]>
git.proxmox.com Git - ovs.git/blob - lib/dpif.c
2 * Copyright (c) 2008, 2009, 2010 Nicira Networks.
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 "ofp-print.h"
36 #include "poll-loop.h"
43 VLOG_DEFINE_THIS_MODULE(dpif
);
45 static const struct dpif_class
*base_dpif_classes
[] = {
52 struct registered_dpif_class
{
53 struct dpif_class dpif_class
;
56 static struct shash dpif_classes
= SHASH_INITIALIZER(&dpif_classes
);
58 /* Rate limit for individual messages going to or from the datapath, output at
59 * DBG level. This is very high because, if these are enabled, it is because
60 * we really need to see them. */
61 static struct vlog_rate_limit dpmsg_rl
= VLOG_RATE_LIMIT_INIT(600, 600);
63 /* Not really much point in logging many dpif errors. */
64 static struct vlog_rate_limit error_rl
= VLOG_RATE_LIMIT_INIT(60, 5);
66 static void log_operation(const struct dpif
*, const char *operation
,
68 static void log_flow_operation(const struct dpif
*, const char *operation
,
69 int error
, struct odp_flow
*flow
);
70 static void log_flow_put(struct dpif
*, int error
,
71 const struct odp_flow_put
*);
72 static bool should_log_flow_message(int error
);
73 static void check_rw_odp_flow(struct odp_flow
*);
78 static int status
= -1;
84 for (i
= 0; i
< ARRAY_SIZE(base_dpif_classes
); i
++) {
85 dp_register_provider(base_dpif_classes
[i
]);
90 /* Performs periodic work needed by all the various kinds of dpifs.
92 * If your program opens any dpifs, it must call both this function and
93 * netdev_run() within its main poll loop. */
97 struct shash_node
*node
;
98 SHASH_FOR_EACH(node
, &dpif_classes
) {
99 const struct registered_dpif_class
*registered_class
= node
->data
;
100 if (registered_class
->dpif_class
.run
) {
101 registered_class
->dpif_class
.run();
106 /* Arranges for poll_block() to wake up when dp_run() needs to be called.
108 * If your program opens any dpifs, it must call both this function and
109 * netdev_wait() within its main poll loop. */
113 struct shash_node
*node
;
114 SHASH_FOR_EACH(node
, &dpif_classes
) {
115 const struct registered_dpif_class
*registered_class
= node
->data
;
116 if (registered_class
->dpif_class
.wait
) {
117 registered_class
->dpif_class
.wait();
122 /* Registers a new datapath provider. After successful registration, new
123 * datapaths of that type can be opened using dpif_open(). */
125 dp_register_provider(const struct dpif_class
*new_class
)
127 struct registered_dpif_class
*registered_class
;
129 if (shash_find(&dpif_classes
, new_class
->type
)) {
130 VLOG_WARN("attempted to register duplicate datapath provider: %s",
135 registered_class
= xmalloc(sizeof *registered_class
);
136 memcpy(®istered_class
->dpif_class
, new_class
,
137 sizeof registered_class
->dpif_class
);
138 registered_class
->refcount
= 0;
140 shash_add(&dpif_classes
, new_class
->type
, registered_class
);
145 /* Unregisters a datapath provider. 'type' must have been previously
146 * registered and not currently be in use by any dpifs. After unregistration
147 * new datapaths of that type cannot be opened using dpif_open(). */
149 dp_unregister_provider(const char *type
)
151 struct shash_node
*node
;
152 struct registered_dpif_class
*registered_class
;
154 node
= shash_find(&dpif_classes
, type
);
156 VLOG_WARN("attempted to unregister a datapath provider that is not "
157 "registered: %s", type
);
161 registered_class
= node
->data
;
162 if (registered_class
->refcount
) {
163 VLOG_WARN("attempted to unregister in use datapath provider: %s", type
);
167 shash_delete(&dpif_classes
, node
);
168 free(registered_class
);
173 /* Clears 'types' and enumerates the types of all currently registered datapath
174 * providers into it. The caller must first initialize the svec. */
176 dp_enumerate_types(struct svec
*types
)
178 struct shash_node
*node
;
183 SHASH_FOR_EACH(node
, &dpif_classes
) {
184 const struct registered_dpif_class
*registered_class
= node
->data
;
185 svec_add(types
, registered_class
->dpif_class
.type
);
189 /* Clears 'names' and enumerates the names of all known created datapaths with
190 * the given 'type'. The caller must first initialize the svec. Returns 0 if
191 * successful, otherwise a positive errno value.
