feat: add approval bound full maintenance script console and executors
Rust development slice / linux (push) Failing after 6s

This commit is contained in:
2026-11-14 16:35:11 +01:00
parent 287a784c8f
commit ff78b6a498
51 changed files with 8880 additions and 278 deletions
+232 -108
View File
@@ -15,11 +15,19 @@ pub struct Output {
pub stderr: Vec<u8>,
pub truncated: bool,
pub timed_out: bool,
pub interrupted: bool,
}
// Nonblocking pipes avoid both unbounded output and reader threads held open by
// grandchildren. The child has a separate process group, cleaned even on success.
pub(crate) fn run(program: &str, args: &[String], timeout: Duration, cap: usize) -> Result<Output> {
run_controlled(program, args, timeout, cap, || Ok(()))
}
pub(crate) fn run_controlled(
program: &str,
args: &[String],
timeout: Duration,
cap: usize,
mut observe: impl FnMut() -> Result<()>,
) -> Result<Output> {
ensure!(!timeout.is_zero() && cap > 0, "invalid process limits");
let mut command = Command::new(program);
command
@@ -34,144 +42,260 @@ pub(crate) fn run(program: &str, args: &[String], timeout: Duration, cap: usize)
.stdout(Stdio::piped())
.stderr(Stdio::piped())
.process_group(0);
observe()?;
let mut child = command
.spawn()
.with_context(|| format!("spawn {program}"))?;
struct Group(u32);
// Keep the group leader unreaped until AFTER the last signal and cleanup.
// Its reserved PID prevents the numerical PGID from becoming a foreign group.
struct Group {
pid: u32,
reserved: bool,
}
impl Group {
fn kill(&self) {
if self.reserved {
unsafe {
libc::kill(-(self.pid as i32), libc::SIGKILL);
}
}
}
}
impl Drop for Group {
fn drop(&mut self) {
// SAFETY: negative PID addresses only the group created for this child.
unsafe {
libc::kill(-(self.0 as i32), libc::SIGKILL);
}
self.kill();
}
}
let group = Group(child.id());
let mut group = Group {
pid: child.id(),
reserved: true,
};
let mut stdout = child.stdout.take().context("missing stdout")?;
let mut stderr = child.stderr.take().context("missing stderr")?;
for fd in [stdout.as_raw_fd(), stderr.as_raw_fd()] {
// SAFETY: these are live owned pipe descriptors; flags are not pointers.
let flags = unsafe { libc::fcntl(fd, libc::F_GETFL) };
if flags < 0 || unsafe { libc::fcntl(fd, libc::F_SETFL, flags | libc::O_NONBLOCK) } < 0 {
drop(group);
let _ = child.wait();
return Err(std::io::Error::last_os_error().into());
}
}
fn drain(
pipe: &mut impl Read,
target: &mut Vec<u8>,
remaining: &mut usize,
truncated: &mut bool,
) -> Result<bool> {
let mut bytes = [0; 4096];
// Bound each pass too: a continuously writing child cannot starve timeout.
for _ in 0..16 {
match pipe.read(&mut bytes) {
Ok(0) => return Ok(true),
Ok(n) => {
let keep = n.min(*remaining);
target.extend_from_slice(&bytes[..keep]);
*remaining -= keep;
*truncated |= keep < n;
}
Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => break,
Err(e) if e.kind() == std::io::ErrorKind::Interrupted => continue,
Err(e) => return Err(e.into()),
}
}
Ok(false)
}
let started = Instant::now();
let mut result = Output {
let began = Instant::now();
let mut output = Output {
code: None,
stdout: vec![],
stderr: vec![],
truncated: false,
timed_out: false,
interrupted: false,
};
let mut remaining = cap;
let status = loop {
let drained = drain(
&mut stdout,
&mut result.stdout,
&mut remaining,
&mut result.truncated,
)
.and_then(|_| {
let mut failure = None;
for fd in [stdout.as_raw_fd(), stderr.as_raw_fd()] {
let flags = unsafe { libc::fcntl(fd, libc::F_GETFL) };
if flags < 0 || unsafe { libc::fcntl(fd, libc::F_SETFL, flags | libc::O_NONBLOCK) } < 0 {
failure = Some(anyhow::Error::from(std::io::Error::last_os_error()));
break;
}
}
if failure.is_none() {
loop {
if let Err(e) = drain(
&mut stdout,
&mut output.stdout,
&mut remaining,
&mut output.truncated,
)
.and_then(|_| {
drain(
&mut stderr,
&mut output.stderr,
&mut remaining,
&mut output.truncated,
)
}) {
failure = Some(e);
break;
}
// WNOWAIT deliberately observes without releasing this child's PID.
