spf-discuss
[Top] [All Lists]

Re: ANNOUNCE: SRS v0.15 documentation and code

2004-02-11 10:01:24
In 
<Pine(_dot_)LNX(_dot_)4(_dot_)53(_dot_)0402111647010(_dot_)29116(_at_)astray(_dot_)com>
 Shevek <spf(_at_)anarres(_dot_)org> writes:

On Wed, 11 Feb 2004, wayne wrote:

In 
<Pine(_dot_)LNX(_dot_)4(_dot_)53(_dot_)0402110952540(_dot_)29116(_at_)astray(_dot_)com>
 Shevek <spf(_at_)anarres(_dot_)org> writes:

As Daniel Roethlisberger points out, you really don't need a complete
MD5 hash.  My gut feel is that even 16 bits of the hash would prevent
any useful spoofing of the SRS system, and I think it would be very
useful to calculate a lower bound.  Then, you can use base32 encoding
or something more resilient to strange MTA manglings.

I've been thinking about this a little bit.

All we need to do is make it impractical to use SRS to create open
relays, not impossible.  As such, I think that 15-20 bits is probably
more than enough.

15 to 20 bits doesn't provide any pretence of security. It can probably be 
cracked in a few milliseconds on an average PC. Since spammers are already 
using lots of virus-infected slave machines to perform their task, this 
much CPU is negligible.

I don't understand this.  I'm talking about using 15-20 bits from the
hash output, not 15-20 bit private keys.  I see no way a spammer could
"crack" SRS with 15-20 bits of output.


Speed is an issue in MTAs.  Could we save a significant amount of CPU
time by using AES in CBC mode vs MD5?  Could we use RC4?

AES is an encryption scheme, not a message authentication scheme.  

All encryption systems can be converted into hash functions.  Unix has
long used this method for storing secure hashes of passwords in
/etc/passwd.



-wayne