X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fust-app.c;h=225d3f86a81a3837b4acbf9ed23c5d3cf01a9416;hp=98ea50fdb250a6f0f89b2c73240eb7dea4df05d9;hb=5d2e1e66a968d9e555f9b8b00d0589ebfaf3de32;hpb=5c3c99d75ba81b265ffc4d41a1f27cf926b7157a diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c index 98ea50fdb..225d3f86a 100644 --- a/src/bin/lttng-sessiond/ust-app.c +++ b/src/bin/lttng-sessiond/ust-app.c @@ -36,6 +36,52 @@ #include "ust-consumer.h" #include "ust-ctl.h" +/* Next available channel key. */ +static unsigned long next_channel_key; + +/* + * Return the atomically incremented value of next_channel_key. + */ +static inline unsigned long get_next_channel_key(void) +{ + return uatomic_add_return(&next_channel_key, 1); +} + +/* + * Return the consumer socket from the given consumer output with the right + * bitness. On error, returns NULL. + * + * The caller MUST acquire a rcu read side lock and keep it until the socket + * object reference is not needed anymore. + */ +static struct consumer_socket *find_consumer_socket_by_bitness(int bits, + struct consumer_output *consumer) +{ + int consumer_fd; + struct consumer_socket *socket = NULL; + + switch (bits) { + case 64: + consumer_fd = uatomic_read(&ust_consumerd64_fd); + break; + case 32: + consumer_fd = uatomic_read(&ust_consumerd32_fd); + break; + default: + assert(0); + goto end; + } + + socket = consumer_find_socket(consumer_fd, consumer); + if (!socket) { + ERR("Consumer socket fd %d not found in consumer obj %p", + consumer_fd, consumer); + } + +end: + return socket; +} + /* * Match function for the hash table lookup. * @@ -100,16 +146,18 @@ no_match: * Unique add of an ust app event in the given ht. This uses the custom * ht_match_ust_app_event match function and the event name as hash. */ -static void add_unique_ust_app_event(struct lttng_ht *ht, +static void add_unique_ust_app_event(struct ust_app_channel *ua_chan, struct ust_app_event *event) { struct cds_lfht_node *node_ptr; struct ust_app_ht_key key; + struct lttng_ht *ht; - assert(ht); - assert(ht->ht); + assert(ua_chan); + assert(ua_chan->events); assert(event); + ht = ua_chan->events; key.name = event->attr.name; key.filter = event->filter; key.loglevel = event->attr.loglevel; @@ -120,6 +168,28 @@ static void add_unique_ust_app_event(struct lttng_ht *ht, assert(node_ptr == &event->node.node); } +/* + * Close the notify socket from the given RCU head object. This MUST be called + * through a call_rcu(). + */ +static void close_notify_sock_rcu(struct rcu_head *head) +{ + int ret; + struct ust_app_notify_sock_obj *obj = + caa_container_of(head, struct ust_app_notify_sock_obj, head); + + /* Must have a valid fd here. */ + assert(obj->fd >= 0); + + ret = close(obj->fd); + if (ret) { + ERR("close notify sock %d RCU", obj->fd); + } + lttng_fd_put(LTTNG_FD_APPS, 1); + + free(obj); +} + /* * Delete ust context safely. RCU read lock must be held before calling * this function. @@ -127,8 +197,16 @@ static void add_unique_ust_app_event(struct lttng_ht *ht, static void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx) { + int ret; + + assert(ua_ctx); + if (ua_ctx->obj) { - ustctl_release_object(sock, ua_ctx->obj); + ret = ustctl_release_object(sock, ua_ctx->obj); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app sock %d release ctx obj handle %d failed with ret %d", + sock, ua_ctx->obj->handle, ret); + } free(ua_ctx->obj); } free(ua_ctx); @@ -141,10 +219,18 @@ void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx) static void delete_ust_app_event(int sock, struct ust_app_event *ua_event) { + int ret; + + assert(ua_event); + free(ua_event->filter); if (ua_event->obj != NULL) { - ustctl_release_object(sock, ua_event->obj); + ret = ustctl_release_object(sock, ua_event->obj); + 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_event->obj); } free(ua_event); @@ -155,10 +241,18 @@ void delete_ust_app_event(int sock, struct ust_app_event *ua_event) * this function. */ static -void delete_ust_app_stream(int sock, struct ltt_ust_stream *stream) +void delete_ust_app_stream(int sock, struct ust_app_stream *stream) { + int ret; + + assert(stream); + if (stream->obj) { - ustctl_release_object(sock, stream->obj); + ret = ustctl_release_object(sock, stream->obj); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app sock %d release stream obj failed with ret %d", + sock, ret); + } lttng_fd_put(LTTNG_FD_APPS, 2); free(stream->obj); } @@ -170,13 +264,18 @@ void delete_ust_app_stream(int sock, struct ltt_ust_stream *stream) * this function. */ static -void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan) +void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, + struct ust_app *app) { int ret; struct lttng_ht_iter iter; struct ust_app_event *ua_event; struct ust_app_ctx *ua_ctx; - struct ltt_ust_stream *stream, *stmp; + struct ust_app_stream *stream, *stmp; + + assert(ua_chan); + + DBG3("UST app deleting channel %s", ua_chan->name); /* Wipe stream */ cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) { @@ -201,49 +300,202 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan) } lttng_ht_destroy(ua_chan->events); + /* Wipe and free registry. */ + ust_registry_channel_destroy(&ua_chan->session->registry, &ua_chan->registry); + if (ua_chan->obj != NULL) { - ustctl_release_object(sock, ua_chan->obj); + /* Remove channel from application UST object descriptor. */ + iter.iter.node = &ua_chan->ust_objd_node.node; + lttng_ht_del(app->ust_objd, &iter); + ret = ustctl_release_object(sock, ua_chan->obj); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app sock %d release channel obj failed with ret %d", + sock, ret); + } lttng_fd_put(LTTNG_FD_APPS, 2); free(ua_chan->obj); } free(ua_chan); } +/* + * For a given application and session, push metadata to consumer. The session + * lock MUST be acquired here before calling this. + * + * Return 0 on success else a negative error. + */ +static int push_metadata(struct ust_app *app, struct ust_app_session *ua_sess) +{ + int ret; + char *metadata_str = NULL; + size_t len, offset; + struct consumer_socket *socket; + + assert(app); + assert(ua_sess); + + if (!ua_sess->consumer || !ua_sess->metadata) { + /* No consumer means no stream associated so just return gracefully. */ + ret = 0; + goto end; + } + + rcu_read_lock(); + + /* Get consumer socket to use to push the metadata.*/ + socket = find_consumer_socket_by_bitness(app->bits_per_long, + ua_sess->consumer); + if (!socket) { + ret = -1; + goto error_rcu_unlock; + } + + /* + * 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); + pthread_mutex_lock(&ua_sess->registry.lock); + + offset = ua_sess->registry.metadata_len_sent; + len = ua_sess->registry.metadata_len - ua_sess->registry.metadata_len_sent; + if (len == 0) { + DBG3("No metadata to push for session id %d", ua_sess->id); + ret = 0; + goto error_reg_unlock; + } + assert(len > 0); + + /* Allocate only what we have to send. */ + metadata_str = zmalloc(len); + if (!metadata_str) { + PERROR("zmalloc ust app metadata string"); + ret = -ENOMEM; + goto error_reg_unlock; + } + /* Copy what we haven't send out. */ + memcpy(metadata_str, ua_sess->registry.metadata + offset, len); + + pthread_mutex_unlock(&ua_sess->registry.lock); + + ret = ust_consumer_push_metadata(socket, ua_sess, metadata_str, len, + offset); + if (ret < 0) { + pthread_mutex_unlock(socket->lock); + goto error_rcu_unlock; + } + + /* Update len sent of the registry. */ + pthread_mutex_lock(&ua_sess->registry.lock); + ua_sess->registry.metadata_len_sent += len; + pthread_mutex_unlock(&ua_sess->registry.lock); + pthread_mutex_unlock(socket->lock); + + rcu_read_unlock(); + free(metadata_str); + return 0; + +error_reg_unlock: + pthread_mutex_unlock(&ua_sess->registry.lock); + pthread_mutex_unlock(socket->lock); +error_rcu_unlock: + rcu_read_unlock(); + free(metadata_str); +end: + return ret; +} + +/* + * 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 + * in the destroy path. + * + * Return 0 on success else a negative value. + */ +static int close_metadata(struct ust_app *app, struct ust_app_session *ua_sess) +{ + int ret; + struct consumer_socket *socket; + + assert(app); + assert(ua_sess); + + /* Ignore if no metadata. Valid since it can be called on unregister. */ + if (!ua_sess->metadata) { + ret = 0; + goto error; + } + + rcu_read_lock(); + + /* Get consumer socket to use to push the metadata.