X-Git-Url: https://git.proxmox.com/?p=pve-storage.git;a=blobdiff_plain;f=PVE%2FStorage%2FZFSPoolPlugin.pm;h=5f6befd23d57bc28b28f69eeca989aa826e19784;hp=6b095f1b62bceaccf5d8118ab0354b9563346b60;hb=95dfa44ca1a41bd5913c5b7c66adcf8291fac6d4;hpb=d6d924d0bf9485329d15bfc29ddc177c3370dc30

diff --git a/PVE/Storage/ZFSPoolPlugin.pm b/PVE/Storage/ZFSPoolPlugin.pm
index 6b095f1..5f6befd 100644
--- a/PVE/Storage/ZFSPoolPlugin.pm
+++ b/PVE/Storage/ZFSPoolPlugin.pm
@@ -2,11 +2,15 @@ package PVE::Storage::ZFSPoolPlugin;
 
 use strict;
 use warnings;
+
 use IO::File;
+use Net::IP;
 use POSIX;
-use PVE::Tools qw(run_command);
-use PVE::Storage::Plugin;
 
+use PVE::ProcFSTools;
+use PVE::RPCEnvironment;
+use PVE::Storage::Plugin;
+use PVE::Tools qw(run_command);
 
 use base qw(PVE::Storage::Plugin);
 
@@ -31,6 +35,10 @@ sub properties {
 	    description => "use sparse volumes",
 	    type => 'boolean',
 	},
+	mountpoint => {
+	    description => "mount point",
+	    type => 'string', format => 'pve-storage-path',
+	},
     };
 }
 
@@ -41,49 +49,14 @@ sub options {
 	sparse => { optional => 1 },
 	nodes => { optional => 1 },
 	disable => { optional => 1 },
-        maxfiles => { optional => 1 },
 	content => { optional => 1 },
+	bwlimit => { optional => 1 },
+	mountpoint => { optional => 1 },
     };
 }
 
 # static zfs helper methods
 
-sub zfs_parse_size {
-    my ($text) = @_;
-
-    return 0 if !$text;
-    
-    if ($text =~ m/^(\d+(\.\d+)?)([TGMK])?$/) {
-
-	my ($size, $reminder, $unit) = ($1, $2, $3);
-	
-	if ($unit) {
-	    if ($unit eq 'K') {
-		$size *= 1024;
-	    } elsif ($unit eq 'M') {
-		$size *= 1024*1024;
-	    } elsif ($unit eq 'G') {
-		$size *= 1024*1024*1024;
-	    } elsif ($unit eq 'T') {
-		$size *= 1024*1024*1024*1024;
-	    } else {
-		die "got unknown zfs size unit '$unit'\n";
-	    }
-	}
-
-	if ($reminder) {
-	    $size = ceil($size);
-	}
-	
-	return $size;
-    
-    }
-
-    warn "unable to parse zfs size '$text'\n";
-
-    return 0;
-}
-
 sub zfs_parse_zvol_list {
     my ($text) = @_;
 
@@ -98,25 +71,24 @@ sub zfs_parse_zvol_list {
 
 	my $zvol = {};
 	my @parts = split /\//, $dataset;
+	next if scalar(@parts) < 2; # we need pool/name
 	my $name = pop @parts;
 	my $pool = join('/', @parts);
 
-	next unless $name =~ m!^(vm|base|subvol)-(\d+)-(\S+)$!;
+	next unless $name =~ m!^(vm|base|subvol|basevol)-(\d+)-(\S+)$!;
 	$zvol->{owner} = $2;
 
-	$name = $pool . '/' . $name;
-
 	$zvol->{pool} = $pool;
 	$zvol->{name} = $name;
 	if ($type eq 'filesystem') {
 	    if ($refquota eq 'none') {
 		$zvol->{size} = 0;
 	    } else {
-		$zvol->{size} = zfs_parse_size($refquota);
+		$zvol->{size} = $refquota + 0;
 	    }
 	    $zvol->{format} = 'subvol';
 	} else {
-	    $zvol->{size} = zfs_parse_size($size);
+	    $zvol->{size} = $size + 0;
 	    $zvol->{format} = 'raw';
 	}
 	if ($origin !~ /^-$/) {
@@ -131,8 +103,10 @@ sub zfs_parse_zvol_list {
 sub parse_volname {
     my ($class, $volname) = @_;
 
