~curtin-dev/curtin/trunk

#!/bin/bash

set -f

VERBOSITY=0
KVM_PID=""
DRY_RUN=false
TEMP_D=""
DEF_BRIDGE="virbr0"
TAPDEVS=( )
# OVS_CLEANUP gets populated with bridge:devname pairs used with ovs
OVS_CLEANUP=( )
MAC_PREFIX="52:54:00:12:34"
KVM="kvm"
declare -A KVM_DEVOPTS

error() { echo "$@" 1>&2; }
fail() { [ $# -eq 0 ] || error "$@"; exit 1; }

bad_Usage() { Usage 1>&2; [ $# -eq 0 ] || error "$@"; exit 1; }
randmac() {
    # return random mac addr within final 3 tokens
    local random=""
    random=$(printf "%02x:%02x:%02x" \
        "$((${RANDOM}%256))" "$((${RANDOM}%256))" "$((${RANDOM}%256))")
    padmac "$random"
}

cleanup() {
    [ -z "${TEMP_D}" -o ! -d "${TEMP_D}" ] || rm -Rf "${TEMP_D}"
    [ -z "${KVM_PID}" ] || kill "$KVM_PID"
    if [ ${#TAPDEVS[@]} -ne 0 ]; then
        local name item
        for item in "${TAPDEVS[@]}"; do
            [ "${item}" = "skip" ] && continue
            debug 1 "removing" "$item"
            name="${item%:*}"
            if $DRY_RUN; then
                error ip tuntap del mode tap "$name"
            else
                ip tuntap del mode tap "$name"
            fi
            [ $? -eq 0 ] || error "failed removal of $name"
        done
        if [ ${#OVS_CLEANUP[@]} -ne 0 ]; then
            # with linux bridges, there seems to be no harm in just deleting
            # the device (not detaching from the bridge).  However, with
            # ovs, you have to remove them from the bridge, or later it
            # will refuse to add the same name.
            error "cleaning up ovs ports: ${OVS_CLEANUP[@]}"
            if ${DRY_RUN}; then
                error sudo "$0" tap-control ovs-cleanup "${OVS_CLEANUP[@]}"
            else
                sudo "$0" tap-control ovs-cleanup "${OVS_CLEANUP[@]}"
            fi
        fi
    fi
}

debug() {
    local level=${1}; shift;
    [ "${level}" -gt "${VERBOSITY}" ] && return
    error "${@}"
}

Usage() {
    cat <<EOF
Usage: ${0##*/} [ options ] -- kvm-args [ ... ]

   run kvm with a tap interface.

   options:
      -n | --netdev NETDEV    netdev can be 'user' or a bridge.
                              default is to bridge to $DEF_BRIDGE
      -d | --disk  DISK.img   attach DISK.img as a disk (via virtio)
           --dry-run          only report what would be done

           --uefi             boot with efi
           --uefi-nvram=FILE  boot with efi, using nvram settings in FILE
                              if FILE not present, copy from defaults.

   NETDEV:
    Above, 'NETDEV' is a comma delimited string
    The first field must be
     * bridge name: (br0 or virbr0): attach a device to this bridge
     * literal 'user': use qemu user networking

    Additional fields are optional, and can be anything that is acceptable
    to kvm either as:
      * '-device virtio-net-pci' option (see 'kvm -device virtio-net-pci,?')
      * '-net [user|tap]' option

   Example:
     * xkvm --netdev br0,macaddr=:05 -- -drive file=disk.img,if=virtio -curses
       attach a tap device to bridge 'br0' with mac address
         '${MAC_PREFIX}:05'

     * xkvm --netdev user,mac=random --netdev br1,model=e1000,mac=auto -- -curses
       attach virtio user networking nic with random mac address
       attach tap device to br1 bridge as e1000 with unspecified mac

