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 "openvswitch/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_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 VLOG_WARN_RL(&error_rl
, "%s: failed to add %s as port: %s",
596 dpif_name(dpif
), netdev_name
, ovs_strerror(error
));
605 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
606 * otherwise a positive errno value. */
608 dpif_port_del(struct dpif
*dpif
, odp_port_t port_no
, bool local_delete
)
612 COVERAGE_INC(dpif_port_del
);
615 error
= dpif
->dpif_class
->port_del(dpif
, port_no
);
617 VLOG_DBG_RL(&dpmsg_rl
, "%s: port_del(%"PRIu32
")",
618 dpif_name(dpif
), port_no
);
620 log_operation(dpif
, "port_del", error
);
624 netdev_ports_remove(port_no
, dpif
->dpif_class
);
628 /* Makes a deep copy of 'src' into 'dst'. */
630 dpif_port_clone(struct dpif_port
*dst
, const struct dpif_port
*src
)
632 dst
->name
= xstrdup(src
->name
);
633 dst
->type
= xstrdup(src
->type
);
634 dst
->port_no
= src
->port_no
;
637 /* Frees memory allocated to members of 'dpif_port'.
639 * Do not call this function on a dpif_port obtained from
640 * dpif_port_dump_next(): that function retains ownership of the data in the
643 dpif_port_destroy(struct dpif_port
*dpif_port
)
645 free(dpif_port
->name
);
646 free(dpif_port
->type
);
649 /* Checks if port named 'devname' exists in 'dpif'. If so, returns
650 * true; otherwise, returns false. */
652 dpif_port_exists(const struct dpif
*dpif
, const char *devname
)
654 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, NULL
);
655 if (error
!= 0 && error
!= ENODEV
) {
656 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %s: %s",
657 dpif_name(dpif
), devname
, ovs_strerror(error
));
663 /* Refreshes configuration of 'dpif's port. */
665 dpif_port_set_config(struct dpif
*dpif
, odp_port_t port_no
,
666 const struct smap
*cfg
)
670 if (dpif
->dpif_class
->port_set_config
) {
671 error
= dpif
->dpif_class
->port_set_config(dpif
, port_no
, cfg
);
673 log_operation(dpif
, "port_set_config", error
);
680 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
681 * initializes '*port' appropriately; on failure, returns a positive errno
684 * Retuns ENODEV if the port doesn't exist.
686 * The caller owns the data in 'port' and must free it with
687 * dpif_port_destroy() when it is no longer needed. */
689 dpif_port_query_by_number(const struct dpif
*dpif
, odp_port_t port_no
,
690 struct dpif_port
*port
)
692 int error
= dpif
->dpif_class
->port_query_by_number(dpif
, port_no
, port
);
694 VLOG_DBG_RL(&dpmsg_rl
, "%s: port %"PRIu32
" is device %s",
695 dpif_name(dpif
), port_no
, port
->name
);
697 memset(port
, 0, sizeof *port
);
698 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %"PRIu32
": %s",
699 dpif_name(dpif
), port_no
, ovs_strerror(error
));
704 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
705 * initializes '*port' appropriately; on failure, returns a positive errno
708 * Retuns ENODEV if the port doesn't exist.
710 * The caller owns the data in 'port' and must free it with
711 * dpif_port_destroy() when it is no longer needed. */
713 dpif_port_query_by_name(const struct dpif
*dpif
, const char *devname
,
714 struct dpif_port
*port
)
716 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, port
);
718 VLOG_DBG_RL(&dpmsg_rl
, "%s: device %s is on port %"PRIu32
,
719 dpif_name(dpif
), devname
, port
->port_no
);
721 memset(port
, 0, sizeof *port
);
723 /* For ENODEV we use DBG level because the caller is probably
724 * interested in whether 'dpif' actually has a port 'devname', so that
725 * it's not an issue worth logging if it doesn't. Other errors are
726 * uncommon and more likely to indicate a real problem. */
727 VLOG_RL(&error_rl
, error
== ENODEV
? VLL_DBG
: VLL_WARN
,
728 "%s: failed to query port %s: %s",
729 dpif_name(dpif
), devname
, ovs_strerror(error
));
734 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
735 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
736 * flows whose packets arrived on port 'port_no'. In the case where the
737 * provider allocates multiple Netlink PIDs to a single port, it may use
738 * 'hash' to spread load among them. The caller need not use a particular
739 * hash function; a 5-tuple hash is suitable.
741 * (The datapath implementation might use some different hash function for
742 * distributing packets received via flow misses among PIDs. This means
743 * that packets received via flow misses might be reordered relative to
744 * packets received via userspace actions. This is not ordinarily a
747 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
748 * allocated to any port, that the client may use for special purposes.
750 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
751 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
752 * disabled and then re-enabled, so a client that does that must be prepared to
753 * update all of the flows that it installed that contain
754 * OVS_ACTION_ATTR_USERSPACE actions. */
756 dpif_port_get_pid(const struct dpif
*dpif
, odp_port_t port_no
, uint32_t hash
)
758 return (dpif
->dpif_class
->port_get_pid
759 ? (dpif
->dpif_class
->port_get_pid
)(dpif
, port_no
, hash
)
763 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
764 * the port's name into the 'name_size' bytes in 'name', ensuring that the
765 * result is null-terminated. On failure, returns a positive errno value and
766 * makes 'name' the empty string. */
768 dpif_port_get_name(struct dpif
*dpif
, odp_port_t port_no
,
769 char *name
, size_t name_size
)
771 struct dpif_port port
;
774 ovs_assert(name_size
> 0);
776 error
= dpif_port_query_by_number(dpif
, port_no
, &port
);
778 ovs_strlcpy(name
, port
.name
, name_size
);
779 dpif_port_destroy(&port
);
786 /* Initializes 'dump' to begin dumping the ports in a dpif.
