2 * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
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"
27 #include "dynamic-string.h"
31 #include "odp-execute.h"
33 #include "ofp-errors.h"
34 #include "ofp-print.h"
38 #include "poll-loop.h"
46 VLOG_DEFINE_THIS_MODULE(dpif);
48 COVERAGE_DEFINE(dpif_destroy);
49 COVERAGE_DEFINE(dpif_port_add);
50 COVERAGE_DEFINE(dpif_port_del);
51 COVERAGE_DEFINE(dpif_flow_flush);
52 COVERAGE_DEFINE(dpif_flow_get);
53 COVERAGE_DEFINE(dpif_flow_put);
54 COVERAGE_DEFINE(dpif_flow_del);
55 COVERAGE_DEFINE(dpif_execute);
56 COVERAGE_DEFINE(dpif_purge);
57 COVERAGE_DEFINE(dpif_execute_with_help);
59 static const struct dpif_class *base_dpif_classes[] = {
64 &dpif_planetlab_class,
67 struct registered_dpif_class {
68 const struct dpif_class *dpif_class;
71 static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes);
72 static struct sset dpif_blacklist = SSET_INITIALIZER(&dpif_blacklist);
74 /* Protects 'dpif_classes', including the refcount, and 'dpif_blacklist'. */
75 static struct ovs_mutex dpif_mutex = OVS_MUTEX_INITIALIZER;
77 /* Rate limit for individual messages going to or from the datapath, output at
78 * DBG level. This is very high because, if these are enabled, it is because
79 * we really need to see them. */
80 static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
82 /* Not really much point in logging many dpif errors. */
83 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
85 static void log_flow_message(const struct dpif *dpif, int error,
86 const char *operation,
87 const struct nlattr *key, size_t key_len,
88 const struct nlattr *mask, size_t mask_len,
89 const struct dpif_flow_stats *stats,
90 const struct nlattr *actions, size_t actions_len);
91 static void log_operation(const struct dpif *, const char *operation,
93 static bool should_log_flow_message(int error);
94 static void log_flow_put_message(struct dpif *, const struct dpif_flow_put *,
96 static void log_flow_del_message(struct dpif *, const struct dpif_flow_del *,
98 static void log_execute_message(struct dpif *, const struct dpif_execute *,
104 static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER;
106 if (ovsthread_once_start(&once)) {
109 for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
110 dp_register_provider(base_dpif_classes[i]);
112 ovsthread_once_done(&once);
117 dp_register_provider__(const struct dpif_class *new_class)
119 struct registered_dpif_class *registered_class;
121 if (sset_contains(&dpif_blacklist, new_class->type)) {
122 VLOG_DBG("attempted to register blacklisted provider: %s",
127 if (shash_find(&dpif_classes, new_class->type)) {
128 VLOG_WARN("attempted to register duplicate datapath provider: %s",
133 registered_class = xmalloc(sizeof *registered_class);
134 registered_class->dpif_class = new_class;
135 registered_class->refcount = 0;
137 shash_add(&dpif_classes, new_class->type, registered_class);
142 /* Registers a new datapath provider. After successful registration, new
143 * datapaths of that type can be opened using dpif_open(). */
145 dp_register_provider(const struct dpif_class *new_class)
149 ovs_mutex_lock(&dpif_mutex);
150 error = dp_register_provider__(new_class);
151 ovs_mutex_unlock(&dpif_mutex);
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 /* Unregisters a datapath provider. 'type' must have been previously
185 * registered and not currently be in use by any dpifs. After unregistration
186 * new datapaths of that type cannot be opened using dpif_open(). */
188 dp_unregister_provider(const char *type)
194 ovs_mutex_lock(&dpif_mutex);
195 error = dp_unregister_provider__(type);
196 ovs_mutex_unlock(&dpif_mutex);
201 /* Blacklists a provider. Causes future calls of dp_register_provider() with
202 * a dpif_class which implements 'type' to fail. */
204 dp_blacklist_provider(const char *type)
206 ovs_mutex_lock(&dpif_mutex);
207 sset_add(&dpif_blacklist, type);
208 ovs_mutex_unlock(&dpif_mutex);
211 /* Clears 'types' and enumerates the types of all currently registered datapath
212 * providers into it. The caller must first initialize the sset. */
214 dp_enumerate_types(struct sset *types)
216 struct shash_node *node;
221 ovs_mutex_lock(&dpif_mutex);
222 SHASH_FOR_EACH(node, &dpif_classes) {
223 const struct registered_dpif_class *registered_class = node->data;
224 sset_add(types, registered_class->dpif_class->type);
226 ovs_mutex_unlock(&dpif_mutex);
230 dp_class_unref(struct registered_dpif_class *rc)
232 ovs_mutex_lock(&dpif_mutex);
233 ovs_assert(rc->refcount);
235 ovs_mutex_unlock(&dpif_mutex);
238 static struct registered_dpif_class *
239 dp_class_lookup(const char *type)
241 struct registered_dpif_class *rc;
243 ovs_mutex_lock(&dpif_mutex);
244 rc = shash_find_data(&dpif_classes, type);
248 ovs_mutex_unlock(&dpif_mutex);
253 /* Clears 'names' and enumerates the names of all known created datapaths with
254 * the given 'type'. The caller must first initialize the sset. Returns 0 if
255 * successful, otherwise a positive errno value.
