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_ALL (MD_FEATURE_BITMAP_OFFSET \
130 |MD_FEATURE_RECOVERY_OFFSET \
131 |MD_FEATURE_RESHAPE_ACTIVE \
132 |MD_FEATURE_BAD_BLOCKS \
133 |MD_FEATURE_REPLACEMENT \
134 |MD_FEATURE_RESHAPE_BACKWARDS \
135 |MD_FEATURE_NEW_OFFSET \
136 |MD_FEATURE_BITMAP_VERSIONED \
139 /* return how many bytes are needed for bitmap, for cluster-md each node
140 * should have it's own bitmap */
141 static unsigned int calc_bitmap_size(bitmap_super_t
*bms
, unsigned int boundary
)
143 unsigned long long bits
, bytes
;
145 bits
= __le64_to_cpu(bms
->sync_size
) / (__le32_to_cpu(bms
->chunksize
)>>9);
146 bytes
= (bits
+7) >> 3;
147 bytes
+= sizeof(bitmap_super_t
);
148 bytes
= ROUND_UP(bytes
, boundary
);
153 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
155 unsigned int disk_csum
, csum
;
156 unsigned long long newcsum
;
157 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
158 unsigned int *isuper
= (unsigned int*)sb
;
160 /* make sure I can count... */
161 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
162 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
163 sizeof(struct mdp_superblock_1
) != 256) {
164 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
167 disk_csum
= sb
->sb_csum
;
170 for (; size
>=4; size
-= 4 ) {
171 newcsum
+= __le32_to_cpu(*isuper
);
176 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
178 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
179 sb
->sb_csum
= disk_csum
;
180 return __cpu_to_le32(csum
);
184 * Information related to file descriptor used for aligned reads/writes.
185 * Cache the block size.
192 static void init_afd(struct align_fd
*afd
, int fd
)
196 if (ioctl(afd
->fd
, BLKSSZGET
, &afd
->blk_sz
) != 0)
200 static char abuf
[4096+4096];
201 static int aread(struct align_fd
*afd
, void *buf
, int len
)
204 * On devices with a 4K sector size, we need to read
205 * the full sector and copy relevant bits into
214 if (!bsize
|| bsize
> 4096 || len
> 4096) {
216 fprintf(stderr
, "WARNING - aread() called with invalid block size\n");
219 b
= ROUND_UP_PTR((char *)abuf
, 4096);
221 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
223 n
= read(afd
->fd
, b
, iosize
);
226 lseek(afd
->fd
, len
- n
, 1);
233 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
236 * On devices with a 4K sector size, we need to write
237 * the full sector. We pre-read if the sector is larger
239 * The address must be sector-aligned.
246 if (!bsize
|| bsize
> 4096 || len
> 4096) {
248 fprintf(stderr
, "WARNING - awrite() called with invalid block size\n");
251 b
= ROUND_UP_PTR((char *)abuf
, 4096);
253 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
257 n
= read(afd
->fd
, b
, iosize
);
260 lseek(afd
->fd
, -n
, 1);
264 n
= write(afd
->fd
, b
, iosize
);
267 lseek(afd
->fd
, len
- n
, 1);
272 static void examine_super1(struct supertype
*st
, char *homehost
)
274 struct mdp_superblock_1
*sb
= st
->sb
;
275 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
282 int l
= homehost
? strlen(homehost
) : 0;
284 unsigned long long sb_offset
;
287 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
288 printf(" Version : 1");
289 sb_offset
= __le64_to_cpu(sb
->super_offset
);
292 else if (sb_offset
<= 8)
296 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
297 printf(" Array UUID : ");
298 for (i
=0; i
<16; i
++) {
299 if ((i
&3)==0 && i
!= 0) printf(":");
300 printf("%02x", sb
->set_uuid
[i
]);
303 printf(" Name : %.32s", sb
->set_name
);
304 if (l
> 0 && l
< 32 &&
305 sb
->set_name
[l
] == ':' &&
306 strncmp(sb
->set_name
, homehost
, l
) == 0)
307 printf(" (local to host %s)", homehost
);
310 printf(" Cluster Name : %-64s\n", bms
->cluster_name
);
311 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
312 printf(" Creation Time : %.24s\n", ctime(&atime
));
313 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
314 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
315 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
317 printf(" Avail Dev Size : %llu%s\n",
318 (unsigned long long)__le64_to_cpu(sb
->data_size
),
319 human_size(__le64_to_cpu(sb
->data_size
)<<9));
320 if (__le32_to_cpu(sb
->level
) > 0) {
321 int ddsks
= 0, ddsks_denom
= 1;
322 switch(__le32_to_cpu(sb
->level
)) {
323 case 1: ddsks
=1;break;
325 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
326 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
328 layout
= __le32_to_cpu(sb
->layout
);
329 ddsks
= __le32_to_cpu(sb
->raid_disks
);
330 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
333 long long asize
= __le64_to_cpu(sb
->size
);
334 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
335 printf(" Array Size : %llu%s\n",
336 asize
>> 10, human_size(asize
));
338 if (sb
->size
!= sb
->data_size
)
339 printf(" Used Dev Size : %llu%s\n",
340 (unsigned long long)__le64_to_cpu(sb
->size
),
341 human_size(__le64_to_cpu(sb
->size
)<<9));
344 printf(" Data Offset : %llu sectors\n",
345 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
346 if (sb
->new_offset
&&
347 (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
)) {
348 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
349 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
350 printf(" New Offset : %llu sectors\n", offset
);
352 printf(" Super Offset : %llu sectors\n",
353 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
354 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
355 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
357 st
->ss
->getinfo_super(st
, &info
, NULL
);
358 if (info
.space_after
!= 1 &&
359 !(__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
360 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
361 info
.space_before
, info
.space_after
);
363 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
364 printf(" Device UUID : ");
365 for (i
=0; i
<16; i
++) {
366 if ((i
&3)==0 && i
!= 0) printf(":");
367 printf("%02x", sb
->device_uuid
[i
]);
371 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
372 printf("Internal Bitmap : %ld sectors from superblock\n",
373 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
375 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
376 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
377 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
378 if (__le32_to_cpu(sb
->delta_disks
)) {
379 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
380 printf(" (%d->%d)\n",
381 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
382 __le32_to_cpu(sb
->raid_disks
));
383 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
