From: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Date: Mon, 8 Aug 2011 05:22:00 +0000 (-0400)
Subject: Fix handling of sessiond respawn after a SIGKILL
X-Git-Tag: v2.0-pre8~3
X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=commitdiff_plain;h=7d8234d9e6f162ee642cdbec911f46c29b012c3d

Fix handling of sessiond respawn after a SIGKILL

Checking if the sessiond responds to connect() lets lttng / ltt-sessiond
check correctly if the socket files are stale files or actively used.

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

diff --git a/liblttngctl/liblttngctl.c b/liblttngctl/liblttngctl.c
index 4faf8414f..2464bce3b 100644
--- a/liblttngctl/liblttngctl.c
+++ b/liblttngctl/liblttngctl.c
@@ -608,13 +608,22 @@ int lttng_session_daemon_alive(void)
 		return ret;
 	}
 
-	/* If socket exist, we consider the daemon started */
+	/* If socket exist, we check if the daemon listens to connect. */
 	ret = access(sessiond_sock_path, F_OK);
 	if (ret < 0) {
 		/* Not alive */
 		return 0;
 	}
 
+	ret = lttcomm_connect_unix_sock(sessiond_sock_path);
+	if (ret < 0) {
+		/* Not alive */
+		return 0;
+	}
+	ret = lttcomm_close_unix_sock(ret);
+	if (ret < 0)
+		perror("lttcomm_close_unix_sock");
+
 	/* Is alive */
 	return 1;
 }
diff --git a/liblttsessiondcomm/liblttsessiondcomm.c b/liblttsessiondcomm/liblttsessiondcomm.c
index 9313a34c5..e7e420433 100644
--- a/liblttsessiondcomm/liblttsessiondcomm.c
+++ b/liblttsessiondcomm/liblttsessiondcomm.c
@@ -186,6 +186,8 @@ int lttcomm_create_unix_sock(const char *pathname)
 	sun.sun_family = AF_UNIX;
 	strncpy(sun.sun_path, pathname, strlen(pathname));
 
+	/* Unlink the old file if present */
+	(void) unlink(pathname);
 	ret = bind(fd, (struct sockaddr *) &sun, sizeof(sun));
 	if (ret < 0) {
 		perror("bind");
diff --git a/ltt-sessiond/Makefile.am b/ltt-sessiond/Makefile.am
index 048084095..268617350 100644
--- a/ltt-sessiond/Makefile.am
+++ b/ltt-sessiond/Makefile.am
@@ -11,7 +11,9 @@ ltt_sessiond_SOURCES = utils.c trace.c session.c traceable-app.c ust-ctl.c \
                        utils.h trace.h session.h traceable-app.h ust-ctl.h \
                        context.h kernel-ctl.h ltt-sessiond.h
 
+# link on liblttngctl for check if sessiond is already alive.
 ltt_sessiond_LDADD = \
 		 $(top_builddir)/liblttsessiondcomm/liblttsessiondcomm.la \
 		 $(top_builddir)/libkernelctl/libkernelctl.la \
-		 $(top_builddir)/libustctl/libustctl.la
+		 $(top_builddir)/libustctl/libustctl.la	\
+		 $(top_builddir)/liblttngctl/liblttngctl.la
diff --git a/ltt-sessiond/main.c b/ltt-sessiond/main.c
index 1139df8b9..1e2a0831d 100644
--- a/ltt-sessiond/main.c
+++ b/ltt-sessiond/main.c
@@ -2418,18 +2418,21 @@ end:
 }
 
 /*
- * Check if the global socket is available.  If yes, error is returned.
+ * Check if the global socket is available, and if a daemon is answering
+ * at the other side. If yes, error is returned.
  */
 static int check_existing_daemon(void)
 {
 	int ret;
 
-	ret = access(client_unix_sock_path, F_OK);
-	if (ret == 0) {
-		ret = access(apps_unix_sock_path, F_OK);
-	}
-
-	return ret;
+	if (access(client_unix_sock_path, F_OK) < 0 &&
+	    access(apps_unix_sock_path, F_OK) < 0)
+		return 0;
+	/* Is there anybody out there ? */
+	if (lttng_session_daemon_alive())
+		return -EEXIST;
+	else
+		return 0;
 }
 
 /*
@@ -2714,11 +2717,9 @@ int main(int argc, char **argv)
 	DBG("Application socket path %s", apps_unix_sock_path);
 
 	/*
-	 * See if daemon already exist. If any of the two socket needed by the
-	 * daemon are present, this test fails. However, if the daemon is killed
-	 * with a SIGKILL, those unix socket must be unlinked by hand.
+	 * See if daemon already exist.
 	 */
-	if ((ret = check_existing_daemon()) == 0) {
+	if ((ret = check_existing_daemon()) < 0) {
 		ERR("Already running daemon.\n");
 		/*
 		 * We do not goto exit because we must not cleanup()