Skip to main content

harn_hostlib/process/
handle.rs

1//! Process abstraction trait used by `tools/proc` and
2//! `tools/long_running`.
3//!
4//! Tier 1C of the de-flake epic (#1057). Production code spawns through
5//! the [`ProcessSpawner`] trait — the default implementation in
6//! `process::real` wraps `std::process::Child` and goes through
7//! `harn_vm::process_sandbox`. Tests install a `MockSpawner` (see
8//! `process::mock`) that returns deterministic [`MockProcess`] handles,
9//! so process-tool tests no longer depend on real subprocess scheduling
10//! or wall-clock timing.
11
12use std::collections::BTreeMap;
13use std::io::{self, Read, Write};
14use std::path::PathBuf;
15use std::sync::Arc;
16use std::time::Duration;
17
18pub use harn_vm::op_interrupt::{ProcessCleanupChild, ProcessCleanupReport};
19
20/// Resolved exit information for a finished process. Mirrors the subset of
21/// `std::process::ExitStatus` that the process-tool builtins surface.
22#[derive(Clone, Copy, Debug, PartialEq, Eq)]
23pub struct ExitStatus {
24    /// Exit code from `exit(2)` / `_exit(2)`. `None` means the process did not
25    /// exit normally (it was terminated by a signal).
26    pub code: Option<i32>,
27    /// Unix signal that terminated the process, when applicable. `None` on
28    /// non-Unix targets or when the process exited normally.
29    pub signal: Option<i32>,
30}
31
32impl ExitStatus {
33    /// Construct a normal exit with the given code.
34    pub fn from_code(code: i32) -> Self {
35        Self {
36            code: Some(code),
37            signal: None,
38        }
39    }
40
41    /// Construct a signal-terminated exit.
42    pub fn from_signal(signal: i32) -> Self {
43        Self {
44            code: None,
45            signal: Some(signal),
46        }
47    }
48}
49
50/// How a spawn should treat the parent's environment. Mirrors the legacy
51/// `EnvMode` from `tools/proc.rs`.
52#[derive(Clone, Copy, Debug, PartialEq, Eq)]
53pub enum EnvMode {
54    /// Inherit the parent's environment, then apply `env` overrides.
55    InheritClean,
56    /// Clear the environment, then apply `env`.
57    Replace,
58    /// Inherit the parent's environment and apply `env` (default behaviour).
59    Patch,
60}
61
62/// What should happen to a spawned process tree if its supervising Harn
63/// process exits without running normal handle cleanup.
64#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
65pub enum OwnerDeathPolicy {
66    /// Do not add an owner-lifetime guard.
67    #[default]
68    None,
69    /// Kill the independently contained worker tree when its owner dies.
70    KillContainment,
71}
72
73/// Explicit secret-bearing environment variable names that the agent's
74/// `run`/`command_run` tool must never leak into a child process (and thus
75/// into the model context, since the child's stdout is returned to the
76/// model as the tool result). These are matched case-insensitively in
77/// addition to the suffix patterns in [`is_sensitive_env_name`].
78const EXPLICIT_SENSITIVE_ENV_NAMES: &[&str] = &[
79    "GITHUB_TOKEN",
80    "GH_TOKEN",
81    "HARN_CLOUD_API_KEY",
82    "BURIN_ADMIN_TOKEN",
83    "AWS_SECRET_ACCESS_KEY",
84    "AWS_SESSION_TOKEN",
85];
86
87/// Provider-namespace prefixes whose entire family of variables is treated
88/// as secret-bearing (e.g. `ANTHROPIC_API_KEY`, `OPENAI_ORG_ID`). Matched
89/// case-insensitively against the start of the variable name.
90const SENSITIVE_ENV_PREFIXES: &[&str] = &[
91    "ANTHROPIC_",
92    "OPENAI_",
93    "OPENROUTER_",
94    "FIREWORKS_",
95    "TOGETHER_",
96    "XAI_",
97    "GROQ_",
98];
99
100/// Returns `true` when an environment variable name looks like it carries a
101/// secret (provider API key, access token, OAuth client secret, etc.) and so
102/// must be stripped from a child process spawned by the agent's `run` tool.
