X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fust-app.c;h=a3601c42455ca3070fec1aca067884dfbd24b746;hp=7c9eb4ff352224b237d9473e3c915c21fc70175f;hb=2463b7879c00298daa79744cdaae82ac061a4ed8;hpb=b9944f51a5ffb49c0f278d7e70f566294036dea1 diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c index 7c9eb4ff3..a3601c424 100644 --- a/src/bin/lttng-sessiond/ust-app.c +++ b/src/bin/lttng-sessiond/ust-app.c @@ -1,46 +1,58 @@ /* - * Copyright (C) 2011 - David Goulet - * Copyright (C) 2016 - Jérémie Galarneau + * 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, - * as published by the Free Software Foundation. + * SPDX-License-Identifier: GPL-2.0-only * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License along - * with this program; if not, write to the Free Software Foundation, Inc., - * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ #define _LGPL_SOURCE #include +#include #include #include #include #include #include +#include #include #include #include #include -#include #include +#include #include +#include +#include +#include +#include +#include +#include +#include +#include #include #include "buffer-registry.h" +#include "condition-internal.h" #include "fd-limit.h" #include "health-sessiond.h" #include "ust-app.h" #include "ust-consumer.h" -#include "ust-ctl.h" +#include "lttng-ust-ctl.h" +#include "lttng-ust-error.h" #include "utils.h" #include "session.h" +#include "lttng-sessiond.h" +#include "notification-thread-commands.h" +#include "rotate.h" +#include "event.h" +#include "trigger-error-accounting.h" + + +struct lttng_ht *ust_app_ht; +struct lttng_ht *ust_app_ht_by_sock; +struct lttng_ht *ust_app_ht_by_notify_sock; static int ust_app_flush_app_session(struct ust_app *app, struct ust_app_session *ua_sess); @@ -90,6 +102,7 @@ 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; } /* @@ -246,7 +259,8 @@ static struct ust_registry_session *get_session_registry( case LTTNG_BUFFER_PER_UID: { struct buffer_reg_uid *reg_uid = buffer_reg_uid_find( - ua_sess->tracing_id, ua_sess->bits_per_long, ua_sess->uid); + ua_sess->tracing_id, ua_sess->bits_per_long, + lttng_credentials_get_uid(&ua_sess->real_credentials)); if (!reg_uid) { goto error; } @@ -314,6 +328,34 @@ void delete_ust_app_event(int sock, struct ust_app_event *ua_event, free(ua_event); } +/* + * Delete ust app token event_rule safely. RCU read lock must be held before calling + * this function. TODO: or does it???? + */ +static +void delete_ust_app_token_event_rule(int sock, struct ust_app_token_event_rule *ua_token, + struct ust_app *app) +{ + int ret; + + assert(ua_token); + + if (ua_token->exclusion != NULL) + free(ua_token->exclusion); + if (ua_token->obj != NULL) { + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_release_object(sock, ua_token->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); + } + free(ua_token->obj); + } + lttng_trigger_put(ua_token->trigger); + free(ua_token); +} + /* * Release ust data object of the given stream. * @@ -376,6 +418,8 @@ void delete_ust_app_channel_rcu(struct rcu_head *head) * 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) @@ -390,12 +434,22 @@ void save_per_pid_lost_discarded_counters(struct ust_app_channel *ua_chan) rcu_read_lock(); session = session_find_by_id(ua_chan->session->tracing_id); - if (!session) { - ERR("Missing LTT session to get discarded events"); - goto end; - } - if (!session->ust_session) { - ERR("Missing UST session to get discarded events"); + 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; } @@ -421,11 +475,16 @@ void save_per_pid_lost_discarded_counters(struct ust_app_channel *ua_chan) end: rcu_read_unlock(); + if (session) { + session_put(session); + } } /* * 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, @@ -468,9 +527,17 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, /* 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, + sock >= 0); + } + /* + * A negative socket can be used by the caller when + * cleaning-up a ua_chan in an error path. Skip the + * accounting in this case. + */ + if (sock >= 0) { + save_per_pid_lost_discarded_counters(ua_chan); } - save_per_pid_lost_discarded_counters(ua_chan); } if (ua_chan->obj != NULL) { @@ -537,7 +604,7 @@ ssize_t ust_app_push_metadata(struct ust_registry_session *registry, char *metadata_str = NULL; size_t len, offset, new_metadata_len_sent; ssize_t ret_val; - uint64_t metadata_key; + uint64_t metadata_key, metadata_version; assert(registry); assert(socket); @@ -552,21 +619,10 @@ ssize_t ust_app_push_metadata(struct ust_registry_session *registry, 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; - } - 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); @@ -603,7 +659,7 @@ push_data: * different bidirectionnal communication sockets. */ ret = consumer_push_metadata(socket, metadata_key, - metadata_str, len, offset); + metadata_str, len, offset, metadata_version); pthread_mutex_lock(®istry->lock); if (ret < 0) { /* @@ -722,6 +778,10 @@ error: * nullified. The session lock MUST be held unless the application is * in the destroy path. * + * Do not hold the registry lock while communicating with the consumerd, because + * doing so causes inter-process deadlocks between consumerd and sessiond with + * the metadata request notification. + * * Return 0 on success else a negative value. */ static int close_metadata(struct ust_registry_session *registry, @@ -729,6 +789,8 @@ static int close_metadata(struct ust_registry_session *registry, { int ret; struct consumer_socket *socket; + uint64_t metadata_key; + bool registry_was_already_closed; assert(registry); assert(consumer); @@ -736,8 +798,19 @@ static int close_metadata(struct ust_registry_session *registry, rcu_read_lock(); pthread_mutex_lock(®istry->lock); + metadata_key = registry->metadata_key; + registry_was_already_closed = registry->metadata_closed; + if (metadata_key != 0) { + /* + * Metadata closed. Even on error this means that the consumer + * is not responding or not found so either way a second close + * should NOT be emit for this registry. + */ + registry->metadata_closed = 1; + } + pthread_mutex_unlock(®istry->lock); - if (!registry->metadata_key || registry->metadata_closed) { + if (metadata_key == 0 || registry_was_already_closed) { ret = 0; goto end; } @@ -747,23 +820,15 @@ static int close_metadata(struct ust_registry_session *registry, consumer); if (!socket) { ret = -1; - goto error; + goto end; } - ret = consumer_close_metadata(socket, registry->metadata_key); + ret = consumer_close_metadata(socket, metadata_key); if (ret < 0) { - goto error; + goto end; } -error: - /* - * Metadata closed. Even on error this means that the consumer is not - * responding or not found so either way a second close should NOT be emit - * for this registry. - */ - registry->metadata_closed = 1; end: - pthread_mutex_unlock(®istry->lock); rcu_read_unlock(); return ret; } @@ -787,6 +852,8 @@ 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, @@ -805,6 +872,7 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, ua_sess->deleted = true; registry = get_session_registry(ua_sess); + /* 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); @@ -832,6 +900,10 @@ 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); } @@ -869,7 +941,14 @@ void delete_ust_app(struct ust_app *app) { int ret, sock; struct ust_app_session *ua_sess, *tmp_ua_sess; + struct lttng_ht_iter iter; + struct ust_app_token_event_rule *token; + /* + * 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; @@ -883,9 +962,26 @@ void delete_ust_app(struct ust_app *app) rcu_read_unlock(); } + /* Wipe token associated with the app */ + cds_lfht_for_each_entry(app->tokens_ht->ht, &iter.iter, token, + node.node) { + ret = lttng_ht_del(app->tokens_ht, &iter); + assert(!ret); + delete_ust_app_token_event_rule(app->sock, token, app); + } + ht_cleanup_push(app->sessions); ht_cleanup_push(app->ust_sessions_objd); ht_cleanup_push(app->ust_objd); + ht_cleanup_push(app->tokens_ht); + + /* This can happen if trigger setup failed. e.g killed app */ + if (app->token_communication.handle) { + ustctl_release_object(sock, app->token_communication.handle); + free(app->token_communication.handle); + } + + lttng_pipe_destroy(app->token_communication.trigger_event_pipe); /* * Wait until we have deleted the application from the sock hash table @@ -908,6 +1004,7 @@ void delete_ust_app(struct ust_app *app) DBG2("UST app pid %d deleted", app->pid); free(app); + session_unlock_list(); } /* @@ -928,6 +1025,8 @@ 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) @@ -956,7 +1055,7 @@ end: * Alloc new UST app session. */ static -struct ust_app_session *alloc_ust_app_session(struct ust_app *app) +struct ust_app_session *alloc_ust_app_session(void) { struct ust_app_session *ua_sess; @@ -982,7 +1081,7 @@ error_free: * Alloc new UST app channel. */ static -struct ust_app_channel *alloc_ust_app_channel(char *name, +struct ust_app_channel *alloc_ust_app_channel(const char *name, struct ust_app_session *ua_sess, struct lttng_ust_channel_attr *attr) { @@ -1019,6 +1118,7 @@ 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; @@ -1065,7 +1165,7 @@ struct ust_app_event *alloc_ust_app_event(char *name, /* Init most of the default value by allocating and zeroing */ ua_event = zmalloc(sizeof(struct ust_app_event)); if (ua_event == NULL) { - PERROR("malloc"); + PERROR("Failed to allocate ust_app_event structure"); goto error; } @@ -1087,6 +1187,52 @@ error: return NULL; } +/* + * Alloc new UST app token event rule. + */ +static struct ust_app_token_event_rule *alloc_ust_app_token_event_rule( + struct lttng_trigger *trigger) +{ + struct ust_app_token_event_rule *ua_token; + struct lttng_condition *condition = NULL; + struct lttng_event_rule *event_rule = NULL; + + ua_token = zmalloc(sizeof(struct ust_app_token_event_rule)); + if (ua_token == NULL) { + PERROR("Failed to allocate ust_app_token_event_rule structure"); + goto error; + } + + /* Get reference of the trigger */ + /* TODO should this be like lttng_event_rule_get with a returned bool? */ + lttng_trigger_get(trigger); + + ua_token->enabled = 1; + ua_token->token = lttng_trigger_get_tracer_token(trigger); + lttng_ht_node_init_u64(&ua_token->node, ua_token->token); + + condition = lttng_trigger_get_condition(trigger); + assert(condition); + assert(lttng_condition_get_type(condition) == LTTNG_CONDITION_TYPE_EVENT_RULE_HIT); + + assert(LTTNG_CONDITION_STATUS_OK == lttng_condition_event_rule_get_rule_mutable(condition, &event_rule)); + assert(event_rule); + + ua_token->trigger = trigger; + ua_token->filter = lttng_event_rule_get_filter_bytecode(event_rule); + ua_token->exclusion = lttng_event_rule_generate_exclusions(event_rule); + ua_token->error_counter_index = lttng_trigger_get_error_counter_index(trigger); + + /* TODO put capture here? or later*/ + + DBG3("UST app token event rule %" PRIu64 " allocated", ua_token->token); + + return ua_token; + +error: + return NULL; +} + /* * Alloc new UST app context. */ @@ -1105,7 +1251,7 @@ struct ust_app_ctx *alloc_ust_app_ctx(struct lttng_ust_context_attr *uctx) 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; + 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); @@ -1128,50 +1274,51 @@ error: } /* - * Allocate a filter and copy the given original filter. + * Create a liblttng-ust filter bytecode from given bytecode. * * Return allocated filter or NULL on error. */ -static struct lttng_filter_bytecode *copy_filter_bytecode( - struct lttng_filter_bytecode *orig_f) +static struct lttng_ust_filter_bytecode * +create_ust_filter_bytecode_from_bytecode(const struct lttng_bytecode *orig_f) { - struct lttng_filter_bytecode *filter = NULL; + struct lttng_ust_filter_bytecode *filter = NULL; /* Copy filter bytecode */ filter = zmalloc(sizeof(*filter) + orig_f->len); if (!filter) { - PERROR("zmalloc alloc filter bytecode"); + PERROR("zmalloc alloc ust filter bytecode"); goto error; } + assert(sizeof(struct lttng_bytecode) == + sizeof(struct lttng_ust_filter_bytecode)); memcpy(filter, orig_f, sizeof(*filter) + orig_f->len); - error: return filter; } /* - * Create a liblttng-ust filter bytecode from given bytecode. + * Create a liblttng-ust capture 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) +static struct lttng_ust_capture_bytecode * +create_ust_capture_bytecode_from_bytecode(const struct lttng_bytecode *orig_f) { - struct lttng_ust_filter_bytecode *filter = NULL; + struct lttng_ust_capture_bytecode *capture = NULL; - /* Copy filter bytecode */ - filter = zmalloc(sizeof(*filter) + orig_f->len); - if (!filter) { - PERROR("zmalloc alloc ust filter bytecode"); + /* Copy capture bytecode */ + capture = zmalloc(sizeof(*capture) + orig_f->len); + if (!capture) { + PERROR("zmalloc alloc ust capture bytecode"); goto error; } - assert(sizeof(struct lttng_filter_bytecode) == - sizeof(struct lttng_ust_filter_bytecode)); - memcpy(filter, orig_f, sizeof(*filter) + orig_f->len); + assert(sizeof(struct lttng_bytecode) == + sizeof(struct lttng_ust_capture_bytecode)); + memcpy(capture, orig_f, sizeof(*capture) + orig_f->len); error: - return filter; + return capture; } /* @@ -1226,7 +1373,7 @@ 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_filter_bytecode *filter, + const char *name, const struct lttng_bytecode *filter, int loglevel_value, const struct lttng_event_exclusion *exclusion) { @@ -1259,6 +1406,32 @@ end: return event; } +/* + * Lookup for an ust app tokens based on a token id. + * + * Return an ust_app_token_event_rule object or NULL on error. + */ +static struct ust_app_token_event_rule *find_ust_app_token_event_rule(struct lttng_ht *ht, + uint64_t token) +{ + struct lttng_ht_iter iter; + struct lttng_ht_node_u64 *node; + struct ust_app_token_event_rule *token_event_rule = NULL; + + assert(ht); + + lttng_ht_lookup(ht, &token, &iter); + node = lttng_ht_iter_get_node_u64(&iter); + if (node == NULL) { + DBG2("UST app token %" PRIu64 " not found", token); + goto end; + } + + token_event_rule = caa_container_of(node, struct ust_app_token_event_rule, node); +end: + return token_event_rule; +} + /* * Create the channel context on the tracer. * @@ -1287,7 +1460,7 @@ int create_ust_channel_context(struct ust_app_channel *ua_chan, * continue normally. */ ret = 0; - DBG3("UST app disable event failed. Application is dead."); + DBG3("UST app add context failed. Application is dead."); } goto error; } @@ -1305,33 +1478,79 @@ error: /* * Set the filter on the tracer. */ -static -int set_ust_event_filter(struct ust_app_event *ua_event, - struct ust_app *app) +static int set_ust_filter(struct ust_app *app, + const struct lttng_bytecode *bytecode, + struct lttng_ust_object_data *ust_object) { int ret; struct lttng_ust_filter_bytecode *ust_bytecode = NULL; health_code_update(); - if (!ua_event->filter) { - ret = 0; + ust_bytecode = create_ust_filter_bytecode_from_bytecode(bytecode); + if (!ust_bytecode) { + ret = -LTTNG_ERR_NOMEM; + goto error; + } + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_set_filter(app->sock, ust_bytecode, + ust_object); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app set filter failed for object %p of app (pid: %d) " + "with ret %d", ust_object, 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 set filter. Application is dead."); + } goto error; } - ust_bytecode = create_ust_bytecode_from_bytecode(ua_event->filter); + DBG2("UST filter set for object %p successfully", ust_object); + +error: + health_code_update(); + free(ust_bytecode); + return ret; +} + +/* + * Set a capture bytecode for the passed object. + * The seqnum enforce the ordering at runtime and on reception. + */ +static int set_ust_capture(struct ust_app *app, + const struct lttng_bytecode *bytecode, + unsigned int seqnum, + struct lttng_ust_object_data *ust_object) +{ + int ret; + struct lttng_ust_capture_bytecode *ust_bytecode = NULL; + + health_code_update(); + + ust_bytecode = create_ust_capture_bytecode_from_bytecode(bytecode); if (!ust_bytecode) { ret = -LTTNG_ERR_NOMEM; goto error; } + + /* Set the seqnum */ + ust_bytecode->seqnum = seqnum; + pthread_mutex_lock(&app->sock_lock); - ret = ustctl_set_filter(app->sock, ust_bytecode, - ua_event->obj); + ret = ustctl_set_capture(app->sock, ust_bytecode, + ust_object); 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); + ERR("UST app set capture failed for object %p of app (pid: %d) " + "with ret %d", ust_object, app->pid, ret); } else { /* * This is normal behavior, an application can die during the @@ -1339,12 +1558,12 @@ int set_ust_event_filter(struct ust_app_event *ua_event, * continue normally. */ ret = 0; - DBG3("UST app filter event failed. Application is dead."); + DBG3("UST app set capture. Application is dead."); } goto error; } - DBG2("UST filter set successfully for event %s", ua_event->name); + DBG2("UST capture set for object %p successfully", ust_object); error: health_code_update(); @@ -1376,33 +1595,30 @@ end: /* * Set event exclusions on the tracer. */ -static -int set_ust_event_exclusion(struct ust_app_event *ua_event, - struct ust_app *app) +static int set_ust_exclusions(struct ust_app *app, + struct lttng_event_exclusion *exclusions, + struct lttng_ust_object_data *ust_object) { int ret; - struct lttng_ust_event_exclusion *ust_exclusion = NULL; + struct lttng_ust_event_exclusion *ust_exclusions = NULL; - health_code_update(); + assert(exclusions && exclusions->count > 0); - if (!ua_event->exclusion || !ua_event->exclusion->count) { - ret = 0; - goto error; - } + health_code_update(); - ust_exclusion = create_ust_exclusion_from_exclusion( - ua_event->exclusion); - if (!ust_exclusion) { + ust_exclusions = create_ust_exclusion_from_exclusion( + exclusions); + if (!ust_exclusions) { ret = -LTTNG_ERR_NOMEM; goto error; } pthread_mutex_lock(&app->sock_lock); - ret = ustctl_set_exclusion(app->sock, ust_exclusion, ua_event->obj); + ret = ustctl_set_exclusion(app->sock, ust_exclusions, ust_object); 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); + ERR("UST app exclusions failed for object %p of app (pid: %d) " + "with ret %d", ust_object, app->pid, ret); } else { /* * This is normal behavior, an application can die during the @@ -1410,37 +1626,36 @@ int set_ust_event_exclusion(struct ust_app_event *ua_event, * continue normally. */ ret = 0; - DBG3("UST app event exclusion failed. Application is dead."); + DBG3("UST app set exclusions failed. Application is dead."); } goto error; } - DBG2("UST exclusion set successfully for event %s", ua_event->name); + DBG2("UST exclusions set successfully for object %p", ust_object); error: health_code_update(); - free(ust_exclusion); + free(ust_exclusions); return ret; } /* * Disable the specified event on to UST tracer for the UST session. */ -static int disable_ust_event(struct ust_app *app, - struct ust_app_session *ua_sess, struct ust_app_event *ua_event) +static int disable_ust_object(struct ust_app *app, + struct lttng_ust_object_data *object) { int ret; health_code_update(); pthread_mutex_lock(&app->sock_lock); - ret = ustctl_disable(app->sock, ua_event->obj); + ret = ustctl_disable(app->sock, object); 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); + ERR("UST app disable failed for object %p app (pid: %d) with ret %d", + object, app->pid, ret); } else { /* * This is normal behavior, an application can die during the @@ -1453,8 +1668,8 @@ static int disable_ust_event(struct ust_app *app, goto error; } - DBG2("UST app event %s disabled successfully for app (pid: %d)", - ua_event->attr.name, app->pid); + DBG2("UST app object %p disabled successfully for app (pid: %d)", + object, app->pid); error: health_code_update(); @@ -1542,21 +1757,19 @@ error: /* * Enable the specified event on to UST tracer for the UST session. */ -static int enable_ust_event(struct ust_app *app, - struct ust_app_session *ua_sess, struct ust_app_event *ua_event) +static int enable_ust_object(struct ust_app *app, struct lttng_ust_object_data *ust_object) { int ret; health_code_update(); pthread_mutex_lock(&app->sock_lock); - ret = ustctl_enable(app->sock, ua_event->obj); + ret = ustctl_enable(app->sock, ust_object); 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); + ERR("UST app enable failed for object %p app (pid: %d) with ret %d", + ust_object, app->pid, ret); } else { /* * This is normal behavior, an application can die during the @@ -1564,13 +1777,13 @@ static int enable_ust_event(struct ust_app *app, * continue normally. */ ret = 0; - DBG3("UST app enable event failed. Application is dead."); + DBG3("UST app enable failed. Application is dead."); } goto error; } - DBG2("UST app event %s enabled successfully for app (pid: %d)", - ua_event->attr.name, app->pid); + DBG2("UST app object %p enabled successfully for app (pid: %d)", + ust_object, app->pid); error: health_code_update(); @@ -1649,6 +1862,7 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + abort(); ERR("Error ustctl create event %s for app pid: %d with ret %d", ua_event->attr.name, app->pid, ret); } else { @@ -1665,14 +1879,14 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, ua_event->handle = ua_event->obj->handle; - DBG2("UST app event %s created successfully for pid:%d", - ua_event->attr.name, app->pid); + DBG2("UST app event %s created successfully for pid:%d object: %p", + ua_event->attr.name, app->pid, ua_event->obj); health_code_update(); /* Set filter if one is present. */ if (ua_event->filter) { - ret = set_ust_event_filter(ua_event, app); + ret = set_ust_filter(app, ua_event->filter, ua_event->obj); if (ret < 0) { goto error; } @@ -1680,7 +1894,7 @@ 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); + ret = set_ust_exclusions(app, ua_event->exclusion, ua_event->obj); if (ret < 0) { goto error; } @@ -1692,7 +1906,7 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, * We now need to explicitly enable the event, since it * is now disabled at creation. */ - ret = enable_ust_event(app, ua_sess, ua_event); + ret = enable_ust_object(app, ua_event->obj); if (ret < 0) { /* * If we hit an EPERM, something is wrong with our enable call. If @@ -1719,105 +1933,257 @@ error: return ret; } -/* - * Copy data between an UST app event and a LTT event. - */ -static void shadow_copy_event(struct ust_app_event *ua_event, - struct ltt_ust_event *uevent) +static +void init_ust_trigger_from_event_rule(const struct lttng_event_rule *rule, struct lttng_ust_trigger *trigger) { - size_t exclusion_alloc_size; + enum lttng_event_rule_status status; + enum lttng_loglevel_type loglevel_type; + enum lttng_ust_loglevel_type ust_loglevel_type = LTTNG_UST_LOGLEVEL_ALL; + int loglevel = -1; + const char *pattern; - strncpy(ua_event->name, uevent->attr.name, sizeof(ua_event->name)); - ua_event->name[sizeof(ua_event->name) - 1] = '\0'; + /* For now only LTTNG_EVENT_RULE_TYPE_TRACEPOINT are supported */ + assert(lttng_event_rule_get_type(rule) == LTTNG_EVENT_RULE_TYPE_TRACEPOINT); - ua_event->enabled = uevent->enabled; + memset(trigger, 0, sizeof(*trigger)); - /* Copy event attributes */ - memcpy(&ua_event->attr, &uevent->attr, sizeof(ua_event->attr)); + if (lttng_event_rule_is_agent(rule)) { + /* + * Special event for agents + * The actual meat of the event is in the filter that will be + * attached later on. + * Set the default values for the agent event. + */ + pattern = event_get_default_agent_ust_name(lttng_event_rule_get_domain_type(rule)); + loglevel = 0; + ust_loglevel_type = LTTNG_UST_LOGLEVEL_ALL; + } else { + status = lttng_event_rule_tracepoint_get_pattern(rule, &pattern); + if (status != LTTNG_EVENT_RULE_STATUS_OK) { + /* At