]>
git.proxmox.com Git - pve-storage.git/blob - PVE/Storage/RBDPlugin.pm
1 package PVE
::Storage
::RBDPlugin
;
7 use PVE
::Tools
qw(run_command trim);
8 use PVE
::Storage
::Plugin
;
9 use PVE
::JSONSchema
qw(get_standard_option);
11 use PVE
::Storage
::CephTools
;
13 use base
qw(PVE::Storage::Plugin);
15 my $rbd_unittobytes = {
18 "G" => 1024*1024*1024,
19 "T" => 1024*1024*1024*1024,
22 my $add_pool_to_disk = sub {
23 my ($scfg, $disk) = @_;
25 my $pool = $scfg->{pool
} ?
$scfg->{pool
} : 'rbd';
31 my ($binary, $scfg, $storeid, $op, @options) = @_;
33 my $cmd_option = PVE
::Storage
::CephTools
::ceph_connect_option
($scfg, $storeid);
34 my $pool = $scfg->{pool
} ?
$scfg->{pool
} : 'rbd';
36 my $cmd = [$binary, '-p', $pool];
38 push @$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
});
46 push @$cmd, @options if scalar(@options);
52 my ($scfg, $storeid, $op, @options) = @_;
54 return $build_cmd->('/usr/bin/rbd', $scfg, $storeid, $op, @options);
58 my ($scfg, $storeid, $op, @options) = @_;
60 return $build_cmd->('/usr/bin/rados', $scfg, $storeid, $op, @options);
63 my $librados_connect = sub {
64 my ($scfg, $storeid, $options) = @_;
66 my $librados_config = PVE
::Storage
::CephTools
::ceph_connect_option
($scfg, $storeid);
68 my $rados = PVE
::RADOS-
>new(%$librados_config);
73 # needed for volumes created using ceph jewel (or higher)
74 my $krbd_feature_disable = sub {
75 my ($scfg, $storeid, $name) = @_;
77 return 1 if !$scfg->{krbd
};
79 my ($major, undef, undef, undef) = ceph_version
();
80 return 1 if $major < 10;
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);
85 my $active_features = { map { $_ => 1 } PVE
::Tools
::split_list
($features)};
86 my $incompatible_features = join(',', grep { %$active_features{$_} } @$krbd_feature_blacklist);
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");
94 my $ceph_version_parser = sub {
96 if ($line =~ m/^ceph version ((\d+)\.(\d+)\.(\d+))(?: \([a-fA-F0-9]+\))/) {
97 return ($2, $3, $4, $1);
99 warn "Could not parse Ceph version: '$line'\n";
106 my $version_string = $cache;
112 if (defined($version_string)) {
113 ($major, $minor, $bugfix, $version_string) = &$ceph_version_parser($version_string);
115 run_command
('ceph --version', outfunc
=> sub {
117 ($major, $minor, $bugfix, $version_string) = &$ceph_version_parser($line);
120 return undef if !defined($version_string);
121 return wantarray ?
($major, $minor, $bugfix, $version_string) : $version_string;
124 sub run_rbd_command
{
125 my ($cmd, %args) = @_;
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 $args{errfunc
} = sub {
134 if ($line =~ m/^\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2}\.\d+ [0-9a-f]+ [\-\d]+ librbd: (.*)$/) {
139 print STDERR
$lasterr;
144 eval { run_command
($cmd, %args); };
146 die $errmsg . $lasterr if length($lasterr);
154 my ($scfg, $storeid) = @_;
156 my $cmd = &$rbd_cmd($scfg, $storeid, 'ls', '-l');
157 my $pool = $scfg->{pool
} ?
$scfg->{pool
} : 'rbd';
164 if ($line =~ m/^((vm|base)-(\d+)-\S+)\s+(\d+)(k|M|G|T)\s((\S+)\/((vm
|base
)-\d
+-\S
+@\S+))?
