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