aboutsummaryrefslogtreecommitdiffstats
path: root/examples/tls_passthrough.py
blob: 20e8f9bec57aefef4e37bf155c749c036cec1ebd (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
"""
This inline script allows conditional TLS Interception based
on a user-defined strategy.

Example:

    > mitmdump -s tls_passthrough.py

    1. curl --proxy http://localhost:8080 https://example.com --insecure
    // works - we'll also see the contents in mitmproxy

    2. curl --proxy http://localhost:8080 https://example.com --insecure
    // still works - we'll also see the contents in mitmproxy

    3. curl --proxy http://localhost:8080 https://example.com
    // fails with a certificate error, which we will also see in mitmproxy

    4. curl --proxy http://localhost:8080 https://example.com
    // works again, but mitmproxy does not intercept and we do *not* see the contents

Authors: Maximilian Hils, Matthew Tuusberg
"""
from __future__ import absolute_import, print_function, division
import collections
import random

import sys
from enum import Enum

import mitmproxy
from mitmproxy.exceptions import TlsProtocolException
from mitmproxy.protocol import TlsLayer, RawTCPLayer


class InterceptionResult(Enum):
    success = True
    failure = False
    skipped = None


class _TlsStrategy(object):
    """
    Abstract base class for interception strategies.
    """

    def __init__(self):
        # A server_address -> interception results mapping
        self.history = collections.defaultdict(lambda: collections.deque(maxlen=200))

    def should_intercept(self, server_address):
        """
        Returns:
            True, if we should attempt to intercept the connection.
            False, if we want to employ pass-through instead.
        """
        raise NotImplementedError()

    def record_success(self, server_address):
        self.history[server_address].append(InterceptionResult.success)

    def record_failure(self, server_address):
        self.history[server_address].append(InterceptionResult.failure)

    def record_skipped(self, server_address):
        self.history[server_address].append(InterceptionResult.skipped)


class ConservativeStrategy(_TlsStrategy):
    """
    Conservative Interception Strategy - only intercept if there haven't been any failed attempts
    in the history.
    """

    def should_intercept(self, server_address):
        if InterceptionResult.failure in self.history[server_address]:
            return False
        return True


class ProbabilisticStrategy(_TlsStrategy):
    """
    Fixed probability that we intercept a given connection.
    """

    def __init__(self, p):
        self.p = p
        super(ProbabilisticStrategy, self).__init__()

    def should_intercept(self, server_address):
        return random.uniform(0, 1) < self.p


class TlsFeedback(TlsLayer):
    """
    Monkey-patch _establish_tls_with_client to get feedback if TLS could be established
    successfully on the client connection (which may fail due to cert pinning).
    """

    def _establish_tls_with_client(self):
        server_address = self.server_conn.address

        try:
            super(TlsFeedback, self)._establish_tls_with_client()
        except TlsProtocolException as e:
            tls_strategy.record_failure(server_address)
            raise e
        else:
            tls_strategy.record_success(server_address)


# inline script hooks below.

tls_strategy = None


def start():
    global tls_strategy
    if len(sys.argv) == 2:
        tls_strategy = ProbabilisticStrategy(float(sys.argv[1]))
    else:
        tls_strategy = ConservativeStrategy()


def next_layer(next_layer):
    """
    This hook does the actual magic - if the next layer is planned to be a TLS layer,
    we check if we want to enter pass-through mode instead.
    """
    if isinstance(next_layer, TlsLayer) and next_layer._client_tls:
        server_address = next_layer.server_conn.address

        if tls_strategy.should_intercept(server_address):
            # We try to intercept.
            # Monkey-Patch the layer to get feedback from the TLSLayer if interception worked.
            next_layer.__class__ = TlsFeedback
        else:
            # We don't intercept - reply with a pass-through layer and add a "skipped" entry.
            mitmproxy.ctx.log("TLS passthrough for %s" % repr(next_layer.server_conn.address), "info")
            next_layer_replacement = RawTCPLayer(next_layer.ctx, ignore=True)
            next_layer.reply.send(next_layer_replacement)
            tls_strategy.record_skipped(server_address)
ering you: * You are stuck with a sub-optimal design and wish you had done more prototyping before it was too late, but prototyping in C++ is by no means "rapid". * Your tests are slow as they depend on too many libraries or use expensive resources (e.g. a database). * Your tests are brittle as some resources they use are unreliable (e.g. the network). * You want to test how your code handles a failure (e.g. a file checksum error), but it's not easy to cause one. * You need to make sure that your module interacts with other modules in the right way, but it's hard to observe the interaction; therefore you resort to observing the side effects at the end of the action, which is awkward at best. * You want to "mock out" your dependencies, except that they don't have mock implementations yet; and, frankly, you aren't thrilled by some of those hand-written mocks. We encourage you to use Google Mock as: * a _design_ tool, for it lets you experiment with your interface design early and often. More iterations lead to better designs! * a _testing_ tool to cut your tests' outbound dependencies and probe the interaction between your module and its collaborators. # Getting Started # Using Google Mock is easy! Inside your C++ source file, just `#include` `"gtest/gtest.h"` and `"gmock/gmock.h"`, and you are ready to go. # A Case for Mock Turtles # Let's look at an example. Suppose you are developing a graphics program that relies on a LOGO-like API for drawing. How would you test that it does the right thing? Well, you can run it and compare the screen with a golden screen snapshot, but let's admit it: tests like this are expensive to run and fragile (What if you just upgraded to a shiny new graphics card that has better anti-aliasing? Suddenly you have to update all your golden images.). It would be too painful if all your tests are like this. Fortunately, you learned about Dependency Injection and know the right thing to do: instead of having your application talk to the drawing API directly, wrap the API in an interface (say, `Turtle`) and code to that interface: ``` class Turtle { ... virtual ~Turtle() {} virtual void PenUp() = 0; virtual void PenDown() = 0; virtual void Forward(int distance) = 0; virtual void Turn(int degrees) = 0; virtual void GoTo(int x, int y) = 0; virtual int GetX() const = 0; virtual int GetY() const = 0; }; ``` (Note that the destructor of `Turtle` **must** be virtual, as is the case for **all** classes you intend to inherit from - otherwise the destructor of the derived class will not be called when you delete an object through a base pointer, and you'll get corrupted program states like memory leaks.) You can control whether the turtle's movement will leave a trace using `PenUp()` and `PenDown()`, and control its movement using `Forward()`, `Turn()`, and `GoTo()`. Finally, `GetX()` and `GetY()` tell you the current position of the turtle. Your program will normally use a real implementation of this interface. In tests, you can use a mock implementation instead. This allows you to easily check what drawing primitives your program is calling, with what arguments, and in which order. Tests written this way are much more robust (they won't break because your new machine does anti-aliasing differently), easier to read and maintain (the intent of a test is expressed in the code, not in some binary images), and run _much, much faster_. # Writing the Mock Class # If you are lucky, the mocks you need to use have already been implemented by some nice people. If, however, you find yourself in the position to write a mock class, relax - Google Mock turns this task into a fun game! (Well, almost.) ## How to Define It ## Using the `Turtle` interface as example, here are the simple steps you need to follow: 1. Derive a class `MockTurtle` from `Turtle`. 