/*
- * Copyright (c) 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* exposed over OpenFlow as a single switch. Datapaths and the collections of
* ports that they contain may be fixed or dynamic. */
-#include <assert.h>
#include "openflow/openflow.h"
#include "dpif.h"
#include "util.h"
static inline void dpif_assert_class(const struct dpif *dpif,
const struct dpif_class *dpif_class)
{
- assert(dpif->dpif_class == dpif_class);
+ ovs_assert(dpif->dpif_class == dpif_class);
}
/* Datapath interface class structure, to be defined by each implementation of
* the type assumed if no type is specified when opening a dpif. */
const char *type;
- /* Performs periodic work needed by dpifs of this class, if any is
- * necessary. */
- void (*run)(void);
-
- /* Arranges for poll_block() to wake up if the "run" member function needs
- * to be called. */
- void (*wait)(void);
-
/* Enumerates the names of all known created datapaths, if possible, into
* 'all_dps'. The caller has already initialized 'all_dps' and other dpif
* classes might already have added names to it.
*
* Some kinds of datapaths might not be practically enumerable, in which
* case this function may be a null pointer. */
- int (*enumerate)(struct svec *all_dps);
+ int (*enumerate)(struct sset *all_dps);
+
+ /* Returns the type to pass to netdev_open() when a dpif of class
+ * 'dpif_class' has a port of type 'type', for a few special cases
+ * when a netdev type differs from a port type. For example, when
+ * using the userspace datapath, a port of type "internal" needs to
+ * be opened as "tap".
+ *
+ * Returns either 'type' itself or a string literal, which must not
+ * be freed. */
+ const char *(*port_open_type)(const struct dpif_class *dpif_class,
+ const char *type);
/* Attempts to open an existing dpif called 'name', if 'create' is false,
* or to open an existing dpif or create a new one, if 'create' is true.
/* Closes 'dpif' and frees associated memory. */
void (*close)(struct dpif *dpif);
- /* Enumerates all names that may be used to open 'dpif' into 'all_names'.
- * The Linux datapath, for example, supports opening a datapath both by
- * number, e.g. "dp0", and by the name of the datapath's local port. For
- * some datapaths, this might be an infinite set (e.g. in a file name,
- * slashes may be duplicated any number of times), in which case only the
- * names most likely to be used should be enumerated.
- *
- * The caller has already initialized 'all_names' and might already have
- * added some names to it. This function should not disturb any existing
- * names in 'all_names'.
- *
- * If a datapath class does not support multiple names for a datapath, this
- * function may be a null pointer.
- *
- * This is used by the vswitch at startup, */
- int (*get_all_names)(const struct dpif *dpif, struct svec *all_names);
-
/* Attempts to destroy the dpif underlying 'dpif'.
*
* If successful, 'dpif' will not be used again except as an argument for
* the 'close' member function. */
int (*destroy)(struct dpif *dpif);
- /* Retrieves statistics for 'dpif' into 'stats'. */
- int (*get_stats)(const struct dpif *dpif, struct odp_stats *stats);
+ /* Performs periodic work needed by 'dpif', if any is necessary. */
+ void (*run)(struct dpif *dpif);
- /* Retrieves 'dpif''s current treatment of IP fragments into '*drop_frags':
- * true indicates that fragments are dropped, false indicates that
- * fragments are treated in the same way as other IP packets (except that
- * the L4 header cannot be read). */
- int (*get_drop_frags)(const struct dpif *dpif, bool *drop_frags);
+ /* Arranges for poll_block() to wake up if the "run" member function needs
+ * to be called for 'dpif'. */
+ void (*wait)(struct dpif *dpif);
- /* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose
- * meaning is the same as for the get_drop_frags member function. */
- int (*set_drop_frags)(struct dpif *dpif, bool drop_frags);
+ /* Retrieves statistics for 'dpif' into 'stats'. */
+ int (*get_stats)(const struct dpif *dpif, struct dpif_dp_stats *stats);
- /* Adds 'netdev' as a new port in 'dpif'. If successful, sets '*port_no'
- * to the new port's port number. */
+ /* Adds 'netdev' as a new port in 'dpif'. If '*port_no' is not
+ * UINT32_MAX, attempts to use that as the port's port number.
+ *
+ * If port is successfully added, sets '*port_no' to the new port's
+ * port number. Returns EBUSY if caller attempted to choose a port
+ * number, and it was in use. */
int (*port_add)(struct dpif *dpif, struct netdev *netdev,
- uint16_t *port_no);
+ odp_port_t *port_no);
/* Removes port numbered 'port_no' from 'dpif'. */
- int (*port_del)(struct dpif *dpif, uint16_t port_no);
+ int (*port_del)(struct dpif *dpif, odp_port_t port_no);
- /* Queries 'dpif' for a port with the given 'port_no' or 'devname'. Stores
- * information about the port into '*port' if successful.