let mut info = std::mem::MaybeUninit::<libc::siginfo_t>::zeroed();
let state = unsafe {
libc::waitid(
libc::P_PID,
child.id(),
info.as_mut_ptr(),
libc::WEXITED | libc::WNOHANG | libc::WNOWAIT,
)
};
if state < 0 {
let e = std::io::Error::last_os_error();
if e.kind() == std::io::ErrorKind::Interrupted {
continue;
}
if e.raw_os_error() == Some(libc::ECHILD) {
group.reserved = false;
}
failure = Some(e.into());
break;
}
if unsafe { info.assume_init().si_pid() } != 0 {
break;
}
let fenced = observe().is_err();
if began.elapsed() >= timeout || fenced {
output.timed_out = began.elapsed() >= timeout;
output.interrupted = fenced;
break;
}
thread::sleep(Duration::from_millis(5));
}
}
if !group.reserved {
// An unexpected external reap destroyed the group identity proof. Never
// signal/adopt a numerical replacement; caller records Unknown.
return Err(failure.unwrap_or_else(|| anyhow::anyhow!("OWNED_GROUP_IDENTITY_LOST")));
}
group.kill();
let cleanup = Instant::now();
loop {
// A setup error may have left a blocking pipe. It must not become an
// unbounded cleanup read; retain only group ownership in that path.
let out_closed = if failure.is_some() {
Ok(true)
} else {
drain(
&mut stdout,
&mut output.stdout,
&mut remaining,
&mut output.truncated,
)
};
let err_closed = if failure.is_some() {
Ok(true)
} else {
drain(
&mut stderr,
&mut result.stderr,
&mut output.stderr,
&mut remaining,
&mut result.truncated,
&mut output.truncated,
)
});
if let Err(e) = drained {
drop(group);
let _ = child.wait();
return Err(e);
};
if out_closed.is_err() || err_closed.is_err() {
output.truncated = true;
}
match child.try_wait() {
Ok(Some(status)) => break status,
Ok(None) => {}
Err(e) => {
drop(group);
let _ = child.wait();
return Err(e.into());
// Only zombies may remain while the *owned* leader stays unreaped.
// They cannot execute or retain pipe writers. Unclear /proc state keeps
// this owner and the caller's executor lock instead of a replacement.
if group_has_no_live_members(group.pid) {
if out_closed.unwrap_or(true) && err_closed.unwrap_or(true) {
break;
}
// A privileged script may deliberately give a pipe to an unrelated
// process. That process is outside our ownership; never kill it.
if cleanup.elapsed() >= Duration::from_millis(250) {
output.truncated = true;
break;
}
}
if started.elapsed() >= timeout {
result.timed_out = true;
// SAFETY: this is the dedicated group created for this child.
unsafe {
libc::kill(-(child.id() as i32), libc::SIGKILL);
}
break child.wait()?;
}
thread::sleep(Duration::from_millis(5));
};
let pgid = child.id();
drop(group);
// SIGKILL delivery is asynchronous. A single WouldBlock is not EOF and
// does not prove that descendants have stopped. Bound both cleanup waiting
// and output draining, and withhold a successful exit code if unconfirmed.
