2 * Copyright (c) 2009, 2010, 2011, 2012 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. */
26 #include "openflow/openflow.h"
34 /* Open vSwitch datapath interface.
36 * This structure should be treated as opaque by dpif implementations. */
38 const struct dpif_class *dpif_class;
41 uint8_t netflow_engine_type;
42 uint8_t netflow_engine_id;
45 void dpif_init(struct dpif *, const struct dpif_class *, const char *name,
46 uint8_t netflow_engine_type, uint8_t netflow_engine_id);
47 void dpif_uninit(struct dpif *dpif, bool close);
49 static inline void dpif_assert_class(const struct dpif *dpif,
50 const struct dpif_class *dpif_class)
52 assert(dpif->dpif_class == dpif_class);
55 /* Datapath interface class structure, to be defined by each implementation of
56 * a datapath interface.
58 * These functions return 0 if successful or a positive errno value on failure,
59 * except where otherwise noted.
61 * These functions are expected to execute synchronously, that is, to block as
62 * necessary to obtain a result. Thus, they may not return EAGAIN or
63 * EWOULDBLOCK or EINPROGRESS. We may relax this requirement in the future if
64 * and when we encounter performance problems. */
66 /* Type of dpif in this class, e.g. "system", "netdev", etc.
68 * One of the providers should supply a "system" type, since this is
69 * the type assumed if no type is specified when opening a dpif. */
72 /* Enumerates the names of all known created datapaths, if possible, into
73 * 'all_dps'. The caller has already initialized 'all_dps' and other dpif
74 * classes might already have added names to it.
76 * This is used by the vswitch at startup, so that it can delete any
77 * datapaths that are not configured.
79 * Some kinds of datapaths might not be practically enumerable, in which
80 * case this function may be a null pointer. */
81 int (*enumerate)(struct sset *all_dps);
83 /* Attempts to open an existing dpif called 'name', if 'create' is false,
84 * or to open an existing dpif or create a new one, if 'create' is true.
86 * 'dpif_class' is the class of dpif to open.
88 * If successful, stores a pointer to the new dpif in '*dpifp', which must
89 * have class 'dpif_class'. On failure there are no requirements on what
90 * is stored in '*dpifp'. */
91 int (*open)(const struct dpif_class *dpif_class,
92 const char *name, bool create, struct dpif **dpifp);
94 /* Closes 'dpif' and frees associated memory. */
95 void (*close)(struct dpif *dpif);
97 /* Attempts to destroy the dpif underlying 'dpif'.
99 * If successful, 'dpif' will not be used again except as an argument for
100 * the 'close' member function. */
101 int (*destroy)(struct dpif *dpif);
103 /* Performs periodic work needed by 'dpif', if any is necessary. */
104 void (*run)(struct dpif *dpif);
106 /* Arranges for poll_block() to wake up if the "run" member function needs
107 * to be called for 'dpif'. */
108 void (*wait)(struct dpif *dpif);
110 /* Retrieves statistics for 'dpif' into 'stats'. */
111 int (*get_stats)(const struct dpif *dpif, struct dpif_dp_stats *stats);
113 /* Adds 'netdev' as a new port in 'dpif'. If '*port_no' is not
114 * UINT16_MAX, attempts to use that as the port's port number.
116 * If port is successfully added, sets '*port_no' to the new port's
117 * port number. Returns EBUSY if caller attempted to choose a port
118 * number, and it was in use. */
119 int (*port_add)(struct dpif *dpif, struct netdev *netdev,
122 /* Removes port numbered 'port_no' from 'dpif'. */
123 int (*port_del)(struct dpif *dpif, uint16_t port_no);
125 /* Queries 'dpif' for a port with the given 'port_no' or 'devname'. Stores
126 * information about the port into '*port' if successful.
128 * The caller takes ownership of data in 'port' and must free it with
129 * dpif_port_destroy() when it is no longer needed. */
130 int (*port_query_by_number)(const struct dpif *dpif, uint16_t port_no,
131 struct dpif_port *port);
132 int (*port_query_by_name)(const struct dpif *dpif, const char *devname,
133 struct dpif_port *port);
135 /* Returns one greater than the largest port number accepted in flow
137 int (*get_max_ports)(const struct dpif *dpif);
139 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
140 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
141 * flows whose packets arrived on port 'port_no'.
143 * A 'port_no' of UINT16_MAX should be treated as a special case. The
144 * implementation should return a reserved PID, not allocated to any port,
145 * that the client may use for special purposes.
147 * The return value only needs to be meaningful when DPIF_UC_ACTION has
148 * been enabled in the 'dpif''s listen mask, and it is allowed to change
149 * when DPIF_UC_ACTION is disabled and then re-enabled.
151 * A dpif provider that doesn't have meaningful Netlink PIDs can use NULL
152 * for this function. This is equivalent to always returning 0. */
153 uint32_t (*port_get_pid)(const struct dpif *dpif, uint16_t port_no);
155 /* Attempts to begin dumping the ports in a dpif. On success, returns 0
156 * and initializes '*statep' with any data needed for iteration. On
157 * failure, returns a positive errno value. */
158 int (*port_dump_start)(const struct dpif *dpif, void **statep);
160 /* Attempts to retrieve another port from 'dpif' for 'state', which was
161 * initialized by a successful call to the 'port_dump_start' function for
162 * 'dpif'. On success, stores a new dpif_port into 'port' and returns 0.
163 * Returns EOF if the end of the port table has been reached, or a positive
164 * errno value on error. This function will not be called again once it
165 * returns nonzero once for a given iteration (but the 'port_dump_done'
166 * function will be called afterward).
168 * The dpif provider retains ownership of the data stored in 'port'. It
169 * must remain valid until at least the next call to 'port_dump_next' or
170 * 'port_dump_done' for 'state'. */
171 int (*port_dump_next)(const struct dpif *dpif, void *state,
172 struct dpif_port *port);
174 /* Releases resources from 'dpif' for 'state', which was initialized by a
175 * successful call to the 'port_dump_start' function for 'dpif'. */
176 int (*port_dump_done)(const struct dpif *dpif, void *state);
178 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
179 * 'dpif' has changed, then this function should do one of the
182 * - Preferably: store the name of the device that was added to or deleted
183 * from 'dpif' in '*devnamep' and return 0. The caller is responsible
184 * for freeing '*devnamep' (with free()) when it no longer needs it.
186 * - Alternatively: return ENOBUFS, without indicating the device that was
189 * Occasional 'false positives', in which the function returns 0 while
190 * indicating a device that was not actually added or deleted or returns
191 * ENOBUFS without any change, are acceptable.
193 * If the set of ports in 'dpif' has not changed, returns EAGAIN. May also
194 * return other positive errno values to indicate that something has gone
196 int (*port_poll)(const struct dpif *dpif, char **devnamep);
198 /* Arranges for the poll loop to wake up when 'port_poll' will return a
199 * value other than EAGAIN. */
200 void (*port_poll_wait)(const struct dpif *dpif);
202 /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
203 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
206 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
207 * failure, returns a positive errno value.
209 * If 'actionsp' is nonnull, then on success '*actionsp' must be set to an
210 * ofpbuf owned by the caller that contains the Netlink attributes for the
211 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete())
212 * when it is no longer needed.
214 * If 'stats' is nonnull, then on success it must be updated with the
215 * flow's statistics. */
216 int (*flow_get)(const struct dpif *dpif,
217 const struct nlattr *key, size_t key_len,
218 struct ofpbuf **actionsp, struct dpif_flow_stats *stats);
220 /* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
221 * attributes with types OVS_KEY_ATTR_* in the 'put->key_len' bytes
222 * starting at 'put->key'. The associated actions are specified by the
223 * Netlink attributes with types OVS_ACTION_ATTR_* in the
224 * 'put->actions_len' bytes starting at 'put->actions'.
226 * - If the flow's key does not exist in 'dpif', then the flow will be
227 * added if 'put->flags' includes DPIF_FP_CREATE. Otherwise the
228 * operation will fail with ENOENT.
230 * If the operation succeeds, then 'put->stats', if nonnull, must be
233 * - If the flow's key does exist in 'dpif', then the flow's actions will
234 * be updated if 'put->flags' includes DPIF_FP_MODIFY. Otherwise the
235 * operation will fail with EEXIST. If the flow's actions are updated,
236 * then its statistics will be zeroed if 'put->flags' includes
237 * DPIF_FP_ZERO_STATS, and left as-is otherwise.
239 * If the operation succeeds, then 'put->stats', if nonnull, must be set
240 * to the flow's statistics before the update.
242 int (*flow_put)(struct dpif *dpif, const struct dpif_flow_put *put);
244 /* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif'
245 * does not contain such a flow. The flow is specified by the Netlink
246 * attributes with types OVS_KEY_ATTR_* in the 'del->key_len' bytes
247 * starting at 'del->key'.
249 * If the operation succeeds, then 'del->stats', if nonnull, must be set to
250 * the flow's statistics before its deletion. */
251 int (*flow_del)(struct dpif *dpif, const struct dpif_flow_del *del);
253 /* Deletes all flows from 'dpif' and clears all of its queues of received
255 int (*flow_flush)(struct dpif *dpif);
257 /* Attempts to begin dumping the flows in a dpif. On success, returns 0
258 * and initializes '*statep' with any data needed for iteration. On
259 * failure, returns a positive errno value. */
260 int (*flow_dump_start)(const struct dpif *dpif, void **statep);
262 /* Attempts to retrieve another flow from 'dpif' for 'state', which was
263 * initialized by a successful call to the 'flow_dump_start' function for
264 * 'dpif'. On success, updates the output parameters as described below
265 * and returns 0. Returns EOF if the end of the flow table has been
266 * reached, or a positive errno value on error. This function will not be
267 * called again once it returns nonzero within a given iteration (but the
268 * 'flow_dump_done' function will be called afterward).
270 * On success, if 'key' and 'key_len' are nonnull then '*key' and
271 * '*key_len' must be set to Netlink attributes with types OVS_KEY_ATTR_*
272 * representing the dumped flow's key. If 'actions' and 'actions_len' are
273 * nonnull then they should be set to Netlink attributes with types
274 * OVS_ACTION_ATTR_* representing the dumped flow's actions. If 'stats'
275 * is nonnull then it should be set to the dumped flow's statistics.
277 * All of the returned data is owned by 'dpif', not by the caller, and the
278 * caller must not modify or free it. 'dpif' must guarantee that it
279 * remains accessible and unchanging until at least the next call to
280 * 'flow_dump_next' or 'flow_dump_done' for 'state'. */
281 int (*flow_dump_next)(const struct dpif *dpif, void *state,
282 const struct nlattr **key, size_t *key_len,
283 const struct nlattr **actions, size_t *actions_len,
284 const struct dpif_flow_stats **stats);
286 /* Releases resources from 'dpif' for 'state', which was initialized by a
287 * successful call to the 'flow_dump_start' function for 'dpif'. */
288 int (*flow_dump_done)(const struct dpif *dpif, void *state);
290 /* Performs the 'execute->actions_len' bytes of actions in
291 * 'execute->actions' on the Ethernet frame specified in 'execute->packet'
292 * taken from the flow specified in the 'execute->key_len' bytes of
293 * 'execute->key'. ('execute->key' is mostly redundant with
294 * 'execute->packet', but it contains some metadata that cannot be
295 * recovered from 'execute->packet', such as tun_id and in_port.) */
296 int (*execute)(struct dpif *dpif, const struct dpif_execute *execute);
298 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order
299 * in which they are specified, placing each operation's results in the
300 * "output" members documented in comments.
302 * This function is optional. It is only worthwhile to implement it if
303 * 'dpif' can perform operations in batch faster than individually. */
304 void (*operate)(struct dpif *dpif, struct dpif_op **ops, size_t n_ops);
306 /* Enables or disables receiving packets with dpif_recv() for 'dpif'.
307 * Turning packet receive off and then back on is allowed to change Netlink
308 * PID assignments (see ->port_get_pid()). The client is responsible for
309 * updating flows as necessary if it does this. */
310 int (*recv_set)(struct dpif *dpif, bool enable);
312 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
313 * priority value used for setting packet priority. */
314 int (*queue_to_priority)(const struct dpif *dpif, uint32_t queue_id,
317 /* Polls for an upcall from 'dpif'. If successful, stores the upcall into
318 * '*upcall', using 'buf' for storage. Should only be called if 'recv_set'
319 * has been used to enable receiving packets from 'dpif'.
321 * The implementation should point 'upcall->packet' and 'upcall->key' into
322 * data in the caller-provided 'buf'. If necessary to make room, the
323 * implementation may expand the data in 'buf'. (This is hardly a great
324 * way to do things but it works out OK for the dpif providers that exist
327 * This function must not block. If no upcall is pending when it is
328 * called, it should return EAGAIN without blocking. */
329 int (*recv)(struct dpif *dpif, struct dpif_upcall *upcall,
332 /* Arranges for the poll loop to wake up when 'dpif' has a message queued
333 * to be received with the recv member function. */
334 void (*recv_wait)(struct dpif *dpif);
336 /* Throws away any queued upcalls that 'dpif' currently has ready to
338 void (*recv_purge)(struct dpif *dpif);
341 extern const struct dpif_class dpif_linux_class;
342 extern const struct dpif_class dpif_netdev_class;
343 extern const struct dpif_class dpif_planetlab_class;
349 #endif /* dpif-provider.h */