*/ + socket = find_consumer_socket_by_bitness(app->bits_per_long, + ua_sess->consumer); + if (!socket) { + ret = -1; + goto error_rcu_unlock; + } + + ret = ust_consumer_close_metadata(socket, ua_sess->metadata); + if (ret < 0) { + goto error_rcu_unlock; + } + +error_rcu_unlock: + /* Destroy metadata on our side since we must not use it anymore. */ + delete_ust_app_channel(-1, ua_sess->metadata, app); + ua_sess->metadata = NULL; + + rcu_read_unlock(); +error: + return ret; +} + /* * Delete ust app session safely. RCU read lock must be held before calling * this function. */ static -void delete_ust_app_session(int sock, struct ust_app_session *ua_sess) +void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, + struct ust_app *app) { int ret; struct lttng_ht_iter iter; struct ust_app_channel *ua_chan; + assert(ua_sess); + if (ua_sess->metadata) { - if (ua_sess->metadata->stream_obj) { - ustctl_release_object(sock, ua_sess->metadata->stream_obj); - lttng_fd_put(LTTNG_FD_APPS, 2); - free(ua_sess->metadata->stream_obj); - } - if (ua_sess->metadata->obj) { - ustctl_release_object(sock, ua_sess->metadata->obj); - lttng_fd_put(LTTNG_FD_APPS, 2); - free(ua_sess->metadata->obj); - } - trace_ust_destroy_metadata(ua_sess->metadata); + /* Push metadata for application before freeing the application. */ + (void) push_metadata(app, ua_sess); + + /* And ask to close it for this session. */ + (void) close_metadata(app, ua_sess); } cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, node.node) { ret = lttng_ht_del(ua_sess->channels, &iter); assert(!ret); - delete_ust_app_channel(sock, ua_chan); + delete_ust_app_channel(sock, ua_chan, app); } lttng_ht_destroy(ua_sess->channels); + ust_registry_session_destroy(&ua_sess->registry); + if (ua_sess->handle != -1) { - ustctl_release_handle(sock, ua_sess->handle); + ret = ustctl_release_handle(sock, ua_sess->handle); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app sock %d release session handle failed with ret %d", + sock, ret); + } } free(ua_sess); } @@ -270,7 +522,7 @@ void delete_ust_app(struct ust_app *app) cds_list_for_each_entry_safe(ua_sess, tmp_ua_sess, &app->teardown_head, teardown_node) { /* Free every object in the session and the session. */ - delete_ust_app_session(sock, ua_sess); + delete_ust_app_session(sock, ua_sess, app); } /* @@ -313,11 +565,38 @@ void delete_ust_app_rcu(struct rcu_head *head) delete_ust_app(app); } +/* + * Delete the session from the application ht and delete the data structure by + * freeing every object inside and releasing them. + */ +static void destroy_app_session(struct ust_app *app, + struct ust_app_session *ua_sess) +{ + int ret; + struct lttng_ht_iter iter; + + assert(app); + assert(ua_sess); + + iter.iter.node = &ua_sess->node.node; + ret = lttng_ht_del(app->sessions, &iter); + if (ret) { + /* Already scheduled for teardown. */ + goto end; + } + + /* Once deleted, free the data structure. */ + delete_ust_app_session(app->sock, ua_sess, app); + +end: + return; +} + /* * Alloc new UST app session. */ static -struct ust_app_session *alloc_ust_app_session(void) +struct ust_app_session *alloc_ust_app_session(struct ust_app *app) { struct ust_app_session *ua_sess; @@ -325,15 +604,28 @@ struct ust_app_session *alloc_ust_app_session(void) ua_sess = zmalloc(sizeof(struct ust_app_session)); if (ua_sess == NULL) { PERROR("malloc"); - goto error; + goto error_free; } ua_sess->handle = -1; ua_sess->channels = lttng_ht_new(0, LTTNG_HT_TYPE_STRING); + pthread_mutex_init(&ua_sess->lock, NULL); + if (ust_registry_session_init(&ua_sess->registry, app, + 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)) { + goto error; + } return ua_sess; error: + free(ua_sess); +error_free: return NULL; } @@ -342,7 +634,8 @@ error: */ static struct ust_app_channel *alloc_ust_app_channel(char *name, - struct lttng_ust_channel *attr) + struct ust_app_session *ua_sess, + struct lttng_ust_channel_attr *attr) { struct ust_app_channel *ua_chan; @@ -359,16 +652,28 @@ struct ust_app_channel *alloc_ust_app_channel(char *name, ua_chan->enabled = 1; ua_chan->handle = -1; + ua_chan->key = get_next_channel_key(); ua_chan->ctx = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); ua_chan->events = lttng_ht_new(0, LTTNG_HT_TYPE_STRING); lttng_ht_node_init_str(&ua_chan->node, ua_chan->name); CDS_INIT_LIST_HEAD(&ua_chan->streams.head); + /* Initialize UST registry. */ + ust_registry_channel_init(&ua_sess->registry, &ua_chan->registry); + /* Copy attributes */ if (attr) { - memcpy(&ua_chan->attr, attr, sizeof(ua_chan->attr)); + /* Translate from lttng_ust_channel to ustctl_consumer_channel_attr. */ + ua_chan->attr.subbuf_size = attr->subbuf_size; + ua_chan->attr.num_subbuf = attr->num_subbuf; + ua_chan->attr.overwrite = attr->overwrite; + 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; } + /* By default, the channel is a per cpu channel. */ + ua_chan->attr.type = LTTNG_UST_CHAN_PER_CPU; DBG3("UST app channel %s allocated", ua_chan->name); @@ -378,6 +683,28 @@ error: return NULL; } +/* + * Allocate and initialize a UST app stream. + * + * Return newly allocated stream pointer or NULL on error. + */ +struct ust_app_stream *ust_app_alloc_stream(void) +{ + struct ust_app_stream *stream = NULL; + + stream = zmalloc(sizeof(*stream)); + if (stream == NULL) { + PERROR("zmalloc ust app stream"); + goto error; + } + + /* Zero could be a valid value for a handle so flag it to -1. */ + stream->handle = -1; + +error: + return stream; +} + /* * Alloc new UST app event. */ @@ -481,6 +808,29 @@ error: return NULL; } +/* + * Find an ust_app using the notify sock and return it. RCU read side lock must + * be held before calling this helper function. + */ +static struct ust_app *find_app_by_notify_sock(int sock) +{ + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter iter; + + lttng_ht_lookup(ust_app_ht_by_notify_sock, (void *)((unsigned long) sock), + &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node == NULL) { + DBG2("UST app find by notify sock %d not found", sock); + goto error; + } + + return caa_container_of(node, struct ust_app, notify_sock_n); + +error: + return NULL; +} + /* * Lookup for an ust app event based on event name, filter bytecode and the * event loglevel. @@ -519,6 +869,8 @@ end: /* * Create the channel context on the tracer. + * + * Called with UST app session lock held. */ static int create_ust_channel_context(struct ust_app_channel *ua_chan, @@ -526,20 +878,27 @@ int create_ust_channel_context(struct ust_app_channel *ua_chan, { int ret; - health_code_update(&health_thread_cmd); + health_code_update(); ret = ustctl_add_context(app->sock, &ua_ctx->ctx, ua_chan->obj, &ua_ctx->obj); if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app create channel context failed for app (pid: %d) " + "with ret %d", app->pid, ret); + } else { + DBG3("UST app disable event failed. Application is dead."); + } goto error; } ua_ctx->handle = ua_ctx->obj->handle; - DBG2("UST app context created successfully for channel %s", ua_chan->name); + DBG2("UST app context handle %d created successfully for channel %s", + ua_ctx->handle, ua_chan->name); error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } @@ -552,7 +911,7 @@ int set_ust_event_filter(struct ust_app_event *ua_event, { int ret; - health_code_update(&health_thread_cmd); + health_code_update(); if (!ua_event->filter) { ret = 0; @@ -562,13 +921,19 @@ int set_ust_event_filter(struct ust_app_event *ua_event, ret = ustctl_set_filter(app->sock, ua_event->filter, ua_event->obj); if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app event %s filter failed for app (pid: %d) " + "with ret %d", ua_event->attr.name, app->pid, ret); + } else { + DBG3("UST app filter event failed. Application is dead."); + } goto error; } DBG2("UST filter set successfully for event %s", ua_event->name); error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } @@ -580,13 +945,17 @@ static int disable_ust_event(struct ust_app *app, { int ret; - health_code_update(&health_thread_cmd); + health_code_update(); ret = ustctl_disable(app->sock, ua_event->obj); if (ret < 0) { - 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); + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app event %s disable failed for app (pid: %d) " + "and session handle %d with ret %d", + ua_event->attr.name, app->pid, ua_sess->handle, ret); + } else { + DBG3("UST app disable event failed. Application is dead."); + } goto error; } @@ -594,7 +963,7 @@ static int disable_ust_event(struct ust_app *app, ua_event->attr.name, app->pid); error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } @@ -606,13 +975,17 @@ static int disable_ust_channel(struct ust_app *app, { int ret; - health_code_update(&health_thread_cmd); + health_code_update(); ret = ustctl_disable(app->sock, ua_chan->obj); if (ret < 0) { - ERR("UST app channel %s disable failed for app (pid: %d) " - "and session handle %d with ret %d", - ua_chan->name, app->pid, ua_sess->handle, ret); + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app channel %s disable failed for app (pid: %d) " + "and session handle %d with ret %d", + ua_chan->name, app->pid, ua_sess->handle, ret); + } else { + DBG3("UST app disable channel failed. Application is dead."); + } goto error; } @@ -620,7 +993,7 @@ static int disable_ust_channel(struct ust_app *app, ua_chan->name, app->pid); error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } @@ -632,13 +1005,17 @@ static int enable_ust_channel(struct ust_app *app, { int ret; - health_code_update(&health_thread_cmd); + health_code_update(); ret = ustctl_enable(app->sock, ua_chan->obj); if (ret < 0) { - ERR("UST app channel %s enable failed for app (pid: %d) " - "and session handle %d with ret %d", - ua_chan->name, app->pid, ua_sess->handle, ret); + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app channel %s enable failed for app (pid: %d) " + "and session handle %d with ret %d", + ua_chan->name, app->pid, ua_sess->handle, ret); + } else { + DBG3("UST app enable channel failed. Application is dead."); + } goto error; } @@ -648,7 +1025,7 @@ static int enable_ust_channel(struct ust_app *app, ua_chan->name, app->pid); error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } @@ -660,13 +1037,17 @@ static int enable_ust_event(struct ust_app *app, { int ret; - health_code_update(&health_thread_cmd); + health_code_update(); ret = ustctl_enable(app->sock, ua_event->obj); if (ret < 0) { - 