gh-87135: Hang non-main threads that attempt to acquire the GIL during finalization (GH-105805)

Instead of surprise crashes and memory corruption, we now hang threads that attempt to re-enter the Python interpreter after Python runtime finalization has started. These are typically daemon threads (our long standing mis-feature) but could also be threads spawned by extension modules that then try to call into Python. This marks the `PyThread_exit_thread` public C API as deprecated as there is no plausible safe way to accomplish that on any supported platform in the face of things like C++ code with finalizers anywhere on a thread's stack. Doing this was the least bad option.

Co-authored-by: Gregory P. Smith <greg@krypto.org>

Doc/c-api/init.rst

@@ -430,7 +430,11 @@ Initializing and finalizing the interpreter
    Some memory allocated by extension modules may not be freed.  Some extensions may not
    work properly if their initialization routine is called more than once; this can
    happen if an application calls :c:func:`Py_Initialize` and :c:func:`Py_FinalizeEx`
-   more than once.
+   more than once.  :c:func:`Py_FinalizeEx` must not be called recursively from
+   within itself.  Therefore, it must not be called by any code that may be run
+   as part of the interpreter shutdown process, such as :py:mod:`atexit`
+   handlers, object finalizers, or any code that may be run while flushing the
+   stdout and stderr files.
 
    .. audit-event:: cpython._PySys_ClearAuditHooks "" c.Py_FinalizeEx
 
@@ -960,6 +964,37 @@ thread, where the CPython global runtime was originally initialized.
 The only exception is if :c:func:`exec` will be called immediately
 after.
 
+.. _cautions-regarding-runtime-finalization:
+
+Cautions regarding runtime finalization
+---------------------------------------
+
+In the late stage of :term:`interpreter shutdown`, after attempting to wait for
+non-daemon threads to exit (though this can be interrupted by
+:class:`KeyboardInterrupt`) and running the :mod:`atexit` functions, the runtime
+is marked as *finalizing*: :c:func:`_Py_IsFinalizing` and
+:func:`sys.is_finalizing` return true.  At this point, only the *finalization
+thread* that initiated finalization (typically the main thread) is allowed to
+acquire the :term:`GIL`.
+
+If any thread, other than the finalization thread, attempts to acquire the GIL
+during finalization, either explicitly via :c:func:`PyGILState_Ensure`,
+:c:macro:`Py_END_ALLOW_THREADS`, :c:func:`PyEval_AcquireThread`, or
+:c:func:`PyEval_AcquireLock`, or implicitly when the interpreter attempts to
+reacquire it after having yielded it, the thread enters **a permanently blocked
+state** where it remains until the program exits.  In most cases this is
+harmless, but this can result in deadlock if a later stage of finalization
+attempts to acquire a lock owned by the blocked thread, or otherwise waits on
+the blocked thread.
+
+Gross? Yes. This prevents random crashes and/or unexpectedly skipped C++
+finalizations further up the call stack when such threads were forcibly exited
+here in CPython 3.13 and earlier. The CPython runtime GIL acquiring C APIs
+have never had any error reporting or handling expectations at GIL acquisition
+time that would've allowed for graceful exit from this situation. Changing that
+would require new stable C APIs and rewriting the majority of C code in the
+CPython ecosystem to use those with error handling.
+
 
 High-level API
 --------------
@@ -1033,11 +1068,14 @@ code, or when embedding the Python interpreter:
    ensues.
 
    .. note::
-      Calling this function from a thread when the runtime is finalizing
+      Calling this function from a thread when the runtime is finalizing will
-      will terminate the thread, even if the thread was not created by Python.
+      hang the thread until the program exits, even if the thread was not
-      You can use :c:func:`Py_IsFinalizing` or :func:`sys.is_finalizing` to
+      created by Python.  Refer to
-      check if the interpreter is in process of being finalized before calling
+      :ref:`cautions-regarding-runtime-finalization` for more details.
-      this function to avoid unwanted termination.
+
+   .. versionchanged:: next
+      Hangs the current thread, rather than terminating it, if called while the
+      interpreter is finalizing.
 
