X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Fnotification-thread-events.c;h=827cf2c1459f5ee4cb683ac3fd5b0699ab5d9b82;hp=6f0ee7d7942cf762b38c6c7edc37ab97caa8972b;hb=f712e5f6c7a77e92abed2f0d0740b9512d1fa4e2;hpb=f37d0f861f20a72a0b77fb43fa27744521dd7995 diff --git a/src/bin/lttng-sessiond/notification-thread-events.c b/src/bin/lttng-sessiond/notification-thread-events.c index 6f0ee7d79..827cf2c14 100644 --- a/src/bin/lttng-sessiond/notification-thread-events.c +++ b/src/bin/lttng-sessiond/notification-thread-events.c @@ -5,6 +5,8 @@ * */ +#include "lttng/action/action.h" +#include "lttng/trigger/trigger-internal.h" #define _LGPL_SOURCE #include #include @@ -19,12 +21,18 @@ #include #include #include +#include +#include #include #include #include #include #include +#include +#include #include +#include +#include #include #include @@ -32,6 +40,8 @@ #include #include +#include "condition-internal.h" +#include "event-notifier-error-accounting.h" #include "notification-thread.h" #include "notification-thread-events.h" #include "notification-thread-commands.h" @@ -41,6 +51,9 @@ #define CLIENT_POLL_MASK_IN (LPOLLIN | LPOLLERR | LPOLLHUP | LPOLLRDHUP) #define CLIENT_POLL_MASK_IN_OUT (CLIENT_POLL_MASK_IN | LPOLLOUT) +/* The tracers currently limit the capture size to PIPE_BUF (4kb on linux). */ +#define MAX_CAPTURE_SIZE (PIPE_BUF) + enum lttng_object_type { LTTNG_OBJECT_TYPE_UNKNOWN, LTTNG_OBJECT_TYPE_NONE, @@ -50,7 +63,7 @@ enum lttng_object_type { struct lttng_trigger_list_element { /* No ownership of the trigger object is assumed. */ - const struct lttng_trigger *trigger; + struct lttng_trigger *trigger; struct cds_list_head node; }; @@ -108,6 +121,8 @@ struct lttng_session_trigger_list { struct lttng_trigger_ht_element { struct lttng_trigger *trigger; struct cds_lfht_node node; + struct cds_lfht_node node_by_name_uid; + struct cds_list_head client_list_trigger_node; /* call_rcu delayed reclaim. */ struct rcu_head rcu_node; }; @@ -117,87 +132,6 @@ struct lttng_condition_list_element { struct cds_list_head node; }; -struct notification_client_list_element { - struct notification_client *client; - struct cds_list_head node; -}; - -struct notification_client_list { - const struct lttng_trigger *trigger; - struct cds_list_head list; - struct cds_lfht_node notification_trigger_ht_node; - /* call_rcu delayed reclaim. */ - struct rcu_head rcu_node; -}; - -struct notification_client { - int socket; - /* Client protocol version. */ - uint8_t major, minor; - uid_t uid; - gid_t gid; - /* - * Indicates if the credentials and versions of the client have been - * checked. - */ - bool validated; - /* - * Conditions to which the client's notification channel is subscribed. - * List of struct lttng_condition_list_node. The condition member is - * owned by the client. - */ - struct cds_list_head condition_list; - struct cds_lfht_node client_socket_ht_node; - struct { - struct { - /* - * During the reception of a message, the reception - * buffers' "size" is set to contain the current - * message's complete payload. - */ - struct lttng_dynamic_buffer buffer; - /* Bytes left to receive for the current message. */ - size_t bytes_to_receive; - /* Type of the message being received. */ - enum lttng_notification_channel_message_type msg_type; - /* - * Indicates whether or not credentials are expected - * from the client. - */ - bool expect_creds; - /* - * Indicates whether or not credentials were received - * from the client. - */ - bool creds_received; - /* Only used during credentials reception. */ - lttng_sock_cred creds; - } inbound; - struct { - /* - * Indicates whether or not a notification addressed to - * this client was dropped because a command reply was - * already buffered. - * - * A notification is dropped whenever the buffer is not - * empty. - */ - bool dropped_notification; - /* - * Indicates whether or not a command reply is already - * buffered. In this case, it means that the client is - * not consuming command replies before emitting a new - * one. This could be caused by a protocol error or a - * misbehaving/malicious client. - */ - bool queued_command_reply; - struct lttng_dynamic_buffer buffer; - } outbound; - } communication; - /* call_rcu delayed reclaim. */ - struct rcu_head rcu_node; -}; - struct channel_state_sample { struct channel_key key; struct cds_lfht_node channel_state_ht_node; @@ -258,20 +192,42 @@ void lttng_session_trigger_list_destroy( struct lttng_session_trigger_list *list); static int lttng_session_trigger_list_add(struct lttng_session_trigger_list *list, - const struct lttng_trigger *trigger); + struct lttng_trigger *trigger); + +static +int client_handle_transmission_status( + struct notification_client *client, + enum client_transmission_status transmission_status, + struct notification_thread_state *state); + +static +int handle_one_event_notifier_notification( + struct notification_thread_state *state, + int pipe, enum lttng_domain_type domain); +static +void free_lttng_trigger_ht_element_rcu(struct rcu_head *node); static -int match_client(struct cds_lfht_node *node, const void *key) +int match_client_socket(struct cds_lfht_node *node, const void *key) { /* This double-cast is intended to supress pointer-to-cast warning. */ - int socket = (int) (intptr_t) key; - struct notification_client *client; + const int socket = (int) (intptr_t) key; + const struct notification_client *client = caa_container_of(node, + struct notification_client, client_socket_ht_node); - client = caa_container_of(node, struct notification_client, - client_socket_ht_node); + return client->socket == socket; +} + +static +int match_client_id(struct cds_lfht_node *node, const void *key) +{ + /* This double-cast is intended to supress pointer-to-cast warning. */ + const notification_client_id id = *((notification_client_id *) key); + const struct notification_client *client = caa_container_of( + node, struct notification_client, client_id_ht_node); - return !!(client->socket == socket); + return client->id == id; } static @@ -326,18 +282,26 @@ int match_channel_info(struct cds_lfht_node *node, const void *key) } static -int match_condition(struct cds_lfht_node *node, const void *key) +int match_trigger(struct cds_lfht_node *node, const void *key) { - struct lttng_condition *condition_key = (struct lttng_condition *) key; - struct lttng_trigger_ht_element *trigger; - struct lttng_condition *condition; + struct lttng_trigger *trigger_key = (struct lttng_trigger *) key; + struct lttng_trigger_ht_element *trigger_ht_element; - trigger = caa_container_of(node, struct lttng_trigger_ht_element, + trigger_ht_element = caa_container_of(node, struct lttng_trigger_ht_element, node); - condition = lttng_trigger_get_condition(trigger->trigger); - assert(condition); - return !!lttng_condition_is_equal(condition_key, condition); + return !!lttng_trigger_is_equal(trigger_key, trigger_ht_element->trigger); +} + +static +int match_trigger_token(struct cds_lfht_node *node, const void *key) +{ + const uint64_t *_key = key; + struct notification_trigger_tokens_ht_element *element; + + element = caa_container_of(node, + struct notification_trigger_tokens_ht_element, node); + return *_key == element->token; } static @@ -350,8 +314,8 @@ int match_client_list_condition(struct cds_lfht_node *node, const void *key) assert(condition_key); client_list = caa_container_of(node, struct notification_client_list, - notification_trigger_ht_node); - condition = lttng_trigger_get_const_condition(client_list->trigger); + notification_trigger_clients_ht_node); + condition = client_list->condition; return !!lttng_condition_is_equal(condition_key, condition); } @@ -367,102 +331,100 @@ int match_session(struct cds_lfht_node *node, const void *key) } static -unsigned long lttng_condition_buffer_usage_hash( - const struct lttng_condition *_condition) +const char *notification_command_type_str( + enum notification_thread_command_type type) { - unsigned long hash; - unsigned long condition_type; - struct lttng_condition_buffer_usage *condition; - - condition = container_of(_condition, - struct lttng_condition_buffer_usage, parent); - - condition_type = (unsigned long) condition->parent.type; - hash = hash_key_ulong((void *) condition_type, lttng_ht_seed); - if (condition->session_name) { - hash ^= hash_key_str(condition->session_name, lttng_ht_seed); - } - if (condition->channel_name) { - hash ^= hash_key_str(condition->channel_name, lttng_ht_seed); - } - if (condition->domain.set) { - hash ^= hash_key_ulong( - (void *) condition->domain.type, - lttng_ht_seed); - } - if (condition->threshold_ratio.set) { - uint64_t val; - - val = condition->threshold_ratio.value * (double) UINT32_MAX; - hash ^= hash_key_u64(&val, lttng_ht_seed); - } else if (condition->threshold_bytes.set) { - uint64_t val; - - val = condition->threshold_bytes.value; - hash ^= hash_key_u64(&val, lttng_ht_seed); + switch (type) { + case NOTIFICATION_COMMAND_TYPE_REGISTER_TRIGGER: + return "REGISTER_TRIGGER"; + case NOTIFICATION_COMMAND_TYPE_UNREGISTER_TRIGGER: + return "UNREGISTER_TRIGGER"; + case NOTIFICATION_COMMAND_TYPE_ADD_CHANNEL: + return "ADD_CHANNEL"; + case NOTIFICATION_COMMAND_TYPE_REMOVE_CHANNEL: + return "REMOVE_CHANNEL"; + case NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_ONGOING: + return "SESSION_ROTATION_ONGOING"; + case NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_COMPLETED: + return "SESSION_ROTATION_COMPLETED"; + case NOTIFICATION_COMMAND_TYPE_ADD_TRACER_EVENT_SOURCE: + return "ADD_TRACER_EVENT_SOURCE"; + case NOTIFICATION_COMMAND_TYPE_REMOVE_TRACER_EVENT_SOURCE: + return "REMOVE_TRACER_EVENT_SOURCE"; + case NOTIFICATION_COMMAND_TYPE_LIST_TRIGGERS: + return "LIST_TRIGGERS"; + case NOTIFICATION_COMMAND_TYPE_QUIT: + return "QUIT"; + case NOTIFICATION_COMMAND_TYPE_CLIENT_COMMUNICATION_UPDATE: + return "CLIENT_COMMUNICATION_UPDATE"; + default: + abort(); } - return hash; } +/* + * Match trigger based on name and credentials only. + * Name duplication is NOT allowed for the same uid. + */ static -unsigned long lttng_condition_session_consumed_size_hash( - const struct lttng_condition *_condition) +int match_trigger_by_name_uid(struct cds_lfht_node *node, + const void *key) { - unsigned long hash; - unsigned long condition_type = - (unsigned long) LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE; - struct lttng_condition_session_consumed_size *condition; - uint64_t val; - - condition = container_of(_condition, - struct lttng_condition_session_consumed_size, parent); - - hash = hash_key_ulong((void *) condition_type, lttng_ht_seed); - if (condition->session_name) { - hash ^= hash_key_str(condition->session_name, lttng_ht_seed); + bool match = false; + const char *name; + const char *key_name; + enum lttng_trigger_status status; + const struct lttng_credentials *key_creds; + const struct lttng_credentials *node_creds; + const struct lttng_trigger *trigger_key = + (const struct lttng_trigger *) key; + const struct lttng_trigger_ht_element *trigger_ht_element = + caa_container_of(node, + struct lttng_trigger_ht_element, + node_by_name_uid); + + status = lttng_trigger_get_name(trigger_ht_element->trigger, &name); + assert(status == LTTNG_TRIGGER_STATUS_OK); + + status = lttng_trigger_get_name(trigger_key, &key_name); + assert(status == LTTNG_TRIGGER_STATUS_OK); + + /* Compare the names. */ + if (strcmp(name, key_name) != 0) { + goto end; } - val = condition->consumed_threshold_bytes.value; - hash ^= hash_key_u64(&val, lttng_ht_seed); - return hash; -} -static -unsigned long lttng_condition_session_rotation_hash( - const struct lttng_condition *_condition) -{ - unsigned long hash, condition_type; - struct lttng_condition_session_rotation *condition; + /* Compare the owners' UIDs. */ + key_creds = lttng_trigger_get_credentials(trigger_key); + node_creds = lttng_trigger_get_credentials(trigger_ht_element->trigger); - condition = container_of(_condition, - struct lttng_condition_session_rotation, parent); - condition_type = (unsigned long) condition->parent.type; - hash = hash_key_ulong((void *) condition_type, lttng_ht_seed); - assert(condition->session_name); - hash ^= hash_key_str(condition->session_name, lttng_ht_seed); - return hash; + match = lttng_credentials_is_equal_uid(key_creds, node_creds); + +end: + return match; } /* - * The lttng_condition hashing code is kept in this file (rather than - * condition.c) since it makes use of GPLv2 code (hashtable utils), which we - * don't want to link in liblttng-ctl. + * Hash trigger based on name and credentials only. */ static -unsigned long lttng_condition_hash(const struct lttng_condition *condition) +unsigned long hash_trigger_by_name_uid(const struct lttng_trigger *trigger) { - switch (condition->type) { - case LTTNG_CONDITION_TYPE_BUFFER_USAGE_LOW: - case LTTNG_CONDITION_TYPE_BUFFER_USAGE_HIGH: - return lttng_condition_buffer_usage_hash(condition); - case LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE: - return lttng_condition_session_consumed_size_hash(condition); - case LTTNG_CONDITION_TYPE_SESSION_ROTATION_ONGOING: - case LTTNG_CONDITION_TYPE_SESSION_ROTATION_COMPLETED: - return lttng_condition_session_rotation_hash(condition); - default: - ERR("[notification-thread] Unexpected condition type caught"); - abort(); + unsigned long hash = 0; + const struct lttng_credentials *trigger_creds; + const char *trigger_name; + enum lttng_trigger_status status; + + status = lttng_trigger_get_name(trigger, &trigger_name); + if (status == LTTNG_TRIGGER_STATUS_OK) { + hash = hash_key_str(trigger_name, lttng_ht_seed); } + + trigger_creds = lttng_trigger_get_credentials(trigger); + hash ^= hash_key_ulong((void *) (unsigned long) LTTNG_OPTIONAL_GET(trigger_creds->uid), + lttng_ht_seed); + + return hash; } static @@ -475,6 +437,18 @@ unsigned long hash_channel_key(struct channel_key *key) return key_hash ^ domain_hash; } +static +unsigned long hash_client_socket(int socket) +{ + return hash_key_ulong((void *) (unsigned long) socket, lttng_ht_seed); +} + +static +unsigned long hash_client_id(notification_client_id id) +{ + return hash_key_u64(&id, lttng_ht_seed); +} + /* * Get the type of object to which a given condition applies. Bindings let * the notification system evaluate a trigger's condition when a given @@ -495,6 +469,8 @@ enum lttng_object_type get_condition_binding_object( case LTTNG_CONDITION_TYPE_SESSION_ROTATION_ONGOING: case LTTNG_CONDITION_TYPE_SESSION_ROTATION_COMPLETED: return LTTNG_OBJECT_TYPE_SESSION; + case LTTNG_CONDITION_TYPE_ON_EVENT: + return LTTNG_OBJECT_TYPE_NONE; default: return LTTNG_OBJECT_TYPE_UNKNOWN; } @@ -666,7 +642,150 @@ error: return NULL; } -/* RCU read lock must be held by the caller. */ +LTTNG_HIDDEN +bool