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 dpif_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 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
146 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
147 * flows whose packets arrived on port 'port_no'.
149 * The return value only needs to be meaningful when DPIF_UC_ACTION has
150 * been enabled in the 'dpif''s listen mask, and it is allowed to change
151 * when DPIF_UC_ACTION is disabled and then re-enabled.
153 * A dpif provider that doesn't have meaningful Netlink PIDs can use NULL
154 * for this function. This is equivalent to always returning 0. */
155 uint32_t (*port_get_pid)(const struct dpif *dpif, uint16_t port_no);
157 /* Attempts to begin dumping the ports in a dpif. On success, returns 0
158 * and initializes '*statep' with any data needed for iteration. On
159 * failure, returns a positive errno value. */
160 int (*port_dump_start)(const struct dpif *dpif, void **statep);
162 /* Attempts to retrieve another port from 'dpif' for 'state', which was
163 * initialized by a successful call to the 'port_dump_start' function for
164 * 'dpif'. On success, stores a new dpif_port into 'port' and returns 0.
165 * Returns EOF if the end of the port table has been reached, or a positive
166 * errno value on error. This function will not be called again once it
167 * returns nonzero once for a given iteration (but the 'port_dump_done'
168 * function will be called afterward).
170 * The dpif provider retains ownership of the data stored in 'port'. It
171 * must remain valid until at least the next call to 'port_dump_next' or
172 * 'port_dump_done' for 'state'. */
173 int (*port_dump_next)(const struct dpif *dpif, void *state,
174 struct dpif_port *port);
176 /* Releases resources from 'dpif' for 'state', which was initialized by a
177 * successful call to the 'port_dump_start' function for 'dpif'. */
178 int (*port_dump_done)(const struct dpif *dpif, void *state);
180 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
181 * 'dpif' has changed, then this function should do one of the
184 * - Preferably: store the name of the device that was added to or deleted
185 * from 'dpif' in '*devnamep' and return 0. The caller is responsible
186 * for freeing '*devnamep' (with free()) when it no longer needs it.
188 * - Alternatively: return ENOBUFS, without indicating the device that was
191 * Occasional 'false positives', in which the function returns 0 while
192 * indicating a device that was not actually added or deleted or returns
193 * ENOBUFS without any change, are acceptable.
195 * If the set of ports in 'dpif' has not changed, returns EAGAIN. May also
196 * return other positive errno values to indicate that something has gone
198 int (*port_poll)(const struct dpif *dpif, char **devnamep);
200 /* Arranges for the poll loop to wake up when 'port_poll' will return a
201 * value other than EAGAIN. */
202 void (*port_poll_wait)(const struct dpif *dpif);
204 /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
205 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
208 * Returns 0 if successful. If no flow matches, returns ENOENT. On other
209 * failure, returns a positive errno value.
211 * If 'actionsp' is nonnull, then on success '*actionsp' must be set to an
212 * ofpbuf owned by the caller that contains the Netlink attributes for the
213 * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete())
214 * when it is no longer needed.
216 * If 'stats' is nonnull, then on success it must be updated with the
217 * flow's statistics. */
218 int (*flow_get)(const struct dpif *dpif,
219 const struct nlattr *key, size_t key_len,
220 struct ofpbuf **actionsp, struct dpif_flow_stats *stats);
222 /* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
223 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
224 * 'key'. The associated actions are specified by the Netlink attributes
225 * with types OVS_ACTION_ATTR_* in the 'actions_len' bytes starting at
228 * - If the flow's key does not exist in 'dpif', then the flow will be
229 * added if 'flags' includes DPIF_FP_CREATE. Otherwise the operation
230 * will fail with ENOENT.
232 * If the operation succeeds, then 'stats', if nonnull, must be zeroed.
234 * - If the flow's key does exist in 'dpif', then the flow's actions will
235 * be updated if 'flags' includes DPIF_FP_MODIFY. Otherwise the
236 * operation will fail with EEXIST. If the flow's actions are updated,
237 * then its statistics will be zeroed if 'flags' includes
238 * DPIF_FP_ZERO_STATS, and left as-is otherwise.
240 * If the operation succeeds, then 'stats', if nonnull, must be set to
241 * the flow's statistics before the update.
