libvnode_xen.pm 159 KB
Newer Older
Jon Rafkind's avatar
Jon Rafkind committed
1
#!/usr/bin/perl -wT
2
#
3
# Copyright (c) 2008-2017 University of Utah and the Flux Group.
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
# 
# {{{EMULAB-LICENSE
# 
# This file is part of the Emulab network testbed software.
# 
# This file is free software: you can redistribute it and/or modify it
# under the terms of the GNU Affero General Public License as published by
# the Free Software Foundation, either version 3 of the License, or (at
# your option) any later version.
# 
# This file is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
# FITNESS FOR A PARTICULAR PURPOSE.  See the GNU Affero General Public
# License for more details.
# 
# You should have received a copy of the GNU Affero General Public License
# along with this file.  If not, see <http://www.gnu.org/licenses/>.
# 
# }}}
23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39
#
# Implements the libvnode API for Xen support in Emulab.
#
# Note that there is no distinguished first or last call of this library
# in the current implementation.  Every vnode creation (through mkvnode.pl)
# will invoke all the root* and vnode* functions.  It is up to us to make
# sure that "one time" operations really are executed only once.
#
# TODO:
# + Clear out old, incorrect state in /var/lib/xend.
#   Maybe have to do this when tearing down (killing) vnodes.
#
# + Make more robust, little turds of state still get left around
#   that wreak havoc on reboot.
#
# + Support image loading.
#
Jon Rafkind's avatar
Jon Rafkind committed
40 41 42 43 44 45
package libvnode_xen;
use Exporter;
@ISA    = "Exporter";
@EXPORT = qw( init setDebug rootPreConfig
              rootPreConfigNetwork rootPostConfig
	      vnodeCreate vnodeDestroy vnodeState
46
	      vnodeBoot vnodePreBoot vnodeHalt vnodeReboot
47
	      vnodeUnmount
Jon Rafkind's avatar
Jon Rafkind committed
48 49
	      vnodePreConfig vnodePreConfigControlNetwork
              vnodePreConfigExpNetwork vnodeConfigResources
Leigh B Stoller's avatar
Leigh B Stoller committed
50
              vnodeConfigDevices vnodePostConfig vnodeExec vnodeTearDown VGNAME
Jon Rafkind's avatar
Jon Rafkind committed
51
	    );
52
use vars qw($VGNAME);
Jon Rafkind's avatar
Jon Rafkind committed
53 54 55 56 57 58 59 60

%ops = ( 'init' => \&init,
         'setDebug' => \&setDebug,
         'rootPreConfig' => \&rootPreConfig,
         'rootPreConfigNetwork' => \&rootPreConfigNetwork,
         'rootPostConfig' => \&rootPostConfig,
         'vnodeCreate' => \&vnodeCreate,
         'vnodeDestroy' => \&vnodeDestroy,
61
	 'vnodeTearDown' => \&vnodeTearDown,
Jon Rafkind's avatar
Jon Rafkind committed
62 63 64
         'vnodeState' => \&vnodeState,
         'vnodeBoot' => \&vnodeBoot,
         'vnodeHalt' => \&vnodeHalt,
65
# XXX needs to be implemented
66
         'vnodeUnmount' => \&vnodeUnmount,
Jon Rafkind's avatar
Jon Rafkind committed
67
         'vnodeReboot' => \&vnodeReboot,
68
# XXX needs to be implemented
69
         'vnodeExec' => \&vnodeExec,
Jon Rafkind's avatar
Jon Rafkind committed
70 71 72 73 74 75 76 77
         'vnodePreConfig' => \&vnodePreConfig,
         'vnodePreConfigControlNetwork' => \&vnodePreConfigControlNetwork,
         'vnodePreConfigExpNetwork' => \&vnodePreConfigExpNetwork,
         'vnodeConfigResources' => \&vnodeConfigResources,
         'vnodeConfigDevices' => \&vnodeConfigDevices,
         'vnodePostConfig' => \&vnodePostConfig,
       );

Jon Rafkind's avatar
Jon Rafkind committed
78

79 80 81 82 83 84 85
use strict;
use English;
use Data::Dumper;
use Socket;
use File::Basename;
use File::Path;
use File::Copy;
86
use File::Temp;
87
use POSIX qw(:signal_h);
88 89

# Pull in libvnode
90
BEGIN { require "/etc/emulab/paths.pm"; import emulabpaths; }
91 92
use libutil;
use libgenvnode;
93 94
use libvnode;
use libtestbed;
95
use libsetup;
96 97 98 99 100 101 102 103 104 105 106 107 108 109 110

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

#
# Load the OS independent support library. It will load the OS dependent
# library and initialize itself. 
# 

##
## Standard utilities and files section
##

111
my $BRCTL = "brctl";
112
my $IFCONFIG = "/sbin/ifconfig";
Leigh B Stoller's avatar
Leigh B Stoller committed
113
my $ETHTOOL = "/sbin/ethtool";
114 115 116 117 118
my $ROUTE = "/sbin/route";
my $SYSCTL = "/sbin/sysctl";
my $VLANCONFIG = "/sbin/vconfig";
my $MODPROBE = "/sbin/modprobe";
my $DHCPCONF_FILE = "/etc/dhcpd.conf";
119
my $NEW_DHCPCONF_FILE = "/etc/dhcp/dhcpd.conf";
120
my $RESTOREVM	= "$BINDIR/restorevm.pl";
Leigh B Stoller's avatar
Leigh B Stoller committed
121
my $LOCALIZEIMG	= "$BINDIR/localize_image";
122
my $IPTABLES	= "/sbin/iptables";
Leigh B Stoller's avatar
Leigh B Stoller committed
123 124 125 126
my $IPBIN	= "/sbin/ip";
my $NETSTAT     = "/bin/netstat";
my $IMAGEZIP    = "/usr/local/bin/imagezip";
my $IMAGEUNZIP  = "/usr/local/bin/imageunzip";
127
my $IMAGEDUMP   = "/usr/local/bin/imagedump";
128
my $XM          = "/usr/sbin/xm";
129
my $FSCK	= "/sbin/e2fsck";
130
my $FSCKUFS	= "/sbin/fsck.ufs";
131
my $debug  = 0;
132
my $lockdebug = 1;
133 134 135 136 137 138 139 140 141 142
my $sleepdebug = 0;

