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_ALL (MD_FEATURE_BITMAP_OFFSET \
129 |MD_FEATURE_RECOVERY_OFFSET \
130 |MD_FEATURE_RESHAPE_ACTIVE \
131 |MD_FEATURE_BAD_BLOCKS \
132 |MD_FEATURE_REPLACEMENT \
133 |MD_FEATURE_RESHAPE_BACKWARDS \
134 |MD_FEATURE_NEW_OFFSET \
137 /* return how many bytes are needed for bitmap, for cluster-md each node
138 * should have it's own bitmap */
139 static unsigned int calc_bitmap_size(bitmap_super_t
*bms
, unsigned int boundary
)
141 unsigned long long bits
, bytes
;
143 bits
= __le64_to_cpu(bms
->sync_size
) / (__le32_to_cpu(bms
->chunksize
)>>9);
144 bytes
= (bits
+7) >> 3;
145 bytes
+= sizeof(bitmap_super_t
);
146 bytes
= ROUND_UP(bytes
, boundary
);
151 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
153 unsigned int disk_csum
, csum
;
154 unsigned long long newcsum
;
155 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
156 unsigned int *isuper
= (unsigned int*)sb
;
158 /* make sure I can count... */
159 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
160 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
161 sizeof(struct mdp_superblock_1
) != 256) {
162 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
165 disk_csum
= sb
->sb_csum
;
168 for (; size
>=4; size
-= 4 ) {
169 newcsum
+= __le32_to_cpu(*isuper
);
174 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
176 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
177 sb
->sb_csum
= disk_csum
;
178 return __cpu_to_le32(csum
);
182 * Information related to file descriptor used for aligned reads/writes.
183 * Cache the block size.
190 static void init_afd(struct align_fd
*afd
, int fd
)
194 if (ioctl(afd
->fd
, BLKSSZGET
, &afd
->blk_sz
) != 0)
198 static char abuf
[4096+4096];
199 static int aread(struct align_fd
*afd
, void *buf
, int len
)
202 * On devices with a 4K sector size, we need to read
203 * the full sector and copy relevant bits into
212 if (!bsize
|| bsize
> 4096 || len
> 4096) {
214 fprintf(stderr
, "WARNING - aread() called with invalid block size\n");
217 b
= ROUND_UP_PTR((char *)abuf
, 4096);
219 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
221 n
= read(afd
->fd
, b
, iosize
);
224 lseek(afd
->fd
, len
- n
, 1);
231 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
234 * On devices with a 4K sector size, we need to write
235 * the full sector. We pre-read if the sector is larger
237 * The address must be sector-aligned.
244 if (!bsize
|| bsize
> 4096 || len
> 4096) {
246 fprintf(stderr
, "WARNING - awrite() called with invalid block size\n");
249 b
= ROUND_UP_PTR((char *)abuf
, 4096);
251 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
255 n
= read(afd
->fd
, b
, iosize
);
258 lseek(afd
->fd
, -n
, 1);
262 n
= write(afd
->fd
, b
, iosize
);
265 lseek(afd
->fd
, len
- n
, 1);
270 static void examine_super1(struct supertype
*st
, char *homehost
)
272 struct mdp_superblock_1
*sb
= st
->sb
;
273 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
280 int l
= homehost
? strlen(homehost
) : 0;
282 unsigned long long sb_offset
;
285 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
286 printf(" Version : 1");
287 sb_offset
= __le64_to_cpu(sb
->super_offset
);
290 else if (sb_offset
<= 8)
294 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
295 printf(" Array UUID : ");
296 for (i
=0; i
<16; i
++) {
297 if ((i
&3)==0 && i
!= 0) printf(":");
298 printf("%02x", sb
->set_uuid
[i
]);
301 printf(" Name : %.32s", sb
->set_name
);
302 if (l
> 0 && l
< 32 &&
303 sb
->set_name
[l
] == ':' &&
304 strncmp(sb
->set_name
, homehost
, l
) == 0)
305 printf(" (local to host %s)", homehost
);
308 printf("Cluster Name : %s", bms
->cluster_name
);
309 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
310 printf(" Creation Time : %.24s\n", ctime(&atime
));
311 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
312 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
313 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
315 printf(" Avail Dev Size : %llu%s\n",
316 (unsigned long long)__le64_to_cpu(sb
->data_size
),
317 human_size(__le64_to_cpu(sb
->data_size
)<<9));
318 if (__le32_to_cpu(sb
->level
) > 0) {
319 int ddsks
= 0, ddsks_denom
= 1;
320 switch(__le32_to_cpu(sb
->level
)) {
321 case 1: ddsks
=1;break;
323 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
324 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
326 layout
= __le32_to_cpu(sb
->layout
);
327 ddsks
= __le32_to_cpu(sb
->raid_disks
);
328 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
331 long long asize
= __le64_to_cpu(sb
->size
);
332 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
333 printf(" Array Size : %llu%s\n",
334 asize
>> 10, human_size(asize
));
336 if (sb
->size
!= sb
->data_size
)
337 printf(" Used Dev Size : %llu%s\n",
338 (unsigned long long)__le64_to_cpu(sb
->size
),
339 human_size(__le64_to_cpu(sb
->size
)<<9));
342 printf(" Data Offset : %llu sectors\n",
343 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
344 if (sb
->new_offset
&&
345 (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
)) {
346 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
347 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
348 printf(" New Offset : %llu sectors\n", offset
);
350 printf(" Super Offset : %llu sectors\n",
351 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
352 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
353 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
355 st
->ss
->getinfo_super(st
, &info
, NULL
);
356 if (info
.space_after
!= 1 &&
357 !(__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
358 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
359 info
.space_before
, info
.space_after
);
361 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
362 printf(" Device UUID : ");
363 for (i
=0; i
<16; i
++) {
364 if ((i
&3)==0 && i
!= 0) printf(":");
365 printf("%02x", sb
->device_uuid
[i
]);
369 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
370 printf("Internal Bitmap : %ld sectors from superblock\n",
371 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
373 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
374 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
375 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
376 if (__le32_to_cpu(sb
->delta_disks
)) {
377 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
378 printf(" (%d->%d)\n",
379 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
380 __le32_to_cpu(sb
->raid_disks
));
381 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
382 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
