fix(child_process): spawn via posix_spawn to avoid a macOS fork/dyld deadlock

changelog.d/7157-macos-child-process-posix-spawn.md

@@ -0,0 +1 @@
+**macOS `child_process` fork/dyld deadlock:** `exec`/`spawn`/`execFile` (and their `Sync` forms) could deadlock a child on macOS. `std::process::Command` falls back from `posix_spawn` to `fork`+`exec` whenever a bare command name is combined with an `env` option (`env_clear()` sets `env_saw_path()`), and `fork` from Perry's multithreaded runtime (async reactor + GC/worker threads) leaves the child holding locks/Mach state from parent threads that no longer exist — so a fast child like `sh -c "echo hi"` hangs post-`exec` in dyld (`RemoteNotificationResponder::blockOnSynchronousEvent`), the reader/waiter threads block in `read()`/`wait4()`, and the main loop idles in `js_wait_for_event`. Perry now resolves a bare command name to its absolute path in the child's effective PATH before building the `Command`, keeping std on the `posix_spawn` fast path (`argv[0]` preserved via `arg0`); the `exec` shell uses the absolute `/bin/sh`. Verified with a dyld interposer: `env`-carrying `exec`/`spawnSync` go from `fork()` to `posix_spawn` with output/exit-code/argv capture unchanged. `detached` (setsid), `fork()`'s IPC `dup2`, and uid/gid necessarily keep std's fork path (not expressible through std's `posix_spawn`) and are documented inline. Linux behavior is unchanged.

crates/perry-runtime/src/child_process/exec.rs

@@ -46,7 +46,10 @@ pub extern "C" fn js_child_process_exec_sync(
     // Execute the command using the shell, honoring `cwd`/`env` options.
     #[cfg(unix)]
     let mut command = {
-        let mut c = Command::new("sh");
+        // Absolute path (Node's `exec` shell) keeps std on `posix_spawn`
+        // instead of the `fork`+`exec` fallback a bare "sh" + `env` triggers
+        // (the macOS fork/dyld deadlock fix — see `cp_command_for_program`).
+        let mut c = Command::new("/bin/sh");
         c.arg("-c").arg(&cmd_str);
         c
     };
@@ -277,7 +280,10 @@ pub extern "C" fn js_child_process_exec(cmd_ptr: *const StringHeader, arg1: f64,
     // `env` from the options are applied here.
     #[cfg(unix)]
     let mut command = {
-        let mut c = Command::new("sh");
+        // Absolute path (Node's `exec` shell) keeps std on `posix_spawn`
+        // instead of the `fork`+`exec` fallback a bare "sh" + `env` triggers
+        // (the macOS fork/dyld deadlock fix — see `cp_command_for_program`).
+        let mut c = Command::new("/bin/sh");
         c.arg("-c").arg(&cmd_str);
         c
     };
@@ -354,7 +360,7 @@ pub extern "C" fn js_child_process_exec_file(
 
     // `cwd`/`env` come from the options slot; when `opts_val` is the callback
     // (`execFile(file, args, cb)`) it's a closure, so the helper no-ops.
-    let mut command = Command::new(&file_str);
+    let mut command = cp_command_for_program(&file_str, opts_val);
     command.args(&arg_strs);
     cp_apply_options(&mut command, opts_val);
     let run_options = cp_read_async_run_options(opts_val);
@@ -393,7 +399,7 @@ pub extern "C" fn js_child_process_exec_file_sync(
         return cp_box_output(b"", &mode);
     }
     let arg_strs = cp_args_from_value(args_val);
-    let mut command = Command::new(&file_str);
+    let mut command = cp_command_for_program(&file_str, opts_val);
     command.args(&arg_strs);
     cp_apply_argv0(&mut command, opts_val);
     cp_apply_options(&mut command, opts_val);
@@ -480,7 +486,10 @@ extern "C" fn cp_promisified_exec(_closure: *const ClosureHeader, cmd_val: f64,
     let cmd = cp_value_to_string(cmd_val).unwrap_or_default();
     #[cfg(unix)]
     let mut command = {
-        let mut c = Command::new("sh");
+        // Absolute path (Node's `exec` shell) keeps std on `posix_spawn`
+        // instead of the `fork`+`exec` fallback a bare "sh" + `env` triggers
+        // (the macOS fork/dyld deadlock fix — see `cp_command_for_program`).
+        let mut c = Command::new("/bin/sh");
         c.arg("-c").arg(&cmd);
         c
     };
@@ -501,7 +510,9 @@ extern "C" fn cp_promisified_exec_file(
 ) -> f64 {
     let file = cp_value_to_string(file_val).unwrap_or_default();
     let arg_strs = cp_args_from_value(args_val);
-    let mut command = Command::new(&file);
+    // The 2-arg promisify(execFile) wrapper has no options slot; resolve a bare
+    // program against the parent PATH to keep std on `posix_spawn`.
+    let mut command = cp_command_for_program(&file, cp_undefined());
     command.args(&arg_strs);
     // The 2-arg promisify(execFile) wrapper has no options slot.
     cp_promisified_run(

crates/perry-runtime/src/child_process/mod.rs

@@ -100,10 +100,13 @@ pub(crate) use builder::{
 };
 