     * xkvm --disk disk1.img
EOF
}

isdevopt() {
    local model="$1" input="${2%%=*}"
    local out="" opt="" opts=()
    if [ -z "${KVM_DEVOPTS[$model]}" ]; then
        out=$($KVM -device "$model,?" 2>&1) &&
            out=$(echo "$out" | sed -e "s,[^.]*[.],," -e 's,=.*,,') &&
            KVM_DEVOPTS[$model]="$out" ||
            { error "bad device model $model?"; exit 1; }
    fi
    opts=( ${KVM_DEVOPTS[$model]} )
    for opt in "${opts[@]}"; do
        [ "$input" = "$opt" ] && return 0
    done
    return 1
}

padmac() {
    # return a full mac, given a subset.
    # assume whatever is input is the last portion to be
    # returned, and fill it out with entries from MAC_PREFIX
    local mac="$1" num="$2" prefix="${3:-$MAC_PREFIX}" itoks="" ptoks=""
    # if input is empty set to :$num
    [ -n "$mac" ] || mac=$(printf "%02x" "$num") || return
    itoks=( ${mac//:/ } )
    ptoks=( ${prefix//:/ } )
    rtoks=( )
    for r in ${ptoks[@]:0:6-${#itoks[@]}} ${itoks[@]}; do
        rtoks[${#rtoks[@]}]="0x$r"
    done
    _RET=$(printf "%02x:%02x:%02x:%02x:%02x:%02x" "${rtoks[@]}")
}

make_nics_Usage() {
    cat <<EOF
Usage: ${0##*/} tap-control make-nics [options] bridge [bridge [..]]

   make a tap device on each of bridges requested
   outputs: 'tapname:type' for each input, or 'skip' if nothing needed.

   type is one of 'brctl' or 'ovs'
EOF
}

make_nics() {
    # takes input of list of bridges to create a tap device on
    # and echos either 'skip' or
    # <tapname>:<type> for each tap created
    # type is one of "ovs" or "brctl"
    local short_opts="v"
    local long_opts="--verbose"
    local getopt_out=""
    getopt_out=$(getopt --name "${0##*/} make-nics" \
        --options "${short_opts}" --long "${long_opts}" -- "$@") &&
        eval set -- "${getopt_out}" || { make_nics_Usage 1>&2; return 1; }

    local cur="" next=""
    while [ $# -ne 0 ]; do
        cur=${1}; next=${2};
        case "$cur" in
            -v|--verbose) VERBOSITY=$((${VERBOSITY}+1));;
            --) shift; break;;
        esac
        shift;
    done

    [ $# -ne 0 ] || {
        make_nics_Usage 1>&2; error "must give bridge";
        return 1;
    }

    local owner="" ovsbrs="" tap="" tapnum="0" brtype="" bridge=""
    [ "$(id -u)" = "0" ] || { error "must be root for make-nics"; return 1; }
    owner="${SUDO_USER:-root}"
    ovsbrs=""
    if command -v ovs-vsctl >/dev/null 2>&1; then
        out=$(ovs-vsctl list-br)
        out=$(echo "$out" | sed "s/\n/,/")
        ovsbrs=",$out,"
    fi
    for bridge in "$@"; do
        [ "$bridge" = "user" ] && echo skip && continue
        [ "${ovsbrs#*,${bridge},}" != "$ovsbrs" ] &&
            btype="ovs" || btype="brctl"
        tapnum=0;
        while [ -e /sys/class/net/tapvm$tapnum ]; do tapnum=$(($tapnum+1)); done
        tap="tapvm$tapnum"
        debug 1 "creating $tap:$btype on $bridge" 1>&2
        ip tuntap add mode tap user "$owner" "$tap" ||
            { error "failed to create tap '$tap' for '$owner'"; return 1; }
        ip link set "$tap" up 1>&2 || {
            error "failed to bring up $tap";
            ip tuntap del mode tap "$tap";
            return 1;
        }
        if [ "$btype" = "ovs" ]; then
            ovs-vsctl add-port "$bridge" "$tap" 1>&2 || {
                error "failed: ovs-vsctl add-port $bridge $tap";
                ovs-vsctl del-port "$bridge" "$tap"
                return 1;
            }
        else
            ip link set "$tap" master "$bridge" 1>&2 || {
                error "failed to add tap '$tap' to '$bridge'"
                ip tuntap del mode tap "$tap";
                return 1
            }
        fi
        echo "$tap:$btype"
    done
}

ovs_cleanup() {
    [ "$(id -u)" = "0" ] ||
        { error "must be root for ovs-cleanup"; return 1; }
    local item="" errors=0
    # TODO: if get owner (SUDO_USERNAME) and if that isn't
    # the owner, then do not delete.
    for item in "$@"; do
        name=${item#*:}
        bridge=${item%:*}
        ovs-vsctl del-port "$bridge" "$name" || errors=$((errors+1))
    done
    return $errors
}

