minhhungg's picture
feat: initial upload for AutoRestTest Track A datasets
6c50d1f
Raw
History Blame Contribute Delete
29.9 kB
import json
import time
import uuid
from datetime import datetime
import sys
import numpy as np
import requests
from model.operation_dependency_graph import OperationDependencyGraph
from validator.validator import validate
from validator.validator_payload import validate_payload
from .normal_executor import SequenceConverter
from .mutation_executor import mutationSequenceConverter
from .runtime_dictionary import RuntimeDictionary
from model.sequence import SequenceOrigin
class APIFuzzer:
def __init__(self, apis=[], specification={}, odg: OperationDependencyGraph = OperationDependencyGraph(),
host_address="", pre_defined_headers={},
time_budget=300):
self.runtime_dict = RuntimeDictionary()
self.apis = apis
self.server_address = host_address
self.protocol = "HTTPS"
self.odg = odg
self.sequences = odg.generate_sequence()
self.odg_sequences = set(self.sequences)
self.specification = specification
self.start_time = time.time()
# for test case limit time
self.time_limit = 2000
self.end_time = self.start_time + time_budget
self.seed_id = uuid.uuid4().__str__()
self.sequence_id = uuid.uuid4().__str__()
self.iteration_id = 0
self.request_count = 0
self.response_count = 0
self.violations = set()
self.error_sequence = []
self.violations_sequence = []
self.pending_add_sequence = set()
self.pending_remove_sequence = set()
self.pending_test_sequence_to_remove = set()
self.status_code_status = {}
self.time_budget = time_budget
self.send_msg_count = 0
self.receive_msg_count = 0
self.success_sequence_count = 0
self.success_apis = set()
self.error_apis = set()
self.client_apis = set()
self.total_apis = set()
self.success_sequence = set()
self.begin = time.time()
self.success_sequence_output = []
self.success_endpoint = set()
self.error_endpoint = set()
self.client_endpoint = set()
self.error_api_curve = []
self.client_api_curve = []
self.success_endpoint_api_curve = []
self.error_endpoint_api_curve = []
self.client_endpoint_api_curve = []
self.pre_defined_headers = pre_defined_headers
self.total_apis_map = {}
self.api_curve = [{
"time": 0,
'count': 0
}]
def run(self):
# Collect All API
for api in self.apis:
for method in api.methods:
self.total_apis.add(method)
self.total_apis_map[method.method_signature] = method
executor = SequenceConverter(self.runtime_dict)
self.begin = time.time()
begin = self.begin
time_budget = self.time_budget
previous_success_size = 0
while time.time() - begin < time_budget:
# Phase 1
for sequence in self.sequences:
try_times = 1 if len(sequence) > 1 else 20
for _ in range(try_times):
res = executor.execute_sequence(self, sequence)
self.request_count += len(res)
self.send_msg_count += 1
self.process_response(res, sequence)
if time.time() - begin > time_budget:
break
if time.time() - begin > time_budget:
break
self.iteration_id += 1
# remove test sequence
self.sequences = self.sequences.difference(self.pending_test_sequence_to_remove)
self.pending_test_sequence_to_remove = set()
# remove sequences
self.sequences = self.sequences.difference(self.pending_remove_sequence)
self.success_sequence = self.success_sequence.union(self.pending_remove_sequence)
self.pending_remove_sequence = set()
# add sequences
self.sequences = self.sequences.union(self.pending_add_sequence)
self.pending_add_sequence = set()
# analyze dependency to add
self.analyze_dependency_to_add()
print('current seq')
for seq in self.sequences:
print(seq.to_str())
# Phase 2
# if previous_success_size != len(self.success_sequence) and np.random.random() > 0.95:
# previous_success_size = len(self.success_sequence)
# new_synthesis_sequence = self.synthesis_resource_sequence(executor)
# self.sequences = self.sequences.union(new_synthesis_sequence)
if len(self.sequences) == 0:
self.sequences = set(self.odg_sequences)
self.write_result()
# executor.draw_response_dict()
def mutation_fuzz_run(self):
print("Continue to mutation fuzzing \n-------------------- \n")
time.sleep(3)
# re_initialize the fuzzer
self.runtime_dict = RuntimeDictionary()
self.mutation_runtime_dict = RuntimeDictionary()
# re-initialize error sequence
self.error_sequence = []
# create a pair of normal sequence executor and mutation sequence executor
normalExecutor = SequenceConverter(self.runtime_dict)
mutationExecutor = mutationSequenceConverter(self.mutation_runtime_dict)