384 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
386 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
387 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
388 printf(" New Level : %s\n", c
?c
:"-unknown-");
390 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
391 if (__le32_to_cpu(sb
->level
) == 5) {
392 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
393 printf(" New Layout : %s\n", c
?c
:"-unknown-");
395 if (__le32_to_cpu(sb
->level
) == 6) {
396 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
397 printf(" New Layout : %s\n", c
?c
:"-unknown-");
399 if (__le32_to_cpu(sb
->level
) == 10) {
400 printf(" New Layout :");
401 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
405 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
406 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
411 if (sb
->devflags
& WriteMostly1
)
412 printf(" write-mostly");
416 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
417 printf(" Update Time : %.24s\n", ctime(&atime
));
419 if (sb
->bblog_size
&& sb
->bblog_offset
) {
420 printf(" Bad Block Log : %d entries available at offset %ld sectors",
421 __le16_to_cpu(sb
->bblog_size
)*512/8,
422 (long)(int32_t)__le32_to_cpu(sb
->bblog_offset
));
423 if (sb
->feature_map
&
424 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
425 printf(" - bad blocks present.");
429 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
430 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
432 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
433 __le32_to_cpu(calc_sb_1_csum(sb
)));
434 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
436 if (__le32_to_cpu(sb
->level
) == 5) {
437 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
438 printf(" Layout : %s\n", c
?c
:"-unknown-");
440 if (__le32_to_cpu(sb
->level
) == 6) {
441 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
442 printf(" Layout : %s\n", c
?c
:"-unknown-");
444 if (__le32_to_cpu(sb
->level
) == 10) {
445 int lo
= __le32_to_cpu(sb
->layout
);
447 print_r10_layout(lo
);
450 switch(__le32_to_cpu(sb
->level
)) {
456 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
459 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
465 /* This turns out to just be confusing */
466 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
467 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
468 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != 0xffff)
470 for (d
=0; d
< i
; d
++) {
471 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
473 if (role
== 0xffff) printf("empty");
474 else if(role
== 0xfffe) printf("failed");
475 else printf("%d", role
);
479 printf(" Device Role : ");
480 d
= __le32_to_cpu(sb
->dev_number
);
481 if (d
< __le32_to_cpu(sb
->max_dev
))
482 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
487 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
488 printf("Replacement device %d\n", role
);
490 printf("Active device %d\n", role
);
492 printf(" Array State : ");
493 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
496 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
497 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
511 /* This is confusing too */
513 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
514 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
518 if (faulty
) printf(" %d failed", faulty
);
520 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
524 static void brief_examine_super1(struct supertype
*st
, int verbose
)
526 struct mdp_superblock_1
*sb
= st
->sb
;
528 unsigned long long sb_offset
;
530 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
532 nm
= strchr(sb
->set_name
, ':');
535 else if (sb
->set_name
[0])
547 printf(" level=%s", c
);
548 sb_offset
= __le64_to_cpu(sb
->super_offset
);
550 printf(" metadata=1.1 ");
551 else if (sb_offset
<= 8)
552 printf(" metadata=1.2 ");
554 printf(" metadata=1.0 ");
556 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
558 for (i
=0; i
<16; i
++) {
559 if ((i
&3)==0 && i
!= 0) printf(":");
560 printf("%02x", sb
->set_uuid
[i
]);
562 if (sb
->set_name
[0]) {
564 print_quoted(sb
->set_name
);
569 static void export_examine_super1(struct supertype
*st
)
571 struct mdp_superblock_1
*sb
= st
->sb
;
576 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
577 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
579 if (sb
->set_name
[i
] == '\n' ||
580 sb
->set_name
[i
] == '\0') {
585 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
586 if (__le32_to_cpu(sb
->level
) > 0) {
587 int ddsks
= 0, ddsks_denom
= 1;
588 switch(__le32_to_cpu(sb
->level
)) {
589 case 1: ddsks
=1;break;
591 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
592 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
594 layout
= __le32_to_cpu(sb
->layout
);
595 ddsks
= __le32_to_cpu(sb
->raid_disks
);
596 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
599 long long asize
= __le64_to_cpu(sb
->size
);
600 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
601 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
605 for (i
=0; i
<16; i
++) {
606 if ((i
&3)==0 && i
!= 0) printf(":");
607 printf("%02x", sb
->set_uuid
[i
]);
610 printf("MD_UPDATE_TIME=%llu\n",
611 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
612 printf("MD_DEV_UUID=");
613 for (i
=0; i
<16; i
++) {
614 if ((i
&3)==0 && i
!= 0) printf(":");
615 printf("%02x", sb
->device_uuid
[i
]);
618 printf("MD_EVENTS=%llu\n",
619 (unsigned long long)__le64_to_cpu(sb
->events
));
622 static int copy_metadata1(struct supertype
*st
, int from
, int to
)
624 /* Read superblock. If it looks good, write it out.
625 * Then if a bitmap is present, copy that.
626 * And if a bad-block-list is present, copy that too.
629 unsigned long long dsize
, sb_offset
;
630 const int bufsize
= 4*1024;
631 struct mdp_superblock_1 super
, *sb
;
633 if (posix_memalign(&buf
, 4096, bufsize
) != 0)
636 if (!get_dev_size(from
, NULL
, &dsize
))
642 switch(st
->minor_version
) {
646 sb_offset
&= ~(4*2-1);
658 if (lseek64(from
, sb_offset
<< 9, 0) < 0LL)
660 if (read(from
, buf
, bufsize
) != bufsize
)
664 super
= *sb
; // save most of sb for when we reuse buf
666 if (__le32_to_cpu(super
.magic
) != MD_SB_MAGIC
||
667 __le32_to_cpu(super
.major_version
) != 1 ||
668 __le64_to_cpu(super
.super_offset
) != sb_offset
||
669 calc_sb_1_csum(sb
) != super
.sb_csum
)
672 if (lseek64(to
, sb_offset
<< 9, 0) < 0LL)
674 if (write(to
, buf
, bufsize
) != bufsize
)
677 if (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
)) {
678 unsigned long long bitmap_offset
= sb_offset
;
679 int bytes
= 4096; // just an estimate.
