PVE/QemuServer.pm

   1 package PVE::QemuServer;
   2
   3 use strict;
   4 use POSIX;
   5 use IO::Handle;
   6 use IO::Select;
   7 use IO::File;
   8 use IO::Dir;
   9 use IO::Socket::UNIX;
  10 use File::Basename;
  11 use File::Path;
  12 use File::stat;
  13 use Getopt::Long;
  14 use Digest::SHA;
  15 use Fcntl ':flock';
  16 use Cwd 'abs_path';
  17 use IPC::Open3;
  18 use JSON;
  19 use Fcntl;
  20 use PVE::SafeSyslog;
  21 use Storable qw(dclone);
  22 use PVE::Exception qw(raise raise_param_exc);
  23 use PVE::Storage;
  24 use PVE::Tools qw(run_command lock_file file_read_firstline);
  25 use PVE::JSONSchema qw(get_standard_option);
  26 use PVE::Cluster qw(cfs_register_file cfs_read_file cfs_write_file cfs_lock_file);
  27 use PVE::INotify;
  28 use PVE::ProcFSTools;
  29 use PVE::QMPClient;
  30 use Time::HiRes qw(gettimeofday);
  31
  32 my $cpuinfo = PVE::ProcFSTools::read_cpuinfo();
  33
  34 # Note about locking: we use flock on the config file protect
  35 # against concurent actions.
  36 # Aditionaly, we have a 'lock' setting in the config file. This
  37 # can be set to 'migrate', 'backup', 'snapshot' or 'rollback'. Most actions are not
  38 # allowed when such lock is set. But you can ignore this kind of
  39 # lock with the --skiplock flag.
  40
  41 cfs_register_file('/qemu-server/',
  42                   \&parse_vm_config,
  43                   \&write_vm_config);
  44
  45 PVE::JSONSchema::register_standard_option('skiplock', {
  46     description => "Ignore locks - only root is allowed to use this option.",
  47     type => 'boolean',
  48     optional => 1,
  49 });
  50
  51 PVE::JSONSchema::register_standard_option('pve-qm-stateuri', {
  52     description => "Some command save/restore state from this location.",
  53     type => 'string',
  54     maxLength => 128,
  55     optional => 1,
  56 });
  57
  58 PVE::JSONSchema::register_standard_option('pve-snapshot-name', {
  59     description => "The name of the snapshot.",
  60     type => 'string', format => 'pve-configid',
  61     maxLength => 40,
  62 });
  63
  64 #no warnings 'redefine';
  65
  66 unless(defined(&_VZSYSCALLS_H_)) {
  67     eval 'sub _VZSYSCALLS_H_ () {1;}' unless defined(&_VZSYSCALLS_H_);
  68     require 'sys/syscall.ph';
  69     if(defined(&__x86_64__)) {
  70         eval 'sub __NR_fairsched_vcpus () {499;}' unless defined(&__NR_fairsched_vcpus);
  71         eval 'sub __NR_fairsched_mknod () {504;}' unless defined(&__NR_fairsched_mknod);
  72         eval 'sub __NR_fairsched_rmnod () {505;}' unless defined(&__NR_fairsched_rmnod);
  73         eval 'sub __NR_fairsched_chwt () {506;}' unless defined(&__NR_fairsched_chwt);
  74         eval 'sub __NR_fairsched_mvpr () {507;}' unless defined(&__NR_fairsched_mvpr);
  75         eval 'sub __NR_fairsched_rate () {508;}' unless defined(&__NR_fairsched_rate);
  76         eval 'sub __NR_setluid () {501;}' unless defined(&__NR_setluid);
  77         eval 'sub __NR_setublimit () {502;}' unless defined(&__NR_setublimit);
  78     }
  79     elsif(defined( &__i386__) ) {
  80         eval 'sub __NR_fairsched_mknod () {500;}' unless defined(&__NR_fairsched_mknod);
  81         eval 'sub __NR_fairsched_rmnod () {501;}' unless defined(&__NR_fairsched_rmnod);
  82         eval 'sub __NR_fairsched_chwt () {502;}' unless defined(&__NR_fairsched_chwt);
  83         eval 'sub __NR_fairsched_mvpr () {503;}' unless defined(&__NR_fairsched_mvpr);
  84         eval 'sub __NR_fairsched_rate () {504;}' unless defined(&__NR_fairsched_rate);
  85         eval 'sub __NR_fairsched_vcpus () {505;}' unless defined(&__NR_fairsched_vcpus);
  86         eval 'sub __NR_setluid () {511;}' unless defined(&__NR_setluid);
  87         eval 'sub __NR_setublimit () {512;}' unless defined(&__NR_setublimit);
  88     } else {
  89         die("no fairsched syscall for this arch");
  90     }
  91     require 'asm/ioctl.ph';
  92     eval 'sub KVM_GET_API_VERSION () { &_IO(0xAE, 0x);}' unless defined(&KVM_GET_API_VERSION);
  93 }
  94
  95 sub fairsched_mknod {
  96     my ($parent, $weight, $desired) = @_;
  97
  98     return syscall(&__NR_fairsched_mknod, int($parent), int($weight), int($desired));
  99 }
 100
 101 sub fairsched_rmnod {
 102     my ($id) = @_;
 103
 104     return syscall(&__NR_fairsched_rmnod, int($id));
 105 }
 106
 107 sub fairsched_mvpr {
 108     my ($pid, $newid) = @_;
 109
 110     return syscall(&__NR_fairsched_mvpr, int($pid), int($newid));
 111 }
 112
 113 sub fairsched_vcpus {
 114     my ($id, $vcpus) = @_;
 115
 116     return syscall(&__NR_fairsched_vcpus, int($id), int($vcpus));
 117 }
 118
 119 sub fairsched_rate {
 120     my ($id, $op, $rate) = @_;
 121
 122     return syscall(&__NR_fairsched_rate, int($id), int($op), int($rate));
 123 }
 124
 125 use constant FAIRSCHED_SET_RATE  => 0;
 126 use constant FAIRSCHED_DROP_RATE => 1;
 127 use constant FAIRSCHED_GET_RATE  => 2;
 128
 129 sub fairsched_cpulimit {
 130     my ($id, $limit) = @_;
 131
 132     my $cpulim1024 = int($limit * 1024 / 100);
 133     my $op = $cpulim1024 ? FAIRSCHED_SET_RATE : FAIRSCHED_DROP_RATE;
 134
 135     return fairsched_rate($id, $op, $cpulim1024);
 136 }
 137
 138 my $nodename = PVE::INotify::nodename();
 139
 140 mkdir "/etc/pve/nodes/$nodename";
 141 my $confdir = "/etc/pve/nodes/$nodename/qemu-server";
 142 mkdir $confdir;
 143
 144 my $var_run_tmpdir = "/var/run/qemu-server";
 145 mkdir $var_run_tmpdir;
 146
 147 my $lock_dir = "/var/lock/qemu-server";
 148 mkdir $lock_dir;
 149
 150 my $pcisysfs = "/sys/bus/pci";
 151
 152 my $confdesc = {
 153     onboot => {
 154         optional => 1,
 155         type => 'boolean',
 156         description => "Specifies whether a VM will be started during system bootup.",
 157         default => 0,
 158     },
 159     autostart => {
 160         optional => 1,
 161         type => 'boolean',
 162         description => "Automatic restart after crash (currently ignored).",
 163         default => 0,
 164     },
 165     hotplug => {
 166         optional => 1,
 167         type => 'boolean',
 168         description => "Activate hotplug for disk and network device",
 169         default => 0,
 170     },
 171     reboot => {
 172         optional => 1,
 173         type => 'boolean',
 174         description => "Allow reboot. If set to '0' the VM exit on reboot.",
 175         default => 1,
 176     },
 177     lock => {
 178         optional => 1,
 179         type => 'string',
 180         description => "Lock/unlock the VM.",
 181         enum => [qw(migrate backup snapshot rollback)],
 182     },
 183     cpulimit => {
 184         optional => 1,
 185         type => 'integer',
 186         description => "Limit of CPU usage in per cent. Note if the computer has 2 CPUs, it has total of 200% CPU time. Value '0' indicates no CPU limit.\n\nNOTE: This option is currently ignored.",
 187         minimum => 0,
 188         default => 0,
 189     },
 190     cpuunits => {
 191         optional => 1,
 192         type => 'integer',
 193         description => "CPU weight for a VM. Argument is used in the kernel fair scheduler. The larger the number is, the more CPU time this VM gets. Number is relative to weights of all the other running VMs.\n\nNOTE: You can disable fair-scheduler configuration by setting this to 0.",
 194         minimum => 0,
 195         maximum => 500000,
 196         default => 1000,
 197     },
 198     memory => {
 199         optional => 1,
 200         type => 'integer',
 201         description => "Amount of RAM for the VM in MB. This is the maximum available memory when you use the balloon device.",
 202         minimum => 16,
 203         default => 512,
 204     },
 205     balloon => {
 206         optional => 1,
 207         type => 'integer',
 208         description => "Amount of target RAM for the VM in MB.",
 209         minimum => 16,
 210     },
 211     keyboard => {
 212         optional => 1,
 213         type => 'string',
 214         description => "Keybord layout for vnc server. Default is read from the datacenter configuration file.",
 215         enum => PVE::Tools::kvmkeymaplist(),
 216         default => 'en-us',
 217     },
 218     name => {
 219         optional => 1,
 220         type => 'string', format => 'dns-name',
 221         description => "Set a name for the VM. Only used on the configuration web interface.",
 222     },
 223     scsihw => {
 224         optional => 1,
 225         type => 'string',
 226         description => "scsi controller model",
 227         enum => [qw(lsi virtio-scsi-pci megasas)],
 228         default => 'lsi',
 229     },
 230     description => {
 231         optional => 1,
 232         type => 'string',
 233         description => "Description for the VM. Only used on the configuration web interface. This is saved as comment inside the configuration file.",
 234     },
 235     ostype => {
 236         optional => 1,
 237         type => 'string',
 238         enum => [qw(other wxp w2k w2k3 w2k8 wvista win7 l24 l26)],
 239         description => <<EODESC,
 240 Used to enable special optimization/features for specific
 241 operating systems:
 242
 243 other  => unspecified OS
 244 wxp    => Microsoft Windows XP
 245 w2k    => Microsoft Windows 2000
 246 w2k3   => Microsoft Windows 2003
 247 w2k8   => Microsoft Windows 2008
 248 wvista => Microsoft Windows Vista
 249 win7   => Microsoft Windows 7
 250 l24    => Linux 2.4 Kernel
 251 l26    => Linux 2.6/3.X Kernel
 252
 253 other|l24|l26                  ... no special behaviour
 254 wxp|w2k|w2k3|w2k8|wvista|win7  ... use --localtime switch
 255 EODESC
 256     },
 257     boot => {
 258         optional => 1,
 259         type => 'string',
 260         description => "Boot on floppy (a), hard disk (c), CD-ROM (d), or network (n).",
 261         pattern => '[acdn]{1,4}',
 262         default => 'cdn',
 263     },
 264     bootdisk => {
 265         optional => 1,
 266         type => 'string', format => 'pve-qm-bootdisk',
 267         description => "Enable booting from specified disk.",
 268         pattern => '(ide|sata|scsi|virtio)\d+',
 269     },
 270     smp => {
 271         optional => 1,
 272         type => 'integer',
 273         description => "The number of CPUs. Please use option -sockets instead.",
 274         minimum => 1,
 275         default => 1,
 276     },
 277     sockets => {
 278         optional => 1,
 279         type => 'integer',
 280         description => "The number of CPU sockets.",
 281         minimum => 1,
 282         default => 1,
 283     },
 284     cores => {
 285         optional => 1,
 286         type => 'integer',
 287         description => "The number of cores per socket.",
 288         minimum => 1,
 289         default => 1,
 290     },
 291     acpi => {
 292         optional => 1,
 293         type => 'boolean',
 294         description => "Enable/disable ACPI.",
 295         default => 1,
 296     },
 297     agent => {
 298         optional => 1,
 299         type => 'boolean',
 300         description => "Enable/disable Qemu GuestAgent.",
 301         default => 0,
 302     },
 303     kvm => {
 304         optional => 1,
 305         type => 'boolean',
 306         description => "Enable/disable KVM hardware virtualization.",
 307         default => 1,
 308     },
 309     tdf => {
 310         optional => 1,
 311         type => 'boolean',
 312         description => "Enable/disable time drift fix. This is ignored for kvm versions newer that 1.0 (not needed anymore).",
 313         default => 1,
 314     },
 315     localtime => {
 316         optional => 1,
 317         type => 'boolean',
 318         description => "Set the real time clock to local time. This is enabled by default if ostype indicates a Microsoft OS.",
 319     },
 320     freeze => {
 321         optional => 1,
 322         type => 'boolean',
 323         description => "Freeze CPU at startup (use 'c' monitor command to start execution).",
 324     },
 325     vga => {
 326         optional => 1,
 327         type => 'string',
 328         description => "Select VGA type. If you want to use high resolution modes (>= 1280x1024x16) then you should use option 'std' or 'vmware'. Default is 'std' for win7/w2k8, and 'cirrur' for other OS types",
 329         enum => [qw(std cirrus vmware)],
 330     },
 331     watchdog => {
 332         optional => 1,
 333         type => 'string', format => 'pve-qm-watchdog',
 334         typetext => '[[model=]i6300esb|ib700] [,[action=]reset|shutdown|poweroff|pause|debug|none]',
 335         description => "Create a virtual hardware watchdog device.  Once enabled (by a guest action), the watchdog must be periodically polled by an agent inside the guest or else the guest will be restarted (or execute the action specified)",
 336     },
 337     startdate => {
 338         optional => 1,
 339         type => 'string',
 340         typetext => "(now | YYYY-MM-DD | YYYY-MM-DDTHH:MM:SS)",
 341         description => "Set the initial date of the real time clock. Valid format for date are: 'now' or '2006-06-17T16:01:21' or '2006-06-17'.",
 342         pattern => '(now|\d{4}-\d{1,2}-\d{1,2}(T\d{1,2}:\d{1,2}:\d{1,2})?)',
 343         default => 'now',
 344     },
 345     startup => {
 346         optional => 1,
 347         type => 'string', format => 'pve-qm-startup',
 348         typetext => '[[order=]\d+] [,up=\d+] [,down=\d+] ',
 349         description => "Startup and shutdown behavior. Order is a non-negative number defining the general startup order. Shutdown in done with reverse ordering. Additionally you can set the 'up' or 'down' delay in seconds, which specifies a delay to wait before the next VM is started or stopped.",
 350     },
 351     args => {
 352         optional => 1,
 353         type => 'string',
 354         description => <<EODESCR,
 355 Note: this option is for experts only. It allows you to pass arbitrary arguments to kvm, for example:
 356
 357 args: -no-reboot -no-hpet
 358 EODESCR
 359     },
 360     tablet => {
 361         optional => 1,
 362         type => 'boolean',
 363         default => 1,
 364         description => "Enable/disable the usb tablet device. This device is usually needed to allow absolute mouse positioning. Else the mouse runs out of sync with normal vnc clients. If you're running lots of console-only guests on one host, you may consider disabling this to save some context switches.",
 365     },
 366     migrate_speed => {
 367         optional => 1,
 368         type => 'integer',
 369         description => "Set maximum speed (in MB/s) for migrations. Value 0 is no limit.",
 370         minimum => 0,
 371         default => 0,
 372     },
 373     migrate_downtime => {
 374         optional => 1,
 375         type => 'integer',
 376         description => "Set maximum tolerated downtime (in seconds) for migrations.",
 377         minimum => 0,
 378         default => 1,
 379     },
 380     cdrom => {
 381         optional => 1,
 382         type => 'string', format => 'pve-qm-drive',
 383         typetext => 'volume',
 384         description => "This is an alias for option -ide2",
 385     },
 386     cpu => {
 387         optional => 1,
 388         description => "Emulated CPU type.",
 389         type => 'string',
 390         enum => [ qw(486 athlon pentium pentium2 pentium3 coreduo core2duo kvm32 kvm64 qemu32 qemu64 phenom cpu64-rhel6 cpu64-rhel5 Conroe Penryn Nehalem Westmere Opteron_G1 Opteron_G2 Opteron_G3 host) ],
 391         default => 'qemu64',
 392     },
 393     parent => get_standard_option('pve-snapshot-name', {
 394         optional => 1,
 395         description => "Parent snapshot name. This is used internally, and should not be modified.",
 396     }),
 397     snaptime => {
 398         optional => 1,
 399         description => "Timestamp for snapshots.",
 400         type => 'integer',
 401         minimum => 0,
 402     },
 403     vmstate => {
 404         optional => 1,
 405         type => 'string', format => 'pve-volume-id',
 406         description => "Reference to a volume which stores the VM state. This is used internally for snapshots.",
 407     },
 408 };
 409
 410 # what about other qemu settings ?
 411 #cpu => 'string',
 412 #machine => 'string',
 413 #fda => 'file',
 414 #fdb => 'file',
 415 #mtdblock => 'file',
 416 #sd => 'file',
 417 #pflash => 'file',
 418 #snapshot => 'bool',
 419 #bootp => 'file',
 420 ##tftp => 'dir',
 421 ##smb => 'dir',
 422 #kernel => 'file',
 423 #append => 'string',
 424 #initrd => 'file',
 425 ##soundhw => 'string',
 426
 427 while (my ($k, $v) = each %$confdesc) {
 428     PVE::JSONSchema::register_standard_option("pve-qm-$k", $v);
 429 }
 430
 431 my $MAX_IDE_DISKS = 4;
 432 my $MAX_SCSI_DISKS = 14;
 433 my $MAX_VIRTIO_DISKS = 16;
 434 my $MAX_SATA_DISKS = 6;
 435 my $MAX_USB_DEVICES = 5;
 436 my $MAX_NETS = 32;
 437 my $MAX_UNUSED_DISKS = 8;
 438 my $MAX_HOSTPCI_DEVICES = 2;
 439 my $MAX_SERIAL_PORTS = 4;
 440 my $MAX_PARALLEL_PORTS = 3;
 441
 442 my $nic_model_list = ['rtl8139', 'ne2k_pci', 'e1000',  'pcnet',  'virtio',
 443                       'ne2k_isa', 'i82551', 'i82557b', 'i82559er'];
 444 my $nic_model_list_txt = join(' ', sort @$nic_model_list);
 445
 446 # fixme:
 447 my $netdesc = {
 448     optional => 1,
 449     type => 'string', format => 'pve-qm-net',
 450     typetext => "MODEL=XX:XX:XX:XX:XX:XX [,bridge=<dev>][,rate=<mbps>][,tag=<vlanid>]",
 451     description => <<EODESCR,
 452 Specify network devices.
 453
 454 MODEL is one of: $nic_model_list_txt
 455
 456 XX:XX:XX:XX:XX:XX should be an unique MAC address. This is
 457 automatically generated if not specified.
