2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 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 "dynamic-string.h"
32 #include "odp-execute.h"
34 #include "ofp-errors.h"
35 #include "ofp-print.h"
38 #include "packet-dpif.h"
40 #include "poll-loop.h"
41 #include "route-table.h"
46 #include "tnl-arp-cache.h"
47 #include "tnl-ports.h"
50 #include "openvswitch/vlog.h"
52 VLOG_DEFINE_THIS_MODULE(dpif
);
54 COVERAGE_DEFINE(dpif_destroy
);
55 COVERAGE_DEFINE(dpif_port_add
);
56 COVERAGE_DEFINE(dpif_port_del
);
57 COVERAGE_DEFINE(dpif_flow_flush
);
58 COVERAGE_DEFINE(dpif_flow_get
);
59 COVERAGE_DEFINE(dpif_flow_put
);
60 COVERAGE_DEFINE(dpif_flow_del
);
61 COVERAGE_DEFINE(dpif_execute
);
62 COVERAGE_DEFINE(dpif_purge
);
63 COVERAGE_DEFINE(dpif_execute_with_help
);
65 static const struct dpif_class
*base_dpif_classes
[] = {
66 #if defined(__linux__) || defined(_WIN32)
72 struct registered_dpif_class
{
73 const struct dpif_class
*dpif_class
;
76 static struct shash dpif_classes
= SHASH_INITIALIZER(&dpif_classes
);
77 static struct sset dpif_blacklist
= SSET_INITIALIZER(&dpif_blacklist
);
79 /* Protects 'dpif_classes', including the refcount, and 'dpif_blacklist'. */
80 static struct ovs_mutex dpif_mutex
= OVS_MUTEX_INITIALIZER
;
82 /* Rate limit for individual messages going to or from the datapath, output at
83 * DBG level. This is very high because, if these are enabled, it is because
84 * we really need to see them. */
85 static struct vlog_rate_limit dpmsg_rl
= VLOG_RATE_LIMIT_INIT(600, 600);
87 /* Not really much point in logging many dpif errors. */
88 static struct vlog_rate_limit error_rl
= VLOG_RATE_LIMIT_INIT(60, 5);
90 static void log_flow_message(const struct dpif
*dpif
, int error
,
91 const char *operation
,
92 const struct nlattr
*key
, size_t key_len
,
93 const struct nlattr
*mask
, size_t mask_len
,
95 const struct dpif_flow_stats
*stats
,
96 const struct nlattr
*actions
, size_t actions_len
);
97 static void log_operation(const struct dpif
*, const char *operation
,
99 static bool should_log_flow_message(int error
);
100 static void log_flow_put_message(struct dpif
*, const struct dpif_flow_put
*,
102 static void log_flow_del_message(struct dpif
*, const struct dpif_flow_del
*,
104 static void log_execute_message(struct dpif
*, const struct dpif_execute
*,
105 bool subexecute
, int error
);
106 static void log_flow_get_message(const struct dpif
*,
107 const struct dpif_flow_get
*, int error
);
109 /* Incremented whenever tnl route, arp, etc changes. */
110 struct seq
*tnl_conf_seq
;
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_arp_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
;
139 if (sset_contains(&dpif_blacklist
, new_class
->type
)) {
140 VLOG_DBG("attempted to register blacklisted provider: %s",
145 if (shash_find(&dpif_classes
, new_class
->type
)) {
146 VLOG_WARN("attempted to register duplicate datapath provider: %s",
151 registered_class
= xmalloc(sizeof *registered_class
);
152 registered_class
->dpif_class
= new_class
;
153 registered_class
->refcount
= 0;
155 shash_add(&dpif_classes
, new_class
->type
, registered_class
);
160 /* Registers a new datapath provider. After successful registration, new
161 * datapaths of that type can be opened using dpif_open(). */
163 dp_register_provider(const struct dpif_class
*new_class
)
167 ovs_mutex_lock(&dpif_mutex
);
168 error
= dp_register_provider__(new_class
);
169 ovs_mutex_unlock(&dpif_mutex
);
174 /* Unregisters a datapath provider. 'type' must have been previously
175 * registered and not currently be in use by any dpifs. After unregistration
176 * new datapaths of that type cannot be opened using dpif_open(). */
178 dp_unregister_provider__(const char *type
)
180 struct shash_node
*node
;
181 struct registered_dpif_class
*registered_class
;
183 node
= shash_find(&dpif_classes
, type
);
185 VLOG_WARN("attempted to unregister a datapath provider that is not "
186 "registered: %s", type
);
190 registered_class
= node
->data
;
191 if (registered_class
->refcount
) {
192 VLOG_WARN("attempted to unregister in use datapath provider: %s", type
);
196 shash_delete(&dpif_classes
, node
);
197 free(registered_class
);
202 /* Unregisters a datapath provider. 'type' must have been previously
203 * registered and not currently be in use by any dpifs. After unregistration
204 * new datapaths of that type cannot be opened using dpif_open(). */
206 dp_unregister_provider(const char *type
)
212 ovs_mutex_lock(&dpif_mutex
);
213 error
= dp_unregister_provider__(type
);
214 ovs_mutex_unlock(&dpif_mutex
);
219 /* Blacklists a provider. Causes future calls of dp_register_provider() with
220 * a dpif_class which implements 'type' to fail. */
222 dp_blacklist_provider(const char *type
)
224 ovs_mutex_lock(&dpif_mutex
);
225 sset_add(&dpif_blacklist
, type
);
226 ovs_mutex_unlock(&dpif_mutex
);
229 /* Adds the types of all currently registered datapath providers to 'types'.
230 * The caller must first initialize the sset. */
232 dp_enumerate_types(struct sset
*types
)
234 struct shash_node
*node
;
238 ovs_mutex_lock(&dpif_mutex
);
239 SHASH_FOR_EACH(node
, &dpif_classes
) {
240 const struct registered_dpif_class
*registered_class
= node
->data
;
241 sset_add(types
, registered_class
->dpif_class
->type
);
243 ovs_mutex_unlock(&dpif_mutex
);
247 dp_class_unref(struct registered_dpif_class
*rc
)
249 ovs_mutex_lock(&dpif_mutex
);
250 ovs_assert(rc
->refcount
);
252 ovs_mutex_unlock(&dpif_mutex
);
255 static struct registered_dpif_class
*
256 dp_class_lookup(const char *type
)
258 struct registered_dpif_class
*rc
;
260 ovs_mutex_lock(&dpif_mutex
);
261 rc
= shash_find_data(&dpif_classes
, type
);
265 ovs_mutex_unlock(&dpif_mutex
);
270 /* Clears 'names' and enumerates the names of all known created datapaths with
271 * the given 'type'. The caller must first initialize the sset. Returns 0 if
272 * successful, otherwise a positive errno value.
