Read attention ASTs on shared listen sockets in TCPIP 5.5

[David Jones_21]

I use a shared listen socket to run multiple instances of a network daemon. Version 5.5 seems to have broken this application because one process deassigning the channel clears the attention AST set by a different process. Since only 1 process is 'at bat' to receive the next connection, losing the attention AST means new connections stay stuck the socket's accept queue.

The AST wasn't lost in previous versions of TCPIP services, so the daemon instances could be rundown or crash without affecting the ability of the other instances to continue getting connections.