On Tue, Jan 3, 2023 at 12:15 PM Linus Torvalds
[off-list ref] wrote:

On Tue, Jan 3, 2023 at 12:03 PM Jason A. Donenfeld [off-list ref] wrote:

quoted

That buffering cannot be done safely currently

.. again, this is "your semantics" (the (b) in my humbug list), not
necessarily reality for anybody else.

Just to make an example: fork() is already problematic for something
as fundamental as <stdio.h>.

That doesn't mean that we do a special fork-safe stdio.h
infrastructure in the kernel. It just means that people have to do
things like fflush(NULL) (or use variations of setbuf() and friends)
when they deal with fork() and stdio interactions.

The random number generator really isn't that different. Periodic
reseeding isn't something stdio has to deal with, but having a
timestamp in user space library and forcing a re-seed isn't unheard of
in other contexts.

You don't even need anything as fancy as an actual timer, because it
doesn't need *active* flushing, just a "oh, it's been too long since
we read the random data, let's do it again".

And yes, I bet it would be a good idea to have an actual library for
this that handles (and *documents*) these kinds of issues - exactly
like a <stdio.h>, just for randomness.

I just don't think it should be involved in the kernel - again exactly
like <stdio.h>.

            Linus