#
# Set to enable vnodesetup to exit before vnode is completely up
# (see vnodesetup::hackwaitandexit). Allows more parallelism during
# boot-time vnode setup. Note that concurrency may still be constrained
# by $MAXCONCURRENT (defined below) which limits how many new VMs can
# be created at once.
#
my $vsrelease = "immediate";	# or "early" or "none"
143

Mike Hibler's avatar
Mike Hibler committed
144 145 146 147 148 149 150
#
# Some commands/subsystems have evolved in incompatible ways over time,
# these vars keep track of such things.
#
my $newsfdisk = 0;
my $newlvm = 0;

151 152 153 154 155 156 157 158 159 160 161 162
#
# Image wait time.
#
# How long (seconds) we will wait to when trying to grab a lock on
# an image. Should be set to the max time you think it could take frisbee
# to download the largest (compressed) OS image you will support in a VM.
# Also consider that there could be multiple frisbees running at once for
# multiple images (currently limited by the vnode create lock concurrency
# ($MAXCONCURRENT) below.
#
my $MAXIMAGEWAIT = 1800;

163 164 165 166 167 168 169 170 171 172 173 174
#
# Serial console handling. We fire up a capture per active vnode.
# We use a fine assortment of capture options:
#
#	-i: standalone mode, don't try to contact capserver directly
#	-l: (added later) set directory where log, ACL, and pid files are kept.
#	-C: use a circular buffer to capture activity while no user
#	    is connected. This gets dumped to the user when they connect.
#	-X: (added later) run in "Xen mode" on the given domain.
#	    Monitors the pty exported by xenconsoled. Note that the
#	    specific pty can change when a domain reboots; capture
#	    deals with this.
Mike Hibler's avatar
Mike Hibler committed
175 176 177 178
#	-T: Put out a timestamp if there has been no previous output
#	    for at least 10 seconds.
#	-L: In conjunction with -T, the timestamp message includes how
#	    long it has been since the last output.
179 180 181 182 183
#	-R: Retry interval of 2 seconds. When capture is disconnected
#	    from the pty (due to domain reboot/shutdowns), this is how
#	    long we wait between attempts to reconnect.
#
my $CAPTURE     = "/usr/local/sbin/capture-nossl";
Mike Hibler's avatar
Mike Hibler committed
184
my $CAPTUREOPTS	= "-i -C -L -T 10 -R 2000";
185

Mike Hibler's avatar
Mike Hibler committed
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207
#
# Create a thin pool with the name $POOL_NAME using not more
# than $POOL_FRAC of any disk.
# 
my $usethin = 1;
my $POOL_NAME = "disk-pool";
my $POOL_FRAC = 0.75;

#
# If set to one, we will destroy a golden disk when no vnode disks
# are derived from it. Otherwise, we leave it around and it must be
# explicitly GCed by some yet-to-be-written daemon. 
#
my $REAP_GDS = 0;

#
# Flags for allocating LVs
#
sub ALLOC_NOPOOL()	{ return 0; }
sub ALLOC_INPOOL()	{ return 1; }
sub ALLOC_PREFERNOPOOL	{ return 2; }
sub ALLOC_PREFERINPOOL	{ return 3; }
208

209 210 211 212 213 214 215 216
##
## Randomly chosen convention section
##

# global lock
my $GLOBAL_CONF_LOCK = "xenconf";

# default image to load on logical disks
217 218
# Just symlink /boot/vmlinuz-xenU and /boot/initrd-xenU
# to the kernel and ramdisk you want to use by default.
219
my %defaultImage = (
220 221 222 223 224 225
    'name'      => "emulab-ops-emulab-ops-XEN-STD",
    'kernel'    => "/boot/vmlinuz-xenU",
    'ramdisk'   => "/boot/initrd-xenU",
    'OSVERSION' => "any",
    'PARTOS'    => "Linux",
    'ISPACKAGE' => 0,
226 227
    'PART'      => 2,
    'BOOTPART'  => 2,
228 229 230
);

# where all our config files go
231 232
my $VMS    = "/var/emulab/vms";
my $VMDIR  = "$VMS/vminfo";
233 234
my $XENDIR = "/var/xen";

235
# Extra space for capture/restore.
236
my $EXTRAFS = "/capture";
237

238
# Extra space for image metadata between reloads.
Leigh B Stoller's avatar
Leigh B Stoller committed
239
my $METAFS = "/metadata";
240 241
# So we can ask this from outside;
sub METAFS()  { return $METAFS; }
Leigh B Stoller's avatar
Leigh B Stoller committed
242

243 244 245
# Extra space for vminfo (/var/emulab/vms) between reloads.
my $INFOFS = "/vminfo";

246 247
# Xen LVM volume group name. Accessible outside this file.
$VGNAME = "xen-vg";
Leigh B Stoller's avatar
Leigh B Stoller committed
248 249
# So we can ask this from outside;
sub VGNAME()  { return $VGNAME; }
250 251 252 253 254 255 256 257

##
## Indefensible, arbitrary constant section
##

# Minimum memory for dom0
my $MIN_MB_DOM0MEM = 256;

Mike Hibler's avatar
Mike Hibler committed
258
#
259
# Minimum acceptible size (in GB) of LVM VG for domUs.
Mike Hibler's avatar
Mike Hibler committed
260 261 262 263 264 265
#
# XXX we used to calculate this in terms of anticipated maximum number
# of vnodes and minimum vnode images size, blah, blah. Now we just pick
# a value that allows us to use a pc3000 node with a single 144GB disk!
#
my $XEN_MIN_VGSIZE = 120;
266

