2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
28 * The version-1 superblock :
29 * All numeric fields are little-endian.
31 * total size: 256 bytes plus 2 per device.
32 * 1K allows 384 devices.
34 struct mdp_superblock_1
{
35 /* constant array information - 128 bytes */
36 __u32 magic
; /* MD_SB_MAGIC: 0xa92b4efc - little endian */
37 __u32 major_version
; /* 1 */
38 __u32 feature_map
; /* 0 for now */
39 __u32 pad0
; /* always set to 0 when writing */
41 __u8 set_uuid
[16]; /* user-space generated. */
42 char set_name
[32]; /* set and interpreted by user-space */
44 __u64 ctime
; /* lo 40 bits are seconds, top 24 are microseconds or 0*/
45 __u32 level
; /* -4 (multipath), -1 (linear), 0,1,4,5 */
46 __u32 layout
; /* only for raid5 currently */
47 __u64 size
; /* used size of component devices, in 512byte sectors */
49 __u32 chunksize
; /* in 512byte sectors */
51 __u32 bitmap_offset
; /* sectors after start of superblock that bitmap starts
52 * NOTE: signed, so bitmap can be before superblock
53 * only meaningful of feature_map[0] is set.
56 /* These are only valid with feature bit '4' */
57 __u32 new_level
; /* new level we are reshaping to */
58 __u64 reshape_position
; /* next address in array-space for reshape */
59 __u32 delta_disks
; /* change in number of raid_disks */
60 __u32 new_layout
; /* new layout */
61 __u32 new_chunk
; /* new chunk size (sectors) */
62 __u32 new_offset
; /* signed number to add to data_offset in new
63 * layout. 0 == no-change. This can be
64 * different on each device in the array.
67 /* constant this-device information - 64 bytes */
68 __u64 data_offset
; /* sector start of data, often 0 */
69 __u64 data_size
; /* sectors in this device that can be used for data */
70 __u64 super_offset
; /* sector start of this superblock */
72 __u64 recovery_offset
;/* sectors before this offset (from data_offset) have been recovered */
73 __u64 journal_tail
;/* journal tail of journal device (from data_offset) */
75 __u32 dev_number
; /* permanent identifier of this device - not role in raid */
76 __u32 cnt_corrected_read
; /* number of read errors that were corrected by re-writing */
77 __u8 device_uuid
[16]; /* user-space setable, ignored by kernel */
78 __u8 devflags
; /* per-device flags. Only one defined...*/
79 #define WriteMostly1 1 /* mask for writemostly flag in above */
80 #define FailFast1 2 /* Device should get FailFast requests */
81 /* bad block log. If there are any bad blocks the feature flag is set.
82 * if offset and size are non-zero, that space is reserved and available.
84 __u8 bblog_shift
; /* shift from sectors to block size for badblocklist */
85 __u16 bblog_size
; /* number of sectors reserved for badblocklist */
86 __u32 bblog_offset
; /* sector offset from superblock to bblog, signed */
88 /* array state information - 64 bytes */
89 __u64 utime
; /* 40 bits second, 24 btes microseconds */
90 __u64 events
; /* incremented when superblock updated */
91 __u64 resync_offset
; /* data before this offset (from data_offset) known to be in sync */
92 __u32 sb_csum
; /* checksum upto dev_roles[max_dev] */
93 __u32 max_dev
; /* size of dev_roles[] array to consider */
94 __u8 pad3
[64-32]; /* set to 0 when writing */
96 /* device state information. Indexed by dev_number.
98 * Note there are no per-device state flags. State information is rolled
99 * into the 'roles' value. If a device is spare or faulty, then it doesn't
100 * have a meaningful role.
102 __u16 dev_roles
[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
105 #define MAX_SB_SIZE 4096
106 /* bitmap super size is 256, but we round up to a sector for alignment */
107 #define BM_SUPER_SIZE 512
108 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
109 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
110 + sizeof(struct misc_dev_info))
112 struct misc_dev_info
{
116 /* feature_map bits */
117 #define MD_FEATURE_BITMAP_OFFSET 1
118 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
121 #define MD_FEATURE_RESHAPE_ACTIVE 4
122 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
123 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
124 * active device with same 'role'.
125 * 'recovery_offset' is also set.
127 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
128 * of devices, but is going
131 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
132 #define MD_FEATURE_BITMAP_VERSIONED 256 /* bitmap version number checked properly */
133 #define MD_FEATURE_JOURNAL 512 /* support write journal */
134 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
135 |MD_FEATURE_RECOVERY_OFFSET \
136 |MD_FEATURE_RESHAPE_ACTIVE \
137 |MD_FEATURE_BAD_BLOCKS \
138 |MD_FEATURE_REPLACEMENT \
139 |MD_FEATURE_RESHAPE_BACKWARDS \
140 |MD_FEATURE_NEW_OFFSET \
141 |MD_FEATURE_BITMAP_VERSIONED \
142 |MD_FEATURE_JOURNAL \
146 static int role_from_sb(struct mdp_superblock_1
*sb
)
151 d
= __le32_to_cpu(sb
->dev_number
);
152 if (d
< __le32_to_cpu(sb
->max_dev
))
153 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
155 role
= MD_DISK_ROLE_SPARE
;
160 /* return how many bytes are needed for bitmap, for cluster-md each node
161 * should have it's own bitmap */
162 static unsigned int calc_bitmap_size(bitmap_super_t
*bms
, unsigned int boundary
)
164 unsigned long long bits
, bytes
;
166 bits
= bitmap_bits(__le64_to_cpu(bms
->sync_size
),
167 __le32_to_cpu(bms
->chunksize
));
168 bytes
= (bits
+7) >> 3;
169 bytes
+= sizeof(bitmap_super_t
);
170 bytes
= ROUND_UP(bytes
, boundary
);
175 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
177 unsigned int disk_csum
, csum
;
178 unsigned long long newcsum
;
179 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
180 unsigned int *isuper
= (unsigned int*)sb
;
182 /* make sure I can count... */
183 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
184 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
185 sizeof(struct mdp_superblock_1
) != 256) {
186 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
189 disk_csum
= sb
->sb_csum
;
192 for (; size
>=4; size
-= 4 ) {
193 newcsum
+= __le32_to_cpu(*isuper
);
198 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
200 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
201 sb
->sb_csum
= disk_csum
;
202 return __cpu_to_le32(csum
);
206 * Information related to file descriptor used for aligned reads/writes.
207 * Cache the block size.
214 static void init_afd(struct align_fd
*afd
, int fd
)
217 if (!get_dev_sector_size(afd
->fd
, NULL
, (unsigned int *)&afd
->blk_sz
))
221 static char abuf
[4096+4096];
222 static int aread(struct align_fd
*afd
, void *buf
, int len
)
225 * On devices with a 4K sector size, we need to read
226 * the full sector and copy relevant bits into
235 if (!bsize
|| bsize
> 4096 || len
> 4096) {
237 fprintf(stderr
, "WARNING - aread() called with invalid block size\n");
240 b
= ROUND_UP_PTR((char *)abuf
, 4096);
242 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
244 n
= read(afd
->fd
, b
, iosize
);
247 lseek(afd
->fd
, len
- n
, 1);
254 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
257 * On devices with a 4K sector size, we need to write
258 * the full sector. We pre-read if the sector is larger
260 * The address must be sector-aligned.
