X-Git-Url: https://git.lttng.org/?a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fevent-notifier-error-accounting.c;h=70fa39e023d4466431614e9499fde0d8797de2c7;hb=a5a212809b470aafd9c7f4cacbebfb4652feb16b;hp=f0e567f72ed110b6348f99e08baa946d20aa05e8;hpb=a55f8ad13c4e5a261f37793236403449399f973f;p=lttng-tools.git diff --git a/src/bin/lttng-sessiond/event-notifier-error-accounting.c b/src/bin/lttng-sessiond/event-notifier-error-accounting.c index f0e567f72..70fa39e02 100644 --- a/src/bin/lttng-sessiond/event-notifier-error-accounting.c +++ b/src/bin/lttng-sessiond/event-notifier-error-accounting.c @@ -31,13 +31,15 @@ struct index_ht_entry { struct rcu_head rcu_head; }; -struct error_account_entry { +struct ust_error_accounting_entry { + uid_t uid; + struct urcu_ref ref; struct lttng_ht_node_u64 node; struct rcu_head rcu_head; struct ustctl_daemon_counter *daemon_counter; /* - * Those `lttng_ust_abi_object_data` are anonymous handles to the counters - * objects. + * Those `lttng_ust_abi_object_data` are anonymous handles to the + * counters objects. * They are only used to be duplicated for each new applications of the * user. To destroy them, call with the `sock` parameter set to -1. * e.g. `ustctl_release_object(-1, data)`; @@ -47,20 +49,24 @@ struct error_account_entry { int nr_counter_cpu_fds; }; -struct kernel_error_account_entry { - int kernel_event_notifier_error_counter_fd; +struct kernel_error_accounting_entry { + int error_counter_fd; }; -static struct kernel_error_account_entry kernel_error_accountant; - -/* Hashtable mapping event notifier token to index_ht_entry. */ -static struct lttng_ht *error_counter_indexes_ht; +static struct kernel_error_accounting_entry kernel_error_accounting_entry; /* Hashtable mapping uid to error_account_entry. */ static struct lttng_ht *error_counter_uid_ht; -static uint64_t error_counter_size; -static struct lttng_index_allocator *index_allocator; +struct error_accounting_state { + struct lttng_index_allocator *index_allocator; + /* Hashtable mapping event notifier token to index_ht_entry. */ + struct lttng_ht *indices_ht; + uint64_t number_indices; +}; + +static struct error_accounting_state ust_state; +static struct error_accounting_state kernel_state; static inline void get_trigger_info_for_log(const struct lttng_trigger *trigger, const char **trigger_name, @@ -73,7 +79,7 @@ static inline void get_trigger_info_for_log(const struct lttng_trigger *trigger, case LTTNG_TRIGGER_STATUS_OK: break; case LTTNG_TRIGGER_STATUS_UNSET: - *trigger_name = "(unset)"; + *trigger_name = "(anonymous)"; break; default: abort(); @@ -106,24 +112,173 @@ const char *error_accounting_status_str( } } +#ifdef HAVE_LIBLTTNG_UST_CTL +struct event_notifier_counter { + pthread_mutex_t lock; + long count; +}; + +static struct event_notifier_counter the_event_notifier_counter; + +static void free_ust_error_accounting_entry(struct rcu_head *head) +{ + int i; + struct ust_error_accounting_entry *entry = + caa_container_of(head, typeof(*entry), rcu_head); + + for (i = 0; i < entry->nr_counter_cpu_fds; i++) { + ustctl_release_object(-1, entry->cpu_counters[i]); + free(entry->cpu_counters[i]); + } + + free(entry->cpu_counters); + + ustctl_release_object(-1, entry->counter); + free(entry->counter); + + ustctl_destroy_counter(entry->daemon_counter); + + free(entry); +} + +static +bool ust_error_accounting_entry_get(struct ust_error_accounting_entry *entry) +{ + return urcu_ref_get_unless_zero(&entry->ref); +} + +static +void ust_error_accounting_entry_release(struct urcu_ref *entry_ref) +{ + struct ust_error_accounting_entry *entry = + container_of(entry_ref, typeof(*entry), ref); + + rcu_read_lock(); + cds_lfht_del(error_counter_uid_ht->ht, &entry->node.node); + call_rcu(&entry->rcu_head, free_ust_error_accounting_entry); + rcu_read_unlock(); +} + + +static +void ust_error_accounting_entry_put(struct ust_error_accounting_entry *entry) +{ + if (!entry) { + return; + } + + urcu_ref_put(&entry->ref, ust_error_accounting_entry_release); +} + +/* + * Put one reference to every UID entries. + */ +static +void put_ref_all_ust_error_accounting_entry(void) +{ + struct lttng_ht_iter iter; + struct ust_error_accounting_entry *uid_entry; + + ASSERT_LOCKED(the_event_notifier_counter.lock); + + rcu_read_lock(); + cds_lfht_for_each_entry(error_counter_uid_ht->ht, &iter.iter, + uid_entry, node.node) { + ust_error_accounting_entry_put(uid_entry); + } + + rcu_read_unlock(); +} + +/* + * Get one reference to every UID entries. + */ +static +void get_ref_all_ust_error_accounting_entry(void) +{ + struct lttng_ht_iter iter; + struct ust_error_accounting_entry *uid_entry; + + ASSERT_LOCKED(the_event_notifier_counter.lock); + + rcu_read_lock(); + cds_lfht_for_each_entry(error_counter_uid_ht->ht, &iter.iter, + uid_entry, node.node) { + ust_error_accounting_entry_get(uid_entry); + } + + rcu_read_unlock(); +} + +#endif /* HAVE_LIBLTTNG_UST_CTL */ + +static enum event_notifier_error_accounting_status -event_notifier_error_accounting_init(uint64_t nb_bucket) +init_error_accounting_state(struct error_accounting_state *state, + uint64_t index_count) { enum event_notifier_error_accounting_status status; - index_allocator = lttng_index_allocator_create(nb_bucket); - if (!index_allocator) { - ERR("Failed to allocate event notifier error counter index"); + assert(state); + + state->number_indices = index_count; + + state->index_allocator = lttng_index_allocator_create(index_count); + if (!state->index_allocator) { + ERR("Failed to allocate event notifier error counter index allocator"); status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_NOMEM; - goto error_index_allocator; + goto end; } - error_counter_indexes_ht = lttng_ht_new( - ERROR_COUNTER_INDEX_HT_INITIAL_SIZE, LTTNG_HT_TYPE_U64); - if (!error_counter_indexes_ht) { + state->indices_ht = lttng_ht_new(ERROR_COUNTER_INDEX_HT_INITIAL_SIZE, + LTTNG_HT_TYPE_U64); + if (!state->indices_ht) { ERR("Failed to allocate error counter indices hash table"); status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_NOMEM; - goto error_index_allocator; + goto error_indices_ht; + } + + status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK; + goto end; + +error_indices_ht: + lttng_index_allocator_destroy(state->index_allocator); + state->index_allocator = NULL; +end: + return status; +} + +static +void fini_error_accounting_state(struct error_accounting_state *state) +{ + assert(state); + + /* + * Will assert if some error counter indices were not released (an + * internal error). + */ + lttng_ht_destroy(state->indices_ht); + lttng_index_allocator_destroy(state->index_allocator); +} + +enum event_notifier_error_accounting_status +event_notifier_error_accounting_init(uint64_t buffer_size_kernel, + uint64_t buffer_size_ust) +{ + enum event_notifier_error_accounting_status status; + + status = init_error_accounting_state(&kernel_state, buffer_size_kernel); + if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { + ERR("Failed to initialize kernel event notifier accounting state: status = %s", + error_accounting_status_str(status)); + goto end; + } + + status = init_error_accounting_state(&ust_state, buffer_size_ust); + if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { + ERR("Failed to initialize UST event notifier accounting state: status = %s", + error_accounting_status_str(status)); + goto error_ust_state; } error_counter_uid_ht = lttng_ht_new( @@ -131,20 +286,28 @@ event_notifier_error_accounting_init(uint64_t nb_bucket) if (!error_counter_uid_ht) { ERR("Failed to allocate UID to error counter accountant hash table"); status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_NOMEM; - goto error_index_allocator; + goto error_uid_ht; } - error_counter_size = nb_bucket; - status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK; + goto end; -error_index_allocator: +error_uid_ht: + fini_error_accounting_state(&ust_state); +error_ust_state: + fini_error_accounting_state(&kernel_state); +end: return status; } +/* + * Return the error counteur index associated to this event notifier tracer + * token. Returns _STATUS_OK if found and _STATUS_NOT_FOUND otherwise. + */ static enum event_notifier_error_accounting_status get_error_counter_index_for_token( - uint64_t tracer_token, uint64_t *error_counter_index) + struct error_accounting_state *state, uint64_t tracer_token, + uint64_t *error_counter_index) { struct lttng_ht_node_u64 *node; struct lttng_ht_iter iter; @@ -152,7 +315,7 @@ enum event_notifier_error_accounting_status get_error_counter_index_for_token( enum event_notifier_error_accounting_status status; rcu_read_lock(); - lttng_ht_lookup(error_counter_indexes_ht, &tracer_token, &iter); + lttng_ht_lookup(state->indices_ht, &tracer_token, &iter); node = lttng_ht_iter_get_node_u64(&iter); if (node) { index_entry = caa_container_of( @@ -168,53 +331,71 @@ enum event_notifier_error_accounting_status get_error_counter_index_for_token( } #ifdef HAVE_LIBLTTNG_UST_CTL +/* + * Find the entry for this app's UID, the caller acquires a reference if the + * entry is found. + */ static -struct error_account_entry *get_uid_accounting_entry(const struct ust_app *app) +struct ust_error_accounting_entry *ust_error_accounting_entry_find( + struct lttng_ht *uid_ht, const struct ust_app *app) { - struct error_account_entry *entry; + struct ust_error_accounting_entry *entry; struct lttng_ht_node_u64 *node; struct lttng_ht_iter iter; uint64_t key = app->uid; - lttng_ht_lookup(error_counter_uid_ht, &key, &iter); + lttng_ht_lookup(uid_ht, &key, &iter); node = lttng_ht_iter_get_node_u64(&iter); if(node == NULL) { entry = NULL; } else { - entry = caa_container_of(node, struct error_account_entry, node); + bool got_ref; + + entry = caa_container_of(node, + struct ust_error_accounting_entry, node); + + got_ref = ust_error_accounting_entry_get(entry); + if (!got_ref) { + entry = NULL; + } } return entry; } +/* + * Create the entry for this app's UID, the caller acquires a reference to the + * entry, + */ static -struct error_account_entry *create_uid_accounting_entry( - const struct ust_app *app) +struct ust_error_accounting_entry *ust_error_accounting_entry_create( + struct lttng_ht *uid_ht, const struct ust_app *app) { - int i, ret; + int i, ret, *cpu_counter_fds = NULL; struct ustctl_daemon_counter *daemon_counter; struct lttng_ust_abi_object_data *counter, **cpu_counters; - int *cpu_counter_fds = NULL; - struct error_account_entry *entry = NULL; + struct ust_error_accounting_entry *entry = NULL; const struct ustctl_counter_dimension dimension = { - .size = error_counter_size, + .size = ust_state.number_indices, .has_underflow = false, .has_overflow = false, }; - entry = zmalloc(sizeof(struct error_account_entry)); + entry = zmalloc(sizeof(struct ust_error_accounting_entry)); if (!entry) { PERROR("Failed to allocate event notifier error acounting entry") goto error; } + urcu_ref_init(&entry->ref); + entry->uid = app->uid; entry->nr_counter_cpu_fds = ustctl_get_nr_cpu_per_counter(); + cpu_counter_fds = zmalloc(entry->nr_counter_cpu_fds * sizeof(*cpu_counter_fds)); if (!cpu_counter_fds) { PERROR("Failed to allocate event notifier error counter file descriptors array: application uid = %d, application name = '%s', pid = %d, allocation size = %zu", (int) app->uid, app->name, (int) app->pid, entry->nr_counter_cpu_fds * sizeof(*cpu_counter_fds)); - ret = -1; goto error_counter_cpu_fds_alloc; } @@ -223,12 +404,11 @@ struct error_account_entry *create_uid_accounting_entry( cpu_counter_fds[i] = -1; } - cpu_counters = zmalloc(entry->nr_counter_cpu_fds * sizeof(**cpu_counters)); + cpu_counters = zmalloc(entry->nr_counter_cpu_fds * sizeof(struct lttng_ust_abi_object_data *)); if (!cpu_counters) { PERROR("Failed to allocate event notifier error counter lttng_ust_abi_object_data array: application uid = %d, application name = '%s', pid = %d, allocation size = %zu", (int) app->uid, app->name, (int) app->pid, - entry->nr_counter_cpu_fds * sizeof(**cpu_counters)); - ret = -1; + entry->nr_counter_cpu_fds * sizeof(struct lttng_ust_abi_object_data *)); goto error_counter_cpus_alloc; } @@ -276,7 +456,7 @@ struct error_account_entry *create_uid_accounting_entry( entry->counter = counter; entry->cpu_counters = cpu_counters; - lttng_ht_node_init_u64(&entry->node, app->uid); + lttng_ht_node_init_u64(&entry->node, entry->uid); lttng_ht_add_unique_u64(error_counter_uid_ht, &entry->node); goto end; @@ -315,7 +495,8 @@ error_shm_alloc: ret = close(cpu_counter_fds[i]); if (ret) { - PERROR("Failed to close error counter per-CPU shm file descriptor: fd = %d", cpu_counter_fds[i]); + PERROR("Failed to close error counter per-CPU shm file descriptor: fd = %d", + cpu_counter_fds[i]); } } } @@ -401,7 +582,7 @@ event_notifier_error_accounting_register_app(struct ust_app *app) int ret; uint64_t i; struct lttng_ust_abi_object_data *new_counter; - struct error_account_entry *entry; + struct ust_error_accounting_entry *entry; enum event_notifier_error_accounting_status status; struct lttng_ust_abi_object_data **cpu_counters; @@ -410,13 +591,33 @@ event_notifier_error_accounting_register_app(struct ust_app *app) * app. If not, create one. */ rcu_read_lock(); - entry = get_uid_accounting_entry(app); + entry = ust_error_accounting_entry_find(error_counter_uid_ht, app); if (entry == NULL) { - entry = create_uid_accounting_entry(app); + /* + * Take the event notifier counter lock before creating the new + * entry to ensure that no event notifier is registered between + * the the entry creation and event notifier count check. + */ + pthread_mutex_lock(&the_event_notifier_counter.lock); + + entry = ust_error_accounting_entry_create(error_counter_uid_ht, + app); if (!entry) { status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_ERR; - goto end; + pthread_mutex_unlock(&the_event_notifier_counter.lock); + goto error_creating_entry; + } + + /* + * We just created a new UID entry, If there are event + * notifiers already registered, take one reference on their + * behalf. + */ + if (the_event_notifier_counter.count > 0) { + ust_error_accounting_entry_get(entry); } + + pthread_mutex_unlock(&the_event_notifier_counter.lock); } /* Duplicate counter object data. */ @@ -426,11 +627,15 @@ event_notifier_error_accounting_register_app(struct ust_app *app) ERR("Failed to duplicate event notifier error accounting counter for application user: application uid = %d, pid = %d, application name = '%s'", (int) app->uid, (int) app->pid, app->name); status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_ERR; - goto end; + goto error_duplicate_counter; } status = send_counter_data_to_ust(app, new_counter); if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { + if (status == EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_APP_DEAD) { + goto error_send_counter_data; + } + ERR("Failed to send counter data to application tracer: status = %s, application uid = %d, pid = %d, application name = '%s'", error_accounting_status_str(status), (int) app->uid, (int) app->pid, app->name); @@ -438,7 +643,7 @@ event_notifier_error_accounting_register_app(struct ust_app *app) goto error_send_counter_data; } - cpu_counters = zmalloc(entry->nr_counter_cpu_fds * sizeof(struct lttng_ust_abi_object_data)); + cpu_counters = zmalloc(entry->nr_counter_cpu_fds * sizeof(struct lttng_ust_abi_object_data *)); if (!cpu_counters) { PERROR("Failed to allocate event notifier error counter lttng_ust_abi_object_data array: application uid = %d, application name = '%s', pid = %d, allocation size = %zu", (int) app->uid, app->name, (int) app->pid, @@ -465,6 +670,10 @@ event_notifier_error_accounting_register_app(struct ust_app *app) status = send_counter_cpu_data_to_ust(app, new_counter, new_counter_cpu); if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { + if (status == EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_APP_DEAD) { + goto error_send_cpu_counter_data; + } + ERR("Failed to send counter cpu data to application tracer: status = %s, application uid = %d, pid = %d, application name = '%s'", error_accounting_status_str(status), (int) app->uid, (int) app->pid, @@ -503,6 +712,10 @@ error_allocate_cpu_counters: error_send_counter_data: ustctl_release_object(-1, new_counter); free(new_counter); +error_duplicate_counter: + ust_error_accounting_entry_put(entry); +error_creating_entry: + app->event_notifier_group.counter = NULL; end: rcu_read_unlock(); return status; @@ -512,16 +725,30 @@ enum event_notifier_error_accounting_status event_notifier_error_accounting_unregister_app(struct ust_app *app) { enum event_notifier_error_accounting_status status; - struct error_account_entry *entry; + struct ust_error_accounting_entry *entry; int i; rcu_read_lock(); - entry = get_uid_accounting_entry(app); + + /* If an error occurred during app registration no entry was created. */ + if (!app->event_notifier_group.counter) { + status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK; + goto end; + } + + entry = ust_error_accounting_entry_find(error_counter_uid_ht, app); if (entry == NULL) { ERR("Failed to find event notitifier error accounting entry on application teardown: pid = %d, application name = '%s'", app->pid, app->name); status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_ERR; goto end; + } else { + /* + * Put the entry twice as we acquired a reference from the + * `ust_error_accounting_entry_find()` above. + */ + ust_error_accounting_entry_put(entry); + ust_error_accounting_entry_put(entry); } for (i = 0; i < app->event_notifier_group.nr_counter_cpu; i++) { @@ -547,29 +774,22 @@ event_notifier_error_accounting_ust_get_count( const struct lttng_trigger *trigger, uint64_t *count) { struct lttng_ht_iter iter; - struct error_account_entry *uid_entry; + struct ust_error_accounting_entry *uid_entry; uint64_t error_counter_index, global_sum = 0; enum event_notifier_error_accounting_status status; size_t dimension_indexes[1]; const uint64_t tracer_token = lttng_trigger_get_tracer_token(trigger); + uid_t trigger_owner_uid; + const char *trigger_name; - /* - * Go over all error counters (ignoring uid) as a trigger (and trigger - * errors) can be generated from any applications that this session - * daemon is managing. - */ rcu_read_lock(); - status = get_error_counter_index_for_token( - tracer_token, + get_trigger_info_for_log(trigger, &trigger_name, &trigger_owner_uid); + + status = get_error_counter_index_for_token(&ust_state, tracer_token, &error_counter_index); if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { - uid_t trigger_owner_uid; - const char *trigger_name; - - get_trigger_info_for_log(trigger, &trigger_name, - &trigger_owner_uid); ERR("Failed to retrieve index for tracer token: token = %" PRIu64 ", trigger name = '%s', trigger owner uid = %d, status = %s", tracer_token, trigger_name, @@ -580,6 +800,13 @@ event_notifier_error_accounting_ust_get_count( dimension_indexes[0] = error_counter_index; + /* + * Iterate over all the UID entries. + * We aggregate the value of all uid entries regardless of if the uid + * matches the trigger's uid because a user that is allowed to register + * a trigger to a given sessiond is also allowed to create an event + * notifier on all apps that this sessiond is aware of. + */ cds_lfht_for_each_entry(error_counter_uid_ht->ht, &iter.iter, uid_entry, node.node) { int ret; @@ -590,12 +817,6 @@ event_notifier_error_accounting_ust_get_count( dimension_indexes, &local_value, &overflow, &underflow); if (ret || local_value < 0) { - uid_t trigger_owner_uid; - const char *trigger_name; - - get_trigger_info_for_log(trigger, &trigger_name, - &trigger_owner_uid); - if (ret) { ERR("Failed to aggregate event notifier error counter values of trigger: trigger name = '%s', trigger owner uid = %d", trigger_name, @@ -615,7 +836,6 @@ event_notifier_error_accounting_ust_get_count( /* Cast is safe