Reset subinterpreters after fork instead of deleting them.

Author: ZeroIntensity

Include/internal/pycore_pystate.h

@@ -22,8 +22,9 @@ extern "C" {
 // transitions between "attached" and "detached" on its own PyThreadState.
 //
 // The "suspended" states are used to implement stop-the-world pauses and to
-// merge biased reference counts on behalf of detached threads. They are only
-// used in `--disable-gil` builds.
+// merge biased reference counts on behalf of detached threads. They are
+// used for global pauses in all builds and per-interpreter pauses in
+// `--disable-gil` builds.
 // They are similar to the "detached" state in that the thread is not allowed
 // to call most Python APIs. A suspended thread trying to attach marks itself
 // as "suspended-waiting". Only the thread responsible for suspending it may
@@ -161,12 +162,12 @@ extern void _PyThreadState_Suspend(PyThreadState *tstate);
 // stop the world while holding the thread in the suspended state.
 // Returns 1 on success, 0 if the thread was not in the "detached" state.
 extern int _PyThreadState_TrySuspendDetached(PyThreadState *tstate);
+#endif
 
 // Resume a thread suspended by _PyThreadState_TrySuspendDetached() or a
 // stop-the-world pause: switch it back to "detached" or "detached-waiting"
 // and wake it if it is waiting to attach.
 extern void _PyThreadState_ResumeDetached(PyThreadState *tstate);
-#endif
 
 // Mark the thread state as "shutting down". This is used during interpreter
 // and runtime finalization. The thread may no longer attach to the
@@ -245,7 +246,9 @@ extern void _PyThreadState_Bind(PyThreadState *tstate);
 PyAPI_FUNC(PyThreadState *) _PyThreadState_NewBound(
     PyInterpreterState *interp,
     int whence);
-extern PyThreadState * _PyThreadState_RemoveExcept(PyThreadState *tstate);
+extern PyThreadState * _PyThreadState_RemoveExcept(
+    PyThreadState *tstate,
+    int is_after_fork);
 extern void _PyThreadState_DeleteList(PyThreadState *list, int is_after_fork);
 extern void _PyThreadState_ClearMimallocHeaps(PyThreadState *tstate);
 
@@ -267,7 +270,6 @@ extern PyThreadState * _PyThreadState_Swap(
 extern PyStatus _PyInterpreterState_Enable(_PyRuntimeState *runtime);
 
 #ifdef HAVE_FORK
-extern PyStatus _PyInterpreterState_DeleteExceptMain(_PyRuntimeState *runtime);
 extern void _PySignal_AfterFork(void);
 #endif
 

Include/internal/pycore_qsbr.h

@@ -180,7 +180,7 @@ extern void
 _Py_qsbr_fini(PyInterpreterState *interp);
 
 extern void
-_Py_qsbr_after_fork(struct _PyThreadStateImpl *tstate);
+_Py_qsbr_after_fork(PyInterpreterState *interp);
 
 #ifdef __cplusplus
 }

Misc/NEWS.d/next/Core_and_Builtins/2026-10-23-12-00-00.gh-issue-139759.fork.rst

@@ -0,0 +1,2 @@
+Fix crashes and hangs when calling :func:`os.fork` with existing
+subinterpreters.

Modules/posixmodule.c

@@ -723,7 +723,9 @@ PyOS_BeforeFork(void)
     _PyImport_AcquireLock(interp);
     _PyEval_StopTheWorldAll(&_PyRuntime);
     HEAD_LOCK(&_PyRuntime);
-    _PyWarnings_BeforeFork(interp);
+    for (PyInterpreterState *i = PyInterpreterState_Head(); i != NULL; i = i->next) {
+        _PyWarnings_BeforeFork(i);
+    }
 }
 
 void
@@ -753,6 +755,78 @@ reset_asyncio_state(_PyThreadStateImpl *tstate)
     tstate->asyncio_running_task = NULL;
 }
 
