March 08, 2017

RPM Packaging Tutorial

Document revisions

Date Changes
13.03.2017 Added links about kmp drivers
08.03.2017 First release

What is a package

A package is a way of distributing software on Linux systems. A single application is distributed as one or more packages, usually the main package containing the program, and then some optional or secondary packages.

On some platforms, applications are self-contained into a directory. This makes installing an app just adding a folder, and uninstalling the app, removing it.

Linux systems tend to share as much as components as possible. This is due partly to some advantages of this philosophy, but it is due mostly because in the Linux eco-system, the whole universe is built by the same entity, except for a few 3rd party applications. This makes easy to assume that a library is available for all applications to consume. In an MacOS system, only the core comes from a single vendor, and all applications come from 3rd parties. It is therefore harder to make assumptions, and they tend to ship their own version of any depending component, with the exception of everything documented as the “platform”.

For example, if an application requires the SSL library and the Qt toolkit. On a Linux system, it will likely use both components from the Linux distribution, while the MacOS version will likely use the SSL library from the OS, but ship its own version of Qt, as it is not a standard MacOS library.

Anatomy of a package

Lets start with a famous UNIX tool: rsync.

So a package is an archive file:


containing all files related to the application:

$ rpm -qpl rsync-3.1.2-1.5.x86_64.rpm


plus some extra metadata. This metadata includes but it is not limited to:

  • Name, Summary, Description, License, etc.
$ rpm -qpi rsync-3.1.2-1.5.x86_64.rpm

Name        : rsync
Version     : 3.1.2
Release     : 1.5
Architecture: x86_64
Install Date: Wed 26 Oct 2016 01:31:12 PM CEST
Group       : Productivity/Networking/Other
Size        : 636561
License     : GPL-3.0+
Signature   : RSA/SHA256, Mon 17 Oct 2016 02:32:40 AM CEST, Key ID b88b2fd43dbdc284
Source RPM  : rsync-3.1.2-1.5.src.rpm
Build Date  : Mon 17 Oct 2016 02:32:26 AM CEST
Build Host  : lamb18
Relocations : (not relocatable)
Packager    :
Vendor      : openSUSE
URL         :
Summary     : Versatile tool for fast incremental file transfer
Description :
Rsync is a fast and extraordinarily versatile file  copying  tool. It can copy
locally, to/from another host over any remote shell, or to/from a remote rsync
daemon. It offers a large number of options that control every aspect of its
behavior and permit very flexible specification of the set of files to be
copied. It is famous for its delta-transfer algorithm, which reduces the amount
of data sent over the network by sending only the differences between the
source files and the existing files in the destination. Rsync is widely used
for backups and mirroring and as an improved copy command for everyday use.
Distribution: openSUSE Tumbleweed
  • What does the package requires to be also installed in order to work (Requires)
$ rpm -qp --requires rsync-3.1.2-1.5.x86_64.rpm
config(rsync) = 3.1.2-1.5
rpmlib(CompressedFileNames) <= 3.0.4-1
rpmlib(PayloadFilesHavePrefix) <= 4.0-1
rpmlib(PayloadIsLzma) <= 4.4.6-1

For example, a package may need a library, or an executable that is called during runtime.

  • What does the package provides for other packages to to work (Provides)
$ rpm -qp --provides rsync-3.1.2-1.5.x86_64.rpm
config(rsync) = 3.1.2-1.5
rsync = 3.1.2-1.5
rsync(x86-64) = 3.1.2-1.5

Installing packages

When a package is installed, the content (list of files) is placed on the system at the location of each file path relative to the root (/) folder.

Additionally, the metadata of the package and the fact that is installed is recorded in a system-wide database located in /var/lib/rpm, managed by the rpm tool, which is the tool that manages packages at the lowest level.

Packages can be installed with the rpm tools:

$ rpm -U rsync-3.1.2-1.5.x86_64.rpm

Once you do this, you can perform the same queries without specifying the -p option and using what is called the NVRA (name-version-release-architecture, rsync-3.1.2-1.5.x86_64) or a subset of it, eg. just name (rsync).

$ rpm -q --provides rsync

The rpm tool will not help you if the dependencies of the package are not met at installation time. It will just refuse to install the package to avoid having the system in an inconsistent state.

Features like automatically finding the required packages and retrieving them, are implemented in higher-level tools like zypper.

