2 * Copyright (c) 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.
17 #ifndef DPIF_PROVIDER_H
18 #define DPIF_PROVIDER_H 1
20 /* Provider interface to dpifs, which provide an interface to an Open vSwitch
21 * datapath. A datapath is a collection of physical or virtual ports that are
22 * exposed over OpenFlow as a single switch. Datapaths and the collections of
23 * ports that they contain may be fixed or dynamic. */
25 #include "openflow/openflow.h"
33 /* Open vSwitch datapath interface.
35 * This structure should be treated as opaque by dpif implementations. */
37 const struct dpif_class *dpif_class;
40 uint8_t netflow_engine_type;
41 uint8_t netflow_engine_id;
44 void dpif_init(struct dpif *, const struct dpif_class *, const char *name,
45 uint8_t netflow_engine_type, uint8_t netflow_engine_id);
46 void dpif_uninit(struct dpif *dpif, bool close);
48 static inline void dpif_assert_class(const struct dpif *dpif,
49 const struct dpif_class *dpif_class)
51 ovs_assert(dpif->dpif_class == dpif_class);
54 /* Datapath interface class structure, to be defined by each implementation of
55 * a datapath interface.
57 * These functions return 0 if successful or a positive errno value on failure,
58 * except where otherwise noted.
60 * These functions are expected to execute synchronously, that is, to block as
61 * necessary to obtain a result. Thus, they may not return EAGAIN or
62 * EWOULDBLOCK or EINPROGRESS. We may relax this requirement in the future if
63 * and when we encounter performance problems. */
65 /* Type of dpif in this class, e.g. "system", "netdev", etc.
67 * One of the providers should supply a "system" type, since this is
68 * the type assumed if no type is specified when opening a dpif. */
71 /* Enumerates the names of all known created datapaths, if possible, into
72 * 'all_dps'. The caller has already initialized 'all_dps' and other dpif
73 * classes might already have added names to it.
75 * This is used by the vswitch at startup, so that it can delete any
76 * datapaths that are not configured.
78 * Some kinds of datapaths might not be practically enumerable, in which
79 * case this function may be a null pointer. */
80 int (*enumerate)(struct sset *all_dps);
82 /* Returns the type to pass to netdev_open() when a dpif of class
83 * 'dpif_class' has a port of type 'type', for a few special cases
84 * when a netdev type differs from a port type. For example, when
85 * using the userspace datapath, a port of type "internal" needs to
88 * Returns either 'type' itself or a string literal, which must not
90 const char *(*port_open_type)(const struct dpif_class *dpif_class,
93 /* Attempts to open an existing dpif called 'name', if 'create' is false,
94 * or to open an existing dpif or create a new one, if 'create' is true.
96 * 'dpif_class' is the class of dpif to open.
98 * If successful, stores a pointer to the new dpif in '*dpifp', which must
99 * have class 'dpif_class'. On failure there are no requirements on what
100 * is stored in '*dpifp'. */
101 int (*open)(const struct dpif_class *dpif_class,
102 const char *name, bool create, struct dpif **dpifp);
104 /* Closes 'dpif' and frees associated memory. */
105 void (*close)(struct dpif *dpif);
107 /* Attempts to destroy the dpif underlying 'dpif'.
109 * If successful, 'dpif' will not be used again except as an argument for
110 * the 'close' member function. */
111 int (*destroy)(struct dpif *dpif);
113 /* Performs periodic work needed by 'dpif', if any is necessary. */
114 void (*run)(struct dpif *dpif);
116 /* Arranges for poll_block() to wake up if the "run" member function needs
117 * to be called for 'dpif'. */
118 void (*wait)(struct dpif *dpif);
120 /* Retrieves statistics for 'dpif' into 'stats'. */
121 int (*get_stats)(const struct dpif *dpif, struct dpif_dp_stats *stats);
123 /* Adds 'netdev' as a new port in 'dpif'. If '*port_no' is not
124 * UINT32_MAX, attempts to use that as the port's port number.
126 * If port is successfully added, sets '*port_no' to the new port's
127 * port number. Returns EBUSY if caller attempted to choose a port
128 * number, and it was in use. */
129 int (*port_add)(struct dpif *dpif, struct netdev *netdev,
130 odp_port_t *port_no);
132 /* Removes port numbered 'port_no' from 'dpif'. */
133 int (*port_del)(struct dpif *dpif, odp_port_t port_no);
135 /* Queries 'dpif' for a port with the given 'port_no' or 'devname'.
