So, serde seems to be downloading and running a binary on the system without informing the user and without any user consent. Does anyone have any background information on why this is, and how this is supposed to be a good idea?
dtolnay seems like a smart guy, so I assume there is a reason for this, but it doesn’t feel ok at all.
I’m a bit confused, proc macros could always execute arbitrary code on developer machines. As long as the source for the precompiled binary is available (which seems to be the case here), how is this any different than what any other proc macro is doing?
Edit: I should add that any package, macro or not, can also do so in a
build.rsscript.One problem is that the build isn’t easily reproducible: there are a few comments in that issue thread from someone trying to reproduce it and failing.
That seems like it could be an issue, but not the issue being raised by the post. The original post was talking about executing binary code on a user’s machine without consent. The thing is, this is how a lot of Rust packages work. Any package can have a
build.rsthat runs arbitrary code on a developer’s machine (that gets compiled into a binary automatically by Cargo). Any proc macro is arbitrary code that gets compiled into a binary and executed on a developer’s machine. In fact, any library, regardless of if there’s abuild.rsor if it’s a proc macro, can have malicious code in it that gets executed when a developer calls a specific method.None of this is new. When done maliciously, it’s called a supply-chain attack. All packages can do this. This is part of why there’s been interest in executing some of this code in WASM runtimes within the compiler, so that developers can explicitly control the level of impact those packages can have on a developer’s machine. That being said, WASM doesn’t solve the fact that any package can just have malicious code in it that gets executed during runtime. This is why people should vet their packages themselves (when it’s important, at least) to ensure that this won’t happen to them.
If the executable were easily reproducible from the source code, then yes, downloading a precompiled binary would be akin to executing code in
build.rsor a proc macro. The fact that it’s not makes these very different, because it makes your suggestion of “vet[ting] their packages themselves” impossible.Maybe I’m missing something, but I’m not seeing where in serde we’re downloading a precompiled binary. I see a script we can execute ourselves in the repository and an alternative serde_derive that uses that executable (after we compile it), but not where the actual published package has the executable.
It’s possible I’m missing something here though.
bsdtar tfv ᐸ(curl -sL https://static.crates.io/crates/serde_derive/serde_derive-1.0.183.crate)Edit: Ogh, using
ᐸwhich is a replacement character because Lemmy escapes the real one. This is annoying.There, you will see that this file exists:
-rwxr-xr-x 0 0 0 690320 Jul 24 2006 serde_derive-1.0.183/serde_derive-x86_64-unknown-linux-gnuYes, that’s a pre-built binary in the crate source release. It’s that bad.
Looks like I missed that, I was checking locally but I must have been checking an outdated version of the package. I’d feel better about it if it compiled on the user’s machine, which is the impression I was getting.
You can read a
build.rsscript.Then there’s this: http://cm.bell-labs.co/who/ken/trust.html
I’m not sure I follow what that link has to do with this, though.
serdeis open source, anyone can go compile it themselves. In fact, from what I can tell, to get the precompiled version ofserde_derive, you need to compile it yourself anyway. Compiling these proc-macros to binaries before executing the code isn’t new, this is what Cargo does with all proc macros.Also, I might be misreading the source here, but it looks like the executable needs to be manually compiled by the user on their own (by running the
precompiled/build.shscript), and they need to manually add the precompiled variant ofserde_deriveas a dependency instead of using the version that’s on crates.io. Am I missing something here? Is this automatically used by the published version ofserdesomewhere?No,
serde_derivecontains the binary and if you are on linux it will try to run it without asking the user. In fact there’s no way to make it so it won’t run.You can read the source of build.rs and and proc macros executed during a build, but do you? Does anyone do that every time they add a new dependency?
When adding a new dependency I almost always go over the source code to see what kind of performance to expect. If
build.rsis there - checking it takes a single click so yes to that too. Derive macro - less frequently, but you have to do it when documentation is non existent.