Windows: RtlGenRandom vs BCryptGenRandom

[newpavlov]

IIUC both functions use the same algorithm under the hood (also see this). The main difference is that BCryptGenRandom will not work on Windows XP, but otherwise it looks like BCryptGenRandom should be preferred. IIRC Rust explicitly does not support Windows XP anymore, so it should not matter.

@chouquette
Can you help here? In rust-lang/rust#60260 you have used BCryptGenRandom for UWP targets and left RtlGenRandom for non-UWP targets. Is there a reason for that?

cc @GabrielMajeri

[chouquette]

Hi,

Mostly because I was asked to :) rust-lang/rust#60260 (comment)

My understanding is that Microsoft advises to use CryptGenRandom over RtlGenRandom, but CryptGenRandom is deprecated and the "next generation cryptography API" should be used instead, and BCryptGenRandom is part of that API, so if you have the choice, BcryptGenRandom is probably to be favored, however it seems in the context of Rust, there are some historical reasons for keeping RtlGenRandom

In the case of UWP the choice is easy, you can't use RtlGenRandom

[ollie27]

I believe XP is still technically supported by std (https://forge.rust-lang.org/platform-support.html#tier-3) and that's why it needs both implementations. When XP support is officially dropped then the RtlGenRandom version can be removed and BCryptGenRandom can be used for all Windows targets. I suggest that getrandom just copies std for now.

[newpavlov]

Ah, indeed. In that case I think we can use RtlGenRandom fallback only for XP targets, and BCryptGenRandom for all other Windows targets.

UPD: Never mind, we can't distinguish between XP and later versions by using only target triplet.

Relevant discussion: https://internals.rust-lang.org/t/8745

[newpavlov]

Relevant issue: openssl/openssl#8644

So we will have to wait for dropping of Windows XP (and hopefully Vista) from the list of supported Rust targets, until then we will have to keep using RtlGenRandom.

[alexcrichton]

I'd recommend using git blame and doing some digging to learn about the rationale for the usage in Rust's standard library. Doing that, for example, leads to rust-lang/rust#45370 which has a lot more information about why things are the way they are.

[newpavlov]

The compiler team considers dropping Windows XP support, assuming it will be done, I think we can drop its support in getrandom v0.2 as well.

[josephlr]

SGTM, how long after they drop support should we update the library? Given that rand 0.8 isn't out yet, I'd be fine doing it sooner.

[newpavlov]

Right after the final decision will be made? IIRC Rust Windows XP support was arguably broken for a long time, so it should not be a problem. I also think that ideally we should do it before releasing rand v0.8.

[bluetech]

In case you're interested, Go had some discussion on this and decided to stick with RtlGenRandom for non-UWP: golang/go#33542