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
);
327 if (bms
->nodes
> 0 && (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
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
);
783 if (bms
->nodes
> 0 && (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
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
;
1103 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1105 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1106 rv
= cluster_get_dlmlock(&lockid
);
1108 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1109 cluster_release_dlmlock(lockid
);
1114 if (strcmp(update
, "homehost") == 0 &&
1116 /* Note that 'homehost' is special as it is really
1121 c
= strchr(sb
->set_name
, ':');
1123 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1125 strncpy(info
->name
, sb
->set_name
, 32);
1129 if (strcmp(update
, "force-one")==0) {
1130 /* Not enough devices for a working array,
1131 * so bring this one up-to-date
1133 if (sb
->events
!= __cpu_to_le64(info
->events
))
1135 sb
->events
= __cpu_to_le64(info
->events
);
1136 } else if (strcmp(update
, "force-array")==0) {
1137 /* Degraded array and 'force' requests to
1138 * maybe need to mark it 'clean'.
1140 switch(__le32_to_cpu(sb
->level
)) {
1141 case 5: case 4: case 6:
1142 /* need to force clean */
1143 if (sb
->resync_offset
!= MaxSector
)
1145 sb
->resync_offset
= MaxSector
;
1147 } else if (strcmp(update
, "assemble")==0) {
1148 int d
= info
->disk
.number
;
1150 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
1151 want
= info
->disk
.raid_disk
;
1152 else if (info
->disk
.state
& (1<<MD_DISK_JOURNAL
))
1153 want
= MD_DISK_ROLE_JOURNAL
;
1155 want
= MD_DISK_ROLE_SPARE
;
1156 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
1157 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
1160 if (info
->reshape_active
&&
1161 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1162 info
->delta_disks
>= 0 &&
1163 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
1164 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1167 if (info
->reshape_active
&&
1168 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1169 info
->delta_disks
< 0 &&
1170 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
1171 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1174 } else if (strcmp(update
, "linear-grow-new") == 0) {
1177 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
1179 for (i
=0 ; i
< max
; i
++)
1180 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= MD_DISK_ROLE_FAULTY
)
1182 sb
->dev_number
= __cpu_to_le32(i
);
1183 info
->disk
.number
= i
;
1184 if (max
>= __le32_to_cpu(sb
->max_dev
))
1185 sb
->max_dev
= __cpu_to_le32(max
+1);
1187 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1188 read(rfd
, sb
->device_uuid
, 16) != 16) {
1189 __u32 r
[4] = {random(), random(), random(), random()};
1190 memcpy(sb
->device_uuid
, r
, 16);
1196 __cpu_to_le16(info
->disk
.raid_disk
);
1198 fd
= open(devname
, O_RDONLY
);
1200 unsigned long long ds
;
1201 get_dev_size(fd
, devname
, &ds
);
1204 if (__le64_to_cpu(sb
->super_offset
) <
1205 __le64_to_cpu(sb
->data_offset
)) {
1206 sb
->data_size
= __cpu_to_le64(
1207 ds
- __le64_to_cpu(sb
->data_offset
));
1210 ds
&= ~(unsigned long long)(4*2-1);
1211 sb
->super_offset
= __cpu_to_le64(ds
);
1212 sb
->data_size
= __cpu_to_le64(
1213 ds
- __le64_to_cpu(sb
->data_offset
));
1216 } else if (strcmp(update
, "linear-grow-update") == 0) {
1217 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1218 sb
->dev_roles
[info
->disk
.number
] =
1219 __cpu_to_le16(info
->disk
.raid_disk
);
1220 } else if (strcmp(update
, "resync") == 0) {
1221 /* make sure resync happens */
1222 sb
->resync_offset
= 0ULL;
1223 } else if (strcmp(update
, "uuid") == 0) {
1224 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1226 if (__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1227 struct bitmap_super_s
*bm
;
1228 bm
= (struct bitmap_super_s
*)(st
->sb
+MAX_SB_SIZE
);
1229 memcpy(bm
->uuid
, sb
->set_uuid
, 16);
1231 } else if (strcmp(update
, "no-bitmap") == 0) {
1232 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1233 } else if (strcmp(update
