]> git.proxmox.com Git - pve-storage.git/blob - PVE/Storage/RBDPlugin.pm
bd8c28d86e3f25815c42e0960d038015974b2428
[pve-storage.git] / PVE / Storage / RBDPlugin.pm
1 package PVE::Storage::RBDPlugin;
2
3 use strict;
4 use warnings;
5 use IO::File;
6 use Net::IP;
7 use PVE::Tools qw(run_command trim);
8 use PVE::Storage::Plugin;
9 use PVE::JSONSchema qw(get_standard_option);
10 use PVE::RADOS;
11 use PVE::Storage::CephTools;
12 use JSON;
13
14 use base qw(PVE::Storage::Plugin);
15
16 my $get_parent_image_name = sub {
17 my ($parent) = @_;
18 return undef if !$parent;
19 return $parent->{image} . "@" . $parent->{snapshot};
20 };
21
22 my $add_pool_to_disk = sub {
23 my ($scfg, $disk) = @_;
24
25 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
26
27 return "$pool/$disk";
28 };
29
30 my $build_cmd = sub {
31 my ($binary, $scfg, $storeid, $op, @options) = @_;
32
33 my $cmd_option = PVE::Storage::CephTools::ceph_connect_option($scfg, $storeid);
34 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
35
36 my $cmd = [$binary, '-p', $pool];
37
38 push @$cmd, '-c', $cmd_option->{ceph_conf} if ($cmd_option->{ceph_conf});
39 push @$cmd, '-m', $cmd_option->{mon_host} if ($cmd_option->{mon_host});
40 push @$cmd, '--auth_supported', $cmd_option->{auth_supported} if ($cmd_option->{auth_supported});
41 push @$cmd, '-n', "client.$cmd_option->{userid}" if ($cmd_option->{userid});
42 push @$cmd, '--keyring', $cmd_option->{keyring} if ($cmd_option->{keyring});
43
44 push @$cmd, $op;
45
46 push @$cmd, @options if scalar(@options);
47
48 return $cmd;
49 };
50
51 my $rbd_cmd = sub {
52 my ($scfg, $storeid, $op, @options) = @_;
53
54 return $build_cmd->('/usr/bin/rbd', $scfg, $storeid, $op, @options);
55 };
56
57 my $rados_cmd = sub {
58 my ($scfg, $storeid, $op, @options) = @_;
59
60 return $build_cmd->('/usr/bin/rados', $scfg, $storeid, $op, @options);
61 };
62
63 my $librados_connect = sub {
64 my ($scfg, $storeid, $options) = @_;
65
66 my $librados_config = PVE::Storage::CephTools::ceph_connect_option($scfg, $storeid);
67
68 my $rados = PVE::RADOS->new(%$librados_config);
69
70 return $rados;
71 };
72
73 # needed for volumes created using ceph jewel (or higher)
74 my $krbd_feature_disable = sub {
75 my ($scfg, $storeid, $name) = @_;
76
77 return 1 if !$scfg->{krbd};
78
79 my ($major, undef, undef, undef) = ceph_version();
80 return 1 if $major < 10;
81
82 my $krbd_feature_blacklist = ['deep-flatten', 'fast-diff', 'object-map', 'exclusive-lock'];
83 my (undef, undef, undef, undef, $features) = rbd_volume_info($scfg, $storeid, $name);
84
85 my $active_features = { map { $_ => 1 } $features };
86 my $incompatible_features = join(',', grep { %$active_features{$_} } @$krbd_feature_blacklist);
87
88 if ($incompatible_features) {
89 my $feature_cmd = &$rbd_cmd($scfg, $storeid, 'feature', 'disable', $name, $incompatible_features);
90 run_rbd_command($feature_cmd, errmsg => "could not disable krbd-incompatible image features of rbd volume $name");
91 }
92 };
93
94 my $ceph_version_parser = sub {
95 my $line = shift;
96 if ($line =~ m/^ceph version ((\d+)\.(\d+)\.(\d+))(?: \([a-fA-F0-9]+\))/) {
97 return ($2, $3, $4, $1);
98 } else {
99 warn "Could not parse Ceph version: '$line'\n";
100 }
101 };
102
103 sub ceph_version {
104 my ($cache) = @_;
105
106 my $version_string = $cache;
107
108 my $major;
109 my $minor;
110 my $bugfix;
111
112 if (defined($version_string)) {
113 ($major, $minor, $bugfix, $version_string) = &$ceph_version_parser($version_string);
114 } else {
115 run_command('ceph --version', outfunc => sub {
116 my $line = shift;
