Skip to main content

running_process_platform_internal/
platform_linux.rs

1//! Linux implementation root for the process capability.
2
3pub fn shell_command(command: &str) -> std::process::Command {
4    let mut shell = std::process::Command::new("/bin/sh");
5    shell.arg("-c").arg(command);
6    shell
7}
8
9pub fn compat_shell_command(command: &str) -> std::process::Command {
10    let mut shell = std::process::Command::new("sh");
11    shell.arg("-lc").arg(command);
12    shell
13}
14
15pub fn configure_native_command(
16    command: &mut std::process::Command,
17    _windows_creation_flags: u32,
18    create_process_group: bool,
19    nice: Option<i32>,
20    address_space_limit_bytes: Option<u64>,
21) {
22    use std::os::unix::process::CommandExt;
23
24    if create_process_group || nice.is_some() || address_space_limit_bytes.is_some() {
25        unsafe {
26            command.pre_exec(move || {
27                if create_process_group && libc::setpgid(0, 0) == -1 {
28                    return Err(std::io::Error::last_os_error());
29                }
30                if let Some(nice) = nice {
31                    let result = libc::setpriority(libc::PRIO_PROCESS, 0, nice);
32                    if result == -1 {
33                        return Err(std::io::Error::last_os_error());
34                    }
35                }
36                if let Some(limit) = address_space_limit_bytes {
37                    let rlim = libc::rlimit {
38                        rlim_cur: limit,
39                        rlim_max: limit,
40                    };
41                    if libc::setrlimit(libc::RLIMIT_AS, &rlim) == -1 {
42                        return Err(std::io::Error::last_os_error());
43                    }
44                }
45                Ok(())
46            });
47        }
48    }
49}
50
51pub fn canonical_environment_pairs(pairs: Vec<(String, String)>) -> Vec<(String, String)> {
52    pairs
53}
54
55pub fn monitor_console_windows(
56    _duration: std::time::Duration,
57) -> Vec<crate::platform::process::ConsoleWindowInfo> {
58    Vec::new()
59}
60
61use std::ffi::OsStr;
62use std::io;
63use std::io::Read;
64use std::os::fd::{AsRawFd, RawFd};
65use std::os::unix::net::UnixStream;
66use std::sync::Mutex;
67
68use tokio::process::{Child, Command};
69
70use crate::SpawnSpec;
71
72#[derive(Default)]
73pub struct CaptureCancellation {
74    wakers: Mutex<CaptureWakers>,
75}
76
77#[derive(Default)]
78struct CaptureWakers {
79    stdout: Option<UnixStream>,
80    stderr: Option<UnixStream>,
81}
82
83struct CancelableCaptureReader<R> {
84    reader: R,
85    wake_reader: UnixStream,
86}
87
88impl<R: Read + AsRawFd> Read for CancelableCaptureReader<R> {
89    fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
90        if buf.is_empty() { return Ok(0); }
91        loop {
92            let mut poll_fds = [
93                libc::pollfd { fd: self.reader.as_raw_fd(), events: libc::POLLIN | libc::POLLHUP | libc::POLLERR, revents: 0 },
94                libc::pollfd { fd: self.wake_reader.as_raw_fd(), events: libc::POLLIN | libc::POLLHUP | libc::POLLERR, revents: 0 },
95            ];
96            // SAFETY: both descriptors remain owned by this reader for the call.