681 struct align_fd afrom
, ato
;
683 init_afd(&afrom
, from
);
686 bitmap_offset
+= (int32_t)__le32_to_cpu(super
.bitmap_offset
);
688 if (lseek64(from
, bitmap_offset
<<9, 0) < 0)
690 if (lseek64(to
, bitmap_offset
<<9, 0) < 0)
693 for (written
= 0; written
< bytes
; ) {
694 int n
= bytes
- written
;
697 if (aread(&afrom
, buf
, n
) != n
)
700 /* have the header, can calculate
701 * correct bitmap bytes */
704 bytes
= calc_bitmap_size(bms
, 512);
708 if (awrite(&ato
, buf
, n
) != n
)
714 if (super
.bblog_size
!= 0 &&
715 __le32_to_cpu(super
.bblog_size
) <= 100 &&
716 super
.bblog_offset
!= 0 &&
717 (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BAD_BLOCKS
))) {
718 /* There is a bad block log */
719 unsigned long long bb_offset
= sb_offset
;
720 int bytes
= __le32_to_cpu(super
.bblog_size
) * 512;
722 struct align_fd afrom
, ato
;
724 init_afd(&afrom
, from
);
727 bb_offset
+= (int32_t)__le32_to_cpu(super
.bblog_offset
);
729 if (lseek64(from
, bb_offset
<<9, 0) < 0)
731 if (lseek64(to
, bb_offset
<<9, 0) < 0)
734 for (written
= 0; written
< bytes
; ) {
735 int n
= bytes
- written
;
738 if (aread(&afrom
, buf
, n
) != n
)
741 if (awrite(&ato
, buf
, n
) != n
)
755 static void detail_super1(struct supertype
*st
, char *homehost
)
757 struct mdp_superblock_1
*sb
= st
->sb
;
758 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
760 int l
= homehost
? strlen(homehost
) : 0;
762 printf(" Name : %.32s", sb
->set_name
);
763 if (l
> 0 && l
< 32 &&
764 sb
->set_name
[l
] == ':' &&
765 strncmp(sb
->set_name
, homehost
, l
) == 0)
766 printf(" (local to host %s)", homehost
);
768 printf("\n Cluster Name : %-64s", bms
->cluster_name
);
769 printf("\n UUID : ");
770 for (i
=0; i
<16; i
++) {
771 if ((i
&3)==0 && i
!= 0) printf(":");
772 printf("%02x", sb
->set_uuid
[i
]);
774 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb
->events
));
777 static void brief_detail_super1(struct supertype
*st
)
779 struct mdp_superblock_1
*sb
= st
->sb
;
782 if (sb
->set_name
[0]) {
784 print_quoted(sb
->set_name
);
787 for (i
=0; i
<16; i
++) {
788 if ((i
&3)==0 && i
!= 0) printf(":");
789 printf("%02x", sb
->set_uuid
[i
]);
793 static void export_detail_super1(struct supertype
*st
)
795 struct mdp_superblock_1
*sb
= st
->sb
;
800 if (sb
->set_name
[i
] == '\n' ||
801 sb
->set_name
[i
] == '\0') {
806 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
809 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
811 struct mdp_superblock_1
*sb
= st
->sb
;
812 unsigned long long offset
;
817 if (!sb
->bblog_size
|| __le32_to_cpu(sb
->bblog_size
) > 100
818 || !sb
->bblog_offset
){
819 printf("No bad-blocks list configured on %s\n", devname
);
822 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
824 printf("Bad-blocks list is empty in %s\n", devname
);
828 size
= __le32_to_cpu(sb
->bblog_size
)* 512;
829 if (posix_memalign((void**)&bbl
, 4096, size
) != 0) {
830 pr_err("could not allocate badblocks list\n");
833 offset
= __le64_to_cpu(sb
->super_offset
) +
834 (int)__le32_to_cpu(sb
->bblog_offset
);
836 if (lseek64(fd
, offset
, 0) < 0) {
837 pr_err("Cannot seek to bad-blocks list\n");
840 if (read(fd
, bbl
, size
) != size
) {
841 pr_err("Cannot read bad-blocks list\n");
844 /* 64bits per entry. 10 bits is block-count, 54 bits is block
845 * offset. Blocks are sectors unless bblog->shift makes them bigger
848 printf("Bad-blocks on %s:\n", devname
);
849 for (i
= 0; i
< size
/8; i
++, bbp
++) {
850 __u64 bb
= __le64_to_cpu(*bbp
);
851 int count
= bb
& 0x3ff;
852 unsigned long long sector
= bb
>> 10;
857 sector
<<= sb
->bblog_shift
;
858 count
<<= sb
->bblog_shift
;
860 printf("%20llu for %d sectors\n", sector
, count
);
867 static int match_home1(struct supertype
*st
, char *homehost
)
869 struct mdp_superblock_1
*sb
= st
->sb
;
870 int l
= homehost
? strlen(homehost
) : 0;
872 return (l
> 0 && l
< 32 &&
873 sb
->set_name
[l
] == ':' &&
874 strncmp(sb
->set_name
, homehost
, l
) == 0);
877 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
879 struct mdp_superblock_1
*super
= st
->sb
;
880 char *cuuid
= (char*)uuid
;
883 cuuid
[i
] = super
->set_uuid
[i
];
886 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
888 struct mdp_superblock_1
*sb
= st
->sb
;
889 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
890 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
894 unsigned int map_disks
= info
->array
.raid_disks
;
895 unsigned long long super_offset
;
896 unsigned long long data_size
;
898 memset(info
, 0, sizeof(*info
));
899 info
->array
.major_version
= 1;
900 info
->array
.minor_version
= st
->minor_version
;
901 info
->array
.patch_version
= 0;
902 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
903 info
->array
.level
= __le32_to_cpu(sb
->level
);
904 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
905 info
->array
.md_minor
= -1;
906 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
907 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
908 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
910 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
912 if (__le32_to_cpu(bsb
->nodes
) > 1)
913 info
->array
.state
|= (1 << MD_SB_CLUSTERED
);
915 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
916 info
->component_size
= __le64_to_cpu(sb
->size
);
917 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
918 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
920 info
->disk
.major
= 0;
921 info
->disk
.minor
= 0;
922 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
923 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
924 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
927 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
929 super_offset
= __le64_to_cpu(sb
->super_offset
);
930 if (info
->array
.level
<= 0)
931 data_size
= __le64_to_cpu(sb
->data_size
);
933 data_size
= __le64_to_cpu(sb
->size
);
934 if (info
->data_offset
< super_offset
) {
935 unsigned long long end
;
936 info
->space_before
= info
->data_offset
;
939 if (sb
->bblog_offset
&& sb
->bblog_size
) {
940 unsigned long long bboffset
= super_offset
;
941 bboffset
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
946 if (super_offset
+ info
->bitmap_offset
< end
)
947 end
= super_offset
+ info
->bitmap_offset
;
949 if (info
->data_offset
+ data_size
< end
)
950 info
->space_after
= end
- data_size
- info
->data_offset
;
952 info
->space_after
= 0;
954 unsigned long long earliest
;
955 earliest
= super_offset
+ (32+4)*2; /* match kernel */
956 if (info
->bitmap_offset
> 0) {
957 unsigned long long bmend
= info
->bitmap_offset
;
958 unsigned long long size
= __le64_to_cpu(bsb
->sync_size
);
959 size
/= __le32_to_cpu(bsb
->chunksize
) >> 9;
960 size
= (size
+ 7) >> 3;
961 size
+= sizeof(bitmap_super_t
);
962 size
= ROUND_UP(size
, 4096);
965 if (bmend
> earliest
)
968 if (sb
->bblog_offset
&& sb
->bblog_size
) {
969 unsigned long long bbend
= super_offset
;
970 bbend
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
971 bbend
+= __le32_to_cpu(sb
->bblog_size
);
972 if (bbend
> earliest
)
975 if (earliest
< info
->data_offset
)
976 info
->space_before
= info
->data_offset
- earliest
;
978 info
->space_before
= 0;
979 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
981 if (info
->space_before
== 0 && info
->space_after
== 0) {
982 /* It will look like we don't support data_offset changes,
983 * be we do - it's just that there is no room.
