Fix cloud-init service-startup dep. loop

In GCP, there are google services that must start on boot but we also
rely on the more generic cloud-init services to setup the VM.
Specifically, this is necessary to attend to rootfs resizing properly
before testing.  However, there is a dependency loop created when both
are installed, that will result in messages like:

...network-online.target: Found ordering cycle on \
    cloud-init.service/start
...network-online.target: Found dependency on \
    google-network-daemon.service/start
...network-online.target: Found dependency on \
    google-instance-setup.service/start
...network-online.target: Found dependency on \
    network-online.target/start
...network-online.target: Job cloud-init.service/start \
    deleted to break ordering cycle starting with \
    network-online.target/start

Previously, this was addressed by depositing a customized
cloud-init.service file for Fedora VMs.  However, at some point in the
past this step was dropped.  Restore it, but with a few sed-commands
instead of hard-coding a new service file in this repo.

Also, add a header-comment to this and another custom service file to
indicate where they came from.

Signed-off-by: Chris Evich <[email protected]>

base_images/fedora_base-setup.sh

@@ -41,6 +41,20 @@ install_automation_tooling
 if ! ((CONTAINER)); then
     custom_cloud_init
 
+    # Be kind to humans, indicate where generated files came from
+    sourcemsg="### File generated during VM Image build by $(basename $SCRIPT_FILEPATH)"
+
+    if ((OS_RELEASE_VER<35)); then
+        # The packaged cloud-init.service unit has a dependency loop
+        # vs google-network-daemon.service.  Fix this by forcing cloud-init
+        # to run after.
+        CLOUD_SERVICE_PATH="systemd/system/cloud-init.service"
+        echo "$sourcemsg" > /etc/$CLOUD_SERVICE_PATH
+        sed -r -e \
+            's/Before=network-online.target/After=google-network-daemon.service' \
+            /lib/$CLOUD_SERVICE_PATH >> /etc/$CLOUD_SERVICE_PATH
+    fi
+
     # ref: https://cloud.google.com/compute/docs/startupscript
     # The mechanism used by Cirrus-CI to execute tasks on the system is through an
     # "agent" process launched as a GCP startup-script (from the metadata service).
@@ -50,9 +64,10 @@ if ! ((CONTAINER)); then
     # run with the proper contexts.
     METADATA_SERVICE_CTX=unconfined_u:unconfined_r:unconfined_t:s0
     METADATA_SERVICE_PATH=systemd/system/google-startup-scripts.service
+    echo "$sourcemsg" > /etc/systemd/system/cloud-init.service
     sed -r -e \
         "s/^Type=oneshot/Type=oneshot\nSELinuxContext=$METADATA_SERVICE_CTX/" \
-        /lib/$METADATA_SERVICE_PATH > /etc/$METADATA_SERVICE_PATH
+        /lib/$METADATA_SERVICE_PATH >> /etc/$METADATA_SERVICE_PATH
 fi
 
 if [[ "$OS_RELEASE_ID" == "fedora" ]] && ((OS_RELEASE_VER>=33)); then