gh-150599: Prevent bz2 decompressor reuse after errors (#150600)

Author: StanFromIreland

Lib/test/test_bz2.py

@@ -1032,6 +1032,21 @@ def test_failure(self):
         # Previously, a second call could crash due to internal inconsistency
         self.assertRaises(Exception, bzd.decompress, self.BAD_DATA * 30)
 
+    def test_decompress_after_data_error(self):
+        data = bytes.fromhex(
+            "425a6839314159265359000000000000007fffff000000000000000000000000"
+            "00000000000000000000000000000000000000e0370000000000000000000000"
+            "000000000000000000000000000000000000000000000000000083f3"
+        )
+        bzd = BZ2Decompressor()
+        with self.assertRaisesRegex(OSError, "Invalid data stream"):
+            bzd.decompress(data)
+        # Previously, a second call could crash due to internal inconsistency
+        self.assertFalse(bzd.needs_input)
+        self.assertFalse(bzd.eof)
+        with self.assertRaisesRegex(ValueError, "previous error"):
+            bzd.decompress(b'\x00' * 18)
+
     @support.refcount_test
     def test_refleaks_in___init__(self):
         gettotalrefcount = support.get_attribute(sys, 'gettotalrefcount')

Misc/NEWS.d/next/Security/2026-05-30-09-36-20.gh-issue-150599.nlHqU-.rst

@@ -0,0 +1,3 @@
+Fix a possible stack buffer overflow in :mod:`bz2` when a
+:class:`bz2.BZ2Decompressor` is reused after a decompression error.
+The decompressor now becomes unusable after libbz2 reports an error.

Modules/_bz2module.c

@@ -108,6 +108,7 @@ typedef struct {
 typedef struct {
     PyObject_HEAD
     bz_stream bzs;
+    int bzerror;
     char eof;           /* Py_T_BOOL expects a char */
     PyObject *unused_data;
     char needs_input;
@@ -435,8 +436,11 @@ decompress_buf(BZ2Decompressor *d, Py_ssize_t max_length)
 
         d->bzs_avail_in_real += bzs->avail_in;
 
-        if (catch_bz2_error(bzret))
+        if (catch_bz2_error(bzret)) {
+            d->bzerror = bzret;
+            FT_ATOMIC_STORE_CHAR_RELAXED(d->needs_input, 0);
             goto error;
+        }
         if (bzret == BZ_STREAM_END) {
             FT_ATOMIC_STORE_CHAR_RELAXED(d->eof, 1);
             break;
@@ -607,10 +611,17 @@ _bz2_BZ2Decompressor_decompress_impl(BZ2Decompressor *self, Py_buffer *data,
     PyObject *result = NULL;
 
     PyMutex_Lock(&self->mutex);
-    if (self->eof)
+    if (self->eof) {
         PyErr_SetString(PyExc_EOFError, "End of stream already reached");
-    else
+    }
+    else if (self->bzerror) {
+        // Re-entering BZ2_bzDecompress() after an error can write out of bounds.
+        PyErr_SetString(PyExc_ValueError,
+                        "Decompressor is unusable after a previous error");
+    }
+    else {
         result = decompress(self, data->buf, data->len, max_length);
+    }
     PyMutex_Unlock(&self->mutex);
     return result;
 }
@@ -638,6 +649,7 @@ _bz2_BZ2Decompressor_impl(PyTypeObject *type)
     }
 
     self->mutex = (PyMutex){0};
+    self->bzerror = 0;
     self->needs_input = 1;
     self->bzs_avail_in_real = 0;
     self->input_buffer = NULL;