274 * Some kinds of datapaths might not be practically enumerable. This is not
275 * considered an error. */
277 dp_enumerate_names(const char *type
, struct sset
*names
)
279 struct registered_dpif_class
*registered_class
;
280 const struct dpif_class
*dpif_class
;
286 registered_class
= dp_class_lookup(type
);
287 if (!registered_class
) {
288 VLOG_WARN("could not enumerate unknown type: %s", type
);
292 dpif_class
= registered_class
->dpif_class
;
293 error
= (dpif_class
->enumerate
294 ? dpif_class
->enumerate(names
, dpif_class
)
297 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class
->type
,
298 ovs_strerror(error
));
300 dp_class_unref(registered_class
);
305 /* Parses 'datapath_name_', which is of the form [type@]name into its
306 * component pieces. 'name' and 'type' must be freed by the caller.
308 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
310 dp_parse_name(const char *datapath_name_
, char **name
, char **type
)
312 char *datapath_name
= xstrdup(datapath_name_
);
315 separator
= strchr(datapath_name
, '@');
318 *type
= datapath_name
;
319 *name
= xstrdup(dpif_normalize_type(separator
+ 1));
321 *name
= datapath_name
;
322 *type
= xstrdup(dpif_normalize_type(NULL
));
327 do_open(const char *name
, const char *type
, bool create
, struct dpif
**dpifp
)
329 struct dpif
*dpif
= NULL
;
331 struct registered_dpif_class
*registered_class
;
335 type
= dpif_normalize_type(type
);
336 registered_class
= dp_class_lookup(type
);
337 if (!registered_class
) {
338 VLOG_WARN("could not create datapath %s of unknown type %s", name
,
340 error
= EAFNOSUPPORT
;
344 error
= registered_class
->dpif_class
->open(registered_class
->dpif_class
,
345 name
, create
, &dpif
);
347 ovs_assert(dpif
->dpif_class
== registered_class
->dpif_class
);
349 dp_class_unref(registered_class
);
353 *dpifp
= error
? NULL
: dpif
;
357 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
358 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
359 * the empty string to specify the default system type. Returns 0 if
360 * successful, otherwise a positive errno value. On success stores a pointer
361 * to the datapath in '*dpifp', otherwise a null pointer. */
363 dpif_open(const char *name
, const char *type
, struct dpif
**dpifp
)
365 return do_open(name
, type
, false, dpifp
);
368 /* Tries to create and open a new datapath with the given 'name' and 'type'.
369 * 'type' may be either NULL or the empty string to specify the default system
370 * type. Will fail if a datapath with 'name' and 'type' already exists.
371 * Returns 0 if successful, otherwise a positive errno value. On success
372 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
374 dpif_create(const char *name
, const char *type
, struct dpif
**dpifp
)
376 return do_open(name
, type
, true, dpifp
);
379 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
380 * does not exist. 'type' may be either NULL or the empty string to specify
381 * the default system type. Returns 0 if successful, otherwise a positive
382 * errno value. On success stores a pointer to the datapath in '*dpifp',
383 * otherwise a null pointer. */
385 dpif_create_and_open(const char *name
, const char *type
, struct dpif
**dpifp
)
389 error
= dpif_create(name
, type
, dpifp
);
390 if (error
== EEXIST
|| error
== EBUSY
) {
391 error
= dpif_open(name
, type
, dpifp
);
393 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
394 name
, ovs_strerror(error
));
397 VLOG_WARN("failed to create datapath %s: %s",
398 name
, ovs_strerror(error
));
403 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
404 * itself; call dpif_delete() first, instead, if that is desirable. */
406 dpif_close(struct dpif
*dpif
)
409 struct registered_dpif_class
*rc
;
411 rc
= shash_find_data(&dpif_classes
, dpif
->dpif_class
->type
);
412 dpif_uninit(dpif
, true);
417 /* Performs periodic work needed by 'dpif'. */
419 dpif_run(struct dpif
*dpif
)
421 if (dpif
->dpif_class
->run
) {
422 return dpif
->dpif_class
->run(dpif
);
427 /* Arranges for poll_block() to wake up when dp_run() needs to be called for
430 dpif_wait(struct dpif
*dpif
)
432 if (dpif
->dpif_class
->wait
) {
433 dpif
->dpif_class
->wait(dpif
);
437 /* Returns the name of datapath 'dpif' prefixed with the type
438 * (for use in log messages). */
440 dpif_name(const struct dpif
*dpif
)
442 return dpif
->full_name
;
445 /* Returns the name of datapath 'dpif' without the type
446 * (for use in device names). */
448 dpif_base_name(const struct dpif
*dpif
)
450 return dpif
->base_name
;
453 /* Returns the type of datapath 'dpif'. */
455 dpif_type(const struct dpif
*dpif
)
457 return dpif
->dpif_class
->type
;
460 /* Returns the fully spelled out name for the given datapath 'type'.
462 * Normalized type string can be compared with strcmp(). Unnormalized type
463 * string might be the same even if they have different spellings. */
465 dpif_normalize_type(const char *type
)
467 return type
&& type
[0] ? type
: "system";
470 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
471 * ports. After calling this function, it does not make sense to pass 'dpif'
472 * to any functions other than dpif_name() or dpif_close(). */
474 dpif_delete(struct dpif
*dpif
)
478 COVERAGE_INC(dpif_destroy
);
480 error
= dpif
->dpif_class
->destroy(dpif
);
481 log_operation(dpif
, "delete", error
);
485 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
486 * otherwise a positive errno value. */
488 dpif_get_dp_stats(const struct dpif
*dpif
, struct dpif_dp_stats
*stats
)
490 int error
= dpif
->dpif_class
->get_stats(dpif
, stats
);
492 memset(stats
, 0, sizeof *stats
);
494 log_operation(dpif
, "get_stats", error
);
499 dpif_port_open_type(const char *datapath_type
, const char *port_type
)
501 struct registered_dpif_class
*rc
;
503 datapath_type
= dpif_normalize_type(datapath_type
);
505 ovs_mutex_lock(&dpif_mutex
);
506 rc
= shash_find_data(&dpif_classes
, datapath_type
);
507 if (rc
&& rc
->dpif_class
->port_open_type
) {
508 port_type
= rc
->dpif_class
->port_open_type(rc
->dpif_class
, port_type
);
510 ovs_mutex_unlock(&dpif_mutex
);
515 /* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
516 * non-null and its value is not ODPP_NONE, then attempts to use the
517 * value as the port number.
