2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
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 */
71 __u64 recovery_offset
;/* sectors before this offset (from data_offset) have been recovered */
72 __u32 dev_number
; /* permanent identifier of this device - not role in raid */
73 __u32 cnt_corrected_read
; /* number of read errors that were corrected by re-writing */
74 __u8 device_uuid
[16]; /* user-space setable, ignored by kernel */
75 __u8 devflags
; /* per-device flags. Only one defined...*/
76 #define WriteMostly1 1 /* mask for writemostly flag in above */
77 /* bad block log. If there are any bad blocks the feature flag is set.
78 * if offset and size are non-zero, that space is reserved and available.
80 __u8 bblog_shift
; /* shift from sectors to block size for badblocklist */
81 __u16 bblog_size
; /* number of sectors reserved for badblocklist */
82 __u32 bblog_offset
; /* sector offset from superblock to bblog, signed */
84 /* array state information - 64 bytes */
85 __u64 utime
; /* 40 bits second, 24 btes microseconds */
86 __u64 events
; /* incremented when superblock updated */
87 __u64 resync_offset
; /* data before this offset (from data_offset) known to be in sync */
88 __u32 sb_csum
; /* checksum upto dev_roles[max_dev] */
89 __u32 max_dev
; /* size of dev_roles[] array to consider */
90 __u8 pad3
[64-32]; /* set to 0 when writing */
92 /* device state information. Indexed by dev_number.
94 * Note there are no per-device state flags. State information is rolled
95 * into the 'roles' value. If a device is spare or faulty, then it doesn't
96 * have a meaningful role.
98 __u16 dev_roles
[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
101 #define MAX_SB_SIZE 4096
102 /* bitmap super size is 256, but we round up to a sector for alignment */
103 #define BM_SUPER_SIZE 512
104 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
105 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
106 + sizeof(struct misc_dev_info))
108 struct misc_dev_info
{
112 /* feature_map bits */
113 #define MD_FEATURE_BITMAP_OFFSET 1
114 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
117 #define MD_FEATURE_RESHAPE_ACTIVE 4
118 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
119 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
120 * active device with same 'role'.
121 * 'recovery_offset' is also set.
123 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
124 * of devices, but is going
127 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
128 #define MD_FEATURE_BITMAP_VERSIONED 256 /* bitmap version number checked properly */
129 #define MD_FEATURE_JOURNAL 512 /* support write journal */
130 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
131 |MD_FEATURE_RECOVERY_OFFSET \
132 |MD_FEATURE_RESHAPE_ACTIVE \
133 |MD_FEATURE_BAD_BLOCKS \
134 |MD_FEATURE_REPLACEMENT \
135 |MD_FEATURE_RESHAPE_BACKWARDS \
136 |MD_FEATURE_NEW_OFFSET \
137 |MD_FEATURE_BITMAP_VERSIONED \
138 |MD_FEATURE_JOURNAL \
141 /* return how many bytes are needed for bitmap, for cluster-md each node
142 * should have it's own bitmap */
143 static unsigned int calc_bitmap_size(bitmap_super_t
*bms
, unsigned int boundary
)
145 unsigned long long bits
, bytes
;
147 bits
= __le64_to_cpu(bms
->sync_size
) / (__le32_to_cpu(bms
->chunksize
)>>9);
148 bytes
= (bits
+7) >> 3;
149 bytes
+= sizeof(bitmap_super_t
);
150 bytes
= ROUND_UP(bytes
, boundary
);
155 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
157 unsigned int disk_csum
, csum
;
158 unsigned long long newcsum
;
159 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
160 unsigned int *isuper
= (unsigned int*)sb
;
162 /* make sure I can count... */
163 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
164 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
165 sizeof(struct mdp_superblock_1
) != 256) {
166 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
169 disk_csum
= sb
->sb_csum
;
172 for (; size
>=4; size
-= 4 ) {
173 newcsum
+= __le32_to_cpu(*isuper
);
178 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
180 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
181 sb
->sb_csum
= disk_csum
;
182 return __cpu_to_le32(csum
);
186 * Information related to file descriptor used for aligned reads/writes.
187 * Cache the block size.
194 static void init_afd(struct align_fd
*afd
, int fd
)
198 if (ioctl(afd
->fd
, BLKSSZGET
, &afd
->blk_sz
) != 0)
202 static char abuf
[4096+4096];
203 static int aread(struct align_fd
*afd
, void *buf
, int len
)
206 * On devices with a 4K sector size, we need to read
207 * the full sector and copy relevant bits into
216 if (!bsize
|| bsize
> 4096 || len
> 4096) {
218 fprintf(stderr
, "WARNING - aread() called with invalid block size\n");
221 b
= ROUND_UP_PTR((char *)abuf
, 4096);
223 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
225 n
= read(afd
->fd
, b
, iosize
);
228 lseek(afd
->fd
, len
- n
, 1);
235 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
238 * On devices with a 4K sector size, we need to write
239 * the full sector. We pre-read if the sector is larger
241 * The address must be sector-aligned.
248 if (!bsize
|| bsize
> 4096 || len
> 4096) {
250 fprintf(stderr
, "WARNING - awrite() called with invalid block size\n");
253 b
= ROUND_UP_PTR((char *)abuf
, 4096);
255 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
259 n
= read(afd
->fd
, b
, iosize
);
262 lseek(afd
->fd
, -n
, 1);
266 n
= write(afd
->fd
, b
, iosize
);
269 lseek(afd
->fd
, len
- n
, 1);
274 static void examine_super1(struct supertype
*st
, char *homehost
)
276 struct mdp_superblock_1
*sb
= st
->sb
;
277 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
284 int l
= homehost
? strlen(homehost
) : 0;
286 unsigned long long sb_offset
;
289 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
290 printf(" Version : 1");
291 sb_offset
= __le64_to_cpu(sb
->super_offset
);
294 else if (sb_offset
<= 8)
298 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
299 printf(" Array UUID : ");
300 for (i
=0; i
<16; i
++) {
301 if ((i
&3)==0 && i
!= 0) printf(":");
302 printf("%02x", sb
->set_uuid
[i
]);
305 printf(" Name : %.32s", sb
->set_name
);
306 if (l
> 0 && l
< 32 &&
307 sb
->set_name
[l
] == ':' &&
308 strncmp(sb
->set_name
, homehost
, l
) == 0)
309 printf(" (local to host %s)", homehost
);
312 printf(" Cluster Name : %-64s\n", bms
->cluster_name
);
313 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
314 printf(" Creation Time : %.24s\n", ctime(&atime
));
315 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
316 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
317 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
319 printf(" Avail Dev Size : %llu%s\n",
320 (unsigned long long)__le64_to_cpu(sb
->data_size
),
321 human_size(__le64_to_cpu(sb
->data_size
)<<9));
322 if (__le32_to_cpu(sb
->level
) > 0) {
323 int ddsks
= 0, ddsks_denom
= 1;
324 switch(__le32_to_cpu(sb
->level
)) {
325 case 1: ddsks
=1;break;
327 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
328 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
330 layout
= __le32_to_cpu(sb
->layout
);
331 ddsks
= __le32_to_cpu(sb
->raid_disks
);
332 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
335 long long asize
= __le64_to_cpu(sb
->size
);
336 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
337 printf(" Array Size : %llu%s\n",
338 asize
>> 10, human_size(asize
));
340 if (sb
->size
!= sb
->data_size
)
341 printf(" Used Dev Size : %llu%s\n",
342 (unsigned long long)__le64_to_cpu(sb
->size
),
343 human_size(__le64_to_cpu(sb
->size
)<<9));
346 printf(" Data Offset : %llu sectors\n",
347 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
348 if (sb
->new_offset
&&
349 (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
)) {
350 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
351 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
352 printf(" New Offset : %llu sectors\n", offset
);
354 printf(" Super Offset : %llu sectors\n",
355 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
