hooks/nvidia

   1 #! /bin/bash
   2
   3 # Copyright (c) 2017, NVIDIA CORPORATION. All rights reserved.
   4
   5 set -eu
   6
   7 # NVIDIA_VISIBLE_DEVICES="" *or* NVIDIA_VISIBLE_DEVICES="void"
   8 # GPU support was explicitly disabled, exit early.
   9 if [ -z "${NVIDIA_VISIBLE_DEVICES-x}" ] || [ "${NVIDIA_VISIBLE_DEVICES:-}" = "void" ]; then
  10     exit 0
  11 fi
  12
  13 # https://github.com/nvidia/nvidia-container-runtime#cuda_version
  14 if [ -n "${CUDA_VERSION:-}" ] && [ -z "${NVIDIA_REQUIRE_CUDA:-}" ]; then
  15     # Legacy CUDA image: default to all devices and all driver capabilities.
  16     if [ -z "${NVIDIA_VISIBLE_DEVICES+x}" ]; then
  17         NVIDIA_VISIBLE_DEVICES="all"
  18     fi
  19     if [ -z "${NVIDIA_DRIVER_CAPABILITIES:-}" ]; then
  20         NVIDIA_DRIVER_CAPABILITIES="all"
  21     fi
  22     if [[ "${CUDA_VERSION}" =~ ^[0-9]+\.[0-9]+ ]]; then
  23         NVIDIA_REQUIRE_CUDA="cuda>=${BASH_REMATCH[0]}"
  24     fi
  25 else
  26     # NVIDIA_VISIBLE_DEVICES unset and it's not a legacy CUDA image.
  27     # This is not a GPU image, exit early.
  28     if [ -z "${NVIDIA_VISIBLE_DEVICES+x}" ]; then
  29         exit 0
  30     fi
  31 fi
  32
  33 export PATH=$PATH:/usr/sbin:/usr/bin:/sbin:/bin
  34 if ! which nvidia-container-cli >/dev/null; then
  35     echo "ERROR: Missing tool nvidia-container-cli, see https://github.com/NVIDIA/libnvidia-container" >&2
  36     exit 1
  37 fi
  38
  39 in_userns() {
  40     [ -e /proc/self/uid_map ] || { echo no; return; }
  41     while read line; do
  42         fields=$(echo $line | awk '{ print $1 " " $2 " " $3 }')
  43         [ "$fields" = "0 0 4294967295" ] && { echo no; return; } || true
  44         echo $fields | grep -q " 0 1$" && { echo userns-root; return; } || true
  45     done < /proc/self/uid_map
  46
  47     [ "$(cat /proc/self/uid_map)" = "$(cat /proc/1/uid_map)" ] && \
  48         { echo userns-root; return; }
  49     echo yes
  50 }
  51
  52 get_ldconfig() {
  53     which "ldconfig.real" || which "ldconfig"
  54     return $?
  55 }
  56
  57 capability_to_cli() {
  58     case "$1" in
  59         compute)  echo "--compute";;
  60         compat32) echo "--compat32";;
  61         graphics) echo "--graphics";;
  62         utility)  echo "--utility";;
  63         video)    echo "--video";;
  64         *)        exit 1;;
  65     esac
  66     return
  67 }
  68
  69 # Same behavior as strconv.ParseBool in golang
  70 parse_bool() {
  71     case "$1" in
  72         1|t|T|TRUE|true|True)    echo "true";;
  73         0|f|F|FALSE|false|False) echo "false";;
  74         *)                       exit 1;;
  75     esac
  76     return
  77 }
  78
  79 usage() {
  80     cat <<EOF
  81 nvidia-container-cli hook for LXC
  82
  83 Special arguments:
  84 [ -h | --help ]: Print this help message and exit.
  85
  86 Optional arguments:
  87 [ --no-load-kmods ]: Do not try to load the NVIDIA kernel modules.
  88 [ --disable-require ]: Disable all the constraints of the form NVIDIA_REQUIRE_*.
  89 [ --debug <path> ]: The path to the log file.
  90 [ --ldconfig <path> ]: The path to the ldconfig binary, use a '@' prefix for a host path.
  91 EOF
  92     return 0
  93 }
  94
  95 options=$(getopt -o h -l help,no-load-kmods,disable-require,debug:,ldconfig: -- "$@")
  96 if [ $? -ne 0 ]; then
  97     usage
  98     exit 1
  99 fi
 100 eval set -- "$options"
 101
 102 CLI_LOAD_KMODS="true"
 103 CLI_DISABLE_REQUIRE="false"
 104 CLI_DEBUG=
 105 CLI_LDCONFIG=
 106
 107 while :; do
 108     case "$1" in
 109         --help)             usage && exit 1;;
 110         --no-load-kmods)    CLI_LOAD_KMODS="false"; shift 1;;
 111         --disable-require)  CLI_DISABLE_REQUIRE="true"; shift 1;;
 112         --debug)            CLI_DEBUG=$2; shift 2;;
 113         --ldconfig)         CLI_LDCONFIG=$2; shift 2;;
 114         --)                 shift 1; break;;
 115         *)                  break;;
 116     esac
 117 done
 118
 119 HOOK_SECTION=
 120 HOOK_TYPE=
 121 case "${LXC_HOOK_VERSION:-0}" in
 122     0) HOOK_SECTION="${2:-}"; HOOK_TYPE="${3:-}";;
 123     1) HOOK_SECTION="${LXC_HOOK_SECTION:-}"; HOOK_TYPE="${LXC_HOOK_TYPE:-}";;
 124     *) echo "ERROR: Unsupported hook version: ${LXC_HOOK_VERSION}." >&2; exit 1;;
 125 esac
 126
 127 if [ "${HOOK_SECTION}" != "lxc" ]; then
 128     echo "ERROR: Not running through LXC." >&2
 129     exit 1
 130 fi
 131
 132 if [ "${HOOK_TYPE}" != "mount" ]; then
 133     echo "ERROR: This hook must be used as a \"mount\" hook." >&2
 134     exit 1
 135 fi
 136
 137 USERNS=$(in_userns)
 138 if [ "${USERNS}" != "yes" ]; then
 139     # This is a limitation of libnvidia-container.
