Re: [PATCH] ext4: guarantee already started handles to successfully finish while ro remounting

[Daeho Jeong <daeho.jeong@samsung.com>]

> The real problem here is that we want emergency unmount to be
> completely safe, butting MS_RDONLY randomly isn't safe against file
> system corruption.  The idea is you do this only when you have no
> other choice, and the consequences would be worse --- and where you
> would be prepared to do a file system consistency check afterwards.

> We can either fix Android userspace, or we could add a per-file system
> callback which tries (as much as possible) to make an emergency
> unmount safe.  In this particular case, it would probably involve
> setting the "file system is corrupt flag" to force a file system
> consistency check at the next reboot.  Because if you use emergency
> unmount, all bets are off, and there may be other problems....

Actually, we had executed power on/off test repeatedly with 10 Android
devices for a month. But, during the test, just this problem only happened
repeatedly, even if it occurred very rarely. So we had concluded that we
had to fix this problem certainly.

However, I can see the point now. Android have misused the emergency
ro-remount and the filesystem crash by emergency ro-remount is not an issue.
I can understand the purpose of the emergency ro-remount.

I heard that the next version of Android will do full scan of ext4 filesystems
using e2fsck every boot-up, so the existing problem will be naturally resolved,
even if it might not be the right solution.

Thank you for your valuable comments.