def _send_request(self, parser, rendered_data):
""" Send a request and invoke the response parser.
@param parser: A parser to parse the data
@type parser: Function pointer
@param rendered_data: The request's rendered data to send
@type rendered_data: Str
@return: The response from the server
@rtype : HttpResponse
"""
try:
sock = messaging.HttpSock(self._connection_settings)
except TransportLayerException as error:
RAW_LOGGING(f"{error!s}")
# connection failed
return HttpResponse()
success, response = sock.sendRecv(
rendered_data,
req_timeout_sec=Settings().max_request_execution_time)
if not success:
RAW_LOGGING(response.to_str)
Monitor().increment_requests_count(self.__class__.__name__)
return response
def run(self):
""" Thread entrance - performs fuzzing
"""
try:
self._num_total_sequences = driver.generate_sequences(
self._fuzzing_requests, self._checkers, self._fuzzing_jobs)
# At the end of everything print out any request that were never
# rendered (because they never had valid constraints).
logger.print_request_rendering_stats_never_rendered_requests(
self._fuzzing_requests,
GrammarRequestCollection().candidate_values_pool, Monitor())
except InvalidDictionaryException:
pass
except Exception as err:
self._exception = traceback.format_exc()
def _render_last_request(self, seq):
""" Render the last request of the sequence and inspect the status
code of the response. If it's any of 20x, we have probably hit a bug.
@param seq: The sequence whose last request we will try to render.
@type seq: Sequence Class object.
@return: None
@rtype : None
"""
request = seq.last_request
for rendered_data, parser in\
request.render_iter(self._req_collection.candidate_values_pool,
skip=request._current_combination_id):
# Hold the lock (because other workers may be rendering the same
# request) and check whether the current rendering is known from the
# past to lead to invalid status codes. If so, skip the current
# rendering.
if self._lock is not None:
self._lock.acquire()
should_skip = Monitor().is_invalid_rendering(request)
if self._lock is not None:
self._lock.release()
# Skip the loop and don't forget to increase the counter.
if should_skip:
RAW_LOGGING("Skipping rendering: {}".\
format(request._current_combination_id))
request._current_combination_id += 1
continue
rendered_data = seq.resolve_dependencies(rendered_data)
response = self._send_request(parser, rendered_data)
request_utilities.call_response_parser(parser, response)
# Append the rendered data to the sent list as we will not be rendering
# with the sequence's render function
seq.append_data_to_sent_list(rendered_data, parser, response)
if response and self._rule_violation(seq, response):
self._print_suspect_sequence(seq, response)
BugBuckets.Instance().update_bug_buckets(
seq, response.status_code, origin=self.__class__.__name__)
def render_parallel(seq_collection, fuzzing_pool, checkers, generation, global_lock):
""" Does rendering work in parallel by invoking "render_one" multiple
times using a pool of python workers. For brevity we skip arguments and
return types, since they are similar with "render_one".
"""
prev_len = len(seq_collection)
result = fuzzing_pool.starmap(render_one,
[(seq_collection, ith,
checkers, generation, global_lock
)\
for ith in range(prev_len)])
seq_collection = list(itertools.chain(*result))
# Increase internal fuzzing generations' counter. Since the
# constructor of RequestCollection starts this counter from zero,
# the counter will be equal to lenght + 1 after the following line.
Monitor().current_fuzzing_generation += 1
return seq_collection
def render_sequential(seq_collection, fuzzing_pool, checkers, generation, global_lock):
""" Does rendering work sequential by invoking "render_one" multiple
times. For brevity we skip arguments and return types, since they are
similar with "render_one".
"""
prev_len = len(seq_collection)
for ith in range(prev_len):
valid_renderings = render_one(seq_collection, ith, checkers, generation, global_lock)
# Extend collection by adding all valid renderings
seq_collection.extend(valid_renderings)
if len(seq_collection[prev_len:]) == 0:
raise ExhaustSeqCollectionException("")
# Increase internal fuzzing generations' counter. Since the
# constructor of RequestCollection starts this counter from zero,
# the counter will be equal to length + 1 after the following line.
Monitor().current_fuzzing_generation += 1
return seq_collection[prev_len:]
def apply_create_once_resources(fuzzing_requests):
""" Attempts to create all of the resources in the 'create_once' endpoints.
