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"
42 VLOG_DEFINE_THIS_MODULE(dpif)
44 static const struct dpif_class *base_dpif_classes[] = {
51 struct registered_dpif_class {
52 struct dpif_class dpif_class;
55 static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes);
57 /* Rate limit for individual messages going to or from the datapath, output at
58 * DBG level. This is very high because, if these are enabled, it is because
59 * we really need to see them. */
60 static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
62 /* Not really much point in logging many dpif errors. */
63 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
65 static void log_operation(const struct dpif *, const char *operation,
67 static void log_flow_operation(const struct dpif *, const char *operation,
68 int error, struct odp_flow *flow);
69 static void log_flow_put(struct dpif *, int error,
70 const struct odp_flow_put *);
71 static bool should_log_flow_message(int error);
72 static void check_rw_odp_flow(struct odp_flow *);
77 static int status = -1;
83 for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
84 dp_register_provider(base_dpif_classes[i]);
89 /* Performs periodic work needed by all the various kinds of dpifs.
91 * If your program opens any dpifs, it must call both this function and
92 * netdev_run() within its main poll loop. */
96 struct shash_node *node;
97 SHASH_FOR_EACH(node, &dpif_classes) {
98 const struct registered_dpif_class *registered_class = node->data;
99 if (registered_class->dpif_class.run) {
100 registered_class->dpif_class.run();
105 /* Arranges for poll_block() to wake up when dp_run() needs to be called.
107 * If your program opens any dpifs, it must call both this function and
108 * netdev_wait() within its main poll loop. */
112 struct shash_node *node;
113 SHASH_FOR_EACH(node, &dpif_classes) {
114 const struct registered_dpif_class *registered_class = node->data;
115 if (registered_class->dpif_class.wait) {
116 registered_class->dpif_class.wait();
121 /* Registers a new datapath provider. After successful registration, new
122 * datapaths of that type can be opened using dpif_open(). */
124 dp_register_provider(const struct dpif_class *new_class)
126 struct registered_dpif_class *registered_class;
128 if (shash_find(&dpif_classes, new_class->type)) {
129 VLOG_WARN("attempted to register duplicate datapath provider: %s",
134 registered_class = xmalloc(sizeof *registered_class);
135 memcpy(®istered_class->dpif_class, new_class,
136 sizeof registered_class->dpif_class);
137 registered_class->refcount = 0;
139 shash_add(&dpif_classes, new_class->type, registered_class);
144 /* Unregisters a datapath provider. 'type' must have been previously
145 * registered and not currently be in use by any dpifs. After unregistration
146 * new datapaths of that type cannot be opened using dpif_open(). */
148 dp_unregister_provider(const char *type)
150 struct shash_node *node;
151 struct registered_dpif_class *registered_class;
153 node = shash_find(&dpif_classes, type);
155 VLOG_WARN("attempted to unregister a datapath provider that is not "
156 "registered: %s", type);
160 registered_class = node->data;
161 if (registered_class->refcount) {
162 VLOG_WARN("attempted to unregister in use datapath provider: %s", type);
166 shash_delete(&dpif_classes, node);
167 free(registered_class);
172 /* Clears 'types' and enumerates the types of all currently registered datapath
173 * providers into it. The caller must first initialize the svec. */
175 dp_enumerate_types(struct svec *types)
177 struct shash_node *node;
182 SHASH_FOR_EACH(node, &dpif_classes) {
183 const struct registered_dpif_class *registered_class = node->data;
184 svec_add(types, registered_class->dpif_class.type);
188 /* Clears 'names' and enumerates the names of all known created datapaths with
189 * the given 'type'. The caller must first initialize the svec. Returns 0 if
190 * successful, otherwise a positive errno value.