quote_cmd() {
    local quote='"' x="" vline=""
    for x in "$@"; do
        if [ "${x#* }" != "${x}" ]; then
            if [ "${x#*$quote}" = "${x}" ]; then
                x="\"$x\""
            else
                x="'$x'"
            fi
        fi
        vline="${vline} $x"
    done
    echo "$vline"
}

get_bios_opts() {
    # get_bios_opts(bios, uefi, nvram)
    # bios is a explicit bios to boot.
    # uefi is boolean indicating uefi
    # nvram is optional and indicates that ovmf vars should be copied
    # to that file if it does not exist. if it exists, use it.
    local bios="$1" uefi="${2:-false}" nvram="$3"
    local ovmf_dir="/usr/share/OVMF"
    local bios_opts="" pflash_common="if=pflash,format=raw"
    unset _RET
    _RET=( )
    if [ -n "$bios" ]; then
        _RET=( -drive "${pflash_common},file=$bios" )
        return 0
    elif ! $uefi; then
        return 0
    fi

    # ovmf in older releases (14.04) shipped only a single file
    #   /usr/share/ovmf/OVMF.fd
    # newer ovmf ships split files
    #   /usr/share/OVMF/OVMF_CODE.fd
    #   /usr/share/OVMF/OVMF_VARS.fd
    # with single file, pass only one file and read-write
    # with split, pass code as readonly and vars as read-write
    local joined="/usr/share/ovmf/OVMF.fd"
    local code="/usr/share/OVMF/OVMF_CODE.fd"
    local vars="/usr/share/OVMF/OVMF_VARS.fd"
    local split="" nvram_src=""
    if [ -e "$code" -o -e "$vars" ]; then
        split=true
        nvram_src="$vars"
    elif [ -e "$joined" ]; then
        split=false
        nvram_src="$joined"
    elif [ -n "$nvram" -a -e "$nvram" ]; then
        error "WARN: nvram given, but did not find expected ovmf files."
        error "      assuming this is code and vars (OVMF.fd)"
        split=false
    else
        error "uefi support requires ovmf bios: apt-get install -qy ovmf"
        return 1
    fi

    if [ -n "$nvram" ]; then
        if [ ! -f "$nvram" ]; then
            cp "$nvram_src" "$nvram" || 
                { error "failed copy $nvram_src to $nvram"; return 1; }
            debug 1 "copied $nvram_src to $nvram"
        fi
    else
        debug 1 "uefi without --uefi-nvram storage." \
            "nvram settings likely will not persist."
        nvram="${nvram_src}"
    fi

    if [ ! -w "$nvram" ]; then
        debug 1 "nvram file ${nvram} is readonly"
        nvram_ro="readonly"
    fi

    if $split; then
        # to ensure bootability firmware must be first, then variables
        _RET=( -drive "${pflash_common},file=$code,readonly" )
    fi
    _RET=( "${_RET[@]}"
           -drive "${pflash_common},file=$nvram${nvram_ro:+,${nvram_ro}}" )
}

main() {
    local short_opts="hd:n:v"
    local long_opts="bios:,help,dowait,disk:,dry-run,kvm:,no-dowait,netdev:,uefi,uefi-nvram:,verbose"
    local getopt_out=""
    getopt_out=$(getopt --name "${0##*/}" \
        --options "${short_opts}" --long "${long_opts}" -- "$@") &&
        eval set -- "${getopt_out}" || { bad_Usage; return 1; }

    local bridge="$DEF_BRIDGE" oifs="$IFS"
    local netdevs="" need_tap="" ret="" p="" i="" pt="" cur="" conn=""
    local kvm="" kvmcmd="" archopts=""
    local def_disk_driver=${DEF_DISK_DRIVER:-"virtio-blk"}
    local def_netmodel=${DEF_NETMODEL:-"virtio-net-pci"}
    local bios="" uefi=false uefi_nvram=""

    archopts=( )
    kvmcmd=( )
    netdevs=( )
    addargs=( )
    diskdevs=( )
    diskargs=( )

