]>
Commit | Line | Data |
---|---|---|
cbba9b5b DC |
1 | package PVE::Diskmanage; |
2 | ||
3 | use strict; | |
4 | use warnings; | |
d5c80a5b | 5 | |
cbba9b5b DC |
6 | use PVE::ProcFSTools; |
7 | use Data::Dumper; | |
8 | use Cwd qw(abs_path); | |
3196c387 | 9 | use Fcntl ':mode'; |
262ad7a9 | 10 | use File::Basename; |
92ae59df | 11 | use File::stat; |
8cd6d7e8 | 12 | use JSON; |
cbba9b5b DC |
13 | |
14 | use PVE::Tools qw(extract_param run_command file_get_contents file_read_firstline dir_glob_regex dir_glob_foreach trim); | |
15 | ||
16 | my $SMARTCTL = "/usr/sbin/smartctl"; | |
17 | my $ZPOOL = "/sbin/zpool"; | |
18 | my $SGDISK = "/sbin/sgdisk"; | |
19 | my $PVS = "/sbin/pvs"; | |
19dcd1ad | 20 | my $LVS = "/sbin/lvs"; |
8cd6d7e8 | 21 | my $LSBLK = "/bin/lsblk"; |
cbba9b5b | 22 | |
a64aedd3 FE |
23 | my sub strip_dev :prototype($) { |
24 | my ($devpath) = @_; | |
25 | $devpath =~ s|^/dev/||; | |
26 | return $devpath; | |
27 | } | |
28 | ||
525b4a6e FE |
29 | sub check_bin { |
30 | my ($path) = @_; | |
525b4a6e FE |
31 | return -x $path; |
32 | } | |
33 | ||
cbba9b5b DC |
34 | sub verify_blockdev_path { |
35 | my ($rel_path) = @_; | |
36 | ||
37 | die "missing path" if !$rel_path; | |
38 | my $path = abs_path($rel_path); | |
39 | die "failed to get absolute path to $rel_path\n" if !$path; | |
40 | ||
41 | die "got unusual device path '$path'\n" if $path !~ m|^/dev/(.*)$|; | |
42 | ||
43 | $path = "/dev/$1"; # untaint | |
44 | ||
45 | assert_blockdev($path); | |
46 | ||
47 | return $path; | |
48 | } | |
49 | ||
50 | sub assert_blockdev { | |
51 | my ($dev, $noerr) = @_; | |
52 | ||
53 | if ($dev !~ m|^/dev/| || !(-b $dev)) { | |
9aff3f3d | 54 | return if $noerr; |
cbba9b5b DC |
55 | die "not a valid block device\n"; |
56 | } | |
57 | ||
58 | return 1; | |
59 | } | |
60 | ||
61 | sub init_disk { | |
62 | my ($disk, $uuid) = @_; | |
63 | ||
64 | assert_blockdev($disk); | |
65 | ||
cc884f73 FE |
66 | # we should already have checked these in the api call, but we check again for safety |
67 | die "$disk is a partition\n" if is_partition($disk); | |
cbba9b5b DC |
68 | die "disk $disk is already in use\n" if disk_is_used($disk); |
69 | ||
70 | my $id = $uuid || 'R'; | |
71 | run_command([$SGDISK, $disk, '-U', $id]); | |
72 | return 1; | |
73 | } | |
74 | ||
75 | sub disk_is_used { | |
76 | my ($disk) = @_; | |
77 | ||
78 | my $dev = $disk; | |
79 | $dev =~ s|^/dev/||; | |
80 | ||
a2c34371 | 81 | my $disklist = get_disks($dev, 1, 1); |
cbba9b5b DC |
82 | |
83 | die "'$disk' is not a valid local disk\n" if !defined($disklist->{$dev}); | |
84 | return 1 if $disklist->{$dev}->{used}; | |
85 | ||
86 | return 0; | |
87 | } | |
88 | ||
89 | sub get_smart_data { | |
dd902da7 | 90 | my ($disk, $healthonly) = @_; |
cbba9b5b DC |
91 | |
92 | assert_blockdev($disk); | |
93 | my $smartdata = {}; | |
dc1311cb | 94 | my $type; |
cbba9b5b | 95 | |
c3442aa5 WB |
96 | if ($disk =~ m!^/dev/(nvme\d+n\d+)$!) { |
97 | my $info = get_sysdir_info("/sys/block/$1"); | |
98 | $disk = "/dev/".($info->{device} | |
99 | or die "failed to get nvme controller device for $disk\n"); | |
100 | } | |
c9bd3d22 | 101 | |
dd902da7 DC |
102 | my $cmd = [$SMARTCTL, '-H']; |
103 | push @$cmd, '-A', '-f', 'brief' if !$healthonly; | |
104 | push @$cmd, $disk; | |
105 | ||
9aff3f3d TL |
106 | my $returncode = eval { |
107 | run_command($cmd, noerr => 1, outfunc => sub { | |
cbba9b5b DC |
108 | my ($line) = @_; |
109 | ||
1c999553 FG |
110 | # ATA SMART attributes, e.g.: |
111 | # ID# ATTRIBUTE_NAME FLAGS VALUE WORST THRESH FAIL RAW_VALUE | |
112 | # 1 Raw_Read_Error_Rate POSR-K 100 100 000 - 0 | |
dc1311cb FG |
113 | # |
114 | # SAS and NVME disks, e.g.: | |
115 | # Data Units Written: 5,584,952 [2.85 TB] | |
116 | # Accumulated start-stop cycles: 34 | |
117 | ||
bd54091c | 118 | if (defined($type) && $type eq 'ata' && $line =~ m/^([ \d]{2}\d)\s+(\S+)\s+(\S{6})\s+(\d+)\s+(\d+)\s+(\S+)\s+(\S+)\s+(.*)$/) { |
cbba9b5b | 119 | my $entry = {}; |
bd54091c | 120 | |
1c999553 FG |
121 | $entry->{name} = $2 if defined $2; |
122 | $entry->{flags} = $3 if defined $3; | |
cbba9b5b | 123 | # the +0 makes a number out of the strings |
47d1125b | 124 | # FIXME: 'value' is depreacated by 'normalized'; remove with PVE 7.0 |
1c999553 | 125 | $entry->{value} = $4+0 if defined $4; |
4c86c711 | 126 | $entry->{normalized} = $4+0 if defined $4; |
1c999553 | 127 | $entry->{worst} = $5+0 if defined $5; |
bd54091c DC |
