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"
32 #include "ofp-errors.h"
33 #include "ofp-print.h"
37 #include "poll-loop.h"
45 VLOG_DEFINE_THIS_MODULE(dpif);
47 COVERAGE_DEFINE(dpif_destroy);
48 COVERAGE_DEFINE(dpif_port_add);
49 COVERAGE_DEFINE(dpif_port_del);
50 COVERAGE_DEFINE(dpif_flow_flush);
51 COVERAGE_DEFINE(dpif_flow_get);
52 COVERAGE_DEFINE(dpif_flow_put);
53 COVERAGE_DEFINE(dpif_flow_del);
54 COVERAGE_DEFINE(dpif_flow_query_list);
55 COVERAGE_DEFINE(dpif_flow_query_list_n);
56 COVERAGE_DEFINE(dpif_execute);
57 COVERAGE_DEFINE(dpif_purge);
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 /* Rate limit for individual messages going to or from the datapath, output at
75 * DBG level. This is very high because, if these are enabled, it is because
76 * we really need to see them. */
77 static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600);
79 /* Not really much point in logging many dpif errors. */
80 static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5);
82 static void log_flow_message(const struct dpif *dpif, int error,
83 const char *operation,
84 const struct nlattr *key, size_t key_len,
85 const struct nlattr *mask, size_t mask_len,
86 const struct dpif_flow_stats *stats,
87 const struct nlattr *actions, size_t actions_len);
88 static void log_operation(const struct dpif *, const char *operation,
90 static bool should_log_flow_message(int error);
91 static void log_flow_put_message(struct dpif *, const struct dpif_flow_put *,
93 static void log_flow_del_message(struct dpif *, const struct dpif_flow_del *,
95 static void log_execute_message(struct dpif *, const struct dpif_execute *,
101 static int status = -1;
107 for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) {
108 dp_register_provider(base_dpif_classes[i]);
113 /* Registers a new datapath provider. After successful registration, new
114 * datapaths of that type can be opened using dpif_open(). */
116 dp_register_provider(const struct dpif_class *new_class)
118 struct registered_dpif_class *registered_class;
120 if (sset_contains(&dpif_blacklist, new_class->type)) {
121 VLOG_DBG("attempted to register blacklisted provider: %s",
126 if (shash_find(&dpif_classes, new_class->type)) {
127 VLOG_WARN("attempted to register duplicate datapath provider: %s",
132 registered_class = xmalloc(sizeof *registered_class);
133 registered_class->dpif_class = new_class;
134 registered_class->refcount = 0;
136 shash_add(&dpif_classes, new_class->type, registered_class);
141 /* Unregisters a datapath provider. 'type' must have been previously
142 * registered and not currently be in use by any dpifs. After unregistration
143 * new datapaths of that type cannot be opened using dpif_open(). */
145 dp_unregister_provider(const char *type)
147 struct shash_node *node;
148 struct registered_dpif_class *registered_class;
150 node = shash_find(&dpif_classes, type);
152 VLOG_WARN("attempted to unregister a datapath provider that is not "
153 "registered: %s", type);
157 registered_class = node->data;
158 if (registered_class->refcount) {
159 VLOG_WARN("attempted to unregister in use datapath provider: %s", type);
163 shash_delete(&dpif_classes, node);
164 free(registered_class);
169 /* Blacklists a provider. Causes future calls of dp_register_provider() with
170 * a dpif_class which implements 'type' to fail. */
172 dp_blacklist_provider(const char *type)
174 sset_add(&dpif_blacklist, type);
177 /* Clears 'types' and enumerates the types of all currently registered datapath
178 * providers into it. The caller must first initialize the sset. */
180 dp_enumerate_types(struct sset *types)
182 struct shash_node *node;
187 SHASH_FOR_EACH(node, &dpif_classes) {
188 const struct registered_dpif_class *registered_class = node->data;
189 sset_add(types, registered_class->dpif_class->type);
193 /* Clears 'names' and enumerates the names of all known created datapaths with
194 * the given 'type'. The caller must first initialize the sset. Returns 0 if
195 * successful, otherwise a positive errno value.
197 * Some kinds of datapaths might not be practically enumerable. This is not
198 * considered an error. */
200 dp_enumerate_names(const char *type, struct sset *names)
202 const struct registered_dpif_class *registered_class;
203 const struct dpif_class *dpif_class;
209 registered_class = shash_find_data(&dpif_classes, type);
210 if (!registered_class) {
211 VLOG_WARN("could not enumerate unknown type: %s", type);
215 dpif_class = registered_class->dpif_class;
216 error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0;
219 VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type,
220 ovs_strerror(error));
226 /* Parses 'datapath_name_', which is of the form [type@]name into its
227 * component pieces. 'name' and 'type' must be freed by the caller.