984 * A change that reduced the number of devices should
985 * still be allowed, so set the otherwise useless value of '1'
987 info
->space_after
= 1;
990 info
->disk
.raid_disk
= -1;
993 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
996 info
->disk
.state
= 1; /* faulty */
999 info
->disk
.state
= 6; /* active and in sync */
1000 info
->disk
.raid_disk
= role
;
1002 if (sb
->devflags
& WriteMostly1
)
1003 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1004 info
->events
= __le64_to_cpu(sb
->events
);
1005 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
1006 info
->safe_mode_delay
= 200;
1008 memcpy(info
->uuid
, sb
->set_uuid
, 16);
1010 strncpy(info
->name
, sb
->set_name
, 32);
1013 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
1014 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
1015 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
1018 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
1019 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
1021 info
->recovery_start
= MaxSector
;
1023 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
1024 info
->reshape_active
= 1;
1025 if ((sb
->feature_map
& __le32_to_cpu(MD_FEATURE_NEW_OFFSET
)) &&
1026 sb
->new_offset
!= 0)
1027 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
1028 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
1029 info
->new_level
= __le32_to_cpu(sb
->new_level
);
1030 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
1031 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
1032 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
1033 if (info
->delta_disks
< 0)
1034 info
->array
.raid_disks
-= info
->delta_disks
;
1036 info
->reshape_active
= 0;
1038 info
->recovery_blocked
= info
->reshape_active
;
1041 for (i
=0; i
<map_disks
; i
++)
1043 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
1044 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
1045 if (/*role == 0xFFFF || */role
< (unsigned) info
->array
.raid_disks
) {
1047 if (map
&& role
< map_disks
)
1052 info
->array
.working_disks
= working
;
1055 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
1057 struct mdinfo
*info
;
1062 info
= xmalloc(sizeof(*info
));
1063 getinfo_super1(st
, info
, NULL
);
1067 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
1069 char *devname
, int verbose
,
1070 int uuid_set
, char *homehost
)
1072 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1073 * if any change was made. For others, the return value is
1077 struct mdp_superblock_1
*sb
= st
->sb
;
1079 if (strcmp(update
, "homehost") == 0 &&
1081 /* Note that 'homehost' is special as it is really
1086 c
= strchr(sb
->set_name
, ':');
1088 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1090 strncpy(info
->name
, sb
->set_name
, 32);
1094 if (strcmp(update
, "force-one")==0) {
1095 /* Not enough devices for a working array,
1096 * so bring this one up-to-date
1098 if (sb
->events
!= __cpu_to_le64(info
->events
))
1100 sb
->events
= __cpu_to_le64(info
->events
);
1101 } else if (strcmp(update
, "force-array")==0) {
1102 /* Degraded array and 'force' requests to
1103 * maybe need to mark it 'clean'.
1105 switch(__le32_to_cpu(sb
->level
)) {
1106 case 5: case 4: case 6:
1107 /* need to force clean */
1108 if (sb
->resync_offset
!= MaxSector
)
1110 sb
->resync_offset
= MaxSector
;
1112 } else if (strcmp(update
, "assemble")==0) {
1113 int d
= info
->disk
.number
;
1115 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
1116 want
= info
->disk
.raid_disk
;
1119 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
1120 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
1123 if (info
->reshape_active
&&
1124 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1125 info
->delta_disks
>= 0 &&
1126 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
1127 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1130 if (info
->reshape_active
&&
1131 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1132 info
->delta_disks
< 0 &&
1133 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
1134 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1137 } else if (strcmp(update
, "linear-grow-new") == 0) {
1140 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
1142 for (i
=0 ; i
< max
; i
++)
1143 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= 0xfffe)
1145 sb
->dev_number
= __cpu_to_le32(i
);
1146 info
->disk
.number
= i
;
1147 if (max
>= __le32_to_cpu(sb
->max_dev
))
1148 sb
->max_dev
= __cpu_to_le32(max
+1);
1150 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1151 read(rfd
, sb
->device_uuid
, 16) != 16) {
1152 __u32 r
[4] = {random(), random(), random(), random()};
1153 memcpy(sb
->device_uuid
, r
, 16);
1159 __cpu_to_le16(info
->disk
.raid_disk
);
1161 fd
= open(devname
, O_RDONLY
);
1163 unsigned long long ds
;
1164 get_dev_size(fd
, devname
, &ds
);
1167 if (__le64_to_cpu(sb
->super_offset
) <
1168 __le64_to_cpu(sb
->data_offset
)) {
1169 sb
->data_size
= __cpu_to_le64(
1170 ds
- __le64_to_cpu(sb
->data_offset
));
1173 ds
&= ~(unsigned long long)(4*2-1);
1174 sb
->super_offset
= __cpu_to_le64(ds
);
1175 sb
->data_size
= __cpu_to_le64(
1176 ds
- __le64_to_cpu(sb
->data_offset
));
1179 } else if (strcmp(update
, "linear-grow-update") == 0) {
1180 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1181 sb
->dev_roles
[info
->disk
.number
] =
1182 __cpu_to_le16(info
->disk
.raid_disk
);
1183 } else if (strcmp(update
, "resync") == 0) {
1184 /* make sure resync happens */
1185 sb
->resync_offset
= 0ULL;
1186 } else if (strcmp(update
, "uuid") == 0) {
1187 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1189 if (__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1190 struct bitmap_super_s
*bm
;
1191 bm
= (struct bitmap_super_s
*)(st
->sb
+MAX_SB_SIZE
);
1192 memcpy(bm
->uuid
, sb
->set_uuid
, 16);
1194 } else if (strcmp(update
, "no-bitmap") == 0) {
1195 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1196 } else if (strcmp(update
, "bbl") == 0) {
1197 /* only possible if there is room after the bitmap, or if
1198 * there is no bitmap
1200 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1201 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1202 long bitmap_offset
= (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
);
1203 long bm_sectors
= 0;
1207 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1208 struct bitmap_super_s
*bsb
;
1209 bsb
= (struct bitmap_super_s
*)(((char*)sb
)+MAX_SB_SIZE
);
1210 bm_sectors
= bitmap_sectors(bsb
);
1213 if (sb_offset
< data_offset
) {
1214 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1217 bb_offset
= sb_offset
+ 8;
1218 if (bm_sectors
&& bitmap_offset
> 0)
1219 bb_offset
= bitmap_offset
+ bm_sectors
;
1220 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1221 && bb_offset
+ 8+8 <= (long)data_offset
)
1222 /* too close to bitmap, and room to grow */
1224 if (bb_offset
+ 8 <= (long)data_offset
) {
1225 sb
->bblog_size
= __cpu_to_le16(8);
1226 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1229 /* 1.0 - Put bbl just before super block */
1230 if (bm_sectors
&& bitmap_offset
< 0)
1231 space
= -bitmap_offset
- bm_sectors
;
1233 space
= sb_offset
- data_offset
-
1234 __le64_to_cpu(sb
->data_size
);
1236 sb
->bblog_size
= __cpu_to_le16(8);
1237 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1240 } else if (strcmp(update
, "no-bbl") == 0) {
1241 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1242 pr_err("Cannot remove active bbl from %s\n",devname
);
1245 sb
->bblog_shift
= 0;
1246 sb
->bblog_offset
= 0;
1248 } else if (strcmp(update
, "name") == 0) {
1249 if (info
->name
[0] == 0)
1250 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1251 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1253 strchr(info
->name
, ':') == NULL
&&
1254 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1255 strcpy(sb
->set_name
, homehost
);
1256 strcat(sb
->set_name
, ":");
1257 strcat(sb
->set_name
, info
->name
);
1259 strcpy(sb
->set_name
, info
->name
);
1260 } else if (strcmp(update
, "devicesize") == 0 &&
1261 __le64_to_cpu(sb
->super_offset
) <
1262 __le64_to_cpu(sb
->data_offset
)) {
1263 /* set data_size to device size less data_offset */
1264 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1265 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1266 sb
->data_size
= __cpu_to_le64(
1267 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1268 } else if (strcmp(update
, "revert-reshape") == 0) {
1270 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1271 pr_err("No active reshape to revert on %s\n",
1275 unsigned long long reshape_sectors
;
1278 /* reshape_position is a little messy.