117 ($major, $minor, $bugfix, $version_string) = &$ceph_version_parser($line);
118 });
119 }
120 return undef if !defined($version_string);
121 return wantarray ? ($major, $minor, $bugfix, $version_string) : $version_string;
122 }
123
124 sub run_rbd_command {
125 my ($cmd, %args) = @_;
126
127 my $lasterr;
128 my $errmsg = $args{errmsg} . ": " || "";
129 if (!exists($args{errfunc})) {
130 # ' error: 2014-02-06 11:51:59.839135 7f09f94d0760 -1 librbd: snap_unprotect: can't unprotect;
131 # at least 1 child(ren) in pool cephstor1
132 $args{errfunc} = sub {
133 my $line = shift;
134 if ($line =~ m/^\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}\.\d+ [0-9a-f]+ [\-\d]+ librbd: (.*)$/) {
135 $lasterr = "$1\n";
136 } else {
137 $lasterr = $line;
138 }
139 print STDERR $lasterr;
140 *STDERR->flush();
141 };
142 }
143
144 eval { run_command($cmd, %args); };
145 if (my $err = $@) {
146 die $errmsg . $lasterr if length($lasterr);
147 die $err;
148 }
149
150 return undef;
151 }
152
153 sub rbd_ls {
154 my ($scfg, $storeid) = @_;
155
156 my $cmd = &$rbd_cmd($scfg, $storeid, 'ls', '-l', '--format', 'json');
157 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
158
159 my $raw = '';
160 my $parser = sub { $raw .= shift };
161
162 eval {
163 run_rbd_command($cmd, errmsg => "rbd error", errfunc => sub {}, outfunc => $parser);
164 };
165 my $err = $@;
166
167 die $err if $err && $err !~ m/doesn't contain rbd images/ ;
168
169 my $result = $raw ne '' ? JSON::decode_json($raw) : [];
170
171 my $list = {};
172
173 foreach my $el (@$result) {
174 next if defined($el->{snapshot});
175
176 my $image = $el->{image};
177
178 my ($owner) = $image =~ m/^(?:vm|base)-(\d+)-/;
179 next if !defined($owner);
180
181 $list->{$pool}->{$image} = {
182 name => $image,
183 size => $el->{size},
184 parent => $get_parent_image_name->($el->{parent}),
185 vmid => $owner
186 };
187 }
188
189 return $list;
190 }
191
192 sub rbd_volume_info {
193 my ($scfg, $storeid, $volname, $snap) = @_;
194
195 my $cmd = undef;
196
197 my @options = ('info', $volname, '--format', 'json');
198 if ($snap) {
199 push @options, '--snap', $snap;
200 }
201
202 $cmd = &$rbd_cmd($scfg, $storeid, @options);
203
204 my $raw = '';
205 my $parser = sub { $raw .= shift };
206
207 run_rbd_command($cmd, errmsg => "rbd error", errfunc => sub {}, outfunc => $parser);
208
209 my $volume = $raw ne '' ? JSON::decode_json($raw) : {};
210
211 $volume->{parent} = $get_parent_image_name->($volume->{parent});
212 $volume->{protected} = defined($volume->{protected}) && $volume->{protected} eq "true" ? 1 : undef;
213
214 return $volume->@{qw(size parent format protected features)};
215 }
216
217 # Configuration
218
219 sub type {
220 return 'rbd';
221 }
222
223 sub plugindata {
224 return {
225 content => [ {images => 1, rootdir => 1}, { images => 1 }],
226 };
227 }
228
229 sub properties {
230 return {
231 monhost => {
232 description => "IP addresses of monitors (for external clusters).",
233 type => 'string', format => 'pve-storage-portal-dns-list',
234 },
235 pool => {
236 description => "Pool.",
237 type => 'string',
238 },
239 username => {
240 description => "RBD Id.",
241 type => 'string',
242 },
243 authsupported => {
244 description => "Authsupported.",
245 type => 'string',
246 },
247 krbd => {
248 description => "Access rbd through krbd kernel module.",
249 type => 'boolean',
250 },
251 };
252 }
253
254 sub options {
255 return {
256 nodes => { optional => 1 },
257 disable => { optional => 1 },
258 monhost => { optional => 1},
259 pool => { optional => 1 },
260 username => { optional => 1 },
