(src
->name
);
638 dst
->type
= xstrdup(src
->type
);
639 dst
->port_no
= src
->port_no
;
642 /* Frees memory allocated to members of 'dpif_port'.
644 * Do not call this function on a dpif_port obtained from
645 * dpif_port_dump_next(): that function retains ownership of the data in the
648 dpif_port_destroy(struct dpif_port
*dpif_port
)
650 free(dpif_port
->name
);
651 free(dpif_port
->type
);
654 /* Checks if port named 'devname' exists in 'dpif'. If so, returns
655 * true; otherwise, returns false. */
657 dpif_port_exists(const struct dpif
*dpif
, const char *devname
)
659 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, NULL
);
660 if (error
!= 0 && error
!= ENODEV
) {
661 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %s: %s",
662 dpif_name(dpif
), devname
, ovs_strerror(error
));
668 /* Refreshes configuration of 'dpif's port. */
670 dpif_port_set_config(struct dpif
*dpif
, odp_port_t port_no
,
671 const struct smap
*cfg
)
675 if (dpif
->dpif_class
->port_set_config
) {
676 error
= dpif
->dpif_class
->port_set_config(dpif
, port_no
, cfg
);
678 log_operation(dpif
, "port_set_config", error
);
685 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
686 * initializes '*port' appropriately; on failure, returns a positive errno
689 * Retuns ENODEV if the port doesn't exist.
691 * The caller owns the data in 'port' and must free it with
692 * dpif_port_destroy() when it is no longer needed. */
694 dpif_port_query_by_number(const struct dpif
*dpif
, odp_port_t port_no
,
695 struct dpif_port
*port
)
697 int error
= dpif
->dpif_class
->port_query_by_number(dpif
, port_no
, port
);
699 VLOG_DBG_RL(&dpmsg_rl
, "%s: port %"PRIu32
" is device %s",
700 dpif_name(dpif
), port_no
, port
->name
);
702 memset(port
, 0, sizeof *port
);
703 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %"PRIu32
": %s",
704 dpif_name(dpif
), port_no
, ovs_strerror(error
));
709 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
710 * initializes '*port' appropriately; on failure, returns a positive errno
713 * Retuns ENODEV if the port doesn't exist.
715 * The caller owns the data in 'port' and must free it with
716 * dpif_port_destroy() when it is no longer needed. */
718 dpif_port_query_by_name(const struct dpif
*dpif
, const char *devname
,
719 struct dpif_port
*port
)
721 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, port
);
723 VLOG_DBG_RL(&dpmsg_rl
, "%s: device %s is on port %"PRIu32
,
724 dpif_name(dpif
), devname
, port
->port_no
);
726 memset(port
, 0, sizeof *port
);
728 /* For ENODEV we use DBG level because the caller is probably
729 * interested in whether 'dpif' actually has a port 'devname', so that
730 * it's not an issue worth logging if it doesn't. Other errors are
731 * uncommon and more likely to indicate a real problem. */
732 VLOG_RL(&error_rl
, error
== ENODEV
? VLL_DBG
: VLL_WARN
,
733 "%s: failed to query port %s: %s",
734 dpif_name(dpif
), devname
, ovs_strerror(error
));
739 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
740 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
741 * flows whose packets arrived on port 'port_no'.
743 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
744 * allocated to any port, that the client may use for special purposes.
746 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
747 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
748 * disabled and then re-enabled, so a client that does that must be prepared to
749 * update all of the flows that it installed that contain
750 * OVS_ACTION_ATTR_USERSPACE actions. */
752 dpif_port_get_pid(const struct dpif
*dpif
, odp_port_t port_no
)
754 return (dpif
->dpif_class
->port_get_pid
755 ? (dpif
->dpif_class
->port_get_pid
)(dpif
, port_no
)
759 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
760 * the port's name into the 'name_size' bytes in 'name', ensuring that the
761 * result is null-terminated. On failure, returns a positive errno value and
762 * makes 'name' the empty string. */
764 dpif_port_get_name(struct dpif
*dpif
, odp_port_t port_no
,
765 char *name
, size_t name_size
)
767 struct dpif_port port
;
770 ovs_assert(name_size
> 0);
772 error
= dpif_port_query_by_number(dpif
, port_no
, &port
);
774 ovs_strlcpy(name
, port
.name
, name_size
);
775 dpif_port_destroy(&port
);
782 /* Initializes 'dump' to begin dumping the ports in a dpif.
784 * This function provides no status indication. An error status for the entire
785 * dump operation is provided when it is completed by calling
786 * dpif_port_dump_done().
789 dpif_port_dump_start(struct dpif_port_dump
*dump
, const struct dpif
*dpif
)
792 dump
->error
= dpif
->dpif_class
->port_dump_start(dpif
, &dump
->state
);
793 log_operation(dpif
, "port_dump_start", dump
->error
);
796 /* Attempts to retrieve another port from 'dump', which must have been
797 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
798 * into 'port' and returns true. On failure, returns false.
800 * Failure might indicate an actual error or merely that the last port has been
801 * dumped. An error status for the entire dump operation is provided when it
802 * is completed by calling dpif_port_dump_done().
