#!/bin/bash
# This script can be used to start, stop, create, and delete containers.
# It is what confd use, with the Finit container@.conf template, to set
# up, run, and delete containers.
#
# NOTE: when creating/deleting containers, remember 'initctl reload' to
#       activate the changes!  In confd this is already handled.
#
DOWNLOADS=/var/lib/containers/oci
BUILTIN=/lib/oci
TMPDIR=/var/tmp
container=$0
checksum=""
extracted=
timeout=30
dir=""
all=""
env=""
port=""
force=

log()
{
    logger -I $PPID -t container -p local1.notice -- "$*"
}

err()
{
    rc=$1; shift
    logger -I $PPID -t container -p local1.err -- "Error: $*"

    if [ -n "$extracted" ]; then
	if [ -d "$TMPDIR/$dir" ]; then
	    log "Cleaning up extracted $dir"
	    rm -rf "$dir"
	fi
    fi

    [ "$rc" -eq 0 ] || exit "$rc"
}

pidfn()
{
    echo "/run/containers/${1}.pid"
}

check()
{
    file=$1

    if [ -z "$checksum" ]; then
	log "no checksum to verify $file against, continuing."
	return 0
    fi

    if echo "${checksum}  ${file}" | "$cmdsum" -c -s; then
        log "$file checksum verified OK."
	return 0
    fi

    got=$("$cmdsum" "${file}" | awk '{print $1}')
    log "$file checksum mismatch, got $got, expected $checksum, removing file."
    rm -f "$file"

    return 1
}

# Fetch an OCI image over ftp/http/https.  Use wget for FTP, which curl
# empirically does not work well with.  Log progress+ & error to syslog.
fetch()
{
    url=$1
    file=$(basename "$url")
    dst="$DOWNLOADS/$file"

    cd "$DOWNLOADS" || return
    if [ -e "$file" ]; then
	log "$file already available."
	if check "$file"; then
	    echo "$dst"
	    return 0
	fi
    fi

    log "Fetching $url"

    if echo "$url" | grep -qE "^ftp://"; then
        cmd="wget -q $url"
    elif echo "$url" | grep -qE "^https?://"; then
        cmd="curl $creds -sSL --fail -o \"$file\" $url"
    else
        log "Unsupported URL scheme: $url"
        return 1
    fi

    if out=$(eval "$cmd" 2>&1); then
	log "$file downloaded successfully."
	if check "$file"; then
	    echo "$dst"
	    return 0
	fi
    fi

    # log error message from backend
    while IFS= read -r line; do
	log "$line"
    done <<EOF
$out
EOF
    return 1
}

# Unpacks a given oci-archive.tar[.gz] in the current directory.  Sanity
# checks, at least one index.json in the top-level dir of the archive.
# If there are more index files, this function does not handle them.
unpack_archive()
{
    uri=$1
    tag=$2
    img=$(basename "$uri")

    # Supported transports for load and create
    case "$uri" in
	oci:*)			# Unpacked OCI image
	    file=${uri#oci:}
	    ;;
	oci-archive:*)		# Packed OCI image, .tar or .tar.gz format
	    file=${uri#oci-archive:}
	    ;;
	ftp://* | http://* | https://*)
            if ! file=$(fetch "$uri"); then
		return 1
	    fi
            ;;
	*)			# docker://*, docker-archive:*, or URL
	    if podman image exists "$img"; then
		echo "$img"
		return 0
	    fi
	    # XXX: use --retry=0 with Podman 5.0 or later.
	    if ! id=$(podman pull --quiet "$uri"); then
		log "Failed pulling $uri"
		return 1
	    fi
	    # Echo image tag to caller
	    podman images --filter id="$id" --format "{{.Repository}}:{{.Tag}}"
	    return 0
	    ;;
    esac

    if [ ! -e "$file" ]; then
	if [ -e "$DOWNLOADS/$file" ]; then
	    file="$DOWNLOADS/$file"
	elif [ -e "$BUILTIN/$file" ]; then
	    file="$BUILTIN/$file"
	else
	    err 1 "cannot find OCI archive $file in URI $uri"
	fi
    fi

    if [ -d "$file" ]; then
	index=$(find "$file" -name index.json)
	if [ -z "$index" ]; then
	    err 1 "cannot find index.json in OCI image $file"
	fi
    else
	cd "$TMPDIR" || err 0 "failed cd $TMPDIR, wiill use $(pwd) for OCI archive extraction."