 .. c:function:: PyThreadState* PyThreadState_Get()
 
@@ -1092,11 +1130,14 @@ with sub-interpreters:
    to call arbitrary Python code.  Failure is a fatal error.
 
    .. note::
-      Calling this function from a thread when the runtime is finalizing
+      Calling this function from a thread when the runtime is finalizing will
-      will terminate the thread, even if the thread was not created by Python.
+      hang the thread until the program exits, even if the thread was not
-      You can use :c:func:`Py_IsFinalizing` or :func:`sys.is_finalizing` to
+      created by Python.  Refer to
-      check if the interpreter is in process of being finalized before calling
+      :ref:`cautions-regarding-runtime-finalization` for more details.
-      this function to avoid unwanted termination.
+
+   .. versionchanged:: next
+      Hangs the current thread, rather than terminating it, if called while the
+      interpreter is finalizing.
 
 .. c:function:: void PyGILState_Release(PyGILState_STATE)
 
@@ -1374,17 +1415,20 @@ All of the following functions must be called after :c:func:`Py_Initialize`.
    If this thread already has the lock, deadlock ensues.
 
    .. note::
-      Calling this function from a thread when the runtime is finalizing
+      Calling this function from a thread when the runtime is finalizing will
-      will terminate the thread, even if the thread was not created by Python.
+      hang the thread until the program exits, even if the thread was not
-      You can use :c:func:`Py_IsFinalizing` or :func:`sys.is_finalizing` to
+      created by Python.  Refer to
-      check if the interpreter is in process of being finalized before calling
+      :ref:`cautions-regarding-runtime-finalization` for more details.
-      this function to avoid unwanted termination.
 
    .. versionchanged:: 3.8
       Updated to be consistent with :c:func:`PyEval_RestoreThread`,
       :c:func:`Py_END_ALLOW_THREADS`, and :c:func:`PyGILState_Ensure`,
       and terminate the current thread if called while the interpreter is finalizing.
 
+   .. versionchanged:: next
+      Hangs the current thread, rather than terminating it, if called while the
+      interpreter is finalizing.
+
    :c:func:`PyEval_RestoreThread` is a higher-level function which is always
    available (even when threads have not been initialized).
 

Include/internal/pycore_pythread.h

@@ -152,6 +152,19 @@ PyAPI_FUNC(int) PyThread_join_thread(PyThread_handle_t);
  * a non-zero value on failure.
  */
 PyAPI_FUNC(int) PyThread_detach_thread(PyThread_handle_t);
+/*
+ * Hangs the thread indefinitely without exiting it.
+ *
+ * gh-87135: There is no safe way to exit a thread other than returning
+ * normally from its start function.  This is used during finalization in lieu
+ * of actually exiting the thread.  Since the program is expected to terminate
+ * soon anyway, it does not matter if the thread stack stays around until then.
+ *
+ * This is unfortunate for embedders who may not be terminating their process
+ * when they're done with the interpreter, but our C API design does not allow
+ * for safely exiting threads attempting to re-enter Python post finalization.
+ */
+void _Py_NO_RETURN PyThread_hang_thread(void);
 