Dependency matching

You saw above that a package has a list of Requires and Provides. Those aren’t package names, but arbitrary symbols. A package can require any string of text and provide anything.

The main rule is that each package provides its own name. So the rsync package Provides: rsync.

You saw that rsync requires /bin/sh. While this looks like a file name, here it is an arbitrary symbol and the meaning is given by the whole distribution. Why it does not require a package named sh instead?. Well, this is for various reasons:

  • It provides a layer of indirection that makes the system cohesive.

/bin/sh is a capability provided by the bash package. This allows rsync to depend on any shell implementation as long as it provides that symbol.

The distribution build system will scan all executables a package installs in a system and inject automatically those Provides, so that the packager does not need to take care of them.

We will see later that the same is done with libraries. Instead of rsync depending on the glibc package. When glibc was built, the build system scanned the content, found /lib64/ and injected a Provides: into the glibc metadata. In the case of shared libraries it is not that important where they are located, as the linker configuration takes care of that. When the rsync package whas built (glibc had to be installed at that point to build it), the build system scanned the executable /usr/lib/rsync and realized it was linked against

$ ldd /usr/bin/rsync (0x00007ffccb34a000) => /lib64/ (0x00007fc406028000) => /usr/lib64/ (0x00007fc405e1b000) => /usr/lib64/ (0x00007fc405c02000) => /lib64/ (0x00007fc405863000) => /lib64/ (0x00007fc40565e000) => /lib64/ (0x00007fc4051c4000) => /lib64/ (0x00007fc404fa7000)
        /lib64/ (0x00005653cd048000) => /lib64/ (0x00007fc404da3000) => /lib64/ (0x00007fc404b8d000)

Therefore, it injected Requires: to the rsync package.

Compare it to other packaging systems. Package musicplayer requires libsound. /usr/bin/musicplayer links to /usr/lib64/ Later, musicplayer is rebuilt against a newer libsound, which is not published. The user installs musicplayer with no issues because it only Requires: libsound (as in package name). Then when she tries to run it:

$ musicplayer
error while loading shared libraries: cannot open shared object file: No such file or directory

The layer of indirection of automatically injected dependencies prevents this manual work of keeping dependencies in sync. Package provide what they really carry (because provides are injected by advanced scanners), and packages require what they really need (because requires are injected by scanning executables, scripts for shebangs, etc.

This makes rpm based distributions using these conventions highly cohesive. It makes less problematic to do upgrades without breaking your system. At the same time, the conventions and indirections between provides and requires, allow for packages to depend on more abstract capabilities, instead of specific package names (which sometimes get renamed, split, obsoleted, etc). Yes, you can be sure the vim package provides vi.

There are also other dependencies with more advances purposes: Conflicts, Obsoletes, etc. You may already guess what purposes they have.

Weak dependencies

Not everything is as strict. Sometimes a package works better if another package is present. Sometimes a package enhances the functionality of another package, however in neither case they are required. For this purpose, packages can have:

  • Recommends: a soft version of requires. If the recommended packages are not installed, the package will be installed anyway. Higher level tools however, may pull automatically recommended packages based on user settings. The reverse of this dependency is Supplements. For example a spellchecker could Supplements an office-suite package.
  • Suggests and Enhances are the forward and backward version of Recommends and Supplements in a weaker version.

Working with packages

For daily system administration and maintenance, the rpm tool is not sufficient. You will quickly fall into what is called the “dependency hell”. Downloading packages by hand in order to satisfy a dependency to quickly realize this new package also requires something else.

This problem is solved by a tool that implementes a solver. The solver takes:

  • The list of installed packages (and therefore all its dependencies)
  • The list of available packages
  • The user request (“install package foo”, “upgrade system”)

The solver performs an operation that is finding the best solution to a problem that has many solutions. Therefore “best” is defined by policies, user settings, the distribution itself, etc.

On SUSE systems, the solver is implemented by the libsolv project. This engine implements both a satisfiability algorithm and an efficient way to represent the problem in memory. It was developed originally by Michael Schroeder at SUSE, but nowadays it powers even other distribution package managers, like Fedora’s dnf.

