bpo-30243: Fixed the possibility of a crash in _json. (#1420)

It was possible to get a core dump by using uninitialized
_json objects. Now __new__ methods create initialized objects.
__init__ methods are removed.

Author: serhiy-storchaka

Misc/NEWS

@@ -317,6 +317,10 @@ Extension Modules
 Library
 -------
 
+- bpo-30243: Removed the __init__ methods of _json's scanner and encoder.
+  Misusing them could cause memory leaks or crashes.  Now scanner and encoder
+  objects are completely initialized in the __new__ methods.
+
 - bpo-30215: Compiled regular expression objects with the re.LOCALE flag no
   longer depend on the locale at compile time.  Only the locale at matching
   time affects the result of matching.

Modules/_json.c

@@ -89,16 +89,12 @@ static PyObject *
 _build_rval_index_tuple(PyObject *rval, Py_ssize_t idx);
 static PyObject *
 scanner_new(PyTypeObject *type, PyObject *args, PyObject *kwds);
-static int
-scanner_init(PyObject *self, PyObject *args, PyObject *kwds);
 static void
 scanner_dealloc(PyObject *self);
 static int
 scanner_clear(PyObject *self);
 static PyObject *
 encoder_new(PyTypeObject *type, PyObject *args, PyObject *kwds);
-static int
-encoder_init(PyObject *self, PyObject *args, PyObject *kwds);
 static void
 encoder_dealloc(PyObject *self);
 static int
@@ -1200,38 +1196,21 @@ static PyObject *
 scanner_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
 {
     PyScannerObject *s;
-    s = (PyScannerObject *)type->tp_alloc(type, 0);
-    if (s != NULL) {
-        s->strict = NULL;
-        s->object_hook = NULL;
-        s->object_pairs_hook = NULL;
-        s->parse_float = NULL;
-        s->parse_int = NULL;
-        s->parse_constant = NULL;
-    }
-    return (PyObject *)s;
-}
-
-static int
-scanner_init(PyObject *self, PyObject *args, PyObject *kwds)
-{
-    /* Initialize Scanner object */
     PyObject *ctx;
     static char *kwlist[] = {"context", NULL};
-    PyScannerObject *s;
-
-    assert(PyScanner_Check(self));
-    s = (PyScannerObject *)self;
 
     if (!PyArg_ParseTupleAndKeywords(args, kwds, "O:make_scanner", kwlist, &ctx))
-        return -1;
+        return NULL;
 
-    if (s->memo == NULL) {
-        s->memo = PyDict_New();
-        if (s->memo == NULL)
-            goto bail;
+    s = (PyScannerObject *)type->tp_alloc(type, 0);
+    if (s == NULL) {
+        return NULL;
     }
 
+    s->memo = PyDict_New();
+    if (s->memo == NULL)
+        goto bail;
+
     /* All of these will fail "gracefully" so we don't need to verify them */
     s->strict = PyObject_GetAttrString(ctx, "strict");
     if (s->strict == NULL)
@@ -1252,16 +1231,11 @@ scanner_init(PyObject *self, PyObject *args, PyObject *kwds)
     if (s->parse_constant == NULL)
         goto bail;
 
-    return 0;
+    return (PyObject *)s;
 
 bail:
-    Py_CLEAR(s->strict);
-    Py_CLEAR(s->object_hook);
-    Py_CLEAR(s->object_pairs_hook);
-    Py_CLEAR(s->parse_float);
-    Py_CLEAR(s->parse_int);
-    Py_CLEAR(s->parse_constant);
-    return -1;
+    Py_DECREF(s);
+    return NULL;
 }
 
 PyDoc_STRVAR(scanner_doc, "JSON scanner object");
@@ -1303,7 +1277,7 @@ PyTypeObject PyScannerType = {
     0,                    /* tp_descr_get */
     0,                    /* tp_descr_set */
     0,                    /* tp_dictoffset */
-    scanner_init,                    /* tp_init */
+    0,                    /* tp_init */
     0,/* PyType_GenericAlloc, */        /* tp_alloc */
     scanner_new,          /* tp_new */
     0,/* PyObject_GC_Del, */              /* tp_free */
@@ -1312,48 +1286,30 @@ PyTypeObject PyScannerType = {
 static PyObject *
 encoder_new(PyTypeObject *type, PyObject *args, PyObject *kwds)
 {
-    PyEncoderObject *s;
-    s = (PyEncoderObject *)type->tp_alloc(type, 0);
-    if (s != NULL) {
-        s->markers = NULL;
-        s->defaultfn = NULL;
-        s->encoder = NULL;
-        s->indent = NULL;
-        s->key_separator = NULL;
-        s->item_separator = NULL;
-        s->sort_keys = NULL;
-        s->skipkeys = NULL;
-    }
-    return (PyObject *)s;
-}
-
-static int
-encoder_init(PyObject *self, PyObject *args, PyObject *kwds)
-{
-    /* initialize Encoder object */
     static char *kwlist[] = {"markers", "default", "encoder", "indent", "key_separator", "item_separator", "sort_keys", "skipkeys", "allow_nan", NULL};
 
     PyEncoderObject *s;
     PyObject *markers, *defaultfn, *encoder, *indent, *key_separator;
     PyObject *item_separator, *sort_keys, *skipkeys;
     int allow_nan;
 
-    assert(PyEncoder_Check(self));
-    s = (PyEncoderObject *)self;
-
     if (!PyArg_ParseTupleAndKeywords(args, kwds, "OOOOUUOOp:make_encoder", kwlist,
         &markers, &defaultfn, &encoder, &indent,
         &key_separator, &item_separator,
         &sort_keys, &skipkeys, &allow_nan))
-        return -1;
+        return NULL;
 
     if (markers != Py_None && !PyDict_Check(markers)) {
         PyErr_Format(PyExc_TypeError,
                      "make_encoder() argument 1 must be dict or None, "
                      "not %.200s", Py_TYPE(markers)->tp_name);
-        return -1;
+        return NULL;
     }
 
+    s = (PyEncoderObject *)type->tp_alloc(type, 0);
+    if (s == NULL)
+        return NULL;
+
     s->markers = markers;
     s->defaultfn = defaultfn;
     s->encoder = encoder;
@@ -1380,7 +1336,7 @@ encoder_init(PyObject *self, PyObject *args, PyObject *kwds)
     Py_INCREF(s->item_separator);
     Py_INCREF(s->sort_keys);
     Py_INCREF(s->skipkeys);
-    return 0;
+    return (PyObject *)s;
 }
 
 static PyObject *
@@ -1911,7 +1867,7 @@ PyTypeObject PyEncoderType = {
     0,                    /* tp_descr_get */
     0,                    /* tp_descr_set */
     0,                    /* tp_dictoffset */
-    encoder_init,         /* tp_init */
+    0,                    /* tp_init */
     0,                    /* tp_alloc */
     encoder_new,          /* tp_new */
     0,                    /* tp_free */
@@ -1954,10 +1910,8 @@ PyInit__json(void)
     PyObject *m = PyModule_Create(&jsonmodule);
     if (!m)
         return NULL;
-    PyScannerType.tp_new = PyType_GenericNew;
     if (PyType_Ready(&PyScannerType) < 0)
         goto fail;
-    PyEncoderType.tp_new = PyType_GenericNew;
     if (PyType_Ready(&PyEncoderType) < 0)
         goto fail;
     Py_INCREF((PyObject*)&PyScannerType);