]> git.proxmox.com Git - pve-storage.git/blob - PVE/Storage/RBDPlugin.pm
rdb: followup cleanups
[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
180 $list->{$pool}->{$image} = {
181 name => $image,
182 size => $el->{size},
183 parent => $get_parent_image_name->($el->{parent}),
184 vmid => $owner
185 };
186 }
187
188 return $list;
189 }
190
191 sub rbd_volume_info {
192 my ($scfg, $storeid, $volname, $snap) = @_;
193
194 my $cmd = undef;
195
196 my @options = ('info', $volname, '--format', 'json');
197 if ($snap) {
198 push @options, '--snap', $snap;
199 }
200
201 $cmd = &$rbd_cmd($scfg, $storeid, @options);
202
203 my $raw = '';
204 my $parser = sub { $raw .= shift };
205
206 run_rbd_command($cmd, errmsg => "rbd error", errfunc => sub {}, outfunc => $parser);
207
208 my $volume = $raw ne '' ? JSON::decode_json($raw) : {};
209
210 $volume->{parent} = $get_parent_image_name->($volume->{parent});
211 $volume->{protected} = defined($volume->{protected}) && $volume->{protected} eq "true" ? 1 : undef;
212
213 return $volume->@{qw(size parent format protected features)};
214 }
215
216 # Configuration
217
218 sub type {
219 return 'rbd';
220 }
221
222 sub plugindata {
223 return {
224 content => [ {images => 1, rootdir => 1}, { images => 1 }],
225 };
226 }
227
228 sub properties {
229 return {
230 monhost => {
231 description => "IP addresses of monitors (for external clusters).",
232 type => 'string', format => 'pve-storage-portal-dns-list',
233 },
234 pool => {
235 description => "Pool.",
236 type => 'string',
237 },
238 username => {
239 description => "RBD Id.",
240 type => 'string',
241 },
242 authsupported => {
243 description => "Authsupported.",
244 type => 'string',
245 },
246 krbd => {
247 description => "Access rbd through krbd kernel module.",
248 type => 'boolean',
249 },
250 };
251 }
252
253 sub options {
254 return {
255 nodes => { optional => 1 },
256 disable => { optional => 1 },
257 monhost => { optional => 1},
258 pool => { optional => 1 },
259 username => { optional => 1 },
260 content => { optional => 1 },
261 krbd => { optional => 1 },
262 bwlimit => { optional => 1 },
263 };
264 }
265
266 # Storage implementation
267
268 sub on_add_hook {
269 my ($class, $storeid, $scfg, %param) = @_;
270
271 return if defined($scfg->{monhost}); # nothing to do if not pve managed ceph
272
273 PVE::Storage::CephTools::ceph_create_keyfile($scfg->{type}, $storeid);
274 }
275
276 sub on_delete_hook {
277 my ($class, $storeid, $scfg) = @_;
278
279 return if defined($scfg->{monhost}); # nothing to do if not pve managed ceph
280
281 PVE::Storage::CephTools::ceph_remove_keyfile($scfg->{type}, $storeid);
282 }
283
284 sub parse_volname {
285 my ($class, $volname) = @_;
286
287 if ($volname =~ m/^((base-(\d+)-\S+)\/)?((base)?(vm)?-(\d+)-\S+)$/) {
288 return ('images', $4, $7, $2, $3, $5, 'raw');
289 }
290
291 die "unable to parse rbd volume name '$volname'\n";
292 }
293
294 sub path {
295 my ($class, $scfg, $volname, $storeid, $snapname) = @_;
296
297 my $cmd_option = PVE::Storage::CephTools::ceph_connect_option($scfg, $storeid);
298 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
299 $name .= '@'.$snapname if $snapname;
300
301 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
302 return ("/dev/rbd/$pool/$name", $vmid, $vtype) if $scfg->{krbd};
303
304 my $path = "rbd:$pool/$name";
305
306 $path .= ":conf=$cmd_option->{ceph_conf}" if $cmd_option->{ceph_conf};
307 if (defined($scfg->{monhost})) {
308 my $monhost = PVE::Storage::CephTools::hostlist($scfg->{monhost}, ';');
