tun(4)/tap(4): allow devices to be configured as transient

[kevans]

Summary

Transient tunnel devices are removed immediately after last close, so
that an application that's created the tunnel could eliminate the need
to manually destroy the tunnel whose lifetime it's already managing.

[zlei]

sys/net/if_tuntap.c

1257

What about making ifc_find_cloner_in_vnet() public and using it here ?
So we do not need complex logic in if_clone_destroyif_flags() .

[glebius]

sys/net/if_tuntap.c

1257

We have the knowledge about the cloner inside the tuntap driver. You don't need to hack if_clone.c to achieve what you want. Let me come with a prototype quickly...

[kevans]

sys/net/if_tuntap.c

1257

Right, I had a version that looked up the cloner, but it was a little gross because you also need to switch back to the home vnet and grab the cloner from that particular vnet, effectively duplicating what's already done in if_clone.c.

[glebius]

Damn, this again comes to the stupid idea of interfaces traveling between vnets. Grrrr. Still, I believe it should be done in the tuntap, not in if_clone.c. The existing SLIST of cloners can be simplified a lot, let me try.

[kevans]

In D44200#1010509, @glebius wrote:

Damn, this again comes to the stupid idea of interfaces traveling between vnets. Grrrr. Still, I believe it should be done in the tuntap, not in if_clone.c. The existing SLIST of cloners can be simplified a lot, let me try.

I don't really feel that strongly about it either way, it just seemed less error prone since what if_clone finds should always match which list it's actually on, in case someone gets a wild tendency and decides that an interface moving between vnets should be reflected in the cloners instead

[glebius]

So, with D44307 in, you can add this:

static struct if_clone *
tuntap_cloner_from_flags(int tun_flags)
{
 
        for (u_int i = 0; i < NDRV; i++)
                if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
                    tuntap_drivers[i].ident_flags)
                        return (V_tuntap_driver_cloners[i]);
 
        return (NULL);
}

And then call

if_clone_destroyif(tuntap_cloner_from_flags(flags), ifp);

This of course won't cover the problematic traveling case. To cover that, you'd need to:

CURVNET_SET_QUIET(ifp->if_home_vnet);
cloner = tuntap_cloner_from_flags(flags)
CURVNET_RESTORE();
if_clone_destroyif(cloner, ifp);

[kevans]

In D44200#1010523, @glebius wrote:

So, with D44307 in, you can add this:

static struct if_clone *
tuntap_cloner_from_flags(int tun_flags)
{
 
        for (u_int i = 0; i < NDRV; i++)
                if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
                    tuntap_drivers[i].ident_flags)
                        return (V_tuntap_driver_cloners[i]);
 
        return (NULL);
}

And then call

if_clone_destroyif(tuntap_cloner_from_flags(flags), ifp);

This of course won't cover the problematic traveling case. To cover that, you'd need to:

CURVNET_SET_QUIET(ifp->if_home_vnet);
cloner = tuntap_cloner_from_flags(flags)
CURVNET_RESTORE();
if_clone_destroyif(cloner, ifp);

Alright, I'll drop the other review and update this one with the latter variant -- though I likely won't have any time to return to this for a couple weeks. I don't think this is a particularly urgent feature anyways, I just like the idea of allowing a program to construct a tunnel and not worry about the lifetime (and it's reportedly the default Linux behavior, so I could see how some tunnelling applications might benefit from a way to match it).

[kevans]

Update following better tracking of our cloners, add a basic test as well

[zlei]

Looks good to me.