xref: /freebsd/libexec/rc/rc.initdiskless (revision cfd6422a5217410fbd66f7a7a8a64d9d85e61229)
1#!/bin/sh
2#
3# Copyright (c) 1999  Matt Dillon
4# All rights reserved.
5#
6# Redistribution and use in source and binary forms, with or without
7# modification, are permitted provided that the following conditions
8# are met:
9# 1. Redistributions of source code must retain the above copyright
10#    notice, this list of conditions and the following disclaimer.
11# 2. Redistributions in binary form must reproduce the above copyright
12#    notice, this list of conditions and the following disclaimer in the
13#    documentation and/or other materials provided with the distribution.
14#
15# THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16# ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18# ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19# FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20# DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21# OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22# HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23# LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24# OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25# SUCH DAMAGE.
26#
27# $FreeBSD$
28
29# On entry to this script the entire system consists of a read-only root
30# mounted via NFS. The kernel has run BOOTP and configured an interface
31# (otherwise it would not have been able to mount the NFS root!)
32#
33# We use the contents of /conf to create and populate memory filesystems
34# that are mounted on top of this root to implement the writable
35# (and host-specific) parts of the root filesystem, and other volatile
36# filesystems.
37#
38# The hierarchy in /conf has the form /conf/T/M/ where M are directories
39# for which memory filesystems will be created and filled,
40# and T is one of the "template" directories below:
41#
42#  base		universal base, typically a replica of the original root;
43#  default	secondary universal base, typically overriding some
44#		of the files in the original root;
45#  ${ipba}	where ${ipba} is the assigned broadcast IP address
46#  bcast/${ipba} same as above
47#  ${class}	where ${class} is a list of directories supplied by
48#		bootp/dhcp through the T134 option.
49#		${ipba} and ${class} are typically used to configure features
50#		for group of diskless clients, or even individual features;
51#  ${ip}	where ${ip} is the machine's assigned IP address, typically
52#		used to set host-specific features;
53#  ip/${ip}	same as above
54#
55# Template directories are scanned in the order they are listed above,
56# with each successive directory overriding (merged into) the previous one;
57# non-existing directories are ignored.  The subdirectory forms exist to
58# help keep the top level /conf manageable in large installations.
59#
60# The existence of a directory /conf/T/M causes this script to create a
61# memory filesystem mounted as /M on the client.
62#
63# Some files in /conf have special meaning, namely:
64#
65# Filename	Action
66# ----------------------------------------------------------------
67# /conf/T/M/remount
68#		The contents of the file is a mount command. E.g. if
69# 		/conf/1.2.3.4/foo/remount contains "mount -o ro /dev/ad0s3",
70#		then /dev/ad0s3 will be mounted on /conf/1.2.3.4/foo/
71#
72# /conf/T/M/remount_optional
73#		If this file exists, then failure to execute the mount
74#		command contained in /conf/T/M/remount is non-fatal.
75#
76# /conf/T/M/remount_subdir
77#		If this file exists, then the behaviour of /conf/T/M/remount
78#		changes as follows:
79#		 1. /conf/T/M/remount is invoked to mount the root of the
80#		    filesystem where the configuration data exists on a
81#		    temporary mountpoint.
82#		 2. /conf/T/M/remount_subdir is then invoked to mount a
83#		    *subdirectory* of the filesystem mounted by
84#		    /conf/T/M/remount on /conf/T/M/.
85#
86# /conf/T/M/diskless_remount
87#		The contents of the file points to an NFS filesystem,
88#		possibly followed by mount_nfs options. If the server name
89#		is omitted, the script will prepend the root path used when
90#		booting. E.g. if you booted from foo.com:/path/to/root,
91#		an entry for /conf/base/etc/diskless_remount could be any of
92#			foo.com:/path/to/root/etc
93#			/etc -o ro
94#		Because mount_nfs understands ".." in paths, it is
95#		possible to mount from locations above the NFS root with
96#		paths such as "/../../etc".
