Hase provides record-replay debugging suitable for all-ways-on in-production recording. It leverages intel processor trace and core dumps that can be recorded with little overhead. On top of that data it performs symbolic execution to recover states prior to the crash.
Workshop presentation on Klee Workshop 2018
- Setup virtualenv
$ virtualenv venv
$ . venv/bin/activate
- Install project into virtualenv
$ pip install -e .
Note: you may need to upgrade your pip >= 9.0.1
Additionally pyqt5 is required and cannot be installed via pip.
- Install test dependencies
$ pip install -e '.[test]'
- Patch the perf-script-sample-addr
git clone https://github.com/torvalds/linux
cd ./linux/tools/perf
cp path-to-your-hase-folder/perf-script-sample-addr.patch .
patch -p3 < perf-script-sample-addr.patch
make
sudo cp perf /usr/bin
Note: some new parse rules are applied recent days, so if you have intel_pt//u parse error, check this patch https://lkml.org/lkml/2018/5/7/94 and solve by git checkout an-eariler-commit-id
- Testing examples
sudo nosetests -w tests/test_record.py
$ sudo ./bin/hase record
Example crash
$ ./tests/bin/loopy/loopy
$ ls -la /var/lib/hase
.rw-rw-rw- 244 root 9 May 3:22 coredump.log
.rw-r--r-- 4 root 9 May 3:22 hase-record.pid
.rw-r--r-- 41M root 9 May 3:22 loopy-20180509T022227.tar.gz
$ sudo ./bin/hase record ls -al
Benchmarks require Pandas, which cannot be installed via pip. Use https://pandas.pydata.org/pandas-docs/stable/install.html instead or install it using your system package manager.