newnode.in 11.7 KB
Newer Older
1 2 3 4
#!/usr/bin/perl -w

#
# EMULAB-COPYRIGHT
5
# Copyright (c) 2003, 2004 University of Utah and the Flux Group.
6 7 8 9 10 11 12
# All rights reserved.
#

#
# newnode - a script for moving nodes from the new_ tables into production.
#

13
use lib '@prefix@/lib';
14 15

use libdb;
Robert Ricci's avatar
Robert Ricci committed
16
use English;
17
use Getopt::Std;
18 19 20

use strict;

21
my $TB = "@prefix@";
22 23

my $switchmac = "$TB/sbin/switchmac";
24 25 26 27
my $os_load = "$TB/bin/os_load";
my $os_select = "$TB/bin/os_select";
my $newnode_reboot = "$TB/sbin/newnode_reboot";
my $named_setup = "$TB/sbin/named_setup";
28 29 30
my $nalloc = "$TB/bin/nalloc";
my $nfree = "$TB/bin/nfree";
my $dhcpd_makeconf = "$TB/sbin/dhcpd_makeconf";
31
my $exports_setup = "$TB/sbin/exports_setup";
32 33 34

my $dhcpd_rc = "/usr/local/etc/rc.d/2.dhcpd.sh";

35
my $sudo = "/usr/local/bin/sudo -S";
36 37

#
38
# MFS to boot the nodes into initially
39
#
40 41
my $MFS_INITIAL = TB_OSID_FREEBSD_MFS();
my $OS_OPSNODE  = "OPSNODE-BSD";
42 43

#
44
# Initial event system state to put the nodes into
45
#
46 47
my $STATE_INITIAL = TBDB_NODESTATE_SHUTDOWN;
my $STATE_OPSNODE = TBDB_NODESTATE_ISUP;
48 49 50 51 52 53

#
# Number of vnodes to create for each physical node
#
my $NUM_VNODES = 50;

54 55 56 57 58
#
# Default number of vnodes that can be hosted on any pnode
#
my $DEFAULT_PCVM_COUNT = 20;

59 60 61 62 63 64 65 66
#
# Experiments we might put nodes into
#
my $PID_HWDOWN = NODEDEAD_PID();
my $EID_HWDOWN = NODEDEAD_EID();
my $PID_OPSNODE = TBOPSPID();
my $EID_OPSNODE = "opsnodes";

67 68
# Turn off line buffering on output
$| = 1;
69

70 71 72 73
if (!TBAdmin()) {
    die "Sorry, only testbed administrators can run this script!\n";
}

74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
#
# Handle command-line options
#
my $force_unconnected_interfaces = 0;
my %options = ();
if (! getopts("f", \%options)) {
    usage();
}

if ($options{f}) {
    $force_unconnected_interfaces = 1;
}

sub usage() {
    die "Usage: $0 [-f] <node_id> ...\n";
}

91 92 93 94 95 96
#
# We seem to have to do this so that ssh gets proper root permissions to read
# the key file. Argh.
#
$UID = $EUID = 0;

97
if (@ARGV < 1) {
98
    usage();
99 100
}

101 102 103
#
# The user has to be able to run sudo, so they can restart dhcpd.
#
104
if (system "$sudo /bin/pwd < /dev/null") {
105 106 107
    die "You must be able to sudo to root to use this script\n";
}

108 109 110 111 112 113 114 115 116 117
#
# For vnodes - figure out the jail IP base
#
my $IPBASE;
if (TBDB_JAILIPBASE =~ /^(\d+).(\d+).(\d+).(\d+)/) {
    $IPBASE = "$1.$2";
} else {
    die "Problem with JAILIPBASE\n"
}

118 119 120 121 122
my @node_ids = @ARGV;

#
# Now, loop through the nodes given, and add each one
#
Robert Ricci's avatar
Robert Ricci committed
123
my (@succeeded_nodes, @reboot_IPs);
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139
NODE: foreach my $node_id (@node_ids) {
    my $query_result;

    #
    # Check to make sure said node does not already exist!
    #
    $query_result = DBQueryFatal("SELECT node_id FROM nodes WHERE " .
	"node_id='$node_id'");
    if ($query_result->num_rows()) {
	warn "Node $node_id failed: a node with that name already exists!\n";
	next NODE;
    }

