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'.
151 * A 'port_no' of UINT32_MAX should be treated as a special case. The
152 * implementation should return a reserved PID, not allocated to any port,
153 * that the client may use for special purposes.
155 * The return value only needs to be meaningful when DPIF_UC_ACTION has
156 * been enabled in the 'dpif''s listen mask, and it is allowed to change
157 * when DPIF_UC_ACTION is disabled and then re-enabled.
159 * A dpif provider that doesn't have meaningful Netlink PIDs can use NULL
160 * for this function. This is equivalent to always returning 0. */
161 uint32_t (*port_get_pid)(const struct dpif *dpif, odp_port_t port_no);
163 /* Attempts to begin dumping the ports in a dpif. On success, returns 0
164 * and initializes '*statep' with any data needed for iteration. On
165 * failure, returns a positive errno value. */
166 int (*port_dump_start)(const struct dpif *dpif, void **statep);
168 /* Attempts to retrieve another port from 'dpif' for 'state', which was
169 * initialized by a successful call to the 'port_dump_start' function for
170 * 'dpif'. On success, stores a new dpif_port into 'port' and returns 0.
171 * Returns EOF if the end of the port table has been reached, or a positive
172 * errno value on error. This function will not be called again once it
173 * returns nonzero once for a given iteration (but the 'port_dump_done'
174 * function will be called afterward).
176 * The dpif provider retains ownership of the data stored in 'port'. It
177 * must remain valid until at least the next call to 'port_dump_next' or
178 * 'port_dump_done' for 'state'. */
179 int (*port_dump_next)(const struct dpif *dpif, void *state,
180 struct dpif_port *port);
182 /* Releases resources from 'dpif' for 'state', which was initialized by a
183 * successful call to the 'port_dump_start' function for 'dpif'. */
184 int (*port_dump_done)(const struct dpif *dpif, void *state);
186 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
187 * 'dpif' has changed, then this function should do one of the
190 * - Preferably: store the name of the device that was added to or deleted
191 * from 'dpif' in '*devnamep' and return 0. The caller is responsible
192 * for freeing '*devnamep' (with free()) when it no longer needs it.
194 * - Alternatively: return ENOBUFS, without indicating the device that was
197 * Occasional 'false positives', in which the function returns 0 while
198 * indicating a device that was not actually added or deleted or returns
199 * ENOBUFS without any change, are acceptable.
201 * If the set of ports in 'dpif' has not changed, returns EAGAIN. May also
202 * return other positive errno values to indicate that something has gone
204 int (*port_poll)(const struct dpif *dpif, char **devnamep);
206 /* Arranges for the poll loop to wake up when 'port_poll' will return a
207 * value other than EAGAIN. */
208 void (*port_poll_wait)(const struct dpif *dpif);
210 /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
211 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
214 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
215 * failure, returns a positive errno value.
217 * If 'actionsp' is nonnull, then on success '*actionsp' must be set to an
218 * ofpbuf owned by the caller that contains the Netlink attributes for the
219 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete())
220 * when it is no longer needed.
222 * If 'stats' is nonnull, then on success it must be updated with the
223 * flow's statistics. */
224 int (*flow_get)(const struct dpif *dpif,
225 const struct nlattr *key, size_t key_len,
226 struct ofpbuf **actionsp, struct dpif_flow_stats *stats);
228 /* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
229 * attributes with types OVS_KEY_ATTR_* in the 'put->key_len' bytes
230 * starting at 'put->key'. The associated actions are specified by the
231 * Netlink attributes with types OVS_ACTION_ATTR_* in the
232 * 'put->actions_len' bytes starting at 'put->actions'.
234 * - If the flow's key does not exist in 'dpif', then the flow will be
235 * added if 'put->flags' includes DPIF_FP_CREATE. Otherwise the
236 * operation will fail with ENOENT.
238 * If the operation succeeds, then 'put->stats', if nonnull, must be
241 * - If the flow's key does exist in 'dpif', then the flow's actions will
242 * be updated if 'put->flags' includes DPIF_FP_MODIFY. Otherwise the
243 * operation will fail with EEXIST. If the flow's actions are updated,
244 * then its statistics will be zeroed if 'put->flags' includes
245 * DPIF_FP_ZERO_STATS, and left as-is otherwise.
247 * If the operation succeeds, then 'put->stats', if nonnull, must be set
248 * to the flow's statistics before the update.
250 int (*flow_put)(struct dpif *dpif, const struct dpif_flow_put *put);
252 /* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif'
253 * does not contain such a flow. The flow is specified by the Netlink
254 * attributes with types OVS_KEY_ATTR_* in the 'del->key_len' bytes
255 * starting at 'del->key'.
257 * If the operation succeeds, then 'del->stats', if nonnull, must be set to
258 * the flow's statistics before its deletion. */
259 int (*flow_del)(struct dpif *dpif, const struct dpif_flow_del *del);
261 /* Deletes all flows from 'dpif' and clears all of its queues of received
263 int (*flow_flush)(struct dpif *dpif);
265 /* Allocates thread-local state for use with the function 'flow_dump_next'.
266 * On return, initializes '*statep' with any private data needed for
268 void (*flow_dump_state_init)(void **statep);
270 /* Attempts to begin dumping the flows in a dpif. On success, returns 0
271 * and initializes '*iterp' with any shared data needed for iteration.
