2 * Copyright (c) 2008, 2009, 2010 Nicira Networks.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
18 #include "dpif-provider.h"
28 #include "dynamic-string.h"
32 #include "ofp-print.h"
36 #include "poll-loop.h"
43 VLOG_DEFINE_THIS_MODULE(dpif);
45 COVERAGE_DEFINE(dpif_destroy);
46 COVERAGE_DEFINE(dpif_port_add);
47 COVERAGE_DEFINE(dpif_port_del);
48 COVERAGE_DEFINE(dpif_flow_flush);
49 COVERAGE_DEFINE(dpif_flow_get);
50 COVERAGE_DEFINE(dpif_flow_put);
51 COVERAGE_DEFINE(dpif_flow_del);
52 COVERAGE_DEFINE(dpif_flow_query_list);
53 COVERAGE_DEFINE(dpif_flow_query_list_n);
54 COVERAGE_DEFINE(dpif_execute);
55 COVERAGE_DEFINE(dpif_purge);
57 static const struct dpif_class *base_dpif_classes[] = {
64 struct registered_dpif_class {
65 const struct dpif_class *dpif_class;
68 static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes);
70 /* Rate limit for individual messages going to or from the datapath, output at
71 * DBG level. This is very high because, if these are enabled, it is because
72 * we really need to see them. */
73 static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
75 /* Not really much point in logging many dpif errors. */
76 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
78 static void log_operation(const struct dpif *, const char *operation,
80 static void log_flow_operation(const struct dpif *, const char *operation,
81 int error, struct odp_flow *flow);
82 static void log_flow_put(struct dpif *, int error,
83 const struct odp_flow_put *);
84 static bool should_log_flow_message(int error);
85 static void check_rw_odp_flow(struct odp_flow *);
90 static int status = -1;
96 for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
97 dp_register_provider(base_dpif_classes[i]);
102 /* Performs periodic work needed by all the various kinds of dpifs.
104 * If your program opens any dpifs, it must call both this function and
105 * netdev_run() within its main poll loop. */
109 struct shash_node *node;
110 SHASH_FOR_EACH(node, &dpif_classes) {
111 const struct registered_dpif_class *registered_class = node->data;
112 if (registered_class->dpif_class->run) {
113 registered_class->dpif_class->run();
118 /* Arranges for poll_block() to wake up when dp_run() needs to be called.
120 * If your program opens any dpifs, it must call both this function and
121 * netdev_wait() within its main poll loop. */
125 struct shash_node *node;
126 SHASH_FOR_EACH(node, &dpif_classes) {
127 const struct registered_dpif_class *registered_class = node->data;
128 if (registered_class->dpif_class->wait) {
129 registered_class->dpif_class->wait();
134 /* Registers a new datapath provider. After successful registration, new
135 * datapaths of that type can be opened using dpif_open(). */
137 dp_register_provider(const struct dpif_class *new_class)
139 struct registered_dpif_class *registered_class;
141 if (shash_find(&dpif_classes, new_class->type)) {
142 VLOG_WARN("attempted to register duplicate datapath provider: %s",
147 registered_class = xmalloc(sizeof *registered_class);
148 registered_class->dpif_class = new_class;
149 registered_class->refcount = 0;
151 shash_add(&dpif_classes, new_class->type, registered_class);
156 /* Unregisters a datapath provider. 'type' must have been previously
157 * registered and not currently be in use by any dpifs. After unregistration
158 * new datapaths of that type cannot be opened using dpif_open(). */
160 dp_unregister_provider(const char *type)
162 struct shash_node *node;
163 struct registered_dpif_class *registered_class;
165 node = shash_find(&dpif_classes, type);
167 VLOG_WARN("attempted to unregister a datapath provider that is not "
168 "registered: %s", type);
172 registered_class = node->data;
173 if (registered_class->refcount) {
174 VLOG_WARN("attempted to unregister in use datapath provider: %s", type);
178 shash_delete(&dpif_classes, node);
179 free(registered_class);
184 /* Clears 'types' and enumerates the types of all currently registered datapath
185 * providers into it. The caller must first initialize the svec. */
187 dp_enumerate_types(struct svec *types)
189 struct shash_node *node;
194 SHASH_FOR_EACH(node, &dpif_classes) {
195 const struct registered_dpif_class *registered_class = node->data;
196 svec_add(types, registered_class->dpif_class->type);
200 /* Clears 'names' and enumerates the names of all known created datapaths with
201 * the given 'type'. The caller must first initialize the svec. Returns 0 if
202 * successful, otherwise a positive errno value.
