mkjail.pl 24.5 KB
Newer Older
1
2
3
#!/usr/bin/perl -w
#
# EMULAB-COPYRIGHT
4
# Copyright (c) 2000-2003 University of Utah and the Flux Group.
5
6
# All rights reserved.
#
7
8
# Kernel, jail, netstat, route, ifconfig, ipfw, header files.
# 
9
10
11
12
13
use English;
use Getopt::Std;
use Fcntl;
use IO::Handle;
use Socket;
14
use Fcntl ':flock';
15

16
17
18
# Drag in path stuff so we can find emulab stuff. Also untaints path.
BEGIN { require "/etc/emulab/paths.pm"; import emulabpaths; }

19
use libsetup qw(JailedMounts REMOTE LOCALROOTFS TMPASSDB TMGROUPDB);
20

21
22
23
24
25
26
27
28
29
30
31
32
33
#
# Questions:
#
# * Whats the hostname for the jail? Perhaps vnodename.emulab.net
# * What should /etc/resolv.conf look like?
# 

#
# Create a jailed environment. There are some stub files stored in
# /etc/jail that copied into the jail.
#
sub usage()
{
34
35
    print("Usage: mkjail.pl [-s] [-i <ipaddr>] [-p <pid>] ".
	  "[-h <hostname>] <vnodeid>\n");
36
37
    exit(-1);
}
38
my  $optlist = "i:p:e:sh:";
39
40
41
42
43
44
45

#
# Only real root can run this script.
#
if ($UID) {
    die("Must be root to run this script!\n");
}
46
system("sysctl jail.set_hostname_allowed=0 >/dev/null 2>&1");
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79

#
# Catch ^C and exit with error. 
#
my $leaveme = 0;
sub handler ($) {
    my ($signame) = @_;
    
    $SIG{INT}  = 'IGNORE';
    $SIG{USR1} = 'IGNORE';
    $SIG{TERM} = 'IGNORE';
    $SIG{HUP}  = 'IGNORE';

    if ($signame eq 'USR1') {
	$leaveme = 1;
    }
    fatal("Caught a SIG${signame}! Killing the jail ...");
}
$SIG{INT}  = \&handler;
$SIG{USR1} = \&handler;
$SIG{HUP}  = \&handler;
$SIG{TERM} = 'IGNORE';

#
# Turn off line buffering on output
#
STDOUT->autoflush(1);
STDERR->autoflush(1);

#
# Locals
#
my $JAILPATH	= "/var/emulab/jails";
80
my $ETCJAIL     = "/etc/jail";
81
my $LOCALFS	= LOCALROOTFS();
82
83
84
my $LOCALMNTPNT = "/local";
my $TMCC	= "$BINDIR/tmcc";
my $JAILCONFIG  = "jailconfig";
85
my @ROOTCPDIRS	= ("etc", "root");
86
87
88
my @ROOTMKDIRS  = ("dev", "tmp", "var", "usr", "proc", "users", 
		   "bin", "sbin", "home", $LOCALMNTPNT);
my @ROOTMNTDIRS = ("bin", "sbin", "usr");
89
my @EMUVARDIRS	= ("logs", "db", "jails", "boot", "lock");
90
my $VNFILESECT  = 64 * ((1024 * 1024) / 512); # 64MB in 512b sectors.
91
my $MAXVNDEVS	= 100;
92
my $IP;
93
94
my $IPALIAS;
my $IPMASK;
95
my $PID;
96
my $jailhostname;
97
98
99
100
my $debug	= 1;
my $cleaning	= 0;
my $vndevice;
my @mntpoints   = ();
101
my $nfsmounts = 0;
102
103
104
my $jailpid;
my $tmccpid;
my $interactive = 0;
105
106
107

# This stuff is passed from tmcd, which we parse into a config string
# and an option set.
108
my %jailconfig  = ();
109
my $jailoptions = "";
110
my $sshdport    = 50000;	# Bogus default, good for testing.
111
my $routetabid;			# Default to main routing table.
112
my $jailflags   = 0;
113
114
115
my @jailips     = ();		# List of jail IPs (for routing table).
my $JAIL_DEVMEM = 0x01;		# We need to know if these options given.
my $JAIL_ROUTING= 0x02;
116
117
118
119
120
121
122
123
124
125
126
127

#
# Parse command arguments. Once we return from getopts, all that should be
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
if (@ARGV != 1) {
    usage();
}
128
my $vnodeid = $ARGV[0];
129
130
131
132

