Fix of use-after-free error #211

[ateska]

Fixes #211

[bk138]

Hi @ateska
were you running a multi-threaded server?

[Quentin01]

The issue is happening on Virtualbox (see Virtualbox Ticket #17396).

Because Virtualbox is using a multi-threaded server with pthread enabled the use-after-free results in a wait to an non-existing condition-variable, leading to a freeze of the Virtual Machine in a buggy state.

This PR correctly fixes the issue even if its incomplete:

• The lock is not really optional, we should add it (there are other parts of the code iterating on the clients and freeing them in a loop)
• Because the clientConnectionGone is freeing the client and all its regions, the clientOutput thread is segfaulting when accessing cl->modifiedRegion.

I do have a fix for both and will provide a PR ASAP.

[ateska]

The issue happens on iOS platform when a server and client that runs in separated threads shutdowns simultaneously. It happens in 100% cases however - not the typical race condition.

[Quentin01]

The issue is also 100% reproducible in Virtualbox. If you do have a client connected to the multithreaded-server during the call to rfbShutdownServer, the process will wait on the non-existing condition-variable forever.

Even if Virtualbox doesn't really crash directly in the first use-after-free.

[bk138]

Closing this in favour of #238, thanks for spotting #211, @ateska!