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 lock_file_full 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 PVE::RPCEnvironment;
  31 use Time::HiRes qw(gettimeofday);
  32
  33 my $cpuinfo = PVE::ProcFSTools::read_cpuinfo();
  34
  35 # Note about locking: we use flock on the config file protect
  36 # against concurent actions.
  37 # Aditionaly, we have a 'lock' setting in the config file. This
  38 # can be set to 'migrate', 'backup', 'snapshot' or 'rollback'. Most actions are not
  39 # allowed when such lock is set. But you can ignore this kind of
  40 # lock with the --skiplock flag.
  41
  42 cfs_register_file('/qemu-server/',
  43                   \&parse_vm_config,
  44                   \&write_vm_config);
  45
  46 PVE::JSONSchema::register_standard_option('skiplock', {
  47     description => "Ignore locks - only root is allowed to use this option.",
  48     type => 'boolean',
  49     optional => 1,
  50 });
  51
  52 PVE::JSONSchema::register_standard_option('pve-qm-stateuri', {
  53     description => "Some command save/restore state from this location.",
  54     type => 'string',
  55     maxLength => 128,
  56     optional => 1,
  57 });
  58
  59 PVE::JSONSchema::register_standard_option('pve-snapshot-name', {
  60     description => "The name of the snapshot.",
  61     type => 'string', format => 'pve-configid',
  62     maxLength => 40,
  63 });
  64
  65 #no warnings 'redefine';
  66
  67 unless(defined(&_VZSYSCALLS_H_)) {
  68     eval 'sub _VZSYSCALLS_H_ () {1;}' unless defined(&_VZSYSCALLS_H_);
  69     require 'sys/syscall.ph';
  70     if(defined(&__x86_64__)) {
  71         eval 'sub __NR_fairsched_vcpus () {499;}' unless defined(&__NR_fairsched_vcpus);
  72         eval 'sub __NR_fairsched_mknod () {504;}' unless defined(&__NR_fairsched_mknod);
  73         eval 'sub __NR_fairsched_rmnod () {505;}' unless defined(&__NR_fairsched_rmnod);
  74         eval 'sub __NR_fairsched_chwt () {506;}' unless defined(&__NR_fairsched_chwt);
  75         eval 'sub __NR_fairsched_mvpr () {507;}' unless defined(&__NR_fairsched_mvpr);
  76         eval 'sub __NR_fairsched_rate () {508;}' unless defined(&__NR_fairsched_rate);
  77         eval 'sub __NR_setluid () {501;}' unless defined(&__NR_setluid);
  78         eval 'sub __NR_setublimit () {502;}' unless defined(&__NR_setublimit);
  79     }
  80     elsif(defined( &__i386__) ) {
  81         eval 'sub __NR_fairsched_mknod () {500;}' unless defined(&__NR_fairsched_mknod);
  82         eval 'sub __NR_fairsched_rmnod () {501;}' unless defined(&__NR_fairsched_rmnod);
  83         eval 'sub __NR_fairsched_chwt () {502;}' unless defined(&__NR_fairsched_chwt);
  84         eval 'sub __NR_fairsched_mvpr () {503;}' unless defined(&__NR_fairsched_mvpr);
  85         eval 'sub __NR_fairsched_rate () {504;}' unless defined(&__NR_fairsched_rate);
  86         eval 'sub __NR_fairsched_vcpus () {505;}' unless defined(&__NR_fairsched_vcpus);
  87         eval 'sub __NR_setluid () {511;}' unless defined(&__NR_setluid);
  88         eval 'sub __NR_setublimit () {512;}' unless defined(&__NR_setublimit);
  89     } else {
  90         die("no fairsched syscall for this arch");
  91     }
  92     require 'asm/ioctl.ph';
  93     eval 'sub KVM_GET_API_VERSION () { &_IO(0xAE, 0x);}' unless defined(&KVM_GET_API_VERSION);
  94 }
  95
  96 sub fairsched_mknod {
  97     my ($parent, $weight, $desired) = @_;
  98
  99     return syscall(&__NR_fairsched_mknod, int($parent), int($weight), int($desired));
 100 }
 101
 102 sub fairsched_rmnod {
 103     my ($id) = @_;
 104
 105     return syscall(&__NR_fairsched_rmnod, int($id));
 106 }
 107
 108 sub fairsched_mvpr {
 109     my ($pid, $newid) = @_;
 110
 111     return syscall(&__NR_fairsched_mvpr, int($pid), int($newid));
 112 }
 113
 114 sub fairsched_vcpus {
 115     my ($id, $vcpus) = @_;
 116
 117     return syscall(&__NR_fairsched_vcpus, int($id), int($vcpus));
 118 }
 119
 120 sub fairsched_rate {
 121     my ($id, $op, $rate) = @_;
 122
 123     return syscall(&__NR_fairsched_rate, int($id), int($op), int($rate));
 124 }
 125
 126 use constant FAIRSCHED_SET_RATE  => 0;
 127 use constant FAIRSCHED_DROP_RATE => 1;
 128 use constant FAIRSCHED_GET_RATE  => 2;
 129
 130 sub fairsched_cpulimit {
 131     my ($id, $limit) = @_;
 132
 133     my $cpulim1024 = int($limit * 1024 / 100);
 134     my $op = $cpulim1024 ? FAIRSCHED_SET_RATE : FAIRSCHED_DROP_RATE;
 135
 136     return fairsched_rate($id, $op, $cpulim1024);
 137 }
 138
 139 my $nodename = PVE::INotify::nodename();
 140
 141 mkdir "/etc/pve/nodes/$nodename";
 142 my $confdir = "/etc/pve/nodes/$nodename/qemu-server";
 143 mkdir $confdir;
 144
 145 my $var_run_tmpdir = "/var/run/qemu-server";
 146 mkdir $var_run_tmpdir;
 147
 148 my $lock_dir = "/var/lock/qemu-server";
 149 mkdir $lock_dir;
 150
 151 my $pcisysfs = "/sys/bus/pci";
 152
 153 my $confdesc = {
 154     onboot => {
 155         optional => 1,
 156         type => 'boolean',
 157         description => "Specifies whether a VM will be started during system bootup.",
 158         default => 0,
 159     },
 160     autostart => {
 161         optional => 1,
 162         type => 'boolean',
 163         description => "Automatic restart after crash (currently ignored).",
 164         default => 0,
 165     },
 166     hotplug => {
 167         optional => 1,
 168         type => 'boolean',
 169         description => "Allow hotplug for disk and network device",
 170         default => 0,
 171     },
 172     reboot => {
 173         optional => 1,
 174         type => 'boolean',
 175         description => "Allow reboot. If set to '0' the VM exit on reboot.",
 176         default => 1,
 177     },
 178     lock => {
 179         optional => 1,
 180         type => 'string',
 181         description => "Lock/unlock the VM.",
 182         enum => [qw(migrate backup snapshot rollback)],
 183     },
 184     cpulimit => {
 185         optional => 1,
 186         type => 'integer',
 187         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.",
 188         minimum => 0,
 189         default => 0,
 190     },
 191     cpuunits => {
 192         optional => 1,
 193         type => 'integer',
 194         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.",
 195         minimum => 0,
 196         maximum => 500000,
 197         default => 1000,
 198     },
 199     memory => {
 200         optional => 1,
 201         type => 'integer',
 202         description => "Amount of RAM for the VM in MB. This is the maximum available memory when you use the balloon device.",
 203         minimum => 16,
 204         default => 512,
 205     },
 206     balloon => {
 207         optional => 1,
 208         type => 'integer',
 209         description => "Amount of target RAM for the VM in MB. Using zero disables the ballon driver.",
 210         minimum => 0,
 211     },
 212     shares => {
 213         optional => 1,
 214         type => 'integer',
 215         description => "Amount of memory shares for auto-ballooning. The larger the number is, the more memory this VM gets. Number is relative to weights of all other running VMs. Using zero disables auto-ballooning",
 216         minimum => 0,
 217         maximum => 50000,
 218         default => 1000,
 219     },
 220     keyboard => {
 221         optional => 1,
 222         type => 'string',
 223         description => "Keybord layout for vnc server. Default is read from the datacenter configuration file.",
 224         enum => PVE::Tools::kvmkeymaplist(),
 225         default => 'en-us',
 226     },
 227     name => {
 228         optional => 1,
 229         type => 'string', format => 'dns-name',
 230         description => "Set a name for the VM. Only used on the configuration web interface.",
 231     },
 232     scsihw => {
 233         optional => 1,
 234         type => 'string',
 235         description => "scsi controller model",
 236         enum => [qw(lsi virtio-scsi-pci megasas)],
 237         default => 'lsi',
 238     },
 239     description => {
 240         optional => 1,
 241         type => 'string',
 242         description => "Description for the VM. Only used on the configuration web interface. This is saved as comment inside the configuration file.",
 243     },
 244     ostype => {
 245         optional => 1,
 246         type => 'string',
 247         enum => [qw(other wxp w2k w2k3 w2k8 wvista win7 win8 l24 l26)],
 248         description => <<EODESC,
 249 Used to enable special optimization/features for specific
 250 operating systems:
 251
 252 other  => unspecified OS
 253 wxp    => Microsoft Windows XP
 254 w2k    => Microsoft Windows 2000
 255 w2k3   => Microsoft Windows 2003
 256 w2k8   => Microsoft Windows 2008
 257 wvista => Microsoft Windows Vista
 258 win7   => Microsoft Windows 7
 259 win8   => Microsoft Windows 8/2012
 260 l24    => Linux 2.4 Kernel
 261 l26    => Linux 2.6/3.X Kernel
 262
 263 other|l24|l26                       ... no special behaviour
 264 wxp|w2k|w2k3|w2k8|wvista|win7|win8  ... use --localtime switch
 265 EODESC
 266     },
 267     boot => {
 268         optional => 1,
 269         type => 'string',
 270         description => "Boot on floppy (a), hard disk (c), CD-ROM (d), or network (n).",
 271         pattern => '[acdn]{1,4}',
 272         default => 'cdn',
 273     },
 274     bootdisk => {
 275         optional => 1,
 276         type => 'string', format => 'pve-qm-bootdisk',
 277         description => "Enable booting from specified disk.",
 278         pattern => '(ide|sata|scsi|virtio)\d+',
 279     },
 280     smp => {
 281         optional => 1,
 282         type => 'integer',
 283         description => "The number of CPUs. Please use option -sockets instead.",
 284         minimum => 1,
 285         default => 1,
 286     },
 287     sockets => {
 288         optional => 1,
 289         type => 'integer',
 290         description => "The number of CPU sockets.",
 291         minimum => 1,
 292         default => 1,
 293     },
 294     cores => {
 295         optional => 1,
 296         type => 'integer',
 297         description => "The number of cores per socket.",
 298         minimum => 1,
 299         default => 1,
 300     },
 301     acpi => {
 302         optional => 1,
 303         type => 'boolean',
 304         description => "Enable/disable ACPI.",
 305         default => 1,
 306     },
 307     agent => {
 308         optional => 1,
 309         type => 'boolean',
 310         description => "Enable/disable Qemu GuestAgent.",
 311         default => 0,
 312     },
 313     kvm => {
 314         optional => 1,
 315         type => 'boolean',
 316         description => "Enable/disable KVM hardware virtualization.",
 317         default => 1,
 318     },
 319     tdf => {
 320         optional => 1,
 321         type => 'boolean',
 322         description => "Enable/disable time drift fix.",
 323         default => 0,
 324     },
 325     localtime => {
 326         optional => 1,
 327         type => 'boolean',
 328         description => "Set the real time clock to local time. This is enabled by default if ostype indicates a Microsoft OS.",
 329     },
 330     freeze => {
 331         optional => 1,
 332         type => 'boolean',
 333         description => "Freeze CPU at startup (use 'c' monitor command to start execution).",
 334     },
 335     vga => {
 336         optional => 1,
 337         type => 'string',
 338         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 win8/win7/w2k8, and 'cirrur' for other OS types. Option 'qxl' enables the SPICE display sever.",
 339         enum => [qw(std cirrus vmware qxl)],
 340     },
 341     watchdog => {
 342         optional => 1,
 343         type => 'string', format => 'pve-qm-watchdog',
 344         typetext => '[[model=]i6300esb|ib700] [,[action=]reset|shutdown|poweroff|pause|debug|none]',
 345         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)",
 346     },
 347     startdate => {
 348         optional => 1,
 349         type => 'string',
 350         typetext => "(now | YYYY-MM-DD | YYYY-MM-DDTHH:MM:SS)",
 351         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'.",
 352         pattern => '(now|\d{4}-\d{1,2}-\d{1,2}(T\d{1,2}:\d{1,2}:\d{1,2})?)',
 353         default => 'now',
 354     },
 355     startup => {
 356         optional => 1,
 357         type => 'string', format => 'pve-qm-startup',
 358         typetext => '[[order=]\d+] [,up=\d+] [,down=\d+] ',
 359         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.",
 360     },
 361     template => {
 362         optional => 1,
 363         type => 'boolean',
 364         description => "Enable/disable Template.",
 365         default => 0,
 366     },
 367     args => {
 368         optional => 1,
 369         type => 'string',
 370         description => <<EODESCR,
 371 Note: this option is for experts only. It allows you to pass arbitrary arguments to kvm, for example:
 372
 373 args: -no-reboot -no-hpet
 374 EODESCR
 375     },
 376     tablet => {
 377         optional => 1,
 378         type => 'boolean',
 379         default => 1,
 380         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.",
 381     },
 382     migrate_speed => {
 383         optional => 1,
 384         type => 'integer',
 385         description => "Set maximum speed (in MB/s) for migrations. Value 0 is no limit.",
 386         minimum => 0,
 387         default => 0,
 388     },
 389     migrate_downtime => {
 390         optional => 1,
 391         type => 'number',
 392         description => "Set maximum tolerated downtime (in seconds) for migrations.",
 393         minimum => 0,
 394         default => 0.1,
 395     },
 396     cdrom => {
 397         optional => 1,
 398         type => 'string', format => 'pve-qm-drive',
 399         typetext => 'volume',
 400         description => "This is an alias for option -ide2",
 401     },
 402     cpu => {
 403         optional => 1,
 404         description => "Emulated CPU type.",
 405         type => 'string',
 406         enum => [ qw(486 athlon pentium pentium2 pentium3 coreduo core2duo kvm32 kvm64 qemu32 qemu64 phenom Conroe Penryn Nehalem Westmere SandyBridge Haswell Opteron_G1 Opteron_G2 Opteron_G3 Opteron_G4 Opteron_G5 host) ],
 407         default => 'kvm64',
 408     },
 409     parent => get_standard_option('pve-snapshot-name', {
 410         optional => 1,
 411         description => "Parent snapshot name. This is used internally, and should not be modified.",
 412     }),
 413     snaptime => {
 414         optional => 1,
 415         description => "Timestamp for snapshots.",
 416         type => 'integer',
 417         minimum => 0,
 418     },
 419     vmstate => {
 420         optional => 1,
 421         type => 'string', format => 'pve-volume-id',
 422         description => "Reference to a volume which stores the VM state. This is used internally for snapshots.",
 423     },
 424     machine => {
 425         description => "Specific the Qemu machine type.",
 426         type => 'string',
 427         pattern => '(pc|pc(-i440fx)?-\d+\.\d+|q35|pc-q35-\d+\.\d+)',
 428         maxLength => 40,
 429         optional => 1,
 430     },
 431 };
 432
 433 # what about other qemu settings ?
 434 #cpu => 'string',
 435 #machine => 'string',
 436 #fda => 'file',
 437 #fdb => 'file',
 438 #mtdblock => 'file',
 439 #sd => 'file',
 440 #pflash => 'file',
 441 #snapshot => 'bool',
 442 #bootp => 'file',
 443 ##tftp => 'dir',
 444 ##smb => 'dir',
 445 #kernel => 'file',
 446 #append => 'string',
 447 #initrd => 'file',
 448 ##soundhw => 'string',
 449
 450 while (my ($k, $v) = each %$confdesc) {
 451     PVE::JSONSchema::register_standard_option("pve-qm-$k", $v);
 452 }
 453
 454 my $MAX_IDE_DISKS = 4;
 455 my $MAX_SCSI_DISKS = 14;
 456 my $MAX_VIRTIO_DISKS = 16;
 457 my $MAX_SATA_DISKS = 6;
 458 my $MAX_USB_DEVICES = 5;
 459 my $MAX_NETS = 32;
 460 my $MAX_UNUSED_DISKS = 8;
 461 my $MAX_HOSTPCI_DEVICES = 2;
 462 my $MAX_SERIAL_PORTS = 4;
 463 my $MAX_PARALLEL_PORTS = 3;
 464
 465 my $nic_model_list = ['rtl8139', 'ne2k_pci', 'e1000',  'pcnet',  'virtio',
 466                       'ne2k_isa', 'i82551', 'i82557b', 'i82559er'];
 467 my $nic_model_list_txt = join(' ', sort @$nic_model_list);
 468
 469 my $netdesc = {
 470     optional => 1,
 471     type => 'string', format => 'pve-qm-net',
 472     typetext => "MODEL=XX:XX:XX:XX:XX:XX [,bridge=<dev>][,rate=<mbps>][,tag=<vlanid>]",
 473     description => <<EODESCR,
 474 Specify network devices.
 475
 476 MODEL is one of: $nic_model_list_txt
 477
 478 XX:XX:XX:XX:XX:XX should be an unique MAC address. This is
 479 automatically generated if not specified.
 480
 481 The bridge parameter can be used to automatically add the interface to a bridge device. The Proxmox VE standard bridge is called 'vmbr0'.
 482
 483 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'.
 484
 485 If you specify no bridge, we create a kvm 'user' (NATed) network device, which provides DHCP and DNS services. The following addresses are used:
 486
 487 10.0.2.2   Gateway
 488 10.0.2.3   DNS Server
 489 10.0.2.4   SMB Server
 490
 491 The DHCP server assign addresses to the guest starting from 10.0.2.15.
 492
 493 EODESCR
 494 };
 495 PVE::JSONSchema::register_standard_option("pve-qm-net", $netdesc);
 496
 497 for (my $i = 0; $i < $MAX_NETS; $i++)  {
 498     $confdesc->{"net$i"} = $netdesc;
 499 }
 500
 501 my $drivename_hash;
 502
 503 my $idedesc = {
 504     optional => 1,
 505     type => 'string', format => 'pve-qm-drive',
 506     typetext => '[volume=]volume,] [,media=cdrom|disk] [,cyls=c,heads=h,secs=s[,trans=t]] [,snapshot=on|off] [,cache=none|writethrough|writeback|unsafe|directsync] [,format=f] [,backup=yes|no] [,rerror=ignore|report|stop] [,werror=enospc|ignore|report|stop] [,aio=native|threads]',
 507     description => "Use volume as IDE hard disk or CD-ROM (n is 0 to " .($MAX_IDE_DISKS -1) . ").",
 508 };
 509 PVE::JSONSchema::register_standard_option("pve-qm-ide", $idedesc);
 510
 511 my $scsidesc = {
 512     optional => 1,
 513     type => 'string', format => 'pve-qm-drive',
 514     typetext => '[volume=]volume,] [,media=cdrom|disk] [,cyls=c,heads=h,secs=s[,trans=t]] [,snapshot=on|off] [,cache=none|writethrough|writeback|unsafe|directsync] [,format=f] [,backup=yes|no] [,rerror=ignore|report|stop] [,werror=enospc|ignore|report|stop] [,aio=native|threads]',
 515     description => "Use volume as SCSI hard disk or CD-ROM (n is 0 to " . ($MAX_SCSI_DISKS - 1) . ").",
 516 };
 517 PVE::JSONSchema::register_standard_option("pve-qm-scsi", $scsidesc);
 518
 519 my $satadesc = {
 520     optional => 1,
 521     type => 'string', format => 'pve-qm-drive',
 522     typetext => '[volume=]volume,] [,media=cdrom|disk] [,cyls=c,heads=h,secs=s[,trans=t]] [,snapshot=on|off] [,cache=none|writethrough|writeback|unsafe|directsync] [,format=f] [,backup=yes|no] [,rerror=ignore|report|stop] [,werror=enospc|ignore|report|stop] [,aio=native|threads]',
 523     description => "Use volume as SATA hard disk or CD-ROM (n is 0 to " . ($MAX_SATA_DISKS - 1). ").",
 524 };
 525 PVE::JSONSchema::register_standard_option("pve-qm-sata", $satadesc);
 526
 527 my $virtiodesc = {
 528     optional => 1,
 529     type => 'string', format => 'pve-qm-drive',
 530     typetext => '[volume=]volume,] [,media=cdrom|disk] [,cyls=c,heads=h,secs=s[,trans=t]] [,snapshot=on|off] [,cache=none|writethrough|writeback|unsafe|directsync] [,format=f] [,backup=yes|no] [,rerror=ignore|report|stop] [,werror=enospc|ignore|report|stop] [,aio=native|threads]',
 531     description => "Use volume as VIRTIO hard disk (n is 0 to " . ($MAX_VIRTIO_DISKS - 1) . ").",
 532 };
 533 PVE::JSONSchema::register_standard_option("pve-qm-virtio", $virtiodesc);
 534
 535 my $usbdesc = {
 536     optional => 1,
 537     type => 'string', format => 'pve-qm-usb-device',
 538     typetext => 'host=HOSTUSBDEVICE',
 539     description => <<EODESCR,
 540 Configure an USB device (n is 0 to 4). This can be used to
 541 pass-through usb devices to the guest. HOSTUSBDEVICE syntax is:
 542
 543 'bus-port(.port)*' (decimal numbers) or
 544 'vendor_id:product_id' (hexadeciaml numbers)
 545
 546 You can use the 'lsusb -t' command to list existing usb devices.
 547
 548 Note: This option allows direct access to host hardware. So it is no longer possible to migrate such machines - use with special care.
 549
 550 EODESCR
 551 };
 552 PVE::JSONSchema::register_standard_option("pve-qm-usb", $usbdesc);
 553
 554 my $hostpcidesc = {
 555         optional => 1,
 556         type => 'string', format => 'pve-qm-hostpci',
 557         typetext => "HOSTPCIDEVICE",
 558         description => <<EODESCR,
 559 Map host pci devices. HOSTPCIDEVICE syntax is:
 560
 561 'bus:dev.func' (hexadecimal numbers)
 562
 563 You can us the 'lspci' command to list existing pci devices.
 564
 565 Note: This option allows direct access to host hardware. So it is no longer possible to migrate such machines - use with special care.
