X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fust-app.c;h=97083e15a95f7c5b1c919143e9724ff51fd4cbcc;hp=223fb77c0c936e0b81d678e4787a2c5f46e80bdc;hb=766048521f3bc7beb1209e2bb939ca2b42532a52;hpb=5c786dedd0156b93984f89ba47ec841277ed7dae diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c index 223fb77c0..97083e15a 100644 --- a/src/bin/lttng-sessiond/ust-app.c +++ b/src/bin/lttng-sessiond/ust-app.c @@ -1,5 +1,6 @@ /* * Copyright (C) 2011 - David Goulet + * Copyright (C) 2016 - Jérémie Galarneau * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License, version 2 only, @@ -15,7 +16,7 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE +#define _LGPL_SOURCE #include #include #include @@ -34,11 +35,18 @@ #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" +#include "session.h" +#include "lttng-sessiond.h" +#include "notification-thread-commands.h" +#include "rotate.h" + +static +int ust_app_flush_app_session(struct ust_app *app, struct ust_app_session *ua_sess); /* Next available channel key. Access under next_channel_key_lock. */ static uint64_t _next_channel_key; @@ -85,6 +93,7 @@ static void copy_channel_attr_to_ustctl( attr->switch_timer_interval = uattr->switch_timer_interval; attr->read_timer_interval = uattr->read_timer_interval; attr->output = uattr->output; + attr->blocking_timeout = uattr->u.s.blocking_timeout; } /* @@ -97,14 +106,16 @@ static int ht_match_ust_app_event(struct cds_lfht_node *node, const void *_key) { struct ust_app_event *event; const struct ust_app_ht_key *key; + int ev_loglevel_value; assert(node); assert(_key); event = caa_container_of(node, struct ust_app_event, node.node); key = _key; + ev_loglevel_value = event->attr.loglevel; - /* 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) { @@ -112,9 +123,10 @@ static int ht_match_ust_app_event(struct cds_lfht_node *node, const void *_key) } /* Event loglevel. */ - if (event->attr.loglevel != key->loglevel) { + if (ev_loglevel_value != key->loglevel_type) { if (event->attr.loglevel_type == LTTNG_UST_LOGLEVEL_ALL - && key->loglevel == 0 && event->attr.loglevel == -1) { + && key->loglevel_type == 0 && + ev_loglevel_value == -1) { /* * Match is accepted. This is because on event creation, the * loglevel is set to -1 if the event loglevel type is ALL so 0 and @@ -140,6 +152,21 @@ 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; @@ -165,7 +192,8 @@ static void add_unique_ust_app_event(struct ust_app_channel *ua_chan, ht = ua_chan->events; key.name = event->attr.name; key.filter = event->filter; - key.loglevel = event->attr.loglevel; + key.loglevel_type = event->attr.loglevel; + key.exclusion = event->exclusion; node_ptr = cds_lfht_add_unique(ht->ht, ht->hash_fct(event->node.key, lttng_ht_seed), @@ -242,14 +270,17 @@ error: * this function. */ static -void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx) +void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx, + struct ust_app *app) { int ret; assert(ua_ctx); if (ua_ctx->obj) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_object(sock, ua_ctx->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release ctx obj handle %d failed with ret %d", sock, ua_ctx->obj->handle, ret); @@ -264,16 +295,20 @@ void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx) * this function. */ static -void delete_ust_app_event(int sock, struct ust_app_event *ua_event) +void delete_ust_app_event(int sock, struct ust_app_event *ua_event, + struct ust_app *app) { int ret; assert(ua_event); free(ua_event->filter); - + if (ua_event->exclusion != NULL) + free(ua_event->exclusion); if (ua_event->obj != NULL) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_object(sock, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release event obj failed with ret %d", sock, ret); @@ -288,14 +323,17 @@ void delete_ust_app_event(int sock, struct ust_app_event *ua_event) * * Return 0 on success or else a negative value. */ -static int release_ust_app_stream(int sock, struct ust_app_stream *stream) +static int release_ust_app_stream(int sock, struct ust_app_stream *stream, + struct ust_app *app) { int ret = 0; assert(stream); if (stream->obj) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_object(sock, stream->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release stream obj failed with ret %d", sock, ret); @@ -312,11 +350,12 @@ static int release_ust_app_stream(int sock, struct ust_app_stream *stream) * this function. */ static -void delete_ust_app_stream(int sock, struct ust_app_stream *stream) +void delete_ust_app_stream(int sock, struct ust_app_stream *stream, + struct ust_app *app) { assert(stream); - (void) release_ust_app_stream(sock, stream); + (void) release_ust_app_stream(sock, stream, app); free(stream); } @@ -337,9 +376,74 @@ void delete_ust_app_channel_rcu(struct rcu_head *head) free(ua_chan); } +/* + * Extract the lost packet or discarded events counter when the channel is + * being deleted and store the value in the parent channel so we can + * access it from lttng list and at stop/destroy. + * + * The session list lock must be held by the caller. + */ +static +void save_per_pid_lost_discarded_counters(struct ust_app_channel *ua_chan) +{ + uint64_t discarded = 0, lost = 0; + struct ltt_session *session; + struct ltt_ust_channel *uchan; + + if (ua_chan->attr.type != LTTNG_UST_CHAN_PER_CPU) { + return; + } + + rcu_read_lock(); + session = session_find_by_id(ua_chan->session->tracing_id); + if (!session || !session->ust_session) { + /* + * Not finding the session is not an error because there are + * multiple ways the channels can be torn down. + * + * 1) The session daemon can initiate the destruction of the + * ust app session after receiving a destroy command or + * during its shutdown/teardown. + * 2) The application, since we are in per-pid tracing, is + * unregistering and tearing down its ust app session. + * + * Both paths are protected by the session list lock which + * ensures that the accounting of lost packets and discarded + * events is done exactly once. The session is then unpublished + * from the session list, resulting in this condition. + */ + goto end; + } + + if (ua_chan->attr.overwrite) { + consumer_get_lost_packets(ua_chan->session->tracing_id, + ua_chan->key, session->ust_session->consumer, + &lost); + } else { + consumer_get_discarded_events(ua_chan->session->tracing_id, + ua_chan->key, session->ust_session->consumer, + &discarded); + } + uchan = trace_ust_find_channel_by_name( + session->ust_session->domain_global.channels, + ua_chan->name); + if (!uchan) { + ERR("Missing UST channel to store discarded counters"); + goto end; + } + + uchan->per_pid_closed_app_discarded += discarded; + uchan->per_pid_closed_app_lost += lost; + +end: + rcu_read_unlock(); +} + /* * Delete ust app channel safely. RCU read lock must be held before calling * this function. + * + * The session list lock must be held by the caller. */ static void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, @@ -359,14 +463,15 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, /* Wipe stream */ cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) { cds_list_del(&stream->list); - delete_ust_app_stream(sock, stream); + delete_ust_app_stream(sock, stream, app); } /* 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); + delete_ust_app_ctx(sock, ua_ctx, app); } /* Wipe events */ @@ -374,22 +479,27 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, node.node) { ret = lttng_ht_del(ua_chan->events, &iter); assert(!ret); - delete_ust_app_event(sock, ua_event); + delete_ust_app_event(sock, ua_event, app); } 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); + ust_registry_channel_del_free(registry, ua_chan->key, + true); } + save_per_pid_lost_discarded_counters(ua_chan); } if (ua_chan->obj != NULL) { /* Remove channel from application UST object descriptor. */ iter.iter.node = &ua_chan->ust_objd_node.node; - lttng_ht_del(app->ust_objd, &iter); + ret = lttng_ht_del(app->ust_objd, &iter); + assert(!ret); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_object(sock, ua_chan->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release channel obj failed with ret %d", sock, ret); @@ -400,42 +510,71 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, call_rcu(&ua_chan->rcu_head, delete_ust_app_channel_rcu); } +int ust_app_register_done(struct ust_app *app) +{ + int ret; + + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_register_done(app->sock); + pthread_mutex_unlock(&app->sock_lock); + return ret; +} + +int ust_app_release_object(struct ust_app *app, struct lttng_ust_object_data *data) +{ + int ret, sock; + + if (app) { + pthread_mutex_lock(&app->sock_lock); + sock = app->sock; + } else { + sock = -1; + } + ret = ustctl_release_object(sock, data); + if (app) { + pthread_mutex_unlock(&app->sock_lock); + } + return ret; +} + /* * Push metadata to consumer socket. * - * The socket lock MUST be acquired. - * The ust app session lock MUST be acquired. + * RCU read-side lock must be held to guarantee existance of socket. + * Must be called with the ust app session lock held. + * Must be called with the registry lock held. * * On success, return the len of metadata pushed or else a negative value. + * Returning a -EPIPE return value means we could not send the metadata, + * but it can be caused by recoverable errors (e.g. the application has + * terminated concurrently). */ ssize_t ust_app_push_metadata(struct ust_registry_session *registry, struct consumer_socket *socket, int send_zero_data) { int ret; char *metadata_str = NULL; - size_t len, offset; + size_t len, offset, new_metadata_len_sent; ssize_t ret_val; + uint64_t metadata_key, metadata_version; assert(registry); assert(socket); + metadata_key = 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. + * Means that no metadata was assigned to the session. This can + * happens if no start has been done previously. */ - if (registry->metadata_closed) { - return -EPIPE; + if (!metadata_key) { + return 0; } - pthread_mutex_lock(®istry->lock); - offset = registry->metadata_len_sent; len = registry->metadata_len - registry->metadata_len_sent; + new_metadata_len_sent = registry->metadata_len; + metadata_version = registry->metadata_version; if (len == 0) { DBG3("No metadata to push for metadata key %" PRIu64, registry->metadata_key); @@ -454,38 +593,99 @@ ssize_t ust_app_push_metadata(struct ust_registry_session *registry, ret_val = -ENOMEM; goto error; } - /* Copy what we haven't send out. */ + /* Copy what we haven't sent out. */ memcpy(metadata_str, registry->metadata + offset, len); - registry->metadata_len_sent += len; push_data: pthread_mutex_unlock(®istry->lock); - ret = consumer_push_metadata(socket, registry->metadata_key, - metadata_str, len, offset); + /* + * We need to unlock the registry while we push metadata to + * break a circular dependency between the consumerd metadata + * lock and the sessiond registry lock. Indeed, pushing metadata + * to the consumerd awaits that it gets pushed all the way to + * relayd, but doing so requires grabbing the metadata lock. If + * a concurrent metadata request is being performed by + * consumerd, this can try to grab the registry lock on the + * sessiond while holding the metadata lock on the consumer + * daemon. Those push and pull schemes are performed on two + * different bidirectionnal communication sockets. + */ + ret = consumer_push_metadata(socket, metadata_key, + metadata_str, len, offset, metadata_version); + pthread_mutex_lock(®istry->lock); if (ret < 0) { + /* + * There is an acceptable race here between the registry + * metadata key assignment and the creation on the + * consumer. The session daemon can concurrently push + * metadata for this registry while being created on the + * consumer since the metadata key of the registry is + * assigned *before* it is setup to avoid the consumer + * to ask for metadata that could possibly be not found + * in the session daemon. + * + * The metadata will get pushed either by the session + * being stopped or the consumer requesting metadata if + * that race is triggered. + */ + if (ret == -LTTCOMM_CONSUMERD_CHANNEL_FAIL) { + ret = 0; + } else { + ERR("Error pushing metadata to consumer"); + } ret_val = ret; goto error_push; + } else { + /* + * Metadata may have been concurrently pushed, since + * we're not holding the registry lock while pushing to + * consumer. This is handled by the fact that we send + * the metadata content, size, and the offset at which + * that metadata belongs. This may arrive out of order + * on the consumer side, and the consumer is able to + * deal with overlapping fragments. The consumer + * supports overlapping fragments, which must be + * contiguous starting from offset 0. We keep the + * largest metadata_len_sent value of the concurrent + * send. + */ + registry->metadata_len_sent = + max_t(size_t, registry->metadata_len_sent, + new_metadata_len_sent); } - free(metadata_str); return len; end: error: - pthread_mutex_unlock(®istry->lock); + if (ret_val) { + /* + * 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; + } error_push: free(metadata_str); return ret_val; } /* - * For a given application and session, push metadata to consumer. The session - * lock MUST be acquired here before calling this. + * For a given application and session, push metadata to consumer. * Either sock or consumer is required : if sock is NULL, the default * socket to send the metadata is retrieved from consumer, if sock * is not NULL we use it to send the metadata. + * RCU read-side lock must be held while calling this function, + * therefore ensuring existance of registry. It also ensures existance + * of socket throughout this function. * * Return 0 on success else a negative error. + * Returning a -EPIPE return value means we could not send the metadata, + * but it can be caused by recoverable errors (e.g. the application has + * terminated concurrently). */ static int push_metadata(struct ust_registry_session *registry, struct consumer_output *consumer) @@ -497,15 +697,10 @@ static int push_metadata(struct ust_registry_session *registry, assert(registry); assert(consumer); - rcu_read_lock(); - - /* - * Means that no metadata was assigned to the session. This can happens if - * no start has been done previously. - */ - if (!registry->metadata_key) { - ret_val = 0; - goto end_rcu_unlock; + pthread_mutex_lock(®istry->lock); + if (registry->metadata_closed) { + ret_val = -EPIPE; + goto error; } /* Get consumer socket to use to push the metadata.