2022-07-26 15:35:51

by Yann Droneaud

[permalink] [raw]
Subject: Re: [PATCH v4] arc4random: simplify design for better safety

Hi,

Le 26/07/2022 à 15:30, Jason A. Donenfeld via Libc-alpha a écrit :
> Rather than buffering 16 MiB of entropy in userspace (by way of
> chacha20), simply call getrandom() every time.


I dislike the wording because

1) the current buffer is only 512 bytes, not 16MiBytes;
2) implementation reads only 48 bytes of "fresh" entropy from
getrandom() each 16MiBytes generated.

I'm thinking "stirring" or "streaming" would better describe what's
happening:

"Rather than stirring 16MiB of random data in userspace before reseeding"


> This approach is doubtlessly slower, for now, but trying to prematurely
> optimize arc4random appears to be leading toward all sorts of nasty
> properties and gotchas. Instead, this patch takes a much more
> conservative approach. The interface is added as a basic loop wrapper
> around getrandom(), and then later, the kernel and libc together can
> work together on optimizing that.
>
> This prevents numerous issues in which userspace is unaware of when it
> really must throw away its buffer, since we avoid buffering all
> together.


I believe the cloned virtual machine issue should be explicitly
described as a major blocker in the commit message.


> Future improvements may include userspace learning more from
> the kernel about when to do that, which might make these sorts of
> chacha20-based optimizations more possible. The current heuristic of 16
> MiB is meaningless garbage that doesn't correspond to anything the
> kernel might know about. So for now, let's just do something
> conservative that we know is correct and won't lead to cryptographic
> issues for users of this function.
>
> This patch might be considered along the lines of, "optimization is the
> root of all evil," in that the much more complex implementation it
> replaces moves too fast without considering security implications,
> whereas the incremental approach done here is a much safer way of going
> about things. Once this lands, we can take our time in optimizing this
> properly using new interplay between the kernel and userspace.
>
> getrandom(0) is used, since that's the one that ensures the bytes
> returned are cryptographically secure. But on systems without it, we
> fallback to using /dev/urandom. This is unfortunate because it means
> opening a file descriptor, but there's not much of a choice. Secondly,
> as part of the fallback, in order to get more or less the same
> properties of getrandom(0), we poll on /dev/random, and if the poll
> succeeds at least once, then we assume the RNG is initialized. This is a
> rough approximation, as the ancient "non-blocking pool" initialized
> after the "blocking pool", not before, and it may not port back to all
> ancient kernels, but it does to a decent swath of them, so generally
> it's the best approximation we can do.
>
> The motivation for including arc4random, in the first place, is to have
> source-level compatibility with existing code. That means this patch
> doesn't attempt to litigate the interface itself. It does, however,
> choose a conservative approach for implementing it.


Sure arc4random() interface is inherited from *BSD, thus we're not free
to improve it. But arc4random() is already here in glibc git, thus I
think the paragraph is of dubious value in the commit message and can be
removed.


