Skip to content

[3.13] gh-124785: alternative interned string clear fix #124796

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

Closed
wants to merge 2 commits into from
Closed
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
4 changes: 4 additions & 0 deletions Include/internal/pycore_interp.h
Original file line number Diff line number Diff line change
Expand Up @@ -123,6 +123,10 @@ struct _is {
/* Has been fully initialized via pylifecycle.c. */
int _ready;
int finalizing;
/* Set if a subinterpreter imports basic single-phase extensions and
* therefore needs to leak interned strings (it's too tricky to safely
* free them). */
int leak_interned_strings;

uintptr_t last_restart_version;
struct pythreads {
Expand Down

This file was deleted.

Original file line number Diff line number Diff line change
@@ -0,0 +1,2 @@
Use simpler fix for interned string cleanup. This avoids the potential
use-after-free crashes but doesn't break the trace-refs debug build.
83 changes: 44 additions & 39 deletions Objects/unicodeobject.c
Original file line number Diff line number Diff line change
Expand Up @@ -277,37 +277,13 @@ hashtable_unicode_compare(const void *key1, const void *key2)
}
}

/* Return true if this interpreter should share the main interpreter's
intern_dict. That's important for interpreters which load basic
single-phase init extension modules (m_size == -1). There could be interned
immortal strings that are shared between interpreters, due to the
PyDict_Update(mdict, m_copy) call in import_find_extension().

It's not safe to deallocate those strings until all interpreters that
potentially use them are freed. By storing them in the main interpreter, we
ensure they get freed after all other interpreters are freed.
*/
static bool
has_shared_intern_dict(PyInterpreterState *interp)
{
PyInterpreterState *main_interp = _PyInterpreterState_Main();
return interp != main_interp && interp->feature_flags & Py_RTFLAGS_USE_MAIN_OBMALLOC;
}

static int
init_interned_dict(PyInterpreterState *interp)
{
assert(get_interned_dict(interp) == NULL);
PyObject *interned;
if (has_shared_intern_dict(interp)) {
interned = get_interned_dict(_PyInterpreterState_Main());
Py_INCREF(interned);
}
else {
interned = PyDict_New();
if (interned == NULL) {
return -1;
}
PyObject *interned = interned = PyDict_New();
if (interned == NULL) {
return -1;
}
_Py_INTERP_CACHED_OBJECT(interp, interned_strings) = interned;
return 0;
Expand All @@ -318,10 +294,7 @@ clear_interned_dict(PyInterpreterState *interp)
{
PyObject *interned = get_interned_dict(interp);
if (interned != NULL) {
if (!has_shared_intern_dict(interp)) {
// only clear if the dict belongs to this interpreter
PyDict_Clear(interned);
}
PyDict_Clear(interned);
Py_DECREF(interned);
_Py_INTERP_CACHED_OBJECT(interp, interned_strings) = NULL;
}
Expand Down Expand Up @@ -15323,6 +15296,42 @@ PyUnicode_InternFromString(const char *cp)
return s;
}

void
clear_interned_dict_legacy(PyInterpreterState *interp)
{
// See GH-116510 for why this extra care is needed. Immortal interned
// strings can be shared between subinterpreters in this case and we
// can't safely free them without a potential use-after-free crash.
PyObject *interned = get_interned_dict(interp);
Py_ssize_t pos = 0;
PyObject *s, *ignored_value;
while (PyDict_Next(interned, &pos, &s, &ignored_value)) {
assert(PyUnicode_IS_READY(s));
switch (PyUnicode_CHECK_INTERNED(s)) {
case SSTATE_INTERNED_IMMORTAL:
case SSTATE_INTERNED_IMMORTAL_STATIC:
// immortal strings leak since we can't safely free them
break;
case SSTATE_INTERNED_MORTAL:
// Restore 2 references held by the interned dict; these will
// be decref'd by clear_interned_dict's PyDict_Clear.
Py_SET_REFCNT(s, Py_REFCNT(s) + 2);
#ifdef Py_REF_DEBUG
/* let's be pedantic with the ref total */
_Py_IncRefTotal(_PyThreadState_GET());
_Py_IncRefTotal(_PyThreadState_GET());
#endif
break;
case SSTATE_NOT_INTERNED:
/* fall through */
default:
Py_UNREACHABLE();
}
_PyUnicode_STATE(s).interned = SSTATE_NOT_INTERNED;
}
PyDict_Clear(interned);
_Py_INTERP_CACHED_OBJECT(interp, interned_strings) = NULL;
}

void
_PyUnicode_ClearInterned(PyInterpreterState *interp)
Expand All @@ -15333,10 +15342,8 @@ _PyUnicode_ClearInterned(PyInterpreterState *interp)
}
assert(PyDict_CheckExact(interned));

if (has_shared_intern_dict(interp)) {
// the dict doesn't belong to this interpreter, skip the debug
// checks on it and just clear the pointer to it
clear_interned_dict(interp);
if (interp->leak_interned_strings) {
clear_interned_dict_legacy(interp);
return;
}

Expand Down Expand Up @@ -15861,10 +15868,8 @@ _PyUnicode_Fini(PyInterpreterState *interp)
{
struct _Py_unicode_state *state = &interp->unicode;

if (!has_shared_intern_dict(interp)) {
// _PyUnicode_ClearInterned() must be called before _PyUnicode_Fini()
assert(get_interned_dict(interp) == NULL);
}
// _PyUnicode_ClearInterned() must be called before _PyUnicode_Fini()
assert(get_interned_dict(interp) == NULL);

_PyUnicode_FiniEncodings(&state->fs_codec);

Expand Down
5 changes: 5 additions & 0 deletions Python/import.c
Original file line number Diff line number Diff line change
Expand Up @@ -1897,6 +1897,11 @@ import_find_extension(PyThreadState *tstate,
return NULL;
}

/* Interned strings used by this module can be shared between
* subinterpreters, due to the PyDict_Update() call in basic single-phase
* module import case. */
tstate->interp->leak_interned_strings = 1;

PyObject *mod = reload_singlephase_extension(tstate, cached, info);
if (mod == NULL) {
return NULL;
Expand Down
Loading