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",
 339         enum => [qw(std cirrus vmware)],
 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) = @_;
2223
2224     my $cmd = [];
2225     my $globalFlags = [];
2226     my $machineFlags = [];
2227     my $rtcFlags = [];
2228     my $devices = [];
2229     my $pciaddr = '';
2230     my $bridges = {};
2231     my $kvmver = kvm_user_version();
2232     my $vernum = 0; # unknown
2233     if ($kvmver =~ m/^(\d+)\.(\d+)$/) {
2234         $vernum = $1*1000000+$2*1000;
2235     } elsif ($kvmver =~ m/^(\d+)\.(\d+)\.(\d+)$/) {
2236         $vernum = $1*1000000+$2*1000+$3;
2237     }
2238
2239     die "detected old qemu-kvm binary ($kvmver)\n" if $vernum < 15000;
2240
2241     my $have_ovz = -f '/proc/vz/vestat';
2242
2243     push @$cmd, '/usr/bin/kvm';
2244
2245     push @$cmd, '-id', $vmid;
2246
2247     my $use_virtio = 0;
2248
2249     my $qmpsocket = qmp_socket($vmid);
2250     push @$cmd, '-chardev', "socket,id=qmp,path=$qmpsocket,server,nowait";
2251     push @$cmd, '-mon', "chardev=qmp,mode=control";
2252
2253     my $socket = vnc_socket($vmid);
2254     push @$cmd,  '-vnc', "unix:$socket,x509,password";
2255
2256     push @$cmd, '-pidfile' , pidfile_name($vmid);
2257
2258     push @$cmd, '-daemonize';
2259
2260     $pciaddr = print_pci_addr("piix3", $bridges);
2261     push @$devices, '-device', "piix3-usb-uhci,id=uhci$pciaddr.0x2";
2262
2263     my $use_usb2 = 0;
2264     for (my $i = 0; $i < $MAX_USB_DEVICES; $i++)  {
2265         next if !$conf->{"usb$i"};
2266         $use_usb2 = 1;
2267     }
2268     # include usb device config
2269     push @$devices, '-readconfig', '/usr/share/qemu-server/pve-usb.cfg' if $use_usb2;
2270
2271     # enable absolute mouse coordinates (needed by vnc)
2272     my $tablet = defined($conf->{tablet}) ? $conf->{tablet} : $defaults->{tablet};
2273     push @$devices, '-device', 'usb-tablet,id=tablet,bus=uhci.0,port=1' if $tablet;
2274
2275     # host pci devices
2276     for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
2277           my $d = parse_hostpci($conf->{"hostpci$i"});
2278           next if !$d;
2279           $pciaddr = print_pci_addr("hostpci$i", $bridges);
2280           push @$devices, '-device', "pci-assign,host=$d->{pciid},id=hostpci$i$pciaddr";
2281     }
2282
2283     # usb devices
2284     for (my $i = 0; $i < $MAX_USB_DEVICES; $i++)  {
2285         my $d = parse_usb_device($conf->{"usb$i"});
2286         next if !$d;
2287         if ($d->{vendorid} && $d->{productid}) {
2288             push @$devices, '-device', "usb-host,vendorid=0x$d->{vendorid},productid=0x$d->{productid}";
2289         } elsif (defined($d->{hostbus}) && defined($d->{hostport})) {
2290             push @$devices, '-device', "usb-host,hostbus=$d->{hostbus},hostport=$d->{hostport}";
2291         }
2292     }
2293
2294     # serial devices
2295     for (my $i = 0; $i < $MAX_SERIAL_PORTS; $i++)  {
2296         if (my $path = $conf->{"serial$i"}) {
2297             die "no such serial device\n" if ! -c $path;
2298             push @$devices, '-chardev', "tty,id=serial$i,path=$path";
2299             push @$devices, '-device', "isa-serial,chardev=serial$i";
2300         }
2301     }
2302
2303     # parallel devices
2304     for (my $i = 0; $i < $MAX_PARALLEL_PORTS; $i++)  {
2305         if (my $path = $conf->{"parallel$i"}) {
2306             die "no such parallel device\n" if ! -c $path;
2307             push @$devices, '-chardev', "parport,id=parallel$i,path=$path";
2308             push @$devices, '-device', "isa-parallel,chardev=parallel$i";
2309         }
2310     }
2311
2312     my $vmname = $conf->{name} || "vm$vmid";
2313
2314     push @$cmd, '-name', $vmname;
2315
2316     my $sockets = 1;
2317     $sockets = $conf->{smp} if $conf->{smp}; # old style - no longer iused
2318     $sockets = $conf->{sockets} if  $conf->{sockets};
2319
2320     my $cores = $conf->{cores} || 1;
2321
2322     push @$cmd, '-smp', "sockets=$sockets,cores=$cores";
2323
2324     push @$cmd, '-cpu', $conf->{cpu} if $conf->{cpu};
2325
2326     push @$cmd, '-nodefaults';
2327
2328     my $bootorder = $conf->{boot} || $confdesc->{boot}->{default};
2329
2330     my $bootindex_hash = {};
2331     my $i = 1;
2332     foreach my $o (split(//, $bootorder)) {
2333         $bootindex_hash->{$o} = $i*100;
2334         $i++;
2335     }
2336
2337     push @$cmd, '-boot', "menu=on";
2338
2339     push @$cmd, '-no-acpi' if defined($conf->{acpi}) && $conf->{acpi} == 0;
2340
2341     push @$cmd, '-no-reboot' if  defined($conf->{reboot}) && $conf->{reboot} == 0;
2342
2343     my $vga = $conf->{vga};
2344     if (!$vga) {
2345         if ($conf->{ostype} && ($conf->{ostype} eq 'win8' || $conf->{ostype} eq 'win7' || $conf->{ostype} eq 'w2k8')) {
2346             $vga = 'std';
2347         } else {
2348             $vga = 'cirrus';
2349         }
2350     }
2351
2352     push @$cmd, '-vga', $vga if $vga; # for kvm 77 and later
2353
2354     # time drift fix
2355     my $tdf = defined($conf->{tdf}) ? $conf->{tdf} : $defaults->{tdf};
2356
2357     my $nokvm = defined($conf->{kvm}) && $conf->{kvm} == 0 ? 1 : 0;
2358     my $useLocaltime = $conf->{localtime};
2359
2360     if (my $ost = $conf->{ostype}) {
2361         # other, wxp, w2k, w2k3, w2k8, wvista, win7, win8, l24, l26
2362
2363         if ($ost =~ m/^w/) { # windows
2364             $useLocaltime = 1 if !defined($conf->{localtime});
2365
2366             # use time drift fix when acpi is enabled
2367             if (!(defined($conf->{acpi}) && $conf->{acpi} == 0)) {
2368                 $tdf = 1 if !defined($conf->{tdf});
2369             }
2370         }
2371
2372         if ($ost eq 'win7' || $ost eq 'win8' || $ost eq 'w2k8' ||
2373             $ost eq 'wvista') {
2374             push @$globalFlags, 'kvm-pit.lost_tick_policy=discard';
2375             push @$cmd, '-no-hpet';
2376         }
2377     }
2378
2379     push @$rtcFlags, 'driftfix=slew' if $tdf;
2380
2381     if ($nokvm) {
2382         push @$machineFlags, 'accel=tcg';
2383     } else {
2384         die "No accelerator found!\n" if !$cpuinfo->{hvm};
2385     }
2386
2387     if ($conf->{machine}) {
2388         push @$machineFlags, "type=$conf->{machine}";
2389     }
2390
2391     if ($conf->{startdate}) {
2392         push @$rtcFlags, "base=$conf->{startdate}";
2393     } elsif ($useLocaltime) {
2394         push @$rtcFlags, 'base=localtime';
2395     }
2396
2397     push @$cmd, '-S' if $conf->{freeze};
2398
2399     # set keyboard layout
2400     my $kb = $conf->{keyboard} || $defaults->{keyboard};
2401     push @$cmd, '-k', $kb if $kb;
2402
2403     # enable sound
2404     #my $soundhw = $conf->{soundhw} || $defaults->{soundhw};
2405     #push @$cmd, '-soundhw', 'es1370';
2406     #push @$cmd, '-soundhw', $soundhw if $soundhw;
2407
2408     if($conf->{agent}) {
2409         my $qgasocket = qga_socket($vmid);
2410         my $pciaddr = print_pci_addr("qga0", $bridges);
2411         push @$devices, '-chardev', "socket,path=$qgasocket,server,nowait,id=qga0";
2412         push @$devices, '-device', "virtio-serial,id=qga0$pciaddr";
2413         push @$devices, '-device', 'virtserialport,chardev=qga0,name=org.qemu.guest_agent.0';
2414     }
2415
2416     # enable balloon by default, unless explicitly disabled
2417     if (!defined($conf->{balloon}) || $conf->{balloon}) {
2418         $pciaddr = print_pci_addr("balloon0", $bridges);
2419         push @$devices, '-device', "virtio-balloon-pci,id=balloon0$pciaddr";
2420     }
2421
2422     if ($conf->{watchdog}) {
2423         my $wdopts = parse_watchdog($conf->{watchdog});
2424         $pciaddr = print_pci_addr("watchdog", $bridges);
2425         my $watchdog = $wdopts->{model} || 'i6300esb';
2426         push @$devices, '-device', "$watchdog$pciaddr";
2427         push @$devices, '-watchdog-action', $wdopts->{action} if $wdopts->{action};
2428     }
2429
2430     my $vollist = [];
2431     my $scsicontroller = {};
2432     my $ahcicontroller = {};
2433     my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : $defaults->{scsihw};
2434
2435     foreach_drive($conf, sub {
2436         my ($ds, $drive) = @_;
2437
2438         if (PVE::Storage::parse_volume_id($drive->{file}, 1)) {
2439             push @$vollist, $drive->{file};
2440         }
2441
2442         $use_virtio = 1 if $ds =~ m/^virtio/;
2443
2444         if (drive_is_cdrom ($drive)) {
2445             if ($bootindex_hash->{d}) {
2446                 $drive->{bootindex} = $bootindex_hash->{d};
2447                 $bootindex_hash->{d} += 1;
2448             }
2449         } else {
2450             if ($bootindex_hash->{c}) {
2451                 $drive->{bootindex} = $bootindex_hash->{c} if $conf->{bootdisk} && ($conf->{bootdisk} eq $ds);
2452                 $bootindex_hash->{c} += 1;
2453             }
2454         }
2455
2456         if ($drive->{interface} eq 'scsi') {
2457
2458             my $maxdev = ($scsihw ne 'lsi') ? 256 : 7;
2459             my $controller = int($drive->{index} / $maxdev);
2460             $pciaddr = print_pci_addr("scsihw$controller", $bridges);
2461             push @$devices, '-device', "$scsihw,id=scsihw$controller$pciaddr" if !$scsicontroller->{$controller};
2462             $scsicontroller->{$controller}=1;
2463         }
2464
2465         if ($drive->{interface} eq 'sata') {
2466            my $controller = int($drive->{index} / $MAX_SATA_DISKS);
2467            $pciaddr = print_pci_addr("ahci$controller", $bridges);
2468            push @$devices, '-device', "ahci,id=ahci$controller,multifunction=on$pciaddr" if !$ahcicontroller->{$controller};
2469            $ahcicontroller->{$controller}=1;
2470         }
2471
2472         push @$devices, '-drive',print_drive_full($storecfg, $vmid, $drive);
2473         push @$devices, '-device',print_drivedevice_full($storecfg, $conf, $vmid, $drive, $bridges);
2474     });
2475
2476     push @$cmd, '-m', $conf->{memory} || $defaults->{memory};
2477
2478     for (my $i = 0; $i < $MAX_NETS; $i++) {
2479          next if !$conf->{"net$i"};
2480          my $d = parse_net($conf->{"net$i"});
2481          next if !$d;
2482
2483          $use_virtio = 1 if $d->{model} eq 'virtio';
2484
2485          if ($bootindex_hash->{n}) {
2486             $d->{bootindex} = $bootindex_hash->{n};
2487             $bootindex_hash->{n} += 1;
2488          }
2489
2490          my $netdevfull = print_netdev_full($vmid,$conf,$d,"net$i");
2491          push @$devices, '-netdev', $netdevfull;
2492
2493          my $netdevicefull = print_netdevice_full($vmid,$conf,$d,"net$i",$bridges);
2494          push @$devices, '-device', $netdevicefull;
2495     }
2496
2497     #bridges
2498     while (my ($k, $v) = each %$bridges) {
2499         $pciaddr = print_pci_addr("pci.$k");
2500         unshift @$devices, '-device', "pci-bridge,id=pci.$k,chassis_nr=$k$pciaddr" if $k > 0;
2501     }
2502
2503
2504     # hack: virtio with fairsched is unreliable, so we do not use fairsched
2505     # when the VM uses virtio devices.