192 * Some kinds of datapaths might not be practically enumerable. This is not
193 * considered an error. */
195 dp_enumerate_names(const char *type, struct svec *names)
197 const struct registered_dpif_class *registered_class;
198 const struct dpif_class *dpif_class;
204 registered_class = shash_find_data(&dpif_classes, type);
205 if (!registered_class) {
206 VLOG_WARN("could not enumerate unknown type: %s", type);
210 dpif_class = ®istered_class->dpif_class;
211 error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
214 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
221 /* Parses 'datapath name', which is of the form type@name into its
222 * component pieces. 'name' and 'type' must be freed by the caller. */
224 dp_parse_name(const char *datapath_name_, char **name, char **type)
226 char *datapath_name = xstrdup(datapath_name_);
229 separator = strchr(datapath_name, '@');
232 *type = datapath_name;
233 *name = xstrdup(separator + 1);
235 *name = datapath_name;
241 do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
243 struct dpif *dpif = NULL;
245 struct registered_dpif_class *registered_class;
249 if (!type || *type == '\0') {
253 registered_class = shash_find_data(&dpif_classes, type);
254 if (!registered_class) {
255 VLOG_WARN("could not create datapath %s of unknown type %s", name,
257 error = EAFNOSUPPORT;
261 error = registered_class->dpif_class.open(name, type, create, &dpif);
263 registered_class->refcount++;
267 *dpifp = error ? NULL : dpif;
271 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
272 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
273 * the empty string to specify the default system type. Returns 0 if
274 * successful, otherwise a positive errno value. On success stores a pointer
275 * to the datapath in '*dpifp', otherwise a null pointer. */
277 dpif_open(const char *name, const char *type, struct dpif **dpifp)
279 return do_open(name, type, false, dpifp);
282 /* Tries to create and open a new datapath with the given 'name' and 'type'.
283 * 'type' may be either NULL or the empty string to specify the default system
284 * type. Will fail if a datapath with 'name' and 'type' already exists.
285 * Returns 0 if successful, otherwise a positive errno value. On success
286 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
288 dpif_create(const char *name, const char *type, struct dpif **dpifp)
290 return do_open(name, type, true, dpifp);
293 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
294 * does not exist. 'type' may be either NULL or the empty string to specify
295 * the default system type. Returns 0 if successful, otherwise a positive
296 * errno value. On success stores a pointer to the datapath in '*dpifp',
297 * otherwise a null pointer. */
299 dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
303 error = dpif_create(name, type, dpifp);
304 if (error == EEXIST || error == EBUSY) {
305 error = dpif_open(name, type, dpifp);
307 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
308 name, strerror(error));
311 VLOG_WARN("failed to create datapath %s: %s", name, strerror(error));
316 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
317 * itself; call dpif_delete() first, instead, if that is desirable. */
319 dpif_close(struct dpif *dpif)
322 struct registered_dpif_class *registered_class;
324 registered_class = shash_find_data(&dpif_classes,
325 dpif->dpif_class->type);
326 assert(registered_class);
327 assert(registered_class->refcount);
329 registered_class->refcount--;
330 dpif_uninit(dpif, true);
334 /* Returns the name of datapath 'dpif' prefixed with the type
335 * (for use in log messages). */
337 dpif_name(const struct dpif *dpif)
339 return dpif->full_name;
342 /* Returns the name of datapath 'dpif' without the type
343 * (for use in device names). */
345 dpif_base_name(const struct dpif *dpif)
347 return dpif->base_name;
350 /* Enumerates all names that may be used to open 'dpif' into 'all_names'. The
351 * Linux datapath, for example, supports opening a datapath both by number,
352 * e.g. "dp0", and by the name of the datapath's local port. For some
353 * datapaths, this might be an infinite set (e.g. in a file name, slashes may
354 * be duplicated any number of times), in which case only the names most likely
355 * to be used will be enumerated.
