[3.14] gh-90949: expose Expat API to tune exponential expansion protections (GH-139368)

Expose the XML Expat 2.7.2 APIs to tune protections against
"billion laughs" [1] attacks.

The exposed APIs are available on Expat parsers, that is,
parsers created by `xml.parsers.expat.ParserCreate()`, as:

- `parser.SetBillionLaughsAttackProtectionActivationThreshold(threshold)`, and
- `parser.SetBillionLaughsAttackProtectionMaximumAmplification(max_factor)`.

This completes the work in f04bea4,
and improves the existing related documentation.

[1]: https://en.wikipedia.org/wiki/Billion_laughs_attack
(cherry picked from commit 6661123)

Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>

Author: picnixz

Doc/library/pyexpat.rst

@@ -238,16 +238,71 @@ XMLParser Objects
    .. versionadded:: 3.13
 
 
-:class:`!xmlparser` objects have the following methods to mitigate some
-common XML vulnerabilities.
+:class:`!xmlparser` objects have the following methods to tune protections
+against some common XML vulnerabilities.
+
+.. method:: xmlparser.SetBillionLaughsAttackProtectionActivationThreshold(threshold, /)
+
+   Sets the number of output bytes needed to activate protection against
+   `billion laughs`_ attacks.
+
+   The number of output bytes includes amplification from entity expansion
+   and reading DTD files.
+
+   Parser objects usually have a protection activation threshold of 8 MiB,
+   but the actual default value depends on the underlying Expat library.
+
+   An :exc:`ExpatError` is raised if this method is called on a
+   |xml-non-root-parser| parser.
+   The corresponding :attr:`~ExpatError.lineno` and :attr:`~ExpatError.offset`
+   should not be used as they may have no special meaning.
+
+   .. note::
+
+      Activation thresholds below 4 MiB are known to break support for DITA 1.3
+      payload and are hence not recommended.
+
+   .. versionadded:: next
+
+.. method:: xmlparser.SetBillionLaughsAttackProtectionMaximumAmplification(max_factor, /)
+
+   Sets the maximum tolerated amplification factor for protection against
+   `billion laughs`_ attacks.
+
+   The amplification factor is calculated as ``(direct + indirect) / direct``
+   while parsing, where ``direct`` is the number of bytes read from
+   the primary document in parsing and ``indirect`` is the number of
+   bytes added by expanding entities and reading of external DTD files.
+
+   The *max_factor* value must be a non-NaN :class:`float` value greater than
+   or equal to 1.0. Peak amplifications of factor 15,000 for the entire payload
+   and of factor 30,000 in the middle of parsing have been observed with small
+   benign files in practice. In particular, the activation threshold should be
+   carefully chosen to avoid false positives.
+
+   Parser objects usually have a maximum amplification factor of 100,
+   but the actual default value depends on the underlying Expat library.
+
+   An :exc:`ExpatError` is raised if this method is called on a
+   |xml-non-root-parser| parser or if *max_factor* is outside the valid range.
+   The corresponding :attr:`~ExpatError.lineno` and :attr:`~ExpatError.offset`
+   should not be used as they may have no special meaning.
+
+   .. note::
+
+      The maximum amplification factor is only considered if the threshold
+      that can be adjusted by :meth:`.SetBillionLaughsAttackProtectionActivationThreshold`
+      is exceeded.
+
+   .. versionadded:: next
 
 .. method:: xmlparser.SetAllocTrackerActivationThreshold(threshold, /)
 
    Sets the number of allocated bytes of dynamic memory needed to activate
    protection against disproportionate use of RAM.
 
-   By default, parser objects have an allocation activation threshold of 64 MiB,
-   or equivalently 67,108,864 bytes.
+   Parser objects usually have an allocation activation threshold of 64 MiB,
+   but the actual default value depends on the underlying Expat library.
 
    An :exc:`ExpatError` is raised if this method is called on a
    |xml-non-root-parser| parser.
@@ -271,7 +326,8 @@ common XML vulnerabilities.
    near the start of parsing even with benign files in practice. In particular,
    the activation threshold should be carefully chosen to avoid false positives.
 
-   By default, parser objects have a maximum amplification factor of 100.0.
+   Parser objects usually have a maximum amplification factor of 100,
+   but the actual default value depends on the underlying Expat library.
 
    An :exc:`ExpatError` is raised if this method is called on a
    |xml-non-root-parser| parser or if *max_factor* is outside the valid range.
@@ -1019,4 +1075,6 @@ The ``errors`` module has the following attributes:
    not. See https://www.w3.org/TR/2006/REC-xml11-20060816/#NT-EncodingDecl
    and https://www.iana.org/assignments/character-sets/character-sets.xhtml.
 
