Back to files

Readme too

Updates should be found at http://www.vanheusden.com/mirrors/ All changes by Folkert van Heusden & Udo van den Heuvel

v0.0.1
I noticed that my firewall's kernel, running audio-entropyd, logged 'Sound: Recording overrun' although the audio-entropyd was sleeping. After some experimenting I found out that when the dsp_fd was closed during the sleep the errors went away.

v0.0.2
The monotonous logging statements annoyed me a bit so I added the option to send either a SIGUSR2 to stop, and a SIGUSR1 to resume the logging.
I.e.: kill -SIGUSRx `cat /var/run/audio-entropyd.pid` to change the logging status.

v0.0.3
I let the daemon wait until the kernel signals that the lower treshold in
/proc/sys/kernel/random/write_wakeup_threshold (number of bits) has been reached;
then start adding until value of
/proc/sys/kernel/random/poolsize (in bytes) has been reached. Also after a timeout still some entropy is added (i.e.--wait), just like in version 0.0.0.
The signals don't cause problems anymore with reading audio from the soundcard.

v0.0.4
16nov2002, Folkert van Heusden <folkert@vanheusden.com>

           Lowered the sample-rate; the larger the samplerate, 
           the more biased the data gets.
           Removed the mixer-code; we should not crank-up
           the volume since that also increases the bias.
           Changed the daemonise-function to use 'daemon(...)
           which is a lot less code.
           Removed hashing-code since the kernel already
           takes care of doing such things.
           Removed lots of code that dumped all kinds of
           data to files. Not needed anyway in my opinion.
           Added code that determines the amount of entropy
           in a buffer of data so that we actually say
           something good to the kernel when submitting that
           data.
           Removed sleeping-code; now only data gets
           submitted when the kernel-entropy-spool gets
           empty.
           Rewrote the method of entropy-gathering; some
           code has been added to remove bias in the data.

A lot of thanks to & new algorithm by: Gertjan van Oosten (gertjan@West.NL)

v0.0.5
22dec2002, Folkert van Heusden <folkert@vanheusden.com>

           Input/output buffers are now locked into
           memory.