openwrt/package/base-files/files/lib/upgrade/nand.sh

475 lines
13 KiB
Bash
Raw Normal View History

# Copyright (C) 2014 OpenWrt.org
#
. /lib/functions.sh
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
# 'kernel' partition or UBI volume on NAND contains the kernel
CI_KERNPART="${CI_KERNPART:-kernel}"
# 'ubi' partition on NAND contains UBI
# There are also CI_KERN_UBIPART and CI_ROOT_UBIPART if kernel
# and rootfs are on separated UBIs.
CI_UBIPART="${CI_UBIPART:-ubi}"
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
# 'rootfs' UBI volume on NAND contains the rootfs
CI_ROOTPART="${CI_ROOTPART:-rootfs}"
ubi_mknod() {
local dir="$1"
local dev="/dev/$(basename $dir)"
[ -e "$dev" ] && return 0
local devid="$(cat $dir/dev)"
local major="${devid%%:*}"
local minor="${devid##*:}"
mknod "$dev" c $major $minor
}
nand_find_volume() {
local ubidevdir ubivoldir
ubidevdir="/sys/class/ubi/"
[ ! -d "$ubidevdir" ] && return 1
for ubivoldir in $ubidevdir/${1}_*; do
[ ! -d "$ubivoldir" ] && continue
if [ "$( cat $ubivoldir/name )" = "$2" ]; then
basename $ubivoldir
ubi_mknod "$ubivoldir"
return 0
fi
done
}
nand_find_ubi() {
local ubidevdir ubidev mtdnum cmtdnum
mtdnum="$( find_mtd_index $1 )"
[ ! "$mtdnum" ] && return 1
for ubidevdir in /sys/class/ubi/ubi*; do
[ ! -e "$ubidevdir/mtd_num" ] && continue
cmtdnum="$( cat $ubidevdir/mtd_num )"
if [ "$mtdnum" = "$cmtdnum" ]; then
ubidev=$( basename $ubidevdir )
ubi_mknod "$ubidevdir"
echo $ubidev
return 0
fi
done
}
nand_get_magic_long() {
(${3}cat "$1" | dd bs=4 "skip=${2:-0}" count=1 | hexdump -v -n 4 -e '1/1 "%02x"') 2> /dev/null
}
get_magic_long_tar() {
(tar xO${3}f "$1" "$2" | dd bs=4 count=1 | hexdump -v -n 4 -e '1/1 "%02x"') 2> /dev/null
}
identify() {
identify_magic_long $(nand_get_magic_long "$@")
}
identify_tar() {
identify_magic_long $(get_magic_long_tar "$@")
}
identify_if_gzip() {
if [ "$(identify "$1")" = gzip ]; then echo -n z; fi
}
nand_restore_config() {
local ubidev=$( nand_find_ubi "${CI_ROOT_UBIPART:-$CI_UBIPART}" )
local ubivol="$( nand_find_volume $ubidev rootfs_data )"
if [ ! "$ubivol" ]; then
ubivol="$( nand_find_volume $ubidev "$CI_ROOTPART" )"
if [ ! "$ubivol" ]; then
echo "cannot find ubifs data volume"
return 1
fi
fi
mkdir /tmp/new_root
if ! mount -t ubifs /dev/$ubivol /tmp/new_root; then
echo "cannot mount ubifs volume $ubivol"
rmdir /tmp/new_root
return 1
fi
if mv "$1" "/tmp/new_root/$BACKUP_FILE"; then
if umount /tmp/new_root; then
echo "configuration saved"
rmdir /tmp/new_root
return 0
fi
else
umount /tmp/new_root
fi
echo "could not save configuration to ubifs volume $ubivol"
rmdir /tmp/new_root
return 1
}
nand_remove_ubiblock() {
local ubivol="$1"
local ubiblk="ubiblock${ubivol:3}"
if [ -e "/dev/$ubiblk" ]; then
umount "/dev/$ubiblk" && echo "unmounted /dev/$ubiblk" || :
if ! ubiblock -r "/dev/$ubivol"; then
echo "cannot remove $ubiblk"
return 1
fi
fi
}
nand_attach_ubi() {
local ubipart="$1"
local has_env="${2:-0}"
local mtdnum="$( find_mtd_index "$ubipart" )"
if [ ! "$mtdnum" ]; then
>&2 echo "cannot find ubi mtd partition $ubipart"
return 1
fi
local ubidev="$( nand_find_ubi "$ubipart" )"
if [ ! "$ubidev" ]; then
>&2 ubiattach -m "$mtdnum"
ubidev="$( nand_find_ubi "$ubipart" )"
if [ ! "$ubidev" ]; then
>&2 ubiformat /dev/mtd$mtdnum -y
>&2 ubiattach -m "$mtdnum"
ubidev="$( nand_find_ubi "$ubipart" )"
if [ ! "$ubidev" ]; then
>&2 echo "cannot attach ubi mtd partition $ubipart"
return 1
fi
if [ "$has_env" -gt 0 ]; then
>&2 ubimkvol /dev/$ubidev -n 0 -N ubootenv -s 1MiB
>&2 ubimkvol /dev/$ubidev -n 1 -N ubootenv2 -s 1MiB
fi
fi
fi
echo "$ubidev"
return 0
}
nand_detach_ubi() {
local ubipart="$1"
