Sysctl addr_gen_mode does not control tunnel link-local addr

Sysctl addr_gen_mode does not control tunnel link-local addr

[Stephen Suryaputra]

Hi,

I noticed that tunnels, especially sit before commit e5dd729460ca8
("ip/ip6_gre: use the same logic as SIT interfaces when computing v6LL
address"), generates link-local addr regardless of addr_gen_mode
setting. Is this a bug, or is there a specific reason?

In my system, the link-local addr are generated by a userspace process.
So, we set net.ipv6.conf.<dev>.addr_gen_mode to 1 (IN6_ADDR_GEN_MODE_NONE).

The commit e5dd729460ca8 doesn't change the behavior as it is renaming
sit_add_v4_addrs() to add_v4_addrs() as make it generic for GRE/IP6GRE
cases. I'm not sure what the current behavior is for GRE, i.e. whether
addr_gen_mode can control the generation of link-local addr.

If this is a bug or oversight, I think this diff should fix it and I can
put a formal patch.

diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
index c6a90b7bbb70..da7e83699eef 100644
--- a/net/ipv6/addrconf.c
+++ b/net/ipv6/addrconf.c
@@ -3392,6 +3392,8 @@ static void addrconf_sit_config(struct net_device
*dev)
                return;
        }

+       if (idev->cnf.addr_gen_mode == IN6_ADDR_GEN_MODE_NONE)
+               return;
        add_v4_addrs(idev);

        if (dev->flags&IFF_POINTOPOINT)

The commit mention about addressing violation of RFC4291 on GRE and the diff
above doesn't cause it to change as the default addr_gen_mode is 0
(IN6_ADDR_GEN_MODE_EUI64).

Thanks,
Stephen.

Re: Sysctl addr_gen_mode does not control tunnel link-local addr

[Antonio Quartulli]

Hi,

On 20/10/2021 03:30, Stephen Suryaputra wrote:
> Hi,
> 
> I noticed that tunnels, especially sit before commit e5dd729460ca8
> ("ip/ip6_gre: use the same logic as SIT interfaces when computing v6LL
> address"), generates link-local addr regardless of addr_gen_mode
> setting. Is this a bug, or is there a specific reason?
> 
> In my system, the link-local addr are generated by a userspace process.
> So, we set net.ipv6.conf.<dev>.addr_gen_mode to 1 (IN6_ADDR_GEN_MODE_NONE).
> 
> The commit e5dd729460ca8 doesn't change the behavior as it is renaming
> sit_add_v4_addrs() to add_v4_addrs() as make it generic for GRE/IP6GRE
> cases. I'm not sure what the current behavior is for GRE, i.e. whether
> addr_gen_mode can control the generation of link-local addr.
> 
> If this is a bug or oversight, I think this diff should fix it and I can
> put a formal patch.

When I implemented this behavior for SIT interfaces I basically ported
what is currently happening for GRE.

Therefore, any change you apply to SIT, should also be reflected in the
GRE code path.

As of now, the SIT/GRE addrgen logic does not consider the mode setting
at all. But I am not sure it'd be easy to support all the values.

For sure, supporting IN6_ADDR_GEN_MODE_NONE is a low hanging fruit and
should be implemented.

> 
> diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
> index c6a90b7bbb70..da7e83699eef 100644
> --- a/net/ipv6/addrconf.c
> +++ b/net/ipv6/addrconf.c
> @@ -3392,6 +3392,8 @@ static void addrconf_sit_config(struct net_device
> *dev)
>                 return;
>         }
> 
> +       if (idev->cnf.addr_gen_mode == IN6_ADDR_GEN_MODE_NONE)
> +               return;

Please add an empty line here.

>         add_v4_addrs(idev);
> 
>         if (dev->flags&IFF_POINTOPOINT)
> 
> The commit mention about addressing violation of RFC4291 on GRE and the diff
> above doesn't cause it to change as the default addr_gen_mode is 0
> (IN6_ADDR_GEN_MODE_EUI64).

Other than my nitpick above, I'd suggest applying the same change to the
GRE code path, as the behavior should be the same.

Moreover, do you think that addrconf_add_mroute() should not be executed
in this case? Otherwise you could simply add your new code into
add_v4_addrs().

Regards,

-- 
Antonio Quartulli

Re: Sysctl addr_gen_mode does not control tunnel link-local addr

[Stephen Suryaputra]

On Wed, Oct 20, 2021 at 02:59:01PM +0200, Antonio Quartulli wrote:

> For sure, supporting IN6_ADDR_GEN_MODE_NONE is a low hanging fruit and
> should be implemented.
> 
Thanks for reaffirming this!

> Other than my nitpick above, I'd suggest applying the same change to the
> GRE code path, as the behavior should be the same.

> Moreover, do you think that addrconf_add_mroute() should not be executed
> in this case? Otherwise you could simply add your new code into
> add_v4_addrs().

Initially I was thinking that why there is a need to call
addrconf_add_mroute() given the link-local addr hasn't been set. On a
second look, yes, the conditional is better be in add_v4_addrs(). It
will take care of GRE as well.

I'll put a formal patch.

Thanks!