multistrap/pod/multistrap

=pod

=head1 Name

multistrap - debootstrap replacement for multiple repository support

=head1 Synopsis

 multistrap [-a ARCH] [-d DIR] -f CONFIG_FILE
 multistrap [--simulate] -f CONFIG_FILE
 multistrap -?|-h|--help|--version

=head1 Options

--dry-run - collate all the configuration settings and output a
bare summary.

--simulate - same as --dry-run

(The following options can also be set in the configuration file.)

--tidy-up - remove apt cache data, downloaded Packages files and
the apt package cache. Same as cleanup=true.

--no-auth - allow the use of unauthenticated repositories. Same
as noauth=true

--sourcedir DIR - move the contents of var/cache/apt/archives/ from
inside the chroot to the specified external directory, then add the
Debian source packages for each used binary. Same as retainsources=DIR
If the specified directory does not exist, nothing is done.

=head1 Description

multistrap provides a debootstrap-like method based on apt and
extended to provide support for multiple repositories, using a
configuration file to specify the relevant suites, architecture,
extra packages and the mirror to use for each bootstrap.

The aim is to create a complete bootstrap / root filesystem with
all packages installed and configured, instead of just the base
system.

Example configuration:

 [General]
 arch=armel
 directory=/opt/multistrap/
 # same as --tidy-up option if set to true
 cleanup=true
 # same as --no-auth option if set to true
 # keyring packages listed in each bootstrap will
 # still be installed.
 noauth=false
 # extract all downloaded archives (default is true)
 unpack=true
 # whether to add the /suite to be explicit about where apt
 # needs to look for packages. Default is false.
 explicitsuite=false
 # aptsources is a list of sections to be used
 # the /etc/apt/sources.list.d/multistrap.sources.list
 # of the target. Order is not important
 aptsources=Debian
 # the bootstrap option determines which repository
 # is used to calculate the list of Priority: required packages
 # and which packages go into the rootfs.
 # The order of sections is not important.
 bootstrap=Debian
 
 [Debian]
 packages=
 source=http://ftp.uk.debian.org/debian
 keyring=debian-archive-keyring
 suite=lenny

This will result in a completely normal debootstrap of Debian lenny from
the specified mirror, for armel in '/opt/multistrap/'.

Specify a package to extend the multistrap to include that package and
all dependencies.

Specify more bootstraps by adding new sections. Section names are used
in the bootstrap general option.

Section names are case-insensitive.

All dependencies are resolved only by apt, using all bootstrap
repositories, to use only the most recent and most suitable
dependencies. Note that multistrap turns off Install-Recommends
so if the multistrap needs a package that is only a Recommended
dependency, the recommended package needs to be specified in the
packages line explicitly.

'Architecture' and 'directory' can be overridden on the command line.
Some other general options also have command line options.

=head1 Repositories

C<aptsources> lists the sections which should be used to create the
F</etc/apt/sources.list.d/multistrap.list> apt sources in the final
system. Not all C<aptsources> have to appear in the C<bootstrap>
section if you have some internal or local sources which are not
accessible to the installed root filesystem.

C<bootstrap> lists the sections which will be used to create the
multistrap itself. Only packages listed in C<bootstrap> will be
downloaded and unpacked by multistrap.

Make sure C<bootstrap> lists all sections you need for apt to be
able to find all the packages to be unpacked for the multistrap.

(Older versions of multistrap supported the same option under the
C<debootstrap> name - this spelling is still supported but new
configuration files should be C<bootstrap> instead.

=head1 General settings:

'arch' can be overridden on the command line using the C<--arch> option.

'directory' specifies the top level directory where the bootstrap
will be created - it is not packed into a .tgz once complete.

'bootstrap' lists the Sections which will be used to specify the packages
which will be downloaded (and optionally unpacked) into the bootstrap.

'aptsources' lists the Sections which will be used to specify the apt
sources in the final system, e.g. if you need to use a local repository
to generate the rootfs which will not be available to the device at
runtime, list that section in C<bootstrap> but not in C<aptsources>.

If you want a package to be in the rootfs, it B<must> be specified in
the C<bootstrap> list under General.

The order of section names in either list is not important.

As with debootstrap, multistrap will continue after errors, as long
as the configuration file can be correctly parsed.

multistrap also implements the machine:variant support originally
used in Emdebian Crush, although in a different implementation. Using
the cascading configuration support, particular machine:variant
combinations can be supported by simple changes on the command line.

Setting C<tarballname> to true also packs up the final filesystem into
a tarball.

