]> git.proxmox.com Git - pve-storage.git/blame - PVE/Storage/ZFSPoolPlugin.pm
allow containers on lvm
[pve-storage.git] / PVE / Storage / ZFSPoolPlugin.pm
CommitLineData
85fda4dd 1package PVE::Storage::ZFSPoolPlugin;
5bb8e010
DM
2
3use strict;
4use warnings;
5use IO::File;
6use POSIX;
7use PVE::Tools qw(run_command);
8use PVE::Storage::Plugin;
9
10
11use base qw(PVE::Storage::Plugin);
12
5bb8e010 13sub type {
85fda4dd 14 return 'zfspool';
5bb8e010
DM
15}
16
17sub plugindata {
18 return {
1ccae449
DM
19 content => [ {images => 1, rootdir => 1}, {images => 1 , rootdir => 1}],
20 format => [ { raw => 1, subvol => 1 } , 'raw' ],
5bb8e010 21 };
85fda4dd 22}
5bb8e010 23
7730694e
DM
24sub properties {
25 return {
26 blocksize => {
27 description => "block size",
28 type => 'string',
29 },
30 sparse => {
31 description => "use sparse volumes",
32 type => 'boolean',
33 },
34 };
35}
36
5bb8e010
DM
37sub options {
38 return {
7730694e
DM
39 pool => { fixed => 1 },
40 blocksize => { optional => 1 },
41 sparse => { optional => 1 },
42 nodes => { optional => 1 },
5bb8e010
DM
43 disable => { optional => 1 },
44 maxfiles => { optional => 1 },
45 content => { optional => 1 },
46 };
47}
48
7730694e
DM
49# static zfs helper methods
50
060ef890
DM
51sub zfs_parse_size {
52 my ($text) = @_;
53
54 return 0 if !$text;
55
56 if ($text =~ m/^(\d+(\.\d+)?)([TGMK])?$/) {
57
58 my ($size, $reminder, $unit) = ($1, $2, $3);
59
60 if ($unit) {
61 if ($unit eq 'K') {
62 $size *= 1024;
63 } elsif ($unit eq 'M') {
64 $size *= 1024*1024;
65 } elsif ($unit eq 'G') {
66 $size *= 1024*1024*1024;
67 } elsif ($unit eq 'T') {
68 $size *= 1024*1024*1024*1024;
69 } else {
70 die "got unknown zfs size unit '$unit'\n";
71 }
72 }
73
74 if ($reminder) {
75 $size = ceil($size);
76 }
77
78 return $size;
79
80 }
81
82 warn "unable to parse zfs size '$text'\n";
83
84 return 0;
85}
86
7730694e
DM
87sub zfs_parse_zvol_list {
88 my ($text) = @_;
89
90 my $list = ();
91
92 return $list if !$text;
93
94 my @lines = split /\n/, $text;
95 foreach my $line (@lines) {
1ccae449
DM
96 my ($dataset, $size, $origin, $type, $refquota) = split(/\s+/, $line);
97 next if !($type eq 'volume' || $type eq 'filesystem');
98
99 my $zvol = {};
100 my @parts = split /\//, $dataset;
dec97937 101 next if scalar(@parts) < 2; # we need pool/name
1ccae449
DM
102 my $name = pop @parts;
103 my $pool = join('/', @parts);
104
105 next unless $name =~ m!^(vm|base|subvol)-(\d+)-(\S+)$!;
106 $zvol->{owner} = $2;
107
1ccae449
DM
108 $zvol->{pool} = $pool;
109 $zvol->{name} = $name;
110 if ($type eq 'filesystem') {
111 if ($refquota eq 'none') {
112 $zvol->{size} = 0;
113 } else {
114 $zvol->{size} = zfs_parse_size($refquota);
7730694e 115 }
1ccae449
DM
116 $zvol->{format} = 'subvol';
117 } else {
118 $zvol->{size} = zfs_parse_size($size);
119 $zvol->{format} = 'raw';
7730694e 120 }
1ccae449
DM
121 if ($origin !~ /^-$/) {
122 $zvol->{origin} = $origin;
123 }
124 push @$list, $zvol;
7730694e
DM
125 }
126
127 return $list;
128}
129
cc80ed9c
WL
130sub parse_volname {
131 my ($class, $volname) = @_;
132
1ccae449 133 if ($volname =~ m/^(((base|vm)-(\d+)-\S+)\/)?((base)?(vm|subvol)?-(\d+)-\S+)$/) {
55525ad2
DM
134 my $format = $7 && $7 eq 'subvol' ? 'subvol' : 'raw';
135 return ('images', $5, $8, $2, $4, $6, $format);
cc80ed9c
WL
136 }
137
138 die "unable to parse zfs volume name '$volname'\n";
139}
140
7730694e
DM
141# virtual zfs methods (subclass can overwrite them)
142
f3e632d0 143sub path {
e67069eb 144 my ($class, $scfg, $volname, $storeid, $snapname) = @_;
f3e632d0
WL
145
146 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
147
148 my $path = '';
149
1ccae449
DM
150 if ($vtype eq "images") {
151 if ($volname =~ m/^subvol-/) {
152 # fixme: we currently assume standard mount point?!
