X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fust-app.c;h=5c4dc5ae85837ca24bbcef8cbb950fe0269e8376;hp=72887633c36ce31b662c92c9240a164921bc6973;hb=2b2694892c10cdb632afac2b2f1aabf7cb9673d9;hpb=d88aee689d5bd0067f362a323cb69c37717df59f diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c index 72887633c..52d1da787 100644 --- a/src/bin/lttng-sessiond/ust-app.c +++ b/src/bin/lttng-sessiond/ust-app.c @@ -1,5 +1,6 @@ /* * Copyright (C) 2011 - David Goulet + * Copyright (C) 2016 - Jérémie Galarneau * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License, version 2 only, @@ -15,8 +16,9 @@ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ -#define _GNU_SOURCE +#define _LGPL_SOURCE #include +#include #include #include #include @@ -26,60 +28,72 @@ #include #include #include +#include #include #include +#include "buffer-registry.h" #include "fd-limit.h" -#include "health.h" +#include "health-sessiond.h" #include "ust-app.h" #include "ust-consumer.h" #include "ust-ctl.h" +#include "utils.h" +#include "session.h" +#include "lttng-sessiond.h" +#include "notification-thread-commands.h" +#include "rotate.h" -/* Next available channel key. */ -static unsigned long next_channel_key; +static +int ust_app_flush_app_session(struct ust_app *app, struct ust_app_session *ua_sess); + +/* Next available channel key. Access under next_channel_key_lock. */ +static uint64_t _next_channel_key; +static pthread_mutex_t next_channel_key_lock = PTHREAD_MUTEX_INITIALIZER; + +/* Next available session ID. Access under next_session_id_lock. */ +static uint64_t _next_session_id; +static pthread_mutex_t next_session_id_lock = PTHREAD_MUTEX_INITIALIZER; /* - * Return the atomically incremented value of next_channel_key. + * Return the incremented value of next_channel_key. */ -static inline unsigned long get_next_channel_key(void) +static uint64_t get_next_channel_key(void) { - return uatomic_add_return(&next_channel_key, 1); + uint64_t ret; + + pthread_mutex_lock(&next_channel_key_lock); + ret = ++_next_channel_key; + pthread_mutex_unlock(&next_channel_key_lock); + return ret; } /* - * 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. + * Return the atomically incremented value of next_session_id. */ -static struct consumer_socket *find_consumer_socket_by_bitness(int bits, - struct consumer_output *consumer) +static uint64_t get_next_session_id(void) { - 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; - } + uint64_t ret; - socket = consumer_find_socket(consumer_fd, consumer); - if (!socket) { - ERR("Consumer socket fd %d not found in consumer obj %p", - consumer_fd, consumer); - } + pthread_mutex_lock(&next_session_id_lock); + ret = ++_next_session_id; + pthread_mutex_unlock(&next_session_id_lock); + return ret; +} -end: - return socket; +static void copy_channel_attr_to_ustctl( + struct ustctl_consumer_channel_attr *attr, + struct lttng_ust_channel_attr *uattr) +{ + /* Copy event attributes since the layout is different. */ + attr->subbuf_size = uattr->subbuf_size; + attr->num_subbuf = uattr->num_subbuf; + attr->overwrite = uattr->overwrite; + attr->switch_timer_interval = uattr->switch_timer_interval; + attr->read_timer_interval = uattr->read_timer_interval; + attr->output = uattr->output; + attr->blocking_timeout = uattr->u.s.blocking_timeout; } /* @@ -92,14 +106,16 @@ static int ht_match_ust_app_event(struct cds_lfht_node *node, const void *_key) { struct ust_app_event *event; const struct ust_app_ht_key *key; + int ev_loglevel_value; assert(node); assert(_key); event = caa_container_of(node, struct ust_app_event, node.node); key = _key; + ev_loglevel_value = event->attr.loglevel; - /* Match the 3 elements of the key: name, filter and loglevel. */ + /* Match the 4 elements of the key: name, filter, loglevel, exclusions */ /* Event name */ if (strncmp(event->attr.name, key->name, sizeof(event->attr.name)) != 0) { @@ -107,9 +123,10 @@ static int ht_match_ust_app_event(struct cds_lfht_node *node, const void *_key) } /* Event loglevel. */ - if (event->attr.loglevel != key->loglevel) { + if (ev_loglevel_value != key->loglevel_type) { if (event->attr.loglevel_type == LTTNG_UST_LOGLEVEL_ALL - && key->loglevel == 0 && event->attr.loglevel == -1) { + && key->loglevel_type == 0 && + ev_loglevel_value == -1) { /* * Match is accepted. This is because on event creation, the * loglevel is set to -1 if the event loglevel type is ALL so 0 and @@ -135,6 +152,21 @@ static int ht_match_ust_app_event(struct cds_lfht_node *node, const void *_key) } } + /* One of the exclusions is NULL, fail. */ + if ((key->exclusion && !event->exclusion) || (!key->exclusion && event->exclusion)) { + goto no_match; + } + + if (key->exclusion && event->exclusion) { + /* Both exclusions exists, check count followed by the names. */ + if (event->exclusion->count != key->exclusion->count || + memcmp(event->exclusion->names, key->exclusion->names, + event->exclusion->count * LTTNG_UST_SYM_NAME_LEN) != 0) { + goto no_match; + } + } + + /* Match. */ return 1; @@ -160,7 +192,8 @@ static void add_unique_ust_app_event(struct ust_app_channel *ua_chan, ht = ua_chan->events; key.name = event->attr.name; key.filter = event->filter; - key.loglevel = event->attr.loglevel; + key.loglevel_type = event->attr.loglevel; + key.exclusion = event->exclusion; node_ptr = cds_lfht_add_unique(ht->ht, ht->hash_fct(event->node.key, lttng_ht_seed), @@ -190,19 +223,64 @@ static void close_notify_sock_rcu(struct rcu_head *head) free(obj); } +/* + * Return the session registry according to the buffer type of the given + * session. + * + * A registry per UID object MUST exists before calling this function or else + * it assert() if not found. RCU read side lock must be acquired. + */ +static struct ust_registry_session *get_session_registry( + struct ust_app_session *ua_sess) +{ + struct ust_registry_session *registry = NULL; + + assert(ua_sess); + + switch (ua_sess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + { + struct buffer_reg_pid *reg_pid = buffer_reg_pid_find(ua_sess->id); + if (!reg_pid) { + goto error; + } + registry = reg_pid->registry->reg.ust; + break; + } + case LTTNG_BUFFER_PER_UID: + { + struct buffer_reg_uid *reg_uid = buffer_reg_uid_find( + ua_sess->tracing_id, ua_sess->bits_per_long, ua_sess->uid); + if (!reg_uid) { + goto error; + } + registry = reg_uid->registry->reg.ust; + break; + } + default: + assert(0); + }; + +error: + return registry; +} + /* * Delete ust context safely. RCU read lock must be held before calling * this function. */ static -void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx) +void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx, + struct ust_app *app) { int ret; assert(ua_ctx); if (ua_ctx->obj) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_object(sock, ua_ctx->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release ctx obj handle %d failed with ret %d", sock, ua_ctx->obj->handle, ret); @@ -217,16 +295,20 @@ void delete_ust_app_ctx(int sock, struct ust_app_ctx *ua_ctx) * this function. */ static -void delete_ust_app_event(int sock, struct ust_app_event *ua_event) +void delete_ust_app_event(int sock, struct ust_app_event *ua_event, + struct ust_app *app) { int ret; assert(ua_event); free(ua_event->filter); - + if (ua_event->exclusion != NULL) + free(ua_event->exclusion); if (ua_event->obj != NULL) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_object(sock, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release event obj failed with ret %d", sock, ret); @@ -237,18 +319,21 @@ void delete_ust_app_event(int sock, struct ust_app_event *ua_event) } /* - * Delete ust app stream safely. RCU read lock must be held before calling - * this function. + * Release ust data object of the given stream. + * + * Return 0 on success or else a negative value. */ -static -void delete_ust_app_stream(int sock, struct ust_app_stream *stream) +static int release_ust_app_stream(int sock, struct ust_app_stream *stream, + struct ust_app *app) { - int ret; + int ret = 0; assert(stream); if (stream->obj) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_object(sock, stream->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release stream obj failed with ret %d", sock, ret); @@ -256,12 +341,109 @@ void delete_ust_app_stream(int sock, struct ust_app_stream *stream) lttng_fd_put(LTTNG_FD_APPS, 2); free(stream->obj); } + + return ret; +} + +/* + * Delete ust app stream safely. RCU read lock must be held before calling + * this function. + */ +static +void delete_ust_app_stream(int sock, struct ust_app_stream *stream, + struct ust_app *app) +{ + assert(stream); + + (void) release_ust_app_stream(sock, stream, app); free(stream); } +/* + * We need to execute ht_destroy outside of RCU read-side critical + * section and outside of call_rcu thread, so we postpone its execution + * using ht_cleanup_push. It is simpler than to change the semantic of + * the many callers of delete_ust_app_session(). + */ +static +void delete_ust_app_channel_rcu(struct rcu_head *head) +{ + struct ust_app_channel *ua_chan = + caa_container_of(head, struct ust_app_channel, rcu_head); + + ht_cleanup_push(ua_chan->ctx); + ht_cleanup_push(ua_chan->events); + free(ua_chan); +} + +/* + * Extract the lost packet or discarded events counter when the channel is + * being deleted and store the value in the parent channel so we can + * access it from lttng list and at stop/destroy. + * + * The session list lock must be held by the caller. + */ +static +void save_per_pid_lost_discarded_counters(struct ust_app_channel *ua_chan) +{ + uint64_t discarded = 0, lost = 0; + struct ltt_session *session; + struct ltt_ust_channel *uchan; + + if (ua_chan->attr.type != LTTNG_UST_CHAN_PER_CPU) { + return; + } + + rcu_read_lock(); + session = session_find_by_id(ua_chan->session->tracing_id); + if (!session || !session->ust_session) { + /* + * Not finding the session is not an error because there are + * multiple ways the channels can be torn down. + * + * 1) The session daemon can initiate the destruction of the + * ust app session after receiving a destroy command or + * during its shutdown/teardown. + * 2) The application, since we are in per-pid tracing, is + * unregistering and tearing down its ust app session. + * + * Both paths are protected by the session list lock which + * ensures that the accounting of lost packets and discarded + * events is done exactly once. The session is then unpublished + * from the session list, resulting in this condition. + */ + goto end; + } + + if (ua_chan->attr.overwrite) { + consumer_get_lost_packets(ua_chan->session->tracing_id, + ua_chan->key, session->ust_session->consumer, + &lost); + } else { + consumer_get_discarded_events(ua_chan->session->tracing_id, + ua_chan->key, session->ust_session->consumer, + &discarded); + } + uchan = trace_ust_find_channel_by_name( + session->ust_session->domain_global.channels, + ua_chan->name); + if (!uchan) { + ERR("Missing UST channel to store discarded counters"); + goto end; + } + + uchan->per_pid_closed_app_discarded += discarded; + uchan->per_pid_closed_app_lost += lost; + +end: + rcu_read_unlock(); +} + /* * Delete ust app channel safely. RCU read lock must be held before calling * this function. + * + * The session list lock must be held by the caller. */ static void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, @@ -272,6 +454,7 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, struct ust_app_event *ua_event; struct ust_app_ctx *ua_ctx; struct ust_app_stream *stream, *stmp; + struct ust_registry_session *registry; assert(ua_chan); @@ -280,187 +463,340 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan, /* Wipe stream */ cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) { cds_list_del(&stream->list); - delete_ust_app_stream(sock, stream); + delete_ust_app_stream(sock, stream, app); } /* Wipe context */ cds_lfht_for_each_entry(ua_chan->ctx->ht, &iter.iter, ua_ctx, node.node) { + cds_list_del(&ua_ctx->list); ret = lttng_ht_del(ua_chan->ctx, &iter); assert(!ret); - delete_ust_app_ctx(sock, ua_ctx); + delete_ust_app_ctx(sock, ua_ctx, app); } - lttng_ht_destroy(ua_chan->ctx); /* Wipe events */ cds_lfht_for_each_entry(ua_chan->events->ht, &iter.iter, ua_event, node.node) { ret = lttng_ht_del(ua_chan->events, &iter); assert(!ret); - delete_ust_app_event(sock, ua_event); + delete_ust_app_event(sock, ua_event, app); } - lttng_ht_destroy(ua_chan->events); - /* Wipe and free registry. */ - ust_registry_channel_destroy(&ua_chan->session->registry, &ua_chan->registry); + if (ua_chan->session->buffer_type == LTTNG_BUFFER_PER_PID) { + /* Wipe and free registry from session registry. */ + registry = get_session_registry(ua_chan->session); + if (registry) { + ust_registry_channel_del_free(registry, ua_chan->key, + true); + } + save_per_pid_lost_discarded_counters(ua_chan); + } if (ua_chan->obj != NULL) { /* Remove channel from application UST object descriptor. */ iter.iter.node = &ua_chan->ust_objd_node.node; - lttng_ht_del(app->ust_objd, &iter); + ret = lttng_ht_del(app->ust_objd, &iter); + assert(!ret); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_object(sock, ua_chan->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release channel obj failed with ret %d", sock, ret); } - lttng_fd_put(LTTNG_FD_APPS, 2); + lttng_fd_put(LTTNG_FD_APPS, 1); free(ua_chan->obj); } - free(ua_chan); + call_rcu(&ua_chan->rcu_head, delete_ust_app_channel_rcu); +} + +int ust_app_register_done(struct ust_app *app) +{ + int ret; + + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_register_done(app->sock); + pthread_mutex_unlock(&app->sock_lock); + return ret; +} + +int ust_app_release_object(struct ust_app *app, struct lttng_ust_object_data *data) +{ + int ret, sock; + + if (app) { + pthread_mutex_lock(&app->sock_lock); + sock = app->sock; + } else { + sock = -1; + } + ret = ustctl_release_object(sock, data); + if (app) { + pthread_mutex_unlock(&app->sock_lock); + } + return ret; } /* - * For a given application and session, push metadata to consumer. The session - * lock MUST be acquired here before calling this. + * Push metadata to consumer socket. * - * Return 0 on success else a negative error. + * 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). */ -static int push_metadata(struct ust_app *app, struct ust_app_session *ua_sess) +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; - 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; - } + size_t len, offset, new_metadata_len_sent; + ssize_t ret_val; + uint64_t metadata_key, metadata_version; - rcu_read_lock(); + assert(registry); + assert(socket); - /* 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; - } + metadata_key = registry->metadata_key; /* - * 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. + * Means that no metadata was assigned to the session. This can + * happens if no start has been done previously. */ - pthread_mutex_lock(socket->lock); - pthread_mutex_lock(&ua_sess->registry.lock); + if (!metadata_key) { + return 0; + } - offset = ua_sess->registry.metadata_len_sent; - len = ua_sess->registry.metadata_len - ua_sess->registry.metadata_len_sent; + offset = registry->metadata_len_sent; + len = registry->metadata_len - registry->metadata_len_sent; + new_metadata_len_sent = registry->metadata_len; + metadata_version = registry->metadata_version; if (len == 0) { - DBG3("No metadata to push for session id %d", ua_sess->id); - ret = 0; - goto error_reg_unlock; + DBG3("No metadata to push for metadata key %" PRIu64, + registry->metadata_key); + ret_val = len; + if (send_zero_data) { + DBG("No metadata to push"); + goto push_data; + } + goto end; } - 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; + ret_val = -ENOMEM; + goto error; } - /* Copy what we haven't send out. */ - memcpy(metadata_str, ua_sess->registry.metadata + offset, len); + /* Copy what we haven't sent out. */ + memcpy(metadata_str, registry->metadata + offset, len); - pthread_mutex_unlock(&ua_sess->registry.lock); - - ret = ust_consumer_push_metadata(socket, ua_sess, metadata_str, len, - offset); +push_data: + pthread_mutex_unlock(®istry->lock); + /* + * We need to unlock the registry while we push metadata to + * break a circular dependency between the consumerd metadata + * lock and the sessiond registry lock. Indeed, pushing metadata + * to the consumerd awaits that it gets pushed all the way to + * relayd, but doing so requires grabbing the metadata lock. If + * a concurrent metadata request is being performed by + * consumerd, this can try to grab the registry lock on the + * sessiond while holding the metadata lock on the consumer + * daemon. Those push and pull schemes are performed on two + * different bidirectionnal communication sockets. + */ + ret = consumer_push_metadata(socket, metadata_key, + metadata_str, len, offset, metadata_version); + pthread_mutex_lock(®istry->lock); if (ret < 0) { - pthread_mutex_unlock(socket->lock); - goto error_rcu_unlock; + /* + * There is an acceptable race here between the registry + * metadata key assignment and the creation on the + * consumer. The session daemon can concurrently push + * metadata for this registry while being created on the + * consumer since the metadata key of the registry is + * assigned *before* it is setup to avoid the consumer + * to ask for metadata that could possibly be not found + * in the session daemon. + * + * The metadata will get pushed either by the session + * being stopped or the consumer requesting metadata if + * that race is triggered. + */ + if (ret == -LTTCOMM_CONSUMERD_CHANNEL_FAIL) { + ret = 0; + } else { + ERR("Error pushing metadata to consumer"); + } + ret_val = ret; + goto error_push; + } else { + /* + * Metadata may have been concurrently pushed, since + * we're not holding the registry lock while pushing to + * consumer. This is handled by the fact that we send + * the metadata content, size, and the offset at which + * that metadata belongs. This may arrive out of order + * on the consumer side, and the consumer is able to + * deal with overlapping fragments. The consumer + * supports overlapping fragments, which must be + * contiguous starting from offset 0. We keep the + * largest metadata_len_sent value of the concurrent + * send. + */ + registry->metadata_len_sent = + max_t(size_t, registry->metadata_len_sent, + new_metadata_len_sent); } + free(metadata_str); + return len; - /* 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(); +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. + */ + registry->metadata_closed = 1; + } +error_push: free(metadata_str); + return ret_val; +} + +/* + * For a given application and session, push metadata to consumer. + * Either sock or consumer is required : if sock is NULL, the default + * socket to send the metadata is retrieved from consumer, if sock + * is not NULL we use it to send the metadata. + * RCU read-side lock must be held while calling this function, + * therefore ensuring existance of registry. It also ensures existance + * of socket throughout this function. + * + * Return 0 on success else a negative error. + * Returning a -EPIPE return value means we could not send the metadata, + * but it can be caused by recoverable errors (e.g. the application has + * terminated concurrently). + */ +static int push_metadata(struct ust_registry_session *registry, + struct consumer_output *consumer) +{ + int ret_val; + ssize_t ret; + struct consumer_socket *socket; + + assert(registry); + assert(consumer); + + pthread_mutex_lock(®istry->lock); + if (registry->metadata_closed) { + 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); + if (!socket) { + ret_val = -1; + goto error; + } + + ret = ust_app_push_metadata(registry, socket, 0); + if (ret < 0) { + ret_val = ret; + goto error; + } + pthread_mutex_unlock(®istry->lock); 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; +error: + pthread_mutex_unlock(®istry->lock); + return ret_val; } /* * Send to the consumer a close metadata command for the given session. Once * done, the metadata channel is deleted and the session metadata pointer is - * nullified. The session lock MUST be acquired here unless the application is + * nullified. The session lock MUST be held unless the application is * in the destroy path. * * Return 0 on success else a negative value. */ -static int close_metadata(struct ust_app *app, struct ust_app_session *ua_sess) +static int close_metadata(struct ust_registry_session *registry, + struct consumer_output *consumer) { int ret; struct consumer_socket *socket; - assert(app); - assert(ua_sess); + assert(registry); + assert(consumer); + + rcu_read_lock(); + + pthread_mutex_lock(®istry->lock); - /* Ignore if no metadata. Valid since it can be called on unregister. */ - if (!ua_sess->metadata) { + if (!registry->metadata_key || registry->metadata_closed) { ret = 0; - goto error; + 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); + socket = consumer_find_socket_by_bitness(registry->bits_per_long, + consumer); if (!socket) { ret = -1; - goto error_rcu_unlock; + goto error; } - ret = ust_consumer_close_metadata(socket, ua_sess->metadata); + ret = consumer_close_metadata(socket, registry->metadata_key); if (ret < 0) { - goto error_rcu_unlock; + goto error; } -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: + /* + * Metadata closed. Even on error this means that the consumer is not + * responding or not found so either way a second close should NOT be emit + * for this registry. + */ + registry->metadata_closed = 1; +end: + pthread_mutex_unlock(®istry->lock); + rcu_read_unlock(); return ret; } +/* + * We need to execute ht_destroy outside of RCU read-side critical + * section and outside of call_rcu thread, so we postpone its execution + * using ht_cleanup_push. It is simpler than to change the semantic of + * the many callers of delete_ust_app_session(). + */ +static +void delete_ust_app_session_rcu(struct rcu_head *head) +{ + struct ust_app_session *ua_sess = + caa_container_of(head, struct ust_app_session, rcu_head); + + ht_cleanup_push(ua_sess->channels); + free(ua_sess); +} + /* * Delete ust app session safely. RCU read lock must be held before calling * this function. + * + * The session list lock must be held by the caller. */ static