]>
Commit | Line | Data |
---|---|---|
af50c2e6 WB |
1 | package PVE::Storage::BTRFSPlugin; |
2 | ||
3 | use strict; | |
4 | use warnings; | |
5 | ||
6 | use base qw(PVE::Storage::Plugin); | |
7 | ||
8 | use Fcntl qw(S_ISDIR O_WRONLY O_CREAT O_EXCL); | |
9524b31e | 9 | use File::Basename qw(basename dirname); |
af50c2e6 WB |
10 | use File::Path qw(mkpath); |
11 | use IO::Dir; | |
a0e3e224 | 12 | use POSIX qw(EEXIST); |
af50c2e6 | 13 | |
a0e3e224 | 14 | use PVE::Tools qw(run_command dir_glob_foreach); |
af50c2e6 WB |
15 | |
16 | use PVE::Storage::DirPlugin; | |
17 | ||
18 | use constant { | |
19 | BTRFS_FIRST_FREE_OBJECTID => 256, | |
20 | FS_NOCOW_FL => 0x00800000, | |
21 | FS_IOC_GETFLAGS => 0x40086602, | |
22 | FS_IOC_SETFLAGS => 0x80086601, | |
f6abd82a | 23 | BTRFS_MAGIC => 0x9123683e, |
af50c2e6 WB |
24 | }; |
25 | ||
26 | # Configuration (similar to DirPlugin) | |
27 | ||
28 | sub type { | |
29 | return 'btrfs'; | |
30 | } | |
31 | ||
32 | sub plugindata { | |
33 | return { | |
34 | content => [ | |
35 | { | |
36 | images => 1, | |
37 | rootdir => 1, | |
38 | vztmpl => 1, | |
39 | iso => 1, | |
40 | backup => 1, | |
41 | snippets => 1, | |
42 | none => 1, | |
43 | }, | |
44 | { images => 1, rootdir => 1 }, | |
45 | ], | |
347e677b | 46 | format => [ { raw => 1, subvol => 1 }, 'raw', ], |
af50c2e6 WB |
47 | }; |
48 | } | |
49 | ||
d3c5cf24 WB |
50 | sub properties { |
51 | return { | |
52 | nocow => { | |
53 | description => "Set the NOCOW flag on files." | |
54 | . " Disables data checksumming and causes data errors to be unrecoverable from" | |
55 | . " while allowing direct I/O. Only use this if data does not need to be any more" | |
56 | . " safe than on a single ext4 formatted disk with no underlying raid system.", | |
57 | type => 'boolean', | |
58 | default => 0, | |
59 | }, | |
60 | }; | |
61 | } | |
62 | ||
af50c2e6 WB |
63 | sub options { |
64 | return { | |
65 | path => { fixed => 1 }, | |
66 | nodes => { optional => 1 }, | |
67 | shared => { optional => 1 }, | |
68 | disable => { optional => 1 }, | |
69 | maxfiles => { optional => 1 }, | |
1c1589e6 | 70 | 'prune-backups'=> { optional => 1 }, |
af50c2e6 WB |
71 | content => { optional => 1 }, |
72 | format => { optional => 1 }, | |
73 | is_mountpoint => { optional => 1 }, | |
d3c5cf24 | 74 | nocow => { optional => 1 }, |
a1234a04 | 75 | mkdir => { optional => 1 }, |
af50c2e6 WB |
76 | # TODO: The new variant of mkdir with `populate` vs `create`... |
77 | }; | |
78 | } | |
79 | ||
80 | # Storage implementation | |
81 | # | |
82 | # We use the same volume names are directory plugins, but map *raw* disk image file names into a | |
83 | # subdirectory. | |
84 | # | |
85 | # `vm-VMID-disk-ID.raw` | |
86 | # -> `images/VMID/vm-VMID-disk-ID/disk.raw` | |
87 | # where the `vm-VMID-disk-ID/` subdirectory is a btrfs subvolume | |
88 | ||
89 | # Reuse `DirPlugin`'s `check_config`. This simply checks for invalid paths. | |
90 | sub check_config { | |
91 | my ($self, $sectionId, $config, $create, $skipSchemaCheck) = @_; | |
92 | return PVE::Storage::DirPlugin::check_config($self, $sectionId, $config, $create, $skipSchemaCheck); | |
93 | } | |
94 | ||
f6abd82a WB |
95 | my sub getfsmagic($) { |
96 | my ($path) = @_; | |
97 | # The field type sizes in `struct statfs` are defined in a rather annoying way, and we only | |
98 | # need the first field, which is a `long` for our supported platforms. | |
99 | # Should be moved to pve-rs, so this can be the problem of the `libc` crate ;-) | |
100 | # Just round up and extract what we need: | |
101 | my $buf = pack('x160'); | |
102 | if (0 != syscall(&PVE::Syscall::SYS_statfs, $path, $buf)) { | |
103 | die "statfs on '$path' failed - $!\n"; | |
104 | } | |
105 | ||
106 | return unpack('L!', $buf); | |
107 | } | |
108 | ||
109 | my sub assert_btrfs($) { | |
110 | my ($path) = @_; | |
111 | die "'$path' is not a btrfs file system\n" | |
112 | if getfsmagic($path) != BTRFS_MAGIC; | |
113 | } | |
114 | ||
af50c2e6 WB |
115 | sub activate_storage { |
116 | my ($class, $storeid, $scfg, $cache) = @_; | |
f449cddc TL |
117 | |
118 | my $path = $scfg->{path}; | |
119 | if (!defined($scfg->{mkdir}) || $scfg->{mkdir}) { | |
120 | mkpath $path; | |