-    if ($volname =~ m/^(((base|vm)-(\d+)-\S+)\/)?((base)?(vm|subvol)?-(\d+)-\S+)$/) {
-	return ('images', $5, $8, $2, $4, $6);
+    if ($volname =~ m/^(((base|basevol)-(\d+)-\S+)\/)?((base|basevol|vm|subvol)-(\d+)-\S+)$/) {
+	my $format = ($6 eq 'subvol' || $6 eq 'basevol') ? 'subvol' : 'raw';
+	my $isBase = ($6 eq 'base' || $6 eq 'basevol');
+	return ('images', $5, $7, $2, $4, $isBase, $format);
     }
 
     die "unable to parse zfs volume name '$volname'\n";
@@ -140,20 +114,45 @@ sub parse_volname {
 
 # virtual zfs methods (subclass can overwrite them)
 
+sub on_add_hook {
+    my ($class, $storeid, $scfg, %param) = @_;
+
+    my $cfg_mountpoint = $scfg->{mountpoint};
+
+    # ignore failure, pool might currently not be imported
+    my $mountpoint;
+    eval {
+	my $res = $class->zfs_get_properties($scfg, 'mountpoint', $scfg->{pool}, 1);
+	$mountpoint = PVE::Storage::Plugin::verify_path($res, 1) if defined($res);
+    };
+
+    if (defined($cfg_mountpoint)) {
+	if (defined($mountpoint) && !($cfg_mountpoint =~ m|^\Q$mountpoint\E/?$|)) {
+	    warn "warning for $storeid - mountpoint: $cfg_mountpoint " .
+		 "does not match current mount point: $mountpoint\n";
+	}
+    } else {
+	$scfg->{mountpoint} = $mountpoint;
+    }
+
+    return;
+}
+
 sub path {
-    my ($class, $scfg, $volname) = @_;
+    my ($class, $scfg, $volname, $storeid, $snapname) = @_;
 
     my ($vtype, $name, $vmid) = $class->parse_volname($volname);
 
     my $path = '';
+    my $mountpoint = $scfg->{mountpoint} // "/$scfg->{pool}";
 
     if ($vtype eq "images") {
-	if ($volname =~ m/^subvol-/) {
-	    # fixme: we currently assume standard mount point?!
-	    $path = "/$scfg->{pool}/$volname";
+	if ($name =~ m/^subvol-/ || $name =~ m/^basevol-/) {
+	    $path = "$mountpoint/$name";
 	} else {
-	    $path = "/dev/zvol/$scfg->{pool}/$volname";
+	    $path = "/dev/zvol/$scfg->{pool}/$name";
 	}
+	$path .= "\@$snapname" if defined($snapname);
     } else {
 	die "$vtype is not allowed in ZFSPool!";
     }
@@ -164,56 +163,72 @@ sub path {
 sub zfs_request {
     my ($class, $scfg, $timeout, $method, @params) = @_;
 
-    $timeout = 5 if !$timeout;
-
     my $cmd = [];
 
     if ($method eq 'zpool_list') {
 	push @$cmd, 'zpool', 'list';
+    } elsif ($method eq 'zpool_import') {
+	push @$cmd, 'zpool', 'import';
+	$timeout = 15 if !$timeout || $timeout < 15;
     } else {
 	push @$cmd, 'zfs', $method;
     }
-
     push @$cmd, @params;
- 
+
     my $msg = '';
+    my $output = sub { $msg .= "$_[0]\n" };
 
-    my $output = sub {
-        my $line = shift;
-        $msg .= "$line\n";
-    };
+    $timeout = PVE::RPCEnvironment->is_worker() ? 60*60 : 5 if !$timeout;
 
     run_command($cmd, errmsg => "zfs error", outfunc => $output, timeout => $timeout);
 
     return $msg;
 }
 
+sub zfs_wait_for_zvol_link {
+    my ($class, $scfg, $volname, $timeout) = @_;
+
+    my $default_timeout = PVE::RPCEnvironment->is_worker() ? 60*5 : 10;
+    $timeout = $default_timeout if !defined($timeout);
+
+    my ($devname, undef, undef) = $class->path($scfg, $volname);
+
+    for (my $i = 1; $i <= $timeout; $i++) {
+	last if -b $devname;
+	die "timeout: no zvol device link for '$volname' found after $timeout sec found.\n"
+	    if $i == $timeout;
+
+	sleep(1);
+    }
+}
+
 sub alloc_image {
     my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_;
 