804 * The dpif owns the data stored in 'port'. It will remain valid until at
805 * least the next time 'dump' is passed to dpif_port_dump_next() or
806 * dpif_port_dump_done(). */
808 dpif_port_dump_next(struct dpif_port_dump
*dump
, struct dpif_port
*port
)
810 const struct dpif
*dpif
= dump
->dpif
;
816 dump
->error
= dpif
->dpif_class
->port_dump_next(dpif
, dump
->state
, port
);
817 if (dump
->error
== EOF
) {
818 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all ports", dpif_name(dpif
));
820 log_operation(dpif
, "port_dump_next", dump
->error
);
824 dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
830 /* Completes port table dump operation 'dump', which must have been initialized
831 * with dpif_port_dump_start(). Returns 0 if the dump operation was
832 * error-free, otherwise a positive errno value describing the problem. */
834 dpif_port_dump_done(struct dpif_port_dump
*dump
)
836 const struct dpif
*dpif
= dump
->dpif
;
838 dump
->error
= dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
839 log_operation(dpif
, "port_dump_done", dump
->error
);
841 return dump
->error
== EOF
? 0 : dump
->error
;
844 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
845 * 'dpif' has changed, this function does one of the following:
847 * - Stores the name of the device that was added to or deleted from 'dpif' in
848 * '*devnamep' and returns 0. The caller is responsible for freeing
849 * '*devnamep' (with free()) when it no longer needs it.
851 * - Returns ENOBUFS and sets '*devnamep' to NULL.
853 * This function may also return 'false positives', where it returns 0 and
854 * '*devnamep' names a device that was not actually added or deleted or it
855 * returns ENOBUFS without any change.
857 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
858 * return other positive errno values to indicate that something has gone
861 dpif_port_poll(const struct dpif
*dpif
, char **devnamep
)
863 int error
= dpif
->dpif_class
->port_poll(dpif
, devnamep
);
870 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
871 * value other than EAGAIN. */
873 dpif_port_poll_wait(const struct dpif
*dpif
)
875 dpif
->dpif_class
->port_poll_wait(dpif
);
878 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
879 * arguments must have been initialized through a call to flow_extract().
880 * 'used' is stored into stats->used. */
882 dpif_flow_stats_extract(const struct flow
*flow
, const struct dp_packet
*packet
,
883 long long int used
, struct dpif_flow_stats
*stats
)
885 stats
->tcp_flags
= ntohs(flow
->tcp_flags
);
886 stats
->n_bytes
= dp_packet_size(packet
);
887 stats
->n_packets
= 1;
891 /* Appends a human-readable representation of 'stats' to 's'. */
893 dpif_flow_stats_format(const struct dpif_flow_stats
*stats
, struct ds
*s
)
895 ds_put_format(s
, "packets:%"PRIu64
", bytes:%"PRIu64
", used:",
896 stats
->n_packets
, stats
->n_bytes
);
898 ds_put_format(s
, "%.3fs", (time_msec() - stats
->used
) / 1000.0);
900 ds_put_format(s
, "never");
902 if (stats
->tcp_flags
) {
903 ds_put_cstr(s
, ", flags:");
904 packet_format_tcp_flags(s
, stats
->tcp_flags
);
908 /* Places the hash of the 'key_len' bytes starting at 'key' into '*hash'. */
910 dpif_flow_hash(const struct dpif
*dpif OVS_UNUSED
,
911 const void *key
, size_t key_len
, ovs_u128
*hash
)
913 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
914 static uint32_t secret
;
916 if (ovsthread_once_start(&once
)) {
917 secret
= random_uint32();
918 ovsthread_once_done(&once
);
920 hash_bytes128(key
, key_len
, secret
, hash
);
921 uuid_set_bits_v4((struct uuid
*)hash
);
924 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
925 * positive errno value. */
927 dpif_flow_flush(struct dpif
*dpif
)
931 COVERAGE_INC(dpif_flow_flush
);
933 error
= dpif
->dpif_class
->flow_flush(dpif
);
934 log_operation(dpif
, "flow_flush", error
);
938 /* Attempts to install 'key' into the datapath, fetches it, then deletes it.
939 * Returns true if the datapath supported installing 'flow', false otherwise.
942 dpif_probe_feature(struct dpif
*dpif
, const char *name
,
943 const struct ofpbuf
*key
, const struct ofpbuf
*actions
,
944 const ovs_u128
*ufid
)
946 struct dpif_flow flow
;
948 uint64_t stub
[DPIF_FLOW_BUFSIZE
/ 8];
949 bool enable_feature
= false;
951 const struct nlattr
*nl_actions
= actions
? actions
->data
: NULL
;
952 const size_t nl_actions_size
= actions
? actions
->size
: 0;
954 /* Use DPIF_FP_MODIFY to cover the case where ovs-vswitchd is killed (and
955 * restarted) at just the right time such that feature probes from the
956 * previous run are still present in the datapath. */
957 error
= dpif_flow_put(dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
| DPIF_FP_PROBE
,
958 key
->data
, key
->size
, NULL
, 0,
959 nl_actions
, nl_actions_size
,
960 ufid
, NON_PMD_CORE_ID
, NULL
);
962 if (error
!= EINVAL
&& error
!= EOVERFLOW
) {
963 VLOG_WARN("%s: %s flow probe failed (%s)",
964 dpif_name(dpif