257 * Some kinds of datapaths might not be practically enumerable. This is not
258 * considered an error. */
260 dp_enumerate_names(const char *type, struct sset *names)
262 struct registered_dpif_class *registered_class;
263 const struct dpif_class *dpif_class;
269 registered_class = dp_class_lookup(type);
270 if (!registered_class) {
271 VLOG_WARN("could not enumerate unknown type: %s", type);
275 dpif_class = registered_class->dpif_class;
276 error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
278 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
279 ovs_strerror(error));
281 dp_class_unref(registered_class);
286 /* Parses 'datapath_name_', which is of the form [type@]name into its
287 * component pieces. 'name' and 'type' must be freed by the caller.
289 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
291 dp_parse_name(const char *datapath_name_, char **name, char **type)
293 char *datapath_name = xstrdup(datapath_name_);
296 separator = strchr(datapath_name, '@');
299 *type = datapath_name;
300 *name = xstrdup(dpif_normalize_type(separator + 1));
302 *name = datapath_name;
303 *type = xstrdup(dpif_normalize_type(NULL));
308 do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
310 struct dpif *dpif = NULL;
312 struct registered_dpif_class *registered_class;
316 type = dpif_normalize_type(type);
317 registered_class = dp_class_lookup(type);
318 if (!registered_class) {
319 VLOG_WARN("could not create datapath %s of unknown type %s", name,
321 error = EAFNOSUPPORT;
325 error = registered_class->dpif_class->open(registered_class->dpif_class,
326 name, create, &dpif);
328 ovs_assert(dpif->dpif_class == registered_class->dpif_class);
330 dp_class_unref(registered_class);
334 *dpifp = error ? NULL : dpif;
338 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
339 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
340 * the empty string to specify the default system type. Returns 0 if
341 * successful, otherwise a positive errno value. On success stores a pointer
342 * to the datapath in '*dpifp', otherwise a null pointer. */
344 dpif_open(const char *name, const char *type, struct dpif **dpifp)
346 return do_open(name, type, false, dpifp);
349 /* Tries to create and open a new datapath with the given 'name' and 'type'.
350 * 'type' may be either NULL or the empty string to specify the default system
351 * type. Will fail if a datapath with 'name' and 'type' already exists.
352 * Returns 0 if successful, otherwise a positive errno value. On success
353 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
355 dpif_create(const char *name, const char *type, struct dpif **dpifp)
357 return do_open(name, type, true, dpifp);
360 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
361 * does not exist. 'type' may be either NULL or the empty string to specify
362 * the default system type. Returns 0 if successful, otherwise a positive
363 * errno value. On success stores a pointer to the datapath in '*dpifp',
364 * otherwise a null pointer. */
366 dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
370 error = dpif_create(name, type, dpifp);
371 if (error == EEXIST || error == EBUSY) {
372 error = dpif_open(name, type, dpifp);
374 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
375 name, ovs_strerror(error));
378 VLOG_WARN("failed to create datapath %s: %s",
379 name, ovs_strerror(error));
384 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
385 * itself; call dpif_delete() first, instead, if that is desirable. */
387 dpif_close(struct dpif *dpif)
390 struct registered_dpif_class *rc;
392 rc = shash_find_data(&dpif_classes, dpif->dpif_class->type);
393 dpif_uninit(dpif, true);
398 /* Performs periodic work needed by 'dpif'. */
400 dpif_run(struct dpif *dpif)
402 if (dpif->dpif_class->run) {
403 dpif->dpif_class->run(dpif);
407 /* Arranges for poll_block() to wake up when dp_run() needs to be called for
410 dpif_wait(struct dpif *dpif)
412 if (dpif->dpif_class->wait) {
413 dpif->dpif_class->wait(dpif);
417 /* Returns the name of datapath 'dpif' prefixed with the type
418 * (for use in log messages). */
420 dpif_name(const struct dpif *dpif)
422 return dpif->full_name;
425 /* Returns the name of datapath 'dpif' without the type
426 * (for use in device names). */
428 dpif_base_name(const struct dpif *dpif)
430 return dpif->base_name;
433 /* Returns the type of datapath 'dpif'. */
435 dpif_type(const struct dpif *dpif)
437 return dpif->dpif_class->type;
440 /* Returns the fully spelled out name for the given datapath 'type'.
