]> 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
mdadm: fixes some trivial typos in comments
[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 /* feature_map bits */
125 #define MD_FEATURE_BITMAP_OFFSET 1
126 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
127 * must be honoured
128 */
129 #define MD_FEATURE_RESHAPE_ACTIVE 4
130 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
131 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
132 * active device with same 'role'.
133 * 'recovery_offset' is also set.
134 */
135 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
136 * of devices, but is going
137 * backwards anyway.
138 */
139 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
140 #define MD_FEATURE_BITMAP_VERSIONED 256 /* bitmap version number checked properly */
141 #define MD_FEATURE_JOURNAL 512 /* support write journal */
142 #define MD_FEATURE_PPL 1024 /* support PPL */
143 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
144 |MD_FEATURE_RECOVERY_OFFSET \
145 |MD_FEATURE_RESHAPE_ACTIVE \
146 |MD_FEATURE_BAD_BLOCKS \
147 |MD_FEATURE_REPLACEMENT \
148 |MD_FEATURE_RESHAPE_BACKWARDS \
149 |MD_FEATURE_NEW_OFFSET \
150 |MD_FEATURE_BITMAP_VERSIONED \
151 |MD_FEATURE_JOURNAL \
152 |MD_FEATURE_PPL \
153 )
154
155 static int role_from_sb(struct mdp_superblock_1 *sb)
156 {
157 unsigned int d;
158 int role;
159
160 d = __le32_to_cpu(sb->dev_number);
161 if (d < __le32_to_cpu(sb->max_dev))
162 role = __le16_to_cpu(sb->dev_roles[d]);
163 else
164 role = MD_DISK_ROLE_SPARE;
165 return role;
166 }
167
168 /* return how many bytes are needed for bitmap, for cluster-md each node
169 * should have it's own bitmap */
170 static unsigned int calc_bitmap_size(bitmap_super_t *bms, unsigned int boundary)
171 {
172 unsigned long long bits, bytes;
173
174 bits = bitmap_bits(__le64_to_cpu(bms->sync_size),
175 __le32_to_cpu(bms->chunksize));
176 bytes = (bits+7) >> 3;
177 bytes += sizeof(bitmap_super_t);
178 bytes = ROUND_UP(bytes, boundary);
179
180 return bytes;
181 }
182
183 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
184 {
185 unsigned int disk_csum, csum;
186 unsigned long long newcsum;
187 int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2;
188 unsigned int *isuper = (unsigned int*)sb;
189
190 /* make sure I can count... */
191 if (offsetof(struct mdp_superblock_1,data_offset) != 128 ||
192 offsetof(struct mdp_superblock_1, utime) != 192 ||
193 sizeof(struct mdp_superblock_1) != 256) {
194 fprintf(stderr, "WARNING - superblock isn't sized correctly\n");
195 }
196
197 disk_csum = sb->sb_csum;
198 sb->sb_csum = 0;
199 newcsum = 0;
200 for (; size>=4; size -= 4 ) {
201 newcsum += __le32_to_cpu(*isuper);
202 isuper++;
203 }
204
205 if (size == 2)
206 newcsum += __le16_to_cpu(*(unsigned short*) isuper);
207
208 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
209 sb->sb_csum = disk_csum;
210 return __cpu_to_le32(csum);
211 }
212
213 /*
214 * Information related to file descriptor used for aligned reads/writes.
215 * Cache the block size.
216 */
217 struct align_fd {
218 int fd;
219 int blk_sz;
220 };
221
222 static void init_afd(struct align_fd *afd, int fd)
223 {
224 afd->fd = fd;
225 if (!get_dev_sector_size(afd->fd, NULL, (unsigned int *)&afd->blk_sz))
226 afd->blk_sz = 512;
227 }
228
229 static char abuf[4096+4096];
230
231 static int aread(struct align_fd *afd, void *buf, int len)
232 {
233 /* aligned read.
234 * On devices with a 4K sector size, we need to read
235 * the full sector and copy relevant bits into
236 * the buffer
237 */
238 int bsize, iosize;
239 char *b;
240 int n;
241
242 bsize = afd->blk_sz;
243
244 if (!bsize || bsize > 4096 || len > 4096) {
245 if (!bsize)
246 fprintf(stderr, "WARNING - aread() called with invalid block size\n");
247 return -1;
248 }
249 b = ROUND_UP_PTR((char *)abuf, 4096);
250
251 for (iosize = 0; iosize < len; iosize += bsize)
252 ;
253 n = read(afd->fd, b, iosize);
254 if (n <= 0)
255 return n;
256 lseek(afd->fd, len - n, 1);
257 if (n > len)
258 n = len;
259 memcpy(buf, b, n);
260 return n;
261 }
262
263 static int awrite(struct align_fd *afd, void *buf, int len)
264 {
265 /* aligned write.
266 * On devices with a 4K sector size, we need to write
267 * the full sector. We pre-read if the sector is larger
268 * than the write.
269 * The address must be sector-aligned.
270 */
271 int bsize, iosize;
272 char *b;
273 int n;
274
275 bsize = afd->blk_sz;
276 if (!bsize || bsize > 4096 || len > 4096) {
277 if (!bsize)
278 fprintf(stderr, "WARNING - awrite() called with invalid block size\n");
279 return -1;
280 }
281 b = ROUND_UP_PTR((char *)abuf, 4096);
282
283 for (iosize = 0; iosize < len ; iosize += bsize)
284 ;
285
286 if (len != iosize) {
287 n = read(afd->fd, b, iosize);
288 if (n <= 0)
289 return n;
290 lseek(afd->fd, -n, 1);
291 }
292
293 memcpy(b, buf, len);
294 n = write(afd->fd, b, iosize);
295 if (n <= 0)
296 return n;
297 lseek(afd->fd, len - n, 1);
298 return len;
299 }
300
301 static inline unsigned int choose_ppl_space(int chunk)
302 {
303 return (PPL_HEADER_SIZE >> 9) + (chunk > 128*2 ? chunk : 128*2);
304 }
305
306 static void examine_super1(struct supertype *st, char *homehost)
307 {
308 struct mdp_superblock_1 *sb = st->sb;
309 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
310 time_t atime;
311 unsigned int d;
312 int role;
313 int delta_extra = 0;
314 int i;
315 char *c;
316 int l = homehost ? strlen(homehost) : 0;
317 int layout;
318 unsigned long long sb_offset;
319 struct mdinfo info;
320
321 printf(" Magic : %08x\n", __le32_to_cpu(sb->magic));
322 printf(" Version : 1");
323 sb_offset = __le64_to_cpu(sb->super_offset);
324 if (sb_offset <= 4)
325 printf(".1\n");
326 else if (sb_offset <= 8)
327 printf(".2\n");
328 else
329 printf(".0\n");
330 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb->feature_map));
331 printf(" Array UUID : ");
332 for (i=0; i<16; i++) {
333 if ((i&3)==0 && i != 0) printf(":");
334 printf("%02x", sb->set_uuid[i]);
335 }
336 printf("\n");
337 printf(" Name : %.32s", sb->set_name);
338 if (l > 0 && l < 32 &&
339 sb->set_name[l] == ':' &&
340 strncmp(sb->set_name, homehost, l) == 0)
341 printf(" (local to host %s)", homehost);
342 printf("\n");
343 if (bms->nodes > 0 &&
344 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
345 printf(" Cluster Name : %-64s\n", bms->cluster_name);
346 atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL;
347 printf(" Creation Time : %.24s\n", ctime(&atime));
348 c=map_num(pers, __le32_to_cpu(sb->level));
349 printf(" Raid Level : %s\n", c?c:"-unknown-");
350 printf(" Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks));
351 printf("\n");
352 printf(" Avail Dev Size : %llu%s\n",
353 (unsigned long long)__le64_to_cpu(sb->data_size),
354 human_size(__le64_to_cpu(sb->data_size)<<9));
355 if (__le32_to_cpu(sb->level) > 0) {
356 int ddsks = 0, ddsks_denom = 1;
357 switch(__le32_to_cpu(sb->level)) {
358 case 1: ddsks=1;break;
359 case 4:
360 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
361 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
362 case 10:
363 layout = __le32_to_cpu(sb->layout);
364 ddsks = __le32_to_cpu(sb->raid_disks);
365 ddsks_denom = (layout&255) * ((layout>>8)&255);
366 }
367 if (ddsks) {
368 long long asize = __le64_to_cpu(sb->size);
369 asize = (asize << 9) * ddsks / ddsks_denom;
370 printf(" Array Size : %llu%s\n",
371 asize >> 10, human_size(asize));
372 }
373 if (sb->size != sb->data_size)
374 printf(" Used Dev Size : %llu%s\n",
375 (unsigned long long)__le64_to_cpu(sb->size),
376 human_size(__le64_to_cpu(sb->size)<<9));
377 }
378 if (sb->data_offset)
379 printf(" Data Offset : %llu sectors\n",
380 (unsigned long long)__le64_to_cpu(sb->data_offset));
381 if (sb->new_offset &&
382 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) {
383 unsigned long long offset = __le64_to_cpu(sb->data_offset);
384 offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset);
385 printf(" New Offset : %llu sectors\n", offset);
386 }
387 printf(" Super Offset : %llu sectors\n",
388 (unsigned long long)__le64_to_cpu(sb->super_offset));
389 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)
390 printf("Recovery Offset : %llu sectors\n",
391 (unsigned long long)__le64_to_cpu(sb->recovery_offset));
392
393 st->ss->getinfo_super(st, &info, NULL);
394 if (info.space_after != 1 &&
395 !(__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
396 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
397 info.space_before, info.space_after);
