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"
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_odp_flow(struct odp_flow *);
91 static int status = -1;
97 for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
98 dp_register_provider(base_dpif_classes[i]);
103 /* Performs periodic work needed by all the various kinds of dpifs.
105 * If your program opens any dpifs, it must call both this function and
106 * netdev_run() within its main poll loop. */
110 struct shash_node *node;
111 SHASH_FOR_EACH(node, &dpif_classes) {
112 const struct registered_dpif_class *registered_class = node->data;
113 if (registered_class->dpif_class->run) {
114 registered_class->dpif_class->run();
119 /* Arranges for poll_block() to wake up when dp_run() needs to be called.
121 * If your program opens any dpifs, it must call both this function and
122 * netdev_wait() within its main poll loop. */
126 struct shash_node *node;
127 SHASH_FOR_EACH(node, &dpif_classes) {
128 const struct registered_dpif_class *registered_class = node->data;
129 if (registered_class->dpif_class->wait) {
130 registered_class->dpif_class->wait();
135 /* Registers a new datapath provider. After successful registration, new
136 * datapaths of that type can be opened using dpif_open(). */
138 dp_register_provider(const struct dpif_class *new_class)
140 struct registered_dpif_class *registered_class;
142 if (shash_find(&dpif_classes, new_class->type)) {
143 VLOG_WARN("attempted to register duplicate datapath provider: %s",
148 registered_class = xmalloc(sizeof *registered_class);
149 registered_class->dpif_class = new_class;
150 registered_class->refcount = 0;
152 shash_add(&dpif_classes, new_class->type, registered_class);
157 /* Unregisters a datapath provider. 'type' must have been previously
158 * registered and not currently be in use by any dpifs. After unregistration
159 * new datapaths of that type cannot be opened using dpif_open(). */
161 dp_unregister_provider(const char *type)
163 struct shash_node *node;
164 struct registered_dpif_class *registered_class;
166 node = shash_find(&dpif_classes, type);
168 VLOG_WARN("attempted to unregister a datapath provider that is not "
169 "registered: %s", type);
173 registered_class = node->data;
174 if (registered_class->refcount) {
175 VLOG_WARN("attempted to unregister in use datapath provider: %s", type);
179 shash_delete(&dpif_classes, node);
180 free(registered_class);
185 /* Clears 'types' and enumerates the types of all currently registered datapath
186 * providers into it. The caller must first initialize the svec. */
188 dp_enumerate_types(struct svec *types)
190 struct shash_node *node;
195 SHASH_FOR_EACH(node, &dpif_classes) {
196 const struct registered_dpif_class *registered_class = node->data;
197 svec_add(types, registered_class->dpif_class->type);
201 /* Clears 'names' and enumerates the names of all known created datapaths with
202 * the given 'type'. The caller must first initialize the svec. Returns 0 if
203 * successful, otherwise a positive errno value.
205 * Some kinds of datapaths might not be practically enumerable. This is not
206 * considered an error. */
208 dp_enumerate_names(const char *type, struct svec *names)
210 const struct registered_dpif_class *registered_class;
211 const struct dpif_class *dpif_class;
217 registered_class = shash_find_data(&dpif_classes, type);
218 if (!registered_class) {
219 VLOG_WARN("could not enumerate unknown type: %s", type);
223 dpif_class = registered_class->dpif_class;
224 error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
227 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
234 /* Parses 'datapath name', which is of the form type@name into its
235 * component pieces. 'name' and 'type' must be freed by the caller. */
237 dp_parse_name(const char *datapath_name_, char **name, char **type)
239 char *datapath_name = xstrdup(datapath_name_);
242 separator = strchr(datapath_name, '@');
245 *type = datapath_name;
246 *name = xstrdup(separator + 1);
248 *name = datapath_name;
254 do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
256 struct dpif *dpif = NULL;
258 struct registered_dpif_class *registered_class;
262 if (!type || *type == '\0') {
266 registered_class = shash_find_data(&dpif_classes, type);
267 if (!registered_class) {
268 VLOG_WARN("could not create datapath %s of unknown type %s", name,
270 error = EAFNOSUPPORT;
274 error = registered_class->dpif_class->open(registered_class->dpif_class,
275 name, create, &dpif);
277 assert(dpif->dpif_class == registered_class->dpif_class);
278 registered_class->refcount++;
282 *dpifp = error ? NULL : dpif;
286 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
287 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
288 * the empty string to specify the default system type. Returns 0 if
289 * successful, otherwise a positive errno value. On success stores a pointer
290 * to the datapath in '*dpifp', otherwise a null pointer. */
292 dpif_open(const char *name, const char *type, struct dpif **dpifp)
294 return do_open(name, type, false, dpifp);
297 /* Tries to create and open a new datapath with the given 'name' and 'type'.