442 * Normalized type string can be compared with strcmp(). Unnormalized type
443 * string might be the same even if they have different spellings. */
445 dpif_normalize_type(const char *type)
447 return type && type[0] ? type : "system";
450 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
451 * ports. After calling this function, it does not make sense to pass 'dpif'
452 * to any functions other than dpif_name() or dpif_close(). */
454 dpif_delete(struct dpif *dpif)
458 COVERAGE_INC(dpif_destroy);
460 error = dpif->dpif_class->destroy(dpif);
461 log_operation(dpif, "delete", error);
465 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
466 * otherwise a positive errno value. */
468 dpif_get_dp_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
470 int error = dpif->dpif_class->get_stats(dpif, stats);
472 memset(stats, 0, sizeof *stats);
474 log_operation(dpif, "get_stats", error);
479 dpif_port_open_type(const char *datapath_type, const char *port_type)
481 struct registered_dpif_class *rc;
483 datapath_type = dpif_normalize_type(datapath_type);
485 ovs_mutex_lock(&dpif_mutex);
486 rc = shash_find_data(&dpif_classes, datapath_type);
487 if (rc && rc->dpif_class->port_open_type) {
488 port_type = rc->dpif_class->port_open_type(rc->dpif_class, port_type);
490 ovs_mutex_unlock(&dpif_mutex);
495 /* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
496 * non-null and its value is not ODPP_NONE, then attempts to use the
497 * value as the port number.
499 * If successful, returns 0 and sets '*port_nop' to the new port's port
500 * number (if 'port_nop' is non-null). On failure, returns a positive
501 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
504 dpif_port_add(struct dpif *dpif, struct netdev *netdev, odp_port_t *port_nop)
506 const char *netdev_name = netdev_get_name(netdev);
507 odp_port_t port_no = ODPP_NONE;
510 COVERAGE_INC(dpif_port_add);
516 error = dpif->dpif_class->port_add(dpif, netdev, &port_no);
518 VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu32,
519 dpif_name(dpif), netdev_name, port_no);
521 VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s",
522 dpif_name(dpif), netdev_name, ovs_strerror(error));
531 /* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful,
532 * otherwise a positive errno value. */
534 dpif_port_del(struct dpif *dpif, odp_port_t port_no)
538 COVERAGE_INC(dpif_port_del);
540 error = dpif->dpif_class->port_del(dpif, port_no);
542 VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu32")",
543 dpif_name(dpif), port_no);
545 log_operation(dpif, "port_del", error);
550 /* Makes a deep copy of 'src' into 'dst'. */
552 dpif_port_clone(struct dpif_port *dst, const struct dpif_port *src)
554 dst->name = xstrdup(src->name);
555 dst->type = xstrdup(src->type);
556 dst->port_no = src->port_no;
559 /* Frees memory allocated to members of 'dpif_port'.
561 * Do not call this function on a dpif_port obtained from
562 * dpif_port_dump_next(): that function retains ownership of the data in the
565 dpif_port_destroy(struct dpif_port *dpif_port)
567 free(dpif_port->name);
568 free(dpif_port->type);
571 /* Checks if port named 'devname' exists in 'dpif'. If so, returns
572 * true; otherwise, returns false. */
574 dpif_port_exists(const struct dpif *dpif, const char *devname)
576 int error = dpif->dpif_class->port_query_by_name(dpif, devname, NULL);
577 if (error != 0 && error != ENOENT && error != ENODEV) {
578 VLOG_WARN_RL(&error_rl, "%s: failed to query port %s: %s",
579 dpif_name(dpif), devname, ovs_strerror(error));
585 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
586 * initializes '*port' appropriately; on failure, returns a positive errno
589 * The caller owns the data in 'port' and must free it with
590 * dpif_port_destroy() when it is no longer needed. */
592 dpif_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
593 struct dpif_port *port)
595 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
597 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu32" is device %s",
598 dpif_name(dpif), port_no, port->name);
600 memset(port, 0, sizeof *port);
601 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu32": %s",
602 dpif_name(dpif), port_no, ovs_strerror(error));
607 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
608 * initializes '*port' appropriately; on failure, returns a positive errno
611 * The caller owns the data in 'port' and must free it with
612 * dpif_port_destroy() when it is no longer needed. */
614 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
615 struct dpif_port *port)
617 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
619 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu32,
620 dpif_name(dpif), devname, port->port_no);
622 memset(port, 0, sizeof *port);
624 /* For ENOENT or ENODEV we use DBG level because the caller is probably
625 * interested in whether 'dpif' actually has a port 'devname', so that
626 * it's not an issue worth logging if it doesn't. Other errors are
627 * uncommon and more likely to indicate a real problem. */
629 error == ENOENT || error == ENODEV ? VLL_DBG : VLL_WARN,
630 "%s: failed to query port %s: %s",
631 dpif_name(dpif), devname, ovs_strerror(error));
636 /* Returns one greater than the maximum port number accepted in flow
639 dpif_get_max_ports(const struct dpif *dpif)
641 return dpif->dpif_class->get_max_ports(dpif);
644 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE actions
645 * as the OVS_USERSPACE_ATTR_PID attribute's value, for use in flows whose
646 * packets arrived on port 'port_no'.
648 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
649 * allocated to any port, that the client may use for special purposes.
