path: root/chapter08/pkgmgt.xml

<?xml version="1.0" encoding="ISO-8859-1"?>
<!DOCTYPE sect1 PUBLIC "-//OASIS//DTD DocBook XML V4.5//EN"
  "http://www.oasis-open.org/docbook/xml/4.5/docbookx.dtd" [
  <!ENTITY % general-entities SYSTEM "../general.ent">
  %general-entities;
]>

<sect1 id="ch-system-pkgmgt">
  <?dbhtml filename="pkgmgt.html"?>

  <title>Package Management</title>

  <para>Package Management is an often requested addition to the LFS Book. A
  Package Manager allows tracking the installation of files making it easy to
  remove and upgrade packages. As well as the binary and library files, a
  package manager will handle the installation of configuration files.  Before
  you begin to wonder, NO&mdash;this section will not talk about nor recommend
  any particular package manager. What it provides is a roundup of the more
  popular techniques and how they work. The perfect package manager for you may
  be among these techniques or may be a combination of two or more of these
  techniques. This section briefly mentions issues that may arise when upgrading
  packages.</para>

  <para>Some reasons why no package manager is mentioned in LFS or BLFS
  include:</para>

  <itemizedlist>
    <listitem>
      <para>Dealing with package management takes the focus away from the goals
      of these books&mdash;teaching how a Linux system is built.</para>
    </listitem>

    <listitem>
      <para>There are multiple solutions for package management, each having
      its strengths and drawbacks.  Including one that satisfies all audiences
      is difficult.</para>
    </listitem>
  </itemizedlist>

  <para>There are some hints written on the topic of package management. Visit
  the <ulink url="&hints-root;">Hints Project</ulink> and see if one of them
  fits your need.</para>

  <sect2 id='pkgmgmt-upgrade-issues'>
    <title>Upgrade Issues</title>

    <para>A Package Manager makes it easy to upgrade to newer versions when they
    are released. Generally the instructions in the LFS and BLFS books can be
    used to upgrade to the newer versions. Here are some points that you should
    be aware of when upgrading packages, especially on a running system.</para>

    <itemizedlist>
      <listitem>
        <para>If Linux kernel needs to be upgraded (for example, from
        5.10.17 to 5.10.18 or 5.11.1), nothing else need to be rebuilt.
        The system will keep working fine thanks to the well-defined border
        between kernel and userspace.  Specifically, Linux API headers
        need not to be (and should not be, see the next item) upgraded
        alongside the kernel.  You'll need to reboot your system to use the
        upgraded kernel.</para>
      </listitem>

      <listitem>
        <para>If Linux API headers or Glibc needs to be upgraded to a newer
        version, (e.g.  from glibc-2.31 to glibc-2.32), it is safer to
        rebuild LFS.  Though you <emphasis>may</emphasis> be able to rebuild
        all the packages in their dependency order, we do not recommend
        it.  </para>
      </listitem>

      <listitem> <para>If a package containing a shared library is updated, and
      if the name of the library changes, then any the packages dynamically
      linked to the library need to be recompiled in order to link against the
      newer library.  (Note that there is no correlation between the package
      version and the name of the library.) For example, consider a package
      foo-1.2.3 that installs a shared library with name <filename
      class='libraryfile'>libfoo.so.1</filename>. If you upgrade the package to
      a newer version foo-1.2.4 that installs a shared library with name
      <filename class='libraryfile'>libfoo.so.2</filename>. In this case, any
      packages that are dynamically linked to <filename
      class='libraryfile'>libfoo.so.1</filename> need to be recompiled to link
      against <filename class='libraryfile'>libfoo.so.2</filename> in order to
      use the new library version. You should not remove the previous
      libraries unless all the dependent packages are recompiled.</para>
      </listitem>

      <listitem> <para>If a package containing a shared library is updated,
      and the name of library doesn't change, but the version number of the
      library <emphasis role="bold">file</emphasis> decreases (for example,
      the name of the library is kept named
      <filename class='libraryfile'>libfoo.so.1</filename>,
      but the name of library file is changed from
      <filename class='libraryfile'>libfoo.so.1.25</filename> to
      <filename class='libraryfile'>libfoo.so.1.24</filename>),
      you should remove the library file from the previously installed version
      (<filename class='libraryfile'>libfoo.so.1.25</filename> in the case).
      Or, a <command>ldconfig</command> run (by yourself using a command
      line, or by the installation of some package) will reset the symlink
      <filename class='libraryfile'>libfoo.so.1</filename> to point to
      the old library file because it seems having a <quote>newer</quote>
      version, as its version number is larger.  This situation may happen if
      you have to downgrade a package, or the package changes the versioning
      scheme of library files suddenly.</para> </listitem>