@param fuzzing_requests: The collection of requests to be fuzzed
@type fuzzing_requests: FuzzingRequestCollection
@return: A list of destructors to use to cleanup the create_once resources
@rtype : list(Request)
"""
def exclude_requests(pre_reqs, post_reqs):
# Exclude any requests that produce or destroy the create_once endpoint
for req_i in pre_reqs:
fuzzing_requests.exclude_preprocessing_request(req_i)
for req_i in post_reqs:
fuzzing_requests.exclude_postprocessing_request(req_i)
create_once_endpoints = Settings().create_once_endpoints
if not create_once_endpoints:
return
logger.create_network_log(logger.LOG_TYPE_PREPROCESSING)
destructors = set()
exclude_reqs = set()
request_count = 0
logger.write_to_main("Rendering for create-once resources:\n")
# Iterate through each 'create_once' endpoint
for endpoint in create_once_endpoints:
# Verify that the endpoint exists in the request collection
if endpoint in GrammarRequestCollection().request_id_collection:
# The create_once resource generator
resource_gen_req = None
# Iterate through each of the requests that contain the create_once endpoint
for req in GrammarRequestCollection(
).request_id_collection[endpoint]:
if req not in fuzzing_requests:
logger.write_to_main(
"Warning: Create-once endpoint is not a request in the fuzzing list\n",
True)
break
if not resource_gen_req and req.is_resource_generator():
resource_gen_req = req
# Compute the sequence necessary to create the create_once resource
req_list = driver.compute_request_goal_seq(
resource_gen_req, fuzzing_requests)
logger.write_to_main(
f"{formatting.timestamp()}: Endpoint - {resource_gen_req.endpoint_no_dynamic_objects}"
)
logger.write_to_main(
f"{formatting.timestamp()}: Hex Def - {resource_gen_req.method_endpoint_hex_definition}"
)
create_once_seq = sequences.Sequence(req_list)
renderings = create_once_seq.render(
GrammarRequestCollection().candidate_values_pool,
None,
preprocessing=True)
# Make sure we were able to successfully create the create_once resource
if not renderings.valid:
logger.write_to_main(
f"{formatting.timestamp()}: Rendering INVALID")
exclude_requests(exclude_reqs, destructors)
raise FailedToCreateResource(destructors)
logger.write_to_main(
f"{formatting.timestamp()}: Rendering VALID")
logger.format_rendering_stats_definition(
resource_gen_req,
GrammarRequestCollection().candidate_values_pool)
if Settings().in_smoke_test_mode():
resource_gen_req.stats.request_order = 'Preprocessing'
resource_gen_req.stats.valid = 1
resource_gen_req.stats.status_code = renderings.final_request_response.status_code
resource_gen_req.stats.status_text = renderings.final_request_response.status_text
resource_gen_req.stats.sample_request.set_request_stats(
renderings.sequence.sent_request_data_list[-1].
rendered_data)
resource_gen_req.stats.sample_request.set_response_stats(
renderings.final_request_response,
renderings.final_response_datetime)
if req.is_destructor():
# Add destructors to the destructor list that will be returned
destructors.add(req)
# Only continue processing if a resource generator was actually found for this endpoint
if not resource_gen_req:
continue
request_count += len(req_list)
# Get the set of all dynamic object names in the endpoint
var_names = resource_gen_req.consumes.union(
resource_gen_req.produces)
# This dictionary will map dynamic object names to the values created during
# this preprocessing create-once step.
dynamic_object_values = {}
for name in var_names:
dynamic_object_values[name] = dependencies.get_variable(name)
# Iterate through the entire request collection, searching for requests that include
# the create_once resource. We want to "lock" the resources in these requests with
# the dynamic object values that were created during this preprocessing step.
for req_i in fuzzing_requests:
# Set the variables in any requests whose consumers were produced
# by the create_once resource generator
if resource_gen_req.produces & req_i.consumes:
req_i.set_id_values_for_create_once_dynamic_objects(
dynamic_object_values, renderings)
# Exclude any requests that produce the create_once object(s)
if resource_gen_req.produces & req_i.produces:
exclude_reqs.add(req_i)
else:
exclude_requests(exclude_reqs, destructors)
raise InvalidCreateOnce(destructors)
exclude_requests(exclude_reqs, destructors)
# Reset all of the dynamic object values that were just created
dependencies.reset_tlb()
# Reset the garbage collector, so it doesn't delete any of the resources that were just created
dependencies.set_saved_dynamic_objects()
logger.print_request_rendering_stats(
GrammarRequestCollection().candidate_values_pool, fuzzing_requests,
Monitor(), request_count, logger.PREPROCESSING_GENERATION, None)
# Return the list of destructors that were removed from the request collection.
# These will be used to cleanup the create_once resources created during preprocessing.
return list(destructors)
def generate_sequences(fuzzing_requests, checkers, fuzzing_jobs=1):
""" Implements core restler algorithm.
@param fuzzing_requests: The collection of requests that will be fuzzed
@type fuzzing_requests: FuzzingRequestCollection
@param checkers: The list of checkers to apply
@type checkers: list[Checker]
@param fuzzing_jobs: Optional number of fuzzing jobs for parallel fuzzing.
Default value passed is one (sequential fuzzing).
@type fuzzing_jobs: Int
@return: None
@rtype : None
"""
if not fuzzing_requests.size:
return
logger.create_network_log(logger.LOG_TYPE_TESTING)
fuzzing_mode = Settings().fuzzing_mode
max_len = Settings().max_sequence_length
if fuzzing_mode == 'directed-smoke-test':
return generate_sequences_directed_smoketest(fuzzing_requests,
checkers)
if fuzzing_jobs > 1:
render = render_parallel
global_lock = multiprocessing.Lock()
fuzzing_pool = ThreadPool(fuzzing_jobs)
else:
global_lock = None
fuzzing_pool = None
render = render_sequential
should_stop = False
timeout_reached = False
seq_collection_exhausted = False
num_total_sequences = 0
while not should_stop:
seq_collection = [sequences.Sequence()]
