Re: [PATCH v1] random: block in /dev/urandom

["Jason A. Donenfeld" <Jason@zx2c4.com>]

Hi Guenter,

On Tue, Mar 22, 2022 at 10:56:44AM -0700, Guenter Roeck wrote:
> Everything - including the various root file systems - is at
> git@github.com:groeck/linux-build-test.git. Look into rootfs/ for the
> various boot tests. I'll be happy to provide some qemu command lines
> if needed.

Thanks. It looks like the "problem" is with this shell script:

  init_rng() {
    if check_file_size; then
      printf 'Initializing random number generator: '
      dd if="$URANDOM_SEED" bs="$pool_size" of=/dev/urandom count=1 2> /dev/null
      status=$?
      if [ "$status" -eq 0 ]; then
        echo "OK"
      else
        echo "FAIL"
      fi
      return "$status"
    fi
  }
  
  save_random_seed() {
    printf 'Saving random seed: '
    if touch "$URANDOM_SEED" 2> /dev/null; then
      old_umask=$(umask)
      umask 077
      dd if=/dev/urandom of="$URANDOM_SEED" bs="$pool_size" count=1 2> /dev/null
      status=$?
      umask "$old_umask"
      if [ "$status" -eq 0 ]; then
        echo "OK"
      else
        echo "FAIL"
      fi
    else
      status=$?
      echo "SKIP (read-only file system detected)"
    fi
    return "$status"
  }

  case "$1" in
    start|restart|reload)
      # Carry a random seed from start-up to start-up
      # Load and then save the whole entropy pool
      init_rng && save_random_seed;;

This code is actually problematic for a number of reasons. (And Linus,
I'm not saying "userspace is wrong" to justify breaking it or something,
don't worry.)

The first `dd if="$URANDOM_SEED" bs="$pool_size" of=/dev/urandom count=1`
will write the seed into the input pool, but:

  - It won't credit the entropy from that seed, so the pool won't
    actually initialize. (You need to use the ioctl to credit it.)
  - Because the pool doesn't initialize, subsequent reads from
    /dev/urandom won't actually use that seed.

The first point is why we had to revert this patch. But the second one
is actually a bit dangerous: you might write in a perfectly good seed to
/dev/urandom, but what you read out for the subsequent seed may be
complete deterministic crap. This is because the call to write_pool()
goes right into the input pool and doesn't touch any of the "fast init"
stuff, where we immediately mutate the crng key during early boot.

As far as I can tell, this has been the behavior for a really long time,
making the above innocuous pattern a pretty old thing that's broken. So
I could perhaps say, "this behavior is so old now, that your userspace
code is just plain broken," but I think I might actually have a very
quick unobtrusive fix for this. I'll mull some things over for rc2 or
later in rc1.

But, anyway, this only fixes the second point mentioned above. The first
one -- which resulted in the revert -- remains a stumper for now.

Jason