398
399 printf(" State : %s\n",
400 (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean");
401 printf(" Device UUID : ");
402 for (i=0; i<16; i++) {
403 if ((i&3)==0 && i != 0)
404 printf(":");
405 printf("%02x", sb->device_uuid[i]);
406 }
407 printf("\n");
408 printf("\n");
409 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
410 printf("Internal Bitmap : %ld sectors from superblock\n",
411 (long)(int32_t)__le32_to_cpu(sb->bitmap_offset));
412 } else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_PPL)) {
413 printf(" PPL : %u sectors at offset %d sectors from superblock\n",
414 __le16_to_cpu(sb->ppl.size),
415 __le16_to_cpu(sb->ppl.offset));
416 }
417 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)) {
418 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)
419 __le64_to_cpu(sb->reshape_position)/2,
420 human_size(__le64_to_cpu(sb->reshape_position)<<9));
421 if (__le32_to_cpu(sb->delta_disks)) {
422 printf(" Delta Devices : %d",
423 __le32_to_cpu(sb->delta_disks));
424 printf(" (%d->%d)\n",
425 __le32_to_cpu(sb->raid_disks) -
426 __le32_to_cpu(sb->delta_disks),
427 __le32_to_cpu(sb->raid_disks));
428 if ((int)__le32_to_cpu(sb->delta_disks) < 0)
429 delta_extra = -__le32_to_cpu(sb->delta_disks);
430 }
431 if (__le32_to_cpu(sb->new_level) != __le32_to_cpu(sb->level)) {
432 c = map_num(pers, __le32_to_cpu(sb->new_level));
433 printf(" New Level : %s\n", c?c:"-unknown-");
434 }
435 if (__le32_to_cpu(sb->new_layout) !=
436 __le32_to_cpu(sb->layout)) {
437 if (__le32_to_cpu(sb->level) == 5) {
438 c = map_num(r5layout,
439 __le32_to_cpu(sb->new_layout));
440 printf(" New Layout : %s\n", c?c:"-unknown-");
441 }
442 if (__le32_to_cpu(sb->level) == 6) {
443 c = map_num(r6layout,
444 __le32_to_cpu(sb->new_layout));
445 printf(" New Layout : %s\n", c?c:"-unknown-");
446 }
447 if (__le32_to_cpu(sb->level) == 10) {
448 printf(" New Layout :");
449 print_r10_layout(__le32_to_cpu(sb->new_layout));
450 printf("\n");
451 }
452 }
453 if (__le32_to_cpu(sb->new_chunk) !=
454 __le32_to_cpu(sb->chunksize))
455 printf(" New Chunksize : %dK\n",
456 __le32_to_cpu(sb->new_chunk)/2);
457 printf("\n");
458 }
459 if (sb->devflags) {
460 printf(" Flags :");
461 if (sb->devflags & WriteMostly1)
462 printf(" write-mostly");
463 if (sb->devflags & FailFast1)
464 printf(" failfast");
465 printf("\n");
466 }
467
468 atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL;
469 printf(" Update Time : %.24s\n", ctime(&atime));
470
471 if (sb->bblog_size && sb->bblog_offset) {
472 printf(" Bad Block Log : %d entries available at offset %ld sectors",
473 __le16_to_cpu(sb->bblog_size)*512/8,
474 (long)(int32_t)__le32_to_cpu(sb->bblog_offset));
475 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
476 printf(" - bad blocks present.");
477 printf("\n");
478 }
479
480 if (calc_sb_1_csum(sb) == sb->sb_csum)
481 printf(" Checksum : %x - correct\n",
482 __le32_to_cpu(sb->sb_csum));
483 else
484 printf(" Checksum : %x - expected %x\n",
485 __le32_to_cpu(sb->sb_csum),
486 __le32_to_cpu(calc_sb_1_csum(sb)));
487 printf(" Events : %llu\n",
488 (unsigned long long)__le64_to_cpu(sb->events));
489 printf("\n");
490 if (__le32_to_cpu(sb->level) == 5) {
491 c = map_num(r5layout, __le32_to_cpu(sb->layout));
492 printf(" Layout : %s\n", c?c:"-unknown-");
493 }
494 if (__le32_to_cpu(sb->level) == 6) {
495 c = map_num(r6layout, __le32_to_cpu(sb->layout));
496 printf(" Layout : %s\n", c?c:"-unknown-");
497 }
498 if (__le32_to_cpu(sb->level) == 10) {
499 int lo = __le32_to_cpu(sb->layout);
500 printf(" Layout :");
501 print_r10_layout(lo);
502 printf("\n");
503 }
504 switch(__le32_to_cpu(sb->level)) {
505 case 0:
506 case 4:
507 case 5:
508 case 6:
509 case 10:
510 printf(" Chunk Size : %dK\n",
511 __le32_to_cpu(sb->chunksize)/2);
512 break;
513 case -1:
514 printf(" Rounding : %dK\n",
515 __le32_to_cpu(sb->chunksize)/2);
516 break;
517 default:
518 break;
519 }
520 printf("\n");
521 #if 0
522 /* This turns out to just be confusing */
523 printf(" Array Slot : %d (", __le32_to_cpu(sb->dev_number));
524 for (i = __le32_to_cpu(sb->max_dev); i> 0 ; i--)
525 if (__le16_to_cpu(sb->dev_roles[i-1]) != MD_DISK_ROLE_SPARE)
526 break;
527 for (d = 0; d < i; d++) {
528 int role = __le16_to_cpu(sb->dev_roles[d]);
529 if (d)
530 printf(", ");
531 if (role == MD_DISK_ROLE_SPARE)
532 printf("empty");
533 else
534 if(role == MD_DISK_ROLE_FAULTY)
535 printf("failed");
536 else
537 printf("%d", role);
538 }
539 printf(")\n");
540 #endif
541 printf(" Device Role : ");
542 role = role_from_sb(sb);
543 if (role >= MD_DISK_ROLE_FAULTY)
544 printf("spare\n");
545 else if (role == MD_DISK_ROLE_JOURNAL)
546 printf("Journal\n");
547 else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_REPLACEMENT))
548 printf("Replacement device %d\n", role);
549 else
550 printf("Active device %d\n", role);
551
552 printf(" Array State : ");
553 for (d = 0; d < __le32_to_cpu(sb->raid_disks) + delta_extra; d++) {
554 int cnt = 0;
555 unsigned int i;
556 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
557 unsigned int role = __le16_to_cpu(sb->dev_roles[i]);
558 if (role == d)
559 cnt++;
560 }
561 if (cnt == 2)
562 printf("R");
563 else if (cnt == 1)
564 printf("A");
565 else if (cnt == 0)
566 printf(".");
567 else
568 printf("?");
569 }
570 #if 0
571 /* This is confusing too */
572 faulty = 0;
573 for (i = 0; i< __le32_to_cpu(sb->max_dev); i++) {
574 int role = __le16_to_cpu(sb->dev_roles[i]);
575 if (role == MD_DISK_ROLE_FAULTY)
576 faulty++;
577 }
578 if (faulty)
579 printf(" %d failed", faulty);
580 #endif
581 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
582 printf("\n");
583 }
584
585 static void brief_examine_super1(struct supertype *st, int verbose)
586 {
587 struct mdp_superblock_1 *sb = st->sb;
588 int i;
589 unsigned long long sb_offset;
590 char *nm;
591 char *c = map_num(pers, __le32_to_cpu(sb->level));
592
593 nm = strchr(sb->set_name, ':');
594 if (nm)
595 nm++;
596 else if (sb->set_name[0])
597 nm = sb->set_name;
598 else
599 nm = NULL;
600
601 printf("ARRAY ");
602 if (nm) {
603 printf("/dev/md/");
604 print_escape(nm);
605 putchar(' ');
606 }
607 if (verbose && c)
608 printf(" level=%s", c);
609 sb_offset = __le64_to_cpu(sb->super_offset);
610 if (sb_offset <= 4)
611 printf(" metadata=1.1 ");
612 else if (sb_offset <= 8)
613 printf(" metadata=1.2 ");
614 else
615 printf(" metadata=1.0 ");
616 if (verbose)
617 printf("num-devices=%d ", __le32_to_cpu(sb->raid_disks));
618 printf("UUID=");
619 for (i = 0; i < 16; i++) {
620 if ((i&3)==0 && i != 0)
621 printf(":");
622 printf("%02x", sb->set_uuid[i]);
623 }
624 if (sb->set_name[0]) {
625 printf(" name=");
626 print_quoted(sb->set_name);
627 }
628 printf("\n");
629 }
630
631 static void export_examine_super1(struct supertype *st)
632 {
633 struct mdp_superblock_1 *sb = st->sb;
634 int i;
635 int len = 32;
636 int layout;
637
638 printf("MD_LEVEL=%s\n", map_num(pers, __le32_to_cpu(sb->level)));
639 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb->raid_disks));
640 for (i = 0; i < 32; i++)
641 if (sb->set_name[i] == '\n' || sb->set_name[i] == '\0') {
642 len = i;
643 break;
644 }
645 if (len)
646 printf("MD_NAME=%.*s\n", len, sb->set_name);
647 if (__le32_to_cpu(sb->level) > 0) {
648 int ddsks = 0, ddsks_denom = 1;
649 switch(__le32_to_cpu(sb->level)) {
650 case 1:
651 ddsks = 1;
652 break;
653 case 4:
654 case 5:
655 ddsks = __le32_to_cpu(sb->raid_disks)-1;
656 break;
657 case 6:
658 ddsks = __le32_to_cpu(sb->raid_disks)-2;
659 break;
660 case 10:
661 layout = __le32_to_cpu(sb->layout);
662 ddsks = __le32_to_cpu(sb->raid_disks);
663 ddsks_denom = (layout&255) * ((layout>>8)&255);
664 }
665 if (ddsks) {
666 long long asize = __le64_to_cpu(sb->size);
667 asize = (asize << 9) * ddsks / ddsks_denom;
668 printf("MD_ARRAY_SIZE=%s\n",
669 human_size_brief(asize, JEDEC));
670 }
671 }
672 printf("MD_UUID=");
673 for (i = 0; i < 16; i++) {
674 if ((i&3) == 0 && i != 0)
675 printf(":");
676 printf("%02x", sb->set_uuid[i]);
677 }
678 printf("\n");
679 printf("MD_UPDATE_TIME=%llu\n",
680 __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL);
681 printf("MD_DEV_UUID=");
682 for (i = 0; i < 16; i++) {
683 if ((i&3) == 0 && i != 0)
684 printf(":");
685 printf("%02x", sb->device_uuid[i]);
686 }
687 printf("\n");
688 printf("MD_EVENTS=%llu\n",
689 (unsigned long long)__le64_to_cpu(sb->events));
690 }
691
692 static int copy_metadata1(struct supertype *st, int from, int to)
693 {
694 /* Read superblock. If it looks good, write it out.