), name
, ovs_strerror(error
));
969 ofpbuf_use_stack(&reply
, &stub
, sizeof stub
);
970 error
= dpif_flow_get(dpif
, key
->data
, key
->size
, ufid
,
971 NON_PMD_CORE_ID
, &reply
, &flow
);
973 && (!ufid
|| (flow
.ufid_present
974 && ovs_u128_equals(*ufid
, flow
.ufid
)))) {
975 enable_feature
= true;
978 error
= dpif_flow_del(dpif
, key
->data
, key
->size
, ufid
,
979 NON_PMD_CORE_ID
, NULL
);
981 VLOG_WARN("%s: failed to delete %s feature probe flow",
982 dpif_name(dpif
), name
);
985 return enable_feature
;
988 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_GET. */
990 dpif_flow_get(struct dpif
*dpif
,
991 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
992 const unsigned pmd_id
, struct ofpbuf
*buf
, struct dpif_flow
*flow
)
997 op
.type
= DPIF_OP_FLOW_GET
;
998 op
.flow_get
.key
= key
;
999 op
.flow_get
.key_len
= key_len
;
1000 op
.flow_get
.ufid
= ufid
;
1001 op
.flow_get
.pmd_id
= pmd_id
;
1002 op
.flow_get
.buffer
= buf
;
1004 memset(flow
, 0, sizeof *flow
);
1005 op
.flow_get
.flow
= flow
;
1006 op
.flow_get
.flow
->key
= key
;
1007 op
.flow_get
.flow
->key_len
= key_len
;
1010 dpif_operate(dpif
, &opp
, 1, DPIF_OFFLOAD_AUTO
);
1015 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_PUT. */
1017 dpif_flow_put(struct dpif
*dpif
, enum dpif_flow_put_flags flags
,
1018 const struct nlattr
*key
, size_t key_len
,
1019 const struct nlattr
*mask
, size_t mask_len
,
1020 const struct nlattr
*actions
, size_t actions_len
,
1021 const ovs_u128
*ufid
, const unsigned pmd_id
,
1022 struct dpif_flow_stats
*stats
)
1024 struct dpif_op
*opp
;
1027 op
.type
= DPIF_OP_FLOW_PUT
;
1028 op
.flow_put
.flags
= flags
;
1029 op
.flow_put
.key
= key
;
1030 op
.flow_put
.key_len
= key_len
;
1031 op
.flow_put
.mask
= mask
;
1032 op
.flow_put
.mask_len
= mask_len
;
1033 op
.flow_put
.actions
= actions
;
1034 op
.flow_put
.actions_len
= actions_len
;
1035 op
.flow_put
.ufid
= ufid
;
1036 op
.flow_put
.pmd_id
= pmd_id
;
1037 op
.flow_put
.stats
= stats
;
1040 dpif_operate(dpif
, &opp
, 1, DPIF_OFFLOAD_AUTO
);
1045 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_DEL. */
1047 dpif_flow_del(struct dpif
*dpif
,
1048 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
1049 const unsigned pmd_id
, struct dpif_flow_stats
*stats
)
1051 struct dpif_op
*opp
;
1054 op
.type
= DPIF_OP_FLOW_DEL
;
1055 op
.flow_del
.key
= key
;
1056 op
.flow_del
.key_len
= key_len
;
1057 op
.flow_del
.ufid
= ufid
;
1058 op
.flow_del
.pmd_id
= pmd_id
;
1059 op
.flow_del
.stats
= stats
;
1060 op
.flow_del
.terse
= false;
1063 dpif_operate(dpif
, &opp
, 1, DPIF_OFFLOAD_AUTO
);
1068 /* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
1069 * flows in 'dpif'. If 'terse' is true, then only UFID and statistics will
1070 * be returned in the dump. Otherwise, all fields will be returned.
1072 * This function always successfully returns a dpif_flow_dump. Error
1073 * reporting is deferred to dpif_flow_dump_destroy(). */
1074 struct dpif_flow_dump
*
1075 dpif_flow_dump_create(const struct dpif
*dpif
, bool terse
,
1076 struct dpif_flow_dump_types
*types
)
1078 return dpif
->dpif_class
->flow_dump_create(dpif
, terse
, types
);
1081 /* Destroys 'dump', which must have been created with dpif_flow_dump_create().
1082 * All dpif_flow_dump_thread structures previously created for 'dump' must
1083 * previously have been destroyed.
1085 * Returns 0 if the dump operation was error-free, otherwise a positive errno
1086 * value describing the problem. */
1088 dpif_flow_dump_destroy(struct dpif_flow_dump
*dump
)
1090 const struct dpif
*dpif
= dump
->dpif
;
1091 int error
= dpif
->dpif_class
->flow_dump_destroy(dump
);
1092 log_operation(dpif
, "flow_dump_destroy", error
);
1093 return error
== EOF
? 0 : error
;
1096 /* Returns new thread-local state for use with dpif_flow_dump_next(). */
1097 struct dpif_flow_dump_thread
*
1098 dpif_flow_dump_thread_create(struct dpif_flow_dump
*dump
)
1100 return dump
->dpif
->dpif_class
->flow_dump_thread_create(dump
);
1103 /* Releases 'thread'. */
1105 dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread
*thread
)
1107 thread
->dpif
->dpif_class
->flow_dump_thread_destroy(thread
);
1110 /* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0
1111 * if and only if no flows remained to be retrieved, otherwise a positive
1112 * number reflecting the number of elements in 'flows[]' that were updated.
1113 * The number of flows returned might be less than 'max_flows' because
1114 * fewer than 'max_flows' remained, because this particular datapath does not
1115 * benefit from batching, or because an error occurred partway through
1116 * retrieval. Thus, the caller should continue calling until a 0 return value,
1117 * even if intermediate return values are less than 'max_flows'.
1119 * No error status is immediately provided. An error status for the entire
1120 * dump operation is provided when it is completed by calling
1121 * dpif_flow_dump_destroy().
