RNG_SystemRNG should fail rather than falling back on rng_systemFromNoise

RNG_SystemRNG should fail rather than falling back on rng_systemFromNoise
if it cannot call RtlGenRandom on Windows.

Remove the obsolete code for Windows versions older than Windows XP.

In unix_rand.c, abort the process under more conditions.

R=rsleevi@chromium.org
BUG=244661
TEST=none

[wtc, 2013-06-17 23:04:17]

https://codereview.chromium.org/17058006/diff/1/nss/lib/freebl/unix_rand.c
File nss/lib/freebl/unix_rand.c (right):

https://codereview.chromium.org/17058006/diff/1/nss/lib/freebl/unix_rand.c#ne...
nss/lib/freebl/unix_rand.c:888: bytes = RNG_FileUpdate("/dev/urandom",
SYSTEM_RNG_SEED_COUNT);

Here we request SYSTEM_RNG_SEED_COUNT bytes from /dev/urandom.
This is why I change to abort the process if |bytes| is not equal
to SYSTEM_RNG_SEED_COUNT.

https://codereview.chromium.org/17058006/diff/1/nss/lib/freebl/unix_rand.c#ne...
nss/lib/freebl/unix_rand.c:929: (int)bytes, SYSTEM_RNG_SEED_COUNT);

|bytes| is a size_t, so we need the (int) cast to match the %d
format specifier.

https://codereview.chromium.org/17058006/diff/1/nss/lib/freebl/win_rand.c
File nss/lib/freebl/win_rand.c (right):

https://codereview.chromium.org/17058006/diff/1/nss/lib/freebl/win_rand.c#new...
nss/lib/freebl/win_rand.c:387: return bytes;

Note that I am not calling abort() in this function.

In my testing with upstream NSS, making RNG_SystemRNG fail
(return 0) is enough to cause subsequent attempts to use NSS
to fail.

[Ryan Sleevi, 2013-06-17 23:19:17]

lgtm

https://codereview.chromium.org/17058006/diff/1/nss/lib/freebl/unix_rand.c
File nss/lib/freebl/unix_rand.c (right):

https://codereview.chromium.org/17058006/diff/1/nss/lib/freebl/unix_rand.c#ne...
nss/lib/freebl/unix_rand.c:929: (int)bytes, SYSTEM_RNG_SEED_COUNT);
On 2013/06/17 23:04:17, wtc wrote:
> 
> |bytes| is a size_t, so we need the (int) cast to match the %d
> format specifier.

Is NSS allowed to use C99, or is it limited to C89?

I'm wondering whether the C99 zu is sufficient here. Still, your int cast is
probably fine too.

[wtc, 2013-06-17 23:51:27]

Good idea! This code is in a Chromium patch, and unix_rand.c
is only compiled by GCC and clang. So it is fine to use %zu.
Thanks for the suggestion.

[commit-bot: I haz the power, 2013-06-17 23:53:15]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/wtc@chromium.org/17058006/6001

[commit-bot: I haz the power, 2013-06-17 23:53:34]

Change committed as 206843