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 /* bad block log. If there are any bad blocks the feature flag is set.
81 * if offset and size are non-zero, that space is reserved and available.
83 __u8 bblog_shift
; /* shift from sectors to block size for badblocklist */
84 __u16 bblog_size
; /* number of sectors reserved for badblocklist */
85 __u32 bblog_offset
; /* sector offset from superblock to bblog, signed */
87 /* array state information - 64 bytes */
88 __u64 utime
; /* 40 bits second, 24 btes microseconds */
89 __u64 events
; /* incremented when superblock updated */
90 __u64 resync_offset
; /* data before this offset (from data_offset) known to be in sync */
91 __u32 sb_csum
; /* checksum upto dev_roles[max_dev] */
92 __u32 max_dev
; /* size of dev_roles[] array to consider */
93 __u8 pad3
[64-32]; /* set to 0 when writing */
95 /* device state information. Indexed by dev_number.
97 * Note there are no per-device state flags. State information is rolled
98 * into the 'roles' value. If a device is spare or faulty, then it doesn't
99 * have a meaningful role.
101 __u16 dev_roles
[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
104 #define MAX_SB_SIZE 4096
105 /* bitmap super size is 256, but we round up to a sector for alignment */
106 #define BM_SUPER_SIZE 512
107 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
108 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
109 + sizeof(struct misc_dev_info))
111 struct misc_dev_info
{
115 /* feature_map bits */
116 #define MD_FEATURE_BITMAP_OFFSET 1
117 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
120 #define MD_FEATURE_RESHAPE_ACTIVE 4
121 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
122 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
123 * active device with same 'role'.
124 * 'recovery_offset' is also set.
126 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
127 * of devices, but is going
130 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
131 #define MD_FEATURE_BITMAP_VERSIONED 256 /* bitmap version number checked properly */
132 #define MD_FEATURE_JOURNAL 512 /* support write journal */
133 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
134 |MD_FEATURE_RECOVERY_OFFSET \
135 |MD_FEATURE_RESHAPE_ACTIVE \
136 |MD_FEATURE_BAD_BLOCKS \
137 |MD_FEATURE_REPLACEMENT \
138 |MD_FEATURE_RESHAPE_BACKWARDS \
139 |MD_FEATURE_NEW_OFFSET \
140 |MD_FEATURE_BITMAP_VERSIONED \
141 |MD_FEATURE_JOURNAL \
145 static int role_from_sb(struct mdp_superblock_1
*sb
)
150 d
= __le32_to_cpu(sb
->dev_number
);
151 if (d
< __le32_to_cpu(sb
->max_dev
))
152 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
154 role
= MD_DISK_ROLE_SPARE
;
159 /* return how many bytes are needed for bitmap, for cluster-md each node
160 * should have it's own bitmap */
161 static unsigned int calc_bitmap_size(bitmap_super_t
*bms
, unsigned int boundary
)
163 unsigned long long bits
, bytes
;
165 bits
= bitmap_bits(__le64_to_cpu(bms
->sync_size
),
166 __le32_to_cpu(bms
->chunksize
));
167 bytes
= (bits
+7) >> 3;
168 bytes
+= sizeof(bitmap_super_t
);
169 bytes
= ROUND_UP(bytes
, boundary
);
174 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
176 unsigned int disk_csum
, csum
;
177 unsigned long long newcsum
;
178 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
179 unsigned int *isuper
= (unsigned int*)sb
;
181 /* make sure I can count... */
182 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
183 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
184 sizeof(struct mdp_superblock_1
) != 256) {
185 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
188 disk_csum
= sb
->sb_csum
;
191 for (; size
>=4; size
-= 4 ) {
192 newcsum
+= __le32_to_cpu(*isuper
);
197 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
199 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
200 sb
->sb_csum
= disk_csum
;
201 return __cpu_to_le32(csum
);
205 * Information related to file descriptor used for aligned reads/writes.
206 * Cache the block size.
213 static void init_afd(struct align_fd
*afd
, int fd
)
216 if (!get_dev_sector_size(afd
->fd
, NULL
, (unsigned int *)&afd
->blk_sz
))
220 static char abuf
[4096+4096];
221 static int aread(struct align_fd
*afd
, void *buf
, int len
)
224 * On devices with a 4K sector size, we need to read
225 * the full sector and copy relevant bits into
234 if (!bsize
|| bsize
> 4096 || len
> 4096) {
236 fprintf(stderr
, "WARNING - aread() called with invalid block size\n");
239 b
= ROUND_UP_PTR((char *)abuf
, 4096);
241 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
243 n
= read(afd
->fd
, b
, iosize
);
246 lseek(afd
->fd
, len
- n
, 1);
253 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
256 * On devices with a 4K sector size, we need to write
257 * the full sector. We pre-read if the sector is larger
259 * The address must be sector-aligned.
266 if (!bsize
|| bsize
> 4096 || len
> 4096) {
268 fprintf(stderr
, "WARNING - awrite() called with invalid block size\n");
271 b
= ROUND_UP_PTR((char *)abuf
, 4096);
273 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
277 n
= read(afd
->fd
, b
, iosize
);
280 lseek(afd
->fd
, -n
, 1);
284 n
= write(afd
->fd
, b
, iosize
);
287 lseek(afd
->fd
, len
- n
, 1);
292 static void examine_super1(struct supertype
*st
, char *homehost
)
294 struct mdp_superblock_1
*sb
= st
->sb
;
295 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
302 int l
= homehost
? strlen(homehost
) : 0;
304 unsigned long long sb_offset
;
307 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
308 printf(" Version : 1");
309 sb_offset
= __le64_to_cpu(sb
->super_offset
);
312 else if (sb_offset
<= 8)
316 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
317 printf(" Array UUID : ");
318 for (i
=0; i
<16; i
++) {
319 if ((i
&3)==0 && i
!= 0) printf(":");
320 printf("%02x", sb
->set_uuid
[i
]);
323 printf(" Name : %.32s", sb
->set_name
);
324 if (l
> 0 && l
< 32 &&
325 sb
->set_name
[l
] == ':' &&
326 strncmp(sb
->set_name
, homehost
, l
) == 0)
327 printf(" (local to host %s)", homehost
);
329 if (bms
->nodes
> 0 && (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
330 printf(" Cluster Name : %-64s\n", bms
->cluster_name
);
331 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
332 printf(" Creation Time : %.24s\n", ctime(&atime
));
333 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
334 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
335 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
337 printf(" Avail Dev Size : %llu%s\n",
338 (unsigned long long)__le64_to_cpu(sb
->data_size
),
339 human_size(__le64_to_cpu(sb
->data_size
)<<9));
340 if (__le32_to_cpu(sb
->level
) > 0) {
341 int ddsks
= 0, ddsks_denom
= 1;
342 switch(__le32_to_cpu(sb
->level
)) {
343 case 1: ddsks
=1;break;
345 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
346 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
348 layout
= __le32_to_cpu(sb
->layout
);
349 ddsks
= __le32_to_cpu(sb
->raid_disks
);
350 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
353 long long asize
= __le64_to_cpu(sb
->size
);
354 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
355 printf(" Array Size : %llu%s\n",
356 asize
>> 10, human_size(asize
));
358 if (sb
->size
!= sb
->data_size
)
359 printf(" Used Dev Size : %llu%s\n",
360 (unsigned long long)__le64_to_cpu(sb
->size
),
361 human_size(__le64_to_cpu(sb
->size
)<<9));
364 printf(" Data Offset : %llu sectors\n",
365 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
366 if (sb
->new_offset
&&
367 (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
)) {
368 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
369 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
370 printf(" New Offset : %llu sectors\n", offset
);
372 printf(" Super Offset : %llu sectors\n",
373 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
