Quick Debian development guide
This guide is also available under the URL https://deb.li/quickdev and as a video presentation https://www.youtube.com/watch?v=O83rIRRJysA. This is a living, changing document, although the video, obviously, isn't.
Introduction
This guides aims to kickstart people with working in existing Debian packages, either to backport software, patch existing packages or work on security issues as part of the security team or the LTS project.
This guide assumes that the Debian package already exists and you are making modifications to it. If you wish to make a new Debian package, there are 3 different guides that you can follow, and I am not going to create a fourth one (even this guide is duplicating existing efforts already). So go see one of those:
- Debian New Maintainer's Guide: the one I followed originally
- Guide for Debian Maintainers: a new version of the above, basically using the new debmake instead of dh-make, untested
- Introduction to Debian packaging: a set of slides, a good primer for people that like slides, although a bit technical
All those are part of the developer's manual suite which also includes the Debian policy and the Debian developer's reference, two more reference manuals which you may find useful when looking for more information.
Minimal packaging workflow
This guides tries to take an opinionated approach to maintaining
Debian packages. It doesn't try to cover all cases, doesn't try to
teach you about debhelper, cdbs, uscan or make. It
assumes you will find that information elsewhere, for example in the
above references, and that you are already somewhat familiar with
Debian systems administration (you know how to use a shell) and Debian
packages as a concept (you know what a .deb
file is and know how to
use dpkg -i
).
This will guide you through a standardized approach to:
- download Debian package source code
- make modifications to packages
- build packages, especially with multiple environments (testing, unstable, backports)
- upload packages
It covers a workflow that could be summarily described by this diagram:
Find the source
In the following, I take the example of building a backport of the Calibre package, which I needed. It's a good example because it does not use a git repository to track the Debian package source code, but Bazaar, which I am not familiar enough with to feel comfortable working on.1
To get the source code on an arbitrary package, visit the package
tracker.2 In this case, we look at the Calibre package
tracker page and find the download links for the release we're
interested in. Since we are doing a backport, we use the testing
download link. If you are looking for an packages not in a
distribution or antique package, you can also find download links on
snapshot.debian.org and archive.debian.net.
debian:
calibre | 0.7.7+dfsg-1squeeze1 | squeeze | source, all
calibre | 0.8.51+dfsg1-0.1 | wheezy | source, all
calibre | 1.22.0+dfsg1-1~bpo70+2 | wheezy-backports | source, all
calibre | 2.5.0+dfsg-1 | jessie-kfreebsd | source, all
calibre | 2.5.0+dfsg-1 | jessie | source, all
calibre | 2.55.0+dfsg-1 | stretch | source, all
calibre | 2.55.0+dfsg-1 | sid | source, all
ubuntu:
calibre | 0.8.38+dfsg-1 | precise/universe | source, all
calibre | 1.25.0+dfsg-1build1 | trusty/universe | source, all
calibre | 1.25.0+dfsg-1ubuntu1 | trusty-updates/universe | source, all
calibre | 2.20.0+dfsg-1 | vivid/universe | source, all
calibre | 2.33.0+dfsg-1build1 | wily/universe | source, all
calibre | 2.55.0+dfsg-1 | xenial/universe | source, all
calibre | 2.55.0+dfsg-1build1 | yakkety/universe | source, all
To get the Ubuntu results, I added the following line to my
~/.devscripts
file:
RMADISON_DEFAULT_URL=debian,ubuntu
What we are looking for is the calibre_2.55.0+dfsg-1.dsc file, the
"source description" file for the 2.55.0+dfsg-1
version that is
currently in stretch. Using that .dsc
file, we can get all we need
to build the package. So the first step is to download the source
code, using dget(1):
$ dget http://httpredir.debian.org/debian/pool/main/c/calibre/calibre_2.55.0+dfsg-1.dsc
dget: retrieving http://httpredir.debian.org/debian/pool/main/c/calibre/calibre_2.55.0+dfsg-1.dsc
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 2620 100 2620 0 0 1349 0 0:00:01 0:00:01 --:--:-- 36388
dget: retrieving http://httpredir.debian.org/debian/pool/main/c/calibre/calibre_2.55.0+dfsg.orig.tar.xz
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 37.3M 100 37.3M 0 0 2931k 0 0:00:13 0:00:13 --:--:-- 3032k
dget: retrieving http://httpredir.debian.org/debian/pool/main/c/calibre/calibre_2.55.0+dfsg-1.debian.tar.xz
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 22800 100 22800 0 0 59423 0 --:--:-- --:--:-- --:--:-- 59423
calibre_2.55.0+dfsg-1.dsc:
Good signature found
validating calibre_2.55.0+dfsg.orig.tar.xz
validating calibre_2.55.0+dfsg-1.debian.tar.xz
All files validated successfully.
