Fix: liblttng-ust process startup hang when sessiond is stopped

Ensure the listener thread owns socket and notify_socket, so they don't
have to hold the ust_lock() while connecting to the sessiond and reading
from this socket.

Therefore, after process fork, we can safely cleanup those retources,
because the thread has been removed by the operating system. On exit,
however, let the OS teardown those sockets, so exit path does not race
with the listener thread.

Fixes #545

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

diff --git a/liblttng-ust/lttng-ust-comm.c b/liblttng-ust/lttng-ust-comm.c
index 4d437fc58ac14d651825fe71788a1674a4126db1..ae2e518965e94e37a9743e7c022023eb880e206a 100644 (file)
--- a/liblttng-ust/lttng-ust-comm.c
+++ b/liblttng-ust/lttng-ust-comm.c
@@ -625,6 +625,25 @@ void cleanup_sock_info(struct sock_info *sock_info, int exiting)
 {
        int ret;
 
+       if (sock_info->root_handle != -1) {
+               ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
+               if (ret) {
+                       ERR("Error unref root handle");
+               }
+               sock_info->root_handle = -1;
+       }
+       sock_info->constructor_sem_posted = 0;
+
+       /*
+        * wait_shm_mmap, socket and notify socket are used by listener
+        * threads outside of the ust lock, so we cannot tear them down
+        * ourselves, because we cannot join on these threads. Leave
+        * responsibility of cleaning up these resources to the OS
+        * process exit.
+        */
+       if (exiting)
+               return;
+
        if (sock_info->socket != -1) {
                ret = ustcomm_close_unix_sock(sock_info->socket);
                if (ret) {
@@ -639,21 +658,7 @@ void cleanup_sock_info(struct sock_info *sock_info, int exiting)
                }
                sock_info->notify_socket = -1;
        }
-       if (sock_info->root_handle != -1) {
-               ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
-               if (ret) {
-                       ERR("Error unref root handle");
-               }
-               sock_info->root_handle = -1;
-       }
-       sock_info->constructor_sem_posted = 0;
-       /*
-        * wait_shm_mmap is used by listener threads outside of the
-        * ust lock, so we cannot tear it down ourselves, because we
-        * cannot join on these threads. Leave this task to the OS
-        * process exit.
-        */
-       if (!exiting && sock_info->wait_shm_mmap) {
+       if (sock_info->wait_shm_mmap) {
                ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
                if (ret) {
                        ERR("Error unmapping wait shm");
@@ -910,11 +915,6 @@ restart:
                has_waited = 1;
                prev_connect_failed = 0;
        }
-       ust_lock();
-
-       if (lttng_ust_comm_should_quit) {
-               goto quit;
-       }
 
        if (sock_info->socket != -1) {
                ret = ustcomm_close_unix_sock(sock_info->socket);
@@ -939,6 +939,13 @@ restart:
                if (ret < 0) {
                        DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
                        prev_connect_failed = 1;
+
+                       ust_lock();
+
+                       if (lttng_ust_comm_should_quit) {
+                               goto quit;
+                       }
+
                        /*
                         * If we cannot find the sessiond daemon, don't delay
                         * constructor execution.
@@ -976,6 +983,12 @@ restart:
                WARN("Unsuppoorted timeout value %ld", timeout);
        }
 
+       ust_lock();
+
+       if (lttng_ust_comm_should_quit) {
+               goto quit;
+       }
+
        /*
         * Create only one root handle per listener thread for the whole
         * process lifetime, so we ensure we get ID which is statically