# Only for bfs: If any checkpoint file is available, load state of
# latest generation. Note that it only makes sense to use checkpoints
# for the bfs exploration method, since it is the only systemic and
# exhaustive method.
min_len = 0
if fuzzing_mode == 'bfs':
req_collection = GrammarRequestCollection()
monitor = Monitor()
req_collection, seq_collection, fuzzing_requests, monitor, min_len =\
saver.load(req_collection, seq_collection, fuzzing_requests, monitor)
requests.GlobalRequestCollection.Instance(
)._req_collection = req_collection
fuzzing_monitor.FuzzingMonitor.__instance = monitor
# Repeat external loop only for random walk
if fuzzing_mode != 'random-walk':
should_stop = True
# Initialize fuzzing schedule
fuzzing_schedule = {}
logger.write_to_main(f"Setting fuzzing schemes: {fuzzing_mode}")
for length in range(min_len, max_len):
fuzzing_schedule[length] = fuzzing_mode
# print(" - {}: {}".format(length + 1, fuzzing_schedule[length]))
# print general request-related stats
logger.print_req_collection_stats(
fuzzing_requests,
GrammarRequestCollection().candidate_values_pool)
generation = 0
for length in range(min_len, max_len):
# we can set this without locking, since noone else writes (main
# driver is single-threaded) and every potential worker will just
# read-access this value.
generation = length + 1
fuzzing_mode = fuzzing_schedule[length]
# extend sequences with new request templates
seq_collection = extend(seq_collection, fuzzing_requests,
global_lock)
print(f"{formatting.timestamp()}: Generation: {generation} ")
logger.write_to_main(
f"{formatting.timestamp()}: Generation: {generation} / "
f"Sequences Collection Size: {len(seq_collection)} "
f"(After {fuzzing_schedule[length]} Extend)")
# render templates
try:
seq_collection_exhausted = False
seq_collection = render(seq_collection, fuzzing_pool, checkers,
generation, global_lock)
except TimeOutException:
logger.write_to_main("Timed out...")
timeout_reached = True
seq_collection_exhausted = True
# Increase fuzzing generation after timeout because the code
# that does it would have never been reached. This is done so
# the previous generation's test summary is logged correctly.
Monitor().current_fuzzing_generation += 1
except ExhaustSeqCollectionException:
logger.write_to_main("Exhausted collection...")
seq_collection = []
seq_collection_exhausted = True
logger.write_to_main(
f"{formatting.timestamp()}: Generation: {generation} / "
f"Sequences Collection Size: {len(seq_collection)} "
f"(After {fuzzing_schedule[length]} Render)")
# saving latest state
saver.save(GrammarRequestCollection(), seq_collection,
fuzzing_requests, Monitor(), generation)
# Print stats for iteration of the current generation
logger.print_generation_stats(GrammarRequestCollection(),
Monitor(), global_lock)
num_total_sequences += len(seq_collection)
logger.print_request_rendering_stats(
GrammarRequestCollection().candidate_values_pool,
fuzzing_requests, Monitor(),
Monitor().num_fully_rendered_requests(
fuzzing_requests.all_requests), generation, global_lock)
if timeout_reached or seq_collection_exhausted:
if timeout_reached:
should_stop = True
break
logger.write_to_main("--\n")
if fuzzing_pool is not None:
fuzzing_pool.close()
fuzzing_pool.join()
return num_total_sequences
def extend(seq_collection, fuzzing_requests, lock):
""" Extends each sequence currently present in collection by any request
from request collection whose dependencies can be resolved if appended at
the end of the target sequence.
@param seq_collection: List of sequences in sequence collection.
@type seq_collection: List
@param fuzzing_requests: The collection of requests to fuzz.
@type fuzzing_requests: FuzzingRequestCollection.
@param lock: Lock object used for sync of more than one fuzzing jobs.
@type lock: thread.Lock object
@return: The list of newly enxtended sequences.
@rtype : List
"""
prev_len = len(seq_collection)
# The functions that access the monitor of renderings (e.g.,
# "is_fully_rendered_request" and "num_fully_rendered_requests") answer
# based on the latest _completed_ generation and the internal
# counter that tracks the latest completed fuzzing generation is increased
# after the end of @function render. However, inside the driver main-loop we
# first run @function extend (since initially we start by an empty
# sequence) and then run @function render, and thus, we need to temporarily
# increase the generation counter in order to get a proper behaviour
# when invoking "is_fully_rendered_request" in here after the first iteration
# of the main-loop.
Monitor().current_fuzzing_generation += 1
for req in fuzzing_requests:
for i in range(prev_len):
seq = seq_collection[i]
# Extend sequence collection by adding requests that have
# valid dependencies and skip the rest
if not validate_dependencies(req, seq)\
and not Settings().ignore_dependencies:
continue
req_copy = copy.copy(req)
req_copy._current_combination_id = 0
if seq.is_empty_sequence():
new_seq = sequences.Sequence(req_copy)
else:
new_seq = seq + sequences.Sequence(req_copy)
seq_collection.append(new_seq)
# Append each request to exactly one sequence
if Settings().fuzzing_mode in ['bfs-fast', 'bfs-minimal']:
break
# See comment above...
Monitor().current_fuzzing_generation -= 1
# In case of random walk, truncate sequence collection to
# one randomly selected sequence
if Settings().fuzzing_mode == 'random-walk':
if len(seq_collection) > 0:
rand_int = random.randint(prev_len, len(seq_collection) - 1)
return seq_collection[rand_int:rand_int + 1]
else:
return []
# Drop previous generation and keep current extended generation
return seq_collection[prev_len:]
def generate_sequences_directed_smoketest(fuzzing_requests, checkers):
""" Checks whether each request can be successfully rendered.
For each request:
- Constructs a sequence that satisfies all dependencies by backtracking.
- Renders this sequence.
This allows debugging rendering on a per-request basis
to resolve configuration or spec issues.
"""
def render_request(request, seq):
""" Helper function that attempts to find a valid rendering for the request.
The do-while loop will render each combination of the request until either
a valid rendering is detected or all combinations have been exhausted.
Side effects: request.stats.status_code updated
request.stats.status_text updated
request.stats updated with concrete response and request text
(valid request or last combination)
@return: Tuple containing rendered sequence object, response body, and
failure information enum.
@rtype : Tuple(RenderedSequence, str, FailureInformation)
"""
response_body = None
rendering_information = None
while True:
renderings = seq.render(candidate_values_pool, global_lock)
if renderings.failure_info:
# Even though we will be returning renderings from this function,
# the renderings object that is returned may be from an unrendered
# sequence. We want to save the most recent info.
rendering_information = renderings.failure_info
# Perform this check/save here in case the last call to seq.render
# returns an empty 'renderings' object. An empty renderings object
# will be returned from seq.render if all request combinations are
# exhausted prior to getting a valid status code.
if renderings.final_request_response:
request.stats.status_code = renderings.final_request_response.status_code
request.stats.status_text = renderings.final_request_response.status_text
# Get the last rendered request. The corresponding response should be
# the last received response.
request.stats.sample_request.set_request_stats(
renderings.sequence.sent_request_data_list[-1].
rendered_data)
request.stats.sample_request.set_response_stats(
renderings.final_request_response,
renderings.final_response_datetime)
response_body = renderings.final_request_response.body
apply_checkers(checkers, renderings, global_lock)
# If a valid rendering was found or the combinations have been
# exhausted (empty rendering), exit the loop.
if renderings.valid or renderings.sequence is None:
return renderings, response_body, rendering_information
global_lock = None
candidate_values_pool = GrammarRequestCollection().candidate_values_pool
# print general request-related stats
logger.print_req_collection_stats(
fuzzing_requests,
GrammarRequestCollection().candidate_values_pool)
logger.write_to_main(
f"\n{formatting.timestamp()}: Starting directed-smoke-test\n")
# Sort the request list prior to computing the request sequences,
# so the prefixes are always in the same order for the algorithm
fuzzing_request_list = list(fuzzing_requests._requests)
fuzzing_request_list.sort(key=lambda x: x.method_endpoint_hex_definition)
# sort the requests in fuzzing_requests by depth
sorted_fuzzing_req_list = []
for request in fuzzing_request_list:
req_list = compute_request_goal_seq(request, fuzzing_request_list)
if len(req_list) > 0:
sorted_fuzzing_req_list.append([len(req_list), request, req_list])
# Else an error message was printed and we skip this request
# now sort by length (secondary sort by a hash of the request definition text)
sorted_fuzzing_req_list.sort(
key=lambda x: (x[0], x[1].method_endpoint_hex_definition))
logger.write_to_main(f"{formatting.timestamp()}: Will attempt to render "
f"{len(sorted_fuzzing_req_list)} requests found\n")
# the two following lists are indexed by request number and are of the same size.
# memoize valid rendered sequences for each request and re-use those when going deeper
valid_rendered_sequences_list = []
# memoize the first invalid prefix for each request
first_invalid_prefix_list = []
# try to render all requests starting with the shallow ones
for idx, request_triple in enumerate(sorted_fuzzing_req_list):
req_list_length = request_triple[0]
request = request_triple[1]
req_list = request_triple[2]
valid = False
first_invalid_prefix = -1 # -1 denotes no invalid prefix by default
request.stats.request_order = idx
Found = False
if (req_list_length > 1):
# search for a valid matching prefix we can re-use;
# unless path_regex is used we should always find a match
# because we start with shallow sequences
req_list_prefix = req_list[:-1]
i = 0
while (not Found) and (i < idx):
if sorted_fuzzing_req_list[i][2] == req_list_prefix:
# we found a match
Found = True
logger.write_to_main(
f"Found a matching prefix for request {idx} with previous request {i}"
)
request.stats.matching_prefix[
"id"] = sorted_fuzzing_req_list[i][
1].method_endpoint_hex_definition
else:
# continue searching
i = i + 1
rendering_information = None
response_body = None
if Found:
if valid_rendered_sequences_list[i].is_empty_sequence():
