xref: /qemu/docs/devel/testing/fuzzing.rst (revision ef834aa2b2b52ac44ca97d70d5ef5e975a229034)
1 ========
2 Fuzzing
3 ========
4 
5 This document describes the virtual-device fuzzing infrastructure in QEMU and
6 how to use it to implement additional fuzzers.
7 
8 Basics
9 ------
10 
11 Fuzzing operates by passing inputs to an entry point/target function. The
12 fuzzer tracks the code coverage triggered by the input. Based on these
13 findings, the fuzzer mutates the input and repeats the fuzzing.
14 
15 To fuzz QEMU, we rely on libfuzzer. Unlike other fuzzers such as AFL, libfuzzer
16 is an *in-process* fuzzer. For the developer, this means that it is their
17 responsibility to ensure that state is reset between fuzzing-runs.
18 
19 Building the fuzzers
20 --------------------
21 
22 To build the fuzzers, install a recent version of clang:
23 Configure with (substitute the clang binaries with the version you installed).
24 Here, enable-sanitizers, is optional but it allows us to reliably detect bugs
25 such as out-of-bounds accesses, use-after-frees, double-frees etc.::
26 
27     CC=clang-8 CXX=clang++-8 /path/to/configure --enable-fuzzing \
28                                                 --enable-sanitizers
29 
30 Fuzz targets are built similarly to system targets::
31 
32     make qemu-fuzz-i386
33 
34 This builds ``./qemu-fuzz-i386``
35 
36 The first option to this command is: ``--fuzz-target=FUZZ_NAME``
37 To list all of the available fuzzers run ``qemu-fuzz-i386`` with no arguments.
38 
39 For example::
40 
41     ./qemu-fuzz-i386 --fuzz-target=virtio-scsi-fuzz
42 
43 Internally, libfuzzer parses all arguments that do not begin with ``"--"``.
44 Information about these is available by passing ``-help=1``
45 
46 Now the only thing left to do is wait for the fuzzer to trigger potential
47 crashes.
48 
49 Useful libFuzzer flags
50 ----------------------
51 
52 As mentioned above, libFuzzer accepts some arguments. Passing ``-help=1`` will
53 list the available arguments. In particular, these arguments might be helpful:
54 
55 * ``CORPUS_DIR/`` : Specify a directory as the last argument to libFuzzer.
56   libFuzzer stores each "interesting" input in this corpus directory. The next
57   time you run libFuzzer, it will read all of the inputs from the corpus, and
58   continue fuzzing from there. You can also specify multiple directories.
59   libFuzzer loads existing inputs from all specified directories, but will only
60   write new ones to the first one specified.
61 
62 * ``-max_len=4096`` : specify the maximum byte-length of the inputs libFuzzer
63   will generate.
64 
65 * ``-close_fd_mask={1,2,3}`` : close, stderr, or both. Useful for targets that
66   trigger many debug/error messages, or create output on the serial console.
67 
68 * ``-jobs=4 -workers=4`` : These arguments configure libFuzzer to run 4 fuzzers in
69   parallel (4 fuzzing jobs in 4 worker processes). Alternatively, with only
70   ``-jobs=N``, libFuzzer automatically spawns a number of workers less than or equal
71   to half the available CPU cores. Replace 4 with a number appropriate for your
72   machine. Make sure to specify a ``CORPUS_DIR``, which will allow the parallel
73   fuzzers to share information about the interesting inputs they find.
74 
75 * ``-use_value_profile=1`` : For each comparison operation, libFuzzer computes
76   ``(caller_pc&4095) | (popcnt(Arg1 ^ Arg2) << 12)`` and places this in the
77   coverage table. Useful for targets with "magic" constants. If Arg1 came from
78   the fuzzer's input and Arg2 is a magic constant, then each time the Hamming
79   distance between Arg1 and Arg2 decreases, libFuzzer adds the input to the
80   corpus.
81 
82 * ``-shrink=1`` : Tries to make elements of the corpus "smaller". Might lead to
83   better coverage performance, depending on the target.
84 
85 Note that libFuzzer's exact behavior will depend on the version of
86 clang and libFuzzer used to build the device fuzzers.
87 
88 Generating Coverage Reports
89 ---------------------------
90 
91 Code coverage is a crucial metric for evaluating a fuzzer's performance.
92 libFuzzer's output provides a "cov: " column that provides a total number of
93 unique blocks/edges covered. To examine coverage on a line-by-line basis we
94 can use Clang coverage:
95 
96  1. Configure libFuzzer to store a corpus of all interesting inputs (see
97     CORPUS_DIR above)
98  2. ``./configure`` the QEMU build with ::
99 
100     --enable-fuzzing \
101     --extra-cflags="-fprofile-instr-generate -fcoverage-mapping"
102 
103  3. Re-run the fuzzer. Specify $CORPUS_DIR/* as an argument, telling libfuzzer
104     to execute all of the inputs in $CORPUS_DIR and exit. Once the process
105     exits, you should find a file, "default.profraw" in the working directory.
106  4. Execute these commands to generate a detailed HTML coverage-report::
107 
108       llvm-profdata merge -output=default.profdata default.profraw
109       llvm-cov show ./path/to/qemu-fuzz-i386 -instr-profile=default.profdata \
110       --format html -output-dir=/path/to/output/report
111 
112 Adding a new fuzzer
113 -------------------
114 
115 Coverage over virtual devices can be improved by adding additional fuzzers.
116 Fuzzers are kept in ``tests/qtest/fuzz/`` and should be added to
117 ``tests/qtest/fuzz/meson.build``
118 
119 Fuzzers can rely on both qtest and libqos to communicate with virtual devices.
120 
121 1. Create a new source file. For example ``tests/qtest/fuzz/foo-device-fuzz.c``.
122 
123 2. Write the fuzzing code using the libqtest/libqos API. See existing fuzzers
124    for reference.
125 
126 3. Add the fuzzer to ``tests/qtest/fuzz/meson.build``.
127 
128 Fuzzers can be more-or-less thought of as special qtest programs which can
129 modify the qtest commands and/or qtest command arguments based on inputs
130 provided by libfuzzer. Libfuzzer passes a byte array and length. Commonly the
131 fuzzer loops over the byte-array interpreting it as a list of qtest commands,
132 addresses, or values.
133 
134 The Generic Fuzzer
135 ------------------
136 
137 Writing a fuzz target can be a lot of effort (especially if a device driver has
138 not be built-out within libqos). Many devices can be fuzzed to some degree,
139 without any device-specific code, using the generic-fuzz target.
140 
141 The generic-fuzz target is capable of fuzzing devices over their PIO, MMIO,
142 and DMA input-spaces. To apply the generic-fuzz to a device, we need to define
143 two env-variables, at minimum:
144 
145 * ``QEMU_FUZZ_ARGS=`` is the set of QEMU arguments used to configure a machine, with
146   the device attached. For example, if we want to fuzz the virtio-net device
147   attached to a pc-i440fx machine, we can specify::
148 
149     QEMU_FUZZ_ARGS="-M pc -nodefaults -netdev user,id=user0 \
150     -device virtio-net,netdev=user0"
151 
152 * ``QEMU_FUZZ_OBJECTS=`` is a set of space-delimited strings used to identify
153   the MemoryRegions that will be fuzzed. These strings are compared against
154   MemoryRegion names and MemoryRegion owner names, to decide whether each
155   MemoryRegion should be fuzzed. These strings support globbing. For the
156   virtio-net example, we could use one of ::
157 
158     QEMU_FUZZ_OBJECTS='virtio-net'
159     QEMU_FUZZ_OBJECTS='virtio*'
160     QEMU_FUZZ_OBJECTS='virtio* pcspk' # Fuzz the virtio devices and the speaker
161     QEMU_FUZZ_OBJECTS='*' # Fuzz the whole machine``
162 
163 The ``"info mtree"`` and ``"info qom-tree"`` monitor commands can be especially
164 useful for identifying the ``MemoryRegion`` and ``Object`` names used for
165 matching.
166 
167 As a generic rule-of-thumb, the more ``MemoryRegions``/Devices we match, the
168 greater the input-space, and the smaller the probability of finding crashing
169 inputs for individual devices. As such, it is usually a good idea to limit the
170 fuzzer to only a few ``MemoryRegions``.
171 
172 To ensure that these env variables have been configured correctly, we can use::
173 
174     ./qemu-fuzz-i386 --fuzz-target=generic-fuzz -runs=0
175 
176 The output should contain a complete list of matched MemoryRegions.
177 
178 OSS-Fuzz
179 --------
180 QEMU is continuously fuzzed on `OSS-Fuzz
181 <https://github.com/google/oss-fuzz>`_.  By default, the OSS-Fuzz build
182 will try to fuzz every fuzz-target. Since the generic-fuzz target
183 requires additional information provided in environment variables, we
184 pre-define some generic-fuzz configs in
185 ``tests/qtest/fuzz/generic_fuzz_configs.h``. Each config must specify:
186 
187 - ``.name``: To identify the fuzzer config
188 
189 - ``.args`` OR ``.argfunc``: A string or pointer to a function returning a
190   string.  These strings are used to specify the ``QEMU_FUZZ_ARGS``
191   environment variable.  ``argfunc`` is useful when the config relies on e.g.
192   a dynamically created temp directory, or a free tcp/udp port.
193 
194 - ``.objects``: A string that specifies the ``QEMU_FUZZ_OBJECTS`` environment
195   variable.
196 
197 To fuzz additional devices/device configuration on OSS-Fuzz, send patches for
198 either a new device-specific fuzzer or a new generic-fuzz config.
