mirror of
https://github.com/Grasscutters/mitmproxy.git
synced 2024-11-23 08:11:00 +00:00
1b961fc4ad
This is a common task in pentesting scenarios. This commit adds the following functions: utils.Headers.replace proxy.Request.replace proxy.Response.replace flow.Flow.replace
329 lines
8.8 KiB
Python
329 lines
8.8 KiB
Python
import textwrap, cStringIO, os, time, re, json
|
|
import libpry
|
|
from libmproxy import utils
|
|
|
|
utils.CERT_SLEEP_TIME = 0
|
|
|
|
|
|
class uformat_timestamp(libpry.AutoTree):
|
|
def test_simple(self):
|
|
assert utils.format_timestamp(utils.timestamp())
|
|
|
|
|
|
class uisBin(libpry.AutoTree):
|
|
def test_simple(self):
|
|
assert not utils.isBin("testing\n\r")
|
|
assert utils.isBin("testing\x01")
|
|
assert utils.isBin("testing\x0e")
|
|
assert utils.isBin("testing\x7f")
|
|
|
|
|
|
class uisXML(libpry.AutoTree):
|
|
def test_simple(self):
|
|
assert not utils.isXML("foo")
|
|
assert utils.isXML("<foo")
|
|
assert utils.isXML(" \n<foo")
|
|
|
|
|
|
class uhexdump(libpry.AutoTree):
|
|
def test_simple(self):
|
|
assert utils.hexdump("one\0"*10)
|
|
|
|
|
|
class upretty_size(libpry.AutoTree):
|
|
def test_simple(self):
|
|
assert utils.pretty_size(100) == "100B"
|
|
assert utils.pretty_size(1024) == "1kB"
|
|
assert utils.pretty_size(1024 + (1024/2)) == "1.5kB"
|
|
assert utils.pretty_size(1024*1024) == "1M"
|
|
|
|
|
|
class uData(libpry.AutoTree):
|
|
def test_nonexistent(self):
|
|
libpry.raises("does not exist", utils.data.path, "nonexistent")
|
|
|
|
|
|
class uHeaders(libpry.AutoTree):
|
|
def setUp(self):
|
|
self.hd = utils.Headers()
|
|
|
|
def test_read_simple(self):
|
|
data = """
|
|
Header: one
|
|
Header2: two
|
|
\r\n
|
|
"""
|
|
data = textwrap.dedent(data)
|
|
data = data.strip()
|
|
s = cStringIO.StringIO(data)
|
|
self.hd.read(s)
|
|
assert self.hd["header"] == ["one"]
|
|
assert self.hd["header2"] == ["two"]
|
|
|
|
def test_read_multi(self):
|
|
data = """
|
|
Header: one
|
|
Header: two
|
|
\r\n
|
|
"""
|
|
data = textwrap.dedent(data)
|
|
data = data.strip()
|
|
s = cStringIO.StringIO(data)
|
|
self.hd.read(s)
|
|
assert self.hd["header"] == ["one", "two"]
|
|
|
|
def test_read_continued(self):
|
|
data = """
|
|
Header: one
|
|
\ttwo
|
|
Header2: three
|
|
\r\n
|
|
"""
|
|
data = textwrap.dedent(data)
|
|
data = data.strip()
|
|
s = cStringIO.StringIO(data)
|
|
self.hd.read(s)
|
|
assert self.hd["header"] == ['one\r\n two']
|
|
|
|
def test_dictToHeader1(self):
|
|
self.hd.add("one", "uno")
|
|
self.hd.add("two", "due")
|
|
self.hd.add("two", "tre")
|
|
expected = [
|
|
"one: uno\r\n",
|
|
"two: due\r\n",
|
|
"two: tre\r\n",
|
|
"\r\n"
|
|
]
|
|
out = repr(self.hd)
|
|
for i in expected:
|
|
assert out.find(i) >= 0
|
|
|
|
def test_dictToHeader2(self):
|
|
self.hd["one"] = ["uno"]
|
|
expected1 = "one: uno\r\n"
|
|
expected2 = "\r\n"
|
|
out = repr(self.hd)