#
# Untaint the arguments.
#
133
134
135
if ($vnodeid =~ /^([-\w\/\.]+)$/) {
    $vnodeid = $1;
    $jailhostname = $1;
136
137
}
else {
138
    die("Tainted argument $vnodeid!\n");
139
140
141
142
143
144
}

if (defined($options{'s'})) {
    $interactive = 1;
}

145
146
147
148
149
150
151
152
153
154
155
156
157
158
#
# Get the parent IP.
# 
my $hostname = `hostname`;
my $hostip;

# Untaint and strip newline.
if ($hostname =~ /^([-\w\.]+)$/) {
    $hostname = $1;

    my (undef,undef,undef,undef,@ipaddrs) = gethostbyname($hostname);
    $hostip = inet_ntoa($ipaddrs[0]);
}

159
#
160
161
# If no IP, then it defaults to our hostname's IP, *if* none is provided
# by the jail configuration file. That check is later. 
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
# 
if (defined($options{'i'})) {
    $IP = $options{'i'};

    if ($IP =~ /^(\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3})$/) {
	$IP = $1;
    }
    else {
	die("Tainted argument $IP!\n");
    }
}

if (defined($options{'p'})) {
    $PID = $options{'p'};

    if ($PID =~ /^([-\@\w]+)$/) {
	$PID = $1;
    }
    else {
	die("Tainted argument $PID.");
    }
}

185
186
187
188
189
190
191
192
193
194
195
if (defined($options{'h'})) {
    $jailhostname = $options{'h'};

    if ($jailhostname =~ /^([-\w\.]+)$/) {
	$jailhostname = $1;
    }
    else {
	die("Tainted argument $jailhostname.");
    }
}

196
#
197
# In most cases, the $vnodeid directory will have been created by the caller,
198
199
# and a config file possibly dropped in.
# When debugging, we have to create it here. 
200
201
202
203
# 
chdir($JAILPATH) or
    die("Could not chdir to $JAILPATH: $!\n");

204
205
206
if (! -e $vnodeid) {
    mkdir($vnodeid, 0770) or
	fatal("Could not mkdir $vnodeid in $JAILPATH: $!");
207
}
208
else {
209
    getjailconfig("$JAILPATH/$vnodeid");
210
211
212
213
214
215
}

#
# See if special options supported, and if so setup args as directed.
#
setjailoptions();
216

217
218
219
220
221
222
223
224
225
226
#
# If no IP set in the jail config, then use hostip.
#
if (!defined($IP)) {
    $IP = $hostip;
}

print("Setting up jail for $vnodeid using $IP\n")
    if ($debug);

227
228
229
#
# Create the "disk";
#
230
if (-e "$vnodeid/root") {
231
232
233
    #
    # Try to pick up where we left off.
    # 
234
    restorerootfs("$JAILPATH/$vnodeid");
235
236
237
238
239
}
else {
    #
    # Create the root filesystem.
    #
240
    mkrootfs("$JAILPATH/$vnodeid");
241
242
}

243
244
#
# If the jail has its own IP, must insert the control network alias.
245
246
# We use a 255.255.255.255 netmask since there might be multiple
# virtual nodes from the same subnet on this node. 
247
248
#
if (defined($IPALIAS)) {
249
250
    mysystem("ifconfig `control_interface` alias $IPALIAS ".
	     "netmask 255.255.255.255");
251
252
253
254
255
256
257
258
259
}

#
# Set up jail interfaces. We pass it the rtabid as well, and the script
# does all the right stuff. 
#
mysystem("ifsetup -i -j $vnodeid ".
	 (($jailflags & $JAIL_ROUTING) ? "-r $routetabid" : ""));

260
261
262
263
#
# Start the tmcc proxy. This path will be valid in both the outer
# environment and in the jail!
#
264
startproxy("$JAILPATH/$vnodeid");
265
266
267
268
269
270
271
272
273
274
275
276
277
278

#
# Start the jail. We do it in a child so we can send a signal to the
# jailed process to force it to shutdown. The jail has to shut itself
# down.
#
$jailpid = fork();
if ($jailpid) {
    # We do not really care about the exit status of the jail.
    waitpid($jailpid, 0);
    undef($jailpid);
}
else {
    $SIG{TERM} = 'DEFAULT';
279
    $ENV{'TMCCVNODEID'} = $vnodeid;
280

