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>
27 * The version-1 superblock :
28 * All numeric fields are little-endian.
30 * total size: 256 bytes plus 2 per device.
31 * 1K allows 384 devices.
33 struct mdp_superblock_1
{
34 /* constant array information - 128 bytes */
35 __u32 magic
; /* MD_SB_MAGIC: 0xa92b4efc - little endian */
36 __u32 major_version
; /* 1 */
37 __u32 feature_map
; /* 0 for now */
38 __u32 pad0
; /* always set to 0 when writing */
40 __u8 set_uuid
[16]; /* user-space generated. */
41 char set_name
[32]; /* set and interpreted by user-space */
43 __u64 ctime
; /* lo 40 bits are seconds, top 24 are microseconds or 0*/
44 __u32 level
; /* -4 (multipath), -1 (linear), 0,1,4,5 */
45 __u32 layout
; /* only for raid5 currently */
46 __u64 size
; /* used size of component devices, in 512byte sectors */
48 __u32 chunksize
; /* in 512byte sectors */
50 __u32 bitmap_offset
; /* sectors after start of superblock that bitmap starts
51 * NOTE: signed, so bitmap can be before superblock
52 * only meaningful of feature_map[0] is set.
55 /* These are only valid with feature bit '4' */
56 __u32 new_level
; /* new level we are reshaping to */
57 __u64 reshape_position
; /* next address in array-space for reshape */
58 __u32 delta_disks
; /* change in number of raid_disks */
59 __u32 new_layout
; /* new layout */
60 __u32 new_chunk
; /* new chunk size (bytes) */
61 __u32 new_offset
; /* signed number to add to data_offset in new
62 * layout. 0 == no-change. This can be
63 * different on each device in the array.
66 /* constant this-device information - 64 bytes */
67 __u64 data_offset
; /* sector start of data, often 0 */
68 __u64 data_size
; /* sectors in this device that can be used for data */
69 __u64 super_offset
; /* sector start of this superblock */
70 __u64 recovery_offset
;/* sectors before this offset (from data_offset) have been recovered */
71 __u32 dev_number
; /* permanent identifier of this device - not role in raid */
72 __u32 cnt_corrected_read
; /* number of read errors that were corrected by re-writing */
73 __u8 device_uuid
[16]; /* user-space setable, ignored by kernel */
74 __u8 devflags
; /* per-device flags. Only one defined...*/
75 #define WriteMostly1 1 /* mask for writemostly flag in above */
76 /* bad block log. If there are any bad blocks the feature flag is set.
77 * if offset and size are non-zero, that space is reserved and available.
79 __u8 bblog_shift
; /* shift from sectors to block size for badblocklist */
80 __u16 bblog_size
; /* number of sectors reserved for badblocklist */
81 __u32 bblog_offset
; /* sector offset from superblock to bblog, signed */
83 /* array state information - 64 bytes */
84 __u64 utime
; /* 40 bits second, 24 btes microseconds */
85 __u64 events
; /* incremented when superblock updated */
86 __u64 resync_offset
; /* data before this offset (from data_offset) known to be in sync */
87 __u32 sb_csum
; /* checksum upto dev_roles[max_dev] */
88 __u32 max_dev
; /* size of dev_roles[] array to consider */
89 __u8 pad3
[64-32]; /* set to 0 when writing */
91 /* device state information. Indexed by dev_number.
93 * Note there are no per-device state flags. State information is rolled
94 * into the 'roles' value. If a device is spare or faulty, then it doesn't
95 * have a meaningful role.
97 __u16 dev_roles
[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
100 #define MAX_SB_SIZE 4096
101 /* bitmap super size is 256, but we round up to a sector for alignment */
102 #define BM_SUPER_SIZE 512
103 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
104 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
105 + sizeof(struct misc_dev_info))
107 struct misc_dev_info
{
111 /* feature_map bits */
112 #define MD_FEATURE_BITMAP_OFFSET 1
113 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
116 #define MD_FEATURE_RESHAPE_ACTIVE 4
117 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
118 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
119 * active device with same 'role'.
120 * 'recovery_offset' is also set.
122 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
123 * of devices, but is going
126 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
127 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
128 |MD_FEATURE_RECOVERY_OFFSET \
129 |MD_FEATURE_RESHAPE_ACTIVE \
130 |MD_FEATURE_BAD_BLOCKS \
131 |MD_FEATURE_REPLACEMENT \
132 |MD_FEATURE_RESHAPE_BACKWARDS \
133 |MD_FEATURE_NEW_OFFSET \
137 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
139 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
141 unsigned int disk_csum
, csum
;
142 unsigned long long newcsum
;
143 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
144 unsigned int *isuper
= (unsigned int*)sb
;
146 /* make sure I can count... */
147 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
148 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
149 sizeof(struct mdp_superblock_1
) != 256) {
150 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
153 disk_csum
= sb
->sb_csum
;
156 for (; size
>=4; size
-= 4 ) {
157 newcsum
+= __le32_to_cpu(*isuper
);
162 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
164 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
165 sb
->sb_csum
= disk_csum
;
166 return __cpu_to_le32(csum
);
170 * Information related to file descriptor used for aligned reads/writes.
171 * Cache the block size.
178 static void init_afd(struct align_fd
*afd
, int fd
)
182 if (ioctl(afd
->fd
, BLKSSZGET
, &afd
->blk_sz
) != 0)
186 static char abuf
[4096+4096];
187 static int aread(struct align_fd
*afd
, void *buf
, int len
)
190 * On devices with a 4K sector size, we need to read
191 * the full sector and copy relevant bits into
200 if (!bsize
|| bsize
> 4096 || len
> 4096) {
202 fprintf(stderr
, "WARNING - aread() called with "
203 "invalid block size\n");
206 b
= ROUND_UP_PTR((char *)abuf
, 4096);
208 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
210 n
= read(afd
->fd
, b
, iosize
);
213 lseek(afd
->fd
, len
- n
, 1);
220 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
223 * On devices with a 4K sector size, we need to write
224 * the full sector. We pre-read if the sector is larger
226 * The address must be sector-aligned.
