svn commit: r239569 - head/etc/rc.d

[RW]

On Tue, 11 Sep 2012 16:01:17 -0700
Xin Li wrote:


> Well, 1:1 correspondence is when we fed full text to /dev/random,
> which we don't, right?  Only the first 4K gets consumed.  So:
> 
> Situation 1: we have 45K of plain text, and only first 4k is fed to
> /dev/random at about 5 bits of entropy per byte;
> 
> Situation 2: we have 45K of plain text, compress to e.g. 25K and only
> first 4k is fed to /dev/random at more than 7.6 bits of entropy per
> byte;
> 
> Therefore I think Situation 2 is better than situation 1.

It's marginally better, but still a very poor solution. You still
lose most of the entropy, and you still end up with a substantial risk
of there being no buffers available for /entropy.

Situation 3: use a hash; all the entropy (up to an overkill amount) ends
up in yarrow, most of the buffer space is left for /entropy.


Compression solves neither of the two problem - hashing solves both.