2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
28 * The version-1 superblock :
29 * All numeric fields are little-endian.
31 * total size: 256 bytes plus 2 per device.
32 * 1K allows 384 devices.
34 struct mdp_superblock_1
{
35 /* constant array information - 128 bytes */
36 __u32 magic
; /* MD_SB_MAGIC: 0xa92b4efc - little endian */
37 __u32 major_version
; /* 1 */
38 __u32 feature_map
; /* 0 for now */
39 __u32 pad0
; /* always set to 0 when writing */
41 __u8 set_uuid
[16]; /* user-space generated. */
42 char set_name
[32]; /* set and interpreted by user-space */
44 __u64 ctime
; /* lo 40 bits are seconds, top 24 are microseconds or 0*/
45 __u32 level
; /* -4 (multipath), -1 (linear), 0,1,4,5 */
46 __u32 layout
; /* only for raid5 currently */
47 __u64 size
; /* used size of component devices, in 512byte sectors */
49 __u32 chunksize
; /* in 512byte sectors */
51 __u32 bitmap_offset
; /* sectors after start of superblock that bitmap starts
52 * NOTE: signed, so bitmap can be before superblock
53 * only meaningful of feature_map[0] is set.
56 /* These are only valid with feature bit '4' */
57 __u32 new_level
; /* new level we are reshaping to */
58 __u64 reshape_position
; /* next address in array-space for reshape */
59 __u32 delta_disks
; /* change in number of raid_disks */
60 __u32 new_layout
; /* new layout */
61 __u32 new_chunk
; /* new chunk size (sectors) */
62 __u32 new_offset
; /* signed number to add to data_offset in new
63 * layout. 0 == no-change. This can be
64 * different on each device in the array.
67 /* constant this-device information - 64 bytes */
68 __u64 data_offset
; /* sector start of data, often 0 */
69 __u64 data_size
; /* sectors in this device that can be used for data */
70 __u64 super_offset
; /* sector start of this superblock */
72 __u64 recovery_offset
;/* sectors before this offset (from data_offset) have been recovered */
73 __u64 journal_tail
;/* journal tail of journal device (from data_offset) */
75 __u32 dev_number
; /* permanent identifier of this device - not role in raid */
76 __u32 cnt_corrected_read
; /* number of read errors that were corrected by re-writing */
77 __u8 device_uuid
[16]; /* user-space setable, ignored by kernel */
78 __u8 devflags
; /* per-device flags. Only one defined...*/
79 #define WriteMostly1 1 /* mask for writemostly flag in above */
80 /* bad block log. If there are any bad blocks the feature flag is set.
81 * if offset and size are non-zero, that space is reserved and available.
83 __u8 bblog_shift
; /* shift from sectors to block size for badblocklist */
84 __u16 bblog_size
; /* number of sectors reserved for badblocklist */
85 __u32 bblog_offset
; /* sector offset from superblock to bblog, signed */
87 /* array state information - 64 bytes */
88 __u64 utime
; /* 40 bits second, 24 btes microseconds */
89 __u64 events
; /* incremented when superblock updated */
90 __u64 resync_offset
; /* data before this offset (from data_offset) known to be in sync */
91 __u32 sb_csum
; /* checksum upto dev_roles[max_dev] */
92 __u32 max_dev
; /* size of dev_roles[] array to consider */
93 __u8 pad3
[64-32]; /* set to 0 when writing */
95 /* device state information. Indexed by dev_number.
97 * Note there are no per-device state flags. State information is rolled
98 * into the 'roles' value. If a device is spare or faulty, then it doesn't
99 * have a meaningful role.
101 __u16 dev_roles
[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
104 #define MAX_SB_SIZE 4096
105 /* bitmap super size is 256, but we round up to a sector for alignment */
106 #define BM_SUPER_SIZE 512
107 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
108 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
109 + sizeof(struct misc_dev_info))
111 struct misc_dev_info
{
115 /* feature_map bits */
116 #define MD_FEATURE_BITMAP_OFFSET 1
117 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
120 #define MD_FEATURE_RESHAPE_ACTIVE 4
121 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
122 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
123 * active device with same 'role'.
124 * 'recovery_offset' is also set.
126 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
127 * of devices, but is going
130 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
131 #define MD_FEATURE_BITMAP_VERSIONED 256 /* bitmap version number checked properly */
132 #define MD_FEATURE_JOURNAL 512 /* support write journal */
133 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
134 |MD_FEATURE_RECOVERY_OFFSET \
135 |MD_FEATURE_RESHAPE_ACTIVE \
136 |MD_FEATURE_BAD_BLOCKS \
137 |MD_FEATURE_REPLACEMENT \
138 |MD_FEATURE_RESHAPE_BACKWARDS \
139 |MD_FEATURE_NEW_OFFSET \
140 |MD_FEATURE_BITMAP_VERSIONED \
141 |MD_FEATURE_JOURNAL \
145 static int role_from_sb(struct mdp_superblock_1
*sb
)
150 d
= __le32_to_cpu(sb
->dev_number
);
151 if (d
< __le32_to_cpu(sb
->max_dev
))
152 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
154 role
= MD_DISK_ROLE_SPARE
;
159 /* return how many bytes are needed for bitmap, for cluster-md each node
160 * should have it's own bitmap */
161 static unsigned int calc_bitmap_size(bitmap_super_t
*bms
, unsigned int boundary
)
163 unsigned long long bits
, bytes
;
165 bits
= bitmap_bits(__le64_to_cpu(bms
->sync_size
),
166 __le32_to_cpu(bms
->chunksize
));
167 bytes
= (bits
+7) >> 3;
168 bytes
+= sizeof(bitmap_super_t
);
169 bytes
= ROUND_UP(bytes
, boundary
);
174 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
176 unsigned int disk_csum
, csum
;
177 unsigned long long newcsum
;
178 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
179 unsigned int *isuper
= (unsigned int*)sb
;
181 /* make sure I can count... */
182 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
183 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
184 sizeof(struct mdp_superblock_1
) != 256) {
185 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
188 disk_csum
= sb
->sb_csum
;
191 for (; size
>=4; size
-= 4 ) {
192 newcsum
+= __le32_to_cpu(*isuper
);
197 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
199 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
200 sb
->sb_csum
= disk_csum
;
201 return __cpu_to_le32(csum
);
205 * Information related to file descriptor used for aligned reads/writes.
206 * Cache the block size.
213 static void init_afd(struct align_fd
*afd
, int fd
)
217 if (ioctl(afd
->fd
, BLKSSZGET
, &afd
->blk_sz
) != 0)
221 static char abuf
[4096+4096];
222 static int aread(struct align_fd
*afd
, void *buf
, int len
)
225 * On devices with a 4K sector size, we need to read
226 * the full sector and copy relevant bits into
235 if (!bsize
|| bsize
> 4096 || len
> 4096) {
237 fprintf(stderr
, "WARNING - aread() called with invalid block size\n");
240 b
= ROUND_UP_PTR((char *)abuf
, 4096);
242 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
244 n
= read(afd
->fd
, b
, iosize
);
247 lseek(afd
->fd
, len
- n
, 1);
254 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
257 * On devices with a 4K sector size, we need to write
258 * the full sector. We pre-read if the sector is larger
260 * The address must be sector-aligned.