193 * Some kinds of datapaths might not be practically enumerable. This is not
194 * considered an error. */
196 dp_enumerate_names(const char *type
, struct svec
*names
)
198 const struct registered_dpif_class
*registered_class
;
199 const struct dpif_class
*dpif_class
;
205 registered_class
= shash_find_data(&dpif_classes
, type
);
206 if (!registered_class
) {
207 VLOG_WARN("could not enumerate unknown type: %s", type
);
211 dpif_class
= ®istered_class
->dpif_class
;
212 error
= dpif_class
->enumerate
? dpif_class
->enumerate(names
) : 0;
215 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class
->type
,
222 /* Parses 'datapath name', which is of the form type@name into its
223 * component pieces. 'name' and 'type' must be freed by the caller. */
225 dp_parse_name(const char *datapath_name_
, char **name
, char **type
)
227 char *datapath_name
= xstrdup(datapath_name_
);
230 separator
= strchr(datapath_name
, '@');
233 *type
= datapath_name
;
234 *name
= xstrdup(separator
+ 1);
236 *name
= datapath_name
;
242 do_open(const char *name
, const char *type
, bool create
, struct dpif
**dpifp
)
244 struct dpif
*dpif
= NULL
;
246 struct registered_dpif_class
*registered_class
;
250 if (!type
|| *type
== '\0') {
254 registered_class
= shash_find_data(&dpif_classes
, type
);
255 if (!registered_class
) {
256 VLOG_WARN("could not create datapath %s of unknown type %s", name
,
258 error
= EAFNOSUPPORT
;
262 error
= registered_class
->dpif_class
.open(name
, type
, create
, &dpif
);
264 registered_class
->refcount
++;
268 *dpifp
= error
? NULL
: dpif
;
272 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
273 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
274 * the empty string to specify the default system type. Returns 0 if
275 * successful, otherwise a positive errno value. On success stores a pointer
276 * to the datapath in '*dpifp', otherwise a null pointer. */
278 dpif_open(const char *name
, const char *type
, struct dpif
**dpifp
)
280 return do_open(name
, type
, false, dpifp
);
283 /* Tries to create and open a new datapath with the given 'name' and 'type'.
284 * 'type' may be either NULL or the empty string to specify the default system
285 * type. Will fail if a datapath with 'name' and 'type' already exists.
286 * Returns 0 if successful, otherwise a positive errno value. On success
287 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
289 dpif_create(const char *name
, const char *type
, struct dpif
**dpifp
)
291 return do_open(name
, type
, true, dpifp
);
294 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
295 * does not exist. 'type' may be either NULL or the empty string to specify
296 * the default system type. Returns 0 if successful, otherwise a positive
297 * errno value. On success stores a pointer to the datapath in '*dpifp',
298 * otherwise a null pointer. */
300 dpif_create_and_open(const char *name
, const char *type
, struct dpif
**dpifp
)
304 error
= dpif_create(name
, type
, dpifp
);
305 if (error
== EEXIST
|| error
== EBUSY
) {
306 error
= dpif_open(name
, type
, dpifp
);
308 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
309 name
, strerror(error
));
312 VLOG_WARN("failed to create datapath %s: %s", name
, strerror(error
));
317 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
318 * itself; call dpif_delete() first, instead, if that is desirable. */
320 dpif_close(struct dpif
*dpif
)
323 struct registered_dpif_class
*registered_class
;
325 registered_class
= shash_find_data(&dpif_classes
,
326 dpif
->dpif_class
->type
);
327 assert(registered_class
);
328 assert(registered_class
->refcount
);
330 registered_class
->refcount
--;
331 dpif_uninit(dpif
, true);
335 /* Returns the name of datapath 'dpif' prefixed with the type
336 * (for use in log messages). */
338 dpif_name(const struct dpif
*dpif
)
340 return dpif
->full_name
;
343 /* Returns the name of datapath 'dpif' without the type
344 * (for use in device names). */
346 dpif_base_name(const struct dpif
*dpif
)
348 return dpif
->base_name
;
351 /* Enumerates all names that may be used to open 'dpif' into 'all_names'. The
352 * Linux datapath, for example, supports opening a datapath both by number,
353 * e.g. "dp0", and by the name of the datapath's local port. For some
354 * datapaths, this might be an infinite set (e.g. in a file name, slashes may
355 * be duplicated any number of times), in which case only the names most likely
356 * to be used will be enumerated.