2506     if (!$use_virtio && $have_ovz) {
2507
2508         my $cpuunits = defined($conf->{cpuunits}) ?
2509             $conf->{cpuunits} : $defaults->{cpuunits};
2510
2511         push @$cmd, '-cpuunits', $cpuunits if $cpuunits;
2512
2513         # fixme: cpulimit is currently ignored
2514         #push @$cmd, '-cpulimit', $conf->{cpulimit} if $conf->{cpulimit};
2515     }
2516
2517     # add custom args
2518     if ($conf->{args}) {
2519         my $aa = PVE::Tools::split_args($conf->{args});
2520         push @$cmd, @$aa;
2521     }
2522
2523     push @$cmd, @$devices;
2524     push @$cmd, '-rtc', join(',', @$rtcFlags)
2525         if scalar(@$rtcFlags);
2526     push @$cmd, '-machine', join(',', @$machineFlags)
2527         if scalar(@$machineFlags);
2528     push @$cmd, '-global', join(',', @$globalFlags)
2529         if scalar(@$globalFlags);
2530
2531     return wantarray ? ($cmd, $vollist) : $cmd;
2532 }
2533
2534 sub vnc_socket {
2535     my ($vmid) = @_;
2536     return "${var_run_tmpdir}/$vmid.vnc";
2537 }
2538
2539 sub qmp_socket {
2540     my ($vmid) = @_;
2541     return "${var_run_tmpdir}/$vmid.qmp";
2542 }
2543
2544 sub qga_socket {
2545     my ($vmid) = @_;
2546     return "${var_run_tmpdir}/$vmid.qga";
2547 }
2548
2549 sub pidfile_name {
2550     my ($vmid) = @_;
2551     return "${var_run_tmpdir}/$vmid.pid";
2552 }
2553
2554 sub vm_devices_list {
2555     my ($vmid) = @_;
2556
2557     my $res = vm_mon_cmd($vmid, 'query-pci');
2558
2559     my $devices = {};
2560     foreach my $pcibus (@$res) {
2561         foreach my $device (@{$pcibus->{devices}}) {
2562             next if !$device->{'qdev_id'};
2563             $devices->{$device->{'qdev_id'}} = $device;
2564         }
2565     }
2566
2567     return $devices;
2568 }
2569
2570 sub vm_deviceplug {
2571     my ($storecfg, $conf, $vmid, $deviceid, $device) = @_;
2572
2573     return 1 if !check_running($vmid);
2574
2575     if ($deviceid eq 'tablet') {
2576         my $devicefull = "usb-tablet,id=tablet,bus=uhci.0,port=1";
2577         qemu_deviceadd($vmid, $devicefull);
2578         return 1;
2579     }
2580
2581     return 1 if !$conf->{hotplug};
2582
2583     my $devices_list = vm_devices_list($vmid);
2584     return 1 if defined($devices_list->{$deviceid});
2585
2586     qemu_bridgeadd($storecfg, $conf, $vmid, $deviceid); #add bridge if we need it for the device
2587
2588     if ($deviceid =~ m/^(virtio)(\d+)$/) {
2589         return undef if !qemu_driveadd($storecfg, $vmid, $device);
2590         my $devicefull = print_drivedevice_full($storecfg, $conf, $vmid, $device);
2591         qemu_deviceadd($vmid, $devicefull);
2592         if(!qemu_deviceaddverify($vmid, $deviceid)) {
2593            qemu_drivedel($vmid, $deviceid);
2594            return undef;
2595         }
2596     }
2597
2598     if ($deviceid =~ m/^(scsihw)(\d+)$/) {
2599         my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : "lsi";
2600         my $pciaddr = print_pci_addr($deviceid);
2601         my $devicefull = "$scsihw,id=$deviceid$pciaddr";
2602         qemu_deviceadd($vmid, $devicefull);
2603         return undef if(!qemu_deviceaddverify($vmid, $deviceid));
2604     }
2605
2606     if ($deviceid =~ m/^(scsi)(\d+)$/) {
2607         return 1 if ($conf->{scsihw} && $conf->{scsihw} ne 'lsi'); #virtio-scsi not yet support hotplug
2608         return undef if !qemu_findorcreatescsihw($storecfg,$conf, $vmid, $device);
2609         return undef if !qemu_driveadd($storecfg, $vmid, $device);
2610         my $devicefull = print_drivedevice_full($storecfg, $conf, $vmid, $device);
2611         if(!qemu_deviceadd($vmid, $devicefull)) {
2612            qemu_drivedel($vmid, $deviceid);
2613            return undef;
2614         }
2615     }
2616
2617     if ($deviceid =~ m/^(net)(\d+)$/) {
2618         return undef if !qemu_netdevadd($vmid, $conf, $device, $deviceid);
2619         my $netdevicefull = print_netdevice_full($vmid, $conf, $device, $deviceid);
2620         qemu_deviceadd($vmid, $netdevicefull);
2621         if(!qemu_deviceaddverify($vmid, $deviceid)) {
2622            qemu_netdevdel($vmid, $deviceid);
2623            return undef;
2624         }
2625     }
2626
2627     if ($deviceid =~ m/^(pci\.)(\d+)$/) {
2628         my $bridgeid = $2;
2629         my $pciaddr = print_pci_addr($deviceid);
2630         my $devicefull = "pci-bridge,id=pci.$bridgeid,chassis_nr=$bridgeid$pciaddr";
2631         qemu_deviceadd($vmid, $devicefull);
2632         return undef if !qemu_deviceaddverify($vmid, $deviceid);
2633     }
2634
2635     return 1;
2636 }
2637
2638 sub vm_deviceunplug {
2639     my ($vmid, $conf, $deviceid) = @_;
2640
2641     return 1 if !check_running ($vmid);
2642
2643     if ($deviceid eq 'tablet') {
2644         qemu_devicedel($vmid, $deviceid);
2645         return 1;
2646     }
2647
2648     return 1 if !$conf->{hotplug};
2649
2650     my $devices_list = vm_devices_list($vmid);
2651     return 1 if !defined($devices_list->{$deviceid});
2652
2653     die "can't unplug bootdisk" if $conf->{bootdisk} && $conf->{bootdisk} eq $deviceid;
2654
2655     if ($deviceid =~ m/^(virtio)(\d+)$/) {
2656         qemu_devicedel($vmid, $deviceid);
2657         return undef if !qemu_devicedelverify($vmid, $deviceid);
2658         return undef if !qemu_drivedel($vmid, $deviceid);
2659     }
2660
2661     if ($deviceid =~ m/^(lsi)(\d+)$/) {
2662         return undef if !qemu_devicedel($vmid, $deviceid);
2663     }
2664
2665     if ($deviceid =~ m/^(scsi)(\d+)$/) {
2666         return undef if !qemu_devicedel($vmid, $deviceid);
2667         return undef if !qemu_drivedel($vmid, $deviceid);
2668     }
2669
2670     if ($deviceid =~ m/^(net)(\d+)$/) {
2671         qemu_devicedel($vmid, $deviceid);
2672         return undef if !qemu_devicedelverify($vmid, $deviceid);
2673         return undef if !qemu_netdevdel($vmid, $deviceid);
2674     }
2675
2676     return 1;
2677 }
2678
2679 sub qemu_deviceadd {
2680     my ($vmid, $devicefull) = @_;
2681
2682     $devicefull = "driver=".$devicefull;
2683     my %options =  split(/[=,]/, $devicefull);
2684
2685     vm_mon_cmd($vmid, "device_add" , %options);
2686     return 1;
2687 }
2688
2689 sub qemu_devicedel {
2690     my($vmid, $deviceid) = @_;
2691     my $ret = vm_mon_cmd($vmid, "device_del", id => $deviceid);
2692     return 1;
2693 }
2694
2695 sub qemu_driveadd {
2696     my($storecfg, $vmid, $device) = @_;
2697
2698     my $drive = print_drive_full($storecfg, $vmid, $device);
2699     my $ret = vm_human_monitor_command($vmid, "drive_add auto $drive");
2700     # If the command succeeds qemu prints: "OK"
2701     if ($ret !~ m/OK/s) {
2702         syslog("err", "adding drive failed: $ret");
2703         return undef;
2704     }
2705     return 1;
2706 }
2707
2708 sub qemu_drivedel {
2709     my($vmid, $deviceid) = @_;
2710
2711     my $ret = vm_human_monitor_command($vmid, "drive_del drive-$deviceid");
2712     $ret =~ s/^\s+//;
2713     if ($ret =~ m/Device \'.*?\' not found/s) {
2714         # NB: device not found errors mean the drive was auto-deleted and we ignore the error
2715     }
2716     elsif ($ret ne "") {
2717       syslog("err", "deleting drive $deviceid failed : $ret");
2718       return undef;
2719     }
2720     return 1;
2721 }
2722
2723 sub qemu_deviceaddverify {
2724     my ($vmid,$deviceid) = @_;
2725
2726     for (my $i = 0; $i <= 5; $i++) {
2727          my $devices_list = vm_devices_list($vmid);
2728          return 1 if defined($devices_list->{$deviceid});
2729          sleep 1;
2730     }
2731     syslog("err", "error on hotplug device $deviceid");
2732     return undef;
2733 }
2734
2735
2736 sub qemu_devicedelverify {
2737     my ($vmid,$deviceid) = @_;
2738
2739     #need to verify the device is correctly remove as device_del is async and empty return is not reliable
2740     for (my $i = 0; $i <= 5; $i++) {
2741          my $devices_list = vm_devices_list($vmid);
2742          return 1 if !defined($devices_list->{$deviceid});
2743          sleep 1;
2744     }
2745     syslog("err", "error on hot-unplugging device $deviceid");
2746     return undef;
2747 }
2748
2749 sub qemu_findorcreatescsihw {
2750     my ($storecfg, $conf, $vmid, $device) = @_;
2751
2752     my $maxdev = ($conf->{scsihw} && $conf->{scsihw} ne 'lsi') ? 256 : 7;
2753     my $controller = int($device->{index} / $maxdev);
2754     my $scsihwid="scsihw$controller";
2755     my $devices_list = vm_devices_list($vmid);
2756
2757     if(!defined($devices_list->{$scsihwid})) {
2758        return undef if !vm_deviceplug($storecfg, $conf, $vmid, $scsihwid);
2759     }
2760     return 1;
2761 }
2762
2763 sub qemu_bridgeadd {
2764     my ($storecfg, $conf, $vmid, $device) = @_;
2765
2766     my $bridges = {};
2767     my $bridgeid = undef;
2768     print_pci_addr($device, $bridges);
2769
2770     while (my ($k, $v) = each %$bridges) {
2771         $bridgeid = $k;
2772     }
2773     return if $bridgeid < 1;
2774     my $bridge = "pci.$bridgeid";
2775     my $devices_list = vm_devices_list($vmid);
2776
2777     if(!defined($devices_list->{$bridge})) {
2778         return undef if !vm_deviceplug($storecfg, $conf, $vmid, $bridge);
2779     }
2780     return 1;
2781 }
2782
2783 sub qemu_netdevadd {
2784     my ($vmid, $conf, $device, $deviceid) = @_;
2785
2786     my $netdev = print_netdev_full($vmid, $conf, $device, $deviceid);
2787     my %options =  split(/[=,]/, $netdev);
2788
2789     vm_mon_cmd($vmid, "netdev_add",  %options);
2790     return 1;
2791 }
2792
2793 sub qemu_netdevdel {
2794     my ($vmid, $deviceid) = @_;
2795
2796     vm_mon_cmd($vmid, "netdev_del", id => $deviceid);
2797     return 1;
2798 }
2799
2800 sub qemu_block_set_io_throttle {
2801     my ($vmid, $deviceid, $bps, $bps_rd, $bps_wr, $iops, $iops_rd, $iops_wr) = @_;
2802
2803     return if !check_running($vmid) ;
2804
2805     $bps = 0 if !$bps;
2806     $bps_rd = 0 if !$bps_rd;
2807     $bps_wr = 0 if !$bps_wr;
2808     $iops = 0 if !$iops;
2809     $iops_rd = 0 if !$iops_rd;
2810     $iops_wr = 0 if !$iops_wr;
2811
2812     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($iops_wr));
2813
2814 }
2815
2816 # old code, only used to shutdown old VM after update
2817 sub __read_avail {
2818     my ($fh, $timeout) = @_;
2819
2820     my $sel = new IO::Select;
2821     $sel->add($fh);
2822
2823     my $res = '';
2824     my $buf;
2825
2826     my @ready;
2827     while (scalar (@ready = $sel->can_read($timeout))) {
2828         my $count;
2829         if ($count = $fh->sysread($buf, 8192)) {
2830             if ($buf =~ /^(.*)\(qemu\) $/s) {
2831                 $res .= $1;
2832                 last;
2833             } else {
2834                 $res .= $buf;