97#
98# /conf/T/M/md_size
99#		The contents of the file specifies the size of the memory
100#		filesystem to be created, in 512 byte blocks.
101#		The default size is 10240 blocks (5MB). E.g. if
102#		/conf/base/etc/md_size contains "30000" then a 15MB MFS
103#		will be created. In case of multiple entries for the same
104#		directory M, the last one in the scanning order is used.
105#		NOTE: If you only need to create a memory filesystem but not
106#		initialize it from a template, it is preferable to specify
107#		it in fstab e.g. as  "md /tmp mfs -s=30m,rw 0 0"
108#
109# /conf/T/SUBDIR.cpio.gz
110#		The file is cpio'd into /SUBDIR (and a memory filesystem is
111#		created for /SUBDIR if necessary). The presence of this file
112#		prevents the copy from /conf/T/SUBDIR/
113#
114# /conf/T/M/extract
115#		This is alternative to SUBDIR.cpio.gz and remount.
116#		Similar to remount case, a memory filesystem is created
117#		for /M and initialized from a template but no mounting
118#		performed. Instead, this file is run passing /M as single
119#		argument. It is expected to extract template override to /M
120#		using auxiliary storage found in some embedded systems
121#		having NVRAM too small to hold mountable file system.
122#
123# /conf/T/SUBDIR.remove
124#		The list of paths contained in the file are rm -rf'd
125#		relative to /SUBDIR.
126#
127# /conf/diskless_remount
128#		Similar to /conf/T/M/diskless_remount above, but allows
129#		all of /conf to be remounted.  This can be used to allow
130#		multiple roots to share the same /conf.
131#
132#
133# You will almost universally want to create the following files under /conf
134#
135# File					Content
136# ----------------------------		----------------------------------
137# /conf/base/etc/md_size		size of /etc filesystem
138# /conf/base/etc/diskless_remount	"/etc"
139# /conf/default/etc/rc.conf		generic diskless config parameters
140# /conf/default/etc/fstab		generic diskless fstab e.g. like this
141#
142#	foo:/root_part			/	nfs	ro		0 0
143#	foo:/usr_part			/usr	nfs     ro		0 0
144#	foo:/home_part			/home   nfs     rw      	0 0
145#	md				/tmp	mfs     -s=30m,rw	0 0
146#	md				/var	mfs	-s=30m,rw	0 0
147#	proc				/proc	procfs	rw		0 0
148#
149# plus, possibly, overrides for password files etc.
150#
151# NOTE!  /var, /tmp, and /dev will be typically created elsewhere, e.g.
152# as entries in the fstab as above.
153# Those filesystems should not be specified in /conf.
154#
155# (end of documentation, now get to the real code)
156
157dlv=`/sbin/sysctl -n vfs.nfs.diskless_valid 2> /dev/null`
158
159# DEBUGGING
160# log something on stdout if verbose.
161o_verbose=0     # set to 1 or 2 if you want more debugging
162log() {
163    [ ${o_verbose} -gt 0 ] && echo "*** $* ***"
164    [ ${o_verbose} -gt 1 ] && read -p "=== Press enter to continue" foo
165}
166
167# chkerr:
168#
169# Routine to check for error
170#
171#	checks error code and drops into shell on failure.
172#	if shell exits, terminates script as well as /etc/rc.
173#	if remount_optional exists under the mountpoint, skip this check.
174#
175chkerr() {
176    lastitem () ( n=$(($# - 1)) ; shift $n ; echo $1 )
177    mountpoint="$(lastitem $2)"
178    [ -r $mountpoint/remount_optional ] && ( echo "$2 failed: ignoring due to remount_optional" ; return )
179    case $1 in
180    0)
181	;;
182    *)
183	echo "$2 failed: dropping into /bin/sh"
184	/bin/sh
185	# RESUME
186	;;
187    esac
188}
189
190# The list of filesystems to umount after the copy
191to_umount=""
192
193handle_remount() { # $1 = mount point
194    local nfspt mountopts b
195    b=$1
196    log handle_remount $1
197    [ -d $b -a -f $b/diskless_remount ] || return
198    read nfspt mountopts < $b/diskless_remount
199    log "nfspt ${nfspt} mountopts ${mountopts}"
200    # prepend the nfs root if not present
201    [ `expr "$nfspt" : '\(.\)'` = "/" ] && nfspt="${nfsroot}${nfspt}"
202    mount_nfs $mountopts $nfspt $b
203    chkerr $? "mount_nfs $nfspt $b"
204    to_umount="$b ${to_umount}"
205}
206
207# Create a generic memory disk.
