lib/dpif-provider.h

   1 /*
   2  * Copyright (c) 2009, 2010, 2011 Nicira Networks.
   3  *
   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:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  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.
  15  */
  16
  17 #ifndef DPIF_PROVIDER_H
  18 #define DPIF_PROVIDER_H 1
  19
  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. */
  24
  25 #include <assert.h>
  26 #include "openflow/openflow.h"
  27 #include "dpif.h"
  28 #include "util.h"
  29
  30 #ifdef  __cplusplus
  31 extern "C" {
  32 #endif
  33
  34 /* Open vSwitch datapath interface.
  35  *
  36  * This structure should be treated as opaque by dpif implementations. */
  37 struct dpif {
  38     const struct dpif_class *dpif_class;
  39     char *base_name;
  40     char *full_name;
  41     uint8_t netflow_engine_type;
  42     uint8_t netflow_engine_id;
  43 };
  44
  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);
  48
  49 static inline void dpif_assert_class(const struct dpif *dpif,
  50                                      const struct dpif_class *dpif_class)
  51 {
  52     assert(dpif->dpif_class == dpif_class);
  53 }
  54
  55 /* Datapath interface class structure, to be defined by each implementation of
  56  * a datapath interface.
  57  *
  58  * These functions return 0 if successful or a positive errno value on failure,
  59  * except where otherwise noted.
  60  *
  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. */
  65 struct dpif_class {
  66     /* Type of dpif in this class, e.g. "system", "netdev", etc.
  67      *
  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. */
  70     const char *type;
  71
  72     /* Performs periodic work needed by dpifs of this class, if any is
  73      * necessary. */
  74     void (*run)(void);
  75
  76     /* Arranges for poll_block() to wake up if the "run" member function needs
  77      * to be called. */
  78     void (*wait)(void);
  79
  80     /* Enumerates the names of all known created datapaths, if possible, into
  81      * 'all_dps'.  The caller has already initialized 'all_dps' and other dpif
  82      * classes might already have added names to it.
  83      *
  84      * This is used by the vswitch at startup, so that it can delete any
  85      * datapaths that are not configured.
  86      *
  87      * Some kinds of datapaths might not be practically enumerable, in which
  88      * case this function may be a null pointer. */
  89     int (*enumerate)(struct svec *all_dps);
  90
  91     /* Attempts to open an existing dpif called 'name', if 'create' is false,
  92      * or to open an existing dpif or create a new one, if 'create' is true.
  93      *
  94      * 'dpif_class' is the class of dpif to open.
  95      *
  96      * If successful, stores a pointer to the new dpif in '*dpifp', which must
  97      * have class 'dpif_class'.  On failure there are no requirements on what
  98      * is stored in '*dpifp'. */
  99     int (*open)(const struct dpif_class *dpif_class,
 100                 const char *name, bool create, struct dpif **dpifp);
 101
 102     /* Closes 'dpif' and frees associated memory. */
 103     void (*close)(struct dpif *dpif);
 104
 105     /* Enumerates all names that may be used to open 'dpif' into 'all_names'.
 106      * The Linux datapath, for example, supports opening a datapath both by
 107      * number, e.g. "dp0", and by the name of the datapath's local port.  For
 108      * some datapaths, this might be an infinite set (e.g. in a file name,
 109      * slashes may be duplicated any number of times), in which case only the
 110      * names most likely to be used should be enumerated.
 111      *
 112      * The caller has already initialized 'all_names' and might already have
 113      * added some names to it.  This function should not disturb any existing
 114      * names in 'all_names'.
 115      *
 116      * If a datapath class does not support multiple names for a datapath, this
 117      * function may be a null pointer.
 118      *
 119      * This is used by the vswitch at startup, */
 120     int (*get_all_names)(const struct dpif *dpif, struct svec *all_names);
 121
 122     /* Attempts to destroy the dpif underlying 'dpif'.
