Fix deadlock in message validator

[dknopik]

I noticed that sometimes, during a burst of incoming messages, Anchor would deadlock. More specifically, threads holding a processor permit would deadlock and the processor queue would become full shortly after.

The issue was doubly locking the network state. As RwLock (and thus tokio::sync::watch) avoids writer starvation by not allowing new readers while a writer is waiting, so trying to (read) borrow the lock while
a) holding a read lock already, and
b) a writer is waiting
is a deadlock, as we are now waiting for the writer, while the writer is waiting for us to release the already held read lock.

This PR fixes the issue by using the existing borrowed value instead of borrowing again. Furthermore, we explicitly drop the read guard to release it as early as possible.

[Copilot]

Copilot reviewed 1 out of 1 changed files in this pull request and generated no comments.

Comments suppressed due to low confidence (1)

anchor/message_validator/src/lib.rs:253

• [nitpick] Consider using a scoped block to limit the lifetime of 'network_state' rather than explicitly calling drop. This approach can help prevent accidental misuse of the lock in future modifications.

drop(network_state);

[diegomrsantos]

That's a good catch, thanks. Is there a way we can avoid repeating this in the future? It still seems error-prone.

[dknopik]

Well, we havelockbud in CI, which is supposed to catch stuff like this, but I guess it has its limits.

Otherwise we just have to stay extra vigilant during review, and adopt patterns like either dropping locks ASAP, or even better, acquiring them in a limited scope such as a function.