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 struct ct_dpif_dump_state
;
79 /* Datapath interface class structure, to be defined by each implementation of
80 * a datapath interface.
82 * These functions return 0 if successful or a positive errno value on failure,
83 * except where otherwise noted.
85 * These functions are expected to execute synchronously, that is, to block as
86 * necessary to obtain a result. Thus, they may not return EAGAIN or
87 * EWOULDBLOCK or EINPROGRESS. We may relax this requirement in the future if
88 * and when we encounter performance problems. */
90 /* Type of dpif in this class, e.g. "system", "netdev", etc.
92 * One of the providers should supply a "system" type, since this is
93 * the type assumed if no type is specified when opening a dpif. */
96 /* Called when the dpif provider is registered, typically at program
97 * startup. Returning an error from this function will prevent any
98 * datapath with this class from being created.
100 * This function may be set to null if a datapath class needs no
101 * initialization at registration time. */
104 /* Enumerates the names of all known created datapaths (of class
105 * 'dpif_class'), if possible, into 'all_dps'. The caller has already
106 * initialized 'all_dps' and other dpif classes might already have added
109 * This is used by the vswitch at startup, so that it can delete any
110 * datapaths that are not configured.
112 * Some kinds of datapaths might not be practically enumerable, in which
113 * case this function may be a null pointer. */
114 int (*enumerate
)(struct sset
*all_dps
, const struct dpif_class
*dpif_class
);
116 /* Returns the type to pass to netdev_open() when a dpif of class
117 * 'dpif_class' has a port of type 'type', for a few special cases
118 * when a netdev type differs from a port type. For example, when
119 * using the userspace datapath, a port of type "internal" needs to
120 * be opened as "tap".
122 * Returns either 'type' itself or a string literal, which must not
124 const char *(*port_open_type
)(const struct dpif_class
*dpif_class
,
127 /* Attempts to open an existing dpif called 'name', if 'create' is false,
128 * or to open an existing dpif or create a new one, if 'create' is true.
130 * 'dpif_class' is the class of dpif to open.
132 * If successful, stores a pointer to the new dpif in '*dpifp', which must
133 * have class 'dpif_class'. On failure there are no requirements on what
134 * is stored in '*dpifp'. */
135 int (*open
)(const struct dpif_class
*dpif_class
,
136 const char *name
, bool create
, struct dpif
**dpifp
);
138 /* Closes 'dpif' and frees associated memory. */
139 void (*close
)(struct dpif
*dpif
);
141 /* Attempts to destroy the dpif underlying 'dpif'.
143 * If successful, 'dpif' will not be used again except as an argument for
144 * the 'close' member function. */
145 int (*destroy
)(struct dpif
*dpif
);
147 /* Performs periodic work needed by 'dpif', if any is necessary.
148 * Returns true if need to revalidate. */
149 bool (*run
)(struct dpif
*dpif
);
151 /* Arranges for poll_block() to wake up if the "run" member function needs
152 * to be called for 'dpif'. */
153 void (*wait
)(struct dpif
*dpif
);
155 /* Retrieves statistics for 'dpif' into 'stats'. */
156 int (*get_stats
)(const struct dpif
*dpif
, struct dpif_dp_stats
*stats
);
158 /* Adds 'netdev' as a new port in 'dpif'. If '*port_no' is not
159 * ODPP_NONE, attempts to use that as the port's port number.
161 * If port is successfully added, sets '*port_no' to the new port's
162 * port number. Returns EBUSY if caller attempted to choose a port
163 * number, and it was in use. */
164 int (*port_add
)(struct dpif
*dpif
, struct netdev
*netdev
,
165 odp_port_t
*port_no
);
167 /* Removes port numbered 'port_no' from 'dpif'. */
168 int (*port_del
)(struct dpif
*dpif
, odp_port_t port_no
);
170 /* Refreshes configuration of 'dpif's port. The implementation might
171 * postpone applying the changes until run() is called. */
172 int (*port_set_config
)(struct dpif
*dpif
, odp_port_t port_no
,
173 const struct smap
*cfg
);
175 /* Queries 'dpif' for a port with the given 'port_no' or 'devname'.
176 * If 'port' is not null, stores information about the port into
177 * '*port' if successful.
179 * If the port doesn't exist, the provider must return ENODEV. Other
180 * error numbers means that something wrong happened and will be
181 * treated differently by upper layers.