*/ @@ -513,46 +708,26 @@ static int push_metadata(struct ust_registry_session *registry, consumer); if (!socket) { ret_val = -1; - goto error_rcu_unlock; + goto error; } - /* - * TODO: Currently, we hold the socket lock around sampling of the next - * metadata segment to ensure we send metadata over the consumer socket in - * the correct order. This makes the registry lock nest inside the socket - * lock. - * - * Please note that this is a temporary measure: we should move this lock - * back into ust_consumer_push_metadata() when the consumer gets the - * ability to reorder the metadata it receives. - */ - pthread_mutex_lock(socket->lock); ret = ust_app_push_metadata(registry, socket, 0); - pthread_mutex_unlock(socket->lock); if (ret < 0) { ret_val = ret; - goto error_rcu_unlock; + goto error; } - - rcu_read_unlock(); + pthread_mutex_unlock(®istry->lock); 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(); +error: + pthread_mutex_unlock(®istry->lock); return ret_val; } /* * Send to the consumer a close metadata command for the given session. Once * done, the metadata channel is deleted and the session metadata pointer is - * nullified. The session lock MUST be acquired here unless the application is + * nullified. The session lock MUST be held unless the application is * in the destroy path. * * Return 0 on success else a negative value. @@ -568,6 +743,8 @@ static int close_metadata(struct ust_registry_session *registry, rcu_read_lock(); + pthread_mutex_lock(®istry->lock); + if (!registry->metadata_key || registry->metadata_closed) { ret = 0; goto end; @@ -594,6 +771,7 @@ error: */ registry->metadata_closed = 1; end: + pthread_mutex_unlock(®istry->lock); rcu_read_unlock(); return ret; } @@ -617,6 +795,8 @@ void delete_ust_app_session_rcu(struct rcu_head *head) /* * Delete ust app session safely. RCU read lock must be held before calling * this function. + * + * The session list lock must be held by the caller. */ static void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, @@ -631,8 +811,12 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, pthread_mutex_lock(&ua_sess->lock); + assert(!ua_sess->deleted); + ua_sess->deleted = true; + registry = get_session_registry(ua_sess); - if (registry && !registry->metadata_closed) { + /* Registry can be null on error path during initialization. */ + if (registry) { /* Push metadata for application before freeing the application. */ (void) push_metadata(registry, ua_sess->consumer); @@ -642,8 +826,7 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, * 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 && - !registry->metadata_closed) { + if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID) { /* And ask to close it for this session registry. */ (void) close_metadata(registry, ua_sess->consumer); } @@ -660,20 +843,33 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, if (ua_sess->buffer_type == LTTNG_BUFFER_PER_PID) { struct buffer_reg_pid *reg_pid = buffer_reg_pid_find(ua_sess->id); if (reg_pid) { + /* + * Registry can be null on error path during + * initialization. + */ buffer_reg_pid_remove(reg_pid); buffer_reg_pid_destroy(reg_pid); } } if (ua_sess->handle != -1) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_handle(sock, ua_sess->handle); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release session handle failed with ret %d", sock, ret); } + /* Remove session from application UST object descriptor. */ + iter.iter.node = &ua_sess->ust_objd_node.node; + ret = lttng_ht_del(app->ust_sessions_objd, &iter); + assert(!ret); } + pthread_mutex_unlock(&ua_sess->lock); + consumer_output_put(ua_sess->consumer); + call_rcu(&ua_sess->rcu_head, delete_ust_app_session_rcu); } @@ -689,6 +885,11 @@ void delete_ust_app(struct ust_app *app) int ret, sock; struct ust_app_session *ua_sess, *tmp_ua_sess; + /* + * The session list lock must be held during this function to guarantee + * the existence of ua_sess. + */ + session_lock_list(); /* Delete ust app sessions info */ sock = app->sock; app->sock = -1; @@ -703,6 +904,7 @@ void delete_ust_app(struct ust_app *app) } ht_cleanup_push(app->sessions); + ht_cleanup_push(app->ust_sessions_objd); ht_cleanup_push(app->ust_objd); /* @@ -726,6 +928,7 @@ void delete_ust_app(struct ust_app *app) DBG2("UST app pid %d deleted", app->pid); free(app); + session_unlock_list(); } /* @@ -746,6 +949,8 @@ void delete_ust_app_rcu(struct rcu_head *head) /* * Delete the session from the application ht and delete the data structure by * freeing every object inside and releasing them. + * + * The session list lock must be held by the caller. */ static void destroy_app_session(struct ust_app *app, struct ust_app_session *ua_sess) @@ -787,6 +992,7 @@ struct ust_app_session *alloc_ust_app_session(struct ust_app *app) ua_sess->handle = -1; ua_sess->channels = lttng_ht_new(0, LTTNG_HT_TYPE_STRING); + ua_sess->metadata_attr.type = LTTNG_UST_CHAN_METADATA; pthread_mutex_init(&ua_sess->lock, NULL); return ua_sess; @@ -825,6 +1031,7 @@ 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) { @@ -835,6 +1042,7 @@ struct ust_app_channel *alloc_ust_app_channel(char *name, ua_chan->attr.switch_timer_interval = attr->switch_timer_interval; ua_chan->attr.read_timer_interval = attr->read_timer_interval; ua_chan->attr.output = attr->output; + ua_chan->attr.blocking_timeout = attr->u.s.blocking_timeout; } /* By default, the channel is a per cpu channel. */ ua_chan->attr.type = LTTNG_UST_CHAN_PER_CPU; @@ -907,7 +1115,7 @@ error: * Alloc new UST app context. */ static -struct ust_app_ctx *alloc_ust_app_ctx(struct lttng_ust_context *uctx) +struct ust_app_ctx *alloc_ust_app_ctx(struct lttng_ust_context_attr *uctx) { struct ust_app_ctx *ua_ctx; @@ -916,14 +1124,31 @@ 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)); + if (uctx->ctx == LTTNG_UST_CONTEXT_APP_CONTEXT) { + char *provider_name = NULL, *ctx_name = NULL; + + provider_name = strdup(uctx->u.app_ctx.provider_name); + ctx_name = strdup(uctx->u.app_ctx.ctx_name); + if (!provider_name || !ctx_name) { + free(provider_name); + free(ctx_name); + goto error; + } + + ua_ctx->ctx.u.app_ctx.provider_name = provider_name; + ua_ctx->ctx.u.app_ctx.ctx_name = ctx_name; + } } DBG3("UST app context %d allocated", ua_ctx->ctx.ctx); - -error: return ua_ctx; +error: + free(ua_ctx); + return NULL; } /* @@ -931,15 +1156,15 @@ error: * * Return allocated filter or NULL on error. */ -static struct lttng_ust_filter_bytecode *alloc_copy_ust_app_filter( - struct lttng_ust_filter_bytecode *orig_f) +static struct lttng_filter_bytecode *copy_filter_bytecode( + struct lttng_filter_bytecode *orig_f) { - struct lttng_ust_filter_bytecode *filter = NULL; + struct lttng_filter_bytecode *filter = NULL; /* Copy filter bytecode */ filter = zmalloc(sizeof(*filter) + orig_f->len); if (!filter) { - PERROR("zmalloc alloc ust app filter"); + PERROR("zmalloc alloc filter bytecode"); goto error; } @@ -949,12 +1174,35 @@ error: return filter; } +/* + * Create a liblttng-ust filter bytecode from given bytecode. + * + * Return allocated filter or NULL on error. + */ +static struct lttng_ust_filter_bytecode *create_ust_bytecode_from_bytecode( + struct lttng_filter_bytecode *orig_f) +{ + struct lttng_ust_filter_bytecode *filter = NULL; + + /* Copy filter bytecode */ + filter = zmalloc(sizeof(*filter) + orig_f->len); + if (!filter) { + PERROR("zmalloc alloc ust filter bytecode"); + goto error; + } + + assert(sizeof(struct lttng_filter_bytecode) == + sizeof(struct lttng_ust_filter_bytecode)); + memcpy(filter, orig_f, sizeof(*filter) + orig_f->len); +error: + return filter; +} + /* * 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; @@ -1002,7 +1250,9 @@ 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_filter_bytecode *filter, + int loglevel_value, + const struct lttng_event_exclusion *exclusion) { struct lttng_ht_iter iter; struct lttng_ht_node_str *node; @@ -1015,7 +1265,9 @@ static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht, /* Setup key for event lookup. */ key.name = name; key.filter = filter; - key.loglevel = loglevel; + key.loglevel_type = loglevel_value; + /* lttng_event_exclusion and lttng_ust_event_exclusion structures are similar */ + key.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), @@ -1044,13 +1296,21 @@ int create_ust_channel_context(struct ust_app_channel *ua_chan, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_add_context(app->sock, &ua_ctx->ctx, ua_chan->obj, &ua_ctx->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { 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; @@ -1074,6 +1334,7 @@ int set_ust_event_filter(struct ust_app_event *ua_event, struct ust_app *app) { int ret; + struct lttng_ust_filter_bytecode *ust_bytecode = NULL; health_code_update(); @@ -1082,13 +1343,26 @@ int set_ust_event_filter(struct ust_app_event *ua_event, goto error; } - ret = ustctl_set_filter(app->sock, ua_event->filter, + ust_bytecode = create_ust_bytecode_from_bytecode(ua_event->filter); + if (!ust_bytecode) { + ret = -LTTNG_ERR_NOMEM; + goto error; + } + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_set_filter(app->sock, ust_bytecode, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { 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; @@ -1098,6 +1372,78 @@ int set_ust_event_filter(struct ust_app_event *ua_event, error: health_code_update(); + free(ust_bytecode); + return ret; +} + +static +struct lttng_ust_event_exclusion *create_ust_exclusion_from_exclusion( + struct lttng_event_exclusion *exclusion) +{ + struct lttng_ust_event_exclusion *ust_exclusion = NULL; + size_t exclusion_alloc_size = sizeof(struct lttng_ust_event_exclusion) + + LTTNG_UST_SYM_NAME_LEN * exclusion->count; + + ust_exclusion = zmalloc(exclusion_alloc_size); + if (!ust_exclusion) { + PERROR("malloc"); + goto end; + } + + assert(sizeof(struct lttng_event_exclusion) == + sizeof(struct lttng_ust_event_exclusion)); + memcpy(ust_exclusion, exclusion, exclusion_alloc_size); +end: + return ust_exclusion; +} + +/* + * Set event exclusions on the tracer. + */ +static +int set_ust_event_exclusion(struct ust_app_event *ua_event, + struct ust_app *app) +{ + int ret; + struct lttng_ust_event_exclusion *ust_exclusion = NULL; + + health_code_update(); + + if (!ua_event->exclusion || !ua_event->exclusion->count) { + ret = 0; + goto error; + } + + ust_exclusion = create_ust_exclusion_from_exclusion( + ua_event->exclusion); + if (!ust_exclusion) { + ret = -LTTNG_ERR_NOMEM; + goto error; + } + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_set_exclusion(app->sock, ust_exclusion, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); + 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(); + free(ust_exclusion); return ret; } @@ -1111,13 +1457,21 @@ static int disable_ust_event(struct ust_app *app, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_disable(app->sock, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app event %s disable failed for app (pid: %d) " "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; @@ -1141,13 +1495,21 @@ static int disable_ust_channel(struct ust_app *app, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_disable(app->sock, ua_chan->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app channel %s disable failed for app (pid: %d) " "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; @@ -1171,13 +1533,21 @@ static int enable_ust_channel(struct ust_app *app, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_enable(app->sock, ua_chan->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app channel %s enable failed for app (pid: %d) " "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; @@ -1203,13 +1573,21 @@ static int enable_ust_event(struct ust_app *app, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_enable(app->sock, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app event %s enable failed for app (pid: %d) " "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; @@ -1245,7 +1623,10 @@ static int send_channel_pid_to_ust(struct ust_app *app, /* Send channel to the application. */ ret = ust_consumer_send_channel_to_ust(app, ua_sess, ua_chan); - if (ret < 0) { + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + goto error; + } else if (ret < 0) { goto error; } @@ -1254,12 +1635,15 @@ static int send_channel_pid_to_ust(struct ust_app *app, /* Send all streams to application. */ cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) { ret = ust_consumer_send_stream_to_ust(app, ua_chan, stream); - if (ret < 0) { + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + goto error; + } else if (ret < 0) { goto error; } /* We don't need the stream anymore once sent to the tracer. */ cds_list_del(&stream->list); - delete_ust_app_stream(-1, stream); + delete_ust_app_stream(-1, stream, app); } /* Flag the channel that it is sent to the application. */ ua_chan->is_sent = 1; @@ -1283,13 +1667,21 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, health_code_update(); /* Create UST event on tracer */ + pthread_mutex_lock(&app->sock_lock); ret = ustctl_create_event(app->sock, &ua_event->attr, ua_chan->obj, &ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { 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; @@ -1310,12 +1702,24 @@ 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); + if (ua_event->enabled) { + /* + * We now need to explicitly enable the event, since it + * is now disabled at creation. + */ + ret = enable_ust_event(app, ua_sess, ua_event); if (ret < 0) { /* - * If we hit an EPERM, something is wrong with our disable call. If + * If we hit an EPERM, something is wrong with our enable call. If * we get an EEXIST, there is a problem on the tracer side since we * just created it. */ @@ -1345,6 +1749,8 @@ 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'; @@ -1355,9 +1761,22 @@ static void shadow_copy_event(struct ust_app_event *ua_event, /* Copy filter bytecode */ if (uevent->filter) { - ua_event->filter = alloc_copy_ust_app_filter(uevent->filter); + ua_event->filter = copy_filter_bytecode(uevent->filter); /* Filter might be NULL here in case of ENONEM. */ } + + /* Copy exclusion data */ + if (uevent->exclusion) { + exclusion_alloc_size = sizeof(struct lttng_event_exclusion) + + LTTNG_UST_SYM_NAME_LEN * uevent->exclusion->count; + ua_event->exclusion = zmalloc(exclusion_alloc_size); + if (ua_event->exclusion == NULL) { + PERROR("malloc"); + } else { + memcpy(ua_event->exclusion, uevent->exclusion, + exclusion_alloc_size); + } + } } /* @@ -1370,7 +1789,6 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, struct ltt_ust_event *uevent; struct ltt_ust_context *uctx; struct ust_app_event *ua_event; - struct ust_app_ctx *ua_ctx; DBG2("UST app shadow copy of channel %s started", ua_chan->name); @@ -1386,7 +1804,10 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, ua_chan->attr.overwrite = uchan->attr.overwrite; ua_chan->attr.switch_timer_interval = uchan->attr.switch_timer_interval; ua_chan->attr.read_timer_interval = uchan->attr.read_timer_interval; + ua_chan->monitor_timer_interval = uchan->monitor_timer_interval; ua_chan->attr.output = uchan->attr.output; + ua_chan->attr.blocking_timeout = uchan->attr.u.s.blocking_timeout; + /* * Note that the attribute channel type is not set since the channel on the * tracing registry side does not have this information. @@ -1395,20 +1816,22 @@ 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) { - ua_ctx = alloc_ust_app_ctx(&uctx->ctx); + cds_list_for_each_entry(uctx, &uchan->ctx_list, list) { + struct