 // options.rs — command option application (cwd/env/uid/gid/argv0/detached/stdio).
+#[cfg(unix)]
+pub(crate) use options::cp_resolve_program_path;
 pub(crate) use options::{
     cp_abort_signal_is_aborted, cp_apply_argv0, cp_apply_detached, cp_apply_live_stdio,
-    cp_apply_options, cp_build_command, cp_read_abort_signal, cp_read_stdio, cp_spawnargs_argv0,
-    cp_stdio_from_fd, cp_stdio_js_value, cp_stdio_stream_fd, CpStdio,
+    cp_apply_options, cp_build_command, cp_command_for_program, cp_read_abort_signal,
+    cp_read_stdio, cp_spawnargs_argv0, cp_stdio_from_fd, cp_stdio_js_value, cp_stdio_stream_fd,
+    CpStdio,
 };
 
 // output.rs — output encoding, error shape, exit decoding.

crates/perry-runtime/src/child_process/options.rs

@@ -140,6 +140,12 @@ pub(crate) fn cp_apply_detached(command: &mut Command, opts_val: f64) {
     #[cfg(unix)]
     {
         use std::os::unix::process::CommandExt;
+        // NOTE: a `pre_exec` closure forces std onto the `fork`+`exec` path
+        // (posix_spawn cannot run arbitrary code), so `detached` children do not
+        // benefit from the posix_spawn fork/dyld-deadlock fix. `setsid` is not
+        // expressible through std's posix_spawn wrapper (no `POSIX_SPAWN_SETSID`
+        // knob), and `detached` is a rare, deliberate full-session-detach —
+        // unlike the common `exec`/`spawn` paths, it is not converted here.
         unsafe {
             command.pre_exec(|| {
                 if libc::setsid() < 0 {
@@ -370,6 +376,98 @@ fn cp_default_shell() -> String {
     }
 }
 
