The Wireguard interfaces should not be managed by Glean as they
are usually configured manually (and they are mainly tunnel
interfaces).

They do present themselves with a permanent address however, which
means the only way to ignore them is by using the ignored list.

Change-Id: Ie0c2b56d78620f6ee562b42de6249b1efd37558e

Oslosphinx seems to no longer work and is breaking our docs builds.
Switch to openstackdocstheme instead.

Change-Id: I72600ce84cb04bfa8350b273c030ae3914f7de40

Mailinglists have been updated. Openstack-discuss replaces openstack-dev.

Change-Id: I5c45bd2d45e40bea9466dcfba934325f469f8038

This skip matches the existing service startup skips we have for Red
Hat and Suse on Debuntu systems.

Change-Id: I1926fd122a5833fb295f3c41292c474e5bc8c996

On Fedora 29+, the network.service scripts have been moved into a
separate deprecated package.  NetworkManager is the supported
interface for configuring networks.  We have ignored NM for a long
time, but we are now running out of options.

Luckily we can support this fairly easily with the ifcfg-rh disto
plugin that NM provides.  Using this, the existing
/etc/sysconfig/network-scripts files are used by NM.  The only slight
difference to the config files is that we need to tell NM it's ok to
use it by removing NM_CONTROLLED=no.  We also need to update the
.service file which needs to call glean with the flag to enable NM,
and doesn't need to call ifup.  We have found that local-fs.target
seems to be the best place to run this so it's nice and early.

This is plumbed through from simple-init in DIB
(I4d76e88ce25e5675fd5ef48924acd09915a62a4b) which will call
glean-install with the new --use-nm flag if
DIB_SIMPLE_INIT_NETWORKMANAGER is set.

Since all Centos and Feodora are using NetworkManager by default,
there is an argument that we should just switch to this and not bother
with a flag.  However, since NM brings in new libraries and possibly
other changes to the base system, it's very possible we will either
want to revert the change temporarily or perhaps even run parallel
with and without.  In these cases we can simply flip the variable in
DIB and rebuild rather than having to revert code and do new glean
releases.  Longer term (say, around time we bring up CentOS 8), I see
this becoming the default.

Other platforms have distro plugins for NM to read their "legacy"
configuration files too in the same way as here.  This is left as
future work.

NetworkManager enabled build and boot for Centos and Fedora is tested
in I640838c68a05f3b22683c1e90279725a77678526

Change-Id: I3d379d35e7b000f32c3f6cc197c8aaafebc683fb

We want to default to running all tox environments under python 3, so
set the basepython value in each environment.

We do not want to specify a minor version number, because we do not
want to have to update the file every time we upgrade python.

We do not want to set the override once in testenv, because that
breaks the more specific versions used in default environments like
py35 and py36.

Change-Id: Icb9bbf7bc7b1f9e1e28cb3c9cff200c237e57f80
Closes-Bug:  #1801657

Write dns info to networkd so it can be consumed by resolved if resolved
is enabled.

Change-Id: Ia7b55e9a538cc6f060bc08da85123d7bb3a4f73e

These functions take a 'distro' argument which comes from args.distro.
In future changes, we may want to send through more arguments from the
command-line to these functions, so pass 'args' directly and
dereference the distro when required.

Change-Id: I0f3d736872492fb8148520c195a977919b9e5786

The argv patching was not actually working; it has something to do
with an interaction between the testscenarios based load_tests we use
and our mocking of argv with

 @mock.patch.object(sys, 'argv', ['./glean', '--hostname'])

What appears to happen is that tests in the same scenario groups get
the *same* reference to the sysv.argument object (i.e. it's only
created once, not once for each call).  Thus each test .append()s to
this same list, and thus we end up appending more and more arguments
to the list.  The sys.argv ends up looking like:

  stdout: {{{['./glean', '--hostname', '--distro=ubuntu', '--distro=ubuntu',
    '--distro=debian', '--distro=redhat', '--distro=redhat', '--distro=centos',
    '--distro=centos', '--distro=gentoo', '--distro=gentoo', '--distro=opensuse', ...

This can't be, Ian, how does it even work?  Well we're lucky in that
for all the existing test-cases, the final arguments appended have
overwritten the prior ones sufficiently that the test-cases work.  If
you start changing the arguments, things go wrong.

To simplify the whole thing, make main() take an argv argument that is
passed to argparse.  For normal use this comes from sys.argv, and in
unit tests we construct our own argv list for each test.

Change-Id: I090164fdd7a46d0c7a29eb6c18f80dc84cfd8469

Previously we were checking Gentoo netifrc files (/etc/conf.d/net.{iname})
and not networkd files <F12>(/etc/systemd/network/{iname}.(netdev|network)).

Change to check for the networkd files.  This is not perfect as the file
name does not directly tie to the device name, but it's a start (it exists
as a Nmme or Match value within the file itself).

Change-Id: I6130102005ab61dfc25cd77c6e8abea38bccf201

Change-Id: I1363ee1db18cc77ec7886ee9141a65ff76de38e9

Some clouds provide incorrect network_data on the config drive because
they are doing an end round neutron to configure networking in their
cloud. Neutron has incorrect data, that ends up in the config drive and
we configure the interfaces improperly.

Address this by checking for a metadata key called
glean_ignore_interfaces. If present then we ignore any interface
configuration information in the config drive and instead fall back to
dhcp.

Change-Id: I20bc7a124891715fa2ead9e4a72956c80d63563d

There are bond details that apply to all the interface types. Rather
than set them separately for each interface type just set them once.

Change-Id: Id1abb85ab32a12adea92e337d00e2a03e877021a

Handle each interface type in its own if block and rely on the use of
the if conditional to control flow rather than continues. This allows us
to write the results once and is easier to read.

Change-Id: I9aec63e7d85eac3704c2c6a5566b669ecc5bbf00

We were setting an identical header for all of the different interface
types. Manage this in a single location which simplifies the control
flow through the different interface types.

Change-Id: Ib0300d46896cf95c83223f1af3e2fe890698af2b

We had a special function for checking if a debian interface config file
exists but also manually constructed the path elsewhere to check if it
exists. Consolidate and drop the function since we already need the full
path in other places and once you have the full path os.path.exists is
the function you need.

Change-Id: I7a800b09699e0e86ee9df5988823b48910b57051

This adds some DRYness to the debian interface config writing around
bond modes. Makes code a bit easier to follow.

Part of a larger refactoring to make this code easier to maintain.

Change-Id: I014e16e4078d4b73cb00a8706eca096e5aad6e16

Since this is intended to be used as a context manager, it should
close the file once we're done with it.  Also, I can't see why we
would want to reset the selinux state if an exception was raised -- it
seems there may be no file to actually do this on.  Add it to the
regular exit path.

Change-Id: I7e7dba66a1c446fa82ca3facde19f89c40e9e935

We need to check the /etc/sysconfig/network/ifcfg-* files for bridges
and vlans similar to regular interfaces. Moreover, the existing code for
checking whether the host is a SUSE distribution did not work for newer
releases (openSUSE Leap 15 is being reported as 'opensuse-leap' and
Tumbleweed is being reported as 'opensuse-tumbleweed') so we need to
refactor it a bit to actually check whether the string 'suse' exists in
the distro variable to simplify things.

Finally, the systemd service was missing the check for the SUSE network
files so we add that as well. systemd should apply a logical AND in
these conditionals so it should work fine on both Red Hat and SUSE
distros.

Change-Id: Ic5c6745dbe0077089ecaa1dd8f9b4949ac80efae

Initially I've only added this for Gentoo (will only be turned on within
Gentoo systemd installs).  Has full ipv6 support too.

Change-Id: I34eeb96eb90850d6a4407559e8a3f9aa74483a95

It seems platform.dist and platform.linux_distribution are broken with
newer operating systems. Specifically /etc/os-release seems to be the
new hotness. EG: opensuse-tumbleweed and fedora-27.

Vendor the distro (1.2.0) package, as not to have external
dependencies installed.

Change-Id: Ib9580aa7fe84a20e44d32248d7b4716456684dbc
Depends-On: https://review.openstack.org/545183/


Signed-off-by: Paul Belanger <pabelanger@redhat.com>