void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, @@ -469,15 +805,31 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, int ret; struct lttng_ht_iter iter; struct ust_app_channel *ua_chan; + struct ust_registry_session *registry; assert(ua_sess); - if (ua_sess->metadata) { + pthread_mutex_lock(&ua_sess->lock); + + assert(!ua_sess->deleted); + ua_sess->deleted = true; + + registry = get_session_registry(ua_sess); + /* Registry can be null on error path during initialization. */ + if (registry) { /* Push metadata for application before freeing the application. */ - (void) push_metadata(app, ua_sess); + (void) push_metadata(registry, ua_sess->consumer); - /* And ask to close it for this session. */ - (void) close_metadata(app, ua_sess); + /* + * Don't ask to close metadata for global per UID buffers. Close + * metadata only on destroy trace session in this case. Also, the + * previous push metadata could have flag the metadata registry to + * close so don't send a close command if closed. + */ + if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID) { + /* And ask to close it for this session registry. */ + (void) close_metadata(registry, ua_sess->consumer); + } } cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, @@ -486,23 +838,46 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess, assert(!ret); delete_ust_app_channel(sock, ua_chan, app); } - lttng_ht_destroy(ua_sess->channels); - ust_registry_session_destroy(&ua_sess->registry); + /* In case of per PID, the registry is kept in the session. */ + if (ua_sess->buffer_type == LTTNG_BUFFER_PER_PID) { + struct buffer_reg_pid *reg_pid = buffer_reg_pid_find(ua_sess->id); + if (reg_pid) { + /* + * Registry can be null on error path during + * initialization. + */ + buffer_reg_pid_remove(reg_pid); + buffer_reg_pid_destroy(reg_pid); + } + } if (ua_sess->handle != -1) { + pthread_mutex_lock(&app->sock_lock); ret = ustctl_release_handle(sock, ua_sess->handle); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app sock %d release session handle failed with ret %d", sock, ret); } + /* Remove session from application UST object descriptor. */ + iter.iter.node = &ua_sess->ust_objd_node.node; + ret = lttng_ht_del(app->ust_sessions_objd, &iter); + assert(!ret); } - free(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); } /* * Delete a traceable application structure from the global list. Never call * this function outside of a call_rcu call. + * + * RCU read side lock should _NOT_ be held when calling this function. */ static void delete_ust_app(struct ust_app *app) @@ -510,21 +885,28 @@ void delete_ust_app(struct ust_app *app) int ret, sock; struct ust_app_session *ua_sess, *tmp_ua_sess; - rcu_read_lock(); - + /* + * The session list lock must be held during this function to guarantee + * the existence of ua_sess. + */ + session_lock_list(); /* Delete ust app sessions info */ sock = app->sock; app->sock = -1; - lttng_ht_destroy(app->sessions); - /* Wipe sessions */ 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. */ + rcu_read_lock(); delete_ust_app_session(sock, ua_sess, app); + rcu_read_unlock(); } + ht_cleanup_push(app->sessions); + ht_cleanup_push(app->ust_sessions_objd); + ht_cleanup_push(app->ust_objd); + /* * Wait until we have deleted the application from the sock hash table * before closing this socket, otherwise an application could re-use the @@ -546,8 +928,7 @@ void delete_ust_app(struct ust_app *app) DBG2("UST app pid %d deleted", app->pid); free(app); - - rcu_read_unlock(); + session_unlock_list(); } /* @@ -568,6 +949,8 @@ void delete_ust_app_rcu(struct rcu_head *head) /* * Delete the session from the application ht and delete the data structure by * freeing every object inside and releasing them. + * + * The session list lock must be held by the caller. */ static void destroy_app_session(struct ust_app *app, struct ust_app_session *ua_sess) @@ -596,7 +979,7 @@ end: * Alloc new UST app session. */ static -struct ust_app_session *alloc_ust_app_session(struct ust_app *app) +struct ust_app_session *alloc_ust_app_session(void) { struct ust_app_session *ua_sess; @@ -609,22 +992,11 @@ struct ust_app_session *alloc_ust_app_session(struct ust_app *app) ua_sess->handle = -1; ua_sess->channels = lttng_ht_new(0, LTTNG_HT_TYPE_STRING); + ua_sess->metadata_attr.type = LTTNG_UST_CHAN_METADATA; pthread_mutex_init(&ua_sess->lock, NULL); - 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; } @@ -652,15 +1024,14 @@ struct ust_app_channel *alloc_ust_app_channel(char *name, ua_chan->enabled = 1; ua_chan->handle = -1; + ua_chan->session = ua_sess; 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); + CDS_INIT_LIST_HEAD(&ua_chan->ctx_list); /* Copy attributes */ if (attr) { @@ -671,6 +1042,7 @@ struct ust_app_channel *alloc_ust_app_channel(char *name, ua_chan->attr.switch_timer_interval = attr->switch_timer_interval; ua_chan->attr.read_timer_interval = attr->read_timer_interval; ua_chan->attr.output = attr->output; + ua_chan->attr.blocking_timeout = attr->u.s.blocking_timeout; } /* By default, the channel is a per cpu channel. */ ua_chan->attr.type = LTTNG_UST_CHAN_PER_CPU; @@ -743,7 +1115,7 @@ error: * Alloc new UST app context. */ static -struct ust_app_ctx *alloc_ust_app_ctx(struct lttng_ust_context *uctx) +struct ust_app_ctx *alloc_ust_app_ctx(struct lttng_ust_context_attr *uctx) { struct ust_app_ctx *ua_ctx; @@ -752,14 +1124,31 @@ struct ust_app_ctx *alloc_ust_app_ctx(struct lttng_ust_context *uctx) goto error; } + CDS_INIT_LIST_HEAD(&ua_ctx->list); + if (uctx) { memcpy(&ua_ctx->ctx, uctx, sizeof(ua_ctx->ctx)); + if (uctx->ctx == LTTNG_UST_CONTEXT_APP_CONTEXT) { + char *provider_name = NULL, *ctx_name = NULL; + + provider_name = strdup(uctx->u.app_ctx.provider_name); + ctx_name = strdup(uctx->u.app_ctx.ctx_name); + if (!provider_name || !ctx_name) { + free(provider_name); + free(ctx_name); + goto error; + } + + ua_ctx->ctx.u.app_ctx.provider_name = provider_name; + ua_ctx->ctx.u.app_ctx.ctx_name = ctx_name; + } } DBG3("UST app context %d allocated", ua_ctx->ctx.ctx); - -error: return ua_ctx; +error: + free(ua_ctx); + return NULL; } /* @@ -767,15 +1156,15 @@ error: * * Return allocated filter or NULL on error. */ -static struct lttng_ust_filter_bytecode *alloc_copy_ust_app_filter( - struct lttng_ust_filter_bytecode *orig_f) +static struct lttng_filter_bytecode *copy_filter_bytecode( + struct lttng_filter_bytecode *orig_f) { - struct lttng_ust_filter_bytecode *filter = NULL; + struct lttng_filter_bytecode *filter = NULL; /* Copy filter bytecode */ filter = zmalloc(sizeof(*filter) + orig_f->len); if (!filter) { - PERROR("zmalloc alloc ust app filter"); + PERROR("zmalloc alloc filter bytecode"); goto error; } @@ -785,12 +1174,35 @@ error: return filter; } +/* + * Create a liblttng-ust filter bytecode from given bytecode. + * + * Return allocated filter or NULL on error. + */ +static struct lttng_ust_filter_bytecode *create_ust_bytecode_from_bytecode( + struct lttng_filter_bytecode *orig_f) +{ + struct lttng_ust_filter_bytecode *filter = NULL; + + /* Copy filter bytecode */ + filter = zmalloc(sizeof(*filter) + orig_f->len); + if (!filter) { + PERROR("zmalloc alloc ust filter bytecode"); + goto error; + } + + assert(sizeof(struct lttng_filter_bytecode) == + sizeof(struct lttng_ust_filter_bytecode)); + memcpy(filter, orig_f, sizeof(*filter) + orig_f->len); +error: + return filter; +} + /* * Find an ust_app using the sock and return it. RCU read side lock must be * held before calling this helper function. */ -static -struct ust_app *find_app_by_sock(int sock) +struct ust_app *ust_app_find_by_sock(int sock) { struct lttng_ht_node_ulong *node; struct lttng_ht_iter iter; @@ -838,7 +1250,9 @@ error: * Return an ust_app_event object or NULL on error. */ static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht, - char *name, struct lttng_ust_filter_bytecode *filter, int loglevel) + char *name, struct lttng_filter_bytecode *filter, + int loglevel_value, + const struct lttng_event_exclusion *exclusion) { struct lttng_ht_iter iter; struct lttng_ht_node_str *node; @@ -851,7 +1265,9 @@ static struct ust_app_event *find_ust_app_event(struct lttng_ht *ht, /* Setup key for event lookup. */ key.name = name; key.filter = filter; - key.loglevel = loglevel; + key.loglevel_type = loglevel_value; + /* lttng_event_exclusion and lttng_ust_event_exclusion structures are similar */ + key.exclusion = exclusion; /* Lookup using the event name as hash and a custom match fct. */ cds_lfht_lookup(ht->ht, ht->hash_fct((void *) name, lttng_ht_seed), @@ -880,13 +1296,21 @@ int create_ust_channel_context(struct ust_app_channel *ua_chan, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_add_context(app->sock, &ua_ctx->ctx, ua_chan->obj, &ua_ctx->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app create channel context failed for app (pid: %d) " "with ret %d", app->pid, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app disable event failed. Application is dead."); } goto error; @@ -910,6 +1334,7 @@ int set_ust_event_filter(struct ust_app_event *ua_event, struct ust_app *app) { int ret; + struct lttng_ust_filter_bytecode *ust_bytecode = NULL; health_code_update(); @@ -918,13 +1343,26 @@ int set_ust_event_filter(struct ust_app_event *ua_event, goto error; } - ret = ustctl_set_filter(app->sock, ua_event->filter, + ust_bytecode = create_ust_bytecode_from_bytecode(ua_event->filter); + if (!ust_bytecode) { + ret = -LTTNG_ERR_NOMEM; + goto error; + } + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_set_filter(app->sock, ust_bytecode, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app event %s filter failed for app (pid: %d) " "with ret %d", ua_event->attr.name, app->pid, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app filter event failed. Application is dead."); } goto error; @@ -934,6 +1372,78 @@ int set_ust_event_filter(struct ust_app_event *ua_event, error: health_code_update(); + free(ust_bytecode); + return ret; +} + +static +struct lttng_ust_event_exclusion *create_ust_exclusion_from_exclusion( + struct lttng_event_exclusion *exclusion) +{ + struct lttng_ust_event_exclusion *ust_exclusion = NULL; + size_t exclusion_alloc_size = sizeof(struct lttng_ust_event_exclusion) + + LTTNG_UST_SYM_NAME_LEN * exclusion->count; + + ust_exclusion = zmalloc(exclusion_alloc_size); + if (!ust_exclusion) { + PERROR("malloc"); + goto end; + } + + assert(sizeof(struct lttng_event_exclusion) == + sizeof(struct lttng_ust_event_exclusion)); + memcpy(ust_exclusion, exclusion, exclusion_alloc_size); +end: + return ust_exclusion; +} + +/* + * Set event exclusions on the tracer. + */ +static +int set_ust_event_exclusion(struct ust_app_event *ua_event, + struct ust_app *app) +{ + int ret; + struct lttng_ust_event_exclusion *ust_exclusion = NULL; + + health_code_update(); + + if (!ua_event->exclusion || !ua_event->exclusion->count) { + ret = 0; + goto error; + } + + ust_exclusion = create_ust_exclusion_from_exclusion( + ua_event->exclusion); + if (!ust_exclusion) { + ret = -LTTNG_ERR_NOMEM; + goto error; + } + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_set_exclusion(app->sock, ust_exclusion, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app event %s exclusions failed for app (pid: %d) " + "with ret %d", ua_event->attr.name, app->pid, ret); + } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; + DBG3("UST app event exclusion failed. Application is dead."); + } + goto error; + } + + DBG2("UST exclusion set successfully for event %s", ua_event->name); + +error: + health_code_update(); + free(ust_exclusion); return ret; } @@ -947,13 +1457,21 @@ static int disable_ust_event(struct ust_app *app, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_disable(app->sock, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app event %s disable failed for app (pid: %d) " "and session handle %d with ret %d", ua_event->attr.name, app->pid, ua_sess->handle, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app disable event failed. Application is dead."); } goto error; @@ -977,13 +1495,21 @@ static int disable_ust_channel(struct ust_app *app, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_disable(app->sock, ua_chan->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app channel %s disable failed for app (pid: %d) " "and session handle %d with ret %d", ua_chan->name, app->pid, ua_sess->handle, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app disable channel failed. Application is dead."); } goto error; @@ -1007,13 +1533,21 @@ static int enable_ust_channel(struct ust_app *app, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_enable(app->sock, ua_chan->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app channel %s enable failed for app (pid: %d) " "and session handle %d with ret %d", ua_chan->name, app->pid, ua_sess->handle, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app enable channel failed. Application is dead."); } goto error; @@ -1039,13 +1573,21 @@ static int enable_ust_event(struct ust_app *app, health_code_update(); + pthread_mutex_lock(&app->sock_lock); ret = ustctl_enable(app->sock, ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("UST app event %s enable failed for app (pid: %d) " "and session handle %d with ret %d", ua_event->attr.name, app->pid, ua_sess->handle, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app enable event failed. Application is dead."); } goto error; @@ -1060,73 +1602,31 @@ error: } /* - * Create the specified channel onto the UST tracer for a UST session. This - * MUST be called with UST app session lock held. + * Send channel and stream buffer to application. * * Return 0 on success. On error, a negative value is returned. */ -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) +static int send_channel_pid_to_ust(struct ust_app *app, + struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan) { int ret; - unsigned int nb_fd = 0; - struct consumer_socket *socket; struct ust_app_stream *stream, *stmp; assert(app); assert(ua_sess); assert(ua_chan); - assert(consumer); - - rcu_read_lock(); - health_code_update(); - - /* 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; - } - - health_code_update(); - - /* - * 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) { - goto error; - } - - /* - * 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; - - /* 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_fd_get; - } health_code_update(); - /* - * 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) { - goto error_destroy; - } + DBG("UST app sending channel %s to UST app sock %d", ua_chan->name, + app->sock); /* 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; } @@ -1135,48 +1635,21 @@ static int create_ust_channel(struct ust_app *app, /* Send all streams to application. */ cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) { ret = ust_consumer_send_stream_to_ust(app, ua_chan, stream); - if (ret < 0) { + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + goto error; + } else if (ret < 0) { goto error; } /* We don't need the stream anymore once sent to the tracer. */ cds_list_del(&stream->list); - delete_ust_app_stream(-1, stream); + delete_ust_app_stream(-1, stream, app); } - /* Flag the channel that it is sent to the application. */ ua_chan->is_sent = 1; - /* 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(); - rcu_read_unlock(); return ret; } @@ -1194,13 +1667,21 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, health_code_update(); /* Create UST event on tracer */ + pthread_mutex_lock(&app->sock_lock); ret = ustctl_create_event(app->sock, &ua_event->attr, ua_chan->obj, &ua_event->obj); + pthread_mutex_unlock(&app->sock_lock); if (ret < 0) { if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { ERR("Error ustctl create event %s for app pid: %d with ret %d", ua_event->attr.name, app->pid, ret); } else { + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + ret = 0; DBG3("UST app create event failed. Application is dead."); } goto error; @@ -1221,12 +1702,24 @@ int create_ust_event(struct ust_app *app, struct ust_app_session *ua_sess, } } + /* Set exclusions for the event */ + if (ua_event->exclusion) { + ret = set_ust_event_exclusion(ua_event, app); + if (ret < 0) { + goto error; + } + } + /* If event not enabled, disable it on the tracer */ - if (ua_event->enabled == 0) { - ret = disable_ust_event(app, ua_sess, ua_event); + if (ua_event->enabled) { + /* + * We now need to explicitly enable the event, since it + * is now disabled at creation. + */ + ret = enable_ust_event(app, ua_sess, ua_event); if (ret < 0) { /* - * If we hit an EPERM, something is wrong with our disable call. If + * If we hit an EPERM, something is wrong with our enable call. If * we get an EEXIST, there is a problem on the tracer side since we * just created it. */ @@ -1256,6 +1749,8 @@ error: static void shadow_copy_event(struct ust_app_event *ua_event, struct ltt_ust_event *uevent) { + size_t exclusion_alloc_size; + strncpy(ua_event->name, uevent->attr.name, sizeof(ua_event->name)); ua_event->name[sizeof(ua_event->name) - 1] = '\0'; @@ -1266,9 +1761,22 @@ static void shadow_copy_event(struct ust_app_event *ua_event, /* Copy filter bytecode */ if (uevent->filter) { - ua_event->filter = alloc_copy_ust_app_filter(uevent->filter); + ua_event->filter = copy_filter_bytecode(uevent->filter); /* Filter might be NULL here in case of ENONEM. */ } + + /* Copy exclusion data */ + if (uevent->exclusion) { + exclusion_alloc_size = sizeof(struct lttng_event_exclusion) + + LTTNG_UST_SYM_NAME_LEN * uevent->exclusion->count; + ua_event->exclusion = zmalloc(exclusion_alloc_size); + if (ua_event->exclusion == NULL) { + PERROR("malloc"); + } else { + memcpy(ua_event->exclusion, uevent->exclusion, + exclusion_alloc_size); + } + } } /* @@ -1281,41 +1789,49 @@ static void shadow_copy_channel(struct ust_app_channel *ua_chan, struct ltt_ust_event *uevent; struct ltt_ust_context *uctx; struct ust_app_event *ua_event; - struct ust_app_ctx *ua_ctx; DBG2("UST app shadow copy of channel %s started", ua_chan->name); strncpy(ua_chan->name, uchan->name, sizeof(ua_chan->name)); ua_chan->name[sizeof(ua_chan->name) - 1] = '\0'; + ua_chan->tracefile_size = uchan->tracefile_size; + ua_chan->tracefile_count = uchan->tracefile_count; + /* Copy event attributes since the layout is different. */ ua_chan->attr.subbuf_size = uchan->attr.subbuf_size; ua_chan->attr.num_subbuf = uchan->attr.num_subbuf; ua_chan->attr.overwrite = uchan->attr.overwrite; ua_chan->attr.switch_timer_interval = uchan->attr.switch_timer_interval; ua_chan->attr.read_timer_interval = uchan->attr.read_timer_interval; + ua_chan->monitor_timer_interval = uchan->monitor_timer_interval; ua_chan->attr.output = uchan->attr.output; + ua_chan->attr.blocking_timeout = uchan->attr.u.s.blocking_timeout; + /* * Note that the attribute channel type is not set since the channel on the * tracing registry side does not have this information. */ ua_chan->enabled = uchan->enabled; + ua_chan->tracing_channel_id = uchan->id; + + cds_list_for_each_entry(uctx, &uchan->ctx_list, list) { + struct ust_app_ctx *ua_ctx = alloc_ust_app_ctx(&uctx->ctx); - cds_lfht_for_each_entry(uchan->ctx->ht, &iter.iter, uctx, node.node) { - ua_ctx = alloc_ust_app_ctx(&uctx->ctx); if (ua_ctx == NULL) { continue; } lttng_ht_node_init_ulong(&ua_ctx->node, (unsigned long) ua_ctx->ctx.ctx); - lttng_ht_add_unique_ulong(ua_chan->ctx, &ua_ctx->node); + lttng_ht_add_ulong(ua_chan->ctx, &ua_ctx->node); + cds_list_add_tail(&ua_ctx->list, &ua_chan->ctx_list); } /* Copy all events from ltt ust channel to ust app channel */ cds_lfht_for_each_entry(uchan->events->ht, &iter.iter, uevent, node.node) { ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel); + uevent->filter, uevent->attr.loglevel, uevent->exclusion); if (ua_event == NULL) { DBG2("UST event %s not found on shadow copy channel", uevent->attr.name); @@ -1345,6 +1861,7 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, struct tm *timeinfo; char datetime[16]; int ret; + char tmp_shm_path[PATH_MAX]; /* Get date and time for unique app path */ time(&rawtime); @@ -1353,19 +1870,75 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, DBG2("Shadow copy of session handle %d", ua_sess->handle); - ua_sess->id = usess->id; - ua_sess->uid = usess->uid; - ua_sess->gid = usess->gid; - - ret = snprintf(ua_sess->path, PATH_MAX, "%s-%d-%s/", app->name, app->pid, - datetime); + ua_sess->tracing_id = usess->id; + ua_sess->id = get_next_session_id(); + ua_sess->uid = app->uid; + ua_sess->gid = app->gid; + ua_sess->euid = usess->uid; + ua_sess->egid = usess->gid; + ua_sess->buffer_type = usess->buffer_type; + ua_sess->bits_per_long = app->bits_per_long; + + /* There is only one consumer object per session possible. */ + consumer_output_get(usess->consumer); + ua_sess->consumer = usess->consumer; + + ua_sess->output_traces = usess->output_traces; + ua_sess->live_timer_interval = usess->live_timer_interval; + copy_channel_attr_to_ustctl(&ua_sess->metadata_attr, + &usess->metadata_attr); + + switch (ua_sess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + ret = snprintf(ua_sess->path, sizeof(ua_sess->path), + DEFAULT_UST_TRACE_PID_PATH "/%s-%d-%s", app->name, app->pid, + datetime); + break; + case LTTNG_BUFFER_PER_UID: + ret = snprintf(ua_sess->path, sizeof(ua_sess->path), + DEFAULT_UST_TRACE_UID_PATH, ua_sess->uid, app->bits_per_long); + break; + default: + assert(0); + goto error; + } if (ret < 0) { PERROR("asprintf UST shadow copy session"); - /* TODO: We cannot return an error from here.. */ assert(0); + goto error; } - /* TODO: support all UST domain */ + 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, @@ -1381,7 +1954,8 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, DBG2("Channel %s not found on shadow session copy, creating it", uchan->name); - ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, &uchan->attr); + ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, + &uchan->attr); if (ua_chan == NULL) { /* malloc failed FIXME: Might want to do handle ENOMEM .. */ continue; @@ -1396,6 +1970,10 @@ static void shadow_copy_session(struct ust_app_session *ua_sess, lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node); } + return; + +error: + consumer_output_put(ua_sess->consumer); } /* @@ -1406,7 +1984,7 @@ void __lookup_session_by_app(struct ltt_ust_session *usess, struct ust_app *app, struct lttng_ht_iter *iter) { /* Get right UST app session from app */ - lttng_ht_lookup(app->sessions, (void *)((unsigned long) usess->id), iter); + lttng_ht_lookup(app->sessions, &usess->id, iter); } /* @@ -1417,10 +1995,10 @@ static struct ust_app_session *lookup_session_by_app( struct ltt_ust_session *usess, struct ust_app *app) { struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; + struct lttng_ht_node_u64 *node; __lookup_session_by_app(usess, app, &iter); - node = lttng_ht_iter_get_node_ulong(&iter); + node = lttng_ht_iter_get_node_u64(&iter); if (node == NULL) { goto error; } @@ -1432,53 +2010,219 @@ error: } /* - * 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. + * Setup buffer registry per PID for the given session and application. If none + * is found, a new one is created, added to the global registry and + * initialized. If regp is valid, it's set with the newly created object. * - * 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. + * Return 0 on success or else a negative value. */ -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) +static int setup_buffer_reg_pid(struct ust_app_session *ua_sess, + struct ust_app *app, struct buffer_reg_pid **regp) { - int ret, created = 0; - struct ust_app_session *ua_sess; + int ret = 0; + struct buffer_reg_pid *reg_pid; - assert(usess); + assert(ua_sess); assert(app); - assert(ua_sess_ptr); - health_code_update(); + rcu_read_lock(); - 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(app); - if (ua_sess == NULL) { - /* Only malloc can failed so something is really wrong */ - ret = -ENOMEM; + reg_pid = buffer_reg_pid_find(ua_sess->id); + if (!reg_pid) { + /* + * 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, + ua_sess->root_shm_path, ua_sess->shm_path); + if (ret < 0) { goto error; } - shadow_copy_session(ua_sess, usess, app); - created = 1; + } else { + goto end; } - health_code_update(); + /* Initialize registry. */ + ret = ust_registry_session_init(®_pid->registry->reg.ust, 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, app->version.major, + app->version.minor, reg_pid->root_shm_path, + reg_pid->shm_path, + ua_sess->euid, ua_sess->egid); + if (ret < 0) { + /* + * reg_pid->registry->reg.ust is NULL upon error, so we need to + * destroy the buffer registry, because it is always expected + * that if the buffer registry can be found, its ust registry is + * non-NULL. + */ + buffer_reg_pid_destroy(reg_pid); + goto error; + } - if (ua_sess->handle == -1) { - ret = ustctl_create_session(app->sock); - if (ret < 0) { - if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { - ERR("Creating session for app pid %d with ret %d", + buffer_reg_pid_add(reg_pid); + + DBG3("UST app buffer registry per PID created successfully"); + +end: + if (regp) { + *regp = reg_pid; + } +error: + rcu_read_unlock(); + return ret; +} + +/* + * Setup buffer registry per UID for the given session and application. If none + * is found, a new one is created, added to the global registry and + * initialized. If regp is valid, it's set with the newly created object. + * + * 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; + struct buffer_reg_uid *reg_uid; + + assert(usess); + assert(app); + + rcu_read_lock(); + + reg_uid = buffer_reg_uid_find(usess->id, app->bits_per_long, app->uid); + if (!reg_uid) { + /* + * 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_uid_create(usess->id, app->bits_per_long, app->uid, + LTTNG_DOMAIN_UST, ®_uid, + ua_sess->root_shm_path, ua_sess->shm_path); + if (ret < 0) { + goto error; + } + } else { + goto end; + } + + /* Initialize registry. */ + ret = ust_registry_session_init(®_uid->registry->reg.ust, NULL, + app->bits_per_long, app->uint8_t_alignment, + app->uint16_t_alignment, app->uint32_t_alignment, + app->uint64_t_alignment, app->long_alignment, + app->byte_order, app->version.major, + app->version.minor, reg_uid->root_shm_path, + reg_uid->shm_path, usess->uid, usess->gid); + if (ret < 0) { + /* + * reg_uid->registry->reg.ust is NULL upon error, so we need to + * destroy the buffer registry, because it is always expected + * that if the buffer registry can be found, its ust registry is + * non-NULL. + */ + buffer_reg_uid_destroy(reg_uid, NULL); + goto error; + } + /* Add node to teardown list of the session. */ + cds_list_add(®_uid->lnode, &usess->buffer_reg_uid_list); + + buffer_reg_uid_add(reg_uid); + + DBG3("UST app buffer registry per UID created successfully"); +end: + if (regp) { + *regp = reg_uid; + } +error: + rcu_read_unlock(); + return ret; +} + +/* + * 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. + * + * Returns 0 on success or else a negative code which is either -ENOMEM or + * -ENOTCONN which is the default code if the ustctl_create_session fails. + */ +static int find_or_create_ust_app_session(struct ltt_ust_session *usess, + struct ust_app *app, struct ust_app_session **ua_sess_ptr, + int *is_created) +{ + int ret, created = 0; + struct ust_app_session *ua_sess; + + 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 %" PRIu64 " not found, creating it", + app->pid, usess->id); + ua_sess = alloc_ust_app_session(); + if (ua_sess == NULL) { + /* Only malloc can failed so something is really wrong */ + ret = -ENOMEM; + goto error; + } + shadow_copy_session(ua_sess, usess, app); + created = 1; + } + + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + /* Init local registry. */ + ret = setup_buffer_reg_pid(ua_sess, app, NULL); + if (ret < 0) { + delete_ust_app_session(-1, ua_sess, app); + goto error; + } + break; + case LTTNG_BUFFER_PER_UID: + /* Look for a global registry. If none exists, create one. */ + ret = setup_buffer_reg_uid(usess, ua_sess, app, NULL); + if (ret < 0) { + delete_ust_app_session(-1, ua_sess, app); + goto error; + } + break; + default: + assert(0); + ret = -EINVAL; + goto error; + } + + health_code_update(); + + if (ua_sess->handle == -1) { + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_create_session(app->sock); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("Creating session for app pid %d with ret %d", app->pid, ret); } else { DBG("UST app creating session failed. Application is dead"); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. This will get flagged ENOTCONN and the + * caller will handle it. + */ + ret = 0; } delete_ust_app_session(-1, ua_sess, app); if (ret != -ENOMEM) { @@ -1494,20 +2238,16 @@ static int create_ust_app_session(struct ltt_ust_session *usess, ua_sess->handle = ret; /* Add ust app session to app's HT */ - lttng_ht_node_init_ulong(&ua_sess->node, (unsigned long) ua_sess->id); - lttng_ht_add_unique_ulong(app->sessions, &ua_sess->node); + lttng_ht_node_init_u64(&ua_sess->node, + ua_sess->tracing_id); + lttng_ht_add_unique_u64(app->sessions, &ua_sess->node); + lttng_ht_node_init_ulong(&ua_sess->ust_objd_node, ua_sess->handle); + lttng_ht_add_unique_ulong(app->ust_sessions_objd, + &ua_sess->ust_objd_node); DBG2("UST app session created successfully with handle %d", ret); } - /* - * 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; @@ -1521,26 +2261,103 @@ error: return ret; } +/* + * Match function for a hash table lookup of ust_app_ctx. + * + * It matches an ust app context based on the context type and, in the case + * of perf counters, their name. + */ +static int ht_match_ust_app_ctx(struct cds_lfht_node *node, const void *_key) +{ + struct ust_app_ctx *ctx; + const struct lttng_ust_context_attr *key; + + assert(node); + assert(_key); + + ctx = caa_container_of(node, struct ust_app_ctx, node.node); + key = _key; + + /* Context type */ + if (ctx->ctx.ctx != key->ctx) { + goto no_match; + } + + switch(key->ctx) { + case LTTNG_UST_CONTEXT_PERF_THREAD_COUNTER: + if (strncmp(key->u.perf_counter.name, + ctx->ctx.u.perf_counter.name, + sizeof(key->u.perf_counter.name))) { + goto no_match; + } + break; + case LTTNG_UST_CONTEXT_APP_CONTEXT: + if (strcmp(key->u.app_ctx.provider_name, + ctx->ctx.u.app_ctx.provider_name) || + strcmp(key->u.app_ctx.ctx_name, + ctx->ctx.u.app_ctx.ctx_name)) { + goto no_match; + } + break; + default: + break; + } + + /* Match. */ + return 1; + +no_match: + return 0; +} + +/* + * Lookup for an ust app context from an lttng_ust_context. + * + * Must be called while holding RCU read side lock. + * Return an ust_app_ctx object or NULL on error. + */ +static +struct ust_app_ctx *find_ust_app_context(struct lttng_ht *ht, + struct lttng_ust_context_attr *uctx) +{ + struct lttng_ht_iter iter; + struct lttng_ht_node_ulong *node; + struct ust_app_ctx *app_ctx = NULL; + + assert(uctx); + assert(ht); + + /* Lookup using the lttng_ust_context_type and a custom match fct. */ + cds_lfht_lookup(ht->ht, ht->hash_fct((void *) uctx->ctx, lttng_ht_seed), + ht_match_ust_app_ctx, uctx, &iter.iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (!node) { + goto end; + } + + app_ctx = caa_container_of(node, struct ust_app_ctx, node); + +end: + return app_ctx; +} + /* * Create a context for the channel on the tracer. * - * Called with UST app session lock held. + * Called with UST app session lock held and a RCU read side lock. */ static -int create_ust_app_channel_context(struct ust_app_session *ua_sess, - struct ust_app_channel *ua_chan, struct lttng_ust_context *uctx, +int create_ust_app_channel_context(struct ust_app_channel *ua_chan, + struct lttng_ust_context_attr *uctx, struct ust_app *app) { int ret = 0; - struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; struct ust_app_ctx *ua_ctx; DBG2("UST app adding context to channel %s", ua_chan->name); - lttng_ht_lookup(ua_chan->ctx, (void *)((unsigned long)uctx->ctx), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node != NULL) { + ua_ctx = find_ust_app_context(ua_chan->ctx, uctx); + if (ua_ctx) { ret = -EEXIST; goto error; } @@ -1548,12 +2365,13 @@ int create_ust_app_channel_context(struct ust_app_session *ua_sess, ua_ctx = alloc_ust_app_ctx(uctx); if (ua_ctx == NULL) { /* malloc failed */ - ret = -1; + ret = -ENOMEM; goto error; } lttng_ht_node_init_ulong(&ua_ctx->node, (unsigned long) ua_ctx->ctx.ctx); - lttng_ht_add_unique_ulong(ua_chan->ctx, &ua_ctx->node); + lttng_ht_add_ulong(ua_chan->ctx, &ua_ctx->node); + cds_list_add_tail(&ua_ctx->list, &ua_chan->ctx_list); ret = create_ust_channel_context(ua_chan, ua_ctx, app); if (ret < 0) { @@ -1626,7 +2444,8 @@ error: } /* - * Lookup ust app channel for session and enable it on the tracer side. + * Lookup ust app channel for session and enable it on the tracer side. This + * MUST be called with a RCU read side lock acquired. */ static int enable_ust_app_channel(struct ust_app_session *ua_sess, struct ltt_ust_channel *uchan, struct ust_app *app) @@ -1639,8 +2458,8 @@ static int enable_ust_app_channel(struct ust_app_session *ua_sess, lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter); ua_chan_node = lttng_ht_iter_get_node_str(&iter); if (ua_chan_node == NULL) { - DBG2("Unable to find channel %s in ust session id %u", - uchan->name, ua_sess->id); + DBG2("Unable to find channel %s in ust session id %" PRIu64, + uchan->name, ua_sess->tracing_id); goto error; } @@ -1656,1109 +2475,2732 @@ error: } /* - * Create UST app channel and create it on the tracer. Set ua_chanp of the - * newly created channel if not NULL. + * Ask the consumer to create a channel and get it if successful. * * Called with UST app session lock held. + * + * Return 0 on success or else a negative value. */ -static int 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) +static int do_consumer_create_channel(struct ltt_ust_session *usess, + struct ust_app_session *ua_sess, struct ust_app_channel *ua_chan, + int bitness, struct ust_registry_session *registry, + uint64_t trace_archive_id) { - int ret = 0; - struct lttng_ht_iter iter; - struct lttng_ht_node_str *ua_chan_node; - struct ust_app_channel *ua_chan; + int ret; + unsigned int nb_fd = 0; + struct consumer_socket *socket; - /* Lookup channel in the ust app session */ - lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter); - ua_chan_node = lttng_ht_iter_get_node_str(&iter); - if (ua_chan_node != NULL) { - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - goto end; - } + assert(usess); + assert(ua_sess); + assert(ua_chan); + assert(registry); - ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, &uchan->attr); - if (ua_chan == NULL) { - /* Only malloc can fail here */ - ret = -ENOMEM; + rcu_read_lock(); + health_code_update(); + + /* Get the right consumer socket for the application. */ + socket = consumer_find_socket_by_bitness(bitness, usess->consumer); + if (!socket) { + ret = -EINVAL; goto error; } - shadow_copy_channel(ua_chan, uchan); - /* Set channel type. */ - ua_chan->attr.type = type; + health_code_update(); - ret = create_ust_channel(app, ua_sess, ua_chan, consumer); + /* Need one fd for the channel. */ + ret = lttng_fd_get(LTTNG_FD_APPS, 1); if (ret < 0) { + ERR("Exhausted number of available FD upon create channel"); goto error; } - DBG2("UST app create channel %s for PID %d completed", ua_chan->name, - app->pid); + /* + * 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, usess->consumer, socket, + registry, trace_archive_id); + if (ret < 0) { + goto error_ask; + } - /* Only add the channel if successful on the tracer side. */ - lttng_ht_add_unique_str(ua_sess->channels, &ua_chan->node); + /* + * 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; -end: - if (ua_chanp) { - *ua_chanp = ua_chan; + /* 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_fd_get_stream; } - /* Everything went well. */ + health_code_update(); + + /* + * Now get the channel from the consumer. This call wil populate the stream + * list of that channel and set the ust objects. + */ + if (usess->consumer->enabled) { + ret = ust_consumer_get_channel(socket, ua_chan); + if (ret < 0) { + goto error_destroy; + } + } + + rcu_read_unlock(); return 0; +error_destroy: + lttng_fd_put(LTTNG_FD_APPS, nb_fd); +error_fd_get_stream: + /* + * 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_ask: + lttng_fd_put(LTTNG_FD_APPS, 1); error: - delete_ust_app_channel(ua_chan->is_sent ? app->sock : -1, ua_chan, app); + health_code_update(); + rcu_read_unlock(); return ret; } /* - * Create UST app event and create it on the tracer side. + * Duplicate the ust data object of the ust app stream and save it in the + * buffer registry stream. * - * Called with ust app session mutex held. + * Return 0 on success or else a negative value. */ -static -int create_ust_app_event(struct ust_app_session *ua_sess, - struct ust_app_channel *ua_chan, struct ltt_ust_event *uevent, - struct ust_app *app) +static int duplicate_stream_object(struct buffer_reg_stream *reg_stream, + struct ust_app_stream *stream) { - int ret = 0; - struct ust_app_event *ua_event; + int ret; - /* Get event node */ - ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel); - if (ua_event != NULL) { - ret = -EEXIST; - goto end; - } + assert(reg_stream); + assert(stream); - /* Does not exist so create one */ - ua_event = alloc_ust_app_event(uevent->attr.name, &uevent->attr); - if (ua_event == NULL) { - /* Only malloc can failed so something is really wrong */ - ret = -ENOMEM; - goto end; + /* Reserve the amount of file descriptor we need. */ + ret = lttng_fd_get(LTTNG_FD_APPS, 2); + if (ret < 0) { + ERR("Exhausted number of available FD upon duplicate stream"); + goto error; } - shadow_copy_event(ua_event, uevent); - /* Create it on the tracer side */ - ret = create_ust_event(app, ua_sess, ua_chan, ua_event); + /* Duplicate object for stream once the original is in the registry. */ + ret = ustctl_duplicate_ust_object_data(&stream->obj, + reg_stream->obj.ust); if (ret < 0) { - /* Not found previously means that it does not exist on the tracer */ - assert(ret != -LTTNG_UST_ERR_EXIST); + ERR("Duplicate stream obj from %p to %p failed with ret %d", + reg_stream->obj.ust, stream->obj, ret); + lttng_fd_put(LTTNG_FD_APPS, 2); goto error; } + stream->handle = stream->obj->handle; - add_unique_ust_app_event(ua_chan, ua_event); +error: + return ret; +} - DBG2("UST app create event %s for PID %d completed", ua_event->name, - app->pid); +/* + * Duplicate the ust data object of the ust app. channel and save it in the + * buffer registry channel. + * + * Return 0 on success or else a negative value. + */ +static int duplicate_channel_object(struct buffer_reg_channel *reg_chan, + struct ust_app_channel *ua_chan) +{ + int ret; -end: - return ret; + assert(reg_chan); + assert(ua_chan); + + /* Need two fds for the channel. */ + ret = lttng_fd_get(LTTNG_FD_APPS, 1); + if (ret < 0) { + ERR("Exhausted number of available FD upon duplicate channel"); + goto error_fd_get; + } + + /* Duplicate object for stream once the original is in the registry. */ + ret = ustctl_duplicate_ust_object_data(&ua_chan->obj, reg_chan->obj.ust); + if (ret < 0) { + ERR("Duplicate channel obj from %p to %p failed with ret: %d", + reg_chan->obj.ust, ua_chan->obj, ret); + goto error; + } + ua_chan->handle = ua_chan->obj->handle; + + return 0; error: - /* Valid. Calling here is already in a read side lock */ - delete_ust_app_event(-1, ua_event); + lttng_fd_put(LTTNG_FD_APPS, 1); +error_fd_get: return ret; } /* - * Create UST metadata and open it on the tracer side. + * For a given channel buffer registry, setup all streams of the given ust + * application channel. * - * Called with UST app session lock held. + * Return 0 on success or else a negative value. */ -static int create_ust_app_metadata(struct ust_app_session *ua_sess, - struct ust_app *app, struct consumer_output *consumer) +static int setup_buffer_reg_streams(struct buffer_reg_channel *reg_chan, + struct ust_app_channel *ua_chan, + struct ust_app *app) { int ret = 0; - struct ust_app_channel *metadata; - struct consumer_socket *socket; + struct ust_app_stream *stream, *stmp; - assert(ua_sess); - assert(app); - assert(consumer); + assert(reg_chan); + assert(ua_chan); - if (ua_sess->metadata) { - /* Already exist. Return success. */ - goto end; - } + DBG2("UST app setup buffer registry stream"); - /* Allocate UST metadata */ - metadata = alloc_ust_app_channel(DEFAULT_METADATA_NAME, ua_sess, NULL); - if (!metadata) { - /* malloc() failed */ - ret = -ENOMEM; - goto error; - } + /* Send all streams to application. */ + cds_list_for_each_entry_safe(stream, stmp, &ua_chan->streams.head, list) { + struct buffer_reg_stream *reg_stream; + + ret = buffer_reg_stream_create(®_stream); + 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_CHANNEL_SWITCH_TIMER; - metadata->attr.read_timer_interval = DEFAULT_CHANNEL_READ_TIMER; - metadata->attr.output = LTTNG_UST_MMAP; - metadata->attr.type = LTTNG_UST_CHAN_METADATA; + /* + * Keep original pointer and nullify it in the stream so the delete + * stream call does not release the object. + */ + reg_stream->obj.ust = stream->obj; + stream->obj = NULL; + buffer_reg_stream_add(reg_stream, reg_chan); - /* 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; + /* We don't need the streams anymore. */ + cds_list_del(&stream->list); + delete_ust_app_stream(-1, stream, app); } - /* - * 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); +error: + return ret; +} + +/* + * Create a buffer registry channel for the given session registry and + * application channel object. If regp pointer is valid, it's set with the + * created object. Important, the created object is NOT added to the session + * registry hash table. + * + * Return 0 on success else a negative value. + */ +static int create_buffer_reg_channel(struct buffer_reg_session *reg_sess, + struct ust_app_channel *ua_chan, struct buffer_reg_channel **regp) +{ + int ret; + struct buffer_reg_channel *reg_chan = NULL; + + assert(reg_sess); + assert(ua_chan); + + DBG2("UST app creating buffer registry channel for %s", ua_chan->name); + + /* Create buffer registry channel. */ + ret = buffer_reg_channel_create(ua_chan->tracing_channel_id, ®_chan); if (ret < 0) { - goto error_consumer; + goto error_create; } + assert(reg_chan); + reg_chan->consumer_key = ua_chan->key; + reg_chan->subbuf_size = ua_chan->attr.subbuf_size; + reg_chan->num_subbuf = ua_chan->attr.num_subbuf; - /* - * 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); + /* Create and add a channel registry to session. */ + ret = ust_registry_channel_add(reg_sess->reg.ust, + ua_chan->tracing_channel_id); if (ret < 0) { - goto error_consumer; + goto error; } + buffer_reg_channel_add(reg_sess, reg_chan); - ua_sess->metadata = metadata; - - DBG2("UST metadata created for app pid %d", app->pid); + if (regp) { + *regp = reg_chan; + } -end: return 0; -error_consumer: - delete_ust_app_channel(-1, metadata, app); + error: + /* Safe because the registry channel object was not added to any HT. */ + buffer_reg_channel_destroy(reg_chan, LTTNG_DOMAIN_UST); +error_create: return ret; } /* - * Return pointer to traceable apps list. + * Setup buffer registry channel for the given session registry and application + * channel object. If regp pointer is valid, it's set with the created object. + * + * Return 0 on success else a negative value. */ -struct lttng_ht *ust_app_get_ht(void) +static int setup_buffer_reg_channel(struct buffer_reg_session *reg_sess, + struct ust_app_channel *ua_chan, struct buffer_reg_channel *reg_chan, + struct ust_app *app) { - return ust_app_ht; -} + int ret; -/* - * 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; + assert(reg_sess); + assert(reg_chan); + assert(ua_chan); + assert(ua_chan->obj); - 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); + DBG2("UST app setup buffer registry channel for %s", ua_chan->name); + + /* Setup all streams for the registry. */ + ret = setup_buffer_reg_streams(reg_chan, ua_chan, app); + if (ret < 0) { goto error; } - DBG2("Found UST app by pid %d", pid); + reg_chan->obj.ust = ua_chan->obj; + ua_chan->obj = NULL; - app = caa_container_of(node, struct ust_app, pid_n); + return 0; error: - return app; + buffer_reg_channel_remove(reg_sess, reg_chan); + buffer_reg_channel_destroy(reg_chan, LTTNG_DOMAIN_UST); + return ret; } /* - * 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. + * Send buffer registry channel to the application. * - * The object is returned on success or else NULL. + * Return 0 on success else a negative value. */ -struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) +static int send_channel_uid_to_ust(struct buffer_reg_channel *reg_chan, + struct ust_app *app, struct ust_app_session *ua_sess, + struct ust_app_channel *ua_chan) { - struct ust_app *lta = NULL; + int ret; + struct buffer_reg_stream *reg_stream; - assert(msg); - assert(sock >= 0); + assert(reg_chan); + assert(app); + assert(ua_sess); + assert(ua_chan); - DBG3("UST app creating application for socket %d", sock); + DBG("UST app sending buffer registry channel to ust sock %d", app->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 size is available.