 566
 567 Experimental: user reported problems with this option.
 568 EODESCR
 569 };
 570 PVE::JSONSchema::register_standard_option("pve-qm-hostpci", $hostpcidesc);
 571
 572 my $serialdesc = {
 573         optional => 1,
 574         type => 'string',
 575         pattern => '/dev/ttyS\d+',
 576         description =>  <<EODESCR,
 577 Map host serial devices (n is 0 to 3).
 578
 579 Note: This option allows direct access to host hardware. So it is no longer possible to migrate such machines - use with special care.
 580
 581 Experimental: user reported problems with this option.
 582 EODESCR
 583 };
 584
 585 my $paralleldesc= {
 586         optional => 1,
 587         type => 'string',
 588         pattern => '/dev/parport\d+',
 589         description =>  <<EODESCR,
 590 Map host parallel devices (n is 0 to 2).
 591
 592 Note: This option allows direct access to host hardware. So it is no longer possible to migrate such machines - use with special care.
 593
 594 Experimental: user reported problems with this option.
 595 EODESCR
 596 };
 597
 598 for (my $i = 0; $i < $MAX_PARALLEL_PORTS; $i++)  {
 599     $confdesc->{"parallel$i"} = $paralleldesc;
 600 }
 601
 602 for (my $i = 0; $i < $MAX_SERIAL_PORTS; $i++)  {
 603     $confdesc->{"serial$i"} = $serialdesc;
 604 }
 605
 606 for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
 607     $confdesc->{"hostpci$i"} = $hostpcidesc;
 608 }
 609
 610 for (my $i = 0; $i < $MAX_IDE_DISKS; $i++)  {
 611     $drivename_hash->{"ide$i"} = 1;
 612     $confdesc->{"ide$i"} = $idedesc;
 613 }
 614
 615 for (my $i = 0; $i < $MAX_SATA_DISKS; $i++)  {
 616     $drivename_hash->{"sata$i"} = 1;
 617     $confdesc->{"sata$i"} = $satadesc;
 618 }
 619
 620 for (my $i = 0; $i < $MAX_SCSI_DISKS; $i++)  {
 621     $drivename_hash->{"scsi$i"} = 1;
 622     $confdesc->{"scsi$i"} = $scsidesc ;
 623 }
 624
 625 for (my $i = 0; $i < $MAX_VIRTIO_DISKS; $i++)  {
 626     $drivename_hash->{"virtio$i"} = 1;
 627     $confdesc->{"virtio$i"} = $virtiodesc;
 628 }
 629
 630 for (my $i = 0; $i < $MAX_USB_DEVICES; $i++)  {
 631     $confdesc->{"usb$i"} = $usbdesc;
 632 }
 633
 634 my $unuseddesc = {
 635     optional => 1,
 636     type => 'string', format => 'pve-volume-id',
 637     description => "Reference to unused volumes.",
 638 };
 639
 640 for (my $i = 0; $i < $MAX_UNUSED_DISKS; $i++)  {
 641     $confdesc->{"unused$i"} = $unuseddesc;
 642 }
 643
 644 my $kvm_api_version = 0;
 645
 646 sub kvm_version {
 647
 648     return $kvm_api_version if $kvm_api_version;
 649
 650     my $fh = IO::File->new("</dev/kvm") ||
 651         return 0;
 652
 653     if (my $v = $fh->ioctl(KVM_GET_API_VERSION(), 0)) {
 654         $kvm_api_version = $v;
 655     }
 656
 657     $fh->close();
 658
 659     return  $kvm_api_version;
 660 }
 661
 662 my $kvm_user_version;
 663
 664 sub kvm_user_version {
 665
 666     return $kvm_user_version if $kvm_user_version;
 667
 668     $kvm_user_version = 'unknown';
 669
 670     my $tmp = `kvm -help 2>/dev/null`;
 671
 672     if ($tmp =~ m/^QEMU( PC)? emulator version (\d+\.\d+(\.\d+)?)[,\s]/) {
 673         $kvm_user_version = $2;
 674     }
 675
 676     return $kvm_user_version;
 677
 678 }
 679
 680 my $kernel_has_vhost_net = -c '/dev/vhost-net';
 681
 682 sub disknames {
 683     # order is important - used to autoselect boot disk
 684     return ((map { "ide$_" } (0 .. ($MAX_IDE_DISKS - 1))),
 685             (map { "scsi$_" } (0 .. ($MAX_SCSI_DISKS - 1))),
 686             (map { "virtio$_" } (0 .. ($MAX_VIRTIO_DISKS - 1))),
 687             (map { "sata$_" } (0 .. ($MAX_SATA_DISKS - 1))));
 688 }
 689
 690 sub valid_drivename {
 691     my $dev = shift;
 692
 693     return defined($drivename_hash->{$dev});
 694 }
 695
 696 sub option_exists {
 697     my $key = shift;
 698     return defined($confdesc->{$key});
 699 }
 700
 701 sub nic_models {
 702     return $nic_model_list;
 703 }
 704
 705 sub os_list_description {
 706
 707     return {
 708         other => 'Other',
 709         wxp => 'Windows XP',
 710         w2k => 'Windows 2000',
 711         w2k3 =>, 'Windows 2003',
 712         w2k8 => 'Windows 2008',
 713         wvista => 'Windows Vista',
 714         win7 => 'Windows 7',
 715         win8 => 'Windows 8/2012',
 716         l24 => 'Linux 2.4',
 717         l26 => 'Linux 2.6',
 718     };
 719 }
 720
 721 my $cdrom_path;
 722
 723 sub get_cdrom_path {
 724
 725     return  $cdrom_path if $cdrom_path;
 726
 727     return $cdrom_path = "/dev/cdrom" if -l "/dev/cdrom";
 728     return $cdrom_path = "/dev/cdrom1" if -l "/dev/cdrom1";
 729     return $cdrom_path = "/dev/cdrom2" if -l "/dev/cdrom2";
 730 }
 731
 732 sub get_iso_path {
 733     my ($storecfg, $vmid, $cdrom) = @_;
 734
 735     if ($cdrom eq 'cdrom') {
 736         return get_cdrom_path();
 737     } elsif ($cdrom eq 'none') {
 738         return '';
 739     } elsif ($cdrom =~ m|^/|) {
 740         return $cdrom;
 741     } else {
 742         return PVE::Storage::path($storecfg, $cdrom);
 743     }
 744 }
 745
 746 # try to convert old style file names to volume IDs
 747 sub filename_to_volume_id {
 748     my ($vmid, $file, $media) = @_;
 749
 750     if (!($file eq 'none' || $file eq 'cdrom' ||
 751           $file =~ m|^/dev/.+| || $file =~ m/^([^:]+):(.+)$/)) {
 752
 753         return undef if $file =~ m|/|;
 754
 755         if ($media && $media eq 'cdrom') {
 756             $file = "local:iso/$file";
 757         } else {
 758             $file = "local:$vmid/$file";
 759         }
 760     }
 761
 762     return $file;
 763 }
 764
 765 sub verify_media_type {
 766     my ($opt, $vtype, $media) = @_;
 767
 768     return if !$media;
 769
 770     my $etype;
 771     if ($media eq 'disk') {
 772         $etype = 'images';
 773     } elsif ($media eq 'cdrom') {
 774         $etype = 'iso';
 775     } else {
 776         die "internal error";
 777     }
 778
 779     return if ($vtype eq $etype);
 780
 781     raise_param_exc({ $opt => "unexpected media type ($vtype != $etype)" });
 782 }
 783
 784 sub cleanup_drive_path {
 785     my ($opt, $storecfg, $drive) = @_;
 786
 787     # try to convert filesystem paths to volume IDs
 788
 789     if (($drive->{file} !~ m/^(cdrom|none)$/) &&
 790         ($drive->{file} !~ m|^/dev/.+|) &&
 791         ($drive->{file} !~ m/^([^:]+):(.+)$/) &&
 792         ($drive->{file} !~ m/^\d+$/)) {
 793         my ($vtype, $volid) = PVE::Storage::path_to_volume_id($storecfg, $drive->{file});
 794         raise_param_exc({ $opt => "unable to associate path '$drive->{file}' to any storage"}) if !$vtype;
 795         $drive->{media} = 'cdrom' if !$drive->{media} && $vtype eq 'iso';
 796         verify_media_type($opt, $vtype, $drive->{media});
 797         $drive->{file} = $volid;
 798     }
 799
 800     $drive->{media} = 'cdrom' if !$drive->{media} && $drive->{file} =~ m/^(cdrom|none)$/;
 801 }
 802
 803 sub create_conf_nolock {
 804     my ($vmid, $settings) = @_;
 805
 806     my $filename = config_file($vmid);
 807
 808     die "configuration file '$filename' already exists\n" if -f $filename;
 809
 810     my $defaults = load_defaults();
 811
 812     $settings->{name} = "vm$vmid" if !$settings->{name};
 813     $settings->{memory} = $defaults->{memory} if !$settings->{memory};
 814
 815     my $data = '';
 816     foreach my $opt (keys %$settings) {
 817         next if !$confdesc->{$opt};
 818
 819         my $value = $settings->{$opt};
 820         next if !$value;
 821
 822         $data .= "$opt: $value\n";
 823     }
 824
 825     PVE::Tools::file_set_contents($filename, $data);
 826 }
 827
 828 my $parse_size = sub {
 829     my ($value) = @_;
 830
 831     return undef if $value !~ m/^(\d+(\.\d+)?)([KMG])?$/;
 832     my ($size, $unit) = ($1, $3);
 833     if ($unit) {
 834         if ($unit eq 'K') {
 835             $size = $size * 1024;
 836         } elsif ($unit eq 'M') {
 837             $size = $size * 1024 * 1024;
 838         } elsif ($unit eq 'G') {
 839             $size = $size * 1024 * 1024 * 1024;
 840         }
 841     }
 842     return int($size);
 843 };
 844
 845 my $format_size = sub {
 846     my ($size) = @_;
 847
 848     $size = int($size);
 849
 850     my $kb = int($size/1024);
 851     return $size if $kb*1024 != $size;
 852
 853     my $mb = int($kb/1024);
 854     return "${kb}K" if $mb*1024 != $kb;
 855
 856     my $gb = int($mb/1024);
 857     return "${mb}M" if $gb*1024 != $mb;
 858
 859     return "${gb}G";
 860 };
 861
 862 # ideX = [volume=]volume-id[,media=d][,cyls=c,heads=h,secs=s[,trans=t]]
 863 #        [,snapshot=on|off][,cache=on|off][,format=f][,backup=yes|no]
 864 #        [,rerror=ignore|report|stop][,werror=enospc|ignore|report|stop]
 865 #        [,aio=native|threads]
 866
 867 sub parse_drive {
 868     my ($key, $data) = @_;
 869
 870     my $res = {};
 871
 872     # $key may be undefined - used to verify JSON parameters
 873     if (!defined($key)) {
 874         $res->{interface} = 'unknown'; # should not harm when used to verify parameters
 875         $res->{index} = 0;
 876     } elsif ($key =~ m/^([^\d]+)(\d+)$/) {
 877         $res->{interface} = $1;
 878         $res->{index} = $2;
 879     } else {
 880         return undef;
 881     }
 882
 883     foreach my $p (split (/,/, $data)) {
 884         next if $p =~ m/^\s*$/;
 885
 886         if ($p =~ m/^(file|volume|cyls|heads|secs|trans|media|snapshot|cache|format|rerror|werror|backup|aio|bps|mbps|bps_rd|mbps_rd|bps_wr|mbps_wr|iops|iops_rd|iops_wr|size)=(.+)$/) {
 887             my ($k, $v) = ($1, $2);
 888
 889             $k = 'file' if $k eq 'volume';
 890
 891             return undef if defined $res->{$k};
 892
 893             if ($k eq 'bps' || $k eq 'bps_rd' || $k eq 'bps_wr') {
 894                 return undef if !$v || $v !~ m/^\d+/;
 895                 $k = "m$k";
 896                 $v = sprintf("%.3f", $v / (1024*1024));
 897             }
 898             $res->{$k} = $v;
 899         } else {
 900             if (!$res->{file} && $p !~ m/=/) {
 901                 $res->{file} = $p;
 902             } else {
 903                 return undef;
 904             }
 905         }
 906     }
 907
 908     return undef if !$res->{file};
 909
 910     if($res->{file} =~ m/\.(raw|cow|qcow|qcow2|vmdk|cloop)$/){
 911         $res->{format} = $1;
 912     }
 913
 914     return undef if $res->{cache} &&
 915         $res->{cache} !~ m/^(off|none|writethrough|writeback|unsafe|directsync)$/;
 916     return undef if $res->{snapshot} && $res->{snapshot} !~ m/^(on|off)$/;
 917     return undef if $res->{cyls} && $res->{cyls} !~ m/^\d+$/;
 918     return undef if $res->{heads} && $res->{heads} !~ m/^\d+$/;
 919     return undef if $res->{secs} && $res->{secs} !~ m/^\d+$/;
 920     return undef if $res->{media} && $res->{media} !~ m/^(disk|cdrom)$/;
 921     return undef if $res->{trans} && $res->{trans} !~ m/^(none|lba|auto)$/;
 922     return undef if $res->{format} && $res->{format} !~ m/^(raw|cow|qcow|qcow2|vmdk|cloop)$/;
 923     return undef if $res->{rerror} && $res->{rerror} !~ m/^(ignore|report|stop)$/;
 924     return undef if $res->{werror} && $res->{werror} !~ m/^(enospc|ignore|report|stop)$/;
 925     return undef if $res->{backup} && $res->{backup} !~ m/^(yes|no)$/;
 926     return undef if $res->{aio} && $res->{aio} !~ m/^(native|threads)$/;
 927
 928
 929     return undef if $res->{mbps_rd} && $res->{mbps};
 930     return undef if $res->{mbps_wr} && $res->{mbps};
 931
 932     return undef if $res->{mbps} && $res->{mbps} !~ m/^\d+(\.\d+)?$/;
 933     return undef if $res->{mbps_rd} && $res->{mbps_rd} !~ m/^\d+(\.\d+)?$/;
 934     return undef if $res->{mbps_wr} && $res->{mbps_wr} !~ m/^\d+(\.\d+)?$/;
 935
 936     return undef if $res->{iops_rd} && $res->{iops};
 937     return undef if $res->{iops_wr} && $res->{iops};
 938     return undef if $res->{iops} && $res->{iops} !~ m/^\d+$/;
 939     return undef if $res->{iops_rd} && $res->{iops_rd} !~ m/^\d+$/;
 940     return undef if $res->{iops_wr} && $res->{iops_wr} !~ m/^\d+$/;
 941
 942
 943     if ($res->{size}) {
 944         return undef if !defined($res->{size} = &$parse_size($res->{size}));
 945     }
 946
 947     if ($res->{media} && ($res->{media} eq 'cdrom')) {
 948         return undef if $res->{snapshot} || $res->{trans} || $res->{format};
 949         return undef if $res->{heads} || $res->{secs} || $res->{cyls};
 950         return undef if $res->{interface} eq 'virtio';
 951     }
 952
 953     # rerror does not work with scsi drives
 954     if ($res->{rerror}) {
 955         return undef if $res->{interface} eq 'scsi';
 956     }
 957
 958     return $res;
 959 }
 960
 961 my @qemu_drive_options = qw(heads secs cyls trans media format cache snapshot rerror werror aio iops iops_rd iops_wr);
 962
 963 sub print_drive {
 964     my ($vmid, $drive) = @_;
 965
 966     my $opts = '';
 967     foreach my $o (@qemu_drive_options, 'mbps', 'mbps_rd', 'mbps_wr', 'backup') {
 968         $opts .= ",$o=$drive->{$o}" if $drive->{$o};
 969     }
 970
 971     if ($drive->{size}) {
 972         $opts .= ",size=" . &$format_size($drive->{size});
 973     }
 974
 975     return "$drive->{file}$opts";
 976 }
 977
 978 sub scsi_inquiry {
 979     my($fh, $noerr) = @_;
 980
 981     my $SG_IO = 0x2285;
 982     my $SG_GET_VERSION_NUM = 0x2282;
 983
 984     my $versionbuf = "\x00" x 8;
 985     my $ret = ioctl($fh, $SG_GET_VERSION_NUM, $versionbuf);
 986     if (!$ret) {
 987         die "scsi ioctl SG_GET_VERSION_NUM failoed - $!\n" if !$noerr;
 988         return undef;
 989     }
 990     my $version = unpack("I", $versionbuf);
 991     if ($version < 30000) {
 992         die "scsi generic interface too old\n"  if !$noerr;
 993         return undef;
 994     }
 995
 996     my $buf = "\x00" x 36;
 997     my $sensebuf = "\x00" x 8;
 998     my $cmd = pack("C x3 C x11", 0x12, 36);
 999
1000     # see /usr/include/scsi/sg.h
1001     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";
1002
1003     my $packet = pack($sg_io_hdr_t, ord('S'), -3, length($cmd),
1004                       length($sensebuf), 0, length($buf), $buf,
1005                       $cmd, $sensebuf, 6000);
1006
1007     $ret = ioctl($fh, $SG_IO, $packet);
1008     if (!$ret) {
1009         die "scsi ioctl SG_IO failed - $!\n" if !$noerr;
1010         return undef;
1011     }
1012
1013     my @res = unpack($sg_io_hdr_t, $packet);
1014     if ($res[17] || $res[18]) {
1015         die "scsi ioctl SG_IO status error - $!\n" if !$noerr;
1016         return undef;
1017     }
1018
1019     my $res = {};
1020     ($res->{device}, $res->{removable}, $res->{venodor},
1021      $res->{product}, $res->{revision}) = unpack("C C x6 A8 A16 A4", $buf);
1022
1023     return $res;
1024 }
1025
1026 sub path_is_scsi {
1027     my ($path) = @_;
1028
1029     my $fh = IO::File->new("+<$path") || return undef;
1030     my $res = scsi_inquiry($fh, 1);
1031     close($fh);
1032
1033     return $res;
1034 }
1035
1036 sub print_drivedevice_full {
1037     my ($storecfg, $conf, $vmid, $drive, $bridges) = @_;
1038
1039     my $device = '';
1040     my $maxdev = 0;
1041
1042     if ($drive->{interface} eq 'virtio') {
1043         my $pciaddr = print_pci_addr("$drive->{interface}$drive->{index}", $bridges);
1044         $device = "virtio-blk-pci,drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}$pciaddr";
1045     } elsif ($drive->{interface} eq 'scsi') {
1046         $maxdev = ($conf->{scsihw} && $conf->{scsihw} ne 'lsi') ? 256 : 7;
1047         my $controller = int($drive->{index} / $maxdev);
1048         my $unit = $drive->{index} % $maxdev;
1049         my $devicetype = 'hd';
1050         my $path = '';
1051         if (drive_is_cdrom($drive)) {
1052               $devicetype = 'cd';
1053           } else {
1054               if ($drive->{file} =~ m|^/|) {
1055                   $path = $drive->{file};
1056               } else {
1057                   $path = PVE::Storage::path($storecfg, $drive->{file});
1058               }
1059
1060               if($path =~ m/^iscsi\:\/\//){
1061                  $devicetype = 'generic';
1062               }
1063               else {
1064                  $devicetype = 'block' if path_is_scsi($path);
1065               }
1066          }
1067
1068         if (!$conf->{scsihw} || $conf->{scsihw} eq 'lsi'){
1069             $device = "scsi-$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';
1070         } else {
1071             $device = "scsi-$devicetype,bus=scsihw$controller.0,channel=0,scsi-id=0,lun=$drive->{index},drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}";
1072         }
1073
1074     } elsif ($drive->{interface} eq 'ide'){
1075         $maxdev = 2;
1076         my $controller = int($drive->{index} / $maxdev);
1077         my $unit = $drive->{index} % $maxdev;
1078         my $devicetype = ($drive->{media} && $drive->{media} eq 'cdrom') ? "cd" : "hd";
1079
1080         $device = "ide-$devicetype,bus=ide.$controller,unit=$unit,drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}";
1081     } elsif ($drive->{interface} eq 'sata'){
1082         my $controller = int($drive->{index} / $MAX_SATA_DISKS);
1083         my $unit = $drive->{index} % $MAX_SATA_DISKS;
1084         $device = "ide-drive,bus=ahci$controller.$unit,drive=drive-$drive->{interface}$drive->{index},id=$drive->{interface}$drive->{index}";
1085     } elsif ($drive->{interface} eq 'usb') {
1086         die "implement me";
1087         #  -device ide-drive,bus=ide.1,unit=0,drive=drive-ide0-1-0,id=ide0-1-0
1088     } else {
1089         die "unsupported interface type";
1090     }
1091
1092     $device .= ",bootindex=$drive->{bootindex}" if $drive->{bootindex};
1093
1094     return $device;
1095 }
1096
1097 sub print_drive_full {
1098     my ($storecfg, $vmid, $drive) = @_;
1099
1100     my $opts = '';
1101     foreach my $o (@qemu_drive_options) {
1102         next if $o eq 'bootindex';
1103         $opts .= ",$o=$drive->{$o}" if $drive->{$o};
1104     }
1105
1106     foreach my $o (qw(bps bps_rd bps_wr)) {
1107         my $v = $drive->{"m$o"};
1108         $opts .= ",$o=" . int($v*1024*1024) if $v;
1109     }
1110
1111     # use linux-aio by default (qemu default is threads)
1112     $opts .= ",aio=native" if !$drive->{aio};
1113
1114     my $path;
1115     my $volid = $drive->{file};
1116     if (drive_is_cdrom($drive)) {
1117         $path = get_iso_path($storecfg, $vmid, $volid);
1118     } else {
1119         if ($volid =~ m|^/|) {
1120             $path = $volid;
1121         } else {
1122             $path = PVE::Storage::path($storecfg, $volid);
1123         }
1124     }
1125
1126     $opts .= ",cache=none" if !$drive->{cache} && !drive_is_cdrom($drive);
1127
1128     my $pathinfo = $path ? "file=$path," : '';
1129
1130     return "${pathinfo}if=none,id=drive-$drive->{interface}$drive->{index}$opts";
1131 }
1132
1133 sub print_netdevice_full {
1134     my ($vmid, $conf, $net, $netid, $bridges) = @_;
1135
1136     my $bootorder = $conf->{boot} || $confdesc->{boot}->{default};
1137
1138     my $device = $net->{model};
1139     if ($net->{model} eq 'virtio') {
1140          $device = 'virtio-net-pci';
1141      };
1142
1143     # qemu > 0.15 always try to boot from network - we disable that by
1144     # not loading the pxe rom file
1145     my $extra = ($bootorder !~ m/n/) ? "romfile=," : '';
1146     my $pciaddr = print_pci_addr("$netid", $bridges);
1147     my $tmpstr = "$device,${extra}mac=$net->{macaddr},netdev=$netid$pciaddr,id=$netid";
1148     $tmpstr .= ",bootindex=$net->{bootindex}" if $net->{bootindex} ;
1149     return $tmpstr;
1150 }
1151
1152 sub print_netdev_full {
1153     my ($vmid, $conf, $net, $netid) = @_;
1154
1155     my $i = '';
1156     if ($netid =~ m/^net(\d+)$/) {
1157         $i = int($1);
1158     }
1159
1160     die "got strange net id '$i'\n" if $i >= ${MAX_NETS};
1161
1162     my $ifname = "tap${vmid}i$i";
1163
1164     # kvm uses TUNSETIFF ioctl, and that limits ifname length
1165     die "interface name '$ifname' is too long (max 15 character)\n"
1166         if length($ifname) >= 16;
1167
1168     my $vhostparam = '';
1169     $vhostparam = ',vhost=on' if $kernel_has_vhost_net && $net->{model} eq 'virtio';
1170
1171     my $vmname = $conf->{name} || "vm$vmid";
1172
1173     if ($net->{bridge}) {
1174         return "type=tap,id=$netid,ifname=${ifname},script=/var/lib/qemu-server/pve-bridge$vhostparam";
1175     } else {
1176         return "type=user,id=$netid,hostname=$vmname";
1177     }
1178 }
1179
1180 sub drive_is_cdrom {
1181     my ($drive) = @_;
1182
1183     return $drive && $drive->{media} && ($drive->{media} eq 'cdrom');
1184
1185 }
1186
1187 sub parse_hostpci {
1188     my ($value) = @_;
1189
1190     return undef if !$value;
1191
1192     my $res = {};
1193
1194     if ($value =~ m/^[a-f0-9]{2}:[a-f0-9]{2}\.[a-f0-9]$/) {
1195        $res->{pciid} = $value;
1196     } else {
1197        return undef;
1198     }
1199
1200     return $res;
1201 }
1202
1203 # netX: e1000=XX:XX:XX:XX:XX:XX,bridge=vmbr0,rate=<mbps>
1204 sub parse_net {
1205     my ($data) = @_;
1206
1207     my $res = {};
1208
1209     foreach my $kvp (split(/,/, $data)) {
1210
1211         if ($kvp =~ m/^(ne2k_pci|e1000|rtl8139|pcnet|virtio|ne2k_isa|i82551|i82557b|i82559er)(=([0-9a-f]{2}(:[0-9a-f]{2}){5}))?$/i) {
1212             my $model = lc($1);
1213             my $mac = defined($3) ? uc($3) : PVE::Tools::random_ether_addr();
1214             $res->{model} = $model;
1215             $res->{macaddr} = $mac;
1216         } elsif ($kvp =~ m/^bridge=(\S+)$/) {
1217             $res->{bridge} = $1;
1218         } elsif ($kvp =~ m/^rate=(\d+(\.\d+)?)$/) {
1219             $res->{rate} = $1;
1220         } elsif ($kvp =~ m/^tag=(\d+)$/) {
1221             $res->{tag} = $1;
1222         } else {
1223             return undef;
1224         }
1225
1226     }
1227
1228     return undef if !$res->{model};
1229
1230     return $res;
1231 }
1232
1233 sub print_net {
1234     my $net = shift;
1235
1236     my $res = "$net->{model}";
1237     $res .= "=$net->{macaddr}" if $net->{macaddr};
1238     $res .= ",bridge=$net->{bridge}" if $net->{bridge};
1239     $res .= ",rate=$net->{rate}" if $net->{rate};
1240     $res .= ",tag=$net->{tag}" if $net->{tag};
1241
1242     return $res;
1243 }
1244
1245 sub add_random_macs {
1246     my ($settings) = @_;
1247
1248     foreach my $opt (keys %$settings) {
1249         next if $opt !~ m/^net(\d+)$/;
1250         my $net = parse_net($settings->{$opt});
1251         next if !$net;
1252         $settings->{$opt} = print_net($net);
1253     }
1254 }
1255
1256 sub add_unused_volume {
1257     my ($config, $volid) = @_;
1258
1259     my $key;
1260     for (my $ind = $MAX_UNUSED_DISKS - 1; $ind >= 0; $ind--) {
1261         my $test = "unused$ind";
1262         if (my $vid = $config->{$test}) {
1263             return if $vid eq $volid; # do not add duplicates
1264         } else {
1265             $key = $test;
1266         }
1267     }
1268
1269     die "To many unused volume - please delete them first.\n" if !$key;
1270
1271     $config->{$key} = $volid;
1272
1273     return $key;
1274 }
1275
1276 PVE::JSONSchema::register_format('pve-qm-bootdisk', \&verify_bootdisk);
1277 sub verify_bootdisk {
1278     my ($value, $noerr) = @_;
1279
1280     return $value if valid_drivename($value);
1281
1282     return undef if $noerr;
1283
1284     die "invalid boot disk '$value'\n";
1285 }
1286
1287 PVE::JSONSchema::register_format('pve-qm-net', \&verify_net);
1288 sub verify_net {
1289     my ($value, $noerr) = @_;
1290
1291     return $value if parse_net($value);
1292
1293     return undef if $noerr;
1294
1295     die "unable to parse network options\n";
1296 }
1297
1298 PVE::JSONSchema::register_format('pve-qm-drive', \&verify_drive);
1299 sub verify_drive {
1300     my ($value, $noerr) = @_;
1301
1302     return $value if parse_drive(undef, $value);
1303
1304     return undef if $noerr;
1305
1306     die "unable to parse drive options\n";
1307 }
1308
1309 PVE::JSONSchema::register_format('pve-qm-hostpci', \&verify_hostpci);
1310 sub verify_hostpci {
1311     my ($value, $noerr) = @_;
1312
1313     return $value if parse_hostpci($value);
1314
1315     return undef if $noerr;
1316
1317     die "unable to parse pci id\n";
1318 }
1319
1320 PVE::JSONSchema::register_format('pve-qm-watchdog', \&verify_watchdog);
1321 sub verify_watchdog {
1322     my ($value, $noerr) = @_;
1323
1324     return $value if parse_watchdog($value);
1325
1326     return undef if $noerr;
1327
1328     die "unable to parse watchdog options\n";
1329 }
1330
1331 sub parse_watchdog {
1332     my ($value) = @_;
1333
1334     return undef if !$value;
1335
1336     my $res = {};
1337
1338     foreach my $p (split(/,/, $value)) {
1339         next if $p =~ m/^\s*$/;
1340
1341         if ($p =~ m/^(model=)?(i6300esb|ib700)$/) {
1342             $res->{model} = $2;
1343         } elsif ($p =~ m/^(action=)?(reset|shutdown|poweroff|pause|debug|none)$/) {
1344             $res->{action} = $2;
1345         } else {
1346             return undef;
1347         }
1348     }
1349
1350     return $res;
1351 }
1352
1353 PVE::JSONSchema::register_format('pve-qm-startup', \&verify_startup);
1354 sub verify_startup {
1355     my ($value, $noerr) = @_;
1356
1357     return $value if parse_startup($value);
1358
1359     return undef if $noerr;
1360
1361     die "unable to parse startup options\n";
1362 }
1363
1364 sub parse_startup {
1365     my ($value) = @_;
1366
1367     return undef if !$value;
1368
1369     my $res = {};
1370
1371     foreach my $p (split(/,/, $value)) {
1372         next if $p =~ m/^\s*$/;
1373
1374         if ($p =~ m/^(order=)?(\d+)$/) {
1375             $res->{order} = $2;
1376         } elsif ($p =~ m/^up=(\d+)$/) {
1377             $res->{up} = $1;
1378         } elsif ($p =~ m/^down=(\d+)$/) {
1379             $res->{down} = $1;
1380         } else {
1381             return undef;
1382         }
1383     }
1384
1385     return $res;
1386 }
1387
1388 sub parse_usb_device {
1389     my ($value) = @_;
1390
1391     return undef if !$value;
1392
1393     my @dl = split(/,/, $value);
1394     my $found;
1395
1396     my $res = {};
1397     foreach my $v (@dl) {
1398         if ($v =~ m/^host=(0x)?([0-9A-Fa-f]{4}):(0x)?([0-9A-Fa-f]{4})$/) {
1399             $found = 1;
1400             $res->{vendorid} = $2;
1401             $res->{productid} = $4;
1402         } elsif ($v =~ m/^host=(\d+)\-(\d+(\.\d+)*)$/) {
1403             $found = 1;
1404             $res->{hostbus} = $1;
1405             $res->{hostport} = $2;
1406         } else {
1407             return undef;
1408         }
1409     }
1410     return undef if !$found;
1411
1412     return $res;
1413 }
1414
1415 PVE::JSONSchema::register_format('pve-qm-usb-device', \&verify_usb_device);
1416 sub verify_usb_device {
1417     my ($value, $noerr) = @_;
1418
1419     return $value if parse_usb_device($value);
1420
1421     return undef if $noerr;
1422
1423     die "unable to parse usb device\n";
1424 }
1425
1426 # add JSON properties for create and set function
1427 sub json_config_properties {
1428     my $prop = shift;
1429
1430     foreach my $opt (keys %$confdesc) {
1431         next if $opt eq 'parent' || $opt eq 'snaptime' || $opt eq 'vmstate';
1432         $prop->{$opt} = $confdesc->{$opt};
1433     }
1434
1435     return $prop;
1436 }
1437
1438 sub check_type {
1439     my ($key, $value) = @_;
1440
1441     die "unknown setting '$key'\n" if !$confdesc->{$key};
1442
1443     my $type = $confdesc->{$key}->{type};
1444
1445     if (!defined($value)) {
1446         die "got undefined value\n";
1447     }
1448
1449     if ($value =~ m/[\n\r]/) {
1450         die "property contains a line feed\n";
1451     }
1452
1453     if ($type eq 'boolean') {
1454         return 1 if ($value eq '1') || ($value =~ m/^(on|yes|true)$/i);
1455         return 0 if ($value eq '0') || ($value =~ m/^(off|no|false)$/i);
1456         die "type check ('boolean') failed - got '$value'\n";
1457     } elsif ($type eq 'integer') {
1458         return int($1) if $value =~ m/^(\d+)$/;
1459         die "type check ('integer') failed - got '$value'\n";
1460     } elsif ($type eq 'number') {
1461         return $value if $value =~ m/^(\d+)(\.\d+)?$/;
1462         die "type check ('number') failed - got '$value'\n";
1463     } elsif ($type eq 'string') {
1464         if (my $fmt = $confdesc->{$key}->{format}) {
1465             if ($fmt eq 'pve-qm-drive') {
1466                 # special case - we need to pass $key to parse_drive()
1467                 my $drive = parse_drive($key, $value);
1468                 return $value if $drive;
1469                 die "unable to parse drive options\n";
1470             }
1471             PVE::JSONSchema::check_format($fmt, $value);
1472             return $value;
1473         }
1474         $value =~ s/^\"(.*)\"$/$1/;
1475         return $value;
1476     } else {
1477         die "internal error"
1478     }
1479 }
1480
1481 sub lock_config_full {
1482     my ($vmid, $timeout, $code, @param) = @_;
1483
1484     my $filename = config_file_lock($vmid);
1485
1486     my $res = lock_file($filename, $timeout, $code, @param);
1487
1488     die $@ if $@;
1489
1490     return $res;
1491 }
1492
1493 sub lock_config_mode {
1494     my ($vmid, $timeout, $shared, $code, @param) = @_;
1495
1496     my $filename = config_file_lock($vmid);
1497
1498     my $res = lock_file_full($filename, $timeout, $shared, $code, @param);
1499
1500     die $@ if $@;
1501
1502     return $res;
1503 }
1504
1505 sub lock_config {
1506     my ($vmid, $code, @param) = @_;
1507
1508     return lock_config_full($vmid, 10, $code, @param);
1509 }
1510
1511 sub cfs_config_path {
1512     my ($vmid, $node) = @_;
1513
1514     $node = $nodename if !$node;
1515     return "nodes/$node/qemu-server/$vmid.conf";
1516 }
1517
1518 sub check_iommu_support{
1519     #fixme : need to check IOMMU support
1520     #http://www.linux-kvm.org/page/How_to_assign_devices_with_VT-d_in_KVM
1521
1522     my $iommu=1;
1523     return $iommu;
1524
1525 }
1526
1527 sub config_file {
1528     my ($vmid, $node) = @_;
1529
1530     my $cfspath = cfs_config_path($vmid, $node);
1531     return "/etc/pve/$cfspath";
1532 }
1533
1534 sub config_file_lock {
1535     my ($vmid) = @_;
1536
1537     return "$lock_dir/lock-$vmid.conf";
1538 }
1539
1540 sub touch_config {
1541     my ($vmid) = @_;
1542
1543     my $conf = config_file($vmid);
1544     utime undef, undef, $conf;
1545 }
1546
1547 sub destroy_vm {
1548     my ($storecfg, $vmid, $keep_empty_config) = @_;
1549
1550     my $conffile = config_file($vmid);
1551
1552     my $conf = load_config($vmid);
1553
1554     check_lock($conf);
1555
1556     # only remove disks owned by this VM
1557     foreach_drive($conf, sub {
1558         my ($ds, $drive) = @_;
1559
1560         return if drive_is_cdrom($drive);
1561
1562         my $volid = $drive->{file};
1563
1564         return if !$volid || $volid =~ m|^/|;
1565
1566         my ($path, $owner) = PVE::Storage::path($storecfg, $volid);
1567         return if !$path || !$owner || ($owner != $vmid);
1568
1569         PVE::Storage::vdisk_free($storecfg, $volid);
1570     });
1571
1572     if ($keep_empty_config) {
1573         PVE::Tools::file_set_contents($conffile, "memory: 128\n");
1574     } else {
1575         unlink $conffile;
1576     }
1577
1578     # also remove unused disk
1579     eval {
1580         my $dl = PVE::Storage::vdisk_list($storecfg, undef, $vmid);
1581
1582         eval {
1583             PVE::Storage::foreach_volid($dl, sub {
1584                 my ($volid, $sid, $volname, $d) = @_;
1585                 PVE::Storage::vdisk_free($storecfg, $volid);
1586             });
1587         };
1588         warn $@ if $@;
1589
1590     };
1591     warn $@ if $@;
1592 }
1593
1594 sub load_config {
1595     my ($vmid, $node) = @_;
1596
1597     my $cfspath = cfs_config_path($vmid, $node);
1598
1599     my $conf = PVE::Cluster::cfs_read_file($cfspath);
1600
1601     die "no such VM ('$vmid')\n" if !defined($conf);
1602
1603     return $conf;
1604 }
1605
1606 sub parse_vm_config {
1607     my ($filename, $raw) = @_;
1608
1609     return undef if !defined($raw);
1610
1611     my $res = {
1612         digest => Digest::SHA::sha1_hex($raw),
1613         snapshots => {},
1614     };
1615
1616     $filename =~ m|/qemu-server/(\d+)\.conf$|
1617         || die "got strange filename '$filename'";
1618
1619     my $vmid = $1;
1620
1621     my $conf = $res;
1622     my $descr = '';
1623
1624     my @lines = split(/\n/, $raw);
1625     foreach my $line (@lines) {
1626         next if $line =~ m/^\s*$/;
1627
1628         if ($line =~ m/^\[([a-z][a-z0-9_\-]+)\]\s*$/i) {
1629             my $snapname = $1;
1630             $conf->{description} = $descr if $descr;
1631             $descr = '';
1632             $conf = $res->{snapshots}->{$snapname} = {};
1633             next;
1634         }
1635
1636         if ($line =~ m/^\#(.*)\s*$/) {
1637             $descr .= PVE::Tools::decode_text($1) . "\n";
1638             next;
1639         }
1640
1641         if ($line =~ m/^(description):\s*(.*\S)\s*$/) {
1642             $descr .= PVE::Tools::decode_text($2);
1643         } elsif ($line =~ m/snapstate:\s*(prepare|delete)\s*$/) {
1644             $conf->{snapstate} = $1;
1645         } elsif ($line =~ m/^(args):\s*(.*\S)\s*$/) {
1646             my $key = $1;
1647             my $value = $2;
1648             $conf->{$key} = $value;
1649         } elsif ($line =~ m/^([a-z][a-z_]*\d*):\s*(\S+)\s*$/) {
1650             my $key = $1;
1651             my $value = $2;
1652             eval { $value = check_type($key, $value); };
1653             if ($@) {
1654                 warn "vm $vmid - unable to parse value of '$key' - $@";
1655             } else {
1656                 my $fmt = $confdesc->{$key}->{format};
1657                 if ($fmt && $fmt eq 'pve-qm-drive') {
1658                     my $v = parse_drive($key, $value);
1659                     if (my $volid = filename_to_volume_id($vmid, $v->{file}, $v->{media})) {
1660                         $v->{file} = $volid;
1661                         $value = print_drive($vmid, $v);
1662                     } else {
1663                         warn "vm $vmid - unable to parse value of '$key'\n";
1664                         next;
1665                     }
1666                 }
1667
1668                 if ($key eq 'cdrom') {
1669                     $conf->{ide2} = $value;
1670                 } else {
1671                     $conf->{$key} = $value;
1672                 }
1673             }
1674         }
1675     }
1676
1677     $conf->{description} = $descr if $descr;
1678
1679     delete $res->{snapstate}; # just to be sure
1680
1681     return $res;
1682 }
1683
1684 sub write_vm_config {
1685     my ($filename, $conf) = @_;
1686
1687     delete $conf->{snapstate}; # just to be sure
1688
1689     if ($conf->{cdrom}) {
1690         die "option ide2 conflicts with cdrom\n" if $conf->{ide2};
1691         $conf->{ide2} = $conf->{cdrom};
1692         delete $conf->{cdrom};
1693     }
1694
1695     # we do not use 'smp' any longer
1696     if ($conf->{sockets}) {
1697         delete $conf->{smp};
1698     } elsif ($conf->{smp}) {
1699         $conf->{sockets} = $conf->{smp};
1700         delete $conf->{cores};
1701         delete $conf->{smp};
1702     }
1703
1704     my $used_volids = {};
1705
1706     my $cleanup_config = sub {
1707         my ($cref, $snapname) = @_;
1708
1709         foreach my $key (keys %$cref) {
1710             next if $key eq 'digest' || $key eq 'description' || $key eq 'snapshots' ||
1711                 $key eq 'snapstate';
1712             my $value = $cref->{$key};
1713             eval { $value = check_type($key, $value); };
1714             die "unable to parse value of '$key' - $@" if $@;
1715
1716             $cref->{$key} = $value;
1717
1718             if (!$snapname && valid_drivename($key)) {
1719                 my $drive = parse_drive($key, $value);
1720                 $used_volids->{$drive->{file}} = 1 if $drive && $drive->{file};
1721             }
1722         }
1723     };
1724
1725     &$cleanup_config($conf);
1726     foreach my $snapname (keys %{$conf->{snapshots}}) {
1727         &$cleanup_config($conf->{snapshots}->{$snapname}, $snapname);
1728     }
1729
1730     # remove 'unusedX' settings if we re-add a volume
1731     foreach my $key (keys %$conf) {
1732         my $value = $conf->{$key};
1733         if ($key =~ m/^unused/ && $used_volids->{$value}) {
1734             delete $conf->{$key};
1735         }
1736     }
1737
1738     my $generate_raw_config = sub {
1739         my ($conf) = @_;
1740
1741         my $raw = '';
1742
1743         # add description as comment to top of file
1744         my $descr = $conf->{description} || '';
1745         foreach my $cl (split(/\n/, $descr)) {
1746             $raw .= '#' .  PVE::Tools::encode_text($cl) . "\n";
1747         }
1748
1749         foreach my $key (sort keys %$conf) {
1750             next if $key eq 'digest' || $key eq 'description' || $key eq 'snapshots';
1751             $raw .= "$key: $conf->{$key}\n";
1752         }
1753         return $raw;
1754     };
1755
1756     my $raw = &$generate_raw_config($conf);
1757     foreach my $snapname (sort keys %{$conf->{snapshots}}) {
1758         $raw .= "\n[$snapname]\n";
1759         $raw .= &$generate_raw_config($conf->{snapshots}->{$snapname});
1760     }
1761
1762     return $raw;
1763 }
1764
1765 sub update_config_nolock {
1766     my ($vmid, $conf, $skiplock) = @_;
1767
1768     check_lock($conf) if !$skiplock;
1769
1770     my $cfspath = cfs_config_path($vmid);
1771
1772     PVE::Cluster::cfs_write_file($cfspath, $conf);
1773 }
1774
1775 sub update_config {
1776     my ($vmid, $conf, $skiplock) = @_;
1777
1778     lock_config($vmid, &update_config_nolock, $conf, $skiplock);
1779 }
1780
1781 sub load_defaults {
1782
1783     my $res = {};
1784
1785     # we use static defaults from our JSON schema configuration
1786     foreach my $key (keys %$confdesc) {
1787         if (defined(my $default = $confdesc->{$key}->{default})) {
1788             $res->{$key} = $default;
1789         }
1790     }
1791
1792     my $conf = PVE::Cluster::cfs_read_file('datacenter.cfg');
1793     $res->{keyboard} = $conf->{keyboard} if $conf->{keyboard};
1794
1795     return $res;
1796 }
1797
1798 sub config_list {
1799     my $vmlist = PVE::Cluster::get_vmlist();
1800     my $res = {};
1801     return $res if !$vmlist || !$vmlist->{ids};
1802     my $ids = $vmlist->{ids};
1803
1804     foreach my $vmid (keys %$ids) {
1805         my $d = $ids->{$vmid};
1806         next if !$d->{node} || $d->{node} ne $nodename;
1807         next if !$d->{type} || $d->{type} ne 'qemu';
1808         $res->{$vmid}->{exists} = 1;
1809     }
1810     return $res;
1811 }
1812
1813 # test if VM uses local resources (to prevent migration)
1814 sub check_local_resources {
1815     my ($conf, $noerr) = @_;
1816
1817     my $loc_res = 0;
1818
1819     $loc_res = 1 if $conf->{hostusb}; # old syntax
1820     $loc_res = 1 if $conf->{hostpci}; # old syntax
1821
1822     foreach my $k (keys %$conf) {
1823         $loc_res = 1 if $k =~ m/^(usb|hostpci|serial|parallel)\d+$/;
1824     }
1825
1826     die "VM uses local resources\n" if $loc_res && !$noerr;
1827
1828     return $loc_res;
1829 }
1830
1831 # check if used storages are available on all nodes (use by migrate)
1832 sub check_storage_availability {
1833     my ($storecfg, $conf, $node) = @_;
1834
1835     foreach_drive($conf, sub {
1836         my ($ds, $drive) = @_;
1837
1838         my $volid = $drive->{file};
1839         return if !$volid;
1840
1841         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1842         return if !$sid;
1843
1844         # check if storage is available on both nodes
1845         my $scfg = PVE::Storage::storage_check_node($storecfg, $sid);
1846         PVE::Storage::storage_check_node($storecfg, $sid, $node);
1847    });
1848 }
1849
1850 # list nodes where all VM images are available (used by has_feature API)
1851 sub shared_nodes {
1852     my ($conf, $storecfg) = @_;
1853
1854     my $nodelist = PVE::Cluster::get_nodelist();
1855     my $nodehash = { map { $_ => 1 } @$nodelist };
1856     my $nodename = PVE::INotify::nodename();
1857
1858     foreach_drive($conf, sub {
1859         my ($ds, $drive) = @_;
1860
1861         my $volid = $drive->{file};
1862         return if !$volid;
1863
1864         my ($storeid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1865         if ($storeid) {
1866             my $scfg = PVE::Storage::storage_config($storecfg, $storeid);
1867             if ($scfg->{disable}) {
1868                 $nodehash = {};
1869             } elsif (my $avail = $scfg->{nodes}) {
1870                 foreach my $node (keys %$nodehash) {
1871                     delete $nodehash->{$node} if !$avail->{$node};
1872                 }
1873             } elsif (!$scfg->{shared}) {
1874                 foreach my $node (keys %$nodehash) {
1875                     delete $nodehash->{$node} if $node ne $nodename
1876                 }
1877             }
1878         }
1879     });
1880
1881     return $nodehash
1882 }
1883
1884 sub check_lock {
1885     my ($conf) = @_;
1886
1887     die "VM is locked ($conf->{lock})\n" if $conf->{lock};
1888 }
1889
1890 sub check_cmdline {
1891     my ($pidfile, $pid) = @_;
1892
1893     my $fh = IO::File->new("/proc/$pid/cmdline", "r");
1894     if (defined($fh)) {
1895         my $line = <$fh>;
1896         $fh->close;
1897         return undef if !$line;
1898         my @param = split(/\0/, $line);
1899
1900         my $cmd = $param[0];
1901         return if !$cmd || ($cmd !~ m|kvm$| && $cmd !~ m|qemu-system-x86_64$|);
1902
1903         for (my $i = 0; $i < scalar (@param); $i++) {
1904             my $p = $param[$i];
1905             next if !$p;
1906             if (($p eq '-pidfile') || ($p eq '--pidfile')) {
1907                 my $p = $param[$i+1];
1908                 return 1 if $p && ($p eq $pidfile);
1909                 return undef;
1910             }
1911         }
1912     }
1913     return undef;
1914 }
1915
1916 sub check_running {
1917     my ($vmid, $nocheck, $node) = @_;
1918
1919     my $filename = config_file($vmid, $node);
1920
1921     die "unable to find configuration file for VM $vmid - no such machine\n"
1922         if !$nocheck && ! -f $filename;
1923
1924     my $pidfile = pidfile_name($vmid);
1925
1926     if (my $fd = IO::File->new("<$pidfile")) {
1927         my $st = stat($fd);
1928         my $line = <$fd>;
1929         close($fd);
1930
1931         my $mtime = $st->mtime;
1932         if ($mtime > time()) {
1933             warn "file '$filename' modified in future\n";
1934         }
1935
1936         if ($line =~ m/^(\d+)$/) {
1937             my $pid = $1;
1938             if (check_cmdline($pidfile, $pid)) {
1939                 if (my $pinfo = PVE::ProcFSTools::check_process_running($pid)) {
1940                     return $pid;
1941                 }
1942             }
1943         }
1944     }
1945
1946     return undef;
1947 }
1948
1949 sub vzlist {
1950
1951     my $vzlist = config_list();
1952
1953     my $fd = IO::Dir->new($var_run_tmpdir) || return $vzlist;
1954
1955     while (defined(my $de = $fd->read)) {
1956         next if $de !~ m/^(\d+)\.pid$/;
1957         my $vmid = $1;
1958         next if !defined($vzlist->{$vmid});
1959         if (my $pid = check_running($vmid)) {
1960             $vzlist->{$vmid}->{pid} = $pid;
1961         }
1962     }
1963
1964     return $vzlist;
1965 }
1966
1967 sub disksize {
1968     my ($storecfg, $conf) = @_;
1969
1970     my $bootdisk = $conf->{bootdisk};
1971     return undef if !$bootdisk;
1972     return undef if !valid_drivename($bootdisk);
1973
1974     return undef if !$conf->{$bootdisk};
1975
1976     my $drive = parse_drive($bootdisk, $conf->{$bootdisk});
1977     return undef if !defined($drive);
1978
1979     return undef if drive_is_cdrom($drive);
1980
1981     my $volid = $drive->{file};
1982     return undef if !$volid;
1983
1984     return $drive->{size};
1985 }
1986
1987 my $last_proc_pid_stat;
1988
1989 # get VM status information
1990 # This must be fast and should not block ($full == false)
1991 # We only query KVM using QMP if $full == true (this can be slow)
1992 sub vmstatus {
1993     my ($opt_vmid, $full) = @_;
1994
1995     my $res = {};
1996
1997     my $storecfg = PVE::Storage::config();
1998
1999     my $list = vzlist();
2000     my ($uptime) = PVE::ProcFSTools::read_proc_uptime(1);
2001
2002     my $cpucount = $cpuinfo->{cpus} || 1;
2003
2004     foreach my $vmid (keys %$list) {
2005         next if $opt_vmid && ($vmid ne $opt_vmid);
2006
2007         my $cfspath = cfs_config_path($vmid);
2008         my $conf = PVE::Cluster::cfs_read_file($cfspath) || {};
2009
2010         my $d = {};
2011         $d->{pid} = $list->{$vmid}->{pid};
2012
2013         # fixme: better status?
