xref: /freebsd/libexec/rc/rc.initdiskless (revision 78ae60b447ebf420dd5cebfec30480866fd5cef4)
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
28# On entry to this script the entire system consists of a read-only root
29# mounted via NFS. The kernel has run BOOTP and configured an interface
30# (otherwise it would not have been able to mount the NFS root!)
31#
32# We use the contents of /conf to create and populate memory filesystems
33# that are mounted on top of this root to implement the writable
34# (and host-specific) parts of the root filesystem, and other volatile
35# filesystems.
36#
37# The hierarchy in /conf has the form /conf/T/M/ where M are directories
38# for which memory filesystems will be created and filled,
39# and T is one of the "template" directories below:
40#
41#  base		universal base, typically a replica of the original root;
42#  default	secondary universal base, typically overriding some
43#		of the files in the original root;
44#  ${ipba}	where ${ipba} is the assigned broadcast IP address
45#  bcast/${ipba} same as above
46#  ${class}	where ${class} is a list of directories supplied by
47#		bootp/dhcp through the T134 option.
48#		${ipba} and ${class} are typically used to configure features
49#		for group of diskless clients, or even individual features;
50#  ${ip}	where ${ip} is the machine's assigned IP address, typically
51#		used to set host-specific features;
52#  ip/${ip}	same as above
53#
54# Template directories are scanned in the order they are listed above,
55# with each successive directory overriding (merged into) the previous one;
56# non-existing directories are ignored.  The subdirectory forms exist to
57# help keep the top level /conf manageable in large installations.
58#
59# The existence of a directory /conf/T/M causes this script to create a
60# memory filesystem mounted as /M on the client.
61#
62# Some files in /conf have special meaning, namely:
63#
64# Filename	Action
65# ----------------------------------------------------------------
66# /conf/T/M/remount
67#		The contents of the file is a mount command. E.g. if
68# 		/conf/1.2.3.4/foo/remount contains "mount -o ro /dev/ad0s3",
69#		then /dev/ad0s3 will be mounted on /conf/1.2.3.4/foo/
70#
71# /conf/T/M/remount_optional
72#		If this file exists, then failure to execute the mount
73#		command contained in /conf/T/M/remount is non-fatal.
74#
75# /conf/T/M/remount_subdir
76#		If this file exists, then the behaviour of /conf/T/M/remount
77#		changes as follows:
78#		 1. /conf/T/M/remount is invoked to mount the root of the
79#		    filesystem where the configuration data exists on a
80#		    temporary mountpoint.
81#		 2. /conf/T/M/remount_subdir is then invoked to mount a
82#		    *subdirectory* of the filesystem mounted by
83#		    /conf/T/M/remount on /conf/T/M/.
84#
85# /conf/T/M/diskless_remount
86#		The contents of the file points to an NFS filesystem,
87#		possibly followed by mount_nfs options. If the server name
88#		is omitted, the script will prepend the root path used when
89#		booting. E.g. if you booted from foo.com:/path/to/root,
90#		an entry for /conf/base/etc/diskless_remount could be any of
91#			foo.com:/path/to/root/etc
92#			/etc -o ro
93#		Because mount_nfs understands ".." in paths, it is
94#		possible to mount from locations above the NFS root with
95#		paths such as "/../../etc".
96#
97# /conf/T/M/md_size
98#		The contents of the file specifies the size of the memory
99#		filesystem to be created, in 512 byte blocks.
100#		The default size is 10240 blocks (5MB). E.g. if
101#		/conf/base/etc/md_size contains "30000" then a 15MB MFS
102#		will be created. In case of multiple entries for the same
103#		directory M, the last one in the scanning order is used.
104#		NOTE: If you only need to create a memory filesystem but not
105#		initialize it from a template, it is preferable to specify
106#		it in fstab e.g. as  "md /tmp mfs -s=30m,rw 0 0"
107#
108# /conf/T/SUBDIR.cpio.gz
109#		The file is cpio'd into /SUBDIR (and a memory filesystem is
110#		created for /SUBDIR if necessary). The presence of this file
111#		prevents the copy from /conf/T/SUBDIR/
112#
113# /conf/T/M/extract
114#		This is alternative to SUBDIR.cpio.gz and remount.
115#		Similar to remount case, a memory filesystem is created
116#		for /M and initialized from a template but no mounting
117#		performed. Instead, this file is run passing /M as single
118#		argument. It is expected to extract template override to /M
119#		using auxiliary storage found in some embedded systems
120#		having NVRAM too small to hold mountable file system.
121#
122# /conf/T/SUBDIR.remove
123#		The list of paths contained in the file are rm -rf'd
124#		relative to /SUBDIR.
125#
126# /conf/diskless_remount
127#		Similar to /conf/T/M/diskless_remount above, but allows
128#		all of /conf to be remounted.  This can be used to allow
129#		multiple roots to share the same /conf.