2835             }
2836         } else {
2837             if (!defined($count)) {
2838                 die "$!\n";
2839             }
2840             last;
2841         }
2842     }
2843
2844     die "monitor read timeout\n" if !scalar(@ready);
2845
2846     return $res;
2847 }
2848
2849 # old code, only used to shutdown old VM after update
2850 sub vm_monitor_command {
2851     my ($vmid, $cmdstr, $nocheck) = @_;
2852
2853     my $res;
2854
2855     eval {
2856         die "VM $vmid not running\n" if !check_running($vmid, $nocheck);
2857
2858         my $sname = "${var_run_tmpdir}/$vmid.mon";
2859
2860         my $sock = IO::Socket::UNIX->new( Peer => $sname ) ||
2861             die "unable to connect to VM $vmid socket - $!\n";
2862
2863         my $timeout = 3;
2864
2865         # hack: migrate sometime blocks the monitor (when migrate_downtime
2866         # is set)
2867         if ($cmdstr =~ m/^(info\s+migrate|migrate\s)/) {
2868             $timeout = 60*60; # 1 hour
2869         }
2870
2871         # read banner;
2872         my $data = __read_avail($sock, $timeout);
2873
2874         if ($data !~ m/^QEMU\s+(\S+)\s+monitor\s/) {
2875             die "got unexpected qemu monitor banner\n";
2876         }
2877
2878         my $sel = new IO::Select;
2879         $sel->add($sock);
2880
2881         if (!scalar(my @ready = $sel->can_write($timeout))) {
2882             die "monitor write error - timeout";
2883         }
2884
2885         my $fullcmd = "$cmdstr\r";
2886
2887         # syslog('info', "VM $vmid monitor command: $cmdstr");
2888
2889         my $b;
2890         if (!($b = $sock->syswrite($fullcmd)) || ($b != length($fullcmd))) {
2891             die "monitor write error - $!";
2892         }
2893
2894         return if ($cmdstr eq 'q') || ($cmdstr eq 'quit');
2895
2896         $timeout = 20;
2897
2898         if ($cmdstr =~ m/^(info\s+migrate|migrate\s)/) {
2899             $timeout = 60*60; # 1 hour
2900         } elsif ($cmdstr =~ m/^(eject|change)/) {
2901             $timeout = 60; # note: cdrom mount command is slow
2902         }
2903         if ($res = __read_avail($sock, $timeout)) {
2904
2905             my @lines = split("\r?\n", $res);
2906
2907             shift @lines if $lines[0] !~ m/^unknown command/; # skip echo
2908
2909             $res = join("\n", @lines);
2910             $res .= "\n";
2911         }
2912     };
2913
2914     my $err = $@;
2915
2916     if ($err) {
2917         syslog("err", "VM $vmid monitor command failed - $err");
2918         die $err;
2919     }
2920
2921     return $res;
2922 }
2923
2924 sub qemu_block_resize {
2925     my ($vmid, $deviceid, $storecfg, $volid, $size) = @_;
2926
2927     my $running = check_running($vmid);
2928
2929     return if !PVE::Storage::volume_resize($storecfg, $volid, $size, $running);
2930
2931     return if !$running;
2932
2933     vm_mon_cmd($vmid, "block_resize", device => $deviceid, size => int($size));
2934
2935 }
2936
2937 sub qemu_volume_snapshot {
2938     my ($vmid, $deviceid, $storecfg, $volid, $snap) = @_;
2939
2940     my $running = check_running($vmid);
2941
2942     return if !PVE::Storage::volume_snapshot($storecfg, $volid, $snap, $running);
2943
2944     return if !$running;
2945
2946     vm_mon_cmd($vmid, "snapshot-drive", device => $deviceid, name => $snap);
2947
2948 }
2949
2950 sub qemu_volume_snapshot_delete {
2951     my ($vmid, $deviceid, $storecfg, $volid, $snap) = @_;
2952
2953     my $running = check_running($vmid);
2954
2955     return if !PVE::Storage::volume_snapshot_delete($storecfg, $volid, $snap, $running);
2956
2957     return if !$running;
2958
2959     vm_mon_cmd($vmid, "delete-drive-snapshot", device => $deviceid, name => $snap);
2960 }
2961
2962 sub qga_freezefs {
2963     my ($vmid) = @_;
2964
2965     #need to impplement call to qemu-ga
2966 }
2967
2968 sub qga_unfreezefs {
2969     my ($vmid) = @_;
2970
2971     #need to impplement call to qemu-ga
2972 }
2973
2974 sub vm_start {
2975     my ($storecfg, $vmid, $statefile, $skiplock, $migratedfrom, $paused) = @_;
2976
2977     lock_config($vmid, sub {
2978         my $conf = load_config($vmid, $migratedfrom);
2979
2980         die "you can't start a vm if it's a template\n" if is_template($conf);
2981
2982         check_lock($conf) if !$skiplock;
2983
2984         die "VM $vmid already running\n" if check_running($vmid, undef, $migratedfrom);
2985
2986         my $defaults = load_defaults();
2987
2988         # set environment variable useful inside network script
2989         $ENV{PVE_MIGRATED_FROM} = $migratedfrom if $migratedfrom;
2990
2991         my ($cmd, $vollist) = config_to_command($storecfg, $vmid, $conf, $defaults);
2992
2993         my $migrate_port = 0;
2994
2995         if ($statefile) {
2996             if ($statefile eq 'tcp') {
2997                 $migrate_port = PVE::Tools::next_migrate_port();
2998                 my $migrate_uri = "tcp:localhost:${migrate_port}";
2999                 push @$cmd, '-incoming', $migrate_uri;
3000                 push @$cmd, '-S';
3001             } else {
3002                 push @$cmd, '-loadstate', $statefile;
3003             }
3004         } elsif ($paused) {
3005             push @$cmd, '-S';
3006         }
3007
3008         # host pci devices
3009         for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
3010           my $d = parse_hostpci($conf->{"hostpci$i"});
3011           next if !$d;
3012           my $info = pci_device_info("0000:$d->{pciid}");
3013           die "IOMMU not present\n" if !check_iommu_support();
3014           die "no pci device info for device '$d->{pciid}'\n" if !$info;
3015           die "can't unbind pci device '$d->{pciid}'\n" if !pci_dev_bind_to_stub($info);
3016           die "can't reset pci device '$d->{pciid}'\n" if !pci_dev_reset($info);
3017         }
3018
3019         PVE::Storage::activate_volumes($storecfg, $vollist);
3020
3021         eval  { run_command($cmd, timeout => $statefile ? undef : 30,
3022                     umask => 0077); };
3023         my $err = $@;
3024         die "start failed: $err" if $err;
3025
3026         print "migration listens on port $migrate_port\n" if $migrate_port;
3027
3028         if ($statefile && $statefile ne 'tcp')  {
3029             eval { vm_mon_cmd_nocheck($vmid, "cont"); };
3030             warn $@ if $@;
3031         }
3032
3033         if($migratedfrom) {
3034             my $capabilities = {};
3035             $capabilities->{capability} =  "xbzrle";
3036             $capabilities->{state} = JSON::true;
3037             eval { vm_mon_cmd_nocheck($vmid, "migrate-set-capabilities", capabilities => [$capabilities]); };
3038         }
3039         else{
3040
3041             if (!$statefile && (!defined($conf->{balloon}) || $conf->{balloon})) {
3042                 vm_mon_cmd_nocheck($vmid, "balloon", value => $conf->{balloon}*1024*1024)
3043                     if $conf->{balloon};
3044                 vm_mon_cmd_nocheck($vmid, 'qom-set',
3045                             path => "machine/peripheral/balloon0",
3046                             property => "guest-stats-polling-interval",
3047                             value => 2);
3048             }
3049         }
3050     });
3051 }
3052
3053 sub vm_mon_cmd {
3054     my ($vmid, $execute, %params) = @_;
3055
3056     my $cmd = { execute => $execute, arguments => \%params };
3057     vm_qmp_command($vmid, $cmd);
3058 }
3059
3060 sub vm_mon_cmd_nocheck {
3061     my ($vmid, $execute, %params) = @_;
3062
3063     my $cmd = { execute => $execute, arguments => \%params };
3064     vm_qmp_command($vmid, $cmd, 1);
3065 }
3066
3067 sub vm_qmp_command {
3068     my ($vmid, $cmd, $nocheck) = @_;
3069
3070     my $res;
3071
3072     my $timeout;
3073     if ($cmd->{arguments} && $cmd->{arguments}->{timeout}) {
3074         $timeout = $cmd->{arguments}->{timeout};
3075         delete $cmd->{arguments}->{timeout};
3076     }
3077
3078     eval {
3079         die "VM $vmid not running\n" if !check_running($vmid, $nocheck);
3080         my $sname = qmp_socket($vmid);
3081         if (-e $sname) {
3082             my $qmpclient = PVE::QMPClient->new();
3083
3084             $res = $qmpclient->cmd($vmid, $cmd, $timeout);
3085         } elsif (-e "${var_run_tmpdir}/$vmid.mon") {
3086             die "can't execute complex command on old monitor - stop/start your vm to fix the problem\n"
3087                 if scalar(%{$cmd->{arguments}});
3088             vm_monitor_command($vmid, $cmd->{execute}, $nocheck);
3089         } else {
3090             die "unable to open monitor socket\n";
3091         }
3092     };
3093     if (my $err = $@) {
3094         syslog("err", "VM $vmid qmp command failed - $err");
3095         die $err;
3096     }
3097
3098     return $res;
3099 }
3100
3101 sub vm_human_monitor_command {
3102     my ($vmid, $cmdline) = @_;
3103
3104     my $res;
3105
3106     my $cmd = {
3107         execute => 'human-monitor-command',
3108         arguments => { 'command-line' => $cmdline},
3109     };
3110
3111     return vm_qmp_command($vmid, $cmd);
3112 }
3113
3114 sub vm_commandline {
3115     my ($storecfg, $vmid) = @_;
3116
3117     my $conf = load_config($vmid);
3118
3119     my $defaults = load_defaults();
3120
3121     my $cmd = config_to_command($storecfg, $vmid, $conf, $defaults);
3122
3123     return join(' ', @$cmd);
3124 }
3125
3126 sub vm_reset {
3127     my ($vmid, $skiplock) = @_;
3128
3129     lock_config($vmid, sub {
3130
3131         my $conf = load_config($vmid);
3132
3133         check_lock($conf) if !$skiplock;
3134
3135         vm_mon_cmd($vmid, "system_reset");
3136     });
3137 }
3138
3139 sub get_vm_volumes {
3140     my ($conf) = @_;
3141
3142     my $vollist = [];
3143     foreach_volid($conf, sub {
3144         my ($volid, $is_cdrom) = @_;
3145
3146         return if $volid =~ m|^/|;
3147
3148         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
3149         return if !$sid;
3150
3151         push @$vollist, $volid;
3152     });
3153
3154     return $vollist;
3155 }
3156
3157 sub vm_stop_cleanup {
3158     my ($storecfg, $vmid, $conf, $keepActive) = @_;
3159
3160     eval {
3161         fairsched_rmnod($vmid); # try to destroy group
3162
3163         if (!$keepActive) {
3164             my $vollist = get_vm_volumes($conf);
3165             PVE::Storage::deactivate_volumes($storecfg, $vollist);
3166         }
3167
3168         foreach my $ext (qw(mon qmp pid vnc qga)) {
3169             unlink "/var/run/qemu-server/${vmid}.$ext";
3170         }
3171     };
3172     warn $@ if $@; # avoid errors - just warn
3173 }
3174
3175 # Note: use $nockeck to skip tests if VM configuration file exists.