this point this is a fatal error */ + assert(0); + } - /* Copy filter bytecode */ - if (uevent->filter) { - ua_event->filter = copy_filter_bytecode(uevent->filter); - /* Filter might be NULL here in case of ENONEM. */ - } + status = lttng_event_rule_tracepoint_get_log_level_type( + rule, &loglevel_type); + if (status != LTTNG_EVENT_RULE_STATUS_OK) { + /* At this point this is a fatal error */ + assert(0); + } - /* 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); + switch (loglevel_type) { + case LTTNG_EVENT_LOGLEVEL_ALL: + ust_loglevel_type = LTTNG_UST_LOGLEVEL_ALL; + break; + case LTTNG_EVENT_LOGLEVEL_RANGE: + ust_loglevel_type = LTTNG_UST_LOGLEVEL_RANGE; + break; + case LTTNG_EVENT_LOGLEVEL_SINGLE: + ust_loglevel_type = LTTNG_UST_LOGLEVEL_SINGLE; + break; + } + + if (loglevel_type != LTTNG_EVENT_LOGLEVEL_ALL) { + status = lttng_event_rule_tracepoint_get_log_level( + rule, &loglevel); + assert(status == LTTNG_EVENT_RULE_STATUS_OK); } } + + trigger->instrumentation = LTTNG_UST_TRACEPOINT; + strncpy(trigger->name, pattern, LTTNG_UST_SYM_NAME_LEN - 1); + trigger->loglevel_type = ust_loglevel_type; + trigger->loglevel = loglevel; } /* - * Copy data between an UST app channel and a LTT channel. + * Create the specified event rule token onto the UST tracer for a UST app. */ -static void shadow_copy_channel(struct ust_app_channel *ua_chan, - struct ltt_ust_channel *uchan) +static +int create_ust_token_event_rule(struct ust_app *app, struct ust_app_token_event_rule *ua_token) { - struct lttng_ht_iter iter; - struct ltt_ust_event *uevent; - struct ltt_ust_context *uctx; - struct ust_app_event *ua_event; - - DBG2("UST app shadow copy of channel %s started", ua_chan->name); + int ret = 0; + struct lttng_ust_trigger trigger; + struct lttng_condition *condition = NULL; + struct lttng_event_rule *event_rule = NULL; + unsigned int capture_bytecode_count = 0; - strncpy(ua_chan->name, uchan->name, sizeof(ua_chan->name)); - ua_chan->name[sizeof(ua_chan->name) - 1] = '\0'; + health_code_update(); + assert(app->token_communication.handle); - ua_chan->tracefile_size = uchan->tracefile_size; - ua_chan->tracefile_count = uchan->tracefile_count; + condition = lttng_trigger_get_condition(ua_token->trigger); + assert(condition); + assert(lttng_condition_get_type(condition) == LTTNG_CONDITION_TYPE_EVENT_RULE_HIT); - /* Copy event attributes since the layout is different. */ - ua_chan->attr.subbuf_size = uchan->attr.subbuf_size; - ua_chan->attr.num_subbuf = uchan->attr.num_subbuf; - 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->attr.output = uchan->attr.output; - /* - * Note that the attribute channel type is not set since the channel on the - * tracing registry side does not have this information. - */ + lttng_condition_event_rule_get_rule_mutable(condition, &event_rule); + assert(event_rule); + assert(lttng_event_rule_get_type(event_rule) == LTTNG_EVENT_RULE_TYPE_TRACEPOINT); + /* Should we also test for UST at this point, or do we trust all the + * upper level? */ - ua_chan->enabled = uchan->enabled; - ua_chan->tracing_channel_id = uchan->id; + init_ust_trigger_from_event_rule(event_rule, &trigger); - cds_list_for_each_entry(uctx, &uchan->ctx_list, list) { - struct ust_app_ctx *ua_ctx = alloc_ust_app_ctx(&uctx->ctx); + trigger.id = ua_token->token; + trigger.error_counter_index = ua_token->error_counter_index; - if (ua_ctx == NULL) { - continue; + /* Create UST trigger on tracer */ + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_create_trigger(app->sock, &trigger, app->token_communication.handle, &ua_token->obj); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + abort(); + ERR("Error ustctl create trigger %s for app pid: %d with ret %d", + trigger.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."); } - lttng_ht_node_init_ulong(&ua_ctx->node, - (unsigned long) ua_ctx->ctx.ctx); - lttng_ht_add_ulong(ua_chan->ctx, &ua_ctx->node); - cds_list_add_tail(&ua_ctx->list, &ua_chan->ctx_list); + goto error; } - /* 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->exclusion); - if (ua_event == NULL) { - DBG2("UST event %s not found on shadow copy channel", - uevent->attr.name); - ua_event = alloc_ust_app_event(uevent->attr.name, &uevent->attr); - if (ua_event == NULL) { - continue; - } - shadow_copy_event(ua_event, uevent); - add_unique_ust_app_event(ua_chan, ua_event); - } - } + ua_token->handle = ua_token->obj->handle; - DBG3("UST app shadow copy of channel %s done", ua_chan->name); + DBG2("UST app event %s created successfully for pid:%d object: %p", + trigger.name, app->pid, ua_token->obj); + + health_code_update(); + + /* Set filter if one is present. */ + if (ua_token->filter) { + ret = set_ust_filter(app, ua_token->filter, ua_token->obj); + if (ret < 0) { + goto error; + } + } + + /* Set exclusions for the event */ + if (ua_token->exclusion) { + ret = set_ust_exclusions(app, ua_token->exclusion, ua_token->obj); + if (ret < 0) { + goto error; + } + } + + /* Set the capture bytecode + * TODO: do we want to emulate what is done with exclusion and provide + * and object with a count of capture bytecode? instead of multiple + * call? + * */ + capture_bytecode_count = lttng_trigger_get_capture_bytecode_count(ua_token->trigger); + for (unsigned int i = 0; i < capture_bytecode_count; i++) { + const struct lttng_bytecode *capture_bytecode = lttng_trigger_get_capture_bytecode_at_index(ua_token->trigger, i); + ret = set_ust_capture(app, capture_bytecode, i, ua_token->obj); + if (ret < 0) { + goto error; + } + } + + /* + * We now need to explicitly enable the event, since it + * is disabled at creation. + */ + ret = enable_ust_object(app, ua_token->obj); + 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; + } + ua_token->enabled = true; + +error: + health_code_update(); + return ret; +} + +/* + * Copy data between an UST app event and a LTT event. + */ +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'; + + ua_event->enabled = uevent->enabled; + + /* Copy event attributes */ + memcpy(&ua_event->attr, &uevent->attr, sizeof(ua_event->attr)); + + /* Copy filter bytecode */ + if (uevent->filter) { + ua_event->filter = bytecode_copy(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); + } + } +} + +/* + * Copy data between an UST app channel and a LTT channel. + */ +static void shadow_copy_channel(struct ust_app_channel *ua_chan, + struct ltt_ust_channel *uchan) +{ + DBG2("UST app shadow copy of channel %s started", ua_chan->name); + + strncpy(ua_chan->name, uchan->name, sizeof(ua_chan->name)); + ua_chan->name[sizeof(ua_chan->name) - 1] = '\0'; + + ua_chan->tracefile_size = uchan->tracefile_size; + ua_chan->tracefile_count = uchan->tracefile_count; + + /* Copy event attributes since the layout is different. */ + ua_chan->attr.subbuf_size = uchan->attr.subbuf_size; + ua_chan->attr.num_subbuf = uchan->attr.num_subbuf; + 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. + */ + + ua_chan->enabled = uchan->enabled; + ua_chan->tracing_channel_id = uchan->id; + + DBG3("UST app shadow copy of channel %s done", ua_chan->name); } /* @@ -1826,29 +2192,22 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, static void shadow_copy_session(struct ust_app_session *ua_sess, struct ltt_ust_session *usess, struct ust_app *app) { - struct lttng_ht_node_str *ua_chan_node; - struct lttng_ht_iter iter; - struct ltt_ust_channel *uchan; - struct ust_app_channel *ua_chan; - time_t rawtime; struct tm *timeinfo; char datetime[16]; int ret; char tmp_shm_path[PATH_MAX]; - /* Get date and time for unique app path */ - time(&rawtime); - timeinfo = localtime(&rawtime); + timeinfo = localtime(&app->registration_time); strftime(datetime, sizeof(datetime), "%Y%m%d-%H%M%S", timeinfo); DBG2("Shadow copy of session handle %d", ua_sess->handle); ua_sess->tracing_id = usess->id; ua_sess->id = get_next_session_id(); - ua_sess->uid = app->uid; - ua_sess->gid = app->gid; - ua_sess->euid = usess->uid; - ua_sess->egid = usess->gid; + LTTNG_OPTIONAL_SET(&ua_sess->real_credentials.uid, app->uid); + LTTNG_OPTIONAL_SET(&ua_sess->real_credentials.gid, app->gid); + LTTNG_OPTIONAL_SET(&ua_sess->effective_credentials.uid, usess->uid); + LTTNG_OPTIONAL_SET(&ua_sess->effective_credentials.gid, usess->gid); ua_sess->buffer_type = usess->buffer_type; ua_sess->bits_per_long = app->bits_per_long; @@ -1869,7 +2228,9 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, break; case LTTNG_BUFFER_PER_UID: ret = snprintf(ua_sess->path, sizeof(ua_sess->path), - DEFAULT_UST_TRACE_UID_PATH, ua_sess->uid, app->bits_per_long); + DEFAULT_UST_TRACE_UID_PATH, + lttng_credentials_get_uid(&ua_sess->real_credentials), + app->bits_per_long); break; default: assert(0); @@ -1891,12 +2252,12 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, 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", + "/" 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, + "/" DEFAULT_UST_TRACE_UID_PATH, app->uid, app->bits_per_long); break; default: @@ -1912,37 +2273,6 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, 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) { - struct lttng_ht_iter uiter; - - lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); - ua_chan_node = lttng_ht_iter_get_node_str(&uiter); - if (ua_chan_node != NULL) { - /* Session exist. Contiuing. */ - continue; - } - - 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); - if (ua_chan == NULL) { - /* malloc failed FIXME: Might want to do handle ENOMEM .. */ - continue; - } - shadow_copy_channel(ua_chan, uchan); - /* - * The concept of metadata channel does not exist on the tracing - * registry side of the session daemon so this can only be a per CPU - * channel and not metadata. - */ - ua_chan->attr.type = LTTNG_UST_CHAN_PER_CPU; - - lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node); - } return; error: @@ -1953,7 +2283,7 @@ error: * Lookup sesison wrapper. */ static -void __lookup_session_by_app(struct ltt_ust_session *usess, +void __lookup_session_by_app(const struct ltt_ust_session *usess, struct ust_app *app, struct lttng_ht_iter *iter) { /* Get right UST app session from app */ @@ -1965,7 +2295,7 @@ void __lookup_session_by_app(struct ltt_ust_session *usess, * id. */ static struct ust_app_session *lookup_session_by_app( - struct ltt_ust_session *usess, struct ust_app *app) + const struct ltt_ust_session *usess, struct ust_app *app) { struct lttng_ht_iter iter; struct lttng_ht_node_u64 *node; @@ -2020,10 +2350,12 @@ static int setup_buffer_reg_pid(struct ust_app_session *ua_sess, app->bits_per_long, app->uint8_t_alignment, app->uint16_t_alignment, app->uint32_t_alignment, app->uint64_t_alignment, app->long_alignment, - app->byte_order, app->version.major, - app->version.minor, reg_pid->root_shm_path, - reg_pid->shm_path, - ua_sess->euid, ua_sess->egid); + app->byte_order, app->version.major, app->version.minor, + reg_pid->root_shm_path, reg_pid->shm_path, + lttng_credentials_get_uid(&ua_sess->effective_credentials), + lttng_credentials_get_gid(&ua_sess->effective_credentials), + ua_sess->tracing_id, + app->uid); if (ret < 0) { /* * reg_pid->registry->reg.ust is NULL upon error, so we need to @@ -2090,7 +2422,8 @@ static int setup_buffer_reg_uid(struct ltt_ust_session *usess, app->uint64_t_alignment, app->long_alignment, app->byte_order, app->version.major, app->version.minor, reg_uid->root_shm_path, - reg_uid->shm_path, usess->uid, usess->gid); + reg_uid->shm_path, usess->uid, usess->gid, + ua_sess->tracing_id, app->uid); if (ret < 0) { /* * reg_uid->registry->reg.ust is NULL upon error, so we need to @@ -2127,7 +2460,7 @@ 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) { @@ -2144,7 +2477,7 @@ static int create_ust_app_session(struct ltt_ust_session *usess, if (ua_sess == NULL) { DBG2("UST