281
282
    my $cmd = "jail $jailoptions $JAILPATH/$vnodeid/root $jailhostname $IP ".
	"/etc/jail/injail -v $vnodeid";
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
    if ($interactive) {
	$cmd .= " /bin/csh";
    }
    exec($cmd);
    die("*** $0:\n".
	"    exec failed to start the jail!\n");
}
	 
#
# Once we exit, cleanup the mess.
#
cleanup();
exit(0);

#
# Create a file for a vnode device, vnconfig it, newfs, and then
# mount it on the "root" directory.
#
sub mkrootfs($)
{
    my ($path) = @_;
304
    my $vnsize = $VNFILESECT;
305
306
307
308
309
310
311
312

    chdir($path) or
	fatal("Could not chdir to $path: $!");

    mkdir("root", 0770) or
	fatal("Could not mkdir 'root' in $path: $!");
    
    #
313
    # Lots of zeros. Note we oseek to create a big hole.
314
    # 
315
    mysystem("dd if=/dev/zero of=root.vnode bs=512 oseek=$vnsize count=1");
316
317
318
319
320

    #
    # Find a free vndevice.
    #
    for (my $i = 0; $i < $MAXVNDEVS; $i++) {
321
322
323
324
	# Make sure the dev entries exist!
	mysystem("(cd /dev; ./MAKEDEV vn${i})");
	
	system("vnconfig -c vn${i} root.vnode");
325
326
327
328
329
330
331
	if (! $?) {
	    $vndevice = $i;
	    last;
	}
    }
    fatal("Could not find a free vn device!") 
	if (!defined($vndevice));
332
333
    print("Using vn${vndevice}\n")
 	if ($debug);
334

335
    mysystem("vnconfig -s labels vn${vndevice} root.vnode");
336
337
    mysystem("disklabel -r -w vn${vndevice} auto");
    mysystem("newfs -b 8192 -f 1024 -i 4096 -c 15 /dev/vn${vndevice}c");
338
    mysystem("tunefs -m 2 -o space /dev/vn${vndevice}c");
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
    mysystem("mount /dev/vn${vndevice}c root");
    push(@mntpoints, "$path/root");

    #
    # Okay, copy in the top level directories. 
    #
    foreach my $dir (@ROOTCPDIRS) {
	mysystem("hier -f -FN cp /$dir root/$dir");
    }

    #
    # Make some other directories that are nice to have!
    #
    foreach my $dir (@ROOTMKDIRS) {
	mkdir("root/$dir", 0755) or
	    fatal("Could not mkdir '$dir' in $path/root: $!");
    }

    #
    # Okay, mount some other directories to save space.
    #
    foreach my $dir (@ROOTMNTDIRS) {
361
362
363
364
365
	if ($nfsmounts) {
	    mysystem("mount -r localhost:/$dir $path/root/$dir");
	} else {
	    mysystem("mount -r -t null /$dir $path/root/$dir");
	}
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
	push(@mntpoints, "$path/root/$dir");
    }

    #
    # The proc FS in the jail is per-jail of course.
    # 
    mysystem("mount -t procfs proc $path/root/proc");
    push(@mntpoints, "$path/root/proc");

    #
    # /tmp is special of course
    #
    mysystem("chmod 1777 root/tmp");

    #
    # /dev is also special. It gets a very restricted set of entries.
382
    # Note that we create some BPF devices since they work in our jails.
383
    #
384
385
386
387
388
389
390
391
    my $makedevs = "bpf31";
    if ($jailflags & $JAIL_DEVMEM) {
	$makedevs .= " std pty0";
    }
    else {
	$makedevs .= " jail";
    }
    mysystem("cd $path/root/dev; cp -p /dev/MAKEDEV .; ./MAKEDEV $makedevs");
392
393
394
395

    # Remove some extraneous devices.
    mysystem("cd $path/root/dev; rm -f pci io klog console; ".
	     "ln -sf null console");
396
397
398
399
400
401
402
403
404
405
    
    #
    # Create stub /var and create the necessary log files.
    #
    # NOTE: I stole this little diddy from /etc/rc.diskless2.
    #
    mysystem("mtree -nqdeU -f /etc/mtree/BSD.var.dist ".
	     "-p $path/root/var >/dev/null 2>&1");
    mysystem("mkdir -p $path/root/$path");