     my $volname = $name;
-    
+
     if ($fmt eq 'raw') {
 
-	die "illegal name '$volname' - sould be 'vm-$vmid-*'\n"
+	die "illegal name '$volname' - should be 'vm-$vmid-*'\n"
 	    if $volname && $volname !~ m/^vm-$vmid-/;
-	$volname = $class->zfs_find_free_diskname($storeid, $scfg, $vmid) 
+	$volname = $class->find_free_diskname($storeid, $scfg, $vmid, $fmt)
 	    if !$volname;
 
 	$class->zfs_create_zvol($scfg, $volname, $size);
-	my $devname = "/dev/zvol/$scfg->{pool}/$volname";
-
-	run_command("udevadm trigger --subsystem-match block");
-	system("udevadm settle --timeout 10 --exit-if-exists=${devname}");
+	$class->zfs_wait_for_zvol_link($scfg, $volname);
 
     } elsif ( $fmt eq 'subvol') {
-	
-	die "subvolume allocation without name\n" if !$volname;
-	die "illegal name '$volname' - sould be 'subvol-$vmid-*'\n"
+
+	die "illegal name '$volname' - should be 'subvol-$vmid-*'\n"
+	    if $volname && $volname !~ m/^subvol-$vmid-/;
+	$volname = $class->find_free_diskname($storeid, $scfg, $vmid, $fmt)
+	    if !$volname;
+
+	die "illegal name '$volname' - should be 'subvol-$vmid-*'\n"
 	    if $volname !~ m/^subvol-$vmid-/;
 
-	$class->zfs_create_subvol($scfg, $volname, $size);	
-	
+	$class->zfs_create_subvol($scfg, $volname, $size);
+
     } else {
 	die "unsupported format '$fmt'";
     }
@@ -242,43 +257,48 @@ sub list_images {
 
 	foreach my $image (keys %$dat) {
 
-	    my $volname = $dat->{$image}->{name};
-	    my $parent = $dat->{$image}->{parent};
+	    my $info = $dat->{$image};
 
-	    my $volid = undef;
-            if ($parent && $parent =~ m/^(\S+)@(\S+)$/) {
+	    my $volname = $info->{name};
+	    my $parent = $info->{parent};
+	    my $owner = $info->{vmid};
+
+	    if ($parent && $parent =~ m/^(\S+)\@__base__$/) {
 		my ($basename) = ($1);
-		$volid = "$storeid:$basename/$volname";
+		$info->{volid} = "$storeid:$basename/$volname";
 	    } else {
-		$volid = "$storeid:$volname";
+		$info->{volid} = "$storeid:$volname";
 	    }
 
-	    my $owner = $dat->{$volname}->{vmid};
 	    if ($vollist) {
-		my $found = grep { $_ eq $volid } @$vollist;
+		my $found = grep { $_ eq $info->{volid} } @$vollist;
 		next if !$found;
 	    } else {
 		next if defined ($vmid) && ($owner ne $vmid);
 	    }
 
-	    my $info = $dat->{$volname};
-	    $info->{volid} = $volid;
 	    push @$res, $info;
 	}
     }
     return $res;
 }
 
+sub zfs_get_properties {
+    my ($class, $scfg, $properties, $dataset, $timeout) = @_;
+
+    my $result = $class->zfs_request($scfg, $timeout, 'get', '-o', 'value',
+				     '-Hp', $properties, $dataset);
+    my @values = split /\n/, $result;
+    return wantarray ? @values : $values[0];
+}
+
 sub zfs_get_pool_stats {
     my ($class, $scfg) = @_;
 
     my $available = 0;
     my $used = 0;
 
-    my $text = $class->zfs_request($scfg, undef, 'get', '-o', 'value', '-Hp',
-               'available,used', $scfg->{pool});
-
-    my @lines = split /\n/, $text;
+    my @lines = $class->zfs_get_properties($scfg, 'available,used', $scfg->{pool});
 
     if($lines[0] =~ /^(\d+)$/) {
 	$available = $1;
@@ -291,21 +311,14 @@ sub zfs_get_pool_stats {
     return ($available, $used);
 }
 