+
+.. _billion laughs: https://en.wikipedia.org/wiki/Billion_laughs_attack
 .. |xml-non-root-parser| replace:: :ref:`non-root <xmlparser-non-root>`

Include/pyexpat.h

@@ -57,6 +57,11 @@ struct PyExpat_CAPI
         XML_Parser parser, unsigned long long activationThresholdBytes);
     XML_Bool (*SetAllocTrackerMaximumAmplification)(
         XML_Parser parser, float maxAmplificationFactor);
+    /* might be NULL for expat < 2.4.0 */
+    XML_Bool (*SetBillionLaughsAttackProtectionActivationThreshold)(
+        XML_Parser parser, unsigned long long activationThresholdBytes);
+    XML_Bool (*SetBillionLaughsAttackProtectionMaximumAmplification)(
+        XML_Parser parser, float maxAmplificationFactor);
     /* always add new stuff to the end! */
 };
 

Lib/test/test_pyexpat.py

@@ -1183,6 +1183,64 @@ def test_set_maximum_amplification__fail_for_subparser(self):
         self.assert_root_parser_failure(setter, 123.45)
 
 
+@unittest.skipIf(expat.version_info < (2, 4, 0), "requires Expat >= 2.4.0")
+class ExpansionProtectionTest(AttackProtectionTestBase, unittest.TestCase):
+
+    def assert_rejected(self, func, /, *args, **kwargs):
+        """Check that func(*args, **kwargs) hits the allocation limit."""
+        msg = (
+            r"limit on input amplification factor \(from DTD and entities\) "
+            r"breached: line \d+, column \d+"
+        )
+        self.assertRaisesRegex(expat.ExpatError, msg, func, *args, **kwargs)
+
+    def set_activation_threshold(self, parser, threshold):
+        return parser.SetBillionLaughsAttackProtectionActivationThreshold(threshold)
+
+    def set_maximum_amplification(self, parser, max_factor):
+        return parser.SetBillionLaughsAttackProtectionMaximumAmplification(max_factor)
+
+    def test_set_activation_threshold__threshold_reached(self):
+        parser = expat.ParserCreate()
+        # Choose a threshold expected to be always reached.
+        self.set_activation_threshold(parser, 3)
+        # Check that the threshold is reached by choosing a small factor
+        # and a payload whose peak amplification factor exceeds it.
+        self.assertIsNone(self.set_maximum_amplification(parser, 1.0))
+        payload = self.exponential_expansion_payload(ncols=10, nrows=4)
+        self.assert_rejected(parser.Parse, payload, True)
+
+    def test_set_activation_threshold__threshold_not_reached(self):
+        parser = expat.ParserCreate()
+        # Choose a threshold expected to be never reached.
+        self.set_activation_threshold(parser, pow(10, 5))
+        # Check that the threshold is reached by choosing a small factor
+        # and a payload whose peak amplification factor exceeds it.
+        self.assertIsNone(self.set_maximum_amplification(parser, 1.0))
+        payload = self.exponential_expansion_payload(ncols=10, nrows=4)
+        self.assertIsNotNone(parser.Parse(payload, True))
+
+    def test_set_maximum_amplification__amplification_exceeded(self):
+        parser = expat.ParserCreate()
+        # Unconditionally enable maximum activation factor.
+        self.set_activation_threshold(parser, 0)
+        # Choose a max amplification factor expected to always be exceeded.
+        self.assertIsNone(self.set_maximum_amplification(parser, 1.0))
+        # Craft a payload for which the peak amplification factor is > 1.0.
+        payload = self.exponential_expansion_payload(ncols=1, nrows=2)
+        self.assert_rejected(parser.Parse, payload, True)
+
+    def test_set_maximum_amplification__amplification_not_exceeded(self):
+        parser = expat.ParserCreate()
+        # Unconditionally enable maximum activation factor.
+        self.set_activation_threshold(parser, 0)
+        # Choose a max amplification factor expected to never be exceeded.
+        self.assertIsNone(self.set_maximum_amplification(parser, 1e4))
+        # Craft a payload for which the peak amplification factor is < 1e4.
+        payload = self.exponential_expansion_payload(ncols=1, nrows=2)
+        self.assertIsNotNone(parser.Parse(payload, True))
+
+
 @unittest.skipIf(not hasattr(expat.XMLParserType,
                              "SetAllocTrackerMaximumAmplification"),
                  "requires Python compiled with Expat >= 2.7.2")

Misc/NEWS.d/3.14.1.rst

@@ -749,9 +749,9 @@ fix: paste zero-width in default repl width is wrong.
 
 Add :meth:`~xml.parsers.expat.xmlparser.SetAllocTrackerActivationThreshold`
 and :meth:`~xml.parsers.expat.xmlparser.SetAllocTrackerMaximumAmplification`
-to :ref:`xmlparser <xmlparser-objects>` objects to prevent use of
-disproportional amounts of dynamic memory from within an Expat parser. Patch
-by Bénédikt Tran.
+to :ref:`xmlparser <xmlparser-objects>` objects to tune protections against
+disproportional amounts of dynamic memory usage from within an Expat parser.
+Patch by Bénédikt Tran.
 
 ..
 

Misc/NEWS.d/next/Library/2025-09-26-18-04-28.gh-issue-90949.YHjSzX.rst

@@ -0,0 +1,7 @@
+Add
+:meth:`~xml.parsers.expat.xmlparser.SetBillionLaughsAttackProtectionActivationThreshold`
+and
+:meth:`~xml.parsers.expat.xmlparser.SetBillionLaughsAttackProtectionMaximumAmplification`
+to :ref:`xmlparser <xmlparser-objects>` objects to tune protections against
+`billion laughs <https://en.wikipedia.org/wiki/Billion_laughs_attack>`_ attacks.
+Patch by Bénédikt Tran.