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); + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app event %s enable failed for app (pid: %d) " + "and session handle %d with ret %d", + ua_event->attr.name, app->pid, ua_sess->handle, ret); + } else { + DBG3("UST app enable event failed. Application is dead."); + } goto error; } @@ -674,134 +1055,135 @@ static int enable_ust_event(struct ust_app *app, ua_event->attr.name, app->pid); error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } /* - * Open metadata onto the UST tracer for a UST session. + * Create the specified channel onto the UST tracer for a UST session. This + * MUST be called with UST app session lock held. + * + * Return 0 on success. On error, a negative value is returned. */ -static int open_ust_metadata(struct ust_app *app, - struct ust_app_session *ua_sess) +static int create_ust_channel(struct ust_app *app, + struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan, + struct consumer_output *consumer) { int ret; - struct lttng_ust_channel_attr uattr; + unsigned int nb_fd = 0; + struct consumer_socket *socket; + struct ust_app_stream *stream, *stmp; - health_code_update(&health_thread_cmd); + assert(app); + assert(ua_sess); + assert(ua_chan); + assert(consumer); - uattr.overwrite = ua_sess->metadata->attr.overwrite; - uattr.subbuf_size = ua_sess->metadata->attr.subbuf_size; - uattr.num_subbuf = ua_sess->metadata->attr.num_subbuf; - uattr.switch_timer_interval = - ua_sess->metadata->attr.switch_timer_interval; - uattr.read_timer_interval = - ua_sess->metadata->attr.read_timer_interval; - uattr.output = ua_sess->metadata->attr.output; + rcu_read_lock(); + health_code_update(); - /* We are going to receive 2 fds, we need to reserve them. */ - ret = lttng_fd_get(LTTNG_FD_APPS, 2); - if (ret < 0) { - ERR("Exhausted number of available FD upon metadata open"); - goto error; - } - /* UST tracer metadata creation */ - ret = ustctl_open_metadata(app->sock, ua_sess->handle, &uattr, - &ua_sess->metadata->obj); - if (ret < 0) { - ERR("UST app open metadata failed for app pid:%d with ret %d", - app->pid, ret); + /* Get the right consumer socket for the application. */ + socket = find_consumer_socket_by_bitness(app->bits_per_long, consumer); + if (!socket) { + ret = -1; goto error; } - ua_sess->metadata->handle = ua_sess->metadata->obj->handle; - -error: - health_code_update(&health_thread_cmd); - return ret; -} - -/* - * Create stream onto the UST tracer for a UST session. - */ -static int create_ust_stream(struct ust_app *app, - struct ust_app_session *ua_sess) -{ - int ret; - - health_code_update(&health_thread_cmd); + health_code_update(); - /* We are going to receive 2 fds, we need to reserve them. */ - ret = lttng_fd_get(LTTNG_FD_APPS, 2); - if (ret < 0) { - ERR("Exhausted number of available FD upon metadata stream create"); - goto error; - } - ret = ustctl_create_stream(app->sock, ua_sess->metadata->obj, - &ua_sess->metadata->stream_obj); + /* + * Ask consumer to create channel. The consumer will return the number of + * stream we have to expect. + */ + ret = ust_consumer_ask_channel(ua_sess, ua_chan, consumer, socket); if (ret < 0) { - ERR("UST create metadata stream failed"); goto error; } -error: - health_code_update(&health_thread_cmd); - return ret; -} - -/* - * Create the specified channel onto the UST tracer for a UST session. - */ -static int create_ust_channel(struct ust_app *app, - struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan) -{ - int ret; - - health_code_update(&health_thread_cmd); - - /* TODO: remove cast and use lttng-ust-abi.h */ + /* + * Compute the number of fd needed before receiving them. It must be 2 per + * stream (2 being the default value here). + */ + nb_fd = DEFAULT_UST_STREAM_FD_NUM * ua_chan->expected_stream_count; - /* We are going to receive 2 fds, we need to reserve them. */ - ret = lttng_fd_get(LTTNG_FD_APPS, 2); + /* Reserve the amount of file descriptor we need. */ + ret = lttng_fd_get(LTTNG_FD_APPS, nb_fd); if (ret < 0) { ERR("Exhausted number of available FD upon create channel"); - goto error; + goto error_fd_get; } - health_code_update(&health_thread_cmd); + health_code_update(); - ret = ustctl_create_channel(app->sock, ua_sess->handle, - (struct lttng_ust_channel_attr *)&ua_chan->attr, &ua_chan->obj); + /* + * Now get the channel from the consumer. This call wil populate the stream + * list of that channel and set the ust object. + */ + ret = ust_consumer_get_channel(socket, ua_chan); if (ret < 0) { - ERR("Creating channel %s for app (pid: %d, sock: %d) " - "and session handle %d with ret %d", - ua_chan->name, app->pid, app->sock, - ua_sess->handle, ret); - lttng_fd_put(LTTNG_FD_APPS, 2); - goto error; + goto error_destroy; } - ua_chan->handle = ua_chan->obj->handle; - - DBG2("UST app channel %s created successfully for pid:%d and sock:%d", - ua_chan->name, app->pid, app->sock); + /* Send channel to the application. */ + ret = ust_consumer_send_channel_to_ust(app, ua_sess, ua_chan); + if (ret < 0) { + goto error; + } - health_code_update(&health_thread_cmd); + health_code_update(); - /* If channel is not enabled, disable it on the tracer */ - if (!ua_chan->enabled) { - ret = disable_ust_channel(app, ua_sess, ua_chan); + /* 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) { + goto error; + } + /* We don't need the stream anymore once sent to the tracer. */ + cds_list_del(&stream->list); + delete_ust_app_stream(-1, stream); + } + + /* Flag the channel that it is sent to the application. */ + ua_chan->is_sent = 1; + /* Assign session to channel. */ + ua_chan->session = ua_sess; + /* Initialize ust objd object using the received handle and add it. */ + lttng_ht_node_init_ulong(&ua_chan->ust_objd_node, ua_chan->handle); + lttng_ht_add_unique_ulong(app->ust_objd, &ua_chan->ust_objd_node); + + health_code_update(); + + /* If channel is not enabled, disable it on the tracer */ + if (!ua_chan->enabled) { + ret = disable_ust_channel(app, ua_sess, ua_chan); if (ret < 0) { goto error; } } + rcu_read_unlock(); + return 0; + +error_destroy: + lttng_fd_put(LTTNG_FD_APPS, nb_fd); +error_fd_get: + /* + * Initiate a destroy channel on the consumer since we had an error + * handling it on our side. The return value is of no importance since we + * already have a ret value set by the previous error that we need to + * return. + */ + (void) ust_consumer_destroy_channel(socket, ua_chan); error: - health_code_update(&health_thread_cmd); + health_code_update(); + rcu_read_unlock(); return ret; } /* * Create the specified event onto the UST tracer for a UST session. + * + * Should be called with session mutex held. */ static int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, @@ -809,14 +1191,18 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, { int ret = 0; - health_code_update(&health_thread_cmd); + health_code_update(); /* Create UST event on tracer */ ret = ustctl_create_event(app->sock, &ua_event->attr, ua_chan->obj, &ua_event->obj); if (ret < 0) { - ERR("Error ustctl create event %s for app pid: %d with ret %d", - ua_event->attr.name, app->pid, ret); + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("Error ustctl create event %s for app pid: %d with ret %d", + ua_event->attr.name, app->pid, ret); + } else { + DBG3("UST app create event failed. Application is dead."); + } goto error; } @@ -825,7 +1211,7 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, DBG2("UST app event %s created successfully for pid:%d", ua_event->attr.name, app->pid); - health_code_update(&health_thread_cmd); + health_code_update(); /* Set filter if one is present. */ if (ua_event->filter) { @@ -860,7 +1246,7 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, } error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } @@ -901,8 +1287,18 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, strncpy(ua_chan->name, uchan->name, sizeof(ua_chan->name)); ua_chan->name[sizeof(ua_chan->name) - 1] = '\0'; - /* Copy event attributes */ - memcpy(&ua_chan->attr, &uchan->attr, sizeof(ua_chan->attr)); + + /* 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. + */ ua_chan->enabled = uchan->enabled; @@ -928,7 +1324,7 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, continue; } shadow_copy_event(ua_event, uevent); - add_unique_ust_app_event(ua_chan->events, ua_event); + add_unique_ust_app_event(ua_chan, ua_event); } } @@ -985,13 +1381,19 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, DBG2("Channel %s not found on shadow session copy, creating it", uchan->name); - ua_chan = alloc_ust_app_channel(uchan->name, &uchan->attr); + ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, &uchan->attr); if (ua_chan == NULL) { /* malloc failed FIXME: Might want to do handle ENOMEM .. */ continue; } - 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); } } @@ -1030,42 +1432,63 @@ error: } /* - * Create a UST session onto the tracer of app and add it the session - * hashtable. + * Create a session on the tracer side for the given app. + * + * On success, ua_sess_ptr is populated with the session pointer or else left + * untouched. If the session was created, is_created is set to 1. On error, + * it's left untouched. Note that ua_sess_ptr is mandatory but is_created can + * be NULL. * - * Return ust app session or NULL on 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 struct ust_app_session *create_ust_app_session( - struct