ip autoconfig skips ipv6 dev initialization

ip autoconfig skips ipv6 dev initialization

[Dale Farnsworth]

I've been doing some ipv6 testing here on some nfs-rooted 2.4.18 systems.
I found that ipv6 initialization isn't performed on an ipv4-autoconfigured
network device, leaving it without a link-local address.  It happens
because ip_auto_config() is executed before inet6_init() has a chance
to register to handle NETDEV_UP notifications.  This only occurs when
ipv6 support is compiled into the kernel.

A fix is to do the ipv6 initialization before ipv4 initialization.
The included patch resolved it for me.  Please apply or suggest an
alternative.

Thanks,
-Dale Farnsworth

--- linux.orig/net/Makefile	Tue May 14 12:33:40 2002
+++ linux/net/Makefile	Tue May 14 12:34:35 2002
@@ -15,10 +15,10 @@
 
 
 subdir-$(CONFIG_NET)		+= 802 sched netlink
+subdir-$(CONFIG_IPV6)		+= ipv6
 subdir-$(CONFIG_INET)		+= ipv4
 subdir-$(CONFIG_NETFILTER)	+= ipv4/netfilter
 subdir-$(CONFIG_UNIX)		+= unix
-subdir-$(CONFIG_IPV6)		+= ipv6
 
 ifneq ($(CONFIG_IPV6),n)
 ifneq ($(CONFIG_IPV6),)

Re: ip autoconfig skips ipv6 dev initialization

[David S. Miller]

   From: "Dale Farnsworth" <Dale.Farnsworth@mvista.com>
   Date: Tue, 14 May 2002 15:32:30 -0700
   
   A fix is to do the ipv6 initialization before ipv4 initialization.
   The included patch resolved it for me.  Please apply or suggest an
   alternative.

Ordering of object files does not guarentee ordering of initcalls.
The linker can legitimately move things around in the final object
which will change the initcall ordering, the linker will in fact
do this on many platforms for performance purposes.

This bug can only be fixed correctly in 2.5.x which has a
prioritization scheme for initcalls.