Go to file
2019-05-20 09:32:01 +02:00
bench-cargo-miri make MSE bench run a bit longer (less noise) 2019-01-07 22:57:05 +01:00
benches Fix benchmark 2019-04-07 19:30:37 -04:00
src typo 2019-05-19 14:04:59 +02:00
test-cargo-miri fix tests 2019-05-17 15:03:34 +02:00
tests give up on two-phase borrows 2019-05-17 15:02:31 +02:00
tex Rename lvalue to place 2017-12-06 08:39:31 +01:00
.appveyor.yml try to fix Windows CI 2019-04-18 16:44:43 +02:00
.editorconfig Add EditorConfig 2016-04-14 14:54:59 +02:00
.gitignore Retag is the only operation that generates new tags 2018-11-15 09:35:40 +01:00
.travis.yml CI: build with debug assertions 2019-04-18 14:58:30 +02:00
build.rs update to vergen 3 2018-10-09 20:17:54 +02:00
Cargo.toml Update cargo_metadata to 0.7 2019-04-27 23:31:48 +02:00
LICENSE-APACHE
LICENSE-MIT
README.md expand explanation of how we treat validity invariants 2019-05-20 09:32:01 +02:00
rust-version bump Rust 2019-05-15 08:58:43 +02:00
travis.sh make run-test runnable from other directories 2019-04-22 11:11:06 +02:00

Miri Build Status Windows build status

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. If you found a bug with Miri, we'd appreciate if you tell us and we'll add it to the list!

Be aware that Miri will not catch all possible errors in your program, and cannot run all programs:

  • There are still plenty of open questions around the basic invariants for some types and when these invariants even have to hold. Miri tries to avoid false positives here, so if you program runs fine in Miri right now that is by no means a guarantee that it is UB-free when these questions get answered. In particular, Miri does currently not check that integers are initialized or that references point to valid data.
  • If the program relies on unspecified details of how data is laid out, it will still run fine in Miri -- but might break (including causing UB) on different compiler versions or different platforms.
  • Miri is fully deterministic and does not actually pick a base address in virtual memory for the program's allocations. If program behavior depends on the base address of an allocation, Miri will stop execution (with a few exceptions to make some common pointer comparisons work).
  • Miri runs the program as a platform-independent interpreter, so the program has no access to any platform-specific APIs or FFI. A few APIs have been implemented (such as printing to stdout) but most have not: for example, Miri currently does not support concurrency, or networking, or file system access, or gathering entropy from the system.

Running Miri on your own project (and its test suite)

Install Miri via rustup:

rustup component add miri

If rustup says the miri component is unavailable, that's because not all nightly releases come with all tools. Check out this website to determine a nightly version that comes with Miri and install that, e.g. using rustup install nightly-2019-03-28.

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 miri test.
  3. If you have a binary project, you can run it through Miri using cargo miri run.

The first time you run Miri, it will perform some extra setup and install some dependencies. It will ask you for confirmation before installing anything. If you run Miri on CI, run cargo miri setup to avoid getting interactive questions.

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 miri run -- -Zmiri-disable-validation runs the program without validation of basic type invariants and references. cargo miri test -- -- -Zunstable-options --exclude-should-panic skips #[should_panic] tests, which is a good idea because Miri does not support unwinding or catching panics.

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.

"no mir for std::rt::lang_start_internal"

This means the sysroot you are using was not compiled with Miri in mind. This should never happen when you use cargo miri because that takes care of setting up the sysroot. If you are using miri (the Miri driver) directly, see below for how to set up the sysroot.

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 or a shim for an external function being called) can be done by working just on the Miri side.

To prepare, make sure you are using a nightly Rust compiler. Then you should be able to just cargo build Miri.

In case this fails, your nightly might be incompatible with Miri master. The rust-version file contains the commit hash of rustc that Miri is currently tested against; you can use that to find a nightly that works or you might have to wait for the next nightly to get released.

Testing the Miri driver

The Miri driver in the miri binary is the "heart" of Miri: it is basically a version of rustc that, instead of compiling your code, runs it. It accepts all the same flags as rustc (though the ones only affecting code generation and linking obviously will have no effect) and more.

To run the Miri driver, you need to have the MIRI_SYSROOT environment variable set to an appropriate sysroot. You can generate such a sysroot with the following incantation:

cargo run --bin cargo-miri -- miri setup

This basically runs the cargo-miri binary (which backs the cargo miri subcommand) with cargo, and asks it to setup. It should in the end print the directory where the libstd was built. In the following, we will assume it is ~/.cache/miri/HOST; you may have to adjust that if you are not using Linux.

Now you can run the driver directly using

MIRI_SYSROOT=~/.cache/miri/HOST cargo run tests/run-pass/format.rs # or whatever test you like

and you can run the test suite using

cargo test

We recommend adding the --release flag to make tests run faster.

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).

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. the following helps 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 raised.

Testing cargo miri

Working with the driver directly gives you full control, but you also lose all the convenience provided by cargo. Once your test case depends on a crate, it is probably easier to test it with the cargo wrapper. You can install your development version of Miri using

cargo install --path . --force

and then you can use it as if it was installed by rustup. Make sure you use the same toolchain when calling cargo miri that you used when installing Miri!

There's a test for the cargo wrapper in the test-cargo-miri directory; run ./run-test.py in there to execute it.

Using a locally built rustc

A big part of the Miri driver lives in rustc, so working on Miri will sometimes 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 above for how to proceed working with the Miri driver. Notice that rustc's sysroot is already built for Miri in this case, so you can set MIRI_SYSROOT=$(rustc --print sysroot).

Running cargo miri in this setup is a bit more complicated, because the Miri binary you just created needs help to find the libraries it links against. On Linux, you can set the rpath to make this "just work":

export RUSTFLAGS="-C link-args=-Wl,-rpath,$(rustc --print sysroot)/lib/rustlib/x86_64-unknown-linux-gnu/lib"
cargo install --path . --force

Miri -Z flags and environment variables

Several -Z flags are relevant for Miri:

  • -Zmiri-seed=<hex> is a custom -Z flag added by Miri. It enables the interpreted program to seed an RNG with system entropy. Miri will keep an RNG on its own that is seeded with the given seed, and use that to generate the "system entropy" that seeds the RNG(s) in the interpreted program. NOTE: This entropy is not good enough for cryptographic use! Do not generate secret keys in Miri or perform other kinds of cryptographic operations that rely on proper random numbers.
  • -Zmiri-disable-validation 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. However, this can also help to make Miri run faster.
  • -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.

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 and beyond, which we collect here.

Definite bugs found:

Violations of Stacked Borrows found that are likely bugs (but Stacked Borrows is currently just an experiment):

License

Licensed under either of

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.