fix: detect closed stdin before Rust sanitizes it to /dev/null

Author: ChrisDryden

Cargo.lock

@@ -24,7 +24,8 @@ memchr = { workspace = true }
 memmap2 = { workspace = true }
 regex = { workspace = true }
 clap = { workspace = true }
-uucore = { workspace = true }
+libc = { workspace = true }
+uucore = { workspace = true, features = ["signals"] }
 thiserror = { workspace = true }
 fluent = { workspace = true }
 

src/uu/tac/Cargo.toml

@@ -4,6 +4,9 @@
 // file that was distributed with this source code.
 
 // spell-checker:ignore (ToDO) sbytes slen dlen memmem memmap Mmap mmap SIGBUS
+#[cfg(unix)]
+uucore::init_stdio_state_capture!();
+
 mod error;
 
 use clap::{Arg, ArgAction, Command};
@@ -15,8 +18,7 @@ use std::{
     fs::{File, read},
     path::Path,
 };
-use uucore::error::UError;
-use uucore::error::UResult;
+use uucore::error::{UError, UResult, set_exit_code};
 use uucore::{format_usage, show};
 
 use crate::error::TacError;
@@ -237,6 +239,17 @@ fn tac(filenames: &[OsString], before: bool, regex: bool, separator: &str) -> UR
         let buf;
 
         let data: &[u8] = if filename == "-" {
+            #[cfg(unix)]
+            if uucore::signals::stdin_was_closed() {
+                let e: Box<dyn UError> = TacError::ReadError(
+                    OsString::from("-"),
+                    std::io::Error::from_raw_os_error(libc::EBADF),
+                )
+                .into();
+                show!(e);
+                set_exit_code(1);
+                continue;
+            }
             if let Some(mmap1) = try_mmap_stdin() {
                 mmap = mmap1;
                 &mmap

src/uu/tac/src/tac.rs

@@ -23,7 +23,7 @@ clap = { workspace = true }
 libc = { workspace = true }
 memchr = { workspace = true }
 notify = { workspace = true }
-uucore = { workspace = true, features = ["fs", "parser-size"] }
+uucore = { workspace = true, features = ["fs", "parser-size", "signals"] }
 same-file = { workspace = true }
 fluent = { workspace = true }
 

src/uu/tail/Cargo.toml

@@ -230,13 +230,5 @@ pub fn path_is_tailable(path: &Path) -> bool {
 
 #[inline]
 pub fn stdin_is_bad_fd() -> bool {
-    // FIXME : Rust's stdlib is reopening fds as /dev/null
-    // see also: https://github.com/uutils/coreutils/issues/2873
-    // (gnu/tests/tail-2/follow-stdin.sh fails because of this)
-    //#[cfg(unix)]
-    {
-        //platform::stdin_is_bad_fd()
-    }
-    //#[cfg(not(unix))]
-    false
+    uucore::signals::stdin_was_closed()
 }

src/uu/tail/src/paths.rs

@@ -33,11 +33,14 @@ use std::fs::File;
 use std::io::{self, BufReader, BufWriter, ErrorKind, Read, Seek, SeekFrom, Write, stdin, stdout};
 use std::path::{Path, PathBuf};
 use uucore::display::Quotable;
-use uucore::error::{FromIo, UResult, USimpleError, get_exit_code, set_exit_code};
+use uucore::error::{FromIo, UResult, USimpleError, set_exit_code};
 use uucore::translate;
 
 use uucore::{show, show_error};
 
+#[cfg(unix)]
+uucore::init_stdio_state_capture!();
+
 #[uucore::main]
 pub fn uumain(args: impl uucore::Args) -> UResult<()> {
     // When we receive a SIGPIPE signal, we want to terminate the process so
@@ -105,10 +108,6 @@ fn uu_tail(settings: &Settings) -> UResult<()> {
         }
     }
 
-    if get_exit_code() > 0 && paths::stdin_is_bad_fd() {
-        show_error!("{}: {}", text::DASH, translate!("tail-bad-fd"));
-    }
-
     Ok(())
 }
 
@@ -227,6 +226,17 @@ fn tail_stdin(
         }
     }
 
+    // Check if stdin was closed before Rust reopened it as /dev/null
+    if paths::stdin_is_bad_fd() {
+        set_exit_code(1);
+        show_error!(
+            "{}",
+            translate!("tail-error-cannot-fstat", "file" => translate!("tail-stdin-header").quote(), "error" => translate!("tail-bad-fd"))
+        );
+        show_error!("{}", translate!("tail-no-files-remaining"));
+        return Ok(());
+    }
+
     match input.resolve() {
         // fifo
         Some(path) => {

src/uu/tail/src/tail.rs

@@ -4,12 +4,15 @@
 // file that was distributed with this source code.
 
 // spell-checker:ignore (ToDO) tstr sigstr cmdname setpgid sigchld getpid
+#[cfg(unix)]
+uucore::init_stdio_state_capture!();
+
 mod status;
 
 use crate::status::ExitStatus;
 use clap::{Arg, ArgAction, Command};
 use std::io::ErrorKind;
-use std::os::unix::process::ExitStatusExt;
+use std::os::unix::process::{CommandExt, ExitStatusExt};
 use std::process::{self, Child, Stdio};
 use std::sync::atomic::{self, AtomicBool};
 use std::time::Duration;
@@ -320,23 +323,34 @@ fn timeout(
     #[cfg(unix)]
     enable_pipe_errors()?;
 