    # dowait: run qemu-system with a '&' and then 'wait' on the pid.
    #  the reason to do this or not do this has to do with interactivity
    #  if detached with &, then user input will not go to xkvm.
    #  if *not* detached, then signal handling is blocked until
    #  the foreground subprocess returns. which means we can't handle
    #  a sigterm and kill the qemu-system process.
    #  We default to dowait=false if input and output are a terminal
    local dowait=""
    [ -t 0 -a -t 1 ] && dowait=false || dowait=true
    while [ $# -ne 0 ]; do
        cur=${1}; next=${2};
        case "$cur" in
            -h|--help) Usage; exit 0;;
            -d|--disk)
                diskdevs[${#diskdevs[@]}]="$next"; shift;;
            --dry-run) DRY_RUN=true;;
            --kvm) kvm="$next"; shift;;
            -n|--netdev)
                netdevs[${#netdevs[@]}]=$next; shift;;
            -v|--verbose) VERBOSITY=$((${VERBOSITY}+1));;
            --dowait) dowait=true;;
            --no-dowait) dowait=false;;
            --bios) bios="$next"; shift;;
            --uefi) uefi=true;;
            --uefi-nvram) uefi=true; uefi_nvram="$next"; shift;;
            --) shift; break;;
        esac
        shift;
    done

    [ ${#netdevs[@]} -eq 0 ] && netdevs=( "${DEF_BRIDGE}" )
    pt=( "$@" )

    local kvm_pkg="" virtio_scsi_bus="virtio-scsi-pci"
    [ -n "$kvm" ] && kvm_pkg="none"
    case $(uname -m) in
        i?86)
            [ -n "$kvm" ] ||
                { kvm="qemu-system-i386"; kvm_pkg="qemu-system-x86"; }
            ;;
        x86_64)
            [ -n "$kvm" ] ||
                { kvm="qemu-system-x86_64"; kvm_pkg="qemu-system-x86"; }
            ;;
        s390x)
            [ -n "$kvm" ] ||
                { kvm="qemu-system-s390x"; kvm_pkg="qemu-system-misc"; }
            def_netmodel=${DEF_NETMODEL:-"virtio-net-ccw"}
            virtio_scsi_bus="virtio-scsi-ccw"
            ;;
        ppc64*)
            [ -n "$kvm" ] ||
                { kvm="qemu-system-ppc64"; kvm_pkg="qemu-system-ppc"; }
            def_netmodel="virtio-net-pci"
            # virtio seems functional on in 14.10, but might want scsi here
            #def_diskif="scsi"
            archopts=( "${archopts[@]}" -machine pseries,usb=off )
            archopts=( "${archopts[@]}" -device spapr-vscsi )
            ;;
        *) kvm=qemu-system-$(uname -m);;
    esac
    KVM="$kvm"
    kvmcmd=( $kvm -enable-kvm )

    local bios_opts=""
    if [ -n "$bios" ] && $uefi; then
        error "--uefi (or --uefi-nvram) is incompatible with --bios"
        return 1
    fi
    get_bios_opts "$bios" "$uefi" "$uefi_nvram" ||
        { error "failed to get bios opts"; return 1; }
    bios_opts=( "${_RET[@]}" )

    local out="" fmt="" bus="" unit="" index="" serial="" driver="" devopts=""
    local busorindex="" driveopts="" cur="" val="" file=""
    for((i=0;i<${#diskdevs[@]};i++)); do
        cur=${diskdevs[$i]}
        IFS=","; set -- $cur; IFS="$oifs"
        driver=""
        id=$(printf "disk%02d" "$i")
        file=""
        fmt=""
        bus=""
        unit=""
        index=""
        serial=""
        for tok in "$@"; do
            [ "${tok#*=}" = "${tok}" -a -f "${tok}" -a -z "$file" ] && file="$tok"
            val=${tok#*=}
            case "$tok" in
                driver=*) driver=$val;;
                if=virtio) driver=virtio-blk;;
                if=scsi) driver=scsi-hd;;
                if=pflash) driver=;;
                if=sd|if=mtd|floppy) fail "do not know what to do with $tok on $cur";;
                id=*) id=$val;;
                file=*) file=$val;;
                fmt=*|format=*) fmt=$val;;
                serial=*) serial=$val;;
                bus=*) bus=$val;;
                unit=*) unit=$val;;
                index=*) index=$val;;
            esac
        done
        [ -z "$file" ] && fail "did not read a file from $cur"
        if [ -f "$file" -a -z "$fmt" ]; then
            out=$(LANG=C qemu-img info "$file") &&
                fmt=$(echo "$out" | awk '$0 ~ /^file format:/ { print $3 }') ||
                { error "failed to determine format of $file"; return 1; }
        else
            fmt=raw
        fi
        if [ -z "$driver" ]; then
            driver="$def_disk_driver"
        fi
        if [ -z "$serial" ]; then
            serial="${file##*/}"
        fi