229 * The returned 'type' is normalized, as if by dpif_normalize_type(). */
231 dp_parse_name(const char *datapath_name_, char **name, char **type)
233 char *datapath_name = xstrdup(datapath_name_);
236 separator = strchr(datapath_name, '@');
239 *type = datapath_name;
240 *name = xstrdup(dpif_normalize_type(separator + 1));
242 *name = datapath_name;
243 *type = xstrdup(dpif_normalize_type(NULL));
248 do_open(const char *name, const char *type, bool create, struct dpif **dpifp)
250 struct dpif *dpif = NULL;
252 struct registered_dpif_class *registered_class;
256 type = dpif_normalize_type(type);
258 registered_class = shash_find_data(&dpif_classes, type);
259 if (!registered_class) {
260 VLOG_WARN("could not create datapath %s of unknown type %s", name,
262 error = EAFNOSUPPORT;
266 error = registered_class->dpif_class->open(registered_class->dpif_class,
267 name, create, &dpif);
269 ovs_assert(dpif->dpif_class == registered_class->dpif_class);
270 registered_class->refcount++;
274 *dpifp = error ? NULL : dpif;
278 /* Tries to open an existing datapath named 'name' and type 'type'. Will fail
279 * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or
280 * the empty string to specify the default system type. Returns 0 if
281 * successful, otherwise a positive errno value. On success stores a pointer
282 * to the datapath in '*dpifp', otherwise a null pointer. */
284 dpif_open(const char *name, const char *type, struct dpif **dpifp)
286 return do_open(name, type, false, dpifp);
289 /* Tries to create and open a new datapath with the given 'name' and 'type'.
290 * 'type' may be either NULL or the empty string to specify the default system
291 * type. Will fail if a datapath with 'name' and 'type' already exists.
292 * Returns 0 if successful, otherwise a positive errno value. On success
293 * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */
295 dpif_create(const char *name, const char *type, struct dpif **dpifp)
297 return do_open(name, type, true, dpifp);
300 /* Tries to open a datapath with the given 'name' and 'type', creating it if it
301 * does not exist. 'type' may be either NULL or the empty string to specify
302 * the default system type. Returns 0 if successful, otherwise a positive
303 * errno value. On success stores a pointer to the datapath in '*dpifp',
304 * otherwise a null pointer. */
306 dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp)
310 error = dpif_create(name, type, dpifp);
311 if (error == EEXIST || error == EBUSY) {
312 error = dpif_open(name, type, dpifp);
314 VLOG_WARN("datapath %s already exists but cannot be opened: %s",
315 name, ovs_strerror(error));
318 VLOG_WARN("failed to create datapath %s: %s",
319 name, ovs_strerror(error));
324 /* Closes and frees the connection to 'dpif'. Does not destroy the datapath
325 * itself; call dpif_delete() first, instead, if that is desirable. */
327 dpif_close(struct dpif *dpif)
330 struct registered_dpif_class *registered_class;
332 registered_class = shash_find_data(&dpif_classes,
333 dpif->dpif_class->type);
334 ovs_assert(registered_class);
335 ovs_assert(registered_class->refcount);
337 registered_class->refcount--;
338 dpif_uninit(dpif, true);
342 /* Performs periodic work needed by 'dpif'. */
344 dpif_run(struct dpif *dpif)
346 if (dpif->dpif_class->run) {
347 dpif->dpif_class->run(dpif);
351 /* Arranges for poll_block() to wake up when dp_run() needs to be called for
354 dpif_wait(struct dpif *dpif)
356 if (dpif->dpif_class->wait) {
357 dpif->dpif_class->wait(dpif);
361 /* Returns the name of datapath 'dpif' prefixed with the type
362 * (for use in log messages). */
364 dpif_name(const struct dpif *dpif)
366 return dpif->full_name;
369 /* Returns the name of datapath 'dpif' without the type
370 * (for use in device names). */
372 dpif_base_name(const struct dpif *dpif)
374 return dpif->base_name;
377 /* Returns the type of datapath 'dpif'. */
379 dpif_type(const struct dpif *dpif)
381 return dpif->dpif_class->type;
384 /* Returns the fully spelled out name for the given datapath 'type'.