notification_client_list_get(struct notification_client_list *list) +{ + return urcu_ref_get_unless_zero(&list->ref); +} + +static +void free_notification_client_list_rcu(struct rcu_head *node) +{ + free(caa_container_of(node, struct notification_client_list, + rcu_node)); +} + +static +void notification_client_list_release(struct urcu_ref *list_ref) +{ + struct notification_client_list *list = + container_of(list_ref, typeof(*list), ref); + struct notification_client_list_element *client_list_element, *tmp; + + lttng_condition_put(list->condition); + + if (list->notification_trigger_clients_ht) { + rcu_read_lock(); + + cds_lfht_del(list->notification_trigger_clients_ht, + &list->notification_trigger_clients_ht_node); + rcu_read_unlock(); + list->notification_trigger_clients_ht = NULL; + } + cds_list_for_each_entry_safe(client_list_element, tmp, + &list->clients_list, node) { + free(client_list_element); + } + + assert(cds_list_empty(&list->triggers_list)); + + pthread_mutex_destroy(&list->lock); + call_rcu(&list->rcu_node, free_notification_client_list_rcu); +} + +static +bool condition_applies_to_client(const struct lttng_condition *condition, + struct notification_client *client) +{ + bool applies = false; + struct lttng_condition_list_element *condition_list_element; + + cds_list_for_each_entry(condition_list_element, &client->condition_list, + node) { + applies = lttng_condition_is_equal( + condition_list_element->condition, + condition); + if (applies) { + break; + } + } + + return applies; +} + +static +struct notification_client_list *notification_client_list_create( + struct notification_thread_state *state, + const struct lttng_condition *condition) +{ + struct notification_client *client; + struct cds_lfht_iter iter; + struct notification_client_list *client_list; + + client_list = zmalloc(sizeof(*client_list)); + if (!client_list) { + PERROR("Failed to allocate notification client list"); + goto end; + } + + pthread_mutex_init(&client_list->lock, NULL); + /* + * The trigger that owns the condition has the first reference to this + * client list. + */ + urcu_ref_init(&client_list->ref); + cds_lfht_node_init(&client_list->notification_trigger_clients_ht_node); + CDS_INIT_LIST_HEAD(&client_list->clients_list); + CDS_INIT_LIST_HEAD(&client_list->triggers_list); + + /* + * Create a copy of the condition so that it's independent of any + * trigger. The client list may outlive the trigger object (which owns + * the condition) that is used to create it. + */ + client_list->condition = lttng_condition_copy(condition); + + /* Build a list of clients to which this new condition applies. */ + cds_lfht_for_each_entry (state->client_socket_ht, &iter, client, + client_socket_ht_node) { + struct notification_client_list_element *client_list_element; + + if (!condition_applies_to_client(condition, client)) { + continue; + } + + client_list_element = zmalloc(sizeof(*client_list_element)); + if (!client_list_element) { + goto error_put_client_list; + } + + CDS_INIT_LIST_HEAD(&client_list_element->node); + client_list_element->client = client; + cds_list_add(&client_list_element->node, &client_list->clients_list); + } + + client_list->notification_trigger_clients_ht = + state->notification_trigger_clients_ht; + + rcu_read_lock(); + /* + * Add the client list to the global list of client list. + */ + cds_lfht_add_unique(state->notification_trigger_clients_ht, + lttng_condition_hash(client_list->condition), + match_client_list_condition, + client_list->condition, + &client_list->notification_trigger_clients_ht_node); + rcu_read_unlock(); + goto end; + +error_put_client_list: + notification_client_list_put(client_list); + client_list = NULL; + +end: + return client_list; +} + +void notification_client_list_put(struct notification_client_list *list) +{ + if (!list) { + return; + } + return urcu_ref_put(&list->ref, notification_client_list_release); +} + +/* Provides a reference to the returned list. */ static struct notification_client_list *get_client_list_from_condition( struct notification_thread_state *state, @@ -674,20 +793,25 @@ struct notification_client_list *get_client_list_from_condition( { struct cds_lfht_node *node; struct cds_lfht_iter iter; + struct notification_client_list *list = NULL; + rcu_read_lock(); cds_lfht_lookup(state->notification_trigger_clients_ht, lttng_condition_hash(condition), match_client_list_condition, condition, &iter); node = cds_lfht_iter_get_node(&iter); + if (node) { + list = container_of(node, struct notification_client_list, + notification_trigger_clients_ht_node); + list = notification_client_list_get(list) ? list : NULL; + } - return node ? caa_container_of(node, - struct notification_client_list, - notification_trigger_ht_node) : NULL; + rcu_read_unlock(); + return list; } -/* This function must be called with the RCU read lock held. */ static int evaluate_channel_condition_for_client( const struct lttng_condition *condition, @@ -703,6 +827,8 @@ int evaluate_channel_condition_for_client( struct channel_state_sample *last_sample = NULL; struct lttng_channel_trigger_list *channel_trigger_list = NULL; + rcu_read_lock(); + /* Find the channel associated with the condition. */ cds_lfht_for_each_entry(state->channel_triggers_ht, &iter, channel_trigger_list, channel_triggers_ht_node) { @@ -777,6 +903,7 @@ int evaluate_channel_condition_for_client( *session_uid = channel_info->session_info->uid; *session_gid = channel_info->session_info->gid; end: + rcu_read_unlock(); return ret; } @@ -812,7 +939,6 @@ end: return session_name; } -/* This function must be called with the RCU read lock held. */ static int evaluate_session_condition_for_client( const struct lttng_condition *condition, @@ -826,6 +952,7 @@ int evaluate_session_condition_for_client( const char *session_name; struct session_info *session_info = NULL; + rcu_read_lock(); session_name = get_condition_session_name(condition); /* Find the session associated with the trigger. */ @@ -879,10 +1006,10 @@ int evaluate_session_condition_for_client( end_session_put: session_info_put(session_info); end: + rcu_read_unlock(); return ret; } -/* This function must be called with the RCU read lock held. */ static int evaluate_condition_for_client(const struct lttng_trigger *trigger, const struct lttng_condition *condition, @@ -891,7 +1018,9 @@ int evaluate_condition_for_client(const struct lttng_trigger *trigger, { int ret; struct lttng_evaluation *evaluation = NULL; - struct notification_client_list client_list = { 0 }; + struct notification_client_list client_list = { + .lock = PTHREAD_MUTEX_INITIALIZER, + }; struct notification_client_list_element client_list_element = { 0 }; uid_t object_uid = 0; gid_t object_gid = 0; @@ -911,6 +1040,7 @@ int evaluate_condition_for_client(const struct lttng_trigger *trigger, &evaluation, &object_uid, &object_gid); break; case LTTNG_OBJECT_TYPE_NONE: + DBG("[notification-thread] Newly subscribed-to condition not bound to object, nothing to evaluate"); ret = 0; goto end; case LTTNG_OBJECT_TYPE_UNKNOWN: @@ -933,13 +1063,12 @@ int evaluate_condition_for_client(const struct lttng_trigger *trigger, * Create a temporary client list with the client currently * subscribing. */ - cds_lfht_node_init(&client_list.notification_trigger_ht_node); - CDS_INIT_LIST_HEAD(&client_list.list); - client_list.trigger = trigger; + cds_lfht_node_init(&client_list.notification_trigger_clients_ht_node); + CDS_INIT_LIST_HEAD(&client_list.clients_list); CDS_INIT_LIST_HEAD(&client_list_element.node); client_list_element.client = client; - cds_list_add(&client_list_element.node, &client_list.list); + cds_list_add(&client_list_element.node, &client_list.clients_list); /* Send evaluation result to the newly-subscribed client. */ DBG("[notification-thread] Newly subscribed-to condition evaluated to true, notifying client"); @@ -957,9 +1086,10 @@ int notification_thread_client_subscribe(struct notification_client *client, enum lttng_notification_channel_status *_status) { int ret = 0; - struct notification_client_list *client_list; + struct notification_client_list *client_list = NULL; struct lttng_condition_list_element *condition_list_element = NULL; struct notification_client_list_element *client_list_element = NULL; + struct lttng_trigger_ht_element *trigger_ht_element; enum lttng_notification_channel_status status = LTTNG_NOTIFICATION_CHANNEL_STATUS_OK; @@ -986,8 +1116,6 @@ int notification_thread_client_subscribe(struct notification_client *client, goto error; } - rcu_read_lock(); - /* * Add the newly-subscribed condition to the client's subscription list. */ @@ -1003,21 +1131,32 @@ int notification_thread_client_subscribe(struct notification_client *client, * since this trigger is not registered yet. */ free(client_list_element); - goto end_unlock; + goto end; } /* * The condition to which the client just subscribed is evaluated * at this point so that conditions that are already TRUE result * in a notification being sent out. + * + * Note the iteration on all triggers which share an identical + * `condition` than the one to which the client is registering. This is + * done to ensure that the client receives a distinct notification for + * all triggers that have a `notify` action that have this condition. */ - if (evaluate_condition_for_client(client_list->trigger, condition, - client, state)) { - WARN("[notification-thread] Evaluation of a condition on client subscription failed, aborting."); - ret = -1; - free(client_list_element); - goto end_unlock; + pthread_mutex_lock(&client_list->lock); + cds_list_for_each_entry(trigger_ht_element, + &client_list->triggers_list, client_list_trigger_node) { + if (evaluate_condition_for_client(trigger_ht_element->trigger, condition, + client, state)) { + WARN("[notification-thread] Evaluation of a condition on client subscription failed, aborting."); + ret = -1; + free(client_list_element); + pthread_mutex_unlock(&client_list->lock); + goto end; + } } + pthread_mutex_unlock(&client_list->lock); /* * Add the client to the list of clients interested in a given trigger @@ -1026,13 +1165,17 @@ int notification_thread_client_subscribe(struct notification_client *client, */ client_list_element->client = client; CDS_INIT_LIST_HEAD(&client_list_element->node); - cds_list_add(&client_list_element->node, &client_list->list); -end_unlock: - rcu_read_unlock(); + + pthread_mutex_lock(&client_list->lock); + cds_list_add(&client_list_element->node, &client_list->clients_list); + pthread_mutex_unlock(&client_list->lock); end: if (_status) { *_status = status; } + if (client_list) { + notification_client_list_put(client_list); + } return ret; error: free(condition_list_element); @@ -1088,23 +1231,24 @@ int notification_thread_client_unsubscribe( * Remove the client from the list of clients interested the trigger * matching the condition. */ - rcu_read_lock(); client_list = get_client_list_from_condition(state, condition); if (!client_list) { - goto end_unlock; + goto end; } + pthread_mutex_lock(&client_list->lock); cds_list_for_each_entry_safe(client_list_element, client_tmp, - &client_list->list, node) { - if (client_list_element->client->socket != client->socket) { + &client_list->clients_list, node) { + if (client_list_element->client->id != client->id) { continue; } cds_list_del(&client_list_element->node); free(client_list_element); break; } -end_unlock: - rcu_read_unlock(); + pthread_mutex_unlock(&client_list->lock); + notification_client_list_put(client_list); + client_list = NULL; end: lttng_condition_destroy(condition); if (_status) { @@ -1123,24 +1267,22 @@ static void notification_client_destroy(struct notification_client *client, struct notification_thread_state *state) { - struct lttng_condition_list_element *condition_list_element, *tmp; - if (!client) { return; } - /* Release all conditions to which the client was subscribed. */ - cds_list_for_each_entry_safe(condition_list_element, tmp, - &client->condition_list, node) { - (void) notification_thread_client_unsubscribe(client, - condition_list_element->condition, state, NULL); - } - + /* + * The client object is not reachable by other threads, no need to lock + * the client here. + */ if (client->socket >= 0) { (void) lttcomm_close_unix_sock(client->socket); + client->socket = -1; } - lttng_dynamic_buffer_reset(&client->communication.inbound.buffer); - lttng_dynamic_buffer_reset(&client->communication.outbound.buffer); + client->communication.active = false; + lttng_payload_reset(&client->communication.inbound.payload); + lttng_payload_reset(&client->communication.outbound.payload); + pthread_mutex_destroy(&client->lock); call_rcu(&client->rcu_node, free_notification_client_rcu); } @@ -1157,8 +1299,8 @@ struct notification_client *get_client_from_socket(int socket, struct notification_client *client = NULL; cds_lfht_lookup(state->client_socket_ht, - hash_key_ulong((void *) (unsigned long) socket, lttng_ht_seed), - match_client, + hash_client_socket(socket), + match_client_socket, (void *) (unsigned long) socket, &iter); node = cds_lfht_iter_get_node(&iter); @@ -1172,14 +1314,42 @@ end: return client; } +/* + * Call with rcu_read_lock held (and hold for the lifetime of the returned + * client pointer). + */ static -bool buffer_usage_condition_applies_to_channel( - const struct lttng_condition *condition, - const struct channel_info *channel_info) +struct notification_client *get_client_from_id(notification_client_id id, + struct notification_thread_state *state) { - enum lttng_condition_status status; - enum lttng_domain_type condition_domain; - const char *condition_session_name = NULL; + struct cds_lfht_iter iter; + struct cds_lfht_node *node; + struct notification_client *client = NULL; + + cds_lfht_lookup(state->client_id_ht, + hash_client_id(id), + match_client_id, + &id, + &iter); + node = cds_lfht_iter_get_node(&iter); + if (!node) { + goto end; + } + + client = caa_container_of(node, struct notification_client, + client_id_ht_node); +end: + return client; +} + +static +bool buffer_usage_condition_applies_to_channel( + const struct lttng_condition *condition, + const struct channel_info *channel_info) +{ + enum lttng_condition_status status; + enum lttng_domain_type condition_domain; + const char *condition_session_name = NULL; const char *condition_channel_name = NULL; status = lttng_condition_buffer_usage_get_domain_type(condition, @@ -1261,25 +1431,6 @@ fail: return false; } -static -bool trigger_applies_to_client(struct lttng_trigger *trigger, - struct notification_client *client) -{ - bool applies = false; - struct lttng_condition_list_element *condition_list_element; - - cds_list_for_each_entry(condition_list_element, &client->condition_list, - node) { - applies = lttng_condition_is_equal( - condition_list_element->condition, - lttng_trigger_get_condition(trigger)); - if (applies) { - break; - } - } - return applies; -} - /* Must be called with RCU read lock held. */ static struct lttng_session_trigger_list *get_session_trigger_list( @@ -1375,7 +1526,7 @@ void lttng_session_trigger_list_destroy(struct lttng_session_trigger_list *list) static int lttng_session_trigger_list_add(struct lttng_session_trigger_list *list, - const struct lttng_trigger *trigger) + struct lttng_trigger *trigger) { int ret = 0; struct lttng_trigger_list_element *new_element = @@ -1549,7 +1700,7 @@ int handle_notification_thread_command_add_channel( DBG("[notification-thread] Adding channel %s from session %s, channel key = %" PRIu64 " in %s domain", channel_name, session_name, channel_key_int, - channel_domain == LTTNG_DOMAIN_KERNEL ? "kernel" : "user space"); + lttng_domain_type_str(channel_domain)); CDS_INIT_LIST_HEAD(&trigger_list); @@ -1650,7 +1801,7 @@ int handle_notification_thread_command_remove_channel( struct channel_info *channel_info; DBG("[notification-thread] Removing channel key = %" PRIu64 " in %s domain", - channel_key, domain == LTTNG_DOMAIN_KERNEL ? "kernel" : "user space"); + channel_key, lttng_domain_type_str(domain)); rcu_read_lock(); @@ -1732,6 +1883,10 @@ int handle_notification_thread_command_session_rotation( struct lttng_session_trigger_list *trigger_list; struct lttng_trigger_list_element *trigger_list_element; struct session_info *session_info; + const struct lttng_credentials session_creds = { + .uid = LTTNG_OPTIONAL_INIT_VALUE(session_uid), + .gid = LTTNG_OPTIONAL_INIT_VALUE(session_gid), + }; rcu_read_lock(); @@ -1757,11 +1912,11 @@ int handle_notification_thread_command_session_rotation( cds_list_for_each_entry(trigger_list_element, &trigger_list->list, node) { const struct lttng_condition *condition; - const struct lttng_action *action; - const struct lttng_trigger *trigger; + struct lttng_trigger *trigger; struct notification_client_list *client_list; struct lttng_evaluation *evaluation = NULL; enum lttng_condition_type condition_type; + enum action_executor_status executor_status; trigger = trigger_list_element->trigger; condition = lttng_trigger_get_const_condition(trigger); @@ -1776,23 +1931,7 @@ int handle_notification_thread_command_session_rotation( continue; } - action = lttng_trigger_get_const_action(trigger); - - /* Notify actions are the only type currently supported. */ - assert(lttng_action_get_type_const(action) == - LTTNG_ACTION_TYPE_NOTIFY); - client_list = get_client_list_from_condition(state, condition); - assert(client_list); - - if (cds_list_empty(&client_list->list)) { - /* - * No clients interested in the evaluation's result, - * skip it. - */ - continue; - } - if (cmd_type == NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_ONGOING) { evaluation = lttng_evaluation_session_rotation_ongoing_create( trace_archive_chunk_id); @@ -1805,17 +1944,47 @@ int handle_notification_thread_command_session_rotation( /* Internal error */ ret = -1; cmd_result = LTTNG_ERR_UNK; - goto end; + goto put_list; + } + + /* + * Ownership of `evaluation` transferred to the action executor + * no matter the result. + */ + executor_status = action_executor_enqueue(state->executor, + trigger, evaluation, &session_creds, + client_list); + evaluation = NULL; + switch (executor_status) { + case ACTION_EXECUTOR_STATUS_OK: + break; + case ACTION_EXECUTOR_STATUS_ERROR: + case ACTION_EXECUTOR_STATUS_INVALID: + /* + * TODO Add trigger identification (name/id) when + * it is added to the API. + */ + ERR("Fatal error occurred while enqueuing action associated with session rotation trigger"); + ret = -1; + goto put_list; + case ACTION_EXECUTOR_STATUS_OVERFLOW: + /* + * TODO Add trigger identification (name/id) when + * it is added to the API. + * + * Not a fatal error. + */ + WARN("No space left when enqueuing action associated with session rotation trigger"); + ret = 0; + goto put_list; + default: + abort(); } - /* Dispatch evaluation result to all clients. */ - ret = send_evaluation_to_clients(trigger_list_element->trigger, - evaluation, client_list, state, - session_info->uid, - session_info->gid); - lttng_evaluation_destroy(evaluation); +put_list: + notification_client_list_put(client_list); if (caa_unlikely(ret)) { - goto end; + break; } } end: @@ -1826,25 +1995,309 @@ end: } static -int condition_is_supported(struct lttng_condition *condition) +int handle_notification_thread_command_add_tracer_event_source( + struct notification_thread_state *state, + int tracer_event_source_fd, + enum lttng_domain_type domain_type, + enum lttng_error_code *_cmd_result) +{ + int ret = 0; + enum lttng_error_code cmd_result = LTTNG_OK; + struct notification_event_tracer_event_source_element *element = NULL; + + element = zmalloc(sizeof(*element)); + if (!element) { + cmd_result = LTTNG_ERR_NOMEM; + ret = -1; + goto end; + } + + element->fd = tracer_event_source_fd; + element->domain = domain_type; + + cds_list_add(&element->node, &state->tracer_event_sources_list); + + DBG3("[notification-thread] Adding tracer event source fd to poll set: tracer_event_source_fd = %d, domain = '%s'", + tracer_event_source_fd, + lttng_domain_type_str(domain_type)); + + /* Adding the read side pipe to the event poll. */ + ret = lttng_poll_add(&state->events, tracer_event_source_fd, LPOLLIN | LPOLLERR); + if (ret < 0) { + ERR("[notification-thread] Failed to add tracer event source to poll set: tracer_event_source_fd = %d, domain = '%s'", + tracer_event_source_fd, + lttng_domain_type_str(element->domain)); + cds_list_del(&element->node); + free(element); + goto end; + } + + element->is_fd_in_poll_set = true; + +end: + *_cmd_result = cmd_result; + return ret; +} + +static +int drain_event_notifier_notification_pipe( + struct notification_thread_state *state, + int pipe, enum lttng_domain_type domain) +{ + struct lttng_poll_event events = {0}; + int ret; + + ret = lttng_poll_create(&events, 1, LTTNG_CLOEXEC); + if (ret < 0) { + ERR("[notification-thread] Error creating lttng_poll_event"); + goto end; + } + + ret = lttng_poll_add(&events, pipe, LPOLLIN); + if (ret < 0) { + ERR("[notification-thread] Error adding fd event notifier notification pipe to lttng_poll_event: fd = %d", + pipe); + goto end; + } + + while (true) { + /* + * Continue to consume notifications as long as there are new + * ones coming in. The tracer has been asked to stop producing + * them. + * + * LPOLLIN is explicitly checked since LPOLLHUP is implicitly + * monitored (on Linux, at least) and will be returned when + * the pipe is closed but empty. + */ + ret = lttng_poll_wait_interruptible(&events, 0); + if (ret == 0 || (LTTNG_POLL_GETEV(&events, 0) & LPOLLIN) == 0) { + /* No more notification to be read on this pipe. */ + ret = 0; + goto end; + } else if (ret < 0) { + PERROR("Failed on lttng_poll_wait_interruptible() call"); + ret = -1; + goto end; + } + + ret = handle_one_event_notifier_notification(state, pipe, domain); + if (ret) { + ERR("[notification-thread] Error consuming an event notifier notification from pipe: fd = %d", + pipe); + } + } +end: + lttng_poll_clean(&events); + return ret; +} + +static +int handle_notification_thread_command_remove_tracer_event_source( + struct notification_thread_state *state, + int tracer_event_source_fd, + enum lttng_error_code *_cmd_result) +{ + int ret = 0; + bool found = false; + enum lttng_error_code cmd_result = LTTNG_OK; + struct notification_event_tracer_event_source_element *source_element = NULL, *tmp; + + cds_list_for_each_entry_safe(source_element, tmp, + &state->tracer_event_sources_list, node) { + if (source_element->fd != tracer_event_source_fd) { + continue; + } + + DBG("[notification-thread] Removed tracer event source from poll set: tracer_event_source_fd = %d, domain = '%s'", + tracer_event_source_fd, + lttng_domain_type_str(source_element->domain)); + cds_list_del(&source_element->node); + found = true; + break; + } + + if (!found) { + /* + * This is temporarily allowed since the poll activity set is + * not properly cleaned-up for the moment. This is adressed in + * an upcoming fix. + */ + source_element = NULL; + goto end; + } + + if (!source_element->is_fd_in_poll_set) { + /* Skip the poll set removal. */ + goto end; + } + + DBG3("[notification-thread] Removing tracer event source from poll set: tracer_event_source_fd = %d, domain = '%s'", + tracer_event_source_fd, + lttng_domain_type_str(source_element->domain)); + + /* Removing the fd from the event poll set. */ + ret = lttng_poll_del(&state->events, tracer_event_source_fd); + if (ret < 0) { + ERR("[notification-thread] Failed to remove tracer event source from poll set: tracer_event_source_fd = %d, domain = '%s'", + tracer_event_source_fd, + lttng_domain_type_str(source_element->domain)); + cmd_result = LTTNG_ERR_FATAL; + goto end; + } + + source_element->is_fd_in_poll_set = false; + + ret = drain_event_notifier_notification_pipe(state, tracer_event_source_fd, + source_element->domain); + if (ret) { + ERR("[notification-thread] Error draining event notifier notification: tracer_event_source_fd = %d, domain = %s", + tracer_event_source_fd, + lttng_domain_type_str(source_element->domain)); + cmd_result = LTTNG_ERR_FATAL; + goto end; + } + + /* + * The drain_event_notifier_notification_pipe() call might have read + * data from an fd that we received in event in the latest _poll_wait() + * call. Make sure the thread call poll_wait() again to ensure we have + * a clean state. + */ + state->restart_poll = true; + +end: + free(source_element); + *_cmd_result = cmd_result; + return ret; +} + +static +int condition_on_event_update_error_count(struct lttng_trigger *trigger) +{ + int ret = 0; + uint64_t error_count = 0; + struct lttng_condition *condition; + enum event_notifier_error_accounting_status status; + + condition = lttng_trigger_get_condition(trigger); + assert(lttng_condition_get_type(condition) == + LTTNG_CONDITION_TYPE_ON_EVENT); + + status = event_notifier_error_accounting_get_count(trigger, &error_count); + if (status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { + uid_t trigger_owner_uid; + const char *trigger_name; + const enum lttng_trigger_status trigger_status = + lttng_trigger_get_owner_uid( + trigger, &trigger_owner_uid); + + assert(trigger_status == LTTNG_TRIGGER_STATUS_OK); + if (lttng_trigger_get_name(trigger, &trigger_name) != LTTNG_TRIGGER_STATUS_OK) { + trigger_name = "(unnamed)"; + } + + ERR("Failed to get event notifier error count of trigger for update: trigger owner = %d, trigger name = '%s'", + trigger_owner_uid, trigger_name); + ret = -1; + } + + lttng_condition_on_event_set_error_count(condition, error_count); + return ret; +} + +int handle_notification_thread_remove_tracer_event_source_no_result( + struct notification_thread_state *state, + int tracer_event_source_fd) { int ret; + enum lttng_error_code cmd_result; + + ret = handle_notification_thread_command_remove_tracer_event_source( + state, tracer_event_source_fd, &cmd_result); + (void) cmd_result; + return ret; +} + +static int handle_notification_thread_command_list_triggers( + struct notification_thread_handle *handle, + struct notification_thread_state *state, + uid_t client_uid, + struct lttng_triggers **triggers, + enum lttng_error_code *_cmd_result) +{ + int ret = 0; + enum lttng_error_code cmd_result = LTTNG_OK; + struct cds_lfht_iter iter; + struct lttng_trigger_ht_element *trigger_ht_element; + struct lttng_triggers *local_triggers = NULL; + const struct lttng_credentials *creds; + + rcu_read_lock(); + + local_triggers = lttng_triggers_create(); + if (!local_triggers) { + /* Not a fatal error. */ + cmd_result = LTTNG_ERR_NOMEM; + goto end; + } + + cds_lfht_for_each_entry(state->triggers_ht, &iter, + trigger_ht_element, node) { + /* + * Only return the triggers to which the client has access. + * The root user has visibility over all triggers. + */ + creds = lttng_trigger_get_credentials(trigger_ht_element->trigger); + if (client_uid != lttng_credentials_get_uid(creds) && client_uid != 0) { + continue; + } + + if (lttng_trigger_needs_tracer_notifier(trigger_ht_element->trigger)) { + ret = condition_on_event_update_error_count( + trigger_ht_element->trigger); + assert(!ret); + } + + ret = lttng_triggers_add(local_triggers, + trigger_ht_element->trigger); + if (ret < 0) { + /* Not a fatal error. */ + ret = 0; + cmd_result = LTTNG_ERR_NOMEM; + goto end; + } + } + + /* Transferring ownership to the caller. */ + *triggers = local_triggers; + local_triggers = NULL; + +end: + rcu_read_unlock(); + lttng_triggers_destroy(local_triggers); + *_cmd_result = cmd_result; + return ret; +} + +static +bool condition_is_supported(struct lttng_condition *condition) +{ + bool is_supported; switch (lttng_condition_get_type(condition)) { case LTTNG_CONDITION_TYPE_BUFFER_USAGE_LOW: case LTTNG_CONDITION_TYPE_BUFFER_USAGE_HIGH: { + int ret; enum lttng_domain_type domain; ret = lttng_condition_buffer_usage_get_domain_type(condition, &domain); - if (ret) { - ret = -1; - goto end; - } + assert(ret == 0); if (domain != LTTNG_DOMAIN_KERNEL) { - ret = 1; + is_supported = true; goto end; } @@ -1852,20 +2305,48 @@ int condition_is_supported(struct lttng_condition *condition) * Older kernel tracers don't expose the API to monitor their * buffers. Therefore, we reject triggers that require that * mechanism to be available to be evaluated. + * + * Assume unsupported on error. + */ + is_supported = kernel_supports_ring_buffer_snapshot_sample_positions() == 1; + break; + } + case LTTNG_CONDITION_TYPE_ON_EVENT: + { + const struct lttng_event_rule *event_rule; + enum lttng_domain_type domain; + const enum lttng_condition_status status = + lttng_condition_on_event_get_rule( + condition, &event_rule); + + assert(status == LTTNG_CONDITION_STATUS_OK); + + domain = lttng_event_rule_get_domain_type(event_rule); + if (domain != LTTNG_DOMAIN_KERNEL) { + is_supported = true; + goto end; + } + + /* + * Older kernel tracers can't emit notification. Therefore, we + * reject triggers that require that mechanism to be available + * to be evaluated. + * + * Assume unsupported on error. */ - ret = kernel_supports_ring_buffer_snapshot_sample_positions(); + is_supported = kernel_supports_event_notifiers() == 1; break; } default: - ret = 1; + is_supported = true; } end: - return ret; + return is_supported; } /* Must be called with RCU read lock held. */ static -int bind_trigger_to_matching_session(const struct lttng_trigger *trigger, +int bind_trigger_to_matching_session(struct lttng_trigger *trigger, struct notification_thread_state *state) { int ret = 0; @@ -1911,7 +2392,7 @@ end: /* Must be called with RCU read lock held. */ static -int