2 * Copyright (c) 2008, 2009, 2010, 2011 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"
33 #include "ofp-print.h"
37 #include "poll-loop.h"
44 VLOG_DEFINE_THIS_MODULE(dpif);
46 COVERAGE_DEFINE(dpif_destroy);
47 COVERAGE_DEFINE(dpif_port_add);
48 COVERAGE_DEFINE(dpif_port_del);
49 COVERAGE_DEFINE(dpif_flow_flush);
50 COVERAGE_DEFINE(dpif_flow_get);
51 COVERAGE_DEFINE(dpif_flow_put);
52 COVERAGE_DEFINE(dpif_flow_del);
53 COVERAGE_DEFINE(dpif_flow_query_list);
54 COVERAGE_DEFINE(dpif_flow_query_list_n);
55 COVERAGE_DEFINE(dpif_execute);
56 COVERAGE_DEFINE(dpif_purge);
58 static const struct dpif_class *base_dpif_classes[] = {
65 struct registered_dpif_class {
66 const struct dpif_class *dpif_class;
69 static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes);
71 /* Rate limit for individual messages going to or from the datapath, output at
72 * DBG level. This is very high because, if these are enabled, it is because
73 * we really need to see them. */
74 static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
76 /* Not really much point in logging many dpif errors. */
77 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
79 static void log_operation(const struct dpif *, const char *operation,
81 static void log_flow_operation(const struct dpif *, const char *operation,
82 int error, struct odp_flow *flow);
83 static void log_flow_put(struct dpif *, int error,
84 const struct odp_flow_put *);
85 static bool should_log_flow_message(int error);
86 static void check_rw_flow_actions(struct odp_flow *);
87 static void check_rw_flow_key(struct odp_flow *);
92 static int status = -1;
98 for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
99 dp_register_provider(base_dpif_classes[i]);
104 /* Performs periodic work needed by all the various kinds of dpifs.
106 * If your program opens any dpifs, it must call both this function and
107 * netdev_run() within its main poll loop. */
111 struct shash_node *node;
112 SHASH_FOR_EACH(node, &dpif_classes) {
113 const struct registered_dpif_class *registered_class = node->data;
114 if (registered_class->dpif_class->run) {
115 registered_class->dpif_class->run();
120 /* Arranges for poll_block() to wake up when dp_run() needs to be called.
122 * If your program opens any dpifs, it must call both this function and
123 * netdev_wait() within its main poll loop. */
127 struct shash_node *node;
128 SHASH_FOR_EACH(node, &dpif_classes) {
129 const struct registered_dpif_class *registered_class = node->data;
130 if (registered_class->dpif_class->wait) {
131 registered_class->dpif_class->wait();
136 /* Registers a new datapath provider. After successful registration, new
137 * datapaths of that type can be opened using dpif_open(). */
139 dp_register_provider(const struct dpif_class *new_class)
141 struct registered_dpif_class *registered_class;
143 if (shash_find(&dpif_classes, new_class->type)) {
144 VLOG_WARN("attempted to register duplicate datapath provider: %s",
149 registered_class = xmalloc(sizeof *registered_class);
150 registered_class->dpif_class = new_class;
151 registered_class->refcount = 0;
153 shash_add(&dpif_classes, new_class->type, registered_class);
158 /* Unregisters a datapath provider. 'type' must have been previously
159 * registered and not currently be in use by any dpifs. After unregistration
160 * new datapaths of that type cannot be opened using dpif_open(). */
162 dp_unregister_provider(const char *type)
164 struct shash_node *node;
165 struct registered_dpif_class *registered_class;
167 node = shash_find(&dpif_classes, type);
169 VLOG_WARN("attempted to unregister a datapath provider that is not "
170 "registered: %s", type);
174 registered_class = node->data;
175 if (registered_class->refcount) {
176 VLOG_WARN("attempted to unregister in use datapath provider: %s", type);
180 shash_delete(&dpif_classes, node);
181 free(registered_class);
186 /* Clears 'types' and enumerates the types of all currently registered datapath
187 * providers into it. The caller must first initialize the svec. */
189 dp_enumerate_types(struct svec *types)
191 struct shash_node *node;
196 SHASH_FOR_EACH(node, &dpif_classes) {
197 const struct registered_dpif_class *registered_class = node->data;
198 svec_add(types, registered_class->dpif_class->type);
202 /* Clears 'names' and enumerates the names of all known created datapaths with
203 * the given 'type'. The caller must first initialize the svec. Returns 0 if
204 * successful, otherwise a positive errno value.
