#!/bin/bash

# Intended to be called from the root.d cloud-image script as follows:
# $TMP_HOOKS_PATH/bin/extract-image $BASE_IMAGE_FILE $BASE_IMAGE_TAR $IMAGE_LOCATION $CACHED_IMAGE

if [ ${DIB_DEBUG_TRACE:-0} -gt 0 ]; then
    set -x
fi
set -eu
set -o pipefail

BASE_IMAGE_FILE=$1
BASE_IMAGE_TAR=$2
IMAGE_LOCATION=$3
CACHED_IMAGE=$4

CACHED_TAR=$DIB_IMAGE_CACHE/$BASE_IMAGE_TAR
DIB_LOCAL_IMAGE=${DIB_LOCAL_IMAGE:-""}
TAR_LOCK=$CACHED_TAR.lock

# GPT GUIDs of interest.
# See https://en.wikipedia.org/wiki/GUID_Partition_Table#Partition_type_GUIDs
# also https://systemd.io/BOOT_LOADER_SPECIFICATION/
GUID_EFI="c12a7328-f81f-11d2-ba4b-00a0c93ec93b"
GUID_LINUX_BOOT="bc13c2ff-59e6-4262-a352-b275fd6f7172"

function extract_image() {
    if [ -n "$DIB_OFFLINE" -a -f "$CACHED_TAR" ] ; then
        echo "Not checking freshness of cached $CACHED_TAR."
    else
        if [ -z "$DIB_LOCAL_IMAGE" ]; then
            echo "Fetching Base Image"

            # There seems to be some bad Fedora mirrors returning http 404's for the cloud image.
            # If the image fails to download due to a 404 we retry once.
            set +e
            $TMP_HOOKS_PATH/bin/cache-url $IMAGE_LOCATION $CACHED_IMAGE
            RV=$?
            set -e

            if [ "$RV" == "44" ] ; then
                $TMP_HOOKS_PATH/bin/cache-url $IMAGE_LOCATION $CACHED_IMAGE
            elif [ "$RV" != "0" ] ; then
                exit 1
            fi
        fi

        if [ ! -f $CACHED_TAR -o \
            $CACHED_IMAGE -nt $CACHED_TAR ] ; then
            echo "Repacking base image as tarball."

            WORKING=$(mktemp --tmpdir=${TMP_DIR:-/tmp} -d)
            EACTION="rm -r $WORKING"
            trap "$EACTION" EXIT
            echo "Working in $WORKING"

            RAW_FILE=$(mktemp --tmpdir=$WORKING XXXXXX.raw)

            if [ "${CACHED_IMAGE: -3}" == ".xz" ] ; then
                QCOW2_FILE=$(mktemp --tmpdir=$WORKING XXXXXX.qcow2)
                # This leaves the old image in place so cache-url wont get it again
                unxz --stdout $CACHED_IMAGE > $QCOW2_FILE
                CACHED_IMAGE=$QCOW2_FILE
            fi

            qemu-img convert -f qcow2 -O raw $CACHED_IMAGE $RAW_FILE

            # kpartx fails if no /dev/loop* exists, "losetup -f" prints first unused
            # loop device and creates it if it doesn't exist
            LOOPDEV_BASE=$(basename $(sudo losetup -f))

            # add partition mappings
            sudo kpartx -av $RAW_FILE

            # If running inside Docker, make our nodes manually, because udev will not be working.
            if [ -f /.dockerenv ]; then
                sudo dmsetup --noudevsync mknodes
            fi
            if ! timeout 5 sh -c "while ! ls /dev/mapper/${LOOPDEV_BASE}p* ; do sleep 1; done"; then
                echo "Error: Could not find any ${LOOPDEV_BASE} devices"
                exit 1
            fi

            EACTION="sudo kpartx -d $RAW_FILE ; $EACTION"
            trap "$EACTION" EXIT

            ROOT_LOOPDEV=""
            BOOT_LOOPDEV=""
            EFI_LOOPDEV=""

            LOOPDEVS=$(ls /dev/mapper/${LOOPDEV_BASE}p* | sort -r)
            LOOPDEV_COUNT=$(echo $LOOPDEVS | wc -w)
            if [ $LOOPDEV_COUNT == "1" ]; then
                # if there is one partition device, assume it is the root device
                ROOT_LOOPDEV=${LOOPDEVS}
                LOOPDEVS=""
            fi

            for LOOPDEV in ${LOOPDEVS}; do
                fstype=$(lsblk --all --nodeps --noheadings --output FSTYPE $LOOPDEV)
                label=$(lsblk --all --nodeps --noheadings --output LABEL $LOOPDEV)
                part_type=$(lsblk --all --nodeps --noheadings --output PARTTYPE $LOOPDEV)

                if [ -z "${fstype}" ]; then
                    # Ignore block device with no filesystem type
                    continue
                fi