183 * If 'port' is not null, the caller takes ownership of data in
184 * 'port' and must free it with dpif_port_destroy() when it is no
186 int (*port_query_by_number
)(const struct dpif
*dpif
, odp_port_t port_no
,
187 struct dpif_port
*port
);
188 int (*port_query_by_name
)(const struct dpif
*dpif
, const char *devname
,
189 struct dpif_port
*port
);
191 /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
192 * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
193 * flows whose packets arrived on port 'port_no'. In the case where the
194 * provider allocates multiple Netlink PIDs to a single port, it may use
195 * 'hash' to spread load among them. The caller need not use a particular
196 * hash function; a 5-tuple hash is suitable.
198 * (The datapath implementation might use some different hash function for
199 * distributing packets received via flow misses among PIDs. This means
200 * that packets received via flow misses might be reordered relative to
201 * packets received via userspace actions. This is not ordinarily a
204 * A 'port_no' of UINT32_MAX should be treated as a special case. The
205 * implementation should return a reserved PID, not allocated to any port,
206 * that the client may use for special purposes.
208 * The return value only needs to be meaningful when DPIF_UC_ACTION has
209 * been enabled in the 'dpif''s listen mask, and it is allowed to change
210 * when DPIF_UC_ACTION is disabled and then re-enabled.
212 * A dpif provider that doesn't have meaningful Netlink PIDs can use NULL
213 * for this function. This is equivalent to always returning 0. */
214 uint32_t (*port_get_pid
)(const struct dpif
*dpif
, odp_port_t port_no
,
217 /* Attempts to begin dumping the ports in a dpif. On success, returns 0
218 * and initializes '*statep' with any data needed for iteration. On
219 * failure, returns a positive errno value. */
220 int (*port_dump_start
)(const struct dpif
*dpif
, void **statep
);
222 /* Attempts to retrieve another port from 'dpif' for 'state', which was
223 * initialized by a successful call to the 'port_dump_start' function for
224 * 'dpif'. On success, stores a new dpif_port into 'port' and returns 0.
225 * Returns EOF if the end of the port table has been reached, or a positive
226 * errno value on error. This function will not be called again once it
227 * returns nonzero once for a given iteration (but the 'port_dump_done'
228 * function will be called afterward).
230 * The dpif provider retains ownership of the data stored in 'port'. It
231 * must remain valid until at least the next call to 'port_dump_next' or
232 * 'port_dump_done' for 'state'. */
233 int (*port_dump_next
)(const struct dpif
*dpif
, void *state
,
234 struct dpif_port
*port
);
236 /* Releases resources from 'dpif' for 'state', which was initialized by a
237 * successful call to the 'port_dump_start' function for 'dpif'. */
238 int (*port_dump_done
)(const struct dpif
*dpif
, void *state
);
240 /* Polls for changes in the set of ports in 'dpif'. If the set of ports in
241 * 'dpif' has changed, then this function should do one of the
244 * - Preferably: store the name of the device that was added to or deleted
245 * from 'dpif' in '*devnamep' and return 0. The caller is responsible
246 * for freeing '*devnamep' (with free()) when it no longer needs it.
248 * - Alternatively: return ENOBUFS, without indicating the device that was
251 * Occasional 'false positives', in which the function returns 0 while
252 * indicating a device that was not actually added or deleted or returns
253 * ENOBUFS without any change, are acceptable.
255 * If the set of ports in 'dpif' has not changed, returns EAGAIN. May also
256 * return other positive errno values to indicate that something has gone
258 int (*port_poll
)(const struct dpif
*dpif
, char **devnamep
);
260 /* Arranges for the poll loop to wake up when 'port_poll' will return a
261 * value other than EAGAIN. */
262 void (*port_poll_wait
)(const struct dpif
*dpif
);
264 /* Deletes all flows from 'dpif' and clears all of its queues of received
266 int (*flow_flush
)(struct dpif
*dpif
);
268 /* Flow dumping interface.
270 * This is the back-end for the flow dumping interface described in
271 * dpif.h. Please read the comments there first, because this code
272 * closely follows it.
274 * 'flow_dump_create' and 'flow_dump_thread_create' must always return an
275 * initialized and usable data structure and defer error return until
276 * flow_dump_destroy(). This hasn't been a problem for the dpifs that
279 * 'flow_dump_create' and 'flow_dump_thread_create' must initialize the
280 * structures that they return with dpif_flow_dump_init() and
281 * dpif_flow_dump_thread_init(), respectively.
