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 */
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 \
144 static int role_from_sb(struct mdp_superblock_1
*sb
)
149 d
= __le32_to_cpu(sb
->dev_number
);
150 if (d
< __le32_to_cpu(sb
->max_dev
))
151 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
153 role
= MD_DISK_ROLE_SPARE
;
157 /* return how many bytes are needed for bitmap, for cluster-md each node
158 * should have it's own bitmap */
159 static unsigned int calc_bitmap_size(bitmap_super_t
*bms
, unsigned int boundary
)
161 unsigned long long bits
, bytes
;
163 bits
= __le64_to_cpu(bms
->sync_size
) / (__le32_to_cpu(bms
->chunksize
)>>9);
164 bytes
= (bits
+7) >> 3;
165 bytes
+= sizeof(bitmap_super_t
);
166 bytes
= ROUND_UP(bytes
, boundary
);
171 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
173 unsigned int disk_csum
, csum
;
174 unsigned long long newcsum
;
175 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
176 unsigned int *isuper
= (unsigned int*)sb
;
178 /* make sure I can count... */
179 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
180 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
181 sizeof(struct mdp_superblock_1
) != 256) {
182 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
185 disk_csum
= sb
->sb_csum
;
188 for (; size
>=4; size
-= 4 ) {
189 newcsum
+= __le32_to_cpu(*isuper
);
194 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
196 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
197 sb
->sb_csum
= disk_csum
;
198 return __cpu_to_le32(csum
);
202 * Information related to file descriptor used for aligned reads/writes.
203 * Cache the block size.
210 static void init_afd(struct align_fd
*afd
, int fd
)
214 if (ioctl(afd
->fd
, BLKSSZGET
, &afd
->blk_sz
) != 0)
218 static char abuf
[4096+4096];
219 static int aread(struct align_fd
*afd
, void *buf
, int len
)
222 * On devices with a 4K sector size, we need to read
223 * the full sector and copy relevant bits into
232 if (!bsize
|| bsize
> 4096 || len
> 4096) {
234 fprintf(stderr
, "WARNING - aread() called with invalid block size\n");
237 b
= ROUND_UP_PTR((char *)abuf
, 4096);
239 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
241 n
= read(afd
->fd
, b
, iosize
);
244 lseek(afd
->fd
, len
- n
, 1);
251 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
254 * On devices with a 4K sector size, we need to write
255 * the full sector. We pre-read if the sector is larger
257 * The address must be sector-aligned.
264 if (!bsize
|| bsize
> 4096 || len
> 4096) {
266 fprintf(stderr
, "WARNING - awrite() called with invalid block size\n");
269 b
= ROUND_UP_PTR((char *)abuf
, 4096);
271 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
275 n
= read(afd
->fd
, b
, iosize
);
278 lseek(afd
->fd
, -n
, 1);
282 n
= write(afd
->fd
, b
, iosize
);
285 lseek(afd
->fd
, len
- n
, 1);
290 static void examine_super1(struct supertype
*st
, char *homehost
)
292 struct mdp_superblock_1
*sb
= st
->sb
;
293 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
300 int l
= homehost
? strlen(homehost
) : 0;
302 unsigned long long sb_offset
;
305 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
306 printf(" Version : 1");
307 sb_offset
= __le64_to_cpu(sb
->super_offset
);
310 else if (sb_offset
<= 8)
314 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
315 printf(" Array UUID : ");
316 for (i
=0; i
<16; i
++) {
317 if ((i
&3)==0 && i
!= 0) printf(":");
318 printf("%02x", sb
->set_uuid
[i
]);
321 printf(" Name : %.32s", sb
->set_name
);
322 if (l
> 0 && l
< 32 &&
323 sb
->set_name
[l
] == ':' &&
324 strncmp(sb
->set_name
, homehost
, l
) == 0)
325 printf(" (local to host %s)", homehost
);
328 printf(" Cluster Name : %-64s\n", bms
->cluster_name
);
329 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
330 printf(" Creation Time : %.24s\n", ctime(&atime
));
331 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
332 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
333 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
335 printf(" Avail Dev Size : %llu%s\n",
336 (unsigned long long)__le64_to_cpu(sb
->data_size
),
337 human_size(__le64_to_cpu(sb
->data_size
)<<9));
338 if (__le32_to_cpu(sb
->level
) > 0) {
339 int ddsks
= 0, ddsks_denom
= 1;
340 switch(__le32_to_cpu(sb
->level
)) {
341 case 1: ddsks
=1;break;
343 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
344 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
346 layout
= __le32_to_cpu(sb
->layout
);
347 ddsks
= __le32_to_cpu(sb
->raid_disks
);
348 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
351 long long asize
= __le64_to_cpu(sb
->size
);
352 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
353 printf(" Array Size : %llu%s\n",
354 asize
>> 10, human_size(asize
));
356 if (sb
->size
!= sb
->data_size
)
357 printf(" Used Dev Size : %llu%s\n",
358 (unsigned long long)__le64_to_cpu(sb
->size
),
359 human_size(__le64_to_cpu(sb
->size
)<<9));
362 printf(" Data Offset : %llu sectors\n",
363 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
364 if (sb
->new_offset
&&
365 (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
)) {
366 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
367 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
368 printf(" New Offset : %llu sectors\n", offset
);
370 printf(" Super Offset : %llu sectors\n",
371 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
372 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
373 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
375 st
->ss
->getinfo_super(st
, &info
, NULL
);
376 if (info
.space_after
!= 1 &&
377 !(__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
378 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
379 info
.space_before
, info
.space_after
);
381 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
382 printf(" Device UUID : ");
383 for (i
=0; i
<16; i
++) {
384 if ((i
&3)==0 && i
!= 0) printf(":");
385 printf("%02x", sb
->device_uuid
[i
]);
389 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
390 printf("Internal Bitmap : %ld sectors from superblock\n",
391 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
393 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
394 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
395 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
396 if (__le32_to_cpu(sb
->delta_disks
)) {
397 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
398 printf(" (%d->%d)\n",
399 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
400 __le32_to_cpu(sb
->raid_disks
));
401 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
402 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
404 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
405 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
406 printf(" New Level : %s\n", c
?c
:"-unknown-");
408 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
409 if (__le32_to_cpu(sb
