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, $forcemachine) = @_;
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     my $machine_type = $forcemachine || $conf->{machine};
2388     if ($machine_type) {
2389         push @$machineFlags, "type=${machine_type}";
2390     }
2391
2392     if ($conf->{startdate}) {
2393         push @$rtcFlags, "base=$conf->{startdate}";
2394     } elsif ($useLocaltime) {
2395         push @$rtcFlags, 'base=localtime';
2396     }
2397
2398     push @$cmd, '-S' if $conf->{freeze};
2399
2400     # set keyboard layout
2401     my $kb = $conf->{keyboard} || $defaults->{keyboard};
2402     push @$cmd, '-k', $kb if $kb;
2403
2404     # enable sound
2405     #my $soundhw = $conf->{soundhw} || $defaults->{soundhw};
2406     #push @$cmd, '-soundhw', 'es1370';
2407     #push @$cmd, '-soundhw', $soundhw if $soundhw;
2408
2409     if($conf->{agent}) {
2410         my $qgasocket = qga_socket($vmid);
2411         my $pciaddr = print_pci_addr("qga0", $bridges);
2412         push @$devices, '-chardev', "socket,path=$qgasocket,server,nowait,id=qga0";
2413         push @$devices, '-device', "virtio-serial,id=qga0$pciaddr";
2414         push @$devices, '-device', 'virtserialport,chardev=qga0,name=org.qemu.guest_agent.0';
2415     }
2416
2417     # enable balloon by default, unless explicitly disabled
2418     if (!defined($conf->{balloon}) || $conf->{balloon}) {
2419         $pciaddr = print_pci_addr("balloon0", $bridges);
2420         push @$devices, '-device', "virtio-balloon-pci,id=balloon0$pciaddr";
2421     }
2422
2423     if ($conf->{watchdog}) {
2424         my $wdopts = parse_watchdog($conf->{watchdog});
2425         $pciaddr = print_pci_addr("watchdog", $bridges);
2426         my $watchdog = $wdopts->{model} || 'i6300esb';
2427         push @$devices, '-device', "$watchdog$pciaddr";
2428         push @$devices, '-watchdog-action', $wdopts->{action} if $wdopts->{action};
2429     }
2430
2431     my $vollist = [];
2432     my $scsicontroller = {};
2433     my $ahcicontroller = {};
2434     my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : $defaults->{scsihw};
2435
2436     foreach_drive($conf, sub {
2437         my ($ds, $drive) = @_;
2438
2439         if (PVE::Storage::parse_volume_id($drive->{file}, 1)) {
2440             push @$vollist, $drive->{file};
2441         }
2442
2443         $use_virtio = 1 if $ds =~ m/^virtio/;
2444
2445         if (drive_is_cdrom ($drive)) {
2446             if ($bootindex_hash->{d}) {
2447                 $drive->{bootindex} = $bootindex_hash->{d};
2448                 $bootindex_hash->{d} += 1;
2449             }
2450         } else {
2451             if ($bootindex_hash->{c}) {
2452                 $drive->{bootindex} = $bootindex_hash->{c} if $conf->{bootdisk} && ($conf->{bootdisk} eq $ds);
2453                 $bootindex_hash->{c} += 1;
2454             }
2455         }
2456
2457         if ($drive->{interface} eq 'scsi') {
2458
2459             my $maxdev = ($scsihw ne 'lsi') ? 256 : 7;
2460             my $controller = int($drive->{index} / $maxdev);
2461             $pciaddr = print_pci_addr("scsihw$controller", $bridges);
2462             push @$devices, '-device', "$scsihw,id=scsihw$controller$pciaddr" if !$scsicontroller->{$controller};
2463             $scsicontroller->{$controller}=1;
2464         }
2465
2466         if ($drive->{interface} eq 'sata') {
2467            my $controller = int($drive->{index} / $MAX_SATA_DISKS);
2468            $pciaddr = print_pci_addr("ahci$controller", $bridges);
2469            push @$devices, '-device', "ahci,id=ahci$controller,multifunction=on$pciaddr" if !$ahcicontroller->{$controller};
2470            $ahcicontroller->{$controller}=1;
2471         }
2472
2473         push @$devices, '-drive',print_drive_full($storecfg, $vmid, $drive);
2474         push @$devices, '-device',print_drivedevice_full($storecfg, $conf, $vmid, $drive, $bridges);
2475     });
2476
2477     push @$cmd, '-m', $conf->{memory} || $defaults->{memory};
2478
2479     for (my $i = 0; $i < $MAX_NETS; $i++) {
2480          next if !$conf->{"net$i"};
2481          my $d = parse_net($conf->{"net$i"});
2482          next if !$d;
2483
2484          $use_virtio = 1 if $d->{model} eq 'virtio';
2485
2486          if ($bootindex_hash->{n}) {
2487             $d->{bootindex} = $bootindex_hash->{n};
2488             $bootindex_hash->{n} += 1;
2489          }
2490
2491          my $netdevfull = print_netdev_full($vmid,$conf,$d,"net$i");
2492          push @$devices, '-netdev', $netdevfull;
2493
2494          my $netdevicefull = print_netdevice_full($vmid,$conf,$d,"net$i",$bridges);
2495          push @$devices, '-device', $netdevicefull;
2496     }
2497
2498     #bridges
2499     while (my ($k, $v) = each %$bridges) {
2500         $pciaddr = print_pci_addr("pci.$k");
2501         unshift @$devices, '-device', "pci-bridge,id=pci.$k,chassis_nr=$k$pciaddr" if $k > 0;
2502     }
2503
2504
2505     # hack: virtio with fairsched is unreliable, so we do not use fairsched
2506     # when the VM uses virtio devices.
2507     if (!$use_virtio && $have_ovz) {
2508
2509         my $cpuunits = defined($conf->{cpuunits}) ?
2510             $conf->{cpuunits} : $defaults->{cpuunits};
2511
2512         push @$cmd, '-cpuunits', $cpuunits if $cpuunits;
2513
2514         # fixme: cpulimit is currently ignored
2515         #push @$cmd, '-cpulimit', $conf->{cpulimit} if $conf->{cpulimit};
2516     }
2517
2518     # add custom args
2519     if ($conf->{args}) {
2520         my $aa = PVE::Tools::split_args($conf->{args});
2521         push @$cmd, @$aa;
2522     }
2523
2524     push @$cmd, @$devices;
2525     push @$cmd, '-rtc', join(',', @$rtcFlags)
2526         if scalar(@$rtcFlags);
2527     push @$cmd, '-machine', join(',', @$machineFlags)
2528         if scalar(@$machineFlags);
2529     push @$cmd, '-global', join(',', @$globalFlags)
2530         if scalar(@$globalFlags);
2531
2532     return wantarray ? ($cmd, $vollist) : $cmd;
2533 }
2534
2535 sub vnc_socket {
2536     my ($vmid) = @_;
2537     return "${var_run_tmpdir}/$vmid.vnc";
2538 }
2539
2540 sub qmp_socket {
2541     my ($vmid) = @_;
2542     return "${var_run_tmpdir}/$vmid.qmp";
2543 }
2544
2545 sub qga_socket {
2546     my ($vmid) = @_;
2547     return "${var_run_tmpdir}/$vmid.qga";
2548 }
2549
2550 sub pidfile_name {
2551     my ($vmid) = @_;
2552     return "${var_run_tmpdir}/$vmid.pid";
2553 }
2554
2555 sub vm_devices_list {
2556     my ($vmid) = @_;
2557
2558     my $res = vm_mon_cmd($vmid, 'query-pci');
2559
2560     my $devices = {};
2561     foreach my $pcibus (@$res) {
2562         foreach my $device (@{$pcibus->{devices}}) {
2563             next if !$device->{'qdev_id'};
2564             $devices->{$device->{'qdev_id'}} = $device;
2565         }
2566     }
2567
2568     return $devices;
2569 }
2570
2571 sub vm_deviceplug {
2572     my ($storecfg, $conf, $vmid, $deviceid, $device) = @_;
2573
2574     return 1 if !check_running($vmid);
2575
2576     if ($deviceid eq 'tablet') {
2577         my $devicefull = "usb-tablet,id=tablet,bus=uhci.0,port=1";
2578         qemu_deviceadd($vmid, $devicefull);
2579         return 1;
2580     }
2581
2582     return 1 if !$conf->{hotplug};
2583
2584     my $devices_list = vm_devices_list($vmid);
2585     return 1 if defined($devices_list->{$deviceid});
2586
2587     qemu_bridgeadd($storecfg, $conf, $vmid, $deviceid); #add bridge if we need it for the device
2588
2589     if ($deviceid =~ m/^(virtio)(\d+)$/) {
2590         return undef if !qemu_driveadd($storecfg, $vmid, $device);
2591         my $devicefull = print_drivedevice_full($storecfg, $conf, $vmid, $device);
2592         qemu_deviceadd($vmid, $devicefull);
2593         if(!qemu_deviceaddverify($vmid, $deviceid)) {
2594            qemu_drivedel($vmid, $deviceid);
2595            return undef;
2596         }
2597     }
2598
2599     if ($deviceid =~ m/^(scsihw)(\d+)$/) {
2600         my $scsihw = defined($conf->{scsihw}) ? $conf->{scsihw} : "lsi";
2601         my $pciaddr = print_pci_addr($deviceid);
2602         my $devicefull = "$scsihw,id=$deviceid$pciaddr";
2603         qemu_deviceadd($vmid, $devicefull);
2604         return undef if(!qemu_deviceaddverify($vmid, $deviceid));
2605     }
2606
2607     if ($deviceid =~ m/^(scsi)(\d+)$/) {
2608         return 1 if ($conf->{scsihw} && $conf->{scsihw} ne 'lsi'); #virtio-scsi not yet support hotplug
2609         return undef if !qemu_findorcreatescsihw($storecfg,$conf, $vmid, $device);
2610         return undef if !qemu_driveadd($storecfg, $vmid, $device);
2611         my $devicefull = print_drivedevice_full($storecfg, $conf, $vmid, $device);
2612         if(!qemu_deviceadd($vmid, $devicefull)) {
2613            qemu_drivedel($vmid, $deviceid);
2614            return undef;
2615         }
2616     }
2617
2618     if ($deviceid =~ m/^(net)(\d+)$/) {
2619         return undef if !qemu_netdevadd($vmid, $conf, $device, $deviceid);
2620         my $netdevicefull = print_netdevice_full($vmid, $conf, $device, $deviceid);
2621         qemu_deviceadd($vmid, $netdevicefull);
2622         if(!qemu_deviceaddverify($vmid, $deviceid)) {
2623            qemu_netdevdel($vmid, $deviceid);
2624            return undef;
2625         }
2626     }
2627
2628     if ($deviceid =~ m/^(pci\.)(\d+)$/) {
2629         my $bridgeid = $2;
2630         my $pciaddr = print_pci_addr($deviceid);
2631         my $devicefull = "pci-bridge,id=pci.$bridgeid,chassis_nr=$bridgeid$pciaddr";
2632         qemu_deviceadd($vmid, $devicefull);
2633         return undef if !qemu_deviceaddverify($vmid, $deviceid);
2634     }
2635
2636     return 1;
2637 }
2638
2639 sub vm_deviceunplug {
2640     my ($vmid, $conf, $deviceid) = @_;
2641
2642     return 1 if !check_running ($vmid);
2643
2644     if ($deviceid eq 'tablet') {
2645         qemu_devicedel($vmid, $deviceid);
2646         return 1;
2647     }
2648
2649     return 1 if !$conf->{hotplug};
2650