128 | # some disks report the default threshold as --- instead of 000 |
129 | if (defined($6) && $6 eq '---') { | |
130 | $entry->{threshold} = 0; | |
131 | } else { | |
132 | $entry->{threshold} = $6+0 if defined $6; | |
133 | } | |
1c999553 FG |
134 | $entry->{fail} = $7 if defined $7; |
135 | $entry->{raw} = $8 if defined $8; | |
136 | $entry->{id} = $1 if defined $1; | |
cbba9b5b | 137 | push @{$smartdata->{attributes}}, $entry; |
5db2d529 | 138 | } elsif ($line =~ m/(?:Health Status|self\-assessment test result): (.*)$/ ) { |
cbba9b5b DC |
139 | $smartdata->{health} = $1; |
140 | } elsif ($line =~ m/Vendor Specific SMART Attributes with Thresholds:/) { | |
dc1311cb FG |
141 | $type = 'ata'; |
142 | delete $smartdata->{text}; | |
143 | } elsif ($line =~ m/=== START OF (READ )?SMART DATA SECTION ===/) { | |
144 | $type = 'text'; | |
145 | } elsif (defined($type) && $type eq 'text') { | |
146 | $smartdata->{text} = '' if !defined $smartdata->{text}; | |
147 | $smartdata->{text} .= "$line\n"; | |
2c048efd DC |
148 | # extract wearout from nvme/sas text, allow for decimal values |
149 | if ($line =~ m/Percentage Used(?: endurance indicator)?:\s*(\d+(?:\.\d+)?)\%/i) { | |
ea928fd4 DC |
150 | $smartdata->{wearout} = 100 - $1; |
151 | } | |
dd902da7 DC |
152 | } elsif ($line =~ m/SMART Disabled/) { |
153 | $smartdata->{health} = "SMART Disabled"; | |
cbba9b5b | 154 | } |
9aff3f3d | 155 | }) |
cbba9b5b | 156 | }; |
9018a4e6 DC |
157 | my $err = $@; |
158 | ||
9aff3f3d | 159 | # bit 0 and 1 mark a fatal error, other bits are for disk status -> ignore (see man 8 smartctl) |
9018a4e6 DC |
160 | if ((defined($returncode) && ($returncode & 0b00000011)) || $err) { |
161 | die "Error getting S.M.A.R.T. data: Exit code: $returncode\n"; | |
162 | } | |
dc1311cb FG |
163 | |
164 | $smartdata->{type} = $type; | |
165 | ||
cbba9b5b DC |
166 | return $smartdata; |
167 | } | |
168 | ||
9aff3f3d | 169 | sub get_lsblk_info { |
59c03cd9 | 170 | my $cmd = [$LSBLK, '--json', '-o', 'path,parttype,fstype']; |
8cd6d7e8 | 171 | my $output = ""; |
9aff3f3d | 172 | eval { run_command($cmd, outfunc => sub { $output .= "$_[0]\n"; }) }; |
8cd6d7e8 | 173 | warn "$@\n" if $@; |
9aff3f3d | 174 | return {} if $output eq ''; |
8cd6d7e8 | 175 | |
9aff3f3d | 176 | my $parsed = eval { decode_json($output) } // {}; |
8cd6d7e8 DC |
177 | warn "$@\n" if $@; |
178 | my $list = $parsed->{blockdevices} // []; | |
179 | ||
9aff3f3d TL |
180 | return { |
181 | map { | |
182 | $_->{path} => { | |
183 | parttype => $_->{parttype}, | |
184 | fstype => $_->{fstype} | |
185 | } | |
186 | } @{$list} | |
187 | }; | |
8cd6d7e8 DC |
188 | } |
189 | ||
9aff3f3d | 190 | my sub get_devices_by_partuuid { |
b6bbc2ab | 191 | my ($lsblk_info, $uuids, $res) = @_; |
8cd6d7e8 DC |
192 | |
193 | $res = {} if !defined($res); | |
194 | ||
b6bbc2ab FE |
195 | foreach my $dev (sort keys %{$lsblk_info}) { |
196 | my $uuid = $lsblk_info->{$dev}->{parttype}; | |
197 | next if !defined($uuid) || !defined($uuids->{$uuid}); | |
198 | $res->{$dev} = $uuids->{$uuid}; | |
8cd6d7e8 DC |
199 | } |
200 | ||
201 | return $res; | |
9aff3f3d | 202 | } |
8cd6d7e8 | 203 | |
cbba9b5b | 204 | sub get_zfs_devices { |
b6bbc2ab | 205 | my ($lsblk_info) = @_; |
8cd6d7e8 | 206 | my $res = {}; |
cbba9b5b | 207 | |
525b4a6e | 208 | return {} if !check_bin($ZPOOL); |
4526dffa | 209 | |
9aff3f3d | 210 | # use zpool and parttype uuid, because log and cache do not have zfs type uuid |
cbba9b5b DC |
211 | eval { |
212 | run_command([$ZPOOL, 'list', '-HPLv'], outfunc => sub { | |
213 | my ($line) = @_; | |
cbba9b5b | 214 | if ($line =~ m|^\t([^\t]+)\t|) { |
8cd6d7e8 | 215 | $res->{$1} = 1; |
cbba9b5b DC |
216 | } |
217 | }); | |
218 | }; | |
219 | ||
9aff3f3d | 220 | # only warn here, because maybe zfs tools are not installed |
cbba9b5b DC |
221 | warn "$@\n" if $@; |
222 | ||
8cd6d7e8 DC |
223 | my $uuids = { |
224 | "6a898cc3-1dd2-11b2-99a6-080020736631" => 1, # apple | |
225 | "516e7cba-6ecf-11d6-8ff8-00022d09712b" => 1, # bsd | |
226 | }; | |
cbba9b5b | 227 | |
cbba9b5b | 228 | |
9aff3f3d | 229 | $res = get_devices_by_partuuid($lsblk_info, $uuids, $res); |
8cd6d7e8 DC |
230 | |
231 | return $res; | |
cbba9b5b DC |
232 | } |
233 | ||
234 | sub get_lvm_devices { | |
b6bbc2ab | 235 | my ($lsblk_info) = @_; |
8cd6d7e8 | 236 | my $res = {}; |
cbba9b5b DC |
237 | eval { |
238 | run_command([$PVS, '--noheadings', '--readonly', '-o', 'pv_name'], outfunc => sub{ | |
239 | my ($line) = @_; | |
240 | $line = trim($line); | |