298 * 'type' may be either NULL or the empty string to specify the default system
299 * type. Will fail if a datapath with 'name' and 'type' already exists.
300 * Returns 0 if successful, otherwise a positive errno value. On success
301 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
303 dpif_create(const char *name, const char *type, struct dpif **dpifp)
305 return do_open(name, type, true, dpifp);
308 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
309 * does not exist. 'type' may be either NULL or the empty string to specify
310 * the default system type. Returns 0 if successful, otherwise a positive
311 * errno value. On success stores a pointer to the datapath in '*dpifp',
312 * otherwise a null pointer. */
314 dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
318 error = dpif_create(name, type, dpifp);
319 if (error == EEXIST || error == EBUSY) {
320 error = dpif_open(name, type, dpifp);
322 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
323 name, strerror(error));
326 VLOG_WARN("failed to create datapath %s: %s", name, strerror(error));
331 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
332 * itself; call dpif_delete() first, instead, if that is desirable. */
334 dpif_close(struct dpif *dpif)
337 struct registered_dpif_class *registered_class;
339 registered_class = shash_find_data(&dpif_classes,
340 dpif->dpif_class->type);
341 assert(registered_class);
342 assert(registered_class->refcount);
344 registered_class->refcount--;
345 dpif_uninit(dpif, true);
349 /* Returns the name of datapath 'dpif' prefixed with the type
350 * (for use in log messages). */
352 dpif_name(const struct dpif *dpif)
354 return dpif->full_name;
357 /* Returns the name of datapath 'dpif' without the type
358 * (for use in device names). */
360 dpif_base_name(const struct dpif *dpif)
362 return dpif->base_name;
365 /* Enumerates all names that may be used to open 'dpif' into 'all_names'. The
366 * Linux datapath, for example, supports opening a datapath both by number,
367 * e.g. "dp0", and by the name of the datapath's local port. For some
368 * datapaths, this might be an infinite set (e.g. in a file name, slashes may
369 * be duplicated any number of times), in which case only the names most likely
370 * to be used will be enumerated.
372 * The caller must already have initialized 'all_names'. Any existing names in
373 * 'all_names' will not be disturbed. */
375 dpif_get_all_names(const struct dpif *dpif, struct svec *all_names)
377 if (dpif->dpif_class->get_all_names) {
378 int error = dpif->dpif_class->get_all_names(dpif, all_names);
380 VLOG_WARN_RL(&error_rl,
381 "failed to retrieve names for datpath %s: %s",
382 dpif_name(dpif), strerror(error));
386 svec_add(all_names, dpif_base_name(dpif));
392 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
393 * ports. After calling this function, it does not make sense to pass 'dpif'
394 * to any functions other than dpif_name() or dpif_close(). */
396 dpif_delete(struct dpif *dpif)
400 COVERAGE_INC(dpif_destroy);
402 error = dpif->dpif_class->destroy(dpif);
403 log_operation(dpif, "delete", error);
407 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
408 * otherwise a positive errno value. */
410 dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats)
412 int error = dpif->dpif_class->get_stats(dpif, stats);
414 memset(stats, 0, sizeof *stats);
416 log_operation(dpif, "get_stats", error);
420 /* Retrieves the current IP fragment handling policy for 'dpif' into
421 * '*drop_frags': true indicates that fragments are dropped, false indicates
422 * that fragments are treated in the same way as other IP packets (except that
423 * the L4 header cannot be read). Returns 0 if successful, otherwise a
424 * positive errno value. */
426 dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags)
428 int error = dpif->dpif_class->get_drop_frags(dpif, drop_frags);
432 log_operation(dpif, "get_drop_frags", error);
436 /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is
437 * the same as for the get_drop_frags member function. Returns 0 if
438 * successful, otherwise a positive errno value. */
440 dpif_set_drop_frags(struct dpif *dpif, bool drop_frags)
442 int error = dpif->dpif_class->set_drop_frags(dpif, drop_frags);
443 log_operation(dpif, "set_drop_frags", error);
447 /* Attempts to add 'netdev' as a port on 'dpif'. If successful, returns 0 and
448 * sets '*port_nop' to the new port's port number (if 'port_nop' is non-null).