app pid: %d session id %" PRIu64 " not found, creating it", app->pid, usess->id); - ua_sess = alloc_ust_app_session(app); + ua_sess = alloc_ust_app_session(); if (ua_sess == NULL) { /* Only malloc can failed so something is really wrong */ ret = -ENOMEM; @@ -2320,9 +2653,8 @@ end: * 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, +int create_ust_app_channel_context(struct ust_app_channel *ua_chan, + struct lttng_ust_context_attr *uctx, struct ust_app *app) { int ret = 0; @@ -2339,7 +2671,7 @@ 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; } @@ -2367,7 +2699,7 @@ int enable_ust_app_event(struct ust_app_session *ua_sess, { int ret; - ret = enable_ust_event(app, ua_sess, ua_event); + ret = enable_ust_object(app, ua_event->obj); if (ret < 0) { goto error; } @@ -2386,7 +2718,7 @@ static int disable_ust_app_event(struct ust_app_session *ua_sess, { int ret; - ret = disable_ust_event(app, ua_sess, ua_event); + ret = disable_ust_object(app, ua_event->obj); if (ret < 0) { goto error; } @@ -2451,11 +2783,14 @@ 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, struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan, - int bitness, struct ust_registry_session *registry) + int bitness, struct ust_registry_session *registry, + uint64_t trace_archive_id) { int ret; unsigned int nb_fd = 0; @@ -2490,7 +2825,7 @@ static int do_consumer_create_channel(struct ltt_ust_session *usess, * stream we have to expect. */ ret = ust_consumer_ask_channel(ua_sess, ua_chan, usess->consumer, socket, - registry); + registry, usess->current_trace_chunk); if (ret < 0) { goto error_ask; } @@ -2796,9 +3131,6 @@ static int send_channel_uid_to_ust(struct buffer_reg_channel *reg_chan, (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; - } else if (ret < 0) { - goto error_stream_unlock; } goto error_stream_unlock; } @@ -2820,6 +3152,9 @@ 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, @@ -2829,6 +3164,9 @@ static int create_channel_per_uid(struct ust_app *app, int ret; struct buffer_reg_uid *reg_uid; struct buffer_reg_channel *reg_chan; + struct ltt_session *session = NULL; + enum lttng_error_code notification_ret; + struct ust_registry_channel *chan_reg; assert(app); assert(usess); @@ -2847,50 +3185,78 @@ 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) { - ERR("Error creating the UST channel \"%s\" registry instance", - ua_chan->name); - 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) { - ERR("Error creating UST channel \"%s\" on the consumer daemon", + /* 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; + } - /* - * 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; - } + session = session_find_by_id(ua_sess->tracing_id); + assert(session); + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); + + /* + * 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, + session->most_recent_chunk_id.value); + 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, app); - if (ret < 0) { - ERR("Error setting up UST channel \"%s\"", - ua_chan->name); - goto error; - } + 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; + } + /* Notify the notification subsystem of the channel's creation. */ + pthread_mutex_lock(®_uid->registry->reg.ust->lock); + chan_reg = ust_registry_channel_find(reg_uid->registry->reg.ust, + ua_chan->tracing_channel_id); + assert(chan_reg); + chan_reg->consumer_key = ua_chan->key; + chan_reg = NULL; + pthread_mutex_unlock(®_uid->registry->reg.ust->lock); + + notification_ret = notification_thread_command_add_channel( + notification_thread_handle, session->name, + lttng_credentials_get_uid(&ua_sess->effective_credentials), + lttng_credentials_get_gid(&ua_sess->effective_credentials), + ua_chan->name, + ua_chan->key, LTTNG_DOMAIN_UST, + ua_chan->attr.subbuf_size * ua_chan->attr.num_subbuf); + if (notification_ret != LTTNG_OK) { + ret = - (int) notification_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) { @@ -2901,12 +3267,18 @@ static int create_channel_per_uid(struct ust_app *app, } error: + if (session) { + session_put(session); + } return ret; } /* * 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, @@ -2915,6 +3287,10 @@ 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 = NULL; + uint64_t chan_reg_key; + struct ust_registry_channel *chan_reg; assert(app); assert(usess); @@ -2926,6 +3302,7 @@ 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. */ @@ -2936,13 +3313,20 @@ static int create_channel_per_pid(struct ust_app *app, goto error; } + session = session_find_by_id(ua_sess->tracing_id); + assert(session); + + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); + /* Create and get channel on the consumer side. */ ret = do_consumer_create_channel(usess, ua_sess, ua_chan, - app->bits_per_long, registry); + app->bits_per_long, registry, + session->most_recent_chunk_id.value); if (ret < 0) { ERR("Error creating UST channel \"%s\" on the consumer daemon", ua_chan->name); - goto error; + goto error_remove_from_registry; } ret = send_channel_pid_to_ust(app, ua_sess, ua_chan); @@ -2950,23 +3334,52 @@ static int create_channel_per_pid(struct ust_app *app, if (ret != -ENOTCONN) { ERR("Error sending channel to application"); } - goto error; + goto error_remove_from_registry; + } + + 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); + + cmd_ret = notification_thread_command_add_channel( + notification_thread_handle, session->name, + lttng_credentials_get_uid(&ua_sess->effective_credentials), + lttng_credentials_get_gid(&ua_sess->effective_credentials), + 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_remove_from_registry; } +error_remove_from_registry: + if (ret) { + ust_registry_channel_del_free(registry, ua_chan->key, false); + } error: rcu_read_unlock(); + if (session) { + session_put(session); + } return ret; } /* * From an already allocated ust app channel, create the channel buffers if - * need and send it to the application. This MUST be called with a RCU read + * needed and send them to the application. This MUST be called with a RCU read * side lock acquired. * + * 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, +static int ust_app_channel_send(struct ust_app *app, struct ltt_ust_session *usess, struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan) { @@ -2974,6 +3387,7 @@ static int do_create_channel(struct ust_app *app, assert(app); assert(usess); + assert(usess->active); assert(ua_sess); assert(ua_chan); @@ -3018,16 +3432,14 @@ error: } /* - * Create UST app channel and create it on the tracer. Set ua_chanp of the - * newly created channel if not NULL. + * Create UST app channel and return it through ua_chanp if not NULL. * * Called with UST app session lock and RCU read-side lock held. * - * Return 0 on success or else a negative value. Returns -ENOTCONN if - * the application exited concurrently. + * Return 0 on success or else a negative value. */ -static int create_ust_app_channel(struct ust_app_session *ua_sess, - struct ltt_ust_channel *uchan, struct ust_app *app, +static int ust_app_channel_allocate(struct ust_app_session *ua_sess, + struct ltt_ust_channel *uchan, enum lttng_ust_chan_type type, struct ltt_ust_session *usess, struct ust_app_channel **ua_chanp) { @@ -3048,24 +3460,15 @@ static int create_ust_app_channel(struct ust_app_session *ua_sess, if (ua_chan == NULL) { /* Only malloc can fail here */ ret = -ENOMEM; - goto error_alloc; + goto error; } shadow_copy_channel(ua_chan, uchan); /* Set channel type. */ ua_chan->attr.type = type; - ret = do_create_channel(app, usess, ua_sess, ua_chan); - if (ret < 0) { - goto error; - } - - DBG2("UST app create channel %s for PID %d completed", ua_chan->name, - app->pid); - /* 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; @@ -3075,8 +3478,6 @@ end: return 0; error: - delete_ust_app_channel(ua_chan->is_sent ? app->sock : -1, ua_chan, app); -error_alloc: return ret; } @@ -3093,18 +3494,9 @@ int create_ust_app_event(struct ust_app_session *ua_sess, int ret = 0; struct ust_app_event *ua_event; - /* Get event node */ - ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel, uevent->exclusion); - if (ua_event != NULL) { - ret = -EEXIST; - goto end; - } - - /* Does not exist so create one */ ua_event = alloc_ust_app_event(uevent->attr.name, &uevent->attr); if (ua_event == NULL) { - /* Only malloc can failed so something is really wrong */ + /* Only failure mode of alloc_ust_app_event(). */ ret = -ENOMEM; goto end; } @@ -3113,8 +3505,19 @@ int create_ust_app_event(struct ust_app_session *ua_sess, /* Create it on the tracer side */ ret = create_ust_event(app, ua_sess, ua_chan, ua_event); if (ret < 0) { - /* Not found previously means that it does not exist on the tracer */ - assert(ret != -LTTNG_UST_ERR_EXIST); + /* + * Not found previously means that it does not exist on the + * tracer. If the application reports that the event existed, + * it means there is a bug in the sessiond or lttng-ust + * (or corruption, etc.) + */ + if (ret == -LTTNG_UST_ERR_EXIST) { + ERR("Tracer for application reported that an event being created already existed: " + "event_name = \"%s\", pid = %d, ppid = %d, uid = %d, gid = %d", + uevent->attr.name, + app->pid, app->ppid, app->uid, + app->gid); + } goto error; } @@ -3132,6 +3535,57 @@ error: return ret; } +/* + * Create UST app event and create it on the tracer side. + * + * Called with ust app session mutex held. + */ +static +int create_ust_app_token_event_rule(struct lttng_trigger *trigger, + struct ust_app *app) +{ + int ret = 0; + struct ust_app_token_event_rule *ua_token; + + ua_token = alloc_ust_app_token_event_rule(trigger); + if (ua_token == NULL) { + ret = -ENOMEM; + goto end; + } + + /* Create it on the tracer side */ + ret = create_ust_token_event_rule(app, ua_token); + if (ret < 0) { + /* + * Not found previously means that it does not exist on the + * tracer. If the application reports that the event existed, + * it means there is a bug in the sessiond or lttng-ust + * (or corruption, etc.) + */ + if (ret == -LTTNG_UST_ERR_EXIST) { + ERR("Tracer for application reported that a token event rule being created already existed: " + "token = \"%" PRIu64 "\", pid = %d, ppid = %d, uid = %d, gid = %d", + lttng_trigger_get_tracer_token(trigger), + app->pid, app->ppid, app->uid, + app->gid); + } + goto error; + } + + lttng_ht_add_unique_u64(app->tokens_ht, &ua_token->node); + + DBG2("UST app create token event rule %" PRIu64 " for PID %d completed", lttng_trigger_get_tracer_token(trigger), + app->pid); + + goto end; + +error: + /* Valid. Calling here is already in a read side lock */ + delete_ust_app_token_event_rule(-1, ua_token, app); +end: + return ret; +} + /* * Create UST metadata and open it on the tracer side. * @@ -3144,12 +3598,14 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, struct ust_app_channel *metadata; struct consumer_socket *socket; struct ust_registry_session *registry; + struct ltt_session *session = NULL; assert(ua_sess); assert(app); 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); @@ -3192,6 +3648,12 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, */ registry->metadata_key = metadata->key; + session = session_find_by_id(ua_sess->tracing_id); + assert(session); + + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); + /* * Ask the metadata channel creation to the consumer. The metadata object * will be created by the consumer and kept their. However, the stream is @@ -3199,7 +3661,7 @@ static int create_ust_app_metadata(struct ust_app_session *ua_sess, * consumer. */ ret = ust_consumer_ask_channel(ua_sess, metadata, consumer, socket, - registry); + registry, session->current_trace_chunk); if (ret < 0) { /* Nullify the metadata key so we don't try to close it later on. */ registry->metadata_key = 0; @@ -3227,6 +3689,9 @@ error_consumer: delete_ust_app_channel(-1, metadata, app); error: pthread_mutex_unlock(®istry->lock); + if (session) { + session_put(session); + } return ret; } @@ -3265,6 +3730,7 @@ error: struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) { struct ust_app *lta = NULL; + struct lttng_pipe *trigger_event_source_pipe = NULL; assert(msg); assert(sock >= 0); @@ -3281,12 +3747,20 @@ struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) goto error; } + trigger_event_source_pipe = lttng_pipe_open(FD_CLOEXEC); + if (!trigger_event_source_pipe) { + PERROR("Open trigger pipe"); + goto error; + } + lta = zmalloc(sizeof(struct ust_app)); if (lta == NULL) { PERROR("malloc"); goto error; } + lta->token_communication.trigger_event_pipe = trigger_event_source_pipe; + lta->ppid = msg->ppid; lta->uid = msg->uid; lta->gid = msg->gid; @@ -3305,6 +3779,7 @@ struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) 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; + lta->tokens_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); /* Copy name and make sure it's NULL terminated. */ strncpy(lta->name, msg->name, sizeof(lta->name)); @@ -3336,6 +3811,8 @@ void ust_app_add(struct ust_app *app) assert(app); assert(app->notify_sock >= 0); + app->registration_time = time(NULL); + rcu_read_lock(); /* @@ -3390,13 +3867,69 @@ int ust_app_version(struct ust_app *app) } /* - * Unregister app by removing it from the global traceable app list and freeing - * the data struct. + * Setup the base trigger group. * - * The socket is already closed at this point so no close to sock. + * Return 0 on success else a negative value either an errno code or a + * LTTng-UST error code. + */ +int ust_app_setup_trigger_group(struct ust_app *app) +{ + int ret; + int writefd; + struct lttng_ust_object_data *group = NULL; + enum lttng_error_code lttng_ret; + enum trigger_error_accounting_status trigger_error_accounting_status; + + assert(app); + + /* Get the write side of the pipe */ + writefd = lttng_pipe_get_writefd(app->token_communication.trigger_event_pipe); + + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_create_trigger_group(app->sock, writefd, &group); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0) { + if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { + ERR("UST app %d create_trigger_group failed with ret %d, trigger pipe %d", app->sock, ret, writefd); + } else { + DBG("UST app %d create trigger group failed. Application is dead", app->sock); + } + goto end; + } + + lttng_ret = notification_thread_command_add_application( + notification_thread_handle, lttng_pipe_get_readfd(app->token_communication.trigger_event_pipe), LTTNG_DOMAIN_UST); + if (lttng_ret != LTTNG_OK) { + /* TODO: error */ + ret = - 1; + ERR("Failed to add channel to notification thread"); + goto end; + } + + /* Assign handle only when the complete setup is valid */ + app->token_communication.handle = group; + + trigger_error_accounting_status = trigger_error_accounting_register_app(app); + if (trigger_error_accounting_status != TRIGGER_ERROR_ACCOUNTING_STATUS_OK) { + ERR("Failed to setup trigger error accouting for app"); + ret = -1; + goto end; + } + + +end: + return ret; +} + +/* + * Unregister app by removing it from the global traceable app list and freeing + * the data struct. + * + * The socket is already closed at this point so no close to sock. */ void ust_app_unregister(int sock) { + enum lttng_error_code ret_code; struct ust_app *lta; struct lttng_ht_node_ulong *node; struct lttng_ht_iter ust_app_sock_iter; @@ -3484,8 +4017,8 @@ void ust_app_unregister(int sock) /* * 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. + * 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); @@ -3502,6 +4035,19 @@ void ust_app_unregister(int sock) lta->pid); } + /* trigger handle can be null in certain scenario such as a dead app */ + if (lta->token_communication.handle) { + int fd = lttng_pipe_get_readfd( + lta->token_communication.trigger_event_pipe); + + ret_code = notification_thread_command_remove_application( + notification_thread_handle, + fd); + if (ret_code != LTTNG_OK) { + ERR("Failed to remove application from notification thread"); + } + } + /* Free memory */ call_rcu(<a->pid_n.head, delete_ust_app_rcu); @@ -3794,6 +4340,24 @@ void ust_app_clean_list(void) rcu_read_lock(); + /* Cleanup notify socket hash table */ + 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) { + struct cds_lfht_node *node; + struct ust_app *app; + + node = cds_lfht_iter_get_node(&iter.iter); + if (!node) { + continue; + } + + app = container_of(node, struct ust_app, + notify_sock_n.node); + ust_app_notify_sock_unregister(app->notify_sock); + } + } + 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); @@ -3811,14 +4375,6 @@ void ust_app_clean_list(void) } } - /* Cleanup notify socket hash table */ - 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 */ @@ -3866,12 +4422,7 @@ int ust_app_disable_channel_glb(struct ltt_ust_session *usess, struct ust_app_session *ua_sess; struct ust_app_channel *ua_chan; - if (usess == NULL || uchan == NULL) { - ERR("Disabling UST global channel with NULL values"); - ret = -1; - goto error; - } - + assert(usess->active); DBG2("UST app disabling channel %s from global domain for session id %" PRIu64, uchan->name, usess->id); @@ -3911,8 +4462,6 @@ int ust_app_disable_channel_glb(struct ltt_ust_session *usess, } rcu_read_unlock(); - -error: return ret; } @@ -3927,12 +4476,7 @@ int ust_app_enable_channel_glb(struct ltt_ust_session *usess, struct ust_app *app; struct ust_app_session *ua_sess; - if (usess == NULL || uchan == NULL) { - ERR("Adding UST global channel to NULL values"); - ret = -1; - goto error; - } - + assert(usess->active); DBG2("UST app enabling channel %s to global domain for session id %" PRIu64, uchan->name, usess->id); @@ -3961,8 +4505,6 @@ int ust_app_enable_channel_glb(struct ltt_ust_session *usess, } rcu_read_unlock(); - -error: return ret; } @@ -3980,6 +4522,7 @@ int ust_app_disable_event_glb(struct ltt_ust_session *usess, struct ust_app_channel *ua_chan; struct ust_app_event *ua_event; + assert(usess->active); DBG("UST app disabling event %s for all apps in channel " "%s for session id %" PRIu64, uevent->attr.name, uchan->name, usess->id); @@ -4028,107 +4571,82 @@ int ust_app_disable_event_glb(struct ltt_ust_session *usess, } 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) +/* The ua_sess lock must be held by the caller. */ +static +int ust_app_channel_create(struct ltt_ust_session *usess, + struct ust_app_session *ua_sess, + struct ltt_ust_channel *uchan, struct ust_app *app, + struct ust_app_channel **_ua_chan) { - 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); + int ret = 0; + struct ust_app_channel *ua_chan = NULL; - rcu_read_lock(); + assert(ua_sess); + ASSERT_LOCKED(ua_sess->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; - } - if (!trace_ust_pid_tracker_lookup(usess, app->pid)) { - /* Skip. */ - continue; - } + if (!strncmp(uchan->name, DEFAULT_METADATA_NAME, + sizeof(uchan->name))) { + copy_channel_attr_to_ustctl(&ua_sess->metadata_attr, + &uchan->attr); + ret = 0; + } else { + struct ltt_ust_context *uctx = NULL; /* - * 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. + * Create channel onto application and synchronize its + * configuration. */ - ret = create_ust_app_session(usess, app, &ua_sess, &created); + ret = ust_app_channel_allocate(ua_sess, uchan, + LTTNG_UST_CHAN_PER_CPU, usess, + &ua_chan); if (ret < 0) { - 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; - } + goto error; } - assert(ua_sess); - pthread_mutex_lock(&ua_sess->lock); - - if (ua_sess->deleted) { - pthread_mutex_unlock(&ua_sess->lock); - continue; + ret = ust_app_channel_send(app, usess, + ua_sess, ua_chan); + if (ret) { + goto error; } - if (!strncmp(uchan->name, DEFAULT_METADATA_NAME, - sizeof(uchan->name))) { - 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, - LTTNG_UST_CHAN_PER_CPU, usess, NULL); - } - pthread_mutex_unlock(&ua_sess->lock); - if (ret < 0) { - /* 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; + /* Add contexts. */ + cds_list_for_each_entry(uctx, &uchan->ctx_list, list) { + ret = create_ust_app_channel_context(ua_chan, + &uctx->ctx, app); + if (ret) { + goto error; } } } -error_rcu_unlock: - rcu_read_unlock(); +error: + if (ret < 0) { + 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. */ + break; + case -ENOMEM: + default: + break; + } + } + + if (ret == 0 && _ua_chan) { + /* + * Only return the application's channel on success. Note + * that the channel can still be part of the application's + * channel hashtable on error. + */ + *_ua_chan = ua_chan; + } return ret; } @@ -4146,6 +4664,7 @@ int ust_app_enable_event_glb(struct ltt_ust_session *usess, struct ust_app_channel *ua_chan; struct ust_app_event *ua_event; + assert(usess->active); DBG("UST app enabling event %s for all apps for session id %" PRIu64, uevent->attr.name, usess->id); @@ -4231,6 +4750,7 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, struct ust_app_session *ua_sess; struct ust_app_channel *ua_chan; + assert(usess->active); DBG("UST app creating event %s for all apps for session id %" PRIu64, uevent->attr.name, usess->id); @@ -4280,12 +4800,14 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, } rcu_read_unlock(); - return ret; } /* * 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) @@ -4314,24 +4836,16 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) goto end; } + if (ua_sess->enabled) { + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } + /* Upon restart, we skip the setup, already done */ if (ua_sess->started) { goto skip_setup; } - /* 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); - if (ret < 0) { - if (errno != EEXIST) { - ERR("Trace directory creation error"); - goto error_unlock; - } - } - } - /* * Create the metadata for the application. This returns gracefully if a * metadata was already set for the session. @@ -4344,7 +4858,7 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) health_code_update(); skip_setup: - /* This start the UST tracing */ + /* This starts the UST tracing */ pthread_mutex_lock(&app->sock_lock); ret = ustctl_start_session(app->sock, ua_sess->handle); pthread_mutex_unlock(&app->sock_lock); @@ -4367,6 +4881,7 @@ skip_setup: /* Indicate that the session has been started once */ ua_sess->started = 1; + ua_sess->enabled = 1; pthread_mutex_unlock(&ua_sess->lock); @@ -4456,6 +4971,7 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) } health_code_update(); + ua_sess->enabled = 0; /* Quiescent wait after stopping trace */ pthread_mutex_lock(&app->sock_lock); @@ -4469,6 +4985,8 @@ 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); /* Push metadata for application before freeing the application. */ @@ -4625,98 +5143,262 @@ int ust_app_flush_session(struct ltt_ust_session *usess) return ret; } -/* - * Destroy a specific UST session in apps. - */ -static int destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) +static +int ust_app_clear_quiescent_app_session(struct ust_app *app, + struct ust_app_session *ua_sess) { - int ret; - struct ust_app_session *ua_sess; + int ret = 0; struct lttng_ht_iter iter; - struct lttng_ht_node_u64 *node; + struct ust_app_channel *ua_chan; + struct consumer_socket *socket; - DBG("Destroy tracing 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; + goto end_not_compatible; } - __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); + pthread_mutex_lock(&ua_sess->lock); - health_code_update(); - destroy_app_session(app, ua_sess); + if (ua_sess->deleted) { + goto end_unlock; + } 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); + 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; } -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(); - 