267 if (!bsize
|| bsize
> 4096 || len
> 4096) {
269 fprintf(stderr
, "WARNING - awrite() called with invalid block size\n");
272 b
= ROUND_UP_PTR((char *)abuf
, 4096);
274 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
278 n
= read(afd
->fd
, b
, iosize
);
281 lseek(afd
->fd
, -n
, 1);
285 n
= write(afd
->fd
, b
, iosize
);
288 lseek(afd
->fd
, len
- n
, 1);
293 static void examine_super1(struct supertype
*st
, char *homehost
)
295 struct mdp_superblock_1
*sb
= st
->sb
;
296 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
303 int l
= homehost
? strlen(homehost
) : 0;
305 unsigned long long sb_offset
;
308 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
309 printf(" Version : 1");
310 sb_offset
= __le64_to_cpu(sb
->super_offset
);
313 else if (sb_offset
<= 8)
317 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
318 printf(" Array UUID : ");
319 for (i
=0; i
<16; i
++) {
320 if ((i
&3)==0 && i
!= 0) printf(":");
321 printf("%02x", sb
->set_uuid
[i
]);
324 printf(" Name : %.32s", sb
->set_name
);
325 if (l
> 0 && l
< 32 &&
326 sb
->set_name
[l
] == ':' &&
327 strncmp(sb
->set_name
, homehost
, l
) == 0)
328 printf(" (local to host %s)", homehost
);
330 if (bms
->nodes
> 0 && (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
331 printf(" Cluster Name : %-64s\n", bms
->cluster_name
);
332 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
333 printf(" Creation Time : %.24s\n", ctime(&atime
));
334 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
335 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
336 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
338 printf(" Avail Dev Size : %llu%s\n",
339 (unsigned long long)__le64_to_cpu(sb
->data_size
),
340 human_size(__le64_to_cpu(sb
->data_size
)<<9));
341 if (__le32_to_cpu(sb
->level
) > 0) {
342 int ddsks
= 0, ddsks_denom
= 1;
343 switch(__le32_to_cpu(sb
->level
)) {
344 case 1: ddsks
=1;break;
346 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
347 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
349 layout
= __le32_to_cpu(sb
->layout
);
350 ddsks
= __le32_to_cpu(sb
->raid_disks
);
351 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
354 long long asize
= __le64_to_cpu(sb
->size
);
355 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
356 printf(" Array Size : %llu%s\n",
357 asize
>> 10, human_size(asize
));
359 if (sb
->size
!= sb
->data_size
)
360 printf(" Used Dev Size : %llu%s\n",
361 (unsigned long long)__le64_to_cpu(sb
->size
),
362 human_size(__le64_to_cpu(sb
->size
)<<9));
365 printf(" Data Offset : %llu sectors\n",
366 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
367 if (sb
->new_offset
&&
368 (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
)) {
369 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
370 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
371 printf(" New Offset : %llu sectors\n", offset
);
373 printf(" Super Offset : %llu sectors\n",
374 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
375 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
376 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
378 st
->ss
->getinfo_super(st
, &info
, NULL
);
379 if (info
.space_after
!= 1 &&
380 !(__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
381 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
382 info
.space_before
, info
.space_after
);
384 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
385 printf(" Device UUID : ");
386 for (i
=0; i
<16; i
++) {
387 if ((i
&3)==0 && i
!= 0) printf(":");
388 printf("%02x", sb
->device_uuid
[i
]);
392 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
393 printf("Internal Bitmap : %ld sectors from superblock\n",
394 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
396 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
397 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
398 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
399 if (__le32_to_cpu(sb
->delta_disks
)) {
400 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
401 printf(" (%d->%d)\n",
402 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
403 __le32_to_cpu(sb
->raid_disks
));
404 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
405 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
407 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
408 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
409 printf(" New Level : %s\n", c
?c
:"-unknown-");
411 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
412 if (__le32_to_cpu(sb
->level
) == 5) {
413 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
414 printf(" New Layout : %s\n", c
?c
:"-unknown-");
416 if (__le32_to_cpu(sb
->level
) == 6) {
417 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
418 printf(" New Layout : %s\n", c
?c
:"-unknown-");
420 if (__le32_to_cpu(sb
->level
) == 10) {
421 printf(" New Layout :");
422 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
426 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
427 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
432 if (sb
->devflags
& WriteMostly1
)
433 printf(" write-mostly");
434 if (sb
->devflags
& FailFast1
)
439 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
440 printf(" Update Time : %.24s\n", ctime(&atime
));
442 if (sb
->bblog_size
&& sb
->bblog_offset
) {
443 printf(" Bad Block Log : %d entries available at offset %ld sectors",
444 __le16_to_cpu(sb
->bblog_size
)*512/8,
445 (long)(int32_t)__le32_to_cpu(sb
->bblog_offset
));
446 if (sb
->feature_map
&
447 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
448 printf(" - bad blocks present.");
452 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
453 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
455 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
456 __le32_to_cpu(calc_sb_1_csum(sb
)));
457 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
459 if (__le32_to_cpu(sb
->level
) == 5) {
460 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
461 printf(" Layout : %s\n", c
?c
:"-unknown-");
463 if (__le32_to_cpu(sb
->level
) == 6) {
464 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
465 printf(" Layout : %s\n", c
?c
:"-unknown-");
467 if (__le32_to_cpu(sb
->level
) == 10) {
468 int lo
= __le32_to_cpu(sb
->layout
);
470 print_r10_layout(lo
);
473 switch(__le32_to_cpu(sb
->level
)) {
479 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
482 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
488 /* This turns out to just be confusing */
489 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
490 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
491 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != MD_DISK_ROLE_SPARE
)
493 for (d
=0; d
< i
; d
++) {
494 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
496 if (role
== MD_DISK_ROLE_SPARE
) printf("empty");
497 else if(role
== MD_DISK_ROLE_FAULTY
) printf("failed");
498 else printf("%d", role
);
502 printf(" Device Role : ");
503 role
= role_from_sb(sb
);
504 if (role
>= MD_DISK_ROLE_FAULTY
)
506 else if (role
== MD_DISK_ROLE_JOURNAL
)
508 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
509 printf("Replacement device %d\n", role
);
511 printf("Active device %d\n", role
);
513 printf(" Array State : ");
514 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
517 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
518 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
532 /* This is confusing too */
534 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
535 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
536 if (role
== MD_DISK_ROLE_FAULTY
)
539 if (faulty
) printf(" %d failed", faulty
);
541 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
545 static void brief_examine_super1(struct supertype
*st
, int verbose
)
547 struct mdp_superblock_1
*sb
= st
->sb
;
549 unsigned long long sb_offset
;
551 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
553 nm
= strchr(sb
->set_name
, ':');
556 else if (sb
->set_name
[0])
568 printf(" level=%s", c
);
569 sb_offset
= __le64_to_cpu(sb
->super_offset
);
571 printf(" metadata=1.1 ");
572 else if (sb_offset
<= 8)
573 printf(" metadata=1.2 ");
575 printf(" metadata=1.0 ");
577 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
579 for (i
=0; i
<16; i
++) {
580 if ((i
&3)==0 && i
!= 0) printf(":");
581 printf("%02x", sb
->set_uuid
[i
]);
583 if (sb
->set_name
[0]) {
585 print_quoted(sb
->set_name
);
590 static void export_examine_super1(struct supertype
*st
)
592 struct mdp_superblock_1
*sb
= st
->sb
;
597 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
598 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
600 if (sb
->set_name
[i
] == '\n' ||
601 sb
->set_name
[i
] == '\0') {
606 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
607 if (__le32_to_cpu(sb
->level
) > 0) {
608 int ddsks
= 0, ddsks_denom
= 1;
609 switch(__le32_to_cpu(sb
->level
)) {
610 case 1: ddsks
=1;break;
612 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
613 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
615 layout
= __le32_to_cpu(sb
->layout
);
616 ddsks
= __le32_to_cpu(sb
->raid_disks
);
617 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
620 long long asize
= __le64_to_cpu(sb
->size
);
621 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
622 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
626 for (i
=0; i
<16; i
++) {
627 if ((i
&3)==0 && i
!= 0) printf(":");
628 printf("%02x", sb
->set_uuid
[i
]);
631 printf("MD_UPDATE_TIME=%llu\n",
632 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
633 printf("MD_DEV_UUID=");
634 for (i
=0; i
<16; i
++) {
635 if ((i
&3)==0 && i
!= 0) printf(":");
636 printf("%02x", sb
->device_uuid
[i
]);
639 printf("MD_EVENTS=%llu\n",
640 (unsigned long long)__le64_to_cpu(sb
->events
));
643 static int copy_metadata1(struct supertype
*st
, int from
, int to
)
645 /* Read superblock. If it looks good, write it out.
646 * Then if a bitmap is present, copy that.
647 * And if a bad-block-list is present, copy that too.
