Ask for help about the change of libnl 3.2.16 or newer version

[Roopa Prabhu]

On 1/27/13 8:50 AM, Andy Wang wrote:
> Hi friends of Libnl organization:
>
> I am a Watchguard employee and quite new to Libnl framework. In about
> Jan 5^th we upgraded to the new version 3.2.16 . The new version is good
> ,but we found we cannot add the IPV6 address in the new version.
>
> Function "rtnl_link_name2i" in this version always return integer value
> *0*which means it fails to get a valid interface index from cache. I
> also tried the newest version 3.2.20, it has the same problem.
>
> I copied the function of our company in which it called Libnl function
> to add IPV6 address. It works well in the past version before 3.2.16,
> but it failed in the line which is drawn as green after it.
>
> -------------------------
>
> /* This adds single address to interface */
>
> int wgnet_add_addr(const char * ifname,struct if_addr * ip) {
>
> struct nl_cache *link_cache = NULL;
>
> struct nl_addr *n_addr = NULL;
>
> struct rtnl_addr *r_addr = NULL;
>
> int ifindex;
>
> int ret=-1;
>
> if (!ip || !ifname) {
>
> return EINVAL;
>
> }
>
> if (!handle && (ret=wgnet_init()) != 0 ) {
>
> return ret;
>
> }
>
> rtnl_link_alloc_cache(handle,ip->ipaddr.sa_family,&link_cache);
>
> r_addr = rtnl_addr_alloc();
>
> /* FIXME we can use local version at some point */
>
> n_addr=nl_addr_alloc(sizeof(ip->ipaddr.ip6));
>
> if (!link_cache || !r_addr || !n_addr) {
>
> ret = ENOMEM;
>
> goto error;
>
> }
>
> **
>
> *ifindex = rtnl_link_name2i(link_cache, ifname); *
>
> if (!ifindex) {
>
> ret = ENOENT;
>
> goto error;
>
> }
>
> /* Copy data from provided data */
>
> nl_addr_set_family(n_addr,ip->ipaddr.sa_family);
>
> nl_addr_set_binary_addr(n_addr,ip->ipaddr.addr,ip->ipaddr.sa_family==AF_INET
> ? sizeof(ip->ipaddr.ip): sizeof(ip->ipaddr.ip6)) ;
>
> nl_addr_set_prefixlen(n_addr,ip->ipaddr.prefix);
>
> /* set netlink message and sent it out */
>
> rtnl_addr_set_local(r_addr,n_addr);
>
> rtnl_addr_set_ifindex(r_addr, ifindex);
>
> if ( ip->scope ) {
>
> rtnl_addr_set_scope(r_addr,ip->scope);
>
> }
>
> if ( ip->flags ) {
>
> rtnl_addr_set_flags(r_addr,ip->flags);
>
> }
>
> ret=rtnl_addr_add(handle, r_addr, 0);
>
> error:
>
> *****
>
> return ret;
>
> }
>
> -------------------------
>
> I compared the source code of the past version 3.2.13 and the current
> version 3.2.16 , and found there are the following modification in
> function “rtnl_link_get_by_name” which “rtnl_link_name2i” calls.
>
> Then I comment this new-added code and build the newlibnl* .so files,
> and use them again. Then the IPV6 address could be added rightly like
> before.
>
> I think there should have some reasons why the new code is added, but I
> can’t understand more.

This was something I added. Sorry if this broke your code.
This code was added during adding AF_BRIDGE support to link cache.

The reason this was added is, link cache can now contain objects of 
AF_BRIDGE, And i wanted to make sure rtnl_link_get_by_name did not break 
and return AF_BRIDGE objects.

But from what you are saying, rtnl_link_get_by_name returned the first 
link object with matching ifindex and that could have also been an 
object of AF_INET6 family. And i assumed that this was only of family 
AF_UNSPEC.

A straightforward fix is to also add the AF_INET6 check in 
rtnl_link_get_by_name there.

Or maybe there is a better solution. Let me think about this some more.


> So I ask help from you and expect to make it more
> clear.
>
> My questions is :
>
> 1.Does my change of rolling back this line reasonable? Will it affect
> other fiends?
>
> 2.If my change is not reasonable, is it possible to change Watchguard
> code to adapt the new version of Libnl. My current purpose is to let the
> adding of IPV6 address work rightly.
>
> Thanks again for your time to read my long Email. Expecting for any
> comments!
>
> *Thanks & Regards! *
>
> *YuTong Wang *
>
>
>
> _______________________________________________
> libnl mailing list
> libnl at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/libnl