]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super1.c
projects / thirdparty / mdadm.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Detail: show consistency policy
[thirdparty/mdadm.git] / super1.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
5 *
6 *
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.
11 *
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.
16 *
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
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24
25 #include <stddef.h>
26 #include "mdadm.h"
27 /*
28 * The version-1 superblock :
29 * All numeric fields are little-endian.
30 *
31 * total size: 256 bytes plus 2 per device.
32 * 1K allows 384 devices.
33 */
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 */
40
41 __u8 set_uuid[16]; /* user-space generated. */
42 char set_name[32]; /* set and interpreted by user-space */
43
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 */
48
49 __u32 chunksize; /* in 512byte sectors */
50 __u32 raid_disks;
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.
54 */
55
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.
65 */
66
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 */
71 union {
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) */
74 };
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 #define FailFast1 2 /* Device should get FailFast requests */
81 /* bad block log. If there are any bad blocks the feature flag is set.
82 * if offset and size are non-zero, that space is reserved and available.
83 */
84 __u8 bblog_shift; /* shift from sectors to block size for badblocklist */
85 __u16 bblog_size; /* number of sectors reserved for badblocklist */
86 __u32 bblog_offset; /* sector offset from superblock to bblog, signed */
87
88 /* array state information - 64 bytes */
89 __u64 utime; /* 40 bits second, 24 btes microseconds */
90 __u64 events; /* incremented when superblock updated */
91 __u64 resync_offset; /* data before this offset (from data_offset) known to be in sync */
92 __u32 sb_csum; /* checksum upto dev_roles[max_dev] */
93 __u32 max_dev; /* size of dev_roles[] array to consider */
94 __u8 pad3[64-32]; /* set to 0 when writing */
95
96 /* device state information. Indexed by dev_number.
97 * 2 bytes per device
98 * Note there are no per-device state flags. State information is rolled
99 * into the 'roles' value. If a device is spare or faulty, then it doesn't
100 * have a meaningful role.
101 */
102 __u16 dev_roles[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
103 };
104
105 #define MAX_SB_SIZE 4096
106 /* bitmap super size is 256, but we round up to a sector for alignment */
107 #define BM_SUPER_SIZE 512
108 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
109 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
110 + sizeof(struct misc_dev_info))
111
112 struct misc_dev_info {
113 __u64 device_size;
114 };
115
116 /* feature_map bits */
117 #define MD_FEATURE_BITMAP_OFFSET 1
118 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
119 * must be honoured
120 */
121 #define MD_FEATURE_RESHAPE_ACTIVE 4
122 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
123 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
124 * active device with same 'role'.
125 * 'recovery_offset' is also set.
126 */
127 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
128 * of devices, but is going
129 * backwards anyway.
130 */
131 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
132 #define MD_FEATURE_BITMAP_VERSIONED 256 /* bitmap version number checked properly */
133 #define MD_FEATURE_JOURNAL 512 /* support write journal */
134 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
135 |MD_FEATURE_RECOVERY_OFFSET \
136 |MD_FEATURE_RESHAPE_ACTIVE \
137 |MD_FEATURE_BAD_BLOCKS \
138 |MD_FEATURE_REPLACEMENT \
139 |MD_FEATURE_RESHAPE_BACKWARDS \
140 |MD_FEATURE_NEW_OFFSET \
141 |MD_FEATURE_BITMAP_VERSIONED \
142 |MD_FEATURE_JOURNAL \
143 )
144
145 #ifndef MDASSEMBLE
146 static int role_from_sb(struct mdp_superblock_1 *sb)
147 {
148 unsigned int d;
149 int role;
150
151 d = __le32_to_cpu(sb->dev_number);
152 if (d < __le32_to_cpu(sb->max_dev))
153 role = __le16_to_cpu(sb->dev_roles[d]);
154 else
155 role = MD_DISK_ROLE_SPARE;
156 return role;
157 }
158 #endif
159
160 /* return how many bytes are needed for bitmap, for cluster-md each node
161 * should have it's own bitmap */
162 static unsigned int calc_bitmap_size(bitmap_super_t *bms, unsigned int boundary)
163 {
164 unsigned long long bits, bytes;
165
166 bits = bitmap_bits(__le64_to_cpu(bms->sync_size),
167 __le32_to_cpu(bms->chunksize));
168 bytes = (bits+7) >> 3;
169 bytes += sizeof(bitmap_super_t);
170 bytes = ROUND_UP(bytes, boundary);
171
172 return bytes;
173 }
174
175 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
176 {
177 unsigned int disk_csum, csum;
178 unsigned long long newcsum;
179 int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2;
180 unsigned int *isuper = (unsigned int*)sb;
181
182 /* make sure I can count... */
183 if (offsetof(struct mdp_superblock_1,data_offset) != 128 ||
184 offsetof(struct mdp_superblock_1, utime) != 192 ||
185 sizeof(struct mdp_superblock_1) != 256) {
186 fprintf(stderr, "WARNING - superblock isn't sized correctly\n");
187 }
188
189 disk_csum = sb->sb_csum;
190 sb->sb_csum = 0;
191 newcsum = 0;
192 for (; size>=4; size -= 4 ) {
193 newcsum += __le32_to_cpu(*isuper);
194 isuper++;
195 }
196
197 if (size == 2)
198 newcsum += __le16_to_cpu(*(unsigned short*) isuper);
199
200 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
201 sb->sb_csum = disk_csum;
202 return __cpu_to_le32(csum);
203 }
204
205 /*
206 * Information related to file descriptor used for aligned reads/writes.
207 * Cache the block size.
208 */
209 struct align_fd {
210 int fd;
211 int blk_sz;
212 };
213
214 static void init_afd(struct align_fd *afd, int fd)
215 {
216 afd->fd = fd;
217 if (!get_dev_sector_size(afd->fd, NULL, (unsigned int *)&afd->blk_sz))
218 afd->blk_sz = 512;
219 }
220
221 static char abuf[4096+4096];
222 static int aread(struct align_fd *afd, void *buf, int len)
223 {
224 /* aligned read.
225 * On devices with a 4K sector size, we need to read
226 * the full sector and copy relevant bits into
227 * the buffer
228 */
229 int bsize, iosize;
230 char *b;
231 int n;
232
233 bsize = afd->blk_sz;
234
235 if (!bsize || bsize > 4096 || len > 4096) {
236 if (!bsize)
237 fprintf(stderr, "WARNING - aread() called with invalid block size\n");
238 return -1;
239 }
240 b = ROUND_UP_PTR((char *)abuf, 4096);
241
242 for (iosize = 0; iosize < len; iosize += bsize)
243 ;
244 n = read(afd->fd, b, iosize);
245 if (n <= 0)
246 return n;
247 lseek(afd->fd, len - n, 1);
248 if (n > len)
249 n = len;
250 memcpy(buf, b, n);
251 return n;
252 }
253
254 static int awrite(struct align_fd *afd, void *buf, int len)
255 {
256 /* aligned write.
257 * On devices with a 4K sector size, we need to write
258 * the full sector. We pre-read if the sector is larger
259 * than the write.
260 * The address must be sector-aligned.
261 */
262 int bsize, iosize;
263 char *b;
264 int n;
265
266 bsize = afd->blk_sz;
267 if (!bsize || bsize > 4096 || len > 4096) {
268 if (!bsize)
269 fprintf(stderr, "WARNING - awrite() called with invalid block size\n");
270 return -1;
271 }
272 b = ROUND_UP_PTR((char *)abuf, 4096);
273
274 for (iosize = 0; iosize < len ; iosize += bsize)
275 ;
276
277 if (len != iosize) {
278 n = read(afd->fd, b, iosize);
279 if (n <= 0)
280 return n;
281 lseek(afd->fd, -n, 1);
282 }
283
284 memcpy(b, buf, len);
285 n = write(afd->fd, b, iosize);
286 if (n <= 0)
287 return n;
288 lseek(afd->fd, len - n, 1);
289 return len;
290 }
291
292 #ifndef MDASSEMBLE
293 static void examine_super1(struct supertype *st, char *homehost)
294 {
295 struct mdp_superblock_1 *sb = st->sb;
296 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
297 time_t atime;
298 unsigned int d;
299 int role;
300 int delta_extra = 0;
301 int i;
302 char *c;
303 int l = homehost ? strlen(homehost) : 0;
304 int layout;
305 unsigned long long sb_offset;
306 struct mdinfo info;
307
308 printf(" Magic : %08x\n", __le32_to_cpu(sb->magic));
309 printf(" Version : 1");
310 sb_offset = __le64_to_cpu(sb->super_offset);
311 if (sb_offset <= 4)
312 printf(".1\n");
313 else if (sb_offset <= 8)
314 printf(".2\n");
315 else
316 printf(".0\n");
317 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb->feature_map));
318 printf(" Array UUID : ");
319 for (i=0; i<16; i++) {
320 if ((i&3)==0 && i != 0) printf(":");
321 printf("%02x", sb->set_uuid[i]);
322 }
323 printf("\n");
324 printf(" Name : %.32s", sb->set_name);
325 if (l > 0 && l < 32 &&
326 sb->set_name[l] == ':' &&
327 strncmp(sb->set_name, homehost, l) == 0)
328 printf(" (local to host %s)", homehost);
329 printf("\n");
330 if (bms->nodes > 0 && (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
331 printf(" Cluster Name : %-64s\n", bms->cluster_name);
332 atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL;
333 printf(" Creation Time : %.24s\n", ctime(&atime));
334 c=map_num(pers, __le32_to_cpu(sb->level));
335 printf(" Raid Level : %s\n", c?c:"-unknown-");
336 printf(" Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks));
337 printf("\n");
338 printf(" Avail Dev Size : %llu%s\n",
339 (unsigned long long)__le64_to_cpu(sb->data_size),
340 human_size(__le64_to_cpu(sb->data_size)<<9));
341 if (__le32_to_cpu(sb->level) > 0) {
342 int ddsks = 0, ddsks_denom = 1;
343 switch(__le32_to_cpu(sb->level)) {
344 case 1: ddsks=1;break;
345 case 4:
346 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
347 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
348 case 10:
349 layout = __le32_to_cpu(sb->layout);
350 ddsks = __le32_to_cpu(sb->raid_disks);
351 ddsks_denom = (layout&255) * ((layout>>8)&255);
352 }
353 if (ddsks) {
354 long long asize = __le64_to_cpu(sb->size);
355 asize = (asize << 9) * ddsks / ddsks_denom;
356 printf(" Array Size : %llu%s\n",
357 asize >> 10, human_size(asize));
358 }
359 if (sb->size != sb->data_size)
360 printf(" Used Dev Size : %llu%s\n",
361 (unsigned long long)__le64_to_cpu(sb->size),
362 human_size(__le64_to_cpu(sb->size)<<9));
363 }
364 if (sb->data_offset)
365 printf(" Data Offset : %llu sectors\n",
366 (unsigned long long)__le64_to_cpu(sb->data_offset));