206 * Some kinds of datapaths might not be practically enumerable. This is not
207 * considered an error. */
209 dp_enumerate_names(const char *type, struct svec *names)
211 const struct registered_dpif_class *registered_class;
212 const struct dpif_class *dpif_class;
218 registered_class = shash_find_data(&dpif_classes, type);
219 if (!registered_class) {
220 VLOG_WARN("could not enumerate unknown type: %s", type);
224 dpif_class = registered_class->dpif_class;
225 error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
228 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
235 /* Parses 'datapath name', which is of the form type@name into its
236 * component pieces. 'name' and 'type' must be freed by the caller. */
238 dp_parse_name(const char *datapath_name_, char **name, char **type)
240 char *datapath_name = xstrdup(datapath_name_);
243 separator = strchr(datapath_name, '@');
246 *type = datapath_name;
247 *name = xstrdup(separator + 1);
249 *name = datapath_name;
255 do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
257 struct dpif *dpif = NULL;
259 struct registered_dpif_class *registered_class;
263 if (!type || *type == '\0') {
267 registered_class = shash_find_data(&dpif_classes, type);
268 if (!registered_class) {
269 VLOG_WARN("could not create datapath %s of unknown type %s", name,
271 error = EAFNOSUPPORT;
275 error = registered_class->dpif_class->open(registered_class->dpif_class,
276 name, create, &dpif);
278 assert(dpif->dpif_class == registered_class->dpif_class);
279 registered_class->refcount++;
283 *dpifp = error ? NULL : dpif;
287 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
288 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
289 * the empty string to specify the default system type. Returns 0 if
290 * successful, otherwise a positive errno value. On success stores a pointer
291 * to the datapath in '*dpifp', otherwise a null pointer. */
293 dpif_open(const char *name, const char *type, struct dpif **dpifp)
295 return do_open(name, type, false, dpifp);
298 /* Tries to create and open a new datapath with the given 'name' and 'type'.
299 * 'type' may be either NULL or the empty string to specify the default system
300 * type. Will fail if a datapath with 'name' and 'type' already exists.
301 * Returns 0 if successful, otherwise a positive errno value. On success
302 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
304 dpif_create(const char *name, const char *type, struct dpif **dpifp)
306 return do_open(name, type, true, dpifp);
309 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
310 * does not exist. 'type' may be either NULL or the empty string to specify
311 * the default system type. Returns 0 if successful, otherwise a positive
312 * errno value. On success stores a pointer to the datapath in '*dpifp',
313 * otherwise a null pointer. */
315 dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
319 error = dpif_create(name, type, dpifp);
320 if (error == EEXIST || error == EBUSY) {
321 error = dpif_open(name, type, dpifp);
323 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
324 name, strerror(error));
327 VLOG_WARN("failed to create datapath %s: %s", name, strerror(error));
332 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
333 * itself; call dpif_delete() first, instead, if that is desirable. */
335 dpif_close(struct dpif *dpif)
338 struct registered_dpif_class *registered_class;
340 registered_class = shash_find_data(&dpif_classes,
341 dpif->dpif_class->type);
342 assert(registered_class);
343 assert(registered_class->refcount);
345 registered_class->refcount--;
346 dpif_uninit(dpif, true);
350 /* Returns the name of datapath 'dpif' prefixed with the type
351 * (for use in log messages). */
353 dpif_name(const struct dpif *dpif)
355 return dpif->full_name;
358 /* Returns the name of datapath 'dpif' without the type
359 * (for use in device names). */
361 dpif_base_name(const struct dpif *dpif)
363 return dpif->base_name;
366 /* Enumerates all names that may be used to open 'dpif' into 'all_names'. The
367 * Linux datapath, for example, supports opening a datapath both by number,
368 * e.g. "dp0", and by the name of the datapath's local port. For some
369 * datapaths, this might be an infinite set (e.g. in a file name, slashes may
370 * be duplicated any number of times), in which case only the names most likely
371 * to be used will be enumerated.
