2 * mdadm - manage Linux "md" devices aka RAID arrays.
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * Email: <neilb@suse.de>
27 * The version-1 superblock :
28 * All numeric fields are little-endian.
30 * total size: 256 bytes plus 2 per device.
31 * 1K allows 384 devices.
33 struct mdp_superblock_1 {
34 /* constant array information - 128 bytes */
35 __u32 magic; /* MD_SB_MAGIC: 0xa92b4efc - little endian */
36 __u32 major_version; /* 1 */
37 __u32 feature_map; /* 0 for now */
38 __u32 pad0; /* always set to 0 when writing */
40 __u8 set_uuid[16]; /* user-space generated. */
41 char set_name[32]; /* set and interpreted by user-space */
43 __u64 ctime; /* lo 40 bits are seconds, top 24 are microseconds or 0*/
44 __u32 level; /* -4 (multipath), -1 (linear), 0,1,4,5 */
45 __u32 layout; /* only for raid5 currently */
46 __u64 size; /* used size of component devices, in 512byte sectors */
48 __u32 chunksize; /* in 512byte sectors */
50 __u32 bitmap_offset; /* sectors after start of superblock that bitmap starts
51 * NOTE: signed, so bitmap can be before superblock
52 * only meaningful of feature_map[0] is set.
55 /* These are only valid with feature bit '4' */
56 __u32 new_level; /* new level we are reshaping to */
57 __u64 reshape_position; /* next address in array-space for reshape */
58 __u32 delta_disks; /* change in number of raid_disks */
59 __u32 new_layout; /* new layout */
60 __u32 new_chunk; /* new chunk size (bytes) */
61 __u32 new_offset; /* signed number to add to data_offset in new
62 * layout. 0 == no-change. This can be
63 * different on each device in the array.
66 /* constant this-device information - 64 bytes */
67 __u64 data_offset; /* sector start of data, often 0 */
68 __u64 data_size; /* sectors in this device that can be used for data */
69 __u64 super_offset; /* sector start of this superblock */
70 __u64 recovery_offset;/* sectors before this offset (from data_offset) have been recovered */
71 __u32 dev_number; /* permanent identifier of this device - not role in raid */
72 __u32 cnt_corrected_read; /* number of read errors that were corrected by re-writing */
73 __u8 device_uuid[16]; /* user-space setable, ignored by kernel */
74 __u8 devflags; /* per-device flags. Only one defined...*/
75 #define WriteMostly1 1 /* mask for writemostly flag in above */
76 /* bad block log. If there are any bad blocks the feature flag is set.
77 * if offset and size are non-zero, that space is reserved and available.
79 __u8 bblog_shift; /* shift from sectors to block size for badblocklist */
80 __u16 bblog_size; /* number of sectors reserved for badblocklist */
81 __u32 bblog_offset; /* sector offset from superblock to bblog, signed */
83 /* array state information - 64 bytes */
84 __u64 utime; /* 40 bits second, 24 btes microseconds */
85 __u64 events; /* incremented when superblock updated */
86 __u64 resync_offset; /* data before this offset (from data_offset) known to be in sync */
87 __u32 sb_csum; /* checksum upto dev_roles[max_dev] */
88 __u32 max_dev; /* size of dev_roles[] array to consider */
89 __u8 pad3[64-32]; /* set to 0 when writing */
91 /* device state information. Indexed by dev_number.
93 * Note there are no per-device state flags. State information is rolled
94 * into the 'roles' value. If a device is spare or faulty, then it doesn't
95 * have a meaningful role.
97 __u16 dev_roles[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
100 #define MAX_SB_SIZE 4096
101 /* bitmap super size is 256, but we round up to a sector for alignment */
102 #define BM_SUPER_SIZE 512
103 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
104 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
105 + sizeof(struct misc_dev_info))
107 struct misc_dev_info {
111 /* feature_map bits */
112 #define MD_FEATURE_BITMAP_OFFSET 1
113 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
116 #define MD_FEATURE_RESHAPE_ACTIVE 4
117 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
118 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
119 * active device with same 'role'.
120 * 'recovery_offset' is also set.
122 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
123 * of devices, but is going
126 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
127 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
128 |MD_FEATURE_RECOVERY_OFFSET \
129 |MD_FEATURE_RESHAPE_ACTIVE \
130 |MD_FEATURE_BAD_BLOCKS \
131 |MD_FEATURE_REPLACEMENT \
132 |MD_FEATURE_RESHAPE_BACKWARDS \
133 |MD_FEATURE_NEW_OFFSET \
137 #define offsetof(t,f) ((size_t)&(((t*)0)->f))
139 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
141 unsigned int disk_csum, csum;
142 unsigned long long newcsum;
143 int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2;
144 unsigned int *isuper = (unsigned int*)sb;
146 /* make sure I can count... */
147 if (offsetof(struct mdp_superblock_1,data_offset) != 128 ||
148 offsetof(struct mdp_superblock_1, utime) != 192 ||
149 sizeof(struct mdp_superblock_1) != 256) {
150 fprintf(stderr, "WARNING - superblock isn't sized correctly\n");
153 disk_csum = sb->sb_csum;
156 for (; size>=4; size -= 4 ) {
157 newcsum += __le32_to_cpu(*isuper);
162 newcsum += __le16_to_cpu(*(unsigned short*) isuper);
164 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
165 sb->sb_csum = disk_csum;
166 return __cpu_to_le32(csum);
170 * Information related to file descriptor used for aligned reads/writes.
171 * Cache the block size.
178 static void init_afd(struct align_fd *afd, int fd)
182 if (ioctl(afd->fd, BLKSSZGET, &afd->blk_sz) != 0)
186 static char abuf[4096+4096];
187 static int aread(struct align_fd *afd, void *buf, int len)
190 * On devices with a 4K sector size, we need to read
191 * the full sector and copy relevant bits into
200 if (!bsize || bsize > 4096 || len > 4096) {
202 fprintf(stderr, "WARNING - aread() called with "
203 "invalid block size\n");
206 b = ROUND_UP_PTR((char *)abuf, 4096);
208 for (iosize = 0; iosize < len; iosize += bsize)
210 n = read(afd->fd, b, iosize);
213 lseek(afd->fd, len - n, 1);
220 static int awrite(struct align_fd *afd, void *buf, int len)
223 * On devices with a 4K sector size, we need to write
224 * the full sector. We pre-read if the sector is larger
226 * The address must be sector-aligned.
