FUZZING

# Introduction and setup

Pidgin has fuzzing support for libpurple via
[Libfuzzer](https://llvm.org/docs/LibFuzzer.html). If you're new to fuzzing with
libfuzzer, there is a fantastic tutorial available
[here](https://github.com/google/fuzzing/blob/master/tutorial/libFuzzerTutorial.md).

The fuzzers reside in libpurples/fuzzers. To build them, you'll need to specify
`clang` as your C compiler as well as pass `--enable-fuzzing` to `./configure`.
Once this is done you can `cd libpurple/fuzzers` and run `make check` to build
the fuzzers.

Example:
```bash
$ CC=clang ./configure --enable-fuzzing --disable-cyrus-sasl --disable-gtkui --disable-gstreamer --disable-vv --disable-idn --disable-meanwhile --disable-avahi --disable-libgadu --disable-dbus --disable-libsecret --disable-gnome-keyring --disable-kwallet --disable-plugins
```

Now that the build system has been configured, you need to build everything,
including the fuzzers. You can do this with the following command. Note that the
`-j $(nproc)` tells make to build with all available cores and is recommended
but optional.

```bash
$ make -j $(nproc) check
```

Now that the fuzzers are built, you can run them directly. There is also an
optional `-dict` parameter that can be used to specify a dictionary to be used
during the process. Also all fuzzers must have a basic corpus to help the
fuzzer find values and should be located in the fuzzers/corpus/<fuzzer-name>
directory.

```bash
$ ./fuzz_xmlnode -dict=dictionaries/xml.dict corpus/xmlnode
```

# Useful options

Because Libfuzzer is a sophisticated program, here are some handy options that
are available in all fuzzers.

 * **-help=1** Print help.
 * **-jobs=1** Number of jobs to run. If jobs >= 1 this will spawn that many jobs in separate worker processes with stdout/stderr redirected to fuzz-JOB.log.
 * **-workers=0** Number of simultaneous worker processes to run the jobs. If zero, `min(jobs,NumberOfCpuCores()/2)` is used.
 * **-max_len=0** Maximum length of the test input. If 0, libFuzzer tries to guess a good value based on the corpus and reports it.

# Adding more fuzzers

Of course, having more fuzzers and covering more areas of the code base is
always a good thing. It's simple to incorporate a fuzzer into the current build
system! If you open the `Makefile.am` file in `libpurple/fuzzers` you'll see a
`check_PROGRAMS` variable, you have to add the name to your new fuzzing harness
in there.

Example:

```
fuzz_programs=\
	fuzz_html_to_xhtml \
	fuzz_jabber_caps \
	fuzz_jabber_id_new \
	fuzz_markup_strip_html \
	fuzz_mime \
	fuzz_xmlnode \
	fuzz_newfuzzer # This is the newly added fuzzer
```

You'll also need to define the sources, which we can do by copying and changing
the lines from an existing fuzzer.

For example we have a `fuzz_xmlnode.c` fuzzer, these are the lines that define
the sources and the flags:

```
fuzz_xmlnode_SOURCES=fuzz_xmlnode.c
fuzz_xmlnode_LDADD=$(check_libpurple_LDADD)
fuzz_xmlnode_CFLAGS=-fsanitize=fuzzer,address $(check_libpurple_CFLAGS)
```

You'll need to change the names of these to match the name of our new fuzzer and
add any necessary flags:

```
fuzz_new_SOURCES=fuzz_new.c
fuzz_new_LDADD=$(common_LDADD)
fuzz_new_CFLAGS=-fsanitize=fuzzer,address $(common_CFLAGS)
```

Now you must include your harness in `fuzz_new.c`, an example of a new harness
could be as follows:

```C
#include <glib.h>
#include <string.h>

#include <purple.h>

gint LLVMFuzzerTestOneInput(const guint8 *data, size_t size);

gint
LLVMFuzzerTestOneInput(const guint8 *data, size_t size) {
        gchar *malicious_input = g_new0(gchar, size + 1);

        memcpy(malicious_input, data, size);
        malicious_input[size] = '\0';

        function_you_want_to_fuzz(malicious_input);

        g_free(malicious_input);

        return 0;
}
```

Make sure to include the relevant headers and then run `make check`. This will
force an update of the build system and build everything that needs to be
rebuilt. If there were no issues, you should now be able to run your new fuzzer
from the `libpurple/fuzzers` directory.