357 * The caller must already have initialized 'all_names'. Any existing names in
358 * 'all_names' will not be disturbed. */
360 dpif_get_all_names(const struct dpif *dpif, struct svec *all_names)
362 if (dpif->dpif_class->get_all_names) {
363 int error = dpif->dpif_class->get_all_names(dpif, all_names);
365 VLOG_WARN_RL(&error_rl,
366 "failed to retrieve names for datpath %s: %s",
367 dpif_name(dpif), strerror(error));
371 svec_add(all_names, dpif_base_name(dpif));
376 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
377 * ports. After calling this function, it does not make sense to pass 'dpif'
378 * to any functions other than dpif_name() or dpif_close(). */
380 dpif_delete(struct dpif *dpif)
384 COVERAGE_INC(dpif_destroy);
386 error = dpif->dpif_class->destroy(dpif);
387 log_operation(dpif, "delete", error);
391 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
392 * otherwise a positive errno value. */
394 dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats)
396 int error = dpif->dpif_class->get_stats(dpif, stats);
398 memset(stats, 0, sizeof *stats);
400 log_operation(dpif, "get_stats", error);
404 /* Retrieves the current IP fragment handling policy for 'dpif' into
405 * '*drop_frags': true indicates that fragments are dropped, false indicates
406 * that fragments are treated in the same way as other IP packets (except that
407 * the L4 header cannot be read). Returns 0 if successful, otherwise a
408 * positive errno value. */
410 dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags)
412 int error = dpif->dpif_class->get_drop_frags(dpif, drop_frags);
416 log_operation(dpif, "get_drop_frags", error);
420 /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is
421 * the same as for the get_drop_frags member function. Returns 0 if
422 * successful, otherwise a positive errno value. */
424 dpif_set_drop_frags(struct dpif *dpif, bool drop_frags)
426 int error = dpif->dpif_class->set_drop_frags(dpif, drop_frags);
427 log_operation(dpif, "set_drop_frags", error);
431 /* Attempts to add 'devname' as a port on 'dpif', given the combination of
432 * ODP_PORT_* flags in 'flags'. If successful, returns 0 and sets '*port_nop'
433 * to the new port's port number (if 'port_nop' is non-null). On failure,
434 * returns a positive errno value and sets '*port_nop' to UINT16_MAX (if
435 * 'port_nop' is non-null). */
437 dpif_port_add(struct dpif *dpif, const char *devname, uint16_t flags,
443 COVERAGE_INC(dpif_port_add);
445 error = dpif->dpif_class->port_add(dpif, devname, flags, &port_no);
447 VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16,
448 dpif_name(dpif), devname, port_no);
450 VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
451 dpif_name(dpif), devname, strerror(error));
452 port_no = UINT16_MAX;
460 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
461 * otherwise a positive errno value. */
463 dpif_port_del(struct dpif *dpif, uint16_t port_no)
467 COVERAGE_INC(dpif_port_del);
469 error = dpif->dpif_class->port_del(dpif, port_no);
470 log_operation(dpif, "port_del", error);
474 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
475 * initializes '*port' appropriately; on failure, returns a positive errno
478 dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
479 struct odp_port *port)
481 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
483 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s",
484 dpif_name(dpif), port_no, port->devname);
486 memset(port, 0, sizeof *port);
487 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s",
488 dpif_name(dpif), port_no, strerror(error));
493 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
494 * initializes '*port' appropriately; on failure, returns a positive errno
497 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
498 struct odp_port *port)
500 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
502 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16,
503 dpif_name(dpif), devname, port->port);
505 memset(port, 0, sizeof *port);
507 /* Log level is DBG here because all the current callers are interested
508 * in whether 'dpif' actually has a port 'devname', so that it's not an
509 * issue worth logging if it doesn't. */
510 VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s",
511 dpif_name(dpif), devname, strerror(error));
516 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
517 * the port's name into the 'name_size' bytes in 'name', ensuring that the
518 * result is null-terminated. On failure, returns a positive errno value and
519 * makes 'name' the empty string. */
521 dpif_port_get_name(struct dpif *dpif, uint16_t port_no,
522 char *name, size_t name_size)
524 struct odp_port port;
527 assert(name_size > 0);
529 error = dpif_port_query_by_number(dpif, port_no, &port);
531 ovs_strlcpy(name, port.devname, name_size);
538 /* Obtains a list of all the ports in 'dpif'.