1. Take a _virtual_ function of `Turtle` (while it's possible to [mock non-virtual methods using templates](http://code.google.com/p/googlemock/wiki/V1_6_CookBook#Mocking_Nonvirtual_Methods), it's much more involved). Count how many arguments it has. 1. In the `public:` section of the child class, write `MOCK_METHODn();` (or `MOCK_CONST_METHODn();` if you are mocking a `const` method), where `n` is the number of the arguments; if you counted wrong, shame on you, and a compiler error will tell you so. 1. Now comes the fun part: you take the function signature, cut-and-paste the _function name_ as the _first_ argument to the macro, and leave what's left as the _second_ argument (in case you're curious, this is the _type of the function_). 1. Repeat until all virtual functions you want to mock are done. After the process, you should have something like: ``` #include "gmock/gmock.h" // Brings in Google Mock. class MockTurtle : public Turtle { public: ... MOCK_METHOD0(PenUp, void()); MOCK_METHOD0(PenDown, void()); MOCK_METHOD1(Forward, void(int distance)); MOCK_METHOD1(Turn, void(int degrees)); MOCK_METHOD2(GoTo, void(int x, int y)); MOCK_CONST_METHOD0(GetX, int()); MOCK_CONST_METHOD0(GetY, int()); }; ``` You don't need to define these mock methods somewhere else - the `MOCK_METHOD*` macros will generate the definitions for you. It's that simple! Once you get the hang of it, you can pump out mock classes faster than your source-control system can handle your check-ins. **Tip:** If even this is too much work for you, you'll find the `gmock_gen.py` tool in Google Mock's `scripts/generator/` directory (courtesy of the [cppclean](http://code.google.com/p/cppclean/) project) useful. This command-line tool requires that you have Python 2.4 installed. You give it a C++ file and the name of an abstract class defined in it, and it will print the definition of the mock class for you. Due to the complexity of the C++ language, this script may not always work, but it can be quite handy when it does. For more details, read the [user documentation](http://code.google.com/p/googlemock/source/browse/trunk/scripts/generator/README). ## Where to Put It ## When you define a mock class, you need to decide where to put its definition. Some people put it in a `*_test.cc`. This is fine when the interface being mocked (say, `Foo`) is owned by the same person or team. Otherwise, when the owner of `Foo` changes it, your test could break. (You can't really expect `Foo`'s maintainer to fix every test that uses `Foo`, can you?) So, the rule of thumb is: if you need to mock `Foo` and it's owned by others, define the mock class in `Foo`'s package (better, in a `testing` sub-package such that you can clearly separate production code and testing utilities), and put it in a `mock_foo.h`. Then everyone can reference `mock_foo.h` from their tests. If `Foo` ever changes, there is only one copy of `MockFoo` to change, and only tests that depend on the changed methods need to be fixed. Another way to do it: you can introduce a thin layer `FooAdaptor` on top of `Foo` and code to this new interface. Since you own `FooAdaptor`, you can absorb changes in `Foo` much more easily. While this is more work initially, carefully choosing the adaptor interface can make your code easier to write and more readable (a net win in the long run), as you can choose `FooAdaptor` to fit your specific domain much better than `Foo` does. # Using Mocks in Tests # Once you have a mock class, using it is easy. The typical work flow is: 1. Import the Google Mock names from the `testing` namespace such that you can use them unqualified (You only have to do it once per file. Remember that namespaces are a good idea and good for your health.). 1. Create some mock objects. 1. Specify your expectations on them (How many times will a method be called? With what arguments? What should it do? etc.). 1. Exercise some code that uses the mocks; optionally, check the result using Google Test assertions. If a mock method is called more than expected or with wrong arguments, you'll get an error immediately. 