A minor fs extend brainwave....
Hi again...
I've just realised that the inode table, block bitmaps and inode bitmaps do not need to follow the superblock/group descriptors at the start of the group.
If they are moved to the end of the file system *BEFORE* the file system is put to use, their is no big problem with extending the file system later, and no large scale move of data (or inodes as I was first thinking) is required, only a maximum of 1 block of user data per group needs to be relocated (128K / GB), and the new group descriptors written in it's place.
I've already hacked up a program to move these tables around (on an empty file system), and the kernel and fsck like it as long as the inode table is not hard up against the end of the group (strange - must be an off by one error somewhere in fsck.ext2).
The short trailing groups at the end of a file system are not a problem, as by default mkfs.ext2 makes sure that they have 50 blocks of data in there, enough to extend a file system 12.5GB (greater than it's size at creation (can ext2 file systems be that big?).
This makes things a lot simpler....
Ideas/Comments?
Mike
_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com