#!/bin/sh

# Copyright (c) 2010 The Chromium OS Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.

# This script can be called during startup to trash the stateful partition
# and possibly reset the other root filesystem

. /usr/share/misc/chromeos-common.sh
SCRIPT="$0"

# stateful partition isn't around for logging, so dump to the screen:
set -x

# Log file to store the output of this run.
CLOBBER_STATE_LOG="/tmp/clobber-state.log"
: ${TTY=/dev/tty1}

# Run "clobber-state fast" to clobber the state quickly but unsecurely.
FACTORY_WIPE=$(echo "$@" | grep -sqw factory && echo "factory")
FAST_WIPE=$(echo "$@" | grep -sqw fast && echo "fast")
KEEPIMG=$(echo "$@" | grep -sqw keepimg && echo "keepimg")
SAFE_WIPE=$(echo "$@" | grep -sqw safe && echo "safe")

PRESERVED_TAR="/tmp/preserve.tar"
LABMACHINE=".labmachine"
STATE_PATH="/mnt/stateful_partition"
POWERWASH_COUNT="${STATE_PATH}/unencrypted/preserve/powerwash_count"

# List of files to preserve relative to /mnt/stateful_partition/
PRESERVED_FILES=""
PRESERVED_LIST=""

ensure_root_account() {
  if [ $(id -u) -ne 0 ]; then
    echo 'You must run this as root'
    exit 1
  fi
}

redirect_log_start() {
  # Redirect stdout to our log.
  exec >"${CLOBBER_STATE_LOG}" 2>&1
}

redirect_log_stop() {
  # Stop appending to the log and preserve it.
  exec >/dev/null 2>&1
  mv -f "${CLOBBER_STATE_LOG}" "${STATE_PATH}/unencrypted/clobber-state.log"
}

preserve_files() {
  # Preserve these files in safe mode.
  if [ "$SAFE_WIPE" = "safe" ]; then
    PRESERVED_FILES="
      unencrypted/preserve/attestation.epb
      unencrypted/preserve/enrollment_state.epb
      unencrypted/preserve/local_tpm_data
      unencrypted/preserve/powerwash_count
      unencrypted/preserve/update_engine/prefs/rollback-version
      home/.shadow/install_attributes.pb
    "

    # Powerwash count is only preserved for "safe" powerwashes.
    COUNT=1
    if [ -f $POWERWASH_COUNT ]; then
      COUNT_UNSANITIZED=$(head -1 $POWERWASH_COUNT | cut -c1-4)
      if [ $(expr "$COUNT_UNSANITIZED" : "^[0-9][0-9]*$") -ne 0 ]; then
        COUNT=$(( COUNT_UNSANITIZED + 1 ))
      fi
    fi
    echo $COUNT > $POWERWASH_COUNT
  fi

  # For a factory wipe (and only factory) preserve seed for CRX cache.
  if [ "$FACTORY_WIPE" = "factory" ]; then
    IMPORT_FILES="$(cd ${STATE_PATH};
                    echo unencrypted/import_extensions/extensions/*.crx)"
    if [ "$IMPORT_FILES" != \
         "unencrypted/import_extensions/extensions/*.crx" ]; then
      PRESERVED_FILES="${PRESERVED_FILES} ${IMPORT_FILES}"
    fi
  fi

  # Test images in the lab enable certain extra behaviors if the
  # .labmachine flag file is present.  Those behaviors include some
  # important recovery behaviors (cf. the recover_duts upstart job).
  # We need those behaviors to survive across power wash, otherwise,
  # the current boot could wind up as a black hole.
  if [ -f "${STATE_PATH}/${LABMACHINE}" ] &&
      crossystem 'debug_build?1'; then
    PRESERVED_FILES="${PRESERVED_FILES} ${LABMACHINE}"
  fi

  if [ -n "$PRESERVED_FILES" ]; then
    # We want to preserve permissions and recreate the directory structure
    # for all of the files in the PRESERVED_FILES variable. In order to do
    # so we run tar --no-recurison and specify the names of each of the
    # parent directories. For example for home/.shadow/install_attributes.pb
    # we pass to tar home home/.shadow home/.shadow/install_attributes.pb
    for file in $PRESERVED_FILES; do
      if [ ! -e "$STATE_PATH/$file" ]; then
        continue
      fi
      path=$file
      while [ "$path" != '.' ]; do
        PRESERVED_LIST="$path $PRESERVED_LIST"
        path=$(dirname $path)
      done
    done
    tar cf $PRESERVED_TAR -C $STATE_PATH --no-recursion -- $PRESERVED_LIST
  fi
}

