Prevent accidental DoS on malformed stacktraces (#3980)

Enormous stacktraces containing a giant array on a single line is
causing bots to freeze.
Although fuzztest really should not be printing an input this large,
let's try to be resilient when it misbehaves.

Fixes: #3978

src/clusterfuzz/_internal/tests/core/crash_analysis/stack_parsing/stack_analyzer_test.py

@@ -3744,6 +3744,12 @@ def test_split_stacktrace(self):
 
  self.assertEqual(actual_split_stacktrace, expected_split_stacktrace)
 
+ def test_split_stacktrace_no_dos(self):
+ """Tests that split_stacktrace wont search forever unless there's a very
+ well crafted input."""
+ stacktrace = "S text\n" + " 42 " * 100000 + "j text\n"
+ StackParser.split_stacktrace(stacktrace)
+
  def test_jazzer_js_javascript(self):
  """Test Jazzer.js JS stacktrace."""
  data = self._read_test_data('jazzer_js_javascript.txt')

src/clusterfuzz/stacktraces/__init__.py

@@ -1407,6 +1407,10 @@ def should_ignore_line_for_crash_processing(line, state):
  if SAN_DEADLYSIGNAL_REGEX.match(line):
  return True
 
+ if len(line) > 1024**2:
+ logs.log_error('Line is too long for a stacktrace.')
+ return True
+
  return False
 
 

src/clusterfuzz/stacktraces/constants.py

@@ -66,8 +66,11 @@
 # The 'assertion .* failed' is suffixed with the failure reason. E.g.,
 # 'assertion __dest < len() failed: sample_array[] index out of bounds',
 # 'assertion !full() failed: sample_function() called on an empty vector'
+# Add the lookahead group to prevent DoS, see
+# https://github.com/google/clusterfuzz/issues/3978.
 ASSERT_REGEX_GLIBC_SUFFIXED = re.compile(
- r'.*\S.*\/.*:\d+:\s*assertion .* failed:\s*(\S.*)')
+ r'(?=.*assertion .* failed:.).*.*\S.*\/.*:\d+:\s*assertion .* failed:\s*(\S.*)' # pylint: disable=line-too-long
+)
 ASSERT_NOT_REACHED_REGEX = re.compile(r'^\s*SHOULD NEVER BE REACHED\s*$')
 CENTIPEDE_TIMEOUT_REGEX = re.compile(r'(?:%s)' % '|'.join([
  r'========= Timeout of \d+ seconds exceeded; exiting',
@@ -200,8 +203,11 @@
  r'[ ]*([^ ]*|Atomic [^ ]*) of size ([^ ]*) at ([^ ]*)')
 SAN_DEADLYSIGNAL_REGEX = re.compile(
  r'(Address|Leak|Memory|UndefinedBehavior|Thread)Sanitizer:DEADLYSIGNAL')
+# Use the lookahead group to prevent DoS, see
+# https://github.com/google/clusterfuzz/issues/3978.
 CONCATENATED_SAN_DEADLYSIGNAL_REGEX = re.compile(
- r'\n([^\n]*\S[^\n]*)(' + SAN_DEADLYSIGNAL_REGEX.pattern + r')\n')
+ r'\n(?=.*Sanitizer\:DEADLYSIGNAL.*)([^\n]*\S[^\n]*)(' +
+ SAN_DEADLYSIGNAL_REGEX.pattern + r')\n')
 SPLIT_CONCATENATED_SAN_DEADLYSIGNAL_REGEX = r'\n\1\n\2\n'
 SAN_FPE_REGEX = re.compile(r'.*[a-zA-Z]+Sanitizer: FPE ')
 SAN_ILL_REGEX = re.compile(r'.*[a-zA-Z]+Sanitizer: ILL ')