519 * If successful, returns 0 and sets '*port_nop' to the new port's port
520 * number (if 'port_nop' is non-null). On failure, returns a positive
521 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
524 dpif_port_add(struct dpif
*dpif
, struct netdev
*netdev
, odp_port_t
*port_nop
)
526 const char *netdev_name
= netdev_get_name(netdev
);
527 odp_port_t port_no
= ODPP_NONE
;
530 COVERAGE_INC(dpif_port_add
);
536 error
= dpif
->dpif_class
->port_add(dpif
, netdev
, &port_no
);
538 VLOG_DBG_RL(&dpmsg_rl
, "%s: added %s as port %"PRIu32
,
539 dpif_name(dpif
), netdev_name
, port_no
);
541 VLOG_WARN_RL(&error_rl
, "%s: failed to add %s as port: %s",
542 dpif_name(dpif
), netdev_name
, ovs_strerror(error
));
551 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
552 * otherwise a positive errno value. */
554 dpif_port_del(struct dpif
*dpif
, odp_port_t port_no
)
558 COVERAGE_INC(dpif_port_del
);
560 error
= dpif
->dpif_class
->port_del(dpif
, port_no
);
562 VLOG_DBG_RL(&dpmsg_rl
, "%s: port_del(%"PRIu32
")",
563 dpif_name(dpif
), port_no
);
565 log_operation(dpif
, "port_del", error
);
570 /* Makes a deep copy of 'src' into 'dst'. */
572 dpif_port_clone(struct dpif_port
*dst
, const struct dpif_port
*src
)
574 dst
->name
= xstrdup(src
->name
);
575 dst
->type
= xstrdup(src
->type
);
576 dst
->port_no
= src
->port_no
;
579 /* Frees memory allocated to members of 'dpif_port'.
581 * Do not call this function on a dpif_port obtained from
582 * dpif_port_dump_next(): that function retains ownership of the data in the
585 dpif_port_destroy(struct dpif_port
*dpif_port
)
587 free(dpif_port
->name
);
588 free(dpif_port
->type
);
591 /* Checks if port named 'devname' exists in 'dpif'. If so, returns
592 * true; otherwise, returns false. */
594 dpif_port_exists(const struct dpif
*dpif
, const char *devname
)
596 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, NULL
);
597 if (error
!= 0 && error
!= ENOENT
&& error
!= ENODEV
) {
598 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %s: %s",
599 dpif_name(dpif
), devname
, ovs_strerror(error
));
605 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
606 * initializes '*port' appropriately; on failure, returns a positive errno
609 * The caller owns the data in 'port' and must free it with
610 * dpif_port_destroy() when it is no longer needed. */
612 dpif_port_query_by_number(const struct dpif
*dpif
, odp_port_t port_no
,
613 struct dpif_port
*port
)
615 int error
= dpif
->dpif_class
->port_query_by_number(dpif
, port_no
, port
);
617 VLOG_DBG_RL(&dpmsg_rl
, "%s: port %"PRIu32
" is device %s",
618 dpif_name(dpif
), port_no
, port
->name
);
620 memset(port
, 0, sizeof *port
);
621 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %"PRIu32
": %s",
622 dpif_name(dpif
), port_no
, ovs_strerror(error
));
627 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
628 * initializes '*port' appropriately; on failure, returns a positive errno
631 * The caller owns the data in 'port' and must free it with
632 * dpif_port_destroy() when it is no longer needed. */
634 dpif_port_query_by_name(const struct dpif
*dpif
, const char *devname
,
635 struct dpif_port
*port
)
637 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, port
);
639 VLOG_DBG_RL(&dpmsg_rl
, "%s: device %s is on port %"PRIu32
,
640 dpif_name(dpif
), devname
, port
->port_no
);
642 memset(port
, 0, sizeof *port
);
644 /* For ENOENT or ENODEV we use DBG level because the caller is probably
645 * interested in whether 'dpif' actually has a port 'devname', so that
646 * it's not an issue worth logging if it doesn't. Other errors are
647 * uncommon and more likely to indicate a real problem. */
649 error
== ENOENT
|| error
== ENODEV
? VLL_DBG
: VLL_WARN
,
650 "%s: failed to query port %s: %s",
651 dpif_name(dpif
), devname
, ovs_strerror(error
));
656 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
657 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
658 * flows whose packets arrived on port 'port_no'. In the case where the
659 * provider allocates multiple Netlink PIDs to a single port, it may use
660 * 'hash' to spread load among them. The caller need not use a particular
661 * hash function; a 5-tuple hash is suitable.
663 * (The datapath implementation might use some different hash function for
664 * distributing packets received via flow misses among PIDs. This means
665 * that packets received via flow misses might be reordered relative to
666 * packets received via userspace actions. This is not ordinarily a
669 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
670 * allocated to any port, that the client may use for special purposes.
672 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
673 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
674 * disabled and then re-enabled, so a client that does that must be prepared to
675 * update all of the flows that it installed that contain
676 * OVS_ACTION_ATTR_USERSPACE actions. */
678 dpif_port_get_pid(const struct dpif
*dpif
, odp_port_t port_no
, uint32_t hash
)
680 return (dpif
->dpif_class
->port_get_pid
681 ? (dpif
->dpif_class
->port_get_pid
)(dpif
, port_no
, hash
)
685 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
686 * the port's name into the 'name_size' bytes in 'name', ensuring that the
687 * result is null-terminated. On failure, returns a positive errno value and
688 * makes 'name' the empty string. */
690 dpif_port_get_name(struct dpif
*dpif
, odp_port_t port_no
,
691 char *name
, size_t name_size
)
693 struct dpif_port port
;
696 ovs_assert(name_size
> 0);
698 error
= dpif_port_query_by_number(dpif
, port_no
, &port
);
700 ovs_strlcpy(name
, port
.name
, name_size
);
701 dpif_port_destroy(&port
);
708 /* Initializes 'dump' to begin dumping the ports in a dpif.
710 * This function provides no status indication. An error status for the entire
711 * dump operation is provided when it is completed by calling
712 * dpif_port_dump_done().
715 dpif_port_dump_start(struct dpif_port_dump
*dump
, const struct dpif
*dpif
)
718 dump
->error
= dpif
->dpif_class
->port_dump_start(dpif
, &dump
->state
);
719 log_operation(dpif
, "port_dump_start", dump
->error
);
722 /* Attempts to retrieve another port from 'dump', which must have been
723 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
724 * into 'port' and returns true. On failure, returns false.
