1 package PVE
::Storage
::ZFSPoolPlugin
;
11 use PVE
::RPCEnvironment
;
12 use PVE
::Storage
::Plugin
;
13 use PVE
::Tools
qw(run_command);
15 use base
qw(PVE::Storage::Plugin);
23 content
=> [ {images
=> 1, rootdir
=> 1}, {images
=> 1 , rootdir
=> 1}],
24 format
=> [ { raw
=> 1, subvol
=> 1 } , 'raw' ],
31 description
=> "block size",
35 description
=> "use sparse volumes",
39 description
=> "mount point",
40 type
=> 'string', format
=> 'pve-storage-path',
47 pool
=> { fixed
=> 1 },
48 blocksize
=> { optional
=> 1 },
49 sparse
=> { optional
=> 1 },
50 nodes
=> { optional
=> 1 },
51 disable
=> { optional
=> 1 },
52 content
=> { optional
=> 1 },
53 bwlimit
=> { optional
=> 1 },
54 mountpoint
=> { optional
=> 1 },
58 # static zfs helper methods
60 sub zfs_parse_zvol_list
{
65 return $list if !$text;
67 my @lines = split /\n/, $text;
68 foreach my $line (@lines) {
69 my ($dataset, $size, $origin, $type, $refquota) = split(/\s+/, $line);
70 next if !($type eq 'volume' || $type eq 'filesystem');
73 my @parts = split /\//, $dataset;
74 next if scalar(@parts) < 2; # we need pool/name
75 my $name = pop @parts;
76 my $pool = join('/', @parts);
78 next unless $name =~ m!^(vm|base|subvol|basevol)-(\d+)-(\S+)$!;
81 $zvol->{pool
} = $pool;
82 $zvol->{name
} = $name;
83 if ($type eq 'filesystem') {
84 if ($refquota eq 'none') {
87 $zvol->{size
} = $refquota + 0;
89 $zvol->{format
} = 'subvol';
91 $zvol->{size
} = $size + 0;
92 $zvol->{format
} = 'raw';
94 if ($origin !~ /^-$/) {
95 $zvol->{origin
} = $origin;
104 my ($class, $volname) = @_;
106 if ($volname =~ m/^(((base|basevol)-(\d+)-\S+)\/)?
((base
|basevol
|vm
|subvol
)-(\d
+)-\S
+)$/) {
107 my $format = ($6 eq 'subvol' || $6 eq 'basevol') ?
'subvol' : 'raw';
108 my $isBase = ($6 eq 'base' || $6 eq 'basevol');
109 return ('images', $5, $7, $2, $4, $isBase, $format);
112 die "unable to parse zfs volume name '$volname'\n";
115 # virtual zfs methods (subclass can overwrite them)
118 my ($class, $storeid, $scfg, %param) = @_;
120 my $cfg_mountpoint = $scfg->{mountpoint
};
122 # ignore failure, pool might currently not be imported
125 my $res = $class->zfs_get_properties($scfg, 'mountpoint', $scfg->{pool
}, 1);
126 $mountpoint = PVE
::Storage
::Plugin
::verify_path
($res, 1) if defined($res);
129 if (defined($cfg_mountpoint)) {
130 if (defined($mountpoint) && !($cfg_mountpoint =~ m
|^\Q
$mountpoint\E/?$|)) {
131 warn "warning for $storeid - mountpoint: $cfg_mountpoint " .
132 "does not match current mount point: $mountpoint\n";
135 $scfg->{mountpoint
} = $mountpoint;
142 my ($class, $scfg, $volname, $storeid, $snapname) = @_;
144 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
147 my $mountpoint = $scfg->{mountpoint
} // "/$scfg->{pool}";
149 if ($vtype eq "images") {
150 if ($name =~ m/^subvol-/ || $name =~ m/^basevol-/) {
151 $path = "$mountpoint/$name";
153 $path = "/dev/zvol/$scfg->{pool}/$name";
155 $path .= "\@$snapname" if defined($snapname);
157 die "$vtype is not allowed in ZFSPool!";
160 return ($path, $vmid, $vtype);
164 my ($class, $scfg, $timeout, $method, @params) = @_;
168 if ($method eq 'zpool_list') {
169 push @$cmd, 'zpool', 'list';
170 } elsif ($method eq 'zpool_import') {
171 push @$cmd, 'zpool', 'import';
172 $timeout = 15 if !$timeout || $timeout < 15;
174 push @$cmd, 'zfs', $method;
179 my $output = sub { $msg .= "$_[0]\n" };
181 $timeout = PVE
::RPCEnvironment-
>is_worker() ?
