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"
35 #include "poll-loop.h"
40 #include "fatal-signal.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);
59 /* Rate limit for individual messages going to or from the datapath, output at
60 * DBG level. This is very high because, if these are enabled, it is because
61 * we really need to see them. */
62 static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
64 /* Not really much point in logging many dpif errors. */
65 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
67 static void log_operation(const struct dpif *, const char *operation,
69 static void log_flow_operation(const struct dpif *, const char *operation,
70 int error, struct odp_flow *flow);
71 static void log_flow_put(struct dpif *, int error,
72 const struct odp_flow_put *);
73 static bool should_log_flow_message(int error);
74 static void check_rw_odp_flow(struct odp_flow *);
79 static int status = -1;
84 struct shash_node *node;
88 for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
89 dp_register_provider(base_dpif_classes[i]);
93 /* register an exit handler for the registered classes */
94 SHASH_FOR_EACH(node, &dpif_classes) {
95 const struct registered_dpif_class *registered_class = node->data;
96 if (registered_class->dpif_class.exit_hook) {
97 fatal_signal_add_hook(registered_class->dpif_class.exit_hook,
107 /* Performs periodic work needed by all the various kinds of dpifs.
109 * If your program opens any dpifs, it must call both this function and
110 * netdev_run() within its main poll loop. */
114 struct shash_node *node;
115 SHASH_FOR_EACH(node, &dpif_classes) {
116 const struct registered_dpif_class *registered_class = node->data;
117 if (registered_class->dpif_class.run) {
118 registered_class->dpif_class.run();
123 /* Arranges for poll_block() to wake up when dp_run() needs to be called.
125 * If your program opens any dpifs, it must call both this function and
126 * netdev_wait() within its main poll loop. */
130 struct shash_node *node;
131 SHASH_FOR_EACH(node, &dpif_classes) {
132 const struct registered_dpif_class *registered_class = node->data;
133 if (registered_class->dpif_class.wait) {
134 registered_class->dpif_class.wait();
140 /* Start the datapath management.
142 * This function has been thought for a scenario in which the management of the
143 * datapath module and the ofproto module are performed in separate
144 * threads/processes module. */
148 struct shash_node *node;
150 SHASH_FOR_EACH(node, &dpif_classes) {
151 const struct registered_dpif_class *registered_class = node->data;
152 if (registered_class->dpif_class.start) {
153 registered_class->dpif_class.start();
160 /* Registers a new datapath provider. After successful registration, new
161 * datapaths of that type can be opened using dpif_open(). */
163 dp_register_provider(const struct dpif_class *new_class)
165 struct registered_dpif_class *registered_class;
167 if (shash_find(&dpif_classes, new_class->type)) {
168 VLOG_WARN("attempted to register duplicate datapath provider: %s",
173 registered_class = xmalloc(sizeof *registered_class);
174 memcpy(®istered_class->dpif_class, new_class,
175 sizeof registered_class->dpif_class);
176 registered_class->refcount = 0;
178 shash_add(&dpif_classes, new_class->type, registered_class);
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 /* Clears 'types' and enumerates the types of all currently registered datapath
212 * providers into it. The caller must first initialize the svec. */
214 dp_enumerate_types(struct svec *types)
216 struct shash_node *node;
221 SHASH_FOR_EACH(node, &dpif_classes) {
222 const struct registered_dpif_class *registered_class = node->data;
223 svec_add(types, registered_class->dpif_class.type);
227 /* Clears 'names' and enumerates the names of all known created datapaths with
228 * the given 'type'. The caller must first initialize the svec. Returns 0 if
229 * successful, otherwise a positive errno value.