406
407
408
409
410
411
412
413
414
415
416
417
418
419
    #
    # Make the emulab directories since they are not in the mtree file.
    #
    if (! -e "$path/root/var/emulab") {
	mkdir("$path/root/var/emulab", 0755) or
	    fatal("Could not mkdir 'emulab' in $path/root/var: $!");
    }
    foreach my $dir (@EMUVARDIRS) {
	if (! -e "$path/root/var/emulab/$dir") {
	    mkdir("$path/root/var/emulab/$dir", 0755) or
		fatal("Could not mkdir 'dir' in $path/root/var/emulab: $!");
	}
    }

420
421
422
423
424
425
426
    #
    # Stash the control net IP if not the same as the host IP
    #
    if ($IP ne $hostip) {
	mysystem("echo $IP >> $path/root/var/emulab/boot/myip");
    }

427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
    #
    # Get a list of all the plain files and create zero length versions
    # in the new var.
    #
    opendir(DIR, "/var/log") or
	fatal("Cannot opendir /var/log: $!");
    my @logs = grep { -f "/var/log/$_" } readdir(DIR);
    closedir(DIR);

    foreach my $log (@logs) {
	mysystem("touch $path/root/var/log/$log");
    }

    #
    # Now a bunch of stuff to set up a nice environment in the jail.
    #
443
444
445
    mysystem("cp -p $ETCJAIL/rc.conf $path/root/etc");
    mysystem("rm -f $path/root/etc/rc.conf.local");
    mysystem("cp -p $ETCJAIL/rc.local $path/root/etc");
446
    mysystem("cp -p $ETCJAIL/crontab $path/root/etc");
447
    mysystem("cp /dev/null $path/root/etc/fstab");
448
    mysystem("ln -s /usr/compat $path/root/compat");
449
450
451
452
453
454

    # No X11 forwarding. 
    mysystem("cat $path/root/etc/ssh/sshd_config | ".
	     "sed -e 's/^X11Forwarding.*yes/X11Forwarding no/' > ".
	     "$path/root/tmp/sshd_foo");
    mysystem("cp -f $path/root/tmp/sshd_foo $path/root/etc/ssh/sshd_config");
455
456
457
458
459
460
461
462
463
464
    
    # Port/Address for sshd.
    if ($IP ne $hostip) {
	mysystem("echo 'ListenAddress $IP' >> ".
		 "$path/root/etc/ssh/sshd_config");
    }
    else {
	mysystem("echo 'sshd_flags=\"\$sshd_flags -p $sshdport\"' >> ".
		 "$path/root/etc/rc.conf");
    }
465
466
467
468
469
470

    # In the jail, 127.0.0.1 refers to the jail, but we want to use the
    # nameserver running *outside* the jail.
    mysystem("cat /etc/resolv.conf | ".
	     "sed -e 's/127\.0\.0\.1/$hostip/' > ".
	     "$path/root/etc/resolv.conf");
471

472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
    #
    # Password/group file stuff. If remote the jail picks them up. Locally
    # we start with current set of accounts on the physnode, since this is
    # more efficient (less tmcd work), and besides, the physnode starts with
    # exactly the same accounts.
    #
    if (REMOTE()) {
	mysystem("cp -p $ETCJAIL/group $path/root/etc");
	mysystem("cp -p $ETCJAIL/master.passwd $path/root/etc");
	mysystem("pwd_mkdb -p -d $path/root/etc $path/root/etc/master.passwd");
    }
    else {
	mysystem("cp -p /etc/group $path/root/etc");
	mysystem("cp -p /etc/master.passwd $path/root/etc");
	mysystem("pwd_mkdb -p -d $path/root/etc $path/root/etc/master.passwd");
	mysystem("cp -p " . TMPASSDB() . ".db $path/root/$DBDIR");
	mysystem("cp -p " . TMGROUPDB() . ".db $path/root/$DBDIR");
    }
    
491
492
    #
    # If the jail gets its own routing table, must arrange for it to
493
    # be populated with some extras when it boots up.
494
495
    # 
    if ($jailflags & $JAIL_ROUTING) {
496
	addroutestorc("$path/root/etc/rc.conf.routes");
497
    }
498

499
500
501
502
    #
    # Give the jail an NFS mount of the local project directory. This one
    # is read-write.
    #
503
504
    if (defined($PID) && -e $LOCALFS && -e "$LOCALFS/$PID") {
	mysystem("mkdir -p $path/root/$LOCALMNTPNT/$PID");
505
506
507
508
509
510
511
	if ($nfsmounts) {
	    mysystem("mount localhost:$LOCALFS/$PID ".
		     "      $path/root/$LOCALMNTPNT/$PID");
	} else {
	    mysystem("mount -t null $LOCALFS/$PID ".
		     "      $path/root/$LOCALMNTPNT/$PID");
	}
512
	push(@mntpoints, "$path/root/$LOCALMNTPNT/$PID");
513

514
	mysystem("ln -s $LOCALMNTPNT/$PID/$vnodeid $path/root/opt");
515
516
517
    }
    else {
	mkdir("$path/root/opt", 0775);
518
519
    }

520
521
522
523
524
525
    #
    # Ug. Until we have SFS working the way I want it, NFS mount the
    # usual directories inside the jail. This duplicates a lot of mounts,
    # but not sure what to do about that. 
    #
    if (! REMOTE()) {
526
	foreach my $dir ( JailedMounts($vnodeid, "$path/root", $nfsmounts) ) {
527
528
529
530
	    push(@mntpoints, "$path/root/$dir");
	}
    }

531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
    cleanmess($path);
    return 0;
}

#
# Restore a jail after a crash.
#
sub restorerootfs($)
{
    my ($path) = @_;

    chdir($path) or
	fatal("Could not chdir to $path: $!");

    #
    # Find a free vndevice.
    #
    for (my $i = 0; $i < $MAXVNDEVS; $i++) {
549
550
551
552
	# Make sure the dev entries exist!
	mysystem("(cd /dev; ./MAKEDEV vn${i})");
	
	system("vnconfig -c vn${i} root.vnode");
553
554
555
556
557
558
559
	if (! $?) {
	    $vndevice = $i;
	    last;
	}
    }
    fatal("Could not find a free vn device!") 
	if (!defined($vndevice));
560
561
    print("Using vn${vndevice}\n")
 	if ($debug);
562

563
    mysystem("vnconfig -s labels vn${vndevice} root.vnode");
564
565
566
567
568
569
570
571
    mysystem("fsck -y /dev/vn${vndevice}c");
    mysystem("mount /dev/vn${vndevice}c root");
    push(@mntpoints, "$path/root");

    #
    # Okay, mount some other directories to save space.
    #
    foreach my $dir (@ROOTMNTDIRS) {
572
573
574
575
576
	if ($nfsmounts) {
	    mysystem("mount -r localhost:/$dir $path/root/$dir");
	} else {
	    mysystem("mount -r -t null /$dir $path/root/$dir");
	}
577
578
579
580
581
582
583
584
585
	push(@mntpoints, "$path/root/$dir");
    }

    #
    # The proc FS in the jail is per-jail of course.
    # 
    mysystem("mount -t procfs proc $path/root/proc");
    push(@mntpoints, "$path/root/proc");

586
587
588
589
590
591
592
    #
    # Must rebuild the routes list cause of swapmodify.
    #
    if ($jailflags & $JAIL_ROUTING) {
	addroutestorc("$path/root/etc/rc.conf.routes");
    }

593
594
595
596
    #
    # Give the jail an NFS mount of the local project directory. This one
    # is read-write.
    #
597
598
    if (defined($PID) && -e $LOCALFS && -e "$LOCALFS/$PID") {
	mysystem("mkdir -p $path/root/$LOCALMNTPNT/$PID");
599
600
601
602
603
604
605
	if ($nfsmounts) {
	    mysystem("mount localhost:$LOCALFS/$PID ".
		     "      $path/root/$LOCALMNTPNT/$PID");
	} else {
	    mysystem("mount -t null $LOCALFS/$PID ".
		     "      $path/root/$LOCALMNTPNT/$PID");
	}
606
	push(@mntpoints, "$path/root/$LOCALMNTPNT/$PID");
607
    }
608
609
610
611
612
613
614

    #
    # Ug. Until we have SFS working the way I want it, NFS mount the
    # usual directories inside the jail. This duplicates a lot of mounts,
    # but not sure what to do about that. 
    #
    if (! REMOTE()) {
615
	foreach my $dir ( JailedMounts($vnodeid, "$path/root", $nfsmounts) ) {
616
617
618
	    push(@mntpoints, "$path/root/$dir");
	}
    }
619
620
621
622
    return 0;
}