121 | } | |
122 | ||
123 | my $mp = PVE::Storage::DirPlugin::parse_is_mountpoint($scfg); | |
38f0f469 | 124 | if (defined($mp) && !PVE::Storage::DirPlugin::path_is_mounted($mp, $cache->{mountdata})) { |
f449cddc TL |
125 | die "unable to activate storage '$storeid' - directory is expected to be a mount point but" |
126 | ." is not mounted: '$mp'\n"; | |
127 | } | |
128 | ||
129 | assert_btrfs($path); # only assert this stuff now, ensures $path is there and better UX | |
130 | ||
131 | $class->SUPER::activate_storage($storeid, $scfg, $cache); | |
af50c2e6 WB |
132 | } |
133 | ||
134 | sub status { | |
135 | my ($class, $storeid, $scfg, $cache) = @_; | |
136 | return PVE::Storage::DirPlugin::status($class, $storeid, $scfg, $cache); | |
137 | } | |
138 | ||
139 | # TODO: sub get_volume_notes {} | |
140 | ||
141 | # TODO: sub update_volume_notes {} | |
142 | ||
143 | # croak would not include the caller from within this module | |
144 | sub __error { | |
145 | my ($msg) = @_; | |
146 | my (undef, $f, $n) = caller(1); | |
147 | die "$msg at $f: $n\n"; | |
148 | } | |
149 | ||
150 | # Given a name (eg. `vm-VMID-disk-ID.raw`), take the part up to the format suffix as the name of | |
151 | # the subdirectory (subvolume). | |
152 | sub raw_name_to_dir($) { | |
153 | my ($raw) = @_; | |
154 | ||
155 | # For the subvolume directory Strip the `.<format>` suffix: | |
156 | if ($raw =~ /^(.*)\.raw$/) { | |
157 | return $1; | |
158 | } | |
159 | ||
160 | __error "internal error: bad disk name: $raw"; | |
161 | } | |
162 | ||
163 | sub raw_file_to_subvol($) { | |
164 | my ($file) = @_; | |
165 | ||
166 | if ($file =~ m|^(.*)/disk\.raw$|) { | |
167 | return "$1"; | |
168 | } | |
169 | ||
170 | __error "internal error: bad raw path: $file"; | |
171 | } | |
172 | ||
173 | sub filesystem_path { | |
174 | my ($class, $scfg, $volname, $snapname) = @_; | |
175 | ||
176 | my ($vtype, $name, $vmid, undef, undef, $isBase, $format) = | |
177 | $class->parse_volname($volname); | |
178 | ||
179 | my $path = $class->get_subdir($scfg, $vtype); | |
180 | ||
181 | $path .= "/$vmid" if $vtype eq 'images'; | |
182 | ||
6c315e45 | 183 | if (defined($format) && $format eq 'raw') { |
af50c2e6 WB |
184 | my $dir = raw_name_to_dir($name); |
185 | if ($snapname) { | |
186 | $dir .= "\@$snapname"; | |
187 | } | |
188 | $path .= "/$dir/disk.raw"; | |
6c315e45 | 189 | } elsif (defined($format) && $format eq 'subvol') { |
af50c2e6 WB |
190 | $path .= "/$name"; |
191 | if ($snapname) { | |
192 | $path .= "\@$snapname"; | |
193 | } | |
194 | } else { | |
195 | $path .= "/$name"; | |
196 | } | |
197 | ||
198 | return wantarray ? ($path, $vmid, $vtype) : $path; | |
199 | } | |
200 | ||
201 | sub btrfs_cmd { | |
202 | my ($class, $cmd, $outfunc) = @_; | |
203 | ||
204 | my $msg = ''; | |
205 | my $func; | |
206 | if (defined($outfunc)) { | |
207 | $func = sub { | |
208 | my $part = &$outfunc(@_); | |
209 | $msg .= $part if defined($part); | |
210 | }; | |
211 | } else { | |
212 | $func = sub { $msg .= "$_[0]\n" }; | |
213 | } | |
214 | run_command(['btrfs', '-q', @$cmd], errmsg => 'btrfs error', outfunc => $func); | |
215 | ||
216 | return $msg; | |
217 | } | |
218 | ||
219 | sub btrfs_get_subvol_id { | |
220 | my ($class, $path) = @_; | |
221 | my $info = $class->btrfs_cmd(['subvolume', 'show', '--', $path]); | |
222 | if ($info !~ /^\s*(?:Object|Subvolume) ID:\s*(\d+)$/m) { | |
223 | die "failed to get btrfs subvolume ID from: $info\n"; | |
224 | } | |
225 | return $1; | |
226 | } | |
227 | ||
228 | my sub chattr : prototype($$$) { | |
229 | my ($fh, $mask, $xor) = @_; | |
230 | ||
231 | my $flags = pack('L!', 0); | |
232 | ioctl($fh, FS_IOC_GETFLAGS, $flags) or die "FS_IOC_GETFLAGS failed - $!\n"; | |
233 | $flags = pack('L!', (unpack('L!', $flags) & $mask) ^ $xor); | |
234 | ioctl($fh, FS_IOC_SETFLAGS, $flags) or die "FS_IOC_SETFLAGS failed - $!\n"; | |
235 | return 1; | |
236 | } | |
237 | ||
238 | sub create_base { | |
239 | my ($class, $storeid, $scfg, $volname) = @_; | |
240 | ||
241 | my ($vtype, $name, $vmid, $basename, $basevmid, $isBase, $format) = | |
242 | $class->parse_volname($volname); | |
243 | ||
244 | my $newname = $name; | |