ltt_ust_session *usess, struct ust_app *app) +static int create_ust_app_session(struct ltt_ust_session *usess, + struct ust_app *app, struct ust_app_session **ua_sess_ptr, + int *is_created) { + int ret, created = 0; struct ust_app_session *ua_sess; - health_code_update(&health_thread_cmd); + assert(usess); + assert(app); + assert(ua_sess_ptr); + + health_code_update(); ua_sess = lookup_session_by_app(usess, app); if (ua_sess == NULL) { DBG2("UST app pid: %d session id %d not found, creating it", app->pid, usess->id); - ua_sess = alloc_ust_app_session(); + ua_sess = alloc_ust_app_session(app); if (ua_sess == NULL) { /* Only malloc can failed so something is really wrong */ - goto end; + ret = -ENOMEM; + goto error; } shadow_copy_session(ua_sess, usess, app); + created = 1; } - health_code_update(&health_thread_cmd); + health_code_update(); if (ua_sess->handle == -1) { - int ret; - ret = ustctl_create_session(app->sock); if (ret < 0) { - ERR("Creating session for app pid %d", app->pid); - delete_ust_app_session(-1, ua_sess); - /* This means that the tracer is gone... */ - ua_sess = (void*) -1UL; - goto end; + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("Creating session for app pid %d with ret %d", + app->pid, ret); + } else { + DBG("UST app creating session failed. Application is dead"); + } + delete_ust_app_session(-1, ua_sess, app); + if (ret != -ENOMEM) { + /* + * Tracer is probably gone or got an internal error so let's + * behave like it will soon unregister or not usable. + */ + ret = -ENOTCONN; + } + goto error; } ua_sess->handle = ret; @@ -1077,13 +1500,31 @@ static struct ust_app_session *create_ust_app_session( DBG2("UST app session created successfully with handle %d", ret); } -end: - health_code_update(&health_thread_cmd); - return ua_sess; + /* + * Assign consumer if not already set. For one application, there is only + * one possible consumer has of now. + */ + if (!ua_sess->consumer) { + ua_sess->consumer = usess->consumer; + } + + *ua_sess_ptr = ua_sess; + if (is_created) { + *is_created = created; + } + + /* Everything went well. */ + ret = 0; + +error: + health_code_update(); + return ret; } /* * Create a context for the channel on the tracer. + * + * Called with UST app session lock held. */ static int create_ust_app_channel_context(struct ust_app_session *ua_sess, @@ -1125,6 +1566,8 @@ error: /* * Enable on the tracer side a ust app event for the session and channel. + * + * Called with UST app session lock held. */ static int enable_ust_app_event(struct ust_app_session *ua_sess, @@ -1215,9 +1658,12 @@ error: /* * Create UST app channel and create it on the tracer. Set ua_chanp of the * newly created channel if not NULL. + * + * Called with UST app session lock held. */ static int create_ust_app_channel(struct ust_app_session *ua_sess, struct ltt_ust_channel *uchan, struct ust_app *app, + struct consumer_output *consumer, enum lttng_ust_chan_type type, struct ust_app_channel **ua_chanp) { int ret = 0; @@ -1233,7 +1679,7 @@ static int create_ust_app_channel(struct ust_app_session *ua_sess, goto end; } - ua_chan = alloc_ust_app_channel(uchan->name, &uchan->attr); + ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, &uchan->attr); if (ua_chan == NULL) { /* Only malloc can fail here */ ret = -ENOMEM; @@ -1241,19 +1687,20 @@ static int create_ust_app_channel(struct ust_app_session *ua_sess, } shadow_copy_channel(ua_chan, uchan); - ret = create_ust_channel(app, ua_sess, ua_chan); + /* Set channel type. */ + ua_chan->attr.type = type; + + ret = create_ust_channel(app, ua_sess, ua_chan, consumer); if (ret < 0) { - /* Not found previously means that it does not exist on the tracer */ - assert(ret != -LTTNG_UST_ERR_EXIST); goto error; } - /* Only add the channel if successful on the tracer side. */ - lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node); - 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; @@ -1263,12 +1710,14 @@ end: return 0; error: - delete_ust_app_channel(-1, ua_chan); + delete_ust_app_channel(ua_chan->is_sent ? app->sock : -1, ua_chan, app); 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_event(struct ust_app_session *ua_sess, @@ -1303,7 +1752,7 @@ int create_ust_app_event(struct ust_app_session *ua_sess, goto error; } - add_unique_ust_app_event(ua_chan->events, ua_event); + add_unique_ust_app_event(ua_chan, ua_event); DBG2("UST app create event %s for PID %d completed", ua_event->name, app->pid); @@ -1319,62 +1768,81 @@ error: /* * Create UST metadata and open it on the tracer side. + * + * Called with UST app session lock held. */ static int create_ust_app_metadata(struct ust_app_session *ua_sess, - char *pathname, struct ust_app *app) + struct ust_app *app, struct consumer_output *consumer) { int ret = 0; + struct ust_app_channel *metadata; + struct consumer_socket *socket; - if (ua_sess->metadata == NULL) { - /* Allocate UST metadata */ - ua_sess->metadata = trace_ust_create_metadata(pathname); - if (ua_sess->metadata == NULL) { - /* malloc() failed */ - goto error; - } - - ret = open_ust_metadata(app, ua_sess); - if (ret < 0) { - DBG3("Opening metadata failed. Cleaning up memory"); + assert(ua_sess); + assert(app); + assert(consumer); - /* Cleanup failed metadata struct */ - free(ua_sess->metadata); - /* - * This is very important because delete_ust_app_session check if - * the pointer is null or not in order to delete the metadata. - */ - ua_sess->metadata = NULL; - goto error; - } + if (ua_sess->metadata) { + /* Already exist. Return success. */ + goto end; + } - DBG2("UST metadata opened for app pid %d", app->pid); + /* Allocate UST metadata */ + metadata = alloc_ust_app_channel(DEFAULT_METADATA_NAME, ua_sess, NULL); + if (!metadata) { + /* malloc() failed */ + ret = -ENOMEM; + goto error; } - /* Open UST metadata stream */ - if (ua_sess->metadata->stream_obj == NULL) { - ret = create_ust_stream(app, ua_sess); - if (ret < 0) { - goto error; - } + /* Set default attributes for metadata. */ + metadata->attr.overwrite = DEFAULT_CHANNEL_OVERWRITE; + metadata->attr.subbuf_size = default_get_metadata_subbuf_size(); + metadata->attr.num_subbuf = DEFAULT_METADATA_SUBBUF_NUM; + metadata->attr.switch_timer_interval = DEFAULT_UST_CHANNEL_SWITCH_TIMER; + metadata->attr.read_timer_interval = DEFAULT_UST_CHANNEL_READ_TIMER; + metadata->attr.output = LTTNG_UST_MMAP; + metadata->attr.type = LTTNG_UST_CHAN_METADATA; - ret = snprintf(ua_sess->metadata->pathname, PATH_MAX, - "%s/metadata", ua_sess->path); - if (ret < 0) { - PERROR("asprintf UST create stream"); - goto error; - } + /* Get the right consumer socket for the application. */ + socket = find_consumer_socket_by_bitness(app->bits_per_long, consumer); + if (!socket) { + ret = -EINVAL; + goto error_consumer; + } - DBG2("UST metadata stream object created for app pid %d", - app->pid); - } else { - ERR("Attempting to create stream without metadata opened"); - goto error; + /* + * Ask the metadata channel creation to the consumer. The metadata object + * will be created by the consumer and kept their. However, the stream is + * never added or monitored until we do a first push metadata to the + * consumer. + */ + ret = ust_consumer_ask_channel(ua_sess, metadata, consumer, socket); + if (ret < 0) { + goto error_consumer; } - return 0; + /* + * The setup command will make the metadata stream be sent to the relayd, + * if applicable, and the thread managing the metadatas. This is important + * because after this point, if an error occurs, the only way the stream + * can be deleted is to be monitored in the consumer. + */ + ret = ust_consumer_setup_metadata(socket, metadata); + if (ret < 0) { + goto error_consumer; + } + + ua_sess->metadata = metadata; + + DBG2("UST metadata created for app pid %d", app->pid); +end: + return 0; +error_consumer: + delete_ust_app_channel(-1, metadata, app); error: - return -1; + return ret; } /* @@ -1386,114 +1854,159 @@ struct lttng_ht *ust_app_get_ht(void) } /* - * Return ust app pointer or NULL if not found. + * Return ust app pointer or NULL if not found. RCU read side lock MUST be + * acquired before calling this function. */ struct ust_app *ust_app_find_by_pid(pid_t pid) { + struct ust_app *app = NULL; struct lttng_ht_node_ulong *node; struct lttng_ht_iter iter; - rcu_read_lock(); lttng_ht_lookup(ust_app_ht, (void *)((unsigned long) pid), &iter); node = lttng_ht_iter_get_node_ulong(&iter); if (node == NULL) { DBG2("UST app no found with pid %d", pid); goto error; } - rcu_read_unlock(); DBG2("Found UST app by pid %d", pid); - return caa_container_of(node, struct ust_app, pid_n); + app = caa_container_of(node, struct ust_app, pid_n); error: - rcu_read_unlock(); - return NULL; + return app; } /* - * Using pid and uid (of the app), allocate a new ust_app struct and - * add it to the global traceable app list. + * Allocate and init an UST app object using the registration information and + * the command socket. This is called when the command socket connects to the + * session daemon. * - * On success, return 0, else return malloc -ENOMEM, or -EINVAL if app - * bitness is not supported. + * The object is returned on success or else NULL. */ -int ust_app_register(struct ust_register_msg *msg, int sock) +struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) { - struct ust_app *lta; - int ret; + struct ust_app *lta = NULL; + + assert(msg); + assert(sock >= 0); + + DBG3("UST app creating application for socket %d", sock); if ((msg->bits_per_long == 64 && (uatomic_read(&ust_consumerd64_fd) == -EINVAL)) || (msg->bits_per_long == 32 && (uatomic_read(&ust_consumerd32_fd) == -EINVAL))) { ERR("Registration failed: application \"%s\" (pid: %d) has " - "%d-bit long, but no consumerd for this long size is available.