X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fust-app.c;h=fc4b7085ec5770e57550c169924e112bb794dd49;hp=41a6d3988750e29a4647e5dced92aedbd516e1f6;hb=027a694fd396114812d927115f3561088e1b769f;hpb=d07ceecd2f35ccf383c7529a18dcc9b9c4b8cb17 diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c index 41a6d3988..fc4b7085e 100644 --- a/src/bin/lttng-sessiond/ust-app.c +++ b/src/bin/lttng-sessiond/ust-app.c @@ -426,46 +426,52 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, /* * Push metadata to consumer socket. * - * The socket lock MUST be acquired. - * The ust app session lock MUST be acquired. + * RCU read-side lock must be held to guarantee existance of socket. + * Must be called with the ust app session lock held. + * Must be called with the registry lock held. * * On success, return the len of metadata pushed or else a negative value. + * Returning a -EPIPE return value means we could not send the metadata, + * but it can be caused by recoverable errors (e.g. the application has + * terminated concurrently). */ ssize_t ust_app_push_metadata(struct ust_registry_session *registry, struct consumer_socket *socket, int send_zero_data) { int ret; char *metadata_str = NULL; - size_t len, offset; + size_t len, offset, new_metadata_len_sent; ssize_t ret_val; + uint64_t metadata_key; assert(registry); assert(socket); - pthread_mutex_lock(®istry->lock); + metadata_key = registry->metadata_key; /* - * Means that no metadata was assigned to the session. This can happens if - * no start has been done previously. + * Means that no metadata was assigned to the session. This can + * happens if no start has been done previously. */ - if (!registry->metadata_key) { - pthread_mutex_unlock(®istry->lock); + if (!metadata_key) { return 0; } /* - * On a push metadata error either the consumer is dead or the metadata - * channel has been destroyed because its endpoint might have died (e.g: - * relayd). 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. + * 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) { - pthread_mutex_unlock(®istry->lock); return -EPIPE; } offset = registry->metadata_len_sent; len = registry->metadata_len - registry->metadata_len_sent; + new_metadata_len_sent = registry->metadata_len; if (len == 0) { DBG3("No metadata to push for metadata key %" PRIu64, registry->metadata_key); @@ -484,38 +490,66 @@ ssize_t ust_app_push_metadata(struct ust_registry_session *registry, ret_val = -ENOMEM; goto error; } - /* Copy what we haven't send out. */ + /* Copy what we haven't sent out. */ memcpy(metadata_str, registry->metadata + offset, len); - registry->metadata_len_sent += len; push_data: pthread_mutex_unlock(®istry->lock); - ret = consumer_push_metadata(socket, registry->metadata_key, + /* + * We need to unlock the registry while we push metadata to + * break a circular dependency between the consumerd metadata + * lock and the sessiond registry lock. Indeed, pushing metadata + * to the consumerd awaits that it gets pushed all the way to + * relayd, but doing so requires grabbing the metadata lock. If + * a concurrent metadata request is being performed by + * consumerd, this can try to grab the registry lock on the + * sessiond while holding the metadata lock on the consumer + * daemon. Those push and pull schemes are performed on two + * different bidirectionnal communication sockets. + */ + ret = consumer_push_metadata(socket, metadata_key, metadata_str, len, offset); + pthread_mutex_lock(®istry->lock); if (ret < 0) { /* - * There is an acceptable race here between the registry metadata key - * assignment and the creation on the consumer. The session daemon can - * concurrently push metadata for this registry while being created on - * the consumer since the metadata key of the registry is assigned - * *before* it is setup to avoid the consumer to ask for metadata that - * could possibly be not found in the session daemon. + * There is an acceptable race here between the registry + * metadata key assignment and the creation on the + * consumer. The session daemon can concurrently push + * metadata for this registry while being created on the + * consumer since the metadata key of the registry