Peter Gutmann writes:
For those of you who don't read sci.crypt, Robert Ames <gnome@istar.ca> has posted an article in which he observes that all WebTV posts contain an X-WebTV-Signature: line containing base64-encoded data. For samples of WebTV-generated posts, try the alt.weemba newsgroup, which is filled with WebTV-user drool (caution: remember to employ protection when exposting your mind to the content of the messages). Some samples:
X-WebTV-Signature: 1 ETAsAhQDqtur/jfleJ2CDOnNrVoeyALEQAIUOQyCBbzjx5HHfxeMERDgCjztXOU= X-WebTV-Signature: 1 ETAtAhUAmCCzQt+Tqt6fNX+L9+gDCECaqQkCFA0YCPz5tk85mUgq7iX/u4vWvOgG
Could someone grep their news spool for a few hundred of these signatures and post them here? We can do statistical analysis on them and determine whether they all appear to be modulo the same q value.
These decode into ASN.1-encoded DSA signatures, eg:
1 30 45: SEQUENCE { 3 02 21: INTEGER : 00 98 20 B3 42 DF 93 AA DE 9F 35 7F 8B F7 E8 03 : 08 40 9A A9 09 26 02 20: INTEGER : 0D 18 08 FC F9 B6 4F 39 99 48 2A EE 25 FF BB 8B : D6 BC E8 06 : }
for the second one. The key isn't included in the header, presumably the @webtv.net address can be tied to the hardware which contains some hardcoded DSA key. I wonder if WebTV users know they're signing each message they send?
If they are DSA signatures, they should all be mod q, where q is some 160 bit prime. Whether everyone uses the same key or different keys, they probably all share p and q. In that case the histogram of the values should be flat up to a cutoff point. We need to collect some hundreds of these values in order to distinguish them from random 160 bit values. The largest q value found in a dozen or so alt.weemba messages from dejanews started with 0xc0, so q must be at least this large if they are DSA sigs. This amount of data was not sufficient to distinguish from alternate theories, such as that each person has a random 160 bit q.