如何评估模糊测试工具-unibench的使用

news2024/9/24 0:21:55

unibench是一个用来评估模糊测试工具的benchmark。这个benchmark集成了20多个常用的测试程序，以及许多模糊测试工具。

这篇文章（https://zhuanlan.zhihu.com/p/421124258）对unibench进行了简单的介绍，本文就不再赘诉，而是侧重于介绍unibench具体是如何用来评估模糊测试工具。

关于unibench更详细的介绍，可以去看2021年发表在Usenix Security上的论文：[https://nesa.zju.edu.cn/download/UNIFUZZ A Holistic and Pragmatic Metrics-Driven Platform for Evaluating Fuzzers.pdf](https://nesa.zju.edu.cn/download/UNIFUZZ A Holistic and Pragmatic Metrics-Driven Platform for Evaluating Fuzzers.pdf)

unibench可以在这里找到：https://github.com/unifuzz

下面将介绍以使用unibench来评估AFL这一经典的模糊测试工具为例来介绍过一下unibench的流程。

1. 构建AFL和测试程序的镜像

运行下面命令，构建AFL和unibench的20个程序的镜像。下面命令需要较长的时间才能完成。

git clone https://github.com/unifuzz/unibench_build.git
cd unibench/afl
docker build -t .

除了上面的方法，unifuzz的作者也在dockerhub上传了build好的镜像可以直接使用。具体可以查看该链接：https://hub.docker.com/r/unifuzz/unibench

docker pull unifuzz/unibench:afl

然后运行下面命令验证是否装好

 docker run -it --rm unifuzz/unibench:afl strings /d/p/justafl/exiv2 |grep afl|head

若出现下面的提示信息，则说明已装好

__afl_global_area_ptr
__afl_maybe_log
__afl_area_ptr
__afl_setup
__afl_store
__afl_prev_loc
__afl_return
__afl_setup_failure
__afl_setup_first
__afl_setup_abort

2. 运行模糊测试实验

2.1 系统配置

运行实验前，先进行基础的系统配置

echo "" | sudo tee /proc/sys/kernel/core_pattern # disable generating of core dump file
echo 0 | sudo tee /proc/sys/kernel/core_uses_pid
echo performance | sudo tee /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor
echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
echo 1 | sudo tee /proc/sys/kernel/sched_child_runs_first # tfuzz require this
echo 0 | sudo tee /proc/sys/kernel/randomize_va_space # vuzzer require this

如果出现No such file or directory的错误信息也没关系，可以无视。

2.2 运行实验

前面搭建好的镜像中已经使用afl-clang插过桩了，所以只需要使用docker run命令来跑afl-fuzz，就可以运行模糊测试的实验了。

mkdir work # 在宿主机创建一个文件夹保留AFL跑实验的结果
cd work 
git clone https://github.com/unifuzz/seeds.git # 获取测试程序所需的种子
docker run -it --cpus=1 -w /work -v `pwd`:/work --privileged unifuzz/unibench:afl \
afl-fuzz -i /work/seeds/general_evaluation/mp3 -o output -m none -t 500+ \
--/d/p/justafl/mp3gain @@

下面介绍下docker run各个选项的意思：

-it：指定交互式终端，并分配伪终端
—cpus=1：只分配一个CPU核
-w /work：指定容器内工作目录为/work
-v pwd:/work ：挂载当前目录（宿主机）为容器的/work目录，使容器内部对/work的访问实际上是对宿主机当前目录的访问
—privileged：赋予容器运行时的特权

然后第二行开始的就是afl-fuzz的选项：

-i /work/seeds/general_evaluation/mp3：指定种子目录为/work/seeds/general_evaluation/mp3
-o output：指定输出目录为 output
-m none：不限制内存
-t 500+：处理种子样例所需的时间

第三行可以理解为AFL会不断fork进程运行第三行的命令，@@则是模糊测试生成的测试样例的占位符。

unibench也给出了其他程序的命令行参数。

data = [
    #id, prog, commandline, seed_folder
    [1, "exiv2", "@@", "jpg"],
    [2,"tiffsplit","@@","tiff"],
    [3,"mp3gain","@@","mp3"],
    [4,"wav2swf","-o /dev/null @@","wav"],
    [5,"pdftotext","@@ /dev/null","pdf"],
    [6,"infotocap","-o /dev/null @@","text"],
    [7,"mp42aac","@@ /dev/null","mp4"],
    [8,"flvmeta","@@","flv"],
    [9,"objdump","-S @@","obj"],
    [14, "tcpdump", "-e -vv -nr @@", "tcpdump100"],
    [15, "ffmpeg", "-y -i @@ -c:v mpeg4 -c:a copy -f mp4 /dev/null", "ffmpeg100"],
    [16, "gdk-pixbuf-pixdata", "@@ /dev/null", "pixbuf"],
    [17, "cflow", "@@", "cflow"],
    [18, "nm-new", "-A -a -l -S -s --special-syms --synthetic --with-symbol-versions -D @@", "nm"],
    [19, "sqlite3", " < @@", "sql"],
    [20, "lame3.99.5", "@@ /dev/null", "lame3.99.5"],
    [21, "jhead", "@@", "jhead"],
    [22, "imginfo", "-f @@", "imginfo"],
    [23, "jq", ". @@", "json"],
    [24, "mujs", "@@", "mujs"],
    # below is the LAVA-M settings
    [10,"uniq","@@","uniq"],
    [11,"base64","-d @@","base64"],
    [12,"md5sum","-c @@","md5sum"],
    [13,"who","@@","who"],
]

运行完上面命令，就会显示AFL的界面。
在这里插入图片描述

如果想后台运行实验，docker run的时候加上-d选项即可

docker run -itd --cpus=1 -w /work -v `pwd`:/work --privileged unifuzz/unibench:afl \
afl-fuzz -i /work/seeds/general_evaluation/mp3 -o output -m none -t 500+ \
--/d/p/justafl/mp3gain @@