373 * The caller must already have initialized 'all_names'. Any existing names in
374 * 'all_names' will not be disturbed. */
376 dpif_get_all_names(const struct dpif *dpif, struct svec *all_names)
378 if (dpif->dpif_class->get_all_names) {
379 int error = dpif->dpif_class->get_all_names(dpif, all_names);
381 VLOG_WARN_RL(&error_rl,
382 "failed to retrieve names for datpath %s: %s",
383 dpif_name(dpif), strerror(error));
387 svec_add(all_names, dpif_base_name(dpif));
393 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
394 * ports. After calling this function, it does not make sense to pass 'dpif'
395 * to any functions other than dpif_name() or dpif_close(). */
397 dpif_delete(struct dpif *dpif)
401 COVERAGE_INC(dpif_destroy);
403 error = dpif->dpif_class->destroy(dpif);
404 log_operation(dpif, "delete", error);
408 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
409 * otherwise a positive errno value. */
411 dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats)
413 int error = dpif->dpif_class->get_stats(dpif, stats);
415 memset(stats, 0, sizeof *stats);
417 log_operation(dpif, "get_stats", error);
421 /* Retrieves the current IP fragment handling policy for 'dpif' into
422 * '*drop_frags': true indicates that fragments are dropped, false indicates
423 * that fragments are treated in the same way as other IP packets (except that
424 * the L4 header cannot be read). Returns 0 if successful, otherwise a
425 * positive errno value. */
427 dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags)
429 int error = dpif->dpif_class->get_drop_frags(dpif, drop_frags);
433 log_operation(dpif, "get_drop_frags", error);
437 /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is
438 * the same as for the get_drop_frags member function. Returns 0 if
439 * successful, otherwise a positive errno value. */
441 dpif_set_drop_frags(struct dpif *dpif, bool drop_frags)
443 int error = dpif->dpif_class->set_drop_frags(dpif, drop_frags);
444 log_operation(dpif, "set_drop_frags", error);
448 /* Attempts to add 'netdev' as a port on 'dpif'. If successful, returns 0 and
449 * sets '*port_nop' to the new port's port number (if 'port_nop' is non-null).
450 * On failure, returns a positive errno value and sets '*port_nop' to
451 * UINT16_MAX (if 'port_nop' is non-null). */
453 dpif_port_add(struct dpif *dpif, struct netdev *netdev, uint16_t *port_nop)
455 const char *netdev_name = netdev_get_name(netdev);
459 COVERAGE_INC(dpif_port_add);
461 error = dpif->dpif_class->port_add(dpif, netdev, &port_no);
463 VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16,
464 dpif_name(dpif), netdev_name, port_no);
466 VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
467 dpif_name(dpif), netdev_name, strerror(error));
468 port_no = UINT16_MAX;
476 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
477 * otherwise a positive errno value. */
479 dpif_port_del(struct dpif *dpif, uint16_t port_no)
483 COVERAGE_INC(dpif_port_del);
485 error = dpif->dpif_class->port_del(dpif, port_no);
487 VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu16")",
488 dpif_name(dpif), port_no);
490 log_operation(dpif, "port_del", error);
495 /* Makes a deep copy of 'src' into 'dst'. */
497 dpif_port_clone(struct dpif_port *dst, const struct dpif_port *src)
499 dst->name = xstrdup(src->name);
500 dst->type = xstrdup(src->type);
501 dst->port_no = src->port_no;
504 /* Frees memory allocated to members of 'dpif_port'.