130#
131#
132# You will almost universally want to create the following files under /conf
133#
134# File					Content
135# ----------------------------		----------------------------------
136# /conf/base/etc/md_size		size of /etc filesystem
137# /conf/base/etc/diskless_remount	"/etc"
138# /conf/default/etc/rc.conf		generic diskless config parameters
139# /conf/default/etc/fstab		generic diskless fstab e.g. like this
140#
141#	foo:/root_part			/	nfs	ro		0 0
142#	foo:/usr_part			/usr	nfs     ro		0 0
143#	foo:/home_part			/home   nfs     rw      	0 0
144#	md				/tmp	mfs     -s=30m,rw	0 0
145#	md				/var	mfs	-s=30m,rw	0 0
146#	proc				/proc	procfs	rw		0 0
147#
148# plus, possibly, overrides for password files etc.
149#
150# NOTE!  /var, /tmp, and /dev will be typically created elsewhere, e.g.
151# as entries in the fstab as above.
152# Those filesystems should not be specified in /conf.
153#
154# (end of documentation, now get to the real code)
155
156dlv=`/sbin/sysctl -n vfs.nfs.diskless_valid 2> /dev/null`
157
158# DEBUGGING
159# log something on stdout if verbose.
160o_verbose=0     # set to 1 or 2 if you want more debugging
161log() {
162    [ ${o_verbose} -gt 0 ] && echo "*** $* ***"
163    [ ${o_verbose} -gt 1 ] && read -p "=== Press enter to continue" foo
164}
165
166# chkerr:
167#
168# Routine to check for error
169#
170#	checks error code and drops into shell on failure.
171#	if shell exits, terminates script as well as /etc/rc.
172#	if remount_optional exists under the mountpoint, skip this check.
173#
174chkerr() {
175    lastitem () ( n=$(($# - 1)) ; shift $n ; echo $1 )
176    mountpoint="$(lastitem $2)"
177    [ -r $mountpoint/remount_optional ] && ( echo "$2 failed: ignoring due to remount_optional" ; return )
178    case $1 in
179    0)
180	;;
181    *)
182	echo "$2 failed: dropping into /bin/sh"
183	/bin/sh
184	# RESUME
185	;;
186    esac
187}
188
189# The list of filesystems to umount after the copy
190to_umount=""
191
192handle_remount() { # $1 = mount point
193    local nfspt mountopts b
194    b=$1
195    log handle_remount $1
196    [ -d $b -a -f $b/diskless_remount ] || return
197    read nfspt mountopts < $b/diskless_remount
198    log "nfspt ${nfspt} mountopts ${mountopts}"
199    # prepend the nfs root if not present
200    [ `expr "$nfspt" : '\(.\)'` = "/" ] && nfspt="${nfsroot}${nfspt}"
201    mount_nfs $mountopts $nfspt $b
202    chkerr $? "mount_nfs $nfspt $b"
203    to_umount="$b ${to_umount}"
204}
205
206# Create a generic memory disk.
207# The 'auto' parameter will attempt to use tmpfs(5), falls back to md(4).
208# $1 is size in 512-byte sectors, $2 is the mount point.
209mount_md() {
210    if [ ${o_verbose} -gt 0 ] ; then
211        /sbin/mdmfs -XL -s $1 auto $2
212    else
213        /sbin/mdmfs -s $1 auto $2
214    fi
215}
216
217# Create the memory filesystem if it has not already been created
218#
219create_md() {
220	[ "x`eval echo \\$md_created_$1`" = "x" ] || return # only once
221	if [ "x`eval echo \\$md_size_$1`" = "x" ]; then
222	    md_size=10240
223	else
224	    md_size=`eval echo \\$md_size_$1`
225	fi
226	log create_md $1 with size $md_size
227	mount_md $md_size /$1
228	/bin/chmod 755 /$1
229	eval md_created_$1=created
230}
231
232# DEBUGGING
233#
234# set -v
235
236# Figure out our interface and IP.
237#
238bootp_ifc=""
239bootp_ipa=""
240bootp_ipbca=""
241class=""
242if [ ${dlv:=0} -ne 0 ] ; then
243	iflist=`ifconfig -l`
244	for i in ${iflist} ; do
245	    set -- `ifconfig ${i}`
246	    while [ $# -ge 1 ] ; do
247		if [ "${bootp_ifc}" = "" -a "$1" = "inet" ] ; then
248		    bootp_ifc=${i} ; bootp_ipa=${2} ; shift
249		fi
250		if [ "${bootp_ipbca}" = "" -a "$1" = "broadcast" ] ; then
251		    bootp_ipbca=$2; shift
252		fi
253		shift
254	    done
255	    if [ "${bootp_ifc}" != "" ] ; then
256		break
257	    fi
258	done
259	# Get the values passed with the T134 bootp cookie.