#
623
624
# Okay, we clean up some of what is in /etc and /etc/emulab so that the
# jail cannot see that stuff. 
625
626
627
628
629
630
631
632
633
634
635
#
sub cleanmess($) {
    my ($path) = @_;

    #
    # And, some security stuff. We want to remove bits of stuff from the
    # jail that would enable it to talk to tmcd directly. 
    #
    mysystem("rm -f $path/root/etc/emulab.cdkey");
    mysystem("rm -f $path/root/etc/emulab.pkey");

636
    # This is /etc/emulab inside the jail.
637
638
639
640
    mysystem("rm -f  $path/root/$ETCDIR/*.pem");
    mysystem("rm -f  $path/root/$ETCDIR/cvsup.auth");
    mysystem("rm -rf $path/root/$ETCDIR/.cvsup");
    mysystem("rm -f  $path/root/$ETCDIR/master.passwd");
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681

    #
    # Copy in emulabman if it exists.
    #
    if (my (undef,undef,undef,undef,undef,undef,undef,$dir) =
	getpwnam("emulabman")) {
	mysystem("hier cp $dir $path/root/home/emulabman");
    }
}

#
# Start the tmcc proxy and insert the unix path into the environment
# for the jail to pick up.
#
sub startproxy($)
{
    my ($dir) = @_;
    my $log   = "$dir/tmcc.log";

    #
    # The point of these paths is so that there is a comman path to
    # the socket both outside and inside the jail. Yuck!
    #
    my $insidepath  = "$dir/tmcc";
    my $outsidepath = "$dir/root/$insidepath";

    $tmccpid = fork();
    if ($tmccpid) {
	#
	# So tmcc will work nicely inside the jail without needing the
	# -l option specified all over.
	#
	$ENV{'TMCCUNIXPATH'} = $insidepath;
	
	select(undef, undef, undef, 0.2);
	return 0;
    }
    $SIG{TERM} = 'DEFAULT';

    # The -o option will cause the proxy to detach but not fork!
    # Eventually change this to standard pid file kill.
682
    exec("$TMCC -d -x $outsidepath -n $vnodeid -o $log");
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
    die("Exec of $TMCC failed! $!\n");
}

#
# Cleanup at exit.
#
sub cleanup()
{
    if ($cleaning) {
	die("*** $0:\n".
	    "    Oops, already cleaning!\n");
    }
    $cleaning = 1;
    
    #
    # Note that the unmounts will fail unless all the processes inside
    # the jail exit!
    # 
    if (defined($tmccpid)) {
	kill('TERM', $tmccpid);
	waitpid($tmccpid, 0);
    }

    if (defined($jailpid)) {
707
	kill('USR1', $jailpid);
708
709
710
	waitpid($jailpid, 0);
    }

711
712
713
714
715
716
    # Clean up interfaces we might have setup
    system("ifsetup -u -j $vnodeid");

    # If the jail has its own IP, clean the alias.
    if (defined($IPALIAS)) {
	system("ifconfig `control_interface` -alias $IPALIAS");
717
718
    }

719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
    while (@mntpoints) {
	my $mntpoint = pop(@mntpoints);

	# If the umounts fail, we do want to continue. Dangerous!
	system("umount $mntpoint");
	if ($?) {
	    # Avoid recursive calls to cleanup; do not use fatal.
	    die("*** $0:\n".
		"    umount '$mntpoint' failed: $!\n");
	}
    }
    if (defined($vndevice)) {
	system("vnconfig -u vn${vndevice}");
    }
    if (!$leaveme) {
734
735
736
        #
        # Ug, with NFS mounts inside the jail, we need to be really careful.
        #
737
738
	if (-d "$JAILPATH/$vnodeid/root" &&
	    !rmdir("$JAILPATH/$vnodeid/root")) {
739
	    die("*** $0:\n".
740
		"    $JAILPATH/$vnodeid/root is not empty! This is bad!\n");
741
	}
742
	system("rm -rf $JAILPATH/$vnodeid");
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
    }
}

#
# Print error and exit.
#
sub fatal($)
{
    my ($msg) = @_;

    cleanup();
    die("*** $0:\n".
	"    $msg\n");
}

#
# Run a command string, redirecting output to a logfile.
#
sub mysystem($)
{
    my ($command) = @_;

    system($command);
    if ($?) {
	fatal("Command failed: $? - $command");
    }
}
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801