506 * Do not call this function on a dpif_port obtained from
507 * dpif_port_dump_next(): that function retains ownership of the data in the
510 dpif_port_destroy(struct dpif_port *dpif_port)
512 free(dpif_port->name);
513 free(dpif_port->type);
516 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
517 * initializes '*port' appropriately; on failure, returns a positive errno
520 * The caller owns the data in 'port' and must free it with
521 * dpif_port_destroy() when it is no longer needed. */
523 dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
524 struct dpif_port *port)
526 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
528 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s",
529 dpif_name(dpif), port_no, port->name);
531 memset(port, 0, sizeof *port);
532 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s",
533 dpif_name(dpif), port_no, strerror(error));
538 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
539 * initializes '*port' appropriately; on failure, returns a positive errno
542 * The caller owns the data in 'port' and must free it with
543 * dpif_port_destroy() when it is no longer needed. */
545 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
546 struct dpif_port *port)
548 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
550 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16,
551 dpif_name(dpif), devname, port->port_no);
553 memset(port, 0, sizeof *port);
555 /* Log level is DBG here because all the current callers are interested
556 * in whether 'dpif' actually has a port 'devname', so that it's not an
557 * issue worth logging if it doesn't. */
558 VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s",
559 dpif_name(dpif), devname, strerror(error));
564 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
565 * the port's name into the 'name_size' bytes in 'name', ensuring that the
566 * result is null-terminated. On failure, returns a positive errno value and
567 * makes 'name' the empty string. */
569 dpif_port_get_name(struct dpif *dpif, uint16_t port_no,
570 char *name, size_t name_size)
572 struct dpif_port port;
575 assert(name_size > 0);
577 error = dpif_port_query_by_number(dpif, port_no, &port);
579 ovs_strlcpy(name, port.name, name_size);
580 dpif_port_destroy(&port);
587 /* Initializes 'dump' to begin dumping the ports in a dpif.
589 * This function provides no status indication. An error status for the entire
590 * dump operation is provided when it is completed by calling
591 * dpif_port_dump_done().
594 dpif_port_dump_start(struct dpif_port_dump *dump, const struct dpif *dpif)
597 dump->error = dpif->dpif_class->port_dump_start(dpif, &dump->state);
598 log_operation(dpif, "port_dump_start", dump->error);
601 /* Attempts to retrieve another port from 'dump', which must have been
602 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
603 * into 'port' and returns true. On failure, returns false.
605 * Failure might indicate an actual error or merely that the last port has been
606 * dumped. An error status for the entire dump operation is provided when it
607 * is completed by calling dpif_port_dump_done().
609 * The dpif owns the data stored in 'port'. It will remain valid until at
610 * least the next time 'dump' is passed to dpif_port_dump_next() or
611 * dpif_port_dump_done(). */
613 dpif_port_dump_next(struct dpif_port_dump *dump, struct dpif_port *port)
615 const struct dpif *dpif = dump->dpif;
621 dump->error = dpif->dpif_class->port_dump_next(dpif, dump->state, port);
622 if (dump->error == EOF) {
623 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all ports", dpif_name(dpif));
625 log_operation(dpif, "port_dump_next", dump->error);
629 dpif->dpif_class->port_dump_done(dpif, dump->state);
635 /* Completes port table dump operation 'dump', which must have been initialized
636 * with dpif_port_dump_start(). Returns 0 if the dump operation was
637 * error-free, otherwise a positive errno value describing the problem. */
639 dpif_port_dump_done(struct dpif_port_dump *dump)
641 const struct dpif *dpif = dump->dpif;
643 dump->error = dpif->dpif_class->port_dump_done(dpif, dump->state);
644 log_operation(dpif, "port_dump_done", dump->error);
646 return dump->error == EOF ? 0 : dump->error;
649 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
650 * 'dpif' has changed, this function does one of the following:
652 * - Stores the name of the device that was added to or deleted from 'dpif' in
653 * '*devnamep' and returns 0. The caller is responsible for freeing
654 * '*devnamep' (with free()) when it no longer needs it.
656 * - Returns ENOBUFS and sets '*devnamep' to NULL.
658 * This function may also return 'false positives', where it returns 0 and
659 * '*devnamep' names a device that was not actually added or deleted or it
660 * returns ENOBUFS without any change.
662 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
663 * return other positive errno values to indicate that something has gone
666 dpif_port_poll(const struct dpif *dpif, char **devnamep)
668 int error = dpif->dpif_class->port_poll(dpif, devnamep);
675 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
676 * value other than EAGAIN. */
678 dpif_port_poll_wait(const struct dpif *dpif)
680 dpif->dpif_class->port_poll_wait(dpif);
683 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
684 * positive errno value. */
686 dpif_flow_flush(struct dpif *dpif)
690 COVERAGE_INC(dpif_flow_flush);
692 error = dpif->dpif_class->flow_flush(dpif);
693 log_operation(dpif, "flow_flush", error);
697 /* Queries 'dpif' for a flow entry matching 'flow->key'.