2014         $d->{status} = $list->{$vmid}->{pid} ? 'running' : 'stopped';
2015
2016         my $size = disksize($storecfg, $conf);
2017         if (defined($size)) {
2018             $d->{disk} = 0; # no info available
2019             $d->{maxdisk} = $size;
2020         } else {
2021             $d->{disk} = 0;
2022             $d->{maxdisk} = 0;
2023         }
2024
2025         $d->{cpus} = ($conf->{sockets} || 1) * ($conf->{cores} || 1);
2026         $d->{cpus} = $cpucount if $d->{cpus} > $cpucount;
2027
2028         $d->{name} = $conf->{name} || "VM $vmid";
2029         $d->{maxmem} = $conf->{memory} ? $conf->{memory}*(1024*1024) : 0;
2030
2031         if ($conf->{balloon}) {
2032             $d->{balloon_min} = $conf->{balloon}*(1024*1024);
2033             $d->{shares} = defined($conf->{shares}) ? $conf->{shares} : 1000;
2034         }
2035
2036         $d->{uptime} = 0;
2037         $d->{cpu} = 0;
2038         $d->{mem} = 0;
2039
2040         $d->{netout} = 0;
2041         $d->{netin} = 0;
2042
2043         $d->{diskread} = 0;
2044         $d->{diskwrite} = 0;
2045
2046         $d->{template} = is_template($conf);
2047
2048         $res->{$vmid} = $d;
2049     }
2050
2051     my $netdev = PVE::ProcFSTools::read_proc_net_dev();
2052     foreach my $dev (keys %$netdev) {
2053         next if $dev !~ m/^tap([1-9]\d*)i/;
2054         my $vmid = $1;
2055         my $d = $res->{$vmid};
2056         next if !$d;
2057
2058         $d->{netout} += $netdev->{$dev}->{receive};
2059         $d->{netin} += $netdev->{$dev}->{transmit};
2060     }
2061
2062     my $ctime = gettimeofday;
2063
2064     foreach my $vmid (keys %$list) {
2065
2066         my $d = $res->{$vmid};
2067         my $pid = $d->{pid};
2068         next if !$pid;
2069
2070         my $pstat = PVE::ProcFSTools::read_proc_pid_stat($pid);
2071         next if !$pstat; # not running
2072
2073         my $used = $pstat->{utime} + $pstat->{stime};
2074
2075         $d->{uptime} = int(($uptime - $pstat->{starttime})/$cpuinfo->{user_hz});
2076
2077         if ($pstat->{vsize}) {
2078             $d->{mem} = int(($pstat->{rss}/$pstat->{vsize})*$d->{maxmem});
2079         }
2080
2081         my $old = $last_proc_pid_stat->{$pid};
2082         if (!$old) {
2083             $last_proc_pid_stat->{$pid} = {
2084                 time => $ctime,
2085                 used => $used,
2086                 cpu => 0,
2087             };
2088             next;
2089         }
2090
2091         my $dtime = ($ctime -  $old->{time}) * $cpucount * $cpuinfo->{user_hz};
2092
2093         if ($dtime > 1000) {
2094             my $dutime = $used -  $old->{used};
2095
2096             $d->{cpu} = (($dutime/$dtime)* $cpucount) / $d->{cpus};
2097             $last_proc_pid_stat->{$pid} = {
2098                 time => $ctime,
2099                 used => $used,
2100                 cpu => $d->{cpu},
2101             };
2102         } else {
2103             $d->{cpu} = $old->{cpu};
2104         }
2105     }
2106
2107     return $res if !$full;
2108
2109     my $qmpclient = PVE::QMPClient->new();
2110
2111     my $ballooncb = sub {
2112         my ($vmid, $resp) = @_;
2113
2114         my $info = $resp->{'return'};
2115         return if !$info->{max_mem};
2116
2117         my $d = $res->{$vmid};
2118
2119         # use memory assigned to VM
2120         $d->{maxmem} = $info->{max_mem};
2121         $d->{balloon} = $info->{actual};
2122
2123         if (defined($info->{total_mem}) && defined($info->{free_mem})) {
2124             $d->{mem} = $info->{total_mem} - $info->{free_mem};
2125             $d->{freemem} = $info->{free_mem};
2126         }
2127
2128     };
2129
2130     my $blockstatscb = sub {
2131         my ($vmid, $resp) = @_;
2132         my $data = $resp->{'return'} || [];
2133         my $totalrdbytes = 0;
2134         my $totalwrbytes = 0;
2135         for my $blockstat (@$data) {
2136             $totalrdbytes = $totalrdbytes + $blockstat->{stats}->{rd_bytes};
2137             $totalwrbytes = $totalwrbytes + $blockstat->{stats}->{wr_bytes};
2138         }
2139         $res->{$vmid}->{diskread} = $totalrdbytes;
2140         $res->{$vmid}->{diskwrite} = $totalwrbytes;
2141     };
2142
2143     my $statuscb = sub {
2144         my ($vmid, $resp) = @_;
2145
2146         $qmpclient->queue_cmd($vmid, $blockstatscb, 'query-blockstats');
2147         # this fails if ballon driver is not loaded, so this must be
2148         # the last commnand (following command are aborted if this fails).
2149         $qmpclient->queue_cmd($vmid, $ballooncb, 'query-balloon');
2150
2151         my $status = 'unknown';
2152         if (!defined($status = $resp->{'return'}->{status})) {
2153             warn "unable to get VM status\n";
2154             return;
2155         }
2156
2157         $res->{$vmid}->{qmpstatus} = $resp->{'return'}->{status};
2158     };
2159
2160     foreach my $vmid (keys %$list) {
2161         next if $opt_vmid && ($vmid ne $opt_vmid);
2162         next if !$res->{$vmid}->{pid}; # not running
2163         $qmpclient->queue_cmd($vmid, $statuscb, 'query-status');
2164     }
2165
2166     $qmpclient->queue_execute();
2167
2168     foreach my $vmid (keys %$list) {
2169         next if $opt_vmid && ($vmid ne $opt_vmid);
2170         $res->{$vmid}->{qmpstatus} = $res->{$vmid}->{status} if !$res->{$vmid}->{qmpstatus};
2171     }
2172
2173     return $res;
2174 }
2175
2176 sub foreach_drive {
2177     my ($conf, $func) = @_;
2178
2179     foreach my $ds (keys %$conf) {
2180         next if !valid_drivename($ds);
2181
2182         my $drive = parse_drive($ds, $conf->{$ds});
2183         next if !$drive;
2184
2185         &$func($ds, $drive);
2186     }
2187 }
2188
2189 sub foreach_volid {
2190     my ($conf, $func) = @_;
2191
2192     my $volhash = {};
2193
2194     my $test_volid = sub {
2195         my ($volid, $is_cdrom) = @_;
2196
2197         return if !$volid;
2198
2199         $volhash->{$volid} = $is_cdrom || 0;
2200     };
2201
2202     foreach_drive($conf, sub {
2203         my ($ds, $drive) = @_;
2204         &$test_volid($drive->{file}, drive_is_cdrom($drive));
2205     });
2206
2207     foreach my $snapname (keys %{$conf->{snapshots}}) {
2208         my $snap = $conf->{snapshots}->{$snapname};
2209         &$test_volid($snap->{vmstate}, 0);
2210         foreach_drive($snap, sub {
2211             my ($ds, $drive) = @_;
2212             &$test_volid($drive->{file}, drive_is_cdrom($drive));
2213         });
2214     }
2215
2216     foreach my $volid (keys %$volhash) {
2217         &$func($volid, $volhash->{$volid});
2218     }
2219 }
2220
2221 sub config_to_command {
2222     my ($storecfg, $vmid, $conf, $defaults, $forcemachine) = @_;
2223
2224     my $cmd = [];
2225     my $globalFlags = [];
2226     my $machineFlags = [];
2227     my $rtcFlags = [];
2228     my $cpuFlags = [];
2229     my $devices = [];
2230     my $pciaddr = '';
2231     my $bridges = {};
2232     my $kvmver = kvm_user_version();
2233     my $vernum = 0; # unknown
2234     if ($kvmver =~ m/^(\d+)\.(\d+)$/) {
2235         $vernum = $1*1000000+$2*1000;
2236     } elsif ($kvmver =~ m/^(\d+)\.(\d+)\.(\d+)$/) {
2237         $vernum = $1*1000000+$2*1000+$3;
2238     }
2239
2240     die "detected old qemu-kvm binary ($kvmver)\n" if $vernum < 15000;
2241
2242     my $have_ovz = -f '/proc/vz/vestat';
2243
2244     push @$cmd, '/usr/bin/kvm';
2245
2246     push @$cmd, '-id', $vmid;
2247
2248     my $use_virtio = 0;
2249
2250     my $qmpsocket = qmp_socket($vmid);
2251     push @$cmd, '-chardev', "socket,id=qmp,path=$qmpsocket,server,nowait";
2252     push @$cmd, '-mon', "chardev=qmp,mode=control";
2253
2254     my $socket = vnc_socket($vmid);
2255     push @$cmd,  '-vnc', "unix:$socket,x509,password";
2256
2257     push @$cmd, '-pidfile' , pidfile_name($vmid);
2258
2259     push @$cmd, '-daemonize';
2260
2261     $pciaddr = print_pci_addr("piix3", $bridges);
2262     push @$devices, '-device', "piix3-usb-uhci,id=uhci$pciaddr.0x2";
2263
2264     my $use_usb2 = 0;
2265     for (my $i = 0; $i < $MAX_USB_DEVICES; $i++)  {
2266         next if !$conf->{"usb$i"};
2267         $use_usb2 = 1;
2268     }
2269     # include usb device config
2270     push @$devices, '-readconfig', '/usr/share/qemu-server/pve-usb.cfg' if $use_usb2;
2271
2272     # enable absolute mouse coordinates (needed by vnc)
2273     my $tablet = defined($conf->{tablet}) ? $conf->{tablet} : $defaults->{tablet};
2274     push @$devices, '-device', 'usb-tablet,id=tablet,bus=uhci.0,port=1' if $tablet;
2275
2276     # host pci devices
2277     for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
2278           my $d = parse_hostpci($conf->{"hostpci$i"});
2279           next if !$d;
2280           $pciaddr = print_pci_addr("hostpci$i", $bridges);
2281           push @$devices, '-device', "pci-assign,host=$d->{pciid},id=hostpci$i$pciaddr";
2282     }
2283
2284     # usb devices
2285     for (my $i = 0; $i < $MAX_USB_DEVICES; $i++)  {
2286         my $d = parse_usb_device($conf->{"usb$i"});
2287         next if !$d;
2288         if ($d->{vendorid} && $d->{productid}) {
2289             push @$devices, '-device', "usb-host,vendorid=0x$d->{vendorid},productid=0x$d->{productid}";
2290         } elsif (defined($d->{hostbus}) && defined($d->{hostport})) {
2291             push @$devices, '-device', "usb-host,hostbus=$d->{hostbus},hostport=$d->{hostport}";
2292         }
2293     }
2294
2295     # serial devices
2296     for (my $i = 0; $i < $MAX_SERIAL_PORTS; $i++)  {
2297         if (my $path = $conf->{"serial$i"}) {
2298             die "no such serial device\n" if ! -c $path;
2299             push @$devices, '-chardev', "tty,id=serial$i,path=$path";
2300             push @$devices, '-device', "isa-serial,chardev=serial$i";
2301         }
2302     }
2303
2304     # parallel devices
2305     for (my $i = 0; $i < $MAX_PARALLEL_PORTS; $i++)  {
2306         if (my $path = $conf->{"parallel$i"}) {
2307             die "no such parallel device\n" if ! -c $path;
2308             push @$devices, '-chardev', "parport,id=parallel$i,path=$path";
2309             push @$devices, '-device', "isa-parallel,chardev=parallel$i";
2310         }
2311     }
2312
2313     my $vmname = $conf->{name} || "vm$vmid";
2314
2315     push @$cmd, '-name', $vmname;
2316
2317     my $sockets = 1;
2318     $sockets = $conf->{smp} if $conf->{smp}; # old style - no longer iused
2319     $sockets = $conf->{sockets} if  $conf->{sockets};
2320
2321     my $cores = $conf->{cores} || 1;
2322     push @$cmd, '-smp', "sockets=$sockets,cores=$cores";
2323
2324     push @$cmd, '-nodefaults';
2325
2326     my $bootorder = $conf->{boot} || $confdesc->{boot}->{default};
2327
2328     my $bootindex_hash = {};
2329     my $i = 1;
2330     foreach my $o (split(//, $bootorder)) {
2331         $bootindex_hash->{$o} = $i*100;
2332         $i++;
2333     }
2334
2335     push @$cmd, '-boot', "menu=on";
2336
2337     push @$cmd, '-no-acpi' if defined($conf->{acpi}) && $conf->{acpi} == 0;
2338
2339     push @$cmd, '-no-reboot' if  defined($conf->{reboot}) && $conf->{reboot} == 0;
2340
2341     my $vga = $conf->{vga};
2342     if (!$vga) {
2343         if ($conf->{ostype} && ($conf->{ostype} eq 'win8' || $conf->{ostype} eq 'win7' || $conf->{ostype} eq 'w2k8')) {
2344             $vga = 'std';
2345         } else {
2346             $vga = 'cirrus';
2347         }
2348     }
2349
2350     push @$cmd, '-vga', $vga if $vga; # for kvm 77 and later
2351
2352     # time drift fix
2353     my $tdf = defined($conf->{tdf}) ? $conf->{tdf} : $defaults->{tdf};
2354
2355     my $nokvm = defined($conf->{kvm}) && $conf->{kvm} == 0 ? 1 : 0;
2356     my $useLocaltime = $conf->{localtime};
2357
2358     if (my $ost = $conf->{ostype}) {
2359         # other, wxp, w2k, w2k3, w2k8, wvista, win7, win8, l24, l26
2360
2361         if ($ost =~ m/^w/) { # windows
2362             $useLocaltime = 1 if !defined($conf->{localtime});
2363
2364             # use time drift fix when acpi is enabled
2365             if (!(defined($conf->{acpi}) && $conf->{acpi} == 0)) {
2366                 $tdf = 1 if !defined($conf->{tdf});
2367             }
2368         }
2369
2370         if ($ost eq 'win7' || $ost eq 'win8' || $ost eq 'w2k8' ||
2371             $ost eq 'wvista') {
2372             push @$globalFlags, 'kvm-pit.lost_tick_policy=discard';
2373             push @$cmd, '-no-hpet';
2374         }
2375     }
2376
2377     push @$rtcFlags, 'driftfix=slew' if $tdf;
2378
2379     if ($nokvm) {
2380         push @$machineFlags, 'accel=tcg';
2381     } else {
2382         die "No accelerator found!\n" if !$cpuinfo->{hvm};
2383     }
2384
2385     my $machine_type = $forcemachine || $conf->{machine};
2386     if ($machine_type) {
2387         push @$machineFlags, "type=${machine_type}";
2388     }
2389
2390     if ($conf->{startdate}) {
2391         push @$rtcFlags, "base=$conf->{startdate}";
2392     } elsif ($useLocaltime) {
2393         push @$rtcFlags, 'base=localtime';
2394     }
2395
2396     my $cpu = $nokvm ? "qemu64" : "kvm64";
2397     $cpu = $conf->{cpu} if $conf->{cpu};
2398
2399     push @$cpuFlags , '+x2apic' if !$nokvm;
2400
2401     push @$cpuFlags, '+sep' if $cpu eq 'kvm64' || $cpu eq 'kvm32';
2402
2403     $cpu .= ",".join(',', @$cpuFlags) if scalar(@$cpuFlags);
2404
2405     push @$cmd, '-cpu', $cpu;
2406
2407     push @$cmd, '-S' if $conf->{freeze};
2408
2409     # set keyboard layout
2410     my $kb = $conf->{keyboard} || $defaults->{keyboard};
2411     push @$cmd, '-k', $kb if $kb;
2412
2413     # enable sound
2414     #my $soundhw = $conf->{soundhw} || $defaults->{soundhw};
2415     #push @$cmd, '-soundhw', 'es1370';
2416     #push @$cmd, '-soundhw', $soundhw if $soundhw;
2417
2418     if($conf->{agent}) {
2419         my $qgasocket = qga_socket($vmid);
2420         my $pciaddr = print_pci_addr("qga0", $bridges);
2421         push @$devices, '-chardev', "socket,path=$qgasocket,server,nowait,id=qga0";
2422         push @$devices, '-device', "virtio-serial,id=qga0$pciaddr";
2423         push @$devices, '-device', 'virtserialport,chardev=qga0,name=org.qemu.guest_agent.0';
2424     }
2425
2426     if ($vga eq 'qxl') {
2427         my $pciaddr = print_pci_addr("spice", $bridges);
2428
2429         # todo: enable tls
2430         #my $x509 = "x509-key-file=/etc/pve/local/pve-ssl.key";
2431         #$x509 .= ",x509-cert-file=/etc/pve/local/pve-ssl.pem";
2432         #$x509 .= ",x509-cacert-file=/etc/pve/pve-root-ca.pem";
2433
2434         my $socket = spice_socket($vmid);
2435         push @$cmd, '-spice', "unix=$socket";
2436         push @$cmd, '-device', "virtio-serial,id=spice$pciaddr";
2437         push @$cmd, '-chardev', "spicevmc,id=vdagent,name=vdagent";
2438         push @$cmd, '-device', "virtserialport,chardev=vdagent,name=com.redhat.spice.0";
2439     }
2440
2441     # enable balloon by default, unless explicitly disabled
2442     if (!defined($conf->{balloon}) || $conf->{balloon}) {
2443         $pciaddr = print_pci_addr("balloon0", $bridges);
2444         push @$devices, '-device', "virtio-balloon-pci,id=balloon0$pciaddr";
2445     }
2446
2447     if ($conf->{watchdog}) {
2448         my $wdopts = parse_watchdog($conf->{watchdog});
2449         $pciaddr = print_pci_addr("watchdog", $bridges);
2450         my $watchdog = $wdopts->{model} || 'i6300esb';
2451         push @$devices, '-device', "$watchdog$pciaddr";
2452         push @$devices, '-watchdog-action', $wdopts->{action} if $wdopts->{action};
2453     }
2454
2455     my $vollist = [];
2456     my $scsicontroller = {};
2457     my $ahcicontroller = {};
2458     my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : $defaults->{scsihw};
2459
2460     foreach_drive($conf, sub {
2461         my ($ds, $drive) = @_;
2462
2463         if (PVE::Storage::parse_volume_id($drive->{file}, 1)) {
2464             push @$vollist, $drive->{file};
2465         }
2466
2467         $use_virtio = 1 if $ds =~ m/^virtio/;
2468
2469         if (drive_is_cdrom ($drive)) {
2470             if ($bootindex_hash->{d}) {
2471                 $drive->{bootindex} = $bootindex_hash->{d};
2472                 $bootindex_hash->{d} += 1;
2473             }
2474         } else {
2475             if ($bootindex_hash->{c}) {
2476                 $drive->{bootindex} = $bootindex_hash->{c} if $conf->{bootdisk} && ($conf->{bootdisk} eq $ds);
2477                 $bootindex_hash->{c} += 1;
2478             }
2479         }
2480
2481         if ($drive->{interface} eq 'scsi') {
2482
2483             my $maxdev = ($scsihw ne 'lsi') ? 256 : 7;
2484             my $controller = int($drive->{index} / $maxdev);
2485             $pciaddr = print_pci_addr("scsihw$controller", $bridges);
2486             push @$devices, '-device', "$scsihw,id=scsihw$controller$pciaddr" if !$scsicontroller->{$controller};
2487             $scsicontroller->{$controller}=1;
2488         }
2489
2490         if ($drive->{interface} eq 'sata') {
2491            my $controller = int($drive->{index} / $MAX_SATA_DISKS);
2492            $pciaddr = print_pci_addr("ahci$controller", $bridges);
2493            push @$devices, '-device', "ahci,id=ahci$controller,multifunction=on$pciaddr" if !$ahcicontroller->{$controller};
2494            $ahcicontroller->{$controller}=1;
2495         }
2496
2497         push @$devices, '-drive',print_drive_full($storecfg, $vmid, $drive);
2498         push @$devices, '-device',print_drivedevice_full($storecfg, $conf, $vmid, $drive, $bridges);
2499     });
2500
2501     push @$cmd, '-m', $conf->{memory} || $defaults->{memory};
2502
2503     for (my $i = 0; $i < $MAX_NETS; $i++) {
2504          next if !$conf->{"net$i"};
2505          my $d = parse_net($conf->{"net$i"});
2506          next if !$d;
2507
2508          $use_virtio = 1 if $d->{model} eq 'virtio';
2509
2510          if ($bootindex_hash->{n}) {
2511             $d->{bootindex} = $bootindex_hash->{n};
2512             $bootindex_hash->{n} += 1;
2513          }
2514
2515          my $netdevfull = print_netdev_full($vmid,$conf,$d,"net$i");
2516          push @$devices, '-netdev', $netdevfull;
2517
2518          my $netdevicefull = print_netdevice_full($vmid,$conf,$d,"net$i",$bridges);
2519          push @$devices, '-device', $netdevicefull;
2520     }
2521
2522     #bridges
2523     while (my ($k, $v) = each %$bridges) {
2524         $pciaddr = print_pci_addr("pci.$k");
2525         unshift @$devices, '-device', "pci-bridge,id=pci.$k,chassis_nr=$k$pciaddr" if $k > 0;
2526     }
2527
2528
2529     # hack: virtio with fairsched is unreliable, so we do not use fairsched
2530     # when the VM uses virtio devices.