695 * Then if a bitmap is present, copy that.
696 * And if a bad-block-list is present, copy that too.
697 */
698 void *buf;
699 unsigned long long dsize, sb_offset;
700 const int bufsize = 4*1024;
701 struct mdp_superblock_1 super, *sb;
702
703 if (posix_memalign(&buf, 4096, bufsize) != 0)
704 return 1;
705
706 if (!get_dev_size(from, NULL, &dsize))
707 goto err;
708
709 dsize >>= 9;
710 if (dsize < 24)
711 goto err;
712 switch(st->minor_version) {
713 case 0:
714 sb_offset = dsize;
715 sb_offset -= 8*2;
716 sb_offset &= ~(4*2-1);
717 break;
718 case 1:
719 sb_offset = 0;
720 break;
721 case 2:
722 sb_offset = 4*2;
723 break;
724 default:
725 goto err;
726 }
727
728 if (lseek64(from, sb_offset << 9, 0) < 0LL)
729 goto err;
730 if (read(from, buf, bufsize) != bufsize)
731 goto err;
732
733 sb = buf;
734 super = *sb; // save most of sb for when we reuse buf
735
736 if (__le32_to_cpu(super.magic) != MD_SB_MAGIC ||
737 __le32_to_cpu(super.major_version) != 1 ||
738 __le64_to_cpu(super.super_offset) != sb_offset ||
739 calc_sb_1_csum(sb) != super.sb_csum)
740 goto err;
741
742 if (lseek64(to, sb_offset << 9, 0) < 0LL)
743 goto err;
744 if (write(to, buf, bufsize) != bufsize)
745 goto err;
746
747 if (super.feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
748 unsigned long long bitmap_offset = sb_offset;
749 int bytes = 4096; // just an estimate.
750 int written = 0;
751 struct align_fd afrom, ato;
752
753 init_afd(&afrom, from);
754 init_afd(&ato, to);
755
756 bitmap_offset += (int32_t)__le32_to_cpu(super.bitmap_offset);
757
758 if (lseek64(from, bitmap_offset<<9, 0) < 0)
759 goto err;
760 if (lseek64(to, bitmap_offset<<9, 0) < 0)
761 goto err;
762
763 for (written = 0; written < bytes ; ) {
764 int n = bytes - written;
765 if (n > 4096)
766 n = 4096;
767 if (aread(&afrom, buf, n) != n)
768 goto err;
769 if (written == 0) {
770 /* have the header, can calculate
771 * correct bitmap bytes */
772 bitmap_super_t *bms;
773 bms = (void*)buf;
774 bytes = calc_bitmap_size(bms, 512);
775 if (n > bytes)
776 n = bytes;
777 }
778 if (awrite(&ato, buf, n) != n)
779 goto err;
780 written += n;
781 }
782 }
783
784 if (super.bblog_size != 0 &&
785 __le16_to_cpu(super.bblog_size) <= 100 &&
786 super.bblog_offset != 0 &&
787 (super.feature_map & __le32_to_cpu(MD_FEATURE_BAD_BLOCKS))) {
788 /* There is a bad block log */
789 unsigned long long bb_offset = sb_offset;
790 int bytes = __le16_to_cpu(super.bblog_size) * 512;
791 int written = 0;
792 struct align_fd afrom, ato;
793
794 init_afd(&afrom, from);
795 init_afd(&ato, to);
796
797 bb_offset += (int32_t)__le32_to_cpu(super.bblog_offset);
798
799 if (lseek64(from, bb_offset<<9, 0) < 0)
800 goto err;
801 if (lseek64(to, bb_offset<<9, 0) < 0)
802 goto err;
803
804 for (written = 0; written < bytes ; ) {
805 int n = bytes - written;
806 if (n > 4096)
807 n = 4096;
808 if (aread(&afrom, buf, n) != n)
809 goto err;
810
811 if (awrite(&ato, buf, n) != n)
812 goto err;
813 written += n;
814 }
815 }
816
817 free(buf);
818 return 0;
819
820 err:
821 free(buf);
822 return 1;
823 }
824
825 static void detail_super1(struct supertype *st, char *homehost)
826 {
827 struct mdp_superblock_1 *sb = st->sb;
828 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
829 int i;
830 int l = homehost ? strlen(homehost) : 0;
831
832 printf(" Name : %.32s", sb->set_name);
833 if (l > 0 && l < 32 && sb->set_name[l] == ':' &&
834 strncmp(sb->set_name, homehost, l) == 0)
835 printf(" (local to host %s)", homehost);
836 if (bms->nodes > 0 &&
837 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
838 printf("\n Cluster Name : %-64s", bms->cluster_name);
839 printf("\n UUID : ");
840 for (i = 0; i < 16; i++) {
841 if ((i&3) == 0 && i != 0)
842 printf(":");
843 printf("%02x", sb->set_uuid[i]);
844 }
845 printf("\n Events : %llu\n\n",
846 (unsigned long long)__le64_to_cpu(sb->events));
847 }
848
849 static void brief_detail_super1(struct supertype *st)
850 {
851 struct mdp_superblock_1 *sb = st->sb;
852 int i;
853
854 if (sb->set_name[0]) {
855 printf(" name=");
856 print_quoted(sb->set_name);
857 }
858 printf(" UUID=");
859 for (i = 0; i < 16; i++) {
860 if ((i & 3) == 0 && i != 0)
861 printf(":");
862 printf("%02x", sb->set_uuid[i]);
863 }
864 }
865
866 static void export_detail_super1(struct supertype *st)
867 {
868 struct mdp_superblock_1 *sb = st->sb;
869 int i;
870 int len = 32;
871
872 for (i = 0; i < 32; i++)
873 if (sb->set_name[i] == '\n' || sb->set_name[i] == '\0') {
874 len = i;
875 break;
876 }
877 if (len)
878 printf("MD_NAME=%.*s\n", len, sb->set_name);
879 }
880
881 static int examine_badblocks_super1(struct supertype *st, int fd, char *devname)
882 {
883 struct mdp_superblock_1 *sb = st->sb;
884 unsigned long long offset;
885 int size;
886 __u64 *bbl, *bbp;
887 int i;
888
889 if (!sb->bblog_size || __le16_to_cpu(sb->bblog_size) > 100 ||
890 !sb->bblog_offset){
891 printf("No bad-blocks list configured on %s\n", devname);
892 return 0;
893 }
894 if ((sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS)) == 0) {
895 printf("Bad-blocks list is empty in %s\n", devname);
896 return 0;
897 }
898
899 size = __le16_to_cpu(sb->bblog_size)* 512;
900 if (posix_memalign((void**)&bbl, 4096, size) != 0) {
901 pr_err("could not allocate badblocks list\n");
902 return 0;
903 }
904 offset = __le64_to_cpu(sb->super_offset) +
905 (int)__le32_to_cpu(sb->bblog_offset);
906 offset <<= 9;
907 if (lseek64(fd, offset, 0) < 0) {
908 pr_err("Cannot seek to bad-blocks list\n");
909 return 1;
910 }
911 if (read(fd, bbl, size) != size) {
912 pr_err("Cannot read bad-blocks list\n");
913 return 1;
914 }
915 /* 64bits per entry. 10 bits is block-count, 54 bits is block
916 * offset. Blocks are sectors unless bblog->shift makes them bigger
917 */
918 bbp = (__u64*)bbl;
919 printf("Bad-blocks on %s:\n", devname);
920 for (i = 0; i < size/8; i++, bbp++) {
921 __u64 bb = __le64_to_cpu(*bbp);
922 int count = bb & 0x3ff;
923 unsigned long long sector = bb >> 10;
924
925 if (bb + 1 == 0)
926 break;
927
928 sector <<= sb->bblog_shift;
929 count <<= sb->bblog_shift;
930
931 printf("%20llu for %d sectors\n", sector, count);
932 }
933 return 0;
934 }
935
936 static int match_home1(struct supertype *st, char *homehost)
937 {
938 struct mdp_superblock_1 *sb = st->sb;
939 int l = homehost ? strlen(homehost) : 0;
940
941 return (l > 0 && l < 32 && sb->set_name[l] == ':' &&
942 strncmp(sb->set_name, homehost, l) == 0);
943 }
944
945 static void uuid_from_super1(struct supertype *st, int uuid[4])
946 {
947 struct mdp_superblock_1 *super = st->sb;
948 char *cuuid = (char*)uuid;
949 int i;
950 for (i = 0; i < 16; i++)
951 cuuid[i] = super->set_uuid[i];
952 }
953
954 static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
955 {
956 struct mdp_superblock_1 *sb = st->sb;
957 struct bitmap_super_s *bsb = (void*)(((char*)sb)+MAX_SB_SIZE);
958 struct misc_dev_info *misc =
959 (void*)(((char*)sb)+MAX_SB_SIZE+BM_SUPER_SIZE);
960 int working = 0;
961 unsigned int i;
962 unsigned int role;
963 unsigned int map_disks = info->array.raid_disks;
964 unsigned long long super_offset;
965 unsigned long long data_size;
966
967 memset(info, 0, sizeof(*info));
968 info->array.major_version = 1;
969 info->array.minor_version = st->minor_version;
970 info->array.patch_version = 0;
