Ok... Yet another round of fixes for smbmount and autofs.

1)  The earlier fix for the smbmount race conditions broke the PID
	registration with smbfs.  That fix has been backed out and
	replaced by a signalling convention from the child smbmount
	process back to the parent telling the parent when it is safe
	to exit.

2)  Fixing all of this uncovered a NASTY deadly embrace between smbmount,
	smbmnt, and autofs.  This was caused by the setsid call in the
	daemon code.  The smbmnt process no longer was registered as
	"magic" because it was no longer in the autofs process group.
	Many many kudos and thanks to H Peter Anvin for giving me the
	clue to solving this agravating puzzle.  The setsid was moved
	down the where the child signals the parent and a warn left in
	its place in the daemonize code.

3)  Fixed (actually worked around with a BUTT UGLY HACK) a problem with
	SMB_GET_MOUNTPID in smbumount.c.  The smb_fs.h header file has
	the parameter to this ioctl defined as a uid_t.  Unfortunately
	that's a 32 bit quantity under glibc and it's currently a 16 bit
	quantity in kernel space.  Undefined the macro and redefined
	it with a parameter of __kernel_uid_t.  That should keep us
	out of trouble till I can have someone fix smb_fs.h in the
	kernel sources...
(This used to be commit c5608093e48ed7fd3b7a1e2a4b1af7e4e59fc2f9)

1 files changed, 55 insertions, 18 deletions

diff --git a/source3/client/smbmount.c b/source3/client/smbmount.c
index 2cfd166688..3c2d207498 100644
--- a/source3/client/smbmount.c
+++ b/source3/client/smbmount.c
@@ -38,7 +38,7 @@ static struct smb_conn_opt conn_options;
 #endif
 
 /* Uncomment this to allow debug the smbmount daemon */
-#define SMBFS_DEBUG 1
+/* #define SMBFS_DEBUG 1 */
 
 pstring cur_dir = "\\";
 pstring cd_path = "";
@@ -71,7 +71,6 @@ extern int name_type;
 extern int max_protocol;
 int port = SMB_PORT;
 
-
 time_t newer_than = 0;
 int archive_level = 0;
 
@@ -182,19 +181,50 @@ static BOOL chkpath(char *path,BOOL report)
 }
 
 static void
+exit_parent( int sig )
+{
+	/* parent simply exits when child says go... */
+	exit(0);
+}
+
+static void
 daemonize(void)
 {
-	int i;
-	if ((i = fork()) < 0)
+	int j, status;
+	pid_t child_pid;
+
+	signal( SIGTERM, exit_parent );
+
+	if ((child_pid = fork()) < 0)
 	{
 		DEBUG(0, ("could not fork\n"));
 	}
-	if (i > 0)
+	if (child_pid > 0)
 	{
-		/* parent simply exits */
-		exit(0);
+		while( 1 ) {
+			j = waitpid( child_pid, &status, 0 );
+			if( j < 0 ) {
+				if( EINTR == errno ) {
+					continue;
+				}
+				status = errno;
+			}
+			break;
+		}
+		/* If we get here - the child exited with some error status */
+		exit(status);
 	}
-	setsid();
+	/* Programmers Note:
+		Danger Will Robinson!  Danger!
+
+		There use to be a call to setsid() here.  This does no
+		harm to normal mount operations, but it broke automounting.
+		The setsid call has been moved to just before the child
+		sends the SIGTERM to the parent.  All of our deadly embrace
+		conditions with autofs will have been cleared by then...
+		-mhw-
+	*/
+	signal( SIGTERM, SIG_DFL );
 	chdir("/");
 }
 
@@ -255,7 +285,8 @@ static void
 send_fs_socket(char *mount_point, char *inbuf, char *outbuf)
 {
 	int fd, closed = 0, res = 1;
-	int first_time = 1;
+
+	pid_t parentpid = getppid();
 
 	while (1)
 	{
@@ -278,6 +309,15 @@ send_fs_socket(char *mount_point, char *inbuf, char *outbuf)
 			DEBUG(0, ("smbmount: ioctl failed, res=%d\n", res));
 		}
 
+		if( parentpid ) {
+			/* Ok...  We are going to kill the parent.  Now
+				is the time to break the process group... */
+			setsid();
+			/* Send a signal to the parent to terminate */
+			kill( parentpid, SIGTERM );
+			parentpid = 0;
+		}
+
 		close_sockets();
 		close(fd);
 		/*
@@ -291,15 +331,6 @@ send_fs_socket(char *mount_point, char *inbuf, char *outbuf)
 		}
 #endif
 
-		if( first_time ) {
-	/*
-	 * Create the background process after trying the mount.
-	 * to avoid race conditions with automount and other processes.
-	 */
-			first_time = 0;
-			daemonize();
-		}
-
 		/*
 		 * Wait for a signal from smbfs ...
 		 */
@@ -368,6 +399,12 @@ static void cmd_mount(char *inbuf,char *outbuf)
 
 	DEBUG(3, ("mount command: %s\n", mount_command));
 
+	/*
+		Set up to return as a daemon child and wait in the parent
+		until the child say it's ready...
+	*/
+	daemonize();
+
 	if ((retval = system(mount_command)) != 0)
 	{
 		DEBUG(0,("mount failed\n"));