[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Mixmaster Security Issues


At 6:17 PM 8/30/95, hroller Mixmaster wrote:
>Apart from thwarting traffic analysis attacks, how does the security
>of a Mixmaster Type II remailer packet compare to that of a
>PGP-chained Type I message?

There is no way in which the security of Mixmaster messages is LESS than
that of type I (cypherpunk) remailers. 

>For example, is each remailer in the path limited to knowing only
>the next remailer in the path?  Is there any way for a remailer
>(except for the first and last in the chain) to know how many hops
>have already occurred or how many remain?  Is there a session key
>chosen via an RNG?  If so, how random is the RNG?  Is it seeded from
>a pseudo-random source that's at least as secure as measuring
>keystroke latencies, as PGP does?

Yes, each remailer is limited to knowing the previous and next

A Mixmaster remailer can only tell if it is first, last, or somewhere in
the middle. No information is leaked about position in the chain. There is
a hard limit of 20 hops. 

>Lance Cottrell's original "remailer essay" which proposed the Type
>II concept envisioned, if I'm not mistaken, the use of PGP
>technology to do the actual encryptions.  Now it seems that another,
>seemingly proprietary, implementation of RSAREF was used, instead.
>What was the reason for this change?

Version 1.0 (which was released but not widely used or promoted) used
PGPTools by Pr0duct Cypher. This is a library which provides hooks for most
of the major PGP routines. The main problem with PGPTools is that I could
not get it to compile on anything but a SUN. The other problem was that it
was difficult to control the encryption so I could avoid any change in the
size of information when it was encrypted. RSAREF is very portable, robust,
supported, easy to work with, and was easy to use for fine control of the
encryption process. RSAREF is also much less of a black box to me. I can
understand what it is doing in detail. 

>Would any security be lost if Type I and II technology were combined
>and a PGP-chained Type I packet were initially sent via Mixmaster?
>This would would seem to provide the necessary protection against
>traffic analysis while bypassing any *POSSIBLE* hidden weaknesses in
>Mixmaster.  IOW, if the outer Mixmaster "envelope" were "steamed
>open", perhasps based on some hidden weakness in Mixmaster, the
>inner, nested PGP envelope(s) would remain intact.

Because of the message size limitations there are some advantages to
sending the mixmaster chain through some type 1 remailers first, rather
than sending a type 1 message in a Mixmaster packet.

>BTW, what volume of message traffic is the Mixmaster network of
>remailers currently handling?  Is much cover traffic necessary to
>minimize delays while providing enough reordering to thwart traffic
>analysis?  (IOW, so a remailer with a reordering pool size of five
>messages, and averaging one REAL message a day, wouldn't have to
>keep a message for an average of five days before sending it on its
>next hop, as a worst-case scenario).

It is very difficult to know what fraction of the traffic I see is cover. I
generate some cover traffic my self, and I know some others do as well.
Right now a reordering pool of 5 messages results in a latency of about 30
min. Mixmaster is no longer a small fraction of the remailer market. A
majority of all public remailers support Mixmaster. 

>Is my math correct in surmising that chaining a message through five
>remailers, each with a reordering pool of five messages, could mean
>that the message eventually leaves the chain as one of 5^5 (3125)
>possible messages?  (My math is a bit weak, so please feel free to
>correct my methodology, if necessary.)  If so, does that work in
>reverse?  Could a given output message that finally surfaced in the
>clear be narrowed down to one of 3125 Mixmaster input messages
>through traffic analysis?  Or would the fact that the attacker
>didn't know the exact number of hops utilized significantly increase
>the odds against identifying the sender?  What effect, if any, would
>increasing the number of available remailers have on traffic

This is not quite correct, at each hop your message could have gone to any
remailer at all. There are now 16 Mixmaster remailers in operation. If you
have two good remailers in your chain (not run by the enemy), then a given
message into the system is probably one of the messages that emerges
between 10 Min and several hours later (with some complex probability
distribution over that time). Note that because of the way the reordering
is done, messages could stay in the pool forever, but this is exponentially
less likely with time.

It turns out that this is good security for one message, but is much less
secure if you continue to communicate with the same person for some time.
Then the attacker can look for correlations between your sending a message,
and everyone who receives them. After several messages in one month, you
will stand out, unless you send cover messages regularly, so you correlate
with everyone all the time (destroying any information about who you
actually correspond with). 

Version: 2.6.2


Lance Cottrell   [email protected]
PGP 2.6 key available by finger or server.
Mixmaster, the next generation remailer, is now available!
http://obscura.com/~loki/Welcome.html or FTP to obscura.com

"Love is a snowmobile racing across the tundra.  Suddenly
it flips over, pinning you underneath.  At night the ice
weasels come."