dpkg-source: info: extraction de calibre dans calibre-2.55.0+dfsg
dpkg-source: info: extraction de calibre_2.55.0+dfsg.orig.tar.xz
dpkg-source: info: extraction de calibre_2.55.0+dfsg-1.debian.tar.xz
dpkg-source: info: mise en place de no_updates_dialog.patch
dpkg-source: info: mise en place de disable_plugins.py
dpkg-source: info: mise en place de use-system-feedparser.patch
dpkg-source: info: mise en place de python_multiarch_inc.patch
dpkg-source: info: mise en place de dont_build_unrar_plugin.patch
dpkg-source: info: mise en place de mips_no_build_threads.patch
dpkg-source: info: mise en place de links-privacy.patch
A lot of stuff has happened here!
First, dget downloaded the .dsc
file, which includes references to
the .orig.tar.xz
file and the debian.tar.xz
files. The
.orig.tar.xz
is the upstream source code,3 and the
.debian.tar.xz
file is basically the content of the debian/
subdirectory, the metadata used to build the debian package, including
all the patches specific to Debian.
Then dget downloads the files .orig.tar.xz
and .debian.tar.xz
files.
Then the OpenPGP signature on the .dsc
file is verified against the
web of trust,4 using dscverify. The .dsc
files includes
checksums for the downloaded files, and those checksums are verified
as well.
Then the files are extracted using dpkg-source -x
.
Notice how dget
is just a shortcut to commands you could all have
ran by hand.
If the version control system the package uses is familiar to you, you can use debcheckout to checkout the source directly. However, keep in mind that it does not ensure end-to-end cryptographic integrity like the previous procedure does, and instead relies on HTTPS-level transport security.
It might be useful if you prefer to collaborate with GitLab merge requests over at salsa.debian.org, but be warned that not all maintainers watch their GitLab projects as closely as the bug tracking system.
Modifying the package
At this point, we have a shiny source tree available in the
calibre-2.55.0+dfsg/
directory:
cd calibre-2.55.0+dfsg/
We can start looking around and make some changes.
Changing version
The first thing we want to make sure we do is to bump the version
number so that we don't mistakenly build a new package with the same
version number but with undocumented changes. The Debian package
version is stored in debian/changelog
.
In the case of calibre, I only needed to change that file to complete
the backport. This is called a "trivial backport": we just need to
recompile against the stable
environment. In other cases,
dependencies need to be modified, patches need to be included and so
on. But let's stick with the version number for now.
The generate the new version, I use dch --bpo
, which chooses the
right version number for me and pops open my $EDITOR
so I can add
things to the changelog:
dch --bpo
In this case, this created an entry for the 2.55.0+dfsg-1~bpo8+1
version5. I also added a Closes
entry to indicate that the
upload will fix a bug report I opened about creating that very
backport. It's important to describe exactly what you are doing in the
changelog and what bugs are being fixed. Here's the result:
calibre (2.55.0+dfsg-1~bpo8+1) jessie-backports; urgency=medium
* Rebuild for jessie-backports (Closes: #818309)
-- Antoine Beaupré <anarcat@debian.org> Tue, 26 Apr 2016 16:49:56 -0400
Note that there are other options you can pass to dch
. I often use:
--bpo
for backports--security, -s
for security uploads--nmu, -n
for non-maintainer uploads--increment, -i
for my own packages
There are more described in the dch manpage. The managing packages section of the developer's reference is useful in crafting packages specific to your situation.