local mtdnum="$( find_mtd_index "$ubipart" )"
if [ ! "$mtdnum" ]; then
echo "cannot find ubi mtd partition $ubipart"
return 1
fi
local ubidev="$( nand_find_ubi "$ubipart" )"
if [ "$ubidev" ]; then
for ubivol in $(find /dev -name "${ubidev}_*" -maxdepth 1 | sort); do
ubivol="${ubivol:5}"
nand_remove_ubiblock "$ubivol" || :
umount "/dev/$ubivol" && echo "unmounted /dev/$ubivol" || :
done
if ! ubidetach -m "$mtdnum"; then
echo "cannot detach ubi mtd partition $ubipart"
return 1
fi
fi
}
nand_upgrade_prepare_ubi() {
local rootfs_length="$1"
local rootfs_type="$2"
local rootfs_data_max="$(fw_printenv -n rootfs_data_max 2> /dev/null)"
[ -n "$rootfs_data_max" ] && rootfs_data_max=$((rootfs_data_max))
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
local kernel_length="$3"
local has_env="${4:-0}"
local kern_ubidev
local root_ubidev
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
[ -n "$rootfs_length" -o -n "$kernel_length" ] || return 1
if [ -n "$CI_KERN_UBIPART" -a -n "$CI_ROOT_UBIPART" ]; then
kern_ubidev="$( nand_attach_ubi "$CI_KERN_UBIPART" "$has_env" )"
[ -n "$kern_ubidev" ] || return 1
root_ubidev="$( nand_attach_ubi "$CI_ROOT_UBIPART" )"
[ -n "$root_ubidev" ] || return 1
else
kern_ubidev="$( nand_attach_ubi "$CI_UBIPART" "$has_env" )"
[ -n "$kern_ubidev" ] || return 1
root_ubidev="$kern_ubidev"
fi
local kern_ubivol="$( nand_find_volume $kern_ubidev "$CI_KERNPART" )"
local root_ubivol="$( nand_find_volume $root_ubidev "$CI_ROOTPART" )"
local data_ubivol="$( nand_find_volume $root_ubidev rootfs_data )"
[ "$root_ubivol" = "$kern_ubivol" ] && root_ubivol=
# remove ubiblocks
[ "$kern_ubivol" ] && { nand_remove_ubiblock $kern_ubivol || return 1; }
[ "$root_ubivol" ] && { nand_remove_ubiblock $root_ubivol || return 1; }
[ "$data_ubivol" ] && { nand_remove_ubiblock $data_ubivol || return 1; }
# kill volumes
[ "$kern_ubivol" ] && ubirmvol /dev/$kern_ubidev -N "$CI_KERNPART" || :
[ "$root_ubivol" ] && ubirmvol /dev/$root_ubidev -N "$CI_ROOTPART" || :
[ "$data_ubivol" ] && ubirmvol /dev/$root_ubidev -N rootfs_data || :
# create kernel vol
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
if [ -n "$kernel_length" ]; then
if ! ubimkvol /dev/$kern_ubidev -N "$CI_KERNPART" -s $kernel_length; then
echo "cannot create kernel volume"
return 1;
fi
fi
# create rootfs vol
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
if [ -n "$rootfs_length" ]; then
local rootfs_size_param
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
if [ "$rootfs_type" = "ubifs" ]; then
rootfs_size_param="-m"
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
else
rootfs_size_param="-s $rootfs_length"
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
fi
if ! ubimkvol /dev/$root_ubidev -N "$CI_ROOTPART" $rootfs_size_param; then
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
echo "cannot create rootfs volume"
return 1;
fi
fi
# create rootfs_data vol for non-ubifs rootfs
if [ "$rootfs_type" != "ubifs" ]; then
local rootfs_data_size_param="-m"
if [ -n "$rootfs_data_max" ]; then
rootfs_data_size_param="-s $rootfs_data_max"
fi
if ! ubimkvol /dev/$root_ubidev -N rootfs_data $rootfs_data_size_param; then
if ! ubimkvol /dev/$root_ubidev -N rootfs_data -m; then
echo "cannot initialize rootfs_data volume"
return 1
fi
fi
fi
return 0
}
# Write the UBI image to MTD ubi partition
nand_upgrade_ubinized() {
local ubi_file="$1"
local gz="$2"
local ubi_length=$( (${gz}cat "$ubi_file" | wc -c) 2> /dev/null)
nand_detach_ubi "$CI_UBIPART" || return 1
local mtdnum="$( find_mtd_index "$CI_UBIPART" )"
${gz}cat "$ubi_file" | ubiformat "/dev/mtd$mtdnum" -S "$ubi_length" -y -f - && ubiattach -m "$mtdnum"