233 if (!bsize
|| bsize
> 4096 || len
> 4096) {
235 fprintf(stderr
, "WARNING - awrite() called with "
236 "invalid block size\n");
239 b
= ROUND_UP_PTR((char *)abuf
, 4096);
241 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
245 n
= read(afd
->fd
, b
, iosize
);
248 lseek(afd
->fd
, -n
, 1);
252 n
= write(afd
->fd
, b
, iosize
);
255 lseek(afd
->fd
, len
- n
, 1);
260 static void examine_super1(struct supertype
*st
, char *homehost
)
262 struct mdp_superblock_1
*sb
= st
->sb
;
269 int l
= homehost
? strlen(homehost
) : 0;
271 unsigned long long sb_offset
;
273 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
274 printf(" Version : 1");
275 sb_offset
= __le64_to_cpu(sb
->super_offset
);
278 else if (sb_offset
<= 8)
282 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
283 printf(" Array UUID : ");
284 for (i
=0; i
<16; i
++) {
285 if ((i
&3)==0 && i
!= 0) printf(":");
286 printf("%02x", sb
->set_uuid
[i
]);
289 printf(" Name : %.32s", sb
->set_name
);
290 if (l
> 0 && l
< 32 &&
291 sb
->set_name
[l
] == ':' &&
292 strncmp(sb
->set_name
, homehost
, l
) == 0)
293 printf(" (local to host %s)", homehost
);
295 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
296 printf(" Creation Time : %.24s\n", ctime(&atime
));
297 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
298 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
299 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
301 printf(" Avail Dev Size : %llu%s\n",
302 (unsigned long long)__le64_to_cpu(sb
->data_size
),
303 human_size(__le64_to_cpu(sb
->data_size
)<<9));
304 if (__le32_to_cpu(sb
->level
) > 0) {
305 int ddsks
= 0, ddsks_denom
= 1;
306 switch(__le32_to_cpu(sb
->level
)) {
307 case 1: ddsks
=1;break;
309 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
310 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
312 layout
= __le32_to_cpu(sb
->layout
);
313 ddsks
= __le32_to_cpu(sb
->raid_disks
);
314 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
317 long long asize
= __le64_to_cpu(sb
->size
);
318 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
319 printf(" Array Size : %llu%s\n",
320 asize
>> 10, human_size(asize
));
322 if (sb
->size
!= sb
->data_size
)
323 printf(" Used Dev Size : %llu%s\n",
324 (unsigned long long)__le64_to_cpu(sb
->size
),
325 human_size(__le64_to_cpu(sb
->size
)<<9));
328 printf(" Data Offset : %llu sectors\n",
329 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
330 if (sb
->new_offset
) {
331 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
332 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
333 printf(" New Offset : %llu sectors\n", offset
);
335 printf(" Super Offset : %llu sectors\n",
336 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
337 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
338 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
339 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
340 printf(" Device UUID : ");
341 for (i
=0; i
<16; i
++) {
342 if ((i
&3)==0 && i
!= 0) printf(":");
343 printf("%02x", sb
->device_uuid
[i
]);
347 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
348 printf("Internal Bitmap : %ld sectors from superblock\n",
349 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
351 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
352 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
353 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
354 if (__le32_to_cpu(sb
->delta_disks
)) {
355 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
356 printf(" (%d->%d)\n",
357 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
358 __le32_to_cpu(sb
->raid_disks
));
359 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
360 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
362 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
363 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
364 printf(" New Level : %s\n", c
?c
:"-unknown-");
366 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
367 if (__le32_to_cpu(sb
->level
) == 5) {
368 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
369 printf(" New Layout : %s\n", c
?c
:"-unknown-");
371 if (__le32_to_cpu(sb
->level
) == 6) {
372 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
373 printf(" New Layout : %s\n", c
?c
:"-unknown-");
375 if (__le32_to_cpu(sb
->level
) == 10) {
376 printf(" New Layout :");
377 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
381 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
382 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
387 if (sb
->devflags
& WriteMostly1
)
388 printf(" write-mostly");
392 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
393 printf(" Update Time : %.24s\n", ctime(&atime
));
395 if (sb
->bblog_size
&& sb
->bblog_offset
) {
396 printf(" Bad Block Log : %d entries available at offset %ld sectors",
397 __le16_to_cpu(sb
->bblog_size
)*512/8,
398 (long)__le32_to_cpu(sb
->bblog_offset
));
399 if (sb
->feature_map
&
400 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
401 printf(" - bad blocks present.");
406 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
407 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
409 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
410 __le32_to_cpu(calc_sb_1_csum(sb
)));
411 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
413 if (__le32_to_cpu(sb
->level
) == 5) {
414 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
415 printf(" Layout : %s\n", c
?c
:"-unknown-");
417 if (__le32_to_cpu(sb
->level
) == 6) {
418 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
419 printf(" Layout : %s\n", c
?c
:"-unknown-");
421 if (__le32_to_cpu(sb
->level
) == 10) {
422 int lo
= __le32_to_cpu(sb
->layout
);
424 print_r10_layout(lo
);
427 switch(__le32_to_cpu(sb
->level
)) {
433 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
436 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
442 /* This turns out to just be confusing */
443 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
444 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
445 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != 0xffff)
447 for (d
=0; d
< i
; d
++) {
448 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
450 if (role
== 0xffff) printf("empty");
451 else if(role
== 0xfffe) printf("failed");
452 else printf("%d", role
);
456 printf(" Device Role : ");
457 d
= __le32_to_cpu(sb
->dev_number
);
458 if (d
< __le32_to_cpu(sb
->max_dev
))
459 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
464 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
465 printf("Replacement device %d\n", role
);
467 printf("Active device %d\n", role
);
469 printf(" Array State : ");
470 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
473 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
474 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
488 /* This is confusing too */