241 | if ($line =~ m|^/dev/|) { | |
8cd6d7e8 | 242 | $res->{$line} = 1; |
cbba9b5b DC |
243 | } |
244 | }); | |
245 | }; | |
246 | ||
9aff3f3d | 247 | # if something goes wrong, we do not want to give up, but indicate an error has occurred |
cbba9b5b DC |
248 | warn "$@\n" if $@; |
249 | ||
8cd6d7e8 DC |
250 | my $uuids = { |
251 | "e6d6d379-f507-44c2-a23c-238f2a3df928" => 1, | |
252 | }; | |
cbba9b5b | 253 | |
9aff3f3d | 254 | $res = get_devices_by_partuuid($lsblk_info, $uuids, $res); |
cbba9b5b | 255 | |
8cd6d7e8 | 256 | return $res; |
cbba9b5b DC |
257 | } |
258 | ||
259 | sub get_ceph_journals { | |
b6bbc2ab | 260 | my ($lsblk_info) = @_; |
8cd6d7e8 DC |
261 | my $res = {}; |
262 | ||
263 | my $uuids = { | |
264 | '45b0969e-9b03-4f30-b4c6-b4b80ceff106' => 1, # journal | |
265 | '30cd0809-c2b2-499c-8879-2d6b78529876' => 2, # db | |
266 | '5ce17fce-4087-4169-b7ff-056cc58473f9' => 3, # wal | |
267 | 'cafecafe-9b03-4f30-b4c6-b4b80ceff106' => 4, # block | |
268 | }; | |
269 | ||
9aff3f3d | 270 | $res = get_devices_by_partuuid($lsblk_info, $uuids, $res); |
cbba9b5b | 271 | |
8cd6d7e8 | 272 | return $res; |
cbba9b5b DC |
273 | } |
274 | ||
19dcd1ad DC |
275 | # reads the lv_tags and matches them with the devices |
276 | sub get_ceph_volume_infos { | |
277 | my $result = {}; | |
278 | ||
248f43f5 TL |
279 | my $cmd = [ $LVS, '-S', 'lv_name=~^osd-', '-o', 'devices,lv_name,lv_tags', |
280 | '--noheadings', '--readonly', '--separator', ';' ]; | |
19dcd1ad DC |
281 | |
282 | run_command($cmd, outfunc => sub { | |
283 | my $line = shift; | |
248f43f5 TL |
284 | $line =~ s/(?:^\s+)|(?:\s+$)//g; # trim whitespaces |
285 | ||
286 | my $fields = [ split(';', $line) ]; | |
19dcd1ad DC |
287 | |
288 | # lvs syntax is /dev/sdX(Y) where Y is the start (which we do not need) | |
41f93ece | 289 | my ($dev) = $fields->[0] =~ m|^(/dev/[a-z]+[^(]*)|; |
19dcd1ad DC |
290 | if ($fields->[1] =~ m|^osd-([^-]+)-|) { |
291 | my $type = $1; | |
248f43f5 | 292 | # $result autovivification is wanted, to not creating empty hashes |
79f4a7bf | 293 | if (($type eq 'block' || $type eq 'data') && $fields->[2] =~ m/ceph.osd_id=([^,]+)/) { |
19dcd1ad DC |
294 | $result->{$dev}->{osdid} = $1; |
295 | $result->{$dev}->{bluestore} = ($type eq 'block'); | |
bfb3d42d DC |
296 | if ($fields->[2] =~ m/ceph\.encrypted=1/) { |
297 | $result->{$dev}->{encrypted} = 1; | |
298 | } | |
19dcd1ad | 299 | } else { |
248f43f5 | 300 | # undef++ becomes '1' (see `perldoc perlop`: Auto-increment) |
19dcd1ad DC |
301 | $result->{$dev}->{$type}++; |
302 | } | |
303 | } | |
304 | }); | |
305 | ||
306 | return $result; | |
307 | } | |
308 | ||
cbba9b5b DC |
309 | sub get_udev_info { |
310 | my ($dev) = @_; | |
311 | ||
312 | my $info = ""; | |
313 | my $data = {}; | |
314 | eval { | |
d3a5e309 | 315 | run_command(['udevadm', 'info', '-p', $dev, '--query', 'all'], outfunc => sub { |
cbba9b5b DC |
316 | my ($line) = @_; |
317 | $info .= "$line\n"; | |
318 | }); | |
319 | }; | |
320 | warn $@ if $@; | |
9aff3f3d | 321 | return if !$info; |
cbba9b5b | 322 | |
9aff3f3d TL |
323 | return if $info !~ m/^E: DEVTYPE=(disk|partition)$/m; |
324 | return if $info =~ m/^E: ID_CDROM/m; | |
cbba9b5b | 325 | |
9aff3f3d | 326 | # we use this, because some disks are not simply in /dev e.g. /dev/cciss/c0d0 |
cbba9b5b DC |
327 | if ($info =~ m/^E: DEVNAME=(\S+)$/m) { |
328 | $data->{devpath} = $1; | |
329 | } | |
330 | return if !defined($data->{devpath}); | |
331 | ||
332 | $data->{serial} = 'unknown'; | |
9aff3f3d | 333 | $data->{serial} = $1 if $info =~ m/^E: ID_SERIAL_SHORT=(\S+)$/m; |
cbba9b5b | 334 | |
9aff3f3d | 335 | $data->{gpt} = $info =~ m/^E: ID_PART_TABLE_TYPE=gpt$/m ? 1 : 0; |
cbba9b5b | 336 | |
cbba9b5b | 337 | $data->{rpm} = -1; |
9aff3f3d | 338 | $data->{rpm} = $1 if $info =~ m/^E: ID_ATA_ROTATION_RATE_RPM=(\d+)$/m; # detects SSD implicit |
cbba9b5b | 339 | |
9aff3f3d | 340 | $data->{usb} = 1 if $info =~ m/^E: ID_BUS=usb$/m; |
cbba9b5b | 341 | |
9aff3f3d | 342 | $data->{model} = $1 if $info =~ m/^E: ID_MODEL=(.+)$/m; |
865bdbd9 | 343 | |
cbba9b5b | 344 | $data->{wwn} = 'unknown'; |
9aff3f3d | 345 | $data->{wwn} = $1 if $info =~ m/^E: ID_WWN=(.*)$/m; |
cbba9b5b | 346 | |
0f0d99a3 SI |
347 | if ($info =~ m/^E: DEVLINKS=(.+)$/m) { |
348 | my @devlinks = grep(m#^/dev/disk/by-id/(ata|scsi|nvme(?!-eui))#, split (/ /, $1)); | |
349 | $data->{by_id_link} = $devlinks[0] if defined($devlinks[0]); | |
350 | } | |
351 | ||
cbba9b5b DC |
352 | return $data; |