231 * Some kinds of datapaths might not be practically enumerable. This is not
232 * considered an error. */
234 dp_enumerate_names(const char *type, struct svec *names)
236 const struct registered_dpif_class *registered_class;
237 const struct dpif_class *dpif_class;
243 registered_class = shash_find_data(&dpif_classes, type);
244 if (!registered_class) {
245 VLOG_WARN("could not enumerate unknown type: %s", type);
249 dpif_class = ®istered_class->dpif_class;
250 error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
253 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
260 /* Parses 'datapath name', which is of the form type@name into its
261 * component pieces. 'name' and 'type' must be freed by the caller. */
263 dp_parse_name(const char *datapath_name_, char **name, char **type)
265 char *datapath_name = xstrdup(datapath_name_);
268 separator = strchr(datapath_name, '@');
271 *type = datapath_name;
272 *name = xstrdup(separator + 1);
274 *name = datapath_name;
280 do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
282 struct dpif *dpif = NULL;
284 struct registered_dpif_class *registered_class;
288 if (!type || *type == '\0') {
292 registered_class = shash_find_data(&dpif_classes, type);
293 if (!registered_class) {
294 VLOG_WARN("could not create datapath %s of unknown type %s", name,
296 error = EAFNOSUPPORT;
300 error = registered_class->dpif_class.open(name, type, create, &dpif);
302 registered_class->refcount++;
306 *dpifp = error ? NULL : dpif;
310 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
311 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
312 * the empty string to specify the default system type. Returns 0 if
313 * successful, otherwise a positive errno value. On success stores a pointer
314 * to the datapath in '*dpifp', otherwise a null pointer. */
316 dpif_open(const char *name, const char *type, struct dpif **dpifp)
318 return do_open(name, type, false, dpifp);
321 /* Tries to create and open a new datapath with the given 'name' and 'type'.
322 * 'type' may be either NULL or the empty string to specify the default system
323 * type. Will fail if a datapath with 'name' and 'type' already exists.
324 * Returns 0 if successful, otherwise a positive errno value. On success
325 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
327 dpif_create(const char *name, const char *type, struct dpif **dpifp)
329 return do_open(name, type, true, dpifp);
332 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
333 * does not exist. 'type' may be either NULL or the empty string to specify
334 * the default system type. Returns 0 if successful, otherwise a positive
335 * errno value. On success stores a pointer to the datapath in '*dpifp',
336 * otherwise a null pointer. */
338 dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
342 error = dpif_create(name, type, dpifp);
343 if (error == EEXIST || error == EBUSY) {
344 error = dpif_open(name, type, dpifp);
346 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
347 name, strerror(error));
350 VLOG_WARN("failed to create datapath %s: %s", name, strerror(error));
355 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
356 * itself; call dpif_delete() first, instead, if that is desirable. */
358 dpif_close(struct dpif *dpif)
361 struct registered_dpif_class *registered_class;
363 registered_class = shash_find_data(&dpif_classes,
364 dpif->dpif_class->type);
365 assert(registered_class);
366 assert(registered_class->refcount);
368 registered_class->refcount--;
369 dpif_uninit(dpif, true);
373 /* Returns the name of datapath 'dpif' prefixed with the type
374 * (for use in log messages). */
376 dpif_name(const struct dpif *dpif)
378 return dpif->full_name;
381 /* Returns the name of datapath 'dpif' without the type
382 * (for use in device names). */
384 dpif_base_name(const struct dpif *dpif)
386 return dpif->base_name;
389 /* Enumerates all names that may be used to open 'dpif' into 'all_names'. The
390 * Linux datapath, for example, supports opening a datapath both by number,
391 * e.g. "dp0", and by the name of the datapath's local port. For some
392 * datapaths, this might be an infinite set (e.g. in a file name, slashes may
393 * be duplicated any number of times), in which case only the names most likely
394 * to be used will be enumerated.