267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
#
# When loading an Emulab partition image, we use a compressed version of our
# standard MBR layout:
#
# MBR 1 or 2 FreeBSD:
#    P1: 6GB (XEN_LDSIZE) offset at 63, OS goes here
#    P2: 1MB (XEN_EMPTYSIZE), as small as we can make it
#    P3: 1GB (XEN_SWAPSIZE), standard MBR2 swap size
# MBR 1 or 2 Linux:
#    P1: 1MB (XEN_EMPTYSIZE), as small as we can make it
#    P2: 6GB (XEN_LDSIZE) offset at 63, OS goes here
#    P3: 1GB (XEN_SWAPSIZE), standard MBR2 swap size
# MBR 3:
#    P1: 16GB (XEN_LDSIZE_3) offset at 2048, standard OS partition
#    P2: 1MB (XEN_EMPTYSIZE), as small as we can make it
#    P3: 1GB (XEN_SWAPSIZE), standard MBR2 swap size
#
284 285 286 287 288
# P4 is sized based on what the user told us. If they do not specify
# XEN_EXTRA, then we default to 1G (XEN_EXTRASIZE). We need enough
# space here to support uses of mkextrafs in the clientside (e.g., for
# "no nfs" experiments where local homedirs are created.
#
289 290 291 292 293 294
# Sizes below are in 1K blocks.
#
my $XEN_LDSIZE    =  6152895;
my $XEN_LDSIZE_3  = 16777216;
my $XEN_SWAPSIZE  =  1048576;
my $XEN_EMPTYSIZE =     1024;
295
my $XEN_EXTRASIZE =  1048576;
296

Leigh B Stoller's avatar
Leigh B Stoller committed
297 298 299 300 301 302 303 304 305 306 307
# IFBs
my $IFBDB      = "/var/emulab/db/ifbdb";
# Kernel auto-creates only two! Sheesh, why a fixed limit?
my $MAXIFB     = 1024;

# Route tables for tunnels
my $RTDB           = "/var/emulab/db/rtdb";
my $RTTABLES       = "/etc/iproute2/rt_tables";
# Temporary; later kernel version increases this.
my $MAXROUTETTABLE = 255;

308 309 310
# Striping
my $STRIPE_COUNT   = 1;

311 312 313
# Setup a RAID10 underneath the LVM
my $LVM_RAID = SHAREDHOST() ? 1 : 0;

314 315 316
# Avoid using SSDs unless there are only SSDs
my $LVM_AVOIDSSD = 1;

317 318 319
# Whether or not to use only unpartitioned (unused) disks to form the Xen VG.
my $LVM_FULLDISKONLY = 0;

320 321
# Whether or not to use partitions only when they are big.
my $LVM_ONLYLARGEPARTS = 1;
322
my $LVM_LARGEPARTPCT = 10;
323 324 325

# In general, you only want to use one partition per disk since we stripe.
my $LVM_ONEPARTPERDISK = 1;
326

327
# Use openvswitch for gre tunnels.
Mike Hibler's avatar
Mike Hibler committed
328
# Use a custom version if present, the standard version otherwise.
329 330
my $OVSCTL   = "/usr/local/bin/ovs-vsctl";
my $OVSSTART = "/usr/local/share/openvswitch/scripts/ovs-ctl";
Mike Hibler's avatar
Mike Hibler committed
331 332 333 334
if (! -x "$OVSCTL") {
    $OVSCTL   = "/usr/bin/ovs-vsctl";
    $OVSSTART = "/usr/share/openvswitch/scripts/ovs-ctl";
}
335

336 337
my $ISREMOTENODE = REMOTEDED();
my $BRIDGENAME   = "xenbr0";
338
my $VIFROUTING   = ((-e "$ETCDIR/xenvifrouting") ? 1 : 0);
339

340 341
my $TMCD_PORT	 = 7777;

342 343 344 345 346
#
# Information about the running Xen hypervisor
#
my %xeninfo = ();

347 348
# Local functions
sub findRoot();
349
sub createAuxDisk($$);
350 351 352 353
sub configFile($);
sub domain0Memory();
sub totalMemory();
sub hostIP($);
354
sub createDHCP();
355 356
sub addDHCP($$$$);
sub subDHCP($$);
357
sub restartDHCP();
358 359
sub formatDHCP($$$);
sub fixupMac($);
360
sub createControlNetworkScript($$$$);
361
sub createExpNetworkScript($$$$$$$$);
362
sub createTunnelScript($$$$$);
Leigh B Stoller's avatar
Leigh B Stoller committed
363
sub createExpBridges($$$);
364
sub destroyExpBridges($$);
365
sub domainStatus($;$);
366 367 368 369
sub domainExists($);
sub addConfig($$$);
sub createXenConfig($$);
sub readXenConfig($);
370
sub lookupXenConfig($$);
371
sub getXenInfo();
Leigh B Stoller's avatar
Leigh B Stoller committed
372 373 374 375 376
sub AllocateIFBs($$$);
sub InitializeRouteTable();
sub AllocateRouteTable($);
sub LookupRouteTable($);
sub FreeRouteTable($);
Mike Hibler's avatar
Mike Hibler committed
377
sub downloadOneImage($$$);
378
sub captureRunning($);
379
sub checkForInterrupt();
380 381 382

sub getXenInfo()
{
383 384
    open(XM,"$XM info|") 
        or die "getXenInfo: could not run '$XM info': $!";
385 386 387 388 389 390 391 392 393

    while (<XM>) {
	    chomp;
	    /^(\S+)\s*:\s+(.*)$/;
	    $xeninfo{$1} = $2;
    }
    
