ThreadLocalRandom clinit troubles (original) (raw)

Peter Levart peter.levart at gmail.com
Tue Jun 24 21:35:37 UTC 2014

On 06/24/2014 06:01 PM, Martin Buchholz wrote:

On Tue, Jun 24, 2014 at 7:03 AM, Peter Levart <peter.levart at gmail.com_ _<mailto:peter.levart at gmail.com>> wrote: I would rather use SecureRandom.generateSeed() instance method instead of SecureRandom.nextBytes(). Why? Because every SecureRandom instance has to initialize it's seed 1st before getBytes() can provide the next random bytes from the PRNG. Since we only need the 1st 8 bytes from the SecureRandom instance to initialize TLR's seeder, we might as well directly call the SecureRandom.generateSeed() method. If I strace this program on Linux using strace -ff -q java SecureRandoms: public class SecureRandoms { public static void main(String[] args) throws Throwable { byte[] bytes = new byte[8]; new java.security.SecureRandom().nextBytes(bytes); } } I see a read from /dev/urandom, but not from /dev/random, so I conclude your intuitive understanding of how the seeding works must be wrong. It makes sense that NativePRNG doesn't need to do any special seeding of its own, since it reuses the operating system's.

You're right. I checked again. The NativePRNG is actually using /dev/urandom (by default unless java.security.egd or securerandom.source is defined). It's mixing the /dev/urandom stream with the stream obtained from SHA1 generator which is seeded by 20 bytes from /dev/urandom too. So by default yes, plain NativePRNG (the default on UNIX-es) is using /dev/urandom for nextBytes(), but this can be changed by defining java.security.egd or securerandom.source system property. I still think that for configuration-independent PRNG seed on UNIX-es it's better to invoke generateSeed() on NativePRNG$NonBlocking, which hard-codes /dev/urandom and doesn't mix it with SHA1 stream.

On Windows, there's a different story, since the default SecureRandom algorithm is SHA1, seeded by SeedGenerator.getSystemEntropy() and SeedGenerator.generateSeed(). The former call includes invoking networking code and resolving local host name. Which we would like to avoid. So I think we need a nicer story on windows then just: new SecureRandom().nextBytes(). I propose requesting explicit algorithm / provider on each particular platform that we know does best what we want, rather than using default which can still be used as a fall-back.

Regards, Peter

More information about the core-libs-dev mailing list