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 "dp-packet.h"
29 #include "dynamic-string.h"
33 #include "odp-execute.h"
35 #include "ofp-errors.h"
36 #include "ofp-print.h"
40 #include "poll-loop.h"
41 #include "route-table.h"
46 #include "tnl-arp-cache.h"
47 #include "tnl-ports.h"
51 #include "openvswitch/vlog.h"
53 VLOG_DEFINE_THIS_MODULE(dpif
);
55 COVERAGE_DEFINE(dpif_destroy
);
56 COVERAGE_DEFINE(dpif_port_add
);
57 COVERAGE_DEFINE(dpif_port_del
);
58 COVERAGE_DEFINE(dpif_flow_flush
);
59 COVERAGE_DEFINE(dpif_flow_get
);
60 COVERAGE_DEFINE(dpif_flow_put
);
61 COVERAGE_DEFINE(dpif_flow_del
);
62 COVERAGE_DEFINE(dpif_execute
);
63 COVERAGE_DEFINE(dpif_purge
);
64 COVERAGE_DEFINE(dpif_execute_with_help
);
66 static const struct dpif_class
*base_dpif_classes
[] = {
67 #if defined(__linux__) || defined(_WIN32)
73 struct registered_dpif_class
{
74 const struct dpif_class
*dpif_class
;
77 static struct shash dpif_classes
= SHASH_INITIALIZER(&dpif_classes
);
78 static struct sset dpif_blacklist
= SSET_INITIALIZER(&dpif_blacklist
);
80 /* Protects 'dpif_classes', including the refcount, and 'dpif_blacklist'. */
81 static struct ovs_mutex dpif_mutex
= OVS_MUTEX_INITIALIZER
;
83 /* Rate limit for individual messages going to or from the datapath, output at
84 * DBG level. This is very high because, if these are enabled, it is because
85 * we really need to see them. */
86 static struct vlog_rate_limit dpmsg_rl
= VLOG_RATE_LIMIT_INIT(600, 600);
88 /* Not really much point in logging many dpif errors. */
89 static struct vlog_rate_limit error_rl
= VLOG_RATE_LIMIT_INIT(60, 5);
91 static void log_flow_message(const struct dpif
*dpif
, int error
,
92 const char *operation
,
93 const struct nlattr
*key
, size_t key_len
,
94 const struct nlattr
*mask
, size_t mask_len
,
96 const struct dpif_flow_stats
*stats
,
97 const struct nlattr
*actions
, size_t actions_len
);
98 static void log_operation(const struct dpif
*, const char *operation
,
100 static bool should_log_flow_message(int error
);
101 static void log_flow_put_message(struct dpif
*, const struct dpif_flow_put
*,
103 static void log_flow_del_message(struct dpif
*, const struct dpif_flow_del
*,
105 static void log_execute_message(struct dpif
*, const struct dpif_execute
*,
106 bool subexecute
, int error
);
107 static void log_flow_get_message(const struct dpif
*,
108 const struct dpif_flow_get
*, int error
);
110 /* Incremented whenever tnl route, arp, etc changes. */
111 struct seq
*tnl_conf_seq
;
116 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
118 if (ovsthread_once_start(&once
)) {
121 tnl_conf_seq
= seq_create();
122 dpctl_unixctl_register();
124 tnl_arp_cache_init();
127 for (i
= 0; i
< ARRAY_SIZE(base_dpif_classes
); i
++) {
128 dp_register_provider(base_dpif_classes
[i
]);
131 ovsthread_once_done(&once
);
136 dp_register_provider__(const struct dpif_class
*new_class
)
138 struct registered_dpif_class
*registered_class
;
141 if (sset_contains(&dpif_blacklist
, new_class
->type
)) {
142 VLOG_DBG("attempted to register blacklisted provider: %s",
147 if (shash_find(&dpif_classes
, new_class
->type
)) {
148 VLOG_WARN("attempted to register duplicate datapath provider: %s",
153 error
= new_class
->init
? new_class
->init() : 0;
155 VLOG_WARN("failed to initialize %s datapath class: %s",
156 new_class
->type
, ovs_strerror(error
));
160 registered_class
= xmalloc(sizeof *registered_class
);
161 registered_class
->dpif_class
= new_class
;
162 registered_class
->refcount
= 0;
164 shash_add(&dpif_classes
, new_class
->type
, registered_class
);
169 /* Registers a new datapath provider. After successful registration, new
170 * datapaths of that type can be opened using dpif_open(). */
172 dp_register_provider(const struct dpif_class
*new_class
)
176 ovs_mutex_lock(&dpif_mutex
);
177 error
= dp_register_provider__(new_class
);
178 ovs_mutex_unlock(&dpif_mutex
);
183 /* Unregisters a datapath provider. 'type' must have been previously
184 * registered and not currently be in use by any dpifs. After unregistration
185 * new datapaths of that type cannot be opened using dpif_open(). */
187 dp_unregister_provider__(const char *type
)
189 struct shash_node
*node
;
190 struct registered_dpif_class
*registered_class
;
192 node
= shash_find(&dpif_classes
, type
);
194 VLOG_WARN("attempted to unregister a datapath provider that is not "
195 "registered: %s", type
);
199 registered_class
= node
->data
;
200 if (registered_class
->refcount
) {
201 VLOG_WARN("attempted to unregister in use datapath provider: %s", type
);
205 shash_delete(&dpif_classes
, node
);
206 free(registered_class
);
211 /* Unregisters a datapath provider. 'type' must have been previously
212 * registered and not currently be in use by any dpifs. After unregistration
213 * new datapaths of that type cannot be opened using dpif_open(). */
215 dp_unregister_provider(const char *type
)
221 ovs_mutex_lock(&dpif_mutex
);
222 error
= dp_unregister_provider__(type
);
223 ovs_mutex_unlock(&dpif_mutex
);
228 /* Blacklists a provider. Causes future calls of dp_register_provider() with
229 * a dpif_class which implements 'type' to fail. */
231 dp_blacklist_provider(const char *type
)
233 ovs_mutex_lock(&dpif_mutex
);
234 sset_add(&dpif_blacklist
, type
);
235 ovs_mutex_unlock(&dpif_mutex
);
238 /* Adds the types of all currently registered datapath providers to 'types'.