136 * If 'port' is not null, stores information about the port into
137 * '*port' if successful.
139 * If 'port' is not null, the caller takes ownership of data in
140 * 'port' and must free it with dpif_port_destroy() when it is no
142 int (*port_query_by_number)(const struct dpif *dpif, odp_port_t port_no,
143 struct dpif_port *port);
144 int (*port_query_by_name)(const struct dpif *dpif, const char *devname,
145 struct dpif_port *port);
147 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
148 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
149 * flows whose packets arrived on port 'port_no'. In the case where the
150 * provider allocates multiple Netlink PIDs to a single port, it may use
151 * 'hash' to spread load among them. The caller need not use a particular
152 * hash function; a 5-tuple hash is suitable.
154 * (The datapath implementation might use some different hash function for
155 * distributing packets received via flow misses among PIDs. This means
156 * that packets received via flow misses might be reordered relative to
157 * packets received via userspace actions. This is not ordinarily a
160 * A 'port_no' of UINT32_MAX should be treated as a special case. The
161 * implementation should return a reserved PID, not allocated to any port,
162 * that the client may use for special purposes.
164 * The return value only needs to be meaningful when DPIF_UC_ACTION has
165 * been enabled in the 'dpif''s listen mask, and it is allowed to change
166 * when DPIF_UC_ACTION is disabled and then re-enabled.
168 * A dpif provider that doesn't have meaningful Netlink PIDs can use NULL
169 * for this function. This is equivalent to always returning 0. */
170 uint32_t (*port_get_pid)(const struct dpif *dpif, odp_port_t port_no,
173 /* Attempts to begin dumping the ports in a dpif. On success, returns 0
174 * and initializes '*statep' with any data needed for iteration. On
175 * failure, returns a positive errno value. */
176 int (*port_dump_start)(const struct dpif *dpif, void **statep);
178 /* Attempts to retrieve another port from 'dpif' for 'state', which was
179 * initialized by a successful call to the 'port_dump_start' function for
180 * 'dpif'. On success, stores a new dpif_port into 'port' and returns 0.
181 * Returns EOF if the end of the port table has been reached, or a positive
182 * errno value on error. This function will not be called again once it
183 * returns nonzero once for a given iteration (but the 'port_dump_done'
184 * function will be called afterward).
186 * The dpif provider retains ownership of the data stored in 'port'. It
187 * must remain valid until at least the next call to 'port_dump_next' or
188 * 'port_dump_done' for 'state'. */
189 int (*port_dump_next)(const struct dpif *dpif, void *state,
190 struct dpif_port *port);
192 /* Releases resources from 'dpif' for 'state', which was initialized by a
193 * successful call to the 'port_dump_start' function for 'dpif'. */
194 int (*port_dump_done)(const struct dpif *dpif, void *state);
196 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
197 * 'dpif' has changed, then this function should do one of the
200 * - Preferably: store the name of the device that was added to or deleted
201 * from 'dpif' in '*devnamep' and return 0. The caller is responsible
202 * for freeing '*devnamep' (with free()) when it no longer needs it.
204 * - Alternatively: return ENOBUFS, without indicating the device that was
207 * Occasional 'false positives', in which the function returns 0 while
208 * indicating a device that was not actually added or deleted or returns
209 * ENOBUFS without any change, are acceptable.
211 * If the set of ports in 'dpif' has not changed, returns EAGAIN. May also
212 * return other positive errno values to indicate that something has gone
214 int (*port_poll)(const struct dpif *dpif, char **devnamep);
216 /* Arranges for the poll loop to wake up when 'port_poll' will return a
217 * value other than EAGAIN. */
218 void (*port_poll_wait)(const struct dpif *dpif);
220 /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
221 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
224 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
225 * failure, returns a positive errno value.
227 * If 'actionsp' is nonnull, then on success '*actionsp' must be set to an
228 * ofpbuf owned by the caller that contains the Netlink attributes for the
229 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete())
230 * when it is no longer needed.
232 * If 'stats' is nonnull, then on success it must be updated with the
233 * flow's statistics. */
234 int (*flow_get)(const struct dpif *dpif,
235 const struct nlattr *key, size_t key_len,
236 struct ofpbuf **actionsp, struct dpif_flow_stats *stats);
238 /* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
239 * attributes with types OVS_KEY_ATTR_* in the 'put->key_len' bytes
240 * starting at 'put->key'. The associated actions are specified by the
241 * Netlink attributes with types OVS_ACTION_ATTR_* in the
242 * 'put->actions_len' bytes starting at 'put->actions'.