788 * This function provides no status indication. An error status for the entire
789 * dump operation is provided when it is completed by calling
790 * dpif_port_dump_done().
793 dpif_port_dump_start(struct dpif_port_dump
*dump
, const struct dpif
*dpif
)
796 dump
->error
= dpif
->dpif_class
->port_dump_start(dpif
, &dump
->state
);
797 log_operation(dpif
, "port_dump_start", dump
->error
);
800 /* Attempts to retrieve another port from 'dump', which must have been
801 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
802 * into 'port' and returns true. On failure, returns false.
804 * Failure might indicate an actual error or merely that the last port has been
805 * dumped. An error status for the entire dump operation is provided when it
806 * is completed by calling dpif_port_dump_done().
808 * The dpif owns the data stored in 'port'. It will remain valid until at
809 * least the next time 'dump' is passed to dpif_port_dump_next() or
810 * dpif_port_dump_done(). */
812 dpif_port_dump_next(struct dpif_port_dump
*dump
, struct dpif_port
*port
)
814 const struct dpif
*dpif
= dump
->dpif
;
820 dump
->error
= dpif
->dpif_class
->port_dump_next(dpif
, dump
->state
, port
);
821 if (dump
->error
== EOF
) {
822 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all ports", dpif_name(dpif
));
824 log_operation(dpif
, "port_dump_next", dump
->error
);
828 dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
834 /* Completes port table dump operation 'dump', which must have been initialized
835 * with dpif_port_dump_start(). Returns 0 if the dump operation was
836 * error-free, otherwise a positive errno value describing the problem. */
838 dpif_port_dump_done(struct dpif_port_dump
*dump
)
840 const struct dpif
*dpif
= dump
->dpif
;
842 dump
->error
= dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
843 log_operation(dpif
, "port_dump_done", dump
->error
);
845 return dump
->error
== EOF
? 0 : dump
->error
;
848 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
849 * 'dpif' has changed, this function does one of the following:
851 * - Stores the name of the device that was added to or deleted from 'dpif' in
852 * '*devnamep' and returns 0. The caller is responsible for freeing
853 * '*devnamep' (with free()) when it no longer needs it.
855 * - Returns ENOBUFS and sets '*devnamep' to NULL.
857 * This function may also return 'false positives', where it returns 0 and
858 * '*devnamep' names a device that was not actually added or deleted or it
859 * returns ENOBUFS without any change.
861 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
862 * return other positive errno values to indicate that something has gone
865 dpif_port_poll(const struct dpif
*dpif
, char **devnamep
)
867 int error
= dpif
->dpif_class
->port_poll(dpif
, devnamep
);
874 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
875 * value other than EAGAIN. */
877 dpif_port_poll_wait(const struct dpif
*dpif
)
879 dpif
->dpif_class
->port_poll_wait(dpif
);
882 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
883 * arguments must have been initialized through a call to flow_extract().
884 * 'used' is stored into stats->used. */
886 dpif_flow_stats_extract(const struct flow
*flow
, const struct dp_packet
*packet
,
887 long long int used
, struct dpif_flow_stats
*stats
)
889 stats
->tcp_flags
= ntohs(flow
->tcp_flags
);
890 stats
->n_bytes
= dp_packet_size(packet
);
891 stats
->n_packets
= 1;
895 /* Appends a human-readable representation of 'stats' to 's'. */
897 dpif_flow_stats_format(const struct dpif_flow_stats
*stats
, struct ds
*s
)
899 ds_put_format(s
, "packets:%"PRIu64
", bytes:%"PRIu64
", used:",
900 stats
->n_packets
, stats
->n_bytes
);
902 ds_put_format(s
, "%.3fs", (time_msec() - stats
->used
) / 1000.0);
904 ds_put_format(s
, "never");
906 if (stats
->tcp_flags
) {
907 ds_put_cstr(s
, ", flags:");
908 packet_format_tcp_flags(s
, stats
->tcp_flags
);
912 /* Places the hash of the 'key_len' bytes starting at 'key' into '*hash'. */
914 dpif_flow_hash(const struct dpif
*dpif OVS_UNUSED
,
915 const void *key
, size_t key_len
, ovs_u128
*hash
)
917 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
918 static uint32_t secret
;
920 if (ovsthread_once_start(&once
)) {
921 secret
= random_uint32();
922 ovsthread_once_done(&once
);
924 hash_bytes128(key
, key_len
, secret
, hash
);
925 uuid_set_bits_v4((struct uuid
*)hash
);
928 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
929 * positive errno value. */
931 dpif_flow_flush(struct dpif
*dpif
)
935 COVERAGE_INC(dpif_flow_flush
);
937 error
= dpif
->dpif_class
->flow_flush(dpif
);
938 log_operation(dpif
, "flow_flush", error
);
942 /* Attempts to install 'key' into the datapath, fetches it, then deletes it.
943 * Returns true if the datapath supported installing 'flow', false otherwise.