386 * Normalized type string can be compared with strcmp(). Unnormalized type
387 * string might be the same even if they have different spellings. */
389 dpif_normalize_type(const char *type)
391 return type && type[0] ? type : "system";
394 /* Destroys the datapath that 'dpif' is connected to, first removing all of its
395 * ports. After calling this function, it does not make sense to pass 'dpif'
396 * to any functions other than dpif_name() or dpif_close(). */
398 dpif_delete(struct dpif *dpif)
402 COVERAGE_INC(dpif_destroy);
404 error = dpif->dpif_class->destroy(dpif);
405 log_operation(dpif, "delete", error);
409 /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful,
410 * otherwise a positive errno value. */
412 dpif_get_dp_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
414 int error = dpif->dpif_class->get_stats(dpif, stats);
416 memset(stats, 0, sizeof *stats);
418 log_operation(dpif, "get_stats", error);
423 dpif_port_open_type(const char *datapath_type, const char *port_type)
425 struct registered_dpif_class *registered_class;
427 datapath_type = dpif_normalize_type(datapath_type);
429 registered_class = shash_find_data(&dpif_classes, datapath_type);
430 if (!registered_class
431 || !registered_class->dpif_class->port_open_type) {
435 return registered_class->dpif_class->port_open_type(
436 registered_class->dpif_class, port_type);
439 /* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is
440 * non-null and its value is not ODPP_NONE, then attempts to use the
441 * value as the port number.
443 * If successful, returns 0 and sets '*port_nop' to the new port's port
444 * number (if 'port_nop' is non-null). On failure, returns a positive
445 * errno value and sets '*port_nop' to ODPP_NONE (if 'port_nop' is
448 dpif_port_add(struct dpif *dpif, struct netdev *netdev, odp_port_t *port_nop)
450 const char *netdev_name = netdev_get_name(netdev);
451 odp_port_t port_no = ODPP_NONE;
454 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 %"PRIu32,
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, ovs_strerror(error));
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, odp_port_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(%"PRIu32")",
487 dpif_name(dpif), port_no);
489 log_operation(dpif, "port_del", error);
494 /* Makes a deep copy of 'src' into 'dst'. */
496 dpif_port_clone(struct dpif_port *dst, const struct dpif_port *src)
498 dst->name = xstrdup(src->name);
499 dst->type = xstrdup(src->type);
500 dst->port_no = src->port_no;
503 /* Frees memory allocated to members of 'dpif_port'.
505 * Do not call this function on a dpif_port obtained from
506 * dpif_port_dump_next(): that function retains ownership of the data in the
509 dpif_port_destroy(struct dpif_port *dpif_port)
511 free(dpif_port->name);
512 free(dpif_port->type);
515 /* Checks if port named 'devname' exists in 'dpif'. If so, returns
516 * true; otherwise, returns false. */
518 dpif_port_exists(const struct dpif *dpif, const char *devname)
520 int error = dpif->dpif_class->port_query_by_name(dpif, devname, NULL);
521 if (error != 0 && error != ENOENT && error != ENODEV) {
522 VLOG_WARN_RL(&error_rl, "%s: failed to query port %s: %s",
523 dpif_name(dpif), devname, ovs_strerror(error));
529 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and
530 * initializes '*port' appropriately; on failure, returns a positive errno
533 * The caller owns the data in 'port' and must free it with
534 * dpif_port_destroy() when it is no longer needed. */
536 dpif_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
537 struct dpif_port *port)
539 int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port);
541 VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu32" is device %s",
542 dpif_name(dpif), port_no, port->name);
544 memset(port, 0, sizeof *port);
545 VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu32": %s",
546 dpif_name(dpif), port_no, ovs_strerror(error));
551 /* Looks up port named 'devname' in 'dpif'. On success, returns 0 and
552 * initializes '*port' appropriately; on failure, returns a positive errno
555 * The caller owns the data in 'port' and must free it with
556 * dpif_port_destroy() when it is no longer needed. */
558 dpif_port_query_by_name(const struct dpif *dpif, const char *devname,
559 struct dpif_port *port)
561 int error = dpif->dpif_class->port_query_by_name(dpif, devname, port);
563 VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu32,
564 dpif_name(dpif), devname, port->port_no);
566 memset(port, 0, sizeof *port);
568 /* For ENOENT or ENODEV we use DBG level because the caller is probably
569 * interested in whether 'dpif' actually has a port 'devname', so that
570 * it's not an issue worth logging if it doesn't. Other errors are
571 * uncommon and more likely to indicate a real problem. */
573 error == ENOENT || error == ENODEV ? VLL_DBG : VLL_WARN,
574 "%s: failed to query port %s: %s",
575 dpif_name(dpif), devname, ovs_strerror(error));
580 /* Returns one greater than the maximum port number accepted in flow
583 dpif_get_max_ports(const struct dpif *dpif)
585 return dpif->dpif_class->get_max_ports(dpif);
588 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE actions
589 * as the OVS_USERSPACE_ATTR_PID attribute's value, for use in flows whose
590 * packets arrived on port 'port_no'.
