X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif.c;h=68af298ffaf9c89df3cf4796e673088e6cd36ae5;hb=7a97619ae4484604c15b846ce7268094e048729a;hp=c225e2bc958c17fbafd8a8833f7121c79063e036;hpb=c97fb13280c565f55ed0de7cf0bf06ffe8320b70;p=sliver-openvswitch.git diff --git a/lib/dpif.c b/lib/dpif.c index c225e2bc9..68af298ff 100644 --- a/lib/dpif.c +++ b/lib/dpif.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. + * Copyright (c) 2008, 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. @@ -30,13 +30,14 @@ #include "netdev.h" #include "netlink.h" #include "odp-util.h" +#include "ofp-errors.h" #include "ofp-print.h" #include "ofp-util.h" #include "ofpbuf.h" #include "packets.h" #include "poll-loop.h" #include "shash.h" -#include "svec.h" +#include "sset.h" #include "timeval.h" #include "util.h" #include "valgrind.h" @@ -57,10 +58,11 @@ COVERAGE_DEFINE(dpif_execute); COVERAGE_DEFINE(dpif_purge); static const struct dpif_class *base_dpif_classes[] = { -#ifdef HAVE_NETLINK +#ifdef LINUX_DATAPATH &dpif_linux_class, #endif &dpif_netdev_class, + &dpif_planetlab_class, }; struct registered_dpif_class { @@ -68,6 +70,7 @@ struct registered_dpif_class { int refcount; }; static struct shash dpif_classes = SHASH_INITIALIZER(&dpif_classes); +static struct sset dpif_blacklist = SSET_INITIALIZER(&dpif_blacklist); /* 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 @@ -85,6 +88,12 @@ static void log_flow_message(const struct dpif *dpif, int error, static void log_operation(const struct dpif *, const char *operation, int error); static bool should_log_flow_message(int error); +static void log_flow_put_message(struct dpif *, const struct dpif_flow_put *, + int error); +static void log_flow_del_message(struct dpif *, const struct dpif_flow_del *, + int error); +static void log_execute_message(struct dpif *, const struct dpif_execute *, + int error); static void dp_initialize(void) @@ -101,38 +110,6 @@ dp_initialize(void) } } -/* Performs periodic work needed by all the various kinds of dpifs. - * - * If your program opens any dpifs, it must call both this function and - * netdev_run() within its main poll loop. */ -void -dp_run(void) -{ - struct shash_node *node; - SHASH_FOR_EACH(node, &dpif_classes) { - const struct registered_dpif_class *registered_class = node->data; - if (registered_class->dpif_class->run) { - registered_class->dpif_class->run(); - } - } -} - -/* Arranges for poll_block() to wake up when dp_run() needs to be called. - * - * If your program opens any dpifs, it must call both this function and - * netdev_wait() within its main poll loop. */ -void -dp_wait(void) -{ - struct shash_node *node; - SHASH_FOR_EACH(node, &dpif_classes) { - const struct registered_dpif_class *registered_class = node->data; - if (registered_class->dpif_class->wait) { - registered_class->dpif_class->wait(); - } - } -} - /* Registers a new datapath provider. After successful registration, new * datapaths of that type can be opened using dpif_open(). */ int @@ -140,6 +117,12 @@ dp_register_provider(const struct dpif_class *new_class) { struct registered_dpif_class *registered_class; + if (sset_contains(&dpif_blacklist, new_class->type)) { + VLOG_DBG("attempted to register blacklisted provider: %s", + new_class->type); + return EINVAL; + } + if (shash_find(&dpif_classes, new_class->type)) { VLOG_WARN("attempted to register duplicate datapath provider: %s", new_class->type); @@ -183,37 +166,45 @@ dp_unregister_provider(const char *type) return 0; } +/* Blacklists a provider. Causes future calls of dp_register_provider() with + * a dpif_class which implements 'type' to fail. */ +void +dp_blacklist_provider(const char *type) +{ + sset_add(&dpif_blacklist, type); +} + /* Clears 'types' and enumerates the types of all currently registered datapath - * providers into it. The caller must first initialize the svec. */ + * providers into it. The caller must first initialize the sset. */ void -dp_enumerate_types(struct svec *types) +dp_enumerate_types(struct sset *types) { struct shash_node *node; dp_initialize(); - svec_clear(types); + sset_clear(types); SHASH_FOR_EACH(node, &dpif_classes) { const