1123 * All of the data stored into 'flows' is owned by the datapath, not by the
1124 * caller, and the caller must not modify or free it. The datapath guarantees
1125 * that it remains accessible and unchanged until the first of:
1126 * - The next call to dpif_flow_dump_next() for 'thread', or
1127 * - The next rcu quiescent period. */
1129 dpif_flow_dump_next(struct dpif_flow_dump_thread
*thread
,
1130 struct dpif_flow
*flows
, int max_flows
)
1132 struct dpif
*dpif
= thread
->dpif
;
1135 ovs_assert(max_flows
> 0);
1136 n
= dpif
->dpif_class
->flow_dump_next(thread
, flows
, max_flows
);
1138 struct dpif_flow
*f
;
1140 for (f
= flows
; f
< &flows
[n
]
1141 && should_log_flow_message(&this_module
, 0); f
++) {
1142 log_flow_message(dpif
, 0, &this_module
, "flow_dump",
1143 f
->key
, f
->key_len
, f
->mask
, f
->mask_len
,
1144 &f
->ufid
, &f
->stats
, f
->actions
, f
->actions_len
);
1147 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all flows", dpif_name(dpif
));
1152 struct dpif_execute_helper_aux
{
1154 const struct flow
*flow
;
1156 const struct nlattr
*meter_action
; /* Non-NULL, if have a meter action. */
1159 /* This is called for actions that need the context of the datapath to be
1162 dpif_execute_helper_cb(void *aux_
, struct dp_packet_batch
*packets_
,
1163 const struct nlattr
*action
, bool should_steal
)
1165 struct dpif_execute_helper_aux
*aux
= aux_
;
1166 int type
= nl_attr_type(action
);
1167 struct dp_packet
*packet
= packets_
->packets
[0];
1169 ovs_assert(packets_
->count
== 1);
1171 switch ((enum ovs_action_attr
)type
) {
1172 case OVS_ACTION_ATTR_METER
:
1173 /* Maintain a pointer to the first meter action seen. */
1174 if (!aux
->meter_action
) {
1175 aux
->meter_action
= action
;
1179 case OVS_ACTION_ATTR_CT
:
1180 case OVS_ACTION_ATTR_OUTPUT
:
1181 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
1182 case OVS_ACTION_ATTR_TUNNEL_POP
:
1183 case OVS_ACTION_ATTR_USERSPACE
:
1184 case OVS_ACTION_ATTR_RECIRC
: {
1185 struct dpif_execute execute
;
1186 struct ofpbuf execute_actions
;
1187 uint64_t stub
[256 / 8];
1188 struct pkt_metadata
*md
= &packet
->md
;
1190 if (flow_tnl_dst_is_set(&md
->tunnel
) || aux
->meter_action
) {
1191 ofpbuf_use_stub(&execute_actions
, stub
, sizeof stub
);
1193 if (aux
->meter_action
) {
1194 const struct nlattr
*a
= aux
->meter_action
;
1196 /* XXX: This code collects meter actions since the last action
1197 * execution via the datapath to be executed right before the
1198 * current action that needs to be executed by the datapath.
1199 * This is only an approximation, but better than nothing.
1200 * Fundamentally, we should have a mechanism by which the
1201 * datapath could return the result of the meter action so that
1202 * we could execute them at the right order. */
1204 ofpbuf_put(&execute_actions
, a
, NLA_ALIGN(a
->nla_len
));
1205 /* Find next meter action before 'action', if any. */
1207 a
= nl_attr_next(a
);
1208 } while (a
!= action
&&
1209 nl_attr_type(a
) != OVS_ACTION_ATTR_METER
);
1210 } while (a
!= action
);
1213 /* The Linux kernel datapath throws away the tunnel information
1214 * that we supply as metadata. We have to use a "set" action to
1216 if (md
->tunnel
.ip_dst
) {
1217 odp_put_tunnel_action(&md
->tunnel
, &execute_actions
, NULL
);
1219 ofpbuf_put(&execute_actions
, action
, NLA_ALIGN(action
->nla_len
));
1221 execute
.actions
= execute_actions
.data
;
1222 execute
.actions_len
= execute_actions
.size
;
1224 execute
.actions
= action
;
1225 execute
.actions_len
= NLA_ALIGN(action
->nla_len
);
1228 struct dp_packet
*clone
= NULL
;
1229 uint32_t cutlen
= dp_packet_get_cutlen(packet
);
1230 if (cutlen
&& (type
== OVS_ACTION_ATTR_OUTPUT
1231 || type
== OVS_ACTION_ATTR_TUNNEL_PUSH
1232 || type
== OVS_ACTION_ATTR_TUNNEL_POP
1233 || type
== OVS_ACTION_ATTR_USERSPACE
)) {
1234 dp_packet_reset_cutlen(packet
);
1235 if (!should_steal
) {
1236 packet
= clone
= dp_packet_clone(packet
);
1238 dp_packet_set_size(packet
, dp_packet_size(packet
) - cutlen
);
1241 execute
.packet
= packet
;
1242 execute
.flow
= aux
->flow
;
1243 execute
.needs_help
= false;
1244 execute
.probe
= false;
1246 aux
->error
= dpif_execute(aux
->dpif
, &execute
);
1247 log_execute_message(aux
->dpif
, &this_module
, &execute
,
1250 dp_packet_delete(clone
);
1252 if (flow_tnl_dst_is_set(&md
->tunnel
) || aux
->meter_action
) {
1253 ofpbuf_uninit(&execute_actions
);
1255 /* Do not re-use the same meters for later output actions. */
1256 aux
->meter_action
= NULL
;
1261 case OVS_ACTION_ATTR_HASH
:
1262 case OVS_ACTION_ATTR_PUSH_VLAN
:
1263 case OVS_ACTION_ATTR_POP_VLAN
:
1264 case OVS_ACTION_ATTR_PUSH_MPLS
:
1265 case OVS_ACTION_ATTR_POP_MPLS
:
1266 case OVS_ACTION_ATTR_SET
:
1267 case OVS_ACTION_ATTR_SET_MASKED
:
1268 case OVS_ACTION_ATTR_SAMPLE
:
1269 case OVS_ACTION_ATTR_TRUNC
:
1270 case OVS_ACTION_ATTR_PUSH_ETH
:
1271 case OVS_ACTION_ATTR_POP_ETH
:
1272 case OVS_ACTION_ATTR_CLONE
:
1273 case OVS_ACTION_ATTR_PUSH_NSH
:
1274 case OVS_ACTION_ATTR_POP_NSH
:
1275 case OVS_ACTION_ATTR_CT_CLEAR
