Allow booting from an image file (/data/postmarketOS.img) instead of flashing to a partition

Created by: ollieparanoid

@NotKit wrote in #postmarketOS:

can something like this be accepted to allow using /data/postmarketOS.img instead of flashing pmOS rootfs to /system partition?

Further discussion with @drebrez was about automatically resizing the image to the full partition width or not, and that it should be made optional (it is not desired in the Android multiboot scenario).

diff --git a/aports/main/postmarketos-mkinitfs/init.sh.in b/aports/main/postmarketos-mkinitfs/init.sh.in
index d0af68c..c4e3c96 100644
--- a/aports/main/postmarketos-mkinitfs/init.sh.in
+++ b/aports/main/postmarketos-mkinitfs/init.sh.in
@@ -13,6 +13,7 @@ setup_log
 # shellcheck disable=SC2154,SC2086
 [ -d /lib/modules ] && modprobe -a ${deviceinfo_modules_initfs} ext4
 setup_mdev
+mount_loopback_device
 mount_subpartitions
 
 # Start msm-fb-refresher workaround for qualcomm devices
diff --git a/aports/main/postmarketos-mkinitfs/init_functions.sh b/aports/main/postmarketos-mkinitfs/init_functions.sh
index a3564dd..5647f47 100644
--- a/aports/main/postmarketos-mkinitfs/init_functions.sh
+++ b/aports/main/postmarketos-mkinitfs/init_functions.sh
@@ -38,6 +38,21 @@ setup_mdev() {
 	mdev -s
 }
 
+mount_loopback_device() {
+	if [ ! -z "$deviceinfo_data_partition" ]; then
+		mkdir /data
+		mount $deviceinfo_data_partition /data
+
+		loopback_img=/data/postmarketOS.img
+
+		if [ -f "$loopback_img" ]; then
+			loopback_device=$(losetup -f)
+			losetup $loopback_device $loopback_img
+			kpartx -afs $loopback_device
+		fi
+	fi
+}
+
 mount_subpartitions() {
 	# Do not create subpartition mappings if pmOS_boot
 	# already exists (e.g. installed on an sdcard)

---

Yes please! in fact a similar idea was proposed here:

[...] have one big ext4/yaffs partition, and include the whole system image file inside that. Then we would adjust the initramfs code to find the boot and root partitions inside that, and teach it how to resize the whole thing to the maximum partition size. We already did some experiments, and so far everything worked (talk to drebrez or ollieparanoid).

Created by: mirh

Take notice that for encryption to work (also performance, and transparent compression!) the only choice for MTD devices is UBIFS.

By * postmarketOS Bot on 2018-03-22T23:14:57

A bit more sophisticated approach to that function to autodetect possible partitions. The issue it has is that blkid doesn't list partition table labels and we don't have udev running in initramfs to check them. Suggestions welcomed.

mount_loopback_device() {
	loopback_img=postmarketOS.img
	labels="userdata linux"

	mkdir /tmpmnt
	parts=$(blkid)

	for part in /dev/mmcblk*; do
		for label in $labels; do
			if echo "$parts" | grep $part | grep -q $label; then
				mount -o ro $part /tmpmnt

				if [ -f "/tmpmnt/$loopback_img" ]; then
					mount -o remount,rw /tmpmnt
					loopback_device=$(losetup -f)
					losetup $loopback_device "/tmpmnt/$loopback_img"
					kpartx -afs $loopback_device
					return 0
				fi

				umount /tmpmnt
			fi
		done
	done
}

By NotKit on 2018-05-07T19:32:42

@NotKit what about trying all the available partitions? it's too slow?

By Daniele Debernardi on 2018-05-07T22:47:13

@drebrez not sure, but I would like to avoid mounting all the possible partitions as not all of them are even supposed to be mounted by system, though mounting r/o and checking for existence of file shouldn't break things.

By NotKit on 2018-05-10T12:36:21

Created by: ollieparanoid

@NotKit: looks good to me!

By * postmarketOS Bot on 2018-05-15T21:58:45

We can consider implementing this feature using an hook which is called before the mount_subpartitions, we should add this use case in #1607 (closed) (initramfs: allow hooks in multiple places)

By Daniele Debernardi on 2018-08-09T23:18:25

mentioned in issue #1743 (closed)

By Daniele Debernardi on 2019-07-23T22:50:58

closed

changed the description

By Robert Eckelmann on 2024-08-16T01:05:44