spastorino / miri

An interpreter for Rust's mid-level intermediate representation

Miri

An experimental interpreter for Rust's mid-level intermediate representation (MIR). It can run binaries and test suites of cargo projects and detect certain classes of undefined behavior, for example:

• Out-of-bounds memory accesses and use-after-free
• Invalid use of uninitialized data
• Violation of intrinsic preconditions (an unreachable_unchecked being reached, calling copy_nonoverlapping with overlapping ranges, ...)
• Not sufficiently aligned memory accesses and references
• Violation of basic type invariants (a bool that is not 0 or 1, for example, or an invalid enum discriminant)
• WIP: Violations of the rules governing aliasing for reference types

Miri has already discovered some real-world bugs.

Running Miri on your own project('s test suite)

Install Miri as a cargo subcommand:

cargo +nightly install --force --git https://github.com/rust-lang/miri miri

If this does not work, try using the nightly version given in this file. CI should ensure that this nightly always works.

You have to use a consistent Rust version for building miri and your project, so remember to either always specify the nightly version manually (like in the example above), overriding it in your project directory as well, or use rustup default nightly (or rustup default nightly-YYYY-MM-DD) to globally make nightly the default toolchain.

Now you can run your project in Miri:

1. Run cargo clean to eliminate any cached dependencies. Miri needs your dependencies to be compiled the right way, that would not happen if they have previously already been compiled.
2. To run all tests in your project through Miri, use cargo +nightly miri test.
3. If you have a binary project, you can run it through Miri using cargo +nightly miri run.

You can pass arguments to Miri after the first --, and pass arguments to the interpreted program or test suite after the second --. For example, cargo +nightly miri run -- -Zmiri-disable-validation runs the program without validation of basic type invariants and references. cargo +nightly miri test -- -- filter passes filter to the test suite the same way cargo test filter would.

When running code via cargo miri, the miri config flag is set. You can use this to exclude test cases that will fail under Miri because they do things Miri does not support:

#[cfg(not(miri))]
#[test]
fn does_not_work_on_miri() {
    let x = 0u8;
    assert!(&x as *const _ as usize % 4 < 4);
}

Common Problems

When using the above instructions, you may encounter a number of confusing compiler errors.

"found possibly newer version of crate std which <dependency> depends on"

Your build directory may contain artifacts from an earlier build that have/have not been built for Miri. Run cargo clean before switching from non-Miri to Miri builds and vice-versa.

"found crate std compiled by an incompatible version of rustc"

You may be running cargo miri with a different compiler version than the one used to build the custom libstd that Miri uses, and Miri failed to detect that. Try deleting ~/.cache/miri.

Development and Debugging

If you want to hack on miri yourself, great! Here are some resources you might find useful.

Using a nightly rustc

Miri heavily relies on internal rustc interfaces to execute MIR. Still, some things (like adding support for a new intrinsic) can be done by working just on the Miri side.

To prepare, make sure you are using a nightly Rust compiler. The most convenient way is to install Miri using cargo, then you can easily run it on other projects:

cargo +nightly install --path "$DIR" --force # or the nightly in `rust-version`
cargo +nightly miri setup

If you want to use a different libstd (not the one that comes with the nightly), you can do that by running

XARGO_RUST_SRC=~/src/rust/rustc/src/ cargo +nightly miri setup

Either way, you can now do cargo +nightly miri run to run Miri with your local changes on whatever project you are debugging.

(We are giving +nightly explicitly here all the time because it is important that all of these commands get executed with the same toolchain.)

cargo miri setup should end in printing the directory where the libstd was built. For the next step to work, set that as your MIRI_SYSROOT environment variable:

export MIRI_SYSROOT=~/.cache/miri/HOST # or whatever the previous command said

Testing Miri

Instead of running an entire project using cargo miri, you can also use the Miri "driver" directly to run just a single file. That can be easier during debugging.

cargo run tests/run-pass/format.rs # or whatever test you like

You can also run the test suite with cargo test --release. cargo test --release FILTER only runs those tests that contain FILTER in their filename (including the base directory, e.g. cargo test --release fail will run all compile-fail tests). We recommend using --release to make test running take less time.