as negative values are checked-for above. */ global_sum += (uint64_t) local_value; - } *count = global_sum; @@ -631,21 +851,14 @@ enum event_notifier_error_accounting_status event_notifier_error_accounting_ust_ const struct lttng_trigger *trigger) { struct lttng_ht_iter iter; - struct error_account_entry *uid_entry; + struct ust_error_accounting_entry *uid_entry; uint64_t error_counter_index; enum event_notifier_error_accounting_status status; size_t dimension_index; const uint64_t tracer_token = lttng_trigger_get_tracer_token(trigger); - /* - * Go over all error counters (ignoring uid) as a trigger (and trigger - * errors) can be generated from any applications that this session - * daemon is managing. - */ - rcu_read_lock(); - status = get_error_counter_index_for_token( - tracer_token, + status = get_error_counter_index_for_token(&ust_state, tracer_token, &error_counter_index); if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { uid_t trigger_owner_uid; @@ -663,6 +876,11 @@ enum event_notifier_error_accounting_status event_notifier_error_accounting_ust_ dimension_index = error_counter_index; + /* + * Go over all error counters (ignoring uid) as a trigger (and trigger + * errors) can be generated from any applications that this session + * daemon is managing. + */ cds_lfht_for_each_entry(error_counter_uid_ht->ht, &iter.iter, uid_entry, node.node) { const int ret = ustctl_counter_clear(uid_entry->daemon_counter, @@ -699,7 +917,7 @@ event_notifier_error_accounting_kernel_clear( enum event_notifier_error_accounting_status status; struct lttng_kernel_counter_clear counter_clear = {}; - status = get_error_counter_index_for_token( + status = get_error_counter_index_for_token(&kernel_state, lttng_trigger_get_tracer_token(trigger), &error_counter_index); if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { @@ -719,7 +937,7 @@ event_notifier_error_accounting_kernel_clear( counter_clear.index.dimension_indexes[0] = error_counter_index; ret = kernctl_counter_clear( - kernel_error_accountant.kernel_event_notifier_error_counter_fd, + kernel_error_accounting_entry.error_counter_fd, &counter_clear); if (ret) { uid_t trigger_owner_uid; @@ -752,7 +970,7 @@ event_notifier_error_accounting_register_kernel( LTTNG_KERNEL_COUNTER_BITNESS_64, .global_sum_step = 0, .number_dimensions = 1, - .dimensions[0].size = error_counter_size, + .dimensions[0].size = kernel_state.number_indices, .dimensions[0].has_underflow = false, .dimensions[0].has_overflow = false, }; @@ -780,7 +998,7 @@ event_notifier_error_accounting_register_kernel( DBG("Created kernel event notifier group error counter: fd = %d", error_counter_fd); - kernel_error_accountant.kernel_event_notifier_error_counter_fd = + kernel_error_accounting_entry.error_counter_fd = error_counter_fd; status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK; @@ -790,22 +1008,25 @@ error: static enum event_notifier_error_accounting_status create_error_counter_index_for_token( - uint64_t tracer_token, uint64_t *error_counter_index) + struct error_accounting_state *state, uint64_t tracer_token, + uint64_t *error_counter_index) { struct index_ht_entry *index_entry; enum lttng_index_allocator_status index_alloc_status; uint64_t local_error_counter_index; enum event_notifier_error_accounting_status status; + assert(state); + /* Allocate a new index for that counter. */ - index_alloc_status = lttng_index_allocator_alloc(index_allocator, + index_alloc_status = lttng_index_allocator_alloc(state->index_allocator, &local_error_counter_index); switch (index_alloc_status) { case LTTNG_INDEX_ALLOCATOR_STATUS_EMPTY: DBG("No indices left in the configured event notifier error counter: " "number-of-indices = %"PRIu64, lttng_index_allocator_get_index_count( - index_allocator)); + state->index_allocator)); status = EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_NO_INDEX_AVAILABLE; goto end; case LTTNG_INDEX_ALLOCATOR_STATUS_OK: @@ -824,7 +1045,7 @@ enum