Changing package metadata
If I needed to modify dependencies, I would have edited
debian/control
directly. Other modifications to the Debian package
would also happen in the debian/
directory. The function of the
various files in that directory vary a lot according to how the
package is built, but a good starting point is
Debian policy §4: Source packages.
Modifying the source code
If I needed to modify the source tree outside debian/
, I can do
the modifications directly, then use dpkg-source --commit
to
generate a patch that will end up added to the quilt patchset in
debian/patches
. New patches should follow the
patch tagging guidelines and dpkg-source --commit
will use that
template when creating a new patch.
Applying patches
If I already have a patch I want to apply to the source tree, then quilt is even more important. The first step is to import the patch:
quilt import ~/patches/CVE-2015-8477.patch
The above simply adds the CVE-2015-8477.patch
to the patch set, but
does not apply. You can apply all patches with:
quilt push -a
Quilt may tell you that the patch fails to apply. You can try to force-apply it with a "fuzz" argument:
quilt push --fuzz 100
Note that if the patch fails to apply, quilt leaves the source tree intact, so you can try that again and again. If the above fails but you are confident you can manually fix the patch, you can force-apply the patch:
quilt push --force
... then apply the modifications by hand. Then you need to refresh the patch to make sure it is updated correctly:
quilt refresh
Review the patch in debian/patches
to make sure it still looks sane.
Notice how you could also have bypassed quilt
completely and applied
the patch with the patch
command directly, then use dpkg-source
--commit
to generate a completely new patch.
Also note that the above does not care where the patch comes from. I
often extract the patch from a Git source tree fetched with
debcheckout
on the side, with, for example:
( cd ../source-git ; git show $hash ) > debian/patches/CVE-2015-8477.patch
quilt import debian/patches/CVE-2015-8477.patch
quilt push
Again, it's useful to add metadata to the patch and follow the patch tagging guidelines.
Building the package
Now that we are satisfied with our modified package, we need to build
it. The generic command to build a Debian package is
dpkg-buildpackage6. So at the root of the source tree
(above the debian/
directory), simply run:
dpkg-buildpackage
dpkg-buildpackage will the .deb
file. It also creates
new .dsc
, .debian.tar.gz
and .changes
files.7 Those
files should all show up in the parent directory.
gbp
in short) for that purpose, but other also use the simpler
git-pkg. I find that gbp
has more error checking and a better
workflow, but there are many opinions on how to do this.8
There are also many other ways of packaging Debian packages in Git,
including dgit and git-dpm, so until Debian standardizes on
one of those, this guide will remain git-agnostic.
In any case, there's a catch here: you need all the build-dependencies for the above builds to succeed. You may not have all of those, so you can try to install them with:
sudo mk-build-deps -i -r calibre
mk-build-deps
makes a dummy package to wrap them all up together, so
they are easy to uninstall. But this still installs a lot of cruft on
your system that you don't want in the first place.
Furthermore, the above procedure doesn't build the package in a "clean
environment". For example, say I am building a package for a regular
upload into unstable
("sid"). Yet, my workstation is running
stable
("jessie", currently). I can't simply build the package in
jessie
and expect it to work in sid
, I need to build it into a sid
environment.
For this, we need more powerful tools.
Building in a clean environment
I am using sbuild to build packages in a dedicated clean build environment. This means I can build packages for arbitrary distributions and also make sure there aren't exotic local configurations that can contaminate the build.
sbuild
takes your source package (the .dsc
file), and builds it in
a clean, temporary chroot
. To create that .dsc
file, you can use
dpkg-buildpackage -S
simply call sbuild
in the source directory
which will create it for you.