267 if (!bsize
|| bsize
> 4096 || len
> 4096) {
269 fprintf(stderr
, "WARNING - awrite() called with invalid block size\n");
272 b
= ROUND_UP_PTR((char *)abuf
, 4096);
274 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
278 n
= read(afd
->fd
, b
, iosize
);
281 lseek(afd
->fd
, -n
, 1);
285 n
= write(afd
->fd
, b
, iosize
);
288 lseek(afd
->fd
, len
- n
, 1);
293 static void examine_super1(struct supertype
*st
, char *homehost
)
295 struct mdp_superblock_1
*sb
= st
->sb
;
296 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
303 int l
= homehost
? strlen(homehost
) : 0;
305 unsigned long long sb_offset
;
308 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
309 printf(" Version : 1");
310 sb_offset
= __le64_to_cpu(sb
->super_offset
);
313 else if (sb_offset
<= 8)
317 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
318 printf(" Array UUID : ");
319 for (i
=0; i
<16; i
++) {
320 if ((i
&3)==0 && i
!= 0) printf(":");
321 printf("%02x", sb
->set_uuid
[i
]);
324 printf(" Name : %.32s", sb
->set_name
);
325 if (l
> 0 && l
< 32 &&
326 sb
->set_name
[l
] == ':' &&
327 strncmp(sb
->set_name
, homehost
, l
) == 0)
328 printf(" (local to host %s)", homehost
);
330 if (bms
->nodes
> 0 && (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
331 printf(" Cluster Name : %-64s\n", bms
->cluster_name
);
332 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
333 printf(" Creation Time : %.24s\n", ctime(&atime
));
334 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
335 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
336 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
338 printf(" Avail Dev Size : %llu%s\n",
339 (unsigned long long)__le64_to_cpu(sb
->data_size
),
340 human_size(__le64_to_cpu(sb
->data_size
)<<9));
341 if (__le32_to_cpu(sb
->level
) > 0) {
342 int ddsks
= 0, ddsks_denom
= 1;
343 switch(__le32_to_cpu(sb
->level
)) {
344 case 1: ddsks
=1;break;
346 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
347 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
349 layout
= __le32_to_cpu(sb
->layout
);
350 ddsks
= __le32_to_cpu(sb
->raid_disks
);
351 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
354 long long asize
= __le64_to_cpu(sb
->size
);
355 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
356 printf(" Array Size : %llu%s\n",
357 asize
>> 10, human_size(asize
));
359 if (sb
->size
!= sb
->data_size
)
360 printf(" Used Dev Size : %llu%s\n",
361 (unsigned long long)__le64_to_cpu(sb
->size
),
362 human_size(__le64_to_cpu(sb
->size
)<<9));
365 printf(" Data Offset : %llu sectors\n",
366 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
367 if (sb
->new_offset
&&
368 (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
)) {
369 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
370 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
371 printf(" New Offset : %llu sectors\n", offset
);
373 printf(" Super Offset : %llu sectors\n",
374 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
375 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
376 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
378 st
->ss
->getinfo_super(st
, &info
, NULL
);
379 if (info
.space_after
!= 1 &&
380 !(__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
381 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
382 info
.space_before
, info
.space_after
);
384 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
385 printf(" Device UUID : ");
386 for (i
=0; i
<16; i
++) {
387 if ((i
&3)==0 && i
!= 0) printf(":");
388 printf("%02x", sb
->device_uuid
[i
]);
392 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
393 printf("Internal Bitmap : %ld sectors from superblock\n",
394 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
396 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
397 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
398 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
399 if (__le32_to_cpu(sb
->delta_disks
)) {
400 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
401 printf(" (%d->%d)\n",
402 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
403 __le32_to_cpu(sb
->raid_disks
));
404 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
405 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
407 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
408 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
409 printf(" New Level : %s\n", c
?c
:"-unknown-");
411 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
412 if (__le32_to_cpu(sb
->level
) == 5) {
413 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
414 printf(" New Layout : %s\n", c
?c
:"-unknown-");
416 if (__le32_to_cpu(sb
->level
) == 6) {
417 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
418 printf(" New Layout : %s\n", c
?c
:"-unknown-");
420 if (__le32_to_cpu(sb
->level
) == 10) {
421 printf(" New Layout :");
422 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
426 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
427 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
432 if (sb
->devflags
& WriteMostly1
)
433 printf(" write-mostly");
437 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
438 printf(" Update Time : %.24s\n", ctime(&atime
));
440 if (sb
->bblog_size
&& sb
->bblog_offset
) {
441 printf(" Bad Block Log : %d entries available at offset %ld sectors",
442 __le16_to_cpu(sb
->bblog_size
)*512/8,
443 (long)(int32_t)__le32_to_cpu(sb
->bblog_offset
));
444 if (sb
->feature_map
&
445 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
446 printf(" - bad blocks present.");
450 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
451 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
453 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
454 __le32_to_cpu(calc_sb_1_csum(sb
)));
455 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
457 if (__le32_to_cpu(sb
->level
) == 5) {
458 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
459 printf(" Layout : %s\n", c
?c
:"-unknown-");
461 if (__le32_to_cpu(sb
->level
) == 6) {
462 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
463 printf(" Layout : %s\n", c
?c
:"-unknown-");
465 if (__le32_to_cpu(sb
->level
) == 10) {
466 int lo
= __le32_to_cpu(sb
->layout
);
468 print_r10_layout(lo
);
471 switch(__le32_to_cpu(sb
->level
)) {
477 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
480 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
486 /* This turns out to just be confusing */
487 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
488 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
489 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != MD_DISK_ROLE_SPARE
)
491 for (d
=0; d
< i
; d
++) {
492 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
494 if (role
== MD_DISK_ROLE_SPARE
) printf("empty");
495 else if(role
== MD_DISK_ROLE_FAULTY
) printf("failed");
496 else printf("%d", role
);
500 printf(" Device Role : ");
501 role
= role_from_sb(sb
);
502 if (role
>= MD_DISK_ROLE_FAULTY
)
504 else if (role
== MD_DISK_ROLE_JOURNAL
)
506 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
507 printf("Replacement device %d\n", role
);
509 printf("Active device %d\n", role
);
511 printf(" Array State : ");
512 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
515 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
516 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
530 /* This is confusing too */
532 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
533 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
534 if (role
== MD_DISK_ROLE_FAULTY
)
537 if (faulty
) printf(" %d failed", faulty
);
539 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
543 static void brief_examine_super1(struct supertype
*st
, int verbose
)
545 struct mdp_superblock_1
*sb
= st
->sb
;
547 unsigned long long sb_offset
;
549 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
551 nm
= strchr(sb
->set_name
, ':');
554 else if (sb
->set_name
[0])
566 printf(" level=%s", c
);
567 sb_offset
= __le64_to_cpu(sb
->super_offset
);
569 printf(" metadata=1.1 ");
570 else if (sb_offset
<= 8)
571 printf(" metadata=1.2 ");
573 printf(" metadata=1.0 ");
575 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
577 for (i
=0; i
<16; i
++) {
578 if ((i
&3)==0 && i
!= 0) printf(":");
579 printf("%02x", sb
->set_uuid
[i
]);
581 if (sb
->set_name
[0]) {
583 print_quoted(sb
->set_name
);
588 static void export_examine_super1(struct supertype
*st
)
590 struct mdp_superblock_1
*sb
= st
->sb
;
595 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
596 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
598 if (sb
->set_name
[i
] == '\n' ||
599 sb
->set_name
[i
] == '\0') {
604 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
605 if (__le32_to_cpu(sb
->level
) > 0) {
606 int ddsks
= 0, ddsks_denom
= 1;
607 switch(__le32_to_cpu(sb
->level
)) {
608 case 1: ddsks
=1;break;
610 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
611 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
613 layout
= __le32_to_cpu(sb
->layout
);
614 ddsks
= __le32_to_cpu(sb
->raid_disks
);
615 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
618 long long asize
= __le64_to_cpu(sb
->size
);
619 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
620 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
624 for (i
=0; i
<16; i
++) {
625 if ((i
&3)==0 && i
!= 0) printf(":");
626 printf("%02x", sb
->set_uuid
[i
]);
629 printf("MD_UPDATE_TIME=%llu\n",
630 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
631 printf("MD_DEV_UUID=");
632 for (i
=0; i
<16; i
++) {
633 if ((i
&3)==0 && i
!= 0) printf(":");
634 printf("%02x", sb
->device_uuid
[i
]);
637 printf("MD_EVENTS=%llu\n",
638 (unsigned long long)__le64_to_cpu(sb
->events
));
641 static int copy_metadata1(struct supertype
*st
, int from
, int to
)
643 /* Read superblock. If it looks good, write it out.
644 * Then if a bitmap is present, copy that.
645 * And if a bad-block-list is present, copy that too.
