xref: /openbmc/qemu/docs/image-fuzzer.txt (revision fc8c745d)
1# Specification for the fuzz testing tool
2#
3# Copyright (C) 2014 Maria Kustova <maria.k@catit.be>
4#
5# This program is free software: you can redistribute it and/or modify
6# it under the terms of the GNU General Public License as published by
7# the Free Software Foundation, either version 2 of the License, or
8# (at your option) any later version.
9#
10# This program is distributed in the hope that it will be useful,
11# but WITHOUT ANY WARRANTY; without even the implied warranty of
12# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13# GNU General Public License for more details.
14#
15# You should have received a copy of the GNU General Public License
16# along with this program.  If not, see <http://www.gnu.org/licenses/>.
17
18
19Image fuzzer
20============
21
22Description
23-----------
24
25The goal of the image fuzzer is to catch crashes of qemu-io/qemu-img
26by providing to them randomly corrupted images.
27Test images are generated from scratch and have valid inner structure with some
28elements, e.g. L1/L2 tables, having random invalid values.
29
30
31Test runner
32-----------
33
34The test runner generates test images, executes tests utilizing generated
35images, indicates their results and collects all test related artifacts (logs,
36core dumps, test images, backing files).
37The test means execution of all available commands under test with the same
38generated test image.
39By default, the test runner generates new tests and executes them until
40keyboard interruption. But if a test seed is specified via the '--seed' runner
41parameter, then only one test with this seed will be executed, after its finish
42the runner will exit.
43
44The runner uses an external image fuzzer to generate test images. An image
45generator should be specified as a mandatory parameter of the test runner.
46Details about interactions between the runner and fuzzers see "Module
47interfaces".
48
49The runner activates generation of core dumps during test executions, but it
50assumes that core dumps will be generated in the current working directory.
51For comprehensive test results, please, set up your test environment
52properly.
53
54Paths to binaries under test (SUTs) qemu-img and qemu-io are retrieved from
55environment variables. If the environment check fails the runner will
56use SUTs installed in system paths.
57qemu-img is required for creation of backing files, so it's mandatory to set
58the related environment variable if it's not installed in the system path.
59For details about environment variables see qemu-iotests/check.
60
61The runner accepts a JSON array of fields expected to be fuzzed via the
62'--config' argument, e.g.
63
64       '[["feature_name_table"], ["header", "l1_table_offset"]]'
65
66Each sublist can have one or two strings defining image structure elements.
67In the latter case a parent element should be placed on the first position,
68and a field name on the second one.
69
70The runner accepts a list of commands under test as a JSON array via
71the '--command' argument. Each command is a list containing a SUT and all its
72arguments, e.g.
73
74       runner.py -c '[["qemu-io", "$test_img", "-c", "write $off $len"]]'
75     /tmp/test ../qcow2
76
77For variable arguments next aliases can be used:
78    - $test_img for a fuzzed img
79    - $off for an offset in the fuzzed image
80    - $len for a data size
81
82Values for last two aliases will be generated based on a size of a virtual
83disk of the generated image.
84In case when no commands are specified the runner will execute commands from
85the default list:
86    - qemu-img check
87    - qemu-img info
88    - qemu-img convert
89    - qemu-io -c read
90    - qemu-io -c write
91    - qemu-io -c aio_read
92    - qemu-io -c aio_write
93    - qemu-io -c flush
94    - qemu-io -c discard
95    - qemu-io -c truncate
96
97
98Qcow2 image generator
99---------------------
100
101The 'qcow2' generator is a Python package providing 'create_image' method as
102a single public API. See details in 'Test runner/image fuzzer' chapter of
103'Module interfaces'.
104
105Qcow2 contains two submodules: fuzz.py and layout.py.
106
107'fuzz.py' contains all fuzzing functions, one per image field. It's assumed
108that after code analysis every field will have own constraints for its value.
109For now only universal potentially dangerous values are used, e.g. type limits
110for integers or unsafe symbols as '%s' for strings. For bitmasks random amount
111of bits are set to ones. All fuzzed values are checked on non-equality to the
112current valid value of the field. In case of equality the value will be
113regenerated.
114
115'layout.py' creates a random valid image, fuzzes a random subset of the image
116fields by 'fuzz.py' module and writes a fuzzed image to the file specified.