245 | $newname =~ s/^vm-/base-/; | |
246 | ||
247 | # If we're not working with a 'raw' file, which is the only thing that's "different" for btrfs, | |
248 | # or a subvolume, we forward to the DirPlugin | |
249 | if ($format ne 'raw' && $format ne 'subvol') { | |
250 | return PVE::Storage::DirPlugin::create_base(@_); | |
251 | } | |
252 | ||
253 | my $path = $class->filesystem_path($scfg, $volname); | |
254 | my $newvolname = $basename ? "$basevmid/$basename/$vmid/$newname" : "$vmid/$newname"; | |
255 | my $newpath = $class->filesystem_path($scfg, $newvolname); | |
256 | ||
257 | my $subvol = $path; | |
258 | my $newsubvol = $newpath; | |
259 | if ($format eq 'raw') { | |
260 | $subvol = raw_file_to_subvol($subvol); | |
261 | $newsubvol = raw_file_to_subvol($newsubvol); | |
262 | } | |
263 | ||
264 | rename($subvol, $newsubvol) | |
265 | || die "rename '$subvol' to '$newsubvol' failed - $!\n"; | |
266 | eval { $class->btrfs_cmd(['property', 'set', $newsubvol, 'ro', 'true']) }; | |
267 | warn $@ if $@; | |
268 | ||
269 | return $newvolname; | |
270 | } | |
271 | ||
272 | sub clone_image { | |
273 | my ($class, $scfg, $storeid, $volname, $vmid, $snap) = @_; | |
274 | ||
275 | my ($vtype, $basename, $basevmid, undef, undef, $isBase, $format) = | |
276 | $class->parse_volname($volname); | |
277 | ||
278 | # If we're not working with a 'raw' file, which is the only thing that's "different" for btrfs, | |
279 | # or a subvolume, we forward to the DirPlugin | |
280 | if ($format ne 'raw' && $format ne 'subvol') { | |
281 | return PVE::Storage::DirPlugin::clone_image(@_); | |
282 | } | |
283 | ||
284 | my $imagedir = $class->get_subdir($scfg, 'images'); | |
285 | $imagedir .= "/$vmid"; | |
286 | mkpath $imagedir; | |
287 | ||
288 | my $path = $class->filesystem_path($scfg, $volname); | |
289 | my $newname = $class->find_free_diskname($storeid, $scfg, $vmid, $format, 1); | |
290 | ||
291 | # For btrfs subvolumes we don't actually need the "link": | |
292 | #my $newvolname = "$basevmid/$basename/$vmid/$newname"; | |
293 | my $newvolname = "$vmid/$newname"; | |
294 | my $newpath = $class->filesystem_path($scfg, $newvolname); | |
295 | ||
296 | my $subvol = $path; | |
297 | my $newsubvol = $newpath; | |
298 | if ($format eq 'raw') { | |
299 | $subvol = raw_file_to_subvol($subvol); | |
300 | $newsubvol = raw_file_to_subvol($newsubvol); | |
301 | } | |
302 | ||
303 | $class->btrfs_cmd(['subvolume', 'snapshot', '--', $subvol, $newsubvol]); | |
304 | ||
305 | return $newvolname; | |
306 | } | |
307 | ||
308 | sub alloc_image { | |
309 | my ($class, $storeid, $scfg, $vmid, $fmt, $name, $size) = @_; | |
310 | ||
311 | if ($fmt ne 'raw' && $fmt ne 'subvol') { | |
08ca3955 | 312 | return $class->SUPER::alloc_image($storeid, $scfg, $vmid, $fmt, $name, $size); |
af50c2e6 WB |
313 | } |
314 | ||
315 | # From Plugin.pm: | |
316 | ||
317 | my $imagedir = $class->get_subdir($scfg, 'images') . "/$vmid"; | |
318 | ||
319 | mkpath $imagedir; | |
320 | ||
321 | $name = $class->find_free_diskname($storeid, $scfg, $vmid, $fmt, 1) if !$name; | |
322 | ||
323 | my (undef, $tmpfmt) = PVE::Storage::Plugin::parse_name_dir($name); | |
324 | ||
325 | die "illegal name '$name' - wrong extension for format ('$tmpfmt != '$fmt')\n" | |
326 | if $tmpfmt ne $fmt; | |
327 | ||
328 | # End copy from Plugin.pm | |
329 | ||
330 | my $subvol = "$imagedir/$name"; | |
331 | # .raw is not part of the directory name | |
332 | $subvol =~ s/\.raw$//; | |
333 | ||
334 | die "disk image '$subvol' already exists\n" if -e $subvol; | |
335 | ||
336 | my $path; | |
337 | if ($fmt eq 'raw') { | |
338 | $path = "$subvol/disk.raw"; | |
339 | } | |
340 | ||
341 | if ($fmt eq 'subvol' && !!$size) { | |
342 | # NOTE: `btrfs send/recv` actually drops quota information so supporting subvolumes with | |
343 | # quotas doesn't play nice with send/recv. | |
344 | die "btrfs quotas are currently not supported, use an unsized subvolume or a raw file\n"; | |
345 | } | |
346 | ||
347 | $class->btrfs_cmd(['subvolume', 'create', '--', $subvol]); | |
348 | ||
349 | eval { | |
350 | if ($fmt eq 'subvol') { | |
351 | # Nothing to do for now... | |
352 | ||
353 | # This is how we *would* do it: | |
354 | # # Use the subvol's default 0/$id qgroup | |
355 | # eval { | |
356 | # # This call should happen at storage creation instead and therefore governed by a | |
357 | # # configuration option! | |
358 | # # $class->btrfs_cmd(['quota', 'enable', $subvol]); | |
359 | # my $id = $class->btrfs_get_subvol_id($subvol); | |
360 | # $class->btrfs_cmd(['qgroup', 'limit', "${size}k", "0/$id", $subvol]); | |
361 | # }; | |
362 | } elsif ($fmt eq 'raw') { | |
363 | sysopen my $fh, $path, O_WRONLY | O_CREAT | O_EXCL | |
364 | or die "failed to create raw file '$path' - $!\n"; | |
d3c5cf24 | 365 | chattr($fh, ~FS_NOCOW_FL, FS_NOCOW_FL) if $scfg->{nocow}; |
af50c2e6 WB |
366 | truncate($fh, $size * 1024) |
367 | or die "failed to set file size for '$path' - $!\n"; | |
368 | close($fh); | |
369 | } else { | |
370 | die "internal format error (format = $fmt)\n"; | |
371 | } | |
372 | }; | |
373 | ||
374 | if (my $err = $@) { | |
375 | eval { $class->btrfs_cmd(['subvolume', 'delete', '--', $subvol]); }; | |
376 | warn $@ if $@; | |
377 | die $err; | |
378 | } | |
379 | ||
380 | return "$vmid/$name"; | |
381 | } | |
382 | ||
383 | # Same as btrfsprogs does: | |
384 | my sub path_is_subvolume : prototype($) { | |
385 | my ($path) = @_; | |
386 | my @stat = stat($path) | |
387 | or die "stat failed on '$path' - $!\n"; | |
388 | my ($ino, $mode) = @stat[1, 2]; | |
389 | return S_ISDIR($mode) && $ino == BTRFS_FIRST_FREE_OBJECTID; | |
390 | } | |
391 | ||
392 | my $BTRFS_VOL_REGEX = qr/((?:vm|base|subvol)-\d+-disk-\d+(?:\.subvol)?)(?:\@(\S+))$/; | |
393 | ||
394 | # Calls `$code->($volume, $name, $snapshot)` for each subvol in a directory matching our volume | |
395 | # regex. | |
396 | my sub foreach_subvol : prototype($$) { | |
397 | my ($dir, $code) = @_; | |
398 | ||
399 | dir_glob_foreach($dir, $BTRFS_VOL_REGEX, sub { | |
400 | my ($volume, $name, $snapshot) = ($1, $2, $3); | |
401 | return if !path_is_subvolume("$dir/$volume"); | |
402 | $code->($volume, $name, $snapshot); | |
403 | }) | |
404 | } | |
405 | ||
406 | sub free_image { | |
407 | my ($class, $storeid, $scfg, $volname, $isBase, $_format) = @_; | |
408 | ||
409 | my (undef, undef, $vmid, undef, undef, undef, $format) = | |
410 | $class->parse_volname($volname); | |
411 | ||
6c315e45 | 412 | if (!defined($format) || ($format ne 'subvol' && $format ne 'raw')) { |
dcd8f3a3 | 413 | return $class->SUPER::free_image($storeid, $scfg, $volname, $isBase, $_format); |
af50c2e6 WB |
414 | } |
415 | ||
416 | my $path = $class->filesystem_path($scfg, $volname); | |
417 | ||
418 | my $subvol = $path; | |
419 | if ($format eq 'raw') { | |
420 | $subvol = raw_file_to_subvol($path); | |
421 | } | |
422 | ||
423 | my $dir = dirname($subvol); | |
9524b31e | 424 | my $basename = basename($subvol); |
af50c2e6 WB |
425 | my @snapshot_vols; |
426 | foreach_subvol($dir, sub { | |
427 | my ($volume, $name, $snapshot) = @_; | |
9524b31e | 428 | return if $name ne $basename; |
af50c2e6 WB |
429 | return if !defined $snapshot; |
430 | push @snapshot_vols, "$dir/$volume"; | |
431 | }); | |
432 | ||
433 | $class->btrfs_cmd(['subvolume', 'delete', '--', @snapshot_vols, $subvol]); | |
434 | # try to cleanup directory to not clutter storage with empty $vmid dirs if | |
435 | # all images from a guest got deleted | |
436 | rmdir($dir); | |
437 | ||
438 | return undef; | |
439 | } | |
440 | ||
441 | # Currently not used because quotas clash with send/recv. | |
442 | # my sub btrfs_subvol_quota { | |
443 | # my ($class, $path) = @_; | |
444 | # my $id = '0/' . $class->btrfs_get_subvol_id($path); | |
445 | # my $search = qr/^\Q$id\E\s+(\d)+\s+\d+\s+(\d+)\s*$/; | |
446 | # my ($used, $size); | |
447 | # $class->btrfs_cmd(['qgroup', 'show', '--raw', '-rf', '--', $path], sub { | |
448 | # return if defined($size); | |
449 | # if ($_[0] =~ $search) { | |
450 | # ($used, $size) = ($1, $2); | |
451 | # } | |
452 | # }); | |
453 | # if (!defined($size)) { | |
454 | # # syslog should include more information: | |
455 | # syslog('err', "failed to get subvolume size for: $path (id $id)"); | |
456 | # # UI should only see the last path component: | |
457 | # $path =~ s|^.*/||; | |
458 | # die "failed to get subvolume size for $path\n"; | |
