]> git.proxmox.com Git - pve-storage.git/blob - PVE/Storage/RBDPlugin.pm
implement map_volume and unmap_volume
[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, $name) = @_;
76
77 my ($major, undef, undef, undef) = ceph_version();
78 return 1 if $major < 10;
79
80 my $krbd_feature_blacklist = ['deep-flatten', 'fast-diff', 'object-map', 'exclusive-lock'];
81 my (undef, undef, undef, undef, $features) = rbd_volume_info($scfg, $storeid, $name);
82
83 my $active_features = { map { $_ => 1 } @$features };
84 my $incompatible_features = join(',', grep { %$active_features{$_} } @$krbd_feature_blacklist);
85
86 if ($incompatible_features) {
87 my $feature_cmd = &$rbd_cmd($scfg, $storeid, 'feature', 'disable', $name, $incompatible_features);
88 run_rbd_command($feature_cmd, errmsg => "could not disable krbd-incompatible image features of rbd volume $name");
89 }
90 };
91
92 my $ceph_version_parser = sub {
93 my $line = shift;
94 if ($line =~ m/^ceph version ((\d+)\.(\d+)\.(\d+))(?: \([a-fA-F0-9]+\))/) {
95 return ($2, $3, $4, $1);
96 } else {
97 warn "Could not parse Ceph version: '$line'\n";
98 }
99 };
100
101 sub ceph_version {
102 my ($cache) = @_;
103
104 my $version_string = $cache;
105
106 my $major;
107 my $minor;
108 my $bugfix;
109
110 if (defined($version_string)) {
111 ($major, $minor, $bugfix, $version_string) = &$ceph_version_parser($version_string);
112 } else {
113 run_command('ceph --version', outfunc => sub {
114 my $line = shift;
115 ($major, $minor, $bugfix, $version_string) = &$ceph_version_parser($line);
116 });
117 }
118 return undef if !defined($version_string);
119 return wantarray ? ($major, $minor, $bugfix, $version_string) : $version_string;
120 }
121
122 sub run_rbd_command {
123 my ($cmd, %args) = @_;
124
125 my $lasterr;
126 my $errmsg = $args{errmsg} . ": " || "";
127 if (!exists($args{errfunc})) {
128 # ' error: 2014-02-06 11:51:59.839135 7f09f94d0760 -1 librbd: snap_unprotect: can't unprotect;
129 # at least 1 child(ren) in pool cephstor1
130 $args{errfunc} = sub {
131 my $line = shift;
132 if ($line =~ m/^\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}\.\d+ [0-9a-f]+ [\-\d]+ librbd: (.*)$/) {
133 $lasterr = "$1\n";
134 } else {
135 $lasterr = $line;
136 }
137 print STDERR $lasterr;
138 *STDERR->flush();
139 };
140 }
141
142 eval { run_command($cmd, %args); };
143 if (my $err = $@) {
144 die $errmsg . $lasterr if length($lasterr);
145 die $err;
146 }
147
148 return undef;
149 }
150
151 sub rbd_ls {
152 my ($scfg, $storeid) = @_;
153
154 my $cmd = &$rbd_cmd($scfg, $storeid, 'ls', '-l', '--format', 'json');
155 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
156
157 my $raw = '';
158 my $parser = sub { $raw .= shift };
159
160 eval {
161 run_rbd_command($cmd, errmsg => "rbd error", errfunc => sub {}, outfunc => $parser);
162 };
163 my $err = $@;
164
165 die $err if $err && $err !~ m/doesn't contain rbd images/ ;
166
167 my $result;
168 if ($raw eq '') {
169 $result = [];
170 } elsif ($raw =~ m/^(\[.*\])$/s) { # untaint
171 $result = JSON::decode_json($1);
172 } else {
173 die "got unexpected data from rbd ls: '$raw'\n";
174 }
175
176 my $list = {};
177
178 foreach my $el (@$result) {
179 next if defined($el->{snapshot});
180
181 my $image = $el->{image};
182
183 my ($owner) = $image =~ m/^(?:vm|base)-(\d+)-/;
184 next if !defined($owner);
185
186 $list->{$pool}->{$image} = {
187 name => $image,
188 size => $el->{size},
189 parent => $get_parent_image_name->($el->{parent}),
190 vmid => $owner
191 };
192 }
193
194 return $list;
195 }
196
197 sub rbd_volume_info {
198 my ($scfg, $storeid, $volname, $snap) = @_;
199
200 my $cmd = undef;
201
202 my @options = ('info', $volname, '--format', 'json');
203 if ($snap) {
204 push @options, '--snap', $snap;
205 }
206
207 $cmd = &$rbd_cmd($scfg, $storeid, @options);