bind_trigger_to_matching_channels(const struct lttng_trigger *trigger, +int bind_trigger_to_matching_channels(struct lttng_trigger *trigger, struct notification_thread_state *state) { int ret = 0; @@ -1955,52 +2436,227 @@ end: return ret; } -/* - * FIXME A client's credentials are not checked when registering a trigger, nor - * are they stored alongside with the trigger. - * - * The effects of this are benign since: - * - The client will succeed in registering the trigger, as it is valid, - * - The trigger will, internally, be bound to the channel/session, - * - The notifications will not be sent since the client's credentials - * are checked against the channel at that moment. - * - * If this function returns a non-zero value, it means something is - * fundamentally broken and the whole subsystem/thread will be torn down. - * - * If a non-fatal error occurs, just set the cmd_result to the appropriate - * error code. - */ static -int handle_notification_thread_command_register_trigger( - struct notification_thread_state *state, - struct lttng_trigger *trigger, - enum lttng_error_code *cmd_result) +bool is_trigger_action_notify(const struct lttng_trigger *trigger) { - int ret = 0; - struct lttng_condition *condition; - struct notification_client *client; - struct notification_client_list *client_list = NULL; + bool is_notify = false; + unsigned int i, count; + enum lttng_action_status action_status; + const struct lttng_action *action = + lttng_trigger_get_const_action(trigger); + enum lttng_action_type action_type; + + assert(action); + action_type = lttng_action_get_type(action); + if (action_type == LTTNG_ACTION_TYPE_NOTIFY) { + is_notify = true; + goto end; + } else if (action_type != LTTNG_ACTION_TYPE_GROUP) { + goto end; + } + + action_status = lttng_action_group_get_count(action, &count); + assert(action_status == LTTNG_ACTION_STATUS_OK); + + for (i = 0; i < count; i++) { + const struct lttng_action *inner_action = + lttng_action_group_get_at_index( + action, i); + + action_type = lttng_action_get_type(inner_action); + if (action_type == LTTNG_ACTION_TYPE_NOTIFY) { + is_notify = true; + goto end; + } + } + +end: + return is_notify; +} + +static bool trigger_name_taken(struct notification_thread_state *state, + const struct lttng_trigger *trigger) +{ + struct cds_lfht_iter iter; + + /* + * No duplicata is allowed in the triggers_by_name_uid_ht. + * The match is done against the trigger name and uid. + */ + cds_lfht_lookup(state->triggers_by_name_uid_ht, + hash_trigger_by_name_uid(trigger), + match_trigger_by_name_uid, + trigger, + &iter); + return !!cds_lfht_iter_get_node(&iter); +} + +static +enum lttng_error_code generate_trigger_name( + struct notification_thread_state *state, + struct lttng_trigger *trigger, const char **name) +{ + enum lttng_error_code ret_code = LTTNG_OK; + bool taken = false; + enum lttng_trigger_status status; + + do { + const int ret = lttng_trigger_generate_name(trigger, + state->trigger_id.name_offset++); + if (ret) { + /* The only reason this can fail right now. */ + ret_code = LTTNG_ERR_NOMEM; + break; + } + + status = lttng_trigger_get_name(trigger, name); + assert(status == LTTNG_TRIGGER_STATUS_OK); + + taken = trigger_name_taken(state, trigger); + } while (taken || state->trigger_id.name_offset == UINT64_MAX); + + return ret_code; +} + +static inline +void notif_thread_state_remove_trigger_ht_elem( + struct notification_thread_state *state, + struct lttng_trigger_ht_element *trigger_ht_element) +{ + assert(state); + assert(trigger_ht_element); + + cds_lfht_del(state->triggers_ht, &trigger_ht_element->node); + cds_lfht_del(state->triggers_by_name_uid_ht, &trigger_ht_element->node_by_name_uid); +} + +static +enum lttng_error_code setup_tracer_notifier( + struct notification_thread_state *state, + struct lttng_trigger *trigger) +{ + enum lttng_error_code ret; + enum event_notifier_error_accounting_status error_accounting_status; + struct cds_lfht_node *node; + uint64_t error_counter_index = 0; + struct lttng_condition *condition = lttng_trigger_get_condition(trigger); + struct notification_trigger_tokens_ht_element *trigger_tokens_ht_element = NULL; + + trigger_tokens_ht_element = zmalloc(sizeof(*trigger_tokens_ht_element)); + if (!trigger_tokens_ht_element) { + ret = LTTNG_ERR_NOMEM; + goto end; + } + + /* Add trigger token to the trigger_tokens_ht. */ + cds_lfht_node_init(&trigger_tokens_ht_element->node); + trigger_tokens_ht_element->token = LTTNG_OPTIONAL_GET(trigger->tracer_token); + trigger_tokens_ht_element->trigger = trigger; + + node = cds_lfht_add_unique(state->trigger_tokens_ht, + hash_key_u64(&trigger_tokens_ht_element->token, lttng_ht_seed), + match_trigger_token, + &trigger_tokens_ht_element->token, + &trigger_tokens_ht_element->node); + if (node != &trigger_tokens_ht_element->node) { + ret = LTTNG_ERR_TRIGGER_EXISTS; + goto error_free_ht_element; + } + + error_accounting_status = event_notifier_error_accounting_register_event_notifier( + trigger, &error_counter_index); + if (error_accounting_status != EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_OK) { + if (error_accounting_status == EVENT_NOTIFIER_ERROR_ACCOUNTING_STATUS_NO_INDEX_AVAILABLE) { + DBG("Trigger group error accounting counter full."); + ret = LTTNG_ERR_EVENT_NOTIFIER_ERROR_ACCOUNTING_FULL; + } else { + ERR("Error registering trigger for error accounting"); + ret = LTTNG_ERR_EVENT_NOTIFIER_REGISTRATION; + } + + goto error_remove_ht_element; + } + + lttng_condition_on_event_set_error_counter_index( + condition, error_counter_index); + + ret = LTTNG_OK; + goto end; + +error_remove_ht_element: + cds_lfht_del(state->trigger_tokens_ht, &trigger_tokens_ht_element->node); +error_free_ht_element: + free(trigger_tokens_ht_element); +end: + return ret; +} + +/* + * FIXME A client's credentials are not checked when registering a trigger. + * + * The effects of this are benign since: + * - The client will succeed in registering the trigger, as it is valid, + * - The trigger will, internally, be bound to the channel/session, + * - The notifications will not be sent since the client's credentials + * are checked against the channel at that moment. + * + * If this function returns a non-zero value, it means something is + * fundamentally broken and the whole subsystem/thread will be torn down. + * + * If a non-fatal error occurs, just set the cmd_result to the appropriate + * error code. + */ +static +int handle_notification_thread_command_register_trigger( + struct notification_thread_state *state, + struct lttng_trigger *trigger, + enum lttng_error_code *cmd_result) +{ + int ret = 0; + struct lttng_condition *condition; + struct notification_client_list *client_list = NULL; struct lttng_trigger_ht_element *trigger_ht_element = NULL; - struct notification_client_list_element *client_list_element, *tmp; struct cds_lfht_node *node; - struct cds_lfht_iter iter; + const char* trigger_name; bool free_trigger = true; + struct lttng_evaluation *evaluation = NULL; + struct lttng_credentials object_creds; + uid_t object_uid; + gid_t object_gid; + enum action_executor_status executor_status; + const uint64_t trigger_tracer_token = + state->trigger_id.next_tracer_token++; rcu_read_lock(); + /* Set the trigger's tracer token. */ + lttng_trigger_set_tracer_token(trigger, trigger_tracer_token); + + if (lttng_trigger_get_name(trigger, &trigger_name) == + LTTNG_TRIGGER_STATUS_UNSET) { + const enum lttng_error_code ret_code = generate_trigger_name( + state, trigger, &trigger_name); + + if (ret_code != LTTNG_OK) { + /* Fatal error. */ + ret = -1; + *cmd_result = ret_code; + goto error; + } + } else if (trigger_name_taken(state, trigger)) { + /* Not a fatal error. */ + *cmd_result = LTTNG_ERR_TRIGGER_EXISTS; + ret = 0; + goto error; + } + condition = lttng_trigger_get_condition(trigger); assert(condition); - ret = condition_is_supported(condition); - if (ret < 0) { - goto error; - } else if (ret == 0) { + /* Some conditions require tracers to implement a minimal ABI version. */ + if (!condition_is_supported(condition)) { *cmd_result = LTTNG_ERR_NOT_SUPPORTED; goto error; - } else { - /* Feature is supported, continue. */ - ret = 0; } trigger_ht_element = zmalloc(sizeof(*trigger_ht_element)); @@ -2011,12 +2667,13 @@ int handle_notification_thread_command_register_trigger( /* Add trigger to the trigger_ht. */ cds_lfht_node_init(&trigger_ht_element->node); + cds_lfht_node_init(&trigger_ht_element->node_by_name_uid); trigger_ht_element->trigger = trigger; node = cds_lfht_add_unique(state->triggers_ht, lttng_condition_hash(condition), - match_condition, - condition, + match_trigger, + trigger, &trigger_ht_element->node); if (node != &trigger_ht_element->node) { /* Not a fatal error, simply report it to the client. */ @@ -2024,54 +2681,84 @@ int handle_notification_thread_command_register_trigger( goto error_free_ht_element; } + node = cds_lfht_add_unique(state->triggers_by_name_uid_ht, + hash_trigger_by_name_uid(trigger), + match_trigger_by_name_uid, + trigger, + &trigger_ht_element->node_by_name_uid); + if (node != &trigger_ht_element->node_by_name_uid) { + /* Not a fatal error, simply report it to the client. */ + cds_lfht_del(state->triggers_ht, &trigger_ht_element->node); + *cmd_result = LTTNG_ERR_TRIGGER_EXISTS; + goto error_free_ht_element; + } + /* - * Ownership of the trigger and of its wrapper was transfered to - * the triggers_ht. + * Some triggers might need a tracer notifier depending on its + * condition and actions. */ - trigger_ht_element = NULL; - free_trigger = false; + if (lttng_trigger_needs_tracer_notifier(trigger)) { + enum lttng_error_code error_code; + + error_code = setup_tracer_notifier(state, trigger); + if (error_code != LTTNG_OK) { + notif_thread_state_remove_trigger_ht_elem(state, + trigger_ht_element); + if (error_code == LTTNG_ERR_NOMEM) { + ret = -1; + } else { + *cmd_result = error_code; + ret = 0; + } + + goto error_free_ht_element; + } + } /* * The rest only applies to triggers that have a "notify" action. * It is not skipped as this is the only action type currently * supported. */ - client_list = zmalloc(sizeof(*client_list)); - if (!client_list) { - ret = -1; - goto error_free_ht_element; - } - cds_lfht_node_init(&client_list->notification_trigger_ht_node); - CDS_INIT_LIST_HEAD(&client_list->list); - client_list->trigger = trigger; - - /* Build a list of clients to which this new trigger applies. */ - cds_lfht_for_each_entry(state->client_socket_ht, &iter, client, - client_socket_ht_node) { - if (!trigger_applies_to_client(trigger, client)) { - continue; + if (is_trigger_action_notify(trigger)) { + /* + * Find or create the client list of this condition. It may + * already be present if another trigger is already registered + * with the same condition. + */ + client_list = get_client_list_from_condition(state, condition); + if (!client_list) { + /* + * No client list for this condition yet. We create new + * one and build it up. + */ + client_list = notification_client_list_create(state, condition); + if (!client_list) { + ERR("Error creating notification client list for trigger %s", trigger->name); + goto error_free_ht_element; + } } - client_list_element = zmalloc(sizeof(*client_list_element)); - if (!client_list_element) { - ret = -1; - goto error_free_client_list; - } - CDS_INIT_LIST_HEAD(&client_list_element->node); - client_list_element->client = client; - cds_list_add(&client_list_element->node, &client_list->list); + CDS_INIT_LIST_HEAD(&trigger_ht_element->client_list_trigger_node); + + pthread_mutex_lock(&client_list->lock); + cds_list_add(&trigger_ht_element->client_list_trigger_node, &client_list->triggers_list); + pthread_mutex_unlock(&client_list->lock); } - cds_lfht_add(state->notification_trigger_clients_ht, - lttng_condition_hash(condition), - &client_list->notification_trigger_ht_node); + /* + * Ownership of the trigger and of its wrapper was transfered to + * the triggers_ht. Same for token ht element if necessary. + */ + trigger_ht_element = NULL; + free_trigger = false; switch (get_condition_binding_object(condition)) { case LTTNG_OBJECT_TYPE_SESSION: /* Add the trigger to the list if it matches a known session. */ ret = bind_trigger_to_matching_session(trigger, state); if (ret) { - goto error_free_client_list; + goto error_free_ht_element; } break; case LTTNG_OBJECT_TYPE_CHANNEL: @@ -2081,21 +2768,24 @@ int handle_notification_thread_command_register_trigger( */ ret = bind_trigger_to_matching_channels(trigger, state); if (ret) { - goto error_free_client_list; + goto error_free_ht_element; } break; case LTTNG_OBJECT_TYPE_NONE: break; default: - ERR("[notification-thread] Unknown object type on which to bind a newly registered trigger was encountered"); + ERR("Unknown object type on which to bind a newly registered trigger was encountered"); ret = -1; - goto error_free_client_list; + goto error_free_ht_element; } /* - * Since there is nothing preventing clients from subscribing to a - * condition before the corresponding trigger is registered, we have - * to evaluate this new condition right away. + * The new trigger's condition must be evaluated against the current + * state. + * + * In the case of `notify` action, nothing preventing clients from + * subscribing to a condition before the corresponding trigger is + * registered, we have to evaluate this new condition right away. * * At some point, we were waiting for the next "evaluation" (e.g. on * reception of a channel sample) to evaluate this new condition, but @@ -2118,67 +2808,153 @@ int handle_notification_thread_command_register_trigger( * that the evaluations remain the same (true for samples n-1 and n) and * the client will never know that the condition has been met. */ - cds_list_for_each_entry_safe(client_list_element, tmp, - &client_list->list, node) { - ret = evaluate_condition_for_client(trigger, condition, - client_list_element->client, state); - if (ret) { - goto error_free_client_list; - } + switch (get_condition_binding_object(condition)) { + case LTTNG_OBJECT_TYPE_SESSION: + ret = evaluate_session_condition_for_client(condition, state, + &evaluation, &object_uid, + &object_gid); + LTTNG_OPTIONAL_SET(&object_creds.uid, object_uid); + LTTNG_OPTIONAL_SET(&object_creds.gid, object_gid); + break; + case LTTNG_OBJECT_TYPE_CHANNEL: + ret = evaluate_channel_condition_for_client(condition, state, + &evaluation, &object_uid, + &object_gid); + LTTNG_OPTIONAL_SET(&object_creds.uid, object_uid); + LTTNG_OPTIONAL_SET(&object_creds.gid, object_gid); + break; + case LTTNG_OBJECT_TYPE_NONE: + ret = 0; + break; + case LTTNG_OBJECT_TYPE_UNKNOWN: + default: + ret = -1; + break; + } + + if (ret) { + /* Fatal error. */ + goto error_free_ht_element; + } + + DBG("Newly registered