node_reboot.in 9.08 KB
Newer Older
1 2 3 4 5
#!/usr/bin/perl -wT
use English;
use Getopt::Std;

#
6 7
# Reboot a node (or nodes). Will power cycle the node as a last resort.
# Use -e option to reboot all nodes in an experiment.
8
#
9
# usage: node_reboot [-d] [-f] node [node ...]
10
#        node_reboot [-d] [-f] -e pid,eid
11 12 13 14 15
#        Exit value is 0 if all nodes reboot okay, or the number of nodes
#        could not be rebooted.
#
sub usage()
{
16
    print STDOUT "Usage: node_reboot [-d] [-f] node [node ...]\n" .
17
	         "       node_reboot [-d] [-f] -e pid,eid\n".
18
	"Use the -d option to turn on debugging\n" .
19
	"Use the -e option to reboot all the nodes in an experiment\n" .
20
	"Use the -f option to shoot the node in the head\n";
21 22
    exit(-1);
}
23
my  $optlist = "dfe:";
24 25 26 27 28

#
# Configure variables
#
my $TB		= "@prefix@";
29 30

#
31
# Testbed Support libraries
32
#
33 34 35
use lib "@prefix@/lib";
use libdb;
use libtestbed;
36
use event;
37

Robert Ricci's avatar
Robert Ricci committed
38
my $ssh		= "$TB/bin/sshtb -n";
39
my $power	= "$TB/bin/power -e";
40 41 42 43 44
my $ipod	= "$TB/sbin/ipod";
my $ping	= "/sbin/ping";
my %pids	= ();
my @row;
my @nodes       = ();
45
my $debug       = 0;
46
my $force       = 0;
47
my $failed      = 0;
48 49 50
my $eidmode     = 0;
my $pid;
my $eid;
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74

# un-taint path
$ENV{'PATH'} = '/bin:/sbin:/usr/bin:/usr/local/bin';
delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

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

#
# We don't want to run this script unless its the real version.
#
if ($EUID != 0) {
    die("Must be root! Maybe its a development version?");
}

#
# Parse command arguments. Once we return from getopts, all that should
# left are the required arguments.
#
%options = ();
if (! getopts($optlist, \%options)) {
    usage();
}
if (defined($options{"d"})) {
75 76 77 78
    $debug = 1;
}
if (defined($options{"f"})) {
    $force = 1;
79
}
80 81 82 83 84 85 86 87 88
if (defined($options{"e"})) {
    if (@ARGV) {
	usage();
    }
    
    $eidmode = $options{"e"};
    if ($eidmode =~ /([-\w]*),([-\w]*)/) {
	$pid = $1;
	$eid = $2;
89 90
    }
    else {
91 92
	print STDOUT "Invalid argument to -e option: $eidmode\n";
	usage();
93
    }
94 95 96 97 98 99 100 101
}

#
# If eidmode, then get the node list out of the DB instead of the command
# line. A proper check is made later, so need to be fancy about the query.
#
if ($eidmode) {
    my @row;
Leigh B. Stoller's avatar
Leigh B. Stoller committed
102 103 104 105 106 107 108 109

    #
    # Verify permission to muck with this experiment.
    #
    if ($UID && !TBAdmin($UID) &&
	! TBExptAccessCheck($UID, $pid, $eid, TB_EXPT_MODIFY)) {
	die("*** You not have permission to reboot nodes in $pid/$eid!\n");
    }
110
    
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
    my $query_result =
	DBQueryFatal("select node_id from reserved where ".
		     "pid='$pid' and eid='$eid'");

    if ($query_result->numrows == 0) {
	print STDOUT "There are no nodes reserved in pid/eid $pid/$eid\n";
	usage();
    }
    while (@row = $query_result->fetchrow_array()) {
	push(@nodes, $row[0]);
    }
}
else {
    if (@ARGV == 0) {
	usage();
    }
    
