[PATCH v2] arc4random: simplify design for better safety

[Mark Harris]

Jason A. Donenfeld wrote:
> +      l = __getrandom_nocancel (p, n, 0);
> +      if (l > 0)
> +       {
> +         if ((size_t) l == n)
> +           return; /* Done reading, success. */
> +         p = (uint8_t *) p + l;
> +         n -= l;
> +         continue; /* Interrupted by a signal; keep going. */
> +       }
> +      else if (l == 0)
> +       arc4random_getrandom_failure (); /* Weird, should never happen. */
> +      else if (errno == ENOSYS)
> +       {
> +         have_getrandom = false;
> +         break; /* No syscall, so fallback to /dev/urandom. */
> +       }
> +      arc4random_getrandom_failure (); /* Unknown error, should never happen. */

Isn't EINTR also possible?  Aborting in that case does not seem reasonable.

Also the __getrandom_nocancel function does not set errno on Linux; it
just returns INTERNAL_SYSCALL_CALL (getrandom, buf, buflen, flags).
So unless that is changed, it doesn't look like this ENOSYS check will
detect old Linux kernels.

> +      struct pollfd pfd = { .events = POLLIN };
> +      pfd.fd = TEMP_FAILURE_RETRY (
> +         __open64_nocancel ("/dev/random", O_RDONLY | O_CLOEXEC | O_NOCTTY));
> +      if (pfd.fd < 0)
> +       arc4random_getrandom_failure ();
> +      if (__poll (&pfd, 1, -1) < 0)
> +       arc4random_getrandom_failure ();
> +      if (__close_nocancel (pfd.fd) < 0)
> +       arc4random_getrandom_failure ();

The TEMP_FAILURE_RETRY handles EINTR on open, but __poll can also
result in EINTR.


 - Mark