Training courses

Kernel and Embedded Linux

Bootlin training courses

Embedded Linux, kernel,
Yocto Project, Buildroot, real-time,
graphics, boot time, debugging...

Bootlin logo

Elixir Cross Referencer

#!/usr/bin/env python

"""
This is a generic fuzz testing tool, see --help for more information.
"""

import os
import sys
import random
import subprocess
import itertools

class TestGenerator:
    def __init__(self, inputs, delete, insert, replace,
                 insert_strings, pick_input):
        self.inputs = [(s, open(s).read()) for s in inputs]

        self.delete = bool(delete)
        self.insert = bool(insert)
        self.replace = bool(replace)
        self.pick_input = bool(pick_input)
        self.insert_strings = list(insert_strings)

        self.num_positions = sum([len(d) for _,d in self.inputs])
        self.num_insert_strings = len(insert_strings)
        self.num_tests = ((delete + (insert + replace)*self.num_insert_strings)
                          * self.num_positions)
        self.num_tests += 1

        if self.pick_input:
            self.num_tests *= self.num_positions

    def position_to_source_index(self, position):
        for i,(s,d) in enumerate(self.inputs):
            n = len(d)
            if position < n:
                return (i,position)
            position -= n
        raise ValueError,'Invalid position.'

    def get_test(self, index):
        assert 0 <= index < self.num_tests

        picked_position = None
        if self.pick_input:
            index,picked_position = divmod(index, self.num_positions)
            picked_position = self.position_to_source_index(picked_position)

        if index == 0:
            return ('nothing', None, None, picked_position)

        index -= 1
        index,position = divmod(index, self.num_positions)
        position = self.position_to_source_index(position)
        if self.delete:
            if index == 0:
                return ('delete', position, None, picked_position)
            index -= 1

        index,insert_index = divmod(index, self.num_insert_strings)
        insert_str = self.insert_strings[insert_index]
        if self.insert:
            if index == 0:
                return ('insert', position, insert_str, picked_position)
            index -= 1

        assert self.replace
        assert index == 0
        return ('replace', position, insert_str, picked_position)

class TestApplication:
    def __init__(self, tg, test):
        self.tg = tg
        self.test = test

    def apply(self):
        if self.test[0] == 'nothing':
            pass
        else:
            i,j = self.test[1]
            name,data = self.tg.inputs[i]
            if self.test[0] == 'delete':
                data = data[:j] + data[j+1:]
            elif self.test[0] == 'insert':
                data = data[:j] + self.test[2] + data[j:]
            elif self.test[0] == 'replace':
                data = data[:j] + self.test[2] + data[j+1:]
            else:
                raise ValueError,'Invalid test %r' % self.test
            open(name,'wb').write(data)

    def revert(self):
        if self.test[0] != 'nothing':
            i,j = self.test[1]
            name,data = self.tg.inputs[i]
            open(name,'wb').write(data)

def quote(str):
    return '"' + str + '"'
        
def run_one_test(test_application, index, input_files, args):
    test = test_application.test

    # Interpolate arguments.
    options = { 'index' : index,
                'inputs' : ' '.join(quote(f) for f in input_files) }

    # Add picked input interpolation arguments, if used.
    if test[3] is not None:
        pos = test[3][1]
        options['picked_input'] = input_files[test[3][0]]
        options['picked_input_pos'] = pos
        # Compute the line and column.
        file_data = test_application.tg.inputs[test[3][0]][1]
        line = column = 1
        for i in range(pos):
            c = file_data[i]
            if c == '\n':
                line += 1
                column = 1
            else:
                column += 1
        options['picked_input_line'] = line
        options['picked_input_col'] = column
        
    test_args = [a % options for a in args]
    if opts.verbose:
        print '%s: note: executing %r' % (sys.argv[0], test_args)

