Re: [RFC] reserved inode fields to use for i_blocks_high

On Mon, Jun 05, 2006 at 02:02:53PM -0600, Andreas Dilger wrote:
> The current "large inode" support is existing since 2.5.44 kernels (Oct
> 2002) and 2.4.24 kernels (Jan 2004) so changing e2fsprogs to format
> filesystems with 256-byte inodes by default would be quite widely
> compatible. It becomes less backward compatible if you start storing
> EAs in the large inode space, 2.6.10 = Nov 2004, not sure about 2.4.

I was just looking at that.  Unfortunately if EA's are enabled, and
you mount a filesystem with 256+ byte inodes, the space starts getting
used automatically.  I believe the result if the filesystem is then
mounted on a pre-2.6.10 system is that the newly added EA's will not
be visible; not catastrophically, bad, but it still could be quite
problematic if (for example) someone were to mount a filesystem under
a modern Fedora system, and then remount it on a RHEL4 system such
that SELINUX information disappears.  

The EA-in-inode code should have been done with a mount option or
re-compat feature-flag (the latter being preferable) to enable or
disable ea-in-inode.  Water under the bridge, sigh....  The one good
thing is that there should be very few systems created on older
distro's with 256+ inodes, but it's still a rather large misfeature.

> Using Ted's new "/etc/mke2fs.opt" (or whatever) file if we don't want
> to change it in the default mke2fs would be one way to do this for a
> distro without enabling it for all e2fsprogs users.

We don't have inode size as a parameter that can be controlled via
/etc/mke2fs.conf yet, but I agree it's a good thing to add.

						- Ted