260	class="`/sbin/sysctl -qn kern.bootp_cookie`"
261
262	echo "Interface ${bootp_ifc} IP-Address ${bootp_ipa} Broadcast ${bootp_ipbca} ${class}"
263fi
264
265log Figure out our NFS root path
266#
267set -- `mount -t nfs`
268while [ $# -ge 1 ] ; do
269    if [ "$2" = "on" -a "$3" = "/" ]; then
270	nfsroot="$1"
271	break
272    fi
273    shift
274done
275
276# The list of directories with template files
277templates="base default"
278if [ -n "${bootp_ipbca}" ]; then
279	templates="${templates} ${bootp_ipbca} bcast/${bootp_ipbca}"
280fi
281if [ -n "${class}" ]; then
282	templates="${templates} ${class}"
283fi
284if [ -n "${bootp_ipa}" ]; then
285	templates="${templates} ${bootp_ipa} ip/${bootp_ipa}"
286fi
287
288# If /conf/diskless_remount exists, remount all of /conf.
289handle_remount /conf
290
291# Resolve templates in /conf/base, /conf/default, /conf/${bootp_ipbca},
292# and /conf/${bootp_ipa}.  For each subdirectory found within these
293# directories:
294#
295# - calculate memory filesystem sizes.  If the subdirectory (prior to
296#   NFS remounting) contains the file 'md_size', the contents specified
297#   in 512 byte sectors will be used to size the memory filesystem.  Otherwise
298#   8192 sectors (4MB) is used.
299#
300# - handle NFS remounts.  If the subdirectory contains the file
301#   diskless_remount, the contents of the file is NFS mounted over
302#   the directory.  For example /conf/base/etc/diskless_remount
303#   might contain 'myserver:/etc'.  NFS remounts allow you to avoid
304#   having to dup your system directories in /conf.  Your server must
305#   be sure to export those filesystems -alldirs, however.
306#   If the diskless_remount file contains a string beginning with a
307#   '/' it is assumed that the local nfsroot should be prepended to
308#   it before attempting to the remount.  This allows the root to be
309#   relocated without needing to change the remount files.
310#
311log "templates are ${templates}"
312for i in ${templates} ; do
313    for j in /conf/$i/* ; do
314	[ -d $j ] || continue
315
316	# memory filesystem size specification
317	subdir=${j##*/}
318	[ -f $j/md_size ] && eval md_size_$subdir=`cat $j/md_size`
319
320	# remount. Beware, the command is in the file itself!
321	if [ -f $j/remount ]; then
322	    if [ -f $j/remount_subdir ]; then
323		k="/conf.tmp/$i/$subdir"
324		[ -d $k ] || continue
325
326		# Mount the filesystem root where the config data is
327		# on the temporary mount point.
328		nfspt=`/bin/cat $j/remount`
329		$nfspt $k
330		chkerr $? "$nfspt $k"
331
332		# Now use a nullfs mount to get the data where we
333		# really want to see it.
334		remount_subdir=`/bin/cat $j/remount_subdir`
335		remount_subdir_cmd="mount -t nullfs $k/$remount_subdir"
336
337		$remount_subdir_cmd $j
338		chkerr $? "$remount_subdir_cmd $j"
339
340		# XXX check order -- we must force $k to be unmounted
341		# after j, as j depends on k.
342		to_umount="$j $k ${to_umount}"
343	    else
344		nfspt=`/bin/cat $j/remount`
345		$nfspt $j
346		chkerr $? "$nfspt $j"
347		to_umount="$j ${to_umount}" # XXX hope it is really a mount!
348	    fi
349	fi
350
351	# NFS remount
352	handle_remount $j
353    done
354done
355
356# - Create all required MFS filesystems and populate them from
357#   our templates.  Support both a direct template and a dir.cpio.gz
358#   archive. Support for auxiliary NVRAM. Support dir.remove files containing
359#   a list of relative paths to remove.
360#
361# The dir.cpio.gz form is there to make the copy process more efficient,
362# so if the cpio archive is present, it prevents the files from dir/
363# from being copied.
364
365PATH=${PATH}:/rescue
366
367for i in ${templates} ; do
368    for j in /conf/$i/* ; do
369	subdir=${j##*/}
370	if [ -d $j -a ! -f $j.cpio.gz  ]; then
371	    create_md $subdir
372	    cp -Rp $j/ /$subdir
373	fi
374    done
375    for j in /conf/$i/*.cpio.gz ; do
376	subdir=${j%*.cpio.gz}
377	subdir=${subdir##*/}
378	if [ -f $j ]; then
379	    create_md $subdir
380	    echo "Loading /$subdir from cpio archive $j"
381	    (cd / ; tar -xpf $j)
382	fi
383    done
384    for j in /conf/$i/*/extract ; do
385	if [ -x $j ]; then
386	    subdir=${j%*/extract}
387	    subdir=${subdir##*/}
388	    create_md $subdir
389	    echo "Loading /$subdir using auxiliary command $j"
390	    $j /$subdir
391	fi
392    done
393    for j in /conf/$i/*.remove ; do
394	subdir=${j%*.remove}
395	subdir=${subdir##*/}
396	if [ -f $j ]; then
397	    # doubly sure it is a memory disk before rm -rf'ing
398	    create_md $subdir
399	    (cd /$subdir; rm -rf `/bin/cat $j`)
400	fi
401    done
402done
403
404# umount partitions used to fill the memory filesystems
405[ -n "${to_umount}" ] && umount $to_umount
406