699 * If a flow matching 'flow->key' exists in 'dpif', stores statistics for the
700 * flow into 'flow->stats'. If 'flow->actions_len' is zero, then
701 * 'flow->actions' is ignored. If 'flow->actions_len' is nonzero, then
702 * 'flow->actions' should point to an array of the specified number of bytes.
703 * At most that many bytes of the flow's actions will be copied into that
704 * array. 'flow->actions_len' will be updated to the number of bytes of
705 * actions actually present in the flow, which may be greater than the amount
706 * stored if the flow has more actions than space available in the array.
708 * If no flow matching 'flow->key' exists in 'dpif', returns ENOENT. On other
709 * failure, returns a positive errno value. */
711 dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
715 COVERAGE_INC(dpif_flow_get);
717 check_rw_flow_actions(flow);
718 error = dpif->dpif_class->flow_get(dpif, flow, 1);
720 error = flow->stats.error;
723 /* Make the results predictable on error. */
724 memset(&flow->stats, 0, sizeof flow->stats);
725 flow->actions_len = 0;
727 if (should_log_flow_message(error)) {
728 log_flow_operation(dpif, "flow_get", error, flow);
733 /* For each flow 'flow' in the 'n' flows in 'flows':
735 * - If a flow matching 'flow->key' exists in 'dpif':
737 * Stores 0 into 'flow->stats.error' and stores statistics for the flow
738 * into 'flow->stats'.
740 * If 'flow->actions_len' is zero, then 'flow->actions' is ignored. If
741 * 'flow->actions_len' is nonzero, then 'flow->actions' should point to an
742 * array of the specified number of bytes. At most that amount of flow's
743 * actions will be copied into that array. 'flow->actions_len' will be
744 * updated to the number of bytes of actions actually present in the flow,
745 * which may be greater than the amount stored if the flow's actions are
746 * longer than the available space.
748 * - Flow-specific errors are indicated by a positive errno value in
749 * 'flow->stats.error'. In particular, ENOENT indicates that no flow
750 * matching 'flow->key' exists in 'dpif'. When an error value is stored, the
751 * contents of 'flow->key' are preserved but other members of 'flow' should
752 * be treated as indeterminate.
754 * Returns 0 if all 'n' flows in 'flows' were updated (whether they were
755 * individually successful or not is indicated by 'flow->stats.error',
756 * however). Returns a positive errno value if an error that prevented this
757 * update occurred, in which the caller must not depend on any elements in
758 * 'flows' being updated or not updated.
761 dpif_flow_get_multiple(const struct dpif *dpif,
762 struct odp_flow flows[], size_t n)
767 COVERAGE_ADD(dpif_flow_get, n);
769 for (i = 0; i < n; i++) {
770 check_rw_flow_actions(&flows[i]);
773 error = dpif->dpif_class->flow_get(dpif, flows, n);
774 log_operation(dpif, "flow_get_multiple", error);
778 /* Adds or modifies a flow in 'dpif' as specified in 'put':
780 * - If the flow specified in 'put->flow' does not exist in 'dpif', then
781 * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': if
782 * it is, the flow will be added, otherwise the operation will fail with
785 * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'.
786 * Behavior in this case depends on whether ODPPF_MODIFY is specified in
787 * 'put->flags': if it is, the flow's actions will be updated, otherwise the
788 * operation will fail with EEXIST. If the flow's actions are updated, then
789 * its statistics will be zeroed if ODPPF_ZERO_STATS is set in 'put->flags',
790 * left as-is otherwise.
792 * Returns 0 if successful, otherwise a positive errno value.
795 dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
799 COVERAGE_INC(dpif_flow_put);
801 error = dpif->dpif_class->flow_put(dpif, put);
802 if (should_log_flow_message(error)) {
803 log_flow_put(dpif, error, put);
808 /* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if 'dpif'
809 * does not contain such a flow.