356 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
357 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
359 st
->ss
->getinfo_super(st
, &info
, NULL
);
360 if (info
.space_after
!= 1 &&
361 !(__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
362 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
363 info
.space_before
, info
.space_after
);
365 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
366 printf(" Device UUID : ");
367 for (i
=0; i
<16; i
++) {
368 if ((i
&3)==0 && i
!= 0) printf(":");
369 printf("%02x", sb
->device_uuid
[i
]);
373 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
374 printf("Internal Bitmap : %ld sectors from superblock\n",
375 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
377 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
378 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
379 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
380 if (__le32_to_cpu(sb
->delta_disks
)) {
381 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
382 printf(" (%d->%d)\n",
383 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
384 __le32_to_cpu(sb
->raid_disks
));
385 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
386 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
388 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
389 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
390 printf(" New Level : %s\n", c
?c
:"-unknown-");
392 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
393 if (__le32_to_cpu(sb
->level
) == 5) {
394 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
395 printf(" New Layout : %s\n", c
?c
:"-unknown-");
397 if (__le32_to_cpu(sb
->level
) == 6) {
398 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
399 printf(" New Layout : %s\n", c
?c
:"-unknown-");
401 if (__le32_to_cpu(sb
->level
) == 10) {
402 printf(" New Layout :");
403 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
407 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
408 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
413 if (sb
->devflags
& WriteMostly1
)
414 printf(" write-mostly");
418 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
419 printf(" Update Time : %.24s\n", ctime(&atime
));
421 if (sb
->bblog_size
&& sb
->bblog_offset
) {
422 printf(" Bad Block Log : %d entries available at offset %ld sectors",
423 __le16_to_cpu(sb
->bblog_size
)*512/8,
424 (long)(int32_t)__le32_to_cpu(sb
->bblog_offset
));
425 if (sb
->feature_map
&
426 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
427 printf(" - bad blocks present.");
431 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
432 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
434 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
435 __le32_to_cpu(calc_sb_1_csum(sb
)));
436 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
438 if (__le32_to_cpu(sb
->level
) == 5) {
439 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
440 printf(" Layout : %s\n", c
?c
:"-unknown-");
442 if (__le32_to_cpu(sb
->level
) == 6) {
443 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
444 printf(" Layout : %s\n", c
?c
:"-unknown-");
446 if (__le32_to_cpu(sb
->level
) == 10) {
447 int lo
= __le32_to_cpu(sb
->layout
);
449 print_r10_layout(lo
);
452 switch(__le32_to_cpu(sb
->level
)) {
458 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
461 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
467 /* This turns out to just be confusing */
468 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
469 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
470 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != MD_DISK_ROLE_SPARE
)
472 for (d
=0; d
< i
; d
++) {
473 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
475 if (role
== MD_DISK_ROLE_SPARE
) printf("empty");
476 else if(role
== MD_DISK_ROLE_FAULTY
) printf("failed");
477 else printf("%d", role
);
481 printf(" Device Role : ");
482 d
= __le32_to_cpu(sb
->dev_number
);
483 if (d
< __le32_to_cpu(sb
->max_dev
))
484 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
486 role
= MD_DISK_ROLE_SPARE
;
487 if (role
>= MD_DISK_ROLE_FAULTY
)
489 else if (role
== MD_DISK_ROLE_JOURNAL
)
491 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
492 printf("Replacement device %d\n", role
);
494 printf("Active device %d\n", role
);
496 printf(" Array State : ");
497 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
500 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
501 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
515 /* This is confusing too */
517 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
518 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
519 if (role
== MD_DISK_ROLE_FAULTY
)
522 if (faulty
) printf(" %d failed", faulty
);
524 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
528 static void brief_examine_super1(struct supertype
*st
, int verbose
)
530 struct mdp_superblock_1
*sb
= st
->sb
;
532 unsigned long long sb_offset
;
534 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
536 nm
= strchr(sb
->set_name
, ':');
539 else if (sb
->set_name
[0])
551 printf(" level=%s", c
);
552 sb_offset
= __le64_to_cpu(sb
->super_offset
);
554 printf(" metadata=1.1 ");
555 else if (sb_offset
<= 8)
556 printf(" metadata=1.2 ");
558 printf(" metadata=1.0 ");
560 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
562 for (i
=0; i
<16; i
++) {
563 if ((i
&3)==0 && i
!= 0) printf(":");
564 printf("%02x", sb
->set_uuid
[i
]);
566 if (sb
->set_name
[0]) {
568 print_quoted(sb
->set_name
);
573 static void export_examine_super1(struct supertype
*st
)
575 struct mdp_superblock_1
*sb
= st
->sb
;
580 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
581 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
583 if (sb
->set_name
[i
] == '\n' ||
584 sb
->set_name
[i
] == '\0') {
589 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
590 if (__le32_to_cpu(sb
->level
) > 0) {
591 int ddsks
= 0, ddsks_denom
= 1;
592 switch(__le32_to_cpu(sb
->level
)) {
593 case 1: ddsks
=1;break;
595 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
596 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
598 layout
= __le32_to_cpu(sb
->layout
);
599 ddsks
= __le32_to_cpu(sb
->raid_disks
);
600 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
603 long long asize
= __le64_to_cpu(sb
->size
);
604 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
605 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
609 for (i
=0; i
<16; i
++) {
610 if ((i
&3)==0 && i
!= 0) printf(":");
611 printf("%02x", sb
->set_uuid
[i
]);
614 printf("MD_UPDATE_TIME=%llu\n",
615 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
616 printf("MD_DEV_UUID=");
617 for (i
=0; i
<16; i
++) {
618 if ((i
&3)==0 && i
!= 0) printf(":");
619 printf("%02x", sb
->device_uuid
[i
]);
622 printf("MD_EVENTS=%llu\n",
623 (unsigned long long)__le64_to_cpu(sb
->events
));
626 static int copy_metadata1(struct supertype
*st
, int from
, int to
)
628 /* Read superblock. If it looks good, write it out.
629 * Then if a bitmap is present, copy that.
630 * And if a bad-block-list is present, copy that too.
633 unsigned long long dsize
, sb_offset
;
634 const int bufsize
= 4*1024;
635 struct mdp_superblock_1 super
, *sb
;
637 if (posix_memalign(&buf
, 4096, bufsize
) != 0)
640 if (!get_dev_size(from
, NULL
, &dsize
))
646 switch(st
->minor_version
) {
650 sb_offset
&= ~(4*2-1);
662 if (lseek64(from
, sb_offset
<< 9, 0) < 0LL)
664 if (read(from
, buf
, bufsize
) != bufsize
)
668 super
= *sb
; // save most of sb for when we reuse buf
670 if (__le32_to_cpu(super
.magic
) != MD_SB_MAGIC
||
671 __le32_to_cpu(super
.major_version
) != 1 ||
672 __le64_to_cpu(super
.super_offset
) != sb_offset
||
673 calc_sb_1_csum(sb
) != super
.sb_csum
)
676 if (lseek64(to
, sb_offset
<< 9, 0) < 0LL)
678 if (write(to
, buf
, bufsize
) != bufsize
)
681 if (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
)) {
682 unsigned long long bitmap_offset
= sb_offset
;
683 int bytes
= 4096; // just an estimate.
685 struct align_fd afrom
, ato
;
687 init_afd(&afrom
, from
);
690 bitmap_offset
+= (int32_t)__le32_to_cpu(super
.bitmap_offset
);
692 if (lseek64(from
, bitmap_offset
<<9, 0) < 0)
694 if (lseek64(to
, bitmap_offset
<<9, 0) < 0)
697 for (written
= 0; written
< bytes
; ) {
698 int n
= bytes
- written
;
701 if (aread(&afrom
, buf
, n
) != n
)
704 /* have the header, can calculate
705 * correct bitmap bytes */
708 bytes
= calc_bitmap_size(bms
, 512);
712 if (awrite(&ato
, buf
, n
) != n
)
718 if (super
.bblog_size
!= 0 &&
719 __le32_to_cpu(super
.bblog_size
) <= 100 &&
720 super
.bblog_offset
!= 0 &&
721 (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BAD_BLOCKS
))) {
722 /* There is a bad block log */
723 unsigned long long bb_offset
= sb_offset
;
724 int bytes
= __le32_to_cpu(super
.bblog_size
) * 512;
726 struct align_fd afrom
, ato
;
728 init_afd(&afrom
, from
);
731 bb_offset
+= (int32_t)__le32_to_cpu(super
.bblog_offset
);
733 if (lseek64(from
, bb_offset
<<9, 0) < 0)
735 if (lseek64(to
, bb_offset
<<9, 0) < 0)
738 for (written
= 0; written
< bytes
; ) {
739 int n
= bytes
- written
;
742 if (aread(&afrom
, buf
, n
) != n
)
745 if (awrite(&ato
, buf
, n
) != n
)
759 static void detail_super1(struct supertype
*st
, char *homehost
)
761 struct mdp_superblock_1
*sb
= st
->sb
;
762 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
764 int l
= homehost
? strlen(homehost
) : 0;
766 printf(" Name : %.32s", sb
->set_name
);
767 if (l
> 0 && l
< 32 &&
768 sb
->set_name
[l
] == ':' &&
769 strncmp(sb
->set_name
, homehost
, l
) == 0)
770 printf(" (local to host %s)", homehost
);
772 printf("\n Cluster Name : %-64s", bms
->cluster_name
);
773 printf("\n UUID : ");
774 for (i
=0; i
<16; i
++) {
775 if ((i
&3)==0 && i
!= 0) printf(":");
776 printf("%02x", sb
->set_uuid
[i
]);
778 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb
->events
));
781 static void brief_detail_super1(struct supertype
*st
)
783 struct mdp_superblock_1
*sb
= st
->sb
;
786 if (sb
->set_name
[0]) {
788 print_quoted(sb
->set_name
);
791 for (i
=0; i
<16; i
++) {
792 if ((i
&3)==0 && i
!= 0) printf(":");
793 printf("%02x", sb
->set_uuid
[i
]);
797 static void export_detail_super1(struct supertype
*st
)
799 struct mdp_superblock_1
*sb
= st
->sb
;
804 if (sb
->set_name
[i
] == '\n' ||
805 sb
->set_name
[i
] == '\0') {
810 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
813 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
815 struct mdp_superblock_1
*sb
= st
->sb
;
816 unsigned long long offset
;
821 if (!sb
->bblog_size
|| __le32_to_cpu(sb
->bblog_size
) > 100
822 || !sb
->bblog_offset
){
823 printf("No bad-blocks list configured on %s\n", devname
);
826 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
828 printf("Bad-blocks list is empty in %s\n", devname
);
832 size
= __le32_to_cpu(sb
->bblog_size
)* 512;
833 if (posix_memalign((void**)&bbl
, 4096, size
) != 0) {
834 pr_err("could not allocate badblocks list\n");
837 offset
= __le64_to_cpu(sb
->super_offset
) +
838 (int)__le32_to_cpu(sb
->bblog_offset
);
840 if (lseek64(fd
, offset
, 0) < 0) {
841 pr_err("Cannot seek to bad-blocks list\n");
844 if (read(fd
, bbl
, size
) != size
) {
845 pr_err("Cannot read bad-blocks list\n");
848 /* 64bits per entry. 10 bits is block-count, 54 bits is block
849 * offset. Blocks are sectors unless bblog->shift makes them bigger
852 printf("Bad-blocks on %s:\n", devname
);
853 for (i
= 0; i
< size
/8; i
++, bbp
++) {
854 __u64 bb
= __le64_to_cpu(*bbp
);
855 int count
= bb
& 0x3ff;
856 unsigned long long sector
= bb
>> 10;
861 sector
<<= sb
->bblog_shift
;
862 count
<<= sb
->bblog_shift
;
864 printf("%20llu for %d sectors\n", sector
, count
);
871 static int match_home1(struct supertype
*st
, char *homehost
)
873 struct mdp_superblock_1
*sb
= st
->sb
;
874 int l
= homehost
? strlen(homehost
) : 0;
876 return (l
> 0 && l
< 32 &&
877 sb
->set_name
[l
] == ':' &&
878 strncmp(sb
->set_name
, homehost
, l
) == 0);
881 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
883 struct mdp_superblock_1
*super
= st
->sb
;
884 char *cuuid
= (char*)uuid
;
887 cuuid
[i
] = super
->set_uuid
[i
];
890 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
892 struct mdp_superblock_1
*sb
= st
->sb
;
893 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
894 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
898 unsigned int map_disks
= info
->array
.raid_disks
;
899 unsigned long long super_offset
;
900 unsigned long long data_size
;
902 memset(info
, 0, sizeof(*info
));
903 info
->array
.major_version
= 1;
904 info
->array
.minor_version
= st
->minor_version
;
905 info
->array
.patch_version
= 0;
906 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
907 info
->array
.level
= __le32_to_cpu(sb
->level
);
908 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
909 info
->array
.md_minor
= -1;
910 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
911 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
912 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
914 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
916 if (__le32_to_cpu(bsb
->nodes
) > 1)
917 info
->array
.state
|= (1 << MD_SB_CLUSTERED
);
919 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
920 info
->component_size
= __le64_to_cpu(sb
->size
);
921 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
922 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
924 info
->disk
.major
= 0;
925 info
->disk
.minor
= 0;
926 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
927 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
928 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
929 role
= MD_DISK_ROLE_FAULTY
;
931 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
933 super_offset
= __le64_to_cpu(sb
->super_offset
);
934 if (info
->array
.level
<= 0)
935 data_size
= __le64_to_cpu(sb
->data_size
);
937 data_size
= __le64_to_cpu(sb
->size
);
938 if (info
->data_offset
< super_offset
) {
939 unsigned long long end
;
940 info
->space_before
= info
->data_offset
;
943 if (sb
->bblog_offset
&& sb
->bblog_size
) {
944 unsigned long long bboffset
= super_offset
;
945 bboffset
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
950 if (super_offset
+ info
->bitmap_offset
< end
)
951 end
= super_offset
+ info
->bitmap_offset
;
953 if (info
->data_offset
+ data_size
< end
)
954 info
->space_after
= end
- data_size
- info
->data_offset
;
956 info
->space_after
= 0;
958 unsigned long long earliest
;
959 earliest
= super_offset
+ (32+4)*2; /* match kernel */
960 if (info
->bitmap_offset
> 0) {
961 unsigned long long bmend
= info
->bitmap_offset
;
962 unsigned long long size
= __le64_to_cpu(bsb
->sync_size
);
963 size
/= __le32_to_cpu(bsb
->chunksize
) >> 9;
964 size
= (size
+ 7) >> 3;
965 size
+= sizeof(bitmap_super_t
);
966 size
= ROUND_UP(size
, 4096);
969 if (bmend
> earliest
)
972 if (sb
->bblog_offset
&& sb
->bblog_size
) {
973 unsigned long long bbend
= super_offset
;
974 bbend
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
975 bbend
+= __le32_to_cpu(sb
->bblog_size
);
976 if (bbend
> earliest
)
979 if (earliest
< info
->data_offset
)
980 info
->space_before
= info
->data_offset
- earliest
;
982 info
->space_before
= 0;
983 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
985 if (info
->space_before
== 0 && info
->space_after
== 0) {
986 /* It will look like we don't support data_offset changes,
987 * be we do - it's just that there is no room.