650 unsigned long long dsize
, sb_offset
;
651 const int bufsize
= 4*1024;
652 struct mdp_superblock_1 super
, *sb
;
654 if (posix_memalign(&buf
, 4096, bufsize
) != 0)
657 if (!get_dev_size(from
, NULL
, &dsize
))
663 switch(st
->minor_version
) {
667 sb_offset
&= ~(4*2-1);
679 if (lseek64(from
, sb_offset
<< 9, 0) < 0LL)
681 if (read(from
, buf
, bufsize
) != bufsize
)
685 super
= *sb
; // save most of sb for when we reuse buf
687 if (__le32_to_cpu(super
.magic
) != MD_SB_MAGIC
||
688 __le32_to_cpu(super
.major_version
) != 1 ||
689 __le64_to_cpu(super
.super_offset
) != sb_offset
||
690 calc_sb_1_csum(sb
) != super
.sb_csum
)
693 if (lseek64(to
, sb_offset
<< 9, 0) < 0LL)
695 if (write(to
, buf
, bufsize
) != bufsize
)
698 if (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
)) {
699 unsigned long long bitmap_offset
= sb_offset
;
700 int bytes
= 4096; // just an estimate.
702 struct align_fd afrom
, ato
;
704 init_afd(&afrom
, from
);
707 bitmap_offset
+= (int32_t)__le32_to_cpu(super
.bitmap_offset
);
709 if (lseek64(from
, bitmap_offset
<<9, 0) < 0)
711 if (lseek64(to
, bitmap_offset
<<9, 0) < 0)
714 for (written
= 0; written
< bytes
; ) {
715 int n
= bytes
- written
;
718 if (aread(&afrom
, buf
, n
) != n
)
721 /* have the header, can calculate
722 * correct bitmap bytes */
725 bytes
= calc_bitmap_size(bms
, 512);
729 if (awrite(&ato
, buf
, n
) != n
)
735 if (super
.bblog_size
!= 0 &&
736 __le16_to_cpu(super
.bblog_size
) <= 100 &&
737 super
.bblog_offset
!= 0 &&
738 (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BAD_BLOCKS
))) {
739 /* There is a bad block log */
740 unsigned long long bb_offset
= sb_offset
;
741 int bytes
= __le16_to_cpu(super
.bblog_size
) * 512;
743 struct align_fd afrom
, ato
;
745 init_afd(&afrom
, from
);
748 bb_offset
+= (int32_t)__le32_to_cpu(super
.bblog_offset
);
750 if (lseek64(from
, bb_offset
<<9, 0) < 0)
752 if (lseek64(to
, bb_offset
<<9, 0) < 0)
755 for (written
= 0; written
< bytes
; ) {
756 int n
= bytes
- written
;
759 if (aread(&afrom
, buf
, n
) != n
)
762 if (awrite(&ato
, buf
, n
) != n
)
776 static void detail_super1(struct supertype
*st
, char *homehost
)
778 struct mdp_superblock_1
*sb
= st
->sb
;
779 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
781 int l
= homehost
? strlen(homehost
) : 0;
783 printf(" Name : %.32s", sb
->set_name
);
784 if (l
> 0 && l
< 32 &&
785 sb
->set_name
[l
] == ':' &&
786 strncmp(sb
->set_name
, homehost
, l
) == 0)
787 printf(" (local to host %s)", homehost
);
788 if (bms
->nodes
> 0 && (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
789 printf("\n Cluster Name : %-64s", bms
->cluster_name
);
790 printf("\n UUID : ");
791 for (i
=0; i
<16; i
++) {
792 if ((i
&3)==0 && i
!= 0) printf(":");
793 printf("%02x", sb
->set_uuid
[i
]);
795 printf("\n Events : %llu\n\n",
796 (unsigned long long)__le64_to_cpu(sb
->events
));
799 static void brief_detail_super1(struct supertype
*st
)
801 struct mdp_superblock_1
*sb
= st
->sb
;
804 if (sb
->set_name
[0]) {
806 print_quoted(sb
->set_name
);
809 for (i
=0; i
<16; i
++) {
810 if ((i
&3)==0 && i
!= 0) printf(":");
811 printf("%02x", sb
->set_uuid
[i
]);
815 static void export_detail_super1(struct supertype
*st
)
817 struct mdp_superblock_1
*sb
= st
->sb
;
822 if (sb
->set_name
[i
] == '\n' ||
823 sb
->set_name
[i
] == '\0') {
828 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
831 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
833 struct mdp_superblock_1
*sb
= st
->sb
;
834 unsigned long long offset
;
839 if (!sb
->bblog_size
|| __le16_to_cpu(sb
->bblog_size
) > 100
840 || !sb
->bblog_offset
){
841 printf("No bad-blocks list configured on %s\n", devname
);
844 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
846 printf("Bad-blocks list is empty in %s\n", devname
);
850 size
= __le16_to_cpu(sb
->bblog_size
)* 512;
851 if (posix_memalign((void**)&bbl
, 4096, size
) != 0) {
852 pr_err("could not allocate badblocks list\n");
855 offset
= __le64_to_cpu(sb
->super_offset
) +
856 (int)__le32_to_cpu(sb
->bblog_offset
);
858 if (lseek64(fd
, offset
, 0) < 0) {
859 pr_err("Cannot seek to bad-blocks list\n");
862 if (read(fd
, bbl
, size
) != size
) {
863 pr_err("Cannot read bad-blocks list\n");
866 /* 64bits per entry. 10 bits is block-count, 54 bits is block
867 * offset. Blocks are sectors unless bblog->shift makes them bigger
870 printf("Bad-blocks on %s:\n", devname
);
871 for (i
= 0; i
< size
/8; i
++, bbp
++) {
872 __u64 bb
= __le64_to_cpu(*bbp
);
873 int count
= bb
& 0x3ff;
874 unsigned long long sector
= bb
>> 10;
879 sector
<<= sb
->bblog_shift
;
880 count
<<= sb
->bblog_shift
;
882 printf("%20llu for %d sectors\n", sector
, count
);
889 static int match_home1(struct supertype
*st
, char *homehost
)
891 struct mdp_superblock_1
*sb
= st
->sb
;
892 int l
= homehost
? strlen(homehost
) : 0;
894 return (l
> 0 && l
< 32 &&
895 sb
->set_name
[l
] == ':' &&
896 strncmp(sb
->set_name
, homehost
, l
) == 0);
899 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
901 struct mdp_superblock_1
*super
= st
->sb
;
902 char *cuuid
= (char*)uuid
;
905 cuuid
[i
] = super
->set_uuid
[i
];
908 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
910 struct mdp_superblock_1
*sb
= st
->sb
;
911 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
912 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
916 unsigned int map_disks
= info
->array
.raid_disks
;
917 unsigned long long super_offset
;
918 unsigned long long data_size
;
920 memset(info
, 0, sizeof(*info
));
921 info
->array
.major_version
= 1;
922 info
->array
.minor_version
= st
->minor_version
;
923 info
->array
.patch_version
= 0;
924 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
925 info
->array
.level
= __le32_to_cpu(sb
->level
);
926 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
927 info
->array
.md_minor
= -1;
928 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
929 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
930 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
932 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
934 if (__le32_to_cpu(bsb
->nodes
) > 1)
935 info
->array
.state
|= (1 << MD_SB_CLUSTERED
);
937 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
938 info
->component_size
= __le64_to_cpu(sb
->size
);
939 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
940 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
942 info
->disk
.major
= 0;
943 info
->disk
.minor
= 0;
944 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
945 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
946 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
947 role
= MD_DISK_ROLE_FAULTY
;
949 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
951 super_offset
= __le64_to_cpu(sb
->super_offset
);
952 if (info
->array
.level
<= 0)
953 data_size
= __le64_to_cpu(sb
->data_size
);
955 data_size
= __le64_to_cpu(sb
->size
);
956 if (info
->data_offset
< super_offset
) {
957 unsigned long long end
;
958 info
->space_before
= info
->data_offset
;
961 if (sb
->bblog_offset
&& sb
->bblog_size
) {
962 unsigned long long bboffset
= super_offset
;
963 bboffset
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
968 if (super_offset
+ info
->bitmap_offset
< end
)
969 end
= super_offset
+ info
->bitmap_offset
;
971 if (info
->data_offset
+ data_size
< end
)
972 info
->space_after
= end
- data_size
- info
->data_offset
;
974 info
->space_after
= 0;
976 unsigned long long earliest
;
977 earliest
= super_offset
+ (32+4)*2; /* match kernel */
978 if (info
->bitmap_offset
> 0) {
979 unsigned long long bmend
= info
->bitmap_offset
;
980 unsigned long long size
= calc_bitmap_size(bsb
, 4096);
983 if (bmend
> earliest
)
986 if (sb
->bblog_offset
&& sb
->bblog_size
) {
987 unsigned long long bbend
= super_offset
;
988 bbend
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
989 bbend
+= __le16_to_cpu(sb
->bblog_size
);
990 if (bbend
> earliest
)
993 if (earliest
< info
->data_offset
)
994 info
->space_before
= info
->data_offset
- earliest
;
996 info
->space_before
= 0;
997 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
999 if (info
->space_before
== 0 && info
->space_after
== 0) {
1000 /* It will look like we don't support data_offset changes,
1001 * be we do - it's just that there is no room.