459 | # } | |
460 | # return wantarray ? ($used, $size) : $size; | |
461 | # } | |
462 | ||
463 | sub volume_size_info { | |
464 | my ($class, $scfg, $storeid, $volname, $timeout) = @_; | |
465 | ||
466 | my $path = $class->filesystem_path($scfg, $volname); | |
467 | ||
468 | my $format = ($class->parse_volname($volname))[6]; | |
469 | ||
6c315e45 | 470 | if (defined($format) && $format eq 'subvol') { |
af50c2e6 WB |
471 | my $ctime = (stat($path))[10]; |
472 | my ($used, $size) = (0, 0); | |
473 | #my ($used, $size) = btrfs_subvol_quota($class, $path); # uses wantarray | |
474 | return wantarray ? ($size, 'subvol', $used, undef, $ctime) : 1; | |
475 | } | |
476 | ||
477 | return PVE::Storage::Plugin::file_size_info($path, $timeout); | |
478 | } | |
479 | ||
480 | sub volume_resize { | |
481 | my ($class, $scfg, $storeid, $volname, $size, $running) = @_; | |
482 | ||
483 | my $format = ($class->parse_volname($volname))[6]; | |
484 | if ($format eq 'subvol') { | |
485 | my $path = $class->filesystem_path($scfg, $volname); | |
486 | my $id = '0/' . $class->btrfs_get_subvol_id($path); | |
487 | $class->btrfs_cmd(['qgroup', 'limit', '--', "${size}k", "0/$id", $path]); | |
488 | return undef; | |
489 | } | |
490 | ||
491 | return PVE::Storage::Plugin::volume_resize(@_); | |
492 | } | |
493 | ||
494 | sub volume_snapshot { | |
495 | my ($class, $scfg, $storeid, $volname, $snap) = @_; | |
496 | ||
497 | my ($name, $vmid, $format) = ($class->parse_volname($volname))[1,2,6]; | |
498 | if ($format ne 'subvol' && $format ne 'raw') { | |
499 | return PVE::Storage::Plugin::volume_snapshot(@_); | |
500 | } | |
501 | ||
502 | my $path = $class->filesystem_path($scfg, $volname); | |
503 | my $snap_path = $class->filesystem_path($scfg, $volname, $snap); | |
504 | ||
505 | if ($format eq 'raw') { | |
506 | $path = raw_file_to_subvol($path); | |
507 | $snap_path = raw_file_to_subvol($snap_path); | |
508 | } | |
509 | ||
510 | my $snapshot_dir = $class->get_subdir($scfg, 'images') . "/$vmid"; | |
511 | mkpath $snapshot_dir; | |
512 | ||
513 | $class->btrfs_cmd(['subvolume', 'snapshot', '-r', '--', $path, $snap_path]); | |
514 | return undef; | |
515 | } | |
516 | ||
517 | sub volume_rollback_is_possible { | |
518 | my ($class, $scfg, $storeid, $volname, $snap) = @_; | |
519 | ||
520 | return 1; | |
521 | } | |
522 | ||
523 | sub volume_snapshot_rollback { | |
524 | my ($class, $scfg, $storeid, $volname, $snap) = @_; | |
525 | ||
526 | my ($name, $format) = ($class->parse_volname($volname))[1,6]; | |
527 | ||
528 | if ($format ne 'subvol' && $format ne 'raw') { | |
529 | return PVE::Storage::Plugin::volume_snapshot_rollback(@_); | |
530 | } | |
531 | ||
532 | my $path = $class->filesystem_path($scfg, $volname); | |
533 | my $snap_path = $class->filesystem_path($scfg, $volname, $snap); | |
534 | ||
535 | if ($format eq 'raw') { | |
536 | $path = raw_file_to_subvol($path); | |
537 | $snap_path = raw_file_to_subvol($snap_path); | |
538 | } | |
539 | ||
540 | # Simple version would be: | |
541 | # rename old to temp | |
542 | # create new | |
543 | # on error rename temp back | |
544 | # But for atomicity in case the rename after create-failure *also* fails, we create the new | |
545 | # subvol first, then use RENAME_EXCHANGE, | |
546 | my $tmp_path = "$path.tmp.$$"; | |
547 | $class->btrfs_cmd(['subvolume', 'snapshot', '--', $snap_path, $tmp_path]); | |
548 | # The paths are absolute, so pass -1 as file descriptors. | |
549 | my $ok = PVE::Tools::renameat2(-1, $tmp_path, -1, $path, &PVE::Tools::RENAME_EXCHANGE); | |
550 | ||
551 | eval { $class->btrfs_cmd(['subvolume', 'delete', '--', $tmp_path]) }; | |
552 | warn "failed to remove '$tmp_path' subvolume: $@" if $@; | |
553 | ||
554 | if (!$ok) { | |
555 | die "failed to rotate '$tmp_path' into place at '$path' - $!\n"; | |
556 | } | |
557 | ||
558 | return undef; | |
559 | } | |
560 | ||
561 | sub volume_snapshot_delete { | |
562 | my ($class, $scfg, $storeid, $volname, $snap, $running) = @_; | |
563 | ||
564 | my ($name, $vmid, $format) = ($class->parse_volname($volname))[1,2,6]; | |
565 | ||
566 | if ($format ne 'subvol' && $format ne 'raw') { | |
567 | return PVE::Storage::Plugin::volume_snapshot_delete(@_); | |