event_notifier_error_accounting_status create_error_counter_index_for_token index_entry->error_counter_index = local_error_counter_index; lttng_ht_node_init_u64(&index_entry->node, tracer_token); - lttng_ht_add_unique_u64(error_counter_indexes_ht, &index_entry->node); + lttng_ht_add_unique_u64(state->indices_ht, &index_entry->node); DBG("Allocated error counter index for tracer token: tracer token = %" PRIu64 ", index = %" PRIu64, tracer_token, local_error_counter_index); @@ -841,12 +1062,24 @@ event_notifier_error_accounting_register_event_notifier( { enum event_notifier_error_accounting_status status; uint64_t local_error_counter_index; + struct error_accounting_state *state; + + switch (lttng_trigger_get_underlying_domain_type_restriction(trigger)) { + case LTTNG_DOMAIN_KERNEL: + state = &kernel_state; + break; + case LTTNG_DOMAIN_UST: + state = &ust_state; + break; + default: + abort(); + } /* * Check if this event notifier already has a error counter index * assigned. */ - status = get_error_counter_index_for_token( + status = get_error_counter_index_for_token(state, lttng_trigger_get_tracer_token(trigger), &local_error_counter_index); switch (status) { @@ -862,7 +1095,7 @@ event_notifier_error_accounting_register_event_notifier( trigger_name, trigger_owner_uid, lttng_trigger_get_tracer_token(trigger)); - status = create_error_counter_index_for_token( + status = create_error_counter_index_for_token(state, lttng_trigger_get_tracer_token(trigger), &local_error_counter_index); if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { @@ -881,6 +1114,28 @@ event_notifier_error_accounting_register_event_notifier( break; } +#ifdef HAVE_LIBLTTNG_UST_CTL + switch (lttng_trigger_get_underlying_domain_type_restriction(trigger)) { + case LTTNG_DOMAIN_UST: + pthread_mutex_lock(&the_event_notifier_counter.lock); + the_event_notifier_counter.count++; + if (the_event_notifier_counter.count == 1) { + /* + * On the first event notifier, we get a reference to + * every existing UID entries. This ensures that the + * entries are kept around if there are still + * registered event notifiers but no apps. + */ + get_ref_all_ust_error_accounting_entry(); + } + pthread_mutex_unlock(&the_event_notifier_counter.lock); + break; + default: + break; + } +#endif /* HAVE_LIBLTTNG_UST_CTL */ + + end: return status; } @@ -895,8 +1150,9 @@ event_notifier_error_accounting_kernel_get_count( uint64_t error_counter_index; int ret; - status = get_error_counter_index_for_token( - lttng_trigger_get_tracer_token(trigger), &error_counter_index); + status = get_error_counter_index_for_token(&kernel_state, + lttng_trigger_get_tracer_token(trigger), + &error_counter_index); if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { ERR("Error getting index for token: status=%s", error_accounting_status_str(status)); @@ -906,10 +1162,10 @@ event_notifier_error_accounting_kernel_get_count( counter_aggregate.index.number_dimensions = 1; counter_aggregate.index.dimension_indexes[0] = error_counter_index; - assert(kernel_error_accountant.kernel_event_notifier_error_counter_fd); + assert(kernel_error_accounting_entry.error_counter_fd); ret = kernctl_counter_get_aggregate_value( - kernel_error_accountant.kernel_event_notifier_error_counter_fd, + kernel_error_accounting_entry.error_counter_fd, &counter_aggregate); if (ret || counter_aggregate.value.value < 0) { uid_t trigger_owner_uid; @@ -951,8 +1207,10 @@ event_notifier_error_accounting_get_count( trigger, count); case LTTNG_DOMAIN_UST: #ifdef HAVE_LIBLTTNG_UST_CTL - return event_notifier_error_accounting_ust_get_count(trigger, count); + return event_notifier_error_accounting_ust_get_count(trigger, + count); #else + *count = 0; return EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK; #endif /* HAVE_LIBLTTNG_UST_CTL */ default: @@ -993,16 +1251,47 @@ void event_notifier_error_accounting_unregister_event_notifier( struct lttng_ht_node_u64 *node; const uint64_t tracer_token = lttng_trigger_get_tracer_token(trigger); enum