 458
 459 The bridge parameter can be used to automatically add the interface to a bridge device. The Proxmox VE standard bridge is called 'vmbr0'.
 460
 461 Option 'rate' is used to limit traffic bandwidth from and to this interface. It is specified as floating point number, unit is 'Megabytes per second'.
 462
 463 If you specify no bridge, we create a kvm 'user' (NATed) network device, which provides DHCP and DNS services. The following addresses are used:
 464
 465 10.0.2.2   Gateway
 466 10.0.2.3   DNS Server
 467 10.0.2.4   SMB Server
 468
 469 The DHCP server assign addresses to the guest starting from 10.0.2.15.
 470
 471 EODESCR
 472 };
 473 PVE::JSONSchema::register_standard_option("pve-qm-net", $netdesc);
 474
 475 for (my $i = 0; $i < $MAX_NETS; $i++)  {
 476     $confdesc->{"net$i"} = $netdesc;
 477 }
 478
 479 my $drivename_hash;
 480
 481 my $idedesc = {
 482     optional => 1,
 483     type => 'string', format => 'pve-qm-drive',
 484     typetext => '[volume=]volume,] [,media=cdrom|disk] [,cyls=c,heads=h,secs=s[,trans=t]] [,snapshot=on|off] [,cache=none|writethrough|writeback|unsafe|directsync] [,format=f] [,backup=yes|no] [,rerror=ignore|report|stop] [,werror=enospc|ignore|report|stop] [,aio=native|threads]',
 485     description => "Use volume as IDE hard disk or CD-ROM (n is 0 to " .($MAX_IDE_DISKS -1) . ").",
 486 };
 487 PVE::JSONSchema::register_standard_option("pve-qm-ide", $idedesc);
 488
 489 my $scsidesc = {
 490     optional => 1,
 491     type => 'string', format => 'pve-qm-drive',
 492     typetext => '[volume=]volume,] [,media=cdrom|disk] [,cyls=c,heads=h,secs=s[,trans=t]] [,snapshot=on|off] [,cache=none|writethrough|writeback|unsafe|directsync] [,format=f] [,backup=yes|no] [,rerror=ignore|report|stop] [,werror=enospc|ignore|report|stop] [,aio=native|threads]',
 493     description => "Use volume as SCSI hard disk or CD-ROM (n is 0 to " . ($MAX_SCSI_DISKS - 1) . ").",
 494 };
 495 PVE::JSONSchema::register_standard_option("pve-qm-scsi", $scsidesc);
 496
 497 my $satadesc = {
 498     optional => 1,
 499     type => 'string', format => 'pve-qm-drive',
 500     typetext => '[volume=]volume,] [,media=cdrom|disk] [,cyls=c,heads=h,secs=s[,trans=t]] [,snapshot=on|off] [,cache=none|writethrough|writeback|unsafe|directsync] [,format=f] [,backup=yes|no] [,rerror=ignore|report|stop] [,werror=enospc|ignore|report|stop] [,aio=native|threads]',
 501     description => "Use volume as SATA hard disk or CD-ROM (n is 0 to " . ($MAX_SATA_DISKS - 1). ").",
 502 };
 503 PVE::JSONSchema::register_standard_option("pve-qm-sata", $satadesc);
 504
 505 my $virtiodesc = {
 506     optional => 1,
 507     type => 'string', format => 'pve-qm-drive',
 508     typetext => '[volume=]volume,] [,media=cdrom|disk] [,cyls=c,heads=h,secs=s[,trans=t]] [,snapshot=on|off] [,cache=none|writethrough|writeback|unsafe|directsync] [,format=f] [,backup=yes|no] [,rerror=ignore|report|stop] [,werror=enospc|ignore|report|stop] [,aio=native|threads]',
 509     description => "Use volume as VIRTIO hard disk (n is 0 to " . ($MAX_VIRTIO_DISKS - 1) . ").",
 510 };
 511 PVE::JSONSchema::register_standard_option("pve-qm-virtio", $virtiodesc);
 512
 513 my $usbdesc = {
 514     optional => 1,
 515     type => 'string', format => 'pve-qm-usb-device',
 516     typetext => 'host=HOSTUSBDEVICE',
 517     description => <<EODESCR,
 518 Configure an USB device (n is 0 to 4). This can be used to
 519 pass-through usb devices to the guest. HOSTUSBDEVICE syntax is:
 520
 521 'bus-port(.port)*' (decimal numbers) or
 522 'vendor_id:product_id' (hexadeciaml numbers)
 523
 524 You can use the 'lsusb -t' command to list existing usb devices.
 525
 526 Note: This option allows direct access to host hardware. So it is no longer possible to migrate such machines - use with special care.
 527
 528 EODESCR
 529 };
 530 PVE::JSONSchema::register_standard_option("pve-qm-usb", $usbdesc);
 531
 532 my $hostpcidesc = {
 533         optional => 1,
 534         type => 'string', format => 'pve-qm-hostpci',
 535         typetext => "HOSTPCIDEVICE",
 536         description => <<EODESCR,
 537 Map host pci devices. HOSTPCIDEVICE syntax is:
 538
 539 'bus:dev.func' (hexadecimal numbers)
 540
 541 You can us the 'lspci' command to list existing pci devices.
 542
 543 Note: This option allows direct access to host hardware. So it is no longer possible to migrate such machines - use with special care.
 544
 545 Experimental: user reported problems with this option.
 546 EODESCR
 547 };
 548 PVE::JSONSchema::register_standard_option("pve-qm-hostpci", $hostpcidesc);
 549
 550 my $serialdesc = {
 551         optional => 1,
 552         type => 'string',
 553         pattern => '/dev/ttyS\d+',
 554         description =>  <<EODESCR,
 555 Map host serial devices (n is 0 to 3).
 556
 557 Note: This option allows direct access to host hardware. So it is no longer possible to migrate such machines - use with special care.
 558
 559 Experimental: user reported problems with this option.
 560 EODESCR
 561 };
 562
 563 my $paralleldesc= {
 564         optional => 1,
 565         type => 'string',
 566         pattern => '/dev/parport\d+',
 567         description =>  <<EODESCR,
 568 Map host parallel devices (n is 0 to 2).
 569
 570 Note: This option allows direct access to host hardware. So it is no longer possible to migrate such machines - use with special care.
 571
 572 Experimental: user reported problems with this option.
 573 EODESCR
 574 };
 575
 576 for (my $i = 0; $i < $MAX_PARALLEL_PORTS; $i++)  {
 577     $confdesc->{"parallel$i"} = $paralleldesc;
 578 }
 579
 580 for (my $i = 0; $i < $MAX_SERIAL_PORTS; $i++)  {
 581     $confdesc->{"serial$i"} = $serialdesc;
 582 }
 583
 584 for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
 585     $confdesc->{"hostpci$i"} = $hostpcidesc;
 586 }
 587
 588 for (my $i = 0; $i < $MAX_IDE_DISKS; $i++)  {
 589     $drivename_hash->{"ide$i"} = 1;
 590     $confdesc->{"ide$i"} = $idedesc;
 591 }
 592
 593 for (my $i = 0; $i < $MAX_SATA_DISKS; $i++)  {
 594     $drivename_hash->{"sata$i"} = 1;
 595     $confdesc->{"sata$i"} = $satadesc;
 596 }
 597
 598 for (my $i = 0; $i < $MAX_SCSI_DISKS; $i++)  {
 599     $drivename_hash->{"scsi$i"} = 1;
 600     $confdesc->{"scsi$i"} = $scsidesc ;
 601 }
 602
 603 for (my $i = 0; $i < $MAX_VIRTIO_DISKS; $i++)  {
 604     $drivename_hash->{"virtio$i"} = 1;
 605     $confdesc->{"virtio$i"} = $virtiodesc;
 606 }
 607
 608 for (my $i = 0; $i < $MAX_USB_DEVICES; $i++)  {
 609     $confdesc->{"usb$i"} = $usbdesc;
 610 }
 611
 612 my $unuseddesc = {
 613     optional => 1,
 614     type => 'string', format => 'pve-volume-id',
 615     description => "Reference to unused volumes.",
 616 };
 617
 618 for (my $i = 0; $i < $MAX_UNUSED_DISKS; $i++)  {
 619     $confdesc->{"unused$i"} = $unuseddesc;
 620 }
 621
 622 my $kvm_api_version = 0;
 623
 624 sub kvm_version {
 625
 626     return $kvm_api_version if $kvm_api_version;
 627
 628     my $fh = IO::File->new("</dev/kvm") ||
 629         return 0;
 630
 631     if (my $v = $fh->ioctl(KVM_GET_API_VERSION(), 0)) {
 632         $kvm_api_version = $v;
 633     }
 634
 635     $fh->close();
 636
 637     return  $kvm_api_version;
 638 }
 639
 640 my $kvm_user_version;
 641
 642 sub kvm_user_version {
 643
 644     return $kvm_user_version if $kvm_user_version;
 645
 646     $kvm_user_version = 'unknown';
 647
 648     my $tmp = `kvm -help 2>/dev/null`;
 649
 650     if ($tmp =~ m/^QEMU( PC)? emulator version (\d+\.\d+(\.\d+)?) /) {
 651         $kvm_user_version = $2;
 652     }
 653
 654     return $kvm_user_version;
 655
 656 }
 657
 658 my $kernel_has_vhost_net = -c '/dev/vhost-net';
 659
 660 sub disknames {
 661     # order is important - used to autoselect boot disk
 662     return ((map { "ide$_" } (0 .. ($MAX_IDE_DISKS - 1))),
 663             (map { "scsi$_" } (0 .. ($MAX_SCSI_DISKS - 1))),
 664             (map { "virtio$_" } (0 .. ($MAX_VIRTIO_DISKS - 1))),
 665             (map { "sata$_" } (0 .. ($MAX_SATA_DISKS - 1))));
 666 }
 667
 668 sub valid_drivename {
 669     my $dev = shift;
 670
 671     return defined($drivename_hash->{$dev});
 672 }
 673
 674 sub option_exists {
 675     my $key = shift;
 676     return defined($confdesc->{$key});
 677 }
 678
 679 sub nic_models {
 680     return $nic_model_list;
 681 }
 682
 683 sub os_list_description {
 684
 685     return {
 686         other => 'Other',
 687         wxp => 'Windows XP',
 688         w2k => 'Windows 2000',
 689         w2k3 =>, 'Windows 2003',
 690         w2k8 => 'Windows 2008',
 691         wvista => 'Windows Vista',
 692         win7 => 'Windows 7',
 693         l24 => 'Linux 2.4',
 694         l26 => 'Linux 2.6',
 695     };
 696 }
 697
 698 my $cdrom_path;
 699
 700 sub get_cdrom_path {
 701
 702     return  $cdrom_path if $cdrom_path;
 703
 704     return $cdrom_path = "/dev/cdrom" if -l "/dev/cdrom";
 705     return $cdrom_path = "/dev/cdrom1" if -l "/dev/cdrom1";
 706     return $cdrom_path = "/dev/cdrom2" if -l "/dev/cdrom2";
 707 }
 708
 709 sub get_iso_path {
 710     my ($storecfg, $vmid, $cdrom) = @_;
 711
 712     if ($cdrom eq 'cdrom') {
 713         return get_cdrom_path();
 714     } elsif ($cdrom eq 'none') {
 715         return '';
 716     } elsif ($cdrom =~ m|^/|) {
 717         return $cdrom;
 718     } else {
 719         return PVE::Storage::path($storecfg, $cdrom);
 720     }
 721 }
 722
 723 # try to convert old style file names to volume IDs
 724 sub filename_to_volume_id {
 725     my ($vmid, $file, $media) = @_;
 726
 727     if (!($file eq 'none' || $file eq 'cdrom' ||
 728           $file =~ m|^/dev/.+| || $file =~ m/^([^:]+):(.+)$/)) {
 729
 730         return undef if $file =~ m|/|;
 731
 732         if ($media && $media eq 'cdrom') {
 733             $file = "local:iso/$file";
 734         } else {
 735             $file = "local:$vmid/$file";
 736         }
 737     }
 738
 739     return $file;
 740 }
 741
 742 sub verify_media_type {
 743     my ($opt, $vtype, $media) = @_;
 744
 745     return if !$media;
 746
 747     my $etype;
 748     if ($media eq 'disk') {
 749         $etype = 'images';
 750     } elsif ($media eq 'cdrom') {
 751         $etype = 'iso';
 752     } else {
 753         die "internal error";
 754     }
 755
 756     return if ($vtype eq $etype);
 757
 758     raise_param_exc({ $opt => "unexpected media type ($vtype != $etype)" });
 759 }
 760
 761 sub cleanup_drive_path {
 762     my ($opt, $storecfg, $drive) = @_;
 763
 764     # try to convert filesystem paths to volume IDs
 765
 766     if (($drive->{file} !~ m/^(cdrom|none)$/) &&
 767         ($drive->{file} !~ m|^/dev/.+|) &&
 768         ($drive->{file} !~ m/^([^:]+):(.+)$/) &&
 769         ($drive->{file} !~ m/^\d+$/)) {
 770         my ($vtype, $volid) = PVE::Storage::path_to_volume_id($storecfg, $drive->{file});
 771         raise_param_exc({ $opt => "unable to associate path '$drive->{file}' to any storage"}) if !$vtype;
 772         $drive->{media} = 'cdrom' if !$drive->{media} && $vtype eq 'iso';
 773         verify_media_type($opt, $vtype, $drive->{media});
 774         $drive->{file} = $volid;
 775     }
 776
 777     $drive->{media} = 'cdrom' if !$drive->{media} && $drive->{file} =~ m/^(cdrom|none)$/;
 778 }
 779
 780 sub create_conf_nolock {
 781     my ($vmid, $settings) = @_;
 782
 783     my $filename = config_file($vmid);
 784
 785     die "configuration file '$filename' already exists\n" if -f $filename;
 786
 787     my $defaults = load_defaults();
 788
 789     $settings->{name} = "vm$vmid" if !$settings->{name};
 790     $settings->{memory} = $defaults->{memory} if !$settings->{memory};
 791
 792     my $data = '';
 793     foreach my $opt (keys %$settings) {
 794         next if !$confdesc->{$opt};
 795
 796         my $value = $settings->{$opt};
 797         next if !$value;
 798
 799         $data .= "$opt: $value\n";
 800     }
 801
 802     PVE::Tools::file_set_contents($filename, $data);
 803 }
 804
 805 my $parse_size = sub {
 806     my ($value) = @_;
 807
 808     return undef if $value !~ m/^(\d+(\.\d+)?)([KMG])?$/;
 809     my ($size, $unit) = ($1, $3);
 810     if ($unit) {
 811         if ($unit eq 'K') {
 812             $size = $size * 1024;
 813         } elsif ($unit eq 'M') {
 814             $size = $size * 1024 * 1024;
 815         } elsif ($unit eq 'G') {
 816             $size = $size * 1024 * 1024 * 1024;
 817         }
 818     }
 819     return int($size);
 820 };
 821
 822 my $format_size = sub {
 823     my ($size) = @_;
 824
 825     $size = int($size);
 826
 827     my $kb = int($size/1024);
 828     return $size if $kb*1024 != $size;
 829
 830     my $mb = int($kb/1024);
 831     return "${kb}K" if $mb*1024 != $kb;
 832
 833     my $gb = int($mb/1024);
 834     return "${mb}M" if $gb*1024 != $mb;
 835
 836     return "${gb}G";
 837 };
 838
 839 # ideX = [volume=]volume-id[,media=d][,cyls=c,heads=h,secs=s[,trans=t]]
 840 #        [,snapshot=on|off][,cache=on|off][,format=f][,backup=yes|no]
 841 #        [,rerror=ignore|report|stop][,werror=enospc|ignore|report|stop]
 842 #        [,aio=native|threads]
 843
 844 sub parse_drive {
 845     my ($key, $data) = @_;
 846
 847     my $res = {};
 848
 849     # $key may be undefined - used to verify JSON parameters
 850     if (!defined($key)) {
 851         $res->{interface} = 'unknown'; # should not harm when used to verify parameters
 852         $res->{index} = 0;
 853     } elsif ($key =~ m/^([^\d]+)(\d+)$/) {
 854         $res->{interface} = $1;
 855         $res->{index} = $2;
 856     } else {
 857         return undef;
 858     }
 859
 860     foreach my $p (split (/,/, $data)) {
 861         next if $p =~ m/^\s*$/;
 862
 863         if ($p =~ m/^(file|volume|cyls|heads|secs|trans|media|snapshot|cache|format|rerror|werror|backup|aio|bps|mbps|bps_rd|mbps_rd|bps_wr|mbps_wr|iops|iops_rd|iops_wr|size)=(.+)$/) {