309 $monhost =~ s/:/\\:/g;
310 $path .= ":mon_host=$monhost";
311 $path .= ":auth_supported=$cmd_option->{auth_supported}";
312 }
313
314 $path .= ":id=$cmd_option->{userid}:keyring=$cmd_option->{keyring}" if ($cmd_option->{keyring});
315
316 return ($path, $vmid, $vtype);
317 }
318
319 my $find_free_diskname = sub {
320 my ($storeid, $scfg, $vmid) = @_;
321
322 my $cmd = &$rbd_cmd($scfg, $storeid, 'ls');
323 my $disk_ids = {};
324
325 my $parser = sub {
326 my $line = shift;
327
328 if ($line =~ m/^(vm|base)-\Q$vmid\E+-disk-(\d+)$/) {
329 $disk_ids->{$2} = 1;
330 }
331 };
332
333 eval {
334 run_rbd_command($cmd, errmsg => "rbd error", errfunc => sub {}, outfunc => $parser);
335 };
336 my $err = $@;
337
338 die $err if $err && $err !~ m/doesn't contain rbd images/;
339
340 #fix: can we search in $rbd hash key with a regex to find (vm|base) ?
341 for (my $i = 1; $i < 100; $i++) {
342 if (!$disk_ids->{$i}) {
343 return "vm-$vmid-disk-$i";
344 }
345 }
346
347 die "unable to allocate an image name for VM $vmid in storage '$storeid'\n";
348 };
349
350 sub create_base {
351 my ($class, $storeid, $scfg, $volname) = @_;
352
353 my $snap = '__base__';
354
355 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
356 $class->parse_volname($volname);
357
358 die "create_base not possible with base image\n" if $isBase;
359
360 my ($size, $parent, $format, undef) = rbd_volume_info($scfg, $storeid, $name);
361 die "rbd volume info on '$name' failed\n" if !($size);
362
363 die "rbd image must be at format V2" if $format ne "2";
364
365 die "volname '$volname' contains wrong information about parent $parent $basename\n"
366 if $basename && (!$parent || $parent ne $basename."@".$snap);
367
368 my $newname = $name;
369 $newname =~ s/^vm-/base-/;
370
371 my $newvolname = $basename ? "$basename/$newname" : "$newname";
372
373 my $cmd = &$rbd_cmd($scfg, $storeid, 'rename', &$add_pool_to_disk($scfg, $name), &$add_pool_to_disk($scfg, $newname));
374 run_rbd_command($cmd, errmsg => "rbd rename '$name' error");
375
376 my $running = undef; #fixme : is create_base always offline ?
377
378 $class->volume_snapshot($scfg, $storeid, $newname, $snap, $running);
379
380 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $newname, $snap);
381
382 if (!$protected){
383 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'protect', $newname, '--snap', $snap);
384 run_rbd_command($cmd, errmsg => "rbd protect $newname snap '$snap' error");
385 }
386
387 return $newvolname;
388
389 }
390
391 sub clone_image {
392 my ($class, $scfg, $storeid, $volname, $vmid, $snapname) = @_;
393
394 my $snap = '__base__';
395 $snap = $snapname if length $snapname;
396
397 my ($vtype, $basename, $basevmid, undef, undef, $isBase) =
398 $class->parse_volname($volname);
399
400 die "$volname is not a base image and snapname is not provided\n"
401 if !$isBase && !length($snapname);
402
403 my $name = &$find_free_diskname($storeid, $scfg, $vmid);
404
405 warn "clone $volname: $basename snapname $snap to $name\n";
406
407 if (length($snapname)) {
408 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $volname, $snapname);
409
410 if (!$protected) {
411 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'protect', $volname, '--snap', $snapname);
412 run_rbd_command($cmd, errmsg => "rbd protect $volname snap $snapname error");
413 }
414 }
415
416 my $newvol = "$basename/$name";
417 $newvol = $name if length($snapname);
418
419 my $cmd = &$rbd_cmd($scfg, $storeid, 'clone', &$add_pool_to_disk($scfg, $basename),