Note that multistrap ignores any unrecognised options in the config
file - this allows for backwards-compatible behaviour as well as
overloading the multistrap config files to support other tools
(like pbuilder).

=head1 Section settings

 [Debian]
 packages=
 source=http://ftp.uk.debian.org/debian
 keyring=debian-archive-keyring
 suite=lenny

The section name (in [] brackets) needs to be unique for this configuration
file and any configuration files which this file includes. Section names
are case insensitive (all comparisons happen after conversion to lower
case).

'packages' is the list of packages to be added when this Section
is listed in C<bootstrap>.

'source' is the apt source to use for this Section. (To use a local
source on the same machine, ensure you use C<copy://> not C<file://>,
so that apt is told to copy the packages into the rootfs instead of
assuming it can try to download them later - because that "later" will
never actually happen.

'keyring' lists the package which contains the key used by the source
listed in this Section. If no keyring is specified, the C<noauth>
option must be set to B<true>. See Secure Apt.

'suite' is the suite to use from this source. Note that this B<must>
be the suite, not the codename.

Suites change from time to time: (oldstable, stable, testing, sid)
The codename (etch, lenny, squeeze, sid) does not change.


=head1 Secure Apt

To use authenticated apt repositories, multistrap either needs to be
able to install an appropriate keyring package from the existing apt
sources *outside the multistrap environment* or have the relevant keys
already configured using apt-key *on the host system*.

If relevant packages exist, specify them in the 'keyring' option for
each repository. multistrap will then check that apt has already
installed this package so that the repository can be authenticated
before any packages are downloaded from it.

Note that *all* repositories to be used with multistrap must be
authenticated or apt will fail. Similarly, secure apt can only be
disabled for all repositories (by using the --no-auth command line
option or setting the general noauth option in the configuration
file), even if only one repository does not have a suitable keyring
available. Not all packages need keyring packages, if you configure
apt-key appropriately.

The keyring package(s) will also be installed inside the multistrap
environment to match the installed apt sources for the multistrap.

All configuration of apt-key needs to be done for the machine
running multistrap itself.

=head1 State

multistrap is stateless - if the directory exists, it will simply
proceed as normal and apt will try to pick up where it left off.

=head1 Configuration

multistrap unpacks the downloaded packages but other stages of
system configuration are not attempted. Examples include:

 /etc/inittab
 /etc/fstab
 /etc/hosts
 /etc/securetty
 /etc/modules
 /etc/hostname
 /etc/network/interfaces
 /etc/init.d
 /etc/dhcp3 

Any device-specific device nodes will also need to be created
using MAKEDEV.

As an alternative, multistrap includes a device-table.pl helper
script that can work around some of the issues with MAKEDEV.
device-table.pl requires a device table file along the lines of
the one in the mtd-utils source package.

Once multistrap has successfully created the basic file and
directory layout, other device-specific scripts are needed before
the filesystem can be packaged up and installed onto the
target device.

Once installed, the packages themselves need to be configured
using the package maintainer scripts and C<dpkg --configure -a>,
unless this is a native multistrap.

For C<dpkg> to work, F</proc> and F</sysfs> must be mounted (or
mountable), F</dev/pts> is also recommended.

See also: http://wiki.debian.org/Multistrap

=head1 Environment

To configure the unpacked packages (whether in native or cross mode),
certain environment variables are needed:

Debconf needs to be told to accept that user interaction is not
desired:

 DEBIAN_FRONTEND=noninteractive DEBCONF_NONINTERACTIVE_SEEN=true

Perl needs to be told to accept that no locales are available inside
the chroot and not to complain:

 LC_ALL=C LANGUAGE=C LANG=C

Then, dpkg can configure the packages:

chroot method (PATH = top directory of chroot):

 DEBIAN_FRONTEND=noninteractive DEBCONF_NONINTERACTIVE_SEEN=true \
 LC_ALL=C LANGUAGE=C LANG=C chroot /PATH/ dpkg --configure -a

at a login shell:

 # export DEBIAN_FRONTEND=noninteractive DEBCONF_NONINTERACTIVE_SEEN=true
 # export LC_ALL=C LANGUAGE=C LANG=C 
 # dpkg --configure -a

(As above, dpkg needs F</proc> and F</sysfs> mounted first.)

=head1 Native mode - multistrap

multistrap was not intended for native support, it was developed for
cross architecture support. In order for multiple repositories to be
used, multistrap only unpacks the packages selected by apt.

