X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fust-app.c;h=d8adccde214925f2f9cd68d20ef65dadc3617951;hp=e36238182d8f04e078d440dce4d3bee94b9c5360;hb=7cc9a73c638bff0a86d7ea802ebd7fa3b4716ca9;hpb=4dc3dfc55223dae057447c03a1e7aadc2c177b3a diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c index e36238182..d8adccde2 100644 --- a/src/bin/lttng-sessiond/ust-app.c +++ b/src/bin/lttng-sessiond/ust-app.c @@ -34,29 +34,44 @@ #include "buffer-registry.h" #include "fd-limit.h" -#include "health.h" +#include "health-sessiond.h" #include "ust-app.h" #include "ust-consumer.h" #include "ust-ctl.h" +#include "utils.h" -/* Next available channel key. */ -static unsigned long next_channel_key; -static unsigned long next_session_id; +/* Next available channel key. Access under next_channel_key_lock. */ +static uint64_t _next_channel_key; +static pthread_mutex_t next_channel_key_lock = PTHREAD_MUTEX_INITIALIZER; + +/* Next available session ID. Access under next_session_id_lock. */ +static uint64_t _next_session_id; +static pthread_mutex_t next_session_id_lock = PTHREAD_MUTEX_INITIALIZER; /* - * Return the atomically incremented value of next_channel_key. + * Return the incremented value of next_channel_key. */ -static inline unsigned long get_next_channel_key(void) +static uint64_t get_next_channel_key(void) { - return uatomic_add_return(&next_channel_key, 1); + uint64_t ret; + + pthread_mutex_lock(&next_channel_key_lock); + ret = ++_next_channel_key; + pthread_mutex_unlock(&next_channel_key_lock); + return ret; } /* * Return the atomically incremented value of next_session_id. */ -static inline unsigned long get_next_session_id(void) +static uint64_t get_next_session_id(void) { - return uatomic_add_return(&next_session_id, 1); + uint64_t ret; + + pthread_mutex_lock(&next_session_id_lock); + ret = ++_next_session_id; + pthread_mutex_unlock(&next_session_id_lock); + return ret; } static void copy_channel_attr_to_ustctl( @@ -89,7 +104,7 @@ static int ht_match_ust_app_event(struct cds_lfht_node *node, const void *_key) event = caa_container_of(node, struct ust_app_event, node.node); key = _key; - /* Match the 3 elements of the key: name, filter and loglevel. */ + /* Match the 4 elements of the key: name, filter, loglevel, exclusions */ /* Event name */ if (strncmp(event->attr.name, key->name, sizeof(event->attr.name)) != 0) { @@ -125,6 +140,21 @@ static int ht_match_ust_app_event(struct cds_lfht_node *node, const void *_key) } } + /* One of the exclusions is NULL, fail. */ + if ((key->exclusion && !event->exclusion) || (!key->exclusion && event->exclusion)) { + goto no_match; + } + + if (key->exclusion && event->exclusion) { + /* Both exclusions exists, check count followed by the names. */ + if (event->exclusion->count != key->exclusion->count || + memcmp(event->exclusion->names, key->exclusion->names, + event->exclusion->count * LTTNG_UST_SYM_NAME_LEN) != 0) { + goto no_match; + } + } + + /* Match. */ return 1; @@ -151,6 +181,7 @@ static void add_unique_ust_app_event(struct ust_app_channel *ua_chan, key.name = event->attr.name; key.filter = event->filter; key.loglevel = event->attr.loglevel; + key.exclusion = event->exclusion; node_ptr = cds_lfht_add_unique(ht->ht, ht->hash_fct(event->node.key, lttng_ht_seed), @@ -256,7 +287,8 @@ void delete_ust_app_event(int sock, struct ust_app_event *ua_event) assert(ua_event); free(ua_event->filter); - + if (ua_event->exclusion != NULL) + free(ua_event->exclusion); if (ua_event->obj != NULL) { ret = ustctl_release_object(sock, ua_event->obj); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { @@ -307,9 +339,9 @@ void delete_ust_app_stream(int sock, struct ust_app_stream *stream) /* * We need to execute ht_destroy outside of RCU read-side critical - * section, so we postpone its execution using call_rcu. It is simpler - * than to change the semantic of the many callers of - * delete_ust_app_channel(). + * section and outside of call_rcu thread, so we postpone its execution + * using ht_cleanup_push. It is simpler than to change the semantic of + * the many callers of delete_ust_app_session(). */ static void delete_ust_app_channel_rcu(struct rcu_head *head) @@ -317,8 +349,8 @@ void delete_ust_app_channel_rcu(struct rcu_head *head) struct ust_app_channel *ua_chan = caa_container_of(head, struct ust_app_channel, rcu_head); - lttng_ht_destroy(ua_chan->ctx); - lttng_ht_destroy(ua_chan->events); + ht_cleanup_push(ua_chan->ctx); + ht_cleanup_push(ua_chan->events); free(ua_chan); } @@ -349,6 +381,7 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, /* Wipe context */ cds_lfht_for_each_entry(ua_chan->ctx->ht, &iter.iter, ua_ctx, node.node) { + cds_list_del(&ua_ctx->list); ret = lttng_ht_del(ua_chan->ctx, &iter); assert(!ret); delete_ust_app_ctx(sock, ua_ctx); @@ -362,10 +395,12 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, delete_ust_app_event(sock, ua_event); } - /* Wipe and free registry from session registry. */ - registry = get_session_registry(ua_chan->session); - if (registry) { - ust_registry_channel_del_free(registry, ua_chan->key); + if (ua_chan->session->buffer_type == LTTNG_BUFFER_PER_PID) { + /* Wipe and free registry from session registry. */ + registry = get_session_registry(ua_chan->session); + if (registry) { + ust_registry_channel_del_free(registry, ua_chan->key); + } } if (ua_chan->obj != NULL) { @@ -384,7 +419,10 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, } /* - * Push metadata to consumer socket. The socket lock MUST be acquired. + * Push metadata to consumer socket. + * + * The socket lock MUST be acquired. + * The ust app session lock MUST be acquired. * * On success, return the len of metadata pushed or else a negative value. */ @@ -398,8 +436,19 @@ ssize_t ust_app_push_metadata(struct ust_registry_session *registry, assert(registry); assert(socket); - /* Should never be 0 which is the initial state. */ - assert(registry->metadata_key); + + /* + * On a push metadata error either the consumer is dead or the metadata + * channel has been destroyed because its endpoint might have died (e.g: + * relayd). If so, the metadata closed flag is set to 1 so we deny pushing + * metadata again which is not valid anymore on the consumer side. + * + * The ust app session mutex locked allows us to make this check without + * the registry lock. + */ + if (registry->metadata_closed) { + return -EPIPE; + } pthread_mutex_lock(®istry->lock); @@ -474,7 +523,7 @@ static int push_metadata(struct ust_registry_session *registry, */ if (!registry->metadata_key) { ret_val = 0; - goto error_rcu_unlock; + goto end_rcu_unlock; } /* Get consumer socket to use to push the metadata.*/ @@ -507,6 +556,13 @@ static int push_metadata(struct ust_registry_session *registry, return 0; error_rcu_unlock: + /* + * On error, flag the registry that the metadata is closed. We were unable + * to push anything and this means that either the consumer is not + * responding or the metadata cache has been destroyed on the consumer. + */ + registry->metadata_closed = 1; +end_rcu_unlock: rcu_read_unlock(); return ret_val; } @@ -532,7 +588,7 @@ static int close_metadata(struct ust_registry_session *registry, if (!registry->metadata_key || registry->metadata_closed) { ret = 0; - goto error; + goto end; } /* Get consumer socket to use to push the metadata.*/ @@ -548,19 +604,23 @@ static int close_metadata(struct ust_registry_session *registry, goto error; } - /* Metadata successfully closed. Flag the registry. */ - registry->metadata_closed = 1; - error: + /* + * Metadata closed. Even on error this means that the consumer is not + * responding or not found so either way a second close should NOT be emit + * for this registry. + */ + registry->metadata_closed = 1; +end: rcu_read_unlock(); return ret; } /* * We need to execute ht_destroy outside of RCU read-side critical - * section, so we postpone its execution using call_rcu. It is simpler - * than to change the semantic of the many callers of - * delete_ust_app_session(). + * section and outside of call_rcu thread, so we postpone its execution + * using ht_cleanup_push. It is simpler than to change the semantic of + * the many callers of delete_ust_app_session(). */ static void delete_ust_app_session_rcu(struct rcu_head *head) @@ -568,7 +628,7 @@ void delete_ust_app_session_rcu(struct rcu_head *head) struct ust_app_session *ua_sess = caa_container_of(head, struct ust_app_session, rcu_head); - lttng_ht_destroy(ua_sess->channels); + ht_cleanup_push(ua_sess->channels); free(ua_sess); } @@ -587,16 +647,21 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, assert(ua_sess); + pthread_mutex_lock(&ua_sess->lock); + registry = get_session_registry(ua_sess); - if (registry) { + if (registry && !registry->metadata_closed) { /* Push metadata for application before freeing the application. */ (void) push_metadata(registry, ua_sess->consumer); /* * Don't ask to close metadata for global per UID buffers. Close - * metadata only on destroy trace session in this case. + * metadata only on destroy trace session in this case. Also, the + * previous push metadata could have flag the metadata registry to + * close so don't send a close command if closed. */ - if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID) { + if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID && + !registry->metadata_closed) { /* And ask to close it for this session registry. */ (void) close_metadata(registry, ua_sess->consumer); } @@ -625,6 +690,8 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, sock, ret); } } + pthread_mutex_unlock(&ua_sess->lock); + call_rcu(&ua_sess->rcu_head, delete_ust_app_session_rcu); } @@ -653,8 +720,8 @@ void delete_ust_app(struct ust_app *app) rcu_read_unlock(); } - lttng_ht_destroy(app->sessions); - lttng_ht_destroy(app->ust_objd); + ht_cleanup_push(app->sessions); + ht_cleanup_push(app->ust_objd); /* * Wait until we have deleted the application from the sock hash table @@ -776,6 +843,7 @@ struct ust_app_channel *alloc_ust_app_channel(char *name, lttng_ht_node_init_str(&ua_chan->node, ua_chan->name); CDS_INIT_LIST_HEAD(&ua_chan->streams.head); + CDS_INIT_LIST_HEAD(&ua_chan->ctx_list); /* Copy attributes */ if (attr) { @@ -867,6 +935,8 @@ struct ust_app_ctx *alloc_ust_app_ctx(struct lttng_ust_context *uctx) goto error; } + CDS_INIT_LIST_HEAD(&ua_ctx->list); + if (uctx) { memcpy(&ua_ctx->ctx, uctx, sizeof(ua_ctx->ctx)); } @@ -904,8 +974,7 @@ error: * Find an ust_app using the sock and return it. RCU read side lock must be * held before calling this helper function. */ -static -struct ust_app *find_app_by_sock(int sock) +struct ust_app *ust_app_find_by_sock(int sock) { struct lttng_ht_node_ulong *node; struct lttng_ht_iter iter; @@ -953,7 +1022,8 @@ error: * Return an ust_app_event object or NULL on error. */ static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht, - char *name, struct lttng_ust_filter_bytecode *filter, int loglevel) + char *name, struct lttng_ust_filter_bytecode *filter, int loglevel, + const struct lttng_event_exclusion *exclusion) { struct lttng_ht_iter iter; struct lttng_ht_node_str *node; @@ -967,6 +1037,8 @@ static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht, key.name = name; key.filter = filter; key.loglevel = loglevel; + /* lttng_event_exclusion and lttng_ust_event_exclusion structures are similar */ + key.exclusion = (struct lttng_ust_event_exclusion *)exclusion; /* Lookup using the event name as hash and a custom match fct. */ cds_lfht_lookup(ht->ht, ht->hash_fct((void *) name, lttng_ht_seed), @@ -1002,6 +1074,12 @@ int create_ust_channel_context(struct ust_app_channel *ua_chan, ERR("UST app create channel context failed for app (pid: %d) " "with ret %d", app->pid, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app disable event failed. Application is dead."); } goto error; @@ -1040,6 +1118,12 @@ int set_ust_event_filter(struct ust_app_event *ua_event, ERR("UST app event %s filter failed for app (pid: %d) " "with ret %d", ua_event->attr.name, app->pid, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app filter event failed. Application is dead."); } goto error; @@ -1052,6 +1136,47 @@ error: return ret; } +/* + * Set event exclusions on the tracer. + */ +static +int set_ust_event_exclusion(struct ust_app_event *ua_event, + struct ust_app *app) +{ + int ret; + + health_code_update(); + + if (!ua_event->exclusion || !ua_event->exclusion->count) { + ret = 0; + goto error; + } + + ret = ustctl_set_exclusion(app->sock, ua_event->exclusion, + ua_event->obj); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app event %s exclusions failed for app (pid: %d) " + "with ret %d", ua_event->attr.name, app->pid, ret); + } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; + DBG3("UST app event exclusion failed. Application is dead."); + } + goto error; + } + + DBG2("UST exclusion set successfully for event %s", ua_event->name); + +error: + health_code_update(); + return ret; +} + /* * Disable the specified event on to UST tracer for the UST session. */ @@ -1069,6 +1194,12 @@ static int disable_ust_event(struct ust_app *app, "and session handle %d with ret %d", ua_event->attr.name, app->pid, ua_sess->handle, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app disable event failed. Application is dead."); } goto error; @@ -1099,6 +1230,12 @@ static int disable_ust_channel(struct ust_app *app, "and session handle %d with ret %d", ua_chan->name, app->pid, ua_sess->handle, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app disable channel failed. Application is dead."); } goto error; @@ -1129,6 +1266,12 @@ static int enable_ust_channel(struct ust_app *app, "and session handle %d with ret %d", ua_chan->name, app->pid, ua_sess->handle, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app enable channel failed. Application is dead."); } goto error; @@ -1161,6 +1304,12 @@ static int enable_ust_event(struct ust_app *app, "and session handle %d with ret %d", ua_event->attr.name, app->pid, ua_sess->handle, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app enable event failed. Application is dead."); } goto error; @@ -1241,6 +1390,12 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, ERR("Error ustctl create event %s for app pid: %d with ret %d", ua_event->attr.name, app->pid, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app create event failed. Application is dead."); } goto error; @@ -1261,6 +1416,14 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, } } + /* Set exclusions for the event */ + if (ua_event->exclusion) { + ret = set_ust_event_exclusion(ua_event, app); + if (ret < 0) { + goto error; + } + } + /* If event not enabled, disable it on the tracer */ if (ua_event->enabled == 0) { ret = disable_ust_event(app, ua_sess, ua_event); @@ -1296,6 +1459,8 @@ error: static void shadow_copy_event(struct ust_app_event *ua_event, struct ltt_ust_event *uevent) { + size_t exclusion_alloc_size; + strncpy(ua_event->name, uevent->attr.name, sizeof(ua_event->name)); ua_event->name[sizeof(ua_event->name) - 1] = '\0'; @@ -1309,6 +1474,16 @@ static void shadow_copy_event(struct ust_app_event *ua_event, ua_event->filter = alloc_copy_ust_app_filter(uevent->filter); /* Filter might be NULL here in case of ENONEM. */ } + + /* Copy exclusion data */ + if (uevent->exclusion) { + exclusion_alloc_size = sizeof(struct lttng_ust_event_exclusion) + + LTTNG_UST_SYM_NAME_LEN * uevent->exclusion->count; + ua_event->exclusion = zmalloc(exclusion_alloc_size); + if (ua_event->exclusion) { + memcpy(ua_event->exclusion, uevent->exclusion, exclusion_alloc_size); + } + } } /* @@ -1346,7 +1521,7 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, ua_chan->enabled = uchan->enabled; ua_chan->tracing_channel_id = uchan->id; - cds_lfht_for_each_entry(uchan->ctx->ht, &iter.iter, uctx, node.node) { + cds_list_for_each_entry(uctx, &uchan->ctx_list, list) { ua_ctx = alloc_ust_app_ctx(&uctx->ctx); if (ua_ctx == NULL) { continue; @@ -1354,12 +1529,13 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, lttng_ht_node_init_ulong(&ua_ctx->node, (unsigned long) ua_ctx->ctx.ctx); lttng_ht_add_unique_ulong(ua_chan->ctx, &ua_ctx->node); + cds_list_add_tail(&ua_ctx->list, &ua_chan->ctx_list); } /* Copy all events from ltt ust channel to ust app channel */ cds_lfht_for_each_entry(uchan->events->ht, &iter.iter, uevent, node.node) { ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel); + uevent->filter, uevent->attr.loglevel, uevent->exclusion); if (ua_event == NULL) { DBG2("UST event %s not found on shadow copy channel", uevent->attr.name); @@ -1407,6 +1583,8 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, ua_sess->bits_per_long = app->bits_per_long; /* There is only one consumer object per session possible. */ ua_sess->consumer = usess->consumer; + ua_sess->output_traces = usess->output_traces; + ua_sess->live_timer_interval = usess->live_timer_interval; switch (ua_sess->buffer_type) { case LTTNG_BUFFER_PER_PID: @@ -1470,7 +1648,7 @@ void __lookup_session_by_app(struct ltt_ust_session *usess, struct ust_app *app, struct lttng_ht_iter *iter) { /* Get right UST app session from app */ - lttng_ht_lookup(app->sessions, (void *)((unsigned long) usess->id), iter); + lttng_ht_lookup(app->sessions, &usess->id, iter); } /* @@ -1481,10 +1659,10 @@ static struct ust_app_session *lookup_session_by_app( struct ltt_ust_session *usess, struct ust_app *app) { struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; __lookup_session_by_app(usess, app, &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + node = lttng_ht_iter_get_node_u64(&iter); if (node == NULL) { goto error; } @@ -1634,7 +1812,7 @@ static int create_ust_app_session(struct ltt_ust_session *usess, ua_sess = lookup_session_by_app(usess, app); if (ua_sess == NULL) { - DBG2("UST app pid: %d session id %d not found, creating it", + DBG2("UST app pid: %d session id %" PRIu64 " not found, creating it", app->pid, usess->id); ua_sess = alloc_ust_app_session(app); if (ua_sess == NULL) { @@ -1677,6 +1855,13 @@ static int create_ust_app_session(struct ltt_ust_session *usess, app->pid, ret); } else { DBG("UST app creating session failed. Application is dead"); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. This will get flagged ENOTCONN and the + * caller will handle it. + */ + ret = 0; } delete_ust_app_session(-1, ua_sess, app); if (ret != -ENOMEM) { @@ -1692,9 +1877,9 @@ static int create_ust_app_session(struct ltt_ust_session *usess, ua_sess->handle = ret; /* Add ust app session to app's HT */ - lttng_ht_node_init_ulong(&ua_sess->node, - (unsigned long) ua_sess->tracing_id); - lttng_ht_add_unique_ulong(app->sessions, &ua_sess->node); + lttng_ht_node_init_u64(&ua_sess->node, + ua_sess->tracing_id); + lttng_ht_add_unique_u64(app->sessions, &ua_sess->node); DBG2("UST app session created successfully with handle %d", ret); } @@ -1745,6 +1930,7 @@ int create_ust_app_channel_context(struct ust_app_session *ua_sess, lttng_ht_node_init_ulong(&ua_ctx->node, (unsigned long) ua_ctx->ctx.ctx); lttng_ht_add_unique_ulong(ua_chan->ctx, &ua_ctx->node); + cds_list_add_tail(&ua_ctx->list, &ua_chan->ctx_list); ret = create_ust_channel_context(ua_chan, ua_ctx, app); if (ret < 0) { @@ -1831,7 +2017,7 @@ static int enable_ust_app_channel(struct ust_app_session *ua_sess, lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter); ua_chan_node = lttng_ht_iter_get_node_str(&iter); if (ua_chan_node == NULL) { - DBG2("Unable to find channel %s in ust session id %u", + DBG2("Unable to find channel %s in ust session id %" PRIu64, uchan->name, ua_sess->tracing_id); goto error; } @@ -1913,9 +2099,11 @@ static int do_consumer_create_channel(struct ltt_ust_session *usess, * Now get the channel from the consumer. This call wil populate the stream * list of that channel and set the ust objects. */ - ret = ust_consumer_get_channel(socket, ua_chan); - if (ret < 0) { - goto error_destroy; + if (usess->consumer->enabled) { + ret = ust_consumer_get_channel(socket, ua_chan); + if (ret < 0) { + goto error_destroy; + } } rcu_read_unlock(); @@ -2082,6 +2270,7 @@ static int create_buffer_reg_channel(struct buffer_reg_session *reg_sess, } assert(reg_chan); reg_chan->consumer_key = ua_chan->key; + reg_chan->subbuf_size = ua_chan->attr.subbuf_size; /* Create and add a channel registry to session. */ ret = ust_registry_channel_add(reg_sess->reg.ust, @@ -2246,6 +2435,14 @@ static int create_channel_per_uid(struct ust_app *app, ret = do_consumer_create_channel(usess, ua_sess, ua_chan, app->bits_per_long, reg_uid->registry->reg.ust); if (ret < 0) { + /* + * Let's remove the previously created buffer registry channel so + * it's not visible anymore in the session registry. + */ + ust_registry_channel_del_free(reg_uid->registry->reg.ust, + ua_chan->tracing_channel_id); + buffer_reg_channel_remove(reg_uid->registry, reg_chan); + buffer_reg_channel_destroy(reg_chan, LTTNG_DOMAIN_UST); goto error; } @@ -2404,7 +2601,7 @@ static int create_ust_app_channel(struct ust_app_session *ua_sess, if (ua_chan == NULL) { /* Only malloc can fail here */ ret = -ENOMEM; - goto error; + goto error_alloc; } shadow_copy_channel(ua_chan, uchan); @@ -2432,6 +2629,7 @@ end: error: delete_ust_app_channel(ua_chan->is_sent ? app->sock : -1, ua_chan, app); +error_alloc: return ret; } @@ -2450,7 +2648,7 @@ int create_ust_app_event(struct ust_app_session *ua_sess, /* Get event node */ ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel); + uevent->filter, uevent->attr.loglevel, uevent->exclusion); if (ua_event != NULL) { ret = -EEXIST; goto end; @@ -2508,8 +2706,8 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, registry = get_session_registry(ua_sess); assert(registry); - /* Metadata already exists for this registry. */ - if (registry->metadata_key) { + /* Metadata already exists for this registry or it was closed previously */ + if (registry->metadata_key || registry->metadata_closed) { ret = 0; goto error; } @@ -2527,8 +2725,8 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, metadata->attr.overwrite = DEFAULT_CHANNEL_OVERWRITE; metadata->attr.subbuf_size = default_get_metadata_subbuf_size(); metadata->attr.num_subbuf = DEFAULT_METADATA_SUBBUF_NUM; - metadata->attr.switch_timer_interval = DEFAULT_UST_CHANNEL_SWITCH_TIMER; - metadata->attr.read_timer_interval = DEFAULT_UST_CHANNEL_READ_TIMER; + metadata->attr.switch_timer_interval = DEFAULT_METADATA_SWITCH_TIMER; + metadata->attr.read_timer_interval = DEFAULT_METADATA_READ_TIMER; metadata->attr.output = LTTNG_UST_MMAP; metadata->attr.type = LTTNG_UST_CHAN_METADATA; } else { @@ -2568,10 +2766,8 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, ret = ust_consumer_ask_channel(ua_sess, metadata, consumer, socket, registry); if (ret < 0) { - /* - * Safe because the metadata obj pointer is not set so the delete below - * will not put a FD back again. - */ + /* Nullify the metadata key so we don't try to close it later on. */ + registry->metadata_key = 0; goto error_consumer; } @@ -2583,10 +2779,8 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, */ ret = consumer_setup_metadata(socket, metadata->key); if (ret < 0) { - /* - * Safe because the metadata obj pointer is not set so the delete below - * will not put a FD back again. - */ + /* Nullify the metadata key so we don't try to close it later on. */ + registry->metadata_key = 0; goto error_consumer; } @@ -2679,7 +2873,7 @@ struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) lta->v_major = msg->major; lta->v_minor = msg->minor; - lta->sessions = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + lta->sessions = lttng_ht_new(0, LTTNG_HT_TYPE_U64); lta->ust_objd = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); lta->notify_sock = -1; @@ -2842,15 +3036,18 @@ void ust_app_unregister(int sock) * session so the delete session will NOT push/close a second time. */ registry = get_session_registry(ua_sess); - if (registry) { + if (registry && !registry->metadata_closed) { /* Push metadata for application before freeing the application. */ (void) push_metadata(registry, ua_sess->consumer); /* * Don't ask to close metadata for global per UID buffers. Close - * metadata only on destroy trace session in this case. + * metadata only on destroy trace session in this case. Also, the + * previous push metadata could have flag the metadata registry to + * close so don't send a close command if closed. */ - if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID) { + if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID && + !registry->metadata_closed) { /* And ask to close it for this session registry. */ (void) close_metadata(registry, ua_sess->consumer); } @@ -2933,6 +3130,12 @@ int ust_app_list_events(struct lttng_event **events) app->sock, ret); } else { DBG3("UST app tp list get failed. Application is dead"); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. Continue normal execution. + */ + break; } goto rcu_error; } @@ -3027,6 +3230,12 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) app->sock, ret); } else { DBG3("UST app tp list field failed. Application is dead"); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + break; } goto rcu_error; } @@ -3111,9 +3320,9 @@ void ust_app_clean_list(void) rcu_read_unlock(); /* Destroy is done only when the ht is empty */ - lttng_ht_destroy(ust_app_ht); - lttng_ht_destroy(ust_app_ht_by_sock); - lttng_ht_destroy(ust_app_ht_by_notify_sock); + ht_cleanup_push(ust_app_ht); + ht_cleanup_push(ust_app_ht_by_sock); + ht_cleanup_push(ust_app_ht_by_notify_sock); } /* @@ -3145,7 +3354,7 @@ int ust_app_disable_channel_glb(struct ltt_ust_session *usess, goto error; } - DBG2("UST app disabling channel %s from global domain for session id %d", + DBG2("UST app disabling channel %s from global domain for session id %" PRIu64, uchan->name, usess->id); rcu_read_lock(); @@ -3206,7 +3415,7 @@ int ust_app_enable_channel_glb(struct ltt_ust_session *usess, goto error; } - DBG2("UST app enabling