X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fust-app.c;h=73d1177be5643cec31e9853bac8c375beb894c9f;hb=9c9abe17599b58b2f8348c23f1dad263e5965cf8;hp=cc611fcf27ad2e85fcc45731827c366b10857629;hpb=1af53eb511455656ea8be615ddb5620adfe66008;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c index cc611fcf2..73d1177be 100644 --- a/src/bin/lttng-sessiond/ust-app.c +++ b/src/bin/lttng-sessiond/ust-app.c @@ -40,6 +40,9 @@ #include "ust-ctl.h" #include "utils.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; static pthread_mutex_t next_channel_key_lock = PTHREAD_MUTEX_INITIALIZER; @@ -181,6 +184,7 @@ static void add_unique_ust_app_event(struct ust_app_channel *ua_chan, key.name = event->attr.name; key.filter = event->filter; key.loglevel = event->attr.loglevel; + key.exclusion = event->exclusion; node_ptr = cds_lfht_add_unique(ht->ht, ht->hash_fct(event->node.key, lttng_ht_seed), @@ -257,14 +261,17 @@ 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); @@ -279,16 +286,20 @@ 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); @@ -303,14 +314,17 @@ 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); @@ -327,11 +341,12 @@ 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); } @@ -374,7 +389,7 @@ 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 */ @@ -382,7 +397,7 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, 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 */ @@ -390,7 +405,7 @@ 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) { @@ -404,8 +419,11 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *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); @@ -416,42 +434,82 @@ 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; 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 (!metadata_key) { + return 0; + } + + /* + * 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), or because the application has + * exited. 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. */ if (registry->metadata_closed) { return -EPIPE; } - 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; if (len == 0) { DBG3("No metadata to push for metadata key %" PRIu64, registry->metadata_key); @@ -470,38 +528,99 @@ 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, + /* + * 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); + 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) @@ -513,15 +632,10 @@ 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.*/ @@ -529,46 +643,26 @@ 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. @@ -584,6 +678,8 @@ 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; @@ -610,6 +706,7 @@ error: */ registry->metadata_closed = 1; end: + pthread_mutex_unlock(®istry->lock); rcu_read_unlock(); return ret; } @@ -647,8 +744,11 @@ 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) { + if (registry) { /* Push metadata for application before freeing the application. */ (void) push_metadata(registry, ua_sess->consumer); @@ -658,8 +758,7 @@ 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); } @@ -682,7 +781,9 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, } 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); @@ -690,6 +791,8 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, } pthread_mutex_unlock(&ua_sess->lock); + consumer_output_put(ua_sess->consumer); + call_rcu(&ua_sess->rcu_head, delete_ust_app_session_rcu); } @@ -803,6 +906,7 @@ 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; @@ -1020,7 +1124,8 @@ error: * Return an ust_app_event object or NULL on error. */ static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht, - char *name, struct lttng_ust_filter_bytecode *filter, int loglevel) + char *name, struct lttng_ust_filter_bytecode *filter, int loglevel, + const struct lttng_event_exclusion *exclusion) { struct lttng_ht_iter iter; struct lttng_ht_node_str *node; @@ -1034,6 +1139,8 @@ static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht, key.name = name; key.filter = filter; key.loglevel = loglevel; + /* lttng_event_exclusion and lttng_ust_event_exclusion structures are similar */ + key.exclusion = (struct lttng_ust_event_exclusion *)exclusion; /* Lookup using the event name as hash and a custom match fct. */ cds_lfht_lookup(ht->ht, ht->hash_fct((void *) name, lttng_ht_seed), @@ -1062,8 +1169,10 @@ 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) " @@ -1106,8 +1215,10 @@ int set_ust_event_filter(struct ust_app_event *ua_event, goto error; } + pthread_mutex_lock(&app->sock_lock); ret = ustctl_set_filter(app->sock, ua_event->filter, 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) " @@ -1131,6 +1242,48 @@ error: return ret; } +/* + * Set event exclusions on the tracer. + */ +static +int set_ust_event_exclusion(struct ust_app_event *ua_event, + struct ust_app *app) +{ + int ret; + + health_code_update(); + + if (!ua_event->exclusion || !ua_event->exclusion->count) { + ret = 0; + goto error; + } + + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_set_exclusion(app->sock, ua_event->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(); + return