Re: Concurrent access to /dev/urandom

[daw@taverner.cs.berkeley.edu (David Wagner)]

Jon Masters  wrote:
>On Sat, 27 Nov 2004 15:45:49 -0500, Bernard Normier <bernard@zeroc.com> wrote:
>> I use /dev/urandom to generate UUIDs by reading 16 random bytes from
>> /dev/urandom (very much like e2fsprogs' libuuid).
>
>Why not use /dev/random for such data instead?

Because /dev/urandom is the right thing to use, and /dev/random is not.

>/dev/urandom suffers from a poor level of entropy if you keep reading
>from it over and over again whereas the alternative can block until it
>has more randomness available.

That's not accurate.  Once it has been properly seeded, /dev/urandom
should be fine for this purpose (assuming no root compromise).  Because
/dev/urandom uses a cryptographically secure PRNG, once it has been securely
seeded with (say) 128 bits of secure entropy, you can generate as much
pseudorandom output as you like without worries (unless someone breaks
the crypto, which is usually considered unlikely).  If the crypto is secure
and /dev/urandom is properly seeded, then its pseudorandom output is
indistinguishable from true random bits; this is true even if you extract
millions of pseudorandom bits.  "Entropy" is often a misleading notion,
when you are dealing with cryptographic PRNGs and computationally bounded
adversaries.