# then the current sequence will also be INVALID.
# propagate the root-cause explaining why the prefix was invalid
first_invalid_prefix = first_invalid_prefix_list[i]
logger.write_to_main(
f"\tbut that prefix was INVALID (root = {first_invalid_prefix})\n"
)
request.stats.matching_prefix["valid"] = 0
# since valid = False by default, nothing else to do here
else:
# re-use the previous VALID prefix
logger.write_to_main("\tand re-using that VALID prefix\n")
request.stats.matching_prefix["valid"] = 1
new_seq = valid_rendered_sequences_list[i]
req_copy = copy.copy(request)
req_copy._current_combination_id = 0
new_seq = new_seq + sequences.Sequence(req_copy)
new_seq.seq_i = 0
renderings, response_body, rendering_information = render_request(
request, new_seq)
valid = renderings.valid
else:
logger.write_to_main(f"Rendering request {idx} from scratch\n")
# render the sequence.
new_seq = sequences.Sequence()
for req in req_list:
req_copy = copy.copy(req)
req_copy._current_combination_id = 0
if new_seq.is_empty_sequence():
new_seq = sequences.Sequence(req_copy)
else:
new_seq = new_seq + sequences.Sequence(req_copy)
new_seq.seq_i = 0
renderings, response_body, rendering_information = render_request(
req, new_seq)
valid = renderings.valid
logger.write_to_main(
f"{formatting.timestamp()}: Request {idx}\n"
f"{formatting.timestamp()}: Endpoint - {request.endpoint_no_dynamic_objects}\n"
f"{formatting.timestamp()}: Hex Def - {request.method_endpoint_hex_definition}\n"
f"{formatting.timestamp()}: Sequence length that satisfies dependencies: {req_list_length}"
)
if valid:
logger.write_to_main(f"{formatting.timestamp()}: Rendering VALID")
request.stats.valid = 1
# remember this valid sequence
valid_rendered_sequences_list.append(new_seq)
first_invalid_prefix_list.append(first_invalid_prefix)
else:
logger.write_to_main(
f"{formatting.timestamp()}: Rendering INVALID")
request.stats.valid = 0
request.stats.error_msg = response_body
# remember RESTler didn't find any valid sequence with an empty request sequence
valid_rendered_sequences_list.append(sequences.Sequence())
if (first_invalid_prefix == -1):
first_invalid_prefix = idx
first_invalid_prefix_list.append(first_invalid_prefix)
if rendering_information:
if rendering_information == FailureInformation.PARSER:
msg = (
"This request received a VALID status code, but the parser failed.\n"
"Because of this, the request was set to INVALID.\n")
elif rendering_information == FailureInformation.RESOURCE_CREATION:
msg = (
"This request received a VALID status code, but the server "
"indicated that there was a failure when creating the resource.\n"
)
elif rendering_information == FailureInformation.SEQUENCE:
msg = (
"This request was never rendered because the sequence failed to re-render.\n"
"Because of this, the request was set to INVALID.\n")
elif rendering_information == FailureInformation.BUG:
msg = "A bug code was received after rendering this request."
else:
msg = "An unknown error occurred when processing this request."
logger.write_to_main(f"{formatting.timestamp()}: {msg}")
request.stats.failure = rendering_information
rendering_information = None
logger.format_rendering_stats_definition(
request,
GrammarRequestCollection().candidate_values_pool)
logger.print_request_rendering_stats(
GrammarRequestCollection().candidate_values_pool, fuzzing_requests,
Monitor(), fuzzing_requests.size_all_requests,
Monitor().current_fuzzing_generation, global_lock)
Monitor().current_fuzzing_generation += 1
return len(valid_rendered_sequences_list)
def render_one(seq_collection, ith, checkers, generation, global_lock):
""" Render ith sequence from sequence collection.
@param seq_collection: List of sequences in sequence collection.
@type seq_collection: List
@param ith: The position of the target sequence (to be rendered) in the
sequence collection.
@type ith: Int
@param checkers: The list of checkers to apply
@type checkers: list[Checker]
@param generation: The fuzzing generation
@type generation: Int
@param global_lock: Lock object used for sync of more than one fuzzing jobs.
@type global_lock: thread.Lock object
@return: The list of sequences with valid renderings.
@rtype : List
Note: Try ith sequence's template with all posible primitive type value
combinations and return only renderings (combinations of primitive type
values) that lead to valid error codes. We keep track of the order of the
current sequence in the collection using "ith" argument for logging
purposes.
"""
# Log memory consumption every hour.
n_minutes = 60
# Static variable used for keeping track of the last time memory consumption was printed
render_one.last_memory_consumption_check = getattr(
render_one, 'last_memory_consumption_check', int(time.time()))
if int(time.time()) - render_one.last_memory_consumption_check > (
n_minutes * 60):
logger.print_memory_consumption(GrammarRequestCollection(), Monitor(),
Settings().fuzzing_mode, generation)
render_one.last_memory_consumption_check = int(time.time())
candidate_values_pool = GrammarRequestCollection().candidate_values_pool
current_seq = seq_collection[ith]
current_seq.seq_i = ith
valid_renderings = []
# Try to find one valid rendering.
n_invalid_renderings = 0
while True:
# Render on a sequence instance will internally iterate over possible
# renderings of current sequence until a valid or an invalid combination
# of values for its primitive types is found -- internal iteration may
# skip some renderings (that are marked to be skipped according to past
# failures) -- that's why we put everything in a while.
renderings = current_seq.render(candidate_values_pool, global_lock)