This manual covers two different ways of using sbuild
:
with a
schroot
backend, which is basically just a chroot wrapperwith a
autopkgtest
andqemu
backend, which enforces much stronger isolation at a (performance) cost
schroot instructions
To use sbuild
with schroot
, you first need to configure a schroot
:
sudo sbuild-createchroot --include=eatmydata,gnupg unstable /srv/chroot/unstable-amd64-sbuild http://deb.debian.org/debian
This assumes that:
you are running Stretch or later (see the sbuild wiki docs for workarounds in Jessie)
sbuild is already installed and you are in the right group, do this otherwise:
sudo apt-get install sbuild sudo sbuild-adduser $LOGNAME
you want to create an "unstable" image in amd64. to change the architecture, use the
--arch
argument, and to change the suite, change it in two the places that sayunstable
, obviouslyyou won't need to use hardlinks. overlay filesystems do not support hardlinks and you may need to switch to a tarball image if you need that feature (e.g. the mercurial test suite relies on this). to create a tarball image, use this:
sudo sbuild-createchroot --make-sbuild-tarball=/srv/chroot/unstable-amd64-sbuild.tar.gz unstable --chroot-prefix unstable-tar `mktemp -d` http://deb.debian.org/debian
You can also use qemu
instead, see below
The above will create chroots for the unstable suite with an amd64
architecture, using debootstrap. You may of course modify this to
taste based on your requirements and available disk space. My build
directories count for around 7GB (including ~3GB of cached .deb
packages) and each chroot is between 500MB and 700MB.
A few handy sbuild-related commands:
sbuild -c bookworm-amd64-sbuild
- build in thebookworm
chroot even though another suite is specified (e.g.UNRElEASED
,bookworm-backports
orbookworm-security
)sbuild --build-dep-resolver=aptitude
- use another solver for dependencies, required for backports, for example. see the manpage for details of those solvers.schroot -c bookworm-amd64-sbuild
- enter thebookworm
chroot to make tests, changes will be discardedsbuild-shell bookworm
- enter thebookworm
chroot to make permanent changes, which will not be discardedsbuild-destroychroot
- supposedly destroys schroots created by sbuild for later rebuilding, but I have found that command to be quite unreliable. besides, all it does is:rm -rf /srv/chroot/unstable-amd64-sbuild /etc/schroot/chroot.d/unstable-amd64-sbuild-*
Also note that it is useful to add aliases to your schroot
configuration files. This allows you, for example, to automatically
build bookworm-security
or bookworm-backports
packages in the bookworm
schroot. Just add this line to the relevant config in
/etc/schroot/chroot.d/
:
aliases=bookworm-security-amd64-sbuild,bookworm-backports-amd64-build
Qemu configuration
To use qemu as a backend, use this instead:
sudo mkdir -p /srv/sbuild/qemu/
sudo apt install sbuild-qemu
sudo sbuild-qemu-create -a amd64 -o /srv/sbuild/qemu/unstable-amd64.img unstable https://deb.debian.org/debian
This will create a single virtual machine image from the Debian
unstable suite with an amd64
architecture. Thanks to qemu
, you can
use this to build or test packages in completely different
architectures.