    # Untaint the nodes.
    foreach my $node ( @ARGV ) {
	if ($node =~ /^([-\@\w]+)$/) {
	    $node = $1;
	}
	else {
	    die("Bad node name: $node.");
	}
    
	push(@nodes, $node);
    }
139

Leigh B. Stoller's avatar
Leigh B. Stoller committed
140 141 142 143 144 145 146
    #
    # Verify permission to reboot these nodes.
    #
    if ($UID && !TBAdmin($UID) &&
	! TBNodeAccessCheck($UID, TB_NODEACCESS_REBOOT, @nodes)) {
	die("You do not have permission to reboot one (or more) ".
	    "of the nodes!\n");
147 148 149
    }
}

150
#
151 152 153 154
# Another shark hack. Well, perhaps not. We really don't want 50 nodes
# all rebooting at the same time, PCs *or* sharks. Lets order them
# so that the shelves are grouped together at least, and issue the reboots
# in batches. 
155
#
156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212
my @sortednodes = sort(@nodes);

while (@sortednodes) {
    my @batch = ();
    my $i     = 0;
    my $lastshelf = 0;
    
    while ($i < 8 && @sortednodes > 0) {
	my $node = shift(@sortednodes);
	my $shelf;
	my $unit;

	#
	# The point of this sillyness is stop at each shelf transition.
	#
	if (IsShelved($node, \$shelf, \$unit)) {
	    if ($lastshelf && $lastshelf ne $shelf) {
		unshift(@sortednodes, $node);
		last;
	    }
	    $lastshelf = $shelf;
	}
	    
	push(@batch, $node);
	$i++;
    }

    if ($force) {
        #
        # In force mode, call the power program for the whole batch, and
	# continue on. We don't wait for them to go down or reboot.
        #
	system("$power cycle @batch");
	if ($?) {
	    exit ($? >> 8);
	}
    }
    else {
        #
        # Fire off a reboot process so that we can overlap them all.
        # We need the pid so we can wait for them all before preceeding.
        #
	foreach my $node ( @batch ) {
	    $mypid = RebootNode($node);
	    $pids{$node} = $mypid;
	}
    }

    # 
    # If there are more nodes to go, then lets pause a bit so that we
    # do not get a flood of machines coming up all at the same exact
    # moment.
    #
    if (@sortednodes) {
	print STDOUT "Pausing to give some nodes time to reboot ...\n";
	if ($lastshelf) {
	    sleep(15);
Mac Newbold's avatar
Mac Newbold committed
213
	} else {
214
	    sleep(10);
Mac Newbold's avatar
Mac Newbold committed
215
	}
216
    }
217 218
}

219
#
220
# In force mode, we are done.
221
#
222 223
if ($force) {
    exit 0;
224 225 226 227 228
}

#
# Wait for all the reboot children to exit before continuing.
#
229
foreach my $node ( sort(@nodes) ) {
230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260
    my $mypid     = $pids{$node};

    waitpid($mypid, 0);
    if ($?) {
	$failed++;
	print STDERR "Reboot of node $node failed!\n";
    }
    else {
	print STDOUT "$node rebooting ...\n";
    }
}

if ($debug && $failed) {
    print STDERR "$failed nodes could not be rebooted\n";
}
exit $failed;

#
# Reboot a node in a child process. Return the pid to the parent so
# that it can wait on all the children later.
# 
sub RebootNode {
    local($pc) = @_;
    local($status, $syspid, $mypid);

    print STDOUT "Rebooting $pc ...\n";

    $mypid = fork();
    if ($mypid) {
	return $mypid;
    }
261
    TBdbfork();
262 263 264 265 266 267 268

    #
    # See if the machine is pingable. If its not pingable, then we just
    # power cycle the machine rather than wait for ssh to time out.
    #
    # ping returns 0 if any packets make it through. 
    #
269
    if (! DoesPing($pc)) {
270 271 272 273 274 275 276 277 278 279 280
	print STDERR "$pc appears to be dead. Power cycling ...\n" if $debug;
	if (PowerCycle($pc)) {
	    exit(-1);
	}
	exit(0);
    }

    #
    # Machine is pingable at least. Try to reboot it gracefully,
    # or power cycle anyway if that does not work. 
    #
281
    print STDERR "Trying ssh reboot of $pc ...\n" if $debug;
282