204 * Some kinds of datapaths might not be practically enumerable. This is not
205 * considered an error. */
207 dp_enumerate_names(const char *type, struct svec *names)
209 const struct registered_dpif_class *registered_class;
210 const struct dpif_class *dpif_class;
216 registered_class = shash_find_data(&dpif_classes, type);
217 if (!registered_class) {
218 VLOG_WARN("could not enumerate unknown type: %s", type);
222 dpif_class = registered_class->dpif_class;
223 error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
226 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
233 /* Parses 'datapath name', which is of the form type@name into its
234 * component pieces. 'name' and 'type' must be freed by the caller. */
236 dp_parse_name(const char *datapath_name_, char **name, char **type)
238 char *datapath_name = xstrdup(datapath_name_);
241 separator = strchr(datapath_name, '@');
244 *type = datapath_name;
245 *name = xstrdup(separator + 1);
247 *name = datapath_name;
253 do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
255 struct dpif *dpif = NULL;
257 struct registered_dpif_class *registered_class;
261 if (!type || *type == '\0') {
265 registered_class = shash_find_data(&dpif_classes, type);
266 if (!registered_class) {
267 VLOG_WARN("could not create datapath %s of unknown type %s", name,
269 error = EAFNOSUPPORT;
273 error = registered_class->dpif_class->open(registered_class->dpif_class,
274 name, create, &dpif);
276 assert(dpif->dpif_class == registered_class->dpif_class);
277 registered_class->refcount++;
281 *dpifp = error ? NULL : dpif;
285 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
286 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
287 * the empty string to specify the default system type. Returns 0 if
288 * successful, otherwise a positive errno value. On success stores a pointer
289 * to the datapath in '*dpifp', otherwise a null pointer. */
291 dpif_open(const char *name, const char *type, struct dpif **dpifp)
293 return do_open(name, type, false, dpifp);
296 /* Tries to create and open a new datapath with the given 'name' and 'type'.
297 * 'type' may be either NULL or the empty string to specify the default system
298 * type. Will fail if a datapath with 'name' and 'type' already exists.
299 * Returns 0 if successful, otherwise a positive errno value. On success
300 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
302 dpif_create(const char *name, const char *type, struct dpif **dpifp)
304 return do_open(name, type, true, dpifp);
307 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
308 * does not exist. 'type' may be either NULL or the empty string to specify
309 * the default system type. Returns 0 if successful, otherwise a positive
310 * errno value. On success stores a pointer to the datapath in '*dpifp',
311 * otherwise a null pointer. */
313 dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
317 error = dpif_create(name, type, dpifp);
318 if (error == EEXIST || error == EBUSY) {
319 error = dpif_open(name, type, dpifp);
321 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
322 name, strerror(error));
325 VLOG_WARN("failed to create datapath %s: %s", name, strerror(error));
330 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
331 * itself; call dpif_delete() first, instead, if that is desirable. */
333 dpif_close(struct dpif *dpif)
336 struct registered_dpif_class *registered_class;
338 registered_class = shash_find_data(&dpif_classes,
339 dpif->dpif_class->type);
340 assert(registered_class);
341 assert(registered_class->refcount);
343 registered_class->refcount--;
344 dpif_uninit(dpif, true);
348 /* Returns the name of datapath 'dpif' prefixed with the type
349 * (for use in log messages). */
351 dpif_name(const struct dpif *dpif)
353 return dpif->full_name;
356 /* Returns the name of datapath 'dpif' without the type
357 * (for use in device names). */
359 dpif_base_name(const struct dpif *dpif)
361 return dpif->base_name;
364 /* Enumerates all names that may be used to open 'dpif' into 'all_names'. The
365 * Linux datapath, for example, supports opening a datapath both by number,
366 * e.g. "dp0", and by the name of the datapath's local port. For some
367 * datapaths, this might be an infinite set (e.g. in a file name, slashes may
368 * be duplicated any number of times), in which case only the names most likely
369 * to be used will be enumerated.