490 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
491 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
495 if (faulty
) printf(" %d failed", faulty
);
497 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
502 static void brief_examine_super1(struct supertype
*st
, int verbose
)
504 struct mdp_superblock_1
*sb
= st
->sb
;
506 unsigned long long sb_offset
;
508 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
510 nm
= strchr(sb
->set_name
, ':');
513 else if (sb
->set_name
[0])
525 printf(" level=%s", c
);
526 sb_offset
= __le64_to_cpu(sb
->super_offset
);
528 printf(" metadata=1.1 ");
529 else if (sb_offset
<= 8)
530 printf(" metadata=1.2 ");
532 printf(" metadata=1.0 ");
534 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
536 for (i
=0; i
<16; i
++) {
537 if ((i
&3)==0 && i
!= 0) printf(":");
538 printf("%02x", sb
->set_uuid
[i
]);
540 if (sb
->set_name
[0]) {
542 print_quoted(sb
->set_name
);
547 static void export_examine_super1(struct supertype
*st
)
549 struct mdp_superblock_1
*sb
= st
->sb
;
554 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
555 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
557 if (sb
->set_name
[i
] == '\n' ||
558 sb
->set_name
[i
] == '\0') {
563 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
564 if (__le32_to_cpu(sb
->level
) > 0) {
565 int ddsks
= 0, ddsks_denom
= 1;
566 switch(__le32_to_cpu(sb
->level
)) {
567 case 1: ddsks
=1;break;
569 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
570 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
572 layout
= __le32_to_cpu(sb
->layout
);
573 ddsks
= __le32_to_cpu(sb
->raid_disks
);
574 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
577 long long asize
= __le64_to_cpu(sb
->size
);
578 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
579 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
583 for (i
=0; i
<16; i
++) {
584 if ((i
&3)==0 && i
!= 0) printf(":");
585 printf("%02x", sb
->set_uuid
[i
]);
588 printf("MD_UPDATE_TIME=%llu\n",
589 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
590 printf("MD_DEV_UUID=");
591 for (i
=0; i
<16; i
++) {
592 if ((i
&3)==0 && i
!= 0) printf(":");
593 printf("%02x", sb
->device_uuid
[i
]);
596 printf("MD_EVENTS=%llu\n",
597 (unsigned long long)__le64_to_cpu(sb
->events
));
600 static void detail_super1(struct supertype
*st
, char *homehost
)
602 struct mdp_superblock_1
*sb
= st
->sb
;
604 int l
= homehost
? strlen(homehost
) : 0;
606 printf(" Name : %.32s", sb
->set_name
);
607 if (l
> 0 && l
< 32 &&
608 sb
->set_name
[l
] == ':' &&
609 strncmp(sb
->set_name
, homehost
, l
) == 0)
610 printf(" (local to host %s)", homehost
);
611 printf("\n UUID : ");
612 for (i
=0; i
<16; i
++) {
613 if ((i
&3)==0 && i
!= 0) printf(":");
614 printf("%02x", sb
->set_uuid
[i
]);
616 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb
->events
));
619 static void brief_detail_super1(struct supertype
*st
)
621 struct mdp_superblock_1
*sb
= st
->sb
;
624 if (sb
->set_name
[0]) {
626 print_quoted(sb
->set_name
);
629 for (i
=0; i
<16; i
++) {
630 if ((i
&3)==0 && i
!= 0) printf(":");
631 printf("%02x", sb
->set_uuid
[i
]);
635 static void export_detail_super1(struct supertype
*st
)
637 struct mdp_superblock_1
*sb
= st
->sb
;
642 if (sb
->set_name
[i
] == '\n' ||
643 sb
->set_name
[i
] == '\0') {
648 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
651 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
653 struct mdp_superblock_1
*sb
= st
->sb
;
654 unsigned long long offset
;
659 if (!sb
->bblog_size
|| __le32_to_cpu(sb
->bblog_size
) > 100
660 || !sb
->bblog_offset
){
661 printf("No bad-blocks list configured on %s\n", devname
);
664 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
666 printf("Bad-blocks list is empty in %s\n", devname
);
670 size
= __le32_to_cpu(sb
->bblog_size
)* 512;
671 posix_memalign((void**)&bbl
, 4096, size
);
672 offset
= __le64_to_cpu(sb
->super_offset
) +
673 (int)__le32_to_cpu(sb
->bblog_offset
);
675 if (lseek64(fd
, offset
, 0) < 0) {
676 pr_err("Cannot seek to bad-blocks list\n");
679 if (read(fd
, bbl
, size
) != size
) {
680 pr_err("Cannot read bad-blocks list\n");
683 /* 64bits per entry. 10 bits is block-count, 54 bits is block
684 * offset. Blocks are sectors unless bblog->shift makes them bigger
687 printf("Bad-blocks on %s:\n", devname
);
688 for (i
= 0; i
< size
/8; i
++, bbp
++) {
689 __u64 bb
= __le64_to_cpu(*bbp
);
690 int count
= bb
& 0x3ff;
691 unsigned long long sector
= bb
>> 10;
696 sector
<<= sb
->bblog_shift
;
697 count
<<= sb
->bblog_shift
;
699 printf("%20llu for %d sectors\n", sector
, count
);
706 static int match_home1(struct supertype
*st
, char *homehost
)
708 struct mdp_superblock_1
*sb
= st
->sb
;
709 int l
= homehost
? strlen(homehost
) : 0;
711 return (l
> 0 && l
< 32 &&
712 sb
->set_name
[l
] == ':' &&
713 strncmp(sb
->set_name
, homehost
, l
) == 0);
716 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
718 struct mdp_superblock_1
*super
= st
->sb
;
719 char *cuuid
= (char*)uuid
;
722 cuuid
[i
] = super
->set_uuid
[i
];
725 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
727 struct mdp_superblock_1
*sb
= st
->sb
;
728 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
729 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
733 unsigned int map_disks
= info
->array
.raid_disks
;
734 unsigned long long super_offset
;
735 unsigned long long data_size
;
737 memset(info
, 0, sizeof(*info
));
738 info
->array
.major_version
= 1;
739 info
->array
.minor_version
= st
->minor_version
;
740 info
->array
.patch_version
= 0;
741 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
742 info
->array
.level
= __le32_to_cpu(sb
->level
);
743 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
744 info
->array
.md_minor
= -1;
745 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
746 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
747 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
749 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
752 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
753 info
->component_size
= __le64_to_cpu(sb
->size
);
754 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
755 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
757 info
->disk
.major
= 0;
758 info
->disk
.minor
= 0;
759 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
760 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
761 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
764 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
766 super_offset
= __le64_to_cpu(sb
->super_offset
);
767 data_size
= __le64_to_cpu(sb
->size
);
768 if (info
->data_offset
< super_offset
) {
769 unsigned long long end
;
770 info
->space_before
= info
->data_offset
;
772 if (info
->bitmap_offset
< 0)
773 end
+= info
->bitmap_offset
;
774 if (info
->data_offset
+ data_size
< end
)
775 info
->space_after
= end
- data_size
- info