946 dpif_probe_feature(struct dpif
*dpif
, const char *name
,
947 const struct ofpbuf
*key
, const struct ofpbuf
*actions
,
948 const ovs_u128
*ufid
)
950 struct dpif_flow flow
;
952 uint64_t stub
[DPIF_FLOW_BUFSIZE
/ 8];
953 bool enable_feature
= false;
955 const struct nlattr
*nl_actions
= actions
? actions
->data
: NULL
;
956 const size_t nl_actions_size
= actions
? actions
->size
: 0;
958 /* Use DPIF_FP_MODIFY to cover the case where ovs-vswitchd is killed (and
959 * restarted) at just the right time such that feature probes from the
960 * previous run are still present in the datapath. */
961 error
= dpif_flow_put(dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
| DPIF_FP_PROBE
,
962 key
->data
, key
->size
, NULL
, 0,
963 nl_actions
, nl_actions_size
,
964 ufid
, NON_PMD_CORE_ID
, NULL
);
966 if (error
!= EINVAL
&& error
!= EOVERFLOW
) {
967 VLOG_WARN("%s: %s flow probe failed (%s)",
968 dpif_name(dpif
), name
, ovs_strerror(error
));
973 ofpbuf_use_stack(&reply
, &stub
, sizeof stub
);
974 error
= dpif_flow_get(dpif
, key
->data
, key
->size
, ufid
,
975 NON_PMD_CORE_ID
, &reply
, &flow
);
977 && (!ufid
|| (flow
.ufid_present
978 && ovs_u128_equals(*ufid
, flow
.ufid
)))) {
979 enable_feature
= true;
982 error
= dpif_flow_del(dpif
, key
->data
, key
->size
, ufid
,
983 NON_PMD_CORE_ID
, NULL
);
985 VLOG_WARN("%s: failed to delete %s feature probe flow",
986 dpif_name(dpif
), name
);
989 return enable_feature
;
992 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_GET. */
994 dpif_flow_get(struct dpif
*dpif
,
995 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
996 const unsigned pmd_id
, struct ofpbuf
*buf
, struct dpif_flow
*flow
)
1001 op
.type
= DPIF_OP_FLOW_GET
;
1002 op
.u
.flow_get
.key
= key
;
1003 op
.u
.flow_get
.key_len
= key_len
;
1004 op
.u
.flow_get
.ufid
= ufid
;
1005 op
.u
.flow_get
.pmd_id
= pmd_id
;
1006 op
.u
.flow_get
.buffer
= buf
;
1008 memset(flow
, 0, sizeof *flow
);
1009 op
.u
.flow_get
.flow
= flow
;
1010 op
.u
.flow_get
.flow
->key
= key
;
1011 op
.u
.flow_get
.flow
->key_len
= key_len
;
1014 dpif_operate(dpif
, &opp
, 1);
1019 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_PUT. */
1021 dpif_flow_put(struct dpif
*dpif
, enum dpif_flow_put_flags flags
,
1022 const struct nlattr
*key
, size_t key_len
,
1023 const struct nlattr
*mask
, size_t mask_len
,
1024 const struct nlattr
*actions
, size_t actions_len
,
1025 const ovs_u128
*ufid
, const unsigned pmd_id
,
1026 struct dpif_flow_stats
*stats
)
1028 struct dpif_op
*opp
;
1031 op
.type
= DPIF_OP_FLOW_PUT
;
1032 op
.u
.flow_put
.flags
= flags
;
1033 op
.u
.flow_put
.key
= key
;
1034 op
.u
.flow_put
.key_len
= key_len
;
1035 op
.u
.flow_put
.mask
= mask
;
1036 op
.u
.flow_put
.mask_len
= mask_len
;
1037 op
.u
.flow_put
.actions
= actions
;
1038 op
.u
.flow_put
.actions_len
= actions_len
;
1039 op
.u
.flow_put
.ufid
= ufid
;
1040 op
.u
.flow_put
.pmd_id
= pmd_id
;
1041 op
.u
.flow_put
.stats
= stats
;
1044 dpif_operate(dpif
, &opp
, 1);
1049 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_DEL. */
1051 dpif_flow_del(struct dpif
*dpif
,
1052 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
1053 const unsigned pmd_id
, struct dpif_flow_stats
*stats
)
1055 struct dpif_op
*opp
;
1058 op
.type
= DPIF_OP_FLOW_DEL
;
1059 op
.u
.flow_del
.key
= key
;
1060 op
.u
.flow_del
.key_len
= key_len
;
1061 op
.u
.flow_del
.ufid
= ufid
;
1062 op
.u
.flow_del
.pmd_id
= pmd_id
;
1063 op
.u
.flow_del
.stats
= stats
;
1064 op
.u
.flow_del
.terse
= false;
1067 dpif_operate(dpif
, &opp
, 1);
1072 /* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
1073 * flows in 'dpif'. If 'terse' is true, then only UFID and statistics will
1074 * be returned in the dump. Otherwise, all fields will be returned.
1076 * This function always successfully returns a dpif_flow_dump. Error
1077 * reporting is deferred to dpif_flow_dump_destroy(). */
1078 struct dpif_flow_dump
*
1079 dpif_flow_dump_create(const struct dpif
*dpif
, bool terse
, char *type
)
1081 return dpif
->dpif_class
->flow_dump_create(dpif
, terse
, type
);
1084 /* Destroys 'dump', which must have been created with dpif_flow_dump_create().
1085 * All dpif_flow_dump_thread structures previously created for 'dump' must
1086 * previously have been destroyed.
1088 * Returns 0 if the dump operation was error-free, otherwise a positive errno
1089 * value describing the problem. */
1091 dpif_flow_dump_destroy(struct dpif_flow_dump
*dump
)
1093 const struct dpif
*dpif
= dump
->dpif
;
1094 int error
= dpif
->dpif_class
->flow_dump_destroy(dump
);
1095 log_operation(dpif
, "flow_dump_destroy", error
);
1096 return error
== EOF
? 0 : error
;
1099 /* Returns new thread-local state for use with dpif_flow_dump_next(). */
1100 struct dpif_flow_dump_thread
*
1101 dpif_flow_dump_thread_create(struct dpif_flow_dump
*dump
)
1103 return dump
->dpif
->dpif_class
->flow_dump_thread_create(dump
);
1106 /* Releases 'thread'. */
1108 dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread
*thread
)
1110 thread
->dpif
->dpif_class
->flow_dump_thread_destroy(thread
);
1113 /* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0
1114 * if and only if no flows remained to be retrieved, otherwise a positive
1115 * number reflecting the number of elements in 'flows[]' that were updated.
