Re: Linux 2.5.5-dj1 - IPv6 not loading correctly.

[Dave Jones]

On Sun, Feb 24, 2002 at 10:16:16PM -0800, Ben Clifford wrote:

 > Looking at the code, the the ICMP6 control socket error is occurring
 > because sock_register isn't called for inet6 until after the ICMP6 control
 > socket is created (in af_inet6.c).
 > However, the ICMP6 control socket create calls sock_create, which requires
 > sock_register to have already been called.

 This probably happened during acme's recent protocol cleanups,
 and is probably a problem in mainline as well as -dj.
 
 > I have made the below change, which moves the protocol family registration
 > higher up in the code.  It seems to make ipv6 work now.
 > 
 > However, I'm concerned that this gives a small amount of time when the
 > family is registered but not fully initialised.
 > Is this bad?

 I'll let davem/acme comment on the correctness of the fix..
 Looks straightforward enough to me, but I'm not as kneedeep in
 networking internals as those two 8-)

 > 
 > - --- /mnt/dev/hda11/2.5.5-dj1-snark-not-changed-much/net/ipv6/af_inet6.c	Tue Feb 19 18:10:53 2002
 > +++ 2.5.5-dj1/net/ipv6/af_inet6.c	Sun Feb 24 22:13:38 2002
 > @@ -675,6 +675,13 @@
 >  	 */
 >  	inet6_register_protosw(&rawv6_protosw);
 > 
 > +	/* register the family here so that the init calls below will
 > +	 * work. ?? is this dangerous ??
 > +	 */
 > +
 > +	(void) sock_register(&inet6_family_ops);
 > +
 > +
 >  	/*
 >  	 *	ipngwg API draft makes clear that the correct semantics
 >  	 *	for TCP and UDP is to consider one TCP and UDP instance
 > @@ -719,9 +726,6 @@
 >  	udpv6_init();
 >  	tcpv6_init();
 > 
 > - -	/* Now the userspace is allowed to create INET6 sockets. */
 > - -	(void) sock_register(&inet6_family_ops);
 > - -
 >  	return 0;
 > 
 >  #ifdef CONFIG_PROC_FS
 > 

-- 
| Dave Jones.        http://www.codemonkey.org.uk
| SuSE Labs