Warning message

Log in to add comments.

Reproducible Builds for Fedora

Red Hat published on 2013-09-18T13:30:32+00:00, last updated 2013-09-18T13:30:32+00:00

It should be possible to reproduce every build of every package in Fedora (strong, long-term goal).  It should be possible for the users to verify that the binary matches what the source intended to produce, in an independent fashion. This is the basic nature of open source, the source code is available, so what can we do with it?

I want to be able to show that our binary was the result of our source code from our compiler and nobody added anything to the binary along the way. Can we show that one of Fedora's RPMs was built from the source RPM shipped?  Fedora shouldn't be forced to say Trust Us when asked about proving the binary RPMs came from the source RPMs.

The steps below use the scripts stored in this git repository: https://github.com/kholia/ReproducibleBuilds

Steps Involved

There are many pieces of information that need to be provided to make builds reproducible.  Luckily many of these pieces already exist:


So how can you make this all happen?

sudo yum install python-celery python-requests koji 
      python-urlgrabber python-setuptools -y

sudo yum install redis python-redis -y  # required for parallel processing

sudo service redis start

sudo chkconfig redis on

# steal "group" information from upstream files

python get-comps.py

ln -s comps-f19.xml comps.xml  # use "comps-f20.xml" file or others if need be

Known Good Versions

Here are the software versions which are know to work 100%. Newer versions should be fine too.

  • python-celery-3.0.15-2.fc19.noarch
  • python-redis-2.7.2-2.fc19.noarch
  • redis-2.6.16-1.fc19.x86_64
  • python-urlgrabber-3.9.1-27.fc19.noarch
  • python-requests-1.2.3-5.fc19.noarch
  • python3-requests-1.2.3-5.fc19.noarch
  • python-setuptools-0.6.36-1.fc19.noarch

Please note that older versions of python-celery package are known to be broken on Fedora 19.

Process Flow

Initially, we have a NVR we want to verify (e.g. git-  Let us assume that we have the corresponding RPM file(s) as well which are provided by the vendor and which we want to verify.

First, obtain the SRPM corresponding to the target RPM file(s).  Next, we need to replicate the original buildroot.  Gather information about the buildroot corresponding to the NVR.

python get-mock-info.py git-

This step will populate the RPM "cache" and generate git- file which contains the buildroot replication information.  Make sure you have celery worker --autoscale=10,0 -A tasks running to do processing in the background.

Make repository which will be used by Mock.

./makerepo.sh git- myrepo

This command will create myrepo folder (which is our repository) by using the RPM "cache".  This command will also create a myrepo.cfg file which is a Mock configuration file.

Build the SRPM using mock.

mock -r myrepo --configdir=. --rebuild git-

Compare upstream build with our local build.

./rpm-compare /upstream/git- 

Current State

As of today, packages like git and john are 100% reproducible as far as code is concerned.  We also support "Recursive Verification".  For example, if building "Z" requires installing "Y" RPM, then, once we have verified that Z is OK, we can ask our tool to verify "Y" too and so on.

Current Challenges

Debian has been working on their own reproducible build problems.  Among the fixes are:

  • python-epydoc will add timestamps to the HTML file it produces (needs FIXING).
  • javadoc will add timestamps to the HTML file it produces (needs FIXING).


For additional information on Reproducible Builds please visit the following links:

It should be noted that this process assumes the toolchain is
trusted. The issue of proving toolchain trust is outside the
scope of this project.

About The Author

jnanadisplayedauthor's picture

Red Hat

This user is used for automation in Pantheon as part of the Docs publishing toolchain.