mutation_time_budget = self.time_budget # make it a time budget parameter later
mutation_fuzz_begin = time.time()
while time.time() - mutation_fuzz_begin < mutation_time_budget:
for sequence in self.success_sequence:
# check pattern
# TODO: update sequence checking
path_to_check = ["post get"]
req_methods_path = ''
for i, method in enumerate(sequence.requests):
req_methods_path += method.method_type + ' '
proceed = False
for path in path_to_check:
if path in req_methods_path:
proceed = True
if not proceed:
continue
for mutation_index, method in enumerate(sequence.requests):
baseline_res = normalExecutor.execute_sequence(self, sequence)
res = mutationExecutor.execute_sequence(self, sequence,
mutation_index) # i labels the position of mutation
self.request_count += len(res)
self.send_msg_count += 1
self.process_mutation_response(baseline_res, res, sequence, mutation_index)
self.write_mutation_result()
print(self.violations)
def synthesis_resource_sequence(self, executor):
print('synthesis resource')
add_sequence_size = 100
matched_pair = {}
success_sequence = list(self.success_sequence)
np.random.shuffle(success_sequence)
success_sequence = success_sequence[:add_sequence_size]
new_sequences = set()
graph = executor.response_resource_graph.get("graph")
# extract method pair
for method_pair in graph.keys():
first_method, second_method = method_pair
first_method_set = matched_pair.get(first_method, set())
second_method_set = matched_pair.get(second_method, set())
matched_pair[first_method] = first_method_set
matched_pair[second_method] = second_method_set
first_method_set.add(method_pair)
second_method_set.add(method_pair)
# iterate each sequence
for sequence in success_sequence:
# omit single element sequence
if len(sequence) == 1:
continue
# we do not want to add more sequences in runtime
if np.random.random() < 0.99:
continue
for i in range(len(sequence)):
current_method = sequence[i]
# omit the last one
if i == len(sequence) - 1:
continue
# go through the
if not (current_method.method_signature in matched_pair):
continue
# by chance to omit current method
# if np.random.random() < 0.5:
# continue
# here we go through the pairs
for method_pair in matched_pair[current_method.method_signature]:
pair_method = None
if method_pair[0] == current_method.method_signature:
pair_method = self.total_apis_map[method_pair[1]]
else:
pair_method = self.total_apis_map[method_pair[0]]
duplicate_sequence = sequence.duplicate()
# Change method
duplicate_sequence.set_method(i, pair_method)
# Check reference
matched_reference_pairs = graph[method_pair]
ref = duplicate_sequence.get_ref(i + 1).get_refs()
for refer_pair in matched_reference_pairs:
first_attribute, second_attribute = refer_pair
if first_attribute in ref:
ref[second_attribute] = ref[first_attribute]
ref.__delitem__(first_attribute)
if second_attribute in ref:
ref[first_attribute] = ref[second_attribute]
ref.__delitem__(second_attribute)
duplicate_sequence.origin = SequenceOrigin.RESOURCE_MERGING
sequence_requests = set()
is_valid = True
# sequence should not contains the same request
for seq_request in duplicate_sequence.requests:
if sequence_requests.__contains__(seq_request):
is_valid = False
break
else:
sequence_requests.add(seq_request)
if not is_valid:
continue
new_sequences.add(duplicate_sequence)
duplicate_sequence.sequence_trace.append([sequence.origin, str(sequence)])
return new_sequences
def write_brief_result(self, target_dir):
orig_stdout = sys.stdout
with open(target_dir + '/errors.json', 'w') as data:
json.dump({"result": self.error_sequence}, data)
with open(target_dir + '/violations_sequence.json', 'w') as data:
json.dump({"result": self.violations_sequence}, data)
f = open(target_dir + '/brief.text', 'w')
sys.stdout = f
self.overall_status()
sys.stdout = orig_stdout
f.close()
def write_result(self):
import json
with open('errors.json', 'w') as data:
json.dump({
"number of errors": len(self.error_sequence),
"result": self.error_sequence
}, data)
with open('success_resource_merging.json', 'w') as data:
json.dump({
"number of success sequence output": len(self.success_sequence_output),
"result": self.success_sequence_output
}, data)
with open('client_errors.json', 'w') as data:
json.dump({
"number of client APIs": len(self.client_apis),
"result": list(self.client_apis)
}, data)
with open('runtime.json', 'w') as data:
json.dump({
"number of keys": len(self.runtime_dict.signature_to_value.keys()),