	index=$(tar tf "$file" |grep index.json)
	if [ -z "$index" ]; then
	    err 1 "invalid OCI archive, cannot find index.json in $file"
	fi

	[ -n "$quiet" ] || log "Extracting OCI archive $file ..."
        tar xf "$file"  || err 1 "failed unpacking $file in $(pwd)"
	extracted=true
    fi

    dir=$(dirname "$index")
    if echo "$dir" | grep -q ":"; then
        if [ -z "$tag" ]; then
            tag="$dir"
        fi
        sanitized_dir=$(echo "$dir" | cut -d':' -f1)
        mv "$dir" "$sanitized_dir" || err 1 "failed renaming $dir to $sanitized_dir"
        dir="$sanitized_dir"
    fi

    [ -n "$quiet" ] || log "Loading OCI image $dir ..."
    podman load -qi "$dir"        >/dev/null

    # Clean up after ourselves
    if [ -n "$extracted" ]; then
	log "Cleaning up extracted $dir"
	rm -rf "$dir"
    fi

    # Retag image from podman default $dir:latest
    if [ -n "$tag" ]; then
	podman tag "$dir" "$tag" >/dev/null
	podman rmi "$dir"        >/dev/null
    else
	tag=$dir
    fi

    echo "$tag"
}

running()
{
    status=$(podman inspect -f '{{.State.Status}}' "$1" 2>/dev/null)
    [ "$status" = "running" ] && return 0
    return 1
}

# shellcheck disable=SC2086
create()
{
    name=$1
    image=$2
    shift 2

    if [ -z "$name" ] || [ -z "$image" ]; then
	echo "Usage:"
	echo "  container create NAME IMAGE"
	exit 1
    fi

    # Unpack and load docker-archive/oci/oci-archive, returning image
    # name, or return docker:// URL for download.
    if ! image=$(unpack_archive "$image"); then
	exit 1
    fi

    if [ -z "$logging" ]; then
	logging="--log-driver syslog"
    fi

    # When we get here we've already fetched, or pulled, the image
    args="$args --read-only --replace --quiet --cgroup-parent=containers $caps"
    args="$args --restart=$restart --systemd=false --tz=local $privileged"
    args="$args $vol $mount $hostname $entrypoint $env $port $logging"
    pidfile=/run/container:${name}.pid

    [ -n "$quiet" ] || log "---------------------------------------"
    [ -n "$quiet" ] || log "Got name: $name image: $image"
    [ -n "$quiet" ] || log "Got networks: $network"

    if [ -n "$network" ]; then
	for net in $network; do
	    args="$args --net=$net"
	done

	for srv in $dns; do
	    args="$args --dns=$srv"
	done

	for domain in $search; do
	    args="$args --dns-search=$domain"
	done
    else
	args="$args --network=none"
    fi

    # shellcheck disable=SC2048
    log "podman create --name $name --conmon-pidfile=$pidfile $args $image $*"
    if podman create --name "$name" --conmon-pidfile="$pidfile" $args "$image" $*; then
	[ -n "$quiet"  ] || log "Successfully created container $name from $image"
	[ -n "$manual" ] || start "$name"

	# Should already be enabled by confd (this is for manual use)
	initctl -bnq enable "container@${name}.conf"
	exit 0
    fi

    err 1 "failed creating container $name, please check the configuration."
}

delete()
{
    name=$1

    if [ -z "$name" ]; then
	echo "Usage:"
	echo "  container delete NAME"
	exit 1
    fi

