X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif.c;h=ad143c8ed21c7ad612d334531921c4909670391c;hb=f613a0d72c521ca3a4eeb2c29ac523f6fdf72667;hp=3dc7af00704bb1d703edd3405b65c438cf4338c8;hpb=335562c0b9b190f61c0cc457a4098c2688561387;p=sliver-openvswitch.git diff --git a/lib/dpif.c b/lib/dpif.c index 3dc7af007..ad143c8ed 100644 --- a/lib/dpif.c +++ b/lib/dpif.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -15,39 +15,59 @@ */ #include -#include "dpif.h" +#include "dpif-provider.h" #include #include #include -#include #include -#include -#include -#include -#include -#include #include #include -#include -#include -#include -#include #include "coverage.h" #include "dynamic-string.h" #include "flow.h" +#include "netdev.h" #include "netlink.h" #include "odp-util.h" #include "ofp-print.h" +#include "ofp-util.h" #include "ofpbuf.h" #include "packets.h" #include "poll-loop.h" +#include "shash.h" +#include "sset.h" +#include "timeval.h" #include "util.h" #include "valgrind.h" - #include "vlog.h" -#define THIS_MODULE VLM_dpif + +VLOG_DEFINE_THIS_MODULE(dpif); + +COVERAGE_DEFINE(dpif_destroy); +COVERAGE_DEFINE(dpif_port_add); +COVERAGE_DEFINE(dpif_port_del); +COVERAGE_DEFINE(dpif_flow_flush); +COVERAGE_DEFINE(dpif_flow_get); +COVERAGE_DEFINE(dpif_flow_put); +COVERAGE_DEFINE(dpif_flow_del); +COVERAGE_DEFINE(dpif_flow_query_list); +COVERAGE_DEFINE(dpif_flow_query_list_n); +COVERAGE_DEFINE(dpif_execute); +COVERAGE_DEFINE(dpif_purge); + +static const struct dpif_class *base_dpif_classes[] = { +#ifdef HAVE_NETLINK + &dpif_linux_class, +#endif + &dpif_netdev_class, +}; + +struct registered_dpif_class { + const struct dpif_class *dpif_class; + int refcount; +}; +static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes); /* Rate limit for individual messages going to or from the datapath, output at * DBG level. This is very high because, if these are enabled, it is because @@ -55,550 +75,862 @@ static struct vlog_rate_limit dpmsg_rl = VLOG_RATE_LIMIT_INIT(600, 600); /* Not really much point in logging many dpif errors. */ -static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); +static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(60, 5); + +static void log_flow_message(const struct dpif *dpif, int error, + const char *operation, + const struct nlattr *key, size_t key_len, + const struct dpif_flow_stats *stats, + const struct nlattr *actions, size_t actions_len); +static void log_operation(const struct dpif *, const char *operation, + int error); +static bool should_log_flow_message(int error); + +static void +dp_initialize(void) +{ + static int status = -1; -static int get_minor_from_name(const char *name, unsigned int *minor); -static int name_to_minor(const char *name, unsigned int *minor); -static int lookup_minor(const char *name, unsigned int *minor); -static int open_by_minor(unsigned int minor, struct dpif *); -static int make_openvswitch_device(unsigned int minor, char **fnp); -static void check_rw_odp_flow(struct odp_flow *); + if (status < 0) { + int i; + status = 0; + for (i = 0; i < ARRAY_SIZE(base_dpif_classes); i++) { + dp_register_provider(base_dpif_classes[i]); + } + } +} + +/* Registers a new datapath provider. After successful registration, new + * datapaths of that type can be opened using dpif_open(). */ int -dpif_open(const char *name, struct dpif *dpif) +dp_register_provider(const struct dpif_class *new_class) { - int listen_mask; + struct registered_dpif_class *registered_class; + + if (shash_find(&dpif_classes, new_class->type)) { + VLOG_WARN("attempted to register duplicate datapath provider: %s", + new_class->type); + return EEXIST; + } + + registered_class = xmalloc(sizeof *registered_class); + registered_class->dpif_class = new_class; + registered_class->refcount = 0; + + shash_add(&dpif_classes, new_class->type, registered_class); + + return 0; +} + +/* Unregisters a datapath provider. 'type' must have been previously + * registered and not currently be in use by any dpifs. After unregistration + * new datapaths of that type cannot be opened using dpif_open(). */ +int +dp_unregister_provider(const char *type) +{ + struct shash_node *node; + struct registered_dpif_class *registered_class; + + node = shash_find(&dpif_classes, type); + if (!node) { + VLOG_WARN("attempted to unregister a datapath provider that is not " + "registered: %s", type); + return EAFNOSUPPORT; + } + + registered_class = node->data; + if (registered_class->refcount) { + VLOG_WARN("attempted to unregister in use datapath provider: %s", type); + return EBUSY; + } + + shash_delete(&dpif_classes, node); + free(registered_class); + + return 0; +} + +/* Clears 'types' and enumerates the types of all currently registered datapath + * providers into it. The caller must first initialize the sset. */ +void +dp_enumerate_types(struct sset *types) +{ + struct shash_node *node; + + dp_initialize(); + sset_clear(types); + + SHASH_FOR_EACH(node, &dpif_classes) { + const struct registered_dpif_class *registered_class = node->data; + sset_add(types, registered_class->dpif_class->type); + } +} + +/* Clears 'names' and enumerates the names of all known created datapaths with + * the given 'type'. The caller must first initialize the sset. Returns 0 if + * successful, otherwise a positive errno value. + * + * Some kinds of datapaths might not be practically enumerable. This is not + * considered an error. */ +int +dp_enumerate_names(const char *type, struct sset *names) +{ + const struct registered_dpif_class *registered_class; + const struct dpif_class *dpif_class; int error; - dpif->fd = -1; + dp_initialize(); + sset_clear(names); - error = name_to_minor(name, &dpif->minor); - if (error) { - return error; + registered_class = shash_find_data(&dpif_classes, type); + if (!registered_class) { + VLOG_WARN("could not enumerate unknown type: %s", type); + return EAFNOSUPPORT; } - error = open_by_minor(dpif->minor, dpif); + dpif_class = registered_class->dpif_class; + error = dpif_class->enumerate ? dpif_class->enumerate(names) : 0; + if (error) { - return error; + VLOG_WARN("failed to enumerate %s datapaths: %s", dpif_class->type, + strerror(error)); } - /* We can open the device, but that doesn't mean that it's been created. - * If it hasn't been, then any command other than ODP_DP_CREATE will - * return ENODEV. Try something innocuous. */ - listen_mask = 0; /* Make Valgrind happy. */ - if (ioctl(dpif->fd, ODP_GET_LISTEN_MASK, &listen_mask)) { - error = errno; - if (error != ENODEV) { - VLOG_WARN("dp%u: probe returned unexpected error: %s", - dpif->minor, strerror(error)); - } - dpif_close(dpif); - return error; - } - return 0; + return error; } +/* Parses 'datapath_name_', which is of the form [type@]name into its + * component pieces. 