208# The 'auto' parameter will attempt to use tmpfs(5), falls back to md(4).
209# $1 is size in 512-byte sectors, $2 is the mount point.
210mount_md() {
211    if [ ${o_verbose} -gt 0 ] ; then
212        /sbin/mdmfs -XL -s $1 auto $2
213    else
214        /sbin/mdmfs -s $1 auto $2
215    fi
216}
217
218# Create the memory filesystem if it has not already been created
219#
220create_md() {
221	[ "x`eval echo \\$md_created_$1`" = "x" ] || return # only once
222	if [ "x`eval echo \\$md_size_$1`" = "x" ]; then
223	    md_size=10240
224	else
225	    md_size=`eval echo \\$md_size_$1`
226	fi
227	log create_md $1 with size $md_size
228	mount_md $md_size /$1
229	/bin/chmod 755 /$1
230	eval md_created_$1=created
231}
232
233# DEBUGGING
234#
235# set -v
236
237# Figure out our interface and IP.
238#
239bootp_ifc=""
240bootp_ipa=""
241bootp_ipbca=""
242class=""
243if [ ${dlv:=0} -ne 0 ] ; then
244	iflist=`ifconfig -l`
245	for i in ${iflist} ; do
246	    set -- `ifconfig ${i}`
247	    while [ $# -ge 1 ] ; do
248		if [ "${bootp_ifc}" = "" -a "$1" = "inet" ] ; then
249		    bootp_ifc=${i} ; bootp_ipa=${2} ; shift
250		fi
251		if [ "${bootp_ipbca}" = "" -a "$1" = "broadcast" ] ; then
252		    bootp_ipbca=$2; shift
253		fi
254		shift
255	    done
256	    if [ "${bootp_ifc}" != "" ] ; then
257		break
258	    fi
259	done
260	# Get the values passed with the T134 bootp cookie.
261	class="`/sbin/sysctl -qn kern.bootp_cookie`"
262
263	echo "Interface ${bootp_ifc} IP-Address ${bootp_ipa} Broadcast ${bootp_ipbca} ${class}"
264fi
265
266log Figure out our NFS root path
267#
268set -- `mount -t nfs`
269while [ $# -ge 1 ] ; do
270    if [ "$2" = "on" -a "$3" = "/" ]; then
271	nfsroot="$1"
272	break
273    fi
274    shift
275done
276
277# The list of directories with template files
278templates="base default"
279if [ -n "${bootp_ipbca}" ]; then
280	templates="${templates} ${bootp_ipbca} bcast/${bootp_ipbca}"
281fi
282if [ -n "${class}" ]; then
283	templates="${templates} ${class}"
284fi
285if [ -n "${bootp_ipa}" ]; then
286	templates="${templates} ${bootp_ipa} ip/${bootp_ipa}"
287fi
288
289# If /conf/diskless_remount exists, remount all of /conf.
290handle_remount /conf
291
292# Resolve templates in /conf/base, /conf/default, /conf/${bootp_ipbca},
293# and /conf/${bootp_ipa}.  For each subdirectory found within these
294# directories:
295#
296# - calculate memory filesystem sizes.  If the subdirectory (prior to
297#   NFS remounting) contains the file 'md_size', the contents specified
298#   in 512 byte sectors will be used to size the memory filesystem.  Otherwise
299#   8192 sectors (4MB) is used.