396 * The caller must already have initialized 'all_names'. Any existing names in
397 * 'all_names' will not be disturbed. */
399 dpif_get_all_names(const struct dpif *dpif, struct svec *all_names)
401 if (dpif->dpif_class->get_all_names) {
402 int error = dpif->dpif_class->get_all_names(dpif, all_names);
404 VLOG_WARN_RL(&error_rl,
405 "failed to retrieve names for datpath %s: %s",
406 dpif_name(dpif), strerror(error));
410 svec_add(all_names, dpif_base_name(dpif));
415 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
416 * ports. After calling this function, it does not make sense to pass 'dpif'
417 * to any functions other than dpif_name() or dpif_close(). */
419 dpif_delete(struct dpif *dpif)
423 COVERAGE_INC(dpif_destroy);
425 error = dpif->dpif_class->destroy(dpif);
426 log_operation(dpif, "delete", error);
430 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
431 * otherwise a positive errno value. */
433 dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats)
435 int error = dpif->dpif_class->get_stats(dpif, stats);
437 memset(stats, 0, sizeof *stats);
439 log_operation(dpif, "get_stats", error);
443 /* Retrieves the current IP fragment handling policy for 'dpif' into
444 * '*drop_frags': true indicates that fragments are dropped, false indicates
445 * that fragments are treated in the same way as other IP packets (except that
446 * the L4 header cannot be read). Returns 0 if successful, otherwise a
447 * positive errno value. */
449 dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags)
451 int error = dpif->dpif_class->get_drop_frags(dpif, drop_frags);
455 log_operation(dpif, "get_drop_frags", error);
459 /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is
460 * the same as for the get_drop_frags member function. Returns 0 if
461 * successful, otherwise a positive errno value. */
463 dpif_set_drop_frags(struct dpif *dpif, bool drop_frags)
465 int error = dpif->dpif_class->set_drop_frags(dpif, drop_frags);
466 log_operation(dpif, "set_drop_frags", error);
470 /* Attempts to add 'devname' as a port on 'dpif', given the combination of
471 * ODP_PORT_* flags in 'flags'. If successful, returns 0 and sets '*port_nop'
472 * to the new port's port number (if 'port_nop' is non-null). On failure,
473 * returns a positive errno value and sets '*port_nop' to UINT16_MAX (if
474 * 'port_nop' is non-null). */
476 dpif_port_add(struct dpif *dpif, const char *devname, uint16_t flags,
482 COVERAGE_INC(dpif_port_add);
484 error = dpif->dpif_class->port_add(dpif, devname, flags, &port_no);
486 VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16,
487 dpif_name(dpif), devname, port_no);
489 VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
490 dpif_name(dpif), devname, strerror(error));
491 port_no = UINT16_MAX;
499 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
500 * otherwise a positive errno value. */
502 dpif_port_del(struct dpif *dpif, uint16_t port_no)
506 COVERAGE_INC(dpif_port_del);
508 error = dpif->dpif_class->port_del(dpif, port_no);
509 log_operation(dpif, "port_del", error);
513 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
514 * initializes '*port' appropriately; on failure, returns a positive errno
517 dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
518 struct odp_port *port)
520 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
522 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s",
523 dpif_name(dpif), port_no, port->devname);
525 memset(port, 0, sizeof *port);
526 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s",
527 dpif_name(dpif), port_no, strerror(error));
532 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
533 * initializes '*port' appropriately; on failure, returns a positive errno
536 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
537 struct odp_port *port)
539 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
541 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16,
542 dpif_name(dpif), devname, port->port);
544 memset(port, 0, sizeof *port);
546 /* Log level is DBG here because all the current callers are interested
547 * in whether 'dpif' actually has a port 'devname', so that it's not an
548 * issue worth logging if it doesn't. */
549 VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s",
550 dpif_name(dpif), devname, strerror(error));
555 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
556 * the port's name into the 'name_size' bytes in 'name', ensuring that the
557 * result is null-terminated. On failure, returns a positive errno value and
558 * makes 'name' the empty string. */
560 dpif_port_get_name(struct dpif *dpif, uint16_t port_no,
561 char *name, size_t name_size)
563 struct odp_port port;
566 assert(name_size > 0);
568 error = dpif_port_query_by_number(dpif, port_no, &port);
570 ovs_strlcpy(name, port.devname, name_size);
577 /* Obtains a list of all the ports in 'dpif'.