239 * The caller must first initialize the sset. */
241 dp_enumerate_types(struct sset
*types
)
243 struct shash_node
*node
;
247 ovs_mutex_lock(&dpif_mutex
);
248 SHASH_FOR_EACH(node
, &dpif_classes
) {
249 const struct registered_dpif_class
*registered_class
= node
->data
;
250 sset_add(types
, registered_class
->dpif_class
->type
);
252 ovs_mutex_unlock(&dpif_mutex
);
256 dp_class_unref(struct registered_dpif_class
*rc
)
258 ovs_mutex_lock(&dpif_mutex
);
259 ovs_assert(rc
->refcount
);
261 ovs_mutex_unlock(&dpif_mutex
);
264 static struct registered_dpif_class
*
265 dp_class_lookup(const char *type
)
267 struct registered_dpif_class
*rc
;
269 ovs_mutex_lock(&dpif_mutex
);
270 rc
= shash_find_data(&dpif_classes
, type
);
274 ovs_mutex_unlock(&dpif_mutex
);
279 /* Clears 'names' and enumerates the names of all known created datapaths with
280 * the given 'type'. The caller must first initialize the sset. Returns 0 if
281 * successful, otherwise a positive errno value.
283 * Some kinds of datapaths might not be practically enumerable. This is not
284 * considered an error. */
286 dp_enumerate_names(const char *type
, struct sset
*names
)
288 struct registered_dpif_class
*registered_class
;
289 const struct dpif_class
*dpif_class
;
295 registered_class
= dp_class_lookup(type
);
296 if (!registered_class
) {
297 VLOG_WARN("could not enumerate unknown type: %s", type
);
301 dpif_class
= registered_class
->dpif_class
;
302 error
= (dpif_class
->enumerate
303 ? dpif_class
->enumerate(names
, dpif_class
)
306 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class
->type
,
307 ovs_strerror(error
));
309 dp_class_unref(registered_class
);
314 /* Parses 'datapath_name_', which is of the form [type@]name into its
315 * component pieces. 'name' and 'type' must be freed by the caller.
317 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
319 dp_parse_name(const char *datapath_name_
, char **name
, char **type
)
321 char *datapath_name
= xstrdup(datapath_name_
);
324 separator
= strchr(datapath_name
, '@');
327 *type
= datapath_name
;
328 *name
= xstrdup(dpif_normalize_type(separator
+ 1));
330 *name
= datapath_name
;
331 *type
= xstrdup(dpif_normalize_type(NULL
));
336 do_open(const char *name
, const char *type
, bool create
, struct dpif
**dpifp
)
338 struct dpif
*dpif
= NULL
;
340 struct registered_dpif_class
*registered_class
;
344 type
= dpif_normalize_type(type
);
345 registered_class
= dp_class_lookup(type
);
346 if (!registered_class
) {
347 VLOG_WARN("could not create datapath %s of unknown type %s", name
,
349 error
= EAFNOSUPPORT
;
353 error
= registered_class
->dpif_class
->open(registered_class
->dpif_class
,
354 name
, create
, &dpif
);
356 ovs_assert(dpif
->dpif_class
== registered_class
->dpif_class
);
358 dp_class_unref(registered_class
);
362 *dpifp
= error
? NULL
: dpif
;
366 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
367 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
368 * the empty string to specify the default system type. Returns 0 if
369 * successful, otherwise a positive errno value. On success stores a pointer
370 * to the datapath in '*dpifp', otherwise a null pointer. */
372 dpif_open(const char *name
, const char *type
, struct dpif
**dpifp
)
374 return do_open(name
, type
, false, dpifp
);
377 /* Tries to create and open a new datapath with the given 'name' and 'type'.
378 * 'type' may be either NULL or the empty string to specify the default system
379 * type. Will fail if a datapath with 'name' and 'type' already exists.
380 * Returns 0 if successful, otherwise a positive errno value. On success
381 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
383 dpif_create(const char *name
, const char *type
, struct dpif
**dpifp
)
385 return do_open(name
, type
, true, dpifp
);
388 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
389 * does not exist. 'type' may be either NULL or the empty string to specify
390 * the default system type. Returns 0 if successful, otherwise a positive
391 * errno value. On success stores a pointer to the datapath in '*dpifp',
392 * otherwise a null pointer. */
394 dpif_create_and_open(const char *name
, const char *type
, struct dpif
**dpifp
)
398 error
= dpif_create(name
, type
, dpifp
);
399 if (error
== EEXIST
|| error
== EBUSY
) {
400 error
= dpif_open(name
, type
, dpifp
);
402 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
403 name
, ovs_strerror(error
));
406 VLOG_WARN("failed to create datapath %s: %s",
407 name
, ovs_strerror(error
));
412 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
413 * itself; call dpif_delete() first, instead, if that is desirable. */
415 dpif_close(struct dpif
*dpif
)
418 struct registered_dpif_class
*rc
;
420 rc
= shash_find_data(&dpif_classes
, dpif
->dpif_class
->type
);
421 dpif_uninit(dpif
, true);
426 /* Performs periodic work needed by 'dpif'. */
428 dpif_run(struct dpif
*dpif
)
430 if (dpif
->dpif_class
->run
) {
431 return dpif
->dpif_class
->run(dpif
);
436 /* Arranges for poll_block() to wake up when dp_run() needs to be called for
439 dpif_wait(struct dpif
*dpif
)
441 if (dpif
->dpif_class
->wait
) {
442 dpif
->dpif_class
->wait(dpif
);
446 /* Returns the name of datapath 'dpif' prefixed with the type
447 * (for use in log messages). */
449 dpif_name(const struct dpif
*dpif
)
451 return dpif
->full_name
;
454 /* Returns the name of datapath 'dpif' without the type
455 * (for use in device names). */
457 dpif_base_name(const struct dpif
*dpif
)
459 return dpif
->base_name
;
462 /* Returns the type of datapath 'dpif'. */
464 dpif_type(const struct dpif
*dpif
)
466 return dpif
->dpif_class
->type
;
469 /* Returns the fully spelled out name for the given datapath 'type'.
471 * Normalized type string can be compared with strcmp(). Unnormalized type
472 * string might be the same even if they have different spellings. */
474 dpif_normalize_type(const char *type
)
476 return type
&& type
[0] ? type
: "system";
479 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
480 * ports. After calling this function, it does not make sense to pass 'dpif'
481 * to any functions other than dpif_name() or dpif_close(). */
483 dpif_delete(struct dpif
*dpif
)
487 COVERAGE_INC(dpif_destroy
);
489 error
= dpif
->dpif_class
->destroy(dpif
);
490 log_operation(dpif
, "delete", error
);
494 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
495 * otherwise a positive errno value. */
497 dpif_get_dp_stats(const struct dpif
*dpif
, struct dpif_dp_stats
*stats
)
499 int error
= dpif
->dpif_class
->get_stats(dpif
, stats
);
501 memset(stats
, 0, sizeof *stats
);
503 log_operation(dpif
, "get_stats", error
);
508 dpif_port_open_type(const char *datapath_type
, const char *port_type
)
510 struct registered_dpif_class
*rc
;
512 datapath_type
= dpif_normalize_type(datapath_type
);
514 ovs_mutex_lock(&dpif_mutex
);
515 rc
= shash_find_data(&dpif_classes
, datapath_type
);
516 if (rc
&& rc
->dpif_class
->port_open_type
) {
517 port_type
= rc
->dpif_class
->port_open_type(rc
->dpif_class
, port_type
);
519 ovs_mutex_unlock(&dpif_mutex
);
524 /* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
525 * non-null and its value is not ODPP_NONE, then attempts to use the
526 * value as the port number.