->level
) == 5) {
410 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
411 printf(" New Layout : %s\n", c
?c
:"-unknown-");
413 if (__le32_to_cpu(sb
->level
) == 6) {
414 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
415 printf(" New Layout : %s\n", c
?c
:"-unknown-");
417 if (__le32_to_cpu(sb
->level
) == 10) {
418 printf(" New Layout :");
419 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
423 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
424 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
429 if (sb
->devflags
& WriteMostly1
)
430 printf(" write-mostly");
434 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
435 printf(" Update Time : %.24s\n", ctime(&atime
));
437 if (sb
->bblog_size
&& sb
->bblog_offset
) {
438 printf(" Bad Block Log : %d entries available at offset %ld sectors",
439 __le16_to_cpu(sb
->bblog_size
)*512/8,
440 (long)(int32_t)__le32_to_cpu(sb
->bblog_offset
));
441 if (sb
->feature_map
&
442 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
443 printf(" - bad blocks present.");
447 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
448 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
450 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
451 __le32_to_cpu(calc_sb_1_csum(sb
)));
452 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
454 if (__le32_to_cpu(sb
->level
) == 5) {
455 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
456 printf(" Layout : %s\n", c
?c
:"-unknown-");
458 if (__le32_to_cpu(sb
->level
) == 6) {
459 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
460 printf(" Layout : %s\n", c
?c
:"-unknown-");
462 if (__le32_to_cpu(sb
->level
) == 10) {
463 int lo
= __le32_to_cpu(sb
->layout
);
465 print_r10_layout(lo
);
468 switch(__le32_to_cpu(sb
->level
)) {
474 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
477 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
483 /* This turns out to just be confusing */
484 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
485 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
486 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != MD_DISK_ROLE_SPARE
)
488 for (d
=0; d
< i
; d
++) {
489 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
491 if (role
== MD_DISK_ROLE_SPARE
) printf("empty");
492 else if(role
== MD_DISK_ROLE_FAULTY
) printf("failed");
493 else printf("%d", role
);
497 printf(" Device Role : ");
498 role
= role_from_sb(sb
);
499 if (role
>= MD_DISK_ROLE_FAULTY
)
501 else if (role
== MD_DISK_ROLE_JOURNAL
)
503 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
504 printf("Replacement device %d\n", role
);
506 printf("Active device %d\n", role
);
508 printf(" Array State : ");
509 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
512 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
513 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
527 /* This is confusing too */
529 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
530 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
531 if (role
== MD_DISK_ROLE_FAULTY
)
534 if (faulty
) printf(" %d failed", faulty
);
536 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
540 static void brief_examine_super1(struct supertype
*st
, int verbose
)
542 struct mdp_superblock_1
*sb
= st
->sb
;
544 unsigned long long sb_offset
;
546 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
548 nm
= strchr(sb
->set_name
, ':');
551 else if (sb
->set_name
[0])
563 printf(" level=%s", c
);
564 sb_offset
= __le64_to_cpu(sb
->super_offset
);
566 printf(" metadata=1.1 ");
567 else if (sb_offset
<= 8)
568 printf(" metadata=1.2 ");
570 printf(" metadata=1.0 ");
572 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
574 for (i
=0; i
<16; i
++) {
575 if ((i
&3)==0 && i
!= 0) printf(":");
576 printf("%02x", sb
->set_uuid
[i
]);
578 if (sb
->set_name
[0]) {
580 print_quoted(sb
->set_name
);
585 static void export_examine_super1(struct supertype
*st
)
587 struct mdp_superblock_1
*sb
= st
->sb
;
592 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
593 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
595 if (sb
->set_name
[i
] == '\n' ||
596 sb
->set_name
[i
] == '\0') {
601 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
602 if (__le32_to_cpu(sb
->level
) > 0) {
603 int ddsks
= 0, ddsks_denom
= 1;
604 switch(__le32_to_cpu(sb
->level
)) {
605 case 1: ddsks
=1;break;
607 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
608 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
610 layout
= __le32_to_cpu(sb
->layout
);
611 ddsks
= __le32_to_cpu(sb
->raid_disks
);
612 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
615 long long asize
= __le64_to_cpu(sb
->size
);
616 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
617 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
621 for (i
=0; i
<16; i
++) {
622 if ((i
&3)==0 && i
!= 0) printf(":");
623 printf("%02x", sb
->set_uuid
[i
]);
626 printf("MD_UPDATE_TIME=%llu\n",
627 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
628 printf("MD_DEV_UUID=");
629 for (i
=0; i
<16; i
++) {
630 if ((i
&3)==0 && i
!= 0) printf(":");
631 printf("%02x", sb
->device_uuid
[i
]);
634 printf("MD_EVENTS=%llu\n",
635 (unsigned long long)__le64_to_cpu(sb
->events
));
638 static int copy_metadata1(struct supertype
*st
, int from
, int to
)
640 /* Read superblock. If it looks good, write it out.
641 * Then if a bitmap is present, copy that.
642 * And if a bad-block-list is present, copy that too.
645 unsigned long long dsize
, sb_offset
;
646 const int bufsize
= 4*1024;
647 struct mdp_superblock_1 super
, *sb
;
649 if (posix_memalign(&buf
, 4096, bufsize
) != 0)
652 if (!get_dev_size(from
, NULL
, &dsize
))
658 switch(st
->minor_version
) {
662 sb_offset
&= ~(4*2-1);
674 if (lseek64(from
, sb_offset
<< 9, 0) < 0LL)
676 if (read(from
, buf
, bufsize
) != bufsize
)
680 super
= *sb
; // save most of sb for when we reuse buf
682 if (__le32_to_cpu(super
.magic
) != MD_SB_MAGIC
||
683 __le32_to_cpu(super
.major_version
) != 1 ||
684 __le64_to_cpu(super
.super_offset
) != sb_offset
||
685 calc_sb_1_csum(sb
) != super
.sb_csum
)
688 if (lseek64(to
, sb_offset
<< 9, 0) < 0LL)
690 if (write(to
, buf
, bufsize
) != bufsize
)
693 if (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
)) {
694 unsigned long long bitmap_offset
= sb_offset
;
695 int bytes
= 4096; // just an estimate.
697 struct align_fd afrom
, ato
;
699 init_afd(&afrom
, from
);
702 bitmap_offset
+= (int32_t)__le32_to_cpu(super
.bitmap_offset
);
704 if (lseek64(from
, bitmap_offset
<<9, 0) < 0)
706 if (lseek64(to
, bitmap_offset
<<9, 0) < 0)
709 for (written
= 0; written
< bytes
; ) {
710 int n
= bytes
- written
;
713 if (aread(&afrom
, buf
, n
) != n
)
716 /* have the header, can calculate
717 * correct bitmap bytes */
720 bytes
= calc_bitmap_size(bms
, 512);
724 if (awrite(&ato
, buf
, n
) != n
)
730 if (super
.bblog_size
!= 0 &&
731 __le32_to_cpu(super