726 * Failure might indicate an actual error or merely that the last port has been
727 * dumped. An error status for the entire dump operation is provided when it
728 * is completed by calling dpif_port_dump_done().
730 * The dpif owns the data stored in 'port'. It will remain valid until at
731 * least the next time 'dump' is passed to dpif_port_dump_next() or
732 * dpif_port_dump_done(). */
734 dpif_port_dump_next(struct dpif_port_dump
*dump
, struct dpif_port
*port
)
736 const struct dpif
*dpif
= dump
->dpif
;
742 dump
->error
= dpif
->dpif_class
->port_dump_next(dpif
, dump
->state
, port
);
743 if (dump
->error
== EOF
) {
744 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all ports", dpif_name(dpif
));
746 log_operation(dpif
, "port_dump_next", dump
->error
);
750 dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
756 /* Completes port table dump operation 'dump', which must have been initialized
757 * with dpif_port_dump_start(). Returns 0 if the dump operation was
758 * error-free, otherwise a positive errno value describing the problem. */
760 dpif_port_dump_done(struct dpif_port_dump
*dump
)
762 const struct dpif
*dpif
= dump
->dpif
;
764 dump
->error
= dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
765 log_operation(dpif
, "port_dump_done", dump
->error
);
767 return dump
->error
== EOF
? 0 : dump
->error
;
770 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
771 * 'dpif' has changed, this function does one of the following:
773 * - Stores the name of the device that was added to or deleted from 'dpif' in
774 * '*devnamep' and returns 0. The caller is responsible for freeing
775 * '*devnamep' (with free()) when it no longer needs it.
777 * - Returns ENOBUFS and sets '*devnamep' to NULL.
779 * This function may also return 'false positives', where it returns 0 and
780 * '*devnamep' names a device that was not actually added or deleted or it
781 * returns ENOBUFS without any change.
783 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
784 * return other positive errno values to indicate that something has gone
787 dpif_port_poll(const struct dpif
*dpif
, char **devnamep
)
789 int error
= dpif
->dpif_class
->port_poll(dpif
, devnamep
);
796 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
797 * value other than EAGAIN. */
799 dpif_port_poll_wait(const struct dpif
*dpif
)
801 dpif
->dpif_class
->port_poll_wait(dpif
);
804 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
805 * arguments must have been initialized through a call to flow_extract().
806 * 'used' is stored into stats->used. */
808 dpif_flow_stats_extract(const struct flow
*flow
, const struct ofpbuf
*packet
,
809 long long int used
, struct dpif_flow_stats
*stats
)
811 stats
->tcp_flags
= ntohs(flow
->tcp_flags
);
812 stats
->n_bytes
= ofpbuf_size(packet
);
813 stats
->n_packets
= 1;
817 /* Appends a human-readable representation of 'stats' to 's'. */
819 dpif_flow_stats_format(const struct dpif_flow_stats
*stats
, struct ds
*s
)
821 ds_put_format(s
, "packets:%"PRIu64
", bytes:%"PRIu64
", used:",
822 stats
->n_packets
, stats
->n_bytes
);
824 ds_put_format(s
, "%.3fs", (time_msec() - stats
->used
) / 1000.0);
826 ds_put_format(s
, "never");
828 if (stats
->tcp_flags
) {
829 ds_put_cstr(s
, ", flags:");
830 packet_format_tcp_flags(s
, stats
->tcp_flags
);
834 /* Places the hash of the 'key_len' bytes starting at 'key' into '*hash'. */
836 dpif_flow_hash(const struct dpif
*dpif OVS_UNUSED
,
837 const void *key
, size_t key_len
, ovs_u128
*hash
)
839 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
840 static uint32_t secret
;
842 if (ovsthread_once_start(&once
)) {
843 secret
= random_uint32();
844 ovsthread_once_done(&once
);
846 hash_bytes128(key
, key_len
, secret
, hash
);
849 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
850 * positive errno value. */
852 dpif_flow_flush(struct dpif
*dpif
)
856 COVERAGE_INC(dpif_flow_flush
);
858 error
= dpif
->dpif_class
->flow_flush(dpif
);
859 log_operation(dpif
, "flow_flush", error
);
863 /* Attempts to install 'key' into the datapath, fetches it, then deletes it.
864 * Returns true if the datapath supported installing 'flow', false otherwise.
867 dpif_probe_feature(struct dpif
*dpif
, const char *name
,
868 const struct ofpbuf
*key
, const ovs_u128
*ufid
)
870 struct dpif_flow flow
;
872 uint64_t stub
[DPIF_FLOW_BUFSIZE
/ 8];
873 bool enable_feature
= false;
876 /* Use DPIF_FP_MODIFY to cover the case where ovs-vswitchd is killed (and
877 * restarted) at just the right time such that feature probes from the
878 * previous run are still present in the datapath. */
879 error
= dpif_flow_put(dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
| DPIF_FP_PROBE
,
880 ofpbuf_data(key
), ofpbuf_size(key
), NULL
, 0, NULL
, 0,
883 if (error
!= EINVAL
) {
884 VLOG_WARN("%s: %s flow probe failed (%s)",
885 dpif_name(dpif
), name
, ovs_strerror(error
));
890 ofpbuf_use_stack(&reply
, &stub
, sizeof stub
);
891 error
= dpif_flow_get(dpif
, ofpbuf_data(key
), ofpbuf_size(key
), ufid
,
894 && (!ufid
|| (flow
.ufid_present
&& ovs_u128_equal(ufid
, &flow
.ufid
)))) {
895 enable_feature
= true;
898 error
= dpif_flow_del(dpif
, ofpbuf_data(key
), ofpbuf_size(key
), ufid
,
901 VLOG_WARN("%s: failed to delete %s feature probe flow",
902 dpif_name(dpif
), name
);
905 return enable_feature
;
908 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_GET. */
910 dpif_flow_get(struct dpif
*dpif
,
911 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
912 struct ofpbuf
*buf
, struct dpif_flow
*flow
)
917 op
.type
= DPIF_OP_FLOW_GET
;
918 op
.u
.flow_get
.key
= key
;
919 op
.u
.flow_get
.key_len
= key_len
;
920 op
.u
.flow_get
.ufid
= ufid
;
921 op
.u
.flow_get
.buffer
= buf
;
923 memset(flow
, 0, sizeof *flow
);
924 op
.u
.flow_get
.flow
= flow
;
925 op
.u
.flow_get
.flow
->key
= key
;
926 op
.u
.flow_get
.flow
->key_len
= key_len
;
929 dpif_operate(dpif
, &opp
, 1);
934 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_PUT. */
936 dpif_flow_put(struct dpif
*dpif
, enum dpif_flow_put_flags flags
,
937 const struct nlattr
*key
, size_t key_len
,
938 const struct nlattr
*mask
, size_t mask_len
,
939 const struct nlattr
*actions
, size_t actions_len
,
940 const ovs_u128
*ufid
, struct dpif_flow_stats
*stats
)
945 op
.type
= DPIF_OP_FLOW_PUT
;
946 op
.u
.flow_put
.flags
= flags
;
947 op
.u
.flow_put
.key
= key
;
948 op
.u
.flow_put
.key_len
= key_len
;
949 op
.u
.flow_put
.mask
= mask
;
950 op
.u
.flow_put
.mask_len
= mask_len
;
951 op
.u
.flow_put
.actions
= actions
;
952 op
.u
.flow_put
.actions_len
= actions_len
;
953 op
.u
.flow_put
.ufid
= ufid
;
954 op
.u
.flow_put
.stats
= stats
;
957 dpif_operate(dpif
, &opp
, 1);
962 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_DEL. */
964 dpif_flow_del(struct dpif
*dpif
,
965 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
966 struct dpif_flow_stats
*stats
)
971 op
.type
= DPIF_OP_FLOW_DEL
;
972 op
.u
.flow_del
.key
= key
;
973 op
.u
.flow_del
.key_len
= key_len
;
974 op
.u
.flow_del
.ufid
= ufid
;
975 op
.u
.flow_del
.stats
= stats
;
976 op
.u
.flow_del
.terse
= false;
979 dpif_operate(dpif
, &opp
, 1);
984 /* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
985 * flows in 'dpif'. If 'terse' is true, then only UFID and statistics will
986 * be returned in the dump. Otherwise, all fields will be returned.
