Enabled user session dbus in base image to support podman rootless mode.
Extension images can now be created from multiple packages by seperating
them with a comma. The podman sysext includes app-containers/podman and
net-misc/passt.
It can be enabled by adding podman to /etc/flatcar/enabled-sysext.conf.
Potential TODO: gpgme had to be added as BDEPEND to podman ebuild.
The if-up-down to trigger the DHCP request causes problems. It's better
to directly ask systemd-networkd to issue the request. It seems that
one needs to use "reconfigure" instead of "forcerenew", so I went with
only that instead of somehow trying to see if "forcerenew" has an
effect.
This reverts commit 9556c7f94b
because waagent looks for internal hostname changes and wants to
propagate them to the Azure VM properties by issuing a new DHCP request.
As Ignition supports KubeVirt, add a custom oem for it and also the
required parts to be able to build an image in .qcow2 format that
is already using internal .qcow2 gzip compression.
Fixes: https://github.com/flatcar/Flatcar/issues/1358
Signed-off-by: Adrian Vladu <avladu@cloudbasesolutions.com>
This is a new package that is being developed to handle providing
symlinks for nvme disks (os,data,temporary) on newer Azure instances.
It needs to be part of the OS, and not oem-azure, because it carries
udev rules.
Signed-off-by: Jeremi Piotrowski <jpiotrowski@microsoft.com>
Flatcar prepares /etc/hostname from the initrd through afterburn. After
switching root, systemd-networkd fetches a dhcp lease with the correct
hostname already. This publishes the hostname to the vnet DNS server as
well. When WALinuxAgent starts, it tries to do the same steps: configure
the hostname, bounce the link to force dhcp lease renewal. This has
caused issues in the past with multi-nic configurations but also with
networked services that are trying to use the network (etcd/flanneld).
The link bouncing by WALinuxAgent is not necessary because of Flatcars
boot design, so return without bouncing the link. Tested that DNS from
other VMs in the same vnet works.
Signed-off-by: Jeremi Piotrowski <jpiotrowski@microsoft.com>
When CoreosCommonUtil was factored out, we missed updating the class
name in a call to super(). This results in an error when executing
`/usr/sbin/waagent -force -deprovision+user`. Fix the class name.
Create a compatibility symlink at the old config file location
(/usr/share/oem/waagent.conf) to handle the case of enabling
auto-updates on the agent. The upstream version of the agent does not
have our downstream patch so doesn't know about the updated config file
location. We should upstream our changes.
Signed-off-by: Jeremi Piotrowski <jpiotrowski@microsoft.com>
In case of sysext of zfs-kmod, pkg_setup() calls linux-info_get_any_version()
of eclass/linux-info.eclass with MERGE_TYPE=binary and
LINUX_INFO_BINARY_RESET=0. That results in KV_FULL and KV_OUT_DIR being
unset, which is not what we want. So set LINUX_INFO_BINARY_RESET=1 to
avoid such side effect.
Without setting them, build_sysext complains like below, i.e. it tries to find
System.map of a wrong Kernel version 6.6.21, when it should find a newer
version like 6.6.23. That results in subsequent zfs CI tests failing due to
missing files like System.map and module deps.
```
System.map for kernel 6.6.21-flatcar was not found, may be due to the
built kernel sources no longer being available and lacking the fallback:
```
Suggested-by: Jeremi Piotrowski <jpiotrowski@microsoft.com>