struct registered_dpif_class *registered_class = node->data; - svec_add(types, registered_class->dpif_class->type); + 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 svec. Returns 0 if + * 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 svec *names) +dp_enumerate_names(const char *type, struct sset *names) { const struct registered_dpif_class *registered_class; const struct dpif_class *dpif_class; int error; dp_initialize(); - svec_clear(names); + sset_clear(names); registered_class = shash_find_data(&dpif_classes, type); if (!registered_class) { @@ -232,8 +223,10 @@ dp_enumerate_names(const char *type, struct svec *names) 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. */ +/* 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 dp_parse_name(const char *datapath_name_, char **name, char **type) { @@ -244,10 +237,10 @@ dp_parse_name(const char *datapath_name_, char **name, char **type) if (separator) { *separator = '\0'; *type = datapath_name; - *name = xstrdup(separator + 1); + *name = xstrdup(dpif_normalize_type(separator + 1)); } else { *name = datapath_name; - *type = NULL; + *type = xstrdup(dpif_normalize_type(NULL)); } } @@ -260,9 +253,7 @@ do_open(const char *name, const char *type, bool create, struct dpif **dpifp) dp_initialize(); - if (!type || *type == '\0') { - type = "system"; - } + type = dpif_normalize_type(type); registered_class = shash_find_data(&dpif_classes, type); if (!registered_class) { @@ -347,6 +338,25 @@ dpif_close(struct dpif *dpif) } } +/* Performs periodic work needed by 'dpif'. */ +void +dpif_run(struct dpif *dpif) +{ + if (dpif->dpif_class->run) { + dpif->dpif_class->run(dpif); + } +} + +/* Arranges for poll_block() to wake up when dp_run() needs to be called for + * 'dpif'. */ +void +dpif_wait(struct dpif *dpif) +{ + 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 * @@ -363,33 +373,16 @@ dpif_base_name(const struct dpif *dpif) return dpif->base_name; } -/* 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 will be enumerated. +/* Returns the fully spelled out name for the given datapath 'type'. * - * The caller must already have initialized 'all_names'. Any existing names in - * 'all_names' will not be disturbed. */ -int -dpif_get_all_names(const struct dpif *dpif, struct svec *all_names) + * 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) { - if (dpif->dpif_class->get_all_names) { - int error = dpif->dpif_class->get_all_names(dpif, all_names); - if (error) { - VLOG_WARN_RL(&error_rl, - "failed to retrieve names for datpath %s: %s", - dpif_name(dpif), strerror(error)); - } - return error; - } else { - svec_add(all_names, dpif_base_name(dpif)); - return 0; - } + 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(). */ @@ -408,7 +401,7 @@ dpif_delete(struct dpif *dpif) /* Retrieves statistics for 'dpif' into 'stats'. Returns 0 if successful, * otherwise a positive errno value. */ int -dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats) +dpif_get_dp_stats(const struct dpif *dpif, struct dpif_dp_stats *stats) { int error = dpif->dpif_class->get_stats(dpif, stats); if (error) { @@ -418,46 +411,27 @@ dpif_get_dp_stats(const struct dpif *dpif, struct odp_stats *stats) 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_drop_frags(const struct dpif *dpif, bool *drop_frags) -{ - int error = dpif->dpif_class->get_drop_frags(dpif, drop_frags); - if (error) { - *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_drop_frags(struct dpif *dpif, bool drop_frags) -{ - 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). */ +/* Attempts to add 'netdev' as a port on 'dpif'. If 'port_nop' is + * non-null and its value is not UINT16_MAX, then attempts to use the + * value as the port number. + * + * 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_port_add(struct dpif *dpif, struct netdev *netdev, uint16_t *port_nop) { const char *netdev_name = netdev_get_name(netdev); - uint16_t port_no; + uint16_t port_no = UINT16_MAX; int error; COVERAGE_INC(dpif_port_add); + if (port_nop) { + port_no = *port_nop; + } + error = dpif->dpif_class->port_add(dpif, netdev, &port_no); if (!error) { VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu16, @@ -552,11 +526,14 @@ dpif_port_query_by_name(const struct dpif *dpif, const char *devname, } else { memset(port, 0, sizeof *port); - /* Log level is DBG here because all the current callers are interested - * in whether 'dpif' actually has a port 'devname', so that it's not an - * issue worth logging if it doesn't. */ - VLOG_DBG_RL(&error_rl, "%s: failed to query port %s: %s", - dpif_name(dpif), devname, strerror(error)); + /* 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; } @@ -569,6 +546,26 @@ dpif_get_max_ports(const struct dpif *dpif) return dpif->dpif_class->get_max_ports(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 UINT16_MAX is a special case: it returns a reserved PID, not + * allocated to any port, that the client may use for special purposes. + * + * The return value is only meaningful when DPIF_UC_ACTION has been enabled in + * the 'dpif''s listen mask. It is allowed to change when DPIF_UC_ACTION is + * disabled and then re-enabled, so a client that does that must be prepared to + * update all of the flows that it installed that contain + * OVS_ACTION_ATTR_USERSPACE actions. */ +uint32_t +dpif_port_get_pid(const struct dpif *dpif, uint16_t port_no) +{ + return (dpif->dpif_class->port_get_pid + ? (dpif->dpif_class->port_get_pid)(dpif, port_no) + : 0); +} + /* 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 @@ -688,6 +685,19 @@ dpif_port_poll_wait(const struct dpif *dpif) dpif->dpif_class->port_poll_wait(dpif); } +/* Extracts the flow stats for a packet. The 'flow' and 'packet' + * arguments must have been initialized through a call to flow_extract(). + * 'used' is stored into stats->used. */ +void +dpif_flow_stats_extract(const struct flow *flow, const struct ofpbuf *packet, + long long int used, struct dpif_flow_stats *stats) +{ + stats->tcp_flags = packet_get_tcp_flags(packet, flow); + stats->n_bytes = packet->size; + stats->n_packets = 1; + stats->used = used; +} + /* Appends a human-readable representation of 'stats' to 's'. */ void dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s) @@ -699,7 +709,10 @@ dpif_flow_stats_format(const struct dpif_flow_stats *stats, struct ds *s) } else { ds_put_format(s, "never"); } - /* XXX tcp_flags? */ + if (stats->tcp_flags) { + ds_put_cstr(s, ", flags:"); + packet_format_tcp_flags(s, stats->tcp_flags); + } } /* Deletes all flows from 'dpif'. Returns 0 if successful, otherwise a @@ -717,7 +730,7 @@ dpif_flow_flush(struct dpif *dpif) } /* Queries 'dpif' for a flow entry. The flow is specified by the Netlink - * attributes with types ODP_KEY_ATTR_* in the 'key_len' bytes starting at + * attributes with types OVS_KEY_ATTR_* in the 'key_len' bytes starting at * 'key'. * * Returns 0 if successful. If no flow matches, returns ENOENT. On other @@ -731,7 +744,7 @@ dpif_flow_flush(struct dpif *dpif) * If 'stats' is nonnull, then on success it will be updated with the flow's * statistics. */ int -dpif_flow_get(const struct dpif *dpif, int flags, +dpif_flow_get(const struct dpif *dpif, const struct nlattr *key, size_t key_len, struct ofpbuf **actionsp, struct dpif_flow_stats *stats) { @@ -739,8 +752,7 @@ dpif_flow_get(const struct dpif *dpif, int flags, COVERAGE_INC(dpif_flow_get); - error = dpif->dpif_class->flow_get(dpif, flags, key, key_len, actionsp, - stats); + error = dpif->dpif_class->flow_get(dpif, key, key_len, actionsp, stats); if (error) { if (actionsp) { *actionsp = NULL; @@ -766,69 +778,78 @@ dpif_flow_get(const struct dpif *dpif, int flags, return error; } +static int +dpif_flow_put__(struct dpif *dpif, const struct dpif_flow_put *put) +{ + int error; + + COVERAGE_INC(dpif_flow_put); + assert(!