[openssl-users] DH_generate_key Hangs

[Jeffrey Walton]

>> You should avoid calls to RAND_poll altogether on Windows. Do so by
>> explicitly seeding the random number generator yourself.
>
> As a starting point, try something like this:
>
> -----
> static ENGINE *rdrand;
>
> void init_prng(void) {
>     /* Try to seed the PRNG with the Intel RDRAND on-chip PRNG */
>     OPENSSL_cpuid_setup();
>     ENGINE_load_rdrand();
>     rdrand = ENGINE_by_id("rdrand");
>     if (rdrand) {
>         int success = 0;
>         if (ENGINE_init(rdrand)) {
>             success = ENGINE_set_default(rdrand, ENGINE_METHOD_RAND);
>         }
>
>         /***
>         Per OpenSSL wiki, call ENGINE_free here regardless of whether we're
>         successfully using rdrand. The "functional reference" to rdrand will
>         be released when we call ENGINE_finish.
>         ***/
>         ENGINE_free(rdrand);
>         if (! success) ENGINE_finish(rdrand), rdrand = NULL;
>     }
>
>     if (!rdrand && !RAND_status()){
>       RAND_screen();   /* this isn't really emough entropy, but it's a start */
>       if (!RAND_status()) {
>          RAND_poll();      /* try to gather additional entropy */
>       }
>    }
> }
>
> void terminate_engines(void) {
>    if (rdrand) ENGINE_finish(rdrand), rdrand = NULL;
>    /* similarly for any other engines you use */
>    ENGINE_cleanup();
> }
> -----
>
> Call init_prng after your OpenSSL initialization code (e.g. after calling OpenSSL_add_all_algorithms), and terminate_engines when you're done using OpenSSL (e.g. just before process exit).
>
> Note that this code uses RAND_screen if RDRAND isn't available. RAND_screen is really not a very good idea; it may be OK on workstations, but rarely provides much entropy on servers because they typically aren't doing much screen output. And if you still need entropy after the RAND_screen call, you'll end up in RAND_poll anyway. The alternative is to write your own code that harvests entropy from some source (or sources).
>
> Other people may have better suggestions.

Headless servers without hw entropy sources are tough. In this case I
use hedging. I've got some patches somewhere for 1.0.1, but they won't
apply to 0.9.8.

Also see:

* When Good Randomness Goes Bad: Virtual Machine Reset Vulnerabilities
and Hedging Deployed Cryptography,
http://pages.cs.wisc.edu/~rist/papers/sslhedge.pdf
* When Virtual is Harder than Real: Security Challenges in Virtual
Machine Based Computing Environments,
http://www.usenix.org/legacy/event/hotos05/final_papers/full_papers/garfinkel/garfinkel.pdf

Jeff