trigger's condition evaluated to %s", + evaluation ? "true" : "false"); + if (!evaluation) { + /* Evaluation yielded nothing. Normal exit. */ + ret = 0; + goto success; } /* - * Client list ownership transferred to the - * notification_trigger_clients_ht. + * Ownership of `evaluation` transferred to the action executor + * no matter the result. */ - client_list = NULL; + executor_status = action_executor_enqueue(state->executor, trigger, + evaluation, &object_creds, client_list); + evaluation = NULL; + switch (executor_status) { + case ACTION_EXECUTOR_STATUS_OK: + break; + case ACTION_EXECUTOR_STATUS_ERROR: + case ACTION_EXECUTOR_STATUS_INVALID: + /* + * TODO Add trigger identification (name/id) when + * it is added to the API. + */ + ERR("Fatal error occurred while enqueuing action associated to newly registered trigger"); + ret = -1; + goto error_free_ht_element; + case ACTION_EXECUTOR_STATUS_OVERFLOW: + /* + * TODO Add trigger identification (name/id) when + * it is added to the API. + * + * Not a fatal error. + */ + WARN("No space left when enqueuing action associated to newly registered trigger"); + ret = 0; + goto success; + default: + abort(); + } +success: *cmd_result = LTTNG_OK; -error_free_client_list: - if (client_list) { - cds_list_for_each_entry_safe(client_list_element, tmp, - &client_list->list, node) { - free(client_list_element); - } - free(client_list); - } + DBG("Registered trigger: name = `%s`, tracer token = %" PRIu64, + trigger_name, trigger_tracer_token); + goto end; + error_free_ht_element: - free(trigger_ht_element); + if (trigger_ht_element) { + /* Delayed removal due to RCU constraint on delete. */ + call_rcu(&trigger_ht_element->rcu_node, + free_lttng_trigger_ht_element_rcu); + } error: if (free_trigger) { lttng_trigger_destroy(trigger); } +end: rcu_read_unlock(); return ret; } static -void free_notification_client_list_rcu(struct rcu_head *node) +void free_lttng_trigger_ht_element_rcu(struct rcu_head *node) { - free(caa_container_of(node, struct notification_client_list, + free(caa_container_of(node, struct lttng_trigger_ht_element, rcu_node)); } static -void free_lttng_trigger_ht_element_rcu(struct rcu_head *node) +void free_notification_trigger_tokens_ht_element_rcu(struct rcu_head *node) { - free(caa_container_of(node, struct lttng_trigger_ht_element, + free(caa_container_of(node, struct notification_trigger_tokens_ht_element, rcu_node)); } +static +void teardown_tracer_notifier(struct notification_thread_state *state, + const struct lttng_trigger *trigger) +{ + struct cds_lfht_iter iter; + struct notification_trigger_tokens_ht_element *trigger_tokens_ht_element; + + cds_lfht_for_each_entry(state->trigger_tokens_ht, &iter, + trigger_tokens_ht_element, node) { + + if (!lttng_trigger_is_equal(trigger, + trigger_tokens_ht_element->trigger)) { + continue; + } + + event_notifier_error_accounting_unregister_event_notifier( + trigger_tokens_ht_element->trigger); + + /* TODO talk to all app and remove it */ + DBG("[notification-thread] Removed trigger from tokens_ht"); + cds_lfht_del(state->trigger_tokens_ht, + &trigger_tokens_ht_element->node); + + call_rcu(&trigger_tokens_ht_element->rcu_node, + free_notification_trigger_tokens_ht_element_rcu); + + break; + } +} + static int handle_notification_thread_command_unregister_trigger( struct notification_thread_state *state, - struct lttng_trigger *trigger, + const struct lttng_trigger *trigger, enum lttng_error_code *_cmd_reply) { struct cds_lfht_iter iter; struct cds_lfht_node *triggers_ht_node; struct lttng_channel_trigger_list *trigger_list; struct notification_client_list *client_list; - struct notification_client_list_element *client_list_element, *tmp; struct lttng_trigger_ht_element *trigger_ht_element = NULL; - struct lttng_condition *condition = lttng_trigger_get_condition( + const struct lttng_condition *condition = lttng_trigger_get_const_condition( trigger); enum lttng_error_code cmd_reply; @@ -2186,8 +2962,8 @@ int handle_notification_thread_command_unregister_trigger( cds_lfht_lookup(state->triggers_ht, lttng_condition_hash(condition), - match_condition, - condition, + match_trigger, + trigger, &iter); triggers_ht_node = cds_lfht_iter_get_node(&iter); if (!triggers_ht_node) { @@ -2204,13 +2980,7 @@ int handle_notification_thread_command_unregister_trigger( cds_list_for_each_entry_safe(trigger_element, tmp, &trigger_list->list, node) { - const struct lttng_condition *current_condition = - lttng_trigger_get_const_condition( - trigger_element->trigger); - - assert(current_condition); - if (!lttng_condition_is_equal(condition, - current_condition)) { + if (!lttng_trigger_is_equal(trigger, trigger_element->trigger)) { continue; } @@ -2221,25 +2991,33 @@ int handle_notification_thread_command_unregister_trigger( } } - /* - * Remove and release the client list from - * notification_trigger_clients_ht. - */ - client_list = get_client_list_from_condition(state, condition); - assert(client_list); - - cds_list_for_each_entry_safe(client_list_element, tmp, - &client_list->list, node) { - free(client_list_element); + if (lttng_trigger_needs_tracer_notifier(trigger)) { + teardown_tracer_notifier(state, trigger); } - cds_lfht_del(state->notification_trigger_clients_ht, - &client_list->notification_trigger_ht_node); - call_rcu(&client_list->rcu_node, free_notification_client_list_rcu); - /* Remove trigger from triggers_ht. */ trigger_ht_element = caa_container_of(triggers_ht_node, struct lttng_trigger_ht_element, node); - cds_lfht_del(state->triggers_ht, triggers_ht_node); + + if (is_trigger_action_notify(trigger)) { + /* + * Remove and release the client list from + * notification_trigger_clients_ht. + */ + client_list = get_client_list_from_condition(state, condition); + assert(client_list); + + pthread_mutex_lock(&client_list->lock); + cds_list_del(&trigger_ht_element->client_list_trigger_node); + pthread_mutex_unlock(&client_list->lock); + + /* Put new reference and the hashtable's reference. */ + notification_client_list_put(client_list); + notification_client_list_put(client_list); + client_list = NULL; + } + + /* Remove trigger from triggers_ht. */ + notif_thread_state_remove_trigger_ht_elem(state, trigger_ht_element); /* Release the ownership of the trigger. */ lttng_trigger_destroy(trigger_ht_element->trigger); @@ -2271,21 +3049,24 @@ int handle_notification_thread_command( pthread_mutex_lock(&handle->cmd_queue.lock); cmd = cds_list_first_entry(&handle->cmd_queue.list, struct notification_thread_command, cmd_list_node); + cds_list_del(&cmd->cmd_list_node); + pthread_mutex_unlock(&handle->cmd_queue.lock); + + DBG("[notification-thread] Received `%s` command", + notification_command_type_str(cmd->type)); switch (cmd->type) { case NOTIFICATION_COMMAND_TYPE_REGISTER_TRIGGER: - DBG("[notification-thread] Received register trigger command"); - ret = handle_notification_thread_command_register_trigger( - state, cmd->parameters.trigger, + ret = handle_notification_thread_command_register_trigger(state, + cmd->parameters.register_trigger.trigger, &cmd->reply_code); break; case NOTIFICATION_COMMAND_TYPE_UNREGISTER_TRIGGER: - DBG("[notification-thread] Received unregister trigger command"); ret = handle_notification_thread_command_unregister_trigger( - state, cmd->parameters.trigger, + state, + cmd->parameters.unregister_trigger.trigger, &cmd->reply_code); break; case NOTIFICATION_COMMAND_TYPE_ADD_CHANNEL: - DBG("[notification-thread] Received add channel command"); ret = handle_notification_thread_command_add_channel( state, cmd->parameters.add_channel.session.name, @@ -2298,7 +3079,6 @@ int handle_notification_thread_command( &cmd->reply_code); break; case NOTIFICATION_COMMAND_TYPE_REMOVE_CHANNEL: - DBG("[notification-thread] Received remove channel command"); ret = handle_notification_thread_command_remove_channel( state, cmd->parameters.remove_channel.key, cmd->parameters.remove_channel.domain, @@ -2306,9 +3086,6 @@ int handle_notification_thread_command( break; case NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_ONGOING: case NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_COMPLETED: - DBG("[notification-thread] Received session rotation %s command", - cmd->type == NOTIFICATION_COMMAND_TYPE_SESSION_ROTATION_ONGOING ? - "ongoing" : "completed"); ret = handle_notification_thread_command_session_rotation( state, cmd->type, @@ -2319,40 +3096,90 @@ int handle_notification_thread_command( cmd->parameters.session_rotation.location, &cmd->reply_code); break; - case NOTIFICATION_COMMAND_TYPE_QUIT: - DBG("[notification-thread] Received quit command"); - cmd->reply_code = LTTNG_OK; - ret = 1; - goto end; - default: - ERR("[notification-thread] Unknown internal command received"); - goto error_unlock; + case NOTIFICATION_COMMAND_TYPE_ADD_TRACER_EVENT_SOURCE: + ret = handle_notification_thread_command_add_tracer_event_source( + state, + cmd->parameters.tracer_event_source.tracer_event_source_fd, + cmd->parameters.tracer_event_source.domain, + &cmd->reply_code); + break; + case NOTIFICATION_COMMAND_TYPE_REMOVE_TRACER_EVENT_SOURCE: + ret = handle_notification_thread_command_remove_tracer_event_source( + state, + cmd->parameters.tracer_event_source.tracer_event_source_fd, + &cmd->reply_code); + break; + case NOTIFICATION_COMMAND_TYPE_LIST_TRIGGERS: + { + struct lttng_triggers *triggers = NULL; + + ret = handle_notification_thread_command_list_triggers( + handle, + state, + cmd->parameters.list_triggers.uid, + &triggers, + &cmd->reply_code); + cmd->reply.list_triggers.triggers = triggers; + ret = 0; + break; + } + case NOTIFICATION_COMMAND_TYPE_QUIT: + cmd->reply_code = LTTNG_OK; + ret = 1; + goto end; + case NOTIFICATION_COMMAND_TYPE_CLIENT_COMMUNICATION_UPDATE: + { + const enum client_transmission_status client_status = + cmd->parameters.client_communication_update + .status; + const notification_client_id client_id = + cmd->parameters.client_communication_update.id; + struct notification_client *client; + + rcu_read_lock(); + client = get_client_from_id(client_id, state); + + if (!client) { + /* + * Client error was probably already picked-up by the + * notification thread or it has disconnected + * gracefully while this command was queued. + */ + DBG("Failed to find notification client to update communication status, client id = %" PRIu64, + client_id); + ret = 0; + } else { + ret = client_handle_transmission_status( + client, client_status, state); + } + rcu_read_unlock(); + break; + } + default: + ERR("[notification-thread] Unknown internal command received"); + goto error_unlock; } if (ret) { goto error_unlock; } end: - cds_list_del(&cmd->cmd_list_node); - lttng_waiter_wake_up(&cmd->reply_waiter); - pthread_mutex_unlock(&handle->cmd_queue.lock); + if (cmd->is_async) { + free(cmd); + cmd = NULL; + } else { + lttng_waiter_wake_up(&cmd->reply_waiter); + } return ret; error_unlock: /* Wake-up and return a fatal error to the calling thread. */ lttng_waiter_wake_up(&cmd->reply_waiter); - pthread_mutex_unlock(&handle->cmd_queue.lock); cmd->reply_code = LTTNG_ERR_FATAL; error: /* Indicate a fatal error to the caller. */ return -1; } -static -unsigned long hash_client_socket(int socket) -{ - return hash_key_ulong((void *) (unsigned long) socket, lttng_ht_seed); -} - static int socket_set_non_blocking(int socket) { @@ -2381,9 +3208,8 @@ int client_reset_inbound_state(struct notification_client *client) { int ret; - ret = lttng_dynamic_buffer_set_size( - &client->communication.inbound.buffer, 0); - assert(!ret); + + lttng_payload_clear(&client->communication.inbound.payload); client->communication.inbound.bytes_to_receive = sizeof(struct lttng_notification_channel_message); @@ -2392,8 +3218,9 @@ int client_reset_inbound_state(struct notification_client *client) LTTNG_SOCK_SET_UID_CRED(&client->communication.inbound.creds, -1); LTTNG_SOCK_SET_GID_CRED(&client->communication.inbound.creds, -1); ret = lttng_dynamic_buffer_set_size( - &client->communication.inbound.buffer, + &client->communication.inbound.payload.buffer, client->communication.inbound.bytes_to_receive); + return ret; } @@ -2411,10 +3238,14 @@ int handle_notification_thread_client_connect( ret = -1; goto error; } + + pthread_mutex_init(&client->lock, NULL); + client->id = state->next_notification_client_id++; CDS_INIT_LIST_HEAD(&client->condition_list); - lttng_dynamic_buffer_init(&client->communication.inbound.buffer); - lttng_dynamic_buffer_init(&client->communication.outbound.buffer); + lttng_payload_init(&client->communication.inbound.payload); + lttng_payload_init(&client->communication.outbound.payload); client->communication.inbound.expect_creds = true; + ret = client_reset_inbound_state(client); if (ret) { ERR("[notification-thread] Failed to reset client communication's inbound state"); @@ -2459,17 +3290,60 @@ int handle_notification_thread_client_connect( cds_lfht_add(state->client_socket_ht, hash_client_socket(client->socket), &client->client_socket_ht_node); + cds_lfht_add(state->client_id_ht, + hash_client_id(client->id), + &client->client_id_ht_node); rcu_read_unlock(); return ret; + error: notification_client_destroy(client, state); return ret; } -int handle_notification_thread_client_disconnect( - int client_socket, +/* + * RCU read-lock must be held by the caller. + * Client lock must _not_ be held by the caller. + */ +static +int notification_thread_client_disconnect( + struct notification_client *client, struct notification_thread_state *state) +{ + int ret; + struct lttng_condition_list_element *condition_list_element, *tmp; + + /* Acquire the client lock to disable its communication atomically. */ + pthread_mutex_lock(&client->lock); + client->communication.active = false; + cds_lfht_del(state->client_socket_ht, &client->client_socket_ht_node); + cds_lfht_del(state->client_id_ht, &client->client_id_ht_node); + pthread_mutex_unlock(&client->lock); + + ret = lttng_poll_del(&state->events, client->socket); + if (ret) { + ERR("[notification-thread] Failed to remove client socket %d from poll set", + client->socket); + } + + /* Release all conditions to which the client was subscribed. */ + cds_list_for_each_entry_safe(condition_list_element, tmp, + &client->condition_list, node) { + (void) notification_thread_client_unsubscribe(client, + condition_list_element->condition, state, NULL); + } + + /* + * Client no longer accessible to other threads (through the + * client lists). + */ + notification_client_destroy(client, state); + return ret; +} + +int handle_notification_thread_client_disconnect( + int client_socket, struct notification_thread_state *state) { int ret = 0; struct notification_client *client; @@ -2486,13 +3360,7 @@ int handle_notification_thread_client_disconnect( goto end; } - ret = lttng_poll_del(&state->events, client_socket); - if (ret) { - ERR("[notification-thread] Failed to remove client socket from poll set"); - } - cds_lfht_del(state->client_socket_ht, - &client->client_socket_ht_node); - notification_client_destroy(client, state); + ret = notification_thread_client_disconnect(client, state); end: rcu_read_unlock(); return ret; @@ -2508,11 +3376,11 @@ int handle_notification_thread_client_disconnect_all( rcu_read_lock(); DBG("[notification-thread] Closing all client connections"); cds_lfht_for_each_entry(state->client_socket_ht, &iter, client, - client_socket_ht_node) { + client_socket_ht_node) { int ret; - ret = handle_notification_thread_client_disconnect( - client->socket, state); + ret = notification_thread_client_disconnect( + client, state); if (ret) { error_encoutered = true; } @@ -2542,76 +3410,185 @@ int handle_notification_thread_trigger_unregister_all( } static -int client_flush_outgoing_queue(struct notification_client *client, +int client_handle_transmission_status( + struct notification_client *client, + enum client_transmission_status transmission_status, struct notification_thread_state *state) +{ + int ret = 0; + + switch (transmission_status) { + case CLIENT_TRANSMISSION_STATUS_COMPLETE: + ret = lttng_poll_mod(&state->events, client->socket, + CLIENT_POLL_MASK_IN); + if (ret) { + goto end; + } + + break; + case CLIENT_TRANSMISSION_STATUS_QUEUED: + /* + * We want to be notified whenever there is buffer space + * available to send the rest of the payload. + */ + ret = lttng_poll_mod(&state->events, client->socket, + CLIENT_POLL_MASK_IN_OUT); + if (ret) { + goto end; + } + break; + case CLIENT_TRANSMISSION_STATUS_FAIL: + ret = notification_thread_client_disconnect(client, state); + if (ret) { + goto end; + } + break; + case CLIENT_TRANSMISSION_STATUS_ERROR: + ret = -1; + goto end; + default: + abort(); + } +end: + return ret; +} + +/* Client lock must be acquired by caller. */ +static +enum client_transmission_status client_flush_outgoing_queue( + struct notification_client *client) { ssize_t ret; size_t to_send_count; + enum client_transmission_status status; + struct lttng_payload_view pv = lttng_payload_view_from_payload( + &client->communication.outbound.payload, 0, -1); + const int fds_to_send_count = + lttng_payload_view_get_fd_handle_count(&pv); + + ASSERT_LOCKED(client->lock); + + if (!client->communication.active) { + status = CLIENT_TRANSMISSION_STATUS_FAIL; + goto end; + } + + if (pv.buffer.size == 0) { + /* + * If both data and fds are equal to zero, we are in an invalid + * state. + */ + assert(fds_to_send_count != 0); + goto send_fds; + } - assert(client->communication.outbound.buffer.size != 0); - to_send_count = client->communication.outbound.buffer.size; + /* Send data. */ + to_send_count = pv.buffer.size; DBG("[notification-thread] Flushing client (socket fd = %i) outgoing queue", client->socket); ret = lttcomm_send_unix_sock_non_block(client->socket, - client->communication.outbound.buffer.data, + pv.buffer.data, to_send_count); if ((ret >= 0 && ret < to_send_count)) { DBG("[notification-thread] Client (socket fd = %i) outgoing queue could not be completely flushed", client->socket); to_send_count -= max(ret, 0); - memcpy(client->communication.outbound.buffer.data, - client->communication.outbound.buffer.data + - client->communication.outbound.buffer.size - to_send_count, + memmove(client->communication.outbound.payload.buffer.data, + pv.buffer.data + + pv.buffer.size - to_send_count, to_send_count); ret = lttng_dynamic_buffer_set_size( - &client->communication.outbound.buffer, + &client->communication.outbound.payload.buffer, to_send_count); if (ret) { goto error; } - /* - * We want to be notified whenever there is buffer space - * available to send the rest of the payload. - */ - ret = lttng_poll_mod(&state->events, client->socket, - CLIENT_POLL_MASK_IN_OUT); - if (ret) { - goto error; - } + status = CLIENT_TRANSMISSION_STATUS_QUEUED; + goto end; } else if (ret < 0) { - /* Generic error, disconnect the client. */ - ERR("[notification-thread] Failed to send flush outgoing queue, disconnecting client (socket fd = %i)", + /* Generic error, disable the client's communication. */ + ERR("[notification-thread] Failed to flush outgoing queue, disconnecting client (socket fd = %i)", client->socket); - ret = handle_notification_thread_client_disconnect( - client->socket, state); - if (ret) { - goto error; - } + client->communication.active = false; + status = CLIENT_TRANSMISSION_STATUS_FAIL; + goto end; } else { - /* No error and flushed the queue completely. */ + /* + * No error and flushed the queue completely. + * + * The payload buffer size is used later to + * check if there is notifications queued. So albeit that the + * direct caller knows that the transmission is complete, we + * need to set the buffer size to zero. + */ ret = lttng_dynamic_buffer_set_size( - &client->communication.outbound.buffer, 0); - if (ret) { - goto error; - } - ret = lttng_poll_mod(&state->events, client->socket, - CLIENT_POLL_MASK_IN); + &client->communication.outbound.payload.buffer, 0); if (ret) { goto error; } + } +send_fds: + /* No fds to send, transmission is complete. */ + if (fds_to_send_count == 0) { + status = CLIENT_TRANSMISSION_STATUS_COMPLETE; + goto end; + } + + ret = lttcomm_send_payload_view_fds_unix_sock_non_block( + client->socket, &pv); + if (ret < 0) { + /* Generic error, disable the client's communication. */ + ERR("[notification-thread] Failed to flush outgoing fds queue, disconnecting client (socket fd = %i)", + client->socket); + client->communication.active = false; + status = CLIENT_TRANSMISSION_STATUS_FAIL; + goto end; + } else if (ret == 0) { + /* Nothing could be sent. */ + status = CLIENT_TRANSMISSION_STATUS_QUEUED; + } else { + /* Fd passing is an all or nothing kind of thing. */ + status = CLIENT_TRANSMISSION_STATUS_COMPLETE; + /* + * The payload _fd_array count is used later to + * check if there is notifications queued. So although the + * direct caller knows that the transmission is complete, we + * need to clear the _fd_array for the queuing check. + */ + lttng_dynamic_pointer_array_clear( + &client->communication.outbound.payload + ._fd_handles); + } + +end: + if (status == CLIENT_TRANSMISSION_STATUS_COMPLETE) { client->communication.outbound.queued_command_reply = false; client->communication.outbound.dropped_notification = false; + lttng_payload_clear(&client->communication.outbound.payload); } - return 0; + return status; error: - return -1; + return CLIENT_TRANSMISSION_STATUS_ERROR; } +static +bool client_has_outbound_data_left( + const struct notification_client *client) +{ + const struct lttng_payload_view pv = lttng_payload_view_from_payload( + &client->communication.outbound.payload, 0, -1); + const bool has_data = pv.buffer.size != 0; + const bool has_fds = lttng_payload_view_get_fd_handle_count(&pv); + + return has_data || has_fds; +} + +/* Client lock must _not_ be held by the caller. */ static int client_send_command_reply(struct notification_client *client, struct notification_thread_state *state, @@ -2626,210 +3603,266 @@ int client_send_command_reply(struct notification_client *client, .size = sizeof(reply), }; char buffer[sizeof(msg) + sizeof(reply)]; - - if (client->communication.outbound.queued_command_reply) { - /* Protocol error. */ - goto error; - } + enum client_transmission_status transmission_status; memcpy(buffer, &msg, sizeof(msg)); memcpy(buffer + sizeof(msg), &reply, sizeof(reply)); DBG("[notification-thread] Send command reply (%i)", (int) status); + pthread_mutex_lock(&client->lock); + if (client->communication.outbound.queued_command_reply) { + /* Protocol error. */ + goto error_unlock; + } + /* Enqueue buffer to outgoing queue and flush it. */ ret = lttng_dynamic_buffer_append( - &client->communication.outbound.buffer, + &client->communication.outbound.payload.buffer, buffer, sizeof(buffer)); if (ret) { - goto error; + goto error_unlock; } - ret = client_flush_outgoing_queue(client, state); - if (ret) { - goto error; - } + transmission_status = client_flush_outgoing_queue(client); - if (client->communication.outbound.buffer.size != 0) { + if (client_has_outbound_data_left(client)) { /* Queue could not be emptied. */ client->communication.outbound.queued_command_reply = true; } + pthread_mutex_unlock(&client->lock); + ret = client_handle_transmission_status( + client, transmission_status, state); + if (ret) { + goto error; + } + return 0; +error_unlock: + pthread_mutex_unlock(&client->lock); error: return -1; } static -int client_dispatch_message(struct notification_client *client, +int client_handle_message_unknown(struct notification_client *client, struct notification_thread_state *state) { - int ret = 0; + int ret; + /* + * Receiving message header. The function will be called again + * once the rest of the message as been received and can be + * interpreted. + */ + const struct lttng_notification_channel_message *msg; - if (client->communication.inbound.msg_type != - LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_HANDSHAKE && - client->communication.inbound.msg_type != - LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_UNKNOWN && - !client->validated) { - WARN("[notification-thread] client attempted a command before handshake"); + assert(sizeof(*msg) == client->communication.inbound.payload.buffer.size); + msg = (const struct lttng_notification_channel_message *) + client->communication.inbound.payload.buffer.data; + + if (msg->size == 0 || + msg->size > DEFAULT_MAX_NOTIFICATION_CLIENT_MESSAGE_PAYLOAD_SIZE) { + ERR("[notification-thread] Invalid notification channel message: length = %u", + msg->size); ret = -1; goto end; } - switch (client->communication.inbound.msg_type) { - case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_UNKNOWN: - { - /* - * Receiving message header. The function will be called again - * once the rest of the message as been received and can be - * interpreted. - */ - const struct lttng_notification_channel_message *msg; - - assert(sizeof(*msg) == - client->communication.inbound.buffer.size); - msg = (const struct lttng_notification_channel_message *) - client->communication.inbound.buffer.data; + switch (msg->type) { + case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_SUBSCRIBE: + case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_UNSUBSCRIBE: + case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_HANDSHAKE: + break; + default: + ret = -1; + ERR("[notification-thread] Invalid notification channel message: unexpected message type"); + goto end; + } - if (msg->size == 0 || msg->size > DEFAULT_MAX_NOTIFICATION_CLIENT_MESSAGE_PAYLOAD_SIZE) { - ERR("[notification-thread] Invalid notification channel message: length = %u", msg->size); - ret = -1; - goto end; - } + client->communication.inbound.bytes_to_receive = msg->size; + client->communication.inbound.fds_to_receive = msg->fds; + client->communication.inbound.msg_type = + (enum lttng_notification_channel_message_type) msg->type; + ret = lttng_dynamic_buffer_set_size( + &client->communication.inbound.payload.buffer, msg->size); - switch (msg->type) { - case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_SUBSCRIBE: - case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_UNSUBSCRIBE: - case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_HANDSHAKE: - break; - default: - ret = -1; - ERR("[notification-thread] Invalid notification channel message: unexpected message type"); - goto end; - } + /* msg is not valid anymore due to lttng_dynamic_buffer_set_size. */ + msg = NULL; +end: + return ret; +} - client->communication.inbound.bytes_to_receive = msg->size; - client->communication.inbound.msg_type = - (enum lttng_notification_channel_message_type) msg->type; - ret = lttng_dynamic_buffer_set_size( - &client->communication.inbound.buffer, msg->size); - if (ret) { - goto end; - } - break; - } - case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_HANDSHAKE: - { - struct lttng_notification_channel_command_handshake *handshake_client; - struct lttng_notification_channel_command_handshake handshake_reply = { +static +int client_handle_message_handshake(struct notification_client *client, + struct notification_thread_state *state) +{ + int ret; + struct lttng_notification_channel_command_handshake *handshake_client; + const struct lttng_notification_channel_command_handshake handshake_reply = { .major = LTTNG_NOTIFICATION_CHANNEL_VERSION_MAJOR, .minor = LTTNG_NOTIFICATION_CHANNEL_VERSION_MINOR, - }; - struct lttng_notification_channel_message msg_header = { + }; + const struct lttng_notification_channel_message msg_header = { .type = LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_HANDSHAKE, .size = sizeof(handshake_reply), - }; - enum lttng_notification_channel_status status = - LTTNG_NOTIFICATION_CHANNEL_STATUS_OK; - char send_buffer[sizeof(msg_header) + sizeof(handshake_reply)]; - - memcpy(send_buffer, &msg_header, sizeof(msg_header)); - memcpy(send_buffer + sizeof(msg_header), &handshake_reply, - sizeof(handshake_reply)); - - handshake_client = - (struct lttng_notification_channel_command_handshake *) - client->communication.inbound.buffer.data; - client->major = handshake_client->major; - client->minor = handshake_client->minor; - if (!client->communication.inbound.creds_received) { - ERR("[notification-thread] No credentials received from client"); - ret = -1; - goto end; - } + }; + enum lttng_notification_channel_status status = + LTTNG_NOTIFICATION_CHANNEL_STATUS_OK; + char send_buffer[sizeof(msg_header) + sizeof(handshake_reply)]; + + memcpy(send_buffer, &msg_header, sizeof(msg_header)); + memcpy(send_buffer + sizeof(msg_header), &handshake_reply, + sizeof(handshake_reply)); + + handshake_client = + (struct lttng_notification_channel_command_handshake *) + client->communication.inbound.payload.buffer + .data; + client->major = handshake_client->major; + client->minor = handshake_client->minor; + if (!client->communication.inbound.creds_received) { + ERR("[notification-thread] No credentials received from client"); + ret = -1; + goto end; + } - client->uid = LTTNG_SOCK_GET_UID_CRED( - &client->communication.inbound.creds); - client->gid = LTTNG_SOCK_GET_GID_CRED( - &client->communication.inbound.creds); - DBG("[notification-thread] Received handshake from client (uid = %u, gid = %u) with version %i.%i", - client->uid, client->gid, (int) client->major, - (int) client->minor); + client->uid = LTTNG_SOCK_GET_UID_CRED( + &client->communication.inbound.creds); + client->gid = LTTNG_SOCK_GET_GID_CRED( + &client->communication.inbound.creds); + DBG("[notification-thread] Received handshake from client (uid = %u, gid = %u) with version %i.