'keys': list(self.runtime_dict.signature_to_value.keys())
}, data)
def write_mutation_result(self):
with open('violations_sequence.json', 'w') as data:
json.dump({"result": self.violations_sequence}, data)
def analyze_dependency_to_add(self):
generated_sequences = set()
if len(self.sequences) > len(self.apis):
return
if np.random.random() > 0.05:
return
success_sequences = list(self.success_sequence)
np.random.shuffle(success_sequences)
for sequence in self.sequences:
# give chance to add longer sequence
if len(sequence) > 1 and np.random.random() > 0.01:
continue
request_nominal_parameters = sequence.get_request_parameter_by_index(0)
response_nominal_runtime_parameters = self.runtime_dict.signature_to_value.keys()
matched_methods = {}
request_parameter_to_methods = {}
method_recorded_response = {}
method_to_request = {}
for request_parameter in request_nominal_parameters:
chunked_token = request_parameter.lower().split('.')[-1]
for response in response_nominal_runtime_parameters:
method_name, response_parameter = response.split(self.runtime_dict.signature_splitter)
chunked_response_token = response_parameter.lower().split('.')[-1]
if chunked_token != chunked_response_token:
continue
matched_parameters = matched_methods.get(method_name, set())
matched_parameters.add(response_parameter)
matched_methods[method_name] = matched_parameters
request_parameter_to_meth = request_parameter_to_methods.get(request_parameter, set())
request_parameter_to_meth.add(method_name)
request_parameter_to_methods[request_parameter] = request_parameter_to_meth
method_to_response = method_recorded_response.get(method_name, set())
method_to_response.add(response_parameter)
method_recorded_response[method_name] = method_to_response
method_to_req = method_to_request.get(method_name, set())
method_to_req.add((request_parameter, response_parameter))
method_to_request[method_name] = method_to_req
# for success_sequence in self.success_sequence:
# if np.random.random() < 0.5:
# break
# if not success_sequence.has_method(method_name):
# continue
# new_seq = success_sequence.slice_by_method_name(method_name)
# new_seq.add_def(len(new_seq) - 1, response_parameter)
# new_seq.add_method(sequence.requests[0])
# new_seq.add_ref(len(new_seq) - 1, request_parameter, response_parameter)
# res.add(new_seq){'placeOrder', 'updateUser', 'createUser'}
# print('add ref in sequence', sequence.requests[0].method_name, "to", method_name,
# f'via {response_parameter} -> {request_parameter}', "new sequence",
# len(new_seq))
for method in method_recorded_response:
matched_seq = None
for success_seq in success_sequences:
if success_seq.has_method(method):
matched_seq = success_seq
break
if matched_seq:
request_parameters = method_to_request[method]
response_parameters = method_recorded_response[method]
new_seq = success_seq.slice_by_method_name(method)
for response_parameter in response_parameters:
new_seq.add_def(len(new_seq) - 1, response_parameter)
new_seq.add_method(sequence.requests[0])
new_seq.origin = SequenceOrigin.DYNAMIC_FIX_ADD
for req, res in request_parameters:
new_seq.add_ref(len(new_seq) - 1, new_seq[len(new_seq) - 1], req, res)
new_seq.sequence_trace.append([success_seq.origin, str(success_sequences)])
sequence_requests = set()
is_valid = True
# sequence should not contains the same request
for seq_request in new_seq.requests:
if sequence_requests.__contains__(seq_request):
is_valid = False
break
else:
sequence_requests.add(seq_request)
if not is_valid:
break
generated_sequences.add(new_seq)
print('add ref in sequence', sequence.requests[0].method_name, "to", method,
f'via {request_parameters}', "new sequence",
len(new_seq))
break
# print('request_parameter_to_methods')
# print(request_parameter_to_methods)
# print('matched_methods')
# print(matched_methods)
# time.sleep(20)
self.sequences = self.sequences.union(generated_sequences)
def process_response(self, resp, sequence):
violations = validate(resp, self.apis)
has_error = False
# FIXME: temp: check 500
for ind, response in enumerate(resp):
status_code = str(response["statusCode"])
self.runtime_dict.parse(sequence[ind], response)
if int(status_code) > 499 and int(status_code) < 600 and not has_error:
self.error_sequence.append({
"testcase": resp
})
has_error = True
result_count = self.status_code_status.get(str(status_code), 0)
self.status_code_status[status_code] = result_count + 1