 #ifdef __cplusplus
 }

Include/pythread.h

@@ -17,7 +17,26 @@ typedef enum PyLockStatus {
 
 PyAPI_FUNC(void) PyThread_init_thread(void);
 PyAPI_FUNC(unsigned long) PyThread_start_new_thread(void (*)(void *), void *);
-PyAPI_FUNC(void) _Py_NO_RETURN PyThread_exit_thread(void);
+/* Terminates the current thread. Considered unsafe.
+ *
+ * WARNING: This function is only safe to call if all functions in the full call
+ * stack are written to safely allow it.  Additionally, the behavior is
+ * platform-dependent.  This function should be avoided, and is no longer called
+ * by Python itself.  It is retained only for compatibility with existing C
+ * extension code.
+ *
+ * With pthreads, calls `pthread_exit` causes some libcs (glibc?) to attempt to
+ * unwind the stack and call C++ destructors; if a `noexcept` function is
+ * reached, they may terminate the process. Others (macOS) do unwinding.
+ *
+ * On Windows, calls `_endthreadex` which kills the thread without calling C++
+ * destructors.
+ *
+ * In either case there is a risk of invalid references remaining to data on the
+ * thread stack.
+ */
+Py_DEPRECATED(3.14) PyAPI_FUNC(void) _Py_NO_RETURN PyThread_exit_thread(void);
+
 PyAPI_FUNC(unsigned long) PyThread_get_thread_ident(void);
 
 #if (defined(__APPLE__) || defined(__linux__) || defined(_WIN32) \

Lib/test/test_threading.py

@@ -1171,6 +1171,76 @@ class ThreadTests(BaseTestCase):
         self.assertEqual(out.strip(), b"OK")
         self.assertIn(b"can't create new thread at interpreter shutdown", err)
 
+    @cpython_only
+    def test_finalize_daemon_thread_hang(self):
+        if support.check_sanitizer(thread=True, memory=True):
+            # the thread running `time.sleep(100)` below will still be alive
+            # at process exit
+            self.skipTest(
+                    "https://github.com/python/cpython/issues/124878 - Known"
+                    " race condition that TSAN identifies.")
+        # gh-87135: tests that daemon threads hang during finalization
+        script = textwrap.dedent('''
+            import os
+            import sys
+            import threading
+            import time
+            import _testcapi
+
+            lock = threading.Lock()
+            lock.acquire()
+            thread_started_event = threading.Event()
+            def thread_func():
+                try:
+                    thread_started_event.set()
+                    _testcapi.finalize_thread_hang(lock.acquire)
+                finally:
+                    # Control must not reach here.
+                    os._exit(2)
+
+            t = threading.Thread(target=thread_func)
+            t.daemon = True
+            t.start()
+            thread_started_event.wait()
+            # Sleep to ensure daemon thread is blocked on `lock.acquire`
+            #
+            # Note: This test is designed so that in the unlikely case that
+            # `0.1` seconds is not sufficient time for the thread to become
+            # blocked on `lock.acquire`, the test will still pass, it just
+            # won't be properly testing the thread behavior during
+            # finalization.
+            time.sleep(0.1)
+
+            def run_during_finalization():
+                # Wake up daemon thread
+                lock.release()
+                # Sleep to give the daemon thread time to crash if it is going
+                # to.
+                #
+                # Note: If due to an exceptionally slow execution this delay is
+                # insufficient, the test will still pass but will simply be
+                # ineffective as a test.
+                time.sleep(0.1)
+                # If control reaches here, the test succeeded.
+                os._exit(0)
+
+            # Replace sys.stderr.flush as a way to run code during finalization
+            orig_flush = sys.stderr.flush
+            def do_flush(*args, **kwargs):
+                orig_flush(*args, **kwargs)
+                if not sys.is_finalizing:
+                    return
+                sys.stderr.flush = orig_flush
+                run_during_finalization()
+
+            sys.stderr.flush = do_flush
+
+            # If the follow exit code is retained, `run_during_finalization`
+            # did not run.
+            sys.exit(1)
+        ''')
+        assert_python_ok("-c", script)
+
 class ThreadJoinOnShutdown(BaseTestCase):
 
     def _run_and_join(self, script):

Misc/NEWS.d/next/C API/2022-08-05-19-41-20.gh-issue-87135.SCNBYj.rst

@@ -0,0 +1,15 @@
+Attempting to acquire the GIL after runtime finalization has begun in a
+different thread now causes the thread to hang rather than terminate, which
+avoids potential crashes or memory corruption caused by attempting to
+terminate a thread that is running code not specifically designed to support
+termination. In most cases this hanging is harmless since the process will
+soon exit anyway.
+
+The ``PyThread_exit_thread`` function is now deprecated.  Its behavior is
+inconsistent across platforms, and it can only be used safely in the
+unlikely case that every function in the entire call stack has been designed
+to support the platform-dependent termination mechanism.  It is recommended
+that users of this function change their design to not require thread
+termination.  In the unlikely case that thread termination is needed and can
+be done safely, users may migrate to calling platform-specific APIs such as
+``pthread_exit`` (POSIX) or ``_endthreadex`` (Windows) directly.

