众所周知,在 python 中可使用 exec
函数来执行包含 python 源代码的字符串:html
>>> code = ''' ...: a = "hello" ...: print(a) ...: ''' >>> exec(code) hello >>> a 'hello'
exec
函数的这个功能非常强大,慎用。若是必定要用的话,那么就须要注意一下下面这些安全相关的问题。python
在 exec
执行的代码中,默承认以访问执行 exec
时的局部变量和全局变量, 一样也会修改全局变量。若是 exec 执行的代码是根据用户提交的数据生产的话,这种默认行为就是一个安全隐患。安全
如何更改这种默认行为呢?能够经过执行 exec
函数的时候再传两个参数的方式来 修改这种行为(详见 以前 关于 exec 的文章):函数
>>> g = {} >>> l = {'b': 'world'} >>> exec('hello = "hello" + b', g, l) >>> l {'b': 'world', 'hello': 'helloworld'} >>> g {'__builtins__': {...}} >>> hello --------------------------------------------------------------------------- NameError Traceback (most recent call last) ... NameError: name 'hello' is not defined
若是要限制使用内置函数的话,能够在 globals 参数中定义一下 __builtins__
这个 key:ui
>>> g = {} >>> l = {} >>> exec('a = int("1")', g, l) >>> l {'a': 1} >>> g = {'__builtins__': {}} >>> exec('a = int("1")', g, l) Traceback (most recent call last): File "<stdin>", line 1, in <module> File "<string>", line 1, in <module> NameError: name 'int' is not defined >>>
如今咱们限制了访问和修改全局变量以及使用内置函数,难道这样就万事大吉了吗? 然而并不是如此,仍是能够经过其余的方式来获取内置函数甚至 os.system
函数。code
经过函数对象:orm
>>> def a(): pass ... >>> a.__globals__['__builtins__'] >>> a.__globals__['__builtins__'].open <built-in function open>
经过内置类型对象:htm
>>> for cls in {}.__class__.__base__.__subclasses__(): ... if cls.__name__ == 'WarningMessage': ... b = cls.__init__.__globals__['__builtins__'] ... b['open'] ... <built-in function open> >>>
获取 os.system
:对象
>>> cls = [x for x in [].__class__.__base__.__subclasses__() if x.__name__ == '_wrap_close'][0] >>> cls.__init__.__globals__['path'].os <module 'os' from '/usr/local/var/pyenv/versions/3.5.1/lib/python3.5/os.py'> >>>
对于这两种办法又如何应对呢? 一种办法就是禁止访问以 _
开头的属性:token
若是能够控制 code 的生成,那么就在生成 code 的时候判断
若是不能的话,能够经过 (dist 没法分析嵌套函数的代码)dis
模块分析生成的 code
使用 tokenize
模块:
In [68]: from io import BytesIO In [69]: code = ''' ....: a = 'b' ....: a.__str__ ....: def b(): ....: b.__get__ ....: ''' In [70]: t = tokenize(BytesIO(code.encode()).readline) In [71]: for x in t: ....: print(x) ....: TokenInfo(type=59 (ENCODING), string='utf-8', start=(0, 0), end=(0, 0), line='') TokenInfo(type=58 (NL), string='\n', start=(1, 0), end=(1, 1), line='\n') TokenInfo(type=1 (NAME), string='a', start=(2, 0), end=(2, 1), line="a = 'b'\n") TokenInfo(type=53 (OP), string='=', start=(2, 2), end=(2, 3), line="a = 'b'\n") TokenInfo(type=3 (STRING), string="'b'", start=(2, 4), end=(2, 7), line="a = 'b'\n") TokenInfo(type=4 (NEWLINE), string='\n', start=(2, 7), end=(2, 8), line="a = 'b'\n") TokenInfo(type=1 (NAME), string='a', start=(3, 0), end=(3, 1), line='a.__str__\n') TokenInfo(type=53 (OP), string='.', start=(3, 1), end=(3, 2), line='a.__str__\n') TokenInfo(type=1 (NAME), string='__str__', start=(3, 2), end=(3, 9), line='a.__str__\n') TokenInfo(type=4 (NEWLINE), string='\n', start=(3, 9), end=(3, 10), line='a.__str__\n') TokenInfo(type=1 (NAME), string='def', start=(4, 0), end=(4, 3), line='def b():\n') TokenInfo(type=1 (NAME), string='b', start=(4, 4), end=(4, 5), line='def b():\n') TokenInfo(type=53 (OP), string='(', start=(4, 5), end=(4, 6), line='def b():\n') TokenInfo(type=53 (OP), string=')', start=(4, 6), end=(4, 7), line='def b():\n') TokenInfo(type=53 (OP), string=':', start=(4, 7), end=(4, 8), line='def b():\n') TokenInfo(type=4 (NEWLINE), string='\n', start=(4, 8), end=(4, 9), line='def b():\n') TokenInfo(type=5 (INDENT), string=' ', start=(5, 0), end=(5, 4), line=' b.__get__\n') TokenInfo(type=1 (NAME), string='b', start=(5, 4), end=(5, 5), line=' b.__get__\n') TokenInfo(type=53 (OP), string='.', start=(5, 5), end=(5, 6), line=' b.__get__\n') TokenInfo(type=1 (NAME), string='__get__', start=(5, 6), end=(5, 13), line=' b.__get__\n') TokenInfo(type=4 (NEWLINE), string='\n', start=(5, 13), end=(5, 14), line=' b.__get__\n') TokenInfo(type=6 (DEDENT), string='', start=(6, 0), end=(6, 0), line='') TokenInfo(type=0 (ENDMARKER), string='', start=(6, 0), end=(6, 0), line='')
从上面的输出咱们能够知道当 type 是 OP 而且 string 等于 '.' 时,下一条记录就是
点以后的属性名称。因此咱们的检查代码能够这样写:
import io import tokenize def check_unsafe_attributes(string): g = tokenize.tokenize(io.BytesIO(string.encode('utf-8')).readline) pre_op = '' for toktype, tokval, _, _, _ in g: if toktype == tokenize.NAME and pre_op == '.' and tokval.startswith('_'): attr = tokval msg = "access to attribute '{0}' is unsafe.".format(attr) raise AttributeError(msg) elif toktype == tokenize.OP: pre_op = tokval
我所知道的使用 exec
函数时须要注意的安全问题就是这些了。 若是你还知道其余须要注意的安全问题的话,欢迎留言告知。