                # look for EFI partition to mount at /boot/efi either by GUID or
                # label convention
                if [ -z "$EFI_LOOPDEV" ]; then
                    if [[ ${part_type} == ${GUID_EFI} ]]; then
                        EFI_LOOPDEV=$LOOPDEV
                        continue
                    fi
                fi

                # look for EFI partition to mount at /boot/efi either by GUID or
                # label convention.
                if [ -z "$BOOT_LOOPDEV" ]; then
                    if [[ ${part_type} == ${GUID_LINUX_BOOT} || ${label} == "boot" ]]; then
                        BOOT_LOOPDEV=$LOOPDEV
                        continue
                    fi
                fi

                if [ -z "$ROOT_LOOPDEV" ]; then
                    ROOT_LOOPDEV=$LOOPDEV
                    continue
                fi
            done

            mkdir $WORKING/mnt
            ROOT_FSTYPE=$(sudo blkid -o value -s TYPE $ROOT_LOOPDEV)
            if [ "xfs" = "$ROOT_FSTYPE" ]; then
                # mount xfs with nouuid, just in case that uuid is already mounted
                # use ro to avoid/workaround xfs uuid issues on older
                # kernels with newer rhel images which seem to set
                # flags to generate unique uuid's:
                #  xfs superblock has incompatible features (0x4)
                # we don't need to worry about this, we just want the data
                MOUNTOPTS="-o nouuid,ro"
            elif [ "btrfs" = "$ROOT_FSTYPE" ]; then
                # Fedora has a btrfs filesystem with a subvolume called root.
                # For now assume there will be a 'root' subvolume, but in the
                # future the subvolume layout may need to be discovered for different
                # images
                MOUNTOPTS="-o subvol=root"
            else
                MOUNTOPTS=""
            fi

            sudo mount $MOUNTOPTS $ROOT_LOOPDEV $WORKING/mnt
            EACTION="sudo umount -f $WORKING/mnt ; $EACTION"
            trap "$EACTION" EXIT

            if [ ! -z "$BOOT_LOOPDEV" ]; then
                # mount to /boot
                BOOT_FSTYPE=$(sudo blkid -o value -s TYPE $ROOT_LOOPDEV)
                if [ "xfs" = "$BOOT_FSTYPE" ]; then
                    BOOT_MOUNTOPTS="-o nouuid,ro"
                    # Similar to root filesystem, if the boot filesystem
                    # is XFS and the base OS is the same as the image being
                    # rebuilt, we need to pass "nouuid" to bypass UUID safety
                    # checks and successfully mounts so we can extract the
                    # contents.
                else
                    BOOT_MOUNTOPTS=""
                fi
                sudo mount $BOOT_MOUNTOPTS $BOOT_LOOPDEV $WORKING/mnt/boot
                EACTION="sudo umount -f $BOOT_LOOPDEV ; $EACTION"
                trap "$EACTION" EXIT
            fi
            if [ ! -z "$EFI_LOOPDEV" ]; then
                # mount to /boot/efi
                sudo mount $EFI_LOOPDEV $WORKING/mnt/boot/efi
                EACTION="sudo umount -f $EFI_LOOPDEV ; $EACTION"
                trap "$EACTION" EXIT
            fi

            # find out if chroot tar has full xattr support
            if [ 0 == `sudo chroot $WORKING/mnt bin/tar --help | grep -c xattrs-exclude` ]; then
                TAROPTS="--no-xattrs"
            else
                TAROPTS="--xattrs --xattrs-include=* --xattrs-exclude=security.selinux"
            fi
            # Chroot in so that we get the correct uid/gid
            sudo chroot $WORKING/mnt bin/tar $TAROPTS -cz . > $WORKING/tmp.tar
            mv $WORKING/tmp.tar $CACHED_TAR
        else
            echo "Using cached tar from $CACHED_TAR"
        fi
    fi

    # Extract the base image (use --numeric-owner to avoid UID/GID mismatch between
    # image tarball and host OS e.g. when building Fedora image on an openSUSE host)
    # Include all xattrs except selinux because the selinux ones cause issues in our
    # chroot environment, and we restore all of those at the end of the build anyway.
    echo "Extracting base root image from $CACHED_TAR"
    sudo tar -C $TARGET_ROOT --numeric-owner --xattrs --xattrs-include='*' --xattrs-exclude='security.selinux' -xzf $CACHED_TAR
}

(
    echo "Getting $TAR_LOCK: $(date)"
    # Wait up to 20 minutes for another process to download
    if ! flock -w 1200 9 ; then
        echo "Did not get $TAR_LOCK: $(date)"
        exit 1
    fi
    extract_image
) 9> $TAR_LOCK