d6d924d0 153 $path = "/$scfg->{pool}/$volname";
1ccae449
DM
154 } else {
155 $path = "/dev/zvol/$scfg->{pool}/$volname";
156 }
e67069eb 157 $path .= "\@$snapname" if defined($snapname);
f3e632d0 158 } else {
85fda4dd 159 die "$vtype is not allowed in ZFSPool!";
f3e632d0
WL
160 }
161
162 return ($path, $vmid, $vtype);
163}
164
7730694e
DM
165sub zfs_request {
166 my ($class, $scfg, $timeout, $method, @params) = @_;
167
168 $timeout = 5 if !$timeout;
169
170 my $cmd = [];
171
172 if ($method eq 'zpool_list') {
86d47239 173 push @$cmd, 'zpool', 'list';
7730694e
DM
174 } else {
175 push @$cmd, 'zfs', $method;
176 }
177
178 push @$cmd, @params;
179
180 my $msg = '';
181
182 my $output = sub {
183 my $line = shift;
184 $msg .= "$line\n";
185 };
186
1f390a30 187 run_command($cmd, errmsg => "zfs error", outfunc => $output, timeout => $timeout);
7730694e
DM
188
189 return $msg;
190}
191
b3ba95e4
WL
192sub alloc_image {
193 my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_;
194
1ccae449
DM
195 my $volname = $name;
196
197 if ($fmt eq 'raw') {
b3ba95e4 198
1ccae449
DM
199 die "illegal name '$volname' - sould be 'vm-$vmid-*'\n"
200 if $volname && $volname !~ m/^vm-$vmid-/;
55525ad2 201 $volname = $class->zfs_find_free_diskname($storeid, $scfg, $vmid, $fmt)
1ccae449 202 if !$volname;
b3ba95e4 203
1ccae449
DM
204 $class->zfs_create_zvol($scfg, $volname, $size);
205 my $devname = "/dev/zvol/$scfg->{pool}/$volname";
82e08809 206
1ccae449
DM
207 run_command("udevadm trigger --subsystem-match block");
208 system("udevadm settle --timeout 10 --exit-if-exists=${devname}");
76fd7dc7 209
1ccae449 210 } elsif ( $fmt eq 'subvol') {
55525ad2
DM
211
212 die "illegal name '$volname' - sould be 'subvol-$vmid-*'\n"
213 if $volname && $volname !~ m/^subvol-$vmid-/;
214 $volname = $class->zfs_find_free_diskname($storeid, $scfg, $vmid, $fmt)
215 if !$volname;
216
1ccae449
DM
217 die "illegal name '$volname' - sould be 'subvol-$vmid-*'\n"
218 if $volname !~ m/^subvol-$vmid-/;
76fd7dc7 219
1ccae449
DM
220 $class->zfs_create_subvol($scfg, $volname, $size);
221
222 } else {
223 die "unsupported format '$fmt'";
224 }
b3ba95e4 225
82e08809 226 return $volname;
b3ba95e4
WL
227}
228
e9565df5
WL
229sub free_image {
230 my ($class, $storeid, $scfg, $volname, $isBase) = @_;
231
232 my (undef, $name, undef) = $class->parse_volname($volname);
233
234 $class->zfs_delete_zvol($scfg, $name);
235
236 return undef;
237}
238
ca04180f
WL
239sub list_images {
240 my ($class, $storeid, $scfg, $vmid, $vollist, $cache) = @_;
241
242 $cache->{zfs} = $class->zfs_list_zvol($scfg) if !$cache->{zfs};
243 my $zfspool = $scfg->{pool};
244 my $res = [];
245
246 if (my $dat = $cache->{zfs}->{$zfspool}) {
247
248 foreach my $image (keys %$dat) {
249
250 my $volname = $dat->{$image}->{name};
251 my $parent = $dat->{$image}->{parent};
252
253 my $volid = undef;
254 if ($parent && $parent =~ m/^(\S+)@(\S+)$/) {
255 my ($basename) = ($1);
256 $volid = "$storeid:$basename/$volname";
