ext4 with journaling - restores (unwanted) old versions of files

Hi,

I'm working on BeagleBone Black, Angstrom system. I have a two partitions:

1. /dev/mmcblk0p2 - rootfs, mounted as read only
2. /dev/mmcblk0p3 - user data, mounter as read-write

Both are mounted with noatime,nodiratime,sync,data=journal,commit=1,barrier=1.

I am doing power failure test since two weeks: power supply is on and off in cycle from 1 s to 55 s (power is on for 1 s and cut off, power is on for 2 s and cut off, power is on for 3 s and cut off... up to 55 s and again from beginning). I have done about 10000 power failure cycles. File system were still consisent and data were not corrupted. Today a have done some modyfications in kernel, recompiled it, copied from host to BeagleBone, and reboot (shutdown -r now). I have run about 10 power failure cycles with new kernel and I noticed that after next boot BeagleBone run on previous version of kernel.

A noticed this occurance before when BeagleBone has changed IP address to previous without any human interference.

I guess that journaling system magically recover old files (or e2fsck running when Angstrom is booting).

Could you explain me is it possible and how to avoid situations like above?

Both are mounted with noatime,nodiratime,sync,data=journal,commit=1,barrier=1.

Should be set to Both are mounted with noatime,sync,data=journal,commit=1,barrier=1.

If the system is not shut down correctly, the journal is restored.

Should be set to Both are mounted with noatime,sync,data=journal,commit=1,barrier=1.

Of course you are right - nodiratime isn't required (I have checked it in kernel source fs/inode.c, function void touch_atime(struct path *path)).

After kernel update I have shutdown system properly (shutdown -r now) and than began test over again.
Angstrom booted with new kernel about 10 times, so I understand that journal didn't do any recovery.

After ten boot I saw that I have previous version of kernel. So I can't undestand why kernel was restored despite proper system shutdown and journal did it until ten boot time.

As I undestand journal cache should be emptied every 1 second (commit=1). So any recovery should be done after first system boot with new kernel, not after 10-th boot. Am I right?

The cache flush time should be selectable.
1s, 2s, 3s, 4s,...

Depends on how save the power supply is.