384 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
385 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
386 printf(" New Level : %s\n", c
?c
:"-unknown-");
388 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
389 if (__le32_to_cpu(sb
->level
) == 5) {
390 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
391 printf(" New Layout : %s\n", c
?c
:"-unknown-");
393 if (__le32_to_cpu(sb
->level
) == 6) {
394 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
395 printf(" New Layout : %s\n", c
?c
:"-unknown-");
397 if (__le32_to_cpu(sb
->level
) == 10) {
398 printf(" New Layout :");
399 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
403 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
404 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
409 if (sb
->devflags
& WriteMostly1
)
410 printf(" write-mostly");
414 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
415 printf(" Update Time : %.24s\n", ctime(&atime
));
417 if (sb
->bblog_size
&& sb
->bblog_offset
) {
418 printf(" Bad Block Log : %d entries available at offset %ld sectors",
419 __le16_to_cpu(sb
->bblog_size
)*512/8,
420 (long)(int32_t)__le32_to_cpu(sb
->bblog_offset
));
421 if (sb
->feature_map
&
422 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
423 printf(" - bad blocks present.");
427 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
428 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
430 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
431 __le32_to_cpu(calc_sb_1_csum(sb
)));
432 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
434 if (__le32_to_cpu(sb
->level
) == 5) {
435 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
436 printf(" Layout : %s\n", c
?c
:"-unknown-");
438 if (__le32_to_cpu(sb
->level
) == 6) {
439 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
440 printf(" Layout : %s\n", c
?c
:"-unknown-");
442 if (__le32_to_cpu(sb
->level
) == 10) {
443 int lo
= __le32_to_cpu(sb
->layout
);
445 print_r10_layout(lo
);
448 switch(__le32_to_cpu(sb
->level
)) {
454 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
457 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
463 /* This turns out to just be confusing */
464 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
465 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
466 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != 0xffff)
468 for (d
=0; d
< i
; d
++) {
469 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
471 if (role
== 0xffff) printf("empty");
472 else if(role
== 0xfffe) printf("failed");
473 else printf("%d", role
);
477 printf(" Device Role : ");
478 d
= __le32_to_cpu(sb
->dev_number
);
479 if (d
< __le32_to_cpu(sb
->max_dev
))
480 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
485 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
486 printf("Replacement device %d\n", role
);
488 printf("Active device %d\n", role
);
490 printf(" Array State : ");
491 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
494 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
495 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
509 /* This is confusing too */
511 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
512 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
516 if (faulty
) printf(" %d failed", faulty
);
518 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
522 static void brief_examine_super1(struct supertype
*st
, int verbose
)
524 struct mdp_superblock_1
*sb
= st
->sb
;
526 unsigned long long sb_offset
;
528 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
530 nm
= strchr(sb
->set_name
, ':');
533 else if (sb
->set_name
[0])
545 printf(" level=%s", c
);
546 sb_offset
= __le64_to_cpu(sb
->super_offset
);
548 printf(" metadata=1.1 ");
549 else if (sb_offset
<= 8)
550 printf(" metadata=1.2 ");
552 printf(" metadata=1.0 ");
554 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
556 for (i
=0; i
<16; i
++) {
557 if ((i
&3)==0 && i
!= 0) printf(":");
558 printf("%02x", sb
->set_uuid
[i
]);
560 if (sb
->set_name
[0]) {
562 print_quoted(sb
->set_name
);
567 static void export_examine_super1(struct supertype
*st
)
569 struct mdp_superblock_1
*sb
= st
->sb
;
574 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
575 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
577 if (sb
->set_name
[i
] == '\n' ||
578 sb
->set_name
[i
] == '\0') {
583 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
584 if (__le32_to_cpu(sb
->level
) > 0) {
585 int ddsks
= 0, ddsks_denom
= 1;
586 switch(__le32_to_cpu(sb
->level
)) {
587 case 1: ddsks
=1;break;
589 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
590 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
592 layout
= __le32_to_cpu(sb
->layout
);
593 ddsks
= __le32_to_cpu(sb
->raid_disks
);
594 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
597 long long asize
= __le64_to_cpu(sb
->size
);
598 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
599 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
603 for (i
=0; i
<16; i
++) {
604 if ((i
&3)==0 && i
!= 0) printf(":");
605 printf("%02x", sb
->set_uuid
[i
]);
608 printf("MD_UPDATE_TIME=%llu\n",
609 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
610 printf("MD_DEV_UUID=");
611 for (i
=0; i
<16; i
++) {
612 if ((i
&3)==0 && i
!= 0) printf(":");
613 printf("%02x", sb
->device_uuid
[i
]);
616 printf("MD_EVENTS=%llu\n",
617 (unsigned long long)__le64_to_cpu(sb
->events
));
620 static int copy_metadata1(struct supertype
*st
, int from
, int to
)
622 /* Read superblock. If it looks good, write it out.
623 * Then if a bitmap is present, copy that.
624 * And if a bad-block-list is present, copy that too.
627 unsigned long long dsize
, sb_offset
;
628 const int bufsize
= 4*1024;
629 struct mdp_superblock_1 super
, *sb
;
631 if (posix_memalign(&buf
, 4096, bufsize
) != 0)
634 if (!get_dev_size(from
, NULL
, &dsize
))
640 switch(st
->minor_version
) {
644 sb_offset
&= ~(4*2-1);
656 if (lseek64(from
, sb_offset
<< 9, 0) < 0LL)
658 if (read(from
, buf
, bufsize
) != bufsize
)
662 super
= *sb
; // save most of sb for when we reuse buf
664 if (__le32_to_cpu(super
.magic
) != MD_SB_MAGIC
||
665 __le32_to_cpu(super
.major_version
) != 1 ||
666 __le64_to_cpu(super
.super_offset
) != sb_offset
||
667 calc_sb_1_csum(sb
) != super
.sb_csum
)
670 if (lseek64(to
, sb_offset
<< 9, 0) < 0LL)
672 if (write(to
, buf
, bufsize
) != bufsize
)
675 if (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
)) {
676 unsigned long long bitmap_offset
= sb_offset
;
677 int bytes
= 4096; // just an estimate.