261 content => { optional => 1 },
262 krbd => { optional => 1 },
263 bwlimit => { optional => 1 },
264 };
265 }
266
267 # Storage implementation
268
269 sub on_add_hook {
270 my ($class, $storeid, $scfg, %param) = @_;
271
272 return if defined($scfg->{monhost}); # nothing to do if not pve managed ceph
273
274 PVE::Storage::CephTools::ceph_create_keyfile($scfg->{type}, $storeid);
275 }
276
277 sub on_delete_hook {
278 my ($class, $storeid, $scfg) = @_;
279
280 return if defined($scfg->{monhost}); # nothing to do if not pve managed ceph
281
282 PVE::Storage::CephTools::ceph_remove_keyfile($scfg->{type}, $storeid);
283 }
284
285 sub parse_volname {
286 my ($class, $volname) = @_;
287
288 if ($volname =~ m/^((base-(\d+)-\S+)\/)?((base)?(vm)?-(\d+)-\S+)$/) {
289 return ('images', $4, $7, $2, $3, $5, 'raw');
290 }
291
292 die "unable to parse rbd volume name '$volname'\n";
293 }
294
295 sub path {
296 my ($class, $scfg, $volname, $storeid, $snapname) = @_;
297
298 my $cmd_option = PVE::Storage::CephTools::ceph_connect_option($scfg, $storeid);
299 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
300 $name .= '@'.$snapname if $snapname;
301
302 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
303 return ("/dev/rbd/$pool/$name", $vmid, $vtype) if $scfg->{krbd};
304
305 my $path = "rbd:$pool/$name";
306
307 $path .= ":conf=$cmd_option->{ceph_conf}" if $cmd_option->{ceph_conf};
308 if (defined($scfg->{monhost})) {
309 my $monhost = PVE::Storage::CephTools::hostlist($scfg->{monhost}, ';');
310 $monhost =~ s/:/\\:/g;
311 $path .= ":mon_host=$monhost";
312 $path .= ":auth_supported=$cmd_option->{auth_supported}";
313 }
314
315 $path .= ":id=$cmd_option->{userid}:keyring=$cmd_option->{keyring}" if ($cmd_option->{keyring});
316
317 return ($path, $vmid, $vtype);
318 }
319
320 my $find_free_diskname = sub {
321 my ($storeid, $scfg, $vmid) = @_;
322
323 my $cmd = &$rbd_cmd($scfg, $storeid, 'ls');
324 my $disk_ids = {};
325
326 my $parser = sub {
327 my $line = shift;
328
329 if ($line =~ m/^(vm|base)-\Q$vmid\E+-disk-(\d+)$/) {
330 $disk_ids->{$2} = 1;
331 }
332 };
333
334 eval {
335 run_rbd_command($cmd, errmsg => "rbd error", errfunc => sub {}, outfunc => $parser);
336 };
337 my $err = $@;
338
339 die $err if $err && $err !~ m/doesn't contain rbd images/;
340
341 #fix: can we search in $rbd hash key with a regex to find (vm|base) ?
342 for (my $i = 1; $i < 100; $i++) {
343 if (!$disk_ids->{$i}) {
344 return "vm-$vmid-disk-$i";
345 }
346 }
347
348 die "unable to allocate an image name for VM $vmid in storage '$storeid'\n";
349 };
350
351 sub create_base {
352 my ($class, $storeid, $scfg, $volname) = @_;
353
354 my $snap = '__base__';
355
356 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
357 $class->parse_volname($volname);
358
359 die "create_base not possible with base image\n" if $isBase;
360
361 my ($size, $parent, $format, undef) = rbd_volume_info($scfg, $storeid, $name);
362 die "rbd volume info on '$name' failed\n" if !($size);
363
364 die "rbd image must be at format V2" if $format ne "2";
365
366 die "volname '$volname' contains wrong information about parent $parent $basename\n"
367 if $basename && (!$parent || $parent ne $basename."@".$snap);
368
369 my $newname = $name;
370 $newname =~ s/^vm-/base-/;
371
372 my $newvolname = $basename ? "$basename/$newname" : "$newname";
373
374 my $cmd = &$rbd_cmd($scfg, $storeid, 'rename', &$add_pool_to_disk($scfg, $name), &$add_pool_to_disk($scfg, $newname));
375 run_rbd_command($cmd, errmsg => "rbd rename '$name' error");
376
377 my $running = undef; #fixme : is create_base always offline ?