let cleanup_started = Instant::now();
loop {
let stdout_closed = drain(
&mut stdout,
&mut result.stdout,
&mut remaining,
&mut result.truncated,
)?;
let stderr_closed = drain(
&mut stderr,
&mut result.stderr,
&mut remaining,
&mut result.truncated,
)?;
// SAFETY: signal zero only probes the dedicated process group.
let absent = unsafe { libc::kill(-(pgid as i32), 0) } == -1
&& std::io::Error::last_os_error().raw_os_error() == Some(libc::ESRCH);
if stdout_closed && stderr_closed && absent {
result.code = status.code();
break;
}
if cleanup_started.elapsed() >= Duration::from_millis(250) {
result.truncated |= !stdout_closed || !stderr_closed;
// None explicitly means no confirmed successful completion. This
// can also occur for lingering zombies: never guess their state.
result.code = None;
break;
if cleanup.elapsed() >= Duration::from_millis(250) {
output.timed_out = true;
}
thread::sleep(Duration::from_millis(5));
}
Ok(result)
// Disarm BEFORE releasing the reserved leader PID: no later Drop can signal
// a reused PGID. All assigned living members have already ended.
group.reserved = false;
let status = child.wait()?;
if let Some(e) = failure {
return Err(e);
}
output.interrupted |= observe().is_err();
output.code = if output.timed_out || began.elapsed() >= timeout {
None
} else {
status.code()
};
Ok(output)
}
fn drain(
pipe: &mut impl Read,
target: &mut Vec<u8>,
remaining: &mut usize,
truncated: &mut bool,
) -> Result<bool> {
let mut bytes = [0; 4096];
for _ in 0..16 {
match pipe.read(&mut bytes) {
Ok(0) => return Ok(true),
Ok(n) => {
let keep = n.min(*remaining);
target.extend_from_slice(&bytes[..keep]);
*remaining -= keep;
*truncated |= keep < n;
}
Err(e) if e.kind() == std::io::ErrorKind::WouldBlock => break,
Err(e) if e.kind() == std::io::ErrorKind::Interrupted => continue,
Err(e) => return Err(e.into()),
}
}
Ok(false)
}
fn group_has_no_live_members(pgid: u32) -> bool {
let Ok(entries) = std::fs::read_dir("/proc") else {
return false;
};
for entry in entries {
let Ok(entry) = entry else {
return false;
};
if entry.file_name().to_string_lossy().parse::<u32>().is_err() {
continue;
}
let bytes = match std::fs::read_to_string(entry.path().join("stat")) {
Ok(v) => v,
Err(e) if e.kind() == std::io::ErrorKind::NotFound => continue,
Err(_) => return false,
};
let Some((_, rest)) = bytes.rsplit_once(") ") else {
return false;
};
let fields: Vec<_> = rest.split_whitespace().take(3).collect();
if fields.len() != 3 {
return false;
}
let Ok(group) = fields[2].parse::<u32>() else {
return false;
};
if group == pgid && fields[0] != "Z" && fields[0] != "X" {
return false;
}
}
true
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn cancellation_observation_terminates_only_owned_group_and_keeps_raw_bounded_output() {
let began = Instant::now();
let output = run_controlled(
"/bin/sh",
&[
"-c".into(),
"printf before; printf err >&2; sleep 10".into(),
],
Duration::from_secs(5),
64,
|| {
ensure!(
began.elapsed() < Duration::from_millis(80),
"TEST_CANCELLED"
);
Ok(())
},
)
.unwrap();
assert!(output.interrupted);
assert!(!output.timed_out);
assert_eq!(output.stdout, b"before");
assert_eq!(output.stderr, b"err");
assert_ne!(output.code, Some(0));
assert!(began.elapsed() < Duration::from_secs(2));
}
#[test]
fn exited_leader_does_not_leave_background_effects_or_open_pipes() {
let directory =