 123      *
 124      * If successful, 'dpif' will not be used again except as an argument for
 125      * the 'close' member function. */
 126     int (*destroy)(struct dpif *dpif);
 127
 128     /* Retrieves statistics for 'dpif' into 'stats'. */
 129     int (*get_stats)(const struct dpif *dpif, struct odp_stats *stats);
 130
 131     /* Retrieves 'dpif''s current treatment of IP fragments into '*drop_frags':
 132      * true indicates that fragments are dropped, false indicates that
 133      * fragments are treated in the same way as other IP packets (except that
 134      * the L4 header cannot be read). */
 135     int (*get_drop_frags)(const struct dpif *dpif, bool *drop_frags);
 136
 137     /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose
 138      * meaning is the same as for the get_drop_frags member function. */
 139     int (*set_drop_frags)(struct dpif *dpif, bool drop_frags);
 140
 141     /* Adds 'netdev' as a new port in 'dpif'.  If successful, sets '*port_no'
 142      * to the new port's port number. */
 143     int (*port_add)(struct dpif *dpif, struct netdev *netdev,
 144                     uint16_t *port_no);
 145
 146     /* Removes port numbered 'port_no' from 'dpif'. */
 147     int (*port_del)(struct dpif *dpif, uint16_t port_no);
 148
 149     /* Queries 'dpif' for a port with the given 'port_no' or 'devname'.  Stores
 150      * information about the port into '*port' if successful.
 151      *
 152      * The caller takes ownership of data in 'port' and must free it with
 153      * dpif_port_destroy() when it is no longer needed. */
 154     int (*port_query_by_number)(const struct dpif *dpif, uint16_t port_no,
 155                                 struct dpif_port *port);
 156     int (*port_query_by_name)(const struct dpif *dpif, const char *devname,
 157                               struct dpif_port *port);
 158
 159     /* Returns one greater than the largest port number accepted in flow
 160      * actions. */
 161     int (*get_max_ports)(const struct dpif *dpif);
 162
 163     /* Attempts to begin dumping the ports in a dpif.  On success, returns 0
 164      * and initializes '*statep' with any data needed for iteration.  On
 165      * failure, returns a positive errno value. */
 166     int (*port_dump_start)(const struct dpif *dpif, void **statep);
 167
 168     /* Attempts to retrieve another port from 'dpif' for 'state', which was
 169      * initialized by a successful call to the 'port_dump_start' function for
 170      * 'dpif'.  On success, stores a new dpif_port into 'port' and returns 0.
 171      * Returns EOF if the end of the port table has been reached, or a positive
 172      * errno value on error.  This function will not be called again once it
 173      * returns nonzero once for a given iteration (but the 'port_dump_done'
 174      * function will be called afterward).
 175      *
 176      * The dpif provider retains ownership of the data stored in 'port'.  It
 177      * must remain valid until at least the next call to 'port_dump_next' or
 178      * 'port_dump_done' for 'state'. */
 179     int (*port_dump_next)(const struct dpif *dpif, void *state,
 180                           struct dpif_port *port);
 181
 182     /* Releases resources from 'dpif' for 'state', which was initialized by a
 183      * successful call to the 'port_dump_start' function for 'dpif'.  */
 184     int (*port_dump_done)(const struct dpif *dpif, void *state);
 185
 186     /* Polls for changes in the set of ports in 'dpif'.  If the set of ports in
 187      * 'dpif' has changed, then this function should do one of the
 188      * following:
 189      *
 190      * - Preferably: store the name of the device that was added to or deleted
 191      *   from 'dpif' in '*devnamep' and return 0.  The caller is responsible
 192      *   for freeing '*devnamep' (with free()) when it no longer needs it.
 193      *
 194      * - Alternatively: return ENOBUFS, without indicating the device that was
 195      *   added or deleted.
 196      *
 197      * Occasional 'false positives', in which the function returns 0 while
 198      * indicating a device that was not actually added or deleted or returns
 199      * ENOBUFS without any change, are acceptable.
 200      *
 201      * If the set of ports in 'dpif' has not changed, returns EAGAIN.  May also
 202      * return other positive errno values to indicate that something has gone
 203      * wrong. */
 204     int (*port_poll)(const struct dpif *dpif, char **devnamep);
 205
 206     /* Arranges for the poll loop to wake up when 'port_poll' will return a
 207      * value other than EAGAIN. */
 208     void (*port_poll_wait)(const struct dpif *dpif);
 209
 210     /* Queries 'dpif' for a flow entry.  The flow is specified by the Netlink
 211      * attributes with types ODP_KEY_ATTR_* in the 'key_len' bytes starting at
 212      * 'key'.