648 unsigned long long dsize
, sb_offset
;
649 const int bufsize
= 4*1024;
650 struct mdp_superblock_1 super
, *sb
;
652 if (posix_memalign(&buf
, 4096, bufsize
) != 0)
655 if (!get_dev_size(from
, NULL
, &dsize
))
661 switch(st
->minor_version
) {
665 sb_offset
&= ~(4*2-1);
677 if (lseek64(from
, sb_offset
<< 9, 0) < 0LL)
679 if (read(from
, buf
, bufsize
) != bufsize
)
683 super
= *sb
; // save most of sb for when we reuse buf
685 if (__le32_to_cpu(super
.magic
) != MD_SB_MAGIC
||
686 __le32_to_cpu(super
.major_version
) != 1 ||
687 __le64_to_cpu(super
.super_offset
) != sb_offset
||
688 calc_sb_1_csum(sb
) != super
.sb_csum
)
691 if (lseek64(to
, sb_offset
<< 9, 0) < 0LL)
693 if (write(to
, buf
, bufsize
) != bufsize
)
696 if (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
)) {
697 unsigned long long bitmap_offset
= sb_offset
;
698 int bytes
= 4096; // just an estimate.
700 struct align_fd afrom
, ato
;
702 init_afd(&afrom
, from
);
705 bitmap_offset
+= (int32_t)__le32_to_cpu(super
.bitmap_offset
);
707 if (lseek64(from
, bitmap_offset
<<9, 0) < 0)
709 if (lseek64(to
, bitmap_offset
<<9, 0) < 0)
712 for (written
= 0; written
< bytes
; ) {
713 int n
= bytes
- written
;
716 if (aread(&afrom
, buf
, n
) != n
)
719 /* have the header, can calculate
720 * correct bitmap bytes */
723 bytes
= calc_bitmap_size(bms
, 512);
727 if (awrite(&ato
, buf
, n
) != n
)
733 if (super
.bblog_size
!= 0 &&
734 __le16_to_cpu(super
.bblog_size
) <= 100 &&
735 super
.bblog_offset
!= 0 &&
736 (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BAD_BLOCKS
))) {
737 /* There is a bad block log */
738 unsigned long long bb_offset
= sb_offset
;
739 int bytes
= __le16_to_cpu(super
.bblog_size
) * 512;
741 struct align_fd afrom
, ato
;
743 init_afd(&afrom
, from
);
746 bb_offset
+= (int32_t)__le32_to_cpu(super
.bblog_offset
);
748 if (lseek64(from
, bb_offset
<<9, 0) < 0)
750 if (lseek64(to
, bb_offset
<<9, 0) < 0)
753 for (written
= 0; written
< bytes
; ) {
754 int n
= bytes
- written
;
757 if (aread(&afrom
, buf
, n
) != n
)
760 if (awrite(&ato
, buf
, n
) != n
)
774 static void detail_super1(struct supertype
*st
, char *homehost
)
776 struct mdp_superblock_1
*sb
= st
->sb
;
777 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
779 int l
= homehost
? strlen(homehost
) : 0;
781 printf(" Name : %.32s", sb
->set_name
);
782 if (l
> 0 && l
< 32 &&
783 sb
->set_name
[l
] == ':' &&
784 strncmp(sb
->set_name
, homehost
, l
) == 0)
785 printf(" (local to host %s)", homehost
);
786 if (bms
->nodes
> 0 && (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
787 printf("\n Cluster Name : %-64s", bms
->cluster_name
);
788 printf("\n UUID : ");
789 for (i
=0; i
<16; i
++) {
790 if ((i
&3)==0 && i
!= 0) printf(":");
791 printf("%02x", sb
->set_uuid
[i
]);
793 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb
->events
));
796 static void brief_detail_super1(struct supertype
*st
)
798 struct mdp_superblock_1
*sb
= st
->sb
;
801 if (sb
->set_name
[0]) {
803 print_quoted(sb
->set_name
);
806 for (i
=0; i
<16; i
++) {
807 if ((i
&3)==0 && i
!= 0) printf(":");
808 printf("%02x", sb
->set_uuid
[i
]);
812 static void export_detail_super1(struct supertype
*st
)
814 struct mdp_superblock_1
*sb
= st
->sb
;
819 if (sb
->set_name
[i
] == '\n' ||
820 sb
->set_name
[i
] == '\0') {
825 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
828 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
830 struct mdp_superblock_1
*sb
= st
->sb
;
831 unsigned long long offset
;
836 if (!sb
->bblog_size
|| __le16_to_cpu(sb
->bblog_size
) > 100
837 || !sb
->bblog_offset
){
838 printf("No bad-blocks list configured on %s\n", devname
);
841 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
843 printf("Bad-blocks list is empty in %s\n", devname
);
847 size
= __le16_to_cpu(sb
->bblog_size
)* 512;
848 if (posix_memalign((void**)&bbl
, 4096, size
) != 0) {
849 pr_err("could not allocate badblocks list\n");
852 offset
= __le64_to_cpu(sb
->super_offset
) +
853 (int)__le32_to_cpu(sb
->bblog_offset
);
855 if (lseek64(fd
, offset
, 0) < 0) {
856 pr_err("Cannot seek to bad-blocks list\n");
859 if (read(fd
, bbl
, size
) != size
) {
860 pr_err("Cannot read bad-blocks list\n");
863 /* 64bits per entry. 10 bits is block-count, 54 bits is block
864 * offset. Blocks are sectors unless bblog->shift makes them bigger
867 printf("Bad-blocks on %s:\n", devname
);
868 for (i
= 0; i
< size
/8; i
++, bbp
++) {
869 __u64 bb
= __le64_to_cpu(*bbp
);
870 int count
= bb
& 0x3ff;
871 unsigned long long sector
= bb
>> 10;
876 sector
<<= sb
->bblog_shift
;
877 count
<<= sb
->bblog_shift
;
879 printf("%20llu for %d sectors\n", sector
, count
);
886 static int match_home1(struct supertype
*st
, char *homehost
)
888 struct mdp_superblock_1
*sb
= st
->sb
;
889 int l
= homehost
? strlen(homehost
) : 0;
891 return (l
> 0 && l
< 32 &&
892 sb
->set_name
[l
] == ':' &&
893 strncmp(sb
->set_name
, homehost
, l
) == 0);
896 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
898 struct mdp_superblock_1
*super
= st
->sb
;
899 char *cuuid
= (char*)uuid
;
902 cuuid
[i
] = super
->set_uuid
[i
];
905 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
907 struct mdp_superblock_1
*sb
= st
->sb
;
908 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
909 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
913 unsigned int map_disks
= info
->array
.raid_disks
;
914 unsigned long long super_offset
;
915 unsigned long long data_size
;
917 memset(info
, 0, sizeof(*info
));
918 info
->array
.major_version
= 1;
919 info
->array
.minor_version
= st
->minor_version
;
920 info
->array
.patch_version
= 0;
921 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
922 info
->array
.level
= __le32_to_cpu(sb
->level
);
923 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
924 info
->array
.md_minor
= -1;
925 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
926 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
927 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
929 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
931 if (__le32_to_cpu(bsb
->nodes
) > 1)
932 info
->array
.state
|= (1 << MD_SB_CLUSTERED
);
934 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
935 info
->component_size
= __le64_to_cpu(sb
->size
);
936 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
937 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
939 info
->disk
.major
= 0;
940 info
->disk
.minor
= 0;
941 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
942 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
943 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
944 role
= MD_DISK_ROLE_FAULTY
;
946 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
948 super_offset
= __le64_to_cpu(sb
->super_offset
);
949 if (info
->array
.level
<= 0)
950 data_size
= __le64_to_cpu(sb
->data_size
);
952 data_size
= __le64_to_cpu(sb
->size
);
953 if (info
->data_offset
< super_offset
) {
954 unsigned long long end
;
955 info
->space_before
= info
->data_offset
;
958 if (sb
->bblog_offset
&& sb
->bblog_size
) {
959 unsigned long long bboffset
= super_offset
;
960 bboffset
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
965 if (super_offset
+ info
->bitmap_offset
< end
)
966 end
= super_offset
+ info
->bitmap_offset
;
968 if (info
->data_offset
+ data_size
< end
)
969 info
->space_after
= end
- data_size
- info
->data_offset
;
971 info
->space_after
= 0;
973 unsigned long long earliest
;
974 earliest
= super_offset
+ (32+4)*2; /* match kernel */
975 if (info
->bitmap_offset
> 0) {
976 unsigned long long bmend
= info
->bitmap_offset
;
977 unsigned long long size
= calc_bitmap_size(bsb
, 4096);
980 if (bmend
> earliest
)
983 if (sb
->bblog_offset
&& sb
->bblog_size
) {
984 unsigned long long bbend
= super_offset
;
985 bbend
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
986 bbend
+= __le16_to_cpu(sb
->bblog_size
);
987 if (bbend
> earliest
)
990 if (earliest
< info
->data_offset
)
991 info
->space_before
= info
->data_offset
- earliest
;
993 info
->space_before
= 0;
994 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
996 if (info
->space_before
== 0 && info
->space_after
== 0) {
997 /* It will look like we don't support data_offset changes,
998 * be we do - it's just that there is no room.