811 * If successful, updates 'flow->stats', 'flow->actions_len', and
812 * 'flow->actions' as described for dpif_flow_get(). */
814 dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
818 COVERAGE_INC(dpif_flow_del);
820 check_rw_flow_actions(flow);
821 memset(&flow->stats, 0, sizeof flow->stats);
823 error = dpif->dpif_class->flow_del(dpif, flow);
824 if (should_log_flow_message(error)) {
825 log_flow_operation(dpif, "delete flow", error, flow);
830 /* Initializes 'dump' to begin dumping the flows in a dpif.
832 * This function provides no status indication. An error status for the entire
833 * dump operation is provided when it is completed by calling
834 * dpif_flow_dump_done().
837 dpif_flow_dump_start(struct dpif_flow_dump *dump, const struct dpif *dpif)
840 dump->error = dpif->dpif_class->flow_dump_start(dpif, &dump->state);
841 log_operation(dpif, "flow_dump_start", dump->error);
844 /* Attempts to retrieve another flow from 'dump', which must have been
845 * initialized with dpif_flow_dump_start(). On success, stores a new odp_flow
846 * into 'flow' and returns true. Failure might indicate an actual error or
847 * merely the end of the flow table. An error status for the entire dump
848 * operation is provided when it is completed by calling dpif_flow_dump_done().
850 * Dumping flow actions is optional. To avoid dumping actions initialize
851 * 'flow->actions' to NULL and 'flow->actions_len' to 0. Otherwise, point
852 * 'flow->actions' to an array of struct nlattr and initialize
853 * 'flow->actions_len' with the number of bytes of Netlink attributes.
854 * dpif_flow_dump_next() will fill in as many actions as will fit into the
855 * provided array and update 'flow->actions_len' with the number of bytes
856 * required (regardless of whether they fit in the provided space). */
858 dpif_flow_dump_next(struct dpif_flow_dump *dump, struct odp_flow *flow)
860 const struct dpif *dpif = dump->dpif;
862 check_rw_flow_actions(flow);
863 check_rw_flow_key(flow);
869 dump->error = dpif->dpif_class->flow_dump_next(dpif, dump->state, flow);
870 if (dump->error == EOF) {
871 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif));
876 if (should_log_flow_message(dump->error)) {
877 log_flow_operation(dpif, "flow_dump_next", dump->error, flow);
882 dpif->dpif_class->flow_dump_done(dpif, dump->state);
888 /* Completes flow table dump operation 'dump', which must have been initialized
889 * with dpif_flow_dump_start(). Returns 0 if the dump operation was
890 * error-free, otherwise a positive errno value describing the problem. */
892 dpif_flow_dump_done(struct dpif_flow_dump *dump)
894 const struct dpif *dpif = dump->dpif;
896 dump->error = dpif->dpif_class->flow_dump_done(dpif, dump->state);
897 log_operation(dpif, "flow_dump_done", dump->error);
899 return dump->error == EOF ? 0 : dump->error;
902 /* Causes 'dpif' to perform the 'actions_len' bytes of actions in 'actions' on
903 * the Ethernet frame specified in 'packet'.
905 * Returns 0 if successful, otherwise a positive errno value. */
907 dpif_execute(struct dpif *dpif,
908 const struct nlattr *actions, size_t actions_len,
909 const struct ofpbuf *buf)
913 COVERAGE_INC(dpif_execute);
914 if (actions_len > 0) {
915 error = dpif->dpif_class->execute(dpif, actions, actions_len, buf);
920 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
921 struct ds ds = DS_EMPTY_INITIALIZER;
922 char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
923 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
924 format_odp_actions(&ds, actions, actions_len);
926 ds_put_format(&ds, " failed (%s)", strerror(error));
928 ds_put_format(&ds, " on packet %s", packet);
929 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
936 /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set
937 * in '*listen_mask' indicates that dpif_recv() will receive messages of that
938 * type. Returns 0 if successful, otherwise a positive errno value. */
940 dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask)
942 int error = dpif->dpif_class->recv_get_mask(dpif, listen_mask);
946 log_operation(dpif, "recv_get_mask", error);
950 /* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in
951 * '*listen_mask' requests that dpif_recv() receive messages of that type.