/) {
165 my ($image, $owner, $size, $unit, $parent) = ($1, $3, $4, $5, $8);
166 return if $image =~ /@/; #skip snapshots
168 $list->{$pool}->{$image} = {
170 size
=> $size*$rbd_unittobytes->{$unit},
178 run_rbd_command
($cmd, errmsg
=> "rbd error", errfunc
=> sub {}, outfunc
=> $parser);
182 die $err if $err && $err !~ m/doesn't contain rbd images/ ;
187 sub rbd_volume_info
{
188 my ($scfg, $storeid, $volname, $snap) = @_;
193 $cmd = &$rbd_cmd($scfg, $storeid, 'info', $volname, '--snap', $snap);
195 $cmd = &$rbd_cmd($scfg, $storeid, 'info', $volname);
201 my $protected = undef;
202 my $features = undef;
207 if ($line =~ m/size (\d+) (k|M|G|T)B in (\d+) objects/) {
208 $size = $1 * $rbd_unittobytes->{$2} if ($1);
209 } elsif ($line =~ m/parent:\s(\S+)\/(\S
+)/) {
211 } elsif ($line =~ m/format:\s(\d+)/) {
213 } elsif ($line =~ m/protected:\s(\S+)/) {
214 $protected = 1 if $1 eq "True";
215 } elsif ($line =~ m/features:\s(.+)/) {
221 run_rbd_command
($cmd, errmsg
=> "rbd error", errfunc
=> sub {}, outfunc
=> $parser);
223 return ($size, $parent, $format, $protected, $features);
234 content
=> [ {images
=> 1, rootdir
=> 1}, { images
=> 1 }],
241 description
=> "IP addresses of monitors (for external clusters).",
242 type
=> 'string', format
=> 'pve-storage-portal-dns-list',
245 description
=> "Pool.",
249 description
=> "RBD Id.",
253 description
=> "Authsupported.",
257 description
=> "Access rbd through krbd kernel module.",
265 nodes
=> { optional
=> 1 },
266 disable
=> { optional
=> 1 },
267 monhost
=> { optional
=> 1},
268 pool
=> { optional
=> 1 },
269 username
=> { optional
=> 1 },
270 content
=> { optional
=> 1 },
271 krbd
=> { optional
=> 1 },
272 bwlimit
=> { optional
=> 1 },
276 # Storage implementation
279 my ($class, $storeid, $scfg, %param) = @_;
281 return if defined($scfg->{monhost
}); # nothing to do if not pve managed ceph
283 my $ceph_admin_keyring = '/etc/pve/priv/ceph.client.admin.keyring';
284 my $ceph_storage_keyring = "/etc/pve/priv/ceph/${storeid}.keyring";
286 die "ceph authx keyring file for storage '$storeid' already exists!\n"
287 if -e
$ceph_storage_keyring;
290 mkdir '/etc/pve/priv/ceph';
291 PVE
::Tools
::file_copy
($ceph_admin_keyring, $ceph_storage_keyring);
294 unlink $ceph_storage_keyring;
295 die "failed to copy ceph authx keyring for storage '$storeid': $err\n";
301 my ($class, $storeid, $scfg) = @_;
303 return if defined($scfg->{monhost
}); # nothing to do if not pve managed ceph
305 my $ceph_storage_keyring = "/etc/pve/priv/ceph/${storeid}.keyring";
306 if (-f
$ceph_storage_keyring) {
307 unlink($ceph_storage_keyring) or warn "removing keyring of storage failed: $!\n";
313 my ($class, $volname) = @_;
315 if ($volname =~ m/^((base-(\d+)-\S+)\/)?
((base
)?
(vm
)?
-(\d
+)-\S
+)$/) {
316 return ('images', $4, $7, $2, $3, $5, 'raw');
319 die "unable to parse rbd volume name '$volname'\n";
323 my ($class, $scfg, $volname, $storeid, $snapname) = @_;
325 my $cmd_option = PVE
::Storage
::CephTools
::ceph_connect_option
($scfg, $storeid);
326 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
327 $name .= '@'.$snapname if $snapname;
329 my $pool = $scfg->{pool
} ?
$scfg->{pool
} : 'rbd';
330 return ("/dev/rbd/$pool/$name", $vmid, $vtype) if $scfg->{krbd
};
332 my $path = "rbd:$pool/$name";
334 $path .= ":conf=$cmd_option->{ceph_conf}" if $cmd_option->{ceph_conf
};
335 if (defined($scfg->{monhost
})) {
336 my $monhost = PVE
::Storage
::CephTools
::hostlist
($scfg->{monhost
}, ';');
337 $monhost =~ s/:/\\:/g;
338 $path .= ":mon_host=$monhost";
339 $path .= ":auth_supported=$cmd_option->{auth_supported}";
342 $path .= ":id=$cmd_option->{userid}:keyring=$cmd_option->{keyring}" if ($cmd_option->{keyring
});
344 return ($path, $vmid, $vtype);
347 my $find_free_diskname = sub {
348 my ($storeid, $scfg, $vmid) = @_;
350 my $cmd = &$rbd_cmd($scfg, $storeid, 'ls');
356 if ($line =~ m/^(vm|base)-\Q$vmid\E+-disk-(\d+)$/) {
362 run_rbd_command
($cmd, errmsg
=> "rbd error", errfunc
=> sub {}, outfunc
=> $parser);
366 die $err if $err && $err !~ m/doesn't contain rbd images/;
368 #fix: can we search in $rbd hash key with a regex to find (vm|base) ?