374 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
375 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
377 st
->ss
->getinfo_super(st
, &info
, NULL
);
378 if (info
.space_after
!= 1 &&
379 !(__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
380 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
381 info
.space_before
, info
.space_after
);
383 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
384 printf(" Device UUID : ");
385 for (i
=0; i
<16; i
++) {
386 if ((i
&3)==0 && i
!= 0) printf(":");
387 printf("%02x", sb
->device_uuid
[i
]);
391 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
392 printf("Internal Bitmap : %ld sectors from superblock\n",
393 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
395 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
396 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
397 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
398 if (__le32_to_cpu(sb
->delta_disks
)) {
399 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
400 printf(" (%d->%d)\n",
401 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
402 __le32_to_cpu(sb
->raid_disks
));
403 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
404 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
406 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
407 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
408 printf(" New Level : %s\n", c
?c
:"-unknown-");
410 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
411 if (__le32_to_cpu(sb
->level
) == 5) {
412 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
413 printf(" New Layout : %s\n", c
?c
:"-unknown-");
415 if (__le32_to_cpu(sb
->level
) == 6) {
416 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
417 printf(" New Layout : %s\n", c
?c
:"-unknown-");
419 if (__le32_to_cpu(sb
->level
) == 10) {
420 printf(" New Layout :");
421 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
425 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
426 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
431 if (sb
->devflags
& WriteMostly1
)
432 printf(" write-mostly");
436 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
437 printf(" Update Time : %.24s\n", ctime(&atime
));
439 if (sb
->bblog_size
&& sb
->bblog_offset
) {
440 printf(" Bad Block Log : %d entries available at offset %ld sectors",
441 __le16_to_cpu(sb
->bblog_size
)*512/8,
442 (long)(int32_t)__le32_to_cpu(sb
->bblog_offset
));
443 if (sb
->feature_map
&
444 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
445 printf(" - bad blocks present.");
449 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
450 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
452 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
453 __le32_to_cpu(calc_sb_1_csum(sb
)));
454 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
456 if (__le32_to_cpu(sb
->level
) == 5) {
457 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
458 printf(" Layout : %s\n", c
?c
:"-unknown-");
460 if (__le32_to_cpu(sb
->level
) == 6) {
461 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
462 printf(" Layout : %s\n", c
?c
:"-unknown-");
464 if (__le32_to_cpu(sb
->level
) == 10) {
465 int lo
= __le32_to_cpu(sb
->layout
);
467 print_r10_layout(lo
);
470 switch(__le32_to_cpu(sb
->level
)) {
476 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
479 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
485 /* This turns out to just be confusing */
486 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
487 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
488 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != MD_DISK_ROLE_SPARE
)
490 for (d
=0; d
< i
; d
++) {
491 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
493 if (role
== MD_DISK_ROLE_SPARE
) printf("empty");
494 else if(role
== MD_DISK_ROLE_FAULTY
) printf("failed");
495 else printf("%d", role
);
499 printf(" Device Role : ");
500 role
= role_from_sb(sb
);
501 if (role
>= MD_DISK_ROLE_FAULTY
)
503 else if (role
== MD_DISK_ROLE_JOURNAL
)
505 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
506 printf("Replacement device %d\n", role
);
508 printf("Active device %d\n", role
);
510 printf(" Array State : ");
511 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
514 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
515 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
529 /* This is confusing too */
531 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
532 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
533 if (role
== MD_DISK_ROLE_FAULTY
)
536 if (faulty
) printf(" %d failed", faulty
);
538 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
542 static void brief_examine_super1(struct supertype
*st
, int verbose
)
544 struct mdp_superblock_1
*sb
= st
->sb
;
546 unsigned long long sb_offset
;
548 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
550 nm
= strchr(sb
->set_name
, ':');
553 else if (sb
->set_name
[0])
565 printf(" level=%s", c
);
566 sb_offset
= __le64_to_cpu(sb
->super_offset
);
568 printf(" metadata=1.1 ");
569 else if (sb_offset
<= 8)
570 printf(" metadata=1.2 ");
572 printf(" metadata=1.0 ");
574 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
576 for (i
=0; i
<16; i
++) {
577 if ((i
&3)==0 && i
!= 0) printf(":");
578 printf("%02x", sb
->set_uuid
[i
]);
580 if (sb
->set_name
[0]) {
582 print_quoted(sb
->set_name
);
587 static void export_examine_super1(struct supertype
*st
)
589 struct mdp_superblock_1
*sb
= st
->sb
;
594 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
595 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
597 if (sb
->set_name
[i
] == '\n' ||
598 sb
->set_name
[i
] == '\0') {
603 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
604 if (__le32_to_cpu(sb
->level
) > 0) {
605 int ddsks
= 0, ddsks_denom
= 1;
606 switch(__le32_to_cpu(sb
->level
)) {
607 case 1: ddsks
=1;break;
609 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
610 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
612 layout
= __le32_to_cpu(sb
->layout
);
613 ddsks
= __le32_to_cpu(sb
->raid_disks
);
614 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
617 long long asize
= __le64_to_cpu(sb
->size
);
618 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
619 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
623 for (i
=0; i
<16; i
++) {
624 if ((i
&3)==0 && i
!= 0) printf(":");
625 printf("%02x", sb
->set_uuid
[i
]);
628 printf("MD_UPDATE_TIME=%llu\n",
629 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
630 printf("MD_DEV_UUID=");
631 for (i
=0; i
<16; i
++) {
632 if ((i
&3)==0 && i
!= 0) printf(":");
633 printf("%02x", sb
->device_uuid
[i
]);
636 printf("MD_EVENTS=%llu\n",
637 (unsigned long long)__le64_to_cpu(sb
->events
));
640 static int copy_metadata1(struct supertype
*st
, int from
, int to
)
642 /* Read superblock. If it looks good, write it out.
643 * Then if a bitmap is present, copy that.
644 * And if a bad-block-list is present, copy that too.
647 unsigned long long dsize
, sb_offset
;
648 const int bufsize
= 4*1024;
649 struct mdp_superblock_1 super
, *sb
;
651 if (posix_memalign(&buf
, 4096, bufsize
) != 0)
654 if (!get_dev_size(from
, NULL
, &dsize
))
660 switch(st
->minor_version
) {
664 sb_offset
&= ~(4*2-1);
676 if (lseek64(from
, sb_offset
<< 9, 0) < 0LL)
678 if (read(from
, buf
, bufsize
) != bufsize
)
682 super
= *sb
; // save most of sb for when we reuse buf
684 if (__le32_to_cpu(super
.magic
) != MD_SB_MAGIC
||
685 __le32_to_cpu(super
.major_version
) != 1 ||
686 __le64_to_cpu(super
.super_offset
) != sb_offset
||
687 calc_sb_1_csum(sb
) != super
.sb_csum
)
690 if (lseek64(to
, sb_offset
<< 9, 0) < 0LL)
692 if (write(to
, buf
, bufsize
) != bufsize
)
695 if (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
)) {
696 unsigned long long bitmap_offset
= sb_offset
;
697 int bytes
= 4096; // just an estimate.
699 struct align_fd afrom
, ato
;
701 init_afd(&afrom
, from
);
704 bitmap_offset
+= (int32_t)__le32_to_cpu(super
.bitmap_offset
);
706 if (lseek64(from
, bitmap_offset
<<9, 0) < 0)
708 if (lseek64(to
, bitmap_offset
<<9, 0) < 0)
711 for (written
= 0; written
< bytes
; ) {
712 int n
= bytes
- written
;
715 if (aread(&afrom
, buf
, n
) != n
)
718 /* have the header, can calculate
719 * correct bitmap bytes */
722 bytes
= calc_bitmap_size(bms
, 512);
726 if (awrite(&ato
, buf
, n
) != n
)
732 if (super
.bblog_size
!= 0 &&
733 __le16_to_cpu(super
.bblog_size
) <= 100 &&
734 super
.bblog_offset
!= 0 &&
735 (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BAD_BLOCKS
))) {
736 /* There is a bad block log */
737 unsigned long long bb_offset
= sb_offset
;
738 int bytes
= __le16_to_cpu(super
.bblog_size
) * 512;
740 struct align_fd afrom
, ato
;
742 init_afd(&afrom
, from
);
745 bb_offset
+= (int32_t)__le32_to_cpu(super
.bblog_offset
);
747 if (lseek64(from
, bb_offset
<<9, 0) < 0)
749 if (lseek64(to
, bb_offset
<<9, 0) < 0)
752 for (written
= 0; written
< bytes
; ) {
753 int n
= bytes
- written
;
756 if (aread(&afrom
, buf
, n
) != n
)
759 if (awrite(&ato
, buf
, n
) != n
)
773 static void detail_super1(struct supertype
*st
, char *homehost
)
775 struct mdp_superblock_1
*sb
= st
->sb
;
776 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
778 int l
= homehost
? strlen(homehost
) : 0;
780 printf(" Name : %.32s", sb
->set_name
);
781 if (l
> 0 && l
< 32 &&
782 sb
->set_name
[l
] == ':' &&
783 strncmp(sb
->set_name
, homehost
, l
) == 0)
784 printf(" (local to host %s)", homehost
);
785 if (bms
->nodes
> 0 && (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
786 printf("\n Cluster Name : %-64s", bms
->cluster_name
);
787 printf("\n UUID : ");
788 for (i
=0; i
<16; i
++) {
789 if ((i
&3)==0 && i
!= 0) printf(":");
790 printf("%02x", sb
->set_uuid
[i
]);
792 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb
->events
));
795 static void brief_detail_super1(struct supertype
*st
)
797 struct mdp_superblock_1
*sb
= st
->sb
;
800 if (sb
->set_name
[0]) {
802 print_quoted(sb
->set_name
);
805 for (i
=0; i
<16; i
++) {
806 if ((i
&3)==0 && i
!= 0) printf(":");
807 printf("%02x", sb
->set_uuid
[i
]);
811 static void export_detail_super1(struct supertype
*st
)
813 struct mdp_superblock_1
*sb
= st
->sb
;
818 if (sb
->set_name
[i
] == '\n' ||
819 sb
->set_name
[i
] == '\0') {
824 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
827 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
829 struct mdp_superblock_1
*sb
= st
->sb
;
830 unsigned long long offset
;
835 if (!sb
->bblog_size
|| __le16_to_cpu(sb
->bblog_size
) > 100
836 || !sb
->bblog_offset
){
837 printf("No bad-blocks list configured on %s\n", devname
);
840 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
842 printf("Bad-blocks list is empty in %s\n", devname
);
846 size
= __le16_to_cpu(sb
->bblog_size
)* 512;
847 if (posix_memalign((void**)&bbl
, 4096, size
) != 0) {
848 pr_err("could not allocate badblocks list\n");
851 offset
= __le64_to_cpu(sb
->super_offset
) +
852 (int)__le32_to_cpu(sb
->bblog_offset
);
854 if (lseek64(fd
, offset
, 0) < 0) {
855 pr_err("Cannot seek to bad-blocks list\n");
858 if (read(fd
, bbl
, size
) != size
) {
859 pr_err("Cannot read bad-blocks list\n");
862 /* 64bits per entry. 10 bits is block-count, 54 bits is block
863 * offset. Blocks are sectors unless bblog->shift makes them bigger
866 printf("Bad-blocks on %s:\n", devname
);
867 for (i
= 0; i
< size
/8; i
++, bbp
++) {
868 __u64 bb
= __le64_to_cpu(*bbp
);
869 int count
= bb
& 0x3ff;
870 unsigned long long sector
= bb
>> 10;
875 sector
<<= sb
->bblog_shift
;
876 count
<<= sb
->bblog_shift
;
878 printf("%20llu for %d sectors\n", sector
, count
);
885 static int match_home1(struct supertype
*st
, char *homehost
)
887 struct mdp_superblock_1
*sb
= st
->sb
;
888 int l
= homehost
? strlen(homehost
) : 0;
890 return (l
> 0 && l
< 32 &&
891 sb
->set_name
[l
] == ':' &&
892 strncmp(sb
->set_name
, homehost
, l
) == 0);
895 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
897 struct mdp_superblock_1
*super
= st
->sb
;
898 char *cuuid
= (char*)uuid
;
901 cuuid
[i
] = super
->set_uuid
[i
];
904 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
906 struct mdp_superblock_1
*sb
= st
->sb
;
907 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
908 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
912 unsigned int map_disks
= info
->array
.raid_disks
;
913 unsigned long long super_offset
;
914 unsigned long long data_size
;
916 memset(info
, 0, sizeof(*info
));
917 info
->array
.major_version
= 1;
918 info
->array
.minor_version
= st
->minor_version
;
919 info
->array
.patch_version
= 0;
920 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
921 info
->array
.level
= __le32_to_cpu(sb
->level
);
922 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
923 info
->array
.md_minor
= -1;
924 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
925 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
926 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
928 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
930 if (__le32_to_cpu(bsb
->nodes
) > 1)
931 info
->array
.state
|= (1 << MD_SB_CLUSTERED
);
933 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
934 info
->component_size
= __le64_to_cpu(sb
->size
);
935 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
936 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
938 info
->disk
.major
= 0;
939 info
->disk
.minor
= 0;
940 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
941 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
942 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
943 role
= MD_DISK_ROLE_FAULTY
;
945 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
947 super_offset
= __le64_to_cpu(sb
->super_offset
);
948 if (info
->array
.level
<= 0)
949 data_size
= __le64_to_cpu(sb
->data_size
);
951 data_size
= __le64_to_cpu(sb
->size
);
952 if (info
->data_offset
< super_offset
) {
953 unsigned long long end
;
954 info
->space_before
= info
->data_offset
;
957 if (sb
->bblog_offset
&& sb
->bblog_size
) {
958 unsigned long long bboffset
= super_offset
;
959 bboffset
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
964 if (super_offset
+ info
->bitmap_offset
< end
)
965 end
= super_offset
+ info
->bitmap_offset
;
967 if (info
->data_offset
+ data_size
< end
)
968 info
->space_after
= end
- data_size
- info
->data_offset
;
970 info
->space_after
= 0;
972 unsigned long long earliest
;
973 earliest
= super_offset
+ (32+4)*2; /* match kernel */
974 if (info
->bitmap_offset
> 0) {
975 unsigned long long bmend
= info
->bitmap_offset
;
976 unsigned long long size
= calc_bitmap_size(bsb
, 4096);
979 if (bmend
> earliest
)
982 if (sb
->bblog_offset
&& sb
->bblog_size
) {
983 unsigned long long bbend
= super_offset
;
984 bbend
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
985 bbend
+= __le16_to_cpu(sb
->bblog_size
);
986 if (bbend
> earliest
)
989 if (earliest
< info
->data_offset
)
990 info
->space_before
= info
->data_offset
- earliest
;
992 info
->space_before
= 0;
993 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
995 if (info
->space_before
== 0 && info
->space_after
== 0) {
996 /* It will look like we don't support data_offset changes,
997 * be we do - it's just that there is no room.
