Add fork handling to GC

[CyberShadow]

Make sure a fork does not happen while GC code is running, thus leaving it in an inconsistent state.

[rainers]

I think this pointer should be reset before gcx is destroyed. BTW: as only members of gcx are affected, how about moving this and the initialization into Gcx.Dtor and initialize, respectively?

[CyberShadow]

The lock is in ConservativeGC, and we need to refer to it from the callbacks. Accessing ConservativeGC from Gcx seemed to me like a layering violation, but I see now there is precedent.

[CyberShadow]

I found a new bug while playing around to create a test case for this PR.

import core.stdc.stdio : printf;
import core.stdc.stdlib : exit;
import core.sys.posix.sys.types : pid_t;
import core.sys.posix.sys.wait : waitpid;
import core.sys.posix.unistd : fork;
import core.thread : Thread, getpid;

extern(C) int syscall(int, ...);
pid_t gettid() { return syscall(186/*SYS_gettid*/); }

void main()
{
	foreach (t; 0 .. 10)
		new Thread({
			auto parent = gettid();
			foreach (n; 0 .. 100)
			{
				foreach (x; 0 .. 100)
					new ubyte[x];
				auto child = fork();
				assert(child >= 0);
				if (child == 0)
				{
					child = gettid();
					printf("%d has forked.\n", child);
					foreach (x; 0 .. 100)
						new ubyte[x];
					printf("%d is exiting.\n", child);
					exit(0);
				}
				else
				{
					printf("%d waiting on %d...\n", parent, child);
					waitpid(child, null, 0);
					printf("%d done waiting on %d.\n", parent, child);
				}
			}
		}).start();
}

This program doesn't exit for me, because some threads are locked in a waitpid. The strange thing, however, is that the process they're waiting on actually has exited and is a zombie. I don't yet understand what is happening.

[dlang-bot]

Thanks for your pull request, @CyberShadow!

Bugzilla references

Auto-close	Bugzilla	Severity	Description
✓	20271	normal	Handle forking in the GC

Testing this PR locally

If you don't have a local development environment setup, you can use Digger to test this PR:

dub fetch digger
dub run digger -- build "stable + druntime#2817"

[CyberShadow]

I will probably later move these out to a general Druntime fork handler which also takes care of the thread stuff.

[rainers]

That probably won't work as the GC can be swapped out at link time.

[CyberShadow]

Not sure what you mean. The atfork notifications would then be part of the abstract GC interface.