    close XM;
}
394

395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496
#
# Things that matter:
#
# - RAM in dom0.
#   Swapping is deadly. Looks like 1024MB is NOT enough based on experience
#   noted below. 4096MB is plenty and seems to override most of the other
#   concerns.
#
# - Number of CPUs.
#   Have not seen any appreciable difference with 32 CPUs vs. 4. Other
#   things cause problems well before this.
#
# - The number of disks in the VG.
#   LVM performance is generally unpredicable. More than one disk is
#   good, but haven't seen much improvement with, e.g., 6 instead of 2.
#   The killer is concurrent frisbees (write to LVM) and even more so,
#   imageunzips (read from and write to LVM).
#
# - The BW from the frisbee server.
#   Possibly an issue if nothing else stands in the way, due to subboss
#   disk speed that tops out at about 150MB/sec. Given random I/O and
#   multiple images, probably going to get less than 50MB/sec.
#
# Random proposal based on tests run on Emulab d710/d820/d430 nodes and
# Apt c6220 nodes:
#
# * Change the arbitrary 164MB write buf memory to an equally arbitrary,
#   but more aestetically pleasing, 128MB (where the hell did 164 come from?)
#
# * Adjust concurrency based on:
#
# if (dom0 physical RAM < 1GB) MAX = 1;
# if (any swap activity) MAX = 1;
#
#    This captures pc3000s/other old machines and overloaded (RAM) machines.
#
# if (# physical CPUs <= 2) MAX = 3;
# if (# physical spindles == 1) MAX = 3;
# if (dom0 physical RAM <= 2GB) MAX = 3;
#
#    This captures d710s, Apt r320, and Cloudlab m510s. We may need to
#    reconsider the latter since its single drive is an NVMe device.
#    But first we have to get Xen working with them (UEFI issues)...
#
# MAX = 5;
#
#    This captures Emulab d430/d820s, Apt c6220s, and probably all
#    Clemson and Wisconsin Cloudlab nodes.
#
# Random observations based on waaay too much time spent on d710s:
#
# Observation: d710 with 5 vnodes and all different images does not
# boot first time with MAXCONCURRENT==5. 4 vnodes appear to be downloading
# their disk image when the BSD domU tries to boot--qemu times out.
# Restarting the vnode later works fine. The reason for this is that dom0
# starts swapping due to imageunzip processes running. 164MB of write
# buffering per imageunzip is too much for 3 imageunzips and 1GB of dom0 mem.
# Even dropping to 128MB of write buffering is not enough. Single-threading
# imageunzip (-n -W 1) works fine, but things go really slow. We need a
# buffering based on the available dom0 RAM and the max number of concurrent
# imageunzips (MAXCONCURRENT) independent of how the latter is calculated.
#
# Observation: qemu processes blow up huge when they first start (500MB) but
# don't require that much afterward (20MB). So most of our d710 problems
# stem from qemu blasting off while imageunzips are running. This happens
# because qemu is outside of the MAXCONCURRENT lock.
#
# Empirically, based on a d710 with 1GB of dom0 RAM, we can pull off 3
# imageunzips + 1 qemu with just a tad of swapping--not enough to cause
# the qemu to timeout. Full imageunzips seem to be about 32MB + writebuf
# memory. No swapping until the qemu starts. The number of qemus we launch
# will be implicitly constrained by this limit as qemu startups (vnodeBoot)
# take less time than vnodeCreate so we should not have more than MAXCONCURRENT
# vnodes in vnodeBoot at once.
#

#
# Historic concurrency value. Should get overwritten in setConcurrency.
#
my $MAXCONCURRENT = 3;

#
# Number of concurrent containers set up in parallel. See the big, long
# navel-gazing comment just above...
#
sub setConcurrency($)
{
    my ($maxval) = @_;
   
    if ($maxval) {
	$MAXCONCURRENT = 5;
    } else {
	my ($ram,$cpus) = domain0Resources();
	my $disks = $STRIPE_COUNT;
	my $hasswapped = domain0Swapping();

	print STDERR "setConcurrency: cpus=$cpus, ram=$ram, disks=$disks hasswapped=$hasswapped\n"
	    if ($debug);

	if ($cpus > 0 && $disks > 0 && $ram > 0) {
	    if ($ram < 1024 || (!SHAREDHOST() && $hasswapped)) {
		$MAXCONCURRENT = 1;
497
	    } elsif ($cpus <= 2 || $disks <= 2 || $ram <= 2048) {
498 499 500 501 502 503 504 505 506
		$MAXCONCURRENT = 3;
	    } else {
		$MAXCONCURRENT = 5;
	    }
	}
    }
    print STDERR "Limiting to $MAXCONCURRENT concurrent vnode creations.\n";
}

507 508 509 510 511 512
sub init($)
{
    my ($pnode_id,) = @_;

    makeIfaceMaps();
    makeBridgeMaps();
513

Mike Hibler's avatar
Mike Hibler committed
514 515 516 517 518 519
    my $toolstack;
    if (-x "/usr/lib/xen-common/bin/xen-toolstack") {
	$toolstack = `/usr/lib/xen-common/bin/xen-toolstack`;
    } else {
	$toolstack = `grep TOOLSTACK /etc/default/xen`;
    }
520 521 522
    if ($toolstack =~ /xl$/) {
	$XM = "/usr/sbin/xl";
    }
523
    getXenInfo();
524

Mike Hibler's avatar
Mike Hibler committed
525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542
    # See which sfdisk we have. Version 2.26 removed some options we used.
    my $out = `sfdisk -v`;
    if (defined($out) && $out =~ /2\.(\d+)(\.\d+)$/) {
	if (int($1) >= 26) {
	    $newsfdisk = 1;
	}
    }

    # See what version of LVM we have. Again, some commands are different.
    $out = `lvm version | grep 'LVM version'`;
    if (defined($out) && $out =~ /LVM version:\s+(\d+)\.(\d+)\.(\d+)/) {
	if (int($1) > 2 ||
	    (int($1) == 2 && int($2) > 2) ||
	    (int($1) == 2 && int($2) == 2 && int($3) >= 99)) {
	    $newlvm = 1;
	}
    }

543 544 545
    # Compute the strip size for new lvms.
    if (-e "/var/run/xen.ready") {
	$STRIPE_COUNT = computeStripeSize($VGNAME);
546
    }
Jon Rafkind's avatar
Jon Rafkind committed
547 548 549
    return 0;
}

550 551 552 553
sub setDebug($)
{
    $debug = shift;
    libvnode::setDebug($debug);
554 555 556 557
    $lockdebug = 1;
    if ($debug > 1) {
	$sleepdebug = 1;
    }
558 559 560 561
    print "libvnode_xen: debug=$debug\n"
	if ($debug);
}

