From mboxrd@z Thu Jan  1 00:00:00 1970
From: "Nigel Roles" <nigel@9fs.org>
To: "Fans of Plan9" <9fans@cse.psu.edu>
Date: Mon,  4 Oct 2004 19:16:23 +0100
Message-ID: <HOEHIDJJJINMLFPOLFJCAEMBDFAA.nigel@9fs.org>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
Subject: [9fans] OT: puzzle
Topicbox-Message-UUID: eca6d87c-eacd-11e9-9e20-41e7f4b1d025

I've been struggling with this one for a bit, so I
thought I'd consult the combined brains of the 9fans list.

Imagine I want to reverse engineer the disk format used
by a hardware raid 5 controller. I plug 3 disks into
a controller, build the array, and then write a test
pattern to the logical file system. I then remove the
raid controller, plug the disks direct into the motherboard
and examine the layout on each disk.

Because of the way raid 5 works (in theory), I expect to
see a mixture of plaintext blocks, blocks which are
xors of the other pairs of blocks, and management blocks
containing proprietary stuff I may well not decipher.

What pattern should I write to each block so that I can
tell

1) whether a block is an xor, plaintext, or management block
2) for plaintext ones, which logical block it holds
3) for xor ones, which two plaintext blocks it is the xor of

1) seems easy as if there is a fixed pattern in every block,
it will appear as nulls in any xor block. So long as the
pattern is reasonably long, it is unlikely to accidentally
match a management block.

2) is simply a matter of writing a block number into the
block as well as the fixed pattern

3) has me stumped.

All suggestions welcome, including "you can't do it".

Nigel