wipe_block_dev() {
  DEV="$1"
  # Wipe the filesystem size if we can determine it. Full partition wipe
  # takes a long time on 16G SSD or rotating media.
  if dumpe2fs ${DEV} ; then
    FS_BLOCKS=$(dumpe2fs -h ${DEV} | grep "Block count" | sed "s/^.*:\W*//")
    FS_BLOCKSIZE=$(dumpe2fs -h ${DEV} | grep "Block size" | sed "s/^.*:\W*//")
  else
    FS_BLOCKS=$(numsectors ${DEV})
    FS_BLOCKSIZE=512
  fi
  BLOCKS_4M=$((4 * 1024 * 1024 / $FS_BLOCKSIZE))  # 4MiB in sectors
  FULL_BLKS=$(($FS_BLOCKS / $BLOCKS_4M))
  REMAINDER_SECTORS=$(($FS_BLOCKS % $BLOCKS_4M))

  if type pv; then
    # Opening a TTY device with no one else attached will allocate and reset
    # terminal attributes. To prevent user input echo and ctrl-breaks, we need
    # to allocate and configure the terminal before using tty for output.
    # Note &2 is currently used for debug messages dumping so we can't redirect
    # subshell by 2>"${TTY}".
    ( stty -F "${TTY}" raw -echo -cread
      pv -etp -s $((4 * 1024 * 1024 * ${FULL_BLKS})) /dev/zero 2>"${TTY}" |
        dd of=${DEV} bs=4M count=${FULL_BLKS} iflag=fullblock oflag=sync
    ) >"${TTY}"
  else
    dd if=/dev/zero of=${DEV} bs=4M count=${FULL_BLKS}
  fi
  dd if=/dev/zero of=${DEV} bs=${FS_BLOCKSIZE} count=${REMAINDER_SECTORS} \
    seek=$(($FULL_BLKS * $BLOCKS_4M))
}

# Get the specified information from the specified UBI volume.
#  $1 the field name such as "reserved_for_bad", or "data_bytes".
#  $2 the volume number such as "1" or "1_0".
get_ubi_var() {
  local field="$1"
  local part="$2"
  cat "/sys/class/ubi/ubi${part}/${field}"
}

get_ubi_reserved_ebs() {
  get_ubi_var reserved_for_bad "$@"
}

get_ubi_volume_size() {
  local part="$1"
  get_ubi_var data_bytes "${part}_0"
}

# Calculate the maximum number of bad blocks per 1024 blocks for UBI.
#  $1 partition number
calculate_ubi_max_beb_per_1024() {
  local part_no mtd_size eb_size nr_blocks
  part_no="$1"
  # The max beb per 1024 is on the total device size, not the partition size.
  mtd_size=$(cat /sys/class/mtd/mtd0/size)
  eb_size=$(cat /sys/class/mtd/mtd0/erasesize)
  nr_blocks=$((mtd_size / eb_size))
  reserved_ebs=$(get_ubi_reserved_ebs "${part_no}")
  echo $((reserved_ebs * 1024 / nr_blocks))
}

# Wipe a UBI device.
#  $1 the device node. This must be in the form /dev/ubiX_0 or /dev/ubiblockX_0.
wipe_ubi_dev() {
  local dev phy_dev part_no part_name
  dev="$1"
  part_no=$(echo "${dev}" | sed -e 's#/dev/ubi\(block\)\?\([0-9]\)_0#\2#')
  phy_dev="/dev/ubi${part_no}"
  # We only wipe state and root partitions in this function.
  case "${part_no}" in
    "${PARTITION_NUM_STATE}")
      part_name="STATE"
      ;;
    "${PARTITION_NUM_ROOT_A}")
      part_name="ROOT_A"
      ;;
    "${PARTITION_NUM_ROOT_B}")
      part_name="ROOT_B"
      ;;
    *)
      part_name="UNKNOWN_${part_no}"
      echo "Do not know how to name UBI partition number ${part_no}."
      ;;
  esac

