source: chapter07/kernfs.xml

<?xml version="1.0" encoding="UTF-8"?>
<!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-tools-kernfs">
  <?dbhtml filename="kernfs.html"?>

  <title>Preparing Virtual Kernel File Systems</title>

  <indexterm zone="ch-tools-kernfs">
    <primary sortas="e-/dev/">/dev/*</primary>
  </indexterm>

    <para>Applications running in userspace utilize various file 
    systems created by the kernel to communicate 
    with the kernel itself. These file systems are virtual: no disk
    space is used for them. The content of these file systems resides in
    memory. These file systems must be mounted in the $LFS directory tree
    so the applications can find them in the chroot environment.</para>

    <para>Begin by creating the directories on which these virtual file systems will be
    mounted:</para>

<screen><userinput>mkdir -pv $LFS/{dev,proc,sys,run}</userinput></screen>

  <sect2 id="ch-tools-bindmount">
    <title>Mounting and Populating /dev</title>

      <para>During a normal boot of an LFS system, the kernel automatically
      mounts the <systemitem class="filesystem">devtmpfs</systemitem>
      file system on the
      <filename class="directory">/dev</filename> directory; the kernel
      creates device nodes on that virtual file system during the boot process,
      or when a device is first detected or accessed. The udev daemon may
      change the ownership or permissions of the device nodes created by the
      kernel, and create new device nodes or symlinks, to ease the work of
      distro maintainers and system administrators.  (See
      <xref linkend='ch-config-udev-device-node-creation'/> for details.)
      If the host kernel supports &devtmpfs;, we can simply mount a
      &devtmpfs; at <filename class='directory'>$LFS/dev</filename> and rely
      on the kernel to populate it.</para>

      <para>But some host kernels lack &devtmpfs; support; these
      host distros use different methods to create the content of
      <filename class="directory">/dev</filename>.
      So the only host-agnostic way to populate the
      <filename class="directory">$LFS/dev</filename> directory is
      by bind mounting the host system's
      <filename class="directory">/dev</filename> directory. A bind mount is
      a special type of mount that makes a directory subtree or a file
      visible at some other location. Use the following
      command to do this.</para>

<screen><userinput>mount -v --bind /dev $LFS/dev</userinput></screen>

  </sect2>

  <sect2 id="ch-tools-kernfsmount">
    <title>Mounting Virtual Kernel File Systems</title>

      <para>Now mount the remaining virtual kernel file systems:</para>

      <!-- Do not put any option after $LFS/${mountpoint} or jhalfs cannot
           handle it! -->

<screen><userinput>mount -vt devpts devpts -o gid=5,mode=0620 $LFS/dev/pts
mount -vt proc proc $LFS/proc
mount -vt sysfs sysfs $LFS/sys
mount -vt tmpfs tmpfs $LFS/run</userinput></screen>

    <variablelist>
      <title>The meaning of the mount options for devpts:</title>

      <varlistentry>
        <term><parameter>gid=5</parameter></term>
        <listitem>
          <para>This ensures that all devpts-created device nodes are owned by
          group ID 5.  This is the ID we will use later on for the <systemitem
          class="groupname">tty</systemitem> group.  We use the group ID instead
          of a name, since the host system might use a different ID for its
          <systemitem class="groupname">tty</systemitem> group.</para>
        </listitem>
      </varlistentry>

      <varlistentry>
        <term><parameter>mode=0620</parameter></term>
        <listitem>
          <para>This ensures that all devpts-created device nodes have mode 0620
          (user readable and writable, group writable).  Together with the
          option above, this ensures that devpts will create device nodes that
          meet the requirements of grantpt(), meaning the Glibc
          <command>pt_chown</command> helper binary (which is not installed by
          default) is not necessary.</para>
        </listitem>
      </varlistentry>

    </variablelist>

      <para>In some host systems, <filename>/dev/shm</filename> is a
      symbolic link to a directory, typically
      <filename class="directory">/run/shm</filename>.
      The /run tmpfs was mounted above so in this case only a
      directory needs to be created with the correct permissions.</para>

      <para>In other host systems <filename>/dev/shm</filename> is a mount point 
      for a tmpfs. In that case the mount of /dev above will only create
      /dev/shm as a directory in the chroot environment. In this situation
      we must explicitly mount a tmpfs:</para>

<screen><userinput>if [ -h $LFS/dev/shm ]; then
  install -v -d -m 1777 $LFS$(realpath /dev/shm)
else
  mount -vt tmpfs -o nosuid,nodev tmpfs $LFS/dev/shm
fi</userinput></screen>

  </sect2>

</sect1>