diff --git a/fail2ban/tests/files/logs/sshd b/fail2ban/tests/files/logs/sshd index fb3defea..b9559359 100644 --- a/fail2ban/tests/files/logs/sshd +++ b/fail2ban/tests/files/logs/sshd @@ -1,3 +1,5 @@ +# filterOptions: [{}, {"mode": "aggressive"}] + #1 # failJSON: { "time": "2005-06-21T16:47:48", "match": true , "host": "192.030.0.6" } Jun 21 16:47:48 digital-mlhhyiqscv sshd[13709]: error: PAM: Authentication failure for myhlj1374 from 192.030.0.6 @@ -189,7 +191,7 @@ Apr 27 13:02:04 host sshd[29116]: Received disconnect from 1.2.3.4: 11: Normal S # failJSON: { "time": "2015-04-16T20:02:50", "match": true , "host": "222.186.21.217", "desc": "Authentication for user failed" } 2015-04-16T18:02:50.321974+00:00 host sshd[2716]: pam_unix(sshd:auth): authentication failure; logname= uid=0 euid=0 tty=ssh ruser= rhost=222.186.21.217 user=root -# filterOptions: {"mode": "ddos"} +# filterOptions: [{"mode": "ddos"}, {"mode": "aggressive"}] # http://forums.powervps.com/showthread.php?t=1667 # failJSON: { "time": "2005-06-07T01:10:56", "match": true , "host": "69.61.56.114" } @@ -218,7 +220,7 @@ Nov 24 23:46:43 host sshd[32686]: fatal: Read from socket failed: Connection res Mar 15 09:20:57 host sshd[28972]: Connection reset by 192.0.2.39 port 14282 [preauth] -# filterOptions: {"mode": "extra"} +# filterOptions: [{"mode": "extra"}, {"mode": "aggressive"}] # several other cases from gh-864: # failJSON: { "time": "2004-11-25T01:34:12", "match": true , "host": "127.0.0.1", "desc": "No supported authentication methods" } diff --git a/fail2ban/tests/samplestestcase.py b/fail2ban/tests/samplestestcase.py index 121c1c5c..5f0a447a 100644 --- a/fail2ban/tests/samplestestcase.py +++ b/fail2ban/tests/samplestestcase.py @@ -49,7 +49,8 @@ class FilterSamplesRegex(unittest.TestCase): def setUp(self): """Call before every test case.""" super(FilterSamplesRegex, self).setUp() - self.filter = None + self._filters = dict() + self._filterTests = None setUpMyTime() def tearDown(self): @@ -79,14 +80,20 @@ class FilterSamplesRegex(unittest.TestCase): RE_WRONG_GREED.search('non-greedy .+? test' + RE_HOST + ' test vary catch-all .* anchored$')) - def _readFilter(self, name, basedir, opts=None): - self.filter = Filter(None) - self.filter.returnRawHost = True - self.filter.checkAllRegex = True - self.filter.checkFindTime = False - self.filter.active = True + def _readFilter(self, fltName, name, basedir, opts=None): + # Check filter with this option combination was already used: + flt = self._filters.get(fltName) + if flt: + return flt + # First time: + flt = Filter(None) + flt.returnRawHost = True + flt.checkAllRegex = True + flt.checkFindTime = False + flt.active = True + # Read filter: if opts is None: opts = dict() - # Check filter exists + opts = opts.copy() filterConf = FilterReader(name, "jail", opts, basedir=basedir, share_config=unittest.F2B.share_config) self.assertEqual(filterConf.getFile(), name) @@ -103,25 +110,28 @@ class FilterSamplesRegex(unittest.TestCase): self.fail('Unexpected config-token %r in stream' % (opt,)) for optval in optval: if opt[2] == "prefregex": - self.filter.prefRegex = optval + flt.prefRegex = optval elif opt[2] == "addfailregex": - self.filter.addFailRegex(optval) + flt.addFailRegex(optval) elif opt[2] == "addignoreregex": - self.filter.addIgnoreRegex(optval) + flt.addIgnoreRegex(optval) elif opt[2] == "maxlines": - self.filter.setMaxLines(optval) + flt.setMaxLines(optval) elif opt[2] == "datepattern": - self.filter.setDatePattern(optval) + flt.setDatePattern(optval) # test regexp contains greedy catch-all before , that is # not hard-anchored at end or has not precise sub expression