562 563 564 565 566 567 568
sub ImageLockName($)
{
    my ($imagename) = @_;

    return "xenimage." .
	(defined($imagename) ? $imagename : $defaultImage{'name'});
}
569 570 571 572 573 574
sub ImageLVName($)
{
    my ($imagename) = @_;

    return "image+" . $imagename;
}
575

576 577 578 579
#
# Called on each vnode, but should only be executed once per boot.
# We use a file in /var/run (cleared on reboots) to ensure this.
#
580
sub rootPreConfig($)
581
{
582
    my $bossip = shift;
583 584 585 586 587
    #
    # Haven't been called yet, grab the lock and double check that someone
    # didn't do it while we were waiting.
    #
    if (! -e "/var/run/xen.ready") {
Mike Hibler's avatar
Mike Hibler committed
588 589
	TBDebugTimeStamp("rootPreConfig: grabbing global lock $GLOBAL_CONF_LOCK")
	    if ($lockdebug);
590 591 592 593 594 595 596 597 598
	my $locked = TBScriptLock($GLOBAL_CONF_LOCK,
				  TBSCRIPTLOCK_GLOBALWAIT(), 900);
	if ($locked != TBSCRIPTLOCK_OKAY()) {
	    return 0
		if ($locked == TBSCRIPTLOCK_IGNORE());
	    print STDERR "Could not get the xeninit lock after a long time!\n";
	    return -1;
	}
    }
Mike Hibler's avatar
Mike Hibler committed
599 600
    TBDebugTimeStamp("  got global lock")
	if ($lockdebug);
601
    if (-e "/var/run/xen.ready") {
Mike Hibler's avatar
Mike Hibler committed
602 603
	TBDebugTimeStamp("  releasing global lock")
	    if ($lockdebug);
604 605 606 607 608 609
        TBScriptUnlock();
        return 0;
    }
    
    print "Configuring root vnode context\n";

610 611 612 613 614 615 616 617 618 619 620
    #
    # For compatibility with existing (physical host) Emulab images,
    # the physical host provides DHCP info for the vnodes. We manage
    # the dhcpd.conf file here. See below. 
    #
    # Note that we must first add an alias to the control net bridge so
    # that we (the physical host) are in the same subnet as the vnodes,
    # otherwise dhcpd will fail.
    #
    my ($alias_iface, $alias_ip, $alias_mask);

621
    #
622 623 624
    # Locally, we just need to add the alias to the control interface
    # (which might be a bridge).
    # 
625
    if (!$ISREMOTENODE) {
626
	my ($cnet_iface) = findControlNet();
627 628

	#
629 630 631
	# We use xen's antispoofing when constructing the guest control net
	# interfaces. This is most useful on a shared host, but no
	# harm in doing it all the time.
632
	#
633 634 635 636 637
	mysystem("$IPTABLES -P FORWARD DROP");
	mysystem("$IPTABLES -F FORWARD");
	# This says to forward traffic across the bridge.
	mysystem("$IPTABLES -A FORWARD ".
		 "-m physdev --physdev-in $cnet_iface -j ACCEPT");
638 639 640 641 642 643 644
	
	if ($VIFROUTING) {
	    mysystem("echo 1 >/proc/sys/net/ipv4/conf/$cnet_iface/proxy_arp");
	    mysystem("echo 1 >/proc/sys/net/ipv4/ip_forward");
	    # This is for arping -A to work. See emulab-cnet.pl
	    mysystem("echo 1 >/proc/sys/net/ipv4/ip_nonlocal_bind");
	}
645 646 647 648 649 650 651 652

	# Set up for metadata server for ec2 support
	print "Setting up redirection for meta server...\n";
	mysystem("$IPBIN addr add 169.254.169.254/32 ".
		 "   scope global dev $cnet_iface");
	mysystem("$IPTABLES -t nat -A PREROUTING -d 169.254.169.254/32 " .
		 "   -p tcp -m tcp --dport 80 -j DNAT ".
		 "   --to-destination ${bossip}:8787");
653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670
    }
    else {
	if (!existsBridge($BRIDGENAME)) {
	    if (mysystem2("$BRCTL addbr $BRIDGENAME")) {
		TBScriptUnlock();
		return -1;
	    }
	    #
	    # We do not set the mac address; we want it to take
	    # on the address of the attached vif interfaces so that
	    # arp works. This is quite kludgy of course, but otherwise
	    # the arp comes into the bridge interface and then kernel
	    # drops it. There is a brouter (ebtables) work around
	    # but not worth worrying about. 
	    #
	}
	(undef,$alias_mask,$alias_ip) = findVirtControlNet();
	$alias_iface = $BRIDGENAME;
671 672 673 674 675

	if (system("ifconfig $alias_iface | grep -q 'inet addr'")) {
	    print "Creating $alias_iface alias...\n";
	    mysystem("ifconfig $alias_iface $alias_ip netmask $alias_mask");
	}
676 677
    }

678 679 680 681
    # For tunnels
    mysystem("$MODPROBE openvswitch");
    mysystem("$OVSSTART --delete-bridges start");

682 683 684 685
    # For gre tunnels to work with iptables
    mysystem("$MODPROBE nf_conntrack_proto_gre");
    mysystem("$MODPROBE nf_conntrack_pptp");

686
    # For bandwidth contraints.
Leigh B Stoller's avatar
Leigh B Stoller committed
687 688 689 690 691 692 693 694 695 696 697 698 699 700 701
    mysystem("$MODPROBE ifb numifbs=$MAXIFB");

    # Create a DB to manage them. 
    my %MDB;
    if (!dbmopen(%MDB, $IFBDB, 0660)) {
	print STDERR "*** Could not create $IFBDB\n";
	TBScriptUnlock();
	return -1;
    }
    for (my $i = 0; $i < $MAXIFB; $i++) {
	$MDB{"$i"} = ""
	    if (!defined($MDB{"$i"}));
    }
    dbmclose(%MDB);
    