3176 # We need that when migration VMs to other nodes (files already moved)
3177 # Note: we set $keepActive in vzdump stop mode - volumes need to stay active
3178 sub vm_stop {
3179     my ($storecfg, $vmid, $skiplock, $nocheck, $timeout, $shutdown, $force, $keepActive, $migratedfrom) = @_;
3180
3181     $force = 1 if !defined($force) && !$shutdown;
3182
3183     if ($migratedfrom){
3184         my $pid = check_running($vmid, $nocheck, $migratedfrom);
3185         kill 15, $pid if $pid;
3186         my $conf = load_config($vmid, $migratedfrom);
3187         vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive);
3188         return;
3189     }
3190
3191     lock_config($vmid, sub {
3192
3193         my $pid = check_running($vmid, $nocheck);
3194         return if !$pid;
3195
3196         my $conf;
3197         if (!$nocheck) {
3198             $conf = load_config($vmid);
3199             check_lock($conf) if !$skiplock;
3200             if (!defined($timeout) && $shutdown && $conf->{startup}) {
3201                 my $opts = parse_startup($conf->{startup});
3202                 $timeout = $opts->{down} if $opts->{down};
3203             }
3204         }
3205
3206         $timeout = 60 if !defined($timeout);
3207
3208         eval {
3209             if ($shutdown) {
3210                 $nocheck ? vm_mon_cmd_nocheck($vmid, "system_powerdown") : vm_mon_cmd($vmid, "system_powerdown");
3211
3212             } else {
3213                 $nocheck ? vm_mon_cmd_nocheck($vmid, "quit") : vm_mon_cmd($vmid, "quit");
3214             }
3215         };
3216         my $err = $@;
3217
3218         if (!$err) {
3219             my $count = 0;
3220             while (($count < $timeout) && check_running($vmid, $nocheck)) {
3221                 $count++;
3222                 sleep 1;
3223             }
3224
3225             if ($count >= $timeout) {
3226                 if ($force) {
3227                     warn "VM still running - terminating now with SIGTERM\n";
3228                     kill 15, $pid;
3229                 } else {
3230                     die "VM quit/powerdown failed - got timeout\n";
3231                 }
3232             } else {
3233                 vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive) if $conf;
3234                 return;
3235             }
3236         } else {
3237             if ($force) {
3238                 warn "VM quit/powerdown failed - terminating now with SIGTERM\n";
3239                 kill 15, $pid;
3240             } else {
3241                 die "VM quit/powerdown failed\n";
3242             }
3243         }
3244
3245         # wait again
3246         $timeout = 10;
3247
3248         my $count = 0;
3249         while (($count < $timeout) && check_running($vmid, $nocheck)) {
3250             $count++;
3251             sleep 1;
3252         }
3253
3254         if ($count >= $timeout) {
3255             warn "VM still running - terminating now with SIGKILL\n";
3256             kill 9, $pid;
3257             sleep 1;
3258         }
3259
3260         vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive) if $conf;
3261    });
3262 }
3263
3264 sub vm_suspend {
3265     my ($vmid, $skiplock) = @_;
3266
3267     lock_config($vmid, sub {
3268
3269         my $conf = load_config($vmid);
3270
3271         check_lock($conf) if !($skiplock || ($conf->{lock} && $conf->{lock} eq 'backup'));
3272
3273         vm_mon_cmd($vmid, "stop");
3274     });
3275 }
3276
3277 sub vm_resume {
3278     my ($vmid, $skiplock) = @_;
3279
3280     lock_config($vmid, sub {
3281
3282         my $conf = load_config($vmid);
3283
3284         check_lock($conf) if !($skiplock || ($conf->{lock} && $conf->{lock} eq 'backup'));
3285
3286         vm_mon_cmd($vmid, "cont");
3287     });
3288 }
3289
3290 sub vm_sendkey {
3291     my ($vmid, $skiplock, $key) = @_;
3292
3293     lock_config($vmid, sub {
3294
3295         my $conf = load_config($vmid);
3296
3297         # there is no qmp command, so we use the human monitor command
3298         vm_human_monitor_command($vmid, "sendkey $key");
3299     });
3300 }
3301
3302 sub vm_destroy {
3303     my ($storecfg, $vmid, $skiplock) = @_;
3304
3305     lock_config($vmid, sub {
3306
3307         my $conf = load_config($vmid);
3308
3309         check_lock($conf) if !$skiplock;
3310
3311         if (!check_running($vmid)) {
3312             fairsched_rmnod($vmid); # try to destroy group
3313             destroy_vm($storecfg, $vmid);
3314         } else {
3315             die "VM $vmid is running - destroy failed\n";
3316         }
3317     });
3318 }
3319
3320 # pci helpers
3321
3322 sub file_write {
3323     my ($filename, $buf) = @_;
3324
3325     my $fh = IO::File->new($filename, "w");
3326     return undef if !$fh;
3327
3328     my $res = print $fh $buf;
3329
3330     $fh->close();
3331
3332     return $res;
3333 }
3334
3335 sub pci_device_info {
3336     my ($name) = @_;
3337
3338     my $res;
3339
3340     return undef if $name !~ m/^([a-f0-9]{4}):([a-f0-9]{2}):([a-f0-9]{2})\.([a-f0-9])$/;
3341     my ($domain, $bus, $slot, $func) = ($1, $2, $3, $4);
3342
3343     my $irq = file_read_firstline("$pcisysfs/devices/$name/irq");
3344     return undef if !defined($irq) || $irq !~ m/^\d+$/;
3345
3346     my $vendor = file_read_firstline("$pcisysfs/devices/$name/vendor");
3347     return undef if !defined($vendor) || $vendor !~ s/^0x//;
3348
3349     my $product = file_read_firstline("$pcisysfs/devices/$name/device");
3350     return undef if !defined($product) || $product !~ s/^0x//;
3351
3352     $res = {
3353         name => $name,
3354         vendor => $vendor,
3355         product => $product,
3356         domain => $domain,
3357         bus => $bus,
3358         slot => $slot,
3359         func => $func,
3360         irq => $irq,
3361         has_fl_reset => -f "$pcisysfs/devices/$name/reset" || 0,
3362     };
3363
3364     return $res;
3365 }
3366
3367 sub pci_dev_reset {
3368     my ($dev) = @_;
3369
3370     my $name = $dev->{name};
3371
3372     my $fn = "$pcisysfs/devices/$name/reset";
3373
3374     return file_write($fn, "1");
3375 }
3376
3377 sub pci_dev_bind_to_stub {
3378     my ($dev) = @_;
3379
3380     my $name = $dev->{name};
3381
3382     my $testdir = "$pcisysfs/drivers/pci-stub/$name";
3383     return 1 if -d $testdir;
3384
3385     my $data = "$dev->{vendor} $dev->{product}";
3386     return undef if !file_write("$pcisysfs/drivers/pci-stub/new_id", $data);
3387
3388     my $fn = "$pcisysfs/devices/$name/driver/unbind";
3389     if (!file_write($fn, $name)) {
3390         return undef if -f $fn;
3391     }
3392
3393     $fn = "$pcisysfs/drivers/pci-stub/bind";
3394     if (! -d $testdir) {
3395         return undef if !file_write($fn, $name);
3396     }
3397
3398     return -d $testdir;
3399 }
3400
3401 sub print_pci_addr {
3402     my ($id, $bridges) = @_;
3403
3404     my $res = '';
3405     my $devices = {
3406         piix3 => { bus => 0, addr => 1 },
3407         #addr2 : first videocard
3408         balloon0 => { bus => 0, addr => 3 },
3409         watchdog => { bus => 0, addr => 4 },
3410         scsihw0 => { bus => 0, addr => 5 },
3411         scsihw1 => { bus => 0, addr => 6 },
3412         ahci0 => { bus => 0, addr => 7 },
3413         qga0 => { bus => 0, addr => 8 },
3414         virtio0 => { bus => 0, addr => 10 },
3415         virtio1 => { bus => 0, addr => 11 },
3416         virtio2 => { bus => 0, addr => 12 },
3417         virtio3 => { bus => 0, addr => 13 },
3418         virtio4 => { bus => 0, addr => 14 },
3419         virtio5 => { bus => 0, addr => 15 },
3420         hostpci0 => { bus => 0, addr => 16 },
3421         hostpci1 => { bus => 0, addr => 17 },
3422         net0 => { bus => 0, addr => 18 },
3423         net1 => { bus => 0, addr => 19 },
3424         net2 => { bus => 0, addr => 20 },
3425         net3 => { bus => 0, addr => 21 },
3426         net4 => { bus => 0, addr => 22 },
3427         net5 => { bus => 0, addr => 23 },
3428         #addr29 : usb-host (pve-usb.cfg)
3429         'pci.1' => { bus => 0, addr => 30 },
3430         'pci.2' => { bus => 0, addr => 31 },
3431         'net6' => { bus => 1, addr => 1 },
3432         'net7' => { bus => 1, addr => 2 },
3433         'net8' => { bus => 1, addr => 3 },
3434         'net9' => { bus => 1, addr => 4 },
3435         'net10' => { bus => 1, addr => 5 },
3436         'net11' => { bus => 1, addr => 6 },
3437         'net12' => { bus => 1, addr => 7 },
3438         'net13' => { bus => 1, addr => 8 },
3439         'net14' => { bus => 1, addr => 9 },
3440         'net15' => { bus => 1, addr => 10 },
3441         'net16' => { bus => 1, addr => 11 },
3442         'net17' => { bus => 1, addr => 12 },
3443         'net18' => { bus => 1, addr => 13 },
3444         'net19' => { bus => 1, addr => 14 },
3445         'net20' => { bus => 1, addr => 15 },
3446         'net21' => { bus => 1, addr => 16 },
3447         'net22' => { bus => 1, addr => 17 },
3448         'net23' => { bus => 1, addr => 18 },
3449         'net24' => { bus => 1, addr => 19 },
3450         'net25' => { bus => 1, addr => 20 },
3451         'net26' => { bus => 1, addr => 21 },
3452         'net27' => { bus => 1, addr => 22 },
3453         'net28' => { bus => 1, addr => 23 },
3454         'net29' => { bus => 1, addr => 24 },
3455         'net30' => { bus => 1, addr => 25 },
3456         'net31' => { bus => 1, addr => 26 },
3457         'virtio6' => { bus => 2, addr => 1 },
3458         'virtio7' => { bus => 2, addr => 2 },
3459         'virtio8' => { bus => 2, addr => 3 },
3460         'virtio9' => { bus => 2, addr => 4 },
3461         'virtio10' => { bus => 2, addr => 5 },
3462         'virtio11' => { bus => 2, addr => 6 },
3463         'virtio12' => { bus => 2, addr => 7 },
3464         'virtio13' => { bus => 2, addr => 8 },
3465         'virtio14' => { bus => 2, addr => 9 },
3466         'virtio15' => { bus => 2, addr => 10 },
3467     };
3468
3469     if (defined($devices->{$id}->{bus}) && defined($devices->{$id}->{addr})) {
3470            my $addr = sprintf("0x%x", $devices->{$id}->{addr});
3471            my $bus = $devices->{$id}->{bus};
3472            $res = ",bus=pci.$bus,addr=$addr";
3473            $bridges->{$bus} = 1 if $bridges;
3474     }
3475     return $res;
3476
3477 }
3478
3479 # vzdump restore implementaion
3480
3481 sub tar_archive_read_firstfile {
3482     my $archive = shift;
3483
3484     die "ERROR: file '$archive' does not exist\n" if ! -f $archive;
3485
3486     # try to detect archive type first
3487     my $pid = open (TMP, "tar tf '$archive'|") ||
3488         die "unable to open file '$archive'\n";
3489     my $firstfile = <TMP>;
3490     kill 15, $pid;
3491     close TMP;
3492
3493     die "ERROR: archive contaions no data\n" if !$firstfile;
3494     chomp $firstfile;
3495
3496     return $firstfile;
3497 }
3498
3499 sub tar_restore_cleanup {
3500     my ($storecfg, $statfile) = @_;
3501
3502     print STDERR "starting cleanup\n";
3503
3504     if (my $fd = IO::File->new($statfile, "r")) {
3505         while (defined(my $line = <$fd>)) {
3506             if ($line =~ m/vzdump:([^\s:]*):(\S+)$/) {
3507                 my $volid = $2;
3508                 eval {
3509                     if ($volid =~ m|^/|) {
3510                         unlink $volid || die 'unlink failed\n';
3511                     } else {
3512                         PVE::Storage::vdisk_free($storecfg, $volid);
3513                     }
3514                     print STDERR "temporary volume '$volid' sucessfuly removed\n";
3515                 };
3516                 print STDERR "unable to cleanup '$volid' - $@" if $@;
3517             } else {
3518                 print STDERR "unable to parse line in statfile - $line";
