These structs have misleading names. An ExitStatus[Error] is actually
a Unix wait status; an ExitCode is actually an exit status.
The Display impls are fixed, but the Debug impls are still misleading,
as reported in #74832.
Fix this by pretending that these internal structs are called
`unix_exit_status` and `unix_wait_status` as applicable. (We can't
actually rename the structs because of the way that the cross-platform
machinery works: the names are cross-platform.)
Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>
Add Linux-specific pidfd process extensions (take 2)
Continuation of #77168.
I addressed the following concerns from the original PR:
- make `CommandExt` and `ChildExt` sealed traits
- wrap file descriptors in `PidFd` struct representing ownership over the fd
- add `take_pidfd` to take the fd out of `Child`
- close fd when dropped
Tracking Issue: #82971
Background:
Over the last year, pidfd support was added to the Linux kernel. This
allows interacting with other processes. In particular, this allows
waiting on a child process with a timeout in a race-free way, bypassing
all of the awful signal-handler tricks that are usually required.
Pidfds can be obtained for a child process (as well as any other
process) via the `pidfd_open` syscall. Unfortunately, this requires
several conditions to hold in order to be race-free (i.e. the pid is not
reused).
Per `man pidfd_open`:
```
· the disposition of SIGCHLD has not been explicitly set to SIG_IGN
(see sigaction(2));
· the SA_NOCLDWAIT flag was not specified while establishing a han‐
dler for SIGCHLD or while setting the disposition of that signal to
SIG_DFL (see sigaction(2)); and
· the zombie process was not reaped elsewhere in the program (e.g.,
either by an asynchronously executed signal handler or by wait(2)
or similar in another thread).
If any of these conditions does not hold, then the child process
(along with a PID file descriptor that refers to it) should instead
be created using clone(2) with the CLONE_PIDFD flag.
```
Sadly, these conditions are impossible to guarantee once any libraries
are used. For example, C code runnng in a different thread could call
`wait()`, which is impossible to detect from Rust code trying to open a
pidfd.
While pid reuse issues should (hopefully) be rare in practice, we can do
better. By passing the `CLONE_PIDFD` flag to `clone()` or `clone3()`, we
can obtain a pidfd for the child process in a guaranteed race-free
manner.
This PR:
This PR adds Linux-specific process extension methods to allow obtaining
pidfds for processes spawned via the standard `Command` API. Other than
being made available to user code, the standard library does not make
use of these pidfds in any way. In particular, the implementation of
`Child::wait` is completely unchanged.
Two Linux-specific helper methods are added: `CommandExt::create_pidfd`
and `ChildExt::pidfd`. These methods are intended to serve as a building
block for libraries to build higher-level abstractions - in particular,
waiting on a process with a timeout.
I've included a basic test, which verifies that pidfds are created iff
the `create_pidfd` method is used. This test is somewhat special - it
should always succeed on systems with the `clone3` system call
available, and always fail on systems without `clone3` available. I'm
not sure how to best ensure this programatically.
This PR relies on the newer `clone3` system call to pass the `CLONE_FD`,
rather than the older `clone` system call. `clone3` was added to Linux
in the same release as pidfds, so this shouldn't unnecessarily limit the
kernel versions that this code supports.
Unresolved questions:
* What should the name of the feature gate be for these newly added
methods?
* Should the `pidfd` method distinguish between an error occurring
and `create_pidfd` not being called?
Use posix_spawn() on unix if program is a path
Previously `Command::spawn` would fall back to the non-posix_spawn based
implementation if the `PATH` environment variable was possibly changed.
On systems with a modern (g)libc `posix_spawn()` can be significantly
faster. If program is a path itself the `PATH` environment variable is
not used for the lookup and it should be safe to use the
`posix_spawnp()` method. [1]
We found this, because we have a cli application that effectively runs a
lot of subprocesses. It would sometimes noticeably hang while printing
output. Profiling showed that the process was spending the majority of
time in the kernel's `copy_page_range` function while spawning
subprocesses. During this time the process is completely blocked from
running, explaining why users were reporting the cli app hanging.
Through this we discovered that `std::process::Command` has a fast and
slow path for process execution. The fast path is backed by
`posix_spawnp()` and the slow path by fork/exec syscalls being called
explicitly. Using fork for process creation is supposed to be fast, but
it slows down as your process uses more memory. It's not because the
kernel copies the actual memory from the parent, but it does need to
copy the references to it (see `copy_page_range` above!). We ended up
using the slow path, because the command spawn implementation in falls
back to the slow path if it suspects the PATH environment variable was
changed.