568 | } | |
569 | ||
570 | my $path = $class->filesystem_path($scfg, $volname, $snap); | |
571 | ||
572 | if ($format eq 'raw') { | |
573 | $path = raw_file_to_subvol($path); | |
574 | } | |
575 | ||
576 | $class->btrfs_cmd(['subvolume', 'delete', '--', $path]); | |
577 | ||
578 | return undef; | |
579 | } | |
580 | ||
581 | sub volume_has_feature { | |
582 | my ($class, $scfg, $feature, $storeid, $volname, $snapname, $running) = @_; | |
583 | ||
584 | my $features = { | |
585 | snapshot => { | |
586 | current => { qcow2 => 1, raw => 1, subvol => 1 }, | |
587 | snap => { qcow2 => 1, raw => 1, subvol => 1 } | |
588 | }, | |
589 | clone => { | |
590 | base => { qcow2 => 1, raw => 1, subvol => 1, vmdk => 1 }, | |
591 | current => { raw => 1 }, | |
592 | snap => { raw => 1 }, | |
593 | }, | |
594 | template => { current => { qcow2 => 1, raw => 1, vmdk => 1, subvol => 1 } }, | |
595 | copy => { | |
596 | base => { qcow2 => 1, raw => 1, subvol => 1, vmdk => 1 }, | |
597 | current => { qcow2 => 1, raw => 1, subvol => 1, vmdk => 1 }, | |
598 | snap => { qcow2 => 1, raw => 1, subvol => 1 }, | |
599 | }, | |
600 | sparseinit => { base => {qcow2 => 1, raw => 1, vmdk => 1 }, | |
601 | current => {qcow2 => 1, raw => 1, vmdk => 1 } }, | |
602 | }; | |
603 | ||
604 | my ($vtype, $name, $vmid, $basename, $basevmid, $isBase, $format) = | |
605 | $class->parse_volname($volname); | |
606 | ||
607 | my $key = undef; | |
608 | if ($snapname) { | |
609 | $key = 'snap'; | |
610 | } else { | |
611 | $key = $isBase ? 'base' : 'current'; | |
612 | } | |
613 | ||
614 | return 1 if defined($features->{$feature}->{$key}->{$format}); | |
615 | ||
616 | return undef; | |
617 | } | |
618 | ||
619 | sub list_images { | |
620 | my ($class, $storeid, $scfg, $vmid, $vollist, $cache) = @_; | |
621 | my $imagedir = $class->get_subdir($scfg, 'images'); | |
622 | ||
623 | my $res = []; | |
624 | ||
625 | # Copied from Plugin.pm, with file_size_info calls adapted: | |
626 | foreach my $fn (<$imagedir/[0-9][0-9]*/*>) { | |
627 | # different to in Plugin.pm the regex below also excludes '@' as valid file name | |
628 | next if $fn !~ m@^(/.+/(\d+)/([^/\@.]+(?:\.(qcow2|vmdk|subvol))?))$@; | |
629 | $fn = $1; # untaint | |
630 | ||
631 | my $owner = $2; | |
632 | my $name = $3; | |
633 | my $ext = $4; | |
634 | ||
635 | next if !$vollist && defined($vmid) && ($owner ne $vmid); | |
636 | ||
637 | my $volid = "$storeid:$owner/$name"; | |
638 | my ($size, $format, $used, $parent, $ctime); | |
639 | ||
640 | if (!$ext) { # raw | |
641 | $volid .= '.raw'; | |
642 | ($size, $format, $used, $parent, $ctime) = PVE::Storage::Plugin::file_size_info("$fn/disk.raw"); | |
643 | } elsif ($ext eq 'subvol') { | |
644 | ($used, $size) = (0, 0); | |
645 | #($used, $size) = btrfs_subvol_quota($class, $fn); | |
646 | $format = 'subvol'; | |
647 | } else { | |
648 | ($size, $format, $used, $parent, $ctime) = PVE::Storage::Plugin::file_size_info($fn); | |
649 | } | |
650 | next if !($format && defined($size)); | |
651 | ||
652 | if ($vollist) { | |
653 | next if ! grep { $_ eq $volid } @$vollist; | |
654 | } | |
655 | ||
656 | my $info = { | |
657 | volid => $volid, format => $format, | |
658 | size => $size, vmid => $owner, used => $used, parent => $parent, | |
659 | }; | |
660 | ||
661 | $info->{ctime} = $ctime if $ctime; | |
662 | ||
663 | push @$res, $info; | |
664 | } | |
665 | ||
666 | return $res; | |
667 | } | |
668 | ||
af50c2e6 | 669 | sub volume_export_formats { |
a0e3e224 | 670 | my ($class, $scfg, $storeid, $volname, $snapshot, $base_snapshot, $with_snapshots) = @_; |
af50c2e6 | 671 | |
a0e3e224 WB |
672 | # We can do whatever `DirPlugin` can do. |
673 | my @result = PVE::Storage::Plugin::volume_export_formats(@_); | |
674 | ||
675 | # `btrfs send` only works on snapshots: | |
676 | return @result if !defined $snapshot; | |
677 | ||
678 | # Incremental stream with snapshots is only supported if the snapshots are listed (new api): | |
679 | return @result if defined($base_snapshot) && $with_snapshots && ref($with_snapshots) ne 'ARRAY'; | |
680 | ||
681 | # Otherwise we do also support `with_snapshots`. | |
682 | ||
