Rust jack deadlocking when dropped

kunerd · April 6, 2025, 8:22pm

I'm trying to integrate rust-jack into my iced app. Unfortunately I ran into a problem with handling the case if the jack server shuts down. I have created a minimal example here: rust_jack_threading/src/main.rs at main · kunerd/rust_jack_threading · GitHub.
The ConnectionLost message is send, but after that the line from the link doesn't get executed.

edit 1
The shutdown callback needs to fulfill these requirements: signal-safety(7) - Linux manual page
I don't know if the usage of a channel in there violates these. Maybe someone has an idea about this.

edit 2
I have updated the code to use the async-signal-safe channel from signal_hook crate, but it still doesn't work. I guess there is something wrong with my thread organization.

Please feel free to throw any idea in, even if it doesn't directly solve the problem.

edit 3
I could reproduce the problem with a stripped down version and even without the shutdown signal handler involved. The root seems to be in async_client.deactivate() which hangs if called, after the server has been shut down.

kunerd · April 7, 2025, 4:57pm

I can get rid of the deadlock by explicitly dropping the mutex guard before this return is called:

github.com/RustAudio/rust-jack

src/client/async_client.rs

main


      
          unsafe fn maybe_deactivate(&mut self) -> Result<Box<CallbackContext<N, P>>, Error> {
              let _m = CREATE_OR_DESTROY_CLIENT_MUTEX.lock().ok();
              if self.callback.is_none() {
                  return Err(Error::ClientIsNoLongerAlive);
              }
              let cb = self.callback.take().ok_or(Error::ClientIsNoLongerAlive)?;
              let client_ptr = cb.client.raw();
          
              // deactivate
              if j::jack_deactivate(client_ptr) != 0 {
                  return Err(Error::ClientDeactivationError);
              }
          
              // clear the callbacks
              clear_callbacks(client_ptr)?;
              // done, take ownership of callback
              if cb.has_panic.load(std::sync::atomic::Ordering::Relaxed) {
                  std::mem::forget(cb);
                  return Err(Error::ClientPanicked);
              }
              Ok(cb)

I hope someone could explain this in more detail, so that I eventually could provide an PR.

edit
I think it deadlocks because cb, which is holding client is dropped when returning. The drop impl of client itself tries to lock and bang...deadlock. This probably needs some refactoring.

It would be nice if someone with more knowledge in this field could confirm my assumption.

kunerd · April 8, 2025, 8:41am

I have done some more digging and was able to track down the bug. I opened a report containing a minimal example here: Thread hang or SEGFAULTs while trying to implement a connection retry · Issue #219 · RustAudio/rust-jack · GitHub

Topic		Replies	Views
[RustBook] 20.3 Multithreaded WebServer sometimes goes into deadlock when shutting down help	7	775	January 12, 2023
How to send close notify in the native-tls? help	3	91	February 22, 2025
Need advice for tracking down a deadlock help	3	953	January 12, 2023
Fix a deadlock in a simple channel help	7	271	May 27, 2024
Is This Deadlocked? help	7	201	January 21, 2025

Rust jack deadlocking when dropped

Related topics