702 703 704 705 706 707 708 709 710 711 712 713 714 715
    #
    # Ensure that LVM is loaded in the kernel and ready.
    #
    print "Enabling LVM...\n"
	if ($debug);

    # We assume our kernels support this.
    mysystem2("$MODPROBE dm-snapshot");
    if ($?) {
	print STDERR "ERROR: could not load snaphot module!\n";
	TBScriptUnlock();
	return -1;
    }

716
    #
717
    # Make sure pieces are at least 5 GiB.
718
    #
719
    my $minpsize = 5 * 1024;
720
    my %devs = libvnode::findSpareDisks($minpsize, $LVM_AVOIDSSD);
721 722 723

    # if ignoring SSDs but came up with nothing, we have to use them!
    if ($LVM_AVOIDSSD && keys(%devs) == 0) {
724
	%devs = libvnode::findSpareDisks($minpsize, 0);
725
    }
726

Leigh B Stoller's avatar
Leigh B Stoller committed
727 728
    #
    # Turn on write caching. Hacky. 
729 730
    # XXX note we do not use the returned "path" here as we need to
    # change the setting on all devices, not just the whole disk devices.
Leigh B Stoller's avatar
Leigh B Stoller committed
731
    #
732
    my %diddev = ();
Leigh B Stoller's avatar
Leigh B Stoller committed
733
    foreach my $dev (keys(%devs)) {
734
	# only mess with the disks we are going to use
735 736
	if (!exists($diddev{$dev}) &&
	    (exists($devs{$dev}{"size"}) || $LVM_FULLDISKONLY == 0)) {
737
	    mysystem2("hdparm -W1 /dev/$dev");
738
	    $diddev{$dev} = 1;
739
	}
Leigh B Stoller's avatar
Leigh B Stoller committed
740
    }
741
    undef %diddev;
Leigh B Stoller's avatar
Leigh B Stoller committed
742

743 744 745 746 747 748 749 750
    #
    # See if our LVM volume group for VMs exists and create it if not.
    #
    my $vg = `vgs | grep $VGNAME`;
    if ($vg !~ /^\s+${VGNAME}\s/) {
	print "Creating volume group...\n"
	    if ($debug);

751
	#
752 753
	# Total up potential maximum size.
	# Also determine mix of SSDs and non-SSDs if required.
754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769
	#
	my $maxtotalSize = 0;
	my $sizeThreshold = 0;
	foreach my $dev (keys(%devs)) {
	    if (defined($devs{$dev}{"size"})) {
		$maxtotalSize += $devs{$dev}{"size"};
	    } else {
		foreach my $part (keys(%{$devs{$dev}})) {
		    $maxtotalSize += $devs{$dev}{$part}{"size"};
		}
	    }
	}
	if ($maxtotalSize > 0) {
	    $sizeThreshold = int($maxtotalSize * $LVM_LARGEPARTPCT / 100.0);
	}

770 771 772 773
	#
	# Find available devices of sufficient size, prepare them,
	# and incorporate them into a volume group.
	#
774
      again:
775
	my $totalSize = 0;
776
	my $minSize;
777
	my @blockdevs = ();
778
	foreach my $dev (sort keys(%devs)) {
779
	    #
780
	    # Whole disk is available, use it.
781
	    #
782
	    if (defined($devs{$dev}{"size"})) {
783 784
		my $psize = $devs{$dev}{"size"};

785
		push(@blockdevs, $devs{$dev}{"path"});
786 787 788 789
		$totalSize += $psize;
		if (!defined($minSize) || $psize < $minSize) {
		    $minSize = $psize;
		}
790
		next;
791
	    }
792

793
	    #
794
	    # Disk contains partitions that are available.
795
	    #
796 797 798 799 800 801 802 803 804 805 806 807
	    my ($lpsize,$lppath);
	    foreach my $part (keys(%{$devs{$dev}})) {
		my $psize = $devs{$dev}{$part}{"size"};
		my $ppath = $devs{$dev}{$part}{"path"};

		#
		# XXX one way to avoid using the system disk, just ignore
		# all partition devices. However, in cases where the
		# remainder of the system disk represents the majority of
		# the available space (e.g., Utah d710s), this is a bad
		# idea.
		#
808
		if ($LVM_FULLDISKONLY || $LVM_RAID) {
809 810 811 812
		    print STDERR
			"WARNING: not using partition $ppath for LVM\n";
		    next;
		}
813

814 815 816 817
		#
		# XXX Another heurstic to try to weed out the system
		# disk whenever feasible: if a partition device represents
		# less than some percentage of the max possible space,
818
		# avoid it. At Utah this one is tuned (10%) to avoid using
819
		# left over space on the system disk of d820s (which have
820 821
		# six other larger drives) or d430s (which have two large
		# disks) while using it on the pc3000s and d710s.
822 823 824 825 826
		#
		if ($LVM_ONLYLARGEPARTS && $psize < $sizeThreshold) {
		    print STDERR "WARNING: not using $ppath for LVM (too small)\n";
		    next;
		}
827

828 829 830 831 832 833 834 835 836 837 838 839
		#
		# XXX If we are only going to use one partition per disk,
		# record the largest one we find here. This check will
		# filter out the small "other OS" partition (3-6GB) in
		# favor of the larger "rest of the disk" partition.
		#
		if ($LVM_ONEPARTPERDISK) {
		    if (!defined($lppath) || $psize > $lpsize) {
			$lppath = $ppath;
			$lpsize = $psize;
		    }
		    next;
840
		}
841 842 843 844 845 846

		#
		# It ran the gauntlet of feeble filters, use it!
		#
		push(@blockdevs, $ppath);
		$totalSize += $psize;
847 848 849
		if (!defined($minSize) || $psize < $minSize) {
		    $minSize = $psize;
		}
850 851 852 853
	    }
	    if ($LVM_ONEPARTPERDISK && defined($lppath)) {
		push(@blockdevs, $lppath);
		$totalSize += $lpsize;
854 855 856
		if (!defined($minSize) || $lpsize < $minSize) {
		    $minSize = $lpsize;
		}
857 858
	    }
	}
859 860 861 862 863 864
	if ($LVM_RAID && @blockdevs < 2) {
	    print STDERR "WARNING: not enough disks available, ".
		"not configuring RAID.\n";
	    $LVM_RAID = 0;
	    goto again;
	}
865
	if (@blockdevs == 0) {
866 867 868 869 870
	    print STDERR "ERROR: findSpareDisks found no disks for LVM!\n";
	    TBScriptUnlock();
	    return -1;
	}
		    