60*60 : 5 if !$timeout;
183 run_command
($cmd, errmsg
=> "zfs error", outfunc
=> $output, timeout
=> $timeout);
188 sub zfs_wait_for_zvol_link
{
189 my ($class, $scfg, $volname, $timeout) = @_;
191 my $default_timeout = PVE
::RPCEnvironment-
>is_worker() ?
60*5 : 10;
192 $timeout = $default_timeout if !defined($timeout);
194 my ($devname, undef, undef) = $class->path($scfg, $volname);
196 for (my $i = 1; $i <= $timeout; $i++) {
198 die "timeout: no zvol device link for '$volname' found after $timeout sec found.\n"
206 my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_;
212 die "illegal name '$volname' - should be 'vm-$vmid-*'\n"
213 if $volname && $volname !~ m/^vm-$vmid-/;
214 $volname = $class->find_free_diskname($storeid, $scfg, $vmid, $fmt)
217 $class->zfs_create_zvol($scfg, $volname, $size);
218 $class->zfs_wait_for_zvol_link($scfg, $volname);
220 } elsif ( $fmt eq 'subvol') {
222 die "illegal name '$volname' - should be 'subvol-$vmid-*'\n"
223 if $volname && $volname !~ m/^subvol-$vmid-/;
224 $volname = $class->find_free_diskname($storeid, $scfg, $vmid, $fmt)
227 die "illegal name '$volname' - should be 'subvol-$vmid-*'\n"
228 if $volname !~ m/^subvol-$vmid-/;
230 $class->zfs_create_subvol($scfg, $volname, $size);
233 die "unsupported format '$fmt'";
240 my ($class, $storeid, $scfg, $volname, $isBase) = @_;
242 my (undef, $name, undef) = $class->parse_volname($volname);
244 $class->zfs_delete_zvol($scfg, $name);
250 my ($class, $storeid, $scfg, $vmid, $vollist, $cache) = @_;
252 $cache->{zfs
} = $class->zfs_list_zvol($scfg) if !$cache->{zfs
};
253 my $zfspool = $scfg->{pool
};
256 if (my $dat = $cache->{zfs
}->{$zfspool}) {
258 foreach my $image (keys %$dat) {
260 my $info = $dat->{$image};
262 my $volname = $info->{name
};
263 my $parent = $info->{parent
};
264 my $owner = $info->{vmid
};
266 if ($parent && $parent =~ m/^(\S+)\@__base__$/) {
267 my ($basename) = ($1);
268 $info->{volid
} = "$storeid:$basename/$volname";
270 $info->{volid
} = "$storeid:$volname";
274 my $found = grep { $_ eq $info->{volid
} } @$vollist;
277 next if defined ($vmid) && ($owner ne $vmid);
286 sub zfs_get_properties
{
287 my ($class, $scfg, $properties, $dataset, $timeout) = @_;
289 my $result = $class->zfs_request($scfg, $timeout, 'get', '-o', 'value',
290 '-Hp', $properties, $dataset);
291 my @values = split /\n/, $result;
292 return wantarray ?
@values : $values[0];
295 sub zfs_get_pool_stats
{
296 my ($class, $scfg) = @_;
301 my @lines = $class->zfs_get_properties($scfg, 'available,used', $scfg->{pool
});
303 if($lines[0] =~ /^(\d+)$/) {
307 if($lines[1] =~ /^(\d+)$/) {
311 return ($available, $used);
314 sub zfs_create_zvol
{
315 my ($class, $scfg, $zvol, $size) = @_;
317 # always align size to 1M as workaround until
318 # https://github.com/zfsonlinux/zfs/issues/8541 is solved
319 my $padding = (1024 - $size % 1024) % 1024;
320 $size = $size + $padding;
322 my $cmd = ['create'];
324 push @$cmd, '-s' if $scfg->{sparse
};
326 push @$cmd, '-b', $scfg->{blocksize
} if $scfg->{blocksize
};
328 push @$cmd, '-V', "${size}k", "$scfg->{pool}/$zvol";
330 $class->zfs_request($scfg, undef, @$cmd);
333 sub zfs_create_subvol
{
334 my ($class, $scfg, $volname, $size) = @_;
336 my $dataset = "$scfg->{pool}/$volname";
338 my $cmd = ['create', '-o', 'acltype=posixacl', '-o', 'xattr=sa',
339 '-o', "refquota=${size}k", $dataset];
341 $class->zfs_request($scfg, undef, @$cmd);
344 sub zfs_delete_zvol
{
345 my ($class, $scfg, $zvol) = @_;
349 for (my $i = 0; $i < 6; $i++) {
351 eval { $class->zfs_request($scfg, undef, 'destroy', '-r', "$scfg->{pool}/$zvol"); };
353 if ($err =~ m/^zfs error:(.*): dataset is busy.*/) {
355 } elsif ($err =~ m/^zfs error:.*: dataset does not exist.*$/) {
370 my ($class, $scfg) = @_;
372 my $text = $class->zfs_request($scfg, 10, 'list', '-o', 'name,volsize,origin,type,refquota', '-t', 'volume,filesystem', '-Hrp');
373 my $zvols = zfs_parse_zvol_list
($text);
374 return undef if !$zvols;
377 foreach my $zvol (@$zvols) {
378 my $pool = $zvol->{pool
};
379 my $name = $zvol->{name
};
380 my $parent = $zvol->{origin
};
381 if($zvol->{origin
} && $zvol->{origin
} =~ m/^$scfg->{pool}\/(\S
+)$/){
385 $list->{$pool}->{$name} = {
387 size
=> $zvol->{size
},
389 format
=> $zvol->{format
},
390 vmid
=> $zvol->{owner
},
397 sub zfs_get_latest_snapshot
{
398 my ($class, $scfg, $volname) = @_;
400 my $vname = ($class->parse_volname($volname))[1];
402 # abort rollback if snapshot is not the latest
403 my @params = ('-t', 'snapshot', '-o', 'name', '-s', 'creation');
404 my $text = $class->zfs_request($scfg, undef, 'list', @params);
405 my @snapshots = split(/\n/, $text);
408 foreach (@snapshots) {
409 if (/$scfg->{pool}\/$vname/) {
419 my ($class, $storeid, $scfg, $cache) = @_;
427 ($free, $used) = $class->zfs_get_pool_stats($scfg);
429 $total = $free + $used;
433 return ($total, $free, $used, $active);
436 sub volume_size_info
{
437 my ($class, $scfg, $storeid, $volname, $timeout) = @_;
439 my (undef, $vname, undef, undef, undef, undef, $format) =
440 $class->parse_volname($volname);
442 my $attr = $format eq 'subvol' ?
'refquota' : 'volsize';
443 my $value = $class->zfs_get_properties($scfg, $attr, "$scfg->{pool}/$vname");
444 if ($value =~ /^(\d+)$/) {
448 die "Could not get zfs volume size\n";
451 sub volume_snapshot
{
452 my ($class, $scfg, $storeid, $volname, $snap) = @_;
454 my $vname = ($class->parse_volname($volname))[1];
456 $class->zfs_request($scfg, undef, 'snapshot', "$scfg->{pool}/$vname\@$snap");
459 sub volume_snapshot_delete
{
460 my ($class, $scfg, $storeid, $volname, $snap, $running) = @_;
462 my $vname = ($class->parse_volname($volname))[1];
464 $class->deactivate_volume($storeid, $scfg, $vname, $snap, {});
465 $class->zfs_request($scfg, undef, 'destroy', "$scfg->{pool}/$vname\@$snap");
468 sub volume_snapshot_rollback
{
469 my ($class, $scfg, $storeid, $volname, $snap) = @_;
471 my (undef, $vname, undef, undef, undef, undef, $format) = $class->parse_volname($volname);
473 my $msg = $class->zfs_request($scfg, undef, 'rollback', "$scfg->{pool}/$vname\@$snap");
475 # we have to unmount rollbacked subvols, to invalidate wrong kernel
476 # caches, they get mounted in activate volume again
477 # see zfs bug #10931 https://github.com/openzfs/zfs/issues/10931
478 if ($format eq 'subvol') {
479 $class->zfs_request($scfg, undef, 'unmount', "$scfg->{pool}/$vname");
485 sub volume_rollback_is_possible
{
486 my ($class, $scfg, $storeid, $volname, $snap) = @_;
488 my $recentsnap = $class->zfs_get_latest_snapshot($scfg, $volname);
490 die "can't rollback, no snapshots exist at all\n"
491 if !defined($recentsnap);
493 die "can't rollback, '$snap' is not most recent snapshot\n"
494 if $snap ne $recentsnap;
499 sub volume_snapshot_list
{
500 my ($class, $scfg, $storeid, $volname) = @_;
502 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
504 my $zpath = "$scfg->{pool}/$name";
508 my $cmd = ['zfs', 'list', '-r', '-H', '-S', 'name', '-t', 'snap', '-o',
514 if ($line =~ m/^\Q$zpath\E@(.*)$/) {
519 eval { run_command
( [$cmd], outfunc
=> $outfunc , errfunc
=> sub{}); };
521 # return an empty array if dataset does not exist.
525 sub activate_storage
{
526 my ($class, $storeid, $scfg, $cache) = @_;
528 # Note: $scfg->{pool} can include dataset <pool>/<dataset>
529 my $dataset = $scfg->{pool
};
530 my $pool = ($dataset =~ s!/.*$!!r);
532 my $dataset_mounted = sub {
534 my $dataset_dec = PVE
::ProcFSTools
::decode_mount
($dataset);
535 my $mounts = eval { PVE
::ProcFSTools
::parse_proc_mounts
() };
537 foreach my $mp (@$mounts) {
538 my ($what, $dir, $fs) = @$mp;
539 next if $fs ne 'zfs';
540 # check for root-dataset of storage or any child-dataset.
541 # root-dataset could have 'canmount=off'. If any child is mounted
542 # heuristically assume that `zfs mount -a` was successful
543 next if $what !~ m!^$dataset_dec(?:/|$)!;
550 my $pool_imported = sub {
551 my @param = ('-o', 'name', '-H', $pool);
552 my $res = eval { $class->zfs_request($scfg, undef, 'zpool_list', @param) };
555 return defined($res) && $res =~ m/$pool/;
558 if (!$dataset_mounted->()) {
559 # import can only be done if not yet imported!
560 my @param = ('-d', '/dev/disk/by-id/', '-o', 'cachefile=none', $pool);
561 eval { $class->zfs_request($scfg, undef, 'zpool_import', @param) };
563 # just could've raced with another import, so recheck if it is imported
564 die "could not activate storage '$storeid', $err\n" if !$pool_imported->();
566 eval { $class->zfs_request($scfg, undef, 'mount', '-a') };
567 die "could not activate storage '$storeid', $@\n" if $@;
572 sub deactivate_storage
{
573 my ($class, $storeid, $scfg, $cache) = @_;
577 sub activate_volume
{
578 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
580 return 1 if defined($snapname);
582 my (undef, $dataset, undef, undef, undef, undef, $format) = $class->parse_volname($volname);
584 if ($format eq 'raw') {
585 $class->zfs_wait_for_zvol_link($scfg, $volname);
586 } elsif ($format eq 'subvol') {
587 my $mounted = $class->zfs_get_properties($scfg, 'mounted', "$scfg->{pool}/$dataset");
588 if ($mounted !~ m/^yes$/) {
589 $class->zfs_request($scfg, undef, 'mount', "$scfg->{pool}/$dataset");
596 sub deactivate_volume
{
597 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
602 my ($class, $scfg, $storeid, $volname, $vmid, $snap) = @_;
604 $snap ||= '__base__';
606 my ($vtype, $basename, $basevmid, undef, undef, $isBase, $format) =
607 $class->parse_volname($volname);
609 die "clone_image only works on base images\n" if !$isBase;
611 my $name = $class->find_free_diskname($storeid, $scfg, $vmid, $format);
613 if ($format eq 'subvol') {
614 my $size = $class->zfs_request($scfg, undef, 'list', '-Hp', '-o', 'refquota', "$scfg->{pool}/$basename");
616 $class->zfs_request($scfg, undef, 'clone', "$scfg->{pool}/$basename\@$snap", "$scfg->{pool}/$name", '-o', "refquota=$size");
618 $class->zfs_request($scfg, undef, 'clone', "$scfg->{pool}/$basename\@$snap", "$scfg->{pool}/$name");
621 return "$basename/$name";
625 my ($class, $storeid, $scfg, $volname) = @_;
627 my $snap = '__base__';
629 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase, $format) =
630 $class->parse_volname($volname);
632 die "create_base not possible with base image\n" if $isBase;
635 if ( $format eq 'subvol' ) {
636 $newname =~ s/^subvol-/basevol-/;
638 $newname =~ s/^vm-/base-/;
640 my $newvolname = $basename ?
"$basename/$newname" : "$newname";
642 $class->zfs_request($scfg, undef, 'rename', "$scfg->{pool}/$name", "$scfg->{pool}/$newname");
644 my $running = undef; #fixme : is create_base always offline ?
646 $class->volume_snapshot($scfg, $storeid, $newname, $snap, $running);
652 my ($class, $scfg, $storeid, $volname, $size, $running) = @_;
654 my $new_size = int($size/1024);
656 my (undef, $vname, undef, undef, undef, undef, $format) =
657 $class->parse_volname($volname);
659 my $attr = $format eq 'subvol' ?
'refquota' : 'volsize';
661 # align size to 1M so we always have a valid multiple of the volume block size
662 if ($format eq 'raw') {
663 my $padding = (1024 - $new_size % 1024) % 1024;
664 $new_size = $new_size + $padding;
667 $class->zfs_request($scfg, undef, 'set', "$attr=${new_size}k", "$scfg->{pool}/$vname");
672 sub storage_can_replicate
{
673 my ($class, $scfg, $storeid, $format) = @_;
675 return 1 if $format eq 'raw' || $format eq 'subvol';
680 sub volume_has_feature
{
681 my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_;
684 snapshot
=> { current
=> 1, snap
=> 1},
685 clone
=> { base
=> 1},
686 template
=> { current
=> 1},
687 copy
=> { base
=> 1, current
=> 1},
688 sparseinit
=> { base
=> 1, current
=> 1},
689 replicate
=> { base
=> 1, current
=> 1},
692 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
693 $class->parse_volname($volname);
700 $key = $isBase ?
'base' : 'current';
703 return 1 if $features->{$feature}->{$key};
709 my ($class, $scfg, $storeid, $fh, $volname, $format, $snapshot, $base_snapshot, $with_snapshots) = @_;
711 die "unsupported export stream format for $class: $format\n"
714 die "$class storage can only export snapshots\n"
715 if !defined($snapshot);
717 my $dataset = ($class->parse_volname($volname))[1];
719 my $fd = fileno($fh);
720 die "internal error: invalid file handle for volume_export\n"
724 # For zfs we always create a replication stream (-R) which means the remote
725 # side will always delete non-existing source snapshots. This should work
726 # for all our use cases.
727 my $cmd = ['zfs', 'send', '-Rpv'];
728 if (defined($base_snapshot)) {
729 my $arg = $with_snapshots ?
'-I' : '-i';
730 push @$cmd, $arg, $base_snapshot;
732 push @$cmd, '--', "$scfg->{pool}/$dataset\@$snapshot";
734 run_command
($cmd, output
=> $fd);
739 sub volume_export_formats
{
740 my ($class, $scfg, $storeid, $volname, $snapshot, $base_snapshot, $with_snapshots) = @_;
742 my @formats = ('zfs');
744 # push @formats, 'fies' if $volname !~ /^(?:basevol|subvol)-/;
745 # push @formats, 'raw' if !$base_snapshot && !$with_snapshots;
750 my ($class, $scfg, $storeid, $fh, $volname, $format, $base_snapshot, $with_snapshots, $allow_rename) = @_;
752 die "unsupported import stream format for $class: $format\n"
755 my $fd = fileno($fh);
756 die "internal error: invalid file handle for volume_import\n"
759 my (undef, $dataset, $vmid) = $class->parse_volname($volname);
760 my $zfspath = "$scfg->{pool}/$dataset";
761 my $suffix = defined($base_snapshot) ?
"\@$base_snapshot" : '';
762 my $exists = 0 == run_command
(['zfs', 'get', '-H', 'name', $zfspath.$suffix],
763 noerr
=> 1, quiet
=> 1);
764 if (defined($base_snapshot)) {
765 die "base snapshot '$zfspath\@$base_snapshot' doesn't exist\n" if !$exists;
767 die "volume '$zfspath' already exists\n" if !$allow_rename;
768 warn "volume '$zfspath' already exists - importing with a different name\n";
769 $dataset = $class->find_free_diskname($storeid, $scfg, $vmid, $format);
770 $zfspath = "$scfg->{pool}/$dataset";
773 eval { run_command
(['zfs', 'recv', '-F', '--', $zfspath], input
=> "<&$fd") };
775 if (defined($base_snapshot)) {
776 eval { run_command
(['zfs', 'rollback', '-r', '--', "$zfspath\@$base_snapshot"]) };
778 eval { run_command
(['zfs', 'destroy', '-r', '--', $zfspath]) };
783 return "$storeid:$dataset";
786 sub volume_import_formats
{
787 my ($class, $scfg, $storeid, $volname, $base_snapshot, $with_snapshots) = @_;
789 return $class->volume_export_formats($scfg, $storeid, $volname, undef, $base_snapshot, $with_snapshots);