358 * The caller must already have initialized 'all_names'. Any existing names in
359 * 'all_names' will not be disturbed. */
361 dpif_get_all_names(const struct dpif
*dpif
, struct svec
*all_names
)
363 if (dpif
->dpif_class
->get_all_names
) {
364 int error
= dpif
->dpif_class
->get_all_names(dpif
, all_names
);
366 VLOG_WARN_RL(&error_rl
,
367 "failed to retrieve names for datpath %s: %s",
368 dpif_name(dpif
), strerror(error
));
372 svec_add(all_names
, dpif_base_name(dpif
));
377 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
378 * ports. After calling this function, it does not make sense to pass 'dpif'
379 * to any functions other than dpif_name() or dpif_close(). */
381 dpif_delete(struct dpif
*dpif
)
385 COVERAGE_INC(dpif_destroy
);
387 error
= dpif
->dpif_class
->destroy(dpif
);
388 log_operation(dpif
, "delete", error
);
392 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
393 * otherwise a positive errno value. */
395 dpif_get_dp_stats(const struct dpif
*dpif
, struct odp_stats
*stats
)
397 int error
= dpif
->dpif_class
->get_stats(dpif
, stats
);
399 memset(stats
, 0, sizeof *stats
);
401 log_operation(dpif
, "get_stats", error
);
405 /* Retrieves the current IP fragment handling policy for 'dpif' into
406 * '*drop_frags': true indicates that fragments are dropped, false indicates
407 * that fragments are treated in the same way as other IP packets (except that
408 * the L4 header cannot be read). Returns 0 if successful, otherwise a
409 * positive errno value. */
411 dpif_get_drop_frags(const struct dpif
*dpif
, bool *drop_frags
)
413 int error
= dpif
->dpif_class
->get_drop_frags(dpif
, drop_frags
);
417 log_operation(dpif
, "get_drop_frags", error
);
421 /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is
422 * the same as for the get_drop_frags member function. Returns 0 if
423 * successful, otherwise a positive errno value. */
425 dpif_set_drop_frags(struct dpif
*dpif
, bool drop_frags
)
427 int error
= dpif
->dpif_class
->set_drop_frags(dpif
, drop_frags
);
428 log_operation(dpif
, "set_drop_frags", error
);
432 /* Attempts to add 'devname' as a port on 'dpif', given the combination of
433 * ODP_PORT_* flags in 'flags'. If successful, returns 0 and sets '*port_nop'
434 * to the new port's port number (if 'port_nop' is non-null). On failure,
435 * returns a positive errno value and sets '*port_nop' to UINT16_MAX (if
436 * 'port_nop' is non-null). */
438 dpif_port_add(struct dpif
*dpif
, const char *devname
, uint16_t flags
,
444 COVERAGE_INC(dpif_port_add
);
446 error
= dpif
->dpif_class
->port_add(dpif
, devname
, flags
, &port_no
);
448 VLOG_DBG_RL(&dpmsg_rl
, "%s: added %s as port %"PRIu16
,
449 dpif_name(dpif
), devname
, port_no
);
451 VLOG_WARN_RL(&error_rl
, "%s: failed to add %s as port: %s",
452 dpif_name(dpif
), devname
, strerror(error
));
453 port_no
= UINT16_MAX
;
461 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
462 * otherwise a positive errno value. */
464 dpif_port_del(struct dpif
*dpif
, uint16_t port_no
)
468 COVERAGE_INC(dpif_port_del
);
470 error
= dpif
->dpif_class
->port_del(dpif
, port_no
);
471 log_operation(dpif
, "port_del", error
);
475 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
476 * initializes '*port' appropriately; on failure, returns a positive errno
479 dpif_port_query_by_number(const struct dpif
*dpif
, uint16_t port_no
,
480 struct odp_port
*port
)
482 int error
= dpif
->dpif_class
->port_query_by_number(dpif
, port_no
, port
);
484 VLOG_DBG_RL(&dpmsg_rl
, "%s: port %"PRIu16
" is device %s",
485 dpif_name(dpif
), port_no
, port
->devname
);
487 memset(port
, 0, sizeof *port
);
488 VLOG_WARN_RL(&error_rl
, "%s: failed to query port %"PRIu16
": %s",
489 dpif_name(dpif
), port_no
, strerror(error
));
494 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
495 * initializes '*port' appropriately; on failure, returns a positive errno
498 dpif_port_query_by_name(const struct dpif
*dpif
, const char *devname
,
499 struct odp_port
*port
)
501 int error
= dpif
->dpif_class
->port_query_by_name(dpif
, devname
, port
);
503 VLOG_DBG_RL(&dpmsg_rl
, "%s: device %s is on port %"PRIu16
,
504 dpif_name(dpif
), devname
, port
->port
);
506 memset(port
, 0, sizeof *port
);
508 /* Log level is DBG here because all the current callers are interested
509 * in whether 'dpif' actually has a port 'devname', so that it's not an
510 * issue worth logging if it doesn't. */
511 VLOG_DBG_RL(&error_rl
, "%s: failed to query port %s: %s",
512 dpif_name(dpif
), devname
, strerror(error
));
517 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
518 * the port's name into the 'name_size' bytes in 'name', ensuring that the
519 * result is null-terminated. On failure, returns a positive errno value and
520 * makes 'name' the empty string. */
522 dpif_port_get_name(struct dpif
*dpif
, uint16_t port_no
,
523 char *name
, size_t name_size
)
525 struct odp_port port
;
528 assert(name_size
> 0);
530 error
= dpif_port_query_by_number(dpif
, port_no
, &port
);
532 ovs_strlcpy(name
, port
.devname
, name_size
);
539 /* Obtains a list of all the ports in 'dpif'.
