]> 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
imsm: validate multiple ppls during assemble
[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 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_ALL (MD_FEATURE_BITMAP_OFFSET \
148 |MD_FEATURE_RECOVERY_OFFSET \
149 |MD_FEATURE_RESHAPE_ACTIVE \
150 |MD_FEATURE_BAD_BLOCKS \
151 |MD_FEATURE_REPLACEMENT \
152 |MD_FEATURE_RESHAPE_BACKWARDS \
153 |MD_FEATURE_NEW_OFFSET \
154 |MD_FEATURE_BITMAP_VERSIONED \
155 |MD_FEATURE_JOURNAL \
156 |MD_FEATURE_PPL \
157 |MD_FEATURE_MULTIPLE_PPLS \
158 )
159
160 static int role_from_sb(struct mdp_superblock_1 *sb)
161 {
162 unsigned int d;
163 int role;
164
165 d = __le32_to_cpu(sb->dev_number);
166 if (d < __le32_to_cpu(sb->max_dev))
167 role = __le16_to_cpu(sb->dev_roles[d]);
168 else
169 role = MD_DISK_ROLE_SPARE;
170 return role;
171 }
172
173 /* return how many bytes are needed for bitmap, for cluster-md each node
174 * should have it's own bitmap */
175 static unsigned int calc_bitmap_size(bitmap_super_t *bms, unsigned int boundary)
176 {
177 unsigned long long bits, bytes;
178
179 bits = bitmap_bits(__le64_to_cpu(bms->sync_size),
180 __le32_to_cpu(bms->chunksize));
181 bytes = (bits+7) >> 3;
182 bytes += sizeof(bitmap_super_t);
183 bytes = ROUND_UP(bytes, boundary);
184
185 return bytes;
186 }
187
188 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
189 {
190 unsigned int disk_csum, csum;
191 unsigned long long newcsum;
192 int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2;
193 unsigned int *isuper = (unsigned int*)sb;
194
195 /* make sure I can count... */
196 if (offsetof(struct mdp_superblock_1,data_offset) != 128 ||
197 offsetof(struct mdp_superblock_1, utime) != 192 ||
198 sizeof(struct mdp_superblock_1) != 256) {
199 fprintf(stderr, "WARNING - superblock isn't sized correctly\n");
200 }
201
202 disk_csum = sb->sb_csum;
203 sb->sb_csum = 0;
204 newcsum = 0;
205 for (; size>=4; size -= 4 ) {
206 newcsum += __le32_to_cpu(*isuper);
207 isuper++;
208 }
209
210 if (size == 2)
211 newcsum += __le16_to_cpu(*(unsigned short*) isuper);
212
213 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
214 sb->sb_csum = disk_csum;
215 return __cpu_to_le32(csum);
216 }
217
218 /*
219 * Information related to file descriptor used for aligned reads/writes.
220 * Cache the block size.
221 */
222 struct align_fd {
223 int fd;
224 int blk_sz;
225 };
226
227 static void init_afd(struct align_fd *afd, int fd)
228 {
229 afd->fd = fd;
230 if (!get_dev_sector_size(afd->fd, NULL, (unsigned int *)&afd->blk_sz))
231 afd->blk_sz = 512;
232 }
233
234 static char abuf[4096+4096];
235
236 static int aread(struct align_fd *afd, void *buf, int len)
237 {
238 /* aligned read.
239 * On devices with a 4K sector size, we need to read
240 * the full sector and copy relevant bits into
241 * the buffer
242 */
243 int bsize, iosize;
244 char *b;
245 int n;
246
247 bsize = afd->blk_sz;
248
249 if (!bsize || bsize > 4096 || len > 4096) {
250 if (!bsize)
251 fprintf(stderr, "WARNING - aread() called with invalid block size\n");
252 return -1;
253 }
254 b = ROUND_UP_PTR((char *)abuf, 4096);
255
256 for (iosize = 0; iosize < len; iosize += bsize)
257 ;
258 n = read(afd->fd, b, iosize);
259 if (n <= 0)
260 return n;
261 lseek(afd->fd, len - n, 1);
262 if (n > len)
263 n = len;
264 memcpy(buf, b, n);
265 return n;
266 }
267
268 static int awrite(struct align_fd *afd, void *buf, int len)
269 {
270 /* aligned write.
271 * On devices with a 4K sector size, we need to write
272 * the full sector. We pre-read if the sector is larger
273 * than the write.
274 * The address must be sector-aligned.
275 */
276 int bsize, iosize;
277 char *b;
278 int n;
279
280 bsize = afd->blk_sz;
281 if (!bsize || bsize > 4096 || len > 4096) {
282 if (!bsize)
283 fprintf(stderr, "WARNING - awrite() called with invalid block size\n");
284 return -1;
285 }
286 b = ROUND_UP_PTR((char *)abuf, 4096);
287
288 for (iosize = 0; iosize < len ; iosize += bsize)
289 ;
290
291 if (len != iosize) {
292 n = read(afd->fd, b, iosize);
293 if (n <= 0)
294 return n;
295 lseek(afd->fd, -n, 1);
296 }
297
298 memcpy(b, buf, len);
299 n = write(afd->fd, b, iosize);
300 if (n <= 0)
301 return n;
302 lseek(afd->fd, len - n, 1);
303 return len;
304 }
305
306 static inline unsigned int md_feature_any_ppl_on(__u32 feature_map)
307 {
308 return ((__cpu_to_le32(feature_map) &
309 (MD_FEATURE_PPL | MD_FEATURE_MUTLIPLE_PPLS)));
310 }
311
312 static inline unsigned int choose_ppl_space(int chunk)
313 {
314 return (PPL_HEADER_SIZE >> 9) + (chunk > 128*2 ? chunk : 128*2);
315 }
316
317 static void examine_super1(struct supertype *st, char *homehost)
318 {
319 struct mdp_superblock_1 *sb = st->sb;
320 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
321 time_t atime;
322 unsigned int d;
323 int role;
324 int delta_extra = 0;
325 int i;
326 char *c;
327 int l = homehost ? strlen(homehost) : 0;
328 int layout;
329 unsigned long long sb_offset;
330 struct mdinfo info;
331
332 printf(" Magic : %08x\n", __le32_to_cpu(sb->magic));
333 printf(" Version : 1");
334 sb_offset = __le64_to_cpu(sb->super_offset);
335 if (sb_offset <= 4)
336 printf(".1\n");
337 else if (sb_offset <= 8)
338 printf(".2\n");
339 else
340 printf(".0\n");
341 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb->feature_map));
342 printf(" Array UUID : ");
343 for (i=0; i<16; i++) {
344 if ((i&3)==0 && i != 0) printf(":");
345 printf("%02x", sb->set_uuid[i]);
346 }
347 printf("\n");
348 printf(" Name : %.32s", sb->set_name);
349 if (l > 0 && l < 32 &&
350 sb->set_name[l] == ':' &&
351 strncmp(sb->set_name, homehost, l) == 0)
352 printf(" (local to host %s)", homehost);
353 printf("\n");
354 if (bms->nodes > 0 &&
355 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
356 printf(" Cluster Name : %-64s\n", bms->cluster_name);
357 atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL;
358 printf(" Creation Time : %.24s\n", ctime(&atime));
359 c=map_num(pers, __le32_to_cpu(sb->level));
360 printf(" Raid Level : %s\n", c?c:"-unknown-");
361 printf(" Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks));
362 printf("\n");
363 printf(" Avail Dev Size : %llu%s\n",
364 (unsigned long long)__le64_to_cpu(sb->data_size),
365 human_size(__le64_to_cpu(sb->data_size)<<9));
366 if (__le32_to_cpu(sb->level) > 0) {
367 int ddsks = 0, ddsks_denom = 1;
368 switch(__le32_to_cpu(sb->level)) {
369 case 1: ddsks=1;break;
370 case 4:
371 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
372 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
373 case 10:
374 layout = __le32_to_cpu(sb->layout);
375 ddsks = __le32_to_cpu(sb->raid_disks);
376 ddsks_denom = (layout&255) * ((layout>>8)&255);
377 }
378 if (ddsks) {
379 long long asize = __le64_to_cpu(sb->size);
380 asize = (asize << 9) * ddsks / ddsks_denom;
381 printf(" Array Size : %llu%s\n",
382 asize >> 10, human_size(asize));
383 }
384 if (sb->size != sb->data_size)
385 printf(" Used Dev Size : %llu%s\n",
386 (unsigned long long)__le64_to_cpu(sb->size),
387 human_size(__le64_to_cpu(sb->size)<<9));
388 }
389 if (sb->data_offset)
390 printf(" Data Offset : %llu sectors\n",
391 (unsigned long long)__le64_to_cpu(sb->data_offset));
392 if (sb->new_offset &&
393 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) {
394 unsigned long long offset = __le64_to_cpu(sb->data_offset);
395 offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset);
396 printf(" New Offset : %llu sectors\n", offset);
397 }
398 printf(" Super Offset : %llu sectors\n",
399 (unsigned long long)__le64_to_cpu(sb->super_offset));
400 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)
401 printf("Recovery Offset : %llu sectors\n",
402 (unsigned long long)__le64_to_cpu(sb->recovery_offset));
403
404 st->ss->getinfo_super(st, &info, NULL);
405 if (info.space_after != 1 &&
406 !(__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
407 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
408 info.space_before, info.space_after);
409
410 printf(" State : %s\n",
411 (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean");
412 printf(" Device UUID : ");
413 for (i=0; i<16; i++) {
414 if ((i&3)==0 && i != 0)
415 printf(":");
416 printf("%02x", sb->device_uuid[i]);
417 }
418 printf("\n");
419 printf("\n");
420 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
421 printf("Internal Bitmap : %ld sectors from superblock\n",
422 (long)(int32_t)__le32_to_cpu(sb->bitmap_offset));
423 } else if (md_feature_any_ppl_on(sb->feature_map)) {
424 printf(" PPL : %u sectors at offset %d sectors from superblock\n",
425 __le16_to_cpu(sb->ppl.size),
426 __le16_to_cpu(sb->ppl.offset));
427 }
428 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)) {
429 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)
430 __le64_to_cpu(sb->reshape_position)/2,
431 human_size(__le64_to_cpu(sb->reshape_position)<<9));
432 if (__le32_to_cpu(sb->delta_disks)) {
433 printf(" Delta Devices : %d",