528 * If successful, returns 0 and sets '*port_nop' to the new port's port
529 * number (if 'port_nop' is non-null). On failure, returns a positive
530 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
533 dpif_port_add(struct dpif
*dpif
, struct netdev
*netdev
, odp_port_t
*port_nop
)
535 const char *netdev_name
= netdev_get_name(netdev
);
536 odp_port_t port_no
= ODPP_NONE
;
539 COVERAGE_INC(dpif_port_add
);
545 error
= dpif
->dpif_class
->port_add(dpif
, netdev
, &port_no
);
547 VLOG_DBG_RL(&dpmsg_rl
, "%s: added %s as port %"PRIu32
,
548 dpif_name(dpif
), netdev_name
, port_no
);
550 VLOG_WARN_RL(&error_rl
, "%s: failed to add %s as port: %s",
551 dpif_name(dpif
), netdev_name
, ovs_strerror(error
));
560 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
561 * otherwise a positive errno value. */
563 dpif_port_del(struct dpif
*dpif
, odp_port_t port_no
)
567 COVERAGE_INC(dpif_port_del
);
569 error
= dpif
->dpif_class
->port_del(dpif
, port_no
);
571 VLOG_DBG_RL(&dpmsg_rl
, "%s: port_del(%"PRIu32
")",
572 dpif_name(dpif
), port_no
);
574 log_operation(dpif
, "port_del", error
);
579 /* Makes a deep copy of 'src' into 'dst'. */
581 dpif_port_clone(struct dpif_port
*dst
, const struct dpif_port
*src
)
583 dst
->name
= xstrdup(src
->name
);
584 dst
->type
= xstrdup(src
->type
);
585 dst
->port_no
= src
->port_no
;
588 /* Frees memory allocated to members of 'dpif_port'.
590 * Do not call this function on a dpif_port obtained from
591 * dpif_port_dump_next(): that function retains ownership of the data in the
594 dpif_port_destroy(struct dpif_port
*dpif_port
)
596 free(dpif_port
->name
);
597 free(dpif_port
->type
);
600 /* Checks if port named 'devname' exists in 'dpif'. If so, returns
601 * true; otherwise, returns false. */
603 dpif_port_exists(const struct dpif
*dpif
, const char *devname
)
605 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, NULL
);
606 if (error
!= 0 && error
!= ENOENT
&& error
!= ENODEV
) {
607 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %s: %s",
608 dpif_name(dpif
), devname
, ovs_strerror(error
));
614 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
615 * initializes '*port' appropriately; on failure, returns a positive errno
618 * The caller owns the data in 'port' and must free it with
619 * dpif_port_destroy() when it is no longer needed. */
621 dpif_port_query_by_number(const struct dpif
*dpif
, odp_port_t port_no
,
622 struct dpif_port
*port
)
624 int error
= dpif
->dpif_class
->port_query_by_number(dpif
, port_no
, port
);
626 VLOG_DBG_RL(&dpmsg_rl
, "%s: port %"PRIu32
" is device %s",
627 dpif_name(dpif
), port_no
, port
->name
);
629 memset(port
, 0, sizeof *port
);
630 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %"PRIu32
": %s",
631 dpif_name(dpif
), port_no
, ovs_strerror(error
));
636 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
637 * initializes '*port' appropriately; on failure, returns a positive errno
640 * The caller owns the data in 'port' and must free it with
641 * dpif_port_destroy() when it is no longer needed. */
643 dpif_port_query_by_name(const struct dpif
*dpif
, const char *devname
,
644 struct dpif_port
*port
)
646 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, port
);
648 VLOG_DBG_RL(&dpmsg_rl
, "%s: device %s is on port %"PRIu32
,
649 dpif_name(dpif
), devname
, port
->port_no
);
651 memset(port
, 0, sizeof *port
);
653 /* For ENOENT or ENODEV we use DBG level because the caller is probably
654 * interested in whether 'dpif' actually has a port 'devname', so that
655 * it's not an issue worth logging if it doesn't. Other errors are
656 * uncommon and more likely to indicate a real problem. */
658 error
== ENOENT
|| error
== ENODEV
? VLL_DBG
: VLL_WARN
,
659 "%s: failed to query port %s: %s",
660 dpif_name(dpif
), devname
, ovs_strerror(error
));
665 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
666 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
667 * flows whose packets arrived on port 'port_no'. In the case where the
668 * provider allocates multiple Netlink PIDs to a single port, it may use
669 * 'hash' to spread load among them. The caller need not use a particular
670 * hash function; a 5-tuple hash is suitable.
672 * (The datapath implementation might use some different hash function for
673 * distributing packets received via flow misses among PIDs. This means
674 * that packets received via flow misses might be reordered relative to
675 * packets received via userspace actions. This is not ordinarily a
678 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
679 * allocated to any port, that the client may use for special purposes.
681 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
682 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
683 * disabled and then re-enabled, so a client that does that must be prepared to
684 * update all of the flows that it installed that contain
685 * OVS_ACTION_ATTR_USERSPACE actions. */
687 dpif_port_get_pid(const struct dpif
*dpif
, odp_port_t port_no
, uint32_t hash
)
689 return (dpif
->dpif_class
->port_get_pid
690 ? (dpif
->dpif_class
->port_get_pid
)(dpif
, port_no
, hash
)
694 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
695 * the port's name into the 'name_size' bytes in 'name', ensuring that the
696 * result is null-terminated. On failure, returns a positive errno value and
697 * makes 'name' the empty string. */
699 dpif_port_get_name(struct dpif
*dpif
, odp_port_t port_no
,
700 char *name
, size_t name_size
)
702 struct dpif_port port
;
705 ovs_assert(name_size
> 0);
707 error
= dpif_port_query_by_number(dpif
, port_no
, &port
);
709 ovs_strlcpy(name
, port
.name
, name_size
);
710 dpif_port_destroy(&port
);
717 /* Initializes 'dump' to begin dumping the ports in a dpif.
719 * This function provides no status indication. An error status for the entire
720 * dump operation is provided when it is completed by calling
721 * dpif_port_dump_done().