540 * If successful, returns 0 and sets '*portsp' to point to an array of
541 * appropriately initialized port structures and '*n_portsp' to the number of
542 * ports in the array. The caller is responsible for freeing '*portp' by
545 * On failure, returns a positive errno value and sets '*portsp' to NULL and
546 * '*n_portsp' to 0. */
548 dpif_port_list(const struct dpif *dpif,
549 struct odp_port **portsp, size_t *n_portsp)
551 struct odp_port *ports;
556 struct odp_stats stats;
559 error = dpif_get_dp_stats(dpif, &stats);
564 ports = xcalloc(stats.n_ports, sizeof *ports);
565 retval = dpif->dpif_class->port_list(dpif, ports, stats.n_ports);
571 } else if (retval <= stats.n_ports) {
577 /* Soft error: port count increased behind our back. Try again. */
590 log_operation(dpif, "port_list", error);
594 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
595 * 'dpif' has changed, this function does one of the following:
597 * - Stores the name of the device that was added to or deleted from 'dpif' in
598 * '*devnamep' and returns 0. The caller is responsible for freeing
599 * '*devnamep' (with free()) when it no longer needs it.
601 * - Returns ENOBUFS and sets '*devnamep' to NULL.
603 * This function may also return 'false positives', where it returns 0 and
604 * '*devnamep' names a device that was not actually added or deleted or it
605 * returns ENOBUFS without any change.
607 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
608 * return other positive errno values to indicate that something has gone
611 dpif_port_poll(const struct dpif *dpif, char **devnamep)
613 int error = dpif->dpif_class->port_poll(dpif, devnamep);
620 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
621 * value other than EAGAIN. */
623 dpif_port_poll_wait(const struct dpif *dpif)
625 dpif->dpif_class->port_poll_wait(dpif);
628 /* Retrieves a list of the port numbers in port group 'group' in 'dpif'.
630 * On success, returns 0 and points '*ports' to a newly allocated array of
631 * integers, each of which is a 'dpif' port number for a port in
632 * 'group'. Stores the number of elements in the array in '*n_ports'. The
633 * caller is responsible for freeing '*ports' by calling free().
635 * On failure, returns a positive errno value and sets '*ports' to NULL and
636 * '*n_ports' to 0. */
638 dpif_port_group_get(const struct dpif *dpif, uint16_t group,
639 uint16_t **ports, size_t *n_ports)
646 int retval = dpif->dpif_class->port_group_get(dpif, group,
655 } else if (retval <= *n_ports) {
661 /* Soft error: there were more ports than we expected in the
662 * group. Try again. */
664 *ports = xcalloc(retval, sizeof **ports);
668 log_operation(dpif, "port_group_get", error);
672 /* Updates port group 'group' in 'dpif', making it contain the 'n_ports' ports
673 * whose 'dpif' port numbers are given in 'n_ports'. Returns 0 if
674 * successful, otherwise a positive errno value.
676 * Behavior is undefined if the values in ports[] are not unique. */
678 dpif_port_group_set(struct dpif *dpif, uint16_t group,
679 const uint16_t ports[], size_t n_ports)
683 COVERAGE_INC(dpif_port_group_set);
685 error = dpif->dpif_class->port_group_set(dpif, group, ports, n_ports);
686 log_operation(dpif, "port_group_set", error);
690 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
691 * positive errno value. */
693 dpif_flow_flush(struct dpif *dpif)
697 COVERAGE_INC(dpif_flow_flush);
699 error = dpif->dpif_class->flow_flush(dpif);
700 log_operation(dpif, "flow_flush", error);
704 /* Queries 'dpif' for a flow entry matching 'flow->key'.
706 * If a flow matching 'flow->key' exists in 'dpif', stores statistics for the
707 * flow into 'flow->stats'. If 'flow->n_actions' is zero, then 'flow->actions'
708 * is ignored. If 'flow->n_actions' is nonzero, then 'flow->actions' should
709 * point to an array of the specified number of actions. At most that many of
710 * the flow's actions will be copied into that array. 'flow->n_actions' will
711 * be updated to the number of actions actually present in the flow, which may
712 * be greater than the number stored if the flow has more actions than space
713 * available in the array.