.bblog_size
) <= 100 &&
732 super
.bblog_offset
!= 0 &&
733 (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BAD_BLOCKS
))) {
734 /* There is a bad block log */
735 unsigned long long bb_offset
= sb_offset
;
736 int bytes
= __le32_to_cpu(super
.bblog_size
) * 512;
738 struct align_fd afrom
, ato
;
740 init_afd(&afrom
, from
);
743 bb_offset
+= (int32_t)__le32_to_cpu(super
.bblog_offset
);
745 if (lseek64(from
, bb_offset
<<9, 0) < 0)
747 if (lseek64(to
, bb_offset
<<9, 0) < 0)
750 for (written
= 0; written
< bytes
; ) {
751 int n
= bytes
- written
;
754 if (aread(&afrom
, buf
, n
) != n
)
757 if (awrite(&ato
, buf
, n
) != n
)
771 static void detail_super1(struct supertype
*st
, char *homehost
)
773 struct mdp_superblock_1
*sb
= st
->sb
;
774 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
776 int l
= homehost
? strlen(homehost
) : 0;
778 printf(" Name : %.32s", sb
->set_name
);
779 if (l
> 0 && l
< 32 &&
780 sb
->set_name
[l
] == ':' &&
781 strncmp(sb
->set_name
, homehost
, l
) == 0)
782 printf(" (local to host %s)", homehost
);
784 printf("\n Cluster Name : %-64s", bms
->cluster_name
);
785 printf("\n UUID : ");
786 for (i
=0; i
<16; i
++) {
787 if ((i
&3)==0 && i
!= 0) printf(":");
788 printf("%02x", sb
->set_uuid
[i
]);
790 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb
->events
));
793 static void brief_detail_super1(struct supertype
*st
)
795 struct mdp_superblock_1
*sb
= st
->sb
;
798 if (sb
->set_name
[0]) {
800 print_quoted(sb
->set_name
);
803 for (i
=0; i
<16; i
++) {
804 if ((i
&3)==0 && i
!= 0) printf(":");
805 printf("%02x", sb
->set_uuid
[i
]);
809 static void export_detail_super1(struct supertype
*st
)
811 struct mdp_superblock_1
*sb
= st
->sb
;
816 if (sb
->set_name
[i
] == '\n' ||
817 sb
->set_name
[i
] == '\0') {
822 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
825 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
827 struct mdp_superblock_1
*sb
= st
->sb
;
828 unsigned long long offset
;
833 if (!sb
->bblog_size
|| __le32_to_cpu(sb
->bblog_size
) > 100
834 || !sb
->bblog_offset
){
835 printf("No bad-blocks list configured on %s\n", devname
);
838 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
840 printf("Bad-blocks list is empty in %s\n", devname
);
844 size
= __le32_to_cpu(sb
->bblog_size
)* 512;
845 if (posix_memalign((void**)&bbl
, 4096, size
) != 0) {
846 pr_err("could not allocate badblocks list\n");
849 offset
= __le64_to_cpu(sb
->super_offset
) +
850 (int)__le32_to_cpu(sb
->bblog_offset
);
852 if (lseek64(fd
, offset
, 0) < 0) {
853 pr_err("Cannot seek to bad-blocks list\n");
856 if (read(fd
, bbl
, size
) != size
) {
857 pr_err("Cannot read bad-blocks list\n");
860 /* 64bits per entry. 10 bits is block-count, 54 bits is block
861 * offset. Blocks are sectors unless bblog->shift makes them bigger
864 printf("Bad-blocks on %s:\n", devname
);
865 for (i
= 0; i
< size
/8; i
++, bbp
++) {
866 __u64 bb
= __le64_to_cpu(*bbp
);
867 int count
= bb
& 0x3ff;
868 unsigned long long sector
= bb
>> 10;
873 sector
<<= sb
->bblog_shift
;
874 count
<<= sb
->bblog_shift
;
876 printf("%20llu for %d sectors\n", sector
, count
);
883 static int match_home1(struct supertype
*st
, char *homehost
)
885 struct mdp_superblock_1
*sb
= st
->sb
;
886 int l
= homehost
? strlen(homehost
) : 0;
888 return (l
> 0 && l
< 32 &&
889 sb
->set_name
[l
] == ':' &&
890 strncmp(sb
->set_name
, homehost
, l
) == 0);
893 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
895 struct mdp_superblock_1
*super
= st
->sb
;
896 char *cuuid
= (char*)uuid
;
899 cuuid
[i
] = super
->set_uuid
[i
];
902 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
904 struct mdp_superblock_1
*sb
= st
->sb
;
905 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
906 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
910 unsigned int map_disks
= info
->array
.raid_disks
;
911 unsigned long long super_offset
;
912 unsigned long long data_size
;
914 memset(info
, 0, sizeof(*info
));
915 info
->array
.major_version
= 1;
916 info
->array
.minor_version
= st
->minor_version
;
917 info
->array
.patch_version
= 0;
918 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
919 info
->array
.level
= __le32_to_cpu(sb
->level
);
920 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
921 info
->array
.md_minor
= -1;
922 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
923 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
924 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
926 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
928 if (__le32_to_cpu(bsb
->nodes
) > 1)
929 info
->array
.state
|= (1 << MD_SB_CLUSTERED
);
931 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
932 info
->component_size
= __le64_to_cpu(sb
->size
);
933 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
934 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
936 info
->disk
.major
= 0;
937 info
->disk
.minor
= 0;
938 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
939 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
940 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
941 role
= MD_DISK_ROLE_FAULTY
;
943 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
945 super_offset
= __le64_to_cpu(sb
->super_offset
);
946 if (info
->array
.level
<= 0)
947 data_size
= __le64_to_cpu(sb
->data_size
);
949 data_size
= __le64_to_cpu(sb
->size
);
950 if (info
->data_offset
< super_offset
) {
951 unsigned long long end
;
952 info
->space_before
= info
->data_offset
;
955 if (sb
->bblog_offset
&& sb
->bblog_size
) {
956 unsigned long long bboffset
= super_offset
;
957 bboffset
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
962 if (super_offset
+ info
->bitmap_offset
< end
)
963 end
= super_offset
+ info
->bitmap_offset
;
965 if (info
->data_offset
+ data_size
< end
)
966 info
->space_after
= end
- data_size
- info
->data_offset
;
968 info
->space_after
= 0;
970 unsigned long long earliest
;
971 earliest
= super_offset
+ (32+4)*2; /* match kernel */
972 if (info
->bitmap_offset
> 0) {
973 unsigned long long bmend
= info
->bitmap_offset
;
974 unsigned long long size
= __le64_to_cpu(bsb
->sync_size
);
975 size
/= __le32_to_cpu(bsb
->chunksize
) >> 9;
976 size
= (size
+ 7) >> 3;
977 size
+= sizeof(bitmap_super_t
);
978 size
= ROUND_UP(size
, 4096);
981 if (bmend
> earliest
)
984 if (sb
->bblog_offset
&& sb
->bblog_size
) {
985 unsigned long long bbend
= super_offset
;
986 bbend
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
987 bbend
+= __le32_to_cpu(sb
->bblog_size
);
988 if (bbend
> earliest
)
991 if (earliest
< info
->data_offset
)
992 info
->space_before
= info
->data_offset
- earliest
;
994 info
->space_before
= 0;
995 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
997 if (info
->space_before
== 0 && info
->space_after
== 0) {
998 /* It will look like we don't support data_offset changes,
999 * be we do - it's just that there is no room.