724 dpif_port_dump_start(struct dpif_port_dump
*dump
, const struct dpif
*dpif
)
727 dump
->error
= dpif
->dpif_class
->port_dump_start(dpif
, &dump
->state
);
728 log_operation(dpif
, "port_dump_start", dump
->error
);
731 /* Attempts to retrieve another port from 'dump', which must have been
732 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
733 * into 'port' and returns true. On failure, returns false.
735 * Failure might indicate an actual error or merely that the last port has been
736 * dumped. An error status for the entire dump operation is provided when it
737 * is completed by calling dpif_port_dump_done().
739 * The dpif owns the data stored in 'port'. It will remain valid until at
740 * least the next time 'dump' is passed to dpif_port_dump_next() or
741 * dpif_port_dump_done(). */
743 dpif_port_dump_next(struct dpif_port_dump
*dump
, struct dpif_port
*port
)
745 const struct dpif
*dpif
= dump
->dpif
;
751 dump
->error
= dpif
->dpif_class
->port_dump_next(dpif
, dump
->state
, port
);
752 if (dump
->error
== EOF
) {
753 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all ports", dpif_name(dpif
));
755 log_operation(dpif
, "port_dump_next", dump
->error
);
759 dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
765 /* Completes port table dump operation 'dump', which must have been initialized
766 * with dpif_port_dump_start(). Returns 0 if the dump operation was
767 * error-free, otherwise a positive errno value describing the problem. */
769 dpif_port_dump_done(struct dpif_port_dump
*dump
)
771 const struct dpif
*dpif
= dump
->dpif
;
773 dump
->error
= dpif
->dpif_class
->port_dump_done(dpif
, dump
->state
);
774 log_operation(dpif
, "port_dump_done", dump
->error
);
776 return dump
->error
== EOF
? 0 : dump
->error
;
779 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
780 * 'dpif' has changed, this function does one of the following:
782 * - Stores the name of the device that was added to or deleted from 'dpif' in
783 * '*devnamep' and returns 0. The caller is responsible for freeing
784 * '*devnamep' (with free()) when it no longer needs it.
786 * - Returns ENOBUFS and sets '*devnamep' to NULL.
788 * This function may also return 'false positives', where it returns 0 and
789 * '*devnamep' names a device that was not actually added or deleted or it
790 * returns ENOBUFS without any change.
792 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
793 * return other positive errno values to indicate that something has gone
796 dpif_port_poll(const struct dpif
*dpif
, char **devnamep
)
798 int error
= dpif
->dpif_class
->port_poll(dpif
, devnamep
);
805 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
806 * value other than EAGAIN. */
808 dpif_port_poll_wait(const struct dpif
*dpif
)
810 dpif
->dpif_class
->port_poll_wait(dpif
);
813 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
814 * arguments must have been initialized through a call to flow_extract().
815 * 'used' is stored into stats->used. */
817 dpif_flow_stats_extract(const struct flow
*flow
, const struct dp_packet
*packet
,
818 long long int used
, struct dpif_flow_stats
*stats
)
820 stats
->tcp_flags
= ntohs(flow
->tcp_flags
);
821 stats
->n_bytes
= dp_packet_size(packet
);
822 stats
->n_packets
= 1;
826 /* Appends a human-readable representation of 'stats' to 's'. */
828 dpif_flow_stats_format(const struct dpif_flow_stats
*stats
, struct ds
*s
)
830 ds_put_format(s
, "packets:%"PRIu64
", bytes:%"PRIu64
", used:",
831 stats
->n_packets
, stats
->n_bytes
);
833 ds_put_format(s
, "%.3fs", (time_msec() - stats
->used
) / 1000.0);
835 ds_put_format(s
, "never");
837 if (stats
->tcp_flags
) {
838 ds_put_cstr(s
, ", flags:");
839 packet_format_tcp_flags(s
, stats
->tcp_flags
);
843 /* Places the hash of the 'key_len' bytes starting at 'key' into '*hash'. */
845 dpif_flow_hash(const struct dpif
*dpif OVS_UNUSED
,
846 const void *key
, size_t key_len
, ovs_u128
*hash
)
848 static struct ovsthread_once once
= OVSTHREAD_ONCE_INITIALIZER
;
849 static uint32_t secret
;
851 if (ovsthread_once_start(&once
)) {
852 secret
= random_uint32();
853 ovsthread_once_done(&once
);
855 hash_bytes128(key
, key_len
, secret
, hash
);
856 uuid_set_bits_v4((struct uuid
*)hash
);
859 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
860 * positive errno value. */
862 dpif_flow_flush(struct dpif
*dpif
)
866 COVERAGE_INC(dpif_flow_flush
);
868 error
= dpif
->dpif_class
->flow_flush(dpif
);
869 log_operation(dpif
, "flow_flush", error
);
873 /* Attempts to install 'key' into the datapath, fetches it, then deletes it.
874 * Returns true if the datapath supported installing 'flow', false otherwise.
