Calling SQL_LockDatabase twice is crashing server

[Bara]

Environment

• SourceMod version: 1.11.0.6916
• Metamod: version: 1.11.0-dev+1148

Description

I randomly found the issue and is reproducible with the following code, sqlite and mysql doesn't matter.

Problematic Code (or Steps to Reproduce)

public void OnPluginStart()
{
  char sError[512];
  Database db = SQL_Connect("clientprefs", true, sError, sizeof(sError));

  if (db == null)
  {
      SetFailState("Error: %s", sError);
      return;
  }

  SQL_LockDatabase(db);
  SQL_LockDatabase(db);
  SQL_UnlockDatabase(db);
}

Last message in console was the following:

sm plugins load crashtest
**** WARNING: Watchdog timer exceeded, aborting!

I'm not really familiar how SQL_LockDatabase works, but this should be checked to avoid a crash like this.

[sirdigbot]

Seems the std::mutex used by LockDatabase has undefined behaviour if the currently-owning thread locks again.
Maybe that's it?

https://en.cppreference.com/w/cpp/thread/mutex/lock

If lock is called by a thread that already owns the mutex, the behavior is undefined: for example, the program may deadlock.

[Wend4r]

There is a similar problem with server crash when reloading plugin that is waiting for a SQL query response.
But I think it needs to be solved globally, rewriting the basis for working with different databases in other threads (one DB/TCP connection = one thread) to don't clog the queue flow (this problem has old server holders with a long response from DB due to many data and DB plugins (which delayed in the queue)).
And from SourceMod view point, which is basically designed to work with the Source Engine, also can use the Valve Thread Pool (aka -threads or IThreadPool) for this (it looks like #1689).

[Bara]

#1937 was merged, so I think this can be closed too

[peace-maker]

Your server would still deadlock, no? Unbalanced Lock/Unlock calls.