Please add retpoline support.

[dancrossnyc]

For a project at Google, we need retpoline support. Taylor Cramer suggested I open an issue; my initial search didn't reveal any pre-existing issues that seemed particularly relevant. There was a reddit thread that stalled after identifying some blocking issues (LLD integration, LLVM upgrades, etc) all of which have, I believe, subsequently been addressed. https://www.reddit.com/r/rust/comments/7saomi/status_of_spectre_variant_2_mitigation_in_rust/

As I understand it, the heavy lifting for this is in LLVM and it simply needs to be plumbed through.

Put another way, just as clang provides -mretpoline (https://reviews.llvm.org/D41723), I'm looking for something similar in the Rust toolchain.

Thanks!

[alexcrichton]

Some simple testing in clang shows that -C target-feature=+retpoline may be sufficient in Rust to work here. Does that work for you?

[eddyb]

cc @rust-lang/compiler

[GabrielMajeri]

I'm no security expert, but from what I can tell from the LLVM code, this specific retpoline patch changes the PLT?

Because it just so happens we have a PR which will disable the PLT by default and it seems that it might reduce the performance impact of retpoline (and reduce the attack surface). If you also build your C/C++ libraries with a global CFLAGS=-fno-plt you should be able to 100% remove the PLT from your binaries.

[nagisa]

We should still support retpolines when @plt is used. That being said, it seems to me that this is almost entirely a linker’s job (i.e. all of the support is being added to lld and I haven’t seen any patches against LLVM itself). This means that implementing this is probably as easy as passing the relevant flags to linker and can be done even now (-Clink-args) without any explicit support from the compiler?

EDIT: OTOH from what @alexcrichton said above, it probably does need compiler support, but still no explicit changes to rustc are necessary.

[dancrossnyc]

Thanks all! I'll take what's written here and see if I can plumb the appropriate flags through cargo etc to get retpolines in the generated binary. I'll report back with results.

[cyplo]

Heya, any news on this one @dancrossnyc ?

[dancrossnyc]

Sorry for the delayed response; I was traveling over the new year.

We put this in a while back and it seems to be working as expected. Thanks, all!