Writing into /dev/urandom doesn't actually credit any entropy bits. And
while it adds that data to the entropy pool, it won't actually be
immediately used when reading from /dev/urandom subsequently. This is
how the kernel's /dev/urandom has always worked, unfortunately.
As a result of this behavior, which may be understandably surprising,
writing a good seed file into /dev/urandom and then saving a new seed
file immediately after is dangerous, because the new seed file may wind
up being entirely deterministic, even if the old seed file was quite
good.
This has been fixed in systemd with
<da2862ef06>,
and fortunately it's possible to do the same thing in shell script here.
Specifically, instead of just saving new /dev/urandom output straight
up, we hash the new /dev/urandom together with the old seed, in order to
produce the new seed. This way the amount of entropy in the new seed
will stay the same or get better, but not appreciably regress.
At the same time, the pool size check in this script is useless. Writing
to /dev/urandom never credits bits anyway, so no matter what, writing
into /dev/urandom is useful and not harmful. There's also not much of a
point in seeding with more than 256 bits, which is what the hashing
operation above produces. So this commit removes the file size check.
As a final note, while this commit improves upon the status quo by
removing a vulnerability, this shell script still does not actually
initialize the RNG like it says it does. For initialization via a seed
file, the RNDADDENTROPY ioctl must be used but there's currently no way
to do that from a shell script for now.
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
f0986de551