579 * If successful, returns 0 and sets '*portsp' to point to an array of
580 * appropriately initialized port structures and '*n_portsp' to the number of
581 * ports in the array. The caller is responsible for freeing '*portp' by
584 * On failure, returns a positive errno value and sets '*portsp' to NULL and
585 * '*n_portsp' to 0. */
587 dpif_port_list(const struct dpif *dpif,
588 struct odp_port **portsp, size_t *n_portsp)
590 struct odp_port *ports;
595 struct odp_stats stats;
598 error = dpif_get_dp_stats(dpif, &stats);
603 ports = xcalloc(stats.n_ports, sizeof *ports);
604 retval = dpif->dpif_class->port_list(dpif, ports, stats.n_ports);
610 } else if (retval <= stats.n_ports) {
616 /* Soft error: port count increased behind our back. Try again. */
629 log_operation(dpif, "port_list", error);
633 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
634 * 'dpif' has changed, this function does one of the following:
636 * - Stores the name of the device that was added to or deleted from 'dpif' in
637 * '*devnamep' and returns 0. The caller is responsible for freeing
638 * '*devnamep' (with free()) when it no longer needs it.
640 * - Returns ENOBUFS and sets '*devnamep' to NULL.
642 * This function may also return 'false positives', where it returns 0 and
643 * '*devnamep' names a device that was not actually added or deleted or it
644 * returns ENOBUFS without any change.
646 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
647 * return other positive errno values to indicate that something has gone
650 dpif_port_poll(const struct dpif *dpif, char **devnamep)
652 int error = dpif->dpif_class->port_poll(dpif, devnamep);
659 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
660 * value other than EAGAIN. */
662 dpif_port_poll_wait(const struct dpif *dpif)
664 dpif->dpif_class->port_poll_wait(dpif);
667 /* Retrieves a list of the port numbers in port group 'group' in 'dpif'.
669 * On success, returns 0 and points '*ports' to a newly allocated array of
670 * integers, each of which is a 'dpif' port number for a port in
671 * 'group'. Stores the number of elements in the array in '*n_ports'. The
672 * caller is responsible for freeing '*ports' by calling free().
674 * On failure, returns a positive errno value and sets '*ports' to NULL and
675 * '*n_ports' to 0. */
677 dpif_port_group_get(const struct dpif *dpif, uint16_t group,
678 uint16_t **ports, size_t *n_ports)
685 int retval = dpif->dpif_class->port_group_get(dpif, group,
694 } else if (retval <= *n_ports) {
700 /* Soft error: there were more ports than we expected in the
701 * group. Try again. */
703 *ports = xcalloc(retval, sizeof **ports);
707 log_operation(dpif, "port_group_get", error);
711 /* Updates port group 'group' in 'dpif', making it contain the 'n_ports' ports
712 * whose 'dpif' port numbers are given in 'n_ports'. Returns 0 if
713 * successful, otherwise a positive errno value.
715 * Behavior is undefined if the values in ports[] are not unique. */
717 dpif_port_group_set(struct dpif *dpif, uint16_t group,
718 const uint16_t ports[], size_t n_ports)
722 COVERAGE_INC(dpif_port_group_set);
724 error = dpif->dpif_class->port_group_set(dpif, group, ports, n_ports);
725 log_operation(dpif, "port_group_set", error);
729 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
730 * positive errno value. */
732 dpif_flow_flush(struct dpif *dpif)
736 COVERAGE_INC(dpif_flow_flush);
738 error = dpif->dpif_class->flow_flush(dpif);
739 log_operation(dpif, "flow_flush", error);
743 /* Queries 'dpif' for a flow entry matching 'flow->key'.
745 * If a flow matching 'flow->key' exists in 'dpif', stores statistics for the
746 * flow into 'flow->stats'. If 'flow->n_actions' is zero, then 'flow->actions'
747 * is ignored. If 'flow->n_actions' is nonzero, then 'flow->actions' should
748 * point to an array of the specified number of actions. At most that many of
749 * the flow's actions will be copied into that array. 'flow->n_actions' will
750 * be updated to the number of actions actually present in the flow, which may
751 * be greater than the number stored if the flow has more actions than space
752 * available in the array.