378
379 $class->volume_snapshot($scfg, $storeid, $newname, $snap, $running);
380
381 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $newname, $snap);
382
383 if (!$protected){
384 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'protect', $newname, '--snap', $snap);
385 run_rbd_command($cmd, errmsg => "rbd protect $newname snap '$snap' error");
386 }
387
388 return $newvolname;
389
390 }
391
392 sub clone_image {
393 my ($class, $scfg, $storeid, $volname, $vmid, $snapname) = @_;
394
395 my $snap = '__base__';
396 $snap = $snapname if length $snapname;
397
398 my ($vtype, $basename, $basevmid, undef, undef, $isBase) =
399 $class->parse_volname($volname);
400
401 die "$volname is not a base image and snapname is not provided\n"
402 if !$isBase && !length($snapname);
403
404 my $name = &$find_free_diskname($storeid, $scfg, $vmid);
405
406 warn "clone $volname: $basename snapname $snap to $name\n";
407
408 if (length($snapname)) {
409 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $volname, $snapname);
410
411 if (!$protected) {
412 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'protect', $volname, '--snap', $snapname);
413 run_rbd_command($cmd, errmsg => "rbd protect $volname snap $snapname error");
414 }
415 }
416
417 my $newvol = "$basename/$name";
418 $newvol = $name if length($snapname);
419
420 my $cmd = &$rbd_cmd($scfg, $storeid, 'clone', &$add_pool_to_disk($scfg, $basename),
421 '--snap', $snap, &$add_pool_to_disk($scfg, $name));
422
423 run_rbd_command($cmd, errmsg => "rbd clone '$basename' error");
424
425 &$krbd_feature_disable($scfg, $storeid, $name);
426
427 return $newvol;
428 }
429
430 sub alloc_image {
431 my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_;
432
433
434 die "illegal name '$name' - should be 'vm-$vmid-*'\n"
435 if $name && $name !~ m/^vm-$vmid-/;
436
437 $name = &$find_free_diskname($storeid, $scfg, $vmid) if !$name;
438
439 my $cmd = &$rbd_cmd($scfg, $storeid, 'create', '--image-format' , 2, '--size', int(($size+1023)/1024), $name);
440 run_rbd_command($cmd, errmsg => "rbd create $name' error");
441
442 &$krbd_feature_disable($scfg, $storeid, $name);
443
444 return $name;
445 }
446
447 sub free_image {
448 my ($class, $storeid, $scfg, $volname, $isBase) = @_;
449
450 my ($vtype, $name, $vmid, undef, undef, undef) =
451 $class->parse_volname($volname);
452
453 if ($isBase) {
454 my $snap = '__base__';
455 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $name, $snap);
456 if ($protected){
457 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'unprotect', $name, '--snap', $snap);
458 run_rbd_command($cmd, errmsg => "rbd unprotect $name snap '$snap' error");
459 }
460 }
461
462 $class->deactivate_volume($storeid, $scfg, $volname);
463
464 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'purge', $name);
465 run_rbd_command($cmd, errmsg => "rbd snap purge '$volname' error");
466
467 $cmd = &$rbd_cmd($scfg, $storeid, 'rm', $name);
468 run_rbd_command($cmd, errmsg => "rbd rm '$volname' error");
469
470 return undef;
471 }
472
473 sub list_images {
474 my ($class, $storeid, $scfg, $vmid, $vollist, $cache) = @_;
475
476 $cache->{rbd} = rbd_ls($scfg, $storeid) if !$cache->{rbd};
477 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
478
479 my $res = [];
480
481 if (my $dat = $cache->{rbd}->{$pool}) {
482 foreach my $image (keys %$dat) {
483
484 my $info = $dat->{$image};
485
486 my $volname = $info->{name};
487 my $parent = $info->{parent};
488 my $owner = $info->{vmid};
489
490 if ($parent && $parent =~ m/^(base-\d+-\S+)\@__base__$/) {
491 $info->{volid} = "$storeid:$1/$volname";
492 } else {
493 $info->{volid} = "$storeid:$volname";
494 }
495
496 if ($vollist) {
497 my $found = grep { $_ eq $info->{volid} } @$vollist;
498 next if !$found;
499 } else {
500 next if defined ($vmid) && ($owner ne $vmid);
501 }
502
503 $info->{format} = 'raw';
504
505 push @$res, $info;
506 }
507 }
508
509 return $res;
510 }
511
512 sub status {
513 my ($class, $storeid, $scfg, $cache) = @_;
514
515
516 my $rados = &$librados_connect($scfg, $storeid);
517 my $df = $rados->mon_command({ prefix => 'df', format => 'json' });
518
519 my ($d) = grep { $_->{name} eq $scfg->{pool} } @{$df->{pools}};
520
521 # max_avail -> max available space for data w/o replication in the pool