371 * The caller must already have initialized 'all_names'. Any existing names in
372 * 'all_names' will not be disturbed. */
374 dpif_get_all_names(const struct dpif *dpif, struct svec *all_names)
376 if (dpif->dpif_class->get_all_names) {
377 int error = dpif->dpif_class->get_all_names(dpif, all_names);
379 VLOG_WARN_RL(&error_rl,
380 "failed to retrieve names for datpath %s: %s",
381 dpif_name(dpif), strerror(error));
385 svec_add(all_names, dpif_base_name(dpif));
391 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
392 * ports. After calling this function, it does not make sense to pass 'dpif'
393 * to any functions other than dpif_name() or dpif_close(). */
395 dpif_delete(struct dpif *dpif)
399 COVERAGE_INC(dpif_destroy);
401 error = dpif->dpif_class->destroy(dpif);
402 log_operation(dpif, "delete", error);
406 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
407 * otherwise a positive errno value. */
409 dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats)
411 int error = dpif->dpif_class->get_stats(dpif, stats);
413 memset(stats, 0, sizeof *stats);
415 log_operation(dpif, "get_stats", error);
419 /* Retrieves the current IP fragment handling policy for 'dpif' into
420 * '*drop_frags': true indicates that fragments are dropped, false indicates
421 * that fragments are treated in the same way as other IP packets (except that
422 * the L4 header cannot be read). Returns 0 if successful, otherwise a
423 * positive errno value. */
425 dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags)
427 int error = dpif->dpif_class->get_drop_frags(dpif, drop_frags);
431 log_operation(dpif, "get_drop_frags", error);
435 /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is
436 * the same as for the get_drop_frags member function. Returns 0 if
437 * successful, otherwise a positive errno value. */
439 dpif_set_drop_frags(struct dpif *dpif, bool drop_frags)
441 int error = dpif->dpif_class->set_drop_frags(dpif, drop_frags);
442 log_operation(dpif, "set_drop_frags", error);
446 /* Attempts to add 'devname' as a port on 'dpif', given the combination of
447 * ODP_PORT_* flags in 'flags'. If successful, returns 0 and sets '*port_nop'
448 * to the new port's port number (if 'port_nop' is non-null). On failure,
449 * returns a positive errno value and sets '*port_nop' to UINT16_MAX (if
450 * 'port_nop' is non-null). */
452 dpif_port_add(struct dpif *dpif, const char *devname, uint16_t flags,
458 COVERAGE_INC(dpif_port_add);
460 error = dpif->dpif_class->port_add(dpif, devname, flags, &port_no);
462 VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16,
463 dpif_name(dpif), devname, port_no);
465 VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
466 dpif_name(dpif), devname, 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);
485 log_operation(dpif, "port_del", error);
489 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
490 * initializes '*port' appropriately; on failure, returns a positive errno
493 dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no,
494 struct odp_port *port)
496 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
498 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s",
499 dpif_name(dpif), port_no, port->devname);
501 memset(port, 0, sizeof *port);
502 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s",
503 dpif_name(dpif), port_no, strerror(error));
508 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
509 * initializes '*port' appropriately; on failure, returns a positive errno
512 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
513 struct odp_port *port)
515 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
517 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16,
518 dpif_name(dpif), devname, port->port);
520 memset(port, 0, sizeof *port);
522 /* Log level is DBG here because all the current callers are interested
523 * in whether 'dpif' actually has a port 'devname', so that it's not an
524 * issue worth logging if it doesn't. */
525 VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s",
526 dpif_name(dpif), devname, strerror(error));
531 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
532 * the port's name into the 'name_size' bytes in 'name', ensuring that the
533 * result is null-terminated. On failure, returns a positive errno value and
534 * makes 'name' the empty string. */
536 dpif_port_get_name(struct dpif *dpif, uint16_t port_no,
537 char *name, size_t name_size)
539 struct odp_port port;
542 assert(name_size > 0);
544 error = dpif_port_query_by_number(dpif, port_no, &port);
546 ovs_strlcpy(name, port.devname, name_size);
553 /* Obtains a list of all the ports in 'dpif'.