208
209 my $raw = '';
210 my $parser = sub { $raw .= shift };
211
212 run_rbd_command($cmd, errmsg => "rbd error", errfunc => sub {}, outfunc => $parser);
213
214 my $volume;
215 if ($raw eq '') {
216 $volume = {};
217 } elsif ($raw =~ m/^(\{.*\})$/s) { # untaint
218 $volume = JSON::decode_json($1);
219 } else {
220 die "got unexpected data from rbd info: '$raw'\n";
221 }
222
223 $volume->{parent} = $get_parent_image_name->($volume->{parent});
224 $volume->{protected} = defined($volume->{protected}) && $volume->{protected} eq "true" ? 1 : undef;
225
226 return $volume->@{qw(size parent format protected features)};
227 }
228
229 # Configuration
230
231 sub type {
232 return 'rbd';
233 }
234
235 sub plugindata {
236 return {
237 content => [ {images => 1, rootdir => 1}, { images => 1 }],
238 };
239 }
240
241 sub properties {
242 return {
243 monhost => {
244 description => "IP addresses of monitors (for external clusters).",
245 type => 'string', format => 'pve-storage-portal-dns-list',
246 },
247 pool => {
248 description => "Pool.",
249 type => 'string',
250 },
251 username => {
252 description => "RBD Id.",
253 type => 'string',
254 },
255 authsupported => {
256 description => "Authsupported.",
257 type => 'string',
258 },
259 krbd => {
260 description => "Always access rbd through krbd kernel module.",
261 type => 'boolean',
262 },
263 };
264 }
265
266 sub options {
267 return {
268 nodes => { optional => 1 },
269 disable => { optional => 1 },
270 monhost => { optional => 1},
271 pool => { optional => 1 },
272 username => { optional => 1 },
273 content => { optional => 1 },
274 krbd => { optional => 1 },
275 bwlimit => { optional => 1 },
276 };
277 }
278
279 # Storage implementation
280
281 sub on_add_hook {
282 my ($class, $storeid, $scfg, %param) = @_;
283
284 return if defined($scfg->{monhost}); # nothing to do if not pve managed ceph
285
286 PVE::Storage::CephTools::ceph_create_keyfile($scfg->{type}, $storeid);
287 }
288
289 sub on_delete_hook {
290 my ($class, $storeid, $scfg) = @_;
291
292 return if defined($scfg->{monhost}); # nothing to do if not pve managed ceph
293
294 PVE::Storage::CephTools::ceph_remove_keyfile($scfg->{type}, $storeid);
295 }
296
297 sub parse_volname {
298 my ($class, $volname) = @_;
299
300 if ($volname =~ m/^((base-(\d+)-\S+)\/)?((base)?(vm)?-(\d+)-\S+)$/) {
301 return ('images', $4, $7, $2, $3, $5, 'raw');
302 }
303
304 die "unable to parse rbd volume name '$volname'\n";
305 }
306
307 sub path {
308 my ($class, $scfg, $volname, $storeid, $snapname) = @_;
309
310 my $cmd_option = PVE::Storage::CephTools::ceph_connect_option($scfg, $storeid);
311 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
312 $name .= '@'.$snapname if $snapname;
313
314 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
315 return ("/dev/rbd/$pool/$name", $vmid, $vtype) if $scfg->{krbd};
316
317 my $path = "rbd:$pool/$name";
318
319 $path .= ":conf=$cmd_option->{ceph_conf}" if $cmd_option->{ceph_conf};
320 if (defined($scfg->{monhost})) {
321 my $monhost = PVE::Storage::CephTools::hostlist($scfg->{monhost}, ';');
322 $monhost =~ s/:/\\:/g;
323 $path .= ":mon_host=$monhost";
324 $path .= ":auth_supported=$cmd_option->{auth_supported}";
325 }
326
327 $path .= ":id=$cmd_option->{userid}:keyring=$cmd_option->{keyring}" if ($cmd_option->{keyring});
328
329 return ($path, $vmid, $vtype);
330 }
331
332 my $find_free_diskname = sub {
333 my ($storeid, $scfg, $vmid) = @_;
334
335 my $cmd = &$rbd_cmd($scfg, $storeid, 'ls');
336 my $disk_list = [];
337
338 my $parser = sub {
339 my $line = shift;
340 if ($line =~ m/^(.*)$/) { # untaint
341 push @$disk_list, $1;
342 }
343 };
344
345 eval {
346 run_rbd_command($cmd, errmsg => "rbd error", errfunc => sub {}, outfunc => $parser);
347 };
348 my $err = $@;
349