988 * A change that reduced the number of devices should
989 * still be allowed, so set the otherwise useless value of '1'
991 info
->space_after
= 1;
994 info
->disk
.raid_disk
= -1;
996 case MD_DISK_ROLE_SPARE
:
997 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
999 case MD_DISK_ROLE_FAULTY
:
1000 info
->disk
.state
= 1; /* faulty */
1002 case MD_DISK_ROLE_JOURNAL
:
1003 info
->disk
.state
= (1 << MD_DISK_JOURNAL
);
1004 info
->disk
.raid_disk
= role
;
1005 info
->space_after
= (misc
->device_size
- info
->data_offset
) % 8; /* journal uses all 4kB blocks*/
1008 info
->disk
.state
= 6; /* active and in sync */
1009 info
->disk
.raid_disk
= role
;
1011 if (sb
->devflags
& WriteMostly1
)
1012 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1013 info
->events
= __le64_to_cpu(sb
->events
);
1014 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
1015 info
->safe_mode_delay
= 200;
1017 memcpy(info
->uuid
, sb
->set_uuid
, 16);
1019 strncpy(info
->name
, sb
->set_name
, 32);
1022 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
1023 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
1024 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
1027 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
1028 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
1030 info
->recovery_start
= MaxSector
;
1032 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
1033 info
->reshape_active
= 1;
1034 if ((sb
->feature_map
& __le32_to_cpu(MD_FEATURE_NEW_OFFSET
)) &&
1035 sb
->new_offset
!= 0)
1036 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
1037 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
1038 info
->new_level
= __le32_to_cpu(sb
->new_level
);
1039 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
1040 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
1041 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
1042 if (info
->delta_disks
< 0)
1043 info
->array
.raid_disks
-= info
->delta_disks
;
1045 info
->reshape_active
= 0;
1047 info
->recovery_blocked
= info
->reshape_active
;
1050 for (i
=0; i
<map_disks
; i
++)
1052 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
1053 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
1054 if (/*role == MD_DISK_ROLE_SPARE || */role
< (unsigned) info
->array
.raid_disks
) {
1056 if (map
&& role
< map_disks
)
1061 info
->array
.working_disks
= working
;
1064 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
1066 struct mdinfo
*info
;
1071 info
= xmalloc(sizeof(*info
));
1072 getinfo_super1(st
, info
, NULL
);
1076 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
1078 char *devname
, int verbose
,
1079 int uuid_set
, char *homehost
)
1081 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1082 * if any change was made. For others, the return value is
1086 struct mdp_superblock_1
*sb
= st
->sb
;
1088 if (strcmp(update
, "homehost") == 0 &&
1090 /* Note that 'homehost' is special as it is really
1095 c
= strchr(sb
->set_name
, ':');
1097 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1099 strncpy(info
->name
, sb
->set_name
, 32);
1103 if (strcmp(update
, "force-one")==0) {
1104 /* Not enough devices for a working array,
1105 * so bring this one up-to-date
1107 if (sb
->events
!= __cpu_to_le64(info
->events
))
1109 sb
->events
= __cpu_to_le64(info
->events
);
1110 } else if (strcmp(update
, "force-array")==0) {
1111 /* Degraded array and 'force' requests to
1112 * maybe need to mark it 'clean'.
1114 switch(__le32_to_cpu(sb
->level
)) {
1115 case 5: case 4: case 6:
1116 /* need to force clean */
1117 if (sb
->resync_offset
!= MaxSector
)
1119 sb
->resync_offset
= MaxSector
;
1121 } else if (strcmp(update
, "assemble")==0) {
1122 int d
= info
->disk
.number
;
1124 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
1125 want
= info
->disk
.raid_disk
;
1127 want
= MD_DISK_ROLE_SPARE
;
1128 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
1129 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
1132 if (info
->reshape_active
&&
1133 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1134 info
->delta_disks
>= 0 &&
1135 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
1136 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1139 if (info
->reshape_active
&&
1140 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1141 info
->delta_disks
< 0 &&
1142 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
1143 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1146 } else if (strcmp(update
, "linear-grow-new") == 0) {
1149 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
1151 for (i
=0 ; i
< max
; i
++)
1152 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= MD_DISK_ROLE_FAULTY
)
1154 sb
->dev_number
= __cpu_to_le32(i
);
1155 info
->disk
.number
= i
;
1156 if (max
>= __le32_to_cpu(sb
->max_dev
))
1157 sb
->max_dev
= __cpu_to_le32(max
+1);
1159 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1160 read(rfd
, sb
->device_uuid
, 16) != 16) {
1161 __u32 r
[4] = {random(), random(), random(), random()};
1162 memcpy(sb
->device_uuid
, r
, 16);
1168 __cpu_to_le16(info
->disk
.raid_disk
);
1170 fd
= open(devname
, O_RDONLY
);
1172 unsigned long long ds
;
1173 get_dev_size(fd
, devname
, &ds
);
1176 if (__le64_to_cpu(sb
->super_offset
) <
1177 __le64_to_cpu(sb
->data_offset
)) {
1178 sb
->data_size
= __cpu_to_le64(
1179 ds
- __le64_to_cpu(sb
->data_offset
));
1182 ds
&= ~(unsigned long long)(4*2-1);
1183 sb
->super_offset
= __cpu_to_le64(ds
);
1184 sb
->data_size
= __cpu_to_le64(
1185 ds
- __le64_to_cpu(sb
->data_offset
));
1188 } else if (strcmp(update
, "linear-grow-update") == 0) {
1189 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1190 sb
->dev_roles
[info
->disk
.number
] =
1191 __cpu_to_le16(info
->disk
.raid_disk
);
1192 } else if (strcmp(update
, "resync") == 0) {
1193 /* make sure resync happens */
1194 sb
->resync_offset
= 0ULL;
1195 } else if (strcmp(update
, "uuid") == 0) {
1196 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1198 if (__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1199 struct bitmap_super_s
*bm
;
1200 bm
= (struct bitmap_super_s
*)(st
->sb
+MAX_SB_SIZE
);
1201 memcpy(bm
->uuid
, sb
->set_uuid
, 16);
1203 } else if (strcmp(update
, "no-bitmap") == 0) {
1204 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1205 } else if (strcmp(update
, "bbl") == 0) {
1206 /* only possible if there is room after the bitmap, or if
1207 * there is no bitmap
1209 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1210 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1211 long bitmap_offset
= (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
);
1212 long bm_sectors
= 0;
1216 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1217 struct bitmap_super_s
*bsb
;
1218 bsb
= (struct bitmap_super_s
*)(((char*)sb
)+MAX_SB_SIZE
);
1219 bm_sectors
= bitmap_sectors(bsb
);
1222 if (sb_offset
< data_offset
) {
1223 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1226 bb_offset
= sb_offset
+ 8;
1227 if (bm_sectors
&& bitmap_offset
> 0)
1228 bb_offset
= bitmap_offset
+ bm_sectors
;
1229 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1230 && bb_offset
+ 8+8 <= (long)data_offset
)
1231 /* too close to bitmap, and room to grow */
1233 if (bb_offset
+ 8 <= (long)data_offset
) {
1234 sb
->bblog_size
= __cpu_to_le16(8);