-sub zfs_get_zvol_size {
-    my ($class, $scfg, $zvol) = @_;
-
-    my $text = $class->zfs_request($scfg, undef, 'get', '-Hp', 'volsize', "$scfg->{pool}/$zvol");
-
-    if ($text =~ /volsize\s(\d+)/) {
-	return $1;
-    }
-
-    die "Could not get zvol size";
-}
-
 sub zfs_create_zvol {
     my ($class, $scfg, $zvol, $size) = @_;
-    
+
+    # always align size to 1M as workaround until
+    # https://github.com/zfsonlinux/zfs/issues/8541 is solved
+    my $padding = (1024 - $size % 1024) % 1024;
+    $size = $size + $padding;
+
     my $cmd = ['create'];
 
     push @$cmd, '-s' if $scfg->{sparse};
@@ -321,8 +334,10 @@ sub zfs_create_subvol {
     my ($class, $scfg, $volname, $size) = @_;
 
     my $dataset = "$scfg->{pool}/$volname";
-    
-    my $cmd = ['create', '-o', "refquota=${size}k", $dataset];
+    my $quota = $size ? "${size}k" : "none";
+
+    my $cmd = ['create', '-o', 'acltype=posixacl', '-o', 'xattr=sa',
+	       '-o', "refquota=${quota}", $dataset];
 
     $class->zfs_request($scfg, undef, @$cmd);
 }
@@ -338,6 +353,9 @@ sub zfs_delete_zvol {
 	if ($err = $@) {
 	    if ($err =~ m/^zfs error:(.*): dataset is busy.*/) {
 		sleep(1);
+	    } elsif ($err =~ m/^zfs error:.*: dataset does not exist.*$/) {
+		$err = undef;
+		last;
 	    } else {
 		die $err;
 	    }
@@ -352,7 +370,7 @@ sub zfs_delete_zvol {
 sub zfs_list_zvol {
     my ($class, $scfg) = @_;
 
-    my $text = $class->zfs_request($scfg, 10, 'list', '-o', 'name,volsize,origin,type,refquota', '-t', 'volume,filesystem', '-Hr');
+    my $text = $class->zfs_request($scfg, 10, 'list', '-o', 'name,volsize,origin,type,refquota', '-t', 'volume,filesystem', '-Hrp');
     my $zvols = zfs_parse_zvol_list($text);
     return undef if !$zvols;
 
@@ -377,48 +395,21 @@ sub zfs_list_zvol {
     return $list;
 }
 
-sub zfs_find_free_diskname {
-    my ($class, $storeid, $scfg, $vmid) = @_;
-
-    my $name = undef;
-    my $volumes = $class->zfs_list_zvol($scfg);
+sub zfs_get_sorted_snapshot_list {
+    my ($class, $scfg, $volname, $sort_params) = @_;
 
-    my $disk_ids = {};
-    my $dat = $volumes->{$scfg->{pool}};
+    my $vname = ($class->parse_volname($volname))[1];
 
-    foreach my $image (keys %$dat) {
-        my $volname = $dat->{$image}->{name};
-        if ($volname =~ m/(vm|base)-$vmid-disk-(\d+)/){
-            $disk_ids->{$2} = 1;
-        }
-    }
-
-    for (my $i = 1; $i < 100; $i++) {
-        if (!$disk_ids->{$i}) {
-            return "vm-$vmid-disk-$i";
-        }
-    }
-
-    die "unable to allocate an image name for VM $vmid in storage '$storeid'\n";
-}
-
-sub zfs_get_latest_snapshot {
-    my ($class, $scfg, $volname) = @_;
-
-    # abort rollback if snapshot is not the latest
-    my @params = ('-t', 'snapshot', '-o', 'name', '-s', 'creation');
-    my $text = zfs_request($class, $scfg, undef, 'list', @params);
+    my @params = ('-H', '-t', 'snapshot', '-o', 'name', $sort_params->@*, "$scfg->{pool}\/$vname");
+    my $text = $class->zfs_request($scfg, undef, 'list', @params);
     my @snapshots = split(/\n/, $text);
 