#
# Read in the jail config file. 
#
sub getjailconfig($)
{
    my ($path) = @_;

    $path .= "/$JAILCONFIG";

    if (! -e $path) {
	return 0;
    }
    
    if (! open(CONFIG, $path)) {
	print("$path could not be opened for reading: $!\n");
	return -1;
    }
    while (<CONFIG>) {
	if ($_ =~ /^(.*)="(.+)"$/ ||
	    $_ =~ /^(.*)=(.+)$/) {
	    $jailconfig{$1} = $2;
	}
    }
    close(CONFIG);
    return 0;
}

#
# See if special jail opts supported.
#
sub setjailoptions() {
802
803
    my $portrange;
    
804
805
806
807
808
809
810
    #
    # Do this all the time, so that we can figure out the sshd port.
    # 
    foreach my $key (keys(%jailconfig)) {
	my $val = $jailconfig{$key};

        SWITCH: for ($key) {
811
812
813
814
815
816
817
818
819
820
821
822
	    /^JAILIP$/ && do {
		if ($val =~ /([\d\.]+),([\d\.]+)/) {
		    # Set the jail IP, but do not override one on the 
		    # command line.
		    if (!defined($IP)) {
			$IP      = $1;
			$IPALIAS = $1;
			$IPMASK  = $2;
		    }
		}
		last SWITCH;
	    };
823
824
	    /^PORTRANGE$/ && do {
		if ($val =~ /(\d+),(\d+)/) {
825
		    $portrange = "$1:$2";
826
827
828
829
830
		}
		last SWITCH;
	    };
	    /^SSHDPORT$/ && do {
		if ($val =~ /(\d+)/) {
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
		    $sshdport     = $1;
		}
		last SWITCH;
	    };
	    /^SYSVIPC$/ && do {
		if ($val) {
		    $jailoptions .= " -o sysvipc";
		}
		else {
		    $jailoptions .= " -o nosysvipc";
		}
		last SWITCH;
	    };
	    /^INETRAW$/ && do {
		if ($val) {
		    $jailoptions .= " -o inetraw";
		}
		else {
		    $jailoptions .= " -o noinetraw";
		}
		last SWITCH;
	    };
	    /^BPFRO$/ && do {
		if ($val) {
		    $jailoptions .= " -o bpfro";
		}
		else {
		    $jailoptions .= " -o nobpfro";
		}
		last SWITCH;
	    };
862
863
864
865
866
867
868
869
870
871
872
	    /^INADDRANY$/ && do {
		if ($val) {
		    $jailoptions .= " -o inaddrany";
		}
		else {
		    $jailoptions .= " -o noinaddrany";
		}
		last SWITCH;
	    };
	    /^ROUTING$/ && do {
		if ($val) {
873
		    $jailoptions .= " -o routing -i 127.0.0.1 ";
874
875
876
877

		    $jailflags |= $JAIL_ROUTING;
		    #
		    # If the jail gets routing privs, then it must get
878
		    # its own routing table. 
879
880
881
882
883
884
885
886
887
888
		    #
		    $routetabid   = getnextrtabid();
		    $jailoptions .= " -r $routetabid";
		}
		last SWITCH;
	    };
	    /^DEVMEM$/ && do {
		$jailflags |= $JAIL_DEVMEM;
		last SWITCH;
	    };
Leigh B. Stoller's avatar
Leigh B. Stoller committed
889
 	    /^IPADDRS$/ && do {
890
891
892
893
894
895
		# Comma separated list of IPs
		my @iplist = split(",", $val);

		foreach my $ip (@iplist) {
		    if ($ip =~ /(\d+\.\d+\.\d+\.\d+)/) {
			$jailoptions .= " -i $1";
896
			push(@jailips, $1);
897
898
899
900
		    }
		}
 		last SWITCH;
 	    };
901
902
	}
    }
903
904
905
906
    #
    # If there is no IP for the jail, must restrict the port range of
    # it. Otherwise it has it own IP, and there is no need to.
    # 
907
908
909
910
911
912
913
914
915
    if (! defined($IPALIAS)) {
	if (defined($portrange)) {
	    $jailoptions .= " -p $portrange";
	}
	else {
	    fatal("Must have a port range if jail ip equals host ip!");
	}
	print("SSHD port is $sshdport\n");
    }
916
917
918

    system("sysctl jail.inetraw_allowed=1 >/dev/null 2>&1");
    system("sysctl jail.bpf_allowed=1 >/dev/null 2>&1");
919
920
921
922
    system("sysctl jail.inaddrany_allowed=1 >/dev/null 2>&1");
    system("sysctl jail.multiip_allowed=1 >/dev/null 2>&1");
    system("sysctl net.link.ether.inet.useloopback=0 >/dev/null 2>&1");