679 struct align_fd afrom
, ato
;
681 init_afd(&afrom
, from
);
684 bitmap_offset
+= (int32_t)__le32_to_cpu(super
.bitmap_offset
);
686 if (lseek64(from
, bitmap_offset
<<9, 0) < 0)
688 if (lseek64(to
, bitmap_offset
<<9, 0) < 0)
691 for (written
= 0; written
< bytes
; ) {
692 int n
= bytes
- written
;
695 if (aread(&afrom
, buf
, n
) != n
)
698 /* have the header, can calculate
699 * correct bitmap bytes */
702 bytes
= calc_bitmap_size(bms
, 512);
706 if (awrite(&ato
, buf
, n
) != n
)
712 if (super
.bblog_size
!= 0 &&
713 __le32_to_cpu(super
.bblog_size
) <= 100 &&
714 super
.bblog_offset
!= 0 &&
715 (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BAD_BLOCKS
))) {
716 /* There is a bad block log */
717 unsigned long long bb_offset
= sb_offset
;
718 int bytes
= __le32_to_cpu(super
.bblog_size
) * 512;
720 struct align_fd afrom
, ato
;
722 init_afd(&afrom
, from
);
725 bb_offset
+= (int32_t)__le32_to_cpu(super
.bblog_offset
);
727 if (lseek64(from
, bb_offset
<<9, 0) < 0)
729 if (lseek64(to
, bb_offset
<<9, 0) < 0)
732 for (written
= 0; written
< bytes
; ) {
733 int n
= bytes
- written
;
736 if (aread(&afrom
, buf
, n
) != n
)
739 if (awrite(&ato
, buf
, n
) != n
)
753 static void detail_super1(struct supertype
*st
, char *homehost
)
755 struct mdp_superblock_1
*sb
= st
->sb
;
756 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
758 int l
= homehost
? strlen(homehost
) : 0;
760 printf(" Name : %.32s", sb
->set_name
);
761 if (l
> 0 && l
< 32 &&
762 sb
->set_name
[l
] == ':' &&
763 strncmp(sb
->set_name
, homehost
, l
) == 0)
764 printf(" (local to host %s)", homehost
);
766 printf("Cluster Name : %64s", bms
->cluster_name
);
767 printf("\n UUID : ");
768 for (i
=0; i
<16; i
++) {
769 if ((i
&3)==0 && i
!= 0) printf(":");
770 printf("%02x", sb
->set_uuid
[i
]);
772 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb
->events
));
775 static void brief_detail_super1(struct supertype
*st
)
777 struct mdp_superblock_1
*sb
= st
->sb
;
780 if (sb
->set_name
[0]) {
782 print_quoted(sb
->set_name
);
785 for (i
=0; i
<16; i
++) {
786 if ((i
&3)==0 && i
!= 0) printf(":");
787 printf("%02x", sb
->set_uuid
[i
]);
791 static void export_detail_super1(struct supertype
*st
)
793 struct mdp_superblock_1
*sb
= st
->sb
;
798 if (sb
->set_name
[i
] == '\n' ||
799 sb
->set_name
[i
] == '\0') {
804 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
807 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
809 struct mdp_superblock_1
*sb
= st
->sb
;
810 unsigned long long offset
;
815 if (!sb
->bblog_size
|| __le32_to_cpu(sb
->bblog_size
) > 100
816 || !sb
->bblog_offset
){
817 printf("No bad-blocks list configured on %s\n", devname
);
820 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
822 printf("Bad-blocks list is empty in %s\n", devname
);
826 size
= __le32_to_cpu(sb
->bblog_size
)* 512;
827 if (posix_memalign((void**)&bbl
, 4096, size
) != 0) {
828 pr_err("could not allocate badblocks list\n");
831 offset
= __le64_to_cpu(sb
->super_offset
) +
832 (int)__le32_to_cpu(sb
->bblog_offset
);
834 if (lseek64(fd
, offset
, 0) < 0) {
835 pr_err("Cannot seek to bad-blocks list\n");
838 if (read(fd
, bbl
, size
) != size
) {
839 pr_err("Cannot read bad-blocks list\n");
842 /* 64bits per entry. 10 bits is block-count, 54 bits is block
843 * offset. Blocks are sectors unless bblog->shift makes them bigger
846 printf("Bad-blocks on %s:\n", devname
);
847 for (i
= 0; i
< size
/8; i
++, bbp
++) {
848 __u64 bb
= __le64_to_cpu(*bbp
);
849 int count
= bb
& 0x3ff;
850 unsigned long long sector
= bb
>> 10;
855 sector
<<= sb
->bblog_shift
;
856 count
<<= sb
->bblog_shift
;
858 printf("%20llu for %d sectors\n", sector
, count
);
865 static int match_home1(struct supertype
*st
, char *homehost
)
867 struct mdp_superblock_1
*sb
= st
->sb
;
868 int l
= homehost
? strlen(homehost
) : 0;
870 return (l
> 0 && l
< 32 &&
871 sb
->set_name
[l
] == ':' &&
872 strncmp(sb
->set_name
, homehost
, l
) == 0);
875 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
877 struct mdp_superblock_1
*super
= st
->sb
;
878 char *cuuid
= (char*)uuid
;
881 cuuid
[i
] = super
->set_uuid
[i
];
884 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
886 struct mdp_superblock_1
*sb
= st
->sb
;
887 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
888 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
892 unsigned int map_disks
= info
->array
.raid_disks
;
893 unsigned long long super_offset
;
894 unsigned long long data_size
;
896 memset(info
, 0, sizeof(*info
));
897 info
->array
.major_version
= 1;
898 info
->array
.minor_version
= st
->minor_version
;
899 info
->array
.patch_version
= 0;
900 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
901 info
->array
.level
= __le32_to_cpu(sb
->level
);
902 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
903 info
->array
.md_minor
= -1;
904 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
905 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
906 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
908 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
910 if (__le32_to_cpu(bsb
->nodes
) > 1)
911 info
->array
.state
|= (1 << MD_SB_CLUSTERED
);
913 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
914 info
->component_size
= __le64_to_cpu(sb
->size
);
915 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
916 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
918 info
->disk
.major
= 0;
919 info
->disk
.minor
= 0;
920 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
921 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
922 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
925 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
927 super_offset
= __le64_to_cpu(sb
->super_offset
);
928 if (info
->array
.level
<= 0)
929 data_size
= __le64_to_cpu(sb
->data_size
);
931 data_size
= __le64_to_cpu(sb
->size
);
932 if (info
->data_offset
< super_offset
) {
933 unsigned long long end
;
934 info
->space_before
= info
->data_offset
;
937 if (sb
->bblog_offset
&& sb
->bblog_size
) {
938 unsigned long long bboffset
= super_offset
;
939 bboffset
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
944 if (super_offset
+ info
->bitmap_offset
< end
)
945 end
= super_offset
+ info
->bitmap_offset
;
947 if (info
->data_offset
+ data_size
< end
)
948 info
->space_after
= end
- data_size
- info
->data_offset
;
950 info
->space_after
= 0;
952 unsigned long long earliest
;
953 earliest
= super_offset
+ (32+4)*2; /* match kernel */
954 if (info
->bitmap_offset
> 0) {
955 unsigned long long bmend
= info
->bitmap_offset
;
956 unsigned long long size
= __le64_to_cpu(bsb
->sync_size
);
957 size
/= __le32_to_cpu(bsb
->chunksize
) >> 9;
958 size
= (size
+ 7) >> 3;
959 size
+= sizeof(bitmap_super_t
);
960 size
= ROUND_UP(size
, 4096);
963 if (bmend
> earliest
)
966 if (sb
->bblog_offset
&& sb
->bblog_size
) {
967 unsigned long long bbend
= super_offset
;
968 bbend
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
969 bbend
+= __le32_to_cpu(sb
->bblog_size
);
970 if (bbend
> earliest
)
973 if (earliest
< info
->data_offset
)
974 info
->space_before
= info
->data_offset
- earliest
;
976 info
->space_before
= 0;
977 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
979 if (info
->space_before
== 0 && info
->space_after
== 0) {
980 /* It will look like we don't support data_offset changes,
981 * be we do - it's just that there is no room.
