Although the loop block device is created before attaching the image
to it, the devices for the partition that the image contains are still
not created. This patch creates those devices as well, when they are
not already available.
Fixes#482
Signed-off-by: Vasilis Tsiligiannis <acinonyx@openwrt.gr>
Format the root partition with the metadata_csum option
Format the boot partition with 4 sectors per cluster for a slight performance boost
Remove 'flush' mount option from the boot partition
When building on a native ARM platform, binfmt_misc is not
required to be loaded. This change checks the machine type
and if it's a ARM platform, skip the binfmt_misc validation.
* Tidy up and simplify function code
* Allow to pass extra arguments to debootstrap, e.g. `--variant minbase`
* Preserve compatibility with existing function calls
* Added Docker support
- replaced necessity for devicemapper (through kpartx) by using parted and
losetup with offsets
- added Dockerfile
- added dependency for parted and grep
- added hints to README.md
- common: loop through unmounts, fix shellcheck warnings
* stage2: use debconf instead of console-setup patch. Fixes#41
* export-noobs/prerun.sh: Use nested mountpoint
While it seems elegant and intuitive to use separate bootfs and rootfs
mountpoints for compressing the partitions, doing so violates a
precondition of unmount_image that they be mounted as a tree. This
causes the image to not be properly unmounted and detached. A better
solution might be to pack up the previous stage's chroot directory, but
that rework can wait for the time being.
scripts/common.sh: Output device name correctly
A misplaced ) in unmount_image caused the loop device to be incorrectly
identified, resulting in a fair bit of chaos trying to unmount other
filesystems on /dev/mapper devices. Such as / on a LUKS-encrypted
installation, for example. The unmount will fail as it should and
build.sh will abort the build without any cleanup. Best to avoid that.
These changes closeRPi-Distro/pi-gen#19