1000 * A change that reduced the number of devices should
1001 * still be allowed, so set the otherwise useless value of '1'
1003 info
->space_after
= 1;
1006 info
->disk
.raid_disk
= -1;
1008 case MD_DISK_ROLE_SPARE
:
1009 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
1011 case MD_DISK_ROLE_FAULTY
:
1012 info
->disk
.state
= 1; /* faulty */
1014 case MD_DISK_ROLE_JOURNAL
:
1015 info
->disk
.state
= (1 << MD_DISK_JOURNAL
);
1016 info
->disk
.raid_disk
= role
;
1017 info
->space_after
= (misc
->device_size
- info
->data_offset
) % 8; /* journal uses all 4kB blocks*/
1020 info
->disk
.state
= 6; /* active and in sync */
1021 info
->disk
.raid_disk
= role
;
1023 if (sb
->devflags
& WriteMostly1
)
1024 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1025 info
->events
= __le64_to_cpu(sb
->events
);
1026 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
1027 info
->safe_mode_delay
= 200;
1029 memcpy(info
->uuid
, sb
->set_uuid
, 16);
1031 strncpy(info
->name
, sb
->set_name
, 32);
1034 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
1035 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
1036 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
1039 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
1040 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
1042 info
->recovery_start
= MaxSector
;
1044 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
1045 info
->reshape_active
= 1;
1046 if ((sb
->feature_map
& __le32_to_cpu(MD_FEATURE_NEW_OFFSET
)) &&
1047 sb
->new_offset
!= 0)
1048 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
1049 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
1050 info
->new_level
= __le32_to_cpu(sb
->new_level
);
1051 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
1052 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
1053 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
1054 if (info
->delta_disks
< 0)
1055 info
->array
.raid_disks
-= info
->delta_disks
;
1057 info
->reshape_active
= 0;
1059 info
->recovery_blocked
= info
->reshape_active
;
1062 for (i
=0; i
<map_disks
; i
++)
1064 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
1065 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
1066 if (/*role == MD_DISK_ROLE_SPARE || */role
< (unsigned) info
->array
.raid_disks
) {
1068 if (map
&& role
< map_disks
)
1073 info
->array
.working_disks
= working
;
1074 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_JOURNAL
))
1075 info
->journal_device_required
= 1;
1076 info
->journal_clean
= 0;
1079 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
1081 struct mdinfo
*info
;
1086 info
= xmalloc(sizeof(*info
));
1087 getinfo_super1(st
, info
, NULL
);
1091 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
1093 char *devname
, int verbose
,
1094 int uuid_set
, char *homehost
)
1096 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1097 * if any change was made. For others, the return value is
1102 struct mdp_superblock_1
*sb
= st
->sb
;
1104 if (is_clustered(st
)) {
1105 rv
= cluster_get_dlmlock(st
, &lockid
);
1107 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1108 cluster_release_dlmlock(st
, 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 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1187 read(rfd
, sb
->device_uuid
, 16) != 16) {
1188 __u32 r
[4] = {random(), random(), random(), random()};
1189 memcpy(sb
->device_uuid
, r
, 16);
1195 __cpu_to_le16(info
->disk
.raid_disk
);
1197 fd
= open(devname
, O_RDONLY
);
1199 unsigned long long ds
;
1200 get_dev_size(fd
, devname
, &ds
);
1203 if (__le64_to_cpu(sb
->super_offset
) <
1204 __le64_to_cpu(sb
->data_offset
)) {
1205 sb
->data_size
= __cpu_to_le64(
1206 ds
- __le64_to_cpu(sb
->data_offset
));
1209 ds
&= ~(unsigned long long)(4*2-1);
1210 sb
->super_offset
= __cpu_to_le64(ds
);
1211 sb
->data_size
= __cpu_to_le64(
1212 ds
- __le64_to_cpu(sb
->data_offset
));
1215 } else if (strcmp(update
, "linear-grow-update") == 0) {
1216 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1217 sb
->dev_roles
[info
->disk
.number
] =
1218 __cpu_to_le16(info
->disk
.raid_disk
);
1219 } else if (strcmp(update
, "resync") == 0) {
1220 /* make sure resync happens */
1221 sb
->resync_offset
= 0ULL;
1222 } else if (strcmp(update
, "uuid") == 0) {
1223 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1225 if (__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1226 struct bitmap_super_s
*bm
;
1227 bm
= (struct bitmap_super_s
*)(st
->sb
+MAX_SB_SIZE
);
1228 memcpy(bm
->uuid
, sb
->set_uuid
, 16);
1230 } else if (strcmp(update
, "no-bitmap") == 0) {
1231 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1232 } else if (strcmp(update
, "bbl") == 0) {
1233 /* only possible if there is room after the bitmap, or if
1234 * there is no bitmap
1236 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1237 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1238 long bitmap_offset
= (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
);
1239 long bm_sectors
= 0;
1243 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1244 struct bitmap_super_s
*bsb
;
1245 bsb
= (struct bitmap_super_s
*)(((char*)sb
)+MAX_SB_SIZE
);
1246 bm_sectors
= bitmap_sectors(bsb
);
1249 if (sb_offset
< data_offset
) {
1250 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1253 bb_offset
= sb_offset
+ 8;
1254 if (bm_sectors
&& bitmap_offset
> 0)
1255 bb_offset
= bitmap_offset
+ bm_sectors
;
1256 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1257 && bb_offset
+ 8+8 <= (long)data_offset
)
1258 /* too close to bitmap, and room to grow */
1260 if (bb_offset
+ 8 <= (long)data_offset
) {
1261 sb
->bblog_size
= __cpu_to_le16(8);
1262 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1265 /* 1.0 - Put bbl just before super block */
1266 if (bm_sectors
&& bitmap_offset
< 0)
1267 space
= -bitmap_offset
- bm_sectors
;
1269 space
= sb_offset
- data_offset
-
1270 __le64_to_cpu(sb
->data_size
);
1272 sb
->bblog_size
= __cpu_to_le16(8);
1273 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1276 } else if (strcmp(update
, "no-bbl") == 0) {
1277 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1278 pr_err("Cannot remove active bbl from %s\n",devname
);
1281 sb
->bblog_shift
= 0;
1282 sb
->bblog_offset
= 0;
1284 } else if (strcmp(update
, "name") == 0) {
1285 if (info
->name
[0] == 0)
1286 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1287 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1289 strchr(info
->name
, ':') == NULL
&&
1290 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1291 strcpy(sb
->set_name
, homehost
);
1292 strcat(sb
->set_name
, ":");
1293 strcat(sb
->set_name
, info
->name
);
1295 strcpy(sb
->set_name
, info
->name
);
1296 } else if (strcmp(update
, "devicesize") == 0 &&
1297 __le64_to_cpu(sb
->super_offset
) <
1298 __le64_to_cpu(sb
->data_offset
)) {
1299 /* set data_size to device size less data_offset */
1300 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1301 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1302 sb
->data_size
= __cpu_to_le64(
1303 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1304 } else if (strcmp(update
, "revert-reshape") == 0) {
1306 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1307 pr_err("No active reshape to revert on %s\n",
1311 unsigned long long reshape_sectors
;
1314 /* reshape_position is a little messy.
1315 * Its value must be a multiple of the larger
1316 * chunk size, and of the "after" data disks.
1317 * So when reverting we need to change it to
1318 * be a multiple of the new "after" data disks,
1319 * which is the old "before".
1320 * If it isn't already a multiple of 'before',
1321 * the only thing we could do would be
1322 * copy some block around on the disks, which
1323 * is easy to get wrong.
1324 * So we reject a revert-reshape unless the
1325 * alignment is good.