\n", - msg->name, msg->pid, msg->bits_per_long); - ret = close(sock); - if (ret) { - PERROR("close"); - } - lttng_fd_put(LTTNG_FD_APPS, 1); - return -EINVAL; - } - if (msg->major != LTTNG_UST_COMM_MAJOR) { - ERR("Registration failed: application \"%s\" (pid: %d) has " - "communication protocol version %u.%u, but sessiond supports 2.x.\n", - msg->name, msg->pid, msg->major, msg->minor); - ret = close(sock); - if (ret) { - PERROR("close"); - } - lttng_fd_put(LTTNG_FD_APPS, 1); - return -EINVAL; + "%d-bit long, but no consumerd for this size is available.\n", + msg->name, msg->pid, msg->bits_per_long); + goto error; } + lta = zmalloc(sizeof(struct ust_app)); if (lta == NULL) { PERROR("malloc"); - return -ENOMEM; + goto error; } lta->ppid = msg->ppid; lta->uid = msg->uid; lta->gid = msg->gid; - lta->compatible = 0; /* Not compatible until proven */ + lta->bits_per_long = msg->bits_per_long; + lta->uint8_t_alignment = msg->uint8_t_alignment; + lta->uint16_t_alignment = msg->uint16_t_alignment; + lta->uint32_t_alignment = msg->uint32_t_alignment; + lta->uint64_t_alignment = msg->uint64_t_alignment; + lta->long_alignment = msg->long_alignment; + lta->byte_order = msg->byte_order; + lta->v_major = msg->major; lta->v_minor = msg->minor; - strncpy(lta->name, msg->name, sizeof(lta->name)); - lta->name[16] = '\0'; lta->sessions = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + lta->ust_objd = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + lta->notify_sock = -1; + + /* Copy name and make sure it's NULL terminated. */ + strncpy(lta->name, msg->name, sizeof(lta->name)); + lta->name[UST_APP_PROCNAME_LEN] = '\0'; + + /* + * Before this can be called, when receiving the registration information, + * the application compatibility is checked. So, at this point, the + * application can work with this session daemon. + */ + lta->compatible = 1; lta->pid = msg->pid; - lttng_ht_node_init_ulong(<a->pid_n, (unsigned long)lta->pid); + lttng_ht_node_init_ulong(<a->pid_n, (unsigned long) lta->pid); lta->sock = sock; - lttng_ht_node_init_ulong(<a->sock_n, (unsigned long)lta->sock); + lttng_ht_node_init_ulong(<a->sock_n, (unsigned long) lta->sock); CDS_INIT_LIST_HEAD(<a->teardown_head); +error: + return lta; +} + +/* + * For a given application object, add it to every hash table. + */ +void ust_app_add(struct ust_app *app) +{ + assert(app); + assert(app->notify_sock >= 0); + rcu_read_lock(); /* * On a re-registration, we want to kick out the previous registration of * that pid */ - lttng_ht_add_replace_ulong(ust_app_ht, <a->pid_n); + lttng_ht_add_replace_ulong(ust_app_ht, &app->pid_n); /* * The socket _should_ be unique until _we_ call close. So, a add_unique * for the ust_app_ht_by_sock is used which asserts fail if the entry was * already in the table. */ - lttng_ht_add_unique_ulong(ust_app_ht_by_sock, <a->sock_n); + lttng_ht_add_unique_ulong(ust_app_ht_by_sock, &app->sock_n); + + /* Add application to the notify socket hash table. */ + lttng_ht_node_init_ulong(&app->notify_sock_n, app->notify_sock); + lttng_ht_add_unique_ulong(ust_app_ht_by_notify_sock, &app->notify_sock_n); + + DBG("App registered with pid:%d ppid:%d uid:%d gid:%d sock:%d name:%s " + "notify_sock:%d (version %d.%d)", app->pid, app->ppid, app->uid, + app->gid, app->sock, app->name, app->notify_sock, app->v_major, + app->v_minor); rcu_read_unlock(); +} - DBG("App registered with pid:%d ppid:%d uid:%d gid:%d sock:%d name:%s" - " (version %d.%d)", lta->pid, lta->ppid, lta->uid, lta->gid, - lta->sock, lta->name, lta->v_major, lta->v_minor); +/* + * Set the application version into the object. + * + * Return 0 on success else a negative value either an errno code or a + * LTTng-UST error code. + */ +int ust_app_version(struct ust_app *app) +{ + int ret; - return 0; + assert(app); + + ret = ustctl_tracer_version(app->sock, &app->version); + if (ret < 0) { + if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { + ERR("UST app %d verson failed with ret %d", app->sock, ret); + } else { + DBG3("UST app %d verion failed. Application is dead", app->sock); + } + } + + return ret; } /* @@ -1515,27 +2028,30 @@ void ust_app_unregister(int sock) /* Get the node reference for a call_rcu */ lttng_ht_lookup(ust_app_ht_by_sock, (void *)((unsigned long) sock), &iter); node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - ERR("Unable to find app by sock %d", sock); - goto error; - } + assert(node); lta = caa_container_of(node, struct ust_app, sock_n); - DBG("PID %d unregistering with sock %d", lta->pid, sock); /* Remove application from PID hash table */ ret = lttng_ht_del(ust_app_ht_by_sock, &iter); assert(!ret); - /* Assign second node for deletion */ - iter.iter.node = <a->pid_n.node; + /* + * Remove application from notify hash table. The thread handling the + * notify socket could have deleted the node so ignore on error because + * either way it's valid. The close of that socket is handled by the other + * thread. + */ + iter.iter.node = <a->notify_sock_n.node; + (void) lttng_ht_del(ust_app_ht_by_notify_sock, &iter); /* * Ignore return value since the node might have been removed before by an * add replace during app registration because the PID can be reassigned by * the OS. */ + iter.iter.node = <a->pid_n.node; ret = lttng_ht_del(ust_app_ht, &iter); if (ret) { DBG3("Unregister app by PID %d failed. This can happen on pid reuse", @@ -1555,13 +2071,29 @@ void ust_app_unregister(int sock) * Add session to list for teardown. This is safe since at this point we * are the only one using this list. */ + pthread_mutex_lock(&ua_sess->lock); + + /* + * Normally, this is done in the delete session process which is + * executed in the call rcu below. However, upon registration we can't + * afford to wait for the grace period before pushing data or else the + * data pending feature can race between the unregistration and stop + * command where the data pending command is sent *before* the grace + * period ended. + * + * The close metadata below nullifies the metadata pointer in the + * session so the delete session will NOT push/close a second time. + */ + (void) push_metadata(lta, ua_sess); + (void) close_metadata(lta, ua_sess); + cds_list_add(&ua_sess->teardown_node, <a->teardown_head); + pthread_mutex_unlock(&ua_sess->lock); } /* Free memory */ call_rcu(<a->pid_n.head, delete_ust_app_rcu); -error: rcu_read_unlock(); return; } @@ -1604,7 +2136,7 @@ int ust_app_list_events(struct lttng_event **events) cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { struct lttng_ust_tracepoint_iter uiter; - health_code_update(&health_thread_cmd); + health_code_update(); if (!app->compatible) { /* @@ -1615,14 +2147,28 @@ int ust_app_list_events(struct lttng_event **events) } handle = ustctl_tracepoint_list(app->sock); if (handle < 0) { - ERR("UST app list events getting handle failed for app pid %d", - app->pid); + if (handle != -EPIPE && handle != -LTTNG_UST_ERR_EXITING) { + ERR("UST app list events getting handle failed for app pid %d", + app->pid); + } continue; } while ((ret = ustctl_tracepoint_list_get(app->sock, handle, &uiter)) != -LTTNG_UST_ERR_NOENT) { - health_code_update(&health_thread_cmd); + /* Handle ustctl error. */ + if (ret < 0) { + free(tmp_event); + if (ret != -LTTNG_UST_ERR_EXITING || ret != -EPIPE) { + ERR("UST app tp list get failed for app %d with ret %d", + app->sock, ret); + } else { + DBG3("UST app tp list get failed. Application is dead"); + } + goto rcu_error; + } + + health_code_update(); if (count >= nbmem) { /* In case the realloc fails, we free the memory */ void *ptr; @@ -1656,7 +2202,7 @@ int ust_app_list_events(struct lttng_event **events) rcu_error: rcu_read_unlock(); error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } @@ -1684,7 +2230,7 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { struct lttng_ust_field_iter uiter; - health_code_update(&health_thread_cmd); + health_code_update(); if (!app->compatible) { /* @@ -1695,14 +2241,28 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) } handle = ustctl_tracepoint_field_list(app->sock); if (handle < 0) { - ERR("UST app list event fields getting handle failed for app pid %d", - app->pid); + if (handle != -EPIPE && handle != -LTTNG_UST_ERR_EXITING) { + ERR("UST app list field getting handle failed for app pid %d", + app->pid); + } continue; } while ((ret = ustctl_tracepoint_field_list_get(app->sock, handle, &uiter)) != -LTTNG_UST_ERR_NOENT) { - health_code_update(&health_thread_cmd); + /* Handle ustctl error. */ + if (ret < 0) { + free(tmp_event); + if (ret != -LTTNG_UST_ERR_EXITING || ret != -EPIPE) { + ERR("UST app tp list field failed for app %d with ret %d", + app->sock, ret); + } else { + DBG3("UST app tp list field failed. Application is dead"); + } + goto rcu_error; + } + + health_code_update(); if (count >= nbmem) { /* In case the realloc fails, we free the memory */ void *ptr; @@ -1741,7 +2301,7 @@ int ust_app_list_event_fields(struct lttng_event_field **fields) rcu_error: rcu_read_unlock(); error: - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } @@ -1771,9 +2331,17 @@ void ust_app_clean_list(void) assert(!ret); } + /* Cleanup