1002 * A change that reduced the number of devices should
1003 * still be allowed, so set the otherwise useless value of '1'
1005 info
->space_after
= 1;
1008 info
->disk
.raid_disk
= -1;
1010 case MD_DISK_ROLE_SPARE
:
1011 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
1013 case MD_DISK_ROLE_FAULTY
:
1014 info
->disk
.state
= 1; /* faulty */
1016 case MD_DISK_ROLE_JOURNAL
:
1017 info
->disk
.state
= (1 << MD_DISK_JOURNAL
);
1018 info
->disk
.raid_disk
= role
;
1019 info
->space_after
= (misc
->device_size
- info
->data_offset
) % 8; /* journal uses all 4kB blocks*/
1022 info
->disk
.state
= 6; /* active and in sync */
1023 info
->disk
.raid_disk
= role
;
1025 if (sb
->devflags
& WriteMostly1
)
1026 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1027 if (sb
->devflags
& FailFast1
)
1028 info
->disk
.state
|= (1 << MD_DISK_FAILFAST
);
1029 info
->events
= __le64_to_cpu(sb
->events
);
1030 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
1031 info
->safe_mode_delay
= 200;
1033 memcpy(info
->uuid
, sb
->set_uuid
, 16);
1035 strncpy(info
->name
, sb
->set_name
, 32);
1038 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
1039 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
1040 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
1043 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
1044 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
1046 info
->recovery_start
= MaxSector
;
1048 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
1049 info
->reshape_active
= 1;
1050 if ((sb
->feature_map
& __le32_to_cpu(MD_FEATURE_NEW_OFFSET
)) &&
1051 sb
->new_offset
!= 0)
1052 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
1053 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
1054 info
->new_level
= __le32_to_cpu(sb
->new_level
);
1055 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
1056 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
1057 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
1058 if (info
->delta_disks
< 0)
1059 info
->array
.raid_disks
-= info
->delta_disks
;
1061 info
->reshape_active
= 0;
1063 info
->recovery_blocked
= info
->reshape_active
;
1066 for (i
=0; i
<map_disks
; i
++)
1068 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
1069 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
1070 if (/*role == MD_DISK_ROLE_SPARE || */role
< (unsigned) info
->array
.raid_disks
) {
1072 if (map
&& role
< map_disks
)
1077 info
->array
.working_disks
= working
;
1078 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_JOURNAL
))
1079 info
->journal_device_required
= 1;
1080 info
->journal_clean
= 0;
1083 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
1085 struct mdinfo
*info
;
1090 info
= xmalloc(sizeof(*info
));
1091 getinfo_super1(st
, info
, NULL
);
1095 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
1097 char *devname
, int verbose
,
1098 int uuid_set
, char *homehost
)
1100 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1101 * if any change was made. For others, the return value is
1106 struct mdp_superblock_1
*sb
= st
->sb
;
1107 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1109 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1110 rv
= cluster_get_dlmlock(&lockid
);
1112 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1113 cluster_release_dlmlock(lockid
);
1118 if (strcmp(update
, "homehost") == 0 &&
1120 /* Note that 'homehost' is special as it is really
1125 c
= strchr(sb
->set_name
, ':');
1127 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1129 strncpy(info
->name
, sb
->set_name
, 32);
1133 if (strcmp(update
, "force-one")==0) {
1134 /* Not enough devices for a working array,
1135 * so bring this one up-to-date
1137 if (sb
->events
!= __cpu_to_le64(info
->events
))
1139 sb
->events
= __cpu_to_le64(info
->events
);
1140 } else if (strcmp(update
, "force-array")==0) {
1141 /* Degraded array and 'force' requests to
1142 * maybe need to mark it 'clean'.
1144 switch(__le32_to_cpu(sb
->level
)) {
1145 case 5: case 4: case 6:
1146 /* need to force clean */
1147 if (sb
->resync_offset
!= MaxSector
)
1149 sb
->resync_offset
= MaxSector
;
1151 } else if (strcmp(update
, "assemble")==0) {
1152 int d
= info
->disk
.number
;
1154 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
1155 want
= info
->disk
.raid_disk
;
1156 else if (info
->disk
.state
& (1<<MD_DISK_JOURNAL
))
1157 want
= MD_DISK_ROLE_JOURNAL
;
1159 want
= MD_DISK_ROLE_SPARE
;
1160 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
1161 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
1164 if (info
->reshape_active
&&
1165 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1166 info
->delta_disks
>= 0 &&
1167 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
1168 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1171 if (info
->reshape_active
&&
1172 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1173 info
->delta_disks
< 0 &&
1174 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
1175 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1178 } else if (strcmp(update
, "linear-grow-new") == 0) {
1181 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
1183 for (i
=0 ; i
< max
; i
++)
1184 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= MD_DISK_ROLE_FAULTY
)
1186 sb
->dev_number
= __cpu_to_le32(i
);
1187 info
->disk
.number
= i
;
1188 if (max
>= __le32_to_cpu(sb
->max_dev
))
1189 sb
->max_dev
= __cpu_to_le32(max
+1);
1191 random_uuid(sb
->device_uuid
);
1194 __cpu_to_le16(info
->disk
.raid_disk
);
1196 fd
= open(devname
, O_RDONLY
);
1198 unsigned long long ds
;
1199 get_dev_size(fd
, devname
, &ds
);
1202 if (__le64_to_cpu(sb
->super_offset
) <
1203 __le64_to_cpu(sb
->data_offset
)) {
1204 sb
->data_size
= __cpu_to_le64(
1205 ds
- __le64_to_cpu(sb
->data_offset
));
1208 ds
&= ~(unsigned long long)(4*2-1);
1209 sb
->super_offset
= __cpu_to_le64(ds
);
1210 sb
->data_size
= __cpu_to_le64(
1211 ds
- __le64_to_cpu(sb
->data_offset
));
1214 } else if (strcmp(update
, "linear-grow-update") == 0) {
1215 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1216 sb
->dev_roles
[info
->disk
.number
] =
1217 __cpu_to_le16(info
->disk
.raid_disk
);
1218 } else if (strcmp(update
, "resync") == 0) {
1219 /* make sure resync happens */
1220 sb
->resync_offset
= 0ULL;
1221 } else if (strcmp(update
, "uuid") == 0) {
1222 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1224 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
)
1225 memcpy(bms
->uuid
, sb
->set_uuid
, 16);
1226 } else if (strcmp(update
, "no-bitmap") == 0) {
1227 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1228 } else if (strcmp(update
, "bbl") == 0) {
1229 /* only possible if there is room after the bitmap, or if
1230 * there is no bitmap
1232 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1233 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1234 long bitmap_offset
= 0;
1235 long bm_sectors
= 0;
1239 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1240 bitmap_offset
= (long)__le32_to_cpu(sb
->bitmap_offset
);
1241 bm_sectors
= calc_bitmap_size(bms
, 4096) >> 9;
1244 if (sb_offset
< data_offset
) {
1245 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1248 bb_offset
= sb_offset
+ 8;
1249 if (bm_sectors
&& bitmap_offset
> 0)
1250 bb_offset
= bitmap_offset
+ bm_sectors
;
1251 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1252 && bb_offset
+ 8+8 <= (long)data_offset
)
1253 /* too close to bitmap, and room to grow */
1255 if (bb_offset
+ 8 <= (long)data_offset
) {
1256 sb
->bblog_size
= __cpu_to_le16(8);
1257 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1260 /* 1.0 - Put bbl just before super block */
1261 if (bm_sectors
&& bitmap_offset
< 0)
1262 space
= -bitmap_offset
- bm_sectors
;
1264 space
= sb_offset
- data_offset
-
1265 __le64_to_cpu(sb
->data_size
);
1267 sb
->bblog_size
= __cpu_to_le16(8);
1268 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1271 } else if (strcmp(update
, "no-bbl") == 0) {
1272 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1273 pr_err("Cannot remove active bbl from %s\n",devname
);
1276 sb
->bblog_shift
= 0;
1277 sb
->bblog_offset
= 0;
1279 } else if (strcmp(update
, "force-no-bbl") == 0) {
1280 sb
->feature_map
&= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1282 sb
->bblog_shift
= 0;
1283 sb
->bblog_offset
= 0;
1284 } else if (strcmp(update
, "name") == 0) {
1285 if (info
->name
[0] == 0)
1286 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1287 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1289 strchr(info
->name
, ':') == NULL
&&
1290 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1291 strcpy(sb
->set_name
, homehost
);
1292 strcat(sb
->set_name
, ":");
1293 strcat(sb
->set_name
, info
->name
);
1295 strncpy(sb
->set_name
, info
->name
, sizeof(sb
->set_name
));
1296 } else if (strcmp(update
, "devicesize") == 0 &&
1297 __le64_to_cpu(sb
->super_offset
) <
1298 __le64_to_cpu(sb
->data_offset
)) {
1299 /* set data_size to device size less data_offset */
1300 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1301 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1302 sb
->data_size
= __cpu_to_le64(
1303 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1304 } else if (strncmp(update
, "revert-reshape", 14) == 0) {
1306 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1307 pr_err("No active reshape to revert on %s\n",
1311 unsigned long long reshape_sectors
;
1314 /* If the reshape hasn't started, just stop it.