 213      *
 214      * Returns 0 if successful.  If no flow matches, returns ENOENT.  On other
 215      * failure, returns a positive errno value.
 216      *
 217      * If 'actionsp' is nonnull, then on success '*actionsp' must be set to an
 218      * ofpbuf owned by the caller that contains the Netlink attributes for the
 219      * flow's actions.  The caller must free the ofpbuf (with ofpbuf_delete())
 220      * when it is no longer needed.
 221      *
 222      * If 'stats' is nonnull, then on success it must be updated with the
 223      * flow's statistics. */
 224     int (*flow_get)(const struct dpif *dpif, int flags,
 225                     const struct nlattr *key, size_t key_len,
 226                     struct ofpbuf **actionsp, struct odp_flow_stats *stats);
 227
 228     /* Adds or modifies a flow in 'dpif'.  The flow is specified by the Netlink
 229      * attributes with types ODP_KEY_ATTR_* in the 'key_len' bytes starting at
 230      * 'key'.  The associated actions are specified by the Netlink attributes
 231      * with types ODPAT_* in the 'actions_len' bytes starting at 'actions'.
 232      *
 233      * - If the flow's key does not exist in 'dpif', then the flow will be
 234      *   added if 'flags' includes ODPPF_CREATE.  Otherwise the operation will
 235      *   fail with ENOENT.
 236      *
 237      *   If the operation succeeds, then 'stats', if nonnull, must be zeroed.
 238      *
 239      * - If the flow's key does exist in 'dpif', then the flow's actions will
 240      *   be updated if 'flags' includes ODPPF_MODIFY.  Otherwise the operation
 241      *   will fail with EEXIST.  If the flow's actions are updated, then its
 242      *   statistics will be zeroed if 'flags' includes ODPPF_ZERO_STATS, and
 243      *   left as-is otherwise.
 244      *
 245      *   If the operation succeeds, then 'stats', if nonnull, must be set to
 246      *   the flow's statistics before the update.
 247      */
 248     int (*flow_put)(struct dpif *dpif, int flags,
 249                     const struct nlattr *key, size_t key_len,
 250                     const struct nlattr *actions, size_t actions_len,
 251                     struct odp_flow_stats *stats);
 252
 253     /* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif'
 254      * does not contain such a flow.  The flow is specified by the Netlink
 255      * attributes with types ODP_KEY_ATTR_* in the 'key_len' bytes starting at
 256      * 'key'.
 257      *
 258      * If the operation succeeds, then 'stats', if nonnull, must be set to the
 259      * flow's statistics before its deletion. */
 260     int (*flow_del)(struct dpif *dpif,
 261                     const struct nlattr *key, size_t key_len,
 262                     struct odp_flow_stats *stats);
 263
 264     /* Deletes all flows from 'dpif' and clears all of its queues of received
 265      * packets. */
 266     int (*flow_flush)(struct dpif *dpif);
 267
 268     /* Attempts to begin dumping the flows in a dpif.  On success, returns 0
 269      * and initializes '*statep' with any data needed for iteration.  On
 270      * failure, returns a positive errno value. */
 271     int (*flow_dump_start)(const struct dpif *dpif, void **statep);
 272
 273     /* Attempts to retrieve another flow from 'dpif' for 'state', which was
 274      * initialized by a successful call to the 'flow_dump_start' function for
 275      * 'dpif'.  On success, updates the output parameters as described below
 276      * and returns 0.  Returns EOF if the end of the flow table has been
 277      * reached, or a positive errno value on error.  This function will not be
 278      * called again once it returns nonzero within a given iteration (but the
 279      * 'flow_dump_done' function will be called afterward).
 280      *
 281      * On success, if 'key' and 'key_len' are nonnull then '*key' and
 282      * '*key_len' must be set to Netlink attributes with types ODP_KEY_ATTR_*
 283      * representing the dumped flow's key.  If 'actions' and 'actions_len' are
 284      * nonnull then they should be set to Netlink attributes with types ODPAT_*
 285      * representing the dumped flow's actions.  If 'stats' is nonnull then it
 286      * should be set to the dumped flow's statistics.