103///
104/// The check is deliberately conservative about credentials but permissive
105/// about ordinary build/toolchain variables: `PATH`, `HOME`, `LANG`,
106/// `CARGO_HOME`, language toolchain vars, etc. are *not* sensitive and stay
107/// in the child environment so builds and tests still work.
108///
109/// Matching is case-insensitive and covers:
110/// - suffix patterns `_API_KEY`, `_TOKEN`, `_SECRET`, `_KEY`, `_PASSWORD`,
111///   `_PASSWD`, `_CREDENTIALS`;
112/// - the provider prefixes in [`SENSITIVE_ENV_PREFIXES`];
113/// - the explicit names in [`EXPLICIT_SENSITIVE_ENV_NAMES`].
114pub fn is_sensitive_env_name(name: &str) -> bool {
115    let upper = name.to_ascii_uppercase();
116    if EXPLICIT_SENSITIVE_ENV_NAMES.contains(&upper.as_str()) {
117        return true;
118    }
119    if SENSITIVE_ENV_PREFIXES
120        .iter()
121        .any(|prefix| upper.starts_with(prefix))
122    {
123        return true;
124    }
125    // Suffix patterns catch the long tail of provider/service credentials
126    // (`*_API_KEY`, `*_TOKEN`, `*_SECRET`, `*_KEY`, `*_PASSWORD`, `*_PASSWD`,
127    // `*_CREDENTIALS`) without enumerating every vendor. `_KEY` is the broadest;
128    // these suffixes still exclude benign names like `PATH`/`HOME`/`LANG` and
129    // `BUILD_KEYBOARD` that don't end in one of them.
130    upper.ends_with("_API_KEY")
131        || upper.ends_with("_TOKEN")
132        || upper.ends_with("_SECRET")
133        || upper.ends_with("_KEY")
134        || upper.ends_with("_PASSWORD")
135        || upper.ends_with("_PASSWD")
136        || upper.ends_with("_CREDENTIALS")
137}
138
139/// Parameters describing a single spawn. The spawner is responsible for any
140/// sandbox setup (Linux seccomp/landlock, macOS sandbox-exec, etc.) and for
141/// configuring the child's process group when requested.
142#[derive(Clone, Debug)]
143pub struct SpawnSpec {
144    /// Builtin name surfaced in error messages (e.g. `"hostlib_tools_run_command"`).
145    pub builtin: &'static str,
146    /// Program to execute. Must be non-empty (validated by the spawner).
147    pub program: String,
148    /// Arguments to pass to the program.
149    pub args: Vec<String>,
150    /// Working directory for the child. `None` inherits the parent's cwd.
151    pub cwd: Option<PathBuf>,
152    /// Environment overrides to apply (interpretation depends on `env_mode`).
153    pub env: BTreeMap<String, String>,
154    /// Variable names to strip from the inherited environment before `env`
155    /// overrides apply. A key present in both `env_remove` and `env` ends up
156    /// set (explicit overrides win). No effect under `EnvMode::Replace`,
157    /// which already starts from an empty environment.
158    pub env_remove: Vec<String>,
159    /// How to treat the parent's environment.
160    pub env_mode: EnvMode,
161    /// Whether stdin is available. Captured-output processes receive a pipe;
162    /// inherited-output processes inherit the caller's terminal. `false`
163    /// connects the child to the null device.
164    pub use_stdin: bool,
165    /// Set the child's process group to its own pid (`setpgid(0, 0)`). Used
166    /// for long-running handles so the kill-by-pgid path works.
167    pub configure_process_group: bool,
168    /// Process-tree behavior when the supervising Harn process dies.
169    pub owner_death: OwnerDeathPolicy,
170    /// How stdout/stderr should be attached for this child.
171    pub output_capture: OutputCapture,
172}
173
174/// Child stdout/stderr attachment strategy for a spawned process.
175#[derive(Clone, Debug)]
176pub enum OutputCapture {
177    /// Inherit the caller's stdout and stderr. When [`SpawnSpec::use_stdin`]
178    /// is true, stdin is inherited too; captured-output modes instead expose a
179    /// writable pipe. Intended for supervised CLI workloads whose output
180    /// remains user-facing rather than model-facing.
181    Inherit,
182    /// Capture stdout/stderr through pipes owned by the parent process.
183    Pipe,
184    /// Capture stdout/stderr through pre-created private files.
185    File {
186        /// Path attached to the child's stdout handle.
