On a board, I would like to flash the (NAND) partition where my current rootfs is without restarting and being able to check if the flash succeeded.
To do so, I try to:
- kill every unnecessary processes
- mount a SD card containing another rootfs (a mount point to the current rootfs will be available on the SD card)
- run pivot_root between the rootfs of the SD card and the original one
- run chroot to use the SD card root.
- unmount the old root partition
- flash a new rootfs
- pivot_root back on the rootfs partition
Unfortunately, I cannot unmount the original root partition.
The beginning is basically the steps described in the pivot_root manual page:
mount /dev/hda1 /new-root
cd /new-root
pivot_root . old_root
exec chroot . sh <dev/console >dev/console 2>&1
umount /old-root # <== unable to unmount
Killing some processes:
ps
PID USER VSZ STAT COMMAND 1 root 1524 S init [3] 2 root 0 SW< [kthreadd] 3 root 0 SW< [ksoftirqd/0] 4 root 0 SW< [watchdog/0] 5 root 0 SW< [events/0] 6 root 0 SW< [khelper] 126 root 0 SW< [kblockd/0] 132 root 0 SW< [kseriod] 136 root 0 SW< [kmmcd] 159 root 0 SW [pdflush] 160 root 0 SW [pdflush] 161 root 0 SW< [kswapd0] 209 root 0 SW< [aio/0] 213 root 0 SW< [nfsiod] 220 root 0 SW< [cifsoplockd] 807 root 0 SW< [kapmd] 873 root 0 SW< [mtdblockd] 919 root 0 SW< [rpciod/0] 925 root 0 SWN [jffs2_gcd_mtd3] 927 root 0 SW< [mmcqd] 1806 root 2908 R -bash 2456 root 2072 R ps
mount /dev/mmcblk0p0 /mnt/disk
umount /sys
umount /tmp
cat /proc/mounts
rootfs / rootfs rw 0 0 /dev/root / jffs2 rw 0 0 /proc /proc proc rw 0 0 /dev/mmcblk0p0 /mnt/disk ext2 rw,errors=continue 0 0
I guess I have to find a way to unmount /dev/root
and /
or rootfs
but won't /mnt/disk
be a problem because it will still be in use but, it is the SD card I want to chroot to?
umount /proc
cd /mnt/disk
pivot_root . old-root
mount -t proc none /proc
ls -l /proc/1
-r-------- 1 root root 0 Nov 30 01:17 auxv --w------- 1 root root 0 Nov 30 01:17 clear_refs -r--r--r-- 1 root root 0 Nov 30 01:16 cmdline -rw-r--r-- 1 root root 0 Nov 30 01:17 coredump_filter lrwxrwxrwx 1 root root 0 Nov 30 01:17 cwd -> // -r-------- 1 root root 0 Nov 30 01:17 environ lrwxrwxrwx 1 root root 0 Nov 30 01:16 exe -> /old_root /sbin/init* dr-x------ 2 root root 0 Nov 30 01:17 fd/ dr-x------ 2 root root 0 Nov 30 01:17 fdinfo/ -r-------- 1 root root 0 Nov 30 01:17 limits -r--r--r-- 1 root root 0 Nov 30 01:17 maps -rw------- 1 root root 0 Nov 30 01:17 mem -r--r--r-- 1 root root 0 Nov 30 01:17 mountinfo -r--r--r-- 1 root root 0 Nov 30 01:17 mounts -r-------- 1 root root 0 Nov 30 01:17 mountstats dr-xr-xr-x 5 root root 0 Nov 30 01:17 net/ -rw-r--r-- 1 root root 0 Nov 30 01:17 oom_adj -r--r--r-- 1 root root 0 Nov 30 01:17 oom_score -r-------- 1 root root 0 Nov 30 01:17 pagemap -r-------- 1 root root 0 Nov 30 01:17 personality lrwxrwxrwx 1 root root 0 Nov 30 01:17 root -> // -rw-r--r-- 1 root root 0 Nov 30 01:17 sched -r--r--r-- 1 root root 0 Nov 30 01:17 smaps -r--r--r-- 1 root root 0 Nov 30 01:16 stat -r--r--r-- 1 root root 0 Nov 30 01:17 statm -r--r--r-- 1 root root 0 Nov 30 01:17 status dr-xr-xr-x 3 root root 0 Nov 30 01:17 task/ -r--r--r-- 1 root root 0 Nov 30 01:17 wchan
cat /proc/mounts
rootfs / rootfs rw 0 0 /dev/root /old_root jffs2 rw 0 0 /dev/mmcblk0p0 / ext2 rw,errors=continue 0 0 none /proc proc rw 0 0
exec chroot . sh <dev/console >dev/console 2>&1
umount /old_root
umount: can't umount /old_root/: Device or resource busy
fuser -m /old_root/
#> 1 # <= issue here
I would like to know if I forgot something (maybe my strategy is simply broken?)
I tried with chroot / switch_root
.
I also tried to mount --move /proc
and /sys
instead of using the SD card rootfs.
Kernel version: 2.6.29.4 ← a rootfs exists so, is it possible to umount the root?
Busybox version: 1.16.1
Best Answer
I believe, you are executing the command
umount /old_root
still from the old root, and therefore it is busy.I once did a similar script, and the following worked for me:
then, inside the new root, the first command the new init executes is
umount /mnt/tmp/
.