svn commit: r249035 - head/lib/libc/stdlib

[Andrey Chernov]

On 03.04.2013 11:04, Bruce Evans wrote:
>> +    mib[0] = CTL_KERN;
>> +    mib[1] = KERN_ARND;
>> +    sysctl(mib, 2, (void *)&next, &len, NULL, 0);
>> }
> 
> The sysctl() is certain to fail on old kernels (like open of /dev/random
> on even older kernels), but there is no longer any error checking or
> handling.  The contents of `next' on error is indeterminate (not documented
> in the man page), but is probably unchanged.  Applications can actually
> detect this error although though the API doesn't support this, by
> using the documented implementation details and assuming that errno
> is properly left changed if the syscall fails (set errno to 0 before
> the call here and check it after).

Even in case sysctl is not fail, kernel's arc4 is very poorly
initialized right after the boot. See the comment in the
libkern/arc4random.c arc4_randomstir(), it is about device not loaded
case, but it is the same even if device is loaded but not harvest
enough. NetBSD implementation of kernel's arc4 is different from ours.
Really I am _very_ tired to tell people about the problem, to write
patches for the problem, etc., so anybody who are interested can search
more details and patches in the mail archives.