952 * Returns 0 if successful, otherwise a positive errno value. */
954 dpif_recv_set_mask(struct dpif *dpif, int listen_mask)
956 int error = dpif->dpif_class->recv_set_mask(dpif, listen_mask);
957 log_operation(dpif, "recv_set_mask", error);
961 /* Retrieve the sFlow sampling probability. '*probability' is expressed as the
962 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
963 * the probability of sampling a given packet.
965 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
966 * indicates that 'dpif' does not support sFlow sampling. */
968 dpif_get_sflow_probability(const struct dpif *dpif, uint32_t *probability)
970 int error = (dpif->dpif_class->get_sflow_probability
971 ? dpif->dpif_class->get_sflow_probability(dpif, probability)
976 log_operation(dpif, "get_sflow_probability", error);
980 /* Set the sFlow sampling probability. 'probability' is expressed as the
981 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
982 * the probability of sampling a given packet.
984 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
985 * indicates that 'dpif' does not support sFlow sampling. */
987 dpif_set_sflow_probability(struct dpif *dpif, uint32_t probability)
989 int error = (dpif->dpif_class->set_sflow_probability
990 ? dpif->dpif_class->set_sflow_probability(dpif, probability)
992 log_operation(dpif, "set_sflow_probability", error);
996 /* Polls for an upcall from 'dpif'. If successful, stores the upcall into
997 * '*upcall'. Only upcalls of the types selected with the set_listen_mask
998 * member function will ordinarily be received (but if a message type is
999 * enabled and then later disabled, some stragglers might pop up).
1001 * The caller takes ownership of the data that 'upcall' points to.
1002 * 'upcall->key' and 'upcall->actions' (if nonnull) point into data owned by
1003 * 'upcall->packet', so their memory cannot be freed separately. (This is
1004 * hardly a great way to do things but it works out OK for the dpif providers
1005 * and clients that exist so far.)
1007 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1008 * if no upcall is immediately available. */
1010 dpif_recv(struct dpif *dpif, struct dpif_upcall *upcall)
1012 int error = dpif->dpif_class->recv(dpif, upcall);
1013 if (!error && !VLOG_DROP_DBG(&dpmsg_rl)) {
1017 s = ofp_packet_to_string(upcall->packet->data,
1018 upcall->packet->size, upcall->packet->size);
1019 odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow);
1021 VLOG_DBG("%s: %s upcall on port %"PRIu16": %s", dpif_name(dpif),
1022 (upcall->type == _ODPL_MISS_NR ? "miss"
1023 : upcall->type == _ODPL_ACTION_NR ? "action"
1024 : upcall->type == _ODPL_SFLOW_NR ? "sFlow"
1032 /* Discards all messages that would otherwise be received by dpif_recv() on
1035 dpif_recv_purge(struct dpif *dpif)
1037 COVERAGE_INC(dpif_purge);
1038 if (dpif->dpif_class->recv_purge) {
1039 dpif->dpif_class->recv_purge(dpif);
1043 /* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
1044 * received with dpif_recv(). */
1046 dpif_recv_wait(struct dpif *dpif)
1048 dpif->dpif_class->recv_wait(dpif);
1051 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1052 * and '*engine_id', respectively. */
1054 dpif_get_netflow_ids(const struct dpif *dpif,
1055 uint8_t *engine_type, uint8_t *engine_id)
1057 *engine_type = dpif->netflow_engine_type;
1058 *engine_id = dpif->netflow_engine_id;
1061 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1062 * value for use in the ODPAT_SET_PRIORITY action. On success, returns 0 and
1063 * stores the priority into '*priority'. On failure, returns a positive errno
1064 * value and stores 0 into '*priority'. */
1066 dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1069 int error = (dpif->dpif_class->queue_to_priority
1070 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1076 log_operation(dpif, "queue_to_priority", error);
1081 dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1083 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1085 dpif->dpif_class = dpif_class;
1086 dpif->base_name = xstrdup(name);
1087 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
1088 dpif->netflow_engine_type = netflow_engine_type;
1089 dpif->netflow_engine_id = netflow_engine_id;
1092 /* Undoes the results of initialization.