1315 * It is conceivable that a stripe was modified but
1316 * the metadata not updated. In that case the backup
1317 * should have been used to get passed the critical stage.
1318 * If that couldn't happen, the "-nobackup" version
1321 if (strcmp(update
, "revert-reshape-nobackup") == 0 &&
1322 sb
->reshape_position
== 0 &&
1323 (__le32_to_cpu(sb
->delta_disks
) > 0 ||
1324 (__le32_to_cpu(sb
->delta_disks
) == 0 &&
1325 !(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
))))) {
1326 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1327 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1328 __le32_to_cpu(sb
->delta_disks
));
1329 sb
->delta_disks
= 0;
1332 /* reshape_position is a little messy.
1333 * Its value must be a multiple of the larger
1334 * chunk size, and of the "after" data disks.
1335 * So when reverting we need to change it to
1336 * be a multiple of the new "after" data disks,
1337 * which is the old "before".
1338 * If it isn't already a multiple of 'before',
1339 * the only thing we could do would be
1340 * copy some block around on the disks, which
1341 * is easy to get wrong.
1342 * So we reject a revert-reshape unless the
1343 * alignment is good.
1345 if (__le32_to_cpu(sb
->level
) >= 4 &&
1346 __le32_to_cpu(sb
->level
) <= 6) {
1347 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1348 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1349 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1350 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1351 if (reshape_sectors
% reshape_chunk
) {
1352 pr_err("Reshape position is not suitably aligned.\n");
1353 pr_err("Try normal assembly and stop again\n");
1357 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1358 __le32_to_cpu(sb
->delta_disks
));
1359 if (sb
->delta_disks
== 0)
1360 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1362 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1364 temp
= sb
->new_layout
;
1365 sb
->new_layout
= sb
->layout
;
1368 temp
= sb
->new_chunk
;
1369 sb
->new_chunk
= sb
->chunksize
;
1370 sb
->chunksize
= temp
;
1372 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1373 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1374 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1375 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1376 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1380 } else if (strcmp(update
, "_reshape_progress")==0)
1381 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1382 else if (strcmp(update
, "writemostly")==0)
1383 sb
->devflags
|= WriteMostly1
;
1384 else if (strcmp(update
, "readwrite")==0)
1385 sb
->devflags
&= ~WriteMostly1
;
1386 else if (strcmp(update
, "failfast") == 0)
1387 sb
->devflags
|= FailFast1
;
1388 else if (strcmp(update
, "nofailfast") == 0)
1389 sb
->devflags
&= ~FailFast1
;
1393 sb
->sb_csum
= calc_sb_1_csum(sb
);
1394 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1395 cluster_release_dlmlock(lockid
);
1400 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1401 struct shape
*s
, char *name
, char *homehost
,
1402 int *uuid
, unsigned long long data_offset
)
1404 struct mdp_superblock_1
*sb
;
1409 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1410 pr_err("could not allocate superblock\n");
1413 memset(sb
, 0, SUPER1_SIZE
);
1417 /* zeroing superblock */
1421 spares
= info
->working_disks
- info
->active_disks
;
1422 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1423 pr_err("too many devices requested: %d+%d > %d\n",
1424 info
->raid_disks
, spares
, MAX_DEVS
);
1428 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1429 sb
->major_version
= __cpu_to_le32(1);
1430 sb
->feature_map
= 0;
1434 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1436 random_uuid(sb
->set_uuid
);;
1438 if (name
== NULL
|| *name
== 0) {
1439 sprintf(defname
, "%d", info
->md_minor
);
1443 strchr(name
, ':')== NULL
&&
1444 strlen(homehost
)+1+strlen(name
) < 32) {
1445 strcpy(sb
->set_name
, homehost
);
1446 strcat(sb
->set_name
, ":");
1447 strcat(sb
->set_name
, name
);
1449 strncpy(sb
->set_name
, name
, sizeof(sb
->set_name
));
1451 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1452 sb
->level
= __cpu_to_le32(info
->level
);
1453 sb
->layout
= __cpu_to_le32(info
->layout
);
1454 sb
->size
= __cpu_to_le64(s
->size
*2ULL);
1455 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1456 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1458 sb
->data_offset
= __cpu_to_le64(data_offset
);
1459 sb
->data_size
= __cpu_to_le64(0);
1460 sb
->super_offset
= __cpu_to_le64(0);
1461 sb
->recovery_offset
= __cpu_to_le64(0);
1463 sb
->utime
= sb
->ctime
;
1464 sb
->events
= __cpu_to_le64(1);
1465 if (info
->state
& (1<<MD_SB_CLEAN
))
1466 sb
->resync_offset
= MaxSector
;
1468 sb
->resync_offset
= 0;
1469 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1470 sbsize
= ROUND_UP(sbsize
, 512);
1471 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1473 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1481 long long data_offset
;
1482 mdu_disk_info_t disk
;
1483 struct devinfo
*next
;
1486 /* Add a device to the superblock being created */
1487 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1488 int fd
, char *devname
, unsigned long long data_offset
)
1490 struct mdp_superblock_1
*sb
= st
->sb
;
1491 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1492 struct devinfo
*di
, **dip
;
1493 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1497 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1498 rv
= cluster_get_dlmlock(&lockid
);
1500 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1501 cluster_release_dlmlock(lockid
);
1506 dk_state
= dk
->state
& ~(1<<MD_DISK_FAILFAST
);
1507 if ((dk_state
& 6) == 6) /* active, sync */
1508 *rp
= __cpu_to_le16(dk
->raid_disk
);
1509 else if (dk_state
& (1<<MD_DISK_JOURNAL
))
1510 *rp
= MD_DISK_ROLE_JOURNAL
;
1511 else if ((dk_state
& ~2) == 0) /* active or idle -> spare */
1512 *rp
= MD_DISK_ROLE_SPARE
;
1514 *rp
= MD_DISK_ROLE_FAULTY
;
1516 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1517 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1518 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1520 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1521 sb
->devflags
= 0; /* don't copy another disks flags */
1522 sb
->sb_csum
= calc_sb_1_csum(sb
);
1524 dip
= (struct devinfo
**)&st
->info
;
1526 dip
= &(*dip
)->next
;
1527 di
= xmalloc(sizeof(struct devinfo
));
1529 di
->devname
= devname
;
1531 di
->data_offset
= data_offset
;
1535 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1536 cluster_release_dlmlock(lockid
);
1542 static int locate_bitmap1(struct supertype
*st
, int fd
, int node_num
);
1544 static int store_super1(struct supertype
*st
, int fd
)
1546 struct mdp_superblock_1
*sb
= st
->sb
;
1547 unsigned long long sb_offset
;
1548 struct align_fd afd
;
1550 unsigned long long dsize
;
1551 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1554 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1555 rv
= cluster_get_dlmlock(&lockid
);
1557 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1558 cluster_release_dlmlock(lockid
);
1563 if (!get_dev_size(fd
, NULL
, &dsize
))
1574 * Calculate the position of the superblock.
