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 * 0, jouranl not required
145 * 1, journal required
146 * 2, no superblock loated (st->sb == NULL)
148 static int require_journal1(struct supertype
*st
)
150 struct mdp_superblock_1
*sb
= st
->sb
;
152 if (sb
->feature_map
& MD_FEATURE_JOURNAL
)
155 return 2; /* no sb loaded */
159 static int role_from_sb(struct mdp_superblock_1
*sb
)
164 d
= __le32_to_cpu(sb
->dev_number
);
165 if (d
< __le32_to_cpu(sb
->max_dev
))
166 role
= __le16_to_cpu(sb
->dev_roles
[d
]);
168 role
= MD_DISK_ROLE_SPARE
;
172 /* return how many bytes are needed for bitmap, for cluster-md each node
173 * should have it's own bitmap */
174 static unsigned int calc_bitmap_size(bitmap_super_t
*bms
, unsigned int boundary
)
176 unsigned long long bits
, bytes
;
178 bits
= __le64_to_cpu(bms
->sync_size
) / (__le32_to_cpu(bms
->chunksize
)>>9);
179 bytes
= (bits
+7) >> 3;
180 bytes
+= sizeof(bitmap_super_t
);
181 bytes
= ROUND_UP(bytes
, boundary
);
186 static unsigned int calc_sb_1_csum(struct mdp_superblock_1
* sb
)
188 unsigned int disk_csum
, csum
;
189 unsigned long long newcsum
;
190 int size
= sizeof(*sb
) + __le32_to_cpu(sb
->max_dev
)*2;
191 unsigned int *isuper
= (unsigned int*)sb
;
193 /* make sure I can count... */
194 if (offsetof(struct mdp_superblock_1
,data_offset
) != 128 ||
195 offsetof(struct mdp_superblock_1
, utime
) != 192 ||
196 sizeof(struct mdp_superblock_1
) != 256) {
197 fprintf(stderr
, "WARNING - superblock isn't sized correctly\n");
200 disk_csum
= sb
->sb_csum
;
203 for (; size
>=4; size
-= 4 ) {
204 newcsum
+= __le32_to_cpu(*isuper
);
209 newcsum
+= __le16_to_cpu(*(unsigned short*) isuper
);
211 csum
= (newcsum
& 0xffffffff) + (newcsum
>> 32);
212 sb
->sb_csum
= disk_csum
;
213 return __cpu_to_le32(csum
);
217 * Information related to file descriptor used for aligned reads/writes.
218 * Cache the block size.
225 static void init_afd(struct align_fd
*afd
, int fd
)
229 if (ioctl(afd
->fd
, BLKSSZGET
, &afd
->blk_sz
) != 0)
233 static char abuf
[4096+4096];
234 static int aread(struct align_fd
*afd
, void *buf
, int len
)
237 * On devices with a 4K sector size, we need to read
238 * the full sector and copy relevant bits into
247 if (!bsize
|| bsize
> 4096 || len
> 4096) {
249 fprintf(stderr
, "WARNING - aread() called with invalid block size\n");
252 b
= ROUND_UP_PTR((char *)abuf
, 4096);
254 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
256 n
= read(afd
->fd
, b
, iosize
);
259 lseek(afd
->fd
, len
- n
, 1);
266 static int awrite(struct align_fd
*afd
, void *buf
, int len
)
269 * On devices with a 4K sector size, we need to write
270 * the full sector. We pre-read if the sector is larger
272 * The address must be sector-aligned.
279 if (!bsize
|| bsize
> 4096 || len
> 4096) {
281 fprintf(stderr
, "WARNING - awrite() called with invalid block size\n");
284 b
= ROUND_UP_PTR((char *)abuf
, 4096);
286 for (iosize
= 0; iosize
< len
; iosize
+= bsize
)
290 n
= read(afd
->fd
, b
, iosize
);
293 lseek(afd
->fd
, -n
, 1);
297 n
= write(afd
->fd
, b
, iosize
);
300 lseek(afd
->fd
, len
- n
, 1);
305 static void examine_super1(struct supertype
*st
, char *homehost
)
307 struct mdp_superblock_1
*sb
= st
->sb
;
308 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
315 int l
= homehost
? strlen(homehost
) : 0;
317 unsigned long long sb_offset
;
320 printf(" Magic : %08x\n", __le32_to_cpu(sb
->magic
));
321 printf(" Version : 1");
322 sb_offset
= __le64_to_cpu(sb
->super_offset
);
325 else if (sb_offset
<= 8)
329 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb
->feature_map
));
330 printf(" Array UUID : ");
331 for (i
=0; i
<16; i
++) {
332 if ((i
&3)==0 && i
!= 0) printf(":");
333 printf("%02x", sb
->set_uuid
[i
]);
336 printf(" Name : %.32s", sb
->set_name
);
337 if (l
> 0 && l
< 32 &&
338 sb
->set_name
[l
] == ':' &&
339 strncmp(sb
->set_name
, homehost
, l
) == 0)
340 printf(" (local to host %s)", homehost
);
343 printf(" Cluster Name : %-64s\n", bms
->cluster_name
);
344 atime
= __le64_to_cpu(sb
->ctime
) & 0xFFFFFFFFFFULL
;
345 printf(" Creation Time : %.24s\n", ctime(&atime
));
346 c
=map_num(pers
, __le32_to_cpu(sb
->level
));
347 printf(" Raid Level : %s\n", c
?c
:"-unknown-");
348 printf(" Raid Devices : %d\n", __le32_to_cpu(sb
->raid_disks
));
350 printf(" Avail Dev Size : %llu%s\n",
351 (unsigned long long)__le64_to_cpu(sb
->data_size
),
352 human_size(__le64_to_cpu(sb
->data_size
)<<9));
353 if (__le32_to_cpu(sb
->level
) > 0) {
354 int ddsks
= 0, ddsks_denom
= 1;
355 switch(__le32_to_cpu(sb
->level
)) {
356 case 1: ddsks
=1;break;
358 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
359 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
361 layout
= __le32_to_cpu(sb
->layout
);
362 ddsks
= __le32_to_cpu(sb
->raid_disks
);
363 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
366 long long asize
= __le64_to_cpu(sb
->size
);
367 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
368 printf(" Array Size : %llu%s\n",
369 asize
>> 10, human_size(asize
));
371 if (sb
->size
!= sb
->data_size
)
372 printf(" Used Dev Size : %llu%s\n",
373 (unsigned long long)__le64_to_cpu(sb
->size
),
374 human_size(__le64_to_cpu(sb
->size
)<<9));
377 printf(" Data Offset : %llu sectors\n",
378 (unsigned long long)__le64_to_cpu(sb
->data_offset
));
379 if (sb
->new_offset
&&
380 (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
)) {
381 unsigned long long offset
= __le64_to_cpu(sb
->data_offset
);
382 offset
+= (signed)(int32_t)__le32_to_cpu(sb
->new_offset
);
383 printf(" New Offset : %llu sectors\n", offset
);
385 printf(" Super Offset : %llu sectors\n",
386 (unsigned long long)__le64_to_cpu(sb
->super_offset
));
387 if (__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_RECOVERY_OFFSET
)
388 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb
->recovery_offset
));
390 st
->ss
->getinfo_super(st
, &info
, NULL
);
391 if (info
.space_after
!= 1 &&
392 !(__le32_to_cpu(sb
->feature_map
) & MD_FEATURE_NEW_OFFSET
))
393 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
394 info
.space_before
, info
.space_after
);
396 printf(" State : %s\n", (__le64_to_cpu(sb
->resync_offset
)+1)? "active":"clean");
397 printf(" Device UUID : ");
398 for (i
=0; i
<16; i
++) {
399 if ((i
&3)==0 && i
!= 0) printf(":");
400 printf("%02x", sb
->device_uuid
[i
]);
404 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
405 printf("Internal Bitmap : %ld sectors from superblock\n",
406 (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
));
408 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)) {
409 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb
->reshape_position
)/2,
410 human_size(__le64_to_cpu(sb
->reshape_position
)<<9));
411 if (__le32_to_cpu(sb
->delta_disks
)) {
412 printf(" Delta Devices : %d", __le32_to_cpu(sb
->delta_disks
));
413 printf(" (%d->%d)\n",
414 __le32_to_cpu(sb
->raid_disks
)-__le32_to_cpu(sb
->delta_disks
),