998 * A change that reduced the number of devices should
999 * still be allowed, so set the otherwise useless value of '1'
1001 info
->space_after
= 1;
1004 info
->disk
.raid_disk
= -1;
1006 case MD_DISK_ROLE_SPARE
:
1007 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
1009 case MD_DISK_ROLE_FAULTY
:
1010 info
->disk
.state
= 1; /* faulty */
1012 case MD_DISK_ROLE_JOURNAL
:
1013 info
->disk
.state
= (1 << MD_DISK_JOURNAL
);
1014 info
->disk
.raid_disk
= role
;
1015 info
->space_after
= (misc
->device_size
- info
->data_offset
) % 8; /* journal uses all 4kB blocks*/
1018 info
->disk
.state
= 6; /* active and in sync */
1019 info
->disk
.raid_disk
= role
;
1021 if (sb
->devflags
& WriteMostly1
)
1022 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1023 info
->events
= __le64_to_cpu(sb
->events
);
1024 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
1025 info
->safe_mode_delay
= 200;
1027 memcpy(info
->uuid
, sb
->set_uuid
, 16);
1029 strncpy(info
->name
, sb
->set_name
, 32);
1032 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
1033 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
1034 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
1037 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
1038 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
1040 info
->recovery_start
= MaxSector
;
1042 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
1043 info
->reshape_active
= 1;
1044 if ((sb
->feature_map
& __le32_to_cpu(MD_FEATURE_NEW_OFFSET
)) &&
1045 sb
->new_offset
!= 0)
1046 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
1047 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
1048 info
->new_level
= __le32_to_cpu(sb
->new_level
);
1049 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
1050 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
1051 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
1052 if (info
->delta_disks
< 0)
1053 info
->array
.raid_disks
-= info
->delta_disks
;
1055 info
->reshape_active
= 0;
1057 info
->recovery_blocked
= info
->reshape_active
;
1060 for (i
=0; i
<map_disks
; i
++)
1062 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
1063 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
1064 if (/*role == MD_DISK_ROLE_SPARE || */role
< (unsigned) info
->array
.raid_disks
) {
1066 if (map
&& role
< map_disks
)
1071 info
->array
.working_disks
= working
;
1072 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_JOURNAL
))
1073 info
->journal_device_required
= 1;
1074 info
->journal_clean
= 0;
1077 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
1079 struct mdinfo
*info
;
1084 info
= xmalloc(sizeof(*info
));
1085 getinfo_super1(st
, info
, NULL
);
1089 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
1091 char *devname
, int verbose
,
1092 int uuid_set
, char *homehost
)
1094 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1095 * if any change was made. For others, the return value is
1100 struct mdp_superblock_1
*sb
= st
->sb
;
1101 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1103 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1104 rv
= cluster_get_dlmlock(&lockid
);
1106 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1107 cluster_release_dlmlock(lockid
);
1112 if (strcmp(update
, "homehost") == 0 &&
1114 /* Note that 'homehost' is special as it is really
1119 c
= strchr(sb
->set_name
, ':');
1121 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1123 strncpy(info
->name
, sb
->set_name
, 32);
1127 if (strcmp(update
, "force-one")==0) {
1128 /* Not enough devices for a working array,
1129 * so bring this one up-to-date
1131 if (sb
->events
!= __cpu_to_le64(info
->events
))
1133 sb
->events
= __cpu_to_le64(info
->events
);
1134 } else if (strcmp(update
, "force-array")==0) {
1135 /* Degraded array and 'force' requests to
1136 * maybe need to mark it 'clean'.
1138 switch(__le32_to_cpu(sb
->level
)) {
1139 case 5: case 4: case 6:
1140 /* need to force clean */
1141 if (sb
->resync_offset
!= MaxSector
)
1143 sb
->resync_offset
= MaxSector
;
1145 } else if (strcmp(update
, "assemble")==0) {
1146 int d
= info
->disk
.number
;
1148 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
1149 want
= info
->disk
.raid_disk
;
1150 else if (info
->disk
.state
& (1<<MD_DISK_JOURNAL
))
1151 want
= MD_DISK_ROLE_JOURNAL
;
1153 want
= MD_DISK_ROLE_SPARE
;
1154 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
1155 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
1158 if (info
->reshape_active
&&
1159 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1160 info
->delta_disks
>= 0 &&
1161 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
1162 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1165 if (info
->reshape_active
&&
1166 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1167 info
->delta_disks
< 0 &&
1168 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
1169 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1172 } else if (strcmp(update
, "linear-grow-new") == 0) {
1175 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
1177 for (i
=0 ; i
< max
; i
++)
1178 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= MD_DISK_ROLE_FAULTY
)
1180 sb
->dev_number
= __cpu_to_le32(i
);
1181 info
->disk
.number
= i
;
1182 if (max
>= __le32_to_cpu(sb
->max_dev
))
1183 sb
->max_dev
= __cpu_to_le32(max
+1);
1185 random_uuid(sb
->device_uuid
);
1188 __cpu_to_le16(info
->disk
.raid_disk
);
1190 fd
= open(devname
, O_RDONLY
);
1192 unsigned long long ds
;
1193 get_dev_size(fd
, devname
, &ds
);
1196 if (__le64_to_cpu(sb
->super_offset
) <
1197 __le64_to_cpu(sb
->data_offset
)) {
1198 sb
->data_size
= __cpu_to_le64(
1199 ds
- __le64_to_cpu(sb
->data_offset
));
1202 ds
&= ~(unsigned long long)(4*2-1);
1203 sb
->super_offset
= __cpu_to_le64(ds
);
1204 sb
->data_size
= __cpu_to_le64(
1205 ds
- __le64_to_cpu(sb
->data_offset
));
1208 } else if (strcmp(update
, "linear-grow-update") == 0) {
1209 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1210 sb
->dev_roles
[info
->disk
.number
] =
1211 __cpu_to_le16(info
->disk
.raid_disk
);
1212 } else if (strcmp(update
, "resync") == 0) {
1213 /* make sure resync happens */
1214 sb
->resync_offset
= 0ULL;
1215 } else if (strcmp(update
, "uuid") == 0) {
1216 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1218 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
)
1219 memcpy(bms
->uuid
, sb
->set_uuid
, 16);
1220 } else if (strcmp(update
, "no-bitmap") == 0) {
1221 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1222 } else if (strcmp(update
, "bbl") == 0) {
1223 /* only possible if there is room after the bitmap, or if
1224 * there is no bitmap
1226 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1227 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1228 long bitmap_offset
= 0;
1229 long bm_sectors
= 0;
1233 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1234 bitmap_offset
= (long)__le32_to_cpu(sb
->bitmap_offset
);
1235 bm_sectors
= calc_bitmap_size(bms
, 4096) >> 9;
1238 if (sb_offset
< data_offset
) {
1239 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1242 bb_offset
= sb_offset
+ 8;
1243 if (bm_sectors
&& bitmap_offset
> 0)
1244 bb_offset
= bitmap_offset
+ bm_sectors
;
1245 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1246 && bb_offset
+ 8+8 <= (long)data_offset
)
1247 /* too close to bitmap, and room to grow */
1249 if (bb_offset
+ 8 <= (long)data_offset
) {
1250 sb
->bblog_size
= __cpu_to_le16(8);
1251 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1254 /* 1.0 - Put bbl just before super block */
1255 if (bm_sectors
&& bitmap_offset
< 0)
1256 space
= -bitmap_offset
- bm_sectors
;
1258 space
= sb_offset
- data_offset
-
1259 __le64_to_cpu(sb
->data_size
);
1261 sb
->bblog_size
= __cpu_to_le16(8);
1262 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1265 } else if (strcmp(update
, "no-bbl") == 0) {
1266 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1267 pr_err("Cannot remove active bbl from %s\n",devname
);
1270 sb
->bblog_shift
= 0;
1271 sb
->bblog_offset
= 0;
1273 } else if (strcmp(update
, "force-no-bbl") == 0) {
1274 sb
->feature_map
&= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1276 sb
->bblog_shift
= 0;
1277 sb
->bblog_offset
= 0;
1278 } else if (strcmp(update
, "name") == 0) {
1279 if (info
->name
[0] == 0)
1280 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1281 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1283 strchr(info
->name
, ':') == NULL
&&
1284 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1285 strcpy(sb
->set_name
, homehost
);
1286 strcat(sb
->set_name
, ":");
1287 strcat(sb
->set_name
, info
->name
);
1289 strncpy(sb
->set_name
, info
->name
, sizeof(sb
->set_name
));
1290 } else if (strcmp(update
, "devicesize") == 0 &&
1291 __le64_to_cpu(sb
->super_offset
) <
1292 __le64_to_cpu(sb
->data_offset
)) {
1293 /* set data_size to device size less data_offset */
1294 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1295 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1296 sb
->data_size
= __cpu_to_le64(
1297 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1298 } else if (strncmp(update
, "revert-reshape", 14) == 0) {
1300 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1301 pr_err("No active reshape to revert on %s\n",
1305 unsigned long long reshape_sectors
;
1308 /* If the reshape hasn't started, just stop it.