  if [ ! -c "${phy_dev}" ]; then
    # Try to attach the volume to obtain info about it.
    ubiattach -m "${part_no}" -d "${part_no}"
  fi

  local max_beb_per_1024 volume_size
  max_beb_per_1024=$(calculate_ubi_max_beb_per_1024 "${part_no}")
  volume_size=$(get_ubi_volume_size "${part_no}")

  ubidetach -d "${part_no}"
  ubiformat -y -e 0 "/dev/mtd${part_no}"

  # We need to attach so that we could set max beb/1024 and create a volume.
  # After a volume is created, we don't need to specify max beb/1024 anymore.
  ubiattach -d "${part_no}" -m "${part_no}" \
            --max-beb-per1024 "${max_beb_per_1024}"

  ubimkvol -s "${volume_size}" -N "${part_name}" "${phy_dev}"
}

# Wipe an MTD device.
#  $1 the device node. This must be /dev/mtdX or /dev/ubiX_0.
wipe_mtd_dev() {
  local dev="$1"
  case "${dev}" in
    "/dev/mtd"*)
      flash_erase "${dev}" 0 0
      ;;
    "/dev/ubi"*)
      wipe_ubi_dev "${dev}"
      ;;
    *)
      echo "Do not know how to wipe ${dev}."
      ;;
  esac
}

# Perform media-dependent wipe of the device. IS_MTD flag controls if the wipe
# should be an MTD wipe or a block device wipe.
#  $1 the device, such as /dev/sda3, /dev/ubi5_0
wipedev() {
  local dev="$1"
  if [ "${IS_MTD}" = "1" ]; then
    wipe_mtd_dev "${dev}"
  else
    wipe_block_dev "${dev}"
  fi
}

# Make sure the active kernel is still bootable after being wiped.
# The system may be in AU state that active kernel does not have "successful"
# bit set to 1 (only tries).
ensure_bootable_kernel() {
  local kernel_num="$1"
  local dst="$2"
  local active_flag="$(cgpt show -S -i "${kernel_num}" "${dst}")"
  local priority="$(cgpt show -P -i "${kernel_num}" "${dst}")"

  if [ "${active_flag}" -lt 1 ]; then
    cgpt add -i "${kernel_num}" "${dst}" -S 1
  fi
  if [ "${priority}" -lt 1 ]; then
    cgpt prioritize -i "${kernel_num}" "${dst}" -P 3
  fi
  sync
}

detect_system_config() {
  # Root devs are /dev/sda3, /dev/ubiblock5_0.
  # Kernel devs to go along with these are /dev/sda2, /dev/mtd4 respectively.