notify socket hash table */ + cds_lfht_for_each_entry(ust_app_ht_by_notify_sock->ht, &iter.iter, app, + notify_sock_n.node) { + ret = lttng_ht_del(ust_app_ht_by_notify_sock, &iter); + assert(!ret); + } + /* Destroy is done only when the ht is empty */ lttng_ht_destroy(ust_app_ht); lttng_ht_destroy(ust_app_ht_by_sock); + lttng_ht_destroy(ust_app_ht_by_notify_sock); rcu_read_unlock(); } @@ -1785,6 +2353,7 @@ void ust_app_ht_alloc(void) { ust_app_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); ust_app_ht_by_sock = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + ust_app_ht_by_notify_sock = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); } /* @@ -2030,10 +2599,10 @@ int ust_app_disable_all_event_glb(struct ltt_ust_session *usess, int ust_app_create_channel_glb(struct ltt_ust_session *usess, struct ltt_ust_channel *uchan) { - int ret = 0; + int ret = 0, created; struct lttng_ht_iter iter; struct ust_app *app; - struct ust_app_session *ua_sess; + struct ust_app_session *ua_sess = NULL; /* Very wrong code flow */ assert(usess); @@ -2058,25 +2627,37 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess, * that if session exist, it will simply return a pointer to the ust * app session. */ - ua_sess = create_ust_app_session(usess, app); - if (ua_sess == NULL) { - /* The malloc() failed. */ - ret = -ENOMEM; - goto error_rcu_unlock; - } else if (ua_sess == (void *) -1UL) { - /* - * The application's socket is not valid. Either a bad socket or a - * timeout on it. We can't inform yet the caller that for a - * specific app, the session failed so we continue here. - */ - continue; + ret = create_ust_app_session(usess, app, &ua_sess, &created); + 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. + */ + continue; + case -ENOMEM: + default: + goto error_rcu_unlock; + } } + assert(ua_sess); + pthread_mutex_lock(&ua_sess->lock); /* Create channel onto application. We don't need the chan ref. */ - ret = create_ust_app_channel(ua_sess, uchan, app, NULL); - if (ret < 0 && ret == -ENOMEM) { - /* No more memory is a fatal error. Stop right now. */ - goto error_rcu_unlock; + ret = create_ust_app_channel(ua_sess, uchan, app, usess->consumer, + LTTNG_UST_CHAN_PER_CPU, NULL); + 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); + } } } @@ -2125,6 +2706,8 @@ int ust_app_enable_event_glb(struct ltt_ust_session *usess, continue; } + pthread_mutex_lock(&ua_sess->lock); + /* 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); @@ -2139,13 +2722,16 @@ int ust_app_enable_event_glb(struct ltt_ust_session *usess, if (ua_event == NULL) { DBG3("UST app enable event %s not found for app PID %d." "Skipping app", uevent->attr.name, app->pid); - continue; + goto next_app; } ret = enable_ust_app_event(ua_sess, ua_event, app); if (ret < 0) { + pthread_mutex_unlock(&ua_sess->lock); goto error; } + next_app: + pthread_mutex_unlock(&ua_sess->lock); } error: @@ -2187,6 +2773,7 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, continue; } + pthread_mutex_lock(&ua_sess->lock); /* 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); @@ -2196,6 +2783,7 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); ret = create_ust_app_event(ua_sess, ua_chan, uevent, app); + pthread_mutex_unlock(&ua_sess->lock); if (ret < 0) { if (ret != -LTTNG_UST_ERR_EXIST) { /* Possible value at this point: -ENOMEM. If so, we stop! */ @@ -2218,11 +2806,7 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) { int ret = 0; - struct lttng_ht_iter iter; struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; - struct ltt_ust_stream *ustream; - struct consumer_socket *socket; DBG("Starting tracing for ust app pid %d", app->pid); @@ -2238,6 +2822,8 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) goto end; } + pthread_mutex_lock(&ua_sess->lock); + /* Upon restart, we skip the setup, already done */ if (ua_sess->started) { goto skip_setup; @@ -2251,125 +2837,55 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) if (ret < 0) { if (ret != -EEXIST) { ERR("Trace directory creation error"); - ret = -1; - goto error_rcu_unlock; - } - } - } - - ret = create_ust_app_metadata(ua_sess, usess->pathname, app); - if (ret < 0) { - ret = LTTNG_ERR_UST_META_FAIL; - goto error_rcu_unlock; - } - - /* For each channel */ - cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, - node.node) { - /* Create all streams */ - while (1) { - /* Create UST stream */ - ustream = zmalloc(sizeof(*ustream)); - if (ustream == NULL) { - PERROR("zmalloc ust stream"); - goto error_rcu_unlock; - } - - /* We are going to receive 2 fds, we need to reserve them. */ - ret = lttng_fd_get(LTTNG_FD_APPS, 2); - if (ret < 0) { - ERR("Exhausted number of available FD upon stream create"); - free(ustream); - goto error_rcu_unlock; - } - - health_code_update(&health_thread_cmd); - - ret = ustctl_create_stream(app->sock, ua_chan->obj, - &ustream->obj); - if (ret < 0) { - /* Got all streams */ - lttng_fd_put(LTTNG_FD_APPS, 2); - free(ustream); - ret = LTTNG_ERR_UST_STREAM_FAIL; - break; - } - ustream->handle = ustream->obj->handle; - - health_code_update(&health_thread_cmd); - - /* Order is important */ - cds_list_add_tail(&ustream->list, &ua_chan->streams.head); - ret = snprintf(ustream->name, sizeof(ustream->name), "%s_%u", - ua_chan->name, ua_chan->streams.count); - ua_chan->streams.count++; - if (ret < 0) { - PERROR("asprintf UST create stream"); - /* - * XXX what should we do here with the - * stream ? - */ - continue; + goto error_unlock; } - DBG2("UST stream %d ready (handle: %d)", ua_chan->streams.count, - ustream->handle); - } - - health_code_update(&health_thread_cmd); - } - - switch (app->bits_per_long) { - case 64: - socket = consumer_find_socket(uatomic_read(&ust_consumerd64_fd), - usess->consumer); - if (socket == NULL) { - goto skip_setup; - } - break; - case 32: - socket = consumer_find_socket(uatomic_read(&ust_consumerd32_fd), - usess->consumer); - if (socket == NULL) { - goto skip_setup; } - break; - default: - ret = -EINVAL; - goto error_rcu_unlock; } - /* Setup UST consumer socket and send fds to it */ - ret = ust_consumer_send_session(ua_sess, usess->consumer, socket); + /* Create the metadata for the application. */ + ret = create_ust_app_metadata(ua_sess, app, usess->consumer); if (ret < 0) { - goto error_rcu_unlock; + goto error_unlock; } - health_code_update(&health_thread_cmd); + health_code_update(); skip_setup: /* This start the UST tracing */ ret = ustctl_start_session(app->sock, ua_sess->handle); if (ret < 0) { - ERR("Error starting tracing for app pid: %d (ret: %d)", app->pid, ret); - goto error_rcu_unlock; + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("Error starting tracing for app pid: %d (ret: %d)", + app->pid, ret); + } else { + DBG("UST app start session failed. Application is dead."); + } + goto error_unlock; } /* Indicate that the session has been started once */ ua_sess->started = 1; - health_code_update(&health_thread_cmd); + pthread_mutex_unlock(&ua_sess->lock); - /* Quiescent wait after starting trace */ - ustctl_wait_quiescent(app->sock); + health_code_update(); -end: + /* Quiescent wait after starting trace */ + ret = ustctl_wait_quiescent(app->sock); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app wait quiescent failed for app pid %d ret %d", + app->pid, ret); + } + +end: rcu_read_unlock(); - health_code_update(&health_thread_cmd); + health_code_update(); return 0; -error_rcu_unlock: +error_unlock: + pthread_mutex_unlock(&ua_sess->lock); rcu_read_unlock(); - health_code_update(&health_thread_cmd); + health_code_update(); return -1; } @@ -2388,14 +2904,16 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) rcu_read_lock(); if (!app->compatible) { - goto end; + goto end_no_session; } ua_sess = lookup_session_by_app(usess, app); if (ua_sess == NULL) { - goto end; + goto end_no_session; } + pthread_mutex_lock(&ua_sess->lock); + /* * 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 @@ -2406,52 +2924,69 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) goto error_rcu_unlock; } - health_code_update(&health_thread_cmd); + health_code_update(); /* This inhibits UST tracing */ ret = ustctl_stop_session(app->sock, ua_sess->handle); if (ret < 0) { - ERR("Error stopping tracing for app pid: %d (ret: %d)", app->pid, ret); + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("Error stopping tracing for app pid: %d (ret: %d)", + app->pid, ret); + } else { + DBG("UST app stop session failed. Application is dead."); + } goto error_rcu_unlock; } - health_code_update(&health_thread_cmd); + health_code_update(); /* Quiescent wait after stopping trace */ - ustctl_wait_quiescent(app->sock); + ret = ustctl_wait_quiescent(app->sock); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app wait quiescent failed for app pid %d ret %d", + app->pid, ret); + } - health_code_update(&health_thread_cmd); + health_code_update(); /* Flushing buffers */ cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, node.node) { - health_code_update(&health_thread_cmd); + health_code_update(); + assert(ua_chan->is_sent); ret = ustctl_sock_flush_buffer(app->sock, ua_chan->obj); if (ret < 0) { - ERR("UST app PID %d channel %s flush failed with ret %d", - app->pid, ua_chan->name, ret); + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app PID %d channel %s flush failed with ret %d", + app->pid, ua_chan->name, ret); + } else { + DBG3("UST app failed