        # make sure we add either bus= or index=
        if [ -n "$bus" -o "$unit" ] && [ -n "$index" ]; then
            fail "bus and index cant be specified together: $cur"
        elif [ -z "$bus" -a -z "$unit" -a -z "$index" ]; then
            index=$i
        elif [ -n "$bus" -a -z "$unit" ]; then
            unit=$i
        fi

        busorindex="${bus:+bus=$bus,unit=$unit}${index:+index=${index}}"
        diskopts="file=${file},id=$id,if=none,format=$fmt,$busorindex"
        devopts="$driver,drive=$id${serial:+,serial=${serial}}"
        for tok in "$@"; do
            case "$tok" in
                id=*|if=*|driver=*|$file|file=*) continue;;
                fmt=*|format=*) continue;;
                serial=*|bus=*|unit=*|index=*) continue;;
            esac
            isdevopt "$driver" "$tok" && devopts="${devopts},$tok" ||
                diskopts="${diskopts},${tok}"
        done

        diskargs=( "${diskargs[@]}" -drive "$diskopts" -device "$devopts" )
    done

    local mnics_vflag=""
    for((i=0;i<${VERBOSITY}-1;i++)); do mnics_vflag="${mnics_vflag}v"; done
    [ -n "$mnics_vflag" ] && mnics_vflag="-${mnics_vflag}"

    # now go through and split out options
    # -device virtio-net-pci,netdev=virtnet0,mac=52:54:31:15:63:02
    # -netdev type=tap,id=virtnet0,vhost=on,script=/etc/kvm/kvm-ifup.br0,downscript=no
    local netopts="" devopts="" id="" need_taps=0 model=""
    local device_args netdev_args
    device_args=( )
    netdev_args=( )
    connections=( )
    for((i=0;i<${#netdevs[@]};i++)); do
        id=$(printf "net%02d" "$i")
        netopts="";
        devopts=""
        # mac=auto is 'unspecified' (let qemu assign one)
        mac="auto"
        #vhost="off"

        IFS=","; set -- ${netdevs[$i]}; IFS="$oifs"
        bridge=$1; shift;
        if [ "$bridge" = "user" ]; then
            netopts="type=user"
            ntype="user"
            connections[$i]="user"
        else
            need_taps=1
            ntype="tap"
            netopts="type=tap"
            connections[$i]="$bridge"
        fi
        netopts="${netopts},id=$id"
        [ "$ntype" = "tap" ] && netopts="${netopts},script=no,downscript=no"

        model="${def_netmodel}"
        for tok in "$@"; do
            [ "${tok#model=}" = "${tok}" ] && continue
            case "${tok#model=}" in
                virtio) model=virtio-net-pci;;
                *) model=${tok#model=};;
            esac
        done

        for tok in "$@"; do
            case "$tok" in
                mac=*) mac="${tok#mac=}"; continue;;
                macaddr=*) mac=${tok#macaddr=}; continue;;
                model=*) continue;;
            esac

            isdevopt "$model" "$tok" && devopts="${devopts},$tok" ||
                netopts="${netopts},${tok}"
        done
        devopts=${devopts#,}
        netopts=${netopts#,}

        if [ "$mac" != "auto" ]; then
            [ "$mac" = "random" ] && randmac && mac="$_RET"
            padmac "$mac" "$i"
            devopts="${devopts:+${devopts},}mac=$_RET"
        fi
        devopts="$model,netdev=$id${devopts:+,${devopts}}"
        #netopts="${netopts},vhost=${vhost}"