3519             }
3520         }
3521         $fd->close();
3522     }
3523 }
3524
3525 sub restore_archive {
3526     my ($archive, $vmid, $user, $opts) = @_;
3527
3528     my $format = $opts->{format};
3529     my $comp;
3530
3531     if ($archive =~ m/\.tgz$/ || $archive =~ m/\.tar\.gz$/) {
3532         $format = 'tar' if !$format;
3533         $comp = 'gzip';
3534     } elsif ($archive =~ m/\.tar$/) {
3535         $format = 'tar' if !$format;
3536     } elsif ($archive =~ m/.tar.lzo$/) {
3537         $format = 'tar' if !$format;
3538         $comp = 'lzop';
3539     } elsif ($archive =~ m/\.vma$/) {
3540         $format = 'vma' if !$format;
3541     } elsif ($archive =~ m/\.vma\.gz$/) {
3542         $format = 'vma' if !$format;
3543         $comp = 'gzip';
3544     } elsif ($archive =~ m/\.vma\.lzo$/) {
3545         $format = 'vma' if !$format;
3546         $comp = 'lzop';
3547     } else {
3548         $format = 'vma' if !$format; # default
3549     }
3550
3551     # try to detect archive format
3552     if ($format eq 'tar') {
3553         return restore_tar_archive($archive, $vmid, $user, $opts);
3554     } else {
3555         return restore_vma_archive($archive, $vmid, $user, $opts, $comp);
3556     }
3557 }
3558
3559 sub restore_update_config_line {
3560     my ($outfd, $cookie, $vmid, $map, $line, $unique) = @_;
3561
3562     return if $line =~ m/^\#qmdump\#/;
3563     return if $line =~ m/^\#vzdump\#/;
3564     return if $line =~ m/^lock:/;
3565     return if $line =~ m/^unused\d+:/;
3566     return if $line =~ m/^parent:/;
3567     return if $line =~ m/^template:/; # restored VM is never a template
3568
3569     if (($line =~ m/^(vlan(\d+)):\s*(\S+)\s*$/)) {
3570         # try to convert old 1.X settings
3571         my ($id, $ind, $ethcfg) = ($1, $2, $3);
3572         foreach my $devconfig (PVE::Tools::split_list($ethcfg)) {
3573             my ($model, $macaddr) = split(/\=/, $devconfig);
3574             $macaddr = PVE::Tools::random_ether_addr() if !$macaddr || $unique;
3575             my $net = {
3576                 model => $model,
3577                 bridge => "vmbr$ind",
3578                 macaddr => $macaddr,
3579             };
3580             my $netstr = print_net($net);
3581
3582             print $outfd "net$cookie->{netcount}: $netstr\n";
3583             $cookie->{netcount}++;
3584         }
3585     } elsif (($line =~ m/^(net\d+):\s*(\S+)\s*$/) && $unique) {
3586         my ($id, $netstr) = ($1, $2);
3587         my $net = parse_net($netstr);
3588         $net->{macaddr} = PVE::Tools::random_ether_addr() if $net->{macaddr};
3589         $netstr = print_net($net);
3590         print $outfd "$id: $netstr\n";
3591     } elsif ($line =~ m/^((ide|scsi|virtio|sata)\d+):\s*(\S+)\s*$/) {
3592         my $virtdev = $1;
3593         my $value = $3;
3594         if ($line =~ m/backup=no/) {
3595             print $outfd "#$line";
3596         } elsif ($virtdev && $map->{$virtdev}) {
3597             my $di = parse_drive($virtdev, $value);
3598             delete $di->{format}; # format can change on restore
3599             $di->{file} = $map->{$virtdev};
3600             $value = print_drive($vmid, $di);
3601             print $outfd "$virtdev: $value\n";
3602         } else {
3603             print $outfd $line;
3604         }
3605     } else {
3606         print $outfd $line;
3607     }
3608 }
3609
3610 sub scan_volids {
3611     my ($cfg, $vmid) = @_;
3612
3613     my $info = PVE::Storage::vdisk_list($cfg, undef, $vmid);
3614
3615     my $volid_hash = {};
3616     foreach my $storeid (keys %$info) {
3617         foreach my $item (@{$info->{$storeid}}) {
3618             next if !($item->{volid} && $item->{size});
3619             $item->{path} = PVE::Storage::path($cfg, $item->{volid});
3620             $volid_hash->{$item->{volid}} = $item;
3621         }
3622     }
3623
3624     return $volid_hash;
3625 }
3626
3627 sub get_used_paths {
3628     my ($vmid, $storecfg, $conf, $scan_snapshots, $skip_drive) = @_;
3629
3630     my $used_path = {};
3631
3632     my $scan_config = sub {
3633         my ($cref, $snapname) = @_;
3634
3635         foreach my $key (keys %$cref) {
3636             my $value = $cref->{$key};
3637             if (valid_drivename($key)) {
3638                 next if $skip_drive && $key eq $skip_drive;
3639                 my $drive = parse_drive($key, $value);
3640                 next if !$drive || !$drive->{file} || drive_is_cdrom($drive);
3641                 if ($drive->{file} =~ m!^/!) {
3642                     $used_path->{$drive->{file}}++; # = 1;
3643                 } else {
3644                     my ($storeid, $volname) = PVE::Storage::parse_volume_id($drive->{file}, 1);
3645                     next if !$storeid;
3646                     my $scfg = PVE::Storage::storage_config($storecfg, $storeid, 1);
3647                     next if !$scfg;
3648                     my $path = PVE::Storage::path($storecfg, $drive->{file}, $snapname);
3649                     $used_path->{$path}++; # = 1;
3650                 }
3651             }
3652         }
3653     };
3654
3655     &$scan_config($conf);
3656
3657     undef $skip_drive;
3658
3659     if ($scan_snapshots) {
3660         foreach my $snapname (keys %{$conf->{snapshots}}) {
3661             &$scan_config($conf->{snapshots}->{$snapname}, $snapname);
3662         }
3663     }
3664
3665     return $used_path;
3666 }
3667
3668 sub update_disksize {
3669     my ($vmid, $conf, $volid_hash) = @_;
3670
3671     my $changes;
3672
3673     my $used = {};
3674
3675     # Note: it is allowed to define multiple storages with same path (alias), so
3676     # we need to check both 'volid' and real 'path' (two different volid can point
3677     # to the same path).
3678
3679     my $usedpath = {};
3680
3681     # update size info
3682     foreach my $opt (keys %$conf) {
3683         if (valid_drivename($opt)) {
3684             my $drive = parse_drive($opt, $conf->{$opt});
3685             my $volid = $drive->{file};
3686             next if !$volid;
3687
3688             $used->{$volid} = 1;
3689             if ($volid_hash->{$volid} &&
3690                 (my $path = $volid_hash->{$volid}->{path})) {
3691                 $usedpath->{$path} = 1;
3692             }
3693
3694             next if drive_is_cdrom($drive);
3695             next if !$volid_hash->{$volid};
3696
3697             $drive->{size} = $volid_hash->{$volid}->{size};
3698             my $new = print_drive($vmid, $drive);
3699             if ($new ne $conf->{$opt}) {
3700                 $changes = 1;
3701                 $conf->{$opt} = $new;
3702             }
3703         }
3704     }
3705
3706     # remove 'unusedX' entry if volume is used
3707     foreach my $opt (keys %$conf) {
3708         next if $opt !~ m/^unused\d+$/;
3709         my $volid = $conf->{$opt};
3710         my $path = $volid_hash->{$volid}->{path} if $volid_hash->{$volid};
3711         if ($used->{$volid} || ($path && $usedpath->{$path})) {
3712             $changes = 1;
3713             delete $conf->{$opt};
3714         }
3715     }
3716
3717     foreach my $volid (sort keys %$volid_hash) {
3718         next if $volid =~ m/vm-$vmid-state-/;
3719         next if $used->{$volid};
3720         my $path = $volid_hash->{$volid}->{path};
3721         next if !$path; # just to be sure
3722         next if $usedpath->{$path};
3723         $changes = 1;
3724         add_unused_volume($conf, $volid);
3725         $usedpath->{$path} = 1; # avoid to add more than once (aliases)
3726     }
3727
3728     return $changes;
3729 }
3730
3731 sub rescan {
3732     my ($vmid, $nolock) = @_;
3733
3734     my $cfg = PVE::Cluster::cfs_read_file("storage.cfg");
3735
3736     my $volid_hash = scan_volids($cfg, $vmid);
3737
3738     my $updatefn =  sub {
3739         my ($vmid) = @_;
3740
3741         my $conf = load_config($vmid);
3742
3743         check_lock($conf);
3744
3745         my $vm_volids = {};
3746         foreach my $volid (keys %$volid_hash) {
3747             my $info = $volid_hash->{$volid};
3748             $vm_volids->{$volid} = $info if $info->{vmid} && $info->{vmid} == $vmid;
3749         }
3750
3751         my $changes = update_disksize($vmid, $conf, $vm_volids);
3752
3753         update_config_nolock($vmid, $conf, 1) if $changes;
3754     };
3755
3756     if (defined($vmid)) {
3757         if ($nolock) {
3758             &$updatefn($vmid);
3759         } else {
3760             lock_config($vmid, $updatefn, $vmid);
3761         }
3762     } else {
3763         my $vmlist = config_list();
3764         foreach my $vmid (keys %$vmlist) {
3765             if ($nolock) {
3766                 &$updatefn($vmid);
3767             } else {
3768                 lock_config($vmid, $updatefn, $vmid);
3769             }
3770         }
3771     }
3772 }
3773
3774 sub restore_vma_archive {
3775     my ($archive, $vmid, $user, $opts, $comp) = @_;
3776
3777     my $input = $archive eq '-' ? "<&STDIN" : undef;
3778     my $readfrom = $archive;
3779
3780     my $uncomp = '';
3781     if ($comp) {
3782         $readfrom = '-';
3783         my $qarchive = PVE::Tools::shellquote($archive);
3784         if ($comp eq 'gzip') {
3785             $uncomp = "zcat $qarchive|";
3786         } elsif ($comp eq 'lzop') {
3787             $uncomp = "lzop -d -c $qarchive|";
3788         } else {
3789             die "unknown compression method '$comp'\n";
3790         }
3791
3792     }
3793
3794     my $tmpdir = "/var/tmp/vzdumptmp$$";
3795     rmtree $tmpdir;
3796
3797     # disable interrupts (always do cleanups)
3798     local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = sub {
3799         warn "got interrupt - ignored\n";
3800     };
3801
3802     my $mapfifo = "/var/tmp/vzdumptmp$$.fifo";
3803     POSIX::mkfifo($mapfifo, 0600);
3804     my $fifofh;
3805
3806     my $openfifo = sub {
3807         open($fifofh, '>', $mapfifo) || die $!;
3808     };
3809
3810     my $cmd = "${uncomp}vma extract -v -r $mapfifo $readfrom $tmpdir";
3811
3812     my $oldtimeout;
3813     my $timeout = 5;
3814
3815     my $devinfo = {};
3816
3817     my $rpcenv = PVE::RPCEnvironment::get();
3818
3819     my $conffile = config_file($vmid);
3820     my $tmpfn = "$conffile.$$.tmp";
3821
3822     # Note: $oldconf is undef if VM does not exists
3823     my $oldconf = PVE::Cluster::cfs_read_file(cfs_config_path($vmid));
3824
3825     my $print_devmap = sub {
3826         my $virtdev_hash = {};
3827
3828         my $cfgfn = "$tmpdir/qemu-server.conf";
3829
3830         # we can read the config - that is already extracted
3831         my $fh = IO::File->new($cfgfn, "r") ||
3832             "unable to read qemu-server.conf - $!\n";
3833
3834         while (defined(my $line = <$fh>)) {
3835             if ($line =~ m/^\#qmdump\#map:(\S+):(\S+):(\S*):(\S*):$/) {
3836                 my ($virtdev, $devname, $storeid, $format) = ($1, $2, $3, $4);
3837                 die "archive does not contain data for drive '$virtdev'\n"
3838                     if !$devinfo->{$devname};
3839                 if (defined($opts->{storage})) {
3840                     $storeid = $opts->{storage} || 'local';
3841                 } elsif (!$storeid) {
3842                     $storeid = 'local';
3843                 }
3844                 $format = 'raw' if !$format;
3845                 $devinfo->{$devname}->{devname} = $devname;
3846                 $devinfo->{$devname}->{virtdev} = $virtdev;
3847                 $devinfo->{$devname}->{format} = $format;
3848                 $devinfo->{$devname}->{storeid} = $storeid;
3849
3850                 # check permission on storage
3851                 my $pool = $opts->{pool}; # todo: do we need that?
3852                 if ($user ne 'root@pam') {
3853                     $rpcenv->check($user, "/storage/$storeid", ['Datastore.AllocateSpace']);
3854                 }
3855
3856                 $virtdev_hash->{$virtdev} = $devinfo->{$devname};
3857             }
3858         }
3859
3860         foreach my $devname (keys %$devinfo) {
3861             die "found no device mapping information for device '$devname'\n"
3862                 if !$devinfo->{$devname}->{virtdev};
3863         }
3864
3865         my $cfg = cfs_read_file('storage.cfg');
3866
3867         # create empty/temp config
3868         if ($oldconf) {
3869             PVE::Tools::file_set_contents($conffile, "memory: 128\n");
3870             foreach_drive($oldconf, sub {
3871                 my ($ds, $drive) = @_;
3872
3873                 return if drive_is_cdrom($drive);
3874
3875                 my $volid = $drive->{file};
3876
3877                 return if !$volid || $volid =~ m|^/|;
3878
3879                 my ($path, $owner) = PVE::Storage::path($cfg, $volid);
3880                 return if !$path || !$owner || ($owner != $vmid);
3881
3882                 # Note: only delete disk we want to restore
3883                 # other volumes will become unused
3884                 if ($virtdev_hash->{$ds}) {
3885                     PVE::Storage::vdisk_free($cfg, $volid);
3886                 }
3887             });
3888         }
3889
3890         my $map = {};
3891         foreach my $virtdev (sort keys %$virtdev_hash) {
3892             my $d = $virtdev_hash->{$virtdev};
3893             my $alloc_size = int(($d->{size} + 1024 - 1)/1024);
3894             my $scfg = PVE::Storage::storage_config($cfg, $d->{storeid});
3895
3896             # test if requested format is supported
3897             my ($defFormat, $validFormats) = PVE::Storage::storage_default_format($cfg, $d->{storeid});
3898             my $supported = grep { $_ eq $d->{format} } @$validFormats;
3899             $d->{format} = $defFormat if !$supported;
3900
3901             my $volid = PVE::Storage::vdisk_alloc($cfg, $d->{storeid}, $vmid,
3902                                                   $d->{format}, undef, $alloc_size);
3903             print STDERR "new volume ID is '$volid'\n";
3904             $d->{volid} = $volid;
3905             my $path = PVE::Storage::path($cfg, $volid);
3906
3907             my $write_zeros = 1;
3908             # fixme: what other storages types initialize volumes with zero?
3909             if ($scfg->{type} eq 'dir' || $scfg->{type} eq 'nfs' ||
3910                 $scfg->{type} eq 'sheepdog' || $scfg->{type} eq 'rbd') {
3911                 $write_zeros = 0;
3912             }
3913
3914             print $fifofh "${write_zeros}:$d->{devname}=$path\n";
3915
3916             print "map '$d->{devname}' to '$path' (write zeros = ${write_zeros})\n";
3917             $map->{$virtdev} = $volid;
3918         }
3919
3920         $fh->seek(0, 0) || die "seek failed - $!\n";
3921
3922         my $outfd = new IO::File ($tmpfn, "w") ||
3923             die "unable to write config for VM $vmid\n";
3924
3925         my $cookie = { netcount => 0 };
3926         while (defined(my $line = <$fh>)) {
3927             restore_update_config_line($outfd, $cookie, $vmid, $map, $line, $opts->{unique});
3928         }
3929
3930         $fh->close();
3931         $outfd->close();
3932     };
3933
3934     eval {
3935         # enable interrupts
3936         local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = $SIG{PIPE} = sub {
3937             die "interrupted by signal\n";
3938         };
3939         local $SIG{ALRM} = sub { die "got timeout\n"; };
3940
3941         $oldtimeout = alarm($timeout);
3942
3943         my $parser = sub {
3944             my $line = shift;
3945
3946             print "$line\n";
3947
3948             if ($line =~ m/^DEV:\sdev_id=(\d+)\ssize:\s(\d+)\sdevname:\s(\S+)$/) {
3949                 my ($dev_id, $size, $devname) = ($1, $2, $3);
3950                 $devinfo->{$devname} = { size => $size, dev_id => $dev_id };
3951             } elsif ($line =~ m/^CTIME: /) {
3952                 &$print_devmap();
3953                 print $fifofh "done\n";
3954                 my $tmp = $oldtimeout || 0;
3955                 $oldtimeout = undef;
3956                 alarm($tmp);
3957                 close($fifofh);
3958             }
3959         };
3960
3961         print "restore vma archive: $cmd\n";
3962         run_command($cmd, input => $input, outfunc => $parser, afterfork => $openfifo);
3963     };
3964     my $err = $@;
3965
3966     alarm($oldtimeout) if $oldtimeout;
3967
3968     unlink $mapfifo;
3969
3970     if ($err) {
3971         rmtree $tmpdir;
3972         unlink $tmpfn;
3973
3974         my $cfg = cfs_read_file('storage.cfg');
3975         foreach my $devname (keys %$devinfo) {
3976             my $volid = $devinfo->{$devname}->{volid};
3977             next if !$volid;
3978             eval {
3979                 if ($volid =~ m|^/|) {
3980                     unlink $volid || die 'unlink failed\n';
3981                 } else {
3982                     PVE::Storage::vdisk_free($cfg, $volid);
3983                 }
3984                 print STDERR "temporary volume '$volid' sucessfuly removed\n";
3985             };
3986             print STDERR "unable to cleanup '$volid' - $@" if $@;
3987         }
3988         die $err;
3989     }
3990
3991     rmtree $tmpdir;
3992
3993     rename($tmpfn, $conffile) ||
3994         die "unable to commit configuration file '$conffile'\n";
3995
3996     PVE::Cluster::cfs_update(); # make sure we read new file
3997
3998     eval { rescan($vmid, 1); };
3999     warn $@ if $@;
4000 }
4001
4002 sub restore_tar_archive {
4003     my ($archive, $vmid, $user, $opts) = @_;
4004
4005     if ($archive ne '-') {
4006         my $firstfile = tar_archive_read_firstfile($archive);
4007         die "ERROR: file '$archive' dos not lock like a QemuServer vzdump backup\n"
4008             if $firstfile ne 'qemu-server.conf';
4009     }
4010
4011     my $storecfg = cfs_read_file('storage.cfg');
4012
4013     # destroy existing data - keep empty config
4014     my $vmcfgfn = PVE::QemuServer::config_file($vmid);
4015     destroy_vm($storecfg, $vmid, 1) if -f $vmcfgfn;
4016
4017     my $tocmd = "/usr/lib/qemu-server/qmextract";
4018
4019     $tocmd .= " --storage " . PVE::Tools::shellquote($opts->{storage}) if $opts->{storage};
4020     $tocmd .= " --pool " . PVE::Tools::shellquote($opts->{pool}) if $opts->{pool};
4021     $tocmd .= ' --prealloc' if $opts->{prealloc};
4022     $tocmd .= ' --info' if $opts->{info};
4023
4024     # tar option "xf" does not autodetect compression when read from STDIN,
4025     # so we pipe to zcat
4026     my $cmd = "zcat -f|tar xf " . PVE::Tools::shellquote($archive) . " " .
4027         PVE::Tools::shellquote("--to-command=$tocmd");
4028
4029     my $tmpdir = "/var/tmp/vzdumptmp$$";
4030     mkpath $tmpdir;
4031
4032     local $ENV{VZDUMP_TMPDIR} = $tmpdir;
4033     local $ENV{VZDUMP_VMID} = $vmid;
4034     local $ENV{VZDUMP_USER} = $user;
4035
4036     my $conffile = config_file($vmid);
4037     my $tmpfn = "$conffile.$$.tmp";
4038
4039     # disable interrupts (always do cleanups)
4040     local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = sub {
4041         print STDERR "got interrupt - ignored\n";
4042     };
4043
4044     eval {
4045         # enable interrupts
4046         local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = $SIG{PIPE} = sub {
4047             die "interrupted by signal\n";
4048         };
4049
4050         if ($archive eq '-') {
4051             print "extracting archive from STDIN\n";
4052             run_command($cmd, input => "<&STDIN");
4053         } else {
4054             print "extracting archive '$archive'\n";
4055             run_command($cmd);
4056         }
4057
4058         return if $opts->{info};
4059
4060         # read new mapping
4061         my $map = {};
4062         my $statfile = "$tmpdir/qmrestore.stat";
4063         if (my $fd = IO::File->new($statfile, "r")) {
4064             while (defined (my $line = <$fd>)) {
4065                 if ($line =~ m/vzdump:([^\s:]*):(\S+)$/) {
4066                     $map->{$1} = $2 if $1;
4067                 } else {
4068                     print STDERR "unable to parse line in statfile - $line\n";
4069                 }
4070             }
4071             $fd->close();
4072         }
4073
4074         my $confsrc = "$tmpdir/qemu-server.conf";
4075
4076         my $srcfd = new IO::File($confsrc, "r") ||
4077             die "unable to open file '$confsrc'\n";
4078
4079         my $outfd = new IO::File ($tmpfn, "w") ||
4080             die "unable to write config for VM $vmid\n";
4081
4082         my $cookie = { netcount => 0 };
4083         while (defined (my $line = <$srcfd>)) {
4084             restore_update_config_line($outfd, $cookie, $vmid, $map, $line, $opts->{unique});
4085         }
4086
4087         $srcfd->close();
4088         $outfd->close();
4089     };
4090     my $err = $@;
4091
4092     if ($err) {
4093
4094         unlink $tmpfn;
4095
4096         tar_restore_cleanup($storecfg, "$tmpdir/qmrestore.stat") if !$opts->{info};
4097
4098         die $err;
4099     }
4100
4101     rmtree $tmpdir;
4102
4103     rename $tmpfn, $conffile ||
4104         die "unable to commit configuration file '$conffile'\n";
4105
4106     PVE::Cluster::cfs_update(); # make sure we read new file
4107
4108     eval { rescan($vmid, 1); };
4109     warn $@ if $@;
4110 };
4111
4112
4113 # Internal snapshots
4114
4115 # NOTE: Snapshot create/delete involves several non-atomic
4116 # action, and can take a long time.
4117 # So we try to avoid locking the file and use 'lock' variable
4118 # inside the config file instead.
4119
4120 my $snapshot_copy_config = sub {
4121     my ($source, $dest) = @_;
4122
4123     foreach my $k (keys %$source) {
4124         next if $k eq 'snapshots';
4125         next if $k eq 'snapstate';
4126         next if $k eq 'snaptime';
4127         next if $k eq 'vmstate';
4128         next if $k eq 'lock';
4129         next if $k eq 'digest';
4130         next if $k eq 'description';
4131         next if $k =~ m/^unused\d+$/;
4132
4133         $dest->{$k} = $source->{$k};
4134     }
4135 };
4136
4137 my $snapshot_apply_config = sub {
4138     my ($conf, $snap) = @_;
4139
4140     # copy snapshot list
4141     my $newconf = {
4142         snapshots => $conf->{snapshots},
4143     };
4144
4145     # keep description and list of unused disks
4146     foreach my $k (keys %$conf) {
4147         next if !($k =~ m/^unused\d+$/ || $k eq 'description');
4148         $newconf->{$k} = $conf->{$k};
4149     }
4150
4151     &$snapshot_copy_config($snap, $newconf);
4152
4153     return $newconf;
4154 };
4155
4156 sub foreach_writable_storage {
4157     my ($conf, $func) = @_;
4158
4159     my $sidhash = {};
4160
4161     foreach my $ds (keys %$conf) {
4162         next if !valid_drivename($ds);
4163
4164         my $drive = parse_drive($ds, $conf->{$ds});
4165         next if !$drive;
4166         next if drive_is_cdrom($drive);
4167
4168         my $volid = $drive->{file};
4169
4170         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
4171         $sidhash->{$sid} = $sid if $sid;
4172     }
4173
4174     foreach my $sid (sort keys %$sidhash) {
4175         &$func($sid);
4176     }
4177 }
4178
4179 my $alloc_vmstate_volid = sub {
4180     my ($storecfg, $vmid, $conf, $snapname) = @_;
4181
4182     # Note: we try to be smart when selecting a $target storage
4183
4184     my $target;
4185
4186     # search shared storage first
4187     foreach_writable_storage($conf, sub {
4188         my ($sid) = @_;
4189         my $scfg = PVE::Storage::storage_config($storecfg, $sid);
4190         return if !$scfg->{shared};
4191
4192         $target = $sid if !$target || $scfg->{path}; # prefer file based storage
4193     });
4194
4195     if (!$target) {
4196         # now search local storage
4197         foreach_writable_storage($conf, sub {
4198             my ($sid) = @_;
4199             my $scfg = PVE::Storage::storage_config($storecfg, $sid);
4200             return if $scfg->{shared};
4201
4202             $target = $sid if !$target || $scfg->{path}; # prefer file based storage;
4203         });
4204     }
4205
4206     $target = 'local' if !$target;
4207
4208     my $driver_state_size = 500; # assume 32MB is enough to safe all driver state;
4209     # we abort live save after $conf->{memory}, so we need at max twice that space
4210     my $size = $conf->{memory}*2 + $driver_state_size;
4211
4212     my $name = "vm-$vmid-state-$snapname";
4213     my $scfg = PVE::Storage::storage_config($storecfg, $target);
4214     $name .= ".raw" if $scfg->{path}; # add filename extension for file base storage
4215     my $volid = PVE::Storage::vdisk_alloc($storecfg, $target, $vmid, 'raw', $name, $size*1024);
4216
4217     return $volid;
4218 };
4219
4220 my $snapshot_prepare = sub {
4221     my ($vmid, $snapname, $save_vmstate, $comment) = @_;
4222
4223     my $snap;
4224
4225     my $updatefn =  sub {
4226
4227         my $conf = load_config($vmid);
4228
4229         die "you can't take a snapshot if it's a template\n"
4230             if is_template($conf);
4231
4232         check_lock($conf);
4233
4234         $conf->{lock} = 'snapshot';
4235
4236         die "snapshot name '$snapname' already used\n"
4237             if defined($conf->{snapshots}->{$snapname});
4238
4239         my $storecfg = PVE::Storage::config();
4240         die "snapshot feature is not available" if !has_feature('snapshot', $conf, $storecfg);
4241
4242         $snap = $conf->{snapshots}->{$snapname} = {};
4243
4244         if ($save_vmstate && check_running($vmid)) {
4245             $snap->{vmstate} = &$alloc_vmstate_volid($storecfg, $vmid, $conf, $snapname);
4246         }
4247
4248         &$snapshot_copy_config($conf, $snap);
4249
4250         $snap->{snapstate} = "prepare";
4251         $snap->{snaptime} = time();
4252         $snap->{description} = $comment if $comment;
4253
4254         update_config_nolock($vmid, $conf, 1);
4255     };
4256
4257     lock_config($vmid, $updatefn);
4258
4259     return $snap;
4260 };
4261
4262 my $snapshot_commit = sub {
4263     my ($vmid, $snapname) = @_;
4264
4265     my $updatefn = sub {
4266
4267         my $conf = load_config($vmid);
4268
4269         die "missing snapshot lock\n"
4270             if !($conf->{lock} && $conf->{lock} eq 'snapshot');
4271
4272         my $snap = $conf->{snapshots}->{$snapname};
4273
4274         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4275
4276         die "wrong snapshot state\n"
4277             if !($snap->{snapstate} && $snap->{snapstate} eq "prepare");
4278
4279         delete $snap->{snapstate};
4280         delete $conf->{lock};
4281
4282         my $newconf = &$snapshot_apply_config($conf, $snap);
4283
4284         $newconf->{parent} = $snapname;
4285
4286         update_config_nolock($vmid, $newconf, 1);
4287     };
4288
4289     lock_config($vmid, $updatefn);
4290 };
4291
4292 sub snapshot_rollback {
4293     my ($vmid, $snapname) = @_;
4294
4295     my $snap;
4296
4297     my $prepare = 1;
4298
4299     my $storecfg = PVE::Storage::config();
4300
4301     my $updatefn = sub {
4302
4303         my $conf = load_config($vmid);
4304
4305         die "you can't rollback if vm is a template\n" if is_template($conf);
4306
4307         $snap = $conf->{snapshots}->{$snapname};
4308
4309         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4310
4311         die "unable to rollback to incomplete snapshot (snapstate = $snap->{snapstate})\n"
4312             if $snap->{snapstate};
4313
4314         if ($prepare) {
4315             check_lock($conf);
4316             vm_stop($storecfg, $vmid, undef, undef, 5, undef, undef);
4317         }
4318
4319         die "unable to rollback vm $vmid: vm is running\n"
4320             if check_running($vmid);
4321
4322         if ($prepare) {
4323             $conf->{lock} = 'rollback';
4324         } else {
4325             die "got wrong lock\n" if !($conf->{lock} && $conf->{lock} eq 'rollback');
4326             delete $conf->{lock};
4327         }
4328
4329         if (!$prepare) {
4330             # copy snapshot config to current config
4331             $conf = &$snapshot_apply_config($conf, $snap);
4332             $conf->{parent} = $snapname;
4333         }
4334
4335         update_config_nolock($vmid, $conf, 1);
4336
4337         if (!$prepare && $snap->{vmstate}) {
4338             my $statefile = PVE::Storage::path($storecfg, $snap->{vmstate});
4339             vm_start($storecfg, $vmid, $statefile);
4340         }
4341     };
4342
4343     lock_config($vmid, $updatefn);
4344
4345     foreach_drive($snap, sub {
4346         my ($ds, $drive) = @_;
4347
4348         return if drive_is_cdrom($drive);
4349
4350         my $volid = $drive->{file};
4351         my $device = "drive-$ds";
4352
4353         PVE::Storage::volume_snapshot_rollback($storecfg, $volid, $snapname);
4354     });
4355
4356     $prepare = 0;
4357     lock_config($vmid, $updatefn);
4358 }
4359
4360 my $savevm_wait = sub {
4361     my ($vmid) = @_;
4362
4363     for(;;) {
4364         my $stat = vm_mon_cmd_nocheck($vmid, "query-savevm");
4365         if (!$stat->{status}) {
4366             die "savevm not active\n";
4367         } elsif ($stat->{status} eq 'active') {
4368             sleep(1);
4369             next;
4370         } elsif ($stat->{status} eq 'completed') {
4371             last;
4372         } else {
4373             die "query-savevm returned status '$stat->{status}'\n";
4374         }
4375     }
4376 };
4377
4378 sub snapshot_create {
4379     my ($vmid, $snapname, $save_vmstate, $freezefs, $comment) = @_;
4380
4381     my $snap = &$snapshot_prepare($vmid, $snapname, $save_vmstate, $comment);
4382
4383     $freezefs = $save_vmstate = 0 if !$snap->{vmstate}; # vm is not running
4384
4385     my $drivehash = {};
4386
4387     my $running = check_running($vmid);
4388
4389     eval {
4390         # create internal snapshots of all drives
4391
4392         my $storecfg = PVE::Storage::config();
4393
4394         if ($running) {
4395             if ($snap->{vmstate}) {
4396                 my $path = PVE::Storage::path($storecfg, $snap->{vmstate});
4397                 vm_mon_cmd($vmid, "savevm-start", statefile => $path);
4398                 &$savevm_wait($vmid);
4399             } else {
4400                 vm_mon_cmd($vmid, "savevm-start");
4401             }
4402         };
4403
4404         qga_freezefs($vmid) if $running && $freezefs;
4405
4406         foreach_drive($snap, sub {
4407             my ($ds, $drive) = @_;
4408
4409             return if drive_is_cdrom($drive);
4410
4411             my $volid = $drive->{file};
4412             my $device = "drive-$ds";
4413
4414             qemu_volume_snapshot($vmid, $device, $storecfg, $volid, $snapname);
4415             $drivehash->{$ds} = 1;
4416        });
4417     };
4418     my $err = $@;
4419
4420     eval { qga_unfreezefs($vmid) if $running && $freezefs; };
4421     warn $@ if $@;
4422
4423     eval { vm_mon_cmd($vmid, "savevm-end") if $running; };
4424     warn $@ if $@;
4425
4426     if ($err) {
4427         warn "snapshot create failed: starting cleanup\n";
4428         eval { snapshot_delete($vmid, $snapname, 0, $drivehash); };
4429         warn $@ if $@;
4430         die $err;
4431     }
4432
4433     &$snapshot_commit($vmid, $snapname);
4434 }
4435
4436 # Note: $drivehash is only set when called from snapshot_create.
4437 sub snapshot_delete {
4438     my ($vmid, $snapname, $force, $drivehash) = @_;
4439
4440     my $prepare = 1;
4441
4442     my $snap;
4443     my $unused = [];
4444
4445     my $unlink_parent = sub {
4446         my ($confref, $new_parent) = @_;
4447
4448         if ($confref->{parent} && $confref->{parent} eq $snapname) {
4449             if ($new_parent) {
4450                 $confref->{parent} = $new_parent;
4451             } else {
4452                 delete $confref->{parent};
4453             }
4454         }
4455     };
4456
4457     my $updatefn =  sub {
4458         my ($remove_drive) = @_;
4459
4460         my $conf = load_config($vmid);
4461
4462         if (!$drivehash) {
4463             check_lock($conf);
4464             die "you can't delete a snapshot if vm is a template\n"
4465                 if is_template($conf);
4466         }
4467
4468         $snap = $conf->{snapshots}->{$snapname};
4469
4470         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4471
4472         # remove parent refs
4473         &$unlink_parent($conf, $snap->{parent});
4474         foreach my $sn (keys %{$conf->{snapshots}}) {
4475             next if $sn eq $snapname;
4476             &$unlink_parent($conf->{snapshots}->{$sn}, $snap->{parent});
4477         }
4478
4479         if ($remove_drive) {
4480             if ($remove_drive eq 'vmstate') {
4481                 delete $snap->{$remove_drive};
4482             } else {
4483                 my $drive = parse_drive($remove_drive, $snap->{$remove_drive});
4484                 my $volid = $drive->{file};
4485                 delete $snap->{$remove_drive};
4486                 add_unused_volume($conf, $volid);
4487             }
4488         }
4489
4490         if ($prepare) {
4491             $snap->{snapstate} = 'delete';
4492         } else {
4493             delete $conf->{snapshots}->{$snapname};
4494             delete $conf->{lock} if $drivehash;
4495             foreach my $volid (@$unused) {
4496                 add_unused_volume($conf, $volid);
4497             }
4498         }
4499
4500         update_config_nolock($vmid, $conf, 1);
4501     };
4502
4503     lock_config($vmid, $updatefn);
4504
4505     # now remove vmstate file
4506
4507     my $storecfg = PVE::Storage::config();
4508
4509     if ($snap->{vmstate}) {
4510         eval {  PVE::Storage::vdisk_free($storecfg, $snap->{vmstate}); };
4511         if (my $err = $@) {
4512             die $err if !$force;
4513             warn $err;
4514         }
4515         # save changes (remove vmstate from snapshot)
4516         lock_config($vmid, $updatefn, 'vmstate') if !$force;
4517     };
4518
4519     # now remove all internal snapshots
4520     foreach_drive($snap, sub {
4521         my ($ds, $drive) = @_;
4522
4523         return if drive_is_cdrom($drive);
4524
4525         my $volid = $drive->{file};
4526         my $device = "drive-$ds";
4527
4528         if (!$drivehash || $drivehash->{$ds}) {
4529             eval { qemu_volume_snapshot_delete($vmid, $device, $storecfg, $volid, $snapname); };
4530             if (my $err = $@) {
4531                 die $err if !$force;
4532                 warn $err;
4533             }
4534         }
4535
4536         # save changes (remove drive fron snapshot)
4537         lock_config($vmid, $updatefn, $ds) if !$force;
4538         push @$unused, $volid;
4539     });
4540
4541     # now cleanup config
4542     $prepare = 0;
4543     lock_config($vmid, $updatefn);
4544 }
4545
4546 sub has_feature {
4547     my ($feature, $conf, $storecfg, $snapname, $running) = @_;
4548
4549     my $err;
4550     foreach_drive($conf, sub {
4551         my ($ds, $drive) = @_;
4552
4553         return if drive_is_cdrom($drive);
4554         my $volid = $drive->{file};
4555         $err = 1 if !PVE::Storage::volume_has_feature($storecfg, $feature, $volid, $snapname, $running);
4556     });
4557
4558     return $err ? 0 : 1;
4559 }
4560
4561 sub template_create {
4562     my ($vmid, $conf, $disk) = @_;
4563
4564     my $storecfg = PVE::Storage::config();
4565
4566     foreach_drive($conf, sub {
4567         my ($ds, $drive) = @_;
4568
4569         return if drive_is_cdrom($drive);
4570         return if $disk && $ds ne $disk;
4571
4572         my $volid = $drive->{file};
4573         return if !PVE::Storage::volume_has_feature($storecfg, 'template', $volid);
4574
4575         my $voliddst = PVE::Storage::vdisk_create_base($storecfg, $volid);
4576         $drive->{file} = $voliddst;
4577         $conf->{$ds} = print_drive($vmid, $drive);
4578         update_config_nolock($vmid, $conf, 1);
4579     });
4580 }
4581
4582 sub is_template {
4583     my ($conf) = @_;
4584
4585     return 1 if defined $conf->{template} && $conf->{template} == 1;
4586 }
4587
4588 sub qemu_img_convert {
4589     my ($src_volid, $dst_volid, $size, $snapname) = @_;
4590
4591     my $storecfg = PVE::Storage::config();
4592     my ($src_storeid, $src_volname) = PVE::Storage::parse_volume_id($src_volid, 1);
4593     my ($dst_storeid, $dst_volname) = PVE::Storage::parse_volume_id($dst_volid, 1);
4594
4595     if ($src_storeid && $dst_storeid) {
4596         my $src_scfg = PVE::Storage::storage_config($storecfg, $src_storeid);
4597         my $dst_scfg = PVE::Storage::storage_config($storecfg, $dst_storeid);
4598
4599         my $src_format = qemu_img_format($src_scfg, $src_volname);
4600         my $dst_format = qemu_img_format($dst_scfg, $dst_volname);
4601
4602         my $src_path = PVE::Storage::path($storecfg, $src_volid, $snapname);
4603         my $dst_path = PVE::Storage::path($storecfg, $dst_volid);
4604
4605         my $cmd = [];
4606         push @$cmd, '/usr/bin/qemu-img', 'convert', '-t', 'writeback', '-p', '-C';
4607         push @$cmd, '-s', $snapname if($snapname && $src_format eq "qcow2");
4608         push @$cmd, '-f', $src_format, '-O', $dst_format, $src_path, $dst_path;
4609
4610         my $parser = sub {
4611             my $line = shift;
4612             if($line =~ m/\((\S+)\/100\%\)/){
4613                 my $percent = $1;
4614                 my $transferred = int($size * $percent / 100);
4615                 my $remaining = $size - $transferred;
4616
4617                 print "transferred: $transferred bytes remaining: $remaining bytes total: $size bytes progression: $percent %\n";
4618             }
4619
4620         };
4621
4622         eval  { run_command($cmd, timeout => undef, outfunc => $parser); };
4623         my $err = $@;
4624         die "copy failed: $err" if $err;
4625     }
4626 }
4627
4628 sub qemu_img_format {
4629     my ($scfg, $volname) = @_;
4630
4631     if ($scfg->{path} && $volname =~ m/\.(raw|qcow2|qed|vmdk)$/) {
4632         return $1;
4633     } elsif ($scfg->{type} eq 'iscsi') {
4634         return "host_device";
4635     } else {
4636         return "raw";
4637     }
4638 }
4639
4640 sub qemu_drive_mirror {
4641     my ($vmid, $drive, $dst_volid, $vmiddst, $maxwait) = @_;
4642
4643     my $count = 1;
4644     my $old_len = 0;
4645     my $frozen = undef;
4646
4647     my $storecfg = PVE::Storage::config();
4648     my ($dst_storeid, $dst_volname) = PVE::Storage::parse_volume_id($dst_volid, 1);
4649
4650     if ($dst_storeid) {
4651         my $dst_scfg = PVE::Storage::storage_config($storecfg, $dst_storeid);
4652
4653         my $format;
4654         if ($dst_volname =~ m/\.(raw|qcow2)$/){
4655             $format = $1;
4656         }
4657
4658         my $dst_path = PVE::Storage::path($storecfg, $dst_volid);
4659
4660         if ($format) {
4661             #fixme : sometime drive-mirror timeout, but works fine after.
4662             # (I have see the problem with big volume > 200GB), so we need to eval
4663             eval { vm_mon_cmd($vmid, "drive-mirror", timeout => 10, device => "drive-$drive", mode => "existing",
4664                               sync => "full", target => $dst_path, format => $format); };
4665         } else {
4666             eval { vm_mon_cmd($vmid, "drive-mirror", timeout => 10, device => "drive-$drive", mode => "existing",
4667                               sync => "full", target => $dst_path); };
4668         }
4669
4670         eval {
4671             while (1) {
4672                 my $stats = vm_mon_cmd($vmid, "query-block-jobs");
4673                 my $stat = @$stats[0];
4674                 die "mirroring job seem to have die. Maybe do you have bad sectors?" if !$stat;
4675                 die "error job is not mirroring" if $stat->{type} ne "mirror";
4676
4677                 my $transferred = $stat->{offset};
4678                 my $total = $stat->{len};
4679                 my $remaining = $total - $transferred;
4680                 my $percent = sprintf "%.2f", ($transferred * 100 / $total);
4681
4682                 print "transferred: $transferred bytes remaining: $remaining bytes total: $total bytes progression: $percent %\n";
4683
4684                 last if ($stat->{len} == $stat->{offset});
4685                 if ($old_len == $stat->{offset}) {
4686                     if ($maxwait && $count > $maxwait) {
4687                     # if writes to disk occurs the disk needs to be freezed
4688                     # to be able to complete the migration
4689                         vm_suspend($vmid,1);
4690                         $count = 0;
4691                         $frozen = 1;
4692                     } else {
4693                         $count++ unless $frozen;
4694                     }
4695                 } elsif ($frozen) {
4696                     vm_resume($vmid,1);
4697                     $count = 0;
4698                 }
4699                 $old_len = $stat->{offset};
4700                 sleep 1;
4701             }
4702
4703             if ($vmiddst == $vmid) {
4704                 # switch the disk if source and destination are on the same guest
4705                 vm_mon_cmd($vmid, "block-job-complete", device => "drive-$drive");
4706             }
4707         };
4708         if (my $err = $@) {
4709             eval { vm_mon_cmd($vmid, "block-job-cancel", device => "drive-$drive"); };
4710             die "mirroring error: $err";
4711         }
4712
4713         if ($vmiddst != $vmid) {
4714             # if we clone a disk for a new target vm, we don't switch the disk
4715             vm_mon_cmd($vmid, "block-job-cancel", device => "drive-$drive");
4716         }
4717     }
4718 }
4719
4720 sub clone_disk {
4721     my ($storecfg, $vmid, $running, $drivename, $drive, $snapname,
4722         $newvmid, $storage, $format, $full, $newvollist) = @_;
4723
4724     my $newvolid;
4725
4726     if (!$full) {
4727         print "create linked clone of drive $drivename ($drive->{file})\n";
4728         $newvolid = PVE::Storage::vdisk_clone($storecfg,  $drive->{file}, $newvmid);
4729         push @$newvollist, $newvolid;
4730     } else {
4731         my ($storeid, $volname) = PVE::Storage::parse_volume_id($drive->{file});
4732         $storeid = $storage if $storage;
4733
4734         my ($defFormat, $validFormats) = PVE::Storage::storage_default_format($storecfg, $storeid);
4735         if (!$format) {
4736             $format = $drive->{format} || $defFormat;
4737         }
4738
4739         # test if requested format is supported - else use default
4740         my $supported = grep { $_ eq $format } @$validFormats;
4741         $format = $defFormat if !$supported;
4742
4743         my ($size) = PVE::Storage::volume_size_info($storecfg, $drive->{file}, 3);
4744
4745         print "create full clone of drive $drivename ($drive->{file})\n";
4746         $newvolid = PVE::Storage::vdisk_alloc($storecfg, $storeid, $newvmid, $format, undef, ($size/1024));
4747         push @$newvollist, $newvolid;
4748
4749         if (!$running || $snapname) {
4750             qemu_img_convert($drive->{file}, $newvolid, $size, $snapname);
4751         } else {
4752             qemu_drive_mirror($vmid, $drivename, $newvolid, $newvmid);
4753         }
4754     }
4755
4756     my ($size) = PVE::Storage::volume_size_info($storecfg, $newvolid, 3);
4757
4758     my $disk = $drive;
4759     $disk->{format} = undef;
4760     $disk->{file} = $newvolid;
4761     $disk->{size} = $size;
4762
4763     return $disk;
4764 }
4765
4766 1;