353 | } | |
354 | ||
40be5c5c FE |
355 | sub get_sysdir_size { |
356 | my ($sysdir) = @_; | |
357 | ||
358 | my $size = file_read_firstline("$sysdir/size"); | |
359 | return if !$size; | |
360 | ||
9aff3f3d | 361 | # linux always considers sectors to be 512 bytes, independently of real block size |
40be5c5c FE |
362 | return $size * 512; |
363 | } | |
364 | ||
cbba9b5b DC |
365 | sub get_sysdir_info { |
366 | my ($sysdir) = @_; | |
367 | ||
9aff3f3d | 368 | return if ! -d "$sysdir/device"; |
461a9fd8 | 369 | |
cbba9b5b DC |
370 | my $data = {}; |
371 | ||
40be5c5c | 372 | $data->{size} = get_sysdir_size($sysdir) or return; |
cbba9b5b DC |
373 | |
374 | # dir/queue/rotational should be 1 for hdd, 0 for ssd | |
571b6f26 | 375 | $data->{rotational} = file_read_firstline("$sysdir/queue/rotational") // -1; |
cbba9b5b DC |
376 | |
377 | $data->{vendor} = file_read_firstline("$sysdir/device/vendor") || 'unknown'; | |
378 | $data->{model} = file_read_firstline("$sysdir/device/model") || 'unknown'; | |
379 | ||
c3442aa5 | 380 | if (defined(my $device = readlink("$sysdir/device"))) { |
9aff3f3d | 381 | ($data->{device}) = $device =~ m!([^/]+)$!; # strip directory and untaint |
c3442aa5 WB |
382 | } |
383 | ||
cbba9b5b DC |
384 | return $data; |
385 | } | |
386 | ||
6965a670 | 387 | sub get_wear_leveling_info { |
dbad606d | 388 | my ($smartdata) = @_; |
ea928fd4 DC |
389 | my $attributes = $smartdata->{attributes}; |
390 | ||
391 | if (defined($smartdata->{wearout})) { | |
392 | return $smartdata->{wearout}; | |
393 | } | |
6965a670 DC |
394 | |
395 | my $wearout; | |
396 | ||
9aff3f3d TL |
397 | # Common register names that represent percentage values of potential failure indicators used |
398 | # in drivedb.h of smartmontool's. Order matters, as some drives may have multiple definitions | |
dbad606d JJS |
399 | my @wearoutregisters = ( |
400 | "Media_Wearout_Indicator", | |
401 | "SSD_Life_Left", | |
402 | "Wear_Leveling_Count", | |
403 | "Perc_Write\/Erase_Ct_BC", | |
404 | "Perc_Rated_Life_Remain", | |
405 | "Remaining_Lifetime_Perc", | |
406 | "Percent_Lifetime_Remain", | |
407 | "Lifetime_Left", | |
408 | "PCT_Life_Remaining", | |
409 | "Lifetime_Remaining", | |
410 | "Percent_Life_Remaining", | |
411 | "Percent_Lifetime_Used", | |
412 | "Perc_Rated_Life_Used" | |
413 | ); | |
414 | ||
415 | # Search for S.M.A.R.T. attributes for known register | |
416 | foreach my $register (@wearoutregisters) { | |
417 | last if defined $wearout; | |
418 | foreach my $attr (@$attributes) { | |
419 | next if $attr->{name} !~ m/$register/; | |
420 | $wearout = $attr->{value}; | |
421 | last; | |
6965a670 DC |
422 | } |
423 | } | |
424 | ||
6965a670 DC |
425 | return $wearout; |
426 | } | |
427 | ||
10a48db5 DC |
428 | sub dir_is_empty { |
429 | my ($dir) = @_; | |
430 | ||
431 | my $dh = IO::Dir->new ($dir); | |
432 | return 1 if !$dh; | |
433 | ||
434 | while (defined(my $tmp = $dh->read)) { | |
435 | next if $tmp eq '.' || $tmp eq '..'; | |
436 | $dh->close; | |
437 | return 0; | |
438 | } | |
439 | $dh->close; | |
440 | return 1; | |
441 | } | |
442 | ||
eebcdb11 DC |
443 | sub is_iscsi { |
444 | my ($sysdir) = @_; | |
445 | ||
446 | if (-l $sysdir && readlink($sysdir) =~ m|host[^/]*/session[^/]*|) { | |
447 | return 1; | |
448 | } | |
449 | ||
450 | return 0; | |
451 | } | |
452 | ||
4731eb11 TL |
453 | my sub is_ssdlike { |
454 | my ($type) = @_; | |
455 | return $type eq 'ssd' || $type eq 'nvme'; | |
456 | } | |
457 | ||
7e14102a | 458 | sub mounted_blockdevs { |
cbba9b5b DC |
459 | my $mounted = {}; |
460 | ||
461 | my $mounts = PVE::ProcFSTools::parse_proc_mounts(); | |
462 | ||
463 | foreach my $mount (@$mounts) { | |
464 | next if $mount->[0] !~ m|^/dev/|; | |
465 | $mounted->{abs_path($mount->[0])} = $mount->[1]; | |
466 | }; | |
467 | ||
7e14102a FE |
468 | return $mounted; |
469 | } | |
470 | ||
4de60025 AL |
471 | # returns hashmap of abs mount path -> first part of /proc/mounts (what) |
472 | sub mounted_paths { | |
473 | my $mounted = {}; | |
474 | ||
475 | my $mounts = PVE::ProcFSTools::parse_proc_mounts(); | |
476 | ||
477 | foreach my $mount (@$mounts) { | |
478 | $mounted->{abs_path($mount->[1])} = $mount->[0]; | |
479 | }; | |
480 | ||
481 | return $mounted; | |
482 | } | |
483 | ||
7e14102a FE |
484 | sub get_disks { |
485 | my ($disks, $nosmart, $include_partitions) = @_; | |
486 | my $disklist = {}; | |
487 | ||
488 | my $mounted = mounted_blockdevs(); | |
489 | ||