ret; +} + /* * Disable the specified event on to UST tracer for the UST session. */ @@ -1141,7 +1294,9 @@ 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) " @@ -1177,7 +1332,9 @@ 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) " @@ -1213,7 +1370,9 @@ 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) " @@ -1251,7 +1410,9 @@ 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) " @@ -1299,7 +1460,10 @@ 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; } @@ -1308,12 +1472,15 @@ 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; @@ -1337,8 +1504,10 @@ 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", @@ -1370,8 +1539,41 @@ 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) { + 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 enable call. If + * we get an EEXIST, there is a problem on the tracer side since we + * just created it. + */ + switch (ret) { + case -LTTNG_UST_ERR_PERM: + /* Code flow problem */ + assert(0); + case -LTTNG_UST_ERR_EXIST: + /* It's OK for our use case. */ + ret = 0; + break; + default: + break; + } + goto error; + } + } else { ret = disable_ust_event(app, ua_sess, ua_event); if (ret < 0) { /* @@ -1405,6 +1607,8 @@ 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'; @@ -1418,6 +1622,19 @@ static void shadow_copy_event(struct ust_app_event *ua_event, ua_event->filter = alloc_copy_ust_app_filter(uevent->filter); /* Filter might be NULL here in case of ENONEM. */ } + + /* Copy exclusion data */ + if (uevent->exclusion) { + exclusion_alloc_size = sizeof(struct lttng_ust_event_exclusion) + + LTTNG_UST_SYM_NAME_LEN * uevent->exclusion->count; + ua_event->exclusion = zmalloc(exclusion_alloc_size); + if (ua_event->exclusion == NULL) { + PERROR("malloc"); + } else { + memcpy(ua_event->exclusion, uevent->exclusion, + exclusion_alloc_size); + } + } } /* @@ -1462,14 +1679,14 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, } lttng_ht_node_init_ulong(&ua_ctx->node, (unsigned long) ua_ctx->ctx.ctx); - lttng_ht_add_unique_ulong(ua_chan->ctx, &ua_ctx->node); + 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); @@ -1515,10 +1732,15 @@ 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: @@ -1569,9 +1791,10 @@ 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); } /* @@ -1782,7 +2005,9 @@ 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", @@ -1831,6 +2056,75 @@ error: return ret; } +/* + * Match function for a hash table lookup of ust_app_ctx. + * + * It matches an ust app context based on the context type and, in the case + * of perf counters, their name. + */ +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 *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; + } + + /* Check the name in the case of perf thread counters. */ + if (key->ctx == 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; + } + } + + /* 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 *uctx) +{ + 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. * @@ -1842,15 +2136,12 @@ int create_ust_app_channel_context(struct ust_app_session *ua_sess, struct ust_app *app) { int ret = 0; - struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; 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; } @@ -1863,7 +2154,7 @@ int create_ust_app_channel_context(struct ust_app_session *ua_sess, } lttng_ht_node_init_ulong(&ua_ctx->node, (unsigned long) ua_ctx->ctx.ctx); - lttng_ht_add_unique_ulong(ua_chan->ctx, &ua_ctx->node); + 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); @@ -2142,7 +2433,8 @@ 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; @@ -2171,7 +2463,7 @@ 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: @@ -2205,6 +2497,7 @@ static int create_buffer_reg_channel(struct buffer_reg_session *reg_sess, assert(reg_chan); reg_chan->consumer_key = ua_chan->key; reg_chan->subbuf_size = ua_chan->attr.subbuf_size; + 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, @@ -2234,7 +2527,8 @@ 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; @@ -2246,7 +2540,7 @@ 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; } @@ -2288,7 +2582,10 @@ 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; } @@ -2306,7 +2603,10 @@ 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; } @@ -2314,7 +2614,7 @@ 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; @@ -2358,6 +2658,8 @@ static int create_channel_per_uid(struct ust_app *app, /* 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; } assert(reg_chan); @@ -2369,6 +2671,9 @@ static int create_channel_per_uid(struct ust_app *app, ret = do_consumer_create_channel(usess, ua_sess, ua_chan, app->bits_per_long, reg_uid->registry->reg.ust); if (ret < 0) { + ERR("Error creating UST channel \"%s\" on the consumer daemon", + ua_chan->name); + /* * Let's remove the previously created buffer registry channel so * it's not visible anymore in the session registry. @@ -2383,8 +2688,11 @@ static int create_channel_per_uid(struct ust_app *app, /* * Setup the streams and add it to the session registry. */ - ret = setup_buffer_reg_channel(reg_uid->registry, ua_chan, reg_chan); + 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; } @@ -2393,6 +2701,9 @@ static int create_channel_per_uid(struct ust_app *app, /* 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; } @@ -2427,6 +2738,8 @@ static int create_channel_per_pid(struct ust_app *app, /* 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; } @@ -2434,11 +2747,16 @@ 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; } @@ -2452,7 +2770,8 @@ 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. + * 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, @@ -2511,7 +2830,8 @@ 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, @@ -2582,7 +2902,7 @@ 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; @@ -2615,7 +2935,7 @@ 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; } @@ -2625,8 +2945,7 @@ 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; @@ -2640,6 +2959,8 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, registry = get_session_registry(ua_sess); 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; @@ -2654,20 +2975,7 @@ 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); @@ -2725,17 +3033,10 @@ 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. @@ -2825,10 +3126,10 @@ 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; } @@ -2880,12 +3181,14 @@ 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); } } @@ -2902,6 +3205,7 @@ 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; @@ -2909,39 +3213,19 @@ 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; @@ -2952,12 +3236,21 @@ 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 @@ -2970,7 +3263,7 @@ 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); @@ -2980,36 +3273,46 @@ 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); } - /* Free memory */ - call_rcu(<a->pid_n.head, delete_ust_app_rcu); + /* Remove application from PID hash table */ + ret = lttng_ht_del(ust_app_ht_by_sock, &ust_app_sock_iter); + assert(!ret); - rcu_read_unlock(); - return; -} + /* + * 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); -/* - * Return traceable_app_count - */ -unsigned long ust_app_list_count(void) -{ - unsigned long count; + /* + * 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); + } - rcu_read_lock(); - count = lttng_ht_get_count(ust_app_ht); - rcu_read_unlock(); + /* Free memory */ + call_rcu(<a->pid_n.head, delete_ust_app_rcu); - return count; + rcu_read_unlock(); + return; } /* @@ -3045,12 +3348,14 @@ 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; } @@ -3058,8 +3363,9 @@ 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 { @@ -3071,25 +3377,48 @@ int ust_app_list_events(struct lttng_event **events) */ 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; @@ -3098,6 +3427,11 @@ 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; @@ -3145,12 +3479,14 @@ 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; } @@ -3158,8 +3494,9 @@ 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 { @@ -3167,42 +3504,73 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) /* * This is normal behavior, an application can die during the * creation process. Don't report an error so the execution can - * continue normally. + * 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; } 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; @@ -3390,7 +3758,7 @@ 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; @@ -3427,14 +3795,14 @@ 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) { @@ -3448,65 +3816,6 @@ int ust_app_disable_event_glb(struct ltt_ust_session *usess, return ret; } -/* - * For a specific UST session and UST channel, the event for all - * registered apps. - */ -int ust_app_disable_all_event_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; - 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 all event for all apps in channel " - "%s for session id %" PRIu64, uchan->name, usess->id); - - rcu_read_lock(); - - /* For all 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; - } - ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ - 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. */ @@ -3550,6 +3859,7 @@ 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: @@ -3559,12 +3869,16 @@ 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, @@ -3572,14 +3886,23 @@ 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; + } } } @@ -3630,17 +3953,29 @@ 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); @@ -3696,6 +4031,12 @@ 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); @@ -3747,6 +4088,11 @@ 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; @@ -3758,7 +4104,7 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) ret = run_as_mkdir_recursive(usess->consumer->dst.trace_path, S_IRWXU | S_IRWXG, ua_sess->euid, ua_sess->egid); if (ret < 0) { - if (ret != -EEXIST) { + if (errno != EEXIST) { ERR("Trace directory creation