555 * If successful, returns 0 and sets '*portsp' to point to an array of
556 * appropriately initialized port structures and '*n_portsp' to the number of
557 * ports in the array. The caller is responsible for freeing '*portp' by
560 * On failure, returns a positive errno value and sets '*portsp' to NULL and
561 * '*n_portsp' to 0. */
563 dpif_port_list(const struct dpif *dpif,
564 struct odp_port **portsp, size_t *n_portsp)
566 struct odp_port *ports;
571 struct odp_stats stats;
574 error = dpif_get_dp_stats(dpif, &stats);
579 ports = xcalloc(stats.n_ports, sizeof *ports);
580 retval = dpif->dpif_class->port_list(dpif, ports, stats.n_ports);
586 } else if (retval <= stats.n_ports) {
592 /* Soft error: port count increased behind our back. Try again. */
605 log_operation(dpif, "port_list", error);
609 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
610 * 'dpif' has changed, this function does one of the following:
612 * - Stores the name of the device that was added to or deleted from 'dpif' in
613 * '*devnamep' and returns 0. The caller is responsible for freeing
614 * '*devnamep' (with free()) when it no longer needs it.
616 * - Returns ENOBUFS and sets '*devnamep' to NULL.
618 * This function may also return 'false positives', where it returns 0 and
619 * '*devnamep' names a device that was not actually added or deleted or it
620 * returns ENOBUFS without any change.
622 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
623 * return other positive errno values to indicate that something has gone
626 dpif_port_poll(const struct dpif *dpif, char **devnamep)
628 int error = dpif->dpif_class->port_poll(dpif, devnamep);
635 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
636 * value other than EAGAIN. */
638 dpif_port_poll_wait(const struct dpif *dpif)
640 dpif->dpif_class->port_poll_wait(dpif);
643 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
644 * positive errno value. */
646 dpif_flow_flush(struct dpif *dpif)
650 COVERAGE_INC(dpif_flow_flush);
652 error = dpif->dpif_class->flow_flush(dpif);
653 log_operation(dpif, "flow_flush", error);
657 /* Queries 'dpif' for a flow entry matching 'flow->key'.
659 * If a flow matching 'flow->key' exists in 'dpif', stores statistics for the
660 * flow into 'flow->stats'. If 'flow->n_actions' is zero, then 'flow->actions'
661 * is ignored. If 'flow->n_actions' is nonzero, then 'flow->actions' should
662 * point to an array of the specified number of actions. At most that many of
663 * the flow's actions will be copied into that array. 'flow->n_actions' will
664 * be updated to the number of actions actually present in the flow, which may
665 * be greater than the number stored if the flow has more actions than space
666 * available in the array.