367 if (sb->new_offset &&
368 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) {
369 unsigned long long offset = __le64_to_cpu(sb->data_offset);
370 offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset);
371 printf(" New Offset : %llu sectors\n", offset);
372 }
373 printf(" Super Offset : %llu sectors\n",
374 (unsigned long long)__le64_to_cpu(sb->super_offset));
375 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)
376 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb->recovery_offset));
377
378 st->ss->getinfo_super(st, &info, NULL);
379 if (info.space_after != 1 &&
380 !(__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
381 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
382 info.space_before, info.space_after);
383
384 printf(" State : %s\n", (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean");
385 printf(" Device UUID : ");
386 for (i=0; i<16; i++) {
387 if ((i&3)==0 && i != 0) printf(":");
388 printf("%02x", sb->device_uuid[i]);
389 }
390 printf("\n");
391 printf("\n");
392 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
393 printf("Internal Bitmap : %ld sectors from superblock\n",
394 (long)(int32_t)__le32_to_cpu(sb->bitmap_offset));
395 }
396 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)) {
397 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb->reshape_position)/2,
398 human_size(__le64_to_cpu(sb->reshape_position)<<9));
399 if (__le32_to_cpu(sb->delta_disks)) {
400 printf(" Delta Devices : %d", __le32_to_cpu(sb->delta_disks));
401 printf(" (%d->%d)\n",
402 __le32_to_cpu(sb->raid_disks)-__le32_to_cpu(sb->delta_disks),
403 __le32_to_cpu(sb->raid_disks));
404 if ((int)__le32_to_cpu(sb->delta_disks) < 0)
405 delta_extra = -__le32_to_cpu(sb->delta_disks);
406 }
407 if (__le32_to_cpu(sb->new_level) != __le32_to_cpu(sb->level)) {
408 c = map_num(pers, __le32_to_cpu(sb->new_level));
409 printf(" New Level : %s\n", c?c:"-unknown-");
410 }
411 if (__le32_to_cpu(sb->new_layout) != __le32_to_cpu(sb->layout)) {
412 if (__le32_to_cpu(sb->level) == 5) {
413 c = map_num(r5layout, __le32_to_cpu(sb->new_layout));
414 printf(" New Layout : %s\n", c?c:"-unknown-");
415 }
416 if (__le32_to_cpu(sb->level) == 6) {
417 c = map_num(r6layout, __le32_to_cpu(sb->new_layout));
418 printf(" New Layout : %s\n", c?c:"-unknown-");
419 }
420 if (__le32_to_cpu(sb->level) == 10) {
421 printf(" New Layout :");
422 print_r10_layout(__le32_to_cpu(sb->new_layout));
423 printf("\n");
424 }
425 }
426 if (__le32_to_cpu(sb->new_chunk) != __le32_to_cpu(sb->chunksize))
427 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb->new_chunk)/2);
428 printf("\n");
429 }
430 if (sb->devflags) {
431 printf(" Flags :");
432 if (sb->devflags & WriteMostly1)
433 printf(" write-mostly");
434 if (sb->devflags & FailFast1)
435 printf(" failfast");
436 printf("\n");
437 }
438
439 atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL;
440 printf(" Update Time : %.24s\n", ctime(&atime));
441
442 if (sb->bblog_size && sb->bblog_offset) {
443 printf(" Bad Block Log : %d entries available at offset %ld sectors",
444 __le16_to_cpu(sb->bblog_size)*512/8,
445 (long)(int32_t)__le32_to_cpu(sb->bblog_offset));
446 if (sb->feature_map &
447 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
448 printf(" - bad blocks present.");
449 printf("\n");
450 }
451
452 if (calc_sb_1_csum(sb) == sb->sb_csum)
453 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb->sb_csum));
454 else
455 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb->sb_csum),
456 __le32_to_cpu(calc_sb_1_csum(sb)));
457 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb->events));
458 printf("\n");
459 if (__le32_to_cpu(sb->level) == 5) {
460 c = map_num(r5layout, __le32_to_cpu(sb->layout));
461 printf(" Layout : %s\n", c?c:"-unknown-");
462 }
463 if (__le32_to_cpu(sb->level) == 6) {
464 c = map_num(r6layout, __le32_to_cpu(sb->layout));
465 printf(" Layout : %s\n", c?c:"-unknown-");
466 }
467 if (__le32_to_cpu(sb->level) == 10) {
468 int lo = __le32_to_cpu(sb->layout);
469 printf(" Layout :");
470 print_r10_layout(lo);
471 printf("\n");
472 }
473 switch(__le32_to_cpu(sb->level)) {
474 case 0:
475 case 4:
476 case 5:
477 case 6:
478 case 10:
479 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb->chunksize)/2);
480 break;
481 case -1:
482 printf(" Rounding : %dK\n", __le32_to_cpu(sb->chunksize)/2);
483 break;
484 default: break;
485 }
486 printf("\n");
487 #if 0
488 /* This turns out to just be confusing */
489 printf(" Array Slot : %d (", __le32_to_cpu(sb->dev_number));
490 for (i= __le32_to_cpu(sb->max_dev); i> 0 ; i--)
491 if (__le16_to_cpu(sb->dev_roles[i-1]) != MD_DISK_ROLE_SPARE)
492 break;
493 for (d=0; d < i; d++) {
494 int role = __le16_to_cpu(sb->dev_roles[d]);
495 if (d) printf(", ");
496 if (role == MD_DISK_ROLE_SPARE) printf("empty");
497 else if(role == MD_DISK_ROLE_FAULTY) printf("failed");
498 else printf("%d", role);
499 }
500 printf(")\n");
501 #endif
502 printf(" Device Role : ");
503 role = role_from_sb(sb);
504 if (role >= MD_DISK_ROLE_FAULTY)
505 printf("spare\n");
506 else if (role == MD_DISK_ROLE_JOURNAL)
507 printf("Journal\n");
508 else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_REPLACEMENT))
509 printf("Replacement device %d\n", role);
510 else
511 printf("Active device %d\n", role);
512
513 printf(" Array State : ");
514 for (d=0; d<__le32_to_cpu(sb->raid_disks) + delta_extra; d++) {
515 int cnt = 0;
516 unsigned int i;
517 for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
518 unsigned int role = __le16_to_cpu(sb->dev_roles[i]);
519 if (role == d)
520 cnt++;
521 }
522 if (cnt == 2)
523 printf("R");
524 else if (cnt == 1)
525 printf("A");
526 else if (cnt == 0)
527 printf(".");
528 else
529 printf("?");
530 }
531 #if 0
532 /* This is confusing too */
533 faulty = 0;
534 for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
535 int role = __le16_to_cpu(sb->dev_roles[i]);
536 if (role == MD_DISK_ROLE_FAULTY)
537 faulty++;
538 }
539 if (faulty) printf(" %d failed", faulty);
540 #endif
541 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
542 printf("\n");
543 }
544
545 static void brief_examine_super1(struct supertype *st, int verbose)
546 {
547 struct mdp_superblock_1 *sb = st->sb;
548 int i;
549 unsigned long long sb_offset;
550 char *nm;
551 char *c=map_num(pers, __le32_to_cpu(sb->level));
552
553 nm = strchr(sb->set_name, ':');
554 if (nm)
555 nm++;
556 else if (sb->set_name[0])
557 nm = sb->set_name;
558 else
559 nm = NULL;
560
561 printf("ARRAY ");
562 if (nm) {
563 printf("/dev/md/");
564 print_escape(nm);
565 putchar(' ');
566 }
567 if (verbose && c)
568 printf(" level=%s", c);
569 sb_offset = __le64_to_cpu(sb->super_offset);
570 if (sb_offset <= 4)
571 printf(" metadata=1.1 ");
572 else if (sb_offset <= 8)
573 printf(" metadata=1.2 ");
574 else
575 printf(" metadata=1.0 ");
576 if (verbose)
577 printf("num-devices=%d ", __le32_to_cpu(sb->raid_disks));
578 printf("UUID=");
579 for (i=0; i<16; i++) {
580 if ((i&3)==0 && i != 0) printf(":");
581 printf("%02x", sb->set_uuid[i]);
582 }
583 if (sb->set_name[0]) {
584 printf(" name=");
585 print_quoted(sb->set_name);
586 }
587 printf("\n");
588 }
589
590 static void export_examine_super1(struct supertype *st)
591 {
592 struct mdp_superblock_1 *sb = st->sb;
593 int i;
594 int len = 32;
595 int layout;
596
597 printf("MD_LEVEL=%s\n", map_num(pers, __le32_to_cpu(sb->level)));
598 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb->raid_disks));
599 for (i=0; i<32; i++)
600 if (sb->set_name[i] == '\n' ||
601 sb->set_name[i] == '\0') {
602 len = i;
603 break;
604 }
605 if (len)
606 printf("MD_NAME=%.*s\n", len, sb->set_name);
607 if (__le32_to_cpu(sb->level) > 0) {
608 int ddsks = 0, ddsks_denom = 1;
609 switch(__le32_to_cpu(sb->level)) {
610 case 1: ddsks=1;break;
611 case 4:
612 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
613 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
614 case 10:
615 layout = __le32_to_cpu(sb->layout);
616 ddsks = __le32_to_cpu(sb->raid_disks);
617 ddsks_denom = (layout&255) * ((layout>>8)&255);
618 }
619 if (ddsks) {
620 long long asize = __le64_to_cpu(sb->size);
621 asize = (asize << 9) * ddsks / ddsks_denom;
622 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize,JEDEC));
623 }
624 }
625 printf("MD_UUID=");
626 for (i=0; i<16; i++) {
627 if ((i&3)==0 && i != 0) printf(":");
628 printf("%02x", sb->set_uuid[i]);
629 }
630 printf("\n");
631 printf("MD_UPDATE_TIME=%llu\n",
632 __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL);
633 printf("MD_DEV_UUID=");
634 for (i=0; i<16; i++) {
635 if ((i&3)==0 && i != 0) printf(":");
636 printf("%02x", sb->device_uuid[i]);
637 }
638 printf("\n");
639 printf("MD_EVENTS=%llu\n",
640 (unsigned long long)__le64_to_cpu(sb->events));
641 }
642
643 static int copy_metadata1(struct supertype *st, int from, int to)
644 {
645 /* Read superblock. If it looks good, write it out.
646 * Then if a bitmap is present, copy that.
647 * And if a bad-block-list is present, copy that too.
648 */
649 void *buf;
650 unsigned long long dsize, sb_offset;
651 const int bufsize = 4*1024;
652 struct mdp_superblock_1 super, *sb;
653
654 if (posix_memalign(&buf, 4096, bufsize) != 0)
655 return 1;
656
657 if (!get_dev_size(from, NULL, &dsize))
658 goto err;
659
660 dsize >>= 9;
661 if (dsize < 24)
662 goto err;
663 switch(st->minor_version) {
664 case 0:
665 sb_offset = dsize;
666 sb_offset -= 8*2;
667 sb_offset &= ~(4*2-1);
668 break;
669 case 1:
670 sb_offset = 0;
671 break;
672 case 2:
673 sb_offset = 4*2;
674 break;
675 default:
676 goto err;
677 }
678
679 if (lseek64(from, sb_offset << 9, 0) < 0LL)
680 goto err;
681 if (read(from, buf, bufsize) != bufsize)
682 goto err;
683
684 sb = buf;
685 super = *sb; // save most of sb for when we reuse buf
686
687 if (__le32_to_cpu(super.magic) != MD_SB_MAGIC ||
688 __le32_to_cpu(super.major_version) != 1 ||
689 __le64_to_cpu(super.super_offset) != sb_offset ||
690 calc_sb_1_csum(sb) != super.sb_csum)
691 goto err;
692
693 if (lseek64(to, sb_offset << 9, 0) < 0LL)
694 goto err;
695 if (write(to, buf, bufsize) != bufsize)
696 goto err;
697
698 if (super.feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
699 unsigned long long bitmap_offset = sb_offset;
700 int bytes = 4096; // just an estimate.