449 * On failure, returns a positive errno value and sets '*port_nop' to
450 * UINT16_MAX (if 'port_nop' is non-null). */
452 dpif_port_add(struct dpif *dpif, struct netdev *netdev, uint16_t *port_nop)
454 const char *netdev_name = netdev_get_name(netdev);
458 COVERAGE_INC(dpif_port_add);
460 error = dpif->dpif_class->port_add(dpif, netdev, &port_no);
462 VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16,
463 dpif_name(dpif), netdev_name, port_no);
465 VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
466 dpif_name(dpif), netdev_name, strerror(error));
467 port_no = UINT16_MAX;
475 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
476 * otherwise a positive errno value. */
478 dpif_port_del(struct dpif *dpif, uint16_t port_no)
482 COVERAGE_INC(dpif_port_del);
484 error = dpif->dpif_class->port_del(dpif, port_no);
486 VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu16")",
487 dpif_name(dpif), port_no);
489 log_operation(dpif, "port_del", error);
494 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
495 * initializes '*port' appropriately; on failure, returns a positive errno
498 dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
499 struct odp_port *port)
501 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
503 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s",
504 dpif_name(dpif), port_no, port->devname);
506 memset(port, 0, sizeof *port);
507 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s",
508 dpif_name(dpif), port_no, strerror(error));
513 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
514 * initializes '*port' appropriately; on failure, returns a positive errno
517 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
518 struct odp_port *port)
520 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
522 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16,
523 dpif_name(dpif), devname, port->port);
525 memset(port, 0, sizeof *port);
527 /* Log level is DBG here because all the current callers are interested
528 * in whether 'dpif' actually has a port 'devname', so that it's not an
529 * issue worth logging if it doesn't. */
530 VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s",
531 dpif_name(dpif), devname, strerror(error));
536 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
537 * the port's name into the 'name_size' bytes in 'name', ensuring that the
538 * result is null-terminated. On failure, returns a positive errno value and
539 * makes 'name' the empty string. */
541 dpif_port_get_name(struct dpif *dpif, uint16_t port_no,
542 char *name, size_t name_size)
544 struct odp_port port;
547 assert(name_size > 0);
549 error = dpif_port_query_by_number(dpif, port_no, &port);
551 ovs_strlcpy(name, port.devname, name_size);
558 /* Obtains a list of all the ports in 'dpif'.
560 * If successful, returns 0 and sets '*portsp' to point to an array of
561 * appropriately initialized port structures and '*n_portsp' to the number of
562 * ports in the array. The caller is responsible for freeing '*portp' by
565 * On failure, returns a positive errno value and sets '*portsp' to NULL and
566 * '*n_portsp' to 0. */
568 dpif_port_list(const struct dpif *dpif,
569 struct odp_port **portsp, size_t *n_portsp)
571 struct odp_port *ports;
576 struct odp_stats stats;
579 error = dpif_get_dp_stats(dpif, &stats);
584 ports = xcalloc(stats.n_ports, sizeof *ports);
585 retval = dpif->dpif_class->port_list(dpif, ports, stats.n_ports);
591 } else if (retval <= stats.n_ports) {
597 /* Soft error: port count increased behind our back. Try again. */
610 log_operation(dpif, "port_list", error);
614 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
615 * 'dpif' has changed, this function does one of the following:
617 * - Stores the name of the device that was added to or deleted from 'dpif' in
618 * '*devnamep' and returns 0. The caller is responsible for freeing
619 * '*devnamep' (with free()) when it no longer needs it.
621 * - Returns ENOBUFS and sets '*devnamep' to NULL.
623 * This function may also return 'false positives', where it returns 0 and
624 * '*devnamep' names a device that was not actually added or deleted or it
625 * returns ENOBUFS without any change.