592 * A 'port_no' of ODPP_NONE is a special case: it returns a reserved PID, not
593 * allocated to any port, that the client may use for special purposes.
595 * The return value is only meaningful when DPIF_UC_ACTION has been enabled in
596 * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is
597 * disabled and then re-enabled, so a client that does that must be prepared to
598 * update all of the flows that it installed that contain
599 * OVS_ACTION_ATTR_USERSPACE actions. */
601 dpif_port_get_pid(const struct dpif *dpif, odp_port_t port_no)
603 return (dpif->dpif_class->port_get_pid
604 ? (dpif->dpif_class->port_get_pid)(dpif, port_no)
608 /* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies
609 * the port's name into the 'name_size' bytes in 'name', ensuring that the
610 * result is null-terminated. On failure, returns a positive errno value and
611 * makes 'name' the empty string. */
613 dpif_port_get_name(struct dpif *dpif, odp_port_t port_no,
614 char *name, size_t name_size)
616 struct dpif_port port;
619 ovs_assert(name_size > 0);
621 error = dpif_port_query_by_number(dpif, port_no, &port);
623 ovs_strlcpy(name, port.name, name_size);
624 dpif_port_destroy(&port);
631 /* Initializes 'dump' to begin dumping the ports in a dpif.
633 * This function provides no status indication. An error status for the entire
634 * dump operation is provided when it is completed by calling
635 * dpif_port_dump_done().
638 dpif_port_dump_start(struct dpif_port_dump *dump, const struct dpif *dpif)
641 dump->error = dpif->dpif_class->port_dump_start(dpif, &dump->state);
642 log_operation(dpif, "port_dump_start", dump->error);
645 /* Attempts to retrieve another port from 'dump', which must have been
646 * initialized with dpif_port_dump_start(). On success, stores a new dpif_port
647 * into 'port' and returns true. On failure, returns false.
649 * Failure might indicate an actual error or merely that the last port has been
650 * dumped. An error status for the entire dump operation is provided when it
651 * is completed by calling dpif_port_dump_done().
653 * The dpif owns the data stored in 'port'. It will remain valid until at
654 * least the next time 'dump' is passed to dpif_port_dump_next() or
655 * dpif_port_dump_done(). */
657 dpif_port_dump_next(struct dpif_port_dump *dump, struct dpif_port *port)
659 const struct dpif *dpif = dump->dpif;
665 dump->error = dpif->dpif_class->port_dump_next(dpif, dump->state, port);
666 if (dump->error == EOF) {
667 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all ports", dpif_name(dpif));
669 log_operation(dpif, "port_dump_next", dump->error);
673 dpif->dpif_class->port_dump_done(dpif, dump->state);
679 /* Completes port table dump operation 'dump', which must have been initialized
680 * with dpif_port_dump_start(). Returns 0 if the dump operation was
681 * error-free, otherwise a positive errno value describing the problem. */
683 dpif_port_dump_done(struct dpif_port_dump *dump)
685 const struct dpif *dpif = dump->dpif;
687 dump->error = dpif->dpif_class->port_dump_done(dpif, dump->state);
688 log_operation(dpif, "port_dump_done", dump->error);
690 return dump->error == EOF ? 0 : dump->error;
693 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
694 * 'dpif' has changed, this function does one of the following:
696 * - Stores the name of the device that was added to or deleted from 'dpif' in
697 * '*devnamep' and returns 0. The caller is responsible for freeing
698 * '*devnamep' (with free()) when it no longer needs it.
700 * - Returns ENOBUFS and sets '*devnamep' to NULL.
702 * This function may also return 'false positives', where it returns 0 and
703 * '*devnamep' names a device that was not actually added or deleted or it
704 * returns ENOBUFS without any change.
706 * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also
707 * return other positive errno values to indicate that something has gone
710 dpif_port_poll(const struct dpif *dpif, char **devnamep)
712 int error = dpif->dpif_class->port_poll(dpif, devnamep);
719 /* Arranges for the poll loop to wake up when port_poll(dpif) will return a
720 * value other than EAGAIN. */
722 dpif_port_poll_wait(const struct dpif *dpif)
724 dpif->dpif_class->port_poll_wait(dpif);
727 /* Extracts the flow stats for a packet. The 'flow' and 'packet'
728 * arguments must have been initialized through a call to flow_extract().