 864             my ($k, $v) = ($1, $2);
 865
 866             $k = 'file' if $k eq 'volume';
 867
 868             return undef if defined $res->{$k};
 869
 870             if ($k eq 'bps' || $k eq 'bps_rd' || $k eq 'bps_wr') {
 871                 return undef if !$v || $v !~ m/^\d+/;
 872                 $k = "m$k";
 873                 $v = sprintf("%.3f", $v / (1024*1024));
 874             }
 875             $res->{$k} = $v;
 876         } else {
 877             if (!$res->{file} && $p !~ m/=/) {
 878                 $res->{file} = $p;
 879             } else {
 880                 return undef;
 881             }
 882         }
 883     }
 884
 885     return undef if !$res->{file};
 886
 887     return undef if $res->{cache} &&
 888         $res->{cache} !~ m/^(off|none|writethrough|writeback|unsafe|directsync)$/;
 889     return undef if $res->{snapshot} && $res->{snapshot} !~ m/^(on|off)$/;
 890     return undef if $res->{cyls} && $res->{cyls} !~ m/^\d+$/;
 891     return undef if $res->{heads} && $res->{heads} !~ m/^\d+$/;
 892     return undef if $res->{secs} && $res->{secs} !~ m/^\d+$/;
 893     return undef if $res->{media} && $res->{media} !~ m/^(disk|cdrom)$/;
 894     return undef if $res->{trans} && $res->{trans} !~ m/^(none|lba|auto)$/;
 895     return undef if $res->{format} && $res->{format} !~ m/^(raw|cow|qcow|qcow2|vmdk|cloop)$/;
 896     return undef if $res->{rerror} && $res->{rerror} !~ m/^(ignore|report|stop)$/;
 897     return undef if $res->{werror} && $res->{werror} !~ m/^(enospc|ignore|report|stop)$/;
 898     return undef if $res->{backup} && $res->{backup} !~ m/^(yes|no)$/;
 899     return undef if $res->{aio} && $res->{aio} !~ m/^(native|threads)$/;
 900
 901
 902     return undef if $res->{mbps_rd} && $res->{mbps};
 903     return undef if $res->{mbps_wr} && $res->{mbps};
 904
 905     return undef if $res->{mbps} && $res->{mbps} !~ m/^\d+(\.\d+)?$/;
 906     return undef if $res->{mbps_rd} && $res->{mbps_rd} !~ m/^\d+(\.\d+)?$/;
 907     return undef if $res->{mbps_wr} && $res->{mbps_wr} !~ m/^\d+(\.\d+)?$/;
 908
 909     return undef if $res->{iops_rd} && $res->{iops};
 910     return undef if $res->{iops_wr} && $res->{iops};
 911     return undef if $res->{iops} && $res->{iops} !~ m/^\d+$/;
 912     return undef if $res->{iops_rd} && $res->{iops_rd} !~ m/^\d+$/;
 913     return undef if $res->{iops_wr} && $res->{iops_wr} !~ m/^\d+$/;
 914
 915
 916     if ($res->{size}) {
 917         return undef if !defined($res->{size} = &$parse_size($res->{size}));
 918     }
 919
 920     if ($res->{media} && ($res->{media} eq 'cdrom')) {
 921         return undef if $res->{snapshot} || $res->{trans} || $res->{format};
 922         return undef if $res->{heads} || $res->{secs} || $res->{cyls};
 923         return undef if $res->{interface} eq 'virtio';
 924     }
 925
 926     # rerror does not work with scsi drives
 927     if ($res->{rerror}) {
 928         return undef if $res->{interface} eq 'scsi';
 929     }
 930
 931     return $res;
 932 }
 933
 934 my @qemu_drive_options = qw(heads secs cyls trans media format cache snapshot rerror werror aio iops iops_rd iops_wr);
 935
 936 sub print_drive {
 937     my ($vmid, $drive) = @_;
 938
 939     my $opts = '';
 940     foreach my $o (@qemu_drive_options, 'mbps', 'mbps_rd', 'mbps_wr', 'backup') {
 941         $opts .= ",$o=$drive->{$o}" if $drive->{$o};
 942     }
 943
 944     if ($drive->{size}) {
 945         $opts .= ",size=" . &$format_size($drive->{size});
 946     }
 947
 948     return "$drive->{file}$opts";
 949 }
 950
 951 sub scsi_inquiry {
 952     my($fh, $noerr) = @_;
 953
 954     my $SG_IO = 0x2285;
 955     my $SG_GET_VERSION_NUM = 0x2282;
 956
 957     my $versionbuf = "\x00" x 8;
 958     my $ret = ioctl($fh, $SG_GET_VERSION_NUM, $versionbuf);
 959     if (!$ret) {
 960         die "scsi ioctl SG_GET_VERSION_NUM failoed - $!\n" if !$noerr;
 961         return undef;
 962     }
 963     my $version = unpack("I", $versionbuf);
 964     if ($version < 30000) {
 965         die "scsi generic interface too old\n"  if !$noerr;
 966         return undef;
 967     }
 968
 969     my $buf = "\x00" x 36;
 970     my $sensebuf = "\x00" x 8;
 971     my $cmd = pack("C x3 C x11", 0x12, 36);
 972
 973     # see /usr/include/scsi/sg.h
 974     my $sg_io_hdr_t = "i i C C s I P P P I I i P C C C C S S i I I";
 975
 976     my $packet = pack($sg_io_hdr_t, ord('S'), -3, length($cmd),
 977                       length($sensebuf), 0, length($buf), $buf,
 978                       $cmd, $sensebuf, 6000);
 979
 980     $ret = ioctl($fh, $SG_IO, $packet);
 981     if (!$ret) {
 982         die "scsi ioctl SG_IO failed - $!\n" if !$noerr;
 983         return undef;
 984     }
 985
 986     my @res = unpack($sg_io_hdr_t, $packet);
 987     if ($res[17] || $res[18]) {
 988         die "scsi ioctl SG_IO status error - $!\n" if !$noerr;
 989         return undef;
 990     }
 991
 992     my $res = {};
 993     ($res->{device}, $res->{removable}, $res->{venodor},
 994      $res->{product}, $res->{revision}) = unpack("C C x6 A8 A16 A4", $buf);
 995
 996     return $res;
 997 }
 998
 999 sub path_is_scsi {
1000     my ($path) = @_;
1001
1002     my $fh = IO::File->new("+<$path") || return undef;
1003     my $res = scsi_inquiry($fh, 1);
1004     close($fh);
1005
1006     return $res;
1007 }
1008
1009 sub print_drivedevice_full {
1010     my ($storecfg, $conf, $vmid, $drive, $bridges) = @_;
1011
1012     my $device = '';
1013     my $maxdev = 0;
1014
1015     if ($drive->{interface} eq 'virtio') {
1016         my $pciaddr = print_pci_addr("$drive->{interface}$drive->{index}", $bridges);
1017         $device = "virtio-blk-pci,drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}$pciaddr";
1018     } elsif ($drive->{interface} eq 'scsi') {
1019         $maxdev = ($conf->{scsihw} && $conf->{scsihw} ne 'lsi') ? 256 : 7;
1020         my $controller = int($drive->{index} / $maxdev);
1021         my $unit = $drive->{index} % $maxdev;
1022         my $devicetype = 'hd';
1023         my $path = '';
1024         if (drive_is_cdrom($drive)) {
1025               $devicetype = 'cd';
1026           } else {
1027               if ($drive->{file} =~ m|^/|) {
1028                   $path = $drive->{file};
1029               } else {
1030                   $path = PVE::Storage::path($storecfg, $drive->{file});
1031               }
1032
1033               if($path =~ m/^iscsi\:\/\//){
1034                  $devicetype = 'generic';
1035               }
1036               else {
1037                  $devicetype = 'block' if path_is_scsi($path);
1038               }
1039          }
1040
1041         if (!$conf->{scsihw} || $conf->{scsihw} eq 'lsi'){
1042             $device = "scsi-$devicetype,bus=scsihw$controller.0,scsi-id=$unit,drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}" if !$conf->{scsihw} || $conf->{scsihw} eq 'lsi';
1043         } else {
1044             $device = "scsi-$devicetype,bus=scsihw$controller.0,channel=0,scsi-id=0,lun=$drive->{index},drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}";
1045         }
1046
1047     } elsif ($drive->{interface} eq 'ide'){
1048         $maxdev = 2;
1049         my $controller = int($drive->{index} / $maxdev);
1050         my $unit = $drive->{index} % $maxdev;
1051         my $devicetype = ($drive->{media} && $drive->{media} eq 'cdrom') ? "cd" : "hd";
1052
1053         $device = "ide-$devicetype,bus=ide.$controller,unit=$unit,drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}";
1054     } elsif ($drive->{interface} eq 'sata'){
1055         my $controller = int($drive->{index} / $MAX_SATA_DISKS);
1056         my $unit = $drive->{index} % $MAX_SATA_DISKS;
1057         $device = "ide-drive,bus=ahci$controller.$unit,drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}";
1058     } elsif ($drive->{interface} eq 'usb') {
1059         die "implement me";
1060         #  -device ide-drive,bus=ide.1,unit=0,drive=drive-ide0-1-0,id=ide0-1-0
1061     } else {
1062         die "unsupported interface type";
1063     }
1064
1065     $device .= ",bootindex=$drive->{bootindex}" if $drive->{bootindex};
1066
1067     return $device;
1068 }
1069
1070 sub print_drive_full {
1071     my ($storecfg, $vmid, $drive) = @_;
1072
1073     my $opts = '';
1074     foreach my $o (@qemu_drive_options) {
1075         next if $o eq 'bootindex';
1076         $opts .= ",$o=$drive->{$o}" if $drive->{$o};
1077     }
1078
1079     foreach my $o (qw(bps bps_rd bps_wr)) {
1080         my $v = $drive->{"m$o"};
1081         $opts .= ",$o=" . int($v*1024*1024) if $v;
1082     }
1083
1084     # use linux-aio by default (qemu default is threads)
1085     $opts .= ",aio=native" if !$drive->{aio};
1086
1087     my $path;
1088     my $volid = $drive->{file};
1089     if (drive_is_cdrom($drive)) {
1090         $path = get_iso_path($storecfg, $vmid, $volid);
1091     } else {
1092         if ($volid =~ m|^/|) {
1093             $path = $volid;
1094         } else {
1095             $path = PVE::Storage::path($storecfg, $volid);
1096         }
1097         if (!$drive->{cache} && ($path =~ m|^/dev/| || $path =~ m|\.raw$|)) {
1098             $opts .= ",cache=none";
1099         }
1100     }
1101
1102     my $pathinfo = $path ? "file=$path," : '';
1103
1104     return "${pathinfo}if=none,id=drive-$drive->{interface}$drive->{index}$opts";
1105 }
1106
1107 sub print_netdevice_full {
1108     my ($vmid, $conf, $net, $netid, $bridges) = @_;
1109
1110     my $bootorder = $conf->{boot} || $confdesc->{boot}->{default};
1111
1112     my $device = $net->{model};
1113     if ($net->{model} eq 'virtio') {
1114          $device = 'virtio-net-pci';
1115      };
1116
1117     # qemu > 0.15 always try to boot from network - we disable that by
1118     # not loading the pxe rom file
1119     my $extra = ($bootorder !~ m/n/) ? "romfile=," : '';
1120     my $pciaddr = print_pci_addr("$netid", $bridges);
1121     my $tmpstr = "$device,${extra}mac=$net->{macaddr},netdev=$netid$pciaddr,id=$netid";
1122     $tmpstr .= ",bootindex=$net->{bootindex}" if $net->{bootindex} ;
1123     return $tmpstr;
1124 }
1125
1126 sub print_netdev_full {
1127     my ($vmid, $conf, $net, $netid) = @_;
1128
1129     my $i = '';
1130     if ($netid =~ m/^net(\d+)$/) {
1131         $i = int($1);
1132     }
1133
1134     die "got strange net id '$i'\n" if $i >= ${MAX_NETS};
1135
1136     my $ifname = "tap${vmid}i$i";
1137
1138     # kvm uses TUNSETIFF ioctl, and that limits ifname length
1139     die "interface name '$ifname' is too long (max 15 character)\n"
1140         if length($ifname) >= 16;
1141
1142     my $vhostparam = '';
1143     $vhostparam = ',vhost=on' if $kernel_has_vhost_net && $net->{model} eq 'virtio';
1144
1145     my $vmname = $conf->{name} || "vm$vmid";
1146
1147     if ($net->{bridge}) {
1148         return "type=tap,id=$netid,ifname=${ifname},script=/var/lib/qemu-server/pve-bridge$vhostparam";
1149     } else {
1150         return "type=user,id=$netid,hostname=$vmname";
1151     }
1152 }
1153
1154 sub drive_is_cdrom {
1155     my ($drive) = @_;
1156
1157     return $drive && $drive->{media} && ($drive->{media} eq 'cdrom');
1158
1159 }
1160
1161 sub parse_hostpci {
1162     my ($value) = @_;
1163
1164     return undef if !$value;
1165
1166     my $res = {};
1167
1168     if ($value =~ m/^[a-f0-9]{2}:[a-f0-9]{2}\.[a-f0-9]$/) {
1169        $res->{pciid} = $value;
1170     } else {
1171        return undef;
1172     }
1173
1174     return $res;
1175 }
1176
1177 # netX: e1000=XX:XX:XX:XX:XX:XX,bridge=vmbr0,rate=<mbps>
1178 sub parse_net {
1179     my ($data) = @_;
1180
1181     my $res = {};
1182
1183     foreach my $kvp (split(/,/, $data)) {
1184
1185         if ($kvp =~ m/^(ne2k_pci|e1000|rtl8139|pcnet|virtio|ne2k_isa|i82551|i82557b|i82559er)(=([0-9a-f]{2}(:[0-9a-f]{2}){5}))?$/i) {
1186             my $model = lc($1);
1187             my $mac = uc($3) || PVE::Tools::random_ether_addr();
1188             $res->{model} = $model;
1189             $res->{macaddr} = $mac;
1190         } elsif ($kvp =~ m/^bridge=(\S+)$/) {
1191             $res->{bridge} = $1;
1192         } elsif ($kvp =~ m/^rate=(\d+(\.\d+)?)$/) {
1193             $res->{rate} = $1;
1194         } elsif ($kvp =~ m/^tag=(\d+)$/) {
1195             $res->{tag} = $1;
1196         } else {
1197             return undef;
1198         }
1199
1200     }
1201
1202     return undef if !$res->{model};
1203
1204     return $res;
1205 }
1206
1207 sub print_net {
1208     my $net = shift;
1209
1210     my $res = "$net->{model}";
1211     $res .= "=$net->{macaddr}" if $net->{macaddr};
1212     $res .= ",bridge=$net->{bridge}" if $net->{bridge};
1213     $res .= ",rate=$net->{rate}" if $net->{rate};
1214     $res .= ",tag=$net->{tag}" if $net->{tag};
1215
1216     return $res;
1217 }
1218
1219 sub add_random_macs {
1220     my ($settings) = @_;
1221
1222     foreach my $opt (keys %$settings) {
1223         next if $opt !~ m/^net(\d+)$/;
1224         my $net = parse_net($settings->{$opt});
1225         next if !$net;
1226         $settings->{$opt} = print_net($net);
1227     }
1228 }
1229
1230 sub add_unused_volume {
1231     my ($config, $volid) = @_;
1232
1233     my $key;
1234     for (my $ind = $MAX_UNUSED_DISKS - 1; $ind >= 0; $ind--) {
1235         my $test = "unused$ind";
1236         if (my $vid = $config->{$test}) {
1237             return if $vid eq $volid; # do not add duplicates
1238         } else {
1239             $key = $test;
1240         }
1241     }
1242
1243     die "To many unused volume - please delete them first.\n" if !$key;
1244
1245     $config->{$key} = $volid;
1246
1247     return $key;
1248 }
1249
1250 # fixme: remove all thos $noerr parameters?