982 * A change that reduced the number of devices should
983 * still be allowed, so set the otherwise useless value of '1'
985 info
->space_after
= 1;
988 info
->disk
.raid_disk
= -1;
991 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
994 info
->disk
.state
= 1; /* faulty */
997 info
->disk
.state
= 6; /* active and in sync */
998 info
->disk
.raid_disk
= role
;
1000 if (sb
->devflags
& WriteMostly1
)
1001 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1002 info
->events
= __le64_to_cpu(sb
->events
);
1003 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
1004 info
->safe_mode_delay
= 200;
1006 memcpy(info
->uuid
, sb
->set_uuid
, 16);
1008 strncpy(info
->name
, sb
->set_name
, 32);
1011 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
1012 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
1013 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
1016 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
1017 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
1019 info
->recovery_start
= MaxSector
;
1021 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
1022 info
->reshape_active
= 1;
1023 if ((sb
->feature_map
& __le32_to_cpu(MD_FEATURE_NEW_OFFSET
)) &&
1024 sb
->new_offset
!= 0)
1025 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
1026 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
1027 info
->new_level
= __le32_to_cpu(sb
->new_level
);
1028 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
1029 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
1030 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
1031 if (info
->delta_disks
< 0)
1032 info
->array
.raid_disks
-= info
->delta_disks
;
1034 info
->reshape_active
= 0;
1036 info
->recovery_blocked
= info
->reshape_active
;
1039 for (i
=0; i
<map_disks
; i
++)
1041 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
1042 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
1043 if (/*role == 0xFFFF || */role
< (unsigned) info
->array
.raid_disks
) {
1045 if (map
&& role
< map_disks
)
1050 info
->array
.working_disks
= working
;
1053 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
1055 struct mdinfo
*info
;
1060 info
= xmalloc(sizeof(*info
));
1061 getinfo_super1(st
, info
, NULL
);
1065 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
1067 char *devname
, int verbose
,
1068 int uuid_set
, char *homehost
)
1070 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1071 * if any change was made. For others, the return value is
1075 struct mdp_superblock_1
*sb
= st
->sb
;
1077 if (strcmp(update
, "homehost") == 0 &&
1079 /* Note that 'homehost' is special as it is really
1084 c
= strchr(sb
->set_name
, ':');
1086 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1088 strncpy(info
->name
, sb
->set_name
, 32);
1092 if (strcmp(update
, "force-one")==0) {
1093 /* Not enough devices for a working array,
1094 * so bring this one up-to-date
1096 if (sb
->events
!= __cpu_to_le64(info
->events
))
1098 sb
->events
= __cpu_to_le64(info
->events
);
1099 } else if (strcmp(update
, "force-array")==0) {
1100 /* Degraded array and 'force' requests to
1101 * maybe need to mark it 'clean'.
1103 switch(__le32_to_cpu(sb
->level
)) {
1104 case 5: case 4: case 6:
1105 /* need to force clean */
1106 if (sb
->resync_offset
!= MaxSector
)
1108 sb
->resync_offset
= MaxSector
;
1110 } else if (strcmp(update
, "assemble")==0) {
1111 int d
= info
->disk
.number
;
1113 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
1114 want
= info
->disk
.raid_disk
;
1117 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
1118 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
1121 if (info
->reshape_active
&&
1122 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1123 info
->delta_disks
>= 0 &&
1124 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
1125 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1128 if (info
->reshape_active
&&
1129 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1130 info
->delta_disks
< 0 &&
1131 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
1132 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1135 } else if (strcmp(update
, "linear-grow-new") == 0) {
1138 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
1140 for (i
=0 ; i
< max
; i
++)
1141 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= 0xfffe)
1143 sb
->dev_number
= __cpu_to_le32(i
);
1144 info
->disk
.number
= i
;
1145 if (max
>= __le32_to_cpu(sb
->max_dev
))
1146 sb
->max_dev
= __cpu_to_le32(max
+1);
1148 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1149 read(rfd
, sb
->device_uuid
, 16) != 16) {
1150 __u32 r
[4] = {random(), random(), random(), random()};
1151 memcpy(sb
->device_uuid
, r
, 16);
1157 __cpu_to_le16(info
->disk
.raid_disk
);
1159 fd
= open(devname
, O_RDONLY
);
1161 unsigned long long ds
;
1162 get_dev_size(fd
, devname
, &ds
);
1165 if (__le64_to_cpu(sb
->super_offset
) <
1166 __le64_to_cpu(sb
->data_offset
)) {
1167 sb
->data_size
= __cpu_to_le64(
1168 ds
- __le64_to_cpu(sb
->data_offset
));
1171 ds
&= ~(unsigned long long)(4*2-1);
1172 sb
->super_offset
= __cpu_to_le64(ds
);
1173 sb
->data_size
= __cpu_to_le64(
1174 ds
- __le64_to_cpu(sb
->data_offset
));
1177 } else if (strcmp(update
, "linear-grow-update") == 0) {
1178 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1179 sb
->dev_roles
[info
->disk
.number
] =
1180 __cpu_to_le16(info
->disk
.raid_disk
);
1181 } else if (strcmp(update
, "resync") == 0) {
1182 /* make sure resync happens */
1183 sb
->resync_offset
= 0ULL;
1184 } else if (strcmp(update
, "uuid") == 0) {
1185 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1187 if (__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1188 struct bitmap_super_s
*bm
;
1189 bm
= (struct bitmap_super_s
*)(st
->sb
+MAX_SB_SIZE
);
1190 memcpy(bm
->uuid
, sb
->set_uuid
, 16);
1192 } else if (strcmp(update
, "no-bitmap") == 0) {
1193 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1194 } else if (strcmp(update
, "bbl") == 0) {
1195 /* only possible if there is room after the bitmap, or if
1196 * there is no bitmap
1198 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1199 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1200 long bitmap_offset
= (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
);
1201 long bm_sectors
= 0;
1205 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1206 struct bitmap_super_s
*bsb
;
1207 bsb
= (struct bitmap_super_s
*)(((char*)sb
)+MAX_SB_SIZE
);
1208 bm_sectors
= bitmap_sectors(bsb
);
1211 if (sb_offset
< data_offset
) {
1212 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1215 bb_offset
= sb_offset
+ 8;
1216 if (bm_sectors
&& bitmap_offset
> 0)
1217 bb_offset
= bitmap_offset
+ bm_sectors
;
1218 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1219 && bb_offset
+ 8+8 <= (long)data_offset
)
1220 /* too close to bitmap, and room to grow */
1222 if (bb_offset
+ 8 <= (long)data_offset
) {
1223 sb
->bblog_size
= __cpu_to_le16(8);
1224 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1227 /* 1.0 - Put bbl just before super block */
1228 if (bm_sectors
&& bitmap_offset
< 0)
1229 space
= -bitmap_offset
- bm_sectors
;
1231 space
= sb_offset
- data_offset
-
1232 __le64_to_cpu(sb
->data_size
);
1234 sb
->bblog_size
= __cpu_to_le16(8);
1235 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1238 } else if (strcmp(update
, "no-bbl") == 0) {
1239 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1240 pr_err("Cannot remove active bbl from %s\n",devname
);
1243 sb
->bblog_shift
= 0;
1244 sb
->bblog_offset
= 0;
1246 } else if (strcmp(update
, "name") == 0) {
1247 if (info
->name
[0] == 0)
1248 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1249 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1251 strchr(info
->name
, ':') == NULL
&&
1252 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1253 strcpy(sb
->set_name
, homehost
);
1254 strcat(sb
->set_name
, ":");
1255 strcat(sb
->set_name
, info
->name
);
1257 strcpy(sb
->set_name
, info
->name
);
1258 } else if (strcmp(update
, "devicesize") == 0 &&
1259 __le64_to_cpu(sb
->super_offset
) <
1260 __le64_to_cpu(sb
->data_offset
)) {
1261 /* set data_size to device size less data_offset */
1262 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1263 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1264 sb
->data_size
= __cpu_to_le64(
1265 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1266 } else if (strcmp(update
, "revert-reshape") == 0) {
1268 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1269 pr_err("No active reshape to revert on %s\n",
1273 unsigned long long reshape_sectors
;
1276 /* reshape_position is a little messy.
