Skip to content

gh-109496: Detect Py_DECREF() after dealloc in debug mode #109539

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Sep 18, 2023
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 4 additions & 6 deletions Include/object.h
Original file line number Diff line number Diff line change
Expand Up @@ -660,17 +660,15 @@ static inline void Py_DECREF(PyObject *op) {
#elif defined(Py_REF_DEBUG)
static inline void Py_DECREF(const char *filename, int lineno, PyObject *op)
{
if (op->ob_refcnt <= 0) {
_Py_NegativeRefcount(filename, lineno, op);
}
if (_Py_IsImmortal(op)) {
return;
}
_Py_DECREF_STAT_INC();
_Py_DECREF_DecRefTotal();
if (--op->ob_refcnt != 0) {
if (op->ob_refcnt < 0) {
_Py_NegativeRefcount(filename, lineno, op);
}
}
else {
if (--op->ob_refcnt == 0) {
_Py_Dealloc(op);
}
}
Expand Down
36 changes: 26 additions & 10 deletions Lib/test/test_capi/test_misc.py
Original file line number Diff line number Diff line change
Expand Up @@ -301,24 +301,40 @@ def test_getitem_with_error(self):
def test_buildvalue_N(self):
_testcapi.test_buildvalue_N()

@unittest.skipUnless(hasattr(_testcapi, 'negative_refcount'),
'need _testcapi.negative_refcount')
def test_negative_refcount(self):
def check_negative_refcount(self, code):
# bpo-35059: Check that Py_DECREF() reports the correct filename
# when calling _Py_NegativeRefcount() to abort Python.
code = textwrap.dedent("""
import _testcapi
from test import support

with support.SuppressCrashReport():
_testcapi.negative_refcount()
""")
code = textwrap.dedent(code)
rc, out, err = assert_python_failure('-c', code)
self.assertRegex(err,
br'_testcapimodule\.c:[0-9]+: '
br'_Py_NegativeRefcount: Assertion failed: '
br'object has negative ref count')

@unittest.skipUnless(hasattr(_testcapi, 'negative_refcount'),
'need _testcapi.negative_refcount()')
def test_negative_refcount(self):
code = """
import _testcapi
from test import support

with support.SuppressCrashReport():
_testcapi.negative_refcount()
"""
self.check_negative_refcount(code)

@unittest.skipUnless(hasattr(_testcapi, 'decref_freed_object'),
'need _testcapi.decref_freed_object()')
def test_decref_freed_object(self):
code = """
import _testcapi
from test import support

with support.SuppressCrashReport():
_testcapi.decref_freed_object()
"""
self.check_negative_refcount(code)

def test_trashcan_subclass(self):
# bpo-35983: Check that the trashcan mechanism for "list" is NOT
# activated when its tp_dealloc is being called by a subclass
Expand Down
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
On a Python built in debug mode, :c:func:`Py_DECREF()` now calls
``_Py_NegativeRefcount()`` if the object is a dangling pointer to
deallocated memory: memory filled with ``0xDD`` "dead byte" by the debug
hook on memory allocators. The fix is to check the reference count *before*
checking for ``_Py_IsImmortal()``. Patch by Victor Stinner.
21 changes: 21 additions & 0 deletions Modules/_testcapimodule.c
Original file line number Diff line number Diff line change
Expand Up @@ -2034,6 +2034,26 @@ negative_refcount(PyObject *self, PyObject *Py_UNUSED(args))

Py_RETURN_NONE;
}

static PyObject *
decref_freed_object(PyObject *self, PyObject *Py_UNUSED(args))
{
PyObject *obj = PyUnicode_FromString("decref_freed_object");
if (obj == NULL) {
return NULL;
}
assert(Py_REFCNT(obj) == 1);

// Deallocate the memory
Py_DECREF(obj);
// obj is a now a dangling pointer
Comment on lines +2048 to +2049
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a chance this could be an actual pointer to deallocated memory, or memory written to by another thread? Would it be safer to just call Py_DECREF on a fake object which is explicitly memset to 0xDD bytes?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I prefer to use a functional test: not mock anything here.

This test is very fragile, it makes sure that a dangling pointer has a determinstic behavior. Problem: by definition, it has an undefined behavior :-) But in practice, almost always, the memory should remain around and should just be filled by 0xDD bytes.

If tomorrow the test starts to fail randomly, sure, we can rewrite it by "mocking" the first DECREF() and just pass memory filled by 0xDD byte pattern.

test_capi should not be run in a process with more than 1 thread. libregrtest works hard to make sure that previous tests don't leak tests. Also, it's strongly recommended to run tests with -jN: each test file is run in a fresh process, so with a known number of threads: just 1 :-)


// gh-109496: If Python is built in debug mode, Py_DECREF() must call
// _Py_NegativeRefcount() and abort Python.
Py_DECREF(obj);

Py_RETURN_NONE;
}
#endif


Expand Down Expand Up @@ -3299,6 +3319,7 @@ static PyMethodDef TestMethods[] = {
{"bad_get", _PyCFunction_CAST(bad_get), METH_FASTCALL},
#ifdef Py_REF_DEBUG
{"negative_refcount", negative_refcount, METH_NOARGS},
{"decref_freed_object", decref_freed_object, METH_NOARGS},
#endif
{"meth_varargs", meth_varargs, METH_VARARGS},
{"meth_varargs_keywords", _PyCFunction_CAST(meth_varargs_keywords), METH_VARARGS|METH_KEYWORDS},
Expand Down