:
1276 case OVS_ACTION_ATTR_UNSPEC
:
1277 case __OVS_ACTION_ATTR_MAX
:
1280 dp_packet_delete_batch(packets_
, should_steal
);
1283 /* Executes 'execute' by performing most of the actions in userspace and
1284 * passing the fully constructed packets to 'dpif' for output and userspace
1287 * This helps with actions that a given 'dpif' doesn't implement directly. */
1289 dpif_execute_with_help(struct dpif
*dpif
, struct dpif_execute
*execute
)
1291 struct dpif_execute_helper_aux aux
= {dpif
, execute
->flow
, 0, NULL
};
1292 struct dp_packet_batch pb
;
1294 COVERAGE_INC(dpif_execute_with_help
);
1296 dp_packet_batch_init_packet(&pb
, execute
->packet
);
1297 odp_execute_actions(&aux
, &pb
, false, execute
->actions
,
1298 execute
->actions_len
, dpif_execute_helper_cb
);
1302 /* Returns true if the datapath needs help executing 'execute'. */
1304 dpif_execute_needs_help(const struct dpif_execute
*execute
)
1306 return execute
->needs_help
|| nl_attr_oversized(execute
->actions_len
);
1309 /* A dpif_operate() wrapper for performing a single DPIF_OP_EXECUTE. */
1311 dpif_execute(struct dpif
*dpif
, struct dpif_execute
*execute
)
1313 if (execute
->actions_len
) {
1314 struct dpif_op
*opp
;
1317 op
.type
= DPIF_OP_EXECUTE
;
1318 op
.execute
= *execute
;
1321 dpif_operate(dpif
, &opp
, 1, DPIF_OFFLOAD_AUTO
);
1329 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1330 * which they are specified. Places each operation's results in the "output"
1331 * members documented in comments, and 0 in the 'error' member on success or a
1332 * positive errno on failure.
1335 dpif_operate(struct dpif
*dpif
, struct dpif_op
**ops
, size_t n_ops
,
1336 enum dpif_offload_type offload_type
)
1338 if (offload_type
== DPIF_OFFLOAD_ALWAYS
&& !netdev_is_flow_api_enabled()) {
1340 for (i
= 0; i
< n_ops
; i
++) {
1341 struct dpif_op
*op
= ops
[i
];
1350 /* Count 'chunk', the number of ops that can be executed without
1351 * needing any help. Ops that need help should be rare, so we
1352 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1353 for (chunk
= 0; chunk
< n_ops
; chunk
++) {
1354 struct dpif_op
*op
= ops
[chunk
];
1356 if (op
->type
== DPIF_OP_EXECUTE
1357 && dpif_execute_needs_help(&op
->execute
)) {
1363 /* Execute a chunk full of ops that the dpif provider can
1364 * handle itself, without help. */
1367 dpif
->dpif_class
->operate(dpif
, ops
, chunk
, offload_type
);
1369 for (i
= 0; i
< chunk
; i
++) {
1370 struct dpif_op
*op
= ops
[i
];
1371 int error
= op
->error
;
1374 case DPIF_OP_FLOW_PUT
: {
1375 struct dpif_flow_put
*put
= &op
->flow_put
;
1377 COVERAGE_INC(dpif_flow_put
);
1378 log_flow_put_message(dpif
, &this_module
, put
, error
);
1379 if (error
&& put
->stats
) {
1380 memset(put
->stats
, 0, sizeof *put
->stats
);
1385 case DPIF_OP_FLOW_GET
: {
1386 struct dpif_flow_get
*get
= &op
->flow_get
;
1388 COVERAGE_INC(dpif_flow_get
);
1390 memset(get
->flow
, 0, sizeof *get
->flow
);
1392 log_flow_get_message(dpif
, &this_module
, get
, error
);
1397 case DPIF_OP_FLOW_DEL
: {
1398 struct dpif_flow_del
*del
= &op
->flow_del
;
1400 COVERAGE_INC(dpif_flow_del
);
1401 log_flow_del_message(dpif
, &this_module
, del
, error
);
1402 if (error
&& del
->stats
) {
1403 memset(del
->stats
, 0, sizeof *del
->stats
);
1408 case DPIF_OP_EXECUTE
:
1409 COVERAGE_INC(dpif_execute
);
1410 log_execute_message(dpif
, &this_module
, &op
->execute
,
1419 /* Help the dpif provider to execute one op. */
1420 struct dpif_op
*op
= ops
[0];
1422 COVERAGE_INC(dpif_execute
);
1423 op
->error
= dpif_execute_with_help(dpif
, &op
->execute
);
1430 /* Returns a string that represents 'type', for use in log messages. */
1432 dpif_upcall_type_to_string(enum dpif_upcall_type type
)
1435 case DPIF_UC_MISS
: return "miss";
1436 case DPIF_UC_ACTION
: return "action";
1437 case DPIF_N_UC_TYPES
: default: return "<unknown>";
1441 /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1442 * if successful, otherwise a positive errno value.
1444 * Turning packet receive off and then back on may change the Netlink PID
1445 * assignments returned by dpif_port_get_pid(). If the client does this, it
1446 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1447 * using the new PID assignment. */
1449 dpif_recv_set(struct dpif
*dpif
, bool enable
)
1453 if (dpif
->dpif_class
->recv_set
) {
1454 error
= dpif
->dpif_class
->recv_set(dpif
, enable
);
1455 log_operation(dpif
, "recv_set", error
);
1460 /* Refreshes the poll loops and Netlink sockets associated to each port,
1461 * when the number of upcall handlers (upcall receiving thread) is changed
1462 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
1465 * Since multiple upcall handlers can read upcalls simultaneously from
1466 * 'dpif', each port can have multiple Netlink sockets, one per upcall
1467 * handler. So, handlers_set() is responsible for the following tasks:
1469 * When receiving upcall is enabled, extends or creates the
1470 * configuration to support:
1472 * - 'n_handlers' Netlink sockets for each port.