369 for (my $i = 1; $i < 100; $i++) {
370 if (!$disk_ids->{$i}) {
371 return "vm-$vmid-disk-$i";
375 die "unable to allocate an image name for VM $vmid in storage '$storeid'\n";
379 my ($class, $storeid, $scfg, $volname) = @_;
381 my $snap = '__base__';
383 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
384 $class->parse_volname($volname);
386 die "create_base not possible with base image\n" if $isBase;
388 my ($size, $parent, $format, undef) = rbd_volume_info
($scfg, $storeid, $name);
389 die "rbd volume info on '$name' failed\n" if !($size);
391 die "rbd image must be at format V2" if $format ne "2";
393 die "volname '$volname' contains wrong information about parent $parent $basename\n"
394 if $basename && (!$parent || $parent ne $basename."@".$snap);
397 $newname =~ s/^vm-/base-/;
399 my $newvolname = $basename ?
"$basename/$newname" : "$newname";
401 my $cmd = &$rbd_cmd($scfg, $storeid, 'rename', &$add_pool_to_disk($scfg, $name), &$add_pool_to_disk($scfg, $newname));
402 run_rbd_command
($cmd, errmsg
=> "rbd rename '$name' error");
404 my $running = undef; #fixme : is create_base always offline ?
406 $class->volume_snapshot($scfg, $storeid, $newname, $snap, $running);
408 my (undef, undef, undef, $protected) = rbd_volume_info
($scfg, $storeid, $newname, $snap);
411 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'protect', $newname, '--snap', $snap);
412 run_rbd_command
($cmd, errmsg
=> "rbd protect $newname snap '$snap' error");
420 my ($class, $scfg, $storeid, $volname, $vmid, $snapname) = @_;
422 my $snap = '__base__';
423 $snap = $snapname if length $snapname;
425 my ($vtype, $basename, $basevmid, undef, undef, $isBase) =
426 $class->parse_volname($volname);
428 die "$volname is not a base image and snapname is not provided\n"
429 if !$isBase && !length($snapname);
431 my $name = &$find_free_diskname($storeid, $scfg, $vmid);
433 warn "clone $volname: $basename snapname $snap to $name\n";
435 if (length($snapname)) {
436 my (undef, undef, undef, $protected) = rbd_volume_info
($scfg, $storeid, $volname, $snapname);
439 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'protect', $volname, '--snap', $snapname);
440 run_rbd_command
($cmd, errmsg
=> "rbd protect $volname snap $snapname error");
444 my $newvol = "$basename/$name";
445 $newvol = $name if length($snapname);
447 my $cmd = &$rbd_cmd($scfg, $storeid, 'clone', &$add_pool_to_disk($scfg, $basename),
448 '--snap', $snap, &$add_pool_to_disk($scfg, $name));
450 run_rbd_command
($cmd, errmsg
=> "rbd clone '$basename' error");
452 &$krbd_feature_disable($scfg, $storeid, $name);
458 my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_;
461 die "illegal name '$name' - should be 'vm-$vmid-*'\n"
462 if $name && $name !~ m/^vm-$vmid-/;
464 $name = &$find_free_diskname($storeid, $scfg, $vmid) if !$name;
466 my $cmd = &$rbd_cmd($scfg, $storeid, 'create', '--image-format' , 2, '--size', int(($size+1023)/1024), $name);
467 run_rbd_command
($cmd, errmsg
=> "rbd create $name' error");
469 &$krbd_feature_disable($scfg, $storeid, $name);
475 my ($class, $storeid, $scfg, $volname, $isBase) = @_;
477 my ($vtype, $name, $vmid, undef, undef, undef) =
478 $class->parse_volname($volname);
481 my $snap = '__base__';
482 my (undef, undef, undef, $protected) = rbd_volume_info
($scfg, $storeid, $name, $snap);
484 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'unprotect', $name, '--snap', $snap);
485 run_rbd_command
($cmd, errmsg
=> "rbd unprotect $name snap '$snap' error");
489 $class->deactivate_volume($storeid, $scfg, $volname);
491 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'purge', $name);
492 run_rbd_command
($cmd, errmsg
=> "rbd snap purge '$volname' error");
494 $cmd = &$rbd_cmd($scfg, $storeid, 'rm', $name);
495 run_rbd_command
($cmd, errmsg
=> "rbd rm '$volname' error");
501 my ($class, $storeid, $scfg, $vmid, $vollist, $cache) = @_;
503 $cache->{rbd
} = rbd_ls
($scfg, $storeid) if !$cache->{rbd
};
504 my $pool = $scfg->{pool
} ?