1277 * Its value must be a multiple of the larger
1278 * chunk size, and of the "after" data disks.
1279 * So when reverting we need to change it to
1280 * be a multiple of the new "after" data disks,
1281 * which is the old "before".
1282 * If it isn't already a multiple of 'before',
1283 * the only thing we could do would be
1284 * copy some block around on the disks, which
1285 * is easy to get wrong.
1286 * So we reject a revert-reshape unless the
1287 * alignment is good.
1289 if (__le32_to_cpu(sb
->level
) >= 4 &&
1290 __le32_to_cpu(sb
->level
) <= 6) {
1291 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1292 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1293 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1294 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1295 if (reshape_sectors
% reshape_chunk
) {
1296 pr_err("Reshape position is not suitably aligned.\n");
1297 pr_err("Try normal assembly and stop again\n");
1301 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1302 __le32_to_cpu(sb
->delta_disks
));
1303 if (sb
->delta_disks
== 0)
1304 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1306 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1308 temp
= sb
->new_layout
;
1309 sb
->new_layout
= sb
->layout
;
1312 temp
= sb
->new_chunk
;
1313 sb
->new_chunk
= sb
->chunksize
;
1314 sb
->chunksize
= temp
;
1316 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1317 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1318 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1319 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1320 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1323 } else if (strcmp(update
, "_reshape_progress")==0)
1324 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1325 else if (strcmp(update
, "writemostly")==0)
1326 sb
->devflags
|= WriteMostly1
;
1327 else if (strcmp(update
, "readwrite")==0)
1328 sb
->devflags
&= ~WriteMostly1
;
1332 sb
->sb_csum
= calc_sb_1_csum(sb
);
1336 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1337 unsigned long long size
, char *name
, char *homehost
,
1338 int *uuid
, unsigned long long data_offset
)
1340 struct mdp_superblock_1
*sb
;
1346 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1347 pr_err("could not allocate superblock\n");
1350 memset(sb
, 0, SUPER1_SIZE
);
1354 /* zeroing superblock */
1358 spares
= info
->working_disks
- info
->active_disks
;
1359 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1360 pr_err("too many devices requested: %d+%d > %d\n",
1361 info
->raid_disks
, spares
, MAX_DEVS
);
1365 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1366 sb
->major_version
= __cpu_to_le32(1);
1367 sb
->feature_map
= 0;
1371 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1373 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1374 read(rfd
, sb
->set_uuid
, 16) != 16) {
1375 __u32 r
[4] = {random(), random(), random(), random()};
1376 memcpy(sb
->set_uuid
, r
, 16);
1378 if (rfd
>= 0) close(rfd
);
1381 if (name
== NULL
|| *name
== 0) {
1382 sprintf(defname
, "%d", info
->md_minor
);
1386 strchr(name
, ':')== NULL
&&
1387 strlen(homehost
)+1+strlen(name
) < 32) {
1388 strcpy(sb
->set_name
, homehost
);
1389 strcat(sb
->set_name
, ":");
1390 strcat(sb
->set_name
, name
);
1392 strcpy(sb
->set_name
, name
);
1394 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1395 sb
->level
= __cpu_to_le32(info
->level
);
1396 sb
->layout
= __cpu_to_le32(info
->layout
);
1397 sb
->size
= __cpu_to_le64(size
*2ULL);
1398 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1399 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1401 sb
->data_offset
= __cpu_to_le64(data_offset
);
1402 sb
->data_size
= __cpu_to_le64(0);
1403 sb
->super_offset
= __cpu_to_le64(0);
1404 sb
->recovery_offset
= __cpu_to_le64(0);
1406 sb
->utime
= sb
->ctime
;
1407 sb
->events
= __cpu_to_le64(1);
1408 if (info
->state
& (1<<MD_SB_CLEAN
))
1409 sb
->resync_offset
= MaxSector
;
1411 sb
->resync_offset
= 0;
1412 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1413 sbsize
= ROUND_UP(sbsize
, 512);
1414 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1416 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1424 long long data_offset
;
1425 mdu_disk_info_t disk
;
1426 struct devinfo
*next
;
1429 /* Add a device to the superblock being created */
1430 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1431 int fd
, char *devname
, unsigned long long data_offset
)
1433 struct mdp_superblock_1
*sb
= st
->sb
;
1434 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1435 struct devinfo
*di
, **dip
;
1437 if ((dk
->state
& 6) == 6) /* active, sync */
1438 *rp
= __cpu_to_le16(dk
->raid_disk
);
1439 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1444 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1445 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1446 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1448 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1449 sb
->devflags
= 0; /* don't copy another disks flags */
1450 sb
->sb_csum
= calc_sb_1_csum(sb
);
1452 dip
= (struct devinfo
**)&st
->info
;
1454 dip
= &(*dip
)->next
;
1455 di
= xmalloc(sizeof(struct devinfo
));
1457 di
->devname
= devname
;
1459 di
->data_offset
= data_offset
;
1467 static void locate_bitmap1(struct supertype
*st
, int fd
);
1469 static int store_super1(struct supertype
*st
, int fd
)
1471 struct mdp_superblock_1
*sb
= st
->sb
;
1472 unsigned long long sb_offset
;
1473 struct align_fd afd
;
1475 unsigned long long dsize
;
1477 if (!get_dev_size(fd
, NULL
, &dsize
))
1488 * Calculate the position of the superblock.
1489 * It is always aligned to a 4K boundary and
1490 * depending on minor_version, it can be:
1491 * 0: At least 8K, but less than 12K, from end of device
1492 * 1: At start of device
1493 * 2: 4K from start of device.
1495 switch(st
->minor_version
) {
1499 sb_offset
&= ~(4*2-1);
1511 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1512 0 != __le64_to_cpu(sb
->super_offset
)
1514 pr_err("internal error - sb_offset is wrong\n");
1518 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1521 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1523 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1526 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1527 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1528 (((char*)sb
)+MAX_SB_SIZE
);
1529 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1530 locate_bitmap1(st
, fd
);
1531 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1539 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1541 static unsigned long choose_bm_space(unsigned long devsize
)
1543 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1544 * if bigger than 200Gig, save 128k
1545 * NOTE: result must be multiple of 4K else bad things happen
1546 * on 4K-sector devices.
