[5.1.x] Fixed CVE-2024-53907 -- Mitigated potential DoS in strip_tags().

sarahboyce · sarahboyce · commit bbc74a7f7eb7 · 2024-12-04T13:47:21.000+01:00
Thanks to jiangniao for the report, and Shai Berger and Natalia Bidart
for the reviews.
diff --git a/django/utils/html.py b/django/utils/html.py
@@ -7,6 +7,7 @@
 from html.parser import HTMLParser
 from urllib.parse import parse_qsl, quote, unquote, urlencode, urlsplit, urlunsplit
 
+from django.core.exceptions import SuspiciousOperation
 from django.utils.deprecation import RemovedInDjango60Warning
 from django.utils.encoding import punycode
 from django.utils.functional import Promise, cached_property, keep_lazy, keep_lazy_text
@@ -39,6 +40,7 @@
 )
 
 MAX_URL_LENGTH = 2048
+MAX_STRIP_TAGS_DEPTH = 50
 
 
 @keep_lazy(SafeString)
@@ -205,15 +207,19 @@ def _strip_once(value):
 @keep_lazy_text
 def strip_tags(value):
     """Return the given HTML with all tags stripped."""
-    # Note: in typical case this loop executes _strip_once once. Loop condition
-    # is redundant, but helps to reduce number of executions of _strip_once.
     value = str(value)
+    # Note: in typical case this loop executes _strip_once twice (the second
+    # execution does not remove any more tags).
+    strip_tags_depth = 0
     while "<" in value and ">" in value:
+        if strip_tags_depth >= MAX_STRIP_TAGS_DEPTH:
+            raise SuspiciousOperation
         new_value = _strip_once(value)
         if value.count("<") == new_value.count("<"):
             # _strip_once wasn't able to detect more tags.
             break
         value = new_value
+        strip_tags_depth += 1
     return value
 
 
diff --git a/docs/releases/4.2.17.txt b/docs/releases/4.2.17.txt
@@ -6,3 +6,19 @@ Django 4.2.17 release notes
 
 Django 4.2.17 fixes one security issue with severity "high" and one security
 issue with severity "moderate" in 4.2.16.
+
+CVE-2024-53907: Denial-of-service possibility in ``strip_tags()``
+=================================================================
+
+:func:`~django.utils.html.strip_tags` would be extremely slow to evaluate
+certain inputs containing large sequences of nested incomplete HTML entities.
+The ``strip_tags()`` method is used to implement the corresponding
+:tfilter:`striptags` template filter, which was thus also vulnerable.
+
+``strip_tags()`` now has an upper limit of recursive calls to ``HTMLParser``
+before raising a :exc:`.SuspiciousOperation` exception.
+
+Remember that absolutely NO guarantee is provided about the results of
+``strip_tags()`` being HTML safe. So NEVER mark safe the result of a
+``strip_tags()`` call without escaping it first, for example with
+:func:`django.utils.html.escape`.
diff --git a/docs/releases/5.0.10.txt b/docs/releases/5.0.10.txt
@@ -6,3 +6,19 @@ Django 5.0.10 release notes
 
 Django 5.0.10 fixes one security issue with severity "high" and one security
 issue with severity "moderate" in 5.0.9.
+
+CVE-2024-53907: Denial-of-service possibility in ``strip_tags()``
+=================================================================
+
+:func:`~django.utils.html.strip_tags` would be extremely slow to evaluate
+certain inputs containing large sequences of nested incomplete HTML entities.
+The ``strip_tags()`` method is used to implement the corresponding
+:tfilter:`striptags` template filter, which was thus also vulnerable.
+
+``strip_tags()`` now has an upper limit of recursive calls to ``HTMLParser``
+before raising a :exc:`.SuspiciousOperation` exception.
+
+Remember that absolutely NO guarantee is provided about the results of
+``strip_tags()`` being HTML safe. So NEVER mark safe the result of a
+``strip_tags()`` call without escaping it first, for example with
+:func:`django.utils.html.escape`.
diff --git a/docs/releases/5.1.4.txt b/docs/releases/5.1.4.txt
@@ -7,6 +7,22 @@ Django 5.1.4 release notes
 Django 5.1.4 fixes one security issue with severity "high", one security issue
 with severity "moderate", and several bugs in 5.1.3.
 
+CVE-2024-53907: Denial-of-service possibility in ``strip_tags()``
+=================================================================
+
+:func:`~django.utils.html.strip_tags` would be extremely slow to evaluate
+certain inputs containing large sequences of nested incomplete HTML entities.
+The ``strip_tags()`` method is used to implement the corresponding
+:tfilter:`striptags` template filter, which was thus also vulnerable.
+
+``strip_tags()`` now has an upper limit of recursive calls to ``HTMLParser``
+before raising a :exc:`.SuspiciousOperation` exception.
+
+Remember that absolutely NO guarantee is provided about the results of
+``strip_tags()`` being HTML safe. So NEVER mark safe the result of a
+``strip_tags()`` call without escaping it first, for example with
+:func:`django.utils.html.escape`.
+
 Bugfixes
 ========
 
diff --git a/tests/utils_tests/test_html.py b/tests/utils_tests/test_html.py
@@ -1,6 +1,7 @@
 import os
 from datetime import datetime
 
+from django.core.exceptions import SuspiciousOperation
 from django.core.serializers.json import DjangoJSONEncoder
 from django.test import SimpleTestCase
 from django.utils.deprecation import RemovedInDjango60Warning
@@ -124,12 +125,18 @@ def test_strip_tags(self):
             ("<script>alert()</script>&h", "alert()h"),
             ("><!" + ("&" * 16000) + "D", "><!" + ("&" * 16000) + "D"),
             ("X<<<<br>br>br>br>X", "XX"),
+            ("<" * 50 + "a>" * 50, ""),
         )
         for value, output in items:
             with self.subTest(value=value, output=output):
                 self.check_output(strip_tags, value, output)
                 self.check_output(strip_tags, lazystr(value), output)
 
+    def test_strip_tags_suspicious_operation(self):
+        value = "<" * 51 + "a>" * 51, "<a>"
+        with self.assertRaises(SuspiciousOperation):
+            strip_tags(value)
+
     def test_strip_tags_files(self):
         # Test with more lengthy content (also catching performance regressions)
         for filename in ("strip_tags1.html", "strip_tags2.txt"):