1474 * - 'n_handlers' poll loops, one for each upcall handler.
1476 * - registering the Netlink sockets for the same upcall handler to
1477 * the corresponding poll loop.
1479 * Returns 0 if successful, otherwise a positive errno value. */
1481 dpif_handlers_set(struct dpif
*dpif
, uint32_t n_handlers
)
1485 if (dpif
->dpif_class
->handlers_set
) {
1486 error
= dpif
->dpif_class
->handlers_set(dpif
, n_handlers
);
1487 log_operation(dpif
, "handlers_set", error
);
1493 dpif_register_dp_purge_cb(struct dpif
*dpif
, dp_purge_callback
*cb
, void *aux
)
1495 if (dpif
->dpif_class
->register_dp_purge_cb
) {
1496 dpif
->dpif_class
->register_dp_purge_cb(dpif
, cb
, aux
);
1501 dpif_register_upcall_cb(struct dpif
*dpif
, upcall_callback
*cb
, void *aux
)
1503 if (dpif
->dpif_class
->register_upcall_cb
) {
1504 dpif
->dpif_class
->register_upcall_cb(dpif
, cb
, aux
);
1509 dpif_enable_upcall(struct dpif
*dpif
)
1511 if (dpif
->dpif_class
->enable_upcall
) {
1512 dpif
->dpif_class
->enable_upcall(dpif
);
1517 dpif_disable_upcall(struct dpif
*dpif
)
1519 if (dpif
->dpif_class
->disable_upcall
) {
1520 dpif
->dpif_class
->disable_upcall(dpif
);
1525 dpif_print_packet(struct dpif
*dpif
, struct dpif_upcall
*upcall
)
1527 if (!VLOG_DROP_DBG(&dpmsg_rl
)) {
1531 packet
= ofp_dp_packet_to_string(&upcall
->packet
);
1534 odp_flow_key_format(upcall
->key
, upcall
->key_len
, &flow
);
1536 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1537 dpif_name(dpif
), dpif_upcall_type_to_string(upcall
->type
),
1538 ds_cstr(&flow
), packet
);
1545 /* Pass custom configuration to the datapath implementation. Some of the
1546 * changes can be postponed until dpif_run() is called. */
1548 dpif_set_config(struct dpif
*dpif
, const struct smap
*cfg
)
1552 if (dpif
->dpif_class
->set_config
) {
1553 error
= dpif
->dpif_class
->set_config(dpif
, cfg
);
1555 log_operation(dpif
, "set_config", error
);
1562 /* Polls for an upcall from 'dpif' for an upcall handler. Since there can
1563 * be multiple poll loops, 'handler_id' is needed as index to identify the
1564 * corresponding poll loop. If successful, stores the upcall into '*upcall',
1565 * using 'buf' for storage. Should only be called if 'recv_set' has been used
1566 * to enable receiving packets from 'dpif'.
1568 * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
1569 * 'buf', so their memory cannot be freed separately from 'buf'.
1571 * The caller owns the data of 'upcall->packet' and may modify it. If
1572 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
1573 * will be reallocated. This requires the data of 'upcall->packet' to be
1574 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
1575 * when an error is returned, the 'upcall->packet' may be uninitialized
1576 * and should not be released.
1578 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1579 * if no upcall is immediately available. */
1581 dpif_recv(struct dpif
*dpif
, uint32_t handler_id
, struct dpif_upcall
*upcall
,
1586 if (dpif
->dpif_class
->recv
) {
1587 error
= dpif
->dpif_class
->recv(dpif
, handler_id
, upcall
, buf
);
1589 dpif_print_packet(dpif
, upcall
);
1590 } else if (error
!= EAGAIN
) {
1591 log_operation(dpif
, "recv", error
);
1597 /* Discards all messages that would otherwise be received by dpif_recv() on
1600 dpif_recv_purge(struct dpif
*dpif
)
1602 COVERAGE_INC(dpif_purge
);
1603 if (dpif
->dpif_class
->recv_purge
) {
1604 dpif
->dpif_class
->recv_purge(dpif
);
1608 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
1609 * 'dpif' has a message queued to be received with the recv member
1610 * function. Since there can be multiple poll loops, 'handler_id' is
1611 * needed as index to identify the corresponding poll loop. */
1613 dpif_recv_wait(struct dpif
*dpif
, uint32_t handler_id
)
1615 if (dpif
->dpif_class
->recv_wait
) {
1616 dpif
->dpif_class
->recv_wait(dpif
, handler_id
);
1621 * Return the datapath version. Caller is responsible for freeing
1625 dpif_get_dp_version(const struct dpif
*dpif
)
1627 char *version
= NULL
;
1629 if (dpif
->dpif_class
->get_datapath_version
) {
1630 version
= dpif
->dpif_class
->get_datapath_version();
1636 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1637 * and '*engine_id', respectively. */
1639 dpif_get_netflow_ids(const struct dpif
*dpif
,
1640 uint8_t *engine_type
, uint8_t *engine_id
)
1642 *engine_type
= dpif
->netflow_engine_type
;
1643 *engine_id
= dpif
->netflow_engine_id
;
1646 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1647 * value used for setting packet priority.
1648 * On success, returns 0 and stores the priority into '*priority'.