channel %s to global domain for session id %d", + DBG2("UST app enabling channel %s to global domain for session id %" PRIu64, uchan->name, usess->id); rcu_read_lock(); @@ -3254,7 +3463,8 @@ int ust_app_disable_event_glb(struct ltt_ust_session *usess, struct ust_app_event *ua_event; DBG("UST app disabling event %s for all apps in channel " - "%s for session id %d", uevent->attr.name, uchan->name, usess->id); + "%s for session id %" PRIu64, + uevent->attr.name, uchan->name, usess->id); rcu_read_lock(); @@ -3277,7 +3487,7 @@ int ust_app_disable_event_glb(struct ltt_ust_session *usess, lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); ua_chan_node = lttng_ht_iter_get_node_str(&uiter); if (ua_chan_node == NULL) { - DBG2("Channel %s not found in session id %d for app pid %d." + DBG2("Channel %s not found in session id %" PRIu64 " for app pid %d." "Skipping", uchan->name, usess->id, app->pid); continue; } @@ -3320,7 +3530,7 @@ int ust_app_disable_all_event_glb(struct ltt_ust_session *usess, struct ust_app_event *ua_event; DBG("UST app disabling all event for all apps in channel " - "%s for session id %d", uchan->name, usess->id); + "%s for session id %" PRIu64, uchan->name, usess->id); rcu_read_lock(); @@ -3378,7 +3588,7 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, assert(usess); assert(uchan); - DBG2("UST app adding channel %s to UST domain for session id %d", + DBG2("UST app adding channel %s to UST domain for session id %" PRIu64, uchan->name, usess->id); rcu_read_lock(); @@ -3458,7 +3668,7 @@ int ust_app_enable_event_glb(struct ltt_ust_session *usess, struct ust_app_channel *ua_chan; struct ust_app_event *ua_event; - DBG("UST app enabling event %s for all apps for session id %d", + DBG("UST app enabling event %s for all apps for session id %" PRIu64, uevent->attr.name, usess->id); /* @@ -3496,7 +3706,7 @@ int ust_app_enable_event_glb(struct ltt_ust_session *usess, /* Get event node */ ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel); + uevent->filter, uevent->attr.loglevel, uevent->exclusion); if (ua_event == NULL) { DBG3("UST app enable event %s not found for app PID %d." "Skipping app", uevent->attr.name, app->pid); @@ -3531,7 +3741,7 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, struct ust_app_session *ua_sess; struct ust_app_channel *ua_chan; - DBG("UST app creating event %s for all apps for session id %d", + DBG("UST app creating event %s for all apps for session id %" PRIu64, uevent->attr.name, usess->id); rcu_read_lock(); @@ -3641,6 +3851,13 @@ skip_setup: app->pid, ret); } else { DBG("UST app start session failed. Application is dead."); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + pthread_mutex_unlock(&ua_sess->lock); + goto end; } goto error_unlock; } @@ -3716,6 +3933,12 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) app->pid, ret); } else { DBG("UST app stop session failed. Application is dead."); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + goto end_unlock; } goto error_rcu_unlock; } @@ -3733,9 +3956,13 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) registry = get_session_registry(ua_sess); assert(registry); - /* Push metadata for application before freeing the application. */ - (void) push_metadata(registry, ua_sess->consumer); + if (!registry->metadata_closed) { + /* Push metadata for application before freeing the application. */ + (void) push_metadata(registry, ua_sess->consumer); + } + +end_unlock: pthread_mutex_unlock(&ua_sess->lock); end_no_session: rcu_read_unlock(); @@ -3790,8 +4017,11 @@ int ust_app_flush_trace(struct ltt_ust_session *usess, struct ust_app *app) } else { DBG3("UST app failed to flush %s. Application is dead.", ua_chan->name); - /* No need to continue. */ - break; + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ } /* Continuing flushing all buffers */ continue; @@ -3815,7 +4045,7 @@ static int destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) int ret; struct ust_app_session *ua_sess; struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; DBG("Destroy tracing for ust app pid %d", app->pid); @@ -3826,7 +4056,7 @@ static int destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) } __lookup_session_by_app(usess, app, &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + node = lttng_ht_iter_get_node_u64(&iter); if (node == NULL) { /* Session is being or is deleted. */ goto end; @@ -3897,7 +4127,7 @@ int ust_app_stop_trace_all(struct ltt_ust_session *usess) } } - /* Flush buffers */ + /* Flush buffers and push metadata (for UID buffers). */ switch (usess->buffer_type) { case LTTNG_BUFFER_PER_UID: { @@ -3905,6 +4135,7 @@ int ust_app_stop_trace_all(struct ltt_ust_session *usess) /* Flush all per UID buffers associated to that session. */ cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct ust_registry_session *ust_session_reg; struct buffer_reg_channel *reg_chan; struct consumer_socket *socket; @@ -3925,7 +4156,14 @@ int ust_app_stop_trace_all(struct ltt_ust_session *usess) */ (void) consumer_flush_channel(socket, reg_chan->consumer_key); } + + ust_session_reg = reg->registry->reg.ust; + if (!ust_session_reg->metadata_closed) { + /* Push metadata. */ + (void) push_metadata(ust_session_reg, usess->consumer); + } } + break; } case LTTNG_BUFFER_PER_PID: @@ -3979,7 +4217,7 @@ int ust_app_destroy_trace_all(struct ltt_ust_session *usess) void ust_app_global_update(struct ltt_ust_session *usess, int sock) { int ret = 0; - struct lttng_ht_iter iter, uiter, iter_ctx; + struct lttng_ht_iter iter, uiter; struct ust_app *app; struct ust_app_session *ua_sess = NULL; struct ust_app_channel *ua_chan; @@ -3989,12 +4227,12 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) assert(usess); assert(sock >= 0); - DBG2("UST app global update for app sock %d for session id %d", sock, + DBG2("UST app global update for app sock %d for session id %" PRIu64, sock, usess->id); rcu_read_lock(); - app = find_app_by_sock(sock); + app = ust_app_find_by_sock(sock); if (app == NULL) { /* * Application can be unregistered before so this is possible hence @@ -4051,8 +4289,11 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) } } - cds_lfht_for_each_entry(ua_chan->ctx->ht, &iter_ctx.iter, ua_ctx, - node.node) { + /* + * Add context using the list so they are enabled in the same order the + * user added them. + */ + cds_list_for_each_entry(ua_ctx, &ua_chan->ctx_list, list) { ret = create_ust_channel_context(ua_chan, ua_ctx, app); if (ret < 0) { goto error_unlock; @@ -4191,7 +4432,7 @@ int ust_app_enable_event_pid(struct ltt_ust_session *usess, ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel); + uevent->filter, uevent->attr.loglevel, uevent->exclusion); if (ua_event == NULL) { ret = create_ust_app_event(ua_sess, ua_chan, uevent, app); if (ret < 0) { @@ -4425,12 +4666,19 @@ static int reply_ust_register_channel(int sock, int sobjd, int cobjd, DBG("Application socket %d is being teardown. Abort event notify", sock); ret = 0; + free(fields); goto error_rcu_unlock; } - /* Lookup channel by UST object descriptor. Should always be found. */ + /* Lookup channel by UST object descriptor. */ ua_chan = find_channel_by_objd(app, cobjd); - assert(ua_chan); + if (!ua_chan) { + DBG("Application channel is being teardown. Abort event notify"); + ret = 0; + free(fields); + goto error_rcu_unlock; + } + assert(ua_chan->session); ua_sess = ua_chan->session; @@ -4464,6 +4712,9 @@ static int reply_ust_register_channel(int sock, int sobjd, int cobjd, } else { /* Get current already assigned values. */ type = chan_reg->header_type; + free(fields); + /* Set to NULL so the error path does not do a double free. */ + fields = NULL; } /* Channel id is set during the object creation. */ chan_id = chan_reg->chan_id; @@ -4499,6 +4750,9 @@ error: pthread_mutex_unlock(®istry->lock); error_rcu_unlock: rcu_read_unlock(); + if (ret) { + free(fields); + } return ret; } @@ -4531,12 +4785,23 @@ static int add_event_ust_registry(int sock, int sobjd, int cobjd, char *name, DBG("Application socket %d is being teardown. Abort event notify", sock); ret = 0; + free(sig); + free(fields); + free(model_emf_uri); goto error_rcu_unlock; } - /* Lookup channel by UST object descriptor. Should always be found. */ + /* Lookup channel by UST object descriptor. */ ua_chan = find_channel_by_objd(app, cobjd); - assert(ua_chan); + if (!ua_chan) { + DBG("Application channel is being teardown. Abort event notify"); + ret = 0; + free(sig); + free(fields); + free(model_emf_uri); + goto error_rcu_unlock; + } + assert(ua_chan->session); ua_sess = ua_chan->session; @@ -4551,9 +4816,15 @@ static int add_event_ust_registry(int sock, int sobjd, int cobjd, char *name, pthread_mutex_lock(®istry->lock); + /* + * From this point on, this call acquires the ownership of the sig, fields + * and model_emf_uri meaning any free are done inside it if needed. These + * three variables MUST NOT be read/write after this. + */ ret_code = ust_registry_create_event(registry, chan_reg_key, sobjd, cobjd, name, sig, nr_fields, fields, loglevel, - model_emf_uri, ua_sess->buffer_type, &event_id); + model_emf_uri, ua_sess->buffer_type, &event_id, + app); /* * The return value is returned to ustctl so in case of an error, the @@ -4627,7 +4898,12 @@ int ust_app_recv_notify(int sock) goto error; } - /* Add event to the UST registry coming from the notify socket. */ + /* + * Add event to the UST registry coming from the notify socket. This + * call will free if needed the sig, fields and model_emf_uri. This + * code path loses the ownsership of these variables and transfer them + * to the this function. + */ ret = add_event_ust_registry(sock, sobjd, cobjd, name, sig, nr_fields, fields, loglevel, model_emf_uri); if (ret < 0) { @@ -4655,6 +4931,11 @@ int ust_app_recv_notify(int sock) goto error; } + /* + * The fields ownership are transfered to this function call meaning + * that if needed it will be freed. After this, it's invalid to access + * fields or clean it up. + */ ret = reply_ust_register_channel(sock, sobjd, cobjd, nr_fields, fields); if (ret < 0) { @@ -4748,3 +5029,235 @@ close_socket: call_rcu(&obj->head, close_notify_sock_rcu); } } + +/* + * Destroy a ust app data structure and free its memory. + */ +void ust_app_destroy(struct ust_app *app) +{ + if (!app) { + return; + } + + call_rcu(&app->pid_n.head, delete_ust_app_rcu); +} + +/* + * Take a snapshot