There are a few reasons the Ubuntu live server ISOs as shipped may not quite be what you want. Here are some examples:
-
You want to make an ISO that does a completely automated install
-
You want packages that are not in the package repository by default to be available during install, even when there is no network (or only a very isolated network).
-
You want to add an argument to the default kernel command line.
-
You want to inject a new version of the subiquity snap for testing.
This script aims to help you making modified versions of the distributed ISOs with changes such as those above.
Some of the things it does probably work on desktop installer ISOs too but I haven't thought very hard about that side of things yet.
Although this program was written by a Canonical employee on company time, it started out as essentially an overgrown shell script used as part of installer developement. It is not a Canonical product and is not supported as such. The "this program is distributed in the hope that it will be useful" part of the GPL stanza is as true as ever but the "without even the implied warranty of FITNESS FOR A PARTICULAR PURPOSE" is perhaps even more true than usual.
This script is pretty Linux-dependent and requires xorriso
and
mksquashfs
to be available on $PATH
. Some actions require the
python3-debian
package to be installed and gpg
command to be
available. Operating on ISOs from some Ubuntu releases will also need
lz4cat (to unpack the initrd) which is found in liblz4-tool
.
It needs to be run as root (although possibly using FUSE variants for all the mounts would allow it to run as a regular user, maybe).
The basic idea behind this tool is that you tell it where to find the source ISO, where to put the modified ISO and a list of actions that make up the modifications. So an invocation always looks somewhat like this:
# livefs-edit $source.iso $dest.iso [actions]
Actions can be specified two ways: on the command line or in a YAML file. Each action has a name and many of them take arguments.
On the command line, actions and arguments are specified like:
--action-name arg1 arg2 --next-action-name
Arguments that are interpreted as boolean interpret 'on', 'yes', 'true' (case insensitively) as true.
Alternatively (if shell quoting starts to get painful), the actions
can be passed as a YAML file, using the --action-yaml
flag:
# livefs-edit $source.iso $dest.iso --action-yaml examples/example.yaml
The YAML file should be a list of mappings. Each mapping names the
actions with name
and lists any arguments by name, for example:
- name: add-cmdline-arg
arg: autoinstall
persist: false
- name: shell
- name: add-packages-to-pool
packages:
- casper
- valgrind
This script does all its work in a temporary directory. Within that
directory the original ISO is mounted at old/iso
and what will be
packed into the new ISO is present at new/iso
(the script uses a lot
of overlayfs
mounts to avoid copying large amounts of data around,
and also only repack things when there are changes).
Many actions require a writable emulation of the root filesystem that
the installer will run in. By default this is created at rootfs
in
the main temporary directory, but this can be customized.
In general, things from the original ISO are mounted in old/
and are
either read-only or should be treated as such. Writable versions for
the new ISO live in new/
(mostly).
argument: target
(default: "rootfs"
)
This action sets up a writable emulation of the root filesystem that
the installer will run in at the directory named by target
. Changes
to this rootfs will be present in the rootfs used by installer on the
modified ISO.
Many actions will do this implicitly but it may be clearer to be
explicit about the target directory name if later shell
or cp
actions refer to paths in the rootfs.
argument: command
(default: null
)
Runs a shell (bash) in the main temporary directory. If command
is
present, this is the command that is run. If not, an interactive shell
is run. If the shell (command or interactive) exits with a non-zero
code, that aborts the run.
argument: source
argument: dest
Copy a file. The paths source
and dest
are interpreted in a special way:
- Absolute paths (i.e. paths starting with '/') are handled as is.
- A magic prefix of '$LAYER[n]' is handled by mounting the n'th layer if needed and interpreting the remainer of the path relative to the base of that layer.
- Other paths are interpreted relative to the main temporary directory.
So something like this:
--cp /my/custom/initrd new/iso/casper/initrd
to replace the initrd. And something like this:
--cp /path/to/sources.list '$LAYER[0]/etc/apt/sources.list'
to overwrite sources.list in the base layer.
argument: path
Remove a file or directory. path
is interpreted the same way as the
paths passed to --cp
.
--rm new/iso/casper
to remove the directory casper.
argument: snap
argument: channel
(default: "stable"
)
Inject the passed snap into the rootfs the installer runs in. This is used to test new versions of subiquity. If there is an assert alongside the snap, this will be copied into the ISO too and the snap set up to track the passed channel, otherwise it is installed unasserted.
argument: snap_name
argument: channel
(default: "stable"
)
A wrapper around --inject-snap
that downloads the specified snap
from the store first.
argument: squash_name
argument: add_sys_mounts
(default: true
)
Mount the squashfs named squash_name
at new/{name}
and arrange for it be
repacked if there are any changes before the new ISO is made.
add_sys_mounts
controls whether the usual chroot setup stuff is done
(mounting /dev, /proc/ etc, setting up /etc/resolv.conf).
argument: arg
argument: persist
(default: true
)
Add an argument to the default kernel command line. If persist
is
true, it will be present on the default kernel command line of the
installed system as well.
argument: autoinstall_config
Add the provided autoinstall config to the ISO so it is used by default. This also adds "autoinstall" to the default kernel command line.
autoinstall_config
is the path to a YAML file which can contain
either the autoinstall config directly or a cloud-init user-data file
(in which case it can contain other configuration for the live
installer session, such as ssh keys to be used for the installer
user).
This will generate a new Ed25519 GPG key, sign the package repository
on the ISO with it, arrange for the public part to end up in
/etc/apt/trusted.gpg.d/custom-iso-key.gpg
in the installed system,
and then throw the private part away. You should be aware of this
change to the default apt configuration! Deleting this file in an
autoinstall late-command
would be a reasonable thing to do, unless
you want the option of using the ISO as an apt repository later on.
argument: list of deb files
Add the passed deb files to the repository on the CD so that they are available for installation while the installer is running, even if the install is done offline.
This calls resign-pool
so do read the note about GPG in the
description of that action.
argument: packages
(list of package names)
This is a wrapper around add-debs-to-pool
which takes package names
rather than deb files. It downloads the listed packages and any others
needed to satisfy their dependencies from the main Ubuntu archive and
passes them to add-debs-to-pool
. Do read the note about GPG in the
description of resign-pool
.
argument: target
(default: "new/initrd"
)
Unpack the initrd using unmkinitramfs into target
(contents will
likely end up in subdirectories called things like early
, early2
and main
, at least on amd64) and arrange for these to be repacked
into a replacement initrd for the modified ISO if any changes are made.
argument: repo
Run add-apt-repository <repo>
in the base layer.
argument: list of deb files
Install the listed deb files in the installer environment.
argument: packages
(list of deb files)
Install the listed packages in the base layer.
argument: <cmd>
(optional)
Execute <cmd>
in a namespace that contains the context object as
ctxt
or start an interactive shell if <cmd>
is not passed.
argument: <flavor>
Modify the ISO so the kernel installed via the metapackage
linux-<flavor>
will be used to boot the ISO and be installed in the
target system.
Mount all squashfses in new/iso/casper
at old/{squash_name}
(this
is mostly to make poking at ISOs in a subsequent interactive --shell
action easier).