[Xorp-hackers] Patch for a netlink problem: Can't set MAC address

[Pavlin Radoslavov]

Pavlin Radoslavov <pavlin at icir.org> wrote:

> > In fea/ifconfig_set_netlink.cc , there is code that attempts to
> > set interface MAC addresses.  This code fails to work correctly
> > because the netlink payload that it sends is a simple MAC address.
> > The kernel code is looking for a struct sockaddr.
> 
> Thank you for the fix. Now it is applied to CVS:
> 
> Revision  Changes                               Path
> 1.28      +35 -4;  commitid: 3cdc44d7ee127ea6; xorp/fea/ifconfig_set_netlink.cc
> 
> > One odd-ball thing is that the payload length (rta_len) must be
> > truncated.  It must be set to the sizeof
> > sa_family+mac_address_length rather than the larger, and more
> > correct, sizeof sockaddr
> 
> Setting the payload length to sa_family+mac_address_length is
> probably intentiional, because this mechanism is suppose to work
> with large hardware addresses as indicated in the following email:
> 
> http://marc2.theaimsgroup.com/?l=linux-kernel&m=104137641925584&w=2
> For the record, the patch in that email is an earlier version of
> what actually was added to the kernel:
> http://www.ussg.iu.edu/hypermail/linux/kernel/0301.1/2436.html
> 
> However, what I found really odd is the fact that the payload is
> suppose to be a MAC address contained inside "struct sockaddr", but
> rta_len only is suppose to be ETH_ALEN, when the correct value
> should be sizeof(sa_family) + mac_address_length.
> I believe this is a bug in the Linux kernel API (at least version
> 2.6.17).
> 
> The particular problematic code in the kernel is inside
> net/core/rtnetlink.c, do_setlink():
>     ...
>     if (ida[IFLA_ADDRESS - 1]) {
>         ...
>         if (ida[IFLA_ADDRESS - 1]->rta_len != RTA_LENGTH(dev->addr_len))
>             goto out;
>             err = dev->set_mac_address(dev, RTA_DATA(ida[IFLA_ADDRESS - 1]));
>             ...
> 
>         [ Where dev->set_mac_address() expects to see a second argument of
>           type "struct sockaddr". ]
> 
> I am going to send an email to the author of that patch about the
> issue.

I sent an email to the linux-kernel ML, and it turned that this is a
Linux kernel bug which will be fixed in both 2.6.18 and
2.6.17.x-stable.
See the following thread for details:
http://www.spinics.net/lists/kernel/msg498873.html

After the fix, the RTM_SETLINK payload for setting the MAC address
will be the MAC address (only) without the "sockaddr" wrapper around
it.

Hence, I am going to reverse the recent change to
fea/ifconfig_set_netlink.cc (rev 1.28) after a Linux kernel with the
fix is released, because the original code will be working correctly
with that kernel.

Pavlin