1327 if (__le32_to_cpu(sb
->level
) >= 4 &&
1328 __le32_to_cpu(sb
->level
) <= 6) {
1329 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1330 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1331 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1332 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1333 if (reshape_sectors
% reshape_chunk
) {
1334 pr_err("Reshape position is not suitably aligned.\n");
1335 pr_err("Try normal assembly and stop again\n");
1339 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1340 __le32_to_cpu(sb
->delta_disks
));
1341 if (sb
->delta_disks
== 0)
1342 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1344 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1346 temp
= sb
->new_layout
;
1347 sb
->new_layout
= sb
->layout
;
1350 temp
= sb
->new_chunk
;
1351 sb
->new_chunk
= sb
->chunksize
;
1352 sb
->chunksize
= temp
;
1354 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1355 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1356 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1357 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1358 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1361 } else if (strcmp(update
, "_reshape_progress")==0)
1362 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1363 else if (strcmp(update
, "writemostly")==0)
1364 sb
->devflags
|= WriteMostly1
;
1365 else if (strcmp(update
, "readwrite")==0)
1366 sb
->devflags
&= ~WriteMostly1
;
1370 sb
->sb_csum
= calc_sb_1_csum(sb
);
1371 if (is_clustered(st
))
1372 cluster_release_dlmlock(st
, lockid
);
1377 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1378 unsigned long long size
, char *name
, char *homehost
,
1379 int *uuid
, unsigned long long data_offset
)
1381 struct mdp_superblock_1
*sb
;
1387 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1388 pr_err("could not allocate superblock\n");
1391 memset(sb
, 0, SUPER1_SIZE
);
1395 /* zeroing superblock */
1399 spares
= info
->working_disks
- info
->active_disks
;
1400 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1401 pr_err("too many devices requested: %d+%d > %d\n",
1402 info
->raid_disks
, spares
, MAX_DEVS
);
1406 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1407 sb
->major_version
= __cpu_to_le32(1);
1408 sb
->feature_map
= 0;
1412 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1414 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1415 read(rfd
, sb
->set_uuid
, 16) != 16) {
1416 __u32 r
[4] = {random(), random(), random(), random()};
1417 memcpy(sb
->set_uuid
, r
, 16);
1419 if (rfd
>= 0) close(rfd
);
1422 if (name
== NULL
|| *name
== 0) {
1423 sprintf(defname
, "%d", info
->md_minor
);
1427 strchr(name
, ':')== NULL
&&
1428 strlen(homehost
)+1+strlen(name
) < 32) {
1429 strcpy(sb
->set_name
, homehost
);
1430 strcat(sb
->set_name
, ":");
1431 strcat(sb
->set_name
, name
);
1433 strcpy(sb
->set_name
, name
);
1435 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1436 sb
->level
= __cpu_to_le32(info
->level
);
1437 sb
->layout
= __cpu_to_le32(info
->layout
);
1438 sb
->size
= __cpu_to_le64(size
*2ULL);
1439 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1440 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1442 sb
->data_offset
= __cpu_to_le64(data_offset
);
1443 sb
->data_size
= __cpu_to_le64(0);
1444 sb
->super_offset
= __cpu_to_le64(0);
1445 sb
->recovery_offset
= __cpu_to_le64(0);
1447 sb
->utime
= sb
->ctime
;
1448 sb
->events
= __cpu_to_le64(1);
1449 if (info
->state
& (1<<MD_SB_CLEAN
))
1450 sb
->resync_offset
= MaxSector
;
1452 sb
->resync_offset
= 0;
1453 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1454 sbsize
= ROUND_UP(sbsize
, 512);
1455 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1457 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1465 long long data_offset
;
1466 mdu_disk_info_t disk
;
1467 struct devinfo
*next
;
1470 /* Add a device to the superblock being created */
1471 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1472 int fd
, char *devname
, unsigned long long data_offset
)
1474 struct mdp_superblock_1
*sb
= st
->sb
;
1475 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1476 struct devinfo
*di
, **dip
;
1479 if (is_clustered(st
)) {
1480 rv
= cluster_get_dlmlock(st
, &lockid
);
1482 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1483 cluster_release_dlmlock(st
, lockid
);
1488 if ((dk
->state
& 6) == 6) /* active, sync */
1489 *rp
= __cpu_to_le16(dk
->raid_disk
);
1490 else if (dk
->state
& (1<<MD_DISK_JOURNAL
))
1491 *rp
= MD_DISK_ROLE_JOURNAL
;
1492 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1493 *rp
= MD_DISK_ROLE_SPARE
;
1495 *rp
= MD_DISK_ROLE_FAULTY
;
1497 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1498 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1499 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1501 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1502 sb
->devflags
= 0; /* don't copy another disks flags */
1503 sb
->sb_csum
= calc_sb_1_csum(sb
);
1505 dip
= (struct devinfo
**)&st
->info
;
1507 dip
= &(*dip
)->next
;
1508 di
= xmalloc(sizeof(struct devinfo
));
1510 di
->devname
= devname
;
1512 di
->data_offset
= data_offset
;
1516 if (is_clustered(st
))
1517 cluster_release_dlmlock(st
, lockid
);
1523 static void locate_bitmap1(struct supertype
*st
, int fd
);
1525 static int store_super1(struct supertype
*st
, int fd
)
1527 struct mdp_superblock_1
*sb
= st
->sb
;
1528 unsigned long long sb_offset
;
1529 struct align_fd afd
;
1531 unsigned long long dsize
;
1534 if (is_clustered(st
)) {
1535 rv
= cluster_get_dlmlock(st
, &lockid
);
1537 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1538 cluster_release_dlmlock(st
, lockid
);
1543 if (!get_dev_size(fd
, NULL
, &dsize
))
1554 * Calculate the position of the superblock.
1555 * It is always aligned to a 4K boundary and
1556 * depending on minor_version, it can be:
1557 * 0: At least 8K, but less than 12K, from end of device
1558 * 1: At start of device
1559 * 2: 4K from start of device.
1561 switch(st
->minor_version
) {
1565 sb_offset
&= ~(4*2-1);
1577 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1578 0 != __le64_to_cpu(sb
->super_offset
)
1580 pr_err("internal error - sb_offset is wrong\n");
1584 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1587 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1589 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1592 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1593 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1594 (((char*)sb
)+MAX_SB_SIZE
);
1595 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1596 locate_bitmap1(st
, fd
);
1597 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1602 if (is_clustered(st
))
1603 cluster_release_dlmlock(st
, lockid
);
1608 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1610 static unsigned long choose_bm_space(unsigned long devsize
)
1612 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1613 * if bigger than 200Gig, save 128k
1614 * NOTE: result must be multiple of 4K else bad things happen
1615 * on 4K-sector devices.
1617 if (devsize
< 64*2) return 0;
1618 if (devsize
- 64*2 >= 200*1024*1024*2)
1620 if (devsize
- 4*2 > 8*1024*1024*2)
1625 static void free_super1(struct supertype
*st
);
1627 #define META_BLOCK_SIZE 4096
1628 unsigned long crc32(
1630 const unsigned char *buf
,
1633 static int write_empty_r5l_meta_block(struct supertype
*st
, int fd
)
1635 struct r5l_meta_block
*mb
;
1636 struct mdp_superblock_1
*sb
= st
->sb
;
1637 struct align_fd afd
;
1642 if (posix_memalign((void**)&mb
, 4096, META_BLOCK_SIZE
) != 0) {
1643 pr_err("Could not allocate memory for the meta block.\n");
1647 memset(mb
, 0, META_BLOCK_SIZE
);
1649 mb
->magic
= __cpu_to_le32(R5LOG_MAGIC
);
1650 mb
->version
= R5LOG_VERSION
;
1651 mb
->meta_size
= __cpu_to_le32(sizeof(struct r5l_meta_block
));
1652 mb
->seq
= __cpu_to_le64(random32());
1653 mb
->position
= __cpu_to_le64(0);
1655 crc
= crc32(0xffffffff, sb
->set_uuid
, sizeof(sb
->set_uuid
));
1656 crc
= crc32(crc
, (void *)mb
, META_BLOCK_SIZE
);
1657 mb
->checksum
= __cpu_to_le32(crc
);
1659 if (lseek64(fd
, (sb
->data_offset
) * 512, 0) < 0LL) {
1660 pr_err("cannot seek to offset of the meta block\n");
1664 if (awrite(&afd
, mb
, META_BLOCK_SIZE
) != META_BLOCK_SIZE
) {
1665 pr_err("failed to store write the meta block \n");
1679 static int write_init_super1(struct supertype
*st
)
1681 struct mdp_superblock_1
*sb
= st
->sb
;
1682 struct supertype
*refst
;
1685 unsigned long long bm_space
;
1687 unsigned long long dsize
, array_size
;
1688 unsigned long long sb_offset
;
1689 unsigned long long data_offset
;
1691 for (di
= st
->info
; di
; di
= di
->next
) {
1692 if (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))
1693 sb
->feature_map
|= MD_FEATURE_JOURNAL
;
1696 for (di
= st
->info
; di
; di
= di
->next
) {
1697 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1702 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1705 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1706 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1707 sb
->devflags
|= WriteMostly1
;
1709 sb
->devflags
&= ~WriteMostly1
;
1711 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1712 read(rfd
, sb
->device_uuid
, 16) != 16) {
1713 __u32 r
[4] = {random(), random(), random(), random()};
1714 memcpy(sb
->device_uuid
, r
, 16);
1721 refst
= dup_super(st
);
1722 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1723 struct mdp_superblock_1
*refsb
= refst
->sb
;
1725 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1726 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1727 /* same array, so preserve events and
1729 sb
->events
= refsb
->events
;
1730 /* bugs in 2.6.17 and earlier mean the
1731 * dev_number chosen in Manage must be preserved
1733 if (get_linux_version() >= 2006018)
1734 sb
->dev_number
= refsb
->dev_number
;
1740 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1753 * Calculate the position of the superblock.
1754 * It is always aligned to a 4K boundary and
1755 * depending on minor_version, it can be:
1756 * 0: At least 8K, but less than 12K, from end of device
1757 * 1: At start of device
1758 * 2: 4K from start of device.
1759 * data_offset has already been set.
1761 array_size
= __le64_to_cpu(sb
->size
);
1762 /* work out how much space we left for a bitmap,
1763 * Add 8 sectors for bad block log */
1764 bm_space
= choose_bm_space(array_size
) + 8;
1766 data_offset
= di
->data_offset
;
1767 if (data_offset
== INVALID_SECTORS
)
1768 data_offset
= st
->data_offset
;
1769 switch(st
->minor_version
) {
1771 if (data_offset
== INVALID_SECTORS
)
1775 sb_offset
&= ~(4*2-1);
1776 sb
->data_offset
= __cpu_to_le64(data_offset
);
1777 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1778 if (sb_offset
< array_size
+ bm_space
)
1779 bm_space
= sb_offset
- array_size
;
1780 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1781 if (bm_space
>= 8) {
1782 sb
->bblog_size
= __cpu_to_le16(8);
1783 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1787 sb
->super_offset
= __cpu_to_le64(0);
1788 if (data_offset
== INVALID_SECTORS
)
1791 sb
->data_offset
= __cpu_to_le64(data_offset
);
1792 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1793 if (data_offset
>= 8 + 32*2 + 8) {
1794 sb
->bblog_size
= __cpu_to_le16(8);
1795 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1796 } else if (data_offset
>= 16) {
1797 sb
->bblog_size
= __cpu_to_le16(8);
1798 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8);
1803 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1804 if (data_offset
== INVALID_SECTORS
)
1807 sb
->data_offset
= __cpu_to_le64(data_offset
);
1808 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1809 if (data_offset
>= 16 + 32*2 + 8) {
1810 sb
->bblog_size
= __cpu_to_le16(8);
1811 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1812 } else if (data_offset
>= 16+16) {
1813 sb
->bblog_size
= __cpu_to_le16(8);
1814 /* '8' sectors for the bblog, and another '8'
1815 * because we want offset from superblock, not
1818 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8-8);
1822 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1823 st
->minor_version
, di
->devname
);
1827 /* Disable badblock log on clusters, or when explicitly requested */
1828 if (st
->nodes
> 0 || conf_get_create_info()->bblist
== 0) {
1830 sb
->bblog_offset
= 0;
1833 sb
->sb_csum
= calc_sb_1_csum(sb
);
1834 rv
= store_super1(st
, di
->fd
);
1836 if (rv
== 0 && (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))) {
1837 rv
= write_empty_r5l_meta_block(st
, di
->fd
);
1842 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1843 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NoUpdate
);
1851 pr_err("Failed to write metadata to %s\n",
1858 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1862 * 0 same, or first was empty, and second was copied
1863 * 1 second had wrong number
1865 * 3 wrong other info
1867 struct mdp_superblock_1
*first
= st
->sb
;
1868 struct mdp_superblock_1
*second
= tst
->sb
;
1870 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1872 if (second
->major_version
!= __cpu_to_le32(1))
1876 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1877 pr_err("could not allocate superblock\n");
1880 memcpy(first
, second
, SUPER1_SIZE
);
1884 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1887 if (first
->ctime
!= second
->ctime
||
1888 first
->level
!= second
->level
||
1889 first
->layout
!= second
->layout
||
1890 first
->size
!= second
->size
||
1891 first
->chunksize
!= second
->chunksize
||
1892 first
->raid_disks
!= second
->raid_disks
)
1897 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1899 unsigned long long dsize
;
1900 unsigned long long sb_offset
;
1901 struct mdp_superblock_1
*super
;
1903 struct bitmap_super_s
*bsb
;
1904 struct misc_dev_info
*misc
;
1905 struct align_fd afd
;
1911 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1913 struct supertype tst
;
1914 __u64 bestctime
= 0;
1915 /* guess... choose latest ctime */
1916 memset(&tst
, 0, sizeof(tst
));
1918 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1919 switch(load_super1(&tst
, fd
, devname
)) {
1920 case 0: super
= tst
.sb
;
1921 if (bestvers
== -1 ||
1922 bestctime
< __le64_to_cpu(super
->ctime
)) {
1923 bestvers
= tst
.minor_version
;
1924 bestctime
= __le64_to_cpu(super
->ctime
);
1929 case 1: return 1; /*bad device */
1930 case 2: break; /* bad, try next */
1933 if (bestvers
!= -1) {
1935 tst
.minor_version
= bestvers
;
1937 tst
.max_devs
= MAX_DEVS
;
1938 rv
= load_super1(&tst
, fd
, devname
);
1945 if (!get_dev_size(fd
, devname
, &dsize
))
1951 pr_err("%s is too small for md: size is %llu sectors.\n",
1957 * Calculate the position of the superblock.
1958 * It is always aligned to a 4K boundary and
1959 * depending on minor_version, it can be:
1960 * 0: At least 8K, but less than 12K, from end of device
1961 * 1: At start of device
1962 * 2: 4K from start of device.