871
	my $blockdevstr = join(' ', sort @blockdevs);
872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896

	#
	# If we are doing a reliable LVM setup, setup an mdadm RAID10
	# with two copies. This seems to be the least expensive form
	# of redundancy we can get (as opposed to using RAID5 when there
	# are three or more disks).
	#
	# Note that this setup is unique to mdadm and is equivalent to
	# RAID1 with two disks, and equivalent to RAID1+0 (i.e., what is
	# typically referred to as "RAID10") with four disks.
	#
	if ($LVM_RAID) {
	    my $rdev = "/dev/md/xen";
	    my $rargs;
	    my $ndevs = int(@blockdevs);

	    mysystem("mdadm --create --verbose $rdev ".
		     "--level=10 --raid-devices=$ndevs $blockdevstr");
	    sleep(1);
	    # XXX raid5 config lists parity drive as spare til initialized
	    mysystem("mdadm --detail --scan | sed -e 's/spares=1//' ".
		     ">>/etc/mdadm/mdadm.conf");
	    $blockdevstr = $rdev;
	}

897 898
	mysystem("pvcreate $blockdevstr");
	mysystem("vgcreate $VGNAME $blockdevstr");
899 900 901

	my $size = lvmVGSize($VGNAME);
	if ($size < $XEN_MIN_VGSIZE) {
Mike Hibler's avatar
Mike Hibler committed
902 903 904 905 906 907 908 909
	    print STDERR "WARNING: physical disk space below the desired ".
		" minimum value ($size < $XEN_MIN_VGSIZE), expect trouble.\n";
	}

	#
	# Create an image pool for golden images.
	# If this fails, we just don't use thin volumes!
	#
910
	if ($usethin && createThinPool($blockdevstr)) {
Mike Hibler's avatar
Mike Hibler committed
911 912 913
	    print STDERR "WARNING: could not create a thin pool, ".
		"disabling golden image support\n";
	    $usethin = 0;
914 915
	}
    }
916 917
    $STRIPE_COUNT = computeStripeSize($VGNAME);
    
918 919 920 921 922 923 924 925
    #
    # Make sure our volumes are active -- they seem to become inactive
    # across reboots
    #
    mysystem("vgchange -a y $VGNAME");

    print "Creating dhcp.conf skeleton...\n"
        if ($debug);
926
    createDHCP();
927

928
    print "Creating scratch FS ...\n";
Leigh B Stoller's avatar
Leigh B Stoller committed
929
    if (createExtraFS($EXTRAFS, $VGNAME, "25G")) {
930 931 932
	TBScriptUnlock();
	return -1;
    }
933
    print "Creating image metadata FS ...\n";
Leigh B Stoller's avatar
Leigh B Stoller committed
934
    if (createExtraFS($METAFS, $VGNAME, "1G")) {
Leigh B Stoller's avatar
Leigh B Stoller committed
935 936 937
	TBScriptUnlock();
	return -1;
    }
938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957
    print "Creating container info FS ...\n";
    if (createExtraFS($INFOFS, $VGNAME, "3G")) {
	TBScriptUnlock();
	return -1;
    }
    if (! -l $VMS) {
	#
	# We need this stuff to be sticky across reloads, so move it
	# into an lvm. If we lose the lvm, well then we are screwed.
	#
	my @files = glob("$VMS/*");
	foreach my $file (@files) {
	    my $base = basename($file);
	    mysystem("/bin/mv $file $INFOFS")
		if (! -e "$INFOFS/$base");
	}
	mysystem("/bin/rm -rf $VMS");
	mysystem("/bin/ln -s $INFOFS $VMS");
    }

Leigh B Stoller's avatar
Leigh B Stoller committed
958 959 960 961 962
    if (InitializeRouteTables()) {
	print STDERR "*** Could not initialize routing table DB\n";
	TBScriptUnlock();
	return -1;
    }
963

964 965 966 967 968
    #
    # Make sure IP forwarding is enabled on the host
    #
    mysystem2("$SYSCTL -w net.ipv4.conf.all.forwarding=1");

Mike Hibler's avatar
Mike Hibler committed
969 970 971 972 973 974
    #
    # Increase socket buffer size for frisbee download of images.
    #
    mysystem2("$SYSCTL -w net.core.rmem_max=1048576");
    mysystem2("$SYSCTL -w net.core.wmem_max=1048576");

975 976 977
    #
    # Need these to avoid overflowing the NAT tables.
    #
Leigh B Stoller's avatar
Leigh B Stoller committed
978 979 980 981 982 983
    mysystem2("$MODPROBE nf_conntrack");
    if ($?) {
	print STDERR "ERROR: could not load nf_conntrack module!\n";
	TBScriptUnlock();
	return -1;
    }
984
    mysystem2("$SYSCTL -w ".
985
	     "  net.netfilter.nf_conntrack_generic_timeout=120");
986
    mysystem2("$SYSCTL -w ".
987
	     "  net.netfilter.nf_conntrack_tcp_timeout_established=54000");
988
    mysystem2("$SYSCTL -w ".
989
	     "  net.netfilter.nf_conntrack_max=131071");
990
    mysystem2("echo 16384 > /sys/module/nf_conntrack/parameters/hashsize");
991 992 993 994 995 996 997
 
    # These might fail on new kernels.  
    mysystem2("$SYSCTL -w ".
	      " net.ipv4.netfilter.ip_conntrack_generic_timeout=120");
    mysystem2("$SYSCTL -w ".
	      " net.ipv4.netfilter.ip_conntrack_tcp_timeout_established=54000");

