PYTHON-5923 Add remaining-buffer bound check in _array_of_documents_to_buffer (#2872)

Signed-off-by: FugoP <264910004+AgentGymLeader@users.noreply.github.com>

Author: AgentGymLeader

bson/_cbsonmodule.c

@@ -3238,6 +3238,15 @@ static PyObject* _cbson_array_of_documents_to_buffer(PyObject* self, PyObject* a
             goto fail;
          }
 
+        if (value_length >= (uint32_t)(size - position)) {
+            PyObject* InvalidBSON = _error("InvalidBSON");
+            if (InvalidBSON) {
+                PyErr_SetString(InvalidBSON, "invalid array content");
+                Py_DECREF(InvalidBSON);
+            }
+            goto fail;
+        }
+
         if (pymongo_buffer_write(buffer, string + position, value_length) == 1) {
             goto fail;
         }

doc/changelog.rst

@@ -17,6 +17,11 @@ Changes in Version 4.18.0
   attempts, so consumers can correlate a retried operation's events. As a
   result, ``operation_id`` is no longer equal to the per-attempt ``request_id``
   for these operations.
+- Fixed a potential out-of-bounds read in the C extension when decoding an
+  array of BSON documents. An embedded document whose declared length exceeds
+  the bytes remaining in the array now raises
+  :class:`~bson.errors.InvalidBSON` instead of reading past the end of the
+  buffer.
 
 Changes in Version 4.17.0 (2026/04/20)
 --------------------------------------

test/test_bson.py

@@ -1758,6 +1758,45 @@ def test_array_of_documents_to_buffer(self):
         with self.assertRaises(InvalidBSON):
             _array_of_documents_to_buffer(buf)
 
+    def test_array_of_documents_to_buffer_rejects_oversized_element(self):
+        # Regression test for the bound check in
+        # _cbson_array_of_documents_to_buffer: an embedded document whose
+        # declared length exceeds the bytes remaining in the array document
+        # must raise InvalidBSON before any copy, rather than reading out of
+        # bounds (CWE-125).
+        doc = dict(a=1)
+        valid = encode({"0": doc})
+        # A well-formed buffer is unaffected by the guard.
+        self.assertEqual(_array_of_documents_to_buffer(valid), encode(doc))
+        # The first embedded document starts after the array header:
+        # 4-byte array length + 1-byte element type (0x03) + "0\x00" key = 7.
+        # Its leading 4 bytes are the embedded document's declared length.
+        offset = 4 + 1 + 2
+        malformed = bytearray(valid)
+        # Claim a length far larger than what remains (but >= BSON_MIN_SIZE so
+        # the existing lower-bound check still passes), forcing the new
+        # upper-bound guard to reject it.
+        malformed[offset : offset + 4] = struct.pack("<i", 0x7FFF)
+        # The C fast-path raises a dedicated "invalid array content" error from the
+        # new bound check; the pure-Python path reaches the same invariant through
+        # _get_object_size, which raises "invalid object length".
+        expected = "invalid array content" if bson.has_c() else "invalid object length"
+        with self.assertRaisesRegex(InvalidBSON, expected):
+            _array_of_documents_to_buffer(bytes(malformed))
+
+    def test_array_of_documents_to_buffer_rejects_element_consuming_terminator(self):
+        doc = dict(a=1)
+        valid = encode({"0": doc})
+        self.assertEqual(_array_of_documents_to_buffer(valid), encode(doc))
+        offset = 4 + 1 + 2
+        malformed = bytearray(valid)
+        value_length = len(valid) - offset
+        malformed[offset : offset + 4] = struct.pack("<i", value_length)
+        # Covers the exact boundary where the embedded doc consumes the array's EOO byte.
+        expected = "invalid array content" if bson.has_c() else "invalid object length"
+        with self.assertRaisesRegex(InvalidBSON, expected):
+            _array_of_documents_to_buffer(bytes(malformed))
+
     def test_datetime_ms_hash(self):
         # Equal values must have equal hashes.
         self.assertEqual(hash(DatetimeMS(0)), hash(DatetimeMS(0)))