283 284 285 286 287 288
    #
    # Must change our real UID to root so that ssh will work. We save the old
    # UID so that we can restore it after we finish the ssh
    #
    my $oldUID = $UID;
    print STDERR "Saved UID: $oldUID\n" if $debug;
289 290
    $UID = 0;
    
291 292 293 294 295 296
    #
    # Run an ssh command in a child process, protected by an alarm to
    # ensure that the ssh is not hung up forever if the machine is in
    # some funky state.
    # 
    $syspid = fork();
297

298 299
    if ($syspid) {
	local $SIG{ALRM} = sub { kill("TERM", $syspid); };
300
	alarm 20;
301 302 303 304 305 306 307
	waitpid($syspid, 0);
	alarm 0;

	#
	# The ssh can return non-zero exit status, but still have worked.
	# FreeBSD for example.
	#
308
	print STDERR "reboot of $pc returned $?.\n" if $debug;
309 310
    
	#
311 312
	# If either ssh is not running or it timed out,
	# send it a ping of death.
313
	# 
314 315 316 317 318 319 320 321
	if ($? == 256 || $? == 15) {
	    if ($? == 256) {
		print STDERR "$pc is not running sshd.\n" if $debug;
	    } else {
		print STDERR "$pc is wedged.\n" if $debug;
	    }
	    print STDERR "Trying Ping-of-Death on $pc ...\n" if $debug;

322 323 324 325 326 327 328 329
	    system("$ipod $pc");
	}
    }
    else {
	exec("$ssh $pc /sbin/reboot");
	exit(0);
    }

330 331 332 333 334 335 336
    #
    # Restore the old UID so that scripts run from this point on get the 
    # user's real UID
    #
    $UID = $oldUID;
    print STDERR "Restored UID: $UID\n" if $debug;

337 338 339 340 341 342
    #
    # Okay, before we power cycle lets really make sure. We wait a while
    # for it to stop responding to pings, and if it never goes silent,
    # punch the power button.
    #
    if (WaitTillDead($pc) == 0) {
343 344 345 346
	EventSendFatal(objtype   => "TBNODESTATE",
	    	       objname   => $pc,
		       eventtype => "REBOOTING",
		       host      => TB_BOSSNODE);
347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
	exit(0);
    }
	
    print STDERR "$pc is still running. Power cycling ...\n" if $debug;
    if (PowerCycle($pc)) {
	exit(-1);
    }
    exit(0);
}

#
# Power cycle a PC using the testbed power program.
#
sub PowerCycle {
    local($pc) = @_;

    system("$power cycle $pc");
    return $? >> 8;
}

#
# Wait until a machine stops returning ping packets.
# 
sub WaitTillDead {
    local($pc) = @_;
    local($status);

    print STDERR "Waiting for $pc to die off\n" if $debug;
    
    #
    # Sigh, a long ping results in the script waiting until all the
    # packets are sent from all the pings, before it will exit. So,
    # loop doing a bunch of shorter pings.
    #
381 382 383
    for ($i = 0; $i < 30; $i++) {
	if (! DoesPing($pc)) {
	    print STDERR "$pc is rebooting.\n" if $debug;
384 385 386 387 388 389 390
	    return 0;
	}
    }
    print STDERR "$pc is still alive.\n" if $debug;
    return 1;
}

391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415
#
# Returns 1 if host is responding to pings, 0 otherwise
#
sub DoesPing {
    local($pc) = @_;
    local($status);
    local($saveuid);

    $saveuid = $UID;
    $UID = 0;
    system("$ping -q -i 0.25 -c 8 -t 2 $pc >/dev/null 2>&1");
    $UID = $saveuid;
    $status = $? >> 8;

    #
    # Returns 0 if any packets are returned. Returns 2 if pingable
    # but no packets are returned. Other non-zero error codes indicate
    # other problems.  Any non-zero return indicates "not pingable" to us.
    # 
    print STDERR "$ping $pc returned $status\n" if $debug;
    if ($status) {
	return 0;
    }
    return 1;
}