233 if (!bsize || bsize > 4096 || len > 4096) {
235 fprintf(stderr, "WARNING - awrite() called with "
236 "invalid block size\n");
239 b = ROUND_UP_PTR((char *)abuf, 4096);
241 for (iosize = 0; iosize < len ; iosize += bsize)
245 n = read(afd->fd, b, iosize);
248 lseek(afd->fd, -n, 1);
252 n = write(afd->fd, b, iosize);
255 lseek(afd->fd, len - n, 1);
260 static void examine_super1(struct supertype *st, char *homehost)
262 struct mdp_superblock_1 *sb = st->sb;
269 int l = homehost ? strlen(homehost) : 0;
271 unsigned long long sb_offset;
273 printf(" Magic : %08x\n", __le32_to_cpu(sb->magic));
274 printf(" Version : 1");
275 sb_offset = __le64_to_cpu(sb->super_offset);
278 else if (sb_offset <= 8)
282 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb->feature_map));
283 printf(" Array UUID : ");
284 for (i=0; i<16; i++) {
285 if ((i&3)==0 && i != 0) printf(":");
286 printf("%02x", sb->set_uuid[i]);
289 printf(" Name : %.32s", sb->set_name);
290 if (l > 0 && l < 32 &&
291 sb->set_name[l] == ':' &&
292 strncmp(sb->set_name, homehost, l) == 0)
293 printf(" (local to host %s)", homehost);
295 atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL;
296 printf(" Creation Time : %.24s\n", ctime(&atime));
297 c=map_num(pers, __le32_to_cpu(sb->level));
298 printf(" Raid Level : %s\n", c?c:"-unknown-");
299 printf(" Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks));
301 printf(" Avail Dev Size : %llu%s\n",
302 (unsigned long long)__le64_to_cpu(sb->data_size),
303 human_size(__le64_to_cpu(sb->data_size)<<9));
304 if (__le32_to_cpu(sb->level) > 0) {
305 int ddsks = 0, ddsks_denom = 1;
306 switch(__le32_to_cpu(sb->level)) {
307 case 1: ddsks=1;break;
309 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
310 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
312 layout = __le32_to_cpu(sb->layout);
313 ddsks = __le32_to_cpu(sb->raid_disks);
314 ddsks_denom = (layout&255) * ((layout>>8)&255);
317 long long asize = __le64_to_cpu(sb->size);
318 asize = (asize << 9) * ddsks / ddsks_denom;
319 printf(" Array Size : %llu%s\n",
320 asize >> 10, human_size(asize));
322 if (sb->size != sb->data_size)
323 printf(" Used Dev Size : %llu%s\n",
324 (unsigned long long)__le64_to_cpu(sb->size),
325 human_size(__le64_to_cpu(sb->size)<<9));
328 printf(" Data Offset : %llu sectors\n",
329 (unsigned long long)__le64_to_cpu(sb->data_offset));
330 if (sb->new_offset) {
331 unsigned long long offset = __le64_to_cpu(sb->data_offset);
332 offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset);
333 printf(" New Offset : %llu sectors\n", offset);
335 printf(" Super Offset : %llu sectors\n",
336 (unsigned long long)__le64_to_cpu(sb->super_offset));
337 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)
338 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb->recovery_offset));
339 printf(" State : %s\n", (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean");
340 printf(" Device UUID : ");
341 for (i=0; i<16; i++) {
342 if ((i&3)==0 && i != 0) printf(":");
343 printf("%02x", sb->device_uuid[i]);
347 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
348 printf("Internal Bitmap : %ld sectors from superblock\n",
349 (long)(int32_t)__le32_to_cpu(sb->bitmap_offset));
351 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)) {
352 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb->reshape_position)/2,
353 human_size(__le64_to_cpu(sb->reshape_position)<<9));
354 if (__le32_to_cpu(sb->delta_disks)) {
355 printf(" Delta Devices : %d", __le32_to_cpu(sb->delta_disks));
356 printf(" (%d->%d)\n",
357 __le32_to_cpu(sb->raid_disks)-__le32_to_cpu(sb->delta_disks),
358 __le32_to_cpu(sb->raid_disks));
359 if ((int)__le32_to_cpu(sb->delta_disks) < 0)
360 delta_extra = -__le32_to_cpu(sb->delta_disks);
362 if (__le32_to_cpu(sb->new_level) != __le32_to_cpu(sb->level)) {
363 c = map_num(pers, __le32_to_cpu(sb->new_level));
364 printf(" New Level : %s\n", c?c:"-unknown-");
366 if (__le32_to_cpu(sb->new_layout) != __le32_to_cpu(sb->layout)) {
367 if (__le32_to_cpu(sb->level) == 5) {
368 c = map_num(r5layout, __le32_to_cpu(sb->new_layout));
369 printf(" New Layout : %s\n", c?c:"-unknown-");
371 if (__le32_to_cpu(sb->level) == 6) {
372 c = map_num(r6layout, __le32_to_cpu(sb->new_layout));
373 printf(" New Layout : %s\n", c?c:"-unknown-");
375 if (__le32_to_cpu(sb->level) == 10) {
376 printf(" New Layout :");
377 print_r10_layout(__le32_to_cpu(sb->new_layout));
381 if (__le32_to_cpu(sb->new_chunk) != __le32_to_cpu(sb->chunksize))
382 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb->new_chunk)/2);
387 if (sb->devflags & WriteMostly1)
388 printf(" write-mostly");
392 atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL;
393 printf(" Update Time : %.24s\n", ctime(&atime));
395 if (sb->bblog_size && sb->bblog_offset) {
396 printf(" Bad Block Log : %d entries available at offset %ld sectors",
397 __le16_to_cpu(sb->bblog_size)*512/8,
398 (long)__le32_to_cpu(sb->bblog_offset));
399 if (sb->feature_map &
400 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
401 printf(" - bad blocks present.");
406 if (calc_sb_1_csum(sb) == sb->sb_csum)
407 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb->sb_csum));
409 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb->sb_csum),
410 __le32_to_cpu(calc_sb_1_csum(sb)));
411 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb->events));
413 if (__le32_to_cpu(sb->level) == 5) {
414 c = map_num(r5layout, __le32_to_cpu(sb->layout));
415 printf(" Layout : %s\n", c?c:"-unknown-");
417 if (__le32_to_cpu(sb->level) == 6) {
418 c = map_num(r6layout, __le32_to_cpu(sb->layout));
419 printf(" Layout : %s\n", c?c:"-unknown-");
421 if (__le32_to_cpu(sb->level) == 10) {
422 int lo = __le32_to_cpu(sb->layout);
424 print_r10_layout(lo);
427 switch(__le32_to_cpu(sb->level)) {
433 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb->chunksize)/2);
436 printf(" Rounding : %dK\n", __le32_to_cpu(sb->chunksize)/2);
442 /* This turns out to just be confusing */
443 printf(" Array Slot : %d (", __le32_to_cpu(sb->dev_number));
444 for (i= __le32_to_cpu(sb->max_dev); i> 0 ; i--)
445 if (__le16_to_cpu(sb->dev_roles[i-1]) != 0xffff)
447 for (d=0; d < i; d++) {
448 int role = __le16_to_cpu(sb->dev_roles[d]);
450 if (role == 0xffff) printf("empty");
451 else if(role == 0xfffe) printf("failed");
452 else printf("%d", role);
456 printf(" Device Role : ");
457 d = __le32_to_cpu(sb->dev_number);
458 if (d < __le32_to_cpu(sb->max_dev))
459 role = __le16_to_cpu(sb->dev_roles[d]);