701 int written = 0;
702 struct align_fd afrom, ato;
703
704 init_afd(&afrom, from);
705 init_afd(&ato, to);
706
707 bitmap_offset += (int32_t)__le32_to_cpu(super.bitmap_offset);
708
709 if (lseek64(from, bitmap_offset<<9, 0) < 0)
710 goto err;
711 if (lseek64(to, bitmap_offset<<9, 0) < 0)
712 goto err;
713
714 for (written = 0; written < bytes ; ) {
715 int n = bytes - written;
716 if (n > 4096)
717 n = 4096;
718 if (aread(&afrom, buf, n) != n)
719 goto err;
720 if (written == 0) {
721 /* have the header, can calculate
722 * correct bitmap bytes */
723 bitmap_super_t *bms;
724 bms = (void*)buf;
725 bytes = calc_bitmap_size(bms, 512);
726 if (n > bytes)
727 n = bytes;
728 }
729 if (awrite(&ato, buf, n) != n)
730 goto err;
731 written += n;
732 }
733 }
734
735 if (super.bblog_size != 0 &&
736 __le16_to_cpu(super.bblog_size) <= 100 &&
737 super.bblog_offset != 0 &&
738 (super.feature_map & __le32_to_cpu(MD_FEATURE_BAD_BLOCKS))) {
739 /* There is a bad block log */
740 unsigned long long bb_offset = sb_offset;
741 int bytes = __le16_to_cpu(super.bblog_size) * 512;
742 int written = 0;
743 struct align_fd afrom, ato;
744
745 init_afd(&afrom, from);
746 init_afd(&ato, to);
747
748 bb_offset += (int32_t)__le32_to_cpu(super.bblog_offset);
749
750 if (lseek64(from, bb_offset<<9, 0) < 0)
751 goto err;
752 if (lseek64(to, bb_offset<<9, 0) < 0)
753 goto err;
754
755 for (written = 0; written < bytes ; ) {
756 int n = bytes - written;
757 if (n > 4096)
758 n = 4096;
759 if (aread(&afrom, buf, n) != n)
760 goto err;
761
762 if (awrite(&ato, buf, n) != n)
763 goto err;
764 written += n;
765 }
766 }
767
768 free(buf);
769 return 0;
770
771 err:
772 free(buf);
773 return 1;
774 }
775
776 static void detail_super1(struct supertype *st, char *homehost)
777 {
778 struct mdp_superblock_1 *sb = st->sb;
779 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
780 int i;
781 int l = homehost ? strlen(homehost) : 0;
782
783 printf(" Name : %.32s", sb->set_name);
784 if (l > 0 && l < 32 &&
785 sb->set_name[l] == ':' &&
786 strncmp(sb->set_name, homehost, l) == 0)
787 printf(" (local to host %s)", homehost);
788 if (bms->nodes > 0 && (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
789 printf("\n Cluster Name : %-64s", bms->cluster_name);
790 printf("\n UUID : ");
791 for (i=0; i<16; i++) {
792 if ((i&3)==0 && i != 0) printf(":");
793 printf("%02x", sb->set_uuid[i]);
794 }
795 printf("\n Events : %llu\n\n",
796 (unsigned long long)__le64_to_cpu(sb->events));
797 }
798
799 static void brief_detail_super1(struct supertype *st)
800 {
801 struct mdp_superblock_1 *sb = st->sb;
802 int i;
803
804 if (sb->set_name[0]) {
805 printf(" name=");
806 print_quoted(sb->set_name);
807 }
808 printf(" UUID=");
809 for (i=0; i<16; i++) {
810 if ((i&3)==0 && i != 0) printf(":");
811 printf("%02x", sb->set_uuid[i]);
812 }
813 }
814
815 static void export_detail_super1(struct supertype *st)
816 {
817 struct mdp_superblock_1 *sb = st->sb;
818 int i;
819 int len = 32;
820
821 for (i=0; i<32; i++)
822 if (sb->set_name[i] == '\n' ||
823 sb->set_name[i] == '\0') {
824 len = i;
825 break;
826 }
827 if (len)
828 printf("MD_NAME=%.*s\n", len, sb->set_name);
829 }
830
831 static int examine_badblocks_super1(struct supertype *st, int fd, char *devname)
832 {
833 struct mdp_superblock_1 *sb = st->sb;
834 unsigned long long offset;
835 int size;
836 __u64 *bbl, *bbp;
837 int i;
838
839 if (!sb->bblog_size || __le16_to_cpu(sb->bblog_size) > 100
840 || !sb->bblog_offset){
841 printf("No bad-blocks list configured on %s\n", devname);
842 return 0;
843 }
844 if ((sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
845 == 0) {
846 printf("Bad-blocks list is empty in %s\n", devname);
847 return 0;
848 }
849
850 size = __le16_to_cpu(sb->bblog_size)* 512;
851 if (posix_memalign((void**)&bbl, 4096, size) != 0) {
852 pr_err("could not allocate badblocks list\n");
853 return 0;
854 }
855 offset = __le64_to_cpu(sb->super_offset) +
856 (int)__le32_to_cpu(sb->bblog_offset);
857 offset <<= 9;
858 if (lseek64(fd, offset, 0) < 0) {
859 pr_err("Cannot seek to bad-blocks list\n");
860 return 1;
861 }
862 if (read(fd, bbl, size) != size) {
863 pr_err("Cannot read bad-blocks list\n");
864 return 1;
865 }
866 /* 64bits per entry. 10 bits is block-count, 54 bits is block
867 * offset. Blocks are sectors unless bblog->shift makes them bigger
868 */
869 bbp = (__u64*)bbl;
870 printf("Bad-blocks on %s:\n", devname);
871 for (i = 0; i < size/8; i++, bbp++) {
872 __u64 bb = __le64_to_cpu(*bbp);
873 int count = bb & 0x3ff;
874 unsigned long long sector = bb >> 10;
875
876 if (bb + 1 == 0)
877 break;
878
879 sector <<= sb->bblog_shift;
880 count <<= sb->bblog_shift;
881
882 printf("%20llu for %d sectors\n", sector, count);
883 }
884 return 0;
885 }
886
887 #endif
888
889 static int match_home1(struct supertype *st, char *homehost)
890 {
891 struct mdp_superblock_1 *sb = st->sb;
892 int l = homehost ? strlen(homehost) : 0;
893
894 return (l > 0 && l < 32 &&
895 sb->set_name[l] == ':' &&
896 strncmp(sb->set_name, homehost, l) == 0);
897 }
898
899 static void uuid_from_super1(struct supertype *st, int uuid[4])
900 {
901 struct mdp_superblock_1 *super = st->sb;
902 char *cuuid = (char*)uuid;
903 int i;
904 for (i=0; i<16; i++)
905 cuuid[i] = super->set_uuid[i];
906 }
907
908 static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
909 {
910 struct mdp_superblock_1 *sb = st->sb;
911 struct bitmap_super_s *bsb = (void*)(((char*)sb)+MAX_SB_SIZE);
912 struct misc_dev_info *misc = (void*)(((char*)sb)+MAX_SB_SIZE+BM_SUPER_SIZE);
913 int working = 0;
914 unsigned int i;
915 unsigned int role;
916 unsigned int map_disks = info->array.raid_disks;
917 unsigned long long super_offset;
918 unsigned long long data_size;
919
920 memset(info, 0, sizeof(*info));
921 info->array.major_version = 1;
922 info->array.minor_version = st->minor_version;
923 info->array.patch_version = 0;
924 info->array.raid_disks = __le32_to_cpu(sb->raid_disks);
925 info->array.level = __le32_to_cpu(sb->level);
926 info->array.layout = __le32_to_cpu(sb->layout);
927 info->array.md_minor = -1;
928 info->array.ctime = __le64_to_cpu(sb->ctime);
929 info->array.utime = __le64_to_cpu(sb->utime);
930 info->array.chunk_size = __le32_to_cpu(sb->chunksize)*512;
931 info->array.state =
932 (__le64_to_cpu(sb->resync_offset) == MaxSector)
933 ? 1 : 0;
934 if (__le32_to_cpu(bsb->nodes) > 1)
935 info->array.state |= (1 << MD_SB_CLUSTERED);
936
937 info->data_offset = __le64_to_cpu(sb->data_offset);
938 info->component_size = __le64_to_cpu(sb->size);
939 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET))
940 info->bitmap_offset = (int32_t)__le32_to_cpu(sb->bitmap_offset);
941
942 info->disk.major = 0;
943 info->disk.minor = 0;
944 info->disk.number = __le32_to_cpu(sb->dev_number);
945 if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) ||
946 __le32_to_cpu(sb->dev_number) >= MAX_DEVS)
947 role = MD_DISK_ROLE_FAULTY;
948 else
949 role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]);
950
951 super_offset = __le64_to_cpu(sb->super_offset);
952 if (info->array.level <= 0)
953 data_size = __le64_to_cpu(sb->data_size);
954 else
955 data_size = __le64_to_cpu(sb->size);
956 if (info->data_offset < super_offset) {
957 unsigned long long end;
958 info->space_before = info->data_offset;
959 end = super_offset;
960
961 if (sb->bblog_offset && sb->bblog_size) {
962 unsigned long long bboffset = super_offset;
963 bboffset += (int32_t)__le32_to_cpu(sb->bblog_offset);
964 if (bboffset < end)
965 end = bboffset;
966 }
967
968 if (super_offset + info->bitmap_offset < end)
969 end = super_offset + info->bitmap_offset;
970
971 if (info->data_offset + data_size < end)
972 info->space_after = end - data_size - info->data_offset;
973 else
974 info->space_after = 0;
975 } else {
976 unsigned long long earliest;
977 earliest = super_offset + (32+4)*2; /* match kernel */
978 if (info->bitmap_offset > 0) {
979 unsigned long long bmend = info->bitmap_offset;
980 unsigned long long size = calc_bitmap_size(bsb, 4096);
981 size /= 512;
982 bmend += size;
983 if (bmend > earliest)
984 earliest = bmend;
985 }
986 if (sb->bblog_offset && sb->bblog_size) {
987 unsigned long long bbend = super_offset;
988 bbend += (int32_t)__le32_to_cpu(sb->bblog_offset);
989 bbend += __le16_to_cpu(sb->bblog_size);
990 if (bbend > earliest)
991 earliest = bbend;
992 }
993 if (earliest < info->data_offset)
994 info->space_before = info->data_offset - earliest;
995 else
996 info->space_before = 0;
997 info->space_after = misc->device_size - data_size - info->data_offset;
998 }
999 if (info->space_before == 0 && info->space_after == 0) {
1000 /* It will look like we don't support data_offset changes,
1001 * be we do - it's just that there is no room.