729 * 'used' is stored into stats->used. */
731 dpif_flow_stats_extract(const struct flow *flow, const struct ofpbuf *packet,
732 long long int used, struct dpif_flow_stats *stats)
734 stats->tcp_flags = packet_get_tcp_flags(packet, flow);
735 stats->n_bytes = packet->size;
736 stats->n_packets = 1;
740 /* Appends a human-readable representation of 'stats' to 's'. */
742 dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s)
744 ds_put_format(s, "packets:%"PRIu64", bytes:%"PRIu64", used:",
745 stats->n_packets, stats->n_bytes);
747 ds_put_format(s, "%.3fs", (time_msec() - stats->used) / 1000.0);
749 ds_put_format(s, "never");
751 if (stats->tcp_flags) {
752 ds_put_cstr(s, ", flags:");
753 packet_format_tcp_flags(s, stats->tcp_flags);
757 /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a
758 * positive errno value. */
760 dpif_flow_flush(struct dpif *dpif)
764 COVERAGE_INC(dpif_flow_flush);
766 error = dpif->dpif_class->flow_flush(dpif);
767 log_operation(dpif, "flow_flush", error);
771 /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
772 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
775 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
776 * failure, returns a positive errno value.
778 * If 'actionsp' is nonnull, then on success '*actionsp' will be set to an
779 * ofpbuf owned by the caller that contains the Netlink attributes for the
780 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete()) when
781 * it is no longer needed.
783 * If 'stats' is nonnull, then on success it will be updated with the flow's
786 dpif_flow_get(const struct dpif *dpif,
787 const struct nlattr *key, size_t key_len,
788 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
792 COVERAGE_INC(dpif_flow_get);
794 error = dpif->dpif_class->flow_get(dpif, key, key_len, actionsp, stats);
800 memset(stats, 0, sizeof *stats);
803 if (should_log_flow_message(error)) {
804 const struct nlattr *actions;
807 if (!error && actionsp) {
808 actions = (*actionsp)->data;
809 actions_len = (*actionsp)->size;
814 log_flow_message(dpif, error, "flow_get", key, key_len,
815 NULL, 0, stats, actions, actions_len);
821 dpif_flow_put__(struct dpif *dpif, const struct dpif_flow_put *put)
825 COVERAGE_INC(dpif_flow_put);
826 ovs_assert(!(put->flags & ~(DPIF_FP_CREATE | DPIF_FP_MODIFY
827 | DPIF_FP_ZERO_STATS)));
829 error = dpif->dpif_class->flow_put(dpif, put);
830 if (error && put->stats) {
831 memset(put->stats, 0, sizeof *put->stats);
833 log_flow_put_message(dpif, put, error);
837 /* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
838 * attribute OVS_FLOW_ATTR_KEY with types OVS_KEY_ATTR_* in the 'key_len' bytes
839 * starting at 'key', and OVS_FLOW_ATTR_MASK with types of OVS_KEY_ATTR_* in the
840 * 'mask_len' bytes starting at 'mask'. The associated actions are specified by
841 * the Netlink attributes with types OVS_ACTION_ATTR_* in the 'actions_len'
842 * bytes starting at 'actions'.
844 * - If the flow's key does not exist in 'dpif', then the flow will be added if
845 * 'flags' includes DPIF_FP_CREATE. Otherwise the operation will fail with
848 * If the operation succeeds, then 'stats', if nonnull, will be zeroed.
850 * - If the flow's key does exist in 'dpif', then the flow's actions will be
851 * updated if 'flags' includes DPIF_FP_MODIFY. Otherwise the operation will
852 * fail with EEXIST. If the flow's actions are updated, then its statistics
853 * will be zeroed if 'flags' includes DPIF_FP_ZERO_STATS, and left as-is
856 * If the operation succeeds, then 'stats', if nonnull, will be set to the
857 * flow's statistics before the update.
860 dpif_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags,
861 const struct nlattr *key, size_t key_len,
862 const struct nlattr *mask, size_t mask_len,
863 const struct nlattr *actions, size_t actions_len,
864 struct dpif_flow_stats *stats)
866 struct dpif_flow_put put;
870 put.key_len = key_len;
872 put.mask_len = mask_len;
873 put.actions = actions;
874 put.actions_len = actions_len;
876 return dpif_flow_put__(dpif, &put);
880 dpif_flow_del__(struct dpif *dpif, struct dpif_flow_del *del)
884 COVERAGE_INC(dpif_flow_del);
886 error = dpif->dpif_class->flow_del(dpif, del);
887 if (error && del->stats) {
888 memset(del->stats, 0, sizeof *del->stats);
890 log_flow_del_message(dpif, del, error);
894 /* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif' does
895 * not contain such a flow. The flow is specified by the Netlink attributes
896 * with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at 'key'.