1575 * It is always aligned to a 4K boundary and
1576 * depending on minor_version, it can be:
1577 * 0: At least 8K, but less than 12K, from end of device
1578 * 1: At start of device
1579 * 2: 4K from start of device.
1581 switch(st
->minor_version
) {
1585 sb_offset
&= ~(4*2-1);
1597 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1598 0 != __le64_to_cpu(sb
->super_offset
)
1600 pr_err("internal error - sb_offset is wrong\n");
1604 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1607 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1609 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1612 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1613 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1614 (((char*)sb
)+MAX_SB_SIZE
);
1615 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1616 locate_bitmap1(st
, fd
, 0);
1617 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1622 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1623 cluster_release_dlmlock(lockid
);
1628 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1630 static unsigned long choose_bm_space(unsigned long devsize
)
1632 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1633 * if bigger than 200Gig, save 128k
1634 * NOTE: result must be multiple of 4K else bad things happen
1635 * on 4K-sector devices.
1639 if (devsize
- 64*2 >= 200*1024*1024*2)
1641 if (devsize
- 4*2 > 8*1024*1024*2)
1646 static void free_super1(struct supertype
*st
);
1648 #define META_BLOCK_SIZE 4096
1649 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
1652 static int write_empty_r5l_meta_block(struct supertype
*st
, int fd
)
1654 struct r5l_meta_block
*mb
;
1655 struct mdp_superblock_1
*sb
= st
->sb
;
1656 struct align_fd afd
;
1661 if (posix_memalign((void**)&mb
, 4096, META_BLOCK_SIZE
) != 0) {
1662 pr_err("Could not allocate memory for the meta block.\n");
1666 memset(mb
, 0, META_BLOCK_SIZE
);
1668 mb
->magic
= __cpu_to_le32(R5LOG_MAGIC
);
1669 mb
->version
= R5LOG_VERSION
;
1670 mb
->meta_size
= __cpu_to_le32(sizeof(struct r5l_meta_block
));
1671 mb
->seq
= __cpu_to_le64(random32());
1672 mb
->position
= __cpu_to_le64(0);
1674 crc
= crc32c_le(0xffffffff, sb
->set_uuid
, sizeof(sb
->set_uuid
));
1675 crc
= crc32c_le(crc
, (void *)mb
, META_BLOCK_SIZE
);
1678 if (lseek64(fd
, (sb
->data_offset
) * 512, 0) < 0LL) {
1679 pr_err("cannot seek to offset of the meta block\n");
1683 if (awrite(&afd
, mb
, META_BLOCK_SIZE
) != META_BLOCK_SIZE
) {
1684 pr_err("failed to store write the meta block \n");
1697 static int write_init_super1(struct supertype
*st
)
1699 struct mdp_superblock_1
*sb
= st
->sb
;
1700 struct supertype
*refst
;
1702 unsigned long long bm_space
;
1704 unsigned long long dsize
, array_size
;
1705 unsigned long long sb_offset
;
1706 unsigned long long data_offset
;
1709 for (di
= st
->info
; di
; di
= di
->next
) {
1710 if (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))
1711 sb
->feature_map
|= MD_FEATURE_JOURNAL
;
1714 for (di
= st
->info
; di
; di
= di
->next
) {
1715 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1720 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1723 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1724 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1725 sb
->devflags
|= WriteMostly1
;
1727 sb
->devflags
&= ~WriteMostly1
;
1728 if (di
->disk
.state
& (1<<MD_DISK_FAILFAST
))
1729 sb
->devflags
|= FailFast1
;
1731 sb
->devflags
&= ~FailFast1
;
1733 random_uuid(sb
->device_uuid
);
1735 if (!(di
->disk
.state
& (1<<MD_DISK_JOURNAL
)))
1738 refst
= dup_super(st
);
1739 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1740 struct mdp_superblock_1
*refsb
= refst
->sb
;
1742 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1743 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1744 /* same array, so preserve events and
1746 sb
->events
= refsb
->events
;
1747 /* bugs in 2.6.17 and earlier mean the
1748 * dev_number chosen in Manage must be preserved
1750 if (get_linux_version() >= 2006018)
1751 sb
->dev_number
= refsb
->dev_number
;
1757 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1770 * Calculate the position of the superblock.
1771 * It is always aligned to a 4K boundary and
1772 * depending on minor_version, it can be:
1773 * 0: At least 8K, but less than 12K, from end of device
1774 * 1: At start of device
1775 * 2: 4K from start of device.
1776 * data_offset has already been set.
1778 array_size
= __le64_to_cpu(sb
->size
);
1780 /* work out how much space we left for a bitmap */
1781 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1782 bitmap_super_t
*bms
= (bitmap_super_t
*)
1783 (((char *)sb
) + MAX_SB_SIZE
);
1784 bm_space
= calc_bitmap_size(bms
, 4096) >> 9;
1785 bm_offset
= (long)__le32_to_cpu(sb
->bitmap_offset
);
1787 bm_space
= choose_bm_space(array_size
);
1791 data_offset
= di
->data_offset
;
1792 if (data_offset
== INVALID_SECTORS
)
1793 data_offset
= st
->data_offset
;
1794 switch(st
->minor_version
) {
1796 /* Add 8 sectors for bad block log */
1798 if (data_offset
== INVALID_SECTORS
)
1802 sb_offset
&= ~(4*2-1);
1803 sb
->data_offset
= __cpu_to_le64(data_offset
);
1804 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1805 if (sb_offset
< array_size
+ bm_space
)
1806 bm_space
= sb_offset
- array_size
;
1807 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1808 if (bm_space
>= 8) {
1809 sb
->bblog_size
= __cpu_to_le16(8);
1810 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1815 sb_offset
= st
->minor_version
== 2 ? 8 : 0;
1816 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1817 if (data_offset
== INVALID_SECTORS
)
1818 data_offset
= sb_offset
+ 16;
1820 sb
->data_offset
= __cpu_to_le64(data_offset
);
1821 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1822 if (data_offset
>= sb_offset
+bm_offset
+bm_space
+8) {
1823 sb
->bblog_size
= __cpu_to_le16(8);
1824 sb
->bblog_offset
= __cpu_to_le32(bm_offset
+
1826 } else if (data_offset
>= sb_offset
+ 16) {
1827 sb
->bblog_size
= __cpu_to_le16(8);
1828 /* '8' sectors for the bblog, and 'sb_offset'
1829 * because we want offset from superblock, not
1832 sb
->bblog_offset
= __cpu_to_le32(data_offset
-
1837 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1838 st
->minor_version
, di
->devname
);
1842 /* Disable badblock log on clusters, or when explicitly requested */
1843 if (st
->nodes
> 0 || conf_get_create_info()->bblist
== 0) {
1845 sb
->bblog_offset
= 0;
1848 sb
->sb_csum
= calc_sb_1_csum(sb
);
1849 rv
= store_super1(st
, di
->fd
);
1851 if (rv
== 0 && (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))) {
1852 rv
= write_empty_r5l_meta_block(st
, di
->fd
);
1857 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1858 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NodeNumUpdate
);
1866 pr_err("Failed to write metadata to %s\n",
1873 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1877 * 0 same, or first was empty, and second was copied
1878 * 1 second had wrong number
1880 * 3 wrong other info
1882 struct mdp_superblock_1
*first
= st
->sb
;
1883 struct mdp_superblock_1
*second
= tst
->sb
;
1885 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1887 if (second
->major_version
!= __cpu_to_le32(1))
1891 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1892 pr_err("could not allocate superblock\n");
1895 memcpy(first
, second
, SUPER1_SIZE
);
1899 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1902 if (first
->ctime
!= second
->ctime
||
1903 first
->level
!= second
->level
||
1904 first
->layout
!= second
->layout
||
1905 first
->size
!= second
->size
||
1906 first
->chunksize
!= second
->chunksize
||
1907 first
->raid_disks
!= second
->raid_disks
)
1912 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1914 unsigned long long dsize
;
1915 unsigned long long sb_offset
;
1916 struct mdp_superblock_1
*super
;
1918 struct bitmap_super_s
*bsb
;
1919 struct misc_dev_info
*misc
;
1920 struct align_fd afd
;
1926 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1928 struct supertype tst
;
1929 __u64 bestctime
= 0;
1930 /* guess... choose latest ctime */
1931 memset(&tst
, 0, sizeof(tst
));
1933 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1934 switch(load_super1(&tst
, fd
, devname
)) {
1935 case 0: super
= tst
.sb
;
1936 if (bestvers
== -1 ||
1937 bestctime
< __le64_to_cpu(super
->ctime
)) {
1938 bestvers
= tst
.minor_version
;
1939 bestctime
= __le64_to_cpu(super
->ctime
);
1944 case 1: return 1; /*bad device */
1945 case 2: break; /* bad, try next */
1948 if (bestvers
!= -1) {
1950 tst
.minor_version
= bestvers
;
1952 tst
.max_devs
= MAX_DEVS
;
1953 rv
= load_super1(&tst
, fd
, devname
);
1960 if (!get_dev_size(fd
, devname
, &dsize
))
1966 pr_err("%s is too small for md: size is %llu sectors.\n",
1972 * Calculate the position of the superblock.