999 * A change that reduced the number of devices should
1000 * still be allowed, so set the otherwise useless value of '1'
1002 info
->space_after
= 1;
1005 info
->disk
.raid_disk
= -1;
1007 case MD_DISK_ROLE_SPARE
:
1008 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
1010 case MD_DISK_ROLE_FAULTY
:
1011 info
->disk
.state
= 1; /* faulty */
1013 case MD_DISK_ROLE_JOURNAL
:
1014 info
->disk
.state
= (1 << MD_DISK_JOURNAL
);
1015 info
->disk
.raid_disk
= role
;
1016 info
->space_after
= (misc
->device_size
- info
->data_offset
) % 8; /* journal uses all 4kB blocks*/
1019 info
->disk
.state
= 6; /* active and in sync */
1020 info
->disk
.raid_disk
= role
;
1022 if (sb
->devflags
& WriteMostly1
)
1023 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1024 info
->events
= __le64_to_cpu(sb
->events
);
1025 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
1026 info
->safe_mode_delay
= 200;
1028 memcpy(info
->uuid
, sb
->set_uuid
, 16);
1030 strncpy(info
->name
, sb
->set_name
, 32);
1033 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
1034 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
1035 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
1038 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
1039 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
1041 info
->recovery_start
= MaxSector
;
1043 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
1044 info
->reshape_active
= 1;
1045 if ((sb
->feature_map
& __le32_to_cpu(MD_FEATURE_NEW_OFFSET
)) &&
1046 sb
->new_offset
!= 0)
1047 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
1048 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
1049 info
->new_level
= __le32_to_cpu(sb
->new_level
);
1050 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
1051 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
1052 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
1053 if (info
->delta_disks
< 0)
1054 info
->array
.raid_disks
-= info
->delta_disks
;
1056 info
->reshape_active
= 0;
1058 info
->recovery_blocked
= info
->reshape_active
;
1061 for (i
=0; i
<map_disks
; i
++)
1063 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
1064 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
1065 if (/*role == MD_DISK_ROLE_SPARE || */role
< (unsigned) info
->array
.raid_disks
) {
1067 if (map
&& role
< map_disks
)
1072 info
->array
.working_disks
= working
;
1073 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_JOURNAL
))
1074 info
->journal_device_required
= 1;
1075 info
->journal_clean
= 0;
1078 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
1080 struct mdinfo
*info
;
1085 info
= xmalloc(sizeof(*info
));
1086 getinfo_super1(st
, info
, NULL
);
1090 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
1092 char *devname
, int verbose
,
1093 int uuid_set
, char *homehost
)
1095 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1096 * if any change was made. For others, the return value is
1101 struct mdp_superblock_1
*sb
= st
->sb
;
1102 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1104 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1105 rv
= cluster_get_dlmlock(&lockid
);
1107 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1108 cluster_release_dlmlock(lockid
);
1113 if (strcmp(update
, "homehost") == 0 &&
1115 /* Note that 'homehost' is special as it is really
1120 c
= strchr(sb
->set_name
, ':');
1122 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1124 strncpy(info
->name
, sb
->set_name
, 32);
1128 if (strcmp(update
, "force-one")==0) {
1129 /* Not enough devices for a working array,
1130 * so bring this one up-to-date
1132 if (sb
->events
!= __cpu_to_le64(info
->events
))
1134 sb
->events
= __cpu_to_le64(info
->events
);
1135 } else if (strcmp(update
, "force-array")==0) {
1136 /* Degraded array and 'force' requests to
1137 * maybe need to mark it 'clean'.
1139 switch(__le32_to_cpu(sb
->level
)) {
1140 case 5: case 4: case 6:
1141 /* need to force clean */
1142 if (sb
->resync_offset
!= MaxSector
)
1144 sb
->resync_offset
= MaxSector
;
1146 } else if (strcmp(update
, "assemble")==0) {
1147 int d
= info
->disk
.number
;
1149 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
1150 want
= info
->disk
.raid_disk
;
1151 else if (info
->disk
.state
& (1<<MD_DISK_JOURNAL
))
1152 want
= MD_DISK_ROLE_JOURNAL
;
1154 want
= MD_DISK_ROLE_SPARE
;
1155 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
1156 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
1159 if (info
->reshape_active
&&
1160 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1161 info
->delta_disks
>= 0 &&
1162 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
1163 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1166 if (info
->reshape_active
&&
1167 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1168 info
->delta_disks
< 0 &&
1169 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
1170 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1173 } else if (strcmp(update
, "linear-grow-new") == 0) {
1176 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
1178 for (i
=0 ; i
< max
; i
++)
1179 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= MD_DISK_ROLE_FAULTY
)
1181 sb
->dev_number
= __cpu_to_le32(i
);
1182 info
->disk
.number
= i
;
1183 if (max
>= __le32_to_cpu(sb
->max_dev
))
1184 sb
->max_dev
= __cpu_to_le32(max
+1);
1186 random_uuid(sb
->device_uuid
);
1189 __cpu_to_le16(info
->disk
.raid_disk
);
1191 fd
= open(devname
, O_RDONLY
);
1193 unsigned long long ds
;
1194 get_dev_size(fd
, devname
, &ds
);
1197 if (__le64_to_cpu(sb
->super_offset
) <
1198 __le64_to_cpu(sb
->data_offset
)) {
1199 sb
->data_size
= __cpu_to_le64(
1200 ds
- __le64_to_cpu(sb
->data_offset
));
1203 ds
&= ~(unsigned long long)(4*2-1);
1204 sb
->super_offset
= __cpu_to_le64(ds
);
1205 sb
->data_size
= __cpu_to_le64(
1206 ds
- __le64_to_cpu(sb
->data_offset
));
1209 } else if (strcmp(update
, "linear-grow-update") == 0) {
1210 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1211 sb
->dev_roles
[info
->disk
.number
] =
1212 __cpu_to_le16(info
->disk
.raid_disk
);
1213 } else if (strcmp(update
, "resync") == 0) {
1214 /* make sure resync happens */
1215 sb
->resync_offset
= 0ULL;
1216 } else if (strcmp(update
, "uuid") == 0) {
1217 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1219 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
)
1220 memcpy(bms
->uuid
, sb
->set_uuid
, 16);
1221 } else if (strcmp(update
, "no-bitmap") == 0) {
1222 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1223 } else if (strcmp(update
, "bbl") == 0) {
1224 /* only possible if there is room after the bitmap, or if
1225 * there is no bitmap
1227 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1228 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1229 long bitmap_offset
= 0;
1230 long bm_sectors
= 0;
1234 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1235 bitmap_offset
= (long)__le32_to_cpu(sb
->bitmap_offset
);
1236 bm_sectors
= calc_bitmap_size(bms
, 4096) >> 9;
1239 if (sb_offset
< data_offset
) {
1240 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1243 bb_offset
= sb_offset
+ 8;
1244 if (bm_sectors
&& bitmap_offset
> 0)
1245 bb_offset
= bitmap_offset
+ bm_sectors
;
1246 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1247 && bb_offset
+ 8+8 <= (long)data_offset
)
1248 /* too close to bitmap, and room to grow */
1250 if (bb_offset
+ 8 <= (long)data_offset
) {
1251 sb
->bblog_size
= __cpu_to_le16(8);
1252 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1255 /* 1.0 - Put bbl just before super block */
1256 if (bm_sectors
&& bitmap_offset
< 0)
1257 space
= -bitmap_offset
- bm_sectors
;
1259 space
= sb_offset
- data_offset
-
1260 __le64_to_cpu(sb
->data_size
);
1262 sb
->bblog_size
= __cpu_to_le16(8);
1263 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1266 } else if (strcmp(update
, "no-bbl") == 0) {
1267 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1268 pr_err("Cannot remove active bbl from %s\n",devname
);
1271 sb
->bblog_shift
= 0;
1272 sb
->bblog_offset
= 0;
1274 } else if (strcmp(update
, "force-no-bbl") == 0) {
1275 sb
->feature_map
&= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1277 sb
->bblog_shift
= 0;
1278 sb
->bblog_offset
= 0;
1279 } else if (strcmp(update
, "name") == 0) {
1280 if (info
->name
[0] == 0)
1281 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1282 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1284 strchr(info
->name
, ':') == NULL
&&
1285 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1286 strcpy(sb
->set_name
, homehost
);
1287 strcat(sb
->set_name
, ":");
1288 strcat(sb
->set_name
, info
->name
);
1290 strncpy(sb
->set_name
, info
->name
, sizeof(sb
->set_name
));
1291 } else if (strcmp(update
, "devicesize") == 0 &&
1292 __le64_to_cpu(sb
->super_offset
) <
1293 __le64_to_cpu(sb
->data_offset
)) {
1294 /* set data_size to device size less data_offset */
1295 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1296 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1297 sb
->data_size
= __cpu_to_le64(
1298 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1299 } else if (strncmp(update
, "revert-reshape", 14) == 0) {
1301 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1302 pr_err("No active reshape to revert on %s\n",
1306 unsigned long long reshape_sectors
;
1309 /* If the reshape hasn't started, just stop it.