1116 * The number of flows returned might be less than 'max_flows' because
1117 * fewer than 'max_flows' remained, because this particular datapath does not
1118 * benefit from batching, or because an error occurred partway through
1119 * retrieval. Thus, the caller should continue calling until a 0 return value,
1120 * even if intermediate return values are less than 'max_flows'.
1122 * No error status is immediately provided. An error status for the entire
1123 * dump operation is provided when it is completed by calling
1124 * dpif_flow_dump_destroy().
1126 * All of the data stored into 'flows' is owned by the datapath, not by the
1127 * caller, and the caller must not modify or free it. The datapath guarantees
1128 * that it remains accessible and unchanged until the first of:
1129 * - The next call to dpif_flow_dump_next() for 'thread', or
1130 * - The next rcu quiescent period. */
1132 dpif_flow_dump_next(struct dpif_flow_dump_thread
*thread
,
1133 struct dpif_flow
*flows
, int max_flows
)
1135 struct dpif
*dpif
= thread
->dpif
;
1138 ovs_assert(max_flows
> 0);
1139 n
= dpif
->dpif_class
->flow_dump_next(thread
, flows
, max_flows
);
1141 struct dpif_flow
*f
;
1143 for (f
= flows
; f
< &flows
[n
]
1144 && should_log_flow_message(&this_module
, 0); f
++) {
1145 log_flow_message(dpif
, 0, &this_module
, "flow_dump",
1146 f
->key
, f
->key_len
, f
->mask
, f
->mask_len
,
1147 &f
->ufid
, &f
->stats
, f
->actions
, f
->actions_len
);
1150 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all flows", dpif_name(dpif
));
1155 struct dpif_execute_helper_aux
{
1157 const struct flow
*flow
;
1159 const struct nlattr
*meter_action
; /* Non-NULL, if have a meter action. */
1162 /* This is called for actions that need the context of the datapath to be
1165 dpif_execute_helper_cb(void *aux_
, struct dp_packet_batch
*packets_
,
1166 const struct nlattr
*action
, bool may_steal
)
1168 struct dpif_execute_helper_aux
*aux
= aux_
;
1169 int type
= nl_attr_type(action
);
1170 struct dp_packet
*packet
= packets_
->packets
[0];
1172 ovs_assert(packets_
->count
== 1);
1174 switch ((enum ovs_action_attr
)type
) {
1175 case OVS_ACTION_ATTR_METER
:
1176 /* Maintain a pointer to the first meter action seen. */
1177 if (!aux
->meter_action
) {
1178 aux
->meter_action
= action
;
1182 case OVS_ACTION_ATTR_CT
:
1183 case OVS_ACTION_ATTR_OUTPUT
:
1184 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
1185 case OVS_ACTION_ATTR_TUNNEL_POP
:
1186 case OVS_ACTION_ATTR_USERSPACE
:
1187 case OVS_ACTION_ATTR_RECIRC
: {
1188 struct dpif_execute execute
;
1189 struct ofpbuf execute_actions
;
1190 uint64_t stub
[256 / 8];
1191 struct pkt_metadata
*md
= &packet
->md
;
1193 if (flow_tnl_dst_is_set(&md
->tunnel
) || aux
->meter_action
) {
1194 ofpbuf_use_stub(&execute_actions
, stub
, sizeof stub
);
1196 if (aux
->meter_action
) {
1197 const struct nlattr
*a
= aux
->meter_action
;
1199 /* XXX: This code collects meter actions since the last action
1200 * execution via the datapath to be executed right before the
1201 * current action that needs to be executed by the datapath.
1202 * This is only an approximation, but better than nothing.
1203 * Fundamentally, we should have a mechanism by which the
1204 * datapath could return the result of the meter action so that
1205 * we could execute them at the right order. */
1207 ofpbuf_put(&execute_actions
, a
, NLA_ALIGN(a
->nla_len
));
1208 /* Find next meter action before 'action', if any. */
1210 a
= nl_attr_next(a
);
1211 } while (a
!= action
&&
1212 nl_attr_type(a
) != OVS_ACTION_ATTR_METER
);
1213 } while (a
!= action
);
1216 /* The Linux kernel datapath throws away the tunnel information
1217 * that we supply as metadata. We have to use a "set" action to
1219 if (md
->tunnel
.ip_dst
) {
1220 odp_put_tunnel_action(&md
->tunnel
, &execute_actions
);
1222 ofpbuf_put(&execute_actions
, action
, NLA_ALIGN(action
->nla_len
));
1224 execute
.actions
= execute_actions
.data
;
1225 execute
.actions_len
= execute_actions
.size
;
1227 execute
.actions
= action
;
1228 execute
.actions_len
= NLA_ALIGN(action
->nla_len
);
1231 struct dp_packet
*clone
= NULL
;
1232 uint32_t cutlen
= dp_packet_get_cutlen(packet
);
1233 if (cutlen
&& (type
== OVS_ACTION_ATTR_OUTPUT
1234 || type
== OVS_ACTION_ATTR_TUNNEL_PUSH
1235 || type
== OVS_ACTION_ATTR_TUNNEL_POP
1236 || type
== OVS_ACTION_ATTR_USERSPACE
)) {
1237 dp_packet_reset_cutlen(packet
);
1239 packet
= clone
= dp_packet_clone(packet
);
1241 dp_packet_set_size(packet
, dp_packet_size(packet
) - cutlen
);
1244 execute
.packet
= packet
;
1245 execute
.flow
= aux
->flow
;
1246 execute
.needs_help
= false;