Then to make this used by default, add this to ~/.sbuildrc
:
# run autopkgtest inside the schroot
$run_autopkgtest = 1;
# tell sbuild to use autopkgtest as a chroot
$chroot_mode = 'autopkgtest';
# tell autopkgtest to use qemu
$autopkgtest_virt_server = 'qemu';
# tell autopkgtest-virt-qemu the path to the image
# use --debug there to show what autopkgtest is doing
$autopkgtest_virt_server_options = [ '--', '/srv/sbuild/qemu/%r-%a.img' ];
# tell plain autopkgtest to use qemu, and the right image
$autopkgtest_opts = [ '--', 'qemu', '/srv/sbuild/qemu/%r-%a.img' ];
# no need to cleanup the chroot after build, we run in a completely clean VM
$purge_build_deps = 'never';
# no need for sudo
$autopkgtest_root_args = '';
Note that the above will use the default autopkgtest (1GB, one core) and qemu (128MB, one core) configuration, which might be a little low on resources. You probably want to be explicit about this, with something like this:
# extra parameters to pass to qemu
# --enable-kvm is not necessary, detected on the fly by autopkgtest
my @_qemu_options = ['--ram-size=4096', '--cpus=2'];
# tell autopkgtest-virt-qemu the path to the image
# use --debug there to show what autopkgtest is doing
$autopkgtest_virt_server_options = [ @_qemu_options, '--', '/srv/sbuild/qemu/%r-%a.img' ];
$autopkgtest_opts = [ '--', 'qemu', @_qemu_options, '/srv/sbuild/qemu/%r-%a.img'];
Also see a more in-depth discussion about this configuration in this blog post.
A few handy qemu
related commands:
enter the VM to make test, changes will be discarded:
sbuild-qemu-boot /srv/sbuild/qemu/unstable-amd64.img
That program is shipped only with bookworm and later, an equivalent command is:
qemu-system-x86_64 -snapshot -enable-kvm -object rng-random,filename=/dev/urandom,id=rng0 -device virtio-rng-pci,rng=rng0,id=rng-device0 -m 2048 -nographic /srv/sbuild/qemu/unstable-amd64.img
The key argument here is
-snapshot
.enter the VM to make permanent changes, which will not be discarded:
sudo sbuild-qemu-boot --read-write /srv/sbuild/qemu/unstable-amd64.img
Equivalent command:
sudo qemu-system-x86_64 -enable-kvm -object rng-random,filename=/dev/urandom,id=rng0 -device virtio-rng-pci,rng=rng0,id=rng-device0 -m 2048 -nographic /srv/sbuild/qemu/unstable-amd64.img
That's the same
qemu
command as above, but without the-snapshot
.update the VM:
sudo sbuild-qemu-update /srv/sbuild/qemu/unstable-amd64.img
build in a specific VM regardless of the suite specified in the changelog (e.g.
UNRELEASED
,bookworm-backports
,bookworm-security
, etc):sbuild --autopkgtest-virt-server-opts="-- qemu /var/lib/sbuild/qemu/bookworm-amd64.img"
Note that you'd also need to pass
--autopkgtest-opts
if you wantautopkgtest
to run in the correct VM as well:sbuild --autopkgtest-opts="-- qemu /var/lib/sbuild/qemu/unstable.img" --autopkgtest-virt-server-opts="-- qemu /var/lib/sbuild/qemu/bookworm-amd64.img"
You might also need parameters like
--ram-size
if you customized it above.
Building with sbuild
Now that a sbuild backend is setup, I can build packages in one of three ways.
If I have a
.dsc
already (again, that can be generated withdpkg-buildpackage -S
in the source tree):sbuild calibre_2.55.0+dfsg-1~bpo8+1.dsc
If I'm in the source tree:
sbuild
With git-buildpackage:
git-buildpackage --git-builder=sbuild
The above can be configured by default in
~/.gbp.conf
:[DEFAULT] builder=sbuild # to force lintian to run since we don't use debuild anymore postbuild = lintian $GBP_CHANGES_FILE
sbuild
will generate your binary package in the parent directory
(..
). It will build your package with the suite specified in the
package's latest changelog
entry.
To pass options to the underlying dpkg-buildpackage
(for example,
you often need -sa
to provide the source tarball with the upload),
you should use --debbuildopts -sa
in sbuild
. For git-buildpackage,
simply add -sa
to the commandline.
Build servers
Sometimes, your machine is too slow to build this stuff yourself. If you have a more powerful machine lying around, you can send a source package to the machine and build the package there.
