Repository navigation
Open
Labels
extension-modulesC modules in the Modules dirC modules in the Modules dirtype-crashA hard crash of the interpreter, possibly with a core dumpA hard crash of the interpreter, possibly with a core dump
Description
What happened?
array.array.tofile computes its chunk count before calling the sink’s write method, so it keeps iterating even if the buffer disappears. A reentrant writer can clear the array on the first callback, leaving self->ob_item null, yet the loop still calls PyBytes_FromStringAndSize on that pointer and crashes in the memcpy path.
Proof of Concept:
import array
CHUNK = 64 * 1024
victim = array.array('B', b'\0' * (CHUNK * 2))
class Writer:
armed = True
def write(self, data):
if Writer.armed:
Writer.armed = False
victim.clear()
return 0
victim.tofile(Writer())Affected Versions:
Details
| Python Version | Status | Exit Code |
|---|---|---|
Python 3.9.24+ (heads/3.9:111bbc15b26, Oct 28 2025, 16:51:20) |
Exception | 1 |
Python 3.10.19+ (heads/3.10:014261980b1, Oct 28 2025, 16:52:08) [Clang 18.1.3 (1ubuntu1)] |
Exception | 1 |
Python 3.11.14+ (heads/3.11:88f3f5b5f11, Oct 28 2025, 16:53:08) [Clang 18.1.3 (1ubuntu1)] |
Exception | 1 |
Python 3.12.12+ (heads/3.12:8cb2092bd8c, Oct 28 2025, 16:54:14) [Clang 18.1.3 (1ubuntu1)] |
Exception | 1 |
Python 3.13.9+ (heads/3.13:9c8eade20c6, Oct 28 2025, 16:55:18) [Clang 18.1.3 (1ubuntu1)] |
ASAN | 1 |
Python 3.14.0+ (heads/3.14:2e216728038, Oct 28 2025, 16:56:16) [Clang 18.1.3 (1ubuntu1)] |
ASAN | 1 |
Python 3.15.0a1+ (heads/main:f5394c257ce, Oct 28 2025, 19:29:54) [GCC 13.3.0] |
ASAN | 1 |
Vulnerable Code:
Details
static PyObject *
array_array_tofile_impl(arrayobject *self, PyTypeObject *cls, PyObject *f)
/*[clinic end generated code: output=4560c628d9c18bc2 input=5a24da7a7b407b52]*/
{
Py_ssize_t nbytes = Py_SIZE(self) * self->ob_descr->itemsize;
/* Write 64K blocks at a time */
/* XXX Make the block size settable */
int BLOCKSIZE = 64*1024;
/* Bug: Caches the nblocks at the beginning */
Py_ssize_t nblocks = (nbytes + BLOCKSIZE - 1) / BLOCKSIZE;
Py_ssize_t i;
if (Py_SIZE(self) == 0)
goto done;
array_state *state = get_array_state_by_class(cls);
assert(state != NULL);
for (i = 0; i < nblocks; i++) {
/* self->ob_item has been null out */
char* ptr = self->ob_item + i*BLOCKSIZE;
Py_ssize_t size = BLOCKSIZE;
PyObject *bytes, *res;
if (i*BLOCKSIZE + size > nbytes)
size = nbytes - i*BLOCKSIZE;
bytes = PyBytes_FromStringAndSize(ptr, size);
if (bytes == NULL)
return NULL;
res = PyObject_CallMethodOneArg(f, state->str_write, bytes);
Py_DECREF(bytes);
if (res == NULL)
return NULL;
Py_DECREF(res); /* drop write result */
}
done:
Py_RETURN_NONE;
}Sanitizer Output:
Details
AddressSanitizer:DEADLYSIGNAL
=================================================================
==1299008==ERROR: AddressSanitizer: SEGV on unknown address 0x000000010000 (pc 0x745059388a4d bp 0x7ffeb5f943e0 sp 0x7ffeb5f943b8 T0)
==1299008==The signal is caused by a READ memory access.