->data_offset
;
777 info
->space_after
= 0;
779 info
->space_before
= (info
->data_offset
-
781 if (info
->bitmap_offset
> 0) {
782 unsigned long long bmend
= info
->bitmap_offset
;
783 unsigned long long size
= __le64_to_cpu(bsb
->sync_size
);
784 size
/= __le32_to_cpu(bsb
->chunksize
) >> 9;
785 size
= (size
+ 7) >> 3;
786 size
+= sizeof(bitmap_super_t
);
787 size
= ROUND_UP(size
, 4096);
790 if (size
< info
->space_before
)
791 info
->space_before
-= size
;
793 info
->space_before
= 0;
795 info
->space_before
-= 8; /* superblock */
796 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
799 info
->disk
.raid_disk
= -1;
802 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
805 info
->disk
.state
= 1; /* faulty */
808 info
->disk
.state
= 6; /* active and in sync */
809 info
->disk
.raid_disk
= role
;
811 if (sb
->devflags
& WriteMostly1
)
812 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
813 info
->events
= __le64_to_cpu(sb
->events
);
814 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
815 info
->safe_mode_delay
= 200;
817 memcpy(info
->uuid
, sb
->set_uuid
, 16);
819 strncpy(info
->name
, sb
->set_name
, 32);
822 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
823 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
824 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
828 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
829 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
831 info
->recovery_start
= MaxSector
;
833 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
834 info
->reshape_active
= 1;
835 if (info
->array
.level
== 10)
836 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
837 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
838 info
->new_level
= __le32_to_cpu(sb
->new_level
);
839 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
840 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
841 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
842 if (info
->delta_disks
< 0)
843 info
->array
.raid_disks
-= info
->delta_disks
;
845 info
->reshape_active
= 0;
847 info
->recovery_blocked
= info
->reshape_active
;
850 for (i
=0; i
<map_disks
; i
++)
852 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
853 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
854 if (/*role == 0xFFFF || */role
< (unsigned) info
->array
.raid_disks
) {
856 if (map
&& role
< map_disks
)
861 info
->array
.working_disks
= working
;
864 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
871 info
= xmalloc(sizeof(*info
));
872 getinfo_super1(st
, info
, NULL
);
876 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
878 char *devname
, int verbose
,
879 int uuid_set
, char *homehost
)
881 /* NOTE: for 'assemble' and 'force' we need to return non-zero
882 * if any change was made. For others, the return value is
886 struct mdp_superblock_1
*sb
= st
->sb
;
888 if (strcmp(update
, "force-one")==0) {
889 /* Not enough devices for a working array,
890 * so bring this one up-to-date
892 if (sb
->events
!= __cpu_to_le64(info
->events
))
894 sb
->events
= __cpu_to_le64(info
->events
);
895 } else if (strcmp(update
, "force-array")==0) {
896 /* Degraded array and 'force' requests to
897 * maybe need to mark it 'clean'.
899 switch(__le32_to_cpu(sb
->level
)) {
900 case 5: case 4: case 6:
901 /* need to force clean */
902 if (sb
->resync_offset
!= MaxSector
)
904 sb
->resync_offset
= MaxSector
;
906 } else if (strcmp(update
, "assemble")==0) {
907 int d
= info
->disk
.number
;
909 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
910 want
= info
->disk
.raid_disk
;
913 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
914 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
917 if (info
->reshape_active
&&
918 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
919 info
->delta_disks
>= 0 &&
920 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
921 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
924 if (info
->reshape_active
&&
925 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
926 info
->delta_disks
< 0 &&
927 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
928 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
931 } else if (strcmp(update
, "linear-grow-new") == 0) {
934 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
936 for (i
=0 ; i
< max
; i
++)
937 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= 0xfffe)
939 sb
->dev_number
= __cpu_to_le32(i
);
940 info
->disk
.number
= i
;
941 if (max
>= __le32_to_cpu(sb
->max_dev
))
942 sb
->max_dev
= __cpu_to_le32(max
+1);
944 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
945 read(rfd
, sb
->device_uuid
, 16) != 16) {
946 __u32 r
[4] = {random(), random(), random(), random()};
947 memcpy(sb
->device_uuid
, r
, 16);
953 __cpu_to_le16(info
->disk
.raid_disk
);
955 fd
= open(devname
, O_RDONLY
);
957 unsigned long long ds
;
958 get_dev_size(fd
, devname
, &ds
);
961 if (__le64_to_cpu(sb
->super_offset
) <
962 __le64_to_cpu(sb
->data_offset
)) {
963 sb
->data_size
= __cpu_to_le64(
964 ds
- __le64_to_cpu(sb
->data_offset
));
967 ds
&= ~(unsigned long long)(4*2-1);
968 sb
->super_offset
= __cpu_to_le64(ds
);
969 sb
->data_size
= __cpu_to_le64(
970 ds
- __le64_to_cpu(sb
->data_offset
));
973 } else if (strcmp(update
, "linear-grow-update") == 0) {
974 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
975 sb
->dev_roles
[info
->disk
.number
] =
976 __cpu_to_le16(info
->disk
.raid_disk
);
977 } else if (strcmp(update
, "resync") == 0) {
978 /* make sure resync happens */
979 sb
->resync_offset
= 0ULL;
980 } else if (strcmp(update
, "uuid") == 0) {
981 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
983 if (__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
984 struct bitmap_super_s
*bm
;
985 bm
= (struct bitmap_super_s
*)(st
->sb
+MAX_SB_SIZE
);
986 memcpy(bm
->uuid
, sb
->set_uuid
, 16);
988 } else if (strcmp(update
, "no-bitmap") == 0) {
989 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
990 } else if (strcmp(update
, "bbl") == 0) {
991 /* only possible if there is room after the bitmap, or if
994 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
995 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
996 long bitmap_offset
= (long)__le64_to_cpu(sb
->bitmap_offset
);
1000 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1001 struct bitmap_super_s
*bsb
;
1002 bsb
= (struct bitmap_super_s
*)(((char*)sb
)+MAX_SB_SIZE
);
1003 bm_sectors
= bitmap_sectors(bsb
);
1006 if (sb_offset
< data_offset
) {
1007 /* 1.1 or 1.2. Put bbl just before data
1010 space
= data_offset
- sb_offset
;
1011 bb_offset
= space
- 8;
1012 if (bm_sectors
&& bitmap_offset
> 0)
1013 space
-= (bitmap_offset
+ bm_sectors
);
1015 space
-= 8; /* The superblock */
1017 sb
->bblog_size
= __cpu_to_le16(8);
1018 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1021 /* 1.0 - Put bbl just before super block */
1022 if (bm_sectors
&& bitmap_offset