877 dpif_probe_feature(struct dpif
*dpif
, const char *name
,
878 const struct ofpbuf
*key
, const ovs_u128
*ufid
)
880 struct dpif_flow flow
;
882 uint64_t stub
[DPIF_FLOW_BUFSIZE
/ 8];
883 bool enable_feature
= false;
886 /* Use DPIF_FP_MODIFY to cover the case where ovs-vswitchd is killed (and
887 * restarted) at just the right time such that feature probes from the
888 * previous run are still present in the datapath. */
889 error
= dpif_flow_put(dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
| DPIF_FP_PROBE
,
890 key
->data
, key
->size
, NULL
, 0, NULL
, 0,
891 ufid
, PMD_ID_NULL
, NULL
);
893 if (error
!= EINVAL
) {
894 VLOG_WARN("%s: %s flow probe failed (%s)",
895 dpif_name(dpif
), name
, ovs_strerror(error
));
900 ofpbuf_use_stack(&reply
, &stub
, sizeof stub
);
901 error
= dpif_flow_get(dpif
, key
->data
, key
->size
, ufid
,
902 PMD_ID_NULL
, &reply
, &flow
);
904 && (!ufid
|| (flow
.ufid_present
&& ovs_u128_equal(ufid
, &flow
.ufid
)))) {
905 enable_feature
= true;
908 error
= dpif_flow_del(dpif
, key
->data
, key
->size
, ufid
,
911 VLOG_WARN("%s: failed to delete %s feature probe flow",
912 dpif_name(dpif
), name
);
915 return enable_feature
;
918 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_GET. */
920 dpif_flow_get(struct dpif
*dpif
,
921 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
922 const unsigned pmd_id
, struct ofpbuf
*buf
, struct dpif_flow
*flow
)
927 op
.type
= DPIF_OP_FLOW_GET
;
928 op
.u
.flow_get
.key
= key
;
929 op
.u
.flow_get
.key_len
= key_len
;
930 op
.u
.flow_get
.ufid
= ufid
;
931 op
.u
.flow_get
.pmd_id
= pmd_id
;
932 op
.u
.flow_get
.buffer
= buf
;
934 memset(flow
, 0, sizeof *flow
);
935 op
.u
.flow_get
.flow
= flow
;
936 op
.u
.flow_get
.flow
->key
= key
;
937 op
.u
.flow_get
.flow
->key_len
= key_len
;
940 dpif_operate(dpif
, &opp
, 1);
945 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_PUT. */
947 dpif_flow_put(struct dpif
*dpif
, enum dpif_flow_put_flags flags
,
948 const struct nlattr
*key
, size_t key_len
,
949 const struct nlattr
*mask
, size_t mask_len
,
950 const struct nlattr
*actions
, size_t actions_len
,
951 const ovs_u128
*ufid
, const unsigned pmd_id
,
952 struct dpif_flow_stats
*stats
)
957 op
.type
= DPIF_OP_FLOW_PUT
;
958 op
.u
.flow_put
.flags
= flags
;
959 op
.u
.flow_put
.key
= key
;
960 op
.u
.flow_put
.key_len
= key_len
;
961 op
.u
.flow_put
.mask
= mask
;
962 op
.u
.flow_put
.mask_len
= mask_len
;
963 op
.u
.flow_put
.actions
= actions
;
964 op
.u
.flow_put
.actions_len
= actions_len
;
965 op
.u
.flow_put
.ufid
= ufid
;
966 op
.u
.flow_put
.pmd_id
= pmd_id
;
967 op
.u
.flow_put
.stats
= stats
;
970 dpif_operate(dpif
, &opp
, 1);
975 /* A dpif_operate() wrapper for performing a single DPIF_OP_FLOW_DEL. */
977 dpif_flow_del(struct dpif
*dpif
,
978 const struct nlattr
*key
, size_t key_len
, const ovs_u128
*ufid
,
979 const unsigned pmd_id
, struct dpif_flow_stats
*stats
)
984 op
.type
= DPIF_OP_FLOW_DEL
;
985 op
.u
.flow_del
.key
= key
;
986 op
.u
.flow_del
.key_len
= key_len
;
987 op
.u
.flow_del
.ufid
= ufid
;
988 op
.u
.flow_del
.pmd_id
= pmd_id
;
989 op
.u
.flow_del
.stats
= stats
;
990 op
.u
.flow_del
.terse
= false;
993 dpif_operate(dpif
, &opp
, 1);
998 /* Creates and returns a new 'struct dpif_flow_dump' for iterating through the
999 * flows in 'dpif'. If 'terse' is true, then only UFID and statistics will
1000 * be returned in the dump. Otherwise, all fields will be returned.
1002 * This function always successfully returns a dpif_flow_dump. Error
1003 * reporting is deferred to dpif_flow_dump_destroy(). */
1004 struct dpif_flow_dump
*
1005 dpif_flow_dump_create(const struct dpif
*dpif
, bool terse
)
1007 return dpif
->dpif_class
->flow_dump_create(dpif
, terse
);
1010 /* Destroys 'dump', which must have been created with dpif_flow_dump_create().
1011 * All dpif_flow_dump_thread structures previously created for 'dump' must
1012 * previously have been destroyed.
1014 * Returns 0 if the dump operation was error-free, otherwise a positive errno
1015 * value describing the problem. */
1017 dpif_flow_dump_destroy(struct dpif_flow_dump
*dump
)
1019 const struct dpif
*dpif
= dump
->dpif
;
1020 int error
= dpif
->dpif_class
->flow_dump_destroy(dump
);
1021 log_operation(dpif
, "flow_dump_destroy", error
);
1022 return error
== EOF
? 0 : error
;
1025 /* Returns new thread-local state for use with dpif_flow_dump_next(). */
1026 struct dpif_flow_dump_thread
*
1027 dpif_flow_dump_thread_create(struct dpif_flow_dump
*dump
)
1029 return dump
->dpif
->dpif_class
->flow_dump_thread_create(dump
);
1032 /* Releases 'thread'. */
1034 dpif_flow_dump_thread_destroy(struct dpif_flow_dump_thread
*thread
)
1036 thread
->dpif
->dpif_class
->flow_dump_thread_destroy(thread
);
1039 /* Attempts to retrieve up to 'max_flows' more flows from 'thread'. Returns 0
1040 * if and only if no flows remained to be retrieved, otherwise a positive
1041 * number reflecting the number of elements in 'flows[]' that were updated.
1042 * The number of flows returned might be less than 'max_flows' because
1043 * fewer than 'max_flows' remained, because this particular datapath does not
1044 * benefit from batching, or because an error occurred partway through
1045 * retrieval. Thus, the caller should continue calling until a 0 return value,
1046 * even if intermediate return values are less than 'max_flows'.
1048 * No error status is immediately provided. An error status for the entire
1049 * dump operation is provided when it is completed by calling
1050 * dpif_flow_dump_destroy().