754 * If no flow matching 'flow->key' exists in 'dpif', returns ENOENT. On other
755 * failure, returns a positive errno value. */
757 dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
761 COVERAGE_INC(dpif_flow_get);
763 check_rw_odp_flow(flow);
764 error = dpif->dpif_class->flow_get(dpif, flow, 1);
766 error = flow->stats.error;
769 /* Make the results predictable on error. */
770 memset(&flow->stats, 0, sizeof flow->stats);
773 if (should_log_flow_message(error)) {
774 log_flow_operation(dpif, "flow_get", error, flow);
779 /* For each flow 'flow' in the 'n' flows in 'flows':
781 * - If a flow matching 'flow->key' exists in 'dpif':
783 * Stores 0 into 'flow->stats.error' and stores statistics for the flow
784 * into 'flow->stats'.
786 * If 'flow->n_actions' is zero, then 'flow->actions' is ignored. If
787 * 'flow->n_actions' is nonzero, then 'flow->actions' should point to an
788 * array of the specified number of actions. At most that many of the
789 * flow's actions will be copied into that array. 'flow->n_actions' will
790 * be updated to the number of actions actually present in the flow, which
791 * may be greater than the number stored if the flow has more actions than
792 * space available in the array.
794 * - Flow-specific errors are indicated by a positive errno value in
795 * 'flow->stats.error'. In particular, ENOENT indicates that no flow
796 * matching 'flow->key' exists in 'dpif'. When an error value is stored, the
797 * contents of 'flow->key' are preserved but other members of 'flow' should
798 * be treated as indeterminate.
800 * Returns 0 if all 'n' flows in 'flows' were updated (whether they were
801 * individually successful or not is indicated by 'flow->stats.error',
802 * however). Returns a positive errno value if an error that prevented this
803 * update occurred, in which the caller must not depend on any elements in
804 * 'flows' being updated or not updated.
807 dpif_flow_get_multiple(const struct dpif *dpif,
808 struct odp_flow flows[], size_t n)
813 COVERAGE_ADD(dpif_flow_get, n);
815 for (i = 0; i < n; i++) {
816 check_rw_odp_flow(&flows[i]);
819 error = dpif->dpif_class->flow_get(dpif, flows, n);
820 log_operation(dpif, "flow_get_multiple", error);
824 /* Adds or modifies a flow in 'dpif' as specified in 'put':
826 * - If the flow specified in 'put->flow' does not exist in 'dpif', then
827 * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': if
828 * it is, the flow will be added, otherwise the operation will fail with
831 * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'.
832 * Behavior in this case depends on whether ODPPF_MODIFY is specified in
833 * 'put->flags': if it is, the flow's actions will be updated, otherwise the
834 * operation will fail with EEXIST. If the flow's actions are updated, then
835 * its statistics will be zeroed if ODPPF_ZERO_STATS is set in 'put->flags',
836 * left as-is otherwise.
838 * Returns 0 if successful, otherwise a positive errno value.
841 dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
845 COVERAGE_INC(dpif_flow_put);
847 error = dpif->dpif_class->flow_put(dpif, put);
848 if (should_log_flow_message(error)) {
849 log_flow_put(dpif, error, put);
854 /* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if 'dpif'
855 * does not contain such a flow.