1279 * Its value must be a multiple of the larger
1280 * chunk size, and of the "after" data disks.
1281 * So when reverting we need to change it to
1282 * be a multiple of the new "after" data disks,
1283 * which is the old "before".
1284 * If it isn't already a multiple of 'before',
1285 * the only thing we could do would be
1286 * copy some block around on the disks, which
1287 * is easy to get wrong.
1288 * So we reject a revert-reshape unless the
1289 * alignment is good.
1291 if (__le32_to_cpu(sb
->level
) >= 4 &&
1292 __le32_to_cpu(sb
->level
) <= 6) {
1293 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1294 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1295 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1296 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1297 if (reshape_sectors
% reshape_chunk
) {
1298 pr_err("Reshape position is not suitably aligned.\n");
1299 pr_err("Try normal assembly and stop again\n");
1303 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1304 __le32_to_cpu(sb
->delta_disks
));
1305 if (sb
->delta_disks
== 0)
1306 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1308 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1310 temp
= sb
->new_layout
;
1311 sb
->new_layout
= sb
->layout
;
1314 temp
= sb
->new_chunk
;
1315 sb
->new_chunk
= sb
->chunksize
;
1316 sb
->chunksize
= temp
;
1318 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1319 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1320 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1321 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1322 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1325 } else if (strcmp(update
, "_reshape_progress")==0)
1326 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1327 else if (strcmp(update
, "writemostly")==0)
1328 sb
->devflags
|= WriteMostly1
;
1329 else if (strcmp(update
, "readwrite")==0)
1330 sb
->devflags
&= ~WriteMostly1
;
1334 sb
->sb_csum
= calc_sb_1_csum(sb
);
1338 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1339 unsigned long long size
, char *name
, char *homehost
,
1340 int *uuid
, unsigned long long data_offset
)
1342 struct mdp_superblock_1
*sb
;
1348 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1349 pr_err("could not allocate superblock\n");
1352 memset(sb
, 0, SUPER1_SIZE
);
1356 /* zeroing superblock */
1360 spares
= info
->working_disks
- info
->active_disks
;
1361 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1362 pr_err("too many devices requested: %d+%d > %d\n",
1363 info
->raid_disks
, spares
, MAX_DEVS
);
1367 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1368 sb
->major_version
= __cpu_to_le32(1);
1369 sb
->feature_map
= 0;
1373 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1375 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1376 read(rfd
, sb
->set_uuid
, 16) != 16) {
1377 __u32 r
[4] = {random(), random(), random(), random()};
1378 memcpy(sb
->set_uuid
, r
, 16);
1380 if (rfd
>= 0) close(rfd
);
1383 if (name
== NULL
|| *name
== 0) {
1384 sprintf(defname
, "%d", info
->md_minor
);
1388 strchr(name
, ':')== NULL
&&
1389 strlen(homehost
)+1+strlen(name
) < 32) {
1390 strcpy(sb
->set_name
, homehost
);
1391 strcat(sb
->set_name
, ":");
1392 strcat(sb
->set_name
, name
);
1394 strcpy(sb
->set_name
, name
);
1396 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1397 sb
->level
= __cpu_to_le32(info
->level
);
1398 sb
->layout
= __cpu_to_le32(info
->layout
);
1399 sb
->size
= __cpu_to_le64(size
*2ULL);
1400 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1401 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1403 sb
->data_offset
= __cpu_to_le64(data_offset
);
1404 sb
->data_size
= __cpu_to_le64(0);
1405 sb
->super_offset
= __cpu_to_le64(0);
1406 sb
->recovery_offset
= __cpu_to_le64(0);
1408 sb
->utime
= sb
->ctime
;
1409 sb
->events
= __cpu_to_le64(1);
1410 if (info
->state
& (1<<MD_SB_CLEAN
))
1411 sb
->resync_offset
= MaxSector
;
1413 sb
->resync_offset
= 0;
1414 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1415 sbsize
= ROUND_UP(sbsize
, 512);
1416 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1418 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1426 long long data_offset
;
1427 mdu_disk_info_t disk
;
1428 struct devinfo
*next
;
1431 /* Add a device to the superblock being created */
1432 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1433 int fd
, char *devname
, unsigned long long data_offset
)
1435 struct mdp_superblock_1
*sb
= st
->sb
;
1436 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1437 struct devinfo
*di
, **dip
;
1439 if ((dk
->state
& 6) == 6) /* active, sync */
1440 *rp
= __cpu_to_le16(dk
->raid_disk
);
1441 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1446 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1447 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1448 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1450 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1451 sb
->devflags
= 0; /* don't copy another disks flags */
1452 sb
->sb_csum
= calc_sb_1_csum(sb
);
1454 dip
= (struct devinfo
**)&st
->info
;
1456 dip
= &(*dip
)->next
;
1457 di
= xmalloc(sizeof(struct devinfo
));
1459 di
->devname
= devname
;
1461 di
->data_offset
= data_offset
;
1469 static void locate_bitmap1(struct supertype
*st
, int fd
);
1471 static int store_super1(struct supertype
*st
, int fd
)
1473 struct mdp_superblock_1
*sb
= st
->sb
;
1474 unsigned long long sb_offset
;
1475 struct align_fd afd
;
1477 unsigned long long dsize
;
1479 if (!get_dev_size(fd
, NULL
, &dsize
))
1490 * Calculate the position of the superblock.
1491 * It is always aligned to a 4K boundary and
1492 * depending on minor_version, it can be:
1493 * 0: At least 8K, but less than 12K, from end of device
1494 * 1: At start of device
1495 * 2: 4K from start of device.
1497 switch(st
->minor_version
) {
1501 sb_offset
&= ~(4*2-1);
1513 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1514 0 != __le64_to_cpu(sb
->super_offset
)
1516 pr_err("internal error - sb_offset is wrong\n");
1520 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1523 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1525 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1528 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1529 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1530 (((char*)sb
)+MAX_SB_SIZE
);
1531 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1532 locate_bitmap1(st
, fd
);
1533 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1541 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1543 static unsigned long choose_bm_space(unsigned long devsize
)
1545 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1546 * if bigger than 200Gig, save 128k
1547 * NOTE: result must be multiple of 4K else bad things happen
1548 * on 4K-sector devices.
1550 if (devsize
< 64*2) return 0;
1551 if (devsize
- 64*2 >= 200*1024*1024*2)
1553 if (devsize
- 4*2 > 8*1024*1024*2)
1558 static void free_super1(struct supertype
*st
);
1561 static int write_init_super1(struct supertype
*st
)
1563 struct mdp_superblock_1
*sb
= st
->sb
;
1564 struct supertype
*refst
;
1567 unsigned long long bm_space
;
1569 unsigned long long dsize
, array_size
;
1570 unsigned long long sb_offset
;
1571 unsigned long long data_offset
;
1573 for (di
= st
->info
; di
; di
= di
->next
) {
1574 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1579 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1582 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1583 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1584 sb
->devflags
|= WriteMostly1
;
1586 sb
->devflags
&= ~WriteMostly1
;
1588 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1589 read(rfd
, sb
->device_uuid
, 16) != 16) {
1590 __u32 r
[4] = {random(), random(), random(), random()};
1591 memcpy(sb
->device_uuid
, r
, 16);
1598 refst
= dup_super(st
);
1599 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1600 struct mdp_superblock_1
*refsb
= refst
->sb
;
1602 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1603 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1604 /* same array, so preserve events and
1606 sb
->events
= refsb
->events
;
1607 /* bugs in 2.6.17 and earlier mean the
1608 * dev_number chosen in Manage must be preserved
1610 if (get_linux_version() >= 2006018)
1611 sb
->dev_number
= refsb
->dev_number
;
1617 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1630 * Calculate the position of the superblock.
1631 * It is always aligned to a 4K boundary and
1632 * depending on minor_version, it can be:
1633 * 0: At least 8K, but less than 12K, from end of device
1634 * 1: At start of device
1635 * 2: 4K from start of device.
1636 * data_offset has already been set.
1638 array_size
= __le64_to_cpu(sb
->size
);
1639 /* work out how much space we left for a bitmap,
1640 * Add 8 sectors for bad block log */
1641 bm_space
= choose_bm_space(array_size
) + 8;
1643 data_offset
= di
->data_offset
;
1644 if (data_offset
== INVALID_SECTORS
)
1645 data_offset
= st
->data_offset
;
1646 switch(st
->minor_version
) {
1648 if (data_offset
== INVALID_SECTORS
)
1652 sb_offset
&= ~(4*2-1);
1653 sb
->data_offset
= __cpu_to_le64(data_offset
);
1654 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1655 if (sb_offset
< array_size
+ bm_space
)
1656 bm_space
= sb_offset
- array_size
;
1657 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1658 if (bm_space
>= 8) {
1659 sb
->bblog_size
= __cpu_to_le16(8);
1660 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1664 sb
->super_offset
= __cpu_to_le64(0);
1665 if (data_offset
== INVALID_SECTORS
)
1668 sb
->data_offset
= __cpu_to_le64(data_offset
);
1669 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1670 if (data_offset
>= 8 + 32*2 + 8) {
1671 sb
->bblog_size
= __cpu_to_le16(8);
1672 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1673 } else if (data_offset
>= 16) {
1674 sb
->bblog_size
= __cpu_to_le16(8);
1675 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8);
1680 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1681 if (data_offset
== INVALID_SECTORS
)
1684 sb
->data_offset
= __cpu_to_le64(data_offset
);
1685 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1686 if (data_offset
>= 16 + 32*2 + 8) {
1687 sb
->bblog_size
= __cpu_to_le16(8);
1688 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1689 } else if (data_offset
>= 16+16) {
1690 sb
->bblog_size
= __cpu_to_le16(8);
1691 /* '8' sectors for the bblog, and another '8'
1692 * because we want offset from superblock, not
1695 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8-8);
1699 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1700 st
->minor_version
, di
->devname
);
1704 /* Disable badblock log on clusters, or when explicitly requested */
1705 if (st
->nodes
> 0 || conf_get_create_info()->bblist
== 0) {
1707 sb
->bblog_offset
= 0;
1710 sb
->sb_csum
= calc_sb_1_csum(sb
);
1711 rv
= store_super1(st
, di
->fd
);
1712 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1713 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NoUpdate
);
1721 pr_err("Failed to write metadata to %s\n",
1728 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1732 * 0 same, or first was empty, and second was copied
1733 * 1 second had wrong number
1735 * 3 wrong other info
1737 struct mdp_superblock_1
*first
= st
->sb
;
1738 struct mdp_superblock_1
*second
= tst
->sb
;
1740 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1742 if (second
->major_version
!= __cpu_to_le32(1))
1746 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1747 pr_err("could not allocate superblock\n");
1750 memcpy(first
, second
, SUPER1_SIZE
);
1754 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1757 if (first
->ctime
!= second
->ctime
||
1758 first
->level
!= second
->level
||
1759 first
->layout
!= second
->layout
||
1760 first
->size
!= second
->size
||
1761 first
->chunksize
!= second
->chunksize
||
1762 first
->raid_disks
!= second
->raid_disks
)
1767 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1769 unsigned long long dsize
;
1770 unsigned long long sb_offset
;
1771 struct mdp_superblock_1
*super
;
1773 struct bitmap_super_s
*bsb
;
1774 struct misc_dev_info
*misc
;
1775 struct align_fd afd
;
1781 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1783 struct supertype tst
;
1784 __u64 bestctime
= 0;
1785 /* guess... choose latest ctime */
1786 memset(&tst
, 0, sizeof(tst
));
1788 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1789 switch(load_super1(&tst
, fd
, devname
)) {
1790 case 0: super
= tst
.sb
;
1791 if (bestvers
== -1 ||
1792 bestctime
< __le64_to_cpu(super
->ctime
)) {
1793 bestvers
= tst
.minor_version
;
1794 bestctime
= __le64_to_cpu(super
->ctime
);
1799 case 1: return 1; /*bad device */
1800 case 2: break; /* bad, try next */
1803 if (bestvers
!= -1) {
1805 tst
.minor_version
= bestvers
;
1807 tst
.max_devs
= MAX_DEVS
;
1808 rv
= load_super1(&tst
, fd
, devname
);
1815 if (!get_dev_size(fd
, devname
, &dsize
))
1821 pr_err("%s is too small for md: size is %llu sectors.\n",
1827 * Calculate the position of the superblock.
1828 * It is always aligned to a 4K boundary and
1829 * depending on minor_version, it can be:
1830 * 0: At least 8K, but less than 12K, from end of device
1831 * 1: At start of device
1832 * 2: 4K from start of device.
1834 switch(st
->minor_version
) {
1838 sb_offset
&= ~(4*2-1);
1850 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1852 pr_err("Cannot seek to superblock on %s: %s\n",
1853 devname
, strerror(errno
));
1857 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
1858 pr_err("could not allocate superblock\n");
1862 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
1864 pr_err("Cannot read superblock on %s\n",
1870 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
1872 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
1873 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
1878 if (__le32_to_cpu(super
->major_version
) != 1) {
1880 pr_err("Cannot interpret superblock on %s - version is %d\n",
1881 devname
, __le32_to_cpu(super
->major_version
));
1885 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
1887 pr_err("No superblock found on %s (super_offset is wrong)\n",
1894 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
1896 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
1897 misc
->device_size
= dsize
;
1898 if (st
->data_offset
== INVALID_SECTORS
)
1899 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
1901 /* Now check on the bitmap superblock */
1902 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
1904 /* Read the bitmap superblock and make sure it looks
1905 * valid. If it doesn't clear the bit. An --assemble --force
1906 * should get that written out.
1908 locate_bitmap1(st
, fd
);
1909 if (aread(&afd
, bsb
, 512) != 512)
1912 uuid_from_super1(st
, uuid
);
1913 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
1914 memcmp(bsb
->uuid
, uuid
, 16) != 0)
1919 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
1920 & ~MD_FEATURE_BITMAP_OFFSET
);
1924 static struct supertype
*match_metadata_desc1(char *arg
)
1926 struct supertype
*st
= xcalloc(1, sizeof(*st
));
1928 st
->container_devnm
[0] = 0;
1930 st
->max_devs
= MAX_DEVS
;
1932 st
->data_offset
= INVALID_SECTORS
;
1933 /* leading zeros can be safely ignored. --detail generates them. */
1936 if (strcmp(arg
, "1.0") == 0 ||
1937 strcmp(arg
, "1.00") == 0) {
1938 st
->minor_version
= 0;
1941 if (strcmp(arg
, "1.1") == 0 ||
1942 strcmp(arg
, "1.01") == 0
1944 st
->minor_version
= 1;
1947 if (strcmp(arg
, "1.2") == 0 ||
1948 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
1949 strcmp(arg
, "default") == 0 ||
1950 #endif /* DEFAULT_OLD_METADATA */
1951 strcmp(arg
, "1.02") == 0) {
1952 st
->minor_version
= 2;
1955 if (strcmp(arg
, "1") == 0 ||
1956 strcmp(arg
, "default") == 0) {
1957 st
->minor_version
= -1;
1965 /* find available size on device with this devsize, using
1966 * superblock type st, and reserving 'reserve' sectors for
1969 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
1970 unsigned long long data_offset
)
1972 struct mdp_superblock_1
*super
= st
->sb
;
1979 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1980 /* hot-add. allow for actual size of bitmap */
1981 struct bitmap_super_s
*bsb
;
1982 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
1983 bmspace
= bitmap_sectors(bsb
);
1986 /* Allow space for bad block log */
1987 if (super
->bblog_size
)
1988 bbspace
= __le16_to_cpu(super
->bblog_size
);
1990 if (st
->minor_version
< 0)
1991 /* not specified, so time to set default */
1992 st
->minor_version
= 2;
1994 if (data_offset
== INVALID_SECTORS
)
1995 data_offset
= st
->data_offset
;
1997 if (data_offset
!= INVALID_SECTORS
)
1998 switch(st
->minor_version
) {
2000 return devsize
- data_offset
- 8*2 - bbspace
;
2003 return devsize
- data_offset
;
2010 switch(st
->minor_version
) {
2013 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2015 /* at start, 4K for superblock and possible bitmap */
2016 return devsize
- 4*2 - bbspace
;
2018 /* 4k from start, 4K for superblock and possible bitmap */
2019 return devsize
- (4+4)*2 - bbspace
;
2025 add_internal_bitmap1(struct supertype
*st
,
2026 int *chunkp
, int delay
, int write_behind
,
2027 unsigned long long size
,
2028 int may_change
, int major
)
2031 * If not may_change, then this is a 'Grow' without sysfs support for
2032 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2033 * If may_change, then this is create or a Grow with sysfs syupport,
2034 * and we can put the bitmap wherever we like.
2036 * size is in sectors, chunk is in bytes !!!
2039 unsigned long long bits
;
2040 unsigned long long max_bits
;
2041 unsigned long long min_chunk
;
2043 long bbl_offset
, bbl_size
;
2044 unsigned long long chunk
= *chunkp
;
2047 struct mdp_superblock_1
*sb
= st
->sb
;
2048 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2051 if (__le64_to_cpu(sb
->data_size
) == 0)
2052 /* Must be creating the array, else data_size would be non-zero */
2054 switch(st
->minor_version
) {
2056 /* either 3K after the superblock (when hot-add),
2057 * or some amount of space before.
2060 /* We are creating array, so we *know* how much room has
2065 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2067 room
= __le64_to_cpu(sb
->super_offset
)
2068 - __le64_to_cpu(sb
->data_offset
)
2069 - __le64_to_cpu(sb
->data_size
);
2070 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2073 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2074 if (bbl_size
< -bbl_offset
)
2075 bbl_size
= -bbl_offset
;
2077 if (!may_change
|| (room
< 3*2 &&
2078 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2083 offset
= 0; /* means movable offset */
2088 case 2: /* between superblock and data */
2092 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2094 room
= __le64_to_cpu(sb
->data_offset
)
2095 - __le64_to_cpu(sb
->super_offset
);
2096 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2098 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2103 room
-= 2; /* Leave 1K for superblock */
2107 room
-= 4*2; /* leave 4K for superblock */
2117 if (chunk
== UnSet
&& room
> 128*2)
2118 /* Limit to 128K of bitmap when chunk size not requested */
2122 /* No room for a bitmap */
2125 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2127 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2128 bits
= (size
*512)/min_chunk
+1;
2129 while (bits
> max_bits
) {
2133 if (chunk
== UnSet
) {
2134 /* For practical purpose, 64Meg is a good
2135 * default chunk size for internal bitmaps.