1235 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1238 /* 1.0 - Put bbl just before super block */
1239 if (bm_sectors
&& bitmap_offset
< 0)
1240 space
= -bitmap_offset
- bm_sectors
;
1242 space
= sb_offset
- data_offset
-
1243 __le64_to_cpu(sb
->data_size
);
1245 sb
->bblog_size
= __cpu_to_le16(8);
1246 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1249 } else if (strcmp(update
, "no-bbl") == 0) {
1250 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1251 pr_err("Cannot remove active bbl from %s\n",devname
);
1254 sb
->bblog_shift
= 0;
1255 sb
->bblog_offset
= 0;
1257 } else if (strcmp(update
, "name") == 0) {
1258 if (info
->name
[0] == 0)
1259 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1260 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1262 strchr(info
->name
, ':') == NULL
&&
1263 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1264 strcpy(sb
->set_name
, homehost
);
1265 strcat(sb
->set_name
, ":");
1266 strcat(sb
->set_name
, info
->name
);
1268 strcpy(sb
->set_name
, info
->name
);
1269 } else if (strcmp(update
, "devicesize") == 0 &&
1270 __le64_to_cpu(sb
->super_offset
) <
1271 __le64_to_cpu(sb
->data_offset
)) {
1272 /* set data_size to device size less data_offset */
1273 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1274 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1275 sb
->data_size
= __cpu_to_le64(
1276 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1277 } else if (strcmp(update
, "revert-reshape") == 0) {
1279 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1280 pr_err("No active reshape to revert on %s\n",
1284 unsigned long long reshape_sectors
;
1287 /* reshape_position is a little messy.
1288 * Its value must be a multiple of the larger
1289 * chunk size, and of the "after" data disks.
1290 * So when reverting we need to change it to
1291 * be a multiple of the new "after" data disks,
1292 * which is the old "before".
1293 * If it isn't already a multiple of 'before',
1294 * the only thing we could do would be
1295 * copy some block around on the disks, which
1296 * is easy to get wrong.
1297 * So we reject a revert-reshape unless the
1298 * alignment is good.
1300 if (__le32_to_cpu(sb
->level
) >= 4 &&
1301 __le32_to_cpu(sb
->level
) <= 6) {
1302 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1303 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1304 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1305 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1306 if (reshape_sectors
% reshape_chunk
) {
1307 pr_err("Reshape position is not suitably aligned.\n");
1308 pr_err("Try normal assembly and stop again\n");
1312 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1313 __le32_to_cpu(sb
->delta_disks
));
1314 if (sb
->delta_disks
== 0)
1315 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1317 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1319 temp
= sb
->new_layout
;
1320 sb
->new_layout
= sb
->layout
;
1323 temp
= sb
->new_chunk
;
1324 sb
->new_chunk
= sb
->chunksize
;
1325 sb
->chunksize
= temp
;
1327 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1328 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1329 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1330 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1331 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1334 } else if (strcmp(update
, "_reshape_progress")==0)
1335 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1336 else if (strcmp(update
, "writemostly")==0)
1337 sb
->devflags
|= WriteMostly1
;
1338 else if (strcmp(update
, "readwrite")==0)
1339 sb
->devflags
&= ~WriteMostly1
;
1343 sb
->sb_csum
= calc_sb_1_csum(sb
);
1347 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1348 unsigned long long size
, char *name
, char *homehost
,
1349 int *uuid
, unsigned long long data_offset
)
1351 struct mdp_superblock_1
*sb
;
1357 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1358 pr_err("could not allocate superblock\n");
1361 memset(sb
, 0, SUPER1_SIZE
);
1365 /* zeroing superblock */
1369 spares
= info
->working_disks
- info
->active_disks
;
1370 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1371 pr_err("too many devices requested: %d+%d > %d\n",
1372 info
->raid_disks
, spares
, MAX_DEVS
);
1376 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1377 sb
->major_version
= __cpu_to_le32(1);
1378 sb
->feature_map
= 0;
1382 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1384 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1385 read(rfd
, sb
->set_uuid
, 16) != 16) {
1386 __u32 r
[4] = {random(), random(), random(), random()};
1387 memcpy(sb
->set_uuid
, r
, 16);
1389 if (rfd
>= 0) close(rfd
);
1392 if (name
== NULL
|| *name
== 0) {
1393 sprintf(defname
, "%d", info
->md_minor
);
1397 strchr(name
, ':')== NULL
&&
1398 strlen(homehost
)+1+strlen(name
) < 32) {
1399 strcpy(sb
->set_name
, homehost
);
1400 strcat(sb
->set_name
, ":");
1401 strcat(sb
->set_name
, name
);
1403 strcpy(sb
->set_name
, name
);
1405 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1406 sb
->level
= __cpu_to_le32(info
->level
);
1407 sb
->layout
= __cpu_to_le32(info
->layout
);
1408 sb
->size
= __cpu_to_le64(size
*2ULL);
1409 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1410 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1412 sb
->data_offset
= __cpu_to_le64(data_offset
);
1413 sb
->data_size
= __cpu_to_le64(0);
1414 sb
->super_offset
= __cpu_to_le64(0);
1415 sb
->recovery_offset
= __cpu_to_le64(0);
1417 sb
->utime
= sb
->ctime
;
1418 sb
->events
= __cpu_to_le64(1);
1419 if (info
->state
& (1<<MD_SB_CLEAN
))
1420 sb
->resync_offset
= MaxSector
;
1422 sb
->resync_offset
= 0;
1423 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1424 sbsize
= ROUND_UP(sbsize
, 512);
1425 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1427 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1435 long long data_offset
;
1436 mdu_disk_info_t disk
;
1437 struct devinfo
*next
;
1440 /* Add a device to the superblock being created */
1441 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1442 int fd
, char *devname
, unsigned long long data_offset
)
1444 struct mdp_superblock_1
*sb
= st
->sb
;
1445 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1446 struct devinfo
*di
, **dip
;
1448 if ((dk
->state
& 6) == 6) /* active, sync */
1449 *rp
= __cpu_to_le16(dk
->raid_disk
);
1450 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1451 *rp
= MD_DISK_ROLE_SPARE
;
1453 *rp
= MD_DISK_ROLE_FAULTY
;
1455 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1456 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1457 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1459 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1460 sb
->devflags
= 0; /* don't copy another disks flags */
1461 sb
->sb_csum
= calc_sb_1_csum(sb
);
1463 dip
= (struct devinfo
**)&st
->info
;
1465 dip
= &(*dip
)->next
;
1466 di
= xmalloc(sizeof(struct devinfo
));
1468 di
->devname
= devname
;
1470 di
->data_offset
= data_offset
;
1478 static void locate_bitmap1(struct supertype
*st
, int fd
);
1480 static int store_super1(struct supertype
*st
, int fd
)
1482 struct mdp_superblock_1
*sb
= st
->sb
;
1483 unsigned long long sb_offset
;
1484 struct align_fd afd
;
1486 unsigned long long dsize
;
1488 if (!get_dev_size(fd
, NULL
, &dsize
))
1499 * Calculate the position of the superblock.