541 * If successful, returns 0 and sets '*portsp' to point to an array of
542 * appropriately initialized port structures and '*n_portsp' to the number of
543 * ports in the array. The caller is responsible for freeing '*portp' by
546 * On failure, returns a positive errno value and sets '*portsp' to NULL and
547 * '*n_portsp' to 0. */
549 dpif_port_list(const struct dpif
*dpif
,
550 struct odp_port
**portsp
, size_t *n_portsp
)
552 struct odp_port
*ports
;
557 struct odp_stats stats
;
560 error
= dpif_get_dp_stats(dpif
, &stats
);
565 ports
= xcalloc(stats
.n_ports
, sizeof *ports
);
566 retval
= dpif
->dpif_class
->port_list(dpif
, ports
, stats
.n_ports
);
572 } else if (retval
<= stats
.n_ports
) {
578 /* Soft error: port count increased behind our back. Try again. */
591 log_operation(dpif
, "port_list", error
);
595 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
596 * 'dpif' has changed, this function does one of the following:
598 * - Stores the name of the device that was added to or deleted from 'dpif' in
599 * '*devnamep' and returns 0. The caller is responsible for freeing
600 * '*devnamep' (with free()) when it no longer needs it.
602 * - Returns ENOBUFS and sets '*devnamep' to NULL.
604 * This function may also return 'false positives', where it returns 0 and
605 * '*devnamep' names a device that was not actually added or deleted or it
606 * returns ENOBUFS without any change.
608 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
609 * return other positive errno values to indicate that something has gone
612 dpif_port_poll(const struct dpif
*dpif
, char **devnamep
)
614 int error
= dpif
->dpif_class
->port_poll(dpif
, devnamep
);
621 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
622 * value other than EAGAIN. */
624 dpif_port_poll_wait(const struct dpif
*dpif
)
626 dpif
->dpif_class
->port_poll_wait(dpif
);
629 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
630 * positive errno value. */
632 dpif_flow_flush(struct dpif
*dpif
)
636 COVERAGE_INC(dpif_flow_flush
);
638 error
= dpif
->dpif_class
->flow_flush(dpif
);
639 log_operation(dpif
, "flow_flush", error
);
643 /* Queries 'dpif' for a flow entry matching 'flow->key'.
645 * If a flow matching 'flow->key' exists in 'dpif', stores statistics for the
646 * flow into 'flow->stats'. If 'flow->n_actions' is zero, then 'flow->actions'
647 * is ignored. If 'flow->n_actions' is nonzero, then 'flow->actions' should
648 * point to an array of the specified number of actions. At most that many of
649 * the flow's actions will be copied into that array. 'flow->n_actions' will
650 * be updated to the number of actions actually present in the flow, which may
651 * be greater than the number stored if the flow has more actions than space
652 * available in the array.