522 # bytes_used -> data w/o replication in the pool
523 my $free = $d->{stats}->{max_avail};
524 my $used = $d->{stats}->{bytes_used};
525 my $total = $used + $free;
526 my $active = 1;
527
528 return ($total, $free, $used, $active);
529 }
530
531 sub activate_storage {
532 my ($class, $storeid, $scfg, $cache) = @_;
533 return 1;
534 }
535
536 sub deactivate_storage {
537 my ($class, $storeid, $scfg, $cache) = @_;
538 return 1;
539 }
540
541 sub activate_volume {
542 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
543
544 return 1 if !$scfg->{krbd};
545
546 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
547 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
548
549 my $path = "/dev/rbd/$pool/$name";
550 $path .= '@'.$snapname if $snapname;
551 return if -b $path;
552
553 $name .= '@'.$snapname if $snapname;
554 my $cmd = &$rbd_cmd($scfg, $storeid, 'map', $name);
555 run_rbd_command($cmd, errmsg => "can't mount rbd volume $name");
556
557 return 1;
558 }
559
560 sub deactivate_volume {
561 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
562
563 return 1 if !$scfg->{krbd};
564
565 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
566 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
567
568 my $path = "/dev/rbd/$pool/$name";
569 $path .= '@'.$snapname if $snapname;
570 return if ! -b $path;
571
572 my $cmd = &$rbd_cmd($scfg, $storeid, 'unmap', $path);
573 run_rbd_command($cmd, errmsg => "can't unmap rbd volume $name");
574
575 return 1;
576 }
577
578 sub volume_size_info {
579 my ($class, $scfg, $storeid, $volname, $timeout) = @_;
580
581 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
582 my ($size, undef) = rbd_volume_info($scfg, $storeid, $name);
583 return $size;
584 }
585
586 sub volume_resize {
587 my ($class, $scfg, $storeid, $volname, $size, $running) = @_;
588
589 return 1 if $running && !$scfg->{krbd};
590
591 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
592
593 my $cmd = &$rbd_cmd($scfg, $storeid, 'resize', '--allow-shrink', '--size', ($size/1024/1024), $name);
594 run_rbd_command($cmd, errmsg => "rbd resize '$volname' error");
595 return undef;
596 }
597
598 sub volume_snapshot {
599 my ($class, $scfg, $storeid, $volname, $snap) = @_;
600
601 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
602
603 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'create', '--snap', $snap, $name);
604 run_rbd_command($cmd, errmsg => "rbd snapshot '$volname' error");
605 return undef;
606 }
607
608 sub volume_snapshot_rollback {
609 my ($class, $scfg, $storeid, $volname, $snap) = @_;
610
611 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
612
613 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'rollback', '--snap', $snap, $name);
614 run_rbd_command($cmd, errmsg => "rbd snapshot $volname to '$snap' error");
615 }
616
617 sub volume_snapshot_delete {
618 my ($class, $scfg, $storeid, $volname, $snap, $running) = @_;
619
620 return 1 if $running && !$scfg->{krbd};
621
622 $class->deactivate_volume($storeid, $scfg, $volname, $snap, {});
623
624 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
625
626 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $name, $snap);
627 if ($protected){
628 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'unprotect', $name, '--snap', $snap);
629 run_rbd_command($cmd, errmsg => "rbd unprotect $name snap '$snap' error");
630 }
631
632 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'rm', '--snap', $snap, $name);
633
634 run_rbd_command($cmd, errmsg => "rbd snapshot '$volname' error");
635
636 return undef;
637 }
638
639 sub volume_has_feature {
640 my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_;
641
642 my $features = {
643 snapshot => { current => 1, snap => 1},
644 clone => { base => 1, snap => 1},
645 template => { current => 1},
646 copy => { base => 1, current => 1, snap => 1},
647 sparseinit => { base => 1, current => 1},
648 };
649
650 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
651 $class->parse_volname($volname);
652
653 my $key = undef;
654 if($snapname){
655 $key = 'snap';
656 }else{
657 $key = $isBase ? 'base' : 'current';
658 }
659 return 1 if $features->{$feature}->{$key};
660
661 return undef;
662 }
663
664 1;