627 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
628 * return other positive errno values to indicate that something has gone
631 dpif_port_poll(const struct dpif *dpif, char **devnamep)
633 int error = dpif->dpif_class->port_poll(dpif, devnamep);
640 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
641 * value other than EAGAIN. */
643 dpif_port_poll_wait(const struct dpif *dpif)
645 dpif->dpif_class->port_poll_wait(dpif);
648 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
649 * positive errno value. */
651 dpif_flow_flush(struct dpif *dpif)
655 COVERAGE_INC(dpif_flow_flush);
657 error = dpif->dpif_class->flow_flush(dpif);
658 log_operation(dpif, "flow_flush", error);
662 /* Queries 'dpif' for a flow entry matching 'flow->key'.
664 * If a flow matching 'flow->key' exists in 'dpif', stores statistics for the
665 * flow into 'flow->stats'. If 'flow->actions_len' is zero, then
666 * 'flow->actions' is ignored. If 'flow->actions_len' is nonzero, then
667 * 'flow->actions' should point to an array of the specified number of bytes.
668 * At most that many bytes of the flow's actions will be copied into that
669 * array. 'flow->actions_len' will be updated to the number of bytes of
670 * actions actually present in the flow, which may be greater than the amount
671 * stored if the flow has more actions than space available in the array.
673 * If no flow matching 'flow->key' exists in 'dpif', returns ENOENT. On other
674 * failure, returns a positive errno value. */
676 dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
680 COVERAGE_INC(dpif_flow_get);
682 check_rw_odp_flow(flow);
683 error = dpif->dpif_class->flow_get(dpif, flow, 1);
685 error = flow->stats.error;
688 /* Make the results predictable on error. */
689 memset(&flow->stats, 0, sizeof flow->stats);
690 flow->actions_len = 0;
692 if (should_log_flow_message(error)) {
693 log_flow_operation(dpif, "flow_get", error, flow);
698 /* For each flow 'flow' in the 'n' flows in 'flows':
700 * - If a flow matching 'flow->key' exists in 'dpif':
702 * Stores 0 into 'flow->stats.error' and stores statistics for the flow
703 * into 'flow->stats'.
705 * If 'flow->actions_len' is zero, then 'flow->actions' is ignored. If
706 * 'flow->actions_len' is nonzero, then 'flow->actions' should point to an
707 * array of the specified number of bytes. At most that amount of flow's
708 * actions will be copied into that array. 'flow->actions_len' will be
709 * updated to the number of bytes of actions actually present in the flow,
710 * which may be greater than the amount stored if the flow's actions are
711 * longer than the available space.
713 * - Flow-specific errors are indicated by a positive errno value in
714 * 'flow->stats.error'. In particular, ENOENT indicates that no flow
715 * matching 'flow->key' exists in 'dpif'. When an error value is stored, the
716 * contents of 'flow->key' are preserved but other members of 'flow' should
717 * be treated as indeterminate.
719 * Returns 0 if all 'n' flows in 'flows' were updated (whether they were
720 * individually successful or not is indicated by 'flow->stats.error',
721 * however). Returns a positive errno value if an error that prevented this
722 * update occurred, in which the caller must not depend on any elements in
723 * 'flows' being updated or not updated.
726 dpif_flow_get_multiple(const struct dpif *dpif,
727 struct odp_flow flows[], size_t n)
732 COVERAGE_ADD(dpif_flow_get, n);
734 for (i = 0; i < n; i++) {
735 check_rw_odp_flow(&flows[i]);
738 error = dpif->dpif_class->flow_get(dpif, flows, n);
739 log_operation(dpif, "flow_get_multiple", error);
743 /* Adds or modifies a flow in 'dpif' as specified in 'put':
745 * - If the flow specified in 'put->flow' does not exist in 'dpif', then
746 * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': if
747 * it is, the flow will be added, otherwise the operation will fail with
750 * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'.
751 * Behavior in this case depends on whether ODPPF_MODIFY is specified in
752 * 'put->flags': if it is, the flow's actions will be updated, otherwise the
753 * operation will fail with EEXIST. If the flow's actions are updated, then
754 * its statistics will be zeroed if ODPPF_ZERO_STATS is set in 'put->flags',
755 * left as-is otherwise.