668 * If no flow matching 'flow->key' exists in 'dpif', returns ENOENT. On other
669 * failure, returns a positive errno value. */
671 dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow)
675 COVERAGE_INC(dpif_flow_get);
677 check_rw_odp_flow(flow);
678 error = dpif->dpif_class->flow_get(dpif, flow, 1);
680 error = flow->stats.error;
683 /* Make the results predictable on error. */
684 memset(&flow->stats, 0, sizeof flow->stats);
687 if (should_log_flow_message(error)) {
688 log_flow_operation(dpif, "flow_get", error, flow);
693 /* For each flow 'flow' in the 'n' flows in 'flows':
695 * - If a flow matching 'flow->key' exists in 'dpif':
697 * Stores 0 into 'flow->stats.error' and stores statistics for the flow
698 * into 'flow->stats'.
700 * If 'flow->n_actions' is zero, then 'flow->actions' is ignored. If
701 * 'flow->n_actions' is nonzero, then 'flow->actions' should point to an
702 * array of the specified number of actions. At most that many of the
703 * flow's actions will be copied into that array. 'flow->n_actions' will
704 * be updated to the number of actions actually present in the flow, which
705 * may be greater than the number stored if the flow has more actions than
706 * space available in the array.
708 * - Flow-specific errors are indicated by a positive errno value in
709 * 'flow->stats.error'. In particular, ENOENT indicates that no flow
710 * matching 'flow->key' exists in 'dpif'. When an error value is stored, the
711 * contents of 'flow->key' are preserved but other members of 'flow' should
712 * be treated as indeterminate.
714 * Returns 0 if all 'n' flows in 'flows' were updated (whether they were
715 * individually successful or not is indicated by 'flow->stats.error',
716 * however). Returns a positive errno value if an error that prevented this
717 * update occurred, in which the caller must not depend on any elements in
718 * 'flows' being updated or not updated.
721 dpif_flow_get_multiple(const struct dpif *dpif,
722 struct odp_flow flows[], size_t n)
727 COVERAGE_ADD(dpif_flow_get, n);
729 for (i = 0; i < n; i++) {
730 check_rw_odp_flow(&flows[i]);
733 error = dpif->dpif_class->flow_get(dpif, flows, n);
734 log_operation(dpif, "flow_get_multiple", error);
738 /* Adds or modifies a flow in 'dpif' as specified in 'put':
740 * - If the flow specified in 'put->flow' does not exist in 'dpif', then
741 * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': if
742 * it is, the flow will be added, otherwise the operation will fail with
745 * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'.
746 * Behavior in this case depends on whether ODPPF_MODIFY is specified in
747 * 'put->flags': if it is, the flow's actions will be updated, otherwise the
748 * operation will fail with EEXIST. If the flow's actions are updated, then
749 * its statistics will be zeroed if ODPPF_ZERO_STATS is set in 'put->flags',
750 * left as-is otherwise.
752 * Returns 0 if successful, otherwise a positive errno value.
755 dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put)
759 COVERAGE_INC(dpif_flow_put);
761 error = dpif->dpif_class->flow_put(dpif, put);
762 if (should_log_flow_message(error)) {
763 log_flow_put(dpif, error, put);
768 /* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if 'dpif'
769 * does not contain such a flow.