ust_app_ctx *ua_ctx = alloc_ust_app_ctx(&uctx->ctx); + if (ua_ctx == NULL) { continue; } 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); + lttng_ht_add_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); @@ -1438,6 +1861,7 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, struct tm *timeinfo; char datetime[16]; int ret; + char tmp_shm_path[PATH_MAX]; /* Get date and time for unique app path */ time(&rawtime); @@ -1454,9 +1878,15 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, ua_sess->egid = usess->gid; ua_sess->buffer_type = usess->buffer_type; ua_sess->bits_per_long = app->bits_per_long; + /* There is only one consumer object per session possible. */ + consumer_output_get(usess->consumer); ua_sess->consumer = usess->consumer; + ua_sess->output_traces = usess->output_traces; + ua_sess->live_timer_interval = usess->live_timer_interval; + copy_channel_attr_to_ustctl(&ua_sess->metadata_attr, + &usess->metadata_attr); switch (ua_sess->buffer_type) { case LTTNG_BUFFER_PER_PID: @@ -1478,6 +1908,38 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, goto error; } + strncpy(ua_sess->root_shm_path, usess->root_shm_path, + sizeof(ua_sess->root_shm_path)); + ua_sess->root_shm_path[sizeof(ua_sess->root_shm_path) - 1] = '\0'; + strncpy(ua_sess->shm_path, usess->shm_path, + sizeof(ua_sess->shm_path)); + ua_sess->shm_path[sizeof(ua_sess->shm_path) - 1] = '\0'; + if (ua_sess->shm_path[0]) { + switch (ua_sess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + ret = snprintf(tmp_shm_path, sizeof(tmp_shm_path), + DEFAULT_UST_TRACE_PID_PATH "/%s-%d-%s", + app->name, app->pid, datetime); + break; + case LTTNG_BUFFER_PER_UID: + ret = snprintf(tmp_shm_path, sizeof(tmp_shm_path), + DEFAULT_UST_TRACE_UID_PATH, + app->uid, app->bits_per_long); + break; + default: + assert(0); + goto error; + } + if (ret < 0) { + PERROR("sprintf UST shadow copy session"); + assert(0); + goto error; + } + strncat(ua_sess->shm_path, tmp_shm_path, + sizeof(ua_sess->shm_path) - strlen(ua_sess->shm_path) - 1); + ua_sess->shm_path[sizeof(ua_sess->shm_path) - 1] = '\0'; + } + /* Iterate over all channels in global domain. */ cds_lfht_for_each_entry(usess->domain_global.channels->ht, &iter.iter, uchan, node.node) { @@ -1492,7 +1954,8 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, DBG2("Channel %s not found on shadow session copy, creating it", uchan->name); - ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, &uchan->attr); + ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, + &uchan->attr); if (ua_chan == NULL) { /* malloc failed FIXME: Might want to do handle ENOMEM .. */ continue; @@ -1507,9 +1970,10 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node); } + return; error: - return; + consumer_output_put(ua_sess->consumer); } /* @@ -1569,11 +2033,11 @@ static int setup_buffer_reg_pid(struct ust_app_session *ua_sess, * This is the create channel path meaning that if there is NO * registry available, we have to create one for this session. */ - ret = buffer_reg_pid_create(ua_sess->id, ®_pid); + ret = buffer_reg_pid_create(ua_sess->id, ®_pid, + ua_sess->root_shm_path, ua_sess->shm_path); if (ret < 0) { goto error; } - buffer_reg_pid_add(reg_pid); } else { goto end; } @@ -1584,11 +2048,22 @@ static int setup_buffer_reg_pid(struct ust_app_session *ua_sess, app->uint16_t_alignment, app->uint32_t_alignment, app->uint64_t_alignment, app->long_alignment, app->byte_order, app->version.major, - app->version.minor); + app->version.minor, reg_pid->root_shm_path, + reg_pid->shm_path, + ua_sess->euid, ua_sess->egid); if (ret < 0) { + /* + * reg_pid->registry->reg.ust is NULL upon error, so we need to + * destroy the buffer registry, because it is always expected + * that if the buffer registry can be found, its ust registry is + * non-NULL. + */ + buffer_reg_pid_destroy(reg_pid); goto error; } + buffer_reg_pid_add(reg_pid); + DBG3("UST app buffer registry per PID created successfully"); end: @@ -1608,6 +2083,7 @@ error: * Return 0 on success or else a negative value. */ static int setup_buffer_reg_uid(struct ltt_ust_session *usess, + struct ust_app_session *ua_sess, struct ust_app *app, struct buffer_reg_uid **regp) { int ret = 0; @@ -1625,11 +2101,11 @@ static int setup_buffer_reg_uid(struct ltt_ust_session *usess, * registry available, we have to create one for this session. */ ret = buffer_reg_uid_create(usess->id, app->bits_per_long, app->uid, - LTTNG_DOMAIN_UST, ®_uid); + LTTNG_DOMAIN_UST, ®_uid, + ua_sess->root_shm_path, ua_sess->shm_path); if (ret < 0) { goto error; } - buffer_reg_uid_add(reg_uid); } else { goto end; } @@ -1640,15 +2116,24 @@ static int setup_buffer_reg_uid(struct ltt_ust_session *usess, app->uint16_t_alignment, app->uint32_t_alignment, app->uint64_t_alignment, app->long_alignment, app->byte_order, app->version.major, - app->version.minor); + app->version.minor, reg_uid->root_shm_path, + reg_uid->shm_path, usess->uid, usess->gid); if (ret < 0) { + /* + * reg_uid->registry->reg.ust is NULL upon error, so we need to + * destroy the buffer registry, because it is always expected + * that if the buffer registry can be found, its ust registry is + * non-NULL. + */ + buffer_reg_uid_destroy(reg_uid, NULL); goto error; } /* Add node to teardown list of the session. */ cds_list_add(®_uid->lnode, &usess->buffer_reg_uid_list); - DBG3("UST app buffer registry per UID created successfully"); + buffer_reg_uid_add(reg_uid); + DBG3("UST app buffer registry per UID created successfully"); end: if (regp) { *regp = reg_uid; @@ -1669,7 +2154,7 @@ error: * Returns 0 on success or else a negative code which is either -ENOMEM or * -ENOTCONN which is the default code if the ustctl_create_session fails. */ -static int create_ust_app_session(struct ltt_ust_session *usess, +static int find_or_create_ust_app_session(struct ltt_ust_session *usess, struct ust_app *app, struct ust_app_session **ua_sess_ptr, int *is_created) { @@ -1701,13 +2186,15 @@ static int create_ust_app_session(struct ltt_ust_session *usess, /* Init local registry. */ ret = setup_buffer_reg_pid(ua_sess, app, NULL); if (ret < 0) { + delete_ust_app_session(-1, ua_sess, app); goto error; } break; case LTTNG_BUFFER_PER_UID: /* Look for a global registry. If none exists, create one. */ - ret = setup_buffer_reg_uid(usess, app, NULL); + ret = setup_buffer_reg_uid(usess, ua_sess, app, NULL); if (ret < 0) { + delete_ust_app_session(-1, ua_sess, app); goto error; } break; @@ -1720,13 +2207,22 @@ static int create_ust_app_session(struct ltt_ust_session *usess, health_code_update(); if (ua_sess->handle == -1) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_create_session(app->sock); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("Creating session for app pid %d with ret %d", 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) { @@ -1745,6 +2241,9 @@ static int create_ust_app_session(struct ltt_ust_session *usess, lttng_ht_node_init_u64(&ua_sess->node, ua_sess->tracing_id); lttng_ht_add_unique_u64(app->sessions, &ua_sess->node); + lttng_ht_node_init_ulong(&ua_sess->ust_objd_node, ua_sess->handle); + lttng_ht_add_unique_ulong(app->ust_sessions_objd, + &ua_sess->ust_objd_node); DBG2("UST app session created successfully with handle %d", ret); } @@ -1763,25 +2262,103 @@ error: } /* - * Create a context for the channel on the tracer. + * Match function for a hash table lookup of ust_app_ctx. * - * Called with UST app session lock held and a RCU read side lock. + * It matches an ust app context based on the context type and, in the case + * of perf counters, their name. */ -static -int create_ust_app_channel_context(struct ust_app_session *ua_sess, - struct ust_app_channel *ua_chan, struct lttng_ust_context *uctx, - struct ust_app *app) +static int ht_match_ust_app_ctx(struct cds_lfht_node *node, const void *_key) +{ + struct ust_app_ctx *ctx; + const struct lttng_ust_context_attr *key; + + assert(node); + assert(_key); + + ctx = caa_container_of(node, struct ust_app_ctx, node.node); + key = _key; + + /* Context type */ + if (ctx->ctx.ctx != key->ctx) { + goto no_match; + } + + switch(key->ctx) { + case LTTNG_UST_CONTEXT_PERF_THREAD_COUNTER: + if (strncmp(key->u.perf_counter.name, + ctx->ctx.u.perf_counter.name, + sizeof(key->u.perf_counter.name))) { + goto no_match; + } + break; + case LTTNG_UST_CONTEXT_APP_CONTEXT: + if (strcmp(key->u.app_ctx.provider_name, + ctx->ctx.u.app_ctx.provider_name) || + strcmp(key->u.app_ctx.ctx_name, + ctx->ctx.u.app_ctx.ctx_name)) { + goto no_match; + } + break; + default: + break; + } + + /* Match. */ + return 1; + +no_match: + return 0; +} + +/* + * Lookup for an ust app context from an lttng_ust_context. + * + * Must be called while holding RCU read side lock. + * Return an ust_app_ctx object or NULL on error. + */ +static +struct ust_app_ctx *find_ust_app_context(struct lttng_ht *ht, + struct lttng_ust_context_attr *uctx) { - int ret = 0; struct lttng_ht_iter iter; struct lttng_ht_node_ulong *node; + struct ust_app_ctx *app_ctx = NULL; + + assert(uctx); + assert(ht); + + /* Lookup using the lttng_ust_context_type and a custom match fct. */ + cds_lfht_lookup(ht->ht, ht->hash_fct((void *) uctx->ctx, lttng_ht_seed), + ht_match_ust_app_ctx, uctx, &iter.iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (!node) { + goto end; + } + + app_ctx = caa_container_of(node, struct ust_app_ctx, node); + +end: + return app_ctx; +} + +/* + * Create a context for the channel on the tracer. + * + * Called with UST app session lock held and a RCU read side lock. + */ +static +int create_ust_app_channel_context(struct ust_app_session *ua_sess, + struct ust_app_channel *ua_chan, + struct lttng_ust_context_attr *uctx, + struct ust_app *app) +{ + int ret = 0; struct ust_app_ctx *ua_ctx; DBG2("UST app adding context to channel %s", ua_chan->name); - lttng_ht_lookup(ua_chan->ctx, (void *)((unsigned long)uctx->ctx), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node != NULL) { + ua_ctx = find_ust_app_context(ua_chan->ctx, uctx); + if (ua_ctx) { ret = -EEXIST; goto error; } @@ -1789,12 +2366,13 @@ int create_ust_app_channel_context(struct ust_app_session *ua_sess, ua_ctx = alloc_ust_app_ctx(uctx); if (ua_ctx == NULL) { /* malloc failed */ - ret = -1; + ret = -ENOMEM; goto error; } 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); + lttng_ht_add_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) { @@ -1900,6 +2478,8 @@ error: /* * Ask the consumer to create a channel and get it if successful. * + * Called with UST app session lock held. + * * Return 0 on success or else a negative value. */ static int do_consumer_create_channel(struct ltt_ust_session *usess, @@ -2072,7 +2652,8 @@ error_fd_get: * Return 0 on success or else a negative value. */ static int setup_buffer_reg_streams(struct buffer_reg_channel *reg_chan, - struct ust_app_channel *ua_chan) + struct ust_app_channel *ua_chan, + struct ust_app *app) { int ret = 0; struct ust_app_stream *stream, *stmp; @@ -2101,7 +2682,7 @@ static int setup_buffer_reg_streams(struct buffer_reg_channel *reg_chan, /* We don't need the streams anymore. */ cds_list_del(&stream->list); - delete_ust_app_stream(-1, stream); + delete_ust_app_stream(-1, stream, app); } error: @@ -2134,6 +2715,8 @@ 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; + reg_chan->num_subbuf = ua_chan->attr.num_subbuf; /* Create and add a channel registry to session. */ ret = ust_registry_channel_add(reg_sess->reg.ust, @@ -2163,7 +2746,8 @@ error_create: * Return 0 on success else a negative value. */ static int setup_buffer_reg_channel(struct buffer_reg_session *reg_sess, - struct ust_app_channel *ua_chan, struct buffer_reg_channel *reg_chan) + struct ust_app_channel *ua_chan, struct buffer_reg_channel *reg_chan, + struct ust_app *app) { int ret; @@ -2175,7 +2759,7 @@ static int setup_buffer_reg_channel(struct buffer_reg_session *reg_sess, DBG2("UST app setup buffer registry channel for %s", ua_chan->name); /* Setup all streams for the registry. */ - ret = setup_buffer_reg_streams(reg_chan, ua_chan); + ret = setup_buffer_reg_streams(reg_chan, ua_chan, app); if (ret < 0) { goto error; } @@ -2217,7 +2801,10 @@ static int send_channel_uid_to_ust(struct buffer_reg_channel *reg_chan, /* Send channel to the application. */ ret = ust_consumer_send_channel_to_ust(app, ua_sess, ua_chan); - if (ret < 0) { + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + goto error; + } else if (ret < 0) { goto error; } @@ -2235,7 +2822,10 @@ static int send_channel_uid_to_ust(struct buffer_reg_channel *reg_chan, ret = ust_consumer_send_stream_to_ust(app, ua_chan, &stream); if (ret < 0) { - (void) release_ust_app_stream(-1, &stream); + (void) release_ust_app_stream(-1, &stream, app); + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + } goto error_stream_unlock; } @@ -2243,7 +2833,7 @@ static int send_channel_uid_to_ust(struct buffer_reg_channel *reg_chan, * The return value is not important here. This function will output an * error if needed. */ - (void) release_ust_app_stream(-1, &stream); + (void) release_ust_app_stream(-1, &stream, app); } ua_chan->is_sent = 1; @@ -2256,6 +2846,9 @@ error: /* * Create and send to the application the created buffers with per UID buffers. * + * This MUST be called with a RCU read side lock acquired. + * The session list lock and the session's lock must be acquired. + * * Return 0 on success else a negative value. */ static int create_channel_per_uid(struct ust_app *app, @@ -2283,45 +2876,91 @@ static int create_channel_per_uid(struct ust_app *app, reg_chan = buffer_reg_channel_find(ua_chan->tracing_channel_id, reg_uid); - if (!reg_chan) { - /* Create the buffer registry channel object. */ - ret = create_buffer_reg_channel(reg_uid->registry, ua_chan, ®_chan); - if (ret < 0) { - goto error; - } - assert(reg_chan); + if (reg_chan) { + goto send_channel; + } - /* - * Create the buffers on the consumer side. This call populates the - * ust app channel object with all streams and data object. - */ - 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; - } + /* Create the buffer registry channel object. */ + ret = create_buffer_reg_channel(reg_uid->registry, ua_chan, ®_chan); + if (ret < 0) { + ERR("Error creating the UST channel \"%s\" registry instance", + ua_chan->name); + goto error; + } + + /* + * Create the buffers on the consumer side. This call populates the + * ust app channel object with all streams and data object. + */ + ret = do_consumer_create_channel(usess, ua_sess, ua_chan, + app->bits_per_long, reg_uid->registry->reg.ust); + if (ret < 0) { + ERR("Error creating UST channel \"%s\" on the consumer daemon", + ua_chan->name); /* - * Setup the streams and add it to the session registry. + * Let's remove the previously created buffer registry channel so + * it's not visible anymore in the session registry. */ - ret = setup_buffer_reg_channel(reg_uid->registry, ua_chan, reg_chan); - if (ret < 0) { + ust_registry_channel_del_free(reg_uid->registry->reg.ust, + ua_chan->tracing_channel_id, false); + buffer_reg_channel_remove(reg_uid->registry, reg_chan); + buffer_reg_channel_destroy(reg_chan, LTTNG_DOMAIN_UST); + goto