+static PyStatus
+clear_threads_after_fork(PyThreadState *tstate)
+{
+    struct after_fork_state {
+        PyThreadState *tstate;
+        PyThreadState *deleted;
+        PyThreadState *running_main;
+    };
+
+    size_t count = 0;
+    for (PyInterpreterState *interp = PyInterpreterState_Head(); interp != NULL;
+         interp = interp->next) {
+        count++;
+    }
+    struct after_fork_state *states = PyMem_RawCalloc(count, sizeof(*states));
+    if (states == NULL) {
+        return _PyStatus_NO_MEMORY();
+    }
+
+    // Unlink all vanished threads before running any finalizers. Keep the
+    // subinterpreters busy until their dead thread states have been cleared.
+    size_t next = 1;
+    for (PyInterpreterState *interp = PyInterpreterState_Head(); interp != NULL;
+         interp = interp->next) {
+        size_t index = (interp == tstate->interp) ? 0 : next++;
+        struct after_fork_state *state = &states[index];
+        state->running_main = interp->threads.main;
+        if (state->running_main != NULL
+            && state->running_main->thread_id != tstate->thread_id) {
+            state->running_main = NULL;
+        }
+        state->tstate = (index == 0) ? tstate : PyThreadState_New(interp);
+        if (state->tstate == NULL) {
+            PyMem_RawFree(states);
+            return _PyStatus_NO_MEMORY();
+        }
+        state->deleted = _PyThreadState_RemoveExcept(state->tstate,
+                                                     /*is_after_fork=*/1);
+        if (index != 0) {
+            _Py_atomic_store_ptr(&interp->threads.main, state->tstate);
+        }
+        Py_ssize_t thread_count = 0;
+        _Py_FOR_EACH_TSTATE_UNLOCKED(interp, p) {
+#ifdef PY_HAVE_THREAD_NATIVE_ID
+            p->native_thread_id = tstate->native_thread_id;
+#endif
+            if (p->_whence == _PyThreadState_WHENCE_THREADING
+                || p->_whence == _PyThreadState_WHENCE_THREADING_DAEMON) {
+                thread_count++;
+            }
+        }
+        _Py_atomic_store_ssize(&interp->threads.count, thread_count);
+    }
+
+    // Clearing thread states can call destructors, so resume before doing so.
+    _PyEval_StartTheWorldAll(&_PyRuntime);
+    _PyThreadState_DeleteList(states[0].deleted, /*is_after_fork=*/1);
+    for (size_t i = 1; i < count; i++) {
+        struct after_fork_state *state = &states[i];
+        PyInterpreterState *interp = state->tstate->interp;
+        PyThreadState_Swap(state->tstate);
+        _PyThreadState_DeleteList(state->deleted, /*is_after_fork=*/1);
+        run_at_forkers(interp->after_forkers_child, 0);
+        _Py_atomic_store_ptr(&interp->threads.main, state->running_main);
+        PyThreadState_Clear(state->tstate);
+        PyThreadState_Swap(tstate);
+        PyThreadState_Delete(state->tstate);
+    }
+    PyMem_RawFree(states);
+    return _PyStatus_OK();
+}
+
 
 void
 PyOS_AfterFork_Child(void)
@@ -774,11 +848,6 @@ PyOS_AfterFork_Child(void)
     tstate->native_thread_id = PyThread_get_thread_native_id();
 #endif
 
-#ifdef Py_GIL_DISABLED
-    _Py_brc_after_fork(tstate->interp);
-    _Py_qsbr_after_fork((_PyThreadStateImpl *)tstate);
-#endif
-
     // Ideally we could guarantee tstate is running main.
     _PyInterpreterState_ReinitRunningMain(tstate);
 
@@ -794,32 +863,20 @@ PyOS_AfterFork_Child(void)
     _Py_jit_debug_mutex = (PyMutex){0};
 #endif
 
-    // Thread-state cleanup can run finalizers that issue warnings.
-    _PyWarnings_AfterFork(tstate->interp);
-
     reset_remotedebug_data(tstate);
 
     reset_asyncio_state((_PyThreadStateImpl *)tstate);
 