1251
1252 PVE::JSONSchema::register_format('pve-qm-bootdisk', \&verify_bootdisk);
1253 sub verify_bootdisk {
1254     my ($value, $noerr) = @_;
1255
1256     return $value if valid_drivename($value);
1257
1258     return undef if $noerr;
1259
1260     die "invalid boot disk '$value'\n";
1261 }
1262
1263 PVE::JSONSchema::register_format('pve-qm-net', \&verify_net);
1264 sub verify_net {
1265     my ($value, $noerr) = @_;
1266
1267     return $value if parse_net($value);
1268
1269     return undef if $noerr;
1270
1271     die "unable to parse network options\n";
1272 }
1273
1274 PVE::JSONSchema::register_format('pve-qm-drive', \&verify_drive);
1275 sub verify_drive {
1276     my ($value, $noerr) = @_;
1277
1278     return $value if parse_drive(undef, $value);
1279
1280     return undef if $noerr;
1281
1282     die "unable to parse drive options\n";
1283 }
1284
1285 PVE::JSONSchema::register_format('pve-qm-hostpci', \&verify_hostpci);
1286 sub verify_hostpci {
1287     my ($value, $noerr) = @_;
1288
1289     return $value if parse_hostpci($value);
1290
1291     return undef if $noerr;
1292
1293     die "unable to parse pci id\n";
1294 }
1295
1296 PVE::JSONSchema::register_format('pve-qm-watchdog', \&verify_watchdog);
1297 sub verify_watchdog {
1298     my ($value, $noerr) = @_;
1299
1300     return $value if parse_watchdog($value);
1301
1302     return undef if $noerr;
1303
1304     die "unable to parse watchdog options\n";
1305 }
1306
1307 sub parse_watchdog {
1308     my ($value) = @_;
1309
1310     return undef if !$value;
1311
1312     my $res = {};
1313
1314     foreach my $p (split(/,/, $value)) {
1315         next if $p =~ m/^\s*$/;
1316
1317         if ($p =~ m/^(model=)?(i6300esb|ib700)$/) {
1318             $res->{model} = $2;
1319         } elsif ($p =~ m/^(action=)?(reset|shutdown|poweroff|pause|debug|none)$/) {
1320             $res->{action} = $2;
1321         } else {
1322             return undef;
1323         }
1324     }
1325
1326     return $res;
1327 }
1328
1329 PVE::JSONSchema::register_format('pve-qm-startup', \&verify_startup);
1330 sub verify_startup {
1331     my ($value, $noerr) = @_;
1332
1333     return $value if parse_startup($value);
1334
1335     return undef if $noerr;
1336
1337     die "unable to parse startup options\n";
1338 }
1339
1340 sub parse_startup {
1341     my ($value) = @_;
1342
1343     return undef if !$value;
1344
1345     my $res = {};
1346
1347     foreach my $p (split(/,/, $value)) {
1348         next if $p =~ m/^\s*$/;
1349
1350         if ($p =~ m/^(order=)?(\d+)$/) {
1351             $res->{order} = $2;
1352         } elsif ($p =~ m/^up=(\d+)$/) {
1353             $res->{up} = $1;
1354         } elsif ($p =~ m/^down=(\d+)$/) {
1355             $res->{down} = $1;
1356         } else {
1357             return undef;
1358         }
1359     }
1360
1361     return $res;
1362 }
1363
1364 sub parse_usb_device {
1365     my ($value) = @_;
1366
1367     return undef if !$value;
1368
1369     my @dl = split(/,/, $value);
1370     my $found;
1371
1372     my $res = {};
1373     foreach my $v (@dl) {
1374         if ($v =~ m/^host=(0x)?([0-9A-Fa-f]{4}):(0x)?([0-9A-Fa-f]{4})$/) {
1375             $found = 1;
1376             $res->{vendorid} = $2;
1377             $res->{productid} = $4;
1378         } elsif ($v =~ m/^host=(\d+)\-(\d+(\.\d+)*)$/) {
1379             $found = 1;
1380             $res->{hostbus} = $1;
1381             $res->{hostport} = $2;
1382         } else {
1383             return undef;
1384         }
1385     }
1386     return undef if !$found;
1387
1388     return $res;
1389 }
1390
1391 PVE::JSONSchema::register_format('pve-qm-usb-device', \&verify_usb_device);
1392 sub verify_usb_device {
1393     my ($value, $noerr) = @_;
1394
1395     return $value if parse_usb_device($value);
1396
1397     return undef if $noerr;
1398
1399     die "unable to parse usb device\n";
1400 }
1401
1402 # add JSON properties for create and set function
1403 sub json_config_properties {
1404     my $prop = shift;
1405
1406     foreach my $opt (keys %$confdesc) {
1407         next if $opt eq 'parent' || $opt eq 'snaptime' || $opt eq 'vmstate';
1408         $prop->{$opt} = $confdesc->{$opt};
1409     }
1410
1411     return $prop;
1412 }
1413
1414 sub check_type {
1415     my ($key, $value) = @_;
1416
1417     die "unknown setting '$key'\n" if !$confdesc->{$key};
1418
1419     my $type = $confdesc->{$key}->{type};
1420
1421     if (!defined($value)) {
1422         die "got undefined value\n";
1423     }
1424
1425     if ($value =~ m/[\n\r]/) {
1426         die "property contains a line feed\n";
1427     }
1428
1429     if ($type eq 'boolean') {
1430         return 1 if ($value eq '1') || ($value =~ m/^(on|yes|true)$/i);
1431         return 0 if ($value eq '0') || ($value =~ m/^(off|no|false)$/i);
1432         die "type check ('boolean') failed - got '$value'\n";
1433     } elsif ($type eq 'integer') {
1434         return int($1) if $value =~ m/^(\d+)$/;
1435         die "type check ('integer') failed - got '$value'\n";
1436     } elsif ($type eq 'string') {
1437         if (my $fmt = $confdesc->{$key}->{format}) {
1438             if ($fmt eq 'pve-qm-drive') {
1439                 # special case - we need to pass $key to parse_drive()
1440                 my $drive = parse_drive($key, $value);
1441                 return $value if $drive;
1442                 die "unable to parse drive options\n";
1443             }
1444             PVE::JSONSchema::check_format($fmt, $value);
1445             return $value;
1446         }
1447         $value =~ s/^\"(.*)\"$/$1/;
1448         return $value;
1449     } else {
1450         die "internal error"
1451     }
1452 }
1453
1454 sub lock_config_full {
1455     my ($vmid, $timeout, $code, @param) = @_;
1456
1457     my $filename = config_file_lock($vmid);
1458
1459     my $res = lock_file($filename, $timeout, $code, @param);
1460
1461     die $@ if $@;
1462
1463     return $res;
1464 }
1465
1466 sub lock_config {
1467     my ($vmid, $code, @param) = @_;
1468
1469     return lock_config_full($vmid, 10, $code, @param);
1470 }
1471
1472 sub cfs_config_path {
1473     my ($vmid, $node) = @_;
1474
1475     $node = $nodename if !$node;
1476     return "nodes/$node/qemu-server/$vmid.conf";
1477 }
1478
1479 sub check_iommu_support{
1480     #fixme : need to check IOMMU support
1481     #http://www.linux-kvm.org/page/How_to_assign_devices_with_VT-d_in_KVM
1482
1483     my $iommu=1;
1484     return $iommu;
1485
1486 }
1487
1488 sub config_file {
1489     my ($vmid, $node) = @_;
1490
1491     my $cfspath = cfs_config_path($vmid, $node);
1492     return "/etc/pve/$cfspath";
1493 }
1494
1495 sub config_file_lock {
1496     my ($vmid) = @_;
1497
1498     return "$lock_dir/lock-$vmid.conf";
1499 }
1500
1501 sub touch_config {
1502     my ($vmid) = @_;
1503
1504     my $conf = config_file($vmid);
1505     utime undef, undef, $conf;
1506 }
1507
1508 sub destroy_vm {
1509     my ($storecfg, $vmid, $keep_empty_config) = @_;
1510
1511     my $conffile = config_file($vmid);
1512
1513     my $conf = load_config($vmid);
1514
1515     check_lock($conf);
1516
1517     # only remove disks owned by this VM
1518     foreach_drive($conf, sub {
1519         my ($ds, $drive) = @_;
1520
1521         return if drive_is_cdrom($drive);
1522
1523         my $volid = $drive->{file};
1524         return if !$volid || $volid =~ m|^/|;
1525
1526         my ($path, $owner) = PVE::Storage::path($storecfg, $volid);
1527         return if !$path || !$owner || ($owner != $vmid);
1528
1529         PVE::Storage::vdisk_free($storecfg, $volid);
1530     });
1531
1532     if ($keep_empty_config) {
1533         PVE::Tools::file_set_contents($conffile, "memory: 128\n");
1534     } else {
1535         unlink $conffile;
1536     }
1537
1538     # also remove unused disk
1539     eval {
1540         my $dl = PVE::Storage::vdisk_list($storecfg, undef, $vmid);
1541
1542         eval {
1543             PVE::Storage::foreach_volid($dl, sub {
1544                 my ($volid, $sid, $volname, $d) = @_;
1545                 PVE::Storage::vdisk_free($storecfg, $volid);
1546             });
1547         };
1548         warn $@ if $@;
1549
1550     };
1551     warn $@ if $@;
1552 }
1553
1554 sub load_config {
1555     my ($vmid, $node) = @_;
1556
1557     my $cfspath = cfs_config_path($vmid, $node);
1558
1559     my $conf = PVE::Cluster::cfs_read_file($cfspath);
1560
1561     die "no such VM ('$vmid')\n" if !defined($conf);
1562
1563     return $conf;
1564 }
1565
1566 sub parse_vm_config {
1567     my ($filename, $raw) = @_;
1568
1569     return undef if !defined($raw);
1570
1571     my $res = {
1572         digest => Digest::SHA::sha1_hex($raw),
1573         snapshots => {},
1574     };
1575
1576     $filename =~ m|/qemu-server/(\d+)\.conf$|
1577         || die "got strange filename '$filename'";
1578
1579     my $vmid = $1;
1580
1581     my $conf = $res;
1582     my $descr = '';
1583
1584     my @lines = split(/\n/, $raw);
1585     foreach my $line (@lines) {
1586         next if $line =~ m/^\s*$/;
1587
1588         if ($line =~ m/^\[([a-z][a-z0-9_\-]+)\]\s*$/i) {
1589             my $snapname = $1;
1590             $conf->{description} = $descr if $descr;
1591             $descr = '';
1592             $conf = $res->{snapshots}->{$snapname} = {};
1593             next;
1594         }
1595
1596         if ($line =~ m/^\#(.*)\s*$/) {
1597             $descr .= PVE::Tools::decode_text($1) . "\n";
1598             next;
1599         }
1600
1601         if ($line =~ m/^(description):\s*(.*\S)\s*$/) {
1602             $descr .= PVE::Tools::decode_text($2);
1603         } elsif ($line =~ m/snapstate:\s*(prepare|delete)\s*$/) {
1604             $conf->{snapstate} = $1;
1605         } elsif ($line =~ m/^(args):\s*(.*\S)\s*$/) {
1606             my $key = $1;
1607             my $value = $2;
1608             $conf->{$key} = $value;
1609         } elsif ($line =~ m/^([a-z][a-z_]*\d*):\s*(\S+)\s*$/) {
1610             my $key = $1;
1611             my $value = $2;
1612             eval { $value = check_type($key, $value); };
1613             if ($@) {
1614                 warn "vm $vmid - unable to parse value of '$key' - $@";
1615             } else {
1616                 my $fmt = $confdesc->{$key}->{format};
1617                 if ($fmt && $fmt eq 'pve-qm-drive') {
1618                     my $v = parse_drive($key, $value);
1619                     if (my $volid = filename_to_volume_id($vmid, $v->{file}, $v->{media})) {
1620                         $v->{file} = $volid;
1621                         $value = print_drive($vmid, $v);
1622                     } else {
1623                         warn "vm $vmid - unable to parse value of '$key'\n";
1624                         next;
1625                     }
1626                 }
1627
1628                 if ($key eq 'cdrom') {
1629                     $conf->{ide2} = $value;
1630                 } else {
1631                     $conf->{$key} = $value;
1632                 }
1633             }
1634         }
1635     }
1636
1637     $conf->{description} = $descr if $descr;
1638
1639     delete $res->{snapstate}; # just to be sure
1640
1641     return $res;
1642 }
1643
1644 sub write_vm_config {
1645     my ($filename, $conf) = @_;
1646
1647     delete $conf->{snapstate}; # just to be sure
1648
1649     if ($conf->{cdrom}) {
1650         die "option ide2 conflicts with cdrom\n" if $conf->{ide2};
1651         $conf->{ide2} = $conf->{cdrom};
1652         delete $conf->{cdrom};
1653     }
1654
1655     # we do not use 'smp' any longer
1656     if ($conf->{sockets}) {
1657         delete $conf->{smp};
1658     } elsif ($conf->{smp}) {
1659         $conf->{sockets} = $conf->{smp};
1660         delete $conf->{cores};
1661         delete $conf->{smp};
1662     }
1663
1664     my $used_volids = {};
1665
1666     my $cleanup_config = sub {
1667         my ($cref) = @_;
1668
1669         foreach my $key (keys %$cref) {
1670             next if $key eq 'digest' || $key eq 'description' || $key eq 'snapshots' ||
1671                 $key eq 'snapstate';
1672             my $value = $cref->{$key};
1673             eval { $value = check_type($key, $value); };
1674             die "unable to parse value of '$key' - $@" if $@;
1675
1676             $cref->{$key} = $value;
1677
1678             if (valid_drivename($key)) {
1679                 my $drive = PVE::QemuServer::parse_drive($key, $value);
1680                 $used_volids->{$drive->{file}} = 1 if $drive && $drive->{file};
1681             }
1682         }
1683     };
1684
1685     &$cleanup_config($conf);
1686     foreach my $snapname (keys %{$conf->{snapshots}}) {
1687         &$cleanup_config($conf->{snapshots}->{$snapname});
1688     }
1689
1690     # remove 'unusedX' settings if we re-add a volume
1691     foreach my $key (keys %$conf) {
1692         my $value = $conf->{$key};
1693         if ($key =~ m/^unused/ && $used_volids->{$value}) {
1694             delete $conf->{$key};
1695         }
1696     }
1697
1698     my $generate_raw_config = sub {
1699         my ($conf) = @_;
1700
1701         my $raw = '';
1702
1703         # add description as comment to top of file
1704         my $descr = $conf->{description} || '';
1705         foreach my $cl (split(/\n/, $descr)) {
1706             $raw .= '#' .  PVE::Tools::encode_text($cl) . "\n";
1707         }
1708
1709         foreach my $key (sort keys %$conf) {
1710             next if $key eq 'digest' || $key eq 'description' || $key eq 'snapshots';
1711             $raw .= "$key: $conf->{$key}\n";
1712         }
1713         return $raw;
1714     };
1715
1716     my $raw = &$generate_raw_config($conf);
1717     foreach my $snapname (sort keys %{$conf->{snapshots}}) {
1718         $raw .= "\n[$snapname]\n";
1719         $raw .= &$generate_raw_config($conf->{snapshots}->{$snapname});
1720     }
1721
1722     return $raw;
1723 }
1724
1725 sub update_config_nolock {
1726     my ($vmid, $conf, $skiplock) = @_;
1727
1728     check_lock($conf) if !$skiplock;
1729
1730     my $cfspath = cfs_config_path($vmid);
1731
1732     PVE::Cluster::cfs_write_file($cfspath, $conf);
1733 }
1734
1735 sub update_config {
1736     my ($vmid, $conf, $skiplock) = @_;
1737
1738     lock_config($vmid, &update_config_nolock, $conf, $skiplock);
1739 }
1740
1741 sub load_defaults {
1742
1743     my $res = {};
1744
1745     # we use static defaults from our JSON schema configuration
1746     foreach my $key (keys %$confdesc) {
1747         if (defined(my $default = $confdesc->{$key}->{default})) {
1748             $res->{$key} = $default;
1749         }
1750     }
1751
1752     my $conf = PVE::Cluster::cfs_read_file('datacenter.cfg');
1753     $res->{keyboard} = $conf->{keyboard} if $conf->{keyboard};
1754
1755     return $res;
1756 }
1757
1758 sub config_list {
1759     my $vmlist = PVE::Cluster::get_vmlist();
1760     my $res = {};
1761     return $res if !$vmlist || !$vmlist->{ids};
1762     my $ids = $vmlist->{ids};
1763
1764     foreach my $vmid (keys %$ids) {
1765         my $d = $ids->{$vmid};
1766         next if !$d->{node} || $d->{node} ne $nodename;
1767         next if !$d->{type} || $d->{type} ne 'qemu';
1768         $res->{$vmid}->{exists} = 1;
1769     }
1770     return $res;
1771 }
1772
1773 # test if VM uses local resources (to prevent migration)
1774 sub check_local_resources {
1775     my ($conf, $noerr) = @_;
1776
1777     my $loc_res = 0;
1778
1779     $loc_res = 1 if $conf->{hostusb}; # old syntax
1780     $loc_res = 1 if $conf->{hostpci}; # old syntax
1781
1782     foreach my $k (keys %$conf) {
1783         $loc_res = 1 if $k =~ m/^(usb|hostpci|serial|parallel)\d+$/;
1784     }
1785
1786     die "VM uses local resources\n" if $loc_res && !$noerr;
1787
1788     return $loc_res;
1789 }
1790
1791 # check is used storages are available on all nodes (use by migrate)
1792 sub check_storage_availability {
1793     my ($storecfg, $conf, $node) = @_;
1794
1795     foreach_drive($conf, sub {
1796         my ($ds, $drive) = @_;
1797
1798         my $volid = $drive->{file};
1799         return if !$volid;
1800
1801         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1802         return if !$sid;
1803
1804         # check if storage is available on both nodes
1805         my $scfg = PVE::Storage::storage_check_node($storecfg, $sid);
1806         PVE::Storage::storage_check_node($storecfg, $sid, $node);
1807    });
1808 }
1809
1810 sub check_lock {
1811     my ($conf) = @_;
1812
1813     die "VM is locked ($conf->{lock})\n" if $conf->{lock};
1814 }
1815
1816 sub check_cmdline {
1817     my ($pidfile, $pid) = @_;
1818
1819     my $fh = IO::File->new("/proc/$pid/cmdline", "r");
1820     if (defined($fh)) {
1821         my $line = <$fh>;
1822         $fh->close;
1823         return undef if !$line;
1824         my @param = split(/\0/, $line);
1825
1826         my $cmd = $param[0];
1827         return if !$cmd || ($cmd !~ m|kvm$|);
1828
1829         for (my $i = 0; $i < scalar (@param); $i++) {
1830             my $p = $param[$i];
1831             next if !$p;
1832             if (($p eq '-pidfile') || ($p eq '--pidfile')) {
1833                 my $p = $param[$i+1];
1834                 return 1 if $p && ($p eq $pidfile);
1835                 return undef;
1836             }
1837         }
1838     }
1839     return undef;
1840 }
1841
1842 sub check_running {
1843     my ($vmid, $nocheck, $node) = @_;
1844
1845     my $filename = config_file($vmid, $node);
1846
1847     die "unable to find configuration file for VM $vmid - no such machine\n"
1848         if !$nocheck && ! -f $filename;
1849
1850     my $pidfile = pidfile_name($vmid);
1851
1852     if (my $fd = IO::File->new("<$pidfile")) {
1853         my $st = stat($fd);
1854         my $line = <$fd>;
1855         close($fd);
1856
1857         my $mtime = $st->mtime;
1858         if ($mtime > time()) {
1859             warn "file '$filename' modified in future\n";
1860         }
1861
1862         if ($line =~ m/^(\d+)$/) {
1863             my $pid = $1;
1864             if (check_cmdline($pidfile, $pid)) {
1865                 if (my $pinfo = PVE::ProcFSTools::check_process_running($pid)) {
1866                     return $pid;
1867                 }
1868             }
1869         }
1870     }
1871
1872     return undef;
1873 }
1874
1875 sub vzlist {
1876
1877     my $vzlist = config_list();
1878
1879     my $fd = IO::Dir->new($var_run_tmpdir) || return $vzlist;
1880
1881     while (defined(my $de = $fd->read)) {
1882         next if $de !~ m/^(\d+)\.pid$/;
1883         my $vmid = $1;
1884         next if !defined($vzlist->{$vmid});
1885         if (my $pid = check_running($vmid)) {
1886             $vzlist->{$vmid}->{pid} = $pid;
1887         }
1888     }
1889
1890     return $vzlist;
1891 }
1892
1893 sub disksize {
1894     my ($storecfg, $conf) = @_;
1895
1896     my $bootdisk = $conf->{bootdisk};
1897     return undef if !$bootdisk;
1898     return undef if !valid_drivename($bootdisk);
1899
1900     return undef if !$conf->{$bootdisk};
1901
1902     my $drive = parse_drive($bootdisk, $conf->{$bootdisk});
1903     return undef if !defined($drive);
1904
1905     return undef if drive_is_cdrom($drive);
1906
1907     my $volid = $drive->{file};
1908     return undef if !$volid;
1909
1910     return $drive->{size};
1911 }
1912
1913 my $last_proc_pid_stat;
1914
1915 # get VM status information
1916 # This must be fast and should not block ($full == false)
1917 # We only query KVM using QMP if $full == true (this can be slow)
1918 sub vmstatus {
1919     my ($opt_vmid, $full) = @_;
1920
1921     my $res = {};
1922
1923     my $storecfg = PVE::Storage::config();
1924
1925     my $list = vzlist();
1926     my ($uptime) = PVE::ProcFSTools::read_proc_uptime(1);
1927
1928     my $cpucount = $cpuinfo->{cpus} || 1;
1929
1930     foreach my $vmid (keys %$list) {
1931         next if $opt_vmid && ($vmid ne $opt_vmid);
1932
1933         my $cfspath = cfs_config_path($vmid);
1934         my $conf = PVE::Cluster::cfs_read_file($cfspath) || {};
1935
1936         my $d = {};
1937         $d->{pid} = $list->{$vmid}->{pid};
1938
1939         # fixme: better status?