event_notifier_error_accounting_status status; + struct error_accounting_state *state; status = event_notifier_error_accounting_clear(trigger); if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { /* Trigger details already logged by callee on error. */ ERR("Failed to clear event notifier error counter during unregistration of event notifier: status = '%s'", error_accounting_status_str(status)); + goto end; } rcu_read_lock(); - lttng_ht_lookup(error_counter_indexes_ht, &tracer_token, &iter); + + switch (lttng_trigger_get_underlying_domain_type_restriction(trigger)) { + case LTTNG_DOMAIN_KERNEL: + state = &kernel_state; + break; +#ifdef HAVE_LIBLTTNG_UST_CTL + case LTTNG_DOMAIN_UST: + state = &ust_state; + + pthread_mutex_lock(&the_event_notifier_counter.lock); + the_event_notifier_counter.count--; + if (the_event_notifier_counter.count == 0) { + + /* + * When unregistering the last event notifier, put one + * reference to every uid entries on the behalf of all + * event notifiers. + */ + put_ref_all_ust_error_accounting_entry(); + } + + pthread_mutex_unlock(&the_event_notifier_counter.lock); + + break; +#endif /* HAVE_LIBLTTNG_UST_CTL */ + default: + abort(); + } + + lttng_ht_lookup(state->indices_ht, &tracer_token, &iter); node = lttng_ht_iter_get_node_u64(&iter); if (node) { int del_ret; @@ -1011,7 +1300,7 @@ void event_notifier_error_accounting_unregister_event_notifier( enum lttng_index_allocator_status index_alloc_status; index_alloc_status = lttng_index_allocator_release( - index_allocator, + state->index_allocator, index_entry->error_counter_index); if (index_alloc_status != LTTNG_INDEX_ALLOCATOR_STATUS_OK) { uid_t trigger_owner_uid; @@ -1026,72 +1315,27 @@ void event_notifier_error_accounting_unregister_event_notifier( /* Don't exit, perform the rest of the clean-up. */ } - del_ret = lttng_ht_del(error_counter_indexes_ht, &iter); + del_ret = lttng_ht_del(state->indices_ht, &iter); assert(!del_ret); call_rcu(&index_entry->rcu_head, free_index_ht_entry); } +end: rcu_read_unlock(); } -#ifdef HAVE_LIBLTTNG_UST_CTL -static void free_error_account_entry(struct rcu_head *head) -{ - int i; - struct error_account_entry *entry = - caa_container_of(head, typeof(*entry), rcu_head); - - for (i = 0; i < entry->nr_counter_cpu_fds; i++) { - ustctl_release_object(-1, entry->cpu_counters[i]); - free(entry->cpu_counters[i]); - } - - free(entry->cpu_counters); - - ustctl_release_object(-1, entry->counter); - free(entry->counter); - - ustctl_destroy_counter(entry->daemon_counter); - - free(entry); -} -#else -/* Not called without UST support. */ -static void free_error_account_entry(struct rcu_head *head) {} -#endif /* HAVE_LIBLTTNG_UST_CTL */ - void event_notifier_error_accounting_fini(void) { - struct lttng_ht_iter iter; - struct error_account_entry *uid_entry; - - lttng_index_allocator_destroy(index_allocator); - - if (kernel_error_accountant.kernel_event_notifier_error_counter_fd) { - const int ret = close(kernel_error_accountant.kernel_event_notifier_error_counter_fd); + if (kernel_error_accounting_entry.error_counter_fd) { + const int ret = close(kernel_error_accounting_entry.error_counter_fd); if (ret) { PERROR("Failed to close kernel event notifier error counter"); } } - /* - * FIXME error account entries are not reference-counted and torn - * down on last use. They exist from the moment of their first use - * up until the teardown of the session daemon. - */ - rcu_read_lock(); - cds_lfht_for_each_entry(error_counter_uid_ht->ht, &iter.iter, - uid_entry, node.node) { - cds_lfht_del(error_counter_uid_ht->ht, &uid_entry->node.node); - call_rcu(&uid_entry->rcu_head, free_error_account_entry); - } - rcu_read_unlock(); lttng_ht_destroy(error_counter_uid_ht); - /* - * Will assert if some error counter indices were not released (an - * internal error). - */ - lttng_ht_destroy(error_counter_indexes_ht); + fini_error_accounting_state(&kernel_state); + fini_error_accounting_state(&ust_state); }