1310 * It is conceivable that a stripe was modified but
1311 * the metadata not updated. In that case the backup
1312 * should have been used to get passed the critical stage.
1313 * If that couldn't happen, the "-nobackup" version
1316 if (strcmp(update
, "revert-reshape-nobackup") == 0 &&
1317 sb
->reshape_position
== 0 &&
1318 (__le32_to_cpu(sb
->delta_disks
) > 0 ||
1319 (__le32_to_cpu(sb
->delta_disks
) == 0 &&
1320 !(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
))))) {
1321 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1322 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1323 __le32_to_cpu(sb
->delta_disks
));
1324 sb
->delta_disks
= 0;
1327 /* reshape_position is a little messy.
1328 * Its value must be a multiple of the larger
1329 * chunk size, and of the "after" data disks.
1330 * So when reverting we need to change it to
1331 * be a multiple of the new "after" data disks,
1332 * which is the old "before".
1333 * If it isn't already a multiple of 'before',
1334 * the only thing we could do would be
1335 * copy some block around on the disks, which
1336 * is easy to get wrong.
1337 * So we reject a revert-reshape unless the
1338 * alignment is good.
1340 if (__le32_to_cpu(sb
->level
) >= 4 &&
1341 __le32_to_cpu(sb
->level
) <= 6) {
1342 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1343 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1344 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1345 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1346 if (reshape_sectors
% reshape_chunk
) {
1347 pr_err("Reshape position is not suitably aligned.\n");
1348 pr_err("Try normal assembly and stop again\n");
1352 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1353 __le32_to_cpu(sb
->delta_disks
));
1354 if (sb
->delta_disks
== 0)
1355 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1357 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1359 temp
= sb
->new_layout
;
1360 sb
->new_layout
= sb
->layout
;
1363 temp
= sb
->new_chunk
;
1364 sb
->new_chunk
= sb
->chunksize
;
1365 sb
->chunksize
= temp
;
1367 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1368 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1369 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1370 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1371 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1375 } else if (strcmp(update
, "_reshape_progress")==0)
1376 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1377 else if (strcmp(update
, "writemostly")==0)
1378 sb
->devflags
|= WriteMostly1
;
1379 else if (strcmp(update
, "readwrite")==0)
1380 sb
->devflags
&= ~WriteMostly1
;
1384 sb
->sb_csum
= calc_sb_1_csum(sb
);
1385 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1386 cluster_release_dlmlock(lockid
);
1391 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1392 unsigned long long size
, char *name
, char *homehost
,
1393 int *uuid
, unsigned long long data_offset
)
1395 struct mdp_superblock_1
*sb
;
1400 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1401 pr_err("could not allocate superblock\n");
1404 memset(sb
, 0, SUPER1_SIZE
);
1408 /* zeroing superblock */
1412 spares
= info
->working_disks
- info
->active_disks
;
1413 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1414 pr_err("too many devices requested: %d+%d > %d\n",
1415 info
->raid_disks
, spares
, MAX_DEVS
);
1419 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1420 sb
->major_version
= __cpu_to_le32(1);
1421 sb
->feature_map
= 0;
1425 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1427 random_uuid(sb
->set_uuid
);;
1429 if (name
== NULL
|| *name
== 0) {
1430 sprintf(defname
, "%d", info
->md_minor
);
1434 strchr(name
, ':')== NULL
&&
1435 strlen(homehost
)+1+strlen(name
) < 32) {
1436 strcpy(sb
->set_name
, homehost
);
1437 strcat(sb
->set_name
, ":");
1438 strcat(sb
->set_name
, name
);
1440 strncpy(sb
->set_name
, name
, sizeof(sb
->set_name
));
1442 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1443 sb
->level
= __cpu_to_le32(info
->level
);
1444 sb
->layout
= __cpu_to_le32(info
->layout
);
1445 sb
->size
= __cpu_to_le64(size
*2ULL);
1446 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1447 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1449 sb
->data_offset
= __cpu_to_le64(data_offset
);
1450 sb
->data_size
= __cpu_to_le64(0);
1451 sb
->super_offset
= __cpu_to_le64(0);
1452 sb
->recovery_offset
= __cpu_to_le64(0);
1454 sb
->utime
= sb
->ctime
;
1455 sb
->events
= __cpu_to_le64(1);
1456 if (info
->state
& (1<<MD_SB_CLEAN
))
1457 sb
->resync_offset
= MaxSector
;
1459 sb
->resync_offset
= 0;
1460 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1461 sbsize
= ROUND_UP(sbsize
, 512);
1462 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1464 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1472 long long data_offset
;
1473 mdu_disk_info_t disk
;
1474 struct devinfo
*next
;
1477 /* Add a device to the superblock being created */
1478 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1479 int fd
, char *devname
, unsigned long long data_offset
)
1481 struct mdp_superblock_1
*sb
= st
->sb
;
1482 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1483 struct devinfo
*di
, **dip
;
1484 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1487 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1488 rv
= cluster_get_dlmlock(&lockid
);
1490 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1491 cluster_release_dlmlock(lockid
);
1496 if ((dk
->state
& 6) == 6) /* active, sync */
1497 *rp
= __cpu_to_le16(dk
->raid_disk
);
1498 else if (dk
->state
& (1<<MD_DISK_JOURNAL
))
1499 *rp
= MD_DISK_ROLE_JOURNAL
;
1500 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1501 *rp
= MD_DISK_ROLE_SPARE
;
1503 *rp
= MD_DISK_ROLE_FAULTY
;
1505 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1506 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1507 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1509 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1510 sb
->devflags
= 0; /* don't copy another disks flags */
1511 sb
->sb_csum
= calc_sb_1_csum(sb
);
1513 dip
= (struct devinfo
**)&st
->info
;
1515 dip
= &(*dip
)->next
;
1516 di
= xmalloc(sizeof(struct devinfo
));
1518 di
->devname
= devname
;
1520 di
->data_offset
= data_offset
;
1524 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1525 cluster_release_dlmlock(lockid
);
1531 static int locate_bitmap1(struct supertype
*st
, int fd
, int node_num
);
1533 static int store_super1(struct supertype
*st
, int fd
)
1535 struct mdp_superblock_1
*sb
= st
->sb
;
1536 unsigned long long sb_offset
;
1537 struct align_fd afd
;
1539 unsigned long long dsize
;
1540 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1543 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1544 rv
= cluster_get_dlmlock(&lockid
);
1546 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1547 cluster_release_dlmlock(lockid
);
1552 if (!get_dev_size(fd
, NULL
, &dsize
))
1563 * Calculate the position of the superblock.
1564 * It is always aligned to a 4K boundary and
1565 * depending on minor_version, it can be:
1566 * 0: At least 8K, but less than 12K, from end of device
1567 * 1: At start of device
1568 * 2: 4K from start of device.