1940         $d->{status} = $list->{$vmid}->{pid} ? 'running' : 'stopped';
1941
1942         my $size = disksize($storecfg, $conf);
1943         if (defined($size)) {
1944             $d->{disk} = 0; # no info available
1945             $d->{maxdisk} = $size;
1946         } else {
1947             $d->{disk} = 0;
1948             $d->{maxdisk} = 0;
1949         }
1950
1951         $d->{cpus} = ($conf->{sockets} || 1) * ($conf->{cores} || 1);
1952         $d->{cpus} = $cpucount if $d->{cpus} > $cpucount;
1953
1954         $d->{name} = $conf->{name} || "VM $vmid";
1955         $d->{maxmem} = $conf->{memory} ? $conf->{memory}*(1024*1024) : 0;
1956
1957         $d->{uptime} = 0;
1958         $d->{cpu} = 0;
1959         $d->{mem} = 0;
1960
1961         $d->{netout} = 0;
1962         $d->{netin} = 0;
1963
1964         $d->{diskread} = 0;
1965         $d->{diskwrite} = 0;
1966
1967         $res->{$vmid} = $d;
1968     }
1969
1970     my $netdev = PVE::ProcFSTools::read_proc_net_dev();
1971     foreach my $dev (keys %$netdev) {
1972         next if $dev !~ m/^tap([1-9]\d*)i/;
1973         my $vmid = $1;
1974         my $d = $res->{$vmid};
1975         next if !$d;
1976
1977         $d->{netout} += $netdev->{$dev}->{receive};
1978         $d->{netin} += $netdev->{$dev}->{transmit};
1979     }
1980
1981     my $ctime = gettimeofday;
1982
1983     foreach my $vmid (keys %$list) {
1984
1985         my $d = $res->{$vmid};
1986         my $pid = $d->{pid};
1987         next if !$pid;
1988
1989         my $pstat = PVE::ProcFSTools::read_proc_pid_stat($pid);
1990         next if !$pstat; # not running
1991
1992         my $used = $pstat->{utime} + $pstat->{stime};
1993
1994         $d->{uptime} = int(($uptime - $pstat->{starttime})/$cpuinfo->{user_hz});
1995
1996         if ($pstat->{vsize}) {
1997             $d->{mem} = int(($pstat->{rss}/$pstat->{vsize})*$d->{maxmem});
1998         }
1999
2000         my $old = $last_proc_pid_stat->{$pid};
2001         if (!$old) {
2002             $last_proc_pid_stat->{$pid} = {
2003                 time => $ctime,
2004                 used => $used,
2005                 cpu => 0,
2006             };
2007             next;
2008         }
2009
2010         my $dtime = ($ctime -  $old->{time}) * $cpucount * $cpuinfo->{user_hz};
2011
2012         if ($dtime > 1000) {
2013             my $dutime = $used -  $old->{used};
2014
2015             $d->{cpu} = (($dutime/$dtime)* $cpucount) / $d->{cpus};
2016             $last_proc_pid_stat->{$pid} = {
2017                 time => $ctime,
2018                 used => $used,
2019                 cpu => $d->{cpu},
2020             };
2021         } else {
2022             $d->{cpu} = $old->{cpu};
2023         }
2024     }
2025
2026     return $res if !$full;
2027
2028     my $qmpclient = PVE::QMPClient->new();
2029
2030     my $blockstatscb = sub {
2031         my ($vmid, $resp) = @_;
2032         my $data = $resp->{'return'} || [];
2033         my $totalrdbytes = 0;
2034         my $totalwrbytes = 0;
2035         for my $blockstat (@$data) {
2036             $totalrdbytes = $totalrdbytes + $blockstat->{stats}->{rd_bytes};
2037             $totalwrbytes = $totalwrbytes + $blockstat->{stats}->{wr_bytes};
2038         }
2039         $res->{$vmid}->{diskread} = $totalrdbytes;
2040         $res->{$vmid}->{diskwrite} = $totalwrbytes;
2041     };
2042
2043     my $statuscb = sub {
2044         my ($vmid, $resp) = @_;
2045         $qmpclient->queue_cmd($vmid, $blockstatscb, 'query-blockstats');
2046
2047         my $status = 'unknown';
2048         if (!defined($status = $resp->{'return'}->{status})) {
2049             warn "unable to get VM status\n";
2050             return;
2051         }
2052
2053         $res->{$vmid}->{qmpstatus} = $resp->{'return'}->{status};
2054     };
2055
2056     foreach my $vmid (keys %$list) {
2057         next if $opt_vmid && ($vmid ne $opt_vmid);
2058         next if !$res->{$vmid}->{pid}; # not running
2059         $qmpclient->queue_cmd($vmid, $statuscb, 'query-status');
2060     }
2061
2062     $qmpclient->queue_execute();
2063
2064     foreach my $vmid (keys %$list) {
2065         next if $opt_vmid && ($vmid ne $opt_vmid);
2066         $res->{$vmid}->{qmpstatus} = $res->{$vmid}->{status} if !$res->{$vmid}->{qmpstatus};
2067     }
2068
2069     return $res;
2070 }
2071
2072 sub foreach_drive {
2073     my ($conf, $func) = @_;
2074
2075     foreach my $ds (keys %$conf) {
2076         next if !valid_drivename($ds);
2077
2078         my $drive = parse_drive($ds, $conf->{$ds});
2079         next if !$drive;
2080
2081         &$func($ds, $drive);
2082     }
2083 }
2084
2085 sub foreach_volid {
2086     my ($conf, $func) = @_;
2087
2088     my $volhash = {};
2089
2090     my $test_volid = sub {
2091         my ($volid, $is_cdrom) = @_;
2092
2093         return if !$volid;
2094
2095         $volhash->{$volid} = $is_cdrom || 0;
2096     };
2097
2098     PVE::QemuServer::foreach_drive($conf, sub {
2099         my ($ds, $drive) = @_;
2100         &$test_volid($drive->{file}, drive_is_cdrom($drive));
2101     });
2102
2103     foreach my $snapname (keys %{$conf->{snapshots}}) {
2104         my $snap = $conf->{snapshots}->{$snapname};
2105         &$test_volid($snap->{vmstate}, 0);
2106         PVE::QemuServer::foreach_drive($snap, sub {
2107             my ($ds, $drive) = @_;
2108             &$test_volid($drive->{file}, drive_is_cdrom($drive));
2109         });
2110     }
2111
2112     foreach my $volid (keys %$volhash) {
2113         &$func($volid, $volhash->{$volid});
2114     }
2115 }
2116
2117 sub config_to_command {
2118     my ($storecfg, $vmid, $conf, $defaults) = @_;
2119
2120     my $cmd = [];
2121     my $devices = [];
2122     my $pciaddr = '';
2123     my $bridges = {};
2124     my $kvmver = kvm_user_version();
2125     my $vernum = 0; # unknown
2126     if ($kvmver =~ m/^(\d+)\.(\d+)$/) {
2127         $vernum = $1*1000000+$2*1000;
2128     } elsif ($kvmver =~ m/^(\d+)\.(\d+)\.(\d+)$/) {
2129         $vernum = $1*1000000+$2*1000+$3;
2130     }
2131
2132     die "detected old qemu-kvm binary ($kvmver)\n" if $vernum < 15000;
2133
2134     my $have_ovz = -f '/proc/vz/vestat';
2135
2136     push @$cmd, '/usr/bin/kvm';
2137
2138     push @$cmd, '-id', $vmid;
2139
2140     my $use_virtio = 0;
2141
2142     my $qmpsocket = qmp_socket($vmid);
2143     push @$cmd, '-chardev', "socket,id=qmp,path=$qmpsocket,server,nowait";
2144     push @$cmd, '-mon', "chardev=qmp,mode=control";
2145
2146     my $socket = vnc_socket($vmid);
2147     push @$cmd,  '-vnc', "unix:$socket,x509,password";
2148
2149     push @$cmd, '-pidfile' , pidfile_name($vmid);
2150
2151     push @$cmd, '-daemonize';
2152
2153     my $use_usb2 = 0;
2154     for (my $i = 0; $i < $MAX_USB_DEVICES; $i++)  {
2155         next if !$conf->{"usb$i"};
2156         $use_usb2 = 1;
2157     }
2158     # include usb device config
2159     push @$devices, '-readconfig', '/usr/share/qemu-server/pve-usb.cfg' if $use_usb2;
2160
2161     # enable absolute mouse coordinates (needed by vnc)
2162     my $tablet = defined($conf->{tablet}) ? $conf->{tablet} : $defaults->{tablet};
2163     if ($tablet) {
2164         if ($use_usb2) {
2165             push @$devices, '-device', 'usb-tablet,bus=ehci.0,port=6';
2166         } else {
2167             push @$devices, '-usbdevice', 'tablet';
2168         }
2169     }
2170
2171     # host pci devices
2172     for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
2173           my $d = parse_hostpci($conf->{"hostpci$i"});
2174           next if !$d;
2175           $pciaddr = print_pci_addr("hostpci$i", $bridges);
2176           push @$devices, '-device', "pci-assign,host=$d->{pciid},id=hostpci$i$pciaddr";
2177     }
2178
2179     # usb devices
2180     for (my $i = 0; $i < $MAX_USB_DEVICES; $i++)  {
2181         my $d = parse_usb_device($conf->{"usb$i"});
2182         next if !$d;
2183         if ($d->{vendorid} && $d->{productid}) {
2184             push @$devices, '-device', "usb-host,vendorid=0x$d->{vendorid},productid=0x$d->{productid}";
2185         } elsif (defined($d->{hostbus}) && defined($d->{hostport})) {
2186             push @$devices, '-device', "usb-host,hostbus=$d->{hostbus},hostport=$d->{hostport}";
2187         }
2188     }
2189
2190     # serial devices
2191     for (my $i = 0; $i < $MAX_SERIAL_PORTS; $i++)  {
2192         if (my $path = $conf->{"serial$i"}) {
2193             die "no such serial device\n" if ! -c $path;
2194             push @$devices, '-chardev', "tty,id=serial$i,path=$path";
2195             push @$devices, '-device', "isa-serial,chardev=serial$i";
2196         }
2197     }
2198
2199     # parallel devices
2200     for (my $i = 0; $i < $MAX_PARALLEL_PORTS; $i++)  {
2201         if (my $path = $conf->{"parallel$i"}) {
2202             die "no such parallel device\n" if ! -c $path;
2203             push @$devices, '-chardev', "parport,id=parallel$i,path=$path";
2204             push @$devices, '-device', "isa-parallel,chardev=parallel$i";
2205         }
2206     }
2207
2208     my $vmname = $conf->{name} || "vm$vmid";
2209
2210     push @$cmd, '-name', $vmname;
2211
2212     my $sockets = 1;
2213     $sockets = $conf->{smp} if $conf->{smp}; # old style - no longer iused
2214     $sockets = $conf->{sockets} if  $conf->{sockets};
2215
2216     my $cores = $conf->{cores} || 1;
2217
2218     push @$cmd, '-smp', "sockets=$sockets,cores=$cores";
2219
2220     push @$cmd, '-cpu', $conf->{cpu} if $conf->{cpu};
2221
2222     push @$cmd, '-nodefaults';
2223
2224     my $bootorder = $conf->{boot} || $confdesc->{boot}->{default};
2225
2226     my $bootindex_hash = {};
2227     my $i = 1;
2228     foreach my $o (split(//, $bootorder)) {
2229         $bootindex_hash->{$o} = $i*100;
2230         $i++;
2231     }
2232
2233     push @$cmd, '-boot', "menu=on";
2234
2235     push @$cmd, '-no-acpi' if defined($conf->{acpi}) && $conf->{acpi} == 0;
2236
2237     push @$cmd, '-no-reboot' if  defined($conf->{reboot}) && $conf->{reboot} == 0;
2238
2239     my $vga = $conf->{vga};
2240     if (!$vga) {
2241         if ($conf->{ostype} && ($conf->{ostype} eq 'win7' || $conf->{ostype} eq 'w2k8')) {
2242             $vga = 'std';
2243         } else {
2244             $vga = 'cirrus';
2245         }
2246     }
2247
2248     push @$cmd, '-vga', $vga if $vga; # for kvm 77 and later
2249
2250     # time drift fix
2251     my $tdf = defined($conf->{tdf}) ? $conf->{tdf} : $defaults->{tdf};
2252     # ignore - no longer supported by newer kvm
2253     # push @$cmd, '-tdf' if $tdf;
2254
2255     my $nokvm = defined($conf->{kvm}) && $conf->{kvm} == 0 ? 1 : 0;
2256
2257     if (my $ost = $conf->{ostype}) {
2258         # other, wxp, w2k, w2k3, w2k8, wvista, win7, l24, l26
2259
2260         if ($ost =~ m/^w/) { # windows
2261             push @$cmd, '-localtime' if !defined($conf->{localtime});
2262
2263             # use rtc-td-hack when acpi is enabled
2264             if (!(defined($conf->{acpi}) && $conf->{acpi} == 0)) {
2265                 push @$cmd, '-rtc-td-hack';
2266             }
2267         }
2268
2269         if ($ost eq 'win7' || $ost eq 'w2k8' || $ost eq 'wvista') {
2270             push @$cmd, '-no-kvm-pit-reinjection';
2271             push @$cmd, '-no-hpet';
2272         }
2273
2274         # -tdf ?
2275         # -no-acpi
2276         # -no-kvm
2277         # -win2k-hack ?
2278     }
2279
2280     if ($nokvm) {
2281         push @$cmd, '-no-kvm';
2282     } else {
2283         die "No accelerator found!\n" if !$cpuinfo->{hvm};
2284     }
2285
2286     push @$cmd, '-localtime' if $conf->{localtime};
2287
2288     push @$cmd, '-startdate', $conf->{startdate} if $conf->{startdate};
2289
2290     push @$cmd, '-S' if $conf->{freeze};
2291
2292     # set keyboard layout
2293     my $kb = $conf->{keyboard} || $defaults->{keyboard};
2294     push @$cmd, '-k', $kb if $kb;
2295
2296     # enable sound
2297     #my $soundhw = $conf->{soundhw} || $defaults->{soundhw};
2298     #push @$cmd, '-soundhw', 'es1370';
2299     #push @$cmd, '-soundhw', $soundhw if $soundhw;
2300
2301     if($conf->{agent}) {
2302         my $qgasocket = qga_socket($vmid);
2303         my $pciaddr = print_pci_addr("qga0", $bridges);
2304         push @$devices, '-chardev', "socket,path=$qgasocket,server,nowait,id=qga0";
2305         push @$devices, '-device', "virtio-serial,id=qga0$pciaddr";
2306         push @$devices, '-device', 'virtserialport,chardev=qga0,name=org.qemu.guest_agent.0';
2307     }
2308
2309     $pciaddr = print_pci_addr("balloon0", $bridges);
2310     push @$devices, '-device', "virtio-balloon-pci,id=balloon0$pciaddr" if $conf->{balloon};
2311
2312     if ($conf->{watchdog}) {
2313         my $wdopts = parse_watchdog($conf->{watchdog});
2314         $pciaddr = print_pci_addr("watchdog", $bridges);
2315         my $watchdog = $wdopts->{model} || 'i6300esb';
2316         push @$devices, '-device', "$watchdog$pciaddr";
2317         push @$devices, '-watchdog-action', $wdopts->{action} if $wdopts->{action};
2318     }
2319
2320     my $vollist = [];
2321     my $scsicontroller = {};
2322     my $ahcicontroller = {};
2323     my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : $defaults->{scsihw};
2324
2325     foreach_drive($conf, sub {
2326         my ($ds, $drive) = @_;
2327
2328         if (PVE::Storage::parse_volume_id($drive->{file}, 1)) {
2329             push @$vollist, $drive->{file};
2330         }
2331
2332         $use_virtio = 1 if $ds =~ m/^virtio/;
2333
2334         if (drive_is_cdrom ($drive)) {
2335             if ($bootindex_hash->{d}) {
2336                 $drive->{bootindex} = $bootindex_hash->{d};
2337                 $bootindex_hash->{d} += 1;
2338             }
2339         } else {
2340             if ($bootindex_hash->{c}) {
2341                 $drive->{bootindex} = $bootindex_hash->{c} if $conf->{bootdisk} && ($conf->{bootdisk} eq $ds);
2342                 $bootindex_hash->{c} += 1;
2343             }
2344         }
2345
2346         if ($drive->{interface} eq 'scsi') {
2347
2348             my $maxdev = ($scsihw ne 'lsi') ? 256 : 7;
2349             my $controller = int($drive->{index} / $maxdev);
2350             $pciaddr = print_pci_addr("scsihw$controller", $bridges);
2351             push @$devices, '-device', "$scsihw,id=scsihw$controller$pciaddr" if !$scsicontroller->{$controller};
2352             $scsicontroller->{$controller}=1;
2353         }
2354
2355         if ($drive->{interface} eq 'sata') {
2356            my $controller = int($drive->{index} / $MAX_SATA_DISKS);
2357            $pciaddr = print_pci_addr("ahci$controller", $bridges);
2358            push @$devices, '-device', "ahci,id=ahci$controller,multifunction=on$pciaddr" if !$ahcicontroller->{$controller};
2359            $ahcicontroller->{$controller}=1;
2360         }
2361
2362         push @$devices, '-drive',print_drive_full($storecfg, $vmid, $drive);
2363         push @$devices, '-device',print_drivedevice_full($storecfg, $conf, $vmid, $drive, $bridges);
2364     });
2365
2366     push @$cmd, '-m', $conf->{memory} || $defaults->{memory};
2367
2368     for (my $i = 0; $i < $MAX_NETS; $i++) {
2369          next if !$conf->{"net$i"};
2370          my $d = parse_net($conf->{"net$i"});
2371          next if !$d;
2372
2373          $use_virtio = 1 if $d->{model} eq 'virtio';
2374
2375          if ($bootindex_hash->{n}) {
2376             $d->{bootindex} = $bootindex_hash->{n};
2377             $bootindex_hash->{n} += 1;
2378          }
2379
2380          my $netdevfull = print_netdev_full($vmid,$conf,$d,"net$i");
2381          push @$devices, '-netdev', $netdevfull;
2382
2383          my $netdevicefull = print_netdevice_full($vmid,$conf,$d,"net$i",$bridges);
2384          push @$devices, '-device', $netdevicefull;
2385     }
2386
2387     #bridges
2388     while (my ($k, $v) = each %$bridges) {
2389         $pciaddr = print_pci_addr("pci.$k");
2390         unshift @$devices, '-device', "pci-bridge,id=pci.$k,chassis_nr=$k$pciaddr" if $k > 0;
2391     }
2392
2393
2394     # hack: virtio with fairsched is unreliable, so we do not use fairsched
2395     # when the VM uses virtio devices.
2396     if (!$use_virtio && $have_ovz) {
2397
2398         my $cpuunits = defined($conf->{cpuunits}) ?
