Merge branch 'releng/11.3' into releng-CDN/11.3

[FreeBSD/FreeBSD.git] / etc / rc.initdiskless

#!/bin/sh
#
# Copyright (c) 1999  Matt Dillon
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# 1. Redistributions of source code must retain the above copyright
#    notice, this list of conditions and the following disclaimer.
# 2. Redistributions in binary form must reproduce the above copyright
#    notice, this list of conditions and the following disclaimer in the
#    documentation and/or other materials provided with the distribution.
#
# THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
# ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
# ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
# OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
# HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
# OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
# SUCH DAMAGE.
#
# $FreeBSD$

# On entry to this script the entire system consists of a read-only root
# mounted via NFS. The kernel has run BOOTP and configured an interface
# (otherwise it would not have been able to mount the NFS root!)
#
# We use the contents of /conf to create and populate memory filesystems
# that are mounted on top of this root to implement the writable
# (and host-specific) parts of the root filesystem, and other volatile
# filesystems.
#
# The hierarchy in /conf has the form /conf/T/M/ where M are directories
# for which memory filesystems will be created and filled,
# and T is one of the "template" directories below:
#
#  base         universal base, typically a replica of the original root;
#  default      secondary universal base, typically overriding some
#               of the files in the original root;
#  ${ipba}      where ${ipba} is the assigned broadcast IP address
#  bcast/${ipba} same as above
#  ${class}     where ${class} is a list of directories supplied by
#               bootp/dhcp through the T134 option.
#               ${ipba} and ${class} are typically used to configure features
#               for group of diskless clients, or even individual features;
#  ${ip}        where ${ip} is the machine's assigned IP address, typically
#               used to set host-specific features;
#  ip/${ip}     same as above
#
# Template directories are scanned in the order they are listed above,
# with each successive directory overriding (merged into) the previous one;
# non-existing directories are ignored.  The subdirectory forms exist to
# help keep the top level /conf manageable in large installations.
#
# The existence of a directory /conf/T/M causes this script to create a
# memory filesystem mounted as /M on the client.
#
# Some files in /conf have special meaning, namely:
#
# Filename      Action
# ----------------------------------------------------------------
# /conf/T/M/remount
#               The contents of the file is a mount command. E.g. if
#               /conf/1.2.3.4/foo/remount contains "mount -o ro /dev/ad0s3",
#               then /dev/ad0s3 will be be mounted on /conf/1.2.3.4/foo/
#
# /conf/T/M/remount_optional
#               If this file exists, then failure to execute the mount
#               command contained in /conf/T/M/remount is non-fatal.
#
# /conf/T/M/remount_subdir
#               If this file exists, then the behaviour of /conf/T/M/remount
#               changes as follows:
#                1. /conf/T/M/remount is invoked to mount the root of the
#                   filesystem where the configuration data exists on a
#                   temporary mountpoint.
#                2. /conf/T/M/remount_subdir is then invoked to mount a
#                   *subdirectory* of the filesystem mounted by
#                   /conf/T/M/remount on /conf/T/M/.
#
# /conf/T/M/diskless_remount
#               The contents of the file points to an NFS filesystem,
#               possibly followed by mount_nfs options. If the server name
#               is omitted, the script will prepend the root path used when
#               booting. E.g. if you booted from foo.com:/path/to/root,
#               an entry for /conf/base/etc/diskless_remount could be any of
#                       foo.com:/path/to/root/etc
#                       /etc -o ro
#               Because mount_nfs understands ".." in paths, it is
#               possible to mount from locations above the NFS root with
#               paths such as "/../../etc".
#
# /conf/T/M/md_size
#               The contents of the file specifies the size of the memory
#               filesystem to be created, in 512 byte blocks.
#               The default size is 10240 blocks (5MB). E.g. if
#               /conf/base/etc/md_size contains "30000" then a 15MB MFS
#               will be created. In case of multiple entries for the same
#               directory M, the last one in the scanning order is used.
#               NOTE: If you only need to create a memory filesystem but not
#               initialize it from a template, it is preferable to specify
#               it in fstab e.g. as  "md /tmp mfs -s=30m,rw 0 0"
#
# /conf/T/SUBDIR.cpio.gz
#               The file is cpio'd into /SUBDIR (and a memory filesystem is
#               created for /SUBDIR if necessary). The presence of this file
#               prevents the copy from /conf/T/SUBDIR/
#
# /conf/T/M/extract
#               This is alternative to SUBDIR.cpio.gz and remount.
#               Similar to remount case, a memory filesystem is created
#               for /M and initialized from a template but no mounting
#               performed. Instead, this file is run passing /M as single
#               argument. It is expected to extract template override to /M
#               using auxiliary storage found in some embedded systems
#               having NVRAM too small to hold mountable file system.
#
# /conf/T/SUBDIR.remove
#               The list of paths contained in the file are rm -rf'd
#               relative to /SUBDIR.
#
# /conf/diskless_remount
#               Similar to /conf/T/M/diskless_remount above, but allows
#               all of /conf to be remounted.  This can be used to allow
#               multiple roots to share the same /conf.
#
#
# You will almost universally want to create the following files under /conf
#
# File                                  Content
# ----------------------------          ----------------------------------
# /conf/base/etc/md_size                size of /etc filesystem
# /conf/base/etc/diskless_remount       "/etc"
# /conf/default/etc/rc.conf             generic diskless config parameters
# /conf/default/etc/fstab               generic diskless fstab e.g. like this
#
#       foo:/root_part                  /       nfs     ro              0 0
#       foo:/usr_part                   /usr    nfs     ro              0 0
#       foo:/home_part                  /home   nfs     rw              0 0
#       md                              /tmp    mfs     -s=30m,rw       0 0
#       md                              /var    mfs     -s=30m,rw       0 0
#       proc                            /proc   procfs  rw              0 0
#
# plus, possibly, overrides for password files etc.
#
# NOTE!  /var, /tmp, and /dev will be typically created elsewhere, e.g.
# as entries in the fstab as above.
# Those filesystems should not be specified in /conf.
#
# (end of documentation, now get to the real code)