to flush %s. Application is dead.", + ua_chan->name); + /* No need to continue. */ + break; + } /* Continuing flushing all buffers */ continue; } } - health_code_update(&health_thread_cmd); + health_code_update(); - /* Flush all buffers before stopping */ - ret = ustctl_sock_flush_buffer(app->sock, ua_sess->metadata->obj); + ret = push_metadata(app, ua_sess); if (ret < 0) { - ERR("UST app PID %d metadata flush failed with ret %d", app->pid, - ret); + goto error_rcu_unlock; } -end: + pthread_mutex_unlock(&ua_sess->lock); +end_no_session: rcu_read_unlock(); - health_code_update(&health_thread_cmd); + health_code_update(); return 0; error_rcu_unlock: + pthread_mutex_unlock(&ua_sess->lock); rcu_read_unlock(); - health_code_update(&health_thread_cmd); + health_code_update(); return -1; } @@ -2460,11 +2995,10 @@ error_rcu_unlock: */ static int destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) { + int ret; struct ust_app_session *ua_sess; - struct lttng_ust_object_data obj; struct lttng_ht_iter iter; struct lttng_ht_node_ulong *node; - int ret; DBG("Destroy tracing for ust app pid %d", app->pid); @@ -2481,28 +3015,21 @@ static int destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) goto end; } ua_sess = caa_container_of(node, struct ust_app_session, node); - ret = lttng_ht_del(app->sessions, &iter); - if (ret) { - /* Already scheduled for teardown. */ - goto end; - } - obj.handle = ua_sess->handle; - obj.shm_fd = -1; - obj.wait_fd = -1; - obj.memory_map_size = 0; - health_code_update(&health_thread_cmd); - ustctl_release_object(app->sock, &obj); + health_code_update(); + destroy_app_session(app, ua_sess); - health_code_update(&health_thread_cmd); - delete_ust_app_session(app->sock, ua_sess); + health_code_update(); /* Quiescent wait after stopping trace */ - ustctl_wait_quiescent(app->sock); - + ret = ustctl_wait_quiescent(app->sock); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app wait quiescent failed for app pid %d ret %d", + app->pid, ret); + } end: rcu_read_unlock(); - health_code_update(&health_thread_cmd); + health_code_update(); return 0; } @@ -2548,6 +3075,7 @@ int ust_app_stop_trace_all(struct ltt_ust_session *usess) 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) { + ERR("UST app stop trace failed with ret %d", ret); /* Continue to next apps even on error */ continue; } @@ -2592,12 +3120,13 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) int ret = 0; struct lttng_ht_iter iter, uiter, iter_ctx; struct ust_app *app; - struct ust_app_session *ua_sess; + 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; assert(usess); + assert(sock >= 0); DBG2("UST app global update for app sock %d for session id %d", sock, usess->id); @@ -2606,7 +3135,11 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) app = find_app_by_sock(sock); if (app == NULL) { - ERR("Failed to update app sock %d", sock); + /* + * 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; } @@ -2614,31 +3147,37 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) goto error; } - ua_sess = create_ust_app_session(usess, app); - if (ua_sess == NULL || ua_sess == (void *) -1UL) { - /* Tracer is gone for this session and has been freed */ + ret = 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); /* - * We can iterate safely here over all UST app session sicne the create ust + * 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. */ cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, node.node) { - ret = create_ust_channel(app, ua_sess, ua_chan); + ret = create_ust_channel(app, ua_sess, ua_chan, usess->consumer); if (ret < 0) { - /* FIXME: Should we quit here or continue... */ - continue; + /* + * Stop everything. On error, the application failed, no more file + * descriptor are available or ENOMEM so stopping here is the only + * thing we can do for now. + */ + goto error_unlock; } cds_lfht_for_each_entry(ua_chan->ctx->ht, &iter_ctx.iter, ua_ctx, node.node) { ret = create_ust_channel_context(ua_chan, ua_ctx, app); if (ret < 0) { - /* FIXME: Should we quit here or continue... */ - continue; + goto error_unlock; } } @@ -2648,18 +3187,13 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) node.node) { ret = create_ust_event(app, ua_sess, ua_chan, ua_event); if (ret < 0) { - /* FIXME: Should we quit here or continue... */ - continue; - } - - ret = set_ust_event_filter(ua_event, app); - if (ret < 0) { - /* FIXME: Should we quit here or continue... */ - continue; + goto error_unlock; } } } + pthread_mutex_unlock(&ua_sess->lock); + if (usess->start_trace) { ret = ust_app_start_trace(usess, app); if (ret < 0) { @@ -2669,7 +3203,16 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock) DBG2("UST trace started for app pid %d", app->pid); } + /* Everything went well at this point. */ + rcu_read_unlock(); + return; + +error_unlock: + pthread_mutex_unlock(&ua_sess->lock); error: + if (ua_sess) { + destroy_app_session(app, ua_sess); + } rcu_read_unlock(); return; } @@ -2702,19 +3245,21 @@ int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess, continue; } + pthread_mutex_lock(&ua_sess->lock); /* 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 (ua_chan_node == NULL) { - continue; + goto next_app; } 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); if (ret < 0) { - continue; + goto next_app; } + next_app: + pthread_mutex_unlock(&ua_sess->lock); } rcu_read_unlock(); @@ -2743,20 +3288,22 @@ int ust_app_enable_event_pid(struct ltt_ust_session *usess, if (app == NULL) { ERR("UST app enable event per PID %d not found", pid); ret = -1; - goto error; + goto end; } if (!app->compatible) { ret = 0; - goto error; + goto end; } ua_sess = lookup_session_by_app(usess, app); if (!ua_sess) { /* The application has problem or is probably dead. */ - goto error; + ret = 0; + goto end; } + pthread_mutex_lock(&ua_sess->lock); /* 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); @@ -2770,16 +3317,18 @@ int ust_app_enable_event_pid(struct ltt_ust_session *usess, if (ua_event == NULL) { ret = create_ust_app_event(ua_sess, ua_chan, uevent, app); if (ret < 0) { - goto error; + goto end_unlock; } } else { ret = enable_ust_app_event(ua_sess, ua_event, app); if (ret < 0) { - goto error; + goto end_unlock; } } -error: +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); +end: rcu_read_unlock(); return ret; } @@ -2847,49 +3396,6 @@ error: return ret; } -/* - * Validate version of UST apps and set the compatible bit. - */ -int ust_app_validate_version(int sock) -{ - int ret; - struct ust_app *app; - - rcu_read_lock(); - - app = find_app_by_sock(sock); - assert(app); - - health_code_update(&health_thread_cmd); - - ret = ustctl_tracer_version(sock, &app->version); - if (ret < 0) { - goto error; - } - - /* Validate version */ - if (app->version.major != UST_APP_MAJOR_VERSION) { - goto error; - } - - DBG2("UST app PID %d is compatible with internal major version %d " - "(supporting == %d)", app->pid, app->version.major, - UST_APP_MAJOR_VERSION); - app->compatible = 1; - rcu_read_unlock(); - health_code_update(&health_thread_cmd); - return 0; - -error: - DBG2("UST app PID %d is not compatible with internal major version %d " - "(supporting == %d)", app->pid, app->version.major, - UST_APP_MAJOR_VERSION); - app->compatible = 0; - rcu_read_unlock(); - health_code_update(&health_thread_cmd); - return -1; -} - /* * Calibrate registered applications. */ @@ -2910,7 +3416,7 @@ int ust_app_calibrate_glb(struct lttng_ust_calibrate *calibrate) continue; } - health_code_update(&health_thread_cmd); + health_code_update(); ret = ustctl_calibrate(app->sock, calibrate); if (ret < 0) { @@ -2920,7 +3426,6 @@ int ust_app_calibrate_glb(struct lttng_ust_calibrate *calibrate) ret = 0; break; default: - /* TODO: Report error to user */ DBG2("Calibrate app PID %d returned with error %d", app->pid, ret); break; @@ -2932,7 +3437,413 @@ int ust_app_calibrate_glb(struct lttng_ust_calibrate *calibrate) rcu_read_unlock(); - health_code_update(&health_thread_cmd); + health_code_update(); return ret; } + +/* + * Receive registration and populate the given msg structure. + * + * On success return 0 else a negative value returned by the ustctl call. + */ +int ust_app_recv_registration(int sock, struct ust_register_msg *msg) +{ + int ret; + uint32_t pid, ppid, uid, gid; + + assert(msg); + + ret = ustctl_recv_reg_msg(sock, &msg->type, &msg->major, &msg->minor, + &pid, &ppid, &uid, &gid, + &msg->bits_per_long, + &msg->uint8_t_alignment, + &msg->uint16_t_alignment, + &msg->uint32_t_alignment, + &msg->uint64_t_alignment, + &msg->long_alignment, + &msg->byte_order, + msg->name); + if (ret < 0) { + switch (-ret) { + case EPIPE: + case ECONNRESET: + case LTTNG_UST_ERR_EXITING: + DBG3("UST app recv reg message failed. Application died"); + break; + case LTTNG_UST_ERR_UNSUP_MAJOR: + ERR("UST app recv reg unsupported version %d.%d. Supporting %d.