1649 * On failure, returns a positive errno value and stores 0 into '*priority'. */
1651 dpif_queue_to_priority(const struct dpif
*dpif
, uint32_t queue_id
,
1654 int error
= (dpif
->dpif_class
->queue_to_priority
1655 ? dpif
->dpif_class
->queue_to_priority(dpif
, queue_id
,
1661 log_operation(dpif
, "queue_to_priority", error
);
1666 dpif_init(struct dpif
*dpif
, const struct dpif_class
*dpif_class
,
1668 uint8_t netflow_engine_type
, uint8_t netflow_engine_id
)
1670 dpif
->dpif_class
= dpif_class
;
1671 dpif
->base_name
= xstrdup(name
);
1672 dpif
->full_name
= xasprintf("%s@%s", dpif_class
->type
, name
);
1673 dpif
->netflow_engine_type
= netflow_engine_type
;
1674 dpif
->netflow_engine_id
= netflow_engine_id
;
1677 /* Undoes the results of initialization.
1679 * Normally this function only needs to be called from dpif_close().
1680 * However, it may be called by providers due to an error on opening
1681 * that occurs after initialization. It this case dpif_close() would
1682 * never be called. */
1684 dpif_uninit(struct dpif
*dpif
, bool close
)
1686 char *base_name
= dpif
->base_name
;
1687 char *full_name
= dpif
->full_name
;
1690 dpif
->dpif_class
->close(dpif
);
1698 log_operation(const struct dpif
*dpif
, const char *operation
, int error
)
1701 VLOG_DBG_RL(&dpmsg_rl
, "%s: %s success", dpif_name(dpif
), operation
);
1702 } else if (ofperr_is_valid(error
)) {
1703 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1704 dpif_name(dpif
), operation
, ofperr_get_name(error
));
1706 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1707 dpif_name(dpif
), operation
, ovs_strerror(error
));
1711 static enum vlog_level
1712 flow_message_log_level(int error
)
1714 /* If flows arrive in a batch, userspace may push down multiple
1715 * unique flow definitions that overlap when wildcards are applied.
1716 * Kernels that support flow wildcarding will reject these flows as
1717 * duplicates (EEXIST), so lower the log level to debug for these
1718 * types of messages. */
1719 return (error
&& error
!= EEXIST
) ? VLL_WARN
: VLL_DBG
;
1723 should_log_flow_message(const struct vlog_module
*module
, int error
)
1725 return !vlog_should_drop(module
, flow_message_log_level(error
),
1726 error
? &error_rl
: &dpmsg_rl
);
1730 log_flow_message(const struct dpif
*dpif
, int error
,
1731 const struct vlog_module
*module
,
1732 const char *operation
,
1733 const struct nlattr
*key
, size_t key_len
,
1734 const struct nlattr
*mask
, size_t mask_len
,
1735 const ovs_u128
*ufid
, const struct dpif_flow_stats
*stats
,
1736 const struct nlattr
*actions
, size_t actions_len
)
1738 struct ds ds
= DS_EMPTY_INITIALIZER
;
1739 ds_put_format(&ds
, "%s: ", dpif_name(dpif
));
1741 ds_put_cstr(&ds
, "failed to ");
1743 ds_put_format(&ds
, "%s ", operation
);
1745 ds_put_format(&ds
, "(%s) ", ovs_strerror(error
));
1748 odp_format_ufid(ufid
, &ds
);
1749 ds_put_cstr(&ds
, " ");
1751 odp_flow_format(key
, key_len
, mask
, mask_len
, NULL
, &ds
, true);
1753 ds_put_cstr(&ds
, ", ");
1754 dpif_flow_stats_format(stats
, &ds
);
1756 if (actions
|| actions_len
) {
1757 ds_put_cstr(&ds
, ", actions:");
1758 format_odp_actions(&ds
, actions
, actions_len
, NULL
);
1760 vlog(module
, flow_message_log_level(error
), "%s", ds_cstr(&ds
));
1765 log_flow_put_message(const struct dpif
*dpif
,
1766 const struct vlog_module
*module
,
1767 const struct dpif_flow_put
*put
,
1770 if (should_log_flow_message(module
, error
)
1771 && !(put
->flags
& DPIF_FP_PROBE
)) {
1775 ds_put_cstr(&s
, "put");
1776 if (put
->flags
& DPIF_FP_CREATE
) {
1777 ds_put_cstr(&s
, "[create]");
1779 if (put
->flags
& DPIF_FP_MODIFY
) {
1780 ds_put_cstr(&s
, "[modify]");
1782 if (put
->flags
& DPIF_FP_ZERO_STATS
) {
1783 ds_put_cstr(&s
, "[zero]");
1785 log_flow_message(dpif
, error
, module
, ds_cstr(&s
),
1786 put
->key
, put
->key_len
, put
->mask
, put
->mask_len
,
1787 put
->ufid
, put
->stats
, put
->actions
,
1794 log_flow_del_message(const struct dpif
*dpif
,
1795 const struct vlog_module
*module
,
1796 const struct dpif_flow_del
*del
,
1799 if (should_log_flow_message(module
, error
)) {
1800 log_flow_message(dpif
, error
, module
, "flow_del",
1801 del
->key
, del
->key_len
,
1802 NULL
, 0, del
->ufid
, !error
? del
->stats
: NULL
,
1807 /* Logs that 'execute' was executed on 'dpif' and completed with errno 'error'
1808 * (0 for success). 'subexecute' should be true if the execution is a result
1809 * of breaking down a larger execution that needed help, false otherwise.
1812 * XXX In theory, the log message could be deceptive because this function is
1813 * called after the dpif_provider's '->execute' function, which is allowed to
1814 * modify execute->packet and execute->md. In practice, though:
1816 * - dpif-netlink doesn't modify execute->packet or execute->md.