< 0)
1023 space
= -bitmap_offset
- bm_sectors
;
1025 space
= sb_offset
- data_offset
-
1026 __le64_to_cpu(sb
->data_size
);
1028 sb
->bblog_size
= __cpu_to_le16(8);
1029 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1032 } else if (strcmp(update
, "no-bbl") == 0) {
1033 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1034 pr_err("Cannot remove active bbl from %s\n",devname
);
1037 sb
->bblog_shift
= 0;
1038 sb
->bblog_offset
= 0;
1040 } else if (strcmp(update
, "homehost") == 0 &&
1044 c
= strchr(sb
->set_name
, ':');
1046 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1048 strncpy(info
->name
, sb
->set_name
, 32);
1050 } else if (strcmp(update
, "name") == 0) {
1051 if (info
->name
[0] == 0)
1052 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1053 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1055 strchr(info
->name
, ':') == NULL
&&
1056 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1057 strcpy(sb
->set_name
, homehost
);
1058 strcat(sb
->set_name
, ":");
1059 strcat(sb
->set_name
, info
->name
);
1061 strcpy(sb
->set_name
, info
->name
);
1062 } else if (strcmp(update
, "devicesize") == 0 &&
1063 __le64_to_cpu(sb
->super_offset
) <
1064 __le64_to_cpu(sb
->data_offset
)) {
1065 /* set data_size to device size less data_offset */
1066 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1067 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1068 printf("Size was %llu\n", (unsigned long long)
1069 __le64_to_cpu(sb
->data_size
));
1070 sb
->data_size
= __cpu_to_le64(
1071 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1072 printf("Size is %llu\n", (unsigned long long)
1073 __le64_to_cpu(sb
->data_size
));
1074 } else if (strcmp(update
, "_reshape_progress")==0)
1075 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1076 else if (strcmp(update
, "writemostly")==0)
1077 sb
->devflags
|= WriteMostly1
;
1078 else if (strcmp(update
, "readwrite")==0)
1079 sb
->devflags
&= ~WriteMostly1
;
1083 sb
->sb_csum
= calc_sb_1_csum(sb
);
1087 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1088 unsigned long long size
, char *name
, char *homehost
,
1089 int *uuid
, unsigned long long data_offset
)
1091 struct mdp_superblock_1
*sb
;
1097 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1098 pr_err("%s could not allocate superblock\n", __func__
);
1101 memset(sb
, 0, SUPER1_SIZE
);
1105 /* zeroing superblock */
1109 spares
= info
->working_disks
- info
->active_disks
;
1110 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1111 pr_err("too many devices requested: %d+%d > %d\n",
1112 info
->raid_disks
, spares
, MAX_DEVS
);
1116 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1117 sb
->major_version
= __cpu_to_le32(1);
1118 sb
->feature_map
= 0;
1122 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1124 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1125 read(rfd
, sb
->set_uuid
, 16) != 16) {
1126 __u32 r
[4] = {random(), random(), random(), random()};
1127 memcpy(sb
->set_uuid
, r
, 16);
1129 if (rfd
>= 0) close(rfd
);
1132 if (name
== NULL
|| *name
== 0) {
1133 sprintf(defname
, "%d", info
->md_minor
);
1137 strchr(name
, ':')== NULL
&&
1138 strlen(homehost
)+1+strlen(name
) < 32) {
1139 strcpy(sb
->set_name
, homehost
);
1140 strcat(sb
->set_name
, ":");
1141 strcat(sb
->set_name
, name
);
1143 strcpy(sb
->set_name
, name
);
1145 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1146 sb
->level
= __cpu_to_le32(info
->level
);
1147 sb
->layout
= __cpu_to_le32(info
->layout
);
1148 sb
->size
= __cpu_to_le64(size
*2ULL);
1149 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1150 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1152 sb
->data_offset
= __cpu_to_le64(data_offset
);
1153 sb
->data_size
= __cpu_to_le64(0);
1154 sb
->super_offset
= __cpu_to_le64(0);
1155 sb
->recovery_offset
= __cpu_to_le64(0);
1157 sb
->utime
= sb
->ctime
;
1158 sb
->events
= __cpu_to_le64(1);
1159 if (info
->state
& (1<<MD_SB_CLEAN
))
1160 sb
->resync_offset
= MaxSector
;
1162 sb
->resync_offset
= 0;
1163 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1164 sbsize
= ROUND_UP(sbsize
, 512);
1165 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1167 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1175 long long data_offset
;
1176 mdu_disk_info_t disk
;
1177 struct devinfo
*next
;
1180 /* Add a device to the superblock being created */
1181 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1182 int fd
, char *devname
, unsigned long long data_offset
)
1184 struct mdp_superblock_1
*sb
= st
->sb
;
1185 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1186 struct devinfo
*di
, **dip
;
1188 if ((dk
->state
& 6) == 6) /* active, sync */
1189 *rp
= __cpu_to_le16(dk
->raid_disk
);
1190 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1195 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1196 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1197 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1199 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1200 sb
->devflags
= 0; /* don't copy another disks flags */
1201 sb
->sb_csum
= calc_sb_1_csum(sb
);
1203 dip
= (struct devinfo
**)&st
->info
;
1205 dip
= &(*dip
)->next
;
1206 di
= xmalloc(sizeof(struct devinfo
));
1208 di
->devname
= devname
;
1210 di
->data_offset
= data_offset
;
1218 static void locate_bitmap1(struct supertype
*st
, int fd
);
1220 static int store_super1(struct supertype
*st
, int fd
)
1222 struct mdp_superblock_1
*sb
= st
->sb
;
1223 unsigned long long sb_offset
;
1224 struct align_fd afd
;
1226 unsigned long long dsize
;
1228 if (!get_dev_size(fd
, NULL
, &dsize
))
1239 * Calculate the position of the superblock.
1240 * It is always aligned to a 4K boundary and
1241 * depending on minor_version, it can be:
1242 * 0: At least 8K, but less than 12K, from end of device
1243 * 1: At start of device
1244 * 2: 4K from start of device.
1246 switch(st
->minor_version
) {
1250 sb_offset
&= ~(4*2-1);
1264 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1265 0 != __le64_to_cpu(sb
->super_offset
)
1267 pr_err("internal error - sb_offset is wrong\n");
1271 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1274 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1276 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1279 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1280 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1281 (((char*)sb
)+MAX_SB_SIZE
);
1282 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1283 locate_bitmap1(st
, fd
);
1284 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1292 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1294 static unsigned long choose_bm_space(unsigned long devsize
)
1296 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1297 * if bigger than 200Gig, save 128k
1298 * NOTE: result must be multiple of 4K else bad things happen
1299 * on 4K-sector devices.
1301 if (devsize
< 64*2) return 0;
1302 if (devsize
- 64*2 >= 200*1024*1024*2)
1304 if (devsize
- 4*2 > 8*1024*1024*2)
1309 static void free_super1(struct supertype
*st
);
1312 static int write_init_super1(struct supertype
*st
)
1314 struct mdp_superblock_1
*sb
= st
->sb
;
1315 struct supertype
*refst
;
1318 unsigned long long bm_space
;
1319 unsigned long long reserved
;
1321 unsigned long long dsize
, array_size
;
1322 unsigned long long sb_offset
, headroom
;
1323 unsigned long long data_offset
;
1325 for (di
= st
->info
; di
; di
= di
->next
) {
1326 if (di
->disk
.state
== 1)
1331 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1334 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1335 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1336 sb
->devflags
|= WriteMostly1
;
1338 sb
->devflags
&= ~WriteMostly1
;
1340 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1341 read(rfd
, sb
->device_uuid
, 16) != 16) {
1342 __u32 r
[4] = {random(), random(), random(), random()};
1343 memcpy(sb
->device_uuid
, r
, 16);
1350 refst
= dup_super(st
);
1351 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1352 struct mdp_superblock_1
*refsb
= refst
->sb
;
1354 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1355 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1356 /* same array, so preserve events and
1358 sb
->events
= refsb
->events
;
1359 /* bugs in 2.6.17 and earlier mean the
1360 * dev_number chosen in Manage must be preserved
1362 if (get_linux_version() >= 2006018)
1363 sb
->dev_number
= refsb
->dev_number
;
1369 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1383 * Calculate the position of the superblock.
1384 * It is always aligned to a 4K boundary and
1385 * depending on minor_version, it can be:
1386 * 0: At least 8K, but less than 12K, from end of device
1387 * 1: At start of device
1388 * 2: 4K from start of device.
1389 * Depending on the array size, we might leave extra space
1391 * Also leave 4K for bad-block log.
1393 array_size
= __le64_to_cpu(sb
->size
);
1394 /* work out how much space we left for a bitmap,
1395 * Add 8 sectors for bad block log */
1396 bm_space
= choose_bm_space(array_size
) + 8;
1398 /* We try to leave 0.1% at the start for reshape
1399 * operations, but limit this to 128Meg (0.1% of 10Gig)
1400 * which is plenty for efficient reshapes
1401 * However we make it at least 2 chunks as one chunk
1402 * is minimum needed for reshape.
1404 headroom
= 128 * 1024 * 2;
1405 while (headroom
<< 10 > array_size
&&
1406 headroom
/2 >= __le32_to_cpu(sb
->chunksize
) * 2)
1409 data_offset
= di
->data_offset
;
1410 switch(st
->minor_version
) {
1414 sb_offset
&= ~(4*2-1);
1415 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1416 if (data_offset
== INVALID_SECTORS
)
1417 sb
->data_offset
= 0;
1418 if (sb_offset
< array_size
+ bm_space
)
1419 bm_space
= sb_offset
- array_size
;
1420 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1421 if (bm_space
>= 8) {
1422 sb
->bblog_size
= __cpu_to_le16(8);
1423 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1427 sb
->super_offset
= __cpu_to_le64(0);
1428 if (data_offset
== INVALID_SECTORS
) {
1429 reserved
= bm_space
+ 4*2;
1430 if (reserved
< headroom
)
1431 reserved
= headroom
;
1432 if (reserved
+ array_size
> dsize
)
1433 reserved
= dsize
- array_size
;
1434 /* Try for multiple of 1Meg so it is nicely aligned */
1435 #define ONE_MEG (2*1024)
1436 if (reserved
> ONE_MEG
)
1437 reserved
= (reserved
/ONE_MEG
) * ONE_MEG
;
1439 /* force 4K alignment */
1443 reserved
= data_offset
;
1445 sb
->data_offset
= __cpu_to_le64(reserved
);
1446 sb
->data_size
= __cpu_to_le64(dsize
- reserved
);
1447 if (reserved
>= 16) {
1448 sb
->bblog_size
= __cpu_to_le16(8);
1449 sb
->bblog_offset
= __cpu_to_le32(reserved
-8);
1454 sb
->super_offset
= __cpu_to_le64(4*2);
1455 if (data_offset
== INVALID_SECTORS
) {
1456 if (4*2 + 4*2 + bm_space
+ array_size
1458 bm_space
= dsize
- array_size
1461 reserved
= bm_space
+ 4*2 + 4*2;
1462 if (reserved
< headroom
)
1463 reserved
= headroom
;
1464 if (reserved
+ array_size
> dsize
)
1465 reserved
= dsize
- array_size
;
1466 /* Try for multiple of 1Meg so it is nicely aligned */
1467 #define ONE_MEG (2*1024)
1468 if (reserved
> ONE_MEG
)
1469 reserved
= (reserved
/ONE_MEG
) * ONE_MEG
;
1471 /* force 4K alignment */
1475 reserved
= data_offset
;
1477 sb
->data_offset
= __cpu_to_le64(reserved
);
1478 sb
->data_size
= __cpu_to_le64(dsize
- reserved
);
1479 if (reserved
>= 16+16) {
1480 sb
->bblog_size
= __cpu_to_le16(8);
1481 /* '8' sectors for the bblog, and another '8'
1482 * because we want offset from superblock, not
1485 sb
->bblog_offset
= __cpu_to_le32(reserved
-8-8);
1489 pr_err("Failed to write invalid "
1490 "metadata format 1.%i to %s\n",
1491 st
->minor_version
, di
->devname
);
1496 sb
->sb_csum
= calc_sb_1_csum(sb
);
1497 rv
= store_super1(st
, di
->fd
);
1498 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1499 rv
= st
->ss
->write_bitmap(st
, di
->fd
);
1507 pr_err("Failed to write metadata to %s\n",
1514 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1518 * 0 same, or first was empty, and second was copied
1519 * 1 second had wrong number
1521 * 3 wrong other info
1523 struct mdp_superblock_1
*first
= st
->sb
;
1524 struct mdp_superblock_1
*second
= tst
->sb
;
1526 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1528 if (second
->major_version
!= __cpu_to_le32(1))
1532 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1533 pr_err("%s could not allocate superblock\n", __func__
);
1536 memcpy(first
, second
, SUPER1_SIZE
);
1540 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1543 if (first
->ctime
!= second
->ctime
||
1544 first
->level
!= second
->level
||
1545 first
->layout
!= second
->layout
||
1546 first
->size
!= second
->size
||
1547 first
->chunksize
!= second
->chunksize
||
1548 first
->raid_disks
!= second
->raid_disks
)
1553 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1555 unsigned long long dsize
;
1556 unsigned long long sb_offset
;
1557 struct mdp_superblock_1
*super
;
1559 struct bitmap_super_s
*bsb
;
1560 struct misc_dev_info
*misc
;
1561 struct align_fd afd
;
1567 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1569 struct supertype tst
;
1570 __u64 bestctime
= 0;
1571 /* guess... choose latest ctime */
1572 memset(&tst
, 0, sizeof(tst
));
1574 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1575 switch(load_super1(&tst
, fd
, devname
)) {
1576 case 0: super
= tst
.sb
;
1577 if (bestvers
== -1 ||
1578 bestctime
< __le64_to_cpu(super
->ctime
)) {
1579 bestvers
= tst
.minor_version
;
1580 bestctime
= __le64_to_cpu(super
->ctime
);
1585 case 1: return 1; /*bad device */
1586 case 2: break; /* bad, try next */
1589 if (bestvers
!= -1) {
1591 tst
.minor_version
= bestvers
;
1593 tst
.max_devs
= MAX_DEVS
;
1594 rv
= load_super1(&tst
, fd
, devname
);
1601 if (!get_dev_size(fd
, devname
, &dsize
))
1607 pr_err("%s is too small for md: size is %llu sectors.\n",
1613 * Calculate the position of the superblock.