-    let process = &mut process::Command::new(&cmd[0])
+    let mut command = process::Command::new(&cmd[0]);
+    command
         .args(&cmd[1..])
         .stdin(Stdio::inherit())
         .stdout(Stdio::inherit())
-        .stderr(Stdio::inherit())
-        .spawn()
-        .map_err(|err| {
-            let status_code = match err.kind() {
-                ErrorKind::NotFound => ExitStatus::CommandNotFound.into(),
-                ErrorKind::PermissionDenied => ExitStatus::CannotInvoke.into(),
-                _ => ExitStatus::CannotInvoke.into(),
-            };
-            USimpleError::new(
-                status_code,
-                translate!("timeout-error-failed-to-execute-process", "error" => err),
-            )
-        })?;
+        .stderr(Stdio::inherit());
+
+    // If stdin was closed before Rust reopened it as /dev/null, close it in child
+    if uucore::signals::stdin_was_closed() {
+        unsafe {
+            command.pre_exec(|| {
+                libc::close(libc::STDIN_FILENO);
+                Ok(())
+            });
+        }
+    }
+
+    let process = &mut command.spawn().map_err(|err| {
+        let status_code = match err.kind() {
+            ErrorKind::NotFound => ExitStatus::CommandNotFound.into(),
+            ErrorKind::PermissionDenied => ExitStatus::CannotInvoke.into(),
+            _ => ExitStatus::CannotInvoke.into(),
+        };
+        USimpleError::new(
+            status_code,
+            translate!("timeout-error-failed-to-execute-process", "error" => err),
+        )
+    })?;
     unblock_sigchld();
     catch_sigterm();
     // Wait for the child process for the specified time period.

src/uu/timeout/src/timeout.rs

@@ -3,7 +3,7 @@
 // For the full copyright and license information, please view the LICENSE
 // file that was distributed with this source code.
 
-// spell-checker:ignore (vars/api) fcntl setrlimit setitimer rubout pollable sysconf pgrp
+// spell-checker:ignore (vars/api) fcntl setrlimit setitimer rubout pollable sysconf pgrp GETFD
 // spell-checker:ignore (vars/signals) ABRT ALRM CHLD SEGV SIGABRT SIGALRM SIGBUS SIGCHLD SIGCONT SIGDANGER SIGEMT SIGFPE SIGHUP SIGILL SIGINFO SIGINT SIGIO SIGIOT SIGKILL SIGMIGRATE SIGMSG SIGPIPE SIGPRE SIGPROF SIGPWR SIGQUIT SIGSEGV SIGSTOP SIGSYS SIGTALRM SIGTERM SIGTRAP SIGTSTP SIGTHR SIGTTIN SIGTTOU SIGURG SIGUSR SIGVIRT SIGVTALRM SIGWINCH SIGXCPU SIGXFSZ STKFLT PWR THR TSTP TTIN TTOU VIRT VTALRM XCPU XFSZ SIGCLD SIGPOLL SIGWAITING SIGAIOCANCEL SIGLWP SIGFREEZE SIGTHAW SIGCANCEL SIGLOST SIGXRES SIGJVM SIGRTMIN SIGRT SIGRTMAX TALRM AIOCANCEL XRES RTMIN RTMAX LTOSTOP
 
 //! This module provides a way to handle signals in a platform-independent way.
@@ -426,6 +426,73 @@ pub fn ignore_interrupts() -> Result<(), Errno> {
     unsafe { signal(SIGINT, SigIgn) }.map(|_| ())
 }
 
+// Detect closed stdin/stdout before Rust reopens them as /dev/null (see issue #2873)
+#[cfg(unix)]
+use std::sync::atomic::{AtomicBool, Ordering};
+
+#[cfg(unix)]
+static STDIN_WAS_CLOSED: AtomicBool = AtomicBool::new(false);
+#[cfg(unix)]
+static STDOUT_WAS_CLOSED: AtomicBool = AtomicBool::new(false);
+
+#[cfg(unix)]
+#[allow(clippy::missing_safety_doc)]
+pub unsafe extern "C" fn capture_stdio_state() {
+    use nix::libc;
+    unsafe {
+        STDIN_WAS_CLOSED.store(
+            libc::fcntl(libc::STDIN_FILENO, libc::F_GETFD) == -1,
+            Ordering::Relaxed,
+        );
+        STDOUT_WAS_CLOSED.store(
+            libc::fcntl(libc::STDOUT_FILENO, libc::F_GETFD) == -1,
+            Ordering::Relaxed,
+        );
+    }
+}
+
+#[macro_export]
+#[cfg(unix)]
+macro_rules! init_stdio_state_capture {
+    () => {
+        #[cfg(not(target_os = "macos"))]
+        #[used]
+        #[unsafe(link_section = ".init_array")]
+        static CAPTURE_STDIO_STATE: unsafe extern "C" fn() = $crate::signals::capture_stdio_state;
+
+        #[cfg(target_os = "macos")]
+        #[used]
+        #[unsafe(link_section = "__DATA,__mod_init_func")]
+        static CAPTURE_STDIO_STATE: unsafe extern "C" fn() = $crate::signals::capture_stdio_state;
+    };
+}
+
+#[macro_export]
+#[cfg(not(unix))]
+macro_rules! init_stdio_state_capture {
+    () => {};
+}
+
+#[cfg(unix)]
+pub fn stdin_was_closed() -> bool {
+    STDIN_WAS_CLOSED.load(Ordering::Relaxed)
+}
+
+#[cfg(not(unix))]
+pub const fn stdin_was_closed() -> bool {
+    false
+}
+
+#[cfg(unix)]
+pub fn stdout_was_closed() -> bool {
+    STDOUT_WAS_CLOSED.load(Ordering::Relaxed)
+}
+
+#[cfg(not(unix))]
+pub const fn stdout_was_closed() -> bool {
+    false
+}
+
 #[test]
 fn signal_by_value() {
     assert_eq!(signal_by_name_or_value("0"), Some(0));