1. When a mock is destructed, Google Mock will automatically check whether all expectations on it have been satisfied. Here's an example: ``` #include "path/to/mock-turtle.h" #include "gmock/gmock.h" #include "gtest/gtest.h" using ::testing::AtLeast; // #1 TEST(PainterTest, CanDrawSomething) { MockTurtle turtle; // #2 EXPECT_CALL(turtle, PenDown()) // #3 .Times(AtLeast(1)); Painter painter(&turtle); // #4 EXPECT_TRUE(painter.DrawCircle(0, 0, 10)); } // #5 int main(int argc, char** argv) { // The following line must be executed to initialize Google Mock // (and Google Test) before running the tests. ::testing::InitGoogleMock(&argc, argv); return RUN_ALL_TESTS(); } ``` As you might have guessed, this test checks that `PenDown()` is called at least once. If the `painter` object didn't call this method, your test will fail with a message like this: ``` path/to/my_test.cc:119: Failure Actual function call count doesn't match this expectation: Actually: never called; Expected: called at least once. ``` **Tip 1:** If you run the test from an Emacs buffer, you can hit `<Enter>` on the line number displayed in the error message to jump right to the failed expectation. **Tip 2:** If your mock objects are never deleted, the final verification won't happen. Therefore it's a good idea to use a heap leak checker in your tests when you allocate mocks on the heap. **Important note:** Google Mock requires expectations to be set **before** the mock functions are called, otherwise the behavior is **undefined**. In particular, you mustn't interleave `EXPECT_CALL()`s and calls to the mock functions. This means `EXPECT_CALL()` should be read as expecting that a call will occur _in the future_, not that a call has occurred. Why does Google Mock work like that? Well, specifying the expectation beforehand allows Google Mock to report a violation as soon as it arises, when the context (stack trace, etc) is still available. This makes debugging much easier. Admittedly, this test is contrived and doesn't do much. You can easily achieve the same effect without using Google Mock. However, as we shall reveal soon, Google Mock allows you to do _much more_ with the mocks. ## Using Google Mock with Any Testing Framework ## If you want to use something other than Google Test (e.g. [CppUnit](http://apps.sourceforge.net/mediawiki/cppunit/index.php?title=Main_Page) or [CxxTest](http://cxxtest.tigris.org/)) as your testing framework, just change the `main()` function in the previous section to: ``` int main(int argc, char** argv) { // The following line causes Google Mock to throw an exception on failure, // which will be interpreted by your testing framework as a test failure. ::testing::GTEST_FLAG(throw_on_failure) = true; ::testing::InitGoogleMock(&argc, argv); ... whatever your testing framework requires ... } ``` This approach has a catch: it makes Google Mock throw an exception from a mock object's destructor sometimes. With some compilers, this sometimes causes the test program to crash. You'll still be able to notice that the test has failed, but it's not a graceful failure. A better solution is to use Google Test's [event listener API](http://code.google.com/p/googletest/wiki/V1_6_AdvancedGuide#Extending_Google_Test_by_Handling_Test_Events) to report a test failure to your testing framework properly. You'll need to implement the `OnTestPartResult()` method of the event listener interface, but it should be straightforward. If this turns out to be too much work, we suggest that you stick with Google Test, which works with Google Mock seamlessly (in fact, it is technically part of Google Mock.). If there is a reason that you cannot use Google Test, please let us know. # Setting Expectations # The key to using a mock object successfully is to set the _right expectations_ on it. If you set the expectations too strict, your test will fail as the result of unrelated changes. If you set them too loose, bugs can slip through. You want to do it just right such that your test can catch exactly the kind of bugs you intend it to catch. Google Mock provides the necessary means for you to do it "just right." ## General Syntax ## In Google Mock we use the `EXPECT_CALL()` macro to set an expectation on a mock method. The general syntax is: ``` EXPECT_CALL(mock_object, method(matchers)) .Times(cardinality) .WillOnce(action) .WillRepeatedly(action); ``` The macro has two arguments: first the mock object, and then the method and its arguments. Note that the two are separated by a comma (`,`), not a period (`.