420 '--snap', $snap, &$add_pool_to_disk($scfg, $name));
421
422 run_rbd_command($cmd, errmsg => "rbd clone '$basename' error");
423
424 &$krbd_feature_disable($scfg, $storeid, $name);
425
426 return $newvol;
427 }
428
429 sub alloc_image {
430 my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_;
431
432
433 die "illegal name '$name' - should be 'vm-$vmid-*'\n"
434 if $name && $name !~ m/^vm-$vmid-/;
435
436 $name = &$find_free_diskname($storeid, $scfg, $vmid) if !$name;
437
438 my $cmd = &$rbd_cmd($scfg, $storeid, 'create', '--image-format' , 2, '--size', int(($size+1023)/1024), $name);
439 run_rbd_command($cmd, errmsg => "rbd create $name' error");
440
441 &$krbd_feature_disable($scfg, $storeid, $name);
442
443 return $name;
444 }
445
446 sub free_image {
447 my ($class, $storeid, $scfg, $volname, $isBase) = @_;
448
449 my ($vtype, $name, $vmid, undef, undef, undef) =
450 $class->parse_volname($volname);
451
452 if ($isBase) {
453 my $snap = '__base__';
454 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $name, $snap);
455 if ($protected){
456 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'unprotect', $name, '--snap', $snap);
457 run_rbd_command($cmd, errmsg => "rbd unprotect $name snap '$snap' error");
458 }
459 }
460
461 $class->deactivate_volume($storeid, $scfg, $volname);
462
463 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'purge', $name);
464 run_rbd_command($cmd, errmsg => "rbd snap purge '$volname' error");
465
466 $cmd = &$rbd_cmd($scfg, $storeid, 'rm', $name);
467 run_rbd_command($cmd, errmsg => "rbd rm '$volname' error");
468
469 return undef;
470 }
471
472 sub list_images {
473 my ($class, $storeid, $scfg, $vmid, $vollist, $cache) = @_;
474
475 $cache->{rbd} = rbd_ls($scfg, $storeid) if !$cache->{rbd};
476 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
477
478 my $res = [];
479
480 if (my $dat = $cache->{rbd}->{$pool}) {
481 foreach my $image (keys %$dat) {
482
483 my $info = $dat->{$image};
484
485 my $volname = $info->{name};
486 my $parent = $info->{parent};
487 my $owner = $info->{vmid};
488
489 if ($parent && $parent =~ m/^(base-\d+-\S+)\@__base__$/) {
490 $info->{volid} = "$storeid:$1/$volname";
491 } else {
492 $info->{volid} = "$storeid:$volname";
493 }
494
495 if ($vollist) {
496 my $found = grep { $_ eq $info->{volid} } @$vollist;
497 next if !$found;
498 } else {
499 next if defined ($vmid) && ($owner ne $vmid);
500 }
501
502 $info->{format} = 'raw';
503
504 push @$res, $info;
505 }
506 }
507
508 return $res;
509 }
510
511 sub status {
512 my ($class, $storeid, $scfg, $cache) = @_;
513
514
515 my $rados = &$librados_connect($scfg, $storeid);
516 my $df = $rados->mon_command({ prefix => 'df', format => 'json' });
517
518 my ($d) = grep { $_->{name} eq $scfg->{pool} } @{$df->{pools}};
519
520 # max_avail -> max available space for data w/o replication in the pool
521 # bytes_used -> data w/o replication in the pool
522 my $free = $d->{stats}->{max_avail};
523 my $used = $d->{stats}->{bytes_used};
524 my $total = $used + $free;
525 my $active = 1;
526
527 return ($total, $free, $used, $active);
528 }
529
530 sub activate_storage {
531 my ($class, $storeid, $scfg, $cache) = @_;
532 return 1;
533 }
534
535 sub deactivate_storage {
536 my ($class, $storeid, $scfg, $cache) = @_;
537 return 1;
538 }
539
540 sub activate_volume {
541 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
542
543 return 1 if !$scfg->{krbd};
544
545 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
546 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
547