    #
    # Grab information about the node from the new_nodes table
    #
140
    $query_result = DBQueryFatal("SELECT new_node_id, type, IP, temporary_IP, ".
141
	"floor, building, loc_x, loc_y, contact, phone, room, role " .
142
	"FROM new_nodes WHERE node_id='$node_id'");
143 144 145 146
    if (!$query_result->num_rows()) {
	warn "Node $node_id failed: No pending node with that name exists!\n";
	next NODE;
    }
147
    my ($new_node_id, $type, $IP, $tempIP, $floor, $building, $loc_x, $loc_y,
148
	$contact, $phone, $room, $role) = $query_result->fetchrow();
149

150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
    #
    # Make sure they've given it an IP address that looks valid
    #
    if ($IP !~ /^(\d+).(\d+).(\d+).(\d+)/) {
	warn "Node $node_id has an invalid IP address ($IP) - skipping\n";
	next NODE;
    }

    #
    # Make sure some other node doesn't already have this IP address on its
    # control interface
    #
    $query_result = DBQueryFatal("SELECT node_id FROM interfaces WHERE " .
	"IP='$IP' AND role='" . TBDB_IFACEROLE_CONTROL() . "'");
    if ($query_result->num_rows()) {
	my ($existing_node_id) = $query_result->fetchrow();
	warn "Node $node_id failed: $existing_node_id is already using IP " .
		"address $IP\n";
	next NODE;
    }

171 172 173 174
    #
    # Make sure that the new node is of a valid type, and grab a few other
    # things to fill in as initial values
    #
175
    $query_result = DBQueryFatal("SELECT control_iface FROM node_types " .
176 177 178 179 180
	"WHERE type='$type'");
    if (!$query_result->num_rows()) {
	warn "Node $node_id failed: Type $type does not exist!\n";
	next NODE;
    }
181
    my ($control_iface) = $query_result->fetchrow();
182 183 184 185

    #
    # Grab the node's MACs from the new_interfaces table
    #
186
    $query_result = DBQueryFatal("SELECT card, MAC, interface_type, " .
187
	"switch_id, switch_card, switch_port, cable, len " .
188
	"FROM new_interfaces WHERE new_node_id='$new_node_id'");
189 190 191 192 193 194
    if (!$query_result->num_rows()) {
	warn "Node $node_id failed: Must have at least one interface!\n";
	next NODE;
    }

    my %interfaces;
195
    while (my ($card, $MAC, $iface_type, $switch_id, $switch_card,
196
	    $switch_port, $cable, $len) = $query_result->fetchrow()) {
197 198 199
	#
	# Get some more information about this interface type
	#
200
	my $iface_query = DBQueryFatal("SELECT max_speed, full_duplex, " .
201
	    "connector FROM interface_types WHERE type='$iface_type'");
202
	if (!$iface_query->num_rows()) {
203
	    warn "Node $node_id failed: Interface $card is of unknown type " .
204
	    	"$iface_type\n";
205 206 207
	    next NODE;
	}

208
	my ($max_speed, $full_duplex, $connector) = $iface_query->fetchrow();
209

210 211 212 213
	#
	# Do a sanity check - make sure that we have a switch recorded for all
	# experimental interfaces
	#
214
	unless ($force_unconnected_interfaces || ($role eq "ctrlnode")) {
215
	    my $iface = "eth$card";
216 217
	    if (($iface ne $control_iface) && ($connector !~ /wireless/i)
		&& !$switch_id) {
218 219 220 221 222 223
		warn "Node $node_id failed: Don't know which switch " .
		    "card $iface is connected to\n";
		next NODE;
	    }
	}

224 225 226
	#
	# Stash it away...
	#
227
	$interfaces{$card} = [$MAC, $iface_type, $max_speed, $full_duplex,
228
		$switch_id, $switch_card, $switch_port, $cable, $len];
229 230 231 232 233 234

    }

    #
    # Make up a priority (just used for sorting)
    #
235 236
    $node_id =~ /^(.*\D)(\d+)$/;
    my ($prefix,$nodenum) = ($1, $2);
237
    my $priority;
238 239
    if (defined $nodenum) {
	$priority = $nodenum;
240 241 242 243
    } else {
	$priority = 1;
    }