\n", - msg->name, msg->pid, msg->bits_per_long); + ret = duplicate_channel_object(reg_chan, ua_chan); + if (ret < 0) { goto error; } - lta = zmalloc(sizeof(struct ust_app)); - if (lta == NULL) { - PERROR("malloc"); + /* Send channel to the application. */ + ret = ust_consumer_send_channel_to_ust(app, ua_sess, ua_chan); + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + goto error; + } else if (ret < 0) { goto error; } - lta->ppid = msg->ppid; - lta->uid = msg->uid; - lta->gid = msg->gid; - - 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; - 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; + health_code_update(); - /* Copy name and make sure it's NULL terminated. */ - strncpy(lta->name, msg->name, sizeof(lta->name)); - lta->name[UST_APP_PROCNAME_LEN] = '\0'; + /* Send all streams to application. */ + pthread_mutex_lock(®_chan->stream_list_lock); + cds_list_for_each_entry(reg_stream, ®_chan->streams, lnode) { + struct ust_app_stream stream; - /* - * 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; + ret = duplicate_stream_object(reg_stream, &stream); + if (ret < 0) { + goto error_stream_unlock; + } - lta->pid = msg->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); + ret = ust_consumer_send_stream_to_ust(app, ua_chan, &stream); + if (ret < 0) { + (void) release_ust_app_stream(-1, &stream, app); + if (ret == -EPIPE || ret == -LTTNG_UST_ERR_EXITING) { + ret = -ENOTCONN; /* Caused by app exiting. */ + } + goto error_stream_unlock; + } - CDS_INIT_LIST_HEAD(<a->teardown_head); + /* + * The return value is not important here. This function will output an + * error if needed. + */ + (void) release_ust_app_stream(-1, &stream, app); + } + ua_chan->is_sent = 1; +error_stream_unlock: + pthread_mutex_unlock(®_chan->stream_list_lock); error: - return lta; + return ret; } /* - * For a given application object, add it to every hash table. + * Create and send to the application the created buffers with per UID buffers. + * + * This MUST be called with a RCU read side lock acquired. + * The session list lock and the session's lock must be acquired. + * + * Return 0 on success else a negative value. */ -void ust_app_add(struct ust_app *app) +static int create_channel_per_uid(struct ust_app *app, + struct ltt_ust_session *usess, struct ust_app_session *ua_sess, + struct ust_app_channel *ua_chan) { + int ret; + struct buffer_reg_uid *reg_uid; + struct buffer_reg_channel *reg_chan; + struct ltt_session *session; + enum lttng_error_code notification_ret; + struct ust_registry_channel *chan_reg; + assert(app); - assert(app->notify_sock >= 0); + assert(usess); + assert(ua_sess); + assert(ua_chan); - rcu_read_lock(); + DBG("UST app creating channel %s with per UID buffers", ua_chan->name); + reg_uid = buffer_reg_uid_find(usess->id, app->bits_per_long, app->uid); /* - * On a re-registration, we want to kick out the previous registration of - * that pid + * The session creation handles the creation of this global registry + * object. If none can be find, there is a code flow problem or a + * teardown race. */ - lttng_ht_add_replace_ulong(ust_app_ht, &app->pid_n); + assert(reg_uid); + + reg_chan = buffer_reg_channel_find(ua_chan->tracing_channel_id, + reg_uid); + if (reg_chan) { + goto send_channel; + } + + /* Create the buffer registry channel object. */ + ret = create_buffer_reg_channel(reg_uid->registry, ua_chan, ®_chan); + if (ret < 0) { + ERR("Error creating the UST channel \"%s\" registry instance", + ua_chan->name); + goto error; + } + + session = session_find_by_id(ua_sess->tracing_id); + assert(session); + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); /* - * 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. + * Create the buffers on the consumer side. This call populates the + * ust app channel object with all streams and data object. */ - lttng_ht_add_unique_ulong(ust_app_ht_by_sock, &app->sock_n); + ret = do_consumer_create_channel(usess, ua_sess, ua_chan, + app->bits_per_long, reg_uid->registry->reg.ust, + session->current_archive_id); + if (ret < 0) { + ERR("Error creating UST channel \"%s\" on the consumer daemon", + ua_chan->name); - /* 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); + /* + * Let's remove the previously created buffer registry channel so + * it's not visible anymore in the session registry. + */ + ust_registry_channel_del_free(reg_uid->registry->reg.ust, + ua_chan->tracing_channel_id, false); + buffer_reg_channel_remove(reg_uid->registry, reg_chan); + buffer_reg_channel_destroy(reg_chan, LTTNG_DOMAIN_UST); + goto error; + } - 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(); -} - -/* - * 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; + /* + * Setup the streams and add it to the session registry. + */ + ret = setup_buffer_reg_channel(reg_uid->registry, + ua_chan, reg_chan, app); + if (ret < 0) { + ERR("Error setting up UST channel \"%s\"", ua_chan->name); + goto error; + } - assert(app); + /* Notify the notification subsystem of the channel's creation. */ + pthread_mutex_lock(®_uid->registry->reg.ust->lock); + chan_reg = ust_registry_channel_find(reg_uid->registry->reg.ust, + ua_chan->tracing_channel_id); + assert(chan_reg); + chan_reg->consumer_key = ua_chan->key; + chan_reg = NULL; + pthread_mutex_unlock(®_uid->registry->reg.ust->lock); + + notification_ret = notification_thread_command_add_channel( + notification_thread_handle, session->name, + ua_sess->euid, ua_sess->egid, + ua_chan->name, + ua_chan->key, + LTTNG_DOMAIN_UST, + ua_chan->attr.subbuf_size * ua_chan->attr.num_subbuf); + if (notification_ret != LTTNG_OK) { + ret = - (int) notification_ret; + ERR("Failed to add channel to notification thread"); + goto error; + } - ret = ustctl_tracer_version(app->sock, &app->version); +send_channel: + /* Send buffers to the application. */ + ret = send_channel_uid_to_ust(reg_chan, app, ua_sess, ua_chan); if (ret < 0) { - if (ret != -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); + if (ret != -ENOTCONN) { + ERR("Error sending channel to application"); } + goto error; } +error: return ret; } /* - * Unregister app by removing it from the global traceable app list and freeing - * the data struct. + * Create and send to the application the created buffers with per PID buffers. * - * The socket is already closed at this point so no close to sock. + * Called with UST app session lock held. + * The session list lock and the session's lock must be acquired. + * + * Return 0 on success else a negative value. */ -void ust_app_unregister(int sock) +static int create_channel_per_pid(struct ust_app *app, + struct ltt_ust_session *usess, struct ust_app_session *ua_sess, + struct ust_app_channel *ua_chan) { - struct ust_app *lta; - struct lttng_ht_node_ulong *node; - struct lttng_ht_iter iter; - struct ust_app_session *ua_sess; int ret; + struct ust_registry_session *registry; + enum lttng_error_code cmd_ret; + struct ltt_session *session; + uint64_t chan_reg_key; + struct ust_registry_channel *chan_reg; - rcu_read_lock(); - - /* Get the node reference for a call_rcu */ - lttng_ht_lookup(ust_app_ht_by_sock, (void *)((unsigned long) sock), &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - assert(node); + assert(app); + assert(usess); + assert(ua_sess); + assert(ua_chan); - lta = caa_container_of(node, struct ust_app, sock_n); - DBG("PID %d unregistering with sock %d", lta->pid, sock); + DBG("UST app creating channel %s with per PID buffers", ua_chan->name); - /* Remove application from PID hash table */ - ret = lttng_ht_del(ust_app_ht_by_sock, &iter); - assert(!ret); + rcu_read_lock(); - /* - * 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); + registry = get_session_registry(ua_sess); + /* The UST app session lock is held, registry shall not be null. */ + assert(registry); - /* - * Ignore return value since the node might have been removed before by an - * add replace during app registration because the PID can be reassigned by - * the OS. - */ - iter.iter.node = <a->pid_n.node; - ret = lttng_ht_del(ust_app_ht, &iter); - if (ret) { - DBG3("Unregister app by PID %d failed. This can happen on pid reuse", - lta->pid); + /* Create and add a new channel registry to session. */ + ret = ust_registry_channel_add(registry, ua_chan->key); + if (ret < 0) { + ERR("Error creating the UST channel \"%s\" registry instance", + ua_chan->name); + goto error; } - /* Remove sessions so they are not visible during deletion.*/ - cds_lfht_for_each_entry(lta->sessions->ht, &iter.iter, ua_sess, - node.node) { - ret = lttng_ht_del(lta->sessions, &iter); - if (ret) { - /* The session was already removed so scheduled for teardown. */ - continue; - } - - /* - * 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); + session = session_find_by_id(ua_sess->tracing_id); + assert(session); - /* - * 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); + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); - cds_list_add(&ua_sess->teardown_node, <a->teardown_head); - pthread_mutex_unlock(&ua_sess->lock); + /* Create and get channel on the consumer side. */ + ret = do_consumer_create_channel(usess, ua_sess, ua_chan, + app->bits_per_long, registry, + session->current_archive_id); + if (ret < 0) { + ERR("Error creating UST channel \"%s\" on the consumer daemon", + ua_chan->name); + goto error; } - /* Free memory */ - call_rcu(<a->pid_n.head, delete_ust_app_rcu); - - rcu_read_unlock(); - return; -} + ret = send_channel_pid_to_ust(app, ua_sess, ua_chan); + if (ret < 0) { + if (ret != -ENOTCONN) { + ERR("Error sending channel to application"); + } + goto error; + } -/* - * Return traceable_app_count - */ -unsigned long ust_app_list_count(void) -{ - unsigned long count; + chan_reg_key = ua_chan->key; + pthread_mutex_lock(®istry->lock); + chan_reg = ust_registry_channel_find(registry, chan_reg_key); + assert(chan_reg); + chan_reg->consumer_key = ua_chan->key; + pthread_mutex_unlock(®istry->lock); + + cmd_ret = notification_thread_command_add_channel( + notification_thread_handle, session->name, + ua_sess->euid, ua_sess->egid, + ua_chan->name, + ua_chan->key, + LTTNG_DOMAIN_UST, + ua_chan->attr.subbuf_size * ua_chan->attr.num_subbuf); + if (cmd_ret != LTTNG_OK) { + ret = - (int) cmd_ret; + ERR("Failed to add channel to notification thread"); + goto error; + } - rcu_read_lock(); - count = lttng_ht_get_count(ust_app_ht); +error: rcu_read_unlock(); - - return count; + return ret; } /* - * Fill events array with all events name of all registered apps. + * From an already allocated ust app channel, create the channel buffers if + * need and send it to the application. This MUST be called with a RCU read + * side lock acquired. + * + * Called with UST app session lock held. + * + * Return 0 on success or else a negative value. Returns -ENOTCONN if + * the application exited concurrently. */ -int ust_app_list_events(struct lttng_event **events) +static int do_create_channel(struct ust_app *app, + struct ltt_ust_session *usess, struct ust_app_session *ua_sess, + struct ust_app_channel *ua_chan) { - int ret, handle; - size_t nbmem, count = 0; - struct lttng_ht_iter iter; - struct ust_app *app; - struct lttng_event *tmp_event; - - nbmem = UST_APP_EVENT_LIST_SIZE; - tmp_event = zmalloc(nbmem * sizeof(struct lttng_event)); - if (tmp_event == NULL) { - PERROR("zmalloc ust app events"); - ret = -ENOMEM; - goto error; - } - - rcu_read_lock(); - - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - struct lttng_ust_tracepoint_iter uiter; + int ret; - health_code_update(); + assert(app); + assert(usess); + assert(ua_sess); + assert(ua_chan); - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; + /* Handle buffer type before sending the channel to the application. */ + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + ret = create_channel_per_uid(app, usess, ua_sess, ua_chan); + if (ret < 0) { + goto error; } - handle = ustctl_tracepoint_list(app->sock); - if (handle < 0) { - if (handle != -EPIPE && handle != -LTTNG_UST_ERR_EXITING) { - ERR("UST app list events getting handle failed for app pid %d", - app->pid); - } - continue; + break; + } + case LTTNG_BUFFER_PER_PID: + { + ret = create_channel_per_pid(app, usess, ua_sess, ua_chan); + if (ret < 0) { + goto error; } + break; + } + default: + assert(0); + ret = -EINVAL; + goto error; + } - while ((ret = ustctl_tracepoint_list_get(app->sock, handle, - &uiter)) != -LTTNG_UST_ERR_NOENT) { - /* 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; + /* 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); - DBG2("Reallocating event list from %zu to %zu entries", nbmem, - 2 * nbmem); - nbmem *= 2; - ptr = realloc(tmp_event, nbmem * sizeof(struct lttng_event)); - if (ptr == NULL) { - PERROR("realloc ust app events"); - free(tmp_event); - ret = -ENOMEM; - goto rcu_error; - } - tmp_event = ptr; - } - memcpy(tmp_event[count].name, uiter.name, LTTNG_UST_SYM_NAME_LEN); - tmp_event[count].loglevel = uiter.loglevel; - tmp_event[count].type = (enum lttng_event_type) LTTNG_UST_TRACEPOINT; - tmp_event[count].pid = app->pid; - tmp_event[count].enabled = -1; - count++; + /* 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; } } - ret = count; - *events = tmp_event; - - DBG2("UST app list events done (%zu events)", count); - -rcu_error: - rcu_read_unlock(); error: - health_code_update(); return ret; } /* - * Fill events array with all events name of all registered apps. + * 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 and RCU read-side lock held. + * + * Return 0 on success or else a negative value. Returns -ENOTCONN if + * the application exited concurrently. */ -int ust_app_list_event_fields(struct lttng_event_field **fields) +static int create_ust_app_channel(struct ust_app_session *ua_sess, + struct ltt_ust_channel *uchan, struct ust_app *app, + enum lttng_ust_chan_type type, struct ltt_ust_session *usess, + struct ust_app_channel **ua_chanp) { - int ret, handle; - size_t nbmem, count = 0; + int ret = 0; struct lttng_ht_iter iter; - struct ust_app *app; - struct lttng_event_field *tmp_event; + struct lttng_ht_node_str *ua_chan_node; + struct ust_app_channel *ua_chan; - nbmem = UST_APP_EVENT_LIST_SIZE; - tmp_event = zmalloc(nbmem * sizeof(struct lttng_event_field)); - if (tmp_event == NULL) { - PERROR("zmalloc ust app event fields"); - ret = -ENOMEM; - goto error; + /* Lookup channel in the ust app session */ + lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter); + ua_chan_node = lttng_ht_iter_get_node_str(&iter); + if (ua_chan_node != NULL) { + ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); + goto end; } - rcu_read_lock(); + ua_chan = alloc_ust_app_channel(uchan->name, ua_sess, &uchan->attr); + if (ua_chan == NULL) { + /* Only malloc can fail here */ + ret = -ENOMEM; + goto error_alloc; + } + shadow_copy_channel(ua_chan, uchan); - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - struct lttng_ust_field_iter uiter; - - health_code_update(); - - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; - } - handle = ustctl_tracepoint_field_list(app->sock); - if (handle < 0) { - if (handle != -EPIPE && handle != -LTTNG_UST_ERR_EXITING) { - ERR("UST app list field getting handle failed for app pid %d", - app->pid); - } - continue; - } - - while ((ret = ustctl_tracepoint_field_list_get(app->sock, handle, - &uiter)) != -LTTNG_UST_ERR_NOENT) { - /* 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; + /* Set channel type. */ + ua_chan->attr.type = type; - DBG2("Reallocating event field list from %zu to %zu entries", nbmem, - 2 * nbmem); - nbmem *= 2; - ptr = realloc(tmp_event, nbmem * sizeof(struct lttng_event_field)); - if (ptr == NULL) { - PERROR("realloc ust app event fields"); - free(tmp_event); - ret = -ENOMEM; - goto rcu_error; - } - tmp_event = ptr; - } + ret = do_create_channel(app, usess, ua_sess, ua_chan); + if (ret < 0) { + goto error; + } - memcpy(tmp_event[count].field_name, uiter.field_name, LTTNG_UST_SYM_NAME_LEN); - tmp_event[count].type = uiter.type; - tmp_event[count].nowrite = uiter.nowrite; + DBG2("UST app create channel %s for PID %d completed", ua_chan->name, + app->pid); - memcpy(tmp_event[count].event.name, uiter.event_name, LTTNG_UST_SYM_NAME_LEN); - tmp_event[count].event.loglevel = uiter.loglevel; - tmp_event[count].event.type = LTTNG_UST_TRACEPOINT; - tmp_event[count].event.pid = app->pid; - tmp_event[count].event.enabled = -1; - count++; - } + /* 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; } - ret = count; - *fields = tmp_event; - - DBG2("UST app list event fields done (%zu events)", count); + /* Everything went well. */ + return 0; -rcu_error: - rcu_read_unlock(); error: - health_code_update(); + delete_ust_app_channel(ua_chan->is_sent ? app->sock : -1, ua_chan, app); +error_alloc: return ret; } /* - * Free and clean all traceable apps of the global list. + * Create UST app event and create it on the tracer side. + * + * Called with ust app session mutex held. */ -void ust_app_clean_list(void) +static +int create_ust_app_event(struct ust_app_session *ua_sess, + struct ust_app_channel *ua_chan, struct ltt_ust_event *uevent, + struct ust_app *app) { - int ret; - struct ust_app *app; - struct lttng_ht_iter iter; - - DBG2("UST app cleaning registered apps hash table"); - - rcu_read_lock(); + int ret = 0; + struct ust_app_event *ua_event; - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = lttng_ht_del(ust_app_ht, &iter); - assert(!ret); - call_rcu(&app->pid_n.head, delete_ust_app_rcu); + /* Get event node */ + ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, + uevent->filter, uevent->attr.loglevel, uevent->exclusion); + if (ua_event != NULL) { + ret = -EEXIST; + goto end; } - /* Cleanup socket hash table */ - cds_lfht_for_each_entry(ust_app_ht_by_sock->ht, &iter.iter, app, - sock_n.node) { - ret = lttng_ht_del(ust_app_ht_by_sock, &iter); - assert(!ret); + /* Does not exist so create one */ + ua_event = alloc_ust_app_event(uevent->attr.name, &uevent->attr); + if (ua_event == NULL) { + /* Only malloc can failed so something is really wrong */ + ret = -ENOMEM; + goto end; } + shadow_copy_event(ua_event, uevent); - /* 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); + /* Create it on the tracer side */ + ret = create_ust_event(app, ua_sess, ua_chan, ua_event); + if (ret < 0) { + /* Not found previously means that it does not exist on the tracer */ + assert(ret != -LTTNG_UST_ERR_EXIST); + goto error; } - /* 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); + add_unique_ust_app_event(ua_chan, ua_event); - rcu_read_unlock(); -} + DBG2("UST app create event %s for PID %d completed", ua_event->name, + app->pid); -/* - * Init UST app hash table. - */ -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); +end: + return ret; + +error: + /* Valid. Calling here is already in a read side lock */ + delete_ust_app_event(-1, ua_event, app); + return ret; } /* - * For a specific UST session, disable the channel for all registered apps. + * Create UST metadata and open it on the tracer side. + * + * Called with UST app session lock held and RCU read side lock. */ -int ust_app_disable_channel_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan) +static int create_ust_app_metadata(struct ust_app_session *ua_sess, + struct ust_app *app, struct consumer_output *consumer) { int ret = 0; - struct lttng_ht_iter iter; - struct lttng_ht_node_str *ua_chan_node; - struct ust_app *app; - struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; + struct ust_app_channel *metadata; + struct consumer_socket *socket; + struct ust_registry_session *registry; + struct ltt_session *session; - if (usess == NULL || uchan == NULL) { - ERR("Disabling UST global channel with NULL values"); - ret = -1; + assert(ua_sess); + assert(app); + assert(consumer); + + registry = get_session_registry(ua_sess); + /* The UST app session is held registry shall not be null. */ + assert(registry); + + pthread_mutex_lock(®istry->lock); + + /* Metadata already exists for this registry or it was closed previously */ + if (registry->metadata_key || registry->metadata_closed) { + ret = 0; goto error; } - DBG2("UST app disabling channel %s from global domain for session id %d", - uchan->name, usess->id); + /* Allocate UST metadata */ + metadata = alloc_ust_app_channel(DEFAULT_METADATA_NAME, ua_sess, NULL); + if (!metadata) { + /* malloc() failed */ + ret = -ENOMEM; + goto error; + } - rcu_read_lock(); + memcpy(&metadata->attr, &ua_sess->metadata_attr, sizeof(metadata->attr)); - /* For every registered applications */ - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - struct lttng_ht_iter uiter; - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; - } - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - continue; - } + /* Need one fd for the channel. */ + ret = lttng_fd_get(LTTNG_FD_APPS, 1); + if (ret < 0) { + ERR("Exhausted number of available FD upon create metadata"); + goto error; + } - /* Get channel */ - lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); - ua_chan_node = lttng_ht_iter_get_node_str(&uiter); - /* If the session if found for the app, the channel must be there */ - assert(ua_chan_node); + /* Get the right consumer socket for the application. */ + socket = consumer_find_socket_by_bitness(app->bits_per_long, consumer); + if (!socket) { + ret = -EINVAL; + goto error_consumer; + } - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - /* The channel must not be already disabled */ - assert(ua_chan->enabled == 1); + /* + * Keep metadata key so we