97            let polled = unsafe { libc::poll(poll_fds.as_mut_ptr(), poll_fds.len() as _, -1) };
98            if polled < 0 {
99                let error = io::Error::last_os_error();
100                if error.kind() == io::ErrorKind::Interrupted { continue; }
101                return Err(error);
102            }
103            if poll_fds[1].revents != 0 {
104                return Err(io::Error::new(io::ErrorKind::Interrupted, "capture reader cancelled"));
105            }
106            if poll_fds[0].revents != 0 {
107                match self.reader.read(buf) {
108                    Err(error) if error.kind() == io::ErrorKind::WouldBlock => continue,
109                    result => return result,
110                }
111            }
112        }
113    }
114}
115
116pub fn prepare_capture_reader<R>(
117    reader: R,
118    cancellation: &CaptureCancellation,
119    stream: crate::platform::process::CaptureStream,
120) -> io::Result<Box<dyn Read + Send>>
121where R: Read + AsRawFd + Send + 'static {
122    set_nonblocking(reader.as_raw_fd())?;
123    let (wake_reader, wake_writer) = UnixStream::pair()?;
124    wake_writer.set_nonblocking(true)?;
125    let mut wakers = cancellation.wakers.lock().expect("capture wakers mutex poisoned");
126    match stream {
127        crate::platform::process::CaptureStream::Stdout => wakers.stdout = Some(wake_writer),
128        crate::platform::process::CaptureStream::Stderr => wakers.stderr = Some(wake_writer),
129    }
130    Ok(Box::new(CancelableCaptureReader { reader, wake_reader }))
131}
132
133pub fn capture_reader_done(cancellation: &CaptureCancellation, stream: crate::platform::process::CaptureStream) {
134    let mut wakers = cancellation.wakers.lock().expect("capture wakers mutex poisoned");
135    match stream {
136        crate::platform::process::CaptureStream::Stdout => wakers.stdout = None,
137        crate::platform::process::CaptureStream::Stderr => wakers.stderr = None,
138    }
139}
140
141pub fn cancel_capture_reader(cancellation: &CaptureCancellation) {
142    let wakers = cancellation.wakers.lock().expect("capture wakers mutex poisoned");
143    let byte = [1_u8; 1];
144    for writer in [&wakers.stdout, &wakers.stderr].into_iter().flatten() {
145        // SAFETY: the stored wake socket stays alive while the mutex is held.
146        let _ = unsafe { libc::write(writer.as_raw_fd(), byte.as_ptr().cast(), byte.len()) };
147    }
148}
149
150fn set_nonblocking(fd: RawFd) -> io::Result<()> {
151    // SAFETY: `fd` is borrowed from a live reader for both calls.
152    let flags = unsafe { libc::fcntl(fd, libc::F_GETFL) };
153    if flags < 0 { return Err(io::Error::last_os_error()); }
154    // SAFETY: `fd` is borrowed from a live reader for both calls.
155    if unsafe { libc::fcntl(fd, libc::F_SETFL, flags | libc::O_NONBLOCK) } < 0 {
156        return Err(io::Error::last_os_error());
157    }
158    Ok(())
159}
160
161#[path = "platform_linux_file_handles.rs"]
162mod file_handles;
163pub use file_handles::read_process_file_handles;
164#[path = "platform_linux_cmdline.rs"]
165mod cmdline;
166pub use cmdline::read_process_cmdline;
167
168#[path = "platform/process_tree.rs"]
169mod process_tree;
170
171pub fn kill_tree(pid: u32, timeout: std::time::Duration) -> io::Result<u32> {
172    process_tree::kill_tree(pid, timeout, |_pid, process| Ok(process.start_time()))
173}
174
175pub fn exit_code(status: std::process::ExitStatus) -> i32 {
176    use std::os::unix::process::ExitStatusExt;
177    status.code().unwrap_or_else(|| -status.signal().unwrap_or(1))
178}
179
180pub fn set_process_name(name: &str) {
181    let truncated: String = name.chars().take(15).collect();
182    let c_name = std::ffi::CString::new(truncated).unwrap_or_default();
183    unsafe { libc::prctl(libc::PR_SET_NAME, c_name.as_ptr() as libc::c_ulong, 0, 0, 0); }
184}
185
186pub fn configure_trampoline_command(_command: &mut std::process::Command) {}
187
188pub fn trampoline_exit_code(status: std::process::ExitStatus) -> i32 {
189    use std::os::unix::process::ExitStatusExt;
190    status.signal().map_or_else(|| status.code().unwrap_or(1), |signal| 128 + signal)
191}
192
193/// Enable Linux's process-wide orphan reparenting for launched-tree observation.
194/// Failure is intentionally best-effort: the observer can still track descendants
195/// whose immediate parents remain alive.
196pub fn enable_descendant_subreaper() {
197    let _ = unsafe { libc::prctl(libc::PR_SET_CHILD_SUBREAPER, 1, 0, 0, 0) };
198}
199
200/// Request a graceful shutdown for a child-owned POSIX process group.
201pub fn soft_terminate_process_group(pid: u32) -> io::Result<()> {
202    // SAFETY: `kill` receives only the numeric child-owned group id; no Rust
203    // references or borrowed state cross the OS boundary.