+ /* Queries 'dpif' for a port with the given 'port_no' or 'devname'.
+ * If 'port' is not null, stores information about the port into
+ * '*port' if successful.
*
- * The caller takes ownership of data in 'port' and must free it with
- * dpif_port_destroy() when it is no longer needed. */
- int (*port_query_by_number)(const struct dpif *dpif, uint16_t port_no,
+ * If 'port' is not null, the caller takes ownership of data in
+ * 'port' and must free it with dpif_port_destroy() when it is no
+ * longer needed. */
+ int (*port_query_by_number)(const struct dpif *dpif, odp_port_t port_no,
struct dpif_port *port);
int (*port_query_by_name)(const struct dpif *dpif, const char *devname,
struct dpif_port *port);
- /* Returns one greater than the largest port number accepted in flow
- * actions. */
- int (*get_max_ports)(const struct dpif *dpif);
+ /* Returns the Netlink PID value to supply in OVS_ACTION_ATTR_USERSPACE
+ * actions as the OVS_USERSPACE_ATTR_PID attribute's value, for use in
+ * flows whose packets arrived on port 'port_no'. In the case where the
+ * provider allocates multiple Netlink PIDs to a single port, it may use
+ * 'hash' to spread load among them. The caller need not use a particular
+ * hash function; a 5-tuple hash is suitable.
+ *
+ * (The datapath implementation might use some different hash function for
+ * distributing packets received via flow misses among PIDs. This means
+ * that packets received via flow misses might be reordered relative to
+ * packets received via userspace actions. This is not ordinarily a
+ * problem.)
+ *
+ * A 'port_no' of UINT32_MAX should be treated as a special case. The
+ * implementation should return a reserved PID, not allocated to any port,
+ * that the client may use for special purposes.
+ *
+ * The return value only needs to be meaningful when DPIF_UC_ACTION has
+ * been enabled in the 'dpif''s listen mask, and it is allowed to change
+ * when DPIF_UC_ACTION is disabled and then re-enabled.
+ *
+ * A dpif provider that doesn't have meaningful Netlink PIDs can use NULL
+ * for this function. This is equivalent to always returning 0. */
+ uint32_t (*port_get_pid)(const struct dpif *dpif, odp_port_t port_no,
+ uint32_t hash);
/* Attempts to begin dumping the ports in a dpif. On success, returns 0
* and initializes '*statep' with any data needed for iteration. On
void (*port_poll_wait)(const struct dpif *dpif);
/* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
- * attributes with types ODP_KEY_ATTR_* in the 'key_len' bytes starting at
+ * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at
* 'key'.
*
* Returns 0 if successful. If no flow matches, returns ENOENT. On other
*
* If 'stats' is nonnull, then on success it must be updated with the
* flow's statistics. */
- int (*flow_get)(const struct dpif *dpif, int flags,
+ int (*flow_get)(const struct dpif *dpif,
const struct nlattr *key, size_t key_len,
- struct ofpbuf **actionsp, struct odp_flow_stats *stats);
+ struct ofpbuf **actionsp, struct dpif_flow_stats *stats);
/* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink
- * attributes with types ODP_KEY_ATTR_* in the 'key_len' bytes starting at
- * 'key'. The associated actions are specified by the Netlink attributes
- * with types ODPAT_* in the 'actions_len' bytes starting at 'actions'.
+ * attributes with types OVS_KEY_ATTR_* in the 'put->key_len' bytes
+ * starting at 'put->key'. The associated actions are specified by the
+ * Netlink attributes with types OVS_ACTION_ATTR_* in the
+ * 'put->actions_len' bytes starting at 'put->actions'.
*
* - If the flow's key does not exist in 'dpif', then the flow will be
- * added if 'flags' includes ODPPF_CREATE. Otherwise the operation will
- * fail with ENOENT.
+ * added if 'put->flags' includes DPIF_FP_CREATE. Otherwise the
+ * operation will fail with ENOENT.
*
- * If the operation succeeds, then 'stats', if nonnull, must be zeroed.
+ * If the operation succeeds, then 'put->stats', if nonnull, must be
+ * zeroed.
*
* - If the flow's key does exist in 'dpif', then the flow's actions will
- * be updated if 'flags' includes ODPPF_MODIFY. Otherwise the operation
- * will fail with EEXIST. If the flow's actions are updated, then its
- * statistics will be zeroed if 'flags' includes ODPPF_ZERO_STATS, and
- * left as-is otherwise.