-    my $recentsnap;
-    foreach (@snapshots) {
-        if (/$scfg->{pool}\/$volname/) {
-            s/^.*@//;
-            $recentsnap = $_;
-        }
+    my $snap_names = [];
+    for my $snapshot (@snapshots) {
+	(my $snap_name = $snapshot) =~ s/^.*@//;
+	push $snap_names->@*, $snap_name;
     }
-
-    return $recentsnap;
+    return $snap_names;
 }
 
 sub status {
@@ -442,52 +433,148 @@ sub status {
 sub volume_size_info {
     my ($class, $scfg, $storeid, $volname, $timeout) = @_;
 
-    return $class->zfs_get_zvol_size($scfg, $volname);
+    my (undef, $vname, undef, undef, undef, undef, $format) =
+        $class->parse_volname($volname);
+
+    my $attr = $format eq 'subvol' ? 'refquota' : 'volsize';
+    my $value = $class->zfs_get_properties($scfg, $attr, "$scfg->{pool}/$vname");
+    if ($value =~ /^(\d+)$/) {
+	return $1;
+    }
+
+    die "Could not get zfs volume size\n";
 }
 
 sub volume_snapshot {
     my ($class, $scfg, $storeid, $volname, $snap) = @_;
 
-    $class->zfs_request($scfg, undef, 'snapshot', "$scfg->{pool}/$volname\@$snap");
+    my $vname = ($class->parse_volname($volname))[1];
+
+    $class->zfs_request($scfg, undef, 'snapshot', "$scfg->{pool}/$vname\@$snap");
 }
 
 sub volume_snapshot_delete {
     my ($class, $scfg, $storeid, $volname, $snap, $running) = @_;
 
-    $class->zfs_request($scfg, undef, 'destroy', "$scfg->{pool}/$volname\@$snap");
+    my $vname = ($class->parse_volname($volname))[1];
+
+    $class->deactivate_volume($storeid, $scfg, $vname, $snap, {});
+    $class->zfs_request($scfg, undef, 'destroy', "$scfg->{pool}/$vname\@$snap");
 }
 
 sub volume_snapshot_rollback {
     my ($class, $scfg, $storeid, $volname, $snap) = @_;
 
-    zfs_request($class, $scfg, undef, 'rollback', "$scfg->{pool}/$volname\@$snap");
+    my (undef, $vname, undef, undef, undef, undef, $format) = $class->parse_volname($volname);
+
+    my $msg = $class->zfs_request($scfg, undef, 'rollback', "$scfg->{pool}/$vname\@$snap");
+
+    # we have to unmount rollbacked subvols, to invalidate wrong kernel
+    # caches, they get mounted in activate volume again
+    # see zfs bug #10931 https://github.com/openzfs/zfs/issues/10931
+    if ($format eq 'subvol') {
+	eval { $class->zfs_request($scfg, undef, 'unmount', "$scfg->{pool}/$vname"); };
+	if (my $err = $@) {
+	    die $err if $err !~ m/not currently mounted$/;
+	}
+    }
+
+    return $msg;
 }
 
 sub volume_rollback_is_possible {
-    my ($class, $scfg, $storeid, $volname, $snap) = @_; 
-    
-    my $recentsnap = $class->zfs_get_latest_snapshot($scfg, $volname);
-    if ($snap ne $recentsnap) {
-	die "can't rollback, more recent snapshots exist\n";
+    my ($class, $scfg, $storeid, $volname, $snap, $blockers) = @_;
+
+    # can't use '-S creation', because zfs list won't reverse the order when the
+    # creation time is the same second, breaking at least our tests.
+    my $snapshots = $class->zfs_get_sorted_snapshot_list($scfg, $volname, ['-s', 'creation']);
+
+    my $found;
+    $blockers //= []; # not guaranteed to be set by caller
+    for my $snapshot ($snapshots->@*) {
+	if ($snapshot eq $snap) {
+	    $found = 1;
+	} elsif ($found) {
+	    push $blockers->@*, $snapshot;
+	}
     }
 