error"); goto error_unlock; } @@ -3769,7 +4115,7 @@ 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; } @@ -3778,7 +4124,9 @@ 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)", @@ -3804,7 +4152,9 @@ 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); @@ -3847,6 +4197,11 @@ 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 @@ -3860,7 +4215,9 @@ 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)", @@ -3880,7 +4237,9 @@ 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); @@ -3891,10 +4250,8 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) registry = get_session_registry(ua_sess); 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); @@ -3910,65 +4267,291 @@ error_rcu_unlock: return -1; } +static +int ust_app_flush_app_session(struct ust_app *app, + struct ust_app_session *ua_sess) +{ + int ret, retval = 0; + struct lttng_ht_iter iter; + struct ust_app_channel *ua_chan; + struct consumer_socket *socket; + + DBG("Flushing app session buffers for ust app pid %d", app->pid); + + rcu_read_lock(); + + if (!app->compatible) { + goto end_not_compatible; + } + + pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + goto end_deleted; + } + + health_code_update(); + + /* Flushing buffers */ + 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(); + assert(ua_chan->is_sent); + ret = consumer_flush_channel(socket, ua_chan->key); + if (ret) { + ERR("Error flushing consumer channel"); + retval = -1; + continue; + } + } + break; + case LTTNG_BUFFER_PER_UID: + default: + assert(0); + break; + } + + health_code_update(); + +end_deleted: + pthread_mutex_unlock(&ua_sess->lock); + +end_not_compatible: + rcu_read_unlock(); + health_code_update(); + return retval; +} + /* - * Flush buffers for a specific UST session and app. + * Flush buffers for all applications for a specific UST session. + * Called with UST session lock held. */ static -int ust_app_flush_trace(struct ltt_ust_session *usess, struct ust_app *app) +int ust_app_flush_session(struct ltt_ust_session *usess) + +{ + int ret = 0; + + DBG("Flushing session buffers for all ust apps"); + + rcu_read_lock(); + + /* 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; + + /* Get consumer socket to use to push the metadata.*/ + 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_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) { + 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(); + health_code_update(); + return ret; +} + +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_session *ua_sess; struct ust_app_channel *ua_chan; + struct consumer_socket *socket; - DBG("Flushing buffers for ust app pid %d", app->pid); + DBG("Clearing stream quiescent state for ust app pid %d", app->pid); rcu_read_lock(); if (!app->compatible) { - goto end_no_session; + goto end_not_compatible; } - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - goto end_no_session; + pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + goto end_unlock; } - pthread_mutex_lock(&ua_sess->lock); + health_code_update(); + + 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(); - /* 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); +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); + +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(); + + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + struct lttng_ht_iter iter; + struct buffer_reg_uid *reg; + + /* + * Clear quiescent for all per UID buffers associated to + * that session. + */ + 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) { /* - * This is normal behavior, an application can die during the - * creation process. Don't report an error so the execution can - * continue normally. + * 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); } - /* Continuing flushing all buffers */ - continue; } + break; } + case LTTNG_BUFFER_PER_PID: + { + struct ust_app_session *ua_sess; + struct lttng_ht_iter iter; + struct ust_app *app; - health_code_update(); + 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; + } - pthread_mutex_unlock(&ua_sess->lock); -end_no_session: rcu_read_unlock(); health_code_update(); - return 0; + return ret; } /* @@ -4003,7 +4586,9 @@ static int destroy_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); @@ -4027,6 +4612,14 @@ int ust_app_start_trace_all(struct ltt_ust_session *usess) 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) { @@ -4042,6 +4635,7 @@ int ust_app_start_trace_all(struct ltt_ust_session *usess) /* * Start tracing for the UST session. + * Called with UST session lock held. */ int ust_app_stop_trace_all(struct ltt_ust_session *usess) { @@ -4061,58 +4655,7 @@ int ust_app_stop_trace_all(struct ltt_ust_session *usess) } } - /* Flush buffers and push metadata (for UID buffers). */ - switch (usess->buffer_type) { - case LTTNG_BUFFER_PER_UID: - { - struct buffer_reg_uid *reg; - - /* 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; - - /* Get consumer socket to use to push the metadata.