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; + /* 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; } - rcu_read_unlock(); + health_code_update(); - return 0; +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); + +end_not_compatible: + rcu_read_unlock(); + health_code_update(); + return ret; } /* - * Start tracing for the UST session. + * Clear quiescent state in each stream for all applications for a + * specific UST session. * Called with UST session lock held. */ -int ust_app_stop_trace_all(struct ltt_ust_session *usess) +static +int ust_app_clear_quiescent_session(struct ltt_ust_session *usess) + { int ret = 0; - struct lttng_ht_iter iter; - struct ust_app *app; - DBG("Stopping all UST traces"); + DBG("Clearing stream quiescent state 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; - } - } + 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) { + /* + * 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 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) +{ + struct lttng_ht_iter iter; + struct ust_app *app; + + DBG("Starting all UST traces"); + + /* + * Even though the start trace might fail, flag this session active so + * other application coming in are started by default. + */ + usess->active = 1; + + 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) { + ust_app_global_update(usess, app); + } + + 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"); + + /* + * Even though the stop trace might fail, flag this session inactive so + * other application coming in are not started by default. + */ + usess->active = 0; + + 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; + } + } (void) ust_app_flush_session(usess); @@ -4751,111 +5433,284 @@ int ust_app_destroy_trace_all(struct ltt_ust_session *usess) return 0; } +/* The ua_sess lock must be held by the caller. */ static -void ust_app_global_create(struct ltt_ust_session *usess, struct ust_app *app) +int find_or_create_ust_app_channel( + struct ltt_ust_session *usess, + struct ust_app_session *ua_sess, + struct ust_app *app, + struct ltt_ust_channel *uchan, + struct ust_app_channel **ua_chan) { 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; + struct lttng_ht_iter iter; + struct lttng_ht_node_str *ua_chan_node; - ret = create_ust_app_session(usess, app, &ua_sess, &is_created); - if (ret < 0) { - /* Tracer is probably gone or ENOMEM. */ - goto error; + lttng_ht_lookup(ua_sess->channels, (void *) uchan->name, &iter); + ua_chan_node = lttng_ht_iter_get_node_str(&iter); + if (ua_chan_node) { + *ua_chan = caa_container_of(ua_chan_node, + struct ust_app_channel, node); + goto end; } - if (!is_created) { - /* App session already created. */ + + ret = ust_app_channel_create(usess, ua_sess, uchan, app, ua_chan); + if (ret) { goto end; } - assert(ua_sess); +end: + return ret; +} - pthread_mutex_lock(&ua_sess->lock); +static +int ust_app_channel_synchronize_event(struct ust_app_channel *ua_chan, + struct ltt_ust_event *uevent, struct ust_app_session *ua_sess, + struct ust_app *app) +{ + int ret = 0; + struct ust_app_event *ua_event = NULL; - if (ua_sess->deleted) { - pthread_mutex_unlock(&ua_sess->lock); - goto end; + ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, + uevent->filter, uevent->attr.loglevel, uevent->exclusion); + if (!ua_event) { + ret = create_ust_app_event(ua_sess, ua_chan, uevent, app); + if (ret < 0) { + goto end; + } + } else { + if (ua_event->enabled != uevent->enabled) { + ret = uevent->enabled ? + enable_ust_app_event(ua_sess, ua_event, app) : + disable_ust_app_event(ua_sess, ua_event, app); + } } - /* - * 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 - * ltt ust session. +end: + return ret; +} + +static +void ust_app_synchronize_tokens(struct ust_app *app) +{ + int ret = 0; + enum lttng_error_code ret_code; + enum lttng_trigger_status t_status; + struct lttng_ht_iter app_trigger_iter; + struct lttng_triggers *triggers; + struct ust_app_token_event_rule *token_event_rule_element; + unsigned int count; + + rcu_read_lock(); + /* TODO: is this necessary to protect against new trigger being added ? + * notification_trigger_tokens_ht is still the backing data structure + * for this listing. Leave it there for now. */ - cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, - node.node) { - 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; + pthread_mutex_lock(¬ification_trigger_tokens_ht_lock); + ret_code = notification_thread_command_get_tokens( + notification_thread_handle, &triggers); + if (ret_code != LTTNG_OK) { + ret = -1; + goto end; + } + + assert(triggers); + + t_status = lttng_triggers_get_count(triggers, &count); + if (t_status != LTTNG_TRIGGER_STATUS_OK) { + ret = -1; + goto end; + } + + for (unsigned int i = 0; i < count; i++) { + struct lttng_condition *condition; + struct lttng_event_rule *event_rule; + struct lttng_trigger *trigger; + struct ust_app_token_event_rule *ua_token; + uint64_t token; + + trigger = lttng_triggers_get_pointer_of_index(triggers, i); + assert(trigger); + + /* TODO: error checking and type checking */ + token = lttng_trigger_get_tracer_token(trigger); + condition = lttng_trigger_get_condition(trigger); + (void) lttng_condition_event_rule_get_rule_mutable(condition, &event_rule); + + if (lttng_event_rule_get_domain_type(event_rule) == LTTNG_DOMAIN_KERNEL) { + /* Skip kernel related trigger */ + continue; } - /* - * 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); + /* Iterate over all known token trigger */ + ua_token = find_ust_app_token_event_rule(app->tokens_ht, token); + if (!ua_token) { + ret = create_ust_app_token_event_rule(trigger, app); if (ret < 0) { - goto error_unlock; + goto end; } } + } + + /* Remove all unknown trigger from the app + * TODO find a way better way then this, do it on the unregister command + * and be specific on the token to remove instead of going over all + * trigger known to the app. This is sub optimal. + */ + cds_lfht_for_each_entry (app->tokens_ht->ht, &app_trigger_iter.iter, + token_event_rule_element, node.node) { + uint64_t token; + bool found = false; + token = token_event_rule_element->token; - /* For each events */ - cds_lfht_for_each_entry(ua_chan->events->ht, &uiter.iter, ua_event, - node.node) { - ret = create_ust_event(app, ua_sess, ua_chan, ua_event); - if (ret < 0) { - goto error_unlock; + /* + * Check if the app event trigger still exists on the + * notification side. + * TODO: might want to change the backing data struct of the + * lttng_triggers object to allow quick lookup? + * For kernel mostly all of this can be removed once we delete + * on a per trigger basis. + */ + + for (unsigned int i = 0; i < count; i++) { + struct lttng_trigger *trigger; + uint64_t inner_token; + + trigger = lttng_triggers_get_pointer_of_index( + triggers, i); + assert(trigger); + + inner_token = lttng_trigger_get_tracer_token(trigger); + + if (inner_token == token) { + found = true; + break; } } - } - pthread_mutex_unlock(&ua_sess->lock); - - if (usess->active) { - ret = ust_app_start_trace(usess, app); - if (ret < 0) { - goto error; + if (found) { + /* Still valid */ + continue; } - DBG2("UST trace started for app pid %d", app->pid); - } -end: - /* Everything went well at this point. */ - return; + /* TODO: This is fucking ugly API for fuck sake */ + assert(!lttng_ht_del(app->tokens_ht, &app_trigger_iter)); -error_unlock: - pthread_mutex_unlock(&ua_sess->lock); -error: - if (ua_sess) { - destroy_app_session(app, ua_sess); + (void) disable_ust_object(app, token_event_rule_element->obj); + + delete_ust_app_token_event_rule(app->sock, token_event_rule_element, app); } +end: + lttng_triggers_destroy(triggers); + rcu_read_unlock(); + pthread_mutex_unlock(¬ification_trigger_tokens_ht_lock); return; } +/* + * The caller must ensure that the application is compatible and is tracked + * by the process attribute trackers. + */ static -void ust_app_global_destroy(struct ltt_ust_session *usess, struct ust_app *app) +void ust_app_synchronize(struct ltt_ust_session *usess, + struct ust_app *app) { - struct ust_app_session *ua_sess; + int ret = 0; + struct cds_lfht_iter uchan_iter; + struct ltt_ust_channel *uchan; + struct ust_app_session *ua_sess = NULL; - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - return; - } - destroy_app_session(app, ua_sess); -} + /* + * The application's configuration should only be synchronized for + * active sessions. + */ + assert(usess->active); + + ret = find_or_create_ust_app_session(usess, app, &ua_sess, NULL); + if (ret < 0) { + /* Tracer is probably gone or ENOMEM. */ + goto error; + } + assert(ua_sess); + + pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } + + rcu_read_lock(); + cds_lfht_for_each_entry(usess->domain_global.channels->ht, &uchan_iter, + uchan, node.node) { + struct ust_app_channel *ua_chan; + struct cds_lfht_iter uevent_iter; + struct ltt_ust_event *uevent; + + /* + * Search for a matching ust_app_channel. If none is found, + * create it. Creating the channel will cause the ua_chan + * structure to be allocated, the channel buffers to be + * allocated (if necessary) and sent to the application, and + * all enabled contexts will be added to the channel. + */ + ret = find_or_create_ust_app_channel(usess, ua_sess, + app, uchan, &ua_chan); + if (ret) { + /* Tracer is probably gone or ENOMEM. */ + goto error_unlock; + } + + if (!ua_chan) { + /* ua_chan will be NULL for the metadata channel */ + continue; + } + + cds_lfht_for_each_entry(uchan->events->ht, &uevent_iter, uevent, + node.node) { + ret = ust_app_channel_synchronize_event(ua_chan, + uevent, ua_sess, app); + if (ret) { + goto error_unlock; + } + } + + if (ua_chan->enabled != uchan->enabled) { + ret = uchan->enabled ? + enable_ust_app_channel(ua_sess, uchan, app) : + disable_ust_app_channel(ua_sess, ua_chan, app); + if (ret) { + goto error_unlock; + } + } + } + rcu_read_unlock(); + +end: + pthread_mutex_unlock(&ua_sess->lock); + /* Everything went well at this point. */ + return; + +error_unlock: + rcu_read_unlock(); + pthread_mutex_unlock(&ua_sess->lock); +error: + if (ua_sess) { + destroy_app_session(app, ua_sess); + } + 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. @@ -4866,6 +5721,7 @@ void ust_app_global_destroy(struct ltt_ust_session *usess, struct ust_app *app) void ust_app_global_update(struct ltt_ust_session *usess, struct ust_app *app) { assert(usess); + assert(usess->active); DBG2("UST app global update for app sock %d for session id %" PRIu64, app->sock, usess->id); @@ -4873,14 +5729,40 @@ void ust_app_global_update(struct ltt_ust_session *usess, struct ust_app *app) if (!app->compatible) { return; } - - if (trace_ust_pid_tracker_lookup(usess, app->pid)) { - ust_app_global_create(usess, app); + if (trace_ust_id_tracker_lookup(LTTNG_PROCESS_ATTR_VIRTUAL_PROCESS_ID, + usess, app->pid) && + trace_ust_id_tracker_lookup( + LTTNG_PROCESS_ATTR_VIRTUAL_USER_ID, + usess, app->uid) && + trace_ust_id_tracker_lookup( + LTTNG_PROCESS_ATTR_VIRTUAL_GROUP_ID, + usess, app->gid)) { + /* + * Synchronize the application's internal tracing configuration + * and start tracing. + */ + ust_app_synchronize(usess, app); + ust_app_start_trace(usess, app); } else { ust_app_global_destroy(usess, app); } } +void ust_app_global_update_tokens(struct ust_app *app) +{ + DBG2("UST app global update token for app sock %d", app->sock); + + if (!app->compatible) { + return; + } + if (app->token_communication.handle == NULL) { + WARN("UST app global update token for app sock %d skipped since communcation handle is null", app->sock); + return; + } + + ust_app_synchronize_tokens(app); +} + /* * Called with session lock held. */ @@ -4896,6 +5778,31 @@ void ust_app_global_update_all(struct ltt_ust_session *usess) rcu_read_unlock(); } +void ust_app_global_update_all_tokens(void) +{ + 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_tokens(app); + } + rcu_read_unlock(); +} + +void ust_app_update_trigger_error_count(struct lttng_trigger *trigger) +{ + uint64_t error_count = 0; + enum trigger_error_accounting_status status; + + status = trigger_error_accounting_get_count(trigger, &error_count); + if (status != TRIGGER_ERROR_ACCOUNTING_STATUS_OK) { + ERR("Error getting trigger error count"); + } + + lttng_trigger_set_error_count(trigger, error_count); +} + /* * Add context to a specific channel for global UST domain. */ @@ -4909,8 +5816,9 @@ int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess, struct ust_app_session *ua_sess; struct ust_app *app; - rcu_read_lock(); + assert(usess->active); + rcu_read_lock(); cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { if (!app->compatible) { /* @@ -4939,7 +5847,7 @@ int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess, } ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - ret = create_ust_app_channel_context(ua_sess, ua_chan, &uctx->ctx, app); + ret = create_ust_app_channel_context(ua_chan, &uctx->ctx, app); if (ret < 0) { goto next_app; } @@ -4951,127 +5859,6 @@ int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess, return ret; } -/* - * Enable event for a channel from a UST session for a specific PID. - */ -int ust_app_enable_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; - 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 enabling 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 enable event per PID %d not found", pid); - ret = -1; - goto end; - } - - if (!app->compatible) { - ret = 0; - goto end; - } - - ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ - ret = 0; - goto end; - } - - 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); - /* 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); - - ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - 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) { - goto end_unlock; - } - } else { - ret = enable_ust_app_event(ua_sess, ua_event, app); - if (ret < 0) { - goto end_unlock; - } - } - -end_unlock: - pthread_mutex_unlock(&ua_sess->lock); -end: - 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(); - - 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: - /* 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. * @@ -5183,11 +5970,11 @@ error: * * On success 0 is returned else a negative value. */ -static int reply_ust_register_channel(int sock, int sobjd, int cobjd, +static int reply_ust_register_channel(int sock, int cobjd, size_t nr_fields, struct ustctl_field *fields) { int ret, ret_code = 0; - uint32_t chan_id, reg_count; + uint32_t chan_id; uint64_t chan_reg_key; enum ustctl_channel_header type; struct ust_app *app; @@ -5201,19 +5988,17 @@ 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; } @@ -5222,7 +6007,11 @@ 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) { @@ -5237,22 +6026,19 @@ static int reply_ust_register_channel(int sock, int sobjd, int cobjd, assert(chan_reg); if (!chan_reg->register_done) { - reg_count = ust_registry_get_event_count(chan_reg); - if (reg_count < 31) { - type = USTCTL_CHANNEL_HEADER_COMPACT; - } else { - type = USTCTL_CHANNEL_HEADER_LARGE; - } - + /* + * TODO: eventually use the registry event count for + * this channel to better guess header type for per-pid + * buffers. + */ + type = USTCTL_CHANNEL_HEADER_LARGE; 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; @@ -5288,9 +6074,7 @@ error: pthread_mutex_unlock(®istry->lock); error_rcu_unlock: rcu_read_unlock(); - if (ret) { - free(fields); - } + free(fields); return ret; } @@ -5320,23 +6104,17 @@ 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; } @@ -5344,7 +6122,11 @@ 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; @@ -5363,6 +6145,9 @@ static int add_event_ust_registry(int sock, int sobjd, int cobjd, char *name, 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 @@ -5390,6 +6175,9 @@ error: pthread_mutex_unlock(®istry->lock); error_rcu_unlock: rcu_read_unlock(); + free(sig); + free(fields); + free(model_emf_uri); return ret; } @@ -5426,13 +6214,17 @@ static int add_enum_ust_registry(int sock, int sobjd, char *name, 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. Aborting enum registration."); + 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); - assert(registry); + 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); @@ -5555,7 +6347,7 @@ int ust_app_recv_notify(int sock) * that if needed it will be freed. After this, it's invalid to access * fields or clean it up. */ - ret = reply_ust_register_channel(sock, sobjd, cobjd, nr_fields, + ret = reply_ust_register_channel(sock, cobjd, nr_fields, fields); if (ret < 0) { goto error; @@ -5640,77 +6432,738 @@ void ust_app_notify_sock_unregister(int sock) 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; - } + 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. + * + * Returns LTTNG_OK on success or a LTTNG_ERR error code. + */ +enum lttng_error_code ust_app_snapshot_record( + const struct ltt_ust_session *usess, + const struct consumer_output *output, int wait, + uint64_t nb_packets_per_stream) +{ + int ret = 0; + enum lttng_error_code status = LTTNG_OK; + struct lttng_ht_iter iter; + struct ust_app *app; + char *trace_path = NULL; + + 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; + char pathname[PATH_MAX]; + size_t consumer_path_offset = 0; + + if (!reg->registry->reg.ust->metadata_key) { + /* Skip since no metadata is present */ + continue; + } + + /* Get consumer socket to use to push the metadata.*/ + socket = consumer_find_socket_by_bitness(reg->bits_per_long, + usess->consumer); + if (!socket) { + status = LTTNG_ERR_INVALID; + 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"); + status = LTTNG_ERR_INVALID; + goto error; + } + /* Free path allowed on previous iteration. */ + free(trace_path); + trace_path = setup_channel_trace_path(usess->consumer, pathname, + &consumer_path_offset); + if (!trace_path) { + status = LTTNG_ERR_INVALID; + 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) { + status = consumer_snapshot_channel(socket, + reg_chan->consumer_key, + output, 0, usess->uid, + usess->gid, &trace_path[consumer_path_offset], wait, + nb_packets_per_stream); + if (status != LTTNG_OK) { + goto error; + } + } + status = consumer_snapshot_channel(socket, + reg->registry->reg.ust->metadata_key, output, 1, + usess->uid, usess->gid, &trace_path[consumer_path_offset], + wait, 0); + if (status != LTTNG_OK) { + 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; + char pathname[PATH_MAX]; + size_t consumer_path_offset = 0; + + 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); + if (!socket) { + status = LTTNG_ERR_INVALID; + 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) { + status = LTTNG_ERR_INVALID; + PERROR("snprintf snapshot path"); + goto error; + } + /* Free path allowed on previous iteration. */ + free(trace_path); + trace_path = setup_channel_trace_path(usess->consumer, pathname, + &consumer_path_offset); + if (!trace_path) { + status = LTTNG_ERR_INVALID; + goto error; + } + cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, + ua_chan, node.node) { + status = consumer_snapshot_channel(socket, + ua_chan->key, output, 0, + lttng_credentials_get_uid(&ua_sess->effective_credentials), + lttng_credentials_get_gid(&ua_sess->effective_credentials), + &trace_path[consumer_path_offset], wait, + nb_packets_per_stream); + switch (status) { + case LTTNG_OK: + break; + case LTTNG_ERR_CHAN_NOT_FOUND: + continue; + default: + goto error; + } + } + + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Skip application."); + continue; + } + status = consumer_snapshot_channel(socket, + registry->metadata_key, output, 1, + lttng_credentials_get_uid(&ua_sess->effective_credentials), + lttng_credentials_get_gid(&ua_sess->effective_credentials), + &trace_path[consumer_path_offset], wait, 0); + switch (status) { + case LTTNG_OK: + break; + case LTTNG_ERR_CHAN_NOT_FOUND: + continue; + default: + goto error; + } + } + break; + } + default: + assert(0); + break; + } + +error: + free(trace_path); + rcu_read_unlock(); + return status; +} + +/* + * Return the size taken by one more packet per stream. + */ +uint64_t ust_app_get_size_one_more_packet_per_stream( + const 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: + assert(0); + break; + } + + return tot_size; +} + +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 ret; + uint64_t consumer_chan_key; + + *discarded = 0; + *lost = 0; + + 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; + } + + if (overwrite) { + ret = consumer_get_lost_packets(ust_session_id, + consumer_chan_key, consumer, lost); + } else { + ret = consumer_get_discarded_events(ust_session_id, + consumer_chan_key, consumer, discarded); + } + +end: + return ret; +} + +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; + + *discarded = 0; + *lost = 0; + + rcu_read_lock(); + /* + * Iterate over every registered applications. Sum counters for + * all applications containing requested session and channel. + */ + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct lttng_ht_iter uiter; + + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + continue; + } + + /* 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; +} + +static +int ust_app_regenerate_statedump(struct ltt_ust_session *usess, + struct ust_app *app) +{ + 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; + } + + 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; +} + +/* + * Regenerate the statedump for each app in the session. + */ +int ust_app_regenerate_statedump_all(struct ltt_ust_session *usess) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + + DBG("Regenerating the metadata for all UST apps"); + + rcu_read_lock(); + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + if (!app->compatible) { + continue; + } + + ret = ust_app_regenerate_statedump(usess, app); + if (ret < 0) { + /* Continue to the next app even on error */ + continue; + } + } + + rcu_read_unlock(); + + return 0; +} + +/* + * Rotate all the channels of a session. + * + * Return LTTNG_OK on success or else an LTTng error code. + */ +enum lttng_error_code ust_app_rotate_session(struct ltt_session *session) +{ + int ret; + enum lttng_error_code cmd_ret = LTTNG_OK; + struct lttng_ht_iter iter; + struct ust_app *app; + struct ltt_ust_session *usess = session->ust_session; + + assert(usess); + + 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; + + if (!reg->registry->reg.ust->metadata_key) { + /* Skip since no metadata is present */ + continue; + } + + /* Get consumer socket to use to push the metadata.