'name' and 'type' must be freed by the caller. + * + * The returned 'type' is normalized, as if by dpif_normalize_type(). */ void -dpif_close(struct dpif *dpif) +dp_parse_name(const char *datapath_name_, char **name, char **type) { - if (dpif) { - close(dpif->fd); - dpif->fd = -1; + char *datapath_name = xstrdup(datapath_name_); + char *separator; + + separator = strchr(datapath_name, '@'); + if (separator) { + *separator = '\0'; + *type = datapath_name; + *name = xstrdup(dpif_normalize_type(separator + 1)); + } else { + *name = datapath_name; + *type = xstrdup(dpif_normalize_type(NULL)); } } static int -do_ioctl(const struct dpif *dpif, int cmd, const char *cmd_name, - const void *arg) +do_open(const char *name, const char *type, bool create, struct dpif **dpifp) { - int error = ioctl(dpif->fd, cmd, arg) ? errno : 0; - if (cmd_name) { - if (error) { - VLOG_WARN_RL(&error_rl, "dp%u: ioctl(%s) failed (%s)", - dpif->minor, cmd_name, strerror(error)); - } else { - VLOG_DBG_RL(&dpmsg_rl, "dp%u: ioctl(%s): success", - dpif->minor, cmd_name); - } + struct dpif *dpif = NULL; + int error; + struct registered_dpif_class *registered_class; + + dp_initialize(); + + type = dpif_normalize_type(type); + + registered_class = shash_find_data(&dpif_classes, type); + if (!registered_class) { + VLOG_WARN("could not create datapath %s of unknown type %s", name, + type); + error = EAFNOSUPPORT; + goto exit; } + + error = registered_class->dpif_class->open(registered_class->dpif_class, + name, create, &dpif); + if (!error) { + assert(dpif->dpif_class == registered_class->dpif_class); + registered_class->refcount++; + } + +exit: + *dpifp = error ? NULL : dpif; return error; } +/* Tries to open an existing datapath named 'name' and type 'type'. Will fail + * if no datapath with 'name' and 'type' exists. 'type' may be either NULL or + * the empty string to specify the default system type. Returns 0 if + * successful, otherwise a positive errno value. On success stores a pointer + * to the datapath in '*dpifp', otherwise a null pointer. */ +int +dpif_open(const char *name, const char *type, struct dpif **dpifp) +{ + return do_open(name, type, false, dpifp); +} + +/* Tries to create and open a new datapath with the given 'name' and 'type'. + * 'type' may be either NULL or the empty string to specify the default system + * type. Will fail if a datapath with 'name' and 'type' already exists. + * Returns 0 if successful, otherwise a positive errno value. On success + * stores a pointer to the datapath in '*dpifp', otherwise a null pointer. */ int -dpif_create(const char *name, struct dpif *dpif) +dpif_create(const char *name, const char *type, struct dpif **dpifp) +{ + return do_open(name, type, true, dpifp); +} + +/* Tries to open a datapath with the given 'name' and 'type', creating it if it + * does not exist. 'type' may be either NULL or the empty string to specify + * the default system type. Returns 0 if successful, otherwise a positive + * errno value. On success stores a pointer to the datapath in '*dpifp', + * otherwise a null pointer. */ +int +dpif_create_and_open(const char *name, const char *type, struct dpif **dpifp) { - unsigned int minor; int error; - if (!get_minor_from_name(name, &minor)) { - /* Minor was specified in 'name', go ahead and create it. */ - error = open_by_minor(minor, dpif); + error = dpif_create(name, type, dpifp); + if (error == EEXIST || error == EBUSY) { + error = dpif_open(name, type, dpifp); if (error) { - return error; + VLOG_WARN("datapath %s already exists but cannot be opened: %s", + name, strerror(error)); } + } else if (error) { + VLOG_WARN("failed to create datapath %s: %s", name, strerror(error)); + } + return error; +} - if (!strncmp(name, "nl:", 3)) { - char devname[128]; - sprintf(devname, "of%u", minor); - error = ioctl(dpif->fd, ODP_DP_CREATE, devname) < 0 ? errno : 0; - } else { - error = ioctl(dpif->fd, ODP_DP_CREATE, name) < 0 ? errno : 0; - } - if (error) { - dpif_close(dpif); - } - return error; - } else { - for (minor = 0; minor < ODP_MAX; minor++) { - error = open_by_minor(minor, dpif); - if (error) { - return error; - } - - error = ioctl(dpif->fd, ODP_DP_CREATE, name) < 0 ? errno : 0; - if (!error) { - return 0; - } - dpif_close(dpif); - if (error != EBUSY) { - return error; - } - } - return ENOBUFS; +/* Closes and frees the connection to 'dpif'. Does not destroy the datapath + * itself; call dpif_delete() first, instead, if that is desirable. */ +void +dpif_close(struct dpif *dpif) +{ + if (dpif) { + struct registered_dpif_class *registered_class; + + registered_class = shash_find_data(&dpif_classes, + dpif->dpif_class->type); + assert(registered_class); + assert(registered_class->refcount); + + registered_class->refcount--; + dpif_uninit(dpif, true); } } -int -dpif_delete(struct dpif *dpif) +/* Performs periodic work needed by 'dpif'. */ +void +dpif_run(struct dpif *dpif) { - COVERAGE_INC(dpif_destroy); - return do_ioctl(dpif, ODP_DP_DESTROY, "ODP_DP_DESTROY", NULL); + if (dpif->dpif_class->run) { + dpif->dpif_class->run(dpif); + } } -int -dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats) +/* Arranges for poll_block() to wake up when dp_run() needs to be called for + * 'dpif'. */ +void +dpif_wait(struct dpif *dpif) { - memset(stats, 0, sizeof *stats); - return do_ioctl(dpif, ODP_DP_STATS, "ODP_DP_STATS", stats); + if (dpif->dpif_class->wait) { + dpif->dpif_class->wait(dpif); + } +} + +/* Returns the name of datapath 'dpif' prefixed with the type + * (for use in log messages). */ +const char * +dpif_name(const struct dpif *dpif) +{ + return dpif->full_name; +} + +/* Returns the name of datapath 'dpif' without the type + * (for use in device names). */ +const char * +dpif_base_name(const struct dpif *dpif) +{ + return dpif->base_name; } +/* Returns the fully spelled out name for the given datapath 'type'. + * + * Normalized type string can be compared with strcmp(). Unnormalized type + * string might be the same even if they have different spellings. */ +const char * +dpif_normalize_type(const char *type) +{ + return type && type[0] ? type : "system"; +} + +/* Destroys the datapath that 'dpif' is connected to, first removing all of its + * ports. After calling this function, it does not make sense to pass 'dpif' + * to any functions other than dpif_name() or dpif_close(). */ int -dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags) +dpif_delete(struct dpif *dpif) { - int tmp; - int error = do_ioctl(dpif, ODP_GET_DROP_FRAGS, "ODP_GET_DROP_FRAGS", &tmp); - *drop_frags = error ? tmp & 1 : false; + int error; + + COVERAGE_INC(dpif_destroy); + + error = dpif->dpif_class->destroy(dpif); + log_operation(dpif, "delete", error); return error; } +/* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful, + * otherwise a positive errno value. */ int -dpif_set_drop_frags(struct dpif *dpif, bool drop_frags) +dpif_get_dp_stats(const struct dpif *dpif, struct ovs_dp_stats *stats) { - int tmp = drop_frags; - return do_ioctl(dpif, ODP_SET_DROP_FRAGS, "ODP_SET_DROP_FRAGS", &tmp); + int error = dpif->dpif_class->get_stats(dpif, stats); + if (error) { + memset(stats, 0, sizeof *stats); + } + log_operation(dpif, "get_stats", error); + return error; } +/* Retrieves the current IP fragment handling policy for 'dpif' 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). Returns 0 if successful, otherwise a + * positive errno value. */ int -dpif_get_listen_mask(const struct dpif *dpif, int *listen_mask) +dpif_get_drop_frags(const struct dpif *dpif, bool *drop_frags) { - int error = do_ioctl(dpif, ODP_GET_LISTEN_MASK, "ODP_GET_LISTEN_MASK", - listen_mask); + int error = dpif->dpif_class->get_drop_frags(dpif, drop_frags); if (error) { - *listen_mask = 0; + *drop_frags = false; } + log_operation(dpif, "get_drop_frags", error); return error; } +/* Changes 'dpif''s treatment of IP fragments to 'drop_frags', whose meaning is + * the same as for the get_drop_frags member function. Returns 0 if + * successful, otherwise a positive errno value. */ int -dpif_set_listen_mask(struct dpif *dpif, int listen_mask) +dpif_set_drop_frags(struct dpif *dpif, bool drop_frags) { - return do_ioctl(dpif, ODP_SET_LISTEN_MASK, "ODP_SET_LISTEN_MASK", - &listen_mask); + int error = dpif->dpif_class->set_drop_frags(dpif, drop_frags); + log_operation(dpif, "set_drop_frags", error); + return error; } +/* Attempts to add 'netdev' as a port on 'dpif'. If successful, returns 0 and + * sets '*port_nop' to the new port's port number (if 'port_nop' is non-null). + * On failure, returns a positive errno value and sets '*port_nop' to + * UINT16_MAX (if 'port_nop' is non-null). */ int -dpif_purge(struct dpif *dpif) +dpif_port_add(struct dpif *dpif, struct netdev *netdev, uint16_t *port_nop) { - struct odp_stats stats; - unsigned int i; + const char *netdev_name = netdev_get_name(netdev); + uint16_t port_no; int error; - COVERAGE_INC(dpif_purge); + COVERAGE_INC(dpif_port_add); - error = dpif_get_dp_stats(dpif, &stats); - if (error) { - return error; + error = dpif->dpif_class->port_add(dpif, netdev, &port_no); + if (!error) { + VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16, + dpif_name(dpif), netdev_name, port_no); + } else { + VLOG_WARN_RL(&error_rl, "%s: failed to add %s as port: %s", + dpif_name(dpif), netdev_name, strerror(error)); + port_no = UINT16_MAX; } - - for (i = 0; i < stats.max_miss_queue + stats.max_action_queue; i++) { - struct ofpbuf *buf; - error = dpif_recv(dpif, &buf); - if (error) { - return error == EAGAIN ? 0 : error; - } - ofpbuf_delete(buf); + if (port_nop) { + *port_nop = port_no; } - return 0; + return error; } +/* Attempts to remove 'dpif''s port number 'port_no'. Returns 0 if successful, + * otherwise a positive errno value. */ int -dpif_port_add(struct dpif *dpif, const char *devname, uint16_t port_no, - uint16_t flags) +dpif_port_del(struct dpif *dpif, uint16_t port_no) { - struct odp_port port; + int error; - COVERAGE_INC(dpif_port_add); - memset(&port, 0, sizeof port); - strncpy(port.devname, devname, sizeof port.devname); - port.port = port_no; - port.flags = flags; - if (!ioctl(dpif->fd, ODP_PORT_ADD, &port)) { - VLOG_DBG_RL(&dpmsg_rl, "dp%u: added %s as port %"PRIu16, - dpif->minor, devname, port_no); - return 0; + COVERAGE_INC(dpif_port_del); + + error = dpif->dpif_class->port_del(dpif, port_no); + if (!error) { + VLOG_DBG_RL(&dpmsg_rl, "%s: port_del(%"PRIu16")", + dpif_name(dpif), port_no); } else { - VLOG_WARN_RL(&error_rl, "dp%u: failed to add %s as port " - "%"PRIu16": %s", dpif->minor, devname, port_no, - strerror(errno)); - return errno; + log_operation(dpif, "port_del", error); } + return error; } -int -dpif_port_del(struct dpif *dpif, uint16_t port_no) +/* Makes a deep copy of 'src' into 'dst'. */ +void +dpif_port_clone(struct dpif_port *dst, const struct dpif_port *src) { - int tmp = port_no; - COVERAGE_INC(dpif_port_del); - return do_ioctl(dpif, ODP_PORT_DEL, "ODP_PORT_DEL", &tmp); + dst->name = xstrdup(src->name); + dst->type = xstrdup(src->type); + dst->port_no = src->port_no; } +/* Frees memory allocated to members of 'dpif_port'. + * + * Do not call this function on a dpif_port obtained from + * dpif_port_dump_next(): that function retains ownership of the data in the + * dpif_port. */ +void +dpif_port_destroy(struct dpif_port *dpif_port) +{ + free(dpif_port->name); + free(dpif_port->type); +} + +/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and + * initializes '*port' appropriately; on failure, returns a positive errno + * value. + * + * The caller owns the data in 'port' and must free it with + * dpif_port_destroy() when it is no longer needed. */ int dpif_port_query_by_number(const struct dpif *dpif, uint16_t port_no, - struct odp_port *port) + struct dpif_port *port) { - memset(port, 0, sizeof *port); - port->port = port_no; - if (!ioctl(dpif->fd, ODP_PORT_QUERY, port)) { - VLOG_DBG_RL(&dpmsg_rl, "dp%u: port %"PRIu16" is device %s", - dpif->minor, port_no, port->devname); - return 0; + int error = dpif->dpif_class->port_query_by_number(dpif, port_no, port); + if (!error) { + VLOG_DBG_RL(&dpmsg_rl, "%s: port %"PRIu16" is device %s", + dpif_name(dpif), port_no, port->name); } else { - VLOG_WARN_RL(&error_rl, "dp%u: failed to query port %"PRIu16": %s", - dpif->minor, port_no, strerror(errno)); - return errno; + memset(port, 0, sizeof *port); + VLOG_WARN_RL(&error_rl, "%s: failed to query port %"PRIu16": %s", + dpif_name(dpif), port_no, strerror(error)); } + return error; } +/* Looks up port named 'devname' in 'dpif'. On success, returns 0 and + * initializes '*port' appropriately; on failure, returns a positive errno + * value. + * + * The caller owns the data in 'port' and must free it with + * dpif_port_destroy() when it is no longer needed. */ int dpif_port_query_by_name(const struct dpif *dpif, const char *devname, - struct odp_port *port) + struct dpif_port *port) { - memset(port, 0, sizeof *port); - strncpy(port->devname, devname, sizeof port->devname); - if (!ioctl(dpif->fd, ODP_PORT_QUERY, port)) { - VLOG_DBG_RL(&dpmsg_rl, "dp%u: device %s is on port %"PRIu16, - dpif->minor, devname, port->port); - return 0; + int error = dpif->dpif_class->port_query_by_name(dpif, devname, port); + if (!error) { + VLOG_DBG_RL(&dpmsg_rl, "%s: device %s is on port %"PRIu16, + dpif_name(dpif), devname, port->port_no); } else { - VLOG_WARN_RL(&error_rl, "dp%u: failed to query port %s: %s", - dpif->minor, devname, strerror(errno)); - return errno; + memset(port, 0, sizeof *port); + + /* For ENOENT or ENODEV we use DBG level because the caller is probably + * interested in whether 'dpif' actually has a port 'devname', so that + * it's not an issue worth logging if it doesn't. Other errors are + * uncommon and more likely to indicate a real problem. */ + VLOG_RL(&error_rl, + error == ENOENT || error == ENODEV ? VLL_DBG : VLL_WARN, + "%s: failed to query port %s: %s", + dpif_name(dpif), devname, strerror(error)); } + return error; } +/* Returns one greater than the maximum port number accepted in flow + * actions. */ +int +dpif_get_max_ports(const struct dpif *dpif) +{ + return dpif->dpif_class->get_max_ports(dpif); +} + +/* Looks up port number 'port_no' in 'dpif'. On success, returns 0 and copies + * the port's name into the 'name_size' bytes in 'name', ensuring that the + * result is null-terminated. On failure, returns a positive errno value and + * makes 'name' the empty string. */ int dpif_port_get_name(struct dpif *dpif, uint16_t port_no, char *name, size_t name_size) { - struct odp_port port; + struct dpif_port port; int error; assert(name_size > 0); error = dpif_port_query_by_number(dpif, port_no, &port); if (!error) { - ovs_strlcpy(name, port.devname, name_size); + ovs_strlcpy(name, port.name, name_size); + dpif_port_destroy(&port); } else { *name = '\0'; } return error; } -int -dpif_port_list(const struct dpif *dpif, - struct odp_port **ports, size_t *n_ports) +/* Initializes 'dump' to begin dumping the ports in a dpif. + * + * This function provides no status indication. An error status for the entire + * dump operation is provided when it is completed by calling + * dpif_port_dump_done(). + */ +void +dpif_port_dump_start(struct dpif_port_dump *dump, const struct dpif *dpif) { - struct odp_portvec pv; - struct odp_stats stats; - int error; + dump->dpif = dpif; + dump->error = dpif->dpif_class->port_dump_start(dpif, &dump->state); + log_operation(dpif, "port_dump_start", dump->error); +} - do { - error = dpif_get_dp_stats(dpif, &stats); - if (error) { - goto error; - } +/* Attempts to retrieve another port from 'dump', which must have been + * initialized with dpif_port_dump_start(). On success, stores a new dpif_port + * into 'port' and returns true. On failure, returns false. + * + * Failure might indicate an actual error or merely that the last port has been + * dumped. An error status for the entire dump operation is provided when it + * is completed by calling dpif_port_dump_done(). + * + * The dpif owns the data stored in 'port'. It will remain valid until at + * least the next time 'dump' is passed to dpif_port_dump_next() or + * dpif_port_dump_done(). */ +bool +dpif_port_dump_next(struct dpif_port_dump *dump, struct dpif_port *port) +{ + const struct dpif *dpif = dump->dpif; - *ports = xcalloc(1, stats.n_ports * sizeof **ports); - pv.ports = *ports; - pv.n_ports = stats.n_ports; - error = do_ioctl(dpif, ODP_PORT_LIST, "ODP_PORT_LIST", &pv); - if (error) { - free(*ports); - goto error; - } - } while (pv.n_ports != stats.n_ports); - *n_ports = pv.n_ports; - return 0; + if (dump->error) { + return false; + } -error: - *ports = NULL; - *n_ports = 0; - return error; + dump->error = dpif->dpif_class->port_dump_next(dpif, dump->state, port); + if (dump->error == EOF) { + VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all ports", dpif_name(dpif)); + } else { + log_operation(dpif, "port_dump_next", dump->error); + } + + if (dump->error) { + dpif->dpif_class->port_dump_done(dpif, dump->state); + return false; + } + return true; } +/* Completes port table dump operation 'dump', which must have been initialized + * with dpif_port_dump_start(). Returns 0 if the dump operation was + * error-free, otherwise a positive errno value describing the problem. */ int -dpif_port_group_set(struct dpif *dpif, uint16_t group, - const uint16_t ports[], size_t n_ports) +dpif_port_dump_done(struct dpif_port_dump *dump) { - struct odp_port_group pg; - - COVERAGE_INC(dpif_port_group_set); - assert(n_ports <= UINT16_MAX); - pg.group = group; - pg.ports = (uint16_t *) ports; - pg.n_ports = n_ports; - return do_ioctl(dpif, ODP_PORT_GROUP_SET, "ODP_PORT_GROUP_SET", &pg); + const struct dpif *dpif = dump->dpif; + if (!dump->error) { + dump->error = dpif->dpif_class->port_dump_done(dpif, dump->state); + log_operation(dpif, "port_dump_done", dump->error); + } + return dump->error == EOF ? 0 : dump->error; } -/* Careful: '*n_out' can be greater than 'n_ports' on return, if 'n_ports' is - * less than the number of ports in 'group'. */ +/* Polls for changes in the set of ports in 'dpif'. If the set of ports in + * 'dpif' has changed, this function does one of the following: + * + * - Stores the name of the device that was added to or deleted from 'dpif' in + * '*devnamep' and returns 0. The caller is responsible for freeing + * '*devnamep' (with free()) when it no longer needs it. + * + * - Returns ENOBUFS and sets '*devnamep' to NULL. + * + * This function may also return 'false positives', where it returns 0 and + * '*devnamep' names a device that was not actually added or deleted or it + * returns ENOBUFS without any change. + * + * Returns EAGAIN if the set of ports in 'dpif' has not changed. May also + * return other positive errno values to indicate that something has gone + * wrong. */ int -dpif_port_group_get(const struct dpif *dpif, uint16_t group, - uint16_t ports[], size_t n_ports, size_t *n_out) +dpif_port_poll(const struct dpif *dpif, char **devnamep) { - struct odp_port_group pg; - int error; - - assert(n_ports <= UINT16_MAX); - pg.group = group; - pg.ports = ports; - pg.n_ports = n_ports; - error = do_ioctl(dpif, ODP_PORT_GROUP_GET, "ODP_PORT_GROUP_GET", &pg); - *n_out = error ? 0 : pg.n_ports; + int error = dpif->dpif_class->port_poll(dpif, devnamep); + if (error) { + *devnamep = NULL; + } return error; } -int -dpif_flow_flush(struct dpif *dpif) +/* Arranges for the poll loop to wake up when port_poll(dpif) will return a + * value other than EAGAIN. */ +void +dpif_port_poll_wait(const struct dpif *dpif) { - COVERAGE_INC(dpif_flow_flush); - return do_ioctl(dpif, ODP_FLOW_FLUSH, "ODP_FLOW_FLUSH", NULL); + dpif->dpif_class->port_poll_wait(dpif); } -static enum vlog_level -flow_message_log_level(int error) +/* Appends a human-readable representation of 'stats' to 's'. */ +void +dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s) { - return error ? VLL_WARN : VLL_DBG; + ds_put_format(s, "packets:%"PRIu64", bytes:%"PRIu64", used:", + stats->n_packets, stats->n_bytes); + if (stats->used) { + ds_put_format(s, "%.3fs", (time_msec() - stats->used) / 1000.0); + } else { + ds_put_format(s, "never"); + } + /* XXX tcp_flags? */ } -static bool -should_log_flow_message(int error) +/* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a + * positive errno value. */ +int +dpif_flow_flush(struct dpif *dpif) { - return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error), - error ? &error_rl : &dpmsg_rl); -} + int error; -static void -log_flow_message(const struct dpif *dpif, int error, - const char *operation, - const flow_t *flow, const struct odp_flow_stats *stats, - const union odp_action *actions, size_t n_actions) -{ - struct ds ds = DS_EMPTY_INITIALIZER; - ds_put_format(&ds, "dp%u: ", dpif->minor); - if (error) { - ds_put_cstr(&ds, "failed to "); - } - ds_put_format(&ds, "%s ", operation); - if (error) { - ds_put_format(&ds, "(%s) ", strerror(error)); - } - flow_format(&ds, flow); - if (stats) { - ds_put_cstr(&ds, ", "); - format_odp_flow_stats(&ds, stats); - } - if (actions || n_actions) { - ds_put_cstr(&ds, ", actions:"); - format_odp_actions(&ds, actions, n_actions); - } - vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds)); - ds_destroy(&ds); + COVERAGE_INC(dpif_flow_flush); + + error = dpif->dpif_class->flow_flush(dpif); + log_operation(dpif, "flow_flush", error); + return error; } -static int -do_flow_ioctl(const struct dpif *dpif, int cmd, struct odp_flow *flow, - const char *operation, bool show_stats) +/* 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' will 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 will be updated with the flow's + * statistics. */ +int +dpif_flow_get(const struct dpif *dpif, + const struct nlattr *key, size_t key_len, + struct ofpbuf **actionsp, struct dpif_flow_stats *stats) { - int