204    let result = unsafe { libc::kill(-(pid as i32), libc::SIGTERM) };
205    if result != 0 {
206        let error = io::Error::last_os_error();
207        if error.raw_os_error() != Some(libc::ESRCH) {
208            return Err(error);
209        }
210    }
211    Ok(())
212}
213
214pub fn process_snapshot() -> Vec<crate::platform::process::ProcessSnapshot> {
215    Vec::new()
216}
217
218pub fn process_snapshot_for_pid(_pid: u32) -> Option<crate::platform::process::ProcessSnapshot> {
219    None
220}
221
222/// Mark inherited descriptors close-on-exec without breaking std's exec-error pipe.
223///
224/// # Safety
225/// This must only be called from a post-fork `pre_exec` closure.
226pub unsafe fn unix_mark_extra_fds_close_on_exec() {
227    #[cfg(any(target_arch = "x86_64", target_arch = "aarch64", target_arch = "x86", target_arch = "arm", target_arch = "riscv64", target_arch = "powerpc64"))]
228    {
229        const SYS_CLOSE_RANGE: libc::c_long = 436;
230        const CLOSE_RANGE_CLOEXEC: libc::c_uint = 4;
231        if libc::syscall(SYS_CLOSE_RANGE, 3u32, libc::c_uint::MAX, CLOSE_RANGE_CLOEXEC) == 0 {
232            return;
233        }
234    }
235    mark_fds_from_directory_or_range();
236}
237
238pub fn configure_sync_daemon_command(command: &mut std::process::Command) -> io::Result<()> {
239    use std::os::unix::process::CommandExt;
240    unsafe {
241        command.pre_exec(|| {
242            let _ = libc::setsid();
243            unix_mark_extra_fds_close_on_exec();
244            Ok(())
245        });
246    }
247    Ok(())
248}
249
250pub fn configure_sync_contained_command(command: &mut std::process::Command) -> io::Result<()> {
251    use std::os::unix::process::CommandExt;
252    unsafe {
253        command.pre_exec(|| {
254            if libc::setpgid(0, 0) == -1 { return Err(io::Error::last_os_error()); }
255            if libc::prctl(libc::PR_SET_PDEATHSIG, libc::SIGKILL) == -1 {
256                return Err(io::Error::last_os_error());
257            }
258            if libc::getppid() == 1 { libc::_exit(1); }
259            unix_mark_extra_fds_close_on_exec();
260            Ok(())
261        });
262    }
263    Ok(())
264}
265
266pub fn parent_has_console() -> bool { false }
267
268pub fn sync_child_native_handle(_child: &std::process::Child) -> usize { 0 }
269
270unsafe fn mark_fds_from_directory_or_range() {
271    let dir = libc::opendir(c"/dev/fd".as_ptr());
272    if !dir.is_null() {
273        let dir_fd = libc::dirfd(dir);
274        loop {
275            let entry = libc::readdir(dir);
276            if entry.is_null() { break; }
277            let mut fd: libc::c_int = 0;
278            let mut cursor = (*entry).d_name.as_ptr();
279            let mut numeric = false;
280            while *cursor != 0 {
281                let byte = *cursor as u8;
282                if !byte.is_ascii_digit() { numeric = false; break; }
283                fd = fd * 10 + (byte - b'0') as libc::c_int;
284                cursor = cursor.add(1);
285                numeric = true;
286            }
287            if numeric && fd > 2 && fd != dir_fd { set_cloexec(fd); }
288        }
289        libc::closedir(dir);
290        return;
291    }
292    let maximum = libc::sysconf(libc::_SC_OPEN_MAX);
293    for fd in 3..if maximum < 0 { 4096 } else { maximum as libc::c_int } { set_cloexec(fd); }
294}
295
296unsafe fn set_cloexec(fd: libc::c_int) {
297    let flags = libc::fcntl(fd, libc::F_GETFD);
298    if flags != -1 { libc::fcntl(fd, libc::F_SETFD, flags | libc::FD_CLOEXEC); }
299}
300pub fn observer_backend(scope: crate::platform::process::ObserverScope, category: crate::platform::process::ObserverCategory) -> crate::platform::process::ObserverBackend {
301    use crate::platform::process::{ObserverBackend as B, ObserverCategory as C, ObserverScope as S, ObserverSupport as P};
302    match (scope, category) {
303        (S::SystemWide, C::File) => B { support:P::Unavailable, backend:"seccomp-user-notify", reason:"Phase 3: Linux seccomp user-notify file backend not yet implemented" },