1247 execute
.probe
= false;
1249 aux
->error
= dpif_execute(aux
->dpif
, &execute
);
1250 log_execute_message(aux
->dpif
, &this_module
, &execute
,
1253 dp_packet_delete(clone
);
1255 if (flow_tnl_dst_is_set(&md
->tunnel
) || aux
->meter_action
) {
1256 ofpbuf_uninit(&execute_actions
);
1258 /* Do not re-use the same meters for later output actions. */
1259 aux
->meter_action
= NULL
;
1264 case OVS_ACTION_ATTR_HASH
:
1265 case OVS_ACTION_ATTR_PUSH_VLAN
:
1266 case OVS_ACTION_ATTR_POP_VLAN
:
1267 case OVS_ACTION_ATTR_PUSH_MPLS
:
1268 case OVS_ACTION_ATTR_POP_MPLS
:
1269 case OVS_ACTION_ATTR_SET
:
1270 case OVS_ACTION_ATTR_SET_MASKED
:
1271 case OVS_ACTION_ATTR_SAMPLE
:
1272 case OVS_ACTION_ATTR_TRUNC
:
1273 case OVS_ACTION_ATTR_PUSH_ETH
:
1274 case OVS_ACTION_ATTR_POP_ETH
:
1275 case OVS_ACTION_ATTR_CLONE
:
1276 case OVS_ACTION_ATTR_ENCAP_NSH
:
1277 case OVS_ACTION_ATTR_DECAP_NSH
:
1278 case OVS_ACTION_ATTR_UNSPEC
:
1279 case __OVS_ACTION_ATTR_MAX
:
1282 dp_packet_delete_batch(packets_
, may_steal
);
1285 /* Executes 'execute' by performing most of the actions in userspace and
1286 * passing the fully constructed packets to 'dpif' for output and userspace
1289 * This helps with actions that a given 'dpif' doesn't implement directly. */
1291 dpif_execute_with_help(struct dpif
*dpif
, struct dpif_execute
*execute
)
1293 struct dpif_execute_helper_aux aux
= {dpif
, execute
->flow
, 0, NULL
};
1294 struct dp_packet_batch pb
;
1296 COVERAGE_INC(dpif_execute_with_help
);
1298 dp_packet_batch_init_packet(&pb
, execute
->packet
);
1299 odp_execute_actions(&aux
, &pb
, false, execute
->actions
,
1300 execute
->actions_len
, dpif_execute_helper_cb
);
1304 /* Returns true if the datapath needs help executing 'execute'. */
1306 dpif_execute_needs_help(const struct dpif_execute
*execute
)
1308 return execute
->needs_help
|| nl_attr_oversized(execute
->actions_len
);
1311 /* A dpif_operate() wrapper for performing a single DPIF_OP_EXECUTE. */
1313 dpif_execute(struct dpif
*dpif
, struct dpif_execute
*execute
)
1315 if (execute
->actions_len
) {
1316 struct dpif_op
*opp
;
1319 op
.type
= DPIF_OP_EXECUTE
;
1320 op
.u
.execute
= *execute
;
1323 dpif_operate(dpif
, &opp
, 1);
1331 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1332 * which they are specified. Places each operation's results in the "output"
1333 * members documented in comments, and 0 in the 'error' member on success or a
1334 * positive errno on failure. */
1336 dpif_operate(struct dpif
*dpif
, struct dpif_op
**ops
, size_t n_ops
)
1341 /* Count 'chunk', the number of ops that can be executed without
1342 * needing any help. Ops that need help should be rare, so we
1343 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1344 for (chunk
= 0; chunk
< n_ops
; chunk
++) {
1345 struct dpif_op
*op
= ops
[chunk
];
1347 if (op
->type
== DPIF_OP_EXECUTE
1348 && dpif_execute_needs_help(&op
->u
.execute
)) {
1354 /* Execute a chunk full of ops that the dpif provider can
1355 * handle itself, without help. */
1358 dpif
->dpif_class
->operate(dpif
, ops
, chunk
);
1360 for (i
= 0; i
< chunk
; i
++) {
1361 struct dpif_op
*op
= ops
[i
];
1362 int error
= op
->error
;
1365 case DPIF_OP_FLOW_PUT
: {
1366 struct dpif_flow_put
*put
= &op
->u
.flow_put
;
1368 COVERAGE_INC(dpif_flow_put
);
1369 log_flow_put_message(dpif
, &this_module
, put
, error
);
1370 if (error
&& put
->stats
) {
1371 memset(put
->stats
, 0, sizeof *put
->stats
);
1376 case DPIF_OP_FLOW_GET
: {
1377 struct dpif_flow_get
*get
= &op
->u
.flow_get
;
1379 COVERAGE_INC(dpif_flow_get
);
1381 memset(get
->flow
, 0, sizeof *get
->flow
);
1383 log_flow_get_message(dpif
, &this_module
, get
, error
);
1388 case DPIF_OP_FLOW_DEL
: {
1389 struct dpif_flow_del
*del
= &op
->u
.flow_del
;
1391 COVERAGE_INC(dpif_flow_del
);
1392 log_flow_del_message(dpif
, &this_module
, del
, error
);
1393 if (error
&& del
->stats
) {
1394 memset(del
->stats
, 0, sizeof *del
->stats
);
1399 case DPIF_OP_EXECUTE
:
1400 COVERAGE_INC(dpif_execute
);
1401 log_execute_message(dpif
, &this_module
, &op
->u
.execute
,
1410 /* Help the dpif provider to execute one op. */
1411 struct dpif_op
*op
= ops
[0];
1413 COVERAGE_INC(dpif_execute
);
1414 op
->error
= dpif_execute_with_help(dpif
, &op
->u
.execute
);
1421 /* Returns a string that represents 'type', for use in log messages. */
1423 dpif_upcall_type_to_string(enum dpif_upcall_type type
)
1426 case DPIF_UC_MISS
: return "miss";
1427 case DPIF_UC_ACTION
: return "action";
1428 case DPIF_N_UC_TYPES
: default: return "<unknown>";
1432 /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1433 * if successful, otherwise a positive errno value.