464 else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_REPLACEMENT))
465 printf("Replacement device %d\n", role);
467 printf("Active device %d\n", role);
469 printf(" Array State : ");
470 for (d=0; d<__le32_to_cpu(sb->raid_disks) + delta_extra; d++) {
473 for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
474 unsigned int role = __le16_to_cpu(sb->dev_roles[i]);
488 /* This is confusing too */
490 for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
491 int role = __le16_to_cpu(sb->dev_roles[i]);
495 if (faulty) printf(" %d failed", faulty);
497 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
502 static void brief_examine_super1(struct supertype *st, int verbose)
504 struct mdp_superblock_1 *sb = st->sb;
506 unsigned long long sb_offset;
508 char *c=map_num(pers, __le32_to_cpu(sb->level));
510 nm = strchr(sb->set_name, ':');
513 else if (sb->set_name[0])
525 printf(" level=%s", c);
526 sb_offset = __le64_to_cpu(sb->super_offset);
528 printf(" metadata=1.1 ");
529 else if (sb_offset <= 8)
530 printf(" metadata=1.2 ");
532 printf(" metadata=1.0 ");
534 printf("num-devices=%d ", __le32_to_cpu(sb->raid_disks));
536 for (i=0; i<16; i++) {
537 if ((i&3)==0 && i != 0) printf(":");
538 printf("%02x", sb->set_uuid[i]);
540 if (sb->set_name[0]) {
542 print_quoted(sb->set_name);
547 static void export_examine_super1(struct supertype *st)
549 struct mdp_superblock_1 *sb = st->sb;
554 printf("MD_LEVEL=%s\n", map_num(pers, __le32_to_cpu(sb->level)));
555 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb->raid_disks));
557 if (sb->set_name[i] == '\n' ||
558 sb->set_name[i] == '\0') {
563 printf("MD_NAME=%.*s\n", len, sb->set_name);
564 if (__le32_to_cpu(sb->level) > 0) {
565 int ddsks = 0, ddsks_denom = 1;
566 switch(__le32_to_cpu(sb->level)) {
567 case 1: ddsks=1;break;
569 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
570 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
572 layout = __le32_to_cpu(sb->layout);
573 ddsks = __le32_to_cpu(sb->raid_disks);
574 ddsks_denom = (layout&255) * ((layout>>8)&255);
577 long long asize = __le64_to_cpu(sb->size);
578 asize = (asize << 9) * ddsks / ddsks_denom;
579 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize,JEDEC));
583 for (i=0; i<16; i++) {
584 if ((i&3)==0 && i != 0) printf(":");
585 printf("%02x", sb->set_uuid[i]);
588 printf("MD_UPDATE_TIME=%llu\n",
589 __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL);
590 printf("MD_DEV_UUID=");
591 for (i=0; i<16; i++) {
592 if ((i&3)==0 && i != 0) printf(":");
593 printf("%02x", sb->device_uuid[i]);
596 printf("MD_EVENTS=%llu\n",
597 (unsigned long long)__le64_to_cpu(sb->events));
600 static void detail_super1(struct supertype *st, char *homehost)
602 struct mdp_superblock_1 *sb = st->sb;
604 int l = homehost ? strlen(homehost) : 0;
606 printf(" Name : %.32s", sb->set_name);
607 if (l > 0 && l < 32 &&
608 sb->set_name[l] == ':' &&
609 strncmp(sb->set_name, homehost, l) == 0)
610 printf(" (local to host %s)", homehost);
611 printf("\n UUID : ");
612 for (i=0; i<16; i++) {
613 if ((i&3)==0 && i != 0) printf(":");
614 printf("%02x", sb->set_uuid[i]);
616 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb->events));
619 static void brief_detail_super1(struct supertype *st)
621 struct mdp_superblock_1 *sb = st->sb;
624 if (sb->set_name[0]) {
626 print_quoted(sb->set_name);
629 for (i=0; i<16; i++) {
630 if ((i&3)==0 && i != 0) printf(":");
631 printf("%02x", sb->set_uuid[i]);
635 static void export_detail_super1(struct supertype *st)
637 struct mdp_superblock_1 *sb = st->sb;
642 if (sb->set_name[i] == '\n' ||
643 sb->set_name[i] == '\0') {
648 printf("MD_NAME=%.*s\n", len, sb->set_name);
651 static int examine_badblocks_super1(struct supertype *st, int fd, char *devname)
653 struct mdp_superblock_1 *sb = st->sb;
654 unsigned long long offset;
659 if (!sb->bblog_size || __le32_to_cpu(sb->bblog_size) > 100
660 || !sb->bblog_offset){
661 printf("No bad-blocks list configured on %s\n", devname);
664 if ((sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
666 printf("Bad-blocks list is empty in %s\n", devname);
670 size = __le32_to_cpu(sb->bblog_size)* 512;
671 posix_memalign((void**)&bbl, 4096, size);
672 offset = __le64_to_cpu(sb->super_offset) +
673 (int)__le32_to_cpu(sb->bblog_offset);
675 if (lseek64(fd, offset, 0) < 0) {
676 pr_err("Cannot seek to bad-blocks list\n");
679 if (read(fd, bbl, size) != size) {
680 pr_err("Cannot read bad-blocks list\n");
683 /* 64bits per entry. 10 bits is block-count, 54 bits is block
684 * offset. Blocks are sectors unless bblog->shift makes them bigger
687 printf("Bad-blocks on %s:\n", devname);
688 for (i = 0; i < size/8; i++, bbp++) {
689 __u64 bb = __le64_to_cpu(*bbp);
690 int count = bb & 0x3ff;
691 unsigned long long sector = bb >> 10;
696 sector <<= sb->bblog_shift;
697 count <<= sb->bblog_shift;
699 printf("%20llu for %d sectors\n", sector, count);
706 static int match_home1(struct supertype *st, char *homehost)
708 struct mdp_superblock_1 *sb = st->sb;
709 int l = homehost ? strlen(homehost) : 0;
711 return (l > 0 && l < 32 &&
712 sb->set_name[l] == ':' &&
713 strncmp(sb->set_name, homehost, l) == 0);
716 static void uuid_from_super1(struct supertype *st, int uuid[4])
718 struct mdp_superblock_1 *super = st->sb;
719 char *cuuid = (char*)uuid;
722 cuuid[i] = super->set_uuid[i];
725 static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
727 struct mdp_superblock_1 *sb = st->sb;
728 struct bitmap_super_s *bsb = (void*)(((char*)sb)+MAX_SB_SIZE);
729 struct misc_dev_info *misc = (void*)(((char*)sb)+MAX_SB_SIZE+BM_SUPER_SIZE);
733 unsigned int map_disks = info->array.raid_disks;
734 unsigned long long super_offset;
735 unsigned long long data_size;
737 memset(info, 0, sizeof(*info));
738 info->array.major_version = 1;
739 info->array.minor_version = st->minor_version;
740 info->array.patch_version = 0;
741 info->array.raid_disks = __le32_to_cpu(sb->raid_disks);
742 info->array.level = __le32_to_cpu(sb->level);
743 info->array.layout = __le32_to_cpu(sb->layout);
744 info->array.md_minor = -1;
745 info->array.ctime = __le64_to_cpu(sb->ctime);
746 info->array.utime = __le64_to_cpu(sb->utime);
747 info->array.chunk_size = __le32_to_cpu(sb->chunksize)*512;
749 (__le64_to_cpu(sb->resync_offset) == MaxSector)
752 info->data_offset = __le64_to_cpu(sb->data_offset);
753 info->component_size = __le64_to_cpu(sb->size);
754 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET))
755 info->bitmap_offset = (int32_t)__le32_to_cpu(sb->bitmap_offset);
757 info->disk.major = 0;
758 info->disk.minor = 0;
759 info->disk.number = __le32_to_cpu(sb->dev_number);
760 if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) ||
761 __le32_to_cpu(sb->dev_number) >= MAX_DEVS)
764 role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]);
766 super_offset = __le64_to_cpu(sb->super_offset);
767 data_size = __le64_to_cpu(sb->size);
768 if (info->data_offset < super_offset) {
769 unsigned long long end;
770 info->space_before = info->data_offset;
772 if (info->bitmap_offset < 0)
773 end += info->bitmap_offset;
774 if (info->data_offset + data_size < end)
775 info->space_after = end - data_size - info->data_offset;
777 info->space_after = 0;
779 info->space_before = (info->data_offset -
781 if (info->bitmap_offset > 0) {
782 unsigned long long bmend = info->bitmap_offset;
783 unsigned long long size = __le64_to_cpu(bsb->sync_size);
784 size /= __le32_to_cpu(bsb->chunksize) >> 9;
785 size = (size + 7) >> 3;
786 size += sizeof(bitmap_super_t);
787 size = ROUND_UP(size, 4096);
790 if (size < info->space_before)
791 info->space_before -= size;
793 info->space_before = 0;
795 info->space_before -= 8; /* superblock */
796 info->space_after = misc->device_size - data_size - info->data_offset;
799 info->disk.raid_disk = -1;
802 info->disk.state = 0; /* spare: not active, not sync, not faulty */
805 info->disk.state = 1; /* faulty */
808 info->disk.state = 6; /* active and in sync */
809 info->disk.raid_disk = role;
811 if (sb->devflags & WriteMostly1)
812 info->disk.state |= (1 << MD_DISK_WRITEMOSTLY);
813 info->events = __le64_to_cpu(sb->events);
814 sprintf(info->text_version, "1.%d", st->minor_version);
815 info->safe_mode_delay = 200;
817 memcpy(info->uuid, sb->set_uuid, 16);
819 strncpy(info->name, sb->set_name, 32);
822 if ((__le32_to_cpu(sb->feature_map)&MD_FEATURE_REPLACEMENT)) {
823 info->disk.state &= ~(1 << MD_DISK_SYNC);
824 info->disk.state |= 1 << MD_DISK_REPLACEMENT;
828 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET))
829 info->recovery_start = __le32_to_cpu(sb->recovery_offset);
831 info->recovery_start = MaxSector;
833 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) {
834 info->reshape_active = 1;
835 if (info->array.level == 10)
836 info->reshape_active |= RESHAPE_NO_BACKUP;
837 info->reshape_progress = __le64_to_cpu(sb->reshape_position);
838 info->new_level = __le32_to_cpu(sb->new_level);
839 info->delta_disks = __le32_to_cpu(sb->delta_disks);
840 info->new_layout = __le32_to_cpu(sb->new_layout);
841 info->new_chunk = __le32_to_cpu(sb->new_chunk)<<9;
842 if (info->delta_disks < 0)
843 info->array.raid_disks -= info->delta_disks;
845 info->reshape_active = 0;
847 info->recovery_blocked = info->reshape_active;
850 for (i=0; i<map_disks; i++)
852 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
853 role = __le16_to_cpu(sb->dev_roles[i]);
854 if (/*role == 0xFFFF || */role < (unsigned) info->array.raid_disks) {
856 if (map && role < map_disks)
861 info->array.working_disks = working;
864 static struct mdinfo *container_content1(struct supertype *st, char *subarray)
871 info = xmalloc(sizeof(*info));
872 getinfo_super1(st, info, NULL);
876 static int update_super1(struct supertype *st, struct mdinfo *info,
878 char *devname, int verbose,
879 int uuid_set, char *homehost)
881 /* NOTE: for 'assemble' and 'force' we need to return non-zero
882 * if any change was made. For others, the return value is
886 struct mdp_superblock_1 *sb = st->sb;
888 if (strcmp(update, "homehost") == 0 &&
890 /* Note that 'homehost' is special as it is really
895 c = strchr(sb->set_name, ':');
897 strncpy(info->name, c+1, 31 - (c-sb->set_name));
899 strncpy(info->name, sb->set_name, 32);
903 if (strcmp(update, "force-one")==0) {
904 /* Not enough devices for a working array,
905 * so bring this one up-to-date
907 if (sb->events != __cpu_to_le64(info->events))
909 sb->events = __cpu_to_le64(info->events);
910 } else if (strcmp(update, "force-array")==0) {
911 /* Degraded array and 'force' requests to
912 * maybe need to mark it 'clean'.
914 switch(__le32_to_cpu(sb->level)) {
915 case 5: case 4: case 6:
916 /* need to force clean */
917 if (sb->resync_offset != MaxSector)
919 sb->resync_offset = MaxSector;
921 } else if (strcmp(update, "assemble")==0) {
922 int d = info->disk.number;
924 if (info->disk.state & (1<<MD_DISK_ACTIVE))
925 want = info->disk.raid_disk;
928 if (sb->dev_roles[d] != __cpu_to_le16(want)) {
929 sb->dev_roles[d] = __cpu_to_le16(want);
932 if (info->reshape_active &&
933 sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
934 info->delta_disks >= 0 &&
935 info->reshape_progress < __le64_to_cpu(sb->reshape_position)) {
936 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
939 if (info->reshape_active &&
940 sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
941 info->delta_disks < 0 &&
942 info->reshape_progress > __le64_to_cpu(sb->reshape_position)) {
943 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
946 } else if (strcmp(update, "linear-grow-new") == 0) {
949 unsigned int max = __le32_to_cpu(sb->max_dev);
951 for (i=0 ; i < max ; i++)
952 if (__le16_to_cpu(sb->dev_roles[i]) >= 0xfffe)
954 sb->dev_number = __cpu_to_le32(i);
955 info->disk.number = i;
956 if (max >= __le32_to_cpu(sb->max_dev))
957 sb->max_dev = __cpu_to_le32(max+1);
959 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
960 read(rfd, sb->device_uuid, 16) != 16) {
961 __u32 r[4] = {random(), random(), random(), random()};
962 memcpy(sb->device_uuid, r, 16);
968 __cpu_to_le16(info->disk.raid_disk);
970 fd = open(devname, O_RDONLY);
972 unsigned long long ds;
973 get_dev_size(fd, devname, &ds);
976 if (__le64_to_cpu(sb->super_offset) <
977 __le64_to_cpu(sb->data_offset)) {
978 sb->data_size = __cpu_to_le64(
979 ds - __le64_to_cpu(sb->data_offset));
982 ds &= ~(unsigned long long)(4*2-1);
983 sb->super_offset = __cpu_to_le64(ds);
984 sb->data_size = __cpu_to_le64(
985 ds - __le64_to_cpu(sb->data_offset));
988 } else if (strcmp(update, "linear-grow-update") == 0) {
989 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
990 sb->dev_roles[info->disk.number] =
991 __cpu_to_le16(info->disk.raid_disk);
992 } else if (strcmp(update, "resync") == 0) {
993 /* make sure resync happens */
994 sb->resync_offset = 0ULL;
995 } else if (strcmp(update, "uuid") == 0) {
996 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
998 if (__le32_to_cpu(sb->feature_map)&MD_FEATURE_BITMAP_OFFSET) {
999 struct bitmap_super_s *bm;
1000 bm = (struct bitmap_super_s*)(st->sb+MAX_SB_SIZE);
1001 memcpy(bm->uuid, sb->set_uuid, 16);
1003 } else if (strcmp(update, "no-bitmap") == 0) {
1004 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1005 } else if (strcmp(update, "bbl") == 0) {
1006 /* only possible if there is room after the bitmap, or if
1007 * there is no bitmap
1009 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1010 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1011 long bitmap_offset = (long)__le64_to_cpu(sb->bitmap_offset);
1012 long bm_sectors = 0;
1015 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1016 struct bitmap_super_s *bsb;
1017 bsb = (struct bitmap_super_s *)(((char*)sb)+MAX_SB_SIZE);
1018 bm_sectors = bitmap_sectors(bsb);
1021 if (sb_offset < data_offset) {
1022 /* 1.1 or 1.2. Put bbl just before data
1025 space = data_offset - sb_offset;
1026 bb_offset = space - 8;
1027 if (bm_sectors && bitmap_offset > 0)
1028 space -= (bitmap_offset + bm_sectors);
1030 space -= 8; /* The superblock */
1032 sb->bblog_size = __cpu_to_le16(8);
1033 sb->bblog_offset = __cpu_to_le32(bb_offset);
1036 /* 1.0 - Put bbl just before super block */
1037 if (bm_sectors && bitmap_offset < 0)
1038 space = -bitmap_offset - bm_sectors;
1040 space = sb_offset - data_offset -
1041 __le64_to_cpu(sb->data_size);
1043 sb->bblog_size = __cpu_to_le16(8);
1044 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1047 } else if (strcmp(update, "no-bbl") == 0) {
1048 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
1049 pr_err("Cannot remove active bbl from %s\n",devname);
1052 sb->bblog_shift = 0;
1053 sb->bblog_offset = 0;
1055 } else if (strcmp(update, "name") == 0) {
1056 if (info->name[0] == 0)
1057 sprintf(info->name, "%d", info->array.md_minor);
1058 memset(sb->set_name, 0, sizeof(sb->set_name));
1060 strchr(info->name, ':') == NULL &&
1061 strlen(homehost)+1+strlen(info->name) < 32) {
1062 strcpy(sb->set_name, homehost);
1063 strcat(sb->set_name, ":");
1064 strcat(sb->set_name, info->name);
1066 strcpy(sb->set_name, info->name);
1067 } else if (strcmp(update, "devicesize") == 0 &&
1068 __le64_to_cpu(sb->super_offset) <
1069 __le64_to_cpu(sb->data_offset)) {
1070 /* set data_size to device size less data_offset */
1071 struct misc_dev_info *misc = (struct misc_dev_info*)
1072 (st->sb + MAX_SB_SIZE + BM_SUPER_SIZE);
1073 printf("Size was %llu\n", (unsigned long long)
1074 __le64_to_cpu(sb->data_size));
1075 sb->data_size = __cpu_to_le64(
1076 misc->device_size - __le64_to_cpu(sb->data_offset));
1077 printf("Size is %llu\n", (unsigned long long)
1078 __le64_to_cpu(sb->data_size));
1079 } else if (strcmp(update, "_reshape_progress")==0)
1080 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1081 else if (strcmp(update, "writemostly")==0)
1082 sb->devflags |= WriteMostly1;
1083 else if (strcmp(update, "readwrite")==0)
1084 sb->devflags &= ~WriteMostly1;
1088 sb->sb_csum = calc_sb_1_csum(sb);
1092 static int init_super1(struct supertype *st, mdu_array_info_t *info,
1093 unsigned long long size, char *name, char *homehost,
1094 int *uuid, unsigned long long data_offset)
1096 struct mdp_superblock_1 *sb;
1102 if (posix_memalign((void**)&sb, 4096, SUPER1_SIZE) != 0) {
1103 pr_err("%s could not allocate superblock\n", __func__);
1106 memset(sb, 0, SUPER1_SIZE);
1110 /* zeroing superblock */
1114 spares = info->working_disks - info->active_disks;
1115 if (info->raid_disks + spares > MAX_DEVS) {
1116 pr_err("too many devices requested: %d+%d > %d\n",
1117 info->raid_disks , spares, MAX_DEVS);
1121 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
1122 sb->major_version = __cpu_to_le32(1);
1123 sb->feature_map = 0;
1127 copy_uuid(sb->set_uuid, uuid, super1.swapuuid);
1129 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
1130 read(rfd, sb->set_uuid, 16) != 16) {
1131 __u32 r[4] = {random(), random(), random(), random()};
1132 memcpy(sb->set_uuid, r, 16);
1134 if (rfd >= 0) close(rfd);
1137 if (name == NULL || *name == 0) {
1138 sprintf(defname, "%d", info->md_minor);
1142 strchr(name, ':')== NULL &&
1143 strlen(homehost)+1+strlen(name) < 32) {
1144 strcpy(sb->set_name, homehost);
1145 strcat(sb->set_name, ":");
1146 strcat(sb->set_name, name);
1148 strcpy(sb->set_name, name);
1150 sb->ctime = __cpu_to_le64((unsigned long long)time(0));
1151 sb->level = __cpu_to_le32(info->level);
1152 sb->layout = __cpu_to_le32(info->layout);
1153 sb->size = __cpu_to_le64(size*2ULL);
1154 sb->chunksize = __cpu_to_le32(info->chunk_size>>9);
1155 sb->raid_disks = __cpu_to_le32(info->raid_disks);
1157 sb->data_offset = __cpu_to_le64(data_offset);
1158 sb->data_size = __cpu_to_le64(0);
1159 sb->super_offset = __cpu_to_le64(0);
1160 sb->recovery_offset = __cpu_to_le64(0);
1162 sb->utime = sb->ctime;
1163 sb->events = __cpu_to_le64(1);
1164 if (info->state & (1<<MD_SB_CLEAN))
1165 sb->resync_offset = MaxSector;
1167 sb->resync_offset = 0;
1168 sbsize = sizeof(struct mdp_superblock_1) + 2 * (info->raid_disks + spares);
1169 sbsize = ROUND_UP(sbsize, 512);
1170 sb->max_dev = __cpu_to_le32((sbsize - sizeof(struct mdp_superblock_1)) / 2);
1172 memset(sb->dev_roles, 0xff, MAX_SB_SIZE - sizeof(struct mdp_superblock_1));
1180 long long data_offset;
1181 mdu_disk_info_t disk;
1182 struct devinfo *next;
1185 /* Add a device to the superblock being created */
1186 static int add_to_super1(struct supertype *st, mdu_disk_info_t *dk,
1187 int fd, char *devname, unsigned long long data_offset)
1189 struct mdp_superblock_1 *sb = st->sb;
1190 __u16 *rp = sb->dev_roles + dk->number;
1191 struct devinfo *di, **dip;
1193 if ((dk->state & 6) == 6) /* active, sync */
1194 *rp = __cpu_to_le16(dk->raid_disk);
1195 else if ((dk->state & ~2) == 0) /* active or idle -> spare */
1200 if (dk->number >= (int)__le32_to_cpu(sb->max_dev) &&
1201 __le32_to_cpu(sb->max_dev) < MAX_DEVS)
1202 sb->max_dev = __cpu_to_le32(dk->number+1);
1204 sb->dev_number = __cpu_to_le32(dk->number);
1205 sb->devflags = 0; /* don't copy another disks flags */
1206 sb->sb_csum = calc_sb_1_csum(sb);
1208 dip = (struct devinfo **)&st->info;
1210 dip = &(*dip)->next;
1211 di = xmalloc(sizeof(struct devinfo));
1213 di->devname = devname;
1215 di->data_offset = data_offset;
1223 static void locate_bitmap1(struct supertype *st, int fd);
1225 static int store_super1(struct supertype *st, int fd)
1227 struct mdp_superblock_1 *sb = st->sb;
1228 unsigned long long sb_offset;
1229 struct align_fd afd;
1231 unsigned long long dsize;
1233 if (!get_dev_size(fd, NULL, &dsize))
1244 * Calculate the position of the superblock.
1245 * It is always aligned to a 4K boundary and
1246 * depending on minor_version, it can be:
1247 * 0: At least 8K, but less than 12K, from end of device
1248 * 1: At start of device
1249 * 2: 4K from start of device.
1251 switch(st->minor_version) {
1255 sb_offset &= ~(4*2-1);
1269 if (sb_offset != __le64_to_cpu(sb->super_offset) &&
1270 0 != __le64_to_cpu(sb->super_offset)
1272 pr_err("internal error - sb_offset is wrong\n");
1276 if (lseek64(fd, sb_offset << 9, 0)< 0LL)
1279 sbsize = ROUND_UP(sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev), 512);
1281 if (awrite(&afd, sb, sbsize) != sbsize)
1284 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1285 struct bitmap_super_s *bm = (struct bitmap_super_s*)
1286 (((char*)sb)+MAX_SB_SIZE);
1287 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC) {
1288 locate_bitmap1(st, fd);
1289 if (awrite(&afd, bm, sizeof(*bm)) != sizeof(*bm))
1297 static int load_super1(struct supertype *st, int fd, char *devname);
1299 static unsigned long choose_bm_space(unsigned long devsize)
1301 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1302 * if bigger than 200Gig, save 128k
1303 * NOTE: result must be multiple of 4K else bad things happen
1304 * on 4K-sector devices.
1306 if (devsize < 64*2) return 0;
1307 if (devsize - 64*2 >= 200*1024*1024*2)
1309 if (devsize - 4*2 > 8*1024*1024*2)
1314 static void free_super1(struct supertype *st);
1317 static int write_init_super1(struct supertype *st)
1319 struct mdp_superblock_1 *sb = st->sb;
1320 struct supertype *refst;
1323 unsigned long long bm_space;
1324 unsigned long long reserved;
1326 unsigned long long dsize, array_size;
1327 unsigned long long sb_offset, headroom;
1328 unsigned long long data_offset;
1330 for (di = st->info; di; di = di->next) {
1331 if (di->disk.state == 1)
1336 while (Kill(di->devname, NULL, 0, -1, 1) == 0)
1339 sb->dev_number = __cpu_to_le32(di->disk.number);
1340 if (di->disk.state & (1<<MD_DISK_WRITEMOSTLY))
1341 sb->devflags |= WriteMostly1;
1343 sb->devflags &= ~WriteMostly1;
1345 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
1346 read(rfd, sb->device_uuid, 16) != 16) {
1347 __u32 r[4] = {random(), random(), random(), random()};
1348 memcpy(sb->device_uuid, r, 16);
1355 refst = dup_super(st);
1356 if (load_super1(refst, di->fd, NULL)==0) {
1357 struct mdp_superblock_1 *refsb = refst->sb;
1359 memcpy(sb->device_uuid, refsb->device_uuid, 16);
1360 if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) {
1361 /* same array, so preserve events and
1363 sb->events = refsb->events;
1364 /* bugs in 2.6.17 and earlier mean the
1365 * dev_number chosen in Manage must be preserved
1367 if (get_linux_version() >= 2006018)
1368 sb->dev_number = refsb->dev_number;
1374 if (!get_dev_size(di->fd, NULL, &dsize)) {
1388 * Calculate the position of the superblock.
1389 * It is always aligned to a 4K boundary and
1390 * depending on minor_version, it can be:
1391 * 0: At least 8K, but less than 12K, from end of device
1392 * 1: At start of device
1393 * 2: 4K from start of device.
1394 * Depending on the array size, we might leave extra space
1396 * Also leave 4K for bad-block log.
1398 array_size = __le64_to_cpu(sb->size);
1399 /* work out how much space we left for a bitmap,
1400 * Add 8 sectors for bad block log */
1401 bm_space = choose_bm_space(array_size) + 8;
1403 /* We try to leave 0.1% at the start for reshape
1404 * operations, but limit this to 128Meg (0.1% of 10Gig)
1405 * which is plenty for efficient reshapes
1406 * However we make it at least 2 chunks as one chunk
1407 * is minimum needed for reshape.
1409 headroom = 128 * 1024 * 2;
1410 while (headroom << 10 > array_size &&
1411 headroom/2 >= __le32_to_cpu(sb->chunksize) * 2)
1414 data_offset = di->data_offset;
1415 switch(st->minor_version) {
1419 sb_offset &= ~(4*2-1);
1420 sb->super_offset = __cpu_to_le64(sb_offset);
1421 if (data_offset == INVALID_SECTORS)
1422 sb->data_offset = 0;
1423 if (sb_offset < array_size + bm_space)
1424 bm_space = sb_offset - array_size;
1425 sb->data_size = __cpu_to_le64(sb_offset - bm_space);
1426 if (bm_space >= 8) {
1427 sb->bblog_size = __cpu_to_le16(8);
1428 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1432 sb->super_offset = __cpu_to_le64(0);
1433 if (data_offset == INVALID_SECTORS) {
1434 reserved = bm_space + 4*2;
1435 if (reserved < headroom)
1436 reserved = headroom;
1437 if (reserved + array_size > dsize)
1438 reserved = dsize - array_size;
1439 /* Try for multiple of 1Meg so it is nicely aligned */
1440 #define ONE_MEG (2*1024)
1441 if (reserved > ONE_MEG)
1442 reserved = (reserved/ONE_MEG) * ONE_MEG;
1444 /* force 4K alignment */
1448 reserved = data_offset;
1450 sb->data_offset = __cpu_to_le64(reserved);
1451 sb->data_size = __cpu_to_le64(dsize - reserved);
1452 if (reserved >= 16) {
1453 sb->bblog_size = __cpu_to_le16(8);
1454 sb->bblog_offset = __cpu_to_le32(reserved-8);
1459 sb->super_offset = __cpu_to_le64(4*2);
1460 if (data_offset == INVALID_SECTORS) {
1461 if (4*2 + 4*2 + bm_space + array_size
1463 bm_space = dsize - array_size
1466 reserved = bm_space + 4*2 + 4*2;
1467 if (reserved < headroom)
1468 reserved = headroom;
1469 if (reserved + array_size > dsize)
1470 reserved = dsize - array_size;
1471 /* Try for multiple of 1Meg so it is nicely aligned */
1472 #define ONE_MEG (2*1024)
1473 if (reserved > ONE_MEG)
1474 reserved = (reserved/ONE_MEG) * ONE_MEG;
1476 /* force 4K alignment */
1480 reserved = data_offset;
1482 sb->data_offset = __cpu_to_le64(reserved);
1483 sb->data_size = __cpu_to_le64(dsize - reserved);
1484 if (reserved >= 16+16) {
1485 sb->bblog_size = __cpu_to_le16(8);
1486 /* '8' sectors for the bblog, and another '8'
1487 * because we want offset from superblock, not
1490 sb->bblog_offset = __cpu_to_le32(reserved-8-8);
1494 pr_err("Failed to write invalid "
1495 "metadata format 1.%i to %s\n",
1496 st->minor_version, di->devname);
1501 sb->sb_csum = calc_sb_1_csum(sb);
1502 rv = store_super1(st, di->fd);
1503 if (rv == 0 && (__le32_to_cpu(sb->feature_map) & 1))
1504 rv = st->ss->write_bitmap(st, di->fd);
1512 pr_err("Failed to write metadata to %s\n",
1519 static int compare_super1(struct supertype *st, struct supertype *tst)
1523 * 0 same, or first was empty, and second was copied
1524 * 1 second had wrong number
1526 * 3 wrong other info
1528 struct mdp_superblock_1 *first = st->sb;
1529 struct mdp_superblock_1 *second = tst->sb;
1531 if (second->magic != __cpu_to_le32(MD_SB_MAGIC))
1533 if (second->major_version != __cpu_to_le32(1))
1537 if (posix_memalign((void**)&first, 4096, SUPER1_SIZE) != 0) {
1538 pr_err("%s could not allocate superblock\n", __func__);
1541 memcpy(first, second, SUPER1_SIZE);
1545 if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0)
1548 if (first->ctime != second->ctime ||
1549 first->level != second->level ||
1550 first->layout != second->layout ||
1551 first->size != second->size ||
1552 first->chunksize != second->chunksize ||
1553 first->raid_disks != second->raid_disks)
1558 static int load_super1(struct supertype *st, int fd, char *devname)
1560 unsigned long long dsize;
1561 unsigned long long sb_offset;
1562 struct mdp_superblock_1 *super;
1564 struct bitmap_super_s *bsb;
1565 struct misc_dev_info *misc;
1566 struct align_fd afd;
1572 if (st->ss == NULL || st->minor_version == -1) {
1574 struct supertype tst;
1575 __u64 bestctime = 0;
1576 /* guess... choose latest ctime */
1577 memset(&tst, 0, sizeof(tst));
1579 for (tst.minor_version = 0; tst.minor_version <= 2 ; tst.minor_version++) {
1580 switch(load_super1(&tst, fd, devname)) {
1581 case 0: super = tst.sb;
1582 if (bestvers == -1 ||
1583 bestctime < __le64_to_cpu(super->ctime)) {
1584 bestvers = tst.minor_version;
1585 bestctime = __le64_to_cpu(super->ctime);
1590 case 1: return 1; /*bad device */
1591 case 2: break; /* bad, try next */
1594 if (bestvers != -1) {
1596 tst.minor_version = bestvers;
1598 tst.max_devs = MAX_DEVS;
1599 rv = load_super1(&tst, fd, devname);
1606 if (!get_dev_size(fd, devname, &dsize))
1612 pr_err("%s is too small for md: size is %llu sectors.\n",
1618 * Calculate the position of the superblock.
1619 * It is always aligned to a 4K boundary and
1620 * depending on minor_version, it can be:
1621 * 0: At least 8K, but less than 12K, from end of device
1622 * 1: At start of device
1623 * 2: 4K from start of device.
1625 switch(st->minor_version) {
1629 sb_offset &= ~(4*2-1);
1641 if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
1643 pr_err("Cannot seek to superblock on %s: %s\n",
1644 devname, strerror(errno));
1648 if (posix_memalign((void**)&super, 4096, SUPER1_SIZE) != 0) {
1649 pr_err("%s could not allocate superblock\n",
1654 if (aread(&afd, super, MAX_SB_SIZE) != MAX_SB_SIZE) {
1656 pr_err("Cannot read superblock on %s\n",
1662 if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) {
1664 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
1665 devname, MD_SB_MAGIC, __le32_to_cpu(super->magic));
1670 if (__le32_to_cpu(super->major_version) != 1) {
1672 pr_err("Cannot interpret superblock on %s - version is %d\n",
1673 devname, __le32_to_cpu(super->major_version));
1677 if (__le64_to_cpu(super->super_offset) != sb_offset) {
1679 pr_err("No superblock found on %s (super_offset is wrong)\n",
1686 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
1688 misc = (struct misc_dev_info*) (((char*)super)+MAX_SB_SIZE+BM_SUPER_SIZE);
1689 misc->device_size = dsize;
1691 /* Now check on the bitmap superblock */
1692 if ((__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) == 0)
1694 /* Read the bitmap superblock and make sure it looks
1695 * valid. If it doesn't clear the bit. An --assemble --force
1696 * should get that written out.
1698 locate_bitmap1(st, fd);
1699 if (aread(&afd, bsb, 512) != 512)
1702 uuid_from_super1(st, uuid);
1703 if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
1704 memcmp(bsb->uuid, uuid, 16) != 0)
1709 super->feature_map = __cpu_to_le32(__le32_to_cpu(super->feature_map)
1710 & ~MD_FEATURE_BITMAP_OFFSET);
1715 static struct supertype *match_metadata_desc1(char *arg)
1717 struct supertype *st = xcalloc(1, sizeof(*st));
1719 st->container_devnm[0] = 0;
1721 st->max_devs = MAX_DEVS;
1723 /* leading zeros can be safely ignored. --detail generates them. */
1726 if (strcmp(arg, "1.0") == 0 ||
1727 strcmp(arg, "1.00") == 0) {
1728 st->minor_version = 0;
1731 if (strcmp(arg, "1.1") == 0 ||
1732 strcmp(arg, "1.01") == 0
1734 st->minor_version = 1;
1737 if (strcmp(arg, "1.2") == 0 ||
1738 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
1739 strcmp(arg, "default") == 0 ||
1740 #endif /* DEFAULT_OLD_METADATA */
1741 strcmp(arg, "1.02") == 0) {
1742 st->minor_version = 2;
1745 if (strcmp(arg, "1") == 0 ||
1746 strcmp(arg, "default") == 0) {
1747 st->minor_version = -1;
1755 /* find available size on device with this devsize, using
1756 * superblock type st, and reserving 'reserve' sectors for
1759 static __u64 _avail_size1(struct supertype *st, __u64 devsize,
1760 unsigned long long data_offset, int chunksize)
1762 struct mdp_superblock_1 *super = st->sb;
1768 /* creating: allow suitable space for bitmap */
1769 bmspace = choose_bm_space(devsize);
1771 else if (__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) {
1772 /* hot-add. allow for actual size of bitmap */
1773 struct bitmap_super_s *bsb;
1774 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
1775 bmspace = bitmap_sectors(bsb);
1778 /* Allow space for bad block log */
1779 if (super && super->bblog_size)
1780 devsize -= __le16_to_cpu(super->bblog_size);
1785 if (st->minor_version < 0)
1786 /* not specified, so time to set default */
1787 st->minor_version = 2;
1789 if (data_offset != INVALID_SECTORS)
1790 switch(st->minor_version) {
1792 return devsize - data_offset - 8*2;
1795 return devsize - data_offset;
1802 if (super == NULL && st->minor_version > 0) {
1803 /* haven't committed to a size yet, so allow some
1804 * slack for space for reshape.
1805 * Limit slack to 128M, but aim for about 0.1%
1807 unsigned long long headroom = 128*1024*2;
1808 while ((headroom << 10) > devsize &&
1810 headroom / 2 >= ((unsigned)chunksize*2)*2))
1812 devsize -= headroom;
1814 switch(st->minor_version) {
1817 return ((devsize - 8*2 ) & ~(4*2-1));
1819 /* at start, 4K for superblock and possible bitmap */
1820 return devsize - 4*2;
1822 /* 4k from start, 4K for superblock and possible bitmap */
1823 return devsize - (4+4)*2;
1827 static __u64 avail_size1(struct supertype *st, __u64 devsize,
1828 unsigned long long data_offset)
1830 return _avail_size1(st, devsize, data_offset, 0);
1834 add_internal_bitmap1(struct supertype *st,
1835 int *chunkp, int delay, int write_behind,
1836 unsigned long long size,
1837 int may_change, int major)
1840 * If not may_change, then this is a 'Grow' without sysfs support for
1841 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
1842 * If may_change, then this is create or a Grow with sysfs syupport,
1843 * and we can put the bitmap wherever we like.
1845 * size is in sectors, chunk is in bytes !!!
1848 unsigned long long bits;
1849 unsigned long long max_bits;
1850 unsigned long long min_chunk;
1852 long bbl_offset, bbl_size;
1853 unsigned long long chunk = *chunkp;
1856 struct mdp_superblock_1 *sb = st->sb;
1857 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1861 if (__le64_to_cpu(sb->data_size) == 0)
1862 /* Must be creating the array, else data_size would be non-zero */
1864 switch(st->minor_version) {
1866 /* either 3K after the superblock (when hot-add),
1867 * or some amount of space before.
1870 /* We are creating array, so we *know* how much room has
1874 room = choose_bm_space(__le64_to_cpu(sb->size));
1877 room = __le64_to_cpu(sb->super_offset)
1878 - __le64_to_cpu(sb->data_offset)
1879 - __le64_to_cpu(sb->data_size);
1880 bbl_size = __le16_to_cpu(sb->bblog_size);
1883 bbl_offset = (__s32)__le32_to_cpu(sb->bblog_offset);
1884 if (bbl_size < -bbl_offset)
1885 bbl_size = -bbl_offset;
1887 if (!may_change || (room < 3*2 &&
1888 __le32_to_cpu(sb->max_dev) <= 384)) {
1893 offset = 0; /* means movable offset */
1898 case 2: /* between superblock and data */
1901 room = choose_bm_space(__le64_to_cpu(sb->size));
1904 room = __le64_to_cpu(sb->data_offset)
1905 - __le64_to_cpu(sb->super_offset);
1906 bbl_size = __le16_to_cpu(sb->bblog_size);
1908 room = __le32_to_cpu(sb->bblog_offset) + bbl_size;
1913 room -= 2; /* Leave 1K for superblock */
1917 room -= 4*2; /* leave 4K for superblock */
1927 if (chunk == UnSet && room > 128*2)
1928 /* Limit to 128K of bitmap when chunk size not requested */
1931 max_bits = (room * 512 - sizeof(bitmap_super_t)) * 8;
1933 min_chunk = 4096; /* sub-page chunks don't work yet.. */
1934 bits = (size*512)/min_chunk +1;
1935 while (bits > max_bits) {
1939 if (chunk == UnSet) {
1940 /* For practical purpose, 64Meg is a good
1941 * default chunk size for internal bitmaps.
1944 if (chunk < 64*1024*1024)
1945 chunk = 64*1024*1024;
1946 } else if (chunk < min_chunk)
1947 return 0; /* chunk size too small */
1948 if (chunk == 0) /* rounding problem */
1952 /* start bitmap on a 4K boundary with enough space for
1955 bits = (size*512) / chunk + 1;
1956 room = ((bits+7)/8 + sizeof(bitmap_super_t) +4095)/4096;
1957 room *= 8; /* convert 4K blocks to sectors */
1958 offset = -room - bbl_size;
1961 sb->bitmap_offset = (int32_t)__cpu_to_le32(offset);
1963 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map)
1964 | MD_FEATURE_BITMAP_OFFSET);
1965 memset(bms, 0, sizeof(*bms));
1966 bms->magic = __cpu_to_le32(BITMAP_MAGIC);
1967 bms->version = __cpu_to_le32(major);
1968 uuid_from_super1(st, uuid);
1969 memcpy(bms->uuid, uuid, 16);
1970 bms->chunksize = __cpu_to_le32(chunk);
1971 bms->daemon_sleep = __cpu_to_le32(delay);
1972 bms->sync_size = __cpu_to_le64(size);
1973 bms->write_behind = __cpu_to_le32(write_behind);
1979 static void locate_bitmap1(struct supertype *st, int fd)
1981 unsigned long long offset;
1982 struct mdp_superblock_1 *sb;
1986 if (st->ss->load_super(st, fd, NULL))
1987 return; /* no error I hope... */
1992 offset = __le64_to_cpu(sb->super_offset);
1993 offset += (int32_t) __le32_to_cpu(sb->bitmap_offset);
1996 lseek64(fd, offset<<9, 0);
1999 static int write_bitmap1(struct supertype *st, int fd)
2001 struct mdp_superblock_1 *sb = st->sb;
2002 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
2006 struct align_fd afd;
2010 locate_bitmap1(st, fd);
2012 if (posix_memalign(&buf, 4096, 4096))
2015 memset(buf, 0xff, 4096);
2016 memcpy(buf, (char *)bms, sizeof(bitmap_super_t));
2018 towrite = __le64_to_cpu(bms->sync_size) / (__le32_to_cpu(bms->chunksize)>>9);
2019 towrite = (towrite+7) >> 3; /* bits to bytes */
2020 towrite += sizeof(bitmap_super_t);
2021 towrite = ROUND_UP(towrite, 512);
2022 while (towrite > 0) {
2026 n = awrite(&afd, buf, n);
2031 memset(buf, 0xff, 4096);
2041 static void free_super1(struct supertype *st)
2046 struct devinfo *di = st->info;
2047 st->info = di->next;
2056 static int validate_geometry1(struct supertype *st, int level,
2057 int layout, int raiddisks,
2058 int *chunk, unsigned long long size,
2059 unsigned long long data_offset,
2060 char *subdev, unsigned long long *freesize,
2063 unsigned long long ldsize;
2066 if (level == LEVEL_CONTAINER) {
2068 pr_err("1.x metadata does not support containers\n");
2071 if (chunk && *chunk == UnSet)
2072 *chunk = DEFAULT_CHUNK;
2077 fd = open(subdev, O_RDONLY|O_EXCL, 0);
2080 pr_err("super1.x cannot open %s: %s\n",
2081 subdev, strerror(errno));
2085 if (!get_dev_size(fd, subdev, &ldsize)) {
2091 *freesize = _avail_size1(st, ldsize >> 9, data_offset, *chunk);
2094 #endif /* MDASSEMBLE */
2096 struct superswitch super1 = {
2098 .examine_super = examine_super1,
2099 .brief_examine_super = brief_examine_super1,
2100 .export_examine_super = export_examine_super1,
2101 .detail_super = detail_super1,
2102 .brief_detail_super = brief_detail_super1,
2103 .export_detail_super = export_detail_super1,
2104 .write_init_super = write_init_super1,
2105 .validate_geometry = validate_geometry1,
2106 .add_to_super = add_to_super1,
2107 .examine_badblocks = examine_badblocks_super1,
2109 .match_home = match_home1,
2110 .uuid_from_super = uuid_from_super1,
2111 .getinfo_super = getinfo_super1,
2112 .container_content = container_content1,
2113 .update_super = update_super1,
2114 .init_super = init_super1,
2115 .store_super = store_super1,
2116 .compare_super = compare_super1,
2117 .load_super = load_super1,
2118 .match_metadata_desc = match_metadata_desc1,
2119 .avail_size = avail_size1,
2120 .add_internal_bitmap = add_internal_bitmap1,
2121 .locate_bitmap = locate_bitmap1,
2122 .write_bitmap = write_bitmap1,
2123 .free_super = free_super1,
2124 #if __BYTE_ORDER == BIG_ENDIAN