757 * Returns 0 if successful, otherwise a positive errno value.
760 dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
764 COVERAGE_INC(dpif_flow_put);
766 error = dpif->dpif_class->flow_put(dpif, put);
767 if (should_log_flow_message(error)) {
768 log_flow_put(dpif, error, put);
773 /* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if 'dpif'
774 * does not contain such a flow.
776 * If successful, updates 'flow->stats', 'flow->actions_len', and
777 * 'flow->actions' as described for dpif_flow_get(). */
779 dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
783 COVERAGE_INC(dpif_flow_del);
785 check_rw_odp_flow(flow);
786 memset(&flow->stats, 0, sizeof flow->stats);
788 error = dpif->dpif_class->flow_del(dpif, flow);
789 if (should_log_flow_message(error)) {
790 log_flow_operation(dpif, "delete flow", error, flow);
795 /* Stores up to 'n' flows in 'dpif' into 'flows', including their statistics
796 * but not including any information about their actions. If successful,
797 * returns 0 and sets '*n_out' to the number of flows actually present in
798 * 'dpif', which might be greater than the number stored (if 'dpif' has more
799 * than 'n' flows). On failure, returns a negative errno value and sets
802 dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n,
808 COVERAGE_INC(dpif_flow_query_list);
809 if (RUNNING_ON_VALGRIND) {
810 memset(flows, 0, n * sizeof *flows);
812 for (i = 0; i < n; i++) {
813 flows[i].actions = NULL;
814 flows[i].actions_len = 0;
817 retval = dpif->dpif_class->flow_list(dpif, flows, n);
820 VLOG_WARN_RL(&error_rl, "%s: flow list failed (%s)",
821 dpif_name(dpif), strerror(-retval));
824 COVERAGE_ADD(dpif_flow_query_list_n, retval);
825 *n_out = MIN(n, retval);
826 VLOG_DBG_RL(&dpmsg_rl, "%s: listed %zu flows (of %d)",
827 dpif_name(dpif), *n_out, retval);
832 /* Retrieves all of the flows in 'dpif'.
834 * If successful, returns 0 and stores in '*flowsp' a pointer to a newly
835 * allocated array of flows, including their statistics but not including any
836 * information about their actions, and sets '*np' to the number of flows in
837 * '*flowsp'. The caller is responsible for freeing '*flowsp' by calling
840 * On failure, returns a positive errno value and sets '*flowsp' to NULL and
843 dpif_flow_list_all(const struct dpif *dpif,
844 struct odp_flow **flowsp, size_t *np)
846 struct odp_stats stats;
847 struct odp_flow *flows;
854 error = dpif_get_dp_stats(dpif, &stats);
859 flows = xmalloc(sizeof *flows * stats.n_flows);
860 error = dpif_flow_list(dpif, flows, stats.n_flows, &n_flows);
866 if (stats.n_flows != n_flows) {
867 VLOG_WARN_RL(&error_rl, "%s: datapath stats reported %"PRIu32" "
868 "flows but flow listing reported %zu",
869 dpif_name(dpif), stats.n_flows, n_flows);
876 /* Causes 'dpif' to perform the 'actions_len' bytes of actions in 'actions' on
877 * the Ethernet frame specified in 'packet'.
879 * Returns 0 if successful, otherwise a positive errno value. */
881 dpif_execute(struct dpif *dpif,
882 const struct nlattr *actions, size_t actions_len,
883 const struct ofpbuf *buf)
887 COVERAGE_INC(dpif_execute);
888 if (actions_len > 0) {
889 error = dpif->dpif_class->execute(dpif, actions, actions_len, buf);
894 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
895 struct ds ds = DS_EMPTY_INITIALIZER;
896 char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
897 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
898 format_odp_actions(&ds, actions, actions_len);
900 ds_put_format(&ds, " failed (%s)", strerror(error));
902 ds_put_format(&ds, " on packet %s", packet);
903 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
910 /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set
911 * in '*listen_mask' indicates that dpif_recv() will receive messages of that
912 * type. Returns 0 if successful, otherwise a positive errno value. */
914 dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask)
916 int error = dpif->dpif_class->recv_get_mask(dpif, listen_mask);
920 log_operation(dpif, "recv_get_mask", error);
924 /* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in
925 * '*listen_mask' requests that dpif_recv() receive messages of that type.