2651     my $devices_list = vm_devices_list($vmid);
2652     return 1 if !defined($devices_list->{$deviceid});
2653
2654     die "can't unplug bootdisk" if $conf->{bootdisk} && $conf->{bootdisk} eq $deviceid;
2655
2656     if ($deviceid =~ m/^(virtio)(\d+)$/) {
2657         qemu_devicedel($vmid, $deviceid);
2658         return undef if !qemu_devicedelverify($vmid, $deviceid);
2659         return undef if !qemu_drivedel($vmid, $deviceid);
2660     }
2661
2662     if ($deviceid =~ m/^(lsi)(\d+)$/) {
2663         return undef if !qemu_devicedel($vmid, $deviceid);
2664     }
2665
2666     if ($deviceid =~ m/^(scsi)(\d+)$/) {
2667         return undef if !qemu_devicedel($vmid, $deviceid);
2668         return undef if !qemu_drivedel($vmid, $deviceid);
2669     }
2670
2671     if ($deviceid =~ m/^(net)(\d+)$/) {
2672         qemu_devicedel($vmid, $deviceid);
2673         return undef if !qemu_devicedelverify($vmid, $deviceid);
2674         return undef if !qemu_netdevdel($vmid, $deviceid);
2675     }
2676
2677     return 1;
2678 }
2679
2680 sub qemu_deviceadd {
2681     my ($vmid, $devicefull) = @_;
2682
2683     $devicefull = "driver=".$devicefull;
2684     my %options =  split(/[=,]/, $devicefull);
2685
2686     vm_mon_cmd($vmid, "device_add" , %options);
2687     return 1;
2688 }
2689
2690 sub qemu_devicedel {
2691     my($vmid, $deviceid) = @_;
2692     my $ret = vm_mon_cmd($vmid, "device_del", id => $deviceid);
2693     return 1;
2694 }
2695
2696 sub qemu_driveadd {
2697     my($storecfg, $vmid, $device) = @_;
2698
2699     my $drive = print_drive_full($storecfg, $vmid, $device);
2700     my $ret = vm_human_monitor_command($vmid, "drive_add auto $drive");
2701     # If the command succeeds qemu prints: "OK"
2702     if ($ret !~ m/OK/s) {
2703         syslog("err", "adding drive failed: $ret");
2704         return undef;
2705     }
2706     return 1;
2707 }
2708
2709 sub qemu_drivedel {
2710     my($vmid, $deviceid) = @_;
2711
2712     my $ret = vm_human_monitor_command($vmid, "drive_del drive-$deviceid");
2713     $ret =~ s/^\s+//;
2714     if ($ret =~ m/Device \'.*?\' not found/s) {
2715         # NB: device not found errors mean the drive was auto-deleted and we ignore the error
2716     }
2717     elsif ($ret ne "") {
2718       syslog("err", "deleting drive $deviceid failed : $ret");
2719       return undef;
2720     }
2721     return 1;
2722 }
2723
2724 sub qemu_deviceaddverify {
2725     my ($vmid,$deviceid) = @_;
2726
2727     for (my $i = 0; $i <= 5; $i++) {
2728          my $devices_list = vm_devices_list($vmid);
2729          return 1 if defined($devices_list->{$deviceid});
2730          sleep 1;
2731     }
2732     syslog("err", "error on hotplug device $deviceid");
2733     return undef;
2734 }
2735
2736
2737 sub qemu_devicedelverify {
2738     my ($vmid,$deviceid) = @_;
2739
2740     #need to verify the device is correctly remove as device_del is async and empty return is not reliable
2741     for (my $i = 0; $i <= 5; $i++) {
2742          my $devices_list = vm_devices_list($vmid);
2743          return 1 if !defined($devices_list->{$deviceid});
2744          sleep 1;
2745     }
2746     syslog("err", "error on hot-unplugging device $deviceid");
2747     return undef;
2748 }
2749
2750 sub qemu_findorcreatescsihw {
2751     my ($storecfg, $conf, $vmid, $device) = @_;
2752
2753     my $maxdev = ($conf->{scsihw} && $conf->{scsihw} ne 'lsi') ? 256 : 7;
2754     my $controller = int($device->{index} / $maxdev);
2755     my $scsihwid="scsihw$controller";
2756     my $devices_list = vm_devices_list($vmid);
2757
2758     if(!defined($devices_list->{$scsihwid})) {
2759        return undef if !vm_deviceplug($storecfg, $conf, $vmid, $scsihwid);
2760     }
2761     return 1;
2762 }
2763
2764 sub qemu_bridgeadd {
2765     my ($storecfg, $conf, $vmid, $device) = @_;
2766
2767     my $bridges = {};
2768     my $bridgeid = undef;
2769     print_pci_addr($device, $bridges);
2770
2771     while (my ($k, $v) = each %$bridges) {
2772         $bridgeid = $k;
2773     }
2774     return if !$bridgeid || $bridgeid < 1;
2775     my $bridge = "pci.$bridgeid";
2776     my $devices_list = vm_devices_list($vmid);
2777
2778     if(!defined($devices_list->{$bridge})) {
2779         return undef if !vm_deviceplug($storecfg, $conf, $vmid, $bridge);
2780     }
2781     return 1;
2782 }
2783
2784 sub qemu_netdevadd {
2785     my ($vmid, $conf, $device, $deviceid) = @_;
2786
2787     my $netdev = print_netdev_full($vmid, $conf, $device, $deviceid);
2788     my %options =  split(/[=,]/, $netdev);
2789
2790     vm_mon_cmd($vmid, "netdev_add",  %options);
2791     return 1;
2792 }
2793
2794 sub qemu_netdevdel {
2795     my ($vmid, $deviceid) = @_;
2796
2797     vm_mon_cmd($vmid, "netdev_del", id => $deviceid);
2798     return 1;
2799 }
2800
2801 sub qemu_block_set_io_throttle {
2802     my ($vmid, $deviceid, $bps, $bps_rd, $bps_wr, $iops, $iops_rd, $iops_wr) = @_;
2803
2804     return if !check_running($vmid) ;
2805
2806     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($iops_wr));
2807
2808 }
2809
2810 # old code, only used to shutdown old VM after update
2811 sub __read_avail {
2812     my ($fh, $timeout) = @_;
2813
2814     my $sel = new IO::Select;
2815     $sel->add($fh);
2816
2817     my $res = '';
2818     my $buf;
2819
2820     my @ready;
2821     while (scalar (@ready = $sel->can_read($timeout))) {
2822         my $count;
2823         if ($count = $fh->sysread($buf, 8192)) {
2824             if ($buf =~ /^(.*)\(qemu\) $/s) {
2825                 $res .= $1;
2826                 last;
2827             } else {
2828                 $res .= $buf;
2829             }
2830         } else {
2831             if (!defined($count)) {
2832                 die "$!\n";
2833             }
2834             last;
2835         }
2836     }
2837
2838     die "monitor read timeout\n" if !scalar(@ready);
2839
2840     return $res;
2841 }
2842
2843 # old code, only used to shutdown old VM after update
2844 sub vm_monitor_command {
2845     my ($vmid, $cmdstr, $nocheck) = @_;
2846
2847     my $res;
2848
2849     eval {
2850         die "VM $vmid not running\n" if !check_running($vmid, $nocheck);
2851
2852         my $sname = "${var_run_tmpdir}/$vmid.mon";
2853
2854         my $sock = IO::Socket::UNIX->new( Peer => $sname ) ||
2855             die "unable to connect to VM $vmid socket - $!\n";
2856
2857         my $timeout = 3;
2858
2859         # hack: migrate sometime blocks the monitor (when migrate_downtime
2860         # is set)
2861         if ($cmdstr =~ m/^(info\s+migrate|migrate\s)/) {
2862             $timeout = 60*60; # 1 hour
2863         }
2864
2865         # read banner;
2866         my $data = __read_avail($sock, $timeout);
2867
2868         if ($data !~ m/^QEMU\s+(\S+)\s+monitor\s/) {
2869             die "got unexpected qemu monitor banner\n";
2870         }
2871
2872         my $sel = new IO::Select;
2873         $sel->add($sock);
2874
2875         if (!scalar(my @ready = $sel->can_write($timeout))) {
2876             die "monitor write error - timeout";
2877         }
2878
2879         my $fullcmd = "$cmdstr\r";
2880
2881         # syslog('info', "VM $vmid monitor command: $cmdstr");
2882
2883         my $b;
2884         if (!($b = $sock->syswrite($fullcmd)) || ($b != length($fullcmd))) {
2885             die "monitor write error - $!";
2886         }
2887
2888         return if ($cmdstr eq 'q') || ($cmdstr eq 'quit');
2889
2890         $timeout = 20;
2891
2892         if ($cmdstr =~ m/^(info\s+migrate|migrate\s)/) {
2893             $timeout = 60*60; # 1 hour
2894         } elsif ($cmdstr =~ m/^(eject|change)/) {
2895             $timeout = 60; # note: cdrom mount command is slow
2896         }
2897         if ($res = __read_avail($sock, $timeout)) {
2898
2899             my @lines = split("\r?\n", $res);
2900
2901             shift @lines if $lines[0] !~ m/^unknown command/; # skip echo
2902
2903             $res = join("\n", @lines);
2904             $res .= "\n";
2905         }
2906     };
2907
2908     my $err = $@;
2909
2910     if ($err) {
2911         syslog("err", "VM $vmid monitor command failed - $err");
2912         die $err;
2913     }
2914
2915     return $res;
2916 }
2917
2918 sub qemu_block_resize {
2919     my ($vmid, $deviceid, $storecfg, $volid, $size) = @_;
2920
2921     my $running = check_running($vmid);
2922
2923     return if !PVE::Storage::volume_resize($storecfg, $volid, $size, $running);
2924
2925     return if !$running;
2926
2927     vm_mon_cmd($vmid, "block_resize", device => $deviceid, size => int($size));
2928
2929 }
2930
2931 sub qemu_volume_snapshot {
2932     my ($vmid, $deviceid, $storecfg, $volid, $snap) = @_;
2933
2934     my $running = check_running($vmid);
2935
2936     return if !PVE::Storage::volume_snapshot($storecfg, $volid, $snap, $running);
2937
2938     return if !$running;
2939
2940     vm_mon_cmd($vmid, "snapshot-drive", device => $deviceid, name => $snap);
2941
2942 }
2943
2944 sub qemu_volume_snapshot_delete {
2945     my ($vmid, $deviceid, $storecfg, $volid, $snap) = @_;
2946
2947     my $running = check_running($vmid);
2948
2949     return if !PVE::Storage::volume_snapshot_delete($storecfg, $volid, $snap, $running);
2950
2951     return if !$running;
2952
2953     vm_mon_cmd($vmid, "delete-drive-snapshot", device => $deviceid, name => $snap);
2954 }
2955
2956 sub qga_freezefs {
2957     my ($vmid) = @_;
2958
2959     #need to impplement call to qemu-ga
2960 }
2961
2962 sub qga_unfreezefs {
2963     my ($vmid) = @_;
2964
2965     #need to impplement call to qemu-ga
2966 }
2967
2968 sub vm_start {
2969     my ($storecfg, $vmid, $statefile, $skiplock, $migratedfrom, $paused, $forcemachine) = @_;
2970
2971     lock_config($vmid, sub {
2972         my $conf = load_config($vmid, $migratedfrom);
2973
2974         die "you can't start a vm if it's a template\n" if is_template($conf);
2975
2976         check_lock($conf) if !$skiplock;
2977