257 } else {
258 $volid = "$storeid:$volname";
259 }
260
261 my $owner = $dat->{$volname}->{vmid};
262 if ($vollist) {
263 my $found = grep { $_ eq $volid } @$vollist;
264 next if !$found;
265 } else {
266 next if defined ($vmid) && ($owner ne $vmid);
267 }
268
269 my $info = $dat->{$volname};
270 $info->{volid} = $volid;
271 push @$res, $info;
272 }
273 }
ca04180f
WL
274 return $res;
275}
276
7730694e
DM
277sub zfs_get_pool_stats {
278 my ($class, $scfg) = @_;
279
280 my $available = 0;
281 my $used = 0;
282
283 my $text = $class->zfs_request($scfg, undef, 'get', '-o', 'value', '-Hp',
284 'available,used', $scfg->{pool});
285
286 my @lines = split /\n/, $text;
287
288 if($lines[0] =~ /^(\d+)$/) {
289 $available = $1;
290 }
291
292 if($lines[1] =~ /^(\d+)$/) {
293 $used = $1;
294 }
295
296 return ($available, $used);
297}
298
299sub zfs_get_zvol_size {
300 my ($class, $scfg, $zvol) = @_;
301
302 my $text = $class->zfs_request($scfg, undef, 'get', '-Hp', 'volsize', "$scfg->{pool}/$zvol");
303
304 if ($text =~ /volsize\s(\d+)/) {
305 return $1;
306 }
307
308 die "Could not get zvol size";
309}
310
311sub zfs_create_zvol {
312 my ($class, $scfg, $zvol, $size) = @_;
313
314 my $cmd = ['create'];
315
316 push @$cmd, '-s' if $scfg->{sparse};
317
318 push @$cmd, '-b', $scfg->{blocksize} if $scfg->{blocksize};
319
320 push @$cmd, '-V', "${size}k", "$scfg->{pool}/$zvol";
321
322 $class->zfs_request($scfg, undef, @$cmd);
323}
324
1ccae449
DM
325sub zfs_create_subvol {
326 my ($class, $scfg, $volname, $size) = @_;
327
328 my $dataset = "$scfg->{pool}/$volname";
329
330 my $cmd = ['create', '-o', "refquota=${size}k", $dataset];
331
332 $class->zfs_request($scfg, undef, @$cmd);
333}
334
7730694e
DM
335sub zfs_delete_zvol {
336 my ($class, $scfg, $zvol) = @_;
337
1f390a30
WL
338 my $err;
339
340 for (my $i = 0; $i < 6; $i++) {
341
342 eval { $class->zfs_request($scfg, undef, 'destroy', '-r', "$scfg->{pool}/$zvol"); };
343 if ($err = $@) {
344 if ($err =~ m/^zfs error:(.*): dataset is busy.*/) {
345 sleep(1);
27ff0e99
WL
346 } elsif ($err =~ m/^zfs error:.*: dataset does not exist.*$/) {
347 $err = undef;
348 last;
1f390a30
WL
349 } else {
350 die $err;
351 }
352 } else {
353 last;
354 }
355 }
356
357 die $err if $err;
7730694e
DM
358}
359
360sub zfs_list_zvol {
361 my ($class, $scfg) = @_;
362
1ccae449 363 my $text = $class->zfs_request($scfg, 10, 'list', '-o', 'name,volsize,origin,type,refquota', '-t', 'volume,filesystem', '-Hr');
7730694e
DM
364 my $zvols = zfs_parse_zvol_list($text);
365 return undef if !$zvols;
366
367 my $list = ();
368 foreach my $zvol (@$zvols) {
1ccae449
DM
369 my $pool = $zvol->{pool};
370 my $name = $zvol->{name};
7730694e
DM
371 my $parent = $zvol->{origin};
372 if($zvol->{origin} && $zvol->{origin} =~ m/^$scfg->{pool}\/(\S+)$/){
373 $parent = $1;
374 }
375
1ccae449
DM
376 $list->{$pool}->{$name} = {
377 name => $name,
7730694e
DM
378 size => $zvol->{size},
