2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 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 struct dpif_flow_dump {
56 bool terse; /* If true, key/mask/actions may be omitted. */
60 dpif_flow_dump_init(struct dpif_flow_dump *dump, const struct dpif *dpif)
62 dump->dpif = CONST_CAST(struct dpif *, dpif);
65 struct dpif_flow_dump_thread {
70 dpif_flow_dump_thread_init(struct dpif_flow_dump_thread *thread,
71 struct dpif_flow_dump *dump)
73 thread->dpif = dump->dpif;
76 /* Datapath interface class structure, to be defined by each implementation of
77 * a datapath interface.
79 * These functions return 0 if successful or a positive errno value on failure,
80 * except where otherwise noted.
82 * These functions are expected to execute synchronously, that is, to block as
83 * necessary to obtain a result. Thus, they may not return EAGAIN or
84 * EWOULDBLOCK or EINPROGRESS. We may relax this requirement in the future if
85 * and when we encounter performance problems. */
87 /* Type of dpif in this class, e.g. "system", "netdev", etc.
89 * One of the providers should supply a "system" type, since this is
90 * the type assumed if no type is specified when opening a dpif. */
93 /* Enumerates the names of all known created datapaths (of class
94 * 'dpif_class'), if possible, into 'all_dps'. The caller has already
95 * initialized 'all_dps' and other dpif classes might already have added
98 * This is used by the vswitch at startup, so that it can delete any
99 * datapaths that are not configured.
101 * Some kinds of datapaths might not be practically enumerable, in which
102 * case this function may be a null pointer. */
103 int (*enumerate)(struct sset *all_dps, const struct dpif_class *dpif_class);
105 /* Returns the type to pass to netdev_open() when a dpif of class
106 * 'dpif_class' has a port of type 'type', for a few special cases
107 * when a netdev type differs from a port type. For example, when
108 * using the userspace datapath, a port of type "internal" needs to
109 * be opened as "tap".
111 * Returns either 'type' itself or a string literal, which must not
113 const char *(*port_open_type)(const struct dpif_class *dpif_class,
116 /* Attempts to open an existing dpif called 'name', if 'create' is false,
117 * or to open an existing dpif or create a new one, if 'create' is true.
119 * 'dpif_class' is the class of dpif to open.
121 * If successful, stores a pointer to the new dpif in '*dpifp', which must
122 * have class 'dpif_class'. On failure there are no requirements on what
123 * is stored in '*dpifp'. */
124 int (*open)(const struct dpif_class *dpif_class,
125 const char *name, bool create, struct dpif **dpifp);
127 /* Closes 'dpif' and frees associated memory. */
128 void (*close)(struct dpif *dpif);
130 /* Attempts to destroy the dpif underlying 'dpif'.
132 * If successful, 'dpif' will not be used again except as an argument for
133 * the 'close' member function. */
134 int (*destroy)(struct dpif *dpif);
136 /* Performs periodic work needed by 'dpif', if any is necessary.
137 * Returns true if need to revalidate. */
138 bool (*run)(struct dpif *dpif);
140 /* Arranges for poll_block() to wake up if the "run" member function needs
141 * to be called for 'dpif'. */
142 void (*wait)(struct dpif *dpif);
144 /* Retrieves statistics for 'dpif' into 'stats'. */
145 int (*get_stats)(const struct dpif *dpif, struct dpif_dp_stats *stats);
147 /* Adds 'netdev' as a new port in 'dpif'. If '*port_no' is not
148 * UINT32_MAX, attempts to use that as the port's port number.
150 * If port is successfully added, sets '*port_no' to the new port's
151 * port number. Returns EBUSY if caller attempted to choose a port
152 * number, and it was in use. */
153 int (*port_add)(struct dpif *dpif, struct netdev *netdev,
154 odp_port_t *port_no);
156 /* Removes port numbered 'port_no' from 'dpif'. */
157 int (*port_del)(struct dpif *dpif, odp_port_t port_no);
159 /* Queries 'dpif' for a port with the given 'port_no' or 'devname'.
160 * If 'port' is not null, stores information about the port into
161 * '*port' if successful.
163 * If 'port' is not null, the caller takes ownership of data in
164 * 'port' and must free it with dpif_port_destroy() when it is no
166 int (*port_query_by_number)(const struct dpif *dpif, odp_port_t port_no,
167 struct dpif_port *port);
168 int (*port_query_by_name)(const struct dpif *dpif, const char *devname,
169 struct dpif_port *port);
171 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
172 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
173 * flows whose packets arrived on port 'port_no'. In the case where the
174 * provider allocates multiple Netlink PIDs to a single port, it may use
175 * 'hash' to spread load among them. The caller need not use a particular
176 * hash function; a 5-tuple hash is suitable.
