in.c autoadding prefix route [PATCH]

[Ruslan Ermilov]

On Wed, Nov 17, 2004 at 07:50:32PM +0300, Vladimir Grebenschikov wrote:
> В чт, 11/11/2004 в 21:24 +0100, Max Laier пишет: 
> > All,
> > 
> > I know I have sent this a couple of times before, but never got anywhere. This 
> > time I am set to commit!
> > 
> > The attached patch (http://people.freebsd.org/~mlaier/in.c.patch) derived from 
> > WIDE via OpenBSD in.c, rev 1.21 improves the handling of automatic prefix 
> > routes.
> > 
> > Right now you can't have two legs into the same network. If you want to, you 
> > must give on of the interfaces a host address only (netmask /32). This way it 
> > is not possible to hand over the route if one of the interfaces is 
> > "removed" (however this is done in the special case).
> > 
> > The patch allows to add more than on IPv4 address with the same prefix. In the 
> > case that there is a route already, we leave it alone and add the new address 
> > without the IFA_ROUTE flag. When we remove an address later on, that has a 
> > route associated, we try to find an alternative address to use for the route 
> > and hand it over.
> > 
> > This is required for CARP, but should be helpful for other situations as well.
> > 
> > Any objections?
> 
> This change actually broke one simple thing:
> 
> # ifconfig lo0 alias 10.0.16.111/32
> # ping 10.0.16.111
> PING 10.0.16.111 (10.0.16.111): 56 data bytes
> ping: sendto: No route to host
> ^C
> #
> 
> You should (anyway should) add routing of interface address itself to
> loop-back interface, like it usually done in all other cases.
> 
> Please fix.
> 
This bug was borrowed from OpenBSD (they still have it).
The problem is that in "struct in_ifaddr", initially

	ia_ifa.ifa_addr points to ia_addr
		- and -
	ia_ifa.ifa_dstaddr points to ia_dstaddr

and the IFF_LOOPBACK code in in_ifinit() just resets
the ia_ifa.ifa_dstaddr pointer to point to ia_ifa.ifa_addr.
But new in_addprefix() code ignores this fact, and just
uses INET fields directly, i.e., ia_dstaddr (which happens
to be 0.0.0.0).  Then all the havoc happens (it matches
another interface address (ia_addr=127.0.0.1, ia_dstaddr=0)
and refuses to install a host route.

The fix is a one line change, and as I later found out,
it also corresponds to NetBSD revision 1.83:

http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/netinet/in.c.diff?r1=1.82&r2=1.83&cvsroot=netbsd

%%%
Index: in.c
===================================================================
RCS file: /home/ncvs/src/sys/netinet/in.c,v
retrieving revision 1.78
diff -u -p -r1.78 in.c
--- in.c	12 Nov 2004 20:53:51 -0000	1.78
+++ in.c	17 Nov 2004 21:42:44 -0000
@@ -759,7 +759,7 @@ in_ifinit(ifp, ia, sin, scrub)
 		ia->ia_netbroadcast.s_addr =
 			htonl(ia->ia_net | ~ ia->ia_netmask);
 	} else if (ifp->if_flags & IFF_LOOPBACK) {
-		ia->ia_ifa.ifa_dstaddr = ia->ia_ifa.ifa_addr;
+		ia->ia_dstaddr = ia->ia_addr;
 		flags |= RTF_HOST;
 	} else if (ifp->if_flags & IFF_POINTOPOINT) {
 		if (ia->ia_dstaddr.sin_family != AF_INET)
%%%

An alternative would be to fix in_addprefix() to pay
attention to ia_ifa pointers, but as this bug shows,
this is error prone.  It's much easier to just
initialize the ia_dstaddr as appropriate.
	

Cheers,
-- 
Ruslan Ermilov
ru at FreeBSD.org
FreeBSD committer
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-net/attachments/20050703/55e8e429/attachment.bin