187        stdout_path: PathBuf,
188        /// Path attached to the child's stderr handle.
189        stderr_path: PathBuf,
190    },
191}
192
193/// Handle to a running (or finished) process. Used by both the synchronous
194/// `proc::run` path and the long-running waiter thread.
195///
196/// The trait is intentionally small: the legacy code already managed
197/// stdout/stderr drain on dedicated threads, and stdin is written once after
198/// spawn — wrapping those reads/writes via boxed trait objects keeps the
199/// real and mock paths uniform without forcing async into the rest of the
200/// hostlib.
201pub trait ProcessHandle: Send {
202    /// OS process id, when available.
203    fn pid(&self) -> Option<u32>;
204
205    /// OS process group id, when available. Falls back to [`Self::pid`] on
206    /// platforms that don't expose process groups.
207    fn process_group_id(&self) -> Option<u32>;
208
209    /// Returns a killer that can terminate the process even after the
210    /// stdout/stderr/wait halves have been moved into the waiter thread.
211    fn killer(&self) -> Arc<dyn ProcessKiller>;
212
213    /// Take ownership of the stdin pipe, if the spawn requested one.
214    fn take_stdin(&mut self) -> Option<Box<dyn Write + Send>>;
215
216    /// Take ownership of the stdout reader.
217    fn take_stdout(&mut self) -> Option<Box<dyn Read + Send>>;
218
219    /// Take ownership of the stderr reader.
220    fn take_stderr(&mut self) -> Option<Box<dyn Read + Send>>;
221
222    /// Wait for the process to exit, optionally with a timeout, while
223    /// polling `interrupt`. On timeout the spawner kills the child process
224    /// tree (SIGKILL, historical semantics) and reports
225    /// [`WaitOutcome::TimedOut`]. When `interrupt` returns `true` (scope
226    /// cancellation, `deadline` expiry — see `harn_vm::op_interrupt`) the
227    /// spawner gracefully terminates the child's process tree (SIGTERM,
228    /// then SIGKILL after `harn_vm::op_interrupt::SUBPROCESS_TERM_GRACE`)
229    /// and reports [`WaitOutcome::Interrupted`].
230    fn wait_with_timeout(
231        &mut self,
232        timeout: Option<Duration>,
233        interrupt: &dyn Fn() -> bool,
234    ) -> io::Result<WaitOutcome>;
235
236    /// Block until the process exits, no timeout, no interrupt polling.
237    /// Used by the background (`background: true`) waiter thread, whose
238    /// children deliberately outlive scope cancellation and deadlines.
239    fn wait(&mut self) -> io::Result<ExitStatus>;
240}
241
242/// How a [`ProcessHandle::wait_with_timeout`] ended.
243#[derive(Clone, Debug, PartialEq, Eq)]
244pub enum WaitOutcome {
245    /// The process exited on its own.
246    Exited(ExitStatus),
247    /// The timeout elapsed; the spawner killed the child process tree.
248    TimedOut(ProcessCleanupReport),
249    /// The interrupt callback fired; the spawner gracefully terminated the
250    /// child's process tree.
251    Interrupted(ProcessCleanupReport),
252}
253
254/// Kill side of a [`ProcessHandle`]. Cloneable via `Arc` so cancellation
255/// works after the waiter thread has taken ownership of the handle itself.
256pub trait ProcessKiller: Send + Sync {
257    /// Send SIGKILL to the process tree/group when applicable.
258    fn kill(&self) -> ProcessCleanupReport;
259}
260
261/// Spawner abstraction: produces [`ProcessHandle`] instances.
262pub trait ProcessSpawner: Send + Sync {
263    /// Spawn the configured process.
264    fn spawn(&self, spec: SpawnSpec) -> Result<Box<dyn ProcessHandle>, ProcessError>;
265}
266
267/// Errors raised by a spawner. These map onto `HostlibError::Backend` /
268/// `HostlibError::InvalidParameter` at the call site so the script-side
269/// surface stays unchanged.
270#[derive(Clone, Debug, thiserror::Error)]
271pub enum ProcessError {
272    /// `argv` was empty or otherwise malformed.
273    #[error("invalid argv: {0}")]
274    InvalidArgv(String),
275    /// Sandbox setup (e.g. landlock policy assembly) failed.