In native mode, various post-multistrap operations are likely to be
needed that debootstrap would do for you:

 1. copy /etc/hosts into the chroot
 2. clean the environment to unset LANGUAGE, LC_ALL and LANG
    to silence nuisance perl warnings that obscure other errors

(An alternative to unset the localisation variables is to add
locales to your multistrap configuration file in the 'packages'
option.

A native multistrap can be used directly with chroot, so
C<multistrap> runs C<dpkg --configure -a> at the end of the
multistrap process. 

=head1 Cascading configuration

To support multiple variants of a basic (common) configuration,
C<multistrap> allows configuration files to include other (more general)
configuration files. i.e. the most detailed / specific configuration
file is specified on the command line and that file includes another
file which is shared by other configurations.

Base file:

 /usr/share/multistrap/crosschroot.conf

Variations:

 /usr/share/multistrap/armel.conf

Specifying just the armel.conf file will get the rest of the settings
from crosschroot.conf so that common changes only need to be made in a
single file.

It is B<strongly> recommended that any changes to the configuration files
involved in any particular cascade are tested using the C<--simulate>
option to multistrap which will output a summary of the options that
have been set once the cascade is complete. Note that multistrap does
B<not warn you> if a configuration file contains an unrecognised
option (for future compatibility with backported configurations), so a
simple typo can result in an option not being set.

=head1 Machine:variant support

The old packages.conf variables from emsandbox can all be converted
into C<multistrap> configuration variables. The machine:variant
support in C<multistrap> concentrates on the scripts, 
F<config.sh> and F<setup.sh>

Once C<multistrap> has unpacked the downloaded packages, the 
C<setup.sh> can be called, passing the location and architecture of
the root filesystem, so that other fine tuning can take place. At
this stage, any operations inside the rootfs must not try to execute
any binaries within the rootfs. As the final stage of the multistrap
process, C<config.sh> is copied into the root directory of the rootfs.

One advantage of using machine:variant support is that the entire
rootfilesystem can be managed by a single call to multistrap - this
is useful when building root filesystems in userspace.

To enable machine:variant support, specify the path to the scripts to
be called in the variant configuration file (General section):

 [General]
 include=/path/to/general.conf
 setupscript=/path/to/setup.sh
 configscript=/path/to/config.sh


=head1 Restricting package selection

C<multistrap> includes Required packages by default, the current list
of packages can be seen using:

 grep-available  -FPriority 'required' -sPackage

If the OmitRequired option is set to true, these packages will not be
added - whilst useful, this option can easily lead to a useless
rootfs. Only the packages specified manually in the configuration
files will be used in the calculations - dependencies of those packages
will be added but no others.

=head1 Recommends behaviour

The Debian default behaviour after the Lenny release was to consider
recommended packages as extra packages to be installed when any one
package is selected. Recommended packages are those which the maintainer
considers that would be present on C<most> installations of that package
and allowing Recommends means allowing Recommends of recommended packages
and so on.

The multistrap default is to turn recommends OFF.

Set the allowrecommends option to true in the General section
to use typical Debian behaviour.

=head1 Explicit suite specification

Sometimes, apt needs to be told to get a particular package from a
particular suite, ignoring a more recent version in another suite
in the same set of sources.

C<multistrap> can operate with and without the explicit suite option,
the default is to let apt use the most recent version from the collection
of specified F<bootstrap> sources.

Explicit suite specification has no effect on the final installed
system - if your aptsources includes a repository which in turn includes
a newer version of the package(s) specified explicitly, the next
C<apt-get upgrade> on the device will bring in the newer version.

Also, when specifying packages to get from a specific suite, apt will
also try and ensure that the dependencies for that package are also
from the same suite and this can cause apt to be unable to resolve the
complete set of dependencies. In this situation, being explicit about
one package selection may require being explicit about some (not
necessarily all) of the dependencies of that package as well.

When using this support in Lenny, ensure that each section uses the
suite (oldstable, stable, testing, sid) and B<not> the codename
(etch, lenny, squeeze, sid) in the C<suite> configuration item
as the version of apt in Lenny and previous cannot use the codename.

To test, on Lenny, try:

 $ sudo apt-get install apt/stable

Compare with

 $ sudo apt-get install apt/lenny

=cut

=head1 Omitting deb-src listings

Some multistrap environments do not need access to the Debian sources
of packages being installed, typically this is required when preparing
a build (or cross-build) chroot using multistrap.

To turn off this additional source (and save both download time and
apt-cache size), use the omitdebsrc field in each Section.

 [Baked]
 packages=
 source=http://www.emdebian.org/baked
 keyring=emdebian-archive-keyring
 suite=testing
 omitdebsrc=true

=cut