Merge pull request #5 from ActiveState/1.28.6-sec-reader-outline-xmp

Security (5/5): outline cycle + XMP entity expansion — CVE-2026-24688, CVE-2026-40260

Author: icanhasmath

PyPDF2/_reader.py

@@ -730,7 +730,7 @@ def outlines(self):
         """Read-only property."""
         return self._get_outlines()
 
-    def _get_outlines(self, node=None, outlines=None):
+    def _get_outlines(self, node=None, outlines=None, visited=None):
         """
         Retrieve the document outline present in the document.
 
@@ -758,16 +758,27 @@ def _get_outlines(self, node=None, outlines=None):
         if node is None:
             return outlines
 
+        if visited is None:
+            visited = set()
+
         # see if there are any more outlines
         while True:
+            # CVE-2026-24688: a crafted outline whose /Next chain loops back to
+            # an already-seen node made this walk run forever. Stop on a repeat.
+            node_id = id(node)
+            if node_id in visited:
+                warnings.warn("Cycle detected in document outline; stopping")
+                break
+            visited.add(node_id)
+
             outline = self._build_outline(node)
             if outline:
                 outlines.append(outline)
 
             # check for sub-outlines
             if "/First" in node:
                 sub_outlines = []
-                self._get_outlines(node["/First"], sub_outlines)
+                self._get_outlines(node["/First"], sub_outlines, visited.copy())
                 if sub_outlines:
                     outlines.append(sub_outlines)
 

PyPDF2/xmp.py

@@ -2,11 +2,32 @@
 import decimal
 import re
 import warnings
-from xml.dom.minidom import parseString
+from xml.dom.expatbuilder import ExpatBuilderNS
+from xml.parsers.expat import ExpatError
 
 from ._utils import DEPR_MSG, u_
 from .generic import PdfObject
 
+
+class _SafeExpatBuilder(ExpatBuilderNS):
+    """Namespace-aware XML builder that refuses all entity declarations.
+
+    CVE-2026-40260: parsing untrusted XMP with the default parser allowed XML
+    entity expansion (quadratic blow-up, which libexpat does not block) to
+    exhaust memory. Rejecting any entity declaration outright closes that.
+    """
+
+    def _deny_entity(self, *args):
+        raise ExpatError("Entity declarations are not permitted in XMP")
+
+    def install(self, parser):
+        ExpatBuilderNS.install(self, parser)
+        parser.EntityDeclHandler = self._deny_entity
+
+
+def _safe_parse_xmp(data):
+    return _SafeExpatBuilder().parseString(data)
+
 RDF_NAMESPACE = "http://www.w3.org/1999/02/22-rdf-syntax-ns#"
 DC_NAMESPACE = "http://purl.org/dc/elements/1.1/"
 XMP_NAMESPACE = "http://ns.adobe.com/xap/1.0/"
@@ -88,7 +109,7 @@ class XmpInformation(PdfObject):
 
     def __init__(self, stream):
         self.stream = stream
-        doc_root = parseString(self.stream.get_data())
+        doc_root = _safe_parse_xmp(self.stream.get_data())
         self.rdfRoot = doc_root.getElementsByTagNameNS(RDF_NAMESPACE, "RDF")[0]
         self.cache = {}
 

Tests/test_security_outline.py

@@ -0,0 +1,19 @@
+# -*- coding: utf-8 -*-
+"""Regression test for CVE-2026-24688: cyclic document outline /Next loop."""
+import warnings
+
+from PyPDF2._reader import PdfReader
+from PyPDF2.generic import DictionaryObject, NameObject
+
+
+def test_get_outlines_cycle_terminates():
+    reader = PdfReader.__new__(PdfReader)
+    reader._build_outline = lambda node: None  # isolate the walk
+    a = DictionaryObject()
+    b = DictionaryObject()
+    a[NameObject("/Next")] = b
+    b[NameObject("/Next")] = a  # cycle that never ends
+    with warnings.catch_warnings():
+        warnings.simplefilter("ignore")
+        result = reader._get_outlines(a, [])  # must terminate, not hang
+    assert result == []

Tests/test_security_xmp.py

@@ -0,0 +1,40 @@
+# -*- coding: utf-8 -*-
+"""Regression test for CVE-2026-40260: XMP XML entity expansion."""
+import pytest
+
+from xml.parsers.expat import ExpatError
+
+from PyPDF2.xmp import XmpInformation
+
+
+class _FakeStream(object):
+    def __init__(self, data):
+        self._data = data
+
+    def get_data(self):
+        return self._data
+
+
+_RDF = (
+    b'<rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">'
+    b"<rdf:Description/></rdf:RDF>"
+)
+
+
+def test_xmp_normal_parses():
+    xml = (
+        b'<?xml version="1.0"?>'
+        b'<x:xmpmeta xmlns:x="adobe:ns:meta/">' + _RDF + b"</x:xmpmeta>"
+    )
+    xmp = XmpInformation(_FakeStream(xml))
+    assert xmp.rdfRoot is not None
+
+
+def test_xmp_entity_declaration_rejected():
+    xml = (
+        b'<?xml version="1.0"?>'
+        b'<!DOCTYPE x [ <!ENTITY boom "AAAA"> ]>'
+        b'<x:xmpmeta xmlns:x="adobe:ns:meta/">' + _RDF + b"</x:xmpmeta>"
+    )
+    with pytest.raises(ExpatError):
+        XmpInformation(_FakeStream(xml))