  # As we move factory wiping from release image to factory test image,
  # clobber-state will be invoked directly under a tmpfs. 'rootdev' cannot
  # report correct output under such a situation. Therefore, the output of
  # 'rootdev' is preserved then assigned to environment variables
  # ${ROOT_DEV}/${ROOT_DISK} for clobber-state. For other cases, the
  # environment variables will be empty and it fallbacks to use 'rootdev'.
  if [ -z "${ROOT_DEV}" ]; then
    ROOT_DEV=$(rootdev -s)
  fi
  if [ -z "${ROOT_DISK}" ]; then
    ROOT_DISK=$(rootdev -d -s)
  fi
  IS_MTD=0
  case "${ROOT_DISK}" in
    "/dev/ubi"*)
      # Special casing for NAND devices.
      IS_MTD=1
      ROOT_DISK="/dev/mtd0"
      STATE_DEV="/dev/ubi${PARTITION_NUM_STATE}_0"
      # On NAND, kernel is stored on /dev/mtdX.
      KERNEL_DEV=$(echo "${ROOT_DEV}" | tr \
          "${PARTITION_NUM_ROOT_A}${PARTITION_NUM_ROOT_B}" \
          "${PARTITION_NUM_KERN_A}${PARTITION_NUM_KERN_B}" | \
          sed -e 's/ubiblock\([0-9]*\)_0/mtd\1/')
      ;;
    *)
      STATE_DEV=${ROOT_DEV%[0-9]*}${PARTITION_NUM_STATE}
      KERNEL_DEV=$(echo "${ROOT_DEV}" | tr \
          "${PARTITION_NUM_ROOT_A}${PARTITION_NUM_ROOT_B}" \
          "${PARTITION_NUM_KERN_A}${PARTITION_NUM_KERN_B}")
      ;;
  esac
  OTHER_ROOT_DEV=$(echo "${ROOT_DEV}" | tr \
      "${PARTITION_NUM_ROOT_A}${PARTITION_NUM_ROOT_B}" \
      "${PARTITION_NUM_ROOT_B}${PARTITION_NUM_ROOT_A}")
  OTHER_KERNEL_DEV=$(echo "${KERNEL_DEV}" | tr \
      "${PARTITION_NUM_KERN_A}${PARTITION_NUM_KERN_B}" \
      "${PARTITION_NUM_KERN_B}${PARTITION_NUM_KERN_A}")
  KERNEL_PART_NUM=${KERNEL_DEV##[/a-z]*[/a-z]}
  WIPE_PART_NUM=${OTHER_ROOT_DEV##[/a-z]*[/a-z]}
  WIPE_PART_NUM=${WIPE_PART_NUM%_0}
  # Save the option before wipe-out
  WIPE_OPTION_FILE="${STATE_PATH}/factory_wipe_option"
  if [ -O ${WIPE_OPTION_FILE} ]; then
    WIPE_OPTION=$(cat ${WIPE_OPTION_FILE})
  else
    WIPE_OPTION=
  fi

  # Discover type of device holding the stateful partition; assume SSD.
  # Since there doesn't seem to be a good way to get from a partition name
  # to the base device name beyond simple heuristics, just find the device
  # with the same major number but with minor 0.
  ROTATIONAL=0
  MAJOR=$(stat -c %t ${STATE_DEV})
  for i in $(find /dev -type b); do
    if [ "$(stat -c %t:%T $i)" = "${MAJOR}:0" ]; then
      ROTATIONAL_PATH="/sys/block/$(basename ${i})/queue/rotational"
      if [ -r "${ROTATIONAL_PATH}" ]; then
        ROTATIONAL=$(cat "${ROTATIONAL_PATH}")
        break
      fi
    fi
  done

  # Sanity check root device partition number.
  if [ "$WIPE_PART_NUM" != "${PARTITION_NUM_ROOT_A}" ] && \
     [ "$WIPE_PART_NUM" != "${PARTITION_NUM_ROOT_B}" ]
  then
    echo "Invalid partition to wipe, $WIPE_PART_NUM (${OTHER_ROOT_DEV})"
    exit 1
  fi
}

wipe_rotational_drive() {
  # If the stateful filesystem is available, do some best-effort content
  # shredding. Since the filesystem is not mounted with "data=journal", the
  # writes really are overwriting the block contents (unlike on an SSD).
  if grep -q " ${STATE_PATH} " /proc/mounts ; then
    (
      # Directly remove things that are already encrypted (which are also the
      # large things), or are static from images.
      rm -rf "${STATE_PATH}/encrypted.block" \
             "${STATE_PATH}/var_overlay" \
             "${STATE_PATH}/dev_image"
      find "${STATE_PATH}/home/.shadow" -maxdepth 2 -type d \
                                        -name vault -print0 |
        xargs -r0 rm -rf
      # Shred everything else. We care about contents not filenames, so do not
      # use "-u" since metadata updates via fdatasync dominate the shred time.
      # Note that if the count-down is interrupted, the reset file continues
      # to exist, which correctly continues to indicate a needed wipe.
      find "${STATE_PATH}"/. -type f -print0 | xargs -r0 shred -fz
      sync
    ) &
  fi
  # Since the above rm/shred combo can be very fast, force a minimum of a 5
  # minute delay for this mode.
  delay=300
  ( stty -F "${TTY}" raw -echo -cread
    while [ "${delay}" -ge 0 ]; do
      printf "%2d:%02d\r" $(( delay / 60 )) $(( delay % 60 ))
      sleep 1
      : $(( delay -= 1 ))
    done
    echo
  ) >"${TTY}"
  wait
}

wipe_stateful_partition() {
  # Order constraints:
  # 1. The crossystem call must happen immediately before the destruction of the
  # stateful partition to minimize the probability of inconsistency between TPM
  # state and presence/absence of install attributes in case of power loss.
  # 2. The destruction of stateful must be last in the sequence in order to
  # minimize the probability of losing the preserved files.
  if [ "$FAST_WIPE" = "fast" ] && [ "${IS_MTD}" = "0" ]; then
    if [ -z "${KEEPIMG}" ]; then
      ensure_bootable_kernel "${KERNEL_PART_NUM}" "${ROOT_DISK}"
      dd bs=4M count=1 if=/dev/zero of=${OTHER_ROOT_DEV}
      dd bs=4M count=1 if=/dev/zero of=${OTHER_KERNEL_DEV}
    fi

    # Clear TPM at next boot to make the TPM state consistent with (the soon to be
    # induced) absence of install_attributes.pb.
    crossystem clear_tpm_owner_request=1

    # Just wipe the start of the partition and remake the fs on
    # the stateful partition.
    dd bs=4M count=1 if=/dev/zero of=${STATE_DEV}
  else
    if [ -z "${KEEPIMG}" ]; then
      ensure_bootable_kernel "${KERNEL_PART_NUM}" "${ROOT_DISK}"
      wipedev ${OTHER_ROOT_DEV}
      wipedev ${OTHER_KERNEL_DEV}
    fi

    # Clear TPM at next boot to make the TPM state consistent with (the soon to be
    # induced) absence of install_attributes.pb.
    crossystem clear_tpm_owner_request=1

    # Wipe everything on the stateful partition.
    wipedev ${STATE_DEV}
  fi
}

create_stateful_file_system() {
  if [ "${IS_MTD}" = "1" ]; then
    mkfs.ubifs -y -x none -R 0 "${STATE_DEV}"
  else
    mkfs.ext4 "${STATE_DEV}"
    # TODO(wad) tune2fs.
  fi
}

restore_preserved_files() {
  # If there were preserved files, restore them.
  if [ -n "${PRESERVED_LIST}" ]; then
    tar xfp ${PRESERVED_TAR} -C ${STATE_PATH}
    touch /mnt/stateful_partition/unencrypted/.powerwash_completed
  fi

  # Restore the log file
  clobber-log --restore "${SCRIPT}" "$@"
}

tag_developer_mode_flags() {
  # Tag that we're in developer mode otherwise we may get wiped again.
  if crossystem "devsw_boot?1" && ! crossystem "mainfw_act?recovery"; then
    touch /mnt/stateful_partition/.developer_mode
  fi
}

board_specific_wiping() {
  # Do any board specific wiping here.
  # board_factory_wipe.sh will be installed by the board overlay if necessary.
  if [ "$FACTORY_WIPE" = "factory" ]; then
    [ -n "${FACTORY_RETURN_AFTER_WIPING}" ] && exit
    BOARD_WIPE=/usr/sbin/board_factory_wipe.sh
    if [ -x "${BOARD_WIPE}" ]; then
      ${BOARD_WIPE} ${WIPE_OPTION}
    fi
  fi
}

main() {
  ensure_root_account
  redirect_log_start

  preserve_files
  detect_system_config

  # Preserve the log file
  clobber-log --preserve "${SCRIPT}" "$@"

  # On a non-fast wipe, rotational drives take too long. Override to run them
  # through "fast" mode, with a forced delay. Sensitive contents should already
  # be encrypted.
  if [ "${FAST_WIPE}" != "fast" ] && [ "${ROTATIONAL}" = "1" ]; then
    wipe_rotational_drive
    FAST_WIPE="fast"
  fi

  # Make sure the stateful partition has been unmounted.
  umount -n "${STATE_PATH}"

  wipe_stateful_partition
  create_stateful_file_system

  # Mount the fresh image for last minute additions.
  mount -n "${STATE_DEV}" "${STATE_PATH}"

  restore_preserved_files
  tag_developer_mode_flags

  # Flush linux caches.
  sync

  board_specific_wiping
  redirect_log_stop

  /sbin/shutdown -r now
  sleep 1d  # Wait for shutdown
}

main "$@"