651 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
652 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
653 * disabled and then re-enabled, so a client that does that must be prepared to
654 * update all of the flows that it installed that contain
655 * OVS_ACTION_ATTR_USERSPACE actions. */
657 dpif_port_get_pid(const struct dpif *dpif, odp_port_t port_no)
659 return (dpif->dpif_class->port_get_pid
660 ? (dpif->dpif_class->port_get_pid)(dpif, port_no)
664 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
665 * the port's name into the 'name_size' bytes in 'name', ensuring that the
666 * result is null-terminated. On failure, returns a positive errno value and
667 * makes 'name' the empty string. */
669 dpif_port_get_name(struct dpif *dpif, odp_port_t port_no,
670 char *name, size_t name_size)
672 struct dpif_port port;
675 ovs_assert(name_size > 0);
677 error = dpif_port_query_by_number(dpif, port_no, &port);
679 ovs_strlcpy(name, port.name, name_size);
680 dpif_port_destroy(&port);
687 /* Initializes 'dump' to begin dumping the ports in a dpif.
689 * This function provides no status indication. An error status for the entire
690 * dump operation is provided when it is completed by calling
691 * dpif_port_dump_done().
694 dpif_port_dump_start(struct dpif_port_dump *dump, const struct dpif *dpif)
697 dump->error = dpif->dpif_class->port_dump_start(dpif, &dump->state);
698 log_operation(dpif, "port_dump_start", dump->error);
701 /* Attempts to retrieve another port from 'dump', which must have been
702 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
703 * into 'port' and returns true. On failure, returns false.
705 * Failure might indicate an actual error or merely that the last port has been
706 * dumped. An error status for the entire dump operation is provided when it
707 * is completed by calling dpif_port_dump_done().
709 * The dpif owns the data stored in 'port'. It will remain valid until at
710 * least the next time 'dump' is passed to dpif_port_dump_next() or
711 * dpif_port_dump_done(). */
713 dpif_port_dump_next(struct dpif_port_dump *dump, struct dpif_port *port)
715 const struct dpif *dpif = dump->dpif;
721 dump->error = dpif->dpif_class->port_dump_next(dpif, dump->state, port);
722 if (dump->error == EOF) {
723 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all ports", dpif_name(dpif));
725 log_operation(dpif, "port_dump_next", dump->error);
729 dpif->dpif_class->port_dump_done(dpif, dump->state);
735 /* Completes port table dump operation 'dump', which must have been initialized
736 * with dpif_port_dump_start(). Returns 0 if the dump operation was
737 * error-free, otherwise a positive errno value describing the problem. */
739 dpif_port_dump_done(struct dpif_port_dump *dump)
741 const struct dpif *dpif = dump->dpif;
743 dump->error = dpif->dpif_class->port_dump_done(dpif, dump->state);
744 log_operation(dpif, "port_dump_done", dump->error);
746 return dump->error == EOF ? 0 : dump->error;
749 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
750 * 'dpif' has changed, this function does one of the following:
752 * - Stores the name of the device that was added to or deleted from 'dpif' in
753 * '*devnamep' and returns 0. The caller is responsible for freeing
754 * '*devnamep' (with free()) when it no longer needs it.
756 * - Returns ENOBUFS and sets '*devnamep' to NULL.
758 * This function may also return 'false positives', where it returns 0 and
759 * '*devnamep' names a device that was not actually added or deleted or it
760 * returns ENOBUFS without any change.
762 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
763 * return other positive errno values to indicate that something has gone
766 dpif_port_poll(const struct dpif *dpif, char **devnamep)
768 int error = dpif->dpif_class->port_poll(dpif, devnamep);
775 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
776 * value other than EAGAIN. */
778 dpif_port_poll_wait(const struct dpif *dpif)
780 dpif->dpif_class->port_poll_wait(dpif);
783 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
784 * arguments must have been initialized through a call to flow_extract().
785 * 'used' is stored into stats->used. */
787 dpif_flow_stats_extract(const struct flow *flow, const struct ofpbuf *packet,
788 long long int used, struct dpif_flow_stats *stats)
790 stats->tcp_flags = packet_get_tcp_flags(packet, flow);
791 stats->n_bytes = packet->size;
792 stats->n_packets = 1;
796 /* Appends a human-readable representation of 'stats' to 's'. */
798 dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s)
800 ds_put_format(s, "packets:%"PRIu64", bytes:%"PRIu64", used:",
801 stats->n_packets, stats->n_bytes);
803 ds_put_format(s, "%.3fs", (time_msec() - stats->used) / 1000.0);
805 ds_put_format(s, "never");
807 if (stats->tcp_flags) {
808 ds_put_cstr(s, ", flags:");
809 packet_format_tcp_flags(s, stats->tcp_flags);
813 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
814 * positive errno value. */
816 dpif_flow_flush(struct dpif *dpif)
820 COVERAGE_INC(dpif_flow_flush);
822 error = dpif->dpif_class->flow_flush(dpif);
823 log_operation(dpif, "flow_flush", error);
827 /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
828 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
831 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
832 * failure, returns a positive errno value.
834 * If 'actionsp' is nonnull, then on success '*actionsp' will be set to an
835 * ofpbuf owned by the caller that contains the Netlink attributes for the
836 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete()) when
837 * it is no longer needed.
