X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-provider.h;h=4f04c95f059532b0d949b65f72c5211af670a850;hb=d1673b006d53fdea72c0744e835362ed1917f879;hp=bd159b2a9d90ecd2d9a738160ad24fafc98820ac;hpb=4d678233e981fa319a338f6b0949e9dc625941a4;p=sliver-openvswitch.git diff --git a/lib/dpif-provider.h b/lib/dpif-provider.h index bd159b2a9..4f04c95f0 100644 --- a/lib/dpif-provider.h +++ b/lib/dpif-provider.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009 Nicira Networks. + * Copyright (c) 2009, 2010, 2011, 2012 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -18,31 +18,41 @@ #define DPIF_PROVIDER_H 1 /* Provider interface to dpifs, which provide an interface to an Open vSwitch - * datapath. */ + * datapath. A datapath is a collection of physical or virtual ports that are + * exposed over OpenFlow as a single switch. Datapaths and the collections of + * ports that they contain may be fixed or dynamic. */ -#include +#include "openflow/openflow.h" #include "dpif.h" +#include "util.h" + +#ifdef __cplusplus +extern "C" { +#endif /* Open vSwitch datapath interface. * * This structure should be treated as opaque by dpif implementations. */ struct dpif { - const struct dpif_class *class; - char *name; + const struct dpif_class *dpif_class; + char *base_name; + char *full_name; uint8_t netflow_engine_type; uint8_t netflow_engine_id; }; void dpif_init(struct dpif *, const struct dpif_class *, const char *name, uint8_t netflow_engine_type, uint8_t netflow_engine_id); +void dpif_uninit(struct dpif *dpif, bool close); + static inline void dpif_assert_class(const struct dpif *dpif, - const struct dpif_class *class) + const struct dpif_class *dpif_class) { - assert(dpif->class == class); + ovs_assert(dpif->dpif_class == dpif_class); } /* Datapath interface class structure, to be defined by each implementation of - * a datapath interface + * a datapath interface. * * These functions return 0 if successful or a positive errno value on failure, * except where otherwise noted. @@ -52,23 +62,11 @@ static inline void dpif_assert_class(const struct dpif *dpif, * EWOULDBLOCK or EINPROGRESS. We may relax this requirement in the future if * and when we encounter performance problems. */ struct dpif_class { - /* Prefix for names of dpifs in this class, e.g. "udatapath:". + /* Type of dpif in this class, e.g. "system", "netdev", etc. * - * One dpif class may have the empty string "" as its prefix, in which case - * that dpif class is associated with dpif names that don't match any other - * class name. */ - const char *prefix; - - /* Class name, for use in error messages. */ - const char *name; - - /* 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); + * One of the providers should supply a "system" type, since this is + * 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 @@ -79,80 +77,115 @@ struct dpif_class { * * 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, if 'create' is false, or to open an - * existing dpif or create a new one, if 'create' is true. 'name' is the - * full dpif name provided by the user, e.g. "udatapath:/var/run/mypath". - * This name is useful for error messages but must not be modified. + /* 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. * - * 'suffix' is a copy of 'name' following the dpif's 'prefix'. + * 'dpif_class' is the class of dpif to open. * - * If successful, stores a pointer to the new dpif in '*dpifp'. On failure - * there are no requirements on what is stored in '*dpifp'. */ - int (*open)(const char *name, char *suffix, bool create, - struct dpif **dpifp); + * If successful, stores a pointer to the new dpif in '*dpifp', which must + * have class 'dpif_class'. On failure there are no requirements on what + * is stored in '*dpifp'. */ + int (*open)(const struct dpif_class *dpif_class, + const char *name, bool create, struct dpif **dpifp); /* 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 (*delete)(struct dpif *dpif); + 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); - /* Creates a new port in 'dpif' connected to network device 'devname'. - * 'flags' is a set of ODP_PORT_* flags. If successful, sets '*port_no' - * to the new port's port number. */ - int (*port_add)(struct dpif *dpif, const char *devname, uint16_t flags, - uint16_t *port_no); + /* 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, + uint32_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, uint32_t port_no); - /* Queries 'dpif' for a port with the given 'port_no' or 'devname'. Stores - * information about the port into '*port' if successful. */ - int (*port_query_by_number)(const struct dpif *dpif, uint16_t port_no, - struct odp_port *port); + /* 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. + * + * 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, uint32_t port_no, + struct dpif_port *port); int (*port_query_by_name)(const struct dpif *dpif, const char *devname, - struct odp_port *port); + struct dpif_port *port); - /* Stores in 'ports' information about up to 'n' ports attached to 'dpif', - * in no particular order. Returns the number of ports attached to 'dpif' - * (not the number stored), if successful, otherwise a negative errno - * value. */ - int (*port_list)(const struct dpif *dpif, struct odp_port *ports, int n); + /* 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'. + * + * 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, uint32_t port_no); + + /* Attempts to begin dumping the ports in a dpif. On success, returns 0 + * and initializes '*statep' with any data needed for iteration. On + * failure, returns a positive errno value. */ + int (*port_dump_start)(const struct dpif *dpif, void **statep); + + /* Attempts to retrieve another port from 'dpif' for 'state', which was + * initialized by a successful call to the 'port_dump_start' function for + * 'dpif'. On success, stores a new dpif_port into 'port' and returns 0. + * Returns EOF if the end of the port table has been reached, or a positive + * errno value on error. This function will not be called again once it + * returns nonzero once for a given iteration (but the 'port_dump_done' + * function will be called afterward). + * + * The dpif provider retains ownership of the data stored in 'port'. It + * must remain valid until at least the next call to 'port_dump_next' or + * 'port_dump_done' for 'state'. */ + int (*port_dump_next)(const struct dpif *dpif, void *state, + struct dpif_port *port); + + /* Releases resources from 'dpif' for 'state', which was initialized by a + * successful call to the 'port_dump_start' function for 'dpif'. */ + int (*port_dump_done)(const struct dpif *dpif, void *state); /* Polls for changes in the set of ports in 'dpif'. If the set of ports in * 'dpif' has changed, then this function should do one of the @@ -178,121 +211,151 @@ struct dpif_class { * value other than EAGAIN. */ void (*port_poll_wait)(const struct dpif *dpif); - /* Stores in 'ports' the port numbers of up to 'n' ports that belong to - * 'group' in 'dpif'. Returns the number of ports in 'group' (not the - * number stored), if successful, otherwise a negative errno value. */ - int (*port_group_get)(const struct dpif *dpif, int group, - uint16_t ports[], int n); - - /* Changes port group 'group' in 'dpif' to consist of the 'n' ports whose - * numbers are given in 'ports'. + /* 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'. * - * Use the get_stats member function to obtain the number of supported port - * groups. */ - int (*port_group_set)(struct dpif *dpif, int group, - const uint16_t ports[], int n); - - /* For each flow 'flow' in the 'n' flows in 'flows': + * Returns 0 if successful. If no flow matches, returns ENOENT. On other + * failure, returns a positive errno value. * - * - If a flow matching 'flow->key' exists in 'dpif': + * 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. * - * Stores 0 into 'flow->stats.error' and stores statistics for the flow - * into 'flow->stats'. + * 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 'flow->n_actions' is zero, then 'flow->actions' is ignored. If - * 'flow->n_actions' is nonzero, then 'flow->actions' should point to - * an array of the specified number of actions. At most that many of - * the flow's actions will be copied into that array. - * 'flow->n_actions' will be updated to the number of actions actually - * present in the flow, which may be greater than the number stored if - * the flow has more actions than space available in the array. + * - 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. * - * - Flow-specific errors are indicated by a positive errno value in - * 'flow->stats.error'. In particular, ENOENT indicates that no flow - * matching 'flow->key' exists in 'dpif'. When an error value is stored, - * the contents of 'flow->key' are preserved but other members of 'flow' - * should be treated as indeterminate. + * If the operation succeeds, then 'put->stats', if nonnull, must be + * zeroed. * - * Returns 0 if all 'n' flows in 'flows' were updated (whether they were - * individually successful or not is indicated by 'flow->stats.error', - * however). Returns