Re: mount ext2/3 as ext4 - no changes to ext2/3 fs structures?

[tytso@mit.edu]

On Sun, Jan 31, 2010 at 09:54:27PM -0800, Michael Rubin wrote:
> On Sun, Jan 31, 2010 at 2:31 PM, Thomas Kupper <thomas@kupper.org> wrote:
> > Consider the following scenario: Assume I have a partition
> > formated with ext2 on a computer with some 2.4.x kernel. Now I
> > mount this ext2 partition on another computer as ext4 using a
> > 2.6.29+ kernel. After that I take the disk back to the older
> > computer and mount the ext2 on the 2.4.x kernel with the ext2
> > driver.
> 
> I have seen this work as long as you don't add the "extents" or other
> ext4 specific mount options. Our kernels are not quite as up to date
> as 2.6.29 and we have not done extensive testing with that kernel. Not
> sure anyone has tested this case thoroughly.

Yep, it's a supported feature that mounting a file system previously
intended for ext2 or ext3 on ext4 shouldn't cause it to become
unmountable elsewhere.

Note however that the "extents" ext4 mount options has gone away; if
you want to enable extents on kernels, the supported way to do this
is:

tune2fs -O extents /dev/sda1

Once you do this (and you can do this on a live-mounted ext4
filesystem and it will take effect immediately), any new inodes that
are created will be extent-mapped.

We currently don't have a way of converting extent-mapped inodes back
to the legacy indirect block mapping scheme (a patch to do an off-line
conversion wouldn't be _that_ hard; patches gratefully accepted), so
this is currently a one-way conversion.

						- Ted