117If a fuzzer configuration is specified, then it has the next interpretation:
118
119    1. If a list contains a parent image element only, then some random portion
120    of fields of this element will be fuzzed every test.
121    The same behavior is applied for the entire image if no configuration is
122    used. This case is useful for the test specialization.
123
124    2. If a list contains a parent element and a field name, then a field
125    will be always fuzzed for every test. This case is useful for regression
126    testing.
127
128The generator can create header fields, header extensions, L1/L2 tables and
129refcount table and blocks.
130
131Module interfaces
132-----------------
133
134* Test runner/image fuzzer
135
136The runner calls an image generator specifying the path to a test image file,
137path to a backing file and its format and a fuzzer configuration.
138An image generator is expected to provide a
139
140   'create_image(test_img_path, backing_file_path=None,
141                 backing_file_format=None, fuzz_config=None)'
142
143method that creates a test image, writes it to the specified file and returns
144the size of the virtual disk.
145The file should be created if it doesn't exist or overwritten otherwise.
146fuzz_config has a form of a list of lists. Every sublist can have one
147or two elements: first element is a name of a parent image element, second one
148if exists is a name of a field in this element.
149Example,
150        [['header', 'l1_table_offset'],
151         ['header', 'nb_snapshots'],
152         ['feature_name_table']]
153
154Random seed is set by the runner at every test execution for the regression
155purpose, so an image generator is not recommended to modify it internally.
156
157
158Overall fuzzer requirements
159===========================
160
161Input data:
162----------
163
164 - image template (generator)
165 - work directory
166 - action vector (optional)
167 - seed (optional)
168 - SUT and its arguments (optional)
169
170
171Fuzzer requirements:
172-------------------
173
1741.  Should be able to inject random data
1752.  Should be able to select a random value from the manually pregenerated
176    vector (boundary values, e.g. max/min cluster size)
1773.  Image template should describe a general structure invariant for all
178    test images (image format description)
1794.  Image template should be autonomous and other fuzzer parts should not
180    rely on it
1815.  Image template should contain reference rules (not only block+size
182    description)
1836.  Should generate the test image with the correct structure based on an image
184    template
1857.  Should accept a seed as an argument (for regression purpose)
1868.  Should generate a seed if it is not specified as an input parameter.
1879.  The same seed should generate the same image for the same action vector,
188    specified or generated.
18910. Should accept a vector of actions as an argument (for test reproducing and
190    for test case specification, e.g. group of tests for header structure,
191    group of test for snapshots, etc)
19211. Action vector should be randomly generated from the pool of available
193    actions, if it is not specified as an input parameter
19412. Pool of actions should be defined automatically based on an image template
19513. Should accept a SUT and its call parameters as an argument or select them
196    randomly otherwise. As far as it's expected to be rarely changed, the list
197    of all possible test commands can be available in the test runner
198    internally.
19914. Should support an external cancellation of a test run
20015. Seed should be logged (for regression purpose)
20116. All files related to a test result should be collected: a test image,
202    SUT logs, fuzzer logs and crash dumps
20317. Should be compatible with python version 2.4-2.7
20418. Usage of external libraries should be limited as much as possible.
205
206
207Image formats:
208-------------
209
210Main target image format is qcow2, but support of image templates should
211provide an ability to add any other image format.
212
213
214Effectiveness:
215-------------
216
217The fuzzer can be controlled via template, seed and action vector;
218it makes the fuzzer itself invariant to an image format and test logic.
219It should be able to perform rather complex and precise tests, that can be
220specified via an action vector. Otherwise, knowledge about an image structure
221allows the fuzzer to generate the pool of all available areas can be fuzzed
222and randomly select some of them and so compose its own action vector.
223Also complexity of a template defines complexity of the fuzzer, so its
224functionality can be varied from simple model-independent fuzzing to smart
225model-based one.
226
227
228Glossary:
229--------
230
231Action vector is a sequence of structure elements retrieved from an image
232format, each of them will be fuzzed for the test image. It's a subset of
233elements of the action pool. Example: header, refcount table, etc.
234Action pool is all available elements of an image structure that generated
235automatically from an image template.
236Image template is a formal description of an image structure and relations
237between image blocks.
238Test image is an output image of the fuzzer defined by the current seed and
239action vector.
240