715 * If no flow matching 'flow->key' exists in 'dpif', returns ENOENT. On other
716 * failure, returns a positive errno value. */
718 dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
722 COVERAGE_INC(dpif_flow_get);
724 check_rw_odp_flow(flow);
725 error = dpif->dpif_class->flow_get(dpif, flow, 1);
727 error = flow->stats.error;
730 /* Make the results predictable on error. */
731 memset(&flow->stats, 0, sizeof flow->stats);
734 if (should_log_flow_message(error)) {
735 log_flow_operation(dpif, "flow_get", error, flow);
740 /* For each flow 'flow' in the 'n' flows in 'flows':
742 * - If a flow matching 'flow->key' exists in 'dpif':
744 * Stores 0 into 'flow->stats.error' and stores statistics for the flow
745 * into 'flow->stats'.
747 * If 'flow->n_actions' is zero, then 'flow->actions' is ignored. If
748 * 'flow->n_actions' is nonzero, then 'flow->actions' should point to an
749 * array of the specified number of actions. At most that many of the
750 * flow's actions will be copied into that array. 'flow->n_actions' will
751 * be updated to the number of actions actually present in the flow, which
752 * may be greater than the number stored if the flow has more actions than
753 * space available in the array.
755 * - Flow-specific errors are indicated by a positive errno value in
756 * 'flow->stats.error'. In particular, ENOENT indicates that no flow
757 * matching 'flow->key' exists in 'dpif'. When an error value is stored, the
758 * contents of 'flow->key' are preserved but other members of 'flow' should
759 * be treated as indeterminate.
761 * Returns 0 if all 'n' flows in 'flows' were updated (whether they were
762 * individually successful or not is indicated by 'flow->stats.error',
763 * however). Returns a positive errno value if an error that prevented this
764 * update occurred, in which the caller must not depend on any elements in
765 * 'flows' being updated or not updated.
768 dpif_flow_get_multiple(const struct dpif *dpif,
769 struct odp_flow flows[], size_t n)
774 COVERAGE_ADD(dpif_flow_get, n);
776 for (i = 0; i < n; i++) {
777 check_rw_odp_flow(&flows[i]);
780 error = dpif->dpif_class->flow_get(dpif, flows, n);
781 log_operation(dpif, "flow_get_multiple", error);
785 /* Adds or modifies a flow in 'dpif' as specified in 'put':
787 * - If the flow specified in 'put->flow' does not exist in 'dpif', then
788 * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': if
789 * it is, the flow will be added, otherwise the operation will fail with
792 * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'.
793 * Behavior in this case depends on whether ODPPF_MODIFY is specified in
794 * 'put->flags': if it is, the flow's actions will be updated, otherwise the
795 * operation will fail with EEXIST. If the flow's actions are updated, then
796 * its statistics will be zeroed if ODPPF_ZERO_STATS is set in 'put->flags',
797 * left as-is otherwise.
799 * Returns 0 if successful, otherwise a positive errno value.
802 dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
806 COVERAGE_INC(dpif_flow_put);
808 error = dpif->dpif_class->flow_put(dpif, put);
809 if (should_log_flow_message(error)) {
810 log_flow_put(dpif, error, put);
815 /* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if 'dpif'
816 * does not contain such a flow.