771 * If successful, updates 'flow->stats', 'flow->n_actions', and 'flow->actions'
772 * as described for dpif_flow_get(). */
774 dpif_flow_del(struct dpif *dpif, struct odp_flow *flow)
778 COVERAGE_INC(dpif_flow_del);
780 check_rw_odp_flow(flow);
781 memset(&flow->stats, 0, sizeof flow->stats);
783 error = dpif->dpif_class->flow_del(dpif, flow);
784 if (should_log_flow_message(error)) {
785 log_flow_operation(dpif, "delete flow", error, flow);
790 /* Stores up to 'n' flows in 'dpif' into 'flows', including their statistics
791 * but not including any information about their actions. If successful,
792 * returns 0 and sets '*n_out' to the number of flows actually present in
793 * 'dpif', which might be greater than the number stored (if 'dpif' has more
794 * than 'n' flows). On failure, returns a negative errno value and sets
797 dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n,
803 COVERAGE_INC(dpif_flow_query_list);
804 if (RUNNING_ON_VALGRIND) {
805 memset(flows, 0, n * sizeof *flows);
807 for (i = 0; i < n; i++) {
808 flows[i].actions = NULL;
809 flows[i].n_actions = 0;
812 retval = dpif->dpif_class->flow_list(dpif, flows, n);
815 VLOG_WARN_RL(&error_rl, "%s: flow list failed (%s)",
816 dpif_name(dpif), strerror(-retval));
819 COVERAGE_ADD(dpif_flow_query_list_n, retval);
820 *n_out = MIN(n, retval);
821 VLOG_DBG_RL(&dpmsg_rl, "%s: listed %zu flows (of %d)",
822 dpif_name(dpif), *n_out, retval);
827 /* Retrieves all of the flows in 'dpif'.
829 * If successful, returns 0 and stores in '*flowsp' a pointer to a newly
830 * allocated array of flows, including their statistics but not including any
831 * information about their actions, and sets '*np' to the number of flows in
832 * '*flowsp'. The caller is responsible for freeing '*flowsp' by calling
835 * On failure, returns a positive errno value and sets '*flowsp' to NULL and
838 dpif_flow_list_all(const struct dpif *dpif,
839 struct odp_flow **flowsp, size_t *np)
841 struct odp_stats stats;
842 struct odp_flow *flows;
849 error = dpif_get_dp_stats(dpif, &stats);
854 flows = xmalloc(sizeof *flows * stats.n_flows);
855 error = dpif_flow_list(dpif, flows, stats.n_flows, &n_flows);
861 if (stats.n_flows != n_flows) {
862 VLOG_WARN_RL(&error_rl, "%s: datapath stats reported %"PRIu32" "
863 "flows but flow listing reported %zu",
864 dpif_name(dpif), stats.n_flows, n_flows);
871 /* Causes 'dpif' to perform the 'n_actions' actions in 'actions' on the
872 * Ethernet frame specified in 'packet'.
874 * Returns 0 if successful, otherwise a positive errno value. */
876 dpif_execute(struct dpif *dpif,
877 const union odp_action actions[], size_t n_actions,
878 const struct ofpbuf *buf)
882 COVERAGE_INC(dpif_execute);
884 error = dpif->dpif_class->execute(dpif, actions, n_actions, buf);
889 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
890 struct ds ds = DS_EMPTY_INITIALIZER;
891 char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size);
892 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
893 format_odp_actions(&ds, actions, n_actions);
895 ds_put_format(&ds, " failed (%s)", strerror(error));
897 ds_put_format(&ds, " on packet %s", packet);
898 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));
905 /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set
906 * in '*listen_mask' indicates that dpif_recv() will receive messages of that
907 * type. Returns 0 if successful, otherwise a positive errno value. */
909 dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask)
911 int error = dpif->dpif_class->recv_get_mask(dpif, listen_mask);
915 log_operation(dpif, "recv_get_mask", error);
919 /* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in
920 * '*listen_mask' requests that dpif_recv() receive messages of that type.
921 * Returns 0 if successful, otherwise a positive errno value. */
923 dpif_recv_set_mask(struct dpif *dpif, int listen_mask)
925 int error = dpif->dpif_class->recv_set_mask(dpif, listen_mask);
926 log_operation(dpif, "recv_set_mask", error);
930 /* Retrieve the sFlow sampling probability. '*probability' is expressed as the
931 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
932 * the probability of sampling a given packet.
934 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
935 * indicates that 'dpif' does not support sFlow sampling. */
937 dpif_get_sflow_probability(const struct dpif *dpif, uint32_t *probability)
939 int error = (dpif->dpif_class->get_sflow_probability
940 ? dpif->dpif_class->get_sflow_probability(dpif, probability)
945 log_operation(dpif, "get_sflow_probability", error);
949 /* Set the sFlow sampling probability. 'probability' is expressed as the
950 * number of packets out of UINT_MAX to sample, e.g. probability/UINT_MAX is
951 * the probability of sampling a given packet.
