1 package PVE
::Storage
::ZFSPoolPlugin
;
7 use PVE
::Tools
qw(run_command);
8 use PVE
::Storage
::Plugin
;
9 use PVE
::RPCEnvironment
;
12 use base
qw(PVE::Storage::Plugin);
20 content
=> [ {images
=> 1, rootdir
=> 1}, {images
=> 1 , rootdir
=> 1}],
21 format
=> [ { raw
=> 1, subvol
=> 1 } , 'raw' ],
28 description
=> "block size",
32 description
=> "use sparse volumes",
40 pool
=> { fixed
=> 1 },
41 blocksize
=> { optional
=> 1 },
42 sparse
=> { optional
=> 1 },
43 nodes
=> { optional
=> 1 },
44 disable
=> { optional
=> 1 },
45 content
=> { optional
=> 1 },
46 bwlimit
=> { optional
=> 1 },
50 # static zfs helper methods
57 if ($text =~ m/^(\d+(\.\d+)?)([TGMK])?$/) {
59 my ($size, $reminder, $unit) = ($1, $2, $3);
64 } elsif ($unit eq 'M') {
66 } elsif ($unit eq 'G') {
67 $size *= 1024*1024*1024;
68 } elsif ($unit eq 'T') {
69 $size *= 1024*1024*1024*1024;
71 die "got unknown zfs size unit '$unit'\n";
83 warn "unable to parse zfs size '$text'\n";
88 sub zfs_parse_zvol_list
{
93 return $list if !$text;
95 my @lines = split /\n/, $text;
96 foreach my $line (@lines) {
97 my ($dataset, $size, $origin, $type, $refquota) = split(/\s+/, $line);
98 next if !($type eq 'volume' || $type eq 'filesystem');
101 my @parts = split /\//, $dataset;
102 next if scalar(@parts) < 2; # we need pool/name
103 my $name = pop @parts;
104 my $pool = join('/', @parts);
106 next unless $name =~ m!^(vm|base|subvol|basevol)-(\d+)-(\S+)$!;
109 $zvol->{pool
} = $pool;
110 $zvol->{name
} = $name;
111 if ($type eq 'filesystem') {
112 if ($refquota eq 'none') {
115 $zvol->{size
} = zfs_parse_size
($refquota);
117 $zvol->{format
} = 'subvol';
119 $zvol->{size
} = zfs_parse_size
($size);
120 $zvol->{format
} = 'raw';
122 if ($origin !~ /^-$/) {
123 $zvol->{origin
} = $origin;
132 my ($class, $volname) = @_;
134 if ($volname =~ m/^(((base|basevol)-(\d+)-\S+)\/)?
((base
|basevol
|vm
|subvol
)-(\d
+)-\S
+)$/) {
135 my $format = ($6 eq 'subvol' || $6 eq 'basevol') ?
'subvol' : 'raw';
136 my $isBase = ($6 eq 'base' || $6 eq 'basevol');
137 return ('images', $5, $7, $2, $4, $isBase, $format);
140 die "unable to parse zfs volume name '$volname'\n";
143 # virtual zfs methods (subclass can overwrite them)
146 my ($class, $scfg, $volname, $storeid, $snapname) = @_;
148 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
152 if ($vtype eq "images") {
153 if ($name =~ m/^subvol-/ || $name =~ m/^basevol-/) {
154 # fixme: we currently assume standard mount point?!
155 $path = "/$scfg->{pool}/$name";
157 $path = "/dev/zvol/$scfg->{pool}/$name";
159 $path .= "\@$snapname" if defined($snapname);
161 die "$vtype is not allowed in ZFSPool!";
164 return ($path, $vmid, $vtype);
168 my ($class, $scfg, $timeout, $method, @params) = @_;
172 if ($method eq 'zpool_list') {
173 push @$cmd, 'zpool', 'list';
174 } elsif ($method eq 'zpool_import') {
175 push @$cmd, 'zpool', 'import';
176 $default_timeout = 15 if $default_timeout < 15;
178 push @$cmd, 'zfs', $method;
183 my $output = sub { $msg .= "$_[0]\n" };
185 $timeout = PVE
::RPCEnvironment-
>is_worker() ?
