GitLab

Since GUIDs are 16 bytes and our current interface MACs are only 12 bytes,
I agonized over whether to grow the mac column to 16 bytes and just treat
it as a unique identifier (which is all we use that column for anyway).
However, in the end I just added a new guid column as there were mac columns
in a variety of other tables and it wasn't clear what the relationship was
and what I might break.

So, the newnode MFS will now report back a GUID for interfaces it recognizes
as IB (FreeBSD-specific right now). The boss-side checkin code with stash
that value in new_interfaces (and later interfaces when added). For possible
backward compat, it will also generate a MAC address from that (possibly
Mellanox-specific) so that all entries in the interfaces table will have
a MAC (yes, it should really be the other way around--all interfaces should
always have a guid).

End of story. We don't do anything else with IB right now other than stash
an interface GUID.

hosts, since we are not using a prerouting rule like we do for XEN
containers. Note that I am using the dom0 ules rules on openvz
physical hosts, but might have to split it out if I get any more
special cases like this.

Needed when boss is the tip server (i.e., for IPMI consoles).

Somehow this vaguely feels like a potential flaw. Prior to this,
a misconfigured sshtb command that specified boss would fail.
Now it will succeed and run in the one place it can cause the
most damage.

Also, fix a couple of nits in the console code.

I changed the initscripts to use -e instead of -E since it seemed to
be common between both pkg_info and "pkg info". But alas, that was not
the case.

each container serially with vnodesetup, just do "xl shutdown -a -w"
cause it way faster. We catch that peoperly and save the state of
the VM when it exists this way.

1450. Note that inside a container, we get back regular interface
lines cause all the container sees is a plain interface, but we have
to reduce the MTU.

of gre so we can tell on the client to reduce the MTU. MTU is a pain.

Add code to avoid the respawn loops that Leigh so presciently predicted.
Change the logging path to /usr/testbed/log/tiplogs so we don't have to
keep putting in the /var/log/tiplogs symlink.

This shows up on FBSD 10.x based elabs that use one big FS for
/, /var, /usr since realpath(/var/tmp/...) resolves to /tmp/...
Originally noticed by Jeff Bowles at NMC, but I didn't make the fix
til it bit me on the ass in APT!

In APT, the management network is not the same as was defined in
dhcpd_makeconf. The values that were hardwired there are now the
defaults.

Emulated RTC is in UTC, so remove the /etc/wall_cmos_clock file which
tells FreeBSD that RTC is in local time.

allows the user to create a new disk image in place, instead of having
to create a new profile (Clone).