1614 * It is always aligned to a 4K boundary and
1615 * depending on minor_version, it can be:
1616 * 0: At least 8K, but less than 12K, from end of device
1617 * 1: At start of device
1618 * 2: 4K from start of device.
1620 switch(st
->minor_version
) {
1624 sb_offset
&= ~(4*2-1);
1636 ioctl(fd
, BLKFLSBUF
, 0); /* make sure we read current data */
1639 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1641 pr_err("Cannot seek to superblock on %s: %s\n",
1642 devname
, strerror(errno
));
1646 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
1647 pr_err("%s could not allocate superblock\n",
1652 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
1654 pr_err("Cannot read superblock on %s\n",
1660 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
1662 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
1663 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
1668 if (__le32_to_cpu(super
->major_version
) != 1) {
1670 pr_err("Cannot interpret superblock on %s - version is %d\n",
1671 devname
, __le32_to_cpu(super
->major_version
));
1675 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
1677 pr_err("No superblock found on %s (super_offset is wrong)\n",
1684 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
1686 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
1687 misc
->device_size
= dsize
;
1689 /* Now check on the bitmap superblock */
1690 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
1692 /* Read the bitmap superblock and make sure it looks
1693 * valid. If it doesn't clear the bit. An --assemble --force
1694 * should get that written out.
1696 locate_bitmap1(st
, fd
);
1697 if (aread(&afd
, bsb
, 512) != 512)
1700 uuid_from_super1(st
, uuid
);
1701 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
1702 memcmp(bsb
->uuid
, uuid
, 16) != 0)
1707 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
1708 & ~MD_FEATURE_BITMAP_OFFSET
);
1713 static struct supertype
*match_metadata_desc1(char *arg
)
1715 struct supertype
*st
= xcalloc(1, sizeof(*st
));
1717 st
->container_dev
= NoMdDev
;
1719 st
->max_devs
= MAX_DEVS
;
1721 /* leading zeros can be safely ignored. --detail generates them. */
1724 if (strcmp(arg
, "1.0") == 0 ||
1725 strcmp(arg
, "1.00") == 0) {
1726 st
->minor_version
= 0;
1729 if (strcmp(arg
, "1.1") == 0 ||
1730 strcmp(arg
, "1.01") == 0
1732 st
->minor_version
= 1;
1735 if (strcmp(arg
, "1.2") == 0 ||
1736 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
1737 strcmp(arg
, "default") == 0 ||
1738 #endif /* DEFAULT_OLD_METADATA */
1739 strcmp(arg
, "1.02") == 0) {
1740 st
->minor_version
= 2;
1743 if (strcmp(arg
, "1") == 0 ||
1744 strcmp(arg
, "default") == 0) {
1745 st
->minor_version
= -1;
1753 /* find available size on device with this devsize, using
1754 * superblock type st, and reserving 'reserve' sectors for
1757 static __u64
_avail_size1(struct supertype
*st
, __u64 devsize
,
1758 unsigned long long data_offset
, int chunksize
)
1760 struct mdp_superblock_1
*super
= st
->sb
;
1766 /* creating: allow suitable space for bitmap */
1767 bmspace
= choose_bm_space(devsize
);
1769 else if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1770 /* hot-add. allow for actual size of bitmap */
1771 struct bitmap_super_s
*bsb
;
1772 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
1773 bmspace
= bitmap_sectors(bsb
);
1776 /* Allow space for bad block log */
1777 if (super
&& super
->bblog_size
)
1778 devsize
-= __le16_to_cpu(super
->bblog_size
);
1783 if (st
->minor_version
< 0)
1784 /* not specified, so time to set default */
1785 st
->minor_version
= 2;
1787 if (data_offset
!= INVALID_SECTORS
)
1788 switch(st
->minor_version
) {
1790 return devsize
- data_offset
- 8*2;
1793 return devsize
- data_offset
;
1800 if (super
== NULL
&& st
->minor_version
> 0) {
1801 /* haven't committed to a size yet, so allow some
1802 * slack for space for reshape.
1803 * Limit slack to 128M, but aim for about 0.1%
1805 unsigned long long headroom
= 128*1024*2;
1806 while ((headroom
<< 10) > devsize
&&
1808 headroom
/ 2 >= ((unsigned)chunksize
*2)*2))
1810 devsize
-= headroom
;
1812 switch(st
->minor_version
) {
1815 return ((devsize
- 8*2 ) & ~(4*2-1));
1817 /* at start, 4K for superblock and possible bitmap */
1818 return devsize
- 4*2;
1820 /* 4k from start, 4K for superblock and possible bitmap */
1821 return devsize
- (4+4)*2;
1825 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
1826 unsigned long long data_offset
)
1828 return _avail_size1(st
, devsize
, data_offset
, 0);
1832 add_internal_bitmap1(struct supertype
*st
,
1833 int *chunkp
, int delay
, int write_behind
,
1834 unsigned long long size
,
1835 int may_change
, int major
)
1838 * If not may_change, then this is a 'Grow' without sysfs support for
1839 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
1840 * If may_change, then this is create or a Grow with sysfs syupport,
1841 * and we can put the bitmap wherever we like.
1843 * size is in sectors, chunk is in bytes !!!