654 * If no flow matching 'flow->key' exists in 'dpif', returns ENOENT. On other
655 * failure, returns a positive errno value. */
657 dpif_flow_get(const struct dpif
*dpif
, struct odp_flow
*flow
)
661 COVERAGE_INC(dpif_flow_get
);
663 check_rw_odp_flow(flow
);
664 error
= dpif
->dpif_class
->flow_get(dpif
, flow
, 1);
666 error
= flow
->stats
.error
;
669 /* Make the results predictable on error. */
670 memset(&flow
->stats
, 0, sizeof flow
->stats
);
673 if (should_log_flow_message(error
)) {
674 log_flow_operation(dpif
, "flow_get", error
, flow
);
679 /* For each flow 'flow' in the 'n' flows in 'flows':
681 * - If a flow matching 'flow->key' exists in 'dpif':
683 * Stores 0 into 'flow->stats.error' and stores statistics for the flow
684 * into 'flow->stats'.
686 * If 'flow->n_actions' is zero, then 'flow->actions' is ignored. If
687 * 'flow->n_actions' is nonzero, then 'flow->actions' should point to an
688 * array of the specified number of actions. At most that many of the
689 * flow's actions will be copied into that array. 'flow->n_actions' will
690 * be updated to the number of actions actually present in the flow, which
691 * may be greater than the number stored if the flow has more actions than
692 * space available in the array.
694 * - Flow-specific errors are indicated by a positive errno value in
695 * 'flow->stats.error'. In particular, ENOENT indicates that no flow
696 * matching 'flow->key' exists in 'dpif'. When an error value is stored, the
697 * contents of 'flow->key' are preserved but other members of 'flow' should
698 * be treated as indeterminate.
700 * Returns 0 if all 'n' flows in 'flows' were updated (whether they were
701 * individually successful or not is indicated by 'flow->stats.error',
702 * however). Returns a positive errno value if an error that prevented this
703 * update occurred, in which the caller must not depend on any elements in
704 * 'flows' being updated or not updated.
707 dpif_flow_get_multiple(const struct dpif
*dpif
,
708 struct odp_flow flows
[], size_t n
)
713 COVERAGE_ADD(dpif_flow_get
, n
);
715 for (i
= 0; i
< n
; i
++) {
716 check_rw_odp_flow(&flows
[i
]);
719 error
= dpif
->dpif_class
->flow_get(dpif
, flows
, n
);
720 log_operation(dpif
, "flow_get_multiple", error
);
724 /* Adds or modifies a flow in 'dpif' as specified in 'put':
726 * - If the flow specified in 'put->flow' does not exist in 'dpif', then
727 * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': if
728 * it is, the flow will be added, otherwise the operation will fail with
731 * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'.
732 * Behavior in this case depends on whether ODPPF_MODIFY is specified in
733 * 'put->flags': if it is, the flow's actions will be updated, otherwise the
734 * operation will fail with EEXIST. If the flow's actions are updated, then
735 * its statistics will be zeroed if ODPPF_ZERO_STATS is set in 'put->flags',
736 * left as-is otherwise.
738 * Returns 0 if successful, otherwise a positive errno value.
741 dpif_flow_put(struct dpif
*dpif
, struct odp_flow_put
*put
)
745 COVERAGE_INC(dpif_flow_put
);
747 error
= dpif
->dpif_class
->flow_put(dpif
, put
);
748 if (should_log_flow_message(error
)) {
749 log_flow_put(dpif
, error
, put
);
754 /* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if 'dpif'
755 * does not contain such a flow.
757 * If successful, updates 'flow->stats', 'flow->n_actions', and 'flow->actions'
758 * as described for dpif_flow_get(). */
760 dpif_flow_del(struct dpif
*dpif
, struct odp_flow
*flow
)
764 COVERAGE_INC(dpif_flow_del
);
766 check_rw_odp_flow(flow
);
767 memset(&flow
->stats
, 0, sizeof flow
->stats
);
769 error
= dpif
->dpif_class
->flow_del(dpif
, flow
);
770 if (should_log_flow_message(error
)) {
771 log_flow_operation(dpif
, "delete flow", error
, flow
);
776 /* Stores up to 'n' flows in 'dpif' into 'flows', including their statistics
777 * but not including any information about their actions. If successful,
778 * returns 0 and sets '*n_out' to the number of flows actually present in
779 * 'dpif', which might be greater than the number stored (if 'dpif' has more
780 * than 'n' flows). On failure, returns a negative errno value and sets
783 dpif_flow_list(const struct dpif
*dpif
, struct odp_flow flows
[], size_t n
,
789 COVERAGE_INC(dpif_flow_query_list
);
790 if (RUNNING_ON_VALGRIND
) {
791 memset(flows
, 0, n
* sizeof *flows
);
793 for (i
= 0; i
< n
; i
++) {
794 flows
[i
].actions
= NULL
;
795 flows
[i
].n_actions
= 0;
798 retval
= dpif
->dpif_class
->flow_list(dpif
, flows
, n
);
801 VLOG_WARN_RL(&error_rl
, "%s: flow list failed (%s)",
802 dpif_name(dpif
), strerror(-retval
));
805 COVERAGE_ADD(dpif_flow_query_list_n
, retval
);
806 *n_out
= MIN(n
, retval
);
807 VLOG_DBG_RL(&dpmsg_rl
, "%s: listed %zu flows (of %d)",
808 dpif_name(dpif
), *n_out
, retval
);
813 /* Retrieves all of the flows in 'dpif'.