      <listitem> <para>If a package containing a shared library is updated,
      and the name of library doesn't change, but a severe issue
      (especially, a security vulnerability) is fixed, all running programs
      linked to the shared library should be restarted.  The following
      command, run as <systemitem class="username">root</systemitem> after
      updating, will list what is using the old versions of those libraries
      (replace <replaceable>libfoo</replaceable> with the name of the
      library):</para>

<screen><userinput role="nodump">grep -l  -e '<replaceable>libfoo</replaceable>.*deleted' /proc/*/maps |
   tr -cd 0-9\\n | xargs -r ps u</userinput></screen>

      <para>
        If <application>OpenSSH</application> is being used for accessing
        the system and it is linked to the updated library, you need to
        restart <command>sshd</command> service, then logout, login again,
        and rerun that command to confirm nothing is still using the
        deleted libraries.
      </para></listitem>

      <listitem>
        <para>If a binary or a shared library is overwrote, the processes
        using the code or data in the binary or library may crash.  The
        correct way to update a binary or a shared library without causing
        the process to crash is: remove it first, then install the new
        version into position. The <command>install</command> command
        provided by <application>Coreutils</application> has already
        implemented this and most packages use it to install binaries and
        libraries.  So you won't be troubled by this issue most of the time.
        However, the install process of some packages (notably Mozilla JS
        in BLFS) just overwrites the file if it exists and causes crash, so
        it's safer to save your work and close unneeded running processes
        before updating a package.</para>
      </listitem>
    </itemizedlist>

  </sect2>

  <sect2>
    <title>Package Management Techniques</title>

    <para>The following are some common package management techniques. Before
    making a decision on a package manager, do some research on the various
    techniques, particularly the drawbacks of the particular scheme.</para>

    <sect3>
      <title>It is All in My Head!</title>

      <para>Yes, this is a package management technique. Some folks do not find
      the need for a package manager because they know the packages intimately
      and know what files are installed by each package. Some users also do not
      need any package management because they plan on rebuilding the entire
      system when a package is changed.</para>

    </sect3>

    <sect3>
      <title>Install in Separate Directories</title>

      <para>This is a simplistic package management that does not need any extra
      package to manage the installations. Each package is installed in a
      separate directory. For example, package foo-1.1 is installed in
      <filename class='directory'>/usr/pkg/foo-1.1</filename>
      and a symlink is made from <filename>/usr/pkg/foo</filename> to
      <filename class='directory'>/usr/pkg/foo-1.1</filename>. When installing
      a new version foo-1.2, it is installed in
      <filename class='directory'>/usr/pkg/foo-1.2</filename> and the previous
      symlink is replaced by a symlink to the new version.</para>

      <para>Environment variables such as <envar>PATH</envar>,
      <envar>LD_LIBRARY_PATH</envar>, <envar>MANPATH</envar>,
      <envar>INFOPATH</envar> and <envar>CPPFLAGS</envar> need to be expanded to
      include <filename>/usr/pkg/foo</filename>. For more than a few packages,
      this scheme becomes unmanageable.</para>

    </sect3>

    <sect3>
      <title>Symlink Style Package Management</title>

      <para>This is a variation of the previous package management technique.
      Each package is installed similar to the previous scheme. But instead of
      making the symlink, each file is symlinked into the
      <filename class='directory'>/usr</filename> hierarchy. This removes the
      need to expand the environment variables. Though the symlinks can be
      created by the user to automate the creation, many package managers have
      been written using this approach. A few of the popular ones include Stow,
      Epkg, Graft, and Depot.</para>

      <para>The installation needs to be faked, so that the package thinks that
      it is installed in <filename class="directory">/usr</filename> though in
      reality it is installed in the
      <filename class="directory">/usr/pkg</filename> hierarchy. Installing in
      this manner is not usually a trivial task. For example, consider that you
      are installing a package libfoo-1.1. The following instructions may
      not install the package properly:</para>

<screen role="nodump"><userinput>./configure --prefix=/usr/pkg/libfoo/1.1
make
make install</userinput></screen>

      <para>The installation will work, but the dependent packages may not link
      to libfoo as you would expect. If you compile a package that links against
      libfoo, you may notice that it is linked to
      <filename class='libraryfile'>/usr/pkg/libfoo/1.1/lib/libfoo.so.1</filename>
      instead of <filename class='libraryfile'>/usr/lib/libfoo.so.1</filename>
      as you would expect. The correct approach is to use the
      <envar>DESTDIR</envar> strategy to fake installation of the package. This
      approach works as follows:</para>

<screen role="nodump"><userinput>./configure --prefix=/usr
make
make DESTDIR=/usr/pkg/libfoo/1.1 install</userinput></screen>

      <para>Most packages support this approach, but there are some which do not.
      For the non-compliant packages, you may either need to manually install the
      package, or you may find that it is easier to install some problematic
      packages into <filename class='directory'>/opt</filename>.</para>