683 | # Finally, `btrfs send` only works on formats where we actually use btrfs subvolumes: | |
684 | my $format = ($class->parse_volname($volname))[6]; | |
685 | return @result if $format ne 'raw' && $format ne 'subvol'; | |
686 | ||
687 | return ('btrfs', @result); | |
af50c2e6 WB |
688 | } |
689 | ||
690 | sub volume_import_formats { | |
a0e3e224 WB |
691 | my ($class, $scfg, $storeid, $volname, $snapshot, $base_snapshot, $with_snapshots) = @_; |
692 | ||
693 | # Same as export-formats, beware the parameter order: | |
694 | return volume_export_formats( | |
695 | $class, | |
696 | $scfg, | |
697 | $storeid, | |
698 | $volname, | |
699 | $snapshot, | |
700 | $base_snapshot, | |
701 | $with_snapshots, | |
702 | ); | |
703 | } | |
704 | ||
705 | sub volume_export { | |
706 | my ( | |
707 | $class, | |
708 | $scfg, | |
709 | $storeid, | |
710 | $fh, | |
711 | $volname, | |
712 | $format, | |
713 | $snapshot, | |
714 | $base_snapshot, | |
715 | $with_snapshots, | |
716 | ) = @_; | |
717 | ||
718 | if ($format ne 'btrfs') { | |
719 | return PVE::Storage::Plugin::volume_export(@_); | |
720 | } | |
721 | ||
722 | die "format 'btrfs' only works on snapshots\n" | |
723 | if !defined $snapshot; | |
724 | ||
725 | die "'btrfs' format in incremental mode requires snapshots to be listed explicitly\n" | |
726 | if defined($base_snapshot) && $with_snapshots && ref($with_snapshots) ne 'ARRAY'; | |
727 | ||
728 | my $volume_format = ($class->parse_volname($volname))[6]; | |
729 | ||
730 | die "btrfs-sending volumes of type $volume_format ('$volname') is not supported\n" | |
731 | if $volume_format ne 'raw' && $volume_format ne 'subvol'; | |
732 | ||
733 | my $path = $class->path($scfg, $volname, $storeid); | |
734 | ||
735 | if ($volume_format eq 'raw') { | |
736 | $path = raw_file_to_subvol($path); | |
737 | } | |
738 | ||
739 | my $cmd = ['btrfs', '-q', 'send', '-e']; | |
740 | if ($base_snapshot) { | |
741 | my $base = $class->path($scfg, $volname, $storeid, $base_snapshot); | |
742 | if ($volume_format eq 'raw') { | |
743 | $base = raw_file_to_subvol($base); | |
744 | } | |
745 | push @$cmd, '-p', $base; | |
746 | } | |
747 | push @$cmd, '--'; | |
748 | if (ref($with_snapshots) eq 'ARRAY') { | |
749 | push @$cmd, (map { "$path\@$_" } ($with_snapshots // [])->@*), $path; | |
750 | } else { | |
751 | dir_glob_foreach(dirname($path), $BTRFS_VOL_REGEX, sub { | |
752 | push @$cmd, "$path\@$_[2]" if !(defined($snapshot) && $_[2] eq $snapshot); | |
753 | }); | |
754 | } | |
755 | $path .= "\@$snapshot" if defined($snapshot); | |
756 | push @$cmd, $path; | |
757 | ||
758 | run_command($cmd, output => '>&'.fileno($fh)); | |
759 | return; | |
af50c2e6 WB |
760 | } |
761 | ||
762 | sub volume_import { | |
a0e3e224 WB |
763 | my ( |
764 | $class, | |
765 | $scfg, | |
766 | $storeid, | |
767 | $fh, | |
768 | $volname, | |
769 | $format, | |
770 | $snapshot, | |
771 | $base_snapshot, | |
772 | $with_snapshots, | |
773 | $allow_rename, | |
774 | ) = @_; | |
775 | ||
776 | if ($format ne 'btrfs') { | |
777 | return PVE::Storage::Plugin::volume_import(@_); | |
778 | } | |
779 | ||
780 | die "format 'btrfs' only works on snapshots\n" | |
781 | if !defined $snapshot; | |
782 | ||
783 | my ($vtype, $name, $vmid, $basename, $basevmid, $isBase, $volume_format) = | |
784 | $class->parse_volname($volname); | |
785 | ||
786 | die "btrfs-receiving volumes of type $volume_format ('$volname') is not supported\n" | |
787 | if $volume_format ne 'raw' && $volume_format ne 'subvol'; | |
788 | ||
789 | if (defined($base_snapshot)) { | |
790 | my $path = $class->path($scfg, $volname, $storeid, $base_snapshot); | |
791 | die "base snapshot '$base_snapshot' not found - no such directory '$path'\n" | |
792 | if !path_is_subvolume($path); | |
793 | } | |
794 | ||
795 | my $destination = $class->filesystem_path($scfg, $volname); | |
796 | if ($volume_format eq 'raw') { | |
797 | $destination = raw_file_to_subvol($destination); | |
798 | } | |
799 | ||
800 | if (!defined($base_snapshot) && -e $destination) { | |
801 | die "volume $volname already exists\n" if !$allow_rename; | |
802 | $volname = $class->find_free_diskname($storeid, $scfg, $vmid, $volume_format, 1); | |
803 | } | |
804 | ||