1500 * It is always aligned to a 4K boundary and
1501 * depending on minor_version, it can be:
1502 * 0: At least 8K, but less than 12K, from end of device
1503 * 1: At start of device
1504 * 2: 4K from start of device.
1506 switch(st
->minor_version
) {
1510 sb_offset
&= ~(4*2-1);
1522 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1523 0 != __le64_to_cpu(sb
->super_offset
)
1525 pr_err("internal error - sb_offset is wrong\n");
1529 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1532 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1534 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1537 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1538 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1539 (((char*)sb
)+MAX_SB_SIZE
);
1540 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1541 locate_bitmap1(st
, fd
);
1542 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1550 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1552 static unsigned long choose_bm_space(unsigned long devsize
)
1554 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1555 * if bigger than 200Gig, save 128k
1556 * NOTE: result must be multiple of 4K else bad things happen
1557 * on 4K-sector devices.
1559 if (devsize
< 64*2) return 0;
1560 if (devsize
- 64*2 >= 200*1024*1024*2)
1562 if (devsize
- 4*2 > 8*1024*1024*2)
1567 static void free_super1(struct supertype
*st
);
1570 static int write_init_super1(struct supertype
*st
)
1572 struct mdp_superblock_1
*sb
= st
->sb
;
1573 struct supertype
*refst
;
1576 unsigned long long bm_space
;
1578 unsigned long long dsize
, array_size
;
1579 unsigned long long sb_offset
;
1580 unsigned long long data_offset
;
1582 for (di
= st
->info
; di
; di
= di
->next
) {
1583 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1588 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1591 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1592 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1593 sb
->devflags
|= WriteMostly1
;
1595 sb
->devflags
&= ~WriteMostly1
;
1597 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1598 read(rfd
, sb
->device_uuid
, 16) != 16) {
1599 __u32 r
[4] = {random(), random(), random(), random()};
1600 memcpy(sb
->device_uuid
, r
, 16);
1607 refst
= dup_super(st
);
1608 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1609 struct mdp_superblock_1
*refsb
= refst
->sb
;
1611 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1612 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1613 /* same array, so preserve events and
1615 sb
->events
= refsb
->events
;
1616 /* bugs in 2.6.17 and earlier mean the
1617 * dev_number chosen in Manage must be preserved
1619 if (get_linux_version() >= 2006018)
1620 sb
->dev_number
= refsb
->dev_number
;
1626 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1639 * Calculate the position of the superblock.
1640 * It is always aligned to a 4K boundary and
1641 * depending on minor_version, it can be:
1642 * 0: At least 8K, but less than 12K, from end of device
1643 * 1: At start of device
1644 * 2: 4K from start of device.
1645 * data_offset has already been set.
1647 array_size
= __le64_to_cpu(sb
->size
);
1648 /* work out how much space we left for a bitmap,
1649 * Add 8 sectors for bad block log */
1650 bm_space
= choose_bm_space(array_size
) + 8;
1652 data_offset
= di
->data_offset
;
1653 if (data_offset
== INVALID_SECTORS
)
1654 data_offset
= st
->data_offset
;
1655 switch(st
->minor_version
) {
1657 if (data_offset
== INVALID_SECTORS
)
1661 sb_offset
&= ~(4*2-1);
1662 sb
->data_offset
= __cpu_to_le64(data_offset
);
1663 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1664 if (sb_offset
< array_size
+ bm_space
)
1665 bm_space
= sb_offset
- array_size
;
1666 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1667 if (bm_space
>= 8) {
1668 sb
->bblog_size
= __cpu_to_le16(8);
1669 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1673 sb
->super_offset
= __cpu_to_le64(0);
1674 if (data_offset
== INVALID_SECTORS
)
1677 sb
->data_offset
= __cpu_to_le64(data_offset
);
1678 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1679 if (data_offset
>= 8 + 32*2 + 8) {
1680 sb
->bblog_size
= __cpu_to_le16(8);
1681 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1682 } else if (data_offset
>= 16) {
1683 sb
->bblog_size
= __cpu_to_le16(8);
1684 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8);
1689 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1690 if (data_offset
== INVALID_SECTORS
)
1693 sb
->data_offset
= __cpu_to_le64(data_offset
);
1694 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1695 if (data_offset
>= 16 + 32*2 + 8) {
1696 sb
->bblog_size
= __cpu_to_le16(8);
1697 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1698 } else if (data_offset
>= 16+16) {
1699 sb
->bblog_size
= __cpu_to_le16(8);
1700 /* '8' sectors for the bblog, and another '8'
1701 * because we want offset from superblock, not
1704 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8-8);
1708 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1709 st
->minor_version
, di
->devname
);
1713 /* Disable badblock log on clusters, or when explicitly requested */
1714 if (st
->nodes
> 0 || conf_get_create_info()->bblist
== 0) {
1716 sb
->bblog_offset
= 0;
1719 sb
->sb_csum
= calc_sb_1_csum(sb
);
1720 rv
= store_super1(st
, di
->fd
);
1721 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1722 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NoUpdate
);
1730 pr_err("Failed to write metadata to %s\n",
1737 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1741 * 0 same, or first was empty, and second was copied
1742 * 1 second had wrong number
1744 * 3 wrong other info
1746 struct mdp_superblock_1
*first
= st
->sb
;
1747 struct mdp_superblock_1
*second
= tst
->sb
;
1749 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1751 if (second
->major_version
!= __cpu_to_le32(1))
1755 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1756 pr_err("could not allocate superblock\n");
1759 memcpy(first
, second
, SUPER1_SIZE
);
1763 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1766 if (first
->ctime
!= second
->ctime
||
1767 first
->level
!= second
->level
||
1768 first
->layout
!= second
->layout
||
1769 first
->size
!= second
->size
||
1770 first
->chunksize
!= second
->chunksize
||
1771 first
->raid_disks
!= second
->raid_disks
)
1776 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1778 unsigned long long dsize
;
1779 unsigned long long sb_offset
;
1780 struct mdp_superblock_1
*super
;
1782 struct bitmap_super_s
*bsb
;
1783 struct misc_dev_info
*misc
;
1784 struct align_fd afd
;
1790 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1792 struct supertype tst
;
1793 __u64 bestctime
= 0;
1794 /* guess... choose latest ctime */
1795 memset(&tst
, 0, sizeof(tst
));
1797 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1798 switch(load_super1(&tst
, fd
, devname
)) {
1799 case 0: super
= tst
.sb
;
1800 if (bestvers
== -1 ||
1801 bestctime
< __le64_to_cpu(super
->ctime
)) {
1802 bestvers
= tst
.minor_version
;
1803 bestctime
= __le64_to_cpu(super
->ctime
);
1808 case 1: return 1; /*bad device */
1809 case 2: break; /* bad, try next */
1812 if (bestvers
!= -1) {
1814 tst
.minor_version
= bestvers
;
1816 tst
.max_devs
= MAX_DEVS
;
1817 rv
= load_super1(&tst
, fd
, devname
);
1824 if (!get_dev_size(fd
, devname
, &dsize
))
1830 pr_err("%s is too small for md: size is %llu sectors.\n",
1836 * Calculate the position of the superblock.