926 * Returns 0 if successful, otherwise a positive errno value. */
928 dpif_recv_set_mask(struct dpif *dpif, int listen_mask)
930 int error = dpif->dpif_class->recv_set_mask(dpif, listen_mask);
931 log_operation(dpif, "recv_set_mask", error);
935 /* Retrieve the sFlow sampling probability. '*probability' is expressed as the
936 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
937 * the probability of sampling a given packet.
939 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
940 * indicates that 'dpif' does not support sFlow sampling. */
942 dpif_get_sflow_probability(const struct dpif *dpif, uint32_t *probability)
944 int error = (dpif->dpif_class->get_sflow_probability
945 ? dpif->dpif_class->get_sflow_probability(dpif, probability)
950 log_operation(dpif, "get_sflow_probability", error);
954 /* Set the sFlow sampling probability. 'probability' is expressed as the
955 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
956 * the probability of sampling a given packet.
958 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
959 * indicates that 'dpif' does not support sFlow sampling. */
961 dpif_set_sflow_probability(struct dpif *dpif, uint32_t probability)
963 int error = (dpif->dpif_class->set_sflow_probability
964 ? dpif->dpif_class->set_sflow_probability(dpif, probability)
966 log_operation(dpif, "set_sflow_probability", error);
970 /* Attempts to receive a message from 'dpif'. If successful, stores the
971 * message into '*packetp'. The message, if one is received, will begin with
972 * 'struct odp_msg' as a header, and will have at least DPIF_RECV_MSG_PADDING
973 * bytes of headroom. Only messages of the types selected with
974 * dpif_set_listen_mask() will ordinarily be received (but if a message type is
975 * enabled and then later disabled, some stragglers might pop up).
977 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
978 * if no message is immediately available. */
980 dpif_recv(struct dpif *dpif, struct ofpbuf **packetp)
982 int error = dpif->dpif_class->recv(dpif, packetp);
984 struct ofpbuf *buf = *packetp;
986 assert(ofpbuf_headroom(buf) >= DPIF_RECV_MSG_PADDING);
987 if (VLOG_IS_DBG_ENABLED()) {
988 struct odp_msg *msg = buf->data;
989 void *payload = msg + 1;
990 size_t payload_len = buf->size - sizeof *msg;
991 char *s = ofp_packet_to_string(payload, payload_len, payload_len);
992 VLOG_DBG_RL(&dpmsg_rl, "%s: received %s message of length "
993 "%zu on port %"PRIu16": %s", dpif_name(dpif),
994 (msg->type == _ODPL_MISS_NR ? "miss"
995 : msg->type == _ODPL_ACTION_NR ? "action"
996 : msg->type == _ODPL_SFLOW_NR ? "sFlow"
998 payload_len, msg->port, s);
1007 /* Discards all messages that would otherwise be received by dpif_recv() on
1008 * 'dpif'. Returns 0 if successful, otherwise a positive errno value. */
1010 dpif_recv_purge(struct dpif *dpif)
1012 struct odp_stats stats;
1016 COVERAGE_INC(dpif_purge);
1018 error = dpif_get_dp_stats(dpif, &stats);
1023 for (i = 0; i < stats.max_miss_queue + stats.max_action_queue + stats.max_sflow_queue; i++) {
1025 error = dpif_recv(dpif, &buf);
1027 return error == EAGAIN ? 0 : error;
1034 /* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
1035 * received with dpif_recv(). */
1037 dpif_recv_wait(struct dpif *dpif)
1039 dpif->dpif_class->recv_wait(dpif);
1042 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1043 * and '*engine_id', respectively. */
1045 dpif_get_netflow_ids(const struct dpif *dpif,
1046 uint8_t *engine_type, uint8_t *engine_id)
1048 *engine_type = dpif->netflow_engine_type;
1049 *engine_id = dpif->netflow_engine_id;
1052 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1053 * value for use in the ODPAT_SET_PRIORITY action. On success, returns 0 and
1054 * stores the priority into '*priority'. On failure, returns a positive errno
1055 * value and stores 0 into '*priority'. */
1057 dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1060 int error = (dpif->dpif_class->queue_to_priority
1061 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1067 log_operation(dpif, "queue_to_priority", error);
1072 dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1074 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1076 dpif->dpif_class = dpif_class;
1077 dpif->base_name = xstrdup(name);
1078 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
1079 dpif->netflow_engine_type = netflow_engine_type;
1080 dpif->netflow_engine_id = netflow_engine_id;
1083 /* Undoes the results of initialization.