998
    mysystem("touch /var/run/xen.ready");
Mike Hibler's avatar
Mike Hibler committed
999 1000
    TBDebugTimeStamp("  releasing global lock")
	if ($lockdebug);
1001
    TBScriptUnlock();
Jon Rafkind's avatar
Jon Rafkind committed
1002 1003 1004
    return 0;
}

1005
sub rootPreConfigNetwork($$$$)
1006
{
1007 1008 1009
    my ($vnode_id, undef, $vnconfig, $private) = @_;
    my @node_ifs = @{ $vnconfig->{'ifconfig'} };
    my @node_lds = @{ $vnconfig->{'ldconfig'} };
Jon Rafkind's avatar
Jon Rafkind committed
1010

Mike Hibler's avatar
Mike Hibler committed
1011 1012
    TBDebugTimeStamp("rootPreConfigNetwork: grabbing global lock $GLOBAL_CONF_LOCK")
	if ($lockdebug);
1013 1014 1015
    if (TBScriptLock($GLOBAL_CONF_LOCK,
		     TBSCRIPTLOCK_INTERRUPTIBLE(), 900) != TBSCRIPTLOCK_OKAY()){
	print STDERR "Could not get the global lock!\n";
1016 1017
	return -1;
    }
Mike Hibler's avatar
Mike Hibler committed
1018 1019
    TBDebugTimeStamp("  got global lock")
	if ($lockdebug);
1020

1021
    createDHCP()
1022
	if (! -e $DHCPCONF_FILE && ! -e $NEW_DHCPCONF_FILE);
1023

1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034
    if (!$ISREMOTENODE) {
	my ($cnet_iface) = findControlNet();
	my ($alias_ip,$alias_mask) = domain0ControlNet();
	my $alias_iface = "$cnet_iface:1";

	if (system("ifconfig $alias_iface | grep -q 'inet addr'")) {
	    print "Creating $alias_iface alias...\n";
	    mysystem("ifconfig $alias_iface $alias_ip netmask $alias_mask");
	}
    }

1035 1036 1037 1038 1039 1040
    #
    # If we blocked, it would be because vnodes have come or gone,
    # so we need to rebuild the maps.
    #
    makeIfaceMaps();
    makeBridgeMaps();
Jon Rafkind's avatar
Jon Rafkind committed
1041

Mike Hibler's avatar
Mike Hibler committed
1042 1043
    TBDebugTimeStamp("  releasing global lock")
	if ($lockdebug);
1044
    TBScriptUnlock();
1045 1046 1047 1048 1049
    # XXX let vnodesetup exit early
    if ($vsrelease eq "immediate") {
	TBDebugTimeStamp("rootPreConfigNetwork: touching $VMS/$vnode_id/running");
	mysystem2("touch $VMS/$vnode_id/running");
    }
Jon Rafkind's avatar
Jon Rafkind committed
1050
    return 0;
Leigh B Stoller's avatar
Leigh B Stoller committed
1051 1052 1053
bad:
    TBScriptUnlock();
    return -1;
Jon Rafkind's avatar
Jon Rafkind committed
1054 1055
}

1056 1057
sub rootPostConfig($)
{
Jon Rafkind's avatar
Jon Rafkind committed
1058 1059 1060
    return 0;
}

1061 1062
#
# Create the basic context for the VM and give it a unique ID for identifying
1063 1064
# "internal" state.  If $raref is set, then we are in a RELOAD state machine
# and need to walk the appropriate states.
1065
#
1066
sub vnodeCreate($$$$)
1067
{
1068
    my ($vnode_id, undef, $vnconfig, $private) = @_;
1069
    my $attributes = $vnconfig->{'attributes'};
1070
    my $imagename = $vnconfig->{'image'};
Leigh B Stoller's avatar
Leigh B Stoller committed
1071
    my $raref = $vnconfig->{'reloadinfo'};
1072
    my $vninfo = $private;
1073
    my %image = %defaultImage;
Leigh B Stoller's avatar
Leigh B Stoller committed
1074
    my $imagemetadata;
1075
    my $lvname;
1076
    my $inreload = 0;
Mike Hibler's avatar
Mike Hibler committed
1077
    my $dothinlv = doingThinLVM();
1078 1079

    my $vmid;
1080
    if ($vnode_id =~ /^[-\w]+\-(\d+)$/) {
1081 1082 1083 1084 1085
	$vmid = $1;
    }
    else {
	fatal("xen_vnodeCreate: bad vnode_id $vnode_id!");
    }
1086 1087
    $vninfo->{'vmid'} = $vmid;

1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
    #
    # XXX future optimization possibility.
    #
    # Try to be smart about holding the vnode creation lock which is not
    # a single lock, but rather a small set of locks intended to limit
    # concurrency in the vnode creation process. Specifically, if we grab
    # a create_vnode lock and then block waiting for our image lock, then
    # we might prevent someone else (using a different image) from making 
    # progress. So we could instead: grab a create_vnode lock, make a short
    # attempt (5-10 seconds) to grab the image lock and, failing that, back
    # off of the create_vnode lock, wait and then try the whole process again.
    #
    # The problem is that we may block again down in downloadOneImage when
    # we try to grab the image lock exclusively. Not sure we can back all
    # the way out easily in that case!
    #
    # This is also a bit of a de-optimization when we have a set of vnodes
    # all using the same image. We just cause a bit of excess context
    # switching in that (probably more common) case.
    #

Leigh B Stoller's avatar
Leigh B Stoller committed
1109
    if (CreateVnodeLock() != 0) {
1110
	fatal("CreateVnodeLock()");
Leigh B Stoller's avatar
Leigh B Stoller committed
1111 1112 1113 1114 1115 1116
    }

    #
    # We need to lock while messing with the image. But we can use
    # shared lock so that others can proceed in parallel. We will have
    # to promote to an exclusive lock if the image has to be changed.
1117
    #
1118
    my $imagelockname = ImageLockName($imagename);
Mike Hibler's avatar
Mike Hibler committed
1119 1120
    TBDebugTimeStamp("grabbing image lock $imagelockname shared")
	if ($lockdebug);