Fix: sessiond: session destroy hang in per-uid when context cannot be added

[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 befbfab32dc2c692b56ef3005e4e46ad0ef9bf2a..35e4e3bceed8913f9b17b6b745ebd149058c84e5 100644 (file)
--- a/src/bin/lttng-sessiond/ust-app.c
+++ b/src/bin/lttng-sessiond/ust-app.c
@@ -2488,7 +2488,7 @@ static int do_consumer_create_channel(struct ltt_ust_session *usess,
        health_code_update();
 
        /*
-        * Now get the channel from the consumer. This call wil populate the stream
+        * Now get the channel from the consumer. This call will populate the stream
         * list of that channel and set the ust objects.
         */
        if (usess->consumer->enabled) {
@@ -4356,15 +4356,6 @@ int ust_app_start_trace(struct ltt_ust_session *usess, struct ust_app *app)
                goto skip_setup;
        }
 
-       /*
-        * Create the metadata for the application. This returns gracefully if a
-        * metadata was already set for the session.
-        */
-       ret = create_ust_app_metadata(ua_sess, app, usess->consumer);
-       if (ret < 0) {
-               goto error_unlock;
-       }
-
        health_code_update();
 
 skip_setup:
@@ -5032,6 +5023,7 @@ void ust_app_synchronize(struct ltt_ust_session *usess,
        }
 
        rcu_read_lock();
+
        cds_lfht_for_each_entry(usess->domain_global.channels->ht, &uchan_iter,
                        uchan, node.node) {
                struct ust_app_channel *ua_chan;
@@ -5075,6 +5067,21 @@ void ust_app_synchronize(struct ltt_ust_session *usess,
                        }
                }
        }
+
+       /*
+        * Create the metadata for the application. This returns gracefully if a
+        * metadata was already set for the session.
+        *
+        * The metadata channel must be created after the data channels as the
+        * consumer daemon assumes this ordering. When interacting with a relay
+        * daemon, the consumer will use this assumption to send the
+        * "STREAMS_SENT" message to the relay daemon.
+        */
+       ret = create_ust_app_metadata(ua_sess, app, usess->consumer);
+       if (ret < 0) {
+               goto error_unlock;
+       }
+
        rcu_read_unlock();
 
 end:
@@ -6270,11 +6277,6 @@ enum lttng_error_code ust_app_rotate_session(struct ltt_session *session)
                        struct buffer_reg_channel *reg_chan;
                        struct consumer_socket *socket;
 
-                       if (!reg->registry->reg.ust->metadata_key) {
-                               /* Skip since no metadata is present */
-                               continue;
-                       }
-
                        /* Get consumer socket to use to push the metadata.*/
                        socket = consumer_find_socket_by_bitness(reg->bits_per_long,
                                        usess->consumer);
@@ -6297,6 +6299,19 @@ enum lttng_error_code ust_app_rotate_session(struct ltt_session *session)
                                }
                        }
 
+                       /*
+                        * The metadata channel might not be present.
+                        *
+                        * Consumer stream allocation can be done
+                        * asynchronously and can fail on intermediary
+                        * operations (i.e add context) and lead to data
+                        * channels created with no metadata channel.
+                        */
+                       if (!reg->registry->reg.ust->metadata_key) {
+                               /* Skip since no metadata is present. */
+                               continue;
+                       }
+
                        (void) push_metadata(reg->registry->reg.ust, usess->consumer);
 
                        ret = consumer_rotate_channel(socket,
@@ -6747,7 +6762,7 @@ enum lttng_error_code ust_app_open_packets(struct ltt_session *session)
                                         * Per-PID buffer and application going
                                         * away.
                                         */
-                                       if (ret == -LTTNG_ERR_CHAN_NOT_FOUND) {
+                                       if (open_ret == -LTTNG_ERR_CHAN_NOT_FOUND) {
                                                continue;
                                        }