1002 * A change that reduced the number of devices should
1003 * still be allowed, so set the otherwise useless value of '1'
1004 */
1005 info->space_after = 1;
1006 }
1007
1008 info->disk.raid_disk = -1;
1009 switch(role) {
1010 case MD_DISK_ROLE_SPARE:
1011 info->disk.state = 0; /* spare: not active, not sync, not faulty */
1012 break;
1013 case MD_DISK_ROLE_FAULTY:
1014 info->disk.state = 1; /* faulty */
1015 break;
1016 case MD_DISK_ROLE_JOURNAL:
1017 info->disk.state = (1 << MD_DISK_JOURNAL);
1018 info->disk.raid_disk = role;
1019 info->space_after = (misc->device_size - info->data_offset) % 8; /* journal uses all 4kB blocks*/
1020 break;
1021 default:
1022 info->disk.state = 6; /* active and in sync */
1023 info->disk.raid_disk = role;
1024 }
1025 if (sb->devflags & WriteMostly1)
1026 info->disk.state |= (1 << MD_DISK_WRITEMOSTLY);
1027 if (sb->devflags & FailFast1)
1028 info->disk.state |= (1 << MD_DISK_FAILFAST);
1029 info->events = __le64_to_cpu(sb->events);
1030 sprintf(info->text_version, "1.%d", st->minor_version);
1031 info->safe_mode_delay = 200;
1032
1033 memcpy(info->uuid, sb->set_uuid, 16);
1034
1035 strncpy(info->name, sb->set_name, 32);
1036 info->name[32] = 0;
1037
1038 if ((__le32_to_cpu(sb->feature_map)&MD_FEATURE_REPLACEMENT)) {
1039 info->disk.state &= ~(1 << MD_DISK_SYNC);
1040 info->disk.state |= 1 << MD_DISK_REPLACEMENT;
1041 }
1042
1043 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET))
1044 info->recovery_start = __le32_to_cpu(sb->recovery_offset);
1045 else
1046 info->recovery_start = MaxSector;
1047
1048 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) {
1049 info->reshape_active = 1;
1050 if ((sb->feature_map & __le32_to_cpu(MD_FEATURE_NEW_OFFSET)) &&
1051 sb->new_offset != 0)
1052 info->reshape_active |= RESHAPE_NO_BACKUP;
1053 info->reshape_progress = __le64_to_cpu(sb->reshape_position);
1054 info->new_level = __le32_to_cpu(sb->new_level);
1055 info->delta_disks = __le32_to_cpu(sb->delta_disks);
1056 info->new_layout = __le32_to_cpu(sb->new_layout);
1057 info->new_chunk = __le32_to_cpu(sb->new_chunk)<<9;
1058 if (info->delta_disks < 0)
1059 info->array.raid_disks -= info->delta_disks;
1060 } else
1061 info->reshape_active = 0;
1062
1063 info->recovery_blocked = info->reshape_active;
1064
1065 if (map)
1066 for (i=0; i<map_disks; i++)
1067 map[i] = 0;
1068 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
1069 role = __le16_to_cpu(sb->dev_roles[i]);
1070 if (/*role == MD_DISK_ROLE_SPARE || */role < (unsigned) info->array.raid_disks) {
1071 working++;
1072 if (map && role < map_disks)
1073 map[role] = 1;
1074 }
1075 }
1076
1077 info->array.working_disks = working;
1078 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_JOURNAL))
1079 info->journal_device_required = 1;
1080 info->journal_clean = 0;
1081 }
1082
1083 static struct mdinfo *container_content1(struct supertype *st, char *subarray)
1084 {
1085 struct mdinfo *info;
1086
1087 if (subarray)
1088 return NULL;
1089
1090 info = xmalloc(sizeof(*info));
1091 getinfo_super1(st, info, NULL);
1092 return info;
1093 }
1094
1095 static int update_super1(struct supertype *st, struct mdinfo *info,
1096 char *update,
1097 char *devname, int verbose,
1098 int uuid_set, char *homehost)
1099 {
1100 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1101 * if any change was made. For others, the return value is
1102 * ignored.
1103 */
1104 int rv = 0;
1105 int lockid;
1106 struct mdp_superblock_1 *sb = st->sb;
1107 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1108
1109 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1110 rv = cluster_get_dlmlock(&lockid);
1111 if (rv) {
1112 pr_err("Cannot get dlmlock in %s return %d\n", __func__, rv);
1113 cluster_release_dlmlock(lockid);
1114 return rv;
1115 }
1116 }
1117
1118 if (strcmp(update, "homehost") == 0 &&
1119 homehost) {
1120 /* Note that 'homehost' is special as it is really
1121 * a "name" update.
1122 */
1123 char *c;
1124 update = "name";
1125 c = strchr(sb->set_name, ':');
1126 if (c)
1127 strncpy(info->name, c+1, 31 - (c-sb->set_name));
1128 else
1129 strncpy(info->name, sb->set_name, 32);
1130 info->name[32] = 0;
1131 }
1132
1133 if (strcmp(update, "force-one")==0) {
1134 /* Not enough devices for a working array,
1135 * so bring this one up-to-date
1136 */
1137 if (sb->events != __cpu_to_le64(info->events))
1138 rv = 1;
1139 sb->events = __cpu_to_le64(info->events);
1140 } else if (strcmp(update, "force-array")==0) {
1141 /* Degraded array and 'force' requests to
1142 * maybe need to mark it 'clean'.
1143 */
1144 switch(__le32_to_cpu(sb->level)) {
1145 case 5: case 4: case 6:
1146 /* need to force clean */
1147 if (sb->resync_offset != MaxSector)
1148 rv = 1;
1149 sb->resync_offset = MaxSector;
1150 }
1151 } else if (strcmp(update, "assemble")==0) {
1152 int d = info->disk.number;
1153 int want;
1154 if (info->disk.state & (1<<MD_DISK_ACTIVE))
1155 want = info->disk.raid_disk;
1156 else if (info->disk.state & (1<<MD_DISK_JOURNAL))
1157 want = MD_DISK_ROLE_JOURNAL;
1158 else
1159 want = MD_DISK_ROLE_SPARE;
1160 if (sb->dev_roles[d] != __cpu_to_le16(want)) {
1161 sb->dev_roles[d] = __cpu_to_le16(want);
1162 rv = 1;
1163 }
1164 if (info->reshape_active &&
1165 sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1166 info->delta_disks >= 0 &&
1167 info->reshape_progress < __le64_to_cpu(sb->reshape_position)) {
1168 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1169 rv = 1;
1170 }
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);
1176 rv = 1;
1177 }
1178 } else if (strcmp(update, "linear-grow-new") == 0) {
1179 unsigned int i;
1180 int fd;
1181 unsigned int max = __le32_to_cpu(sb->max_dev);
1182
1183 for (i=0 ; i < max ; i++)
1184 if (__le16_to_cpu(sb->dev_roles[i]) >= MD_DISK_ROLE_FAULTY)
1185 break;
1186 sb->dev_number = __cpu_to_le32(i);
1187 info->disk.number = i;
1188 if (max >= __le32_to_cpu(sb->max_dev))
1189 sb->max_dev = __cpu_to_le32(max+1);
1190
1191 random_uuid(sb->device_uuid);
1192
1193 sb->dev_roles[i] =
1194 __cpu_to_le16(info->disk.raid_disk);
1195
1196 fd = open(devname, O_RDONLY);
1197 if (fd >= 0) {
1198 unsigned long long ds;
1199 get_dev_size(fd, devname, &ds);
1200 close(fd);
1201 ds >>= 9;
1202 if (__le64_to_cpu(sb->super_offset) <
1203 __le64_to_cpu(sb->data_offset)) {
1204 sb->data_size = __cpu_to_le64(
1205 ds - __le64_to_cpu(sb->data_offset));
1206 } else {
1207 ds -= 8*2;
1208 ds &= ~(unsigned long long)(4*2-1);
1209 sb->super_offset = __cpu_to_le64(ds);
1210 sb->data_size = __cpu_to_le64(
1211 ds - __le64_to_cpu(sb->data_offset));
1212 }
1213 }
1214 } else if (strcmp(update, "linear-grow-update") == 0) {
1215 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
1216 sb->dev_roles[info->disk.number] =
1217 __cpu_to_le16(info->disk.raid_disk);
1218 } else if (strcmp(update, "resync") == 0) {
1219 /* make sure resync happens */
1220 sb->resync_offset = 0ULL;
1221 } else if (strcmp(update, "uuid") == 0) {
1222 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
1223
1224 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)
1225 memcpy(bms->uuid, sb->set_uuid, 16);
1226 } else if (strcmp(update, "no-bitmap") == 0) {
1227 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1228 } else if (strcmp(update, "bbl") == 0) {
1229 /* only possible if there is room after the bitmap, or if
1230 * there is no bitmap
1231 */
1232 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1233 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1234 long bitmap_offset = 0;
1235 long bm_sectors = 0;
1236 long space;
1237
1238 #ifndef MDASSEMBLE
1239 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1240 bitmap_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1241 bm_sectors = calc_bitmap_size(bms, 4096) >> 9;
1242 }
1243 #endif
1244 if (sb_offset < data_offset) {
1245 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1246 */
1247 long bb_offset;
1248 bb_offset = sb_offset + 8;
1249 if (bm_sectors && bitmap_offset > 0)
1250 bb_offset = bitmap_offset + bm_sectors;
1251 while (bb_offset < (long)sb_offset + 8 + 32*2
1252 && bb_offset + 8+8 <= (long)data_offset)
1253 /* too close to bitmap, and room to grow */
1254 bb_offset += 8;
1255 if (bb_offset + 8 <= (long)data_offset) {
1256 sb->bblog_size = __cpu_to_le16(8);
1257 sb->bblog_offset = __cpu_to_le32(bb_offset);
1258 }
1259 } else {
1260 /* 1.0 - Put bbl just before super block */
1261 if (bm_sectors && bitmap_offset < 0)
1262 space = -bitmap_offset - bm_sectors;
1263 else
1264 space = sb_offset - data_offset -
1265 __le64_to_cpu(sb->data_size);
1266 if (space >= 8) {
1267 sb->bblog_size = __cpu_to_le16(8);
1268 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1269 }
1270 }
1271 } else if (strcmp(update, "no-bbl") == 0) {
1272 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
1273 pr_err("Cannot remove active bbl from %s\n",devname);
1274 else {
1275 sb->bblog_size = 0;
1276 sb->bblog_shift = 0;
1277 sb->bblog_offset = 0;
1278 }
1279 } else if (strcmp(update, "force-no-bbl") == 0) {
1280 sb->feature_map &= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1281 sb->bblog_size = 0;
1282 sb->bblog_shift = 0;
1283 sb->bblog_offset = 0;
1284 } else if (strcmp(update, "name") == 0) {
1285 if (info->name[0] == 0)
1286 sprintf(info->name, "%d", info->array.md_minor);
1287 memset(sb->set_name, 0, sizeof(sb->set_name));
1288 if (homehost &&
1289 strchr(info->name, ':') == NULL &&
1290 strlen(homehost)+1+strlen(info->name) < 32) {
1291 strcpy(sb->set_name, homehost);
1292 strcat(sb->set_name, ":");
1293 strcat(sb->set_name, info->name);
1294 } else
1295 strncpy(sb->set_name, info->name, sizeof(sb->set_name));
1296 } else if (strcmp(update, "devicesize") == 0 &&
1297 __le64_to_cpu(sb->super_offset) <
1298 __le64_to_cpu(sb->data_offset)) {
1299 /* set data_size to device size less data_offset */
1300 struct misc_dev_info *misc = (struct misc_dev_info*)
1301 (st->sb + MAX_SB_SIZE + BM_SUPER_SIZE);
1302 sb->data_size = __cpu_to_le64(
1303 misc->device_size - __le64_to_cpu(sb->data_offset));
1304 } else if (strncmp(update, "revert-reshape", 14) == 0) {
1305 rv = -2;
1306 if (!(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)))
1307 pr_err("No active reshape to revert on %s\n",
1308 devname);
1309 else {
1310 __u32 temp;
1311 unsigned long long reshape_sectors;
1312 long reshape_chunk;
1313 rv = 0;
1314 /* If the reshape hasn't started, just stop it.
1315 * It is conceivable that a stripe was modified but
1316 * the metadata not updated. In that case the backup
1317 * should have been used to get passed the critical stage.
1318 * If that couldn't happen, the "-nobackup" version
1319 * will be used.
1320 */
1321 if (strcmp(update, "revert-reshape-nobackup") == 0 &&
1322 sb->reshape_position == 0 &&
1323 (__le32_to_cpu(sb->delta_disks) > 0 ||
1324 (__le32_to_cpu(sb->delta_disks) == 0 &&
1325 !(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS))))) {
1326 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1327 sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1328 __le32_to_cpu(sb->delta_disks));
1329 sb->delta_disks = 0;
1330 goto done;
1331 }
1332 /* reshape_position is a little messy.
1333 * Its value must be a multiple of the larger
1334 * chunk size, and of the "after" data disks.
1335 * So when reverting we need to change it to
1336 * be a multiple of the new "after" data disks,
1337 * which is the old "before".
1338 * If it isn't already a multiple of 'before',
1339 * the only thing we could do would be
1340 * copy some block around on the disks, which
1341 * is easy to get wrong.
1342 * So we reject a revert-reshape unless the
1343 * alignment is good.
