mirror of
https://github.com/AFLplusplus/AFLplusplus.git
synced 2025-06-12 01:58:17 +00:00
Mutation Chain Tool (#2281)
* Mutation Chain Tool * Address comments domenukk * Address comments domenukk 2
This commit is contained in:
@ -56,6 +56,10 @@ Here's a quick overview of the stuff you can find in this directory:
|
||||
|
||||
- libpng_no_checksum - a sample patch for removing CRC checks in libpng.
|
||||
|
||||
- mutation_chain - a tool that backtraces the mutation chain of AFL
|
||||
crash files based on the crash/queue file naming
|
||||
standards and outputs this in a json format.
|
||||
|
||||
- persistent_mode - an example of how to use the LLVM persistent process
|
||||
mode to speed up certain fuzzing jobs.
|
||||
|
||||
|
154
utils/mutation_chain/mutation_chain.py
Normal file
154
utils/mutation_chain/mutation_chain.py
Normal file
@ -0,0 +1,154 @@
|
||||
#!/usr/bin/env python3
|
||||
#
|
||||
# MUTATION CHAIN COMPUTATION TOOL
|
||||
#
|
||||
# Ever wondered what the complete history of your AFL crash file looks like?
|
||||
# Now you can!
|
||||
#
|
||||
# This tool is developed to support file structures for parallel fuzzing runs using the
|
||||
# naming of main/secondary nodes as stated in the AFL docs (fuzzer01, fuzzer02 etc...)
|
||||
# In case you want to use it for single node runs just recreate the directory structure
|
||||
# which is used when parallel fuzzing is used (dump your results in a dir called fuzzer01).
|
||||
#
|
||||
# author: Maarten Dekker
|
||||
|
||||
# import required modules
|
||||
import os, re, json
|
||||
import argparse
|
||||
|
||||
crashes = {}
|
||||
queues = {}
|
||||
|
||||
def fillDictWithFilenameKeys(dir):
|
||||
dict = {}
|
||||
for filename in os.listdir(dir):
|
||||
if re.match("^id:\\d+", filename):
|
||||
dict[filename] = None
|
||||
return dict
|
||||
|
||||
# recusively compute the chain of queue items that led to the AFL crash file
|
||||
def compute_mutation_chain(filename, current_fuzzer, n):
|
||||
|
||||
if re.match(".*src:(\\d+),", filename):
|
||||
|
||||
source_id = re.match(".*src:(\\d+),", filename).group(1)
|
||||
file_we_look_for_rex = "^id:" + source_id + ","
|
||||
|
||||
fuzzer_queue = None
|
||||
|
||||
# determine if we need to look in the queue of another fuzzer instance
|
||||
if re.match(".*sync:(fuzzer\\d+),", filename):
|
||||
fuzzer_queue = re.match(".*sync:(fuzzer\\d+),", filename).group(1)
|
||||
else:
|
||||
fuzzer_queue = current_fuzzer
|
||||
|
||||
for k,v in queues[fuzzer_queue].items():
|
||||
|
||||
if re.match(file_we_look_for_rex, k):
|
||||
|
||||
retval = {}
|
||||
retval[k] = compute_mutation_chain(k, fuzzer_queue, n+1)
|
||||
return retval
|
||||
|
||||
# if the mutation result is a splice it thas 2 sources
|
||||
elif re.match(".*src:(\\d+)\\+(\\d+)", filename):
|
||||
|
||||
sources = re.match(".*src:(\\d+)\\+(\\d+)", filename)
|
||||
|
||||
source_id_1 = sources.group(1)
|
||||
source_id_2 = sources.group(2)
|
||||
|
||||
file_we_look_for_1_rex = "^id:" + source_id_1 + ","
|
||||
file_we_look_for_2_rex = "^id:" + source_id_2 + ","
|
||||
|
||||
# for mutation with two sources, the sources are never synced form other queues
|
||||
retval = {}
|
||||
|
||||
for k,v in queues[current_fuzzer].items():
|
||||
|
||||
if re.match(file_we_look_for_1_rex, k):
|
||||
retval[k] = compute_mutation_chain(k, current_fuzzer, n+1)
|
||||
|
||||
elif re.match(file_we_look_for_2_rex, k):
|
||||
retval[k] = compute_mutation_chain(k, current_fuzzer, n+1)
|
||||
|
||||
return retval
|
||||
|
||||
else:
|
||||
return "seed"
|
||||
|
||||
|
||||
def main():
|
||||
|
||||
parser = argparse.ArgumentParser(
|
||||
prog='mutation_chain.py',
|
||||
description='Compute the mutation chain of AFL crash files to visulise the mutation history from seed files to crash' +
|
||||
'This tool just dump json data to the CLI, it is advised to echo them into a file for further analysis (i.e. [command] >> your_file.json)',
|
||||
epilog='Greetings from old zealand'
|
||||
)
|
||||
|
||||
parser.add_argument(
|
||||
"-m", "--mode",
|
||||
choices = ['single', 'all'],
|
||||
help = 'compute chain for one file or all crash files in supplied directory. In single mode the -f argument is required',
|
||||
required = True
|
||||
)
|
||||
|
||||
parser.add_argument(
|
||||
"-i", "--input",
|
||||
action = 'store',
|
||||
help = 'Input directory for the mutation chain tool (the fuzzer\'s output directory)',
|
||||
required = True
|
||||
)
|
||||
|
||||
parser.add_argument(
|
||||
"-n", "--node",
|
||||
action = 'store',
|
||||
help = '[Only used in single mode; optinal] name of the fuzzer node that contains the crash file supplied in the --file argument (e.g. \'fuzzer03\'). Defaults to \'fuzzer01\' if not supplied',
|
||||
required = False
|
||||
)
|
||||
|
||||
parser.add_argument(
|
||||
"-f", "--file",
|
||||
action = 'store',
|
||||
help = '[Only used in single mode; required] filename of specific crash file (e.g. \'id:000008,sig:06,src:000005,op:havoc,rep:8\')',
|
||||
required = False
|
||||
)
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
if args.mode == "single":
|
||||
|
||||
if args.node == None:
|
||||
args.node = "fuzzer01"
|
||||
|
||||
if args.file == None:
|
||||
parser.error("'--mode single' requires the '--file' argument.")
|
||||
|
||||
crash_file_path = args.input + '/' + args.node + '/crashes/' + args.file
|
||||
if not os.path.isfile(crash_file_path):
|
||||
print("Error: \'" + crash_file_path + "\' does not exist.\nPlease verify whether the node and filename are correct.")
|
||||
return
|
||||
|
||||
# Create the interal representation of the various queues of parallel fuzzing nodes
|
||||
for dir in os.listdir(args.input):
|
||||
if re.match("^fuzzer\\d+", dir):
|
||||
queues[dir] = fillDictWithFilenameKeys(args.input + '/' + dir + '/queue')
|
||||
|
||||
if args.mode == "all":
|
||||
|
||||
for dir in os.listdir(args.input):
|
||||
if re.match("^fuzzer\\d+", dir):
|
||||
for filename in os.listdir(args.input + '/' + dir + "/crashes"):
|
||||
if re.match("^id:\\d+", filename):
|
||||
print(filename)
|
||||
crashes[filename] = compute_mutation_chain(filename, dir, 0)
|
||||
|
||||
elif args.mode == "single":
|
||||
|
||||
crashes[args.file] = compute_mutation_chain(args.file, args.node, 0)
|
||||
|
||||
print(json.dumps(crashes, sort_keys=True, indent=4))
|
||||
|
||||
if __name__ == '__main__':
|
||||
main()
|
Reference in New Issue
Block a user