199 
200 Build details:
201 
202 - The Dockerfile that sets up the environment for building QEMU's
203   fuzzers on OSS-Fuzz can be fund in the OSS-Fuzz repository
204   __(https://github.com/google/oss-fuzz/blob/master/projects/qemu/Dockerfile)
205 
206 - The script responsible for building the fuzzers can be found in the
207   QEMU source tree at ``scripts/oss-fuzz/build.sh``
208 
209 Building Crash Reproducers
210 -----------------------------------------
211 When we find a crash, we should try to create an independent reproducer, that
212 can be used on a non-fuzzer build of QEMU. This filters out any potential
213 false-positives, and improves the debugging experience for developers.
214 Here are the steps for building a reproducer for a crash found by the
215 generic-fuzz target.
216 
217 - Ensure the crash reproduces::
218 
219     qemu-fuzz-i386 --fuzz-target... ./crash-...
220 
221 - Gather the QTest output for the crash::
222 
223     QEMU_FUZZ_TIMEOUT=0 QTEST_LOG=1 FUZZ_SERIALIZE_QTEST=1 \
224     qemu-fuzz-i386 --fuzz-target... ./crash-... &> /tmp/trace
225 
226 - Reorder and clean-up the resulting trace::
227 
228     scripts/oss-fuzz/reorder_fuzzer_qtest_trace.py /tmp/trace > /tmp/reproducer
229 
230 - Get the arguments needed to start qemu, and provide a path to qemu::
231 
232     less /tmp/trace # The args should be logged at the top of this file
233     export QEMU_ARGS="-machine ..."
234     export QEMU_PATH="path/to/qemu-system"
235 
236 - Ensure the crash reproduces in qemu-system::
237 
238     $QEMU_PATH $QEMU_ARGS -qtest stdio < /tmp/reproducer
239 
240 - From the crash output, obtain some string that identifies the crash. This
241   can be a line in the stack-trace, for example::
242 
243     export CRASH_TOKEN="hw/usb/hcd-xhci.c:1865"
244 
245 - Minimize the reproducer::
246 
247     scripts/oss-fuzz/minimize_qtest_trace.py -M1 -M2 \
248       /tmp/reproducer /tmp/reproducer-minimized
249 
250 - Confirm that the minimized reproducer still crashes::
251 
252     $QEMU_PATH $QEMU_ARGS -qtest stdio < /tmp/reproducer-minimized
253 
254 - Create a one-liner reproducer that can be sent over email::
255 
256     ./scripts/oss-fuzz/output_reproducer.py -bash /tmp/reproducer-minimized
257 
258 - Output the C source code for a test case that will reproduce the bug::
259 
260     ./scripts/oss-fuzz/output_reproducer.py -owner "John Smith <john@smith.com>"\
261       -name "test_function_name" /tmp/reproducer-minimized
262 
263 - Report the bug and send a patch with the C reproducer upstream
264 
265 Implementation Details / Fuzzer Lifecycle
266 -----------------------------------------
267 
268 The fuzzer has two entrypoints that libfuzzer calls. libfuzzer provides it's
269 own ``main()``, which performs some setup, and calls the entrypoints:
270 
271 ``LLVMFuzzerInitialize``: called prior to fuzzing. Used to initialize all of the
272 necessary state
273 
274 ``LLVMFuzzerTestOneInput``: called for each fuzzing run. Processes the input and
275 resets the state at the end of each run.
276 
277 In more detail:
278 
279 ``LLVMFuzzerInitialize`` parses the arguments to the fuzzer (must start with two
280 dashes, so they are ignored by libfuzzer ``main()``). Currently, the arguments
281 select the fuzz target. Then, the qtest client is initialized. If the target
282 requires qos, qgraph is set up and the QOM/LIBQOS modules are initialized.
283 Then the QGraph is walked and the QEMU cmd_line is determined and saved.
284 
285 After this, the ``vl.c:main`` is called to set up the guest. There are
286 target-specific hooks that can be called before and after main, for
287 additional setup(e.g. PCI setup, or VM snapshotting).
288 
289 ``LLVMFuzzerTestOneInput``: Uses qtest/qos functions to act based on the fuzz
290 input. It is also responsible for manually calling ``main_loop_wait`` to ensure
291 that bottom halves are executed and any cleanup required before the next input.
292 
293 Since the same process is reused for many fuzzing runs, QEMU state needs to
294 be reset at the end of each run. For example, this can be done by rebooting the
295 VM, after each run.
296 
297   - *Pros*: Straightforward and fast for simple fuzz targets.
298 
299   - *Cons*: Depending on the device, does not reset all device state. If the
300     device requires some initialization prior to being ready for fuzzing (common
301     for QOS-based targets), this initialization needs to be done after each
302     reboot.
303 
304   - *Example target*: ``i440fx-qtest-reboot-fuzz``
305