988 * This function always successfully returns a dpif_flow_dump. Error
989 * reporting is deferred to dpif_flow_dump_destroy(). */
990 struct dpif_flow_dump
*
991 dpif_flow_dump_create(const struct dpif
*dpif
, bool terse
)
993 return dpif
->dpif_class
->flow_dump_create(dpif
, terse
);
996 /* Destroys 'dump', which must have been created with dpif_flow_dump_create().
997 * All dpif_flow_dump_thread structures previously created for 'dump' must
998 * previously have been destroyed.
1000 * Returns 0 if the dump operation was error-free, otherwise a positive errno
1001 * value describing the problem. */
1003 dpif_flow_dump_destroy(struct dpif_flow_dump
*dump
)
1005 const struct dpif
*dpif
= dump
->dpif
;
1006 int error
= dpif
->dpif_class
->flow_dump_destroy(dump
);
1007 log_operation(dpif
, "flow_dump_destroy", error
);
1008 return error
== EOF
? 0 : error
;
1011 /* Returns new thread-local state for use with dpif_flow_dump_next(). */
1012 struct dpif_flow_dump_thread
*
1013 dpif_flow_dump_thread_create(struct dpif_flow_dump
*dump
)
1015 return dump
->dpif
->dpif_class
->flow_dump_thread_create(dump
);
1018 /* Releases 'thread'. */
1020 dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread
*thread
)
1022 thread
->dpif
->dpif_class
->flow_dump_thread_destroy(thread
);
1025 /* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0
1026 * if and only if no flows remained to be retrieved, otherwise a positive
1027 * number reflecting the number of elements in 'flows[]' that were updated.
1028 * The number of flows returned might be less than 'max_flows' because
1029 * fewer than 'max_flows' remained, because this particular datapath does not
1030 * benefit from batching, or because an error occurred partway through
1031 * retrieval. Thus, the caller should continue calling until a 0 return value,
1032 * even if intermediate return values are less than 'max_flows'.
1034 * No error status is immediately provided. An error status for the entire
1035 * dump operation is provided when it is completed by calling
1036 * dpif_flow_dump_destroy().
1038 * All of the data stored into 'flows' is owned by the datapath, not by the
1039 * caller, and the caller must not modify or free it. The datapath guarantees
1040 * that it remains accessible and unchanged until the first of:
1041 * - The next call to dpif_flow_dump_next() for 'thread', or
1042 * - The next rcu quiescent period. */
1044 dpif_flow_dump_next(struct dpif_flow_dump_thread
*thread
,
1045 struct dpif_flow
*flows
, int max_flows
)
1047 struct dpif
*dpif
= thread
->dpif
;
1050 ovs_assert(max_flows
> 0);
1051 n
= dpif
->dpif_class
->flow_dump_next(thread
, flows
, max_flows
);
1053 struct dpif_flow
*f
;
1055 for (f
= flows
; f
< &flows
[n
] && should_log_flow_message(0); f
++) {
1056 log_flow_message(dpif
, 0, "flow_dump",
1057 f
->key
, f
->key_len
, f
->mask
, f
->mask_len
,
1058 &f
->ufid
, &f
->stats
, f
->actions
, f
->actions_len
);
1061 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all flows", dpif_name(dpif
));
1066 struct dpif_execute_helper_aux
{
1071 /* This is called for actions that need the context of the datapath to be
1074 dpif_execute_helper_cb(void *aux_
, struct dpif_packet
**packets
, int cnt
,
1075 const struct nlattr
*action
, bool may_steal OVS_UNUSED
)
1077 struct dpif_execute_helper_aux
*aux
= aux_
;
1078 int type
= nl_attr_type(action
);
1079 struct ofpbuf
*packet
= &packets
[0]->ofpbuf
;
1080 struct pkt_metadata
*md
= &packets
[0]->md
;
1082 ovs_assert(cnt
== 1);
1084 switch ((enum ovs_action_attr
)type
) {
1085 case OVS_ACTION_ATTR_OUTPUT
:
1086 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
1087 case OVS_ACTION_ATTR_TUNNEL_POP
:
1088 case OVS_ACTION_ATTR_USERSPACE
:
1089 case OVS_ACTION_ATTR_RECIRC
: {
1090 struct dpif_execute execute
;
1091 struct ofpbuf execute_actions
;
1092 uint64_t stub
[256 / 8];
1094 if (md
->tunnel
.ip_dst
) {
1095 /* The Linux kernel datapath throws away the tunnel information
1096 * that we supply as metadata. We have to use a "set" action to
1098 ofpbuf_use_stub(&execute_actions
, stub
, sizeof stub
);
1099 odp_put_tunnel_action(&md
->tunnel
, &execute_actions
);
1100 ofpbuf_put(&execute_actions
, action
, NLA_ALIGN(action
->nla_len
));
1102 execute
.actions
= ofpbuf_data(&execute_actions
);
1103 execute
.actions_len
= ofpbuf_size(&execute_actions
);
1105 execute
.actions
= action
;
1106 execute
.actions_len
= NLA_ALIGN(action
->nla_len
);
1109 execute