434 __le32_to_cpu(sb->delta_disks));
435 printf(" (%d->%d)\n",
436 __le32_to_cpu(sb->raid_disks) -
437 __le32_to_cpu(sb->delta_disks),
438 __le32_to_cpu(sb->raid_disks));
439 if ((int)__le32_to_cpu(sb->delta_disks) < 0)
440 delta_extra = -__le32_to_cpu(sb->delta_disks);
441 }
442 if (__le32_to_cpu(sb->new_level) != __le32_to_cpu(sb->level)) {
443 c = map_num(pers, __le32_to_cpu(sb->new_level));
444 printf(" New Level : %s\n", c?c:"-unknown-");
445 }
446 if (__le32_to_cpu(sb->new_layout) !=
447 __le32_to_cpu(sb->layout)) {
448 if (__le32_to_cpu(sb->level) == 5) {
449 c = map_num(r5layout,
450 __le32_to_cpu(sb->new_layout));
451 printf(" New Layout : %s\n", c?c:"-unknown-");
452 }
453 if (__le32_to_cpu(sb->level) == 6) {
454 c = map_num(r6layout,
455 __le32_to_cpu(sb->new_layout));
456 printf(" New Layout : %s\n", c?c:"-unknown-");
457 }
458 if (__le32_to_cpu(sb->level) == 10) {
459 printf(" New Layout :");
460 print_r10_layout(__le32_to_cpu(sb->new_layout));
461 printf("\n");
462 }
463 }
464 if (__le32_to_cpu(sb->new_chunk) !=
465 __le32_to_cpu(sb->chunksize))
466 printf(" New Chunksize : %dK\n",
467 __le32_to_cpu(sb->new_chunk)/2);
468 printf("\n");
469 }
470 if (sb->devflags) {
471 printf(" Flags :");
472 if (sb->devflags & WriteMostly1)
473 printf(" write-mostly");
474 if (sb->devflags & FailFast1)
475 printf(" failfast");
476 printf("\n");
477 }
478
479 atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL;
480 printf(" Update Time : %.24s\n", ctime(&atime));
481
482 if (sb->bblog_size && sb->bblog_offset) {
483 printf(" Bad Block Log : %d entries available at offset %ld sectors",
484 __le16_to_cpu(sb->bblog_size)*512/8,
485 (long)(int32_t)__le32_to_cpu(sb->bblog_offset));
486 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
487 printf(" - bad blocks present.");
488 printf("\n");
489 }
490
491 if (calc_sb_1_csum(sb) == sb->sb_csum)
492 printf(" Checksum : %x - correct\n",
493 __le32_to_cpu(sb->sb_csum));
494 else
495 printf(" Checksum : %x - expected %x\n",
496 __le32_to_cpu(sb->sb_csum),
497 __le32_to_cpu(calc_sb_1_csum(sb)));
498 printf(" Events : %llu\n",
499 (unsigned long long)__le64_to_cpu(sb->events));
500 printf("\n");
501 if (__le32_to_cpu(sb->level) == 5) {
502 c = map_num(r5layout, __le32_to_cpu(sb->layout));
503 printf(" Layout : %s\n", c?c:"-unknown-");
504 }
505 if (__le32_to_cpu(sb->level) == 6) {
506 c = map_num(r6layout, __le32_to_cpu(sb->layout));
507 printf(" Layout : %s\n", c?c:"-unknown-");
508 }
509 if (__le32_to_cpu(sb->level) == 10) {
510 int lo = __le32_to_cpu(sb->layout);
511 printf(" Layout :");
512 print_r10_layout(lo);
513 printf("\n");
514 }
515 switch(__le32_to_cpu(sb->level)) {
516 case 0:
517 case 4:
518 case 5:
519 case 6:
520 case 10:
521 printf(" Chunk Size : %dK\n",
522 __le32_to_cpu(sb->chunksize)/2);
523 break;
524 case -1:
525 printf(" Rounding : %dK\n",
526 __le32_to_cpu(sb->chunksize)/2);
527 break;
528 default:
529 break;
530 }
531 printf("\n");
532 #if 0
533 /* This turns out to just be confusing */
534 printf(" Array Slot : %d (", __le32_to_cpu(sb->dev_number));
535 for (i = __le32_to_cpu(sb->max_dev); i> 0 ; i--)
536 if (__le16_to_cpu(sb->dev_roles[i-1]) != MD_DISK_ROLE_SPARE)
537 break;
538 for (d = 0; d < i; d++) {
539 int role = __le16_to_cpu(sb->dev_roles[d]);
540 if (d)
541 printf(", ");
542 if (role == MD_DISK_ROLE_SPARE)
543 printf("empty");
544 else
545 if(role == MD_DISK_ROLE_FAULTY)
546 printf("failed");
547 else
548 printf("%d", role);
549 }
550 printf(")\n");
551 #endif
552 printf(" Device Role : ");
553 role = role_from_sb(sb);
554 if (role >= MD_DISK_ROLE_FAULTY)
555 printf("spare\n");
556 else if (role == MD_DISK_ROLE_JOURNAL)
557 printf("Journal\n");
558 else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_REPLACEMENT))
559 printf("Replacement device %d\n", role);
560 else
561 printf("Active device %d\n", role);
562
563 printf(" Array State : ");
564 for (d = 0; d < __le32_to_cpu(sb->raid_disks) + delta_extra; d++) {
565 int cnt = 0;
566 unsigned int i;
567 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
568 unsigned int role = __le16_to_cpu(sb->dev_roles[i]);
569 if (role == d)
570 cnt++;
571 }
572 if (cnt == 2)
573 printf("R");
574 else if (cnt == 1)
575 printf("A");
576 else if (cnt == 0)
577 printf(".");
578 else
579 printf("?");
580 }
581 #if 0
582 /* This is confusing too */
583 faulty = 0;
584 for (i = 0; i< __le32_to_cpu(sb->max_dev); i++) {
585 int role = __le16_to_cpu(sb->dev_roles[i]);
586 if (role == MD_DISK_ROLE_FAULTY)
587 faulty++;
588 }
589 if (faulty)
590 printf(" %d failed", faulty);
591 #endif
592 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
593 printf("\n");
594 }
595
596 static void brief_examine_super1(struct supertype *st, int verbose)
597 {
598 struct mdp_superblock_1 *sb = st->sb;
599 int i;
600 unsigned long long sb_offset;
601 char *nm;
602 char *c = map_num(pers, __le32_to_cpu(sb->level));
603
604 nm = strchr(sb->set_name, ':');
605 if (nm)
606 nm++;
607 else if (sb->set_name[0])
608 nm = sb->set_name;
609 else
610 nm = NULL;
611
612 printf("ARRAY ");
613 if (nm) {
614 printf("/dev/md/");
615 print_escape(nm);
616 putchar(' ');
617 }
618 if (verbose && c)
619 printf(" level=%s", c);
620 sb_offset = __le64_to_cpu(sb->super_offset);
621 if (sb_offset <= 4)
622 printf(" metadata=1.1 ");
623 else if (sb_offset <= 8)
624 printf(" metadata=1.2 ");
625 else
626 printf(" metadata=1.0 ");
627 if (verbose)
628 printf("num-devices=%d ", __le32_to_cpu(sb->raid_disks));
629 printf("UUID=");
630 for (i = 0; i < 16; i++) {
631 if ((i&3)==0 && i != 0)
632 printf(":");
633 printf("%02x", sb->set_uuid[i]);
634 }
635 if (sb->set_name[0]) {
636 printf(" name=");
637 print_quoted(sb->set_name);
638 }
639 printf("\n");
640 }
641
642 static void export_examine_super1(struct supertype *st)
643 {
644 struct mdp_superblock_1 *sb = st->sb;
645 int i;
646 int len = 32;
647 int layout;
648
649 printf("MD_LEVEL=%s\n", map_num(pers, __le32_to_cpu(sb->level)));
650 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb->raid_disks));
651 for (i = 0; i < 32; i++)
652 if (sb->set_name[i] == '\n' || sb->set_name[i] == '\0') {
653 len = i;
654 break;
655 }
656 if (len)
657 printf("MD_NAME=%.*s\n", len, sb->set_name);
658 if (__le32_to_cpu(sb->level) > 0) {
659 int ddsks = 0, ddsks_denom = 1;
660 switch(__le32_to_cpu(sb->level)) {
661 case 1:
662 ddsks = 1;
663 break;
664 case 4:
665 case 5:
666 ddsks = __le32_to_cpu(sb->raid_disks)-1;
667 break;
668 case 6:
669 ddsks = __le32_to_cpu(sb->raid_disks)-2;
670 break;
671 case 10:
672 layout = __le32_to_cpu(sb->layout);
673 ddsks = __le32_to_cpu(sb->raid_disks);
674 ddsks_denom = (layout&255) * ((layout>>8)&255);
675 }
676 if (ddsks) {
677 long long asize = __le64_to_cpu(sb->size);
678 asize = (asize << 9) * ddsks / ddsks_denom;
679 printf("MD_ARRAY_SIZE=%s\n",
680 human_size_brief(asize, JEDEC));
681 }
682 }
683 printf("MD_UUID=");
684 for (i = 0; i < 16; i++) {
685 if ((i&3) == 0 && i != 0)
686 printf(":");
687 printf("%02x", sb->set_uuid[i]);
688 }
689 printf("\n");
690 printf("MD_UPDATE_TIME=%llu\n",
691 __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL);
692 printf("MD_DEV_UUID=");
693 for (i = 0; i < 16; i++) {
694 if ((i&3) == 0 && i != 0)
695 printf(":");
696 printf("%02x", sb->device_uuid[i]);
697 }
698 printf("\n");
699 printf("MD_EVENTS=%llu\n",
700 (unsigned long long)__le64_to_cpu(sb->events));
701 }
702
703 static int copy_metadata1(struct supertype *st, int from, int to)
704 {
705 /* Read superblock. If it looks good, write it out.
706 * Then if a bitmap is present, copy that.
707 * And if a bad-block-list is present, copy that too.
708 */
709 void *buf;
710 unsigned long long dsize, sb_offset;
711 const int bufsize = 4*1024;
712 struct mdp_superblock_1 super, *sb;
713
714 if (posix_memalign(&buf, 4096, bufsize) != 0)
715 return 1;
716
717 if (!get_dev_size(from, NULL, &dsize))
718 goto err;
719
720 dsize >>= 9;
721 if (dsize < 24)
722 goto err;
723 switch(st->minor_version) {
724 case 0:
725 sb_offset = dsize;
726 sb_offset -= 8*2;
727 sb_offset &= ~(4*2-1);
728 break;
729 case 1:
730 sb_offset = 0;
731 break;
732 case 2:
733 sb_offset = 4*2;
734 break;
735 default:
736 goto err;
737 }
738
739 if (lseek64(from, sb_offset << 9, 0) < 0LL)
740 goto err;
741 if (read(from, buf, bufsize) != bufsize)
742 goto err;
743
744 sb = buf;
745 super = *sb; // save most of sb for when we reuse buf
746
747 if (__le32_to_cpu(super.magic) != MD_SB_MAGIC ||
748 __le32_to_cpu(super.major_version) != 1 ||
749 __le64_to_cpu(super.super_offset) != sb_offset ||
750 calc_sb_1_csum(sb) != super.sb_csum)
751 goto err;
752
753 if (lseek64(to, sb_offset << 9, 0) < 0LL)
754 goto err;
755 if (write(to, buf, bufsize) != bufsize)
756 goto err;
757
758 if (super.feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
759 unsigned long long bitmap_offset = sb_offset;
760 int bytes = 4096; // just an estimate.