1548 if (devsize
< 64*2) return 0;
1549 if (devsize
- 64*2 >= 200*1024*1024*2)
1551 if (devsize
- 4*2 > 8*1024*1024*2)
1556 static void free_super1(struct supertype
*st
);
1559 static int write_init_super1(struct supertype
*st
)
1561 struct mdp_superblock_1
*sb
= st
->sb
;
1562 struct supertype
*refst
;
1565 unsigned long long bm_space
;
1567 unsigned long long dsize
, array_size
;
1568 unsigned long long sb_offset
;
1569 unsigned long long data_offset
;
1571 for (di
= st
->info
; di
; di
= di
->next
) {
1572 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1577 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1580 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1581 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1582 sb
->devflags
|= WriteMostly1
;
1584 sb
->devflags
&= ~WriteMostly1
;
1586 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1587 read(rfd
, sb
->device_uuid
, 16) != 16) {
1588 __u32 r
[4] = {random(), random(), random(), random()};
1589 memcpy(sb
->device_uuid
, r
, 16);
1596 refst
= dup_super(st
);
1597 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1598 struct mdp_superblock_1
*refsb
= refst
->sb
;
1600 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1601 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1602 /* same array, so preserve events and
1604 sb
->events
= refsb
->events
;
1605 /* bugs in 2.6.17 and earlier mean the
1606 * dev_number chosen in Manage must be preserved
1608 if (get_linux_version() >= 2006018)
1609 sb
->dev_number
= refsb
->dev_number
;
1615 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1628 * Calculate the position of the superblock.
1629 * It is always aligned to a 4K boundary and
1630 * depending on minor_version, it can be:
1631 * 0: At least 8K, but less than 12K, from end of device
1632 * 1: At start of device
1633 * 2: 4K from start of device.
1634 * data_offset has already been set.
1636 array_size
= __le64_to_cpu(sb
->size
);
1637 /* work out how much space we left for a bitmap,
1638 * Add 8 sectors for bad block log */
1639 bm_space
= choose_bm_space(array_size
) + 8;
1641 data_offset
= di
->data_offset
;
1642 if (data_offset
== INVALID_SECTORS
)
1643 data_offset
= st
->data_offset
;
1644 switch(st
->minor_version
) {
1646 if (data_offset
== INVALID_SECTORS
)
1650 sb_offset
&= ~(4*2-1);
1651 sb
->data_offset
= __cpu_to_le64(data_offset
);
1652 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1653 if (sb_offset
< array_size
+ bm_space
)
1654 bm_space
= sb_offset
- array_size
;
1655 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1656 if (bm_space
>= 8) {
1657 sb
->bblog_size
= __cpu_to_le16(8);
1658 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1662 sb
->super_offset
= __cpu_to_le64(0);
1663 if (data_offset
== INVALID_SECTORS
)
1666 sb
->data_offset
= __cpu_to_le64(data_offset
);
1667 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1668 if (data_offset
>= 8 + 32*2 + 8) {
1669 sb
->bblog_size
= __cpu_to_le16(8);
1670 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1671 } else if (data_offset
>= 16) {
1672 sb
->bblog_size
= __cpu_to_le16(8);
1673 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8);
1678 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1679 if (data_offset
== INVALID_SECTORS
)
1682 sb
->data_offset
= __cpu_to_le64(data_offset
);
1683 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1684 if (data_offset
>= 16 + 32*2 + 8) {
1685 sb
->bblog_size
= __cpu_to_le16(8);
1686 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1687 } else if (data_offset
>= 16+16) {
1688 sb
->bblog_size
= __cpu_to_le16(8);
1689 /* '8' sectors for the bblog, and another '8'
1690 * because we want offset from superblock, not
1693 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8-8);
1697 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1698 st
->minor_version
, di
->devname
);
1702 if (conf_get_create_info()->bblist
== 0) {
1704 sb
->bblog_offset
= 0;
1707 sb
->sb_csum
= calc_sb_1_csum(sb
);
1708 rv
= store_super1(st
, di
->fd
);
1709 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1710 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NoUpdate
);
1718 pr_err("Failed to write metadata to %s\n",
1725 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1729 * 0 same, or first was empty, and second was copied
1730 * 1 second had wrong number
1732 * 3 wrong other info
1734 struct mdp_superblock_1
*first
= st
->sb
;
1735 struct mdp_superblock_1
*second
= tst
->sb
;
1737 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1739 if (second
->major_version
!= __cpu_to_le32(1))
1743 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1744 pr_err("could not allocate superblock\n");
1747 memcpy(first
, second
, SUPER1_SIZE
);
1751 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1754 if (first
->ctime
!= second
->ctime
||
1755 first
->level
!= second
->level
||
1756 first
->layout
!= second
->layout
||
1757 first
->size
!= second
->size
||
1758 first
->chunksize
!= second
->chunksize
||
1759 first
->raid_disks
!= second
->raid_disks
)
1764 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1766 unsigned long long dsize
;
1767 unsigned long long sb_offset
;
1768 struct mdp_superblock_1
*super
;
1770 struct bitmap_super_s
*bsb
;
1771 struct misc_dev_info
*misc
;
1772 struct align_fd afd
;
1778 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1780 struct supertype tst
;
1781 __u64 bestctime
= 0;
1782 /* guess... choose latest ctime */
1783 memset(&tst
, 0, sizeof(tst
));
1785 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1786 switch(load_super1(&tst
, fd
, devname
)) {
1787 case 0: super
= tst
.sb
;
1788 if (bestvers
== -1 ||
1789 bestctime
< __le64_to_cpu(super
->ctime
)) {
1790 bestvers
= tst
.minor_version
;
1791 bestctime
= __le64_to_cpu(super
->ctime
);
1796 case 1: return 1; /*bad device */
1797 case 2: break; /* bad, try next */
1800 if (bestvers
!= -1) {
1802 tst
.minor_version
= bestvers
;
1804 tst
.max_devs
= MAX_DEVS
;
1805 rv
= load_super1(&tst
, fd
, devname
);
1812 if (!get_dev_size(fd
, devname
, &dsize
))
1818 pr_err("%s is too small for md: size is %llu sectors.\n",
1824 * Calculate the position of the superblock.
1825 * It is always aligned to a 4K boundary and
1826 * depending on minor_version, it can be:
1827 * 0: At least 8K, but less than 12K, from end of device
1828 * 1: At start of device
1829 * 2: 4K from start of device.
1831 switch(st
->minor_version
) {
1835 sb_offset
&= ~(4*2-1);
1847 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1849 pr_err("Cannot seek to superblock on %s: %s\n",
1850 devname
, strerror(errno
));
1854 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
1855 pr_err("could not allocate superblock\n");
1859 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
1861 pr_err("Cannot read superblock on %s\n",
1867 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
1869 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
1870 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
1875 if (__le32_to_cpu(super
->major_version
) != 1) {
1877 pr_err("Cannot interpret superblock on %s - version is %d\n",
1878 devname
, __le32_to_cpu(super
->major_version
));
1882 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
1884 pr_err("No superblock found on %s (super_offset is wrong)\n",
1891 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
1893 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
1894 misc
->device_size
= dsize
;
1895 if (st
->data_offset
== INVALID_SECTORS
)
1896 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
1898 /* Now check on the bitmap superblock */
1899 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
1901 /* Read the bitmap superblock and make sure it looks
1902 * valid. If it doesn't clear the bit. An --assemble --force
1903 * should get that written out.