60*60 : 5 if !$timeout;
187 run_command
($cmd, errmsg
=> "zfs error", outfunc
=> $output, timeout
=> $timeout);
193 my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_;
199 die "illegal name '$volname' - should be 'vm-$vmid-*'\n"
200 if $volname && $volname !~ m/^vm-$vmid-/;
201 $volname = $class->zfs_find_free_diskname($storeid, $scfg, $vmid, $fmt)
204 $class->zfs_create_zvol($scfg, $volname, $size);
205 my $devname = "/dev/zvol/$scfg->{pool}/$volname";
207 my $timeout = PVE
::RPCEnvironment-
>is_worker() ?
60*5 : 10;
208 for (my $i = 1; $i <= $timeout; $i++) {
210 die "Timeout: no zvol after $timeout sec found.\n"
215 } elsif ( $fmt eq 'subvol') {
217 die "illegal name '$volname' - should be 'subvol-$vmid-*'\n"
218 if $volname && $volname !~ m/^subvol-$vmid-/;
219 $volname = $class->zfs_find_free_diskname($storeid, $scfg, $vmid, $fmt)
222 die "illegal name '$volname' - should be 'subvol-$vmid-*'\n"
223 if $volname !~ m/^subvol-$vmid-/;
225 $class->zfs_create_subvol($scfg, $volname, $size);
228 die "unsupported format '$fmt'";
235 my ($class, $storeid, $scfg, $volname, $isBase) = @_;
237 my (undef, $name, undef) = $class->parse_volname($volname);
239 $class->zfs_delete_zvol($scfg, $name);
245 my ($class, $storeid, $scfg, $vmid, $vollist, $cache) = @_;
247 $cache->{zfs
} = $class->zfs_list_zvol($scfg) if !$cache->{zfs
};
248 my $zfspool = $scfg->{pool
};
251 if (my $dat = $cache->{zfs
}->{$zfspool}) {
253 foreach my $image (keys %$dat) {
255 my $info = $dat->{$image};
257 my $volname = $info->{name
};
258 my $parent = $info->{parent
};
259 my $owner = $info->{vmid
};
261 if ($parent && $parent =~ m/^(\S+)\@__base__$/) {
262 my ($basename) = ($1);
263 $info->{volid
} = "$storeid:$basename/$volname";
265 $info->{volid
} = "$storeid:$volname";
269 my $found = grep { $_ eq $info->{volid
} } @$vollist;
272 next if defined ($vmid) && ($owner ne $vmid);
281 sub zfs_get_pool_stats
{
282 my ($class, $scfg) = @_;
287 my $text = $class->zfs_request($scfg, undef, 'get', '-o', 'value', '-Hp',
288 'available,used', $scfg->{pool
});
290 my @lines = split /\n/, $text;
292 if($lines[0] =~ /^(\d+)$/) {
296 if($lines[1] =~ /^(\d+)$/) {
300 return ($available, $used);
303 sub zfs_create_zvol
{
304 my ($class, $scfg, $zvol, $size) = @_;
306 # always align size to 1M as workaround until
307 # https://github.com/zfsonlinux/zfs/issues/8541 is solved
308 my $padding = (1024 - $size % 1024) % 1024;
309 $size = $size + $padding;
311 my $cmd = ['create'];
313 push @$cmd, '-s' if $scfg->{sparse
};
315 push @$cmd, '-b', $scfg->{blocksize
} if $scfg->{blocksize
};
317 push @$cmd, '-V', "${size}k", "$scfg->{pool}/$zvol";
319 $class->zfs_request($scfg, undef, @$cmd);
322 sub zfs_create_subvol
{
323 my ($class, $scfg, $volname, $size) = @_;
325 my $dataset = "$scfg->{pool}/$volname";
327 my $cmd = ['create', '-o', 'acltype=posixacl', '-o', 'xattr=sa',
328 '-o', "refquota=${size}k", $dataset];
330 $class->zfs_request($scfg, undef, @$cmd);
333 sub zfs_delete_zvol
{
334 my ($class, $scfg, $zvol) = @_;
338 for (my $i = 0; $i < 6; $i++) {
340 eval { $class->zfs_request($scfg, undef, 'destroy', '-r', "$scfg->{pool}/$zvol"); };
342 if ($err =~ m/^zfs error:(.*): dataset is busy.*/) {
344 } elsif ($err =~ m/^zfs error:.*: dataset does not exist.*$/) {
359 my ($class, $scfg) = @_;
361 my $text = $class->zfs_request($scfg, 10, 'list', '-o', 'name,volsize,origin,type,refquota', '-t', 'volume,filesystem', '-Hr');
362 my $zvols = zfs_parse_zvol_list
($text);
363 return undef if !$zvols;
366 foreach my $zvol (@$zvols) {
367 my $pool = $zvol->{pool
};
368 my $name = $zvol->{name
};
369 my $parent = $zvol->{origin
};
370 if($zvol->{origin
} && $zvol->{origin
} =~ m/^$scfg->{pool}\/(\S
+)$/){
374 $list->{$pool}->{$name} = {
376 size
=> $zvol->{size
},
378 format
=> $zvol->{format
},
379 vmid
=> $zvol->{owner
},
386 sub zfs_find_free_diskname
{
387 my ($class, $storeid, $scfg, $vmid, $format) = @_;
389 my $volumes = $class->zfs_list_zvol($scfg);
390 my $dat = $volumes->{$scfg->{pool
}};
392 my $disk_list = [ keys %$dat ];
393 return PVE
::Storage
::Plugin
::get_next_vm_diskname
($disk_list, $storeid, $vmid, $format, $scfg);
396 sub zfs_get_latest_snapshot
{
397 my ($class, $scfg, $volname) = @_;
399 my $vname = ($class->parse_volname($volname))[1];
401 # abort rollback if snapshot is not the latest
402 my @params = ('-t', 'snapshot', '-o', 'name', '-s', 'creation');
403 my $text = $class->zfs_request($scfg, undef, 'list', @params);
404 my @snapshots = split(/\n/, $text);
407 foreach (@snapshots) {
408 if (/$scfg->{pool}\/$vname/) {
418 my ($class, $storeid, $scfg, $cache) = @_;
426 ($free, $used) = $class->zfs_get_pool_stats($scfg);
428 $total = $free + $used;
432 return ($total, $free, $used, $active);
435 sub volume_size_info
{
436 my ($class, $scfg, $storeid, $volname, $timeout) = @_;
438 my (undef, $vname, undef, undef, undef, undef, $format) =
439 $class->parse_volname($volname);
441 my $attr = $format eq 'subvol' ?
'refquota' : 'volsize';
442 my $text = $class->zfs_request($scfg, undef, 'get', '-Hp', $attr, "$scfg->{pool}/$vname");
443 if ($text =~ /\s$attr\s(\d+)\s/) {
447 die "Could not get zfs volume size\n";
450 sub volume_snapshot
{
451 my ($class, $scfg, $storeid, $volname, $snap) = @_;
453 my $vname = ($class->parse_volname($volname))[1];
455 $class->zfs_request($scfg, undef, 'snapshot', "$scfg->{pool}/$vname\@$snap");
458 sub volume_snapshot_delete
{
459 my ($class, $scfg, $storeid, $volname, $snap, $running) = @_;
461 my $vname = ($class->parse_volname($volname))[1];
463 $class->deactivate_volume($storeid, $scfg, $vname, $snap, {});
464 $class->zfs_request($scfg, undef, 'destroy', "$scfg->{pool}/$vname\@$snap");
467 sub volume_snapshot_rollback
{
468 my ($class, $scfg, $storeid, $volname, $snap) = @_;
470 my $vname = ($class->parse_volname($volname))[1];
472 $class->zfs_request($scfg, undef, 'rollback', "$scfg->{pool}/$vname\@$snap");
475 sub volume_rollback_is_possible
{
476 my ($class, $scfg, $storeid, $volname, $snap) = @_;
478 my $recentsnap = $class->zfs_get_latest_snapshot($scfg, $volname);
479 if ($snap ne $recentsnap) {
480 die "can't rollback, more recent snapshots exist\n";
486 sub volume_snapshot_list
{
487 my ($class, $scfg, $storeid, $volname) = @_;
489 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
491 my $zpath = "$scfg->{pool}/$name";
495 my $cmd = ['zfs', 'list', '-r', '-H', '-S', 'name', '-t', 'snap', '-o',
501 if ($line =~ m/^\Q$zpath\E@(.*)$/) {
506 eval { run_command
( [$cmd], outfunc
=> $outfunc , errfunc
=> sub{}); };
508 # return an empty array if dataset does not exist.
512 sub activate_storage
{
513 my ($class, $storeid, $scfg, $cache) = @_;
515 # Note: $scfg->{pool} can include dataset <pool>/<dataset>
516 my $pool = $scfg->{pool
};
519 my @param = ('-o', 'name', '-H', "$pool");
522 $res = $class->zfs_request($scfg, undef, 'zpool_list', @param);
525 if ($err || !defined($res) || $res !~ $pool) {
527 @param = ('-d', '/dev/disk/by-id/', '-o', 'cachefile=none', "$pool");
528 $class->zfs_request($scfg, undef, 'zpool_import', @param);
532 die "could not activate storage '$storeid', $@\n";
538 sub deactivate_storage
{
539 my ($class, $storeid, $scfg, $cache) = @_;
543 sub activate_volume
{
544 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
548 sub deactivate_volume
{
549 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
554 my ($class, $scfg, $storeid, $volname, $vmid, $snap) = @_;
556 $snap ||= '__base__';
558 my ($vtype, $basename, $basevmid, undef, undef, $isBase, $format) =
559 $class->parse_volname($volname);
561 die "clone_image only works on base images\n" if !$isBase;
563 my $name = $class->zfs_find_free_diskname($storeid, $scfg, $vmid, $format);
565 if ($format eq 'subvol') {
566 my $size = $class->zfs_request($scfg, undef, 'list', '-H', '-o', 'refquota', "$scfg->{pool}/$basename");
568 $class->zfs_request($scfg, undef, 'clone', "$scfg->{pool}/$basename\@$snap", "$scfg->{pool}/$name", '-o', "refquota=$size");
570 $class->zfs_request($scfg, undef, 'clone', "$scfg->{pool}/$basename\@$snap", "$scfg->{pool}/$name");
573 return "$basename/$name";
577 my ($class, $storeid, $scfg, $volname) = @_;
579 my $snap = '__base__';
581 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase, $format) =
582 $class->parse_volname($volname);
584 die "create_base not possible with base image\n" if $isBase;
587 if ( $format eq 'subvol' ) {
588 $newname =~ s/^subvol-/basevol-/;
590 $newname =~ s/^vm-/base-/;
592 my $newvolname = $basename ?
"$basename/$newname" : "$newname";
594 $class->zfs_request($scfg, undef, 'rename', "$scfg->{pool}/$name", "$scfg->{pool}/$newname");
596 my $running = undef; #fixme : is create_base always offline ?
598 $class->volume_snapshot($scfg, $storeid, $newname, $snap, $running);
604 my ($class, $scfg, $storeid, $volname, $size, $running) = @_;
606 my $new_size = int($size/1024);
608 my (undef, $vname, undef, undef, undef, undef, $format) =
609 $class->parse_volname($volname);
611 my $attr = $format eq 'subvol' ?
'refquota' : 'volsize';
613 $class->zfs_request($scfg, undef, 'set', "$attr=${new_size}k", "$scfg->{pool}/$vname");
618 sub storage_can_replicate
{
619 my ($class, $scfg, $storeid, $format) = @_;
621 return 1 if $format eq 'raw' || $format eq 'subvol';
626 sub volume_has_feature
{
627 my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_;
630 snapshot
=> { current
=> 1, snap
=> 1},
631 clone
=> { base
=> 1},
632 template
=> { current
=> 1},
633 copy
=> { base
=> 1, current
=> 1},
634 sparseinit
=> { base
=> 1, current
=> 1},
635 replicate
=> { base
=> 1, current
=> 1},
638 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
639 $class->parse_volname($volname);
646 $key = $isBase ?
'base' : 'current';
649 return 1 if $features->{$feature}->{$key};
655 my ($class, $scfg, $storeid, $fh, $volname, $format, $snapshot, $base_snapshot, $with_snapshots) = @_;
657 die "unsupported export stream format for $class: $format\n"
660 die "$class storage can only export snapshots\n"
661 if !defined($snapshot);
663 my $dataset = ($class->parse_volname($volname))[1];
665 my $fd = fileno($fh);
666 die "internal error: invalid file handle for volume_export\n"
670 # For zfs we always create a replication stream (-R) which means the remote
671 # side will always delete non-existing source snapshots. This should work
672 # for all our use cases.
673 my $cmd = ['zfs', 'send', '-Rpv'];
674 if (defined($base_snapshot)) {
675 my $arg = $with_snapshots ?
'-I' : '-i';
676 push @$cmd, $arg, $base_snapshot;
678 push @$cmd, '--', "$scfg->{pool}/$dataset\@$snapshot";
680 run_command
($cmd, output
=> $fd);
685 sub volume_export_formats
{
686 my ($class, $scfg, $storeid, $volname, $snapshot, $base_snapshot, $with_snapshots) = @_;
688 my @formats = ('zfs');
690 # push @formats, 'fies' if $volname !~ /^(?:basevol|subvol)-/;
691 # push @formats, 'raw' if !$base_snapshot && !$with_snapshots;
696 my ($class, $scfg, $storeid, $fh, $volname, $format, $base_snapshot, $with_snapshots) = @_;
698 die "unsupported import stream format for $class: $format\n"
701 my $fd = fileno($fh);
702 die "internal error: invalid file handle for volume_import\n"
705 my $dataset = ($class->parse_volname($volname))[1];
706 my $zfspath = "$scfg->{pool}/$dataset";
707 my $suffix = defined($base_snapshot) ?
"\@$base_snapshot" : '';
708 my $exists = 0 == run_command
(['zfs', 'get', '-H', 'name', $zfspath.$suffix],
709 noerr
=> 1, errfunc
=> sub {});
710 if (defined($base_snapshot)) {
711 die "base snapshot '$zfspath\@$base_snapshot' doesn't exist\n" if !$exists;
713 die "volume '$zfspath' already exists\n" if $exists;
716 eval { run_command
(['zfs', 'recv', '-F', '--', $zfspath], input
=> "<&$fd") };
718 if (defined($base_snapshot)) {
719 eval { run_command
(['zfs', 'rollback', '-r', '--', "$zfspath\@$base_snapshot"]) };
721 eval { run_command
(['zfs', 'destroy', '-r', '--', $zfspath]) };
729 sub volume_import_formats
{
730 my ($class, $scfg, $storeid, $volname, $base_snapshot, $with_snapshots) = @_;
732 return $class->volume_export_formats($scfg, $storeid, $volname, undef, $base_snapshot, $with_snapshots);