You first need to create a source package (the .dsc
file created
with dpkg-buildpackage -S
) and transfer the files over. An easy way to do
the latter is with the dcmd command. For example, this will
create a source package, transfer it to the remote host
example.net
and build it:
foo-1.0$ dpkg-buildpackage -S
foo-1.0$ dcmd scp ../foo_1.0.dsc example.net:dist/
foo-1.0$ ssh example.net sbuild dist/foo_1.0.dsc
The above might cause trouble if you are working within a git
repository, as dpkg-source
might bundle .git
files that should be
ignored.
To build from git, you first use gitpkg
to generate a .dsc
file
from the git tree, where rev
is the current commit of the debian
package (can be master
or a specific tag) and upstream
is a
revision pointing at the upstream release to generate the .orig
tarball if not already present:
gitpkg rev upstream
Then you can use the resulting .dsc
file as normal.
If you use cowbuilder, you can also use cowpoke instead of the
above. Since I started using sbuild
, I do not have a cowbuilder
setup anymore.
By default, cowpoke
logs into a remote server and uses sudo
to
call cowbuilder
to build a chroot. For example, this will build
calibre on the remote host buildd.example.com
, in a jessie/amd64
chroot:
cowpoke --buildd buildd.example.com --dist sid --arch amd64 calibre_2.55.0+dfsg-1~bpo8+1.dsc
This assume the chroot already exists of course. You can create it by
using the --create
argument.
If you do not have your own host to build packages, you can upload
source packages to another buildd using dput
, for example through
debomatic. You need to request access first, however. More
documentation is available on the Deb-o-Matic site.
In my experiments, Debomatic was way faster than compiling on my laptop, so I sometimes use it for larger packages. For example, building Xen on Debomatic takes around 6 minutes while on my laptop it takes almost triple that time (17 minutes). Plus Debomatic runs lintian and piuparts.
The obvious downside is that I need to trust the remote server to generate the same package as I would do locally. Even if the package is reproducible (which is not always the case!), I would still have to build the package locally to be able to compare the results...
Testing packages
Some packages have a built-in test suite which you should make sure runs properly during the build. Sometimes, backporting that test suite explicitly can be useful to ensure that everything works properly after a backport or LTS security upload. A key component of this is DEP8, currently implemented as autopkgtest. piuparts can also be used to see if the package cleans up properly after itself.
With autopkgtest
When a package self-testing enabled, it will be ran by Debian CI at various times. While there can be build-time tests, CI runs more often (when dependencies are updated, for example), and on the installed package, which is a different environment. To reproduce test failures or make sure changes won't break the test suite, autopkgtest can be ran locally. First, install it:
sudo apt install autopkgtest
Then you can run the tests against the built package:
autopkgtest libotr_4.1.1-3_amd64.changes -- schroot unstable-amd64-sbuild
That's it! Tests can also be ran against the current directory, in which case autopkgtest will build the package for you first, also in the virtual environment.
Note that if you followed the sbuild-qemu
configuration above,
this is already done by default at the end of your sbuild
run, using
qemu
.
Testing by hand
In some cases, however, those tests are unavailable or insufficient and you need to actually install and run the package somewhere.
Backports can obviously be tested directly on your local machine if you are running stable (which is likely if you are building backports, unless you are doing them for someone else of course).
Otherwise, it is likely that you will need to build a separate environment to test the package if it is built for another distribution. For this, you can use the debootstrap and chroot commands. But it is probably better to run tests within a completely isolated environment, often called a "Virtual Machine".
Vagrant virtual machines
Hashicorp's Vagrant is a useful shortcut you can use to build virtual machines consistently. You can get started by using the following commands to get SSH into a Bookworm machine, for example:
sudo apt install vagrant
mkdir bookworm64; cd bookworm64
vagrant init debian/bookworm64
vagrant up
vagrant ssh
And that's it. vagrant init
creates a Vagrantfile that basically
describes how vagrant up
can recreate the VM. You can
use shell commands, Ansible, Puppet
or other provisionning tools to automatically configure the VM, of
course. Use vagrant halt
to stop the VM and vagrant destroy
to
actually remove all data associated with the VM.
Note that the Vagrantfile
directory is,
by default, shared with the virtual machine, in the /vagrant
directory. This allows you to share files between the host and the VM,
and allow for files to "stick" when destroying the VM. This is
synchronized using vagrant rsync, see
the official Debian baseboxes documentation for more
information. There are also ways to setup realtime filesharing between
the host and the VM using NFS or 9p filesystems, but those are more
complicated and considered out of scope of this tutorial.
If the official Debian images are not up to your taste, you can build your own or choose another one from the Hashicorp Atlas.
Also note that Vagrant can use a libvirt backend, with the following
Vagrantfile
snippet:
config.vm.provider :libvirt do |libvirt|
libvirt.graphics_type = "spice"
libvirt.video_type = "qxl"
libvirt.channel :type => 'spicevmc', :target_name => 'com.redhat.spice.0', :target_type => 'virtio'
# add two devices using spicevmc channel
(1..2).each do
libvirt.redirdev :type => "spicevmc"
end
libvirt.memory = "2096"
end
Qemu
Another simple approach is to use plain Qemu. We will need to use a special tool to create the virtual machine as debootstrap only creates a chroot, which virtual machines do not necessarily understand.
With sbuild-qemu
, above, you already have a qemu image, built with
vmdb2.
The qemu section above also has good tips on how to start those images as well.
There is another tool that accomplished similar things
called grml-debootstrap. I do not use it because it doesn't create
a minimal image by default. vmdebootstrap
is also destined to be the
main tool used to create Debian Live official images which makes
it interesting in the long term. I have used the following commandline
when using grml-deboostrap:
sudo grml-deboostratp --vmfile --bootappend console=ttyS0 --arch $ARCH --release $DIST --target $DIST.qcow2
Also note that the Debian Cloud team is considering using FAI for this in the future, see this post for details and other ideas. This is all quite in flux still.
To boot those images with Qemu, use:
qemu-system-x86_64 -snapshot -enable-kvm -display none -serial mon:stdio $DIST-$ARCH.qcow2
-snapshot
makes the image read-only, so it can be readily reused
without worring about contaminating the environment. KVM is
obviously optional here, but usually works in my tests and is much
faster than non-HVM usage. The remaining options are to make sure
I get a regular terminal from Qemu instead of a graphical window. This
requires serial console to be configured in the image, otherwise you will get no
output at all. Also, if you are testing GUIs, you will obviously want
to remove those options and install a bunch of packages on top of the
minimal install.
To transfer data between the host and the virtual machines, the simplest way I could find is with netcat. On the host:
nc -q 0 -l -p 10080 < /var/cache/pbuilder/wheezy-amd64/result/phpmyadmin_3.4.11.1-2+deb7u4_all.deb
In the VM:
nc 10.0.2.2 10080 > phpmyadmin_3.4.11.1-2+deb7u4_all.deb
The IP address may change, use, ip route
to find the address of the
host, which should be the gateway. 10080
is an arbitrary port above
1024
.
Ports can also be forwarded from the host to the VM using the -net
command. For example, -net user,hostfwd=tcp::10022-:22 -net nic
would allow the host to connect to the VM's SSH server. I ended up
setting up the following shell alias:
# qemu: specify architecture, enable serial port and common port
# forwards (HTTP and SSH), enable KVM support and don't write the
# image by default (can be worked around with C-a s at
# runtime). graphical display still enabled for POLA (some VMs don't
# have serial), can be turned off with -display none.
alias qemu="qemu-system-x86_64 -serial mon:stdio -net user,hostfwd=tcp::10080-:80 -net user,hostfwd=tcp::10022-:22 -net nic -enable-kvm -snapshot"
.deb
files can be installed with dpkg -i
, which will likely fail
because of missing dependency, so you need to also run apt-get
install -f
.
unshare
Another option is to use the "unshare" command, which launches another command in a different namespace:
sudo unshare --ipc --mount --net --pid --uts --user --cgroup --time --fork chroot /path/mountpoint qemu-arm-static /bin/bash
The above launches the bash
binary using the a ARM emulation wrapper
(qemu-arm-static
) inside a chroot at /path/mountpoint
in a
different namespace.