The rest of the package manager includes:

  • Handling of package repositories
  • Checking the integrity of packages
  • Fetching remote packages
  • Reading and honouring user/system policies

This functionality in SUSE systems is implemented by the ZYpp library, which also includes a command-line tool called zypper. While tools like YaST also interact with ZYpp, on the console you will likely interact with zypper.

$ zypper install rsync-3.1.2-1.5.x86_64.rpm

Will, unlike rpm, figure out what else your system is missing, retrieve it, and then install all the required packages in the right order. It will also warn you if another package conflicts with what you are installing, of if the operation has more than one solution, and ask you for decisions on what to do.

Now, when we say “retrieve other packages”, the question arises. From where?


You can see that zypper can install a package directly from an rpm file. Now, if there is the need for installing dependencies or retrieving packages ie. when you upgrade a system, you will need a “library” of packages. This is what is called a repository. A repository is:

  • A collection of packages
  • A set of metadata files

The metadata is nothing more than the information present in the rpm file (Name, Description, Dependencies). The metadata allows the package manager to operate with the repository without having all rpm files locally. Every operation is processed used what we know about the package, and then rpm files are retrieved on demand at installation time.

$ zypper lr

# | Alias          | Name           | Enabled | GPG Check | Refresh
1 | non-oss        | NON-OSS        | Yes     | ( p) Yes  | Yes
2 | oss            | OSS            | Yes     | ( p) Yes  | Yes
3 | oss-update     | OSS Update     | Yes     | ( p) Yes  | Yes
4 | update-non-oss | Update Non-Oss | Yes     | ( p) Yes  | Yes

A system normally will have the following repositories:

  • The base repository, which contains all the distribution packages.
  • Additional modules, addons products or extensions.
  • An update repository for each base product or extension

Running list repositories with -u ie. zypper lr -u will show you the URI of the repository. eg. If you visit this URI, you will see:

  • a x86_64 directory containing all architecture-dependent packages (ie. ones that contain executables, shared libraries, etc)
  • a noarch directory containing architecture-independent packages (ie. ones containing data or scripts)
  • a repodata directory, containing the metadata for all packages.

The metadata for this type of repositories consists in a repodata/repomd.xml file index, that is signed (repomd.xml.asc) using a key already present in the original system. repodata/repomd.xml refers to other metadata file with their checksums. The most important is primary.xml which contains all package dependencies.

If you have a directory with rpm packages, you can create the metadata for them using the createrepo tool. After that you can serve that repository via HTTP.

If you have a directory with rpms you want to use a repository, you don’t need to add metadata. ZYpp allows to have a plain local directory as a repository, and will read the metadata directly from the rpm files into its cache.

Refreshing a repository

$ zypper ref

While the base repository of the distribution is normally immutable, repositories like the one containing updates get new content often. The meaning of refreshing a repository is to get the up to date version of the metadata locally, so that all operations (solving, retrieval) match the current content of the repository.

If a repository is out of date, it means the local metadata represents a previous version of the repository content. We would solve, and likely fetch packages, but those packages may not exists on the repository anymore, so you will get an error at retrieval time.

The list of repositories of the system is kept in /etc/zypp/repos.d. zypper provides most of repository operations in a safer way than messing with those files by hand.

During refresh, metadata is cached locally at /var/cache/zypp/raw and converted to an efficient format for solving operations in /var/cache/zypp/solv.


Services are a higher-level version of repositories. It is another index that lists repositores. When the system is subscribed from to a service, refreshing the service will result in a new list of repositories, and the package manager will add new ones or remove obsolete ones.

Services are used for example on SUSE Linux Enterprise with the SUSE Customer Center. A customer is subscribed to a service provided by SCC using a credential. The customer, based on his entitlements, can “activate” a new product. SUSE Customer Center knows about those activations, and on service refresh, it will provide a new list of repositories that includes the new activated product.

Services can be remote (like SCC), or locally, via a plugin installed on the system. The package manager asks the plugin for a list of repositories. It is up to the plugin to build that list. This is normally used for integration with other systems. The connectivity between zypper and Spacewalk/SUSE Manager was originally implemented using a local plugin.

Repository sources

If you are using SUSE Linux Enterprise, your repositories will appear after the SUSEConnect tool registers your product against the SUSE Customer Center.

If you are on openSUSE, the default installation will setup the base and update repositories. Additionally, there is a lot of content published by the community on the build service projects or via projects like packman.