}
# Write the UBIFS image to UBI rootfs volume
nand_upgrade_ubifs() {
local ubifs_file="$1"
local gz="$2"
local ubifs_length=$( (${gz}cat "$ubifs_file" | wc -c) 2> /dev/null)
nand_upgrade_prepare_ubi "$ubifs_length" "ubifs" "" "" || return 1
local ubidev="$( nand_find_ubi "$CI_UBIPART" )"
local root_ubivol="$(nand_find_volume $ubidev "$CI_ROOTPART")"
${gz}cat "$ubifs_file" | ubiupdatevol /dev/$root_ubivol -s "$ubifs_length" -
}
# Write the FIT image to UBI kernel volume
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
nand_upgrade_fit() {
local fit_file="$1"
local gz="$2"
local fit_length=$( (${gz}cat "$fit_file" | wc -c) 2> /dev/null)
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
nand_upgrade_prepare_ubi "" "" "$fit_length" "1" || return 1
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
local fit_ubidev="$(nand_find_ubi "$CI_UBIPART")"
local fit_ubivol="$(nand_find_volume $fit_ubidev "$CI_KERNPART")"
${gz}cat "$fit_file" | ubiupdatevol /dev/$fit_ubivol -s "$fit_length" -
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
}
# Write images in the TAR file to MTD partitions and/or UBI volumes as required
nand_upgrade_tar() {
local tar_file="$1"
local gz="$2"
# WARNING: This fails if tar contains more than one 'sysupgrade-*' directory.
local board_dir="$(tar t${gz}f "$tar_file" | grep -m 1 '^sysupgrade-.*/$')"
board_dir="${board_dir%/}"
local kernel_mtd kernel_length
if [ "$CI_KERNPART" != "none" ]; then
kernel_mtd="$(find_mtd_index "$CI_KERNPART")"
kernel_length=$( (tar xO${gz}f "$tar_file" "$board_dir/kernel" | wc -c) 2> /dev/null)
[ "$kernel_length" = 0 ] && kernel_length=
fi
local rootfs_length=$( (tar xO${gz}f "$tar_file" "$board_dir/root" | wc -c) 2> /dev/null)
[ "$rootfs_length" = 0 ] && rootfs_length=
image: add support for building FIT image with filesystem Allow for single (external-data) FIT image to hold kernel, dtb and squashfs. In that way, the bootloader verifies the system integrity including the rootfs, because what's the point of checking that the hash of the kernel is correct if it won't boot in case of squashfs being corrupted? Better allow bootloader to check everything needed to make it at least up to failsafe mode. As a positive side effect this change also makes the sysupgrade process on nand potentially much easier as it is now. In short: mkimage has a parameter '-E' which allows generating FIT images with 'external' data rather than embedding the data into the device-tree blob itself. In this way, the FIT structure itself remains small and can be parsed easily (rather than having to page around megabytes of image content). This patch makes use of that and adds support for adding sub-images of type 'filesystem' which are used to store the squashfs. Now U-Boot can verify the whole OS and the new partition parsers added in the Linux kernel can detect the filesystem sub-images, create partitions for them, and select the active rootfs volume based on the configuration in FIT (passing configuration via device tree could be implemented easily at a later stage). This new FIT partition parser works for NOR flash (on top of mtdblock), NAND flash (on top of ubiblock) as well as classic block devices (ie. eMMC, SDcard, SATA, NVME, ...). It could even be used to mount such FIT images via `losetup -P` on a user PC if this patch gets included in Linux upstream one day ;) Signed-off-by: John Crispin <john@phrozen.org> Signed-off-by: Daniel Golle <daniel@makrotopia.org>
2021-02-15 14:37:17 +00:00
local rootfs_type
[ "$rootfs_length" ] && rootfs_type="$(identify_tar "$tar_file" "$board_dir/root" "$gz")"