857 * If successful, updates 'flow->stats', 'flow->n_actions', and 'flow->actions'
858 * as described for dpif_flow_get(). */
860 dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
864 COVERAGE_INC(dpif_flow_del);
866 check_rw_odp_flow(flow);
867 memset(&flow->stats, 0, sizeof flow->stats);
869 error = dpif->dpif_class->flow_del(dpif, flow);
870 if (should_log_flow_message(error)) {
871 log_flow_operation(dpif, "delete flow", error, flow);
876 /* Stores up to 'n' flows in 'dpif' into 'flows', including their statistics
877 * but not including any information about their actions. If successful,
878 * returns 0 and sets '*n_out' to the number of flows actually present in
879 * 'dpif', which might be greater than the number stored (if 'dpif' has more
880 * than 'n' flows). On failure, returns a negative errno value and sets
883 dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n,
889 COVERAGE_INC(dpif_flow_query_list);
890 if (RUNNING_ON_VALGRIND) {
891 memset(flows, 0, n * sizeof *flows);
893 for (i = 0; i < n; i++) {
894 flows[i].actions = NULL;
895 flows[i].n_actions = 0;
898 retval = dpif->dpif_class->flow_list(dpif, flows, n);
901 VLOG_WARN_RL(&error_rl, "%s: flow list failed (%s)",
902 dpif_name(dpif), strerror(-retval));
905 COVERAGE_ADD(dpif_flow_query_list_n, retval);
906 *n_out = MIN(n, retval);
907 VLOG_DBG_RL(&dpmsg_rl, "%s: listed %zu flows (of %d)",
908 dpif_name(dpif), *n_out, retval);
913 /* Retrieves all of the flows in 'dpif'.
915 * If successful, returns 0 and stores in '*flowsp' a pointer to a newly
916 * allocated array of flows, including their statistics but not including any
917 * information about their actions, and sets '*np' to the number of flows in
918 * '*flowsp'. The caller is responsible for freeing '*flowsp' by calling
921 * On failure, returns a positive errno value and sets '*flowsp' to NULL and
924 dpif_flow_list_all(const struct dpif *dpif,
925 struct odp_flow **flowsp, size_t *np)
927 struct odp_stats stats;
928 struct odp_flow *flows;
935 error = dpif_get_dp_stats(dpif, &stats);
940 flows = xmalloc(sizeof *flows * stats.n_flows);
941 error = dpif_flow_list(dpif, flows, stats.n_flows, &n_flows);
947 if (stats.n_flows != n_flows) {
948 VLOG_WARN_RL(&error_rl, "%s: datapath stats reported %"PRIu32" "
949 "flows but flow listing reported %zu",
950 dpif_name(dpif), stats.n_flows, n_flows);
957 /* Causes 'dpif' to perform the 'n_actions' actions in 'actions' on the
958 * Ethernet frame specified in 'packet'.
960 * Pretends that the frame was originally received on the port numbered
961 * 'in_port'. This affects only ODPAT_OUTPUT_GROUP actions, which will not
962 * send a packet out their input port. Specify the number of an unused port
963 * (e.g. UINT16_MAX is currently always unused) to avoid this behavior.
965 * Returns 0 if successful, otherwise a positive errno value. */
967 dpif_execute(struct dpif *dpif, uint16_t in_port,
968 const union odp_action actions[], size_t n_actions,
969 const struct ofpbuf *buf)
973 COVERAGE_INC(dpif_execute);
975 error = dpif->dpif_class->execute(dpif, in_port, actions,
981 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
982 struct ds ds = DS_EMPTY_INITIALIZER;
983 char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
984 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
985 format_odp_actions(&ds, actions, n_actions);
987 ds_put_format(&ds, " failed (%s)", strerror(error));
989 ds_put_format(&ds, " on packet %s", packet);
990 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
997 /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set
998 * in '*listen_mask' indicates that dpif_recv() will receive messages of that
999 * type. Returns 0 if successful, otherwise a positive errno value. */
1001 dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask)
1003 int error = dpif->dpif_class->recv_get_mask(dpif, listen_mask);
1007 log_operation(dpif, "recv_get_mask", error);
1011 /* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in
1012 * '*listen_mask' requests that dpif_recv() receive messages of that type.
1013 * Returns 0 if successful, otherwise a positive errno value. */
1015 dpif_recv_set_mask(struct dpif *dpif, int listen_mask)
1017 int error = dpif->dpif_class->recv_set_mask(dpif, listen_mask);
1018 log_operation(dpif, "recv_set_mask", error);
1022 /* Retrieve the sFlow sampling probability. '*probability' is expressed as the
1023 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
1024 * the probability of sampling a given packet.
