This repository contains 26 programs used by Moss. If you want to know how to use this benchmark, please refer to Moss.
Note: The Dockerfile provided in Moss already includes all the adjustments. Here is just a detailed explanation of each adjustment we made.
Our adjustments: Since docker containers default to the root user, some of the test cases for chown-8.2 involves operations related to the root user. To address this, we created a new user group and user named mychown, and adjusted the test cases to serve more effectively as oracles.
What you need to do before debloating:
Before debloating chown-8.2, you must create a user mychown that belongs to a user group mychown.
Our adjustments: Due to subprocess issues, we were unable to fully trace all the code covered by the test cases in vim-5.8 using LLVM. After manual analysis, we found that it was because vim-5.8 creates the subprocess using fork at line 100313, which prevented LLVM from successfully tracing the subprocess's code. Therefore, we manually created a coverage version of the vim-5.8 for debloating.
What you need to do before debloating:
①You need to download ncurses library and expect.
apt install libncurses-dev expect parallel
②You need to set timeout more than 1 second (in test.sh and generate_cov.py), because open file in vim cost some time.
③Vim use expect script, and it launch many vim-5.8 thread without tty. Take careful to close them. We have used cleanup script in vim-5.8 directory to do this. Please modify the path in cleanup.
Our adjustments: We were also unable to correctly obtain the bash-2.05 code covered by all test cases using LLVM, but we cannot confirm which code needs to be added back through debugging. On one hand, debugging requires using clang -g, while Moss uses clang -O3 during debloating. Therefore, just because bash-2.05 works normally with clang -g doesn't mean it will work correctly with clang -O3. On the other hand, since bash-2.05's test cases are interactively verified based on console output, we cannot use instrumentation for manual debugging. Consequently, we only use the 845 test cases that can be correctly handled by the coverage version of bash-2.05 to calculate the generality of bash-2.05.
What you need to do before debloating:
You need to modify the path of lib in compile.sh.
Our adjustments: None.
What you need to do before debloating:
①You need to modify the path of lib in compile.sh.
②date-8.21 and make-3.79 have crisis. date-8.12 may change the system time which is used in make's testscripts. Don't run them together.
Our adjustments: We use rsync to remove garbage files in /, as the debloated program of mkdir-5.2.1 may generated lots of garbage files in /.
What you need to do before debloating:
Before running mkdir-5.2.1, you must run apt install rsync.
Our adjustments: We use chattr +/-i to keep /bin, /usr, and working directory unchanged while debloating, as the debloated programs of these four programs may mistakenly compress files in these folders.
What you need to do before debloating:
None.
Our adjustments: Due to the impact of long comment blocks on parsing code, we use postgresql-12.14/deletecomments.py to delete all comments in the source code.
What you need to do before debloating:
Use "/postgresql-12.14" (absolute path) as working directory, or you need to change path in postgresql-12.14/path_generator/generate_cov.py and postgresql-12.14/cleanup to your path.
PROGNAME: The program name.version: The debloating strategy (3 is Moss, 2 is Moss-s1, 1 is Moss-s2, 0 is Moss-s3).debop_samplenum: The sample number of the second and third stages.domgad_samplenum: The sample number of the first stage.alphas: Alpha value (weight for attack surface reduction).ks: K value (for computing density values).betas: Beta value (weight for generality).DEBOP_DIR: The path of CovBlock_Stmt.DOMGAD_DIR: The path of CovPath.COV: The path of cov.iternum: The iteration number of each stage of Moss.realorcov: Use real or top program as the baseline.
Please refer to here.