548 my $path = "/dev/rbd/$pool/$name";
549 $path .= '@'.$snapname if $snapname;
550 return if -b $path;
551
552 $name .= '@'.$snapname if $snapname;
553 my $cmd = &$rbd_cmd($scfg, $storeid, 'map', $name);
554 run_rbd_command($cmd, errmsg => "can't mount rbd volume $name");
555
556 return 1;
557 }
558
559 sub deactivate_volume {
560 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
561
562 return 1 if !$scfg->{krbd};
563
564 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
565 my $pool = $scfg->{pool} ? $scfg->{pool} : 'rbd';
566
567 my $path = "/dev/rbd/$pool/$name";
568 $path .= '@'.$snapname if $snapname;
569 return if ! -b $path;
570
571 my $cmd = &$rbd_cmd($scfg, $storeid, 'unmap', $path);
572 run_rbd_command($cmd, errmsg => "can't unmap rbd volume $name");
573
574 return 1;
575 }
576
577 sub volume_size_info {
578 my ($class, $scfg, $storeid, $volname, $timeout) = @_;
579
580 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
581 my ($size, undef) = rbd_volume_info($scfg, $storeid, $name);
582 return $size;
583 }
584
585 sub volume_resize {
586 my ($class, $scfg, $storeid, $volname, $size, $running) = @_;
587
588 return 1 if $running && !$scfg->{krbd};
589
590 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
591
592 my $cmd = &$rbd_cmd($scfg, $storeid, 'resize', '--allow-shrink', '--size', ($size/1024/1024), $name);
593 run_rbd_command($cmd, errmsg => "rbd resize '$volname' error");
594 return undef;
595 }
596
597 sub volume_snapshot {
598 my ($class, $scfg, $storeid, $volname, $snap) = @_;
599
600 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
601
602 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'create', '--snap', $snap, $name);
603 run_rbd_command($cmd, errmsg => "rbd snapshot '$volname' error");
604 return undef;
605 }
606
607 sub volume_snapshot_rollback {
608 my ($class, $scfg, $storeid, $volname, $snap) = @_;
609
610 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
611
612 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'rollback', '--snap', $snap, $name);
613 run_rbd_command($cmd, errmsg => "rbd snapshot $volname to '$snap' error");
614 }
615
616 sub volume_snapshot_delete {
617 my ($class, $scfg, $storeid, $volname, $snap, $running) = @_;
618
619 return 1 if $running && !$scfg->{krbd};
620
621 $class->deactivate_volume($storeid, $scfg, $volname, $snap, {});
622
623 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
624
625 my (undef, undef, undef, $protected) = rbd_volume_info($scfg, $storeid, $name, $snap);
626 if ($protected){
627 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'unprotect', $name, '--snap', $snap);
628 run_rbd_command($cmd, errmsg => "rbd unprotect $name snap '$snap' error");
629 }
630
631 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'rm', '--snap', $snap, $name);
632
633 run_rbd_command($cmd, errmsg => "rbd snapshot '$volname' error");
634
635 return undef;
636 }
637
638 sub volume_has_feature {
639 my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_;
640
641 my $features = {
642 snapshot => { current => 1, snap => 1},
643 clone => { base => 1, snap => 1},
644 template => { current => 1},
645 copy => { base => 1, current => 1, snap => 1},
646 sparseinit => { base => 1, current => 1},
647 };
648
649 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
650 $class->parse_volname($volname);
651
652 my $key = undef;
653 if($snapname){
654 $key = 'snap';
655 }else{
656 $key = $isBase ? 'base' : 'current';
657 }
658 return 1 if $features->{$feature}->{$key};
659
660 return undef;
661 }
662
663 1;