+/// Fallback search path when a child's environment carries no `PATH` — the same
+/// default `execvp(3)` uses (`_PATH_DEFPATH`).
+#[cfg(unix)]
+const CP_DEFAULT_PATH: &str = "/usr/bin:/bin:/usr/sbin:/sbin";
+
+/// The child's effective `PATH` for resolving a bare command name. When an `env`
+/// option is present the child's environment *replaces* the parent's (Node
+/// semantics), so its `PATH` — not the parent's — governs the lookup; a missing
+/// `PATH` falls back to the `execvp` default. With no `env` option the parent's
+/// `PATH` applies.
+#[cfg(unix)]
+fn cp_effective_path(opts_val: f64) -> String {
+    if cp_object_ptr(opts_val).is_some() {
+        let env_val = cp_get_field(opts_val, b"env");
+        if cp_object_ptr(env_val).is_some() {
+            if let Some(p) = cp_value_to_string(cp_get_field(env_val, b"PATH")) {
+                if !p.is_empty() {
+                    return p;
+                }
+            }
+            return CP_DEFAULT_PATH.to_string();
+        }
+    }
+    std::env::var("PATH").unwrap_or_else(|_| CP_DEFAULT_PATH.to_string())
+}
+
+/// Whether `path` names an executable regular file (following symlinks).
+#[cfg(unix)]
+fn cp_is_executable(path: &std::path::Path) -> bool {
+    use std::os::unix::fs::PermissionsExt;
+    std::fs::metadata(path)
+        .map(|m| m.is_file() && (m.permissions().mode() & 0o111 != 0))
+        .unwrap_or(false)
+}
+
+/// Resolve a bare command name to an absolute path by walking `path` (a
+/// colon-separated `PATH`), returning the first executable match. An empty
+/// `PATH` element means the current directory (POSIX `execvp` semantics).
+#[cfg(unix)]
+pub(crate) fn cp_resolve_program_path(program: &str, path: &str) -> Option<String> {
+    for dir in path.split(':') {
+        let base = if dir.is_empty() { "." } else { dir };
+        let candidate = std::path::Path::new(base).join(program);
+        if cp_is_executable(&candidate) {
+            return candidate.into_os_string().into_string().ok();
+        }
+    }
+    None
+}
+
+/// Build a `Command` for `program`, resolving a bare command name to its
+/// absolute path in the child's effective `PATH`.
+///
+/// This is the macOS fork/dyld deadlock fix. `std::process::Command` uses
+/// `posix_spawn` only when the program is given as a path *and* no
+/// `pre_exec`/uid/gid closures are set; a bare command name combined with an
+/// `env` option (which triggers `env_clear()`) drops it onto the `fork`+`exec`
+/// fallback (see `library/std/src/sys/pal/unix/process/process_unix.rs`,
+/// `env_saw_path() && !program_is_path()`). On macOS a `fork` from Perry's
+/// multithreaded runtime (async reactor + GC/worker threads) can deadlock the
+/// child post-`exec` in dyld (`RemoteNotificationResponder::
+/// blockOnSynchronousEvent`) when the process is being observed by a Mach
+/// notification port (telemetry, a crash reporter, a debugger): the child
+/// inherits locks/Mach state from parent threads that don't exist after
+/// `fork`. Resolving the name to an absolute path here keeps std on the
+/// `posix_spawn` fast path.
+///
+/// The original name is preserved as `argv[0]` (`arg0`) so the child sees the
+/// same `process.argv[0]` it would have gotten from the bare name. When nothing
+/// resolves we fall back to the bare name unchanged — a genuine `ENOENT` never
+/// `exec`s a real image, so it cannot hit the dyld hang, and the error surface
+/// stays identical.
+pub(crate) fn cp_command_for_program(program: &str, opts_val: f64) -> Command {
+    #[cfg(unix)]
+    {
+        if !program.is_empty() && !program.contains('/') {
+            let path = cp_effective_path(opts_val);
+            if let Some(abs) = cp_resolve_program_path(program, &path) {
+                use std::os::unix::process::CommandExt;
+                let mut command = Command::new(abs);
+                command.arg0(program);
+                return command;
+            }
+        }
+    }
+    #[cfg(not(unix))]
+    {
+        let _ = opts_val;
+    }
+    Command::new(program)
+}
+
 /// Whether a self-launch uses a Node CLI mode that evaluates source text.
 fn cp_should_use_node_interpreter(cmd: &str, args: &[String]) -> bool {
     let is_self = std::env::args().next().as_deref() == Some(cmd)
@@ -423,14 +521,16 @@ pub(crate) fn cp_build_command(cmd: &str, args: &[String], opts_val: f64) -> Com
             line.push(' ');
             line.push_str(a);
         }
-        let mut c = Command::new(shell_bin);
+        // Resolve a bare shell name to an absolute path so std stays on
+        // `posix_spawn` (see `cp_command_for_program`).
+        let mut c = cp_command_for_program(&shell_bin, opts_val);
         #[cfg(windows)]
         c.arg("/d").arg("/s").arg("/c").arg(line);
         #[cfg(not(windows))]
         c.arg("-c").arg(line);
         c
     } else {
-        let mut c = Command::new(program);
+        let mut c = cp_command_for_program(&program, opts_val);
         c.args(args);
         c
     };
@@ -441,6 +541,71 @@ pub(crate) fn cp_build_command(cmd: &str, args: &[String], opts_val: f64) -> Com
     command
 }
 
