Implement environment variable emulation on Windows

[RalfJung]

Currently, GetEnvironmentVariableW just always returns "env var not found".

Mentoring instructions available

[JOE1994]

Hello, I'm interested in working on this issue. @RalfJung , may I ask which file I should be working on for this issue?? Thank you :)

[RalfJung]

@JOE1994 great! Do you have a Windows system to do the development on? Without that, the testing will be a bit annoying as we'll have to use AppVeyor and bors for that. OTOH, I have zero experience with development on Windows, so you might have to do some experiments to get Miri to build and run locally.

The env var implementation for Unix is at

miri/src/shims/foreign_items.rs

Lines 414 to 493 in ce5e910

	"getenv" => {
	let result = {
	let name_ptr = this.read_scalar(args[0])?.to_ptr()?;
	let name = this.memory().get(name_ptr.alloc_id)?.read_c_str(tcx, name_ptr)?;
	match this.machine.env_vars.get(name) {
	Some(&var) => Scalar::Ptr(var),
	None => Scalar::ptr_null(&*this.tcx),
	}
	};
	this.write_scalar(result, dest)?;
	}
	"unsetenv" => {
	let mut success = None;
	{
	let name_ptr = this.read_scalar(args[0])?.not_undef()?;
	if !name_ptr.is_null_ptr(this) {
	let name_ptr = name_ptr.to_ptr()?;
	let name = this
	.memory()
	.get(name_ptr.alloc_id)?
	.read_c_str(tcx, name_ptr)?
	.to_owned();
	if !name.is_empty() && !name.contains(&b'=') {
	success = Some(this.machine.env_vars.remove(&name));
	}
	}
	}
	if let Some(old) = success {
	if let Some(var) = old {
	this.memory_mut().deallocate(var, None, MiriMemoryKind::Env.into())?;
	}
	this.write_null(dest)?;
	} else {
	this.write_scalar(Scalar::from_int(-1, dest.layout.size), dest)?;
	}
	}
	"setenv" => {
	let mut new = None;
	{
	let name_ptr = this.read_scalar(args[0])?.not_undef()?;
	let value_ptr = this.read_scalar(args[1])?.to_ptr()?;
	let value = this.memory().get(value_ptr.alloc_id)?.read_c_str(tcx, value_ptr)?;
	if !name_ptr.is_null_ptr(this) {
	let name_ptr = name_ptr.to_ptr()?;
	let name = this.memory().get(name_ptr.alloc_id)?.read_c_str(tcx, name_ptr)?;
	if !name.is_empty() && !name.contains(&b'=') {
	new = Some((name.to_owned(), value.to_owned()));
	}
	}
	}
	if let Some((name, value)) = new {
	// `+1` for the null terminator.
	let value_copy = this.memory_mut().allocate(
	Size::from_bytes((value.len() + 1) as u64),
	Align::from_bytes(1).unwrap(),
	MiriMemoryKind::Env.into(),
	);
	// We just allocated these, so the write cannot fail.
	let alloc = this.memory_mut().get_mut(value_copy.alloc_id).unwrap();
	alloc.write_bytes(tcx, value_copy, &value).unwrap();
	let trailing_zero_ptr = value_copy.offset(
	Size::from_bytes(value.len() as u64),
	tcx,
	).unwrap();
	alloc.write_bytes(tcx, trailing_zero_ptr, &[0]).unwrap();
	if let Some(var) = this.machine.env_vars.insert(
	name.to_owned(),
	value_copy,
	)
	{
	this.memory_mut().deallocate(var, None, MiriMemoryKind::Env.into())?;
	}
	this.write_null(dest)?;
	} else {
	this.write_scalar(Scalar::from_int(-1, dest.layout.size), dest)?;
	}
	}

It would probably be a good idea to start by moving as much of this as possible into a new env.rs module (inside src/shims). foreign_items.rs is already way too long anyway. ;) And then you'll have to figure out which foreign functions libstd calls on Windows for reading and writing environment variables, and emulate them. You can use this test case for that, running it in Miri will tell you which foreign function gets called that is not implemented. GetEnvironmentVariableW will likely be the first one.

[pvdrz]

Hi, @JOE1994 did you get a chance to look at this? I have been working on the linux side of this and I could give you a hand.

[JOE1994]

Hello @christianpoveda ,

Hi, @JOE1994 did you get a chance to look at this? I have been working on the linux side of this and I could give you a hand.

That is really nice of you! 👍 I apologize for being irresponsible in working on this issue..
Can I get back to you in a few days after I review the MIRI code and get a better understanding of it?

[pvdrz]

That is really nice of you! +1 I apologize for being irresponsible in working on this issue..

Don't worry, that happens from time to time.

Can I get back to you in a few days after I review the MIRI code and get a better understanding of it?

Sure, feel free to ask any questions :)

[JOE1994]

On Widows 10, using rustc 1.40.0-nightly (032a53a06 2019-10-03).
Currently failing call to "SetEnviromentVariableW"

[pvdrz]

Yes so SetEnvironmentVariableW and GetEnvironmentVariableW are the first two shims that need to be implemented. Those are the windows equivalents of getenv and setenv in linux (which can be found here). These shims call some methods inside this module. You need to do the same for the windows shims but taking care of the encoding.

[JOE1994]

@christianpoveda Thanks for your help!
I'm currently trying to write code for handling GetEnvironmentVariableW in Windows.
If I call the same handling logic as getenv for GetEnvironmentVariableW, I get a size mismatch error.
It seems like args[0] and dest should have a same size, in Linux the sizes are the same, but in Windows the sizes are different as below (8 vs 4). It's embarassing to keep asking questions for an E-easy issue, but can I ask you a question? 😀 😀

args[0] : &[rustc_mir::interpret::OpTy is an immutable reference, and
dest : rustc_mir::interpret::PlaceTy also contains a member variable place which seems to contain an index to a mir::Local. I'm not sure what I should do here to make args[0] and dest compatible in terms of their layout and size.

[pvdrz]

Hi. So the problem seems to be that GetEnvironmentVariableW has an slightly different behavior than getenv.

What GetEnvironmentVariableW does, is taking 3 arguments. A pointer to the name of the variable, a pointer to a buffer where the value of the variable will be written, and the size of such buffer (let's say n) . Then the function writes at most n bytes to the buffer and returns an integer to indicate if the operation was successful. More details here.

In contrast, what getenv does is taking as argument a pointer to the name of the variable and then returning a pointer to the value of the variable.

So what we can do here is implement the windows shim in a different function first, be sure that it works. and then try to deduplicate the code that is also used in the Unix shim.

About asking questions, is not embarrassing at all, we are all learning here :)

I'll write more detailed instructions about how to write to a buffer and such tomorrow. I'm in my phone right now :P

[RalfJung]

It's embarassing to keep asking questions for an E-easy issue

It totally isn't! Asking legitimate questions like yours is never embarrassing.

Also when I assigned the E-easy label, that was a guess. And even if this issue is easy as Miri issues go, there's still a lot of API surface to learn here.

---

Another problem you might run into is that GetEnvironmentVariableW probably works with wide strings (the W is suspicious), meaning 2 bytes per character. You'll need something like the Windows OsStringExt to convert that to something you can use in the interpreter.