839 * If 'stats' is nonnull, then on success it will be updated with the flow's
842 dpif_flow_get(const struct dpif *dpif,
843 const struct nlattr *key, size_t key_len,
844 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
848 COVERAGE_INC(dpif_flow_get);
850 error = dpif->dpif_class->flow_get(dpif, key, key_len, actionsp, stats);
856 memset(stats, 0, sizeof *stats);
859 if (should_log_flow_message(error)) {
860 const struct nlattr *actions;
863 if (!error && actionsp) {
864 actions = (*actionsp)->data;
865 actions_len = (*actionsp)->size;
870 log_flow_message(dpif, error, "flow_get", key, key_len,
871 NULL, 0, stats, actions, actions_len);
877 dpif_flow_put__(struct dpif *dpif, const struct dpif_flow_put *put)
881 COVERAGE_INC(dpif_flow_put);
882 ovs_assert(!(put->flags & ~(DPIF_FP_CREATE | DPIF_FP_MODIFY
883 | DPIF_FP_ZERO_STATS)));
885 error = dpif->dpif_class->flow_put(dpif, put);
886 if (error && put->stats) {
887 memset(put->stats, 0, sizeof *put->stats);
889 log_flow_put_message(dpif, put, error);
893 /* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
894 * attribute OVS_FLOW_ATTR_KEY with types OVS_KEY_ATTR_* in the 'key_len' bytes
895 * starting at 'key', and OVS_FLOW_ATTR_MASK with types of OVS_KEY_ATTR_* in the
896 * 'mask_len' bytes starting at 'mask'. The associated actions are specified by
897 * the Netlink attributes with types OVS_ACTION_ATTR_* in the 'actions_len'
898 * bytes starting at 'actions'.
900 * - If the flow's key does not exist in 'dpif', then the flow will be added if
901 * 'flags' includes DPIF_FP_CREATE. Otherwise the operation will fail with
904 * If the operation succeeds, then 'stats', if nonnull, will be zeroed.
906 * - If the flow's key does exist in 'dpif', then the flow's actions will be
907 * updated if 'flags' includes DPIF_FP_MODIFY. Otherwise the operation will
908 * fail with EEXIST. If the flow's actions are updated, then its statistics
909 * will be zeroed if 'flags' includes DPIF_FP_ZERO_STATS, and left as-is
912 * If the operation succeeds, then 'stats', if nonnull, will be set to the
913 * flow's statistics before the update.
916 dpif_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
917 const struct nlattr *key, size_t key_len,
918 const struct nlattr *mask, size_t mask_len,
919 const struct nlattr *actions, size_t actions_len,
920 struct dpif_flow_stats *stats)
922 struct dpif_flow_put put;
926 put.key_len = key_len;
928 put.mask_len = mask_len;
929 put.actions = actions;
930 put.actions_len = actions_len;
932 return dpif_flow_put__(dpif, &put);
936 dpif_flow_del__(struct dpif *dpif, struct dpif_flow_del *del)
940 COVERAGE_INC(dpif_flow_del);
942 error = dpif->dpif_class->flow_del(dpif, del);
943 if (error && del->stats) {
944 memset(del->stats, 0, sizeof *del->stats);
946 log_flow_del_message(dpif, del, error);
950 /* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif' does
951 * not contain such a flow. The flow is specified by the Netlink attributes
952 * with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at 'key'.
954 * If the operation succeeds, then 'stats', if nonnull, will be set to the
955 * flow's statistics before its deletion. */
957 dpif_flow_del(struct dpif *dpif,
958 const struct nlattr *key, size_t key_len,
959 struct dpif_flow_stats *stats)
961 struct dpif_flow_del del;
964 del.key_len = key_len;
966 return dpif_flow_del__(dpif, &del);
969 /* Initializes 'dump' to begin dumping the flows in a dpif.
971 * This function provides no status indication. An error status for the entire
972 * dump operation is provided when it is completed by calling
973 * dpif_flow_dump_done().
976 dpif_flow_dump_start(struct dpif_flow_dump *dump, const struct dpif *dpif)
979 dump->error = dpif->dpif_class->flow_dump_start(dpif, &dump->state);
980 log_operation(dpif, "flow_dump_start", dump->error);
983 /* Attempts to retrieve another flow from 'dump', which must have been
984 * initialized with dpif_flow_dump_start(). On success, updates the output
985 * parameters as described below and returns true. Otherwise, returns false.
986 * Failure might indicate an actual error or merely the end of the flow table.
987 * An error status for the entire dump operation is provided when it is
988 * completed by calling dpif_flow_dump_done().
990 * On success, if 'key' and 'key_len' are nonnull then '*key' and '*key_len'
991 * will be set to Netlink attributes with types OVS_KEY_ATTR_* representing the
992 * dumped flow's key. If 'actions' and 'actions_len' are nonnull then they are
993 * set to Netlink attributes with types OVS_ACTION_ATTR_* representing the
994 * dumped flow's actions. If 'stats' is nonnull then it will be set to the
995 * dumped flow's statistics.