2399             $conf->{cpuunits} : $defaults->{cpuunits};
2400
2401         push @$cmd, '-cpuunits', $cpuunits if $cpuunits;
2402
2403         # fixme: cpulimit is currently ignored
2404         #push @$cmd, '-cpulimit', $conf->{cpulimit} if $conf->{cpulimit};
2405     }
2406
2407     # add custom args
2408     if ($conf->{args}) {
2409         my $aa = PVE::Tools::split_args($conf->{args});
2410         push @$cmd, @$aa;
2411     }
2412
2413     push @$cmd, @$devices;
2414     return wantarray ? ($cmd, $vollist) : $cmd;
2415 }
2416
2417 sub vnc_socket {
2418     my ($vmid) = @_;
2419     return "${var_run_tmpdir}/$vmid.vnc";
2420 }
2421
2422 sub qmp_socket {
2423     my ($vmid) = @_;
2424     return "${var_run_tmpdir}/$vmid.qmp";
2425 }
2426
2427 sub qga_socket {
2428     my ($vmid) = @_;
2429     return "${var_run_tmpdir}/$vmid.qga";
2430 }
2431
2432 sub pidfile_name {
2433     my ($vmid) = @_;
2434     return "${var_run_tmpdir}/$vmid.pid";
2435 }
2436
2437 sub next_migrate_port {
2438
2439     for (my $p = 60000; $p < 60010; $p++) {
2440
2441         my $sock = IO::Socket::INET->new(Listen => 5,
2442                                          LocalAddr => 'localhost',
2443                                          LocalPort => $p,
2444                                          ReuseAddr => 1,
2445                                          Proto     => 0);
2446
2447         if ($sock) {
2448             close($sock);
2449             return $p;
2450         }
2451     }
2452
2453     die "unable to find free migration port";
2454 }
2455
2456 sub vm_devices_list {
2457     my ($vmid) = @_;
2458
2459     my $res = vm_mon_cmd($vmid, 'query-pci');
2460
2461     my $devices = {};
2462     foreach my $pcibus (@$res) {
2463         foreach my $device (@{$pcibus->{devices}}) {
2464             next if !$device->{'qdev_id'};
2465             $devices->{$device->{'qdev_id'}} = $device;
2466         }
2467     }
2468
2469     return $devices;
2470 }
2471
2472 sub vm_deviceplug {
2473     my ($storecfg, $conf, $vmid, $deviceid, $device) = @_;
2474
2475     return 1 if !check_running($vmid) || !$conf->{hotplug};
2476
2477     my $devices_list = vm_devices_list($vmid);
2478     return 1 if defined($devices_list->{$deviceid});
2479
2480     qemu_bridgeadd($storecfg, $conf, $vmid, $deviceid); #add bridge if we need it for the device
2481
2482     if ($deviceid =~ m/^(virtio)(\d+)$/) {
2483         return undef if !qemu_driveadd($storecfg, $vmid, $device);
2484         my $devicefull = print_drivedevice_full($storecfg, $conf, $vmid, $device);
2485         qemu_deviceadd($vmid, $devicefull);
2486         if(!qemu_deviceaddverify($vmid, $deviceid)) {
2487            qemu_drivedel($vmid, $deviceid);
2488            return undef;
2489         }
2490     }
2491
2492     if ($deviceid =~ m/^(scsihw)(\d+)$/) {
2493         my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : "lsi";
2494         my $pciaddr = print_pci_addr($deviceid);
2495         my $devicefull = "$scsihw,id=$deviceid$pciaddr";
2496         qemu_deviceadd($vmid, $devicefull);
2497         return undef if(!qemu_deviceaddverify($vmid, $deviceid));
2498     }
2499
2500     if ($deviceid =~ m/^(scsi)(\d+)$/) {
2501         return 1 if ($conf->{scsihw} && $conf->{scsihw} ne 'lsi'); #virtio-scsi not yet support hotplug
2502         return undef if !qemu_findorcreatescsihw($storecfg,$conf, $vmid, $device);
2503         return undef if !qemu_driveadd($storecfg, $vmid, $device);
2504         my $devicefull = print_drivedevice_full($storecfg, $conf, $vmid, $device);
2505         if(!qemu_deviceadd($vmid, $devicefull)) {
2506            qemu_drivedel($vmid, $deviceid);
2507            return undef;
2508         }
2509     }
2510
2511     if ($deviceid =~ m/^(net)(\d+)$/) {
2512         return undef if !qemu_netdevadd($vmid, $conf, $device, $deviceid);
2513         my $netdevicefull = print_netdevice_full($vmid, $conf, $device, $deviceid);
2514         qemu_deviceadd($vmid, $netdevicefull);
2515         if(!qemu_deviceaddverify($vmid, $deviceid)) {
2516            qemu_netdevdel($vmid, $deviceid);
2517            return undef;
2518         }
2519     }
2520
2521     if ($deviceid =~ m/^(pci\.)(\d+)$/) {
2522         my $bridgeid = $2;
2523         my $pciaddr = print_pci_addr($deviceid);
2524         my $devicefull = "pci-bridge,id=pci.$bridgeid,chassis_nr=$bridgeid$pciaddr";
2525         qemu_deviceadd($vmid, $devicefull);
2526         return undef if !qemu_deviceaddverify($vmid, $deviceid);
2527     }
2528
2529     return 1;
2530 }
2531
2532 sub vm_deviceunplug {
2533     my ($vmid, $conf, $deviceid) = @_;
2534
2535     return 1 if !check_running ($vmid) || !$conf->{hotplug};
2536
2537     my $devices_list = vm_devices_list($vmid);
2538     return 1 if !defined($devices_list->{$deviceid});
2539
2540     die "can't unplug bootdisk" if $conf->{bootdisk} && $conf->{bootdisk} eq $deviceid;
2541
2542     if ($deviceid =~ m/^(virtio)(\d+)$/) {
2543         return undef if !qemu_drivedel($vmid, $deviceid);
2544         qemu_devicedel($vmid, $deviceid);
2545         return undef if !qemu_devicedelverify($vmid, $deviceid);
2546     }
2547
2548     if ($deviceid =~ m/^(lsi)(\d+)$/) {
2549         return undef if !qemu_devicedel($vmid, $deviceid);
2550     }
2551
2552     if ($deviceid =~ m/^(scsi)(\d+)$/) {
2553         return undef if !qemu_devicedel($vmid, $deviceid);
2554         return undef if !qemu_drivedel($vmid, $deviceid);
2555     }
2556
2557     if ($deviceid =~ m/^(net)(\d+)$/) {
2558         return undef if !qemu_netdevdel($vmid, $deviceid);
2559         qemu_devicedel($vmid, $deviceid);
2560         return undef if !qemu_devicedelverify($vmid, $deviceid);
2561     }
2562
2563     return 1;
2564 }
2565
2566 sub qemu_deviceadd {
2567     my ($vmid, $devicefull) = @_;
2568
2569     my $ret = vm_human_monitor_command($vmid, "device_add $devicefull");
2570     $ret =~ s/^\s+//;
2571     # Otherwise, if the command succeeds, no output is sent. So any non-empty string shows an error
2572     return 1 if $ret eq "";
2573     syslog("err", "error on hotplug device : $ret");
2574     return undef;
2575
2576 }
2577
2578 sub qemu_devicedel {
2579     my($vmid, $deviceid) = @_;
2580
2581     my $ret = vm_human_monitor_command($vmid, "device_del $deviceid");
2582     $ret =~ s/^\s+//;
2583     return 1 if $ret eq "";
2584     syslog("err", "detaching device $deviceid failed : $ret");
2585     return undef;
2586 }
2587
2588 sub qemu_driveadd {
2589     my($storecfg, $vmid, $device) = @_;
2590
2591     my $drive = print_drive_full($storecfg, $vmid, $device);
2592     my $ret = vm_human_monitor_command($vmid, "drive_add auto $drive");
2593     # If the command succeeds qemu prints: "OK"
2594     if ($ret !~ m/OK/s) {
2595         syslog("err", "adding drive failed: $ret");
2596         return undef;
2597     }
2598     return 1;
2599 }
2600
2601 sub qemu_drivedel {
2602     my($vmid, $deviceid) = @_;
2603
2604     my $ret = vm_human_monitor_command($vmid, "drive_del drive-$deviceid");
2605     $ret =~ s/^\s+//;
2606     if ($ret =~ m/Device \'.*?\' not found/s) {
2607         # NB: device not found errors mean the drive was auto-deleted and we ignore the error
2608     }
2609     elsif ($ret ne "") {
2610       syslog("err", "deleting drive $deviceid failed : $ret");
2611       return undef;
2612     }
2613     return 1;
2614 }
2615
2616 sub qemu_deviceaddverify {
2617     my ($vmid,$deviceid) = @_;
2618
2619     for (my $i = 0; $i <= 5; $i++) {
2620          my $devices_list = vm_devices_list($vmid);
2621          return 1 if defined($devices_list->{$deviceid});
2622          sleep 1;
2623     }
2624     syslog("err", "error on hotplug device $deviceid");
2625     return undef;
2626 }
2627
2628
2629 sub qemu_devicedelverify {
2630     my ($vmid,$deviceid) = @_;
2631
2632     #need to verify the device is correctly remove as device_del is async and empty return is not reliable
2633     for (my $i = 0; $i <= 5; $i++) {
2634          my $devices_list = vm_devices_list($vmid);
2635          return 1 if !defined($devices_list->{$deviceid});
2636          sleep 1;
2637     }
2638     syslog("err", "error on hot-unplugging device $deviceid");
2639     return undef;
2640 }
2641
2642 sub qemu_findorcreatescsihw {
2643     my ($storecfg, $conf, $vmid, $device) = @_;
2644
2645     my $maxdev = ($conf->{scsihw} && $conf->{scsihw} ne 'lsi') ? 256 : 7;
2646     my $controller = int($device->{index} / $maxdev);
2647     my $scsihwid="scsihw$controller";
2648     my $devices_list = vm_devices_list($vmid);
2649
2650     if(!defined($devices_list->{$scsihwid})) {
2651        return undef if !vm_deviceplug($storecfg, $conf, $vmid, $scsihwid);
2652     }
2653     return 1;
2654 }
2655
2656 sub qemu_bridgeadd {
2657     my ($storecfg, $conf, $vmid, $device) = @_;
2658
2659     my $bridges = {};
2660     my $bridgeid = undef;
2661     print_pci_addr($device, $bridges);
2662
2663     while (my ($k, $v) = each %$bridges) {
2664         $bridgeid = $k;
2665     }
2666     return if $bridgeid < 1;
2667     my $bridge = "pci.$bridgeid";
2668     my $devices_list = vm_devices_list($vmid);
2669
2670     if(!defined($devices_list->{$bridge})) {
2671         return undef if !vm_deviceplug($storecfg, $conf, $vmid, $bridge);
2672     }
2673     return 1;
2674 }
2675
2676 sub qemu_netdevadd {
2677     my ($vmid, $conf, $device, $deviceid) = @_;
2678
2679     my $netdev = print_netdev_full($vmid, $conf, $device, $deviceid);
2680     my $ret = vm_human_monitor_command($vmid, "netdev_add $netdev");
2681     $ret =~ s/^\s+//;
2682
2683     #if the command succeeds, no output is sent. So any non-empty string shows an error
2684     return 1 if $ret eq "";
2685     syslog("err", "adding netdev failed: $ret");
2686     return undef;
2687 }
2688
2689 sub qemu_netdevdel {
2690     my ($vmid, $deviceid) = @_;
2691
2692     my $ret = vm_human_monitor_command($vmid, "netdev_del $deviceid");
2693     $ret =~ s/^\s+//;
2694     #if the command succeeds, no output is sent. So any non-empty string shows an error
2695     return 1 if $ret eq "";
2696     syslog("err", "deleting netdev failed: $ret");
2697     return undef;
2698 }
2699
2700 sub qemu_block_set_io_throttle {
2701     my ($vmid, $deviceid, $bps, $bps_rd, $bps_wr, $iops, $iops_rd, $iops_wr) = @_;
2702
2703     return if !check_running($vmid) ;
2704
2705     $bps = 0 if !$bps;
2706     $bps_rd = 0 if !$bps_rd;
2707     $bps_wr = 0 if !$bps_wr;
2708     $iops = 0 if !$iops;
2709     $iops_rd = 0 if !$iops_rd;
2710     $iops_wr = 0 if !$iops_wr;
2711
2712     vm_mon_cmd($vmid, "block_set_io_throttle", device => $deviceid, bps => int($bps), bps_rd => int($bps_rd), bps_wr => int($bps_wr), iops => int($iops), iops_rd => int($iops_rd), iops_wr => int($iops_wr));
2713
2714 }
2715
2716 # old code, only used to shutdown old VM after update
2717 sub __read_avail {
2718     my ($fh, $timeout) = @_;
2719
2720     my $sel = new IO::Select;
2721     $sel->add($fh);
2722
2723     my $res = '';
2724     my $buf;
2725
2726     my @ready;
2727     while (scalar (@ready = $sel->can_read($timeout))) {
2728         my $count;
2729         if ($count = $fh->sysread($buf, 8192)) {
2730             if ($buf =~ /^(.*)\(qemu\) $/s) {
2731                 $res .= $1;
2732                 last;
2733             } else {
2734                 $res .= $buf;
2735             }
2736         } else {
2737             if (!defined($count)) {
2738                 die "$!\n";
2739             }
2740             last;
2741         }
2742     }
2743
2744     die "monitor read timeout\n" if !scalar(@ready);
2745
2746     return $res;
2747 }
2748
2749 # old code, only used to shutdown old VM after update
2750 sub vm_monitor_command {
2751     my ($vmid, $cmdstr, $nocheck) = @_;
2752
2753     my $res;
2754
2755     eval {
2756         die "VM $vmid not running\n" if !check_running($vmid, $nocheck);
2757
2758         my $sname = "${var_run_tmpdir}/$vmid.mon";
2759
2760         my $sock = IO::Socket::UNIX->new( Peer => $sname ) ||
2761             die "unable to connect to VM $vmid socket - $!\n";
2762
2763         my $timeout = 3;
2764
2765         # hack: migrate sometime blocks the monitor (when migrate_downtime
2766         # is set)
2767         if ($cmdstr =~ m/^(info\s+migrate|migrate\s)/) {
2768             $timeout = 60*60; # 1 hour
2769         }
2770
2771         # read banner;
2772         my $data = __read_avail($sock, $timeout);
2773
2774         if ($data !~ m/^QEMU\s+(\S+)\s+monitor\s/) {
2775             die "got unexpected qemu monitor banner\n";
2776         }
2777
2778         my $sel = new IO::Select;
2779         $sel->add($sock);
2780
2781         if (!scalar(my @ready = $sel->can_write($timeout))) {
2782             die "monitor write error - timeout";
2783         }
2784
2785         my $fullcmd = "$cmdstr\r";
2786
2787         # syslog('info', "VM $vmid monitor command: $cmdstr");
2788
2789         my $b;
2790         if (!($b = $sock->syswrite($fullcmd)) || ($b != length($fullcmd))) {
2791             die "monitor write error - $!";
2792         }
2793
2794         return if ($cmdstr eq 'q') || ($cmdstr eq 'quit');
2795
2796         $timeout = 20;
2797
2798         if ($cmdstr =~ m/^(info\s+migrate|migrate\s)/) {
2799             $timeout = 60*60; # 1 hour
2800         } elsif ($cmdstr =~ m/^(eject|change)/) {
2801             $timeout = 60; # note: cdrom mount command is slow
2802         }
2803         if ($res = __read_avail($sock, $timeout)) {
2804
2805             my @lines = split("\r?\n", $res);
2806
2807             shift @lines if $lines[0] !~ m/^unknown command/; # skip echo
2808
2809             $res = join("\n", @lines);
2810             $res .= "\n";
2811         }
2812     };
2813
2814     my $err = $@;
2815
2816     if ($err) {
2817         syslog("err", "VM $vmid monitor command failed - $err");
2818         die $err;
2819     }
2820
2821     return $res;
2822 }
2823
2824 sub qemu_block_resize {
2825     my ($vmid, $deviceid, $storecfg, $volid, $size) = @_;
2826
2827     my $running = PVE::QemuServer::check_running($vmid);
2828
2829     return if !PVE::Storage::volume_resize($storecfg, $volid, $size, $running);
2830
2831     return if !$running;
2832
2833     vm_mon_cmd($vmid, "block_resize", device => $deviceid, size => int($size));
2834
2835 }
2836
2837 sub qemu_volume_snapshot {
2838     my ($vmid, $deviceid, $storecfg, $volid, $snap) = @_;
2839
2840     my $running = PVE::QemuServer::check_running($vmid);
2841
2842     return if !PVE::Storage::volume_snapshot($storecfg, $volid, $snap, $running);
2843
2844     return if !$running;
2845
2846     vm_mon_cmd($vmid, "snapshot-drive", device => $deviceid, name => $snap);
2847
2848 }
2849
2850 sub qemu_volume_snapshot_delete {
2851     my ($vmid, $deviceid, $storecfg, $volid, $snap) = @_;
2852
2853     my $running = PVE::QemuServer::check_running($vmid);
2854
2855     return if !PVE::Storage::volume_snapshot_delete($storecfg, $volid, $snap, $running);
2856
2857     return if !$running;
2858
2859     vm_mon_cmd($vmid, "delete-drive-snapshot", device => $deviceid, name => $snap);
2860 }
2861
2862 sub qga_freezefs {
2863     my ($vmid) = @_;
2864
2865     #need to impplement call to qemu-ga
2866 }
2867
2868 sub qga_unfreezefs {
2869     my ($vmid) = @_;
2870
2871     #need to impplement call to qemu-ga
2872 }
2873
2874 sub vm_start {
2875     my ($storecfg, $vmid, $statefile, $skiplock, $migratedfrom) = @_;
2876
2877     lock_config($vmid, sub {
2878         my $conf = load_config($vmid, $migratedfrom);
2879
2880         check_lock($conf) if !$skiplock;
2881
2882         die "VM $vmid already running\n" if check_running($vmid, undef, $migratedfrom);
2883
2884         my $defaults = load_defaults();
2885
2886         # set environment variable useful inside network script
2887         $ENV{PVE_MIGRATED_FROM} = $migratedfrom if $migratedfrom;
2888
2889         my ($cmd, $vollist) = config_to_command($storecfg, $vmid, $conf, $defaults);
2890
2891         my $migrate_port = 0;
2892
2893         if ($statefile) {
2894             if ($statefile eq 'tcp') {
2895                 $migrate_port = next_migrate_port();
2896                 my $migrate_uri = "tcp:localhost:${migrate_port}";
2897                 push @$cmd, '-incoming', $migrate_uri;
2898                 push @$cmd, '-S';
2899             } else {
2900                 push @$cmd, '-loadstate', $statefile;
2901             }
2902         }
2903
2904         # host pci devices
2905         for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
2906           my $d = parse_hostpci($conf->{"hostpci$i"});
2907           next if !$d;
2908           my $info = pci_device_info("0000:$d->{pciid}");
2909           die "IOMMU not present\n" if !check_iommu_support();
2910           die "no pci device info for device '$d->{pciid}'\n" if !$info;
2911           die "can't unbind pci device '$d->{pciid}'\n" if !pci_dev_bind_to_stub($info);
2912           die "can't reset pci device '$d->{pciid}'\n" if !pci_dev_reset($info);
2913         }
2914
2915         PVE::Storage::activate_volumes($storecfg, $vollist);
2916
2917         eval  { run_command($cmd, timeout => $statefile ? undef : 30); };
2918         my $err = $@;
2919         die "start failed: $err" if $err;
2920
2921         print "migration listens on port $migrate_port\n" if $migrate_port;
2922
2923         if ($statefile && $statefile ne 'tcp')  {
2924             eval { vm_mon_cmd($vmid, "cont"); };
2925             warn $@ if $@;
2926         }
2927
2928         # always set migrate speed (overwrite kvm default of 32m)
2929         # we set a very hight default of 8192m which is basically unlimited
2930         my $migrate_speed = $defaults->{migrate_speed} || 8192;
2931         $migrate_speed = $conf->{migrate_speed} || $migrate_speed;
2932         $migrate_speed = $migrate_speed * 1048576;
2933         eval {
2934             vm_mon_cmd($vmid, "migrate_set_speed", value => $migrate_speed);
2935         };
2936
2937         my $migrate_downtime = $defaults->{migrate_downtime};
2938         $migrate_downtime = $conf->{migrate_downtime} if defined($conf->{migrate_downtime});
2939         if (defined($migrate_downtime)) {
2940             eval { vm_mon_cmd($vmid, "migrate_set_downtime", value => $migrate_downtime); };
2941         }
2942
2943         if($migratedfrom) {
2944             my $capabilities = {};
2945             $capabilities->{capability} =  "xbzrle";
2946             $capabilities->{state} = JSON::true;
2947             eval { PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "migrate-set-capabilities", capabilities => [$capabilities]); };
2948         }
2949
2950         vm_balloonset($vmid, $conf->{balloon}) if $conf->{balloon};
2951
2952     });
2953 }
2954
2955 sub vm_mon_cmd {
2956     my ($vmid, $execute, %params) = @_;
2957
2958     my $cmd = { execute => $execute, arguments => \%params };
2959     vm_qmp_command($vmid, $cmd);
2960 }
2961
2962 sub vm_mon_cmd_nocheck {
2963     my ($vmid, $execute, %params) = @_;
2964
2965     my $cmd = { execute => $execute, arguments => \%params };
2966     vm_qmp_command($vmid, $cmd, 1);
2967 }
2968
2969 sub vm_qmp_command {
2970     my ($vmid, $cmd, $nocheck) = @_;
2971
2972     my $res;
2973
2974     my $timeout;
2975     if ($cmd->{arguments} && $cmd->{arguments}->{timeout}) {
2976         $timeout = $cmd->{arguments}->{timeout};
2977         delete $cmd->{arguments}->{timeout};
2978     }
2979
2980     eval {
2981         die "VM $vmid not running\n" if !check_running($vmid, $nocheck);
2982         my $sname = PVE::QemuServer::qmp_socket($vmid);
2983         if (-e $sname) {
2984             my $qmpclient = PVE::QMPClient->new();
2985
2986             $res = $qmpclient->cmd($vmid, $cmd, $timeout);
2987         } elsif (-e "${var_run_tmpdir}/$vmid.mon") {
2988             die "can't execute complex command on old monitor - stop/start your vm to fix the problem\n"
2989                 if scalar(%{$cmd->{arguments}});
2990             vm_monitor_command($vmid, $cmd->{execute}, $nocheck);
2991         } else {
2992             die "unable to open monitor socket\n";
2993         }
2994     };
2995     if (my $err = $@) {
2996         syslog("err", "VM $vmid qmp command failed - $err");
2997         die $err;
2998     }
2999
3000     return $res;
3001 }
3002
3003 sub vm_human_monitor_command {
3004     my ($vmid, $cmdline) = @_;
3005
3006     my $res;
3007
3008     my $cmd = {
3009         execute => 'human-monitor-command',
3010         arguments => { 'command-line' => $cmdline},
3011     };
3012
3013     return vm_qmp_command($vmid, $cmd);
3014 }
3015
3016 sub vm_commandline {
3017     my ($storecfg, $vmid) = @_;
3018
3019     my $conf = load_config($vmid);
3020
3021     my $defaults = load_defaults();
3022
3023     my $cmd = config_to_command($storecfg, $vmid, $conf, $defaults);
3024
3025     return join(' ', @$cmd);
3026 }
3027
3028 sub vm_reset {
3029     my ($vmid, $skiplock) = @_;
3030
3031     lock_config($vmid, sub {
3032
3033         my $conf = load_config($vmid);
3034
3035         check_lock($conf) if !$skiplock;
3036
3037         vm_mon_cmd($vmid, "system_reset");
3038     });
3039 }
3040
3041 sub get_vm_volumes {
3042     my ($conf) = @_;
3043
3044     my $vollist = [];
3045     foreach_volid($conf, sub {
3046         my ($volid, $is_cdrom) = @_;
3047
3048         return if $volid =~ m|^/|;
3049
3050         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
3051         return if !$sid;
3052
3053         push @$vollist, $volid;
3054     });
3055
3056     return $vollist;
3057 }
3058
3059 sub vm_stop_cleanup {
3060     my ($storecfg, $vmid, $conf, $keepActive) = @_;
3061
3062     eval {
3063         fairsched_rmnod($vmid); # try to destroy group
3064
3065         if (!$keepActive) {
3066             my $vollist = get_vm_volumes($conf);
3067             PVE::Storage::deactivate_volumes($storecfg, $vollist);
3068         }
3069
3070         foreach my $ext (qw(mon qmp pid vnc qga)) {
3071             unlink "/var/run/qemu-server/${vmid}.$ext";
3072         }
3073     };
3074     warn $@ if $@; # avoid errors - just warn
3075 }
3076
3077 # Note: use $nockeck to skip tests if VM configuration file exists.