1837 * It is always aligned to a 4K boundary and
1838 * depending on minor_version, it can be:
1839 * 0: At least 8K, but less than 12K, from end of device
1840 * 1: At start of device
1841 * 2: 4K from start of device.
1843 switch(st
->minor_version
) {
1847 sb_offset
&= ~(4*2-1);
1859 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1861 pr_err("Cannot seek to superblock on %s: %s\n",
1862 devname
, strerror(errno
));
1866 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
1867 pr_err("could not allocate superblock\n");
1871 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
1873 pr_err("Cannot read superblock on %s\n",
1879 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
1881 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
1882 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
1887 if (__le32_to_cpu(super
->major_version
) != 1) {
1889 pr_err("Cannot interpret superblock on %s - version is %d\n",
1890 devname
, __le32_to_cpu(super
->major_version
));
1894 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
1896 pr_err("No superblock found on %s (super_offset is wrong)\n",
1903 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
1905 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
1906 misc
->device_size
= dsize
;
1907 if (st
->data_offset
== INVALID_SECTORS
)
1908 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
1910 /* Now check on the bitmap superblock */
1911 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
1913 /* Read the bitmap superblock and make sure it looks
1914 * valid. If it doesn't clear the bit. An --assemble --force
1915 * should get that written out.
1917 locate_bitmap1(st
, fd
);
1918 if (aread(&afd
, bsb
, 512) != 512)
1921 uuid_from_super1(st
, uuid
);
1922 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
1923 memcmp(bsb
->uuid
, uuid
, 16) != 0)
1928 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
1929 & ~MD_FEATURE_BITMAP_OFFSET
);
1933 static struct supertype
*match_metadata_desc1(char *arg
)
1935 struct supertype
*st
= xcalloc(1, sizeof(*st
));
1937 st
->container_devnm
[0] = 0;
1939 st
->max_devs
= MAX_DEVS
;
1941 st
->data_offset
= INVALID_SECTORS
;
1942 /* leading zeros can be safely ignored. --detail generates them. */
1945 if (strcmp(arg
, "1.0") == 0 ||
1946 strcmp(arg
, "1.00") == 0) {
1947 st
->minor_version
= 0;
1950 if (strcmp(arg
, "1.1") == 0 ||
1951 strcmp(arg
, "1.01") == 0
1953 st
->minor_version
= 1;
1956 if (strcmp(arg
, "1.2") == 0 ||
1957 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
1958 strcmp(arg
, "default") == 0 ||
1959 #endif /* DEFAULT_OLD_METADATA */
1960 strcmp(arg
, "1.02") == 0) {
1961 st
->minor_version
= 2;
1964 if (strcmp(arg
, "1") == 0 ||
1965 strcmp(arg
, "default") == 0) {
1966 st
->minor_version
= -1;
1974 /* find available size on device with this devsize, using
1975 * superblock type st, and reserving 'reserve' sectors for
1978 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
1979 unsigned long long data_offset
)
1981 struct mdp_superblock_1
*super
= st
->sb
;
1988 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1989 /* hot-add. allow for actual size of bitmap */
1990 struct bitmap_super_s
*bsb
;
1991 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
1992 bmspace
= bitmap_sectors(bsb
);
1995 /* Allow space for bad block log */
1996 if (super
->bblog_size
)
1997 bbspace
= __le16_to_cpu(super
->bblog_size
);
1999 if (st
->minor_version
< 0)
2000 /* not specified, so time to set default */
2001 st
->minor_version
= 2;
2003 if (data_offset
== INVALID_SECTORS
)
2004 data_offset
= st
->data_offset
;
2006 if (data_offset
!= INVALID_SECTORS
)
2007 switch(st
->minor_version
) {
2009 return devsize
- data_offset
- 8*2 - bbspace
;
2012 return devsize
- data_offset
;
2019 switch(st
->minor_version
) {
2022 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2024 /* at start, 4K for superblock and possible bitmap */
2025 return devsize
- 4*2 - bbspace
;
2027 /* 4k from start, 4K for superblock and possible bitmap */
2028 return devsize
- (4+4)*2 - bbspace
;
2034 add_internal_bitmap1(struct supertype
*st
,
2035 int *chunkp
, int delay
, int write_behind
,
2036 unsigned long long size
,
2037 int may_change
, int major
)
2040 * If not may_change, then this is a 'Grow' without sysfs support for
2041 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2042 * If may_change, then this is create or a Grow with sysfs syupport,
2043 * and we can put the bitmap wherever we like.
2045 * size is in sectors, chunk is in bytes !!!
2048 unsigned long long bits
;
2049 unsigned long long max_bits
;
2050 unsigned long long min_chunk
;
2052 long bbl_offset
, bbl_size
;
2053 unsigned long long chunk
= *chunkp
;
2056 struct mdp_superblock_1
*sb
= st
->sb
;
2057 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2060 if (__le64_to_cpu(sb
->data_size
) == 0)
2061 /* Must be creating the array, else data_size would be non-zero */
2063 switch(st
->minor_version
) {
2065 /* either 3K after the superblock (when hot-add),
2066 * or some amount of space before.
2069 /* We are creating array, so we *know* how much room has
2074 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2076 room
= __le64_to_cpu(sb
->super_offset
)
2077 - __le64_to_cpu(sb
->data_offset
)
2078 - __le64_to_cpu(sb
->data_size
);
2079 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2082 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2083 if (bbl_size
< -bbl_offset
)
2084 bbl_size
= -bbl_offset
;
2086 if (!may_change
|| (room
< 3*2 &&
2087 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2092 offset
= 0; /* means movable offset */
2097 case 2: /* between superblock and data */
2101 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2103 room
= __le64_to_cpu(sb
->data_offset
)
2104 - __le64_to_cpu(sb
->super_offset
);
2105 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2107 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2112 room
-= 2; /* Leave 1K for superblock */
2116 room
-= 4*2; /* leave 4K for superblock */
2126 if (chunk
== UnSet
&& room
> 128*2)
2127 /* Limit to 128K of bitmap when chunk size not requested */
2131 /* No room for a bitmap */
2134 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2136 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2137 bits
= (size
*512)/min_chunk
+1;
2138 while (bits
> max_bits
) {
2142 if (chunk
== UnSet
) {
2143 /* For practical purpose, 64Meg is a good
2144 * default chunk size for internal bitmaps.