$scfg->{pool
} : 'rbd';
508 if (my $dat = $cache->{rbd
}->{$pool}) {
509 foreach my $image (keys %$dat) {
511 my $info = $dat->{$image};
513 my $volname = $info->{name
};
514 my $parent = $info->{parent
};
515 my $owner = $info->{vmid
};
517 if ($parent && $parent =~ m/^(base-\d+-\S+)\@__base__$/) {
518 $info->{volid
} = "$storeid:$1/$volname";
520 $info->{volid
} = "$storeid:$volname";
524 my $found = grep { $_ eq $info->{volid
} } @$vollist;
527 next if defined ($vmid) && ($owner ne $vmid);
530 $info->{format
} = 'raw';
540 my ($class, $storeid, $scfg, $cache) = @_;
543 my $rados = &$librados_connect($scfg, $storeid);
544 my $df = $rados->mon_command({ prefix
=> 'df', format
=> 'json' });
546 my ($d) = grep { $_->{name
} eq $scfg->{pool
} } @{$df->{pools
}};
548 # max_avail -> max available space for data w/o replication in the pool
549 # bytes_used -> data w/o replication in the pool
550 my $free = $d->{stats
}->{max_avail
};
551 my $used = $d->{stats
}->{bytes_used
};
552 my $total = $used + $free;
555 return ($total, $free, $used, $active);
558 sub activate_storage
{
559 my ($class, $storeid, $scfg, $cache) = @_;
563 sub deactivate_storage
{
564 my ($class, $storeid, $scfg, $cache) = @_;
568 sub activate_volume
{
569 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
571 return 1 if !$scfg->{krbd
};
573 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
574 my $pool = $scfg->{pool
} ?
$scfg->{pool
} : 'rbd';
576 my $path = "/dev/rbd/$pool/$name";
577 $path .= '@'.$snapname if $snapname;
580 $name .= '@'.$snapname if $snapname;
581 my $cmd = &$rbd_cmd($scfg, $storeid, 'map', $name);
582 run_rbd_command
($cmd, errmsg
=> "can't mount rbd volume $name");
587 sub deactivate_volume
{
588 my ($class, $storeid, $scfg, $volname, $snapname, $cache) = @_;
590 return 1 if !$scfg->{krbd
};
592 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
593 my $pool = $scfg->{pool
} ?
$scfg->{pool
} : 'rbd';
595 my $path = "/dev/rbd/$pool/$name";
596 $path .= '@'.$snapname if $snapname;
597 return if ! -b
$path;
599 my $cmd = &$rbd_cmd($scfg, $storeid, 'unmap', $path);
600 run_rbd_command
($cmd, errmsg
=> "can't unmap rbd volume $name");
605 sub volume_size_info
{
606 my ($class, $scfg, $storeid, $volname, $timeout) = @_;
608 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
609 my ($size, undef) = rbd_volume_info
($scfg, $storeid, $name);
614 my ($class, $scfg, $storeid, $volname, $size, $running) = @_;
616 return 1 if $running && !$scfg->{krbd
};
618 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
620 my $cmd = &$rbd_cmd($scfg, $storeid, 'resize', '--allow-shrink', '--size', ($size/1024/1024), $name);
621 run_rbd_command
($cmd, errmsg
=> "rbd resize '$volname' error");
625 sub volume_snapshot
{
626 my ($class, $scfg, $storeid, $volname, $snap) = @_;
628 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
630 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'create', '--snap', $snap, $name);
631 run_rbd_command
($cmd, errmsg
=> "rbd snapshot '$volname' error");
635 sub volume_snapshot_rollback
{
636 my ($class, $scfg, $storeid, $volname, $snap) = @_;
638 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
640 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'rollback', '--snap', $snap, $name);
641 run_rbd_command
($cmd, errmsg
=> "rbd snapshot $volname to '$snap' error");
644 sub volume_snapshot_delete
{
645 my ($class, $scfg, $storeid, $volname, $snap, $running) = @_;
647 return 1 if $running && !$scfg->{krbd
};
649 $class->deactivate_volume($storeid, $scfg, $volname, $snap, {});
651 my ($vtype, $name, $vmid) = $class->parse_volname($volname);
653 my (undef, undef, undef, $protected) = rbd_volume_info
($scfg, $storeid, $name, $snap);
655 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'unprotect', $name, '--snap', $snap);
656 run_rbd_command
($cmd, errmsg
=> "rbd unprotect $name snap '$snap' error");
659 my $cmd = &$rbd_cmd($scfg, $storeid, 'snap', 'rm', '--snap', $snap, $name);
661 run_rbd_command
($cmd, errmsg
=> "rbd snapshot '$volname' error");
666 sub volume_has_feature
{
667 my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_;
670 snapshot
=> { current
=> 1, snap
=> 1},
671 clone
=> { base
=> 1, snap
=> 1},
672 template
=> { current
=> 1},
673 copy
=> { base
=> 1, current
=> 1, snap
=> 1},
674 sparseinit
=> { base
=> 1, current
=> 1},
677 my ($vtype, $name, $vmid, $basename, $basevmid, $isBase) =
678 $class->parse_volname($volname);
684 $key = $isBase ?
'base' : 'current';
686 return 1 if $features->{$feature}->{$key};