3078 # We need that when migration VMs to other nodes (files already moved)
3079 # Note: we set $keepActive in vzdump stop mode - volumes need to stay active
3080 sub vm_stop {
3081     my ($storecfg, $vmid, $skiplock, $nocheck, $timeout, $shutdown, $force, $keepActive, $migratedfrom) = @_;
3082
3083     $force = 1 if !defined($force) && !$shutdown;
3084
3085     if ($migratedfrom){
3086         my $pid = check_running($vmid, $nocheck, $migratedfrom);
3087         kill 15, $pid if $pid;
3088         my $conf = load_config($vmid, $migratedfrom);
3089         vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive);
3090         return;
3091     }
3092
3093     lock_config($vmid, sub {
3094
3095         my $pid = check_running($vmid, $nocheck);
3096         return if !$pid;
3097
3098         my $conf;
3099         if (!$nocheck) {
3100             $conf = load_config($vmid);
3101             check_lock($conf) if !$skiplock;
3102             if (!defined($timeout) && $shutdown && $conf->{startup}) {
3103                 my $opts = parse_startup($conf->{startup});
3104                 $timeout = $opts->{down} if $opts->{down};
3105             }
3106         }
3107
3108         $timeout = 60 if !defined($timeout);
3109
3110         eval {
3111             if ($shutdown) {
3112                 $nocheck ? vm_mon_cmd_nocheck($vmid, "system_powerdown") : vm_mon_cmd($vmid, "system_powerdown");
3113
3114             } else {
3115                 $nocheck ? vm_mon_cmd_nocheck($vmid, "quit") : vm_mon_cmd($vmid, "quit");
3116             }
3117         };
3118         my $err = $@;
3119
3120         if (!$err) {
3121             my $count = 0;
3122             while (($count < $timeout) && check_running($vmid, $nocheck)) {
3123                 $count++;
3124                 sleep 1;
3125             }
3126
3127             if ($count >= $timeout) {
3128                 if ($force) {
3129                     warn "VM still running - terminating now with SIGTERM\n";
3130                     kill 15, $pid;
3131                 } else {
3132                     die "VM quit/powerdown failed - got timeout\n";
3133                 }
3134             } else {
3135                 vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive) if $conf;
3136                 return;
3137             }
3138         } else {
3139             if ($force) {
3140                 warn "VM quit/powerdown failed - terminating now with SIGTERM\n";
3141                 kill 15, $pid;
3142             } else {
3143                 die "VM quit/powerdown failed\n";
3144             }
3145         }
3146
3147         # wait again
3148         $timeout = 10;
3149
3150         my $count = 0;
3151         while (($count < $timeout) && check_running($vmid, $nocheck)) {
3152             $count++;
3153             sleep 1;
3154         }
3155
3156         if ($count >= $timeout) {
3157             warn "VM still running - terminating now with SIGKILL\n";
3158             kill 9, $pid;
3159             sleep 1;
3160         }
3161
3162         vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive) if $conf;
3163    });
3164 }
3165
3166 sub vm_suspend {
3167     my ($vmid, $skiplock) = @_;
3168
3169     lock_config($vmid, sub {
3170
3171         my $conf = load_config($vmid);
3172
3173         check_lock($conf) if !$skiplock;
3174
3175         vm_mon_cmd($vmid, "stop");
3176     });
3177 }
3178
3179 sub vm_resume {
3180     my ($vmid, $skiplock) = @_;
3181
3182     lock_config($vmid, sub {
3183
3184         my $conf = load_config($vmid);
3185
3186         check_lock($conf) if !$skiplock;
3187
3188         vm_mon_cmd($vmid, "cont");
3189     });
3190 }
3191
3192 sub vm_sendkey {
3193     my ($vmid, $skiplock, $key) = @_;
3194
3195     lock_config($vmid, sub {
3196
3197         my $conf = load_config($vmid);
3198
3199         # there is no qmp command, so we use the human monitor command
3200         vm_human_monitor_command($vmid, "sendkey $key");
3201     });
3202 }
3203
3204 sub vm_destroy {
3205     my ($storecfg, $vmid, $skiplock) = @_;
3206
3207     lock_config($vmid, sub {
3208
3209         my $conf = load_config($vmid);
3210
3211         check_lock($conf) if !$skiplock;
3212
3213         if (!check_running($vmid)) {
3214             fairsched_rmnod($vmid); # try to destroy group
3215             destroy_vm($storecfg, $vmid);
3216         } else {
3217             die "VM $vmid is running - destroy failed\n";
3218         }
3219     });
3220 }
3221
3222 # pci helpers
3223
3224 sub file_write {
3225     my ($filename, $buf) = @_;
3226
3227     my $fh = IO::File->new($filename, "w");
3228     return undef if !$fh;
3229
3230     my $res = print $fh $buf;
3231
3232     $fh->close();
3233
3234     return $res;
3235 }
3236
3237 sub pci_device_info {
3238     my ($name) = @_;
3239
3240     my $res;
3241
3242     return undef if $name !~ m/^([a-f0-9]{4}):([a-f0-9]{2}):([a-f0-9]{2})\.([a-f0-9])$/;
3243     my ($domain, $bus, $slot, $func) = ($1, $2, $3, $4);
3244
3245     my $irq = file_read_firstline("$pcisysfs/devices/$name/irq");
3246     return undef if !defined($irq) || $irq !~ m/^\d+$/;
3247
3248     my $vendor = file_read_firstline("$pcisysfs/devices/$name/vendor");
3249     return undef if !defined($vendor) || $vendor !~ s/^0x//;
3250
3251     my $product = file_read_firstline("$pcisysfs/devices/$name/device");
3252     return undef if !defined($product) || $product !~ s/^0x//;
3253
3254     $res = {
3255         name => $name,
3256         vendor => $vendor,
3257         product => $product,
3258         domain => $domain,
3259         bus => $bus,
3260         slot => $slot,
3261         func => $func,
3262         irq => $irq,
3263         has_fl_reset => -f "$pcisysfs/devices/$name/reset" || 0,
3264     };
3265
3266     return $res;
3267 }
3268
3269 sub pci_dev_reset {
3270     my ($dev) = @_;
3271
3272     my $name = $dev->{name};
3273
3274     my $fn = "$pcisysfs/devices/$name/reset";
3275
3276     return file_write($fn, "1");
3277 }
3278
3279 sub pci_dev_bind_to_stub {
3280     my ($dev) = @_;
3281
3282     my $name = $dev->{name};
3283
3284     my $testdir = "$pcisysfs/drivers/pci-stub/$name";
3285     return 1 if -d $testdir;
3286
3287     my $data = "$dev->{vendor} $dev->{product}";
3288     return undef if !file_write("$pcisysfs/drivers/pci-stub/new_id", $data);
3289
3290     my $fn = "$pcisysfs/devices/$name/driver/unbind";
3291     if (!file_write($fn, $name)) {
3292         return undef if -f $fn;
3293     }
3294
3295     $fn = "$pcisysfs/drivers/pci-stub/bind";
3296     if (! -d $testdir) {
3297         return undef if !file_write($fn, $name);
3298     }
3299
3300     return -d $testdir;
3301 }
3302
3303 sub print_pci_addr {
3304     my ($id, $bridges) = @_;
3305
3306     my $res = '';
3307     my $devices = {
3308         #addr1 : ide,parallel,serial (motherboard)
3309         #addr2 : first videocard
3310         balloon0 => { bus => 0, addr => 3 },
3311         watchdog => { bus => 0, addr => 4 },
3312         scsihw0 => { bus => 0, addr => 5 },
3313         scsihw1 => { bus => 0, addr => 6 },
3314         ahci0 => { bus => 0, addr => 7 },
3315         qga0 => { bus => 0, addr => 8 },
3316         virtio0 => { bus => 0, addr => 10 },
3317         virtio1 => { bus => 0, addr => 11 },
3318         virtio2 => { bus => 0, addr => 12 },
3319         virtio3 => { bus => 0, addr => 13 },
3320         virtio4 => { bus => 0, addr => 14 },
3321         virtio5 => { bus => 0, addr => 15 },
3322         hostpci0 => { bus => 0, addr => 16 },
3323         hostpci1 => { bus => 0, addr => 17 },
3324         net0 => { bus => 0, addr => 18 },
3325         net1 => { bus => 0, addr => 19 },
3326         net2 => { bus => 0, addr => 20 },
3327         net3 => { bus => 0, addr => 21 },
3328         net4 => { bus => 0, addr => 22 },
3329         net5 => { bus => 0, addr => 23 },
3330         #addr29 : usb-host (pve-usb.cfg)
3331         'pci.1' => { bus => 0, addr => 30 },
3332         'pci.2' => { bus => 0, addr => 31 },
3333         'net6' => { bus => 1, addr => 1 },
3334         'net7' => { bus => 1, addr => 2 },
3335         'net8' => { bus => 1, addr => 3 },
3336         'net9' => { bus => 1, addr => 4 },
3337         'net10' => { bus => 1, addr => 5 },
3338         'net11' => { bus => 1, addr => 6 },
3339         'net12' => { bus => 1, addr => 7 },
3340         'net13' => { bus => 1, addr => 8 },
3341         'net14' => { bus => 1, addr => 9 },
3342         'net15' => { bus => 1, addr => 10 },
3343         'net16' => { bus => 1, addr => 11 },
3344         'net17' => { bus => 1, addr => 12 },
3345         'net18' => { bus => 1, addr => 13 },
3346         'net19' => { bus => 1, addr => 14 },
3347         'net20' => { bus => 1, addr => 15 },
3348         'net21' => { bus => 1, addr => 16 },
3349         'net22' => { bus => 1, addr => 17 },
3350         'net23' => { bus => 1, addr => 18 },
3351         'net24' => { bus => 1, addr => 19 },
3352         'net25' => { bus => 1, addr => 20 },
3353         'net26' => { bus => 1, addr => 21 },
3354         'net27' => { bus => 1, addr => 22 },
3355         'net28' => { bus => 1, addr => 23 },
3356         'net29' => { bus => 1, addr => 24 },
3357         'net30' => { bus => 1, addr => 25 },
3358         'net31' => { bus => 1, addr => 26 },
3359         'virtio6' => { bus => 2, addr => 1 },
3360         'virtio7' => { bus => 2, addr => 2 },
3361         'virtio8' => { bus => 2, addr => 3 },
3362         'virtio9' => { bus => 2, addr => 4 },
3363         'virtio10' => { bus => 2, addr => 5 },
3364         'virtio11' => { bus => 2, addr => 6 },
3365         'virtio12' => { bus => 2, addr => 7 },
3366         'virtio13' => { bus => 2, addr => 8 },
3367         'virtio14' => { bus => 2, addr => 9 },
3368         'virtio15' => { bus => 2, addr => 10 },
3369     };
3370
3371     if (defined($devices->{$id}->{bus}) && defined($devices->{$id}->{addr})) {
3372            my $addr = sprintf("0x%x", $devices->{$id}->{addr});
3373            my $bus = $devices->{$id}->{bus};
3374            $res = ",bus=pci.$bus,addr=$addr";
3375            $bridges->{$bus} = 1 if $bridges;
3376     }
3377     return $res;
3378
3379 }
3380
3381 sub vm_balloonset {
3382     my ($vmid, $value) = @_;
3383
3384     vm_mon_cmd($vmid, "balloon", value => $value);
3385 }
3386
3387 # vzdump restore implementaion
3388
3389 sub archive_read_firstfile {
3390     my $archive = shift;
3391
3392     die "ERROR: file '$archive' does not exist\n" if ! -f $archive;
3393
3394     # try to detect archive type first
3395     my $pid = open (TMP, "tar tf '$archive'|") ||
3396         die "unable to open file '$archive'\n";
3397     my $firstfile = <TMP>;
3398     kill 15, $pid;
3399     close TMP;
3400
3401     die "ERROR: archive contaions no data\n" if !$firstfile;
3402     chomp $firstfile;
3403
3404     return $firstfile;
3405 }
3406
3407 sub restore_cleanup {
3408     my $statfile = shift;
3409
3410     print STDERR "starting cleanup\n";
3411
3412     if (my $fd = IO::File->new($statfile, "r")) {
3413         while (defined(my $line = <$fd>)) {
3414             if ($line =~ m/vzdump:([^\s:]*):(\S+)$/) {
3415                 my $volid = $2;
3416                 eval {
3417                     if ($volid =~ m|^/|) {
3418                         unlink $volid || die 'unlink failed\n';
3419                     } else {
3420                         my $cfg = cfs_read_file('storage.cfg');
3421                         PVE::Storage::vdisk_free($cfg, $volid);
3422                     }
3423                     print STDERR "temporary volume '$volid' sucessfuly removed\n";
3424                 };
3425                 print STDERR "unable to cleanup '$volid' - $@" if $@;
3426             } else {
3427                 print STDERR "unable to parse line in statfile - $line";
3428             }
3429         }
3430         $fd->close();
3431     }
3432 }
3433
3434 sub restore_archive {
3435     my ($archive, $vmid, $user, $opts) = @_;
3436
3437     if ($archive ne '-') {
3438         my $firstfile = archive_read_firstfile($archive);
3439         die "ERROR: file '$archive' dos not lock like a QemuServer vzdump backup\n"
3440             if $firstfile ne 'qemu-server.conf';
3441     }
3442
3443     my $tocmd = "/usr/lib/qemu-server/qmextract";
3444
3445     $tocmd .= " --storage " . PVE::Tools::shellquote($opts->{storage}) if $opts->{storage};
3446     $tocmd .= " --pool " . PVE::Tools::shellquote($opts->{pool}) if $opts->{pool};
3447     $tocmd .= ' --prealloc' if $opts->{prealloc};
3448     $tocmd .= ' --info' if $opts->{info};
3449
3450     # tar option "xf" does not autodetect compression when read from STDIN,
3451     # so we pipe to zcat
3452     my $cmd = "zcat -f|tar xf " . PVE::Tools::shellquote($archive) . " " .
