[3.11]pythonGH-112275: Fix HEAD_LOCK deadlock in child process after …

…fork (pythonGH-112336)

HEAD_LOCK is called from _PyEval_ReInitThreads->_PyThreadState_DeleteExcept before _PyRuntimeState_ReInitThreads reinit runtime->interpreters.mutex which might be locked before fork.

(cherry picked from commit 522799a)

Co-authored-by: ChuBoning <102216855+ChuBoning@users.noreply.github.com>
Co-authored-by: Łukasz Langa <lukasz@langa.pl>

Misc/NEWS.d/next/Core_and_Builtins/2024-09-04-18-20-11.gh-issue-112275.W_iMiB.rst

@@ -0,0 +1,3 @@
+A deadlock involving ``pystate.c``'s ``HEAD_LOCK`` in ``posixmodule.c``
+at fork is now fixed. Patch by ChuBoning based on previous Python 3.12
+fix by Victor Stinner.

Modules/posixmodule.c

@@ -598,6 +598,11 @@ PyOS_AfterFork_Child(void)
         goto fatal_error;
     }
 
+    status = _PyRuntimeState_ReInitThreads(runtime);
+    if (_PyStatus_EXCEPTION(status)) {
+        goto fatal_error;
+    }
+
     PyThreadState *tstate = _PyThreadState_GET();
     _Py_EnsureTstateNotNULL(tstate);
 
@@ -613,11 +618,6 @@ PyOS_AfterFork_Child(void)
 
     _PySignal_AfterFork();
 
-    status = _PyRuntimeState_ReInitThreads(runtime);
-    if (_PyStatus_EXCEPTION(status)) {
-        goto fatal_error;
-    }
-
     status = _PyInterpreterState_DeleteExceptMain(runtime);
     if (_PyStatus_EXCEPTION(status)) {
         goto fatal_error;