.packet
= packet
;
1111 execute
.needs_help
= false;
1112 execute
.probe
= false;
1113 aux
->error
= dpif_execute(aux
->dpif
, &execute
);
1114 log_execute_message(aux
->dpif
, &execute
, true, aux
->error
);
1116 if (md
->tunnel
.ip_dst
) {
1117 ofpbuf_uninit(&execute_actions
);
1122 case OVS_ACTION_ATTR_HASH
:
1123 case OVS_ACTION_ATTR_PUSH_VLAN
:
1124 case OVS_ACTION_ATTR_POP_VLAN
:
1125 case OVS_ACTION_ATTR_PUSH_MPLS
:
1126 case OVS_ACTION_ATTR_POP_MPLS
:
1127 case OVS_ACTION_ATTR_SET
:
1128 case OVS_ACTION_ATTR_SET_MASKED
:
1129 case OVS_ACTION_ATTR_SAMPLE
:
1130 case OVS_ACTION_ATTR_UNSPEC
:
1131 case __OVS_ACTION_ATTR_MAX
:
1136 /* Executes 'execute' by performing most of the actions in userspace and
1137 * passing the fully constructed packets to 'dpif' for output and userspace
1140 * This helps with actions that a given 'dpif' doesn't implement directly. */
1142 dpif_execute_with_help(struct dpif
*dpif
, struct dpif_execute
*execute
)
1144 struct dpif_execute_helper_aux aux
= {dpif
, 0};
1145 struct dpif_packet packet
, *pp
;
1147 COVERAGE_INC(dpif_execute_with_help
);
1149 packet
.ofpbuf
= *execute
->packet
;
1150 packet
.md
= execute
->md
;
1153 odp_execute_actions(&aux
, &pp
, 1, false, execute
->actions
,
1154 execute
->actions_len
, dpif_execute_helper_cb
);
1156 /* Even though may_steal is set to false, some actions could modify or
1157 * reallocate the ofpbuf memory. We need to pass those changes to the
1159 *execute
->packet
= packet
.ofpbuf
;
1160 execute
->md
= packet
.md
;
1165 /* Returns true if the datapath needs help executing 'execute'. */
1167 dpif_execute_needs_help(const struct dpif_execute
*execute
)
1169 return execute
->needs_help
|| nl_attr_oversized(execute
->actions_len
);
1172 /* A dpif_operate() wrapper for performing a single DPIF_OP_EXECUTE. */
1174 dpif_execute(struct dpif
*dpif
, struct dpif_execute
*execute
)
1176 if (execute
->actions_len
) {
1177 struct dpif_op
*opp
;
1180 op
.type
= DPIF_OP_EXECUTE
;
1181 op
.u
.execute
= *execute
;
1184 dpif_operate(dpif
, &opp
, 1);
1192 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1193 * which they are specified. Places each operation's results in the "output"
1194 * members documented in comments, and 0 in the 'error' member on success or a
1195 * positive errno on failure. */
1197 dpif_operate(struct dpif
*dpif
, struct dpif_op
**ops
, size_t n_ops
)
1202 /* Count 'chunk', the number of ops that can be executed without
1203 * needing any help. Ops that need help should be rare, so we
1204 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1205 for (chunk
= 0; chunk
< n_ops
; chunk
++) {
1206 struct dpif_op
*op
= ops
[chunk
];
1208 if (op
->type
== DPIF_OP_EXECUTE
1209 && dpif_execute_needs_help(&op
->u
.execute
)) {
1215 /* Execute a chunk full of ops that the dpif provider can
1216 * handle itself, without help. */
1219 dpif
->dpif_class
->operate(dpif
, ops
, chunk
);
1221 for (i
= 0; i
< chunk
; i
++) {
1222 struct dpif_op
*op
= ops
[i
];
1223 int error
= op
->error
;
1226 case DPIF_OP_FLOW_PUT
: {
1227 struct dpif_flow_put
*put
= &op
->u
.flow_put
;
1229 COVERAGE_INC(dpif_flow_put
);
1230 log_flow_put_message(dpif
, put
, error
);
1231 if (error
&& put
->stats
) {
1232 memset(put
->stats
, 0, sizeof *put
->stats
);
1237 case DPIF_OP_FLOW_GET
: {
1238 struct dpif_flow_get
*get
= &op
->u
.flow_get
;
1240 COVERAGE_INC(dpif_flow_get
);
1242 memset(get
->flow
, 0, sizeof *get
->flow
);
1244 log_flow_get_message(dpif
, get
, error
);
1249 case DPIF_OP_FLOW_DEL
: {
1250 struct dpif_flow_del
*del
= &op
->u
.flow_del
;
1252 COVERAGE_INC(dpif_flow_del
);
1253 log_flow_del_message(dpif
, del
, error
);
1254 if (error
&& del
->stats
) {
1255 memset(del
->stats
, 0, sizeof *del
->stats
);
1260 case DPIF_OP_EXECUTE
:
1261 COVERAGE_INC(dpif_execute
);
1262 log_execute_message(dpif
, &op
->u
.execute
, false, error
);
1270 /* Help the dpif provider to execute one op. */
1271 struct dpif_op
*op
= ops
[0];
1273 COVERAGE_INC(dpif_execute
);
1274 op
->error
= dpif_execute_with_help(dpif
, &op
->u
.execute
);
1281 /* Returns a string that represents 'type', for use in log messages. */
1283 dpif_upcall_type_to_string(enum dpif_upcall_type type
)
1286 case DPIF_UC_MISS
: return "miss";
1287 case DPIF_UC_ACTION
: return "action";
1288 case DPIF_N_UC_TYPES
: default: return "<unknown>";
1292 /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1293 * if successful, otherwise a positive errno value.
