BIND and Solaris rand()

[Chris Thompson]

On Apr 7 2009, Stacey Jonathan Marshall wrote:

>On 04/06/09 16:06, Chris Thompson wrote:
>> It turns out that rand(3c) in even recent Solaris versions returns
>> values in the range 0..32767 only. I suppose this is part of Sun's
>> rather extreme paranoia about backwards compatibility with programs
>> written before the flood.

>rand(3C) - simple random-number generator returns successive 
>pseudo-random numbers in the range of 0 to  RAND_MAX (defined as 32767).

Yes, it's documented - I should perhaps have said that.

However, RAND_MAX on some other Unix-flavoured systems (including some,
but apparently not all, Linuxes) is 0x7FFFFFFF so that one gets random
31-bit numbers rather than 15-bit ones. It seems that the writers
of BIND's lib/isc/random.c were assuming that (the comments go on
about the top bit being 0 and the bottom bits less random).

>BIND 8 was changed to use /dev/urandom on Solaris around the time when 
>arc4random() was added to it.
>Perhaps a similar solution could be used here?

I'm not quite sure what you are suggesting here. BIND needs (fast)
weak pseudo-random numbers as well as (slow) cryptographically strong
ones. The routines in lib/isc/random.c can get hot, e.g. if they
are used to randomise the order that RRsets are returned in responses.
I don't think one would want to use /dev/urandom there (except to
seed the generator, perhaps).

-- 
Chris Thompson
Email: cet1 at cam.ac.uk