283 * If 'terse' is true, then only UID and statistics will
284 * be returned in the dump. Otherwise, all fields will be returned.
286 * If 'type' isn't null, dumps only the flows of the given type. */
287 struct dpif_flow_dump
*(*flow_dump_create
)(const struct dpif
*dpif
,
288 bool terse
, char *type
);
289 int (*flow_dump_destroy
)(struct dpif_flow_dump
*dump
);
291 struct dpif_flow_dump_thread
*(*flow_dump_thread_create
)(
292 struct dpif_flow_dump
*dump
);
293 void (*flow_dump_thread_destroy
)(struct dpif_flow_dump_thread
*thread
);
295 int (*flow_dump_next
)(struct dpif_flow_dump_thread
*thread
,
296 struct dpif_flow
*flows
, int max_flows
);
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 and the 'error' member of each
302 void (*operate
)(struct dpif
*dpif
, struct dpif_op
**ops
, size_t n_ops
);
304 /* Enables or disables receiving packets with dpif_recv() for 'dpif'.
305 * Turning packet receive off and then back on is allowed to change Netlink
306 * PID assignments (see ->port_get_pid()). The client is responsible for
307 * updating flows as necessary if it does this. */
308 int (*recv_set
)(struct dpif
*dpif
, bool enable
);
310 /* Refreshes the poll loops and Netlink sockets associated to each port,
311 * when the number of upcall handlers (upcall receiving thread) is changed
312 * to 'n_handlers' and receiving packets for 'dpif' is enabled by
315 * Since multiple upcall handlers can read upcalls simultaneously from
316 * 'dpif', each port can have multiple Netlink sockets, one per upcall
317 * handler. So, handlers_set() is responsible for the following tasks:
319 * When receiving upcall is enabled, extends or creates the
320 * configuration to support:
322 * - 'n_handlers' Netlink sockets for each port.
324 * - 'n_handlers' poll loops, one for each upcall handler.
326 * - registering the Netlink sockets for the same upcall handler to
327 * the corresponding poll loop.
329 int (*handlers_set
)(struct dpif
*dpif
, uint32_t n_handlers
);
331 /* Pass custom configuration options to the datapath. The implementation
332 * might postpone applying the changes until run() is called. */
333 int (*set_config
)(struct dpif
*dpif
, const struct smap
*other_config
);
335 /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
336 * priority value used for setting packet priority. */
337 int (*queue_to_priority
)(const struct dpif
*dpif
, uint32_t queue_id
,
340 /* Polls for an upcall from 'dpif' for an upcall handler. Since there
341 * can be multiple poll loops (see ->handlers_set()), 'handler_id' is
342 * needed as index to identify the corresponding poll loop. If
343 * successful, stores the upcall into '*upcall', using 'buf' for
344 * storage. Should only be called if 'recv_set' has been used to enable
345 * receiving packets from 'dpif'.
347 * The implementation should point 'upcall->key' and 'upcall->userdata'
348 * (if any) into data in the caller-provided 'buf'. The implementation may
349 * also use 'buf' for storing the data of 'upcall->packet'. If necessary
350 * to make room, the implementation may reallocate the data in 'buf'.
352 * The caller owns the data of 'upcall->packet' and may modify it. If
353 * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
354 * will be reallocated. This requires the data of 'upcall->packet' to be
355 * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
356 * when an error is returned, the 'upcall->packet' may be uninitialized
357 * and should not be released.
359 * This function must not block. If no upcall is pending when it is
360 * called, it should return EAGAIN without blocking. */
361 int (*recv
)(struct dpif
*dpif
, uint32_t handler_id
,
362 struct dpif_upcall
*upcall
, struct ofpbuf
*buf
);
364 /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
365 * has a message queued to be received with the recv member functions.
366 * Since there can be multiple poll loops (see ->handlers_set()),
367 * 'handler_id' is needed as index to identify the corresponding poll loop.
369 void (*recv_wait
)(struct dpif
*dpif
, uint32_t handler_id
);
371 /* Throws away any queued upcalls that 'dpif' currently has ready to
373 void (*recv_purge
)(struct dpif
*dpif
);
375 /* When 'dpif' is about to purge the datapath, the higher layer may want
376 * to be notified so that it could try reacting accordingly (e.g. grabbing
377 * all flow stats before they are gone).