    # Should already be stopped, but if not ...
    log "$name: should already be stopped, double checking ..."
    container stop "$name" >/dev/null

    while running "$name"; do
	log "$name: still running, waiting for it to stop ..."
	_=$((timeout -= 1))
	if [ $timeout -le 0 ]; then
	    err 1 "timed out waiting for container $1 to stop before deleting it."
	fi
	sleep 1
    done

    log "$name: calling podman rm -vif ..."
    podman rm -vif "$name" >/dev/null 2>&1
    [ -n "$quiet" ] || log "Container $name has been removed."
}

waitfor()
{
    while [ ! -f "$1" ]; do
	_=$((timeout -= 1))
	if [ $timeout -le 0 ]; then
	    err 1 "timed out waiting for $1, aborting!"
	fi
	sleep 1;
    done
}

start()
{
    name=$1

    if running "$name"; then
	[ -n "$quiet" ] || echo "$name: already running."
	return
    fi

    initctl start container:$name
    # Real work is done by wrap() courtesy of finit sysv emulation
}

stop()
{
    name=$1

    if ! running "$name"; then
	[ -n "$quiet" ] || echo "$name: not running."
	return
    fi

    initctl stop container:$name
    # Real work is done by wrap() courtesy of finit sysv emulation
}

wrap()
{
    name=$1
    cmd=$2
    pidfile=$(pidfn "$name")

    # Containers have three phases: setup, running, and teardown.

    # The setup phase may run forever in the background trying to fetch
    # the image.  It saves its PID in /run/containers/${name}.pid
    if [ "$cmd" = "stop" ] && [ -f "$pidfile" ]; then
        pid=$(cat "$pidfile")

	# Check if setup is still running ...
	if kill -0 "$pid" 2>/dev/null; then
            kill "$pid"
            wait "$pid" 2>/dev/null
        fi

        rm -f "$pidfile"
        return 0
    fi

    # Skip "echo $name" from podman start in log
    podman "$cmd" "$name" >/dev/null
}

# Removes network $1 from all containers
netwrm()
{
    net=$1

    for c in $(podman ps $all --format "{{.Names}}"); do
	for n in $(podman inspect "$c" |jq -r '.[].NetworkSettings.Networks | keys[]'); do
	    if [ "$n" = "$net" ]; then
		podman network disconnect $force "$n" "$c" >/dev/null
	    fi
	done
    done
}

# Schedule restart of (any) container using network $1 to activate network changes
netrestart()
{
    net=$1

    for c in $(podman ps $all --format "{{.Names}}"); do
	for n in $(podman inspect "$c" |jq -r '.[].NetworkSettings.Networks | keys[]'); do
	    if [ "$n" = "$net" ]; then
		initctl -nbq touch "container@$c"
	    fi
	done
    done
}

cleanup()
{
    pidfile=$(pidfn "$name")

    log "Received signal, exiting."
    if [ -n "$name" ] && [ -f "$pidfile" ]; then
	log "$name: in setup phase, removing $pidfile ..."
	rm -f "$pidfile"
    fi

    exit 1
}

usage()
{
	cat <<EOF
usage:
  container [opt] cmd [arg]

options:
  -a, --all                Show all, of something
      --dns NAMESERVER     Set nameserver(s) when creating a container
      --dns-search LIST    Set host lookup search list when creating container
      --cap-add CAP        Add capability to unprivileged container
      --cap-drop CAP       Drop capability, for privileged containter
      --checksum TYPE:SUM  Use md5/sha256/sha512 to verify ftp/http/https archives
  -c, --creds USR[:PWD]    Credentials to pass to curl -u for remote ops
  -d, --detach             Detach a container started with 'run IMG [CMD]'
  -e, --env FILE           Environment variables when creating container
      --entrypoint         Disable container image's ENTRYPOINT, run cmd + arg
  -f, --force              Force operation, e.g. remove
  -h, --help               Show this help text
      --hostname NAME      Set hostname when creating container
      --net NETWORK        Network interface(s) when creating or finding container
  -l, --log-driver DRV     Log driver to use
      --log-opt OPT        Logging options to log driver
      --log-path PATH      Path for k8s-file log pipe
  -m, --mount HOST:DEST    Bind mount a read-only file inside a container
      --manual             Do not start container automatically after creation
  -n, --name NAME          Alternative way of supplying name to start/stop/restart
      --privileged         Give container extended privileges
  -p, --publish PORT       Publish ports when creating container
                           Syntax: [[ip:][hostPort]:]containerPort[/protocol]
  -q, --quiet              Quiet operation, called from confd
  -r, --restart POLICY     One of "no", "always", or "on-failure:NUM"
  -s, --simple             Show output in simplified format
  -t, --timeout SEC        Set timeout for delete/restart commands, default: 20
  -v, --volume NAME:PATH   Create named volume mounted inside container on PATH