+#[cfg(all(test, unix))]
+mod posix_spawn_tests {
+    use super::{cp_command_for_program, cp_resolve_program_path};
+    use crate::child_process::cp_undefined;
+
+    /// A bare command name in a real `PATH` resolves to an executable absolute
+    /// path — the precondition std needs to pick `posix_spawn` over `fork`.
+    #[test]
+    fn resolves_bare_name_to_absolute_executable() {
+        let resolved = cp_resolve_program_path("sh", "/nonexistent:/bin:/usr/bin")
+            .expect("sh should resolve on a POSIX system");
+        assert!(
+            resolved.starts_with('/'),
+            "expected an absolute path, got {resolved}"
+        );
+        assert!(resolved.ends_with("/sh"));
+        assert!(std::path::Path::new(&resolved).exists());
+    }
+
+    /// A name that cannot be found returns `None` (caller falls back to the bare
+    /// name, which then fails ENOENT before exec'ing any real image).
+    #[test]
+    fn missing_program_does_not_resolve() {
+        assert!(cp_resolve_program_path("perry-definitely-missing-xyz", "/bin:/usr/bin").is_none());
+    }
+
+    /// `cp_command_for_program` rewrites a bare name to an absolute path so std
+    /// stays on `posix_spawn`; an already-absolute program is passed through
+    /// unchanged.
+    #[test]
+    fn command_program_is_absolute_for_bare_name() {
+        let cmd = cp_command_for_program("sh", cp_undefined());
+        let program = cmd.get_program().to_string_lossy().into_owned();
+        assert!(
+            program.starts_with('/') && program.ends_with("/sh"),
+            "bare name should resolve to an absolute path; got {program}"
+        );
+
+        let passthrough = cp_command_for_program("/bin/sh", cp_undefined());
+        assert_eq!(passthrough.get_program().to_string_lossy(), "/bin/sh");
+    }
+
+    /// End-to-end: the resolved (absolute-path, no-`pre_exec`) command spawns via
+    /// std's `posix_spawn` path and captures output correctly. This is exactly
+    /// the shape that deadlocked in dyld when std took the `fork`+`exec` fallback
+    /// on macOS. (Full GC-stress N/N verification uses the compiled repro under
+    /// `PERRY_GC_FORCE_EVACUATE=1`; a raw unit test cannot drive Perry's
+    /// thread-local GC without runtime init.)
+    #[test]
+    fn resolved_command_runs_and_captures_output() {
+        let mut cmd = cp_command_for_program("sh", cp_undefined());
+        assert!(
+            cmd.get_program().to_string_lossy().starts_with('/'),
+            "resolved program must be an absolute path to keep std on posix_spawn"
+        );
+        let out = cmd
+            .arg("-c")
+            .arg("printf ok-%s 42")
+            .output()
+            .expect("spawn resolved sh");
+        assert!(out.status.success());
+        assert_eq!(out.stdout, b"ok-42");
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::cp_should_use_node_interpreter;

crates/perry-runtime/src/child_process/registry.rs

@@ -85,7 +85,55 @@ pub extern "C" fn js_child_process_spawn_background(
             None => return std::ptr::null_mut(),
         };
 
-        let mut command = Command::new(&cmd_str);
+        // Parse the env JSON up front so we can read its `PATH` for command
+        // resolution below (an env override with a `PATH` key is what pushes
+        // std onto the `fork`+`exec` fallback for a bare command name).
+        let env_map = {
+            let env_bits = env_json_val.to_bits();
+            if env_bits != TAG_NULL_BITS && env_bits != TAG_UNDEFINED_BITS {
+                extract_string_from_nanboxed(env_json_val).and_then(|env_json| {
+                    serde_json::from_str::<serde_json::Map<String, serde_json::Value>>(&env_json)
+                        .ok()
+                })
+            } else {
+                None
+            }
+        };
+
+        // Resolve a bare command name to an absolute path so std uses
+        // `posix_spawn` instead of the `fork`+`exec` fallback that an env
+        // override triggers — the macOS fork/dyld deadlock fix (see
+        // `options::cp_command_for_program`). `arg0` preserves argv[0].
+        let mut command = {
+            #[cfg(unix)]
+            {
+                let resolved = if cmd_str.contains('/') {
+                    None
+                } else {
+                    let path = env_map
+                        .as_ref()
+                        .and_then(|m| m.get("PATH"))
+                        .and_then(|v| v.as_str())
+                        .map(str::to_string)
+                        .or_else(|| std::env::var("PATH").ok())
+                        .unwrap_or_default();
+                    super::cp_resolve_program_path(&cmd_str, &path)
+                };
+                match resolved {
+                    Some(abs) => {
+                        use std::os::unix::process::CommandExt;
+                        let mut c = Command::new(abs);
+                        c.arg0(&cmd_str);
+                        c
+                    }
+                    None => Command::new(&cmd_str),
+                }
+            }
+            #[cfg(not(unix))]
+            {
+                Command::new(&cmd_str)
+            }
+        };
 
         // Add arguments if provided
         if args_ptr != 0 {
@@ -102,18 +150,11 @@ pub extern "C" fn js_child_process_spawn_background(
             }
         }
 
-        // Parse env JSON if provided (not null/undefined)
-        let env_bits = env_json_val.to_bits();
-        if env_bits != TAG_NULL_BITS && env_bits != TAG_UNDEFINED_BITS {
-            if let Some(env_json) = extract_string_from_nanboxed(env_json_val) {
-                if let Ok(map) =
-                    serde_json::from_str::<serde_json::Map<String, serde_json::Value>>(&env_json)
-                {
-                    for (k, v) in map {
-                        if let Some(val_str) = v.as_str() {
-                            command.env(k, val_str);
-                        }
-                    }
+        // Apply the parsed env (string values only), matching prior behavior.
+        if let Some(map) = env_map {
+            for (k, v) in map {
+                if let Some(val_str) = v.as_str() {
+                    command.env(k, val_str);
                 }
             }
         }