243 int (*flow_put)(struct dpif *dpif, enum dpif_flow_put_flags flags,
244 const struct nlattr *key, size_t key_len,
245 const struct nlattr *actions, size_t actions_len,
246 struct dpif_flow_stats *stats);
248 /* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif'
249 * does not contain such a flow. The flow is specified by the Netlink
250 * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
253 * If the operation succeeds, then 'stats', if nonnull, must be set to the
254 * flow's statistics before its deletion. */
255 int (*flow_del)(struct dpif *dpif,
256 const struct nlattr *key, size_t key_len,
257 struct dpif_flow_stats *stats);
259 /* Deletes all flows from 'dpif' and clears all of its queues of received
261 int (*flow_flush)(struct dpif *dpif);
263 /* Attempts to begin dumping the flows in a dpif. On success, returns 0
264 * and initializes '*statep' with any data needed for iteration. On
265 * failure, returns a positive errno value. */
266 int (*flow_dump_start)(const struct dpif *dpif, void **statep);
268 /* Attempts to retrieve another flow from 'dpif' for 'state', which was
269 * initialized by a successful call to the 'flow_dump_start' function for
270 * 'dpif'. On success, updates the output parameters as described below
271 * and returns 0. Returns EOF if the end of the flow table has been
272 * reached, or a positive errno value on error. This function will not be
273 * called again once it returns nonzero within a given iteration (but the
274 * 'flow_dump_done' function will be called afterward).
276 * On success, if 'key' and 'key_len' are nonnull then '*key' and
277 * '*key_len' must be set to Netlink attributes with types OVS_KEY_ATTR_*
278 * representing the dumped flow's key. If 'actions' and 'actions_len' are
279 * nonnull then they should be set to Netlink attributes with types
280 * OVS_ACTION_ATTR_* representing the dumped flow's actions. If 'stats'
281 * is nonnull then it should be set to the dumped flow's statistics.
283 * All of the returned data is owned by 'dpif', not by the caller, and the
284 * caller must not modify or free it. 'dpif' must guarantee that it
285 * remains accessible and unchanging until at least the next call to
286 * 'flow_dump_next' or 'flow_dump_done' for 'state'. */
287 int (*flow_dump_next)(const struct dpif *dpif, void *state,
288 const struct nlattr **key, size_t *key_len,
289 const struct nlattr **actions, size_t *actions_len,
290 const struct dpif_flow_stats **stats);
292 /* Releases resources from 'dpif' for 'state', which was initialized by a
293 * successful call to the 'flow_dump_start' function for 'dpif'. */
294 int (*flow_dump_done)(const struct dpif *dpif, void *state);
296 /* Performs the 'actions_len' bytes of actions in 'actions' on the Ethernet
297 * frame specified in 'packet' taken from the flow specified in the
298 * 'key_len' bytes of 'key'. ('key' is mostly redundant with 'packet', but
299 * it contains some metadata that cannot be recovered from 'packet', such
300 * as tun_id and in_port.) */
301 int (*execute)(struct dpif *dpif,
302 const struct nlattr *key, size_t key_len,
303 const struct nlattr *actions, size_t actions_len,
304 const struct ofpbuf *packet);
306 /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. A 1-bit of value
307 * 2**X set in '*listen_mask' indicates that 'dpif' will receive messages
308 * of the type (from "enum dpif_upcall_type") with value X when its 'recv'
309 * function is called. */
310 int (*recv_get_mask)(const struct dpif *dpif, int *listen_mask);
312 /* Sets 'dpif''s "listen mask" to 'listen_mask'. A 1-bit of value 2**X set
313 * in '*listen_mask' requests that 'dpif' will receive messages of the type
314 * (from "enum dpif_upcall_type") with value X when its 'recv' function is
317 * Turning DPIF_UC_ACTION off and then back on is allowed to change Netlink
318 * PID assignments (see ->port_get_pid()). The client is responsible for
319 * updating flows as necessary if it does this. */
320 int (*recv_set_mask)(struct dpif *dpif, int listen_mask);
322 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
323 * priority value for use in the OVS_ACTION_ATTR_SET_PRIORITY action in
325 int (*queue_to_priority)(const struct dpif *dpif, uint32_t queue_id,
328 /* Polls for an upcall from 'dpif'. If successful, stores the upcall into
329 * '*upcall'. Only upcalls of the types selected with the set_listen_mask
330 * member function should be received.
332 * The caller takes ownership of the data that 'upcall' points to.
333 * 'upcall->key' and 'upcall->actions' (if nonnull) point into data owned
334 * by 'upcall->packet', so their memory cannot be freed separately. (This
335 * is hardly a great way to do things but it works out OK for the dpif
336 * providers that exist so far.)
338 * For greatest efficiency, 'upcall->packet' should have at least
339 * offsetof(struct ofp_packet_in, data) bytes of headroom.
341 * This function must not block. If no upcall is pending when it is
342 * called, it should return EAGAIN without blocking. */
343 int (*recv)(struct dpif *dpif, struct dpif_upcall *upcall);
345 /* Arranges for the poll loop to wake up when 'dpif' has a message queued
346 * to be received with the recv member function. */
347 void (*recv_wait)(struct dpif *dpif);
349 /* Throws away any queued upcalls that 'dpif' currently has ready to
351 void (*recv_purge)(struct dpif *dpif);
354 extern const struct dpif_class dpif_linux_class;
355 extern const struct dpif_class dpif_netdev_class;
361 #endif /* dpif-provider.h */