Hello,

On Friday 24 August 2012 10:22:50, Artem Bityutskiy wrote:

On Tue, 2012-07-31 at 13:24 +0200, Alexander Stein wrote:

quoted

upon testing do_emergency_remount from a power fail interrupt (using the 
workqueue of course), we noticed UBIFS is not remounted read-only afterwards.
The current code in do_emergency_remount checks if the kernel actually needs 
to remount a filesystem using the following code:

quoted

if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
    !(sb->s_flags & MS_RDONLY)) {
        /*
         * What lock protects sb->s_flags??
         */
        do_remount_sb(sb, MS_RDONLY, NULL, 1);
}

I'm not in the details of this part of the kernel, but I suspect that sb-

quoted

s_bdev is NULL for UBIFS as it has no block device, but the character device 

/dev/ubiX_Y instead. I have the information from a collegue that removing the 
check for testing purposes for sb->s_bdev mounts UBIFS read-only.
Any comments/ideas how to remount UBIFS as read-only from 
do_emergency_remount?

I think this check is there to prevent file-systems like sysfs, tmpfs,
procfs, cgroup and debugfs from re-mounting R/O. If we just remove the
'sb->s_bdev' check - they all become R/O as well. For me it sounds OK,
but I've never used this functionality, so not sure.

I agree this check filters out file system without a backing storage (block device), which is fine. It just doesn't make a difference if those are remounted R/O.
I'm just wondering how to make the emergeny_remount usable with ubifs which has a backing storage, obviously, but no block device.

Regards,
Alexander