1435 * Turning packet receive off and then back on may change the Netlink PID
1436 * assignments returned by dpif_port_get_pid(). If the client does this, it
1437 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1438 * using the new PID assignment. */
1440 dpif_recv_set(struct dpif
*dpif
, bool enable
)
1444 if (dpif
->dpif_class
->recv_set
) {
1445 error
= dpif
->dpif_class
->recv_set(dpif
, enable
);
1446 log_operation(dpif
, "recv_set", error
);
1451 /* Refreshes the poll loops and Netlink sockets associated to each port,
1452 * when the number of upcall handlers (upcall receiving thread) is changed
1453 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
1456 * Since multiple upcall handlers can read upcalls simultaneously from
1457 * 'dpif', each port can have multiple Netlink sockets, one per upcall
1458 * handler. So, handlers_set() is responsible for the following tasks:
1460 * When receiving upcall is enabled, extends or creates the
1461 * configuration to support:
1463 * - 'n_handlers' Netlink sockets for each port.
1465 * - 'n_handlers' poll loops, one for each upcall handler.
1467 * - registering the Netlink sockets for the same upcall handler to
1468 * the corresponding poll loop.
1470 * Returns 0 if successful, otherwise a positive errno value. */
1472 dpif_handlers_set(struct dpif
*dpif
, uint32_t n_handlers
)
1476 if (dpif
->dpif_class
->handlers_set
) {
1477 error
= dpif
->dpif_class
->handlers_set(dpif
, n_handlers
);
1478 log_operation(dpif
, "handlers_set", error
);
1484 dpif_register_dp_purge_cb(struct dpif
*dpif
, dp_purge_callback
*cb
, void *aux
)
1486 if (dpif
->dpif_class
->register_dp_purge_cb
) {
1487 dpif
->dpif_class
->register_dp_purge_cb(dpif
, cb
, aux
);
1492 dpif_register_upcall_cb(struct dpif
*dpif
, upcall_callback
*cb
, void *aux
)
1494 if (dpif
->dpif_class
->register_upcall_cb
) {
1495 dpif
->dpif_class
->register_upcall_cb(dpif
, cb
, aux
);
1500 dpif_enable_upcall(struct dpif
*dpif
)
1502 if (dpif
->dpif_class
->enable_upcall
) {
1503 dpif
->dpif_class
->enable_upcall(dpif
);
1508 dpif_disable_upcall(struct dpif
*dpif
)
1510 if (dpif
->dpif_class
->disable_upcall
) {
1511 dpif
->dpif_class
->disable_upcall(dpif
);
1516 dpif_print_packet(struct dpif
*dpif
, struct dpif_upcall
*upcall
)
1518 if (!VLOG_DROP_DBG(&dpmsg_rl
)) {
1522 packet
= ofp_dp_packet_to_string(&upcall
->packet
);
1525 odp_flow_key_format(upcall
->key
, upcall
->key_len
, &flow
);
1527 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1528 dpif_name(dpif
), dpif_upcall_type_to_string(upcall
->type
),
1529 ds_cstr(&flow
), packet
);
1536 /* Pass custom configuration to the datapath implementation. Some of the
1537 * changes can be postponed until dpif_run() is called. */
1539 dpif_set_config(struct dpif
*dpif
, const struct smap
*cfg
)
1543 if (dpif
->dpif_class
->set_config
) {
1544 error
= dpif
->dpif_class
->set_config(dpif
, cfg
);
1546 log_operation(dpif
, "set_config", error
);
1553 /* Polls for an upcall from 'dpif' for an upcall handler. Since there can
1554 * be multiple poll loops, 'handler_id' is needed as index to identify the
1555 * corresponding poll loop. If successful, stores the upcall into '*upcall',
1556 * using 'buf' for storage. Should only be called if 'recv_set' has been used
1557 * to enable receiving packets from 'dpif'.
1559 * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
1560 * 'buf', so their memory cannot be freed separately from 'buf'.
1562 * The caller owns the data of 'upcall->packet' and may modify it. If
1563 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
1564 * will be reallocated. This requires the data of 'upcall->packet' to be
1565 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
1566 * when an error is returned, the 'upcall->packet' may be uninitialized
1567 * and should not be released.
1569 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1570 * if no upcall is immediately available. */
1572 dpif_recv(struct dpif
*dpif
, uint32_t handler_id
, struct dpif_upcall
*upcall
,
1577 if (dpif
->dpif_class
->recv
) {
1578 error
= dpif
->dpif_class
->recv(dpif
, handler_id
, upcall
, buf
);
1580 dpif_print_packet(dpif
, upcall
);
1581 } else if (error
!= EAGAIN
) {
1582 log_operation(dpif
, "recv", error
);
1588 /* Discards all messages that would otherwise be received by dpif_recv() on
1591 dpif_recv_purge(struct dpif
*dpif
)
1593 COVERAGE_INC(dpif_purge
);
1594 if (dpif
->dpif_class
->recv_purge
) {
1595 dpif
->dpif_class
->recv_purge(dpif
);
1599 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
1600 * 'dpif' has a message queued to be received with the recv member
1601 * function. Since there can be multiple poll loops, 'handler_id' is
1602 * needed as index to identify the corresponding poll loop. */
1604 dpif_recv_wait(struct dpif
*dpif
, uint32_t handler_id
)
1606 if (dpif
->dpif_class
->recv_wait
) {
1607 dpif
->dpif_class
->recv_wait(dpif
, handler_id
);
1612 * Return the datapath version. Caller is responsible for freeing
1616 dpif_get_dp_version(const struct dpif
*dpif
)
1618 char *version
= NULL
;
1620 if (dpif
->dpif_class
->get_datapath_version
) {
1621 version
= dpif
->dpif_class
->get_datapath_version();
1627 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1628 * and '*engine_id', respectively. */
1630 dpif_get_netflow_ids(const struct dpif
*dpif
,
1631 uint8_t *engine_type
, uint8_t *engine_id
)
1633 *engine_type
= dpif
->netflow_engine_type
;
1634 *engine_id
= dpif
->netflow_engine_id
;
1637 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1638 * value used for setting packet priority.