*/ - 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_flush_channel(socket, reg_chan->consumer_key); - } - - ust_session_reg = reg->registry->reg.ust; - if (!ust_session_reg->metadata_closed) { - /* Push metadata. */ - (void) push_metadata(ust_session_reg, usess->consumer); - } - } - - break; - } - case LTTNG_BUFFER_PER_PID: - 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 */ - continue; - } - } - break; - default: - assert(0); - break; - } + (void) ust_app_flush_session(usess); rcu_read_unlock(); @@ -4189,6 +4732,11 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto error; + } + /* * 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 @@ -4196,31 +4744,17 @@ 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; } /* @@ -4247,7 +4781,7 @@ 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; @@ -4299,6 +4833,12 @@ 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); @@ -4357,6 +4897,12 @@ 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); @@ -4366,7 +4912,7 @@ 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) { @@ -4386,69 +4932,6 @@ 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. */ @@ -4471,7 +4954,9 @@ int ust_app_calibrate_glb(struct lttng_ust_calibrate *calibrate) health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_calibrate(app->sock, calibrate); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { switch (ret) { case -ENOSYS: @@ -4983,27 +5468,19 @@ void ust_app_destroy(struct ust_app *app) * Return 0 on success or else a negative value. */ int ust_app_snapshot_record(struct ltt_ust_session *usess, - struct snapshot_output *output, int wait, unsigned int nb_streams) + 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]; - uint64_t max_stream_size = 0; assert(usess); assert(output); rcu_read_lock(); - /* - * Compute the maximum size of a single stream if a max size is asked by - * the caller. - */ - if (output->max_size > 0 && nb_streams > 0) { - max_stream_size = output->max_size / nb_streams; - } - switch (usess->buffer_type) { case LTTNG_BUFFER_PER_UID: { @@ -5033,30 +5510,16 @@ int ust_app_snapshot_record(struct ltt_ust_session *usess, /* Add the UST default trace dir to path. */ cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, reg_chan, node.node) { - - /* - * Make sure the maximum stream size is not lower than the - * subbuffer size or else it's an error since we won't be able to - * snapshot anything. - */ - if (max_stream_size && - reg_chan->subbuf_size > max_stream_size) { - ret = -EINVAL; - DBG3("UST app snapshot record maximum stream size %" PRIu64 - " is smaller than subbuffer size of %zu", - max_stream_size, reg_chan->subbuf_size); - goto error; - } - ret = consumer_snapshot_channel(socket, reg_chan->consumer_key, output, 0, - usess->uid, usess->gid, pathname, wait, - max_stream_size); + 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, - max_stream_size); + 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; } @@ -5097,23 +5560,9 @@ int ust_app_snapshot_record(struct ltt_ust_session *usess, cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, ua_chan, node.node) { - /* - * Make sure the maximum stream size is not lower than the - * subbuffer size or else it's an error since we won't be able to - * snapshot anything. - */ - if (max_stream_size && - ua_chan->attr.subbuf_size > max_stream_size) { - ret = -EINVAL; - DBG3("UST app snapshot record maximum stream size %" PRIu64 - " is smaller than subbuffer size of %" PRIu64, - max_stream_size, ua_chan->attr.subbuf_size); - goto error; - } - - ret = consumer_snapshot_channel(socket, ua_chan->key, output, 0, - ua_sess->euid, ua_sess->egid, pathname, wait, - max_stream_size); + 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; } @@ -5122,8 +5571,7 @@ int ust_app_snapshot_record(struct ltt_ust_session *usess, 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); + 1, ua_sess->euid, ua_sess->egid, pathname, wait, 0); if (ret < 0) { goto error; } @@ -5141,11 +5589,12 @@ error: } /* - * Return the number of streams for a UST session. + * Return the size taken by one more packet per stream. */ -unsigned int ust_app_get_nb_stream(struct ltt_ust_session *usess) +uint64_t ust_app_get_size_one_more_packet_per_stream(struct ltt_ust_session *usess, + uint64_t cur_nr_packets) { - unsigned int ret = 0; + uint64_t tot_size = 0; struct ust_app *app; struct lttng_ht_iter iter; @@ -5159,10 +5608,19 @@ 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; + rcu_read_lock(); cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, reg_chan, node.node) { - ret += reg_chan->stream_count; + 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; } @@ -5182,7 +5640,14 @@ unsigned int ust_app_get_nb_stream(struct ltt_ust_session *usess) cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, ua_chan, node.node) { - ret += ua_chan->streams.count; + 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(); @@ -5193,5 +5658,5 @@ unsigned int ust_app_get_nb_stream(struct ltt_ust_session *usess) break; } - return ret; + return tot_size; }