2531     if (!$use_virtio && $have_ovz) {
2532
2533         my $cpuunits = defined($conf->{cpuunits}) ?
2534             $conf->{cpuunits} : $defaults->{cpuunits};
2535
2536         push @$cmd, '-cpuunits', $cpuunits if $cpuunits;
2537
2538         # fixme: cpulimit is currently ignored
2539         #push @$cmd, '-cpulimit', $conf->{cpulimit} if $conf->{cpulimit};
2540     }
2541
2542     # add custom args
2543     if ($conf->{args}) {
2544         my $aa = PVE::Tools::split_args($conf->{args});
2545         push @$cmd, @$aa;
2546     }
2547
2548     push @$cmd, @$devices;
2549     push @$cmd, '-rtc', join(',', @$rtcFlags)
2550         if scalar(@$rtcFlags);
2551     push @$cmd, '-machine', join(',', @$machineFlags)
2552         if scalar(@$machineFlags);
2553     push @$cmd, '-global', join(',', @$globalFlags)
2554         if scalar(@$globalFlags);
2555
2556     return wantarray ? ($cmd, $vollist) : $cmd;
2557 }
2558
2559 sub vnc_socket {
2560     my ($vmid) = @_;
2561     return "${var_run_tmpdir}/$vmid.vnc";
2562 }
2563
2564 sub spice_socket {
2565     my ($vmid) = @_;
2566     return "${var_run_tmpdir}/$vmid.spice";
2567 }
2568
2569 sub qmp_socket {
2570     my ($vmid) = @_;
2571     return "${var_run_tmpdir}/$vmid.qmp";
2572 }
2573
2574 sub qga_socket {
2575     my ($vmid) = @_;
2576     return "${var_run_tmpdir}/$vmid.qga";
2577 }
2578
2579 sub pidfile_name {
2580     my ($vmid) = @_;
2581     return "${var_run_tmpdir}/$vmid.pid";
2582 }
2583
2584 sub vm_devices_list {
2585     my ($vmid) = @_;
2586
2587     my $res = vm_mon_cmd($vmid, 'query-pci');
2588
2589     my $devices = {};
2590     foreach my $pcibus (@$res) {
2591         foreach my $device (@{$pcibus->{devices}}) {
2592             next if !$device->{'qdev_id'};
2593             $devices->{$device->{'qdev_id'}} = $device;
2594         }
2595     }
2596
2597     return $devices;
2598 }
2599
2600 sub vm_deviceplug {
2601     my ($storecfg, $conf, $vmid, $deviceid, $device) = @_;
2602
2603     return 1 if !check_running($vmid);
2604
2605     if ($deviceid eq 'tablet') {
2606         my $devicefull = "usb-tablet,id=tablet,bus=uhci.0,port=1";
2607         qemu_deviceadd($vmid, $devicefull);
2608         return 1;
2609     }
2610
2611     return 1 if !$conf->{hotplug};
2612
2613     my $devices_list = vm_devices_list($vmid);
2614     return 1 if defined($devices_list->{$deviceid});
2615
2616     qemu_bridgeadd($storecfg, $conf, $vmid, $deviceid); #add bridge if we need it for the device
2617
2618     if ($deviceid =~ m/^(virtio)(\d+)$/) {
2619         return undef if !qemu_driveadd($storecfg, $vmid, $device);
2620         my $devicefull = print_drivedevice_full($storecfg, $conf, $vmid, $device);
2621         qemu_deviceadd($vmid, $devicefull);
2622         if(!qemu_deviceaddverify($vmid, $deviceid)) {
2623            qemu_drivedel($vmid, $deviceid);
2624            return undef;
2625         }
2626     }
2627
2628     if ($deviceid =~ m/^(scsihw)(\d+)$/) {
2629         my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : "lsi";
2630         my $pciaddr = print_pci_addr($deviceid);
2631         my $devicefull = "$scsihw,id=$deviceid$pciaddr";
2632         qemu_deviceadd($vmid, $devicefull);
2633         return undef if(!qemu_deviceaddverify($vmid, $deviceid));
2634     }
2635
2636     if ($deviceid =~ m/^(scsi)(\d+)$/) {
2637         return 1 if ($conf->{scsihw} && $conf->{scsihw} ne 'lsi'); #virtio-scsi not yet support hotplug
2638         return undef if !qemu_findorcreatescsihw($storecfg,$conf, $vmid, $device);
2639         return undef if !qemu_driveadd($storecfg, $vmid, $device);
2640         my $devicefull = print_drivedevice_full($storecfg, $conf, $vmid, $device);
2641         if(!qemu_deviceadd($vmid, $devicefull)) {
2642            qemu_drivedel($vmid, $deviceid);
2643            return undef;
2644         }
2645     }
2646
2647     if ($deviceid =~ m/^(net)(\d+)$/) {
2648         return undef if !qemu_netdevadd($vmid, $conf, $device, $deviceid);
2649         my $netdevicefull = print_netdevice_full($vmid, $conf, $device, $deviceid);
2650         qemu_deviceadd($vmid, $netdevicefull);
2651         if(!qemu_deviceaddverify($vmid, $deviceid)) {
2652            qemu_netdevdel($vmid, $deviceid);
2653            return undef;
2654         }
2655     }
2656
2657     if ($deviceid =~ m/^(pci\.)(\d+)$/) {
2658         my $bridgeid = $2;
2659         my $pciaddr = print_pci_addr($deviceid);
2660         my $devicefull = "pci-bridge,id=pci.$bridgeid,chassis_nr=$bridgeid$pciaddr";
2661         qemu_deviceadd($vmid, $devicefull);
2662         return undef if !qemu_deviceaddverify($vmid, $deviceid);
2663     }
2664
2665     return 1;
2666 }
2667
2668 sub vm_deviceunplug {
2669     my ($vmid, $conf, $deviceid) = @_;
2670
2671     return 1 if !check_running ($vmid);
2672
2673     if ($deviceid eq 'tablet') {
2674         qemu_devicedel($vmid, $deviceid);
2675         return 1;
2676     }
2677
2678     return 1 if !$conf->{hotplug};
2679
2680     my $devices_list = vm_devices_list($vmid);
2681     return 1 if !defined($devices_list->{$deviceid});
2682
2683     die "can't unplug bootdisk" if $conf->{bootdisk} && $conf->{bootdisk} eq $deviceid;
2684
2685     if ($deviceid =~ m/^(virtio)(\d+)$/) {
2686         qemu_devicedel($vmid, $deviceid);
2687         return undef if !qemu_devicedelverify($vmid, $deviceid);
2688         return undef if !qemu_drivedel($vmid, $deviceid);
2689     }
2690
2691     if ($deviceid =~ m/^(lsi)(\d+)$/) {
2692         return undef if !qemu_devicedel($vmid, $deviceid);
2693     }
2694
2695     if ($deviceid =~ m/^(scsi)(\d+)$/) {
2696         return undef if !qemu_devicedel($vmid, $deviceid);
2697         return undef if !qemu_drivedel($vmid, $deviceid);
2698     }
2699
2700     if ($deviceid =~ m/^(net)(\d+)$/) {
2701         qemu_devicedel($vmid, $deviceid);
2702         return undef if !qemu_devicedelverify($vmid, $deviceid);
2703         return undef if !qemu_netdevdel($vmid, $deviceid);
2704     }
2705
2706     return 1;
2707 }
2708
2709 sub qemu_deviceadd {
2710     my ($vmid, $devicefull) = @_;
2711
2712     $devicefull = "driver=".$devicefull;
2713     my %options =  split(/[=,]/, $devicefull);
2714
2715     vm_mon_cmd($vmid, "device_add" , %options);
2716     return 1;
2717 }
2718
2719 sub qemu_devicedel {
2720     my($vmid, $deviceid) = @_;
2721     my $ret = vm_mon_cmd($vmid, "device_del", id => $deviceid);
2722     return 1;
2723 }
2724
2725 sub qemu_driveadd {
2726     my($storecfg, $vmid, $device) = @_;
2727
2728     my $drive = print_drive_full($storecfg, $vmid, $device);
2729     my $ret = vm_human_monitor_command($vmid, "drive_add auto $drive");
2730     # If the command succeeds qemu prints: "OK"
2731     if ($ret !~ m/OK/s) {
2732         syslog("err", "adding drive failed: $ret");
2733         return undef;
2734     }
2735     return 1;
2736 }
2737
2738 sub qemu_drivedel {
2739     my($vmid, $deviceid) = @_;
2740
2741     my $ret = vm_human_monitor_command($vmid, "drive_del drive-$deviceid");
2742     $ret =~ s/^\s+//;
2743     if ($ret =~ m/Device \'.*?\' not found/s) {
2744         # NB: device not found errors mean the drive was auto-deleted and we ignore the error
2745     }
2746     elsif ($ret ne "") {
2747       syslog("err", "deleting drive $deviceid failed : $ret");
2748       return undef;
2749     }
2750     return 1;
2751 }
2752
2753 sub qemu_deviceaddverify {
2754     my ($vmid,$deviceid) = @_;
2755
2756     for (my $i = 0; $i <= 5; $i++) {
2757          my $devices_list = vm_devices_list($vmid);
2758          return 1 if defined($devices_list->{$deviceid});
2759          sleep 1;
2760     }
2761     syslog("err", "error on hotplug device $deviceid");
2762     return undef;
2763 }
2764
2765
2766 sub qemu_devicedelverify {
2767     my ($vmid,$deviceid) = @_;
2768
2769     #need to verify the device is correctly remove as device_del is async and empty return is not reliable
2770     for (my $i = 0; $i <= 5; $i++) {
2771          my $devices_list = vm_devices_list($vmid);
2772          return 1 if !defined($devices_list->{$deviceid});
2773          sleep 1;
2774     }
2775     syslog("err", "error on hot-unplugging device $deviceid");
2776     return undef;
2777 }
2778
2779 sub qemu_findorcreatescsihw {
2780     my ($storecfg, $conf, $vmid, $device) = @_;
2781
2782     my $maxdev = ($conf->{scsihw} && $conf->{scsihw} ne 'lsi') ? 256 : 7;
2783     my $controller = int($device->{index} / $maxdev);
2784     my $scsihwid="scsihw$controller";
2785     my $devices_list = vm_devices_list($vmid);
2786
2787     if(!defined($devices_list->{$scsihwid})) {
2788        return undef if !vm_deviceplug($storecfg, $conf, $vmid, $scsihwid);
2789     }
2790     return 1;
2791 }
2792
2793 sub qemu_bridgeadd {
2794     my ($storecfg, $conf, $vmid, $device) = @_;
2795
2796     my $bridges = {};
2797     my $bridgeid = undef;
2798     print_pci_addr($device, $bridges);
2799
2800     while (my ($k, $v) = each %$bridges) {
2801         $bridgeid = $k;
2802     }
2803     return if !$bridgeid || $bridgeid < 1;
2804     my $bridge = "pci.$bridgeid";
2805     my $devices_list = vm_devices_list($vmid);
2806
2807     if(!defined($devices_list->{$bridge})) {
2808         return undef if !vm_deviceplug($storecfg, $conf, $vmid, $bridge);
2809     }
2810     return 1;
2811 }
2812
2813 sub qemu_netdevadd {
2814     my ($vmid, $conf, $device, $deviceid) = @_;
2815
2816     my $netdev = print_netdev_full($vmid, $conf, $device, $deviceid);
2817     my %options =  split(/[=,]/, $netdev);
2818
2819     vm_mon_cmd($vmid, "netdev_add",  %options);
2820     return 1;
2821 }
2822
2823 sub qemu_netdevdel {
2824     my ($vmid, $deviceid) = @_;
2825
2826     vm_mon_cmd($vmid, "netdev_del", id => $deviceid);
2827     return 1;
2828 }
2829
2830 sub qemu_block_set_io_throttle {
2831     my ($vmid, $deviceid, $bps, $bps_rd, $bps_wr, $iops, $iops_rd, $iops_wr) = @_;
2832
2833     return if !check_running($vmid) ;
2834
2835     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($iops_wr));
2836
2837 }
2838
2839 # old code, only used to shutdown old VM after update
2840 sub __read_avail {
2841     my ($fh, $timeout) = @_;
2842
2843     my $sel = new IO::Select;
2844     $sel->add($fh);
2845
2846     my $res = '';
2847     my $buf;
2848
2849     my @ready;
2850     while (scalar (@ready = $sel->can_read($timeout))) {
2851         my $count;
2852         if ($count = $fh->sysread($buf, 8192)) {
2853             if ($buf =~ /^(.*)\(qemu\) $/s) {
2854                 $res .= $1;
2855                 last;
2856             } else {
2857                 $res .= $buf;
2858             }
2859         } else {
2860             if (!defined($count)) {
2861                 die "$!\n";
2862             }
2863             last;
2864         }
2865     }
2866
2867     die "monitor read timeout\n" if !scalar(@ready);
2868
2869     return $res;
2870 }
2871
2872 # old code, only used to shutdown old VM after update
2873 sub vm_monitor_command {
2874     my ($vmid, $cmdstr, $nocheck) = @_;
2875
2876     my $res;
2877
2878     eval {
2879         die "VM $vmid not running\n" if !check_running($vmid, $nocheck);
2880
2881         my $sname = "${var_run_tmpdir}/$vmid.mon";
2882
2883         my $sock = IO::Socket::UNIX->new( Peer => $sname ) ||
2884             die "unable to connect to VM $vmid socket - $!\n";
2885
2886         my $timeout = 3;
2887
2888         # hack: migrate sometime blocks the monitor (when migrate_downtime
2889         # is set)
2890         if ($cmdstr =~ m/^(info\s+migrate|migrate\s)/) {
2891             $timeout = 60*60; # 1 hour
2892         }
2893
2894         # read banner;
2895         my $data = __read_avail($sock, $timeout);
2896
2897         if ($data !~ m/^QEMU\s+(\S+)\s+monitor\s/) {
2898             die "got unexpected qemu monitor banner\n";
2899         }
2900
2901         my $sel = new IO::Select;
2902         $sel->add($sock);
2903
2904         if (!scalar(my @ready = $sel->can_write($timeout))) {
2905             die "monitor write error - timeout";
2906         }
2907
2908         my $fullcmd = "$cmdstr\r";
2909
2910         # syslog('info', "VM $vmid monitor command: $cmdstr");
2911
2912         my $b;
2913         if (!($b = $sock->syswrite($fullcmd)) || ($b != length($fullcmd))) {
2914             die "monitor write error - $!";
2915         }
2916
2917         return if ($cmdstr eq 'q') || ($cmdstr eq 'quit');
2918
2919         $timeout = 20;
2920
2921         if ($cmdstr =~ m/^(info\s+migrate|migrate\s)/) {
2922             $timeout = 60*60; # 1 hour
2923         } elsif ($cmdstr =~ m/^(eject|change)/) {
2924             $timeout = 60; # note: cdrom mount command is slow
2925         }
2926         if ($res = __read_avail($sock, $timeout)) {
2927
2928             my @lines = split("\r?\n", $res);
2929
2930             shift @lines if $lines[0] !~ m/^unknown command/; # skip echo
2931
2932             $res = join("\n", @lines);
2933             $res .= "\n";
2934         }
2935     };
2936
2937     my $err = $@;
2938
2939     if ($err) {
2940         syslog("err", "VM $vmid monitor command failed - $err");
2941         die $err;
2942     }
2943
2944     return $res;
2945 }
2946
2947 sub qemu_block_resize {
2948     my ($vmid, $deviceid, $storecfg, $volid, $size) = @_;
2949
2950     my $running = check_running($vmid);
2951
2952     return if !PVE::Storage::volume_resize($storecfg, $volid, $size, $running);
2953
2954     return if !$running;
2955
2956     vm_mon_cmd($vmid, "block_resize", device => $deviceid, size => int($size));
2957
2958 }
2959
2960 sub qemu_volume_snapshot {
2961     my ($vmid, $deviceid, $storecfg, $volid, $snap) = @_;
2962
2963     my $running = check_running($vmid);
2964
2965     return if !PVE::Storage::volume_snapshot($storecfg, $volid, $snap, $running);
2966
2967     return if !$running;
2968
2969     vm_mon_cmd($vmid, "snapshot-drive", device => $deviceid, name => $snap);
2970
2971 }
2972
2973 sub qemu_volume_snapshot_delete {
2974     my ($vmid, $deviceid, $storecfg, $volid, $snap) = @_;
2975
2976     my $running = check_running($vmid);
2977
2978     return if !PVE::Storage::volume_snapshot_delete($storecfg, $volid, $snap, $running);
2979
2980     return if !$running;
2981
2982     vm_mon_cmd($vmid, "delete-drive-snapshot", device => $deviceid, name => $snap);
2983 }
2984
2985 sub qga_freezefs {
2986     my ($vmid) = @_;
2987
2988     #need to impplement call to qemu-ga
2989 }
2990
2991 sub qga_unfreezefs {
2992     my ($vmid) = @_;
2993
2994     #need to impplement call to qemu-ga
2995 }
2996
2997 sub vm_start {
2998     my ($storecfg, $vmid, $statefile, $skiplock, $migratedfrom, $paused, $forcemachine) = @_;
2999
3000     lock_config($vmid, sub {
3001         my $conf = load_config($vmid, $migratedfrom);
3002
3003         die "you can't start a vm if it's a template\n" if is_template($conf);
3004
3005         check_lock($conf) if !$skiplock;
3006
3007         die "VM $vmid already running\n" if check_running($vmid, undef, $migratedfrom);
3008
3009         my $defaults = load_defaults();
3010
3011         # set environment variable useful inside network script
3012         $ENV{PVE_MIGRATED_FROM} = $migratedfrom if $migratedfrom;
3013
3014         my ($cmd, $vollist) = config_to_command($storecfg, $vmid, $conf, $defaults, $forcemachine);
3015
3016         my $migrate_port = 0;
3017
3018         if ($statefile) {
3019             if ($statefile eq 'tcp') {
3020                 $migrate_port = PVE::Tools::next_migrate_port();
3021                 my $migrate_uri = "tcp:localhost:${migrate_port}";
3022                 push @$cmd, '-incoming', $migrate_uri;
3023                 push @$cmd, '-S';
3024             } else {
3025                 push @$cmd, '-loadstate', $statefile;
3026             }
3027         } elsif ($paused) {
3028             push @$cmd, '-S';
3029         }
3030
3031         # host pci devices
3032         for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
3033           my $d = parse_hostpci($conf->{"hostpci$i"});
3034           next if !$d;
3035           my $info = pci_device_info("0000:$d->{pciid}");
3036           die "IOMMU not present\n" if !check_iommu_support();
3037           die "no pci device info for device '$d->{pciid}'\n" if !$info;
3038           die "can't unbind pci device '$d->{pciid}'\n" if !pci_dev_bind_to_stub($info);
3039           die "can't reset pci device '$d->{pciid}'\n" if !pci_dev_reset($info);
3040         }
3041
3042         PVE::Storage::activate_volumes($storecfg, $vollist);
3043
3044         eval  { run_command($cmd, timeout => $statefile ? undef : 30,
3045                     umask => 0077); };
3046         my $err = $@;
3047         die "start failed: $err" if $err;
3048
3049         print "migration listens on port $migrate_port\n" if $migrate_port;
3050
3051         if ($statefile && $statefile ne 'tcp')  {
3052             eval { vm_mon_cmd_nocheck($vmid, "cont"); };
3053             warn $@ if $@;
3054         }
3055
3056         if($migratedfrom) {
3057             my $capabilities = {};
3058             $capabilities->{capability} =  "xbzrle";
3059             $capabilities->{state} = JSON::true;
3060             eval { vm_mon_cmd_nocheck($vmid, "migrate-set-capabilities", capabilities => [$capabilities]); };
3061         }
3062         else{
3063
3064             if (!$statefile && (!defined($conf->{balloon}) || $conf->{balloon})) {
3065                 vm_mon_cmd_nocheck($vmid, "balloon", value => $conf->{balloon}*1024*1024)
3066                     if $conf->{balloon};
3067                 vm_mon_cmd_nocheck($vmid, 'qom-set',
3068                             path => "machine/peripheral/balloon0",
3069                             property => "guest-stats-polling-interval",
3070                             value => 2);
3071             }
3072         }
3073     });
3074 }
3075
3076 sub vm_mon_cmd {
3077     my ($vmid, $execute, %params) = @_;
3078
3079     my $cmd = { execute => $execute, arguments => \%params };
3080     vm_qmp_command($vmid, $cmd);
3081 }
3082
3083 sub vm_mon_cmd_nocheck {
3084     my ($vmid, $execute, %params) = @_;
3085
3086     my $cmd = { execute => $execute, arguments => \%params };
3087     vm_qmp_command($vmid, $cmd, 1);
3088 }
3089
3090 sub vm_qmp_command {
3091     my ($vmid, $cmd, $nocheck) = @_;
3092
3093     my $res;
3094
3095     my $timeout;
3096     if ($cmd->{arguments} && $cmd->{arguments}->{timeout}) {
3097         $timeout = $cmd->{arguments}->{timeout};
3098         delete $cmd->{arguments}->{timeout};
3099     }
3100
3101     eval {
3102         die "VM $vmid not running\n" if !check_running($vmid, $nocheck);
3103         my $sname = qmp_socket($vmid);
3104         if (-e $sname) {
3105             my $qmpclient = PVE::QMPClient->new();
3106
3107             $res = $qmpclient->cmd($vmid, $cmd, $timeout);
3108         } elsif (-e "${var_run_tmpdir}/$vmid.mon") {
3109             die "can't execute complex command on old monitor - stop/start your vm to fix the problem\n"
3110                 if scalar(%{$cmd->{arguments}});
3111             vm_monitor_command($vmid, $cmd->{execute}, $nocheck);
3112         } else {
3113             die "unable to open monitor socket\n";
3114         }
3115     };
3116     if (my $err = $@) {
3117         syslog("err", "VM $vmid qmp command failed - $err");
3118         die $err;
3119     }
3120
3121     return $res;
3122 }
3123