2147 if (chunk
< 64*1024*1024)
2148 chunk
= 64*1024*1024;
2149 } else if (chunk
< min_chunk
)
2150 return 0; /* chunk size too small */
2151 if (chunk
== 0) /* rounding problem */
2155 /* start bitmap on a 4K boundary with enough space for
2158 bits
= (size
*512) / chunk
+ 1;
2159 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2160 room
*= 8; /* convert 4K blocks to sectors */
2161 offset
= -room
- bbl_size
;
2164 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2166 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2167 | MD_FEATURE_BITMAP_OFFSET
);
2168 memset(bms
, 0, sizeof(*bms
));
2169 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2170 bms
->version
= __cpu_to_le32(major
);
2171 uuid_from_super1(st
, uuid
);
2172 memcpy(bms
->uuid
, uuid
, 16);
2173 bms
->chunksize
= __cpu_to_le32(chunk
);
2174 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2175 bms
->sync_size
= __cpu_to_le64(size
);
2176 bms
->write_behind
= __cpu_to_le32(write_behind
);
2177 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2179 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2180 | MD_FEATURE_BITMAP_VERSIONED
);
2181 if (st
->cluster_name
)
2182 strncpy((char *)bms
->cluster_name
,
2183 st
->cluster_name
, strlen(st
->cluster_name
));
2189 static void locate_bitmap1(struct supertype
*st
, int fd
)
2191 unsigned long long offset
;
2192 struct mdp_superblock_1
*sb
;
2196 if (st
->ss
->load_super(st
, fd
, NULL
))
2197 return; /* no error I hope... */
2202 offset
= __le64_to_cpu(sb
->super_offset
);
2203 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
);
2206 lseek64(fd
, offset
<<9, 0);
2209 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2211 struct mdp_superblock_1
*sb
= st
->sb
;
2212 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2216 struct align_fd afd
;
2218 unsigned long long total_bm_space
, bm_space_per_node
;
2222 /* update cluster name */
2223 if (st
->cluster_name
) {
2224 memset((char *)bms
->cluster_name
, 0, sizeof(bms
->cluster_name
));
2225 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, 64);
2229 /* cluster md only supports superblock 1.2 now */
2230 if (st
->minor_version
!= 2) {
2231 pr_err("Warning: cluster md only works with superblock 1.2\n");
2235 /* Each node has an independent bitmap, it is necessary to calculate the
2236 * space is enough or not, first get how many bytes for the total bitmap */
2237 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2239 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2240 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2242 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2243 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2244 total_bm_space
/ bm_space_per_node
);
2248 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2257 locate_bitmap1(st
, fd
);
2259 if (posix_memalign(&buf
, 4096, 4096))
2263 /* Only the bitmap[0] should resync
2264 * whole device on initial assembly
2267 memset(buf
, 0x00, 4096);
2269 memset(buf
, 0xff, 4096);
2270 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2272 towrite
= calc_bitmap_size(bms
, 4096);
2273 while (towrite
> 0) {
2277 n
= awrite(&afd
, buf
, n
);
2283 memset(buf
, 0x00, 4096);
2285 memset(buf
, 0xff, 4096);
2292 } while (++i
< __le32_to_cpu(bms
->nodes
));
2298 static void free_super1(struct supertype
*st
)
2303 struct devinfo
*di
= st
->info
;
2304 st
->info
= di
->next
;
2313 static int validate_geometry1(struct supertype
*st
, int level
,
2314 int layout
, int raiddisks
,
2315 int *chunk
, unsigned long long size
,
2316 unsigned long long data_offset
,
2317 char *subdev
, unsigned long long *freesize
,
2320 unsigned long long ldsize
, devsize
;
2322 unsigned long long headroom
;
2325 if (level
== LEVEL_CONTAINER
) {
2327 pr_err("1.x metadata does not support containers\n");
2330 if (*chunk
== UnSet
)
2331 *chunk
= DEFAULT_CHUNK
;
2336 if (st
->minor_version
< 0)
2337 /* not specified, so time to set default */
2338 st
->minor_version
= 2;
2340 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2343 pr_err("super1.x cannot open %s: %s\n",
2344 subdev
, strerror(errno
));
2348 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2354 devsize
= ldsize
>> 9;
2360 /* creating: allow suitable space for bitmap */
2361 bmspace
= choose_bm_space(devsize
);
2363 if (data_offset
== INVALID_SECTORS
)
2364 data_offset
= st
->data_offset
;
2365 if (data_offset
== INVALID_SECTORS
)
2366 switch (st
->minor_version
) {
2372 /* Choose data offset appropriate for this device
2373 * and use as default for whole array.
2374 * The data_offset must allow for bitmap space
2375 * and base metadata, should allow for some headroom
2376 * for reshape, and should be rounded to multiple
2378 * Headroom is limited to 128M, but aim for about 0.1%
2380 headroom
= 128*1024*2;
2381 while ((headroom
<< 10) > devsize
&&
2383 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2385 data_offset
= 12*2 + bmspace
+ headroom
;
2386 #define ONE_MEG (2*1024)
2387 if (data_offset
> ONE_MEG
)
2388 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2391 if (st
->data_offset
== INVALID_SECTORS
)
2392 st
->data_offset
= data_offset
;
2393 switch(st
->minor_version
) {
2394 case 0: /* metadata at end. Round down and subtract space to reserve */
2395 devsize
= (devsize
& ~(4ULL*2-1));
2396 /* space for metadata, bblog, bitmap */
2397 devsize
-= 8*2 + 8 + bmspace
;
2401 devsize
-= data_offset
;
2404 *freesize
= devsize
;
2407 #endif /* MDASSEMBLE */
2409 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2411 /* Create a v1.0 superblock based on 'info'*/
2413 struct mdp_superblock_1
*sb
;
2416 unsigned long long offset
;
2418 if (posix_memalign(&ret
, 4096, 1024) != 0)
2421 memset(ret
, 0, 1024);
2422 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2423 sb
->major_version
= __cpu_to_le32(1);
2425 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2426 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2427 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2428 sb
->level
= __cpu_to_le32(info
->array
.level
);
2429 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2430 sb
->size
= __cpu_to_le64(info
->component_size
);
2431 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2432 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2433 if (info
->array
.level
> 0)
2434 sb
->data_size
= sb
->size
;
2436 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2437 sb
->resync_offset
= MaxSector
;
2438 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2439 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2440 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2442 offset
= st
->devsize
/512 - 8*2;
2444 sb
->super_offset
= __cpu_to_le64(offset
);
2445 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2447 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
2448 read(rfd
, sb
->device_uuid
, 16) != 16) {
2449 __u32 r
[4] = {random(), random(), random(), random()};
2450 memcpy(sb
->device_uuid
, r
, 16);
2455 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2456 int state
= sb0
->disks
[i
].state
;
2457 sb
->dev_roles
[i
] = MD_DISK_ROLE_SPARE
;
2458 if ((state
& (1<<MD_DISK_SYNC
)) &&
2459 !(state
& (1<<MD_DISK_FAULTY
)))
2460 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2462 sb
->sb_csum
= calc_sb_1_csum(sb
);
2466 struct superswitch super1
= {
2468 .examine_super
= examine_super1
,
2469 .brief_examine_super
= brief_examine_super1
,
2470 .export_examine_super
= export_examine_super1
,
2471 .detail_super
= detail_super1
,
2472 .brief_detail_super
= brief_detail_super1
,
2473 .export_detail_super
= export_detail_super1
,
2474 .write_init_super
= write_init_super1
,
2475 .validate_geometry
= validate_geometry1
,
2476 .add_to_super
= add_to_super1
,
2477 .examine_badblocks
= examine_badblocks_super1
,
2478 .copy_metadata
= copy_metadata1
,
2480 .match_home
= match_home1
,
2481 .uuid_from_super
= uuid_from_super1
,
2482 .getinfo_super
= getinfo_super1
,
2483 .container_content
= container_content1
,
2484 .update_super
= update_super1
,
2485 .init_super
= init_super1
,
2486 .store_super
= store_super1
,
2487 .compare_super
= compare_super1
,
2488 .load_super
= load_super1
,
2489 .match_metadata_desc
= match_metadata_desc1
,
2490 .avail_size
= avail_size1
,
2491 .add_internal_bitmap
= add_internal_bitmap1
,
2492 .locate_bitmap
= locate_bitmap1
,
2493 .write_bitmap
= write_bitmap1
,
2494 .free_super
= free_super1
,
2495 #if __BYTE_ORDER == BIG_ENDIAN