178 * (The datapath implementation might use some different hash function for
179 * distributing packets received via flow misses among PIDs. This means
180 * that packets received via flow misses might be reordered relative to
181 * packets received via userspace actions. This is not ordinarily a
184 * A 'port_no' of UINT32_MAX should be treated as a special case. The
185 * implementation should return a reserved PID, not allocated to any port,
186 * that the client may use for special purposes.
188 * The return value only needs to be meaningful when DPIF_UC_ACTION has
189 * been enabled in the 'dpif''s listen mask, and it is allowed to change
190 * when DPIF_UC_ACTION is disabled and then re-enabled.
192 * A dpif provider that doesn't have meaningful Netlink PIDs can use NULL
193 * for this function. This is equivalent to always returning 0. */
194 uint32_t (*port_get_pid)(const struct dpif *dpif, odp_port_t port_no,
197 /* Attempts to begin dumping the ports in a dpif. On success, returns 0
198 * and initializes '*statep' with any data needed for iteration. On
199 * failure, returns a positive errno value. */
200 int (*port_dump_start)(const struct dpif *dpif, void **statep);
202 /* Attempts to retrieve another port from 'dpif' for 'state', which was
203 * initialized by a successful call to the 'port_dump_start' function for
204 * 'dpif'. On success, stores a new dpif_port into 'port' and returns 0.
205 * Returns EOF if the end of the port table has been reached, or a positive
206 * errno value on error. This function will not be called again once it
207 * returns nonzero once for a given iteration (but the 'port_dump_done'
208 * function will be called afterward).
210 * The dpif provider retains ownership of the data stored in 'port'. It
211 * must remain valid until at least the next call to 'port_dump_next' or
212 * 'port_dump_done' for 'state'. */
213 int (*port_dump_next)(const struct dpif *dpif, void *state,
214 struct dpif_port *port);
216 /* Releases resources from 'dpif' for 'state', which was initialized by a
217 * successful call to the 'port_dump_start' function for 'dpif'. */
218 int (*port_dump_done)(const struct dpif *dpif, void *state);
220 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
221 * 'dpif' has changed, then this function should do one of the
224 * - Preferably: store the name of the device that was added to or deleted
225 * from 'dpif' in '*devnamep' and return 0. The caller is responsible
226 * for freeing '*devnamep' (with free()) when it no longer needs it.
228 * - Alternatively: return ENOBUFS, without indicating the device that was
231 * Occasional 'false positives', in which the function returns 0 while
232 * indicating a device that was not actually added or deleted or returns
233 * ENOBUFS without any change, are acceptable.
235 * If the set of ports in 'dpif' has not changed, returns EAGAIN. May also
236 * return other positive errno values to indicate that something has gone
238 int (*port_poll)(const struct dpif *dpif, char **devnamep);
240 /* Arranges for the poll loop to wake up when 'port_poll' will return a
241 * value other than EAGAIN. */
242 void (*port_poll_wait)(const struct dpif *dpif);
244 /* Deletes all flows from 'dpif' and clears all of its queues of received
246 int (*flow_flush)(struct dpif *dpif);
248 /* Flow dumping interface.
250 * This is the back-end for the flow dumping interface described in
251 * dpif.h. Please read the comments there first, because this code
252 * closely follows it.
254 * 'flow_dump_create' and 'flow_dump_thread_create' must always return an
255 * initialized and usable data structure and defer error return until
256 * flow_dump_destroy(). This hasn't been a problem for the dpifs that
259 * 'flow_dump_create' and 'flow_dump_thread_create' must initialize the
260 * structures that they return with dpif_flow_dump_init() and
261 * dpif_flow_dump_thread_init(), respectively.
263 * If 'terse' is true, then only UID and statistics will
264 * be returned in the dump. Otherwise, all fields will be returned. */
265 struct dpif_flow_dump *(*flow_dump_create)(const struct dpif *dpif,
267 int (*flow_dump_destroy)(struct dpif_flow_dump *dump);
269 struct dpif_flow_dump_thread *(*flow_dump_thread_create)(
270 struct dpif_flow_dump *dump);
271 void (*flow_dump_thread_destroy)(struct dpif_flow_dump_thread *thread);
273 int (*flow_dump_next)(struct dpif_flow_dump_thread *thread,
274 struct dpif_flow *flows, int max_flows);
276 /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order
277 * in which they are specified, placing each operation's results in the
278 * "output" members documented in comments and the 'error' member of each
280 void (*operate)(struct dpif *dpif, struct dpif_op **ops, size_t n_ops);
282 /* Enables or disables receiving packets with dpif_recv() for 'dpif'.