818 * If successful, updates 'flow->stats', 'flow->n_actions', and 'flow->actions'
819 * as described for dpif_flow_get(). */
821 dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
825 COVERAGE_INC(dpif_flow_del);
827 check_rw_odp_flow(flow);
828 memset(&flow->stats, 0, sizeof flow->stats);
830 error = dpif->dpif_class->flow_del(dpif, flow);
831 if (should_log_flow_message(error)) {
832 log_flow_operation(dpif, "delete flow", error, flow);
837 /* Stores up to 'n' flows in 'dpif' into 'flows', including their statistics
838 * but not including any information about their actions. If successful,
839 * returns 0 and sets '*n_out' to the number of flows actually present in
840 * 'dpif', which might be greater than the number stored (if 'dpif' has more
841 * than 'n' flows). On failure, returns a negative errno value and sets
844 dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n,
850 COVERAGE_INC(dpif_flow_query_list);
851 if (RUNNING_ON_VALGRIND) {
852 memset(flows, 0, n * sizeof *flows);
854 for (i = 0; i < n; i++) {
855 flows[i].actions = NULL;
856 flows[i].n_actions = 0;
859 retval = dpif->dpif_class->flow_list(dpif, flows, n);
862 VLOG_WARN_RL(&error_rl, "%s: flow list failed (%s)",
863 dpif_name(dpif), strerror(-retval));
866 COVERAGE_ADD(dpif_flow_query_list_n, retval);
867 *n_out = MIN(n, retval);
868 VLOG_DBG_RL(&dpmsg_rl, "%s: listed %zu flows (of %d)",
869 dpif_name(dpif), *n_out, retval);
874 /* Retrieves all of the flows in 'dpif'.
876 * If successful, returns 0 and stores in '*flowsp' a pointer to a newly
877 * allocated array of flows, including their statistics but not including any
878 * information about their actions, and sets '*np' to the number of flows in
879 * '*flowsp'. The caller is responsible for freeing '*flowsp' by calling
882 * On failure, returns a positive errno value and sets '*flowsp' to NULL and
885 dpif_flow_list_all(const struct dpif *dpif,
886 struct odp_flow **flowsp, size_t *np)
888 struct odp_stats stats;
889 struct odp_flow *flows;
896 error = dpif_get_dp_stats(dpif, &stats);
901 flows = xmalloc(sizeof *flows * stats.n_flows);
902 error = dpif_flow_list(dpif, flows, stats.n_flows, &n_flows);
908 if (stats.n_flows != n_flows) {
909 VLOG_WARN_RL(&error_rl, "%s: datapath stats reported %"PRIu32" "
910 "flows but flow listing reported %zu",
911 dpif_name(dpif), stats.n_flows, n_flows);
918 /* Causes 'dpif' to perform the 'n_actions' actions in 'actions' on the
919 * Ethernet frame specified in 'packet'.
921 * Pretends that the frame was originally received on the port numbered
922 * 'in_port'. This affects only ODPAT_OUTPUT_GROUP actions, which will not
923 * send a packet out their input port. Specify the number of an unused port
924 * (e.g. UINT16_MAX is currently always unused) to avoid this behavior.
926 * Returns 0 if successful, otherwise a positive errno value. */
928 dpif_execute(struct dpif *dpif, uint16_t in_port,
929 const union odp_action actions[], size_t n_actions,
930 const struct ofpbuf *buf)
934 COVERAGE_INC(dpif_execute);
936 error = dpif->dpif_class->execute(dpif, in_port, actions,
942 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
943 struct ds ds = DS_EMPTY_INITIALIZER;
944 char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
945 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
946 format_odp_actions(&ds, actions, n_actions);
948 ds_put_format(&ds, " failed (%s)", strerror(error));
950 ds_put_format(&ds, " on packet %s", packet);
951 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
958 /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set
959 * in '*listen_mask' indicates that dpif_recv() will receive messages of that
960 * type. Returns 0 if successful, otherwise a positive errno value. */
962 dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask)
964 int error = dpif->dpif_class->recv_get_mask(dpif, listen_mask);
968 log_operation(dpif, "recv_get_mask", error);
972 /* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in
973 * '*listen_mask' requests that dpif_recv() receive messages of that type.
974 * Returns 0 if successful, otherwise a positive errno value. */
976 dpif_recv_set_mask(struct dpif *dpif, int listen_mask)
978 int error = dpif->dpif_class->recv_set_mask(dpif, listen_mask);
979 log_operation(dpif, "recv_set_mask", error);
983 /* Retrieve the sFlow sampling probability. '*probability' is expressed as the
984 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
985 * the probability of sampling a given packet.