1052 * All of the data stored into 'flows' is owned by the datapath, not by the
1053 * caller, and the caller must not modify or free it. The datapath guarantees
1054 * that it remains accessible and unchanged until the first of:
1055 * - The next call to dpif_flow_dump_next() for 'thread', or
1056 * - The next rcu quiescent period. */
1058 dpif_flow_dump_next(struct dpif_flow_dump_thread
*thread
,
1059 struct dpif_flow
*flows
, int max_flows
)
1061 struct dpif
*dpif
= thread
->dpif
;
1064 ovs_assert(max_flows
> 0);
1065 n
= dpif
->dpif_class
->flow_dump_next(thread
, flows
, max_flows
);
1067 struct dpif_flow
*f
;
1069 for (f
= flows
; f
< &flows
[n
] && should_log_flow_message(0); f
++) {
1070 log_flow_message(dpif
, 0, "flow_dump",
1071 f
->key
, f
->key_len
, f
->mask
, f
->mask_len
,
1072 &f
->ufid
, &f
->stats
, f
->actions
, f
->actions_len
);
1075 VLOG_DBG_RL(&dpmsg_rl
, "%s: dumped all flows", dpif_name(dpif
));
1080 struct dpif_execute_helper_aux
{
1085 /* This is called for actions that need the context of the datapath to be
1088 dpif_execute_helper_cb(void *aux_
, struct dp_packet
**packets
, int cnt
,
1089 const struct nlattr
*action
, bool may_steal OVS_UNUSED
)
1091 struct dpif_execute_helper_aux
*aux
= aux_
;
1092 int type
= nl_attr_type(action
);
1093 struct dp_packet
*packet
= *packets
;
1095 ovs_assert(cnt
== 1);
1097 switch ((enum ovs_action_attr
)type
) {
1098 case OVS_ACTION_ATTR_OUTPUT
:
1099 case OVS_ACTION_ATTR_TUNNEL_PUSH
:
1100 case OVS_ACTION_ATTR_TUNNEL_POP
:
1101 case OVS_ACTION_ATTR_USERSPACE
:
1102 case OVS_ACTION_ATTR_RECIRC
: {
1103 struct dpif_execute execute
;
1104 struct ofpbuf execute_actions
;
1105 uint64_t stub
[256 / 8];
1106 struct pkt_metadata
*md
= &packet
->md
;
1108 if (md
->tunnel
.ip_dst
) {
1109 /* The Linux kernel datapath throws away the tunnel information
1110 * that we supply as metadata. We have to use a "set" action to
1112 ofpbuf_use_stub(&execute_actions
, stub
, sizeof stub
);
1113 odp_put_tunnel_action(&md
->tunnel
, &execute_actions
);
1114 ofpbuf_put(&execute_actions
, action
, NLA_ALIGN(action
->nla_len
));
1116 execute
.actions
= execute_actions
.data
;
1117 execute
.actions_len
= execute_actions
.size
;
1119 execute
.actions
= action
;
1120 execute
.actions_len
= NLA_ALIGN(action
->nla_len
);
1123 execute
.packet
= packet
;
1124 execute
.needs_help
= false;
1125 execute
.probe
= false;
1126 aux
->error
= dpif_execute(aux
->dpif
, &execute
);
1127 log_execute_message(aux
->dpif
, &execute
, true, aux
->error
);
1129 if (md
->tunnel
.ip_dst
) {
1130 ofpbuf_uninit(&execute_actions
);
1135 case OVS_ACTION_ATTR_HASH
:
1136 case OVS_ACTION_ATTR_PUSH_VLAN
:
1137 case OVS_ACTION_ATTR_POP_VLAN
:
1138 case OVS_ACTION_ATTR_PUSH_MPLS
:
1139 case OVS_ACTION_ATTR_POP_MPLS
:
1140 case OVS_ACTION_ATTR_SET
:
1141 case OVS_ACTION_ATTR_SET_MASKED
:
1142 case OVS_ACTION_ATTR_SAMPLE
:
1143 case OVS_ACTION_ATTR_UNSPEC
:
1144 case __OVS_ACTION_ATTR_MAX
:
1149 /* Executes 'execute' by performing most of the actions in userspace and
1150 * passing the fully constructed packets to 'dpif' for output and userspace
1153 * This helps with actions that a given 'dpif' doesn't implement directly. */
1155 dpif_execute_with_help(struct dpif
*dpif
, struct dpif_execute
*execute
)
1157 struct dpif_execute_helper_aux aux
= {dpif
, 0};
1158 struct dp_packet
*pp
;
1160 COVERAGE_INC(dpif_execute_with_help
);
1162 pp
= execute
->packet
;
1163 odp_execute_actions(&aux
, &pp
, 1, false, execute
->actions
,
1164 execute
->actions_len
, dpif_execute_helper_cb
);
1168 /* Returns true if the datapath needs help executing 'execute'. */
1170 dpif_execute_needs_help(const struct dpif_execute
*execute
)
1172 return execute
->needs_help
|| nl_attr_oversized(execute
->actions_len
);
1175 /* A dpif_operate() wrapper for performing a single DPIF_OP_EXECUTE. */
1177 dpif_execute(struct dpif
*dpif
, struct dpif_execute
*execute
)
1179 if (execute
->actions_len
) {
1180 struct dpif_op
*opp
;
1183 op
.type
= DPIF_OP_EXECUTE
;
1184 op
.u
.execute
= *execute
;
1187 dpif_operate(dpif
, &opp
, 1);
1195 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1196 * which they are specified. Places each operation's results in the "output"
1197 * members documented in comments, and 0 in the 'error' member on success or a
1198 * positive errno on failure. */
1200 dpif_operate(struct dpif
*dpif
, struct dpif_op
**ops
, size_t n_ops
)
1205 /* Count 'chunk', the number of ops that can be executed without
1206 * needing any help. Ops that need help should be rare, so we
1207 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1208 for (chunk
= 0; chunk
< n_ops
; chunk
++) {
1209 struct dpif_op
*op
= ops
[chunk
];
1211 if (op
->type
== DPIF_OP_EXECUTE
1212 && dpif_execute_needs_help(&op
->u
.execute
)) {
1218 /* Execute a chunk full of ops that the dpif provider can
1219 * handle itself, without help. */
1222 dpif
->dpif_class
->operate(dpif
, ops
, chunk
);
1224 for (i
= 0; i
< chunk
; i
++) {
1225 struct dpif_op
*op
= ops
[i
];
1226 int error
= op
->error
;
1229 case DPIF_OP_FLOW_PUT
: {
1230 struct dpif_flow_put
*put
= &op
->u
.flow_put
;
1232 COVERAGE_INC(dpif_flow_put
);
1233 log_flow_put_message(dpif
, put
, error
);
1234 if (error
&& put
->stats
) {
1235 memset(put
->stats
, 0, sizeof *put
->stats
);
1240 case DPIF_OP_FLOW_GET
: {
1241 struct dpif_flow_get
*get
= &op
->u
.flow_get
;
1243 COVERAGE_INC(dpif_flow_get
);
1245 memset(get
->flow
, 0, sizeof *get
->flow
);
1247 log_flow_get_message(dpif
, get
, error
);
1252 case DPIF_OP_FLOW_DEL
: {
1253 struct dpif_flow_del
*del
= &op
->u
.flow_del
;
1255 COVERAGE_INC(dpif_flow_del
);
1256 log_flow_del_message(dpif
, del
, error
);
1257 if (error
&& del
->stats
) {
1258 memset(del
->stats
, 0, sizeof *del
->stats
);
1263 case DPIF_OP_EXECUTE
:
1264 COVERAGE_INC(dpif_execute
);
1265 log_execute_message(dpif
, &op
->u
.execute
, false, error
);
1273 /* Help the dpif provider to execute one op. */
1274 struct dpif_op
*op
= ops
[0];
1276 COVERAGE_INC(dpif_execute
);
1277 op
->error
= dpif_execute_with_help(dpif
, &op
->u
.execute
);
1284 /* Returns a string that represents 'type', for use in log messages. */
1286 dpif_upcall_type_to_string(enum dpif_upcall_type type
)
1289 case DPIF_UC_MISS
: return "miss";
1290 case DPIF_UC_ACTION
: return "action";
1291 case DPIF_N_UC_TYPES
: default: return "<unknown>";
1295 /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1296 * if successful, otherwise a positive errno value.