1309 * It is conceivable that a stripe was modified but
1310 * the metadata not updated. In that case the backup
1311 * should have been used to get passed the critical stage.
1312 * If that couldn't happen, the "-nobackup" version
1315 if (strcmp(update
, "revert-reshape-nobackup") == 0 &&
1316 sb
->reshape_position
== 0 &&
1317 (__le32_to_cpu(sb
->delta_disks
) > 0 ||
1318 (__le32_to_cpu(sb
->delta_disks
) == 0 &&
1319 !(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
))))) {
1320 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1321 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1322 __le32_to_cpu(sb
->delta_disks
));
1323 sb
->delta_disks
= 0;
1326 /* reshape_position is a little messy.
1327 * Its value must be a multiple of the larger
1328 * chunk size, and of the "after" data disks.
1329 * So when reverting we need to change it to
1330 * be a multiple of the new "after" data disks,
1331 * which is the old "before".
1332 * If it isn't already a multiple of 'before',
1333 * the only thing we could do would be
1334 * copy some block around on the disks, which
1335 * is easy to get wrong.
1336 * So we reject a revert-reshape unless the
1337 * alignment is good.
1339 if (__le32_to_cpu(sb
->level
) >= 4 &&
1340 __le32_to_cpu(sb
->level
) <= 6) {
1341 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1342 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1343 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1344 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1345 if (reshape_sectors
% reshape_chunk
) {
1346 pr_err("Reshape position is not suitably aligned.\n");
1347 pr_err("Try normal assembly and stop again\n");
1351 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1352 __le32_to_cpu(sb
->delta_disks
));
1353 if (sb
->delta_disks
== 0)
1354 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1356 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1358 temp
= sb
->new_layout
;
1359 sb
->new_layout
= sb
->layout
;
1362 temp
= sb
->new_chunk
;
1363 sb
->new_chunk
= sb
->chunksize
;
1364 sb
->chunksize
= temp
;
1366 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1367 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1368 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1369 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1370 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1374 } else if (strcmp(update
, "_reshape_progress")==0)
1375 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1376 else if (strcmp(update
, "writemostly")==0)
1377 sb
->devflags
|= WriteMostly1
;
1378 else if (strcmp(update
, "readwrite")==0)
1379 sb
->devflags
&= ~WriteMostly1
;
1383 sb
->sb_csum
= calc_sb_1_csum(sb
);
1384 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1385 cluster_release_dlmlock(lockid
);
1390 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1391 unsigned long long size
, char *name
, char *homehost
,
1392 int *uuid
, unsigned long long data_offset
)
1394 struct mdp_superblock_1
*sb
;
1399 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1400 pr_err("could not allocate superblock\n");
1403 memset(sb
, 0, SUPER1_SIZE
);
1407 /* zeroing superblock */
1411 spares
= info
->working_disks
- info
->active_disks
;
1412 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1413 pr_err("too many devices requested: %d+%d > %d\n",
1414 info
->raid_disks
, spares
, MAX_DEVS
);
1418 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1419 sb
->major_version
= __cpu_to_le32(1);
1420 sb
->feature_map
= 0;
1424 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1426 random_uuid(sb
->set_uuid
);;
1428 if (name
== NULL
|| *name
== 0) {
1429 sprintf(defname
, "%d", info
->md_minor
);
1433 strchr(name
, ':')== NULL
&&
1434 strlen(homehost
)+1+strlen(name
) < 32) {
1435 strcpy(sb
->set_name
, homehost
);
1436 strcat(sb
->set_name
, ":");
1437 strcat(sb
->set_name
, name
);
1439 strncpy(sb
->set_name
, name
, sizeof(sb
->set_name
));
1441 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1442 sb
->level
= __cpu_to_le32(info
->level
);
1443 sb
->layout
= __cpu_to_le32(info
->layout
);
1444 sb
->size
= __cpu_to_le64(size
*2ULL);
1445 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1446 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1448 sb
->data_offset
= __cpu_to_le64(data_offset
);
1449 sb
->data_size
= __cpu_to_le64(0);
1450 sb
->super_offset
= __cpu_to_le64(0);
1451 sb
->recovery_offset
= __cpu_to_le64(0);
1453 sb
->utime
= sb
->ctime
;
1454 sb
->events
= __cpu_to_le64(1);
1455 if (info
->state
& (1<<MD_SB_CLEAN
))
1456 sb
->resync_offset
= MaxSector
;
1458 sb
->resync_offset
= 0;
1459 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1460 sbsize
= ROUND_UP(sbsize
, 512);
1461 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1463 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1471 long long data_offset
;
1472 mdu_disk_info_t disk
;
1473 struct devinfo
*next
;
1476 /* Add a device to the superblock being created */
1477 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1478 int fd
, char *devname
, unsigned long long data_offset
)
1480 struct mdp_superblock_1
*sb
= st
->sb
;
1481 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1482 struct devinfo
*di
, **dip
;
1483 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1486 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1487 rv
= cluster_get_dlmlock(&lockid
);
1489 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1490 cluster_release_dlmlock(lockid
);
1495 if ((dk
->state
& 6) == 6) /* active, sync */
1496 *rp
= __cpu_to_le16(dk
->raid_disk
);
1497 else if (dk
->state
& (1<<MD_DISK_JOURNAL
))
1498 *rp
= MD_DISK_ROLE_JOURNAL
;
1499 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1500 *rp
= MD_DISK_ROLE_SPARE
;
1502 *rp
= MD_DISK_ROLE_FAULTY
;
1504 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1505 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1506 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1508 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1509 sb
->devflags
= 0; /* don't copy another disks flags */
1510 sb
->sb_csum
= calc_sb_1_csum(sb
);
1512 dip
= (struct devinfo
**)&st
->info
;
1514 dip
= &(*dip
)->next
;
1515 di
= xmalloc(sizeof(struct devinfo
));
1517 di
->devname
= devname
;
1519 di
->data_offset
= data_offset
;
1523 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1524 cluster_release_dlmlock(lockid
);
1530 static int locate_bitmap1(struct supertype
*st
, int fd
, int node_num
);
1532 static int store_super1(struct supertype
*st
, int fd
)
1534 struct mdp_superblock_1
*sb
= st
->sb
;
1535 unsigned long long sb_offset
;
1536 struct align_fd afd
;
1538 unsigned long long dsize
;
1539 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1542 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1543 rv
= cluster_get_dlmlock(&lockid
);
1545 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1546 cluster_release_dlmlock(lockid
);
1551 if (!get_dev_size(fd
, NULL
, &dsize
))
1562 * Calculate the position of the superblock.
1563 * It is always aligned to a 4K boundary and
1564 * depending on minor_version, it can be:
1565 * 0: At least 8K, but less than 12K, from end of device
1566 * 1: At start of device
1567 * 2: 4K from start of device.
1569 switch(st
->minor_version
) {
1573 sb_offset
&= ~(4*2-1);
1585 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1586 0 != __le64_to_cpu(sb
->super_offset
)
1588 pr_err("internal error - sb_offset is wrong\n");
1592 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1595 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1597 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1600 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1601 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1602 (((char*)sb
)+MAX_SB_SIZE
);
1603 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1604 locate_bitmap1(st
, fd
, 0);
1605 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1610 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1611 cluster_release_dlmlock(lockid
);
1616 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1618 static unsigned long choose_bm_space(unsigned long devsize
)
1620 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1621 * if bigger than 200Gig, save 128k
1622 * NOTE: result must be multiple of 4K else bad things happen
1623 * on 4K-sector devices.