b6bbc2ab | 490 | my $lsblk_info = get_lsblk_info(); |
8cd6d7e8 | 491 | |
b6bbc2ab | 492 | my $journalhash = get_ceph_journals($lsblk_info); |
19dcd1ad | 493 | my $ceph_volume_infos = get_ceph_volume_infos(); |
cbba9b5b | 494 | |
b6bbc2ab | 495 | my $zfshash = get_zfs_devices($lsblk_info); |
cbba9b5b | 496 | |
b6bbc2ab | 497 | my $lvmhash = get_lvm_devices($lsblk_info); |
cbba9b5b | 498 | |
52a064af DC |
499 | my $disk_regex = ".*"; |
500 | if (defined($disks)) { | |
501 | if (!ref($disks)) { | |
502 | $disks = [ $disks ]; | |
503 | } elsif (ref($disks) ne 'ARRAY') { | |
504 | die "disks is not a string or array reference\n"; | |
505 | } | |
506 | # we get cciss/c0d0 but need cciss!c0d0 | |
5045e0b7 | 507 | $_ =~ s|cciss/|cciss!| for @$disks; |
52a064af | 508 | |
a64aedd3 FE |
509 | if ($include_partitions) { |
510 | # Proper blockdevice is needed for the regex, use parent for partitions. | |
511 | for my $disk ($disks->@*) { | |
512 | next if !is_partition("/dev/$disk"); | |
513 | $disk = strip_dev(get_blockdev("/dev/$disk")); | |
514 | } | |
515 | } | |
516 | ||
52a064af | 517 | $disk_regex = "(?:" . join('|', @$disks) . ")"; |
1590fc13 DC |
518 | } |
519 | ||
52a064af | 520 | dir_glob_foreach('/sys/block', $disk_regex, sub { |
cbba9b5b | 521 | my ($dev) = @_; |
cbba9b5b | 522 | # whitelisting following devices |
9aff3f3d TL |
523 | # - hdX ide block device |
524 | # - sdX scsi/sata block device | |
525 | # - vdX virtIO block device | |
526 | # - xvdX: xen virtual block device | |
527 | # - nvmeXnY: nvme devices | |
528 | # - cciss!cXnY cciss devices | |
cbba9b5b DC |
529 | return if $dev !~ m/^(h|s|x?v)d[a-z]+$/ && |
530 | $dev !~ m/^nvme\d+n\d+$/ && | |
38ddd4ce | 531 | $dev !~ m/^cciss\!c\d+d\d+$/; |
cbba9b5b | 532 | |
9aff3f3d | 533 | my $data = get_udev_info("/sys/block/$dev") // return; |
cbba9b5b DC |
534 | my $devpath = $data->{devpath}; |
535 | ||
536 | my $sysdir = "/sys/block/$dev"; | |
537 | ||
cbba9b5b | 538 | # we do not want iscsi devices |
eebcdb11 | 539 | return if is_iscsi($sysdir); |
cbba9b5b DC |
540 | |
541 | my $sysdata = get_sysdir_info($sysdir); | |
542 | return if !defined($sysdata); | |
543 | ||
544 | my $type = 'unknown'; | |
545 | ||
546 | if ($sysdata->{rotational} == 0) { | |
547 | $type = 'ssd'; | |
4731eb11 | 548 | $type = 'nvme' if $dev =~ m/^nvme\d+n\d+$/; |
cbba9b5b DC |
549 | $data->{rpm} = 0; |
550 | } elsif ($sysdata->{rotational} == 1) { | |
551 | if ($data->{rpm} != -1) { | |
552 | $type = 'hdd'; | |
553 | } elsif ($data->{usb}) { | |
554 | $type = 'usb'; | |
555 | $data->{rpm} = 0; | |
556 | } | |
557 | } | |
558 | ||
9aff3f3d | 559 | my ($health, $wearout) = ('UNKNOWN', 'N/A'); |
7a98a62d FG |
560 | if (!$nosmart) { |
561 | eval { | |
4731eb11 | 562 | my $smartdata = get_smart_data($devpath, !is_ssdlike($type)); |
dd902da7 DC |
563 | $health = $smartdata->{health} if $smartdata->{health}; |
564 | ||
9aff3f3d TL |
565 | if (is_ssdlike($type)) { # if we have an ssd we try to get the wearout indicator |
566 | my $wear_level = get_wear_leveling_info($smartdata); | |
567 | $wearout = $wear_level if defined($wear_level); | |
acd3d916 | 568 | } |
7a98a62d FG |
569 | }; |
570 | } | |
cbba9b5b | 571 | |
9aff3f3d TL |
572 | # we replaced cciss/ with cciss! above, but in the result we need cciss/ again because the |
573 | # caller might want to check the result again with the original parameter | |
fc7c0e05 DC |
574 | if ($dev =~ m|^cciss!|) { |
575 | $dev =~ s|^cciss!|cciss/|; | |
576 | } | |
577 | ||
cbba9b5b DC |
578 | $disklist->{$dev} = { |
579 | vendor => $sysdata->{vendor}, | |
865bdbd9 | 580 | model => $data->{model} || $sysdata->{model}, |
cbba9b5b DC |
581 | size => $sysdata->{size}, |
582 | serial => $data->{serial}, | |
583 | gpt => $data->{gpt}, | |
584 | rpm => $data->{rpm}, | |
585 | type => $type, | |
586 | wwn => $data->{wwn}, | |
587 | health => $health, | |
588 | devpath => $devpath, | |
589 | wearout => $wearout, | |
590 | }; | |
2949acd6 | 591 | $disklist->{$dev}->{mounted} = 1 if exists $mounted->{$devpath}; |
cbba9b5b | 592 | |
0f0d99a3 SI |
593 | my $by_id_link = $data->{by_id_link}; |
594 | $disklist->{$dev}->{by_id_link} = $by_id_link if defined($by_id_link); | |
595 | ||
9aff3f3d TL |
596 | my ($osdid, $bluestore, $osdencrypted) = (-1, 0, 0); |
597 | my ($journal_count, $db_count, $wal_count) = (0, 0, 0); | |
cbba9b5b | 598 | |
cbba9b5b | 599 | my $partpath = $devpath; |
9aff3f3d | 600 | # remove trailing part to get the partition base path, e.g. /dev/cciss/c0d0 -> /dev/cciss |