971 info->array.raid_disks = __le32_to_cpu(sb->raid_disks);
972 info->array.level = __le32_to_cpu(sb->level);
973 info->array.layout = __le32_to_cpu(sb->layout);
974 info->array.md_minor = -1;
975 info->array.ctime = __le64_to_cpu(sb->ctime);
976 info->array.utime = __le64_to_cpu(sb->utime);
977 info->array.chunk_size = __le32_to_cpu(sb->chunksize)*512;
978 info->array.state =
979 (__le64_to_cpu(sb->resync_offset) == MaxSector) ? 1 : 0;
980
981 super_offset = __le64_to_cpu(sb->super_offset);
982 info->data_offset = __le64_to_cpu(sb->data_offset);
983 info->component_size = __le64_to_cpu(sb->size);
984 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
985 info->bitmap_offset = (int32_t)__le32_to_cpu(sb->bitmap_offset);
986 if (__le32_to_cpu(bsb->nodes) > 1)
987 info->array.state |= (1 << MD_SB_CLUSTERED);
988 } else if (sb->feature_map & __le32_to_cpu(MD_FEATURE_PPL)) {
989 info->ppl_offset = __le16_to_cpu(sb->ppl.offset);
990 info->ppl_size = __le16_to_cpu(sb->ppl.size);
991 info->ppl_sector = super_offset + info->ppl_offset;
992 }
993
994 info->disk.major = 0;
995 info->disk.minor = 0;
996 info->disk.number = __le32_to_cpu(sb->dev_number);
997 if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) ||
998 __le32_to_cpu(sb->dev_number) >= MAX_DEVS)
999 role = MD_DISK_ROLE_FAULTY;
1000 else
1001 role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]);
1002
1003 if (info->array.level <= 0)
1004 data_size = __le64_to_cpu(sb->data_size);
1005 else
1006 data_size = __le64_to_cpu(sb->size);
1007 if (info->data_offset < super_offset) {
1008 unsigned long long end;
1009 info->space_before = info->data_offset;
1010 end = super_offset;
1011
1012 if (sb->bblog_offset && sb->bblog_size) {
1013 unsigned long long bboffset = super_offset;
1014 bboffset += (int32_t)__le32_to_cpu(sb->bblog_offset);
1015 if (bboffset < end)
1016 end = bboffset;
1017 }
1018
1019 if (super_offset + info->bitmap_offset + info->ppl_offset < end)
1020 end = super_offset + info->bitmap_offset +
1021 info->ppl_offset;
1022
1023 if (info->data_offset + data_size < end)
1024 info->space_after = end - data_size - info->data_offset;
1025 else
1026 info->space_after = 0;
1027 } else {
1028 unsigned long long earliest;
1029 earliest = super_offset + (32+4)*2; /* match kernel */
1030 if (info->bitmap_offset > 0) {
1031 unsigned long long bmend = info->bitmap_offset;
1032 unsigned long long size = calc_bitmap_size(bsb, 4096);
1033 size /= 512;
1034 bmend += size;
1035 if (bmend > earliest)
1036 earliest = bmend;
1037 } else if (info->ppl_offset > 0) {
1038 unsigned long long pplend;
1039
1040 pplend = info->ppl_offset + info->ppl_size;
1041 if (pplend > earliest)
1042 earliest = pplend;
1043 }
1044 if (sb->bblog_offset && sb->bblog_size) {
1045 unsigned long long bbend = super_offset;
1046 bbend += (int32_t)__le32_to_cpu(sb->bblog_offset);
1047 bbend += __le16_to_cpu(sb->bblog_size);
1048 if (bbend > earliest)
1049 earliest = bbend;
1050 }
1051 if (earliest < info->data_offset)
1052 info->space_before = info->data_offset - earliest;
1053 else
1054 info->space_before = 0;
1055 info->space_after = misc->device_size - data_size -
1056 info->data_offset;
1057 }
1058 if (info->space_before == 0 && info->space_after == 0) {
1059 /* It will look like we don't support data_offset changes,
1060 * be we do - it's just that there is no room.
1061 * A change that reduced the number of devices should
1062 * still be allowed, so set the otherwise useless value of '1'
1063 */
1064 info->space_after = 1;
1065 }
1066
1067 info->disk.raid_disk = -1;
1068 switch(role) {
1069 case MD_DISK_ROLE_SPARE:
1070 /* spare: not active, not sync, not faulty */
1071 info->disk.state = 0;
1072 break;
1073 case MD_DISK_ROLE_FAULTY:
1074 info->disk.state = (1 << MD_DISK_FAULTY); /* faulty */
1075 break;
1076 case MD_DISK_ROLE_JOURNAL:
1077 info->disk.state = (1 << MD_DISK_JOURNAL);
1078 info->disk.raid_disk = role;
1079 /* journal uses all 4kB blocks*/
1080 info->space_after = (misc->device_size - info->data_offset) % 8;
1081 break;
1082 default:
1083 info->disk.state = 6; /* active and in sync */
1084 info->disk.raid_disk = role;
1085 }
1086 if (sb->devflags & WriteMostly1)
1087 info->disk.state |= (1 << MD_DISK_WRITEMOSTLY);
1088 if (sb->devflags & FailFast1)
1089 info->disk.state |= (1 << MD_DISK_FAILFAST);
1090 info->events = __le64_to_cpu(sb->events);
1091 sprintf(info->text_version, "1.%d", st->minor_version);
1092 info->safe_mode_delay = 200;
1093
1094 memcpy(info->uuid, sb->set_uuid, 16);
1095
1096 strncpy(info->name, sb->set_name, 32);
1097 info->name[32] = 0;
1098
1099 if ((__le32_to_cpu(sb->feature_map)&MD_FEATURE_REPLACEMENT)) {
1100 info->disk.state &= ~(1 << MD_DISK_SYNC);
1101 info->disk.state |= 1 << MD_DISK_REPLACEMENT;
1102 }
1103
1104 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET))
1105 info->recovery_start = __le32_to_cpu(sb->recovery_offset);
1106 else
1107 info->recovery_start = MaxSector;
1108
1109 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) {
1110 info->reshape_active = 1;
1111 if ((sb->feature_map & __le32_to_cpu(MD_FEATURE_NEW_OFFSET)) &&
1112 sb->new_offset != 0)
1113 info->reshape_active |= RESHAPE_NO_BACKUP;
1114 info->reshape_progress = __le64_to_cpu(sb->reshape_position);
1115 info->new_level = __le32_to_cpu(sb->new_level);
1116 info->delta_disks = __le32_to_cpu(sb->delta_disks);
1117 info->new_layout = __le32_to_cpu(sb->new_layout);
1118 info->new_chunk = __le32_to_cpu(sb->new_chunk)<<9;
1119 if (info->delta_disks < 0)
1120 info->array.raid_disks -= info->delta_disks;
1121 } else
1122 info->reshape_active = 0;
1123
1124 info->recovery_blocked = info->reshape_active;
1125
1126 if (map)
1127 for (i=0; i<map_disks; i++)
1128 map[i] = 0;
1129 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
1130 role = __le16_to_cpu(sb->dev_roles[i]);
1131 if (/*role == MD_DISK_ROLE_SPARE || */role < (unsigned) info->array.raid_disks) {
1132 working++;
1133 if (map && role < map_disks)
1134 map[role] = 1;
1135 }
1136 }
1137
1138 info->array.working_disks = working;
1139
1140 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_JOURNAL)) {
1141 info->journal_device_required = 1;
1142 info->consistency_policy = CONSISTENCY_POLICY_JOURNAL;
1143 } else if (sb->feature_map & __le32_to_cpu(MD_FEATURE_PPL)) {
1144 info->consistency_policy = CONSISTENCY_POLICY_PPL;
1145 } else if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
1146 info->consistency_policy = CONSISTENCY_POLICY_BITMAP;
1147 } else if (info->array.level <= 0) {
1148 info->consistency_policy = CONSISTENCY_POLICY_NONE;
1149 } else {
1150 info->consistency_policy = CONSISTENCY_POLICY_RESYNC;
1151 }
1152
1153 info->journal_clean = 0;
1154 }
1155
1156 static struct mdinfo *container_content1(struct supertype *st, char *subarray)
1157 {
1158 struct mdinfo *info;
1159
1160 if (subarray)
1161 return NULL;
1162
1163 info = xmalloc(sizeof(*info));
1164 getinfo_super1(st, info, NULL);
1165 return info;
1166 }
1167
1168 static int update_super1(struct supertype *st, struct mdinfo *info,
1169 char *update, char *devname, int verbose,
1170 int uuid_set, char *homehost)
1171 {
1172 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1173 * if any change was made. For others, the return value is
1174 * ignored.