+ * be updated if 'put->flags' includes DPIF_FP_MODIFY. Otherwise the
+ * operation will fail with EEXIST. If the flow's actions are updated,
+ * then its statistics will be zeroed if 'put->flags' includes
+ * DPIF_FP_ZERO_STATS, and left as-is otherwise.
*
- * If the operation succeeds, then 'stats', if nonnull, must be set to
- * the flow's statistics before the update.
+ * If the operation succeeds, then 'put->stats', if nonnull, must be set
+ * to the flow's statistics before the update.
*/
- int (*flow_put)(struct dpif *dpif, int flags,
- const struct nlattr *key, size_t key_len,
- const struct nlattr *actions, size_t actions_len,
- struct odp_flow_stats *stats);
+ int (*flow_put)(struct dpif *dpif, const struct dpif_flow_put *put);
/* Deletes a flow from 'dpif' and returns 0, or returns ENOENT if 'dpif'
* does not contain such a flow. The flow is specified by the Netlink
- * attributes with types ODP_KEY_ATTR_* in the 'key_len' bytes starting at
- * 'key'.
+ * attributes with types OVS_KEY_ATTR_* in the 'del->key_len' bytes
+ * starting at 'del->key'.
*
- * If the operation succeeds, then 'stats', if nonnull, must be set to the
- * flow's statistics before its deletion. */
- int (*flow_del)(struct dpif *dpif,
- const struct nlattr *key, size_t key_len,
- struct odp_flow_stats *stats);
+ * If the operation succeeds, then 'del->stats', if nonnull, must be set to
+ * the flow's statistics before its deletion. */
+ int (*flow_del)(struct dpif *dpif, const struct dpif_flow_del *del);
/* Deletes all flows from 'dpif' and clears all of its queues of received
* packets. */
int (*flow_flush)(struct dpif *dpif);
+ /* Allocates thread-local state for use with the function 'flow_dump_next'.
+ * On return, initializes '*statep' with any private data needed for
+ * iteration. */
+ void (*flow_dump_state_init)(void **statep);
+
/* Attempts to begin dumping the flows in a dpif. On success, returns 0
- * and initializes '*statep' with any data needed for iteration. On
- * failure, returns a positive errno value. */
- int (*flow_dump_start)(const struct dpif *dpif, void **statep);
-
- /* Attempts to retrieve another flow from 'dpif' for 'state', which was
- * initialized by a successful call to the 'flow_dump_start' function for
- * 'dpif'. On success, updates the output parameters as described below
- * and returns 0. Returns EOF if the end of the flow table has been
- * reached, or a positive errno value on error. This function will not be
- * called again once it returns nonzero within a given iteration (but the
- * 'flow_dump_done' function will be called afterward).
- *
- * On success, if 'key' and 'key_len' are nonnull then '*key' and
- * '*key_len' must be set to Netlink attributes with types ODP_KEY_ATTR_*
- * representing the dumped flow's key. If 'actions' and 'actions_len' are
- * nonnull then they should be set to Netlink attributes with types ODPAT_*
- * representing the dumped flow's actions. If 'stats' is nonnull then it
- * should be set to the dumped flow's statistics.
+ * and initializes '*iterp' with any shared data needed for iteration.
+ * On failure, returns a positive errno value. */
+ int (*flow_dump_start)(const struct dpif *dpif, void **iterp);
+
+ /* Attempts to retrieve another flow from 'dpif' for 'iter', using
+ * 'state' for storage. 'iter' must have been initialized by a successful
+ * call to the 'flow_dump_start' function for 'dpif'. 'state' must have
+ * been initialised with a call to the 'flow_dump_state_init' function for
+ * 'dpif.
+ *
+ * On success, updates the output parameters as described below and returns
+ * 0. Returns EOF if the end of the flow table has been reached, or a
+ * positive errno value on error. Multiple threads may use the same 'dpif'
+ * and 'iter' with this function, but all other parameters must be
+ * different for each thread. If this function returns non-zero,
+ * subsequent calls with the same arguments will also return non-zero.
+ *
+ * On success:
+ *
+ * - If 'key' and 'key_len' are nonnull, then '*key' and '*key_len'
+ * must be set to Netlink attributes with types OVS_KEY_ATTR_*
+ * representing the dumped flow's key.
+ *
+ * - If 'mask' and 'mask_len' are nonnull then '*mask' and '*mask_len'
+ * must be set to Netlink attributes with types of OVS_KEY_ATTR_*
+ * representing the dumped flow's mask.
+ *
+ * - If 'actions' and 'actions_len' are nonnull then they should be set
+ * to Netlink attributes with types OVS_ACTION_ATTR_* representing
+ * the dumped flow's actions.
+ *
+ * - If 'stats' is nonnull then it should be set to the dumped flow's
+ * statistics.
*
* All of the returned data is owned by 'dpif', not by the caller, and the
* caller must not modify or free it. 'dpif' must guarantee that it
* remains accessible and unchanging until at least the next call to
- * 'flow_dump_next' or 'flow_dump_done' for 'state'. */
- int (*flow_dump_next)(const struct dpif *dpif, void *state,
+ * 'flow_dump_next' or 'flow_dump_done' for 'iter' and 'state'. */
+ int (*flow_dump_next)(const struct dpif *dpif, void *iter, void *state,
const struct nlattr **key, size_t *key_len,
+ const struct nlattr **mask, size_t *mask_len,
const struct nlattr **actions, size_t *actions_len,
- const struct odp_flow_stats **stats);
+ const struct dpif_flow_stats **stats);
- /* Releases resources from 'dpif' for 'state', which was initialized by a
- * successful call to the 'flow_dump_start' function for 'dpif'. */
- int (*flow_dump_done)(const struct dpif *dpif, void *state);
-
- /* Performs the 'actions_len' bytes of actions in 'actions' on the Ethernet
- * frame specified in 'packet'. */
- int (*execute)(struct dpif *dpif, const struct nlattr *actions,
- size_t actions_len, const struct ofpbuf *packet);
-
- /* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit
- * set in '*listen_mask' indicates the 'dpif' will receive messages of the
- * corresponding type when it calls the recv member function. */
- int (*recv_get_mask)(const struct dpif *dpif, int *listen_mask);
-
- /* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in
- * 'listen_mask' indicates the 'dpif' will receive messages of the
- * corresponding type when it calls the recv member function. */
- int (*recv_set_mask)(struct dpif *dpif, int listen_mask);
-
- /* Retrieves 'dpif''s sFlow sampling probability into '*probability'.
- * Return value is 0 or a positive errno value. EOPNOTSUPP indicates that
- * the datapath does not support sFlow, as does a null pointer.
- *
- * '*probability' is expressed as the number of packets out of UINT_MAX to
- * sample, e.g. probability/UINT_MAX is the probability of sampling a given
- * packet. */
- int (*get_sflow_probability)(const struct dpif *dpif,
- uint32_t *probability);
-
- /* Sets 'dpif''s sFlow sampling probability to 'probability'. Return value
- * is 0 or a positive errno value. EOPNOTSUPP indicates that the datapath
- * does not support sFlow, as does a null pointer.
- *
- * 'probability' is expressed as the number of packets out of UINT_MAX to
- * sample, e.g. probability/UINT_MAX is the probability of sampling a given
- * packet. */
- int (*set_sflow_probability)(struct dpif *dpif, uint32_t probability);
+ /* Determines whether the next call to 'flow_dump_next' with 'state' will
+ * modify or free the keys that it previously returned. 'state' must have
+ * been initialized by a call to 'flow_dump_state_init' for 'dpif'.
+ *
+ * 'dpif' guarantees that data returned by flow_dump_next() will remain
+ * accessible and unchanging until the next call. This function provides a
+ * way for callers to determine whether that guarantee extends beyond the
+ * next call.
+ *
+ * Returns true if the next call to flow_dump_next() is expected to be
+ * destructive to previously returned keys for 'state', false otherwise. */
+ bool (*flow_dump_next_may_destroy_keys)(void *state);
+
+ /* Releases resources from 'dpif' for 'iter', which was initialized by a
+ * successful call to the 'flow_dump_start' function for 'dpif'. Callers
+ * must ensure that this function is called once within a given iteration,
+ * as the final flow dump operation. */
+ int (*flow_dump_done)(const struct dpif *dpif, void *iter);
+
+ /* Releases 'state' which was initialized by a call to the
+ * 'flow_dump_state_init' function for this 'dpif'. */
+ void (*flow_dump_state_uninit)(void *statep);
+
+ /* Performs the 'execute->actions_len' bytes of actions in
+ * 'execute->actions' on the Ethernet frame in 'execute->packet'
+ * and on the packet metadata in 'execute->md'.
+ * May modify both packet and metadata. */
+ int (*execute)(struct dpif *dpif, struct dpif_execute *execute);
+
+ /* Executes each of the 'n_ops' operations in 'ops' on 'dpif', in the order
+ * in which they are specified, placing each operation's results in the
+ * "output" members documented in comments.
