trying to use libsodium from micropython

[stef]

did anyone ever try this? succeed? searchengines are not helpful in finding earlier success. i'm trying now as a natmod, but maybe there's an easier way? thanks for any hint dear community hivemind.

[Josverl]

Which platforms are of interest to you. Should be possible on Unix and windows ports, but for the bare metal ports you would likely need to make use of hardware support in specific processors to be able to make it work with any reasonable performance.

what do you intend to use it for?

[stef]

esp32. the only hw support needed is the rng. the rest is not very performance critical, it's not gonna be a webserver that will serve thousands of connections. only human initiated small calculations. i don't worry about performance (i had libsodium working on a bare bones stm32 metal 15 years ago, and it was fine, post-quantum algos were not, libsodium stuff was)

[stef]

the only thing that might be a performance bottleneck is exactly two ristretto255 scalar multiplications per request, for which no cpu (except the risc-v in the betrusted.io precursor) has hw support for. and if someone chooses a micropython based esp32 for this, they will make that speed trade-off knowingly.

[stef]

btw i am the author and maintainer for pysodium, so for unix/windows ports my libsodium python bindings needs are self-serving :) - why would i use micropython on those platforms?

[stef]

my idea was, that if i wrap my protocol + pylibsodium with a micropython frontend, then auditors will have lower cognitive costs in reviewing the "frontend" in micropython than if i implement the whole thing in c with zephyr. (note the protocol it self is in c, and portable across other platforms, i'm just looking for a way to expose the protocol from common and widely supported mcus), but i ran into a brickwall when trying to build the libsodium deps, since these limits that .ro.data .data must be empty for example, that would mean i have to change code in libsodium, and that means increase of auditors cognitive costs, since they also have to audit libsodium changes, and not just simply use the stock release.

[stef]

btw one detail that i forgot to mention, i do not need to expose libsodium to python scripts in micropython, libsodium is a dep of the protocol implementation in c, i want to expose only the protocol, not libsodium.

[andrewleech]

The STM32 port actually already uses libhydrogen (essentially the embedded version of libsodium) for its encrypted/signed dfu bootloader support. I never had time/need to expose it in python however, though I did think it might be a useful tool to have one day.

[stef]

i started working on this, although only as a dep to the protocol i'm trying to expose. but i believe the results can be plucked and cherry-picked into something more generic.

[stef]

i think i found a solution to the "limits that .ro.data .data must be empty" but how to access any existing RNG hw devices on the mcu is an open question, see also https://github.com/orgs/micropython/discussions/17360