1570 switch(st
->minor_version
) {
1574 sb_offset
&= ~(4*2-1);
1586 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1587 0 != __le64_to_cpu(sb
->super_offset
)
1589 pr_err("internal error - sb_offset is wrong\n");
1593 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1596 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1598 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1601 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1602 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1603 (((char*)sb
)+MAX_SB_SIZE
);
1604 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1605 locate_bitmap1(st
, fd
, 0);
1606 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1611 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1612 cluster_release_dlmlock(lockid
);
1617 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1619 static unsigned long choose_bm_space(unsigned long devsize
)
1621 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1622 * if bigger than 200Gig, save 128k
1623 * NOTE: result must be multiple of 4K else bad things happen
1624 * on 4K-sector devices.
1628 if (devsize
- 64*2 >= 200*1024*1024*2)
1630 if (devsize
- 4*2 > 8*1024*1024*2)
1635 static void free_super1(struct supertype
*st
);
1637 #define META_BLOCK_SIZE 4096
1638 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
1641 static int write_empty_r5l_meta_block(struct supertype
*st
, int fd
)
1643 struct r5l_meta_block
*mb
;
1644 struct mdp_superblock_1
*sb
= st
->sb
;
1645 struct align_fd afd
;
1650 if (posix_memalign((void**)&mb
, 4096, META_BLOCK_SIZE
) != 0) {
1651 pr_err("Could not allocate memory for the meta block.\n");
1655 memset(mb
, 0, META_BLOCK_SIZE
);
1657 mb
->magic
= __cpu_to_le32(R5LOG_MAGIC
);
1658 mb
->version
= R5LOG_VERSION
;
1659 mb
->meta_size
= __cpu_to_le32(sizeof(struct r5l_meta_block
));
1660 mb
->seq
= __cpu_to_le64(random32());
1661 mb
->position
= __cpu_to_le64(0);
1663 crc
= crc32c_le(0xffffffff, sb
->set_uuid
, sizeof(sb
->set_uuid
));
1664 crc
= crc32c_le(crc
, (void *)mb
, META_BLOCK_SIZE
);
1667 if (lseek64(fd
, (sb
->data_offset
) * 512, 0) < 0LL) {
1668 pr_err("cannot seek to offset of the meta block\n");
1672 if (awrite(&afd
, mb
, META_BLOCK_SIZE
) != META_BLOCK_SIZE
) {
1673 pr_err("failed to store write the meta block \n");
1686 static int write_init_super1(struct supertype
*st
)
1688 struct mdp_superblock_1
*sb
= st
->sb
;
1689 struct supertype
*refst
;
1691 unsigned long long bm_space
;
1693 unsigned long long dsize
, array_size
;
1694 unsigned long long sb_offset
;
1695 unsigned long long data_offset
;
1697 for (di
= st
->info
; di
; di
= di
->next
) {
1698 if (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))
1699 sb
->feature_map
|= MD_FEATURE_JOURNAL
;
1702 for (di
= st
->info
; di
; di
= di
->next
) {
1703 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1708 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1711 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1712 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1713 sb
->devflags
|= WriteMostly1
;
1715 sb
->devflags
&= ~WriteMostly1
;
1717 random_uuid(sb
->device_uuid
);
1719 if (!(di
->disk
.state
& (1<<MD_DISK_JOURNAL
)))
1722 refst
= dup_super(st
);
1723 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1724 struct mdp_superblock_1
*refsb
= refst
->sb
;
1726 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1727 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1728 /* same array, so preserve events and
1730 sb
->events
= refsb
->events
;
1731 /* bugs in 2.6.17 and earlier mean the
1732 * dev_number chosen in Manage must be preserved
1734 if (get_linux_version() >= 2006018)
1735 sb
->dev_number
= refsb
->dev_number
;
1741 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1754 * Calculate the position of the superblock.
1755 * It is always aligned to a 4K boundary and
1756 * depending on minor_version, it can be:
1757 * 0: At least 8K, but less than 12K, from end of device
1758 * 1: At start of device
1759 * 2: 4K from start of device.
1760 * data_offset has already been set.
1762 array_size
= __le64_to_cpu(sb
->size
);
1763 /* work out how much space we left for a bitmap,
1764 * Add 8 sectors for bad block log */
1765 bm_space
= choose_bm_space(array_size
) + 8;
1767 data_offset
= di
->data_offset
;
1768 if (data_offset
== INVALID_SECTORS
)
1769 data_offset
= st
->data_offset
;
1770 switch(st
->minor_version
) {
1772 if (data_offset
== INVALID_SECTORS
)
1776 sb_offset
&= ~(4*2-1);
1777 sb
->data_offset
= __cpu_to_le64(data_offset
);
1778 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1779 if (sb_offset
< array_size
+ bm_space
)
1780 bm_space
= sb_offset
- array_size
;
1781 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1782 if (bm_space
>= 8) {
1783 sb
->bblog_size
= __cpu_to_le16(8);
1784 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1788 sb
->super_offset
= __cpu_to_le64(0);
1789 if (data_offset
== INVALID_SECTORS
)
1792 sb
->data_offset
= __cpu_to_le64(data_offset
);
1793 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1794 if (data_offset
>= 8 + 32*2 + 8) {
1795 sb
->bblog_size
= __cpu_to_le16(8);
1796 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1797 } else if (data_offset
>= 16) {
1798 sb
->bblog_size
= __cpu_to_le16(8);
1799 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8);
1804 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1805 if (data_offset
== INVALID_SECTORS
)
1808 sb
->data_offset
= __cpu_to_le64(data_offset
);
1809 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1810 if (data_offset
>= 16 + 32*2 + 8) {
1811 sb
->bblog_size
= __cpu_to_le16(8);
1812 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1813 } else if (data_offset
>= 16+16) {
1814 sb
->bblog_size
= __cpu_to_le16(8);
1815 /* '8' sectors for the bblog, and another '8'
1816 * because we want offset from superblock, not
1819 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8-8);
1823 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1824 st
->minor_version
, di
->devname
);
1828 /* Disable badblock log on clusters, or when explicitly requested */
1829 if (st
->nodes
> 0 || conf_get_create_info()->bblist
== 0) {
1831 sb
->bblog_offset
= 0;
1834 sb
->sb_csum
= calc_sb_1_csum(sb
);
1835 rv
= store_super1(st
, di
->fd
);
1837 if (rv
== 0 && (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))) {
1838 rv
= write_empty_r5l_meta_block(st
, di
->fd
);
1843 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1844 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NodeNumUpdate
);
1852 pr_err("Failed to write metadata to %s\n",
1859 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1863 * 0 same, or first was empty, and second was copied
1864 * 1 second had wrong number
1866 * 3 wrong other info
1868 struct mdp_superblock_1
*first
= st
->sb
;
1869 struct mdp_superblock_1
*second
= tst
->sb
;
1871 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1873 if (second
->major_version
!= __cpu_to_le32(1))
1877 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1878 pr_err("could not allocate superblock\n");
1881 memcpy(first
, second
, SUPER1_SIZE
);
1885 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1888 if (first
->ctime
!= second
->ctime
||
1889 first
->level
!= second
->level
||
1890 first
->layout
!= second
->layout
||
1891 first
->size
!= second
->size
||
1892 first
->chunksize
!= second
->chunksize
||
1893 first
->raid_disks
!= second
->raid_disks
)
1898 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1900 unsigned long long dsize
;
1901 unsigned long long sb_offset
;
1902 struct mdp_superblock_1
*super
;
1904 struct bitmap_super_s
*bsb
;
1905 struct misc_dev_info
*misc
;
1906 struct align_fd afd
;
1912 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1914 struct supertype tst
;
1915 __u64 bestctime
= 0;
1916 /* guess... choose latest ctime */
1917 memset(&tst
, 0, sizeof(tst
));
1919 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1920 switch(load_super1(&tst
, fd
, devname
)) {
1921 case 0: super
= tst
.sb
;
1922 if (bestvers
== -1 ||
1923 bestctime
< __le64_to_cpu(super
->ctime
)) {
1924 bestvers
= tst
.minor_version
;
1925 bestctime
= __le64_to_cpu(super
->ctime
);
1930 case 1: return 1; /*bad device */
1931 case 2: break; /* bad, try next */
1934 if (bestvers
!= -1) {
1936 tst
.minor_version
= bestvers
;
1938 tst
.max_devs
= MAX_DEVS
;
1939 rv
= load_super1(&tst
, fd
, devname
);
1946 if (!get_dev_size(fd
, devname
, &dsize
))
1952 pr_err("%s is too small for md: size is %llu sectors.\n",
1958 * Calculate the position of the superblock.