# Note that this loop will keep running as long as we hit invalid
# renderings and we will end up reapplying the leakage rule a billion
# times for very similar 404s. To control this, when in bfs-cheap, we
# apply the checkers only on the first invalid rendering.
if Settings().fuzzing_mode not in ['bfs-cheap', 'bfs-minimal']\
or renderings.valid or n_invalid_renderings < 1:
apply_checkers(checkers, renderings, global_lock)
# If renderings.sequence is None it means there is nothing left to render.
if renderings.valid or renderings.sequence is None:
break
# This line will only be reached only if we have an invalid rendering.
n_invalid_renderings += 1
# for random-walk and cheap fuzzing, one valid rendering is enough.
if Settings().fuzzing_mode in ['random-walk', 'bfs-cheap', 'bfs-minimal']:
if renderings.valid:
valid_renderings.append(renderings.sequence)
# bfs needs to be exhaustive to provide full grammar coverage
elif Settings().fuzzing_mode in ['bfs', 'bfs-fast']:
# This loop will iterate over possible remaining renderings of the
# current sequence.
while renderings.sequence is not None:
if renderings.valid:
valid_renderings.append(renderings.sequence)
renderings = current_seq.render(candidate_values_pool, global_lock)
apply_checkers(checkers, renderings, global_lock)
else:
print("Unsupported fuzzing_mode:", Settings().fuzzing_mode)
assert False
return valid_renderings
def _exec_request_with_new_body(self,
request,
body_blocks,
tracker,
valid_is_violation=False):
""" Render and send the new request and analyze the response
@param request: Seed request
@type request: Request
@param body_blocks: Definition (request blocks) of the new body
@type body_blocks: List
@param tracker: Response tracker for this run
@type tracker: ResponseTracker
@param valid_is_violation: If valid response is violation
@type valid_is_violation: Bool
@return: None
@rtype: None
"""
# substitute to the original request
new_request = substitute_body(request, body_blocks)
seq = copy(self._sequence)
cnt = 0
# iterate through different value combinations
for rendered_data, parser in new_request.render_iter(
self._req_collection.candidate_values_pool):
# check time budget
if Monitor().remaining_time_budget <= 0:
raise TimeOutException('Exceed Timeout')
# stop fuzzing when reaching the bound
if cnt > int(Settings().max_combinations):
break
cnt += 1
# stop fuzzing when reaching the global bound
if self._global_bound > 0 and self._global_count > self._global_bound:
break
self._global_count += 1
# refresh the sequence to make sure the resource is not garbage collected
if self._refresh_req:
seq = self._refresh(request)
# render the data
rendered_data = seq.resolve_dependencies(rendered_data)
# substitute if there is UUID suffix
original_rendered_data = rendered_data
uuid4_suffix_dict = self._get_custom_payload_uuid4_suffix()
for uuid4_suffix in uuid4_suffix_dict:
suffix = uuid4_suffix_dict[uuid4_suffix]
len_suffix = len(suffix)
# need the query to partition path and body
try:
partition = rendered_data.index('?')
if suffix in rendered_data[:partition]:
new_val_start = rendered_data[:partition].index(suffix)
if new_val_start + len_suffix + 10 > partition:
self._log('unexpected uuid')
continue
new_val = rendered_data[new_val_start:new_val_start +
len_suffix + 10]
# find all occurence in the body
suffix_in_body = [
m.start()
for m in re.finditer(suffix, rendered_data)
][1:]
for si in suffix_in_body:
old_val = rendered_data[si:si + len_suffix + 10]
rendered_data = rendered_data.replace(
old_val, new_val)
except Exception:
rendered_data = original_rendered_data
# send out the request
response = self._send_request(parser, rendered_data)
request_utilities.call_response_parser(parser, response)
self._set_refresh_req(request, response)
if not response or not response.status_code:
self._log('ERROR: no response received')
continue
# analyze response -- coverage
tracker.process_response(response)
if self._acc_response:
hints = self._map_response_to_current_body_schema(response)
for tag in hints:
self._response_values[tag] = hints[tag]
# analyze response -- error
if self._rule_violation(seq, response, valid_is_violation):
# Append the new request to the sequence before filing the bug
seq.replace_last_sent_request_data(rendered_data, parser,
response)
err_seq = sequences.Sequence(seq.requests[:-1] + [new_request])
err_seq.set_sent_requests_for_replay(
seq.sent_request_data_list)
self._print_suspect_sequence(err_seq, response)
bug_info = self._buckets.add_bug(request, rendered_data)
if bug_info is not None:
error_str = bug_info[0]
new_body = bug_info[1]
log_str = f'{error_str}\n{new_body}'
BugBuckets.Instance().update_bug_buckets(
err_seq,
response.status_code,
origin=self.__class__.__name__,
checker_str=error_str,
additional_log_str=log_str)
self._refresh_req = True
def render(self,
candidate_values_pool,
lock,
preprocessing=False,
postprocessing=False):
""" Core routine that performs the rendering of restler sequences. In
principal all requests of a sequence are being constantly rendered with
a specific values combination @param request._current_combination_id
which we know in the past led to a valid rendering and only the last
request of the sequence is being rendered iteratively with all feasible
value combinations. Each time a "valid rendering" is found for the last
request of the sequence (where "valid rendering" is defined according
to "VALID_CODES"), the routine returns a new sequence which has an
end-to-end (i.e., all requests) "valid rendering" and can be added in
the sequences collection in order to be used in the future as a building
block for longer sequences.