|
|
assert out.find(expected1) >= 0
|
|
assert out.find(expected2) >= 0
|
|
|
|
def test_match_re(self):
|
|
h = utils.Headers()
|
|
h.add("one", "uno")
|
|
h.add("two", "due")
|
|
h.add("two", "tre")
|
|
assert h.match_re("uno")
|
|
assert h.match_re("two: due")
|
|
assert not h.match_re("nonono")
|
|
|
|
def test_getset_state(self):
|
|
self.hd.add("foo", 1)
|
|
self.hd.add("foo", 2)
|
|
self.hd.add("bar", 3)
|
|
state = self.hd.get_state()
|
|
nd = utils.Headers.from_state(state)
|
|
assert nd == self.hd
|
|
|
|
def test_copy(self):
|
|
self.hd.add("foo", 1)
|
|
self.hd.add("foo", 2)
|
|
self.hd.add("bar", 3)
|
|
assert self.hd == self.hd.copy()
|
|
|
|
def test_del(self):
|
|
self.hd.add("foo", 1)
|
|
self.hd.add("Foo", 2)
|
|
self.hd.add("bar", 3)
|
|
del self.hd["foo"]
|
|
assert len(self.hd.lst) == 1
|
|
|
|
def test_replace(self):
|
|
self.hd.add("one", "two")
|
|
self.hd.add("two", "one")
|
|
assert self.hd.replace("one", "vun") == 2
|
|
assert self.hd.lst == [
|
|
["vun", "two"],
|
|
["two", "vun"],
|
|
]
|
|
|
|
|
|
class uisStringLike(libpry.AutoTree):
|
|
def test_all(self):
|
|
assert utils.isStringLike("foo")
|
|
assert not utils.isStringLike([1, 2, 3])
|
|
assert not utils.isStringLike((1, 2, 3))
|
|
assert not utils.isStringLike(["1", "2", "3"])
|
|
|
|
|
|
class uisSequenceLike(libpry.AutoTree):
|
|
def test_all(self):
|
|
assert utils.isSequenceLike([1, 2, 3])
|
|
assert utils.isSequenceLike((1, 2, 3))
|
|
assert not utils.isSequenceLike("foobar")
|
|
assert utils.isSequenceLike(["foobar", "foo"])
|
|
x = iter([1, 2, 3])
|
|
assert utils.isSequenceLike(x)
|
|
assert not utils.isSequenceLike(1)
|
|
|
|
|
|
|
|
class upretty_xmlish(libpry.AutoTree):
|
|
def test_tagre(self):
|
|
def f(s):
|
|
return re.search(utils.TAG, s, re.VERBOSE|re.MULTILINE)
|
|
assert f(r"<body>")
|
|
assert f(r"<body/>")
|
|
assert f(r"< body/>")
|
|
assert f(r"< body/ >")
|
|
assert f(r"< body / >")
|
|
assert f(r"<foo a=b>")
|
|
assert f(r"<foo a='b'>")
|
|
assert f(r"<foo a='b\"'>")
|
|
assert f(r'<a b=(a.b) href="foo">')
|
|
assert f('<td width=25%>')
|
|
assert f('<form name="search" action="/search.php" method="get" accept-charset="utf-8" class="search">')
|
|
assert f('<img src="gif" width="125" height="16" alt="" />')
|
|
|
|
|
|
def test_all(self):
|
|
def isbalanced(ret):
|
|
# The last tag should have no indent
|
|
assert ret[-1].strip() == ret[-1]
|
|
|
|
s = "<html><br><br></br><p>one</p></html>"
|
|
ret = utils.pretty_xmlish(s)
|
|
isbalanced(ret)
|
|
|
|
s = r"""
|
|
<body bgcolor=#ffffff text=#000000 link=#0000cc vlink=#551a8b alink=#ff0000 onload="document.f.q.focus();if(document.images)new Image().src='/images/srpr/nav_logo27.png'" ><textarea id=csi style=display:none></textarea></body>
|
|
"""
|
|
isbalanced(utils.pretty_xmlish(textwrap.dedent(s)))
|
|
|
|
s = r"""
|
|
