Attachment 198661 Details for Bug 233481 – Don't leak any file descriptors on exec()

[patch] Don't leak any file descriptors on exec()

autofs-5.0.1-dont-leak-fds-on-exec.patch (text/plain), 1.33 KB, created by Jeff Moyer on 2007-09-18 18:14:46 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Jeff Moyer

Created: 2007-09-18 18:14:46 UTC

Size: 1.33 KB

patch

obsolete

>--- autofs-5.0.1/daemon/spawn.c.fd-leak	2007-09-18 12:20:33.000000000 -0400
>+++ autofs-5.0.1/daemon/spawn.c	2007-09-18 14:07:32.000000000 -0400
>@@ -89,7 +89,7 @@ void reset_signals(void)
> static int do_spawn(logger *log, unsigned int options, const char *prog, const char *const *argv)
> {
> 	pid_t f;
>-	int ret, status, pipefd[2];
>+	int ret, status, pipefd[2], fd;
> 	char errbuf[ERRBUFSIZ + 1], *p, *sp;
> 	int errp, errn;
> 	int cancel_state;
>@@ -158,6 +158,24 @@ static int do_spawn(logger *log, unsigne
> 			setpgid(0, pgrp);
> 		}
> 
>+		/*
>+		 * The daemon is multi-threaded, and thus we can end
>+		 * up in a situation where another lookup is happening
>+		 * while this code runs.  What this means is that the
>+		 * other lookup could end up opening up file
>+		 * descriptors (sockets and the like) in order to do
>+		 * DNS requests, LDAP connections, or NIS lookups.  If
>+		 * we are lucky enough, then we call exec() with these
>+		 * file descriptors open (and FD_CLOEXEC not set) so
>+		 * we leak them to the exec'd process.  The most
>+		 * pragmatic way around this, given the current
>+		 * structure of the code, is to close all file
>+		 * descriptors above 2 (since we've dup'd
>+		 * stdin,out,err to 0, 1, 2).
>+		 */
>+		for (fd = 3; fd < 1024; fd++)
>+			close(fd);
>+
> 		execv(prog, (char *const *) argv);
> 		_exit(255);	/* execv() failed */
> 	} else {

Actions: View | Diff

Attachments on bug 233481: 198661 | 275541