    stdout = None
    stderr = None
    if opts.log_dir:
        stdout_log_path = os.path.join(opts.log_dir, '%s.out' % index)
        stderr_log_path = os.path.join(opts.log_dir, '%s.err' % index)
        stdout = open(stdout_log_path, 'wb')
        stderr = open(stderr_log_path, 'wb')
    else:
        sys.stdout.flush()
    p = subprocess.Popen(test_args, stdout=stdout, stderr=stderr)
    p.communicate()
    exit_code = p.wait()

    test_result = (exit_code == opts.expected_exit_code or
                   exit_code in opts.extra_exit_codes)

    if stdout is not None:
        stdout.close()
        stderr.close()

        # Remove the logs for passes, unless logging all results.
        if not opts.log_all and test_result:
            os.remove(stdout_log_path)
            os.remove(stderr_log_path)

    if not test_result:
        print 'FAIL: %d' % index
    elif not opts.succinct:
        print 'PASS: %d' % index
    return test_result

def main():
    global opts
    from optparse import OptionParser, OptionGroup
    parser = OptionParser("""%prog [options] ... test command args ...

%prog is a tool for fuzzing inputs and testing them.

The most basic usage is something like:

  $ %prog --file foo.txt ./test.sh

which will run a default list of fuzzing strategies on the input. For each
fuzzed input, it will overwrite the input files (in place), run the test script,
then restore the files back to their original contents.

NOTE: You should make sure you have a backup copy of your inputs, in case
something goes wrong!!!

You can cause the fuzzing to not restore the original files with
'--no-revert'. Generally this is used with '--test <index>' to run one failing
test and then leave the fuzzed inputs in place to examine the failure.

For each fuzzed input, %prog will run the test command given on the command
line. Each argument in the command is subject to string interpolation before
being executed. The syntax is "%(VARIABLE)FORMAT" where FORMAT is a standard
printf format, and VARIABLE is one of:

  'index' - the test index being run
  'inputs' - the full list of test inputs
  'picked_input'      - (with --pick-input) the selected input file
  'picked_input_pos'  - (with --pick-input) the selected input position
  'picked_input_line' - (with --pick-input) the selected input line
  'picked_input_col'  - (with --pick-input) the selected input column

By default, the script will run forever continually picking new tests to
run. You can limit the number of tests that are run with '--max-tests <number>',
and you can run a particular test with '--test <index>'.

You can specify '--stop-on-fail' to stop the script on the first failure
without reverting the changes.

""")
    parser.add_option("-v", "--verbose", help="Show more output",
                      action='store_true', dest="verbose", default=False)
    parser.add_option("-s", "--succinct",  help="Reduce amount of output",
                      action="store_true", dest="succinct", default=False)

    group = OptionGroup(parser, "Test Execution")
    group.add_option("", "--expected-exit-code", help="Set expected exit code",
                     type=int, dest="expected_exit_code",
                     default=0)
    group.add_option("", "--extra-exit-code",
                     help="Set additional expected exit code",
                     type=int, action="append", dest="extra_exit_codes",
                     default=[])
    group.add_option("", "--log-dir",
                     help="Capture test logs to an output directory",
                     type=str, dest="log_dir",
                     default=None)
    group.add_option("", "--log-all",
                     help="Log all outputs (not just failures)",
                     action="store_true", dest="log_all", default=False)
    parser.add_option_group(group)

    group = OptionGroup(parser, "Input Files")
    group.add_option("", "--file", metavar="PATH",
                     help="Add an input file to fuzz",
                     type=str, action="append", dest="input_files", default=[])
    group.add_option("", "--filelist", metavar="LIST",
                     help="Add a list of inputs files to fuzz (one per line)",
                     type=str, action="append", dest="filelists", default=[])
    parser.add_option_group(group)

