Fix: set app socket timeout just after accept()

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

diff --git a/src/bin/lttng-sessiond/main.c b/src/bin/lttng-sessiond/main.c
index 6b1035c34c13ab11a8eec2d0fdc29cf6c0021f00..a70ecda4cd04494e62f266698257a28910944ac7 100644 (file)
--- a/src/bin/lttng-sessiond/main.c
+++ b/src/bin/lttng-sessiond/main.c
@@ -1292,12 +1292,6 @@ static void *thread_manage_apps(void *data)
                                                goto error;
                                        }
 
-                                       /* Set socket timeout for both receiving and ending */
-                                       (void) lttcomm_setsockopt_rcv_timeout(sock,
-                                                       app_socket_timeout);
-                                       (void) lttcomm_setsockopt_snd_timeout(sock,
-                                                       app_socket_timeout);
-
                                        DBG("Apps with sock %d added to poll set", sock);
 
                                        health_code_update();
@@ -1355,6 +1349,9 @@ error_testpoint:
  * Send a socket to a thread This is called from the dispatch UST registration
  * thread once all sockets are set for the application.
  *
+ * The sock value can be invalid, we don't really care, the thread will handle
+ * it and make the necessary cleanup if so.
+ *
  * On success, return 0 else a negative value being the errno message of the
  * write().
  */
@@ -1362,9 +1359,14 @@ static int send_socket_to_thread(int fd, int sock)
 {
        int ret;
 
-       /* Sockets MUST be set or else this should not have been called. */
-       assert(fd >= 0);
-       assert(sock >= 0);
+       /*
+        * It's possible that the FD is set as invalid with -1 concurrently just
+        * before calling this function being a shutdown state of the thread.
+        */
+       if (fd < 0) {
+               ret = -EBADF;
+               goto error;
+       }
 
        do {
                ret = write(fd, &sock, sizeof(sock));
@@ -1627,7 +1629,12 @@ static void *thread_dispatch_ust_registration(void *data)
                                if (ret < 0) {
                                        rcu_read_unlock();
                                        session_unlock_list();
-                                       /* No notify thread, stop the UST tracing. */
+                                       /*
+                                        * No notify thread, stop the UST tracing. However, this is
+                                        * not an internal error of the this thread thus setting
+                                        * the health error code to a normal exit.
+                                        */
+                                       err = 0;
                                        goto error;
                                }
 
@@ -1652,7 +1659,12 @@ static void *thread_dispatch_ust_registration(void *data)
                                if (ret < 0) {
                                        rcu_read_unlock();
                                        session_unlock_list();
-                                       /* No apps. thread, stop the UST tracing. */
+                                       /*
+                                        * No apps. thread, stop the UST tracing. However, this is
+                                        * not an internal error of the this thread thus setting
+                                        * the health error code to a normal exit.
+                                        */
+                                       err = 0;
                                        goto error;
                                }
 
@@ -1782,6 +1794,18 @@ static void *thread_registration_apps(void *data)
                                                goto error;
                                        }
 
+                                       /*
+                                        * Set socket timeout for both receiving and ending.
+                                        * app_socket_timeout is in seconds, whereas
+                                        * lttcomm_setsockopt_rcv_timeout and
+                                        * lttcomm_setsockopt_snd_timeout expect msec as
+                                        * parameter.
+                                        */
+                                       (void) lttcomm_setsockopt_rcv_timeout(sock,
+                                                       app_socket_timeout * 1000);
+                                       (void) lttcomm_setsockopt_snd_timeout(sock,
+                                                       app_socket_timeout * 1000);
+
                                        /*
                                         * Set the CLOEXEC flag. Return code is useless because
                                         * either way, the show must go on.