283 * Turning packet receive off and then back on is allowed to change Netlink
284 * PID assignments (see ->port_get_pid()). The client is responsible for
285 * updating flows as necessary if it does this. */
286 int (*recv_set)(struct dpif *dpif, bool enable);
288 /* Refreshes the poll loops and Netlink sockets associated to each port,
289 * when the number of upcall handlers (upcall receiving thread) is changed
290 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
293 * Since multiple upcall handlers can read upcalls simultaneously from
294 * 'dpif', each port can have multiple Netlink sockets, one per upcall
295 * handler. So, handlers_set() is responsible for the following tasks:
297 * When receiving upcall is enabled, extends or creates the
298 * configuration to support:
300 * - 'n_handlers' Netlink sockets for each port.
302 * - 'n_handlers' poll loops, one for each upcall handler.
304 * - registering the Netlink sockets for the same upcall handler to
305 * the corresponding poll loop.
307 int (*handlers_set)(struct dpif *dpif, uint32_t n_handlers);
309 /* If 'dpif' creates its own I/O polling threads, refreshes poll threads
310 * configuration. 'n_rxqs' configures the number of rx_queues, which
311 * are distributed among threads. 'cmask' configures the cpu mask
312 * for setting the polling threads' cpu affinity. */
313 int (*poll_threads_set)(struct dpif *dpif, unsigned int n_rxqs,
316 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
317 * priority value used for setting packet priority. */
318 int (*queue_to_priority)(const struct dpif *dpif, uint32_t queue_id,
321 /* Polls for an upcall from 'dpif' for an upcall handler. Since there
322 * can be multiple poll loops (see ->handlers_set()), 'handler_id' is
323 * needed as index to identify the corresponding poll loop. If
324 * successful, stores the upcall into '*upcall', using 'buf' for
325 * storage. Should only be called if 'recv_set' has been used to enable
326 * receiving packets from 'dpif'.
328 * The implementation should point 'upcall->key' and 'upcall->userdata'
329 * (if any) into data in the caller-provided 'buf'. The implementation may
330 * also use 'buf' for storing the data of 'upcall->packet'. If necessary
331 * to make room, the implementation may reallocate the data in 'buf'.
333 * The caller owns the data of 'upcall->packet' and may modify it. If
334 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
335 * will be reallocated. This requires the data of 'upcall->packet' to be
336 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
337 * when an error is returned, the 'upcall->packet' may be uninitialized
338 * and should not be released.
340 * This function must not block. If no upcall is pending when it is
341 * called, it should return EAGAIN without blocking. */
342 int (*recv)(struct dpif *dpif, uint32_t handler_id,
343 struct dpif_upcall *upcall, struct ofpbuf *buf);
345 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
346 * has a message queued to be received with the recv member functions.
347 * Since there can be multiple poll loops (see ->handlers_set()),
348 * 'handler_id' is needed as index to identify the corresponding poll loop.
350 void (*recv_wait)(struct dpif *dpif, uint32_t handler_id);
352 /* Throws away any queued upcalls that 'dpif' currently has ready to
354 void (*recv_purge)(struct dpif *dpif);
356 /* For datapaths that run in userspace (i.e. dpif-netdev), threads polling
357 * for incoming packets can directly call upcall functions instead of
358 * offloading packet processing to separate handler threads. Datapaths
359 * that directly call upcall functions should use the functions below to
360 * to register an upcall function and enable / disable upcalls.
362 * Registers an upcall callback function with 'dpif'. This is only used if
363 * if 'dpif' directly executes upcall functions. 'aux' is passed to the
364 * callback on invocation. */
365 void (*register_upcall_cb)(struct dpif *, upcall_callback *, void *aux);
367 /* Enables upcalls if 'dpif' directly executes upcall functions. */
368 void (*enable_upcall)(struct dpif *);
370 /* Disables upcalls if 'dpif' directly executes upcall functions. */
371 void (*disable_upcall)(struct dpif *);
373 /* Get datapath version. Caller is responsible for freeing the string
375 char *(*get_datapath_version)(void);
378 extern const struct dpif_class dpif_netlink_class;
379 extern const struct dpif_class dpif_netdev_class;
385 #endif /* dpif-provider.h */