@param candidate_values_pool: The pool of values for primitive types.
@type candidate_values_pool: Dict
@param lock: Lock object used for sync of more than one fuzzing jobs.
@type lock: thread.Lock object
@param preprocessing: Set to true if rendering during preprocessing
@type preprocessing: Bool
@return: A RenderedSequence object containing the sequence, the final
request's response, whether or not the final request received
a valid status code, and a FailureInformation enum if there was
a failure or bug detected during rendering.
@rtype : RenderedSequence
"""
# Try rendering all primitive type value combinations for last request
request = self.last_request
# for clarity reasons, don't log requests whose render iterator is over
if request._current_combination_id <\
request.num_combinations(candidate_values_pool):
CUSTOM_LOGGING(self, candidate_values_pool)
self._sent_request_data_list = []
for rendered_data, parser in\
request.render_iter(candidate_values_pool,
skip=request._current_combination_id,
preprocessing=preprocessing):
# Hold the lock (because other workers may be rendering the same
# request) and check whether the current rendering is known from the
# past to lead to invalid status codes. If so, skip the current
# rendering.
if lock is not None:
lock.acquire()
should_skip = Monitor().is_invalid_rendering(request)
if lock is not None:
lock.release()
# Skip the loop and don't forget to increase the counter.
if should_skip:
RAW_LOGGING("Skipping rendering: {}".\
format(request._current_combination_id))
request._current_combination_id += 1
continue
# Clean up internal state
self.status_codes = []
dependencies.reset_tlb()
sequence_failed = False
# Step A: Static template rendering
# Render last known valid combination of primitive type values
# for every request until the last
for i in range(len(self.requests) - 1):
prev_request = self.requests[i]
prev_rendered_data, prev_parser =\
prev_request.render_current(candidate_values_pool,
preprocessing=preprocessing)
# substitute reference placeholders with resolved values
if not Settings().ignore_dependencies:
prev_rendered_data =\
self.resolve_dependencies(prev_rendered_data)
prev_req_async_wait = Settings(
).get_max_async_resource_creation_time(prev_request.request_id)
prev_producer_timing_delay = Settings(
).get_producer_timing_delay(prev_request.request_id)
prev_response = request_utilities.send_request_data(
prev_rendered_data)
prev_response_to_parse, resource_error, async_waited = async_request_utilities.try_async_poll(
prev_rendered_data, prev_response, prev_req_async_wait)
prev_parser_threw_exception = False
# Response may not exist if there was an error sending the request or a timeout
if prev_parser and prev_response_to_parse:
prev_parser_threw_exception = not request_utilities.call_response_parser(
prev_parser, prev_response_to_parse, prev_request)
prev_status_code = prev_response.status_code
# If the async logic waited for the resource, this wait already included the required
# producer timing delay. Here, set the producer timing delay to zero, so this wait is
# skipped both below for this request and during replay
if async_waited:
prev_producer_timing_delay = 0
else:
prev_req_async_wait = 0
self.append_data_to_sent_list(prev_rendered_data, prev_parser,
prev_response,
prev_producer_timing_delay,
prev_req_async_wait)
if not prev_status_code:
logger.write_to_main(
f"Error: Failed to get status code during valid sequence re-rendering.\n"
)
sequence_failed = True
break
if prev_response.has_bug_code():
BugBuckets.Instance().update_bug_buckets(self,
prev_status_code,
reproduce=False,
lock=lock)
sequence_failed = True
break
if prev_parser_threw_exception:
logger.write_to_main(
"Error: Parser exception occurred during valid sequence re-rendering.\n"
)
sequence_failed = True
break
if resource_error:
logger.write_to_main(
"Error: The resource was left in a Failed state after creation during valid sequence re-rendering.\n"
)
sequence_failed = True
break
# If the previous request is a resource generator and we did not perform an async resource
# creation wait, then wait for the specified duration in order for the backend to have a
# chance to create the resource.
if prev_producer_timing_delay > 0 and prev_request.is_resource_generator(
):
print(
f"Pausing for {prev_producer_timing_delay} seconds, request is a generator..."
)
time.sleep(prev_producer_timing_delay)
# register latest client/server interaction
timestamp_micro = int(time.time() * 10**6)
self.status_codes.append(
status_codes_monitor.RequestExecutionStatus(
timestamp_micro,
prev_request.hex_definition, prev_status_code,
prev_response.has_valid_code(), False))
if sequence_failed:
self.status_codes.append(
status_codes_monitor.RequestExecutionStatus(
int(time.time() * 10**6), request.hex_definition,
RESTLER_INVALID_CODE, False, True))
Monitor().update_status_codes_monitor(self, self.status_codes,
lock)
return RenderedSequence(
failure_info=FailureInformation.SEQUENCE)
# Step B: Dynamic template rendering
# substitute reference placeholders with ressoved values
# for the last request
if not Settings().ignore_dependencies:
rendered_data = self.resolve_dependencies(rendered_data)
# Render candidate value combinations seeking for valid error codes
request._current_combination_id += 1
req_async_wait = Settings().get_max_async_resource_creation_time(
request.request_id)
response = request_utilities.send_request_data(rendered_data)
response_to_parse, resource_error, _ = async_request_utilities.try_async_poll(
rendered_data, response, req_async_wait)
parser_exception_occurred = False
# Response may not exist if there was an error sending the request or a timeout
if parser and response_to_parse:
parser_exception_occurred = not request_utilities.call_response_parser(
parser, response_to_parse, request)
status_code = response.status_code
if not status_code:
return RenderedSequence(None)
self.append_data_to_sent_list(rendered_data,
parser,
response,
max_async_wait_time=req_async_wait)
rendering_is_valid = not parser_exception_occurred\
and not resource_error\
and response.has_valid_code()
# register latest client/server interaction and add to the status codes list
response_datetime = datetime.datetime.now(datetime.timezone.utc)
timestamp_micro = int(response_datetime.timestamp() * 10**6)
self.status_codes.append(
status_codes_monitor.RequestExecutionStatus(
timestamp_micro, request.hex_definition, status_code,
rendering_is_valid, False))