1639 * On success, returns 0 and stores the priority into '*priority'.
1640 * On failure, returns a positive errno value and stores 0 into '*priority'. */
1642 dpif_queue_to_priority(const struct dpif
*dpif
, uint32_t queue_id
,
1645 int error
= (dpif
->dpif_class
->queue_to_priority
1646 ? dpif
->dpif_class
->queue_to_priority(dpif
, queue_id
,
1652 log_operation(dpif
, "queue_to_priority", error
);
1657 dpif_init(struct dpif
*dpif
, const struct dpif_class
*dpif_class
,
1659 uint8_t netflow_engine_type
, uint8_t netflow_engine_id
)
1661 dpif
->dpif_class
= dpif_class
;
1662 dpif
->base_name
= xstrdup(name
);
1663 dpif
->full_name
= xasprintf("%s@%s", dpif_class
->type
, name
);
1664 dpif
->netflow_engine_type
= netflow_engine_type
;
1665 dpif
->netflow_engine_id
= netflow_engine_id
;
1668 /* Undoes the results of initialization.
1670 * Normally this function only needs to be called from dpif_close().
1671 * However, it may be called by providers due to an error on opening
1672 * that occurs after initialization. It this case dpif_close() would
1673 * never be called. */
1675 dpif_uninit(struct dpif
*dpif
, bool close
)
1677 char *base_name
= dpif
->base_name
;
1678 char *full_name
= dpif
->full_name
;
1681 dpif
->dpif_class
->close(dpif
);
1689 log_operation(const struct dpif
*dpif
, const char *operation
, int error
)
1692 VLOG_DBG_RL(&dpmsg_rl
, "%s: %s success", dpif_name(dpif
), operation
);
1693 } else if (ofperr_is_valid(error
)) {
1694 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1695 dpif_name(dpif
), operation
, ofperr_get_name(error
));
1697 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1698 dpif_name(dpif
), operation
, ovs_strerror(error
));
1702 static enum vlog_level
1703 flow_message_log_level(int error
)
1705 /* If flows arrive in a batch, userspace may push down multiple
1706 * unique flow definitions that overlap when wildcards are applied.
1707 * Kernels that support flow wildcarding will reject these flows as
1708 * duplicates (EEXIST), so lower the log level to debug for these
1709 * types of messages. */
1710 return (error
&& error
!= EEXIST
) ? VLL_WARN
: VLL_DBG
;
1714 should_log_flow_message(const struct vlog_module
*module
, int error
)
1716 return !vlog_should_drop(module
, flow_message_log_level(error
),
1717 error
? &error_rl
: &dpmsg_rl
);
1721 log_flow_message(const struct dpif
*dpif
, int error
,
1722 const struct vlog_module
*module
,
1723 const char *operation
,
1724 const struct nlattr
*key
, size_t key_len
,
1725 const struct nlattr
*mask
, size_t mask_len
,
1726 const ovs_u128
*ufid
, const struct dpif_flow_stats
*stats
,
1727 const struct nlattr
*actions
, size_t actions_len
)
1729 struct ds ds
= DS_EMPTY_INITIALIZER
;
1730 ds_put_format(&ds
, "%s: ", dpif_name(dpif
));
1732 ds_put_cstr(&ds
, "failed to ");
1734 ds_put_format(&ds
, "%s ", operation
);
1736 ds_put_format(&ds
, "(%s) ", ovs_strerror(error
));
1739 odp_format_ufid(ufid
, &ds
);
1740 ds_put_cstr(&ds
, " ");
1742 odp_flow_format(key
, key_len
, mask
, mask_len
, NULL
, &ds
, true);
1744 ds_put_cstr(&ds
, ", ");
1745 dpif_flow_stats_format(stats
, &ds
);
1747 if (actions
|| actions_len
) {
1748 ds_put_cstr(&ds
, ", actions:");
1749 format_odp_actions(&ds
, actions
, actions_len
, NULL
);
1751 vlog(module
, flow_message_log_level(error
), "%s", ds_cstr(&ds
));
1756 log_flow_put_message(const struct dpif
*dpif
,
1757 const struct vlog_module
*module
,
1758 const struct dpif_flow_put
*put
,
1761 if (should_log_flow_message(module
, error
)
1762 && !(put
->flags
& DPIF_FP_PROBE
)) {
1766 ds_put_cstr(&s
, "put");
1767 if (put
->flags
& DPIF_FP_CREATE
) {
1768 ds_put_cstr(&s
, "[create]");
1770 if (put
->flags
& DPIF_FP_MODIFY
) {
1771 ds_put_cstr(&s
, "[modify]");
1773 if (put
->flags
& DPIF_FP_ZERO_STATS
) {
1774 ds_put_cstr(&s
, "[zero]");
1776 log_flow_message(dpif
, error
, module
, ds_cstr(&s
),
1777 put
->key
, put
->key_len
, put
->mask
, put
->mask_len
,
1778 put
->ufid
, put
->stats
, put
->actions
,
1785 log_flow_del_message(const struct dpif
*dpif
,
1786 const struct vlog_module
*module
,
1787 const struct dpif_flow_del
*del
,
1790 if (should_log_flow_message(module
, error
)) {
1791 log_flow_message(dpif
, error
, module
, "flow_del",
1792 del
->key
, del
->key_len
,
1793 NULL
, 0, del
->ufid
, !error
? del
->stats
: NULL
,
1798 /* Logs that 'execute' was executed on 'dpif' and completed with errno 'error'
1799 * (0 for success). 'subexecute' should be true if the execution is a result
1800 * of breaking down a larger execution that needed help, false otherwise.