commands:
  create   NAME IMAGE NET  Create container NAME using IMAGE with networks NET
  delete  [network] NAME   Remove container NAME or network NAME from all containers
  exec     NAME CMD        Run a command inside a container
  flush                    Clean up lingering containers and associated anonymous volumes
  find    [ifname PID]     Find PID of container where '--net IFNAME' currently lives
                           or, find the name of our IFNAME inside the container @PID
  help                     Show this help text
  list    [image | oci]    List names (only) of containers, images, or OCI archives
  load    [NAME | URL] NM  Load OCI tarball fileNAME or URL to image NM
  locate                   Find container that currently owns '--net IFNAME'
  remove   IMAGE           Remove an (unused) container image
  restart [network] NAME   Restart a (crashed) container or container(s) using network
  run      NAME [CMD]      Run a container interactively, with an optional command
  save     IMAGE FILE      Save a container image to an OCI tarball FILE[.tar.gz]
  setup    NAME            Create and set up container as a Finit task
  shell   [CMD]            Start a shell, or run CMD, inside a container
  show    [image | volume] Show containers, images, or volumes
  stat                     Show continuous stats about containers (Ctrl-C aborts)
  start   [NAME]           Start a container, see -n
  stop    [NAME]           Stop a container, see -n
  upgrade  NAME            Upgrade a running container (stop, pull, restart)
  volume  [prune]          Prune unused volumes
EOF
}

while [ "$1" != "" ]; do
    case $1 in
	-a | --all)
	    all="-a"
	    ;;
	--cap-add)
	    shift
	    caps="$caps --cap-add=$1"
	    ;;
	--cap-drop)
	    shift
	    caps="$caps --cap-drop=$1"
	    ;;
	--checksum)
	    shift
	    type="${1%%:*}"
	    checksum="${1#*:}"
	    case "$type" in
		md5)
		    cmdsum=md5sum
		    ;;
		sha256)
		    cmdsum=sha256sum
		    ;;
		sha512)
		    cmdsum=sha512sum
		    ;;
		*)
		    err 1 "Unsupported checksum type: $type"
		    ;;
	    esac
	    ;;
	-c | --creds)
	    shift
	    creds="-u $1"
	    ;;
	-d | --detach)
	    detach="-d"
	    ;;
	--dns)
	    shift
	    dns="$dns $1"
	    ;;
	--dns-search)
	    shift
	    search="$search $1"
	    ;;
	-e | --env)
	    shift
	    env="$env --env-file=$1"
	    ;;
	--entrypoint)
	    entrypoint="--entrypoint=\"\""
	    ;;
	-f | --force)
	    force="-f"
	    ;;
	-h | --help)
	    usage
	    exit 0
	    ;;
	--hostname)
	    shift
	    hostname="--hostname $1"
	    ;;
	-l | --log-driver)
	    shift
	    logging=" --log-driver=$1"
	    ;;
	--log-opt)
	    shift
	    logging="$logging --log-opt $1"
	    ;;
	--log-path)
	    shift
	    logging="$logging --log-opt path=$1"
	    ;;
	-m | --mount)
	    shift
	    mount="$mount --mount=$1"
	    ;;
	--manual)
	    manual=true
	    ;;
	-n | --name)
	    shift
	    name="$1"
	    ;;
	--net)
	    shift
	    if [ -n "$network" ]; then
		network="$network $1"
	    else
		network=$1
	    fi
	    ;;
	--privileged)
	    privileged="--privileged=true"
	    ;;
	-p | --publish)
	    shift
	    port="$port -p $1"
	    ;;
	-q | --quiet)
	    quiet="-q"
	    ;;
	-r | --restart)
	    shift
	    restart=$1
	    ;;
	-s | --simple)
	    simple=true
	    ;;
	-t | --timeout)
	    shift
	    timeout=$1
	    ;;
	-v | --volume)
	    shift
	    vol="$vol -v $1"
	    ;;
	*)
	    break
	    ;;
    esac
    shift
done

