Skip to content

Commit ab0893f

Browse files
serhiy-storchakaserhiy-storchaka
authored
[3.12] gh-135462: Fix quadratic complexity in processing special input in HTMLParser (GH-135464) (GH-135483)
End-of-file errors are now handled according to the HTML5 specs -- comments and declarations are automatically closed, tags are ignored. (cherry picked from commit 6eb6c5d)
1 parent 033aa5c commit ab0893f

File tree

3 files changed

+116
-23
lines changed

3 files changed

+116
-23
lines changed

‎Lib/html/parser.py

Lines changed: 30 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -25,6 +25,7 @@
2525
charref = re.compile('&#(?:[0-9]+|[xX][0-9a-fA-F]+)[^0-9a-fA-F]')
2626

2727
starttagopen = re.compile('<[a-zA-Z]')
28+
endtagopen = re.compile('</[a-zA-Z]')
2829
piclose = re.compile('>')
2930
commentclose = re.compile(r'--\s*>')
3031
# Note:
@@ -177,25 +178,43 @@ def goahead(self, end):
177178
k = self.parse_pi(i)
178179
elif startswith("<!", i):
179180
k = self.parse_html_declaration(i)
180-
elif (i + 1) < n:
181+
elif (i + 1) < n or end:
181182
self.handle_data("<")
182183
k = i + 1
183184
else:
184185
break
185186
if k < 0:
186187
if not end:
187188
break
188-
k = rawdata.find('>', i + 1)
189-
if k < 0:
190-
k = rawdata.find('<', i + 1)
191-
if k < 0:
192-
k = i + 1
193-
else:
194-
k += 1
195-
if self.convert_charrefs and not self.cdata_elem:
196-
self.handle_data(unescape(rawdata[i:k]))
189+
if starttagopen.match(rawdata, i): # < + letter
190+
pass
191+
elif startswith("</", i):
192+
if i + 2 == n:
193+
self.handle_data("</")
194+
elif endtagopen.match(rawdata, i): # </ + letter
195+
pass
196+
else:
197+
# bogus comment
198+
self.handle_comment(rawdata[i+2:])
199+
elif startswith("<!--", i):
200+
j = n
201+
for suffix in ("--!", "--", "-"):
202+
if rawdata.endswith(suffix, i+4):
203+
j -= len(suffix)
204+
break
205+
self.handle_comment(rawdata[i+4:j])
206+
elif startswith("<![CDATA[", i):
207+
self.unknown_decl(rawdata[i+3:])
208+
elif rawdata[i:i+9].lower() == '<!doctype':
209+
self.handle_decl(rawdata[i+2:])
210+
elif startswith("<!", i):
211+
# bogus comment
212+
self.handle_comment(rawdata[i+2:])
213+
elif startswith("<?", i):
214+
self.handle_pi(rawdata[i+2:])
197215
else:
198-
self.handle_data(rawdata[i:k])
216+
raise AssertionError("we should not get here!")
217+
k = n
199218
i = self.updatepos(i, k)
200219
elif startswith("&#", i):
201220
match = charref.match(rawdata, i)

‎Lib/test/test_htmlparser.py

Lines changed: 82 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,7 @@
55
import unittest
66

77
from unittest.mock import patch
8+
from test import support
89

910

1011
class EventCollector(html.parser.HTMLParser):
@@ -393,28 +394,34 @@ def test_tolerant_parsing(self):
393394
('data', '<'),
394395
('starttag', 'bc<', [('a', None)]),
395396
('endtag', 'html'),
396-
('data', '\n<img src="URL>'),
397-
('comment', '/img'),
398-
('endtag', 'html<')])
397+
('data', '\n')])
399398

400399
def test_starttag_junk_chars(self):
400+
self._run_check("<", [('data', '<')])
401+
self._run_check("<>", [('data', '<>')])
402+
self._run_check("< >", [('data', '< >')])
403+
self._run_check("< ", [('data', '< ')])
401404
self._run_check("</>", [])
405+
self._run_check("<$>", [('data', '<$>')])
402406
self._run_check("</$>", [('comment', '$')])
403407
self._run_check("</", [('data', '</')])
404-
self._run_check("</a", [('data', '</a')])
408+
self._run_check("</a", [])
409+
self._run_check("</ a>", [('endtag', 'a')])
410+
self._run_check("</ a", [('comment', ' a')])
405411
self._run_check("<a<a>", [('starttag', 'a<a', [])])
406412
self._run_check("</a<a>", [('endtag', 'a<a')])
407-
self._run_check("<!", [('data', '<!')])
408-
self._run_check("<a", [('data', '<a')])
409-
self._run_check("<a foo='bar'", [('data', "<a foo='bar'")])
410-
self._run_check("<a foo='bar", [('data', "<a foo='bar")])
411-
self._run_check("<a foo='>'", [('data', "<a foo='>'")])
412-
self._run_check("<a foo='>", [('data', "<a foo='>")])
413+
self._run_check("<!", [('comment', '')])
414+
self._run_check("<a", [])
415+
self._run_check("<a foo='bar'", [])
416+
self._run_check("<a foo='bar", [])
417+
self._run_check("<a foo='>'", [])
418+
self._run_check("<a foo='>", [])
413419
self._run_check("<a$>", [('starttag', 'a$', [])])
414420
self._run_check("<a$b>", [('starttag', 'a$b', [])])
415421
self._run_check("<a$b/>", [('startendtag', 'a$b', [])])
416422
self._run_check("<a$b >", [('starttag', 'a$b', [])])
417423
self._run_check("<a$b />", [('startendtag', 'a$b', [])])
424+
self._run_check("</a$b>", [('endtag', 'a$b')])
418425