*/ + socket = consumer_find_socket_by_bitness(reg->bits_per_long, + usess->consumer); + if (!socket) { + cmd_ret = LTTNG_ERR_INVALID; + goto error; + } + + /* Rotate the data channels. */ + cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, + reg_chan, node.node) { + ret = consumer_rotate_channel(socket, + reg_chan->consumer_key, + usess->uid, usess->gid, + usess->consumer, + /* is_metadata_channel */ false); + if (ret < 0) { + cmd_ret = LTTNG_ERR_ROTATION_FAIL_CONSUMER; + 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, + /* is_metadata_channel */ true); + if (ret < 0) { + cmd_ret = LTTNG_ERR_ROTATION_FAIL_CONSUMER; + 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, + usess->consumer); + if (!socket) { + cmd_ret = LTTNG_ERR_INVALID; + goto error; + } + + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Skip application."); + continue; + } + + /* Rotate the data channels. */ + cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, + ua_chan, node.node) { + ret = consumer_rotate_channel(socket, + ua_chan->key, + lttng_credentials_get_uid(&ua_sess->effective_credentials), + lttng_credentials_get_gid(&ua_sess->effective_credentials), + ua_sess->consumer, + /* is_metadata_channel */ false); + if (ret < 0) { + /* Per-PID buffer and application going away. */ + if (ret == -LTTNG_ERR_CHAN_NOT_FOUND) + continue; + cmd_ret = LTTNG_ERR_ROTATION_FAIL_CONSUMER; + goto error; + } + } + + /* Rotate the metadata channel. */ + (void) push_metadata(registry, usess->consumer); + ret = consumer_rotate_channel(socket, + registry->metadata_key, + lttng_credentials_get_uid(&ua_sess->effective_credentials), + lttng_credentials_get_gid(&ua_sess->effective_credentials), + ua_sess->consumer, + /* is_metadata_channel */ true); + if (ret < 0) { + /* Per-PID buffer and application going away. */ + if (ret == -LTTNG_ERR_CHAN_NOT_FOUND) + continue; + cmd_ret = LTTNG_ERR_ROTATION_FAIL_CONSUMER; + goto error; + } + } + break; + } + default: + assert(0); + break; + } + + cmd_ret = LTTNG_OK; + +error: + rcu_read_unlock(); + return cmd_ret; +} + +enum lttng_error_code ust_app_create_channel_subdirectories( + const struct ltt_ust_session *usess) +{ + enum lttng_error_code ret = LTTNG_OK; + struct lttng_ht_iter iter; + enum lttng_trace_chunk_status chunk_status; + char *pathname_index; + int fmt_ret; + + assert(usess->current_trace_chunk); + 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) { + fmt_ret = asprintf(&pathname_index, + DEFAULT_UST_TRACE_DIR "/" DEFAULT_UST_TRACE_UID_PATH "/" DEFAULT_INDEX_DIR, + reg->uid, reg->bits_per_long); + if (fmt_ret < 0) { + ERR("Failed to format channel index directory"); + ret = LTTNG_ERR_CREATE_DIR_FAIL; + goto error; + } + + /* + * Create the index subdirectory which will take care + * of implicitly creating the channel's path. + */ + chunk_status = lttng_trace_chunk_create_subdirectory( + usess->current_trace_chunk, + pathname_index); + free(pathname_index); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = LTTNG_ERR_CREATE_DIR_FAIL; + goto error; + } + } + break; + } + case LTTNG_BUFFER_PER_PID: + { + struct ust_app *app; + + /* + * Create the toplevel ust/ directory in case no apps are running. + */ + chunk_status = lttng_trace_chunk_create_subdirectory( + usess->current_trace_chunk, + DEFAULT_UST_TRACE_DIR); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = LTTNG_ERR_CREATE_DIR_FAIL; + goto error; + } - iter.iter.node = &app->notify_sock_n.node; + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, + pid_n.node) { + struct ust_app_session *ua_sess; + struct ust_registry_session *registry; - /* - * 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); + ua_sess = lookup_session_by_app(usess, app); + if (!ua_sess) { + /* Session not associated with this app. */ + continue; + } -close_socket: - rcu_read_unlock(); + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Skip application."); + 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); + fmt_ret = asprintf(&pathname_index, + DEFAULT_UST_TRACE_DIR "/%s/" DEFAULT_INDEX_DIR, + ua_sess->path); + if (fmt_ret < 0) { + ERR("Failed to format channel index directory"); + ret = LTTNG_ERR_CREATE_DIR_FAIL; + goto error; + } + /* + * Create the index subdirectory which will take care + * of implicitly creating the channel's path. + */ + chunk_status = lttng_trace_chunk_create_subdirectory( + usess->current_trace_chunk, + pathname_index); + free(pathname_index); + if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { + ret = LTTNG_ERR_CREATE_DIR_FAIL; + goto error; + } + } + break; } -} - -/* - * Destroy a ust app data structure and free its memory. - */ -void ust_app_destroy(struct ust_app *app) -{ - if (!app) { - return; + default: + abort(); } - call_rcu(&app->pid_n.head, delete_ust_app_rcu); + ret = LTTNG_OK; +error: + rcu_read_unlock(); + return ret; } /* - * Take a snapshot for a given UST session. The snapshot is sent to the given - * output. + * Clear all the channels of a session. * - * Return 0 on success or else a negative value. + * Return LTTNG_OK on success or else an LTTng error code. */ -int ust_app_snapshot_record(struct ltt_ust_session *usess, - struct snapshot_output *output, int wait, - uint64_t nb_packets_per_stream) +enum lttng_error_code ust_app_clear_session(struct ltt_session *session) { - int ret = 0; - unsigned int snapshot_done = 0; + int ret; + enum lttng_error_code cmd_ret = LTTNG_OK; struct lttng_ht_iter iter; struct ust_app *app; - char pathname[PATH_MAX]; + struct ltt_ust_session *usess = session->ust_session; assert(usess); - assert(output); rcu_read_lock(); + if (usess->active) { + ERR("Expecting inactive session %s (%" PRIu64 ")", session->name, session->id); + cmd_ret = LTTNG_ERR_FATAL; + goto end; + } + switch (usess->buffer_type) { case LTTNG_BUFFER_PER_UID: { @@ -5724,36 +7177,32 @@ int ust_app_snapshot_record(struct ltt_ust_session *usess, 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; + cmd_ret = LTTNG_ERR_INVALID; + goto error_socket; } - /* Add the UST default trace dir to path. */ + /* Clear the data channels. */ 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); + ret = consumer_clear_channel(socket, + reg_chan->consumer_key); 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); + + (void) push_metadata(reg->registry->reg.ust, usess->consumer); + + /* + * Clear the metadata channel. + * Metadata channel is not cleared per se but we still need to + * perform a rotation operation on it behind the scene. + */ + ret = consumer_clear_channel(socket, + reg->registry->reg.ust->metadata_key); if (ret < 0) { goto error; } - snapshot_done = 1; } break; } @@ -5774,39 +7223,46 @@ int ust_app_snapshot_record(struct ltt_ust_session *usess, /* Get the right consumer socket for the application. */ socket = consumer_find_socket_by_bitness(app->bits_per_long, - output->consumer); + usess->consumer); if (!socket) { - ret = -EINVAL; - goto error; + cmd_ret = LTTNG_ERR_INVALID; + goto error_socket; } - /* 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; + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Skip application."); + continue; } + /* Clear the data channels. */ 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); + ret = consumer_clear_channel(socket, ua_chan->key); if (ret < 0) { + /* Per-PID buffer and application going away. */ + if (ret == -LTTNG_ERR_CHAN_NOT_FOUND) { + continue; + } 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, 0); + (void) push_metadata(registry, usess->consumer); + + /* + * Clear the metadata channel. + * Metadata channel is not cleared per se but we still need to + * perform rotation operation on it behind the scene. + */ + ret = consumer_clear_channel(socket, registry->metadata_key); if (ret < 0) { + /* Per-PID buffer and application going away. */ + if (ret == -LTTNG_ERR_CHAN_NOT_FOUND) { + continue; + } goto error; } - snapshot_done = 1; } break; } @@ -5815,63 +7271,93 @@ int ust_app_snapshot_record(struct ltt_ust_session *usess, break; } - if (!snapshot_done) { - /* - * If no snapshot was made and we are not in the error path, this means - * that there are no buffers thus no (prior) application to snapshot - * data from so we have simply NO data. - */ - ret = -ENODATA; - } + cmd_ret = LTTNG_OK; + goto end; error: + switch (-ret) { + case LTTCOMM_CONSUMERD_RELAYD_CLEAR_DISALLOWED: + cmd_ret = LTTNG_ERR_CLEAR_RELAY_DISALLOWED; + break; + default: + cmd_ret = LTTNG_ERR_CLEAR_FAIL_CONSUMER; + } + +error_socket: +end: rcu_read_unlock(); - return ret; + return cmd_ret; } /* - * Return the size taken by one more packet per stream. + * This function skips the metadata channel as the begin/end timestamps of a + * metadata packet are useless. + * + * Moreover, opening a packet after a "clear" will cause problems for live + * sessions as it will introduce padding that was not part of the first trace + * chunk. The relay daemon expects the content of the metadata stream of + * successive metadata trace chunks to be strict supersets of one another. + * + * For example, flushing a packet at the beginning of the metadata stream of + * a trace chunk resulting from a "clear" session command will cause the + * size of the metadata stream of the new trace chunk to not match the size of + * the metadata stream of the original chunk. This will confuse the relay + * daemon as the same "offset" in a metadata stream will no longer point + * to the same content. */ -uint64_t ust_app_get_size_one_more_packet_per_stream(struct ltt_ust_session *usess, - uint64_t cur_nr_packets) +enum lttng_error_code ust_app_open_packets(struct ltt_session *session) { - uint64_t tot_size = 0; - struct ust_app *app; + enum lttng_error_code ret = LTTNG_OK; struct lttng_ht_iter iter; + struct ltt_ust_session *usess = session->ust_session; assert(usess); + 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) { + cds_list_for_each_entry ( + reg, &usess->buffer_reg_uid_list, lnode) { struct buffer_reg_channel *reg_chan; + struct consumer_socket *socket; - 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; + socket = consumer_find_socket_by_bitness( + reg->bits_per_long, usess->consumer); + if (!socket) { + ret = LTTNG_ERR_FATAL; + goto error; + } + + cds_lfht_for_each_entry(reg->registry->channels->ht, + &iter.iter, reg_chan, node.node) { + const int open_ret = + consumer_open_channel_packets( + socket, + reg_chan->consumer_key); + + if (open_ret < 0) { + ret = LTTNG_ERR_UNK; + goto error; } - 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 *app; + + 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) { @@ -5879,101 +7365,49 @@ uint64_t ust_app_get_size_one_more_packet_per_stream(struct ltt_ust_session *use 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) { + /* Get the right consumer socket for the application. */ + socket = consumer_find_socket_by_bitness( + app->bits_per_long, usess->consumer); + if (!socket) { + ret = LTTNG_ERR_FATAL; + goto error; + } + + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Skip application."); + continue; + } + + cds_lfht_for_each_entry(ua_sess->channels->ht, + &chan_iter.iter, ua_chan, node.node) { + const int open_ret = + consumer_open_channel_packets( + socket, + ua_chan->key); + + if (open_ret < 0) { /* - * Don't take channel into account if we - * already grab all its packets. + * Per-PID buffer and application going + * away. */ - continue; + if (open_ret == -LTTNG_ERR_CHAN_NOT_FOUND) { + continue; + } + + ret = LTTNG_ERR_UNK; + goto error; } - tot_size += ua_chan->attr.subbuf_size * ua_chan->streams.count; } } - rcu_read_unlock(); break; } default: - assert(0); + abort(); break; } - return tot_size; -} - -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 ret; - uint64_t consumer_chan_key; - - ret = buffer_reg_uid_consumer_channel_key( - buffer_reg_uid_list, ust_session_id, - uchan_id, &consumer_chan_key); - if (ret < 0) { - goto end; - } - - if (overwrite) { - ret = consumer_get_lost_packets(ust_session_id, - consumer_chan_key, consumer, lost); - } else { - ret = consumer_get_discarded_events(ust_session_id, - consumer_chan_key, consumer, discarded); - } - -end: - return ret; -} - -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; - - rcu_read_lock(); - /* - * Iterate over every registered applications, return when we - * found one in the right session and channel. - */ - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - struct lttng_ht_iter uiter; - - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - continue; - } - - /* 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) { - ret = consumer_get_lost_packets(usess->id, ua_chan->key, - consumer, lost); - goto end; - } else { - ret = consumer_get_discarded_events(usess->id, - ua_chan->key, consumer, discarded); - goto end; - } - } - -end: +error: rcu_read_unlock(); return ret; }