1298 * Turning packet receive off and then back on may change the Netlink PID
1299 * assignments returned by dpif_port_get_pid(). If the client does this, it
1300 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1301 * using the new PID assignment. */
1303 dpif_recv_set(struct dpif
*dpif
, bool enable
)
1307 if (dpif
->dpif_class
->recv_set
) {
1308 error
= dpif
->dpif_class
->recv_set(dpif
, enable
);
1309 log_operation(dpif
, "recv_set", error
);
1314 /* Refreshes the poll loops and Netlink sockets associated to each port,
1315 * when the number of upcall handlers (upcall receiving thread) is changed
1316 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
1319 * Since multiple upcall handlers can read upcalls simultaneously from
1320 * 'dpif', each port can have multiple Netlink sockets, one per upcall
1321 * handler. So, handlers_set() is responsible for the following tasks:
1323 * When receiving upcall is enabled, extends or creates the
1324 * configuration to support:
1326 * - 'n_handlers' Netlink sockets for each port.
1328 * - 'n_handlers' poll loops, one for each upcall handler.
1330 * - registering the Netlink sockets for the same upcall handler to
1331 * the corresponding poll loop.
1333 * Returns 0 if successful, otherwise a positive errno value. */
1335 dpif_handlers_set(struct dpif
*dpif
, uint32_t n_handlers
)
1339 if (dpif
->dpif_class
->handlers_set
) {
1340 error
= dpif
->dpif_class
->handlers_set(dpif
, n_handlers
);
1341 log_operation(dpif
, "handlers_set", error
);
1347 dpif_register_upcall_cb(struct dpif
*dpif
, upcall_callback
*cb
, void *aux
)
1349 if (dpif
->dpif_class
->register_upcall_cb
) {
1350 dpif
->dpif_class
->register_upcall_cb(dpif
, cb
, aux
);
1355 dpif_enable_upcall(struct dpif
*dpif
)
1357 if (dpif
->dpif_class
->enable_upcall
) {
1358 dpif
->dpif_class
->enable_upcall(dpif
);
1363 dpif_disable_upcall(struct dpif
*dpif
)
1365 if (dpif
->dpif_class
->disable_upcall
) {
1366 dpif
->dpif_class
->disable_upcall(dpif
);
1371 dpif_print_packet(struct dpif
*dpif
, struct dpif_upcall
*upcall
)
1373 if (!VLOG_DROP_DBG(&dpmsg_rl
)) {
1377 packet
= ofp_packet_to_string(dp_packet_data(&upcall
->packet
),
1378 dp_packet_size(&upcall
->packet
));
1381 odp_flow_key_format(upcall
->key
, upcall
->key_len
, &flow
);
1383 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1384 dpif_name(dpif
), dpif_upcall_type_to_string(upcall
->type
),
1385 ds_cstr(&flow
), packet
);
1392 /* If 'dpif' creates its own I/O polling threads, refreshes poll threads
1395 dpif_poll_threads_set(struct dpif
*dpif
, unsigned int n_rxqs
,
1400 if (dpif
->dpif_class
->poll_threads_set
) {
1401 error
= dpif
->dpif_class
->poll_threads_set(dpif
, n_rxqs
, cmask
);
1403 log_operation(dpif
, "poll_threads_set", error
);
1410 /* Polls for an upcall from 'dpif' for an upcall handler. Since there
1411 * there can be multiple poll loops, 'handler_id' is needed as index to
1412 * identify the corresponding poll loop. If successful, stores the upcall
1413 * into '*upcall', using 'buf' for storage. Should only be called if
1414 * 'recv_set' has been used to enable receiving packets from 'dpif'.
1416 * 'upcall->key' and 'upcall->userdata' point into data in the caller-provided
1417 * 'buf', so their memory cannot be freed separately from 'buf'.
1419 * The caller owns the data of 'upcall->packet' and may modify it. If
1420 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
1421 * will be reallocated. This requires the data of 'upcall->packet' to be
1422 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
1423 * when an error is returned, the 'upcall->packet' may be uninitialized
1424 * and should not be released.
1426 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1427 * if no upcall is immediately available. */
1429 dpif_recv(struct dpif
*dpif
, uint32_t handler_id
, struct dpif_upcall
*upcall
,
1434 if (dpif
->dpif_class
->recv
) {
1435 error
= dpif
->dpif_class
->recv(dpif
, handler_id
, upcall
, buf
);
1437 dpif_print_packet(dpif
, upcall
);
1438 } else if (error
!= EAGAIN
) {
1439 log_operation(dpif
, "recv", error
);
1445 /* Discards all messages that would otherwise be received by dpif_recv() on
1448 dpif_recv_purge(struct dpif
*dpif
)
1450 COVERAGE_INC(dpif_purge
);
1451 if (dpif
->dpif_class
->recv_purge
) {
1452 dpif
->dpif_class
->recv_purge(dpif
);
1456 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
1457 * 'dpif' has a message queued to be received with the recv member
1458 * function. Since there can be multiple poll loops, 'handler_id' is
1459 * needed as index to identify the corresponding poll loop. */
1461 dpif_recv_wait(struct dpif
*dpif
, uint32_t handler_id
)
1463 if (dpif
->dpif_class
->recv_wait
) {
1464 dpif
->dpif_class
->recv_wait(dpif
, handler_id
);
1469 * Return the datapath version. Caller is responsible for freeing
1473 dpif_get_dp_version(const struct dpif
*dpif
)
1475 char *version
= NULL
;
1477 if (dpif
->dpif_class
->get_datapath_version
) {
1478 version
= dpif
->dpif_class
->get_datapath_version();
1484 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1485 * and '*engine_id', respectively. */
1487 dpif_get_netflow_ids(const struct dpif
*dpif
,
1488 uint8_t *engine_type
, uint8_t *engine_id
)
1490 *engine_type
= dpif
->netflow_engine_type
;
1491 *engine_id
= dpif
->netflow_engine_id
;
1494 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1495 * value used for setting packet priority.
1496 * On success, returns 0 and stores the priority into '*priority'.
1497 * On failure, returns a positive errno value and stores 0 into '*priority'. */
1499 dpif_queue_to_priority(const struct dpif
*dpif
, uint32_t queue_id
,
1502 int error
= (dpif
->dpif_class
->queue_to_priority
1503 ? dpif
->dpif_class
->queue_to_priority(dpif
, queue_id
,
1509 log_operation(dpif
, "queue_to_priority", error
);
1514 dpif_init(struct dpif
*dpif
, const struct dpif_class
*dpif_class
,
1516 uint8_t netflow_engine_type
, uint8_t netflow_engine_id
)
1518 dpif
->dpif_class
= dpif_class
;
1519 dpif
->base_name
= xstrdup(name
);
1520 dpif
->full_name
= xasprintf("%s@%s", dpif_class
->type
, name
);
1521 dpif
->netflow_engine_type
= netflow_engine_type
;
1522 dpif
->netflow_engine_id
= netflow_engine_id
;
1525 /* Undoes the results of initialization.
1527 * Normally this function only needs to be called from dpif_close().
1528 * However, it may be called by providers due to an error on opening
1529 * that occurs after initialization. It this case dpif_close() would
1530 * never be called. */
1532 dpif_uninit(struct dpif
*dpif
, bool close
)
1534 char *base_name
= dpif
->base_name
;
1535 char *full_name
= dpif
->full_name
;
1538 dpif
->dpif_class
->close(dpif
);
1546 log_operation(const struct dpif
*dpif
, const char *operation
, int error
)
1549 VLOG_DBG_RL(&dpmsg_rl
, "%s: %s success", dpif_name(dpif
), operation
);
1550 } else if (ofperr_is_valid(error
)) {
1551 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1552 dpif_name(dpif
), operation
, ofperr_get_name(error
));
1554 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1555 dpif_name(dpif
), operation
, ovs_strerror(error
));
1559 static enum vlog_level
1560 flow_message_log_level(int error
)
1562 /* If flows arrive in a batch, userspace may push down multiple
1563 * unique flow definitions that overlap when wildcards are applied.
1564 * Kernels that support flow wildcarding will reject these flows as
1565 * duplicates (EEXIST), so lower the log level to debug for these
1566 * types of messages. */
1567 return (error
&& error
!= EEXIST
) ? VLL_WARN
: VLL_DBG
;
1571 should_log_flow_message(int error
)
1573 return !vlog_should_drop(THIS_MODULE
, flow_message_log_level(error
),
1574 error
? &error_rl
: &dpmsg_rl
);
1578 log_flow_message(const struct dpif
*dpif
, int error
, const char *operation
,
1579 const struct nlattr
*key
, size_t key_len
,
1580 const struct nlattr
*mask
, size_t mask_len
,
1581 const ovs_u128
*ufid
, const struct dpif_flow_stats
*stats
,
1582 const struct nlattr
*actions
, size_t actions_len
)
1584 struct ds ds
= DS_EMPTY_INITIALIZER
;
1585 ds_put_format(&ds
, "%s: ", dpif_name(dpif
));
1587 ds_put_cstr(&ds
, "failed to ");
1589 ds_put_format(&ds
, "%s ", operation
);
1591 ds_put_format(&ds
, "(%s) ", ovs_strerror(error
));
1594 odp_format_ufid(ufid
, &ds
);
1595 ds_put_cstr(&ds
, " ");
1597 odp_flow_format(key
, key_len
, mask
, mask_len
, NULL
, &ds
, true);
1599 ds_put_cstr(&ds
, ", ");
1600 dpif_flow_stats_format(stats
, &ds
);
1602 if (actions
|| actions_len
) {
1603 ds_put_cstr(&ds
, ", actions:");
1604 format_odp_actions(&ds
, actions
, actions_len
);
1606 vlog(THIS_MODULE
, flow_message_log_level(error
), "%s", ds_cstr(&ds
));
1611 log_flow_put_message(struct dpif
*dpif
, const struct dpif_flow_put
*put
,
1614 if (should_log_flow_message(error
) && !(put
->flags
& DPIF_FP_PROBE
)) {
1618 ds_put_cstr(&s
, "put");
1619 if (put
->flags
& DPIF_FP_CREATE
) {
1620 ds_put_cstr(&s
, "[create]");
1622 if (put
->flags
& DPIF_FP_MODIFY
) {
1623 ds_put_cstr(&s
, "[modify]");
1625 if (put
->flags
& DPIF_FP_ZERO_STATS
) {
1626 ds_put_cstr(&s
, "[zero]");
1628 log_flow_message(dpif
, error
, ds_cstr(&s
),
1629 put
->key
, put
->key_len
, put
->mask
, put
->mask_len
,
1630 put
->ufid
, put
->stats
, put
->actions
,
1637 log_flow_del_message(struct dpif
*dpif
, const struct dpif_flow_del
*del
,
1640 if (should_log_flow_message(error
)) {
1641 log_flow_message(dpif
, error
, "flow_del", del
->key
, del
->key_len
,
1642 NULL
, 0, del
->ufid
, !error
? del
->stats
: NULL
,
1647 /* Logs that 'execute' was executed on 'dpif' and completed with errno 'error'
1648 * (0 for success). 'subexecute' should be true if the execution is a result
1649 * of breaking down a larger execution that needed help, false otherwise.
1652 * XXX In theory, the log message could be deceptive because this function is
1653 * called after the dpif_provider's '->execute' function, which is allowed to
1654 * modify execute->packet and execute->md. In practice, though:
1656 * - dpif-netlink doesn't modify execute->packet or execute->md.
1658 * - dpif-netdev does modify them but it is less likely to have problems
1659 * because it is built into ovs-vswitchd and cannot have version skew,
1662 * It would still be better to avoid the potential problem. I don't know of a
1663 * good way to do that, though, that isn't expensive. */
1665 log_execute_message(struct dpif
*dpif
, const struct dpif_execute
*execute
,
1666 bool subexecute
, int error
)
1668 if (!(error
? VLOG_DROP_WARN(&error_rl
) : VLOG_DROP_DBG(&dpmsg_rl
))
1669 && !execute
->probe
) {
1670 struct ds ds
= DS_EMPTY_INITIALIZER
;
1673 packet
= ofp_packet_to_string(dp_packet_data(execute
->packet
),
1674 dp_packet_size(execute
->packet
));
1675 ds_put_format(&ds
, "%s: %sexecute ",
1677 (subexecute
? "sub-"
1678 : dpif_execute_needs_help(execute
) ? "super-"
1680 format_odp_actions(&ds
, execute
->actions
, execute
->actions_len
);
1682 ds_put_format(&ds
, " failed (%s)", ovs_strerror(error
));
1684 ds_put_format(&ds
, " on packet %s", packet
);
1685 vlog(THIS_MODULE
, error
? VLL_WARN
: VLL_DBG
, "%s", ds_cstr(&ds
));
1692 log_flow_get_message(const struct dpif
*dpif
, const struct dpif_flow_get
*get
,
1695 if (should_log_flow_message(error
)) {
1696 log_flow_message(dpif
, error
, "flow_get",
1697 get
->key
, get
->key_len
,
1698 get
->flow
->mask
, get
->flow
->mask_len
,
1699 get
->ufid
, &get
->flow
->stats
,
1700 get
->flow
->actions
, get
->flow
->actions_len
);
1705 dpif_supports_tnl_push_pop(const struct dpif
*dpif
)
1707 return !strcmp(dpif
->dpif_class
->type
, "netdev") ||
1708 !strcmp(dpif
->dpif_class
->type
, "dummy");