dlv=`/sbin/sysctl -n vfs.nfs.diskless_valid 2> /dev/null`

# DEBUGGING
# log something on stdout if verbose.
o_verbose=0     # set to 1 or 2 if you want more debugging
log() {
    [ ${o_verbose} -gt 0 ] && echo "*** $* ***"
    [ ${o_verbose} -gt 1 ] && read -p "=== Press enter to continue" foo
}

# chkerr:
#
# Routine to check for error
#
#       checks error code and drops into shell on failure.
#       if shell exits, terminates script as well as /etc/rc.
#       if remount_optional exists under the mountpoint, skip this check.
#
chkerr() {
    lastitem () ( n=$(($# - 1)) ; shift $n ; echo $1 )
    mountpoint="$(lastitem $2)"
    [ -r $mountpoint/remount_optional ] && ( echo "$2 failed: ignoring due to remount_optional" ; return )
    case $1 in
    0)
        ;;
    *)
        echo "$2 failed: dropping into /bin/sh"
        /bin/sh
        # RESUME
        ;;
    esac
}

# The list of filesystems to umount after the copy
to_umount=""

handle_remount() { # $1 = mount point
    local nfspt mountopts b
    b=$1
    log handle_remount $1
    [ -d $b -a -f $b/diskless_remount ] || return
    read nfspt mountopts < $b/diskless_remount
    log "nfspt ${nfspt} mountopts ${mountopts}"
    # prepend the nfs root if not present
    [ `expr "$nfspt" : '\(.\)'` = "/" ] && nfspt="${nfsroot}${nfspt}"
    mount_nfs $mountopts $nfspt $b
    chkerr $? "mount_nfs $nfspt $b"
    to_umount="$b ${to_umount}"
}

# Create a generic memory disk.
# The 'auto' parameter will attempt to use tmpfs(5), falls back to md(4).
# $1 is size in 512-byte sectors, $2 is the mount point.
mount_md() {
    /sbin/mdmfs -s $1 auto $2
}

# Create the memory filesystem if it has not already been created
#
create_md() {
        [ "x`eval echo \\$md_created_$1`" = "x" ] || return # only once
        if [ "x`eval echo \\$md_size_$1`" = "x" ]; then
            md_size=10240
        else
            md_size=`eval echo \\$md_size_$1`
        fi
        log create_md $1 with size $md_size
        mount_md $md_size /$1
        /bin/chmod 755 /$1
        eval md_created_$1=created
}

# DEBUGGING
#
# set -v

# Figure out our interface and IP.
#
bootp_ifc=""
bootp_ipa=""
bootp_ipbca=""
class=""
if [ ${dlv:=0} -ne 0 ] ; then
        iflist=`ifconfig -l`
        for i in ${iflist} ; do
            set -- `ifconfig ${i}`
            while [ $# -ge 1 ] ; do
                if [ "${bootp_ifc}" = "" -a "$1" = "inet" ] ; then
                    bootp_ifc=${i} ; bootp_ipa=${2} ; shift
                fi
                if [ "${bootp_ipbca}" = "" -a "$1" = "broadcast" ] ; then
                    bootp_ipbca=$2; shift
                fi
                shift
            done
            if [ "${bootp_ifc}" != "" ] ; then
                break
            fi
        done
        # Get the values passed with the T134 bootp cookie.
        class="`/sbin/sysctl -qn kern.bootp_cookie`"

        echo "Interface ${bootp_ifc} IP-Address ${bootp_ipa} Broadcast ${bootp_ipbca} ${class}"
fi

log Figure out our NFS root path
#
set -- `mount -t nfs`
while [ $# -ge 1 ] ; do
    if [ "$2" = "on" -a "$3" = "/" ]; then
        nfsroot="$1"
        break
    fi
    shift
done