Now you are set up! You can write a failing test case, and tweak miri until it fails no more. You can get a trace of which MIR statements are being executed by setting the MIRI_LOG environment variable. For example:

MIRI_LOG=info cargo run tests/run-pass/vecs.rs

Setting MIRI_LOG like this will configure logging for miri itself as well as the rustc::mir::interpret and rustc_mir::interpret modules in rustc. You can also do more targeted configuration, e.g. to debug the stacked borrows implementation:

MIRI_LOG=rustc_mir::interpret=info,miri::stacked_borrows cargo run tests/run-pass/vecs.rs

In addition, you can set MIRI_BACKTRACE=1 to get a backtrace of where an evaluation error was originally created.

Using a locally built rustc

Since the heart of Miri (the main interpreter engine) lives in rustc, working on Miri will often require using a locally built rustc. The bug you want to fix may actually be on the rustc side, or you just need to get more detailed trace of the execution than what is possible with release builds -- in both cases, you should develop miri against a rustc you compiled yourself, with debug assertions (and hence tracing) enabled.

The setup for a local rustc works as follows:

git clone https://github.com/rust-lang/rust/ rustc
cd rustc
cp config.toml.example config.toml
# Now edit `config.toml` and set `debug-assertions = true` and `test-miri = true`.
# The latter is important to build libstd with the right flags for miri.
# This step can take 30 minutes and more.
./x.py build src/rustc
# If you change something, you can get a faster rebuild by doing
./x.py --keep-stage 0 build src/rustc
# You may have to change the architecture in the next command
rustup toolchain link custom build/x86_64-unknown-linux-gnu/stage2
# Now cd to your Miri directory, then configure rustup
rustup override set custom

With this, you should now have a working development setup! See "Testing Miri" above for how to proceed.

Running cargo miri in this setup is a bit more complicated, because the Miri binary you just created does not actually run without some enviroment variables. But you can contort cargo into calling cargo miri the right way for you:

# in some other project's directory, to run `cargo miri test`:
MIRI_SYSROOT=$(rustc +custom --print sysroot) cargo +custom run --manifest-path /path/to/miri/Cargo.toml --bin cargo-miri --release -- miri test

Miri -Z flags and environment variables

Several -Z flags are relevant for Miri:

• -Zmir-opt-level controls how many MIR optimizations are performed. miri overrides the default to be 0; be advised that using any higher level can make miri miss bugs in your program because they got optimized away.
• -Zalways-encode-mir makes rustc dump MIR even for completely monomorphic functions. This is needed so that miri can execute such functions, so miri sets this flag per default.
• -Zmiri-disable-validation is a custom -Z flag added by miri. It disables enforcing the validity invariant, which is enforced by default. This is mostly useful for debugging; it means miri will miss bugs in your program.

Moreover, Miri recognizes some environment variables:

• MIRI_SYSROOT (recognized by miri, cargo miri and the test suite) indicates the sysroot to use.
• MIRI_TARGET (recognized by the test suite) indicates which target architecture to test against. miri and cargo miri accept the --target flag for the same purpose.

Contributing and getting help

Check out the issues on this GitHub repository for some ideas. There's lots that needs to be done that I haven't documented in the issues yet, however. For more ideas or help with running or hacking on Miri, you can open an issue here on GitHub or contact us (oli-obk and RalfJ) on the Rust Zulip.

History

This project began as part of an undergraduate research course in 2015 by @solson at the University of Saskatchewan. There are slides and a report available from that project. In 2016, @oli-obk joined to prepare miri for eventually being used as const evaluator in the Rust compiler itself (basically, for const and static stuff), replacing the old evaluator that worked directly on the AST. In 2017, @RalfJung did an internship with Mozilla and began developing miri towards a tool for detecting undefined behavior, and also using miri as a way to explore the consequences of various possible definitions for undefined behavior in Rust. @oli-obk's move of the miri engine into the compiler finally came to completion in early 2018. Meanwhile, later that year, @RalfJung did a second internship, developing miri further with support for checking basic type invariants and verifying that references are used according to their aliasing restrictions.

Bugs found by Miri

Miri has already found a number of bugs in the Rust standard library, which we collect here.

• Debug for vec_deque::Iter accessing uninitialized memory
• From<&[T]> for Rc creating a not sufficiently aligned reference
• BTreeMap creating a shared reference pointing to a too small allocation
• VecDeque creating overlapping mutable references
• Futures turning a shared reference into a mutable one
• str turning a shared reference into a mutable one
• BTreeMap creating mutable references that overlap with shared references

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT) at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you shall be dual licensed as above, without any additional terms or conditions.