SUSE Linux Enterprise users can take advantage of the community content via the Package Hub.

Other package manager operations

You can use zypper lu to list updates. zypper up to update them.

You can lock packages to avoid them being removed or pulled-in using zypper addlock zypper removelock. You can also list active locks with zypper locks.

The distribution upgrade operation dup is used to do destructive upgrades. This means packages may be suggested for removal as dependencies like Obsoletes are taken into account. It is usually used for upgrading major releases or to update rolling distributions like Tumbleweed. It has to be used with care.

Other solvable types (Products, Patterns, System)

The package manager solver loads all available and installed packages to do the solving. However, there are other entities similar to packages that also have dependencies.


Patterns are used to install a collection of software in a comfortable way. For example you can install a working Laptop-oriented system with:

$ zypper install -t pattern laptop

But where do patterns come from? They don’t exists on their own. The package managers creates them dynamically from packages names patterns-XXXXXX which have a special set of dependencies. So installing a pattern would actually install the package representing that pattern. The other way around is true, if you install the package representing the pattern, it will make the system look like the pattern is installed.

$ zypper info --provides patterns-openSUSE-laptop would show you some of the magic behind patterns (equivalent torpm -q --provides patterns-openSUSE-laptop).


Similar to patterns, products can be queried with:

$ zypper search -t product

Product come from a package called XXXXXX-release which has some special dependencies (rpm -q --provides openSUSE-release). The release package/product installs some information in /etc/products.d that is used by other tools get information about the base and addons products installed.


Patches are used for updates and described by the updateinfo.xml section of the metadata. They represent an entity that conflicts with older versions of one or more packages. Installing a patch does not install packages, but generates a conflict in the solver that ends with the affected version of packages being upgraded.

Patches also carry additional property, like the CVE identifiers of the issues they fix or links to bug tracker incidents.


During solving, there is one entity providing dependencies that is used to match locale and hardware information. If you have a WLAN card, the package manager will dynamically read /sys/devices and make this entity have provides like:

Provides :modalias(pci:v0000104Cd0000840[01]sv*sd*bc*sc*i*)

Then, a package providing a WLAN driver for some cards (eg. wlan-kmp-default), could have the following dependencies:

Supplements: modalias(kernel-default:pci:v0000104Cd0000840[01]sv*sd*bc*sc*i*)
Supplements: modalias(kernel-default:pci:v0000104Cd00009066sv*sd*bc*sc*i*)
Supplements: modalias(kernel-default:pci:v000010B7d00006000sv*sd*bc*sc*i*)

Which results in that at solving time, if the hardware is present, the driver will be selected automatically.

Something similar is done with translation packages and the current configured system locale.

Important notes

What is important to note is that all those types are only really present at solving time. In reality, your system is only packages, and all information comes from the installed packages. Every operation on patches, patterns and products result in a package operation. This is to make the package manager compatible with the lower level rpm tool.

Creating packages

Packages are created providing a so called .spec file. A spec file defines the attributes of the package, explicit dependencies (others are injected as we already mentioned), and how the content of the package is created. A very simple spec file would be:

Name:           mypackage
Version:        1.0
Release:        0
License:        MIT
Summary:        Dummy package
BuildRoot:      %{_tmppath}/%{name}-%{version}-build

Dummy text

mkdir -p %{buildroot}%{_datadir}/%{name}
touch %{buildroot}%{_datadir}/%{name}/CONTENT



This spec file just creates a directory /usr/share/mypackage and puts a dummy CONTENT file in it.

spec files are heavily defined by macros that make sure that paths and values are specified by the distribution. Those macros are shipped by the base distribution and are located in /usr/lib/rpm and /etc/rpm. Other packages may contribute more macros. For example the macros defined in /usr/lib/rpm/golang-macros.rb are provided by the golang-packaging package and are useful to create packages that use the Go language.

Common macros

When building spec files, you should be familiar with macros like %{_prefix}, %{_datadir}, %{_mandir}, %{_libdir}, %{_bindir}, etc. You can evaluate a macro like this:

$ rpm --eval "%{_libdir}"


Sometimes from a single source you will build multiple components that are independent of each other.

