grub_install: Add LOOP_NO_UDEV for builds in a container

A lot of the SDK instructions seem to assume you're in a "classic" Linux login
session. I use Fedora Atomic Workstation and do all of my development in "dev
containers" (currently docker, in the process of switching to podman).

In my setup the `/dev` setup is separate and won't pick up udev changes.
(Ideally I'd filter out a lot of host devices, that's another issue)

Anyone in a similar situation (which I assume would also include trying
to do the SDK build inside a Docker container on CoreOS) can do:
`env LOOP_NO_UDEV=1 ./build_images.sh`
to have it manually set up the partition mounts.

Loopback mounts with containers in general are ugly since they're not
namespaced.

A whole better solution to this IMO is to use something like
http://libguestfs.org/ which basically spawns a VM, although it doesn't support
grub2. So we'd really have to do instead something like what Fedora does with
using Anaconda. Or the "helper VM" could probably just be an existing CoreOS
qcow2.

Author: cgwalters

build_library/grub_install.sh

@@ -52,6 +52,12 @@ CORE_NAME=
 # fixed up the board root's grub will always be used.
 BOARD_GRUB=0
 
+# This may be useful if you're running the build process inside a container
+# (e.g. docker/podman/etc) distinct from the host; in that case you'll have
+# a distinct /dev normally and won't pick up udev probed devices.  But if you
+# do the probe from the host side, it should show up in the container.
+LOOP_NO_UDEV=${LOOP_NO_UDEV:-0}
+
 case "${FLAGS_target}" in
     i386-pc)
         CORE_MODULES+=( biosdisk serial )
@@ -110,6 +116,21 @@ info "Installing GRUB ${FLAGS_target} in ${FLAGS_disk_image##*/}"
 LOOP_DEV=$(sudo losetup --find --show --partscan "${FLAGS_disk_image}")
 ESP_DIR=$(mktemp --directory)
 
+# Hack available for doing the build in a container; see the LOOP_NO_UDEV
+# comments above.
+if [[ ${LOOP_NO_UDEV} -eq 1 ]]; then
+    echo "LOOP_NO_UDEV enabled, creating device nodes directly"
+    LOOPNUM=$(echo ${LOOP_DEV} | sed -e 's,/dev/loop,,')
+    for d in /sys/block/loop${LOOPNUM}/loop${LOOPNUM}p*; do
+        p=$(cat ${d}/partition)
+        dev=$(cat ${d}/dev)
+        min=${dev##*:}
+        maj=${dev%:*}
+        devpath=/dev/loop${LOOPNUM}p${p}
+        sudo rm -f ${devpath}
+        sudo mknod -m 660 ${devpath} b ${maj} ${min}
+    done
+fi
 # work around slow/buggy udev, make sure the node is there before mounting
 if [[ ! -b "${LOOP_DEV}p1" ]]; then
     # sleep a little just in case udev is ok but just not finished yet
@@ -119,11 +140,12 @@ if [[ ! -b "${LOOP_DEV}p1" ]]; then
         if [[ -b "${LOOP_DEV}p1" ]]; then
             break
         fi
-        warn "looback device node still ${LOOP_DEV}p1 missing, reprobing..."
+        warn "loopback device node still ${LOOP_DEV}p1 missing, reprobing..."
         sudo blockdev --rereadpt ${LOOP_DEV}
         sleep 0.5
     done
     if [[ ! -b "${LOOP_DEV}p1" ]]; then
+        ls -al /dev/loop*
         failboat "${LOOP_DEV}p1 where art thou? udev has forsaken us!"
     fi
 fi