 287      *
 288      * All of the returned data is owned by 'dpif', not by the caller, and the
 289      * caller must not modify or free it.  'dpif' must guarantee that it
 290      * remains accessible and unchanging until at least the next call to
 291      * 'flow_dump_next' or 'flow_dump_done' for 'state'. */
 292     int (*flow_dump_next)(const struct dpif *dpif, void *state,
 293                           const struct nlattr **key, size_t *key_len,
 294                           const struct nlattr **actions, size_t *actions_len,
 295                           const struct odp_flow_stats **stats);
 296
 297     /* Releases resources from 'dpif' for 'state', which was initialized by a
 298      * successful call to the 'flow_dump_start' function for 'dpif'.  */
 299     int (*flow_dump_done)(const struct dpif *dpif, void *state);
 300
 301     /* Performs the 'actions_len' bytes of actions in 'actions' on the Ethernet
 302      * frame specified in 'packet'. */
 303     int (*execute)(struct dpif *dpif, const struct nlattr *actions,
 304                    size_t actions_len, const struct ofpbuf *packet);
 305
 306     /* Retrieves 'dpif''s "listen mask" into '*listen_mask'.  Each ODPL_* bit
 307      * set in '*listen_mask' indicates the 'dpif' will receive messages of the
 308      * corresponding type when it calls the recv member function. */
 309     int (*recv_get_mask)(const struct dpif *dpif, int *listen_mask);
 310
 311     /* Sets 'dpif''s "listen mask" to 'listen_mask'.  Each ODPL_* bit set in
 312      * 'listen_mask' indicates the 'dpif' will receive messages of the
 313      * corresponding type when it calls the recv member function. */
 314     int (*recv_set_mask)(struct dpif *dpif, int listen_mask);
 315
 316     /* Retrieves 'dpif''s sFlow sampling probability into '*probability'.
 317      * Return value is 0 or a positive errno value.  EOPNOTSUPP indicates that
 318      * the datapath does not support sFlow, as does a null pointer.
 319      *
 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
 322      * packet. */
 323     int (*get_sflow_probability)(const struct dpif *dpif,
 324                                  uint32_t *probability);
 325
 326     /* Sets 'dpif''s sFlow sampling probability to 'probability'.  Return value
 327      * is 0 or a positive errno value.  EOPNOTSUPP indicates that the datapath
 328      * does not support sFlow, as does a null pointer.
 329      *
 330      * 'probability' is expressed as the number of packets out of UINT_MAX to
 331      * sample, e.g. probability/UINT_MAX is the probability of sampling a given
 332      * packet. */
 333     int (*set_sflow_probability)(struct dpif *dpif, uint32_t probability);
 334
 335     /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
 336      * priority value for use in the ODPAT_SET_PRIORITY action in
 337      * '*priority'. */
 338     int (*queue_to_priority)(const struct dpif *dpif, uint32_t queue_id,
 339                              uint32_t *priority);
 340
 341     /* Polls for an upcall from 'dpif'.  If successful, stores the upcall into
 342      * '*upcall'.  Only upcalls of the types selected with the set_listen_mask
 343      * member function should be received.
 344      *
 345      * The caller takes ownership of the data that 'upcall' points to.
 346      * 'upcall->key' and 'upcall->actions' (if nonnull) point into data owned
 347      * by 'upcall->packet', so their memory cannot be freed separately.  (This
 348      * is hardly a great way to do things but it works out OK for the dpif
 349      * providers that exist so far.)
 350      *
 351      * For greatest efficiency, 'upcall->packet' should have at least
 352      * offsetof(struct ofp_packet_in, data) bytes of headroom.
 353      *
 354      * This function must not block.  If no upcall is pending when it is
 355      * called, it should return EAGAIN without blocking. */
 356     int (*recv)(struct dpif *dpif, struct dpif_upcall *upcall);
 357
 358     /* Arranges for the poll loop to wake up when 'dpif' has a message queued
 359      * to be received with the recv member function. */
 360     void (*recv_wait)(struct dpif *dpif);
 361
 362     /* Throws away any queued upcalls that 'dpif' currently has ready to
 363      * return. */
 364     void (*recv_purge)(struct dpif *dpif);
 365 };
 366
 367 extern const struct dpif_class dpif_linux_class;
 368 extern const struct dpif_class dpif_netdev_class;
 369
 370 #ifdef  __cplusplus
 371 }
 372 #endif
 373
 374 #endif /* dpif-provider.h */