1175 */
1176 int rv = 0;
1177 int lockid;
1178 struct mdp_superblock_1 *sb = st->sb;
1179 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1180
1181 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1182 rv = cluster_get_dlmlock(&lockid);
1183 if (rv) {
1184 pr_err("Cannot get dlmlock in %s return %d\n",
1185 __func__, rv);
1186 cluster_release_dlmlock(lockid);
1187 return rv;
1188 }
1189 }
1190
1191 if (strcmp(update, "homehost") == 0 &&
1192 homehost) {
1193 /* Note that 'homehost' is special as it is really
1194 * a "name" update.
1195 */
1196 char *c;
1197 update = "name";
1198 c = strchr(sb->set_name, ':');
1199 if (c)
1200 strncpy(info->name, c+1, 31 - (c-sb->set_name));
1201 else
1202 strncpy(info->name, sb->set_name, 32);
1203 info->name[32] = 0;
1204 }
1205
1206 if (strcmp(update, "force-one")==0) {
1207 /* Not enough devices for a working array,
1208 * so bring this one up-to-date
1209 */
1210 if (sb->events != __cpu_to_le64(info->events))
1211 rv = 1;
1212 sb->events = __cpu_to_le64(info->events);
1213 } else if (strcmp(update, "force-array")==0) {
1214 /* Degraded array and 'force' requests to
1215 * maybe need to mark it 'clean'.
1216 */
1217 switch(__le32_to_cpu(sb->level)) {
1218 case 4:
1219 case 5:
1220 case 6:
1221 /* need to force clean */
1222 if (sb->resync_offset != MaxSector)
1223 rv = 1;
1224 sb->resync_offset = MaxSector;
1225 }
1226 } else if (strcmp(update, "assemble")==0) {
1227 int d = info->disk.number;
1228 int want;
1229 if (info->disk.state & (1<<MD_DISK_ACTIVE))
1230 want = info->disk.raid_disk;
1231 else if (info->disk.state & (1<<MD_DISK_JOURNAL))
1232 want = MD_DISK_ROLE_JOURNAL;
1233 else
1234 want = MD_DISK_ROLE_SPARE;
1235 if (sb->dev_roles[d] != __cpu_to_le16(want)) {
1236 sb->dev_roles[d] = __cpu_to_le16(want);
1237 rv = 1;
1238 }
1239 if (info->reshape_active &&
1240 sb->feature_map &
1241 __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1242 info->delta_disks >= 0 &&
1243 info->reshape_progress <
1244 __le64_to_cpu(sb->reshape_position)) {
1245 sb->reshape_position =
1246 __cpu_to_le64(info->reshape_progress);
1247 rv = 1;
1248 }
1249 if (info->reshape_active &&
1250 sb->feature_map &
1251 __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1252 info->delta_disks < 0 &&
1253 info->reshape_progress >
1254 __le64_to_cpu(sb->reshape_position)) {
1255 sb->reshape_position =
1256 __cpu_to_le64(info->reshape_progress);
1257 rv = 1;
1258 }
1259 } else if (strcmp(update, "linear-grow-new") == 0) {
1260 unsigned int i;
1261 int fd;
1262 unsigned int max = __le32_to_cpu(sb->max_dev);
1263
1264 for (i = 0; i < max; i++)
1265 if (__le16_to_cpu(sb->dev_roles[i]) >=
1266 MD_DISK_ROLE_FAULTY)
1267 break;
1268 sb->dev_number = __cpu_to_le32(i);
1269 info->disk.number = i;
1270 if (i >= max) {
1271 sb->max_dev = __cpu_to_le32(max+1);
1272 }
1273
1274 random_uuid(sb->device_uuid);
1275
1276 sb->dev_roles[i] = __cpu_to_le16(info->disk.raid_disk);
1277
1278 fd = open(devname, O_RDONLY);
1279 if (fd >= 0) {
1280 unsigned long long ds;
1281 get_dev_size(fd, devname, &ds);
1282 close(fd);
1283 ds >>= 9;
1284 if (__le64_to_cpu(sb->super_offset) <
1285 __le64_to_cpu(sb->data_offset)) {
1286 sb->data_size = __cpu_to_le64(
1287 ds - __le64_to_cpu(sb->data_offset));
1288 } else {
1289 ds -= 8*2;
1290 ds &= ~(unsigned long long)(4*2-1);
1291 sb->super_offset = __cpu_to_le64(ds);
1292 sb->data_size = __cpu_to_le64(
1293 ds - __le64_to_cpu(sb->data_offset));
1294 }
1295 }
1296 } else if (strcmp(update, "linear-grow-update") == 0) {
1297 int max = __le32_to_cpu(sb->max_dev);
1298 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
1299 if (info->array.raid_disks > max) {
1300 sb->max_dev = __cpu_to_le32(max+1);
1301 }
1302 sb->dev_roles[info->disk.number] =
1303 __cpu_to_le16(info->disk.raid_disk);
1304 } else if (strcmp(update, "resync") == 0) {
1305 /* make sure resync happens */
1306 sb->resync_offset = 0ULL;
1307 } else if (strcmp(update, "uuid") == 0) {
1308 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
1309
1310 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)
1311 memcpy(bms->uuid, sb->set_uuid, 16);
1312 } else if (strcmp(update, "no-bitmap") == 0) {
1313 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1314 } else if (strcmp(update, "bbl") == 0) {
1315 /* only possible if there is room after the bitmap, or if
1316 * there is no bitmap
1317 */
1318 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1319 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1320 long bitmap_offset = 0;
1321 long bm_sectors = 0;
1322 long space;
1323
1324 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1325 bitmap_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1326 bm_sectors = calc_bitmap_size(bms, 4096) >> 9;
1327 } else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_PPL)) {
1328 bitmap_offset = (long)__le16_to_cpu(sb->ppl.offset);
1329 bm_sectors = (long)__le16_to_cpu(sb->ppl.size);
1330 }
1331
1332 if (sb_offset < data_offset) {
1333 /*
1334 * 1.1 or 1.2. Put bbl after bitmap leaving
1335 * at least 32K
1336 */
1337 long bb_offset;
1338 bb_offset = sb_offset + 8;
1339 if (bm_sectors && bitmap_offset > 0)
1340 bb_offset = bitmap_offset + bm_sectors;
1341 while (bb_offset < (long)sb_offset + 8 + 32*2 &&
1342 bb_offset + 8+8 <= (long)data_offset)
1343 /* too close to bitmap, and room to grow */
1344 bb_offset += 8;
1345 if (bb_offset + 8 <= (long)data_offset) {
1346 sb->bblog_size = __cpu_to_le16(8);
1347 sb->bblog_offset = __cpu_to_le32(bb_offset);
1348 }
1349 } else {
1350 /* 1.0 - Put bbl just before super block */
1351 if (bm_sectors && bitmap_offset < 0)
1352 space = -bitmap_offset - bm_sectors;
1353 else
1354 space = sb_offset - data_offset -
1355 __le64_to_cpu(sb->data_size);
1356 if (space >= 8) {
1357 sb->bblog_size = __cpu_to_le16(8);
1358 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1359 }
1360 }
1361 } else if (strcmp(update, "no-bbl") == 0) {
1362 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
1363 pr_err("Cannot remove active bbl from %s\n",devname);
1364 else {
1365 sb->bblog_size = 0;
1366 sb->bblog_shift = 0;
1367 sb->bblog_offset = 0;
1368 }
1369 } else if (strcmp(update, "force-no-bbl") == 0) {
1370 sb->feature_map &= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1371 sb->bblog_size = 0;
1372 sb->bblog_shift = 0;
1373 sb->bblog_offset = 0;
1374 } else if (strcmp(update, "ppl") == 0) {
1375 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1376 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1377 unsigned long long data_size = __le64_to_cpu(sb->data_size);