1846 unsigned long long bits
;
1847 unsigned long long max_bits
;
1848 unsigned long long min_chunk
;
1850 long bbl_offset
, bbl_size
;
1851 unsigned long long chunk
= *chunkp
;
1854 struct mdp_superblock_1
*sb
= st
->sb
;
1855 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1859 if (__le64_to_cpu(sb
->data_size
) == 0)
1860 /* Must be creating the array, else data_size would be non-zero */
1862 switch(st
->minor_version
) {
1864 /* either 3K after the superblock (when hot-add),
1865 * or some amount of space before.
1868 /* We are creating array, so we *know* how much room has
1872 room
= choose_bm_space(__le64_to_cpu(sb
->size
));
1875 room
= __le64_to_cpu(sb
->super_offset
)
1876 - __le64_to_cpu(sb
->data_offset
)
1877 - __le64_to_cpu(sb
->data_size
);
1878 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
1881 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
1882 if (bbl_size
< -bbl_offset
)
1883 bbl_size
= -bbl_offset
;
1885 if (!may_change
|| (room
< 3*2 &&
1886 __le32_to_cpu(sb
->max_dev
) <= 384)) {
1891 offset
= 0; /* means movable offset */
1896 case 2: /* between superblock and data */
1899 room
= choose_bm_space(__le64_to_cpu(sb
->size
));
1902 room
= __le64_to_cpu(sb
->data_offset
)
1903 - __le64_to_cpu(sb
->super_offset
);
1904 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
1906 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
1911 room
-= 2; /* Leave 1K for superblock */
1915 room
-= 4*2; /* leave 4K for superblock */
1925 if (chunk
== UnSet
&& room
> 128*2)
1926 /* Limit to 128K of bitmap when chunk size not requested */
1929 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
1931 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
1932 bits
= (size
*512)/min_chunk
+1;
1933 while (bits
> max_bits
) {
1937 if (chunk
== UnSet
) {
1938 /* For practical purpose, 64Meg is a good
1939 * default chunk size for internal bitmaps.
1942 if (chunk
< 64*1024*1024)
1943 chunk
= 64*1024*1024;
1944 } else if (chunk
< min_chunk
)
1945 return 0; /* chunk size too small */
1946 if (chunk
== 0) /* rounding problem */
1950 /* start bitmap on a 4K boundary with enough space for
1953 bits
= (size
*512) / chunk
+ 1;
1954 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
1955 room
*= 8; /* convert 4K blocks to sectors */
1956 offset
= -room
- bbl_size
;
1959 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
1961 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
1962 | MD_FEATURE_BITMAP_OFFSET
);
1963 memset(bms
, 0, sizeof(*bms
));
1964 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
1965 bms
->version
= __cpu_to_le32(major
);
1966 uuid_from_super1(st
, uuid
);
1967 memcpy(bms
->uuid
, uuid
, 16);
1968 bms
->chunksize
= __cpu_to_le32(chunk
);
1969 bms
->daemon_sleep
= __cpu_to_le32(delay
);
1970 bms
->sync_size
= __cpu_to_le64(size
);
1971 bms
->write_behind
= __cpu_to_le32(write_behind
);
1977 static void locate_bitmap1(struct supertype
*st
, int fd
)
1979 unsigned long long offset
;
1980 struct mdp_superblock_1
*sb
;
1984 if (st
->ss
->load_super(st
, fd
, NULL
))
1985 return; /* no error I hope... */
1990 offset
= __le64_to_cpu(sb
->super_offset
);
1991 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
);
1994 lseek64(fd
, offset
<<9, 0);
1997 static int write_bitmap1(struct supertype
*st
, int fd
)
1999 struct mdp_superblock_1
*sb
= st
->sb
;
2000 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2004 struct align_fd afd
;
2008 locate_bitmap1(st
, fd
);
2010 if (posix_memalign(&buf
, 4096, 4096))
2013 memset(buf
, 0xff, 4096);
2014 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2016 towrite
= __le64_to_cpu(bms
->sync_size
) / (__le32_to_cpu(bms
->chunksize
)>>9);
2017 towrite
= (towrite
+7) >> 3; /* bits to bytes */
2018 towrite
+= sizeof(bitmap_super_t
);
2019 towrite
= ROUND_UP(towrite
, 512);
2020 while (towrite
> 0) {
2024 n
= awrite(&afd
, buf
, n
);
2029 memset(buf
, 0xff, 4096);
2039 static void free_super1(struct supertype
*st
)
2044 struct devinfo
*di
= st
->info
;
2045 st
->info
= di
->next
;
2054 static int validate_geometry1(struct supertype
*st
, int level
,
2055 int layout
, int raiddisks
,
2056 int *chunk
, unsigned long long size
,
2057 unsigned long long data_offset
,
2058 char *subdev
, unsigned long long *freesize
,
2061 unsigned long long ldsize
;
2064 if (level
== LEVEL_CONTAINER
) {
2066 pr_err("1.x metadata does not support containers\n");
2069 if (chunk
&& *chunk
== UnSet
)
2070 *chunk
= DEFAULT_CHUNK
;
2075 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2078 pr_err("super1.x cannot open %s: %s\n",
2079 subdev
, strerror(errno
));
2083 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2089 *freesize
= _avail_size1(st
, ldsize
>> 9, data_offset
, *chunk
);
2092 #endif /* MDASSEMBLE */
2094 struct superswitch super1
= {
2096 .examine_super
= examine_super1
,
2097 .brief_examine_super
= brief_examine_super1
,
2098 .export_examine_super
= export_examine_super1
,
2099 .detail_super
= detail_super1
,
2100 .brief_detail_super
= brief_detail_super1
,
2101 .export_detail_super
= export_detail_super1
,
2102 .write_init_super
= write_init_super1
,
2103 .validate_geometry
= validate_geometry1
,
2104 .add_to_super
= add_to_super1
,
2105 .examine_badblocks
= examine_badblocks_super1
,
2107 .match_home
= match_home1
,
2108 .uuid_from_super
= uuid_from_super1
,
2109 .getinfo_super
= getinfo_super1
,
2110 .container_content
= container_content1
,
2111 .update_super
= update_super1
,
2112 .init_super
= init_super1
,
2113 .store_super
= store_super1
,
2114 .compare_super
= compare_super1
,
2115 .load_super
= load_super1
,
2116 .match_metadata_desc
= match_metadata_desc1
,
2117 .avail_size
= avail_size1
,
2118 .add_internal_bitmap
= add_internal_bitmap1
,
2119 .locate_bitmap
= locate_bitmap1
,
2120 .write_bitmap
= write_bitmap1
,
2121 .free_super
= free_super1
,
2122 #if __BYTE_ORDER == BIG_ENDIAN