cmd=$1
if [ -n "$cmd" ]; then
    shift
fi

trap cleanup INT HUP TERM

case $cmd in
    # Does not work atm., cannot attach to TTY because
    # we monitor 'podman start -ai foo' with Finit.
    # attach)
    # 	podman attach "$1"
    # 	;;
    create)
	[ -n "$quiet" ] || log "Got create args: $*"
	create "$@"
	;;
    delete)
	cmd=$1
	[ -n "$name" ] || name=$2
	if [ "$cmd" = "network" ] && [ -n "$name" ]; then
	    netwrm "$name"
	else
	    [ -n "$name" ] || name=$1
	    delete "$name"
	fi
	;;
    exec)
	if [ -z "$name" ]; then
	    name="$1"
	    shift
	fi
	podman exec -i "$name" "$@"
	;;
    flush)
	echo "Cleaning up any lingering containers";
	podman rm -av $force
	;;
    find)
	cmd=$1
	pid=$2
	if [ "$cmd" = "ifname" ] && [ -n "$pid" ]; then
	    nsenter -t "$pid" -n ip -d -j link | \
		jq --arg ifname "$network" -r '.[] | select(.ifalias==$ifname) | .ifname'
	else
	    containers=$(podman ps $all --format "{{.Names}}")
	    for c in $containers; do
		json=$(podman inspect "$c")
		nets=$(echo "$json" |jq -r '.[].NetworkSettings.Networks | keys[]' 2>/dev/null)
		for n in $nets; do
		    if [ "$network" = "$n" ]; then
			pid=$(echo "$json" | jq .[].State.Pid)
			echo "$pid"
			exit 0
		    fi
		done
	    done
	fi
	;;
    help)
	usage
	;;
    load)
	# shellcheck disable=SC2086
	name=$(unpack_archive "$1" $2)
	[ -n "$name" ] || exit 1

	# Show resulting image(s) matching $name
	podman images -n "$name"
	;;
    locate)			# Find where the host's ifname lives
	if [ -z "$network" ]; then
	    echo "Missing --net IFNAME option."
	    exit 1
	fi
	containers=$(podman ps $all --format "{{.Names}}")
	for c in $containers; do
	    json=$(podman inspect "$c")
	    nets=$(echo "$json" |jq -r '.[].NetworkSettings.Networks | keys[]' 2>/dev/null)
	    for n in $nets; do
		if [ "$network" = "$n" ]; then
		    echo "$c"
		    exit 0;
		fi
	    done
	done
	;;
    ls | list)
	cmd=$1
	[ -n "$cmd" ] && shift
	case $cmd in
	    image*)
		podman images $all --format "{{.Repository}}:{{.Tag}}"
		;;
	    oci)
		find $BUILTIN $DOWNLOADS -type f 2>/dev/null
		;;
	    *)
		podman ps $all --format "{{.Names}}"
		;;
	esac
	;;
    pull)
	podman pull "$@"
	;;
    remove)
	podman rmi $all $force -i "$1"
	;;
    run)
	img=$1
	cmd=$2
	[ -n "$port" ] || port="-P"
	if [ -n "$cmd" ]; then
	    shift 2
	    [ -n "$detach" ] || echo "Starting $img ENTRYPOINT $cmd :: use Ctrl-p Ctrl-q to detach"
	    podman run -it --rm $detach $port --entrypoint="$cmd" "$img" "$@"
	else
	    [ -n "$detach" ] || echo "Starting $img :: use Ctrl-p Ctrl-q to detach"
	    podman run -it --rm $detach $port "$img"
	fi
	;;
    save)
	name=$1
	file=$2
	if echo "$file" | grep -q ".gz"; then
	    file=${file%%.gz}
	    gzip=true
	fi
	if ! echo "$file" | grep -q ".tar"; then
	    file=${file}.tar
	    gzip=true
	fi
	podman save -o "$file" "$name"
	if [ -s "$file" ] && [ -n "$gzip" ]; then
	    gzip "$file"
	fi
	;;
    setup)
	[ -n "$name" ] || err 1 "setup: missing container name."
	script=/run/containers/${name}.sh
	[ -x "$script" ] || err 1 "setup: $script does not exist or is not executable."