1378 long bb_offset = __le32_to_cpu(sb->bblog_offset);
1379 int space;
1380 int optimal_space;
1381 int offset;
1382
1383 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1384 pr_err("Cannot add PPL to array with bitmap\n");
1385 return -2;
1386 }
1387
1388 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_JOURNAL)) {
1389 pr_err("Cannot add PPL to array with journal\n");
1390 return -2;
1391 }
1392
1393 if (sb_offset < data_offset) {
1394 if (bb_offset)
1395 space = bb_offset - 8;
1396 else
1397 space = data_offset - sb_offset - 8;
1398 offset = 8;
1399 } else {
1400 offset = -(sb_offset - data_offset - data_size);
1401 if (offset < INT16_MIN)
1402 offset = INT16_MIN;
1403 space = -(offset - bb_offset);
1404 }
1405
1406 if (space < (PPL_HEADER_SIZE >> 9) + 8) {
1407 pr_err("Not enough space to add ppl\n");
1408 return -2;
1409 }
1410
1411 optimal_space = choose_ppl_space(__le32_to_cpu(sb->chunksize));
1412
1413 if (space > optimal_space)
1414 space = optimal_space;
1415 if (space > UINT16_MAX)
1416 space = UINT16_MAX;
1417
1418 sb->ppl.offset = __cpu_to_le16(offset);
1419 sb->ppl.size = __cpu_to_le16(space);
1420 sb->feature_map |= __cpu_to_le32(MD_FEATURE_PPL);
1421 } else if (strcmp(update, "no-ppl") == 0) {
1422 sb->feature_map &= ~ __cpu_to_le32(MD_FEATURE_PPL);
1423 } else if (strcmp(update, "name") == 0) {
1424 if (info->name[0] == 0)
1425 sprintf(info->name, "%d", info->array.md_minor);
1426 memset(sb->set_name, 0, sizeof(sb->set_name));
1427 if (homehost &&
1428 strchr(info->name, ':') == NULL &&
1429 strlen(homehost)+1+strlen(info->name) < 32) {
1430 strcpy(sb->set_name, homehost);
1431 strcat(sb->set_name, ":");
1432 strcat(sb->set_name, info->name);
1433 } else
1434 strncpy(sb->set_name, info->name, sizeof(sb->set_name));
1435 } else if (strcmp(update, "devicesize") == 0 &&
1436 __le64_to_cpu(sb->super_offset) <
1437 __le64_to_cpu(sb->data_offset)) {
1438 /* set data_size to device size less data_offset */
1439 struct misc_dev_info *misc = (struct misc_dev_info*)
1440 (st->sb + MAX_SB_SIZE + BM_SUPER_SIZE);
1441 sb->data_size = __cpu_to_le64(
1442 misc->device_size - __le64_to_cpu(sb->data_offset));
1443 } else if (strncmp(update, "revert-reshape", 14) == 0) {
1444 rv = -2;
1445 if (!(sb->feature_map &
1446 __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)))
1447 pr_err("No active reshape to revert on %s\n",
1448 devname);
1449 else {
1450 __u32 temp;
1451 unsigned long long reshape_sectors;
1452 long reshape_chunk;
1453 rv = 0;
1454 /* If the reshape hasn't started, just stop it.
1455 * It is conceivable that a stripe was modified but
1456 * the metadata not updated. In that case the backup
1457 * should have been used to get passed the critical stage.
1458 * If that couldn't happen, the "-nobackup" version
1459 * will be used.
1460 */
1461 if (strcmp(update, "revert-reshape-nobackup") == 0 &&
1462 sb->reshape_position == 0 &&
1463 (__le32_to_cpu(sb->delta_disks) > 0 ||
1464 (__le32_to_cpu(sb->delta_disks) == 0 &&
1465 !(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS))))) {
1466 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1467 sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1468 __le32_to_cpu(sb->delta_disks));
1469 sb->delta_disks = 0;
1470 goto done;
1471 }
1472 /* reshape_position is a little messy.
1473 * Its value must be a multiple of the larger
1474 * chunk size, and of the "after" data disks.
1475 * So when reverting we need to change it to
1476 * be a multiple of the new "after" data disks,
1477 * which is the old "before".
1478 * If it isn't already a multiple of 'before',
1479 * the only thing we could do would be
1480 * copy some block around on the disks, which
1481 * is easy to get wrong.
1482 * So we reject a revert-reshape unless the
1483 * alignment is good.
1484 */
1485 if (__le32_to_cpu(sb->level) >= 4 &&
1486 __le32_to_cpu(sb->level) <= 6) {
1487 reshape_sectors =
1488 __le64_to_cpu(sb->reshape_position);
1489 reshape_chunk = __le32_to_cpu(sb->new_chunk);
1490 reshape_chunk *= __le32_to_cpu(sb->raid_disks) -
1491 __le32_to_cpu(sb->delta_disks) -
1492 (__le32_to_cpu(sb->level)==6 ? 2 : 1);
1493 if (reshape_sectors % reshape_chunk) {
1494 pr_err("Reshape position is not suitably aligned.\n");
1495 pr_err("Try normal assembly and stop again\n");
1496 return -2;
1497 }
1498 }
1499 sb->raid_disks =
1500 __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1501 __le32_to_cpu(sb->delta_disks));
1502 if (sb->delta_disks == 0)
1503 sb->feature_map ^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1504 else
1505 sb->delta_disks = __cpu_to_le32(-__le32_to_cpu(sb->delta_disks));
1506
1507 temp = sb->new_layout;
1508 sb->new_layout = sb->layout;
1509 sb->layout = temp;
1510
1511 temp = sb->new_chunk;
1512 sb->new_chunk = sb->chunksize;
1513 sb->chunksize = temp;
1514
1515 if (sb->feature_map &
1516 __cpu_to_le32(MD_FEATURE_NEW_OFFSET)) {
1517 long offset_delta =
1518 (int32_t)__le32_to_cpu(sb->new_offset);
1519 sb->data_offset = __cpu_to_le64(__le64_to_cpu(sb->data_offset) + offset_delta);
1520 sb->new_offset = __cpu_to_le32(-offset_delta);
1521 sb->data_size = __cpu_to_le64(__le64_to_cpu(sb->data_size) - offset_delta);
1522 }
1523 done:;
1524 }
1525 } else if (strcmp(update, "_reshape_progress") == 0)
1526 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1527 else if (strcmp(update, "writemostly") == 0)
1528 sb->devflags |= WriteMostly1;
1529 else if (strcmp(update, "readwrite") == 0)
1530 sb->devflags &= ~WriteMostly1;
1531 else if (strcmp(update, "failfast") == 0)
1532 sb->devflags |= FailFast1;
1533 else if (strcmp(update, "nofailfast") == 0)
1534 sb->devflags &= ~FailFast1;
1535 else
1536 rv = -1;
1537
1538 sb->sb_csum = calc_sb_1_csum(sb);
1539 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1540 cluster_release_dlmlock(lockid);
1541
1542 return rv;
1543 }
1544
1545 static int init_super1(struct supertype *st, mdu_array_info_t *info,
1546 struct shape *s, char *name, char *homehost,
1547 int *uuid, unsigned long long data_offset)
1548 {
1549 struct mdp_superblock_1 *sb;
1550 int spares;
1551 char defname[10];
1552 int sbsize;
1553
1554 if (posix_memalign((void**)&sb, 4096, SUPER1_SIZE) != 0) {
1555 pr_err("could not allocate superblock\n");
1556 return 0;
1557 }
1558 memset(sb, 0, SUPER1_SIZE);
1559
1560 st->sb = sb;
1561 if (info == NULL) {
1562 /* zeroing superblock */
1563 return 0;
1564 }
1565
1566 spares = info->working_disks - info->active_disks;
