Function drive_socket_blocking 

pub fn drive_socket_blocking(
    pool: &WorkerPool<2>,
    stream: TcpStream,
    label: &str,
    config: &PumpConfig,
) -> Result<(GuardedReader, GuardedWriter)>

Drive one already-connected socket with two blocking pumps, returning the two owning adapters.

This is the whole seam for a caller that has no teardown thread: hand it a connected TcpStream, receive an AsyncRead and an AsyncWrite that the protocol code cannot tell from a split socket, with both pump threads owned by the values returned.

The socket’s blocking mode is taken over here (own_blocking_mode, crate-private), and fatally: it is what makes both pumps’ bounds hold by construction rather than wherever a flag or option is honoured, so a target that refused it would be a target this seam cannot bound, and that is worth a failed dial rather than a silent park. SO_RCVTIMEO is still set, also fatally, but as the value the reader’s wait uses and as the mechanism on the not(unix) arm; on unix the wait is a POLLIN poll. There is no send-side counterpart: write_frame_deadline owns its own bound and needs no socket option, so there is nothing here for a target that implements fewer of them to switch off.

The pool, and the two bands it carries

The two pumps come from one leased worker set of pool, taken atomically so a circuit at capacity can never hold one pump and block for the other. Their bands are the pool roster’s, not this function’s: at least one caller’s upstream C derives two — libca gives a circuit’s receive thread highestPriorityLevelBelow(initializing thread) and its send thread lowestPriorityLevelAbove(...) (tcpiiu.cpp:677-682), so the sender sits above the receiver and can always drain a queue the receiver’s work is filling — and a caller whose upstream uses one band for both (pvxs, one reactor thread) declares the same band twice in its roster. The Err is now admission’s: io::ErrorKind::WouldBlock when the circuit pool is full, otherwise a socket-option or thread-creation failure.