Subject: Re: 2.0 and >2T filesystems
To: der Mouse <email@example.com>
From: Richard Earnshaw <Richard.Earnshaw@buzzard.freeserve.co.uk>
Date: 11/24/2005 00:33:31
On Wed, 23 Nov 2005 14:47:05 EST, der Mouse wrote:
> My tests show that ffs1, at least, is broken with >2T filesystems. I
> know ffs1 has on-disk 32-bit values in places like the dblock and
> iblock arrays, but I specifically built the filesystem 4K frags and 32K
> blocks, and 2^32 frags of 4K each is 16T, while the filesystem is well
> under 3T, so those on-disk values are within even *signed* 32-bit.
> But when I wrote about 2.5T of nonrepeating but easy-to-recompute data,
> I got corrupt data back on read. Then, like an idiot, I tried to rm
> -rf stuff, and the machine promptly panicked, presumably because some
> inode block got stomped on with data.
IIRC, since the sector size is 512 bytes, 2T is 2^32 sectors. So the
inodes can't address more than that in ffs1.