[BusyBox] nc problem still stumps me

[Jim Treadway]

On Fri, 3 May 2002, Tom Oehser wrote:

> Well, I looked at the strace and the source for the original nc and
> the busybox nc, and I don't see anything obvious why the busybox nc
> doesn't work the same way.  If I have something like:
> 
> while [ true ] ; do nc -l -p 11111 -e /bin/true ; done
> 
> and repeatedly connect to it, the real nc will continue to work every
> time, but the busybox nc will work once and then fail saying that the
> port is already in use and it can't bind.  What I don't understand is
> that it seems to be the condition busybox *leaves* it in that is bad,
> not the way it initializes it- that is, after the busybox one runs,
> the real nc can't bind it, either.
> 
> But, what confuses me is that nc just hands the file descriptor off
> to an execed process- so nc should have *nothing* to do with how it
> is handled when that process ends.  There must be some flag or socket
> option or setup that affects the socket itself?
> 
> I'll keep at it, but I'm open to ideas...

I think it just needs to set SO_REUSEADDR on the socket; the following 
patch should fix it.

Jim

--- ./networking/nc.c.orig	Mon May  6 12:46:52 2002
+++ ./networking/nc.c	Mon May  6 12:43:39 2002
@@ -47,7 +47,8 @@
 	int do_listen = 0, lport = 0, delay = 0, tmpfd, opt, sfd, x;
 	char buf[BUFSIZ];
 #ifdef GAPING_SECURITY_HOLE
-	char * pr00gie = NULL;                  
+	char *pr00gie = NULL;
+	int flags;
 #endif
 
 	struct sockaddr_in address;
@@ -100,6 +101,12 @@
 
 		if (bind(sfd, (struct sockaddr *) &address, sizeof(address)) < 0)
 			perror_msg_and_die("bind");
+#ifdef GAPING_SECURITY_HOLE
+		flags = 1;
+		if (setsockopt(sfd, SOL_SOCKET, SO_REUSEADDR, &flags,
+			       sizeof(flags)))
+			perror_msg_and_die("setsockopt(SO_REUSEADDR)");
+#endif
 	}
 
 	if (do_listen) {