%i", + client->uid, client->gid, (int) client->major, + (int) client->minor); - if (handshake_client->major != LTTNG_NOTIFICATION_CHANNEL_VERSION_MAJOR) { - status = LTTNG_NOTIFICATION_CHANNEL_STATUS_UNSUPPORTED_VERSION; - } + if (handshake_client->major != + LTTNG_NOTIFICATION_CHANNEL_VERSION_MAJOR) { + status = LTTNG_NOTIFICATION_CHANNEL_STATUS_UNSUPPORTED_VERSION; + } - ret = lttng_dynamic_buffer_append(&client->communication.outbound.buffer, - send_buffer, sizeof(send_buffer)); - if (ret) { - ERR("[notification-thread] Failed to send protocol version to notification channel client"); - goto end; - } + pthread_mutex_lock(&client->lock); + /* Outgoing queue will be flushed when the command reply is sent. */ + ret = lttng_dynamic_buffer_append( + &client->communication.outbound.payload.buffer, send_buffer, + sizeof(send_buffer)); + if (ret) { + ERR("[notification-thread] Failed to send protocol version to notification channel client"); + goto end_unlock; + } - ret = client_flush_outgoing_queue(client, state); - if (ret) { - goto end; - } + client->validated = true; + client->communication.active = true; + pthread_mutex_unlock(&client->lock); - ret = client_send_command_reply(client, state, status); - if (ret) { - ERR("[notification-thread] Failed to send reply to notification channel client"); - goto end; - } + /* Set reception state to receive the next message header. */ + ret = client_reset_inbound_state(client); + if (ret) { + ERR("[notification-thread] Failed to reset client communication's inbound state"); + goto end; + } - /* Set reception state to receive the next message header. */ - ret = client_reset_inbound_state(client); - if (ret) { - ERR("[notification-thread] Failed to reset client communication's inbound state"); - goto end; - } - client->validated = true; - break; + /* Flushes the outgoing queue. */ + ret = client_send_command_reply(client, state, status); + if (ret) { + ERR("[notification-thread] Failed to send reply to notification channel client"); + goto end; } - case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_SUBSCRIBE: - case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_UNSUBSCRIBE: - { - struct lttng_condition *condition; - enum lttng_notification_channel_status status = - LTTNG_NOTIFICATION_CHANNEL_STATUS_OK; - struct lttng_payload_view condition_view = - lttng_payload_view_from_dynamic_buffer( - &client->communication.inbound.buffer, + + goto end; +end_unlock: + pthread_mutex_unlock(&client->lock); +end: + return ret; +} + +static +int client_handle_message_subscription( + struct notification_client *client, + enum lttng_notification_channel_message_type msg_type, + struct notification_thread_state *state) +{ + int ret; + struct lttng_condition *condition; + enum lttng_notification_channel_status status = + LTTNG_NOTIFICATION_CHANNEL_STATUS_OK; + struct lttng_payload_view condition_view = + lttng_payload_view_from_payload( + &client->communication.inbound.payload, 0, -1); - size_t expected_condition_size = - client->communication.inbound.buffer.size; + size_t expected_condition_size; - ret = lttng_condition_create_from_payload(&condition_view, - &condition); - if (ret != expected_condition_size) { - ERR("[notification-thread] Malformed condition received from client"); - goto end; - } + /* + * No need to lock client to sample the inbound state as the only + * other thread accessing clients (action executor) only uses the + * outbound state. + */ + expected_condition_size = client->communication.inbound.payload.buffer.size; + ret = lttng_condition_create_from_payload(&condition_view, &condition); + if (ret != expected_condition_size) { + ERR("[notification-thread] Malformed condition received from client"); + goto end; + } - if (client->communication.inbound.msg_type == - LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_SUBSCRIBE) { - ret = notification_thread_client_subscribe(client, - condition, state, &status); - } else { - ret = notification_thread_client_unsubscribe(client, - condition, state, &status); - } - if (ret) { - goto end; - } + if (msg_type == LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_SUBSCRIBE) { + ret = notification_thread_client_subscribe( + client, condition, state, &status); + } else { + ret = notification_thread_client_unsubscribe( + client, condition, state, &status); + } - ret = client_send_command_reply(client, state, status); - if (ret) { - ERR("[notification-thread] Failed to send reply to notification channel client"); - goto end; - } + if (ret) { + goto end; + } - /* Set reception state to receive the next message header. */ - ret = client_reset_inbound_state(client); - if (ret) { - ERR("[notification-thread] Failed to reset client communication's inbound state"); - goto end; - } + /* Set reception state to receive the next message header. */ + ret = client_reset_inbound_state(client); + if (ret) { + ERR("[notification-thread] Failed to reset client communication's inbound state"); + goto end; + } + + ret = client_send_command_reply(client, state, status); + if (ret) { + ERR("[notification-thread] Failed to send reply to notification channel client"); + goto end; + } + +end: + return ret; +} + +static +int client_dispatch_message(struct notification_client *client, + struct notification_thread_state *state) +{ + int ret = 0; + + if (client->communication.inbound.msg_type != + LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_HANDSHAKE && + client->communication.inbound.msg_type != + LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_UNKNOWN && + !client->validated) { + WARN("[notification-thread] client attempted a command before handshake"); + ret = -1; + goto end; + } + + switch (client->communication.inbound.msg_type) { + case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_UNKNOWN: + { + ret = client_handle_message_unknown(client, state); + break; + } + case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_HANDSHAKE: + { + ret = client_handle_message_handshake(client, state); + break; + } + case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_SUBSCRIBE: + case LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_UNSUBSCRIBE: + { + ret = client_handle_message_subscription(client, + client->communication.inbound.msg_type, state); break; } default: @@ -2848,6 +3881,7 @@ int handle_notification_thread_client_in( ssize_t recv_ret; size_t offset; + rcu_read_lock(); client = get_client_from_socket(socket, state); if (!client) { /* Internal error, abort. */ @@ -2855,11 +3889,11 @@ int handle_notification_thread_client_in( goto end; } - offset = client->communication.inbound.buffer.size - + offset = client->communication.inbound.payload.buffer.size - client->communication.inbound.bytes_to_receive; if (client->communication.inbound.expect_creds) { recv_ret = lttcomm_recv_creds_unix_sock(socket, - client->communication.inbound.buffer.data + offset, + client->communication.inbound.payload.buffer.data + offset, client->communication.inbound.bytes_to_receive, &client->communication.inbound.creds); if (recv_ret > 0) { @@ -2868,31 +3902,69 @@ int handle_notification_thread_client_in( } } else { recv_ret = lttcomm_recv_unix_sock_non_block(socket, - client->communication.inbound.buffer.data + offset, + client->communication.inbound.payload.buffer.data + offset, client->communication.inbound.bytes_to_receive); } - if (recv_ret < 0) { + if (recv_ret >= 0) { + client->communication.inbound.bytes_to_receive -= recv_ret; + } else { goto error_disconnect_client; } - client->communication.inbound.bytes_to_receive -= recv_ret; - if (client->communication.inbound.bytes_to_receive == 0) { - ret = client_dispatch_message(client, state); - if (ret) { + if (client->communication.inbound.bytes_to_receive != 0) { + /* Message incomplete wait for more data. */ + ret = 0; + goto end; + } + + assert(client->communication.inbound.bytes_to_receive == 0); + + /* Receive fds. */ + if (client->communication.inbound.fds_to_receive != 0) { + ret = lttcomm_recv_payload_fds_unix_sock_non_block( + client->socket, + client->communication.inbound.fds_to_receive, + &client->communication.inbound.payload); + if (ret > 0) { /* - * Only returns an error if this client must be - * disconnected. + * Fds received. non blocking fds passing is all + * or nothing. */ + ssize_t expected_size; + + expected_size = sizeof(int) * + client->communication.inbound + .fds_to_receive; + assert(ret == expected_size); + client->communication.inbound.fds_to_receive = 0; + } else if (ret == 0) { + /* Received nothing. */ + ret = 0; + goto end; + } else { goto error_disconnect_client; } - } else { - goto end; } + + /* At this point the message is complete.*/ + assert(client->communication.inbound.bytes_to_receive == 0 && + client->communication.inbound.fds_to_receive == 0); + ret = client_dispatch_message(client, state); + if (ret) { + /* + * Only returns an error if this client must be + * disconnected. + */ + goto error_disconnect_client; + } + end: + rcu_read_unlock(); return ret; + error_disconnect_client: - ret = handle_notification_thread_client_disconnect(socket, state); - return ret; + ret = notification_thread_client_disconnect(client, state); + goto end; } /* Client ready to receive outgoing data. */ @@ -2901,7 +3973,9 @@ int handle_notification_thread_client_out( { int ret; struct notification_client *client; + enum client_transmission_status transmission_status; + rcu_read_lock(); client = get_client_from_socket(socket, state); if (!client) { /* Internal error, abort. */ @@ -2909,11 +3983,17 @@ int handle_notification_thread_client_out( goto end; } - ret = client_flush_outgoing_queue(client, state); + pthread_mutex_lock(&client->lock); + transmission_status = client_flush_outgoing_queue(client); + pthread_mutex_unlock(&client->lock); + + ret = client_handle_transmission_status( + client, transmission_status, state); if (ret) { goto end; } end: + rcu_read_unlock(); return ret; } @@ -3080,20 +4160,65 @@ end: } static -int client_enqueue_dropped_notification(struct notification_client *client) +int client_notification_overflow(struct notification_client *client) { - int ret; - struct lttng_notification_channel_message msg = { + int ret = 0; + const struct lttng_notification_channel_message msg = { .type = (int8_t) LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_NOTIFICATION_DROPPED, - .size = 0, }; + ASSERT_LOCKED(client->lock); + + DBG("Dropping notification addressed to client (socket fd = %i)", + client->socket); + if (client->communication.outbound.dropped_notification) { + /* + * The client already has a "notification dropped" message + * in its outgoing queue. Nothing to do since all + * of those messages are coalesced. + */ + goto end; + } + + client->communication.outbound.dropped_notification = true; ret = lttng_dynamic_buffer_append( - &client->communication.outbound.buffer, &msg, + &client->communication.outbound.payload.buffer, &msg, sizeof(msg)); + if (ret) { + PERROR("Failed to enqueue \"dropped notification\" message in client's (socket fd = %i) outgoing queue", + client->socket); + } +end: return ret; } +static int client_handle_transmission_status_wrapper( + struct notification_client *client, + enum client_transmission_status status, + void *user_data) +{ + return client_handle_transmission_status(client, status, + (struct notification_thread_state *) user_data); +} + +static +int send_evaluation_to_clients(const struct lttng_trigger *trigger, + const struct lttng_evaluation *evaluation, + struct notification_client_list* client_list, + struct notification_thread_state *state, + uid_t object_uid, gid_t object_gid) +{ + const struct lttng_credentials creds = { + .uid = LTTNG_OPTIONAL_INIT_VALUE(object_uid), + .gid = LTTNG_OPTIONAL_INIT_VALUE(object_gid), + }; + + return notification_client_list_send_evaluation(client_list, + trigger, evaluation, + &creds, + client_handle_transmission_status_wrapper, state); +} + /* * Permission checks relative to notification channel clients are performed * here. Notice how object, client, and trigger credentials are involved in @@ -3127,24 +4252,27 @@ int client_enqueue_dropped_notification(struct notification_client *client) * interference from external users (those could, for instance, unregister * their triggers). */ -static -int send_evaluation_to_clients(const struct lttng_trigger *trigger, +LTTNG_HIDDEN +int notification_client_list_send_evaluation( + struct notification_client_list *client_list, + const struct lttng_trigger *trigger, const struct lttng_evaluation *evaluation, - struct notification_client_list* client_list, - struct notification_thread_state *state, - uid_t object_uid, gid_t object_gid) + const struct lttng_credentials *source_object_creds, + report_client_transmission_result_cb client_report, + void *user_data) { int ret = 0; struct lttng_payload msg_payload; struct notification_client_list_element *client_list_element, *tmp; const struct lttng_notification notification = { - .condition = (struct lttng_condition *) lttng_trigger_get_const_condition(trigger), + .trigger = (struct lttng_trigger *) trigger, .evaluation = (struct lttng_evaluation *) evaluation, }; struct lttng_notification_channel_message msg_header = { .type = (int8_t) LTTNG_NOTIFICATION_CHANNEL_MESSAGE_TYPE_NOTIFICATION, }; - const struct lttng_credentials *trigger_creds = lttng_trigger_get_credentials(trigger); + const struct lttng_credentials *trigger_creds = + lttng_trigger_get_credentials(trigger); lttng_payload_init(&msg_payload); @@ -3162,29 +4290,60 @@ int send_evaluation_to_clients(const struct lttng_trigger *trigger, } /* Update payload size. */ - ((struct lttng_notification_channel_message * ) msg_payload.buffer.data)->size = - (uint32_t) (msg_payload.buffer.size - sizeof(msg_header)); + ((struct lttng_notification_channel_message *) msg_payload.buffer.data) + ->size = (uint32_t)( + msg_payload.buffer.size - sizeof(msg_header)); + /* Update the payload number of fds. */ + { + const struct lttng_payload_view pv = lttng_payload_view_from_payload( + &msg_payload, 0, -1); + + ((struct lttng_notification_channel_message *) + msg_payload.buffer.data)->fds = (uint32_t) + lttng_payload_view_get_fd_handle_count(&pv); + } + + pthread_mutex_lock(&client_list->lock); cds_list_for_each_entry_safe(client_list_element, tmp, - &client_list->list, node) { + &client_list->clients_list, node) { + enum client_transmission_status transmission_status; struct notification_client *client = client_list_element->client; - if (client->uid != object_uid && client->gid != object_gid && - client->uid != 0) { - /* Client is not allowed to monitor this channel. */ - DBG("[notification-thread] Skipping client at it does not have the object permission to receive notification