815 * If successful, returns 0 and stores in '*flowsp' a pointer to a newly
816 * allocated array of flows, including their statistics but not including any
817 * information about their actions, and sets '*np' to the number of flows in
818 * '*flowsp'. The caller is responsible for freeing '*flowsp' by calling
821 * On failure, returns a positive errno value and sets '*flowsp' to NULL and
824 dpif_flow_list_all(const struct dpif
*dpif
,
825 struct odp_flow
**flowsp
, size_t *np
)
827 struct odp_stats stats
;
828 struct odp_flow
*flows
;
835 error
= dpif_get_dp_stats(dpif
, &stats
);
840 flows
= xmalloc(sizeof *flows
* stats
.n_flows
);
841 error
= dpif_flow_list(dpif
, flows
, stats
.n_flows
, &n_flows
);
847 if (stats
.n_flows
!= n_flows
) {
848 VLOG_WARN_RL(&error_rl
, "%s: datapath stats reported %"PRIu32
" "
849 "flows but flow listing reported %zu",
850 dpif_name(dpif
), stats
.n_flows
, n_flows
);
857 /* Causes 'dpif' to perform the 'n_actions' actions in 'actions' on the
858 * Ethernet frame specified in 'packet'.
860 * Returns 0 if successful, otherwise a positive errno value. */
862 dpif_execute(struct dpif
*dpif
,
863 const union odp_action actions
[], size_t n_actions
,
864 const struct ofpbuf
*buf
)
868 COVERAGE_INC(dpif_execute
);
870 error
= dpif
->dpif_class
->execute(dpif
, actions
, n_actions
, buf
);
875 if (!(error
? VLOG_DROP_WARN(&error_rl
) : VLOG_DROP_DBG(&dpmsg_rl
))) {
876 struct ds ds
= DS_EMPTY_INITIALIZER
;
877 char *packet
= ofp_packet_to_string(buf
->data
, buf
->size
, buf
->size
);
878 ds_put_format(&ds
, "%s: execute ", dpif_name(dpif
));
879 format_odp_actions(&ds
, actions
, n_actions
);
881 ds_put_format(&ds
, " failed (%s)", strerror(error
));
883 ds_put_format(&ds
, " on packet %s", packet
);
884 vlog(THIS_MODULE
, error
? VLL_WARN
: VLL_DBG
, "%s", ds_cstr(&ds
));
891 /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set
892 * in '*listen_mask' indicates that dpif_recv() will receive messages of that
893 * type. Returns 0 if successful, otherwise a positive errno value. */
895 dpif_recv_get_mask(const struct dpif
*dpif
, int *listen_mask
)
897 int error
= dpif
->dpif_class
->recv_get_mask(dpif
, listen_mask
);
901 log_operation(dpif
, "recv_get_mask", error
);
905 /* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in
906 * '*listen_mask' requests that dpif_recv() receive messages of that type.
907 * Returns 0 if successful, otherwise a positive errno value. */
909 dpif_recv_set_mask(struct dpif
*dpif
, int listen_mask
)
911 int error
= dpif
->dpif_class
->recv_set_mask(dpif
, listen_mask
);
912 log_operation(dpif
, "recv_set_mask", error
);
916 /* Retrieve the sFlow sampling probability. '*probability' is expressed as the
917 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
918 * the probability of sampling a given packet.
920 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
921 * indicates that 'dpif' does not support sFlow sampling. */
923 dpif_get_sflow_probability(const struct dpif
*dpif
, uint32_t *probability
)
925 int error
= (dpif
->dpif_class
->get_sflow_probability
926 ? dpif
->dpif_class
->get_sflow_probability(dpif
, probability
)
931 log_operation(dpif
, "get_sflow_probability", error
);
935 /* Set the sFlow sampling probability. 'probability' is expressed as the
936 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
937 * the probability of sampling a given packet.