350 die $err if $err && $err !~ m/doesn't contain rbd images/;
351
352 return PVE::Storage::Plugin::get_next_vm_diskname($disk_list, $storeid, $vmid, undef, $scfg);
353 };
354
355 sub create_base {
356 my ($class, $storeid, $scfg, $volname) = @_;
357
358 my $snap = '__base__';
359
360 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
361 $class->parse_volname($volname);
362
363 die "create_base not possible with base image\n" if $isBase;
364
365 my ($size, $parent, $format, undef) = rbd_volume_info($scfg, $storeid, $name);
366 die "rbd volume info on '$name' failed\n" if !($size);
367
368 die "rbd image must be at format V2" if $format ne "2";
369
370 die "volname '$volname' contains wrong information about parent $parent $basename\n"
371 if $basename && (!$parent || $parent ne $basename."@".$snap);
372
373 my $newname = $name;
374 $newname =~ s/^vm-/base-/;
375
376 my $newvolname = $basename ? "$basename/$newname" : "$newname";
377
378 my $cmd = &$rbd_cmd($scfg, $storeid, 'rename', &$add_pool_to_disk($scfg, $name), &$add_pool_to_disk($scfg, $newname));
379 run_rbd_command($cmd, errmsg => "rbd rename '$name' error");
380
381 my $running = undef; #fixme : is create_base always offline ?
382
383 $class->volume_snapshot($scfg, $storeid, $newname, $snap, $running);
384
385 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $newname, $snap);
386
387 if (!$protected){
388 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'protect', $newname, '--snap', $snap);
389 run_rbd_command($cmd, errmsg => "rbd protect $newname snap '$snap' error");
390 }
391
392 return $newvolname;
393
394 }
395
396 sub clone_image {
397 my ($class, $scfg, $storeid, $volname, $vmid, $snapname) = @_;
398
399 my $snap = '__base__';
400 $snap = $snapname if length $snapname;
401
402 my ($vtype, $basename, $basevmid, undef, undef, $isBase) =
403 $class->parse_volname($volname);
404
405 die "$volname is not a base image and snapname is not provided\n"
406 if !$isBase && !length($snapname);
407
408 my $name = $find_free_diskname->($storeid, $scfg, $vmid);
409
410 warn "clone $volname: $basename snapname $snap to $name\n";
411
412 if (length($snapname)) {
413 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $volname, $snapname);
414
415 if (!$protected) {
416 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'protect', $volname, '--snap', $snapname);
417 run_rbd_command($cmd, errmsg => "rbd protect $volname snap $snapname error");
418 }
419 }
420
421 my $newvol = "$basename/$name";
422 $newvol = $name if length($snapname);
423
424 my $cmd = &$rbd_cmd($scfg, $storeid, 'clone', &$add_pool_to_disk($scfg, $basename),
425 '--snap', $snap, &$add_pool_to_disk($scfg, $name));
426
427 run_rbd_command($cmd, errmsg => "rbd clone '$basename' error");
428
429 return $newvol;
430 }
431
432 sub alloc_image {
433 my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_;
434
435
436 die "illegal name '$name' - should be 'vm-$vmid-*'\n"
437 if $name && $name !~ m/^vm-$vmid-/;
438
439 $name = $find_free_diskname->($storeid, $scfg, $vmid) if !$name;
440
441 my $cmd = &$rbd_cmd($scfg, $storeid, 'create', '--image-format' , 2, '--size', int(($size+1023)/1024), $name);
442 run_rbd_command($cmd, errmsg => "rbd create $name' error");
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 my $get_kernel_device_name = sub {
542 my ($pool, $name) = @_;
543
544 return "/dev/rbd/$pool/$name";
545 };
546
547 sub map_volume {
548 my ($class, $storeid, $scfg, $volname, $snapname) = @_;
549
550 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
551 $name .= '@'.$snapname if $snapname;
552
553 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
554
555 my $kerneldev = $get_kernel_device_name->($pool, $name);
556
557 return $kerneldev if -b $kerneldev; # already mapped
558
559 &$krbd_feature_disable($scfg, $storeid, $name);
