Changes:

1. Changed m_GlobControlLock to SharedMutex. Most of the uses are still exclusive, except:
   • locateSocket and other lookup functions - these use SharedLock on it
   • closeAllSockets calls closing function on all sockets first under SharedLock, then runs the loop again this time on ExclusiveLock
2. Removed locking of m_ControlLock in setListenerState. Instead the exclusive access is ensured through compare_exchange call on m_bListening atomic field with a fallback in case when the "winner" happened to fail setting up the listener.
3. A fix that should clear out the problem, which in some corner situation has led to a segfault, observed here. The problem isn't proven to be solved this way, but at least a path to an undefined (unwanted) behavior has been blocked

The ThreadSanitizer reports one situation described as "T1: Shared(A), Shared(B)", "T2: Shared(B), Exclusive(A)". According to the analysis this doesn't form a potential deadlock, although ThreadSanitizer reports it as such.