cbba9b5b DC |
601 | $partpath =~ s/\/[^\/]+$//; |
602 | ||
01aa7d75 FE |
603 | my $determine_usage = sub { |
604 | my ($devpath, $sysdir, $is_partition) = @_; | |
605 | ||
606 | return 'LVM' if $lvmhash->{$devpath}; | |
607 | return 'ZFS' if $zfshash->{$devpath}; | |
608 | ||
609 | my $info = $lsblk_info->{$devpath} // {}; | |
d3857eeb | 610 | |
9aff3f3d TL |
611 | if (defined(my $parttype = $info->{parttype})) { |
612 | return 'BIOS boot'if $parttype eq '21686148-6449-6e6f-744e-656564454649'; | |
613 | return 'EFI' if $parttype eq 'c12a7328-f81f-11d2-ba4b-00a0c93ec93b'; | |
614 | return 'ZFS reserved' if $parttype eq '6a945a3b-1dd2-11b2-99a6-080020736631'; | |
d3857eeb FE |
615 | } |
616 | ||
9aff3f3d | 617 | return "$info->{fstype}" if defined($info->{fstype}); |
01aa7d75 FE |
618 | return 'mounted' if $mounted->{$devpath}; |
619 | ||
620 | return if !$is_partition; | |
621 | ||
622 | # for devices, this check is done explicitly later | |
623 | return 'Device Mapper' if !dir_is_empty("$sysdir/holders"); | |
624 | ||
ff91cfae | 625 | return; # unused partition |
01aa7d75 FE |
626 | }; |
627 | ||
41f93ece FE |
628 | my $collect_ceph_info = sub { |
629 | my ($devpath) = @_; | |
630 | ||
631 | my $ceph_volume = $ceph_volume_infos->{$devpath} or return; | |
632 | $journal_count += $ceph_volume->{journal} // 0; | |
633 | $db_count += $ceph_volume->{db} // 0; | |
634 | $wal_count += $ceph_volume->{wal} // 0; | |
635 | if (defined($ceph_volume->{osdid})) { | |
636 | $osdid = $ceph_volume->{osdid}; | |
637 | $bluestore = 1 if $ceph_volume->{bluestore}; | |
638 | $osdencrypted = 1 if $ceph_volume->{encrypted}; | |
639 | } | |
6a1919b1 FE |
640 | |
641 | my $result = { %{$ceph_volume} }; | |
9aff3f3d | 642 | $result->{journals} = delete $result->{journal} if $result->{journal}; |
6a1919b1 | 643 | return $result; |
41f93ece FE |
644 | }; |
645 | ||
89c27ea8 | 646 | my $partitions = {}; |
cbba9b5b DC |
647 | dir_glob_foreach("$sysdir", "$dev.+", sub { |
648 | my ($part) = @_; | |
649 | ||
6a1919b1 FE |
650 | $partitions->{$part} = $collect_ceph_info->("$partpath/$part"); |
651 | my $lvm_based_osd = defined($partitions->{$part}); | |
652 | ||
89c27ea8 | 653 | $partitions->{$part}->{devpath} = "$partpath/$part"; |
2949c537 | 654 | $partitions->{$part}->{parent} = "$devpath"; |
2949acd6 | 655 | $partitions->{$part}->{mounted} = 1 if exists $mounted->{"$partpath/$part"}; |
89c27ea8 | 656 | $partitions->{$part}->{gpt} = $data->{gpt}; |
31ed94cc | 657 | $partitions->{$part}->{type} = 'partition'; |
9aff3f3d TL |
658 | $partitions->{$part}->{size} = get_sysdir_size("$sysdir/$part") // 0; |
659 | $partitions->{$part}->{used} = $determine_usage->("$partpath/$part", "$sysdir/$part", 1); | |
6a1919b1 | 660 | $partitions->{$part}->{osdid} //= -1; |
41f93ece | 661 | |
9aff3f3d | 662 | # avoid counting twice (e.g. partition with the LVM for the DB OSD is in $journalhash) |
41f93ece FE |
663 | return if $lvm_based_osd; |
664 | ||
665 | # Legacy handling for non-LVM based OSDs | |
0cca5356 | 666 | if (my $mp = $mounted->{"$partpath/$part"}) { |
cbba9b5b DC |
667 | if ($mp =~ m|^/var/lib/ceph/osd/ceph-(\d+)$|) { |
668 | $osdid = $1; | |
6a1919b1 | 669 | $partitions->{$part}->{osdid} = $osdid; |
cbba9b5b DC |
670 | } |
671 | } | |
672 | ||
0180fa42 TL |
673 | if (my $journal_part = $journalhash->{"$partpath/$part"}) { |
674 | $journal_count++ if $journal_part == 1; | |
675 | $db_count++ if $journal_part == 2; | |
676 | $wal_count++ if $journal_part == 3; | |
677 | $bluestore = 1 if $journal_part == 4; | |
6a1919b1 FE |
678 | |
679 | $partitions->{$part}->{journals} = 1 if $journal_part == 1; | |
680 | $partitions->{$part}->{db} = 1 if $journal_part == 2; | |
681 | $partitions->{$part}->{wal} = 1 if $journal_part == 3; | |
682 | $partitions->{$part}->{bluestore} = 1 if $journal_part == 4; | |
e2bd817c | 683 | } |
cbba9b5b DC |
684 | }); |
685 | ||
01aa7d75 | 686 | my $used = $determine_usage->($devpath, $sysdir, 0); |
2949c537 FE |
687 | if (!$include_partitions) { |
688 | foreach my $part (sort keys %{$partitions}) { | |
2949c537 FE |
689 | $used //= $partitions->{$part}->{used}; |
690 | } | |
415dc398 FE |
691 | } else { |
692 | # fstype might be set even if there are partitions, but showing that is confusing | |
693 | $used = 'partitions' if scalar(keys %{$partitions}); | |
01aa7d75 FE |
694 | } |
695 | $used //= 'partitions' if scalar(keys %{$partitions}); | |
