paste::get_contents leaks file descriptors to unix pipes

[e00E]

sway version 1.4

[package]
name = "clipboard"
version = "0.1.0"
edition = "2018"

[dependencies]
wl-clipboard-rs = "0.4"

use wl_clipboard_rs::paste::{get_contents, ClipboardType, MimeType, Seat};

fn main() {
    let mut i: u64 = 0;
    loop {
        println!("iteration {}", i);
        // Sometimes the program hangs here forever after exhausting file descriptors.
        let result = get_contents(ClipboardType::Regular, Seat::Unspecified, MimeType::Text);
        if let Err(err) = result {
            println!("first error: {}", err);
            // Sleep so that we have time to inspect number of file descriptors.
            std::thread::sleep(std::time::Duration::from_secs(10));
            return;
        }
        i += 1;
    }
}

cargo build && target/debug/clipboard

On my machine I usually see output like:

...
iteration 454
[wayland-client] Protocol error 1 on object wl_display@1: invalid arguments for [email protected]
first error: Wayland compositor communication error

When this happens I check the open file descriptors (the binary is called clipboard):

lsof -p `pidof clipboard`

COMMAND      PID USER   FD   TYPE             DEVICE SIZE/OFF     NODE NAME
...
clipboard 265393    user  457u  unix 0x0000000066c7d124      0t0  9900913 type=STREAM

This number is going to depend on your system's file descriptor limit.

As mentioned in the comment, sometimes get_contents hangs forever instead. If you remove the return in the error branch and continue calling get_contents more errors will happen and get_contents will eventually hang.

[YaLTeR]

Thanks for the detailed report!

I think I need to finish updating to the newest wayland-rs before I investigate this. I should have most of the update done in the wayland-rs-master branch.

[YaLTeR]

I've updated wayland-rs, do you mind re-testing on latest master in case it was an old wayland-rs issue?

[e00E]

Still happens

[YaLTeR]

I investigated, looks like it's Smithay/wayland-rs#235 again. Using the native_lib feature makes the issue go away.

[YaLTeR]

Upon bringing it up with the wayland-rs maintainer, looks like the issue isn't that, but runs a bit deeper: Smithay/wayland-rs#376

[YaLTeR]

A possible workaround is to have a long-lived connection in wl-clipboard-rs instead of creating a new one every time, while making sure to destroy all objects when they are no longer needed.

[dnut]

I'm seeing the same problem in the latest version of wl-clipboard-rs. Looks like the issue is fixed in wayland-rs. Should we upgrade to wayland-client 0.29?

[YaLTeR]

I think it's only supposed to be fixed in wayland-client 0.30, which is the "full rework of the crate and API" version and AFAICT is not released yet.

[YaLTeR]

I've just finished porting to wayland-rs 0.30. Could you check if this is still an issue with the latest git commit?

[dnut]

I see EMFILE panics many times per day, which I believe is caused by this. I'll test this over the next few days and share my results.

[YaLTeR]

Hey, so how'd it go?