1959 * It is always aligned to a 4K boundary and
1960 * depending on minor_version, it can be:
1961 * 0: At least 8K, but less than 12K, from end of device
1962 * 1: At start of device
1963 * 2: 4K from start of device.
1965 switch(st
->minor_version
) {
1969 sb_offset
&= ~(4*2-1);
1981 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1983 pr_err("Cannot seek to superblock on %s: %s\n",
1984 devname
, strerror(errno
));
1988 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
1989 pr_err("could not allocate superblock\n");
1993 memset(super
, 0, SUPER1_SIZE
);
1995 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
1997 pr_err("Cannot read superblock on %s\n",
2003 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
2005 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2006 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
2011 if (__le32_to_cpu(super
->major_version
) != 1) {
2013 pr_err("Cannot interpret superblock on %s - version is %d\n",
2014 devname
, __le32_to_cpu(super
->major_version
));
2018 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
2020 pr_err("No superblock found on %s (super_offset is wrong)\n",
2027 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2029 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
2030 misc
->device_size
= dsize
;
2031 if (st
->data_offset
== INVALID_SECTORS
)
2032 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
2034 /* Now check on the bitmap superblock */
2035 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
2037 /* Read the bitmap superblock and make sure it looks
2038 * valid. If it doesn't clear the bit. An --assemble --force
2039 * should get that written out.
2041 locate_bitmap1(st
, fd
, 0);
2042 if (aread(&afd
, bsb
, 512) != 512)
2045 uuid_from_super1(st
, uuid
);
2046 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
2047 memcmp(bsb
->uuid
, uuid
, 16) != 0)
2052 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
2053 & ~MD_FEATURE_BITMAP_OFFSET
);
2057 static struct supertype
*match_metadata_desc1(char *arg
)
2059 struct supertype
*st
= xcalloc(1, sizeof(*st
));
2061 st
->container_devnm
[0] = 0;
2063 st
->max_devs
= MAX_DEVS
;
2065 st
->data_offset
= INVALID_SECTORS
;
2066 /* leading zeros can be safely ignored. --detail generates them. */
2069 if (strcmp(arg
, "1.0") == 0 ||
2070 strcmp(arg
, "1.00") == 0) {
2071 st
->minor_version
= 0;
2074 if (strcmp(arg
, "1.1") == 0 ||
2075 strcmp(arg
, "1.01") == 0
2077 st
->minor_version
= 1;
2080 if (strcmp(arg
, "1.2") == 0 ||
2081 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2082 strcmp(arg
, "default") == 0 ||
2083 #endif /* DEFAULT_OLD_METADATA */
2084 strcmp(arg
, "1.02") == 0) {
2085 st
->minor_version
= 2;
2088 if (strcmp(arg
, "1") == 0 ||
2089 strcmp(arg
, "default") == 0) {
2090 st
->minor_version
= -1;
2098 /* find available size on device with this devsize, using
2099 * superblock type st, and reserving 'reserve' sectors for
2102 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
2103 unsigned long long data_offset
)
2105 struct mdp_superblock_1
*super
= st
->sb
;
2112 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
2113 /* hot-add. allow for actual size of bitmap */
2114 struct bitmap_super_s
*bsb
;
2115 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2116 bmspace
= calc_bitmap_size(bsb
, 4096) >> 9;
2119 /* Allow space for bad block log */
2120 if (super
->bblog_size
)
2121 bbspace
= __le16_to_cpu(super
->bblog_size
);
2123 if (st
->minor_version
< 0)
2124 /* not specified, so time to set default */
2125 st
->minor_version
= 2;
2127 if (data_offset
== INVALID_SECTORS
)
2128 data_offset
= st
->data_offset
;
2130 if (data_offset
!= INVALID_SECTORS
)
2131 switch(st
->minor_version
) {
2133 return devsize
- data_offset
- 8*2 - bbspace
;
2136 return devsize
- data_offset
;
2143 switch(st
->minor_version
) {
2146 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2148 /* at start, 4K for superblock and possible bitmap */
2149 return devsize
- 4*2 - bbspace
;
2151 /* 4k from start, 4K for superblock and possible bitmap */
2152 return devsize
- (4+4)*2 - bbspace
;
2158 add_internal_bitmap1(struct supertype
*st
,
2159 int *chunkp
, int delay
, int write_behind
,
2160 unsigned long long size
,
2161 int may_change
, int major
)
2164 * If not may_change, then this is a 'Grow' without sysfs support for
2165 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2166 * If may_change, then this is create or a Grow with sysfs syupport,
2167 * and we can put the bitmap wherever we like.
2169 * size is in sectors, chunk is in bytes !!!
2172 unsigned long long bits
;
2173 unsigned long long max_bits
;
2174 unsigned long long min_chunk
;
2176 long bbl_offset
, bbl_size
;
2177 unsigned long long chunk
= *chunkp
;
2181 struct mdp_superblock_1
*sb
= st
->sb
;
2182 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2185 if (__le64_to_cpu(sb
->data_size
) == 0)
2186 /* Must be creating the array, else data_size would be non-zero */
2188 switch(st
->minor_version
) {
2190 /* either 3K after the superblock (when hot-add),
2191 * or some amount of space before.
2194 /* We are creating array, so we *know* how much room has
2199 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2201 room
= __le64_to_cpu(sb
->super_offset
)
2202 - __le64_to_cpu(sb
->data_offset
)
2203 - __le64_to_cpu(sb
->data_size
);
2204 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2207 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2208 if (bbl_size
< -bbl_offset
)
2209 bbl_size
= -bbl_offset
;
2211 if (!may_change
|| (room
< 3*2 &&
2212 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2217 offset
= 0; /* means movable offset */
2222 case 2: /* between superblock and data */
2226 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2228 room
= __le64_to_cpu(sb
->data_offset
)
2229 - __le64_to_cpu(sb
->super_offset
);
2230 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2232 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2237 room
-= 2; /* Leave 1K for superblock */
2241 room
-= 4*2; /* leave 4K for superblock */
2251 if (chunk
== UnSet
&& room
> 128*2)
2252 /* Limit to 128K of bitmap when chunk size not requested */
2256 /* No room for a bitmap */
2259 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2261 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2262 bits
= (size
*512)/min_chunk
+1;
2263 while (bits
> max_bits
) {
2267 if (chunk
== UnSet
) {
2268 /* For practical purpose, 64Meg is a good
2269 * default chunk size for internal bitmaps.