cbba9b5b DC |
696 | # multipath, software raid, etc. |
697 | # this check comes in last, to show more specific info | |
698 | # if we have it | |
01aa7d75 | 699 | $used //= 'Device Mapper' if !dir_is_empty("$sysdir/holders"); |
cbba9b5b DC |
700 | |
701 | $disklist->{$dev}->{used} = $used if $used; | |
41f93ece FE |
702 | |
703 | $collect_ceph_info->($devpath); | |
704 | ||
cbba9b5b | 705 | $disklist->{$dev}->{osdid} = $osdid; |
e2bd817c DC |
706 | $disklist->{$dev}->{journals} = $journal_count if $journal_count; |
707 | $disklist->{$dev}->{bluestore} = $bluestore if $osdid != -1; | |
bfb3d42d | 708 | $disklist->{$dev}->{osdencrypted} = $osdencrypted if $osdid != -1; |
e2bd817c DC |
709 | $disklist->{$dev}->{db} = $db_count if $db_count; |
710 | $disklist->{$dev}->{wal} = $wal_count if $wal_count; | |
2949c537 FE |
711 | |
712 | if ($include_partitions) { | |
9aff3f3d | 713 | $disklist->{$_} = $partitions->{$_} for keys %{$partitions}; |
2949c537 | 714 | } |
cbba9b5b DC |
715 | }); |
716 | ||
717 | return $disklist; | |
cbba9b5b DC |
718 | } |
719 | ||
3196c387 WL |
720 | sub get_partnum { |
721 | my ($part_path) = @_; | |
722 | ||
92ae59df | 723 | my $st = stat($part_path); |
3196c387 | 724 | |
ceb7b1ed FE |
725 | die "error detecting block device '$part_path'\n" |
726 | if !$st || !$st->mode || !S_ISBLK($st->mode) || !$st->rdev; | |
727 | ||
92ae59df AA |
728 | my $major = PVE::Tools::dev_t_major($st->rdev); |
729 | my $minor = PVE::Tools::dev_t_minor($st->rdev); | |
3196c387 WL |
730 | my $partnum_path = "/sys/dev/block/$major:$minor/"; |
731 | ||
9aff3f3d | 732 | my $partnum = file_read_firstline("${partnum_path}partition"); |
481f6177 | 733 | die "Partition does not exist\n" if !defined($partnum); |
9aff3f3d TL |
734 | die "Failed to get partition number\n" if $partnum !~ m/(\d+)/; # untaint |
735 | $partnum = $1; | |
736 | die "Partition number $partnum is invalid\n" if $partnum > 128; | |
3196c387 WL |
737 | |
738 | return $partnum; | |
739 | } | |
740 | ||
0d28307d WL |
741 | sub get_blockdev { |
742 | my ($part_path) = @_; | |
743 | ||
1207620c TL |
744 | my ($dev, $block_dev); |
745 | if ($part_path =~ m|^/dev/(.*)$|) { | |
746 | $dev = $1; | |
747 | my $link = readlink "/sys/class/block/$dev"; | |
748 | $block_dev = $1 if $link =~ m|([^/]*)/$dev$|; | |
749 | } | |
0d28307d WL |
750 | |
751 | die "Can't parse parent device\n" if !defined($block_dev); | |
752 | die "No valid block device\n" if index($dev, $block_dev) == -1; | |
753 | ||
754 | $block_dev = "/dev/$block_dev"; | |
ffc31266 | 755 | die "Block device does not exists\n" if !(-b $block_dev); |
0d28307d WL |
756 | |
757 | return $block_dev; | |
758 | } | |
759 | ||
e8df8fb1 FE |
760 | sub is_partition { |
761 | my ($dev_path) = @_; | |
762 | ||
763 | return defined(eval { get_partnum($dev_path) }); | |
764 | } | |
765 | ||
e39e8ee2 DC |
766 | sub locked_disk_action { |
767 | my ($sub) = @_; | |
768 | my $res = PVE::Tools::lock_file('/run/lock/pve-diskmanage.lck', undef, $sub); | |
769 | die $@ if $@; | |
770 | return $res; | |
771 | } | |
772 | ||
0370861c | 773 | sub assert_disk_unused { |
76c1e57b | 774 | my ($dev) = @_; |
0370861c | 775 | die "device '$dev' is already in use\n" if disk_is_used($dev); |
9aff3f3d | 776 | return; |
76c1e57b DC |
777 | } |
778 | ||
1dc3038d DC |
779 | sub append_partition { |
780 | my ($dev, $size) = @_; | |
781 | ||
782 | my $devname = $dev; | |
783 | $devname =~ s|^/dev/||; | |
784 | ||
785 | my $newpartid = 1; | |
786 | dir_glob_foreach("/sys/block/$devname", qr/\Q$devname\E.*?(\d+)/, sub { | |
787 | my ($part, $partid) = @_; | |
788 | ||
789 | if ($partid >= $newpartid) { | |
790 | $newpartid = $partid + 1; | |
791 | } | |
792 | }); | |
793 | ||
794 | $size = PVE::Tools::convert_size($size, 'b' => 'mb'); | |
795 | ||
796 | run_command([ $SGDISK, '-n', "$newpartid:0:+${size}M", $dev ], | |
797 | errmsg => "error creating partition '$newpartid' on '$dev'"); | |
798 | ||
799 | my $partition; | |
800 | ||
ffc31266 | 801 | # loop again to detect the real partition device which does not always follow |
1dc3038d DC |
802 | # a strict $devname$partition scheme like /dev/nvme0n1 -> /dev/nvme0n1p1 |
803 | dir_glob_foreach("/sys/block/$devname", qr/\Q$devname\E.*$newpartid/, sub { | |
804 | my ($part) = @_; | |
805 | ||
806 | $partition = "/dev/$part"; | |
807 | }); | |
808 | ||
809 | return $partition; | |
810 | } | |
811 | ||
cb057e21 FE |
812 | # Check if a disk or any of its partitions has a holder. |
