[cfarm-users] having a bit of trouble with getting into the new netbsd and freebsd virtual machines

[Olly Betts]

On Sat, Oct 06, 2018 at 04:04:52PM -0500, Segher Boessenkool wrote:
> On Sat, Oct 06, 2018 at 09:25:25PM +0100, Olly Betts via cfarm-users wrote:
> > gcc300$ getaddrinfo -f inet -t stream ::1
> > stream inet tcp 192.168.66.50 0
> 
> $ route -n show
> 
> tells you why this is.

Thanks, but I'm failing to understand the explanation if it is there
(I'd actually even already tried that command).

On Sat, Oct 06, 2018 at 03:14:12PM -0600, Assaf Gordon wrote:
> On 06/10/18 03:06 PM, Assaf Gordon wrote:
> > The physical server hosting the VMs is 192.168.66.50.

OK, so it isn't just an arbitrary IPv4 address on the same network.

> > The NetBSD VM is 192.168.66.202.
> > The public IP (184.68.105.38) machine forwards the TCP ports (e.g. 2400)
> > to the internal VMs.
> > 
> > This seems pretty standard to me, but if there are possible
> > improvements to the VM's configuration I'm happy to try them.

It sounds like an entirely reasonable configuration to me.

> I should add that the physical server is a Debian 9 (stretch),
> using libvirt as the virtual machine infrastructure,
> and the networking is setup to "bridge" mode, as explained here:
> 
> https://wiki.libvirt.org/page/Networking#Bridged_networking_.28aka_.22shared_physical_device.22.29
> 
> Perhaps that is why, under certain circumstances, you get areply
> with the physical host IP (192.168.66.50) instead of the VM's IP.

Perhaps, though it seems very odd that IPv6 addresses get resolved
with "-f inet" at all.  E.g. on the FreeBSD VM (gcc303):

$ getaddrinfo -f inet -t stream ::1
getaddrinfo: Non-recoverable failure in name resolution

And based on what I see in my code calling the getaddrinfo() function,
Linux fails with EAI_ADDRFAMILY and macOS with EAI_NONAME.

> Hope this helps,

Somewhat.  I'm still not really understanding why this happens, but
it makes a certain amount of sense that it's the IP for the host the VM
networking is bridged to, and I can easily avoid trying to resolve IPv6
addresses with AF_INET without knowing exactly what is happening here.
Mostly I'm just happier when I actually understand stuff.

Cheers,
    Olly

P.S.  Thanks very much for hosting these VMs.  Having access to a
variety of platforms just an ssh away is really useful for debugging
portability issues.