923
924
925
    if ($?) {
	print("Special jail options are NOT supported!\n");
	$jailoptions = "";
926
	$jailflags   = 0;
927
928
929
	return 0;
    }
    print("Special jail options are supported: '$jailoptions'\n");
930
931
932
933
934

    return 0;
}

#
935
936
# Append a list of static routes to the rc.conf file. This basically
# augments the list of static routes that tmcd tells us to install. 
937
938
939
940
941
942
#
sub addroutestorc($rc)
{
    my ($rc)   = @_;
    my $count  = 0;

943
    open(RC, ">$rc") or
944
945
	fatal("Could not open $rc to append static routes");

946
    my $routerip  = `cat $BOOTDIR/routerip`;
947
    chomp($routerip);
948
    my $hostip    = `cat $BOOTDIR/myip`;
949
950
    chomp($hostip);

951
952
953
954
    #
    # First the set of routes that all jails get. 
    # 
    print RC "static_routes=\"default lo0 host\"\n";
955
    print RC "route_default=\"default $routerip\"\n";
956
957
    print RC "route_lo0=\"localhost -interface lo0\"\n";
    print RC "route_host=\"$hostip localhost\"\n";
958
    if ($IP ne $hostip) {
959
960
961
962
963
964
965
966
967
968
	# Setup a route for all jails on this node, to the loopback.
	print RC "static_routes=\"\$static_routes jailnet\"\n";
	print RC "route_jailnet=\"-net $IP -interface lo0 255.255.255.0\"\n";

	# Need a route for the private network.
	my $ctrliface = `control_interface`;
	chomp($ctrliface);
	
	print RC "static_routes=\"\$static_routes privnet\"\n";
	print RC "route_privnet=\"-net $IP -interface $ctrliface $IPMASK\"\n";
969
    }
970
971
972

    #
    # Now a list of routes for each of the IPs the jail has access
973
974
975
976
    # to. The idea here is to override the interface route such that
    # traffic to the local interface goes through lo0 instead. This
    # avoids going through traffic shaping when, say, pinging your own
    # interface!
977
978
979
    # 
    foreach my $ip (@jailips) {
	print RC "static_routes=\"ip${count} \$static_routes\"\n";
980
	print RC "route_ip${count}=\"$ip -interface lo0\"\n";
981
982
983
	$count++;
    }
    close(RC);
984
985
    return 0;
}
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033

#
# Ug, with NFS mounts inside the jail, we need to be really careful.
#
sub removevnodedir($)
{
    my ($dir) = @_;

    if (-d "$dir/root" && !rmdir("$dir/root")) {
	die("*** $0:\n".
	    "    $dir/root is not empty! This is very bad!\n");
    }
    system("rm -rf $dir");
}

#
# Get a free routing table ID. This should eventually come from the
# kernel, but for now just use a file with a number in it. 
#
sub getnextrtabid()
{
    my $nextrtabid = 1;

    #
    # The chances of a race are low, but need to deal with it anyway.
    #
    my $lockfile  = "/var/emulab/lock/rtabid";
    my $rtabidfile= "/var/emulab/db/rtabid";

    open(LOCK, ">>$lockfile") ||
	fatal("Could not open $lockfile\n");
    
    while (flock(LOCK, LOCK_EX|LOCK_NB) == 0) {
	print "Waiting to get lock for $lockfile\n";
	sleep(1);
    }
    if (-e $rtabidfile) {
	my $rtabid = `cat $rtabidfile`;
	if ($rtabid =~ /^(\d*)$/) {
	    $nextrtabid = $1 + 1;
	}
	else {
	    close(LOCK);
	    fatal("Bad data in $rtabidfile!");
	}
    }
    system("echo $nextrtabid > $rtabidfile");
    close(LOCK);
1034
1035
1036
1037
1038
1039
    #
    # Flush that routing table so its pristine. Again, this should be
    # handled in the kernel.
    #
    system("route flush -rtabid $nextrtabid");
    
1040
1041
    return $nextrtabid;
}