Stream: git-wasmtime

Topic: wasmtime / Issue #2669 Regression: hard links created by ...

Wasmtime GitHub notifications bot (Feb 19 2021 at 13:48):

RReverser edited Issue #2669:

_Originally reported on the Rust repo at https://github.com/rust-lang/rust/issues/82277_

I tried this code:

fn main() {
    std::fs::hard_link("/a", "/b").unwrap();
}

I compiled it with --target wasm32-wasi and executed with latest Wasmtime 0.23.0 with wasmtime --mapdir=/::. where the current directory contains valid file a.

I expected to see this happen:

A hard link b is created that points to a.

Instead, this happened:

It fails with an error:

thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Os { code: 28, kind: InvalidInput, message: "Invalid argument" }', temp.rs:2:36
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Error: failed to run main module `temp.wasm`

Caused by:
    0: failed to invoke command default
    1: wasm trap: unreachable
       wasm backtrace:
           0: 0x7b4c - <unknown>!__rust_start_panic
           1: 0x75e5 - <unknown>!rust_panic
           2: 0x71bd - <unknown>!std::panicking::rust_panic_with_hook::hb9189a2ae1f00dd6
           3: 0x6928 - <unknown>!std::panicking::begin_panic_handler::{{closure}}::ha38be87363679287
           4: 0x6869 - <unknown>!std::sys_common::backtrace::__rust_end_short_backtrace::h955bbbeea72dc4a0
           5: 0x7061 - <unknown>!rust_begin_unwind
           6: 0xd0a6 - <unknown>!core::panicking::panic_fmt::h904ce09f3cb14707
           7: 0xdbcc - <unknown>!core::option::expect_none_failed::ha29c9c4296a18e58
           8:  0xfee - <unknown>!core::result::Result<T,E>::unwrap::hcf72e0af954224e8
           9:  0x9b7 - <unknown>!temp::main::hdc57b3c35eac1777
          10:  0xac2 - <unknown>!core::ops::function::FnOnce::call_once::h9d9b1f2044973e11
          11:  0x8ac - <unknown>!std::sys_common::backtrace::__rust_begin_short_backtrace::h0d198c55a40ffd2e
          12:  0xdf2 - <unknown>!std::rt::lang_start::{{closure}}::he287b5106de2dbae
          13: 0x76f6 - <unknown>!std::rt::lang_start_internal::h81003f47a7b429a5
          14:  0xdc6 - <unknown>!std::rt::lang_start::h75812a9706ce1699
          15:  0x9e8 - <unknown>!__original_main
          16:  0x3f9 - <unknown>!_start
       note: run with `WASMTIME_BACKTRACE_DETAILS=1` environment variable to display more information

Meta

rustc --version --verbose:

rustc 1.49.0 (e1884a8e3 2020-12-29)
binary: rustc
commit-hash: e1884a8e3c3e813aada8254edfa120e85bf5ffca
commit-date: 2020-12-29
host: x86_64-unknown-linux-gnu
release: 1.49.0

wasmtime --version:

wasmtime 0.23.0

</p>
</details>

Other attempts

Equivalent C code

I've also tried to compile an equivalent C code:

#include <unistd.h>

int main() {
    return link("/a", "/b");
}

When this is compiled with WASI SDK clang --target=wasm32-wasi and executed with the same command in Wasmtime, it successfully creates the hardlink.

Node.js WASI

I've also tried to execute those Wasm files with Node.js WASI support, and both Wasm files create hard links successfully.

Older Wasmtime

Finally, I've tried to execute both with Wasmtime 0.22, and both also create hardlinks successfully, indicating a regression in Wasmtime itself.

Wasmtime GitHub notifications bot (Feb 19 2021 at 16:01):

sunfishcode commented on Issue #2669:

The short answer is it's a bug in Rust's WASI support; see https://github.com/rust-lang/rust/pull/81984 for a fix. Shorter-term fixes are also possible, so if you need something sooner, let's discuss it.

The longer answer is:

POSIX's linkat function has a flags argument which can be AT_SYMLINK_FOLLOW, allowing the user to control whether symlinks are followed. Unlike open's O_NOFOLLOW however, which fails on a symlink, linkat without AT_SYMLINK_FOLLOW succeeds and simply creates a link to the symlink.

WASI's link function is modeled after CloudABI's which is modeled after POSIX's, so it has an equivalent of AT_SYMLINK_FOLLOW. However, it's not safe to implement it in terms of a host call to linkat with AT_SYMLINK_FOLLOW set, because that would blindly follow the symlink and could link to something outside the sandbox. Since there's no way to make linkat fail on a symlink, there's no way to implement a sandboxed linkat with AT_SYMLINK_FOLLOW without introducing race conditions.

linkat with AT_SYMLINK_FOLLOW is rare in practice anyway, so I'm preparing to propose a change to WASI to not support it. Wasmtime is admittedly ahead of the curve in that it already doesn't support it. Creating hard links is pretty uncommon in general, and even moreso in code intended to be portable, so I guessed that this might not come up, but I appear to have been wrong :-}.

Wasmtime GitHub notifications bot (Feb 19 2021 at 17:16):

RReverser commented on Issue #2669:

@sunfishcode Thanks for the detailed explanation! It's always exciting to be hitting bugs that live in a crossover of spec and several implementations :sweat_smile:

Last updated: Dec 23 2024 at 12:05 UTC