997 * All of the returned data is owned by 'dpif', not by the caller, and the
998 * caller must not modify or free it. 'dpif' guarantees that it remains
999 * accessible and unchanging until at least the next call to 'flow_dump_next'
1000 * or 'flow_dump_done' for 'dump'. */
1002 dpif_flow_dump_next(struct dpif_flow_dump *dump,
1003 const struct nlattr **key, size_t *key_len,
1004 const struct nlattr **mask, size_t *mask_len,
1005 const struct nlattr **actions, size_t *actions_len,
1006 const struct dpif_flow_stats **stats)
1008 const struct dpif *dpif = dump->dpif;
1009 int error = dump->error;
1012 error = dpif->dpif_class->flow_dump_next(dpif, dump->state,
1015 actions, actions_len,
1018 dpif->dpif_class->flow_dump_done(dpif, dump->state);
1040 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif));
1041 } else if (should_log_flow_message(error)) {
1042 log_flow_message(dpif, error, "flow_dump",
1043 key ? *key : NULL, key ? *key_len : 0,
1044 mask ? *mask : NULL, mask ? *mask_len : 0,
1045 stats ? *stats : NULL, actions ? *actions : NULL,
1046 actions ? *actions_len : 0);
1049 dump->error = error;
1053 /* Completes flow table dump operation 'dump', which must have been initialized
1054 * with dpif_flow_dump_start(). Returns 0 if the dump operation was
1055 * error-free, otherwise a positive errno value describing the problem. */
1057 dpif_flow_dump_done(struct dpif_flow_dump *dump)
1059 const struct dpif *dpif = dump->dpif;
1061 dump->error = dpif->dpif_class->flow_dump_done(dpif, dump->state);
1062 log_operation(dpif, "flow_dump_done", dump->error);
1064 return dump->error == EOF ? 0 : dump->error;
1067 struct dpif_execute_helper_aux {
1073 dpif_execute_helper_execute__(void *aux_, struct ofpbuf *packet,
1074 const struct flow *flow,
1075 const struct nlattr *actions, size_t actions_len)
1077 struct dpif_execute_helper_aux *aux = aux_;
1078 struct dpif_execute execute;
1079 struct odputil_keybuf key_stub;
1083 ofpbuf_use_stub(&key, &key_stub, sizeof key_stub);
1084 odp_flow_key_from_flow(&key, flow, flow->in_port.odp_port);
1086 execute.key = key.data;
1087 execute.key_len = key.size;
1088 execute.actions = actions;
1089 execute.actions_len = actions_len;
1090 execute.packet = packet;
1091 execute.needs_help = false;
1093 error = aux->dpif->dpif_class->execute(aux->dpif, &execute);
1100 dpif_execute_helper_output_cb(void *aux, struct ofpbuf *packet,
1101 const struct flow *flow, odp_port_t out_port)
1103 uint64_t actions_stub[DIV_ROUND_UP(NL_A_U32_SIZE, 8)];
1104 struct ofpbuf actions;
1106 ofpbuf_use_stack(&actions, actions_stub, sizeof actions_stub);
1107 nl_msg_put_u32(&actions, OVS_ACTION_ATTR_OUTPUT, odp_to_u32(out_port));
1109 dpif_execute_helper_execute__(aux, packet, flow,
1110 actions.data, actions.size);
1114 dpif_execute_helper_userspace_cb(void *aux, struct ofpbuf *packet,
1115 const struct flow *flow,
1116 const struct nlattr *action)
1118 dpif_execute_helper_execute__(aux, packet, flow,
1119 action, NLA_ALIGN(action->nla_len));
1122 /* Executes 'execute' by performing most of the actions in userspace and
1123 * passing the fully constructed packets to 'dpif' for output and userspace
1126 * This helps with actions that a given 'dpif' doesn't implement directly. */
1128 dpif_execute_with_help(struct dpif *dpif, const struct dpif_execute *execute)
1130 struct dpif_execute_helper_aux aux;
1131 enum odp_key_fitness fit;
1132 struct ofpbuf *packet;
1135 COVERAGE_INC(dpif_execute_with_help);
1137 fit = odp_flow_key_to_flow(execute->key, execute->key_len, &flow);
1138 if (fit == ODP_FIT_ERROR) {
1145 packet = ofpbuf_clone_with_headroom(execute->packet, VLAN_HEADER_LEN);
1146 odp_execute_actions(&aux, packet, &flow,
1147 execute->actions, execute->actions_len,
1148 dpif_execute_helper_output_cb,
1149 dpif_execute_helper_userspace_cb);
1150 ofpbuf_delete(packet);
1156 dpif_execute__(struct dpif *dpif, const struct dpif_execute *execute)
1160 COVERAGE_INC(dpif_execute);
1161 if (execute->actions_len > 0) {
1162 error = (execute->needs_help
1163 ? dpif_execute_with_help(dpif, execute)
1164 : dpif->dpif_class->execute(dpif, execute));
1169 log_execute_message(dpif, execute, error);
1174 /* Causes 'dpif' to perform the 'actions_len' bytes of actions in 'actions' on
1175 * the Ethernet frame specified in 'packet' taken from the flow specified in
1176 * the 'key_len' bytes of 'key'. ('key' is mostly redundant with 'packet', but
1177 * it contains some metadata that cannot be recovered from 'packet', such as
1178 * tunnel and in_port.)