953 * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP
954 * indicates that 'dpif' does not support sFlow sampling. */
956 dpif_set_sflow_probability(struct dpif *dpif, uint32_t probability)
958 int error = (dpif->dpif_class->set_sflow_probability
959 ? dpif->dpif_class->set_sflow_probability(dpif, probability)
961 log_operation(dpif, "set_sflow_probability", error);
965 /* Attempts to receive a message from 'dpif'. If successful, stores the
966 * message into '*packetp'. The message, if one is received, will begin with
967 * 'struct odp_msg' as a header, and will have at least DPIF_RECV_MSG_PADDING
968 * bytes of headroom. Only messages of the types selected with
969 * dpif_set_listen_mask() will ordinarily be received (but if a message type is
970 * enabled and then later disabled, some stragglers might pop up).
972 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
973 * if no message is immediately available. */
975 dpif_recv(struct dpif *dpif, struct ofpbuf **packetp)
977 int error = dpif->dpif_class->recv(dpif, packetp);
979 struct ofpbuf *buf = *packetp;
981 assert(ofpbuf_headroom(buf) >= DPIF_RECV_MSG_PADDING);
982 if (VLOG_IS_DBG_ENABLED()) {
983 struct odp_msg *msg = buf->data;
984 void *payload = msg + 1;
985 size_t payload_len = buf->size - sizeof *msg;
986 char *s = ofp_packet_to_string(payload, payload_len, payload_len);
987 VLOG_DBG_RL(&dpmsg_rl, "%s: received %s message of length "
988 "%zu on port %"PRIu16": %s", dpif_name(dpif),
989 (msg->type == _ODPL_MISS_NR ? "miss"
990 : msg->type == _ODPL_ACTION_NR ? "action"
991 : msg->type == _ODPL_SFLOW_NR ? "sFlow"
993 payload_len, msg->port, s);
1002 /* Discards all messages that would otherwise be received by dpif_recv() on
1003 * 'dpif'. Returns 0 if successful, otherwise a positive errno value. */
1005 dpif_recv_purge(struct dpif *dpif)
1007 struct odp_stats stats;
1011 COVERAGE_INC(dpif_purge);
1013 error = dpif_get_dp_stats(dpif, &stats);
1018 for (i = 0; i < stats.max_miss_queue + stats.max_action_queue + stats.max_sflow_queue; i++) {
1020 error = dpif_recv(dpif, &buf);
1022 return error == EAGAIN ? 0 : error;
1029 /* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
1030 * received with dpif_recv(). */
1032 dpif_recv_wait(struct dpif *dpif)
1034 dpif->dpif_class->recv_wait(dpif);
1037 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1038 * and '*engine_id', respectively. */
1040 dpif_get_netflow_ids(const struct dpif *dpif,
1041 uint8_t *engine_type, uint8_t *engine_id)
1043 *engine_type = dpif->netflow_engine_type;
1044 *engine_id = dpif->netflow_engine_id;
1047 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1048 * value for use in the ODPAT_SET_PRIORITY action. On success, returns 0 and
1049 * stores the priority into '*priority'. On failure, returns a positive errno
1050 * value and stores 0 into '*priority'. */
1052 dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1055 int error = (dpif->dpif_class->queue_to_priority
1056 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1062 log_operation(dpif, "queue_to_priority", error);
1067 dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1069 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1071 dpif->dpif_class = dpif_class;
1072 dpif->base_name = xstrdup(name);
1073 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
1074 dpif->netflow_engine_type = netflow_engine_type;
1075 dpif->netflow_engine_id = netflow_engine_id;
1078 /* Undoes the results of initialization.
