GNU ld has prefixed the contents of .gnu.warning.SYMBOL sections
with "warning: " since 2003, so the messages themselves need not
contain the prefix anymore.

From Scott Cheloha
ok jca, deraadt

Prefer _MUTEX_*LOCK over _THREAD_PRIVATE_MUTEX_*LOCK() when thread-specific
data isn't necessary.

ok mpi@, ok&tweak natano@

Move to the <limits.h> universe.
review by millert, binary checking process with doug, concept with guenther

typo; ok deraadt

warn for correct symbol

improve warnings from rand_r(), rand(), and random()
It may take a few iterations to get the tone right.
previously discussed with millert

Change rand(), random(), drand48(), lrand48(), mrand48(), and srand48()
to returning strong random by default, source from arc4random(3).
Parameters to the seeding functions are ignored, and the subsystems remain
in strong random mode. If you wish the standardized deterministic mode,
call srand_deterministic(), srandom_determistic(), srand48_deterministic(),
seed48_deterministic() or lcong48_deterministic() instead.
The re-entrant functions rand_r(), erand48(), nrand48(), jrand48() are
unaffected by this change and remain in deterministic mode (for now).

Verified as a good roadmap forward by auditing 8800 pieces of software.
Roughly 60 pieces of software will need adaptation to request the
deterministic mode.

Violates POSIX and C89, which violate best practice in this century.
ok guenther tedu millert

remove unused variable

ok tedu@

once srandomdev() is called, switch to using arc4random() but mask off the
high bit as required by posix. wouldn't want to break any standards.
idea and ok deraadt

oops, typo. James Hartley is fast at trying -current

In srandomdev(), use arc4random_buf() instead of from the kernel.
discussion with matthew

use getentropy; from matthew

Add linker warnings for rand() and random() and various related functions.

ok deraadt@

threadsafe random().
reported by ajacoutot and (much earlier, sorry) by Alexey Suslikov

Improve standards conformance: ecvt(), fcvt(), gcvt(), mktemp(), ttyslot(),
and valloc() are not in the current version, while posix_memalign() mkstemp(),
and mkdtemp() are, and setstate()'s argument has lost a bogus 'const'.

ok millert@ jmc@ espie@ kettenis@; ports build testing by naddy@

Fix a bug where random() always returns 0 when srandom() is seeded
with 0. Use 1 and not 0 as the first element of the state array,
similar to what glibc does. OK nicm@

Use sysctl(KERN_ARND) to get n bytes, instead of just 4 at a time
and remove fallback code. If somebody is dumb enough to make the
sysctl fail using systrace, he deserves what he gets. Saves 7 syscalls
on process startup.
looks good miod@ ok deraadt@ tedu@

zap remaining rcsid.

Kill old files that are no longer compiled.

okay theo

ansi + de-register

ok otto deraadt

Remove the advertising clause in the UCB license which Berkeley
rescinded 22 July 1999. Proofed by myself and Theo.

Use int32_t, not long since this deals with 32bit quantities.
Inspired by a change in NetBSD and reported by Jan Johansson.

In srandomdev(), if we can't access /dev/arandom, use the sysctl() instead.
We don't want to use the sysctl() by default since we are reading more
than just a few bytes of entropy when setting up the state.

Fix the leak for real (that's what I get for hacking when i can't sleep).

Fix an fd leak if the read from /dev/arandom fails. Pointed out by
Markus Friedl.

Add srandomdev() from FreeBSD for use by sendmail and others.

More XPG4.2 --
setstate takes a const parameter
don't ever spew to stderr, just return NULL

size_t n in initstate(); XPG

Remove dead code
Remove unused variables
Silence some warnings
lint(1) is your friend

Fix RCS ids
Make sure everything uses {SYS,}LIBC_SCCS properly

Substantially improve random number generation by using the largest prime
that fits inside 32 bits as the denominator; take care not to overflow.
Regenerate initial seed after replacing the generator

branches: 1.1.1;
Initial revision