svcj: correctly handle kernels without INET or INET6

[ivy]

Summary

if either INET or INET6 is not enabled in the kernel, then the jail(8)
options ip4=<new|inherit> resp. ip6=<new|inherit> are not available.
detect this case and don't try to provide those options, otherwise
svcjs will not start.

do this automatically (without a warning) so that net_basic, which
includes both netv4 and netv6, continues to work as expected.

if _svcj_ipaddrs is explicitly configured with an address for an IP
version not supported by the kernel, issue a warning but continue to
start the service. this can result in the service being started with
fewer addresses than expected, but never more.

[des]

Wouldn't it be better to check at point of use rather than at point of initialization?

[netchild]

libexec/rc/rc.subr

1253

What if inet is enabled but inet6 is not, but there are ip6 addrs? If I'm not mistaken, in that case you have set this it to ip6=new above, and this branch is wrongly taken.
To me it looks it's better to simply set a "new" flag, and set both to new after this check and before the "translate service jail options" part.

[ivy]

libexec/rc/rc.subr

1253

yes, you're right. i'll rethinking this and see if i can address des comment at the same time.

[ivy]

i think this is a more reasonable way to do it, and seems to work in testing.
i'd like to add a proper set of tests for this, but since the test jails run
with path=/, i'm not sure there's a reasonable way to test this kind of thing.

[netchild]

In D49976#1139747, @des wrote:

Wouldn't it be better to check at point of use rather than at point of initialization?

I see your point. On the other hand I think the part which sets "the other one" may be more easy at the current place than at the place of use. I also like the current way of having the case statement "nice" (not convoluted with checks but simply listing what it entails).