Unrolled build for #150412

Rollup merge of #150412 - the8472:pidfd-spawn, r=tgross35

use PIDFD_GET_INFO ioctl when available

This way using pidfd_spawnp won't have to rely on procfs, avoiding an unpleasant edge-case where the child is spawned but we can't get the pid. And pidfd.{try_}wait will be able to return the exit status even after a process has been reaped. At least on newer kernels.

Tracking issue: #82971

Author: rust-timer

library/std/src/os/linux/process.rs

@@ -67,8 +67,10 @@ impl PidFd {
     /// Waits for the child to exit completely, returning the status that it exited with.
     ///
     /// Unlike [`Child::wait`] it does not ensure that the stdin handle is closed.
-    /// Additionally it will not return an `ExitStatus` if the child
-    /// has already been reaped. Instead an error will be returned.
+    ///
+    /// Additionally on kernels prior to 6.15 only the first attempt to
+    /// reap a child will return an ExitStatus, further attempts
+    /// will return an Error.
     ///
     /// [`Child::wait`]: process::Child::wait
     pub fn wait(&self) -> Result<ExitStatus> {
@@ -77,8 +79,8 @@ impl PidFd {
 
     /// Attempts to collect the exit status of the child if it has already exited.
     ///
-    /// Unlike [`Child::try_wait`] this method will return an Error
-    /// if the child has already been reaped.
+    /// On kernels prior to 6.15, and unlike [`Child::try_wait`], only the first attempt
+    /// to reap a child will return an ExitStatus, further attempts will return an Error.
     ///
     /// [`Child::try_wait`]: process::Child::try_wait
     pub fn try_wait(&self) -> Result<Option<ExitStatus>> {

library/std/src/sys/pal/unix/linux/pidfd.rs

@@ -1,5 +1,5 @@
 use crate::io;
-use crate::os::fd::{AsRawFd, FromRawFd, RawFd};
+use crate::os::fd::{AsRawFd, FromRawFd, IntoRawFd, RawFd};
 use crate::sys::fd::FileDesc;
 use crate::sys::process::ExitStatus;
 use crate::sys::{AsInner, FromInner, IntoInner, cvt};
@@ -15,6 +15,73 @@ impl PidFd {
         self.send_signal(libc::SIGKILL)
     }
 
+    #[cfg(any(test, target_env = "gnu", target_env = "musl"))]
+    pub(crate) fn current_process() -> io::Result<PidFd> {
+        let pid = crate::process::id();
+        let pidfd = cvt(unsafe { libc::syscall(libc::SYS_pidfd_open, pid, 0) })?;
+        Ok(unsafe { PidFd::from_raw_fd(pidfd as RawFd) })
+    }
+
+    #[cfg(any(test, target_env = "gnu", target_env = "musl"))]
+    pub(crate) fn pid(&self) -> io::Result<u32> {
+        use crate::sys::weak::weak;
+
+        // since kernel 6.13
+        // https://lore.kernel.org/all/[email protected]/
+        let mut pidfd_info: libc::pidfd_info = unsafe { crate::mem::zeroed() };
+        pidfd_info.mask = libc::PIDFD_INFO_PID as u64;
+        match cvt(unsafe { libc::ioctl(self.0.as_raw_fd(), libc::PIDFD_GET_INFO, &mut pidfd_info) })
+        {
+            Ok(_) => {}
+            Err(e) if e.raw_os_error() == Some(libc::EINVAL) => {
+                // kernel doesn't support that ioctl, try the glibc helper that looks at procfs
+                weak!(
+                    fn pidfd_getpid(pidfd: RawFd) -> libc::pid_t;
+                );
+                if let Some(pidfd_getpid) = pidfd_getpid.get() {
+                    let pid: libc::c_int = cvt(unsafe { pidfd_getpid(self.0.as_raw_fd()) })?;
+                    return Ok(pid as u32);
+                }
+                return Err(e);
+            }
+            Err(e) => return Err(e),
+        }
+
+        Ok(pidfd_info.pid)
+    }
+
+    fn exit_for_reaped_child(&self) -> io::Result<ExitStatus> {
+        // since kernel 6.15
+        // https://lore.kernel.org/linux-fsdevel/20250305-work-pidfs-kill_on_last_close-v3-0-c8c3d8361705@kernel.org/T/
+        let mut pidfd_info: libc::pidfd_info = unsafe { crate::mem::zeroed() };
+        pidfd_info.mask = libc::PIDFD_INFO_EXIT as u64;
+        cvt(unsafe { libc::ioctl(self.0.as_raw_fd(), libc::PIDFD_GET_INFO, &mut pidfd_info) })?;
+        Ok(ExitStatus::new(pidfd_info.exit_code))
+    }
+
+    fn waitid(&self, options: libc::c_int) -> io::Result<Option<ExitStatus>> {
+        let mut siginfo: libc::siginfo_t = unsafe { crate::mem::zeroed() };
+        let r = cvt(unsafe {
+            libc::waitid(libc::P_PIDFD, self.0.as_raw_fd() as u32, &mut siginfo, options)
+        });
+        match r {
+            Err(waitid_err) if waitid_err.raw_os_error() == Some(libc::ECHILD) => {
+                // already reaped
+                match self.exit_for_reaped_child() {
+                    Ok(exit_status) => return Ok(Some(exit_status)),
+                    Err(_) => return Err(waitid_err),
+                }
+            }
+            Err(e) => return Err(e),
+            Ok(_) => {}
+        }
+        if unsafe { siginfo.si_pid() } == 0 {
+            Ok(None)
+        } else {
+            Ok(Some(ExitStatus::from_waitid_siginfo(siginfo)))
+        }
+    }
+
     pub(crate) fn send_signal(&self, signal: i32) -> io::Result<()> {
         cvt(unsafe {
             libc::syscall(
@@ -29,29 +96,15 @@ impl PidFd {
     }
 
