Unpickler.find_class() can use freed compatibility-mapping values

[Nievesjyl]

Description

With fix_imports enabled for older protocols, Unpickler.find_class() retrieves borrowed values from the public _compat_pickle.NAME_MAPPING and IMPORT_MAPPING dictionaries and uses them without locking or ownership promotion. Concurrent replacement can destroy the tuple, string, or tuple fields before import and attribute lookup.

Observed Behavior

On a free-threaded build, replacing entries with fresh but equal values makes both the name-remapping and import-remapping modes terminate with SIGSEGV in under 0.10 seconds. Both compatibility-GIL controls complete normally with the expected globals.

Affected Version

CPython 3.14.7 at commit 823f0323ee6ec1402088b73bce1a38473cac36dc, using the free-threaded build with PYTHON_GIL=0.

Reproduction

Run both modes:

PYTHON_GIL=0 python3.14t poc/reproduce.py 10 name
PYTHON_GIL=0 python3.14t poc/reproduce.py 10 import

PoC Source Code

poc/reproduce.py:

#!/usr/bin/env python3
"""Race Python-2 global remapping with public compatibility-table updates."""

import _compat_pickle
import pickle
import sys
import threading
import time


MODE = sys.argv[2] if len(sys.argv) > 2 else "name"
if MODE == "name":
    TABLE = _compat_pickle.NAME_MAPPING
    KEY = ("__builtin__", "xrange")
    PAYLOAD = b"\x80\x02c__builtin__\nxrange\n."
    EXPECTED = range
elif MODE == "import":
    TABLE = _compat_pickle.IMPORT_MAPPING
    KEY = "__builtin__"
    PAYLOAD = b"\x80\x02c__builtin__\nlen\n."
    EXPECTED = len
else:
    raise ValueError("mode must be 'name' or 'import'")

ORIGINAL = TABLE[KEY]
STOP = threading.Event()
FAILURES = []


def mutate():
    while not STOP.is_set():
        if MODE == "name":
            TABLE[KEY] = ("".join(("built", "ins")), "".join(("ra", "nge")))
        else:
            TABLE[KEY] = "".join(("built", "ins"))


def load_worker(deadline):
    count = 0
    try:
        while time.monotonic() < deadline and not STOP.is_set():
            result = pickle.loads(PAYLOAD)
            if result is not EXPECTED:
                raise AssertionError(result)
            count += 1
    except BaseException as exc:
        FAILURES.append((type(exc).__name__, repr(exc), count))
        STOP.set()


def main():
    seconds = float(sys.argv[1]) if len(sys.argv) > 1 else 10.0
    deadline = time.monotonic() + seconds
    threads = [threading.Thread(target=mutate)]
    threads += [threading.Thread(target=load_worker, args=(deadline,)) for _ in range(8)]
    for thread in threads:
        thread.start()
    for thread in threads[1:]:
        thread.join()
    STOP.set()
    threads[0].join()
    TABLE[KEY] = ORIGINAL
    print("mode=", MODE, "gil=", sys._is_gil_enabled(), "failures=", FAILURES)
    return bool(FAILURES)


raise SystemExit(main())

[ZeroIntensity]

Please slow down.

[StanFromIreland]

@Nievesjyl please stop submitting issues for a moment.