# add sequence's error codes to bug buckets.
if response.has_bug_code():
BugBuckets.Instance().update_bug_buckets(self,
status_code,
lock=lock)
Monitor().update_status_codes_monitor(self, self.status_codes,
lock)
# Register current rendering's status.
if lock is not None:
lock.acquire()
Monitor().update_renderings_monitor(request, rendering_is_valid)
if lock is not None:
lock.release()
if Monitor().remaining_time_budget <= 0 and not postprocessing:
raise TimeOutException("Exceeded Timeout")
if lock is not None:
lock.acquire()
# Deep copying here will try copying anything the class has access
# to including the shared client monitor, which we update in the
# above code block holding the lock, but then we release the
# lock and one thread can be updating while another is copying.
# This is a typlical nasty read after write syncronization bug.
duplicate = copy.deepcopy(self)
if lock is not None:
lock.release()
datetime_format = "%Y-%m-%d %H:%M:%S"
# return a rendered clone if response indicates a valid status code
if rendering_is_valid or Settings().ignore_feedback:
return RenderedSequence(
duplicate,
valid=True,
final_request_response=response,
response_datetime=response_datetime.strftime(
datetime_format))
else:
information = None
if response.has_valid_code():
if parser_exception_occurred:
information = FailureInformation.PARSER
elif resource_error:
information = FailureInformation.RESOURCE_CREATION
elif response.has_bug_code():
information = FailureInformation.BUG
return RenderedSequence(
duplicate,
valid=False,
failure_info=information,
final_request_response=response,
response_datetime=response_datetime.strftime(
datetime_format))
return RenderedSequence(None)
def delete_create_once_resources(destructors, fuzzing_requests):
""" Iterates through each destructor request and sends it to the server
@param destructors: A list of destructor requests to send
@type destructors: list(Request)
@param fuzzing_requests: The global collection of requests to fuzz
@type fuzzing_requests: FuzzingRequestCollection
@return: None
@rtype : None
"""
if not destructors:
return
candidate_values_pool = GrammarRequestCollection().candidate_values_pool
logger.write_to_main("\nRendering for create-once resource destructors:\n")
for destructor in destructors:
status_codes = []
try:
logger.write_to_main(
f"{formatting.timestamp()}: Endpoint - {destructor.endpoint_no_dynamic_objects}"
)
logger.write_to_main(
f"{formatting.timestamp()}: Hex Def - {destructor.method_endpoint_hex_definition}"
)
seq = sequences.Sequence([destructor])
renderings = seq.render(
GrammarRequestCollection().candidate_values_pool,
None,
postprocessing=True)
if not renderings.valid:
logger.write_to_main(
f"{formatting.timestamp()}: Rendering INVALID")
else:
logger.write_to_main(
f"{formatting.timestamp()}: Rendering VALID")
logger.format_rendering_stats_definition(
destructor,
GrammarRequestCollection().candidate_values_pool)
if Settings().in_smoke_test_mode():
destructor.stats.request_order = 'Postprocessing'
destructor.stats.valid = 1
destructor.stats.status_code = renderings.final_request_response.status_code
destructor.stats.status_text = renderings.final_request_response.status_text
destructor.stats.sample_request.set_request_stats(
renderings.sequence.sent_request_data_list[-1].
rendered_data)
destructor.stats.sample_request.set_response_stats(
renderings.final_request_response,
renderings.final_response_datetime)
except Exception as error:
msg = f"Failed to delete create_once resource: {error!s}"
logger.raw_network_logging(msg)
logger.write_to_main(msg, print_to_console=True)
if Settings().in_smoke_test_mode():
destructor.stats.request_order = 'Postprocessing'
destructor.stats.valid = 0
if renderings and renderings.final_request_response:
destructor.stats.status_code = renderings.final_request_response.status_code
destructor.stats.status_text = renderings.final_request_response.status_text
destructor.stats.error_msg = renderings.final_request_response.body
destructor.stats.sample_request.set_request_stats(
renderings.sequence.sent_request_data_list[-1].
rendered_data)
destructor.stats.sample_request.set_response_stats(
renderings.final_request_response,
renderings.final_response_datetime)
pass
Monitor().current_fuzzing_generation += 1
logger.print_request_rendering_stats(candidate_values_pool,
fuzzing_requests, Monitor(),
fuzzing_requests.size_all_requests,
logger.POSTPROCESSING_GENERATION,
None)