error; + } + + /* + * Setup the streams and add it to the session registry. + */ + ret = setup_buffer_reg_channel(reg_uid->registry, + ua_chan, reg_chan, app); + if (ret < 0) { + ERR("Error setting up UST channel \"%s\"", ua_chan->name); + goto error; + } + + { + enum lttng_error_code cmd_ret; + struct ltt_session *session; + uint64_t chan_reg_key; + struct ust_registry_channel *chan_reg; + + chan_reg_key = ua_chan->tracing_channel_id; + + pthread_mutex_lock(®_uid->registry->reg.ust->lock); + chan_reg = ust_registry_channel_find(reg_uid->registry->reg.ust, + chan_reg_key); + assert(chan_reg); + chan_reg->consumer_key = ua_chan->key; + chan_reg = NULL; + pthread_mutex_unlock(®_uid->registry->reg.ust->lock); + + session = session_find_by_id(ua_sess->tracing_id); + assert(session); + + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); + cmd_ret = notification_thread_command_add_channel( + notification_thread_handle, session->name, + ua_sess->euid, ua_sess->egid, + ua_chan->name, + ua_chan->key, + LTTNG_DOMAIN_UST, + ua_chan->attr.subbuf_size * ua_chan->attr.num_subbuf); + if (cmd_ret != LTTNG_OK) { + ret = - (int) cmd_ret; + ERR("Failed to add channel to notification thread"); goto error; } - } +send_channel: /* Send buffers to the application. */ ret = send_channel_uid_to_ust(reg_chan, app, ua_sess, ua_chan); if (ret < 0) { + if (ret != -ENOTCONN) { + ERR("Error sending channel to application"); + } goto error; } @@ -2332,6 +2971,9 @@ error: /* * Create and send to the application the created buffers with per PID buffers. * + * Called with UST app session lock held. + * The session list lock and the session's lock must be acquired. + * * Return 0 on success else a negative value. */ static int create_channel_per_pid(struct ust_app *app, @@ -2340,6 +2982,10 @@ static int create_channel_per_pid(struct ust_app *app, { int ret; struct ust_registry_session *registry; + enum lttng_error_code cmd_ret; + struct ltt_session *session; + uint64_t chan_reg_key; + struct ust_registry_channel *chan_reg; assert(app); assert(usess); @@ -2351,11 +2997,14 @@ static int create_channel_per_pid(struct ust_app *app, rcu_read_lock(); registry = get_session_registry(ua_sess); + /* The UST app session lock is held, registry shall not be null. */ assert(registry); /* Create and add a new channel registry to session. */ ret = ust_registry_channel_add(registry, ua_chan->key); if (ret < 0) { + ERR("Error creating the UST channel \"%s\" registry instance", + ua_chan->name); goto error; } @@ -2363,11 +3012,42 @@ static int create_channel_per_pid(struct ust_app *app, ret = do_consumer_create_channel(usess, ua_sess, ua_chan, app->bits_per_long, registry); if (ret < 0) { + ERR("Error creating UST channel \"%s\" on the consumer daemon", + ua_chan->name); goto error; } ret = send_channel_pid_to_ust(app, ua_sess, ua_chan); if (ret < 0) { + if (ret != -ENOTCONN) { + ERR("Error sending channel to application"); + } + goto error; + } + + session = session_find_by_id(ua_sess->tracing_id); + assert(session); + + chan_reg_key = ua_chan->key; + pthread_mutex_lock(®istry->lock); + chan_reg = ust_registry_channel_find(registry, chan_reg_key); + assert(chan_reg); + chan_reg->consumer_key = ua_chan->key; + pthread_mutex_unlock(®istry->lock); + + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); + + cmd_ret = notification_thread_command_add_channel( + notification_thread_handle, session->name, + ua_sess->euid, ua_sess->egid, + ua_chan->name, + ua_chan->key, + LTTNG_DOMAIN_UST, + ua_chan->attr.subbuf_size * ua_chan->attr.num_subbuf); + if (cmd_ret != LTTNG_OK) { + ret = - (int) cmd_ret; + ERR("Failed to add channel to notification thread"); goto error; } @@ -2381,7 +3061,10 @@ error: * need and send it to the application. This MUST be called with a RCU read * side lock acquired. * - * Return 0 on success or else a negative value. + * Called with UST app session lock held. + * + * Return 0 on success or else a negative value. Returns -ENOTCONN if + * the application exited concurrently. */ static int do_create_channel(struct ust_app *app, struct ltt_ust_session *usess, struct ust_app_session *ua_sess, @@ -2440,7 +3123,8 @@ error: * * Called with UST app session lock and RCU read-side lock held. * - * Return 0 on success or else a negative value. + * Return 0 on success or else a negative value. Returns -ENOTCONN if + * the application exited concurrently. */ static int create_ust_app_channel(struct ust_app_session *ua_sess, struct ltt_ust_channel *uchan, struct ust_app *app, @@ -2481,7 +3165,6 @@ static int create_ust_app_channel(struct ust_app_session *ua_sess, /* Only add the channel if successful on the tracer side. */ lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node); - end: if (ua_chanp) { *ua_chanp = ua_chan; @@ -2511,7 +3194,7 @@ 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; @@ -2544,7 +3227,7 @@ end: error: /* Valid. Calling here is already in a read side lock */ - delete_ust_app_event(-1, ua_event); + delete_ust_app_event(-1, ua_event, app); return ret; } @@ -2554,8 +3237,7 @@ error: * Called with UST app session lock held and RCU read side lock. */ static int create_ust_app_metadata(struct ust_app_session *ua_sess, - struct ust_app *app, struct consumer_output *consumer, - struct ustctl_consumer_channel_attr *attr) + struct ust_app *app, struct consumer_output *consumer) { int ret = 0; struct ust_app_channel *metadata; @@ -2567,8 +3249,11 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, assert(consumer); registry = get_session_registry(ua_sess); + /* The UST app session is held registry shall not be null. */ assert(registry); + pthread_mutex_lock(®istry->lock); + /* Metadata already exists for this registry or it was closed previously */ if (registry->metadata_key || registry->metadata_closed) { ret = 0; @@ -2583,20 +3268,7 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, goto error; } - if (!attr) { - /* Set default attributes for metadata. */ - 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_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 { - memcpy(&metadata->attr, attr, sizeof(metadata->attr)); - metadata->attr.output = LTTNG_UST_MMAP; - metadata->attr.type = LTTNG_UST_CHAN_METADATA; - } + memcpy(&metadata->attr, &ua_sess->metadata_attr, sizeof(metadata->attr)); /* Need one fd for the channel. */ ret = lttng_fd_get(LTTNG_FD_APPS, 1); @@ -2654,17 +3326,10 @@ error_consumer: lttng_fd_put(LTTNG_FD_APPS, 1); delete_ust_app_channel(-1, metadata, app); error: + pthread_mutex_unlock(®istry->lock); return ret; } -/* - * Return pointer to traceable apps list. - */ -struct lttng_ht *ust_app_get_ht(void) -{ - return ust_app_ht; -} - /* * Return ust app pointer or NULL if not found. RCU read side lock MUST be * acquired before calling this function. @@ -2738,6 +3403,7 @@ struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) lta->v_minor = msg->minor; lta->sessions = lttng_ht_new(0, LTTNG_HT_TYPE_U64); lta->ust_objd = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + lta->ust_sessions_objd = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); lta->notify_sock = -1; /* Copy name and make sure it's NULL terminated. */ @@ -2754,10 +3420,10 @@ struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) lta->pid = msg->pid; lttng_ht_node_init_ulong(<a->pid_n, (unsigned long) lta->pid); lta->sock = sock; + pthread_mutex_init(<a->sock_lock, NULL); lttng_ht_node_init_ulong(<a->sock_n, (unsigned long) lta->sock); CDS_INIT_LIST_HEAD(<a->teardown_head); - error: return lta; } @@ -2809,12 +3475,14 @@ int ust_app_version(struct ust_app *app) assert(app); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_tracer_version(app->sock, &app->version); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { - ERR("UST app %d verson failed with ret %d", app->sock, ret); + ERR("UST app %d version failed with ret %d", app->sock, ret); } else { - DBG3("UST app %d verion failed. Application is dead", app->sock); + DBG3("UST app %d version failed. Application is dead", app->sock); } } @@ -2831,6 +3499,7 @@ void ust_app_unregister(int sock) { struct ust_app *lta; struct lttng_ht_node_ulong *node; + struct lttng_ht_iter ust_app_sock_iter; struct lttng_ht_iter iter; struct ust_app_session *ua_sess; int ret; @@ -2838,39 +3507,19 @@ void ust_app_unregister(int sock) rcu_read_lock(); /* Get the node reference for a call_rcu */ - lttng_ht_lookup(ust_app_ht_by_sock, (void *)((unsigned long) sock), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + lttng_ht_lookup(ust_app_ht_by_sock, (void *)((unsigned long) sock), &ust_app_sock_iter); + node = lttng_ht_iter_get_node_ulong(&ust_app_sock_iter); assert(node); lta = caa_container_of(node, struct ust_app, sock_n); DBG("PID %d unregistering with sock %d", lta->pid, sock); - /* Remove application from PID hash table */ - ret = lttng_ht_del(ust_app_ht_by_sock, &iter); - assert(!ret); - - /* - * Remove application from notify hash table. The thread handling the - * notify socket could have deleted the node so ignore on error because - * either way it's valid. The close of that socket is handled by the other - * thread. - */ - iter.iter.node = <a->notify_sock_n.node; - (void) lttng_ht_del(ust_app_ht_by_notify_sock, &iter); - /* - * Ignore return value since the node might have been removed before by an - * add replace during app registration because the PID can be reassigned by - * the OS. + * For per-PID buffers, perform "push metadata" and flush all + * application streams before removing app from hash tables, + * ensuring proper behavior of data_pending check. + * Remove sessions so they are not visible during deletion. */ - iter.iter.node = <a->pid_n.node; - ret = lttng_ht_del(ust_app_ht, &iter); - if (ret) { - DBG3("Unregister app by PID %d failed. This can happen on pid reuse", - lta->pid); - } - - /* Remove sessions so they are not visible during deletion.*/ cds_lfht_for_each_entry(lta->sessions->ht, &iter.iter, ua_sess, node.node) { struct ust_registry_session *registry; @@ -2881,12 +3530,21 @@ void ust_app_unregister(int sock) continue; } + if (ua_sess->buffer_type == LTTNG_BUFFER_PER_PID) { + (void) ust_app_flush_app_session(lta, ua_sess); + } + /* * Add session to list for teardown. This is safe since at this point we * are the only one using this list. */ pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + /* * Normally, this is done in the delete session process which is * executed in the call rcu below. However, upon registration we can't @@ -2899,7 +3557,7 @@ 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 && !registry->metadata_closed) { + if (registry) { /* Push metadata for application before freeing the application. */ (void) push_metadata(registry, ua_sess->consumer); @@ -2909,17 +3567,41 @@ void ust_app_unregister(int sock) * 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 && - !registry->metadata_closed) { + if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID) { /* And ask to close it for this session registry. */ (void) close_metadata(registry, ua_sess->consumer); } } - cds_list_add(&ua_sess->teardown_node, <a->teardown_head); + pthread_mutex_unlock(&ua_sess->lock); } + /* Remove application from PID hash table */ + ret = lttng_ht_del(ust_app_ht_by_sock, &ust_app_sock_iter); + assert(!ret); + + /* + * Remove application from notify hash table. The thread handling the + * notify socket could have deleted the node so ignore on error because + * either way it's valid. The close of that socket is handled by the + * apps_notify_thread. + */ + iter.iter.node = <a->notify_sock_n.node; + (void) lttng_ht_del(ust_app_ht_by_notify_sock, &iter); + + /* + * Ignore return value since the node might have been removed before by an + * add replace during app registration because the PID can be reassigned by + * the OS. + */ + iter.iter.node = <a->pid_n.node; + ret = lttng_ht_del(ust_app_ht, &iter); + if (ret) { + DBG3("Unregister app by PID %d failed. This can happen on pid reuse", + lta->pid); + } + /* Free memory */ call_rcu(<a->pid_n.head, delete_ust_app_rcu); @@ -2927,20 +3609,6 @@ void ust_app_unregister(int sock) return; } -/* - * Return traceable_app_count - */ -unsigned long ust_app_list_count(void) -{ - unsigned long count; - - rcu_read_lock(); - count = lttng_ht_get_count(ust_app_ht); - rcu_read_unlock(); - - return count; -} - /* * Fill events array with all events name of all registered apps. */ @@ -2974,12 +3642,14 @@ int ust_app_list_events(struct lttng_event **events) */ continue; } + pthread_mutex_lock(&app->sock_lock); handle = ustctl_tracepoint_list(app->sock); if (handle < 0) { if (handle != -EPIPE && handle != -LTTNG_UST_ERR_EXITING) { ERR("UST app list events getting handle failed for app pid %d", app->pid); } + pthread_mutex_unlock(&app->sock_lock); continue; } @@ -2987,32 +3657,62 @@ int ust_app_list_events(struct lttng_event **events) &uiter)) != -LTTNG_UST_ERR_NOENT) { /* Handle ustctl error. */ if (ret < 0) { - free(tmp_event); - if (ret != -LTTNG_UST_ERR_EXITING || ret != -EPIPE) { + int release_ret; + + if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { ERR("UST app tp list get failed for app %d with ret %d", 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; + } + free(tmp_event); + release_ret = ustctl_release_handle(app->sock, handle); + if (release_ret < 0 && + release_ret != -LTTNG_UST_ERR_EXITING && + release_ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, release_ret); } + pthread_mutex_unlock(&app->sock_lock); goto rcu_error; } health_code_update(); if (count >= nbmem) { /* In case the realloc fails, we free the memory */ - void *ptr; + struct lttng_event *new_tmp_event; + size_t new_nbmem; + + new_nbmem = nbmem << 1; + DBG2("Reallocating event list from %zu to %zu entries", + nbmem, new_nbmem); + new_tmp_event = realloc(tmp_event, + new_nbmem * sizeof(struct lttng_event)); + if (new_tmp_event == NULL) { + int release_ret; - DBG2("Reallocating event list from %zu to %zu entries", nbmem, - 2 * nbmem); - nbmem *= 2; - ptr = realloc(tmp_event, nbmem * sizeof(struct lttng_event)); - if (ptr == NULL) { PERROR("realloc ust app events"); free(tmp_event); ret = -ENOMEM; + release_ret = ustctl_release_handle(app->sock, handle); + if (release_ret < 0 && + release_ret != -LTTNG_UST_ERR_EXITING && + release_ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, release_ret); + } + pthread_mutex_unlock(&app->sock_lock); goto rcu_error; } - tmp_event = ptr; + /* Zero the new memory */ + memset(new_tmp_event + nbmem, 0, + (new_nbmem - nbmem) * sizeof(struct lttng_event)); + nbmem = new_nbmem; + tmp_event = new_tmp_event; } memcpy(tmp_event[count].name, uiter.name, LTTNG_UST_SYM_NAME_LEN); tmp_event[count].loglevel = uiter.loglevel; @@ -3021,6 +3721,11 @@ int ust_app_list_events(struct lttng_event **events) tmp_event[count].enabled = -1; count++; } + ret = ustctl_release_handle(app->sock, handle); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0 && ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, ret); + } } ret = count; @@ -3068,12 +3773,14 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) */ continue; } + pthread_mutex_lock(&app->sock_lock); handle = ustctl_tracepoint_field_list(app->sock); if (handle < 0) { if (handle != -EPIPE && handle != -LTTNG_UST_ERR_EXITING) { ERR("UST app list field getting handle failed for app pid %d", app->pid); } + pthread_mutex_unlock(&app->sock_lock); continue; } @@ -3081,12 +3788,27 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) &uiter)) != -LTTNG_UST_ERR_NOENT) { /* Handle ustctl error. */ if (ret < 0) { - free(tmp_event); - if (ret != -LTTNG_UST_ERR_EXITING || ret != -EPIPE) { + int release_ret; + + if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { ERR("UST app tp list field failed for app %d with ret %d", 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. Reset list and count for next app. + */ + break; + } + free(tmp_event); + release_ret = ustctl_release_handle(app->sock, handle); + pthread_mutex_unlock(&app->sock_lock); + if (release_ret < 0 && + release_ret != -LTTNG_UST_ERR_EXITING && + release_ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, release_ret); } goto rcu_error; } @@ -3094,32 +3816,55 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) health_code_update(); if (count >= nbmem) { /* In case the realloc fails, we free the memory */ - void *ptr; + struct lttng_event_field *new_tmp_event; + size_t new_nbmem; + + new_nbmem = nbmem << 1; + DBG2("Reallocating event field list from %zu to %zu entries", + nbmem, new_nbmem); + new_tmp_event = realloc(tmp_event, + new_nbmem * sizeof(struct lttng_event_field)); + if (new_tmp_event == NULL) { + int release_ret; - DBG2("Reallocating event field list from %zu to %zu entries", nbmem, - 2 * nbmem); - nbmem *= 2; - ptr = realloc(tmp_event, nbmem * sizeof(struct lttng_event_field)); - if (ptr == NULL) { PERROR("realloc ust app event fields"); free(tmp_event); ret = -ENOMEM; + release_ret = ustctl_release_handle(app->sock, handle); + pthread_mutex_unlock(&app->sock_lock); + if (release_ret && + release_ret != -LTTNG_UST_ERR_EXITING && + release_ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, release_ret); + } goto rcu_error; } - tmp_event = ptr; + /* Zero the new memory */ + memset(new_tmp_event + nbmem, 0, + (new_nbmem - nbmem) * sizeof(struct lttng_event_field)); + nbmem = new_nbmem; + tmp_event = new_tmp_event; } memcpy(tmp_event[count].field_name, uiter.field_name, LTTNG_UST_SYM_NAME_LEN); - tmp_event[count].type = uiter.type; + /* Mapping between these enums matches 1 to 1. */ + tmp_event[count].type = (enum lttng_event_field_type) uiter.type; tmp_event[count].nowrite = uiter.nowrite; memcpy(tmp_event[count].event.name, uiter.event_name, LTTNG_UST_SYM_NAME_LEN); tmp_event[count].event.loglevel = uiter.loglevel; - tmp_event[count].event.type = LTTNG_UST_TRACEPOINT; + tmp_event[count].event.type = LTTNG_EVENT_TRACEPOINT; tmp_event[count].event.pid = app->pid; tmp_event[count].event.enabled = -1; count++; } + ret = ustctl_release_handle(app->sock, handle); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0 && + ret != -LTTNG_UST_ERR_EXITING && + ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, ret); + } } ret = count; @@ -3149,41 +3894,63 @@ void ust_app_clean_list(void) rcu_read_lock(); - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = lttng_ht_del(ust_app_ht, &iter); - assert(!ret); - call_rcu(&app->pid_n.head, delete_ust_app_rcu); + if (ust_app_ht) { + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ret = lttng_ht_del(ust_app_ht, &iter); + assert(!ret); + call_rcu(&app->pid_n.head, delete_ust_app_rcu); + } } /* Cleanup socket hash table */ - cds_lfht_for_each_entry(ust_app_ht_by_sock->ht, &iter.iter, app, - sock_n.node) { - ret = lttng_ht_del(ust_app_ht_by_sock, &iter); - assert(!ret); + if (ust_app_ht_by_sock) { + cds_lfht_for_each_entry(ust_app_ht_by_sock->ht, &iter.iter, app, + sock_n.node) { + ret = lttng_ht_del(ust_app_ht_by_sock, &iter); + assert(!ret); + } } /* Cleanup notify socket hash table */ - cds_lfht_for_each_entry(ust_app_ht_by_notify_sock->ht, &iter.iter, app, - notify_sock_n.node) { - ret = lttng_ht_del(ust_app_ht_by_notify_sock, &iter); - assert(!ret); + if (ust_app_ht_by_notify_sock) { + cds_lfht_for_each_entry(ust_app_ht_by_notify_sock->ht, &iter.iter, app, + notify_sock_n.node) { + ret = lttng_ht_del(ust_app_ht_by_notify_sock, &iter); + assert(!ret); + } } rcu_read_unlock(); /* Destroy is done only when the ht is empty */ - ht_cleanup_push(ust_app_ht); - ht_cleanup_push(ust_app_ht_by_sock); - ht_cleanup_push(ust_app_ht_by_notify_sock); + if (ust_app_ht) { + ht_cleanup_push(ust_app_ht); + } + if (ust_app_ht_by_sock) { + ht_cleanup_push(ust_app_ht_by_sock); + } + if (ust_app_ht_by_notify_sock) { + ht_cleanup_push(ust_app_ht_by_notify_sock); + } } /* * Init UST app hash table. */ -void ust_app_ht_alloc(void) +int ust_app_ht_alloc(void) { ust_app_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!ust_app_ht) { + return -1; + } ust_app_ht_by_sock = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!ust_app_ht_by_sock) { + return -1; + } ust_app_ht_by_notify_sock = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!ust_app_ht_by_notify_sock) { + return -1; + } + return 0; } /* @@ -3307,7 +4074,7 @@ int ust_app_disable_event_glb(struct ltt_ust_session *usess, { int ret = 0; struct lttng_ht_iter iter, uiter; - struct lttng_ht_node_str *ua_chan_node, *ua_event_node; + struct lttng_ht_node_str *ua_chan_node; struct ust_app *app; struct ust_app_session *ua_sess; struct ust_app_channel *ua_chan; @@ -3344,14 +4111,14 @@ int ust_app_disable_event_glb(struct ltt_ust_session *usess, } ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - lttng_ht_lookup(ua_chan->events, (void *)uevent->attr.name, &uiter); - ua_event_node = lttng_ht_iter_get_node_str(&uiter); - if (ua_event_node == NULL) { + ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, + uevent->filter, uevent->attr.loglevel, + uevent->exclusion); + if (ua_event == NULL) { DBG2("Event %s not found in channel %s for app pid %d." "Skipping", uevent->attr.name, uchan->name, app->pid); continue; } - ua_event = caa_container_of(ua_event_node, struct ust_app_event, node); ret = disable_ust_app_event(ua_sess, ua_event, app); if (ret < 0) { @@ -3366,26 +4133,26 @@ int ust_app_disable_event_glb(struct ltt_ust_session *usess, } /* - * For a specific UST session and UST channel, the event for all - * registered apps. + * For a specific UST session, create the channel for all registered apps. */ -int ust_app_disable_all_event_glb(struct ltt_ust_session *usess, +int ust_app_create_channel_glb(struct ltt_ust_session *usess, struct ltt_ust_channel *uchan) { - int ret = 0; - struct lttng_ht_iter iter, uiter; - struct lttng_ht_node_str *ua_chan_node; + int ret = 0, created; + struct lttng_ht_iter iter; struct ust_app *app; - struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; - struct ust_app_event *ua_event; + struct ust_app_session *ua_sess = NULL; + + /* Very wrong code flow */ + assert(usess); + assert(uchan); - DBG("UST app disabling all event for all apps in channel " - "%s for session id %" PRIu64, uchan->name, usess->id); + DBG2("UST app adding channel %s to UST domain for session id %" PRIu64, + uchan->name, usess->id); rcu_read_lock(); - /* For all registered applications */ + /* For every registered applications */ cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { if (!app->compatible) { /* @@ -3394,71 +4161,17 @@ int ust_app_disable_all_event_glb(struct ltt_ust_session *usess, */ continue; } - ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ + if (!trace_ust_pid_tracker_lookup(usess, app->pid)) { + /* Skip. */ continue; } - /* Lookup channel in the ust app session */ - lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); - ua_chan_node = lttng_ht_iter_get_node_str(&uiter); - /* If the channel is not found, there is a code flow error */ - assert(ua_chan_node); - - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - - /* Disable each events of channel */ - cds_lfht_for_each_entry(ua_chan->events->ht, &uiter.iter, ua_event, - node.node) { - ret = disable_ust_app_event(ua_sess, ua_event, app); - if (ret < 0) { - /* XXX: Report error someday... */ - continue; - } - } - } - - rcu_read_unlock(); - - return ret; -} - -/* - * For a specific UST session, create the channel for all registered apps. - */ -int ust_app_create_channel_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan) -{ - int ret = 0, created; - struct lttng_ht_iter iter; - struct ust_app *app; - struct ust_app_session *ua_sess = NULL; - - /* Very wrong code flow */ - assert(usess); - assert(uchan); - - DBG2("UST app adding channel %s to UST domain for session id %" PRIu64, - uchan->name, usess->id); - - rcu_read_lock(); - - /* For every registered applications */ - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; - } /* * Create session on the tracer side and add it to app session HT. Note * that if session exist, it will simply return a pointer to the ust * app session. */ - ret = create_ust_app_session(usess, app, &ua_sess, &created); + ret = find_or_create_ust_app_session(usess, app, &ua_sess, &created); if (ret < 0) { switch (ret) { case -ENOTCONN: @@ -3467,6 +4180,7 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, * or a timeout on it. We can't inform the caller that for a * specific app, the session failed so lets continue here. */ + ret = 0; /* Not an error. */ continue; case -ENOMEM: default: @@ -3476,12 +4190,16 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, assert(ua_sess); pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + if (!strncmp(uchan->name, DEFAULT_METADATA_NAME, sizeof(uchan->name))) { - struct ustctl_consumer_channel_attr attr; - copy_channel_attr_to_ustctl(&attr, &uchan->attr); - ret = create_ust_app_metadata(ua_sess, app, usess->consumer, - &attr); + copy_channel_attr_to_ustctl(&ua_sess->metadata_attr, &uchan->attr); + ret = 0; } else { /* Create channel onto application. We don't need the chan ref. */ ret = create_ust_app_channel(ua_sess, uchan, app, @@ -3489,14 +4207,23 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, } pthread_mutex_unlock(&ua_sess->lock); if (ret < 0) { - if (ret == -ENOMEM) { - /* No more memory is a fatal error. Stop right now. */ - goto error_rcu_unlock; - } /* Cleanup the created session if it's the case. */ if (created) { destroy_app_session(app, ua_sess); } + switch (ret) { + case -ENOTCONN: + /* + * The application's socket is not valid. Either a bad socket + * or a timeout on it. We can't inform the caller that for a + * specific app, the session failed so lets continue here. + */ + ret = 0; /* Not an error. */ + continue; + case -ENOMEM: + default: + goto error_rcu_unlock; + } } } @@ -3547,17 +4274,29 @@ int ust_app_enable_event_glb(struct ltt_ust_session *usess, pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + /* Lookup channel in the ust app session */ lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); ua_chan_node = lttng_ht_iter_get_node_str(&uiter); - /* If the channel is not found, there is a code flow error */ - assert(ua_chan_node); + /* + * It is possible that the channel cannot be found is + * the channel/event creation occurs concurrently with + * an application exit. + */ + if (!ua_chan_node) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); /* 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); @@ -3613,6 +4352,12 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, } pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + /* Lookup channel in the ust app session */ lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); ua_chan_node = lttng_ht_iter_get_node_str(&uiter); @@ -3641,6 +4386,9 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, /* * Start tracing for a specific UST session and app. + * + * Called with UST app session lock held. + * */ static int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) @@ -3664,6 +4412,11 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } + /* Upon restart, we skip the setup, already done */ if (ua_sess->started) { goto skip_setup; @@ -3671,11 +4424,34 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) /* Create directories if consumer is LOCAL and has a path defined. */ if (usess->consumer->type == CONSUMER_DST_LOCAL && - strlen(usess->consumer->dst.trace_path) > 0) { - ret = run_as_mkdir_recursive(usess->consumer->dst.trace_path, - S_IRWXU | S_IRWXG, ua_sess->euid, ua_sess->egid); + usess->consumer->dst.session_root_path[0] != '\0') { + char *tmp_path; + + tmp_path = zmalloc(LTTNG_PATH_MAX); + if (!tmp_path) { + ERR("Alloc tmp_path"); + goto error_unlock; + } + ret = snprintf(tmp_path, LTTNG_PATH_MAX, "%s%s%s", + usess->consumer->dst.session_root_path, + usess->consumer->chunk_path, + usess->consumer->subdir); + if (ret >= LTTNG_PATH_MAX) { + ERR("Local destination path exceeds the maximal allowed length of %i bytes (needs %i bytes) with path = \"%s%s%s\"", + LTTNG_PATH_MAX, ret, + usess->consumer->dst.session_root_path, + usess->consumer->chunk_path, + usess->consumer->subdir); + goto error_unlock; + } + + DBG("Creating directory path for local tracing: \"%s\"", + tmp_path); + ret = run_as_mkdir_recursive(tmp_path, S_IRWXU | S_IRWXG, + ua_sess->euid, ua_sess->egid); + free(tmp_path); if (ret < 0) { - if (ret != -EEXIST) { + if (errno != EEXIST) { ERR("Trace directory creation error"); goto error_unlock; } @@ -3686,7 +4462,7 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) * Create the metadata for the application. This returns gracefully if a * metadata was already set for the session. */ - ret = create_ust_app_metadata(ua_sess, app, usess->consumer, NULL); + ret = create_ust_app_metadata(ua_sess, app, usess->consumer); if (ret < 0) { goto error_unlock; } @@ -3695,13 +4471,22 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) skip_setup: /* This start the UST tracing */ + pthread_mutex_lock(&app->sock_lock); ret = ustctl_start_session(app->sock, ua_sess->handle); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("Error starting tracing for app pid: %d (ret: %d)", 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; } @@ -3714,7 +4499,9 @@ skip_setup: health_code_update(); /* Quiescent wait after starting trace */ + pthread_mutex_lock(&app->sock_lock); ret = ustctl_wait_quiescent(app->sock); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app wait quiescent failed for app pid %d ret %d", app->pid, ret); @@ -3757,6 +4544,11 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end_no_session; + } + /* * If started = 0, it means that stop trace has been called for a session * that was never started. It's possible since we can have a fail start @@ -3770,13 +4562,21 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) health_code_update(); /* This inhibits UST tracing */ + pthread_mutex_lock(&app->sock_lock); ret = ustctl_stop_session(app->sock, ua_sess->handle); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("Error