for this trigger"); - continue; + ret = 0; + pthread_mutex_lock(&client->lock); + if (!client->communication.active) { + /* + * Skip inactive client (protocol error or + * disconnecting). + */ + DBG("Skipping client at it is marked as inactive"); + goto skip_client; } - if (client->uid != trigger_creds->uid && client->gid != trigger_creds->gid) { + if (source_object_creds) { + if (client->uid != lttng_credentials_get_uid(source_object_creds) && + client->gid != lttng_credentials_get_gid(source_object_creds) && + client->uid != 0) { + /* + * Client is not allowed to monitor this + * object. + */ + DBG("[notification-thread] Skipping client at it does not have the object permission to receive notification for this trigger"); + goto skip_client; + } + } + + if (client->uid != lttng_credentials_get_uid(trigger_creds)) { DBG("[notification-thread] Skipping client at it does not have the permission to receive notification for this trigger"); - continue; + goto skip_client; } DBG("[notification-thread] Sending notification to client (fd = %i, %zu bytes)", client->socket, msg_payload.buffer.size); - if (client->communication.outbound.buffer.size) { + + if (client_has_outbound_data_left(client)) { /* * Outgoing data is already buffered for this client; * drop the notification and enqueue a "dropped @@ -3192,37 +4351,324 @@ int send_evaluation_to_clients(const struct lttng_trigger *trigger, * notification since the socket spilled-over to the * queue. */ - DBG("[notification-thread] Dropping notification addressed to client (socket fd = %i)", - client->socket); - if (!client->communication.outbound.dropped_notification) { - client->communication.outbound.dropped_notification = true; - ret = client_enqueue_dropped_notification( - client); - if (ret) { - goto end; - } + ret = client_notification_overflow(client); + if (ret) { + /* Fatal error. */ + goto skip_client; } - continue; } - ret = lttng_dynamic_buffer_append_buffer( - &client->communication.outbound.buffer, - &msg_payload.buffer); + ret = lttng_payload_copy(&msg_payload, &client->communication.outbound.payload); if (ret) { - goto end; + /* Fatal error. */ + goto skip_client; } - ret = client_flush_outgoing_queue(client, state); + transmission_status = client_flush_outgoing_queue(client); + pthread_mutex_unlock(&client->lock); + ret = client_report(client, transmission_status, user_data); if (ret) { - goto end; + /* Fatal error. */ + goto end_unlock_list; + } + + continue; + +skip_client: + pthread_mutex_unlock(&client->lock); + if (ret) { + /* Fatal error. */ + goto end_unlock_list; } } ret = 0; + +end_unlock_list: + pthread_mutex_unlock(&client_list->lock); end: lttng_payload_reset(&msg_payload); return ret; } +static +struct lttng_event_notifier_notification *recv_one_event_notifier_notification( + int notification_pipe_read_fd, enum lttng_domain_type domain) +{ + int ret; + uint64_t token; + struct lttng_event_notifier_notification *notification = NULL; + char *capture_buffer = NULL; + size_t capture_buffer_size; + void *reception_buffer; + size_t reception_size; + + struct lttng_ust_abi_event_notifier_notification ust_notification; + struct lttng_kernel_event_notifier_notification kernel_notification; + + /* Init lttng_event_notifier_notification */ + switch(domain) { + case LTTNG_DOMAIN_UST: + reception_buffer = (void *) &ust_notification; + reception_size = sizeof(ust_notification); + break; + case LTTNG_DOMAIN_KERNEL: + reception_buffer = (void *) &kernel_notification; + reception_size = sizeof(kernel_notification); + break; + default: + abort(); + } + + /* + * The monitoring pipe only holds messages smaller than PIPE_BUF, + * ensuring that read/write of tracer notifications are atomic. + */ + ret = lttng_read(notification_pipe_read_fd, reception_buffer, + reception_size); + if (ret != reception_size) { + PERROR("Failed to read from event source notification pipe: fd = %d, size to read = %zu, ret = %d", + notification_pipe_read_fd, reception_size, ret); + ret = -1; + goto end; + } + + switch(domain) { + case LTTNG_DOMAIN_UST: + token = ust_notification.token; + capture_buffer_size = ust_notification.capture_buf_size; + break; + case LTTNG_DOMAIN_KERNEL: + token = kernel_notification.token; + capture_buffer_size = kernel_notification.capture_buf_size; + break; + default: + abort(); + } + + if (capture_buffer_size == 0) { + capture_buffer = NULL; + goto skip_capture; + } + + if (capture_buffer_size > MAX_CAPTURE_SIZE) { + ERR("[notification-thread] Event notifier has a capture payload size which exceeds the maximum allowed size: capture_payload_size = %zu bytes, max allowed size = %d bytes", + capture_buffer_size, MAX_CAPTURE_SIZE); + goto end; + } + + capture_buffer = zmalloc(capture_buffer_size); + if (!capture_buffer) { + ERR("[notification-thread] Failed to allocate capture buffer"); + goto end; + } + + /* Fetch additional payload (capture). */ + ret = lttng_read(notification_pipe_read_fd, capture_buffer, capture_buffer_size); + if (ret != capture_buffer_size) { + ERR("[notification-thread] Failed to read from event source pipe (fd = %i)", + notification_pipe_read_fd); + goto end; + } + +skip_capture: + notification = lttng_event_notifier_notification_create(token, domain, + capture_buffer, capture_buffer_size); + if (notification == NULL) { + goto end; + } + + /* + * Ownership transfered to the lttng_event_notifier_notification object. + */ + capture_buffer = NULL; + +end: + free(capture_buffer); + return notification; +} + +static +int dispatch_one_event_notifier_notification(struct notification_thread_state *state, + struct lttng_event_notifier_notification *notification) +{ + struct cds_lfht_node *node; + struct cds_lfht_iter iter; + struct notification_trigger_tokens_ht_element *element; + enum lttng_trigger_status trigger_status; + struct lttng_evaluation *evaluation = NULL; + enum action_executor_status executor_status; + struct notification_client_list *client_list = NULL; + const char *trigger_name; + int ret; + unsigned int capture_count = 0; + + /* Find triggers associated with this token. */ + rcu_read_lock(); + cds_lfht_lookup(state->trigger_tokens_ht, + hash_key_u64(¬ification->tracer_token, lttng_ht_seed), + match_trigger_token, ¬ification->tracer_token, &iter); + node = cds_lfht_iter_get_node(&iter); + if (caa_unlikely(!node)) { + /* + * This is not an error, slow consumption of the tracer + * notifications can lead to situations where a trigger is + * removed but we still get tracer notifications matching a + * trigger that no longer exists. + */ + ret = 0; + goto end_unlock; + } + + element = caa_container_of(node, + struct notification_trigger_tokens_ht_element, + node); + + if (!lttng_trigger_should_fire(element->trigger)) { + ret = 0; + goto end_unlock; + } + + lttng_trigger_fire(element->trigger); + + trigger_status = lttng_trigger_get_name(element->trigger, &trigger_name); + assert(trigger_status == LTTNG_TRIGGER_STATUS_OK); + + if (lttng_condition_on_event_get_capture_descriptor_count( + lttng_trigger_get_const_condition(element->trigger), + &capture_count) != LTTNG_CONDITION_STATUS_OK) { + ERR("Failed to get capture count"); + ret = -1; + goto end; + } + + if (!notification->capture_buffer && capture_count != 0) { + ERR("Expected capture but capture buffer is null"); + ret = -1; + goto end; + } + + evaluation = lttng_evaluation_on_event_create( + container_of(lttng_trigger_get_const_condition( + element->trigger), + struct lttng_condition_on_event, + parent), + notification->capture_buffer, + notification->capture_buf_size, false); + + if (evaluation == NULL) { + ERR("[notification-thread] Failed to create event rule hit evaluation while creating and enqueuing action executor job"); + ret = -1; + goto end_unlock; + } + client_list = get_client_list_from_condition(state, + lttng_trigger_get_const_condition(element->trigger)); + executor_status = action_executor_enqueue(state->executor, + element->trigger, evaluation, NULL, client_list); + switch (executor_status) { + case ACTION_EXECUTOR_STATUS_OK: + ret = 0; + break; + case ACTION_EXECUTOR_STATUS_OVERFLOW: + { + struct notification_client_list_element *client_list_element, + *tmp; + + /* + * Not a fatal error; this is expected and simply means the + * executor has too much work queued already. + */ + ret = 0; + + /* No clients subscribed to notifications for this trigger. */ + if (!client_list) { + break; + } + + /* Warn clients that a notification (or more) was dropped. */ + pthread_mutex_lock(&client_list->lock); + cds_list_for_each_entry_safe(client_list_element, tmp, + &client_list->clients_list, node) { + enum client_transmission_status transmission_status; + struct notification_client *client = + client_list_element->client; + + pthread_mutex_lock(&client->lock); + ret = client_notification_overflow(client); + if (ret) { + /* Fatal error. */ + goto next_client; + } + + transmission_status = + client_flush_outgoing_queue(client); + ret = client_handle_transmission_status( + client, transmission_status, state); + if (ret) { + /* Fatal error. */ + goto next_client; + } +next_client: + pthread_mutex_unlock(&client->lock); + if (ret) { + break; + } + } + + pthread_mutex_unlock(&client_list->lock); + break; + } + case ACTION_EXECUTOR_STATUS_INVALID: + case ACTION_EXECUTOR_STATUS_ERROR: + /* Fatal error, shut down everything. */ + ERR("Fatal error encoutered while enqueuing action to the action executor"); + ret = -1; + goto end_unlock; + default: + /* Unhandled error. */ + abort(); + } + +end_unlock: + notification_client_list_put(client_list); + rcu_read_unlock(); +end: + return ret; +} + +static +int handle_one_event_notifier_notification( + struct notification_thread_state *state, + int pipe, enum lttng_domain_type domain) +{ + int ret = 0; + struct lttng_event_notifier_notification *notification = NULL; + + notification = recv_one_event_notifier_notification(pipe, domain); + if (notification == NULL) { + /* Reception failed, don't consider it fatal. */ + ERR("[notification-thread] Error receiving an event notifier notification from tracer: fd = %i, domain = %s", + pipe, lttng_domain_type_str(domain)); + goto end; + } + + ret = dispatch_one_event_notifier_notification(state, notification); + if (ret) { + ERR("[notification-thread] Error dispatching an event notifier notification from tracer: fd = %i, domain = %s", + pipe, lttng_domain_type_str(domain)); + goto end; + } + +end: + lttng_event_notifier_notification_destroy(notification); + return ret; +} + +int handle_notification_thread_event_notification(struct notification_thread_state *state, + int pipe, enum lttng_domain_type domain) +{ + return handle_one_event_notifier_notification(state, pipe, domain); +} + int handle_notification_thread_channel_sample( struct notification_thread_state *state, int pipe, enum lttng_domain_type domain) @@ -3237,6 +4683,7 @@ int handle_notification_thread_channel_sample( bool previous_sample_available = false; struct channel_state_sample previous_sample, latest_sample; uint64_t previous_session_consumed_total, latest_session_consumed_total; + struct lttng_credentials channel_creds; /* * The monitoring pipe only holds messages smaller than PIPE_BUF, @@ -3275,8 +4722,7 @@ int handle_notification_thread_channel_sample( */ DBG("[notification-thread] Received a sample for an unknown channel from consumerd, key = %" PRIu64 " in %s domain", latest_sample.key.key, - domain == LTTNG_DOMAIN_KERNEL ? "kernel" : - "user space"); + lttng_domain_type_str(domain)); goto end_unlock; } channel_info = caa_container_of(node, struct channel_info, @@ -3355,38 +4801,31 @@ int handle_notification_thread_channel_sample( goto end_unlock; } + channel_creds = (typeof(channel_creds)) { + .uid = LTTNG_OPTIONAL_INIT_VALUE(channel_info->session_info->uid), + .gid = LTTNG_OPTIONAL_INIT_VALUE(channel_info->session_info->gid), + }; + trigger_list = caa_container_of(node, struct lttng_channel_trigger_list, channel_triggers_ht_node); cds_list_for_each_entry(trigger_list_element, &trigger_list->list, node) { const struct lttng_condition *condition; - const struct lttng_action *action; - const struct lttng_trigger *trigger; - struct notification_client_list *client_list; + struct lttng_trigger *trigger; + struct notification_client_list *client_list = NULL; struct lttng_evaluation *evaluation = NULL; + enum action_executor_status executor_status; + ret = 0; trigger = trigger_list_element->trigger; condition = lttng_trigger_get_const_condition(trigger); assert(condition); - action = lttng_trigger_get_const_action(trigger); - - /* Notify actions are the only type currently supported. */ - assert(lttng_action_get_type_const(action) == - LTTNG_ACTION_TYPE_NOTIFY); /* * Check if any client is subscribed to the result of this * evaluation. */ client_list = get_client_list_from_condition(state, condition); - assert(client_list); - if (cds_list_empty(&client_list->list)) { - /* - * No clients interested in the evaluation's result, - * skip it. - */ - continue; - } ret = evaluate_buffer_condition(condition, &evaluation, state, previous_sample_available ? &previous_sample : NULL, @@ -3395,21 +4834,57 @@ int handle_notification_thread_channel_sample( latest_session_consumed_total, channel_info); if (caa_unlikely(ret)) { - goto end_unlock; + goto put_list; } if (caa_likely(!evaluation)) { - continue; + goto put_list; + } + + if (!lttng_trigger_should_fire(trigger)) { + goto put_list; } - /* Dispatch evaluation result to all clients. */ - ret = send_evaluation_to_clients(trigger_list_element->trigger, - evaluation, client_list, state, - channel_info->session_info->uid, - channel_info->session_info->gid); - lttng_evaluation_destroy(evaluation); + lttng_trigger_fire(trigger); + + /* + * Ownership of `evaluation` transferred to the action executor + * no matter the result. + */ + executor_status = action_executor_enqueue(state->executor, + trigger, evaluation, &channel_creds, + client_list); + evaluation = NULL; + switch (executor_status) { + case ACTION_EXECUTOR_STATUS_OK: + break; + case ACTION_EXECUTOR_STATUS_ERROR: + case ACTION_EXECUTOR_STATUS_INVALID: + /* + * TODO Add trigger identification (name/id) when + * it is added to the API. + */ + ERR("Fatal error occurred while enqueuing action associated with buffer-condition trigger"); + ret = -1; + goto put_list; + case ACTION_EXECUTOR_STATUS_OVERFLOW: + /* + * TODO Add trigger identification (name/id) when + * it is added to the API. + * + * Not a fatal error. + */ + WARN("No space left when enqueuing action associated with buffer-condition trigger"); + ret = 0; + goto put_list; + default: + abort(); + } + +put_list: + notification_client_list_put(client_list); if (caa_unlikely(ret)) { - goto end_unlock; + break; } } end_unlock: