[Core]Fix coredump when m_pCryptoControl is Null

[kyolong]

#1182
maybe fixed ~

[maxsharabayko]

By the time this function CUDT::packData(..) is called, the m_pCryptoControl should exist. However, if you somehow manage to close a socket while being in this function, the m_pCryptoControl might have been destroyed in another thread.
It looks like something is wrong with mutexes protecting both cases.

[ethouris]

I'd keenly see the debug log - it will be cleanly seen in what order things happen and therefore how it was possible that packData was called at all on a socket being closed. AFAIR there should be first the m_bClosing and m_bBroken flags set first, then the crypto object deleted. And every function that would do any next thing with the data should first check these closing/broken flags before doing anything.

[ethouris]

My suggestion:

For testing, keep this change, but in the condition when !m_pCryptoControl add a log message so that you can see that it has happened, but print also the state. Something like:

LOGC(mglog.Error, log << "CryptoControl DELETED. connected=" << m_bConnected << " broken=" << m_bBroken << " opened=" << m_bOpened);

When you can see this log during testing, please see if when it happens the broken flag is set and possibly also connected flag is clear. If it is so, it means that the code works still correctly predictably, so the proper fix should be:

if (m_bBroken || !m_bConnected)
{
       return std::make_pair(0, enter_time);
}

in the beginning of packData method (at least after enter_time was assigned). With that, checking for m_pCryptoControl shall be no longer necessary.

[maxsharabayko]

Closing as fixed by #1315