On Thu, Nov 17, 2016 at 11:42:38AM +0000, Stuart Henderson wrote:
> On 2016/11/10 00:57, Jeremie Courreges-Anglas wrote:
> >
> > The following diff adds support for listening multiple addresses (thus
> > for dual-stack setups). Multiple "listen on" settings are allowed, the
> > default is to listen on 0.0.0.0 and :: (currently, only 0.0.0.0).
> > A single "listen on hostname" line arbitrarily supports up to 16
> > addresses.
> >
> > It also tweaks the host*() functions so that addresses are used in the
> > order where they are resolved*. This affects bind addresses, but also
> > "trap receiver" addresses. So in "trap receiver hostname", if hostname
> > resolves to both an IPv4 and an IPv6 address, the address that is picked
> > up respects the order defined by the "family" keyword in resolv.conf.
> > This *could* break existing setups.
>
> I think this behaviour is completely sane for "trap receiver". If something
> else is needed, it can be tied to address-family (e.g. "source-address ::"
> or "source-address 0.0.0.0") or of course a specific address.
>
> Using "listen" with a hostname seems odd anyway. Does anyone reading use
> this? I can't think of anywhere that I'd do so (and would expect it to be
> fragile if I did). I don't think that's a barrier for your diff. (We could
> avoid ambiguity by disallowing names there, as is already the case for
> "source-address", but I don't think it really hurts either way).
>
People sometimes use it and it makes sense in combination with
/etc/hosts. Especially with IPv6 ... and snmpd currently lacks the
interface/group lookup that allows other daemons to
"listen on egress"
Anyway, I think it is OK to listen on hostnames here.
> I've tested it, including with various combinations of name and IP address
> in "trap receiver" and either v4 or v6 in "source-address" - everything worked
> as expected and diff reads good to me.
>
> OK sthen@
>
Not yet from me, see other mail.
Reyk