1180 * Some dpif providers do not implement every action. The Linux kernel
1181 * datapath, in particular, does not implement ARP field modification. If
1182 * 'needs_help' is true, the dpif layer executes in userspace all of the
1183 * actions that it can, and for OVS_ACTION_ATTR_OUTPUT and
1184 * OVS_ACTION_ATTR_USERSPACE actions it passes the packet through to the dpif
1187 * Returns 0 if successful, otherwise a positive errno value. */
1189 dpif_execute(struct dpif *dpif,
1190 const struct nlattr *key, size_t key_len,
1191 const struct nlattr *actions, size_t actions_len,
1192 const struct ofpbuf *buf,
1195 struct dpif_execute execute;
1198 execute.key_len = key_len;
1199 execute.actions = actions;
1200 execute.actions_len = actions_len;
1201 execute.packet = buf;
1202 execute.needs_help = needs_help;
1203 return dpif_execute__(dpif, &execute);
1206 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1207 * which they are specified, placing each operation's results in the "output"
1208 * members documented in comments.
1210 * This function exists because some datapaths can perform batched operations
1211 * faster than individual operations. */
1213 dpif_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1215 if (dpif->dpif_class->operate) {
1219 /* Count 'chunk', the number of ops that can be executed without
1220 * needing any help. Ops that need help should be rare, so we
1221 * expect this to ordinarily be 'n_ops', that is, all the ops. */
1222 for (chunk = 0; chunk < n_ops; chunk++) {
1223 struct dpif_op *op = ops[chunk];
1225 if (op->type == DPIF_OP_EXECUTE && op->u.execute.needs_help) {
1231 /* Execute a chunk full of ops that the dpif provider can
1232 * handle itself, without help. */
1235 dpif->dpif_class->operate(dpif, ops, chunk);
1237 for (i = 0; i < chunk; i++) {
1238 struct dpif_op *op = ops[i];
1241 case DPIF_OP_FLOW_PUT:
1242 log_flow_put_message(dpif, &op->u.flow_put, op->error);
1245 case DPIF_OP_FLOW_DEL:
1246 log_flow_del_message(dpif, &op->u.flow_del, op->error);
1249 case DPIF_OP_EXECUTE:
1250 log_execute_message(dpif, &op->u.execute, op->error);
1258 /* Help the dpif provider to execute one op. */
1259 struct dpif_op *op = ops[0];
1261 op->error = dpif_execute__(dpif, &op->u.execute);
1269 for (i = 0; i < n_ops; i++) {
1270 struct dpif_op *op = ops[i];
1273 case DPIF_OP_FLOW_PUT:
1274 op->error = dpif_flow_put__(dpif, &op->u.flow_put);
1277 case DPIF_OP_FLOW_DEL:
1278 op->error = dpif_flow_del__(dpif, &op->u.flow_del);
1281 case DPIF_OP_EXECUTE:
1282 op->error = dpif_execute__(dpif, &op->u.execute);
1292 /* Returns a string that represents 'type', for use in log messages. */
1294 dpif_upcall_type_to_string(enum dpif_upcall_type type)
1297 case DPIF_UC_MISS: return "miss";
1298 case DPIF_UC_ACTION: return "action";
1299 case DPIF_N_UC_TYPES: default: return "<unknown>";
1303 /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1304 * if successful, otherwise a positive errno value.
1306 * Turning packet receive off and then back on may change the Netlink PID
1307 * assignments returned by dpif_port_get_pid(). If the client does this, it
1308 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1309 * using the new PID assignment. */
1311 dpif_recv_set(struct dpif *dpif, bool enable)
1313 int error = dpif->dpif_class->recv_set(dpif, enable);
1314 log_operation(dpif, "recv_set", error);
1318 /* Polls for an upcall from 'dpif'. If successful, stores the upcall into
1319 * '*upcall', using 'buf' for storage. Should only be called if
1320 * dpif_recv_set() has been used to enable receiving packets on 'dpif'.
1322 * 'upcall->packet' and 'upcall->key' point into data in the caller-provided
1323 * 'buf', so their memory cannot be freed separately from 'buf'. (This is
1324 * hardly a great way to do things but it works out OK for the dpif providers
1325 * and clients that exist so far.)
1327 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1328 * if no upcall is immediately available. */
1330 dpif_recv(struct dpif *dpif, struct dpif_upcall *upcall, struct ofpbuf *buf)
1332 int error = dpif->dpif_class->recv(dpif, upcall, buf);
1333 if (!error && !VLOG_DROP_DBG(&dpmsg_rl)) {
1337 packet = ofp_packet_to_string(upcall->packet->data,
1338 upcall->packet->size);
1341 odp_flow_key_format(upcall->key, upcall->key_len, &flow);
1343 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1344 dpif_name(dpif), dpif_upcall_type_to_string(upcall->type),
1345 ds_cstr(&flow), packet);
1349 } else if (error && error != EAGAIN) {
1350 log_operation(dpif, "recv", error);
1355 /* Discards all messages that would otherwise be received by dpif_recv() on
1358 dpif_recv_purge(struct dpif *dpif)
1360 COVERAGE_INC(dpif_purge);
1361 if (dpif->dpif_class->recv_purge) {
1362 dpif->dpif_class->recv_purge(dpif);
1366 /* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
1367 * received with dpif_recv(). */
1369 dpif_recv_wait(struct dpif *dpif)
1371 dpif->dpif_class->recv_wait(dpif);
1374 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1375 * and '*engine_id', respectively. */
1377 dpif_get_netflow_ids(const struct dpif *dpif,
1378 uint8_t *engine_type, uint8_t *engine_id)
1380 *engine_type = dpif->netflow_engine_type;
1381 *engine_id = dpif->netflow_engine_id;
1384 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1385 * value used for setting packet priority.
