x86_64-unknown-linux-musl binaries occasionally segfault during very-large searches #3494

Description

Please tick this box to confirm you have reviewed the above.

I have a different issue.

What version of ripgrep are you using?

ripgrep 15.2.0 (rev e89fff8)

features:+pcre2

simd(compile):+SSE2,-SSSE3,-AVX2

simd(runtime):+SSE2,+SSSE3,+AVX2

PCRE2 10.45 is available (JIT is available)

How did you install ripgrep?

I originally encountered this bug in the rg bundled with OpenAI Codex. That binary is byte-for-byte identical with the one in https://github.com/BurntSushi/ripgrep/releases/download/15.2.0/ripgrep-15.2.0-x86_64-unknown-linux-musl.tar.gz and I've reproduced the bug from that independently of any Codex dependency. For the analysis below, I built rg-15.2 with debug symbols included by way of CROSS_CONTAINER_ENGINE=podman CARGO_PROFILE_RELEASE_DEBUG=true ~/.cargo/bin/cross build --release --target x86_64-unknown-linux-musl.

What operating system are you using ripgrep on?

OpenSUSE Tumbleweed Linux x86_64

Describe your bug.

Ripgrep built for x86_64-unknown-linux-musl occasionally crashes with a SIGSEGV when searching very-large trees at a high degree of concurrency. The crashing line is an integrity assertion regarding heap metadata inside MUSL's mallocng, in a calloc call made from opendir. The complete backtrace is below.

What are the steps to reproduce the behavior?

Having a sufficiently large search tree seems to be essential for reproduction. Run the attached generate_repro_tree.py. This is an LLM-written program which produces a tree full of random files which mimic the statistics of the repo in which I originally encountered the bug. It will produce a tree containing roughly 20GiB of data across 1.8M files.

Then from the root of that tree, run rg in a loop, searching for some arbitrary literal string that isn't present in the tree: while true; do rg tnoheueunotshisnthukoethnsueothnsiuothonesuioseuinth; done. On my 24-core system, having enough free RAM for the search tree to fit in the kernel's block cache, it typically takes about a minute for the SIGSEGV to appear.

What is the actual behavior?

I get a coredump with the following backtrace:

```

0 get_meta () at ../src_musl/src/malloc/mallocng/meta.h:141

1 __malloc_allzerop () at ../src_musl/src/malloc/mallocng/malloc.c:384

2 0x00007f71f8381b2d in calloc () at ../src_musl/src/malloc/calloc.c:41

3 0x00007f71f83810f4 in opendir () at ../src_musl/src/dirent/opendir.c:15

4 0x00007f71f835c133 in std::sys::fs::unix::readdir::{closure#0} () at library/std/src/sys/fs/unix.rs:2081

5 std::sys::helpers::small_c_string::run_with_cstr_stack<*mut libc::unix::DIR> () at library/std/src/sys/helpers/small_c_string.rs:48

6 std::sys::helpers::small_c_string::run_with_cstr<*mut libc::unix::DIR> () at library/std/src/sys/helpers/small_c_string.rs:28

7 std::sys::helpers::small_c_string::run_path_with_cstr<*mut libc::unix::DIR> () at library/std/src/sys/helpers/small_c_string.rs:18

8 std::sys::fs::unix::readdir () at library/std/src/sys/fs/unix.rs:2081

9 std::sys::fs::read_dir () at library/std/src/sys/fs/mod.rs:68

10 0x00007f71f8206b5c in std::fs::read_dir<&std::path::Path> (path=...)

at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/fs.rs:3265

11 ignore::walk::Work::read_dir (self=0x7f71f5bfeb20) at crates/ignore/src/walk.rs:1551

12 ignore::walk::Worker::run_one (self=0x7f71f5bfef08, work=...) at crates/ignore/src/walk.rs:1749

13 ignore::walk::Worker::run (self=...) at crates/ignore/src/walk.rs:1697

14 0x00007f71f821c866 in ignore::walk::{impl#15}::visit::{closure#0}::{closure#1}::{closure#0} () at crates/ignore/src/walk.rs:1463

15 std::sys::backtrace::__rust_begin_short_backtrace (f=...)

at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/sys/backtrace.rs:166

16 0x00007f71f8224596 in std::thread::lifecycle::spawn_unchecked::{closure#1}::{closure#0}

() at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/thread/lifecycle.rs:70

17 core::panic::unwind_safe::{impl#23}::call_once<(), std::thread::lifecycle::spawn_unchecked::{closure#1}::{closure_env#0}> (self=...)

at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/panic/unwind_safe.rs:275

18 std::panicking::catch_unwind::do_call>, ()> (data=)

at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/panicking.rs:581

19 std::panicking::catch_unwind<(), core::panic::unwind_safe::AssertUnwindSafe>> (f=...)

at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/panicking.rs:544

20 std::panic::catch_unwind>, ()> (f=...)

at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/panic.rs:359

21 std::thread::lifecycle::spawn_unchecked::{closure#1} ()

at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/std/src/thread/lifecycle.rs:68

22 core::ops::function::FnOnce::call_once, ()> () at /home/dfranke/.rustup/toolchains/stable-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/core/src/ops/function.rs:250

23 0x00007f71f8361fcf in alloc::boxed::{impl#31}::call_once<(), (dyn core::ops::function::FnOnce<(), Output=()> + core::marker::Send), alloc::alloc::Global> ()

at library/alloc/src/boxed.rs:2275

24 std::sys::thread::unix::{impl#2}::new::thread_start () at library/std/src/sys/thread/unix.rs:118

25 0x00007f71f8388788 in start () at ../src_musl/src/thread/pthread_create.c:207

26 0x00007f71f8389e6c in __clone () at ../src_musl/src/thread/x86_64/clone.s:22

```
Here is the core dump and the corresponding rg binary which produced it.

What is the expected behavior?

Not a segfault.

Newspaper

RipGrep musl binaries occasionally segfault during very-large searches