3124 sub vm_human_monitor_command {
3125     my ($vmid, $cmdline) = @_;
3126
3127     my $res;
3128
3129     my $cmd = {
3130         execute => 'human-monitor-command',
3131         arguments => { 'command-line' => $cmdline},
3132     };
3133
3134     return vm_qmp_command($vmid, $cmd);
3135 }
3136
3137 sub vm_commandline {
3138     my ($storecfg, $vmid) = @_;
3139
3140     my $conf = load_config($vmid);
3141
3142     my $defaults = load_defaults();
3143
3144     my $cmd = config_to_command($storecfg, $vmid, $conf, $defaults);
3145
3146     return join(' ', @$cmd);
3147 }
3148
3149 sub vm_reset {
3150     my ($vmid, $skiplock) = @_;
3151
3152     lock_config($vmid, sub {
3153
3154         my $conf = load_config($vmid);
3155
3156         check_lock($conf) if !$skiplock;
3157
3158         vm_mon_cmd($vmid, "system_reset");
3159     });
3160 }
3161
3162 sub get_vm_volumes {
3163     my ($conf) = @_;
3164
3165     my $vollist = [];
3166     foreach_volid($conf, sub {
3167         my ($volid, $is_cdrom) = @_;
3168
3169         return if $volid =~ m|^/|;
3170
3171         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
3172         return if !$sid;
3173
3174         push @$vollist, $volid;
3175     });
3176
3177     return $vollist;
3178 }
3179
3180 sub vm_stop_cleanup {
3181     my ($storecfg, $vmid, $conf, $keepActive) = @_;
3182
3183     eval {
3184         fairsched_rmnod($vmid); # try to destroy group
3185
3186         if (!$keepActive) {
3187             my $vollist = get_vm_volumes($conf);
3188             PVE::Storage::deactivate_volumes($storecfg, $vollist);
3189         }
3190
3191         foreach my $ext (qw(mon qmp pid vnc qga)) {
3192             unlink "/var/run/qemu-server/${vmid}.$ext";
3193         }
3194     };
3195     warn $@ if $@; # avoid errors - just warn
3196 }
3197
3198 # Note: use $nockeck to skip tests if VM configuration file exists.
3199 # We need that when migration VMs to other nodes (files already moved)
3200 # Note: we set $keepActive in vzdump stop mode - volumes need to stay active
3201 sub vm_stop {
3202     my ($storecfg, $vmid, $skiplock, $nocheck, $timeout, $shutdown, $force, $keepActive, $migratedfrom) = @_;
3203
3204     $force = 1 if !defined($force) && !$shutdown;
3205
3206     if ($migratedfrom){
3207         my $pid = check_running($vmid, $nocheck, $migratedfrom);
3208         kill 15, $pid if $pid;
3209         my $conf = load_config($vmid, $migratedfrom);
3210         vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive);
3211         return;
3212     }
3213
3214     lock_config($vmid, sub {
3215
3216         my $pid = check_running($vmid, $nocheck);
3217         return if !$pid;
3218
3219         my $conf;
3220         if (!$nocheck) {
3221             $conf = load_config($vmid);
3222             check_lock($conf) if !$skiplock;
3223             if (!defined($timeout) && $shutdown && $conf->{startup}) {
3224                 my $opts = parse_startup($conf->{startup});
3225                 $timeout = $opts->{down} if $opts->{down};
3226             }
3227         }
3228
3229         $timeout = 60 if !defined($timeout);
3230
3231         eval {
3232             if ($shutdown) {
3233                 $nocheck ? vm_mon_cmd_nocheck($vmid, "system_powerdown") : vm_mon_cmd($vmid, "system_powerdown");
3234
3235             } else {
3236                 $nocheck ? vm_mon_cmd_nocheck($vmid, "quit") : vm_mon_cmd($vmid, "quit");
3237             }
3238         };
3239         my $err = $@;
3240
3241         if (!$err) {
3242             my $count = 0;
3243             while (($count < $timeout) && check_running($vmid, $nocheck)) {
3244                 $count++;
3245                 sleep 1;
3246             }
3247
3248             if ($count >= $timeout) {
3249                 if ($force) {
3250                     warn "VM still running - terminating now with SIGTERM\n";
3251                     kill 15, $pid;
3252                 } else {
3253                     die "VM quit/powerdown failed - got timeout\n";
3254                 }
3255             } else {
3256                 vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive) if $conf;
3257                 return;
3258             }
3259         } else {
3260             if ($force) {
3261                 warn "VM quit/powerdown failed - terminating now with SIGTERM\n";
3262                 kill 15, $pid;
3263             } else {
3264                 die "VM quit/powerdown failed\n";
3265             }
3266         }
3267
3268         # wait again
3269         $timeout = 10;
3270
3271         my $count = 0;
3272         while (($count < $timeout) && check_running($vmid, $nocheck)) {
3273             $count++;
3274             sleep 1;
3275         }
3276
3277         if ($count >= $timeout) {
3278             warn "VM still running - terminating now with SIGKILL\n";
3279             kill 9, $pid;
3280             sleep 1;
3281         }
3282
3283         vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive) if $conf;
3284    });
3285 }
3286
3287 sub vm_suspend {
3288     my ($vmid, $skiplock) = @_;
3289
3290     lock_config($vmid, sub {
3291
3292         my $conf = load_config($vmid);
3293
3294         check_lock($conf) if !($skiplock || ($conf->{lock} && $conf->{lock} eq 'backup'));
3295
3296         vm_mon_cmd($vmid, "stop");
3297     });
3298 }
3299
3300 sub vm_resume {
3301     my ($vmid, $skiplock) = @_;
3302
3303     lock_config($vmid, sub {
3304
3305         my $conf = load_config($vmid);
3306
3307         check_lock($conf) if !($skiplock || ($conf->{lock} && $conf->{lock} eq 'backup'));
3308
3309         vm_mon_cmd($vmid, "cont");
3310     });
3311 }
3312
3313 sub vm_sendkey {
3314     my ($vmid, $skiplock, $key) = @_;
3315
3316     lock_config($vmid, sub {
3317
3318         my $conf = load_config($vmid);
3319
3320         # there is no qmp command, so we use the human monitor command
3321         vm_human_monitor_command($vmid, "sendkey $key");
3322     });
3323 }
3324
3325 sub vm_destroy {
3326     my ($storecfg, $vmid, $skiplock) = @_;
3327
3328     lock_config($vmid, sub {
3329
3330         my $conf = load_config($vmid);
3331
3332         check_lock($conf) if !$skiplock;
3333
3334         if (!check_running($vmid)) {
3335             fairsched_rmnod($vmid); # try to destroy group
3336             destroy_vm($storecfg, $vmid);
3337         } else {
3338             die "VM $vmid is running - destroy failed\n";
3339         }
3340     });
3341 }
3342
3343 # pci helpers
3344
3345 sub file_write {
3346     my ($filename, $buf) = @_;
3347
3348     my $fh = IO::File->new($filename, "w");
3349     return undef if !$fh;
3350
3351     my $res = print $fh $buf;
3352
3353     $fh->close();
3354
3355     return $res;
3356 }
3357
3358 sub pci_device_info {
3359     my ($name) = @_;
3360
3361     my $res;
3362
3363     return undef if $name !~ m/^([a-f0-9]{4}):([a-f0-9]{2}):([a-f0-9]{2})\.([a-f0-9])$/;
3364     my ($domain, $bus, $slot, $func) = ($1, $2, $3, $4);
3365
3366     my $irq = file_read_firstline("$pcisysfs/devices/$name/irq");
3367     return undef if !defined($irq) || $irq !~ m/^\d+$/;
3368
3369     my $vendor = file_read_firstline("$pcisysfs/devices/$name/vendor");
3370     return undef if !defined($vendor) || $vendor !~ s/^0x//;
3371
3372     my $product = file_read_firstline("$pcisysfs/devices/$name/device");
3373     return undef if !defined($product) || $product !~ s/^0x//;
3374
3375     $res = {
3376         name => $name,
3377         vendor => $vendor,
3378         product => $product,
3379         domain => $domain,
3380         bus => $bus,
3381         slot => $slot,
3382         func => $func,
3383         irq => $irq,
3384         has_fl_reset => -f "$pcisysfs/devices/$name/reset" || 0,
3385     };
3386
3387     return $res;
3388 }
3389
3390 sub pci_dev_reset {
3391     my ($dev) = @_;
3392
3393     my $name = $dev->{name};
3394
3395     my $fn = "$pcisysfs/devices/$name/reset";
3396
3397     return file_write($fn, "1");
3398 }
3399
3400 sub pci_dev_bind_to_stub {
3401     my ($dev) = @_;
3402
3403     my $name = $dev->{name};
3404
3405     my $testdir = "$pcisysfs/drivers/pci-stub/$name";
3406     return 1 if -d $testdir;
3407
3408     my $data = "$dev->{vendor} $dev->{product}";
3409     return undef if !file_write("$pcisysfs/drivers/pci-stub/new_id", $data);
3410
3411     my $fn = "$pcisysfs/devices/$name/driver/unbind";
3412     if (!file_write($fn, $name)) {
3413         return undef if -f $fn;
3414     }
3415
3416     $fn = "$pcisysfs/drivers/pci-stub/bind";
3417     if (! -d $testdir) {
3418         return undef if !file_write($fn, $name);
3419     }
3420
3421     return -d $testdir;
3422 }
3423
3424 sub print_pci_addr {
3425     my ($id, $bridges) = @_;
3426
3427     my $res = '';
3428     my $devices = {
3429         piix3 => { bus => 0, addr => 1 },
3430         #addr2 : first videocard
3431         balloon0 => { bus => 0, addr => 3 },
3432         watchdog => { bus => 0, addr => 4 },
3433         scsihw0 => { bus => 0, addr => 5 },
3434         scsihw1 => { bus => 0, addr => 6 },
3435         ahci0 => { bus => 0, addr => 7 },
3436         qga0 => { bus => 0, addr => 8 },
3437         spice => { bus => 0, addr => 9 },
3438         virtio0 => { bus => 0, addr => 10 },
3439         virtio1 => { bus => 0, addr => 11 },
3440         virtio2 => { bus => 0, addr => 12 },
3441         virtio3 => { bus => 0, addr => 13 },
3442         virtio4 => { bus => 0, addr => 14 },
3443         virtio5 => { bus => 0, addr => 15 },
3444         hostpci0 => { bus => 0, addr => 16 },
3445         hostpci1 => { bus => 0, addr => 17 },
3446         net0 => { bus => 0, addr => 18 },
3447         net1 => { bus => 0, addr => 19 },
3448         net2 => { bus => 0, addr => 20 },
3449         net3 => { bus => 0, addr => 21 },
3450         net4 => { bus => 0, addr => 22 },
3451         net5 => { bus => 0, addr => 23 },
3452         #addr29 : usb-host (pve-usb.cfg)
3453         'pci.1' => { bus => 0, addr => 30 },
3454         'pci.2' => { bus => 0, addr => 31 },
3455         'net6' => { bus => 1, addr => 1 },
3456         'net7' => { bus => 1, addr => 2 },
3457         'net8' => { bus => 1, addr => 3 },
3458         'net9' => { bus => 1, addr => 4 },
3459         'net10' => { bus => 1, addr => 5 },
3460         'net11' => { bus => 1, addr => 6 },
3461         'net12' => { bus => 1, addr => 7 },
3462         'net13' => { bus => 1, addr => 8 },
3463         'net14' => { bus => 1, addr => 9 },
3464         'net15' => { bus => 1, addr => 10 },
3465         'net16' => { bus => 1, addr => 11 },
3466         'net17' => { bus => 1, addr => 12 },
3467         'net18' => { bus => 1, addr => 13 },
3468         'net19' => { bus => 1, addr => 14 },
3469         'net20' => { bus => 1, addr => 15 },
3470         'net21' => { bus => 1, addr => 16 },
3471         'net22' => { bus => 1, addr => 17 },
3472         'net23' => { bus => 1, addr => 18 },
3473         'net24' => { bus => 1, addr => 19 },
3474         'net25' => { bus => 1, addr => 20 },
3475         'net26' => { bus => 1, addr => 21 },
3476         'net27' => { bus => 1, addr => 22 },
3477         'net28' => { bus => 1, addr => 23 },
3478         'net29' => { bus => 1, addr => 24 },
3479         'net30' => { bus => 1, addr => 25 },
3480         'net31' => { bus => 1, addr => 26 },
3481         'virtio6' => { bus => 2, addr => 1 },
3482         'virtio7' => { bus => 2, addr => 2 },
3483         'virtio8' => { bus => 2, addr => 3 },
3484         'virtio9' => { bus => 2, addr => 4 },
3485         'virtio10' => { bus => 2, addr => 5 },
3486         'virtio11' => { bus => 2, addr => 6 },
3487         'virtio12' => { bus => 2, addr => 7 },
3488         'virtio13' => { bus => 2, addr => 8 },
3489         'virtio14' => { bus => 2, addr => 9 },
3490         'virtio15' => { bus => 2, addr => 10 },
3491     };
3492
3493     if (defined($devices->{$id}->{bus}) && defined($devices->{$id}->{addr})) {
3494            my $addr = sprintf("0x%x", $devices->{$id}->{addr});
3495            my $bus = $devices->{$id}->{bus};
3496            $res = ",bus=pci.$bus,addr=$addr";
3497            $bridges->{$bus} = 1 if $bridges;
3498     }
3499     return $res;
3500
3501 }
3502
3503 # vzdump restore implementaion
3504
3505 sub tar_archive_read_firstfile {
3506     my $archive = shift;
3507
3508     die "ERROR: file '$archive' does not exist\n" if ! -f $archive;
3509
3510     # try to detect archive type first
3511     my $pid = open (TMP, "tar tf '$archive'|") ||
3512         die "unable to open file '$archive'\n";
3513     my $firstfile = <TMP>;
3514     kill 15, $pid;
3515     close TMP;
3516
3517     die "ERROR: archive contaions no data\n" if !$firstfile;
3518     chomp $firstfile;
3519
3520     return $firstfile;
3521 }
3522
3523 sub tar_restore_cleanup {
3524     my ($storecfg, $statfile) = @_;
3525
3526     print STDERR "starting cleanup\n";
3527
3528     if (my $fd = IO::File->new($statfile, "r")) {
3529         while (defined(my $line = <$fd>)) {
3530             if ($line =~ m/vzdump:([^\s:]*):(\S+)$/) {
3531                 my $volid = $2;
3532                 eval {
3533                     if ($volid =~ m|^/|) {
3534                         unlink $volid || die 'unlink failed\n';
3535                     } else {
3536                         PVE::Storage::vdisk_free($storecfg, $volid);
3537                     }
3538                     print STDERR "temporary volume '$volid' sucessfuly removed\n";
3539                 };
3540                 print STDERR "unable to cleanup '$volid' - $@" if $@;
3541             } else {
3542                 print STDERR "unable to parse line in statfile - $line";
3543             }
3544         }
3545         $fd->close();
3546     }
3547 }
3548
3549 sub restore_archive {
3550     my ($archive, $vmid, $user, $opts) = @_;
3551
3552     my $format = $opts->{format};
3553     my $comp;
3554
3555     if ($archive =~ m/\.tgz$/ || $archive =~ m/\.tar\.gz$/) {
3556         $format = 'tar' if !$format;
3557         $comp = 'gzip';
3558     } elsif ($archive =~ m/\.tar$/) {
3559         $format = 'tar' if !$format;
3560     } elsif ($archive =~ m/.tar.lzo$/) {
3561         $format = 'tar' if !$format;
3562         $comp = 'lzop';
3563     } elsif ($archive =~ m/\.vma$/) {
3564         $format = 'vma' if !$format;
3565     } elsif ($archive =~ m/\.vma\.gz$/) {
3566         $format = 'vma' if !$format;
3567         $comp = 'gzip';
3568     } elsif ($archive =~ m/\.vma\.lzo$/) {
3569         $format = 'vma' if !$format;
3570         $comp = 'lzop';
3571     } else {
3572         $format = 'vma' if !$format; # default
3573     }
3574
3575     # try to detect archive format
3576     if ($format eq 'tar') {
3577         return restore_tar_archive($archive, $vmid, $user, $opts);
3578     } else {
3579         return restore_vma_archive($archive, $vmid, $user, $opts, $comp);
3580     }
3581 }
3582
3583 sub restore_update_config_line {
3584     my ($outfd, $cookie, $vmid, $map, $line, $unique) = @_;
3585
3586     return if $line =~ m/^\#qmdump\#/;
3587     return if $line =~ m/^\#vzdump\#/;
3588     return if $line =~ m/^lock:/;
3589     return if $line =~ m/^unused\d+:/;
3590     return if $line =~ m/^parent:/;
3591     return if $line =~ m/^template:/; # restored VM is never a template
3592
3593     if (($line =~ m/^(vlan(\d+)):\s*(\S+)\s*$/)) {
3594         # try to convert old 1.X settings
3595         my ($id, $ind, $ethcfg) = ($1, $2, $3);
3596         foreach my $devconfig (PVE::Tools::split_list($ethcfg)) {
3597             my ($model, $macaddr) = split(/\=/, $devconfig);
3598             $macaddr = PVE::Tools::random_ether_addr() if !$macaddr || $unique;
3599             my $net = {
3600                 model => $model,
3601                 bridge => "vmbr$ind",
3602                 macaddr => $macaddr,
3603             };
3604             my $netstr = print_net($net);
3605
3606             print $outfd "net$cookie->{netcount}: $netstr\n";
3607             $cookie->{netcount}++;
3608         }
3609     } elsif (($line =~ m/^(net\d+):\s*(\S+)\s*$/) && $unique) {
3610         my ($id, $netstr) = ($1, $2);
3611         my $net = parse_net($netstr);
3612         $net->{macaddr} = PVE::Tools::random_ether_addr() if $net->{macaddr};
3613         $netstr = print_net($net);
3614         print $outfd "$id: $netstr\n";
3615     } elsif ($line =~ m/^((ide|scsi|virtio|sata)\d+):\s*(\S+)\s*$/) {
3616         my $virtdev = $1;
3617         my $value = $3;
3618         if ($line =~ m/backup=no/) {
3619             print $outfd "#$line";
3620         } elsif ($virtdev && $map->{$virtdev}) {
3621             my $di = parse_drive($virtdev, $value);
3622             delete $di->{format}; # format can change on restore
3623             $di->{file} = $map->{$virtdev};
3624             $value = print_drive($vmid, $di);
3625             print $outfd "$virtdev: $value\n";
3626         } else {
3627             print $outfd $line;
3628         }
3629     } else {
3630         print $outfd $line;
3631     }
3632 }
3633
3634 sub scan_volids {
3635     my ($cfg, $vmid) = @_;
3636
3637     my $info = PVE::Storage::vdisk_list($cfg, undef, $vmid);
3638
3639     my $volid_hash = {};
3640     foreach my $storeid (keys %$info) {
3641         foreach my $item (@{$info->{$storeid}}) {
3642             next if !($item->{volid} && $item->{size});
3643             $item->{path} = PVE::Storage::path($cfg, $item->{volid});
3644             $volid_hash->{$item->{volid}} = $item;
3645         }
3646     }
3647
3648     return $volid_hash;
3649 }
3650
3651 sub get_used_paths {
3652     my ($vmid, $storecfg, $conf, $scan_snapshots, $skip_drive) = @_;
3653
3654     my $used_path = {};
3655
3656     my $scan_config = sub {
3657         my ($cref, $snapname) = @_;
3658
3659         foreach my $key (keys %$cref) {
3660             my $value = $cref->{$key};
3661             if (valid_drivename($key)) {
3662                 next if $skip_drive && $key eq $skip_drive;
3663                 my $drive = parse_drive($key, $value);
3664                 next if !$drive || !$drive->{file} || drive_is_cdrom($drive);
3665                 if ($drive->{file} =~ m!^/!) {
3666                     $used_path->{$drive->{file}}++; # = 1;
3667                 } else {
3668                     my ($storeid, $volname) = PVE::Storage::parse_volume_id($drive->{file}, 1);
3669                     next if !$storeid;
3670                     my $scfg = PVE::Storage::storage_config($storecfg, $storeid, 1);
3671                     next if !$scfg;
3672                     my $path = PVE::Storage::path($storecfg, $drive->{file}, $snapname);
3673                     $used_path->{$path}++; # = 1;
3674                 }
3675             }
3676         }
3677     };
3678
3679     &$scan_config($conf);
3680
3681     undef $skip_drive;
3682
3683     if ($scan_snapshots) {
3684         foreach my $snapname (keys %{$conf->{snapshots}}) {
3685             &$scan_config($conf->{snapshots}->{$snapname}, $snapname);
3686         }
3687     }
3688
3689     return $used_path;
3690 }
3691
3692 sub update_disksize {
3693     my ($vmid, $conf, $volid_hash) = @_;
3694
3695     my $changes;
3696
3697     my $used = {};
3698
3699     # Note: it is allowed to define multiple storages with same path (alias), so
3700     # we need to check both 'volid' and real 'path' (two different volid can point
3701     # to the same path).