1026 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
1027 * indicates that 'dpif' does not support sFlow sampling. */
1029 dpif_get_sflow_probability(const struct dpif *dpif, uint32_t *probability)
1031 int error = (dpif->dpif_class->get_sflow_probability
1032 ? dpif->dpif_class->get_sflow_probability(dpif, probability)
1037 log_operation(dpif, "get_sflow_probability", error);
1041 /* Set the sFlow sampling probability. 'probability' is expressed as the
1042 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
1043 * the probability of sampling a given packet.
1045 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
1046 * indicates that 'dpif' does not support sFlow sampling. */
1048 dpif_set_sflow_probability(struct dpif *dpif, uint32_t probability)
1050 int error = (dpif->dpif_class->set_sflow_probability
1051 ? dpif->dpif_class->set_sflow_probability(dpif, probability)
1053 log_operation(dpif, "set_sflow_probability", error);
1057 /* Attempts to receive a message from 'dpif'. If successful, stores the
1058 * message into '*packetp'. The message, if one is received, will begin with
1059 * 'struct odp_msg' as a header, and will have at least DPIF_RECV_MSG_PADDING
1060 * bytes of headroom. Only messages of the types selected with
1061 * dpif_set_listen_mask() will ordinarily be received (but if a message type is
1062 * enabled and then later disabled, some stragglers might pop up).
1064 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1065 * if no message is immediately available. */
1067 dpif_recv(struct dpif *dpif, struct ofpbuf **packetp)
1069 int error = dpif->dpif_class->recv(dpif, packetp);
1071 struct ofpbuf *buf = *packetp;
1073 assert(ofpbuf_headroom(buf) >= DPIF_RECV_MSG_PADDING);
1074 if (VLOG_IS_DBG_ENABLED()) {
1075 struct odp_msg *msg = buf->data;
1076 void *payload = msg + 1;
1077 size_t payload_len = buf->size - sizeof *msg;
1078 char *s = ofp_packet_to_string(payload, payload_len, payload_len);
1079 VLOG_DBG_RL(&dpmsg_rl, "%s: received %s message of length "
1080 "%zu on port %"PRIu16": %s", dpif_name(dpif),
1081 (msg->type == _ODPL_MISS_NR ? "miss"
1082 : msg->type == _ODPL_ACTION_NR ? "action"
1083 : msg->type == _ODPL_SFLOW_NR ? "sFlow"
1085 payload_len, msg->port, s);
1094 /* Discards all messages that would otherwise be received by dpif_recv() on
1095 * 'dpif'. Returns 0 if successful, otherwise a positive errno value. */
1097 dpif_recv_purge(struct dpif *dpif)
1099 struct odp_stats stats;
1103 COVERAGE_INC(dpif_purge);
1105 error = dpif_get_dp_stats(dpif, &stats);
1110 for (i = 0; i < stats.max_miss_queue + stats.max_action_queue + stats.max_sflow_queue; i++) {
1112 error = dpif_recv(dpif, &buf);
1114 return error == EAGAIN ? 0 : error;
1121 /* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
1122 * received with dpif_recv(). */
1124 dpif_recv_wait(struct dpif *dpif)
1126 dpif->dpif_class->recv_wait(dpif);
1129 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1130 * and '*engine_id', respectively. */
1132 dpif_get_netflow_ids(const struct dpif *dpif,
1133 uint8_t *engine_type, uint8_t *engine_id)
1135 *engine_type = dpif->netflow_engine_type;
1136 *engine_id = dpif->netflow_engine_id;
1139 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1140 * value for use in the ODPAT_SET_PRIORITY action. On success, returns 0 and
1141 * stores the priority into '*priority'. On failure, returns a positive errno
1142 * value and stores 0 into '*priority'. */
1144 dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1147 int error = (dpif->dpif_class->queue_to_priority
1148 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1154 log_operation(dpif, "queue_to_priority", error);
1159 dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1161 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1163 dpif->dpif_class = dpif_class;
1164 dpif->base_name = xstrdup(name);
1165 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
1166 dpif->netflow_engine_type = netflow_engine_type;
1167 dpif->netflow_engine_id = netflow_engine_id;
1170 /* Undoes the results of initialization.