is + * assigned *before* it is setup to avoid the consumer + * to ask for metadata that could possibly be not found + * in the session daemon. * - * The metadata will get pushed either by the session being stopped or - * the consumer requesting metadata if that race is triggered. + * The metadata will get pushed either by the session + * being stopped or the consumer requesting metadata if + * that race is triggered. */ if (ret == -LTTCOMM_CONSUMERD_CHANNEL_FAIL) { ret = 0; + } else { + ERR("Error pushing metadata to consumer"); } - - /* Update back the actual metadata len sent since it failed here. */ - pthread_mutex_lock(®istry->lock); - registry->metadata_len_sent -= len; - pthread_mutex_unlock(®istry->lock); ret_val = ret; goto error_push; + } else { + /* + * Metadata may have been concurrently pushed, since + * we're not holding the registry lock while pushing to + * consumer. This is handled by the fact that we send + * the metadata content, size, and the offset at which + * that metadata belongs. This may arrive out of order + * on the consumer side, and the consumer is able to + * deal with overlapping fragments. The consumer + * supports overlapping fragments, which must be + * contiguous starting from offset 0. We keep the + * largest metadata_len_sent value of the concurrent + * send. + */ + registry->metadata_len_sent = + max_t(size_t, registry->metadata_len_sent, + new_metadata_len_sent); } - free(metadata_str); return len; @@ -523,13 +557,14 @@ end: error: if (ret_val) { /* - * On error, flag the registry that the metadata is closed. We were unable - * to push anything and this means that either the consumer is not - * responding or the metadata cache has been destroyed on the consumer. + * On error, flag the registry that the metadata is + * closed. We were unable to push anything and this + * means that either the consumer is not responding or + * the metadata cache has been destroyed on the + * consumer. */ registry->metadata_closed = 1; } - pthread_mutex_unlock(®istry->lock); error_push: free(metadata_str); return ret_val; @@ -541,9 +576,13 @@ error_push: * socket to send the metadata is retrieved from consumer, if sock * is not NULL we use it to send the metadata. * RCU read-side lock must be held while calling this function, - * therefore ensuring existance of registry. + * therefore ensuring existance of registry. It also ensures existance + * of socket throughout this function. * * Return 0 on success else a negative error. + * Returning a -EPIPE return value means we could not send the metadata, + * but it can be caused by recoverable errors (e.g. the application has + * terminated concurrently). */ static int push_metadata(struct ust_registry_session *registry, struct consumer_output *consumer) @@ -556,50 +595,36 @@ static int push_metadata(struct ust_registry_session *registry, assert(consumer); pthread_mutex_lock(®istry->lock); - if (registry->metadata_closed) { - pthread_mutex_unlock(®istry->lock); - return -EPIPE; + ret_val = -EPIPE; + goto error; } /* Get consumer socket to use to push the metadata.*/ socket = consumer_find_socket_by_bitness(registry->bits_per_long, consumer); - pthread_mutex_unlock(®istry->lock); if (!socket) { ret_val = -1; goto error; } - /* - * TODO: Currently, we hold the socket lock around sampling of the next - * metadata segment to ensure we send metadata over the consumer socket in - * the correct order. This makes the registry lock nest inside the socket - * lock. - * - * Please note that this is a temporary measure: we should move this lock - * back into ust_consumer_push_metadata() when the consumer gets the - * ability to reorder the metadata it receives. - */ - pthread_mutex_lock(socket->lock); ret = ust_app_push_metadata(registry, socket, 0); - pthread_mutex_unlock(socket->lock); if (ret < 0) { ret_val = ret; goto error; } - + pthread_mutex_unlock(®istry->lock); return 0; error: -end: + pthread_mutex_unlock(®istry->lock); return ret_val; } /* * Send to the consumer a close metadata command for the given session. Once * done, the metadata channel is deleted and the session metadata pointer is - * nullified. The session lock MUST be acquired