1964 switch(st
->minor_version
) {
1968 sb_offset
&= ~(4*2-1);
1980 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1982 pr_err("Cannot seek to superblock on %s: %s\n",
1983 devname
, strerror(errno
));
1987 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
1988 pr_err("could not allocate superblock\n");
1992 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
1994 pr_err("Cannot read superblock on %s\n",
2000 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
2002 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2003 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
2008 if (__le32_to_cpu(super
->major_version
) != 1) {
2010 pr_err("Cannot interpret superblock on %s - version is %d\n",
2011 devname
, __le32_to_cpu(super
->major_version
));
2015 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
2017 pr_err("No superblock found on %s (super_offset is wrong)\n",
2024 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2026 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
2027 misc
->device_size
= dsize
;
2028 if (st
->data_offset
== INVALID_SECTORS
)
2029 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
2031 /* Now check on the bitmap superblock */
2032 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
2034 /* Read the bitmap superblock and make sure it looks
2035 * valid. If it doesn't clear the bit. An --assemble --force
2036 * should get that written out.
2038 locate_bitmap1(st
, fd
);
2039 if (aread(&afd
, bsb
, 512) != 512)
2042 uuid_from_super1(st
, uuid
);
2043 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
2044 memcmp(bsb
->uuid
, uuid
, 16) != 0)
2049 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
2050 & ~MD_FEATURE_BITMAP_OFFSET
);
2054 static struct supertype
*match_metadata_desc1(char *arg
)
2056 struct supertype
*st
= xcalloc(1, sizeof(*st
));
2058 st
->container_devnm
[0] = 0;
2060 st
->max_devs
= MAX_DEVS
;
2062 st
->data_offset
= INVALID_SECTORS
;
2063 /* leading zeros can be safely ignored. --detail generates them. */
2066 if (strcmp(arg
, "1.0") == 0 ||
2067 strcmp(arg
, "1.00") == 0) {
2068 st
->minor_version
= 0;
2071 if (strcmp(arg
, "1.1") == 0 ||
2072 strcmp(arg
, "1.01") == 0
2074 st
->minor_version
= 1;
2077 if (strcmp(arg
, "1.2") == 0 ||
2078 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2079 strcmp(arg
, "default") == 0 ||
2080 #endif /* DEFAULT_OLD_METADATA */
2081 strcmp(arg
, "1.02") == 0) {
2082 st
->minor_version
= 2;
2085 if (strcmp(arg
, "1") == 0 ||
2086 strcmp(arg
, "default") == 0) {
2087 st
->minor_version
= -1;
2095 /* find available size on device with this devsize, using
2096 * superblock type st, and reserving 'reserve' sectors for
2099 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
2100 unsigned long long data_offset
)
2102 struct mdp_superblock_1
*super
= st
->sb
;
2109 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
2110 /* hot-add. allow for actual size of bitmap */
2111 struct bitmap_super_s
*bsb
;
2112 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2113 bmspace
= bitmap_sectors(bsb
);
2116 /* Allow space for bad block log */
2117 if (super
->bblog_size
)
2118 bbspace
= __le16_to_cpu(super
->bblog_size
);
2120 if (st
->minor_version
< 0)
2121 /* not specified, so time to set default */
2122 st
->minor_version
= 2;
2124 if (data_offset
== INVALID_SECTORS
)
2125 data_offset
= st
->data_offset
;
2127 if (data_offset
!= INVALID_SECTORS
)
2128 switch(st
->minor_version
) {
2130 return devsize
- data_offset
- 8*2 - bbspace
;
2133 return devsize
- data_offset
;
2140 switch(st
->minor_version
) {
2143 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2145 /* at start, 4K for superblock and possible bitmap */
2146 return devsize
- 4*2 - bbspace
;
2148 /* 4k from start, 4K for superblock and possible bitmap */
2149 return devsize
- (4+4)*2 - bbspace
;
2155 add_internal_bitmap1(struct supertype
*st
,
2156 int *chunkp
, int delay
, int write_behind
,
2157 unsigned long long size
,
2158 int may_change
, int major
)
2161 * If not may_change, then this is a 'Grow' without sysfs support for
2162 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2163 * If may_change, then this is create or a Grow with sysfs syupport,
2164 * and we can put the bitmap wherever we like.
2166 * size is in sectors, chunk is in bytes !!!
2169 unsigned long long bits
;
2170 unsigned long long max_bits
;
2171 unsigned long long min_chunk
;
2173 long bbl_offset
, bbl_size
;
2174 unsigned long long chunk
= *chunkp
;
2177 struct mdp_superblock_1
*sb
= st
->sb
;
2178 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2181 if (__le64_to_cpu(sb
->data_size
) == 0)
2182 /* Must be creating the array, else data_size would be non-zero */
2184 switch(st
->minor_version
) {
2186 /* either 3K after the superblock (when hot-add),
2187 * or some amount of space before.
2190 /* We are creating array, so we *know* how much room has
2195 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2197 room
= __le64_to_cpu(sb
->super_offset
)
2198 - __le64_to_cpu(sb
->data_offset
)
2199 - __le64_to_cpu(sb
->data_size
);
2200 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2203 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2204 if (bbl_size
< -bbl_offset
)
2205 bbl_size
= -bbl_offset
;
2207 if (!may_change
|| (room
< 3*2 &&
2208 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2213 offset
= 0; /* means movable offset */
2218 case 2: /* between superblock and data */
2222 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2224 room
= __le64_to_cpu(sb
->data_offset
)
2225 - __le64_to_cpu(sb
->super_offset
);
2226 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2228 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2233 room
-= 2; /* Leave 1K for superblock */
2237 room
-= 4*2; /* leave 4K for superblock */
2247 if (chunk
== UnSet
&& room
> 128*2)
2248 /* Limit to 128K of bitmap when chunk size not requested */
2252 /* No room for a bitmap */
2255 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2257 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2258 bits
= (size
*512)/min_chunk
+1;
2259 while (bits
> max_bits
) {
2263 if (chunk
== UnSet
) {
2264 /* For practical purpose, 64Meg is a good
2265 * default chunk size for internal bitmaps.