244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
    #
    # Figure out which project and experiment this node needs to go into
    #
    my ($pid, $eid);
    if ($role eq "ctrlnode") {
	$pid = $PID_OPSNODE;
	$eid = $EID_OPSNODE;
    } else {
	$pid = $PID_HWDOWN;
	$eid = $EID_HWDOWN;
    }

    #
    # Figure out what OS this node should run
    #
    my $osid;
    if ($role eq "ctrlnode") {
	$osid = $OS_OPSNODE;
    } else {
	$osid = $MFS_INITIAL;
    }

    #
    # Figure out which opmode the OS runs in
    #
    my $result = DBQueryFatal("select op_mode from os_info where " .
	" osid='$osid'");
    if ($result->numrows() != 1) {
	die "Unable to find OS information for $osid\n";
    }
    my ($opmode) = $result->fetchrow();

    #
    # Figure out what state to start the node in
    #
    my $state;
    if ($role eq "ctrlnode") {
	$state = $STATE_OPSNODE;
    } else {
	$state = $STATE_INITIAL;
    }

286 287 288 289 290
    #
    # Okay, time to actually add the node!
    #
    
    DBQueryFatal("INSERT INTO nodes SET node_id='$node_id', type='$type', " .
291 292 293
	"phys_nodeid='$node_id', role='$role', priority=$priority, " .
    	"eventstate='$state', op_mode='$opmode', " .
	"def_boot_osid='$MFS_INITIAL', " .
294 295
	"state_timestamp=unix_timestamp(NOW()), " .
	"op_mode_timestamp=unix_timestamp(NOW())");
296

297 298
    DBQueryFatal("insert into node_hostkeys (node_id) ".
		 "values ('$node_id')");
299 300 301 302 303
    DBQueryFatal("insert into node_status ".
		 "(node_id, status, status_timestamp) ".
		 "values ('$node_id', 'down', now()) ");
    DBQueryFatal("insert into node_activity ".
		 "(node_id) values ('$node_id')");
304

305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
    #
    # Copy data into the location_info table, if they provided any
    #
    if ($floor || $building || $loc_x || $loc_y || $contact
	|| $phone || $room) {

	#
	# Handle columns that could have come to us NULL (and thus are undef)
	#
	if (defined $floor) {
	    $floor = "'$floor'";
	} else {
	    $floor = "NULL";
	}
	if (defined $building) {
	    $building = "'$building'";
	} else {
	    $building = "NULL";
	}
	if (defined $room) {
	    $room = "'$room'";
	} else {
	    $room = "NULL";
	}

	DBQueryFatal("INSERT INTO location_info SET node_id='$node_id', " .
	    "floor=$floor, building=$building, loc_x=$loc_x, loc_y=$loc_y, ".
	    "contact='$contact', phone='$phone', room=$room;");
    }

335 336 337
    #
    # Add some vnodes
    #
338
    if (defined $nodenum && ($role eq "testnode")) {
339 340 341 342
	my $vtype = $type;
	if (!($vtype =~ s/pc/pcvm/)) {
	    $vtype = "$vtype-vm";
	}
343 344 345 346
	DBQueryFatal("INSERT INTO node_auxtypes set node_id='$node_id', " .
	    "type='pcvm', count=$DEFAULT_PCVM_COUNT");
	DBQueryFatal("INSERT INTO node_auxtypes set node_id='$node_id', " .
	    "type='$vtype', count=$DEFAULT_PCVM_COUNT");
347 348
    }

349
    while (my ($card, $aref) = each %interfaces) {
350
	my ($MAC, $iface_type, $max_speed, $duplex, $switch_id, $switch_card,
351
	    $switch_port, $cable, $len) = @$aref;
352
	my $iface = "eth$card";
353 354
	my $iface_IP = "";
	my $wire_type = "Node";
355
	my $iface_role = TBDB_IFACEROLE_EXPERIMENT();
356
	if ($iface eq $control_iface) {
357 358
	    $iface_IP = $IP;
	    $wire_type = "Control";
359
	    $iface_role = TBDB_IFACEROLE_CONTROL();
360
	}
361 362
	else {
	    # Experimental interfaces start with no speed set. 
Kirk Webb's avatar
Kirk Webb committed
363
	    $max_speed = 0;
364
	}
365 366 367
	DBQueryFatal("INSERT INTO interfaces SET node_id='$node_id', " .
	    "card=$card, port=1, mac='$MAC', IP='$iface_IP', " .
	    "interface_type='$iface_type', iface='$iface', " .
Kirk Webb's avatar
Kirk Webb committed
368
	    "current_speed='$max_speed', duplex=$duplex, role='$iface_role'");
369