%d", + msg->major, msg->minor, LTTNG_UST_ABI_MAJOR_VERSION, + LTTNG_UST_ABI_MINOR_VERSION); + break; + default: + ERR("UST app recv reg message failed with ret %d", ret); + break; + } + goto error; + } + msg->pid = (pid_t) pid; + msg->ppid = (pid_t) ppid; + msg->uid = (uid_t) uid; + msg->gid = (gid_t) gid; + +error: + return ret; +} + +/* + * Return a ust app channel object using the application object and the channel + * object descriptor has a key. If not found, NULL is returned. A RCU read side + * lock MUST be acquired before calling this function. + */ +static struct ust_app_channel *find_channel_by_objd(struct ust_app *app, + int objd) +{ + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter iter; + struct ust_app_channel *ua_chan = NULL; + + assert(app); + + lttng_ht_lookup(app->ust_objd, (void *)((unsigned long) objd), &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node == NULL) { + DBG2("UST app channel find by objd %d not found", objd); + goto error; + } + + ua_chan = caa_container_of(node, struct ust_app_channel, ust_objd_node); + +error: + return ua_chan; +} + +/* + * Reply to a register channel notification from an application on the notify + * socket. The channel metadata is also created. + * + * The session UST registry lock is acquired in this function. + * + * On success 0 is returned else a negative value. + */ +static int reply_ust_register_channel(int sock, int sobjd, int cobjd, + size_t nr_fields, struct ustctl_field *fields) +{ + int ret, ret_code = 0; + uint32_t chan_id, reg_count; + enum ustctl_channel_header type; + struct ust_app *app; + struct ust_app_channel *ua_chan; + struct ust_app_session *ua_sess; + + rcu_read_lock(); + + /* 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", + sock); + ret = 0; + goto error_rcu_unlock; + } + + /* Lookup channel by UST object descriptor. Should always be found. */ + ua_chan = find_channel_by_objd(app, cobjd); + assert(ua_chan); + assert(ua_chan->session); + ua_sess = ua_chan->session; + assert(ua_sess); + + pthread_mutex_lock(&ua_sess->registry.lock); + + if (ust_registry_is_max_id(ua_chan->session->registry.used_channel_id)) { + ret_code = -1; + chan_id = -1U; + type = -1; + goto reply; + } + + /* Don't assign ID to metadata. */ + if (ua_chan->attr.type == LTTNG_UST_CHAN_METADATA) { + chan_id = -1U; + } else { + chan_id = ust_registry_get_next_chan_id(&ua_chan->session->registry); + } + + reg_count = ust_registry_get_event_count(&ua_chan->registry); + if (reg_count < 31) { + type = USTCTL_CHANNEL_HEADER_COMPACT; + } else { + type = USTCTL_CHANNEL_HEADER_LARGE; + } + + ua_chan->registry.nr_ctx_fields = nr_fields; + ua_chan->registry.ctx_fields = fields; + ua_chan->registry.chan_id = chan_id; + ua_chan->registry.header_type = type; + + /* Append to metadata */ + if (!ret_code) { + ret_code = ust_metadata_channel_statedump(&ua_chan->session->registry, + &ua_chan->registry); + if (ret_code) { + ERR("Error appending channel metadata (errno = %d)", ret_code); + goto reply; + } + } + +reply: + DBG3("UST app replying to register channel with id %u, type: %d, ret: %d", + chan_id, type, ret_code); + + ret = ustctl_reply_register_channel(sock, chan_id, type, ret_code); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app reply channel failed with ret %d", ret); + } else { + DBG3("UST app reply channel failed. Application died"); + } + goto error; + } + +error: + pthread_mutex_unlock(&ua_sess->registry.lock); +error_rcu_unlock: + rcu_read_unlock(); + return ret; +} + +/* + * Add event to the UST channel registry. When the event is added to the + * registry, the metadata is also created. Once done, this replies to the + * application with the appropriate error code. + * + * The session UST registry lock is acquired in the function. + * + * On success 0 is returned else a negative value. + */ +static int add_event_ust_registry(int sock, int sobjd, int cobjd, char *name, + char *sig, size_t nr_fields, struct ustctl_field *fields, int loglevel, + char *model_emf_uri) +{ + int ret, ret_code; + uint32_t event_id = 0; + struct ust_app *app; + struct ust_app_channel *ua_chan; + struct ust_app_session *ua_sess; + + rcu_read_lock(); + + /* 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", + sock); + ret = 0; + goto error_rcu_unlock; + } + + /* Lookup channel by UST object descriptor. Should always be found. */ + ua_chan = find_channel_by_objd(app, cobjd); + assert(ua_chan); + assert(ua_chan->session); + ua_sess = ua_chan->session; + + pthread_mutex_lock(&ua_sess->registry.lock); + + ret_code = ust_registry_create_event(&ua_sess->registry, + &ua_chan->registry, sobjd, cobjd, name, sig, nr_fields, fields, + loglevel, model_emf_uri, &event_id); + + /* + * The return value is returned to ustctl so in case of an error, the + * application can be notified. In case of an error, it's important not to + * return a negative error or else the application will get closed. + */ + ret = ustctl_reply_register_event(sock, event_id, ret_code); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app reply event failed with ret %d", ret); + } else { + DBG3("UST app reply event failed. Application died"); + } + /* + * No need to wipe the create event since the application socket will + * get close on error hence cleaning up everything by itself. + */ + goto error; + } + + DBG3("UST registry event %s has been added successfully", name); + +error: + pthread_mutex_unlock(&ua_sess->registry.lock); +error_rcu_unlock: + rcu_read_unlock(); + return ret; +} + +/* + * Handle application notification through the given notify socket. + * + * Return 0 on success or else a negative value. + */ +int ust_app_recv_notify(int sock) +{ + int ret; + enum ustctl_notify_cmd cmd; + + DBG3("UST app receiving notify from sock %d", sock); + + ret = ustctl_recv_notify(sock, &cmd); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app recv notify failed with ret %d", ret); + } else { + DBG3("UST app recv notify failed. Application died"); + } + goto error; + } + + switch (cmd) { + case USTCTL_NOTIFY_CMD_EVENT: + { + int sobjd, cobjd, loglevel; + char name[LTTNG_UST_SYM_NAME_LEN], *sig, *model_emf_uri; + size_t nr_fields; + struct ustctl_field *fields; + + DBG2("UST app ustctl register event received"); + + ret = ustctl_recv_register_event(sock, &sobjd, &cobjd, name, &loglevel, + &sig, &nr_fields, &fields, &model_emf_uri); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app recv event failed with ret %d", ret); + } else { + DBG3("UST app recv event failed. Application died"); + } + goto error; + } + + /* Add event to the UST registry coming from the notify socket. */ + ret = add_event_ust_registry(sock, sobjd, cobjd, name, sig, nr_fields, + fields, loglevel, model_emf_uri); + if (ret < 0) { + goto error; + } + + break; + } + case USTCTL_NOTIFY_CMD_CHANNEL: + { + int sobjd, cobjd; + size_t nr_fields; + struct ustctl_field *fields; + + DBG2("UST app ustctl register channel received"); + + ret = ustctl_recv_register_channel(sock, &sobjd, &cobjd, &nr_fields, + &fields); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app recv channel failed with ret %d", ret); + } else { + DBG3("UST app recv channel failed. Application died"); + } + goto error; + } + + ret = reply_ust_register_channel(sock, sobjd, cobjd, nr_fields, + fields); + if (ret < 0) { + goto error; + } + + break; + } + default: + /* Should NEVER happen. */ + assert(0); + } + +error: + return ret; +} + +/* + * Once the notify socket hangs up, this is called. First, it tries to find the + * corresponding application. On failure, the call_rcu to close the socket is + * executed. If an application is found, it tries to delete it from the notify + * socket hash table. Whathever the result, it proceeds to the call_rcu. + * + * Note that an object needs to be allocated here so on ENOMEM failure, the + * call RCU is not done but the rest of the cleanup is. + */ +void ust_app_notify_sock_unregister(int sock) +{ + int err_enomem = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + struct ust_app_notify_sock_obj *obj; + + assert(sock >= 0); + + rcu_read_lock(); + + obj = zmalloc(sizeof(*obj)); + if (!obj) { + /* + * An ENOMEM is kind of uncool. If this strikes we continue the + * procedure but the call_rcu will not be called. In this case, we + * accept the fd leak rather than possibly creating an unsynchronized + * state between threads. + * + * TODO: The notify object should be created once the notify socket is + * registered and stored independantely from the ust app object. The + * tricky part is to synchronize the teardown of the application and + * this notify object. Let's keep that in mind so we can avoid this + * kind of shenanigans with ENOMEM in the teardown path. + */ + err_enomem = 1; + } else { + obj->fd = sock; + } + + DBG("UST app notify socket unregister %d", sock); + + /* + * Lookup application by notify socket. If this fails, this means that the + * hash table delete has already been done by the application + * unregistration process so we can safely close the notify socket in a + * call RCU. + */ + app = find_app_by_notify_sock(sock); + if (!app) { + goto close_socket; + } + + iter.iter.node = &app->notify_sock_n.node; + + /* + * Whatever happens here either we fail or succeed, in both cases we have + * to close the socket after a grace period to continue to the call RCU + * here. If the deletion is successful, the application is not visible + * anymore by other threads and is it fails it means that it was already + * deleted from the hash table so either way we just have to close the + * socket. + */ + (void) lttng_ht_del(ust_app_ht_by_notify_sock, &iter); + +close_socket: + rcu_read_unlock(); + + /* + * Close socket after a grace period to avoid for the socket to be reused + * before the application object is freed creating potential race between + * threads trying to add unique in the global hash table. + */ + if (!err_enomem) { + call_rcu(&obj->head, close_notify_sock_rcu); + } +}