local ubi_kernel_length
if [ "$kernel_length" ]; then
if [ "$kernel_mtd" ]; then
# On some devices, the raw kernel and ubi partitions overlap.
# These devices brick if the kernel partition is erased.
# Hence only invalidate kernel for now.
dd if=/dev/zero bs=4096 count=1 2> /dev/null | \
mtd write - "$CI_KERNPART"
else
ubi_kernel_length="$kernel_length"
fi
fi
local has_env=0
nand_upgrade_prepare_ubi "$rootfs_length" "$rootfs_type" "$ubi_kernel_length" "$has_env" || return 1
if [ "$rootfs_length" ]; then
local ubidev="$( nand_find_ubi "${CI_ROOT_UBIPART:-$CI_UBIPART}" )"
local root_ubivol="$( nand_find_volume $ubidev "$CI_ROOTPART" )"
tar xO${gz}f "$tar_file" "$board_dir/root" | \
ubiupdatevol /dev/$root_ubivol -s "$rootfs_length" -
fi
if [ "$kernel_length" ]; then
if [ "$kernel_mtd" ]; then
tar xO${gz}f "$tar_file" "$board_dir/kernel" | \
mtd write - "$CI_KERNPART"
else
local ubidev="$( nand_find_ubi "${CI_KERN_UBIPART:-$CI_UBIPART}" )"
local kern_ubivol="$( nand_find_volume $ubidev "$CI_KERNPART" )"
tar xO${gz}f "$tar_file" "$board_dir/kernel" | \
ubiupdatevol /dev/$kern_ubivol -s "$kernel_length" -
fi
fi
return 0
}
nand_verify_if_gzip_file() {
local file="$1"
local gz="$2"
if [ "$gz" = z ]; then
echo "verifying compressed sysupgrade file integrity"
if ! gzip -t "$file"; then
echo "corrupted compressed sysupgrade file"
return 1
fi
fi
}
nand_verify_tar_file() {
local file="$1"
local gz="$2"
echo "verifying sysupgrade tar file integrity"
if ! tar xO${gz}f "$file" > /dev/null; then
echo "corrupted sysupgrade tar file"
return 1
fi
}
nand_do_flash_file() {
local file="$1"
local gz="$(identify_if_gzip "$file")"
local file_type="$(identify "$file" "" "$gz")"
[ ! "$(find_mtd_index "$CI_UBIPART")" ] && CI_UBIPART=rootfs
case "$file_type" in
"fit")
nand_verify_if_gzip_file "$file" "$gz" || return 1
nand_upgrade_fit "$file" "$gz"
;;
"ubi")
nand_verify_if_gzip_file "$file" "$gz" || return 1
nand_upgrade_ubinized "$file" "$gz"
;;
"ubifs")
nand_verify_if_gzip_file "$file" "$gz" || return 1
nand_upgrade_ubifs "$file" "$gz"
;;
*)
nand_verify_tar_file "$file" "$gz" || return 1
nand_upgrade_tar "$file" "$gz"
;;
esac
}
nand_do_restore_config() {
local conf_tar="/tmp/sysupgrade.tgz"
[ ! -f "$conf_tar" ] || nand_restore_config "$conf_tar"
}
# Recognize type of passed file and start the upgrade process
nand_do_upgrade() {
local file="$1"
sync
nand_do_flash_file "$file" && nand_do_upgrade_success
nand_do_upgrade_failed
}
nand_do_upgrade_success() {
if nand_do_restore_config && sync; then
echo "sysupgrade successful"
umount -a
reboot -f
fi
nand_do_upgrade_failed
}
nand_do_upgrade_failed() {
sync
echo "sysupgrade failed"
# Should we reboot or bring up some failsafe mode instead?
umount -a
reboot -f
}
# Check if passed file is a valid one for NAND sysupgrade.
# Currently it accepts 4 types of files:
# 1) UBI: a ubinized image containing required UBI volumes.
# 2) UBIFS: a UBIFS rootfs volume image.
# 3) FIT: a FIT image containing kernel and rootfs.
# 4) TAR: an archive that includes directory "sysupgrade-${BOARD_NAME}" containing
# a non-empty "CONTROL" file and required partition and/or volume images.
#
# You usually want to call this function in platform_check_image.
#
# $(1): board name, used in case of passing TAR file
# $(2): file to be checked
nand_do_platform_check() {
local board_name="$1"
local file="$2"
local gz="$(identify_if_gzip "$file")"
local file_type="$(identify "$file" "" "$gz")"
local control_length=$( (tar xO${gz}f "$file" "sysupgrade-${board_name//,/_}/CONTROL" | wc -c) 2> /dev/null)
if [ "$control_length" = 0 ]; then
control_length=$( (tar xO${gz}f "$file" "sysupgrade-${board_name//_/,}/CONTROL" | wc -c) 2> /dev/null)
fi
if [ "$control_length" != 0 ]; then
nand_verify_tar_file "$file" "$gz" || return 1
else
nand_verify_if_gzip_file "$file" "$gz" || return 1
if [ "$file_type" != "fit" -a "$file_type" != "ubi" -a "$file_type" != "ubifs" ]; then
echo "invalid sysupgrade file"
return 1
fi
fi
return 0
}