[quagga-dev 5017] Re: A patch for 0.99.8 to fix IPv6 PtP static routes

On Fri, Aug 03, 2007 at 03:41:16PM +0200, Emmanuel Vize wrote:
> I am currently merging my Quagga version with the 0.99.8 release.
> I found one problem: Static IPv6 routes in 6in6 tunnels (PtP) are not 
> working anymore, the routes via the other endpoint of the interface are 
> inactive.
> I observe this on Linux.
> 
> I would suggest the patch below:
> 
> 
> --- rt_netlink.c.orig   Fri Aug  3 14:27:05 2007
> +++ rt_netlink.c        Fri Aug  3 15:20:51 2007
>  <at>  <at>  -712,12 +712,10  <at>  <at> 
>   addr = (tb[IFA_LOCAL] ? RTA_DATA(tb[IFA_LOCAL]) : NULL);
> 
>   /* is there a peer address? */
> -  /* N.B. I do not understand why the memcmp compares 4 bytes regardless
> -     of address family, but this is exactly how it appears in
> -     print_addrinfo.  I wonder if it should be RTA_PAYLOAD(tb[IFA_ADDRESS])
> -     instead of 4... */
>   if (tb[IFA_ADDRESS] &&
> -      memcmp(RTA_DATA(tb[IFA_ADDRESS]), RTA_DATA(tb[IFA_LOCAL]), 4))
> +      memcmp(RTA_DATA(tb[IFA_ADDRESS]),RTA_DATA(tb[IFA_LOCAL]),
> +                  ifa->ifa_family == AF_INET ? 4 :
> +                  ifa->ifa_family == AF_INET6 ? 16 : /* fall back*/ 4))
>     {
>       broad = RTA_DATA(tb[IFA_ADDRESS]);
>       SET_FLAG (flags, ZEBRA_IFA_PEER);
> 
> 
> 
> With this patch, it is working much better.
> 
> If no objections, I will commit next week.

I copied this code originally from iproute2/ip/ipaddress.c:print_addrinfo
(as indicated in comment).  The code there looks like this:

	if (rta_tb[IFA_LOCAL]) {
		fprintf(fp, "%s", rt_addr_n2a(ifa->ifa_family,
					      RTA_PAYLOAD(rta_tb[IFA_LOCAL]),
					      RTA_DATA(rta_tb[IFA_LOCAL]),
					      abuf, sizeof(abuf)));

		if (rta_tb[IFA_ADDRESS] == NULL ||
		    memcmp(RTA_DATA(rta_tb[IFA_ADDRESS]), RTA_DATA(rta_tb[IFA_LOCAL]), 4) == 0) {
			fprintf(fp, "/%d ", ifa->ifa_prefixlen);
		} else {
			fprintf(fp, " peer %s/%d ",
				rt_addr_n2a(ifa->ifa_family,
					    RTA_PAYLOAD(rta_tb[IFA_ADDRESS]),
					    RTA_DATA(rta_tb[IFA_ADDRESS]),
					    abuf, sizeof(abuf)),
				ifa->ifa_prefixlen);
		}
	}

It always seemed strange to me that the 3rd argument to memcmp is 4
(as I also indicated in the comment), but your patch also seems strange
to me: wy do you have 

                  ifa->ifa_family == AF_INET ? 4 :
                  ifa->ifa_family == AF_INET6 ? 16 : /* fall back*/ 4))

instead of something more general like RTA_PAYLOAD(tb[IFA_ADDRESS])?

And just out of curiosity, on the system that shows this bug, does
'/sbin/ip addr ls' give incorrect results?

Regards,
Andy