     pub fn wait(&self) -> io::Result<ExitStatus> {
-        let mut siginfo: libc::siginfo_t = unsafe { crate::mem::zeroed() };
-        cvt(unsafe {
-            libc::waitid(libc::P_PIDFD, self.0.as_raw_fd() as u32, &mut siginfo, libc::WEXITED)
-        })?;
-        Ok(ExitStatus::from_waitid_siginfo(siginfo))
+        let r = self.waitid(libc::WEXITED)?;
+        match r {
+            Some(exit_status) => Ok(exit_status),
+            None => unreachable!("waitid with WEXITED should not return None"),
+        }
     }
 
     pub fn try_wait(&self) -> io::Result<Option<ExitStatus>> {
-        let mut siginfo: libc::siginfo_t = unsafe { crate::mem::zeroed() };
-
-        cvt(unsafe {
-            libc::waitid(
-                libc::P_PIDFD,
-                self.0.as_raw_fd() as u32,
-                &mut siginfo,
-                libc::WEXITED | libc::WNOHANG,
-            )
-        })?;
-        if unsafe { siginfo.si_pid() } == 0 {
-            Ok(None)
-        } else {
-            Ok(Some(ExitStatus::from_waitid_siginfo(siginfo)))
-        }
+        self.waitid(libc::WEXITED | libc::WNOHANG)
     }
 }
 
@@ -78,3 +131,9 @@ impl FromRawFd for PidFd {
         Self(FileDesc::from_raw_fd(fd))
     }
 }
+
+impl IntoRawFd for PidFd {
+    fn into_raw_fd(self) -> RawFd {
+        self.0.into_raw_fd()
+    }
+}

library/std/src/sys/pal/unix/linux/pidfd/tests.rs

@@ -1,8 +1,11 @@
+use super::PidFd as InternalPidFd;
 use crate::assert_matches::assert_matches;
-use crate::os::fd::{AsRawFd, RawFd};
+use crate::io::ErrorKind;
+use crate::os::fd::AsRawFd;
 use crate::os::linux::process::{ChildExt, CommandExt as _};
 use crate::os::unix::process::{CommandExt as _, ExitStatusExt};
 use crate::process::Command;
+use crate::sys::AsInner;
 
 #[test]
 fn test_command_pidfd() {
@@ -48,11 +51,22 @@ fn test_command_pidfd() {
     let mut cmd = Command::new("false");
     let mut child = unsafe { cmd.pre_exec(|| Ok(())) }.create_pidfd(true).spawn().unwrap();
 
-    assert!(child.id() > 0 && child.id() < -1i32 as u32);
+    let id = child.id();
+
+    assert!(id > 0 && id < -1i32 as u32, "spawning with pidfd still returns a sane pid");
 
     if pidfd_open_available {
         assert!(child.pidfd().is_ok())
     }
+
+    if let Ok(pidfd) = child.pidfd() {
+        match pidfd.as_inner().pid() {
+            Ok(pid) => assert_eq!(pid, id),
+            Err(e) if e.kind() == ErrorKind::InvalidInput => { /* older kernel */ }
+            Err(e) => panic!("unexpected error getting pid from pidfd: {}", e),
+        }
+    }
+
     child.wait().expect("error waiting on child");
 }
 
@@ -77,23 +91,21 @@ fn test_pidfd() {
     assert_eq!(status.signal(), Some(libc::SIGKILL));
 