1386 * On success, returns 0 and stores the priority into '*priority'.
1387 * On failure, returns a positive errno value and stores 0 into '*priority'. */
1389 dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1392 int error = (dpif->dpif_class->queue_to_priority
1393 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1399 log_operation(dpif, "queue_to_priority", error);
1404 dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1406 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1408 dpif->dpif_class = dpif_class;
1409 dpif->base_name = xstrdup(name);
1410 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
1411 dpif->netflow_engine_type = netflow_engine_type;
1412 dpif->netflow_engine_id = netflow_engine_id;
1415 /* Undoes the results of initialization.
1417 * Normally this function only needs to be called from dpif_close().
1418 * However, it may be called by providers due to an error on opening
1419 * that occurs after initialization. It this case dpif_close() would
1420 * never be called. */
1422 dpif_uninit(struct dpif *dpif, bool close)
1424 char *base_name = dpif->base_name;
1425 char *full_name = dpif->full_name;
1428 dpif->dpif_class->close(dpif);
1436 log_operation(const struct dpif *dpif, const char *operation, int error)
1439 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
1440 } else if (ofperr_is_valid(error)) {
1441 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1442 dpif_name(dpif), operation, ofperr_get_name(error));
1444 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1445 dpif_name(dpif), operation, ovs_strerror(error));
1449 static enum vlog_level
1450 flow_message_log_level(int error)
1452 /* If flows arrive in a batch, userspace may push down multiple
1453 * unique flow definitions that overlap when wildcards are applied.
1454 * Kernels that support flow wildcarding will reject these flows as
1455 * duplicates (EEXIST), so lower the log level to debug for these
1456 * types of messages. */
1457 return (error && error != EEXIST) ? VLL_WARN : VLL_DBG;
1461 should_log_flow_message(int error)
1463 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1464 error ? &error_rl : &dpmsg_rl);
1468 log_flow_message(const struct dpif *dpif, int error, const char *operation,
1469 const struct nlattr *key, size_t key_len,
1470 const struct nlattr *mask, size_t mask_len,
1471 const struct dpif_flow_stats *stats,
1472 const struct nlattr *actions, size_t actions_len)
1474 struct ds ds = DS_EMPTY_INITIALIZER;
1475 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1477 ds_put_cstr(&ds, "failed to ");
1479 ds_put_format(&ds, "%s ", operation);
1481 ds_put_format(&ds, "(%s) ", ovs_strerror(error));
1483 odp_flow_format(key, key_len, mask, mask_len, NULL, &ds, true);
1485 ds_put_cstr(&ds, ", ");
1486 dpif_flow_stats_format(stats, &ds);
1488 if (actions || actions_len) {
1489 ds_put_cstr(&ds, ", actions:");
1490 format_odp_actions(&ds, actions, actions_len);
1492 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1497 log_flow_put_message(struct dpif *dpif, const struct dpif_flow_put *put,
1500 if (should_log_flow_message(error)) {
1504 ds_put_cstr(&s, "put");
1505 if (put->flags & DPIF_FP_CREATE) {
1506 ds_put_cstr(&s, "[create]");
1508 if (put->flags & DPIF_FP_MODIFY) {
1509 ds_put_cstr(&s, "[modify]");
1511 if (put->flags & DPIF_FP_ZERO_STATS) {
1512 ds_put_cstr(&s, "[zero]");
1514 log_flow_message(dpif, error, ds_cstr(&s),
1515 put->key, put->key_len, put->mask, put->mask_len,
1516 put->stats, put->actions, put->actions_len);
1522 log_flow_del_message(struct dpif *dpif, const struct dpif_flow_del *del,
1525 if (should_log_flow_message(error)) {
1526 log_flow_message(dpif, error, "flow_del", del->key, del->key_len,
1527 NULL, 0, !error ? del->stats : NULL, NULL, 0);
1532 log_execute_message(struct dpif *dpif, const struct dpif_execute *execute,
1535 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
1536 struct ds ds = DS_EMPTY_INITIALIZER;
1539 packet = ofp_packet_to_string(execute->packet->data,
1540 execute->packet->size);
1541 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
1542 format_odp_actions(&ds, execute->actions, execute->actions_len);
1544 ds_put_format(&ds, " failed (%s)", ovs_strerror(error));
1546 ds_put_format(&ds, " on packet %s", packet);
1547 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));