272 * On failure, returns a positive errno value. */
273 int (*flow_dump_start)(const struct dpif *dpif, void **iterp);
275 /* Attempts to retrieve another flow from 'dpif' for 'iter', using
276 * 'state' for storage. 'iter' must have been initialized by a successful
277 * call to the 'flow_dump_start' function for 'dpif'. 'state' must have
278 * been initialised with a call to the 'flow_dump_state_init' function for
281 * On success, updates the output parameters as described below and returns
282 * 0. Returns EOF if the end of the flow table has been reached, or a
283 * positive errno value on error. Multiple threads may use the same 'dpif'
284 * and 'iter' with this function, but all other parameters must be
285 * different for each thread. If this function returns non-zero,
286 * subsequent calls with the same arguments will also return non-zero.
290 * - If 'key' and 'key_len' are nonnull, then '*key' and '*key_len'
291 * must be set to Netlink attributes with types OVS_KEY_ATTR_*
292 * representing the dumped flow's key.
294 * - If 'mask' and 'mask_len' are nonnull then '*mask' and '*mask_len'
295 * must be set to Netlink attributes with types of OVS_KEY_ATTR_*
296 * representing the dumped flow's mask.
298 * - If 'actions' and 'actions_len' are nonnull then they should be set
299 * to Netlink attributes with types OVS_ACTION_ATTR_* representing
300 * the dumped flow's actions.
302 * - If 'stats' is nonnull then it should be set to the dumped flow's
305 * All of the returned data is owned by 'dpif', not by the caller, and the
306 * caller must not modify or free it. 'dpif' must guarantee that it
307 * remains accessible and unchanging until at least the next call to
308 * 'flow_dump_next' or 'flow_dump_done' for 'iter' and 'state'. */
309 int (*flow_dump_next)(const struct dpif *dpif, void *iter, void *state,
310 const struct nlattr **key, size_t *key_len,
311 const struct nlattr **mask, size_t *mask_len,
312 const struct nlattr **actions, size_t *actions_len,
313 const struct dpif_flow_stats **stats);
315 /* Determines whether the next call to 'flow_dump_next' with 'state' will
316 * modify or free the keys that it previously returned. 'state' must have
317 * been initialized by a call to 'flow_dump_state_init' for 'dpif'.
319 * 'dpif' guarantees that data returned by flow_dump_next() will remain
320 * accessible and unchanging until the next call. This function provides a
321 * way for callers to determine whether that guarantee extends beyond the
324 * Returns true if the next call to flow_dump_next() is expected to be
325 * destructive to previously returned keys for 'state', false otherwise. */
326 bool (*flow_dump_next_may_destroy_keys)(void *state);
328 /* Releases resources from 'dpif' for 'iter', which was initialized by a
329 * successful call to the 'flow_dump_start' function for 'dpif'. Callers
330 * must ensure that this function is called once within a given iteration,
331 * as the final flow dump operation. */
332 int (*flow_dump_done)(const struct dpif *dpif, void *iter);
334 /* Releases 'state' which was initialized by a call to the
335 * 'flow_dump_state_init' function for this 'dpif'. */
336 void (*flow_dump_state_uninit)(void *statep);
338 /* Performs the 'execute->actions_len' bytes of actions in
339 * 'execute->actions' on the Ethernet frame in 'execute->packet'
340 * and on the packet metadata in 'execute->md'.
341 * May modify both packet and metadata. */
342 int (*execute)(struct dpif *dpif, struct dpif_execute *execute);
344 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order
345 * in which they are specified, placing each operation's results in the
346 * "output" members documented in comments.
348 * This function is optional. It is only worthwhile to implement it if
349 * 'dpif' can perform operations in batch faster than individually. */
350 void (*operate)(struct dpif *dpif, struct dpif_op **ops, size_t n_ops);
352 /* Enables or disables receiving packets with dpif_recv() for 'dpif'.
353 * Turning packet receive off and then back on is allowed to change Netlink
354 * PID assignments (see ->port_get_pid()). The client is responsible for
355 * updating flows as necessary if it does this. */
356 int (*recv_set)(struct dpif *dpif, bool enable);
358 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
359 * priority value used for setting packet priority. */
360 int (*queue_to_priority)(const struct dpif *dpif, uint32_t queue_id,
363 /* Polls for an upcall from 'dpif'. If successful, stores the upcall into
364 * '*upcall', using 'buf' for storage. Should only be called if 'recv_set'
365 * has been used to enable receiving packets from 'dpif'.
367 * The implementation should point 'upcall->key' and 'upcall->userdata'
368 * (if any) into data in the caller-provided 'buf'. The implementation may
369 * also use 'buf' for storing the data of 'upcall->packet'. If necessary
370 * to make room, the implementation may reallocate the data in 'buf'.
372 * The caller owns the data of 'upcall->packet' and may modify it. If
373 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
374 * will be reallocated. This requires the data of 'upcall->packet' to be
375 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
376 * when an error is returned, the 'upcall->packet' may be uninitialized
377 * and should not be released.
379 * This function must not block. If no upcall is pending when it is
380 * called, it should return EAGAIN without blocking. */
381 int (*recv)(struct dpif *dpif, struct dpif_upcall *upcall,
384 /* Arranges for the poll loop to wake up when 'dpif' has a message queued
385 * to be received with the recv member function. */
386 void (*recv_wait)(struct dpif *dpif);
388 /* Throws away any queued upcalls that 'dpif' currently has ready to
390 void (*recv_purge)(struct dpif *dpif);
393 extern const struct dpif_class dpif_linux_class;
394 extern const struct dpif_class dpif_netdev_class;
400 #endif /* dpif-provider.h */