	# Save our PID in case we get stuck here and someone wants to
	# stop us, e.g., due to reconfiguration or reboot.
	pidfile=$(pidfn "${name}")
	echo $$ > "$pidfile"

	while ! "$script"; do
            log "${name}: setup failed, waiting for network changes ..."
	    read -t 60 _ < <(ip monitor address route)

	    # On IP address/route changes, wait a few seconds more to ensure
	    # the system has ample time to react and set things up for us.
	    log "${name}: retrying ..."
	    sleep 2
	done

	rm -f "$pidfile"
	;;
    shell)
	if [ -z "$name" ]; then
	    name="$1"
	    shift
	fi
	if [ $# -gt 0 ]; then
	    podman exec -i "$name" sh -c "$*"
	else
	    podman exec -it "$name" sh -l
	fi
	;;
    show)
	cmd=$1
	[ -n "$cmd" ] && shift
	case $cmd in
	    image*)
		if [ -n "$simple" ]; then
		    podman images $all --format "{{.Names}} {{.Size}}" \
			| sed 's/\[\(.*\)\] /\1 /g' \
			| awk '{ printf "%-60s %s %s\n", $1, $2, $3}'
		else
		    podman images $all
		fi
		;;
	    volume*)
		printf "%-20s  CONTAINER\n" "VOLUME"
		for v in $(podman volume ls --format "{{.Name}}"); do
		    printf "%-20s" "$v"
		    podman ps -a --filter volume="$v"  --format '{{.Names}}' | sed 's/^/  /'
		done
		;;
	    *)
		if [ -n "$simple" ]; then
		    podman ps $all --format "{{.ID}}  {{.Names}}  {{.Image}}" \
			| awk '{ printf "%s  %-30s %s\n", $1, $2, $3}'
		else
		    podman ps $all
		fi
		;;
	esac
	;;
    start)
	if [ -n "$name" ]; then
	    wrap "$name" start
	elif [ -n "$1" ]; then
	    start "$1"
	else
	    usage
	    exit 1
	fi
	;;
    restart)
	if [ -n "$name" ]; then
	    wrap "$name" restart
	elif [ -n "$1" ]; then
	    cmd=$1
	    name=$2
	    if [ "$cmd" = "network" ] && [ -n "$name" ]; then
		netrestart "$name"
	    else
		name=$1
		stop "$name"
		while running "$name"; do
		    _=$((timeout -= 1))
		    if [ $timeout -le 0 ]; then
			err 1 "timed out waiting for container $1 to stop before restarting it."
		    fi
		    sleep 1
		done
		start "$name"
	    fi
	else
	    usage
	    exit 1
	fi
	;;
    stop)
	if [ -n "$name" ]; then
	    wrap "$name" stop
	elif [ -n "$1" ]; then
	    stop "$1"
	else
	    usage
	    exit 1
	fi
	;;
    stat*)
	podman stats -i 2
	;;
    upgrade)
	# Start script used to initially create container
	script=/run/containers/${1}.sh

	# Find container image
	img=$(podman inspect "$1" | jq -r .[].ImageName)
	if [ -z "$img" ]; then
	    echo "No such container ($1), or invalid ImageName.  Cannot upgrade."
	    exit 1;
	fi

	# Likely an OCI archive, or local directory, assume user has updated image.
	if echo "$img" | grep -Eq '^localhost/'; then
	    file=$(awk '/^# meta-image:/ {print $3}' "$script")
	    echo ">> Upgrading container $1 using $file ..."
	else
	    printf ">> Stopping ... "
	    podman stop "$1"
	    printf ">> "
	    podman pull "$img" || (echo "Failed fetching $img, check your network (settings)."; exit 1)
	    echo ">> Starting $1 ..."
	fi
	if ! "$script"; then
	    echo ">> Failed recreating container $1"
	    exit 1
	fi
	echo ">> Done."
	;;
    volume)
	cmd=$1
	[ -n "$cmd" ] && shift
	case $cmd in
	    prune)
		podman volume prune $force
		;;
	    *)
		false
		;;
	esac
	;;
    *)
	if [ -n "$SERVICE_SCRIPT_TYPE" ] && [ -n "$SERVICE_ID" ]; then
	    case "$SERVICE_SCRIPT_TYPE" in
		pre)
		    # Called as pre-script from Finit service
		    exec $container -q -n "$SERVICE_ID" setup
		    ;;
		cleanup)
		    # Called as cleanup-script from Finit service
		    log "Calling $container -n $SERVICE_ID delete"
		    exec $container -q -n "$SERVICE_ID" delete
		    ;;
		*)
		    false
		    ;;
	    esac
	fi
	usage
	exit 1
	;;
esac