300#
301# - handle NFS remounts.  If the subdirectory contains the file
302#   diskless_remount, the contents of the file is NFS mounted over
303#   the directory.  For example /conf/base/etc/diskless_remount
304#   might contain 'myserver:/etc'.  NFS remounts allow you to avoid
305#   having to dup your system directories in /conf.  Your server must
306#   be sure to export those filesystems -alldirs, however.
307#   If the diskless_remount file contains a string beginning with a
308#   '/' it is assumed that the local nfsroot should be prepended to
309#   it before attemping to the remount.  This allows the root to be
310#   relocated without needing to change the remount files.
311#
312log "templates are ${templates}"
313for i in ${templates} ; do
314    for j in /conf/$i/* ; do
315	[ -d $j ] || continue
316
317	# memory filesystem size specification
318	subdir=${j##*/}
319	[ -f $j/md_size ] && eval md_size_$subdir=`cat $j/md_size`
320
321	# remount. Beware, the command is in the file itself!
322	if [ -f $j/remount ]; then
323	    if [ -f $j/remount_subdir ]; then
324		k="/conf.tmp/$i/$subdir"
325		[ -d $k ] || continue
326
327		# Mount the filesystem root where the config data is
328		# on the temporary mount point.
329		nfspt=`/bin/cat $j/remount`
330		$nfspt $k
331		chkerr $? "$nfspt $k"
332
333		# Now use a nullfs mount to get the data where we
334		# really want to see it.
335		remount_subdir=`/bin/cat $j/remount_subdir`
336		remount_subdir_cmd="mount -t nullfs $k/$remount_subdir"
337
338		$remount_subdir_cmd $j
339		chkerr $? "$remount_subdir_cmd $j"
340
341		# XXX check order -- we must force $k to be unmounted
342		# after j, as j depends on k.
343		to_umount="$j $k ${to_umount}"
344	    else
345		nfspt=`/bin/cat $j/remount`
346		$nfspt $j
347		chkerr $? "$nfspt $j"
348		to_umount="$j ${to_umount}" # XXX hope it is really a mount!
349	    fi
350	fi
351
352	# NFS remount
353	handle_remount $j
354    done
355done
356
357# - Create all required MFS filesystems and populate them from
358#   our templates.  Support both a direct template and a dir.cpio.gz
359#   archive. Support for auxiliary NVRAM. Support dir.remove files containing
360#   a list of relative paths to remove.
361#
362# The dir.cpio.gz form is there to make the copy process more efficient,
363# so if the cpio archive is present, it prevents the files from dir/
364# from being copied.
365
366for i in ${templates} ; do
367    for j in /conf/$i/* ; do
368	subdir=${j##*/}
369	if [ -d $j -a ! -f $j.cpio.gz  ]; then
370	    create_md $subdir
371	    cp -Rp $j/ /$subdir
372	fi
373    done
374    for j in /conf/$i/*.cpio.gz ; do
375	subdir=${j%*.cpio.gz}
376	subdir=${subdir##*/}
377	if [ -f $j ]; then
378	    create_md $subdir
379	    echo "Loading /$subdir from cpio archive $j"
380	    (cd / ; /rescue/tar -xpf $j)
381	fi
382    done
383    for j in /conf/$i/*/extract ; do
384	if [ -x $j ]; then
385	    subdir=${j%*/extract}
386	    subdir=${subdir##*/}
387	    create_md $subdir
388	    echo "Loading /$subdir using auxiliary command $j"
389	    $j /$subdir
390	fi
391    done
392    for j in /conf/$i/*.remove ; do
393	subdir=${j%*.remove}
394	subdir=${subdir##*/}
395	if [ -f $j ]; then
396	    # doubly sure it is a memory disk before rm -rf'ing
397	    create_md $subdir
398	    (cd /$subdir; rm -rf `/bin/cat $j`)
399	fi
400    done
401done
402
403# umount partitions used to fill the memory filesystems
404[ -n "${to_umount}" ] && umount $to_umount
405