Fix: sessiond ust-app session teardown race

[lttng-tools.git] / src / bin / lttng-sessiond / ust-app.c

diff --git a/src/bin/lttng-sessiond/ust-app.c b/src/bin/lttng-sessiond/ust-app.c
index 52193f278b83ca00e4e9714204b1f0f127506486..22f25da35a1c0d5d2bcf2d1c4df18430d124b69d 100644 (file)
--- a/src/bin/lttng-sessiond/ust-app.c
+++ b/src/bin/lttng-sessiond/ust-app.c
@@ -430,6 +430,9 @@ void delete_ust_app_channel(int sock, struct ust_app_channel *ua_chan,
  * Must be called with the registry lock held.
  *
  * On success, return the len of metadata pushed or else a negative value.
+ * Returning a -EPIPE return value means we could not send the metadata,
+ * but it can be caused by recoverable errors (e.g. the application has
+ * terminated concurrently).
  */
 ssize_t ust_app_push_metadata(struct ust_registry_session *registry,
                struct consumer_socket *socket, int send_zero_data)
@@ -453,9 +456,10 @@ ssize_t ust_app_push_metadata(struct ust_registry_session *registry,
        /*
         * On a push metadata error either the consumer is dead or the
         * metadata channel has been destroyed because its endpoint
-        * might have died (e.g: relayd). If so, the metadata closed
-        * flag is set to 1 so we deny pushing metadata again which is
-        * not valid anymore on the consumer side.
+        * might have died (e.g: relayd), or because the application has
+        * exited. If so, the metadata closed flag is set to 1 so we
+        * deny pushing metadata again which is not valid anymore on the
+        * consumer side.
         */
        if (registry->metadata_closed) {
                return -EPIPE;
@@ -546,6 +550,9 @@ error_push:
  * of socket throughout this function.
  *
  * Return 0 on success else a negative error.
+ * Returning a -EPIPE return value means we could not send the metadata,
+ * but it can be caused by recoverable errors (e.g. the application has
+ * terminated concurrently).
  */
 static int push_metadata(struct ust_registry_session *registry,
                struct consumer_output *consumer)
@@ -580,7 +587,6 @@ static int push_metadata(struct ust_registry_session *registry,
        return 0;
 
 error:
-end:
        pthread_mutex_unlock(&registry->lock);
        return ret_val;
 }
@@ -670,6 +676,9 @@ void delete_ust_app_session(int sock, struct ust_app_session *ua_sess,
 
        pthread_mutex_lock(&ua_sess->lock);
 
+       assert(!ua_sess->deleted);
+       ua_sess->deleted = true;
+
        registry = get_session_registry(ua_sess);
        if (registry) {
                /* Push metadata for application before freeing the application. */
@@ -3127,6 +3136,11 @@ void ust_app_unregister(int sock)
                 */
                pthread_mutex_lock(&ua_sess->lock);
 
+               if (ua_sess->deleted) {
+                       pthread_mutex_unlock(&ua_sess->lock);
+                       continue;
+               }
+
                /*
                 * Normally, this is done in the delete session process which is
                 * executed in the call rcu below. However, upon registration we can't
@@ -3692,6 +3706,12 @@ int ust_app_create_channel_glb(struct ltt_ust_session *usess,
                assert(ua_sess);
 
                pthread_mutex_lock(&ua_sess->lock);
+
+               if (ua_sess->deleted) {
+                       pthread_mutex_unlock(&ua_sess->lock);
+                       continue;
+               }
+
                if (!strncmp(uchan->name, DEFAULT_METADATA_NAME,
                                        sizeof(uchan->name))) {
                        copy_channel_attr_to_ustctl(&ua_sess->metadata_attr, &uchan->attr);
@@ -3761,6 +3781,11 @@ int ust_app_enable_event_glb(struct ltt_ust_session *usess,
 
                pthread_mutex_lock(&ua_sess->lock);
 
+               if (ua_sess->deleted) {
+                       pthread_mutex_unlock(&ua_sess->lock);
+                       continue;
+               }
+
                /* Lookup channel in the ust app session */
                lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter);
                ua_chan_node = lttng_ht_iter_get_node_str(&uiter);
@@ -3827,6 +3852,12 @@ int ust_app_create_event_glb(struct ltt_ust_session *usess,
                }
 
                pthread_mutex_lock(&ua_sess->lock);
+
+               if (ua_sess->deleted) {
+                       pthread_mutex_unlock(&ua_sess->lock);
+                       continue;
+               }
+
                /* Lookup channel in the ust app session */
                lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter);
                ua_chan_node = lttng_ht_iter_get_node_str(&uiter);
@@ -3878,6 +3909,11 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app)
 
        pthread_mutex_lock(&ua_sess->lock);
 
+       if (ua_sess->deleted) {
+               pthread_mutex_unlock(&ua_sess->lock);
+               goto end;
+       }
+
        /* Upon restart, we skip the setup, already done */
        if (ua_sess->started) {
                goto skip_setup;
@@ -3978,6 +4014,11 @@ int ust_app_stop_trace(struct ltt_ust_session *usess, struct ust_app *app)
 
        pthread_mutex_lock(&ua_sess->lock);
 
+       if (ua_sess->deleted) {
+               pthread_mutex_unlock(&ua_sess->lock);
+               goto end_no_session;
+       }
+
        /*
         * If started = 0, it means that stop trace has been called for a session
         * that was never started. It's possible since we can have a fail start
@@ -4058,6 +4099,10 @@ int ust_app_flush_app_session(struct ust_app *app,
 
        pthread_mutex_lock(&ua_sess->lock);
 
+       if (ua_sess->deleted) {
+               goto end_deleted;
+       }
+
        health_code_update();
 
        /* Flushing buffers */
@@ -4087,6 +4132,7 @@ int ust_app_flush_app_session(struct ust_app *app,
 
        health_code_update();
 
+end_deleted:
        pthread_mutex_unlock(&ua_sess->lock);
 
 end_not_compatible:
@@ -4167,7 +4213,6 @@ int ust_app_flush_session(struct ltt_ust_session *usess)
                break;
        }
 
-end_no_session:
        rcu_read_unlock();
        health_code_update();
        return ret;
@@ -4341,6 +4386,11 @@ void ust_app_global_update(struct ltt_ust_session *usess, int sock)
 
        pthread_mutex_lock(&ua_sess->lock);
 
+       if (ua_sess->deleted) {
+               pthread_mutex_unlock(&ua_sess->lock);
+               goto error;
+       }
+
        /*
         * We can iterate safely here over all UST app session since the create ust
         * app session above made a shadow copy of the UST global domain from the
@@ -4434,6 +4484,12 @@ int ust_app_add_ctx_channel_glb(struct ltt_ust_session *usess,
                }
 
                pthread_mutex_lock(&ua_sess->lock);
+
+               if (ua_sess->deleted) {
+                       pthread_mutex_unlock(&ua_sess->lock);
+                       continue;
+               }
+
                /* Lookup channel in the ust app session */
                lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &uiter);
                ua_chan_node = lttng_ht_iter_get_node_str(&uiter);
@@ -4492,6 +4548,12 @@ int ust_app_enable_event_pid(struct ltt_ust_session *usess,
        }
 
        pthread_mutex_lock(&ua_sess->lock);
+
+       if (ua_sess->deleted) {
+               ret = 0;
+               goto end_unlock;
+       }
+
        /* Lookup channel in the ust app session */
        lttng_ht_lookup(ua_sess->channels, (void *)uchan->name, &iter);
        ua_chan_node = lttng_ht_iter_get_node_str(&iter);