244 * - If the flow's key does not exist in 'dpif', then the flow will be
245 * added if 'put->flags' includes DPIF_FP_CREATE. Otherwise the
246 * operation will fail with ENOENT.
248 * If the operation succeeds, then 'put->stats', if nonnull, must be
251 * - If the flow's key does exist in 'dpif', then the flow's actions will
252 * be updated if 'put->flags' includes DPIF_FP_MODIFY. Otherwise the
253 * operation will fail with EEXIST. If the flow's actions are updated,
254 * then its statistics will be zeroed if 'put->flags' includes
255 * DPIF_FP_ZERO_STATS, and left as-is otherwise.
257 * If the operation succeeds, then 'put->stats', if nonnull, must be set
258 * to the flow's statistics before the update.
260 int (*flow_put)(struct dpif *dpif, const struct dpif_flow_put *put);
262 /* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif'
263 * does not contain such a flow. The flow is specified by the Netlink
264 * attributes with types OVS_KEY_ATTR_* in the 'del->key_len' bytes
265 * starting at 'del->key'.
267 * If the operation succeeds, then 'del->stats', if nonnull, must be set to
268 * the flow's statistics before its deletion. */
269 int (*flow_del)(struct dpif *dpif, const struct dpif_flow_del *del);
271 /* Deletes all flows from 'dpif' and clears all of its queues of received
273 int (*flow_flush)(struct dpif *dpif);
275 /* Allocates thread-local state for use with the function 'flow_dump_next'.
276 * On return, initializes '*statep' with any private data needed for
278 void (*flow_dump_state_init)(void **statep);
280 /* Attempts to begin dumping the flows in a dpif. On success, returns 0
281 * and initializes '*iterp' with any shared data needed for iteration.
282 * On failure, returns a positive errno value. */
283 int (*flow_dump_start)(const struct dpif *dpif, void **iterp);
285 /* Attempts to retrieve another flow from 'dpif' for 'iter', using
286 * 'state' for storage. 'iter' must have been initialized by a successful
287 * call to the 'flow_dump_start' function for 'dpif'. 'state' must have
288 * been initialised with a call to the 'flow_dump_state_init' function for
291 * On success, updates the output parameters as described below and returns
292 * 0. Returns EOF if the end of the flow table has been reached, or a
293 * positive errno value on error. Multiple threads may use the same 'dpif'
294 * and 'iter' with this function, but all other parameters must be
295 * different for each thread. If this function returns non-zero,
296 * subsequent calls with the same arguments will also return non-zero.
300 * - If 'key' and 'key_len' are nonnull, then '*key' and '*key_len'
301 * must be set to Netlink attributes with types OVS_KEY_ATTR_*
302 * representing the dumped flow's key.
304 * - If 'mask' and 'mask_len' are nonnull then '*mask' and '*mask_len'
305 * must be set to Netlink attributes with types of OVS_KEY_ATTR_*
306 * representing the dumped flow's mask.
308 * - If 'actions' and 'actions_len' are nonnull then they should be set
309 * to Netlink attributes with types OVS_ACTION_ATTR_* representing
310 * the dumped flow's actions.
312 * - If 'stats' is nonnull then it should be set to the dumped flow's
315 * All of the returned data is owned by 'dpif', not by the caller, and the
316 * caller must not modify or free it. 'dpif' must guarantee that it
317 * remains accessible and unchanging until at least the next call to
318 * 'flow_dump_next' or 'flow_dump_done' for 'iter' and 'state'. */
319 int (*flow_dump_next)(const struct dpif *dpif, void *iter, void *state,
320 const struct nlattr **key, size_t *key_len,
321 const struct nlattr **mask, size_t *mask_len,
322 const struct nlattr **actions, size_t *actions_len,
323 const struct dpif_flow_stats **stats);
325 /* Determines whether the next call to 'flow_dump_next' with 'state' will
326 * modify or free the keys that it previously returned. 'state' must have
327 * been initialized by a call to 'flow_dump_state_init' for 'dpif'.