1905 locate_bitmap1(st
, fd
);
1906 if (aread(&afd
, bsb
, 512) != 512)
1909 uuid_from_super1(st
, uuid
);
1910 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
1911 memcmp(bsb
->uuid
, uuid
, 16) != 0)
1916 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
1917 & ~MD_FEATURE_BITMAP_OFFSET
);
1921 static struct supertype
*match_metadata_desc1(char *arg
)
1923 struct supertype
*st
= xcalloc(1, sizeof(*st
));
1925 st
->container_devnm
[0] = 0;
1927 st
->max_devs
= MAX_DEVS
;
1929 st
->data_offset
= INVALID_SECTORS
;
1930 /* leading zeros can be safely ignored. --detail generates them. */
1933 if (strcmp(arg
, "1.0") == 0 ||
1934 strcmp(arg
, "1.00") == 0) {
1935 st
->minor_version
= 0;
1938 if (strcmp(arg
, "1.1") == 0 ||
1939 strcmp(arg
, "1.01") == 0
1941 st
->minor_version
= 1;
1944 if (strcmp(arg
, "1.2") == 0 ||
1945 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
1946 strcmp(arg
, "default") == 0 ||
1947 #endif /* DEFAULT_OLD_METADATA */
1948 strcmp(arg
, "1.02") == 0) {
1949 st
->minor_version
= 2;
1952 if (strcmp(arg
, "1") == 0 ||
1953 strcmp(arg
, "default") == 0) {
1954 st
->minor_version
= -1;
1962 /* find available size on device with this devsize, using
1963 * superblock type st, and reserving 'reserve' sectors for
1966 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
1967 unsigned long long data_offset
)
1969 struct mdp_superblock_1
*super
= st
->sb
;
1976 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1977 /* hot-add. allow for actual size of bitmap */
1978 struct bitmap_super_s
*bsb
;
1979 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
1980 bmspace
= bitmap_sectors(bsb
);
1983 /* Allow space for bad block log */
1984 if (super
->bblog_size
)
1985 bbspace
= __le16_to_cpu(super
->bblog_size
);
1987 if (st
->minor_version
< 0)
1988 /* not specified, so time to set default */
1989 st
->minor_version
= 2;
1991 if (data_offset
== INVALID_SECTORS
)
1992 data_offset
= st
->data_offset
;
1994 if (data_offset
!= INVALID_SECTORS
)
1995 switch(st
->minor_version
) {
1997 return devsize
- data_offset
- 8*2 - bbspace
;
2000 return devsize
- data_offset
;
2007 switch(st
->minor_version
) {
2010 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2012 /* at start, 4K for superblock and possible bitmap */
2013 return devsize
- 4*2 - bbspace
;
2015 /* 4k from start, 4K for superblock and possible bitmap */
2016 return devsize
- (4+4)*2 - bbspace
;
2022 add_internal_bitmap1(struct supertype
*st
,
2023 int *chunkp
, int delay
, int write_behind
,
2024 unsigned long long size
,
2025 int may_change
, int major
)
2028 * If not may_change, then this is a 'Grow' without sysfs support for
2029 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2030 * If may_change, then this is create or a Grow with sysfs syupport,
2031 * and we can put the bitmap wherever we like.
2033 * size is in sectors, chunk is in bytes !!!
2036 unsigned long long bits
;
2037 unsigned long long max_bits
;
2038 unsigned long long min_chunk
;
2040 long bbl_offset
, bbl_size
;
2041 unsigned long long chunk
= *chunkp
;
2044 struct mdp_superblock_1
*sb
= st
->sb
;
2045 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2048 if (__le64_to_cpu(sb
->data_size
) == 0)
2049 /* Must be creating the array, else data_size would be non-zero */
2051 switch(st
->minor_version
) {
2053 /* either 3K after the superblock (when hot-add),
2054 * or some amount of space before.
2057 /* We are creating array, so we *know* how much room has
2062 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2064 room
= __le64_to_cpu(sb
->super_offset
)
2065 - __le64_to_cpu(sb
->data_offset
)
2066 - __le64_to_cpu(sb
->data_size
);
2067 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2070 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2071 if (bbl_size
< -bbl_offset
)
2072 bbl_size
= -bbl_offset
;
2074 if (!may_change
|| (room
< 3*2 &&
2075 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2080 offset
= 0; /* means movable offset */
2085 case 2: /* between superblock and data */
2089 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2091 room
= __le64_to_cpu(sb
->data_offset
)
2092 - __le64_to_cpu(sb
->super_offset
);
2093 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2095 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2100 room
-= 2; /* Leave 1K for superblock */
2104 room
-= 4*2; /* leave 4K for superblock */
2114 if (chunk
== UnSet
&& room
> 128*2)
2115 /* Limit to 128K of bitmap when chunk size not requested */
2119 /* No room for a bitmap */
2122 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2124 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2125 bits
= (size
*512)/min_chunk
+1;
2126 while (bits
> max_bits
) {
2130 if (chunk
== UnSet
) {
2131 /* For practical purpose, 64Meg is a good
2132 * default chunk size for internal bitmaps.
2135 if (chunk
< 64*1024*1024)
2136 chunk
= 64*1024*1024;
2137 } else if (chunk
< min_chunk
)
2138 return 0; /* chunk size too small */
2139 if (chunk
== 0) /* rounding problem */
2143 /* start bitmap on a 4K boundary with enough space for
2146 bits
= (size
*512) / chunk
+ 1;
2147 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2148 room
*= 8; /* convert 4K blocks to sectors */
2149 offset
= -room
- bbl_size
;
2152 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2154 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2155 | MD_FEATURE_BITMAP_OFFSET
);
2156 memset(bms
, 0, sizeof(*bms
));
2157 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2158 bms
->version
= __cpu_to_le32(major
);
2159 uuid_from_super1(st
, uuid
);
2160 memcpy(bms
->uuid
, uuid
, 16);
2161 bms
->chunksize
= __cpu_to_le32(chunk
);
2162 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2163 bms
->sync_size
= __cpu_to_le64(size
);
2164 bms
->write_behind
= __cpu_to_le32(write_behind
);
2165 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2166 if (st
->cluster_name
)
2167 strncpy((char *)bms
->cluster_name
,
2168 st
->cluster_name
, strlen(st
->cluster_name
));
2174 static void locate_bitmap1(struct supertype
*st
, int fd
)
2176 unsigned long long offset
;
2177 struct mdp_superblock_1
*sb
;
2181 if (st
->ss
->load_super(st
, fd
, NULL
))
2182 return; /* no error I hope... */
2187 offset
= __le64_to_cpu(sb
->super_offset
);
2188 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
);
2191 lseek64(fd
, offset
<<9, 0);
2194 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2196 struct mdp_superblock_1
*sb
= st
->sb
;
2197 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2201 struct align_fd afd
;
2203 unsigned long long total_bm_space
, bm_space_per_node
;
2207 /* update cluster name */
2208 if (st
->cluster_name
) {
2209 memset((char *)bms
->cluster_name
, 0, sizeof(bms
->cluster_name
));
2210 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, 64);
2214 /* cluster md only supports superblock 1.2 now */
2215 if (st
->minor_version
!= 2) {
2216 pr_err("Warning: cluster md only works with superblock 1.2\n");
2220 /* Each node has an independent bitmap, it is necessary to calculate the
2221 * space is enough or not, first get how many bytes for the total bitmap */
2222 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2224 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2225 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2227 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2228 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2229 total_bm_space
/ bm_space_per_node
);
2233 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2242 locate_bitmap1(st
, fd
);
2244 if (posix_memalign(&buf
, 4096, 4096))
2248 /* Only the bitmap[0] should resync
2249 * whole device on initial assembly
2252 memset(buf
, 0x00, 4096);
2254 memset(buf
, 0xff, 4096);
2255 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2257 towrite
= calc_bitmap_size(bms
, 4096);
2258 while (towrite
> 0) {
2262 n
= awrite(&afd
, buf
, n
);
2268 memset(buf
, 0x00, 4096);
2270 memset(buf
, 0xff, 4096);
2277 } while (++i
< __le32_to_cpu(bms
->nodes
));
2283 static void free_super1(struct supertype
*st
)
2288 struct devinfo
*di
= st
->info
;
2289 st
->info
= di
->next
;
2298 static int validate_geometry1(struct supertype
*st
, int level
,
2299 int layout
, int raiddisks
,
2300 int *chunk
, unsigned long long size
,
2301 unsigned long long data_offset
,
2302 char *subdev
, unsigned long long *freesize
,
2305 unsigned long long ldsize
, devsize
;
2307 unsigned long long headroom
;
2310 if (level
== LEVEL_CONTAINER
) {
2312 pr_err("1.x metadata does not support containers\n");
2315 if (*chunk
== UnSet
)
2316 *chunk
= DEFAULT_CHUNK
;
2321 if (st
->minor_version
< 0)
2322 /* not specified, so time to set default */
2323 st
->minor_version
= 2;
2325 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2328 pr_err("super1.x cannot open %s: %s\n",
2329 subdev
, strerror(errno
));
2333 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2339 devsize
= ldsize
>> 9;
2345 /* creating: allow suitable space for bitmap */
2346 bmspace
= choose_bm_space(devsize
);
2348 if (data_offset
== INVALID_SECTORS
)
2349 data_offset
= st
->data_offset
;
2350 if (data_offset
== INVALID_SECTORS
)
2351 switch (st
->minor_version
) {
2357 /* Choose data offset appropriate for this device
2358 * and use as default for whole array.
2359 * The data_offset must allow for bitmap space
2360 * and base metadata, should allow for some headroom
2361 * for reshape, and should be rounded to multiple
2363 * Headroom is limited to 128M, but aim for about 0.1%
2365 headroom
= 128*1024*2;
2366 while ((headroom
<< 10) > devsize
&&
2368 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2370 data_offset
= 12*2 + bmspace
+ headroom
;
2371 #define ONE_MEG (2*1024)
2372 if (data_offset
> ONE_MEG
)
2373 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2376 if (st
->data_offset
== INVALID_SECTORS
)
2377 st
->data_offset
= data_offset
;
2378 switch(st
->minor_version
) {
2379 case 0: /* metadata at end. Round down and subtract space to reserve */
2380 devsize
= (devsize
& ~(4ULL*2-1));
2381 /* space for metadata, bblog, bitmap */
2382 devsize
-= 8*2 + 8 + bmspace
;
2386 devsize
-= data_offset
;
2389 *freesize
= devsize
;
2392 #endif /* MDASSEMBLE */
2394 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2396 /* Create a v1.0 superblock based on 'info'*/
2398 struct mdp_superblock_1
*sb
;
2401 unsigned long long offset
;
2403 if (posix_memalign(&ret
, 4096, 1024) != 0)
2406 memset(ret
, 0, 1024);
2407 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2408 sb
->major_version
= __cpu_to_le32(1);
2410 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2411 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2412 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2413 sb
->level
= __cpu_to_le32(info
->array
.level
);
2414 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2415 sb
->size
= __cpu_to_le64(info
->component_size
);
2416 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2417 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2418 if (info
->array
.level
> 0)
2419 sb
->data_size
= sb
->size
;
2421 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2422 sb
->resync_offset
= MaxSector
;
2423 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2424 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2425 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2427 offset
= st
->devsize
/512 - 8*2;
2429 sb
->super_offset
= __cpu_to_le64(offset
);
2430 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2432 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
2433 read(rfd
, sb
->device_uuid
, 16) != 16) {
2434 __u32 r
[4] = {random(), random(), random(), random()};
2435 memcpy(sb
->device_uuid
, r
, 16);
2440 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2441 int state
= sb0
->disks
[i
].state
;
2442 sb
->dev_roles
[i
] = 0xFFFF;
2443 if ((state
& (1<<MD_DISK_SYNC
)) &&
2444 !(state
& (1<<MD_DISK_FAULTY
)))
2445 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2447 sb
->sb_csum
= calc_sb_1_csum(sb
);
2451 struct superswitch super1
= {
2453 .examine_super
= examine_super1
,
2454 .brief_examine_super
= brief_examine_super1
,
2455 .export_examine_super
= export_examine_super1
,
2456 .detail_super
= detail_super1
,
2457 .brief_detail_super
= brief_detail_super1
,
2458 .export_detail_super
= export_detail_super1
,
2459 .write_init_super
= write_init_super1
,
2460 .validate_geometry
= validate_geometry1
,
2461 .add_to_super
= add_to_super1
,
2462 .examine_badblocks
= examine_badblocks_super1
,
2463 .copy_metadata
= copy_metadata1
,
2465 .match_home
= match_home1
,
2466 .uuid_from_super
= uuid_from_super1
,
2467 .getinfo_super
= getinfo_super1
,
2468 .container_content
= container_content1
,
2469 .update_super
= update_super1
,
2470 .init_super
= init_super1
,
2471 .store_super
= store_super1
,
2472 .compare_super
= compare_super1
,
2473 .load_super
= load_super1
,
2474 .match_metadata_desc
= match_metadata_desc1
,
2475 .avail_size
= avail_size1
,
2476 .add_internal_bitmap
= add_internal_bitmap1
,
2477 .locate_bitmap
= locate_bitmap1
,
2478 .write_bitmap
= write_bitmap1
,
2479 .free_super
= free_super1
,
2480 #if __BYTE_ORDER == BIG_ENDIAN