The current mess of code seems like a hodgepodge of complex ideas,
partially copied from systemd, but then subtly different, and it's a
mess. Let's simplify this drastically.
First, assume that getrandom() is always available. If the kernel is too
old, we have an unoptimized slowpath for still supporting ancient
kernels, a path that should be removed at some point. If getrandom()
isn't available and the fallback path doesn't work, the system has much
larger problems, so just crash. This should basically never happen.
getrandom() and having randomness available in general is a critical
system API that should be expected to be available on any functioning
system.
Second, assume that the rng is initialized, so that asking for random
numbers should never block. This is virtually always true on modern
kernels. On ancient kernels, it usually becomes true. But, more
importantly, this is not the responsibility of various daemons, even
ones that run at boot. Instead, this is something for the kernel and/or
init to ensure.
Putting these together, we adopt new behavior:
- First, try getrandom(..., ..., 0). The 0 flags field means that this
call will only return good random bytes, not insecure ones.
- If this fails for some reason that isn't ENOSYS, crash.
- If this fails due to ENOSYS, poll on /dev/random until 1 byte is
available, suggesting that subsequent reads from the rng will almost
have good random bytes. If this fails, crash. Then, read from
/dev/urandom. If this fails, crash.
We don't bother caching when getrandom() returns ENOSYS. We don't apply
any other fancy optimizations to the slow fallback path. We keep that as
barebones and minimal as we can. It works. It's for ancient kernels. It
should be removed soon. It's not worth spending cycles over. Instead,
the goal is to eventually reduce all of this down to a simple boring
call to getrandom(..., ..., 0).
https://gitlab.freedesktop.org/NetworkManager/NetworkManager/-/merge_requests/2127
Heavily inspired by systemd ([1]).
We now also have nm_random_get_bytes{,_full}() and
nm_random_get_crypto_bytes(), like systemd's random_bytes()
and crypto_random_bytes(), respectively.
Differences:
- instead of systemd's random_bytes(), our nm_random_get_bytes_full()
also estimates whether the output is of high quality. The caller
may find that interesting. Due to that, we will first try to call
getrandom(GRND_NONBLOCK) before getrandom(GRND_INSECURE). That is
reversed from systemd's random_bytes(), because we want to find
out whether we can get good random numbers. In most cases, kernel
should have entropy already, and it makes no difference.
Otherwise, heavily rework the code. It should be easy to understand
and correct.
There is also a major bugfix here. Previously, if getrandom() failed
with ENOSYS and we fell back to /dev/urandom, we would assume that we
have high quality random numbers. That assumption is not warranted.
Now instead poll on /dev/random to find out.
[1] a268e7f402/src/basic/random-util.c (L81)