1295 * Turning packet receive off and then back on may change the Netlink PID
1296 * assignments returned by dpif_port_get_pid(). If the client does this, it
1297 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1298 * using the new PID assignment. */
1300 dpif_recv_set(struct dpif
*dpif
, bool enable
)
1304 if (dpif
->dpif_class
->recv_set
) {
1305 error
= dpif
->dpif_class
->recv_set(dpif
, enable
);
1306 log_operation(dpif
, "recv_set", error
);
1311 /* Refreshes the poll loops and Netlink sockets associated to each port,
1312 * when the number of upcall handlers (upcall receiving thread) is changed
1313 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
1316 * Since multiple upcall handlers can read upcalls simultaneously from
1317 * 'dpif', each port can have multiple Netlink sockets, one per upcall
1318 * handler. So, handlers_set() is responsible for the following tasks:
1320 * When receiving upcall is enabled, extends or creates the
1321 * configuration to support:
1323 * - 'n_handlers' Netlink sockets for each port.
1325 * - 'n_handlers' poll loops, one for each upcall handler.
1327 * - registering the Netlink sockets for the same upcall handler to
1328 * the corresponding poll loop.
1330 * Returns 0 if successful, otherwise a positive errno value. */
1332 dpif_handlers_set(struct dpif
*dpif
, uint32_t n_handlers
)
1336 if (dpif
->dpif_class
->handlers_set
) {
1337 error
= dpif
->dpif_class
->handlers_set(dpif
, n_handlers
);
1338 log_operation(dpif
, "handlers_set", error
);
1344 dpif_register_upcall_cb(struct dpif
*dpif
, upcall_callback
*cb
, void *aux
)
1346 if (dpif
->dpif_class
->register_upcall_cb
) {
1347 dpif
->dpif_class
->register_upcall_cb(dpif
, cb
, aux
);
1352 dpif_enable_upcall(struct dpif
*dpif
)
1354 if (dpif
->dpif_class
->enable_upcall
) {
1355 dpif
->dpif_class
->enable_upcall(dpif
);
1360 dpif_disable_upcall(struct dpif
*dpif
)
1362 if (dpif
->dpif_class
->disable_upcall
) {
1363 dpif
->dpif_class
->disable_upcall(dpif
);
1368 dpif_print_packet(struct dpif
*dpif
, struct dpif_upcall
*upcall
)
1370 if (!VLOG_DROP_DBG(&dpmsg_rl
)) {
1374 packet
= ofp_packet_to_string(ofpbuf_data(&upcall
->packet
),
1375 ofpbuf_size(&upcall
->packet
));
1378 odp_flow_key_format(upcall
->key
, upcall
->key_len
, &flow
);
1380 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1381 dpif_name(dpif
), dpif_upcall_type_to_string(upcall
->type
),
1382 ds_cstr(&flow
), packet
);
1389 /* If 'dpif' creates its own I/O polling threads, refreshes poll threads
1392 dpif_poll_threads_set(struct dpif
*dpif
, unsigned int n_rxqs
,
1397 if (dpif
->dpif_class
->poll_threads_set
) {
1398 error
= dpif
->dpif_class
->poll_threads_set(dpif
, n_rxqs
, cmask
);
1400 log_operation(dpif
, "poll_threads_set", error
);
1407 /* Polls for an upcall from 'dpif' for an upcall handler. Since there
1408 * there can be multiple poll loops, 'handler_id' is needed as index to
1409 * identify the corresponding poll loop. If successful, stores the upcall
1410 * into '*upcall', using 'buf' for storage. Should only be called if
1411 * 'recv_set' has been used to enable receiving packets from 'dpif'.
1413 * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
1414 * 'buf', so their memory cannot be freed separately from 'buf'.
1416 * The caller owns the data of 'upcall->packet' and may modify it. If
1417 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
1418 * will be reallocated. This requires the data of 'upcall->packet' to be
1419 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
1420 * when an error is returned, the 'upcall->packet' may be uninitialized
1421 * and should not be released.
1423 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1424 * if no upcall is immediately available. */
1426 dpif_recv(struct dpif
*dpif
, uint32_t handler_id
, struct dpif_upcall
*upcall
,
1431 if (dpif
->dpif_class
->recv
) {
1432 error
= dpif
->dpif_class
->recv(dpif
, handler_id
, upcall
, buf
);
1434 dpif_print_packet(dpif
, upcall
);
1435 } else if (error
!= EAGAIN
) {
1436 log_operation(dpif
, "recv", error
);
1442 /* Discards all messages that would otherwise be received by dpif_recv() on
1445 dpif_recv_purge(struct dpif
*dpif
)
1447 COVERAGE_INC(dpif_purge
);
1448 if (dpif
->dpif_class
->recv_purge
) {
1449 dpif
->dpif_class
->recv_purge(dpif
);
1453 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
1454 * 'dpif' has a message queued to be received with the recv member
1455 * function. Since there can be multiple poll loops, 'handler_id' is
1456 * needed as index to identify the corresponding poll loop. */
1458 dpif_recv_wait(struct dpif
*dpif
, uint32_t handler_id
)
1460 if (dpif
->dpif_class
->recv_wait
) {
1461 dpif
->dpif_class
->recv_wait(dpif
, handler_id
);
1466 * Return the datapath version. Caller is responsible for freeing
1470 dpif_get_dp_version(const struct dpif
*dpif
)
1472 char *version
= NULL
;
1474 if (dpif
->dpif_class
->get_datapath_version
) {
1475 version
= dpif
->dpif_class
->get_datapath_version();
1481 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1482 * and '*engine_id', respectively. */
1484 dpif_get_netflow_ids(const struct dpif
*dpif
,
1485 uint8_t *engine_type
, uint8_t *engine_id
)
1487 *engine_type
= dpif
->netflow_engine_type
;
1488 *engine_id
= dpif
->netflow_engine_id
;
1491 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1492 * value used for setting packet priority.
1493 * On success, returns 0 and stores the priority into '*priority'.
1494 * On failure, returns a positive errno value and stores 0 into '*priority'. */
1496 dpif_queue_to_priority(const struct dpif
*dpif
, uint32_t queue_id
,
1499 int error
= (dpif
->dpif_class
->queue_to_priority
1500 ? dpif
->dpif_class
->queue_to_priority(dpif
, queue_id
,
1506 log_operation(dpif
, "queue_to_priority", error
);
1511 dpif_init(struct dpif
*dpif
, const struct dpif_class
*dpif_class
,
1513 uint8_t netflow_engine_type
, uint8_t netflow_engine_id
)
1515 dpif
->dpif_class
= dpif_class
;
1516 dpif
->base_name
= xstrdup(name
);
1517 dpif
->full_name
= xasprintf("%s@%s", dpif_class
->type
, name
);
1518 dpif
->netflow_engine_type
= netflow_engine_type
;
1519 dpif
->netflow_engine_id
= netflow_engine_id
;
1522 /* Undoes the results of initialization.
1524 * Normally this function only needs to be called from dpif_close().