> Cc: Adhemerval Zanella Netto <[email protected]>
> Cc: Florian Weimer <[email protected]>
> Cc: Cristian Rodríguez <[email protected]>
> Cc: Paul Eggert <[email protected]>
> Cc: Mark Harris <[email protected]>
> Cc: Eric Biggers <[email protected]>
> Cc: [email protected]
> Signed-off-by: Jason A. Donenfeld <[email protected]>
> ---
> LICENSES | 23 -
> NEWS | 4 +-
> include/stdlib.h | 3 -
> manual/math.texi | 13 +-
> stdlib/Makefile | 2 -
> stdlib/arc4random.c | 205 ++-----
> stdlib/arc4random.h | 48 --
> stdlib/chacha20.c | 191 ------
> stdlib/tst-arc4random-chacha20.c | 167 -----
> sysdeps/aarch64/Makefile | 4 -
> sysdeps/aarch64/chacha20-aarch64.S | 314 ----------
> sysdeps/aarch64/chacha20_arch.h | 40 --
> sysdeps/generic/tls-internal-struct.h | 1 -
> sysdeps/generic/tls-internal.c | 10 -
> sysdeps/mach/hurd/_Fork.c | 2 -
> sysdeps/mach/hurd/kernel-features.h | 1 +
> sysdeps/nptl/_Fork.c | 2 -
> .../powerpc/powerpc64/be/multiarch/Makefile | 4 -
> .../powerpc64/be/multiarch/chacha20-ppc.c | 1 -
> .../powerpc64/be/multiarch/chacha20_arch.h | 42 --
> sysdeps/powerpc/powerpc64/power8/Makefile | 5 -
> .../powerpc/powerpc64/power8/chacha20-ppc.c | 256 --------
> .../powerpc/powerpc64/power8/chacha20_arch.h | 37 --
> sysdeps/s390/s390-64/Makefile | 6 -
> sysdeps/s390/s390-64/chacha20-s390x.S | 573 ------------------
> sysdeps/s390/s390-64/chacha20_arch.h | 45 --
> sysdeps/unix/sysv/linux/Makefile | 3 +-
> sysdeps/unix/sysv/linux/Versions | 1 +
> sysdeps/unix/sysv/linux/kernel-features.h | 7 +
> sysdeps/unix/sysv/linux/not-cancel.h | 6 +
> .../sysv/linux/ppoll_nocancel.c} | 19 +-
> sysdeps/unix/sysv/linux/tls-internal.c | 10 -
> sysdeps/unix/sysv/linux/tls-internal.h | 1 -
> sysdeps/x86_64/Makefile | 7 -
> sysdeps/x86_64/chacha20-amd64-avx2.S | 328 ----------
> sysdeps/x86_64/chacha20-amd64-sse2.S | 311 ----------
> sysdeps/x86_64/chacha20_arch.h | 55 --
> 37 files changed, 89 insertions(+), 2658 deletions(-)
> delete mode 100644 stdlib/arc4random.h
> delete mode 100644 stdlib/chacha20.c
> delete mode 100644 stdlib/tst-arc4random-chacha20.c
> delete mode 100644 sysdeps/aarch64/chacha20-aarch64.S
> delete mode 100644 sysdeps/aarch64/chacha20_arch.h
> delete mode 100644 sysdeps/powerpc/powerpc64/be/multiarch/Makefile
> delete mode 100644 sysdeps/powerpc/powerpc64/be/multiarch/chacha20-ppc.c
> delete mode 100644 sysdeps/powerpc/powerpc64/be/multiarch/chacha20_arch.h
> delete mode 100644 sysdeps/powerpc/powerpc64/power8/chacha20-ppc.c
> delete mode 100644 sysdeps/powerpc/powerpc64/power8/chacha20_arch.h
> delete mode 100644 sysdeps/s390/s390-64/chacha20-s390x.S
> delete mode 100644 sysdeps/s390/s390-64/chacha20_arch.h
> rename sysdeps/{generic/chacha20_arch.h => unix/sysv/linux/ppoll_nocancel.c} (62%)
> delete mode 100644 sysdeps/x86_64/chacha20-amd64-avx2.S
> delete mode 100644 sysdeps/x86_64/chacha20-amd64-sse2.S
> delete mode 100644 sysdeps/x86_64/chacha20_arch.h
>
> diff --git a/manual/math.texi b/manual/math.texi
> index 141695cc30..6d69bbff66 100644
> --- a/manual/math.texi
> +++ b/manual/math.texi
> @@ -1993,17 +1993,10 @@ This section describes the random number functions provided as a GNU
> extension, based on OpenBSD interfaces.
>
> @Theglibc{} uses kernel entropy obtained either through @code{getrandom}
> -or by reading @file{/dev/urandom} to seed and periodically re-seed the
> -internal state. A per-thread data pool is used, which allows fast output
> -generation.
> +or by reading @file{/dev/urandom} to seed.
>
> -Although these functions provide higher random quality than ISO, BSD, and
> -SVID functions, these still use a Pseudo-Random generator and should not
> -be used in cryptographic contexts.
> -
> -The internal state is cleared and reseeded with kernel entropy on @code{fork}
> -and @code{_Fork}. It is not cleared on either a direct @code{clone} syscall
> -or when using @theglibc{} @code{syscall} function.
> +These functions provide higher random quality than ISO, BSD, and SVID
> +functions, and may be used in cryptographic contexts.

+ "provided getrandom() and /dev/urandom() could be used in such
context." ;)


Thanks for the improvements, can't wait for a vDSO getrandom() optimized
for reading 1,2,4,8 bytes :)


Regards.


--

Yann Droneaud

OPTEYA