939 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
940 * indicates that 'dpif' does not support sFlow sampling. */
942 dpif_set_sflow_probability(struct dpif
*dpif
, uint32_t probability
)
944 int error
= (dpif
->dpif_class
->set_sflow_probability
945 ? dpif
->dpif_class
->set_sflow_probability(dpif
, probability
)
947 log_operation(dpif
, "set_sflow_probability", error
);
951 /* Attempts to receive a message from 'dpif'. If successful, stores the
952 * message into '*packetp'. The message, if one is received, will begin with
953 * 'struct odp_msg' as a header, and will have at least DPIF_RECV_MSG_PADDING
954 * bytes of headroom. Only messages of the types selected with
955 * dpif_set_listen_mask() will ordinarily be received (but if a message type is
956 * enabled and then later disabled, some stragglers might pop up).
958 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
959 * if no message is immediately available. */
961 dpif_recv(struct dpif
*dpif
, struct ofpbuf
**packetp
)
963 int error
= dpif
->dpif_class
->recv(dpif
, packetp
);
965 struct ofpbuf
*buf
= *packetp
;
967 assert(ofpbuf_headroom(buf
) >= DPIF_RECV_MSG_PADDING
);
968 if (VLOG_IS_DBG_ENABLED()) {
969 struct odp_msg
*msg
= buf
->data
;
970 void *payload
= msg
+ 1;
971 size_t payload_len
= buf
->size
- sizeof *msg
;
972 char *s
= ofp_packet_to_string(payload
, payload_len
, payload_len
);
973 VLOG_DBG_RL(&dpmsg_rl
, "%s: received %s message of length "
974 "%zu on port %"PRIu16
": %s", dpif_name(dpif
),
975 (msg
->type
== _ODPL_MISS_NR
? "miss"
976 : msg
->type
== _ODPL_ACTION_NR
? "action"
977 : msg
->type
== _ODPL_SFLOW_NR
? "sFlow"
979 payload_len
, msg
->port
, s
);
988 /* Discards all messages that would otherwise be received by dpif_recv() on
989 * 'dpif'. Returns 0 if successful, otherwise a positive errno value. */
991 dpif_recv_purge(struct dpif
*dpif
)
993 struct odp_stats stats
;
997 COVERAGE_INC(dpif_purge
);
999 error
= dpif_get_dp_stats(dpif
, &stats
);
1004 for (i
= 0; i
< stats
.max_miss_queue
+ stats
.max_action_queue
+ stats
.max_sflow_queue
; i
++) {
1006 error
= dpif_recv(dpif
, &buf
);
1008 return error
== EAGAIN
? 0 : error
;
1015 /* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
1016 * received with dpif_recv(). */
1018 dpif_recv_wait(struct dpif
*dpif
)
1020 dpif
->dpif_class
->recv_wait(dpif
);
1023 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1024 * and '*engine_id', respectively. */
1026 dpif_get_netflow_ids(const struct dpif
*dpif
,
1027 uint8_t *engine_type
, uint8_t *engine_id
)
1029 *engine_type
= dpif
->netflow_engine_type
;
1030 *engine_id
= dpif
->netflow_engine_id
;
1033 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1034 * value for use in the ODPAT_SET_PRIORITY action. On success, returns 0 and
1035 * stores the priority into '*priority'. On failure, returns a positive errno
1036 * value and stores 0 into '*priority'. */
1038 dpif_queue_to_priority(const struct dpif
*dpif
, uint32_t queue_id
,
1041 int error
= (dpif
->dpif_class
->queue_to_priority
1042 ? dpif
->dpif_class
->queue_to_priority(dpif
, queue_id
,
1048 log_operation(dpif
, "queue_to_priority", error
);
1053 dpif_init(struct dpif
*dpif
, const struct dpif_class
*dpif_class
,
1055 uint8_t netflow_engine_type
, uint8_t netflow_engine_id
)
1057 dpif
->dpif_class
= dpif_class
;
1058 dpif
->base_name
= xstrdup(name
);
1059 dpif
->full_name
= xasprintf("%s@%s", dpif_class
->type
, name
);
1060 dpif
->netflow_engine_type
= netflow_engine_type
;
1061 dpif
->netflow_engine_id
= netflow_engine_id
;
1064 /* Undoes the results of initialization.
1066 * Normally this function only needs to be called from dpif_close().