1627 if (devsize
- 64*2 >= 200*1024*1024*2)
1629 if (devsize
- 4*2 > 8*1024*1024*2)
1634 static void free_super1(struct supertype
*st
);
1636 #define META_BLOCK_SIZE 4096
1637 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
1640 static int write_empty_r5l_meta_block(struct supertype
*st
, int fd
)
1642 struct r5l_meta_block
*mb
;
1643 struct mdp_superblock_1
*sb
= st
->sb
;
1644 struct align_fd afd
;
1649 if (posix_memalign((void**)&mb
, 4096, META_BLOCK_SIZE
) != 0) {
1650 pr_err("Could not allocate memory for the meta block.\n");
1654 memset(mb
, 0, META_BLOCK_SIZE
);
1656 mb
->magic
= __cpu_to_le32(R5LOG_MAGIC
);
1657 mb
->version
= R5LOG_VERSION
;
1658 mb
->meta_size
= __cpu_to_le32(sizeof(struct r5l_meta_block
));
1659 mb
->seq
= __cpu_to_le64(random32());
1660 mb
->position
= __cpu_to_le64(0);
1662 crc
= crc32c_le(0xffffffff, sb
->set_uuid
, sizeof(sb
->set_uuid
));
1663 crc
= crc32c_le(crc
, (void *)mb
, META_BLOCK_SIZE
);
1666 if (lseek64(fd
, (sb
->data_offset
) * 512, 0) < 0LL) {
1667 pr_err("cannot seek to offset of the meta block\n");
1671 if (awrite(&afd
, mb
, META_BLOCK_SIZE
) != META_BLOCK_SIZE
) {
1672 pr_err("failed to store write the meta block \n");
1685 static int write_init_super1(struct supertype
*st
)
1687 struct mdp_superblock_1
*sb
= st
->sb
;
1688 struct supertype
*refst
;
1690 unsigned long long bm_space
;
1692 unsigned long long dsize
, array_size
;
1693 unsigned long long sb_offset
;
1694 unsigned long long data_offset
;
1697 for (di
= st
->info
; di
; di
= di
->next
) {
1698 if (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))
1699 sb
->feature_map
|= MD_FEATURE_JOURNAL
;
1702 for (di
= st
->info
; di
; di
= di
->next
) {
1703 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1708 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1711 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1712 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1713 sb
->devflags
|= WriteMostly1
;
1715 sb
->devflags
&= ~WriteMostly1
;
1717 random_uuid(sb
->device_uuid
);
1719 if (!(di
->disk
.state
& (1<<MD_DISK_JOURNAL
)))
1722 refst
= dup_super(st
);
1723 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1724 struct mdp_superblock_1
*refsb
= refst
->sb
;
1726 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1727 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1728 /* same array, so preserve events and
1730 sb
->events
= refsb
->events
;
1731 /* bugs in 2.6.17 and earlier mean the
1732 * dev_number chosen in Manage must be preserved
1734 if (get_linux_version() >= 2006018)
1735 sb
->dev_number
= refsb
->dev_number
;
1741 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1754 * Calculate the position of the superblock.
1755 * It is always aligned to a 4K boundary and
1756 * depending on minor_version, it can be:
1757 * 0: At least 8K, but less than 12K, from end of device
1758 * 1: At start of device
1759 * 2: 4K from start of device.
1760 * data_offset has already been set.
1762 array_size
= __le64_to_cpu(sb
->size
);
1764 /* work out how much space we left for a bitmap */
1765 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1766 bitmap_super_t
*bms
= (bitmap_super_t
*)
1767 (((char *)sb
) + MAX_SB_SIZE
);
1768 bm_space
= calc_bitmap_size(bms
, 4096) >> 9;
1769 bm_offset
= (long)__le32_to_cpu(sb
->bitmap_offset
);
1771 bm_space
= choose_bm_space(array_size
);
1775 data_offset
= di
->data_offset
;
1776 if (data_offset
== INVALID_SECTORS
)
1777 data_offset
= st
->data_offset
;
1778 switch(st
->minor_version
) {
1780 /* Add 8 sectors for bad block log */
1782 if (data_offset
== INVALID_SECTORS
)
1786 sb_offset
&= ~(4*2-1);
1787 sb
->data_offset
= __cpu_to_le64(data_offset
);
1788 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1789 if (sb_offset
< array_size
+ bm_space
)
1790 bm_space
= sb_offset
- array_size
;
1791 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1792 if (bm_space
>= 8) {
1793 sb
->bblog_size
= __cpu_to_le16(8);
1794 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1799 sb_offset
= st
->minor_version
== 2 ? 8 : 0;
1800 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1801 if (data_offset
== INVALID_SECTORS
)
1802 data_offset
= sb_offset
+ 16;
1804 sb
->data_offset
= __cpu_to_le64(data_offset
);
1805 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1806 if (data_offset
>= sb_offset
+bm_offset
+bm_space
+8) {
1807 sb
->bblog_size
= __cpu_to_le16(8);
1808 sb
->bblog_offset
= __cpu_to_le32(bm_offset
+
1810 } else if (data_offset
>= sb_offset
+ 16) {
1811 sb
->bblog_size
= __cpu_to_le16(8);
1812 /* '8' sectors for the bblog, and 'sb_offset'
1813 * because we want offset from superblock, not
1816 sb
->bblog_offset
= __cpu_to_le32(data_offset
-
1821 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1822 st
->minor_version
, di
->devname
);
1826 /* Disable badblock log on clusters, or when explicitly requested */
1827 if (st
->nodes
> 0 || conf_get_create_info()->bblist
== 0) {
1829 sb
->bblog_offset
= 0;
1832 sb
->sb_csum
= calc_sb_1_csum(sb
);
1833 rv
= store_super1(st
, di
->fd
);
1835 if (rv
== 0 && (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))) {
1836 rv
= write_empty_r5l_meta_block(st
, di
->fd
);
1841 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1842 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NodeNumUpdate
);
1850 pr_err("Failed to write metadata to %s\n",
1857 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1861 * 0 same, or first was empty, and second was copied
1862 * 1 second had wrong number
1864 * 3 wrong other info
1866 struct mdp_superblock_1
*first
= st
->sb
;
1867 struct mdp_superblock_1
*second
= tst
->sb
;
1869 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1871 if (second
->major_version
!= __cpu_to_le32(1))
1875 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1876 pr_err("could not allocate superblock\n");
1879 memcpy(first
, second
, SUPER1_SIZE
);
1883 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1886 if (first
->ctime
!= second
->ctime
||
1887 first
->level
!= second
->level
||
1888 first
->layout
!= second
->layout
||
1889 first
->size
!= second
->size
||
1890 first
->chunksize
!= second
->chunksize
||
1891 first
->raid_disks
!= second
->raid_disks
)
1896 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1898 unsigned long long dsize
;
1899 unsigned long long sb_offset
;
1900 struct mdp_superblock_1
*super
;
1902 struct bitmap_super_s
*bsb
;
1903 struct misc_dev_info
*misc
;
1904 struct align_fd afd
;
1910 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1912 struct supertype tst
;
1913 __u64 bestctime
= 0;
1914 /* guess... choose latest ctime */
1915 memset(&tst
, 0, sizeof(tst
));
1917 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1918 switch(load_super1(&tst
, fd
, devname
)) {
1919 case 0: super
= tst
.sb
;
1920 if (bestvers
== -1 ||
1921 bestctime
< __le64_to_cpu(super
->ctime
)) {
1922 bestvers
= tst
.minor_version
;
1923 bestctime
= __le64_to_cpu(super
->ctime
);
1928 case 1: return 1; /*bad device */
1929 case 2: break; /* bad, try next */
1932 if (bestvers
!= -1) {
1934 tst
.minor_version
= bestvers
;
1936 tst
.max_devs
= MAX_DEVS
;
1937 rv
= load_super1(&tst
, fd
, devname
);
1944 if (!get_dev_size(fd
, devname
, &dsize
))
1950 pr_err("%s is too small for md: size is %llu sectors.\n",
1956 * Calculate the position of the superblock.
1957 * It is always aligned to a 4K boundary and
1958 * depending on minor_version, it can be:
1959 * 0: At least 8K, but less than 12K, from end of device
1960 * 1: At start of device
1961 * 2: 4K from start of device.
