search

kalibera / rchk

103 stars 10 forks source link

readme

This project consists of several bug-finding tools that look for memory protection errors in C source code using R API, that is in the source code of R itself and packages. The tools perform whole-program static analysis on LLVM bitcode and run on Linux. About 200-300 memory protection bugs have been found using rchk and fixed in R. rchk is now regularly used to check CRAN packages.

To use the tool, one needs to build R from source using a special compiler wrapper, which builds LLVM bitcode in addition to native code (both shared libraries and executables). R packages are then installed using this version of R, providing LLVM bitcode for their shared libraries as well. The core of rchk is implemented in C++ and analyzes the LLVM bitcode of R packages and R itself. Several installation options are provided, including containers.

Installation

The tool is available in a pre-built Docker container, for non-interactive use. The container is invoked as a command to check a particular package:

docker pull kalibera/rchk:latest
docker run kalibera/rchk:latest audio

mkdir packages
cp lazy_1.2-16.tar.gz packages
docker run -v `pwd`/packages:/rchk/packages kalibera/rchk:latest /rchk/packages/lazy_1.2-16.tar.gz

For more details, see Docker rchk container. This setup is good for occasional checking of a single package. Docker clients are available for Linux, macOS and Windows.

The tool can also be used interactively in a virtual machine running Ubuntu, which can be automatically installed using Vagrant scripts. This setup is good for Linux, Windows and macOS users and makes it faster to repeatedly check the same package and easier to customize the process. See Automated installation (Docker/Virtualbox) for interactive use.

Finally, the tool can be installed natively on Linux, compiled from source. This setup is good for interactive use and reduces disk space overhead. The setup is not automated, but only requires several steps described for recent Linux distributions (on latest distributions with LLVM > 14, one however has to compile LLVM 14 from source). See Native installation on Linux for interactive use.

An alternative docker image is also available from third parties on R-hub (rhub/ubuntu-rchk, source).

Checking the first package (interactive use)

This part applies to interactive installation of rchk (natively or automated install in Docker/Virtualbox). For this that one also needs to install subversion, rsync (apt-get install subversion rsync, but already available in the automated install). More importantly, one also needs any dependencies needed by that package.

  1. Build R producing also LLVM bitcode
    • svn checkout https://svn.r-project.org/R/trunk
    • cd trunk
    • . ../scripts/config.inc (in automated install, . /opt/rchk/scripts/config.inc)
    • . ../scripts/cmpconfig.inc (in automated install, . /opt/rchk/scripts/cmpconfig.inc)
    • ../scripts/build_r.sh (in automated install, /opt/rchk/scripts/build_r.sh)
  2. Install and check the package
    • echo 'install.packages("jpeg",repos="http://cloud.r-project.org")' | ./bin/R --no-echo
    • ../scripts/check_package.sh jpeg (in automated install, /opt/rchk/scripts/check_package.sh jpeg)

The output of the checking is in files packages/lib/jpeg/libs/jpeg.so.*check. For version 0.1-8 of the package, jpeg.so.maacheck includes

WARNING Suspicious call (two or more unprotected arguments) to Rf_setAttrib at read_jpeg /rchk/trunk/packages/build/IsnsJjDm/jpeg/src/read.c:131

which is a true error. bcheck does not find any errors, jpeg.so.bcheck only contains something like

Analyzed 15 functions, traversed 1938 states.

Version 0.1-10 of the package no longer has the error, jpeg.so.bcheck currently contains something like

ERROR: too many states (abstraction error?) in function strptime_internal
ERROR: too many states (abstraction error?) in function StringValue
ERROR: too many states (abstraction error?) in function RunGenCollect
ERROR: too many states (abstraction error?) in function tre_tnfa_run_parallel
Analyzed 17 functions, traversed 815 states.

Errors about "too many states" can be ignored, this means that the tool could not analyze some R functions in the memory limit provided.

To check the next package, just follow the same steps, installing it into this customized version of R. When checking a tarball, one would typically first install the CRAN/BIOC version of the package to get all dependencies in, and then use R CMD INSTALL to install the newest version to check from the tarball.

Warnings like objdump: Warning: Unrecognized form: 0x22 should be safe to ignore.

One can reduce the number of required R package dependencies by only installing LinkingTo dependencies of the package and then installing the package with --libs-only option (only shared libraries are built and installed). This is enough to build shared libraries of most but not all packages. The Docker container for non-interactive use does this, see scripts/utils.r and definitions of the container for more details.

Further information: