Branches
Comments
[»]
Not quite as strong
by Richard Clark - Jan 27th 2005 12:27:31
You may have already considered this, but "resetting" the
encryption I assume means returning to the IV generated from the
passphrase. If you do this every 8k for example, you provide any attacker
with a large set of similarly produced ciphertexts. In addition, every 8k
block that is equal will encrypt to the same value allowing the attacker to
make inferences about the contents of the file from the prevalence of
particular encrypted results.
I'm not convinced either of these issues is a particularly big deal in
this case, but it might be worth noting somewhere prominent that block
ciphers are chained for these exact reasons, and that the user should
understand that the resulting encrypted file is not as strong as one
produced normally. I think it's more than fair to say (assuming you're
using a decent cipher :) that it is still plenty strong enough for regular
data, although I'd be worried about anything that someone might take a few
months to try and break.
[reply]
[top]
[»]
Re: Not quite as strong
by Alien Science - Jan 27th 2005 13:22:35
This is true. The repeated block issue hadn't occured to me either -- and
thats something that will never go away since its the very thing that makes
the files useful with rsync. Currently, there is a disclaimer at the bottom
of the manpage, however I'll add it to the homepage in the description of
murk's operation.
My only idea for getting round the weaknesses, in resetting the
encryption, is to have a different key for each block. What I haven't
worked out is how to generate these keys in a predicable way so that
different generations of a file can be efficiently rsynced. By predicable,
I mean a block of data always gets encrypted with the same key.
However, it is interesting you mention the importance of the iv being
reset to its original value. Would there be any mileage in resetting the iv
to, say, a checksum or digest of the plain text block? Indentical blocks
would encrypt identically but similar blocks would give away less about
their contents.
[reply]
[top]
|