Modules/_testcapimodule.c

@@ -3310,6 +3310,35 @@ test_critical_sections(PyObject *module, PyObject *Py_UNUSED(args))
     Py_RETURN_NONE;
 }
 
+// Used by `finalize_thread_hang`.
+#ifdef _POSIX_THREADS
+static void finalize_thread_hang_cleanup_callback(void *Py_UNUSED(arg)) {
+    // Should not reach here.
+    Py_FatalError("pthread thread termination was triggered unexpectedly");
+}
+#endif
+
+// Tests that finalization does not trigger pthread cleanup.
+//
+// Must be called with a single nullary callable function that should block
+// (with GIL released) until finalization is in progress.
+static PyObject *
+finalize_thread_hang(PyObject *self, PyObject *callback)
+{
+    // WASI builds some pthread stuff but doesn't have these APIs today?
+#if defined(_POSIX_THREADS) && !defined(__wasi__)
+    pthread_cleanup_push(finalize_thread_hang_cleanup_callback, NULL);
+#endif
+    PyObject_CallNoArgs(callback);
+    // Should not reach here.
+    Py_FatalError("thread unexpectedly did not hang");
+#if defined(_POSIX_THREADS) && !defined(__wasi__)
+    pthread_cleanup_pop(0);
+#endif
+    Py_RETURN_NONE;
+}
+
+
 static PyMethodDef TestMethods[] = {
     {"set_errno",               set_errno,                       METH_VARARGS},
     {"test_config",             test_config,                     METH_NOARGS},
@@ -3449,6 +3478,7 @@ static PyMethodDef TestMethods[] = {
     {"test_weakref_capi", test_weakref_capi, METH_NOARGS},
     {"function_set_warning", function_set_warning, METH_NOARGS},
     {"test_critical_sections", test_critical_sections, METH_NOARGS},
+    {"finalize_thread_hang", finalize_thread_hang, METH_O, NULL},
     {NULL, NULL} /* sentinel */
 };
 

Python/ceval_gil.c

@@ -7,6 +7,7 @@
 #include "pycore_pylifecycle.h"   // _PyErr_Print()
 #include "pycore_pymem.h"         // _PyMem_IsPtrFreed()
 #include "pycore_pystats.h"       // _Py_PrintSpecializationStats()
+#include "pycore_pythread.h"      // PyThread_hang_thread()
 
 /*
    Notes about the implementation:
@@ -277,10 +278,9 @@ drop_gil(PyInterpreterState *interp, PyThreadState *tstate, int final_release)
 /* Take the GIL.
 
    The function saves errno at entry and restores its value at exit.
+   It may hang rather than return if the interpreter has been finalized.
 
-   tstate must be non-NULL.
+   tstate must be non-NULL. */
-
-   Returns 1 if the GIL was acquired, or 0 if not. */
 static void
 take_gil(PyThreadState *tstate)
 {
@@ -293,12 +293,18 @@ take_gil(PyThreadState *tstate)
 
     if (_PyThreadState_MustExit(tstate)) {
         /* bpo-39877: If Py_Finalize() has been called and tstate is not the
-           thread which called Py_Finalize(), exit immediately the thread.
+           thread which called Py_Finalize(), this thread cannot continue.
 