#0 0x745059388a4d in __memmove_avx_unaligned_erms ../sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:265
#1 0x5e74a5aadddd in memcpy /usr/include/x86_64-linux-gnu/bits/string_fortified.h:29
#2 0x5e74a5aadddd in PyBytes_FromStringAndSize Objects/bytesobject.c:162
#3 0x745058bca48b in array_array_tofile_impl Modules/arraymodule.c:1611
#4 0x745058bca675 in array_array_tofile Modules/clinic/arraymodule.c.h:472
#5 0x5e74a5ade3c9 in method_vectorcall_FASTCALL_KEYWORDS_METHOD Objects/descrobject.c:381
#6 0x5e74a5abee7f in _PyObject_VectorcallTstate Include/internal/pycore_call.h:169
#7 0x5e74a5abef72 in PyObject_Vectorcall Objects/call.c:327
#8 0x5e74a5d3d056 in _PyEval_EvalFrameDefault Python/generated_cases.c.h:1620
#9 0x5e74a5d80e54 in _PyEval_EvalFrame Include/internal/pycore_ceval.h:121
#10 0x5e74a5d81148 in _PyEval_Vector Python/ceval.c:2001
#11 0x5e74a5d813f8 in PyEval_EvalCode Python/ceval.c:884
#12 0x5e74a5e78507 in run_eval_code_obj Python/pythonrun.c:1365
#13 0x5e74a5e78723 in run_mod Python/pythonrun.c:1459
#14 0x5e74a5e7957a in pyrun_file Python/pythonrun.c:1293
#15 0x5e74a5e7c220 in _PyRun_SimpleFileObject Python/pythonrun.c:521
#16 0x5e74a5e7c4f6 in _PyRun_AnyFileObject Python/pythonrun.c:81
#17 0x5e74a5ecd74d in pymain_run_file_obj Modules/main.c:410
#18 0x5e74a5ecd9b4 in pymain_run_file Modules/main.c:429
#19 0x5e74a5ecf1b2 in pymain_run_python Modules/main.c:691
#20 0x5e74a5ecf842 in Py_RunMain Modules/main.c:772
#21 0x5e74a5ecfa2e in pymain_main Modules/main.c:802
#22 0x5e74a5ecfdb3 in Py_BytesMain Modules/main.c:826
#23 0x5e74a5953645 in main Programs/python.c:15
#24 0x74505922a1c9 in __libc_start_call_main ../sysdeps/nptl/libc_start_call_main.h:58
#25 0x74505922a28a in __libc_start_main_impl ../csu/libc-start.c:360
#26 0x5e74a5953574 in _start (/home/jackfromeast/Desktop/entropy/targets/grammar-afl++-latest/targets/cpython/python+0x2dd574) (BuildId: 202d5dbb945f6d5f5a66ad50e2688d56affd6ecb)
AddressSanitizer can not provide additional info.
SUMMARY: AddressSanitizer: SEGV ../sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:265 in __memmove_avx_unaligned_erms
==1299008==ABORTING
Linked PRs
Metadata
Metadata
Assignees
Labels
extension-modulesC modules in the Modules dirC modules in the Modules dirtype-crashA hard crash of the interpreter, possibly with a core dumpA hard crash of the interpreter, possibly with a core dump
Projects
- StatusShow more project fieldsTodo
Activity
pythongh-142884: Use-After-Free Vulnerability Fixed in CPython array.…
pythongh-142884: Use-After-Free Vulnerability Fixed in CPython array.…
pythongh-142884: Use-After-Free Vulnerability Fixed in CPython array.…
serhiy-storchaka commentedon Jan 9, 2026
The simplest way to fix this is perhaps to temporary increase the exported buffer counter.
Today, it is more efficient to write using
f.write(arr)thanarr.tofile(f). The latter is only needed if your file-like object does not support the buffer protocol, which is uncommon.array.tofile()crashes due to use-after-free whenwrite()callback modifies the array #144922 as a duplicate of this issuemjbommar commentedon Apr 16, 2026
Found this fuzzing and saw the other dupes marked, but it seems like a PR was never accepted and 3.15 is still impacted.
@serhiy-storchaka and @jackfromeast , happy to help get an acceptable fix over the line if that would be useful.
My only concern is that it seems like there might be disagreement between @picnixz and @serhiy-storchaka about what the preferred fix and result should be. Do we want to take the extra Py_SIZE calls inside the loop or just bump ob_exports and throw BufferError?