3453         PVE::Tools::shellquote("--to-command=$tocmd");
3454
3455     my $tmpdir = "/var/tmp/vzdumptmp$$";
3456     mkpath $tmpdir;
3457
3458     local $ENV{VZDUMP_TMPDIR} = $tmpdir;
3459     local $ENV{VZDUMP_VMID} = $vmid;
3460     local $ENV{VZDUMP_USER} = $user;
3461
3462     my $conffile = PVE::QemuServer::config_file($vmid);
3463     my $tmpfn = "$conffile.$$.tmp";
3464
3465     # disable interrupts (always do cleanups)
3466     local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = sub {
3467         print STDERR "got interrupt - ignored\n";
3468     };
3469
3470     eval {
3471         # enable interrupts
3472         local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = $SIG{PIPE} = sub {
3473             die "interrupted by signal\n";
3474         };
3475
3476         if ($archive eq '-') {
3477             print "extracting archive from STDIN\n";
3478             run_command($cmd, input => "<&STDIN");
3479         } else {
3480             print "extracting archive '$archive'\n";
3481             run_command($cmd);
3482         }
3483
3484         return if $opts->{info};
3485
3486         # read new mapping
3487         my $map = {};
3488         my $statfile = "$tmpdir/qmrestore.stat";
3489         if (my $fd = IO::File->new($statfile, "r")) {
3490             while (defined (my $line = <$fd>)) {
3491                 if ($line =~ m/vzdump:([^\s:]*):(\S+)$/) {
3492                     $map->{$1} = $2 if $1;
3493                 } else {
3494                     print STDERR "unable to parse line in statfile - $line\n";
3495                 }
3496             }
3497             $fd->close();
3498         }
3499
3500         my $confsrc = "$tmpdir/qemu-server.conf";
3501
3502         my $srcfd = new IO::File($confsrc, "r") ||
3503             die "unable to open file '$confsrc'\n";
3504
3505         my $outfd = new IO::File ($tmpfn, "w") ||
3506             die "unable to write config for VM $vmid\n";
3507
3508         my $netcount = 0;
3509
3510         while (defined (my $line = <$srcfd>)) {
3511             next if $line =~ m/^\#vzdump\#/;
3512             next if $line =~ m/^lock:/;
3513             next if $line =~ m/^unused\d+:/;
3514
3515             if (($line =~ m/^(vlan(\d+)):\s*(\S+)\s*$/)) {
3516                 # try to convert old 1.X settings
3517                 my ($id, $ind, $ethcfg) = ($1, $2, $3);
3518                 foreach my $devconfig (PVE::Tools::split_list($ethcfg)) {
3519                     my ($model, $macaddr) = split(/\=/, $devconfig);
3520                     $macaddr = PVE::Tools::random_ether_addr() if !$macaddr || $opts->{unique};
3521                     my $net = {
3522                         model => $model,
3523                         bridge => "vmbr$ind",
3524                         macaddr => $macaddr,
3525                     };
3526                     my $netstr = print_net($net);
3527                     print $outfd "net${netcount}: $netstr\n";
3528                     $netcount++;
3529                 }
3530             } elsif (($line =~ m/^(net\d+):\s*(\S+)\s*$/) && ($opts->{unique})) {
3531                 my ($id, $netstr) = ($1, $2);
3532                 my $net = parse_net($netstr);
3533                 $net->{macaddr} = PVE::Tools::random_ether_addr() if $net->{macaddr};
3534                 $netstr = print_net($net);
3535                 print $outfd "$id: $netstr\n";
3536             } elsif ($line =~ m/^((ide|scsi|virtio)\d+):\s*(\S+)\s*$/) {
3537                 my $virtdev = $1;
3538                 my $value = $2;
3539                 if ($line =~ m/backup=no/) {
3540                     print $outfd "#$line";
3541                 } elsif ($virtdev && $map->{$virtdev}) {
3542                     my $di = PVE::QemuServer::parse_drive($virtdev, $value);
3543                     $di->{file} = $map->{$virtdev};
3544                     $value = PVE::QemuServer::print_drive($vmid, $di);
3545                     print $outfd "$virtdev: $value\n";
3546                 } else {
3547                     print $outfd $line;
3548                 }
3549             } else {
3550                 print $outfd $line;
3551             }
3552         }
3553
3554         $srcfd->close();
3555         $outfd->close();
3556     };
3557     my $err = $@;
3558
3559     if ($err) {
3560
3561         unlink $tmpfn;
3562
3563         restore_cleanup("$tmpdir/qmrestore.stat") if !$opts->{info};
3564
3565         die $err;
3566     }
3567
3568     rmtree $tmpdir;
3569
3570     rename $tmpfn, $conffile ||
3571         die "unable to commit configuration file '$conffile'\n";
3572 };
3573
3574
3575 # Internal snapshots
3576
3577 # NOTE: Snapshot create/delete involves several non-atomic
3578 # action, and can take a long time.
3579 # So we try to avoid locking the file and use 'lock' variable
3580 # inside the config file instead.
3581
3582 my $snapshot_copy_config = sub {
3583     my ($source, $dest) = @_;
3584
3585     foreach my $k (keys %$source) {
3586         next if $k eq 'snapshots';
3587         next if $k eq 'snapstate';
3588         next if $k eq 'snaptime';
3589         next if $k eq 'vmstate';
3590         next if $k eq 'lock';
3591         next if $k eq 'digest';
3592         next if $k eq 'description';
3593         next if $k =~ m/^unused\d+$/;
3594
3595         $dest->{$k} = $source->{$k};
3596     }
3597 };
3598
3599 my $snapshot_apply_config = sub {
3600     my ($conf, $snap) = @_;
3601
3602     # copy snapshot list
3603     my $newconf = {
3604         snapshots => $conf->{snapshots},
3605     };
3606
3607     # keep description and list of unused disks
3608     foreach my $k (keys %$conf) {
3609         next if !($k =~ m/^unused\d+$/ || $k eq 'description');
3610         $newconf->{$k} = $conf->{$k};
3611     }
3612
3613     &$snapshot_copy_config($snap, $newconf);
3614
3615     return $newconf;
3616 };
3617
3618 sub foreach_writable_storage {
3619     my ($conf, $func) = @_;
3620
3621     my $sidhash = {};
3622
3623     foreach my $ds (keys %$conf) {
3624         next if !valid_drivename($ds);
3625
3626         my $drive = parse_drive($ds, $conf->{$ds});
3627         next if !$drive;
3628         next if drive_is_cdrom($drive);
3629
3630         my $volid = $drive->{file};
3631
3632         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
3633         $sidhash->{$sid} = $sid if $sid;
3634     }
3635
3636     foreach my $sid (sort keys %$sidhash) {
3637         &$func($sid);
3638     }
3639 }
3640
3641 my $alloc_vmstate_volid = sub {
3642     my ($storecfg, $vmid, $conf, $snapname) = @_;
3643
3644     # Note: we try to be smart when selecting a $target storage
3645
3646     my $target;
3647
3648     # search shared storage first
3649     foreach_writable_storage($conf, sub {
3650         my ($sid) = @_;
3651         my $scfg = PVE::Storage::storage_config($storecfg, $sid);
3652         return if !$scfg->{shared};
3653
3654         $target = $sid if !$target || $scfg->{path}; # prefer file based storage
3655     });
3656
3657     if (!$target) {
3658         # now search local storage
3659         foreach_writable_storage($conf, sub {
3660             my ($sid) = @_;
3661             my $scfg = PVE::Storage::storage_config($storecfg, $sid);
3662             return if $scfg->{shared};
3663
3664             $target = $sid if !$target || $scfg->{path}; # prefer file based storage;
3665         });
3666     }
3667
3668     $target = 'local' if !$target;
3669
3670     my $driver_state_size = 32; # assume 32MB is enough to safe all driver state;
3671     my $size = $conf->{memory} + $driver_state_size;
3672
3673     my $name = "vm-$vmid-state-$snapname";
3674     my $scfg = PVE::Storage::storage_config($storecfg, $target);
3675     $name .= ".raw" if $scfg->{path}; # add filename extension for file base storage
3676     my $volid = PVE::Storage::vdisk_alloc($storecfg, $target, $vmid, 'raw', $name, $size*1024);
3677
3678     return $volid;
3679 };
3680
3681 my $snapshot_prepare = sub {
3682     my ($vmid, $snapname, $save_vmstate, $comment) = @_;
3683
3684     my $snap;
3685
3686     my $updatefn =  sub {
3687
3688         my $conf = load_config($vmid);
3689
3690         check_lock($conf);
3691
3692         $conf->{lock} = 'snapshot';
3693
3694         die "snapshot name '$snapname' already used\n"
3695             if defined($conf->{snapshots}->{$snapname});
3696
3697         my $storecfg = PVE::Storage::config();
3698
3699         foreach_drive($conf, sub {
3700             my ($ds, $drive) = @_;
3701
3702             return if drive_is_cdrom($drive);
3703             my $volid = $drive->{file};
3704
3705             my ($storeid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
3706             if ($storeid) {
3707                 my $scfg = PVE::Storage::storage_config($storecfg, $storeid);
3708                 die "can't snapshot volume '$volid'\n"
3709                     if !(($scfg->{path} && $volname =~ m/\.qcow2$/) ||
3710                          ($scfg->{type} eq 'nexenta') ||
3711                          ($scfg->{type} eq 'rbd') ||
3712                          ($scfg->{type} eq 'sheepdog'));
3713             } elsif ($volid =~ m|^(/.+)$| && -e $volid) {
3714                 die "snapshot device '$volid' is not possible\n";
3715             } else {
3716                 die "can't snapshot volume '$volid'\n";
3717             }
3718         });
3719
3720
3721         $snap = $conf->{snapshots}->{$snapname} = {};
3722
3723         if ($save_vmstate && check_running($vmid)) {
3724             $snap->{vmstate} = &$alloc_vmstate_volid($storecfg, $vmid, $conf, $snapname);
3725         }
3726
3727         &$snapshot_copy_config($conf, $snap);
3728
3729         $snap->{snapstate} = "prepare";
3730         $snap->{snaptime} = time();
3731         $snap->{description} = $comment if $comment;
3732
3733         update_config_nolock($vmid, $conf, 1);
3734     };
3735
3736     lock_config($vmid, $updatefn);
3737
3738     return $snap;
3739 };
3740
3741 my $snapshot_commit = sub {
3742     my ($vmid, $snapname) = @_;
3743
3744     my $updatefn = sub {
3745
3746         my $conf = load_config($vmid);
3747
3748         die "missing snapshot lock\n"
3749             if !($conf->{lock} && $conf->{lock} eq 'snapshot');
3750
3751         my $snap = $conf->{snapshots}->{$snapname};
3752
3753         die "snapshot '$snapname' does not exist\n" if !defined($snap);
3754
3755         die "wrong snapshot state\n"
3756             if !($snap->{snapstate} && $snap->{snapstate} eq "prepare");
3757
3758         delete $snap->{snapstate};
3759         delete $conf->{lock};
3760
3761         my $newconf = &$snapshot_apply_config($conf, $snap);
3762
3763         $newconf->{parent} = $snapname;
3764
3765         update_config_nolock($vmid, $newconf, 1);
3766     };
3767
3768     lock_config($vmid, $updatefn);
3769 };
3770
3771 sub snapshot_rollback {
3772     my ($vmid, $snapname) = @_;
3773
3774     my $snap;
3775
3776     my $prepare = 1;
3777
3778     my $storecfg = PVE::Storage::config();
3779
3780     my $updatefn = sub {
3781
3782         my $conf = load_config($vmid);
3783
3784         $snap = $conf->{snapshots}->{$snapname};
3785
3786         die "snapshot '$snapname' does not exist\n" if !defined($snap);
3787
3788         die "unable to rollback to incomplete snapshot (snapstate = $snap->{snapstate})\n"
3789             if $snap->{snapstate};
3790
3791         if ($prepare) {
3792             check_lock($conf);
3793             vm_stop($storecfg, $vmid, undef, undef, 5, undef, undef);
3794         }
3795
3796         die "unable to rollback vm $vmid: vm is running\n"
3797             if check_running($vmid);
3798
3799         if ($prepare) {
3800             $conf->{lock} = 'rollback';
3801         } else {
3802             die "got wrong lock\n" if !($conf->{lock} && $conf->{lock} eq 'rollback');
3803             delete $conf->{lock};
3804         }
3805
3806         if (!$prepare) {
3807             # copy snapshot config to current config
3808             $conf = &$snapshot_apply_config($conf, $snap);
3809             $conf->{parent} = $snapname;
3810         }
3811
3812         update_config_nolock($vmid, $conf, 1);
3813
3814         if (!$prepare && $snap->{vmstate}) {
3815             my $statefile = PVE::Storage::path($storecfg, $snap->{vmstate});
3816             # fixme: this only forws for files currently
3817             vm_start($storecfg, $vmid, $statefile);
3818         }
3819
3820     };
3821
3822     lock_config($vmid, $updatefn);
3823
3824     foreach_drive($snap, sub {
3825         my ($ds, $drive) = @_;
3826
3827         return if drive_is_cdrom($drive);
3828
3829         my $volid = $drive->{file};
3830         my $device = "drive-$ds";
3831
3832         PVE::Storage::volume_snapshot_rollback($storecfg, $volid, $snapname);
3833     });
3834
3835     $prepare = 0;
3836     lock_config($vmid, $updatefn);
3837 }
3838
3839 my $savevm_wait = sub {
3840     my ($vmid) = @_;
3841
3842     for(;;) {
3843         my $stat = PVE::QemuServer::vm_mon_cmd_nocheck($vmid, "query-savevm");
3844         if (!$stat->{status}) {
3845             die "savevm not active\n";
3846         } elsif ($stat->{status} eq 'active') {
3847             sleep(1);
3848             next;
3849         } elsif ($stat->{status} eq 'completed') {
3850             last;
3851         } else {
3852             die "query-savevm returned status '$stat->{status}'\n";
3853         }
3854     }
3855 };
3856
3857 sub snapshot_create {
3858     my ($vmid, $snapname, $save_vmstate, $freezefs, $comment) = @_;
3859
3860     my $snap = &$snapshot_prepare($vmid, $snapname, $save_vmstate, $comment);
3861
3862     $freezefs = $save_vmstate = 0 if !$snap->{vmstate}; # vm is not running
3863
3864     my $drivehash = {};
3865
3866     my $running = check_running($vmid);
3867
3868     eval {
3869         # create internal snapshots of all drives
3870
3871         my $storecfg = PVE::Storage::config();
3872
3873         if ($running) {
3874             if ($snap->{vmstate}) {
3875                 my $path = PVE::Storage::path($storecfg, $snap->{vmstate});
3876                 vm_mon_cmd($vmid, "savevm-start", statefile => $path);
3877                 &$savevm_wait($vmid);
3878             } else {
3879                 vm_mon_cmd($vmid, "savevm-start");
3880             }
3881         };
3882
3883         qga_freezefs($vmid) if $running && $freezefs;
3884
3885         foreach_drive($snap, sub {
3886             my ($ds, $drive) = @_;
3887
3888             return if drive_is_cdrom($drive);
3889
3890             my $volid = $drive->{file};
3891             my $device = "drive-$ds";
3892
3893             qemu_volume_snapshot($vmid, $device, $storecfg, $volid, $snapname);
3894             $drivehash->{$ds} = 1;
3895        });
3896     };
3897     my $err = $@;
3898
3899     eval { gqa_unfreezefs($vmid) if $running && $freezefs; };
3900     warn $@ if $@;
3901
3902     eval { vm_mon_cmd($vmid, "savevm-end") if $running; };
3903     warn $@ if $@;
3904
3905     if ($err) {
3906         warn "snapshot create failed: starting cleanup\n";
3907         eval { snapshot_delete($vmid, $snapname, 0, $drivehash); };
3908         warn $@ if $@;
3909         die $err;
3910     }
3911
3912     &$snapshot_commit($vmid, $snapname);
3913 }
3914
3915 # Note: $drivehash is only set when called from snapshot_create.
3916 sub snapshot_delete {
3917     my ($vmid, $snapname, $force, $drivehash) = @_;
3918
3919     my $prepare = 1;
3920
3921     my $snap;
3922     my $unused = [];
3923
3924     my $unlink_parent = sub {
3925         my ($confref, $new_parent) = @_;
3926
3927         if ($confref->{parent} && $confref->{parent} eq $snapname) {
3928             if ($new_parent) {
3929                 $confref->{parent} = $new_parent;
3930             } else {
3931                 delete $confref->{parent};
3932             }
3933         }
3934     };
3935
3936     my $updatefn =  sub {
3937         my ($remove_drive) = @_;
3938
3939         my $conf = load_config($vmid);
3940
3941         check_lock($conf) if !$drivehash;
3942
3943         $snap = $conf->{snapshots}->{$snapname};
3944
3945         die "snapshot '$snapname' does not exist\n" if !defined($snap);
3946
3947         # remove parent refs
3948         &$unlink_parent($conf, $snap->{parent});
3949         foreach my $sn (keys %{$conf->{snapshots}}) {
3950             next if $sn eq $snapname;
3951             &$unlink_parent($conf->{snapshots}->{$sn}, $snap->{parent});
3952         }
3953
3954         if ($remove_drive) {
3955             if ($remove_drive eq 'vmstate') {
3956                 delete $snap->{$remove_drive};
3957             } else {
3958                 my $drive = parse_drive($remove_drive, $snap->{$remove_drive});
3959                 my $volid = $drive->{file};
3960                 delete $snap->{$remove_drive};
3961                 add_unused_volume($conf, $volid);
3962             }
3963         }
3964
3965         if ($prepare) {
3966             $snap->{snapstate} = 'delete';
3967         } else {
3968             delete $conf->{snapshots}->{$snapname};
3969             delete $conf->{lock} if $drivehash;
3970             foreach my $volid (@$unused) {
3971                 add_unused_volume($conf, $volid);
3972             }
3973         }
3974
3975         update_config_nolock($vmid, $conf, 1);
3976     };
3977
3978     lock_config($vmid, $updatefn);
3979
3980     # now remove vmstate file
3981
3982     my $storecfg = PVE::Storage::config();
3983
3984     if ($snap->{vmstate}) {
3985         eval {  PVE::Storage::vdisk_free($storecfg, $snap->{vmstate}); };
3986         if (my $err = $@) {
3987             die $err if !$force;
3988             warn $err;
3989         }
3990         # save changes (remove vmstate from snapshot)
3991         lock_config($vmid, $updatefn, 'vmstate') if !$force;
3992     };
3993
3994     # now remove all internal snapshots
3995     foreach_drive($snap, sub {
3996         my ($ds, $drive) = @_;
3997
3998         return if drive_is_cdrom($drive);
3999
4000         my $volid = $drive->{file};
4001         my $device = "drive-$ds";
4002
4003         if (!$drivehash || $drivehash->{$ds}) {
4004             eval { qemu_volume_snapshot_delete($vmid, $device, $storecfg, $volid, $snapname); };
4005             if (my $err = $@) {
4006                 die $err if !$force;
4007                 warn $err;
4008             }
4009         }
4010
4011         # save changes (remove drive fron snapshot)
4012         lock_config($vmid, $updatefn, $ds) if !$force;
4013         push @$unused, $volid;
4014     });
4015
4016     # now cleanup config
4017     $prepare = 0;
4018     lock_config($vmid, $updatefn);
4019 }
4020
4021 1;