Mongos (and potentially mongod?), when hitting open file limits, quickly creates a huge number (thousands per second) of log messages like :

Tue Aug 30 11:23:39 [mongosMain] Listener: accept() returns -1 errno:24 Too many open files

The messages are triggered from the event loop of Listener::initAndListen(), when a new client connection is detected but cannot be accepted for lack of fds, we log an error. However the event loop continues and the error continues to be retriggered.

Probably needs special-casing as done for other error types.

Untested on 1.9/2.0.