3702
3703     my $usedpath = {};
3704
3705     # update size info
3706     foreach my $opt (keys %$conf) {
3707         if (valid_drivename($opt)) {
3708             my $drive = parse_drive($opt, $conf->{$opt});
3709             my $volid = $drive->{file};
3710             next if !$volid;
3711
3712             $used->{$volid} = 1;
3713             if ($volid_hash->{$volid} &&
3714                 (my $path = $volid_hash->{$volid}->{path})) {
3715                 $usedpath->{$path} = 1;
3716             }
3717
3718             next if drive_is_cdrom($drive);
3719             next if !$volid_hash->{$volid};
3720
3721             $drive->{size} = $volid_hash->{$volid}->{size};
3722             my $new = print_drive($vmid, $drive);
3723             if ($new ne $conf->{$opt}) {
3724                 $changes = 1;
3725                 $conf->{$opt} = $new;
3726             }
3727         }
3728     }
3729
3730     # remove 'unusedX' entry if volume is used
3731     foreach my $opt (keys %$conf) {
3732         next if $opt !~ m/^unused\d+$/;
3733         my $volid = $conf->{$opt};
3734         my $path = $volid_hash->{$volid}->{path} if $volid_hash->{$volid};
3735         if ($used->{$volid} || ($path && $usedpath->{$path})) {
3736             $changes = 1;
3737             delete $conf->{$opt};
3738         }
3739     }
3740
3741     foreach my $volid (sort keys %$volid_hash) {
3742         next if $volid =~ m/vm-$vmid-state-/;
3743         next if $used->{$volid};
3744         my $path = $volid_hash->{$volid}->{path};
3745         next if !$path; # just to be sure
3746         next if $usedpath->{$path};
3747         $changes = 1;
3748         add_unused_volume($conf, $volid);
3749         $usedpath->{$path} = 1; # avoid to add more than once (aliases)
3750     }
3751
3752     return $changes;
3753 }
3754
3755 sub rescan {
3756     my ($vmid, $nolock) = @_;
3757
3758     my $cfg = PVE::Cluster::cfs_read_file("storage.cfg");
3759
3760     my $volid_hash = scan_volids($cfg, $vmid);
3761
3762     my $updatefn =  sub {
3763         my ($vmid) = @_;
3764
3765         my $conf = load_config($vmid);
3766
3767         check_lock($conf);
3768
3769         my $vm_volids = {};
3770         foreach my $volid (keys %$volid_hash) {
3771             my $info = $volid_hash->{$volid};
3772             $vm_volids->{$volid} = $info if $info->{vmid} && $info->{vmid} == $vmid;
3773         }
3774
3775         my $changes = update_disksize($vmid, $conf, $vm_volids);
3776
3777         update_config_nolock($vmid, $conf, 1) if $changes;
3778     };
3779
3780     if (defined($vmid)) {
3781         if ($nolock) {
3782             &$updatefn($vmid);
3783         } else {
3784             lock_config($vmid, $updatefn, $vmid);
3785         }
3786     } else {
3787         my $vmlist = config_list();
3788         foreach my $vmid (keys %$vmlist) {
3789             if ($nolock) {
3790                 &$updatefn($vmid);
3791             } else {
3792                 lock_config($vmid, $updatefn, $vmid);
3793             }
3794         }
3795     }
3796 }
3797
3798 sub restore_vma_archive {
3799     my ($archive, $vmid, $user, $opts, $comp) = @_;
3800
3801     my $input = $archive eq '-' ? "<&STDIN" : undef;
3802     my $readfrom = $archive;
3803
3804     my $uncomp = '';
3805     if ($comp) {
3806         $readfrom = '-';
3807         my $qarchive = PVE::Tools::shellquote($archive);
3808         if ($comp eq 'gzip') {
3809             $uncomp = "zcat $qarchive|";
3810         } elsif ($comp eq 'lzop') {
3811             $uncomp = "lzop -d -c $qarchive|";
3812         } else {
3813             die "unknown compression method '$comp'\n";
3814         }
3815
3816     }
3817
3818     my $tmpdir = "/var/tmp/vzdumptmp$$";
3819     rmtree $tmpdir;
3820
3821     # disable interrupts (always do cleanups)
3822     local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = sub {
3823         warn "got interrupt - ignored\n";
3824     };
3825
3826     my $mapfifo = "/var/tmp/vzdumptmp$$.fifo";
3827     POSIX::mkfifo($mapfifo, 0600);
3828     my $fifofh;
3829
3830     my $openfifo = sub {
3831         open($fifofh, '>', $mapfifo) || die $!;
3832     };
3833
3834     my $cmd = "${uncomp}vma extract -v -r $mapfifo $readfrom $tmpdir";
3835
3836     my $oldtimeout;
3837     my $timeout = 5;
3838
3839     my $devinfo = {};
3840
3841     my $rpcenv = PVE::RPCEnvironment::get();
3842
3843     my $conffile = config_file($vmid);
3844     my $tmpfn = "$conffile.$$.tmp";
3845
3846     # Note: $oldconf is undef if VM does not exists
3847     my $oldconf = PVE::Cluster::cfs_read_file(cfs_config_path($vmid));
3848
3849     my $print_devmap = sub {
3850         my $virtdev_hash = {};
3851
3852         my $cfgfn = "$tmpdir/qemu-server.conf";
3853
3854         # we can read the config - that is already extracted
3855         my $fh = IO::File->new($cfgfn, "r") ||
3856             "unable to read qemu-server.conf - $!\n";
3857
3858         while (defined(my $line = <$fh>)) {
3859             if ($line =~ m/^\#qmdump\#map:(\S+):(\S+):(\S*):(\S*):$/) {
3860                 my ($virtdev, $devname, $storeid, $format) = ($1, $2, $3, $4);
3861                 die "archive does not contain data for drive '$virtdev'\n"
3862                     if !$devinfo->{$devname};
3863                 if (defined($opts->{storage})) {
3864                     $storeid = $opts->{storage} || 'local';
3865                 } elsif (!$storeid) {
3866                     $storeid = 'local';
3867                 }
3868                 $format = 'raw' if !$format;
3869                 $devinfo->{$devname}->{devname} = $devname;
3870                 $devinfo->{$devname}->{virtdev} = $virtdev;
3871                 $devinfo->{$devname}->{format} = $format;
3872                 $devinfo->{$devname}->{storeid} = $storeid;
3873
3874                 # check permission on storage
3875                 my $pool = $opts->{pool}; # todo: do we need that?
3876                 if ($user ne 'root@pam') {
3877                     $rpcenv->check($user, "/storage/$storeid", ['Datastore.AllocateSpace']);
3878                 }
3879
3880                 $virtdev_hash->{$virtdev} = $devinfo->{$devname};
3881             }
3882         }
3883
3884         foreach my $devname (keys %$devinfo) {
3885             die "found no device mapping information for device '$devname'\n"
3886                 if !$devinfo->{$devname}->{virtdev};
3887         }
3888
3889         my $cfg = cfs_read_file('storage.cfg');
3890
3891         # create empty/temp config
3892         if ($oldconf) {
3893             PVE::Tools::file_set_contents($conffile, "memory: 128\n");
3894             foreach_drive($oldconf, sub {
3895                 my ($ds, $drive) = @_;
3896
3897                 return if drive_is_cdrom($drive);
3898
3899                 my $volid = $drive->{file};
3900
3901                 return if !$volid || $volid =~ m|^/|;
3902
3903                 my ($path, $owner) = PVE::Storage::path($cfg, $volid);
3904                 return if !$path || !$owner || ($owner != $vmid);
3905
3906                 # Note: only delete disk we want to restore
3907                 # other volumes will become unused
3908                 if ($virtdev_hash->{$ds}) {
3909                     PVE::Storage::vdisk_free($cfg, $volid);
3910                 }
3911             });
3912         }
3913
3914         my $map = {};
3915         foreach my $virtdev (sort keys %$virtdev_hash) {
3916             my $d = $virtdev_hash->{$virtdev};
3917             my $alloc_size = int(($d->{size} + 1024 - 1)/1024);
3918             my $scfg = PVE::Storage::storage_config($cfg, $d->{storeid});
3919
3920             # test if requested format is supported
3921             my ($defFormat, $validFormats) = PVE::Storage::storage_default_format($cfg, $d->{storeid});
3922             my $supported = grep { $_ eq $d->{format} } @$validFormats;
3923             $d->{format} = $defFormat if !$supported;
3924
3925             my $volid = PVE::Storage::vdisk_alloc($cfg, $d->{storeid}, $vmid,
3926                                                   $d->{format}, undef, $alloc_size);
3927             print STDERR "new volume ID is '$volid'\n";
3928             $d->{volid} = $volid;
3929             my $path = PVE::Storage::path($cfg, $volid);
3930
3931             my $write_zeros = 1;
3932             # fixme: what other storages types initialize volumes with zero?
3933             if ($scfg->{type} eq 'dir' || $scfg->{type} eq 'nfs' ||
3934                 $scfg->{type} eq 'sheepdog' || $scfg->{type} eq 'rbd') {
3935                 $write_zeros = 0;
3936             }
3937
3938             print $fifofh "${write_zeros}:$d->{devname}=$path\n";
3939
3940             print "map '$d->{devname}' to '$path' (write zeros = ${write_zeros})\n";
3941             $map->{$virtdev} = $volid;
3942         }
3943
3944         $fh->seek(0, 0) || die "seek failed - $!\n";
3945
3946         my $outfd = new IO::File ($tmpfn, "w") ||
3947             die "unable to write config for VM $vmid\n";
3948
3949         my $cookie = { netcount => 0 };
3950         while (defined(my $line = <$fh>)) {
3951             restore_update_config_line($outfd, $cookie, $vmid, $map, $line, $opts->{unique});
3952         }
3953
3954         $fh->close();
3955         $outfd->close();
3956     };
3957
3958     eval {
3959         # enable interrupts
3960         local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = $SIG{PIPE} = sub {
3961             die "interrupted by signal\n";
3962         };
3963         local $SIG{ALRM} = sub { die "got timeout\n"; };
3964
3965         $oldtimeout = alarm($timeout);
3966
3967         my $parser = sub {
3968             my $line = shift;
3969
3970             print "$line\n";
3971
3972             if ($line =~ m/^DEV:\sdev_id=(\d+)\ssize:\s(\d+)\sdevname:\s(\S+)$/) {
3973                 my ($dev_id, $size, $devname) = ($1, $2, $3);
3974                 $devinfo->{$devname} = { size => $size, dev_id => $dev_id };
3975             } elsif ($line =~ m/^CTIME: /) {
3976                 &$print_devmap();
3977                 print $fifofh "done\n";
3978                 my $tmp = $oldtimeout || 0;
3979                 $oldtimeout = undef;
3980                 alarm($tmp);
3981                 close($fifofh);
3982             }
3983         };
3984
3985         print "restore vma archive: $cmd\n";
3986         run_command($cmd, input => $input, outfunc => $parser, afterfork => $openfifo);
3987     };
3988     my $err = $@;
3989
3990     alarm($oldtimeout) if $oldtimeout;
3991
3992     unlink $mapfifo;
3993
3994     if ($err) {
3995         rmtree $tmpdir;
3996         unlink $tmpfn;
3997
3998         my $cfg = cfs_read_file('storage.cfg');
3999         foreach my $devname (keys %$devinfo) {
4000             my $volid = $devinfo->{$devname}->{volid};
4001             next if !$volid;
4002             eval {
4003                 if ($volid =~ m|^/|) {
4004                     unlink $volid || die 'unlink failed\n';
4005                 } else {
4006                     PVE::Storage::vdisk_free($cfg, $volid);
4007                 }
4008                 print STDERR "temporary volume '$volid' sucessfuly removed\n";
4009             };
4010             print STDERR "unable to cleanup '$volid' - $@" if $@;
4011         }
4012         die $err;
4013     }
4014
4015     rmtree $tmpdir;
4016
4017     rename($tmpfn, $conffile) ||
4018         die "unable to commit configuration file '$conffile'\n";
4019
4020     PVE::Cluster::cfs_update(); # make sure we read new file
4021
4022     eval { rescan($vmid, 1); };
4023     warn $@ if $@;
4024 }
4025
4026 sub restore_tar_archive {
4027     my ($archive, $vmid, $user, $opts) = @_;
4028
4029     if ($archive ne '-') {
4030         my $firstfile = tar_archive_read_firstfile($archive);
4031         die "ERROR: file '$archive' dos not lock like a QemuServer vzdump backup\n"
4032             if $firstfile ne 'qemu-server.conf';
4033     }
4034
4035     my $storecfg = cfs_read_file('storage.cfg');
4036
4037     # destroy existing data - keep empty config
4038     my $vmcfgfn = PVE::QemuServer::config_file($vmid);
4039     destroy_vm($storecfg, $vmid, 1) if -f $vmcfgfn;
4040
4041     my $tocmd = "/usr/lib/qemu-server/qmextract";
4042
4043     $tocmd .= " --storage " . PVE::Tools::shellquote($opts->{storage}) if $opts->{storage};
4044     $tocmd .= " --pool " . PVE::Tools::shellquote($opts->{pool}) if $opts->{pool};
4045     $tocmd .= ' --prealloc' if $opts->{prealloc};
4046     $tocmd .= ' --info' if $opts->{info};
4047
4048     # tar option "xf" does not autodetect compression when read from STDIN,
4049     # so we pipe to zcat
4050     my $cmd = "zcat -f|tar xf " . PVE::Tools::shellquote($archive) . " " .
4051         PVE::Tools::shellquote("--to-command=$tocmd");
4052
4053     my $tmpdir = "/var/tmp/vzdumptmp$$";
4054     mkpath $tmpdir;
4055
4056     local $ENV{VZDUMP_TMPDIR} = $tmpdir;
4057     local $ENV{VZDUMP_VMID} = $vmid;
4058     local $ENV{VZDUMP_USER} = $user;
4059
4060     my $conffile = config_file($vmid);
4061     my $tmpfn = "$conffile.$$.tmp";
4062
4063     # disable interrupts (always do cleanups)
4064     local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = sub {
4065         print STDERR "got interrupt - ignored\n";
4066     };
4067
4068     eval {
4069         # enable interrupts
4070         local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = $SIG{PIPE} = sub {
4071             die "interrupted by signal\n";
4072         };
4073
4074         if ($archive eq '-') {
4075             print "extracting archive from STDIN\n";
4076             run_command($cmd, input => "<&STDIN");
4077         } else {
4078             print "extracting archive '$archive'\n";
4079             run_command($cmd);
4080         }
4081
4082         return if $opts->{info};
4083
4084         # read new mapping
4085         my $map = {};
4086         my $statfile = "$tmpdir/qmrestore.stat";
4087         if (my $fd = IO::File->new($statfile, "r")) {
4088             while (defined (my $line = <$fd>)) {
4089                 if ($line =~ m/vzdump:([^\s:]*):(\S+)$/) {
4090                     $map->{$1} = $2 if $1;
4091                 } else {
4092                     print STDERR "unable to parse line in statfile - $line\n";
4093                 }
4094             }
4095             $fd->close();
4096         }
4097
4098         my $confsrc = "$tmpdir/qemu-server.conf";
4099
4100         my $srcfd = new IO::File($confsrc, "r") ||
4101             die "unable to open file '$confsrc'\n";
4102
4103         my $outfd = new IO::File ($tmpfn, "w") ||
4104             die "unable to write config for VM $vmid\n";
4105
4106         my $cookie = { netcount => 0 };
4107         while (defined (my $line = <$srcfd>)) {
4108             restore_update_config_line($outfd, $cookie, $vmid, $map, $line, $opts->{unique});
4109         }
4110
4111         $srcfd->close();
4112         $outfd->close();
4113     };
4114     my $err = $@;
4115
4116     if ($err) {
4117
4118         unlink $tmpfn;
4119
4120         tar_restore_cleanup($storecfg, "$tmpdir/qmrestore.stat") if !$opts->{info};
4121
4122         die $err;
4123     }
4124
4125     rmtree $tmpdir;
4126
4127     rename $tmpfn, $conffile ||
4128         die "unable to commit configuration file '$conffile'\n";
4129
4130     PVE::Cluster::cfs_update(); # make sure we read new file
4131
4132     eval { rescan($vmid, 1); };
4133     warn $@ if $@;
4134 };
4135
4136
4137 # Internal snapshots
4138
4139 # NOTE: Snapshot create/delete involves several non-atomic
4140 # action, and can take a long time.
4141 # So we try to avoid locking the file and use 'lock' variable
4142 # inside the config file instead.