2138 if (chunk
< 64*1024*1024)
2139 chunk
= 64*1024*1024;
2140 } else if (chunk
< min_chunk
)
2141 return 0; /* chunk size too small */
2142 if (chunk
== 0) /* rounding problem */
2146 /* start bitmap on a 4K boundary with enough space for
2149 bits
= (size
*512) / chunk
+ 1;
2150 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2151 room
*= 8; /* convert 4K blocks to sectors */
2152 offset
= -room
- bbl_size
;
2155 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2157 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2158 | MD_FEATURE_BITMAP_OFFSET
);
2159 memset(bms
, 0, sizeof(*bms
));
2160 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2161 bms
->version
= __cpu_to_le32(major
);
2162 uuid_from_super1(st
, uuid
);
2163 memcpy(bms
->uuid
, uuid
, 16);
2164 bms
->chunksize
= __cpu_to_le32(chunk
);
2165 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2166 bms
->sync_size
= __cpu_to_le64(size
);
2167 bms
->write_behind
= __cpu_to_le32(write_behind
);
2168 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2170 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2171 | MD_FEATURE_BITMAP_VERSIONED
);
2172 if (st
->cluster_name
)
2173 strncpy((char *)bms
->cluster_name
,
2174 st
->cluster_name
, strlen(st
->cluster_name
));
2180 static void locate_bitmap1(struct supertype
*st
, int fd
)
2182 unsigned long long offset
;
2183 struct mdp_superblock_1
*sb
;
2187 if (st
->ss
->load_super(st
, fd
, NULL
))
2188 return; /* no error I hope... */
2193 offset
= __le64_to_cpu(sb
->super_offset
);
2194 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
);
2197 lseek64(fd
, offset
<<9, 0);
2200 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2202 struct mdp_superblock_1
*sb
= st
->sb
;
2203 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2207 struct align_fd afd
;
2209 unsigned long long total_bm_space
, bm_space_per_node
;
2213 /* update cluster name */
2214 if (st
->cluster_name
) {
2215 memset((char *)bms
->cluster_name
, 0, sizeof(bms
->cluster_name
));
2216 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, 64);
2220 /* cluster md only supports superblock 1.2 now */
2221 if (st
->minor_version
!= 2) {
2222 pr_err("Warning: cluster md only works with superblock 1.2\n");
2226 /* Each node has an independent bitmap, it is necessary to calculate the
2227 * space is enough or not, first get how many bytes for the total bitmap */
2228 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2230 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2231 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2233 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2234 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2235 total_bm_space
/ bm_space_per_node
);
2239 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2248 locate_bitmap1(st
, fd
);
2250 if (posix_memalign(&buf
, 4096, 4096))
2254 /* Only the bitmap[0] should resync
2255 * whole device on initial assembly
2258 memset(buf
, 0x00, 4096);
2260 memset(buf
, 0xff, 4096);
2261 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2263 towrite
= calc_bitmap_size(bms
, 4096);
2264 while (towrite
> 0) {
2268 n
= awrite(&afd
, buf
, n
);
2274 memset(buf
, 0x00, 4096);
2276 memset(buf
, 0xff, 4096);
2283 } while (++i
< __le32_to_cpu(bms
->nodes
));
2289 static void free_super1(struct supertype
*st
)
2294 struct devinfo
*di
= st
->info
;
2295 st
->info
= di
->next
;
2304 static int validate_geometry1(struct supertype
*st
, int level
,
2305 int layout
, int raiddisks
,
2306 int *chunk
, unsigned long long size
,
2307 unsigned long long data_offset
,
2308 char *subdev
, unsigned long long *freesize
,
2311 unsigned long long ldsize
, devsize
;
2313 unsigned long long headroom
;
2316 if (level
== LEVEL_CONTAINER
) {
2318 pr_err("1.x metadata does not support containers\n");
2321 if (*chunk
== UnSet
)
2322 *chunk
= DEFAULT_CHUNK
;
2327 if (st
->minor_version
< 0)
2328 /* not specified, so time to set default */
2329 st
->minor_version
= 2;
2331 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2334 pr_err("super1.x cannot open %s: %s\n",
2335 subdev
, strerror(errno
));
2339 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2345 devsize
= ldsize
>> 9;
2351 /* creating: allow suitable space for bitmap */
2352 bmspace
= choose_bm_space(devsize
);
2354 if (data_offset
== INVALID_SECTORS
)
2355 data_offset
= st
->data_offset
;
2356 if (data_offset
== INVALID_SECTORS
)
2357 switch (st
->minor_version
) {
2363 /* Choose data offset appropriate for this device
2364 * and use as default for whole array.
2365 * The data_offset must allow for bitmap space
2366 * and base metadata, should allow for some headroom
2367 * for reshape, and should be rounded to multiple
2369 * Headroom is limited to 128M, but aim for about 0.1%
2371 headroom
= 128*1024*2;
2372 while ((headroom
<< 10) > devsize
&&
2374 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2376 data_offset
= 12*2 + bmspace
+ headroom
;
2377 #define ONE_MEG (2*1024)
2378 if (data_offset
> ONE_MEG
)
2379 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2382 if (st
->data_offset
== INVALID_SECTORS
)
2383 st
->data_offset
= data_offset
;
2384 switch(st
->minor_version
) {
2385 case 0: /* metadata at end. Round down and subtract space to reserve */
2386 devsize
= (devsize
& ~(4ULL*2-1));
2387 /* space for metadata, bblog, bitmap */
2388 devsize
-= 8*2 + 8 + bmspace
;
2392 devsize
-= data_offset
;
2395 *freesize
= devsize
;
2398 #endif /* MDASSEMBLE */
2400 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2402 /* Create a v1.0 superblock based on 'info'*/
2404 struct mdp_superblock_1
*sb
;
2407 unsigned long long offset
;
2409 if (posix_memalign(&ret
, 4096, 1024) != 0)
2412 memset(ret
, 0, 1024);
2413 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2414 sb
->major_version
= __cpu_to_le32(1);
2416 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2417 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2418 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2419 sb
->level
= __cpu_to_le32(info
->array
.level
);
2420 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2421 sb
->size
= __cpu_to_le64(info
->component_size
);
2422 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2423 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2424 if (info
->array
.level
> 0)
2425 sb
->data_size
= sb
->size
;
2427 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2428 sb
->resync_offset
= MaxSector
;
2429 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2430 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2431 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2433 offset
= st
->devsize
/512 - 8*2;
2435 sb
->super_offset
= __cpu_to_le64(offset
);
2436 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2438 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
2439 read(rfd
, sb
->device_uuid
, 16) != 16) {
2440 __u32 r
[4] = {random(), random(), random(), random()};
2441 memcpy(sb
->device_uuid
, r
, 16);
2446 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2447 int state
= sb0
->disks
[i
].state
;
2448 sb
->dev_roles
[i
] = 0xFFFF;
2449 if ((state
& (1<<MD_DISK_SYNC
)) &&
2450 !(state
& (1<<MD_DISK_FAULTY
)))
2451 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2453 sb
->sb_csum
= calc_sb_1_csum(sb
);
2457 struct superswitch super1
= {
2459 .examine_super
= examine_super1
,
2460 .brief_examine_super
= brief_examine_super1
,
2461 .export_examine_super
= export_examine_super1
,
2462 .detail_super
= detail_super1
,
2463 .brief_detail_super
= brief_detail_super1
,
2464 .export_detail_super
= export_detail_super1
,
2465 .write_init_super
= write_init_super1
,
2466 .validate_geometry
= validate_geometry1
,
2467 .add_to_super
= add_to_super1
,
2468 .examine_badblocks
= examine_badblocks_super1
,
2469 .copy_metadata
= copy_metadata1
,
2471 .match_home
= match_home1
,
2472 .uuid_from_super
= uuid_from_super1
,
2473 .getinfo_super
= getinfo_super1
,
2474 .container_content
= container_content1
,
2475 .update_super
= update_super1
,
2476 .init_super
= init_super1
,
2477 .store_super
= store_super1
,
2478 .compare_super
= compare_super1
,
2479 .load_super
= load_super1
,
2480 .match_metadata_desc
= match_metadata_desc1
,
2481 .avail_size
= avail_size1
,
2482 .add_internal_bitmap
= add_internal_bitmap1
,
2483 .locate_bitmap
= locate_bitmap1
,
2484 .write_bitmap
= write_bitmap1
,
2485 .free_super
= free_super1
,
2486 #if __BYTE_ORDER == BIG_ENDIAN