805 | my $imagedir = $class->get_subdir($scfg, $vtype); | |
806 | $imagedir .= "/$vmid" if $vtype eq 'images'; | |
807 | ||
808 | my $tmppath = "$imagedir/recv.$vmid.tmp"; | |
809 | mkdir($imagedir); # FIXME: if $scfg->{mkdir}; | |
810 | if (!mkdir($tmppath)) { | |
811 | die "temp receive directory already exists at '$tmppath', incomplete concurrent import?\n" | |
812 | if $! == EEXIST; | |
813 | die "failed to create temporary receive directory at '$tmppath' - $!\n"; | |
814 | } | |
815 | ||
816 | my $dh = IO::Dir->new($tmppath) | |
817 | or die "failed to open temporary receive directory '$tmppath' - $!\n"; | |
818 | eval { | |
819 | run_command(['btrfs', '-q', 'receive', '-e', '--', $tmppath], input => '<&'.fileno($fh)); | |
820 | ||
821 | # Analyze the received subvolumes; | |
822 | my ($diskname, $found_snapshot, @snapshots); | |
823 | $dh->rewind; | |
824 | while (defined(my $entry = $dh->read)) { | |
825 | next if $entry eq '.' || $entry eq '..'; | |
826 | next if $entry !~ /^$BTRFS_VOL_REGEX$/; | |
827 | my ($cur_diskname, $cur_snapshot) = ($1, $2); | |
828 | ||
829 | die "send stream included a non-snapshot subvolume\n" | |
830 | if !defined($cur_snapshot); | |
831 | ||
832 | if (!defined($diskname)) { | |
833 | $diskname = $cur_diskname; | |
834 | } else { | |
835 | die "multiple disks contained in stream ('$diskname' vs '$cur_diskname')\n" | |
836 | if $diskname ne $cur_diskname; | |
837 | } | |
838 | ||
839 | if ($cur_snapshot eq $snapshot) { | |
840 | $found_snapshot = 1; | |
841 | } else { | |
842 | push @snapshots, $cur_snapshot; | |
843 | } | |
844 | } | |
845 | ||
846 | die "send stream did not contain the expected current snapshot '$snapshot'\n" | |
847 | if !$found_snapshot; | |
848 | ||
849 | # Rotate the disk into place, first the current state: | |
850 | # Note that read-only subvolumes cannot be moved into different directories, but for the | |
851 | # "current" state we also want a writable copy, so start with that: | |
852 | $class->btrfs_cmd(['property', 'set', "$tmppath/$diskname\@$snapshot", 'ro', 'false']); | |
853 | PVE::Tools::renameat2( | |
854 | -1, | |
855 | "$tmppath/$diskname\@$snapshot", | |
856 | -1, | |
857 | $destination, | |
858 | &PVE::Tools::RENAME_NOREPLACE, | |
859 | ) or die "failed to move received snapshot '$tmppath/$diskname\@$snapshot'" | |
860 | . " into place at '$destination' - $!\n"; | |
861 | ||
862 | # Now recreate the actual snapshot: | |
863 | $class->btrfs_cmd([ | |
864 | 'subvolume', | |
865 | 'snapshot', | |
866 | '-r', | |
867 | '--', | |
868 | $destination, | |
869 | "$destination\@$snapshot", | |
870 | ]); | |
871 | ||
872 | # Now go through the remaining snapshots (if any) | |
873 | foreach my $snap (@snapshots) { | |
874 | $class->btrfs_cmd(['property', 'set', "$tmppath/$diskname\@$snap", 'ro', 'false']); | |
875 | PVE::Tools::renameat2( | |
876 | -1, | |
877 | "$tmppath/$diskname\@$snap", | |
878 | -1, | |
879 | "$destination\@$snap", | |
880 | &PVE::Tools::RENAME_NOREPLACE, | |
881 | ) or die "failed to move received snapshot '$tmppath/$diskname\@$snap'" | |
882 | . " into place at '$destination\@$snap' - $!\n"; | |
883 | eval { $class->btrfs_cmd(['property', 'set', "$destination\@$snap", 'ro', 'true']) }; | |
884 | warn "failed to make $destination\@$snap read-only - $!\n" if $@; | |
885 | } | |
886 | }; | |
887 | my $err = $@; | |
888 | ||
889 | eval { | |
890 | # Cleanup all the received snapshots we did not move into place, so we can remove the temp | |
891 | # directory. | |
892 | if ($dh) { | |
893 | $dh->rewind; | |
894 | while (defined(my $entry = $dh->read)) { | |
895 | next if $entry eq '.' || $entry eq '..'; | |
896 | eval { $class->btrfs_cmd(['subvolume', 'delete', '--', "$tmppath/$entry"]) }; | |
897 | warn $@ if $@; | |
898 | } | |
899 | $dh->close; undef $dh; | |
900 | } | |
901 | if (!rmdir($tmppath)) { | |
902 | warn "failed to remove temporary directory '$tmppath' - $!\n" | |
903 | } | |
904 | }; | |
905 | warn $@ if $@; | |
906 | if ($err) { | |
907 | # clean up if the directory ended up being empty after an error | |
908 | rmdir($tmppath); | |
909 | die $err; | |
910 | } | |
911 | ||
912 | return "$storeid:$volname"; | |
af50c2e6 WB |
913 | } |
914 | ||
915 | 1 |