1567 if (info->raid_disks + spares > MAX_DEVS) {
1568 pr_err("too many devices requested: %d+%d > %d\n",
1569 info->raid_disks , spares, MAX_DEVS);
1570 return 0;
1571 }
1572
1573 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
1574 sb->major_version = __cpu_to_le32(1);
1575 sb->feature_map = 0;
1576 sb->pad0 = 0;
1577
1578 if (uuid)
1579 copy_uuid(sb->set_uuid, uuid, super1.swapuuid);
1580 else
1581 random_uuid(sb->set_uuid);;
1582
1583 if (name == NULL || *name == 0) {
1584 sprintf(defname, "%d", info->md_minor);
1585 name = defname;
1586 }
1587 if (homehost &&
1588 strchr(name, ':')== NULL &&
1589 strlen(homehost)+1+strlen(name) < 32) {
1590 strcpy(sb->set_name, homehost);
1591 strcat(sb->set_name, ":");
1592 strcat(sb->set_name, name);
1593 } else
1594 strncpy(sb->set_name, name, sizeof(sb->set_name));
1595
1596 sb->ctime = __cpu_to_le64((unsigned long long)time(0));
1597 sb->level = __cpu_to_le32(info->level);
1598 sb->layout = __cpu_to_le32(info->layout);
1599 sb->size = __cpu_to_le64(s->size*2ULL);
1600 sb->chunksize = __cpu_to_le32(info->chunk_size>>9);
1601 sb->raid_disks = __cpu_to_le32(info->raid_disks);
1602
1603 sb->data_offset = __cpu_to_le64(data_offset);
1604 sb->data_size = __cpu_to_le64(0);
1605 sb->super_offset = __cpu_to_le64(0);
1606 sb->recovery_offset = __cpu_to_le64(0);
1607
1608 sb->utime = sb->ctime;
1609 sb->events = __cpu_to_le64(1);
1610 if (info->state & (1<<MD_SB_CLEAN))
1611 sb->resync_offset = MaxSector;
1612 else
1613 sb->resync_offset = 0;
1614 sbsize = sizeof(struct mdp_superblock_1) +
1615 2 * (info->raid_disks + spares);
1616 sbsize = ROUND_UP(sbsize, 512);
1617 sb->max_dev =
1618 __cpu_to_le32((sbsize - sizeof(struct mdp_superblock_1)) / 2);
1619
1620 memset(sb->dev_roles, 0xff,
1621 MAX_SB_SIZE - sizeof(struct mdp_superblock_1));
1622
1623 if (s->consistency_policy == CONSISTENCY_POLICY_PPL)
1624 sb->feature_map |= __cpu_to_le32(MD_FEATURE_PPL);
1625
1626 return 1;
1627 }
1628
1629 struct devinfo {
1630 int fd;
1631 char *devname;
1632 long long data_offset;
1633 mdu_disk_info_t disk;
1634 struct devinfo *next;
1635 };
1636
1637 /* Add a device to the superblock being created */
1638 static int add_to_super1(struct supertype *st, mdu_disk_info_t *dk,
1639 int fd, char *devname, unsigned long long data_offset)
1640 {
1641 struct mdp_superblock_1 *sb = st->sb;
1642 __u16 *rp = sb->dev_roles + dk->number;
1643 struct devinfo *di, **dip;
1644 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1645 int rv, lockid;
1646 int dk_state;
1647
1648 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1649 rv = cluster_get_dlmlock(&lockid);
1650 if (rv) {
1651 pr_err("Cannot get dlmlock in %s return %d\n",
1652 __func__, rv);
1653 cluster_release_dlmlock(lockid);
1654 return rv;
1655 }
1656 }
1657
1658 dk_state = dk->state & ~(1<<MD_DISK_FAILFAST);
1659 if ((dk_state & (1<<MD_DISK_ACTIVE)) &&
1660 (dk_state & (1<<MD_DISK_SYNC)))/* active, sync */
1661 *rp = __cpu_to_le16(dk->raid_disk);
1662 else if (dk_state & (1<<MD_DISK_JOURNAL))
1663 *rp = MD_DISK_ROLE_JOURNAL;
1664 else if ((dk_state & ~(1<<MD_DISK_ACTIVE)) == 0)
1665 /* active or idle -> spare */
1666 *rp = MD_DISK_ROLE_SPARE;
1667 else
1668 *rp = MD_DISK_ROLE_FAULTY;
1669
1670 if (dk->number >= (int)__le32_to_cpu(sb->max_dev) &&
1671 __le32_to_cpu(sb->max_dev) < MAX_DEVS)
1672 sb->max_dev = __cpu_to_le32(dk->number+1);
1673
1674 sb->dev_number = __cpu_to_le32(dk->number);
1675 sb->devflags = 0; /* don't copy another disks flags */
1676 sb->sb_csum = calc_sb_1_csum(sb);
1677
1678 dip = (struct devinfo **)&st->info;
1679 while (*dip)
1680 dip = &(*dip)->next;
1681 di = xmalloc(sizeof(struct devinfo));
1682 di->fd = fd;
1683 di->devname = devname;
1684 di->disk = *dk;
1685 di->data_offset = data_offset;
1686 di->next = NULL;
1687 *dip = di;
1688
1689 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1690 cluster_release_dlmlock(lockid);
1691
1692 return 0;
1693 }
1694
1695 static int locate_bitmap1(struct supertype *st, int fd, int node_num);
1696
1697 static int store_super1(struct supertype *st, int fd)
1698 {
1699 struct mdp_superblock_1 *sb = st->sb;
1700 unsigned long long sb_offset;
1701 struct align_fd afd;
1702 int sbsize;
1703 unsigned long long dsize;
1704 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1705 int rv, lockid;
1706
1707 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1708 rv = cluster_get_dlmlock(&lockid);
1709 if (rv) {
1710 pr_err("Cannot get dlmlock in %s return %d\n",
1711 __func__, rv);
1712 cluster_release_dlmlock(lockid);
1713 return rv;
1714 }
1715 }
1716
1717 if (!get_dev_size(fd, NULL, &dsize))
1718 return 1;
1719
1720 dsize >>= 9;
1721
1722 if (dsize < 24)
1723 return 2;
1724
1725 init_afd(&afd, fd);
1726
1727 /*
1728 * Calculate the position of the superblock.
1729 * It is always aligned to a 4K boundary and
1730 * depending on minor_version, it can be:
1731 * 0: At least 8K, but less than 12K, from end of device
1732 * 1: At start of device
1733 * 2: 4K from start of device.
1734 */
1735 switch(st->minor_version) {
1736 case 0:
1737 sb_offset = dsize;
1738 sb_offset -= 8*2;
1739 sb_offset &= ~(4*2-1);
1740 break;
1741 case 1:
1742 sb_offset = 0;
1743 break;
1744 case 2:
1745 sb_offset = 4*2;
1746 break;
1747 default:
1748 return -EINVAL;
1749 }
1750
1751 if (sb_offset != __le64_to_cpu(sb->super_offset) &&
1752 0 != __le64_to_cpu(sb->super_offset)
1753 ) {
1754 pr_err("internal error - sb_offset is wrong\n");
1755 abort();
1756 }
1757
1758 if (lseek64(fd, sb_offset << 9, 0)< 0LL)
1759 return 3;
1760
1761 sbsize = ROUND_UP(sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev), 512);
1762
1763 if (awrite(&afd, sb, sbsize) != sbsize)
1764 return 4;
1765
1766 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1767 struct bitmap_super_s *bm = (struct bitmap_super_s*)
1768 (((char*)sb)+MAX_SB_SIZE);
1769 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC) {
1770 locate_bitmap1(st, fd, 0);
1771 if (awrite(&afd, bm, sizeof(*bm)) != sizeof(*bm))
1772 return 5;
1773 }
1774 }
1775 fsync(fd);
1776 if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1777 cluster_release_dlmlock(lockid);
1778
1779 return 0;
1780 }
1781
1782 static int load_super1(struct supertype *st, int fd, char *devname);
1783
1784 static unsigned long choose_bm_space(unsigned long devsize)
1785 {
1786 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1787 * if bigger than 200Gig, save 128k
1788 * NOTE: result must be multiple of 4K else bad things happen
1789 * on 4K-sector devices.
