1 package PVE
::QemuMigrate
;
8 use POSIX
qw( WNOHANG );
9 use Time
::HiRes
qw( usleep );
13 use PVE
::RPCEnvironment
;
15 use PVE
::ReplicationConfig
;
16 use PVE
::ReplicationState
;
21 use PVE
::QemuServer
::CPUConfig
;
22 use PVE
::QemuServer
::Drive
;
23 use PVE
::QemuServer
::Helpers
qw(min_version);
24 use PVE
::QemuServer
::Machine
;
25 use PVE
::QemuServer
::Monitor
qw(mon_cmd);
28 use PVE
::AbstractMigrate
;
29 use base
qw(PVE::AbstractMigrate);
31 sub fork_command_pipe
{
32 my ($self, $cmd) = @_;
34 my $reader = IO
::File-
>new();
35 my $writer = IO
::File-
>new();
41 eval { $cpid = open2
($reader, $writer, @$cmd); };
46 if ($orig_pid != $$) {
47 $self->log('err', "can't fork command pipe\n");
54 return { writer
=> $writer, reader
=> $reader, pid
=> $cpid };
57 sub finish_command_pipe
{
58 my ($self, $cmdpipe, $timeout) = @_;
60 my $cpid = $cmdpipe->{pid
};
61 return if !defined($cpid);
63 my $writer = $cmdpipe->{writer
};
64 my $reader = $cmdpipe->{reader
};
69 my $collect_child_process = sub {
70 my $res = waitpid($cpid, WNOHANG
);
71 if (defined($res) && ($res == $cpid)) {
72 delete $cmdpipe->{cpid
};
80 for (my $i = 0; $i < $timeout; $i++) {
81 return if &$collect_child_process();
86 $self->log('info', "ssh tunnel still running - terminating now with SIGTERM\n");
90 for (my $i = 0; $i < 10; $i++) {
91 return if &$collect_child_process();
95 $self->log('info', "ssh tunnel still running - terminating now with SIGKILL\n");
99 $self->log('err', "ssh tunnel child process (PID $cpid) couldn't be collected\n")
100 if !&$collect_child_process();
104 my ($self, $tunnel, $timeout) = @_;
106 $timeout = 60 if !defined($timeout);
108 my $reader = $tunnel->{reader
};
112 PVE
::Tools
::run_with_timeout
($timeout, sub { $output = <$reader>; });
114 die "reading from tunnel failed: $@\n" if $@;
122 my ($self, $tunnel, $timeout, $command) = @_;
124 $timeout = 60 if !defined($timeout);
126 my $writer = $tunnel->{writer
};
129 PVE
::Tools
::run_with_timeout
($timeout, sub {
130 print $writer "$command\n";
134 die "writing to tunnel failed: $@\n" if $@;
136 if ($tunnel->{version
} && $tunnel->{version
} >= 1) {
137 my $res = eval { $self->read_tunnel($tunnel, 10); };
138 die "no reply to command '$command': $@\n" if $@;
143 die "tunnel replied '$res' to command '$command'\n";
149 my ($self, $tunnel_addr) = @_;
151 my @localtunnelinfo = ();
152 foreach my $addr (@$tunnel_addr) {
153 push @localtunnelinfo, '-L', $addr;
156 my $cmd = [@{$self->{rem_ssh
}}, '-o ExitOnForwardFailure=yes', @localtunnelinfo, '/usr/sbin/qm', 'mtunnel' ];
158 my $tunnel = $self->fork_command_pipe($cmd);
161 my $helo = $self->read_tunnel($tunnel, 60);
162 die "no reply\n" if !$helo;
163 die "no quorum on target node\n" if $helo =~ m/^no quorum$/;
164 die "got strange reply from mtunnel ('$helo')\n"
165 if $helo !~ m/^tunnel online$/;
170 my $ver = $self->read_tunnel($tunnel, 10);
171 if ($ver =~ /^ver (\d+)$/) {
172 $tunnel->{version
} = $1;
173 $self->log('info', "ssh tunnel $ver\n");
175 $err = "received invalid tunnel version string '$ver'\n" if !$err;
180 $self->finish_command_pipe($tunnel);
181 die "can't open migration tunnel - $err";
187 my ($self, $tunnel) = @_;
189 eval { $self->write_tunnel($tunnel, 30, 'quit'); };
192 $self->finish_command_pipe($tunnel, 30);
194 if ($tunnel->{sock_addr
}) {
195 # ssh does not clean up on local host
196 my $cmd = ['rm', '-f', @{$tunnel->{sock_addr
}}]; #
197 PVE
::Tools
::run_command
($cmd);
199 # .. and just to be sure check on remote side
200 unshift @{$cmd}, @{$self->{rem_ssh
}};
201 PVE
::Tools
::run_command
($cmd);
208 my ($self, $vmid, $code, @param) = @_;
210 return PVE
::QemuConfig-
>lock_config($vmid, $code, @param);
214 my ($self, $vmid) = @_;
216 my $online = $self->{opts
}->{online
};
218 $self->{storecfg
} = PVE
::Storage
::config
();
221 my $conf = $self->{vmconf
} = PVE
::QemuConfig-
>load_config($vmid);
223 PVE
::QemuConfig-
>check_lock($conf);
226 if (my $pid = PVE
::QemuServer
::check_running
($vmid)) {
227 die "can't migrate running VM without --online\n" if !$online;
230 $self->{forcemachine
} = PVE
::QemuServer
::Machine
::qemu_machine_pxe
($vmid, $conf);
232 # To support custom CPU types, we keep QEMU's "-cpu" parameter intact.
233 # Since the parameter itself contains no reference to a custom model,
234 # this makes migration independent of changes to "cpu-models.conf".
236 my $cpuconf = PVE
::QemuServer
::CPUConfig
::parse_cpu_conf_basic
($conf->{cpu
});
237 if ($cpuconf && PVE
::QemuServer
::CPUConfig
::is_custom_model
($cpuconf->{cputype
})) {
238 $self->{forcecpu
} = PVE
::QemuServer
::CPUConfig
::get_cpu_from_running_vm
($pid);
243 my $loc_res = PVE
::QemuServer
::check_local_resources
($conf, 1);
244 if (scalar @$loc_res) {
245 if ($self->{running
} || !$self->{opts
}->{force
}) {
246 die "can't migrate VM which uses local devices: " . join(", ", @$loc_res) . "\n";
248 $self->log('info', "migrating VM which uses local devices");
252 my $vollist = PVE
::QemuServer
::get_vm_volumes
($conf);
254 my $need_activate = [];
255 foreach my $volid (@$vollist) {
256 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
258 # check if storage is available on both nodes
259 my $targetsid = PVE
::QemuServer
::map_storage
($self->{opts
}->{storagemap
}, $sid);
261 my $scfg = PVE
::Storage
::storage_check_node
($self->{storecfg
}, $sid);
262 PVE
::Storage
::storage_check_node
($self->{storecfg
}, $targetsid, $self->{node
});
264 if ($scfg->{shared
}) {
265 # PVE::Storage::activate_storage checks this for non-shared storages
266 my $plugin = PVE
::Storage
::Plugin-
>lookup($scfg->{type
});
267 warn "Used shared storage '$sid' is not online on source node!\n"
268 if !$plugin->check_connection($sid, $scfg);
270 # only activate if not shared
271 next if ($volid =~ m/vm-\d+-cloudinit/);
272 push @$need_activate, $volid;
277 PVE
::Storage
::activate_volumes
($self->{storecfg
}, $need_activate);
279 # test ssh connection
280 my $cmd = [ @{$self->{rem_ssh
}}, '/bin/true' ];
281 eval { $self->cmd_quiet($cmd); };
282 die "Can't connect to destination address using public key\n" if $@;
288 my ($self, $vmid) = @_;
290 my $conf = $self->{vmconf
};
292 # local volumes which have been copied
293 # and their old_id => new_id pairs
294 $self->{volumes
} = [];
295 $self->{volume_map
} = {};
297 my $storecfg = $self->{storecfg
};
300 # found local volumes and their origin
301 my $local_volumes = {};
302 my $local_volumes_errors = {};
303 my $other_errors = [];
306 my $log_error = sub {
307 my ($msg, $volid) = @_;
309 if (defined($volid)) {
310 $local_volumes_errors->{$volid} = $msg;
312 push @$other_errors, $msg;
317 my @sids = PVE
::Storage
::storage_ids
($storecfg);
318 foreach my $storeid (@sids) {
319 my $scfg = PVE
::Storage
::storage_config
($storecfg, $storeid);
320 next if $scfg->{shared
};
321 next if !PVE
::Storage
::storage_check_enabled
($storecfg, $storeid, undef, 1);
323 # get list from PVE::Storage (for unused volumes)
324 my $dl = PVE
::Storage
::vdisk_list
($storecfg, $storeid, $vmid);
326 next if @{$dl->{$storeid}} == 0;
328 my $targetsid = PVE
::QemuServer
::map_storage
($self->{opts
}->{storagemap
}, $storeid);
329 # check if storage is available on target node
330 PVE
::Storage
::storage_check_node
($storecfg, $targetsid, $self->{node
});
332 # grandfather in existing mismatches
333 if ($targetsid ne $storeid) {
334 my $target_scfg = PVE
::Storage
::storage_config
($storecfg, $targetsid);
335 die "content type 'images' is not available on storage '$targetsid'\n"
336 if !$target_scfg->{content
}->{images
};
339 PVE
::Storage
::foreach_volid
($dl, sub {
340 my ($volid, $sid, $volinfo) = @_;
342 $local_volumes->{$volid}->{ref} = 'storage';
344 # If with_snapshots is not set for storage migrate, it tries to use
345 # a raw+size stream, but on-the-fly conversion from qcow2 to raw+size
346 # back to qcow2 is currently not possible.
347 $local_volumes->{$volid}->{snapshots
} = ($volinfo->{format
} =~ /^(?:qcow2|vmdk)$/);
348 $local_volumes->{$volid}->{format
} = $volinfo->{format
};
352 my $rep_cfg = PVE
::ReplicationConfig-
>new();
353 my $replication_jobcfg = $rep_cfg->find_local_replication_job($vmid, $self->{node
});
354 my $replicatable_volumes = !$replication_jobcfg ?
{}
355 : PVE
::QemuConfig-
>get_replicatable_volumes($storecfg, $vmid, $conf, 0, 1);
357 my $test_volid = sub {
358 my ($volid, $attr) = @_;
360 if ($volid =~ m
|^/|) {
361 return if $attr->{shared
};
362 $local_volumes->{$volid}->{ref} = 'config';
363 die "local file/device\n";
366 my $snaprefs = $attr->{referenced_in_snapshot
};
368 if ($attr->{cdrom
}) {
369 if ($volid eq 'cdrom') {
370 my $msg = "can't migrate local cdrom drive";
371 if (defined($snaprefs) && !$attr->{referenced_in_config
}) {
372 my $snapnames = join(', ', sort keys %$snaprefs);
373 $msg .= " (referenced in snapshot - $snapnames)";
375 &$log_error("$msg\n");
378 return if $volid eq 'none';
381 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
383 my $targetsid = PVE
::QemuServer
::map_storage
($self->{opts
}->{storagemap
}, $sid);
384 # check if storage is available on both nodes
385 my $scfg = PVE
::Storage
::storage_check_node
($storecfg, $sid);
386 PVE
::Storage
::storage_check_node
($storecfg, $targetsid, $self->{node
});
388 return if $scfg->{shared
};
390 $local_volumes->{$volid}->{ref} = $attr->{referenced_in_config
} ?
'config' : 'snapshot';
392 if ($attr->{cdrom
}) {
393 if ($volid =~ /vm-\d+-cloudinit/) {
394 $local_volumes->{$volid}->{ref} = 'generated';
397 die "local cdrom image\n";
400 my ($path, $owner) = PVE
::Storage
::path
($storecfg, $volid);
402 die "owned by other VM (owner = VM $owner)\n"
403 if !$owner || ($owner != $vmid);
405 if (defined($snaprefs)) {
406 $local_volumes->{$volid}->{snapshots
} = 1;
408 # we cannot migrate shapshots on local storage
409 # exceptions: 'zfspool' or 'qcow2' files (on directory storage)
411 die "online storage migration not possible if snapshot exists\n" if $self->{running
};
412 if (!($scfg->{type
} eq 'zfspool' || $local_volumes->{$volid}->{format
} eq 'qcow2')) {
413 die "non-migratable snapshot exists\n";
417 die "referenced by linked clone(s)\n"
418 if PVE
::Storage
::volume_is_base_and_used
($storecfg, $volid);
421 PVE
::QemuServer
::foreach_volid
($conf, sub {
422 my ($volid, $attr) = @_;
423 eval { $test_volid->($volid, $attr); };
425 &$log_error($err, $volid);
429 foreach my $vol (sort keys %$local_volumes) {
430 my $type = $replicatable_volumes->{$vol} ?
'local, replicated' : 'local';
431 my $ref = $local_volumes->{$vol}->{ref};
432 if ($ref eq 'storage') {
433 $self->log('info', "found $type disk '$vol' (via storage)\n");
434 } elsif ($ref eq 'config') {
435 &$log_error("can't live migrate attached local disks without with-local-disks option\n", $vol)
436 if $self->{running
} && !$self->{opts
}->{"with-local-disks"};
437 $self->log('info', "found $type disk '$vol' (in current VM config)\n");
438 } elsif ($ref eq 'snapshot') {
439 $self->log('info', "found $type disk '$vol' (referenced by snapshot(s))\n");
440 } elsif ($ref eq 'generated') {
441 $self->log('info', "found generated disk '$vol' (in current VM config)\n");
443 $self->log('info', "found $type disk '$vol'\n");
447 foreach my $vol (sort keys %$local_volumes_errors) {
448 $self->log('warn', "can't migrate local disk '$vol': $local_volumes_errors->{$vol}");
450 foreach my $err (@$other_errors) {
451 $self->log('warn', "$err");
455 die "can't migrate VM - check log\n";
458 # additional checks for local storage
459 foreach my $volid (keys %$local_volumes) {
460 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
461 my $scfg = PVE
::Storage
::storage_config
($storecfg, $sid);
463 my $migratable = $scfg->{type
} =~ /^(?:dir|zfspool|lvmthin|lvm)$/;
465 die "can't migrate '$volid' - storage type '$scfg->{type}' not supported\n"
468 # image is a linked clone on local storage, se we can't migrate.
469 if (my $basename = (PVE
::Storage
::parse_volname
($storecfg, $volid))[3]) {
470 die "can't migrate '$volid' as it's a clone of '$basename'";
474 if ($replication_jobcfg) {
475 if ($self->{running
}) {
477 my $version = PVE
::QemuServer
::kvm_user_version
();
478 if (!min_version
($version, 4, 2)) {
479 die "can't live migrate VM with replicated volumes, pve-qemu to old (< 4.2)!\n"
482 my $live_replicatable_volumes = {};
483 PVE
::QemuConfig-
>foreach_volume($conf, sub {
484 my ($ds, $drive) = @_;
486 my $volid = $drive->{file
};
487 $live_replicatable_volumes->{$ds} = $volid
488 if defined($replicatable_volumes->{$volid});
490 foreach my $drive (keys %$live_replicatable_volumes) {
491 my $volid = $live_replicatable_volumes->{$drive};
493 my $bitmap = "repl_$drive";
495 # start tracking before replication to get full delta + a few duplicates
496 $self->log('info', "$drive: start tracking writes using block-dirty-bitmap '$bitmap'");
497 mon_cmd
($vmid, 'block-dirty-bitmap-add', node
=> "drive-$drive", name
=> $bitmap);
499 # other info comes from target node in phase 2
500 $self->{target_drive
}->{$drive}->{bitmap
} = $bitmap;
503 $self->log('info', "replicating disk images");
505 my $start_time = time();
506 my $logfunc = sub { $self->log('info', shift) };
507 $self->{replicated_volumes
} = PVE
::Replication
::run_replication
(
508 'PVE::QemuConfig', $replication_jobcfg, $start_time, $start_time, $logfunc);
511 # sizes in config have to be accurate for remote node to correctly
512 # allocate disks, rescan to be sure
513 my $volid_hash = PVE
::QemuServer
::scan_volids
($storecfg, $vmid);
514 PVE
::QemuConfig-
>foreach_volume($conf, sub {
515 my ($key, $drive) = @_;
516 my ($updated, $old_size, $new_size) = PVE
::QemuServer
::Drive
::update_disksize
($drive, $volid_hash);
517 if (defined($updated)) {
518 $conf->{$key} = PVE
::QemuServer
::print_drive
($updated);
519 $self->log('info', "size of disk '$updated->{file}' ($key) updated from $old_size to $new_size\n");
523 # we want to set the efidisk size in the config to the size of the
524 # real OVMF_VARS.fd image, else we can create a too big image, which does not work
525 if (defined($conf->{efidisk0
})) {
526 PVE
::QemuServer
::update_efidisk_size
($conf);
529 $self->log('info', "copying local disk images") if scalar(%$local_volumes);
531 foreach my $volid (keys %$local_volumes) {
532 my ($sid, $volname) = PVE
::Storage
::parse_volume_id
($volid);
533 my $targetsid = PVE
::QemuServer
::map_storage
($self->{opts
}->{storagemap
}, $sid);
534 my $ref = $local_volumes->{$volid}->{ref};
535 if ($self->{running
} && $ref eq 'config') {
536 push @{$self->{online_local_volumes
}}, $volid;
537 } elsif ($ref eq 'generated') {
538 die "can't live migrate VM with local cloudinit disk. use a shared storage instead\n" if $self->{running
};
539 # skip all generated volumes but queue them for deletion in phase3_cleanup
540 push @{$self->{volumes
}}, $volid;
543 next if $self->{replicated_volumes
}->{$volid};
544 push @{$self->{volumes
}}, $volid;
545 my $opts = $self->{opts
};
546 # use 'migrate' limit for transfer to other node
547 my $bwlimit = PVE
::Storage
::get_bandwidth_limit
('migration', [$targetsid, $sid], $opts->{bwlimit
});
548 # JSONSchema and get_bandwidth_limit use kbps - storage_migrate bps
549 $bwlimit = $bwlimit * 1024 if defined($bwlimit);
551 my $storage_migrate_opts = {
552 'bwlimit' => $bwlimit,
553 'insecure' => $opts->{migration_type
} eq 'insecure',
554 'with_snapshots' => $local_volumes->{$volid}->{snapshots
},
557 my $new_volid = PVE
::Storage
::storage_migrate
($storecfg, $volid, $self->{ssh_info
},
558 $targetsid, $storage_migrate_opts);
560 $self->{volume_map
}->{$volid} = $new_volid;
561 $self->log('info', "volume '$volid' is '$new_volid' on the target\n");
565 die "Failed to sync data - $@" if $@;
568 sub cleanup_remotedisks
{
571 foreach my $target_drive (keys %{$self->{target_drive
}}) {
572 my $drivestr = $self->{target_drive
}->{$target_drive}->{drivestr
};
573 next if !defined($drivestr);
575 my $drive = PVE
::QemuServer
::parse_drive
($target_drive, $drivestr);
577 # don't clean up replicated disks!
578 next if defined($self->{replicated_volumes
}->{$drive->{file
}});
580 my ($storeid, $volname) = PVE
::Storage
::parse_volume_id
($drive->{file
});
582 my $cmd = [@{$self->{rem_ssh
}}, 'pvesm', 'free', "$storeid:$volname"];
584 eval{ PVE
::Tools
::run_command
($cmd, outfunc
=> sub {}, errfunc
=> sub {}) };
586 $self->log('err', $err);
592 sub cleanup_bitmaps
{
594 foreach my $drive (keys %{$self->{target_drive
}}) {
595 my $bitmap = $self->{target_drive
}->{$drive}->{bitmap
};
597 $self->log('info', "$drive: removing block-dirty-bitmap '$bitmap'");
598 mon_cmd
($self->{vmid
}, 'block-dirty-bitmap-remove', node
=> "drive-$drive", name
=> $bitmap);
603 my ($self, $vmid) = @_;
605 $self->log('info', "starting migration of VM $vmid to node '$self->{node}' ($self->{nodeip})");
607 my $conf = $self->{vmconf
};
609 # set migrate lock in config file
610 $conf->{lock} = 'migrate';
611 PVE
::QemuConfig-
>write_config($vmid, $conf);
613 sync_disks
($self, $vmid);
615 # sync_disks fixes disk sizes to match their actual size, write changes so
616 # target allocates correct volumes
617 PVE
::QemuConfig-
>write_config($vmid, $conf);
621 my ($self, $vmid, $err) = @_;
623 $self->log('info', "aborting phase 1 - cleanup resources");
625 my $conf = $self->{vmconf
};
626 delete $conf->{lock};
627 eval { PVE
::QemuConfig-
>write_config($vmid, $conf) };
629 $self->log('err', $err);
632 if ($self->{volumes
}) {
633 foreach my $volid (@{$self->{volumes
}}) {
634 $self->log('err', "found stale volume copy '$volid' on node '$self->{node}'");
635 # fixme: try to remove ?
639 eval { $self->cleanup_bitmaps() };
641 $self->log('err', $err);
647 my ($self, $vmid) = @_;
649 my $conf = $self->{vmconf
};
651 $self->log('info', "starting VM $vmid on remote node '$self->{node}'");
655 my $ruri; # the whole migration dst. URI (protocol:address[:port])
656 my $nodename = PVE
::INotify
::nodename
();
658 ## start on remote node
659 my $cmd = [@{$self->{rem_ssh
}}];
662 if (PVE
::QemuServer
::vga_conf_has_spice
($conf->{vga
})) {
663 my $res = mon_cmd
($vmid, 'query-spice');
664 $spice_ticket = $res->{ticket
};
667 push @$cmd , 'qm', 'start', $vmid, '--skiplock', '--migratedfrom', $nodename;
669 my $migration_type = $self->{opts
}->{migration_type
};
671 push @$cmd, '--migration_type', $migration_type;
673 push @$cmd, '--migration_network', $self->{opts
}->{migration_network
}
674 if $self->{opts
}->{migration_network
};
676 if ($migration_type eq 'insecure') {
677 push @$cmd, '--stateuri', 'tcp';
679 push @$cmd, '--stateuri', 'unix';
682 if ($self->{forcemachine
}) {
683 push @$cmd, '--machine', $self->{forcemachine
};
686 if ($self->{forcecpu
}) {
687 push @$cmd, '--force-cpu', $self->{forcecpu
};
690 if ($self->{online_local_volumes
}) {
691 push @$cmd, '--targetstorage', ($self->{opts
}->{targetstorage
} // '1');
695 my $tunnel_addr = [];
697 # version > 0 for unix socket support
698 my $nbd_protocol_version = 1;
699 # TODO change to 'spice_ticket: <ticket>\n' in 7.0
700 my $input = $spice_ticket ?
"$spice_ticket\n" : "\n";
701 $input .= "nbd_protocol_version: $nbd_protocol_version\n";
702 foreach my $volid (keys %{$self->{replicated_volumes
}}) {
703 $input .= "replicated_volume: $volid\n";
706 my $target_replicated_volumes = {};
708 # Note: We try to keep $spice_ticket secret (do not pass via command line parameter)
709 # instead we pipe it through STDIN
710 my $exitcode = PVE
::Tools
::run_command
($cmd, input
=> $input, outfunc
=> sub {
713 if ($line =~ m/^migration listens on tcp:(localhost|[\d\.]+|\[[\d\.:a-fA-F]+\]):(\d+)$/) {
716 $ruri = "tcp:$raddr:$rport";
718 elsif ($line =~ m!^migration listens on unix:(/run/qemu-server/(\d+)\.migrate)$!) {
720 die "Destination UNIX sockets VMID does not match source VMID" if $vmid ne $2;
721 $ruri = "unix:$raddr";
723 elsif ($line =~ m/^migration listens on port (\d+)$/) {
724 $raddr = "localhost";
726 $ruri = "tcp:$raddr:$rport";
728 elsif ($line =~ m/^spice listens on port (\d+)$/) {
729 $spice_port = int($1);
731 elsif ($line =~ m/^storage migration listens on nbd:(localhost|[\d\.]+|\[[\d\.:a-fA-F]+\]):(\d+):exportname=(\S+) volume:(\S+)$/) {
733 my $nbd_uri = "nbd:$1:$2:exportname=$3";
734 my $targetdrive = $3;
735 $targetdrive =~ s/drive-//g;
737 $self->{target_drive
}->{$targetdrive}->{drivestr
} = $drivestr;
738 $self->{target_drive
}->{$targetdrive}->{nbd_uri
} = $nbd_uri;
739 } elsif ($line =~ m!^storage migration listens on nbd:unix:(/run/qemu-server/(\d+)_nbd\.migrate):exportname=(\S+) volume:(\S+)$!) {
741 die "Destination UNIX socket's VMID does not match source VMID" if $vmid ne $2;
742 my $nbd_unix_addr = $1;
743 my $nbd_uri = "nbd:unix:$nbd_unix_addr:exportname=$3";
744 my $targetdrive = $3;
745 $targetdrive =~ s/drive-//g;
747 $self->{target_drive
}->{$targetdrive}->{drivestr
} = $drivestr;
748 $self->{target_drive
}->{$targetdrive}->{nbd_uri
} = $nbd_uri;
749 push @$tunnel_addr, "$nbd_unix_addr:$nbd_unix_addr";
750 push @$sock_addr, $nbd_unix_addr;
751 } elsif ($line =~ m/^re-using replicated volume: (\S+) - (.*)$/) {
754 $target_replicated_volumes->{$volid} = $drive;
755 } elsif ($line =~ m/^QEMU: (.*)$/) {
756 $self->log('info', "[$self->{node}] $1\n");
760 $self->log('info', "[$self->{node}] $line");
763 die "remote command failed with exit code $exitcode\n" if $exitcode;
765 die "unable to detect remote migration address\n" if !$raddr;
767 if (scalar(keys %$target_replicated_volumes) != scalar(keys %{$self->{replicated_volumes
}})) {
768 die "number of replicated disks on source and target node do not match - target node too old?\n"
771 $self->log('info', "start remote tunnel");
773 if ($migration_type eq 'secure') {
775 if ($ruri =~ /^unix:/) {
777 push @$tunnel_addr, "$raddr:$raddr";
778 $self->{tunnel
} = $self->fork_tunnel($tunnel_addr);
779 push @$sock_addr, $raddr;
781 my $unix_socket_try = 0; # wait for the socket to become ready
782 while ($unix_socket_try <= 100) {
785 foreach my $sock (@$sock_addr) {
791 if ($available == @$sock_addr) {
797 if ($unix_socket_try > 100) {
799 $self->finish_tunnel($self->{tunnel
});
800 die "Timeout, migration socket $ruri did not get ready";
803 } elsif ($ruri =~ /^tcp:/) {
805 if ($raddr eq "localhost") {
806 # for backwards compatibility with older qemu-server versions
807 my $pfamily = PVE
::Tools
::get_host_address_family
($nodename);
808 my $lport = PVE
::Tools
::next_migrate_port
($pfamily);
809 $tunnel_addr = "$lport:localhost:$rport";
812 $self->{tunnel
} = $self->fork_tunnel($tunnel_addr);
815 die "unsupported protocol in migration URI: $ruri\n";
818 #fork tunnel for insecure migration, to send faster commands like resume
819 $self->{tunnel
} = $self->fork_tunnel();
821 $self->{tunnel
}->{sock_addr
} = $sock_addr if (@$sock_addr);
825 my $opt_bwlimit = $self->{opts
}->{bwlimit
};
827 if (defined($self->{online_local_volumes
})) {
828 $self->{storage_migration
} = 1;
829 $self->{storage_migration_jobs
} = {};
830 $self->log('info', "starting storage migration");
832 die "The number of local disks does not match between the source and the destination.\n"
833 if (scalar(keys %{$self->{target_drive
}}) != scalar @{$self->{online_local_volumes
}});
834 foreach my $drive (keys %{$self->{target_drive
}}){
835 my $target = $self->{target_drive
}->{$drive};
836 my $nbd_uri = $target->{nbd_uri
};
838 my $source_drive = PVE
::QemuServer
::parse_drive
($drive, $conf->{$drive});
839 my $target_drive = PVE
::QemuServer
::parse_drive
($drive, $target->{drivestr
});
841 my $source_volid = $source_drive->{file
};
842 my $target_volid = $target_drive->{file
};
844 my $source_sid = PVE
::Storage
::Plugin
::parse_volume_id
($source_volid);
845 my $target_sid = PVE
::Storage
::Plugin
::parse_volume_id
($target_volid);
847 my $bwlimit = PVE
::Storage
::get_bandwidth_limit
('migration', [$source_sid, $target_sid], $opt_bwlimit);
848 my $bitmap = $target->{bitmap
};
850 $self->log('info', "$drive: start migration to $nbd_uri");
851 PVE
::QemuServer
::qemu_drive_mirror
($vmid, $drive, $nbd_uri, $vmid, undef, $self->{storage_migration_jobs
}, 'skip', undef, $bwlimit, $bitmap);
853 $self->{volume_map
}->{$source_volid} = $target_volid;
854 $self->log('info', "volume '$source_volid' is '$target_volid' on the target\n");
858 $self->log('info', "starting online/live migration on $ruri");
859 $self->{livemigration
} = 1;
862 my $defaults = PVE
::QemuServer
::load_defaults
();
864 $self->log('info', "set migration_caps");
866 PVE
::QemuServer
::set_migration_caps
($vmid);
870 my $qemu_migrate_params = {};
872 # migrate speed can be set via bwlimit (datacenter.cfg and API) and via the
873 # migrate_speed parameter in qm.conf - take the lower of the two.
874 my $bwlimit = PVE
::Storage
::get_bandwidth_limit
('migration', undef, $opt_bwlimit) // 0;
875 my $migrate_speed = $conf->{migrate_speed
} // $bwlimit;
876 # migrate_speed is in MB/s, bwlimit in KB/s
877 $migrate_speed *= 1024;
879 $migrate_speed = ($bwlimit < $migrate_speed) ?
$bwlimit : $migrate_speed;
881 # always set migrate speed (overwrite kvm default of 32m) we set a very high
882 # default of 8192m which is basically unlimited
883 $migrate_speed ||= ($defaults->{migrate_speed
} || 8192) * 1024;
885 # qmp takes migrate_speed in B/s.
886 $migrate_speed *= 1024;
887 $self->log('info', "migration speed limit: $migrate_speed B/s");
888 $qemu_migrate_params->{'max-bandwidth'} = int($migrate_speed);
890 my $migrate_downtime = $defaults->{migrate_downtime
};
891 $migrate_downtime = $conf->{migrate_downtime
} if defined($conf->{migrate_downtime
});
892 # migrate-set-parameters expects limit in ms
893 $migrate_downtime *= 1000;
894 $self->log('info', "migration downtime limit: $migrate_downtime ms");
895 $qemu_migrate_params->{'downtime-limit'} = int($migrate_downtime);
897 # set cachesize to 10% of the total memory
898 my $memory = $conf->{memory
} || $defaults->{memory
};
899 my $cachesize = int($memory * 1048576 / 10);
900 $cachesize = round_powerof2
($cachesize);
902 $self->log('info', "migration cachesize: $cachesize B");
903 $qemu_migrate_params->{'xbzrle-cache-size'} = int($cachesize);
905 $self->log('info', "set migration parameters");
907 mon_cmd
($vmid, "migrate-set-parameters", %{$qemu_migrate_params});
909 $self->log('info', "migrate-set-parameters error: $@") if $@;
911 if (PVE
::QemuServer
::vga_conf_has_spice
($conf->{vga
})) {
912 my $rpcenv = PVE
::RPCEnvironment
::get
();
913 my $authuser = $rpcenv->get_user();
915 my (undef, $proxyticket) = PVE
::AccessControl
::assemble_spice_ticket
($authuser, $vmid, $self->{node
});
917 my $filename = "/etc/pve/nodes/$self->{node}/pve-ssl.pem";
918 my $subject = PVE
::AccessControl
::read_x509_subject_spice
($filename);
920 $self->log('info', "spice client_migrate_info");
923 mon_cmd
($vmid, "client_migrate_info", protocol
=> 'spice',
924 hostname
=> $proxyticket, 'port' => 0, 'tls-port' => $spice_port,
925 'cert-subject' => $subject);
927 $self->log('info', "client_migrate_info error: $@") if $@;
931 $self->log('info', "start migrate command to $ruri");
933 mon_cmd
($vmid, "migrate", uri
=> $ruri);
936 $self->log('info', "migrate uri => $ruri failed: $merr") if $merr;
939 my $usleep = 1000000;
943 my $downtimecounter = 0;
946 my $avglstat = $lstat/$i if $lstat;
951 $stat = mon_cmd
($vmid, "query-migrate");
955 warn "query migrate failed: $err\n";
956 $self->log('info', "query migrate failed: $err");
957 if ($err_count <= 5) {
961 die "too many query migrate failures - aborting\n";
964 if (defined($stat->{status
}) && $stat->{status
} =~ m/^(setup)$/im) {
969 if (defined($stat->{status
}) && $stat->{status
} =~ m/^(active|completed|failed|cancelled)$/im) {
972 if ($stat->{status
} eq 'completed') {
973 my $delay = time() - $start;
975 my $mbps = sprintf "%.2f", $memory / $delay;
976 my $downtime = $stat->{downtime
} || 0;
977 $self->log('info', "migration speed: $mbps MB/s - downtime $downtime ms");
981 if ($stat->{status
} eq 'failed' || $stat->{status
} eq 'cancelled') {
982 $self->log('info', "migration status error: $stat->{status}");
986 if ($stat->{status
} ne 'active') {
987 $self->log('info', "migration status: $stat->{status}");
991 if ($stat->{ram
}->{transferred
} ne $lstat) {
992 my $trans = $stat->{ram
}->{transferred
} || 0;
993 my $rem = $stat->{ram
}->{remaining
} || 0;
994 my $total = $stat->{ram
}->{total
} || 0;
995 my $xbzrlecachesize = $stat->{"xbzrle-cache"}->{"cache-size"} || 0;
996 my $xbzrlebytes = $stat->{"xbzrle-cache"}->{"bytes"} || 0;
997 my $xbzrlepages = $stat->{"xbzrle-cache"}->{"pages"} || 0;
998 my $xbzrlecachemiss = $stat->{"xbzrle-cache"}->{"cache-miss"} || 0;
999 my $xbzrleoverflow = $stat->{"xbzrle-cache"}->{"overflow"} || 0;
1000 # reduce sleep if remainig memory is lower than the average transfer speed
1001 $usleep = 100000 if $avglstat && $rem < $avglstat;
1003 $self->log('info', "migration status: $stat->{status} (transferred ${trans}, " .
1004 "remaining ${rem}), total ${total})");
1006 if (${xbzrlecachesize
}) {
1007 $self->log('info', "migration xbzrle cachesize: ${xbzrlecachesize} transferred ${xbzrlebytes} pages ${xbzrlepages} cachemiss ${xbzrlecachemiss} overflow ${xbzrleoverflow}");
1010 if (($lastrem && $rem > $lastrem ) || ($rem == 0)) {
1015 if ($downtimecounter > 5) {
1016 $downtimecounter = 0;
1017 $migrate_downtime *= 2;
1018 $self->log('info', "auto-increased downtime to continue migration: $migrate_downtime ms");
1020 # migrate-set-parameters does not touch values not
1021 # specified, so this only changes downtime-limit
1022 mon_cmd
($vmid, "migrate-set-parameters", 'downtime-limit' => int($migrate_downtime));
1024 $self->log('info', "migrate-set-parameters error: $@") if $@;
1030 $lstat = $stat->{ram
}->{transferred
};
1034 die "unable to parse migration status '$stat->{status}' - aborting\n";
1039 sub phase2_cleanup
{
1040 my ($self, $vmid, $err) = @_;
1042 return if !$self->{errors
};
1043 $self->{phase2errors
} = 1;
1045 $self->log('info', "aborting phase 2 - cleanup resources");
1047 $self->log('info', "migrate_cancel");
1049 mon_cmd
($vmid, "migrate_cancel");
1051 $self->log('info', "migrate_cancel error: $@") if $@;
1053 my $conf = $self->{vmconf
};
1054 delete $conf->{lock};
1055 eval { PVE
::QemuConfig-
>write_config($vmid, $conf) };
1057 $self->log('err', $err);
1060 # cleanup ressources on target host
1061 if ($self->{storage_migration
}) {
1062 eval { PVE
::QemuServer
::qemu_blockjobs_cancel
($vmid, $self->{storage_migration_jobs
}) };
1064 $self->log('err', $err);
1068 eval { $self->cleanup_bitmaps() };
1070 $self->log('err', $err);
1073 my $nodename = PVE
::INotify
::nodename
();
1075 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'stop', $vmid, '--skiplock', '--migratedfrom', $nodename];
1076 eval{ PVE
::Tools
::run_command
($cmd, outfunc
=> sub {}, errfunc
=> sub {}) };
1078 $self->log('err', $err);
1079 $self->{errors
} = 1;
1082 # cleanup after stopping, otherwise disks might be in-use by target VM!
1083 eval { PVE
::QemuMigrate
::cleanup_remotedisks
($self) };
1085 $self->log('err', $err);
1089 if ($self->{tunnel
}) {
1090 eval { finish_tunnel
($self, $self->{tunnel
}); };
1092 $self->log('err', $err);
1093 $self->{errors
} = 1;
1099 my ($self, $vmid) = @_;
1101 my $volids = $self->{volumes
};
1102 return if $self->{phase2errors
};
1104 # destroy local copies
1105 foreach my $volid (@$volids) {
1106 eval { PVE
::Storage
::vdisk_free
($self->{storecfg
}, $volid); };
1108 $self->log('err', "removing local copy of '$volid' failed - $err");
1109 $self->{errors
} = 1;
1110 last if $err =~ /^interrupted by signal$/;
1115 sub phase3_cleanup
{
1116 my ($self, $vmid, $err) = @_;
1118 my $conf = $self->{vmconf
};
1119 return if $self->{phase2errors
};
1121 my $tunnel = $self->{tunnel
};
1123 if ($self->{storage_migration
}) {
1124 # finish block-job with block-job-cancel, to disconnect source VM from NBD
1125 # to avoid it trying to re-establish it. We are in blockjob ready state,
1126 # thus, this command changes to it to blockjob complete (see qapi docs)
1127 eval { PVE
::QemuServer
::qemu_drive_mirror_monitor
($vmid, undef, $self->{storage_migration_jobs
}, 'cancel'); };
1130 eval { PVE
::QemuServer
::qemu_blockjobs_cancel
($vmid, $self->{storage_migration_jobs
}) };
1131 eval { PVE
::QemuMigrate
::cleanup_remotedisks
($self) };
1132 die "Failed to complete storage migration: $err\n";
1136 if ($self->{volume_map
}) {
1137 PVE
::QemuConfig-
>update_volume_ids($conf, $self->{volume_map
});
1138 PVE
::QemuConfig-
>write_config($vmid, $conf);
1141 # transfer replication state before move config
1142 $self->transfer_replication_state() if $self->{replicated_volumes
};
1144 # move config to remote node
1145 my $conffile = PVE
::QemuConfig-
>config_file($vmid);
1146 my $newconffile = PVE
::QemuConfig-
>config_file($vmid, $self->{node
});
1148 die "Failed to move config to node '$self->{node}' - rename failed: $!\n"
1149 if !rename($conffile, $newconffile);
1151 $self->switch_replication_job_target() if $self->{replicated_volumes
};
1153 if ($self->{livemigration
}) {
1154 if ($self->{storage_migration
}) {
1155 # stop nbd server on remote vm - requirement for resume since 2.9
1156 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'nbdstop', $vmid];
1158 eval{ PVE
::Tools
::run_command
($cmd, outfunc
=> sub {}, errfunc
=> sub {}) };
1160 $self->log('err', $err);
1161 $self->{errors
} = 1;
1165 # config moved and nbd server stopped - now we can resume vm on target
1166 if ($tunnel && $tunnel->{version
} && $tunnel->{version
} >= 1) {
1168 $self->write_tunnel($tunnel, 30, "resume $vmid");
1171 $self->log('err', $err);
1172 $self->{errors
} = 1;
1175 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'resume', $vmid, '--skiplock', '--nocheck'];
1178 $self->log('err', $line);
1180 eval { PVE
::Tools
::run_command
($cmd, outfunc
=> sub {}, errfunc
=> $logf); };
1182 $self->log('err', $err);
1183 $self->{errors
} = 1;
1187 if ($self->{storage_migration
} && PVE
::QemuServer
::parse_guest_agent
($conf)->{fstrim_cloned_disks
} && $self->{running
}) {
1188 my $cmd = [@{$self->{rem_ssh
}}, 'qm', 'guest', 'cmd', $vmid, 'fstrim'];
1189 eval{ PVE
::Tools
::run_command
($cmd, outfunc
=> sub {}, errfunc
=> sub {}) };
1193 # close tunnel on successful migration, on error phase2_cleanup closed it
1195 eval { finish_tunnel
($self, $tunnel); };
1197 $self->log('err', $err);
1198 $self->{errors
} = 1;
1204 if (PVE
::QemuServer
::vga_conf_has_spice
($conf->{vga
}) && $self->{running
}) {
1205 $self->log('info', "Waiting for spice server migration");
1207 my $res = mon_cmd
($vmid, 'query-spice');
1208 last if int($res->{'migrated'}) == 1;
1209 last if $timer > 50;
1216 # always stop local VM
1217 eval { PVE
::QemuServer
::vm_stop
($self->{storecfg
}, $vmid, 1, 1); };
1219 $self->log('err', "stopping vm failed - $err");
1220 $self->{errors
} = 1;
1223 # always deactivate volumes - avoid lvm LVs to be active on several nodes
1225 my $vollist = PVE
::QemuServer
::get_vm_volumes
($conf);
1226 PVE
::Storage
::deactivate_volumes
($self->{storecfg
}, $vollist);
1229 $self->log('err', $err);
1230 $self->{errors
} = 1;
1233 if($self->{storage_migration
}) {
1234 # destroy local copies
1235 my $volids = $self->{online_local_volumes
};
1237 foreach my $volid (@$volids) {
1238 # keep replicated volumes!
1239 next if $self->{replicated_volumes
}->{$volid};
1241 eval { PVE
::Storage
::vdisk_free
($self->{storecfg
}, $volid); };
1243 $self->log('err', "removing local copy of '$volid' failed - $err");
1244 $self->{errors
} = 1;
1245 last if $err =~ /^interrupted by signal$/;
1251 # clear migrate lock
1252 my $cmd = [ @{$self->{rem_ssh
}}, 'qm', 'unlock', $vmid ];
1253 $self->cmd_logerr($cmd, errmsg
=> "failed to clear migrate lock");
1257 my ($self, $vmid) = @_;
1262 sub round_powerof2
{
1263 return 1 if $_[0] < 2;
1264 return 2 << int(log($_[0]-1)/log(2));