2272 if (chunk
< 64*1024*1024)
2273 chunk
= 64*1024*1024;
2274 } else if (chunk
< min_chunk
)
2275 return -EINVAL
; /* chunk size too small */
2276 if (chunk
== 0) /* rounding problem */
2280 /* start bitmap on a 4K boundary with enough space for
2283 bits
= (size
*512) / chunk
+ 1;
2284 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2285 room
*= 8; /* convert 4K blocks to sectors */
2286 offset
= -room
- bbl_size
;
2289 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2291 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2292 | MD_FEATURE_BITMAP_OFFSET
);
2293 memset(bms
, 0, sizeof(*bms
));
2294 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2295 bms
->version
= __cpu_to_le32(major
);
2296 uuid_from_super1(st
, uuid
);
2297 memcpy(bms
->uuid
, uuid
, 16);
2298 bms
->chunksize
= __cpu_to_le32(chunk
);
2299 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2300 bms
->sync_size
= __cpu_to_le64(size
);
2301 bms
->write_behind
= __cpu_to_le32(write_behind
);
2302 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2304 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2305 | MD_FEATURE_BITMAP_VERSIONED
);
2306 if (st
->cluster_name
) {
2307 len
= sizeof(bms
->cluster_name
);
2308 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, len
);
2309 bms
->cluster_name
[len
- 1] = '\0';
2316 static int locate_bitmap1(struct supertype
*st
, int fd
, int node_num
)
2318 unsigned long long offset
;
2319 struct mdp_superblock_1
*sb
;
2324 if (st
->ss
->load_super(st
, fd
, NULL
))
2325 return -1; /* no error I hope... */
2330 if ((__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
2334 offset
= __le64_to_cpu(sb
->super_offset
);
2335 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
) * (node_num
+ 1);
2338 lseek64(fd
, offset
<<9, 0);
2342 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2344 struct mdp_superblock_1
*sb
= st
->sb
;
2345 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2348 int towrite
, n
, len
;
2349 struct align_fd afd
;
2351 unsigned long long total_bm_space
, bm_space_per_node
;
2355 /* update cluster name */
2356 if (st
->cluster_name
) {
2357 len
= sizeof(bms
->cluster_name
);
2358 memset((char *)bms
->cluster_name
, 0, len
);
2359 strncpy((char *)bms
->cluster_name
,
2360 st
->cluster_name
, len
);
2361 bms
->cluster_name
[len
- 1] = '\0';
2365 /* cluster md only supports superblock 1.2 now */
2366 if (st
->minor_version
!= 2 && bms
->version
== BITMAP_MAJOR_CLUSTERED
) {
2367 pr_err("Warning: cluster md only works with superblock 1.2\n");
2371 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
) {
2372 if (st
->nodes
== 1) {
2373 /* the parameter for nodes is not valid */
2374 pr_err("Warning: cluster-md at least needs two nodes\n");
2376 } else if (st
->nodes
== 0)
2377 /* --nodes is not specified */
2379 else if (__cpu_to_le32(st
->nodes
) < bms
->nodes
) {
2380 /* Since the nodes num is not increased, no need to check the space
2381 * is enough or not, just update bms->nodes */
2382 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2386 /* no need to change bms->nodes for other bitmap types */
2388 pr_err("Warning: --nodes option is only suitable for clustered bitmap\n");
2392 /* Each node has an independent bitmap, it is necessary to calculate the
2393 * space is enough or not, first get how many bytes for the total bitmap */
2394 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2396 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2397 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2399 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2400 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2401 total_bm_space
/ bm_space_per_node
);
2405 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2414 locate_bitmap1(st
, fd
, 0);
2416 if (posix_memalign(&buf
, 4096, 4096))
2420 /* Only the bitmap[0] should resync
2421 * whole device on initial assembly
2424 memset(buf
, 0x00, 4096);
2426 memset(buf
, 0xff, 4096);
2427 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2430 * use 4096 boundary if bitmap_offset is aligned
2431 * with 8 sectors, then it should compatible with
2434 if (__le32_to_cpu(sb
->bitmap_offset
) & 7)
2435 towrite
= calc_bitmap_size(bms
, 512);
2437 towrite
= calc_bitmap_size(bms
, 4096);
2438 while (towrite
> 0) {
2442 n
= awrite(&afd
, buf
, n
);
2448 memset(buf
, 0x00, 4096);
2450 memset(buf
, 0xff, 4096);
2457 } while (++i
< __le32_to_cpu(bms
->nodes
));
2463 static void free_super1(struct supertype
*st
)
2469 struct devinfo
*di
= st
->info
;
2470 st
->info
= di
->next
;
2479 static int validate_geometry1(struct supertype
*st
, int level
,
2480 int layout
, int raiddisks
,
2481 int *chunk
, unsigned long long size
,
2482 unsigned long long data_offset
,
2483 char *subdev
, unsigned long long *freesize
,
2486 unsigned long long ldsize
, devsize
;
2488 unsigned long long headroom
;
2491 if (level
== LEVEL_CONTAINER
) {
2493 pr_err("1.x metadata does not support containers\n");
2496 if (*chunk
== UnSet
)
2497 *chunk
= DEFAULT_CHUNK
;
2502 if (st
->minor_version
< 0)
2503 /* not specified, so time to set default */
2504 st
->minor_version
= 2;
2506 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2509 pr_err("super1.x cannot open %s: %s\n",
2510 subdev
, strerror(errno
));
2514 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2520 devsize
= ldsize
>> 9;
2526 /* creating: allow suitable space for bitmap */
2527 bmspace
= choose_bm_space(devsize
);
2529 if (data_offset
== INVALID_SECTORS
)
2530 data_offset
= st
->data_offset
;
2531 if (data_offset
== INVALID_SECTORS
)
2532 switch (st
->minor_version
) {
2538 /* Choose data offset appropriate for this device
2539 * and use as default for whole array.
2540 * The data_offset must allow for bitmap space
2541 * and base metadata, should allow for some headroom
2542 * for reshape, and should be rounded to multiple
2544 * Headroom is limited to 128M, but aim for about 0.1%
2546 headroom
= 128*1024*2;
2547 while ((headroom
<< 10) > devsize
&&
2549 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2551 data_offset
= 12*2 + bmspace
+ headroom
;
2552 #define ONE_MEG (2*1024)
2553 if (data_offset
> ONE_MEG
)
2554 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2557 if (st
->data_offset
== INVALID_SECTORS
)
2558 st
->data_offset
= data_offset
;
2559 switch(st
->minor_version
) {
2560 case 0: /* metadata at end. Round down and subtract space to reserve */
2561 devsize
= (devsize
& ~(4ULL*2-1));
2562 /* space for metadata, bblog, bitmap */
2563 devsize
-= 8*2 + 8 + bmspace
;
2567 devsize
-= data_offset
;
2570 *freesize
= devsize
;
2573 #endif /* MDASSEMBLE */
2575 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2577 /* Create a v1.0 superblock based on 'info'*/
2579 struct mdp_superblock_1
*sb
;
2581 unsigned long long offset
;
2583 if (posix_memalign(&ret
, 4096, 1024) != 0)
2586 memset(ret
, 0, 1024);
2587 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2588 sb
->major_version
= __cpu_to_le32(1);
2590 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2591 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2592 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2593 sb
->level
= __cpu_to_le32(info
->array
.level
);
2594 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2595 sb
->size
= __cpu_to_le64(info
->component_size
);
2596 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2597 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2598 if (info
->array
.level
> 0)
2599 sb
->data_size
= sb
->size
;
2601 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2602 sb
->resync_offset
= MaxSector
;
2603 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2604 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2605 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2607 offset
= st
->devsize
/512 - 8*2;
2609 sb
->super_offset
= __cpu_to_le64(offset
);
2610 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2612 random_uuid(sb
->device_uuid
);
2614 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2615 int state
= sb0
->disks
[i
].state
;
2616 sb
->dev_roles
[i
] = MD_DISK_ROLE_SPARE
;
2617 if ((state
& (1<<MD_DISK_SYNC
)) &&
2618 !(state
& (1<<MD_DISK_FAULTY
)))
2619 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2621 sb
->sb_csum
= calc_sb_1_csum(sb
);
2625 struct superswitch super1
= {
2627 .examine_super
= examine_super1
,
2628 .brief_examine_super
= brief_examine_super1
,
2629 .export_examine_super
= export_examine_super1
,
2630 .detail_super
= detail_super1
,
2631 .brief_detail_super
= brief_detail_super1
,
2632 .export_detail_super
= export_detail_super1
,
2633 .write_init_super
= write_init_super1
,
2634 .validate_geometry
= validate_geometry1
,
2635 .add_to_super
= add_to_super1
,
2636 .examine_badblocks
= examine_badblocks_super1
,
2637 .copy_metadata
= copy_metadata1
,
2639 .match_home
= match_home1
,
2640 .uuid_from_super
= uuid_from_super1
,
2641 .getinfo_super
= getinfo_super1
,
2642 .container_content
= container_content1
,
2643 .update_super
= update_super1
,
2644 .init_super
= init_super1
,
2645 .store_super
= store_super1
,
2646 .compare_super
= compare_super1
,
2647 .load_super
= load_super1
,
2648 .match_metadata_desc
= match_metadata_desc1
,
2649 .avail_size
= avail_size1
,
2650 .add_internal_bitmap
= add_internal_bitmap1
,
2651 .locate_bitmap
= locate_bitmap1
,
2652 .write_bitmap
= write_bitmap1
,
2653 .free_super
= free_super1
,
2654 #if __BYTE_ORDER == BIG_ENDIAN