The sources for an Office Suite may result in:

  • A Word Processor
  • A Spreadsheet
  • Common libraries
  • Development files

For this, you can declare subpackages, which is an independent description and attributes section for each component. The build section is common, and then again in the %files section, you will declare which files go to each subpackage. In this case you would end with something like:

  • office-wordprocessor
  • office-spreadsheet
  • liboffice
  • office-devel

Building with rpmbuild

You can build a package with the rpmbuild tool. It requires the spec file to be in a specific location, so I usually tweak the standard configuration so that spec files are searched in the current directory:

$ cat ~/.rpmmacros
%topdir /space/packages
%_builddir %{topdir}/build
%_rpmdir %{topdir}/rpms
%_sourcedir %(echo $PWD)
%_specdir %(echo $PWD)
%_srcrpmdir %{topdir}/rpms

As you can see, I also configure it so that built packages are saved in /space/packages. Tweak that to your own preferences.

Once I have this setup, I can do:

$ rpmbuild -bb mypackage.spec
Executing(%install): /bin/sh -e /var/tmp/rpm-tmp.lVzwnj
+ umask 022
+ cd /space/packages/build
+ mkdir -p /home/duncan/rpmbuild/BUILDROOT/mypackage-1.0-0.x86_64/usr/share/mypackage
+ touch /home/duncan/rpmbuild/BUILDROOT/mypackage-1.0-0.x86_64/usr/share/mypackage/CONTENT
+ /usr/lib/rpm/brp-compress
+ /usr/lib/rpm/brp-suse
Processing files: mypackage-1.0-0.x86_64
Provides: mypackage = 1.0-0 mypackage(x86-64) = 1.0-0
Requires(rpmlib): rpmlib(CompressedFileNames) <= 3.0.4-1 rpmlib(PayloadFilesHavePrefix) <= 4.0-1
Checking for unpackaged file(s): /usr/lib/rpm/check-files /home/duncan/rpmbuild/BUILDROOT/mypackage-1.0-0.x86_64
Wrote: /space/packages/rpms/x86_64/mypackage-1.0-0.x86_64.rpm
Executing(%clean): /bin/sh -e /var/tmp/rpm-tmp.0xLGri
+ umask 022
+ cd /space/packages/build
+ /usr/bin/rm -rf /home/duncan/rpmbuild/BUILDROOT/mypackage-1.0-0.x86_64
+ rm -rf filelists

Now I can verify the content of the package:

% rpm -qpl /space/packages/rpms/x86_64/mypackage-1.0-0.x86_64.rpm

As you can see, everything that we put into the %{buildroot} ended as content of the package.

Now the term “building a package” can have two meanings. One is assembling the package from existing content. You could build your application in Jenkins, take the built artifacts and use the spec file to package it.

However, where rpm really shines is that you can build the application in the spec file itself, and use the distribution and dependencies to setup the build environment.

A common use case to illustrate this is the typical Linux application built with configure && make && make install. Lets try to build a package for gqlplus, an alternative client for Oracle databases.

Provided that you have readline and ncurses development headers, you can build this application just unpacking the tarball and performing the commands mentioned above. Some programs require an extra step with autoconf in order to generate the configure script. This is specific to buildin software and has nothing to do with packaging. Certainly builting a Qt based program will be a different experience, and if you build a Java application, you will have to deal with other tools.

When you do ./configure you would need to pass the right --prefix. This is where macros help you. You could do configure --prefix=%{_prefix}, however, there is a better macro called %configure which takes care and sets most of the configure options (try expanding it: echo $(rpm --eval '%configure'))

Now, the package can’t build if some libraries are not present, a C compiler is there, and the basic build tools (make) are not available. That is what BuildRequires are for. They define what packages are needed for building, but not necessarily at runtime.

On the other hand, the original oracle-instantclient-sqlplus package is required at runtime, but we don’t need it to build our package.

Name:           gqlplus
Version:	    1.15
Release:	    0
License:	    GPL-2.0
Summary:	    A drop-in replacement for sqlplus, an Oracle SQL client
Group:		    Productivity/Databases/Clients
Source0:	    %{name}-%{version}.tar.bz2
BuildRequires:  readline-devel
BuildRequires:  ncurses-devel
BuildRequires:  gcc make autoconf automake
BuildRoot:      %{_tmppath}/%{name}-%{version}-build
Requires:       oracle-instantclient-sqlplus
GQLPlus is a drop-in replacement for sqlplus, an Oracle SQL client, for UNIX and UNIX-like platforms. The difference between GQLPlus and sqlplus is command-line editing and history, plus table-name and column-name completion.