379 parent => $parent,
1ccae449
DM
380 format => $zvol->{format},
381 vmid => $zvol->{owner},
7730694e
DM
382 };
383 }
384
385 return $list;
386}
387
388sub zfs_find_free_diskname {
55525ad2 389 my ($class, $storeid, $scfg, $vmid, $format) = @_;
7730694e
DM
390
391 my $name = undef;
392 my $volumes = $class->zfs_list_zvol($scfg);
393
394 my $disk_ids = {};
395 my $dat = $volumes->{$scfg->{pool}};
396
397 foreach my $image (keys %$dat) {
398 my $volname = $dat->{$image}->{name};
55525ad2 399 if ($volname =~ m/(vm|base|subvol)-$vmid-disk-(\d+)/){
7730694e
DM
400 $disk_ids->{$2} = 1;
401 }
402 }
403
404 for (my $i = 1; $i < 100; $i++) {
405 if (!$disk_ids->{$i}) {
55525ad2 406 return $format eq 'subvol' ? "subvol-$vmid-disk-$i" : "vm-$vmid-disk-$i";
7730694e
DM
407 }
408 }
409
410 die "unable to allocate an image name for VM $vmid in storage '$storeid'\n";
411}
412
2fc59177
DM
413sub zfs_get_latest_snapshot {
414 my ($class, $scfg, $volname) = @_;
415
416 # abort rollback if snapshot is not the latest
417 my @params = ('-t', 'snapshot', '-o', 'name', '-s', 'creation');
418 my $text = zfs_request($class, $scfg, undef, 'list', @params);
419 my @snapshots = split(/\n/, $text);
420
421 my $recentsnap;
422 foreach (@snapshots) {
423 if (/$scfg->{pool}\/$volname/) {
424 s/^.*@//;
425 $recentsnap = $_;
426 }
427 }
428
429 return $recentsnap;
430}
431
b5e5f7e3
DM
432sub status {
433 my ($class, $storeid, $scfg, $cache) = @_;
434
435 my $total = 0;
436 my $free = 0;
437 my $used = 0;
438 my $active = 0;
439
440 eval {
441 ($free, $used) = $class->zfs_get_pool_stats($scfg);
442 $active = 1;
443 $total = $free + $used;
444 };
445 warn $@ if $@;
446
447 return ($total, $free, $used, $active);
448}
449
450sub volume_size_info {
451 my ($class, $scfg, $storeid, $volname, $timeout) = @_;
452
453 return $class->zfs_get_zvol_size($scfg, $volname);
454}
455
456sub volume_snapshot {
f5640e7d 457 my ($class, $scfg, $storeid, $volname, $snap) = @_;
b5e5f7e3
DM
458
459 $class->zfs_request($scfg, undef, 'snapshot', "$scfg->{pool}/$volname\@$snap");
460}
461
462sub volume_snapshot_delete {
463 my ($class, $scfg, $storeid, $volname, $snap, $running) = @_;
464
399581a2 465 $class->deactivate_volume($storeid, $scfg, $volname, $snap, {});
b5e5f7e3
DM
466 $class->zfs_request($scfg, undef, 'destroy', "$scfg->{pool}/$volname\@$snap");
467}
468
2b40ffae
WL
469sub volume_snapshot_rollback {
470 my ($class, $scfg, $storeid, $volname, $snap) = @_;
471
1597f1f9
WL
472 zfs_request($class, $scfg, undef, 'rollback', "$scfg->{pool}/$volname\@$snap");
473}
474
475sub volume_rollback_is_possible {
476 my ($class, $scfg, $storeid, $volname, $snap) = @_;
477
2fc59177 478 my $recentsnap = $class->zfs_get_latest_snapshot($scfg, $volname);
2b40ffae 479 if ($snap ne $recentsnap) {
1597f1f9 480 die "can't rollback, more recent snapshots exist\n";
2b40ffae
WL
481 }
482
1597f1f9 483 return 1;
2b40ffae
WL
484}
485
0a3d992f
DM
486sub activate_storage {