            This code path can be reached by a daemon thread after Py_Finalize()
            completes. In this case, tstate is a dangling pointer: points to
-           PyThreadState freed memory. */
+           PyThreadState freed memory.
-        PyThread_exit_thread();
+
+           This used to call a *thread_exit API, but that was not safe as it
+           lacks stack unwinding and local variable destruction important to
+           C++. gh-87135: The best that can be done is to hang the thread as
+           the public APIs calling this have no error reporting mechanism (!).
+         */
+        PyThread_hang_thread();
     }
 
     assert(_PyThreadState_CheckConsistency(tstate));
@@ -342,7 +348,9 @@ take_gil(PyThreadState *tstate)
                 if (drop_requested) {
                     _Py_unset_eval_breaker_bit(holder_tstate, _PY_GIL_DROP_REQUEST_BIT);
                 }
-                PyThread_exit_thread();
+                // gh-87135: hang the thread as *thread_exit() is not a safe
+                // API. It lacks stack unwind and local variable destruction.
+                PyThread_hang_thread();
             }
             assert(_PyThreadState_CheckConsistency(tstate));
 
@@ -383,7 +391,7 @@ take_gil(PyThreadState *tstate)
 
     if (_PyThreadState_MustExit(tstate)) {
         /* bpo-36475: If Py_Finalize() has been called and tstate is not
-           the thread which called Py_Finalize(), exit immediately the
+           the thread which called Py_Finalize(), gh-87135: hang the
            thread.
 
            This code path can be reached by a daemon thread which was waiting
@@ -393,7 +401,7 @@ take_gil(PyThreadState *tstate)
         /* tstate could be a dangling pointer, so don't pass it to
            drop_gil(). */
         drop_gil(interp, NULL, 1);
-        PyThread_exit_thread();
+        PyThread_hang_thread();
     }
     assert(_PyThreadState_CheckConsistency(tstate));
 

Python/pylifecycle.c

@@ -2020,7 +2020,7 @@ _Py_Finalize(_PyRuntimeState *runtime)
     /* Ensure that remaining threads are detached */
     _PyEval_StopTheWorldAll(runtime);
 
-    /* Remaining daemon threads will automatically exit
+    /* Remaining daemon threads will be trapped in PyThread_hang_thread
        when they attempt to take the GIL (ex: PyEval_RestoreThread()). */
     _PyInterpreterState_SetFinalizing(tstate->interp, tstate);
     _PyRuntimeState_SetFinalizing(runtime, tstate);

Python/thread_nt.h

@@ -291,6 +291,14 @@ PyThread_exit_thread(void)
     _endthreadex(0);
 }
 
+void _Py_NO_RETURN
+PyThread_hang_thread(void)
+{
+    while (1) {
+        SleepEx(INFINITE, TRUE);
+    }
+}
+
 /*
  * Lock support. It has to be implemented as semaphores.
  * I [Dag] tried to implement it with mutex but I could find a way to

Python/thread_pthread.h

@@ -16,6 +16,7 @@
 #undef destructor
 #endif
 #include <signal.h>
+#include <unistd.h>             /* pause(), also getthrid() on OpenBSD */
 
 #if defined(__linux__)
 #   include <sys/syscall.h>     /* syscall(SYS_gettid) */
@@ -23,8 +24,6 @@
 #   include <pthread_np.h>      /* pthread_getthreadid_np() */
 #elif defined(__FreeBSD_kernel__)
 #   include <sys/syscall.h>     /* syscall(SYS_thr_self) */
-#elif defined(__OpenBSD__)
-#   include <unistd.h>          /* getthrid() */
 #elif defined(_AIX)
 #   include <sys/thread.h>      /* thread_self() */
 #elif defined(__NetBSD__)
@@ -419,6 +418,18 @@ PyThread_exit_thread(void)
 #endif
 }
 
+void _Py_NO_RETURN
+PyThread_hang_thread(void)
+{
+    while (1) {
+#if defined(__wasi__)
+        sleep(9999999);  // WASI doesn't have pause() ?!
+#else
+        pause();
+#endif
+    }
+}
+
 #ifdef USE_SEMAPHORES
 
 /*