`). (Why using a comma? The answer is that it was necessary for technical reasons.) The macro can be followed by some optional _clauses_ that provide more information about the expectation. We'll discuss how each clause works in the coming sections. This syntax is designed to make an expectation read like English. For example, you can probably guess that ``` using ::testing::Return;... EXPECT_CALL(turtle, GetX()) .Times(5) .WillOnce(Return(100)) .WillOnce(Return(150)) .WillRepeatedly(Return(200)); ``` says that the `turtle` object's `GetX()` method will be called five times, it will return 100 the first time, 150 the second time, and then 200 every time. Some people like to call this style of syntax a Domain-Specific Language (DSL). **Note:** Why do we use a macro to do this? It serves two purposes: first it makes expectations easily identifiable (either by `grep` or by a human reader), and second it allows Google Mock to include the source file location of a failed expectation in messages, making debugging easier. ## Matchers: What Arguments Do We Expect? ## When a mock function takes arguments, we must specify what arguments we are expecting; for example: ``` // Expects the turtle to move forward by 100 units. EXPECT_CALL(turtle, Forward(100)); ``` Sometimes you may not want to be too specific (Remember that talk about tests being too rigid? Over specification leads to brittle tests and obscures the intent of tests. Therefore we encourage you to specify only what's necessary - no more, no less.). If you care to check that `Forward()` will be called but aren't interested in its actual argument, write `_` as the argument, which means "anything goes": ``` using ::testing::_; ... // Expects the turtle to move forward. EXPECT_CALL(turtle, Forward(_)); ``` `_` is an instance of what we call **matchers**. A matcher is like a predicate and can test whether an argument is what we'd expect. You can use a matcher inside `EXPECT_CALL()` wherever a function argument is expected. A list of built-in matchers can be found in the [CheatSheet](V1_6_CheatSheet.md). For example, here's the `Ge` (greater than or equal) matcher: ``` using ::testing::Ge;... EXPECT_CALL(turtle, Forward(Ge(100))); ``` This checks that the turtle will be told to go forward by at least 100 units. ## Cardinalities: How Many Times Will It Be Called? ## The first clause we can specify following an `EXPECT_CALL()` is `Times()`. We call its argument a **cardinality** as it tells _how many times_ the call should occur. It allows us to repeat an expectation many times without actually writing it as many times. More importantly, a cardinality can be "fuzzy", just like a matcher can be. This allows a user to express the intent of a test exactly. An interesting special case is when we say `Times(0)`. You may have guessed - it means that the function shouldn't be called with the given arguments at all, and Google Mock will report a Google Test failure whenever the function is (wrongfully) called. We've seen `AtLeast(n)` as an example of fuzzy cardinalities earlier. For the list of built-in cardinalities you can use, see the [CheatSheet](V1_6_CheatSheet.md). The `Times()` clause can be omitted. **If you omit `Times()`, Google Mock will infer the cardinality for you.** The rules are easy to remember: * If **neither** `WillOnce()` **nor** `WillRepeatedly()` is in the `EXPECT_CALL()`, the inferred cardinality is `Times(1)`. * If there are `n WillOnce()`'s but **no** `WillRepeatedly()`, where `n` >= 1, the cardinality is `Times(n)`. * If there are `n WillOnce()`'s and **one** `WillRepeatedly()`, where `n` >= 0, the cardinality is `Times(AtLeast(n))`. **Quick quiz:** what do you think will happen if a function is expected to be called twice but actually called four times? ## Actions: What Should It Do? ## Remember that a mock object doesn't really have a working implementation? We as users have to tell it what to do when a method is invoked. This is easy in Google Mock. First, if the return type of a mock function is a built-in type or a pointer, the function has a **default action** (a `void` function will just return, a `bool` function will return `false`, and other functions will return 0). If you don't say anything, this behavior will be used. Second, if a mock function doesn't have a default action, or the default action doesn't suit you, you can specify the action to be taken each time the expectation matches using a series of `WillOnce()` clauses followed by an optional `WillRepeatedly()`. For example, ``` using ::testing::Return;... EXPECT_CALL(turtle, GetX()) .WillOnce(Return(100)) .WillOnce(Return(200)) .WillOnce(Return(300)); ``` This says that `turtle.GetX()` will be called _exactly three times_ (Google Mock inferred this from how many `WillOnce()` clauses we've written, since we didn't explicitly write `Times()`), and will return 100, 200, and 300 respectively.