329 * 'dpif' guarantees that data returned by flow_dump_next() will remain
330 * accessible and unchanging until the next call. This function provides a
331 * way for callers to determine whether that guarantee extends beyond the
334 * Returns true if the next call to flow_dump_next() is expected to be
335 * destructive to previously returned keys for 'state', false otherwise. */
336 bool (*flow_dump_next_may_destroy_keys)(void *state);
338 /* Releases resources from 'dpif' for 'iter', which was initialized by a
339 * successful call to the 'flow_dump_start' function for 'dpif'. Callers
340 * must ensure that this function is called once within a given iteration,
341 * as the final flow dump operation. */
342 int (*flow_dump_done)(const struct dpif *dpif, void *iter);
344 /* Releases 'state' which was initialized by a call to the
345 * 'flow_dump_state_init' function for this 'dpif'. */
346 void (*flow_dump_state_uninit)(void *statep);
348 /* Performs the 'execute->actions_len' bytes of actions in
349 * 'execute->actions' on the Ethernet frame in 'execute->packet'
350 * and on the packet metadata in 'execute->md'.
351 * May modify both packet and metadata. */
352 int (*execute)(struct dpif *dpif, struct dpif_execute *execute);
354 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order
355 * in which they are specified, placing each operation's results in the
356 * "output" members documented in comments.
358 * This function is optional. It is only worthwhile to implement it if
359 * 'dpif' can perform operations in batch faster than individually. */
360 void (*operate)(struct dpif *dpif, struct dpif_op **ops, size_t n_ops);
362 /* Enables or disables receiving packets with dpif_recv() for 'dpif'.
363 * Turning packet receive off and then back on is allowed to change Netlink
364 * PID assignments (see ->port_get_pid()). The client is responsible for
365 * updating flows as necessary if it does this. */
366 int (*recv_set)(struct dpif *dpif, bool enable);
368 /* Refreshes the poll loops and Netlink sockets associated to each port,
369 * when the number of upcall handlers (upcall receiving thread) is changed
370 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
373 * Since multiple upcall handlers can read upcalls simultaneously from
374 * 'dpif', each port can have multiple Netlink sockets, one per upcall
375 * handler. So, handlers_set() is responsible for the following tasks:
377 * When receiving upcall is enabled, extends or creates the
378 * configuration to support:
380 * - 'n_handlers' Netlink sockets for each port.
382 * - 'n_handlers' poll loops, one for each upcall handler.
384 * - registering the Netlink sockets for the same upcall handler to
385 * the corresponding poll loop.
387 int (*handlers_set)(struct dpif *dpif, uint32_t n_handlers);
389 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
390 * priority value used for setting packet priority. */
391 int (*queue_to_priority)(const struct dpif *dpif, uint32_t queue_id,
394 /* Polls for an upcall from 'dpif' for an upcall handler. Since there
395 * can be multiple poll loops (see ->handlers_set()), 'handler_id' is
396 * needed as index to identify the corresponding poll loop. If
397 * successful, stores the upcall into '*upcall', using 'buf' for
398 * storage. Should only be called if 'recv_set' has been used to enable
399 * receiving packets from 'dpif'.
401 * The implementation should point 'upcall->key' and 'upcall->userdata'
402 * (if any) into data in the caller-provided 'buf'. The implementation may
403 * also use 'buf' for storing the data of 'upcall->packet'. If necessary
404 * to make room, the implementation may reallocate the data in 'buf'.
406 * The caller owns the data of 'upcall->packet' and may modify it. If
407 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
408 * will be reallocated. This requires the data of 'upcall->packet' to be
409 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
410 * when an error is returned, the 'upcall->packet' may be uninitialized
411 * and should not be released.
413 * This function must not block. If no upcall is pending when it is
414 * called, it should return EAGAIN without blocking. */
415 int (*recv)(struct dpif *dpif, uint32_t handler_id,
416 struct dpif_upcall *upcall, struct ofpbuf *buf);
418 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
419 * has a message queued to be received with the recv member functions.
420 * Since there can be multiple poll loops (see ->handlers_set()),
421 * 'handler_id' is needed as index to identify the corresponding poll loop.
423 void (*recv_wait)(struct dpif *dpif, uint32_t handler_id);
425 /* Throws away any queued upcalls that 'dpif' currently has ready to
427 void (*recv_purge)(struct dpif *dpif);
430 extern const struct dpif_class dpif_linux_class;
431 extern const struct dpif_class dpif_netdev_class;
437 #endif /* dpif-provider.h */