487 my ($class, $storeid, $scfg, $cache) = @_;
86d47239
WL
488
489 my @param = ('-o', 'name', '-H');
490
491 my $text = zfs_request($class, $scfg, undef, 'zpool_list', @param);
93124ef4
DM
492
493 # Note: $scfg->{pool} can include dataset <pool>/<dataset>
494 my $pool = $scfg->{pool};
495 $pool =~ s!/.*$!!;
496
497 if ($text !~ $pool) {
86d47239
WL
498 run_command("zpool import -d /dev/disk/by-id/ -a");
499 }
0a3d992f
DM
500 return 1;
501}
502
503sub deactivate_storage {
504 my ($class, $storeid, $scfg, $cache) = @_;
505 return 1;
506}
507
d4c63dc1 508sub activate_volume {
02e797b8 509 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
d4c63dc1
WL
510 return 1;
511}
512
513sub deactivate_volume {
02e797b8 514 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
d4c63dc1
WL
515 return 1;
516}
5bb8e010 517
d3a282e8
WL
518sub clone_image {
519 my ($class, $scfg, $storeid, $volname, $vmid, $snap) = @_;
520
521 $snap ||= '__base__';
522
55525ad2 523 my ($vtype, $basename, $basevmid, undef, undef, $isBase, $format) =
d3a282e8
WL
524 $class->parse_volname($volname);
525
526 die "clone_image only works on base images\n" if !$isBase;
527
55525ad2 528 my $name = $class->zfs_find_free_diskname($storeid, $scfg, $vmid, $format);
d3a282e8 529
d3a282e8
WL
530 $class->zfs_request($scfg, undef, 'clone', "$scfg->{pool}/$basename\@$snap", "$scfg->{pool}/$name");
531
532 return $name;
533}
534
535sub create_base {
536 my ($class, $storeid, $scfg, $volname) = @_;
537
538 my $snap = '__base__';
539
540 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
541 $class->parse_volname($volname);
542
543 die "create_base not possible with base image\n" if $isBase;
544
545 my $newname = $name;
546 $newname =~ s/^vm-/base-/;
547
548 my $newvolname = $basename ? "$basename/$newname" : "$newname";
549
550 $class->zfs_request($scfg, undef, 'rename', "$scfg->{pool}/$name", "$scfg->{pool}/$newname");
551
552 my $running = undef; #fixme : is create_base always offline ?
553
554 $class->volume_snapshot($scfg, $storeid, $newname, $snap, $running);
555
556 return $newvolname;
557}
558
a4034b9f
WL
559sub volume_resize {
560 my ($class, $scfg, $storeid, $volname, $size, $running) = @_;
561
562 my $new_size = ($size/1024);
563
564 $class->zfs_request($scfg, undef, 'set', 'volsize=' . $new_size . 'k', "$scfg->{pool}/$volname");
565
566 return $new_size;
567}
568
2b40ffae
WL
569sub volume_has_feature {
570 my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_;
571
572 my $features = {
573 snapshot => { current => 1, snap => 1},
574 clone => { base => 1},
575 template => { current => 1},
576 copy => { base => 1, current => 1},
577 };
578
579 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
580 $class->parse_volname($volname);
581
582 my $key = undef;
583
584 if ($snapname) {
585 $key = 'snap';
586 } else {
587 $key = $isBase ? 'base' : 'current';
588 }
589
590 return 1 if $features->{$feature}->{$key};
591
592 return undef;
593}
594
5bb8e010 5951;