1094 * Normally this function only needs to be called from dpif_close().
1095 * However, it may be called by providers due to an error on opening
1096 * that occurs after initialization. It this case dpif_close() would
1097 * never be called. */
1099 dpif_uninit(struct dpif *dpif, bool close)
1101 char *base_name = dpif->base_name;
1102 char *full_name = dpif->full_name;
1105 dpif->dpif_class->close(dpif);
1113 log_operation(const struct dpif *dpif, const char *operation, int error)
1116 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
1117 } else if (is_errno(error)) {
1118 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1119 dpif_name(dpif), operation, strerror(error));
1121 VLOG_WARN_RL(&error_rl, "%s: %s failed (%d/%d)",
1122 dpif_name(dpif), operation,
1123 get_ofp_err_type(error), get_ofp_err_code(error));
1127 static enum vlog_level
1128 flow_message_log_level(int error)
1130 return error ? VLL_WARN : VLL_DBG;
1134 should_log_flow_message(int error)
1136 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1137 error ? &error_rl : &dpmsg_rl);
1141 log_flow_message(const struct dpif *dpif, int error, const char *operation,
1142 const struct nlattr *key, size_t key_len,
1143 const struct odp_flow_stats *stats,
1144 const struct nlattr *actions, size_t actions_len)
1146 struct ds ds = DS_EMPTY_INITIALIZER;
1147 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1149 ds_put_cstr(&ds, "failed to ");
1151 ds_put_format(&ds, "%s ", operation);
1153 ds_put_format(&ds, "(%s) ", strerror(error));
1155 odp_flow_key_format(key, key_len, &ds);
1157 ds_put_cstr(&ds, ", ");
1158 format_odp_flow_stats(&ds, stats);
1160 if (actions || actions_len) {
1161 ds_put_cstr(&ds, ", actions:");
1162 format_odp_actions(&ds, actions, actions_len);
1164 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1169 log_flow_operation(const struct dpif *dpif, const char *operation, int error,
1170 struct odp_flow *flow)
1174 flow->actions_len = 0;
1176 log_flow_message(dpif, error, operation,
1177 flow->key, flow->key_len, !error ? &flow->stats : NULL,
1178 flow->actions, flow->actions_len);
1182 log_flow_put(struct dpif *dpif, int error, const struct odp_flow_put *put)
1184 enum { ODPPF_ALL = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS };
1188 ds_put_cstr(&s, "put");
1189 if (put->flags & ODPPF_CREATE) {
1190 ds_put_cstr(&s, "[create]");
1192 if (put->flags & ODPPF_MODIFY) {
1193 ds_put_cstr(&s, "[modify]");
1195 if (put->flags & ODPPF_ZERO_STATS) {
1196 ds_put_cstr(&s, "[zero]");
1198 if (put->flags & ~ODPPF_ALL) {
1199 ds_put_format(&s, "[%x]", put->flags & ~ODPPF_ALL);
1201 log_flow_message(dpif, error, ds_cstr(&s),
1202 put->flow.key, put->flow.key_len,
1203 !error ? &put->flow.stats : NULL,
1204 put->flow.actions, put->flow.actions_len);
1208 /* There is a tendency to construct odp_flow objects on the stack and to
1209 * forget to properly initialize their "actions" and "actions_len" members.
1210 * When this happens, we get memory corruption because the kernel
1211 * writes through the random pointer that is in the "actions" member.
1213 * This function attempts to combat the problem by:
1215 * - Forcing a segfault if "actions" points to an invalid region (instead
1216 * of just getting back EFAULT, which can be easily missed in the log).
1218 * - Storing a distinctive value that is likely to cause an
1219 * easy-to-identify error later if it is dereferenced, etc.
1221 * - Triggering a warning on uninitialized memory from Valgrind if
1222 * "actions" or "actions_len" was not initialized.
1225 check_rw_flow_actions(struct odp_flow *flow)
1227 if (flow->actions_len) {
1228 memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);
1232 /* Same as check_rw_flow_actions() but for flow->key. */
1234 check_rw_flow_key(struct odp_flow *flow)
1236 if (flow->key_len) {
1237 memset(&flow->key[0], 0xcc, sizeof flow->key[0]);