after : - regexList = self.filter.getFailRegex() + regexList = flt.getFailRegex() for fr in regexList: if RE_WRONG_GREED.search(fr): # pragma: no cover raise AssertionError("Following regexp of \"%s\" contains greedy catch-all before , " "that is not hard-anchored at end or has not precise sub expression after :\n%s" % - (name, str(fr).replace(RE_HOST, ''))) - return regexList + (fltName, str(fr).replace(RE_HOST, ''))) + # Cache within used filter combinations and return: + flt = [flt, set()] + self._filters[fltName] = flt + return flt def testSampleRegexsFactory(name, basedir): def testFilter(self): @@ -129,18 +139,11 @@ def testSampleRegexsFactory(name, basedir): self.assertTrue( os.path.isfile(os.path.join(TEST_FILES_DIR, "logs", name)), "No sample log file available for '%s' filter" % name) - - regexList = None - regexsUsedIdx = set() - regexsUsedRe = set() + filenames = [name] + regexsUsedRe = set() - def _testMissingSamples(): - for failRegexIndex, failRegex in enumerate(regexList): - self.assertTrue( - failRegexIndex in regexsUsedIdx or failRegex in regexsUsedRe, - "Regex for filter '%s' has no samples: %i: %r" % - (name, failRegexIndex, failRegex)) + # process each test-file (note: array filenames can grow during processing): i = 0 while i < len(filenames): filename = filenames[i]; i += 1; @@ -154,13 +157,17 @@ def testSampleRegexsFactory(name, basedir): faildata = json.loads(jsonREMatch.group(2)) # filterOptions - dict in JSON to control filter options (e. g. mode, etc.): if jsonREMatch.group(1) == 'filterOptions': - # another filter mode - we should check previous also: - if self.filter is not None: - _testMissingSamples() - regexsUsedIdx = set() # clear used indices (possible overlapping by mode change) - # read filter with another setting: - self.filter = None - regexList = self._readFilter(name, basedir, opts=faildata) + # following lines with another filter options: + self._filterTests = [] + for opts in (faildata if isinstance(faildata, list) else [faildata]): + # unique filter name (using options combination): + self.assertTrue(isinstance(opts, dict)) + fltName = opts.get('filterName') + if not fltName: fltName = str(opts) if opts else '' + fltName = name + fltName + # read it: + flt = self._readFilter(fltName, name, basedir, opts=opts) + self._filterTests.append((fltName, flt)) continue # addFILE - filename to "include" test-files should be additionally parsed: if jsonREMatch.group(1) == 'addFILE': @@ -176,65 +183,81 @@ def testSampleRegexsFactory(name, basedir): else: # pragma: no cover - normally unreachable faildata = {} - if self.filter is None: - regexList = self._readFilter(name, basedir, opts=None) + # if filter options was not yet specified: + if not self._filterTests: + fltName = name + flt = self._readFilter(fltName, name, basedir, opts=None) + self._filterTests = [(fltName, flt)] + + # process line using several filter options (if specified in the test-file): + for fltName, flt in self._filterTests: + flt, regexsUsedIdx = flt + regexList = flt.getFailRegex() + + try: + ret = flt.processLine(line) + if not ret: + # Bypass if filter constraint specified: + if faildata.get('filter') and name != faildata.get('filter'): + continue + # Check line is flagged as none match + self.assertFalse(faildata.get('match', True), + "Line not matched when should have") + continue - try: - ret = self.filter.processLine(line) - if not ret: - # Bypass if filter constraint specified: - if faildata.get('filter') and name != faildata.get('filter'): + failregex, fid, fail2banTime, fail = ret[0] + # Bypass no failure helpers-regexp: + if not faildata.get('match', False) and (fid is None or fail.get('nofail')): + regexsUsedIdx.add(failregex) + regexsUsedRe.add(regexList[failregex]) continue - # Check line is flagged as none match - self.assertFalse(faildata.get('match', True), - "Line not matched when should have") - continue - - failregex, fid, fail2banTime, fail = ret[0] - # Bypass no failure helpers-regexp: - if not faildata.get('match', False) and (fid is None or fail.get('nofail')): + + # Check line is flagged to match + self.assertTrue(faildata.get('match', False), + "Line matched when shouldn't have") + self.assertEqual(len(ret), 1, + "Multiple regexs matched %r" % (map(lambda x: x[0], ret))) + + # Verify match captures (at least fid/host) and timestamp as expected + for k, v in faildata.iteritems(): + if k not in ("time", "match", "desc", "filter"): + fv = fail.get(k, None) + # Fallback for backwards compatibility (previously no fid, was host only): + if k == "host" and fv is None: + fv = fid + self.assertEqual(fv, v) + + t = faildata.get("time", None) + try: + jsonTimeLocal = datetime.datetime.strptime(t, "%Y-%m-%dT%H:%M:%S") + except ValueError: + jsonTimeLocal = datetime.datetime.strptime(t, "%Y-%m-%dT%H:%M:%S.%f") + + jsonTime = time.mktime(jsonTimeLocal.timetuple()) + + jsonTime += jsonTimeLocal.microsecond / 1000000 + + self.assertEqual(fail2banTime, jsonTime, + "UTC Time mismatch %s (%s) != %s (%s) (diff %.3f seconds)" % + (fail2banTime, time.strftime("%Y-%m-%dT%H:%M:%S", time.gmtime(fail2banTime)), + jsonTime, time.strftime("%Y-%m-%dT%H:%M:%S", time.gmtime(jsonTime)), + fail2banTime - jsonTime) ) + regexsUsedIdx.add(failregex) regexsUsedRe.add(regexList[failregex]) - continue - - # Check line is flagged to match - self.assertTrue(faildata.get('match', False), - "Line matched when shouldn't have") - self.assertEqual(len(ret), 1, - "Multiple regexs matched %r" % (map(lambda x: x[0], ret))) - - # Verify match captures (at least fid/host) and timestamp as expected - for k, v in faildata.iteritems(): - if k not in ("time", "match", "desc", "filter"): - fv = fail.get(k, None) - # Fallback for backwards compatibility (previously no fid, was host only): - if k == "host" and fv is None: - fv = fid - self.assertEqual(fv, v) - - t = faildata.get("time", None) - try: - jsonTimeLocal = datetime.datetime.strptime(t, "%Y-%m-%dT%H:%M:%S") - except ValueError: - jsonTimeLocal = datetime.datetime.strptime(t, "%Y-%m-%dT%H:%M:%S.%f") - - jsonTime = time.mktime(jsonTimeLocal.timetuple()) - - jsonTime += jsonTimeLocal.microsecond / 1000000 - - self.assertEqual(fail2banTime, jsonTime, - "UTC Time mismatch %s (%s) != %s (%s) (diff %.3f seconds)" % - (fail2banTime, time.strftime("%Y-%m-%dT%H:%M:%S", time.gmtime(fail2banTime)), - jsonTime, time.strftime("%Y-%m-%dT%H:%M:%S", time.gmtime(jsonTime)), - fail2banTime - jsonTime) ) - - regexsUsedIdx.add(failregex) - regexsUsedRe.add(regexList[failregex]) - except AssertionError as e: # pragma: no cover - raise AssertionError("%s on: %s:%i, line:\n%s" % ( - e, logFile.filename(), logFile.filelineno(), line)) - - _testMissingSamples() + except AssertionError as e: # pragma: no cover + raise AssertionError("%s: %s on: %s:%i, line:\n%s" % ( + fltName, e, logFile.filename(), logFile.filelineno(), line)) + + # check missing samples for regex using each filter-options combination: + for fltName, flt in self._filters.iteritems(): + flt, regexsUsedIdx = flt + regexList = flt.getFailRegex() + for failRegexIndex, failRegex in enumerate(regexList): + self.assertTrue( + failRegexIndex in regexsUsedIdx or failRegex in regexsUsedRe, + "%s: Regex has no samples: %i: %r" % + (fltName, failRegexIndex, failRegex)) return testFilter