error = do_ioctl(dpif, cmd, NULL, flow); - if (error && show_stats) { - flow->n_actions = 0; + int error; + + COVERAGE_INC(dpif_flow_get); + + error = dpif->dpif_class->flow_get(dpif, key, key_len, actionsp, stats); + if (error) { + if (actionsp) { + *actionsp = NULL; + } + if (stats) { + memset(stats, 0, sizeof *stats); + } } if (should_log_flow_message(error)) { - log_flow_message(dpif, error, operation, &flow->key, - show_stats && !error ? &flow->stats : NULL, - flow->actions, flow->n_actions); + const struct nlattr *actions; + size_t actions_len; + + if (!error && actionsp) { + actions = (*actionsp)->data; + actions_len = (*actionsp)->size; + } else { + actions = NULL; + actions_len = 0; + } + log_flow_message(dpif, error, "flow_get", key, key_len, stats, + actions, actions_len); } return error; } +/* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink + * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at + * 'key'. The associated actions are specified by the Netlink attributes with + * types OVS_ACTION_ATTR_* in the 'actions_len' bytes starting at 'actions'. + * + * - If the flow's key does not exist in 'dpif', then the flow will be added if + * 'flags' includes DPIF_FP_CREATE. Otherwise the operation will fail with + * ENOENT. + * + * If the operation succeeds, then 'stats', if nonnull, will be zeroed. + * + * - If the flow's key does exist in 'dpif', then the flow's actions will be + * updated if '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 'flags' includes DPIF_FP_ZERO_STATS, and left as-is + * otherwise. + * + * If the operation succeeds, then 'stats', if nonnull, will be set to the + * flow's statistics before the update. + */ int -dpif_flow_put(struct dpif *dpif, struct odp_flow_put *put) +dpif_flow_put(struct dpif *dpif, enum dpif_flow_put_flags flags, + const struct nlattr *key, size_t key_len, + const struct nlattr *actions, size_t actions_len, + struct dpif_flow_stats *stats) { - int error = do_ioctl(dpif, ODP_FLOW_PUT, NULL, put); + int error; + COVERAGE_INC(dpif_flow_put); + assert(!(flags & ~(DPIF_FP_CREATE | DPIF_FP_MODIFY | DPIF_FP_ZERO_STATS))); + + error = dpif->dpif_class->flow_put(dpif, flags, key, key_len, + actions, actions_len, stats); + if (error && stats) { + memset(stats, 0, sizeof *stats); + } if (should_log_flow_message(error)) { - struct ds operation = DS_EMPTY_INITIALIZER; - ds_put_cstr(&operation, "put"); - if (put->flags & ODPPF_CREATE) { - ds_put_cstr(&operation, "[create]"); - } - if (put->flags & ODPPF_MODIFY) { - ds_put_cstr(&operation, "[modify]"); + struct ds s; + + ds_init(&s); + ds_put_cstr(&s, "put"); + if (flags & DPIF_FP_CREATE) { + ds_put_cstr(&s, "[create]"); } - if (put->flags & ODPPF_ZERO_STATS) { - ds_put_cstr(&operation, "[zero]"); + if (flags & DPIF_FP_MODIFY) { + ds_put_cstr(&s, "[modify]"); } -#define ODPPF_ALL (ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS) - if (put->flags & ~ODPPF_ALL) { - ds_put_format(&operation, "[%x]", put->flags & ~ODPPF_ALL); + if (flags & DPIF_FP_ZERO_STATS) { + ds_put_cstr(&s, "[zero]"); } - log_flow_message(dpif, error, ds_cstr(&operation), &put->flow.key, - !error ? &put->flow.stats : NULL, - put->flow.actions, put->flow.n_actions); - ds_destroy(&operation); + log_flow_message(dpif, error, ds_cstr(&s), key, key_len, stats, + actions, actions_len); + ds_destroy(&s); } return error; } +/* 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 'key_len' bytes starting at 'key'. + * + * If the operation succeeds, then 'stats', if nonnull, will be set to the + * flow's statistics before its deletion. */ int -dpif_flow_del(struct dpif *dpif, struct odp_flow *flow) +dpif_flow_del(struct dpif *dpif, + const struct nlattr *key, size_t key_len, + struct dpif_flow_stats *stats) { + int error; + COVERAGE_INC(dpif_flow_del); - check_rw_odp_flow(flow); - memset(&flow->stats, 0, sizeof flow->stats); - return do_flow_ioctl(dpif, ODP_FLOW_DEL, flow, "delete flow", true); -} -int -dpif_flow_get(const struct dpif *dpif, struct odp_flow *flow) -{ - COVERAGE_INC(dpif_flow_query); - check_rw_odp_flow(flow); - memset(&flow->stats, 0, sizeof flow->stats); - return do_flow_ioctl(dpif, ODP_FLOW_GET, flow, "get flow", true); + error = dpif->dpif_class->flow_del(dpif, key, key_len, stats); + if (error && stats) { + memset(stats, 0, sizeof *stats); + } + if (should_log_flow_message(error)) { + log_flow_message(dpif, error, "flow_del", key, key_len, + !error ? stats : NULL, NULL, 0); + } + return error; } -int -dpif_flow_get_multiple(const struct dpif *dpif, - struct odp_flow flows[], size_t n) +/* Initializes 'dump' to begin dumping the flows in a dpif. + * + * This function provides no status indication. An error status for the entire + * dump operation is provided when it is completed by calling + * dpif_flow_dump_done(). + */ +void +dpif_flow_dump_start(struct dpif_flow_dump *dump, const struct dpif *dpif) { - struct odp_flowvec fv; - size_t i; - - COVERAGE_ADD(dpif_flow_query_multiple, n); - fv.flows = flows; - fv.n_flows = n; - for (i = 0; i < n; i++) { - check_rw_odp_flow(&flows[i]); - } - return do_ioctl(dpif, ODP_FLOW_GET_MULTIPLE, "ODP_FLOW_GET_MULTIPLE", - &fv); + dump->dpif = dpif; + dump->error = dpif->dpif_class->flow_dump_start(dpif, &dump->state); + log_operation(dpif, "flow_dump_start", dump->error); } -int -dpif_flow_list(const struct dpif *dpif, struct odp_flow flows[], size_t n, - size_t *n_out) +/* Attempts to retrieve another flow from 'dump', which must have been + * initialized with dpif_flow_dump_start(). On success, updates the output + * parameters as described below and returns true. Otherwise, returns false. + * Failure might indicate an actual error or merely the end of the flow table. + * An error status for the entire dump operation is provided when it is + * completed by calling dpif_flow_dump_done(). + * + * On success, if 'key' and 'key_len' are nonnull then '*key' and '*key_len' + * will 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 are + * set to Netlink attributes with types OVS_ACTION_ATTR_* representing the + * dumped flow's actions. If 'stats' is nonnull then it will 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' guarantees that it remains + * accessible and unchanging until at least the next call to 'flow_dump_next' + * or 'flow_dump_done' for 'dump'. */ +bool +dpif_flow_dump_next(struct dpif_flow_dump *dump, + const struct nlattr **key, size_t *key_len, + const struct nlattr **actions, size_t *actions_len, + const struct dpif_flow_stats **stats) { - struct odp_flowvec fv; - uint32_t i; - int error; + const struct dpif *dpif = dump->dpif; + int error = dump->error; - COVERAGE_INC(dpif_flow_query_list); - fv.flows = flows; - fv.n_flows = n; - if (RUNNING_ON_VALGRIND) { - memset(flows, 0, n * sizeof *flows); - } else { - for (i = 0; i < n; i++) { - flows[i].actions = NULL; - flows[i].n_actions = 0; + if (!error) { + error = dpif->dpif_class->flow_dump_next(dpif, dump->state, + key, key_len, + actions, actions_len, + stats); + if (error) { + dpif->dpif_class->flow_dump_done(dpif, dump->state); } } - error = do_ioctl(dpif, ODP_FLOW_LIST, NULL, &fv); if (error) { - *n_out = 0; - VLOG_WARN_RL(&error_rl, "dp%u: flow list failed (%s)", - dpif->minor, strerror(error)); - } else { - COVERAGE_ADD(dpif_flow_query_list_n, fv.n_flows); - *n_out = fv.n_flows; - VLOG_DBG_RL(&dpmsg_rl, "dp%u: listed %zu flows", dpif->minor, *n_out); + if (key) { + *key = NULL; + *key_len = 0; + } + if (actions) { + *actions = NULL; + *actions_len = 0; + } + if (stats) { + *stats = NULL; + } } - return error; + if (!dump->error) { + if (error == EOF) { + VLOG_DBG_RL(&dpmsg_rl, "%s: dumped all flows", dpif_name(dpif)); + } else if (should_log_flow_message(error)) { + log_flow_message(dpif, error, "flow_dump", + key ? *key : NULL, key ? *key_len : 0, + stats ? *stats : NULL, actions ? *actions : NULL, + actions ? *actions_len : 0); + } + } + dump->error = error; + return !error; } +/* Completes flow table dump operation 'dump', which must have been initialized + * with dpif_flow_dump_start(). Returns 0 if the dump operation was + * error-free, otherwise a positive errno value describing the problem. */ int -dpif_flow_list_all(const struct dpif *dpif, - struct odp_flow **flowsp, size_t *np) +dpif_flow_dump_done(struct dpif_flow_dump *dump) { - struct odp_stats stats; - struct odp_flow *flows; - size_t n_flows; - int error; - - *flowsp = NULL; - *np = 0; - - error = dpif_get_dp_stats(dpif, &stats); - if (error) { - return error; - } - - flows = xmalloc(sizeof *flows * stats.n_flows); - error = dpif_flow_list(dpif, flows, stats.n_flows, &n_flows); - if (error) { - free(flows); - return error; + const struct dpif *dpif = dump->dpif; + if (!dump->error) { + dump->error = dpif->dpif_class->flow_dump_done(dpif, dump->state); + log_operation(dpif, "flow_dump_done", dump->error); } - - if (stats.n_flows != n_flows) { - VLOG_WARN_RL(&error_rl, "dp%u: datapath stats reported %"PRIu32" " - "flows but flow listing reported %zu", - dpif->minor, stats.n_flows, n_flows); - } - *flowsp = flows; - *np = n_flows; - return 0; + return dump->error == EOF ? 0 : dump->error; } +/* Causes 'dpif' to perform the 'actions_len' bytes of actions in 'actions' on + * the Ethernet frame specified in 'packet' taken from the flow specified in + * the 'key_len' bytes of 'key'. ('key' is mostly redundant with 'packet', but + * it contains some metadata that cannot be recovered from 'packet', such as + * tun_id and in_port.) + * + * Returns 0 if successful, otherwise a positive errno value. */ int -dpif_execute(struct dpif *dpif, uint16_t in_port, - const union odp_action actions[], size_t n_actions, +dpif_execute(struct dpif *dpif, + const struct nlattr *key, size_t key_len, + const struct nlattr *actions, size_t actions_len, const struct ofpbuf *buf) { int error; COVERAGE_INC(dpif_execute); - if (n_actions > 0) { - struct odp_execute execute; - memset(&execute, 0, sizeof execute); - execute.in_port = in_port; - execute.actions = (union odp_action *) actions; - execute.n_actions = n_actions; - execute.data = buf->data; - execute.length = buf->size; - error = do_ioctl(dpif, ODP_EXECUTE, NULL, &execute); + if (actions_len > 0) { + error = dpif->dpif_class->execute(dpif, key, key_len, + actions, actions_len, buf); } else { error = 0; } @@ -606,8 +938,8 @@ dpif_execute(struct dpif *dpif, uint16_t in_port, if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) { struct ds ds = DS_EMPTY_INITIALIZER; char *packet = ofp_packet_to_string(buf->data, buf->size, buf->size); - ds_put_format(&ds, "dp%u: execute ", dpif->minor); - format_odp_actions(&ds, actions, n_actions); + ds_put_format(&ds, "%s: execute ", dpif_name(dpif)); + format_odp_actions(&ds, actions, actions_len); if (error) { ds_put_format(&ds, " failed (%s)", strerror(error)); } @@ -619,445 +951,263 @@ dpif_execute(struct dpif *dpif, uint16_t in_port, return error; } -int -dpif_recv(struct dpif *dpif, struct ofpbuf **bufp) +/* Returns a string that represents 'type', for use in log messages. */ +const char * +dpif_upcall_type_to_string(enum dpif_upcall_type type) { - struct ofpbuf *buf; - int retval; - int error; - - buf = ofpbuf_new(65536); - retval = read(dpif->fd, ofpbuf_tail(buf), ofpbuf_tailroom(buf)); - if (retval < 0) { - error = errno; - if (error != EAGAIN) { - VLOG_WARN_RL(&error_rl, "dp%u: read failed: %s", - dpif->minor, strerror(error)); - } - } else if (retval >= sizeof(struct odp_msg)) { - struct odp_msg *msg = buf->data; - if (msg->length <= retval) { - buf->size += retval; - if (VLOG_IS_DBG_ENABLED()) { - void *payload = msg + 1; - size_t length = buf->size - sizeof *msg; - char *s = ofp_packet_to_string(payload, length, length); - VLOG_DBG_RL(&dpmsg_rl, "dp%u: received %s message of length " - "%zu on port %"PRIu16": %s", dpif->minor, - (msg->type == _ODPL_MISS_NR ? "miss" - : msg->type == _ODPL_ACTION_NR ? "action" - : ""), - msg->length - sizeof(struct odp_msg), - msg->port, s); - free(s); - } - *bufp = buf; - COVERAGE_INC(dpif_recv); - return 0; - } else { - VLOG_WARN_RL(&error_rl, "dp%u: discarding message truncated " - "from %zu bytes to %d", - dpif->minor, msg->length, retval); - error = ERANGE; - } - } else if (!retval) { - VLOG_WARN_RL(&error_rl, "dp%u: unexpected end of file", dpif->minor); - error = EPROTO; - } else { - VLOG_WARN_RL(&error_rl, - "dp%u: discarding too-short message (%d bytes)", - dpif->minor, retval); - error = ERANGE; + switch (type) { + case DPIF_UC_MISS: return "miss"; + case DPIF_UC_ACTION: return "action"; + case DPIF_UC_SAMPLE: return "sample"; + case DPIF_N_UC_TYPES: default: return ""; } +} - *bufp = NULL; - ofpbuf_delete(buf); - return error; +static bool OVS_UNUSED +is_valid_listen_mask(int listen_mask) +{ + return !(listen_mask & ~((1u << DPIF_UC_MISS) | + (1u << DPIF_UC_ACTION) | + (1u << DPIF_UC_SAMPLE))); } -void -dpif_recv_wait(struct dpif *dpif) +/* Retrieves 'dpif''s "listen mask" into '*listen_mask'. A 1-bit of value 2**X + * set in '*listen_mask' indicates that dpif_recv() will receive messages of + * the type (from "enum dpif_upcall_type") with value X. Returns 0 if + * successful, otherwise a positive errno value. */ +int +dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask) { - poll_fd_wait(dpif->fd, POLLIN); + int error = dpif->dpif_class->recv_get_mask(dpif, listen_mask); + if (error) { + *listen_mask = 0; + } + assert(is_valid_listen_mask(*listen_mask)); + log_operation(dpif, "recv_get_mask", error); + return error; } - -struct dpifmon { - struct dpif dpif; - struct nl_sock *sock; - int local_ifindex; -}; +/* Sets 'dpif''s "listen mask" to 'listen_mask'. A 1-bit of value 2**X set in + * '*listen_mask' requests that dpif_recv() will receive messages of the type + * (from "enum dpif_upcall_type") with value X. Returns 0 if successful, + * otherwise a positive errno value. */ int -dpifmon_create(const char *datapath_name, struct dpifmon **monp) +dpif_recv_set_mask(struct dpif *dpif, int listen_mask) { - struct dpifmon *mon; - char local_name[IFNAMSIZ]; int error; - mon = *monp = xmalloc(sizeof *mon); + assert(is_valid_listen_mask(listen_mask)); - error = dpif_open(datapath_name, &mon->dpif); - if (error) { - goto error; - } - error = dpif_port_get_name(&mon->dpif, ODPP_LOCAL, - local_name, sizeof local_name); - if (error) { - goto error_close_dpif; - } - - mon->local_ifindex = if_nametoindex(local_name); - if (!mon->local_ifindex) { - error = errno; - VLOG_WARN("could not get ifindex of %s device: %s", - local_name, strerror(errno)); - goto error_close_dpif; - } + error = dpif->dpif_class->recv_set_mask(dpif, listen_mask); + log_operation(dpif, "recv_set_mask", error); + return error; +} - error = nl_sock_create(NETLINK_ROUTE, RTNLGRP_LINK, 0, 0, &mon->sock); +/* Retrieve the sFlow sampling probability. '*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. + * + * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP + * indicates that 'dpif' does not support sFlow sampling. */ +int +dpif_get_sflow_probability(const struct dpif *dpif, uint32_t *probability) +{ + int error = (dpif->dpif_class->get_sflow_probability + ? dpif->dpif_class->get_sflow_probability(dpif, probability) + : EOPNOTSUPP); if (error) { - VLOG_WARN("could not create rtnetlink socket: %s", strerror(error)); - goto error_close_dpif; + *probability = 0; } - - return 0; - -error_close_dpif: - dpif_close(&mon->dpif); -error: - free(mon); - *monp = NULL; + log_operation(dpif, "get_sflow_probability", error); return error; } -void -dpifmon_destroy(struct dpifmon *mon) +/* Set the sFlow sampling probability. '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. + * + * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP + * indicates that 'dpif' does not support sFlow sampling. */ +int +dpif_set_sflow_probability(struct dpif *dpif, uint32_t probability) { - if (mon) { - dpif_close(&mon->dpif); - nl_sock_destroy(mon->sock); - } + int error = (dpif->dpif_class->set_sflow_probability + ? dpif->dpif_class->set_sflow_probability(dpif, probability) + : EOPNOTSUPP); + log_operation(dpif, "set_sflow_probability", error); + return error; } +/* Polls for an upcall from 'dpif'. If successful, stores the upcall into + * '*upcall'. Only upcalls of the types selected with dpif_recv_set_mask() + * member function will ordinarily be received (but if a message type is + * enabled and then later disabled, some stragglers might pop up). + * + * 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 + * and clients that exist so far.) + * + * Returns 0 if successful, otherwise a positive errno value. Returns EAGAIN + * if no upcall is immediately available. */ int -dpifmon_poll(struct dpifmon *mon, char **devnamep) +dpif_recv(struct dpif *dpif, struct dpif_upcall *upcall) { - static struct vlog_rate_limit slow_rl = VLOG_RATE_LIMIT_INIT(1, 5); - static const struct nl_policy rtnlgrp_link_policy[] = { - [IFLA_IFNAME] = { .type = NL_A_STRING }, - [IFLA_MASTER] = { .type = NL_A_U32, .optional = true }, - }; - struct nlattr *attrs[ARRAY_SIZE(rtnlgrp_link_policy)]; - struct ofpbuf *buf; - int error; + int error = dpif->dpif_class->recv(dpif, upcall); + if (!error && !VLOG_DROP_DBG(&dpmsg_rl)) { + struct ds flow; + char *packet; - *devnamep = NULL; -again: - error = nl_sock_recv(mon->sock, &buf, false); - switch (error) { - case 0: - if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg), - rtnlgrp_link_policy, - attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { - VLOG_WARN_RL(&slow_rl, "received bad rtnl message"); - error = ENOBUFS; - } else { - const char *devname = nl_attr_get_string(attrs[IFLA_IFNAME]); - bool for_us; - - if (attrs[IFLA_MASTER]) { - uint32_t master_ifindex = nl_attr_get_u32(attrs[IFLA_MASTER]); - for_us = master_ifindex == mon->local_ifindex; - } else { - /* It's for us if that device is one of our ports. This is - * open-coded instead of using dpif_port_query_by_name() to - * avoid logging a warning on failure. */ - struct odp_port port; - memset(&port, 0, sizeof port); - strncpy(port.devname, devname, sizeof port.devname); - for_us = !ioctl(mon->dpif.fd, ODP_PORT_QUERY, &port); - } - - if (!for_us) { - /* Not for us, try again. */ - ofpbuf_delete(buf); - COVERAGE_INC(dpifmon_poll_false_wakeup); - goto again; - } - COVERAGE_INC(dpifmon_poll_changed); - *devnamep = xstrdup(devname); - } - ofpbuf_delete(buf); - break; + packet = ofp_packet_to_string(upcall->packet->data, + upcall->packet->size, + upcall->packet->size); - case EAGAIN: - /* Nothing to do. */ - break; + ds_init(&flow); + odp_flow_key_format(upcall->key, upcall->key_len, &flow); - case ENOBUFS: - VLOG_WARN_RL(&slow_rl, "dpifmon socket overflowed"); - break; + VLOG_DBG("%s: %s upcall:\n%s\n%s", + dpif_name(dpif), dpif_upcall_type_to_string(upcall->type), + ds_cstr(&flow), packet); - default: - VLOG_WARN_RL(&slow_rl, "error on dpifmon socket: %s", strerror(error)); - break; + ds_destroy(&flow); + free(packet); } return error; } +/* Discards all messages that would otherwise be received by dpif_recv() on + * 'dpif'. */ void -dpifmon_run(struct dpifmon *mon UNUSED) +dpif_recv_purge(struct dpif *dpif) { - /* Nothing to do in this implementation. */ + COVERAGE_INC(dpif_purge); + if (dpif->dpif_class->recv_purge) { + dpif->dpif_class->recv_purge(dpif); + } } +/* Arranges for the poll loop to wake up when 'dpif' has a message queued to be + * received with dpif_recv(). */ void -dpifmon_wait(struct dpifmon *mon) +dpif_recv_wait(struct dpif *dpif) { - nl_sock_wait(mon->sock, POLLIN); + dpif->dpif_class->recv_wait(dpif); } - -static int get_openvswitch_major(void); -static int get_major(const char *target, int default_major); -static int -lookup_minor(const char *name, unsigned int *minor) +/* Obtains the NetFlow engine type and engine ID for 'dpif' into '*engine_type' + * and '*engine_id', respectively. */ +void +dpif_get_netflow_ids(const struct dpif *dpif, + uint8_t *engine_type, uint8_t *engine_id) { - struct ethtool_drvinfo drvinfo; - struct ifreq ifr; - int error; - int sock; - - *minor = -1; - sock = socket(AF_INET, SOCK_DGRAM, 0); - if (sock < 0) { - VLOG_WARN("socket(AF_INET) failed: %s", strerror(errno)); - error = errno; - goto error; - } - - memset(&ifr, 0, sizeof ifr); - strncpy(ifr.ifr_name, name, sizeof ifr.ifr_name); - ifr.ifr_data = (caddr_t) &drvinfo; - - memset(&drvinfo, 0, sizeof drvinfo); - drvinfo.cmd = ETHTOOL_GDRVINFO; - if (ioctl(sock, SIOCETHTOOL, &ifr)) { - VLOG_WARN("ioctl(SIOCETHTOOL) failed: %s", strerror(errno)); - error = errno; - goto error_close_sock; - } - - if (strcmp(drvinfo.driver, "openvswitch")) { - VLOG_WARN("%s is not an openvswitch device", name); - error = EOPNOTSUPP; - goto error_close_sock; - } - - if (!isdigit(drvinfo.bus_info[0])) { - VLOG_WARN("%s ethtool info does not contain an openvswitch minor", - name); - error = EPROTOTYPE; - goto error_close_sock; - } - - *minor = atoi(drvinfo.bus_info); - close(sock); - return 0; - -error_close_sock: - close(sock); -error: - return error; + *engine_type = dpif->netflow_engine_type; + *engine_id = dpif->netflow_engine_id; } -static int -make_openvswitch_device(unsigned int minor, char **fnp) +/* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority + * value for use in the OVS_ACTION_ATTR_SET_PRIORITY action. On success, + * returns 0 and stores the priority into '*priority'. On failure, returns a + * positive errno value and stores 0 into '*priority'. */ +int +dpif_queue_to_priority(const