/*
- * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
+ * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
ovs_assert(dpif->dpif_class == dpif_class);
}
+struct dpif_flow_dump {
+ struct dpif *dpif;
+ bool terse; /* If true, key/mask/actions may be omitted. */
+};
+
+static inline void
+dpif_flow_dump_init(struct dpif_flow_dump *dump, const struct dpif *dpif)
+{
+ dump->dpif = CONST_CAST(struct dpif *, dpif);
+}
+
+struct dpif_flow_dump_thread {
+ struct dpif *dpif;
+};
+
+static inline void
+dpif_flow_dump_thread_init(struct dpif_flow_dump_thread *thread,
+ struct dpif_flow_dump *dump)
+{
+ thread->dpif = dump->dpif;
+}
+
+struct ct_dpif_dump_state;
+struct ct_dpif_entry;
+
/* Datapath interface class structure, to be defined by each implementation of
* a datapath interface.
*
* the type assumed if no type is specified when opening a dpif. */
const char *type;
- /* 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.
+ /* Called when the dpif provider is registered, typically at program
+ * startup. Returning an error from this function will prevent any
+ * datapath with this class from being created.
+ *
+ * This function may be set to null if a datapath class needs no
+ * initialization at registration time. */
+ int (*init)(void);
+
+ /* Enumerates the names of all known created datapaths (of class
+ * 'dpif_class'), if possible, into 'all_dps'. The caller has already
+ * initialized 'all_dps' and other dpif classes might already have added
+ * names to it.
*
* This is used by the vswitch at startup, so that it can delete any
* datapaths that are not configured.
*
* Some kinds of datapaths might not be practically enumerable, in which
* case this function may be a null pointer. */
- int (*enumerate)(struct sset *all_dps);
+ int (*enumerate)(struct sset *all_dps, const struct dpif_class *dpif_class);
/* 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
* the 'close' member function. */
int (*destroy)(struct dpif *dpif);
- /* Performs periodic work needed by 'dpif', if any is necessary. */
- void (*run)(struct dpif *dpif);
+ /* Performs periodic work needed by 'dpif', if any is necessary.
+ * Returns true if need to revalidate. */
+ bool (*run)(struct dpif *dpif);
/* Arranges for poll_block() to wake up if the "run" member function needs
* to be called for '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'.
+ * 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,
*
* 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 (*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
* value other than EAGAIN. */
void (*port_poll_wait)(const struct dpif *dpif);
- /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink
- * 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
- * failure, returns a positive errno value.
- *
- * If 'actionsp' is nonnull, then on success '*actionsp' must be set to an
- * ofpbuf owned by the caller that contains the Netlink attributes for the
- * flow's actions. The caller must free the ofpbuf (with ofpbuf_delete())
- * when it is no longer needed.
- *
- * If 'stats' is nonnull, then on success it must be updated with the
- * flow's statistics. */
- int (*flow_get)(const struct dpif *dpif,
- const struct nlattr *key, size_t key_len,
- 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 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 'put->flags' includes DPIF_FP_CREATE. Otherwise the
- * operation will fail with ENOENT.
- *
- * 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 '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 'put->stats', if nonnull, must be set
- * to the flow's statistics before the update.
- */
- 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 OVS_KEY_ATTR_* in the 'del->key_len' bytes
- * starting at 'del->key'.
- *
- * 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);
- /* 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 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,
- 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 dpif_flow_stats **stats);
+ /* Flow dumping interface.
+ *
+ * This is the back-end for the flow dumping interface described in
+ * dpif.h. Please read the comments there first, because this code
+ * closely follows it.
+ *
+ * 'flow_dump_create' and 'flow_dump_thread_create' must always return an
+ * initialized and usable data structure and defer error return until
+ * flow_dump_destroy(). This hasn't been a problem for the dpifs that
+ * exist so far.
+ *
+ * 'flow_dump_create' and 'flow_dump_thread_create' must initialize the
+ * structures that they return with dpif_flow_dump_init() and
+ * dpif_flow_dump_thread_init(), respectively.