here unless the application is + * nullified. The session lock MUST be held unless the application is * in the destroy path. * * Return 0 on success else a negative value. @@ -681,6 +706,9 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, pthread_mutex_lock(&ua_sess->lock); + assert(!ua_sess->deleted); + ua_sess->deleted = true; + registry = get_session_registry(ua_sess); if (registry) { /* Push metadata for application before freeing the application. */ @@ -723,6 +751,8 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, } pthread_mutex_unlock(&ua_sess->lock); + consumer_output_put(ua_sess->consumer); + call_rcu(&ua_sess->rcu_head, delete_ust_app_session_rcu); } @@ -984,15 +1014,15 @@ error: * * Return allocated filter or NULL on error. */ -static struct lttng_ust_filter_bytecode *alloc_copy_ust_app_filter( - struct lttng_ust_filter_bytecode *orig_f) +static struct lttng_filter_bytecode *copy_filter_bytecode( + struct lttng_filter_bytecode *orig_f) { - struct lttng_ust_filter_bytecode *filter = NULL; + struct lttng_filter_bytecode *filter = NULL; /* Copy filter bytecode */ filter = zmalloc(sizeof(*filter) + orig_f->len); if (!filter) { - PERROR("zmalloc alloc ust app filter"); + PERROR("zmalloc alloc filter bytecode"); goto error; } @@ -1002,6 +1032,30 @@ error: return filter; } +/* + * Create a liblttng-ust filter bytecode from given bytecode. + * + * Return allocated filter or NULL on error. + */ +static struct lttng_ust_filter_bytecode *create_ust_bytecode_from_bytecode( + struct lttng_filter_bytecode *orig_f) +{ + struct lttng_ust_filter_bytecode *filter = NULL; + + /* Copy filter bytecode */ + filter = zmalloc(sizeof(*filter) + orig_f->len); + if (!filter) { + PERROR("zmalloc alloc ust filter bytecode"); + goto error; + } + + assert(sizeof(struct lttng_filter_bytecode) == + sizeof(struct lttng_ust_filter_bytecode)); + memcpy(filter, orig_f, sizeof(*filter) + orig_f->len); +error: + return filter; +} + /* * Find an ust_app using the sock and return it. RCU read side lock must be * held before calling this helper function. @@ -1054,7 +1108,7 @@ 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_ust_filter_bytecode *filter, int loglevel, + char *name, struct lttng_filter_bytecode *filter, int loglevel, const struct lttng_event_exclusion *exclusion) { struct lttng_ht_iter iter; @@ -1070,7 +1124,7 @@ static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht, key.filter = filter; key.loglevel = loglevel; /* lttng_event_exclusion and lttng_ust_event_exclusion structures are similar */ - key.exclusion = (struct lttng_ust_event_exclusion *)exclusion; + key.exclusion = exclusion; /* Lookup using the event name as hash and a custom match fct. */ cds_lfht_lookup(ht->ht, ht->hash_fct((void *) name, lttng_ht_seed), @@ -1135,6 +1189,7 @@ int set_ust_event_filter(struct ust_app_event *ua_event, struct ust_app *app) { int ret; + struct lttng_ust_filter_bytecode *ust_bytecode = NULL; health_code_update(); @@ -1143,7 +1198,12 @@ int set_ust_event_filter(struct ust_app_event *ua_event, goto error; } - ret = ustctl_set_filter(app->sock, ua_event->filter, + ust_bytecode = create_ust_bytecode_from_bytecode(ua_event->filter); + if (!ust_bytecode) { + ret = -LTTNG_ERR_NOMEM; + goto error; + } + ret = ustctl_set_filter(app->sock, ust_bytecode, ua_event->obj); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { @@ -1165,9 +1225,31 @@ int set_ust_event_filter(struct ust_app_event *ua_event, error: health_code_update(); + free(ust_bytecode); return ret; } +static +struct lttng_ust_event_exclusion *create_ust_exclusion_from_exclusion( + struct lttng_event_exclusion *exclusion) +{ + struct lttng_ust_event_exclusion *ust_exclusion = NULL; + size_t exclusion_alloc_size = sizeof(struct lttng_ust_event_exclusion) + + LTTNG_UST_SYM_NAME_LEN * exclusion->count; + + ust_exclusion = zmalloc(exclusion_alloc_size); + if (!ust_exclusion) { + PERROR("malloc"); + goto end; + } + + assert(sizeof(struct lttng_event_exclusion) == + sizeof(struct lttng_ust_event_exclusion)); + memcpy(ust_exclusion, exclusion, exclusion_alloc_size); +end: + return ust_exclusion; +} + /* * Set event exclusions on the tracer. */ @@ -1176,6 +1258,7 @@ int set_ust_event_exclusion(struct ust_app_event *ua_event, struct ust_app *app) { int ret; + struct lttng_ust_event_exclusion *ust_exclusion = NULL; health_code_update(); @@ -1184,8 +1267,13 @@ int set_ust_event_exclusion(struct ust_app_event *ua_event, goto error; } - ret = ustctl_set_exclusion(app->sock, ua_event->exclusion, - ua_event->obj); + ust_exclusion = create_ust_exclusion_from_exclusion( + ua_event->exclusion); + if (!ust_exclusion) { + ret = -LTTNG_ERR_NOMEM; + goto error; + } + ret = ustctl_set_exclusion(app->sock, ust_exclusion, ua_event->obj); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app event %s exclusions failed for app (pid: %d) " @@ -1206,6 +1294,7 @@ int set_ust_event_exclusion(struct ust_app_event *ua_event, error: health_code_update(); + free(ust_exclusion); return ret; } @@ -1377,7 +1466,10 @@ static int send_channel_pid_to_ust(struct ust_app *app, /* Send channel to the application. */ ret = ust_consumer_send_channel_to_ust(app, ua_sess, ua_chan); - if (ret < 0) { + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + goto error; + } else if (ret < 0) { goto error; } @@ -1386,7 +1478,10 @@ static int send_channel_pid_to_ust(struct ust_app *app, /* Send all streams to application. */ cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) { ret = ust_consumer_send_stream_to_ust(app, ua_chan, stream); - if (ret < 0) { + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + goto error; + } else if (ret < 0) { goto error; } /* We don't need the stream anymore once sent to the tracer. */ @@ -1507,13 +1602,13 @@ static void shadow_copy_event(struct ust_app_event *ua_event, /* Copy filter bytecode */ if (uevent->filter) { - ua_event->filter = alloc_copy_ust_app_filter(uevent->filter); + ua_event->filter = copy_filter_bytecode(uevent->filter); /* Filter might be NULL here in case of ENONEM. */ } /* Copy exclusion data */ if (uevent->exclusion) { - exclusion_alloc_size = sizeof(struct lttng_ust_event_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) { @@ -1604,6 +1699,7 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, struct tm *timeinfo; char datetime[16]; int ret; + char tmp_shm_path[PATH_MAX]; /* Get date and time for unique app path */ time(&rawtime); @@ -1620,8 +1716,11 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, ua_sess->egid = usess->gid; ua_sess->buffer_type = usess->buffer_type; ua_sess->bits_per_long = app->bits_per_long; + /* There is only one consumer object per session possible. */ + consumer_output_get(usess->consumer); ua_sess->consumer = usess->consumer; + ua_sess->output_traces = usess->output_traces; ua_sess->live_timer_interval = usess->live_timer_interval; copy_channel_attr_to_ustctl(&ua_sess->metadata_attr, @@ -1647,6 +1746,38 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, goto error; } + strncpy(ua_sess->root_shm_path, usess->root_shm_path, + sizeof(ua_sess->root_shm_path)); + ua_sess->root_shm_path[sizeof(ua_sess->root_shm_path) - 1] = '\0'; + strncpy(ua_sess->shm_path, usess->shm_path, + sizeof(ua_sess->shm_path)); + ua_sess->shm_path[sizeof(ua_sess->shm_path) - 1] = '\0'; + if (ua_sess->shm_path[0]) { + switch (ua_sess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + ret = snprintf(tmp_shm_path, sizeof(tmp_shm_path), + DEFAULT_UST_TRACE_PID_PATH "/%s-%d-%s", + app->name, app->pid, datetime); + break; + case LTTNG_BUFFER_PER_UID: + ret = snprintf(tmp_shm_path, sizeof(tmp_shm_path), + DEFAULT_UST_TRACE_UID_PATH, + app->uid, app->bits_per_long); + break; + default: + assert(0); + goto error; + } + if (ret < 0) { + PERROR("sprintf UST shadow copy session"); + assert(0); + goto error; + } + strncat(ua_sess->shm_path, tmp_shm_path, + sizeof(ua_sess->shm_path) - strlen(ua_sess->shm_path) - 1); + ua_sess->shm_path[sizeof(ua_sess->shm_path) - 1] = '\0'; + } + /* Iterate over all channels in global domain. */ cds_lfht_for_each_entry(usess->domain_global.channels->ht, &iter.iter, uchan, node.node) { @@ -1676,9 +1807,10 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node); } + return; error: - return; + consumer_output_put(ua_sess->consumer); } /* @@ -1738,7 +1870,8 @@ static int setup_buffer_reg_pid(struct