1085 * Normally this function only needs to be called from dpif_close().
1086 * However, it may be called by providers due to an error on opening
1087 * that occurs after initialization. It this case dpif_close() would
1088 * never be called. */
1090 dpif_uninit(struct dpif *dpif, bool close)
1092 char *base_name = dpif->base_name;
1093 char *full_name = dpif->full_name;
1096 dpif->dpif_class->close(dpif);
1104 log_operation(const struct dpif *dpif, const char *operation, int error)
1107 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
1108 } else if (is_errno(error)) {
1109 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1110 dpif_name(dpif), operation, strerror(error));
1112 VLOG_WARN_RL(&error_rl, "%s: %s failed (%d/%d)",
1113 dpif_name(dpif), operation,
1114 get_ofp_err_type(error), get_ofp_err_code(error));
1118 static enum vlog_level
1119 flow_message_log_level(int error)
1121 return error ? VLL_WARN : VLL_DBG;
1125 should_log_flow_message(int error)
1127 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1128 error ? &error_rl : &dpmsg_rl);
1132 log_flow_message(const struct dpif *dpif, int error, const char *operation,
1133 const struct odp_flow_key *flow,
1134 const struct odp_flow_stats *stats,
1135 const struct nlattr *actions, size_t actions_len)
1137 struct ds ds = DS_EMPTY_INITIALIZER;
1138 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1140 ds_put_cstr(&ds, "failed to ");
1142 ds_put_format(&ds, "%s ", operation);
1144 ds_put_format(&ds, "(%s) ", strerror(error));
1146 format_odp_flow_key(&ds, flow);
1148 ds_put_cstr(&ds, ", ");
1149 format_odp_flow_stats(&ds, stats);
1151 if (actions || actions_len) {
1152 ds_put_cstr(&ds, ", actions:");
1153 format_odp_actions(&ds, actions, actions_len);
1155 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1160 log_flow_operation(const struct dpif *dpif, const char *operation, int error,
1161 struct odp_flow *flow)
1164 flow->actions_len = 0;
1166 log_flow_message(dpif, error, operation, &flow->key,
1167 !error ? &flow->stats : NULL,
1168 flow->actions, flow->actions_len);
1172 log_flow_put(struct dpif *dpif, int error, const struct odp_flow_put *put)
1174 enum { ODPPF_ALL = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS };
1178 ds_put_cstr(&s, "put");
1179 if (put->flags & ODPPF_CREATE) {
1180 ds_put_cstr(&s, "[create]");
1182 if (put->flags & ODPPF_MODIFY) {
1183 ds_put_cstr(&s, "[modify]");
1185 if (put->flags & ODPPF_ZERO_STATS) {
1186 ds_put_cstr(&s, "[zero]");
1188 if (put->flags & ~ODPPF_ALL) {
1189 ds_put_format(&s, "[%x]", put->flags & ~ODPPF_ALL);
1191 log_flow_message(dpif, error, ds_cstr(&s), &put->flow.key,
1192 !error ? &put->flow.stats : NULL,
1193 put->flow.actions, put->flow.actions_len);
1197 /* There is a tendency to construct odp_flow objects on the stack and to
1198 * forget to properly initialize their "actions" and "actions_len" members.
1199 * When this happens, we get memory corruption because the kernel
1200 * writes through the random pointer that is in the "actions" member.
1202 * This function attempts to combat the problem by:
1204 * - Forcing a segfault if "actions" points to an invalid region (instead
1205 * of just getting back EFAULT, which can be easily missed in the log).
1207 * - Storing a distinctive value that is likely to cause an
1208 * easy-to-identify error later if it is dereferenced, etc.
1210 * - Triggering a warning on uninitialized memory from Valgrind if
1211 * "actions" or "actions_len" was not initialized.
1214 check_rw_odp_flow(struct odp_flow *flow)
1216 if (flow->actions_len) {
1217 memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);