+ *
+ * This function is optional. It is only worthwhile to implement it if
+ * 'dpif' can perform operations in batch faster than individually. */
+ void (*operate)(struct dpif *dpif, struct dpif_op **ops, size_t n_ops);
+
+ /* Enables or disables receiving packets with dpif_recv() for 'dpif'.
+ * Turning packet receive off and then back on is allowed to change Netlink
+ * PID assignments (see ->port_get_pid()). The client is responsible for
+ * updating flows as necessary if it does this. */
+ int (*recv_set)(struct dpif *dpif, bool enable);
+
+ /* Refreshes the poll loops and Netlink sockets associated to each port,
+ * when the number of upcall handlers (upcall receiving thread) is changed
+ * to 'n_handlers' and receiving packets for 'dpif' is enabled by
+ * recv_set().
+ *
+ * Since multiple upcall handlers can read upcalls simultaneously from
+ * 'dpif', each port can have multiple Netlink sockets, one per upcall
+ * handler. So, handlers_set() is responsible for the following tasks:
+ *
+ * When receiving upcall is enabled, extends or creates the
+ * configuration to support:
+ *
+ * - 'n_handlers' Netlink sockets for each port.
+ *
+ * - 'n_handlers' poll loops, one for each upcall handler.
+ *
+ * - registering the Netlink sockets for the same upcall handler to
+ * the corresponding poll loop.
+ * */
+ int (*handlers_set)(struct dpif *dpif, uint32_t n_handlers);
/* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
- * priority value for use in the ODPAT_SET_PRIORITY action in
- * '*priority'. */
+ * priority value used for setting packet priority. */
int (*queue_to_priority)(const struct dpif *dpif, uint32_t queue_id,
uint32_t *priority);
- /* Polls for an upcall from 'dpif'. If successful, stores the upcall into
- * '*upcall'. Only upcalls of the types selected with the set_listen_mask
- * member function should be received.
+ /* Polls for an upcall from 'dpif' for an upcall handler. Since there
+ * can be multiple poll loops (see ->handlers_set()), 'handler_id' is
+ * needed as index to identify the corresponding poll loop. If
+ * successful, stores the upcall into '*upcall', using 'buf' for
+ * storage. Should only be called if 'recv_set' has been used to enable
+ * receiving packets from 'dpif'.
*
- * The caller takes ownership of the data that 'upcall' points to.
- * 'upcall->key' and 'upcall->actions' (if nonnull) point into data owned
- * by 'upcall->packet', so their memory cannot be freed separately. (This
- * is hardly a great way to do things but it works out OK for the dpif
- * providers that exist so far.)
+ * The implementation should point 'upcall->key' and 'upcall->userdata'
+ * (if any) into data in the caller-provided 'buf'. The implementation may
+ * also use 'buf' for storing the data of 'upcall->packet'. If necessary
+ * to make room, the implementation may reallocate the data in 'buf'.
*
- * For greatest efficiency, 'upcall->packet' should have at least
- * offsetof(struct ofp_packet_in, data) bytes of headroom.
+ * The caller owns the data of 'upcall->packet' and may modify it. If
+ * packet's headroom is exhausted as it is manipulated, 'upcall->packet'
+ * will be reallocated. This requires the data of 'upcall->packet' to be
+ * released with ofpbuf_uninit() before 'upcall' is destroyed. However,
+ * when an error is returned, the 'upcall->packet' may be uninitialized
+ * and should not be released.
*
* This function must not block. If no upcall is pending when it is
* called, it should return EAGAIN without blocking. */
- int (*recv)(struct dpif *dpif, struct dpif_upcall *upcall);
+ int (*recv)(struct dpif *dpif, uint32_t handler_id,
+ struct dpif_upcall *upcall, struct ofpbuf *buf);
- /* Arranges for the poll loop to wake up when 'dpif' has a message queued
- * to be received with the recv member function. */
- void (*recv_wait)(struct dpif *dpif);
+ /* Arranges for the poll loop for an upcall handler to wake up when 'dpif'
+ * has a message queued to be received with the recv member functions.
+ * Since there can be multiple poll loops (see ->handlers_set()),
+ * 'handler_id' is needed as index to identify the corresponding poll loop.
+ * */
+ void (*recv_wait)(struct dpif *dpif, uint32_t handler_id);
/* Throws away any queued upcalls that 'dpif' currently has ready to
* return. */
extern const struct dpif_class dpif_linux_class;
extern const struct dpif_class dpif_netdev_class;
+extern const struct dpif_class dpif_planetlab_class;
#ifdef __cplusplus
}