 140     echo "FIXME: This hook currently only works in unprivileged mode." >&2
 141     exit 1
 142 fi
 143
 144 if [ "${USERNS}" = "yes" ]; then
 145     CLI_LOAD_KMODS="false"
 146     if ! grep -q nvidia_uvm /proc/modules; then
 147         echo "WARN: Kernel module nvidia_uvm is not loaded, nvidia-container-cli might fail. Make sure the NVIDIA device driver is installed and loaded." >&2
 148     fi
 149 fi
 150
 151 # https://github.com/nvidia/nvidia-container-runtime#nvidia_disable_require
 152 if [ -n "${NVIDIA_DISABLE_REQUIRE:-}" ]; then
 153     if [ "$(parse_bool "${NVIDIA_DISABLE_REQUIRE}")" = "true" ]; then
 154         CLI_DISABLE_REQUIRE="true"
 155     fi
 156 fi
 157
 158 if [ -z "${CLI_DEBUG}" ]; then
 159     if [ "${LXC_LOG_LEVEL}" = "DEBUG" ] || [ "${LXC_LOG_LEVEL}" = "TRACE" ]; then
 160         rootfs_path="${LXC_ROOTFS_PATH#*:}"
 161         hookdir="${rootfs_path/%rootfs/hook}"
 162         if mkdir -p "${hookdir}"; then
 163             CLI_DEBUG="${hookdir}/nvidia.log"
 164         fi
 165     fi
 166 fi
 167
 168 # A '@' prefix means a host path.
 169 if [ -z "${CLI_LDCONFIG}" ]; then
 170     if host_ldconfig=$(get_ldconfig); then
 171         CLI_LDCONFIG="@${host_ldconfig}"
 172     fi
 173 fi
 174
 175 # https://github.com/nvidia/nvidia-container-runtime#nvidia_visible_devices
 176 CLI_DEVICES="${NVIDIA_VISIBLE_DEVICES}"
 177
 178 # https://github.com/nvidia/nvidia-container-runtime#nvidia_driver_capabilities
 179 CLI_CAPABILITIES=
 180 if [ -n "${NVIDIA_DRIVER_CAPABILITIES:-}" ]; then
 181      CLI_CAPABILITIES="${NVIDIA_DRIVER_CAPABILITIES//,/ }"
 182 fi
 183
 184 # https://github.com/nvidia/nvidia-container-runtime#nvidia_require_
 185 CLI_REQUIREMENTS=
 186 for req in $(compgen -e "NVIDIA_REQUIRE_"); do
 187     CLI_REQUIREMENTS="${CLI_REQUIREMENTS} ${!req}"
 188 done
 189
 190 if [ "${CLI_CAPABILITIES}" = "all" ]; then
 191     CLI_CAPABILITIES="compute compat32 graphics utility video"
 192 fi
 193
 194 if [ -z "${CLI_CAPABILITIES}" ]; then
 195     CLI_CAPABILITIES="utility"
 196 fi
 197
 198 global_args=("")
 199 configure_args=("")
 200
 201 if [ -n "${CLI_DEBUG}" ]; then
 202     echo "INFO: Writing nvidia-container-cli log at ${CLI_DEBUG}." >&2
 203     global_args+=("--debug=${CLI_DEBUG}")
 204 fi
 205
 206 if [ "${CLI_LOAD_KMODS}" = "true" ]; then
 207     global_args+=(--load-kmods)
 208 fi
 209
 210 if [ "${USERNS}" = "yes" ]; then
 211     global_args+=(--user)
 212     configure_args+=(--no-cgroups)
 213 fi
 214
 215 if [ -n "${CLI_LDCONFIG}" ]; then
 216     configure_args+=(--ldconfig="${CLI_LDCONFIG}")
 217 fi
 218
 219 if [ -n "${CLI_DEVICES}" ] && [ "${CLI_DEVICES}" != "none" ]; then
 220     configure_args+=(--device="${CLI_DEVICES}")
 221 fi
 222
 223 for cap in ${CLI_CAPABILITIES}; do
 224     if arg=$(capability_to_cli "${cap}"); then
 225         configure_args+=("${arg}")
 226     else
 227         echo "ERROR: Unknown driver capability \"${cap}\"." >&2
 228         exit 1
 229     fi
 230 done
 231
 232 if [ "${CLI_DISABLE_REQUIRE}" = "false" ]; then
 233     for req in ${CLI_REQUIREMENTS}; do
 234         configure_args+=(--require="${req}")
 235     done
 236 fi
 237
 238 set -x
 239 exec nvidia-container-cli ${global_args[@]} configure ${configure_args[@]} "${LXC_ROOTFS_MOUNT}"