# The list of directories with template files
templates="base default"
if [ -n "${bootp_ipbca}" ]; then
        templates="${templates} ${bootp_ipbca} bcast/${bootp_ipbca}"
fi
if [ -n "${class}" ]; then
        templates="${templates} ${class}"
fi
if [ -n "${bootp_ipa}" ]; then
        templates="${templates} ${bootp_ipa} ip/${bootp_ipa}"
fi

# If /conf/diskless_remount exists, remount all of /conf.
handle_remount /conf

# Resolve templates in /conf/base, /conf/default, /conf/${bootp_ipbca},
# and /conf/${bootp_ipa}.  For each subdirectory found within these
# directories:
#
# - calculate memory filesystem sizes.  If the subdirectory (prior to
#   NFS remounting) contains the file 'md_size', the contents specified
#   in 512 byte sectors will be used to size the memory filesystem.  Otherwise
#   8192 sectors (4MB) is used.
#
# - handle NFS remounts.  If the subdirectory contains the file
#   diskless_remount, the contents of the file is NFS mounted over
#   the directory.  For example /conf/base/etc/diskless_remount
#   might contain 'myserver:/etc'.  NFS remounts allow you to avoid
#   having to dup your system directories in /conf.  Your server must
#   be sure to export those filesystems -alldirs, however.
#   If the diskless_remount file contains a string beginning with a
#   '/' it is assumed that the local nfsroot should be prepended to
#   it before attemping to the remount.  This allows the root to be
#   relocated without needing to change the remount files.
#
log "templates are ${templates}"
for i in ${templates} ; do
    for j in /conf/$i/* ; do
        [ -d $j ] || continue

        # memory filesystem size specification
        subdir=${j##*/}
        [ -f $j/md_size ] && eval md_size_$subdir=`cat $j/md_size`

        # remount. Beware, the command is in the file itself!
        if [ -f $j/remount ]; then
            if [ -f $j/remount_subdir ]; then
                k="/conf.tmp/$i/$subdir"
                [ -d $k ] || continue

                # Mount the filesystem root where the config data is
                # on the temporary mount point.
                nfspt=`/bin/cat $j/remount`
                $nfspt $k
                chkerr $? "$nfspt $k"

                # Now use a nullfs mount to get the data where we
                # really want to see it.
                remount_subdir=`/bin/cat $j/remount_subdir`
                remount_subdir_cmd="mount -t nullfs $k/$remount_subdir"

                $remount_subdir_cmd $j
                chkerr $? "$remount_subdir_cmd $j"

                # XXX check order -- we must force $k to be unmounted
                # after j, as j depends on k.
                to_umount="$j $k ${to_umount}"
            else
                nfspt=`/bin/cat $j/remount`
                $nfspt $j
                chkerr $? "$nfspt $j"
                to_umount="$j ${to_umount}" # XXX hope it is really a mount!
            fi
        fi

        # NFS remount
        handle_remount $j
    done
done

# - Create all required MFS filesystems and populate them from
#   our templates.  Support both a direct template and a dir.cpio.gz
#   archive. Support for auxiliary NVRAM. Support dir.remove files containing
#   a list of relative paths to remove.
#
# The dir.cpio.gz form is there to make the copy process more efficient,
# so if the cpio archive is present, it prevents the files from dir/
# from being copied.

for i in ${templates} ; do
    for j in /conf/$i/* ; do
        subdir=${j##*/}
        if [ -d $j -a ! -f $j.cpio.gz  ]; then
            create_md $subdir
            cp -Rp $j/ /$subdir
        fi
    done
    for j in /conf/$i/*.cpio.gz ; do
        subdir=${j%*.cpio.gz}
        subdir=${subdir##*/}
        if [ -f $j ]; then
            create_md $subdir
            echo "Loading /$subdir from cpio archive $j"
            (cd / ; /rescue/tar -xpf $j)
        fi
    done
    for j in /conf/$i/*/extract ; do
        if [ -x $j ]; then
            subdir=${j%*/extract}
            subdir=${subdir##*/}
            create_md $subdir
            echo "Loading /$subdir using auxiliary command $j"
            $j /$subdir
        fi
    done
    for j in /conf/$i/*.remove ; do
        subdir=${j%*.remove}
        subdir=${subdir##*/}
        if [ -f $j ]; then
            # doubly sure it is a memory disk before rm -rf'ing
            create_md $subdir
            (cd /$subdir; rm -rf `/bin/cat $j`)
        fi
    done
done

# umount partitions used to fill the memory filesystems
[ -n "${to_umount}" ] && umount $to_umount