#22233: Only split headers on \r and/or \n, per email RFCs.

Original patch by Martin Panter, new policy fixes by me.

Lib/email/feedparser.py

@@ -27,6 +27,7 @@ from email import errors
 from email import message
 from email._policybase import compat32
 from collections import deque
+from io import StringIO
 
 NLCRE = re.compile('\r\n|\r|\n')
 NLCRE_bol = re.compile('(\r\n|\r|\n)')
@@ -51,8 +52,9 @@ class BufferedSubFile(object):
     simple abstraction -- it parses until EOF closes the current message.
     """
     def __init__(self):
-        # Chunks of the last partial line pushed into this object.
-        self._partial = []
+        # Text stream of the last partial line pushed into this object.
+        # See issue 22233 for why this is a text stream and not a list.
+        self._partial = StringIO(newline='')
         # A deque of full, pushed lines
         self._lines = deque()
         # The stack of false-EOF checking predicates.
@@ -68,8 +70,10 @@ class BufferedSubFile(object):
 
     def close(self):
         # Don't forget any trailing partial line.
-        self.pushlines(''.join(self._partial).splitlines(True))
-        self._partial = []
+        self._partial.seek(0)
+        self.pushlines(self._partial.readlines())
+        self._partial.seek(0)
+        self._partial.truncate()
         self._closed = True
 
     def readline(self):
@@ -97,26 +101,23 @@ class BufferedSubFile(object):
 
     def push(self, data):
         """Push some new data into this object."""
-        # Crack into lines, but preserve the linesep characters on the end of each
-        parts = data.splitlines(True)
-
-        if not parts or not parts[0].endswith(('\n', '\r')):
-            # No new complete lines, so just accumulate partials
-            self._partial += parts
+        self._partial.write(data)
+        if '\n' not in data and '\r' not in data:
+            # No new complete lines, wait for more.
             return
 
-        if self._partial:
-            # If there are previous leftovers, complete them now
-            self._partial.append(parts[0])
-            parts[0:1] = ''.join(self._partial).splitlines(True)
-            del self._partial[:]
+        # Crack into lines, preserving the linesep characters.
+        self._partial.seek(0)
+        parts = self._partial.readlines()
+        self._partial.seek(0)
+        self._partial.truncate()
 
         # If the last element of the list does not end in a newline, then treat
         # it as a partial line.  We only check for '\n' here because a line
         # ending with '\r' might be a line that was split in the middle of a
         # '\r\n' sequence (see bugs 1555570 and 1721862).
         if not parts[-1].endswith('\n'):
-            self._partial = [parts.pop()]
+            self._partial.write(parts.pop())
         self.pushlines(parts)
 
     def pushlines(self, lines):

Lib/email/policy.py

@@ -2,6 +2,7 @@
 code that adds all the email6 features.
 """
 
+import re
 from email._policybase import Policy, Compat32, compat32, _extend_docstrings
 from email.utils import _has_surrogates
 from email.headerregistry import HeaderRegistry as HeaderRegistry
@@ -18,6 +19,8 @@ __all__ = [
     'HTTP',
     ]
 
+linesep_splitter = re.compile(r'\n|\r')
+
 @_extend_docstrings
 class EmailPolicy(Policy):
 
@@ -135,6 +138,8 @@ class EmailPolicy(Policy):
         if hasattr(value, 'name') and value.name.lower() == name.lower():
             return (name, value)
         if isinstance(value, str) and len(value.splitlines())>1:
+            # XXX this error message isn't quite right when we use splitlines
+            # (see issue 22233), but I'm not sure what should happen here.
             raise ValueError("Header values may not contain linefeed "
                              "or carriage return characters")
         return (name, self.header_factory(name, value))
@@ -150,7 +155,9 @@ class EmailPolicy(Policy):
         """
         if hasattr(value, 'name'):
             return value
-        return self.header_factory(name, ''.join(value.splitlines()))
+        # We can't use splitlines here because it splits on more than \r and \n.
+        value = ''.join(linesep_splitter.split(value))
+        return self.header_factory(name, value)
 
     def fold(self, name, value):
         """+

Lib/test/test_email/test_email.py

@@ -3444,10 +3444,12 @@ class TestFeedParsers(TestEmailBase):
         self.assertEqual(m.keys(), ['a', 'b'])
         m = self.parse(['a:\r', '\nb:\n'])
         self.assertEqual(m.keys(), ['a', 'b'])
+
+        # Only CR and LF should break header fields
         m = self.parse(['a:\x85b:\u2028c:\n'])
-        self.assertEqual(m.items(), [('a', '\x85'), ('b', '\u2028'), ('c', '')])
+        self.assertEqual(m.items(), [('a', '\x85b:\u2028c:')])
         m = self.parse(['a:\r', 'b:\x85', 'c:\n'])
-        self.assertEqual(m.items(), [('a', ''), ('b', '\x85'), ('c', '')])
+        self.assertEqual(m.items(), [('a', ''), ('b', '\x85c:')])
 
     def test_long_lines(self):
         # Expected peak memory use on 32-bit platform: 6*N*M bytes.