898 * If the operation succeeds, then 'stats', if nonnull, will be set to the
899 * flow's statistics before its deletion. */
901 dpif_flow_del(struct dpif *dpif,
902 const struct nlattr *key, size_t key_len,
903 struct dpif_flow_stats *stats)
905 struct dpif_flow_del del;
908 del.key_len = key_len;
910 return dpif_flow_del__(dpif, &del);
913 /* Initializes 'dump' to begin dumping the flows in a dpif.
915 * This function provides no status indication. An error status for the entire
916 * dump operation is provided when it is completed by calling
917 * dpif_flow_dump_done().
920 dpif_flow_dump_start(struct dpif_flow_dump *dump, const struct dpif *dpif)
923 dump->error = dpif->dpif_class->flow_dump_start(dpif, &dump->state);
924 log_operation(dpif, "flow_dump_start", dump->error);
927 /* Attempts to retrieve another flow from 'dump', which must have been
928 * initialized with dpif_flow_dump_start(). On success, updates the output
929 * parameters as described below and returns true. Otherwise, returns false.
930 * Failure might indicate an actual error or merely the end of the flow table.
931 * An error status for the entire dump operation is provided when it is
932 * completed by calling dpif_flow_dump_done().
934 * On success, if 'key' and 'key_len' are nonnull then '*key' and '*key_len'
935 * will be set to Netlink attributes with types OVS_KEY_ATTR_* representing the
936 * dumped flow's key. If 'actions' and 'actions_len' are nonnull then they are
937 * set to Netlink attributes with types OVS_ACTION_ATTR_* representing the
938 * dumped flow's actions. If 'stats' is nonnull then it will be set to the
939 * dumped flow's statistics.
941 * All of the returned data is owned by 'dpif', not by the caller, and the
942 * caller must not modify or free it. 'dpif' guarantees that it remains
943 * accessible and unchanging until at least the next call to 'flow_dump_next'
944 * or 'flow_dump_done' for 'dump'. */
946 dpif_flow_dump_next(struct dpif_flow_dump *dump,
947 const struct nlattr **key, size_t *key_len,
948 const struct nlattr **mask, size_t *mask_len,
949 const struct nlattr **actions, size_t *actions_len,
950 const struct dpif_flow_stats **stats)
952 const struct dpif *dpif = dump->dpif;
953 int error = dump->error;
956 error = dpif->dpif_class->flow_dump_next(dpif, dump->state,
959 actions, actions_len,
962 dpif->dpif_class->flow_dump_done(dpif, dump->state);
984 VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif));
985 } else if (should_log_flow_message(error)) {
986 log_flow_message(dpif, error, "flow_dump",
987 key ? *key : NULL, key ? *key_len : 0,
988 mask ? *mask : NULL, mask ? *mask_len : 0,
989 stats ? *stats : NULL, actions ? *actions : NULL,
990 actions ? *actions_len : 0);
997 /* Completes flow table dump operation 'dump', which must have been initialized
998 * with dpif_flow_dump_start(). Returns 0 if the dump operation was
999 * error-free, otherwise a positive errno value describing the problem. */
1001 dpif_flow_dump_done(struct dpif_flow_dump *dump)
1003 const struct dpif *dpif = dump->dpif;
1005 dump->error = dpif->dpif_class->flow_dump_done(dpif, dump->state);
1006 log_operation(dpif, "flow_dump_done", dump->error);
1008 return dump->error == EOF ? 0 : dump->error;
1012 dpif_execute__(struct dpif *dpif, const struct dpif_execute *execute)
1016 COVERAGE_INC(dpif_execute);
1017 if (execute->actions_len > 0) {
1018 error = dpif->dpif_class->execute(dpif, execute);
1023 log_execute_message(dpif, execute, error);
1028 /* Causes 'dpif' to perform the 'actions_len' bytes of actions in 'actions' on
1029 * the Ethernet frame specified in 'packet' taken from the flow specified in
1030 * the 'key_len' bytes of 'key'. ('key' is mostly redundant with 'packet', but
1031 * it contains some metadata that cannot be recovered from 'packet', such as
1032 * tunnel and in_port.)
1034 * Returns 0 if successful, otherwise a positive errno value. */
1036 dpif_execute(struct dpif *dpif,
1037 const struct nlattr *key, size_t key_len,
1038 const struct nlattr *actions, size_t actions_len,
1039 const struct ofpbuf *buf)
1041 struct dpif_execute execute;
1044 execute.key_len = key_len;
1045 execute.actions = actions;
1046 execute.actions_len = actions_len;
1047 execute.packet = buf;
1048 return dpif_execute__(dpif, &execute);
1051 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order in
1052 * which they are specified, placing each operation's results in the "output"
1053 * members documented in comments.