-    return 1; 
+    my $volid = "${storeid}:${volname}";
+
+    die "can't rollback, snapshot '$snap' does not exist on '$volid'\n"
+	if !$found;
+
+    die "can't rollback, '$snap' is not most recent snapshot on '$volid'\n"
+	if scalar($blockers->@*) > 0;
+
+    return 1;
+}
+
+sub volume_snapshot_info {
+    my ($class, $scfg, $storeid, $volname) = @_;
+
+    my $vname = ($class->parse_volname($volname))[1];
+
+    my @params = ('-Hp', '-t', 'snapshot', '-o', 'name,guid,creation', "$scfg->{pool}\/$vname");
+    my $text = $class->zfs_request($scfg, undef, 'list', @params);
+    my @lines = split(/\n/, $text);
+
+    my $info = {};
+    for my $line (@lines) {
+	my ($snapshot, $guid, $creation) = split(/\s+/, $line);
+	(my $snap_name = $snapshot) =~ s/^.*@//;
+
+	$info->{$snap_name} = {
+	    id => $guid,
+	    timestamp => $creation,
+	};
+    }
+    return $info;
+}
+
+my sub dataset_mounted_heuristic {
+    my ($dataset) = @_;
+
+    my $mounts = PVE::ProcFSTools::parse_proc_mounts();
+    for my $mp (@$mounts) {
+	my ($what, $dir, $fs) = $mp->@*;
+	next if $fs ne 'zfs';
+	# check for root-dataset or any child-dataset (root-dataset could have 'canmount=off')
+	# If any child is mounted heuristically assume that `zfs mount -a` was successful
+	next if $what !~ m!^$dataset(?:/|$)!;
+	return 1;
+    }
+    return 0;
 }
 
 sub activate_storage {
     my ($class, $storeid, $scfg, $cache) = @_;
 
-    my @param = ('-o', 'name', '-H');
+    # Note: $scfg->{pool} can include dataset <pool>/<dataset>
+    my $dataset = $scfg->{pool};
+    my $pool = ($dataset =~ s!/.*$!!r);
+
+    return 1 if dataset_mounted_heuristic($dataset); # early return
 
-    my $text = zfs_request($class, $scfg, undef, 'zpool_list', @param);
+    my $pool_imported = sub {
+	my @param = ('-o', 'name', '-H', $pool);
+	my $res = eval { $class->zfs_request($scfg, undef, 'zpool_list', @param) };
+	warn "$@\n" if $@;
 
-    # Note: $scfg->{pool} can include dataset <pool>/<dataset>
-    my $pool = $scfg->{pool};
-    $pool =~ s!/.*$!!;
+	return defined($res) && $res =~ m/$pool/;
+    };
 
-    if ($text !~ $pool) {
-	run_command("zpool import -d /dev/disk/by-id/ -a");
+    if (!$pool_imported->()) {
+	# import can only be done if not yet imported!
+	my @param = ('-d', '/dev/disk/by-id/', '-o', 'cachefile=none', $pool);
+	eval { $class->zfs_request($scfg, undef, 'zpool_import', @param) };
+	if (my $err = $@) {
+	    # just could've raced with another import, so recheck if it is imported
+	    die "could not activate storage '$storeid', $err\n" if !$pool_imported->();
+	}
     }
+    eval { $class->zfs_request($scfg, undef, 'mount', '-a') };
+    die "could not activate storage '$storeid', $@\n" if $@;
     return 1;
 }
 
@@ -497,12 +584,26 @@ sub deactivate_storage {
 }
 
 sub activate_volume {
-    my ($class, $storeid, $scfg, $volname, $exclusive, $cache) = @_;
+    my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
+
+    return 1 if defined($snapname);
+
+    my (undef, $dataset, undef, undef, undef, undef, $format) = $class->parse_volname($volname);
+
+    if ($format eq 'raw') {
+	$class->zfs_wait_for_zvol_link($scfg, $volname);
+    } elsif ($format eq 'subvol') {
+	my $mounted = $class->zfs_get_properties($scfg, 'mounted', "$scfg->{pool}/$dataset");
+	if ($mounted !~ m/^yes$/) {
+	    $class->zfs_request($scfg, undef, 'mount', "$scfg->{pool}/$dataset");
+	}
+    }
+
     return 1;
 }
 
 sub deactivate_volume {
-    my ($class, $storeid, $scfg, $volname, $exclusive, $cache) = @_;
+    my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
     return 1;
 }
 
@@ -511,16 +612,22 @@ sub clone_image {
 
     $snap ||= '__base__';
 
-    my ($vtype, $basename, $basevmid, undef, undef, $isBase) =
+    my ($vtype, $basename, $basevmid, undef, undef, $isBase, $format) =
         $class->parse_volname($volname);
 
     die "clone_image only works on base images\n" if !$isBase;
 
-    my $name = $class->zfs_find_free_diskname($storeid, $scfg, $vmid);
+    my $name = $class->find_free_diskname($storeid, $scfg, $vmid, $format);
 
-    $class->zfs_request($scfg, undef, 'clone', "$scfg->{pool}/$basename\@$snap", "$scfg->{pool}/$name");
+    if ($format eq 'subvol') {
+	my $size = $class->zfs_request($scfg, undef, 'list', '-Hp', '-o', 'refquota', "$scfg->{pool}/$basename");
+	chomp($size);
+	$class->zfs_request($scfg, undef, 'clone', "$scfg->{pool}/$basename\@$snap", "$scfg->{pool}/$name", '-o', "refquota=$size");
+    } else {
+	$class->zfs_request($scfg, undef, 'clone', "$scfg->{pool}/$basename\@$snap", "$scfg->{pool}/$name");
+    }
 
-    return $name;
+    return "$basename/$name";
 }
 
 sub create_base {
@@ -528,14 +635,17 @@ sub create_base {
 
     my $snap = '__base__';
 
-    my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
+    my ($vtype, $name, $vmid, $basename, $basevmid, $isBase, $format) =
         $class->parse_volname($volname);
 
     die "create_base not possible with base image\n" if $isBase;
 
     my $newname = $name;
-    $newname =~ s/^vm-/base-/;
-
+    if ( $format eq 'subvol' ) {
+	$newname =~ s/^subvol-/basevol-/;
+    } else {
+	$newname =~ s/^vm-/base-/;
+    }
     my $newvolname = $basename ? "$basename/$newname" : "$newname";
 
     $class->zfs_request($scfg, undef, 'rename', "$scfg->{pool}/$name", "$scfg->{pool}/$newname");
@@ -550,13 +660,32 @@ sub create_base {
 sub volume_resize {
     my ($class, $scfg, $storeid, $volname, $size, $running) = @_;
 
-    my $new_size = ($size/1024);
+    my $new_size = int($size/1024);
+
+    my (undef, $vname, undef, undef, undef, undef, $format) =
+        $class->parse_volname($volname);
+
+    my $attr = $format eq 'subvol' ? 'refquota' : 'volsize';
+
+    # align size to 1M so we always have a valid multiple of the volume block size
+    if ($format eq 'raw') {
+	my $padding = (1024 - $new_size % 1024) % 1024;
+	$new_size = $new_size + $padding;
+    }
 
-    $class->zfs_request($scfg, undef, 'set', 'volsize=' . $new_size . 'k', "$scfg->{pool}/$volname");
+    $class->zfs_request($scfg, undef, 'set', "$attr=${new_size}k", "$scfg->{pool}/$vname");
 
     return $new_size;
 }
 
+sub storage_can_replicate {
+    my ($class, $scfg, $storeid, $format) = @_;
+
+    return 1 if $format eq 'raw' || $format eq 'subvol';
+
+    return 0;
+}
+
 sub volume_has_feature {
     my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_;
 
@@ -565,6 +694,9 @@ sub volume_has_feature {
 	clone => { base => 1},
 	template => { current => 1},
 	copy => { base => 1, current => 1},
+	sparseinit => { base => 1, current => 1},
+	replicate => { base => 1, current => 1},
+	rename => {current => 1},
     };
 
     my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
@@ -583,4 +715,118 @@ sub volume_has_feature {
     return undef;
 }
 