Yes, there should really be an --all
option here. A previous version
of this document omitted the --user --cgroup --time
flags, for
example.
This can reuse previously created chroots, but the filesystem
separation works only if /path/mountpoint
is really a different
mountpoint. Otherwise changes in the filesystem affect the parent
host, in which case you can just copy over the chroot.
Uploading packages
Uploading packages can be done on your own personal archive if you
have a webserver, using the following ~/.dput.cf
configuration:
[people]
fqdn = people.debian.org
method = scp
incoming = /home/anarcat/public_html/debian/wheezy-lts/
run_dinstall = 0
progress_indicator = 2
The above archive is not usable in a sources.list
file, because that
would involve an incoming queue and all sorts of complicated
things. Instead, people are expected to use dget
to download the
.changes
file and check the signatures. See the HEADER.html on my
personal archive for an example.
You can also upload to the official Debian archives and backports, naturally. If you do not have access to those, you can also upload to mentors and request sponsorship.
Failing all that, you can upload packages to a Launchpad PPA or host your own Debian repository using reprepro (Koumbit has some good documentation) or aptly (haven't tested it myself).
Further work and remaining issues
This guide should be integrated into the official documentation or the Debian wiki. It is eerily similar to this guide which itself is a duplicate of this other guide.
A blog post goes into depth about the
alternatives to qemu
and sbuild
.
- Well, it's not exactly true: I know bzr well enough, but I'm lazy. Besides, the point is to have a procedure that works regardless of the version control used. Furthermore, some packages do not use version control at all!↩
-
we use the tracker instead of
apt-get source
ordget package=version
because those rely on asources.list
with all possible distributions enabled, something that is really inconvenient and harder to configure and document.↩ -
Well, not exactly: in this case, it's a modification of the
upstream source code, prepared specifically to remove non-free
software, hence the
+dfsg
suffix, which is an acronym for Debian Free Software Guidelines. The+dfsg
is simply a naming convention used to designated such modified tarballs.↩ -
In my case, this works cleanly, but that is only because the key is known on my system.
dget
actually offloads that work todscverify
which looks into the official keyrings in the debian-keyring package. This package may be missing some keys because the maintainers are new and were not in the keyring when the Debian version you are using was released. In this case, you need to find the key yourself, add it to your keyring, and then adding the following to~/.devscripts
will leverage your personal keys into the web of trust:DSCVERIFY_KEYRINGS=~/.gnupg/pubring.gpg
Note that this is NOT an environment variable, it needs to be put in the file... An alternative is to inject keys into the
~/.gnupg/trustedkeys.gpg
files which is checked bydscverify
by default.You can also use
↩dscverify --keyring key.gpg *.dsc
to check the signature by hand against a given key file. -
The "tilde" character to indicate that this is a lower
version than the version I am backporting from, so that when the
users eventually upgrade to the next stable (
stretch
, in this case), they will actually upgrade to the real version in stretch, and not keep the backport lying around. There is a detailed algorithm description of how version number are compared, which you can test usingdpkg --compare-versions
if you are unsure.↩ -
You can also use
debuild
instead ofdpkg-buildpackage
because it also runs lintian and signs the binary package with debsign.↩ -
the
.changes
file is similar to the.dsc
file, but also covers the.deb
file. So it's kind of the.dsc
file for the binary package (except there's also a.changes
file for source-only uploads, so not really).↩ - git-pkg actually only extracts a source package from your git tree, and nothing else. There are hooks to trigger builds and so on, but it's basically expected that you do that yourself, and gitpkg is just there to clean things up for your.↩