# analyze sequence
self.analysis_sequence(resp, sequence)
# traverse violation
for violation in violations:
signature = violation.signature()
self.violations.add(signature)
# try to get success and error api status
is_all_success = True
for response_ind, response in enumerate(resp):
status_code = response["statusCode"]
if status_code > 299:
is_all_success = False
if 199 < status_code < 300:
if not self.success_apis.__contains__(response['apiName']):
self.api_curve.append({
"time": time.time() - self.begin,
'count': len(self.success_apis) + 1
})
method_path = sequence[response_ind].method_path
if not self.success_endpoint.__contains__(method_path):
self.success_endpoint_api_curve.append({
"time": time.time() - self.begin,
'count': len(self.success_endpoint) + 1
})
print(method_path, 'method path')
self.success_endpoint.add(method_path)
self.success_apis.add(response["apiName"])
if 399 < status_code < 500:
if not self.client_apis.__contains__(response['apiName']):
self.client_api_curve.append({
"time": time.time() - self.begin,
'count': len(self.client_apis) + 1
})
method_path = sequence[response_ind].method_path
if not self.client_endpoint.__contains__(method_path):
self.client_endpoint_api_curve.append({
"time": time.time() - self.begin,
'count': len(self.client_endpoint) + 1
})
self.client_endpoint.add(method_path)
self.client_apis.add(response["apiName"])
if 499 < status_code < 600:
if not self.error_apis.__contains__(response['apiName']):
self.error_api_curve.append({
"time": time.time() - self.begin,
'count': len(self.error_apis) + 1
})
method_path = sequence[response_ind].method_path
if not self.error_endpoint.__contains__(method_path):
self.error_endpoint_api_curve.append({
"time": time.time() - self.begin,
'count': len(self.error_endpoint) + 1
})
self.error_endpoint.add(method_path)
self.error_apis.add(response['apiName'])
if is_all_success:
self.success_sequence_count += 1
self.success_sequence_output.append({
"testcase": resp,
"source": sequence.origin,
"testcase_length": len(sequence),
"sequence_trace": sequence.sequence_trace,
"sequence": str(sequence)
})
self.receive_msg_count += 1
self.response_count += len(resp)
self.overall_status()
def process_mutation_response(self, baseline_resp, resp, sequence, mutation_index):
violations = validate_payload(baseline_resp, resp, mutation_index, self.apis)
has_error = False
# FIXME: temp: check 500
for ind, response in enumerate(resp):
status_code = str(response["statusCode"])
self.runtime_dict.parse(sequence[ind], response)
if int(status_code) > 499 and int(status_code) < 600 and not has_error:
self.error_sequence.append({
"testcase": resp
})
has_error = True
result_count = self.status_code_status.get(str(status_code), 0)
self.status_code_status[status_code] = result_count + 1
# analyze sequence
self.analysis_sequence(resp, sequence)
# traverse violation
for violation in violations:
signature = violation.signature()
self.violations.add(signature)
# store violation result
if len(violations) != 0:
current_violation_signatures = set()
for violation in violations:
signature = violation.signature()
current_violation_signatures.add(signature)
self.violations_sequence.append({
"violation": str(current_violation_signatures),
"testcase": resp,
"baseline": baseline_resp
})
# try to get success and error api status
is_all_success = True
for response in resp:
status_code = response["statusCode"]
if status_code > 299:
is_all_success = False
if status_code > 199 and status_code < 300:
if not self.success_apis.__contains__(response['apiName']):
self.api_curve.append({
"time": time.time() - self.begin,
'count': len(self.success_apis) + 1
})
self.success_apis.add(response["apiName"])
if status_code > 499 and status_code < 600:
self.error_apis.add(response['apiName'])
if is_all_success:
self.success_sequence_count += 1
self.success_sequence_output.append({
"testcase": resp
})
self.receive_msg_count += 1
self.response_count += len(resp)
self.overall_status() # verbose set to true/false to print status
def has_success_api(self, responses):
for resp in responses:
status_code = resp["statusCode"]
if status_code < 300 and status_code > 199:
return True
return False
def chunk_responses(self, responses):
success_sequence = []
fail_sequence = []
current_fail_sequence = []
current_success_sequence = []
# handle whole error status
if not self.has_success_api(responses):
for i in range(len(responses)):