1803 * XXX In theory, the log message could be deceptive because this function is
1804 * called after the dpif_provider's '->execute' function, which is allowed to
1805 * modify execute->packet and execute->md. In practice, though:
1807 * - dpif-netlink doesn't modify execute->packet or execute->md.
1809 * - dpif-netdev does modify them but it is less likely to have problems
1810 * because it is built into ovs-vswitchd and cannot have version skew,
1813 * It would still be better to avoid the potential problem. I don't know of a
1814 * good way to do that, though, that isn't expensive. */
1816 log_execute_message(const struct dpif
*dpif
,
1817 const struct vlog_module
*module
,
1818 const struct dpif_execute
*execute
,
1819 bool subexecute
, int error
)
1821 if (!(error
? VLOG_DROP_WARN(&error_rl
) : VLOG_DROP_DBG(&dpmsg_rl
))
1822 && !execute
->probe
) {
1823 struct ds ds
= DS_EMPTY_INITIALIZER
;
1825 uint64_t stub
[1024 / 8];
1826 struct ofpbuf md
= OFPBUF_STUB_INITIALIZER(stub
);
1828 packet
= ofp_packet_to_string(dp_packet_data(execute
->packet
),
1829 dp_packet_size(execute
->packet
),
1830 execute
->packet
->packet_type
);
1831 odp_key_from_dp_packet(&md
, execute
->packet
);
1832 ds_put_format(&ds
, "%s: %sexecute ",
1834 (subexecute
? "sub-"
1835 : dpif_execute_needs_help(execute
) ? "super-"
1837 format_odp_actions(&ds
, execute
->actions
, execute
->actions_len
, NULL
);
1839 ds_put_format(&ds
, " failed (%s)", ovs_strerror(error
));
1841 ds_put_format(&ds
, " on packet %s", packet
);
1842 ds_put_format(&ds
, " with metadata ");
1843 odp_flow_format(md
.data
, md
.size
, NULL
, 0, NULL
, &ds
, true);
1844 ds_put_format(&ds
, " mtu %d", execute
->mtu
);
1845 vlog(module
, error
? VLL_WARN
: VLL_DBG
, "%s", ds_cstr(&ds
));
1853 log_flow_get_message(const struct dpif
*dpif
,
1854 const struct vlog_module
*module
,
1855 const struct dpif_flow_get
*get
,
1858 if (should_log_flow_message(module
, error
)) {
1859 log_flow_message(dpif
, error
, module
, "flow_get",
1860 get
->key
, get
->key_len
,
1861 get
->flow
->mask
, get
->flow
->mask_len
,
1862 get
->ufid
, &get
->flow
->stats
,
1863 get
->flow
->actions
, get
->flow
->actions_len
);
1868 dpif_supports_tnl_push_pop(const struct dpif
*dpif
)
1870 return dpif_is_netdev(dpif
);
1875 dpif_meter_get_features(const struct dpif
*dpif
,
1876 struct ofputil_meter_features
*features
)
1878 memset(features
, 0, sizeof *features
);
1879 if (dpif
->dpif_class
->meter_get_features
) {
1880 dpif
->dpif_class
->meter_get_features(dpif
, features
);
1884 /* Adds or modifies meter identified by 'meter_id' in 'dpif'. If '*meter_id'
1885 * is UINT32_MAX, adds a new meter, otherwise modifies an existing meter.
1887 * If meter is successfully added, sets '*meter_id' to the new meter's
1890 dpif_meter_set(struct dpif
*dpif
, ofproto_meter_id
*meter_id
,
1891 struct ofputil_meter_config
*config
)
1895 COVERAGE_INC(dpif_meter_set
);
1897 error
= dpif
->dpif_class
->meter_set(dpif
, meter_id
, config
);
1899 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" set",
1900 dpif_name(dpif
), meter_id
->uint32
);
1902 VLOG_WARN_RL(&error_rl
, "%s: failed to set DPIF meter %"PRIu32
": %s",
1903 dpif_name(dpif
), meter_id
->uint32
, ovs_strerror(error
));
1904 meter_id
->uint32
= UINT32_MAX
;
1910 dpif_meter_get(const struct dpif
*dpif
, ofproto_meter_id meter_id
,
1911 struct ofputil_meter_stats
*stats
, uint16_t n_bands
)
1915 COVERAGE_INC(dpif_meter_get
);
1917 error
= dpif
->dpif_class
->meter_get(dpif
, meter_id
, stats
, n_bands
);
1919 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" get stats",
1920 dpif_name(dpif
), meter_id
.uint32
);
1922 VLOG_WARN_RL(&error_rl
,
1923 "%s: failed to get DPIF meter %"PRIu32
" stats: %s",
1924 dpif_name(dpif
), meter_id
.uint32
, ovs_strerror(error
));
1925 stats
->packet_in_count
= ~0;
1926 stats
->byte_in_count
= ~0;
1933 dpif_meter_del(struct dpif
*dpif
, ofproto_meter_id meter_id
,
1934 struct ofputil_meter_stats
*stats
, uint16_t n_bands
)
1938 COVERAGE_INC(dpif_meter_del
);
1940 error
= dpif
->dpif_class
->meter_del(dpif
, meter_id
, stats
, n_bands
);
1942 VLOG_DBG_RL(&dpmsg_rl
, "%s: DPIF meter %"PRIu32
" deleted",
1943 dpif_name(dpif
), meter_id
.uint32
);
1945 VLOG_WARN_RL(&error_rl
,
1946 "%s: failed to delete DPIF meter %"PRIu32
": %s",
1947 dpif_name(dpif
), meter_id
.uint32
, ovs_strerror(error
));
1949 stats
->packet_in_count
= ~0;
1950 stats
->byte_in_count
= ~0;