Process-wide DNS resolver queue exhaustion denies outbound RTMP connects

[cursor]

Summary

All Client::connect() calls in a process share a single DNS worker queue capped at 32 jobs. The 33rd concurrent connect attempt fails immediately with ErrorCode::Timeout, even when DNS itself is healthy.

Attacker

Any workload in the same process that can trigger many concurrent outbound Client::connect() calls (remote user requesting many streams, compromised thread, or hostile co-hosted workload in the embedder).

Controlled input

33+ simultaneous connect attempts requiring DNS resolution.

Attack path

1. All clients share one mpsc::sync_channel DNS worker (DNS_TX, depth 32) in resolve_socket_addrs().
2. try_send on the 33rd job returns Err → ErrorCode::Timeout.
3. Legitimate concurrent outbound connects in the same process fail even though DNS is available.

Impact

Cross-client denial of service inside a single embedder process (e.g. media gateway spawning many outbound RTMP pulls). Unrelated legitimate connects fail when the queue is saturated.

Location

src/client/mod.rs

Remediation

Increase queue depth, block with a deadline instead of failing immediately on try_send, or isolate DNS resolution per connect with a global concurrency limit separate from the queue depth.