379 * Registers an upcall callback function with 'dpif'. This is only used
380 * if 'dpif' needs to notify the purging of datapath. 'aux' is passed to
381 * the callback on invocation. */
382 void (*register_dp_purge_cb
)(struct dpif
*, dp_purge_callback
*, void *aux
);
384 /* For datapaths that run in userspace (i.e. dpif-netdev), threads polling
385 * for incoming packets can directly call upcall functions instead of
386 * offloading packet processing to separate handler threads. Datapaths
387 * that directly call upcall functions should use the functions below to
388 * to register an upcall function and enable / disable upcalls.
390 * Registers an upcall callback function with 'dpif'. This is only used
391 * if 'dpif' directly executes upcall functions. 'aux' is passed to the
392 * callback on invocation. */
393 void (*register_upcall_cb
)(struct dpif
*, upcall_callback
*, void *aux
);
395 /* Enables upcalls if 'dpif' directly executes upcall functions. */
396 void (*enable_upcall
)(struct dpif
*);
398 /* Disables upcalls if 'dpif' directly executes upcall functions. */
399 void (*disable_upcall
)(struct dpif
*);
401 /* Get datapath version. Caller is responsible for freeing the string
403 char *(*get_datapath_version
)(void);
405 /* Conntrack entry dumping interface.
407 * These functions are used by ct-dpif.c to provide a datapath-agnostic
408 * dumping interface to the connection trackers provided by the
411 * ct_dump_start() should put in '*state' a pointer to a newly allocated
412 * stucture that will be passed by the caller to ct_dump_next() and
413 * ct_dump_done(). If 'zone' is not NULL, only the entries in '*zone'
416 * ct_dump_next() should fill 'entry' with information from a connection
417 * and prepare to dump the next one on a subsequest invocation.
419 * ct_dump_done() should perform any cleanup necessary (including
420 * deallocating the 'state' structure, if applicable). */
421 int (*ct_dump_start
)(struct dpif
*, struct ct_dpif_dump_state
**state
,
422 const uint16_t *zone
, int *);
423 int (*ct_dump_next
)(struct dpif
*, struct ct_dpif_dump_state
*state
,
424 struct ct_dpif_entry
*entry
);
425 int (*ct_dump_done
)(struct dpif
*, struct ct_dpif_dump_state
*state
);
427 /* Flushes the connection tracking tables. If 'zone' is not NULL,
428 * only deletes connections in '*zone'. */
429 int (*ct_flush
)(struct dpif
*, const uint16_t *zone
);
433 /* Queries 'dpif' for supported meter features.
434 * NULL pointer means no meter features are supported. */
435 void (*meter_get_features
)(const struct dpif
*,
436 struct ofputil_meter_features
*);
438 /* Adds or modifies 'meter' in 'dpif'. If '*meter_id' is UINT32_MAX,
439 * adds a new meter, otherwise modifies an existing meter.
441 * If meter is successfully added, sets '*meter_id' to the new meter's
442 * meter id selected by 'dpif'. */
443 int (*meter_set
)(struct dpif
*, ofproto_meter_id
*meter_id
,
444 struct ofputil_meter_config
*);
446 /* Queries 'dpif' for meter stats with the given 'meter_id'. Stores
447 * maximum of 'n_bands' meter statistics, returning the number of band
448 * stats returned in 'stats->n_bands' if successful. */
449 int (*meter_get
)(const struct dpif
*, ofproto_meter_id meter_id
,
450 struct ofputil_meter_stats
*, uint16_t n_bands
);
452 /* Removes meter 'meter_id' from 'dpif'. Stores meter and band statistics
453 * (for maximum of 'n_bands', returning the number of band stats returned
454 * in 'stats->n_bands' if successful. 'stats' may be passed in as NULL if
455 * no stats are needed, in which case 'n_bands' must be passed in as
457 int (*meter_del
)(struct dpif
*, ofproto_meter_id meter_id
,
458 struct ofputil_meter_stats
*, uint16_t n_bands
);
461 extern const struct dpif_class dpif_netlink_class
;
462 extern const struct dpif_class dpif_netdev_class
;
468 #endif /* dpif-provider.h */