    </sect3>

    <sect3>
      <title>Timestamp Based</title>

      <para>In this technique, a file is timestamped before the installation of
      the package. After the installation, a simple use of the
      <command>find</command> command with the appropriate options can generate
      a log of all the files installed after the timestamp file was created. A
      package manager written with this approach is install-log.</para>

      <para>Though this scheme has the advantage of being simple, it has two
      drawbacks. If, during installation, the files are installed with any
      timestamp other than the current time, those files will not be tracked by
      the package manager. Also, this scheme can only be used when one package
      is installed at a time. The logs are not reliable if two packages are
      being installed on two different consoles.</para>

    </sect3>

    <sect3>
      <title>Tracing Installation Scripts</title>

      <para>In this approach, the commands that the installation scripts perform
      are recorded.  There are two techniques that one can use:</para>

      <para>The <envar>LD_PRELOAD</envar> environment variable can be set to
      point to a library to be preloaded before installation.  During
      installation, this library tracks the packages that are being installed by
      attaching itself to various executables such as <command>cp</command>,
      <command>install</command>, <command>mv</command> and tracking the system
      calls that modify the filesystem. For this approach to work, all the
      executables need to be dynamically linked without the suid or sgid bit.
      Preloading the library may cause some unwanted side-effects during
      installation. Therefore, it is advised that one performs some tests to
      ensure that the package manager does not break anything and logs all the
      appropriate files.</para>

      <para>The second technique is to use <command>strace</command>, which
      logs all system calls made during the execution of the installation
      scripts.</para>
    </sect3>

    <sect3>
      <title>Creating Package Archives</title>

      <para>In this scheme, the package installation is faked into a separate
      tree as described in the Symlink style package management. After the
      installation, a package archive is created using the installed files.
      This archive is then used to install the package either on the local
      machine or can even be used to install the package on other machines.</para>

      <para>This approach is used by most of the package managers found in the
      commercial distributions. Examples of package managers that follow this
      approach are RPM (which, incidentally, is required by the <ulink
      url="http://refspecs.linuxfoundation.org/lsb.shtml">Linux
      Standard Base Specification</ulink>), pkg-utils, Debian's apt, and
      Gentoo's Portage system.  A hint describing how to adopt this style of
      package management for LFS systems is located at <ulink
      url="&hints-root;fakeroot.txt"/>.</para>

      <para>Creation of package files that include dependency information is
      complex and is beyond the scope of LFS.</para>

      <para>Slackware uses a <command>tar</command> based system for package
      archives.  This system purposely does not handle package dependencies
      as more complex package managers do.  For details of Slackware package
      management, see <ulink
      url="http://www.slackbook.org/html/package-management.html"/>.</para>
    </sect3>

    <sect3>
      <title>User Based Management</title>

      <para>This scheme, unique to LFS, was devised by Matthias Benkmann, and is
      available from the <ulink url="&hints-root;">Hints Project</ulink>. In
      this scheme, each package is installed as a separate user into the
      standard locations. Files belonging to a package are easily identified by
      checking the user ID. The features and shortcomings of this approach are
      too complex to describe in this section. For the details please see the
      hint at <ulink url="&hints-root;more_control_and_pkg_man.txt"/>.</para>

    </sect3>

  </sect2>

  <sect2>
    <title>Deploying LFS on Multiple Systems</title>

    <para>One of the advantages of an LFS system is that there are no files that
    depend on the position of files on a disk system.  Cloning an LFS build to
    another computer with the same architecture as the base system is as
    simple as using <command>tar</command> on the LFS partition that contains
    the root directory (about 250MB uncompressed for a base LFS build), copying
    that file via network transfer or CD-ROM to the new system and expanding
    it.  From that point, a few configuration files will have to be changed.
    Configuration files that may need to be updated include:
    <filename>/etc/hosts</filename>,
    <filename>/etc/fstab</filename>,
    <filename>/etc/passwd</filename>,
    <filename>/etc/group</filename>,
    <phrase revision="systemd">
      <filename>/etc/shadow</filename>, and
      <filename>/etc/ld.so.conf</filename>.
    </phrase>
    <phrase revision="sysv">
      <filename>/etc/shadow</filename>,
      <filename>/etc/ld.so.conf</filename>,
      <filename>/etc/sysconfig/rc.site</filename>,
      <filename>/etc/sysconfig/network</filename>, and
      <filename>/etc/sysconfig/ifconfig.eth0</filename>.
    </phrase>
    </para>

    <para>A custom kernel may need to be built for the new system depending on
    differences in system hardware and the original kernel
    configuration.</para>

    <note><para>There have been some reports of issues when copying between
    similar but not identical architectures. For instance, the instruction set
    for an Intel system is not identical with an AMD processor and later
    versions of some processors may have instructions that are unavailable in
    earlier versions.</para></note>

    <para>Finally the new system has to be made bootable via <xref
    linkend="ch-bootable-grub"/>.</para>

  </sect2>

</sect1>