, "bbl") == 0) {
1234 /* only possible if there is room after the bitmap, or if
1235 * there is no bitmap
1237 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1238 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1239 long bitmap_offset
= (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
);
1240 long bm_sectors
= 0;
1244 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1245 struct bitmap_super_s
*bsb
;
1246 bsb
= (struct bitmap_super_s
*)(((char*)sb
)+MAX_SB_SIZE
);
1247 bm_sectors
= bitmap_sectors(bsb
);
1250 if (sb_offset
< data_offset
) {
1251 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1254 bb_offset
= sb_offset
+ 8;
1255 if (bm_sectors
&& bitmap_offset
> 0)
1256 bb_offset
= bitmap_offset
+ bm_sectors
;
1257 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1258 && bb_offset
+ 8+8 <= (long)data_offset
)
1259 /* too close to bitmap, and room to grow */
1261 if (bb_offset
+ 8 <= (long)data_offset
) {
1262 sb
->bblog_size
= __cpu_to_le16(8);
1263 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1266 /* 1.0 - Put bbl just before super block */
1267 if (bm_sectors
&& bitmap_offset
< 0)
1268 space
= -bitmap_offset
- bm_sectors
;
1270 space
= sb_offset
- data_offset
-
1271 __le64_to_cpu(sb
->data_size
);
1273 sb
->bblog_size
= __cpu_to_le16(8);
1274 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1277 } else if (strcmp(update
, "no-bbl") == 0) {
1278 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1279 pr_err("Cannot remove active bbl from %s\n",devname
);
1282 sb
->bblog_shift
= 0;
1283 sb
->bblog_offset
= 0;
1285 } else if (strcmp(update
, "force-no-bbl") == 0) {
1286 sb
->feature_map
&= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
);
1288 sb
->bblog_shift
= 0;
1289 sb
->bblog_offset
= 0;
1290 } else if (strcmp(update
, "name") == 0) {
1291 if (info
->name
[0] == 0)
1292 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1293 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1295 strchr(info
->name
, ':') == NULL
&&
1296 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1297 strcpy(sb
->set_name
, homehost
);
1298 strcat(sb
->set_name
, ":");
1299 strcat(sb
->set_name
, info
->name
);
1301 strcpy(sb
->set_name
, info
->name
);
1302 } else if (strcmp(update
, "devicesize") == 0 &&
1303 __le64_to_cpu(sb
->super_offset
) <
1304 __le64_to_cpu(sb
->data_offset
)) {
1305 /* set data_size to device size less data_offset */
1306 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1307 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1308 sb
->data_size
= __cpu_to_le64(
1309 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1310 } else if (strncmp(update
, "revert-reshape", 14) == 0) {
1312 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1313 pr_err("No active reshape to revert on %s\n",
1317 unsigned long long reshape_sectors
;
1320 /* If the reshape hasn't started, just stop it.
1321 * It is conceivable that a stripe was modified but
1322 * the metadata not updated. In that case the backup
1323 * should have been used to get passed the critical stage.
1324 * If that couldn't happen, the "-nobackup" version
1327 if (strcmp(update
, "revert-reshape-nobackup") == 0 &&
1328 sb
->reshape_position
== 0 &&
1329 (__le32_to_cpu(sb
->delta_disks
) > 0 ||
1330 (__le32_to_cpu(sb
->delta_disks
) == 0 &&
1331 !(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
))))) {
1332 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
);
1333 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1334 __le32_to_cpu(sb
->delta_disks
));
1335 sb
->delta_disks
= 0;
1338 /* reshape_position is a little messy.
1339 * Its value must be a multiple of the larger
1340 * chunk size, and of the "after" data disks.
1341 * So when reverting we need to change it to
1342 * be a multiple of the new "after" data disks,
1343 * which is the old "before".
1344 * If it isn't already a multiple of 'before',
1345 * the only thing we could do would be
1346 * copy some block around on the disks, which
1347 * is easy to get wrong.
1348 * So we reject a revert-reshape unless the
1349 * alignment is good.