813 | # Can also be called with a partition. | |
814 | # Expected to be called with a result of verify_blockdev_path(). | |
815 | sub has_holder { | |
816 | my ($devpath) = @_; | |
817 | ||
70dc7098 | 818 | my $dev = strip_dev($devpath); |
cb057e21 | 819 | |
70dc7098 | 820 | return $devpath if !dir_is_empty("/sys/class/block/${dev}/holders"); |
cb057e21 FE |
821 | |
822 | my $found; | |
cb057e21 FE |
823 | dir_glob_foreach("/sys/block/${dev}", "${dev}.+", sub { |
824 | my ($part) = @_; | |
70dc7098 | 825 | $found = "/dev/${part}" if !dir_is_empty("/sys/class/block/${part}/holders"); |
cb057e21 FE |
826 | }); |
827 | ||
828 | return $found; | |
829 | } | |
830 | ||
3bf7f889 FE |
831 | # Basic check if a disk or any of its partitions is mounted. |
832 | # Can also be called with a partition. | |
833 | # Expected to be called with a result of verify_blockdev_path(). | |
834 | sub is_mounted { | |
835 | my ($devpath) = @_; | |
836 | ||
837 | my $mounted = mounted_blockdevs(); | |
838 | ||
839 | return $devpath if $mounted->{$devpath}; | |
840 | ||
70dc7098 | 841 | my $dev = strip_dev($devpath); |
3bf7f889 FE |
842 | |
843 | my $found; | |
3bf7f889 FE |
844 | dir_glob_foreach("/sys/block/${dev}", "${dev}.+", sub { |
845 | my ($part) = @_; | |
3bf7f889 FE |
846 | my $partpath = "/dev/${part}"; |
847 | ||
848 | $found = $partpath if $mounted->{$partpath}; | |
849 | }); | |
850 | ||
851 | return $found; | |
852 | } | |
853 | ||
e8df8fb1 FE |
854 | # Currently only supports GPT-partitioned disks. |
855 | sub change_parttype { | |
856 | my ($partpath, $parttype) = @_; | |
857 | ||
858 | my $err = "unable to change partition type for $partpath"; | |
859 | ||
860 | my $partnum = get_partnum($partpath); | |
861 | my $blockdev = get_blockdev($partpath); | |
862 | my $dev = strip_dev($blockdev); | |
863 | ||
864 | my $info = get_disks($dev, 1); | |
865 | die "$err - unable to get disk info for '$blockdev'\n" if !defined($info->{$dev}); | |
866 | die "$err - disk '$blockdev' is not GPT partitioned\n" if !$info->{$dev}->{gpt}; | |
867 | ||
868 | run_command(['sgdisk', "-t${partnum}:${parttype}", $blockdev], errmsg => $err); | |
869 | } | |
870 | ||
262ad7a9 | 871 | # Wipes all labels and the first 200 MiB of a disk/partition (or the whole if it is smaller). |
bd46e59b | 872 | # If called with a partition, also sets the partition type to 0x83 'Linux filesystem'. |
262ad7a9 FE |
873 | # Expected to be called with a result of verify_blockdev_path(). |
874 | sub wipe_blockdev { | |
875 | my ($devpath) = @_; | |
876 | ||
262ad7a9 FE |
877 | my $devname = basename($devpath); |
878 | my $dev_size = PVE::Tools::file_get_contents("/sys/class/block/$devname/size"); | |
879 | ||
880 | ($dev_size) = $dev_size =~ m|(\d+)|; # untaint $dev_size | |
881 | die "Couldn't get the size of the device $devname\n" if !defined($dev_size); | |
882 | ||
883 | my $size = ($dev_size * 512 / 1024 / 1024); | |
884 | my $count = ($size < 200) ? $size : 200; | |
885 | ||
839afff8 TL |
886 | my $to_wipe = []; |
887 | dir_glob_foreach("/sys/class/block/${devname}", "${devname}.+", sub { | |
888 | my ($part) = @_; | |
889 | push $to_wipe->@*, "/dev/${part}" if -b "/dev/${part}"; | |
890 | }); | |
891 | ||
f7a95153 | 892 | if (scalar($to_wipe->@*) > 0) { |
d9381782 | 893 | print "found child partitions to wipe: ". join(', ', $to_wipe->@*) ."\n"; |
839afff8 TL |
894 | } |
895 | push $to_wipe->@*, $devpath; # put actual device last | |
896 | ||
897 | print "wiping block device ${devpath}\n"; | |
262ad7a9 | 898 | |
839afff8 | 899 | run_command(['wipefs', '--all', $to_wipe->@*], errmsg => "error wiping '${devpath}'"); |
fa6d05ab TL |
900 | |
901 | run_command( | |
902 | ['dd', 'if=/dev/zero', "of=${devpath}", 'bs=1M', 'conv=fdatasync', "count=${count}"], | |
903 | errmsg => "error wiping '${devpath}'", | |
904 | ); | |
bd46e59b FE |
905 | |
906 | if (is_partition($devpath)) { | |
907 | eval { change_parttype($devpath, '8300'); }; | |
908 | warn $@ if $@; | |
909 | } | |
262ad7a9 FE |
910 | } |
911 | ||
26082b7d FE |
912 | # FIXME: Remove once we depend on systemd >= v249. |
913 | # Work around udev bug https://github.com/systemd/systemd/issues/18525 ensuring database is updated. | |
914 | sub udevadm_trigger { | |
915 | my @devs = @_; | |
916 | ||
917 | return if scalar(@devs) == 0; | |
918 | ||
919 | eval { run_command(['udevadm', 'trigger', @devs]); }; | |
920 | warn $@ if $@; | |
921 | } | |
922 | ||
cbba9b5b | 923 | 1; |