370 371 372 373
	if (!$switch_id) {
	    print "No switch found for ${node_id}:$iface - skipping\n";
	    next;
	}
374 375 376 377 378 379 380 381

	my $cable_len = "";
	if ($cable) {
	    $cable_len .= ", cable=$cable";
	}
	if ($len) {
	    $cable_len .= ", len=$len";
	}
382 383 384
	DBQueryFatal("INSERT INTO wires SET type='$wire_type', " .
	    "node_id1='$node_id', card1=$card, port1=1, " .
	    "node_id2='$switch_id', card2='$switch_card', " .
385
	    "port2='$switch_port' $cable_len");
386 387 388
    }

    #
389 390
    # Put it into hwdown for now - I would put them in reloading, but I'm
    # afriad the reload_daemon might do the wrong thing to them
391
    #
392
    system "$nalloc $pid $eid $node_id";
393 394 395 396

    #
    # Remove the node from the new_ tables
    #
397
    DBQueryFatal("DELETE FROM new_nodes WHERE new_node_id=$new_node_id");
398
    DBQueryFatal("DELETE FROM new_interfaces WHERE new_node_id=$new_node_id");
399

400
    print "$node_id succesfully added!\n";
401 402

    push @succeeded_nodes, $node_id;
Robert Ricci's avatar
Robert Ricci committed
403 404 405 406 407 408 409

    #
    # Only reboot regular nodes, not ops nodes
    #
    if ($role ne "ctrlnode") {
	push @reboot_IPs, $tempIP;
    }
410 411 412 413 414 415 416 417 418 419 420 421
}

#
# No point in restarting dhcpd, etc. if there are no nodes that succeeded
#
if (!@succeeded_nodes) {
    die "No nodes succeeded, exiting early\n";
}

#
# Re-generate dhcpd.conf
#
422
print "Re-generating dhcpd.conf\n";
423 424 425
if (system("$dhcpd_makeconf -i")) {
    die "Unable to open regenerate dhcpd config file\n";
}
426 427 428 429 430 431 432 433 434 435 436 437

print "Restarting dhcpd: $sudo $dhcpd_rc stop\n";
my $sudo_rv = system "$sudo $dhcpd_rc stop";
if ($sudo_rv) {
    die "Error stopping dhcpd - return value was $sudo_rv\n";
}
sleep 2;
print "Restarting dhcpd: $sudo $dhcpd_rc start\n";
$sudo_rv = system "$sudo $dhcpd_rc start";
if ($sudo_rv) {
    die "Error starting dhcpd - return value was $sudo_rv\n";
}
438

439 440 441 442 443
print "Setting up nameserver\n";
my $named_rv = system "$named_setup";
if ($named_rv) {
    die "Error running named_setup - return value was $named_rv\n";
}
444

445 446 447 448 449 450 451 452 453
#
# Before we boot nodes into the MFS, we have to make sure they can mount
# NFS filesystems
#
print "Running exports_setup\n";
my $exports_rv = system "$exports_setup";
if ($exports_rv) {
    warn "WARNING - exports_setup returned $exports_rv";
}
454

455 456 457 458
#
# Start rebooting nodes
#
print "Rebooting nodes...\n";
Robert Ricci's avatar
Robert Ricci committed
459
foreach my $IP (@reboot_IPs) {
460 461
    print "Rebooting $IP\n";
    system "$newnode_reboot $IP\n";
462 463
}

464 465 466 467 468
print "\n\n";
print "Finished - when you are satisifed that the nodes are working\n";
print "correctly, use nfree on boss to free them from the emulab-ops/hwdown\n";
print "experiment.\n";

469
#
470 471 472
# TODO -
#    disable interfaces
#    console setup
473
#