a positive errno value if an error that prevented - * this update occurred, in which the caller must not depend on any - * elements in 'flows' being updated or not updated. - */ - int (*flow_get)(const struct dpif *dpif, struct odp_flow flows[], int n); - - /* Adds or modifies a flow in 'dpif' as specified in 'put': + * - 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 flow specified in 'put->flow' does not exist in 'dpif', then - * behavior depends on whether ODPPF_CREATE is specified in 'put->flags': - * if it is, the flow will be added, otherwise the operation will fail - * with ENOENT. - * - * - Otherwise, the flow specified in 'put->flow' does exist in 'dpif'. - * Behavior in this case depends on whether ODPPF_MODIFY is specified in - * 'put->flags': if it is, the flow's actions will be updated, otherwise - * the operation will fail with EEXIST. If the flow's actions are - * updated, then its statistics will be zeroed if ODPPF_ZERO_STATS is set - * in 'put->flags', 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, struct odp_flow_put *put); + int (*flow_put)(struct dpif *dpif, const struct dpif_flow_put *put); - /* Deletes a flow matching 'flow->key' from 'dpif' or returns ENOENT if - * 'dpif' does not contain such a flow. + /* 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 successful, updates 'flow->stats', 'flow->n_actions', and - * 'flow->actions' as described in more detail under the flow_get member - * function below. */ - int (*flow_del)(struct dpif *dpif, struct odp_flow *flow); + * 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); - /* Stores up to 'n' flows in 'dpif' into 'flows', updating their statistics - * and actions as described under the flow_get member function. If - * successful, returns the number of flows actually present in 'dpif', - * which might be greater than the number stored (if 'dpif' has more than - * 'n' flows). On failure, returns a negative errno value. */ - int (*flow_list)(const struct dpif *dpif, struct odp_flow flows[], int n); - - /* Performs the 'n_actions' actions in 'actions' on the Ethernet frame - * specified in 'packet'. + /* 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 '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 **actions, size_t *actions_len, + 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 'execute->actions_len' bytes of actions in + * 'execute->actions' on the Ethernet frame specified in 'execute->packet' + * taken from the flow specified in the 'execute->key_len' bytes of + * 'execute->key'. ('execute->key' is mostly redundant with + * 'execute->packet', but it contains some metadata that cannot be + * recovered from 'execute->packet', such as tunnel and in_port.) */ + int (*execute)(struct dpif *dpif, const 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. * - * Pretends that the frame was originally received on the port numbered - * 'in_port'. This affects only ODPAT_OUTPUT_GROUP actions, which will not - * send a packet out their input port. Specify the number of an unused - * port (e.g. UINT16_MAX is currently always unused) to avoid this - * behavior. */ - int (*execute)(struct dpif *dpif, uint16_t in_port, - const union odp_action actions[], int n_actions, - 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); - - /* Attempts to receive a message from 'dpif'. If successful, stores the - * message into '*packetp'. The message, if one is received, must begin - * with 'struct odp_msg' as a header. Only messages of the types selected - * with the set_listen_mask member function should be received. + * 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); + + /* 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'. * - * This function must not block. If no message is ready to be received - * when it is called, it should return EAGAIN without blocking. */ - int (*recv)(struct dpif *dpif, struct ofpbuf **packetp); + * The implementation should point 'upcall->packet' and 'upcall->key' into + * data in the caller-provided 'buf'. If necessary to make room, the + * implementation may expand the data in 'buf'. (This is hardly a great + * way to do things but it works out OK for the dpif providers that exist + * so far.) + * + * 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); /* 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); + + /* Throws away any queued upcalls that 'dpif' currently has ready to + * return. */ + void (*recv_purge)(struct dpif *dpif); }; 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 +} +#endif #endif /* dpif-provider.h */