4143
4144 my $snapshot_copy_config = sub {
4145     my ($source, $dest) = @_;
4146
4147     foreach my $k (keys %$source) {
4148         next if $k eq 'snapshots';
4149         next if $k eq 'snapstate';
4150         next if $k eq 'snaptime';
4151         next if $k eq 'vmstate';
4152         next if $k eq 'lock';
4153         next if $k eq 'digest';
4154         next if $k eq 'description';
4155         next if $k =~ m/^unused\d+$/;
4156
4157         $dest->{$k} = $source->{$k};
4158     }
4159 };
4160
4161 my $snapshot_apply_config = sub {
4162     my ($conf, $snap) = @_;
4163
4164     # copy snapshot list
4165     my $newconf = {
4166         snapshots => $conf->{snapshots},
4167     };
4168
4169     # keep description and list of unused disks
4170     foreach my $k (keys %$conf) {
4171         next if !($k =~ m/^unused\d+$/ || $k eq 'description');
4172         $newconf->{$k} = $conf->{$k};
4173     }
4174
4175     &$snapshot_copy_config($snap, $newconf);
4176
4177     return $newconf;
4178 };
4179
4180 sub foreach_writable_storage {
4181     my ($conf, $func) = @_;
4182
4183     my $sidhash = {};
4184
4185     foreach my $ds (keys %$conf) {
4186         next if !valid_drivename($ds);
4187
4188         my $drive = parse_drive($ds, $conf->{$ds});
4189         next if !$drive;
4190         next if drive_is_cdrom($drive);
4191
4192         my $volid = $drive->{file};
4193
4194         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
4195         $sidhash->{$sid} = $sid if $sid;
4196     }
4197
4198     foreach my $sid (sort keys %$sidhash) {
4199         &$func($sid);
4200     }
4201 }
4202
4203 my $alloc_vmstate_volid = sub {
4204     my ($storecfg, $vmid, $conf, $snapname) = @_;
4205
4206     # Note: we try to be smart when selecting a $target storage
4207
4208     my $target;
4209
4210     # search shared storage first
4211     foreach_writable_storage($conf, sub {
4212         my ($sid) = @_;
4213         my $scfg = PVE::Storage::storage_config($storecfg, $sid);
4214         return if !$scfg->{shared};
4215
4216         $target = $sid if !$target || $scfg->{path}; # prefer file based storage
4217     });
4218
4219     if (!$target) {
4220         # now search local storage
4221         foreach_writable_storage($conf, sub {
4222             my ($sid) = @_;
4223             my $scfg = PVE::Storage::storage_config($storecfg, $sid);
4224             return if $scfg->{shared};
4225
4226             $target = $sid if !$target || $scfg->{path}; # prefer file based storage;
4227         });
4228     }
4229
4230     $target = 'local' if !$target;
4231
4232     my $driver_state_size = 500; # assume 32MB is enough to safe all driver state;
4233     # we abort live save after $conf->{memory}, so we need at max twice that space
4234     my $size = $conf->{memory}*2 + $driver_state_size;
4235
4236     my $name = "vm-$vmid-state-$snapname";
4237     my $scfg = PVE::Storage::storage_config($storecfg, $target);
4238     $name .= ".raw" if $scfg->{path}; # add filename extension for file base storage
4239     my $volid = PVE::Storage::vdisk_alloc($storecfg, $target, $vmid, 'raw', $name, $size*1024);
4240
4241     return $volid;
4242 };
4243
4244 my $snapshot_prepare = sub {
4245     my ($vmid, $snapname, $save_vmstate, $comment) = @_;
4246
4247     my $snap;
4248
4249     my $updatefn =  sub {
4250
4251         my $conf = load_config($vmid);
4252
4253         die "you can't take a snapshot if it's a template\n"
4254             if is_template($conf);
4255
4256         check_lock($conf);
4257
4258         $conf->{lock} = 'snapshot';
4259
4260         die "snapshot name '$snapname' already used\n"
4261             if defined($conf->{snapshots}->{$snapname});
4262
4263         my $storecfg = PVE::Storage::config();
4264         die "snapshot feature is not available" if !has_feature('snapshot', $conf, $storecfg);
4265
4266         $snap = $conf->{snapshots}->{$snapname} = {};
4267
4268         if ($save_vmstate && check_running($vmid)) {
4269             $snap->{vmstate} = &$alloc_vmstate_volid($storecfg, $vmid, $conf, $snapname);
4270         }
4271
4272         &$snapshot_copy_config($conf, $snap);
4273
4274         $snap->{snapstate} = "prepare";
4275         $snap->{snaptime} = time();
4276         $snap->{description} = $comment if $comment;
4277
4278         # always overwrite machine if we save vmstate. This makes sure we
4279         # can restore it later using correct machine type
4280         $snap->{machine} = get_current_qemu_machine($vmid) if $snap->{vmstate};
4281
4282         update_config_nolock($vmid, $conf, 1);
4283     };
4284
4285     lock_config($vmid, $updatefn);
4286
4287     return $snap;
4288 };
4289
4290 my $snapshot_commit = sub {
4291     my ($vmid, $snapname) = @_;
4292
4293     my $updatefn = sub {
4294
4295         my $conf = load_config($vmid);
4296
4297         die "missing snapshot lock\n"
4298             if !($conf->{lock} && $conf->{lock} eq 'snapshot');
4299
4300         my $snap = $conf->{snapshots}->{$snapname};
4301
4302         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4303
4304         die "wrong snapshot state\n"
4305             if !($snap->{snapstate} && $snap->{snapstate} eq "prepare");
4306
4307         delete $snap->{snapstate};
4308         delete $conf->{lock};
4309
4310         my $newconf = &$snapshot_apply_config($conf, $snap);
4311
4312         $newconf->{parent} = $snapname;
4313
4314         update_config_nolock($vmid, $newconf, 1);
4315     };
4316
4317     lock_config($vmid, $updatefn);
4318 };
4319
4320 sub snapshot_rollback {
4321     my ($vmid, $snapname) = @_;
4322
4323     my $snap;
4324
4325     my $prepare = 1;
4326
4327     my $storecfg = PVE::Storage::config();
4328
4329     my $updatefn = sub {
4330
4331         my $conf = load_config($vmid);
4332
4333         die "you can't rollback if vm is a template\n" if is_template($conf);
4334
4335         $snap = $conf->{snapshots}->{$snapname};
4336
4337         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4338
4339         die "unable to rollback to incomplete snapshot (snapstate = $snap->{snapstate})\n"
4340             if $snap->{snapstate};
4341
4342         if ($prepare) {
4343             check_lock($conf);
4344             vm_stop($storecfg, $vmid, undef, undef, 5, undef, undef);
4345         }
4346
4347         die "unable to rollback vm $vmid: vm is running\n"
4348             if check_running($vmid);
4349
4350         if ($prepare) {
4351             $conf->{lock} = 'rollback';
4352         } else {
4353             die "got wrong lock\n" if !($conf->{lock} && $conf->{lock} eq 'rollback');
4354             delete $conf->{lock};
4355         }
4356
4357         my $forcemachine;
4358
4359         if (!$prepare) {
4360             my $has_machine_config = defined($conf->{machine});
4361
4362             # copy snapshot config to current config
4363             $conf = &$snapshot_apply_config($conf, $snap);
4364             $conf->{parent} = $snapname;
4365
4366             # Note: old code did not store 'machine', so we try to be smart
4367             # and guess the snapshot was generated with kvm 1.4 (pc-i440fx-1.4).
4368             $forcemachine = $conf->{machine} || 'pc-i440fx-1.4';
4369             # we remove the 'machine' configuration if not explicitly specified
4370             # in the original config.
4371             delete $conf->{machine} if $snap->{vmstate} && !$has_machine_config;
4372         }
4373
4374         update_config_nolock($vmid, $conf, 1);
4375
4376         if (!$prepare && $snap->{vmstate}) {
4377             my $statefile = PVE::Storage::path($storecfg, $snap->{vmstate});
4378             vm_start($storecfg, $vmid, $statefile, undef, undef, undef, $forcemachine);
4379         }
4380     };
4381
4382     lock_config($vmid, $updatefn);
4383
4384     foreach_drive($snap, sub {
4385         my ($ds, $drive) = @_;
4386
4387         return if drive_is_cdrom($drive);
4388
4389         my $volid = $drive->{file};
4390         my $device = "drive-$ds";
4391
4392         PVE::Storage::volume_snapshot_rollback($storecfg, $volid, $snapname);
4393     });
4394
4395     $prepare = 0;
4396     lock_config($vmid, $updatefn);
4397 }
4398
4399 my $savevm_wait = sub {
4400     my ($vmid) = @_;
4401
4402     for(;;) {
4403         my $stat = vm_mon_cmd_nocheck($vmid, "query-savevm");
4404         if (!$stat->{status}) {
4405             die "savevm not active\n";
4406         } elsif ($stat->{status} eq 'active') {
4407             sleep(1);
4408             next;
4409         } elsif ($stat->{status} eq 'completed') {
4410             last;
4411         } else {
4412             die "query-savevm returned status '$stat->{status}'\n";
4413         }
4414     }
4415 };
4416
4417 sub snapshot_create {
4418     my ($vmid, $snapname, $save_vmstate, $freezefs, $comment) = @_;
4419
4420     my $snap = &$snapshot_prepare($vmid, $snapname, $save_vmstate, $comment);
4421
4422     $freezefs = $save_vmstate = 0 if !$snap->{vmstate}; # vm is not running
4423
4424     my $drivehash = {};
4425
4426     my $running = check_running($vmid);
4427
4428     eval {
4429         # create internal snapshots of all drives
4430
4431         my $storecfg = PVE::Storage::config();
4432
4433         if ($running) {
4434             if ($snap->{vmstate}) {
4435                 my $path = PVE::Storage::path($storecfg, $snap->{vmstate});
4436                 vm_mon_cmd($vmid, "savevm-start", statefile => $path);
4437                 &$savevm_wait($vmid);
4438             } else {
4439                 vm_mon_cmd($vmid, "savevm-start");
4440             }
4441         };
4442
4443         qga_freezefs($vmid) if $running && $freezefs;
4444
4445         foreach_drive($snap, sub {
4446             my ($ds, $drive) = @_;
4447
4448             return if drive_is_cdrom($drive);
4449
4450             my $volid = $drive->{file};
4451             my $device = "drive-$ds";
4452
4453             qemu_volume_snapshot($vmid, $device, $storecfg, $volid, $snapname);
4454             $drivehash->{$ds} = 1;
4455        });
4456     };
4457     my $err = $@;
4458
4459     eval { qga_unfreezefs($vmid) if $running && $freezefs; };
4460     warn $@ if $@;
4461
4462     eval { vm_mon_cmd($vmid, "savevm-end") if $running; };
4463     warn $@ if $@;
4464
4465     if ($err) {
4466         warn "snapshot create failed: starting cleanup\n";
4467         eval { snapshot_delete($vmid, $snapname, 0, $drivehash); };
4468         warn $@ if $@;
4469         die $err;
4470     }
4471
4472     &$snapshot_commit($vmid, $snapname);
4473 }
4474
4475 # Note: $drivehash is only set when called from snapshot_create.
4476 sub snapshot_delete {
4477     my ($vmid, $snapname, $force, $drivehash) = @_;
4478
4479     my $prepare = 1;
4480
4481     my $snap;
4482     my $unused = [];
4483
4484     my $unlink_parent = sub {
4485         my ($confref, $new_parent) = @_;
4486
4487         if ($confref->{parent} && $confref->{parent} eq $snapname) {
4488             if ($new_parent) {
4489                 $confref->{parent} = $new_parent;
4490             } else {
4491                 delete $confref->{parent};
4492             }
4493         }
4494     };
4495
4496     my $updatefn =  sub {
4497         my ($remove_drive) = @_;
4498
4499         my $conf = load_config($vmid);
4500
4501         if (!$drivehash) {
4502             check_lock($conf);
4503             die "you can't delete a snapshot if vm is a template\n"
4504                 if is_template($conf);
4505         }
4506
4507         $snap = $conf->{snapshots}->{$snapname};
4508
4509         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4510
4511         # remove parent refs
4512         &$unlink_parent($conf, $snap->{parent});
4513         foreach my $sn (keys %{$conf->{snapshots}}) {
4514             next if $sn eq $snapname;
4515             &$unlink_parent($conf->{snapshots}->{$sn}, $snap->{parent});
4516         }
4517
4518         if ($remove_drive) {
4519             if ($remove_drive eq 'vmstate') {
4520                 delete $snap->{$remove_drive};
4521             } else {
4522                 my $drive = parse_drive($remove_drive, $snap->{$remove_drive});
4523                 my $volid = $drive->{file};
4524                 delete $snap->{$remove_drive};
4525                 add_unused_volume($conf, $volid);
4526             }
4527         }
4528
4529         if ($prepare) {
4530             $snap->{snapstate} = 'delete';
4531         } else {
4532             delete $conf->{snapshots}->{$snapname};
4533             delete $conf->{lock} if $drivehash;
4534             foreach my $volid (@$unused) {
4535                 add_unused_volume($conf, $volid);
4536             }
4537         }
4538
4539         update_config_nolock($vmid, $conf, 1);
4540     };
4541
4542     lock_config($vmid, $updatefn);
4543
4544     # now remove vmstate file
4545
4546     my $storecfg = PVE::Storage::config();
4547
4548     if ($snap->{vmstate}) {
4549         eval {  PVE::Storage::vdisk_free($storecfg, $snap->{vmstate}); };
4550         if (my $err = $@) {
4551             die $err if !$force;
4552             warn $err;
4553         }
4554         # save changes (remove vmstate from snapshot)
4555         lock_config($vmid, $updatefn, 'vmstate') if !$force;
4556     };
4557
4558     # now remove all internal snapshots
4559     foreach_drive($snap, sub {
4560         my ($ds, $drive) = @_;
4561
4562         return if drive_is_cdrom($drive);
4563
4564         my $volid = $drive->{file};
4565         my $device = "drive-$ds";
4566
4567         if (!$drivehash || $drivehash->{$ds}) {
4568             eval { qemu_volume_snapshot_delete($vmid, $device, $storecfg, $volid, $snapname); };
4569             if (my $err = $@) {
4570                 die $err if !$force;
4571                 warn $err;
4572             }
4573         }
4574
4575         # save changes (remove drive fron snapshot)
4576         lock_config($vmid, $updatefn, $ds) if !$force;
4577         push @$unused, $volid;
4578     });
4579
4580     # now cleanup config
4581     $prepare = 0;
4582     lock_config($vmid, $updatefn);
4583 }
4584
4585 sub has_feature {
4586     my ($feature, $conf, $storecfg, $snapname, $running) = @_;
4587
4588     my $err;
4589     foreach_drive($conf, sub {
4590         my ($ds, $drive) = @_;
4591
4592         return if drive_is_cdrom($drive);
4593         my $volid = $drive->{file};
4594         $err = 1 if !PVE::Storage::volume_has_feature($storecfg, $feature, $volid, $snapname, $running);
4595     });
4596
4597     return $err ? 0 : 1;
4598 }
4599
4600 sub template_create {
4601     my ($vmid, $conf, $disk) = @_;
4602
4603     my $storecfg = PVE::Storage::config();
4604
4605     foreach_drive($conf, sub {
4606         my ($ds, $drive) = @_;
4607
4608         return if drive_is_cdrom($drive);
4609         return if $disk && $ds ne $disk;
4610
4611         my $volid = $drive->{file};
4612         return if !PVE::Storage::volume_has_feature($storecfg, 'template', $volid);
4613
4614         my $voliddst = PVE::Storage::vdisk_create_base($storecfg, $volid);
4615         $drive->{file} = $voliddst;
4616         $conf->{$ds} = print_drive($vmid, $drive);
4617         update_config_nolock($vmid, $conf, 1);
4618     });
4619 }
4620
4621 sub is_template {
4622     my ($conf) = @_;
4623
4624     return 1 if defined $conf->{template} && $conf->{template} == 1;
4625 }
4626
4627 sub qemu_img_convert {
4628     my ($src_volid, $dst_volid, $size, $snapname) = @_;
4629
4630     my $storecfg = PVE::Storage::config();
4631     my ($src_storeid, $src_volname) = PVE::Storage::parse_volume_id($src_volid, 1);
4632     my ($dst_storeid, $dst_volname) = PVE::Storage::parse_volume_id($dst_volid, 1);
4633
4634     if ($src_storeid && $dst_storeid) {
4635         my $src_scfg = PVE::Storage::storage_config($storecfg, $src_storeid);
4636         my $dst_scfg = PVE::Storage::storage_config($storecfg, $dst_storeid);
4637
4638         my $src_format = qemu_img_format($src_scfg, $src_volname);
4639         my $dst_format = qemu_img_format($dst_scfg, $dst_volname);
4640
4641         my $src_path = PVE::Storage::path($storecfg, $src_volid, $snapname);
4642         my $dst_path = PVE::Storage::path($storecfg, $dst_volid);
4643
4644         my $cmd = [];
4645         push @$cmd, '/usr/bin/qemu-img', 'convert', '-t', 'writeback', '-p', '-C';
4646         push @$cmd, '-s', $snapname if($snapname && $src_format eq "qcow2");
4647         push @$cmd, '-f', $src_format, '-O', $dst_format, $src_path, $dst_path;
4648
4649         my $parser = sub {
4650             my $line = shift;
4651             if($line =~ m/\((\S+)\/100\%\)/){
4652                 my $percent = $1;
4653                 my $transferred = int($size * $percent / 100);
4654                 my $remaining = $size - $transferred;
4655
4656                 print "transferred: $transferred bytes remaining: $remaining bytes total: $size bytes progression: $percent %\n";
4657             }
4658
4659         };
4660
4661         eval  { run_command($cmd, timeout => undef, outfunc => $parser); };
4662         my $err = $@;
4663         die "copy failed: $err" if $err;
4664     }
4665 }
4666
4667 sub qemu_img_format {
4668     my ($scfg, $volname) = @_;
4669
4670     if ($scfg->{path} && $volname =~ m/\.(raw|qcow2|qed|vmdk)$/) {
4671         return $1;
4672     } elsif ($scfg->{type} eq 'iscsi') {
4673         return "host_device";
4674     } else {
4675         return "raw";
4676     }
4677 }
4678
4679 sub qemu_drive_mirror {
4680     my ($vmid, $drive, $dst_volid, $vmiddst, $maxwait) = @_;
4681
4682     my $count = 1;
4683     my $old_len = 0;
4684     my $frozen = undef;
4685
4686     my $storecfg = PVE::Storage::config();
4687     my ($dst_storeid, $dst_volname) = PVE::Storage::parse_volume_id($dst_volid, 1);
4688
4689     if ($dst_storeid) {
4690         my $dst_scfg = PVE::Storage::storage_config($storecfg, $dst_storeid);
4691
4692         my $format;
4693         if ($dst_volname =~ m/\.(raw|qcow2)$/){
4694             $format = $1;
4695         }
4696
4697         my $dst_path = PVE::Storage::path($storecfg, $dst_volid);
4698
4699         if ($format) {
4700             #fixme : sometime drive-mirror timeout, but works fine after.
4701             # (I have see the problem with big volume > 200GB), so we need to eval
4702             eval { vm_mon_cmd($vmid, "drive-mirror", timeout => 10, device => "drive-$drive", mode => "existing",
4703                               sync => "full", target => $dst_path, format => $format); };
4704         } else {
4705             eval { vm_mon_cmd($vmid, "drive-mirror", timeout => 10, device => "drive-$drive", mode => "existing",
4706                               sync => "full", target => $dst_path); };
4707         }
4708
4709         eval {
4710             while (1) {
4711                 my $stats = vm_mon_cmd($vmid, "query-block-jobs");
4712                 my $stat = @$stats[0];
4713                 die "mirroring job seem to have die. Maybe do you have bad sectors?" if !$stat;
4714                 die "error job is not mirroring" if $stat->{type} ne "mirror";
4715
4716                 my $transferred = $stat->{offset};
4717                 my $total = $stat->{len};
4718                 my $remaining = $total - $transferred;
4719                 my $percent = sprintf "%.2f", ($transferred * 100 / $total);
4720
4721                 print "transferred: $transferred bytes remaining: $remaining bytes total: $total bytes progression: $percent %\n";
4722
4723                 last if ($stat->{len} == $stat->{offset});
4724                 if ($old_len == $stat->{offset}) {
4725                     if ($maxwait && $count > $maxwait) {
4726                     # if writes to disk occurs the disk needs to be freezed
4727                     # to be able to complete the migration
4728                         vm_suspend($vmid,1);
4729                         $count = 0;
4730                         $frozen = 1;
4731                     } else {
4732                         $count++ unless $frozen;
4733                     }
4734                 } elsif ($frozen) {
4735                     vm_resume($vmid,1);
4736                     $count = 0;
4737                 }
4738                 $old_len = $stat->{offset};
4739                 sleep 1;
4740             }
4741
4742             if ($vmiddst == $vmid) {
4743                 # switch the disk if source and destination are on the same guest
4744                 vm_mon_cmd($vmid, "block-job-complete", device => "drive-$drive");
4745             }
4746         };
4747         if (my $err = $@) {
4748             eval { vm_mon_cmd($vmid, "block-job-cancel", device => "drive-$drive"); };
4749             die "mirroring error: $err";
4750         }
4751
4752         if ($vmiddst != $vmid) {
4753             # if we clone a disk for a new target vm, we don't switch the disk
4754             vm_mon_cmd($vmid, "block-job-cancel", device => "drive-$drive");
4755         }
4756     }
4757 }
4758
4759 sub clone_disk {
4760     my ($storecfg, $vmid, $running, $drivename, $drive, $snapname,
4761         $newvmid, $storage, $format, $full, $newvollist) = @_;
4762
4763     my $newvolid;
4764
4765     if (!$full) {
4766         print "create linked clone of drive $drivename ($drive->{file})\n";
4767         $newvolid = PVE::Storage::vdisk_clone($storecfg,  $drive->{file}, $newvmid);
4768         push @$newvollist, $newvolid;
4769     } else {
4770         my ($storeid, $volname) = PVE::Storage::parse_volume_id($drive->{file});
4771         $storeid = $storage if $storage;
4772
4773         my ($defFormat, $validFormats) = PVE::Storage::storage_default_format($storecfg, $storeid);
4774         if (!$format) {
4775             $format = $drive->{format} || $defFormat;
4776         }
4777
4778         # test if requested format is supported - else use default
4779         my $supported = grep { $_ eq $format } @$validFormats;
4780         $format = $defFormat if !$supported;
4781
4782         my ($size) = PVE::Storage::volume_size_info($storecfg, $drive->{file}, 3);
4783
4784         print "create full clone of drive $drivename ($drive->{file})\n";
4785         $newvolid = PVE::Storage::vdisk_alloc($storecfg, $storeid, $newvmid, $format, undef, ($size/1024));
4786         push @$newvollist, $newvolid;
4787
4788         if (!$running || $snapname) {
4789             qemu_img_convert($drive->{file}, $newvolid, $size, $snapname);
4790         } else {
4791             qemu_drive_mirror($vmid, $drivename, $newvolid, $newvmid);
4792         }
4793     }
4794
4795     my ($size) = PVE::Storage::volume_size_info($storecfg, $newvolid, 3);
4796
4797     my $disk = $drive;
4798     $disk->{format} = undef;
4799     $disk->{file} = $newvolid;
4800     $disk->{size} = $size;
4801
4802     return $disk;
4803 }
4804
4805 # this only works if VM is running
4806 sub get_current_qemu_machine {
4807     my ($vmid) = @_;
4808
4809     my $cmd = { execute => 'query-machines', arguments => {} };
4810     my $res = PVE::QemuServer::vm_qmp_command($vmid, $cmd);
4811
4812     my ($current, $default);
4813     foreach my $e (@$res) {
4814         $default = $e->{name} if $e->{'is-default'};
4815         $current = $e->{name} if $e->{'is-current'};
4816     }
4817
4818     # fallback to the default machine if current is not supported by qemu
4819     return $current || $default || 'pc';
4820 }
4821
4822 1;