1351 if (__le32_to_cpu(sb
->level
) >= 4 &&
1352 __le32_to_cpu(sb
->level
) <= 6) {
1353 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1354 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1355 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1356 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1357 if (reshape_sectors
% reshape_chunk
) {
1358 pr_err("Reshape position is not suitably aligned.\n");
1359 pr_err("Try normal assembly and stop again\n");
1363 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1364 __le32_to_cpu(sb
->delta_disks
));
1365 if (sb
->delta_disks
== 0)
1366 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1368 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1370 temp
= sb
->new_layout
;
1371 sb
->new_layout
= sb
->layout
;
1374 temp
= sb
->new_chunk
;
1375 sb
->new_chunk
= sb
->chunksize
;
1376 sb
->chunksize
= temp
;
1378 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1379 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1380 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1381 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1382 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1386 } else if (strcmp(update
, "_reshape_progress")==0)
1387 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1388 else if (strcmp(update
, "writemostly")==0)
1389 sb
->devflags
|= WriteMostly1
;
1390 else if (strcmp(update
, "readwrite")==0)
1391 sb
->devflags
&= ~WriteMostly1
;
1395 sb
->sb_csum
= calc_sb_1_csum(sb
);
1396 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1397 cluster_release_dlmlock(lockid
);
1402 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1403 unsigned long long size
, char *name
, char *homehost
,
1404 int *uuid
, unsigned long long data_offset
)
1406 struct mdp_superblock_1
*sb
;
1412 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1413 pr_err("could not allocate superblock\n");
1416 memset(sb
, 0, SUPER1_SIZE
);
1420 /* zeroing superblock */
1424 spares
= info
->working_disks
- info
->active_disks
;
1425 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1426 pr_err("too many devices requested: %d+%d > %d\n",
1427 info
->raid_disks
, spares
, MAX_DEVS
);
1431 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1432 sb
->major_version
= __cpu_to_le32(1);
1433 sb
->feature_map
= 0;
1437 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1439 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1440 read(rfd
, sb
->set_uuid
, 16) != 16) {
1441 __u32 r
[4] = {random(), random(), random(), random()};
1442 memcpy(sb
->set_uuid
, r
, 16);
1444 if (rfd
>= 0) close(rfd
);
1447 if (name
== NULL
|| *name
== 0) {
1448 sprintf(defname
, "%d", info
->md_minor
);
1452 strchr(name
, ':')== NULL
&&
1453 strlen(homehost
)+1+strlen(name
) < 32) {
1454 strcpy(sb
->set_name
, homehost
);
1455 strcat(sb
->set_name
, ":");
1456 strcat(sb
->set_name
, name
);
1458 strcpy(sb
->set_name
, name
);
1460 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1461 sb
->level
= __cpu_to_le32(info
->level
);
1462 sb
->layout
= __cpu_to_le32(info
->layout
);
1463 sb
->size
= __cpu_to_le64(size
*2ULL);
1464 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1465 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1467 sb
->data_offset
= __cpu_to_le64(data_offset
);
1468 sb
->data_size
= __cpu_to_le64(0);
1469 sb
->super_offset
= __cpu_to_le64(0);
1470 sb
->recovery_offset
= __cpu_to_le64(0);
1472 sb
->utime
= sb
->ctime
;
1473 sb
->events
= __cpu_to_le64(1);
1474 if (info
->state
& (1<<MD_SB_CLEAN
))
1475 sb
->resync_offset
= MaxSector
;
1477 sb
->resync_offset
= 0;
1478 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1479 sbsize
= ROUND_UP(sbsize
, 512);
1480 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1482 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1490 long long data_offset
;
1491 mdu_disk_info_t disk
;
1492 struct devinfo
*next
;
1495 /* Add a device to the superblock being created */
1496 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1497 int fd
, char *devname
, unsigned long long data_offset
)
1499 struct mdp_superblock_1
*sb
= st
->sb
;
1500 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1501 struct devinfo
*di
, **dip
;
1502 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1505 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1506 rv
= cluster_get_dlmlock(&lockid
);
1508 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1509 cluster_release_dlmlock(lockid
);
1514 if ((dk
->state
& 6) == 6) /* active, sync */
1515 *rp
= __cpu_to_le16(dk
->raid_disk
);
1516 else if (dk
->state
& (1<<MD_DISK_JOURNAL
))
1517 *rp
= MD_DISK_ROLE_JOURNAL
;
1518 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1519 *rp
= MD_DISK_ROLE_SPARE
;
1521 *rp
= MD_DISK_ROLE_FAULTY
;
1523 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1524 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1525 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1527 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1528 sb
->devflags
= 0; /* don't copy another disks flags */
1529 sb
->sb_csum
= calc_sb_1_csum(sb
);
1531 dip
= (struct devinfo
**)&st
->info
;
1533 dip
= &(*dip
)->next
;
1534 di
= xmalloc(sizeof(struct devinfo
));
1536 di
->devname
= devname
;
1538 di
->data_offset
= data_offset
;
1542 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1543 cluster_release_dlmlock(lockid
);
1549 static int locate_bitmap1(struct supertype
*st
, int fd
);
1551 static int store_super1(struct supertype
*st
, int fd
)
1553 struct mdp_superblock_1
*sb
= st
->sb
;
1554 unsigned long long sb_offset
;
1555 struct align_fd afd
;
1557 unsigned long long dsize
;
1558 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
1561 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready()) {
1562 rv
= cluster_get_dlmlock(&lockid
);
1564 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1565 cluster_release_dlmlock(lockid
);
1570 if (!get_dev_size(fd
, NULL
, &dsize
))
1581 * Calculate the position of the superblock.
1582 * It is always aligned to a 4K boundary and
1583 * depending on minor_version, it can be:
1584 * 0: At least 8K, but less than 12K, from end of device
1585 * 1: At start of device
1586 * 2: 4K from start of device.
1588 switch(st
->minor_version
) {
1592 sb_offset
&= ~(4*2-1);
1604 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1605 0 != __le64_to_cpu(sb
->super_offset
)
1607 pr_err("internal error - sb_offset is wrong\n");
1611 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1614 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1616 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1619 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1620 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1621 (((char*)sb
)+MAX_SB_SIZE
);
1622 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1623 locate_bitmap1(st
, fd
);
1624 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1629 if (bms
->version
== BITMAP_MAJOR_CLUSTERED
&& dlm_funs_ready())
1630 cluster_release_dlmlock(lockid
);
1635 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1637 static unsigned long choose_bm_space(unsigned long devsize
)
1639 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1640 * if bigger than 200Gig, save 128k
1641 * NOTE: result must be multiple of 4K else bad things happen
1642 * on 4K-sector devices.
1644 if (devsize
< 64*2) return 0;
1645 if (devsize
- 64*2 >= 200*1024*1024*2)
1647 if (devsize
- 4*2 > 8*1024*1024*2)
1652 static void free_super1(struct supertype
*st
);
1654 #define META_BLOCK_SIZE 4096
1655 __u32
crc32c_le(__u32 crc
, unsigned char const *p
, size_t len
);
1657 static int write_empty_r5l_meta_block(struct supertype
*st
, int fd
)
1659 struct r5l_meta_block
*mb
;
1660 struct mdp_superblock_1
*sb
= st
->sb
;
1661 struct align_fd afd
;
1666 if (posix_memalign((void**)&mb
, 4096, META_BLOCK_SIZE
) != 0) {
1667 pr_err("Could not allocate memory for the meta block.\n");
1671 memset(mb
, 0, META_BLOCK_SIZE
);
1673 mb
->magic
= __cpu_to_le32(R5LOG_MAGIC
);
1674 mb
->version
= R5LOG_VERSION
;
1675 mb
->meta_size
= __cpu_to_le32(sizeof(struct r5l_meta_block
));
1676 mb
->seq
= __cpu_to_le64(random32());
1677 mb
->position
= __cpu_to_le64(0);
1679 crc
= crc32c_le(0xffffffff, sb
->set_uuid
, sizeof(sb
->set_uuid
));
1680 crc
= crc32c_le(crc
, (void *)mb
, META_BLOCK_SIZE
);
1683 if (lseek64(fd
, (sb
->data_offset
) * 512, 0) < 0LL) {
1684 pr_err("cannot seek to offset of the meta block\n");
1688 if (awrite(&afd
, mb
, META_BLOCK_SIZE
) != META_BLOCK_SIZE
) {
1689 pr_err("failed to store write the meta block \n");
1703 static int write_init_super1(struct supertype
*st
)
1705 struct mdp_superblock_1
*sb
= st
->sb
;
1706 struct supertype
*refst
;
1709 unsigned long long bm_space
;
1711 unsigned long long dsize
, array_size
;
1712 unsigned long long sb_offset
;
1713 unsigned long long data_offset
;
1715 for (di
= st
->info
; di
; di
= di
->next
) {
1716 if (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))
1717 sb
->feature_map
|= MD_FEATURE_JOURNAL
;
1720 for (di
= st
->info
; di
; di
= di
->next
) {
1721 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1726 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1729 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1730 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1731 sb
->devflags
|= WriteMostly1
;
1733 sb
->devflags
&= ~WriteMostly1
;
1735 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1736 read(rfd
, sb
->device_uuid
, 16) != 16) {
1737 __u32 r
[4] = {random(), random(), random(), random()};
1738 memcpy(sb
->device_uuid
, r
, 16);
1743 if (!(di
->disk
.state
& (1<<MD_DISK_JOURNAL
)))
1746 refst
= dup_super(st
);
1747 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1748 struct mdp_superblock_1
*refsb
= refst
->sb
;
1750 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1751 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1752 /* same array, so preserve events and
1754 sb
->events
= refsb
->events
;
1755 /* bugs in 2.6.17 and earlier mean the
1756 * dev_number chosen in Manage must be preserved
1758 if (get_linux_version() >= 2006018)
1759 sb
->dev_number
= refsb
->dev_number
;
1765 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1778 * Calculate the position of the superblock.
1779 * It is always aligned to a 4K boundary and
1780 * depending on minor_version, it can be:
1781 * 0: At least 8K, but less than 12K, from end of device
1782 * 1: At start of device
1783 * 2: 4K from start of device.
1784 * data_offset has already been set.
1786 array_size
= __le64_to_cpu(sb
->size
);
1787 /* work out how much space we left for a bitmap,
1788 * Add 8 sectors for bad block log */
1789 bm_space
= choose_bm_space(array_size
) + 8;
1791 data_offset
= di
->data_offset
;
1792 if (data_offset
== INVALID_SECTORS
)
1793 data_offset
= st
->data_offset
;
1794 switch(st
->minor_version
) {
1796 if (data_offset
== INVALID_SECTORS
)
1800 sb_offset
&= ~(4*2-1);
1801 sb
->data_offset
= __cpu_to_le64(data_offset
);
1802 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1803 if (sb_offset
< array_size
+ bm_space
)
1804 bm_space
= sb_offset
- array_size
;
1805 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1806 if (bm_space
>= 8) {
1807 sb
->bblog_size
= __cpu_to_le16(8);
1808 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1812 sb
->super_offset
= __cpu_to_le64(0);
1813 if (data_offset
== INVALID_SECTORS
)
1816 sb
->data_offset
= __cpu_to_le64(data_offset
);
1817 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1818 if (data_offset
>= 8 + 32*2 + 8) {
1819 sb
->bblog_size
= __cpu_to_le16(8);
1820 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1821 } else if (data_offset
>= 16) {
1822 sb
->bblog_size
= __cpu_to_le16(8);
1823 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8);
1828 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1829 if (data_offset
== INVALID_SECTORS
)
1832 sb
->data_offset
= __cpu_to_le64(data_offset
);
1833 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1834 if (data_offset
>= 16 + 32*2 + 8) {
1835 sb
->bblog_size
= __cpu_to_le16(8);
1836 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1837 } else if (data_offset
>= 16+16) {
1838 sb
->bblog_size
= __cpu_to_le16(8);
1839 /* '8' sectors for the bblog, and another '8'
1840 * because we want offset from superblock, not
1843 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8-8);
1847 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1848 st
->minor_version
, di
->devname
);
1852 /* Disable badblock log on clusters, or when explicitly requested */
1853 if (st
->nodes
> 0 || conf_get_create_info()->bblist
== 0) {
1855 sb
->bblog_offset
= 0;
1858 sb
->sb_csum
= calc_sb_1_csum(sb
);
1859 rv
= store_super1(st
, di
->fd
);
1861 if (rv
== 0 && (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))) {
1862 rv
= write_empty_r5l_meta_block(st
, di
->fd
);
1867 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1868 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NoUpdate
);
1876 pr_err("Failed to write metadata to %s\n",
1883 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1887 * 0 same, or first was empty, and second was copied
1888 * 1 second had wrong number
1890 * 3 wrong other info
1892 struct mdp_superblock_1
*first
= st
->sb
;
1893 struct mdp_superblock_1
*second
= tst
->sb
;
1895 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1897 if (second
->major_version
!= __cpu_to_le32(1))
1901 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1902 pr_err("could not allocate superblock\n");
1905 memcpy(first
, second
, SUPER1_SIZE
);
1909 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1912 if (first
->ctime
!= second
->ctime
||
1913 first
->level
!= second
->level
||
1914 first
->layout
!= second
->layout
||
1915 first
->size
!= second
->size
||
1916 first
->chunksize
!= second
->chunksize
||
1917 first
->raid_disks
!= second
->raid_disks
)
1922 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1924 unsigned long long dsize
;
1925 unsigned long long sb_offset
;
1926 struct mdp_superblock_1
*super
;
1928 struct bitmap_super_s
*bsb
;
1929 struct misc_dev_info
*misc
;
1930 struct align_fd afd
;
1936 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1938 struct supertype tst
;
1939 __u64 bestctime
= 0;
1940 /* guess... choose latest ctime */
1941 memset(&tst
, 0, sizeof(tst
));
1943 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1944 switch(load_super1(&tst
, fd
, devname
)) {
1945 case 0: super
= tst
.sb
;
1946 if (bestvers
== -1 ||
1947 bestctime
< __le64_to_cpu(super
->ctime
)) {
1948 bestvers
= tst
.minor_version
;
1949 bestctime
= __le64_to_cpu(super
->ctime
);
1954 case 1: return 1; /*bad device */
1955 case 2: break; /* bad, try next */
1958 if (bestvers
!= -1) {
1960 tst
.minor_version
= bestvers
;
1962 tst
.max_devs
= MAX_DEVS
;
1963 rv
= load_super1(&tst
, fd
, devname
);
1970 if (!get_dev_size(fd
, devname
, &dsize
))
1976 pr_err("%s is too small for md: size is %llu sectors.\n",
1982 * Calculate the position of the superblock.
1983 * It is always aligned to a 4K boundary and
1984 * depending on minor_version, it can be:
1985 * 0: At least 8K, but less than 12K, from end of device
1986 * 1: At start of device
1987 * 2: 4K from start of device.
1989 switch(st
->minor_version
) {
1993 sb_offset
&= ~(4*2-1);
2005 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
2007 pr_err("Cannot seek to superblock on %s: %s\n",
2008 devname
, strerror(errno
));
2012 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
2013 pr_err("could not allocate superblock\n");
2017 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
2019 pr_err("Cannot read superblock on %s\n",
2025 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
2027 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2028 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
2033 if (__le32_to_cpu(super
->major_version
) != 1) {
2035 pr_err("Cannot interpret superblock on %s - version is %d\n",
2036 devname
, __le32_to_cpu(super
->major_version
));
2040 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
2042 pr_err("No superblock found on %s (super_offset is wrong)\n",
2049 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2051 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
2052 misc
->device_size
= dsize
;
2053 if (st
->data_offset
== INVALID_SECTORS
)
2054 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
2056 /* Now check on the bitmap superblock */
2057 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
2059 /* Read the bitmap superblock and make sure it looks
2060 * valid. If it doesn't clear the bit. An --assemble --force
2061 * should get that written out.
2063 locate_bitmap1(st
, fd
);
2064 if (aread(&afd
, bsb
, 512) != 512)
2067 uuid_from_super1(st
, uuid
);
2068 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
2069 memcmp(bsb
->uuid
, uuid
, 16) != 0)
2074 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
2075 & ~MD_FEATURE_BITMAP_OFFSET
);
2079 static struct supertype
*match_metadata_desc1(char *arg
)
2081 struct supertype
*st
= xcalloc(1, sizeof(*st
));
2083 st
->container_devnm
[0] = 0;
2085 st
->max_devs
= MAX_DEVS
;
2087 st
->data_offset
= INVALID_SECTORS
;
2088 /* leading zeros can be safely ignored. --detail generates them. */
2091 if (strcmp(arg
, "1.0") == 0 ||
2092 strcmp(arg
, "1.00") == 0) {
2093 st
->minor_version
= 0;
2096 if (strcmp(arg
, "1.1") == 0 ||
2097 strcmp(arg
, "1.01") == 0
2099 st
->minor_version
= 1;
2102 if (strcmp(arg
, "1.2") == 0 ||
2103 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2104 strcmp(arg
, "default") == 0 ||
2105 #endif /* DEFAULT_OLD_METADATA */
2106 strcmp(arg
, "1.02") == 0) {
2107 st
->minor_version
= 2;
2110 if (strcmp(arg
, "1") == 0 ||
2111 strcmp(arg
, "default") == 0) {
2112 st
->minor_version
= -1;
2120 /* find available size on device with this devsize, using
2121 * superblock type st, and reserving 'reserve' sectors for
2124 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
2125 unsigned long long data_offset
)
2127 struct mdp_superblock_1
*super
= st
->sb
;
2134 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
2135 /* hot-add. allow for actual size of bitmap */
2136 struct bitmap_super_s
*bsb
;
2137 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2138 bmspace
= bitmap_sectors(bsb
);
2141 /* Allow space for bad block log */
2142 if (super
->bblog_size
)
2143 bbspace
= __le16_to_cpu(super
->bblog_size
);
2145 if (st
->minor_version
< 0)
2146 /* not specified, so time to set default */
2147 st
->minor_version
= 2;
2149 if (data_offset
== INVALID_SECTORS
)
2150 data_offset
= st
->data_offset
;
2152 if (data_offset
!= INVALID_SECTORS
)
2153 switch(st
->minor_version
) {
2155 return devsize
- data_offset
- 8*2 - bbspace
;
2158 return devsize
- data_offset
;
2165 switch(st
->minor_version
) {
2168 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2170 /* at start, 4K for superblock and possible bitmap */
2171 return devsize
- 4*2 - bbspace
;
2173 /* 4k from start, 4K for superblock and possible bitmap */
2174 return devsize
- (4+4)*2 - bbspace
;
2180 add_internal_bitmap1(struct supertype
*st
,
2181 int *chunkp
, int delay
, int write_behind
,
2182 unsigned long long size
,
2183 int may_change
, int major
)
2186 * If not may_change, then this is a 'Grow' without sysfs support for
2187 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2188 * If may_change, then this is create or a Grow with sysfs syupport,
2189 * and we can put the bitmap wherever we like.
2191 * size is in sectors, chunk is in bytes !!!
2194 unsigned long long bits
;
2195 unsigned long long max_bits
;
2196 unsigned long long min_chunk
;
2198 long bbl_offset
, bbl_size
;
2199 unsigned long long chunk
= *chunkp
;
2202 struct mdp_superblock_1
*sb
= st
->sb
;
2203 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2206 if (__le64_to_cpu(sb
->data_size
) == 0)
2207 /* Must be creating the array, else data_size would be non-zero */
2209 switch(st
->minor_version
) {
2211 /* either 3K after the superblock (when hot-add),
2212 * or some amount of space before.
2215 /* We are creating array, so we *know* how much room has
2220 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2222 room
= __le64_to_cpu(sb
->super_offset
)
2223 - __le64_to_cpu(sb
->data_offset
)
2224 - __le64_to_cpu(sb
->data_size
);
2225 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2228 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2229 if (bbl_size
< -bbl_offset
)
2230 bbl_size
= -bbl_offset
;
2232 if (!may_change
|| (room
< 3*2 &&
2233 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2238 offset
= 0; /* means movable offset */
2243 case 2: /* between superblock and data */
2247 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2249 room
= __le64_to_cpu(sb
->data_offset
)
2250 - __le64_to_cpu(sb
->super_offset
);
2251 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2253 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2258 room
-= 2; /* Leave 1K for superblock */
2262 room
-= 4*2; /* leave 4K for superblock */
2272 if (chunk
== UnSet
&& room
> 128*2)
2273 /* Limit to 128K of bitmap when chunk size not requested */
2277 /* No room for a bitmap */
2280 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2282 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2283 bits
= (size
*512)/min_chunk
+1;
2284 while (bits
> max_bits
) {
2288 if (chunk
== UnSet
) {
2289 /* For practical purpose, 64Meg is a good
2290 * default chunk size for internal bitmaps.
2293 if (chunk
< 64*1024*1024)
2294 chunk
= 64*1024*1024;
2295 } else if (chunk
< min_chunk
)
2296 return 0; /* chunk size too small */
2297 if (chunk
== 0) /* rounding problem */
2301 /* start bitmap on a 4K boundary with enough space for
2304 bits
= (size
*512) / chunk
+ 1;
2305 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2306 room
*= 8; /* convert 4K blocks to sectors */
2307 offset
= -room
- bbl_size
;
2310 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2312 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2313 | MD_FEATURE_BITMAP_OFFSET
);
2314 memset(bms
, 0, sizeof(*bms
));
2315 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2316 bms
->version
= __cpu_to_le32(major
);
2317 uuid_from_super1(st
, uuid
);
2318 memcpy(bms
->uuid
, uuid
, 16);
2319 bms
->chunksize
= __cpu_to_le32(chunk
);
2320 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2321 bms
->sync_size
= __cpu_to_le64(size
);
2322 bms
->write_behind
= __cpu_to_le32(write_behind
);
2323 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2325 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2326 | MD_FEATURE_BITMAP_VERSIONED
);
2327 if (st
->cluster_name
)
2328 strncpy((char *)bms
->cluster_name
,
2329 st
->cluster_name
, strlen(st
->cluster_name
));
2335 static int locate_bitmap1(struct supertype
*st
, int fd
)
2337 unsigned long long offset
;
2338 struct mdp_superblock_1
*sb
;
2343 if (st
->ss
->load_super(st
, fd
, NULL
))
2344 return -1; /* no error I hope... */
2349 if ((__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_BITMAP_OFFSET
))
2353 offset
= __le64_to_cpu(sb
->super_offset
);
2354 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
);
2357 lseek64(fd
, offset
<<9, 0);
2361 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2363 struct mdp_superblock_1
*sb
= st
->sb
;
2364 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2368 struct align_fd afd
;
2370 unsigned long long total_bm_space
, bm_space_per_node
;
2374 /* update cluster name */
2375 if (st
->cluster_name
) {
2376 memset((char *)bms
->cluster_name
, 0, sizeof(bms
->cluster_name
));
2377 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, 64);
2381 /* cluster md only supports superblock 1.2 now */
2382 if (st
->minor_version
!= 2) {
2383 pr_err("Warning: cluster md only works with superblock 1.2\n");
2387 /* Each node has an independent bitmap, it is necessary to calculate the
2388 * space is enough or not, first get how many bytes for the total bitmap */
2389 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2391 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2392 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2394 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2395 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2396 total_bm_space
/ bm_space_per_node
);
2400 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2409 locate_bitmap1(st
, fd
);
2411 if (posix_memalign(&buf
, 4096, 4096))
2415 /* Only the bitmap[0] should resync
2416 * whole device on initial assembly
2419 memset(buf
, 0x00, 4096);
2421 memset(buf
, 0xff, 4096);
2422 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2424 towrite
= calc_bitmap_size(bms
, 4096);
2425 while (towrite
> 0) {
2429 n
= awrite(&afd
, buf
, n
);
2435 memset(buf
, 0x00, 4096);
2437 memset(buf
, 0xff, 4096);
2444 } while (++i
< __le32_to_cpu(bms
->nodes
));
2450 static void free_super1(struct supertype
*st
)
2456 struct devinfo
*di
= st
->info
;
2457 st
->info
= di
->next
;
2466 static int validate_geometry1(struct supertype
*st
, int level
,
2467 int layout
, int raiddisks
,
2468 int *chunk
, unsigned long long size
,
2469 unsigned long long data_offset
,
2470 char *subdev
, unsigned long long *freesize
,
2473 unsigned long long ldsize
, devsize
;
2475 unsigned long long headroom
;
2478 if (level
== LEVEL_CONTAINER
) {
2480 pr_err("1.x metadata does not support containers\n");
2483 if (*chunk
== UnSet
)
2484 *chunk
= DEFAULT_CHUNK
;
2489 if (st
->minor_version
< 0)
2490 /* not specified, so time to set default */
2491 st
->minor_version
= 2;
2493 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2496 pr_err("super1.x cannot open %s: %s\n",
2497 subdev
, strerror(errno
));
2501 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2507 devsize
= ldsize
>> 9;
2513 /* creating: allow suitable space for bitmap */
2514 bmspace
= choose_bm_space(devsize
);
2516 if (data_offset
== INVALID_SECTORS
)
2517 data_offset
= st
->data_offset
;
2518 if (data_offset
== INVALID_SECTORS
)
2519 switch (st
->minor_version
) {
2525 /* Choose data offset appropriate for this device
2526 * and use as default for whole array.
2527 * The data_offset must allow for bitmap space
2528 * and base metadata, should allow for some headroom
2529 * for reshape, and should be rounded to multiple
2531 * Headroom is limited to 128M, but aim for about 0.1%
2533 headroom
= 128*1024*2;
2534 while ((headroom
<< 10) > devsize
&&
2536 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2538 data_offset
= 12*2 + bmspace
+ headroom
;
2539 #define ONE_MEG (2*1024)
2540 if (data_offset
> ONE_MEG
)
2541 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2544 if (st
->data_offset
== INVALID_SECTORS
)
2545 st
->data_offset
= data_offset
;
2546 switch(st
->minor_version
) {
2547 case 0: /* metadata at end. Round down and subtract space to reserve */
2548 devsize
= (devsize
& ~(4ULL*2-1));
2549 /* space for metadata, bblog, bitmap */
2550 devsize
-= 8*2 + 8 + bmspace
;
2554 devsize
-= data_offset
;
2557 *freesize
= devsize
;
2560 #endif /* MDASSEMBLE */
2562 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2564 /* Create a v1.0 superblock based on 'info'*/
2566 struct mdp_superblock_1
*sb
;
2569 unsigned long long offset
;
2571 if (posix_memalign(&ret
, 4096, 1024) != 0)
2574 memset(ret
, 0, 1024);
2575 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2576 sb
->major_version
= __cpu_to_le32(1);
2578 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2579 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2580 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2581 sb
->level
= __cpu_to_le32(info
->array
.level
);
2582 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2583 sb
->size
= __cpu_to_le64(info
->component_size
);
2584 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2585 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2586 if (info
->array
.level
> 0)
2587 sb
->data_size
= sb
->size
;
2589 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2590 sb
->resync_offset
= MaxSector
;
2591 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2592 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2593 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2595 offset
= st
->devsize
/512 - 8*2;
2597 sb
->super_offset
= __cpu_to_le64(offset
);
2598 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2600 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
2601 read(rfd
, sb
->device_uuid
, 16) != 16) {
2602 __u32 r
[4] = {random(), random(), random(), random()};
2603 memcpy(sb
->device_uuid
, r
, 16);
2608 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2609 int state
= sb0
->disks
[i
].state
;
2610 sb
->dev_roles
[i
] = MD_DISK_ROLE_SPARE
;
2611 if ((state
& (1<<MD_DISK_SYNC
)) &&
2612 !(state
& (1<<MD_DISK_FAULTY
)))
2613 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2615 sb
->sb_csum
= calc_sb_1_csum(sb
);
2619 struct superswitch super1
= {
2621 .examine_super
= examine_super1
,
2622 .brief_examine_super
= brief_examine_super1
,
2623 .export_examine_super
= export_examine_super1
,
2624 .detail_super
= detail_super1
,
2625 .brief_detail_super
= brief_detail_super1
,
2626 .export_detail_super
= export_detail_super1
,
2627 .write_init_super
= write_init_super1
,
2628 .validate_geometry
= validate_geometry1
,
2629 .add_to_super
= add_to_super1
,
2630 .examine_badblocks
= examine_badblocks_super1
,
2631 .copy_metadata
= copy_metadata1
,
2633 .match_home
= match_home1
,
2634 .uuid_from_super
= uuid_from_super1
,
2635 .getinfo_super
= getinfo_super1
,
2636 .container_content
= container_content1
,
2637 .update_super
= update_super1
,
2638 .init_super
= init_super1
,
2639 .store_super
= store_super1
,
2640 .compare_super
= compare_super1
,
2641 .load_super
= load_super1
,
2642 .match_metadata_desc
= match_metadata_desc1
,
2643 .avail_size
= avail_size1
,
2644 .add_internal_bitmap
= add_internal_bitmap1
,
2645 .locate_bitmap
= locate_bitmap1
,
2646 .write_bitmap
= write_bitmap1
,
2647 .free_super
= free_super1
,
2648 #if __BYTE_ORDER == BIG_ENDIAN