%setup -q

aclocal && autoconf
automake --add-missing
make %{?_smp_mflags}




The Source0 section specifies a source that you can refer later using the %SOURCE0 or %{S:0} macros. You can have more than one source (Source1, etc).

The prep section uses the %setup macro to unpack the sources. You could as well operate directly on the source files if you need to do something unconventional.

As we need make install to install the files inside %{buildroot}, we should call make install DESTDIR=%{buildroot}, but %makeinstall is just a macro for that.

The files section list the files rpmbuild should expect to find inside %{buildroot} that will be the content of the package.

Note also that we don’t need to add a runtime Requires to the readline and ncurses libraries. Because the executable is linked against the ones installed by the -devel packages, it will be scanned and the right Requires injected:

$ rpm -qp --requires gqlplus-1.15-0.x86_64.rpm

This symbols are provided by the right package, so the solver will match them:

rpm -q --whatprovides ''

For more information on how to build packages for various types of software, visit the openSUSE Packaging Guidelines.

Building in a real build environment

Building this way means the build environment is our system. If a package is in BuildRequires, you will have to install it in your system first.

If the software you are building links against some library only if it is available, even if you don’t mention it in your BuildRequires, if that library is present in your system, it will taint the build and make configure find it.

What if you want to build against only the packages that are in the build requirements?.

The Open Build Service

The Open Build Service allows to build packages for multiple distributions and architectures. Visit the Material section of the website for a deeper introduction. For the package we are building, you can get an account at the openSUSE Build Service instance. Go to your “Home Project”, “Create New Package”. Upload the spec and sources.

You need then to configure some target distributions for your home project. That can be one base distribution, or another project. This shows the power by allowing building based on layers that can override things from previous layers.

Add the most popular SUSE distributions (latest Leap and Tumbleweed) and your package will be built automatically. A repository will be published automatically and made available for public consumption.

Every time the sources changes, the package will be rebuilt, and if you have more packages in the same project, they will be rebuilt in the right order, and re-published.

gqlplus on OBS

Open Build Service can not only build packages, but also images from those packages. All SUSE products and the openSUSE distributions are built using the Open Build Service. Contributors basically submit new sources, and the Open Build Service takes care of assembling it all (openQA takes later care that it works).

Using the Build Service locally

With the osc tool you can checkout packages from OBS, make changes and resubmit them.

$ osc co home:dmacvicar gqlplus
A    home:dmacvicar
A    home:dmacvicar/gqlplus
A    home:dmacvicar/gqlplus/gqlplus-1.15.tar.bz2
A    home:dmacvicar/gqlplus/gqlplus.changes
A    home:dmacvicar/gqlplus/gqlplus.spec
At revision 4.

The most interesting feature is the ability to build locally. “We already did that!” you may think (rpmbuild). However, osc allows you to build in an issolated environment (either a chroot jail or a virtual machine), setting up that environment automatically using the BuildRequires of the spec file, and also allowing you to build against a different distribution than the one you are running.

$ cd home:dmacvicar/gqlplus
$ osc build openSUSE_Leap_42.2

Improving the package

When you build a package in the build service, you will find out that in addition to the automated magic that injects dependencies, there is a bunch of checks being done to the package.

Yes, those checks are quite pedantic. It is the only way to ensure quality and consistency when a product is assembled from thousands of sources by hundred of contributors.

The spec-cleaner tool can help you keeping your spec file in shape:

$ spec-cleaner -i gqlplus.spec

For example, it can help you converting BuildRequires: foo-devel dependencies to BuildRequires: pkgconfig(foo). If a -devel package installs a pkg-config module, a Provides: pkgconfig(foo) is automatically injected. If the build process (./configure or Makefile) uses pkg-config to find the software, it makes more sense and it is closer to reality to depend on pkgconfig(foo) being present, regardless of which -devel package provides it.

For post-build checks, you can get more information about how to fix them in the openSUSE Packaging Checks page.


Until now we have left the %changelog section empty. Some distributions write there the history for the package. SUSE-flavored distributions keep the changelog in a separate .changes file. To quickly generate or update it, you can use osc vc in the directory containing the spec file and the sources.