1344 */
1345 if (__le32_to_cpu(sb->level) >= 4 &&
1346 __le32_to_cpu(sb->level) <= 6) {
1347 reshape_sectors = __le64_to_cpu(sb->reshape_position);
1348 reshape_chunk = __le32_to_cpu(sb->new_chunk);
1349 reshape_chunk *= __le32_to_cpu(sb->raid_disks) - __le32_to_cpu(sb->delta_disks) -
1350 (__le32_to_cpu(sb->level)==6 ? 2 : 1);
1351 if (reshape_sectors % reshape_chunk) {
1352 pr_err("Reshape position is not suitably aligned.\n");
1353 pr_err("Try normal assembly and stop again\n");
1354 return -2;
1355 }
1356 }
1357 sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1358 __le32_to_cpu(sb->delta_disks));
1359 if (sb->delta_disks == 0)
1360 sb->feature_map ^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1361 else
1362 sb->delta_disks = __cpu_to_le32(-__le32_to_cpu(sb->delta_disks));
1363
1364 temp = sb->new_layout;
1365 sb->new_layout = sb->layout;
1366 sb->layout = temp;
1367
1368 temp = sb->new_chunk;
1369 sb->new_chunk = sb->chunksize;
1370 sb->chunksize = temp;
1371
1372 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_NEW_OFFSET)) {
1373 long offset_delta = (int32_t)__le32_to_cpu(sb->new_offset);
1374 sb->data_offset = __cpu_to_le64(__le64_to_cpu(sb->data_offset) + offset_delta);
1375 sb->new_offset = __cpu_to_le32(-offset_delta);
1376 sb->data_size = __cpu_to_le64(__le64_to_cpu(sb->data_size) - offset_delta);
1377 }
1378 done:;
1379 }
1380 } else if (strcmp(update, "_reshape_progress")==0)
1381 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1382 else if (strcmp(update, "writemostly")==0)
1383 sb->devflags |= WriteMostly1;
1384 else if (strcmp(update, "readwrite")==0)
1385 sb->devflags &= ~WriteMostly1;
1386 else if (strcmp(update, "failfast") == 0)
1387 sb->devflags |= FailFast1;
1388 else if (strcmp(update, "nofailfast") == 0)
1389 sb->devflags &= ~FailFast1;
1390 else
1391 rv = -1;
1392
1393 sb->sb_csum = calc_sb_1_csum(sb);
1394 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1395 cluster_release_dlmlock(lockid);
1396
1397 return rv;
1398 }
1399
1400 static int init_super1(struct supertype *st, mdu_array_info_t *info,
1401 struct shape *s, char *name, char *homehost,
1402 int *uuid, unsigned long long data_offset)
1403 {
1404 struct mdp_superblock_1 *sb;
1405 int spares;
1406 char defname[10];
1407 int sbsize;
1408
1409 if (posix_memalign((void**)&sb, 4096, SUPER1_SIZE) != 0) {
1410 pr_err("could not allocate superblock\n");
1411 return 0;
1412 }
1413 memset(sb, 0, SUPER1_SIZE);
1414
1415 st->sb = sb;
1416 if (info == NULL) {
1417 /* zeroing superblock */
1418 return 0;
1419 }
1420
1421 spares = info->working_disks - info->active_disks;
1422 if (info->raid_disks + spares > MAX_DEVS) {
1423 pr_err("too many devices requested: %d+%d > %d\n",
1424 info->raid_disks , spares, MAX_DEVS);
1425 return 0;
1426 }
1427
1428 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
1429 sb->major_version = __cpu_to_le32(1);
1430 sb->feature_map = 0;
1431 sb->pad0 = 0;
1432
1433 if (uuid)
1434 copy_uuid(sb->set_uuid, uuid, super1.swapuuid);
1435 else
1436 random_uuid(sb->set_uuid);;
1437
1438 if (name == NULL || *name == 0) {
1439 sprintf(defname, "%d", info->md_minor);
1440 name = defname;
1441 }
1442 if (homehost &&
1443 strchr(name, ':')== NULL &&
1444 strlen(homehost)+1+strlen(name) < 32) {
1445 strcpy(sb->set_name, homehost);
1446 strcat(sb->set_name, ":");
1447 strcat(sb->set_name, name);
1448 } else
1449 strncpy(sb->set_name, name, sizeof(sb->set_name));
1450
1451 sb->ctime = __cpu_to_le64((unsigned long long)time(0));
1452 sb->level = __cpu_to_le32(info->level);
1453 sb->layout = __cpu_to_le32(info->layout);
1454 sb->size = __cpu_to_le64(s->size*2ULL);
1455 sb->chunksize = __cpu_to_le32(info->chunk_size>>9);
1456 sb->raid_disks = __cpu_to_le32(info->raid_disks);
1457
1458 sb->data_offset = __cpu_to_le64(data_offset);
1459 sb->data_size = __cpu_to_le64(0);
1460 sb->super_offset = __cpu_to_le64(0);
1461 sb->recovery_offset = __cpu_to_le64(0);
1462
1463 sb->utime = sb->ctime;
1464 sb->events = __cpu_to_le64(1);
1465 if (info->state & (1<<MD_SB_CLEAN))
1466 sb->resync_offset = MaxSector;
1467 else
1468 sb->resync_offset = 0;
1469 sbsize = sizeof(struct mdp_superblock_1) + 2 * (info->raid_disks + spares);
1470 sbsize = ROUND_UP(sbsize, 512);
1471 sb->max_dev = __cpu_to_le32((sbsize - sizeof(struct mdp_superblock_1)) / 2);
1472
1473 memset(sb->dev_roles, 0xff, MAX_SB_SIZE - sizeof(struct mdp_superblock_1));
1474
1475 return 1;
1476 }
1477
1478 struct devinfo {
1479 int fd;
1480 char *devname;
1481 long long data_offset;
1482 mdu_disk_info_t disk;
1483 struct devinfo *next;
1484 };
1485 #ifndef MDASSEMBLE
1486 /* Add a device to the superblock being created */
1487 static int add_to_super1(struct supertype *st, mdu_disk_info_t *dk,
1488 int fd, char *devname, unsigned long long data_offset)
1489 {
1490 struct mdp_superblock_1 *sb = st->sb;
1491 __u16 *rp = sb->dev_roles + dk->number;
1492 struct devinfo *di, **dip;
1493 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1494 int rv, lockid;
1495 int dk_state;
1496
1497 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1498 rv = cluster_get_dlmlock(&lockid);
1499 if (rv) {
1500 pr_err("Cannot get dlmlock in %s return %d\n", __func__, rv);
1501 cluster_release_dlmlock(lockid);
1502 return rv;
1503 }
1504 }
1505
1506 dk_state = dk->state & ~(1<<MD_DISK_FAILFAST);
1507 if ((dk_state & 6) == 6) /* active, sync */
1508 *rp = __cpu_to_le16(dk->raid_disk);
1509 else if (dk_state & (1<<MD_DISK_JOURNAL))
1510 *rp = MD_DISK_ROLE_JOURNAL;
1511 else if ((dk_state & ~2) == 0) /* active or idle -> spare */
1512 *rp = MD_DISK_ROLE_SPARE;
1513 else
1514 *rp = MD_DISK_ROLE_FAULTY;
1515
1516 if (dk->number >= (int)__le32_to_cpu(sb->max_dev) &&
1517 __le32_to_cpu(sb->max_dev) < MAX_DEVS)
1518 sb->max_dev = __cpu_to_le32(dk->number+1);
1519
1520 sb->dev_number = __cpu_to_le32(dk->number);
1521 sb->devflags = 0; /* don't copy another disks flags */
1522 sb->sb_csum = calc_sb_1_csum(sb);
1523
1524 dip = (struct devinfo **)&st->info;
1525 while (*dip)
1526 dip = &(*dip)->next;
1527 di = xmalloc(sizeof(struct devinfo));
1528 di->fd = fd;
1529 di->devname = devname;
1530 di->disk = *dk;
1531 di->data_offset = data_offset;
1532 di->next = NULL;
1533 *dip = di;
1534
1535 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1536 cluster_release_dlmlock(lockid);
1537
1538 return 0;
1539 }
1540 #endif
1541
1542 static int locate_bitmap1(struct supertype *st, int fd, int node_num);
1543
1544 static int store_super1(struct supertype *st, int fd)
1545 {
1546 struct mdp_superblock_1 *sb = st->sb;
1547 unsigned long long sb_offset;
1548 struct align_fd afd;
1549 int sbsize;
1550 unsigned long long dsize;
1551 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1552 int rv, lockid;
1553
1554 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1555 rv = cluster_get_dlmlock(&lockid);
1556 if (rv) {
1557 pr_err("Cannot get dlmlock in %s return %d\n", __func__, rv);
1558 cluster_release_dlmlock(lockid);
1559 return rv;
1560 }
1561 }
1562
1563 if (!get_dev_size(fd, NULL, &dsize))
1564 return 1;
1565
1566 dsize >>= 9;
1567
1568 if (dsize < 24)
1569 return 2;
1570
1571 init_afd(&afd, fd);
1572
1573 /*
1574 * Calculate the position of the superblock.
1575 * It is always aligned to a 4K boundary and
1576 * depending on minor_version, it can be:
1577 * 0: At least 8K, but less than 12K, from end of device
1578 * 1: At start of device
1579 * 2: 4K from start of device.
1580 */
1581 switch(st->minor_version) {
1582 case 0:
1583 sb_offset = dsize;
1584 sb_offset -= 8*2;
1585 sb_offset &= ~(4*2-1);
1586 break;
1587 case 1:
1588 sb_offset = 0;
1589 break;
1590 case 2:
1591 sb_offset = 4*2;
1592 break;
1593 default:
1594 return -EINVAL;
1595 }
1596
1597 if (sb_offset != __le64_to_cpu(sb->super_offset) &&
1598 0 != __le64_to_cpu(sb->super_offset)
1599 ) {
1600 pr_err("internal error - sb_offset is wrong\n");
1601 abort();
1602 }
1603
1604 if (lseek64(fd, sb_offset << 9, 0)< 0LL)
1605 return 3;
1606
1607 sbsize = ROUND_UP(sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev), 512);
1608
1609 if (awrite(&afd, sb, sbsize) != sbsize)
1610 return 4;
1611
1612 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1613 struct bitmap_super_s *bm = (struct bitmap_super_s*)
1614 (((char*)sb)+MAX_SB_SIZE);
1615 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC) {
1616 locate_bitmap1(st, fd, 0);
1617 if (awrite(&afd, bm, sizeof(*bm)) != sizeof(*bm))
1618 return 5;
1619 }
1620 }
1621 fsync(fd);
1622 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1623 cluster_release_dlmlock(lockid);
1624
1625 return 0;
1626 }
1627
1628 static int load_super1(struct supertype *st, int fd, char *devname);
1629
1630 static unsigned long choose_bm_space(unsigned long devsize)
1631 {
1632 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1633 * if bigger than 200Gig, save 128k
1634 * NOTE: result must be multiple of 4K else bad things happen
1635 * on 4K-sector devices.
1636 */
1637 if (devsize < 64*2)
1638 return 0;
1639 if (devsize - 64*2 >= 200*1024*1024*2)
1640 return 128*2;
1641 if (devsize - 4*2 > 8*1024*1024*2)
1642 return 64*2;
1643 return 4*2;
1644 }
1645
1646 static void free_super1(struct supertype *st);
1647
1648 #define META_BLOCK_SIZE 4096
1649 __u32 crc32c_le(__u32 crc, unsigned char const *p, size_t len);
1650
1651 #ifndef MDASSEMBLE
1652 static int write_empty_r5l_meta_block(struct supertype *st, int fd)
1653 {
1654 struct r5l_meta_block *mb;
1655 struct mdp_superblock_1 *sb = st->sb;
1656 struct align_fd afd;
1657 __u32 crc;
1658
1659 init_afd(&afd, fd);
1660
1661 if (posix_memalign((void**)&mb, 4096, META_BLOCK_SIZE) != 0) {
1662 pr_err("Could not allocate memory for the meta block.\n");
1663 return 1;
1664 }
1665
1666 memset(mb, 0, META_BLOCK_SIZE);
1667
1668 mb->magic = __cpu_to_le32(R5LOG_MAGIC);
1669 mb->version = R5LOG_VERSION;
1670 mb->meta_size = __cpu_to_le32(sizeof(struct r5l_meta_block));
1671 mb->seq = __cpu_to_le64(random32());
1672 mb->position = __cpu_to_le64(0);
1673
1674 crc = crc32c_le(0xffffffff, sb->set_uuid, sizeof(sb->set_uuid));
1675 crc = crc32c_le(crc, (void *)mb, META_BLOCK_SIZE);
1676 mb->checksum = crc;
1677
1678 if (lseek64(fd, (sb->data_offset) * 512, 0) < 0LL) {
1679 pr_err("cannot seek to offset of the meta block\n");
1680 goto fail_to_write;
1681 }
1682
1683 if (awrite(&afd, mb, META_BLOCK_SIZE) != META_BLOCK_SIZE) {
1684 pr_err("failed to store write the meta block \n");
1685 goto fail_to_write;
1686 }
1687 fsync(fd);
1688
1689 free(mb);
1690 return 0;
1691
1692 fail_to_write:
1693 free(mb);
1694 return 1;
1695 }
1696
1697 static int write_init_super1(struct supertype *st)
1698 {
1699 struct mdp_superblock_1 *sb = st->sb;
1700 struct supertype *refst;
1701 int rv = 0;
1702 unsigned long long bm_space;
1703 struct devinfo *di;
1704 unsigned long long dsize, array_size;
1705 unsigned long long sb_offset;
1706 unsigned long long data_offset;
1707 long bm_offset;
1708
1709 for (di = st->info; di; di = di->next) {
1710 if (di->disk.state & (1 << MD_DISK_JOURNAL))
1711 sb->feature_map |= MD_FEATURE_JOURNAL;
1712 }
1713
1714 for (di = st->info; di; di = di->next) {
1715 if (di->disk.state & (1 << MD_DISK_FAULTY))
1716 continue;
1717 if (di->fd < 0)
1718 continue;
1719
1720 while (Kill(di->devname, NULL, 0, -1, 1) == 0)
1721 ;
1722
1723 sb->dev_number = __cpu_to_le32(di->disk.number);
1724 if (di->disk.state & (1<<MD_DISK_WRITEMOSTLY))
1725 sb->devflags |= WriteMostly1;
1726 else
1727 sb->devflags &= ~WriteMostly1;
1728 if (di->disk.state & (1<<MD_DISK_FAILFAST))
1729 sb->devflags |= FailFast1;
1730 else
1731 sb->devflags &= ~FailFast1;
1732
1733 random_uuid(sb->device_uuid);
1734
1735 if (!(di->disk.state & (1<<MD_DISK_JOURNAL)))
1736 sb->events = 0;
1737
1738 refst = dup_super(st);
1739 if (load_super1(refst, di->fd, NULL)==0) {
1740 struct mdp_superblock_1 *refsb = refst->sb;
1741
1742 memcpy(sb->device_uuid, refsb->device_uuid, 16);
1743 if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) {
1744 /* same array, so preserve events and
1745 * dev_number */
1746 sb->events = refsb->events;
1747 /* bugs in 2.6.17 and earlier mean the
1748 * dev_number chosen in Manage must be preserved
1749 */
1750 if (get_linux_version() >= 2006018)
1751 sb->dev_number = refsb->dev_number;
1752 }
1753 free_super1(refst);
1754 }
1755 free(refst);
1756
1757 if (!get_dev_size(di->fd, NULL, &dsize)) {
1758 rv = 1;
1759 goto error_out;
1760 }
1761 dsize >>= 9;
1762
1763 if (dsize < 24) {
1764 close(di->fd);
1765 rv = 2;
1766 goto error_out;
1767 }
1768
1769 /*
1770 * Calculate the position of the superblock.
1771 * It is always aligned to a 4K boundary and
1772 * depending on minor_version, it can be:
1773 * 0: At least 8K, but less than 12K, from end of device
1774 * 1: At start of device
1775 * 2: 4K from start of device.
1776 * data_offset has already been set.
1777 */
1778 array_size = __le64_to_cpu(sb->size);
1779
1780 /* work out how much space we left for a bitmap */
1781 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1782 bitmap_super_t *bms = (bitmap_super_t *)
1783 (((char *)sb) + MAX_SB_SIZE);
1784 bm_space = calc_bitmap_size(bms, 4096) >> 9;
1785 bm_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1786 } else {
1787 bm_space = choose_bm_space(array_size);
1788 bm_offset = 8;
1789 }
1790
1791 data_offset = di->data_offset;
1792 if (data_offset == INVALID_SECTORS)
1793 data_offset = st->data_offset;
1794 switch(st->minor_version) {
1795 case 0:
1796 /* Add 8 sectors for bad block log */
1797 bm_space += 8;
1798 if (data_offset == INVALID_SECTORS)
1799 data_offset = 0;
1800 sb_offset = dsize;
1801 sb_offset -= 8*2;
1802 sb_offset &= ~(4*2-1);
1803 sb->data_offset = __cpu_to_le64(data_offset);
1804 sb->super_offset = __cpu_to_le64(sb_offset);
1805 if (sb_offset < array_size + bm_space)
1806 bm_space = sb_offset - array_size;
1807 sb->data_size = __cpu_to_le64(sb_offset - bm_space);
1808 if (bm_space >= 8) {
1809 sb->bblog_size = __cpu_to_le16(8);
1810 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1811 }
1812 break;
1813 case 1:
1814 case 2:
1815 sb_offset = st->minor_version == 2 ? 8 : 0;
1816 sb->super_offset = __cpu_to_le64(sb_offset);
1817 if (data_offset == INVALID_SECTORS)
1818 data_offset = sb_offset + 16;
1819
1820 sb->data_offset = __cpu_to_le64(data_offset);
1821 sb->data_size = __cpu_to_le64(dsize - data_offset);
1822 if (data_offset >= sb_offset+bm_offset+bm_space+8) {
1823 sb->bblog_size = __cpu_to_le16(8);
1824 sb->bblog_offset = __cpu_to_le32(bm_offset +
1825 bm_space);
1826 } else if (data_offset >= sb_offset + 16) {
1827 sb->bblog_size = __cpu_to_le16(8);
1828 /* '8' sectors for the bblog, and 'sb_offset'
1829 * because we want offset from superblock, not
1830 * start of device.
1831 */
1832 sb->bblog_offset = __cpu_to_le32(data_offset -
1833 8 - sb_offset);
1834 }
1835 break;
1836 default:
1837 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1838 st->minor_version, di->devname);
1839 rv = -EINVAL;
1840 goto out;
1841 }
1842 /* Disable badblock log on clusters, or when explicitly requested */
1843 if (st->nodes > 0 || conf_get_create_info()->bblist == 0) {
1844 sb->bblog_size = 0;
1845 sb->bblog_offset = 0;
1846 }
1847
1848 sb->sb_csum = calc_sb_1_csum(sb);
1849 rv = store_super1(st, di->fd);
1850
1851 if (rv == 0 && (di->disk.state & (1 << MD_DISK_JOURNAL))) {
1852 rv = write_empty_r5l_meta_block(st, di->fd);
1853 if (rv)
1854 goto error_out;
1855 }
1856
1857 if (rv == 0 && (__le32_to_cpu(sb->feature_map) & 1))
1858 rv = st->ss->write_bitmap(st, di->fd, NodeNumUpdate);
1859 close(di->fd);
1860 di->fd = -1;
1861 if (rv)
1862 goto error_out;
1863 }
1864 error_out:
1865 if (rv)
1866 pr_err("Failed to write metadata to %s\n",
1867 di->devname);
1868 out:
1869 return rv;
1870 }
1871 #endif
1872
1873 static int compare_super1(struct supertype *st, struct supertype *tst)
1874 {
1875 /*
1876 * return:
1877 * 0 same, or first was empty, and second was copied
1878 * 1 second had wrong number
1879 * 2 wrong uuid
1880 * 3 wrong other info
1881 */
1882 struct mdp_superblock_1 *first = st->sb;
1883 struct mdp_superblock_1 *second = tst->sb;
1884
1885 if (second->magic != __cpu_to_le32(MD_SB_MAGIC))
1886 return 1;
1887 if (second->major_version != __cpu_to_le32(1))
1888 return 1;
1889
1890 if (!first) {
1891 if (posix_memalign((void**)&first, 4096, SUPER1_SIZE) != 0) {
1892 pr_err("could not allocate superblock\n");
1893 return 1;
1894 }
1895 memcpy(first, second, SUPER1_SIZE);
1896 st->sb = first;
1897 return 0;
1898 }
1899 if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0)
1900 return 2;
1901
1902 if (first->ctime != second->ctime ||
1903 first->level != second->level ||
1904 first->layout != second->layout ||
1905 first->size != second->size ||
1906 first->chunksize != second->chunksize ||
1907 first->raid_disks != second->raid_disks)
1908 return 3;
1909 return 0;
1910 }
1911
1912 static int load_super1(struct supertype *st, int fd, char *devname)
1913 {
1914 unsigned long long dsize;
1915 unsigned long long sb_offset;
1916 struct mdp_superblock_1 *super;
1917 int uuid[4];
1918 struct bitmap_super_s *bsb;
1919 struct misc_dev_info *misc;
1920 struct align_fd afd;
1921
1922 free_super1(st);
1923
1924 init_afd(&afd, fd);
1925
1926 if (st->ss == NULL || st->minor_version == -1) {
1927 int bestvers = -1;
1928 struct supertype tst;
1929 __u64 bestctime = 0;
1930 /* guess... choose latest ctime */
1931 memset(&tst, 0, sizeof(tst));
1932 tst.ss = &super1;
1933 for (tst.minor_version = 0; tst.minor_version <= 2 ; tst.minor_version++) {
1934 switch(load_super1(&tst, fd, devname)) {
1935 case 0: super = tst.sb;
1936 if (bestvers == -1 ||
1937 bestctime < __le64_to_cpu(super->ctime)) {
1938 bestvers = tst.minor_version;
1939 bestctime = __le64_to_cpu(super->ctime);
1940 }
1941 free(super);
1942 tst.sb = NULL;
1943 break;
1944 case 1: return 1; /*bad device */
1945 case 2: break; /* bad, try next */
1946 }
1947 }
1948 if (bestvers != -1) {
1949 int rv;
1950 tst.minor_version = bestvers;
1951 tst.ss = &super1;
1952 tst.max_devs = MAX_DEVS;
1953 rv = load_super1(&tst, fd, devname);
1954 if (rv == 0)
1955 *st = tst;
1956 return rv;
1957 }
1958 return 2;
1959 }
1960 if (!get_dev_size(fd, devname, &dsize))
1961 return 1;
1962 dsize >>= 9;
1963
1964 if (dsize < 24) {
1965 if (devname)
1966 pr_err("%s is too small for md: size is %llu sectors.\n",
1967 devname, dsize);
1968 return 1;
1969 }
1970
1971 /*
1972 * Calculate the position of the superblock.
1973 * It is always aligned to a 4K boundary and
1974 * depending on minor_version, it can be:
1975 * 0: At least 8K, but less than 12K, from end of device
1976 * 1: At start of device
1977 * 2: 4K from start of device.
1978 */
1979 switch(st->minor_version) {
1980 case 0:
1981 sb_offset = dsize;
1982 sb_offset -= 8*2;
1983 sb_offset &= ~(4*2-1);
1984 break;
1985 case 1:
1986 sb_offset = 0;
1987 break;
1988 case 2:
1989 sb_offset = 4*2;
1990 break;
1991 default:
1992 return -EINVAL;
1993 }
1994
1995 if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
1996 if (devname)
1997 pr_err("Cannot seek to superblock on %s: %s\n",
1998 devname, strerror(errno));
1999 return 1;
2000 }
2001
2002 if (posix_memalign((void**)&super, 4096, SUPER1_SIZE) != 0) {
2003 pr_err("could not allocate superblock\n");
2004 return 1;
2005 }
2006
2007 memset(super, 0, SUPER1_SIZE);
2008
2009 if (aread(&afd, super, MAX_SB_SIZE) != MAX_SB_SIZE) {
2010 if (devname)
2011 pr_err("Cannot read superblock on %s\n",
2012 devname);
2013 free(super);
2014 return 1;
2015 }
2016
2017 if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) {
2018 if (devname)
2019 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2020 devname, MD_SB_MAGIC, __le32_to_cpu(super->magic));
2021 free(super);
2022 return 2;
2023 }
2024
2025 if (__le32_to_cpu(super->major_version) != 1) {
2026 if (devname)
2027 pr_err("Cannot interpret superblock on %s - version is %d\n",
2028 devname, __le32_to_cpu(super->major_version));
2029 free(super);
2030 return 2;
2031 }
2032 if (__le64_to_cpu(super->super_offset) != sb_offset) {
2033 if (devname)
2034 pr_err("No superblock found on %s (super_offset is wrong)\n",
2035 devname);
2036 free(super);
2037 return 2;
2038 }
2039 st->sb = super;
2040
2041 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
2042
2043 misc = (struct misc_dev_info*) (((char*)super)+MAX_SB_SIZE+BM_SUPER_SIZE);
2044 misc->device_size = dsize;
2045 if (st->data_offset == INVALID_SECTORS)
2046 st->data_offset = __le64_to_cpu(super->data_offset);
2047
2048 /* Now check on the bitmap superblock */
2049 if ((__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) == 0)
2050 return 0;
2051 /* Read the bitmap superblock and make sure it looks
2052 * valid. If it doesn't clear the bit. An --assemble --force
2053 * should get that written out.
2054 */
2055 locate_bitmap1(st, fd, 0);
2056 if (aread(&afd, bsb, 512) != 512)
2057 goto no_bitmap;
2058
2059 uuid_from_super1(st, uuid);
2060 if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
2061 memcmp(bsb->uuid, uuid, 16) != 0)
2062 goto no_bitmap;
2063 return 0;
2064
2065 no_bitmap:
2066 super->feature_map = __cpu_to_le32(__le32_to_cpu(super->feature_map)
2067 & ~MD_FEATURE_BITMAP_OFFSET);
2068 return 0;
2069 }
2070
2071 static struct supertype *match_metadata_desc1(char *arg)
2072 {
2073 struct supertype *st = xcalloc(1, sizeof(*st));
2074
2075 st->container_devnm[0] = 0;
2076 st->ss = &super1;
2077 st->max_devs = MAX_DEVS;
2078 st->sb = NULL;
2079 st->data_offset = INVALID_SECTORS;
2080 /* leading zeros can be safely ignored. --detail generates them. */
2081 while (*arg == '0')
2082 arg++;
2083 if (strcmp(arg, "1.0") == 0 ||
2084 strcmp(arg, "1.00") == 0) {
2085 st->minor_version = 0;
2086 return st;
2087 }
2088 if (strcmp(arg, "1.1") == 0 ||
2089 strcmp(arg, "1.01") == 0
2090 ) {
2091 st->minor_version = 1;
2092 return st;
2093 }
2094 if (strcmp(arg, "1.2") == 0 ||
2095 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2096 strcmp(arg, "default") == 0 ||
2097 #endif /* DEFAULT_OLD_METADATA */
2098 strcmp(arg, "1.02") == 0) {
2099 st->minor_version = 2;
2100 return st;
2101 }
2102 if (strcmp(arg, "1") == 0 ||
2103 strcmp(arg, "default") == 0) {
2104 st->minor_version = -1;
2105 return st;
2106 }
2107
2108 free(st);
2109 return NULL;
2110 }
2111
2112 /* find available size on device with this devsize, using
2113 * superblock type st, and reserving 'reserve' sectors for
2114 * a possible bitmap
2115 */
2116 static __u64 avail_size1(struct supertype *st, __u64 devsize,
2117 unsigned long long data_offset)
2118 {
2119 struct mdp_superblock_1 *super = st->sb;
2120 int bmspace = 0;
2121 int bbspace = 0;
2122 if (devsize < 24)
2123 return 0;
2124
2125 #ifndef MDASSEMBLE
2126 if (__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) {
2127 /* hot-add. allow for actual size of bitmap */
2128 struct bitmap_super_s *bsb;
2129 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
2130 bmspace = calc_bitmap_size(bsb, 4096) >> 9;
2131 }
2132 #endif
2133 /* Allow space for bad block log */
2134 if (super->bblog_size)
2135 bbspace = __le16_to_cpu(super->bblog_size);
2136
2137 if (st->minor_version < 0)
2138 /* not specified, so time to set default */
2139 st->minor_version = 2;
2140
2141 if (data_offset == INVALID_SECTORS)
2142 data_offset = st->data_offset;
2143
2144 if (data_offset != INVALID_SECTORS)
2145 switch(st->minor_version) {
2146 case 0:
2147 return devsize - data_offset - 8*2 - bbspace;
2148 case 1:
2149 case 2:
2150 return devsize - data_offset;
2151 default:
2152 return 0;
2153 }
2154
2155 devsize -= bmspace;
2156
2157 switch(st->minor_version) {
2158 case 0:
2159 /* at end */
2160 return ((devsize - 8*2 - bbspace ) & ~(4*2-1));
2161 case 1:
2162 /* at start, 4K for superblock and possible bitmap */
2163 return devsize - 4*2 - bbspace;
2164 case 2:
2165 /* 4k from start, 4K for superblock and possible bitmap */
2166 return devsize - (4+4)*2 - bbspace;
2167 }
2168 return 0;
2169 }
2170
2171 static int
2172 add_internal_bitmap1(struct supertype *st,
2173 int *chunkp, int delay, int write_behind,
2174 unsigned long long size,
2175 int may_change, int major)
2176 {
2177 /*
2178 * If not may_change, then this is a 'Grow' without sysfs support for
2179 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2180 * If may_change, then this is create or a Grow with sysfs syupport,
2181 * and we can put the bitmap wherever we like.
2182 *
2183 * size is in sectors, chunk is in bytes !!!
2184 */
2185
2186 unsigned long long bits;
2187 unsigned long long max_bits;
2188 unsigned long long min_chunk;
2189 long offset;
2190 long bbl_offset, bbl_size;
2191 unsigned long long chunk = *chunkp;
2192 int room = 0;
2193 int creating = 0;
2194 int len;
2195 struct mdp_superblock_1 *sb = st->sb;
2196 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
2197 int uuid[4];
2198
2199 if (__le64_to_cpu(sb->data_size) == 0)
2200 /* Must be creating the array, else data_size would be non-zero */
2201 creating = 1;
2202 switch(st->minor_version) {
2203 case 0:
2204 /* either 3K after the superblock (when hot-add),
2205 * or some amount of space before.
2206 */
2207 if (creating) {
2208 /* We are creating array, so we *know* how much room has
2209 * been left.
2210 */
2211 offset = 0;
2212 bbl_size = 8;
2213 room = choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2214 } else {
2215 room = __le64_to_cpu(sb->super_offset)
2216 - __le64_to_cpu(sb->data_offset)
2217 - __le64_to_cpu(sb->data_size);
2218 bbl_size = __le16_to_cpu(sb->bblog_size);
2219 if (bbl_size < 8)
2220 bbl_size = 8;
2221 bbl_offset = (__s32)__le32_to_cpu(sb->bblog_offset);
2222 if (bbl_size < -bbl_offset)
2223 bbl_size = -bbl_offset;
2224
2225 if (!may_change || (room < 3*2 &&
2226 __le32_to_cpu(sb->max_dev) <= 384)) {
2227 room = 3*2;
2228 offset = 1*2;
2229 bbl_size = 0;
2230 } else {
2231 offset = 0; /* means movable offset */
2232 }
2233 }
2234 break;
2235 case 1:
2236 case 2: /* between superblock and data */
2237 if (creating) {
2238 offset = 4*2;
2239 bbl_size = 8;
2240 room = choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2241 } else {
2242 room = __le64_to_cpu(sb->data_offset)
2243 - __le64_to_cpu(sb->super_offset);
2244 bbl_size = __le16_to_cpu(sb->bblog_size);
2245 if (bbl_size)
2246 room = __le32_to_cpu(sb->bblog_offset) + bbl_size;
2247 else
2248 bbl_size = 8;
2249
2250 if (!may_change) {
2251 room -= 2; /* Leave 1K for superblock */
2252 offset = 2;
2253 bbl_size = 0;
2254 } else {
2255 room -= 4*2; /* leave 4K for superblock */
2256 offset = 4*2;
2257 }
2258 }
2259 break;
2260 default:
2261 return -ENOSPC;
2262 }
2263
2264 room -= bbl_size;
2265 if (chunk == UnSet && room > 128*2)
2266 /* Limit to 128K of bitmap when chunk size not requested */
2267 room = 128*2;
2268
2269 if (room <= 1)
2270 /* No room for a bitmap */
2271 return -ENOSPC;
2272
2273 max_bits = (room * 512 - sizeof(bitmap_super_t)) * 8;
2274
2275 min_chunk = 4096; /* sub-page chunks don't work yet.. */
2276 bits = (size*512)/min_chunk +1;
2277 while (bits > max_bits) {
2278 min_chunk *= 2;
2279 bits = (bits+1)/2;
2280 }
2281 if (chunk == UnSet) {
2282 /* For practical purpose, 64Meg is a good
2283 * default chunk size for internal bitmaps.
2284 */
2285 chunk = min_chunk;
2286 if (chunk < 64*1024*1024)
2287 chunk = 64*1024*1024;
2288 } else if (chunk < min_chunk)
2289 return -EINVAL; /* chunk size too small */
2290 if (chunk == 0) /* rounding problem */
2291 return -EINVAL;
2292
2293 if (offset == 0) {
2294 /* start bitmap on a 4K boundary with enough space for
2295 * the bitmap
2296 */
2297 bits = (size*512) / chunk + 1;
2298 room = ((bits+7)/8 + sizeof(bitmap_super_t) +4095)/4096;
2299 room *= 8; /* convert 4K blocks to sectors */
2300 offset = -room - bbl_size;
2301 }
2302
2303 sb->bitmap_offset = (int32_t)__cpu_to_le32(offset);
2304
2305 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map)
2306 | MD_FEATURE_BITMAP_OFFSET);
2307 memset(bms, 0, sizeof(*bms));
2308 bms->magic = __cpu_to_le32(BITMAP_MAGIC);
2309 bms->version = __cpu_to_le32(major);
2310 uuid_from_super1(st, uuid);
2311 memcpy(bms->uuid, uuid, 16);
2312 bms->chunksize = __cpu_to_le32(chunk);
2313 bms->daemon_sleep = __cpu_to_le32(delay);
2314 bms->sync_size = __cpu_to_le64(size);
2315 bms->write_behind = __cpu_to_le32(write_behind);
2316 bms->nodes = __cpu_to_le32(st->nodes);
2317 if (st->nodes)
2318 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map)
2319 | MD_FEATURE_BITMAP_VERSIONED);
2320 if (st->cluster_name) {
2321 len = sizeof(bms->cluster_name);
2322 strncpy((char *)bms->cluster_name, st->cluster_name, len);
2323 bms->cluster_name[len - 1] = '\0';
2324 }
2325
2326 *chunkp = chunk;
2327 return 0;
2328 }
2329
2330 static int locate_bitmap1(struct supertype *st, int fd, int node_num)
2331 {
2332 unsigned long long offset;
2333 struct mdp_superblock_1 *sb;
2334 int mustfree = 0;
2335 int ret;
2336
2337 if (!st->sb) {
2338 if (st->ss->load_super(st, fd, NULL))
2339 return -1; /* no error I hope... */
2340 mustfree = 1;
2341 }
2342 sb = st->sb;
2343
2344 if ((__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
2345 ret = 0;
2346 else
2347 ret = -1;
2348 offset = __le64_to_cpu(sb->super_offset);
2349 offset += (int32_t) __le32_to_cpu(sb->bitmap_offset) * (node_num + 1);
2350 if (mustfree)
2351 free(sb);
2352 lseek64(fd, offset<<9, 0);
2353 return ret;
2354 }
2355
2356 static int write_bitmap1(struct supertype *st, int fd, enum bitmap_update update)
2357 {
2358 struct mdp_superblock_1 *sb = st->sb;
2359 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
2360 int rv = 0;
2361 void *buf;
2362 int towrite, n, len;
2363 struct align_fd afd;
2364 unsigned int i = 0;
2365 unsigned long long total_bm_space, bm_space_per_node;
2366
2367 switch (update) {
2368 case NameUpdate:
2369 /* update cluster name */
2370 if (st->cluster_name) {
2371 len = sizeof(bms->cluster_name);
2372 memset((char *)bms->cluster_name, 0, len);
2373 strncpy((char *)bms->cluster_name,
2374 st->cluster_name, len);
2375 bms->cluster_name[len - 1] = '\0';
2376 }
2377 break;
2378 case NodeNumUpdate:
2379 /* cluster md only supports superblock 1.2 now */
2380 if (st->minor_version != 2 && bms->version == BITMAP_MAJOR_CLUSTERED) {
2381 pr_err("Warning: cluster md only works with superblock 1.2\n");
2382 return -EINVAL;
2383 }
2384
2385 if (bms->version == BITMAP_MAJOR_CLUSTERED) {
2386 if (__cpu_to_le32(st->nodes) < bms->nodes) {
2387 /* Since the nodes num is not increased, no need to check the space
2388 * is enough or not, just update bms->nodes */
2389 bms->nodes = __cpu_to_le32(st->nodes);
2390 break;
2391 }
2392 } else {
2393 /* no need to change bms->nodes for other bitmap types */
2394 if (st->nodes)
2395 pr_err("Warning: --nodes option is only suitable for clustered bitmap\n");
2396 break;
2397 }
2398
2399 /* Each node has an independent bitmap, it is necessary to calculate the
2400 * space is enough or not, first get how many bytes for the total bitmap */
2401 bm_space_per_node = calc_bitmap_size(bms, 4096);
2402
2403 total_bm_space = 512 * (__le64_to_cpu(sb->data_offset) - __le64_to_cpu(sb->super_offset));
2404 total_bm_space = total_bm_space - 4096; /* leave another 4k for superblock */
2405
2406 if (bm_space_per_node * st->nodes > total_bm_space) {
2407 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2408 total_bm_space / bm_space_per_node);
2409 return -ENOMEM;
2410 }
2411
2412 bms->nodes = __cpu_to_le32(st->nodes);
2413 break;
2414 case NoUpdate:
2415 default:
2416 break;
2417 }
2418
2419 init_afd(&afd, fd);
2420
2421 locate_bitmap1(st, fd, 0);
2422
2423 if (posix_memalign(&buf, 4096, 4096))
2424 return -ENOMEM;
2425
2426 do {
2427 /* Only the bitmap[0] should resync
2428 * whole device on initial assembly
2429 */
2430 if (i)
2431 memset(buf, 0x00, 4096);
2432 else
2433 memset(buf, 0xff, 4096);
2434 memcpy(buf, (char *)bms, sizeof(bitmap_super_t));
2435
2436 /*
2437 * use 4096 boundary if bitmap_offset is aligned
2438 * with 8 sectors, then it should compatible with
2439 * older mdadm.
2440 */
2441 if (__le32_to_cpu(sb->bitmap_offset) & 7)
2442 towrite = calc_bitmap_size(bms, 512);
2443 else
2444 towrite = calc_bitmap_size(bms, 4096);
2445 while (towrite > 0) {
2446 n = towrite;
2447 if (n > 4096)
2448 n = 4096;
2449 n = awrite(&afd, buf, n);
2450 if (n > 0)
2451 towrite -= n;
2452 else
2453 break;
2454 if (i)
2455 memset(buf, 0x00, 4096);
2456 else
2457 memset(buf, 0xff, 4096);
2458 }
2459 fsync(fd);
2460 if (towrite) {
2461 rv = -2;
2462 break;
2463 }
2464 } while (++i < __le32_to_cpu(bms->nodes));
2465
2466 free(buf);
2467 return rv;
2468 }
2469
2470 static void free_super1(struct supertype *st)
2471 {
2472
2473 if (st->sb)
2474 free(st->sb);
2475 while (st->info) {
2476 struct devinfo *di = st->info;
2477 st->info = di->next;
2478 if (di->fd >= 0)
2479 close(di->fd);
2480 free(di);
2481 }
2482 st->sb = NULL;
2483 }
2484
2485 #ifndef MDASSEMBLE
2486 static int validate_geometry1(struct supertype *st, int level,
2487 int layout, int raiddisks,
2488 int *chunk, unsigned long long size,
2489 unsigned long long data_offset,
2490 char *subdev, unsigned long long *freesize,
2491 int consistency_policy, int verbose)
2492 {
2493 unsigned long long ldsize, devsize;
2494 int bmspace;
2495 unsigned long long headroom;
2496 int fd;
2497
2498 if (level == LEVEL_CONTAINER) {
2499 if (verbose)
2500 pr_err("1.x metadata does not support containers\n");
2501 return 0;
2502 }
2503 if (*chunk == UnSet)
2504 *chunk = DEFAULT_CHUNK;
2505
2506 if (!subdev)
2507 return 1;
2508
2509 if (st->minor_version < 0)
2510 /* not specified, so time to set default */
2511 st->minor_version = 2;
2512
2513 fd = open(subdev, O_RDONLY|O_EXCL, 0);
2514 if (fd < 0) {
2515 if (verbose)
2516 pr_err("super1.x cannot open %s: %s\n",
2517 subdev, strerror(errno));
2518 return 0;
2519 }
2520
2521 if (!get_dev_size(fd, subdev, &ldsize)) {
2522 close(fd);
2523 return 0;
2524 }
2525 close(fd);
2526
2527 devsize = ldsize >> 9;
2528 if (devsize < 24) {
2529 *freesize = 0;
2530 return 0;
2531 }
2532
2533 /* creating: allow suitable space for bitmap */
2534 bmspace = choose_bm_space(devsize);
2535
2536 if (data_offset == INVALID_SECTORS)
2537 data_offset = st->data_offset;
2538 if (data_offset == INVALID_SECTORS)
2539 switch (st->minor_version) {
2540 case 0:
2541 data_offset = 0;
2542 break;
2543 case 1:
2544 case 2:
2545 /* Choose data offset appropriate for this device
2546 * and use as default for whole array.
2547 * The data_offset must allow for bitmap space
2548 * and base metadata, should allow for some headroom
2549 * for reshape, and should be rounded to multiple
2550 * of 1M.
2551 * Headroom is limited to 128M, but aim for about 0.1%
2552 */
2553 headroom = 128*1024*2;
2554 while ((headroom << 10) > devsize &&
2555 (*chunk == 0 ||
2556 headroom / 2 >= ((unsigned)(*chunk)*2)*2))
2557 headroom >>= 1;
2558 data_offset = 12*2 + bmspace + headroom;
2559 #define ONE_MEG (2*1024)
2560 if (data_offset > ONE_MEG)
2561 data_offset = (data_offset / ONE_MEG) * ONE_MEG;
2562 break;
2563 }
2564 if (st->data_offset == INVALID_SECTORS)
2565 st->data_offset = data_offset;
2566 switch(st->minor_version) {
2567 case 0: /* metadata at end. Round down and subtract space to reserve */
2568 devsize = (devsize & ~(4ULL*2-1));
2569 /* space for metadata, bblog, bitmap */
2570 devsize -= 8*2 + 8 + bmspace;
2571 break;
2572 case 1:
2573 case 2:
2574 devsize -= data_offset;
2575 break;
2576 }
2577 *freesize = devsize;
2578 return 1;
2579 }
2580 #endif /* MDASSEMBLE */
2581
2582 void *super1_make_v0(struct supertype *st, struct mdinfo *info, mdp_super_t *sb0)
2583 {
2584 /* Create a v1.0 superblock based on 'info'*/
2585 void *ret;
2586 struct mdp_superblock_1 *sb;
2587 int i;
2588 unsigned long long offset;
2589
2590 if (posix_memalign(&ret, 4096, 1024) != 0)
2591 return NULL;
2592 sb = ret;
2593 memset(ret, 0, 1024);
2594 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
2595 sb->major_version = __cpu_to_le32(1);
2596
2597 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
2598 sprintf(sb->set_name, "%d", sb0->md_minor);
2599 sb->ctime = __cpu_to_le32(info->array.ctime+1);
2600 sb->level = __cpu_to_le32(info->array.level);
2601 sb->layout = __cpu_to_le32(info->array.layout);
2602 sb->size = __cpu_to_le64(info->component_size);
2603 sb->chunksize = __cpu_to_le32(info->array.chunk_size/512);
2604 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
2605 if (info->array.level > 0)
2606 sb->data_size = sb->size;
2607 else
2608 sb->data_size = st->ss->avail_size(st, st->devsize/512, 0);
2609 sb->resync_offset = MaxSector;
2610 sb->max_dev = __cpu_to_le32(MD_SB_DISKS);
2611 sb->dev_number = __cpu_to_le32(info->disk.number);
2612 sb->utime = __cpu_to_le64(info->array.utime);
2613
2614 offset = st->devsize/512 - 8*2;
2615 offset &= ~(4*2-1);
2616 sb->super_offset = __cpu_to_le64(offset);
2617 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2618
2619 random_uuid(sb->device_uuid);
2620
2621 for (i = 0; i < MD_SB_DISKS; i++) {
2622 int state = sb0->disks[i].state;
2623 sb->dev_roles[i] = MD_DISK_ROLE_SPARE;
2624 if ((state & (1<<MD_DISK_SYNC)) &&
2625 !(state & (1<<MD_DISK_FAULTY)))
2626 sb->dev_roles[i] = __cpu_to_le16(sb0->disks[i].raid_disk);
2627 }
2628 sb->sb_csum = calc_sb_1_csum(sb);
2629 return ret;
2630 }
2631
2632 struct superswitch super1 = {
2633 #ifndef MDASSEMBLE
2634 .examine_super = examine_super1,
2635 .brief_examine_super = brief_examine_super1,
2636 .export_examine_super = export_examine_super1,
2637 .detail_super = detail_super1,
2638 .brief_detail_super = brief_detail_super1,
2639 .export_detail_super = export_detail_super1,
2640 .write_init_super = write_init_super1,
2641 .validate_geometry = validate_geometry1,
2642 .add_to_super = add_to_super1,
2643 .examine_badblocks = examine_badblocks_super1,
2644 .copy_metadata = copy_metadata1,
2645 #endif
2646 .match_home = match_home1,
2647 .uuid_from_super = uuid_from_super1,
2648 .getinfo_super = getinfo_super1,
2649 .container_content = container_content1,
2650 .update_super = update_super1,
2651 .init_super = init_super1,
2652 .store_super = store_super1,
2653 .compare_super = compare_super1,
2654 .load_super = load_super1,
2655 .match_metadata_desc = match_metadata_desc1,
2656 .avail_size = avail_size1,
2657 .add_internal_bitmap = add_internal_bitmap1,
2658 .locate_bitmap = locate_bitmap1,
2659 .write_bitmap = write_bitmap1,
2660 .free_super = free_super1,
2661 #if __BYTE_ORDER == BIG_ENDIAN
2662 .swapuuid = 0,
2663 #else
2664 .swapuuid = 1,
2665 #endif
2666 .name = "1.x",
2667 };
Unnamed repository; edit this file 'description' to name the repository.