(put->flags & ~(DPIF_FP_CREATE | DPIF_FP_MODIFY + | DPIF_FP_ZERO_STATS))); + + error = dpif->dpif_class->flow_put(dpif, put); + if (error && put->stats) { + memset(put->stats, 0, sizeof *put->stats); + } + log_flow_put_message(dpif, put, error); + return error; +} + /* Adds or modifies a flow in 'dpif'. The flow is specified by the Netlink - * attributes with types ODP_KEY_ATTR_* in the 'key_len' bytes starting at + * 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 ODPAT_* in the 'actions_len' bytes starting at 'actions'. + * 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 ODPPF_CREATE. Otherwise the operation will fail with + * '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 ODPPF_MODIFY. Otherwise the operation will + * 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 ODPPF_ZERO_STATS, and left as-is + * 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, int flags, +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) +{ + struct dpif_flow_put put; + + put.flags = flags; + put.key = key; + put.key_len = key_len; + put.actions = actions; + put.actions_len = actions_len; + put.stats = stats; + return dpif_flow_put__(dpif, &put); +} + +static int +dpif_flow_del__(struct dpif *dpif, struct dpif_flow_del *del) { int error; - COVERAGE_INC(dpif_flow_put); + COVERAGE_INC(dpif_flow_del); - 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)) { - enum { ODPPF_ALL = ODPPF_CREATE | ODPPF_MODIFY | ODPPF_ZERO_STATS }; - struct ds s; - - ds_init(&s); - ds_put_cstr(&s, "put"); - if (flags & ODPPF_CREATE) { - ds_put_cstr(&s, "[create]"); - } - if (flags & ODPPF_MODIFY) { - ds_put_cstr(&s, "[modify]"); - } - if (flags & ODPPF_ZERO_STATS) { - ds_put_cstr(&s, "[zero]"); - } - if (flags & ~ODPPF_ALL) { - ds_put_format(&s, "[%x]", flags & ~ODPPF_ALL); - } - log_flow_message(dpif, error, ds_cstr(&s), key, key_len, stats, - actions, actions_len); - ds_destroy(&s); + error = dpif->dpif_class->flow_del(dpif, del); + if (error && del->stats) { + memset(del->stats, 0, sizeof *del->stats); } + log_flow_del_message(dpif, del, error); 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 ODP_KEY_ATTR_* in the 'key_len' bytes starting at 'key'. + * 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. */ @@ -837,19 +858,12 @@ dpif_flow_del(struct dpif *dpif, const struct nlattr *key, size_t key_len, struct dpif_flow_stats *stats) { - int error; + struct dpif_flow_del del; - COVERAGE_INC(dpif_flow_del); - - 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; + del.key = key; + del.key_len = key_len; + del.stats = stats; + return dpif_flow_del__(dpif, &del); } /* Initializes 'dump' to begin dumping the flows in a dpif. @@ -874,11 +888,11 @@ dpif_flow_dump_start(struct dpif_flow_dump *dump, const struct dpif *dpif) * 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 ODP_KEY_ATTR_* representing the + * 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 ODPAT_* representing the dumped flow's - * actions. If 'stats' is nonnull then it will be set to the dumped flow's - * statistics. + * 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 @@ -943,132 +957,162 @@ dpif_flow_dump_done(struct dpif_flow_dump *dump) 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'. - * - * Returns 0 if successful, otherwise a positive errno value. */ -int -dpif_execute(struct dpif *dpif, - const struct nlattr *actions, size_t actions_len, - const struct ofpbuf *buf) +static int +dpif_execute__(struct dpif *dpif, const struct dpif_execute *execute) { int error; COVERAGE_INC(dpif_execute); - if (actions_len > 0) { - error = dpif->dpif_class->execute(dpif, actions, actions_len, buf); + if (execute->actions_len > 0) { + error = dpif->dpif_class->execute(dpif, execute); } else { error = 0; } - 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, "%s: execute ", dpif_name(dpif)); - format_odp_actions(&ds, actions, actions_len); - if (error) { - ds_put_format(&ds, " failed (%s)", strerror(error)); - } - ds_put_format(&ds, " on packet %s", packet); - vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds)); - ds_destroy(&ds); - free(packet); - } - return error; -} + log_execute_message(dpif, execute, error); -/* Retrieves 'dpif''s "listen mask" into '*listen_mask'. Each ODPL_* bit set - * in '*listen_mask' indicates that dpif_recv() will receive messages of that - * type. Returns 0 if successful, otherwise a positive errno value. */ -int -dpif_recv_get_mask(const struct dpif *dpif, int *listen_mask) -{ - int error = dpif->dpif_class->recv_get_mask(dpif, listen_mask); - if (error) { - *listen_mask = 0; - } - log_operation(dpif, "recv_get_mask", error); return error; } -/* Sets 'dpif''s "listen mask" to 'listen_mask'. Each ODPL_* bit set in - * '*listen_mask' requests that dpif_recv() receive messages of that type. +/* 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 + * tunnel and in_port.) + * * Returns 0 if successful, otherwise a positive errno value. */ int -dpif_recv_set_mask(struct dpif *dpif, int listen_mask) +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 = dpif->dpif_class->recv_set_mask(dpif, listen_mask); - log_operation(dpif, "recv_set_mask", error); - return error; + struct dpif_execute execute; + + execute.key = key; + execute.key_len = key_len; + execute.actions = actions; + execute.actions_len = actions_len; + execute.packet = buf; + return dpif_execute__(dpif, &execute); } -/* 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. +/* 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. * - * 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) + * This function exists because some datapaths can perform batched operations + * faster than individual operations. */ +void +dpif_operate(struct dpif *dpif, struct dpif_op **ops, size_t n_ops) { - int error = (dpif->dpif_class->get_sflow_probability - ? dpif->dpif_class->get_sflow_probability(dpif, probability) - : EOPNOTSUPP); - if (error) { - *probability = 0; + size_t i; + + if (dpif->dpif_class->operate) { + dpif->dpif_class->operate(dpif, ops, n_ops); + + for (i = 0; i < n_ops; i++) { + struct dpif_op *op = ops[i]; + + switch (op->type) { + case DPIF_OP_FLOW_PUT: + log_flow_put_message(dpif, &op->u.flow_put, op->error); + break; + + case DPIF_OP_FLOW_DEL: + log_flow_del_message(dpif, &op->u.flow_del, op->error); + break; + + case DPIF_OP_EXECUTE: + log_execute_message(dpif, &op->u.execute, op->error); + break; + } + } + return; + } + + for (i = 0; i < n_ops; i++) { + struct dpif_op *op = ops[i]; + + switch (op->type) { + case DPIF_OP_FLOW_PUT: + op->error = dpif_flow_put__(dpif, &op->u.flow_put); + break; + + case DPIF_OP_FLOW_DEL: + op->error = dpif_flow_del__(dpif, &op->u.flow_del); + break; + + case DPIF_OP_EXECUTE: + op->error = dpif_execute__(dpif, &op->u.execute); + break; + + default: + NOT_REACHED(); + } } - log_operation(dpif, "get_sflow_probability", error); - return error; } -/* 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 a string that represents 'type', for use in log messages. */ +const char * +dpif_upcall_type_to_string(enum dpif_upcall_type type) +{ + switch (type) { + case DPIF_UC_MISS: return "miss"; + case DPIF_UC_ACTION: return "action"; + case DPIF_N_UC_TYPES: default: return ""; + } +} + +/* Enables or disables receiving packets with dpif_recv() on 'dpif'. Returns 0 + * if successful, otherwise a positive errno value. * - * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP - * indicates that 'dpif' does not support sFlow sampling. */ + * Turning packet receive off and then back on may change the Netlink PID + * assignments returned by dpif_port_get_pid(). If the client does this, it + * must update all of the flows that have OVS_ACTION_ATTR_USERSPACE actions + * using the new PID assignment. */ int -dpif_set_sflow_probability(struct dpif *dpif, uint32_t probability) +dpif_recv_set(struct dpif *dpif, bool enable) { - int error = (dpif->dpif_class->set_sflow_probability - ? dpif->dpif_class->set_sflow_probability(dpif, probability) - : EOPNOTSUPP); - log_operation(dpif, "set_sflow_probability", error); + int error = dpif->dpif_class->recv_set(dpif, enable); + log_operation(dpif, "recv_set", error); return error; } /* Polls for an upcall from 'dpif'. If successful, stores the upcall into - * '*upcall'. Only upcalls of the types selected with the set_listen_mask - * member function will ordinarily be received (but if a message type is - * enabled and then later disabled, some stragglers might pop up). + * '*upcall', using 'buf' for storage. Should only be called if + * dpif_recv_set() has been used to enable receiving packets on 'dpif'. * - * The caller takes ownership of the data that 'upcall' points