rafaenoor/occam

OCCAM: Object Culling and Concretization for Assurance Maximization

Description

OCCAM is a whole-program partial evaluator for LLVM bitcode that aims at debloating programs and shared/static libraries running in a specific deployment context.

OCCAM architecture

Docker

A pre-built and installed version of OCCAM can be obtained using Docker:

docker pull sricsl/occam:bionic
docker run -v `pwd`:/host -it sricsl/occam:bionic

Alternatively, it can be built and installed from source as follows.

Prerequisites

OCCAM currently works on Linux, macOS, and FreeBSD. It depends on an installation of LLVM. OCCAM currently requires llvm-5.0. You will also need the Google protocol buffer compiler protoc and the corresponding Python package.

If you need to generate application bitcode (that OCCAM operates on), you will want to install WLLVM, either from the the pip package or the GitHub repository.

The test harness also requires lit and FileCheck. FileCheck can often be found in the binary directory of your LLVM installation. However, if you built your own, you may need to read this. Hint: the build produces it, but does not install it. (Try locate FileCheck, then copy it to the bin directory.)

Detailed configuration instructions for Ubuntu 14.04 can be gleaned from bootstrap.sh as well as the Travis CI scripts for each branch .travis.yml.

Building and Installing

Set where OCCAM's library will be stored:

  export OCCAM_HOME={path to location in your home directory}

Point to your LLVM's location, if non-standard:

  export LLVM_HOME=/usr/local/llvm-5.0
  export LLVM_CONFIG=llvm-config-5.0

Set where system libraries, including Google Protocol Buffers, are located:

  export LD_FLAGS='-L/usr/local/lib'

Clone, build, and install OCCAM with:

  git clone --recurse-submodules https://github.com/SRI-CSL/OCCAM.git
  make
  make install
  make test

Using OCCAM

You can choose to record logs from OCCAM by setting the following variables:

  export OCCAM_LOGFILE={absolute path to log location}
  export OCCAM_LOGLEVEL={INFO, WARNING, or ERROR}

Using razor

razor is a pip package that relies on the same dynamic library as occam. So you should first build and install occam as described above. razor provides the commandline tool slash for end users. You can either install razor from this repository, or you can use:

pip install razor

To install an editable version from this repository:

make -f Makefile develop

This may require sudo priviliges. Either way you can now use slash:

slash [--work-dir=<dir>]  [--force] [--no-strip] [--intra-spec-policy=<type>] [--inter-spec-policy=<type>] <manifest>

where

type=none|aggressive|nonrec-aggressive

The value none will prevent any inter or intra-module specialization. The value aggressive specializes a call if any parameter is a constant. The value nonrec-aggressive specializes a call if the function is non-recursive and any parameter is a constant.

To function correctly slash calls LLVM tools such as opt and clang++. These should be available in your PATH, and be the currently supported version (5.0). Like wllvm, slash, will pay attention to the environment variables LLVM_OPT_NAME and LLVM_CXX_NAME if your version of these tools is adorned with suffixes.

The Manifest(o)

The manifest for slash should be valid JSON. The following keys have meaning:

• main : a path to the bitcode module containing the main entry point.

• modules: a list of paths to the other bitcode modules needed.

• binary : the name of the desired executable.

• native_libs : a list of flags (-lm, -lc, -lpthread) or paths to native objects (.o, .a, .so, .dylib)

• ldflags: a list of linker flags such as --static, --nostdlib

• args : the list of arguments you wish to specialize in the main() of main.

• constraints : a list consisting of a positive integer, followed by some number of strings. The number indicates the expected number of arguments the specialized program will receive, and the remaing strings are the specialized arguments to the original program.

Note that args and constraints are mutually exclusive. If you use one you should not use the other.

As an example, (see examples/linux/apache), to previrtualize apache:

{ "main" : "httpd.bc"
, "binary"  : "httpd_slashed"
, "modules"    : ["libapr-1.so.bc", "libaprutil-1.so.bc", "libpcre.so.bc"]
, "native_libs" : ["-lcrypt", "-ldl", "-lpthread"]
, "args"    : ["-d", "/var/www"]
, "name"    : "httpd"
}

Another example, (see examples/linux/musl_nweb), specializes nweb with musl libc.c:

{ "main" :  "nweb.o.bc"
, "binary"  : "nweb_razor"
, "modules"    : ["libc.a.bc"]
, "native_libs" : ["crt1.o", "libc.a"]
, "ldflags" : ["-static", "-nostdlib"]
, "args"    : ["8181", "./root"]
, "name"    : "nweb"
}

A third example, (see examples/portfolio/tree), illustrates the use of the constraints field to partially specialize the arguments to the tree utility.

{ "main" : "tree.bc"
, "binary"  : "tree"
, "modules"    : []
, "native_libs" : []
, "ldflags" : [ "-O2" ]
, "name"    : "tree"
, "constraints" : ["1", "tree", "-J", "-h"]
}

The specialized program will output its results in JSON notation (-J) that will include a human readable size field (-h). The specialized program expects one extra argument, either a directory or another flag to output the contents of the current working directory.

---

This material is based upon work supported by the National Science Foundation under Grant ACI-1440800. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.