1818 * - dpif-netdev does modify them but it is less likely to have problems
1819 * because it is built into ovs-vswitchd and cannot have version skew,
1822 * It would still be better to avoid the potential problem. I don't know of a
1823 * good way to do that, though, that isn't expensive. */
1825 log_execute_message(const struct dpif
*dpif
,
1826 const struct vlog_module
*module
,
1827 const struct dpif_execute
*execute
,
1828 bool subexecute
, int error
)
1830 if (!(error
? VLOG_DROP_WARN(&error_rl
) : VLOG_DROP_DBG(&dpmsg_rl
))
1831 && !execute
->probe
) {
1832 struct ds ds
= DS_EMPTY_INITIALIZER
;
1834 uint64_t stub
[1024 / 8];
1835 struct ofpbuf md
= OFPBUF_STUB_INITIALIZER(stub
);
1837 packet
= ofp_packet_to_string(dp_packet_data(execute
->packet
),
1838 dp_packet_size(execute
->packet
),
1839 execute
->packet
->packet_type
);
1840 odp_key_from_dp_packet(&md
, execute
->packet
);
1841 ds_put_format(&ds
, "%s: %sexecute ",
1843 (subexecute
? "sub-"
1844 : dpif_execute_needs_help(execute
) ? "super-"
1846 format_odp_actions(&ds
, execute
->actions
, execute
->actions_len
, NULL
);
1848 ds_put_format(&ds
, " failed (%s)", ovs_strerror(error
));
1850 ds_put_format(&ds
, " on packet %s", packet
);
1851 ds_put_format(&ds
, " with metadata ");
1852 odp_flow_format(md
.data
, md
.size
, NULL
, 0, NULL
, &ds
, true);
1853 ds_put_format(&ds
, " mtu %d", execute
->mtu
);
1854 vlog(module
, error
? VLL_WARN
: VLL_DBG
, "%s", ds_cstr(&ds
));
1862 log_flow_get_message(const struct dpif
*dpif
,
1863 const struct vlog_module
*module
,
1864 const struct dpif_flow_get
*get
,
1867 if (should_log_flow_message(module
, error
)) {
1868 log_flow_message(dpif
, error
, module
, "flow_get",
1869 get
->key
, get
->key_len
,
1870 get
->flow
->mask
, get
->flow
->mask_len
,
1871 get
->ufid
, &get
->flow
->stats
,
1872 get
->flow
->actions
, get
->flow
->actions_len
);
1877 dpif_supports_tnl_push_pop(const struct dpif
*dpif
)
1879 return dpif_is_netdev(dpif
);
1884 dpif_meter_get_features(const struct dpif
*dpif
,
1885 struct ofputil_meter_features
*features
)
1887 memset(features
, 0, sizeof *features
);
1888 if (dpif
->dpif_class
->meter_get_features
) {
1889 dpif
->dpif_class
->meter_get_features(dpif
, features
);
1893 /* Adds or modifies the meter in 'dpif' with the given 'meter_id' and
1894 * the configuration in 'config'.
1896 * The meter id specified through 'config->meter_id' is ignored. */
1898 dpif_meter_set(struct dpif
*dpif
, ofproto_meter_id meter_id
,
1899 struct ofputil_meter_config
*config
)
1901 COVERAGE_INC(dpif_meter_set
);
1903 if (!(config
->flags
& (OFPMF13_KBPS
| OFPMF13_PKTPS
))) {
1904 return EBADF
; /* Rate unit type not set. */
1907 if ((config
->flags
& OFPMF13_KBPS
) && (config
->flags
& OFPMF13_PKTPS
)) {
1908 return EBADF
; /* Both rate units may not be set. */
1911 if (config
->n_bands
== 0) {
1915 for (size_t i
= 0; i
< config
->n_bands
; i
++) {
1916 if (config
->bands
[i
].rate
== 0) {
1917 return EDOM
; /* Rate must be non-zero */
1921 int error
= dpif
->dpif_class
->meter_set(dpif
, meter_id
, config
);
1923 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" set",
1924 dpif_name(dpif
), meter_id
.uint32
);
1926 VLOG_WARN_RL(&error_rl
, "%s: failed to set DPIF meter %"PRIu32
": %s",
1927 dpif_name(dpif
), meter_id
.uint32
, ovs_strerror(error
));
1933 dpif_meter_get(const struct dpif
*dpif
, ofproto_meter_id meter_id
,
1934 struct ofputil_meter_stats
*stats
, uint16_t n_bands
)
1938 COVERAGE_INC(dpif_meter_get
);
1940 error
= dpif
->dpif_class
->meter_get(dpif
, meter_id
, stats
, n_bands
);
1942 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" get stats",
1943 dpif_name(dpif
), meter_id
.uint32
);
1945 VLOG_WARN_RL(&error_rl
,
1946 "%s: failed to get DPIF meter %"PRIu32
" stats: %s",
1947 dpif_name(dpif
), meter_id
.uint32
, ovs_strerror(error
));
1948 stats
->packet_in_count
= ~0;
1949 stats
->byte_in_count
= ~0;
1956 dpif_meter_del(struct dpif
*dpif
, ofproto_meter_id meter_id
,
1957 struct ofputil_meter_stats
*stats
, uint16_t n_bands
)
1961 COVERAGE_INC(dpif_meter_del
);
1963 error
= dpif
->dpif_class
->meter_del(dpif
, meter_id
, stats
, n_bands
);
1965 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" deleted",
1966 dpif_name(dpif
), meter_id
.uint32
);
1968 VLOG_WARN_RL(&error_rl
,
1969 "%s: failed to delete DPIF meter %"PRIu32
": %s",
1970 dpif_name(dpif
), meter_id
.uint32
, ovs_strerror(error
));
1972 stats
->packet_in_count
= ~0;
1973 stats
->byte_in_count
= ~0;