to. - * 'upcall->key' and 'upcall->actions' (if nonnull) point into data owned by - * 'upcall->packet', so their memory cannot be freed separately. (This is + * 'upcall->packet' and 'upcall->key' point into data in the caller-provided + * 'buf', so their memory cannot be freed separately from 'buf'. (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 -dpif_recv(struct dpif *dpif, struct dpif_upcall *upcall) +dpif_recv(struct dpif *dpif, struct dpif_upcall *upcall, struct ofpbuf *buf) { - int error = dpif->dpif_class->recv(dpif, upcall); + int error = dpif->dpif_class->recv(dpif, upcall, buf); if (!error && !VLOG_DROP_DBG(&dpmsg_rl)) { - struct flow flow; - char *s; - - s = ofp_packet_to_string(upcall->packet->data, - upcall->packet->size, upcall->packet->size); - odp_flow_key_to_flow(upcall->key, upcall->key_len, &flow); - - VLOG_DBG("%s: %s upcall on port %"PRIu16": %s", dpif_name(dpif), - (upcall->type == _ODPL_MISS_NR ? "miss" - : upcall->type == _ODPL_ACTION_NR ? "action" - : upcall->type == _ODPL_SFLOW_NR ? "sFlow" - : ""), - flow.in_port, s); - free(s); + struct ds flow; + char *packet; + + packet = ofp_packet_to_string(upcall->packet->data, + upcall->packet->size); + + ds_init(&flow); + odp_flow_key_format(upcall->key, upcall->key_len, &flow); + + VLOG_DBG("%s: %s upcall:\n%s\n%s", + dpif_name(dpif), dpif_upcall_type_to_string(upcall->type), + ds_cstr(&flow), packet); + + ds_destroy(&flow); + free(packet); + } else if (error && error != EAGAIN) { + log_operation(dpif, "recv", error); } return error; } @@ -1103,9 +1147,9 @@ dpif_get_netflow_ids(const struct dpif *dpif, } /* Translates OpenFlow queue ID 'queue_id' (in host byte order) into a priority - * value for use in the ODPAT_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'. */ + * value used for setting packet priority. + * 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) @@ -1158,13 +1202,12 @@ log_operation(const struct dpif *dpif, const char *operation, int error) { if (!error) { VLOG_DBG_RL(&dpmsg_rl, "%s: %s success", dpif_name(dpif), operation); - } else if (is_errno(error)) { + } else if (ofperr_is_valid(error)) { VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)", - dpif_name(dpif), operation, strerror(error)); + dpif_name(dpif), operation, ofperr_get_name(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)); + VLOG_WARN_RL(&error_rl, "%s: %s failed (%s)", + dpif_name(dpif), operation, strerror(error)); } } @@ -1208,3 +1251,60 @@ log_flow_message(const struct dpif *dpif, int error, const char *operation, vlog(THIS_MODULE, flow_message_log_level(error), "%s", ds_cstr(&ds)); ds_destroy(&ds); } + +static void +log_flow_put_message(struct dpif *dpif, const struct dpif_flow_put *put, + int error) +{ + if (should_log_flow_message(error)) { + struct ds s; + + ds_init(&s); + ds_put_cstr(&s, "put"); + if (put->flags & DPIF_FP_CREATE) { + ds_put_cstr(&s, "[create]"); + } + if (put->flags & DPIF_FP_MODIFY) { + ds_put_cstr(&s, "[modify]"); + } + if (put->flags & DPIF_FP_ZERO_STATS) { + ds_put_cstr(&s, "[zero]"); + } + log_flow_message(dpif, error, ds_cstr(&s), + put->key, put->key_len, put->stats, + put->actions, put->actions_len); + ds_destroy(&s); + } +} + +static void +log_flow_del_message(struct dpif *dpif, const struct dpif_flow_del *del, + int error) +{ + if (should_log_flow_message(error)) { + log_flow_message(dpif, error, "flow_del", del->key, del->key_len, + !error ? del->stats : NULL, NULL, 0); + } +} + +static void +log_execute_message(struct dpif *dpif, const struct dpif_execute *execute, + int error) +{ + if (!(error ? VLOG_DROP_WARN(&error_rl) : VLOG_DROP_DBG(&dpmsg_rl))) { + struct ds ds = DS_EMPTY_INITIALIZER; + char *packet; + + packet = ofp_packet_to_string(execute->packet->data, + execute->packet->size); + ds_put_format(&ds, "%s: execute ", dpif_name(dpif)); + format_odp_actions(&ds, execute->actions, execute->actions_len); + if (error) { + ds_put_format(&ds, " failed (%s)", strerror(error)); + } + ds_put_format(&ds, " on packet %s", packet); + vlog(THIS_MODULE, error ? VLL_WARN : VLL_DBG, "%s", ds_cstr(&ds)); + ds_destroy(&ds); + free(packet); + } +}