ust_app_session *ua_sess, * This is the create channel path meaning that if there is NO * registry available, we have to create one for this session. */ - ret = buffer_reg_pid_create(ua_sess->id, ®_pid); + ret = buffer_reg_pid_create(ua_sess->id, ®_pid, + ua_sess->root_shm_path, ua_sess->shm_path); if (ret < 0) { goto error; } @@ -1752,7 +1885,9 @@ static int setup_buffer_reg_pid(struct ust_app_session *ua_sess, app->uint16_t_alignment, app->uint32_t_alignment, app->uint64_t_alignment, app->long_alignment, app->byte_order, app->version.major, - app->version.minor); + app->version.minor, reg_pid->root_shm_path, + reg_pid->shm_path, + ua_sess->euid, ua_sess->egid); if (ret < 0) { /* * reg_pid->registry->reg.ust is NULL upon error, so we need to @@ -1785,6 +1920,7 @@ error: * Return 0 on success or else a negative value. */ static int setup_buffer_reg_uid(struct ltt_ust_session *usess, + struct ust_app_session *ua_sess, struct ust_app *app, struct buffer_reg_uid **regp) { int ret = 0; @@ -1802,7 +1938,8 @@ static int setup_buffer_reg_uid(struct ltt_ust_session *usess, * registry available, we have to create one for this session. */ ret = buffer_reg_uid_create(usess->id, app->bits_per_long, app->uid, - LTTNG_DOMAIN_UST, ®_uid); + LTTNG_DOMAIN_UST, ®_uid, + ua_sess->root_shm_path, ua_sess->shm_path); if (ret < 0) { goto error; } @@ -1816,7 +1953,8 @@ static int setup_buffer_reg_uid(struct ltt_ust_session *usess, app->uint16_t_alignment, app->uint32_t_alignment, app->uint64_t_alignment, app->long_alignment, app->byte_order, app->version.major, - app->version.minor); + app->version.minor, reg_uid->root_shm_path, + reg_uid->shm_path, usess->uid, usess->gid); if (ret < 0) { /* * reg_uid->registry->reg.ust is NULL upon error, so we need to @@ -1891,7 +2029,7 @@ static int create_ust_app_session(struct ltt_ust_session *usess, break; case LTTNG_BUFFER_PER_UID: /* Look for a global registry. If none exists, create one. */ - ret = setup_buffer_reg_uid(usess, app, NULL); + ret = setup_buffer_reg_uid(usess, ua_sess, app, NULL); if (ret < 0) { delete_ust_app_session(-1, ua_sess, app); goto error; @@ -2479,7 +2617,10 @@ static int send_channel_uid_to_ust(struct buffer_reg_channel *reg_chan, /* Send channel to the application. */ ret = ust_consumer_send_channel_to_ust(app, ua_sess, ua_chan); - if (ret < 0) { + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + goto error; + } else if (ret < 0) { goto error; } @@ -2498,6 +2639,12 @@ static int send_channel_uid_to_ust(struct buffer_reg_channel *reg_chan, ret = ust_consumer_send_stream_to_ust(app, ua_chan, &stream); if (ret < 0) { (void) release_ust_app_stream(-1, &stream); + 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; } @@ -2591,10 +2738,9 @@ static int create_channel_per_uid(struct ust_app *app, /* Send buffers to the application. */ ret = send_channel_uid_to_ust(reg_chan, app, ua_sess, ua_chan); if (ret < 0) { - /* - * Don't report error to the console, since it may be - * caused by application concurrently exiting. - */ + if (ret != -ENOTCONN) { + ERR("Error sending channel to application"); + } goto error; } @@ -2645,10 +2791,9 @@ static int create_channel_per_pid(struct ust_app *app, ret = send_channel_pid_to_ust(app, ua_sess, ua_chan); if (ret < 0) { - /* - * Don't report error to the console, since it may be - * caused by application concurrently exiting. - */ + if (ret != -ENOTCONN) { + ERR("Error sending channel to application"); + } goto error; } @@ -2662,7 +2807,8 @@ error: * need and send it to the application. This MUST be called with a RCU read * side lock acquired. * - * Return 0 on success or else a negative value. + * Return 0 on success or else a negative value. Returns -ENOTCONN if + * the application exited concurrently. */ static int do_create_channel(struct ust_app *app, struct ltt_ust_session *usess, struct ust_app_session *ua_sess, @@ -2721,7 +2867,8 @@ error: * * Called with UST app session lock and RCU read-side lock held. * - * Return 0 on success or else a negative value. + * Return 0 on success or else a negative value. Returns -ENOTCONN if + * the application exited concurrently. */ static int create_ust_app_channel(struct ust_app_session *ua_sess, struct ltt_ust_channel *uchan, struct ust_app *app, @@ -3134,6 +3281,11 @@ void ust_app_unregister(int sock) */ pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + /* * Normally, this is done in the delete session process which is * executed in the call rcu below. However, upon registration we can't @@ -3698,6 +3850,11 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, */ continue; } + if (!trace_ust_pid_tracker_lookup(usess, app->pid)) { + /* Skip. */ + continue; + } + /* * Create session on the tracer side and add it to app session HT. Note * that if session exist, it will simply return a pointer to the ust @@ -3712,6 +3869,7 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, * or a timeout on it. We can't inform the caller that for a * specific app, the session failed so lets continue here. */ + ret = 0; /* Not an error. */ continue; case -ENOMEM: default: @@ -3721,6 +3879,12 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, assert(ua_sess); pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + if (!strncmp(uchan->name, DEFAULT_METADATA_NAME, sizeof(uchan->name))) { copy_channel_attr_to_ustctl(&ua_sess->metadata_attr, &uchan->attr); @@ -3732,14 +3896,23 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, } pthread_mutex_unlock(&ua_sess->lock); if (ret < 0) { - if (ret == -ENOMEM) { - /* No more memory is a fatal error. Stop right now. */ - goto error_rcu_unlock; - } /* Cleanup the created session if it's the case. */ if (created) { destroy_app_session(app, ua_sess); } + switch (ret) { + case -ENOTCONN: + /* + * The application's socket is not valid. Either a bad socket + * or a timeout on it. We can't inform the caller that for a + * specific app, the session failed so lets continue here. + */ + ret = 0; /* Not an error. */ + continue; + case -ENOMEM: + default: + goto error_rcu_unlock; + } } } @@ -3790,11 +3963,23 @@ int ust_app_enable_event_glb(struct ltt_ust_session *usess, pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + /* Lookup channel in the ust app session */ lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); ua_chan_node = lttng_ht_iter_get_node_str(&uiter); - /* If the channel is not found, there is a code flow error */ - assert(ua_chan_node); + /* + * It is possible that the channel cannot be found is + * the channel/event creation occurs concurrently with + * an application exit. + */ + if (!ua_chan_node) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); @@ -3856,6 +4041,12 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, } pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + /* Lookup channel in the ust app session */ lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); ua_chan_node = lttng_ht_iter_get_node_str(&uiter); @@ -3907,6 +4098,11 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } + /* Upon restart, we skip the setup, already done */ if (ua_sess->started) { goto skip_setup; @@ -3918,7 +4114,7 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) ret = run_as_mkdir_recursive(usess->consumer->dst.trace_path, S_IRWXU | S_IRWXG, ua_sess->euid, ua_sess->egid); if (ret < 0) { - if (ret != -EEXIST) { + if (errno != EEXIST) { ERR("Trace directory creation error"); goto error_unlock; } @@ -4007,6 +4203,11 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end_no_session; + } + /* * If started = 0, it means that stop trace has been called for a session * that was never started. It's possible since we can have a fail start @@ -4087,6 +4288,10 @@ int ust_app_flush_app_session(struct ust_app *app, pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + goto end_deleted; + } + health_code_update(); /* Flushing buffers */ @@ -4116,6 +4321,7 @@ int ust_app_flush_app_session(struct ust_app *app, health_code_update(); +end_deleted: pthread_mutex_unlock(&ua_sess->lock); end_not_compatible: @@ -4132,7 +4338,7 @@ static int ust_app_flush_session(struct ltt_ust_session *usess) { - int ret; + int ret = 0; DBG("Flushing session buffers for all ust apps"); @@ -4173,7 +4379,6 @@ int ust_app_flush_session(struct ltt_ust_session *usess) /* Push metadata. */ (void) push_metadata(ust_session_reg, usess->consumer); } - ret = 