+sub volume_export {
+    my ($class, $scfg, $storeid, $fh, $volname, $format, $snapshot, $base_snapshot, $with_snapshots) = @_;
+
+    die "unsupported export stream format for $class: $format\n"
+	if $format ne 'zfs';
+
+    die "$class storage can only export snapshots\n"
+	if !defined($snapshot);
+
+    my $dataset = ($class->parse_volname($volname))[1];
+
+    my $fd = fileno($fh);
+    die "internal error: invalid file handle for volume_export\n"
+	if !defined($fd);
+    $fd = ">&$fd";
+
+    # For zfs we always create a replication stream (-R) which means the remote
+    # side will always delete non-existing source snapshots. This should work
+    # for all our use cases.
+    my $cmd = ['zfs', 'send', '-Rpv'];
+    if (defined($base_snapshot)) {
+	my $arg = $with_snapshots ? '-I' : '-i';
+	push @$cmd, $arg, $base_snapshot;
+    }
+    push @$cmd, '--', "$scfg->{pool}/$dataset\@$snapshot";
+
+    run_command($cmd, output => $fd);
+
+    return;
+}
+
+sub volume_export_formats {
+    my ($class, $scfg, $storeid, $volname, $snapshot, $base_snapshot, $with_snapshots) = @_;
+
+    my @formats = ('zfs');
+    # TODOs:
+    # push @formats, 'fies' if $volname !~ /^(?:basevol|subvol)-/;
+    # push @formats, 'raw' if !$base_snapshot && !$with_snapshots;
+    return @formats;
+}
+
+sub volume_import {
+    my ($class, $scfg, $storeid, $fh, $volname, $format, $snapshot, $base_snapshot, $with_snapshots, $allow_rename) = @_;
+
+    die "unsupported import stream format for $class: $format\n"
+	if $format ne 'zfs';
+
+    my $fd = fileno($fh);
+    die "internal error: invalid file handle for volume_import\n"
+	if !defined($fd);
+
+    my (undef, $dataset, $vmid) = $class->parse_volname($volname);
+    my $zfspath = "$scfg->{pool}/$dataset";
+    my $suffix = defined($base_snapshot) ? "\@$base_snapshot" : '';
+    my $exists = 0 == run_command(['zfs', 'get', '-H', 'name', $zfspath.$suffix],
+				  noerr => 1, quiet => 1);
+    if (defined($base_snapshot)) {
+	die "base snapshot '$zfspath\@$base_snapshot' doesn't exist\n" if !$exists;
+    } elsif ($exists) {
+	die "volume '$zfspath' already exists\n" if !$allow_rename;
+	warn "volume '$zfspath' already exists - importing with a different name\n";
+	$dataset = $class->find_free_diskname($storeid, $scfg, $vmid, $format);
+	$zfspath = "$scfg->{pool}/$dataset";
+    }
+
+    eval { run_command(['zfs', 'recv', '-F', '--', $zfspath], input => "<&$fd") };
+    if (my $err = $@) {
+	if (defined($base_snapshot)) {
+	    eval { run_command(['zfs', 'rollback', '-r', '--', "$zfspath\@$base_snapshot"]) };
+	} else {
+	    eval { run_command(['zfs', 'destroy', '-r', '--', $zfspath]) };
+	}
+	die $err;
+    }
+
+    return "$storeid:$dataset";
+}
+
+sub volume_import_formats {
+    my ($class, $scfg, $storeid, $volname, $snapshot, $base_snapshot, $with_snapshots) = @_;
+
+    return $class->volume_export_formats($scfg, $storeid, $volname, $snapshot, $base_snapshot, $with_snapshots);
+}
+
+sub rename_volume {
+    my ($class, $scfg, $storeid, $source_volname, $target_vmid, $target_volname) = @_;
+
+    my (
+	undef,
+	$source_image,
+	$source_vmid,
+	$base_name,
+	$base_vmid,
+	undef,
+	$format
+    ) = $class->parse_volname($source_volname);
+    $target_volname = $class->find_free_diskname($storeid, $scfg, $target_vmid, $format)
+	if !$target_volname;
+
+    my $pool = $scfg->{pool};
+    my $source_zfspath = "${pool}/${source_image}";
+    my $target_zfspath = "${pool}/${target_volname}";
+
+    my $exists = 0 == run_command(['zfs', 'get', '-H', 'name', $target_zfspath],
+				  noerr => 1, quiet => 1);
+    die "target volume '${target_volname}' already exists\n" if $exists;
+
+    $class->zfs_request($scfg, 5, 'rename', ${source_zfspath}, ${target_zfspath});
+
+    $base_name = $base_name ? "${base_name}/" : '';
+
+    return "${storeid}:${base_name}${target_volname}";
+}
+
 1;