304        (S::SystemWide, C::Network) => B { support:P::Unavailable, backend:"ebpf", reason:"Phase 3: Linux eBPF network backend not yet implemented" },
305        (S::SystemWide, C::Process) => B { support:P::Unavailable, backend:"seccomp-user-notify", reason:"Phase 3: Linux seccomp user-notify process backend not yet implemented" },
306        (S::LaunchedProcessTree, C::File) => B { support:P::Partial, backend:"proc-fd-snapshot", reason:"Linux /proc/<pid>/fd/* snapshot via read_process_file_handles (#539 slice 6 follow-up; no streaming file events)" },
307        (S::LaunchedProcessTree, C::Network) => B { support:P::Unavailable, backend:"none", reason:"#539: no-admin per-child network backend deferred to a follow-up issue" },
308        (S::LaunchedProcessTree, C::Process) => B { support:P::Supported, backend:"subreaper-proc-poll", reason:"Linux PR_SET_CHILD_SUBREAPER + /proc descendant polling (#539 slice 5)" },
309    }
310}
311
312pub fn unix_set_priority(pid: u32, nice: i32) -> io::Result<()> {
313    if unsafe { libc::setpriority(libc::PRIO_PROCESS, pid, nice) } == -1 { Err(io::Error::last_os_error()) } else { Ok(()) }
314}
315pub fn unix_signal_process(pid: u32, signal: crate::platform::process::UnixSignalKind) -> io::Result<()> {
316    if unsafe { libc::kill(pid as i32, unix_signal_raw(signal)) } == -1 { Err(io::Error::last_os_error()) } else { Ok(()) }
317}
318pub fn unix_signal_process_group(pid: i32, signal: crate::platform::process::UnixSignalKind) -> io::Result<()> {
319    if unsafe { libc::killpg(pid, unix_signal_raw(signal)) } == -1 { Err(io::Error::last_os_error()) } else { Ok(()) }
320}
321pub fn unix_signal_raw(signal: crate::platform::process::UnixSignalKind) -> i32 {
322    match signal { crate::platform::process::UnixSignalKind::Interrupt => libc::SIGINT, crate::platform::process::UnixSignalKind::Terminate => libc::SIGTERM, crate::platform::process::UnixSignalKind::Kill => libc::SIGKILL }
323}
324
325pub fn configure_compat_tokio_command(
326    command: &mut Command,
327    _show_console: bool,
328    kill_when_owner_dies: bool,
329) -> io::Result<()> {
330    configure_command(command, false, kill_when_owner_dies)
331}
332
333pub fn after_compat_tokio_spawn(_child: &Child, _kill_when_owner_dies: bool) {}
334
335pub(crate) fn configure_command(
336    command: &mut Command,
337    create_process_group: bool,
338    kill_when_owner_dies: bool,
339) -> io::Result<()> {
340    if create_process_group {
341        command.process_group(0);
342    }
343    if kill_when_owner_dies {
344        let owner_pid = unsafe { libc::getpid() };
345        // SAFETY: the closure invokes only async-signal-safe libc calls.
346        unsafe {
347            command.pre_exec(move || {
348                if libc::prctl(
349                    libc::PR_SET_PDEATHSIG,
350                    libc::SIGTERM as libc::c_ulong,
351                    0,
352                    0,
353                    0,
354                ) == -1
355                {
356                    return Err(io::Error::last_os_error());
357                }
358                if libc::getppid() != owner_pid {
359                    libc::kill(libc::getpid(), libc::SIGTERM);
360                }
361                Ok(())
362            });
363        }
364    }
365    Ok(())
366}
367
368pub(crate) fn after_spawn(_child: &Child, _kill_when_owner_dies: bool) {}
369
370pub(crate) fn signal_process(pid: u32) -> io::Result<()> {
371    unix_kill(pid as i32, libc::SIGKILL)
372}
373
374pub(crate) fn signal_process_group(pid: u32) -> io::Result<()> {
375    unix_kill(-(pid as i32), libc::SIGTERM)
376}
377
378fn unix_kill(target: i32, signal: i32) -> io::Result<()> {
379    let result = unsafe { libc::kill(target, signal) };
380    if result == 0 {
381        return Ok(());
382    }
383    let error = io::Error::last_os_error();
384    if error.raw_os_error() == Some(libc::ESRCH) {
385        Ok(())
386    } else {
387        Err(error)
388    }
389}
390
391pub(crate) fn shell_spec(command: &OsStr) -> SpawnSpec {
392    SpawnSpec::new("/bin/sh").arg("-c").arg(command)
393}
394#[path = "sync_spawn_group.rs"]
395mod sync_spawn;
396pub use sync_spawn::{spawn_sync, spawn_sync_daemon};