for a given UST session. The snapshot is sent to the given + * output. + * + * Return 0 on success or else a negative value. + */ +int ust_app_snapshot_record(struct ltt_ust_session *usess, + struct snapshot_output *output, int wait, unsigned int nb_streams) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + char pathname[PATH_MAX]; + uint64_t max_stream_size = 0; + + assert(usess); + assert(output); + + rcu_read_lock(); + + /* + * Compute the maximum size of a single stream if a max size is asked by + * the caller. + */ + if (output->max_size > 0 && nb_streams > 0) { + max_stream_size = output->max_size / nb_streams; + } + + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + struct buffer_reg_uid *reg; + + cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct buffer_reg_channel *reg_chan; + struct consumer_socket *socket; + + /* Get consumer socket to use to push the metadata.*/ + socket = consumer_find_socket_by_bitness(reg->bits_per_long, + usess->consumer); + if (!socket) { + ret = -EINVAL; + goto error; + } + + memset(pathname, 0, sizeof(pathname)); + ret = snprintf(pathname, sizeof(pathname), + DEFAULT_UST_TRACE_DIR "/" DEFAULT_UST_TRACE_UID_PATH, + reg->uid, reg->bits_per_long); + if (ret < 0) { + PERROR("snprintf snapshot path"); + goto error; + } + + /* Add the UST default trace dir to path. */ + cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, + reg_chan, node.node) { + + /* + * Make sure the maximum stream size is not lower than the + * subbuffer size or else it's an error since we won't be able to + * snapshot anything. + */ + if (max_stream_size && + reg_chan->subbuf_size > max_stream_size) { + ret = -EINVAL; + DBG3("UST app snapshot record maximum stream size %" PRIu64 + " is smaller than subbuffer size of %zu", + max_stream_size, reg_chan->subbuf_size); + goto error; + } + ret = consumer_snapshot_channel(socket, reg_chan->consumer_key, output, 0, + usess->uid, usess->gid, pathname, wait, + max_stream_size); + if (ret < 0) { + goto error; + } + } + ret = consumer_snapshot_channel(socket, reg->registry->reg.ust->metadata_key, output, + 1, usess->uid, usess->gid, pathname, wait, + max_stream_size); + if (ret < 0) { + goto error; + } + } + break; + } + case LTTNG_BUFFER_PER_PID: + { + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct consumer_socket *socket; + struct lttng_ht_iter chan_iter; + struct ust_app_channel *ua_chan; + struct ust_app_session *ua_sess; + struct ust_registry_session *registry; + + ua_sess = lookup_session_by_app(usess, app); + if (!ua_sess) { + /* Session not associated with this app. */ + continue; + } + + /* Get the right consumer socket for the application. */ + socket = consumer_find_socket_by_bitness(app->bits_per_long, + output->consumer); + if (!socket) { + ret = -EINVAL; + goto error; + } + + /* Add the UST default trace dir to path. */ + memset(pathname, 0, sizeof(pathname)); + ret = snprintf(pathname, sizeof(pathname), DEFAULT_UST_TRACE_DIR "/%s", + ua_sess->path); + if (ret < 0) { + PERROR("snprintf snapshot path"); + goto error; + } + + cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, + ua_chan, node.node) { + /* + * Make sure the maximum stream size is not lower than the + * subbuffer size or else it's an error since we won't be able to + * snapshot anything. + */ + if (max_stream_size && + ua_chan->attr.subbuf_size > max_stream_size) { + ret = -EINVAL; + DBG3("UST app snapshot record maximum stream size %" PRIu64 + " is smaller than subbuffer size of %" PRIu64, + max_stream_size, ua_chan->attr.subbuf_size); + goto error; + } + + ret = consumer_snapshot_channel(socket, ua_chan->key, output, 0, + ua_sess->euid, ua_sess->egid, pathname, wait, + max_stream_size); + if (ret < 0) { + goto error; + } + } + + registry = get_session_registry(ua_sess); + assert(registry); + ret = consumer_snapshot_channel(socket, registry->metadata_key, output, + 1, ua_sess->euid, ua_sess->egid, pathname, wait, + max_stream_size); + if (ret < 0) { + goto error; + } + } + break; + } + default: + assert(0); + break; + } + +error: + rcu_read_unlock(); + return ret; +} + +/* + * Return the number of streams for a UST session. + */ +unsigned int ust_app_get_nb_stream(struct ltt_ust_session *usess) +{ + unsigned int ret = 0; + struct ust_app *app; + struct lttng_ht_iter iter; + + assert(usess); + + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + struct buffer_reg_uid *reg; + + cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct buffer_reg_channel *reg_chan; + + cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, + reg_chan, node.node) { + ret += reg_chan->stream_count; + } + } + break; + } + case LTTNG_BUFFER_PER_PID: + { + rcu_read_lock(); + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct ust_app_channel *ua_chan; + struct ust_app_session *ua_sess; + struct lttng_ht_iter chan_iter; + + ua_sess = lookup_session_by_app(usess, app); + if (!ua_sess) { + /* Session not associated with this app. */ + continue; + } + + cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, + ua_chan, node.node) { + ret += ua_chan->streams.count; + } + } + rcu_read_unlock(); + break; + } + default: + assert(0); + break; + } + + return ret; +}