2978         die "VM $vmid already running\n" if check_running($vmid, undef, $migratedfrom);
2979
2980         my $defaults = load_defaults();
2981
2982         # set environment variable useful inside network script
2983         $ENV{PVE_MIGRATED_FROM} = $migratedfrom if $migratedfrom;
2984
2985         my ($cmd, $vollist) = config_to_command($storecfg, $vmid, $conf, $defaults, $forcemachine);
2986
2987         my $migrate_port = 0;
2988
2989         if ($statefile) {
2990             if ($statefile eq 'tcp') {
2991                 $migrate_port = PVE::Tools::next_migrate_port();
2992                 my $migrate_uri = "tcp:localhost:${migrate_port}";
2993                 push @$cmd, '-incoming', $migrate_uri;
2994                 push @$cmd, '-S';
2995             } else {
2996                 push @$cmd, '-loadstate', $statefile;
2997             }
2998         } elsif ($paused) {
2999             push @$cmd, '-S';
3000         }
3001
3002         # host pci devices
3003         for (my $i = 0; $i < $MAX_HOSTPCI_DEVICES; $i++)  {
3004           my $d = parse_hostpci($conf->{"hostpci$i"});
3005           next if !$d;
3006           my $info = pci_device_info("0000:$d->{pciid}");
3007           die "IOMMU not present\n" if !check_iommu_support();
3008           die "no pci device info for device '$d->{pciid}'\n" if !$info;
3009           die "can't unbind pci device '$d->{pciid}'\n" if !pci_dev_bind_to_stub($info);
3010           die "can't reset pci device '$d->{pciid}'\n" if !pci_dev_reset($info);
3011         }
3012
3013         PVE::Storage::activate_volumes($storecfg, $vollist);
3014
3015         eval  { run_command($cmd, timeout => $statefile ? undef : 30,
3016                     umask => 0077); };
3017         my $err = $@;
3018         die "start failed: $err" if $err;
3019
3020         print "migration listens on port $migrate_port\n" if $migrate_port;
3021
3022         if ($statefile && $statefile ne 'tcp')  {
3023             eval { vm_mon_cmd_nocheck($vmid, "cont"); };
3024             warn $@ if $@;
3025         }
3026
3027         if($migratedfrom) {
3028             my $capabilities = {};
3029             $capabilities->{capability} =  "xbzrle";
3030             $capabilities->{state} = JSON::true;
3031             eval { vm_mon_cmd_nocheck($vmid, "migrate-set-capabilities", capabilities => [$capabilities]); };
3032         }
3033         else{
3034
3035             if (!$statefile && (!defined($conf->{balloon}) || $conf->{balloon})) {
3036                 vm_mon_cmd_nocheck($vmid, "balloon", value => $conf->{balloon}*1024*1024)
3037                     if $conf->{balloon};
3038                 vm_mon_cmd_nocheck($vmid, 'qom-set',
3039                             path => "machine/peripheral/balloon0",
3040                             property => "guest-stats-polling-interval",
3041                             value => 2);
3042             }
3043         }
3044     });
3045 }
3046
3047 sub vm_mon_cmd {
3048     my ($vmid, $execute, %params) = @_;
3049
3050     my $cmd = { execute => $execute, arguments => \%params };
3051     vm_qmp_command($vmid, $cmd);
3052 }
3053
3054 sub vm_mon_cmd_nocheck {
3055     my ($vmid, $execute, %params) = @_;
3056
3057     my $cmd = { execute => $execute, arguments => \%params };
3058     vm_qmp_command($vmid, $cmd, 1);
3059 }
3060
3061 sub vm_qmp_command {
3062     my ($vmid, $cmd, $nocheck) = @_;
3063
3064     my $res;
3065
3066     my $timeout;
3067     if ($cmd->{arguments} && $cmd->{arguments}->{timeout}) {
3068         $timeout = $cmd->{arguments}->{timeout};
3069         delete $cmd->{arguments}->{timeout};
3070     }
3071
3072     eval {
3073         die "VM $vmid not running\n" if !check_running($vmid, $nocheck);
3074         my $sname = qmp_socket($vmid);
3075         if (-e $sname) {
3076             my $qmpclient = PVE::QMPClient->new();
3077
3078             $res = $qmpclient->cmd($vmid, $cmd, $timeout);
3079         } elsif (-e "${var_run_tmpdir}/$vmid.mon") {
3080             die "can't execute complex command on old monitor - stop/start your vm to fix the problem\n"
3081                 if scalar(%{$cmd->{arguments}});
3082             vm_monitor_command($vmid, $cmd->{execute}, $nocheck);
3083         } else {
3084             die "unable to open monitor socket\n";
3085         }
3086     };
3087     if (my $err = $@) {
3088         syslog("err", "VM $vmid qmp command failed - $err");
3089         die $err;
3090     }
3091
3092     return $res;
3093 }
3094
3095 sub vm_human_monitor_command {
3096     my ($vmid, $cmdline) = @_;
3097
3098     my $res;
3099
3100     my $cmd = {
3101         execute => 'human-monitor-command',
3102         arguments => { 'command-line' => $cmdline},
3103     };
3104
3105     return vm_qmp_command($vmid, $cmd);
3106 }
3107
3108 sub vm_commandline {
3109     my ($storecfg, $vmid) = @_;
3110
3111     my $conf = load_config($vmid);
3112
3113     my $defaults = load_defaults();
3114
3115     my $cmd = config_to_command($storecfg, $vmid, $conf, $defaults);
3116
3117     return join(' ', @$cmd);
3118 }
3119
3120 sub vm_reset {
3121     my ($vmid, $skiplock) = @_;
3122
3123     lock_config($vmid, sub {
3124
3125         my $conf = load_config($vmid);
3126
3127         check_lock($conf) if !$skiplock;
3128
3129         vm_mon_cmd($vmid, "system_reset");
3130     });
3131 }
3132
3133 sub get_vm_volumes {
3134     my ($conf) = @_;
3135
3136     my $vollist = [];
3137     foreach_volid($conf, sub {
3138         my ($volid, $is_cdrom) = @_;
3139
3140         return if $volid =~ m|^/|;
3141
3142         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
3143         return if !$sid;
3144
3145         push @$vollist, $volid;
3146     });
3147
3148     return $vollist;
3149 }
3150
3151 sub vm_stop_cleanup {
3152     my ($storecfg, $vmid, $conf, $keepActive) = @_;
3153
3154     eval {
3155         fairsched_rmnod($vmid); # try to destroy group
3156
3157         if (!$keepActive) {
3158             my $vollist = get_vm_volumes($conf);
3159             PVE::Storage::deactivate_volumes($storecfg, $vollist);
3160         }
3161
3162         foreach my $ext (qw(mon qmp pid vnc qga)) {
3163             unlink "/var/run/qemu-server/${vmid}.$ext";
3164         }
3165     };
3166     warn $@ if $@; # avoid errors - just warn
3167 }
3168
3169 # Note: use $nockeck to skip tests if VM configuration file exists.
3170 # We need that when migration VMs to other nodes (files already moved)
3171 # Note: we set $keepActive in vzdump stop mode - volumes need to stay active
3172 sub vm_stop {
3173     my ($storecfg, $vmid, $skiplock, $nocheck, $timeout, $shutdown, $force, $keepActive, $migratedfrom) = @_;
3174
3175     $force = 1 if !defined($force) && !$shutdown;
3176
3177     if ($migratedfrom){
3178         my $pid = check_running($vmid, $nocheck, $migratedfrom);
3179         kill 15, $pid if $pid;
3180         my $conf = load_config($vmid, $migratedfrom);
3181         vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive);
3182         return;
3183     }
3184
3185     lock_config($vmid, sub {
3186
3187         my $pid = check_running($vmid, $nocheck);
3188         return if !$pid;
3189
3190         my $conf;
3191         if (!$nocheck) {
3192             $conf = load_config($vmid);
3193             check_lock($conf) if !$skiplock;
3194             if (!defined($timeout) && $shutdown && $conf->{startup}) {
3195                 my $opts = parse_startup($conf->{startup});
3196                 $timeout = $opts->{down} if $opts->{down};
3197             }
3198         }
3199
3200         $timeout = 60 if !defined($timeout);
3201
3202         eval {
3203             if ($shutdown) {
3204                 $nocheck ? vm_mon_cmd_nocheck($vmid, "system_powerdown") : vm_mon_cmd($vmid, "system_powerdown");
3205
3206             } else {
3207                 $nocheck ? vm_mon_cmd_nocheck($vmid, "quit") : vm_mon_cmd($vmid, "quit");
3208             }
3209         };
3210         my $err = $@;
3211
3212         if (!$err) {
3213             my $count = 0;
3214             while (($count < $timeout) && check_running($vmid, $nocheck)) {
3215                 $count++;
3216                 sleep 1;
3217             }
3218
3219             if ($count >= $timeout) {
3220                 if ($force) {
3221                     warn "VM still running - terminating now with SIGTERM\n";
3222                     kill 15, $pid;
3223                 } else {
3224                     die "VM quit/powerdown failed - got timeout\n";
3225                 }
3226             } else {
3227                 vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive) if $conf;
3228                 return;
3229             }
3230         } else {
3231             if ($force) {
3232                 warn "VM quit/powerdown failed - terminating now with SIGTERM\n";
3233                 kill 15, $pid;
3234             } else {
3235                 die "VM quit/powerdown failed\n";
3236             }
3237         }
3238
3239         # wait again
3240         $timeout = 10;
3241
3242         my $count = 0;
3243         while (($count < $timeout) && check_running($vmid, $nocheck)) {
3244             $count++;
3245             sleep 1;
3246         }
3247
3248         if ($count >= $timeout) {
3249             warn "VM still running - terminating now with SIGKILL\n";
3250             kill 9, $pid;
3251             sleep 1;
3252         }
3253
3254         vm_stop_cleanup($storecfg, $vmid, $conf, $keepActive) if $conf;
3255    });
3256 }
3257
3258 sub vm_suspend {
3259     my ($vmid, $skiplock) = @_;
3260
3261     lock_config($vmid, sub {
3262
3263         my $conf = load_config($vmid);
3264
3265         check_lock($conf) if !($skiplock || ($conf->{lock} && $conf->{lock} eq 'backup'));
3266
3267         vm_mon_cmd($vmid, "stop");
3268     });
3269 }
3270
3271 sub vm_resume {
3272     my ($vmid, $skiplock) = @_;
3273
3274     lock_config($vmid, sub {
3275
3276         my $conf = load_config($vmid);
3277
3278         check_lock($conf) if !($skiplock || ($conf->{lock} && $conf->{lock} eq 'backup'));
3279
3280         vm_mon_cmd($vmid, "cont");
3281     });
3282 }
3283
3284 sub vm_sendkey {
3285     my ($vmid, $skiplock, $key) = @_;
3286
3287     lock_config($vmid, sub {
3288
3289         my $conf = load_config($vmid);
3290
3291         # there is no qmp command, so we use the human monitor command
3292         vm_human_monitor_command($vmid, "sendkey $key");
3293     });
3294 }
3295
3296 sub vm_destroy {
3297     my ($storecfg, $vmid, $skiplock) = @_;
3298
3299     lock_config($vmid, sub {
3300
3301         my $conf = load_config($vmid);
3302
3303         check_lock($conf) if !$skiplock;
3304
3305         if (!check_running($vmid)) {
3306             fairsched_rmnod($vmid); # try to destroy group
3307             destroy_vm($storecfg, $vmid);
3308         } else {
3309             die "VM $vmid is running - destroy failed\n";
3310         }
3311     });
3312 }
3313
3314 # pci helpers
3315
3316 sub file_write {
3317     my ($filename, $buf) = @_;
3318
3319     my $fh = IO::File->new($filename, "w");
3320     return undef if !$fh;
3321
3322     my $res = print $fh $buf;
3323
3324     $fh->close();
3325
3326     return $res;
3327 }
3328
3329 sub pci_device_info {
3330     my ($name) = @_;
3331
3332     my $res;
3333
3334     return undef if $name !~ m/^([a-f0-9]{4}):([a-f0-9]{2}):([a-f0-9]{2})\.([a-f0-9])$/;
3335     my ($domain, $bus, $slot, $func) = ($1, $2, $3, $4);
3336
3337     my $irq = file_read_firstline("$pcisysfs/devices/$name/irq");
3338     return undef if !defined($irq) || $irq !~ m/^\d+$/;
3339
3340     my $vendor = file_read_firstline("$pcisysfs/devices/$name/vendor");
3341     return undef if !defined($vendor) || $vendor !~ s/^0x//;
3342
3343     my $product = file_read_firstline("$pcisysfs/devices/$name/device");
3344     return undef if !defined($product) || $product !~ s/^0x//;
3345
3346     $res = {
3347         name => $name,
3348         vendor => $vendor,
3349         product => $product,
3350         domain => $domain,
3351         bus => $bus,
3352         slot => $slot,
3353         func => $func,
3354         irq => $irq,
3355         has_fl_reset => -f "$pcisysfs/devices/$name/reset" || 0,
3356     };
3357
3358     return $res;
3359 }
3360
3361 sub pci_dev_reset {
3362     my ($dev) = @_;
3363
3364     my $name = $dev->{name};
3365
3366     my $fn = "$pcisysfs/devices/$name/reset";
3367
3368     return file_write($fn, "1");
3369 }
3370
3371 sub pci_dev_bind_to_stub {
3372     my ($dev) = @_;
3373
3374     my $name = $dev->{name};
3375
3376     my $testdir = "$pcisysfs/drivers/pci-stub/$name";
3377     return 1 if -d $testdir;
3378
3379     my $data = "$dev->{vendor} $dev->{product}";
3380     return undef if !file_write("$pcisysfs/drivers/pci-stub/new_id", $data);
3381
3382     my $fn = "$pcisysfs/devices/$name/driver/unbind";
3383     if (!file_write($fn, $name)) {
3384         return undef if -f $fn;
3385     }
3386
3387     $fn = "$pcisysfs/drivers/pci-stub/bind";
3388     if (! -d $testdir) {
3389         return undef if !file_write($fn, $name);
3390     }
3391
3392     return -d $testdir;
3393 }
3394
3395 sub print_pci_addr {
3396     my ($id, $bridges) = @_;
3397
3398     my $res = '';
3399     my $devices = {
3400         piix3 => { bus => 0, addr => 1 },
3401         #addr2 : first videocard
3402         balloon0 => { bus => 0, addr => 3 },
3403         watchdog => { bus => 0, addr => 4 },
3404         scsihw0 => { bus => 0, addr => 5 },
3405         scsihw1 => { bus => 0, addr => 6 },
3406         ahci0 => { bus => 0, addr => 7 },
3407         qga0 => { bus => 0, addr => 8 },
3408         virtio0 => { bus => 0, addr => 10 },
3409         virtio1 => { bus => 0, addr => 11 },
3410         virtio2 => { bus => 0, addr => 12 },
3411         virtio3 => { bus => 0, addr => 13 },
3412         virtio4 => { bus => 0, addr => 14 },
3413         virtio5 => { bus => 0, addr => 15 },
3414         hostpci0 => { bus => 0, addr => 16 },
3415         hostpci1 => { bus => 0, addr => 17 },
3416         net0 => { bus => 0, addr => 18 },
3417         net1 => { bus => 0, addr => 19 },
3418         net2 => { bus => 0, addr => 20 },
3419         net3 => { bus => 0, addr => 21 },
3420         net4 => { bus => 0, addr => 22 },
3421         net5 => { bus => 0, addr => 23 },
3422         #addr29 : usb-host (pve-usb.cfg)
3423         'pci.1' => { bus => 0, addr => 30 },
3424         'pci.2' => { bus => 0, addr => 31 },
3425         'net6' => { bus => 1, addr => 1 },
3426         'net7' => { bus => 1, addr => 2 },
3427         'net8' => { bus => 1, addr => 3 },
3428         'net9' => { bus => 1, addr => 4 },
3429         'net10' => { bus => 1, addr => 5 },
3430         'net11' => { bus => 1, addr => 6 },
3431         'net12' => { bus => 1, addr => 7 },
3432         'net13' => { bus => 1, addr => 8 },
3433         'net14' => { bus => 1, addr => 9 },
3434         'net15' => { bus => 1, addr => 10 },
3435         'net16' => { bus => 1, addr => 11 },
3436         'net17' => { bus => 1, addr => 12 },
3437         'net18' => { bus => 1, addr => 13 },
3438         'net19' => { bus => 1, addr => 14 },
3439         'net20' => { bus => 1, addr => 15 },
3440         'net21' => { bus => 1, addr => 16 },
3441         'net22' => { bus => 1, addr => 17 },
3442         'net23' => { bus => 1, addr => 18 },
3443         'net24' => { bus => 1, addr => 19 },
3444         'net25' => { bus => 1, addr => 20 },
3445         'net26' => { bus => 1, addr => 21 },
3446         'net27' => { bus => 1, addr => 22 },
3447         'net28' => { bus => 1, addr => 23 },
3448         'net29' => { bus => 1, addr => 24 },
3449         'net30' => { bus => 1, addr => 25 },
3450         'net31' => { bus => 1, addr => 26 },
3451         'virtio6' => { bus => 2, addr => 1 },
3452         'virtio7' => { bus => 2, addr => 2 },
3453         'virtio8' => { bus => 2, addr => 3 },
3454         'virtio9' => { bus => 2, addr => 4 },
3455         'virtio10' => { bus => 2, addr => 5 },
3456         'virtio11' => { bus => 2, addr => 6 },
3457         'virtio12' => { bus => 2, addr => 7 },
3458         'virtio13' => { bus => 2, addr => 8 },
3459         'virtio14' => { bus => 2, addr => 9 },
3460         'virtio15' => { bus => 2, addr => 10 },
3461     };
3462
3463     if (defined($devices->{$id}->{bus}) && defined($devices->{$id}->{addr})) {
3464            my $addr = sprintf("0x%x", $devices->{$id}->{addr});
3465            my $bus = $devices->{$id}->{bus};
3466            $res = ",bus=pci.$bus,addr=$addr";
3467            $bridges->{$bus} = 1 if $bridges;
3468     }
3469     return $res;
3470
3471 }
3472
3473 # vzdump restore implementaion
3474
3475 sub tar_archive_read_firstfile {
3476     my $archive = shift;
3477
3478     die "ERROR: file '$archive' does not exist\n" if ! -f $archive;
3479
3480     # try to detect archive type first
3481     my $pid = open (TMP, "tar tf '$archive'|") ||
3482         die "unable to open file '$archive'\n";
3483     my $firstfile = <TMP>;
3484     kill 15, $pid;
3485     close TMP;
3486
3487     die "ERROR: archive contaions no data\n" if !$firstfile;
3488     chomp $firstfile;
3489
3490     return $firstfile;
3491 }
3492
3493 sub tar_restore_cleanup {
3494     my ($storecfg, $statfile) = @_;
3495
3496     print STDERR "starting cleanup\n";
3497
3498     if (my $fd = IO::File->new($statfile, "r")) {
3499         while (defined(my $line = <$fd>)) {
3500             if ($line =~ m/vzdump:([^\s:]*):(\S+)$/) {
3501                 my $volid = $2;
3502                 eval {
3503                     if ($volid =~ m|^/|) {
3504                         unlink $volid || die 'unlink failed\n';
3505                     } else {
3506                         PVE::Storage::vdisk_free($storecfg, $volid);
3507                     }
3508                     print STDERR "temporary volume '$volid' sucessfuly removed\n";
3509                 };
3510                 print STDERR "unable to cleanup '$volid' - $@" if $@;
3511             } else {
3512                 print STDERR "unable to parse line in statfile - $line";
3513             }
3514         }
3515         $fd->close();
3516     }
3517 }
3518
3519 sub restore_archive {
3520     my ($archive, $vmid, $user, $opts) = @_;
3521
3522     my $format = $opts->{format};
3523     my $comp;
3524
3525     if ($archive =~ m/\.tgz$/ || $archive =~ m/\.tar\.gz$/) {
3526         $format = 'tar' if !$format;
3527         $comp = 'gzip';
3528     } elsif ($archive =~ m/\.tar$/) {
3529         $format = 'tar' if !$format;
3530     } elsif ($archive =~ m/.tar.lzo$/) {
3531         $format = 'tar' if !$format;
3532         $comp = 'lzop';
3533     } elsif ($archive =~ m/\.vma$/) {
3534         $format = 'vma' if !$format;
3535     } elsif ($archive =~ m/\.vma\.gz$/) {
3536         $format = 'vma' if !$format;
3537         $comp = 'gzip';
3538     } elsif ($archive =~ m/\.vma\.lzo$/) {
3539         $format = 'vma' if !$format;
3540         $comp = 'lzop';
3541     } else {
3542         $format = 'vma' if !$format; # default
3543     }
3544
3545     # try to detect archive format
3546     if ($format eq 'tar') {
3547         return restore_tar_archive($archive, $vmid, $user, $opts);
3548     } else {
3549         return restore_vma_archive($archive, $vmid, $user, $opts, $comp);
3550     }
3551 }
3552
3553 sub restore_update_config_line {
3554     my ($outfd, $cookie, $vmid, $map, $line, $unique) = @_;
3555
3556     return if $line =~ m/^\#qmdump\#/;
3557     return if $line =~ m/^\#vzdump\#/;
3558     return if $line =~ m/^lock:/;
3559     return if $line =~ m/^unused\d+:/;
3560     return if $line =~ m/^parent:/;
3561     return if $line =~ m/^template:/; # restored VM is never a template
3562
3563     if (($line =~ m/^(vlan(\d+)):\s*(\S+)\s*$/)) {
3564         # try to convert old 1.X settings
3565         my ($id, $ind, $ethcfg) = ($1, $2, $3);
3566         foreach my $devconfig (PVE::Tools::split_list($ethcfg)) {
3567             my ($model, $macaddr) = split(/\=/, $devconfig);
3568             $macaddr = PVE::Tools::random_ether_addr() if !$macaddr || $unique;
3569             my $net = {
3570                 model => $model,
3571                 bridge => "vmbr$ind",
3572                 macaddr => $macaddr,
3573             };
3574             my $netstr = print_net($net);
3575
3576             print $outfd "net$cookie->{netcount}: $netstr\n";
3577             $cookie->{netcount}++;
3578         }
3579     } elsif (($line =~ m/^(net\d+):\s*(\S+)\s*$/) && $unique) {
3580         my ($id, $netstr) = ($1, $2);
3581         my $net = parse_net($netstr);
3582         $net->{macaddr} = PVE::Tools::random_ether_addr() if $net->{macaddr};
3583         $netstr = print_net($net);
3584         print $outfd "$id: $netstr\n";
3585     } elsif ($line =~ m/^((ide|scsi|virtio|sata)\d+):\s*(\S+)\s*$/) {
3586         my $virtdev = $1;
3587         my $value = $3;
3588         if ($line =~ m/backup=no/) {
3589             print $outfd "#$line";
3590         } elsif ($virtdev && $map->{$virtdev}) {
3591             my $di = parse_drive($virtdev, $value);
3592             delete $di->{format}; # format can change on restore
3593             $di->{file} = $map->{$virtdev};
3594             $value = print_drive($vmid, $di);
3595             print $outfd "$virtdev: $value\n";
3596         } else {
3597             print $outfd $line;
3598         }
3599     } else {
3600         print $outfd $line;
3601     }
3602 }
3603
3604 sub scan_volids {
3605     my ($cfg, $vmid) = @_;
3606
3607     my $info = PVE::Storage::vdisk_list($cfg, undef, $vmid);
3608
3609     my $volid_hash = {};
3610     foreach my $storeid (keys %$info) {
3611         foreach my $item (@{$info->{$storeid}}) {
3612             next if !($item->{volid} && $item->{size});
3613             $item->{path} = PVE::Storage::path($cfg, $item->{volid});
3614             $volid_hash->{$item->{volid}} = $item;
3615         }
3616     }
3617
3618     return $volid_hash;
3619 }
3620
3621 sub get_used_paths {
3622     my ($vmid, $storecfg, $conf, $scan_snapshots, $skip_drive) = @_;
3623
3624     my $used_path = {};
3625
3626     my $scan_config = sub {
3627         my ($cref, $snapname) = @_;
3628
3629         foreach my $key (keys %$cref) {
3630             my $value = $cref->{$key};
3631             if (valid_drivename($key)) {
3632                 next if $skip_drive && $key eq $skip_drive;
3633                 my $drive = parse_drive($key, $value);
3634                 next if !$drive || !$drive->{file} || drive_is_cdrom($drive);
3635                 if ($drive->{file} =~ m!^/!) {
3636                     $used_path->{$drive->{file}}++; # = 1;
3637                 } else {
3638                     my ($storeid, $volname) = PVE::Storage::parse_volume_id($drive->{file}, 1);
3639                     next if !$storeid;
3640                     my $scfg = PVE::Storage::storage_config($storecfg, $storeid, 1);
3641                     next if !$scfg;
3642                     my $path = PVE::Storage::path($storecfg, $drive->{file}, $snapname);
3643                     $used_path->{$path}++; # = 1;
3644                 }
3645             }
3646         }
3647     };
3648
3649     &$scan_config($conf);
3650
3651     undef $skip_drive;
3652
3653     if ($scan_snapshots) {
3654         foreach my $snapname (keys %{$conf->{snapshots}}) {
3655             &$scan_config($conf->{snapshots}->{$snapname}, $snapname);
3656         }
3657     }
3658
3659     return $used_path;
3660 }
3661
3662 sub update_disksize {
3663     my ($vmid, $conf, $volid_hash) = @_;
3664
3665     my $changes;
3666
3667     my $used = {};
3668
3669     # Note: it is allowed to define multiple storages with same path (alias), so
3670     # we need to check both 'volid' and real 'path' (two different volid can point
3671     # to the same path).
3672
3673     my $usedpath = {};
3674
3675     # update size info
3676     foreach my $opt (keys %$conf) {
3677         if (valid_drivename($opt)) {
3678             my $drive = parse_drive($opt, $conf->{$opt});
3679             my $volid = $drive->{file};
3680             next if !$volid;
3681
3682             $used->{$volid} = 1;
3683             if ($volid_hash->{$volid} &&
3684                 (my $path = $volid_hash->{$volid}->{path})) {
3685                 $usedpath->{$path} = 1;
3686             }
3687
3688             next if drive_is_cdrom($drive);
3689             next if !$volid_hash->{$volid};
3690
3691             $drive->{size} = $volid_hash->{$volid}->{size};
3692             my $new = print_drive($vmid, $drive);
3693             if ($new ne $conf->{$opt}) {
3694                 $changes = 1;
3695                 $conf->{$opt} = $new;
3696             }
3697         }
3698     }
3699
3700     # remove 'unusedX' entry if volume is used
3701     foreach my $opt (keys %$conf) {
3702         next if $opt !~ m/^unused\d+$/;
3703         my $volid = $conf->{$opt};
3704         my $path = $volid_hash->{$volid}->{path} if $volid_hash->{$volid};
3705         if ($used->{$volid} || ($path && $usedpath->{$path})) {
3706             $changes = 1;
3707             delete $conf->{$opt};
3708         }
3709     }
3710
3711     foreach my $volid (sort keys %$volid_hash) {
3712         next if $volid =~ m/vm-$vmid-state-/;
3713         next if $used->{$volid};
3714         my $path = $volid_hash->{$volid}->{path};
3715         next if !$path; # just to be sure
3716         next if $usedpath->{$path};
3717         $changes = 1;
3718         add_unused_volume($conf, $volid);
3719         $usedpath->{$path} = 1; # avoid to add more than once (aliases)
3720     }
3721
3722     return $changes;
3723 }
3724
3725 sub rescan {
3726     my ($vmid, $nolock) = @_;
3727
3728     my $cfg = PVE::Cluster::cfs_read_file("storage.cfg");
3729
3730     my $volid_hash = scan_volids($cfg, $vmid);
3731
3732     my $updatefn =  sub {
3733         my ($vmid) = @_;
3734
3735         my $conf = load_config($vmid);
3736
3737         check_lock($conf);
3738
3739         my $vm_volids = {};
3740         foreach my $volid (keys %$volid_hash) {
3741             my $info = $volid_hash->{$volid};
3742             $vm_volids->{$volid} = $info if $info->{vmid} && $info->{vmid} == $vmid;
3743         }
3744
3745         my $changes = update_disksize($vmid, $conf, $vm_volids);
3746
3747         update_config_nolock($vmid, $conf, 1) if $changes;
3748     };
3749
3750     if (defined($vmid)) {
3751         if ($nolock) {
3752             &$updatefn($vmid);
3753         } else {
3754             lock_config($vmid, $updatefn, $vmid);
3755         }
3756     } else {
3757         my $vmlist = config_list();
3758         foreach my $vmid (keys %$vmlist) {
3759             if ($nolock) {
3760                 &$updatefn($vmid);
3761             } else {
3762                 lock_config($vmid, $updatefn, $vmid);
3763             }
3764         }
3765     }
3766 }
3767
3768 sub restore_vma_archive {
3769     my ($archive, $vmid, $user, $opts, $comp) = @_;
3770
3771     my $input = $archive eq '-' ? "<&STDIN" : undef;
3772     my $readfrom = $archive;
3773
3774     my $uncomp = '';
3775     if ($comp) {
3776         $readfrom = '-';
3777         my $qarchive = PVE::Tools::shellquote($archive);
3778         if ($comp eq 'gzip') {
3779             $uncomp = "zcat $qarchive|";
3780         } elsif ($comp eq 'lzop') {
3781             $uncomp = "lzop -d -c $qarchive|";
3782         } else {
3783             die "unknown compression method '$comp'\n";
3784         }
3785
3786     }
3787
3788     my $tmpdir = "/var/tmp/vzdumptmp$$";
3789     rmtree $tmpdir;
3790
3791     # disable interrupts (always do cleanups)
3792     local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = sub {
3793         warn "got interrupt - ignored\n";
3794     };
3795
3796     my $mapfifo = "/var/tmp/vzdumptmp$$.fifo";
3797     POSIX::mkfifo($mapfifo, 0600);
3798     my $fifofh;
3799
3800     my $openfifo = sub {
3801         open($fifofh, '>', $mapfifo) || die $!;
3802     };
3803
3804     my $cmd = "${uncomp}vma extract -v -r $mapfifo $readfrom $tmpdir";
3805
3806     my $oldtimeout;
3807     my $timeout = 5;
3808
3809     my $devinfo = {};
3810
3811     my $rpcenv = PVE::RPCEnvironment::get();
3812
3813     my $conffile = config_file($vmid);
3814     my $tmpfn = "$conffile.$$.tmp";
3815
3816     # Note: $oldconf is undef if VM does not exists
3817     my $oldconf = PVE::Cluster::cfs_read_file(cfs_config_path($vmid));
3818
3819     my $print_devmap = sub {
3820         my $virtdev_hash = {};
3821
3822         my $cfgfn = "$tmpdir/qemu-server.conf";
3823
3824         # we can read the config - that is already extracted
3825         my $fh = IO::File->new($cfgfn, "r") ||
3826             "unable to read qemu-server.conf - $!\n";
3827
3828         while (defined(my $line = <$fh>)) {
3829             if ($line =~ m/^\#qmdump\#map:(\S+):(\S+):(\S*):(\S*):$/) {
3830                 my ($virtdev, $devname, $storeid, $format) = ($1, $2, $3, $4);
3831                 die "archive does not contain data for drive '$virtdev'\n"
3832                     if !$devinfo->{$devname};
3833                 if (defined($opts->{storage})) {
3834                     $storeid = $opts->{storage} || 'local';
3835                 } elsif (!$storeid) {
3836                     $storeid = 'local';
3837                 }
3838                 $format = 'raw' if !$format;
3839                 $devinfo->{$devname}->{devname} = $devname;
3840                 $devinfo->{$devname}->{virtdev} = $virtdev;
3841                 $devinfo->{$devname}->{format} = $format;
3842                 $devinfo->{$devname}->{storeid} = $storeid;
3843
3844                 # check permission on storage
3845                 my $pool = $opts->{pool}; # todo: do we need that?
3846                 if ($user ne 'root@pam') {
3847                     $rpcenv->check($user, "/storage/$storeid", ['Datastore.AllocateSpace']);
3848                 }
3849
3850                 $virtdev_hash->{$virtdev} = $devinfo->{$devname};
3851             }
3852         }
3853
3854         foreach my $devname (keys %$devinfo) {
3855             die "found no device mapping information for device '$devname'\n"
3856                 if !$devinfo->{$devname}->{virtdev};
3857         }
3858
3859         my $cfg = cfs_read_file('storage.cfg');
3860
3861         # create empty/temp config
3862         if ($oldconf) {
3863             PVE::Tools::file_set_contents($conffile, "memory: 128\n");
3864             foreach_drive($oldconf, sub {
3865                 my ($ds, $drive) = @_;
3866
3867                 return if drive_is_cdrom($drive);
3868
3869                 my $volid = $drive->{file};
3870
3871                 return if !$volid || $volid =~ m|^/|;
3872
3873                 my ($path, $owner) = PVE::Storage::path($cfg, $volid);
3874                 return if !$path || !$owner || ($owner != $vmid);
3875
3876                 # Note: only delete disk we want to restore
3877                 # other volumes will become unused
3878                 if ($virtdev_hash->{$ds}) {
3879                     PVE::Storage::vdisk_free($cfg, $volid);
3880                 }
3881             });
3882         }
3883
3884         my $map = {};
3885         foreach my $virtdev (sort keys %$virtdev_hash) {
3886             my $d = $virtdev_hash->{$virtdev};
3887             my $alloc_size = int(($d->{size} + 1024 - 1)/1024);
3888             my $scfg = PVE::Storage::storage_config($cfg, $d->{storeid});
3889
3890             # test if requested format is supported
3891             my ($defFormat, $validFormats) = PVE::Storage::storage_default_format($cfg, $d->{storeid});
3892             my $supported = grep { $_ eq $d->{format} } @$validFormats;
3893             $d->{format} = $defFormat if !$supported;
3894
3895             my $volid = PVE::Storage::vdisk_alloc($cfg, $d->{storeid}, $vmid,
3896                                                   $d->{format}, undef, $alloc_size);
3897             print STDERR "new volume ID is '$volid'\n";
3898             $d->{volid} = $volid;
3899             my $path = PVE::Storage::path($cfg, $volid);
3900
3901             my $write_zeros = 1;
3902             # fixme: what other storages types initialize volumes with zero?
3903             if ($scfg->{type} eq 'dir' || $scfg->{type} eq 'nfs' ||
3904                 $scfg->{type} eq 'sheepdog' || $scfg->{type} eq 'rbd') {
3905                 $write_zeros = 0;
3906             }
3907
3908             print $fifofh "${write_zeros}:$d->{devname}=$path\n";
3909
3910             print "map '$d->{devname}' to '$path' (write zeros = ${write_zeros})\n";
3911             $map->{$virtdev} = $volid;
3912         }
3913
3914         $fh->seek(0, 0) || die "seek failed - $!\n";
3915
3916         my $outfd = new IO::File ($tmpfn, "w") ||
3917             die "unable to write config for VM $vmid\n";
3918
3919         my $cookie = { netcount => 0 };
3920         while (defined(my $line = <$fh>)) {
3921             restore_update_config_line($outfd, $cookie, $vmid, $map, $line, $opts->{unique});
3922         }
3923
3924         $fh->close();
3925         $outfd->close();
3926     };
3927
3928     eval {
3929         # enable interrupts
3930         local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = $SIG{PIPE} = sub {
3931             die "interrupted by signal\n";
3932         };
3933         local $SIG{ALRM} = sub { die "got timeout\n"; };
3934
3935         $oldtimeout = alarm($timeout);
3936
3937         my $parser = sub {
3938             my $line = shift;
3939
3940             print "$line\n";
3941
3942             if ($line =~ m/^DEV:\sdev_id=(\d+)\ssize:\s(\d+)\sdevname:\s(\S+)$/) {
3943                 my ($dev_id, $size, $devname) = ($1, $2, $3);
3944                 $devinfo->{$devname} = { size => $size, dev_id => $dev_id };
3945             } elsif ($line =~ m/^CTIME: /) {
3946                 &$print_devmap();
3947                 print $fifofh "done\n";
3948                 my $tmp = $oldtimeout || 0;
3949                 $oldtimeout = undef;
3950                 alarm($tmp);
3951                 close($fifofh);
3952             }
3953         };
3954
3955         print "restore vma archive: $cmd\n";
3956         run_command($cmd, input => $input, outfunc => $parser, afterfork => $openfifo);
3957     };
3958     my $err = $@;
3959
3960     alarm($oldtimeout) if $oldtimeout;
3961
3962     unlink $mapfifo;
3963
3964     if ($err) {
3965         rmtree $tmpdir;
3966         unlink $tmpfn;
3967
3968         my $cfg = cfs_read_file('storage.cfg');
3969         foreach my $devname (keys %$devinfo) {
3970             my $volid = $devinfo->{$devname}->{volid};
3971             next if !$volid;
3972             eval {
3973                 if ($volid =~ m|^/|) {
3974                     unlink $volid || die 'unlink failed\n';
3975                 } else {
3976                     PVE::Storage::vdisk_free($cfg, $volid);
3977                 }
3978                 print STDERR "temporary volume '$volid' sucessfuly removed\n";
3979             };
3980             print STDERR "unable to cleanup '$volid' - $@" if $@;
3981         }
3982         die $err;
3983     }
3984
3985     rmtree $tmpdir;
3986
3987     rename($tmpfn, $conffile) ||
3988         die "unable to commit configuration file '$conffile'\n";
3989
3990     PVE::Cluster::cfs_update(); # make sure we read new file
3991
3992     eval { rescan($vmid, 1); };
3993     warn $@ if $@;
3994 }
3995
3996 sub restore_tar_archive {
3997     my ($archive, $vmid, $user, $opts) = @_;
3998
3999     if ($archive ne '-') {
4000         my $firstfile = tar_archive_read_firstfile($archive);
4001         die "ERROR: file '$archive' dos not lock like a QemuServer vzdump backup\n"
4002             if $firstfile ne 'qemu-server.conf';
4003     }
4004
4005     my $storecfg = cfs_read_file('storage.cfg');
4006
4007     # destroy existing data - keep empty config
4008     my $vmcfgfn = PVE::QemuServer::config_file($vmid);
4009     destroy_vm($storecfg, $vmid, 1) if -f $vmcfgfn;
4010
4011     my $tocmd = "/usr/lib/qemu-server/qmextract";
4012
4013     $tocmd .= " --storage " . PVE::Tools::shellquote($opts->{storage}) if $opts->{storage};
4014     $tocmd .= " --pool " . PVE::Tools::shellquote($opts->{pool}) if $opts->{pool};
4015     $tocmd .= ' --prealloc' if $opts->{prealloc};
4016     $tocmd .= ' --info' if $opts->{info};
4017
4018     # tar option "xf" does not autodetect compression when read from STDIN,
4019     # so we pipe to zcat
4020     my $cmd = "zcat -f|tar xf " . PVE::Tools::shellquote($archive) . " " .
4021         PVE::Tools::shellquote("--to-command=$tocmd");
4022
4023     my $tmpdir = "/var/tmp/vzdumptmp$$";
4024     mkpath $tmpdir;
4025
4026     local $ENV{VZDUMP_TMPDIR} = $tmpdir;
4027     local $ENV{VZDUMP_VMID} = $vmid;
4028     local $ENV{VZDUMP_USER} = $user;
4029
4030     my $conffile = config_file($vmid);
4031     my $tmpfn = "$conffile.$$.tmp";
4032
4033     # disable interrupts (always do cleanups)
4034     local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = sub {
4035         print STDERR "got interrupt - ignored\n";
4036     };
4037
4038     eval {
4039         # enable interrupts
4040         local $SIG{INT} = $SIG{TERM} = $SIG{QUIT} = $SIG{HUP} = $SIG{PIPE} = sub {
4041             die "interrupted by signal\n";
4042         };
4043
4044         if ($archive eq '-') {
4045             print "extracting archive from STDIN\n";
4046             run_command($cmd, input => "<&STDIN");
4047         } else {
4048             print "extracting archive '$archive'\n";
4049             run_command($cmd);
4050         }
4051
4052         return if $opts->{info};
4053
4054         # read new mapping
4055         my $map = {};
4056         my $statfile = "$tmpdir/qmrestore.stat";
4057         if (my $fd = IO::File->new($statfile, "r")) {
4058             while (defined (my $line = <$fd>)) {
4059                 if ($line =~ m/vzdump:([^\s:]*):(\S+)$/) {
4060                     $map->{$1} = $2 if $1;
4061                 } else {
4062                     print STDERR "unable to parse line in statfile - $line\n";
4063                 }
4064             }
4065             $fd->close();
4066         }
4067
4068         my $confsrc = "$tmpdir/qemu-server.conf";
4069
4070         my $srcfd = new IO::File($confsrc, "r") ||
4071             die "unable to open file '$confsrc'\n";
4072
4073         my $outfd = new IO::File ($tmpfn, "w") ||
4074             die "unable to write config for VM $vmid\n";
4075
4076         my $cookie = { netcount => 0 };
4077         while (defined (my $line = <$srcfd>)) {
4078             restore_update_config_line($outfd, $cookie, $vmid, $map, $line, $opts->{unique});
4079         }
4080
4081         $srcfd->close();
4082         $outfd->close();
4083     };
4084     my $err = $@;
4085
4086     if ($err) {
4087
4088         unlink $tmpfn;
4089
4090         tar_restore_cleanup($storecfg, "$tmpdir/qmrestore.stat") if !$opts->{info};
4091
4092         die $err;
4093     }
4094
4095     rmtree $tmpdir;
4096
4097     rename $tmpfn, $conffile ||
4098         die "unable to commit configuration file '$conffile'\n";
4099
4100     PVE::Cluster::cfs_update(); # make sure we read new file
4101
4102     eval { rescan($vmid, 1); };
4103     warn $@ if $@;
4104 };
4105
4106
4107 # Internal snapshots
4108
4109 # NOTE: Snapshot create/delete involves several non-atomic
4110 # action, and can take a long time.
4111 # So we try to avoid locking the file and use 'lock' variable
4112 # inside the config file instead.
4113
4114 my $snapshot_copy_config = sub {
4115     my ($source, $dest) = @_;
4116
4117     foreach my $k (keys %$source) {
4118         next if $k eq 'snapshots';
4119         next if $k eq 'snapstate';
4120         next if $k eq 'snaptime';
4121         next if $k eq 'vmstate';
4122         next if $k eq 'lock';
4123         next if $k eq 'digest';
4124         next if $k eq 'description';
4125         next if $k =~ m/^unused\d+$/;
4126
4127         $dest->{$k} = $source->{$k};
4128     }
4129 };
4130
4131 my $snapshot_apply_config = sub {
4132     my ($conf, $snap) = @_;
4133
4134     # copy snapshot list
4135     my $newconf = {
4136         snapshots => $conf->{snapshots},
4137     };
4138
4139     # keep description and list of unused disks
4140     foreach my $k (keys %$conf) {
4141         next if !($k =~ m/^unused\d+$/ || $k eq 'description');
4142         $newconf->{$k} = $conf->{$k};
4143     }
4144
4145     &$snapshot_copy_config($snap, $newconf);
4146
4147     return $newconf;
4148 };
4149
4150 sub foreach_writable_storage {
4151     my ($conf, $func) = @_;
4152
4153     my $sidhash = {};
4154
4155     foreach my $ds (keys %$conf) {
4156         next if !valid_drivename($ds);
4157
4158         my $drive = parse_drive($ds, $conf->{$ds});
4159         next if !$drive;
4160         next if drive_is_cdrom($drive);
4161
4162         my $volid = $drive->{file};
4163
4164         my ($sid, $volname) = PVE::Storage::parse_volume_id($volid, 1);
4165         $sidhash->{$sid} = $sid if $sid;
4166     }
4167
4168     foreach my $sid (sort keys %$sidhash) {
4169         &$func($sid);
4170     }
4171 }
4172
4173 my $alloc_vmstate_volid = sub {
4174     my ($storecfg, $vmid, $conf, $snapname) = @_;
4175
4176     # Note: we try to be smart when selecting a $target storage
4177
4178     my $target;
4179
4180     # search shared storage first
4181     foreach_writable_storage($conf, sub {
4182         my ($sid) = @_;
4183         my $scfg = PVE::Storage::storage_config($storecfg, $sid);
4184         return if !$scfg->{shared};
4185
4186         $target = $sid if !$target || $scfg->{path}; # prefer file based storage
4187     });
4188
4189     if (!$target) {
4190         # now search local storage
4191         foreach_writable_storage($conf, sub {
4192             my ($sid) = @_;
4193             my $scfg = PVE::Storage::storage_config($storecfg, $sid);
4194             return if $scfg->{shared};
4195
4196             $target = $sid if !$target || $scfg->{path}; # prefer file based storage;
4197         });
4198     }
4199
4200     $target = 'local' if !$target;
4201
4202     my $driver_state_size = 500; # assume 32MB is enough to safe all driver state;
4203     # we abort live save after $conf->{memory}, so we need at max twice that space
4204     my $size = $conf->{memory}*2 + $driver_state_size;
4205
4206     my $name = "vm-$vmid-state-$snapname";
4207     my $scfg = PVE::Storage::storage_config($storecfg, $target);
4208     $name .= ".raw" if $scfg->{path}; # add filename extension for file base storage
4209     my $volid = PVE::Storage::vdisk_alloc($storecfg, $target, $vmid, 'raw', $name, $size*1024);
4210
4211     return $volid;
4212 };
4213
4214 my $snapshot_prepare = sub {
4215     my ($vmid, $snapname, $save_vmstate, $comment) = @_;
4216
4217     my $snap;
4218
4219     my $updatefn =  sub {
4220
4221         my $conf = load_config($vmid);
4222
4223         die "you can't take a snapshot if it's a template\n"
4224             if is_template($conf);
4225
4226         check_lock($conf);
4227
4228         $conf->{lock} = 'snapshot';
4229
4230         die "snapshot name '$snapname' already used\n"
4231             if defined($conf->{snapshots}->{$snapname});
4232
4233         my $storecfg = PVE::Storage::config();
4234         die "snapshot feature is not available" if !has_feature('snapshot', $conf, $storecfg);
4235
4236         $snap = $conf->{snapshots}->{$snapname} = {};
4237
4238         if ($save_vmstate && check_running($vmid)) {
4239             $snap->{vmstate} = &$alloc_vmstate_volid($storecfg, $vmid, $conf, $snapname);
4240         }
4241
4242         &$snapshot_copy_config($conf, $snap);
4243
4244         $snap->{snapstate} = "prepare";
4245         $snap->{snaptime} = time();
4246         $snap->{description} = $comment if $comment;
4247
4248         # always overwrite machine if we save vmstate. This makes sure we
4249         # can restore it later using correct machine type
4250         $snap->{machine} = get_current_qemu_machine($vmid) if $snap->{vmstate};
4251
4252         update_config_nolock($vmid, $conf, 1);
4253     };
4254
4255     lock_config($vmid, $updatefn);
4256
4257     return $snap;
4258 };
4259
4260 my $snapshot_commit = sub {
4261     my ($vmid, $snapname) = @_;
4262
4263     my $updatefn = sub {
4264
4265         my $conf = load_config($vmid);
4266
4267         die "missing snapshot lock\n"
4268             if !($conf->{lock} && $conf->{lock} eq 'snapshot');
4269
4270         my $snap = $conf->{snapshots}->{$snapname};
4271
4272         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4273
4274         die "wrong snapshot state\n"
4275             if !($snap->{snapstate} && $snap->{snapstate} eq "prepare");
4276
4277         delete $snap->{snapstate};
4278         delete $conf->{lock};
4279
4280         my $newconf = &$snapshot_apply_config($conf, $snap);
4281
4282         $newconf->{parent} = $snapname;
4283
4284         update_config_nolock($vmid, $newconf, 1);
4285     };
4286
4287     lock_config($vmid, $updatefn);
4288 };
4289
4290 sub snapshot_rollback {
4291     my ($vmid, $snapname) = @_;
4292
4293     my $snap;
4294
4295     my $prepare = 1;
4296
4297     my $storecfg = PVE::Storage::config();
4298
4299     my $updatefn = sub {
4300
4301         my $conf = load_config($vmid);
4302
4303         die "you can't rollback if vm is a template\n" if is_template($conf);
4304
4305         $snap = $conf->{snapshots}->{$snapname};
4306
4307         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4308
4309         die "unable to rollback to incomplete snapshot (snapstate = $snap->{snapstate})\n"
4310             if $snap->{snapstate};
4311
4312         if ($prepare) {
4313             check_lock($conf);
4314             vm_stop($storecfg, $vmid, undef, undef, 5, undef, undef);
4315         }
4316
4317         die "unable to rollback vm $vmid: vm is running\n"
4318             if check_running($vmid);
4319
4320         if ($prepare) {
4321             $conf->{lock} = 'rollback';
4322         } else {
4323             die "got wrong lock\n" if !($conf->{lock} && $conf->{lock} eq 'rollback');
4324             delete $conf->{lock};
4325         }
4326
4327         my $forcemachine;
4328
4329         if (!$prepare) {
4330             my $has_machine_config = defined($conf->{machine});
4331
4332             # copy snapshot config to current config
4333             $conf = &$snapshot_apply_config($conf, $snap);
4334             $conf->{parent} = $snapname;
4335
4336             # Note: old code did not store 'machine', so we try to be smart
4337             # and guess the snapshot was generated with kvm 1.4 (pc-i440fx-1.4).
4338             $forcemachine = $conf->{machine} || 'pc-i440fx-1.4';
4339             # we remove the 'machine' configuration if not explicitly specified
4340             # in the original config.
4341             delete $conf->{machine} if $snap->{vmstate} && !$has_machine_config;
4342         }
4343
4344         update_config_nolock($vmid, $conf, 1);
4345
4346         if (!$prepare && $snap->{vmstate}) {
4347             my $statefile = PVE::Storage::path($storecfg, $snap->{vmstate});
4348             vm_start($storecfg, $vmid, $statefile, undef, undef, undef, $forcemachine);
4349         }
4350     };
4351
4352     lock_config($vmid, $updatefn);
4353
4354     foreach_drive($snap, sub {
4355         my ($ds, $drive) = @_;
4356
4357         return if drive_is_cdrom($drive);
4358
4359         my $volid = $drive->{file};
4360         my $device = "drive-$ds";
4361
4362         PVE::Storage::volume_snapshot_rollback($storecfg, $volid, $snapname);
4363     });
4364
4365     $prepare = 0;
4366     lock_config($vmid, $updatefn);
4367 }
4368
4369 my $savevm_wait = sub {
4370     my ($vmid) = @_;
4371
4372     for(;;) {
4373         my $stat = vm_mon_cmd_nocheck($vmid, "query-savevm");
4374         if (!$stat->{status}) {
4375             die "savevm not active\n";
4376         } elsif ($stat->{status} eq 'active') {
4377             sleep(1);
4378             next;
4379         } elsif ($stat->{status} eq 'completed') {
4380             last;
4381         } else {
4382             die "query-savevm returned status '$stat->{status}'\n";
4383         }
4384     }
4385 };
4386
4387 sub snapshot_create {
4388     my ($vmid, $snapname, $save_vmstate, $freezefs, $comment) = @_;
4389
4390     my $snap = &$snapshot_prepare($vmid, $snapname, $save_vmstate, $comment);
4391
4392     $freezefs = $save_vmstate = 0 if !$snap->{vmstate}; # vm is not running
4393
4394     my $drivehash = {};
4395
4396     my $running = check_running($vmid);
4397
4398     eval {
4399         # create internal snapshots of all drives
4400
4401         my $storecfg = PVE::Storage::config();
4402
4403         if ($running) {
4404             if ($snap->{vmstate}) {
4405                 my $path = PVE::Storage::path($storecfg, $snap->{vmstate});
4406                 vm_mon_cmd($vmid, "savevm-start", statefile => $path);
4407                 &$savevm_wait($vmid);
4408             } else {
4409                 vm_mon_cmd($vmid, "savevm-start");
4410             }
4411         };
4412
4413         qga_freezefs($vmid) if $running && $freezefs;
4414
4415         foreach_drive($snap, sub {
4416             my ($ds, $drive) = @_;
4417
4418             return if drive_is_cdrom($drive);
4419
4420             my $volid = $drive->{file};
4421             my $device = "drive-$ds";
4422
4423             qemu_volume_snapshot($vmid, $device, $storecfg, $volid, $snapname);
4424             $drivehash->{$ds} = 1;
4425        });
4426     };
4427     my $err = $@;
4428
4429     eval { qga_unfreezefs($vmid) if $running && $freezefs; };
4430     warn $@ if $@;
4431
4432     eval { vm_mon_cmd($vmid, "savevm-end") if $running; };
4433     warn $@ if $@;
4434
4435     if ($err) {
4436         warn "snapshot create failed: starting cleanup\n";
4437         eval { snapshot_delete($vmid, $snapname, 0, $drivehash); };
4438         warn $@ if $@;
4439         die $err;
4440     }
4441
4442     &$snapshot_commit($vmid, $snapname);
4443 }
4444
4445 # Note: $drivehash is only set when called from snapshot_create.
4446 sub snapshot_delete {
4447     my ($vmid, $snapname, $force, $drivehash) = @_;
4448
4449     my $prepare = 1;
4450
4451     my $snap;
4452     my $unused = [];
4453
4454     my $unlink_parent = sub {
4455         my ($confref, $new_parent) = @_;
4456
4457         if ($confref->{parent} && $confref->{parent} eq $snapname) {
4458             if ($new_parent) {
4459                 $confref->{parent} = $new_parent;
4460             } else {
4461                 delete $confref->{parent};
4462             }
4463         }
4464     };
4465
4466     my $updatefn =  sub {
4467         my ($remove_drive) = @_;
4468
4469         my $conf = load_config($vmid);
4470
4471         if (!$drivehash) {
4472             check_lock($conf);
4473             die "you can't delete a snapshot if vm is a template\n"
4474                 if is_template($conf);
4475         }
4476
4477         $snap = $conf->{snapshots}->{$snapname};
4478
4479         die "snapshot '$snapname' does not exist\n" if !defined($snap);
4480
4481         # remove parent refs
4482         &$unlink_parent($conf, $snap->{parent});
4483         foreach my $sn (keys %{$conf->{snapshots}}) {
4484             next if $sn eq $snapname;
4485             &$unlink_parent($conf->{snapshots}->{$sn}, $snap->{parent});
4486         }
4487
4488         if ($remove_drive) {
4489             if ($remove_drive eq 'vmstate') {
4490                 delete $snap->{$remove_drive};
4491             } else {
4492                 my $drive = parse_drive($remove_drive, $snap->{$remove_drive});
4493                 my $volid = $drive->{file};
4494                 delete $snap->{$remove_drive};
4495                 add_unused_volume($conf, $volid);
4496             }
4497         }
4498
4499         if ($prepare) {
4500             $snap->{snapstate} = 'delete';
4501         } else {
4502             delete $conf->{snapshots}->{$snapname};
4503             delete $conf->{lock} if $drivehash;
4504             foreach my $volid (@$unused) {
4505                 add_unused_volume($conf, $volid);
4506             }
4507         }
4508
4509         update_config_nolock($vmid, $conf, 1);
4510     };
4511
4512     lock_config($vmid, $updatefn);
4513
4514     # now remove vmstate file
4515
4516     my $storecfg = PVE::Storage::config();
4517
4518     if ($snap->{vmstate}) {
4519         eval {  PVE::Storage::vdisk_free($storecfg, $snap->{vmstate}); };
4520         if (my $err = $@) {
4521             die $err if !$force;
4522             warn $err;
4523         }
4524         # save changes (remove vmstate from snapshot)
4525         lock_config($vmid, $updatefn, 'vmstate') if !$force;
4526     };
4527
4528     # now remove all internal snapshots
4529     foreach_drive($snap, sub {
4530         my ($ds, $drive) = @_;
4531
4532         return if drive_is_cdrom($drive);
4533
4534         my $volid = $drive->{file};
4535         my $device = "drive-$ds";
4536
4537         if (!$drivehash || $drivehash->{$ds}) {
4538             eval { qemu_volume_snapshot_delete($vmid, $device, $storecfg, $volid, $snapname); };
4539             if (my $err = $@) {
4540                 die $err if !$force;
4541                 warn $err;
4542             }
4543         }
4544
4545         # save changes (remove drive fron snapshot)
4546         lock_config($vmid, $updatefn, $ds) if !$force;
4547         push @$unused, $volid;
4548     });
4549
4550     # now cleanup config
4551     $prepare = 0;
4552     lock_config($vmid, $updatefn);
4553 }
4554
4555 sub has_feature {
4556     my ($feature, $conf, $storecfg, $snapname, $running) = @_;
4557
4558     my $err;
4559     foreach_drive($conf, sub {
4560         my ($ds, $drive) = @_;
4561
4562         return if drive_is_cdrom($drive);
4563         my $volid = $drive->{file};
4564         $err = 1 if !PVE::Storage::volume_has_feature($storecfg, $feature, $volid, $snapname, $running);
4565     });
4566
4567     return $err ? 0 : 1;
4568 }
4569
4570 sub template_create {
4571     my ($vmid, $conf, $disk) = @_;
4572
4573     my $storecfg = PVE::Storage::config();
4574
4575     foreach_drive($conf, sub {
4576         my ($ds, $drive) = @_;
4577
4578         return if drive_is_cdrom($drive);
4579         return if $disk && $ds ne $disk;
4580
4581         my $volid = $drive->{file};
4582         return if !PVE::Storage::volume_has_feature($storecfg, 'template', $volid);
4583
4584         my $voliddst = PVE::Storage::vdisk_create_base($storecfg, $volid);
4585         $drive->{file} = $voliddst;
4586         $conf->{$ds} = print_drive($vmid, $drive);
4587         update_config_nolock($vmid, $conf, 1);
4588     });
4589 }
4590
4591 sub is_template {
4592     my ($conf) = @_;
4593
4594     return 1 if defined $conf->{template} && $conf->{template} == 1;
4595 }
4596
4597 sub qemu_img_convert {
4598     my ($src_volid, $dst_volid, $size, $snapname) = @_;
4599
4600     my $storecfg = PVE::Storage::config();
4601     my ($src_storeid, $src_volname) = PVE::Storage::parse_volume_id($src_volid, 1);
4602     my ($dst_storeid, $dst_volname) = PVE::Storage::parse_volume_id($dst_volid, 1);
4603
4604     if ($src_storeid && $dst_storeid) {
4605         my $src_scfg = PVE::Storage::storage_config($storecfg, $src_storeid);
4606         my $dst_scfg = PVE::Storage::storage_config($storecfg, $dst_storeid);
4607
4608         my $src_format = qemu_img_format($src_scfg, $src_volname);
4609         my $dst_format = qemu_img_format($dst_scfg, $dst_volname);
4610
4611         my $src_path = PVE::Storage::path($storecfg, $src_volid, $snapname);
4612         my $dst_path = PVE::Storage::path($storecfg, $dst_volid);
4613
4614         my $cmd = [];
4615         push @$cmd, '/usr/bin/qemu-img', 'convert', '-t', 'writeback', '-p', '-C';
4616         push @$cmd, '-s', $snapname if($snapname && $src_format eq "qcow2");
4617         push @$cmd, '-f', $src_format, '-O', $dst_format, $src_path, $dst_path;
4618
4619         my $parser = sub {
4620             my $line = shift;
4621             if($line =~ m/\((\S+)\/100\%\)/){
4622                 my $percent = $1;
4623                 my $transferred = int($size * $percent / 100);
4624                 my $remaining = $size - $transferred;
4625
4626                 print "transferred: $transferred bytes remaining: $remaining bytes total: $size bytes progression: $percent %\n";
4627             }
4628
4629         };
4630
4631         eval  { run_command($cmd, timeout => undef, outfunc => $parser); };
4632         my $err = $@;
4633         die "copy failed: $err" if $err;
4634     }
4635 }
4636
4637 sub qemu_img_format {
4638     my ($scfg, $volname) = @_;
4639
4640     if ($scfg->{path} && $volname =~ m/\.(raw|qcow2|qed|vmdk)$/) {
4641         return $1;
4642     } elsif ($scfg->{type} eq 'iscsi') {
4643         return "host_device";
4644     } else {
4645         return "raw";
4646     }
4647 }
4648
4649 sub qemu_drive_mirror {
4650     my ($vmid, $drive, $dst_volid, $vmiddst, $maxwait) = @_;
4651
4652     my $count = 1;
4653     my $old_len = 0;
4654     my $frozen = undef;
4655
4656     my $storecfg = PVE::Storage::config();
4657     my ($dst_storeid, $dst_volname) = PVE::Storage::parse_volume_id($dst_volid, 1);
4658
4659     if ($dst_storeid) {
4660         my $dst_scfg = PVE::Storage::storage_config($storecfg, $dst_storeid);
4661
4662         my $format;
4663         if ($dst_volname =~ m/\.(raw|qcow2)$/){
4664             $format = $1;
4665         }
4666
4667         my $dst_path = PVE::Storage::path($storecfg, $dst_volid);
4668
4669         if ($format) {
4670             #fixme : sometime drive-mirror timeout, but works fine after.
4671             # (I have see the problem with big volume > 200GB), so we need to eval
4672             eval { vm_mon_cmd($vmid, "drive-mirror", timeout => 10, device => "drive-$drive", mode => "existing",
4673                               sync => "full", target => $dst_path, format => $format); };
4674         } else {
4675             eval { vm_mon_cmd($vmid, "drive-mirror", timeout => 10, device => "drive-$drive", mode => "existing",
4676                               sync => "full", target => $dst_path); };
4677         }
4678
4679         eval {
4680             while (1) {
4681                 my $stats = vm_mon_cmd($vmid, "query-block-jobs");
4682                 my $stat = @$stats[0];
4683                 die "mirroring job seem to have die. Maybe do you have bad sectors?" if !$stat;
4684                 die "error job is not mirroring" if $stat->{type} ne "mirror";
4685
4686                 my $transferred = $stat->{offset};
4687                 my $total = $stat->{len};
4688                 my $remaining = $total - $transferred;
4689                 my $percent = sprintf "%.2f", ($transferred * 100 / $total);
4690
4691                 print "transferred: $transferred bytes remaining: $remaining bytes total: $total bytes progression: $percent %\n";
4692
4693                 last if ($stat->{len} == $stat->{offset});
4694                 if ($old_len == $stat->{offset}) {
4695                     if ($maxwait && $count > $maxwait) {
4696                     # if writes to disk occurs the disk needs to be freezed
4697                     # to be able to complete the migration
4698                         vm_suspend($vmid,1);
4699                         $count = 0;
4700                         $frozen = 1;
4701                     } else {
4702                         $count++ unless $frozen;
4703                     }
4704                 } elsif ($frozen) {
4705                     vm_resume($vmid,1);
4706                     $count = 0;
4707                 }
4708                 $old_len = $stat->{offset};
4709                 sleep 1;
4710             }
4711
4712             if ($vmiddst == $vmid) {
4713                 # switch the disk if source and destination are on the same guest
4714                 vm_mon_cmd($vmid, "block-job-complete", device => "drive-$drive");
4715             }
4716         };
4717         if (my $err = $@) {
4718             eval { vm_mon_cmd($vmid, "block-job-cancel", device => "drive-$drive"); };
4719             die "mirroring error: $err";
4720         }
4721
4722         if ($vmiddst != $vmid) {
4723             # if we clone a disk for a new target vm, we don't switch the disk
4724             vm_mon_cmd($vmid, "block-job-cancel", device => "drive-$drive");
4725         }
4726     }
4727 }
4728
4729 sub clone_disk {
4730     my ($storecfg, $vmid, $running, $drivename, $drive, $snapname,
4731         $newvmid, $storage, $format, $full, $newvollist) = @_;
4732
4733     my $newvolid;
4734
4735     if (!$full) {
4736         print "create linked clone of drive $drivename ($drive->{file})\n";
4737         $newvolid = PVE::Storage::vdisk_clone($storecfg,  $drive->{file}, $newvmid);
4738         push @$newvollist, $newvolid;
4739     } else {
4740         my ($storeid, $volname) = PVE::Storage::parse_volume_id($drive->{file});
4741         $storeid = $storage if $storage;
4742
4743         my ($defFormat, $validFormats) = PVE::Storage::storage_default_format($storecfg, $storeid);
4744         if (!$format) {
4745             $format = $drive->{format} || $defFormat;
4746         }
4747
4748         # test if requested format is supported - else use default
4749         my $supported = grep { $_ eq $format } @$validFormats;
4750         $format = $defFormat if !$supported;
4751
4752         my ($size) = PVE::Storage::volume_size_info($storecfg, $drive->{file}, 3);
4753
4754         print "create full clone of drive $drivename ($drive->{file})\n";
4755         $newvolid = PVE::Storage::vdisk_alloc($storecfg, $storeid, $newvmid, $format, undef, ($size/1024));
4756         push @$newvollist, $newvolid;
4757
4758         if (!$running || $snapname) {
4759             qemu_img_convert($drive->{file}, $newvolid, $size, $snapname);
4760         } else {
4761             qemu_drive_mirror($vmid, $drivename, $newvolid, $newvmid);
4762         }
4763     }
4764
4765     my ($size) = PVE::Storage::volume_size_info($storecfg, $newvolid, 3);
4766
4767     my $disk = $drive;
4768     $disk->{format} = undef;
4769     $disk->{file} = $newvolid;
4770     $disk->{size} = $size;
4771
4772     return $disk;
4773 }
4774
4775 # this only works if VM is running
4776 sub get_current_qemu_machine {
4777     my ($vmid) = @_;
4778
4779     my $cmd = { execute => 'query-machines', arguments => {} };
4780     my $res = PVE::QemuServer::vm_qmp_command($vmid, $cmd);
4781
4782     my ($current, $default);
4783     foreach my $e (@$res) {
4784         $default = $e->{name} if $e->{'is-default'};
4785         $current = $e->{name} if $e->{'is-current'};
4786     }
4787
4788     # fallback to the default machine if current is not supported by qemu
4789     return $current || $default || 'pc';
4790 }
4791
4792 1;