]> 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
Create.c: fix uclibc build
[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; /* used for raid5, raid6, raid10, and raid0 */
47 __u64 size; /* used size of component devices, in 512byte sectors */
48
49 __u32 chunksize; /* in 512byte sectors */
50 __u32 raid_disks;
51 union {
52 __u32 bitmap_offset; /* sectors after start of superblock that bitmap starts
53 * NOTE: signed, so bitmap can be before superblock
54 * only meaningful of feature_map[0] is set.
55 */
56
57 /* only meaningful when feature_map[MD_FEATURE_PPL] is set */
58 struct {
59 __s16 offset; /* sectors from start of superblock that ppl starts */
60 __u16 size; /* ppl size in sectors */
61 } ppl;
62 };
63
64 /* These are only valid with feature bit '4' */
65 __u32 new_level; /* new level we are reshaping to */
66 __u64 reshape_position; /* next address in array-space for reshape */
67 __u32 delta_disks; /* change in number of raid_disks */
68 __u32 new_layout; /* new layout */
69 __u32 new_chunk; /* new chunk size (sectors) */
70 __u32 new_offset; /* signed number to add to data_offset in new
71 * layout. 0 == no-change. This can be
72 * different on each device in the array.
73 */
74
75 /* constant this-device information - 64 bytes */
76 __u64 data_offset; /* sector start of data, often 0 */
77 __u64 data_size; /* sectors in this device that can be used for data */
78 __u64 super_offset; /* sector start of this superblock */
79 union {
80 __u64 recovery_offset;/* sectors before this offset (from data_offset) have been recovered */
81 __u64 journal_tail;/* journal tail of journal device (from data_offset) */
82 };
83 __u32 dev_number; /* permanent identifier of this device - not role in raid */
84 __u32 cnt_corrected_read; /* number of read errors that were corrected by re-writing */
85 __u8 device_uuid[16]; /* user-space setable, ignored by kernel */
86 __u8 devflags; /* per-device flags. Only one defined...*/
87 #define WriteMostly1 1 /* mask for writemostly flag in above */
88 #define FailFast1 2 /* Device should get FailFast requests */
89 /* bad block log. If there are any bad blocks the feature flag is set.
90 * if offset and size are non-zero, that space is reserved and available.
91 */
92 __u8 bblog_shift; /* shift from sectors to block size for badblock list */
93 __u16 bblog_size; /* number of sectors reserved for badblock list */
94 __u32 bblog_offset; /* sector offset from superblock to bblog, signed */
95
96 /* array state information - 64 bytes */
97 __u64 utime; /* 40 bits second, 24 bits microseconds */
98 __u64 events; /* incremented when superblock updated */
99 __u64 resync_offset; /* data before this offset (from data_offset) known to be in sync */
100 __u32 sb_csum; /* checksum upto dev_roles[max_dev] */
101 __u32 max_dev; /* size of dev_roles[] array to consider */
102 __u8 pad3[64-32]; /* set to 0 when writing */
103
104 /* device state information. Indexed by dev_number.
105 * 2 bytes per device
106 * Note there are no per-device state flags. State information is rolled
107 * into the 'roles' value. If a device is spare or faulty, then it doesn't
108 * have a meaningful role.
109 */
110 __u16 dev_roles[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
111 };
112
113 #define MAX_SB_SIZE 4096
114 /* bitmap super size is 256, but we round up to a sector for alignment */
115 #define BM_SUPER_SIZE 512
116 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
117 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
118 + sizeof(struct misc_dev_info))
119
120 struct misc_dev_info {
121 __u64 device_size;
122 };
123
124 #define MULTIPLE_PPL_AREA_SIZE_SUPER1 (1024 * 1024) /* Size of the whole
125 * mutliple PPL area
126 */
127 /* feature_map bits */
128 #define MD_FEATURE_BITMAP_OFFSET 1
129 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
130 * must be honoured
131 */
132 #define MD_FEATURE_RESHAPE_ACTIVE 4
133 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
134 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
135 * active device with same 'role'.
136 * 'recovery_offset' is also set.
137 */
138 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
139 * of devices, but is going
140 * backwards anyway.
141 */
142 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
143 #define MD_FEATURE_BITMAP_VERSIONED 256 /* bitmap version number checked properly */
144 #define MD_FEATURE_JOURNAL 512 /* support write journal */
145 #define MD_FEATURE_PPL 1024 /* support PPL */
146 #define MD_FEATURE_MUTLIPLE_PPLS 2048 /* support for multiple PPLs */
147 #define MD_FEATURE_RAID0_LAYOUT 4096 /* layout is meaningful in RAID0 */
148 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
149 |MD_FEATURE_RECOVERY_OFFSET \
150 |MD_FEATURE_RESHAPE_ACTIVE \
151 |MD_FEATURE_BAD_BLOCKS \
152 |MD_FEATURE_REPLACEMENT \
153 |MD_FEATURE_RESHAPE_BACKWARDS \
154 |MD_FEATURE_NEW_OFFSET \
155 |MD_FEATURE_BITMAP_VERSIONED \
156 |MD_FEATURE_JOURNAL \
157 |MD_FEATURE_PPL \
158 |MD_FEATURE_MULTIPLE_PPLS \
159 |MD_FEATURE_RAID0_LAYOUT \
160 )
161
162 static int role_from_sb(struct mdp_superblock_1 *sb)
163 {
164 unsigned int d;
165 int role;
166
167 d = __le32_to_cpu(sb->dev_number);
168 if (d < __le32_to_cpu(sb->max_dev))
169 role = __le16_to_cpu(sb->dev_roles[d]);
170 else
171 role = MD_DISK_ROLE_SPARE;
172 return role;
173 }
174
175 /* return how many bytes are needed for bitmap, for cluster-md each node
176 * should have it's own bitmap */
177 static unsigned int calc_bitmap_size(bitmap_super_t *bms, unsigned int boundary)
178 {
179 unsigned long long bits, bytes;
180
181 bits = bitmap_bits(__le64_to_cpu(bms->sync_size),
182 __le32_to_cpu(bms->chunksize));
183 bytes = (bits+7) >> 3;
184 bytes += sizeof(bitmap_super_t);
185 bytes = ROUND_UP(bytes, boundary);
186
187 return bytes;
188 }
189
190 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
191 {
192 unsigned int disk_csum, csum;
193 unsigned long long newcsum;
194 int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2;
195 unsigned int *isuper = (unsigned int *)sb;
196
197 /* make sure I can count... */
198 if (offsetof(struct mdp_superblock_1,data_offset) != 128 ||
199 offsetof(struct mdp_superblock_1, utime) != 192 ||
200 sizeof(struct mdp_superblock_1) != 256) {
201 fprintf(stderr, "WARNING - superblock isn't sized correctly\n");
202 }
203
204 disk_csum = sb->sb_csum;
205 sb->sb_csum = 0;
206 newcsum = 0;
207 for (; size >= 4; size -= 4) {
208 newcsum += __le32_to_cpu(*isuper);
209 isuper++;
210 }
211
212 if (size == 2)
213 newcsum += __le16_to_cpu(*(unsigned short*) isuper);
214
215 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
216 sb->sb_csum = disk_csum;
217 return __cpu_to_le32(csum);
218 }
219
220 /*
221 * Information related to file descriptor used for aligned reads/writes.
222 * Cache the block size.
223 */
224 struct align_fd {
225 int fd;
226 int blk_sz;
227 };
228
229 static void init_afd(struct align_fd *afd, int fd)
230 {
231 afd->fd = fd;
232 if (!get_dev_sector_size(afd->fd, NULL, (unsigned int *)&afd->blk_sz))
233 afd->blk_sz = 512;
234 }
235
236 static char abuf[4096+4096];
237
238 static int aread(struct align_fd *afd, void *buf, int len)
239 {
240 /* aligned read.
241 * On devices with a 4K sector size, we need to read
242 * the full sector and copy relevant bits into
243 * the buffer
244 */
245 int bsize, iosize;
246 char *b;
247 int n;
248
249 bsize = afd->blk_sz;
250
251 if (!bsize || bsize > 4096 || len > 4096) {
252 if (!bsize)
253 fprintf(stderr, "WARNING - aread() called with invalid block size\n");
254 return -1;
255 }
256 b = ROUND_UP_PTR((char *)abuf, 4096);
257
258 for (iosize = 0; iosize < len; iosize += bsize)
259 ;
260 n = read(afd->fd, b, iosize);
261 if (n <= 0)
262 return n;
263 lseek(afd->fd, len - n, 1);
264 if (n > len)
265 n = len;
266 memcpy(buf, b, n);
267 return n;
268 }
269
270 static int awrite(struct align_fd *afd, void *buf, int len)
271 {
272 /* aligned write.
273 * On devices with a 4K sector size, we need to write
274 * the full sector. We pre-read if the sector is larger
275 * than the write.
276 * The address must be sector-aligned.
277 */
278 int bsize, iosize;
279 char *b;
280 int n;
281
282 bsize = afd->blk_sz;
283 if (!bsize || bsize > 4096 || len > 4096) {
284 if (!bsize)
285 fprintf(stderr, "WARNING - awrite() called with invalid block size\n");
286 return -1;
287 }
288 b = ROUND_UP_PTR((char *)abuf, 4096);
289
290 for (iosize = 0; iosize < len ; iosize += bsize)
291 ;
292
293 if (len != iosize) {
294 n = read(afd->fd, b, iosize);
295 if (n <= 0)
296 return n;
297 lseek(afd->fd, -n, 1);
298 }
299
300 memcpy(b, buf, len);
301 n = write(afd->fd, b, iosize);
302 if (n <= 0)
303 return n;
304 lseek(afd->fd, len - n, 1);
305 return len;
306 }
307
308 static inline unsigned int md_feature_any_ppl_on(__u32 feature_map)
309 {
310 return ((__cpu_to_le32(feature_map) &
311 (MD_FEATURE_PPL | MD_FEATURE_MUTLIPLE_PPLS)));
312 }
313
314 static inline unsigned int choose_ppl_space(int chunk)
315 {
316 return (PPL_HEADER_SIZE >> 9) + (chunk > 128*2 ? chunk : 128*2);
317 }
318
319 static void examine_super1(struct supertype *st, char *homehost)
320 {
321 struct mdp_superblock_1 *sb = st->sb;
322 bitmap_super_t *bms = (bitmap_super_t *)(((char *)sb) + MAX_SB_SIZE);
323 time_t atime;
324 unsigned int d;
325 int role;
326 int delta_extra = 0;
327 int i;
328 char *c;
329 int l = homehost ? strlen(homehost) : 0;
330 int layout;
331 unsigned long long sb_offset;
332 struct mdinfo info;
333 int inconsistent = 0;
334
335 printf(" Magic : %08x\n", __le32_to_cpu(sb->magic));
336 printf(" Version : 1");
337 sb_offset = __le64_to_cpu(sb->super_offset);
338 if (sb_offset <= 4)
339 printf(".1\n");
340 else if (sb_offset <= 8)
341 printf(".2\n");
342 else
343 printf(".0\n");
344 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb->feature_map));
345 printf(" Array UUID : ");
346 for (i = 0; i < 16; i++) {
347 if ((i & 3) == 0 && i != 0)
348 printf(":");
349 printf("%02x", sb->set_uuid[i]);
350 }
351 printf("\n");
352 printf(" Name : %.32s", sb->set_name);
353 if (l > 0 && l < 32 &&
354 sb->set_name[l] == ':' &&
355 strncmp(sb->set_name, homehost, l) == 0)
356 printf(" (local to host %s)", homehost);
357 printf("\n");
358 if (bms->nodes > 0 &&
359 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
360 printf(" Cluster Name : %-64s\n", bms->cluster_name);
361 atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL;
362 printf(" Creation Time : %.24s\n", ctime(&atime));
363 c=map_num(pers, __le32_to_cpu(sb->level));
364 printf(" Raid Level : %s\n", c?c:"-unknown-");
365 printf(" Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks));
366 printf("\n");
367 printf(" Avail Dev Size : %llu sectors%s\n",
368 (unsigned long long)__le64_to_cpu(sb->data_size),
369 human_size(__le64_to_cpu(sb->data_size)<<9));
370 if (__le32_to_cpu(sb->level) > 0) {
371 int ddsks = 0, ddsks_denom = 1;
372 switch(__le32_to_cpu(sb->level)) {
373 case 1: ddsks=1;break;
374 case 4:
375 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
376 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
377 case 10:
378 layout = __le32_to_cpu(sb->layout);
379 ddsks = __le32_to_cpu(sb->raid_disks);
380 ddsks_denom = (layout&255) * ((layout>>8)&255);
381 }
382 if (ddsks) {
383 long long asize = __le64_to_cpu(sb->size);
384 asize = (asize << 9) * ddsks / ddsks_denom;
385 printf(" Array Size : %llu KiB%s\n",
386 asize >> 10, human_size(asize));
387 }
388 if (sb->size != sb->data_size)
389 printf(" Used Dev Size : %llu sectors%s\n",
390 (unsigned long long)__le64_to_cpu(sb->size),
391 human_size(__le64_to_cpu(sb->size)<<9));
392 }
393 if (sb->data_offset)
394 printf(" Data Offset : %llu sectors\n",
395 (unsigned long long)__le64_to_cpu(sb->data_offset));
396 if (sb->new_offset &&
397 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) {
398 unsigned long long offset = __le64_to_cpu(sb->data_offset);
399 offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset);
400 printf(" New Offset : %llu sectors\n", offset);
401 }
402 printf(" Super Offset : %llu sectors\n",
403 (unsigned long long)__le64_to_cpu(sb->super_offset));
404 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)
405 printf("Recovery Offset : %llu sectors\n",
406 (unsigned long long)__le64_to_cpu(sb->recovery_offset));
407
408 st->ss->getinfo_super(st, &info, NULL);
409 if (info.space_after != 1 &&
410 !(__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) {
411 printf(" Unused Space : before=%llu sectors, ",
412 info.space_before);
413 if (info.space_after < INT64_MAX)
414 printf("after=%llu sectors\n", info.space_after);
415 else
416 printf("after=-%llu sectors DEVICE TOO SMALL\n",
417 UINT64_MAX - info.space_after);
418 }
419 printf(" State : %s%s\n",
420 (__le64_to_cpu(sb->resync_offset) + 1) ? "active":"clean",
421 (info.space_after > INT64_MAX) ? " TRUNCATED DEVICE" : "");
422 printf(" Device UUID : ");
423 for (i = 0; i < 16; i++) {
424 if ((i & 3)==0 && i != 0)
425 printf(":");
426 printf("%02x", sb->device_uuid[i]);
427 }
428 printf("\n");
429 printf("\n");
430 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
431 printf("Internal Bitmap : %ld sectors from superblock\n",
432 (long)(int32_t)__le32_to_cpu(sb->bitmap_offset));
433 } else if (md_feature_any_ppl_on(sb->feature_map)) {
434 printf(" PPL : %u sectors at offset %d sectors from superblock\n",
435 __le16_to_cpu(sb->ppl.size),
436 __le16_to_cpu(sb->ppl.offset));
437 }
438 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)) {
439 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)
440 __le64_to_cpu(sb->reshape_position)/2,
441 human_size(__le64_to_cpu(sb->reshape_position)<<9));
442 if (__le32_to_cpu(sb->delta_disks)) {
443 printf(" Delta Devices : %d",
444 __le32_to_cpu(sb->delta_disks));
445 printf(" (%d->%d)\n",
446 __le32_to_cpu(sb->raid_disks) -
447 __le32_to_cpu(sb->delta_disks),
448 __le32_to_cpu(sb->raid_disks));
449 if ((int)__le32_to_cpu(sb->delta_disks) < 0)
450 delta_extra = -__le32_to_cpu(sb->delta_disks);
451 }
452 if (__le32_to_cpu(sb->new_level) != __le32_to_cpu(sb->level)) {
453 c = map_num(pers, __le32_to_cpu(sb->new_level));
454 printf(" New Level : %s\n", c?c:"-unknown-");
455 }
456 if (__le32_to_cpu(sb->new_layout) !=
457 __le32_to_cpu(sb->layout)) {
458 if (__le32_to_cpu(sb->level) == 5) {
459 c = map_num(r5layout,
460 __le32_to_cpu(sb->new_layout));
461 printf(" New Layout : %s\n", c?c:"-unknown-");
462 }
463 if (__le32_to_cpu(sb->level) == 6) {
464 c = map_num(r6layout,
465 __le32_to_cpu(sb->new_layout));
466 printf(" New Layout : %s\n", c?c:"-unknown-");
467 }
468 if (__le32_to_cpu(sb->level) == 10) {
469 printf(" New Layout :");
470 print_r10_layout(__le32_to_cpu(sb->new_layout));
471 printf("\n");
472 }
473 }
474 if (__le32_to_cpu(sb->new_chunk) !=
475 __le32_to_cpu(sb->chunksize))
476 printf(" New Chunksize : %dK\n",
477 __le32_to_cpu(sb->new_chunk)/2);
478 printf("\n");
479 }
480 if (sb->devflags) {
481 printf(" Flags :");
482 if (sb->devflags & WriteMostly1)
483 printf(" write-mostly");
484 if (sb->devflags & FailFast1)
485 printf(" failfast");
486 printf("\n");
487 }
488
489 atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL;
490 printf(" Update Time : %.24s\n", ctime(&atime));
491
492 if (sb->bblog_size && sb->bblog_offset) {
493 printf(" Bad Block Log : %d entries available at offset %ld sectors",
494 __le16_to_cpu(sb->bblog_size)*512/8,
495 (long)(int32_t)__le32_to_cpu(sb->bblog_offset));
496 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
497 printf(" - bad blocks present.");
498 printf("\n");
499 }
500
501 if (calc_sb_1_csum(sb) == sb->sb_csum)
502 printf(" Checksum : %x - correct\n",
503 __le32_to_cpu(sb->sb_csum));
504 else
505 printf(" Checksum : %x - expected %x\n",
506 __le32_to_cpu(sb->sb_csum),
507 __le32_to_cpu(calc_sb_1_csum(sb)));
508 printf(" Events : %llu\n",
509 (unsigned long long)__le64_to_cpu(sb->events));
510 printf("\n");
511 if (__le32_to_cpu(sb->level) == 0 &&
512 (sb->feature_map & __cpu_to_le32(MD_FEATURE_RAID0_LAYOUT))) {
513 c = map_num(r0layout, __le32_to_cpu(sb->layout));
514 printf(" Layout : %s\n", c?c:"-unknown-");
515 }
516 if (__le32_to_cpu(sb->level) == 5) {
517 c = map_num(r5layout, __le32_to_cpu(sb->layout));
518 printf(" Layout : %s\n", c?c:"-unknown-");
519 }
520 if (__le32_to_cpu(sb->level) == 6) {
521 c = map_num(r6layout, __le32_to_cpu(sb->layout));
522 printf(" Layout : %s\n", c?c:"-unknown-");
523 }
524 if (__le32_to_cpu(sb->level) == 10) {
525 int lo = __le32_to_cpu(sb->layout);
526 printf(" Layout :");
527 print_r10_layout(lo);
528 printf("\n");
529 }
530 switch(__le32_to_cpu(sb->level)) {
531 case 0:
532 case 4:
533 case 5:
534 case 6:
535 case 10:
536 printf(" Chunk Size : %dK\n",
537 __le32_to_cpu(sb->chunksize)/2);
538 break;
539 case -1:
540 printf(" Rounding : %dK\n",
541 __le32_to_cpu(sb->chunksize)/2);
542 break;
543 default:
544 break;
545 }
546 printf("\n");
547 printf(" Device Role : ");
548 role = role_from_sb(sb);
549 if (role >= MD_DISK_ROLE_FAULTY)
550 printf("spare\n");
551 else if (role == MD_DISK_ROLE_JOURNAL)
552 printf("Journal\n");
553 else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_REPLACEMENT))
554 printf("Replacement device %d\n", role);
555 else
556 printf("Active device %d\n", role);
557
558 printf(" Array State : ");
559 for (d = 0; d < __le32_to_cpu(sb->raid_disks) + delta_extra; d++) {
560 int cnt = 0;
561 unsigned int i;
562 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
563 unsigned int role = __le16_to_cpu(sb->dev_roles[i]);
564 if (role == d)
565 cnt++;
566 }
567 if (cnt == 2 && __le32_to_cpu(sb->level) > 0)
568 printf("R");
569 else if (cnt == 1)
570 printf("A");
571 else if (cnt == 0)
572 printf(".");
573 else {
574 printf("?");
575 inconsistent = 1;
576 }
577 }
578 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
579 printf("\n");
580 for (d = 0; d < __le32_to_cpu(sb->max_dev); d++) {
581 unsigned int r = __le16_to_cpu(sb->dev_roles[d]);
582 if (r <= MD_DISK_ROLE_MAX &&
583 r > __le32_to_cpu(sb->raid_disks) + delta_extra)
584 inconsistent = 1;
585 }
586 if (inconsistent) {
587 printf("WARNING Array state is inconsistent - each number should appear only once\n");
588 for (d = 0; d < __le32_to_cpu(sb->max_dev); d++)
589 if (__le16_to_cpu(sb->dev_roles[d]) >=
590 MD_DISK_ROLE_FAULTY)
591 printf(" %d:-", d);
592 else
593 printf(" %d:%d", d,
594 __le16_to_cpu(sb->dev_roles[d]));
595 printf("\n");
596 }
597 }
598
599 static void brief_examine_super1(struct supertype *st, int verbose)
600 {
601 struct mdp_superblock_1 *sb = st->sb;
602 int i;
603 unsigned long long sb_offset;
604 char *nm;
605 char *c = map_num(pers, __le32_to_cpu(sb->level));
606
607 nm = strchr(sb->set_name, ':');
608 if (nm)
609 nm++;
610 else if (sb->set_name[0])
611 nm = sb->set_name;
612 else
613 nm = NULL;
614
615 printf("ARRAY ");
616 if (nm) {
617 printf(DEV_MD_DIR "%s", nm);
618 putchar(' ');
619 }
620 if (verbose && c)
621 printf(" level=%s", c);
622 sb_offset = __le64_to_cpu(sb->super_offset);
623 if (sb_offset <= 4)
624 printf(" metadata=1.1 ");
625 else if (sb_offset <= 8)
626 printf(" metadata=1.2 ");
627 else
628 printf(" metadata=1.0 ");
629 if (verbose)
630 printf("num-devices=%d ", __le32_to_cpu(sb->raid_disks));
631 printf("UUID=");
632 for (i = 0; i < 16; i++) {
633 if ((i & 3)==0 && i != 0)
634 printf(":");
635 printf("%02x", sb->set_uuid[i]);
636 }
637 printf("\n");
638 }
639
640 static void export_examine_super1(struct supertype *st)
641 {
642 struct mdp_superblock_1 *sb = st->sb;
643 int i;
644 int len = 32;
645 int layout;
646
647 printf("MD_LEVEL=%s\n", map_num_s(pers, __le32_to_cpu(sb->level)));
648 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb->raid_disks));
649 for (i = 0; i < 32; i++)
650 if (sb->set_name[i] == '\n' || sb->set_name[i] == '\0') {
651 len = i;
652 break;
653 }
654 if (len)
655 printf("MD_NAME=%.*s\n", len, sb->set_name);
656 if (__le32_to_cpu(sb->level) > 0) {
657 int ddsks = 0, ddsks_denom = 1;
658 switch(__le32_to_cpu(sb->level)) {
659 case 1:
660 ddsks = 1;
661 break;
662 case 4:
663 case 5:
664 ddsks = __le32_to_cpu(sb->raid_disks)-1;
665 break;
666 case 6:
667 ddsks = __le32_to_cpu(sb->raid_disks)-2;
668 break;
669 case 10:
670 layout = __le32_to_cpu(sb->layout);
671 ddsks = __le32_to_cpu(sb->raid_disks);
672 ddsks_denom = (layout&255) * ((layout>>8)&255);
673 }
674 if (ddsks) {
675 long long asize = __le64_to_cpu(sb->size);
676 asize = (asize << 9) * ddsks / ddsks_denom;
677 printf("MD_ARRAY_SIZE=%s\n",
678 human_size_brief(asize, JEDEC));
679 }
680 }
681 printf("MD_UUID=");
682 for (i = 0; i < 16; i++) {
683 if ((i & 3) == 0 && i != 0)
684 printf(":");
685 printf("%02x", sb->set_uuid[i]);
686 }
687 printf("\n");
688 printf("MD_UPDATE_TIME=%llu\n",
689 __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL);
690 printf("MD_DEV_UUID=");
691 for (i = 0; i < 16; i++) {
692 if ((i & 3) == 0 && i != 0)
693 printf(":");
694 printf("%02x", sb->device_uuid[i]);
695 }
696 printf("\n");
697 printf("MD_EVENTS=%llu\n",
698 (unsigned long long)__le64_to_cpu(sb->events));
699 }
700
701 static int copy_metadata1(struct supertype *st, int from, int to)
702 {
703 /* Read superblock. If it looks good, write it out.
704 * Then if a bitmap is present, copy that.
705 * And if a bad-block-list is present, copy that too.
706 */
707 void *buf;
708 unsigned long long dsize, sb_offset;
709 const int bufsize = 4*1024;
710 struct mdp_superblock_1 super, *sb;
711
712 if (posix_memalign(&buf, 4096, bufsize) != 0)
713 return 1;
714
715 if (!get_dev_size(from, NULL, &dsize))
716 goto err;
717
718 dsize >>= 9;
719 if (dsize < 24)
720 goto err;
721 switch(st->minor_version) {
722 case 0:
723 sb_offset = dsize;
724 sb_offset -= 8*2;
725 sb_offset &= ~(4*2-1);
726 break;
727 case 1:
728 sb_offset = 0;
729 break;
730 case 2:
731 sb_offset = 4*2;
732 break;
733 default:
734 goto err;
735 }
736
737 if (lseek64(from, sb_offset << 9, 0) < 0LL)
738 goto err;
739 if (read(from, buf, bufsize) != bufsize)
740 goto err;
741
742 sb = buf;
743 super = *sb; // save most of sb for when we reuse buf
744
745 if (__le32_to_cpu(super.magic) != MD_SB_MAGIC ||
746 __le32_to_cpu(super.major_version) != 1 ||
747 __le64_to_cpu(super.super_offset) != sb_offset ||
748 calc_sb_1_csum(sb) != super.sb_csum)
749 goto err;
750
751 if (lseek64(to, sb_offset << 9, 0) < 0LL)
752 goto err;
753 if (write(to, buf, bufsize) != bufsize)
754 goto err;
755
756 if (super.feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
757 unsigned long long bitmap_offset = sb_offset;
758 int bytes = 4096; // just an estimate.
759 int written = 0;
760 struct align_fd afrom, ato;
761
762 init_afd(&afrom, from);
763 init_afd(&ato, to);
764
765 bitmap_offset += (int32_t)__le32_to_cpu(super.bitmap_offset);
766
767 if (lseek64(from, bitmap_offset<<9, 0) < 0)
768 goto err;
769 if (lseek64(to, bitmap_offset<<9, 0) < 0)
770 goto err;
771
772 for (written = 0; written < bytes ; ) {
773 int n = bytes - written;
774 if (n > 4096)
775 n = 4096;
776 if (aread(&afrom, buf, n) != n)
777 goto err;
778 if (written == 0) {
779 /* have the header, can calculate
780 * correct bitmap bytes */
781 bitmap_super_t *bms;
782 bms = (void *)buf;
783 bytes = calc_bitmap_size(bms, 512);
784 if (n > bytes)
785 n = bytes;
786 }
787 if (awrite(&ato, buf, n) != n)
788 goto err;
789 written += n;
790 }
791 }
792
793 if (super.bblog_size != 0 &&
794 __le16_to_cpu(super.bblog_size) <= 100 &&
795 super.bblog_offset != 0 &&
796 (super.feature_map & __le32_to_cpu(MD_FEATURE_BAD_BLOCKS))) {
797 /* There is a bad block log */
798 unsigned long long bb_offset = sb_offset;
799 int bytes = __le16_to_cpu(super.bblog_size) * 512;
800 int written = 0;
801 struct align_fd afrom, ato;
802
803 init_afd(&afrom, from);
804 init_afd(&ato, to);
805
806 bb_offset += (int32_t)__le32_to_cpu(super.bblog_offset);
807
808 if (lseek64(from, bb_offset<<9, 0) < 0)
809 goto err;
810 if (lseek64(to, bb_offset<<9, 0) < 0)
811 goto err;
812
813 for (written = 0; written < bytes ; ) {
814 int n = bytes - written;
815 if (n > 4096)
816 n = 4096;
817 if (aread(&afrom, buf, n) != n)
818 goto err;
819
820 if (awrite(&ato, buf, n) != n)
821 goto err;
822 written += n;
823 }
824 }
825
826 free(buf);
827 return 0;
828
829 err:
830 free(buf);
831 return 1;
832 }
833
834 static void detail_super1(struct supertype *st, char *homehost, char *subarray)
835 {
836 struct mdp_superblock_1 *sb = st->sb;
837 bitmap_super_t *bms = (bitmap_super_t *)(((char *)sb) + MAX_SB_SIZE);
838 int i;
839 int l = homehost ? strlen(homehost) : 0;
840
841 printf(" Name : %.32s", sb->set_name);
842 if (l > 0 && l < 32 && sb->set_name[l] == ':' &&
843 strncmp(sb->set_name, homehost, l) == 0)
844 printf(" (local to host %s)", homehost);
845 if (bms->nodes > 0 &&
846 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
847 printf("\n Cluster Name : %-64s", bms->cluster_name);
848 printf("\n UUID : ");
849 for (i = 0; i < 16; i++) {
850 if ((i & 3) == 0 && i != 0)
851 printf(":");
852 printf("%02x", sb->set_uuid[i]);
853 }
854 printf("\n Events : %llu\n\n",
855 (unsigned long long)__le64_to_cpu(sb->events));
856 }
857
858 static void brief_detail_super1(struct supertype *st, char *subarray)
859 {
860 struct mdp_superblock_1 *sb = st->sb;
861 int i;
862
863 printf(" UUID=");
864 for (i = 0; i < 16; i++) {
865 if ((i & 3) == 0 && i != 0)
866 printf(":");
867 printf("%02x", sb->set_uuid[i]);
868 }
869 }
870
871 static void export_detail_super1(struct supertype *st)
872 {
873 struct mdp_superblock_1 *sb = st->sb;
874 int i;
875 int len = 32;
876
877 for (i = 0; i < 32; i++)
878 if (sb->set_name[i] == '\n' || sb->set_name[i] == '\0') {
879 len = i;
880 break;
881 }
882 if (len)
883 printf("MD_NAME=%.*s\n", len, sb->set_name);
884 }
885
886 static int examine_badblocks_super1(struct supertype *st, int fd, char *devname)
887 {
888 struct mdp_superblock_1 *sb = st->sb;
889 unsigned long long offset;
890 int size;
891 __u64 *bbl, *bbp;
892 int i;
893
894 if (!sb->bblog_size || __le16_to_cpu(sb->bblog_size) > 100 ||
895 !sb->bblog_offset){
896 printf("No bad-blocks list configured on %s\n", devname);
897 return 0;
898 }
899 if ((sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS)) == 0) {
900 printf("Bad-blocks list is empty in %s\n", devname);
901 return 0;
902 }
903
904 size = __le16_to_cpu(sb->bblog_size)* 512;
905 if (posix_memalign((void **)&bbl, 4096, size) != 0) {
906 pr_err("could not allocate badblocks list\n");
907 return 0;
908 }
909 offset = __le64_to_cpu(sb->super_offset) +
910 (int)__le32_to_cpu(sb->bblog_offset);
911 offset <<= 9;
912 if (lseek64(fd, offset, 0) < 0) {
913 pr_err("Cannot seek to bad-blocks list\n");
914 return 1;
915 }
916 if (read(fd, bbl, size) != size) {
917 pr_err("Cannot read bad-blocks list\n");
918 return 1;
919 }
920 /* 64bits per entry. 10 bits is block-count, 54 bits is block
921 * offset. Blocks are sectors unless bblog->shift makes them bigger
922 */
923 bbp = (__u64*)bbl;
924 printf("Bad-blocks on %s:\n", devname);
925 for (i = 0; i < size/8; i++, bbp++) {
926 __u64 bb = __le64_to_cpu(*bbp);
927 int count = bb & 0x3ff;
928 unsigned long long sector = bb >> 10;
929
930 if (bb + 1 == 0)
931 break;
932
933 sector <<= sb->bblog_shift;
934 count <<= sb->bblog_shift;
935
936 printf("%20llu for %d sectors\n", sector, count);
937 }
938 return 0;
939 }
940
941 static int match_home1(struct supertype *st, char *homehost)
942 {
943 struct mdp_superblock_1 *sb = st->sb;
944 int l = homehost ? strlen(homehost) : 0;
945
946 return (l > 0 && l < 32 && sb->set_name[l] == ':' &&
947 strncmp(sb->set_name, homehost, l) == 0);
948 }
949
950 static void uuid_from_super1(struct supertype *st, int uuid[4])
951 {
952 struct mdp_superblock_1 *super = st->sb;
953 char *cuuid = (char *)uuid;
954 int i;
955 for (i = 0; i < 16; i++)
956 cuuid[i] = super->set_uuid[i];
957 }
958
959 static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
960 {
961 struct mdp_superblock_1 *sb = st->sb;
962 struct bitmap_super_s *bsb = (void *)(((char *)sb) + MAX_SB_SIZE);
963 struct misc_dev_info *misc =
964 (void *)(((char *)sb) + MAX_SB_SIZE+BM_SUPER_SIZE);
965 int working = 0;
966 unsigned int i;
967 unsigned int role;
968 unsigned int map_disks = info->array.raid_disks;
969 unsigned long long super_offset;
970 unsigned long long data_size;
971
972 memset(info, 0, sizeof(*info));
973 info->array.major_version = 1;
974 info->array.minor_version = st->minor_version;
975 info->array.patch_version = 0;
976 info->array.raid_disks = __le32_to_cpu(sb->raid_disks);
977 info->array.level = __le32_to_cpu(sb->level);
978 info->array.layout = __le32_to_cpu(sb->layout);
979 info->array.md_minor = -1;
980 info->array.ctime = __le64_to_cpu(sb->ctime);
981 info->array.utime = __le64_to_cpu(sb->utime);
982 info->array.chunk_size = __le32_to_cpu(sb->chunksize)*512;
983 info->array.state =
984 (__le64_to_cpu(sb->resync_offset) == MaxSector) ? 1 : 0;
985
986 super_offset = __le64_to_cpu(sb->super_offset);
987 info->data_offset = __le64_to_cpu(sb->data_offset);
988 info->component_size = __le64_to_cpu(sb->size);
989 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
990 info->bitmap_offset = (int32_t)__le32_to_cpu(sb->bitmap_offset);
991 if (__le32_to_cpu(bsb->nodes) > 1)
992 info->array.state |= (1 << MD_SB_CLUSTERED);
993 } else if (md_feature_any_ppl_on(sb->feature_map)) {
994 info->ppl_offset = __le16_to_cpu(sb->ppl.offset);
995 info->ppl_size = __le16_to_cpu(sb->ppl.size);
996 info->ppl_sector = super_offset + info->ppl_offset;
997 }
998
999 info->disk.major = 0;
1000 info->disk.minor = 0;
1001 info->disk.number = __le32_to_cpu(sb->dev_number);
1002 if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) ||
1003 __le32_to_cpu(sb->dev_number) >= MAX_DEVS)
1004 role = MD_DISK_ROLE_FAULTY;
1005 else
1006 role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]);
1007
1008 if (info->array.level <= 0)
1009 data_size = __le64_to_cpu(sb->data_size);
1010 else
1011 data_size = __le64_to_cpu(sb->size);
1012 if (info->data_offset < super_offset) {
1013 unsigned long long end;
1014 info->space_before = info->data_offset;
1015 end = super_offset;
1016
1017 if (sb->bblog_offset && sb->bblog_size) {
1018 unsigned long long bboffset = super_offset;
1019 bboffset += (int32_t)__le32_to_cpu(sb->bblog_offset);
1020 if (bboffset < end)
1021 end = bboffset;
1022 }
1023
1024 if (super_offset + info->bitmap_offset + info->ppl_offset < end)
1025 end = super_offset + info->bitmap_offset +
1026 info->ppl_offset;
1027
1028 if (info->data_offset + data_size < end)
1029 info->space_after = end - data_size - info->data_offset;
1030 else
1031 info->space_after = 0;
1032 } else {
1033 unsigned long long earliest;
1034 earliest = super_offset + (32+4)*2; /* match kernel */
1035 if (info->bitmap_offset > 0) {
1036 unsigned long long bmend = info->bitmap_offset;
1037 unsigned long long size = calc_bitmap_size(bsb, 4096);
1038 size /= 512;
1039 bmend += size;
1040 if (bmend > earliest)
1041 earliest = bmend;
1042 } else if (info->ppl_offset > 0) {
1043 unsigned long long pplend;
1044
1045 pplend = info->ppl_offset + info->ppl_size;
1046 if (pplend > earliest)
1047 earliest = pplend;
1048 }
1049 if (sb->bblog_offset && sb->bblog_size) {
1050 unsigned long long bbend = super_offset;
1051 bbend += (int32_t)__le32_to_cpu(sb->bblog_offset);
1052 bbend += __le16_to_cpu(sb->bblog_size);
1053 if (bbend > earliest)
1054 earliest = bbend;
1055 }
1056 if (earliest < info->data_offset)
1057 info->space_before = info->data_offset - earliest;
1058 else
1059 info->space_before = 0;
1060 info->space_after = misc->device_size - data_size -
1061 info->data_offset;
1062 }
1063 if (info->space_before == 0 && info->space_after == 0) {
1064 /* It will look like we don't support data_offset changes,
1065 * be we do - it's just that there is no room.
1066 * A change that reduced the number of devices should
1067 * still be allowed, so set the otherwise useless value of '1'
1068 */
1069 info->space_after = 1;
1070 }
1071
1072 info->disk.raid_disk = -1;
1073 switch(role) {
1074 case MD_DISK_ROLE_SPARE:
1075 /* spare: not active, not sync, not faulty */
1076 info->disk.state = 0;
1077 break;
1078 case MD_DISK_ROLE_FAULTY:
1079 info->disk.state = (1 << MD_DISK_FAULTY); /* faulty */
1080 break;
1081 case MD_DISK_ROLE_JOURNAL:
1082 info->disk.state = (1 << MD_DISK_JOURNAL);
1083 info->disk.raid_disk = role;
1084 /* journal uses all 4kB blocks*/
1085 info->space_after = (misc->device_size - info->data_offset) % 8;
1086 break;
1087 default:
1088 info->disk.state = 6; /* active and in sync */
1089 info->disk.raid_disk = role;
1090 }
1091 if (sb->devflags & WriteMostly1)
1092 info->disk.state |= (1 << MD_DISK_WRITEMOSTLY);
1093 if (sb->devflags & FailFast1)
1094 info->disk.state |= (1 << MD_DISK_FAILFAST);
1095 info->events = __le64_to_cpu(sb->events);
1096 sprintf(info->text_version, "1.%d", st->minor_version);
1097 info->safe_mode_delay = 200;
1098
1099 memcpy(info->uuid, sb->set_uuid, 16);
1100
1101 strncpy(info->name, sb->set_name, 32);
1102 info->name[32] = 0;
1103
1104 if ((__le32_to_cpu(sb->feature_map)&MD_FEATURE_REPLACEMENT)) {
1105 info->disk.state &= ~(1 << MD_DISK_SYNC);
1106 info->disk.state |= 1 << MD_DISK_REPLACEMENT;
1107 }
1108
1109 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET))
1110 info->recovery_start = __le32_to_cpu(sb->recovery_offset);
1111 else
1112 info->recovery_start = MaxSector;
1113
1114 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) {
1115 info->reshape_active = 1;
1116 if ((sb->feature_map & __le32_to_cpu(MD_FEATURE_NEW_OFFSET)) &&
1117 sb->new_offset != 0)
1118 info->reshape_active |= RESHAPE_NO_BACKUP;
1119 info->reshape_progress = __le64_to_cpu(sb->reshape_position);
1120 info->new_level = __le32_to_cpu(sb->new_level);
1121 info->delta_disks = __le32_to_cpu(sb->delta_disks);
1122 info->new_layout = __le32_to_cpu(sb->new_layout);
1123 info->new_chunk = __le32_to_cpu(sb->new_chunk)<<9;
1124 if (info->delta_disks < 0)
1125 info->array.raid_disks -= info->delta_disks;
1126 } else
1127 info->reshape_active = 0;
1128
1129 info->recovery_blocked = info->reshape_active;
1130
1131 if (map)
1132 for (i = 0; i < map_disks; i++)
1133 map[i] = 0;
1134 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
1135 role = __le16_to_cpu(sb->dev_roles[i]);
1136 if (/*role == MD_DISK_ROLE_SPARE || */role < (unsigned) info->array.raid_disks) {
1137 working++;
1138 if (map && role < map_disks)
1139 map[role] = 1;
1140 }
1141 }
1142
1143 info->array.working_disks = working;
1144
1145 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_JOURNAL)) {
1146 info->journal_device_required = 1;
1147 info->consistency_policy = CONSISTENCY_POLICY_JOURNAL;
1148 } else if (md_feature_any_ppl_on(sb->feature_map)) {
1149 info->consistency_policy = CONSISTENCY_POLICY_PPL;
1150 } else if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
1151 info->consistency_policy = CONSISTENCY_POLICY_BITMAP;
1152 } else if (info->array.level <= 0) {
1153 info->consistency_policy = CONSISTENCY_POLICY_NONE;
1154 } else {
1155 info->consistency_policy = CONSISTENCY_POLICY_RESYNC;
1156 }
1157
1158 info->journal_clean = 0;
1159 }
1160
1161 static struct mdinfo *container_content1(struct supertype *st, char *subarray)
1162 {
1163 struct mdinfo *info;
1164
1165 if (subarray)
1166 return NULL;
1167
1168 info = xmalloc(sizeof(*info));
1169 getinfo_super1(st, info, NULL);
1170 return info;
1171 }
1172
1173 static int update_super1(struct supertype *st, struct mdinfo *info,
1174 enum update_opt update, char *devname, int verbose,
1175 int uuid_set, char *homehost)
1176 {
1177 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1178 * if any change was made. For others, the return value is
1179 * ignored.
1180 */
1181 int rv = 0;
1182 struct mdp_superblock_1 *sb = st->sb;
1183 bitmap_super_t *bms = (bitmap_super_t *)(((char *)sb) + MAX_SB_SIZE);
1184
1185 if (update == UOPT_HOMEHOST && homehost) {
1186 /*
1187 * Note that 'homehost' is special as it is really
1188 * a "name" update.
1189 */
1190 char *c;
1191 update = UOPT_NAME;
1192 c = strchr(sb->set_name, ':');
1193 if (c)
1194 snprintf(info->name, sizeof(info->name), "%s", c + 1);
1195 else
1196 snprintf(info->name, sizeof(info->name), "%s",
1197 sb->set_name);
1198 }
1199
1200 switch (update) {
1201 case UOPT_NAME: {
1202 int namelen;
1203
1204 if (!info->name[0])
1205 snprintf(info->name, sizeof(info->name), "%d", info->array.md_minor);
1206 memset(sb->set_name, 0, sizeof(sb->set_name));
1207
1208 namelen = strnlen(homehost, MD_NAME_MAX) + 1 + strnlen(info->name, MD_NAME_MAX);
1209 if (homehost &&
1210 strchr(info->name, ':') == NULL &&
1211 namelen < MD_NAME_MAX) {
1212 strcpy(sb->set_name, homehost);
1213 strcat(sb->set_name, ":");
1214 strcat(sb->set_name, info->name);
1215 } else {
1216 namelen = min((int)strnlen(info->name, MD_NAME_MAX),
1217 (int)sizeof(sb->set_name) - 1);
1218 memcpy(sb->set_name, info->name, namelen);
1219 memset(&sb->set_name[namelen], '\0',
1220 sizeof(sb->set_name) - namelen);
1221 }
1222 break;
1223 }
1224 case UOPT_SPEC_FORCE_ONE:
1225 /* Not enough devices for a working array,
1226 * so bring this one up-to-date
1227 */
1228 if (sb->events != __cpu_to_le64(info->events))
1229 rv = 1;
1230 sb->events = __cpu_to_le64(info->events);
1231 break;
1232 case UOPT_SPEC_FORCE_ARRAY:
1233 /* Degraded array and 'force' requests to
1234 * maybe need to mark it 'clean'.
1235 */
1236 switch(__le32_to_cpu(sb->level)) {
1237 case 4:
1238 case 5:
1239 case 6:
1240 /* need to force clean */
1241 if (sb->resync_offset != MaxSector)
1242 rv = 1;
1243 sb->resync_offset = MaxSector;
1244 }
1245 break;
1246 case UOPT_SPEC_ASSEMBLE: {
1247 int d = info->disk.number;
1248 int want;
1249 if (info->disk.state & (1<<MD_DISK_ACTIVE))
1250 want = info->disk.raid_disk;
1251 else if (info->disk.state & (1<<MD_DISK_JOURNAL))
1252 want = MD_DISK_ROLE_JOURNAL;
1253 else
1254 want = MD_DISK_ROLE_SPARE;
1255 if (sb->dev_roles[d] != __cpu_to_le16(want)) {
1256 sb->dev_roles[d] = __cpu_to_le16(want);
1257 rv = 1;
1258 }
1259 if (info->reshape_active &&
1260 sb->feature_map &
1261 __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1262 info->delta_disks >= 0 &&
1263 info->reshape_progress <
1264 __le64_to_cpu(sb->reshape_position)) {
1265 sb->reshape_position =
1266 __cpu_to_le64(info->reshape_progress);
1267 rv = 1;
1268 }
1269 if (info->reshape_active &&
1270 sb->feature_map &
1271 __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1272 info->delta_disks < 0 &&
1273 info->reshape_progress >
1274 __le64_to_cpu(sb->reshape_position)) {
1275 sb->reshape_position =
1276 __cpu_to_le64(info->reshape_progress);
1277 rv = 1;
1278 }
1279 break;
1280 }
1281 case UOPT_SPEC_LINEAR_GROW_NEW: {
1282 int i;
1283 int fd;
1284 int max = __le32_to_cpu(sb->max_dev);
1285
1286 if (max > MAX_DEVS)
1287 return -2;
1288
1289 for (i = 0; i < max; i++)
1290 if (__le16_to_cpu(sb->dev_roles[i]) >=
1291 MD_DISK_ROLE_FAULTY)
1292 break;
1293 if (i != info->disk.number)
1294 return -2;
1295 sb->dev_number = __cpu_to_le32(i);
1296
1297 if (i == max)
1298 sb->max_dev = __cpu_to_le32(max + 1);
1299 if (i > max)
1300 return -2;
1301
1302 random_uuid(sb->device_uuid);
1303
1304 sb->dev_roles[i] = __cpu_to_le16(info->disk.raid_disk);
1305
1306 fd = open(devname, O_RDONLY);
1307 if (fd >= 0) {
1308 unsigned long long ds;
1309 get_dev_size(fd, devname, &ds);
1310 close(fd);
1311 ds >>= 9;
1312 if (__le64_to_cpu(sb->super_offset) <
1313 __le64_to_cpu(sb->data_offset)) {
1314 sb->data_size = __cpu_to_le64(
1315 ds - __le64_to_cpu(sb->data_offset));
1316 } else {
1317 ds -= 8 * 2;
1318 ds &= ~(unsigned long long)(4 * 2 - 1);
1319 sb->super_offset = __cpu_to_le64(ds);
1320 sb->data_size = __cpu_to_le64(
1321 ds - __le64_to_cpu(sb->data_offset));
1322 }
1323 }
1324 break;
1325 }
1326 case UOPT_SPEC_LINEAR_GROW_UPDATE: {
1327 int max = __le32_to_cpu(sb->max_dev);
1328 int i = info->disk.number;
1329 if (max > MAX_DEVS || i > MAX_DEVS)
1330 return -2;
1331 if (i > max)
1332 return -2;
1333 if (i == max)
1334 sb->max_dev = __cpu_to_le32(max + 1);
1335 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
1336 sb->dev_roles[info->disk.number] =
1337 __cpu_to_le16(info->disk.raid_disk);
1338 break;
1339 }
1340 case UOPT_RESYNC:
1341 /* make sure resync happens */
1342 sb->resync_offset = 0;
1343 break;
1344 case UOPT_UUID:
1345 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
1346
1347 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)
1348 memcpy(bms->uuid, sb->set_uuid, 16);
1349 break;
1350 case UOPT_NO_BITMAP:
1351 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1352 if (bms->version == BITMAP_MAJOR_CLUSTERED && !IsBitmapDirty(devname))
1353 sb->resync_offset = MaxSector;
1354 break;
1355 case UOPT_BBL: {
1356 /* only possible if there is room after the bitmap, or if
1357 * there is no bitmap
1358 */
1359 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1360 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1361 long bitmap_offset = 0;
1362 long bm_sectors = 0;
1363 long space;
1364
1365 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1366 bitmap_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1367 bm_sectors = calc_bitmap_size(bms, 4096) >> 9;
1368 } else if (md_feature_any_ppl_on(sb->feature_map)) {
1369 bitmap_offset = (long)__le16_to_cpu(sb->ppl.offset);
1370 bm_sectors = (long)__le16_to_cpu(sb->ppl.size);
1371 }
1372
1373 if (sb_offset < data_offset) {
1374 /*
1375 * 1.1 or 1.2. Put bbl after bitmap leaving
1376 * at least 32K
1377 */
1378 long bb_offset;
1379 bb_offset = sb_offset + 8;
1380 if (bm_sectors && bitmap_offset > 0)
1381 bb_offset = bitmap_offset + bm_sectors;
1382 while (bb_offset < (long)sb_offset + 8 + 32*2 &&
1383 bb_offset + 8+8 <= (long)data_offset)
1384 bb_offset += 8;
1385 if (bb_offset + 8 <= (long)data_offset) {
1386 sb->bblog_size = __cpu_to_le16(8);
1387 sb->bblog_offset = __cpu_to_le32(bb_offset);
1388 }
1389 } else {
1390 if (bm_sectors && bitmap_offset < 0)
1391 space = -bitmap_offset - bm_sectors;
1392 else
1393 space = sb_offset - data_offset -
1394 __le64_to_cpu(sb->data_size);
1395 if (space >= 8) {
1396 sb->bblog_size = __cpu_to_le16(8);
1397 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1398 }
1399 }
1400 break;
1401 }
1402 case UOPT_NO_BBL:
1403 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
1404 pr_err("Cannot remove active bbl from %s\n",devname);
1405 else {
1406 sb->bblog_size = 0;
1407 sb->bblog_shift = 0;
1408 sb->bblog_offset = 0;
1409 }
1410 break;
1411 case UOPT_FORCE_NO_BBL:
1412 sb->feature_map &= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1413 sb->bblog_size = 0;
1414 sb->bblog_shift = 0;
1415 sb->bblog_offset = 0;
1416 break;
1417 case UOPT_PPL: {
1418 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1419 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1420 unsigned long long data_size = __le64_to_cpu(sb->data_size);
1421 long bb_offset = __le32_to_cpu(sb->bblog_offset);
1422 int space;
1423 int offset;
1424
1425 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1426 pr_err("Cannot add PPL to array with bitmap\n");
1427 return -2;
1428 }
1429
1430 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_JOURNAL)) {
1431 pr_err("Cannot add PPL to array with journal\n");
1432 return -2;
1433 }
1434
1435 if (sb_offset < data_offset) {
1436 if (bb_offset)
1437 space = bb_offset - 8;
1438 else
1439 space = data_offset - sb_offset - 8;
1440 offset = 8;
1441 } else {
1442 offset = -(sb_offset - data_offset - data_size);
1443 if (offset < INT16_MIN)
1444 offset = INT16_MIN;
1445 space = -(offset - bb_offset);
1446 }
1447
1448 if (space < (PPL_HEADER_SIZE >> 9) + 8) {
1449 pr_err("Not enough space to add ppl\n");
1450 return -2;
1451 }
1452
1453 if (space >= (MULTIPLE_PPL_AREA_SIZE_SUPER1 >> 9)) {
1454 space = (MULTIPLE_PPL_AREA_SIZE_SUPER1 >> 9);
1455 } else {
1456 int optimal_space = choose_ppl_space(
1457 __le32_to_cpu(sb->chunksize));
1458 if (space > optimal_space)
1459 space = optimal_space;
1460 if (space > UINT16_MAX)
1461 space = UINT16_MAX;
1462 }
1463
1464 sb->ppl.offset = __cpu_to_le16(offset);
1465 sb->ppl.size = __cpu_to_le16(space);
1466 sb->feature_map |= __cpu_to_le32(MD_FEATURE_PPL);
1467 break;
1468 }
1469 case UOPT_NO_PPL:
1470 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_PPL |
1471 MD_FEATURE_MUTLIPLE_PPLS);
1472 break;
1473 case UOPT_DEVICESIZE:
1474 if (__le64_to_cpu(sb->super_offset) >=
1475 __le64_to_cpu(sb->data_offset))
1476 break;
1477 /*
1478 * set data_size to device size less data_offset
1479 */
1480 struct misc_dev_info *misc = (struct misc_dev_info*)
1481 (st->sb + MAX_SB_SIZE + BM_SUPER_SIZE);
1482 sb->data_size = __cpu_to_le64(
1483 misc->device_size - __le64_to_cpu(sb->data_offset));
1484 break;
1485 case UOPT_SPEC_REVERT_RESHAPE_NOBACKUP:
1486 case UOPT_REVERT_RESHAPE:
1487 rv = -2;
1488 if (!(sb->feature_map &
1489 __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)))
1490 pr_err("No active reshape to revert on %s\n",
1491 devname);
1492 else {
1493 __u32 temp;
1494 unsigned long long reshape_sectors;
1495 long reshape_chunk;
1496 rv = 0;
1497 /* If the reshape hasn't started, just stop it.
1498 * It is conceivable that a stripe was modified but
1499 * the metadata not updated. In that case the backup
1500 * should have been used to get passed the critical stage.
1501 * If that couldn't happen, the "-nobackup" version
1502 * will be used.
1503 */
1504 if (update == UOPT_SPEC_REVERT_RESHAPE_NOBACKUP &&
1505 sb->reshape_position == 0 &&
1506 (__le32_to_cpu(sb->delta_disks) > 0 ||
1507 (__le32_to_cpu(sb->delta_disks) == 0 &&
1508 !(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS))))) {
1509 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1510 sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1511 __le32_to_cpu(sb->delta_disks));
1512 sb->delta_disks = 0;
1513 goto done;
1514 }
1515 /* reshape_position is a little messy.
1516 * Its value must be a multiple of the larger
1517 * chunk size, and of the "after" data disks.
1518 * So when reverting we need to change it to
1519 * be a multiple of the new "after" data disks,
1520 * which is the old "before".
1521 * If it isn't already a multiple of 'before',
1522 * the only thing we could do would be
1523 * copy some block around on the disks, which
1524 * is easy to get wrong.
1525 * So we reject a revert-reshape unless the
1526 * alignment is good.
1527 */
1528 if (is_level456(__le32_to_cpu(sb->level))) {
1529 reshape_sectors =
1530 __le64_to_cpu(sb->reshape_position);
1531 reshape_chunk = __le32_to_cpu(sb->new_chunk);
1532 reshape_chunk *= __le32_to_cpu(sb->raid_disks) -
1533 __le32_to_cpu(sb->delta_disks) -
1534 (__le32_to_cpu(sb->level)==6 ? 2 : 1);
1535 if (reshape_sectors % reshape_chunk) {
1536 pr_err("Reshape position is not suitably aligned.\n");
1537 pr_err("Try normal assembly and stop again\n");
1538 return -2;
1539 }
1540 }
1541 sb->raid_disks =
1542 __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1543 __le32_to_cpu(sb->delta_disks));
1544 if (sb->delta_disks == 0)
1545 sb->feature_map ^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1546 else
1547 sb->delta_disks = __cpu_to_le32(-__le32_to_cpu(sb->delta_disks));
1548
1549 temp = sb->new_layout;
1550 sb->new_layout = sb->layout;
1551 sb->layout = temp;
1552
1553 temp = sb->new_chunk;
1554 sb->new_chunk = sb->chunksize;
1555 sb->chunksize = temp;
1556
1557 if (sb->feature_map &
1558 __cpu_to_le32(MD_FEATURE_NEW_OFFSET)) {
1559 long offset_delta =
1560 (int32_t)__le32_to_cpu(sb->new_offset);
1561 sb->data_offset = __cpu_to_le64(__le64_to_cpu(sb->data_offset) + offset_delta);
1562 sb->new_offset = __cpu_to_le32(-offset_delta);
1563 sb->data_size = __cpu_to_le64(__le64_to_cpu(sb->data_size) - offset_delta);
1564 }
1565 done:;
1566 }
1567 break;
1568 case UOPT_SPEC__RESHAPE_PROGRESS:
1569 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1570 break;
1571 case UOPT_SPEC_WRITEMOSTLY:
1572 sb->devflags |= WriteMostly1;
1573 break;
1574 case UOPT_SPEC_READWRITE:
1575 sb->devflags &= ~WriteMostly1;
1576 break;
1577 case UOPT_SPEC_FAILFAST:
1578 sb->devflags |= FailFast1;
1579 break;
1580 case UOPT_SPEC_NOFAILFAST:
1581 sb->devflags &= ~FailFast1;
1582 break;
1583 case UOPT_LAYOUT_ORIGINAL:
1584 case UOPT_LAYOUT_ALTERNATE:
1585 case UOPT_LAYOUT_UNSPECIFIED:
1586 if (__le32_to_cpu(sb->level) != 0) {
1587 pr_err("%s: %s only supported for RAID0\n",
1588 devname ?: "", map_num(update_options, update));
1589 rv = -1;
1590 } else if (update == UOPT_LAYOUT_UNSPECIFIED) {
1591 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_RAID0_LAYOUT);
1592 sb->layout = 0;
1593 } else {
1594 sb->feature_map |= __cpu_to_le32(MD_FEATURE_RAID0_LAYOUT);
1595 sb->layout = __cpu_to_le32(update == UOPT_LAYOUT_ORIGINAL ? 1 : 2);
1596 }
1597 break;
1598 default:
1599 rv = -1;
1600 }
1601
1602 sb->sb_csum = calc_sb_1_csum(sb);
1603
1604 return rv;
1605 }
1606
1607 static int init_super1(struct supertype *st, mdu_array_info_t *info,
1608 struct shape *s, char *name, char *homehost,
1609 int *uuid, unsigned long long data_offset)
1610 {
1611 struct mdp_superblock_1 *sb;
1612 int spares;
1613 char defname[10];
1614 int sbsize;
1615
1616 if (posix_memalign((void **)&sb, 4096, SUPER1_SIZE) != 0) {
1617 pr_err("could not allocate superblock\n");
1618 return 0;
1619 }
1620 memset(sb, 0, SUPER1_SIZE);
1621
1622 st->sb = sb;
1623 if (info == NULL) {
1624 /* zeroing superblock */
1625 return 0;
1626 }
1627
1628 spares = info->working_disks - info->active_disks;
1629 if (info->raid_disks + spares > MAX_DEVS) {
1630 pr_err("too many devices requested: %d+%d > %d\n",
1631 info->raid_disks , spares, MAX_DEVS);
1632 return 0;
1633 }
1634
1635 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
1636 sb->major_version = __cpu_to_le32(1);
1637 sb->feature_map = 0;
1638 sb->pad0 = 0;
1639
1640 if (uuid)
1641 copy_uuid(sb->set_uuid, uuid, super1.swapuuid);
1642 else
1643 random_uuid(sb->set_uuid);;
1644
1645 if (name == NULL || *name == 0) {
1646 sprintf(defname, "%d", info->md_minor);
1647 name = defname;
1648 }
1649 if (homehost &&
1650 strchr(name, ':') == NULL &&
1651 strlen(homehost) + 1 + strlen(name) < 32) {
1652 strcpy(sb->set_name, homehost);
1653 strcat(sb->set_name, ":");
1654 strcat(sb->set_name, name);
1655 } else {
1656 int namelen;
1657
1658 namelen = min((int)strlen(name),
1659 (int)sizeof(sb->set_name) - 1);
1660 memcpy(sb->set_name, name, namelen);
1661 memset(&sb->set_name[namelen], '\0',
1662 sizeof(sb->set_name) - namelen);
1663 }
1664
1665 sb->ctime = __cpu_to_le64((unsigned long long)time(0));
1666 sb->level = __cpu_to_le32(info->level);
1667 sb->layout = __cpu_to_le32(info->layout);
1668 sb->size = __cpu_to_le64(s->size*2ULL);
1669 sb->chunksize = __cpu_to_le32(info->chunk_size>>9);
1670 sb->raid_disks = __cpu_to_le32(info->raid_disks);
1671
1672 sb->data_offset = __cpu_to_le64(data_offset);
1673 sb->data_size = __cpu_to_le64(0);
1674 sb->super_offset = __cpu_to_le64(0);
1675 sb->recovery_offset = __cpu_to_le64(0);
1676
1677 sb->utime = sb->ctime;
1678 sb->events = __cpu_to_le64(1);
1679 if (info->state & (1<<MD_SB_CLEAN))
1680 sb->resync_offset = MaxSector;
1681 else
1682 sb->resync_offset = 0;
1683 sbsize = sizeof(struct mdp_superblock_1) +
1684 2 * (info->raid_disks + spares);
1685 sbsize = ROUND_UP(sbsize, 512);
1686 sb->max_dev =
1687 __cpu_to_le32((sbsize - sizeof(struct mdp_superblock_1)) / 2);
1688
1689 memset(sb->dev_roles, 0xff,
1690 MAX_SB_SIZE - sizeof(struct mdp_superblock_1));
1691
1692 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
1693 sb->feature_map |= __cpu_to_le32(MD_FEATURE_PPL);
1694
1695 return 1;
1696 }
1697
1698 struct devinfo {
1699 int fd;
1700 char *devname;
1701 long long data_offset;
1702 unsigned long long dev_size;
1703 mdu_disk_info_t disk;
1704 struct devinfo *next;
1705 };
1706
1707 /* Add a device to the superblock being created */
1708 static int add_to_super1(struct supertype *st, mdu_disk_info_t *dk,
1709 int fd, char *devname, unsigned long long data_offset)
1710 {
1711 struct mdp_superblock_1 *sb = st->sb;
1712 __u16 *rp = sb->dev_roles + dk->number;
1713 struct devinfo *di, **dip;
1714 int dk_state;
1715
1716 dk_state = dk->state & ~(1<<MD_DISK_FAILFAST);
1717 if ((dk_state & (1<<MD_DISK_ACTIVE)) &&
1718 (dk_state & (1<<MD_DISK_SYNC)))/* active, sync */
1719 *rp = __cpu_to_le16(dk->raid_disk);
1720 else if (dk_state & (1<<MD_DISK_JOURNAL))
1721 *rp = MD_DISK_ROLE_JOURNAL;
1722 else if ((dk_state & ~(1<<MD_DISK_ACTIVE)) == 0)
1723 /* active or idle -> spare */
1724 *rp = MD_DISK_ROLE_SPARE;
1725 else
1726 *rp = MD_DISK_ROLE_FAULTY;
1727
1728 if (dk->number >= (int)__le32_to_cpu(sb->max_dev) &&
1729 __le32_to_cpu(sb->max_dev) < MAX_DEVS)
1730 sb->max_dev = __cpu_to_le32(dk->number + 1);
1731
1732 sb->dev_number = __cpu_to_le32(dk->number);
1733 sb->devflags = 0; /* don't copy another disks flags */
1734 sb->sb_csum = calc_sb_1_csum(sb);
1735
1736 dip = (struct devinfo **)&st->info;
1737 while (*dip)
1738 dip = &(*dip)->next;
1739 di = xmalloc(sizeof(struct devinfo));
1740 di->fd = fd;
1741 di->devname = devname;
1742 di->disk = *dk;
1743 di->data_offset = data_offset;
1744
1745 if (is_fd_valid(fd))
1746 get_dev_size(fd, NULL, &di->dev_size);
1747
1748 di->next = NULL;
1749 *dip = di;
1750
1751 return 0;
1752 }
1753
1754 static int locate_bitmap1(struct supertype *st, int fd, int node_num);
1755
1756 static int store_super1(struct supertype *st, int fd)
1757 {
1758 struct mdp_superblock_1 *sb = st->sb;
1759 unsigned long long sb_offset;
1760 struct align_fd afd;
1761 int sbsize;
1762 unsigned long long dsize;
1763
1764 if (!get_dev_size(fd, NULL, &dsize))
1765 return 1;
1766
1767 dsize >>= 9;
1768
1769 if (dsize < 24)
1770 return 2;
1771
1772 init_afd(&afd, fd);
1773
1774 /*
1775 * Calculate the position of the superblock.
1776 * It is always aligned to a 4K boundary and
1777 * depending on minor_version, it can be:
1778 * 0: At least 8K, but less than 12K, from end of device
1779 * 1: At start of device
1780 * 2: 4K from start of device.
1781 */
1782 switch(st->minor_version) {
1783 case 0:
1784 sb_offset = dsize;
1785 sb_offset -= 8*2;
1786 sb_offset &= ~(4*2-1);
1787 break;
1788 case 1:
1789 sb_offset = 0;
1790 break;
1791 case 2:
1792 sb_offset = 4*2;
1793 break;
1794 default:
1795 return -EINVAL;
1796 }
1797
1798 if (sb_offset != __le64_to_cpu(sb->super_offset) &&
1799 0 != __le64_to_cpu(sb->super_offset)
1800 ) {
1801 pr_err("internal error - sb_offset is wrong\n");
1802 abort();
1803 }
1804
1805 if (lseek64(fd, sb_offset << 9, 0)< 0LL)
1806 return 3;
1807
1808 sbsize = ROUND_UP(sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev), 512);
1809
1810 if (awrite(&afd, sb, sbsize) != sbsize)
1811 return 4;
1812
1813 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1814 struct bitmap_super_s *bm;
1815 bm = (struct bitmap_super_s *)(((char *)sb) + MAX_SB_SIZE);
1816 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC) {
1817 locate_bitmap1(st, fd, 0);
1818 if (awrite(&afd, bm, sizeof(*bm)) != sizeof(*bm))
1819 return 5;
1820 }
1821 }
1822 fsync(fd);
1823
1824 return 0;
1825 }
1826
1827 static int load_super1(struct supertype *st, int fd, char *devname);
1828
1829 static unsigned long choose_bm_space(unsigned long devsize)
1830 {
1831 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1832 * if bigger than 200Gig, save 128k
1833 * NOTE: result must be multiple of 4K else bad things happen
1834 * on 4K-sector devices.
1835 */
1836 if (devsize < 64*2)
1837 return 0;
1838 if (devsize - 64*2 >= 200*1024*1024*2)
1839 return 128*2;
1840 if (devsize - 4*2 > 8*1024*1024*2)
1841 return 64*2;
1842 return 4*2;
1843 }
1844
1845 static void free_super1(struct supertype *st);
1846
1847 __u32 crc32c_le(__u32 crc, unsigned char const *p, size_t len);
1848
1849 static int write_init_ppl1(struct supertype *st, struct mdinfo *info, int fd)
1850 {
1851 struct mdp_superblock_1 *sb = st->sb;
1852 void *buf;
1853 struct ppl_header *ppl_hdr;
1854 int ret;
1855
1856 /* first clear entire ppl space */
1857 ret = zero_disk_range(fd, info->ppl_sector, info->ppl_size);
1858 if (ret)
1859 return ret;
1860
1861 ret = posix_memalign(&buf, 4096, PPL_HEADER_SIZE);
1862 if (ret) {
1863 pr_err("Failed to allocate PPL header buffer\n");
1864 return ret;
1865 }
1866
1867 memset(buf, 0, PPL_HEADER_SIZE);
1868 ppl_hdr = buf;
1869 memset(ppl_hdr->reserved, 0xff, PPL_HDR_RESERVED);
1870 ppl_hdr->signature = __cpu_to_le32(~crc32c_le(~0, sb->set_uuid,
1871 sizeof(sb->set_uuid)));
1872 ppl_hdr->checksum = __cpu_to_le32(~crc32c_le(~0, buf, PPL_HEADER_SIZE));
1873
1874 if (lseek64(fd, info->ppl_sector * 512, SEEK_SET) < 0) {
1875 ret = errno;
1876 perror("Failed to seek to PPL header location");
1877 }
1878
1879 if (!ret && write(fd, buf, PPL_HEADER_SIZE) != PPL_HEADER_SIZE) {
1880 ret = errno;
1881 perror("Write PPL header failed");
1882 }
1883
1884 if (!ret)
1885 fsync(fd);
1886
1887 free(buf);
1888 return ret;
1889 }
1890
1891 #define META_BLOCK_SIZE 4096
1892
1893 static int write_empty_r5l_meta_block(struct supertype *st, int fd)
1894 {
1895 struct r5l_meta_block *mb;
1896 struct mdp_superblock_1 *sb = st->sb;
1897 struct align_fd afd;
1898 __u32 crc;
1899
1900 init_afd(&afd, fd);
1901
1902 if (posix_memalign((void **)&mb, 4096, META_BLOCK_SIZE) != 0) {
1903 pr_err("Could not allocate memory for the meta block.\n");
1904 return 1;
1905 }
1906
1907 memset(mb, 0, META_BLOCK_SIZE);
1908
1909 mb->magic = __cpu_to_le32(R5LOG_MAGIC);
1910 mb->version = R5LOG_VERSION;
1911 mb->meta_size = __cpu_to_le32(sizeof(struct r5l_meta_block));
1912 mb->seq = __cpu_to_le64(random32());
1913 mb->position = __cpu_to_le64(0);
1914
1915 crc = crc32c_le(0xffffffff, sb->set_uuid, sizeof(sb->set_uuid));
1916 crc = crc32c_le(crc, (void *)mb, META_BLOCK_SIZE);
1917 mb->checksum = crc;
1918
1919 if (lseek64(fd, __le64_to_cpu(sb->data_offset) * 512, 0) < 0LL) {
1920 pr_err("cannot seek to offset of the meta block\n");
1921 goto fail_to_write;
1922 }
1923
1924 if (awrite(&afd, mb, META_BLOCK_SIZE) != META_BLOCK_SIZE) {
1925 pr_err("failed to store write the meta block \n");
1926 goto fail_to_write;
1927 }
1928 fsync(fd);
1929
1930 free(mb);
1931 return 0;
1932
1933 fail_to_write:
1934 free(mb);
1935 return 1;
1936 }
1937
1938 static bool has_raid0_layout(struct mdp_superblock_1 *sb)
1939 {
1940 if (sb->level == 0 && sb->layout != 0)
1941 return true;
1942 else
1943 return false;
1944 }
1945
1946 static int write_init_super1(struct supertype *st)
1947 {
1948 struct mdp_superblock_1 *sb = st->sb;
1949 struct supertype *refst;
1950 int rv = 0;
1951 unsigned long long bm_space;
1952 struct devinfo *di;
1953 unsigned long long dsize, array_size;
1954 unsigned long long sb_offset;
1955 unsigned long long data_offset;
1956 long bm_offset;
1957 bool raid0_need_layout = false;
1958
1959 /* Since linux kernel v5.4, raid0 always has a layout */
1960 if (has_raid0_layout(sb) && get_linux_version() >= 5004000)
1961 raid0_need_layout = true;
1962
1963 for (di = st->info; di; di = di->next) {
1964 if (di->disk.state & (1 << MD_DISK_JOURNAL))
1965 sb->feature_map |= __cpu_to_le32(MD_FEATURE_JOURNAL);
1966 if (has_raid0_layout(sb) && !raid0_need_layout) {
1967
1968 struct devinfo *di2 = st->info;
1969 unsigned long long s1, s2;
1970 s1 = di->dev_size;
1971 if (di->data_offset != INVALID_SECTORS)
1972 s1 -= di->data_offset;
1973 s1 /= __le32_to_cpu(sb->chunksize);
1974 s2 = di2->dev_size;
1975 if (di2->data_offset != INVALID_SECTORS)
1976 s2 -= di2->data_offset;
1977 s2 /= __le32_to_cpu(sb->chunksize);
1978 if (s1 != s2)
1979 raid0_need_layout = true;
1980 }
1981 }
1982
1983 for (di = st->info; di; di = di->next) {
1984 if (di->disk.state & (1 << MD_DISK_FAULTY))
1985 continue;
1986 if (di->fd < 0)
1987 continue;
1988
1989 while (Kill(di->devname, NULL, 0, -1, 1) == 0)
1990 ;
1991
1992 sb->dev_number = __cpu_to_le32(di->disk.number);
1993 if (di->disk.state & (1<<MD_DISK_WRITEMOSTLY))
1994 sb->devflags |= WriteMostly1;
1995 else
1996 sb->devflags &= ~WriteMostly1;
1997 if (di->disk.state & (1<<MD_DISK_FAILFAST))
1998 sb->devflags |= FailFast1;
1999 else
2000 sb->devflags &= ~FailFast1;
2001
2002 random_uuid(sb->device_uuid);
2003
2004 if (!(di->disk.state & (1<<MD_DISK_JOURNAL)))
2005 sb->events = 0;
2006
2007 refst = dup_super(st);
2008 if (load_super1(refst, di->fd, NULL)==0) {
2009 struct mdp_superblock_1 *refsb = refst->sb;
2010
2011 memcpy(sb->device_uuid, refsb->device_uuid, 16);
2012 if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) {
2013 /* same array, so preserve events and
2014 * dev_number */
2015 sb->events = refsb->events;
2016 }
2017 free_super1(refst);
2018 }
2019 free(refst);
2020
2021 if (!get_dev_size(di->fd, NULL, &dsize)) {
2022 rv = 1;
2023 goto error_out;
2024 }
2025 dsize >>= 9;
2026
2027 if (dsize < 24) {
2028 close(di->fd);
2029 rv = 2;
2030 goto error_out;
2031 }
2032
2033 /*
2034 * Calculate the position of the superblock.
2035 * It is always aligned to a 4K boundary and
2036 * depending on minor_version, it can be:
2037 * 0: At least 8K, but less than 12K, from end of device
2038 * 1: At start of device
2039 * 2: 4K from start of device.
2040 * data_offset has already been set.
2041 */
2042 array_size = __le64_to_cpu(sb->size);
2043
2044 /* work out how much space we left for a bitmap */
2045 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
2046 bitmap_super_t *bms = (bitmap_super_t *)
2047 (((char *)sb) + MAX_SB_SIZE);
2048 bm_space = calc_bitmap_size(bms, 4096) >> 9;
2049 bm_offset = (long)__le32_to_cpu(sb->bitmap_offset);
2050 } else if (md_feature_any_ppl_on(sb->feature_map)) {
2051 bm_space = MULTIPLE_PPL_AREA_SIZE_SUPER1 >> 9;
2052 if (st->minor_version == 0)
2053 bm_offset = -bm_space - 8;
2054 else
2055 bm_offset = 8;
2056 sb->ppl.offset = __cpu_to_le16(bm_offset);
2057 sb->ppl.size = __cpu_to_le16(bm_space);
2058 } else {
2059 bm_space = choose_bm_space(array_size);
2060 bm_offset = 8;
2061 }
2062
2063 data_offset = di->data_offset;
2064 if (data_offset == INVALID_SECTORS)
2065 data_offset = st->data_offset;
2066 switch(st->minor_version) {
2067 case 0:
2068 /* Add 8 sectors for bad block log */
2069 bm_space += 8;
2070 if (data_offset == INVALID_SECTORS)
2071 data_offset = 0;
2072 sb_offset = dsize;
2073 sb_offset -= 8*2;
2074 sb_offset &= ~(4*2-1);
2075 sb->data_offset = __cpu_to_le64(data_offset);
2076 sb->super_offset = __cpu_to_le64(sb_offset);
2077 if (sb_offset < array_size + bm_space)
2078 bm_space = sb_offset - array_size;
2079 sb->data_size = __cpu_to_le64(sb_offset - bm_space);
2080 if (bm_space >= 8) {
2081 sb->bblog_size = __cpu_to_le16(8);
2082 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
2083 }
2084 break;
2085 case 1:
2086 case 2:
2087 sb_offset = st->minor_version == 2 ? 8 : 0;
2088 sb->super_offset = __cpu_to_le64(sb_offset);
2089 if (data_offset == INVALID_SECTORS)
2090 data_offset = sb_offset + 16;
2091
2092 sb->data_offset = __cpu_to_le64(data_offset);
2093 sb->data_size = __cpu_to_le64(dsize - data_offset);
2094 if (data_offset >= sb_offset+bm_offset+bm_space+8) {
2095 sb->bblog_size = __cpu_to_le16(8);
2096 sb->bblog_offset = __cpu_to_le32(bm_offset +
2097 bm_space);
2098 } else if (data_offset >= sb_offset + 16) {
2099 sb->bblog_size = __cpu_to_le16(8);
2100 /* '8' sectors for the bblog, and 'sb_offset'
2101 * because we want offset from superblock, not
2102 * start of device.
2103 */
2104 sb->bblog_offset = __cpu_to_le32(data_offset -
2105 8 - sb_offset);
2106 }
2107 break;
2108 default:
2109 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
2110 st->minor_version, di->devname);
2111 rv = -EINVAL;
2112 goto out;
2113 }
2114 /*
2115 * Disable badblock log on clusters, or when
2116 * explicitly requested
2117 */
2118 if (st->nodes > 0 || conf_get_create_info()->bblist == 0) {
2119 sb->bblog_size = 0;
2120 sb->bblog_offset = 0;
2121 }
2122
2123 /* RAID0 needs a layout if devices aren't all the same size */
2124 if (raid0_need_layout)
2125 sb->feature_map |= __cpu_to_le32(MD_FEATURE_RAID0_LAYOUT);
2126
2127 sb->sb_csum = calc_sb_1_csum(sb);
2128 rv = store_super1(st, di->fd);
2129
2130 if (rv == 0 && (di->disk.state & (1 << MD_DISK_JOURNAL))) {
2131 rv = write_empty_r5l_meta_block(st, di->fd);
2132 if (rv)
2133 goto error_out;
2134 }
2135
2136 if (rv == 0 &&
2137 (__le32_to_cpu(sb->feature_map) &
2138 MD_FEATURE_BITMAP_OFFSET)) {
2139 rv = st->ss->write_bitmap(st, di->fd, NodeNumUpdate);
2140 } else if (rv == 0 &&
2141 md_feature_any_ppl_on(sb->feature_map)) {
2142 struct mdinfo info;
2143
2144 st->ss->getinfo_super(st, &info, NULL);
2145 rv = st->ss->write_init_ppl(st, &info, di->fd);
2146 }
2147
2148 close(di->fd);
2149 di->fd = -1;
2150 if (rv)
2151 goto error_out;
2152 }
2153 error_out:
2154 if (rv)
2155 pr_err("Failed to write metadata to %s\n", di->devname);
2156 out:
2157 return rv;
2158 }
2159
2160 static int compare_super1(struct supertype *st, struct supertype *tst,
2161 int verbose)
2162 {
2163 /*
2164 * return:
2165 * 0 same, or first was empty, and second was copied
2166 * 1 second had wrong number
2167 * 2 wrong uuid
2168 * 3 wrong other info
2169 */
2170 struct mdp_superblock_1 *first = st->sb;
2171 struct mdp_superblock_1 *second = tst->sb;
2172
2173 if (second->magic != __cpu_to_le32(MD_SB_MAGIC))
2174 return 1;
2175 if (second->major_version != __cpu_to_le32(1))
2176 return 1;
2177
2178 if (!first) {
2179 if (posix_memalign((void **)&first, 4096, SUPER1_SIZE) != 0) {
2180 pr_err("could not allocate superblock\n");
2181 return 1;
2182 }
2183 memcpy(first, second, SUPER1_SIZE);
2184 st->sb = first;
2185 return 0;
2186 }
2187 if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0)
2188 return 2;
2189
2190 if (first->ctime != second->ctime ||
2191 first->level != second->level ||
2192 first->layout != second->layout ||
2193 first->size != second->size ||
2194 first->chunksize != second->chunksize ||
2195 first->raid_disks != second->raid_disks)
2196 return 3;
2197 return 0;
2198 }
2199
2200 static int load_super1(struct supertype *st, int fd, char *devname)
2201 {
2202 unsigned long long dsize;
2203 unsigned long long sb_offset;
2204 struct mdp_superblock_1 *super;
2205 int uuid[4];
2206 struct bitmap_super_s *bsb;
2207 struct misc_dev_info *misc;
2208 struct align_fd afd;
2209
2210 free_super1(st);
2211
2212 init_afd(&afd, fd);
2213
2214 if (st->ss == NULL || st->minor_version == -1) {
2215 int bestvers = -1;
2216 struct supertype tst;
2217 __u64 bestctime = 0;
2218 /* guess... choose latest ctime */
2219 memset(&tst, 0, sizeof(tst));
2220 tst.ss = &super1;
2221 for (tst.minor_version = 0; tst.minor_version <= 2;
2222 tst.minor_version++) {
2223 tst.ignore_hw_compat = st->ignore_hw_compat;
2224 switch(load_super1(&tst, fd, devname)) {
2225 case 0: super = tst.sb;
2226 if (bestvers == -1 ||
2227 bestctime < __le64_to_cpu(super->ctime)) {
2228 bestvers = tst.minor_version;
2229 bestctime = __le64_to_cpu(super->ctime);
2230 }
2231 free(super);
2232 tst.sb = NULL;
2233 break;
2234 case 1: return 1; /*bad device */
2235 case 2: break; /* bad, try next */
2236 }
2237 }
2238 if (bestvers != -1) {
2239 int rv;
2240 tst.minor_version = bestvers;
2241 tst.ss = &super1;
2242 tst.max_devs = MAX_DEVS;
2243 rv = load_super1(&tst, fd, devname);
2244 if (rv == 0)
2245 *st = tst;
2246 return rv;
2247 }
2248 return 2;
2249 }
2250 if (!get_dev_size(fd, devname, &dsize))
2251 return 1;
2252 dsize >>= 9;
2253
2254 if (dsize < 24) {
2255 if (devname)
2256 pr_err("%s is too small for md: size is %llu sectors.\n",
2257 devname, dsize);
2258 return 1;
2259 }
2260
2261 /*
2262 * Calculate the position of the superblock.
2263 * It is always aligned to a 4K boundary and
2264 * depending on minor_version, it can be:
2265 * 0: At least 8K, but less than 12K, from end of device
2266 * 1: At start of device
2267 * 2: 4K from start of device.
2268 */
2269 switch(st->minor_version) {
2270 case 0:
2271 sb_offset = dsize;
2272 sb_offset -= 8*2;
2273 sb_offset &= ~(4*2-1);
2274 break;
2275 case 1:
2276 sb_offset = 0;
2277 break;
2278 case 2:
2279 sb_offset = 4*2;
2280 break;
2281 default:
2282 return -EINVAL;
2283 }
2284
2285 if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
2286 if (devname)
2287 pr_err("Cannot seek to superblock on %s: %s\n",
2288 devname, strerror(errno));
2289 return 1;
2290 }
2291
2292 if (posix_memalign((void **)&super, 4096, SUPER1_SIZE) != 0) {
2293 pr_err("could not allocate superblock\n");
2294 return 1;
2295 }
2296
2297 memset(super, 0, SUPER1_SIZE);
2298
2299 if (aread(&afd, super, MAX_SB_SIZE) != MAX_SB_SIZE) {
2300 if (devname)
2301 pr_err("Cannot read superblock on %s\n",
2302 devname);
2303 free(super);
2304 return 1;
2305 }
2306
2307 if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) {
2308 if (devname)
2309 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2310 devname, MD_SB_MAGIC,
2311 __le32_to_cpu(super->magic));
2312 free(super);
2313 return 2;
2314 }
2315
2316 if (__le32_to_cpu(super->major_version) != 1) {
2317 if (devname)
2318 pr_err("Cannot interpret superblock on %s - version is %d\n",
2319 devname, __le32_to_cpu(super->major_version));
2320 free(super);
2321 return 2;
2322 }
2323 if (__le64_to_cpu(super->super_offset) != sb_offset) {
2324 if (devname)
2325 pr_err("No superblock found on %s (super_offset is wrong)\n",
2326 devname);
2327 free(super);
2328 return 2;
2329 }
2330
2331 bsb = (struct bitmap_super_s *)(((char *)super) + MAX_SB_SIZE);
2332
2333 misc = (struct misc_dev_info*)
2334 (((char *)super) + MAX_SB_SIZE+BM_SUPER_SIZE);
2335 misc->device_size = dsize;
2336 if (st->data_offset == INVALID_SECTORS)
2337 st->data_offset = __le64_to_cpu(super->data_offset);
2338
2339 if (st->minor_version >= 1 &&
2340 st->ignore_hw_compat == 0 &&
2341 ((role_from_sb(super) != MD_DISK_ROLE_JOURNAL &&
2342 dsize < (__le64_to_cpu(super->data_offset) +
2343 __le64_to_cpu(super->size))) ||
2344 dsize < (__le64_to_cpu(super->data_offset) +
2345 __le64_to_cpu(super->data_size)))) {
2346 if (devname)
2347 pr_err("Device %s is not large enough for data described in superblock\n",
2348 devname);
2349 free(super);
2350 return 2;
2351 }
2352 st->sb = super;
2353
2354 /* Now check on the bitmap superblock */
2355 if ((__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) == 0)
2356 return 0;
2357 /* Read the bitmap superblock and make sure it looks
2358 * valid. If it doesn't clear the bit. An --assemble --force
2359 * should get that written out.
2360 */
2361 locate_bitmap1(st, fd, 0);
2362 if (aread(&afd, bsb, 512) != 512)
2363 goto no_bitmap;
2364
2365 uuid_from_super1(st, uuid);
2366 if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
2367 memcmp(bsb->uuid, uuid, 16) != 0)
2368 goto no_bitmap;
2369 return 0;
2370
2371 no_bitmap:
2372 super->feature_map = __cpu_to_le32(__le32_to_cpu(super->feature_map) &
2373 ~MD_FEATURE_BITMAP_OFFSET);
2374 return 0;
2375 }
2376
2377 static struct supertype *match_metadata_desc1(char *arg)
2378 {
2379 struct supertype *st = xcalloc(1, sizeof(*st));
2380
2381 st->container_devnm[0] = 0;
2382 st->ss = &super1;
2383 st->max_devs = MAX_DEVS;
2384 st->sb = NULL;
2385 st->data_offset = INVALID_SECTORS;
2386 /* leading zeros can be safely ignored. --detail generates them. */
2387 while (*arg == '0')
2388 arg++;
2389 if (strcmp(arg, "1.0") == 0 || strcmp(arg, "1.00") == 0) {
2390 st->minor_version = 0;
2391 return st;
2392 }
2393 if (strcmp(arg, "1.1") == 0 || strcmp(arg, "1.01") == 0
2394 ) {
2395 st->minor_version = 1;
2396 return st;
2397 }
2398 if (strcmp(arg, "1.2") == 0 ||
2399 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2400 strcmp(arg, "default") == 0 ||
2401 #endif /* DEFAULT_OLD_METADATA */
2402 strcmp(arg, "1.02") == 0) {
2403 st->minor_version = 2;
2404 return st;
2405 }
2406 if (strcmp(arg, "1") == 0 || strcmp(arg, "default") == 0) {
2407 st->minor_version = -1;
2408 return st;
2409 }
2410
2411 free(st);
2412 return NULL;
2413 }
2414
2415 /* find available size on device with this devsize, using
2416 * superblock type st, and reserving 'reserve' sectors for
2417 * a possible bitmap
2418 */
2419 static __u64 avail_size1(struct supertype *st, __u64 devsize,
2420 unsigned long long data_offset)
2421 {
2422 struct mdp_superblock_1 *super = st->sb;
2423 int bmspace = 0;
2424 int bbspace = 0;
2425 if (devsize < 24)
2426 return 0;
2427
2428 if (__le32_to_cpu(super->feature_map) & MD_FEATURE_BITMAP_OFFSET) {
2429 /* hot-add. allow for actual size of bitmap */
2430 struct bitmap_super_s *bsb;
2431 bsb = (struct bitmap_super_s *)(((char *)super) + MAX_SB_SIZE);
2432 bmspace = calc_bitmap_size(bsb, 4096) >> 9;
2433 } else if (md_feature_any_ppl_on(super->feature_map)) {
2434 bmspace = __le16_to_cpu(super->ppl.size);
2435 }
2436
2437 /* Allow space for bad block log */
2438 if (super->bblog_size)
2439 bbspace = __le16_to_cpu(super->bblog_size);
2440
2441 if (st->minor_version < 0)
2442 /* not specified, so time to set default */
2443 st->minor_version = 2;
2444
2445 if (data_offset == INVALID_SECTORS)
2446 data_offset = st->data_offset;
2447
2448 if (data_offset != INVALID_SECTORS)
2449 switch(st->minor_version) {
2450 case 0:
2451 return devsize - data_offset - 8*2 - bbspace;
2452 case 1:
2453 case 2:
2454 return devsize - data_offset;
2455 default:
2456 return 0;
2457 }
2458
2459 devsize -= bmspace;
2460
2461 switch(st->minor_version) {
2462 case 0:
2463 /* at end */
2464 return ((devsize - 8*2 - bbspace ) & ~(4*2-1));
2465 case 1:
2466 /* at start, 4K for superblock and possible bitmap */
2467 return devsize - 4*2 - bbspace;
2468 case 2:
2469 /* 4k from start, 4K for superblock and possible bitmap */
2470 return devsize - (4+4)*2 - bbspace;
2471 }
2472 return 0;
2473 }
2474
2475 static int
2476 add_internal_bitmap1(struct supertype *st,
2477 int *chunkp, int delay, int write_behind,
2478 unsigned long long size,
2479 int may_change, int major)
2480 {
2481 /*
2482 * If not may_change, then this is a 'Grow' without sysfs support for
2483 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2484 * If may_change, then this is create or a Grow with sysfs support,
2485 * and we can put the bitmap wherever we like.
2486 *
2487 * size is in sectors, chunk is in bytes !!!
2488 */
2489
2490 unsigned long long bits;
2491 unsigned long long max_bits;
2492 unsigned long long min_chunk;
2493 long offset;
2494 long bbl_offset, bbl_size;
2495 unsigned long long chunk = *chunkp;
2496 int room = 0;
2497 int creating = 0;
2498 int len;
2499 struct mdp_superblock_1 *sb = st->sb;
2500 bitmap_super_t *bms = (bitmap_super_t *)(((char *)sb) + MAX_SB_SIZE);
2501 int uuid[4];
2502
2503 if (__le64_to_cpu(sb->data_size) == 0)
2504 /*
2505 * Must be creating the array, else data_size
2506 * would be non-zero
2507 */
2508 creating = 1;
2509 switch(st->minor_version) {
2510 case 0:
2511 /*
2512 * either 3K after the superblock (when hot-add),
2513 * or some amount of space before.
2514 */
2515 if (creating) {
2516 /*
2517 * We are creating array, so we *know* how much room has
2518 * been left.
2519 */
2520 offset = 0;
2521 bbl_size = 8;
2522 room =
2523 choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2524 } else {
2525 room = __le64_to_cpu(sb->super_offset)
2526 - __le64_to_cpu(sb->data_offset)
2527 - __le64_to_cpu(sb->data_size);
2528 bbl_size = __le16_to_cpu(sb->bblog_size);
2529 if (bbl_size < 8)
2530 bbl_size = 8;
2531 bbl_offset = (__s32)__le32_to_cpu(sb->bblog_offset);
2532 if (bbl_size < -bbl_offset)
2533 bbl_size = -bbl_offset;
2534
2535 if (!may_change ||
2536 (room < 3*2 && __le32_to_cpu(sb->max_dev) <= 384)) {
2537 room = 3*2;
2538 offset = 1*2;
2539 bbl_size = 0;
2540 } else {
2541 offset = 0; /* means movable offset */
2542 }
2543 }
2544 break;
2545 case 1:
2546 case 2: /* between superblock and data */
2547 if (creating) {
2548 offset = 4*2;
2549 bbl_size = 8;
2550 room =
2551 choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2552 } else {
2553 room = __le64_to_cpu(sb->data_offset)
2554 - __le64_to_cpu(sb->super_offset);
2555 bbl_size = __le16_to_cpu(sb->bblog_size);
2556 if (bbl_size)
2557 room =
2558 __le32_to_cpu(sb->bblog_offset) + bbl_size;
2559 else
2560 bbl_size = 8;
2561
2562 if (!may_change) {
2563 room -= 2; /* Leave 1K for superblock */
2564 offset = 2;
2565 bbl_size = 0;
2566 } else {
2567 room -= 4*2; /* leave 4K for superblock */
2568 offset = 4*2;
2569 }
2570 }
2571 break;
2572 default:
2573 return -ENOSPC;
2574 }
2575
2576 room -= bbl_size;
2577 if (chunk == UnSet && room > 128*2)
2578 /* Limit to 128K of bitmap when chunk size not requested */
2579 room = 128*2;
2580
2581 if (room <= 1)
2582 /* No room for a bitmap */
2583 return -ENOSPC;
2584
2585 max_bits = (room * 512 - sizeof(bitmap_super_t)) * 8;
2586
2587 min_chunk = 4096; /* sub-page chunks don't work yet.. */
2588 bits = (size * 512) / min_chunk + 1;
2589 while (bits > max_bits) {
2590 min_chunk *= 2;
2591 bits = (bits + 1) / 2;
2592 }
2593 if (chunk == UnSet) {
2594 /* For practical purpose, 64Meg is a good
2595 * default chunk size for internal bitmaps.
2596 */
2597 chunk = min_chunk;
2598 if (chunk < 64*1024*1024)
2599 chunk = 64*1024*1024;
2600 } else if (chunk < min_chunk)
2601 return -EINVAL; /* chunk size too small */
2602 if (chunk == 0) /* rounding problem */
2603 return -EINVAL;
2604
2605 if (offset == 0) {
2606 /* start bitmap on a 4K boundary with enough space for
2607 * the bitmap
2608 */
2609 bits = (size * 512) / chunk + 1;
2610 room = ((bits + 7) / 8 + sizeof(bitmap_super_t) + 4095) / 4096;
2611 room *= 8; /* convert 4K blocks to sectors */
2612 offset = -room - bbl_size;
2613 }
2614
2615 sb->bitmap_offset = (int32_t)__cpu_to_le32(offset);
2616
2617 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map) |
2618 MD_FEATURE_BITMAP_OFFSET);
2619 memset(bms, 0, sizeof(*bms));
2620 bms->magic = __cpu_to_le32(BITMAP_MAGIC);
2621 bms->version = __cpu_to_le32(major);
2622 uuid_from_super1(st, uuid);
2623 memcpy(bms->uuid, uuid, 16);
2624 bms->chunksize = __cpu_to_le32(chunk);
2625 bms->daemon_sleep = __cpu_to_le32(delay);
2626 bms->sync_size = __cpu_to_le64(size);
2627 bms->write_behind = __cpu_to_le32(write_behind);
2628 bms->nodes = __cpu_to_le32(st->nodes);
2629 if (st->nodes)
2630 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map) |
2631 MD_FEATURE_BITMAP_VERSIONED);
2632 if (st->cluster_name) {
2633 len = sizeof(bms->cluster_name);
2634 strncpy((char *)bms->cluster_name, st->cluster_name, len);
2635 bms->cluster_name[len - 1] = '\0';
2636 }
2637
2638 *chunkp = chunk;
2639 return 0;
2640 }
2641
2642 static int locate_bitmap1(struct supertype *st, int fd, int node_num)
2643 {
2644 unsigned long long offset, bm_sectors_per_node;
2645 struct mdp_superblock_1 *sb;
2646 bitmap_super_t *bms;
2647 int mustfree = 0;
2648 int ret;
2649
2650 if (!st->sb) {
2651 if (st->ss->load_super(st, fd, NULL))
2652 return -1; /* no error I hope... */
2653 mustfree = 1;
2654 }
2655 sb = st->sb;
2656
2657 if ((__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
2658 ret = 0;
2659 else
2660 ret = -1;
2661
2662 offset = __le64_to_cpu(sb->super_offset) + (int32_t)__le32_to_cpu(sb->bitmap_offset);
2663 if (node_num) {
2664 bms = (bitmap_super_t *)(((char *)sb) + MAX_SB_SIZE);
2665 bm_sectors_per_node = calc_bitmap_size(bms, 4096) >> 9;
2666 offset += bm_sectors_per_node * node_num;
2667 }
2668 if (mustfree)
2669 free(sb);
2670 lseek64(fd, offset<<9, 0);
2671 return ret;
2672 }
2673
2674 static int write_bitmap1(struct supertype *st, int fd, enum bitmap_update update)
2675 {
2676 struct mdp_superblock_1 *sb = st->sb;
2677 bitmap_super_t *bms = (bitmap_super_t *)(((char *)sb) + MAX_SB_SIZE);
2678 int rv = 0;
2679 void *buf;
2680 int towrite, n, len;
2681 struct align_fd afd;
2682 unsigned int i = 0;
2683 unsigned long long total_bm_space, bm_space_per_node;
2684
2685 switch (update) {
2686 case NameUpdate:
2687 /* update cluster name */
2688 if (st->cluster_name) {
2689 len = sizeof(bms->cluster_name);
2690 memset((char *)bms->cluster_name, 0, len);
2691 strncpy((char *)bms->cluster_name,
2692 st->cluster_name, len);
2693 bms->cluster_name[len - 1] = '\0';
2694 }
2695 break;
2696 case NodeNumUpdate:
2697 /* cluster md only supports superblock 1.2 now */
2698 if (st->minor_version != 2 &&
2699 bms->version == BITMAP_MAJOR_CLUSTERED) {
2700 pr_err("Warning: cluster md only works with superblock 1.2\n");
2701 return -EINVAL;
2702 }
2703
2704 if (bms->version == BITMAP_MAJOR_CLUSTERED) {
2705 if (st->nodes == 1) {
2706 /* the parameter for nodes is not valid */
2707 pr_err("Warning: cluster-md at least needs two nodes\n");
2708 return -EINVAL;
2709 } else if (st->nodes == 0) {
2710 /*
2711 * parameter "--nodes" is not specified, (eg, add a disk to
2712 * clustered raid)
2713 */
2714 break;
2715 } else if (__cpu_to_le32(st->nodes) < bms->nodes) {
2716 /*
2717 * Since the nodes num is not increased, no
2718 * need to check the space enough or not,
2719 * just update bms->nodes
2720 */
2721 bms->nodes = __cpu_to_le32(st->nodes);
2722 break;
2723 }
2724 } else {
2725 /*
2726 * no need to change bms->nodes for other
2727 * bitmap types
2728 */
2729 if (st->nodes)
2730 pr_err("Warning: --nodes option is only suitable for clustered bitmap\n");
2731 break;
2732 }
2733
2734 /*
2735 * Each node has an independent bitmap, it is necessary to
2736 * calculate the space is enough or not, first get how many
2737 * bytes for the total bitmap
2738 */
2739 bm_space_per_node = calc_bitmap_size(bms, 4096);
2740
2741 total_bm_space = 512 * (__le64_to_cpu(sb->data_offset) -
2742 __le64_to_cpu(sb->super_offset));
2743 /* leave another 4k for superblock */
2744 total_bm_space = total_bm_space - 4096;
2745
2746 if (bm_space_per_node * st->nodes > total_bm_space) {
2747 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2748 total_bm_space / bm_space_per_node);
2749 return -ENOMEM;
2750 }
2751
2752 bms->nodes = __cpu_to_le32(st->nodes);
2753 break;
2754 case NoUpdate:
2755 default:
2756 break;
2757 }
2758
2759 init_afd(&afd, fd);
2760
2761 if (locate_bitmap1(st, fd, 0) < 0) {
2762 pr_err("Error: Invalid bitmap\n");
2763 return -EINVAL;
2764 }
2765
2766 if (posix_memalign(&buf, 4096, 4096))
2767 return -ENOMEM;
2768
2769 do {
2770 /* Only the bitmap[0] should resync
2771 * whole device on initial assembly
2772 */
2773 if (i)
2774 memset(buf, 0x00, 4096);
2775 else
2776 memset(buf, 0xff, 4096);
2777 memcpy(buf, (char *)bms, sizeof(bitmap_super_t));
2778
2779 /*
2780 * use 4096 boundary if bitmap_offset is aligned
2781 * with 8 sectors, then it should compatible with
2782 * older mdadm.
2783 */
2784 if (__le32_to_cpu(sb->bitmap_offset) & 7)
2785 towrite = calc_bitmap_size(bms, 512);
2786 else
2787 towrite = calc_bitmap_size(bms, 4096);
2788 while (towrite > 0) {
2789 n = towrite;
2790 if (n > 4096)
2791 n = 4096;
2792 n = awrite(&afd, buf, n);
2793 if (n > 0)
2794 towrite -= n;
2795 else
2796 break;
2797 if (i)
2798 memset(buf, 0x00, 4096);
2799 else
2800 memset(buf, 0xff, 4096);
2801 }
2802 fsync(fd);
2803 if (towrite) {
2804 rv = -2;
2805 break;
2806 }
2807 } while (++i < __le32_to_cpu(bms->nodes));
2808
2809 free(buf);
2810 return rv;
2811 }
2812
2813 static void free_super1(struct supertype *st)
2814 {
2815
2816 if (st->sb)
2817 free(st->sb);
2818 while (st->info) {
2819 struct devinfo *di = st->info;
2820 st->info = di->next;
2821 if (di->fd >= 0)
2822 close(di->fd);
2823 free(di);
2824 }
2825 st->sb = NULL;
2826 }
2827
2828 static int validate_geometry1(struct supertype *st, int level,
2829 int layout, int raiddisks,
2830 int *chunk, unsigned long long size,
2831 unsigned long long data_offset,
2832 char *subdev, unsigned long long *freesize,
2833 int consistency_policy, int verbose)
2834 {
2835 unsigned long long ldsize, devsize;
2836 int bmspace;
2837 unsigned long long headroom;
2838 unsigned long long overhead;
2839 int fd;
2840
2841 if (is_container(level)) {
2842 if (verbose)
2843 pr_err("1.x metadata does not support containers\n");
2844 return 0;
2845 }
2846 if (*chunk == UnSet)
2847 *chunk = DEFAULT_CHUNK;
2848
2849 if (!subdev)
2850 return 1;
2851
2852 if (st->minor_version < 0)
2853 /* not specified, so time to set default */
2854 st->minor_version = 2;
2855
2856 fd = open(subdev, O_RDONLY|O_EXCL, 0);
2857 if (fd < 0) {
2858 if (verbose)
2859 pr_err("super1.x cannot open %s: %s\n",
2860 subdev, strerror(errno));
2861 return 0;
2862 }
2863
2864 if (!get_dev_size(fd, subdev, &ldsize)) {
2865 close(fd);
2866 return 0;
2867 }
2868 close(fd);
2869
2870 devsize = ldsize >> 9;
2871
2872 /* creating: allow suitable space for bitmap or PPL */
2873 if (consistency_policy == CONSISTENCY_POLICY_PPL)
2874 bmspace = MULTIPLE_PPL_AREA_SIZE_SUPER1 >> 9;
2875 else
2876 bmspace = choose_bm_space(devsize);
2877
2878 if (data_offset == INVALID_SECTORS)
2879 data_offset = st->data_offset;
2880 if (data_offset == INVALID_SECTORS)
2881 switch (st->minor_version) {
2882 case 0:
2883 data_offset = 0;
2884 break;
2885 case 1:
2886 case 2:
2887 /* Choose data offset appropriate for this device
2888 * and use as default for whole array.
2889 * The data_offset must allow for bitmap space
2890 * and base metadata, should allow for some headroom
2891 * for reshape, and should be rounded to multiple
2892 * of 1M.
2893 * Headroom is limited to 128M, but aim for about 0.1%
2894 */
2895 headroom = 128*1024*2;
2896 while ((headroom << 10) > devsize &&
2897 (*chunk == 0 ||
2898 headroom / 2 >= ((unsigned)(*chunk)*2)*2))
2899 headroom >>= 1;
2900 data_offset = 12*2 + bmspace + headroom;
2901 #define ONE_MEG (2*1024)
2902 data_offset = ROUND_UP(data_offset, ONE_MEG);
2903 break;
2904 }
2905 if (st->data_offset == INVALID_SECTORS)
2906 st->data_offset = data_offset;
2907 switch(st->minor_version) {
2908 case 0: /* metadata at end. Round down and subtract space to reserve */
2909 devsize = (devsize & ~(4ULL*2-1));
2910 /* space for metadata, bblog, bitmap/ppl */
2911 overhead = 8*2 + 8 + bmspace;
2912 if (devsize < overhead) /* detect underflow */
2913 goto dev_too_small_err;
2914 devsize -= overhead;
2915 break;
2916 case 1:
2917 case 2:
2918 if (devsize < data_offset) /* detect underflow */
2919 goto dev_too_small_err;
2920 devsize -= data_offset;
2921 break;
2922 }
2923 *freesize = devsize;
2924 return 1;
2925
2926 /* Error condition, device cannot even hold the overhead. */
2927 dev_too_small_err:
2928 fprintf(stderr, "device %s is too small (%lluK) for "
2929 "required metadata!\n", subdev, devsize>>1);
2930 *freesize = 0;
2931 return 0;
2932 }
2933
2934 void *super1_make_v0(struct supertype *st, struct mdinfo *info, mdp_super_t *sb0)
2935 {
2936 /* Create a v1.0 superblock based on 'info'*/
2937 void *ret;
2938 struct mdp_superblock_1 *sb;
2939 int i;
2940 unsigned long long offset;
2941
2942 if (posix_memalign(&ret, 4096, 1024) != 0)
2943 return NULL;
2944 sb = ret;
2945 memset(ret, 0, 1024);
2946 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
2947 sb->major_version = __cpu_to_le32(1);
2948
2949 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
2950 sprintf(sb->set_name, "%d", sb0->md_minor);
2951 sb->ctime = __cpu_to_le32(info->array.ctime + 1);
2952 sb->level = __cpu_to_le32(info->array.level);
2953 sb->layout = __cpu_to_le32(info->array.layout);
2954 sb->size = __cpu_to_le64(info->component_size);
2955 sb->chunksize = __cpu_to_le32(info->array.chunk_size / 512);
2956 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
2957 if (info->array.level > 0)
2958 sb->data_size = sb->size;
2959 else
2960 sb->data_size = st->ss->avail_size(st, st->devsize / 512, 0);
2961 sb->resync_offset = MaxSector;
2962 sb->max_dev = __cpu_to_le32(MD_SB_DISKS);
2963 sb->dev_number = __cpu_to_le32(info->disk.number);
2964 sb->utime = __cpu_to_le64(info->array.utime);
2965
2966 offset = st->devsize/512 - 8*2;
2967 offset &= ~(4*2-1);
2968 sb->super_offset = __cpu_to_le64(offset);
2969 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2970
2971 random_uuid(sb->device_uuid);
2972
2973 for (i = 0; i < MD_SB_DISKS; i++) {
2974 int state = sb0->disks[i].state;
2975 sb->dev_roles[i] = MD_DISK_ROLE_SPARE;
2976 if ((state & (1<<MD_DISK_SYNC)) &&
2977 !(state & (1<<MD_DISK_FAULTY)))
2978 sb->dev_roles[i] = __cpu_to_le16(sb0->disks[i].raid_disk);
2979 }
2980 sb->sb_csum = calc_sb_1_csum(sb);
2981 return ret;
2982 }
2983
2984 struct superswitch super1 = {
2985 .examine_super = examine_super1,
2986 .brief_examine_super = brief_examine_super1,
2987 .export_examine_super = export_examine_super1,
2988 .detail_super = detail_super1,
2989 .brief_detail_super = brief_detail_super1,
2990 .export_detail_super = export_detail_super1,
2991 .write_init_super = write_init_super1,
2992 .validate_geometry = validate_geometry1,
2993 .add_to_super = add_to_super1,
2994 .examine_badblocks = examine_badblocks_super1,
2995 .copy_metadata = copy_metadata1,
2996 .write_init_ppl = write_init_ppl1,
2997 .match_home = match_home1,
2998 .uuid_from_super = uuid_from_super1,
2999 .getinfo_super = getinfo_super1,
3000 .container_content = container_content1,
3001 .update_super = update_super1,
3002 .init_super = init_super1,
3003 .store_super = store_super1,
3004 .compare_super = compare_super1,
3005 .load_super = load_super1,
3006 .match_metadata_desc = match_metadata_desc1,
3007 .avail_size = avail_size1,
3008 .add_internal_bitmap = add_internal_bitmap1,
3009 .locate_bitmap = locate_bitmap1,
3010 .write_bitmap = write_bitmap1,
3011 .free_super = free_super1,
3012 #if __BYTE_ORDER == BIG_ENDIAN
3013 .swapuuid = 0,
3014 #else
3015 .swapuuid = 1,
3016 #endif
3017 .name = "1.x",
3018 };
Unnamed repository; edit this file 'description' to name the repository.