<a href="http://foo.com" target="">
|
|
<img src="http://foo.gif" alt="bar" height="25" width="132">
|
|
</a>
|
|
"""
|
|
isbalanced(utils.pretty_xmlish(textwrap.dedent(s)))
|
|
|
|
s = r"""
|
|
<!DOCTYPE html PUBLIC \"-//W3C//DTD XHTML 1.0 Strict//EN\"
|
|
\"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd\">
|
|
<html></html>
|
|
"""
|
|
ret = utils.pretty_xmlish(textwrap.dedent(s))
|
|
isbalanced(ret)
|
|
|
|
s = "<html><br/><p>one</p></html>"
|
|
ret = utils.pretty_xmlish(s)
|
|
assert len(ret) == 6
|
|
isbalanced(ret)
|
|
|
|
s = "gobbledygook"
|
|
assert utils.pretty_xmlish(s) == ["gobbledygook"]
|
|
|
|
|
|
class upretty_json(libpry.AutoTree):
|
|
def test_one(self):
|
|
s = json.dumps({"foo": 1})
|
|
assert utils.pretty_json(s)
|
|
assert not utils.pretty_json("moo")
|
|
|
|
|
|
class u_urldecode(libpry.AutoTree):
|
|
def test_one(self):
|
|
s = "one=two&three=four"
|
|
assert len(utils.urldecode(s)) == 2
|
|
|
|
|
|
class udummy_ca(libpry.AutoTree):
|
|
def test_all(self):
|
|
d = self.tmpdir()
|
|
path = os.path.join(d, "foo/cert.cnf")
|
|
assert utils.dummy_ca(path)
|
|
assert os.path.exists(path)
|
|
|
|
path = os.path.join(d, "foo/cert2.pem")
|
|
assert utils.dummy_ca(path)
|
|
assert os.path.exists(path)
|
|
assert os.path.exists(os.path.join(d, "foo/cert2-cert.pem"))
|
|
assert os.path.exists(os.path.join(d, "foo/cert2-cert.p12"))
|
|
|
|
|
|
class udummy_cert(libpry.AutoTree):
|
|
def test_with_ca(self):
|
|
d = self.tmpdir()
|
|
cacert = os.path.join(d, "foo/cert.cnf")
|
|
assert utils.dummy_ca(cacert)
|
|
assert utils.dummy_cert(
|
|
os.path.join(d, "foo"),
|
|
cacert,
|
|
"foo.com"
|
|
)
|
|
assert os.path.exists(os.path.join(d, "foo", "foo.com.pem"))
|
|
# Short-circuit
|
|
assert utils.dummy_cert(
|
|
os.path.join(d, "foo"),
|
|
cacert,
|
|
"foo.com"
|
|
)
|
|
|
|
def test_no_ca(self):
|
|
d = self.tmpdir()
|
|
assert utils.dummy_cert(
|
|
d,
|
|
None,
|
|
"foo.com"
|
|
)
|
|
assert os.path.exists(os.path.join(d, "foo.com.pem"))
|
|
|
|
|
|
class uLRUCache(libpry.AutoTree):
|
|
def test_one(self):
|
|
class Foo:
|
|
ran = False
|
|
@utils.LRUCache(2)
|
|
def one(self, x):
|
|
self.ran = True
|
|
return x
|
|
|
|
f = Foo()
|
|
assert f.one(1) == 1
|
|
assert f.ran
|
|
f.ran = False
|
|
assert f.one(1) == 1
|
|
assert not f.ran
|
|
|
|
f.ran = False
|
|
assert f.one(1) == 1
|
|
assert not f.ran
|
|
assert f.one(2) == 2
|
|
assert f.one(3) == 3
|
|
assert f.ran
|
|
|
|
f.ran = False
|
|
assert f.one(1) == 1
|
|
assert f.ran
|
|
|
|
assert len(f._cached_one) == 2
|
|
assert len(f._cachelist_one) == 2
|
|
|
|
|
|
tests = [
|
|
uformat_timestamp(),
|
|
uisBin(),
|
|
uisXML(),
|
|
uhexdump(),
|
|
upretty_size(),
|
|
uisStringLike(),
|
|
uisSequenceLike(),
|
|
uHeaders(),
|
|
uData(),
|
|
upretty_xmlish(),
|
|
upretty_json(),
|
|
u_urldecode(),
|
|
udummy_ca(),
|
|
udummy_cert(),
|
|
uLRUCache(),
|
|
]
|