987 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
988 * indicates that 'dpif' does not support sFlow sampling. */
990 dpif_get_sflow_probability(const struct dpif *dpif, uint32_t *probability)
992 int error = (dpif->dpif_class->get_sflow_probability
993 ? dpif->dpif_class->get_sflow_probability(dpif, probability)
998 log_operation(dpif, "get_sflow_probability", error);
1002 /* Set the sFlow sampling probability. 'probability' is expressed as the
1003 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
1004 * the probability of sampling a given packet.
1006 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
1007 * indicates that 'dpif' does not support sFlow sampling. */
1009 dpif_set_sflow_probability(struct dpif *dpif, uint32_t probability)
1011 int error = (dpif->dpif_class->set_sflow_probability
1012 ? dpif->dpif_class->set_sflow_probability(dpif, probability)
1014 log_operation(dpif, "set_sflow_probability", error);
1018 /* Attempts to receive a message from 'dpif'. If successful, stores the
1019 * message into '*packetp'. The message, if one is received, will begin with
1020 * 'struct odp_msg' as a header, and will have at least DPIF_RECV_MSG_PADDING
1021 * bytes of headroom. Only messages of the types selected with
1022 * dpif_set_listen_mask() will ordinarily be received (but if a message type is
1023 * enabled and then later disabled, some stragglers might pop up).
1025 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1026 * if no message is immediately available. */
1028 dpif_recv(struct dpif *dpif, struct ofpbuf **packetp)
1030 int error = dpif->dpif_class->recv(dpif, packetp);
1032 struct ofpbuf *buf = *packetp;
1034 assert(ofpbuf_headroom(buf) >= DPIF_RECV_MSG_PADDING);
1035 if (VLOG_IS_DBG_ENABLED()) {
1036 struct odp_msg *msg = buf->data;
1037 void *payload = msg + 1;
1038 size_t payload_len = buf->size - sizeof *msg;
1039 char *s = ofp_packet_to_string(payload, payload_len, payload_len);
1040 VLOG_DBG_RL(&dpmsg_rl, "%s: received %s message of length "
1041 "%zu on port %"PRIu16": %s", dpif_name(dpif),
1042 (msg->type == _ODPL_MISS_NR ? "miss"
1043 : msg->type == _ODPL_ACTION_NR ? "action"
1044 : msg->type == _ODPL_SFLOW_NR ? "sFlow"
1046 payload_len, msg->port, s);
1055 /* Discards all messages that would otherwise be received by dpif_recv() on
1056 * 'dpif'. Returns 0 if successful, otherwise a positive errno value. */
1058 dpif_recv_purge(struct dpif *dpif)
1060 struct odp_stats stats;
1064 COVERAGE_INC(dpif_purge);
1066 error = dpif_get_dp_stats(dpif, &stats);
1071 for (i = 0; i < stats.max_miss_queue + stats.max_action_queue + stats.max_sflow_queue; i++) {
1073 error = dpif_recv(dpif, &buf);
1075 return error == EAGAIN ? 0 : error;
1082 /* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
1083 * received with dpif_recv(). */
1085 dpif_recv_wait(struct dpif *dpif)
1087 dpif->dpif_class->recv_wait(dpif);
1090 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1091 * and '*engine_id', respectively. */
1093 dpif_get_netflow_ids(const struct dpif *dpif,
1094 uint8_t *engine_type, uint8_t *engine_id)
1096 *engine_type = dpif->netflow_engine_type;
1097 *engine_id = dpif->netflow_engine_id;
1100 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1101 * value for use in the ODPAT_SET_PRIORITY action. On success, returns 0 and
1102 * stores the priority into '*priority'. On failure, returns a positive errno
1103 * value and stores 0 into '*priority'. */
1105 dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1108 int error = (dpif->dpif_class->queue_to_priority
1109 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1115 log_operation(dpif, "queue_to_priority", error);
1120 dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1122 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1124 dpif->dpif_class = dpif_class;
1125 dpif->base_name = xstrdup(name);
1126 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
1127 dpif->netflow_engine_type = netflow_engine_type;
1128 dpif->netflow_engine_id = netflow_engine_id;
1131 /* Undoes the results of initialization.