276    #[error("sandbox setup failed: {0}")]
277    SandboxSetup(String),
278    /// Sandbox rejected the supplied cwd.
279    #[error("sandbox cwd rejected: {0}")]
280    SandboxCwd(String),
281    /// Sandbox rejected the spawn at execve time.
282    #[error("sandbox rejected spawn: {0}")]
283    SandboxSpawn(String),
284    /// Generic spawn failure (typically io::Error from `Command::spawn`).
285    #[error("spawn failed: {0}")]
286    Spawn(String),
287    /// A never-approvable UNIVERSAL catastrophic command (machine/disk/data
288    /// destruction) was rejected by the floor BEFORE spawning. Enforced
289    /// unconditionally at [`spawn_process`] — no `command_policy` required —
290    /// so it is universal across every hostlib process tool, embedders, and
291    /// standalone Harn. See
292    /// [`harn_vm::orchestration::universal_catastrophic_reason`].
293    #[error("{0}")]
294    CatastrophicFloor(String),
295}
296
297use std::cell::RefCell;
298
299thread_local! {
300    static THREAD_SPAWNER: RefCell<Option<Arc<dyn ProcessSpawner>>> = const { RefCell::new(None) };
301}
302
303/// Install a per-thread spawner used by `spawn_process` from this thread.
304/// Returns a guard that restores the previous spawner on drop. Tests use
305/// this to install a [`super::mock::MockSpawner`]; production never calls
306/// it (the default real spawner runs whenever no per-thread spawner is
307/// installed).
308///
309/// Thread-local rather than global so parallel test execution is safe.
310/// Process-tool spawns happen on the test's thread; the long-running
311/// waiter threads operate on the handle that was already returned, so
312/// they don't perform spawner lookups themselves.
313pub fn install_spawner(spawner: Arc<dyn ProcessSpawner>) -> SpawnerGuard {
314    let prev = THREAD_SPAWNER.with(|slot| slot.replace(Some(spawner)));
315    SpawnerGuard { prev: Some(prev) }
316}
317
318/// Guard returned by [`install_spawner`]. Restores the previous spawner on
319/// drop so installs nest correctly across tests.
320pub struct SpawnerGuard {
321    // Outer Option distinguishes "guard already restored" (None) from
322    // "guard owes a restore" (Some(_)); inner Option carries the previous
323    // spawner slot value (which can itself be None when no spawner was set).
324    #[allow(clippy::option_option)]
325    prev: Option<Option<Arc<dyn ProcessSpawner>>>,
326}
327
328impl Drop for SpawnerGuard {
329    fn drop(&mut self) {
330        if let Some(prev) = self.prev.take() {
331            THREAD_SPAWNER.with(|slot| {
332                *slot.borrow_mut() = prev;
333            });
334        }
335    }
336}
337
338/// Return the currently installed spawner for this thread, falling back
339/// to the default real spawner.
340pub fn current_spawner() -> Arc<dyn ProcessSpawner> {
341    THREAD_SPAWNER
342        .with(|slot| slot.borrow().clone())
343        .unwrap_or_else(super::real::default_spawner)
344}
345
346/// Spawn a process via the currently installed spawner.
347///
348/// This is the single chokepoint every hostlib process tool funnels through
349/// (`run_command`, `run_test`, `run_build_command`, `manage_packages`, and the
350/// long-running background path). The UNIVERSAL catastrophic-command floor is
351/// enforced HERE, UNCONDITIONALLY — no `command_policy` on the stack is
352/// required — so a machine/disk/data-destroying command (`rm -rf /`, fork bomb,
353/// `mkfs`, `dd of=<device>`, `chmod -R 000`, `truncate -s 0` of a source file,
354/// redirect-over-source, project-root delete) and the textual git-destructive
355/// family (`git reset --hard`, `git clean -fd`, force-push) is rejected before
356/// the child is ever created, on standalone Harn and under every embedder
357/// alike. Structured git builtins remain the reviewed path for legitimate
358/// force-with-lease workflows. The spec's raw `program`/`args` are classified
359/// BEFORE any sandbox wrapper is applied, so a `sandbox-exec`/`bwrap` prefix
360/// can't bury the real command.