560
561 my $cmd = &$rbd_cmd($scfg, $storeid, 'map', $name);
562 run_rbd_command($cmd, errmsg => "can't map rbd volume $name");
563
564 return $kerneldev;
565 }
566
567 sub unmap_volume {
568 my ($class, $storeid, $scfg, $volname, $snapname) = @_;
569
570 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
571 $name .= '@'.$snapname if $snapname;
572
573 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
574
575 my $kerneldev = $get_kernel_device_name->($pool, $name);
576
577 if (-b $kerneldev) {
578 my $cmd = &$rbd_cmd($scfg, $storeid, 'unmap', $kerneldev);
579 run_rbd_command($cmd, errmsg => "can't unmap rbd device $kerneldev");
580 }
581
582 return 1;
583 }
584
585 sub activate_volume {
586 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
587
588 $class->map_volume($storeid, $scfg, $volname, $snapname) if $scfg->{krbd};
589
590 return 1;
591 }
592
593 sub deactivate_volume {
594 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
595
596 $class->unmap_volume($storeid, $scfg, $volname, $snapname);
597
598 return 1;
599 }
600
601 sub volume_size_info {
602 my ($class, $scfg, $storeid, $volname, $timeout) = @_;
603
604 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
605 my ($size, undef) = rbd_volume_info($scfg, $storeid, $name);
606 return $size;
607 }
608
609 sub volume_resize {
610 my ($class, $scfg, $storeid, $volname, $size, $running) = @_;
611
612 return 1 if $running && !$scfg->{krbd}; # FIXME???
613
614 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
615
616 my $cmd = &$rbd_cmd($scfg, $storeid, 'resize', '--allow-shrink', '--size', ($size/1024/1024), $name);
617 run_rbd_command($cmd, errmsg => "rbd resize '$volname' error");
618 return undef;
619 }
620
621 sub volume_snapshot {
622 my ($class, $scfg, $storeid, $volname, $snap) = @_;
623
624 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
625
626 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'create', '--snap', $snap, $name);
627 run_rbd_command($cmd, errmsg => "rbd snapshot '$volname' error");
628 return undef;
629 }
630
631 sub volume_snapshot_rollback {
632 my ($class, $scfg, $storeid, $volname, $snap) = @_;
633
634 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
635
636 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'rollback', '--snap', $snap, $name);
637 run_rbd_command($cmd, errmsg => "rbd snapshot $volname to '$snap' error");
638 }
639
640 sub volume_snapshot_delete {
641 my ($class, $scfg, $storeid, $volname, $snap, $running) = @_;
642
643 return 1 if $running && !$scfg->{krbd}; # FIXME: ????
644
645 $class->deactivate_volume($storeid, $scfg, $volname, $snap, {});
646
647 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
648
649 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $name, $snap);
650 if ($protected){
651 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'unprotect', $name, '--snap', $snap);
652 run_rbd_command($cmd, errmsg => "rbd unprotect $name snap '$snap' error");
653 }
654
655 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'rm', '--snap', $snap, $name);
656
657 run_rbd_command($cmd, errmsg => "rbd snapshot '$volname' error");
658
659 return undef;
660 }
661
662 sub volume_has_feature {
663 my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_;
664
665 my $features = {
666 snapshot => { current => 1, snap => 1},
667 clone => { base => 1, snap => 1},
668 template => { current => 1},
669 copy => { base => 1, current => 1, snap => 1},
670 sparseinit => { base => 1, current => 1},
671 };
672
673 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
674 $class->parse_volname($volname);
675
676 my $key = undef;
677 if($snapname){
678 $key = 'snap';
679 }else{
680 $key = $isBase ? 'base' : 'current';
681 }
682 return 1 if $features->{$feature}->{$key};
683
684 return undef;
685 }
686
687 1;