        device_args[$i]="$devopts"
        netdev_args[$i]="$netopts"
    done

    trap cleanup EXIT

    reqs=( "$kvm" )
    pkgs=( "$kvm_pkg" )
    for((i=0;i<${#reqs[@]};i++)); do
        req=${reqs[$i]}
        pkg=${pkgs[$i]}
        [ "$pkg" = "none" ] && continue
        command -v "$req" >/dev/null || {
            missing="${missing:+${missing} }${req}"
            missing_pkgs="${missing_pkgs:+${missing_pkgs} }$pkg"
        }
    done
    if [ -n "$missing" ]; then
        local reply cmd=""
        cmd=( sudo apt-get --quiet install ${missing_pkgs} )
        error "missing prereqs: $missing";
        error "install them now with the following?: ${cmd[*]}"
        read reply && [ "$reply" = "y" -o "$reply" = "Y" ] ||
            { error "run: apt-get install ${missing_pkgs}"; return 1; }
        "${cmd[@]}" || { error "failed to install packages"; return 1; }
    fi

    if [ $need_taps -ne 0 ]; then
        local missing="" missing_pkgs="" reqs="" req="" pkgs="" pkg=""
        for i in "${connections[*]}"; do
            [ "$i" = "user" -o -e "/sys/class/net/dev/$i" ] ||
                missing="${missing} $i"
        done
        [ -z "$missing" ] || {
            error "cannot create connection on ${missing# }."
            error "bridges do not exist.";
            return 1;
        }
        error "creating tap devices: ${connections[*]}"
        if $DRY_RUN; then
            error "sudo $0 tap-control make-nics" \
                $mnics_vflag "${connections[@]}"
            taps=""
            for((i=0;i<${#connections[@]};i++)); do
                if [ "${connections[$i]}" = "user" ]; then
                    taps="${taps} skip"
                else
                    taps="${taps} dryruntap$i:brctl"
                fi
            done
        else
            taps=$(sudo "$0" tap-control make-nics \
                   ${mnics_vflag} "${connections[@]}") ||
                { error "$failed to make-nics ${connections[*]}"; return 1; }
        fi
        TAPDEVS=( ${taps} )
        for((i=0;i<${#TAPDEVS[@]};i++)); do
            cur=${TAPDEVS[$i]}
            [ "${cur#*:}" = "ovs" ] || continue
            conn=${connections[$i]}
            OVS_CLEANUP[${#OVS_CLEANUP[@]}]="${conn}:${cur%:*}"
        done

        debug 2 "tapdevs='${TAPDEVS[@]}'"
        [ ${#OVS_CLEANUP[@]} -eq 0 ] || error "OVS_CLEANUP='${OVS_CLEANUP[*]}'"

        for((i=0;i<${#TAPDEVS[@]};i++)); do
            cur=${TAPDEVS[$i]}
            [ "$cur" = "skip" ] && continue
            netdev_args[$i]="${netdev_args[$i]},ifname=${cur%:*}";
        done
    fi

    netargs=()
    for((i=0;i<${#device_args[@]};i++)); do
        netargs=( "${netargs[@]}" -device "${device_args[$i]}"
                  -netdev "${netdev_args[$i]}")
    done

    local bus_devices
    bus_devices=( -device "$virtio_scsi_bus,id=virtio-scsi-xkvm" )
    cmd=( "${kvmcmd[@]}" "${archopts[@]}" 
          "${bios_opts[@]}"
          "${bus_devices[@]}"
          "${netargs[@]}"
          "${diskargs[@]}" "${pt[@]}" )
    local pcmd=$(quote_cmd "${cmd[@]}")
    error "$pcmd"
    ${DRY_RUN} && return 0

    if $dowait; then
        "${cmd[@]}" &
        KVM_PID=$!
        debug 1 "kvm pid=$KVM_PID. my pid=$$"
        wait
        ret=$?
        KVM_PID=""
    else
        "${cmd[@]}"
        ret=$?
    fi
    return $ret
}


if [ "$1" = "tap-control" ]; then
    shift
    mode=$1
    shift || fail "must give mode to tap-control"
    case "$mode" in
        make-nics) make_nics "$@";;
        ovs-cleanup) ovs_cleanup "$@";;
        *) fail "tap mode must be either make-nics or ovs-cleanup";;
    esac
else
    main "$@"
fi

# vi: ts=4 expandtab