From: Damien Miller Newsgroups: isaac.lists.linux-ipsec Subject: linux-ipsec: audio-entropyd Date: 15 Aug 1999 07:32:12 -0700 Message-ID: With all this talk about /dev/random, I have decided to do a bit of hacking which should help mitigate kernel random number generator depletion attacks. audio-entropyd reads a user-defined amount data from the soundcard, takes the difference between the left and the right channels, hashes this into a (much) smaller user-defined buffer using SHA1 and credits a user-defined subset of this into the KRNG. It then sleeps for a user-defined period and starts the process again. The differencing between stereo channels should limit some external signals such as mains noise, etc. audio-entropyd performs a few other tricks to help prevent it getting fooled by regular signals. It can be configured to dump a lot of random bits into the KRNG at once - this should prevent the iterated guessing attack people have been talking about. This is an alpha release intended for people to review and find "improvement opportunities". Please do not use this on a production server. Consider this code guilty until proven innocent :) http://www.ilogic.com.au/~dmiller/files/audio-entropyd-0.0.0.tar.gz I am particulaly curious about what people consider a safe entropy estimation. My soundcard gets about 3 bits per byte (37.5%) entropy on differenced input as calculated by the 'ent' tool. The default estimate used in audio-entropyd is much more conservative - around 0.06%. The README file encourages users to base their settings on their own entropy measurements. Regards, Damien Miller -- | "Bombay is 250ms from New York in the new world order" - Alan Cox | Damien Miller - http://www.ilogic.com.au/~dmiller | Email: dmiller@ilogic.com.au (home) -or- damien@ibs.com.au (work)