Lib/test/test_email/test_parser.py

@@ -2,6 +2,7 @@ import io
 import email
 import unittest
 from email.message import Message
+from email.policy import default
 from test.test_email import TestEmailBase
 
 
@@ -32,5 +33,45 @@ class TestCustomMessage(TestEmailBase):
     # XXX add tests for other functions that take Message arg.
 
 
+class TestParserBase:
+
+    def test_only_split_on_cr_lf(self):
+        # The unicode line splitter splits on unicode linebreaks, which are
+        # more numerous than allowed by the email RFCs; make sure we are only
+        # splitting on those two.
+        msg = self.parser(
+            "Next-Line: not\x85broken\r\n"
+            "Null: not\x00broken\r\n"
+            "Vertical-Tab: not\vbroken\r\n"
+            "Form-Feed: not\fbroken\r\n"
+            "File-Separator: not\x1Cbroken\r\n"
+            "Group-Separator: not\x1Dbroken\r\n"
+            "Record-Separator: not\x1Ebroken\r\n"
+            "Line-Separator: not\u2028broken\r\n"
+            "Paragraph-Separator: not\u2029broken\r\n"
+            "\r\n",
+            policy=default,
+        )
+        self.assertEqual(msg.items(), [
+            ("Next-Line", "not\x85broken"),
+            ("Null", "not\x00broken"),
+            ("Vertical-Tab", "not\vbroken"),
+            ("Form-Feed", "not\fbroken"),
+            ("File-Separator", "not\x1Cbroken"),
+            ("Group-Separator", "not\x1Dbroken"),
+            ("Record-Separator", "not\x1Ebroken"),
+            ("Line-Separator", "not\u2028broken"),
+            ("Paragraph-Separator", "not\u2029broken"),
+        ])
+        self.assertEqual(msg.get_payload(), "")
+
+class TestParser(TestParserBase, TestEmailBase):
+    parser = staticmethod(email.message_from_string)
+
+class TestBytesParser(TestParserBase, TestEmailBase):
+    def parser(self, s, *args, **kw):
+        return email.message_from_bytes(s.encode(), *args, **kw)
+
+
 if __name__ == '__main__':
     unittest.main()

Lib/test/test_httplib.py

@@ -283,6 +283,36 @@ class HeaderTests(TestCase):
         self.assertEqual(resp.getheader('First'), 'val')
         self.assertEqual(resp.getheader('Second'), 'val')
 
+    def test_parse_all_octets(self):
+        # Ensure no valid header field octet breaks the parser
+        body = (
+            b'HTTP/1.1 200 OK\r\n'
+            b"!#$%&'*+-.^_`|~: value\r\n"  # Special token characters
+            b'VCHAR: ' + bytes(range(0x21, 0x7E + 1)) + b'\r\n'
+            b'obs-text: ' + bytes(range(0x80, 0xFF + 1)) + b'\r\n'
+            b'obs-fold: text\r\n'
+            b' folded with space\r\n'
+            b'\tfolded with tab\r\n'
+            b'Content-Length: 0\r\n'
+            b'\r\n'
+        )
+        sock = FakeSocket(body)
+        resp = client.HTTPResponse(sock)
+        resp.begin()
+        self.assertEqual(resp.getheader('Content-Length'), '0')
+        self.assertEqual(resp.msg['Content-Length'], '0')
+        self.assertEqual(resp.getheader("!#$%&'*+-.^_`|~"), 'value')
+        self.assertEqual(resp.msg["!#$%&'*+-.^_`|~"], 'value')
+        vchar = ''.join(map(chr, range(0x21, 0x7E + 1)))
+        self.assertEqual(resp.getheader('VCHAR'), vchar)
+        self.assertEqual(resp.msg['VCHAR'], vchar)
+        self.assertIsNotNone(resp.getheader('obs-text'))
+        self.assertIn('obs-text', resp.msg)
+        for folded in (resp.getheader('obs-fold'), resp.msg['obs-fold']):
+            self.assertTrue(folded.startswith('text'))
+            self.assertIn(' folded with space', folded)
+            self.assertTrue(folded.endswith('folded with tab'))
+
     def test_invalid_headers(self):
         conn = client.HTTPConnection('example.com')
         conn.sock = FakeSocket('')

Misc/NEWS

@@ -62,6 +62,10 @@ Core and Builtins
 Library
 -------
 
+- Issue #22233: Break email header lines *only* on the RFC specified CR and LF
+  characters, not on arbitrary unicode line breaks.  This also fixes a bug in
+  HTTP header parsing.
+
 - Issue 27988: Fix email iter_attachments incorrect mutation of payload list.
 
 - Issue #27691: Fix ssl module's parsing of GEN_RID subject alternative name