419426
def test_slashes_in_starttag(self):
420427
self._run_check('<a foo="var"/>', [('startendtag', 'a', [('foo', 'var')])])
@@ -539,13 +546,56 @@ def test_EOF_in_charref(self):
539546
for html, expected in data:
540547
self._run_check(html, expected)
541548

542-
def test_broken_comments(self):
543-
html = ('<! not really a comment >'
549+
def test_eof_in_comments(self):
550+
data = [
551+
('<!--', [('comment', '')]),
552+
('<!---', [('comment', '')]),
553+
('<!----', [('comment', '')]),
554+
('<!-----', [('comment', '-')]),
555+
('<!------', [('comment', '--')]),
556+
('<!----!', [('comment', '')]),
557+
('<!---!', [('comment', '-!')]),
558+
('<!---!>', [('comment', '-!>')]),
559+
('<!--foo', [('comment', 'foo')]),
560+
('<!--foo-', [('comment', 'foo')]),
561+
('<!--foo--', [('comment', 'foo')]),
562+
('<!--foo--!', [('comment', 'foo')]),
563+
('<!--<!--', [('comment', '<!')]),
564+
('<!--<!--!', [('comment', '<!')]),
565+
]
566+
for html, expected in data:
567+
self._run_check(html, expected)
568+
569+
def test_eof_in_declarations(self):
570+
data = [
571+
('<!', [('comment', '')]),
572+
('<!-', [('comment', '-')]),
573+
('<![', [('comment', '[')]),
574+
('<![CDATA[', [('unknown decl', 'CDATA[')]),
575+
('<![CDATA[x', [('unknown decl', 'CDATA[x')]),
576+
('<![CDATA[x]', [('unknown decl', 'CDATA[x]')]),
577+
('<![CDATA[x]]', [('unknown decl', 'CDATA[x]]')]),
578+
('<!DOCTYPE', [('decl', 'DOCTYPE')]),
579+
('<!DOCTYPE ', [('decl', 'DOCTYPE ')]),
580+
('<!DOCTYPE html', [('decl', 'DOCTYPE html')]),
581+
('<!DOCTYPE html ', [('decl', 'DOCTYPE html ')]),
582+
('<!DOCTYPE html PUBLIC', [('decl', 'DOCTYPE html PUBLIC')]),
583+
('<!DOCTYPE html PUBLIC "foo', [('decl', 'DOCTYPE html PUBLIC "foo')]),
584+
('<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01//EN" "foo',
585+
[('decl', 'DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01//EN" "foo')]),
586+
]
587+
for html, expected in data:
588+
self._run_check(html, expected)
589+
590+
def test_bogus_comments(self):
591+
html = ('<!ELEMENT br EMPTY>'
592+
'<! not really a comment >'
544593
'<! not a comment either -->'
545594
'<! -- close enough -->'
546595
'<!><!<-- this was an empty comment>'
547596
'<!!! another bogus comment !!!>')
548597
expected = [
598+
('comment', 'ELEMENT br EMPTY'),
549599
('comment', ' not really a comment '),
550600
('comment', ' not a comment either --'),
551601
('comment', ' -- close enough --'),
@@ -600,6 +650,26 @@ def test_convert_charrefs_dropped_text(self):
600650
('endtag', 'a'), ('data', ' bar & baz')]
601651
)
602652

653+
@support.requires_resource('cpu')
654+
def test_eof_no_quadratic_complexity(self):
655+
# Each of these examples used to take about an hour.
656+
# Now they take a fraction of a second.
657+
def check(source):
658+
parser = html.parser.HTMLParser()
659+
parser.feed(source)
660+
parser.close()
661+
n = 120_000
662+
check("<a " * n)
663+
check("<a a=" * n)
664+
check("</a " * 14 * n)
665+
check("</a a=" * 11 * n)
666+
check("<!--" * 4 * n)
667+
check("<!" * 60 * n)
668+
check("<?" * 19 * n)
669+
check("</$" * 15 * n)
670+
check("<![CDATA[" * 9 * n)
671+
check("<!doctype" * 35 * n)
672+
603673

604674
class AttributesTestCase(TestCaseBase):
605675

‎Misc/NEWS.d/next/Security/2025-06-13-15-55-22.gh-issue-135462.KBeJpc.rst

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,4 @@
1+
Fix quadratic complexity in processing specially crafted input in
2+
:class:`html.parser.HTMLParser`. End-of-file errors are now handled according
3+
to the HTML5 specs -- comments and declarations are automatically closed,
4+
tags are ignored.

0 commit comments

Comments
 (0)