1525 * However, it may be called by providers due to an error on opening
1526 * that occurs after initialization. It this case dpif_close() would
1527 * never be called. */
1529 dpif_uninit(struct dpif
*dpif
, bool close
)
1531 char *base_name
= dpif
->base_name
;
1532 char *full_name
= dpif
->full_name
;
1535 dpif
->dpif_class
->close(dpif
);
1543 log_operation(const struct dpif
*dpif
, const char *operation
, int error
)
1546 VLOG_DBG_RL(&dpmsg_rl
, "%s: %s success", dpif_name(dpif
), operation
);
1547 } else if (ofperr_is_valid(error
)) {
1548 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1549 dpif_name(dpif
), operation
, ofperr_get_name(error
));
1551 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1552 dpif_name(dpif
), operation
, ovs_strerror(error
));
1556 static enum vlog_level
1557 flow_message_log_level(int error
)
1559 /* If flows arrive in a batch, userspace may push down multiple
1560 * unique flow definitions that overlap when wildcards are applied.
1561 * Kernels that support flow wildcarding will reject these flows as
1562 * duplicates (EEXIST), so lower the log level to debug for these
1563 * types of messages. */
1564 return (error
&& error
!= EEXIST
) ? VLL_WARN
: VLL_DBG
;
1568 should_log_flow_message(int error
)
1570 return !vlog_should_drop(THIS_MODULE
, flow_message_log_level(error
),
1571 error
? &error_rl
: &dpmsg_rl
);
1575 log_flow_message(const struct dpif
*dpif
, int error
, const char *operation
,
1576 const struct nlattr
*key
, size_t key_len
,
1577 const struct nlattr
*mask
, size_t mask_len
,
1578 const ovs_u128
*ufid
, const struct dpif_flow_stats
*stats
,
1579 const struct nlattr
*actions
, size_t actions_len
)
1581 struct ds ds
= DS_EMPTY_INITIALIZER
;
1582 ds_put_format(&ds
, "%s: ", dpif_name(dpif
));
1584 ds_put_cstr(&ds
, "failed to ");
1586 ds_put_format(&ds
, "%s ", operation
);
1588 ds_put_format(&ds
, "(%s) ", ovs_strerror(error
));
1591 odp_format_ufid(ufid
, &ds
);
1592 ds_put_cstr(&ds
, " ");
1594 odp_flow_format(key
, key_len
, mask
, mask_len
, NULL
, &ds
, true);
1596 ds_put_cstr(&ds
, ", ");
1597 dpif_flow_stats_format(stats
, &ds
);
1599 if (actions
|| actions_len
) {
1600 ds_put_cstr(&ds
, ", actions:");
1601 format_odp_actions(&ds
, actions
, actions_len
);
1603 vlog(THIS_MODULE
, flow_message_log_level(error
), "%s", ds_cstr(&ds
));
1608 log_flow_put_message(struct dpif
*dpif
, const struct dpif_flow_put
*put
,
1611 if (should_log_flow_message(error
) && !(put
->flags
& DPIF_FP_PROBE
)) {
1615 ds_put_cstr(&s
, "put");
1616 if (put
->flags
& DPIF_FP_CREATE
) {
1617 ds_put_cstr(&s
, "[create]");
1619 if (put
->flags
& DPIF_FP_MODIFY
) {
1620 ds_put_cstr(&s
, "[modify]");
1622 if (put
->flags
& DPIF_FP_ZERO_STATS
) {
1623 ds_put_cstr(&s
, "[zero]");
1625 log_flow_message(dpif
, error
, ds_cstr(&s
),
1626 put
->key
, put
->key_len
, put
->mask
, put
->mask_len
,
1627 put
->ufid
, put
->stats
, put
->actions
,
1634 log_flow_del_message(struct dpif
*dpif
, const struct dpif_flow_del
*del
,
1637 if (should_log_flow_message(error
)) {
1638 log_flow_message(dpif
, error
, "flow_del", del
->key
, del
->key_len
,
1639 NULL
, 0, del
->ufid
, !error
? del
->stats
: NULL
,
1644 /* Logs that 'execute' was executed on 'dpif' and completed with errno 'error'
1645 * (0 for success). 'subexecute' should be true if the execution is a result
1646 * of breaking down a larger execution that needed help, false otherwise.
1649 * XXX In theory, the log message could be deceptive because this function is
1650 * called after the dpif_provider's '->execute' function, which is allowed to
1651 * modify execute->packet and execute->md. In practice, though:
1653 * - dpif-netlink doesn't modify execute->packet or execute->md.
1655 * - dpif-netdev does modify them but it is less likely to have problems
1656 * because it is built into ovs-vswitchd and cannot have version skew,
1659 * It would still be better to avoid the potential problem. I don't know of a
1660 * good way to do that, though, that isn't expensive. */
1662 log_execute_message(struct dpif
*dpif
, const struct dpif_execute
*execute
,
1663 bool subexecute
, int error
)
1665 if (!(error
? VLOG_DROP_WARN(&error_rl
) : VLOG_DROP_DBG(&dpmsg_rl
))
1666 && !execute
->probe
) {
1667 struct ds ds
= DS_EMPTY_INITIALIZER
;
1670 packet
= ofp_packet_to_string(ofpbuf_data(execute
->packet
),
1671 ofpbuf_size(execute
->packet
));
1672 ds_put_format(&ds
, "%s: %sexecute ",
1674 (subexecute
? "sub-"
1675 : dpif_execute_needs_help(execute
) ? "super-"
1677 format_odp_actions(&ds
, execute
->actions
, execute
->actions_len
);
1679 ds_put_format(&ds
, " failed (%s)", ovs_strerror(error
));
1681 ds_put_format(&ds
, " on packet %s", packet
);
1682 vlog(THIS_MODULE
, error
? VLL_WARN
: VLL_DBG
, "%s", ds_cstr(&ds
));
1689 log_flow_get_message(const struct dpif
*dpif
, const struct dpif_flow_get
*get
,
1692 if (should_log_flow_message(error
)) {
1693 log_flow_message(dpif
, error
, "flow_get",
1694 get
->key
, get
->key_len
,
1695 get
->flow
->mask
, get
->flow
->mask_len
,
1696 get
->ufid
, &get
->flow
->stats
,
1697 get
->flow
->actions
, get
->flow
->actions_len
);
1702 dpif_supports_tnl_push_pop(const struct dpif
*dpif
)
1704 return !strcmp(dpif
->dpif_class
->type
, "netdev") ||
1705 !strcmp(dpif
->dpif_class
->type
, "dummy");