2268 if (chunk
< 64*1024*1024)
2269 chunk
= 64*1024*1024;
2270 } else if (chunk
< min_chunk
)
2271 return 0; /* chunk size too small */
2272 if (chunk
== 0) /* rounding problem */
2276 /* start bitmap on a 4K boundary with enough space for
2279 bits
= (size
*512) / chunk
+ 1;
2280 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2281 room
*= 8; /* convert 4K blocks to sectors */
2282 offset
= -room
- bbl_size
;
2285 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2287 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2288 | MD_FEATURE_BITMAP_OFFSET
);
2289 memset(bms
, 0, sizeof(*bms
));
2290 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2291 bms
->version
= __cpu_to_le32(major
);
2292 uuid_from_super1(st
, uuid
);
2293 memcpy(bms
->uuid
, uuid
, 16);
2294 bms
->chunksize
= __cpu_to_le32(chunk
);
2295 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2296 bms
->sync_size
= __cpu_to_le64(size
);
2297 bms
->write_behind
= __cpu_to_le32(write_behind
);
2298 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2300 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2301 | MD_FEATURE_BITMAP_VERSIONED
);
2302 if (st
->cluster_name
)
2303 strncpy((char *)bms
->cluster_name
,
2304 st
->cluster_name
, strlen(st
->cluster_name
));
2310 static void locate_bitmap1(struct supertype
*st
, int fd
)
2312 unsigned long long offset
;
2313 struct mdp_superblock_1
*sb
;
2317 if (st
->ss
->load_super(st
, fd
, NULL
))
2318 return; /* no error I hope... */
2323 offset
= __le64_to_cpu(sb
->super_offset
);
2324 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
);
2327 lseek64(fd
, offset
<<9, 0);
2330 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2332 struct mdp_superblock_1
*sb
= st
->sb
;
2333 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2337 struct align_fd afd
;
2339 unsigned long long total_bm_space
, bm_space_per_node
;
2343 /* update cluster name */
2344 if (st
->cluster_name
) {
2345 memset((char *)bms
->cluster_name
, 0, sizeof(bms
->cluster_name
));
2346 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, 64);
2350 /* cluster md only supports superblock 1.2 now */
2351 if (st
->minor_version
!= 2) {
2352 pr_err("Warning: cluster md only works with superblock 1.2\n");
2356 /* Each node has an independent bitmap, it is necessary to calculate the
2357 * space is enough or not, first get how many bytes for the total bitmap */
2358 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2360 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2361 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2363 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2364 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2365 total_bm_space
/ bm_space_per_node
);
2369 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2378 locate_bitmap1(st
, fd
);
2380 if (posix_memalign(&buf
, 4096, 4096))
2384 /* Only the bitmap[0] should resync
2385 * whole device on initial assembly
2388 memset(buf
, 0x00, 4096);
2390 memset(buf
, 0xff, 4096);
2391 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2393 towrite
= calc_bitmap_size(bms
, 4096);
2394 while (towrite
> 0) {
2398 n
= awrite(&afd
, buf
, n
);
2404 memset(buf
, 0x00, 4096);
2406 memset(buf
, 0xff, 4096);
2413 } while (++i
< __le32_to_cpu(bms
->nodes
));
2419 static void free_super1(struct supertype
*st
)
2422 if (is_clustered(st
)) {
2423 rv
= cluster_get_dlmlock(st
, &lockid
);
2425 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
2426 cluster_release_dlmlock(st
, lockid
);
2434 struct devinfo
*di
= st
->info
;
2435 st
->info
= di
->next
;
2441 if (is_clustered(st
))
2442 cluster_release_dlmlock(st
, lockid
);
2446 static int validate_geometry1(struct supertype
*st
, int level
,
2447 int layout
, int raiddisks
,
2448 int *chunk
, unsigned long long size
,
2449 unsigned long long data_offset
,
2450 char *subdev
, unsigned long long *freesize
,
2453 unsigned long long ldsize
, devsize
;
2455 unsigned long long headroom
;
2458 if (level
== LEVEL_CONTAINER
) {
2460 pr_err("1.x metadata does not support containers\n");
2463 if (*chunk
== UnSet
)
2464 *chunk
= DEFAULT_CHUNK
;
2469 if (st
->minor_version
< 0)
2470 /* not specified, so time to set default */
2471 st
->minor_version
= 2;
2473 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2476 pr_err("super1.x cannot open %s: %s\n",
2477 subdev
, strerror(errno
));
2481 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2487 devsize
= ldsize
>> 9;
2493 /* creating: allow suitable space for bitmap */
2494 bmspace
= choose_bm_space(devsize
);
2496 if (data_offset
== INVALID_SECTORS
)
2497 data_offset
= st
->data_offset
;
2498 if (data_offset
== INVALID_SECTORS
)
2499 switch (st
->minor_version
) {
2505 /* Choose data offset appropriate for this device
2506 * and use as default for whole array.
2507 * The data_offset must allow for bitmap space
2508 * and base metadata, should allow for some headroom
2509 * for reshape, and should be rounded to multiple
2511 * Headroom is limited to 128M, but aim for about 0.1%
2513 headroom
= 128*1024*2;
2514 while ((headroom
<< 10) > devsize
&&
2516 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2518 data_offset
= 12*2 + bmspace
+ headroom
;
2519 #define ONE_MEG (2*1024)
2520 if (data_offset
> ONE_MEG
)
2521 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2524 if (st
->data_offset
== INVALID_SECTORS
)
2525 st
->data_offset
= data_offset
;
2526 switch(st
->minor_version
) {
2527 case 0: /* metadata at end. Round down and subtract space to reserve */
2528 devsize
= (devsize
& ~(4ULL*2-1));
2529 /* space for metadata, bblog, bitmap */
2530 devsize
-= 8*2 + 8 + bmspace
;
2534 devsize
-= data_offset
;
2537 *freesize
= devsize
;
2540 #endif /* MDASSEMBLE */
2542 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2544 /* Create a v1.0 superblock based on 'info'*/
2546 struct mdp_superblock_1
*sb
;
2549 unsigned long long offset
;
2551 if (posix_memalign(&ret
, 4096, 1024) != 0)
2554 memset(ret
, 0, 1024);
2555 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2556 sb
->major_version
= __cpu_to_le32(1);
2558 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2559 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2560 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2561 sb
->level
= __cpu_to_le32(info
->array
.level
);
2562 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2563 sb
->size
= __cpu_to_le64(info
->component_size
);
2564 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2565 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2566 if (info
->array
.level
> 0)
2567 sb
->data_size
= sb
->size
;
2569 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2570 sb
->resync_offset
= MaxSector
;
2571 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2572 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2573 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2575 offset
= st
->devsize
/512 - 8*2;
2577 sb
->super_offset
= __cpu_to_le64(offset
);
2578 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2580 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
2581 read(rfd
, sb
->device_uuid
, 16) != 16) {
2582 __u32 r
[4] = {random(), random(), random(), random()};
2583 memcpy(sb
->device_uuid
, r
, 16);
2588 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2589 int state
= sb0
->disks
[i
].state
;
2590 sb
->dev_roles
[i
] = MD_DISK_ROLE_SPARE
;
2591 if ((state
& (1<<MD_DISK_SYNC
)) &&
2592 !(state
& (1<<MD_DISK_FAULTY
)))
2593 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2595 sb
->sb_csum
= calc_sb_1_csum(sb
);
2599 struct superswitch super1
= {
2601 .examine_super
= examine_super1
,
2602 .brief_examine_super
= brief_examine_super1
,
2603 .export_examine_super
= export_examine_super1
,
2604 .detail_super
= detail_super1
,
2605 .brief_detail_super
= brief_detail_super1
,
2606 .export_detail_super
= export_detail_super1
,
2607 .write_init_super
= write_init_super1
,
2608 .validate_geometry
= validate_geometry1
,
2609 .add_to_super
= add_to_super1
,
2610 .examine_badblocks
= examine_badblocks_super1
,
2611 .copy_metadata
= copy_metadata1
,
2613 .match_home
= match_home1
,
2614 .uuid_from_super
= uuid_from_super1
,
2615 .getinfo_super
= getinfo_super1
,
2616 .container_content
= container_content1
,
2617 .update_super
= update_super1
,
2618 .init_super
= init_super1
,
2619 .store_super
= store_super1
,
2620 .compare_super
= compare_super1
,
2621 .load_super
= load_super1
,
2622 .match_metadata_desc
= match_metadata_desc1
,
2623 .avail_size
= avail_size1
,
2624 .add_internal_bitmap
= add_internal_bitmap1
,
2625 .locate_bitmap
= locate_bitmap1
,
2626 .write_bitmap
= write_bitmap1
,
2627 .free_super
= free_super1
,
2628 #if __BYTE_ORDER == BIG_ENDIAN