1080 * Normally this function only needs to be called from dpif_close().
1081 * However, it may be called by providers due to an error on opening
1082 * that occurs after initialization. It this case dpif_close() would
1083 * never be called. */
1085 dpif_uninit(struct dpif *dpif, bool close)
1087 char *base_name = dpif->base_name;
1088 char *full_name = dpif->full_name;
1091 dpif->dpif_class->close(dpif);
1099 log_operation(const struct dpif *dpif, const char *operation, int error)
1102 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
1103 } else if (is_errno(error)) {
1104 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1105 dpif_name(dpif), operation, strerror(error));
1107 VLOG_WARN_RL(&error_rl, "%s: %s failed (%d/%d)",
1108 dpif_name(dpif), operation,
1109 get_ofp_err_type(error), get_ofp_err_code(error));
1113 static enum vlog_level
1114 flow_message_log_level(int error)
1116 return error ? VLL_WARN : VLL_DBG;
1120 should_log_flow_message(int error)
1122 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1123 error ? &error_rl : &dpmsg_rl);
1127 log_flow_message(const struct dpif *dpif, int error, const char *operation,
1128 const struct odp_flow_key *flow,
1129 const struct odp_flow_stats *stats,
1130 const union odp_action *actions, size_t n_actions)
1132 struct ds ds = DS_EMPTY_INITIALIZER;
1133 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1135 ds_put_cstr(&ds, "failed to ");
1137 ds_put_format(&ds, "%s ", operation);
1139 ds_put_format(&ds, "(%s) ", strerror(error));
1141 format_odp_flow_key(&ds, flow);
1143 ds_put_cstr(&ds, ", ");
1144 format_odp_flow_stats(&ds, stats);
1146 if (actions || n_actions) {
1147 ds_put_cstr(&ds, ", actions:");
1148 format_odp_actions(&ds, actions, n_actions);
1150 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1155 log_flow_operation(const struct dpif *dpif, const char *operation, int error,
1156 struct odp_flow *flow)
1159 flow->n_actions = 0;
1161 log_flow_message(dpif, error, operation, &flow->key,
1162 !error ? &flow->stats : NULL,
1163 flow->actions, flow->n_actions);
1167 log_flow_put(struct dpif *dpif, int error, const struct odp_flow_put *put)
1169 enum { ODPPF_ALL = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS };
1173 ds_put_cstr(&s, "put");
1174 if (put->flags & ODPPF_CREATE) {
1175 ds_put_cstr(&s, "[create]");
1177 if (put->flags & ODPPF_MODIFY) {
1178 ds_put_cstr(&s, "[modify]");
1180 if (put->flags & ODPPF_ZERO_STATS) {
1181 ds_put_cstr(&s, "[zero]");
1183 if (put->flags & ~ODPPF_ALL) {
1184 ds_put_format(&s, "[%x]", put->flags & ~ODPPF_ALL);
1186 log_flow_message(dpif, error, ds_cstr(&s), &put->flow.key,
1187 !error ? &put->flow.stats : NULL,
1188 put->flow.actions, put->flow.n_actions);
1192 /* There is a tendency to construct odp_flow objects on the stack and to
1193 * forget to properly initialize their "actions" and "n_actions" members.
1194 * When this happens, we get memory corruption because the kernel
1195 * writes through the random pointer that is in the "actions" member.
1197 * This function attempts to combat the problem by:
1199 * - Forcing a segfault if "actions" points to an invalid region (instead
1200 * of just getting back EFAULT, which can be easily missed in the log).
1202 * - Storing a distinctive value that is likely to cause an
1203 * easy-to-identify error later if it is dereferenced, etc.
1205 * - Triggering a warning on uninitialized memory from Valgrind if
1206 * "actions" or "n_actions" was not initialized.
1209 check_rw_odp_flow(struct odp_flow *flow)
1211 if (flow->n_actions) {
1212 memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]);