Here is a smallish program demonstrating the slowdown before this code
change:
```
use std::process::Command;
use std::time::Instant;
fn main() {
let mut args = std::env::args().skip(1);
if let Some(size) = args.next() {
// Allocate some memory
let _xs: Vec<_> = std::iter::repeat(0)
.take(size.parse().expect("valid number"))
.collect();
let mut command = Command::new("/bin/sh");
command
.arg("-c")
.arg("echo hello");
if args.next().is_some() {
println!("Overriding PATH");
command.env("PATH", std::env::var("PATH").expect("PATH env var"));
}
let now = Instant::now();
let child = command
.spawn()
.expect("failed to execute process");
println!("Spawn took: {:?}", now.elapsed());
let output = child.wait_with_output().expect("failed to wait on process");
println!("Output: {:?}", output);
} else {
eprintln!("Usage: prog [size]");
std::process::exit(1);
}
()
}
```
Running it and passing different amounts of elements to use to allocate
memory shows that the time taken for `spawn()` can differ quite
significantly. In latter case the `posix_spawnp()` implementation is 30x
faster:
```
$ cargo run --release 10000000
...
Spawn took: 324.275µs
hello
$ cargo run --release 10000000 changepath
...
Overriding PATH
Spawn took: 2.346809ms
hello
$ cargo run --release 100000000
...
Spawn took: 387.842µs
hello
$ cargo run --release 100000000 changepath
...
Overriding PATH
Spawn took: 13.434677ms
hello
```
[1]: 5f72f9800b/posix/execvpe.c (L81)
Previously `Command::spawn` would fall back to the non-posix_spawn based
implementation if the `PATH` environment variable was possibly changed.
On systems with a modern (g)libc `posix_spawn()` can be significantly
faster. If program is a path itself the `PATH` environment variable is
not used for the lookup and it should be safe to use the
`posix_spawnp()` method. [1]
We found this, because we have a cli application that effectively runs a
lot of subprocesses. It would sometimes noticeably hang while printing
output. Profiling showed that the process was spending the majority of
time in the kernel's `copy_page_range` function while spawning
subprocesses. During this time the process is completely blocked from
running, explaining why users were reporting the cli app hanging.
Through this we discovered that `std::process::Command` has a fast and
slow path for process execution. The fast path is backed by
`posix_spawnp()` and the slow path by fork/exec syscalls being called
explicitly. Using fork for process creation is supposed to be fast, but
it slows down as your process uses more memory. It's not because the
kernel copies the actual memory from the parent, but it does need to
copy the references to it (see `copy_page_range` above!). We ended up
using the slow path, because the command spawn implementation in falls
back to the slow path if it suspects the PATH environment variable was
changed.
Here is a smallish program demonstrating the slowdown before this code
change:
```
use std::process::Command;
use std::time::Instant;
fn main() {
let mut args = std::env::args().skip(1);
if let Some(size) = args.next() {
// Allocate some memory
let _xs: Vec<_> = std::iter::repeat(0)
.take(size.parse().expect("valid number"))
.collect();
let mut command = Command::new("/bin/sh");
command
.arg("-c")
.arg("echo hello");
if args.next().is_some() {
println!("Overriding PATH");
command.env("PATH", std::env::var("PATH").expect("PATH env var"));
}
let now = Instant::now();
let child = command
.spawn()
.expect("failed to execute process");
println!("Spawn took: {:?}", now.elapsed());
let output = child.wait_with_output().expect("failed to wait on process");
println!("Output: {:?}", output);
} else {
eprintln!("Usage: prog [size]");
std::process::exit(1);
}
()
}
```
Running it and passing different amounts of elements to use to allocate
memory shows that the time taken for `spawn()` can differ quite
significantly. In latter case the `posix_spawnp()` implementation is 30x
faster:
```
$ cargo run --release 10000000
...
Spawn took: 324.275µs
hello
$ cargo run --release 10000000 changepath
...
Overriding PATH
Spawn took: 2.346809ms
hello
$ cargo run --release 100000000
...
Spawn took: 387.842µs
hello
$ cargo run --release 100000000 changepath
...
Overriding PATH
Spawn took: 13.434677ms
hello
```
[1]: 5f72f9800b/posix/execvpe.c (L81)
