Now that we assume ownership of /boot/initrd.img-<version> for all
installed kernel versions, an update or deletion command with "-k all"
should be applied to all such files, even if we didn't create them.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

In Debian, update-initramfs has always refused to overwrite or delete
an initramfs image that it didn't create, unless the -t option is
used.  On Ubuntu, this check is apparently disabled.  The commit log
doesn't provide much of a clue as to why.  Looking at the the state of
Debian and the linux-2.6 package at the time, I think the reasoning
must have been something like:

(1) Multiple initramfs/initrd generators were available and could be
coinstalled.  The configuration file /etc/kernel-img.conf specified
which one kernel packages should invoke ("ramdisk" variable).

(2) Packages that hooked into initramfs-tools would always run update-
initramfs if it existed, and it would immediately update the
initramfs, because triggers hadn't yet been invented.

(3) To keep the alternate initramfs generators working, update-
initramfs therefore had to either (a) check the configuration file to
find out whether initramfs-tools was meant to be used, or (b) keep
track of which initramfs images it created and therefore "owned".

At this point, (1) is no longer true; the "ramdisk" variable is not
honoured and the initramfs generator to be invoked is chosen by
installing one of several conflicting packages.  I think this means
that (3) no longer follows, and whichever of the automatic initramfs
generator packages is currently installed should be considered to
"own" /boot/initrd.img-<version> for all kernel versions.

This commit only disables the checks.  The state directory and the
state updates are retained for now, in case it becomes necessary to
revert this change.

Closes: #765750, #804295
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

zram is ephemeral (it's a compressed RAM disk) so we must not use it.

Closes: #861228
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

If we reject the largest swap device then we should look at any
other swap devices too.

This is unlikely to make a difference by itself, since we only
reject swap devices encrypted with a random key, and there are
unlikely to be other swap devices configured differently.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

On a usr-merged filesystem /bin, /lib*, /sbin are symlinks to the
corresponding subdirectories of /usr.  When the host filesystem looks
like this, copy_file may be asked to copy both /lib/foo and
/usr/lib/foo.  It's too late at this point to make /lib a symlink
in the initramfs, and copy_file isn't general enough to do that.

Avoid this by setting up the initramfs as usr-merged.  This could
possibly cause file conflicts, but it's unlikely because:

* Policy forbids such conflicts on the host filesystem
* Few executables and libraries that go into the initramfs are
  explicitly installed in /usr on the host filesystem, because
  of historical support for having the init system mount /usr
* mkinitramfs never used to create /usr/bin or /usr/lib

Closes: #874523
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Gbp-Dch: ignore
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Change filename matching in manual_add_modules, copy_modules_dir,
hidden_dep_add_modules to allow a compression suffix after '.ko'
in filenames.

kmod is responsible for decompressing modules, and Debian's kmod
currently only supports xz compression.  We allow any compression
suffix here, so if kmod is rebuilt with gzip/zlib support enabled then
initramfs-tools should automatically support it too.

Closes: #891560
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Exclusions may include wildcards, which should be matched by find
and not the shell.  Change the argument processing to manipulate
the argument array, instead of building a single variable which
cannot maintain quoting.

(This would be easier in bash - we could use a named array variable.)

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

"rmdir --ignore-fail-on-non-empty" won't report an error if there
is ever another file in /var/cache/ldconfig, but it will report an
error if the directory doesn't exist at all.

"rm -df" has the opposite problem.

Add a test for directory existence around the aux-cache removal.

Closes: #904065
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Gbp-Dch: ignore
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

upstart will never be fixed, so it is correct to not specify a version.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

The document it refers to is no longer packaged.

Gbp-Dch: ignore
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Set Rules-Requires-Root: no.  No other policy changes seem to be
relevant.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

The broken cryptsetup, elilo, lilo, s390-tools, console-setup,
systemd-sysv, and lvm2 versions are all older than jessie (current
oldstable).  We can assume that they will have been upgraded without
an explicit Breaks.

kmod replaced module-init-tools in wheezy (current oldoldstable)
so there is no need to allow the latter as an alternative.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

I updated the wrong filename here when moving the initramfs-tools man
page.

Gbp-Dch: ignore
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

The initramfs-tools(7) manual page is far more useful.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

maintainer-notes.md (formerly .html) is only useful for those working
on the source package.

HACKING might be slightly useful to maintainers of other packages, but
initramfs-tools(7) is much more useful.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Set the .TH source parameter for all pages to "initramfs-tools".
Set the .TH manual parameter to a commonly used name for the man page
section.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Set each page's date to the date of the last substantive
change, i.e. excluding cross-reference updates and spelling fixes.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Manual pages that describe a general topic belong in section 7, not 8.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

An initramfs image is not a ramdisk, even though it is passed to the
kernel in the same way as an initrd.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

The --supported-{host,target}-version option were only supported by
mkinitramfs-kpkg, which was removed in version 0.99.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Kristian Klausen pointed out in #845034 that even after the preceding
commit /var/cache/ldconfig/aux-cache is unreproducible.  ldconfig
creates this file if /var/cache already exists.

The reason that it's unreproducible is that it's an associative array
keyed by file attributes including device and inode number, which will
change on each build.  They will also be different when the initramfs
image is unpacked, making the cache useless at boot time.

Kristian's proposal was to disable ldconfig creating caches.  But
/etc/ld.so.cache is useful and seems to be reproducible.  Therefore,
keep running ldconfig but always remove the aux-cache file.

Thanks: Kristian Klausen <klausenbusk@hotmail.com>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

[bwh: Resolved conflicts with lz4 support and updated Depends]
Closes: #845034
Thanks: Chris Lamb <lamby@debian.org>
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

[bwh: Also unset it before invoking init]
Closes: #883664
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Closes: #887211
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

It is possible for the early initramfs to consist of multiple
sections, each a valid uncompressed cpio archive, for example if
amd64-microcode and intel-microcode are both installed and explicitly
enabled.  Each cpio archive ends with an EOF marker consisting of four
zero bytes and possibly additional zero padding.

The kernel's early microcode loader simply skips the zero bytes, but
cpio will stop.  So whenever we find such a group of zero bytes, we
need to stop parsing cpio headers, run cpio over what we found, and
then repeat the process on the following bytes.

Closes: #886424
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Include vmd in the block drivers group, and add the dependency to
hidden_dep_add_modules.

Thanks: Mario Limonciello <Mario.Limonciello@dell.com>
Closes: #886349
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

On some systems the keyboard controller doesn't behave properly until
psmouse has also done its initialisation.  We already include psmouse
in MODULES=most builds for this reason.

Close: #878829
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Built-in "modules" can have weak dependencies on modules, and we
will need to handle that here.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Module installation directories reflect their source directory, and
since the kernel source is sometimes reorganised it isn't very
future-proof to hard-code directories here.

Rewrite the file tests as an iteration over all modules already
installed.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Currently the "bugreports @ ubuntu" link goes to the Launchpad
package overview for initramfs-tools, not the bug list.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

I think tracker.debian.org shows all the information on the old PTS,
and more.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>

The existing link specifies the package name initramfs-tools, so the
resulting list won't include bugs assigned to initramfs-tools-core.
We have to specify source package name (src query parameter) to get
everything.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>