Re: [BUG] gre interface incorrectly generates link-local addresses

[Jakub Kicinski <kuba@kernel.org>]

Adding Thomas as well.

On Fri, 24 Mar 2023 19:35:06 +0300 Aleksey Shumnik wrote:
> Dear Maintainers,
> 
> I found that GRE arbitrarily hangs IP addresses from other interfaces
> described in /etc/network/interfaces above itself (from bottom to
> top). Moreover, this error occurs on both ip4gre and ip6gre.
> 
> Example of mgre interface:
> 
> 13: mgre1@NONE: <MULTICAST,NOARP,UP,LOWER_UP> mtu 1400 qdisc noqueue
> state UNKNOWN group default qlen 1000
>     link/gre 0.0.0.0 brd 0.0.0.0
>     inet 10.10.10.100/8 brd 10.255.255.255 scope global mgre1
>        valid_lft forever preferred_lft forever
>     inet6 fe80::a0a:a64/64 scope link
>        valid_lft forever preferred_lft forever
>     inet6 fe80::7f00:1/64 scope host
>        valid_lft forever preferred_lft forever
>     inet6 fe80::a0:6842/64 scope host
>        valid_lft forever preferred_lft forever
>     inet6 fe80::c0a8:1264/64 scope host
>        valid_lft forever preferred_lft forever
> 
> It seems that after the corrections in the following commits
> https://github.com/torvalds/linux/commit/e5dd729460ca8d2da02028dbf264b65be8cd4b5f
> https://github.com/torvalds/linux/commit/30e2291f61f93f7132c060190f8360df52644ec1
> https://github.com/torvalds/linux/commit/23ca0c2c93406bdb1150659e720bda1cec1fad04
> 
> in function add_v4_addrs() instead of stopping after this check:
> 
> if (addr.s6_addr32[3]) {
>                 add_addr(idev, &addr, plen, scope, IFAPROT_UNSPEC);
>                 addrconf_prefix_route(&addr, plen, 0, idev->dev, 0, pflags,
>                                                                 GFP_KERNEL);
>                  return;
> }
> 
> it goes further and in this cycle hangs addresses from all interfaces on the gre
> 
> for_each_netdev(net, dev) {
>       struct in_device *in_dev = __in_dev_get_rtnl(dev);
>       if (in_dev && (dev->flags & IFF_UP)) {
>       struct in_ifaddr *ifa;
>       int flag = scope;
>       in_dev_for_each_ifa_rtnl(ifa, in_dev) {
>             addr.s6_addr32[3] = ifa->ifa_local;
>             if (ifa->ifa_scope == RT_SCOPE_LINK)
>                      continue;
>             if (ifa->ifa_scope >= RT_SCOPE_HOST) {
>                      if (idev->dev->flags&IFF_POINTOPOINT)
>                               continue;
>                      flag |= IFA_HOST;
>             }
>             add_addr(idev, &addr, plen, flag,
>                                     IFAPROT_UNSPEC);
>             addrconf_prefix_route(&addr, plen, 0, idev->dev,
>                                      0, pflags, GFP_KERNEL);
>             }
> }
> 
> Moreover, before switching to Debian 12 kernel version 6.1.15, I used
> Debian 11 on 5.10.140, and there was no error described in the commit
> https://github.com/torvalds/linux/commit/e5dd729460ca8d2da02028dbf264b65be8cd4b5f.
> One link-local address was always generated on the gre interface,
> regardless of whether the destination or the local address of the
> tunnel was specified.
> 
> Which linux distribution did you use when you found an error with the
> lack of link-local address generation on the gre interface?
> After fixing the error, only one link-local address is generated?
> I think this is a bug and most likely the problem is in generating
> dev->dev_addr, since link-local is formed from it.
> 
> I suggest solving this problem or roll back the code changes made in
> the comments above.