1963 switch(st
->minor_version
) {
1967 sb_offset
&= ~(4*2-1);
1979 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1981 pr_err("Cannot seek to superblock on %s: %s\n",
1982 devname
, strerror(errno
));
1986 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
1987 pr_err("could not allocate superblock\n");
1991 memset(super
, 0, SUPER1_SIZE
);
1993 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
1995 pr_err("Cannot read superblock on %s\n",
2001 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
2003 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2004 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
2009 if (__le32_to_cpu(super
->major_version
) != 1) {
2011 pr_err("Cannot interpret superblock on %s - version is %d\n",
2012 devname
, __le32_to_cpu(super
->major_version
));
2016 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
2018 pr_err("No superblock found on %s (super_offset is wrong)\n",
2025 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2027 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
2028 misc
->device_size
= dsize
;
2029 if (st
->data_offset
== INVALID_SECTORS
)
2030 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
2032 /* Now check on the bitmap superblock */
2033 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
2035 /* Read the bitmap superblock and make sure it looks
2036 * valid. If it doesn't clear the bit. An --assemble --force
2037 * should get that written out.
2039 locate_bitmap1(st
, fd
, 0);
2040 if (aread(&afd
, bsb
, 512) != 512)
2043 uuid_from_super1(st
, uuid
);
2044 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
2045 memcmp(bsb
->uuid
, uuid
, 16) != 0)
2050 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
2051 & ~MD_FEATURE_BITMAP_OFFSET
);
2055 static struct supertype
*match_metadata_desc1(char *arg
)
2057 struct supertype
*st
= xcalloc(1, sizeof(*st
));
2059 st
->container_devnm
[0] = 0;
2061 st
->max_devs
= MAX_DEVS
;
2063 st
->data_offset
= INVALID_SECTORS
;
2064 /* leading zeros can be safely ignored. --detail generates them. */
2067 if (strcmp(arg
, "1.0") == 0 ||
2068 strcmp(arg
, "1.00") == 0) {
2069 st
->minor_version
= 0;
2072 if (strcmp(arg
, "1.1") == 0 ||
2073 strcmp(arg
, "1.01") == 0
2075 st
->minor_version
= 1;
2078 if (strcmp(arg
, "1.2") == 0 ||
2079 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2080 strcmp(arg
, "default") == 0 ||
2081 #endif /* DEFAULT_OLD_METADATA */
2082 strcmp(arg
, "1.02") == 0) {
2083 st
->minor_version
= 2;
2086 if (strcmp(arg
, "1") == 0 ||
2087 strcmp(arg
, "default") == 0) {
2088 st
->minor_version
= -1;
2096 /* find available size on device with this devsize, using
2097 * superblock type st, and reserving 'reserve' sectors for
2100 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
2101 unsigned long long data_offset
)
2103 struct mdp_superblock_1
*super
= st
->sb
;
2110 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
2111 /* hot-add. allow for actual size of bitmap */
2112 struct bitmap_super_s
*bsb
;
2113 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2114 bmspace
= calc_bitmap_size(bsb
, 4096) >> 9;
2117 /* Allow space for bad block log */
2118 if (super
->bblog_size
)
2119 bbspace
= __le16_to_cpu(super
->bblog_size
);
2121 if (st
->minor_version
< 0)
2122 /* not specified, so time to set default */
2123 st
->minor_version
= 2;
2125 if (data_offset
== INVALID_SECTORS
)
2126 data_offset
= st
->data_offset
;
2128 if (data_offset
!= INVALID_SECTORS
)
2129 switch(st
->minor_version
) {
2131 return devsize
- data_offset
- 8*2 - bbspace
;
2134 return devsize
- data_offset
;
2141 switch(st
->minor_version
) {
2144 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2146 /* at start, 4K for superblock and possible bitmap */
2147 return devsize
- 4*2 - bbspace
;
2149 /* 4k from start, 4K for superblock and possible bitmap */
2150 return devsize
- (4+4)*2 - bbspace
;
2156 add_internal_bitmap1(struct supertype
*st
,
2157 int *chunkp
, int delay
, int write_behind
,
2158 unsigned long long size
,
2159 int may_change
, int major
)
2162 * If not may_change, then this is a 'Grow' without sysfs support for
2163 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2164 * If may_change, then this is create or a Grow with sysfs syupport,
2165 * and we can put the bitmap wherever we like.
2167 * size is in sectors, chunk is in bytes !!!
2170 unsigned long long bits
;
2171 unsigned long long max_bits
;
2172 unsigned long long min_chunk
;
2174 long bbl_offset
, bbl_size
;
2175 unsigned long long chunk
= *chunkp
;
2179 struct mdp_superblock_1
*sb
= st
->sb
;
2180 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2183 if (__le64_to_cpu(sb
->data_size
) == 0)
2184 /* Must be creating the array, else data_size would be non-zero */
2186 switch(st
->minor_version
) {
2188 /* either 3K after the superblock (when hot-add),
2189 * or some amount of space before.
2192 /* We are creating array, so we *know* how much room has
2197 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2199 room
= __le64_to_cpu(sb
->super_offset
)
2200 - __le64_to_cpu(sb
->data_offset
)
2201 - __le64_to_cpu(sb
->data_size
);
2202 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2205 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2206 if (bbl_size
< -bbl_offset
)
2207 bbl_size
= -bbl_offset
;
2209 if (!may_change
|| (room
< 3*2 &&
2210 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2215 offset
= 0; /* means movable offset */
2220 case 2: /* between superblock and data */
2224 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2226 room
= __le64_to_cpu(sb
->data_offset
)
2227 - __le64_to_cpu(sb
->super_offset
);
2228 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2230 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2235 room
-= 2; /* Leave 1K for superblock */
2239 room
-= 4*2; /* leave 4K for superblock */
2249 if (chunk
== UnSet
&& room
> 128*2)
2250 /* Limit to 128K of bitmap when chunk size not requested */
2254 /* No room for a bitmap */
2257 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2259 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2260 bits
= (size
*512)/min_chunk
+1;
2261 while (bits
> max_bits
) {
2265 if (chunk
== UnSet
) {
2266 /* For practical purpose, 64Meg is a good
2267 * default chunk size for internal bitmaps.