     // Trying to wait again for a reaped child is safe since there's no pid-recycling race.
-    // But doing so will return an error.
+    // But doing so may return an error.
     let res = fd.wait();
-    assert_matches!(res, Err(e) if e.raw_os_error() == Some(libc::ECHILD));
+    match res {
+        // older kernels
+        Err(e) if e.raw_os_error() == Some(libc::ECHILD) => {}
+        // 6.15+
+        Ok(exit) if exit.signal() == Some(libc::SIGKILL) => {}
+        other => panic!("expected ECHILD error, got {:?}", other),
+    }
 
     // Ditto for additional attempts to kill an already-dead child.
     let res = fd.kill();
     assert_matches!(res, Err(e) if e.raw_os_error() == Some(libc::ESRCH));
 }
 
 fn probe_pidfd_support() -> bool {
-    // pidfds require the pidfd_open syscall
-    let our_pid = crate::process::id();
-    let pidfd = unsafe { libc::syscall(libc::SYS_pidfd_open, our_pid, 0) };
-    if pidfd >= 0 {
-        unsafe { libc::close(pidfd as RawFd) };
-        true
-    } else {
-        false
-    }
+    InternalPidFd::current_process().is_ok()
 }

library/std/src/sys/process/unix/unix.rs

@@ -482,10 +482,6 @@ impl Command {
                     ) -> libc::c_int;
                 );
 
-                weak!(
-                    fn pidfd_getpid(pidfd: libc::c_int) -> libc::c_int;
-                );
-
                 static PIDFD_SUPPORTED: Atomic<u8> = AtomicU8::new(0);
                 const UNKNOWN: u8 = 0;
                 const SPAWN: u8 = 1;
@@ -502,24 +498,26 @@ impl Command {
                     }
                     if support == UNKNOWN {
                         support = NO;
-                        let our_pid = crate::process::id();
-                        let pidfd = cvt(unsafe { libc::syscall(libc::SYS_pidfd_open, our_pid, 0) } as c_int);
-                        match pidfd {
+
+                        match PidFd::current_process() {
                             Ok(pidfd) => {
+                                // if pidfd_open works then we at least know the fork path is available.
                                 support = FORK_EXEC;
-                                if let Some(Ok(pid)) = pidfd_getpid.get().map(|f| cvt(unsafe { f(pidfd) } as i32)) {
-                                    if pidfd_spawnp.get().is_some() && pid as u32 == our_pid {
-                                        support = SPAWN
-                                    }
+                                // but for the fast path we need both spawnp and the
+                                // pidfd -> pid conversion to work.
+                                if pidfd_spawnp.get().is_some() && let Ok(pid) = pidfd.pid() {
+                                    assert_eq!(pid, crate::process::id(), "sanity check");
+                                    support = SPAWN;
                                 }
-                                unsafe { libc::close(pidfd) };
                             }
                             Err(e) if e.raw_os_error() == Some(libc::EMFILE) => {
-                                // We're temporarily(?) out of file descriptors.  In this case obtaining a pidfd would also fail
+                                // We're temporarily(?) out of file descriptors. In this case pidfd_spawnp would also fail
                                 // Don't update the support flag so we can probe again later.
                                 return Err(e)
                             }
-                            _ => {}
+                            _ => {
+                                // pidfd_open not available? likely an old kernel without pidfd support.
+                            }
                         }
                         PIDFD_SUPPORTED.store(support, Ordering::Relaxed);
                         if support == FORK_EXEC {
@@ -791,21 +789,25 @@ impl Command {
                 }
                 spawn_res?;
 
-                let pid = match cvt(pidfd_getpid.get().unwrap()(pidfd)) {
+                use crate::os::fd::{FromRawFd, IntoRawFd};
+
+                let pidfd = PidFd::from_raw_fd(pidfd);
+                let pid = match pidfd.pid() {
                     Ok(pid) => pid,
                     Err(e) => {
                         // The child has been spawned and we are holding its pidfd.
-                        // But we cannot obtain its pid even though pidfd_getpid support was verified earlier.
-                        // This might happen if libc can't open procfs because the file descriptor limit has been reached.
-                        libc::close(pidfd);
+                        // But we cannot obtain its pid even though pidfd_spawnp and getpid support
+                        // was verified earlier.
+                        // This is quite unlikely, but might happen if the ioctl is not supported,
+                        // glibc tries to use procfs and we're out of file descriptors.
                         return Err(Error::new(
                             e.kind(),
                             "pidfd_spawnp succeeded but the child's PID could not be obtained",
                         ));
                     }
                 };
 
-                return Ok(Some(Process::new(pid, pidfd)));
+                return Ok(Some(Process::new(pid as i32, pidfd.into_raw_fd())));
             }
 
             // Safety: -1 indicates we don't have a pidfd.