0; break; } case LTTNG_BUFFER_PER_PID: @@ -4192,11 +4397,11 @@ int ust_app_flush_session(struct ltt_ust_session *usess) break; } default: + ret = -1; assert(0); break; } -end_no_session: rcu_read_unlock(); health_code_update(); return ret; @@ -4326,50 +4531,35 @@ int ust_app_destroy_trace_all(struct ltt_ust_session *usess) return 0; } -/* - * Add channels/events from UST global domain to registered apps at sock. - */ -void ust_app_global_update(struct ltt_ust_session *usess, int sock) +static +void ust_app_global_create(struct ltt_ust_session *usess, struct ust_app *app) { int ret = 0; struct lttng_ht_iter iter, uiter; - struct ust_app *app; struct ust_app_session *ua_sess = NULL; struct ust_app_channel *ua_chan; struct ust_app_event *ua_event; struct ust_app_ctx *ua_ctx; + int is_created = 0; - assert(usess); - assert(sock >= 0); - - DBG2("UST app global update for app sock %d for session id %" PRIu64, sock, - usess->id); - - rcu_read_lock(); - - app = ust_app_find_by_sock(sock); - if (app == NULL) { - /* - * Application can be unregistered before so this is possible hence - * simply stopping the update. - */ - DBG3("UST app update failed to find app sock %d", sock); - goto error; - } - - if (!app->compatible) { - goto error; - } - - ret = create_ust_app_session(usess, app, &ua_sess, NULL); + ret = create_ust_app_session(usess, app, &ua_sess, &is_created); if (ret < 0) { /* Tracer is probably gone or ENOMEM. */ goto error; } + if (!is_created) { + /* App session already created. */ + goto end; + } assert(ua_sess); pthread_mutex_lock(&ua_sess->lock); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } + /* * We can iterate safely here over all UST app session since the create ust * app session above made a shadow copy of the UST global domain from the @@ -4378,11 +4568,14 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, node.node) { ret = do_create_channel(app, usess, ua_sess, ua_chan); - if (ret < 0) { + 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. + * 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; } @@ -4419,9 +4612,8 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) DBG2("UST trace started for app pid %d", app->pid); } - +end: /* Everything went well at this point. */ - rcu_read_unlock(); return; error_unlock: @@ -4430,10 +4622,60 @@ error: if (ua_sess) { destroy_app_session(app, ua_sess); } - rcu_read_unlock(); return; } +static +void ust_app_global_destroy(struct ltt_ust_session *usess, struct ust_app *app) +{ + struct ust_app_session *ua_sess; + + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + return; + } + destroy_app_session(app, ua_sess); +} + +/* + * Add channels/events from UST global domain to registered apps at sock. + * + * Called with session lock held. + * Called with RCU read-side lock held. + */ +void ust_app_global_update(struct ltt_ust_session *usess, struct ust_app *app) +{ + assert(usess); + + DBG2("UST app global update for app sock %d for session id %" PRIu64, + app->sock, usess->id); + + if (!app->compatible) { + return; + } + + if (trace_ust_pid_tracker_lookup(usess, app->pid)) { + ust_app_global_create(usess, app); + } else { + ust_app_global_destroy(usess, app); + } +} + +/* + * Called with session lock held. + */ +void ust_app_global_update_all(struct ltt_ust_session *usess) +{ + struct lttng_ht_iter iter; + struct ust_app *app; + + rcu_read_lock(); + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ust_app_global_update(usess, app); + } + rcu_read_unlock(); +} + /* * Add context to a specific channel for global UST domain. */ @@ -4463,6 +4705,12 @@ int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess, } pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + /* Lookup channel in the ust app session */ lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); ua_chan_node = lttng_ht_iter_get_node_str(&uiter); @@ -4521,6 +4769,12 @@ int ust_app_enable_event_pid(struct ltt_ust_session *usess, } pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + ret = 0; + goto end_unlock; + } + /* Lookup channel in the ust app session */ lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter); ua_chan_node = lttng_ht_iter_get_node_str(&iter);