1973 * It is always aligned to a 4K boundary and
1974 * depending on minor_version, it can be:
1975 * 0: At least 8K, but less than 12K, from end of device
1976 * 1: At start of device
1977 * 2: 4K from start of device.
1979 switch(st
->minor_version
) {
1983 sb_offset
&= ~(4*2-1);
1995 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1997 pr_err("Cannot seek to superblock on %s: %s\n",
1998 devname
, strerror(errno
));
2002 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
2003 pr_err("could not allocate superblock\n");
2007 memset(super
, 0, SUPER1_SIZE
);
2009 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
2011 pr_err("Cannot read superblock on %s\n",
2017 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
2019 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2020 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
2025 if (__le32_to_cpu(super
->major_version
) != 1) {
2027 pr_err("Cannot interpret superblock on %s - version is %d\n",
2028 devname
, __le32_to_cpu(super
->major_version
));
2032 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
2034 pr_err("No superblock found on %s (super_offset is wrong)\n",
2041 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2043 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
2044 misc
->device_size
= dsize
;
2045 if (st
->data_offset
== INVALID_SECTORS
)
2046 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
2048 /* Now check on the bitmap superblock */
2049 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
2051 /* Read the bitmap superblock and make sure it looks
2052 * valid. If it doesn't clear the bit. An --assemble --force
2053 * should get that written out.
2055 locate_bitmap1(st
, fd
, 0);
2056 if (aread(&afd
, bsb
, 512) != 512)
2059 uuid_from_super1(st
, uuid
);
2060 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
2061 memcmp(bsb
->uuid
, uuid
, 16) != 0)
2066 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
2067 & ~MD_FEATURE_BITMAP_OFFSET
);
2071 static struct supertype
*match_metadata_desc1(char *arg
)
2073 struct supertype
*st
= xcalloc(1, sizeof(*st
));
2075 st
->container_devnm
[0] = 0;
2077 st
->max_devs
= MAX_DEVS
;
2079 st
->data_offset
= INVALID_SECTORS
;
2080 /* leading zeros can be safely ignored. --detail generates them. */
2083 if (strcmp(arg
, "1.0") == 0 ||
2084 strcmp(arg
, "1.00") == 0) {
2085 st
->minor_version
= 0;
2088 if (strcmp(arg
, "1.1") == 0 ||
2089 strcmp(arg
, "1.01") == 0
2091 st
->minor_version
= 1;
2094 if (strcmp(arg
, "1.2") == 0 ||
2095 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2096 strcmp(arg
, "default") == 0 ||
2097 #endif /* DEFAULT_OLD_METADATA */
2098 strcmp(arg
, "1.02") == 0) {
2099 st
->minor_version
= 2;
2102 if (strcmp(arg
, "1") == 0 ||
2103 strcmp(arg
, "default") == 0) {
2104 st
->minor_version
= -1;
2112 /* find available size on device with this devsize, using
2113 * superblock type st, and reserving 'reserve' sectors for
2116 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
2117 unsigned long long data_offset
)
2119 struct mdp_superblock_1
*super
= st
->sb
;
2126 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
2127 /* hot-add. allow for actual size of bitmap */
2128 struct bitmap_super_s
*bsb
;
2129 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2130 bmspace
= calc_bitmap_size(bsb
, 4096) >> 9;
2133 /* Allow space for bad block log */
2134 if (super
->bblog_size
)
2135 bbspace
= __le16_to_cpu(super
->bblog_size
);
2137 if (st
->minor_version
< 0)
2138 /* not specified, so time to set default */
2139 st
->minor_version
= 2;
2141 if (data_offset
== INVALID_SECTORS
)
2142 data_offset
= st
->data_offset
;
2144 if (data_offset
!= INVALID_SECTORS
)
2145 switch(st
->minor_version
) {
2147 return devsize
- data_offset
- 8*2 - bbspace
;
2150 return devsize
- data_offset
;
2157 switch(st
->minor_version
) {
2160 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2162 /* at start, 4K for superblock and possible bitmap */
2163 return devsize
- 4*2 - bbspace
;
2165 /* 4k from start, 4K for superblock and possible bitmap */
2166 return devsize
- (4+4)*2 - bbspace
;
2172 add_internal_bitmap1(struct supertype
*st
,
2173 int *chunkp
, int delay
, int write_behind
,
2174 unsigned long long size
,
2175 int may_change
, int major
)
2178 * If not may_change, then this is a 'Grow' without sysfs support for
2179 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2180 * If may_change, then this is create or a Grow with sysfs syupport,
2181 * and we can put the bitmap wherever we like.
2183 * size is in sectors, chunk is in bytes !!!
2186 unsigned long long bits
;
2187 unsigned long long max_bits
;
2188 unsigned long long min_chunk
;
2190 long bbl_offset
, bbl_size
;
2191 unsigned long long chunk
= *chunkp
;
2195 struct mdp_superblock_1
*sb
= st
->sb
;
2196 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2199 if (__le64_to_cpu(sb
->data_size
) == 0)
2200 /* Must be creating the array, else data_size would be non-zero */
2202 switch(st
->minor_version
) {
2204 /* either 3K after the superblock (when hot-add),
2205 * or some amount of space before.
2208 /* We are creating array, so we *know* how much room has
2213 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2215 room
= __le64_to_cpu(sb
->super_offset
)
2216 - __le64_to_cpu(sb
->data_offset
)
2217 - __le64_to_cpu(sb
->data_size
);
2218 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2221 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2222 if (bbl_size
< -bbl_offset
)
2223 bbl_size
= -bbl_offset
;
2225 if (!may_change
|| (room
< 3*2 &&
2226 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2231 offset
= 0; /* means movable offset */
2236 case 2: /* between superblock and data */
2240 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2242 room
= __le64_to_cpu(sb
->data_offset
)
2243 - __le64_to_cpu(sb
->super_offset
);
2244 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2246 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2251 room
-= 2; /* Leave 1K for superblock */
2255 room
-= 4*2; /* leave 4K for superblock */
2265 if (chunk
== UnSet
&& room
> 128*2)
2266 /* Limit to 128K of bitmap when chunk size not requested */
2270 /* No room for a bitmap */
2273 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2275 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2276 bits
= (size
*512)/min_chunk
+1;
2277 while (bits
> max_bits
) {
2281 if (chunk
== UnSet
) {
2282 /* For practical purpose, 64Meg is a good
2283 * default chunk size for internal bitmaps.