fail_sequence.append([i])
return success_sequence, fail_sequence
for index in range(len(responses)):
status_code = responses[index]["statusCode"]
if status_code < 300 and status_code > 199:
current_success_sequence.append(index)
if len(current_fail_sequence) > 0:
fail_sequence.append(current_fail_sequence)
current_fail_sequence = []
else:
current_fail_sequence.append(index)
if len(current_success_sequence) > 0:
success_sequence.append(current_success_sequence)
current_success_sequence = []
if len(current_success_sequence) > 0:
success_sequence.append(current_success_sequence)
else:
fail_sequence.append(current_fail_sequence)
return success_sequence, fail_sequence
def analysis_sequence(self, responses, test_sequence):
success_converted_sequences = []
failed_converted_sequences = []
success_sequences, fail_sequences = self.chunk_responses(responses)
assert test_sequence.origin is not None
# convert sequence
for seq in success_sequences:
new_seq = test_sequence.sub_sequence(seq)
if len(seq) == len(test_sequence):
new_seq.origin = test_sequence.origin
else:
new_seq.origin = SequenceOrigin.DYNAMIC_FIX_DELETE
new_seq.sequence_trace.append([test_sequence.origin, str(test_sequence)])
success_converted_sequences.append(new_seq)
for seq in fail_sequences:
new_seq = test_sequence.sub_sequence(seq)
new_seq.origin = SequenceOrigin.DYNAMIC_FIX_DELETE
new_seq.sequence_trace.append([test_sequence.origin, str(test_sequence)])
failed_converted_sequences.append(new_seq)
self.pending_add_sequence = self.pending_add_sequence.union(failed_converted_sequences)
self.pending_remove_sequence = self.pending_remove_sequence.union(success_converted_sequences)
self.pending_test_sequence_to_remove.add(test_sequence)
def overall_status(self):
status_stat = ""
elapsed = time.time() - self.begin
for status_code in self.status_code_status.keys():
status_stat += f'{status_code}:{self.status_code_status[status_code]},' \
f'{float(self.status_code_status[status_code]) / self.response_count} '
# collect the sequence source
sequence_source = {}
for seq in self.success_sequence:
sequence_source[seq.origin] = sequence_source.get(seq.origin, 0) + 1
print(
f'{datetime.now().strftime("%Y/%m/%d %H:%M:%S")}, Fuzzing Time: {elapsed}s, Send Msg Count: {self.send_msg_count}, '
f'Receive Msg Count: {self.receive_msg_count}, Success Response Sequence Count: {self.success_sequence_count}, '
f'Success Sequence Rate: {float(self.success_sequence_count) / self.receive_msg_count}, '
f'Success API: {float(len(self.success_apis)) / len(self.total_apis)}'
f' ({len(self.success_apis)}/{len(self.total_apis)}),'
f'Error API: {float(len(self.error_apis)) / len(self.total_apis)}'
f' ({len(self.error_apis)}/{len(self.total_apis)}),'
#f'Client Error API: {float(len(self.client_apis)) / len(self.total_apis)}'
#f' ({len(self.client_apis)}/{len(self.total_apis)}),'
f'Success Endpoint: {float(len(self.success_endpoint)) / len(self.apis)}'
f' ({len(self.success_endpoint)}/{len(self.apis)}),'
f'Error Endpoint: {float(len(self.error_endpoint)) / len(self.apis)}'
f' ({len(self.error_endpoint)}/{len(self.apis)}),'
#f'Client Error Endpoint: {float(len(self.client_endpoint)) / len(self.apis)}'
#f' ({len(self.client_endpoint)}/{len(self.apis)}),'
f' Tested Error and Success Unique API: {float(len(set.union(self.success_apis, self.error_apis))) / len(self.total_apis)}'
f' ({len(set.union(self.success_apis, self.error_apis))}/{len(self.total_apis)}),'
f' Request Count: {self.request_count}, '
f'Response Count: {self.response_count}, Response / Request: {float(self.response_count) / self.request_count}, '
f'Request(QPS): {float(self.request_count) / elapsed}, Response(QPS):'
f' {float(self.response_count) / elapsed}, Violation Count: {len(self.violations)}'
f' Remain Sequence Count: {len(self.sequences)}'
f' Success Sequence: {len(self.success_sequence)}')
print('success')
print(self.success_apis)
print('errors')
print(self.error_apis)
print('client errors')
print(self.client_apis)
print('==API Curve')
print({
"success_api": self.api_curve,
"error_api": self.error_api_curve,
#"client_api": self.client_api_curve,
"success_endpoint": self.success_endpoint_api_curve,
"error_endpoint": self.error_endpoint_api_curve,
#"client_endpoint": self.client_endpoint_api_curve
})
print(self.runtime_dict.signature_to_value.keys())
print(status_stat)
# print(sequence_source)
print("sequence source", sequence_source)