1172 * Normally this function only needs to be called from dpif_close().
1173 * However, it may be called by providers due to an error on opening
1174 * that occurs after initialization. It this case dpif_close() would
1175 * never be called. */
1177 dpif_uninit(struct dpif *dpif, bool close)
1179 char *base_name = dpif->base_name;
1180 char *full_name = dpif->full_name;
1183 dpif->dpif_class->close(dpif);
1191 log_operation(const struct dpif *dpif, const char *operation, int error)
1194 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
1196 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1197 dpif_name(dpif), operation, strerror(error));
1201 static enum vlog_level
1202 flow_message_log_level(int error)
1204 return error ? VLL_WARN : VLL_DBG;
1208 should_log_flow_message(int error)
1210 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1211 error ? &error_rl : &dpmsg_rl);
1215 log_flow_message(const struct dpif *dpif, int error, const char *operation,
1216 const flow_t *flow, const struct odp_flow_stats *stats,
1217 const union odp_action *actions, size_t n_actions)
1219 struct ds ds = DS_EMPTY_INITIALIZER;
1220 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1222 ds_put_cstr(&ds, "failed to ");
1224 ds_put_format(&ds, "%s ", operation);
1226 ds_put_format(&ds, "(%s) ", strerror(error));
1228 flow_format(&ds, flow);
1230 ds_put_cstr(&ds, ", ");
1231 format_odp_flow_stats(&ds, stats);
1233 if (actions || n_actions) {
1234 ds_put_cstr(&ds, ", actions:");
1235 format_odp_actions(&ds, actions, n_actions);
1237 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1242 log_flow_operation(const struct dpif *dpif, const char *operation, int error,
1243 struct odp_flow *flow)
1246 flow->n_actions = 0;
1248 log_flow_message(dpif, error, operation, &flow->key,
1249 !error ? &flow->stats : NULL,
1250 flow->actions, flow->n_actions);
1254 log_flow_put(struct dpif *dpif, int error, const struct odp_flow_put *put)
1256 enum { ODPPF_ALL = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS };
1260 ds_put_cstr(&s, "put");
1261 if (put->flags & ODPPF_CREATE) {
1262 ds_put_cstr(&s, "[create]");
1264 if (put->flags & ODPPF_MODIFY) {
1265 ds_put_cstr(&s, "[modify]");
1267 if (put->flags & ODPPF_ZERO_STATS) {
1268 ds_put_cstr(&s, "[zero]");
1270 if (put->flags & ~ODPPF_ALL) {
1271 ds_put_format(&s, "[%x]", put->flags & ~ODPPF_ALL);
1273 log_flow_message(dpif, error, ds_cstr(&s), &put->flow.key,
1274 !error ? &put->flow.stats : NULL,
1275 put->flow.actions, put->flow.n_actions);
1279 /* There is a tendency to construct odp_flow objects on the stack and to
1280 * forget to properly initialize their "actions" and "n_actions" members.
1281 * When this happens, we get memory corruption because the kernel
1282 * writes through the random pointer that is in the "actions" member.
1284 * This function attempts to combat the problem by:
1286 * - Forcing a segfault if "actions" points to an invalid region (instead
1287 * of just getting back EFAULT, which can be easily missed in the log).
1289 * - Storing a distinctive value that is likely to cause an
1290 * easy-to-identify error later if it is dereferenced, etc.
1292 * - Triggering a warning on uninitialized memory from Valgrind if
1293 * "actions" or "n_actions" was not initialized.
1296 check_rw_odp_flow(struct odp_flow *flow)
1298 if (flow->n_actions) {
1299 memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);