415 __le32_to_cpu(sb
->raid_disks
));
416 if ((int)__le32_to_cpu(sb
->delta_disks
) < 0)
417 delta_extra
= -__le32_to_cpu(sb
->delta_disks
);
419 if (__le32_to_cpu(sb
->new_level
) != __le32_to_cpu(sb
->level
)) {
420 c
= map_num(pers
, __le32_to_cpu(sb
->new_level
));
421 printf(" New Level : %s\n", c
?c
:"-unknown-");
423 if (__le32_to_cpu(sb
->new_layout
) != __le32_to_cpu(sb
->layout
)) {
424 if (__le32_to_cpu(sb
->level
) == 5) {
425 c
= map_num(r5layout
, __le32_to_cpu(sb
->new_layout
));
426 printf(" New Layout : %s\n", c
?c
:"-unknown-");
428 if (__le32_to_cpu(sb
->level
) == 6) {
429 c
= map_num(r6layout
, __le32_to_cpu(sb
->new_layout
));
430 printf(" New Layout : %s\n", c
?c
:"-unknown-");
432 if (__le32_to_cpu(sb
->level
) == 10) {
433 printf(" New Layout :");
434 print_r10_layout(__le32_to_cpu(sb
->new_layout
));
438 if (__le32_to_cpu(sb
->new_chunk
) != __le32_to_cpu(sb
->chunksize
))
439 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb
->new_chunk
)/2);
444 if (sb
->devflags
& WriteMostly1
)
445 printf(" write-mostly");
449 atime
= __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
;
450 printf(" Update Time : %.24s\n", ctime(&atime
));
452 if (sb
->bblog_size
&& sb
->bblog_offset
) {
453 printf(" Bad Block Log : %d entries available at offset %ld sectors",
454 __le16_to_cpu(sb
->bblog_size
)*512/8,
455 (long)(int32_t)__le32_to_cpu(sb
->bblog_offset
));
456 if (sb
->feature_map
&
457 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
458 printf(" - bad blocks present.");
462 if (calc_sb_1_csum(sb
) == sb
->sb_csum
)
463 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb
->sb_csum
));
465 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb
->sb_csum
),
466 __le32_to_cpu(calc_sb_1_csum(sb
)));
467 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb
->events
));
469 if (__le32_to_cpu(sb
->level
) == 5) {
470 c
= map_num(r5layout
, __le32_to_cpu(sb
->layout
));
471 printf(" Layout : %s\n", c
?c
:"-unknown-");
473 if (__le32_to_cpu(sb
->level
) == 6) {
474 c
= map_num(r6layout
, __le32_to_cpu(sb
->layout
));
475 printf(" Layout : %s\n", c
?c
:"-unknown-");
477 if (__le32_to_cpu(sb
->level
) == 10) {
478 int lo
= __le32_to_cpu(sb
->layout
);
480 print_r10_layout(lo
);
483 switch(__le32_to_cpu(sb
->level
)) {
489 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
492 printf(" Rounding : %dK\n", __le32_to_cpu(sb
->chunksize
)/2);
498 /* This turns out to just be confusing */
499 printf(" Array Slot : %d (", __le32_to_cpu(sb
->dev_number
));
500 for (i
= __le32_to_cpu(sb
->max_dev
); i
> 0 ; i
--)
501 if (__le16_to_cpu(sb
->dev_roles
[i
-1]) != MD_DISK_ROLE_SPARE
)
503 for (d
=0; d
< i
; d
++) {
504 int role
= __le16_to_cpu(sb
->dev_roles
[d
]);
506 if (role
== MD_DISK_ROLE_SPARE
) printf("empty");
507 else if(role
== MD_DISK_ROLE_FAULTY
) printf("failed");
508 else printf("%d", role
);
512 printf(" Device Role : ");
513 role
= role_from_sb(sb
);
514 if (role
>= MD_DISK_ROLE_FAULTY
)
516 else if (role
== MD_DISK_ROLE_JOURNAL
)
518 else if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_REPLACEMENT
))
519 printf("Replacement device %d\n", role
);
521 printf("Active device %d\n", role
);
523 printf(" Array State : ");
524 for (d
=0; d
<__le32_to_cpu(sb
->raid_disks
) + delta_extra
; d
++) {
527 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
528 unsigned int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
542 /* This is confusing too */
544 for (i
=0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
545 int role
= __le16_to_cpu(sb
->dev_roles
[i
]);
546 if (role
== MD_DISK_ROLE_FAULTY
)
549 if (faulty
) printf(" %d failed", faulty
);
551 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
555 static void brief_examine_super1(struct supertype
*st
, int verbose
)
557 struct mdp_superblock_1
*sb
= st
->sb
;
559 unsigned long long sb_offset
;
561 char *c
=map_num(pers
, __le32_to_cpu(sb
->level
));
563 nm
= strchr(sb
->set_name
, ':');
566 else if (sb
->set_name
[0])
578 printf(" level=%s", c
);
579 sb_offset
= __le64_to_cpu(sb
->super_offset
);
581 printf(" metadata=1.1 ");
582 else if (sb_offset
<= 8)
583 printf(" metadata=1.2 ");
585 printf(" metadata=1.0 ");
587 printf("num-devices=%d ", __le32_to_cpu(sb
->raid_disks
));
589 for (i
=0; i
<16; i
++) {
590 if ((i
&3)==0 && i
!= 0) printf(":");
591 printf("%02x", sb
->set_uuid
[i
]);
593 if (sb
->set_name
[0]) {
595 print_quoted(sb
->set_name
);
600 static void export_examine_super1(struct supertype
*st
)
602 struct mdp_superblock_1
*sb
= st
->sb
;
607 printf("MD_LEVEL=%s\n", map_num(pers
, __le32_to_cpu(sb
->level
)));
608 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb
->raid_disks
));
610 if (sb
->set_name
[i
] == '\n' ||
611 sb
->set_name
[i
] == '\0') {
616 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
617 if (__le32_to_cpu(sb
->level
) > 0) {
618 int ddsks
= 0, ddsks_denom
= 1;
619 switch(__le32_to_cpu(sb
->level
)) {
620 case 1: ddsks
=1;break;
622 case 5: ddsks
= __le32_to_cpu(sb
->raid_disks
)-1; break;
623 case 6: ddsks
= __le32_to_cpu(sb
->raid_disks
)-2; break;
625 layout
= __le32_to_cpu(sb
->layout
);
626 ddsks
= __le32_to_cpu(sb
->raid_disks
);
627 ddsks_denom
= (layout
&255) * ((layout
>>8)&255);
630 long long asize
= __le64_to_cpu(sb
->size
);
631 asize
= (asize
<< 9) * ddsks
/ ddsks_denom
;
632 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize
,JEDEC
));
636 for (i
=0; i
<16; i
++) {
637 if ((i
&3)==0 && i
!= 0) printf(":");
638 printf("%02x", sb
->set_uuid
[i
]);
641 printf("MD_UPDATE_TIME=%llu\n",
642 __le64_to_cpu(sb
->utime
) & 0xFFFFFFFFFFULL
);
643 printf("MD_DEV_UUID=");
644 for (i
=0; i
<16; i
++) {
645 if ((i
&3)==0 && i
!= 0) printf(":");
646 printf("%02x", sb
->device_uuid
[i
]);
649 printf("MD_EVENTS=%llu\n",
650 (unsigned long long)__le64_to_cpu(sb
->events
));
653 static int copy_metadata1(struct supertype
*st
, int from
, int to
)
655 /* Read superblock. If it looks good, write it out.
656 * Then if a bitmap is present, copy that.
657 * And if a bad-block-list is present, copy that too.
660 unsigned long long dsize
, sb_offset
;
661 const int bufsize
= 4*1024;
662 struct mdp_superblock_1 super
, *sb
;
664 if (posix_memalign(&buf
, 4096, bufsize
) != 0)
667 if (!get_dev_size(from
, NULL
, &dsize
))
673 switch(st
->minor_version
) {
677 sb_offset
&= ~(4*2-1);
689 if (lseek64(from
, sb_offset
<< 9, 0) < 0LL)
691 if (read(from
, buf
, bufsize
) != bufsize
)
695 super
= *sb
; // save most of sb for when we reuse buf
697 if (__le32_to_cpu(super
.magic
) != MD_SB_MAGIC
||
698 __le32_to_cpu(super
.major_version
) != 1 ||
699 __le64_to_cpu(super
.super_offset
) != sb_offset
||
700 calc_sb_1_csum(sb
) != super
.sb_csum
)
703 if (lseek64(to
, sb_offset
<< 9, 0) < 0LL)
705 if (write(to
, buf
, bufsize
) != bufsize
)
708 if (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
)) {
709 unsigned long long bitmap_offset
= sb_offset
;
710 int bytes
= 4096; // just an estimate.
712 struct align_fd afrom
, ato
;
714 init_afd(&afrom
, from
);
717 bitmap_offset
+= (int32_t)__le32_to_cpu(super
.bitmap_offset
);
719 if (lseek64(from
, bitmap_offset
<<9, 0) < 0)
721 if (lseek64(to
, bitmap_offset
<<9, 0) < 0)
724 for (written
= 0; written
< bytes
; ) {
725 int n
= bytes
- written
;
728 if (aread(&afrom
, buf
, n
) != n
)
731 /* have the header, can calculate
732 * correct bitmap bytes */
735 bytes
= calc_bitmap_size(bms
, 512);
739 if (awrite(&ato
, buf
, n
) != n
)
745 if (super
.bblog_size
!= 0 &&
746 __le32_to_cpu(super
.bblog_size
) <= 100 &&
747 super
.bblog_offset
!= 0 &&
748 (super
.feature_map
& __le32_to_cpu(MD_FEATURE_BAD_BLOCKS
))) {
749 /* There is a bad block log */
750 unsigned long long bb_offset
= sb_offset
;
751 int bytes
= __le32_to_cpu(super
.bblog_size
) * 512;
753 struct align_fd afrom
, ato
;
755 init_afd(&afrom
, from
);
758 bb_offset
+= (int32_t)__le32_to_cpu(super
.bblog_offset
);
760 if (lseek64(from
, bb_offset
<<9, 0) < 0)
762 if (lseek64(to
, bb_offset
<<9, 0) < 0)
765 for (written
= 0; written
< bytes
; ) {
766 int n
= bytes
- written
;
769 if (aread(&afrom
, buf
, n
) != n
)
772 if (awrite(&ato
, buf
, n
) != n
)
786 static void detail_super1(struct supertype
*st
, char *homehost
)
788 struct mdp_superblock_1
*sb
= st
->sb
;
789 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
791 int l
= homehost
? strlen(homehost
) : 0;
793 printf(" Name : %.32s", sb
->set_name
);
794 if (l
> 0 && l
< 32 &&
795 sb
->set_name
[l
] == ':' &&
796 strncmp(sb
->set_name
, homehost
, l
) == 0)
797 printf(" (local to host %s)", homehost
);
799 printf("\n Cluster Name : %-64s", bms
->cluster_name
);
800 printf("\n UUID : ");
801 for (i
=0; i
<16; i
++) {
802 if ((i
&3)==0 && i
!= 0) printf(":");
803 printf("%02x", sb
->set_uuid
[i
]);
805 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb
->events
));
808 static void brief_detail_super1(struct supertype
*st
)
810 struct mdp_superblock_1
*sb
= st
->sb
;
813 if (sb
->set_name
[0]) {
815 print_quoted(sb
->set_name
);
818 for (i
=0; i
<16; i
++) {
819 if ((i
&3)==0 && i
!= 0) printf(":");
820 printf("%02x", sb
->set_uuid
[i
]);
824 static void export_detail_super1(struct supertype
*st
)
826 struct mdp_superblock_1
*sb
= st
->sb
;
831 if (sb
->set_name
[i
] == '\n' ||
832 sb
->set_name
[i
] == '\0') {
837 printf("MD_NAME=%.*s\n", len
, sb
->set_name
);
840 static int examine_badblocks_super1(struct supertype
*st
, int fd
, char *devname
)
842 struct mdp_superblock_1
*sb
= st
->sb
;
843 unsigned long long offset
;
848 if (!sb
->bblog_size
|| __le32_to_cpu(sb
->bblog_size
) > 100
849 || !sb
->bblog_offset
){
850 printf("No bad-blocks list configured on %s\n", devname
);
853 if ((sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
855 printf("Bad-blocks list is empty in %s\n", devname
);
859 size
= __le32_to_cpu(sb
->bblog_size
)* 512;
860 if (posix_memalign((void**)&bbl
, 4096, size
) != 0) {
861 pr_err("could not allocate badblocks list\n");
864 offset
= __le64_to_cpu(sb
->super_offset
) +
865 (int)__le32_to_cpu(sb
->bblog_offset
);
867 if (lseek64(fd
, offset
, 0) < 0) {
868 pr_err("Cannot seek to bad-blocks list\n");
871 if (read(fd
, bbl
, size
) != size
) {
872 pr_err("Cannot read bad-blocks list\n");
875 /* 64bits per entry. 10 bits is block-count, 54 bits is block
876 * offset. Blocks are sectors unless bblog->shift makes them bigger
879 printf("Bad-blocks on %s:\n", devname
);
880 for (i
= 0; i
< size
/8; i
++, bbp
++) {
881 __u64 bb
= __le64_to_cpu(*bbp
);
882 int count
= bb
& 0x3ff;
883 unsigned long long sector
= bb
>> 10;
888 sector
<<= sb
->bblog_shift
;
889 count
<<= sb
->bblog_shift
;
891 printf("%20llu for %d sectors\n", sector
, count
);
898 static int match_home1(struct supertype
*st
, char *homehost
)
900 struct mdp_superblock_1
*sb
= st
->sb
;
901 int l
= homehost
? strlen(homehost
) : 0;
903 return (l
> 0 && l
< 32 &&
904 sb
->set_name
[l
] == ':' &&
905 strncmp(sb
->set_name
, homehost
, l
) == 0);
908 static void uuid_from_super1(struct supertype
*st
, int uuid
[4])
910 struct mdp_superblock_1
*super
= st
->sb
;
911 char *cuuid
= (char*)uuid
;
914 cuuid
[i
] = super
->set_uuid
[i
];
917 static void getinfo_super1(struct supertype
*st
, struct mdinfo
*info
, char *map
)
919 struct mdp_superblock_1
*sb
= st
->sb
;
920 struct bitmap_super_s
*bsb
= (void*)(((char*)sb
)+MAX_SB_SIZE
);
921 struct misc_dev_info
*misc
= (void*)(((char*)sb
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
925 unsigned int map_disks
= info
->array
.raid_disks
;
926 unsigned long long super_offset
;
927 unsigned long long data_size
;
929 memset(info
, 0, sizeof(*info
));
930 info
->array
.major_version
= 1;
931 info
->array
.minor_version
= st
->minor_version
;
932 info
->array
.patch_version
= 0;
933 info
->array
.raid_disks
= __le32_to_cpu(sb
->raid_disks
);
934 info
->array
.level
= __le32_to_cpu(sb
->level
);
935 info
->array
.layout
= __le32_to_cpu(sb
->layout
);
936 info
->array
.md_minor
= -1;
937 info
->array
.ctime
= __le64_to_cpu(sb
->ctime
);
938 info
->array
.utime
= __le64_to_cpu(sb
->utime
);
939 info
->array
.chunk_size
= __le32_to_cpu(sb
->chunksize
)*512;
941 (__le64_to_cpu(sb
->resync_offset
) == MaxSector
)
943 if (__le32_to_cpu(bsb
->nodes
) > 1)
944 info
->array
.state
|= (1 << MD_SB_CLUSTERED
);
946 info
->data_offset
= __le64_to_cpu(sb
->data_offset
);
947 info
->component_size
= __le64_to_cpu(sb
->size
);
948 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET
))
949 info
->bitmap_offset
= (int32_t)__le32_to_cpu(sb
->bitmap_offset
);
951 info
->disk
.major
= 0;
952 info
->disk
.minor
= 0;
953 info
->disk
.number
= __le32_to_cpu(sb
->dev_number
);
954 if (__le32_to_cpu(sb
->dev_number
) >= __le32_to_cpu(sb
->max_dev
) ||
955 __le32_to_cpu(sb
->dev_number
) >= MAX_DEVS
)
956 role
= MD_DISK_ROLE_FAULTY
;
958 role
= __le16_to_cpu(sb
->dev_roles
[__le32_to_cpu(sb
->dev_number
)]);
960 super_offset
= __le64_to_cpu(sb
->super_offset
);
961 if (info
->array
.level
<= 0)
962 data_size
= __le64_to_cpu(sb
->data_size
);
964 data_size
= __le64_to_cpu(sb
->size
);
965 if (info
->data_offset
< super_offset
) {
966 unsigned long long end
;
967 info
->space_before
= info
->data_offset
;
970 if (sb
->bblog_offset
&& sb
->bblog_size
) {
971 unsigned long long bboffset
= super_offset
;
972 bboffset
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
977 if (super_offset
+ info
->bitmap_offset
< end
)
978 end
= super_offset
+ info
->bitmap_offset
;
980 if (info
->data_offset
+ data_size
< end
)
981 info
->space_after
= end
- data_size
- info
->data_offset
;
983 info
->space_after
= 0;
985 unsigned long long earliest
;
986 earliest
= super_offset
+ (32+4)*2; /* match kernel */
987 if (info
->bitmap_offset
> 0) {
988 unsigned long long bmend
= info
->bitmap_offset
;
989 unsigned long long size
= __le64_to_cpu(bsb
->sync_size
);
990 size
/= __le32_to_cpu(bsb
->chunksize
) >> 9;
991 size
= (size
+ 7) >> 3;
992 size
+= sizeof(bitmap_super_t
);
993 size
= ROUND_UP(size
, 4096);
996 if (bmend
> earliest
)
999 if (sb
->bblog_offset
&& sb
->bblog_size
) {
1000 unsigned long long bbend
= super_offset
;
1001 bbend
+= (int32_t)__le32_to_cpu(sb
->bblog_offset
);
1002 bbend
+= __le32_to_cpu(sb
->bblog_size
);
1003 if (bbend
> earliest
)
1006 if (earliest
< info
->data_offset
)
1007 info
->space_before
= info
->data_offset
- earliest
;
1009 info
->space_before
= 0;
1010 info
->space_after
= misc
->device_size
- data_size
- info
->data_offset
;
1012 if (info
->space_before
== 0 && info
->space_after
== 0) {
1013 /* It will look like we don't support data_offset changes,
1014 * be we do - it's just that there is no room.
1015 * A change that reduced the number of devices should
1016 * still be allowed, so set the otherwise useless value of '1'
1018 info
->space_after
= 1;
1021 info
->disk
.raid_disk
= -1;
1023 case MD_DISK_ROLE_SPARE
:
1024 info
->disk
.state
= 0; /* spare: not active, not sync, not faulty */
1026 case MD_DISK_ROLE_FAULTY
:
1027 info
->disk
.state
= 1; /* faulty */
1029 case MD_DISK_ROLE_JOURNAL
:
1030 info
->disk
.state
= (1 << MD_DISK_JOURNAL
);
1031 info
->disk
.raid_disk
= role
;
1032 info
->space_after
= (misc
->device_size
- info
->data_offset
) % 8; /* journal uses all 4kB blocks*/
1035 info
->disk
.state
= 6; /* active and in sync */
1036 info
->disk
.raid_disk
= role
;
1038 if (sb
->devflags
& WriteMostly1
)
1039 info
->disk
.state
|= (1 << MD_DISK_WRITEMOSTLY
);
1040 info
->events
= __le64_to_cpu(sb
->events
);
1041 sprintf(info
->text_version
, "1.%d", st
->minor_version
);
1042 info
->safe_mode_delay
= 200;
1044 memcpy(info
->uuid
, sb
->set_uuid
, 16);
1046 strncpy(info
->name
, sb
->set_name
, 32);
1049 if ((__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_REPLACEMENT
)) {
1050 info
->disk
.state
&= ~(1 << MD_DISK_SYNC
);
1051 info
->disk
.state
|= 1 << MD_DISK_REPLACEMENT
;
1054 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET
))
1055 info
->recovery_start
= __le32_to_cpu(sb
->recovery_offset
);
1057 info
->recovery_start
= MaxSector
;
1059 if (sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
)) {
1060 info
->reshape_active
= 1;
1061 if ((sb
->feature_map
& __le32_to_cpu(MD_FEATURE_NEW_OFFSET
)) &&
1062 sb
->new_offset
!= 0)
1063 info
->reshape_active
|= RESHAPE_NO_BACKUP
;
1064 info
->reshape_progress
= __le64_to_cpu(sb
->reshape_position
);
1065 info
->new_level
= __le32_to_cpu(sb
->new_level
);
1066 info
->delta_disks
= __le32_to_cpu(sb
->delta_disks
);
1067 info
->new_layout
= __le32_to_cpu(sb
->new_layout
);
1068 info
->new_chunk
= __le32_to_cpu(sb
->new_chunk
)<<9;
1069 if (info
->delta_disks
< 0)
1070 info
->array
.raid_disks
-= info
->delta_disks
;
1072 info
->reshape_active
= 0;
1074 info
->recovery_blocked
= info
->reshape_active
;
1077 for (i
=0; i
<map_disks
; i
++)
1079 for (i
= 0; i
< __le32_to_cpu(sb
->max_dev
); i
++) {
1080 role
= __le16_to_cpu(sb
->dev_roles
[i
]);
1081 if (/*role == MD_DISK_ROLE_SPARE || */role
< (unsigned) info
->array
.raid_disks
) {
1083 if (map
&& role
< map_disks
)
1088 info
->array
.working_disks
= working
;
1091 static struct mdinfo
*container_content1(struct supertype
*st
, char *subarray
)
1093 struct mdinfo
*info
;
1098 info
= xmalloc(sizeof(*info
));
1099 getinfo_super1(st
, info
, NULL
);
1103 static int update_super1(struct supertype
*st
, struct mdinfo
*info
,
1105 char *devname
, int verbose
,
1106 int uuid_set
, char *homehost
)
1108 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1109 * if any change was made. For others, the return value is
1114 struct mdp_superblock_1
*sb
= st
->sb
;
1116 if (is_clustered(st
)) {
1117 rv
= cluster_get_dlmlock(st
, &lockid
);
1119 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1120 cluster_release_dlmlock(st
, lockid
);
1125 if (strcmp(update
, "homehost") == 0 &&
1127 /* Note that 'homehost' is special as it is really
1132 c
= strchr(sb
->set_name
, ':');
1134 strncpy(info
->name
, c
+1, 31 - (c
-sb
->set_name
));
1136 strncpy(info
->name
, sb
->set_name
, 32);
1140 if (strcmp(update
, "force-one")==0) {
1141 /* Not enough devices for a working array,
1142 * so bring this one up-to-date
1144 if (sb
->events
!= __cpu_to_le64(info
->events
))
1146 sb
->events
= __cpu_to_le64(info
->events
);
1147 } else if (strcmp(update
, "force-array")==0) {
1148 /* Degraded array and 'force' requests to
1149 * maybe need to mark it 'clean'.
1151 switch(__le32_to_cpu(sb
->level
)) {
1152 case 5: case 4: case 6:
1153 /* need to force clean */
1154 if (sb
->resync_offset
!= MaxSector
)
1156 sb
->resync_offset
= MaxSector
;
1158 } else if (strcmp(update
, "assemble")==0) {
1159 int d
= info
->disk
.number
;
1161 if (info
->disk
.state
& (1<<MD_DISK_ACTIVE
))
1162 want
= info
->disk
.raid_disk
;
1163 else if (info
->disk
.state
& (1<<MD_DISK_JOURNAL
))
1164 want
= MD_DISK_ROLE_JOURNAL
;
1166 want
= MD_DISK_ROLE_SPARE
;
1167 if (sb
->dev_roles
[d
] != __cpu_to_le16(want
)) {
1168 sb
->dev_roles
[d
] = __cpu_to_le16(want
);
1171 if (info
->reshape_active
&&
1172 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1173 info
->delta_disks
>= 0 &&
1174 info
->reshape_progress
< __le64_to_cpu(sb
->reshape_position
)) {
1175 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1178 if (info
->reshape_active
&&
1179 sb
->feature_map
& __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE
) &&
1180 info
->delta_disks
< 0 &&
1181 info
->reshape_progress
> __le64_to_cpu(sb
->reshape_position
)) {
1182 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1185 } else if (strcmp(update
, "linear-grow-new") == 0) {
1188 unsigned int max
= __le32_to_cpu(sb
->max_dev
);
1190 for (i
=0 ; i
< max
; i
++)
1191 if (__le16_to_cpu(sb
->dev_roles
[i
]) >= MD_DISK_ROLE_FAULTY
)
1193 sb
->dev_number
= __cpu_to_le32(i
);
1194 info
->disk
.number
= i
;
1195 if (max
>= __le32_to_cpu(sb
->max_dev
))
1196 sb
->max_dev
= __cpu_to_le32(max
+1);
1198 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1199 read(rfd
, sb
->device_uuid
, 16) != 16) {
1200 __u32 r
[4] = {random(), random(), random(), random()};
1201 memcpy(sb
->device_uuid
, r
, 16);
1207 __cpu_to_le16(info
->disk
.raid_disk
);
1209 fd
= open(devname
, O_RDONLY
);
1211 unsigned long long ds
;
1212 get_dev_size(fd
, devname
, &ds
);
1215 if (__le64_to_cpu(sb
->super_offset
) <
1216 __le64_to_cpu(sb
->data_offset
)) {
1217 sb
->data_size
= __cpu_to_le64(
1218 ds
- __le64_to_cpu(sb
->data_offset
));
1221 ds
&= ~(unsigned long long)(4*2-1);
1222 sb
->super_offset
= __cpu_to_le64(ds
);
1223 sb
->data_size
= __cpu_to_le64(
1224 ds
- __le64_to_cpu(sb
->data_offset
));
1227 } else if (strcmp(update
, "linear-grow-update") == 0) {
1228 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
1229 sb
->dev_roles
[info
->disk
.number
] =
1230 __cpu_to_le16(info
->disk
.raid_disk
);
1231 } else if (strcmp(update
, "resync") == 0) {
1232 /* make sure resync happens */
1233 sb
->resync_offset
= 0ULL;
1234 } else if (strcmp(update
, "uuid") == 0) {
1235 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
1237 if (__le32_to_cpu(sb
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
1238 struct bitmap_super_s
*bm
;
1239 bm
= (struct bitmap_super_s
*)(st
->sb
+MAX_SB_SIZE
);
1240 memcpy(bm
->uuid
, sb
->set_uuid
, 16);
1242 } else if (strcmp(update
, "no-bitmap") == 0) {
1243 sb
->feature_map
&= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
);
1244 } else if (strcmp(update
, "bbl") == 0) {
1245 /* only possible if there is room after the bitmap, or if
1246 * there is no bitmap
1248 unsigned long long sb_offset
= __le64_to_cpu(sb
->super_offset
);
1249 unsigned long long data_offset
= __le64_to_cpu(sb
->data_offset
);
1250 long bitmap_offset
= (long)(int32_t)__le32_to_cpu(sb
->bitmap_offset
);
1251 long bm_sectors
= 0;
1255 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1256 struct bitmap_super_s
*bsb
;
1257 bsb
= (struct bitmap_super_s
*)(((char*)sb
)+MAX_SB_SIZE
);
1258 bm_sectors
= bitmap_sectors(bsb
);
1261 if (sb_offset
< data_offset
) {
1262 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1265 bb_offset
= sb_offset
+ 8;
1266 if (bm_sectors
&& bitmap_offset
> 0)
1267 bb_offset
= bitmap_offset
+ bm_sectors
;
1268 while (bb_offset
< (long)sb_offset
+ 8 + 32*2
1269 && bb_offset
+ 8+8 <= (long)data_offset
)
1270 /* too close to bitmap, and room to grow */
1272 if (bb_offset
+ 8 <= (long)data_offset
) {
1273 sb
->bblog_size
= __cpu_to_le16(8);
1274 sb
->bblog_offset
= __cpu_to_le32(bb_offset
);
1277 /* 1.0 - Put bbl just before super block */
1278 if (bm_sectors
&& bitmap_offset
< 0)
1279 space
= -bitmap_offset
- bm_sectors
;
1281 space
= sb_offset
- data_offset
-
1282 __le64_to_cpu(sb
->data_size
);
1284 sb
->bblog_size
= __cpu_to_le16(8);
1285 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1288 } else if (strcmp(update
, "no-bbl") == 0) {
1289 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BAD_BLOCKS
))
1290 pr_err("Cannot remove active bbl from %s\n",devname
);
1293 sb
->bblog_shift
= 0;
1294 sb
->bblog_offset
= 0;
1296 } else if (strcmp(update
, "name") == 0) {
1297 if (info
->name
[0] == 0)
1298 sprintf(info
->name
, "%d", info
->array
.md_minor
);
1299 memset(sb
->set_name
, 0, sizeof(sb
->set_name
));
1301 strchr(info
->name
, ':') == NULL
&&
1302 strlen(homehost
)+1+strlen(info
->name
) < 32) {
1303 strcpy(sb
->set_name
, homehost
);
1304 strcat(sb
->set_name
, ":");
1305 strcat(sb
->set_name
, info
->name
);
1307 strcpy(sb
->set_name
, info
->name
);
1308 } else if (strcmp(update
, "devicesize") == 0 &&
1309 __le64_to_cpu(sb
->super_offset
) <
1310 __le64_to_cpu(sb
->data_offset
)) {
1311 /* set data_size to device size less data_offset */
1312 struct misc_dev_info
*misc
= (struct misc_dev_info
*)
1313 (st
->sb
+ MAX_SB_SIZE
+ BM_SUPER_SIZE
);
1314 sb
->data_size
= __cpu_to_le64(
1315 misc
->device_size
- __le64_to_cpu(sb
->data_offset
));
1316 } else if (strcmp(update
, "revert-reshape") == 0) {
1318 if (!(sb
->feature_map
& __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE
)))
1319 pr_err("No active reshape to revert on %s\n",
1323 unsigned long long reshape_sectors
;
1326 /* reshape_position is a little messy.
1327 * Its value must be a multiple of the larger
1328 * chunk size, and of the "after" data disks.
1329 * So when reverting we need to change it to
1330 * be a multiple of the new "after" data disks,
1331 * which is the old "before".
1332 * If it isn't already a multiple of 'before',
1333 * the only thing we could do would be
1334 * copy some block around on the disks, which
1335 * is easy to get wrong.
1336 * So we reject a revert-reshape unless the
1337 * alignment is good.
1339 if (__le32_to_cpu(sb
->level
) >= 4 &&
1340 __le32_to_cpu(sb
->level
) <= 6) {
1341 reshape_sectors
= __le64_to_cpu(sb
->reshape_position
);
1342 reshape_chunk
= __le32_to_cpu(sb
->new_chunk
);
1343 reshape_chunk
*= __le32_to_cpu(sb
->raid_disks
) - __le32_to_cpu(sb
->delta_disks
) -
1344 (__le32_to_cpu(sb
->level
)==6 ? 2 : 1);
1345 if (reshape_sectors
% reshape_chunk
) {
1346 pr_err("Reshape position is not suitably aligned.\n");
1347 pr_err("Try normal assembly and stop again\n");
1351 sb
->raid_disks
= __cpu_to_le32(__le32_to_cpu(sb
->raid_disks
) -
1352 __le32_to_cpu(sb
->delta_disks
));
1353 if (sb
->delta_disks
== 0)
1354 sb
->feature_map
^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS
);
1356 sb
->delta_disks
= __cpu_to_le32(-__le32_to_cpu(sb
->delta_disks
));
1358 temp
= sb
->new_layout
;
1359 sb
->new_layout
= sb
->layout
;
1362 temp
= sb
->new_chunk
;
1363 sb
->new_chunk
= sb
->chunksize
;
1364 sb
->chunksize
= temp
;
1366 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_NEW_OFFSET
)) {
1367 long offset_delta
= (int32_t)__le32_to_cpu(sb
->new_offset
);
1368 sb
->data_offset
= __cpu_to_le64(__le64_to_cpu(sb
->data_offset
) + offset_delta
);
1369 sb
->new_offset
= __cpu_to_le32(-offset_delta
);
1370 sb
->data_size
= __cpu_to_le64(__le64_to_cpu(sb
->data_size
) - offset_delta
);
1373 } else if (strcmp(update
, "_reshape_progress")==0)
1374 sb
->reshape_position
= __cpu_to_le64(info
->reshape_progress
);
1375 else if (strcmp(update
, "writemostly")==0)
1376 sb
->devflags
|= WriteMostly1
;
1377 else if (strcmp(update
, "readwrite")==0)
1378 sb
->devflags
&= ~WriteMostly1
;
1382 sb
->sb_csum
= calc_sb_1_csum(sb
);
1383 if (is_clustered(st
))
1384 cluster_release_dlmlock(st
, lockid
);
1389 static int init_super1(struct supertype
*st
, mdu_array_info_t
*info
,
1390 unsigned long long size
, char *name
, char *homehost
,
1391 int *uuid
, unsigned long long data_offset
)
1393 struct mdp_superblock_1
*sb
;
1399 if (posix_memalign((void**)&sb
, 4096, SUPER1_SIZE
) != 0) {
1400 pr_err("could not allocate superblock\n");
1403 memset(sb
, 0, SUPER1_SIZE
);
1407 /* zeroing superblock */
1411 spares
= info
->working_disks
- info
->active_disks
;
1412 if (info
->raid_disks
+ spares
> MAX_DEVS
) {
1413 pr_err("too many devices requested: %d+%d > %d\n",
1414 info
->raid_disks
, spares
, MAX_DEVS
);
1418 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
1419 sb
->major_version
= __cpu_to_le32(1);
1420 sb
->feature_map
= 0;
1424 copy_uuid(sb
->set_uuid
, uuid
, super1
.swapuuid
);
1426 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1427 read(rfd
, sb
->set_uuid
, 16) != 16) {
1428 __u32 r
[4] = {random(), random(), random(), random()};
1429 memcpy(sb
->set_uuid
, r
, 16);
1431 if (rfd
>= 0) close(rfd
);
1434 if (name
== NULL
|| *name
== 0) {
1435 sprintf(defname
, "%d", info
->md_minor
);
1439 strchr(name
, ':')== NULL
&&
1440 strlen(homehost
)+1+strlen(name
) < 32) {
1441 strcpy(sb
->set_name
, homehost
);
1442 strcat(sb
->set_name
, ":");
1443 strcat(sb
->set_name
, name
);
1445 strcpy(sb
->set_name
, name
);
1447 sb
->ctime
= __cpu_to_le64((unsigned long long)time(0));
1448 sb
->level
= __cpu_to_le32(info
->level
);
1449 sb
->layout
= __cpu_to_le32(info
->layout
);
1450 sb
->size
= __cpu_to_le64(size
*2ULL);
1451 sb
->chunksize
= __cpu_to_le32(info
->chunk_size
>>9);
1452 sb
->raid_disks
= __cpu_to_le32(info
->raid_disks
);
1454 sb
->data_offset
= __cpu_to_le64(data_offset
);
1455 sb
->data_size
= __cpu_to_le64(0);
1456 sb
->super_offset
= __cpu_to_le64(0);
1457 sb
->recovery_offset
= __cpu_to_le64(0);
1459 sb
->utime
= sb
->ctime
;
1460 sb
->events
= __cpu_to_le64(1);
1461 if (info
->state
& (1<<MD_SB_CLEAN
))
1462 sb
->resync_offset
= MaxSector
;
1464 sb
->resync_offset
= 0;
1465 sbsize
= sizeof(struct mdp_superblock_1
) + 2 * (info
->raid_disks
+ spares
);
1466 sbsize
= ROUND_UP(sbsize
, 512);
1467 sb
->max_dev
= __cpu_to_le32((sbsize
- sizeof(struct mdp_superblock_1
)) / 2);
1469 memset(sb
->dev_roles
, 0xff, MAX_SB_SIZE
- sizeof(struct mdp_superblock_1
));
1477 long long data_offset
;
1478 mdu_disk_info_t disk
;
1479 struct devinfo
*next
;
1482 /* Add a device to the superblock being created */
1483 static int add_to_super1(struct supertype
*st
, mdu_disk_info_t
*dk
,
1484 int fd
, char *devname
, unsigned long long data_offset
)
1486 struct mdp_superblock_1
*sb
= st
->sb
;
1487 __u16
*rp
= sb
->dev_roles
+ dk
->number
;
1488 struct devinfo
*di
, **dip
;
1491 if (is_clustered(st
)) {
1492 rv
= cluster_get_dlmlock(st
, &lockid
);
1494 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1495 cluster_release_dlmlock(st
, lockid
);
1500 if ((dk
->state
& 6) == 6) /* active, sync */
1501 *rp
= __cpu_to_le16(dk
->raid_disk
);
1502 else if (dk
->state
& (1<<MD_DISK_JOURNAL
))
1503 *rp
= MD_DISK_ROLE_JOURNAL
;
1504 else if ((dk
->state
& ~2) == 0) /* active or idle -> spare */
1505 *rp
= MD_DISK_ROLE_SPARE
;
1507 *rp
= MD_DISK_ROLE_FAULTY
;
1509 if (dk
->number
>= (int)__le32_to_cpu(sb
->max_dev
) &&
1510 __le32_to_cpu(sb
->max_dev
) < MAX_DEVS
)
1511 sb
->max_dev
= __cpu_to_le32(dk
->number
+1);
1513 sb
->dev_number
= __cpu_to_le32(dk
->number
);
1514 sb
->devflags
= 0; /* don't copy another disks flags */
1515 sb
->sb_csum
= calc_sb_1_csum(sb
);
1517 dip
= (struct devinfo
**)&st
->info
;
1519 dip
= &(*dip
)->next
;
1520 di
= xmalloc(sizeof(struct devinfo
));
1522 di
->devname
= devname
;
1524 di
->data_offset
= data_offset
;
1528 if (is_clustered(st
))
1529 cluster_release_dlmlock(st
, lockid
);
1535 static void locate_bitmap1(struct supertype
*st
, int fd
);
1537 static int store_super1(struct supertype
*st
, int fd
)
1539 struct mdp_superblock_1
*sb
= st
->sb
;
1540 unsigned long long sb_offset
;
1541 struct align_fd afd
;
1543 unsigned long long dsize
;
1546 if (is_clustered(st
)) {
1547 rv
= cluster_get_dlmlock(st
, &lockid
);
1549 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
1550 cluster_release_dlmlock(st
, lockid
);
1555 if (!get_dev_size(fd
, NULL
, &dsize
))
1566 * Calculate the position of the superblock.
1567 * It is always aligned to a 4K boundary and
1568 * depending on minor_version, it can be:
1569 * 0: At least 8K, but less than 12K, from end of device
1570 * 1: At start of device
1571 * 2: 4K from start of device.
1573 switch(st
->minor_version
) {
1577 sb_offset
&= ~(4*2-1);
1589 if (sb_offset
!= __le64_to_cpu(sb
->super_offset
) &&
1590 0 != __le64_to_cpu(sb
->super_offset
)
1592 pr_err("internal error - sb_offset is wrong\n");
1596 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL)
1599 sbsize
= ROUND_UP(sizeof(*sb
) + 2 * __le32_to_cpu(sb
->max_dev
), 512);
1601 if (awrite(&afd
, sb
, sbsize
) != sbsize
)
1604 if (sb
->feature_map
& __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET
)) {
1605 struct bitmap_super_s
*bm
= (struct bitmap_super_s
*)
1606 (((char*)sb
)+MAX_SB_SIZE
);
1607 if (__le32_to_cpu(bm
->magic
) == BITMAP_MAGIC
) {
1608 locate_bitmap1(st
, fd
);
1609 if (awrite(&afd
, bm
, sizeof(*bm
)) != sizeof(*bm
))
1614 if (is_clustered(st
))
1615 cluster_release_dlmlock(st
, lockid
);
1620 static int load_super1(struct supertype
*st
, int fd
, char *devname
);
1622 static unsigned long choose_bm_space(unsigned long devsize
)
1624 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1625 * if bigger than 200Gig, save 128k
1626 * NOTE: result must be multiple of 4K else bad things happen
1627 * on 4K-sector devices.
1629 if (devsize
< 64*2) return 0;
1630 if (devsize
- 64*2 >= 200*1024*1024*2)
1632 if (devsize
- 4*2 > 8*1024*1024*2)
1637 static void free_super1(struct supertype
*st
);
1639 #define META_BLOCK_SIZE 4096
1640 unsigned long crc32(
1642 const unsigned char *buf
,
1645 static int write_empty_r5l_meta_block(struct supertype
*st
, int fd
)
1647 struct r5l_meta_block
*mb
;
1648 struct mdp_superblock_1
*sb
= st
->sb
;
1649 struct align_fd afd
;
1654 if (posix_memalign((void**)&mb
, 4096, META_BLOCK_SIZE
) != 0) {
1655 pr_err("Could not allocate memory for the meta block.\n");
1659 memset(mb
, 0, META_BLOCK_SIZE
);
1661 mb
->magic
= __cpu_to_le32(R5LOG_MAGIC
);
1662 mb
->version
= R5LOG_VERSION
;
1663 mb
->meta_size
= __cpu_to_le32(sizeof(struct r5l_meta_block
));
1664 mb
->seq
= __cpu_to_le64(random32());
1665 mb
->position
= __cpu_to_le64(0);
1667 crc
= crc32(0xffffffff, sb
->set_uuid
, sizeof(sb
->set_uuid
));
1668 crc
= crc32(crc
, (void *)mb
, META_BLOCK_SIZE
);
1669 mb
->checksum
= __cpu_to_le32(crc
);
1671 if (lseek64(fd
, (sb
->data_offset
) * 512, 0) < 0LL) {
1672 pr_err("cannot seek to offset of the meta block\n");
1676 if (awrite(&afd
, mb
, META_BLOCK_SIZE
) != META_BLOCK_SIZE
) {
1677 pr_err("failed to store write the meta block \n");
1691 static int write_init_super1(struct supertype
*st
)
1693 struct mdp_superblock_1
*sb
= st
->sb
;
1694 struct supertype
*refst
;
1697 unsigned long long bm_space
;
1699 unsigned long long dsize
, array_size
;
1700 unsigned long long sb_offset
;
1701 unsigned long long data_offset
;
1703 for (di
= st
->info
; di
; di
= di
->next
) {
1704 if (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))
1705 sb
->feature_map
|= MD_FEATURE_JOURNAL
;
1708 for (di
= st
->info
; di
; di
= di
->next
) {
1709 if (di
->disk
.state
& (1 << MD_DISK_FAULTY
))
1714 while (Kill(di
->devname
, NULL
, 0, -1, 1) == 0)
1717 sb
->dev_number
= __cpu_to_le32(di
->disk
.number
);
1718 if (di
->disk
.state
& (1<<MD_DISK_WRITEMOSTLY
))
1719 sb
->devflags
|= WriteMostly1
;
1721 sb
->devflags
&= ~WriteMostly1
;
1723 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
1724 read(rfd
, sb
->device_uuid
, 16) != 16) {
1725 __u32 r
[4] = {random(), random(), random(), random()};
1726 memcpy(sb
->device_uuid
, r
, 16);
1733 refst
= dup_super(st
);
1734 if (load_super1(refst
, di
->fd
, NULL
)==0) {
1735 struct mdp_superblock_1
*refsb
= refst
->sb
;
1737 memcpy(sb
->device_uuid
, refsb
->device_uuid
, 16);
1738 if (memcmp(sb
->set_uuid
, refsb
->set_uuid
, 16)==0) {
1739 /* same array, so preserve events and
1741 sb
->events
= refsb
->events
;
1742 /* bugs in 2.6.17 and earlier mean the
1743 * dev_number chosen in Manage must be preserved
1745 if (get_linux_version() >= 2006018)
1746 sb
->dev_number
= refsb
->dev_number
;
1752 if (!get_dev_size(di
->fd
, NULL
, &dsize
)) {
1765 * Calculate the position of the superblock.
1766 * It is always aligned to a 4K boundary and
1767 * depending on minor_version, it can be:
1768 * 0: At least 8K, but less than 12K, from end of device
1769 * 1: At start of device
1770 * 2: 4K from start of device.
1771 * data_offset has already been set.
1773 array_size
= __le64_to_cpu(sb
->size
);
1774 /* work out how much space we left for a bitmap,
1775 * Add 8 sectors for bad block log */
1776 bm_space
= choose_bm_space(array_size
) + 8;
1778 data_offset
= di
->data_offset
;
1779 if (data_offset
== INVALID_SECTORS
)
1780 data_offset
= st
->data_offset
;
1781 switch(st
->minor_version
) {
1783 if (data_offset
== INVALID_SECTORS
)
1787 sb_offset
&= ~(4*2-1);
1788 sb
->data_offset
= __cpu_to_le64(data_offset
);
1789 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1790 if (sb_offset
< array_size
+ bm_space
)
1791 bm_space
= sb_offset
- array_size
;
1792 sb
->data_size
= __cpu_to_le64(sb_offset
- bm_space
);
1793 if (bm_space
>= 8) {
1794 sb
->bblog_size
= __cpu_to_le16(8);
1795 sb
->bblog_offset
= __cpu_to_le32((unsigned)-8);
1799 sb
->super_offset
= __cpu_to_le64(0);
1800 if (data_offset
== INVALID_SECTORS
)
1803 sb
->data_offset
= __cpu_to_le64(data_offset
);
1804 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1805 if (data_offset
>= 8 + 32*2 + 8) {
1806 sb
->bblog_size
= __cpu_to_le16(8);
1807 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1808 } else if (data_offset
>= 16) {
1809 sb
->bblog_size
= __cpu_to_le16(8);
1810 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8);
1815 sb
->super_offset
= __cpu_to_le64(sb_offset
);
1816 if (data_offset
== INVALID_SECTORS
)
1819 sb
->data_offset
= __cpu_to_le64(data_offset
);
1820 sb
->data_size
= __cpu_to_le64(dsize
- data_offset
);
1821 if (data_offset
>= 16 + 32*2 + 8) {
1822 sb
->bblog_size
= __cpu_to_le16(8);
1823 sb
->bblog_offset
= __cpu_to_le32(8 + 32*2);
1824 } else if (data_offset
>= 16+16) {
1825 sb
->bblog_size
= __cpu_to_le16(8);
1826 /* '8' sectors for the bblog, and another '8'
1827 * because we want offset from superblock, not
1830 sb
->bblog_offset
= __cpu_to_le32(data_offset
-8-8);
1834 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1835 st
->minor_version
, di
->devname
);
1839 /* Disable badblock log on clusters, or when explicitly requested */
1840 if (st
->nodes
> 0 || conf_get_create_info()->bblist
== 0) {
1842 sb
->bblog_offset
= 0;
1845 sb
->sb_csum
= calc_sb_1_csum(sb
);
1846 rv
= store_super1(st
, di
->fd
);
1848 if (rv
== 0 && (di
->disk
.state
& (1 << MD_DISK_JOURNAL
))) {
1849 rv
= write_empty_r5l_meta_block(st
, di
->fd
);
1854 if (rv
== 0 && (__le32_to_cpu(sb
->feature_map
) & 1))
1855 rv
= st
->ss
->write_bitmap(st
, di
->fd
, NoUpdate
);
1863 pr_err("Failed to write metadata to %s\n",
1870 static int compare_super1(struct supertype
*st
, struct supertype
*tst
)
1874 * 0 same, or first was empty, and second was copied
1875 * 1 second had wrong number
1877 * 3 wrong other info
1879 struct mdp_superblock_1
*first
= st
->sb
;
1880 struct mdp_superblock_1
*second
= tst
->sb
;
1882 if (second
->magic
!= __cpu_to_le32(MD_SB_MAGIC
))
1884 if (second
->major_version
!= __cpu_to_le32(1))
1888 if (posix_memalign((void**)&first
, 4096, SUPER1_SIZE
) != 0) {
1889 pr_err("could not allocate superblock\n");
1892 memcpy(first
, second
, SUPER1_SIZE
);
1896 if (memcmp(first
->set_uuid
, second
->set_uuid
, 16)!= 0)
1899 if (first
->ctime
!= second
->ctime
||
1900 first
->level
!= second
->level
||
1901 first
->layout
!= second
->layout
||
1902 first
->size
!= second
->size
||
1903 first
->chunksize
!= second
->chunksize
||
1904 first
->raid_disks
!= second
->raid_disks
)
1909 static int load_super1(struct supertype
*st
, int fd
, char *devname
)
1911 unsigned long long dsize
;
1912 unsigned long long sb_offset
;
1913 struct mdp_superblock_1
*super
;
1915 struct bitmap_super_s
*bsb
;
1916 struct misc_dev_info
*misc
;
1917 struct align_fd afd
;
1923 if (st
->ss
== NULL
|| st
->minor_version
== -1) {
1925 struct supertype tst
;
1926 __u64 bestctime
= 0;
1927 /* guess... choose latest ctime */
1928 memset(&tst
, 0, sizeof(tst
));
1930 for (tst
.minor_version
= 0; tst
.minor_version
<= 2 ; tst
.minor_version
++) {
1931 switch(load_super1(&tst
, fd
, devname
)) {
1932 case 0: super
= tst
.sb
;
1933 if (bestvers
== -1 ||
1934 bestctime
< __le64_to_cpu(super
->ctime
)) {
1935 bestvers
= tst
.minor_version
;
1936 bestctime
= __le64_to_cpu(super
->ctime
);
1941 case 1: return 1; /*bad device */
1942 case 2: break; /* bad, try next */
1945 if (bestvers
!= -1) {
1947 tst
.minor_version
= bestvers
;
1949 tst
.max_devs
= MAX_DEVS
;
1950 rv
= load_super1(&tst
, fd
, devname
);
1957 if (!get_dev_size(fd
, devname
, &dsize
))
1963 pr_err("%s is too small for md: size is %llu sectors.\n",
1969 * Calculate the position of the superblock.
1970 * It is always aligned to a 4K boundary and
1971 * depending on minor_version, it can be:
1972 * 0: At least 8K, but less than 12K, from end of device
1973 * 1: At start of device
1974 * 2: 4K from start of device.
1976 switch(st
->minor_version
) {
1980 sb_offset
&= ~(4*2-1);
1992 if (lseek64(fd
, sb_offset
<< 9, 0)< 0LL) {
1994 pr_err("Cannot seek to superblock on %s: %s\n",
1995 devname
, strerror(errno
));
1999 if (posix_memalign((void**)&super
, 4096, SUPER1_SIZE
) != 0) {
2000 pr_err("could not allocate superblock\n");
2004 if (aread(&afd
, super
, MAX_SB_SIZE
) != MAX_SB_SIZE
) {
2006 pr_err("Cannot read superblock on %s\n",
2012 if (__le32_to_cpu(super
->magic
) != MD_SB_MAGIC
) {
2014 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2015 devname
, MD_SB_MAGIC
, __le32_to_cpu(super
->magic
));
2020 if (__le32_to_cpu(super
->major_version
) != 1) {
2022 pr_err("Cannot interpret superblock on %s - version is %d\n",
2023 devname
, __le32_to_cpu(super
->major_version
));
2027 if (__le64_to_cpu(super
->super_offset
) != sb_offset
) {
2029 pr_err("No superblock found on %s (super_offset is wrong)\n",
2036 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2038 misc
= (struct misc_dev_info
*) (((char*)super
)+MAX_SB_SIZE
+BM_SUPER_SIZE
);
2039 misc
->device_size
= dsize
;
2040 if (st
->data_offset
== INVALID_SECTORS
)
2041 st
->data_offset
= __le64_to_cpu(super
->data_offset
);
2043 /* Now check on the bitmap superblock */
2044 if ((__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) == 0)
2046 /* Read the bitmap superblock and make sure it looks
2047 * valid. If it doesn't clear the bit. An --assemble --force
2048 * should get that written out.
2050 locate_bitmap1(st
, fd
);
2051 if (aread(&afd
, bsb
, 512) != 512)
2054 uuid_from_super1(st
, uuid
);
2055 if (__le32_to_cpu(bsb
->magic
) != BITMAP_MAGIC
||
2056 memcmp(bsb
->uuid
, uuid
, 16) != 0)
2061 super
->feature_map
= __cpu_to_le32(__le32_to_cpu(super
->feature_map
)
2062 & ~MD_FEATURE_BITMAP_OFFSET
);
2066 static struct supertype
*match_metadata_desc1(char *arg
)
2068 struct supertype
*st
= xcalloc(1, sizeof(*st
));
2070 st
->container_devnm
[0] = 0;
2072 st
->max_devs
= MAX_DEVS
;
2074 st
->data_offset
= INVALID_SECTORS
;
2075 /* leading zeros can be safely ignored. --detail generates them. */
2078 if (strcmp(arg
, "1.0") == 0 ||
2079 strcmp(arg
, "1.00") == 0) {
2080 st
->minor_version
= 0;
2083 if (strcmp(arg
, "1.1") == 0 ||
2084 strcmp(arg
, "1.01") == 0
2086 st
->minor_version
= 1;
2089 if (strcmp(arg
, "1.2") == 0 ||
2090 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2091 strcmp(arg
, "default") == 0 ||
2092 #endif /* DEFAULT_OLD_METADATA */
2093 strcmp(arg
, "1.02") == 0) {
2094 st
->minor_version
= 2;
2097 if (strcmp(arg
, "1") == 0 ||
2098 strcmp(arg
, "default") == 0) {
2099 st
->minor_version
= -1;
2107 /* find available size on device with this devsize, using
2108 * superblock type st, and reserving 'reserve' sectors for
2111 static __u64
avail_size1(struct supertype
*st
, __u64 devsize
,
2112 unsigned long long data_offset
)
2114 struct mdp_superblock_1
*super
= st
->sb
;
2121 if (__le32_to_cpu(super
->feature_map
)&MD_FEATURE_BITMAP_OFFSET
) {
2122 /* hot-add. allow for actual size of bitmap */
2123 struct bitmap_super_s
*bsb
;
2124 bsb
= (struct bitmap_super_s
*)(((char*)super
)+MAX_SB_SIZE
);
2125 bmspace
= bitmap_sectors(bsb
);
2128 /* Allow space for bad block log */
2129 if (super
->bblog_size
)
2130 bbspace
= __le16_to_cpu(super
->bblog_size
);
2132 if (st
->minor_version
< 0)
2133 /* not specified, so time to set default */
2134 st
->minor_version
= 2;
2136 if (data_offset
== INVALID_SECTORS
)
2137 data_offset
= st
->data_offset
;
2139 if (data_offset
!= INVALID_SECTORS
)
2140 switch(st
->minor_version
) {
2142 return devsize
- data_offset
- 8*2 - bbspace
;
2145 return devsize
- data_offset
;
2152 switch(st
->minor_version
) {
2155 return ((devsize
- 8*2 - bbspace
) & ~(4*2-1));
2157 /* at start, 4K for superblock and possible bitmap */
2158 return devsize
- 4*2 - bbspace
;
2160 /* 4k from start, 4K for superblock and possible bitmap */
2161 return devsize
- (4+4)*2 - bbspace
;
2167 add_internal_bitmap1(struct supertype
*st
,
2168 int *chunkp
, int delay
, int write_behind
,
2169 unsigned long long size
,
2170 int may_change
, int major
)
2173 * If not may_change, then this is a 'Grow' without sysfs support for
2174 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2175 * If may_change, then this is create or a Grow with sysfs syupport,
2176 * and we can put the bitmap wherever we like.
2178 * size is in sectors, chunk is in bytes !!!
2181 unsigned long long bits
;
2182 unsigned long long max_bits
;
2183 unsigned long long min_chunk
;
2185 long bbl_offset
, bbl_size
;
2186 unsigned long long chunk
= *chunkp
;
2189 struct mdp_superblock_1
*sb
= st
->sb
;
2190 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
) + MAX_SB_SIZE
);
2193 if (__le64_to_cpu(sb
->data_size
) == 0)
2194 /* Must be creating the array, else data_size would be non-zero */
2196 switch(st
->minor_version
) {
2198 /* either 3K after the superblock (when hot-add),
2199 * or some amount of space before.
2202 /* We are creating array, so we *know* how much room has
2207 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2209 room
= __le64_to_cpu(sb
->super_offset
)
2210 - __le64_to_cpu(sb
->data_offset
)
2211 - __le64_to_cpu(sb
->data_size
);
2212 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2215 bbl_offset
= (__s32
)__le32_to_cpu(sb
->bblog_offset
);
2216 if (bbl_size
< -bbl_offset
)
2217 bbl_size
= -bbl_offset
;
2219 if (!may_change
|| (room
< 3*2 &&
2220 __le32_to_cpu(sb
->max_dev
) <= 384)) {
2225 offset
= 0; /* means movable offset */
2230 case 2: /* between superblock and data */
2234 room
= choose_bm_space(__le64_to_cpu(sb
->size
)) + bbl_size
;
2236 room
= __le64_to_cpu(sb
->data_offset
)
2237 - __le64_to_cpu(sb
->super_offset
);
2238 bbl_size
= __le16_to_cpu(sb
->bblog_size
);
2240 room
= __le32_to_cpu(sb
->bblog_offset
) + bbl_size
;
2245 room
-= 2; /* Leave 1K for superblock */
2249 room
-= 4*2; /* leave 4K for superblock */
2259 if (chunk
== UnSet
&& room
> 128*2)
2260 /* Limit to 128K of bitmap when chunk size not requested */
2264 /* No room for a bitmap */
2267 max_bits
= (room
* 512 - sizeof(bitmap_super_t
)) * 8;
2269 min_chunk
= 4096; /* sub-page chunks don't work yet.. */
2270 bits
= (size
*512)/min_chunk
+1;
2271 while (bits
> max_bits
) {
2275 if (chunk
== UnSet
) {
2276 /* For practical purpose, 64Meg is a good
2277 * default chunk size for internal bitmaps.
2280 if (chunk
< 64*1024*1024)
2281 chunk
= 64*1024*1024;
2282 } else if (chunk
< min_chunk
)
2283 return 0; /* chunk size too small */
2284 if (chunk
== 0) /* rounding problem */
2288 /* start bitmap on a 4K boundary with enough space for
2291 bits
= (size
*512) / chunk
+ 1;
2292 room
= ((bits
+7)/8 + sizeof(bitmap_super_t
) +4095)/4096;
2293 room
*= 8; /* convert 4K blocks to sectors */
2294 offset
= -room
- bbl_size
;
2297 sb
->bitmap_offset
= (int32_t)__cpu_to_le32(offset
);
2299 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2300 | MD_FEATURE_BITMAP_OFFSET
);
2301 memset(bms
, 0, sizeof(*bms
));
2302 bms
->magic
= __cpu_to_le32(BITMAP_MAGIC
);
2303 bms
->version
= __cpu_to_le32(major
);
2304 uuid_from_super1(st
, uuid
);
2305 memcpy(bms
->uuid
, uuid
, 16);
2306 bms
->chunksize
= __cpu_to_le32(chunk
);
2307 bms
->daemon_sleep
= __cpu_to_le32(delay
);
2308 bms
->sync_size
= __cpu_to_le64(size
);
2309 bms
->write_behind
= __cpu_to_le32(write_behind
);
2310 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2312 sb
->feature_map
= __cpu_to_le32(__le32_to_cpu(sb
->feature_map
)
2313 | MD_FEATURE_BITMAP_VERSIONED
);
2314 if (st
->cluster_name
)
2315 strncpy((char *)bms
->cluster_name
,
2316 st
->cluster_name
, strlen(st
->cluster_name
));
2322 static void locate_bitmap1(struct supertype
*st
, int fd
)
2324 unsigned long long offset
;
2325 struct mdp_superblock_1
*sb
;
2329 if (st
->ss
->load_super(st
, fd
, NULL
))
2330 return; /* no error I hope... */
2335 offset
= __le64_to_cpu(sb
->super_offset
);
2336 offset
+= (int32_t) __le32_to_cpu(sb
->bitmap_offset
);
2339 lseek64(fd
, offset
<<9, 0);
2342 static int write_bitmap1(struct supertype
*st
, int fd
, enum bitmap_update update
)
2344 struct mdp_superblock_1
*sb
= st
->sb
;
2345 bitmap_super_t
*bms
= (bitmap_super_t
*)(((char*)sb
)+MAX_SB_SIZE
);
2349 struct align_fd afd
;
2351 unsigned long long total_bm_space
, bm_space_per_node
;
2355 /* update cluster name */
2356 if (st
->cluster_name
) {
2357 memset((char *)bms
->cluster_name
, 0, sizeof(bms
->cluster_name
));
2358 strncpy((char *)bms
->cluster_name
, st
->cluster_name
, 64);
2362 /* cluster md only supports superblock 1.2 now */
2363 if (st
->minor_version
!= 2) {
2364 pr_err("Warning: cluster md only works with superblock 1.2\n");
2368 /* Each node has an independent bitmap, it is necessary to calculate the
2369 * space is enough or not, first get how many bytes for the total bitmap */
2370 bm_space_per_node
= calc_bitmap_size(bms
, 4096);
2372 total_bm_space
= 512 * (__le64_to_cpu(sb
->data_offset
) - __le64_to_cpu(sb
->super_offset
));
2373 total_bm_space
= total_bm_space
- 4096; /* leave another 4k for superblock */
2375 if (bm_space_per_node
* st
->nodes
> total_bm_space
) {
2376 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2377 total_bm_space
/ bm_space_per_node
);
2381 bms
->nodes
= __cpu_to_le32(st
->nodes
);
2390 locate_bitmap1(st
, fd
);
2392 if (posix_memalign(&buf
, 4096, 4096))
2396 /* Only the bitmap[0] should resync
2397 * whole device on initial assembly
2400 memset(buf
, 0x00, 4096);
2402 memset(buf
, 0xff, 4096);
2403 memcpy(buf
, (char *)bms
, sizeof(bitmap_super_t
));
2405 towrite
= calc_bitmap_size(bms
, 4096);
2406 while (towrite
> 0) {
2410 n
= awrite(&afd
, buf
, n
);
2416 memset(buf
, 0x00, 4096);
2418 memset(buf
, 0xff, 4096);
2425 } while (++i
< __le32_to_cpu(bms
->nodes
));
2431 static void free_super1(struct supertype
*st
)
2434 if (is_clustered(st
)) {
2435 rv
= cluster_get_dlmlock(st
, &lockid
);
2437 pr_err("Cannot get dlmlock in %s return %d\n", __func__
, rv
);
2438 cluster_release_dlmlock(st
, lockid
);
2446 struct devinfo
*di
= st
->info
;
2447 st
->info
= di
->next
;
2453 if (is_clustered(st
))
2454 cluster_release_dlmlock(st
, lockid
);
2458 static int validate_geometry1(struct supertype
*st
, int level
,
2459 int layout
, int raiddisks
,
2460 int *chunk
, unsigned long long size
,
2461 unsigned long long data_offset
,
2462 char *subdev
, unsigned long long *freesize
,
2465 unsigned long long ldsize
, devsize
;
2467 unsigned long long headroom
;
2470 if (level
== LEVEL_CONTAINER
) {
2472 pr_err("1.x metadata does not support containers\n");
2475 if (*chunk
== UnSet
)
2476 *chunk
= DEFAULT_CHUNK
;
2481 if (st
->minor_version
< 0)
2482 /* not specified, so time to set default */
2483 st
->minor_version
= 2;
2485 fd
= open(subdev
, O_RDONLY
|O_EXCL
, 0);
2488 pr_err("super1.x cannot open %s: %s\n",
2489 subdev
, strerror(errno
));
2493 if (!get_dev_size(fd
, subdev
, &ldsize
)) {
2499 devsize
= ldsize
>> 9;
2505 /* creating: allow suitable space for bitmap */
2506 bmspace
= choose_bm_space(devsize
);
2508 if (data_offset
== INVALID_SECTORS
)
2509 data_offset
= st
->data_offset
;
2510 if (data_offset
== INVALID_SECTORS
)
2511 switch (st
->minor_version
) {
2517 /* Choose data offset appropriate for this device
2518 * and use as default for whole array.
2519 * The data_offset must allow for bitmap space
2520 * and base metadata, should allow for some headroom
2521 * for reshape, and should be rounded to multiple
2523 * Headroom is limited to 128M, but aim for about 0.1%
2525 headroom
= 128*1024*2;
2526 while ((headroom
<< 10) > devsize
&&
2528 headroom
/ 2 >= ((unsigned)(*chunk
)*2)*2))
2530 data_offset
= 12*2 + bmspace
+ headroom
;
2531 #define ONE_MEG (2*1024)
2532 if (data_offset
> ONE_MEG
)
2533 data_offset
= (data_offset
/ ONE_MEG
) * ONE_MEG
;
2536 if (st
->data_offset
== INVALID_SECTORS
)
2537 st
->data_offset
= data_offset
;
2538 switch(st
->minor_version
) {
2539 case 0: /* metadata at end. Round down and subtract space to reserve */
2540 devsize
= (devsize
& ~(4ULL*2-1));
2541 /* space for metadata, bblog, bitmap */
2542 devsize
-= 8*2 + 8 + bmspace
;
2546 devsize
-= data_offset
;
2549 *freesize
= devsize
;
2552 #endif /* MDASSEMBLE */
2554 void *super1_make_v0(struct supertype
*st
, struct mdinfo
*info
, mdp_super_t
*sb0
)
2556 /* Create a v1.0 superblock based on 'info'*/
2558 struct mdp_superblock_1
*sb
;
2561 unsigned long long offset
;
2563 if (posix_memalign(&ret
, 4096, 1024) != 0)
2566 memset(ret
, 0, 1024);
2567 sb
->magic
= __cpu_to_le32(MD_SB_MAGIC
);
2568 sb
->major_version
= __cpu_to_le32(1);
2570 copy_uuid(sb
->set_uuid
, info
->uuid
, super1
.swapuuid
);
2571 sprintf(sb
->set_name
, "%d", sb0
->md_minor
);
2572 sb
->ctime
= __cpu_to_le32(info
->array
.ctime
+1);
2573 sb
->level
= __cpu_to_le32(info
->array
.level
);
2574 sb
->layout
= __cpu_to_le32(info
->array
.layout
);
2575 sb
->size
= __cpu_to_le64(info
->component_size
);
2576 sb
->chunksize
= __cpu_to_le32(info
->array
.chunk_size
/512);
2577 sb
->raid_disks
= __cpu_to_le32(info
->array
.raid_disks
);
2578 if (info
->array
.level
> 0)
2579 sb
->data_size
= sb
->size
;
2581 sb
->data_size
= st
->ss
->avail_size(st
, st
->devsize
/512, 0);
2582 sb
->resync_offset
= MaxSector
;
2583 sb
->max_dev
= __cpu_to_le32(MD_SB_DISKS
);
2584 sb
->dev_number
= __cpu_to_le32(info
->disk
.number
);
2585 sb
->utime
= __cpu_to_le64(info
->array
.utime
);
2587 offset
= st
->devsize
/512 - 8*2;
2589 sb
->super_offset
= __cpu_to_le64(offset
);
2590 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2592 if ((rfd
= open("/dev/urandom", O_RDONLY
)) < 0 ||
2593 read(rfd
, sb
->device_uuid
, 16) != 16) {
2594 __u32 r
[4] = {random(), random(), random(), random()};
2595 memcpy(sb
->device_uuid
, r
, 16);
2600 for (i
= 0; i
< MD_SB_DISKS
; i
++) {
2601 int state
= sb0
->disks
[i
].state
;
2602 sb
->dev_roles
[i
] = MD_DISK_ROLE_SPARE
;
2603 if ((state
& (1<<MD_DISK_SYNC
)) &&
2604 !(state
& (1<<MD_DISK_FAULTY
)))
2605 sb
->dev_roles
[i
] = __cpu_to_le16(sb0
->disks
[i
].raid_disk
);
2607 sb
->sb_csum
= calc_sb_1_csum(sb
);
2611 struct superswitch super1
= {
2613 .examine_super
= examine_super1
,
2614 .brief_examine_super
= brief_examine_super1
,
2615 .export_examine_super
= export_examine_super1
,
2616 .detail_super
= detail_super1
,
2617 .brief_detail_super
= brief_detail_super1
,
2618 .export_detail_super
= export_detail_super1
,
2619 .write_init_super
= write_init_super1
,
2620 .validate_geometry
= validate_geometry1
,
2621 .add_to_super
= add_to_super1
,
2622 .examine_badblocks
= examine_badblocks_super1
,
2623 .copy_metadata
= copy_metadata1
,
2625 .match_home
= match_home1
,
2626 .uuid_from_super
= uuid_from_super1
,
2627 .getinfo_super
= getinfo_super1
,
2628 .container_content
= container_content1
,
2629 .update_super
= update_super1
,
2630 .init_super
= init_super1
,
2631 .store_super
= store_super1
,
2632 .compare_super
= compare_super1
,
2633 .load_super
= load_super1
,
2634 .match_metadata_desc
= match_metadata_desc1
,
2635 .avail_size
= avail_size1
,
2636 .add_internal_bitmap
= add_internal_bitmap1
,
2637 .locate_bitmap
= locate_bitmap1
,
2638 .write_bitmap
= write_bitmap1
,
2639 .free_super
= free_super1
,
2640 .require_journal
= require_journal1
,
2641 #if __BYTE_ORDER == BIG_ENDIAN