761 int written = 0;
762 struct align_fd afrom, ato;
763
764 init_afd(&afrom, from);
765 init_afd(&ato, to);
766
767 bitmap_offset += (int32_t)__le32_to_cpu(super.bitmap_offset);
768
769 if (lseek64(from, bitmap_offset<<9, 0) < 0)
770 goto err;
771 if (lseek64(to, bitmap_offset<<9, 0) < 0)
772 goto err;
773
774 for (written = 0; written < bytes ; ) {
775 int n = bytes - written;
776 if (n > 4096)
777 n = 4096;
778 if (aread(&afrom, buf, n) != n)
779 goto err;
780 if (written == 0) {
781 /* have the header, can calculate
782 * correct bitmap bytes */
783 bitmap_super_t *bms;
784 bms = (void*)buf;
785 bytes = calc_bitmap_size(bms, 512);
786 if (n > bytes)
787 n = bytes;
788 }
789 if (awrite(&ato, buf, n) != n)
790 goto err;
791 written += n;
792 }
793 }
794
795 if (super.bblog_size != 0 &&
796 __le16_to_cpu(super.bblog_size) <= 100 &&
797 super.bblog_offset != 0 &&
798 (super.feature_map & __le32_to_cpu(MD_FEATURE_BAD_BLOCKS))) {
799 /* There is a bad block log */
800 unsigned long long bb_offset = sb_offset;
801 int bytes = __le16_to_cpu(super.bblog_size) * 512;
802 int written = 0;
803 struct align_fd afrom, ato;
804
805 init_afd(&afrom, from);
806 init_afd(&ato, to);
807
808 bb_offset += (int32_t)__le32_to_cpu(super.bblog_offset);
809
810 if (lseek64(from, bb_offset<<9, 0) < 0)
811 goto err;
812 if (lseek64(to, bb_offset<<9, 0) < 0)
813 goto err;
814
815 for (written = 0; written < bytes ; ) {
816 int n = bytes - written;
817 if (n > 4096)
818 n = 4096;
819 if (aread(&afrom, buf, n) != n)
820 goto err;
821
822 if (awrite(&ato, buf, n) != n)
823 goto err;
824 written += n;
825 }
826 }
827
828 free(buf);
829 return 0;
830
831 err:
832 free(buf);
833 return 1;
834 }
835
836 static void detail_super1(struct supertype *st, char *homehost)
837 {
838 struct mdp_superblock_1 *sb = st->sb;
839 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
840 int i;
841 int l = homehost ? strlen(homehost) : 0;
842
843 printf(" Name : %.32s", sb->set_name);
844 if (l > 0 && l < 32 && sb->set_name[l] == ':' &&
845 strncmp(sb->set_name, homehost, l) == 0)
846 printf(" (local to host %s)", homehost);
847 if (bms->nodes > 0 &&
848 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
849 printf("\n Cluster Name : %-64s", bms->cluster_name);
850 printf("\n UUID : ");
851 for (i = 0; i < 16; i++) {
852 if ((i&3) == 0 && i != 0)
853 printf(":");
854 printf("%02x", sb->set_uuid[i]);
855 }
856 printf("\n Events : %llu\n\n",
857 (unsigned long long)__le64_to_cpu(sb->events));
858 }
859
860 static void brief_detail_super1(struct supertype *st)
861 {
862 struct mdp_superblock_1 *sb = st->sb;
863 int i;
864
865 if (sb->set_name[0]) {
866 printf(" name=");
867 print_quoted(sb->set_name);
868 }
869 printf(" UUID=");
870 for (i = 0; i < 16; i++) {
871 if ((i & 3) == 0 && i != 0)
872 printf(":");
873 printf("%02x", sb->set_uuid[i]);
874 }
875 }
876
877 static void export_detail_super1(struct supertype *st)
878 {
879 struct mdp_superblock_1 *sb = st->sb;
880 int i;
881 int len = 32;
882
883 for (i = 0; i < 32; i++)
884 if (sb->set_name[i] == '\n' || sb->set_name[i] == '\0') {
885 len = i;
886 break;
887 }
888 if (len)
889 printf("MD_NAME=%.*s\n", len, sb->set_name);
890 }
891
892 static int examine_badblocks_super1(struct supertype *st, int fd, char *devname)
893 {
894 struct mdp_superblock_1 *sb = st->sb;
895 unsigned long long offset;
896 int size;
897 __u64 *bbl, *bbp;
898 int i;
899
900 if (!sb->bblog_size || __le16_to_cpu(sb->bblog_size) > 100 ||
901 !sb->bblog_offset){
902 printf("No bad-blocks list configured on %s\n", devname);
903 return 0;
904 }
905 if ((sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS)) == 0) {
906 printf("Bad-blocks list is empty in %s\n", devname);
907 return 0;
908 }
909
910 size = __le16_to_cpu(sb->bblog_size)* 512;
911 if (posix_memalign((void**)&bbl, 4096, size) != 0) {
912 pr_err("could not allocate badblocks list\n");
913 return 0;
914 }
915 offset = __le64_to_cpu(sb->super_offset) +
916 (int)__le32_to_cpu(sb->bblog_offset);
917 offset <<= 9;
918 if (lseek64(fd, offset, 0) < 0) {
919 pr_err("Cannot seek to bad-blocks list\n");
920 return 1;
921 }
922 if (read(fd, bbl, size) != size) {
923 pr_err("Cannot read bad-blocks list\n");
924 return 1;
925 }
926 /* 64bits per entry. 10 bits is block-count, 54 bits is block
927 * offset. Blocks are sectors unless bblog->shift makes them bigger
928 */
929 bbp = (__u64*)bbl;
930 printf("Bad-blocks on %s:\n", devname);
931 for (i = 0; i < size/8; i++, bbp++) {
932 __u64 bb = __le64_to_cpu(*bbp);
933 int count = bb & 0x3ff;
934 unsigned long long sector = bb >> 10;
935
936 if (bb + 1 == 0)
937 break;
938
939 sector <<= sb->bblog_shift;
940 count <<= sb->bblog_shift;
941
942 printf("%20llu for %d sectors\n", sector, count);
943 }
944 return 0;
945 }
946
947 static int match_home1(struct supertype *st, char *homehost)
948 {
949 struct mdp_superblock_1 *sb = st->sb;
950 int l = homehost ? strlen(homehost) : 0;
951
952 return (l > 0 && l < 32 && sb->set_name[l] == ':' &&
953 strncmp(sb->set_name, homehost, l) == 0);
954 }
955
956 static void uuid_from_super1(struct supertype *st, int uuid[4])
957 {
958 struct mdp_superblock_1 *super = st->sb;
959 char *cuuid = (char*)uuid;
960 int i;
961 for (i = 0; i < 16; i++)
962 cuuid[i] = super->set_uuid[i];
963 }
964
965 static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
966 {
967 struct mdp_superblock_1 *sb = st->sb;
968 struct bitmap_super_s *bsb = (void*)(((char*)sb)+MAX_SB_SIZE);
969 struct misc_dev_info *misc =
970 (void*)(((char*)sb)+MAX_SB_SIZE+BM_SUPER_SIZE);
971 int working = 0;
972 unsigned int i;
973 unsigned int role;
974 unsigned int map_disks = info->array.raid_disks;
975 unsigned long long super_offset;
976 unsigned long long data_size;
977
978 memset(info, 0, sizeof(*info));
979 info->array.major_version = 1;
980 info->array.minor_version = st->minor_version;
981 info->array.patch_version = 0;
982 info->array.raid_disks = __le32_to_cpu(sb->raid_disks);
983 info->array.level = __le32_to_cpu(sb->level);
984 info->array.layout = __le32_to_cpu(sb->layout);
985 info->array.md_minor = -1;
986 info->array.ctime = __le64_to_cpu(sb->ctime);
987 info->array.utime = __le64_to_cpu(sb->utime);
988 info->array.chunk_size = __le32_to_cpu(sb->chunksize)*512;
989 info->array.state =
990 (__le64_to_cpu(sb->resync_offset) == MaxSector) ? 1 : 0;
991
992 super_offset = __le64_to_cpu(sb->super_offset);
993 info->data_offset = __le64_to_cpu(sb->data_offset);
994 info->component_size = __le64_to_cpu(sb->size);
995 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
996 info->bitmap_offset = (int32_t)__le32_to_cpu(sb->bitmap_offset);
997 if (__le32_to_cpu(bsb->nodes) > 1)
998 info->array.state |= (1 << MD_SB_CLUSTERED);
999 } else if (md_feature_any_ppl_on(sb->feature_map)) {
1000 info->ppl_offset = __le16_to_cpu(sb->ppl.offset);
1001 info->ppl_size = __le16_to_cpu(sb->ppl.size);
1002 info->ppl_sector = super_offset + info->ppl_offset;
1003 }
1004
1005 info->disk.major = 0;
1006 info->disk.minor = 0;
1007 info->disk.number = __le32_to_cpu(sb->dev_number);
1008 if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) ||
1009 __le32_to_cpu(sb->dev_number) >= MAX_DEVS)
1010 role = MD_DISK_ROLE_FAULTY;
1011 else
1012 role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]);
1013
1014 if (info->array.level <= 0)
1015 data_size = __le64_to_cpu(sb->data_size);
1016 else
1017 data_size = __le64_to_cpu(sb->size);
1018 if (info->data_offset < super_offset) {
1019 unsigned long long end;
1020 info->space_before = info->data_offset;
1021 end = super_offset;
1022
1023 if (sb->bblog_offset && sb->bblog_size) {
1024 unsigned long long bboffset = super_offset;
1025 bboffset += (int32_t)__le32_to_cpu(sb->bblog_offset);
1026 if (bboffset < end)
1027 end = bboffset;
1028 }
1029
1030 if (super_offset + info->bitmap_offset + info->ppl_offset < end)
1031 end = super_offset + info->bitmap_offset +
1032 info->ppl_offset;
1033
1034 if (info->data_offset + data_size < end)
1035 info->space_after = end - data_size - info->data_offset;
1036 else
1037 info->space_after = 0;
1038 } else {
1039 unsigned long long earliest;
1040 earliest = super_offset + (32+4)*2; /* match kernel */
1041 if (info->bitmap_offset > 0) {
1042 unsigned long long bmend = info->bitmap_offset;
1043 unsigned long long size = calc_bitmap_size(bsb, 4096);
1044 size /= 512;
1045 bmend += size;
1046 if (bmend > earliest)
1047 earliest = bmend;
1048 } else if (info->ppl_offset > 0) {
1049 unsigned long long pplend;
1050
1051 pplend = info->ppl_offset + info->ppl_size;
1052 if (pplend > earliest)
1053 earliest = pplend;
1054 }
1055 if (sb->bblog_offset && sb->bblog_size) {
1056 unsigned long long bbend = super_offset;
1057 bbend += (int32_t)__le32_to_cpu(sb->bblog_offset);
1058 bbend += __le16_to_cpu(sb->bblog_size);
1059 if (bbend > earliest)
1060 earliest = bbend;
1061 }
1062 if (earliest < info->data_offset)
1063 info->space_before = info->data_offset - earliest;
1064 else
1065 info->space_before = 0;
1066 info->space_after = misc->device_size - data_size -
1067 info->data_offset;
1068 }
1069 if (info->space_before == 0 && info->space_after == 0) {
1070 /* It will look like we don't support data_offset changes,
1071 * be we do - it's just that there is no room.
1072 * A change that reduced the number of devices should
1073 * still be allowed, so set the otherwise useless value of '1'
1074 */
1075 info->space_after = 1;
1076 }
1077
1078 info->disk.raid_disk = -1;
1079 switch(role) {
1080 case MD_DISK_ROLE_SPARE:
1081 /* spare: not active, not sync, not faulty */
1082 info->disk.state = 0;
1083 break;
1084 case MD_DISK_ROLE_FAULTY:
1085 info->disk.state = (1 << MD_DISK_FAULTY); /* faulty */
1086 break;
1087 case MD_DISK_ROLE_JOURNAL:
1088 info->disk.state = (1 << MD_DISK_JOURNAL);
1089 info->disk.raid_disk = role;
1090 /* journal uses all 4kB blocks*/
1091 info->space_after = (misc->device_size - info->data_offset) % 8;
1092 break;
1093 default:
1094 info->disk.state = 6; /* active and in sync */
1095 info->disk.raid_disk = role;
1096 }
1097 if (sb->devflags & WriteMostly1)
1098 info->disk.state |= (1 << MD_DISK_WRITEMOSTLY);
1099 if (sb->devflags & FailFast1)
1100 info->disk.state |= (1 << MD_DISK_FAILFAST);
1101 info->events = __le64_to_cpu(sb->events);
1102 sprintf(info->text_version, "1.%d", st->minor_version);
1103 info->safe_mode_delay = 200;
1104
1105 memcpy(info->uuid, sb->set_uuid, 16);
1106
1107 strncpy(info->name, sb->set_name, 32);
1108 info->name[32] = 0;
1109
1110 if ((__le32_to_cpu(sb->feature_map)&MD_FEATURE_REPLACEMENT)) {
1111 info->disk.state &= ~(1 << MD_DISK_SYNC);
1112 info->disk.state |= 1 << MD_DISK_REPLACEMENT;
1113 }
1114
1115 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET))
1116 info->recovery_start = __le32_to_cpu(sb->recovery_offset);
1117 else
1118 info->recovery_start = MaxSector;
1119
1120 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) {
1121 info->reshape_active = 1;
1122 if ((sb->feature_map & __le32_to_cpu(MD_FEATURE_NEW_OFFSET)) &&
1123 sb->new_offset != 0)
1124 info->reshape_active |= RESHAPE_NO_BACKUP;
1125 info->reshape_progress = __le64_to_cpu(sb->reshape_position);
1126 info->new_level = __le32_to_cpu(sb->new_level);
1127 info->delta_disks = __le32_to_cpu(sb->delta_disks);
1128 info->new_layout = __le32_to_cpu(sb->new_layout);
1129 info->new_chunk = __le32_to_cpu(sb->new_chunk)<<9;
1130 if (info->delta_disks < 0)
1131 info->array.raid_disks -= info->delta_disks;
1132 } else
1133 info->reshape_active = 0;
1134
1135 info->recovery_blocked = info->reshape_active;
1136
1137 if (map)
1138 for (i=0; i<map_disks; i++)
1139 map[i] = 0;
1140 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
1141 role = __le16_to_cpu(sb->dev_roles[i]);
1142 if (/*role == MD_DISK_ROLE_SPARE || */role < (unsigned) info->array.raid_disks) {
1143 working++;
1144 if (map && role < map_disks)
1145 map[role] = 1;
1146 }
1147 }
1148
1149 info->array.working_disks = working;
1150
1151 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_JOURNAL)) {
1152 info->journal_device_required = 1;
1153 info->consistency_policy = CONSISTENCY_POLICY_JOURNAL;
1154 } else if (md_feature_any_ppl_on(sb->feature_map)) {
1155 info->consistency_policy = CONSISTENCY_POLICY_PPL;
1156 } else if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
1157 info->consistency_policy = CONSISTENCY_POLICY_BITMAP;
1158 } else if (info->array.level <= 0) {
1159 info->consistency_policy = CONSISTENCY_POLICY_NONE;
1160 } else {
1161 info->consistency_policy = CONSISTENCY_POLICY_RESYNC;
1162 }
1163
1164 info->journal_clean = 0;
1165 }
1166
1167 static struct mdinfo *container_content1(struct supertype *st, char *subarray)
1168 {
1169 struct mdinfo *info;
1170
1171 if (subarray)
1172 return NULL;
1173
1174 info = xmalloc(sizeof(*info));
1175 getinfo_super1(st, info, NULL);
1176 return info;
1177 }
1178
1179 static int update_super1(struct supertype *st, struct mdinfo *info,
1180 char *update, char *devname, int verbose,
1181 int uuid_set, char *homehost)
1182 {
1183 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1184 * if any change was made. For others, the return value is
1185 * ignored.
1186 */
1187 int rv = 0;
1188 int lockid;
1189 struct mdp_superblock_1 *sb = st->sb;
1190 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1191
1192 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1193 rv = cluster_get_dlmlock(&lockid);
1194 if (rv) {
1195 pr_err("Cannot get dlmlock in %s return %d\n",
1196 __func__, rv);
1197 cluster_release_dlmlock(lockid);
1198 return rv;
1199 }
1200 }
1201
1202 if (strcmp(update, "homehost") == 0 &&
1203 homehost) {
1204 /* Note that 'homehost' is special as it is really
1205 * a "name" update.
1206 */
1207 char *c;
1208 update = "name";
1209 c = strchr(sb->set_name, ':');
1210 if (c)
1211 strncpy(info->name, c+1, 31 - (c-sb->set_name));
1212 else
1213 strncpy(info->name, sb->set_name, 32);
1214 info->name[32] = 0;
1215 }
1216
1217 if (strcmp(update, "force-one")==0) {
1218 /* Not enough devices for a working array,
1219 * so bring this one up-to-date
1220 */
1221 if (sb->events != __cpu_to_le64(info->events))
1222 rv = 1;
1223 sb->events = __cpu_to_le64(info->events);
1224 } else if (strcmp(update, "force-array")==0) {
1225 /* Degraded array and 'force' requests to
1226 * maybe need to mark it 'clean'.
1227 */
1228 switch(__le32_to_cpu(sb->level)) {
1229 case 4:
1230 case 5:
1231 case 6:
1232 /* need to force clean */
1233 if (sb->resync_offset != MaxSector)
1234 rv = 1;
1235 sb->resync_offset = MaxSector;
1236 }
1237 } else if (strcmp(update, "assemble")==0) {
1238 int d = info->disk.number;
1239 int want;
1240 if (info->disk.state & (1<<MD_DISK_ACTIVE))
1241 want = info->disk.raid_disk;
1242 else if (info->disk.state & (1<<MD_DISK_JOURNAL))
1243 want = MD_DISK_ROLE_JOURNAL;
1244 else
1245 want = MD_DISK_ROLE_SPARE;
1246 if (sb->dev_roles[d] != __cpu_to_le16(want)) {
1247 sb->dev_roles[d] = __cpu_to_le16(want);
1248 rv = 1;
1249 }
1250 if (info->reshape_active &&
1251 sb->feature_map &
1252 __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1253 info->delta_disks >= 0 &&
1254 info->reshape_progress <
1255 __le64_to_cpu(sb->reshape_position)) {
1256 sb->reshape_position =
1257 __cpu_to_le64(info->reshape_progress);
1258 rv = 1;
1259 }
1260 if (info->reshape_active &&
1261 sb->feature_map &
1262 __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1263 info->delta_disks < 0 &&
1264 info->reshape_progress >
1265 __le64_to_cpu(sb->reshape_position)) {
1266 sb->reshape_position =
1267 __cpu_to_le64(info->reshape_progress);
1268 rv = 1;
1269 }
1270 } else if (strcmp(update, "linear-grow-new") == 0) {
1271 unsigned int i;
1272 int fd;
1273 unsigned int max = __le32_to_cpu(sb->max_dev);
1274
1275 for (i = 0; i < max; i++)
1276 if (__le16_to_cpu(sb->dev_roles[i]) >=
1277 MD_DISK_ROLE_FAULTY)
1278 break;
1279 sb->dev_number = __cpu_to_le32(i);
1280 info->disk.number = i;
1281 if (i >= max) {
1282 sb->max_dev = __cpu_to_le32(max+1);
1283 }
1284
1285 random_uuid(sb->device_uuid);
1286
1287 sb->dev_roles[i] = __cpu_to_le16(info->disk.raid_disk);
1288
1289 fd = open(devname, O_RDONLY);
1290 if (fd >= 0) {
1291 unsigned long long ds;
1292 get_dev_size(fd, devname, &ds);
1293 close(fd);
1294 ds >>= 9;
1295 if (__le64_to_cpu(sb->super_offset) <
1296 __le64_to_cpu(sb->data_offset)) {
1297 sb->data_size = __cpu_to_le64(
1298 ds - __le64_to_cpu(sb->data_offset));
1299 } else {
1300 ds -= 8*2;
1301 ds &= ~(unsigned long long)(4*2-1);
1302 sb->super_offset = __cpu_to_le64(ds);
1303 sb->data_size = __cpu_to_le64(
1304 ds - __le64_to_cpu(sb->data_offset));
1305 }
1306 }
1307 } else if (strcmp(update, "linear-grow-update") == 0) {
1308 int max = __le32_to_cpu(sb->max_dev);
1309 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
1310 if (info->array.raid_disks > max) {
1311 sb->max_dev = __cpu_to_le32(max+1);
1312 }
1313 sb->dev_roles[info->disk.number] =
1314 __cpu_to_le16(info->disk.raid_disk);
1315 } else if (strcmp(update, "resync") == 0) {
1316 /* make sure resync happens */
1317 sb->resync_offset = 0ULL;
1318 } else if (strcmp(update, "uuid") == 0) {
1319 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
1320
1321 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)
1322 memcpy(bms->uuid, sb->set_uuid, 16);
1323 } else if (strcmp(update, "no-bitmap") == 0) {
1324 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1325 } else if (strcmp(update, "bbl") == 0) {
1326 /* only possible if there is room after the bitmap, or if
1327 * there is no bitmap
1328 */
1329 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1330 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1331 long bitmap_offset = 0;
1332 long bm_sectors = 0;
1333 long space;
1334
1335 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1336 bitmap_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1337 bm_sectors = calc_bitmap_size(bms, 4096) >> 9;
1338 } else if (md_feature_any_ppl_on(sb->feature_map)) {
1339 bitmap_offset = (long)__le16_to_cpu(sb->ppl.offset);
1340 bm_sectors = (long)__le16_to_cpu(sb->ppl.size);
1341 }
1342
1343 if (sb_offset < data_offset) {
1344 /*
1345 * 1.1 or 1.2. Put bbl after bitmap leaving
1346 * at least 32K
1347 */
1348 long bb_offset;
1349 bb_offset = sb_offset + 8;
1350 if (bm_sectors && bitmap_offset > 0)
1351 bb_offset = bitmap_offset + bm_sectors;
1352 while (bb_offset < (long)sb_offset + 8 + 32*2 &&
1353 bb_offset + 8+8 <= (long)data_offset)
1354 /* too close to bitmap, and room to grow */
1355 bb_offset += 8;
1356 if (bb_offset + 8 <= (long)data_offset) {
1357 sb->bblog_size = __cpu_to_le16(8);
1358 sb->bblog_offset = __cpu_to_le32(bb_offset);
1359 }
1360 } else {
1361 /* 1.0 - Put bbl just before super block */
1362 if (bm_sectors && bitmap_offset < 0)
1363 space = -bitmap_offset - bm_sectors;
1364 else
1365 space = sb_offset - data_offset -
1366 __le64_to_cpu(sb->data_size);
1367 if (space >= 8) {
1368 sb->bblog_size = __cpu_to_le16(8);
1369 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1370 }
1371 }
1372 } else if (strcmp(update, "no-bbl") == 0) {
1373 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
1374 pr_err("Cannot remove active bbl from %s\n",devname);
1375 else {
1376 sb->bblog_size = 0;
1377 sb->bblog_shift = 0;
1378 sb->bblog_offset = 0;
1379 }
1380 } else if (strcmp(update, "force-no-bbl") == 0) {
1381 sb->feature_map &= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1382 sb->bblog_size = 0;
1383 sb->bblog_shift = 0;
1384 sb->bblog_offset = 0;
1385 } else if (strcmp(update, "ppl") == 0) {
1386 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1387 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1388 unsigned long long data_size = __le64_to_cpu(sb->data_size);
1389 long bb_offset = __le32_to_cpu(sb->bblog_offset);
1390 int space;
1391 int offset;
1392
1393 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1394 pr_err("Cannot add PPL to array with bitmap\n");
1395 return -2;
1396 }
1397
1398 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_JOURNAL)) {
1399 pr_err("Cannot add PPL to array with journal\n");
1400 return -2;
1401 }
1402
1403 if (sb_offset < data_offset) {
1404 if (bb_offset)
1405 space = bb_offset - 8;
1406 else
1407 space = data_offset - sb_offset - 8;
1408 offset = 8;
1409 } else {
1410 offset = -(sb_offset - data_offset - data_size);
1411 if (offset < INT16_MIN)
1412 offset = INT16_MIN;
1413 space = -(offset - bb_offset);
1414 }
1415
1416 if (space < (PPL_HEADER_SIZE >> 9) + 8) {
1417 pr_err("Not enough space to add ppl\n");
1418 return -2;
1419 }
1420
1421 if (space >= (MULTIPLE_PPL_AREA_SIZE_SUPER1 >> 9)) {
1422 space = (MULTIPLE_PPL_AREA_SIZE_SUPER1 >> 9);
1423 } else {
1424 int optimal_space = choose_ppl_space(
1425 __le32_to_cpu(sb->chunksize));
1426 if (space > optimal_space)
1427 space = optimal_space;
1428 if (space > UINT16_MAX)
1429 space = UINT16_MAX;
1430 }
1431
1432 sb->ppl.offset = __cpu_to_le16(offset);
1433 sb->ppl.size = __cpu_to_le16(space);
1434 sb->feature_map |= __cpu_to_le32(MD_FEATURE_PPL);
1435 } else if (strcmp(update, "no-ppl") == 0) {
1436 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_PPL |
1437 MD_FEATURE_MUTLIPLE_PPLS);
1438 } else if (strcmp(update, "name") == 0) {
1439 if (info->name[0] == 0)
1440 sprintf(info->name, "%d", info->array.md_minor);
1441 memset(sb->set_name, 0, sizeof(sb->set_name));
1442 if (homehost &&
1443 strchr(info->name, ':') == NULL &&
1444 strlen(homehost)+1+strlen(info->name) < 32) {
1445 strcpy(sb->set_name, homehost);
1446 strcat(sb->set_name, ":");
1447 strcat(sb->set_name, info->name);
1448 } else
1449 strncpy(sb->set_name, info->name, sizeof(sb->set_name));
1450 } else if (strcmp(update, "devicesize") == 0 &&
1451 __le64_to_cpu(sb->super_offset) <
1452 __le64_to_cpu(sb->data_offset)) {
1453 /* set data_size to device size less data_offset */
1454 struct misc_dev_info *misc = (struct misc_dev_info*)
1455 (st->sb + MAX_SB_SIZE + BM_SUPER_SIZE);
1456 sb->data_size = __cpu_to_le64(
1457 misc->device_size - __le64_to_cpu(sb->data_offset));
1458 } else if (strncmp(update, "revert-reshape", 14) == 0) {
1459 rv = -2;
1460 if (!(sb->feature_map &
1461 __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)))
1462 pr_err("No active reshape to revert on %s\n",
1463 devname);
1464 else {
1465 __u32 temp;
1466 unsigned long long reshape_sectors;
1467 long reshape_chunk;
1468 rv = 0;
1469 /* If the reshape hasn't started, just stop it.
1470 * It is conceivable that a stripe was modified but
1471 * the metadata not updated. In that case the backup
1472 * should have been used to get passed the critical stage.
1473 * If that couldn't happen, the "-nobackup" version
1474 * will be used.
1475 */
1476 if (strcmp(update, "revert-reshape-nobackup") == 0 &&
1477 sb->reshape_position == 0 &&
1478 (__le32_to_cpu(sb->delta_disks) > 0 ||
1479 (__le32_to_cpu(sb->delta_disks) == 0 &&
1480 !(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS))))) {
1481 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1482 sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1483 __le32_to_cpu(sb->delta_disks));
1484 sb->delta_disks = 0;
1485 goto done;
1486 }
1487 /* reshape_position is a little messy.
1488 * Its value must be a multiple of the larger
1489 * chunk size, and of the "after" data disks.
1490 * So when reverting we need to change it to
1491 * be a multiple of the new "after" data disks,
1492 * which is the old "before".
1493 * If it isn't already a multiple of 'before',
1494 * the only thing we could do would be
1495 * copy some block around on the disks, which
1496 * is easy to get wrong.
1497 * So we reject a revert-reshape unless the
1498 * alignment is good.
1499 */
1500 if (__le32_to_cpu(sb->level) >= 4 &&
1501 __le32_to_cpu(sb->level) <= 6) {
1502 reshape_sectors =
1503 __le64_to_cpu(sb->reshape_position);
1504 reshape_chunk = __le32_to_cpu(sb->new_chunk);
1505 reshape_chunk *= __le32_to_cpu(sb->raid_disks) -
1506 __le32_to_cpu(sb->delta_disks) -
1507 (__le32_to_cpu(sb->level)==6 ? 2 : 1);
1508 if (reshape_sectors % reshape_chunk) {
1509 pr_err("Reshape position is not suitably aligned.\n");
1510 pr_err("Try normal assembly and stop again\n");
1511 return -2;
1512 }
1513 }
1514 sb->raid_disks =
1515 __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1516 __le32_to_cpu(sb->delta_disks));
1517 if (sb->delta_disks == 0)
1518 sb->feature_map ^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1519 else
1520 sb->delta_disks = __cpu_to_le32(-__le32_to_cpu(sb->delta_disks));
1521
1522 temp = sb->new_layout;
1523 sb->new_layout = sb->layout;
1524 sb->layout = temp;
1525
1526 temp = sb->new_chunk;
1527 sb->new_chunk = sb->chunksize;
1528 sb->chunksize = temp;
1529
1530 if (sb->feature_map &
1531 __cpu_to_le32(MD_FEATURE_NEW_OFFSET)) {
1532 long offset_delta =
1533 (int32_t)__le32_to_cpu(sb->new_offset);
1534 sb->data_offset = __cpu_to_le64(__le64_to_cpu(sb->data_offset) + offset_delta);
1535 sb->new_offset = __cpu_to_le32(-offset_delta);
1536 sb->data_size = __cpu_to_le64(__le64_to_cpu(sb->data_size) - offset_delta);
1537 }
1538 done:;
1539 }
1540 } else if (strcmp(update, "_reshape_progress") == 0)
1541 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1542 else if (strcmp(update, "writemostly") == 0)
1543 sb->devflags |= WriteMostly1;
1544 else if (strcmp(update, "readwrite") == 0)
1545 sb->devflags &= ~WriteMostly1;
1546 else if (strcmp(update, "failfast") == 0)
1547 sb->devflags |= FailFast1;
1548 else if (strcmp(update, "nofailfast") == 0)
1549 sb->devflags &= ~FailFast1;
1550 else
1551 rv = -1;
1552
1553 sb->sb_csum = calc_sb_1_csum(sb);
1554 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1555 cluster_release_dlmlock(lockid);
1556
1557 return rv;
1558 }
1559
1560 static int init_super1(struct supertype *st, mdu_array_info_t *info,
1561 struct shape *s, char *name, char *homehost,
1562 int *uuid, unsigned long long data_offset)
1563 {
1564 struct mdp_superblock_1 *sb;
1565 int spares;
1566 char defname[10];
1567 int sbsize;
1568
1569 if (posix_memalign((void**)&sb, 4096, SUPER1_SIZE) != 0) {
1570 pr_err("could not allocate superblock\n");
1571 return 0;
1572 }
1573 memset(sb, 0, SUPER1_SIZE);
1574
1575 st->sb = sb;
1576 if (info == NULL) {
1577 /* zeroing superblock */
1578 return 0;
1579 }
1580
1581 spares = info->working_disks - info->active_disks;
1582 if (info->raid_disks + spares > MAX_DEVS) {
1583 pr_err("too many devices requested: %d+%d > %d\n",
1584 info->raid_disks , spares, MAX_DEVS);
1585 return 0;
1586 }
1587
1588 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
1589 sb->major_version = __cpu_to_le32(1);
1590 sb->feature_map = 0;
1591 sb->pad0 = 0;
1592
1593 if (uuid)
1594 copy_uuid(sb->set_uuid, uuid, super1.swapuuid);
1595 else
1596 random_uuid(sb->set_uuid);;
1597
1598 if (name == NULL || *name == 0) {
1599 sprintf(defname, "%d", info->md_minor);
1600 name = defname;
1601 }
1602 if (homehost &&
1603 strchr(name, ':')== NULL &&
1604 strlen(homehost)+1+strlen(name) < 32) {
1605 strcpy(sb->set_name, homehost);
1606 strcat(sb->set_name, ":");
1607 strcat(sb->set_name, name);
1608 } else
1609 strncpy(sb->set_name, name, sizeof(sb->set_name));
1610
1611 sb->ctime = __cpu_to_le64((unsigned long long)time(0));
1612 sb->level = __cpu_to_le32(info->level);
1613 sb->layout = __cpu_to_le32(info->layout);
1614 sb->size = __cpu_to_le64(s->size*2ULL);
1615 sb->chunksize = __cpu_to_le32(info->chunk_size>>9);
1616 sb->raid_disks = __cpu_to_le32(info->raid_disks);
1617
1618 sb->data_offset = __cpu_to_le64(data_offset);
1619 sb->data_size = __cpu_to_le64(0);
1620 sb->super_offset = __cpu_to_le64(0);
1621 sb->recovery_offset = __cpu_to_le64(0);
1622
1623 sb->utime = sb->ctime;
1624 sb->events = __cpu_to_le64(1);
1625 if (info->state & (1<<MD_SB_CLEAN))
1626 sb->resync_offset = MaxSector;
1627 else
1628 sb->resync_offset = 0;
1629 sbsize = sizeof(struct mdp_superblock_1) +
1630 2 * (info->raid_disks + spares);
1631 sbsize = ROUND_UP(sbsize, 512);
1632 sb->max_dev =
1633 __cpu_to_le32((sbsize - sizeof(struct mdp_superblock_1)) / 2);
1634
1635 memset(sb->dev_roles, 0xff,
1636 MAX_SB_SIZE - sizeof(struct mdp_superblock_1));
1637
1638 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
1639 sb->feature_map |= __cpu_to_le32(MD_FEATURE_PPL);
1640
1641 return 1;
1642 }
1643
1644 struct devinfo {
1645 int fd;
1646 char *devname;
1647 long long data_offset;
1648 mdu_disk_info_t disk;
1649 struct devinfo *next;
1650 };
1651
1652 /* Add a device to the superblock being created */
1653 static int add_to_super1(struct supertype *st, mdu_disk_info_t *dk,
1654 int fd, char *devname, unsigned long long data_offset)
1655 {
1656 struct mdp_superblock_1 *sb = st->sb;
1657 __u16 *rp = sb->dev_roles + dk->number;
1658 struct devinfo *di, **dip;
1659 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1660 int rv, lockid;
1661 int dk_state;
1662
1663 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1664 rv = cluster_get_dlmlock(&lockid);
1665 if (rv) {
1666 pr_err("Cannot get dlmlock in %s return %d\n",
1667 __func__, rv);
1668 cluster_release_dlmlock(lockid);
1669 return rv;
1670 }
1671 }
1672
1673 dk_state = dk->state & ~(1<<MD_DISK_FAILFAST);
1674 if ((dk_state & (1<<MD_DISK_ACTIVE)) &&
1675 (dk_state & (1<<MD_DISK_SYNC)))/* active, sync */
1676 *rp = __cpu_to_le16(dk->raid_disk);
1677 else if (dk_state & (1<<MD_DISK_JOURNAL))
1678 *rp = MD_DISK_ROLE_JOURNAL;
1679 else if ((dk_state & ~(1<<MD_DISK_ACTIVE)) == 0)
1680 /* active or idle -> spare */
1681 *rp = MD_DISK_ROLE_SPARE;
1682 else
1683 *rp = MD_DISK_ROLE_FAULTY;
1684
1685 if (dk->number >= (int)__le32_to_cpu(sb->max_dev) &&
1686 __le32_to_cpu(sb->max_dev) < MAX_DEVS)
1687 sb->max_dev = __cpu_to_le32(dk->number+1);
1688
1689 sb->dev_number = __cpu_to_le32(dk->number);
1690 sb->devflags = 0; /* don't copy another disks flags */
1691 sb->sb_csum = calc_sb_1_csum(sb);
1692
1693 dip = (struct devinfo **)&st->info;
1694 while (*dip)
1695 dip = &(*dip)->next;
1696 di = xmalloc(sizeof(struct devinfo));
1697 di->fd = fd;
1698 di->devname = devname;
1699 di->disk = *dk;
1700 di->data_offset = data_offset;
1701 di->next = NULL;
1702 *dip = di;
1703
1704 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1705 cluster_release_dlmlock(lockid);
1706
1707 return 0;
1708 }
1709
1710 static int locate_bitmap1(struct supertype *st, int fd, int node_num);
1711
1712 static int store_super1(struct supertype *st, int fd)
1713 {
1714 struct mdp_superblock_1 *sb = st->sb;
1715 unsigned long long sb_offset;
1716 struct align_fd afd;
1717 int sbsize;
1718 unsigned long long dsize;
1719 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1720 int rv, lockid;
1721
1722 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1723 rv = cluster_get_dlmlock(&lockid);
1724 if (rv) {
1725 pr_err("Cannot get dlmlock in %s return %d\n",
1726 __func__, rv);
1727 cluster_release_dlmlock(lockid);
1728 return rv;
1729 }
1730 }
1731
1732 if (!get_dev_size(fd, NULL, &dsize))
1733 return 1;
1734
1735 dsize >>= 9;
1736
1737 if (dsize < 24)
1738 return 2;
1739
1740 init_afd(&afd, fd);
1741
1742 /*
1743 * Calculate the position of the superblock.
1744 * It is always aligned to a 4K boundary and
1745 * depending on minor_version, it can be:
1746 * 0: At least 8K, but less than 12K, from end of device
1747 * 1: At start of device
1748 * 2: 4K from start of device.
1749 */
1750 switch(st->minor_version) {
1751 case 0:
1752 sb_offset = dsize;
1753 sb_offset -= 8*2;
1754 sb_offset &= ~(4*2-1);
1755 break;
1756 case 1:
1757 sb_offset = 0;
1758 break;
1759 case 2:
1760 sb_offset = 4*2;
1761 break;
1762 default:
1763 return -EINVAL;
1764 }
1765
1766 if (sb_offset != __le64_to_cpu(sb->super_offset) &&
1767 0 != __le64_to_cpu(sb->super_offset)
1768 ) {
1769 pr_err("internal error - sb_offset is wrong\n");
1770 abort();
1771 }
1772
1773 if (lseek64(fd, sb_offset << 9, 0)< 0LL)
1774 return 3;
1775
1776 sbsize = ROUND_UP(sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev), 512);
1777
1778 if (awrite(&afd, sb, sbsize) != sbsize)
1779 return 4;
1780
1781 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1782 struct bitmap_super_s *bm = (struct bitmap_super_s*)
1783 (((char*)sb)+MAX_SB_SIZE);
1784 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC) {
1785 locate_bitmap1(st, fd, 0);
1786 if (awrite(&afd, bm, sizeof(*bm)) != sizeof(*bm))
1787 return 5;
1788 }
1789 }
1790 fsync(fd);
1791 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1792 cluster_release_dlmlock(lockid);
1793
1794 return 0;
1795 }
1796
1797 static int load_super1(struct supertype *st, int fd, char *devname);
1798
1799 static unsigned long choose_bm_space(unsigned long devsize)
1800 {
1801 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1802 * if bigger than 200Gig, save 128k
1803 * NOTE: result must be multiple of 4K else bad things happen
1804 * on 4K-sector devices.
1805 */
1806 if (devsize < 64*2)
1807 return 0;
1808 if (devsize - 64*2 >= 200*1024*1024*2)
1809 return 128*2;
1810 if (devsize - 4*2 > 8*1024*1024*2)
1811 return 64*2;
1812 return 4*2;
1813 }
1814
1815 static void free_super1(struct supertype *st);
1816
1817 __u32 crc32c_le(__u32 crc, unsigned char const *p, size_t len);
1818
1819 static int write_init_ppl1(struct supertype *st, struct mdinfo *info, int fd)
1820 {
1821 struct mdp_superblock_1 *sb = st->sb;
1822 void *buf;
1823 struct ppl_header *ppl_hdr;
1824 int ret;
1825
1826 ret = posix_memalign(&buf, 4096, PPL_HEADER_SIZE);
1827 if (ret) {
1828 pr_err("Failed to allocate PPL header buffer\n");
1829 return ret;
1830 }
1831
1832 memset(buf, 0, PPL_HEADER_SIZE);
1833 ppl_hdr = buf;
1834 memset(ppl_hdr->reserved, 0xff, PPL_HDR_RESERVED);
1835 ppl_hdr->signature = __cpu_to_le32(~crc32c_le(~0, sb->set_uuid,
1836 sizeof(sb->set_uuid)));
1837 ppl_hdr->checksum = __cpu_to_le32(~crc32c_le(~0, buf, PPL_HEADER_SIZE));
1838
1839 if (lseek64(fd, info->ppl_sector * 512, SEEK_SET) < 0) {
1840 ret = errno;
1841 perror("Failed to seek to PPL header location");
1842 }
1843
1844 if (!ret && write(fd, buf, PPL_HEADER_SIZE) != PPL_HEADER_SIZE) {
1845 ret = errno;
1846 perror("Write PPL header failed");
1847 }
1848
1849 if (!ret)
1850 fsync(fd);
1851
1852 free(buf);
1853 return ret;
1854 }
1855
1856 #define META_BLOCK_SIZE 4096
1857
1858 static int write_empty_r5l_meta_block(struct supertype *st, int fd)
1859 {
1860 struct r5l_meta_block *mb;
1861 struct mdp_superblock_1 *sb = st->sb;
1862 struct align_fd afd;
1863 __u32 crc;
1864
1865 init_afd(&afd, fd);
1866
1867 if (posix_memalign((void**)&mb, 4096, META_BLOCK_SIZE) != 0) {
1868 pr_err("Could not allocate memory for the meta block.\n");
1869 return 1;
1870 }
1871
1872 memset(mb, 0, META_BLOCK_SIZE);
1873
1874 mb->magic = __cpu_to_le32(R5LOG_MAGIC);
1875 mb->version = R5LOG_VERSION;
1876 mb->meta_size = __cpu_to_le32(sizeof(struct r5l_meta_block));
1877 mb->seq = __cpu_to_le64(random32());
1878 mb->position = __cpu_to_le64(0);
1879
1880 crc = crc32c_le(0xffffffff, sb->set_uuid, sizeof(sb->set_uuid));
1881 crc = crc32c_le(crc, (void *)mb, META_BLOCK_SIZE);
1882 mb->checksum = crc;
1883
1884 if (lseek64(fd, __le64_to_cpu(sb->data_offset) * 512, 0) < 0LL) {
1885 pr_err("cannot seek to offset of the meta block\n");
1886 goto fail_to_write;
1887 }
1888
1889 if (awrite(&afd, mb, META_BLOCK_SIZE) != META_BLOCK_SIZE) {
1890 pr_err("failed to store write the meta block \n");
1891 goto fail_to_write;
1892 }
1893 fsync(fd);
1894
1895 free(mb);
1896 return 0;
1897
1898 fail_to_write:
1899 free(mb);
1900 return 1;
1901 }
1902
1903 static int write_init_super1(struct supertype *st)
1904 {
1905 struct mdp_superblock_1 *sb = st->sb;
1906 struct supertype *refst;
1907 int rv = 0;
1908 unsigned long long bm_space;
1909 struct devinfo *di;
1910 unsigned long long dsize, array_size;
1911 unsigned long long sb_offset;
1912 unsigned long long data_offset;
1913 long bm_offset;
1914
1915 for (di = st->info; di; di = di->next) {
1916 if (di->disk.state & (1 << MD_DISK_JOURNAL))
1917 sb->feature_map |= __cpu_to_le32(MD_FEATURE_JOURNAL);
1918 }
1919
1920 for (di = st->info; di; di = di->next) {
1921 if (di->disk.state & (1 << MD_DISK_FAULTY))
1922 continue;
1923 if (di->fd < 0)
1924 continue;
1925
1926 while (Kill(di->devname, NULL, 0, -1, 1) == 0)
1927 ;
1928
1929 sb->dev_number = __cpu_to_le32(di->disk.number);
1930 if (di->disk.state & (1<<MD_DISK_WRITEMOSTLY))
1931 sb->devflags |= WriteMostly1;
1932 else
1933 sb->devflags &= ~WriteMostly1;
1934 if (di->disk.state & (1<<MD_DISK_FAILFAST))
1935 sb->devflags |= FailFast1;
1936 else
1937 sb->devflags &= ~FailFast1;
1938
1939 random_uuid(sb->device_uuid);
1940
1941 if (!(di->disk.state & (1<<MD_DISK_JOURNAL)))
1942 sb->events = 0;
1943
1944 refst = dup_super(st);
1945 if (load_super1(refst, di->fd, NULL)==0) {
1946 struct mdp_superblock_1 *refsb = refst->sb;
1947
1948 memcpy(sb->device_uuid, refsb->device_uuid, 16);
1949 if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) {
1950 /* same array, so preserve events and
1951 * dev_number */
1952 sb->events = refsb->events;
1953 /* bugs in 2.6.17 and earlier mean the
1954 * dev_number chosen in Manage must be preserved
1955 */
1956 if (get_linux_version() >= 2006018)
1957 sb->dev_number = refsb->dev_number;
1958 }
1959 free_super1(refst);
1960 }
1961 free(refst);
1962
1963 if (!get_dev_size(di->fd, NULL, &dsize)) {
1964 rv = 1;
1965 goto error_out;
1966 }
1967 dsize >>= 9;
1968
1969 if (dsize < 24) {
1970 close(di->fd);
1971 rv = 2;
1972 goto error_out;
1973 }
1974
1975 /*
1976 * Calculate the position of the superblock.
1977 * It is always aligned to a 4K boundary and
1978 * depending on minor_version, it can be:
1979 * 0: At least 8K, but less than 12K, from end of device
1980 * 1: At start of device
1981 * 2: 4K from start of device.
1982 * data_offset has already been set.
1983 */
1984 array_size = __le64_to_cpu(sb->size);
1985
1986 /* work out how much space we left for a bitmap */
1987 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1988 bitmap_super_t *bms = (bitmap_super_t *)
1989 (((char *)sb) + MAX_SB_SIZE);
1990 bm_space = calc_bitmap_size(bms, 4096) >> 9;
1991 bm_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1992 } else if (md_feature_any_ppl_on(sb->feature_map)) {
1993 bm_space = MULTIPLE_PPL_AREA_SIZE_SUPER1 >> 9;
1994 if (st->minor_version == 0)
1995 bm_offset = -bm_space - 8;
1996 else
1997 bm_offset = 8;
1998 sb->ppl.offset = __cpu_to_le16(bm_offset);
1999 sb->ppl.size = __cpu_to_le16(bm_space);
2000 } else {
2001 bm_space = choose_bm_space(array_size);
2002 bm_offset = 8;
2003 }
2004
2005 data_offset = di->data_offset;
2006 if (data_offset == INVALID_SECTORS)
2007 data_offset = st->data_offset;
2008 switch(st->minor_version) {
2009 case 0:
2010 /* Add 8 sectors for bad block log */
2011 bm_space += 8;
2012 if (data_offset == INVALID_SECTORS)
2013 data_offset = 0;
2014 sb_offset = dsize;
2015 sb_offset -= 8*2;
2016 sb_offset &= ~(4*2-1);
2017 sb->data_offset = __cpu_to_le64(data_offset);
2018 sb->super_offset = __cpu_to_le64(sb_offset);
2019 if (sb_offset < array_size + bm_space)
2020 bm_space = sb_offset - array_size;
2021 sb->data_size = __cpu_to_le64(sb_offset - bm_space);
2022 if (bm_space >= 8) {
2023 sb->bblog_size = __cpu_to_le16(8);
2024 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
2025 }
2026 break;
2027 case 1:
2028 case 2:
2029 sb_offset = st->minor_version == 2 ? 8 : 0;
2030 sb->super_offset = __cpu_to_le64(sb_offset);
2031 if (data_offset == INVALID_SECTORS)
2032 data_offset = sb_offset + 16;
2033
2034 sb->data_offset = __cpu_to_le64(data_offset);
2035 sb->data_size = __cpu_to_le64(dsize - data_offset);
2036 if (data_offset >= sb_offset+bm_offset+bm_space+8) {
2037 sb->bblog_size = __cpu_to_le16(8);
2038 sb->bblog_offset = __cpu_to_le32(bm_offset +
2039 bm_space);
2040 } else if (data_offset >= sb_offset + 16) {
2041 sb->bblog_size = __cpu_to_le16(8);
2042 /* '8' sectors for the bblog, and 'sb_offset'
2043 * because we want offset from superblock, not
2044 * start of device.
2045 */
2046 sb->bblog_offset = __cpu_to_le32(data_offset -
2047 8 - sb_offset);
2048 }
2049 break;
2050 default:
2051 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
2052 st->minor_version, di->devname);
2053 rv = -EINVAL;
2054 goto out;
2055 }
2056 /*
2057 * Disable badblock log on clusters, or when
2058 * explicitly requested
2059 */
2060 if (st->nodes > 0 || conf_get_create_info()->bblist == 0) {
2061 sb->bblog_size = 0;
2062 sb->bblog_offset = 0;
2063 }
2064
2065 sb->sb_csum = calc_sb_1_csum(sb);
2066 rv = store_super1(st, di->fd);
2067
2068 if (rv == 0 && (di->disk.state & (1 << MD_DISK_JOURNAL))) {
2069 rv = write_empty_r5l_meta_block(st, di->fd);
2070 if (rv)
2071 goto error_out;
2072 }
2073
2074 if (rv == 0 &&
2075 (__le32_to_cpu(sb->feature_map) &
2076 MD_FEATURE_BITMAP_OFFSET)) {
2077 rv = st->ss->write_bitmap(st, di->fd, NodeNumUpdate);
2078 } else if (rv == 0 &&
2079 md_feature_any_ppl_on(sb->feature_map)) {
2080 struct mdinfo info;
2081
2082 st->ss->getinfo_super(st, &info, NULL);
2083 rv = st->ss->write_init_ppl(st, &info, di->fd);
2084 }
2085
2086 close(di->fd);
2087 di->fd = -1;
2088 if (rv)
2089 goto error_out;
2090 }
2091 error_out:
2092 if (rv)
2093 pr_err("Failed to write metadata to %s\n", di->devname);
2094 out:
2095 return rv;
2096 }
2097
2098 static int compare_super1(struct supertype *st, struct supertype *tst)
2099 {
2100 /*
2101 * return:
2102 * 0 same, or first was empty, and second was copied
2103 * 1 second had wrong number
2104 * 2 wrong uuid
2105 * 3 wrong other info
2106 */
2107 struct mdp_superblock_1 *first = st->sb;
2108 struct mdp_superblock_1 *second = tst->sb;
2109
2110 if (second->magic != __cpu_to_le32(MD_SB_MAGIC))
2111 return 1;
2112 if (second->major_version != __cpu_to_le32(1))
2113 return 1;
2114
2115 if (!first) {
2116 if (posix_memalign((void**)&first, 4096, SUPER1_SIZE) != 0) {
2117 pr_err("could not allocate superblock\n");
2118 return 1;
2119 }
2120 memcpy(first, second, SUPER1_SIZE);
2121 st->sb = first;
2122 return 0;
2123 }
2124 if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0)
2125 return 2;
2126
2127 if (first->ctime != second->ctime ||
2128 first->level != second->level ||
2129 first->layout != second->layout ||
2130 first->size != second->size ||
2131 first->chunksize != second->chunksize ||
2132 first->raid_disks != second->raid_disks)
2133 return 3;
2134 return 0;
2135 }
2136
2137 static int load_super1(struct supertype *st, int fd, char *devname)
2138 {
2139 unsigned long long dsize;
2140 unsigned long long sb_offset;
2141 struct mdp_superblock_1 *super;
2142 int uuid[4];
2143 struct bitmap_super_s *bsb;
2144 struct misc_dev_info *misc;
2145 struct align_fd afd;
2146
2147 free_super1(st);
2148
2149 init_afd(&afd, fd);
2150
2151 if (st->ss == NULL || st->minor_version == -1) {
2152 int bestvers = -1;
2153 struct supertype tst;
2154 __u64 bestctime = 0;
2155 /* guess... choose latest ctime */
2156 memset(&tst, 0, sizeof(tst));
2157 tst.ss = &super1;
2158 for (tst.minor_version = 0; tst.minor_version <= 2;
2159 tst.minor_version++) {
2160 switch(load_super1(&tst, fd, devname)) {
2161 case 0: super = tst.sb;
2162 if (bestvers == -1 ||
2163 bestctime < __le64_to_cpu(super->ctime)) {
2164 bestvers = tst.minor_version;
2165 bestctime = __le64_to_cpu(super->ctime);
2166 }
2167 free(super);
2168 tst.sb = NULL;
2169 break;
2170 case 1: return 1; /*bad device */
2171 case 2: break; /* bad, try next */
2172 }
2173 }
2174 if (bestvers != -1) {
2175 int rv;
2176 tst.minor_version = bestvers;
2177 tst.ss = &super1;
2178 tst.max_devs = MAX_DEVS;
2179 rv = load_super1(&tst, fd, devname);
2180 if (rv == 0)
2181 *st = tst;
2182 return rv;
2183 }
2184 return 2;
2185 }
2186 if (!get_dev_size(fd, devname, &dsize))
2187 return 1;
2188 dsize >>= 9;
2189
2190 if (dsize < 24) {
2191 if (devname)
2192 pr_err("%s is too small for md: size is %llu sectors.\n",
2193 devname, dsize);
2194 return 1;
2195 }
2196
2197 /*
2198 * Calculate the position of the superblock.
2199 * It is always aligned to a 4K boundary and
2200 * depending on minor_version, it can be:
2201 * 0: At least 8K, but less than 12K, from end of device
2202 * 1: At start of device
2203 * 2: 4K from start of device.
2204 */
2205 switch(st->minor_version) {
2206 case 0:
2207 sb_offset = dsize;
2208 sb_offset -= 8*2;
2209 sb_offset &= ~(4*2-1);
2210 break;
2211 case 1:
2212 sb_offset = 0;
2213 break;
2214 case 2:
2215 sb_offset = 4*2;
2216 break;
2217 default:
2218 return -EINVAL;
2219 }
2220
2221 if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
2222 if (devname)
2223 pr_err("Cannot seek to superblock on %s: %s\n",
2224 devname, strerror(errno));
2225 return 1;
2226 }
2227
2228 if (posix_memalign((void**)&super, 4096, SUPER1_SIZE) != 0) {
2229 pr_err("could not allocate superblock\n");
2230 return 1;
2231 }
2232
2233 memset(super, 0, SUPER1_SIZE);
2234
2235 if (aread(&afd, super, MAX_SB_SIZE) != MAX_SB_SIZE) {
2236 if (devname)
2237 pr_err("Cannot read superblock on %s\n",
2238 devname);
2239 free(super);
2240 return 1;
2241 }
2242
2243 if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) {
2244 if (devname)
2245 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2246 devname, MD_SB_MAGIC,
2247 __le32_to_cpu(super->magic));
2248 free(super);
2249 return 2;
2250 }
2251
2252 if (__le32_to_cpu(super->major_version) != 1) {
2253 if (devname)
2254 pr_err("Cannot interpret superblock on %s - version is %d\n",
2255 devname, __le32_to_cpu(super->major_version));
2256 free(super);
2257 return 2;
2258 }
2259 if (__le64_to_cpu(super->super_offset) != sb_offset) {
2260 if (devname)
2261 pr_err("No superblock found on %s (super_offset is wrong)\n",
2262 devname);
2263 free(super);
2264 return 2;
2265 }
2266 st->sb = super;
2267
2268 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
2269
2270 misc = (struct misc_dev_info*)
2271 (((char*)super)+MAX_SB_SIZE+BM_SUPER_SIZE);
2272 misc->device_size = dsize;
2273 if (st->data_offset == INVALID_SECTORS)
2274 st->data_offset = __le64_to_cpu(super->data_offset);
2275
2276 /* Now check on the bitmap superblock */
2277 if ((__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) == 0)
2278 return 0;
2279 /* Read the bitmap superblock and make sure it looks
2280 * valid. If it doesn't clear the bit. An --assemble --force
2281 * should get that written out.
2282 */
2283 locate_bitmap1(st, fd, 0);
2284 if (aread(&afd, bsb, 512) != 512)
2285 goto no_bitmap;
2286
2287 uuid_from_super1(st, uuid);
2288 if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
2289 memcmp(bsb->uuid, uuid, 16) != 0)
2290 goto no_bitmap;
2291 return 0;
2292
2293 no_bitmap:
2294 super->feature_map = __cpu_to_le32(__le32_to_cpu(super->feature_map)
2295 & ~MD_FEATURE_BITMAP_OFFSET);
2296 return 0;
2297 }
2298
2299 static struct supertype *match_metadata_desc1(char *arg)
2300 {
2301 struct supertype *st = xcalloc(1, sizeof(*st));
2302
2303 st->container_devnm[0] = 0;
2304 st->ss = &super1;
2305 st->max_devs = MAX_DEVS;
2306 st->sb = NULL;
2307 st->data_offset = INVALID_SECTORS;
2308 /* leading zeros can be safely ignored. --detail generates them. */
2309 while (*arg == '0')
2310 arg++;
2311 if (strcmp(arg, "1.0") == 0 || strcmp(arg, "1.00") == 0) {
2312 st->minor_version = 0;
2313 return st;
2314 }
2315 if (strcmp(arg, "1.1") == 0 || strcmp(arg, "1.01") == 0
2316 ) {
2317 st->minor_version = 1;
2318 return st;
2319 }
2320 if (strcmp(arg, "1.2") == 0 ||
2321 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2322 strcmp(arg, "default") == 0 ||
2323 #endif /* DEFAULT_OLD_METADATA */
2324 strcmp(arg, "1.02") == 0) {
2325 st->minor_version = 2;
2326 return st;
2327 }
2328 if (strcmp(arg, "1") == 0 || strcmp(arg, "default") == 0) {
2329 st->minor_version = -1;
2330 return st;
2331 }
2332
2333 free(st);
2334 return NULL;
2335 }
2336
2337 /* find available size on device with this devsize, using
2338 * superblock type st, and reserving 'reserve' sectors for
2339 * a possible bitmap
2340 */
2341 static __u64 avail_size1(struct supertype *st, __u64 devsize,
2342 unsigned long long data_offset)
2343 {
2344 struct mdp_superblock_1 *super = st->sb;
2345 int bmspace = 0;
2346 int bbspace = 0;
2347 if (devsize < 24)
2348 return 0;
2349
2350 if (__le32_to_cpu(super->feature_map) & MD_FEATURE_BITMAP_OFFSET) {
2351 /* hot-add. allow for actual size of bitmap */
2352 struct bitmap_super_s *bsb;
2353 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
2354 bmspace = calc_bitmap_size(bsb, 4096) >> 9;
2355 } else if (md_feature_any_ppl_on(super->feature_map)) {
2356 bmspace = __le16_to_cpu(super->ppl.size);
2357 }
2358
2359 /* Allow space for bad block log */
2360 if (super->bblog_size)
2361 bbspace = __le16_to_cpu(super->bblog_size);
2362
2363 if (st->minor_version < 0)
2364 /* not specified, so time to set default */
2365 st->minor_version = 2;
2366
2367 if (data_offset == INVALID_SECTORS)
2368 data_offset = st->data_offset;
2369
2370 if (data_offset != INVALID_SECTORS)
2371 switch(st->minor_version) {
2372 case 0:
2373 return devsize - data_offset - 8*2 - bbspace;
2374 case 1:
2375 case 2:
2376 return devsize - data_offset;
2377 default:
2378 return 0;
2379 }
2380
2381 devsize -= bmspace;
2382
2383 switch(st->minor_version) {
2384 case 0:
2385 /* at end */
2386 return ((devsize - 8*2 - bbspace ) & ~(4*2-1));
2387 case 1:
2388 /* at start, 4K for superblock and possible bitmap */
2389 return devsize - 4*2 - bbspace;
2390 case 2:
2391 /* 4k from start, 4K for superblock and possible bitmap */
2392 return devsize - (4+4)*2 - bbspace;
2393 }
2394 return 0;
2395 }
2396
2397 static int
2398 add_internal_bitmap1(struct supertype *st,
2399 int *chunkp, int delay, int write_behind,
2400 unsigned long long size,
2401 int may_change, int major)
2402 {
2403 /*
2404 * If not may_change, then this is a 'Grow' without sysfs support for
2405 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2406 * If may_change, then this is create or a Grow with sysfs support,
2407 * and we can put the bitmap wherever we like.
2408 *
2409 * size is in sectors, chunk is in bytes !!!
2410 */
2411
2412 unsigned long long bits;
2413 unsigned long long max_bits;
2414 unsigned long long min_chunk;
2415 long offset;
2416 long bbl_offset, bbl_size;
2417 unsigned long long chunk = *chunkp;
2418 int room = 0;
2419 int creating = 0;
2420 int len;
2421 struct mdp_superblock_1 *sb = st->sb;
2422 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
2423 int uuid[4];
2424
2425 if (__le64_to_cpu(sb->data_size) == 0)
2426 /*
2427 * Must be creating the array, else data_size
2428 * would be non-zero
2429 */
2430 creating = 1;
2431 switch(st->minor_version) {
2432 case 0:
2433 /*
2434 * either 3K after the superblock (when hot-add),
2435 * or some amount of space before.
2436 */
2437 if (creating) {
2438 /*
2439 * We are creating array, so we *know* how much room has
2440 * been left.
2441 */
2442 offset = 0;
2443 bbl_size = 8;
2444 room =
2445 choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2446 } else {
2447 room = __le64_to_cpu(sb->super_offset)
2448 - __le64_to_cpu(sb->data_offset)
2449 - __le64_to_cpu(sb->data_size);
2450 bbl_size = __le16_to_cpu(sb->bblog_size);
2451 if (bbl_size < 8)
2452 bbl_size = 8;
2453 bbl_offset = (__s32)__le32_to_cpu(sb->bblog_offset);
2454 if (bbl_size < -bbl_offset)
2455 bbl_size = -bbl_offset;
2456
2457 if (!may_change ||
2458 (room < 3*2 && __le32_to_cpu(sb->max_dev) <= 384)) {
2459 room = 3*2;
2460 offset = 1*2;
2461 bbl_size = 0;
2462 } else {
2463 offset = 0; /* means movable offset */
2464 }
2465 }
2466 break;
2467 case 1:
2468 case 2: /* between superblock and data */
2469 if (creating) {
2470 offset = 4*2;
2471 bbl_size = 8;
2472 room =
2473 choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2474 } else {
2475 room = __le64_to_cpu(sb->data_offset)
2476 - __le64_to_cpu(sb->super_offset);
2477 bbl_size = __le16_to_cpu(sb->bblog_size);
2478 if (bbl_size)
2479 room =
2480 __le32_to_cpu(sb->bblog_offset) + bbl_size;
2481 else
2482 bbl_size = 8;
2483
2484 if (!may_change) {
2485 room -= 2; /* Leave 1K for superblock */
2486 offset = 2;
2487 bbl_size = 0;
2488 } else {
2489 room -= 4*2; /* leave 4K for superblock */
2490 offset = 4*2;
2491 }
2492 }
2493 break;
2494 default:
2495 return -ENOSPC;
2496 }
2497
2498 room -= bbl_size;
2499 if (chunk == UnSet && room > 128*2)
2500 /* Limit to 128K of bitmap when chunk size not requested */
2501 room = 128*2;
2502
2503 if (room <= 1)
2504 /* No room for a bitmap */
2505 return -ENOSPC;
2506
2507 max_bits = (room * 512 - sizeof(bitmap_super_t)) * 8;
2508
2509 min_chunk = 4096; /* sub-page chunks don't work yet.. */
2510 bits = (size*512)/min_chunk +1;
2511 while (bits > max_bits) {
2512 min_chunk *= 2;
2513 bits = (bits+1)/2;
2514 }
2515 if (chunk == UnSet) {
2516 /* For practical purpose, 64Meg is a good
2517 * default chunk size for internal bitmaps.
2518 */
2519 chunk = min_chunk;
2520 if (chunk < 64*1024*1024)
2521 chunk = 64*1024*1024;
2522 } else if (chunk < min_chunk)
2523 return -EINVAL; /* chunk size too small */
2524 if (chunk == 0) /* rounding problem */
2525 return -EINVAL;
2526
2527 if (offset == 0) {
2528 /* start bitmap on a 4K boundary with enough space for
2529 * the bitmap
2530 */
2531 bits = (size*512) / chunk + 1;
2532 room = ((bits+7)/8 + sizeof(bitmap_super_t) +4095)/4096;
2533 room *= 8; /* convert 4K blocks to sectors */
2534 offset = -room - bbl_size;
2535 }
2536
2537 sb->bitmap_offset = (int32_t)__cpu_to_le32(offset);
2538
2539 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map) |
2540 MD_FEATURE_BITMAP_OFFSET);
2541 memset(bms, 0, sizeof(*bms));
2542 bms->magic = __cpu_to_le32(BITMAP_MAGIC);
2543 bms->version = __cpu_to_le32(major);
2544 uuid_from_super1(st, uuid);
2545 memcpy(bms->uuid, uuid, 16);
2546 bms->chunksize = __cpu_to_le32(chunk);
2547 bms->daemon_sleep = __cpu_to_le32(delay);
2548 bms->sync_size = __cpu_to_le64(size);
2549 bms->write_behind = __cpu_to_le32(write_behind);
2550 bms->nodes = __cpu_to_le32(st->nodes);
2551 if (st->nodes)
2552 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map) |
2553 MD_FEATURE_BITMAP_VERSIONED);
2554 if (st->cluster_name) {
2555 len = sizeof(bms->cluster_name);
2556 strncpy((char *)bms->cluster_name, st->cluster_name, len);
2557 bms->cluster_name[len - 1] = '\0';
2558 }
2559
2560 *chunkp = chunk;
2561 return 0;
2562 }
2563
2564 static int locate_bitmap1(struct supertype *st, int fd, int node_num)
2565 {
2566 unsigned long long offset;
2567 struct mdp_superblock_1 *sb;
2568 int mustfree = 0;
2569 int ret;
2570
2571 if (!st->sb) {
2572 if (st->ss->load_super(st, fd, NULL))
2573 return -1; /* no error I hope... */
2574 mustfree = 1;
2575 }
2576 sb = st->sb;
2577
2578 if ((__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
2579 ret = 0;
2580 else
2581 ret = -1;
2582 offset = __le64_to_cpu(sb->super_offset);
2583 offset += (int32_t) __le32_to_cpu(sb->bitmap_offset) * (node_num + 1);
2584 if (mustfree)
2585 free(sb);
2586 lseek64(fd, offset<<9, 0);
2587 return ret;
2588 }
2589
2590 static int write_bitmap1(struct supertype *st, int fd, enum bitmap_update update)
2591 {
2592 struct mdp_superblock_1 *sb = st->sb;
2593 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
2594 int rv = 0;
2595 void *buf;
2596 int towrite, n, len;
2597 struct align_fd afd;
2598 unsigned int i = 0;
2599 unsigned long long total_bm_space, bm_space_per_node;
2600
2601 switch (update) {
2602 case NameUpdate:
2603 /* update cluster name */
2604 if (st->cluster_name) {
2605 len = sizeof(bms->cluster_name);
2606 memset((char *)bms->cluster_name, 0, len);
2607 strncpy((char *)bms->cluster_name,
2608 st->cluster_name, len);
2609 bms->cluster_name[len - 1] = '\0';
2610 }
2611 break;
2612 case NodeNumUpdate:
2613 /* cluster md only supports superblock 1.2 now */
2614 if (st->minor_version != 2 &&
2615 bms->version == BITMAP_MAJOR_CLUSTERED) {
2616 pr_err("Warning: cluster md only works with superblock 1.2\n");
2617 return -EINVAL;
2618 }
2619
2620 if (bms->version == BITMAP_MAJOR_CLUSTERED) {
2621 if (__cpu_to_le32(st->nodes) < bms->nodes) {
2622 /*
2623 * Since the nodes num is not increased, no
2624 * need to check the space enough or not,
2625 * just update bms->nodes
2626 */
2627 bms->nodes = __cpu_to_le32(st->nodes);
2628 break;
2629 }
2630 } else {
2631 /*
2632 * no need to change bms->nodes for other
2633 * bitmap types
2634 */
2635 if (st->nodes)
2636 pr_err("Warning: --nodes option is only suitable for clustered bitmap\n");
2637 break;
2638 }
2639
2640 /*
2641 * Each node has an independent bitmap, it is necessary to
2642 * calculate the space is enough or not, first get how many
2643 * bytes for the total bitmap
2644 */
2645 bm_space_per_node = calc_bitmap_size(bms, 4096);
2646
2647 total_bm_space = 512 * (__le64_to_cpu(sb->data_offset) -
2648 __le64_to_cpu(sb->super_offset));
2649 /* leave another 4k for superblock */
2650 total_bm_space = total_bm_space - 4096;
2651
2652 if (bm_space_per_node * st->nodes > total_bm_space) {
2653 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2654 total_bm_space / bm_space_per_node);
2655 return -ENOMEM;
2656 }
2657
2658 bms->nodes = __cpu_to_le32(st->nodes);
2659 break;
2660 case NoUpdate:
2661 default:
2662 break;
2663 }
2664
2665 init_afd(&afd, fd);
2666
2667 locate_bitmap1(st, fd, 0);
2668
2669 if (posix_memalign(&buf, 4096, 4096))
2670 return -ENOMEM;
2671
2672 do {
2673 /* Only the bitmap[0] should resync
2674 * whole device on initial assembly
2675 */
2676 if (i)
2677 memset(buf, 0x00, 4096);
2678 else
2679 memset(buf, 0xff, 4096);
2680 memcpy(buf, (char *)bms, sizeof(bitmap_super_t));
2681
2682 /*
2683 * use 4096 boundary if bitmap_offset is aligned
2684 * with 8 sectors, then it should compatible with
2685 * older mdadm.
2686 */
2687 if (__le32_to_cpu(sb->bitmap_offset) & 7)
2688 towrite = calc_bitmap_size(bms, 512);
2689 else
2690 towrite = calc_bitmap_size(bms, 4096);
2691 while (towrite > 0) {
2692 n = towrite;
2693 if (n > 4096)
2694 n = 4096;
2695 n = awrite(&afd, buf, n);
2696 if (n > 0)
2697 towrite -= n;
2698 else
2699 break;
2700 if (i)
2701 memset(buf, 0x00, 4096);
2702 else
2703 memset(buf, 0xff, 4096);
2704 }
2705 fsync(fd);
2706 if (towrite) {
2707 rv = -2;
2708 break;
2709 }
2710 } while (++i < __le32_to_cpu(bms->nodes));
2711
2712 free(buf);
2713 return rv;
2714 }
2715
2716 static void free_super1(struct supertype *st)
2717 {
2718
2719 if (st->sb)
2720 free(st->sb);
2721 while (st->info) {
2722 struct devinfo *di = st->info;
2723 st->info = di->next;
2724 if (di->fd >= 0)
2725 close(di->fd);
2726 free(di);
2727 }
2728 st->sb = NULL;
2729 }
2730
2731 static int validate_geometry1(struct supertype *st, int level,
2732 int layout, int raiddisks,
2733 int *chunk, unsigned long long size,
2734 unsigned long long data_offset,
2735 char *subdev, unsigned long long *freesize,
2736 int consistency_policy, int verbose)
2737 {
2738 unsigned long long ldsize, devsize;
2739 int bmspace;
2740 unsigned long long headroom;
2741 int fd;
2742
2743 if (level == LEVEL_CONTAINER) {
2744 if (verbose)
2745 pr_err("1.x metadata does not support containers\n");
2746 return 0;
2747 }
2748 if (*chunk == UnSet)
2749 *chunk = DEFAULT_CHUNK;
2750
2751 if (!subdev)
2752 return 1;
2753
2754 if (st->minor_version < 0)
2755 /* not specified, so time to set default */
2756 st->minor_version = 2;
2757
2758 fd = open(subdev, O_RDONLY|O_EXCL, 0);
2759 if (fd < 0) {
2760 if (verbose)
2761 pr_err("super1.x cannot open %s: %s\n",
2762 subdev, strerror(errno));
2763 return 0;
2764 }
2765
2766 if (!get_dev_size(fd, subdev, &ldsize)) {
2767 close(fd);
2768 return 0;
2769 }
2770 close(fd);
2771
2772 devsize = ldsize >> 9;
2773 if (devsize < 24) {
2774 *freesize = 0;
2775 return 0;
2776 }
2777
2778 /* creating: allow suitable space for bitmap or PPL */
2779 if (consistency_policy == CONSISTENCY_POLICY_PPL)
2780 bmspace = MULTIPLE_PPL_AREA_SIZE_SUPER1 >> 9;
2781 else
2782 bmspace = choose_bm_space(devsize);
2783
2784 if (data_offset == INVALID_SECTORS)
2785 data_offset = st->data_offset;
2786 if (data_offset == INVALID_SECTORS)
2787 switch (st->minor_version) {
2788 case 0:
2789 data_offset = 0;
2790 break;
2791 case 1:
2792 case 2:
2793 /* Choose data offset appropriate for this device
2794 * and use as default for whole array.
2795 * The data_offset must allow for bitmap space
2796 * and base metadata, should allow for some headroom
2797 * for reshape, and should be rounded to multiple
2798 * of 1M.
2799 * Headroom is limited to 128M, but aim for about 0.1%
2800 */
2801 headroom = 128*1024*2;
2802 while ((headroom << 10) > devsize &&
2803 (*chunk == 0 ||
2804 headroom / 2 >= ((unsigned)(*chunk)*2)*2))
2805 headroom >>= 1;
2806 data_offset = 12*2 + bmspace + headroom;
2807 #define ONE_MEG (2*1024)
2808 data_offset = ROUND_UP(data_offset, ONE_MEG);
2809 break;
2810 }
2811 if (st->data_offset == INVALID_SECTORS)
2812 st->data_offset = data_offset;
2813 switch(st->minor_version) {
2814 case 0: /* metadata at end. Round down and subtract space to reserve */
2815 devsize = (devsize & ~(4ULL*2-1));
2816 /* space for metadata, bblog, bitmap/ppl */
2817 devsize -= 8*2 + 8 + bmspace;
2818 break;
2819 case 1:
2820 case 2:
2821 devsize -= data_offset;
2822 break;
2823 }
2824 *freesize = devsize;
2825 return 1;
2826 }
2827
2828 void *super1_make_v0(struct supertype *st, struct mdinfo *info, mdp_super_t *sb0)
2829 {
2830 /* Create a v1.0 superblock based on 'info'*/
2831 void *ret;
2832 struct mdp_superblock_1 *sb;
2833 int i;
2834 unsigned long long offset;
2835
2836 if (posix_memalign(&ret, 4096, 1024) != 0)
2837 return NULL;
2838 sb = ret;
2839 memset(ret, 0, 1024);
2840 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
2841 sb->major_version = __cpu_to_le32(1);
2842
2843 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
2844 sprintf(sb->set_name, "%d", sb0->md_minor);
2845 sb->ctime = __cpu_to_le32(info->array.ctime+1);
2846 sb->level = __cpu_to_le32(info->array.level);
2847 sb->layout = __cpu_to_le32(info->array.layout);
2848 sb->size = __cpu_to_le64(info->component_size);
2849 sb->chunksize = __cpu_to_le32(info->array.chunk_size/512);
2850 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
2851 if (info->array.level > 0)
2852 sb->data_size = sb->size;
2853 else
2854 sb->data_size = st->ss->avail_size(st, st->devsize/512, 0);
2855 sb->resync_offset = MaxSector;
2856 sb->max_dev = __cpu_to_le32(MD_SB_DISKS);
2857 sb->dev_number = __cpu_to_le32(info->disk.number);
2858 sb->utime = __cpu_to_le64(info->array.utime);
2859
2860 offset = st->devsize/512 - 8*2;
2861 offset &= ~(4*2-1);
2862 sb->super_offset = __cpu_to_le64(offset);
2863 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2864
2865 random_uuid(sb->device_uuid);
2866
2867 for (i = 0; i < MD_SB_DISKS; i++) {
2868 int state = sb0->disks[i].state;
2869 sb->dev_roles[i] = MD_DISK_ROLE_SPARE;
2870 if ((state & (1<<MD_DISK_SYNC)) &&
2871 !(state & (1<<MD_DISK_FAULTY)))
2872 sb->dev_roles[i] = __cpu_to_le16(sb0->disks[i].raid_disk);
2873 }
2874 sb->sb_csum = calc_sb_1_csum(sb);
2875 return ret;
2876 }
2877
2878 struct superswitch super1 = {
2879 .examine_super = examine_super1,
2880 .brief_examine_super = brief_examine_super1,
2881 .export_examine_super = export_examine_super1,
2882 .detail_super = detail_super1,
2883 .brief_detail_super = brief_detail_super1,
2884 .export_detail_super = export_detail_super1,
2885 .write_init_super = write_init_super1,
2886 .validate_geometry = validate_geometry1,
2887 .add_to_super = add_to_super1,
2888 .examine_badblocks = examine_badblocks_super1,
2889 .copy_metadata = copy_metadata1,
2890 .write_init_ppl = write_init_ppl1,
2891 .match_home = match_home1,
2892 .uuid_from_super = uuid_from_super1,
2893 .getinfo_super = getinfo_super1,
2894 .container_content = container_content1,
2895 .update_super = update_super1,
2896 .init_super = init_super1,
2897 .store_super = store_super1,
2898 .compare_super = compare_super1,
2899 .load_super = load_super1,
2900 .match_metadata_desc = match_metadata_desc1,
2901 .avail_size = avail_size1,
2902 .add_internal_bitmap = add_internal_bitmap1,
2903 .locate_bitmap = locate_bitmap1,
2904 .write_bitmap = write_bitmap1,
2905 .free_super = free_super1,
2906 #if __BYTE_ORDER == BIG_ENDIAN
2907 .swapuuid = 0,
2908 #else
2909 .swapuuid = 1,
2910 #endif
2911 .name = "1.x",
2912 };
Unnamed repository; edit this file 'description' to name the repository.