1055 * This function exists because some datapaths can perform batched operations
1056 * faster than individual operations. */
1058 dpif_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops)
1062 if (dpif->dpif_class->operate) {
1063 dpif->dpif_class->operate(dpif, ops, n_ops);
1065 for (i = 0; i < n_ops; i++) {
1066 struct dpif_op *op = ops[i];
1069 case DPIF_OP_FLOW_PUT:
1070 log_flow_put_message(dpif, &op->u.flow_put, op->error);
1073 case DPIF_OP_FLOW_DEL:
1074 log_flow_del_message(dpif, &op->u.flow_del, op->error);
1077 case DPIF_OP_EXECUTE:
1078 log_execute_message(dpif, &op->u.execute, op->error);
1085 for (i = 0; i < n_ops; i++) {
1086 struct dpif_op *op = ops[i];
1089 case DPIF_OP_FLOW_PUT:
1090 op->error = dpif_flow_put__(dpif, &op->u.flow_put);
1093 case DPIF_OP_FLOW_DEL:
1094 op->error = dpif_flow_del__(dpif, &op->u.flow_del);
1097 case DPIF_OP_EXECUTE:
1098 op->error = dpif_execute__(dpif, &op->u.execute);
1108 /* Returns a string that represents 'type', for use in log messages. */
1110 dpif_upcall_type_to_string(enum dpif_upcall_type type)
1113 case DPIF_UC_MISS: return "miss";
1114 case DPIF_UC_ACTION: return "action";
1115 case DPIF_N_UC_TYPES: default: return "<unknown>";
1119 /* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0
1120 * if successful, otherwise a positive errno value.
1122 * Turning packet receive off and then back on may change the Netlink PID
1123 * assignments returned by dpif_port_get_pid(). If the client does this, it
1124 * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions
1125 * using the new PID assignment. */
1127 dpif_recv_set(struct dpif *dpif, bool enable)
1129 int error = dpif->dpif_class->recv_set(dpif, enable);
1130 log_operation(dpif, "recv_set", error);
1134 /* Polls for an upcall from 'dpif'. If successful, stores the upcall into
1135 * '*upcall', using 'buf' for storage. Should only be called if
1136 * dpif_recv_set() has been used to enable receiving packets on 'dpif'.
1138 * 'upcall->packet' and 'upcall->key' point into data in the caller-provided
1139 * 'buf', so their memory cannot be freed separately from 'buf'. (This is
1140 * hardly a great way to do things but it works out OK for the dpif providers
1141 * and clients that exist so far.)
1143 * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN
1144 * if no upcall is immediately available. */
1146 dpif_recv(struct dpif *dpif, struct dpif_upcall *upcall, struct ofpbuf *buf)
1148 int error = dpif->dpif_class->recv(dpif, upcall, buf);
1149 if (!error && !VLOG_DROP_DBG(&dpmsg_rl)) {
1153 packet = ofp_packet_to_string(upcall->packet->data,
1154 upcall->packet->size);
1157 odp_flow_key_format(upcall->key, upcall->key_len, &flow);
1159 VLOG_DBG("%s: %s upcall:\n%s\n%s",
1160 dpif_name(dpif), dpif_upcall_type_to_string(upcall->type),
1161 ds_cstr(&flow), packet);
1165 } else if (error && error != EAGAIN) {
1166 log_operation(dpif, "recv", error);
1171 /* Discards all messages that would otherwise be received by dpif_recv() on
1174 dpif_recv_purge(struct dpif *dpif)
1176 COVERAGE_INC(dpif_purge);
1177 if (dpif->dpif_class->recv_purge) {
1178 dpif->dpif_class->recv_purge(dpif);
1182 /* Arranges for the poll loop to wake up when 'dpif' has a message queued to be
1183 * received with dpif_recv(). */
1185 dpif_recv_wait(struct dpif *dpif)
1187 dpif->dpif_class->recv_wait(dpif);
1190 /* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type'
1191 * and '*engine_id', respectively. */
1193 dpif_get_netflow_ids(const struct dpif *dpif,
1194 uint8_t *engine_type, uint8_t *engine_id)
1196 *engine_type = dpif->netflow_engine_type;
1197 *engine_id = dpif->netflow_engine_id;
1200 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority
1201 * value used for setting packet priority.
1202 * On success, returns 0 and stores the priority into '*priority'.