361pub fn spawn_process(spec: SpawnSpec) -> Result<Box<dyn ProcessHandle>, ProcessError> {
362    validate_process_spec(&spec)?;
363    current_spawner().spawn(spec)
364}
365
366pub(crate) fn validate_process_spec(spec: &SpawnSpec) -> Result<(), ProcessError> {
367    let workspace_roots: Vec<String> = spec
368        .cwd
369        .as_ref()
370        .map(|cwd| vec![cwd.display().to_string()])
371        .unwrap_or_default();
372    if let Some(reason) = harn_vm::orchestration::universal_catastrophic_reason(
373        &spec.program,
374        &spec.args,
375        &workspace_roots,
376    ) {
377        return Err(ProcessError::CatastrophicFloor(reason));
378    }
379    Ok(())
380}
381
382#[cfg(test)]
383mod tests {
384    use super::is_sensitive_env_name;
385
386    #[test]
387    fn denies_secret_bearing_names() {
388        // Suffix patterns.
389        assert!(is_sensitive_env_name("ANTHROPIC_API_KEY"));
390        assert!(is_sensitive_env_name("OPENAI_API_KEY"));
391        assert!(is_sensitive_env_name("SOME_VENDOR_TOKEN"));
392        assert!(is_sensitive_env_name("MY_CLIENT_SECRET"));
393        assert!(is_sensitive_env_name("RANDOM_KEY"));
394        assert!(is_sensitive_env_name("DOCKER_PASSWORD"));
395        assert!(is_sensitive_env_name("TWINE_PASSWORD"));
396        assert!(is_sensitive_env_name("DATABASE_PASSWORD"));
397        assert!(is_sensitive_env_name("MYSQL_PASSWD"));
398        assert!(is_sensitive_env_name("GCP_CREDENTIALS"));
399        // Explicit names.
400        assert!(is_sensitive_env_name("GITHUB_TOKEN"));
401        assert!(is_sensitive_env_name("GH_TOKEN"));
402        assert!(is_sensitive_env_name("HARN_CLOUD_API_KEY"));
403        assert!(is_sensitive_env_name("BURIN_ADMIN_TOKEN"));
404        assert!(is_sensitive_env_name("AWS_SECRET_ACCESS_KEY"));
405        assert!(is_sensitive_env_name("AWS_SESSION_TOKEN"));
406        // Provider prefixes (even without a key/token suffix).
407        assert!(is_sensitive_env_name("OPENROUTER_BASE_URL"));
408        assert!(is_sensitive_env_name("FIREWORKS_ACCOUNT"));
409        assert!(is_sensitive_env_name("TOGETHER_ORG"));
410        assert!(is_sensitive_env_name("XAI_REGION"));
411        assert!(is_sensitive_env_name("GROQ_PROJECT"));
412    }
413
414    #[test]
415    fn allows_benign_build_and_toolchain_names() {
416        assert!(!is_sensitive_env_name("PATH"));
417        assert!(!is_sensitive_env_name("HOME"));
418        assert!(!is_sensitive_env_name("CARGO_HOME"));
419        assert!(!is_sensitive_env_name("LANG"));
420        assert!(!is_sensitive_env_name("LC_ALL"));
421        assert!(!is_sensitive_env_name("TERM"));
422        assert!(!is_sensitive_env_name("USER"));
423        assert!(!is_sensitive_env_name("RUSTUP_HOME"));
424        assert!(!is_sensitive_env_name("CARGO_TARGET_DIR"));
425        assert!(!is_sensitive_env_name("SHELL"));
426        // `_KEY`/`_PASSWORD` suffixes match only at the end, so words that
427        // merely embed those substrings are not swept.
428        assert!(!is_sensitive_env_name("BUILD_KEYBOARD"));
429        assert!(!is_sensitive_env_name("PASSWORD_HASH_ROUNDS"));
430        // Bare names without an underscore-prefixed suffix are not caught,
431        // matching the existing convention (bare `TOKEN` is not swept either).
432        assert!(!is_sensitive_env_name("PASSWORD"));
433        assert!(!is_sensitive_env_name("TOKEN"));
434    }
435
436    #[test]
437    fn matches_case_insensitively() {
438        assert!(is_sensitive_env_name("anthropic_api_key"));
439        assert!(is_sensitive_env_name("github_token"));
440        assert!(!is_sensitive_env_name("path"));
441    }
442}