diff --git a/Misc/NEWS.d/next/Core and Builtins/2019-09-30-09-33-21.bpo-38006.UYlJum.rst b/Misc/NEWS.d/next/Core and Builtins/2019-09-30-09-33-21.bpo-38006.UYlJum.rst new file mode 100644 index 00000000000..c2dafae0afa --- /dev/null +++ b/Misc/NEWS.d/next/Core and Builtins/2019-09-30-09-33-21.bpo-38006.UYlJum.rst @@ -0,0 +1,3 @@ +Fix a bug due to the interaction of weakrefs and the cyclic garbage +collector. We must clear any weakrefs in garbage in order to prevent their +callbacks from executing and causing a crash. diff --git a/Modules/gcmodule.c b/Modules/gcmodule.c index f6e841d52ff..b95f676663a 100644 --- a/Modules/gcmodule.c +++ b/Modules/gcmodule.c @@ -678,6 +678,21 @@ handle_weakrefs(PyGC_Head *unreachable, PyGC_Head *old) op = FROM_GC(gc); next = GC_NEXT(gc); + if (PyWeakref_Check(op)) { + /* A weakref inside the unreachable set must be cleared. If we + * allow its callback to execute inside delete_garbage(), it + * could expose objects that have tp_clear already called on + * them. Or, it could resurrect unreachable objects. One way + * this can happen is if some container objects do not implement + * tp_traverse. Then, wr_object can be outside the unreachable + * set but can be deallocated as a result of breaking the + * reference cycle. If we don't clear the weakref, the callback + * will run and potentially cause a crash. See bpo-38006 for + * one example. + */ + _PyWeakref_ClearRef((PyWeakReference *)op); + } + if (! PyType_SUPPORTS_WEAKREFS(Py_TYPE(op))) continue; @@ -733,6 +748,8 @@ handle_weakrefs(PyGC_Head *unreachable, PyGC_Head *old) * is moved to wrcb_to_call in this case. */ if (gc_is_collecting(AS_GC(wr))) { + /* it should already have been cleared above */ + assert(wr->wr_object == Py_None); continue; }