2270 if (chunk
< 64*1024*1024)
2271 chunk
= 64*1024*1024;
2272 } else if (chunk
< min_chunk
)
2273 return -EINVAL
; /* chunk size too small */
2274 if (chunk
== 0) /* rounding problem */
2278 /* start bitmap on a 4K boundary with enough space for
2281 bits
= (size
*512) / chunk
+ 1;
2282 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2283 room
*= 8; /* convert 4K blocks to sectors */
2284 offset
= -room
- bbl_size
;
2287 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2289 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2290 | MD_FEATURE_BITMAP_OFFSET
);
2291 memset(bms
, 0, sizeof(*bms
));
2292 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2293 bms
->version
= __cpu_to_le32(major
);
2294 uuid_from_super1(st
, uuid
);
2295 memcpy(bms
->uuid
, uuid
, 16);
2296 bms
->chunksize
= __cpu_to_le32(chunk
);
2297 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2298 bms
->sync_size
= __cpu_to_le64(size
);
2299 bms
->write_behind
= __cpu_to_le32(write_behind
);
2300 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2302 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2303 | MD_FEATURE_BITMAP_VERSIONED
);
2304 if (st
->cluster_name
) {
2305 len
= sizeof(bms
->cluster_name
);
2306 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, len
);
2307 bms
->cluster_name
[len
- 1] = '\0';
2314 static int locate_bitmap1(struct supertype
*st
, int fd
, int node_num
)
2316 unsigned long long offset
;
2317 struct mdp_superblock_1
*sb
;
2322 if (st
->ss
->load_super(st
, fd
, NULL
))
2323 return -1; /* no error I hope... */
2328 if ((__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
2332 offset
= __le64_to_cpu(sb
->super_offset
);
2333 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
) * (node_num
+ 1);
2336 lseek64(fd
, offset
<<9, 0);
2340 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2342 struct mdp_superblock_1
*sb
= st
->sb
;
2343 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2346 int towrite
, n
, len
;
2347 struct align_fd afd
;
2349 unsigned long long total_bm_space
, bm_space_per_node
;
2353 /* update cluster name */
2354 if (st
->cluster_name
) {
2355 len
= sizeof(bms
->cluster_name
);
2356 memset((char *)bms
->cluster_name
, 0, len
);
2357 strncpy((char *)bms
->cluster_name
,
2358 st
->cluster_name
, len
);
2359 bms
->cluster_name
[len
- 1] = '\0';
2363 /* cluster md only supports superblock 1.2 now */
2364 if (st
->minor_version
!= 2 && bms
->version
== BITMAP_MAJOR_CLUSTERED
) {
2365 pr_err("Warning: cluster md only works with superblock 1.2\n");
2369 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
) {
2370 if (st
->nodes
== 1) {
2371 /* the parameter for nodes is not valid */
2372 pr_err("Warning: cluster-md at least needs two nodes\n");
2374 } else if (st
->nodes
== 0)
2375 /* --nodes is not specified */
2377 else if (__cpu_to_le32(st
->nodes
) < bms
->nodes
) {
2378 /* Since the nodes num is not increased, no need to check the space
2379 * is enough or not, just update bms->nodes */
2380 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2384 /* no need to change bms->nodes for other bitmap types */
2386 pr_err("Warning: --nodes option is only suitable for clustered bitmap\n");
2390 /* Each node has an independent bitmap, it is necessary to calculate the
2391 * space is enough or not, first get how many bytes for the total bitmap */
2392 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2394 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2395 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2397 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2398 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2399 total_bm_space
/ bm_space_per_node
);
2403 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2412 locate_bitmap1(st
, fd
, 0);
2414 if (posix_memalign(&buf
, 4096, 4096))
2418 /* Only the bitmap[0] should resync
2419 * whole device on initial assembly
2422 memset(buf
, 0x00, 4096);
2424 memset(buf
, 0xff, 4096);
2425 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2428 * use 4096 boundary if bitmap_offset is aligned
2429 * with 8 sectors, then it should compatible with
2432 if (__le32_to_cpu(sb
->bitmap_offset
) & 7)
2433 towrite
= calc_bitmap_size(bms
, 512);
2435 towrite
= calc_bitmap_size(bms
, 4096);
2436 while (towrite
> 0) {
2440 n
= awrite(&afd
, buf
, n
);
2446 memset(buf
, 0x00, 4096);
2448 memset(buf
, 0xff, 4096);
2455 } while (++i
< __le32_to_cpu(bms
->nodes
));
2461 static void free_super1(struct supertype
*st
)
2467 struct devinfo
*di
= st
->info
;
2468 st
->info
= di
->next
;
2477 static int validate_geometry1(struct supertype
*st
, int level
,
2478 int layout
, int raiddisks
,
2479 int *chunk
, unsigned long long size
,
2480 unsigned long long data_offset
,
2481 char *subdev
, unsigned long long *freesize
,
2484 unsigned long long ldsize
, devsize
;
2486 unsigned long long headroom
;
2489 if (level
== LEVEL_CONTAINER
) {
2491 pr_err("1.x metadata does not support containers\n");
2494 if (*chunk
== UnSet
)
2495 *chunk
= DEFAULT_CHUNK
;
2500 if (st
->minor_version
< 0)
2501 /* not specified, so time to set default */
2502 st
->minor_version
= 2;
2504 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2507 pr_err("super1.x cannot open %s: %s\n",
2508 subdev
, strerror(errno
));
2512 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2518 devsize
= ldsize
>> 9;
2524 /* creating: allow suitable space for bitmap */
2525 bmspace
= choose_bm_space(devsize
);
2527 if (data_offset
== INVALID_SECTORS
)
2528 data_offset
= st
->data_offset
;
2529 if (data_offset
== INVALID_SECTORS
)
2530 switch (st
->minor_version
) {
2536 /* Choose data offset appropriate for this device
2537 * and use as default for whole array.
2538 * The data_offset must allow for bitmap space
2539 * and base metadata, should allow for some headroom
2540 * for reshape, and should be rounded to multiple
2542 * Headroom is limited to 128M, but aim for about 0.1%
2544 headroom
= 128*1024*2;
2545 while ((headroom
<< 10) > devsize
&&
2547 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2549 data_offset
= 12*2 + bmspace
+ headroom
;
2550 #define ONE_MEG (2*1024)
2551 if (data_offset
> ONE_MEG
)
2552 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2555 if (st
->data_offset
== INVALID_SECTORS
)
2556 st
->data_offset
= data_offset
;
2557 switch(st
->minor_version
) {
2558 case 0: /* metadata at end. Round down and subtract space to reserve */
2559 devsize
= (devsize
& ~(4ULL*2-1));
2560 /* space for metadata, bblog, bitmap */
2561 devsize
-= 8*2 + 8 + bmspace
;
2565 devsize
-= data_offset
;
2568 *freesize
= devsize
;
2571 #endif /* MDASSEMBLE */
2573 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2575 /* Create a v1.0 superblock based on 'info'*/
2577 struct mdp_superblock_1
*sb
;
2579 unsigned long long offset
;
2581 if (posix_memalign(&ret
, 4096, 1024) != 0)
2584 memset(ret
, 0, 1024);
2585 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2586 sb
->major_version
= __cpu_to_le32(1);
2588 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2589 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2590 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2591 sb
->level
= __cpu_to_le32(info
->array
.level
);
2592 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2593 sb
->size
= __cpu_to_le64(info
->component_size
);
2594 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2595 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2596 if (info
->array
.level
> 0)
2597 sb
->data_size
= sb
->size
;
2599 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2600 sb
->resync_offset
= MaxSector
;
2601 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2602 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2603 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2605 offset
= st
->devsize
/512 - 8*2;
2607 sb
->super_offset
= __cpu_to_le64(offset
);
2608 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2610 random_uuid(sb
->device_uuid
);
2612 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2613 int state
= sb0
->disks
[i
].state
;
2614 sb
->dev_roles
[i
] = MD_DISK_ROLE_SPARE
;
2615 if ((state
& (1<<MD_DISK_SYNC
)) &&
2616 !(state
& (1<<MD_DISK_FAULTY
)))
2617 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2619 sb
->sb_csum
= calc_sb_1_csum(sb
);
2623 struct superswitch super1
= {
2625 .examine_super
= examine_super1
,
2626 .brief_examine_super
= brief_examine_super1
,
2627 .export_examine_super
= export_examine_super1
,
2628 .detail_super
= detail_super1
,
2629 .brief_detail_super
= brief_detail_super1
,
2630 .export_detail_super
= export_detail_super1
,
2631 .write_init_super
= write_init_super1
,
2632 .validate_geometry
= validate_geometry1
,
2633 .add_to_super
= add_to_super1
,
2634 .examine_badblocks
= examine_badblocks_super1
,
2635 .copy_metadata
= copy_metadata1
,
2637 .match_home
= match_home1
,
2638 .uuid_from_super
= uuid_from_super1
,
2639 .getinfo_super
= getinfo_super1
,
2640 .container_content
= container_content1
,
2641 .update_super
= update_super1
,
2642 .init_super
= init_super1
,
2643 .store_super
= store_super1
,
2644 .compare_super
= compare_super1
,
2645 .load_super
= load_super1
,
2646 .match_metadata_desc
= match_metadata_desc1
,
2647 .avail_size
= avail_size1
,
2648 .add_internal_bitmap
= add_internal_bitmap1
,
2649 .locate_bitmap
= locate_bitmap1
,
2650 .write_bitmap
= write_bitmap1
,
2651 .free_super
= free_super1
,
2652 #if __BYTE_ORDER == BIG_ENDIAN