stopping tracing for app pid: %d (ret: %d)", 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; } @@ -3784,7 +4584,9 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) health_code_update(); /* Quiescent wait after stopping trace */ + pthread_mutex_lock(&app->sock_lock); ret = ustctl_wait_quiescent(app->sock); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app wait quiescent failed for app pid %d ret %d", app->pid, ret); @@ -3793,13 +4595,14 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) health_code_update(); registry = get_session_registry(ua_sess); + + /* The UST app session is held registry shall not be null. */ assert(registry); - if (!registry->metadata_closed) { - /* Push metadata for application before freeing the application. */ - (void) push_metadata(registry, ua_sess->consumer); - } + /* 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(); @@ -3813,162 +4616,90 @@ error_rcu_unlock: return -1; } -/* - * Flush buffers for a specific UST session and app. - */ static -int ust_app_flush_trace(struct ltt_ust_session *usess, struct ust_app *app) +int ust_app_flush_app_session(struct ust_app *app, + struct ust_app_session *ua_sess) { - int ret = 0; + int ret, retval = 0; struct lttng_ht_iter iter; - struct ust_app_session *ua_sess; struct ust_app_channel *ua_chan; + struct consumer_socket *socket; - DBG("Flushing buffers for ust app pid %d", app->pid); + DBG("Flushing app session buffers for ust app pid %d", app->pid); rcu_read_lock(); if (!app->compatible) { - goto end_no_session; - } - - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - goto end_no_session; + goto end_not_compatible; } pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + goto end_deleted; + } + health_code_update(); /* Flushing buffers */ - cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, - node.node) { - health_code_update(); - assert(ua_chan->is_sent); - ret = ustctl_sock_flush_buffer(app->sock, ua_chan->obj); - if (ret < 0) { - if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { - ERR("UST app PID %d channel %s flush failed with ret %d", - app->pid, ua_chan->name, ret); - } else { - DBG3("UST app failed to flush %s. Application is dead.", - ua_chan->name); - /* No need to continue. */ - break; + socket = consumer_find_socket_by_bitness(app->bits_per_long, + ua_sess->consumer); + + /* Flush buffers and push metadata. */ + switch (ua_sess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, + node.node) { + health_code_update(); + ret = consumer_flush_channel(socket, ua_chan->key); + if (ret) { + ERR("Error flushing consumer channel"); + retval = -1; + continue; } - /* Continuing flushing all buffers */ - continue; } + break; + case LTTNG_BUFFER_PER_UID: + default: + assert(0); + break; } health_code_update(); +end_deleted: pthread_mutex_unlock(&ua_sess->lock); -end_no_session: - rcu_read_unlock(); - health_code_update(); - return 0; -} - -/* - * Destroy a specific UST session in apps. - */ -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_u64 *node; - - DBG("Destroy tracing for ust app pid %d", app->pid); - - rcu_read_lock(); - - if (!app->compatible) { - goto end; - } - - __lookup_session_by_app(usess, app, &iter); - node = lttng_ht_iter_get_node_u64(&iter); - if (node == NULL) { - /* Session is being or is deleted. */ - goto end; - } - ua_sess = caa_container_of(node, struct ust_app_session, node); - - health_code_update(); - destroy_app_session(app, ua_sess); - health_code_update(); - - /* Quiescent wait after stopping trace */ - ret = ustctl_wait_quiescent(app->sock); - if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { - ERR("UST app wait quiescent failed for app pid %d ret %d", - app->pid, ret); - } -end: +end_not_compatible: rcu_read_unlock(); health_code_update(); - return 0; + return retval; } /* - * Start tracing for the UST session. + * Flush buffers for all applications for a specific UST session. + * Called with UST session lock held. */ -int ust_app_start_trace_all(struct ltt_ust_session *usess) -{ - int ret = 0; - struct lttng_ht_iter iter; - struct ust_app *app; - - DBG("Starting all UST traces"); - - rcu_read_lock(); - - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = ust_app_start_trace(usess, app); - if (ret < 0) { - /* Continue to next apps even on error */ - continue; - } - } - - rcu_read_unlock(); - - return 0; -} +static +int ust_app_flush_session(struct ltt_ust_session *usess) -/* - * Start tracing for the UST session. - */ -int ust_app_stop_trace_all(struct ltt_ust_session *usess) { int ret = 0; - struct lttng_ht_iter iter; - struct ust_app *app; - DBG("Stopping all UST traces"); + DBG("Flushing session buffers for all ust apps"); rcu_read_lock(); - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = ust_app_stop_trace(usess, app); - if (ret < 0) { - /* Continue to next apps even on error */ - continue; - } - } - - /* Flush buffers */ + /* Flush buffers and push metadata. */ switch (usess->buffer_type) { case LTTNG_BUFFER_PER_UID: { struct buffer_reg_uid *reg; + struct lttng_ht_iter iter; /* 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; @@ -3989,98 +4720,351 @@ 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; + /* Push metadata. */ + (void) push_metadata(ust_session_reg, usess->consumer); } break; } case LTTNG_BUFFER_PER_PID: + { + struct ust_app_session *ua_sess; + struct lttng_ht_iter iter; + struct ust_app *app; + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = ust_app_flush_trace(usess, app); - if (ret < 0) { - /* Continue to next apps even on error */ + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { continue; } + (void) ust_app_flush_app_session(app, ua_sess); } break; + } default: + ret = -1; assert(0); break; } rcu_read_unlock(); - - return 0; + health_code_update(); + return ret; } -/* - * Destroy app UST session. - */ -int ust_app_destroy_trace_all(struct ltt_ust_session *usess) +static +int ust_app_clear_quiescent_app_session(struct ust_app *app, + struct ust_app_session *ua_sess) { int ret = 0; struct lttng_ht_iter iter; - struct ust_app *app; + struct ust_app_channel *ua_chan; + struct consumer_socket *socket; - DBG("Destroy all UST traces"); + DBG("Clearing stream quiescent state for ust app pid %d", app->pid); rcu_read_lock(); - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = destroy_trace(usess, app); - if (ret < 0) { - /* Continue to next apps even on error */ - continue; - } + if (!app->compatible) { + goto end_not_compatible; } - rcu_read_unlock(); + pthread_mutex_lock(&ua_sess->lock); - return 0; -} + if (ua_sess->deleted) { + goto end_unlock; + } -/* - * Add channels/events from UST global domain to registered apps at sock. - */ -void ust_app_global_update(struct ltt_ust_session *usess, int sock) -{ - int ret = 0; - struct lttng_ht_iter iter, uiter, iter_ctx; - struct ust_app *app; - struct ust_app_session *ua_sess = NULL; - struct ust_app_channel *ua_chan; - struct ust_app_event *ua_event; - struct ust_app_ctx *ua_ctx; + health_code_update(); - assert(usess); - assert(sock >= 0); + socket = consumer_find_socket_by_bitness(app->bits_per_long, + ua_sess->consumer); + if (!socket) { + ERR("Failed to find consumer (%" PRIu32 ") socket", + app->bits_per_long); + ret = -1; + goto end_unlock; + } + + /* Clear quiescent state. */ + switch (ua_sess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, + ua_chan, node.node) { + health_code_update(); + ret = consumer_clear_quiescent_channel(socket, + ua_chan->key); + if (ret) { + ERR("Error clearing quiescent state for consumer channel"); + ret = -1; + continue; + } + } + break; + case LTTNG_BUFFER_PER_UID: + default: + assert(0); + ret = -1; + break; + } + + health_code_update(); + +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); - DBG2("UST app global update for app sock %d for session id %" PRIu64, sock, - usess->id); +end_not_compatible: + rcu_read_unlock(); + health_code_update(); + return ret; +} + +/* + * Clear quiescent state in each stream for all applications for a + * specific UST session. + * Called with UST session lock held. + */ +static +int ust_app_clear_quiescent_session(struct ltt_ust_session *usess) + +{ + int ret = 0; + + DBG("Clearing stream quiescent state for all ust apps"); rcu_read_lock(); - app = find_app_by_sock(sock); - if (app == NULL) { + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + struct lttng_ht_iter iter; + struct buffer_reg_uid *reg; + /* - * Application can be unregistered before so this is possible hence - * simply stopping the update. + * Clear quiescent for all per UID buffers associated to + * that session. */ - DBG3("UST app update failed to find app sock %d", sock); - goto error; + cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct consumer_socket *socket; + struct buffer_reg_channel *reg_chan; + + /* Get associated consumer socket.*/ + socket = consumer_find_socket_by_bitness( + reg->bits_per_long, usess->consumer); + if (!socket) { + /* + * Ignore request if no consumer is found for + * the session. + */ + continue; + } + + cds_lfht_for_each_entry(reg->registry->channels->ht, + &iter.iter, reg_chan, node.node) { + /* + * The following call will print error values so + * the return code is of little importance + * because whatever happens, we have to try them + * all. + */ + (void) consumer_clear_quiescent_channel(socket, + reg_chan->consumer_key); + } + } + break; + } + case LTTNG_BUFFER_PER_PID: + { + struct ust_app_session *ua_sess; + struct lttng_ht_iter iter; + struct ust_app *app; + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, + pid_n.node) { + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + continue; + } + (void) ust_app_clear_quiescent_app_session(app, + ua_sess); + } + break; + } + default: + ret = -1; + assert(0); + break; } + rcu_read_unlock(); + health_code_update(); + return ret; +} + +/* + * Destroy a specific UST session in apps. + */ +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_u64 *node; + + DBG("Destroy tracing for ust app pid %d", app->pid); + + rcu_read_lock(); + if (!app->compatible) { - goto error; + goto end; + } + + __lookup_session_by_app(usess, app, &iter); + node = lttng_ht_iter_get_node_u64(&iter); + if (node == NULL) { + /* Session is being or is deleted. */ + goto end; + } + ua_sess = caa_container_of(node, struct ust_app_session, node); + + health_code_update(); + destroy_app_session(app, ua_sess); + + health_code_update(); + + /* Quiescent wait after stopping trace */ + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_wait_quiescent(app->sock); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app wait quiescent failed for app pid %d ret %d", + app->pid, ret); + } +end: + rcu_read_unlock(); + health_code_update(); + return 0; +} + +/* + * Start tracing for the UST session. + */ +int ust_app_start_trace_all(struct ltt_ust_session *usess) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + + DBG("Starting all UST traces"); + + rcu_read_lock(); + + /* + * In a start-stop-start use-case, we need to clear the quiescent state + * of each channel set by the prior stop command, thus ensuring that a + * following stop or destroy is sure to grab a timestamp_end near those + * operations, even if the packet is empty. + */ + (void) ust_app_clear_quiescent_session(usess); + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ret = ust_app_start_trace(usess, app); + if (ret < 0) { + /* Continue to next apps even on error */ + continue; + } + } + + rcu_read_unlock(); + + return 0; +} + +/* + * Start tracing for the UST session. + * Called with UST session lock held. + */ +int ust_app_stop_trace_all(struct ltt_ust_session *usess) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + + DBG("Stopping all UST traces"); + + rcu_read_lock(); + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ret = ust_app_stop_trace(usess, app); + if (ret < 0) { + /* Continue to next apps even on error */ + continue; + } } - ret = create_ust_app_session(usess, app, &ua_sess, NULL); + (void) ust_app_flush_session(usess); + + rcu_read_unlock(); + + return 0; +} + +/* + * Destroy app UST session. + */ +int ust_app_destroy_trace_all(struct ltt_ust_session *usess) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + + DBG("Destroy all UST traces"); + + rcu_read_lock(); + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ret = destroy_trace(usess, app); + if (ret < 0) { + /* Continue to next apps even on error */ + continue; + } + } + + rcu_read_unlock(); + + return 0; +} + +static +void ust_app_global_create(struct ltt_ust_session *usess, struct ust_app *app) +{ + int ret = 0; + struct lttng_ht_iter iter, uiter; + struct ust_app_session *ua_sess = NULL; + struct ust_app_channel *ua_chan; + struct ust_app_event *ua_event; + struct ust_app_ctx *ua_ctx; + int is_created = 0; + + ret = find_or_create_ust_app_session(usess, app, &ua_sess, &is_created); if (ret < 0) { /* Tracer is probably gone or ENOMEM. */ goto error; } + if (!is_created) { + /* App session already created. */ + goto end; + } assert(ua_sess); pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } + /* * We can iterate safely here over all UST app session since the create ust * app session above made a shadow copy of the UST global domain from the @@ -4088,35 +5072,24 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) */ cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, node.node) { - /* - * For a metadata channel, handle it differently. - */ - if (!strncmp(ua_chan->name, DEFAULT_METADATA_NAME, - sizeof(ua_chan->name))) { - ret = create_ust_app_metadata(ua_sess, app, usess->consumer, - &ua_chan->attr); - if (ret < 0) { - goto error_unlock; - } - /* Remove it from the hash table and continue!. */ - ret = lttng_ht_del(ua_sess->channels, &iter); - assert(!ret); - delete_ust_app_channel(-1, ua_chan, app); - continue; - } else { - ret = do_create_channel(app, usess, ua_sess, ua_chan); - if (ret < 0) { - /* - * Stop everything. On error, the application failed, no more - * file descriptor are available or ENOMEM so stopping here is - * the only thing we can do for now. - */ - goto error_unlock; - } + ret = do_create_channel(app, usess, ua_sess, ua_chan); + if (ret < 0 && ret != -ENOTCONN) { + /* + * Stop everything. On error, the application + * failed, no more file descriptor are available + * or ENOMEM so stopping here is the only thing + * we can do for now. The only exception is + * -ENOTCONN, which indicates that the application + * has exit. + */ + goto error_unlock; } - 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; @@ -4136,7 +5109,7 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) pthread_mutex_unlock(&ua_sess->lock); - if (usess->start_trace) { + if (usess->active) { ret = ust_app_start_trace(usess, app); if (ret < 0) { goto error; @@ -4144,9 +5117,8 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) DBG2("UST trace started for app pid %d", app->pid); } - +end: /* Everything went well at this point. */ - rcu_read_unlock(); return; error_unlock: @@ -4155,10 +5127,60 @@ error: if (ua_sess) { destroy_app_session(app, ua_sess); } - rcu_read_unlock(); return; } +static +void ust_app_global_destroy(struct ltt_ust_session *usess, struct ust_app *app) +{ + struct ust_app_session *ua_sess; + + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + return; + } + destroy_app_session(app, ua_sess); +} + +/* + * Add channels/events from UST global domain to registered apps at sock. + * + * Called with session lock held. + * Called with RCU read-side lock held. + */ +void ust_app_global_update(struct ltt_ust_session *usess, struct ust_app *app) +{ + assert(usess); + + DBG2("UST app global update for app sock %d for session id %" PRIu64, + app->sock, usess->id); + + if (!app->compatible) { + return; + } + + if (trace_ust_pid_tracker_lookup(usess, app->pid)) { + ust_app_global_create(usess, app); + } else { + ust_app_global_destroy(usess, app); + } +} + +/* + * Called with session lock held. + */ +void ust_app_global_update_all(struct ltt_ust_session *usess) +{ + struct lttng_ht_iter iter; + struct ust_app *app; + + rcu_read_lock(); + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ust_app_global_update(usess, app); + } + rcu_read_unlock(); +} + /* * Add context to a specific channel for global UST domain. */ @@ -4188,6 +5210,12 @@ int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess, } pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + /* Lookup channel in the ust app session */ lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); ua_chan_node = lttng_ht_iter_get_node_str(&uiter); @@ -4246,6 +5274,12 @@ int ust_app_enable_event_pid(struct ltt_ust_session *usess, } pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + ret = 0; + goto end_unlock; + } + /* Lookup channel in the ust app session */ lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter); ua_chan_node = lttng_ht_iter_get_node_str(&iter); @@ -4255,7 +5289,7 @@ 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) { @@ -4275,115 +5309,6 @@ end: return ret; } -/* - * Disable event for a channel from a UST session for a specific PID. - */ -int ust_app_disable_event_pid(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent, pid_t pid) -{ - int ret = 0; - struct lttng_ht_iter iter; - struct lttng_ht_node_str *ua_chan_node, *ua_event_node; - struct ust_app *app; - struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; - struct ust_app_event *ua_event; - - DBG("UST app disabling event %s for PID %d", uevent->attr.name, pid); - - rcu_read_lock(); - - app = ust_app_find_by_pid(pid); - if (app == NULL) { - ERR("UST app disable event per PID %d not found", pid); - ret = -1; - goto error; - } - - if (!app->compatible) { - ret = 0; - goto error; - } - - ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ - goto error; - } - - /* Lookup channel in the ust app session */ - 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) { - /* Channel does not exist, skip disabling */ - goto error; - } - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - - lttng_ht_lookup(ua_chan->events, (void *)uevent->attr.name, &iter); - ua_event_node = lttng_ht_iter_get_node_str(&iter); - if (ua_event_node == NULL) { - /* Event does not exist, skip disabling */ - goto error; - } - ua_event = caa_container_of(ua_event_node, struct ust_app_event, node); - - ret = disable_ust_app_event(ua_sess, ua_event, app); - if (ret < 0) { - goto error; - } - -error: - rcu_read_unlock(); - return ret; -} - -/* - * Calibrate registered applications. - */ -int ust_app_calibrate_glb(struct lttng_ust_calibrate *calibrate) -{ - int ret = 0; - struct lttng_ht_iter iter; - struct ust_app *app; - - rcu_read_lock(); - - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; - } - - health_code_update(); - - ret = ustctl_calibrate(app->sock, calibrate); - if (ret < 0) { - switch (ret) { - case -ENOSYS: - /* Means that it's not implemented on the tracer side. */ - ret = 0; - break; - default: - DBG2("Calibrate app PID %d returned with error %d", - app->pid, ret); - break; - } - } - } - - DBG("UST app global domain calibration finished"); - - rcu_read_unlock(); - - health_code_update(); - - return ret; -} - /* * Receive registration and populate the given msg structure. * @@ -4433,6 +5358,33 @@ error: return ret; } +/* + * Return a ust app session object using the application object and the + * session object descriptor has a key. If not found, NULL is returned. + * A RCU read side lock MUST be acquired when calling this function. +*/ +static struct ust_app_session *find_session_by_objd(struct ust_app *app, + int objd) +{ + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter iter; + struct ust_app_session *ua_sess = NULL; + + assert(app); + + lttng_ht_lookup(app->ust_sessions_objd, (void *)((unsigned long) objd), &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node == NULL) { + DBG2("UST app session find by objd %d not found", objd); + goto error; + } + + ua_sess = caa_container_of(node, struct ust_app_session, ust_objd_node); + +error: + return ua_sess; +} + /* * Return a ust app channel object using the application object and the channel * object descriptor has a key. If not found, NULL is returned. A RCU read side @@ -4486,19 +5438,17 @@ static int reply_ust_register_channel(int sock, int sobjd, int cobjd, /* Lookup application. If not found, there is a code flow error. */ app = find_app_by_notify_sock(sock); if (!app) { - DBG("Application socket %d is being teardown. Abort event notify", + DBG("Application socket %d is being torn down. Abort event notify", sock); ret = 0; - free(fields); goto error_rcu_unlock; } /* Lookup channel by UST object descriptor. */ ua_chan = find_channel_by_objd(app, cobjd); if (!ua_chan) { - DBG("Application channel is being teardown. Abort event notify"); + DBG("Application channel is being torn down. Abort event notify"); ret = 0; - free(fields); goto error_rcu_unlock; } @@ -4507,7 +5457,11 @@ static int reply_ust_register_channel(int sock, int sobjd, int cobjd, /* Get right session registry depending on the session buffer type. */ registry = get_session_registry(ua_sess); - assert(registry); + if (!registry) { + DBG("Application session is being torn down. Abort event notify"); + ret = 0; + goto error_rcu_unlock; + }; /* Depending on the buffer type, a different channel key is used. */ if (ua_sess->buffer_type == LTTNG_BUFFER_PER_UID) { @@ -4531,13 +5485,11 @@ static int reply_ust_register_channel(int sock, int sobjd, int cobjd, chan_reg->nr_ctx_fields = nr_fields; chan_reg->ctx_fields = fields; + fields = NULL; chan_reg->header_type = type; } 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; @@ -4573,9 +5525,7 @@ error: pthread_mutex_unlock(®istry->lock); error_rcu_unlock: rcu_read_unlock(); - if (ret) { - free(fields); - } + free(fields); return ret; } @@ -4589,8 +5539,8 @@ error_rcu_unlock: * On success 0 is returned else a negative value. */ static int add_event_ust_registry(int sock, int sobjd, int cobjd, char *name, - char *sig, size_t nr_fields, struct ustctl_field *fields, int loglevel, - char *model_emf_uri) + char *sig, size_t nr_fields, struct ustctl_field *fields, + int loglevel_value, char *model_emf_uri) { int ret, ret_code; uint32_t event_id = 0; @@ -4605,23 +5555,17 @@ static int add_event_ust_registry(int sock, int sobjd, int cobjd, char *name, /* Lookup application. If not found, there is a code flow error. */ app = find_app_by_notify_sock(sock); if (!app) { - DBG("Application socket %d is being teardown. Abort event notify", + DBG("Application socket %d is being torn down. Abort event notify", sock); ret = 0; - free(sig); - free(fields); - free(model_emf_uri); goto error_rcu_unlock; } /* Lookup channel by UST object descriptor. */ ua_chan = find_channel_by_objd(app, cobjd); if (!ua_chan) { - DBG("Application channel is being teardown. Abort event notify"); + DBG("Application channel is being torn down. Abort event notify"); ret = 0; - free(sig); - free(fields); - free(model_emf_uri); goto error_rcu_unlock; } @@ -4629,7 +5573,11 @@ static int add_event_ust_registry(int sock, int sobjd, int cobjd, char *name, ua_sess = ua_chan->session; registry = get_session_registry(ua_sess); - assert(registry); + if (!registry) { + DBG("Application session is being torn down. Abort event notify"); + ret = 0; + goto error_rcu_unlock; + } if (ua_sess->buffer_type == LTTNG_BUFFER_PER_UID) { chan_reg_key = ua_chan->tracing_channel_id; @@ -4640,35 +5588,126 @@ 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. + * 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_value, model_emf_uri, ua_sess->buffer_type, + &event_id, app); + sig = NULL; + fields = NULL; + model_emf_uri = NULL; + + /* + * The return value is returned to ustctl so in case of an error, the + * application can be notified. In case of an error, it's important not to + * return a negative error or else the application will get closed. + */ + ret = ustctl_reply_register_event(sock, event_id, ret_code); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app reply event failed with ret %d", ret); + } else { + DBG3("UST app reply event failed. Application died"); + } + /* + * No need to wipe the create event since the application socket will + * get close on error hence cleaning up everything by itself. + */ + goto error; + } + + DBG3("UST registry event %s with id %" PRId32 " added successfully", + name, event_id); + +error: + pthread_mutex_unlock(®istry->lock); +error_rcu_unlock: + rcu_read_unlock(); + free(sig); + free(fields); + free(model_emf_uri); + return ret; +} + +/* + * Add enum to the UST session registry. Once done, this replies to the + * application with the appropriate error code. + * + * The session UST registry lock is acquired within this function. + * + * On success 0 is returned else a negative value. + */ +static int add_enum_ust_registry(int sock, int sobjd, char *name, + struct ustctl_enum_entry *entries, size_t nr_entries) +{ + int ret = 0, ret_code; + struct ust_app *app; + struct ust_app_session *ua_sess; + struct ust_registry_session *registry; + uint64_t enum_id = -1ULL; + + rcu_read_lock(); + + /* Lookup application. If not found, there is a code flow error. */ + app = find_app_by_notify_sock(sock); + if (!app) { + /* Return an error since this is not an error */ + DBG("Application socket %d is being torn down. Aborting enum registration", + sock); + free(entries); + goto error_rcu_unlock; + } + + /* Lookup session by UST object descriptor. */ + ua_sess = find_session_by_objd(app, sobjd); + if (!ua_sess) { + /* Return an error since this is not an error */ + DBG("Application session is being torn down (session not found). Aborting enum registration."); + free(entries); + goto error_rcu_unlock; + } + + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down (registry not found). Aborting enum registration."); + free(entries); + goto error_rcu_unlock; + } + + pthread_mutex_lock(®istry->lock); + + /* + * From this point on, the callee acquires the ownership of + * entries. The variable entries MUST NOT be read/written after + * call. */ - 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); + ret_code = ust_registry_create_or_find_enum(registry, sobjd, name, + entries, nr_entries, &enum_id); + entries = NULL; /* * The return value is returned to ustctl so in case of an error, the * application can be notified. In case of an error, it's important not to * return a negative error or else the application will get closed. */ - ret = ustctl_reply_register_event(sock, event_id, ret_code); + ret = ustctl_reply_register_enum(sock, enum_id, ret_code); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { - ERR("UST app reply event failed with ret %d", ret); + ERR("UST app reply enum failed with ret %d", ret); } else { - DBG3("UST app reply event failed. Application died"); + DBG3("UST app reply enum failed. Application died"); } /* - * No need to wipe the create event since the application socket will + * No need to wipe the create enum since the application socket will * get close on error hence cleaning up everything by itself. */ goto error; } - DBG3("UST registry event %s with id %" PRId32 " added successfully", - name, event_id); + DBG3("UST registry enum %s added successfully or already found", name); error: pthread_mutex_unlock(®istry->lock); @@ -4702,15 +5741,16 @@ int ust_app_recv_notify(int sock) switch (cmd) { case USTCTL_NOTIFY_CMD_EVENT: { - int sobjd, cobjd, loglevel; + int sobjd, cobjd, loglevel_value; char name[LTTNG_UST_SYM_NAME_LEN], *sig, *model_emf_uri; size_t nr_fields; struct ustctl_field *fields; DBG2("UST app ustctl register event received"); - ret = ustctl_recv_register_event(sock, &sobjd, &cobjd, name, &loglevel, - &sig, &nr_fields, &fields, &model_emf_uri); + ret = ustctl_recv_register_event(sock, &sobjd, &cobjd, name, + &loglevel_value, &sig, &nr_fields, &fields, + &model_emf_uri); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app recv event failed with ret %d", ret); @@ -4727,7 +5767,7 @@ int ust_app_recv_notify(int sock) * to the this function. */ ret = add_event_ust_registry(sock, sobjd, cobjd, name, sig, nr_fields, - fields, loglevel, model_emf_uri); + fields, loglevel_value, model_emf_uri); if (ret < 0) { goto error; } @@ -4766,212 +5806,512 @@ int ust_app_recv_notify(int sock) break; } + case USTCTL_NOTIFY_CMD_ENUM: + { + int sobjd; + char name[LTTNG_UST_SYM_NAME_LEN]; + size_t nr_entries; + struct ustctl_enum_entry *entries; + + DBG2("UST app ustctl register enum received"); + + ret = ustctl_recv_register_enum(sock, &sobjd, name, + &entries, &nr_entries); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app recv enum failed with ret %d", ret); + } else { + DBG3("UST app recv enum failed. Application died"); + } + goto error; + } + + /* Callee assumes ownership of entries */ + ret = add_enum_ust_registry(sock, sobjd, name, + entries, nr_entries); + if (ret < 0) { + goto error; + } + + break; + } + default: + /* Should NEVER happen. */ + assert(0); + } + +error: + return ret; +} + +/* + * Once the notify socket hangs up, this is called. First, it tries to find the + * corresponding application. On failure, the call_rcu to close the socket is + * executed. If an application is found, it tries to delete it from the notify + * socket hash table. Whathever the result, it proceeds to the call_rcu. + * + * Note that an object needs to be allocated here so on ENOMEM failure, the + * call RCU is not done but the rest of the cleanup is. + */ +void ust_app_notify_sock_unregister(int sock) +{ + int err_enomem = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + struct ust_app_notify_sock_obj *obj; + + assert(sock >= 0); + + rcu_read_lock(); + + obj = zmalloc(sizeof(*obj)); + if (!obj) { + /* + * An ENOMEM is kind of uncool. If this strikes we continue the + * procedure but the call_rcu will not be called. In this case, we + * accept the fd leak rather than possibly creating an unsynchronized + * state between threads. + * + * TODO: The notify object should be created once the notify socket is + * registered and stored independantely from the ust app object. The + * tricky part is to synchronize the teardown of the application and + * this notify object. Let's keep that in mind so we can avoid this + * kind of shenanigans with ENOMEM in the teardown path. + */ + err_enomem = 1; + } else { + obj->fd = sock; + } + + DBG("UST app notify socket unregister %d", sock); + + /* + * Lookup application by notify socket. If this fails, this means that the + * hash table delete has already been done by the application + * unregistration process so we can safely close the notify socket in a + * call RCU. + */ + app = find_app_by_notify_sock(sock); + if (!app) { + goto close_socket; + } + + iter.iter.node = &app->notify_sock_n.node; + + /* + * Whatever happens here either we fail or succeed, in both cases we have + * to close the socket after a grace period to continue to the call RCU + * here. If the deletion is successful, the application is not visible + * anymore by other threads and is it fails it means that it was already + * deleted from the hash table so either way we just have to close the + * socket. + */ + (void) lttng_ht_del(ust_app_ht_by_notify_sock, &iter); + +close_socket: + rcu_read_unlock(); + + /* + * Close socket after a grace period to avoid for the socket to be reused + * before the application object is freed creating potential race between + * threads trying to add unique in the global hash table. + */ + if (!err_enomem) { + 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, + uint64_t nb_packets_per_stream) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + char pathname[PATH_MAX]; + + assert(usess); + assert(output); + + rcu_read_lock(); + + 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) { + ret = consumer_snapshot_channel(socket, reg_chan->consumer_key, + output, 0, usess->uid, usess->gid, pathname, wait, + nb_packets_per_stream); + if (ret < 0) { + goto error; + } + } + ret = consumer_snapshot_channel(socket, + reg->registry->reg.ust->metadata_key, output, 1, + usess->uid, usess->gid, pathname, wait, 0); + 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) { + ret = consumer_snapshot_channel(socket, ua_chan->key, output, + 0, ua_sess->euid, ua_sess->egid, pathname, wait, + nb_packets_per_stream); + if (ret < 0) { + goto error; + } + } + + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Abort snapshot record."); + ret = -1; + goto error; + } + ret = consumer_snapshot_channel(socket, registry->metadata_key, output, + 1, ua_sess->euid, ua_sess->egid, pathname, wait, 0); + if (ret < 0) { + goto error; + } + } + break; + } + default: + assert(0); + break; + } + +error: + rcu_read_unlock(); + return ret; +} + +/* + * Return the size taken by one more packet per stream. + */ +uint64_t ust_app_get_size_one_more_packet_per_stream(struct ltt_ust_session *usess, + uint64_t cur_nr_packets) +{ + uint64_t tot_size = 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; + + rcu_read_lock(); + cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, + reg_chan, node.node) { + if (cur_nr_packets >= reg_chan->num_subbuf) { + /* + * Don't take channel into account if we + * already grab all its packets. + */ + continue; + } + tot_size += reg_chan->subbuf_size * reg_chan->stream_count; + } + rcu_read_unlock(); + } + 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) { + if (cur_nr_packets >= ua_chan->attr.num_subbuf) { + /* + * Don't take channel into account if we + * already grab all its packets. + */ + continue; + } + tot_size += ua_chan->attr.subbuf_size * ua_chan->streams.count; + } + } + rcu_read_unlock(); + break; + } default: - /* Should NEVER happen. */ assert(0); + break; } -error: - return ret; + return tot_size; } -/* - * Once the notify socket hangs up, this is called. First, it tries to find the - * corresponding application. On failure, the call_rcu to close the socket is - * executed. If an application is found, it tries to delete it from the notify - * socket hash table. Whathever the result, it proceeds to the call_rcu. - * - * Note that an object needs to be allocated here so on ENOMEM failure, the - * call RCU is not done but the rest of the cleanup is. - */ -void ust_app_notify_sock_unregister(int sock) +int ust_app_uid_get_channel_runtime_stats(uint64_t ust_session_id, + struct cds_list_head *buffer_reg_uid_list, + struct consumer_output *consumer, uint64_t uchan_id, + int overwrite, uint64_t *discarded, uint64_t *lost) { - int err_enomem = 0; - struct lttng_ht_iter iter; - struct ust_app *app; - struct ust_app_notify_sock_obj *obj; + int ret; + uint64_t consumer_chan_key; - assert(sock >= 0); + *discarded = 0; + *lost = 0; - rcu_read_lock(); + ret = buffer_reg_uid_consumer_channel_key( + buffer_reg_uid_list, uchan_id, &consumer_chan_key); + if (ret < 0) { + /* Not found */ + ret = 0; + goto end; + } - obj = zmalloc(sizeof(*obj)); - if (!obj) { - /* - * An ENOMEM is kind of uncool. If this strikes we continue the - * procedure but the call_rcu will not be called. In this case, we - * accept the fd leak rather than possibly creating an unsynchronized - * state between threads. - * - * TODO: The notify object should be created once the notify socket is - * registered and stored independantely from the ust app object. The - * tricky part is to synchronize the teardown of the application and - * this notify object. Let's keep that in mind so we can avoid this - * kind of shenanigans with ENOMEM in the teardown path. - */ - err_enomem = 1; + if (overwrite) { + ret = consumer_get_lost_packets(ust_session_id, + consumer_chan_key, consumer, lost); } else { - obj->fd = sock; + ret = consumer_get_discarded_events(ust_session_id, + consumer_chan_key, consumer, discarded); } - DBG("UST app notify socket unregister %d", sock); +end: + return ret; +} - /* - * Lookup application by notify socket. If this fails, this means that the - * hash table delete has already been done by the application - * unregistration process so we can safely close the notify socket in a - * call RCU. - */ - app = find_app_by_notify_sock(sock); - if (!app) { - goto close_socket; - } +int ust_app_pid_get_channel_runtime_stats(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan, + struct consumer_output *consumer, int overwrite, + uint64_t *discarded, uint64_t *lost) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct lttng_ht_node_str *ua_chan_node; + struct ust_app *app; + struct ust_app_session *ua_sess; + struct ust_app_channel *ua_chan; - iter.iter.node = &app->notify_sock_n.node; + *discarded = 0; + *lost = 0; + rcu_read_lock(); /* - * Whatever happens here either we fail or succeed, in both cases we have - * to close the socket after a grace period to continue to the call RCU - * here. If the deletion is successful, the application is not visible - * anymore by other threads and is it fails it means that it was already - * deleted from the hash table so either way we just have to close the - * socket. + * Iterate over every registered applications. Sum counters for + * all applications containing requested session and channel. */ - (void) lttng_ht_del(ust_app_ht_by_notify_sock, &iter); + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct lttng_ht_iter uiter; -close_socket: - rcu_read_unlock(); + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + continue; + } - /* - * Close socket after a grace period to avoid for the socket to be reused - * before the application object is freed creating potential race between - * threads trying to add unique in the global hash table. - */ - if (!err_enomem) { - call_rcu(&obj->head, close_notify_sock_rcu); + /* Get channel */ + lttng_ht_lookup(ua_sess->channels, (void *) uchan->name, &uiter); + ua_chan_node = lttng_ht_iter_get_node_str(&uiter); + /* If the session is found for the app, the channel must be there */ + assert(ua_chan_node); + + ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); + + if (overwrite) { + uint64_t _lost; + + ret = consumer_get_lost_packets(usess->id, ua_chan->key, + consumer, &_lost); + if (ret < 0) { + break; + } + (*lost) += _lost; + } else { + uint64_t _discarded; + + ret = consumer_get_discarded_events(usess->id, + ua_chan->key, consumer, &_discarded); + if (ret < 0) { + break; + } + (*discarded) += _discarded; + } } + + rcu_read_unlock(); + return ret; } -/* - * Destroy a ust app data structure and free its memory. - */ -void ust_app_destroy(struct ust_app *app) +static +int ust_app_regenerate_statedump(struct ltt_ust_session *usess, + struct ust_app *app) { - if (!app) { - return; + int ret = 0; + struct ust_app_session *ua_sess; + + DBG("Regenerating the metadata for ust app pid %d", app->pid); + + rcu_read_lock(); + + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + /* The session is in teardown process. Ignore and continue. */ + goto end; } - call_rcu(&app->pid_n.head, delete_ust_app_rcu); + pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + goto end_unlock; + } + + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_regenerate_statedump(app->sock, ua_sess->handle); + pthread_mutex_unlock(&app->sock_lock); + +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); + +end: + rcu_read_unlock(); + health_code_update(); + return ret; } /* - * 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. + * Regenerate the statedump for each app in the session. */ -int ust_app_snapshot_record(struct ltt_ust_session *usess, - struct snapshot_output *output, int wait, unsigned int nb_streams) +int ust_app_regenerate_statedump_all(struct ltt_ust_session *usess) { 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); + DBG("Regenerating the metadata for all UST apps"); 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; - } - 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. */ + if (!app->compatible) { 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 (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); + ret = ust_app_regenerate_statedump(usess, app); if (ret < 0) { - goto error; + /* Continue to the next app even on error */ + continue; } - } -error: rcu_read_unlock(); - return ret; + + return 0; } /* - * Return the number of streams for a UST session. + * Rotate all the channels of a session. + * + * Return 0 on success or else a negative value. */ -unsigned int ust_app_get_nb_stream(struct ltt_ust_session *usess) +int ust_app_rotate_session(struct ltt_session *session, bool *ust_active) { - unsigned int ret = 0; - struct ust_app *app; + int ret = 0; struct lttng_ht_iter iter; + struct ust_app *app; + struct ltt_ust_session *usess = session->ust_session; + char pathname[LTTNG_PATH_MAX]; assert(usess); + rcu_read_lock(); + switch (usess->buffer_type) { case LTTNG_BUFFER_PER_UID: { @@ -4979,34 +6319,168 @@ unsigned int ust_app_get_nb_stream(struct ltt_ust_session *usess) 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; + } + + /* + * Account the metadata channel first to make sure the + * number of channels waiting for a rotation cannot + * reach 0 before we complete the iteration over all + * the channels. + */ + ret = rotate_add_channel_pending( + reg->registry->reg.ust->metadata_key, + LTTNG_DOMAIN_UST, session); + if (ret < 0) { + ret = reg->bits_per_long == 32 ? + -LTTNG_ERR_UST_CONSUMER32_FAIL : + -LTTNG_ERR_UST_CONSUMER64_FAIL; + goto error; + } + + ret = snprintf(pathname, sizeof(pathname), + DEFAULT_UST_TRACE_DIR "/" DEFAULT_UST_TRACE_UID_PATH, + reg->uid, reg->bits_per_long); + if (ret < 0 || ret == sizeof(pathname)) { + PERROR("Failed to format rotation path"); + goto error; + } + /* Rotate the data channels. */ cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, reg_chan, node.node) { - ret += reg_chan->stream_count; + ret = rotate_add_channel_pending( + reg_chan->consumer_key, + LTTNG_DOMAIN_UST, session); + if (ret < 0) { + ret = reg->bits_per_long == 32 ? + -LTTNG_ERR_UST_CONSUMER32_FAIL : + -LTTNG_ERR_UST_CONSUMER64_FAIL; + goto error; + } + ret = consumer_rotate_channel(socket, + reg_chan->consumer_key, + usess->uid, usess->gid, + usess->consumer, pathname, + /* is_metadata_channel */ false, + session->current_archive_id, + &session->rotate_pending_relay); + if (ret < 0) { + goto error; + } + } + + (void) push_metadata(reg->registry->reg.ust, usess->consumer); + + ret = consumer_rotate_channel(socket, + reg->registry->reg.ust->metadata_key, + usess->uid, usess->gid, + usess->consumer, pathname, + /* is_metadata_channel */ true, + session->current_archive_id, + &session->rotate_pending_relay); + if (ret < 0) { + goto error; } + *ust_active = true; } 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 consumer_socket *socket; + struct lttng_ht_iter chan_iter; struct ust_app_channel *ua_chan; struct ust_app_session *ua_sess; - struct lttng_ht_iter chan_iter; + struct ust_registry_session *registry; ua_sess = lookup_session_by_app(usess, app); if (!ua_sess) { /* Session not associated with this app. */ continue; } + ret = snprintf(pathname, sizeof(pathname), + DEFAULT_UST_TRACE_DIR "/%s", + ua_sess->path); + if (ret < 0 || ret == sizeof(pathname)) { + PERROR("Failed to format rotation path"); + goto error; + } + + /* Get the right consumer socket for the application. */ + socket = consumer_find_socket_by_bitness(app->bits_per_long, + usess->consumer); + if (!socket) { + ret = -EINVAL; + goto error; + } + + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Abort snapshot record."); + ret = -1; + goto error; + } + /* + * Account the metadata channel first to make sure the + * number of channels waiting for a rotation cannot + * reach 0 before we complete the iteration over all + * the channels. + */ + ret = rotate_add_channel_pending(registry->metadata_key, + LTTNG_DOMAIN_UST, session); + if (ret < 0) { + ret = app->bits_per_long == 32 ? + -LTTNG_ERR_UST_CONSUMER32_FAIL : + -LTTNG_ERR_UST_CONSUMER64_FAIL; + goto error; + } + + /* Rotate the data channels. */ cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, ua_chan, node.node) { - ret += ua_chan->streams.count; + ret = rotate_add_channel_pending( + ua_chan->key, LTTNG_DOMAIN_UST, + session); + if (ret < 0) { + ret = app->bits_per_long == 32 ? + -LTTNG_ERR_UST_CONSUMER32_FAIL : + -LTTNG_ERR_UST_CONSUMER64_FAIL; + goto error; + } + ret = consumer_rotate_channel(socket, ua_chan->key, + ua_sess->euid, ua_sess->egid, + ua_sess->consumer, pathname, + /* is_metadata_channel */ false, + session->current_archive_id, + &session->rotate_pending_relay); + if (ret < 0) { + goto error; + } } + + /* Rotate the metadata channel. */ + (void) push_metadata(registry, usess->consumer); + ret = consumer_rotate_channel(socket, registry->metadata_key, + ua_sess->euid, ua_sess->egid, + ua_sess->consumer, pathname, + /* is_metadata_channel */ true, + session->current_archive_id, + &session->rotate_pending_relay); + if (ret < 0) { + goto error; + } + *ust_active = true; } - rcu_read_unlock(); break; } default: @@ -5014,5 +6488,9 @@ unsigned int ust_app_get_nb_stream(struct ltt_ust_session *usess) break; } + ret = LTTNG_OK; + +error: + rcu_read_unlock(); return ret; }