    group = OptionGroup(parser, "Fuzz Options")
    group.add_option("", "--replacement-chars", dest="replacement_chars",
                     help="Characters to insert/replace",
                     default="0{}[]<>\;@#$^%& ")
    group.add_option("", "--replacement-string", dest="replacement_strings",
                     action="append", help="Add a replacement string to use",
                     default=[])
    group.add_option("", "--replacement-list", dest="replacement_lists",
                     help="Add a list of replacement strings (one per line)",
                     action="append", default=[])
    group.add_option("", "--no-delete", help="Don't delete characters",
                     action='store_false', dest="enable_delete", default=True)
    group.add_option("", "--no-insert", help="Don't insert strings",
                     action='store_false', dest="enable_insert", default=True)
    group.add_option("", "--no-replace", help="Don't replace strings",
                     action='store_false', dest="enable_replace", default=True)
    group.add_option("", "--no-revert", help="Don't revert changes",
                     action='store_false', dest="revert", default=True)
    group.add_option("", "--stop-on-fail", help="Stop on first failure",
                     action='store_true', dest="stop_on_fail", default=False)
    parser.add_option_group(group)

    group = OptionGroup(parser, "Test Selection")
    group.add_option("", "--test", help="Run a particular test",
                     type=int, dest="test", default=None, metavar="INDEX")
    group.add_option("", "--max-tests", help="Maximum number of tests",
                     type=int, dest="max_tests", default=None, metavar="COUNT")
    group.add_option("", "--pick-input",
                     help="Randomly select an input byte as well as fuzzing",
                     action='store_true', dest="pick_input", default=False)
    parser.add_option_group(group)

    parser.disable_interspersed_args()

    (opts, args) = parser.parse_args()

    if not args:
        parser.error("Invalid number of arguments")

    # Collect the list of inputs.
    input_files = list(opts.input_files)
    for filelist in opts.filelists:
        f = open(filelist)
        try:
            for ln in f:
                ln = ln.strip()
                if ln:
                    input_files.append(ln)
        finally:
            f.close()
    input_files.sort()

    if not input_files:
        parser.error("No input files!")

    print '%s: note: fuzzing %d files.' % (sys.argv[0], len(input_files))

    # Make sure the log directory exists if used.
    if opts.log_dir:
        if not os.path.exists(opts.log_dir):
            try:
                os.mkdir(opts.log_dir)
            except OSError:
                print "%s: error: log directory couldn't be created!" % (
                    sys.argv[0],)
                raise SystemExit,1

    # Get the list if insert/replacement strings.
    replacements = list(opts.replacement_chars)
    replacements.extend(opts.replacement_strings)
    for replacement_list in opts.replacement_lists:
        f = open(replacement_list)
        try:
            for ln in f:
                ln = ln[:-1]
                if ln:
                    replacements.append(ln)
        finally:
            f.close()

    # Unique and order the replacement list.
    replacements = list(set(replacements))
    replacements.sort()

    # Create the test generator.
    tg = TestGenerator(input_files, opts.enable_delete, opts.enable_insert,
                       opts.enable_replace, replacements, opts.pick_input)

    print '%s: note: %d input bytes.' % (sys.argv[0], tg.num_positions)
    print '%s: note: %d total tests.' % (sys.argv[0], tg.num_tests)
    if opts.test is not None:
        it = [opts.test]
    elif opts.max_tests is not None:
        it = itertools.imap(random.randrange,
                            itertools.repeat(tg.num_tests, opts.max_tests))
    else:
        it = itertools.imap(random.randrange, itertools.repeat(tg.num_tests))
    for test in it:
        t = tg.get_test(test)

        if opts.verbose:
            print '%s: note: running test %d: %r' % (sys.argv[0], test, t)
        ta = TestApplication(tg, t)
        try:
            ta.apply()
            test_result = run_one_test(ta, test, input_files, args)
            if not test_result and opts.stop_on_fail:
                opts.revert = False
                sys.exit(1)
        finally:
            if opts.revert:
                ta.revert()

        sys.stdout.flush()

if __name__ == '__main__':
    main()