+ *
+ * If 'terse' is true, then only UID and statistics will
+ * be returned in the dump. Otherwise, all fields will be returned. */
+ struct dpif_flow_dump *(*flow_dump_create)(const struct dpif *dpif,
+ bool terse);
+ int (*flow_dump_destroy)(struct dpif_flow_dump *dump);
- /* 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);
+ struct dpif_flow_dump_thread *(*flow_dump_thread_create)(
+ struct dpif_flow_dump *dump);
+ void (*flow_dump_thread_destroy)(struct dpif_flow_dump_thread *thread);
- /* 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);
+ int (*flow_dump_next)(struct dpif_flow_dump_thread *thread,
+ struct dpif_flow *flows, int max_flows);
/* 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. */
+ * "output" members documented in comments and the 'error' member of each
+ * dpif_op. */
void (*operate)(struct dpif *dpif, struct dpif_op **ops, size_t n_ops);
/* Enables or disables receiving packets with dpif_recv() for 'dpif'.
* 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);
+
+ /* If 'dpif' creates its own I/O polling threads, refreshes poll threads
+ * configuration. 'n_rxqs' configures the number of rx_queues, which
+ * are distributed among threads. 'cmask' configures the cpu mask
+ * for setting the polling threads' cpu affinity. */
+ int (*poll_threads_set)(struct dpif *dpif, unsigned int n_rxqs,
+ const char *cmask);
+
/* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a
* 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', using 'buf' for storage. Should only be called if 'recv_set'
- * has been used to enable receiving packets from 'dpif'.
+ /* 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 implementation should point 'upcall->key' and 'upcall->userdata'
* (if any) into data in the caller-provided 'buf'. The implementation may
*
* 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,
- struct ofpbuf *buf);
+ 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. */
void (*recv_purge)(struct dpif *dpif);
+
+ /* When 'dpif' is about to purge the datapath, the higher layer may want
+ * to be notified so that it could try reacting accordingly (e.g. grabbing
+ * all flow stats before they are gone).
+ *
+ * Registers an upcall callback function with 'dpif'. This is only used
+ * if 'dpif' needs to notify the purging of datapath. 'aux' is passed to
+ * the callback on invocation. */
+ void (*register_dp_purge_cb)(struct dpif *, dp_purge_callback *, void *aux);
+
+ /* For datapaths that run in userspace (i.e. dpif-netdev), threads polling
+ * for incoming packets can directly call upcall functions instead of
+ * offloading packet processing to separate handler threads. Datapaths
+ * that directly call upcall functions should use the functions below to
+ * to register an upcall function and enable / disable upcalls.
+ *
+ * Registers an upcall callback function with 'dpif'. This is only used
+ * if 'dpif' directly executes upcall functions. 'aux' is passed to the
+ * callback on invocation. */
+ void (*register_upcall_cb)(struct dpif *, upcall_callback *, void *aux);
+
+ /* Enables upcalls if 'dpif' directly executes upcall functions. */
+ void (*enable_upcall)(struct dpif *);
+
+ /* Disables upcalls if 'dpif' directly executes upcall functions. */
+ void (*disable_upcall)(struct dpif *);
+
+ /* Get datapath version. Caller is responsible for freeing the string
+ * returned. */
+ char *(*get_datapath_version)(void);
+
+ /* Conntrack entry dumping interface.
+ *
+ * These functions are used by ct-dpif.c to provide a datapath-agnostic
+ * dumping interface to the connection trackes provided by the
+ * datapaths.
+ *
+ * ct_dump_start() should put in '*state' a pointer to a newly allocated
+ * stucture that will be passed by the caller to ct_dump_next() and
+ * ct_dump_done(). If 'zone' is not NULL, only the entries in '*zone'
+ * should be dumped.
+ *
+ * ct_dump_next() should fill 'entry' with information from a connection
+ * and prepare to dump the next one on a subsequest invocation.
+ *
+ * ct_dump_done should perform any cleanup necessary (including
+ * deallocating the 'state' structure, if applicable). */
+ int (*ct_dump_start)(struct dpif *, struct ct_dpif_dump_state **state,
+ const uint16_t *zone);
+ int (*ct_dump_next)(struct dpif *, struct ct_dpif_dump_state *,
+ struct ct_dpif_entry *entry);
+ int (*ct_dump_done)(struct dpif *, struct ct_dpif_dump_state *state);
+
+ /* Flushes the connection tracking tables. If 'zone' is not NULL,
+ * only deletes connections in '*zone'. */
+ int (*ct_flush)(struct dpif *, const uint16_t *zone);
};
-extern const struct dpif_class dpif_linux_class;
+extern const struct dpif_class dpif_netlink_class;
extern const struct dpif_class dpif_netdev_class;
#ifdef __cplusplus