1133 * Normally this function only needs to be called from dpif_close().
1134 * However, it may be called by providers due to an error on opening
1135 * that occurs after initialization. It this case dpif_close() would
1136 * never be called. */
1138 dpif_uninit(struct dpif *dpif, bool close)
1140 char *base_name = dpif->base_name;
1141 char *full_name = dpif->full_name;
1144 dpif->dpif_class->close(dpif);
1152 log_operation(const struct dpif *dpif, const char *operation, int error)
1155 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
1157 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1158 dpif_name(dpif), operation, strerror(error));
1162 static enum vlog_level
1163 flow_message_log_level(int error)
1165 return error ? VLL_WARN : VLL_DBG;
1169 should_log_flow_message(int error)
1171 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1172 error ? &error_rl : &dpmsg_rl);
1176 log_flow_message(const struct dpif *dpif, int error, const char *operation,
1177 const flow_t *flow, const struct odp_flow_stats *stats,
1178 const union odp_action *actions, size_t n_actions)
1180 struct ds ds = DS_EMPTY_INITIALIZER;
1181 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1183 ds_put_cstr(&ds, "failed to ");
1185 ds_put_format(&ds, "%s ", operation);
1187 ds_put_format(&ds, "(%s) ", strerror(error));
1189 flow_format(&ds, flow);
1191 ds_put_cstr(&ds, ", ");
1192 format_odp_flow_stats(&ds, stats);
1194 if (actions || n_actions) {
1195 ds_put_cstr(&ds, ", actions:");
1196 format_odp_actions(&ds, actions, n_actions);
1198 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1203 log_flow_operation(const struct dpif *dpif, const char *operation, int error,
1204 struct odp_flow *flow)
1207 flow->n_actions = 0;
1209 log_flow_message(dpif, error, operation, &flow->key,
1210 !error ? &flow->stats : NULL,
1211 flow->actions, flow->n_actions);
1215 log_flow_put(struct dpif *dpif, int error, const struct odp_flow_put *put)
1217 enum { ODPPF_ALL = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS };
1221 ds_put_cstr(&s, "put");
1222 if (put->flags & ODPPF_CREATE) {
1223 ds_put_cstr(&s, "[create]");
1225 if (put->flags & ODPPF_MODIFY) {
1226 ds_put_cstr(&s, "[modify]");
1228 if (put->flags & ODPPF_ZERO_STATS) {
1229 ds_put_cstr(&s, "[zero]");
1231 if (put->flags & ~ODPPF_ALL) {
1232 ds_put_format(&s, "[%x]", put->flags & ~ODPPF_ALL);
1234 log_flow_message(dpif, error, ds_cstr(&s), &put->flow.key,
1235 !error ? &put->flow.stats : NULL,
1236 put->flow.actions, put->flow.n_actions);
1240 /* There is a tendency to construct odp_flow objects on the stack and to
1241 * forget to properly initialize their "actions" and "n_actions" members.
1242 * When this happens, we get memory corruption because the kernel
1243 * writes through the random pointer that is in the "actions" member.
1245 * This function attempts to combat the problem by:
1247 * - Forcing a segfault if "actions" points to an invalid region (instead
1248 * of just getting back EFAULT, which can be easily missed in the log).
1250 * - Storing a distinctive value that is likely to cause an
1251 * easy-to-identify error later if it is dereferenced, etc.
1253 * - Triggering a warning on uninitialized memory from Valgrind if
1254 * "actions" or "n_actions" was not initialized.
1257 check_rw_odp_flow(struct odp_flow *flow)
1259 if (flow->n_actions) {
1260 memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);