-    // Remove the dead thread states. We "start the world" once we are the only
-    // thread state left to undo the stop the world call in `PyOS_BeforeFork`.
-    // That needs to happen before `_PyThreadState_DeleteList`, because that
-    // may call destructors.
-    PyThreadState *list = _PyThreadState_RemoveExcept(tstate);
-    _PyEval_StartTheWorldAll(&_PyRuntime);
-    _PyThreadState_DeleteList(list, /*is_after_fork=*/1);
+    status = clear_threads_after_fork(tstate);
+    if (_PyStatus_EXCEPTION(status)) {
+        goto fatal_error;
+    }
 
     _PyImport_ReInitLock(tstate->interp);
     _PyImport_ReleaseLock(tstate->interp);
 
     _PySignal_AfterFork();
 
-    status = _PyInterpreterState_DeleteExceptMain(runtime);
-    if (_PyStatus_EXCEPTION(status)) {
-        goto fatal_error;
-    }
-    assert(_PyThreadState_GET() == tstate);
-
     status = _PyPerfTrampoline_AfterFork_Child();
     if (_PyStatus_EXCEPTION(status)) {
         goto fatal_error;

Python/ceval_gil.c

@@ -618,24 +618,26 @@ PyEval_ReleaseThread(PyThreadState *tstate)
 }
 
 #ifdef HAVE_FORK
-/* This function is called from PyOS_AfterFork_Child to re-initialize the
-   GIL and pending calls lock. */
+/* This function is called from PyOS_AfterFork_Child to re-initialize all GILs. */
 PyStatus
 _PyEval_ReInitThreads(PyThreadState *tstate)
 {
     assert(tstate->interp == _PyInterpreterState_Main());
 
-    struct _gil_runtime_state *gil = tstate->interp->ceval.gil;
-    if (!gil_created(gil)) {
-        return _PyStatus_OK();
+    for (PyInterpreterState *interp = PyInterpreterState_Head(); interp != NULL;
+         interp = interp->next) {
+        struct _gil_runtime_state *gil = interp->ceval.gil;
+        if (interp->ceval.own_gil && gil_created(gil)) {
+            recreate_gil(gil);
+        }
+        _Py_FOR_EACH_TSTATE_UNLOCKED(interp, p) {
+            p->holds_gil = 0;
+            p->gil_requested = 0;
+        }
     }
-    recreate_gil(gil);
 
     take_gil(tstate);
 
-    struct _pending_calls *pending = &tstate->interp->ceval.pending;
-    _PyMutex_at_fork_reinit(&pending->mutex);
-
     return _PyStatus_OK();
 }
 #endif

Python/pylifecycle.c

@@ -2357,9 +2357,8 @@ make_pre_finalization_calls(PyThreadState *tstate, int subinterpreters)
         }
 
         /* Stop the world to prevent other threads from creating threads or
-         * atexit callbacks. On the default build, this is simply locked by
-         * the GIL. For pending calls, we acquire the dedicated mutex, because
-         * Py_AddPendingCall() can be called without an attached thread state.
+         * atexit callbacks. For pending calls, we acquire the dedicated mutex,
+         * because Py_AddPendingCall() can be called without an attached thread state.
          */
         PyMutex_Lock(&interp->ceval.pending.mutex);
         _PyEval_StopTheWorldAll(interp->runtime);
@@ -2444,7 +2443,7 @@ _Py_Finalize(_PyRuntimeState *runtime)
        except if wait_for_thread_shutdown() has been cancelled by CTRL+C.
        We start the world once we are the only thread state left,
        before we call destructors. */
-    PyThreadState *list = _PyThreadState_RemoveExcept(tstate);
+    PyThreadState *list = _PyThreadState_RemoveExcept(tstate, /*is_after_fork=*/0);
     for (PyThreadState *p = list; p != NULL; p = p->next) {
         _PyThreadState_SetShuttingDown(p);
     }
@@ -2825,7 +2824,7 @@ Py_EndInterpreter(PyThreadState *tstate)
        when they attempt to take the GIL (ex: PyEval_RestoreThread()). */
     _PyInterpreterState_SetFinalizing(interp, tstate);
 
-    PyThreadState *list = _PyThreadState_RemoveExcept(tstate);
+    PyThreadState *list = _PyThreadState_RemoveExcept(tstate, /*is_after_fork=*/0);
     for (PyThreadState *p = list; p != NULL; p = p->next) {
         _PyThreadState_SetShuttingDown(p);
     }