2286 if (chunk
< 64*1024*1024)
2287 chunk
= 64*1024*1024;
2288 } else if (chunk
< min_chunk
)
2289 return -EINVAL
; /* chunk size too small */
2290 if (chunk
== 0) /* rounding problem */
2294 /* start bitmap on a 4K boundary with enough space for
2297 bits
= (size
*512) / chunk
+ 1;
2298 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2299 room
*= 8; /* convert 4K blocks to sectors */
2300 offset
= -room
- bbl_size
;
2303 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2305 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2306 | MD_FEATURE_BITMAP_OFFSET
);
2307 memset(bms
, 0, sizeof(*bms
));
2308 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2309 bms
->version
= __cpu_to_le32(major
);
2310 uuid_from_super1(st
, uuid
);
2311 memcpy(bms
->uuid
, uuid
, 16);
2312 bms
->chunksize
= __cpu_to_le32(chunk
);
2313 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2314 bms
->sync_size
= __cpu_to_le64(size
);
2315 bms
->write_behind
= __cpu_to_le32(write_behind
);
2316 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2318 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2319 | MD_FEATURE_BITMAP_VERSIONED
);
2320 if (st
->cluster_name
) {
2321 len
= sizeof(bms
->cluster_name
);
2322 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, len
);
2323 bms
->cluster_name
[len
- 1] = '\0';
2330 static int locate_bitmap1(struct supertype
*st
, int fd
, int node_num
)
2332 unsigned long long offset
;
2333 struct mdp_superblock_1
*sb
;
2338 if (st
->ss
->load_super(st
, fd
, NULL
))
2339 return -1; /* no error I hope... */
2344 if ((__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
2348 offset
= __le64_to_cpu(sb
->super_offset
);
2349 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
) * (node_num
+ 1);
2352 lseek64(fd
, offset
<<9, 0);
2356 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2358 struct mdp_superblock_1
*sb
= st
->sb
;
2359 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2362 int towrite
, n
, len
;
2363 struct align_fd afd
;
2365 unsigned long long total_bm_space
, bm_space_per_node
;
2369 /* update cluster name */
2370 if (st
->cluster_name
) {
2371 len
= sizeof(bms
->cluster_name
);
2372 memset((char *)bms
->cluster_name
, 0, len
);
2373 strncpy((char *)bms
->cluster_name
,
2374 st
->cluster_name
, len
);
2375 bms
->cluster_name
[len
- 1] = '\0';
2379 /* cluster md only supports superblock 1.2 now */
2380 if (st
->minor_version
!= 2 && bms
->version
== BITMAP_MAJOR_CLUSTERED
) {
2381 pr_err("Warning: cluster md only works with superblock 1.2\n");
2385 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
) {
2386 if (__cpu_to_le32(st
->nodes
) < bms
->nodes
) {
2387 /* Since the nodes num is not increased, no need to check the space
2388 * is enough or not, just update bms->nodes */
2389 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2393 /* no need to change bms->nodes for other bitmap types */
2395 pr_err("Warning: --nodes option is only suitable for clustered bitmap\n");
2399 /* Each node has an independent bitmap, it is necessary to calculate the
2400 * space is enough or not, first get how many bytes for the total bitmap */
2401 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2403 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2404 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2406 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2407 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2408 total_bm_space
/ bm_space_per_node
);
2412 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2421 locate_bitmap1(st
, fd
, 0);
2423 if (posix_memalign(&buf
, 4096, 4096))
2427 /* Only the bitmap[0] should resync
2428 * whole device on initial assembly
2431 memset(buf
, 0x00, 4096);
2433 memset(buf
, 0xff, 4096);
2434 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2437 * use 4096 boundary if bitmap_offset is aligned
2438 * with 8 sectors, then it should compatible with
2441 if (__le32_to_cpu(sb
->bitmap_offset
) & 7)
2442 towrite
= calc_bitmap_size(bms
, 512);
2444 towrite
= calc_bitmap_size(bms
, 4096);
2445 while (towrite
> 0) {
2449 n
= awrite(&afd
, buf
, n
);
2455 memset(buf
, 0x00, 4096);
2457 memset(buf
, 0xff, 4096);
2464 } while (++i
< __le32_to_cpu(bms
->nodes
));
2470 static void free_super1(struct supertype
*st
)
2476 struct devinfo
*di
= st
->info
;
2477 st
->info
= di
->next
;
2486 static int validate_geometry1(struct supertype
*st
, int level
,
2487 int layout
, int raiddisks
,
2488 int *chunk
, unsigned long long size
,
2489 unsigned long long data_offset
,
2490 char *subdev
, unsigned long long *freesize
,
2491 int consistency_policy
, int verbose
)
2493 unsigned long long ldsize
, devsize
;
2495 unsigned long long headroom
;
2498 if (level
== LEVEL_CONTAINER
) {
2500 pr_err("1.x metadata does not support containers\n");
2503 if (*chunk
== UnSet
)
2504 *chunk
= DEFAULT_CHUNK
;
2509 if (st
->minor_version
< 0)
2510 /* not specified, so time to set default */
2511 st
->minor_version
= 2;
2513 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2516 pr_err("super1.x cannot open %s: %s\n",
2517 subdev
, strerror(errno
));
2521 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2527 devsize
= ldsize
>> 9;
2533 /* creating: allow suitable space for bitmap */
2534 bmspace
= choose_bm_space(devsize
);
2536 if (data_offset
== INVALID_SECTORS
)
2537 data_offset
= st
->data_offset
;
2538 if (data_offset
== INVALID_SECTORS
)
2539 switch (st
->minor_version
) {
2545 /* Choose data offset appropriate for this device
2546 * and use as default for whole array.
2547 * The data_offset must allow for bitmap space
2548 * and base metadata, should allow for some headroom
2549 * for reshape, and should be rounded to multiple
2551 * Headroom is limited to 128M, but aim for about 0.1%
2553 headroom
= 128*1024*2;
2554 while ((headroom
<< 10) > devsize
&&
2556 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2558 data_offset
= 12*2 + bmspace
+ headroom
;
2559 #define ONE_MEG (2*1024)
2560 if (data_offset
> ONE_MEG
)
2561 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2564 if (st
->data_offset
== INVALID_SECTORS
)
2565 st
->data_offset
= data_offset
;
2566 switch(st
->minor_version
) {
2567 case 0: /* metadata at end. Round down and subtract space to reserve */
2568 devsize
= (devsize
& ~(4ULL*2-1));
2569 /* space for metadata, bblog, bitmap */
2570 devsize
-= 8*2 + 8 + bmspace
;
2574 devsize
-= data_offset
;
2577 *freesize
= devsize
;
2580 #endif /* MDASSEMBLE */
2582 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2584 /* Create a v1.0 superblock based on 'info'*/
2586 struct mdp_superblock_1
*sb
;
2588 unsigned long long offset
;
2590 if (posix_memalign(&ret
, 4096, 1024) != 0)
2593 memset(ret
, 0, 1024);
2594 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2595 sb
->major_version
= __cpu_to_le32(1);
2597 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2598 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2599 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2600 sb
->level
= __cpu_to_le32(info
->array
.level
);
2601 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2602 sb
->size
= __cpu_to_le64(info
->component_size
);
2603 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2604 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2605 if (info
->array
.level
> 0)
2606 sb
->data_size
= sb
->size
;
2608 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2609 sb
->resync_offset
= MaxSector
;
2610 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2611 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2612 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2614 offset
= st
->devsize
/512 - 8*2;
2616 sb
->super_offset
= __cpu_to_le64(offset
);
2617 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2619 random_uuid(sb
->device_uuid
);
2621 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2622 int state
= sb0
->disks
[i
].state
;
2623 sb
->dev_roles
[i
] = MD_DISK_ROLE_SPARE
;
2624 if ((state
& (1<<MD_DISK_SYNC
)) &&
2625 !(state
& (1<<MD_DISK_FAULTY
)))
2626 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2628 sb
->sb_csum
= calc_sb_1_csum(sb
);
2632 struct superswitch super1
= {
2634 .examine_super
= examine_super1
,
2635 .brief_examine_super
= brief_examine_super1
,
2636 .export_examine_super
= export_examine_super1
,
2637 .detail_super
= detail_super1
,
2638 .brief_detail_super
= brief_detail_super1
,
2639 .export_detail_super
= export_detail_super1
,
2640 .write_init_super
= write_init_super1
,
2641 .validate_geometry
= validate_geometry1
,
2642 .add_to_super
= add_to_super1
,
2643 .examine_badblocks
= examine_badblocks_super1
,
2644 .copy_metadata
= copy_metadata1
,
2646 .match_home
= match_home1
,
2647 .uuid_from_super
= uuid_from_super1
,
2648 .getinfo_super
= getinfo_super1
,
2649 .container_content
= container_content1
,
2650 .update_super
= update_super1
,
2651 .init_super
= init_super1
,
2652 .store_super
= store_super1
,
2653 .compare_super
= compare_super1
,
2654 .load_super
= load_super1
,
2655 .match_metadata_desc
= match_metadata_desc1
,
2656 .avail_size
= avail_size1
,
2657 .add_internal_bitmap
= add_internal_bitmap1
,
2658 .locate_bitmap
= locate_bitmap1
,
2659 .write_bitmap
= write_bitmap1
,
2660 .free_super
= free_super1
,
2661 #if __BYTE_ORDER == BIG_ENDIAN