can identify it on the consumer side. Assign it + * to the registry *before* we ask the consumer so we avoid the race of the + * consumer requesting the metadata and the ask_channel call on our side + * did not returned yet. + */ + registry->metadata_key = metadata->key; - /* Disable channel onto application */ - ret = disable_ust_app_channel(ua_sess, ua_chan, app); - if (ret < 0) { - /* XXX: We might want to report this error at some point... */ - continue; - } + session = session_find_by_id(ua_sess->tracing_id); + assert(session); + + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); + + /* + * Ask the metadata channel creation to the consumer. The metadata object + * will be created by the consumer and kept their. However, the stream is + * never added or monitored until we do a first push metadata to the + * consumer. + */ + ret = ust_consumer_ask_channel(ua_sess, metadata, consumer, socket, + registry, session->current_archive_id); + if (ret < 0) { + /* Nullify the metadata key so we don't try to close it later on. */ + registry->metadata_key = 0; + goto error_consumer; } - rcu_read_unlock(); + /* + * 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 = consumer_setup_metadata(socket, metadata->key); + if (ret < 0) { + /* Nullify the metadata key so we don't try to close it later on. */ + registry->metadata_key = 0; + goto error_consumer; + } + + DBG2("UST metadata with key %" PRIu64 " created for app pid %d", + metadata->key, app->pid); +error_consumer: + lttng_fd_put(LTTNG_FD_APPS, 1); + delete_ust_app_channel(-1, metadata, app); error: + pthread_mutex_unlock(®istry->lock); return ret; } /* - * For a specific UST session, enable the channel for all registered apps. + * Return ust app pointer or NULL if not found. RCU read side lock MUST be + * acquired before calling this function. */ -int ust_app_enable_channel_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan) +struct ust_app *ust_app_find_by_pid(pid_t pid) { - int ret = 0; + struct ust_app *app = NULL; + struct lttng_ht_node_ulong *node; struct lttng_ht_iter iter; - struct ust_app *app; - struct ust_app_session *ua_sess; - if (usess == NULL || uchan == NULL) { - ERR("Adding UST global channel to NULL values"); - ret = -1; + 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; } - DBG2("UST app enabling channel %s to global domain for session id %d", - uchan->name, usess->id); + DBG2("Found UST app by pid %d", pid); - rcu_read_lock(); + app = caa_container_of(node, struct ust_app, pid_n); - /* For every registered applications */ - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by +error: + return app; +} + +/* + * 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. + * + * The object is returned on success or else NULL. + */ +struct ust_app *ust_app_create(struct ust_register_msg *msg, int sock) +{ + 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 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"); + goto error; + } + + lta->ppid = msg->ppid; + lta->uid = msg->uid; + lta->gid = msg->gid; + + 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; + lta->sessions = lttng_ht_new(0, LTTNG_HT_TYPE_U64); + lta->ust_objd = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + lta->ust_sessions_objd = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + lta->notify_sock = -1; + + /* Copy name and make sure it's NULL terminated. */ + 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); + lta->sock = sock; + pthread_mutex_init(<a->sock_lock, NULL); + lttng_ht_node_init_ulong(<a->sock_n, (unsigned long) lta->sock); + + CDS_INIT_LIST_HEAD(<a->teardown_head); +error: + return lta; +} + +/* + * 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, &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, &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(); +} + +/* + * 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; + + assert(app); + + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_tracer_version(app->sock, &app->version); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0) { + if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { + ERR("UST app %d version failed with ret %d", app->sock, ret); + } else { + DBG3("UST app %d version failed. Application is dead", app->sock); + } + } + + return ret; +} + +/* + * Unregister app by removing it from the global traceable app list and freeing + * the data struct. + * + * The socket is already closed at this point so no close to sock. + */ +void ust_app_unregister(int sock) +{ + struct ust_app *lta; + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter ust_app_sock_iter; + struct lttng_ht_iter iter; + struct ust_app_session *ua_sess; + int ret; + + rcu_read_lock(); + + /* Get the node reference for a call_rcu */ + lttng_ht_lookup(ust_app_ht_by_sock, (void *)((unsigned long) sock), &ust_app_sock_iter); + node = lttng_ht_iter_get_node_ulong(&ust_app_sock_iter); + assert(node); + + lta = caa_container_of(node, struct ust_app, sock_n); + DBG("PID %d unregistering with sock %d", lta->pid, sock); + + /* + * For per-PID buffers, perform "push metadata" and flush all + * application streams before removing app from hash tables, + * ensuring proper behavior of data_pending check. + * Remove sessions so they are not visible during deletion. + */ + cds_lfht_for_each_entry(lta->sessions->ht, &iter.iter, ua_sess, + node.node) { + struct ust_registry_session *registry; + + ret = lttng_ht_del(lta->sessions, &iter); + if (ret) { + /* The session was already removed so scheduled for teardown. */ + continue; + } + + if (ua_sess->buffer_type == LTTNG_BUFFER_PER_PID) { + (void) ust_app_flush_app_session(lta, ua_sess); + } + + /* + * Add session to list for teardown. This is safe since at this point we + * are the only one using this list. + */ + pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + + /* + * Normally, this is done in the delete session process which is + * executed in the call rcu below. However, upon registration we can't + * 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. + */ + registry = get_session_registry(ua_sess); + if (registry) { + /* Push metadata for application before freeing the application. */ + (void) push_metadata(registry, ua_sess->consumer); + + /* + * Don't ask to close metadata for global per UID buffers. Close + * metadata only on destroy trace session in this case. Also, the + * previous push metadata could have flag the metadata registry to + * close so don't send a close command if closed. + */ + if (ua_sess->buffer_type != LTTNG_BUFFER_PER_UID) { + /* And ask to close it for this session registry. */ + (void) close_metadata(registry, ua_sess->consumer); + } + } + cds_list_add(&ua_sess->teardown_node, <a->teardown_head); + + pthread_mutex_unlock(&ua_sess->lock); + } + + /* Remove application from PID hash table */ + ret = lttng_ht_del(ust_app_ht_by_sock, &ust_app_sock_iter); + assert(!ret); + + /* + * Remove application from notify hash table. The thread handling the + * notify socket could have deleted the node so ignore on error because + * either way it's valid. The close of that socket is handled by the + * apps_notify_thread. + */ + iter.iter.node = <a->notify_sock_n.node; + (void) lttng_ht_del(ust_app_ht_by_notify_sock, &iter); + + /* + * Ignore return value since the node might have been removed before by an + * add replace during app registration because the PID can be reassigned by + * the OS. + */ + iter.iter.node = <a->pid_n.node; + ret = lttng_ht_del(ust_app_ht, &iter); + if (ret) { + DBG3("Unregister app by PID %d failed. This can happen on pid reuse", + lta->pid); + } + + /* Free memory */ + call_rcu(<a->pid_n.head, delete_ust_app_rcu); + + rcu_read_unlock(); + return; +} + +/* + * Fill events array with all events name of all registered apps. + */ +int ust_app_list_events(struct lttng_event **events) +{ + int ret, handle; + size_t nbmem, count = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + struct lttng_event *tmp_event; + + nbmem = UST_APP_EVENT_LIST_SIZE; + tmp_event = zmalloc(nbmem * sizeof(struct lttng_event)); + if (tmp_event == NULL) { + PERROR("zmalloc ust app events"); + ret = -ENOMEM; + goto error; + } + + rcu_read_lock(); + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct lttng_ust_tracepoint_iter uiter; + + health_code_update(); + + if (!app->compatible) { + /* + * TODO: In time, we should notice the caller of this error by + * telling him that this is a version error. + */ + continue; + } + pthread_mutex_lock(&app->sock_lock); + handle = ustctl_tracepoint_list(app->sock); + if (handle < 0) { + if (handle != -EPIPE && handle != -LTTNG_UST_ERR_EXITING) { + ERR("UST app list events getting handle failed for app pid %d", + app->pid); + } + pthread_mutex_unlock(&app->sock_lock); + continue; + } + + while ((ret = ustctl_tracepoint_list_get(app->sock, handle, + &uiter)) != -LTTNG_UST_ERR_NOENT) { + /* Handle ustctl error. */ + if (ret < 0) { + int release_ret; + + if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { + ERR("UST app tp list get failed for app %d with ret %d", + app->sock, ret); + } else { + DBG3("UST app tp list get failed. Application is dead"); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. Continue normal execution. + */ + break; + } + free(tmp_event); + release_ret = ustctl_release_handle(app->sock, handle); + if (release_ret < 0 && + release_ret != -LTTNG_UST_ERR_EXITING && + release_ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, release_ret); + } + pthread_mutex_unlock(&app->sock_lock); + goto rcu_error; + } + + health_code_update(); + if (count >= nbmem) { + /* In case the realloc fails, we free the memory */ + struct lttng_event *new_tmp_event; + size_t new_nbmem; + + new_nbmem = nbmem << 1; + DBG2("Reallocating event list from %zu to %zu entries", + nbmem, new_nbmem); + new_tmp_event = realloc(tmp_event, + new_nbmem * sizeof(struct lttng_event)); + if (new_tmp_event == NULL) { + int release_ret; + + PERROR("realloc ust app events"); + free(tmp_event); + ret = -ENOMEM; + release_ret = ustctl_release_handle(app->sock, handle); + if (release_ret < 0 && + release_ret != -LTTNG_UST_ERR_EXITING && + release_ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, release_ret); + } + pthread_mutex_unlock(&app->sock_lock); + goto rcu_error; + } + /* Zero the new memory */ + memset(new_tmp_event + nbmem, 0, + (new_nbmem - nbmem) * sizeof(struct lttng_event)); + nbmem = new_nbmem; + tmp_event = new_tmp_event; + } + memcpy(tmp_event[count].name, uiter.name, LTTNG_UST_SYM_NAME_LEN); + tmp_event[count].loglevel = uiter.loglevel; + tmp_event[count].type = (enum lttng_event_type) LTTNG_UST_TRACEPOINT; + tmp_event[count].pid = app->pid; + tmp_event[count].enabled = -1; + count++; + } + ret = ustctl_release_handle(app->sock, handle); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0 && ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, ret); + } + } + + ret = count; + *events = tmp_event; + + DBG2("UST app list events done (%zu events)", count); + +rcu_error: + rcu_read_unlock(); +error: + health_code_update(); + return ret; +} + +/* + * Fill events array with all events name of all registered apps. + */ +int ust_app_list_event_fields(struct lttng_event_field **fields) +{ + int ret, handle; + size_t nbmem, count = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + struct lttng_event_field *tmp_event; + + nbmem = UST_APP_EVENT_LIST_SIZE; + tmp_event = zmalloc(nbmem * sizeof(struct lttng_event_field)); + if (tmp_event == NULL) { + PERROR("zmalloc ust app event fields"); + ret = -ENOMEM; + goto error; + } + + rcu_read_lock(); + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct lttng_ust_field_iter uiter; + + health_code_update(); + + if (!app->compatible) { + /* + * TODO: In time, we should notice the caller of this error by + * telling him that this is a version error. + */ + continue; + } + pthread_mutex_lock(&app->sock_lock); + handle = ustctl_tracepoint_field_list(app->sock); + if (handle < 0) { + if (handle != -EPIPE && handle != -LTTNG_UST_ERR_EXITING) { + ERR("UST app list field getting handle failed for app pid %d", + app->pid); + } + pthread_mutex_unlock(&app->sock_lock); + continue; + } + + while ((ret = ustctl_tracepoint_field_list_get(app->sock, handle, + &uiter)) != -LTTNG_UST_ERR_NOENT) { + /* Handle ustctl error. */ + if (ret < 0) { + int release_ret; + + if (ret != -LTTNG_UST_ERR_EXITING && ret != -EPIPE) { + ERR("UST app tp list field failed for app %d with ret %d", + app->sock, ret); + } else { + DBG3("UST app tp list field failed. Application is dead"); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. Reset list and count for next app. + */ + break; + } + free(tmp_event); + release_ret = ustctl_release_handle(app->sock, handle); + pthread_mutex_unlock(&app->sock_lock); + if (release_ret < 0 && + release_ret != -LTTNG_UST_ERR_EXITING && + release_ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, release_ret); + } + goto rcu_error; + } + + health_code_update(); + if (count >= nbmem) { + /* In case the realloc fails, we free the memory */ + struct lttng_event_field *new_tmp_event; + size_t new_nbmem; + + new_nbmem = nbmem << 1; + DBG2("Reallocating event field list from %zu to %zu entries", + nbmem, new_nbmem); + new_tmp_event = realloc(tmp_event, + new_nbmem * sizeof(struct lttng_event_field)); + if (new_tmp_event == NULL) { + int release_ret; + + PERROR("realloc ust app event fields"); + free(tmp_event); + ret = -ENOMEM; + release_ret = ustctl_release_handle(app->sock, handle); + pthread_mutex_unlock(&app->sock_lock); + if (release_ret && + release_ret != -LTTNG_UST_ERR_EXITING && + release_ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, release_ret); + } + goto rcu_error; + } + /* Zero the new memory */ + memset(new_tmp_event + nbmem, 0, + (new_nbmem - nbmem) * sizeof(struct lttng_event_field)); + nbmem = new_nbmem; + tmp_event = new_tmp_event; + } + + memcpy(tmp_event[count].field_name, uiter.field_name, LTTNG_UST_SYM_NAME_LEN); + /* Mapping between these enums matches 1 to 1. */ + tmp_event[count].type = (enum lttng_event_field_type) uiter.type; + tmp_event[count].nowrite = uiter.nowrite; + + memcpy(tmp_event[count].event.name, uiter.event_name, LTTNG_UST_SYM_NAME_LEN); + tmp_event[count].event.loglevel = uiter.loglevel; + tmp_event[count].event.type = LTTNG_EVENT_TRACEPOINT; + tmp_event[count].event.pid = app->pid; + tmp_event[count].event.enabled = -1; + count++; + } + ret = ustctl_release_handle(app->sock, handle); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0 && + ret != -LTTNG_UST_ERR_EXITING && + ret != -EPIPE) { + ERR("Error releasing app handle for app %d with ret %d", app->sock, ret); + } + } + + ret = count; + *fields = tmp_event; + + DBG2("UST app list event fields done (%zu events)", count); + +rcu_error: + rcu_read_unlock(); +error: + health_code_update(); + return ret; +} + +/* + * Free and clean all traceable apps of the global list. + * + * Should _NOT_ be called with RCU read-side lock held. + */ +void ust_app_clean_list(void) +{ + int ret; + struct ust_app *app; + struct lttng_ht_iter iter; + + DBG2("UST app cleaning registered apps hash table"); + + rcu_read_lock(); + + if (ust_app_ht) { + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ret = lttng_ht_del(ust_app_ht, &iter); + assert(!ret); + call_rcu(&app->pid_n.head, delete_ust_app_rcu); + } + } + + /* Cleanup socket hash table */ + if (ust_app_ht_by_sock) { + cds_lfht_for_each_entry(ust_app_ht_by_sock->ht, &iter.iter, app, + sock_n.node) { + ret = lttng_ht_del(ust_app_ht_by_sock, &iter); + assert(!ret); + } + } + + /* Cleanup notify socket hash table */ + if (ust_app_ht_by_notify_sock) { + cds_lfht_for_each_entry(ust_app_ht_by_notify_sock->ht, &iter.iter, app, + notify_sock_n.node) { + ret = lttng_ht_del(ust_app_ht_by_notify_sock, &iter); + assert(!ret); + } + } + rcu_read_unlock(); + + /* Destroy is done only when the ht is empty */ + if (ust_app_ht) { + ht_cleanup_push(ust_app_ht); + } + if (ust_app_ht_by_sock) { + ht_cleanup_push(ust_app_ht_by_sock); + } + if (ust_app_ht_by_notify_sock) { + ht_cleanup_push(ust_app_ht_by_notify_sock); + } +} + +/* + * Init UST app hash table. + */ +int ust_app_ht_alloc(void) +{ + ust_app_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!ust_app_ht) { + return -1; + } + ust_app_ht_by_sock = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!ust_app_ht_by_sock) { + return -1; + } + ust_app_ht_by_notify_sock = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); + if (!ust_app_ht_by_notify_sock) { + return -1; + } + return 0; +} + +/* + * For a specific UST session, disable the channel for all registered apps. + */ +int ust_app_disable_channel_glb(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct lttng_ht_node_str *ua_chan_node; + struct ust_app *app; + struct ust_app_session *ua_sess; + struct ust_app_channel *ua_chan; + + if (usess == NULL || uchan == NULL) { + ERR("Disabling UST global channel with NULL values"); + ret = -1; + goto error; + } + + DBG2("UST app disabling channel %s from global domain for session id %" PRIu64, + uchan->name, usess->id); + + rcu_read_lock(); + + /* For every registered applications */ + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct lttng_ht_iter uiter; + if (!app->compatible) { + /* + * TODO: In time, we should notice the caller of this error by + * telling him that this is a version error. + */ + continue; + } + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + continue; + } + + /* Get channel */ + lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); + ua_chan_node = lttng_ht_iter_get_node_str(&uiter); + /* If the session if found for the app, the channel must be there */ + assert(ua_chan_node); + + ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); + /* The channel must not be already disabled */ + assert(ua_chan->enabled == 1); + + /* Disable channel onto application */ + ret = disable_ust_app_channel(ua_sess, ua_chan, app); + if (ret < 0) { + /* XXX: We might want to report this error at some point... */ + continue; + } + } + + rcu_read_unlock(); + +error: + return ret; +} + +/* + * For a specific UST session, enable the channel for all registered apps. + */ +int ust_app_enable_channel_glb(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + struct ust_app_session *ua_sess; + + if (usess == NULL || uchan == NULL) { + ERR("Adding UST global channel to NULL values"); + ret = -1; + goto error; + } + + DBG2("UST app enabling channel %s to global domain for session id %" PRIu64, + uchan->name, usess->id); + + rcu_read_lock(); + + /* For every registered applications */ + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + if (!app->compatible) { + /* + * TODO: In time, we should notice the caller of this error by + * telling him that this is a version error. + */ + continue; + } + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + continue; + } + + /* Enable channel onto application */ + ret = enable_ust_app_channel(ua_sess, uchan, app); + if (ret < 0) { + /* XXX: We might want to report this error at some point... */ + continue; + } + } + + rcu_read_unlock(); + +error: + return ret; +} + +/* + * Disable an event in a channel and for a specific session. + */ +int ust_app_disable_event_glb(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent) +{ + int ret = 0; + struct lttng_ht_iter iter, uiter; + struct lttng_ht_node_str *ua_chan_node; + struct ust_app *app; + struct ust_app_session *ua_sess; + struct ust_app_channel *ua_chan; + struct ust_app_event *ua_event; + + DBG("UST app disabling event %s for all apps in channel " + "%s for session id %" PRIu64, + uevent->attr.name, uchan->name, usess->id); + + rcu_read_lock(); + + /* For all registered applications */ + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + if (!app->compatible) { + /* + * TODO: In time, we should notice the caller of this error by + * telling him that this is a version error. + */ + continue; + } + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + /* Next app */ + 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 (ua_chan_node == NULL) { + DBG2("Channel %s not found in session id %" PRIu64 " for app pid %d." + "Skipping", uchan->name, usess->id, app->pid); + continue; + } + ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); + + ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, + uevent->filter, uevent->attr.loglevel, + uevent->exclusion); + if (ua_event == NULL) { + DBG2("Event %s not found in channel %s for app pid %d." + "Skipping", uevent->attr.name, uchan->name, app->pid); + continue; + } + + ret = disable_ust_app_event(ua_sess, ua_event, app); + if (ret < 0) { + /* XXX: Report error someday... */ + continue; + } + } + + rcu_read_unlock(); + + return ret; +} + +/* + * For a specific UST session, create the channel for all registered apps. + */ +int ust_app_create_channel_glb(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan) +{ + int ret = 0, created; + struct lttng_ht_iter iter; + struct ust_app *app; + struct ust_app_session *ua_sess = NULL; + + /* Very wrong code flow */ + assert(usess); + assert(uchan); + + DBG2("UST app adding channel %s to UST domain for session id %" PRIu64, + uchan->name, usess->id); + + rcu_read_lock(); + + /* For every registered applications */ + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + if (!app->compatible) { + /* + * TODO: In time, we should notice the caller of this error by + * telling him that this is a version error. + */ + continue; + } + 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 + * app session. + */ + ret = find_or_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. + */ + ret = 0; /* Not an error. */ + continue; + case -ENOMEM: + default: + goto error_rcu_unlock; + } + } + 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); + ret = 0; + } else { + /* Create channel onto application. We don't need the chan ref. */ + ret = create_ust_app_channel(ua_sess, uchan, app, + LTTNG_UST_CHAN_PER_CPU, usess, NULL); + } + pthread_mutex_unlock(&ua_sess->lock); + if (ret < 0) { + /* Cleanup the created session if it's the case. */ + if (created) { + destroy_app_session(app, ua_sess); + } + switch (ret) { + case -ENOTCONN: + /* + * The application's socket is not valid. Either a bad socket + * or a timeout on it. We can't inform the caller that for a + * specific app, the session failed so lets continue here. + */ + ret = 0; /* Not an error. */ + continue; + case -ENOMEM: + default: + goto error_rcu_unlock; + } + } + } + +error_rcu_unlock: + rcu_read_unlock(); + return ret; +} + +/* + * Enable event for a specific session and channel on the tracer. + */ +int ust_app_enable_event_glb(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent) +{ + int ret = 0; + struct lttng_ht_iter iter, uiter; + struct lttng_ht_node_str *ua_chan_node; + struct ust_app *app; + struct ust_app_session *ua_sess; + struct ust_app_channel *ua_chan; + struct ust_app_event *ua_event; + + DBG("UST app enabling event %s for all apps for session id %" PRIu64, + uevent->attr.name, usess->id); + + /* + * NOTE: At this point, this function is called only if the session and + * channel passed are already created for all apps. and enabled on the + * tracer also. + */ + + rcu_read_lock(); + + /* For all registered applications */ + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + if (!app->compatible) { + /* + * TODO: In time, we should notice the caller of this error by + * telling him that this is a version error. + */ + continue; + } + ua_sess = lookup_session_by_app(usess, app); + if (!ua_sess) { + /* The application has problem or is probably dead. */ + continue; + } + + 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); + /* + * It is possible that the channel cannot be found is + * the channel/event creation occurs concurrently with + * an application exit. + */ + if (!ua_chan_node) { + pthread_mutex_unlock(&ua_sess->lock); + continue; + } + + ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); + + /* Get event node */ + ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, + uevent->filter, uevent->attr.loglevel, uevent->exclusion); + if (ua_event == NULL) { + DBG3("UST app enable event %s not found for app PID %d." + "Skipping app", uevent->attr.name, app->pid); + 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: + rcu_read_unlock(); + return ret; +} + +/* + * For a specific existing UST session and UST channel, creates the event for + * all registered apps. + */ +int ust_app_create_event_glb(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent) +{ + int ret = 0; + struct lttng_ht_iter iter, uiter; + struct lttng_ht_node_str *ua_chan_node; + struct ust_app *app; + struct ust_app_session *ua_sess; + struct ust_app_channel *ua_chan; + + DBG("UST app creating event %s for all apps for session id %" PRIu64, + uevent->attr.name, usess->id); + + rcu_read_lock(); + + /* For all registered applications */ + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + if (!app->compatible) { + /* + * TODO: In time, we should notice the caller of this error by * telling him that this is a version error. */ - continue; + continue; + } + ua_sess = lookup_session_by_app(usess, app); + if (!ua_sess) { + /* The application has problem or is probably dead. */ + continue; + } + + 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); + + 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! */ + break; + } + DBG2("UST app event %s already exist on app PID %d", + uevent->attr.name, app->pid); + continue; + } + } + + rcu_read_unlock(); + + return ret; +} + +/* + * Start tracing for a specific UST session and app. + * + * Called with UST app session lock held. + * + */ +static +int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) +{ + int ret = 0; + struct ust_app_session *ua_sess; + + DBG("Starting tracing for ust app pid %d", app->pid); + + rcu_read_lock(); + + if (!app->compatible) { + goto end; + } + + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + /* The session is in teardown process. Ignore and continue. */ + goto end; + } + + pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } + + /* Upon restart, we skip the setup, already done */ + if (ua_sess->started) { + goto skip_setup; + } + + /* Create directories if consumer is LOCAL and has a path defined. */ + if (usess->consumer->type == CONSUMER_DST_LOCAL && + usess->consumer->dst.session_root_path[0] != '\0') { + char *tmp_path; + + tmp_path = zmalloc(LTTNG_PATH_MAX); + if (!tmp_path) { + ERR("Alloc tmp_path"); + goto error_unlock; + } + ret = snprintf(tmp_path, LTTNG_PATH_MAX, "%s%s%s", + usess->consumer->dst.session_root_path, + usess->consumer->chunk_path, + usess->consumer->subdir); + if (ret >= LTTNG_PATH_MAX) { + ERR("Local destination path exceeds the maximal allowed length of %i bytes (needs %i bytes) with path = \"%s%s%s\"", + LTTNG_PATH_MAX, ret, + usess->consumer->dst.session_root_path, + usess->consumer->chunk_path, + usess->consumer->subdir); + free(tmp_path); + goto error_unlock; + } + + DBG("Creating directory path for local tracing: \"%s\"", + tmp_path); + ret = run_as_mkdir_recursive(tmp_path, S_IRWXU | S_IRWXG, + ua_sess->euid, ua_sess->egid); + free(tmp_path); + if (ret < 0) { + if (errno != EEXIST) { + ERR("Trace directory creation error"); + goto error_unlock; + } + } + } + + /* + * Create the metadata for the application. This returns gracefully if a + * metadata was already set for the session. + */ + ret = create_ust_app_metadata(ua_sess, app, usess->consumer); + if (ret < 0) { + goto error_unlock; + } + + health_code_update(); + +skip_setup: + /* This start the UST tracing */ + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_start_session(app->sock, ua_sess->handle); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("Error starting tracing for app pid: %d (ret: %d)", + app->pid, ret); + } else { + DBG("UST app start session failed. Application is dead."); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } + goto error_unlock; + } + + /* Indicate that the session has been started once */ + ua_sess->started = 1; + + pthread_mutex_unlock(&ua_sess->lock); + + health_code_update(); + + /* Quiescent wait after starting trace */ + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_wait_quiescent(app->sock); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app wait quiescent failed for app pid %d ret %d", + app->pid, ret); + } + +end: + rcu_read_unlock(); + health_code_update(); + return 0; + +error_unlock: + pthread_mutex_unlock(&ua_sess->lock); + rcu_read_unlock(); + health_code_update(); + return -1; +} + +/* + * Stop tracing for a specific UST session and app. + */ +static +int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) +{ + int ret = 0; + struct ust_app_session *ua_sess; + struct ust_registry_session *registry; + + DBG("Stopping tracing for ust app pid %d", app->pid); + + rcu_read_lock(); + + if (!app->compatible) { + goto end_no_session; + } + + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + goto end_no_session; + } + + 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 + * from either the application manager thread or the command thread. Simply + * indicate that this is a stop error. + */ + if (!ua_sess->started) { + goto error_rcu_unlock; + } + + health_code_update(); + + /* This inhibits UST tracing */ + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_stop_session(app->sock, ua_sess->handle); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("Error stopping tracing for app pid: %d (ret: %d)", + app->pid, ret); + } else { + DBG("UST app stop session failed. Application is dead."); + /* + * This is normal behavior, an application can die during the + * creation process. Don't report an error so the execution can + * continue normally. + */ + goto end_unlock; } - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - continue; + goto error_rcu_unlock; + } + + health_code_update(); + + /* Quiescent wait after stopping trace */ + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_wait_quiescent(app->sock); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app wait quiescent failed for app pid %d ret %d", + app->pid, ret); + } + + health_code_update(); + + registry = get_session_registry(ua_sess); + + /* The UST app session is held registry shall not be null. */ + assert(registry); + + /* Push metadata for application before freeing the application. */ + (void) push_metadata(registry, ua_sess->consumer); + +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); +end_no_session: + rcu_read_unlock(); + health_code_update(); + return 0; + +error_rcu_unlock: + pthread_mutex_unlock(&ua_sess->lock); + rcu_read_unlock(); + health_code_update(); + return -1; +} + +static +int ust_app_flush_app_session(struct ust_app *app, + struct ust_app_session *ua_sess) +{ + int ret, retval = 0; + struct lttng_ht_iter iter; + struct ust_app_channel *ua_chan; + struct consumer_socket *socket; + + DBG("Flushing app session buffers for ust app pid %d", app->pid); + + rcu_read_lock(); + + if (!app->compatible) { + goto end_not_compatible; + } + + pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + goto end_deleted; + } + + health_code_update(); + + /* Flushing buffers */ + socket = consumer_find_socket_by_bitness(app->bits_per_long, + ua_sess->consumer); + + /* Flush buffers and push metadata. */ + switch (ua_sess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, + node.node) { + health_code_update(); + ret = consumer_flush_channel(socket, ua_chan->key); + if (ret) { + ERR("Error flushing consumer channel"); + retval = -1; + continue; + } } + break; + case LTTNG_BUFFER_PER_UID: + default: + assert(0); + break; + } - /* Enable channel onto application */ - ret = enable_ust_app_channel(ua_sess, uchan, app); - if (ret < 0) { - /* XXX: We might want to report this error at some point... */ - continue; + health_code_update(); + +end_deleted: + pthread_mutex_unlock(&ua_sess->lock); + +end_not_compatible: + rcu_read_unlock(); + health_code_update(); + return retval; +} + +/* + * Flush buffers for all applications for a specific UST session. + * Called with UST session lock held. + */ +static +int ust_app_flush_session(struct ltt_ust_session *usess) + +{ + int ret = 0; + + DBG("Flushing session buffers for all ust apps"); + + rcu_read_lock(); + + /* Flush buffers and push metadata. */ + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + struct buffer_reg_uid *reg; + struct lttng_ht_iter iter; + + /* Flush all per UID buffers associated to that session. */ + cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct ust_registry_session *ust_session_reg; + struct buffer_reg_channel *reg_chan; + struct consumer_socket *socket; + + /* Get consumer socket to use to push the metadata.*/ + socket = consumer_find_socket_by_bitness(reg->bits_per_long, + usess->consumer); + if (!socket) { + /* Ignore request if no consumer is found for the session. */ + continue; + } + + cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, + reg_chan, node.node) { + /* + * The following call will print error values so the return + * code is of little importance because whatever happens, we + * have to try them all. + */ + (void) consumer_flush_channel(socket, reg_chan->consumer_key); + } + + ust_session_reg = reg->registry->reg.ust; + /* Push metadata. */ + (void) push_metadata(ust_session_reg, usess->consumer); + } + break; + } + case LTTNG_BUFFER_PER_PID: + { + struct ust_app_session *ua_sess; + struct lttng_ht_iter iter; + struct ust_app *app; + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + continue; + } + (void) ust_app_flush_app_session(app, ua_sess); + } + break; + } + default: + ret = -1; + assert(0); + break; + } + + rcu_read_unlock(); + health_code_update(); + return ret; +} + +static +int ust_app_clear_quiescent_app_session(struct ust_app *app, + struct ust_app_session *ua_sess) +{ + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app_channel *ua_chan; + struct consumer_socket *socket; + + DBG("Clearing stream quiescent state for ust app pid %d", app->pid); + + rcu_read_lock(); + + if (!app->compatible) { + goto end_not_compatible; + } + + pthread_mutex_lock(&ua_sess->lock); + + if (ua_sess->deleted) { + goto end_unlock; + } + + health_code_update(); + + socket = consumer_find_socket_by_bitness(app->bits_per_long, + ua_sess->consumer); + if (!socket) { + ERR("Failed to find consumer (%" PRIu32 ") socket", + app->bits_per_long); + ret = -1; + goto end_unlock; + } + + /* Clear quiescent state. */ + switch (ua_sess->buffer_type) { + case LTTNG_BUFFER_PER_PID: + cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, + ua_chan, node.node) { + health_code_update(); + ret = consumer_clear_quiescent_channel(socket, + ua_chan->key); + if (ret) { + ERR("Error clearing quiescent state for consumer channel"); + ret = -1; + continue; + } } + break; + case LTTNG_BUFFER_PER_UID: + default: + assert(0); + ret = -1; + break; } + health_code_update(); + +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); + +end_not_compatible: rcu_read_unlock(); + health_code_update(); + return ret; +} -error: +/* + * Clear quiescent state in each stream for all applications for a + * specific UST session. + * Called with UST session lock held. + */ +static +int ust_app_clear_quiescent_session(struct ltt_ust_session *usess) + +{ + int ret = 0; + + DBG("Clearing stream quiescent state for all ust apps"); + + rcu_read_lock(); + + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + struct lttng_ht_iter iter; + struct buffer_reg_uid *reg; + + /* + * Clear quiescent for all per UID buffers associated to + * that session. + */ + cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct consumer_socket *socket; + struct buffer_reg_channel *reg_chan; + + /* Get associated consumer socket.*/ + socket = consumer_find_socket_by_bitness( + reg->bits_per_long, usess->consumer); + if (!socket) { + /* + * Ignore request if no consumer is found for + * the session. + */ + continue; + } + + cds_lfht_for_each_entry(reg->registry->channels->ht, + &iter.iter, reg_chan, node.node) { + /* + * The following call will print error values so + * the return code is of little importance + * because whatever happens, we have to try them + * all. + */ + (void) consumer_clear_quiescent_channel(socket, + reg_chan->consumer_key); + } + } + break; + } + case LTTNG_BUFFER_PER_PID: + { + struct ust_app_session *ua_sess; + struct lttng_ht_iter iter; + struct ust_app *app; + + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, + pid_n.node) { + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + continue; + } + (void) ust_app_clear_quiescent_app_session(app, + ua_sess); + } + break; + } + default: + ret = -1; + assert(0); + break; + } + + rcu_read_unlock(); + health_code_update(); return ret; } /* - * Disable an event in a channel and for a specific session. + * Destroy a specific UST session in apps. + */ +static int destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) +{ + int ret; + struct ust_app_session *ua_sess; + struct lttng_ht_iter iter; + struct lttng_ht_node_u64 *node; + + DBG("Destroy tracing for ust app pid %d", app->pid); + + rcu_read_lock(); + + if (!app->compatible) { + goto end; + } + + __lookup_session_by_app(usess, app, &iter); + node = lttng_ht_iter_get_node_u64(&iter); + if (node == NULL) { + /* Session is being or is deleted. */ + goto end; + } + ua_sess = caa_container_of(node, struct ust_app_session, node); + + health_code_update(); + destroy_app_session(app, ua_sess); + + health_code_update(); + + /* Quiescent wait after stopping trace */ + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_wait_quiescent(app->sock); + pthread_mutex_unlock(&app->sock_lock); + if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app wait quiescent failed for app pid %d ret %d", + app->pid, ret); + } +end: + rcu_read_unlock(); + health_code_update(); + return 0; +} + +/* + * Start tracing for the UST session. */ -int ust_app_disable_event_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent) +int ust_app_start_trace_all(struct ltt_ust_session *usess) { int ret = 0; - struct lttng_ht_iter iter, uiter; - struct lttng_ht_node_str *ua_chan_node, *ua_event_node; + struct lttng_ht_iter iter; struct ust_app *app; - struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; - struct ust_app_event *ua_event; - DBG("UST app disabling event %s for all apps in channel " - "%s for session id %d", uevent->attr.name, uchan->name, usess->id); + DBG("Starting all UST traces"); rcu_read_lock(); - /* For all registered applications */ - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; - } - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - /* Next app */ - 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 (ua_chan_node == NULL) { - DBG2("Channel %s not found in session id %d for app pid %d." - "Skipping", uchan->name, usess->id, app->pid); - continue; - } - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - - lttng_ht_lookup(ua_chan->events, (void *)uevent->attr.name, &uiter); - ua_event_node = lttng_ht_iter_get_node_str(&uiter); - if (ua_event_node == NULL) { - DBG2("Event %s not found in channel %s for app pid %d." - "Skipping", uevent->attr.name, uchan->name, app->pid); - continue; - } - ua_event = caa_container_of(ua_event_node, struct ust_app_event, node); + /* + * In a start-stop-start use-case, we need to clear the quiescent state + * of each channel set by the prior stop command, thus ensuring that a + * following stop or destroy is sure to grab a timestamp_end near those + * operations, even if the packet is empty. + */ + (void) ust_app_clear_quiescent_session(usess); - ret = disable_ust_app_event(ua_sess, ua_event, app); + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + ret = ust_app_start_trace(usess, app); if (ret < 0) { - /* XXX: Report error someday... */ + /* Continue to next apps even on error */ continue; } } rcu_read_unlock(); - return ret; + return 0; } /* - * For a specific UST session and UST channel, the event for all - * registered apps. + * Start tracing for the UST session. + * Called with UST session lock held. */ -int ust_app_disable_all_event_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan) +int ust_app_stop_trace_all(struct ltt_ust_session *usess) { int ret = 0; - struct lttng_ht_iter iter, uiter; - struct lttng_ht_node_str *ua_chan_node; + struct lttng_ht_iter iter; struct ust_app *app; - struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; - struct ust_app_event *ua_event; - DBG("UST app disabling all event for all apps in channel " - "%s for session id %d", uchan->name, usess->id); + DBG("Stopping all UST traces"); rcu_read_lock(); - /* For all registered applications */ cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; - } - ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ + ret = ust_app_stop_trace(usess, app); + if (ret < 0) { + /* Continue to next apps even on error */ continue; } - - /* Lookup channel in the ust app session */ - lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter); - ua_chan_node = lttng_ht_iter_get_node_str(&uiter); - /* If the channel is not found, there is a code flow error */ - assert(ua_chan_node); - - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - - /* Disable each events of channel */ - cds_lfht_for_each_entry(ua_chan->events->ht, &uiter.iter, ua_event, - node.node) { - ret = disable_ust_app_event(ua_sess, ua_event, app); - if (ret < 0) { - /* XXX: Report error someday... */ - continue; - } - } } + (void) ust_app_flush_session(usess); + rcu_read_unlock(); - return ret; + return 0; } /* - * For a specific UST session, create the channel for all registered apps. + * Destroy app UST session. */ -int ust_app_create_channel_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan) +int ust_app_destroy_trace_all(struct ltt_ust_session *usess) { - int ret = 0, created; + int ret = 0; struct lttng_ht_iter iter; struct ust_app *app; - struct ust_app_session *ua_sess = NULL; - - /* Very wrong code flow */ - assert(usess); - assert(uchan); - DBG2("UST app adding channel %s to global domain for session id %d", - uchan->name, usess->id); + DBG("Destroy all UST traces"); rcu_read_lock(); - /* For every registered applications */ cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; - } - /* - * Create session on the tracer side and add it to app session HT. Note - * that if session exist, it will simply return a pointer to the ust - * app session. - */ - ret = create_ust_app_session(usess, app, &ua_sess, &created); - 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, usess->consumer, - LTTNG_UST_CHAN_PER_CPU, NULL); - pthread_mutex_unlock(&ua_sess->lock); + ret = destroy_trace(usess, app); 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); - } + /* Continue to next apps even on error */ + continue; } } -error_rcu_unlock: rcu_read_unlock(); - return ret; + + return 0; } -/* - * Enable event for a specific session and channel on the tracer. - */ -int ust_app_enable_event_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent) +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 lttng_ht_node_str *ua_chan_node; - 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; + int is_created = 0; - DBG("UST app enabling event %s for all apps for session id %d", - uevent->attr.name, usess->id); + ret = find_or_create_ust_app_session(usess, app, &ua_sess, &is_created); + if (ret < 0) { + /* Tracer is probably gone or ENOMEM. */ + goto error; + } + if (!is_created) { + /* App session already created. */ + goto end; + } + assert(ua_sess); - /* - * NOTE: At this point, this function is called only if the session and - * channel passed are already created for all apps. and enabled on the - * tracer also. - */ + pthread_mutex_lock(&ua_sess->lock); - rcu_read_lock(); + if (ua_sess->deleted) { + pthread_mutex_unlock(&ua_sess->lock); + goto end; + } - /* For all registered applications */ - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { + /* + * 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 = do_create_channel(app, usess, ua_sess, ua_chan); + if (ret < 0 && ret != -ENOTCONN) { /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. + * 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. */ - continue; + goto error_unlock; } - ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ - continue; + + /* + * Add context using the list so they are enabled in the same order the + * user added them. + */ + cds_list_for_each_entry(ua_ctx, &ua_chan->ctx_list, list) { + ret = create_ust_channel_context(ua_chan, ua_ctx, app); + if (ret < 0) { + goto error_unlock; + } } - 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 the channel is not found, there is a code flow error */ - assert(ua_chan_node); + /* For each events */ + cds_lfht_for_each_entry(ua_chan->events->ht, &uiter.iter, ua_event, + node.node) { + ret = create_ust_event(app, ua_sess, ua_chan, ua_event); + if (ret < 0) { + goto error_unlock; + } + } + } + + pthread_mutex_unlock(&ua_sess->lock); + + if (usess->active) { + ret = ust_app_start_trace(usess, app); + if (ret < 0) { + goto error; + } + + DBG2("UST trace started for app pid %d", app->pid); + } +end: + /* Everything went well at this point. */ + return; + +error_unlock: + pthread_mutex_unlock(&ua_sess->lock); +error: + if (ua_sess) { + destroy_app_session(app, ua_sess); + } + return; +} + +static +void ust_app_global_destroy(struct ltt_ust_session *usess, struct ust_app *app) +{ + struct ust_app_session *ua_sess; + + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + return; + } + destroy_app_session(app, ua_sess); +} + +/* + * Add channels/events from UST global domain to registered apps at sock. + * + * 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); - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); + DBG2("UST app global update for app sock %d for session id %" PRIu64, + app->sock, usess->id); - /* Get event node */ - ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel); - if (ua_event == NULL) { - DBG3("UST app enable event %s not found for app PID %d." - "Skipping app", uevent->attr.name, app->pid); - goto next_app; - } + if (!app->compatible) { + return; + } - 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); + if (trace_ust_pid_tracker_lookup(usess, app->pid)) { + ust_app_global_create(usess, app); + } else { + ust_app_global_destroy(usess, app); } +} -error: +/* + * 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(); - return ret; } /* - * For a specific existing UST session and UST channel, creates the event for - * all registered apps. + * Add context to a specific channel for global UST domain. */ -int ust_app_create_event_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent) +int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan, struct ltt_ust_context *uctx) { int ret = 0; - struct lttng_ht_iter iter, uiter; struct lttng_ht_node_str *ua_chan_node; - struct ust_app *app; + struct lttng_ht_iter iter, uiter; + struct ust_app_channel *ua_chan = NULL; struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; - - DBG("UST app creating event %s for all apps for session id %d", - uevent->attr.name, usess->id); + struct ust_app *app; rcu_read_lock(); - /* For all registered applications */ cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { if (!app->compatible) { /* @@ -2768,627 +5210,881 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess, continue; } ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ + if (ua_sess == NULL) { continue; } 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); - - 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 (ua_chan_node == NULL) { + goto next_app; + } + ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, + node); + ret = create_ust_app_channel_context(ua_chan, &uctx->ctx, app); if (ret < 0) { - if (ret != -LTTNG_UST_ERR_EXIST) { - /* Possible value at this point: -ENOMEM. If so, we stop! */ - break; - } - DBG2("UST app event %s already exist on app PID %d", - uevent->attr.name, app->pid); - continue; + goto next_app; } + next_app: + pthread_mutex_unlock(&ua_sess->lock); } rcu_read_unlock(); - return ret; } /* - * Start tracing for a specific UST session and app. + * Enable event for a channel from a UST session for a specific PID. */ -int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app) +int ust_app_enable_event_pid(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent, pid_t pid) { int ret = 0; + struct lttng_ht_iter iter; + struct lttng_ht_node_str *ua_chan_node; + struct ust_app *app; struct ust_app_session *ua_sess; + struct ust_app_channel *ua_chan; + struct ust_app_event *ua_event; - DBG("Starting tracing for ust app pid %d", app->pid); + DBG("UST app enabling event %s for PID %d", uevent->attr.name, pid); rcu_read_lock(); + app = ust_app_find_by_pid(pid); + if (app == NULL) { + ERR("UST app enable event per PID %d not found", pid); + ret = -1; + goto end; + } + if (!app->compatible) { + ret = 0; goto end; } ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - /* The session is in teardown process. Ignore and continue. */ + if (!ua_sess) { + /* The application has problem or is probably dead. */ + ret = 0; goto end; } pthread_mutex_lock(&ua_sess->lock); - /* Upon restart, we skip the setup, already done */ - if (ua_sess->started) { - goto skip_setup; + if (ua_sess->deleted) { + ret = 0; + goto end_unlock; } - /* Create directories if consumer is LOCAL and has a path defined. */ - if (usess->consumer->type == CONSUMER_DST_LOCAL && - strlen(usess->consumer->dst.trace_path) > 0) { - ret = run_as_mkdir_recursive(usess->consumer->dst.trace_path, - S_IRWXU | S_IRWXG, usess->uid, usess->gid); + /* Lookup channel in the ust app session */ + lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter); + ua_chan_node = lttng_ht_iter_get_node_str(&iter); + /* If the channel is not found, there is a code flow error */ + assert(ua_chan_node); + + ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); + + ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, + uevent->filter, uevent->attr.loglevel, uevent->exclusion); + if (ua_event == NULL) { + ret = create_ust_app_event(ua_sess, ua_chan, uevent, app); if (ret < 0) { - if (ret != -EEXIST) { - ERR("Trace directory creation error"); - goto error_unlock; - } + goto end_unlock; + } + } else { + ret = enable_ust_app_event(ua_sess, ua_event, app); + if (ret < 0) { + goto end_unlock; } } - /* Create the metadata for the application. */ - ret = create_ust_app_metadata(ua_sess, app, usess->consumer); - if (ret < 0) { - goto error_unlock; - } +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); +end: + rcu_read_unlock(); + return ret; +} - health_code_update(); +/* + * 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; -skip_setup: - /* This start the UST tracing */ - ret = ustctl_start_session(app->sock, ua_sess->handle); + 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) { - 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."); + 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_unlock; + 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 session object using the application object and the + * session object descriptor has a key. If not found, NULL is returned. + * A RCU read side lock MUST be acquired when calling this function. +*/ +static struct ust_app_session *find_session_by_objd(struct ust_app *app, + int objd) +{ + struct lttng_ht_node_ulong *node; + struct lttng_ht_iter iter; + struct ust_app_session *ua_sess = NULL; + + assert(app); + + lttng_ht_lookup(app->ust_sessions_objd, (void *)((unsigned long) objd), &iter); + node = lttng_ht_iter_get_node_ulong(&iter); + if (node == NULL) { + DBG2("UST app session find by objd %d not found", objd); + goto error; } - /* Indicate that the session has been started once */ - ua_sess->started = 1; + ua_sess = caa_container_of(node, struct ust_app_session, ust_objd_node); - pthread_mutex_unlock(&ua_sess->lock); +error: + return ua_sess; +} - health_code_update(); +/* + * 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; - /* 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); + 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; } -end: - rcu_read_unlock(); - health_code_update(); - return 0; + ua_chan = caa_container_of(node, struct ust_app_channel, ust_objd_node); -error_unlock: - pthread_mutex_unlock(&ua_sess->lock); - rcu_read_unlock(); - health_code_update(); - return -1; +error: + return ua_chan; } /* - * Stop tracing for a specific UST session and app. + * 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. */ -int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app) +static int reply_ust_register_channel(int sock, int cobjd, + size_t nr_fields, struct ustctl_field *fields) { - int ret = 0; - struct lttng_ht_iter iter; - struct ust_app_session *ua_sess; + int ret, ret_code = 0; + uint32_t chan_id, reg_count; + uint64_t chan_reg_key; + enum ustctl_channel_header type; + struct ust_app *app; struct ust_app_channel *ua_chan; - - DBG("Stopping tracing for ust app pid %d", app->pid); + struct ust_app_session *ua_sess; + struct ust_registry_session *registry; + struct ust_registry_channel *chan_reg; rcu_read_lock(); - if (!app->compatible) { - goto end_no_session; + /* 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 torn down. Abort event notify", + sock); + ret = 0; + goto error_rcu_unlock; } - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - goto end_no_session; + /* Lookup channel by UST object descriptor. */ + ua_chan = find_channel_by_objd(app, cobjd); + if (!ua_chan) { + DBG("Application channel is being torn down. Abort event notify"); + ret = 0; + goto error_rcu_unlock; } - pthread_mutex_lock(&ua_sess->lock); + assert(ua_chan->session); + ua_sess = ua_chan->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 - * from either the application manager thread or the command thread. Simply - * indicate that this is a stop error. - */ - if (!ua_sess->started) { + /* Get right session registry depending on the session buffer type. */ + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Abort event notify"); + ret = 0; goto error_rcu_unlock; + }; + + /* Depending on the buffer type, a different channel key is used. */ + if (ua_sess->buffer_type == LTTNG_BUFFER_PER_UID) { + chan_reg_key = ua_chan->tracing_channel_id; + } else { + chan_reg_key = ua_chan->key; } - health_code_update(); + pthread_mutex_lock(®istry->lock); - /* This inhibits UST tracing */ - ret = ustctl_stop_session(app->sock, ua_sess->handle); - if (ret < 0) { - if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { - ERR("Error stopping tracing for app pid: %d (ret: %d)", - app->pid, ret); + chan_reg = ust_registry_channel_find(registry, chan_reg_key); + assert(chan_reg); + + if (!chan_reg->register_done) { + reg_count = ust_registry_get_event_count(chan_reg); + if (reg_count < 31) { + type = USTCTL_CHANNEL_HEADER_COMPACT; } else { - DBG("UST app stop session failed. Application is dead."); + type = USTCTL_CHANNEL_HEADER_LARGE; } - goto error_rcu_unlock; - } - - health_code_update(); - /* Quiescent wait after stopping trace */ - ret = ustctl_wait_quiescent(app->sock); - if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { - ERR("UST app wait quiescent failed for app pid %d ret %d", - app->pid, ret); + chan_reg->nr_ctx_fields = nr_fields; + chan_reg->ctx_fields = fields; + fields = NULL; + chan_reg->header_type = type; + } else { + /* Get current already assigned values. */ + type = chan_reg->header_type; } + /* Channel id is set during the object creation. */ + chan_id = chan_reg->chan_id; - health_code_update(); - - /* Flushing buffers */ - cds_lfht_for_each_entry(ua_sess->channels->ht, &iter.iter, ua_chan, - node.node) { - health_code_update(); - assert(ua_chan->is_sent); - ret = ustctl_sock_flush_buffer(app->sock, ua_chan->obj); - if (ret < 0) { - if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { - ERR("UST app PID %d channel %s flush failed with ret %d", - app->pid, ua_chan->name, ret); - } else { - DBG3("UST app failed to flush %s. Application is dead.", - ua_chan->name); - /* No need to continue. */ - break; - } - /* Continuing flushing all buffers */ - continue; + /* Append to metadata */ + if (!chan_reg->metadata_dumped) { + ret_code = ust_metadata_channel_statedump(registry, chan_reg); + if (ret_code) { + ERR("Error appending channel metadata (errno = %d)", ret_code); + goto reply; } } - health_code_update(); +reply: + DBG3("UST app replying to register channel key %" PRIu64 + " with id %u, type: %d, ret: %d", chan_reg_key, chan_id, type, + ret_code); - ret = push_metadata(app, ua_sess); + ret = ustctl_reply_register_channel(sock, chan_id, type, ret_code); if (ret < 0) { - goto error_rcu_unlock; + 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; } - pthread_mutex_unlock(&ua_sess->lock); -end_no_session: - rcu_read_unlock(); - health_code_update(); - return 0; + /* This channel registry registration is completed. */ + chan_reg->register_done = 1; +error: + pthread_mutex_unlock(®istry->lock); error_rcu_unlock: - pthread_mutex_unlock(&ua_sess->lock); rcu_read_unlock(); - health_code_update(); - return -1; + free(fields); + return ret; } /* - * Destroy a specific UST session in apps. + * 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 destroy_trace(struct ltt_ust_session *usess, struct ust_app *app) +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_value, char *model_emf_uri) { - int ret; + int ret, ret_code; + uint32_t event_id = 0; + uint64_t chan_reg_key; + struct ust_app *app; + struct ust_app_channel *ua_chan; struct ust_app_session *ua_sess; - struct lttng_ht_iter iter; - struct lttng_ht_node_ulong *node; - - DBG("Destroy tracing for ust app pid %d", app->pid); + struct ust_registry_session *registry; rcu_read_lock(); - if (!app->compatible) { - goto end; + /* 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 torn down. Abort event notify", + sock); + ret = 0; + goto error_rcu_unlock; } - __lookup_session_by_app(usess, app, &iter); - node = lttng_ht_iter_get_node_ulong(&iter); - if (node == NULL) { - /* Session is being or is deleted. */ - goto end; + /* Lookup channel by UST object descriptor. */ + ua_chan = find_channel_by_objd(app, cobjd); + if (!ua_chan) { + DBG("Application channel is being torn down. Abort event notify"); + ret = 0; + goto error_rcu_unlock; } - ua_sess = caa_container_of(node, struct ust_app_session, node); - - health_code_update(); - destroy_app_session(app, ua_sess); - health_code_update(); + assert(ua_chan->session); + ua_sess = ua_chan->session; - /* Quiescent wait after stopping trace */ - ret = ustctl_wait_quiescent(app->sock); - if (ret < 0 && ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { - ERR("UST app wait quiescent failed for app pid %d ret %d", - app->pid, ret); + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Abort event notify"); + ret = 0; + goto error_rcu_unlock; } -end: - rcu_read_unlock(); - health_code_update(); - return 0; -} -/* - * Start tracing for the UST session. - */ -int ust_app_start_trace_all(struct ltt_ust_session *usess) -{ - int ret = 0; - struct lttng_ht_iter iter; - struct ust_app *app; + if (ua_sess->buffer_type == LTTNG_BUFFER_PER_UID) { + chan_reg_key = ua_chan->tracing_channel_id; + } else { + chan_reg_key = ua_chan->key; + } - DBG("Starting all UST traces"); + pthread_mutex_lock(®istry->lock); - rcu_read_lock(); + /* + * From this point on, this call acquires the ownership of the sig, fields + * and model_emf_uri meaning any free are done inside it if needed. These + * three variables MUST NOT be read/write after this. + */ + ret_code = ust_registry_create_event(registry, chan_reg_key, + sobjd, cobjd, name, sig, nr_fields, fields, + loglevel_value, model_emf_uri, ua_sess->buffer_type, + &event_id, app); + sig = NULL; + fields = NULL; + model_emf_uri = NULL; - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = ust_app_start_trace(usess, app); - if (ret < 0) { - /* Continue to next apps even on error */ - continue; + /* + * 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; } - rcu_read_unlock(); + DBG3("UST registry event %s with id %" PRId32 " added successfully", + name, event_id); - return 0; +error: + pthread_mutex_unlock(®istry->lock); +error_rcu_unlock: + rcu_read_unlock(); + free(sig); + free(fields); + free(model_emf_uri); + return ret; } /* - * Start tracing for the UST session. + * Add enum to the UST session registry. Once done, this replies to the + * application with the appropriate error code. + * + * The session UST registry lock is acquired within this function. + * + * On success 0 is returned else a negative value. */ -int ust_app_stop_trace_all(struct ltt_ust_session *usess) +static int add_enum_ust_registry(int sock, int sobjd, char *name, + struct ustctl_enum_entry *entries, size_t nr_entries) { - int ret = 0; - struct lttng_ht_iter iter; + int ret = 0, ret_code; struct ust_app *app; - - DBG("Stopping all UST traces"); + struct ust_app_session *ua_sess; + struct ust_registry_session *registry; + uint64_t enum_id = -1ULL; rcu_read_lock(); - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = ust_app_stop_trace(usess, app); - if (ret < 0) { - ERR("UST app stop trace failed with ret %d", ret); - /* Continue to next apps even on error */ - continue; - } + /* Lookup application. If not found, there is a code flow error. */ + app = find_app_by_notify_sock(sock); + if (!app) { + /* Return an error since this is not an error */ + DBG("Application socket %d is being torn down. Aborting enum registration", + sock); + free(entries); + goto error_rcu_unlock; } - rcu_read_unlock(); - - return 0; -} + /* Lookup session by UST object descriptor. */ + ua_sess = find_session_by_objd(app, sobjd); + if (!ua_sess) { + /* Return an error since this is not an error */ + DBG("Application session is being torn down (session not found). Aborting enum registration."); + free(entries); + goto error_rcu_unlock; + } -/* - * Destroy app UST session. - */ -int ust_app_destroy_trace_all(struct ltt_ust_session *usess) -{ - int ret = 0; - struct lttng_ht_iter iter; - struct ust_app *app; + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down (registry not found). Aborting enum registration."); + free(entries); + goto error_rcu_unlock; + } - DBG("Destroy all UST traces"); + pthread_mutex_lock(®istry->lock); - rcu_read_lock(); + /* + * From this point on, the callee acquires the ownership of + * entries. The variable entries MUST NOT be read/written after + * call. + */ + ret_code = ust_registry_create_or_find_enum(registry, sobjd, name, + entries, nr_entries, &enum_id); + entries = NULL; - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - ret = destroy_trace(usess, app); - if (ret < 0) { - /* Continue to next apps even on error */ - continue; + /* + * 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_enum(sock, enum_id, ret_code); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app reply enum failed with ret %d", ret); + } else { + DBG3("UST app reply enum failed. Application died"); } + /* + * No need to wipe the create enum since the application socket will + * get close on error hence cleaning up everything by itself. + */ + goto error; } + DBG3("UST registry enum %s added successfully or already found", name); + +error: + pthread_mutex_unlock(®istry->lock); +error_rcu_unlock: rcu_read_unlock(); - - return 0; + return ret; } /* - * Add channels/events from UST global domain to registered apps at sock. + * Handle application notification through the given notify socket. + * + * Return 0 on success or else a negative value. */ -void ust_app_global_update(struct ltt_ust_session *usess, int sock) +int ust_app_recv_notify(int sock) { - int ret = 0; - struct lttng_ht_iter iter, uiter, iter_ctx; - struct ust_app *app; - struct ust_app_session *ua_sess = NULL; - struct ust_app_channel *ua_chan; - struct ust_app_event *ua_event; - struct ust_app_ctx *ua_ctx; - - assert(usess); - assert(sock >= 0); - - DBG2("UST app global update for app sock %d for session id %d", sock, - usess->id); - - rcu_read_lock(); - - app = find_app_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; - } + int ret; + enum ustctl_notify_cmd cmd; - if (!app->compatible) { - goto error; - } + DBG3("UST app receiving notify from sock %d", sock); - ret = create_ust_app_session(usess, app, &ua_sess, NULL); + ret = ustctl_recv_notify(sock, &cmd); if (ret < 0) { - /* Tracer is probably gone or ENOMEM. */ + 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; } - assert(ua_sess); - pthread_mutex_lock(&ua_sess->lock); + switch (cmd) { + case USTCTL_NOTIFY_CMD_EVENT: + { + int sobjd, cobjd, loglevel_value; + char name[LTTNG_UST_SYM_NAME_LEN], *sig, *model_emf_uri; + size_t nr_fields; + struct ustctl_field *fields; - /* - * 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, usess->consumer); - if (ret < 0) { - /* - * Stop everything. On error, the application failed, no more file - * descriptor are available or ENOMEM so stopping here is the only - * thing we can do for now. - */ - goto error_unlock; - } + DBG2("UST app ustctl register event received"); - 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) { - goto error_unlock; + ret = ustctl_recv_register_event(sock, &sobjd, &cobjd, name, + &loglevel_value, &sig, &nr_fields, &fields, + &model_emf_uri); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app recv event failed with ret %d", ret); + } else { + DBG3("UST app recv event failed. Application died"); } + goto error; } - - /* For each events */ - cds_lfht_for_each_entry(ua_chan->events->ht, &uiter.iter, ua_event, - node.node) { - ret = create_ust_event(app, ua_sess, ua_chan, ua_event); - if (ret < 0) { - goto error_unlock; - } + /* + * Add event to the UST registry coming from the notify socket. This + * call will free if needed the sig, fields and model_emf_uri. This + * code path loses the ownsership of these variables and transfer them + * to the this function. + */ + ret = add_event_ust_registry(sock, sobjd, cobjd, name, sig, nr_fields, + fields, loglevel_value, 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; - pthread_mutex_unlock(&ua_sess->lock); + DBG2("UST app ustctl register channel received"); - if (usess->start_trace) { - ret = ust_app_start_trace(usess, app); + 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; } - DBG2("UST trace started for app pid %d", app->pid); - } - - /* Everything went well at this point. */ - rcu_read_unlock(); - return; + /* + * The fields ownership are transfered to this function call meaning + * that if needed it will be freed. After this, it's invalid to access + * fields or clean it up. + */ + ret = reply_ust_register_channel(sock, cobjd, nr_fields, + fields); + if (ret < 0) { + goto error; + } -error_unlock: - pthread_mutex_unlock(&ua_sess->lock); -error: - if (ua_sess) { - destroy_app_session(app, ua_sess); + break; } - rcu_read_unlock(); - return; -} - -/* - * Add context to a specific channel for global UST domain. - */ -int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan, struct ltt_ust_context *uctx) -{ - int ret = 0; - struct lttng_ht_node_str *ua_chan_node; - struct lttng_ht_iter iter, uiter; - struct ust_app_channel *ua_chan = NULL; - struct ust_app_session *ua_sess; - struct ust_app *app; + case USTCTL_NOTIFY_CMD_ENUM: + { + int sobjd; + char name[LTTNG_UST_SYM_NAME_LEN]; + size_t nr_entries; + struct ustctl_enum_entry *entries; - rcu_read_lock(); + DBG2("UST app ustctl register enum received"); - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; - } - ua_sess = lookup_session_by_app(usess, app); - if (ua_sess == NULL) { - continue; + ret = ustctl_recv_register_enum(sock, &sobjd, name, + &entries, &nr_entries); + if (ret < 0) { + if (ret != -EPIPE && ret != -LTTNG_UST_ERR_EXITING) { + ERR("UST app recv enum failed with ret %d", ret); + } else { + DBG3("UST app recv enum failed. Application died"); + } + goto error; } - 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) { - 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); + /* Callee assumes ownership of entries */ + ret = add_enum_ust_registry(sock, sobjd, name, + entries, nr_entries); if (ret < 0) { - goto next_app; + goto error; } - next_app: - pthread_mutex_unlock(&ua_sess->lock); + + break; + } + default: + /* Should NEVER happen. */ + assert(0); } - rcu_read_unlock(); +error: return ret; } /* - * Enable event for a channel from a UST session for a specific PID. + * 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. */ -int ust_app_enable_event_pid(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent, pid_t pid) +void ust_app_notify_sock_unregister(int sock) { - int ret = 0; + int err_enomem = 0; struct lttng_ht_iter iter; - struct lttng_ht_node_str *ua_chan_node; struct ust_app *app; - struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; - struct ust_app_event *ua_event; + struct ust_app_notify_sock_obj *obj; - DBG("UST app enabling event %s for PID %d", uevent->attr.name, pid); + assert(sock >= 0); rcu_read_lock(); - app = ust_app_find_by_pid(pid); - if (app == NULL) { - ERR("UST app enable event per PID %d not found", pid); - ret = -1; - goto end; + 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; } - if (!app->compatible) { - ret = 0; - goto end; - } + DBG("UST app notify socket unregister %d", sock); - ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ - ret = 0; - goto end; + /* + * 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; } - 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); - /* If the channel is not found, there is a code flow error */ - assert(ua_chan_node); + iter.iter.node = &app->notify_sock_n.node; - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, 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); + } +} - ua_event = find_ust_app_event(ua_chan->events, uevent->attr.name, - uevent->filter, uevent->attr.loglevel); - if (ua_event == NULL) { - ret = create_ust_app_event(ua_sess, ua_chan, uevent, app); - if (ret < 0) { - goto end_unlock; - } - } else { - ret = enable_ust_app_event(ua_sess, ua_event, app); - if (ret < 0) { - goto end_unlock; - } +/* + * Destroy a ust app data structure and free its memory. + */ +void ust_app_destroy(struct ust_app *app) +{ + if (!app) { + return; } -end_unlock: - pthread_mutex_unlock(&ua_sess->lock); -end: - rcu_read_unlock(); - return ret; + call_rcu(&app->pid_n.head, delete_ust_app_rcu); } /* - * Disable event for a channel from a UST session for a specific PID. + * Take a snapshot for a given UST session. The snapshot is sent to the given + * output. + * + * Return 0 on success or else a negative value. */ -int ust_app_disable_event_pid(struct ltt_ust_session *usess, - struct ltt_ust_channel *uchan, struct ltt_ust_event *uevent, pid_t pid) +int ust_app_snapshot_record(struct ltt_ust_session *usess, + struct snapshot_output *output, int wait, + uint64_t nb_packets_per_stream) { int ret = 0; struct lttng_ht_iter iter; - struct lttng_ht_node_str *ua_chan_node, *ua_event_node; struct ust_app *app; - struct ust_app_session *ua_sess; - struct ust_app_channel *ua_chan; - struct ust_app_event *ua_event; + char pathname[PATH_MAX]; + struct ltt_session *session; + uint64_t trace_archive_id; - DBG("UST app disabling event %s for PID %d", uevent->attr.name, pid); + assert(usess); + assert(output); rcu_read_lock(); - app = ust_app_find_by_pid(pid); - if (app == NULL) { - ERR("UST app disable event per PID %d not found", pid); - ret = -1; - goto error; - } + session = session_find_by_id(usess->id); + assert(session); + assert(pthread_mutex_trylock(&session->lock)); + assert(session_trylock_list()); + trace_archive_id = session->current_archive_id; - if (!app->compatible) { - ret = 0; - goto error; - } + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + struct buffer_reg_uid *reg; - ua_sess = lookup_session_by_app(usess, app); - if (!ua_sess) { - /* The application has problem or is probably dead. */ - goto error; - } + cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct buffer_reg_channel *reg_chan; + struct consumer_socket *socket; - /* Lookup channel in the ust app session */ - lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter); - ua_chan_node = lttng_ht_iter_get_node_str(&iter); - if (ua_chan_node == NULL) { - /* Channel does not exist, skip disabling */ - goto error; - } - ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); + if (!reg->registry->reg.ust->metadata_key) { + /* Skip since no metadata is present */ + continue; + } - lttng_ht_lookup(ua_chan->events, (void *)uevent->attr.name, &iter); - ua_event_node = lttng_ht_iter_get_node_str(&iter); - if (ua_event_node == NULL) { - /* Event does not exist, skip disabling */ - goto error; + /* Get consumer socket to use to push the metadata.*/ + socket = consumer_find_socket_by_bitness(reg->bits_per_long, + usess->consumer); + if (!socket) { + ret = -EINVAL; + goto error; + } + + memset(pathname, 0, sizeof(pathname)); + ret = snprintf(pathname, sizeof(pathname), + DEFAULT_UST_TRACE_DIR "/" DEFAULT_UST_TRACE_UID_PATH, + reg->uid, reg->bits_per_long); + if (ret < 0) { + PERROR("snprintf snapshot path"); + goto error; + } + + /* Add the UST default trace dir to path. */ + cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, + reg_chan, node.node) { + ret = consumer_snapshot_channel(socket, + reg_chan->consumer_key, + output, 0, usess->uid, + usess->gid, pathname, wait, + nb_packets_per_stream, + trace_archive_id); + if (ret < 0) { + goto error; + } + } + ret = consumer_snapshot_channel(socket, + reg->registry->reg.ust->metadata_key, output, 1, + usess->uid, usess->gid, pathname, wait, 0, + trace_archive_id); + if (ret < 0) { + goto error; + } + } + break; } - ua_event = caa_container_of(ua_event_node, struct ust_app_event, node); + case LTTNG_BUFFER_PER_PID: + { + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct consumer_socket *socket; + struct lttng_ht_iter chan_iter; + struct ust_app_channel *ua_chan; + struct ust_app_session *ua_sess; + struct ust_registry_session *registry; + + ua_sess = lookup_session_by_app(usess, app); + if (!ua_sess) { + /* Session not associated with this app. */ + continue; + } - ret = disable_ust_app_event(ua_sess, ua_event, app); - if (ret < 0) { - goto error; + /* Get the right consumer socket for the application. */ + socket = consumer_find_socket_by_bitness(app->bits_per_long, + output->consumer); + if (!socket) { + ret = -EINVAL; + goto error; + } + + /* Add the UST default trace dir to path. */ + memset(pathname, 0, sizeof(pathname)); + ret = snprintf(pathname, sizeof(pathname), DEFAULT_UST_TRACE_DIR "/%s", + ua_sess->path); + if (ret < 0) { + PERROR("snprintf snapshot path"); + goto error; + } + + cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, + ua_chan, node.node) { + ret = consumer_snapshot_channel(socket, + ua_chan->key, output, + 0, ua_sess->euid, ua_sess->egid, + pathname, wait, + nb_packets_per_stream, + trace_archive_id); + if (ret < 0) { + goto error; + } + } + + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Abort snapshot record."); + ret = -1; + goto error; + } + ret = consumer_snapshot_channel(socket, + registry->metadata_key, output, + 1, ua_sess->euid, ua_sess->egid, + pathname, wait, 0, + trace_archive_id); + if (ret < 0) { + goto error; + } + } + break; + } + default: + assert(0); + break; } error: @@ -3397,453 +6093,432 @@ error: } /* - * Calibrate registered applications. + * Return the size taken by one more packet per stream. */ -int ust_app_calibrate_glb(struct lttng_ust_calibrate *calibrate) +uint64_t ust_app_get_size_one_more_packet_per_stream(struct ltt_ust_session *usess, + uint64_t cur_nr_packets) { - int ret = 0; - struct lttng_ht_iter iter; + uint64_t tot_size = 0; struct ust_app *app; + struct lttng_ht_iter iter; - rcu_read_lock(); + assert(usess); - cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { - if (!app->compatible) { - /* - * TODO: In time, we should notice the caller of this error by - * telling him that this is a version error. - */ - continue; + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: + { + struct buffer_reg_uid *reg; + + cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct buffer_reg_channel *reg_chan; + + rcu_read_lock(); + cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, + reg_chan, node.node) { + if (cur_nr_packets >= reg_chan->num_subbuf) { + /* + * Don't take channel into account if we + * already grab all its packets. + */ + continue; + } + tot_size += reg_chan->subbuf_size * reg_chan->stream_count; + } + rcu_read_unlock(); } + break; + } + case LTTNG_BUFFER_PER_PID: + { + rcu_read_lock(); + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct ust_app_channel *ua_chan; + struct ust_app_session *ua_sess; + struct lttng_ht_iter chan_iter; + + ua_sess = lookup_session_by_app(usess, app); + if (!ua_sess) { + /* Session not associated with this app. */ + continue; + } - health_code_update(); - - ret = ustctl_calibrate(app->sock, calibrate); - if (ret < 0) { - switch (ret) { - case -ENOSYS: - /* Means that it's not implemented on the tracer side. */ - ret = 0; - break; - default: - DBG2("Calibrate app PID %d returned with error %d", - app->pid, ret); - break; + cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, + ua_chan, node.node) { + if (cur_nr_packets >= ua_chan->attr.num_subbuf) { + /* + * Don't take channel into account if we + * already grab all its packets. + */ + continue; + } + tot_size += ua_chan->attr.subbuf_size * ua_chan->streams.count; } } + rcu_read_unlock(); + break; + } + default: + assert(0); + break; } - DBG("UST app global domain calibration finished"); - - rcu_read_unlock(); - - health_code_update(); - - return ret; + return tot_size; } -/* - * 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 ust_app_uid_get_channel_runtime_stats(uint64_t ust_session_id, + struct cds_list_head *buffer_reg_uid_list, + struct consumer_output *consumer, uint64_t uchan_id, + int overwrite, uint64_t *discarded, uint64_t *lost) { int ret; - uint32_t pid, ppid, uid, gid; + uint64_t consumer_chan_key; - assert(msg); + *discarded = 0; + *lost = 0; - 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); + ret = buffer_reg_uid_consumer_channel_key( + buffer_reg_uid_list, uchan_id, &consumer_chan_key); 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; + /* Not found */ + ret = 0; + goto end; } - msg->pid = (pid_t) pid; - msg->ppid = (pid_t) ppid; - msg->uid = (uid_t) uid; - msg->gid = (gid_t) gid; -error: + if (overwrite) { + ret = consumer_get_lost_packets(ust_session_id, + consumer_chan_key, consumer, lost); + } else { + ret = consumer_get_discarded_events(ust_session_id, + consumer_chan_key, consumer, discarded); + } + +end: return ret; } -/* - * 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) +int ust_app_pid_get_channel_runtime_stats(struct ltt_ust_session *usess, + struct ltt_ust_channel *uchan, + struct consumer_output *consumer, int overwrite, + uint64_t *discarded, uint64_t *lost) { - struct lttng_ht_node_ulong *node; + int ret = 0; struct lttng_ht_iter iter; - struct ust_app_channel *ua_chan = NULL; + struct lttng_ht_node_str *ua_chan_node; + struct ust_app *app; + struct ust_app_session *ua_sess; + struct ust_app_channel *ua_chan; - assert(app); + *discarded = 0; + *lost = 0; - 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; - } + rcu_read_lock(); + /* + * Iterate over every registered applications. Sum counters for + * all applications containing requested session and channel. + */ + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct lttng_ht_iter uiter; - ua_chan = caa_container_of(node, struct ust_app_channel, ust_objd_node); + ua_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + continue; + } -error: - return ua_chan; -} + /* Get channel */ + lttng_ht_lookup(ua_sess->channels, (void *) uchan->name, &uiter); + ua_chan_node = lttng_ht_iter_get_node_str(&uiter); + /* If the session is found for the app, the channel must be there */ + assert(ua_chan_node); -/* - * 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; + ua_chan = caa_container_of(ua_chan_node, struct ust_app_channel, node); - rcu_read_lock(); + if (overwrite) { + uint64_t _lost; - /* 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; + ret = consumer_get_lost_packets(usess->id, ua_chan->key, + consumer, &_lost); + if (ret < 0) { + break; + } + (*lost) += _lost; + } else { + uint64_t _discarded; + + ret = consumer_get_discarded_events(usess->id, + ua_chan->key, consumer, &_discarded); + if (ret < 0) { + break; + } + (*discarded) += _discarded; + } } - /* 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); + rcu_read_unlock(); + return ret; +} - pthread_mutex_lock(&ua_sess->registry.lock); +static +int ust_app_regenerate_statedump(struct ltt_ust_session *usess, + struct ust_app *app) +{ + int ret = 0; + struct ust_app_session *ua_sess; - if (ust_registry_is_max_id(ua_chan->session->registry.used_channel_id)) { - ret_code = -1; - chan_id = -1U; - type = -1; - goto reply; - } + DBG("Regenerating the metadata for ust app pid %d", app->pid); - /* 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); - } + rcu_read_lock(); - 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_sess = lookup_session_by_app(usess, app); + if (ua_sess == NULL) { + /* The session is in teardown process. Ignore and continue. */ + goto end; } - 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; + pthread_mutex_lock(&ua_sess->lock); - /* 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; - } + if (ua_sess->deleted) { + goto end_unlock; } -reply: - DBG3("UST app replying to register channel with id %u, type: %d, ret: %d", - chan_id, type, ret_code); + pthread_mutex_lock(&app->sock_lock); + ret = ustctl_regenerate_statedump(app->sock, ua_sess->handle); + pthread_mutex_unlock(&app->sock_lock); - 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; - } +end_unlock: + pthread_mutex_unlock(&ua_sess->lock); -error: - pthread_mutex_unlock(&ua_sess->registry.lock); -error_rcu_unlock: +end: rcu_read_unlock(); + health_code_update(); 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. + * Regenerate the statedump for each app in the session. */ -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 ust_app_regenerate_statedump_all(struct ltt_ust_session *usess) { - int ret, ret_code; - uint32_t event_id = 0; + int ret = 0; + struct lttng_ht_iter iter; 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; + DBG("Regenerating the metadata for all UST apps"); - pthread_mutex_lock(&ua_sess->registry.lock); + rcu_read_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); + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + if (!app->compatible) { + continue; + } - /* - * 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"); + ret = ust_app_regenerate_statedump(usess, app); + if (ret < 0) { + /* Continue to the next app even on error */ + continue; } - /* - * 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; + + return 0; } /* - * Handle application notification through the given notify socket. + * Rotate all the channels of a session. * * Return 0 on success or else a negative value. */ -int ust_app_recv_notify(int sock) +int ust_app_rotate_session(struct ltt_session *session, bool *ust_active) { - int ret; - enum ustctl_notify_cmd cmd; + int ret = 0; + struct lttng_ht_iter iter; + struct ust_app *app; + struct ltt_ust_session *usess = session->ust_session; + char pathname[LTTNG_PATH_MAX]; - DBG3("UST app receiving notify from sock %d", sock); + assert(usess); - 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; - } + rcu_read_lock(); - switch (cmd) { - case USTCTL_NOTIFY_CMD_EVENT: + switch (usess->buffer_type) { + case LTTNG_BUFFER_PER_UID: { - int sobjd, cobjd, loglevel; - char name[LTTNG_UST_SYM_NAME_LEN], *sig, *model_emf_uri; - size_t nr_fields; - struct ustctl_field *fields; + struct buffer_reg_uid *reg; + + cds_list_for_each_entry(reg, &usess->buffer_reg_uid_list, lnode) { + struct buffer_reg_channel *reg_chan; + struct consumer_socket *socket; + + /* Get consumer socket to use to push the metadata.*/ + socket = consumer_find_socket_by_bitness(reg->bits_per_long, + usess->consumer); + if (!socket) { + ret = -EINVAL; + goto error; + } - DBG2("UST app ustctl register event received"); + /* + * Account the metadata channel first to make sure the + * number of channels waiting for a rotation cannot + * reach 0 before we complete the iteration over all + * the channels. + */ + ret = rotate_add_channel_pending( + reg->registry->reg.ust->metadata_key, + LTTNG_DOMAIN_UST, session); + if (ret < 0) { + ret = reg->bits_per_long == 32 ? + -LTTNG_ERR_UST_CONSUMER32_FAIL : + -LTTNG_ERR_UST_CONSUMER64_FAIL; + goto error; + } - ret = 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"); + ret = snprintf(pathname, sizeof(pathname), + DEFAULT_UST_TRACE_DIR "/" DEFAULT_UST_TRACE_UID_PATH, + reg->uid, reg->bits_per_long); + if (ret < 0 || ret == sizeof(pathname)) { + PERROR("Failed to format rotation path"); + goto error; } - 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; - } + /* Rotate the data channels. */ + cds_lfht_for_each_entry(reg->registry->channels->ht, &iter.iter, + reg_chan, node.node) { + ret = rotate_add_channel_pending( + reg_chan->consumer_key, + LTTNG_DOMAIN_UST, session); + if (ret < 0) { + ret = reg->bits_per_long == 32 ? + -LTTNG_ERR_UST_CONSUMER32_FAIL : + -LTTNG_ERR_UST_CONSUMER64_FAIL; + goto error; + } + ret = consumer_rotate_channel(socket, + reg_chan->consumer_key, + usess->uid, usess->gid, + usess->consumer, pathname, + /* is_metadata_channel */ false, + session->current_archive_id, + &session->rotate_pending_relay); + if (ret < 0) { + goto error; + } + } + (void) push_metadata(reg->registry->reg.ust, usess->consumer); + + ret = consumer_rotate_channel(socket, + reg->registry->reg.ust->metadata_key, + usess->uid, usess->gid, + usess->consumer, pathname, + /* is_metadata_channel */ true, + session->current_archive_id, + &session->rotate_pending_relay); + if (ret < 0) { + goto error; + } + *ust_active = true; + } break; } - case USTCTL_NOTIFY_CMD_CHANNEL: + case LTTNG_BUFFER_PER_PID: { - int sobjd, cobjd; - size_t nr_fields; - struct ustctl_field *fields; + cds_lfht_for_each_entry(ust_app_ht->ht, &iter.iter, app, pid_n.node) { + struct consumer_socket *socket; + struct lttng_ht_iter chan_iter; + struct ust_app_channel *ua_chan; + struct ust_app_session *ua_sess; + struct ust_registry_session *registry; + + ua_sess = lookup_session_by_app(usess, app); + if (!ua_sess) { + /* Session not associated with this app. */ + continue; + } + ret = snprintf(pathname, sizeof(pathname), + DEFAULT_UST_TRACE_DIR "/%s", + ua_sess->path); + if (ret < 0 || ret == sizeof(pathname)) { + PERROR("Failed to format rotation path"); + goto error; + } - DBG2("UST app ustctl register channel received"); + /* Get the right consumer socket for the application. */ + socket = consumer_find_socket_by_bitness(app->bits_per_long, + usess->consumer); + if (!socket) { + ret = -EINVAL; + goto error; + } - 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"); + registry = get_session_registry(ua_sess); + if (!registry) { + DBG("Application session is being torn down. Abort snapshot record."); + ret = -1; + goto error; } - goto error; - } - ret = reply_ust_register_channel(sock, sobjd, cobjd, nr_fields, - fields); - if (ret < 0) { - goto error; - } + /* + * Account the metadata channel first to make sure the + * number of channels waiting for a rotation cannot + * reach 0 before we complete the iteration over all + * the channels. + */ + ret = rotate_add_channel_pending(registry->metadata_key, + LTTNG_DOMAIN_UST, session); + if (ret < 0) { + ret = app->bits_per_long == 32 ? + -LTTNG_ERR_UST_CONSUMER32_FAIL : + -LTTNG_ERR_UST_CONSUMER64_FAIL; + goto error; + } + /* Rotate the data channels. */ + cds_lfht_for_each_entry(ua_sess->channels->ht, &chan_iter.iter, + ua_chan, node.node) { + ret = rotate_add_channel_pending( + ua_chan->key, LTTNG_DOMAIN_UST, + session); + if (ret < 0) { + ret = app->bits_per_long == 32 ? + -LTTNG_ERR_UST_CONSUMER32_FAIL : + -LTTNG_ERR_UST_CONSUMER64_FAIL; + goto error; + } + ret = consumer_rotate_channel(socket, ua_chan->key, + ua_sess->euid, ua_sess->egid, + ua_sess->consumer, pathname, + /* is_metadata_channel */ false, + session->current_archive_id, + &session->rotate_pending_relay); + if (ret < 0) { + goto error; + } + } + + /* Rotate the metadata channel. */ + (void) push_metadata(registry, usess->consumer); + ret = consumer_rotate_channel(socket, registry->metadata_key, + ua_sess->euid, ua_sess->egid, + ua_sess->consumer, pathname, + /* is_metadata_channel */ true, + session->current_archive_id, + &session->rotate_pending_relay); + if (ret < 0) { + goto error; + } + *ust_active = true; + } break; } default: - /* Should NEVER happen. */ assert(0); + break; } -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); + ret = LTTNG_OK; -close_socket: +error: 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); - } + return ret; }