struct dpif *dpif, uint32_t queue_id, + uint32_t *priority) { - dev_t dev = makedev(get_openvswitch_major(), minor); - const char dirname[] = "/dev/net"; - struct stat s; - char fn[128]; - - *fnp = NULL; - sprintf(fn, "%s/dp%d", dirname, minor); - if (!stat(fn, &s)) { - if (!S_ISCHR(s.st_mode)) { - VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing", - fn); - } else if (s.st_rdev != dev) { - VLOG_WARN_RL(&error_rl, - "%s is device %u:%u instead of %u:%u, fixing", - fn, major(s.st_rdev), minor(s.st_rdev), - major(dev), minor(dev)); - } else { - goto success; - } - if (unlink(fn)) { - VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)", - fn, strerror(errno)); - return errno; - } - } else if (errno == ENOENT) { - if (stat(dirname, &s)) { - if (errno == ENOENT) { - if (mkdir(dirname, 0755)) { - VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)", - dirname, strerror(errno)); - return errno; - } - } else { - VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", - dirname, strerror(errno)); - return errno; - } - } - } else { - VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno)); - return errno; - } - - /* The device needs to be created. */ - if (mknod(fn, S_IFCHR | 0700, dev)) { - VLOG_WARN_RL(&error_rl, - "%s: creating character device %u:%u failed (%s)", - fn, major(dev), minor(dev), strerror(errno)); - return errno; + int error = (dpif->dpif_class->queue_to_priority + ? dpif->dpif_class->queue_to_priority(dpif, queue_id, + priority) + : EOPNOTSUPP); + if (error) { + *priority = 0; } - -success: - *fnp = xstrdup(fn); - return 0; + log_operation(dpif, "queue_to_priority", error); + return error; } - - -static int -get_openvswitch_major(void) + +void +dpif_init(struct dpif *dpif, const struct dpif_class *dpif_class, + const char *name, + uint8_t netflow_engine_type, uint8_t netflow_engine_id) { - static unsigned int openvswitch_major; - if (!openvswitch_major) { - enum { DEFAULT_MAJOR = 248 }; - openvswitch_major = get_major("openvswitch", DEFAULT_MAJOR); - } - return openvswitch_major; + dpif->dpif_class = dpif_class; + dpif->base_name = xstrdup(name); + dpif->full_name = xasprintf("%s@%s", dpif_class->type, name); + dpif->netflow_engine_type = netflow_engine_type; + dpif->netflow_engine_id = netflow_engine_id; } -static int -get_major(const char *target, int default_major) +/* Undoes the results of initialization. + * + * Normally this function only needs to be called from dpif_close(). + * However, it may be called by providers due to an error on opening + * that occurs after initialization. It this case dpif_close() would + * never be called. */ +void +dpif_uninit(struct dpif *dpif, bool close) { - const char fn[] = "/proc/devices"; - char line[128]; - FILE *file; - int ln; - - file = fopen(fn, "r"); - if (!file) { - VLOG_ERR("opening %s failed (%s)", fn, strerror(errno)); - goto error; - } + char *base_name = dpif->base_name; + char *full_name = dpif->full_name; - for (ln = 1; fgets(line, sizeof line, file); ln++) { - char name[64]; - int major; - - if (!strncmp(line, "Character", 9) || line[0] == '\0') { - /* Nothing to do. */ - } else if (!strncmp(line, "Block", 5)) { - /* We only want character devices, so skip the rest of the file. */ - break; - } else if (sscanf(line, "%d %63s", &major, name)) { - if (!strcmp(name, target)) { - fclose(file); - return major; - } - } else { - static bool warned; - if (!warned) { - VLOG_WARN("%s:%d: syntax error", fn, ln); - } - warned = true; - } + if (close) { + dpif->dpif_class->close(dpif); } - VLOG_ERR("%s: %s major not found (is the module loaded?), using " - "default major %d", fn, target, default_major); -error: - VLOG_INFO("using default major %d for %s", default_major, target); - return default_major; + free(base_name); + free(full_name); } - -static int -name_to_minor(const char *name, unsigned int *minor) + +static void +log_operation(const struct dpif *dpif, const char *operation, int error) { - if (!get_minor_from_name(name, minor)) { - return 0; + if (!error) { + VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation); + } else if (is_errno(error)) { + VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)", + dpif_name(dpif), operation, strerror(error)); + } else { + VLOG_WARN_RL(&error_rl, "%s: %s failed (%d/%d)", + dpif_name(dpif), operation, + get_ofp_err_type(error), get_ofp_err_code(error)); } - return lookup_minor(name, minor); } -static int -get_minor_from_name(const char *name, unsigned int *minor) +static enum vlog_level +flow_message_log_level(int error) { - if (!strncmp(name, "dp", 2) && isdigit(name[2])) { - *minor = atoi(name + 2); - return 0; - } else if (!strncmp(name, "nl:", 3) && isdigit(name[3])) { - /* This is for compatibility only and will be dropped. */ - *minor = atoi(name + 3); - return 0; - } else { - return EINVAL; - } + return error ? VLL_WARN : VLL_DBG; } -static int -open_by_minor(unsigned int minor, struct dpif *dpif) +static bool +should_log_flow_message(int error) { - int error; - char *fn; - int fd; + return !vlog_should_drop(THIS_MODULE, flow_message_log_level(error), + error ? &error_rl : &dpmsg_rl); +} - dpif->minor = -1; - dpif->fd = -1; - error = make_openvswitch_device(minor, &fn); +static void +log_flow_message(const struct dpif *dpif, int error, const char *operation, + const struct nlattr *key, size_t key_len, + const struct dpif_flow_stats *stats, + const struct nlattr *actions, size_t actions_len) +{ + struct ds ds = DS_EMPTY_INITIALIZER; + ds_put_format(&ds, "%s: ", dpif_name(dpif)); if (error) { - return error; + ds_put_cstr(&ds, "failed to "); } - - fd = open(fn, O_RDONLY | O_NONBLOCK); - if (fd < 0) { - error = errno; - VLOG_WARN("%s: open failed (%s)", fn, strerror(error)); - free(fn); - return error; + ds_put_format(&ds, "%s ", operation); + if (error) { + ds_put_format(&ds, "(%s) ", strerror(error)); } - - free(fn); - dpif->minor = minor; - dpif->fd = fd; - return 0; -} - -/* There is a tendency to construct odp_flow objects on the stack and to - * forget to properly initialize their "actions" and "n_actions" members. - * When this happens, we get memory corruption because the kernel - * writes through the random pointer that is in the "actions" member. - * - * This function attempts to combat the problem by: - * - * - Forcing a segfault if "actions" points to an invalid region (instead - * of just getting back EFAULT, which can be easily missed in the log). - * - * - Storing a distinctive value that is likely to cause an - * easy-to-identify error later if it is dereferenced, etc. - * - * - Triggering a warning on uninitialized memory from Valgrind if - * "actions" or "n_actions" was not initialized. - */ -static void -check_rw_odp_flow(struct odp_flow *flow) -{ - if (flow->n_actions) { - memset(&flow->actions[0], 0xcc, sizeof flow->actions[0]); + odp_flow_key_format(key, key_len, &ds); + if (stats) { + ds_put_cstr(&ds, ", "); + dpif_flow_stats_format(stats, &ds); + } + if (actions || actions_len) { + ds_put_cstr(&ds, ", actions:"); + format_odp_actions(&ds, actions, actions_len); } + vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds)); + ds_destroy(&ds); }