1067 * However, it may be called by providers due to an error on opening
1068 * that occurs after initialization. It this case dpif_close() would
1069 * never be called. */
1071 dpif_uninit(struct dpif
*dpif
, bool close
)
1073 char *base_name
= dpif
->base_name
;
1074 char *full_name
= dpif
->full_name
;
1077 dpif
->dpif_class
->close(dpif
);
1085 log_operation(const struct dpif
*dpif
, const char *operation
, int error
)
1088 VLOG_DBG_RL(&dpmsg_rl
, "%s: %s success", dpif_name(dpif
), operation
);
1089 } else if (is_errno(error
)) {
1090 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%s)",
1091 dpif_name(dpif
), operation
, strerror(error
));
1093 VLOG_WARN_RL(&error_rl
, "%s: %s failed (%d/%d)",
1094 dpif_name(dpif
), operation
,
1095 get_ofp_err_type(error
), get_ofp_err_code(error
));
1099 static enum vlog_level
1100 flow_message_log_level(int error
)
1102 return error
? VLL_WARN
: VLL_DBG
;
1106 should_log_flow_message(int error
)
1108 return !vlog_should_drop(THIS_MODULE
, flow_message_log_level(error
),
1109 error
? &error_rl
: &dpmsg_rl
);
1113 log_flow_message(const struct dpif
*dpif
, int error
, const char *operation
,
1114 const struct odp_flow_key
*flow
,
1115 const struct odp_flow_stats
*stats
,
1116 const union odp_action
*actions
, size_t n_actions
)
1118 struct ds ds
= DS_EMPTY_INITIALIZER
;
1119 ds_put_format(&ds
, "%s: ", dpif_name(dpif
));
1121 ds_put_cstr(&ds
, "failed to ");
1123 ds_put_format(&ds
, "%s ", operation
);
1125 ds_put_format(&ds
, "(%s) ", strerror(error
));
1127 format_odp_flow_key(&ds
, flow
);
1129 ds_put_cstr(&ds
, ", ");
1130 format_odp_flow_stats(&ds
, stats
);
1132 if (actions
|| n_actions
) {
1133 ds_put_cstr(&ds
, ", actions:");
1134 format_odp_actions(&ds
, actions
, n_actions
);
1136 vlog(THIS_MODULE
, flow_message_log_level(error
), "%s", ds_cstr(&ds
));
1141 log_flow_operation(const struct dpif
*dpif
, const char *operation
, int error
,
1142 struct odp_flow
*flow
)
1145 flow
->n_actions
= 0;
1147 log_flow_message(dpif
, error
, operation
, &flow
->key
,
1148 !error
? &flow
->stats
: NULL
,
1149 flow
->actions
, flow
->n_actions
);
1153 log_flow_put(struct dpif
*dpif
, int error
, const struct odp_flow_put
*put
)
1155 enum { ODPPF_ALL
= ODPPF_CREATE
| ODPPF_MODIFY
| ODPPF_ZERO_STATS
};
1159 ds_put_cstr(&s
, "put");
1160 if (put
->flags
& ODPPF_CREATE
) {
1161 ds_put_cstr(&s
, "[create]");
1163 if (put
->flags
& ODPPF_MODIFY
) {
1164 ds_put_cstr(&s
, "[modify]");
1166 if (put
->flags
& ODPPF_ZERO_STATS
) {
1167 ds_put_cstr(&s
, "[zero]");
1169 if (put
->flags
& ~ODPPF_ALL
) {
1170 ds_put_format(&s
, "[%x]", put
->flags
& ~ODPPF_ALL
);
1172 log_flow_message(dpif
, error
, ds_cstr(&s
), &put
->flow
.key
,
1173 !error
? &put
->flow
.stats
: NULL
,
1174 put
->flow
.actions
, put
->flow
.n_actions
);
1178 /* There is a tendency to construct odp_flow objects on the stack and to
1179 * forget to properly initialize their "actions" and "n_actions" members.
1180 * When this happens, we get memory corruption because the kernel
1181 * writes through the random pointer that is in the "actions" member.
1183 * This function attempts to combat the problem by:
1185 * - Forcing a segfault if "actions" points to an invalid region (instead
1186 * of just getting back EFAULT, which can be easily missed in the log).
1188 * - Storing a distinctive value that is likely to cause an
1189 * easy-to-identify error later if it is dereferenced, etc.
1191 * - Triggering a warning on uninitialized memory from Valgrind if
1192 * "actions" or "n_actions" was not initialized.
1195 check_rw_odp_flow(struct odp_flow
*flow
)
1197 if (flow
->n_actions
) {
1198 memset(&flow
->actions
[0], 0xcc, sizeof flow
->actions
[0]);