1203 * On failure, returns a positive errno value and stores 0 into '*priority'. */
1205 dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id,
1208 int error = (dpif->dpif_class->queue_to_priority
1209 ? dpif->dpif_class->queue_to_priority(dpif, queue_id,
1215 log_operation(dpif, "queue_to_priority", error);
1220 dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class,
1222 uint8_t netflow_engine_type, uint8_t netflow_engine_id)
1224 dpif->dpif_class = dpif_class;
1225 dpif->base_name = xstrdup(name);
1226 dpif->full_name = xasprintf("%s@%s", dpif_class->type, name);
1227 dpif->netflow_engine_type = netflow_engine_type;
1228 dpif->netflow_engine_id = netflow_engine_id;
1231 /* Undoes the results of initialization.
1233 * Normally this function only needs to be called from dpif_close().
1234 * However, it may be called by providers due to an error on opening
1235 * that occurs after initialization. It this case dpif_close() would
1236 * never be called. */
1238 dpif_uninit(struct dpif *dpif, bool close)
1240 char *base_name = dpif->base_name;
1241 char *full_name = dpif->full_name;
1244 dpif->dpif_class->close(dpif);
1252 log_operation(const struct dpif *dpif, const char *operation, int error)
1255 VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation);
1256 } else if (ofperr_is_valid(error)) {
1257 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1258 dpif_name(dpif), operation, ofperr_get_name(error));
1260 VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)",
1261 dpif_name(dpif), operation, ovs_strerror(error));
1265 static enum vlog_level
1266 flow_message_log_level(int error)
1268 /* If flows arrive in a batch, userspace may push down multiple
1269 * unique flow definitions that overlap when wildcards are applied.
1270 * Kernels that support flow wildcarding will reject these flows as
1271 * duplicates (EEXIST), so lower the log level to debug for these
1272 * types of messages. */
1273 return (error && error != EEXIST) ? VLL_WARN : VLL_DBG;
1277 should_log_flow_message(int error)
1279 return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error),
1280 error ? &error_rl : &dpmsg_rl);
1284 log_flow_message(const struct dpif *dpif, int error, const char *operation,
1285 const struct nlattr *key, size_t key_len,
1286 const struct nlattr *mask, size_t mask_len,
1287 const struct dpif_flow_stats *stats,
1288 const struct nlattr *actions, size_t actions_len)
1290 struct ds ds = DS_EMPTY_INITIALIZER;
1291 ds_put_format(&ds, "%s: ", dpif_name(dpif));
1293 ds_put_cstr(&ds, "failed to ");
1295 ds_put_format(&ds, "%s ", operation);
1297 ds_put_format(&ds, "(%s) ", ovs_strerror(error));
1299 odp_flow_format(key, key_len, mask, mask_len, &ds);
1301 ds_put_cstr(&ds, ", ");
1302 dpif_flow_stats_format(stats, &ds);
1304 if (actions || actions_len) {
1305 ds_put_cstr(&ds, ", actions:");
1306 format_odp_actions(&ds, actions, actions_len);
1308 vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds));
1313 log_flow_put_message(struct dpif *dpif, const struct dpif_flow_put *put,
1316 if (should_log_flow_message(error)) {
1320 ds_put_cstr(&s, "put");
1321 if (put->flags & DPIF_FP_CREATE) {
1322 ds_put_cstr(&s, "[create]");
1324 if (put->flags & DPIF_FP_MODIFY) {
1325 ds_put_cstr(&s, "[modify]");
1327 if (put->flags & DPIF_FP_ZERO_STATS) {
1328 ds_put_cstr(&s, "[zero]");
1330 log_flow_message(dpif, error, ds_cstr(&s),
1331 put->key, put->key_len, put->mask, put->mask_len,
1332 put->stats, put->actions, put->actions_len);
1338 log_flow_del_message(struct dpif *dpif, const struct dpif_flow_del *del,
1341 if (should_log_flow_message(error)) {
1342 log_flow_message(dpif, error, "flow_del", del->key, del->key_len,
1343 NULL, 0, !error ? del->stats : NULL, NULL, 0);
1348 log_execute_message(struct dpif *dpif, const struct dpif_execute *execute,
1351 if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) {
1352 struct ds ds = DS_EMPTY_INITIALIZER;
1355 packet = ofp_packet_to_string(execute->packet->data,
1356 execute->packet->size);
1357 ds_put_format(&ds, "%s: execute ", dpif_name(dpif));
1358 format_odp_actions(&ds, execute->actions, execute->actions_len);
1360 ds_put_format(&ds, " failed (%s)", ovs_strerror(error));
1362 ds_put_format(&ds, " on packet %s", packet);
1363 vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds));