2 * Copyright (c) 2009, 2010, 2011 Nicira Networks.
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 ovs_dp_stats *stats);
113 /* Retrieves 'dpif''s current treatment of IP fragments into '*drop_frags':
114 * true indicates that fragments are dropped, false indicates that
115 * fragments are treated in the same way as other IP packets (except that
116 * the L4 header cannot be read). */
117 int (*get_drop_frags)(const struct dpif *dpif, bool *drop_frags);
119 /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose
120 * meaning is the same as for the get_drop_frags member function. */
121 int (*set_drop_frags)(struct dpif *dpif, bool drop_frags);
123 /* Adds 'netdev' as a new port in 'dpif'. If successful, sets '*port_no'
124 * to the new port's port number. */
125 int (*port_add)(struct dpif *dpif, struct netdev *netdev,
128 /* Removes port numbered 'port_no' from 'dpif'. */
129 int (*port_del)(struct dpif *dpif, uint16_t port_no);
131 /* Queries 'dpif' for a port with the given 'port_no' or 'devname'. Stores
132 * information about the port into '*port' if successful.
134 * The caller takes ownership of data in 'port' and must free it with
135 * dpif_port_destroy() when it is no longer needed. */
136 int (*port_query_by_number)(const struct dpif *dpif, uint16_t port_no,
137 struct dpif_port *port);
138 int (*port_query_by_name)(const struct dpif *dpif, const char *devname,
139 struct dpif_port *port);
141 /* Returns one greater than the largest port number accepted in flow
143 int (*get_max_ports)(const struct dpif *dpif);
145 /* Attempts to begin dumping the ports in a dpif. On success, returns 0
146 * and initializes '*statep' with any data needed for iteration. On
147 * failure, returns a positive errno value. */
148 int (*port_dump_start)(const struct dpif *dpif, void **statep);
150 /* Attempts to retrieve another port from 'dpif' for 'state', which was
151 * initialized by a successful call to the 'port_dump_start' function for
152 * 'dpif'. On success, stores a new dpif_port into 'port' and returns 0.
153 * Returns EOF if the end of the port table has been reached, or a positive
154 * errno value on error. This function will not be called again once it
155 * returns nonzero once for a given iteration (but the 'port_dump_done'
156 * function will be called afterward).
158 * The dpif provider retains ownership of the data stored in 'port'. It
159 * must remain valid until at least the next call to 'port_dump_next' or
160 * 'port_dump_done' for 'state'. */
161 int (*port_dump_next)(const struct dpif *dpif, void *state,
162 struct dpif_port *port);
164 /* Releases resources from 'dpif' for 'state', which was initialized by a
165 * successful call to the 'port_dump_start' function for 'dpif'. */
166 int (*port_dump_done)(const struct dpif *dpif, void *state);
168 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
169 * 'dpif' has changed, then this function should do one of the
172 * - Preferably: store the name of the device that was added to or deleted
173 * from 'dpif' in '*devnamep' and return 0. The caller is responsible
174 * for freeing '*devnamep' (with free()) when it no longer needs it.
176 * - Alternatively: return ENOBUFS, without indicating the device that was
179 * Occasional 'false positives', in which the function returns 0 while
180 * indicating a device that was not actually added or deleted or returns
181 * ENOBUFS without any change, are acceptable.
183 * If the set of ports in 'dpif' has not changed, returns EAGAIN. May also
184 * return other positive errno values to indicate that something has gone
186 int (*port_poll)(const struct dpif *dpif, char **devnamep);
188 /* Arranges for the poll loop to wake up when 'port_poll' will return a
189 * value other than EAGAIN. */
190 void (*port_poll_wait)(const struct dpif *dpif);
192 /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
193 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
196 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
197 * failure, returns a positive errno value.
199 * If 'actionsp' is nonnull, then on success '*actionsp' must be set to an
200 * ofpbuf owned by the caller that contains the Netlink attributes for the
201 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete())
202 * when it is no longer needed.
204 * If 'stats' is nonnull, then on success it must be updated with the
205 * flow's statistics. */
206 int (*flow_get)(const struct dpif *dpif,
207 const struct nlattr *key, size_t key_len,
208 struct ofpbuf **actionsp, struct dpif_flow_stats *stats);
210 /* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
211 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
212 * 'key'. The associated actions are specified by the Netlink attributes
213 * with types OVS_ACTION_ATTR_* in the 'actions_len' bytes starting at
216 * - If the flow's key does not exist in 'dpif', then the flow will be
217 * added if 'flags' includes DPIF_FP_CREATE. Otherwise the operation
218 * will fail with ENOENT.
220 * If the operation succeeds, then 'stats', if nonnull, must be zeroed.
222 * - If the flow's key does exist in 'dpif', then the flow's actions will
223 * be updated if 'flags' includes DPIF_FP_MODIFY. Otherwise the
224 * operation will fail with EEXIST. If the flow's actions are updated,
225 * then its statistics will be zeroed if 'flags' includes
226 * DPIF_FP_ZERO_STATS, and left as-is otherwise.
228 * If the operation succeeds, then 'stats', if nonnull, must be set to
229 * the flow's statistics before the update.
231 int (*flow_put)(struct dpif *dpif, enum dpif_flow_put_flags flags,
232 const struct nlattr *key, size_t key_len,
233 const struct nlattr *actions, size_t actions_len,
234 struct dpif_flow_stats *stats);
236 /* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif'
237 * does not contain such a flow. The flow is specified by the Netlink
238 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
241 * If the operation succeeds, then 'stats', if nonnull, must be set to the
242 * flow's statistics before its deletion. */
243 int (*flow_del)(struct dpif *dpif,
244 const struct nlattr *key, size_t key_len,
245 struct dpif_flow_stats *stats);
247 /* Deletes all flows from 'dpif' and clears all of its queues of received
249 int (*flow_flush)(struct dpif *dpif);
251 /* Attempts to begin dumping the flows in a dpif. On success, returns 0
252 * and initializes '*statep' with any data needed for iteration. On
253 * failure, returns a positive errno value. */
254 int (*flow_dump_start)(const struct dpif *dpif, void **statep);
256 /* Attempts to retrieve another flow from 'dpif' for 'state', which was
257 * initialized by a successful call to the 'flow_dump_start' function for
258 * 'dpif'. On success, updates the output parameters as described below
259 * and returns 0. Returns EOF if the end of the flow table has been
260 * reached, or a positive errno value on error. This function will not be
261 * called again once it returns nonzero within a given iteration (but the
262 * 'flow_dump_done' function will be called afterward).
264 * On success, if 'key' and 'key_len' are nonnull then '*key' and
265 * '*key_len' must be set to Netlink attributes with types OVS_KEY_ATTR_*
266 * representing the dumped flow's key. If 'actions' and 'actions_len' are
267 * nonnull then they should be set to Netlink attributes with types
268 * OVS_ACTION_ATTR_* representing the dumped flow's actions. If 'stats'
269 * is nonnull then it should be set to the dumped flow's statistics.
271 * All of the returned data is owned by 'dpif', not by the caller, and the
272 * caller must not modify or free it. 'dpif' must guarantee that it
273 * remains accessible and unchanging until at least the next call to
274 * 'flow_dump_next' or 'flow_dump_done' for 'state'. */
275 int (*flow_dump_next)(const struct dpif *dpif, void *state,
276 const struct nlattr **key, size_t *key_len,
277 const struct nlattr **actions, size_t *actions_len,
278 const struct dpif_flow_stats **stats);
280 /* Releases resources from 'dpif' for 'state', which was initialized by a
281 * successful call to the 'flow_dump_start' function for 'dpif'. */
282 int (*flow_dump_done)(const struct dpif *dpif, void *state);
284 /* Performs the 'actions_len' bytes of actions in 'actions' on the Ethernet
285 * frame specified in 'packet' taken from the flow specified in the
286 * 'key_len' bytes of 'key'. ('key' is mostly redundant with 'packet', but
287 * it contains some metadata that cannot be recovered from 'packet', such
288 * as tun_id and in_port.) */
289 int (*execute)(struct dpif *dpif,
290 const struct nlattr *key, size_t key_len,
291 const struct nlattr *actions, size_t actions_len,
292 const struct ofpbuf *packet);
294 /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. A 1-bit of value
295 * 2**X set in '*listen_mask' indicates that 'dpif' will receive messages
296 * of the type (from "enum dpif_upcall_type") with value X when its 'recv'
297 * function is called. */
298 int (*recv_get_mask)(const struct dpif *dpif, int *listen_mask);
300 /* Sets 'dpif''s "listen mask" to 'listen_mask'. A 1-bit of value 2**X set
301 * in '*listen_mask' requests that 'dpif' will receive messages of the type
302 * (from "enum dpif_upcall_type") with value X when its 'recv' function is
304 int (*recv_set_mask)(struct dpif *dpif, int listen_mask);
306 /* Retrieves 'dpif''s sFlow sampling probability into '*probability'.
307 * Return value is 0 or a positive errno value. EOPNOTSUPP indicates that
308 * the datapath does not support sFlow, as does a null pointer.
310 * '*probability' is expressed as the number of packets out of UINT_MAX to
311 * sample, e.g. probability/UINT_MAX is the probability of sampling a given
313 int (*get_sflow_probability)(const struct dpif *dpif,
314 uint32_t *probability);
316 /* Sets 'dpif''s sFlow sampling probability to 'probability'. Return value
317 * is 0 or a positive errno value. EOPNOTSUPP indicates that the datapath
318 * does not support sFlow, as does a null pointer.
320 * 'probability' is expressed as the number of packets out of UINT_MAX to
321 * sample, e.g. probability/UINT_MAX is the probability of sampling a given
323 int (*set_sflow_probability)(struct dpif *dpif, uint32_t probability);
325 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
326 * priority value for use in the OVS_ACTION_ATTR_SET_PRIORITY action in
328 int (*queue_to_priority)(const struct dpif *dpif, uint32_t queue_id,
331 /* Polls for an upcall from 'dpif'. If successful, stores the upcall into
332 * '*upcall'. Only upcalls of the types selected with the set_listen_mask
333 * member function should be received.
335 * The caller takes ownership of the data that 'upcall' points to.
336 * 'upcall->key' and 'upcall->actions' (if nonnull) point into data owned
337 * by 'upcall->packet', so their memory cannot be freed separately. (This
338 * is hardly a great way to do things but it works out OK for the dpif
339 * providers that exist so far.)
341 * For greatest efficiency, 'upcall->packet' should have at least
342 * offsetof(struct ofp_packet_in, data) bytes of headroom.
344 * This function must not block. If no upcall is pending when it is
345 * called, it should return EAGAIN without blocking. */
346 int (*recv)(struct dpif *dpif, struct dpif_upcall *upcall);
348 /* Arranges for the poll loop to wake up when 'dpif' has a message queued
349 * to be received with the recv member function. */
350 void (*recv_wait)(struct dpif *dpif);
352 /* Throws away any queued upcalls that 'dpif' currently has ready to
354 void (*recv_purge)(struct dpif *dpif);
357 extern const struct dpif_class dpif_linux_class;
358 extern const struct dpif_class dpif_netdev_class;
364 #endif /* dpif-provider.h */