1790 */
1791 if (devsize < 64*2)
1792 return 0;
1793 if (devsize - 64*2 >= 200*1024*1024*2)
1794 return 128*2;
1795 if (devsize - 4*2 > 8*1024*1024*2)
1796 return 64*2;
1797 return 4*2;
1798 }
1799
1800 static void free_super1(struct supertype *st);
1801
1802 __u32 crc32c_le(__u32 crc, unsigned char const *p, size_t len);
1803
1804 static int write_init_ppl1(struct supertype *st, struct mdinfo *info, int fd)
1805 {
1806 struct mdp_superblock_1 *sb = st->sb;
1807 void *buf;
1808 struct ppl_header *ppl_hdr;
1809 int ret;
1810
1811 ret = posix_memalign(&buf, 4096, PPL_HEADER_SIZE);
1812 if (ret) {
1813 pr_err("Failed to allocate PPL header buffer\n");
1814 return ret;
1815 }
1816
1817 memset(buf, 0, PPL_HEADER_SIZE);
1818 ppl_hdr = buf;
1819 memset(ppl_hdr->reserved, 0xff, PPL_HDR_RESERVED);
1820 ppl_hdr->signature = __cpu_to_le32(~crc32c_le(~0, sb->set_uuid,
1821 sizeof(sb->set_uuid)));
1822 ppl_hdr->checksum = __cpu_to_le32(~crc32c_le(~0, buf, PPL_HEADER_SIZE));
1823
1824 if (lseek64(fd, info->ppl_sector * 512, SEEK_SET) < 0) {
1825 ret = errno;
1826 perror("Failed to seek to PPL header location");
1827 }
1828
1829 if (!ret && write(fd, buf, PPL_HEADER_SIZE) != PPL_HEADER_SIZE) {
1830 ret = errno;
1831 perror("Write PPL header failed");
1832 }
1833
1834 if (!ret)
1835 fsync(fd);
1836
1837 free(buf);
1838 return ret;
1839 }
1840
1841 #define META_BLOCK_SIZE 4096
1842
1843 static int write_empty_r5l_meta_block(struct supertype *st, int fd)
1844 {
1845 struct r5l_meta_block *mb;
1846 struct mdp_superblock_1 *sb = st->sb;
1847 struct align_fd afd;
1848 __u32 crc;
1849
1850 init_afd(&afd, fd);
1851
1852 if (posix_memalign((void**)&mb, 4096, META_BLOCK_SIZE) != 0) {
1853 pr_err("Could not allocate memory for the meta block.\n");
1854 return 1;
1855 }
1856
1857 memset(mb, 0, META_BLOCK_SIZE);
1858
1859 mb->magic = __cpu_to_le32(R5LOG_MAGIC);
1860 mb->version = R5LOG_VERSION;
1861 mb->meta_size = __cpu_to_le32(sizeof(struct r5l_meta_block));
1862 mb->seq = __cpu_to_le64(random32());
1863 mb->position = __cpu_to_le64(0);
1864
1865 crc = crc32c_le(0xffffffff, sb->set_uuid, sizeof(sb->set_uuid));
1866 crc = crc32c_le(crc, (void *)mb, META_BLOCK_SIZE);
1867 mb->checksum = crc;
1868
1869 if (lseek64(fd, __le64_to_cpu(sb->data_offset) * 512, 0) < 0LL) {
1870 pr_err("cannot seek to offset of the meta block\n");
1871 goto fail_to_write;
1872 }
1873
1874 if (awrite(&afd, mb, META_BLOCK_SIZE) != META_BLOCK_SIZE) {
1875 pr_err("failed to store write the meta block \n");
1876 goto fail_to_write;
1877 }
1878 fsync(fd);
1879
1880 free(mb);
1881 return 0;
1882
1883 fail_to_write:
1884 free(mb);
1885 return 1;
1886 }
1887
1888 static int write_init_super1(struct supertype *st)
1889 {
1890 struct mdp_superblock_1 *sb = st->sb;
1891 struct supertype *refst;
1892 int rv = 0;
1893 unsigned long long bm_space;
1894 struct devinfo *di;
1895 unsigned long long dsize, array_size;
1896 unsigned long long sb_offset;
1897 unsigned long long data_offset;
1898 long bm_offset;
1899
1900 for (di = st->info; di; di = di->next) {
1901 if (di->disk.state & (1 << MD_DISK_JOURNAL))
1902 sb->feature_map |= __cpu_to_le32(MD_FEATURE_JOURNAL);
1903 }
1904
1905 for (di = st->info; di; di = di->next) {
1906 if (di->disk.state & (1 << MD_DISK_FAULTY))
1907 continue;
1908 if (di->fd < 0)
1909 continue;
1910
1911 while (Kill(di->devname, NULL, 0, -1, 1) == 0)
1912 ;
1913
1914 sb->dev_number = __cpu_to_le32(di->disk.number);
1915 if (di->disk.state & (1<<MD_DISK_WRITEMOSTLY))
1916 sb->devflags |= WriteMostly1;
1917 else
1918 sb->devflags &= ~WriteMostly1;
1919 if (di->disk.state & (1<<MD_DISK_FAILFAST))
1920 sb->devflags |= FailFast1;
1921 else
1922 sb->devflags &= ~FailFast1;
1923
1924 random_uuid(sb->device_uuid);
1925
1926 if (!(di->disk.state & (1<<MD_DISK_JOURNAL)))
1927 sb->events = 0;
1928
1929 refst = dup_super(st);
1930 if (load_super1(refst, di->fd, NULL)==0) {
1931 struct mdp_superblock_1 *refsb = refst->sb;
1932
1933 memcpy(sb->device_uuid, refsb->device_uuid, 16);
1934 if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) {
1935 /* same array, so preserve events and
1936 * dev_number */
1937 sb->events = refsb->events;
1938 /* bugs in 2.6.17 and earlier mean the
1939 * dev_number chosen in Manage must be preserved
1940 */
1941 if (get_linux_version() >= 2006018)
1942 sb->dev_number = refsb->dev_number;
1943 }
1944 free_super1(refst);
1945 }
1946 free(refst);
1947
1948 if (!get_dev_size(di->fd, NULL, &dsize)) {
1949 rv = 1;
1950 goto error_out;
1951 }
1952 dsize >>= 9;
1953
1954 if (dsize < 24) {
1955 close(di->fd);
1956 rv = 2;
1957 goto error_out;
1958 }
1959
1960 /*
1961 * Calculate the position of the superblock.
1962 * It is always aligned to a 4K boundary and
1963 * depending on minor_version, it can be:
1964 * 0: At least 8K, but less than 12K, from end of device
1965 * 1: At start of device
1966 * 2: 4K from start of device.
1967 * data_offset has already been set.
1968 */
1969 array_size = __le64_to_cpu(sb->size);
1970
1971 /* work out how much space we left for a bitmap */
1972 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1973 bitmap_super_t *bms = (bitmap_super_t *)
1974 (((char *)sb) + MAX_SB_SIZE);
1975 bm_space = calc_bitmap_size(bms, 4096) >> 9;
1976 bm_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1977 } else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_PPL)) {
1978 bm_space =
1979 choose_ppl_space(__le32_to_cpu(sb->chunksize));
1980 if (bm_space > UINT16_MAX)
1981 bm_space = UINT16_MAX;
1982 if (st->minor_version == 0) {
1983 bm_offset = -bm_space - 8;
1984 if (bm_offset < INT16_MIN) {
1985 bm_offset = INT16_MIN;
1986 bm_space = -bm_offset - 8;
1987 }
1988 } else {
1989 bm_offset = 8;
1990 }
1991 sb->ppl.offset = __cpu_to_le16(bm_offset);
1992 sb->ppl.size = __cpu_to_le16(bm_space);
1993 } else {
1994 bm_space = choose_bm_space(array_size);
1995 bm_offset = 8;
1996 }
1997
1998 data_offset = di->data_offset;
1999 if (data_offset == INVALID_SECTORS)
2000 data_offset = st->data_offset;
2001 switch(st->minor_version) {
2002 case 0:
2003 /* Add 8 sectors for bad block log */
2004 bm_space += 8;
2005 if (data_offset == INVALID_SECTORS)
2006 data_offset = 0;
2007 sb_offset = dsize;
2008 sb_offset -= 8*2;
2009 sb_offset &= ~(4*2-1);
2010 sb->data_offset = __cpu_to_le64(data_offset);
2011 sb->super_offset = __cpu_to_le64(sb_offset);
2012 if (sb_offset < array_size + bm_space)
2013 bm_space = sb_offset - array_size;
2014 sb->data_size = __cpu_to_le64(sb_offset - bm_space);
2015 if (bm_space >= 8) {
2016 sb->bblog_size = __cpu_to_le16(8);
2017 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
2018 }
2019 break;
2020 case 1:
2021 case 2:
2022 sb_offset = st->minor_version == 2 ? 8 : 0;
2023 sb->super_offset = __cpu_to_le64(sb_offset);
2024 if (data_offset == INVALID_SECTORS)
2025 data_offset = sb_offset + 16;
2026
2027 sb->data_offset = __cpu_to_le64(data_offset);
2028 sb->data_size = __cpu_to_le64(dsize - data_offset);
2029 if (data_offset >= sb_offset+bm_offset+bm_space+8) {
2030 sb->bblog_size = __cpu_to_le16(8);
2031 sb->bblog_offset = __cpu_to_le32(bm_offset +
2032 bm_space);
2033 } else if (data_offset >= sb_offset + 16) {
2034 sb->bblog_size = __cpu_to_le16(8);
2035 /* '8' sectors for the bblog, and 'sb_offset'
2036 * because we want offset from superblock, not
2037 * start of device.
2038 */
2039 sb->bblog_offset = __cpu_to_le32(data_offset -
2040 8 - sb_offset);
2041 }
2042 break;
2043 default:
2044 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
2045 st->minor_version, di->devname);
2046 rv = -EINVAL;
2047 goto out;
2048 }
2049 /*
2050 * Disable badblock log on clusters, or when
2051 * explicitly requested
2052 */
2053 if (st->nodes > 0 || conf_get_create_info()->bblist == 0) {
2054 sb->bblog_size = 0;
2055 sb->bblog_offset = 0;
2056 }
2057
2058 sb->sb_csum = calc_sb_1_csum(sb);
2059 rv = store_super1(st, di->fd);
2060
2061 if (rv == 0 && (di->disk.state & (1 << MD_DISK_JOURNAL))) {
2062 rv = write_empty_r5l_meta_block(st, di->fd);
2063 if (rv)
2064 goto error_out;
2065 }
2066
2067 if (rv == 0 &&
2068 (__le32_to_cpu(sb->feature_map) &
2069 MD_FEATURE_BITMAP_OFFSET)) {
2070 rv = st->ss->write_bitmap(st, di->fd, NodeNumUpdate);
2071 } else if (rv == 0 &&
2072 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL)) {
2073 struct mdinfo info;
2074
2075 st->ss->getinfo_super(st, &info, NULL);
2076 rv = st->ss->write_init_ppl(st, &info, di->fd);
2077 }
2078
2079 close(di->fd);
2080 di->fd = -1;
2081 if (rv)
2082 goto error_out;
2083 }
2084 error_out:
2085 if (rv)
2086 pr_err("Failed to write metadata to %s\n", di->devname);
2087 out:
2088 return rv;
2089 }
2090
2091 static int compare_super1(struct supertype *st, struct supertype *tst)
2092 {
2093 /*
2094 * return:
2095 * 0 same, or first was empty, and second was copied
2096 * 1 second had wrong number
2097 * 2 wrong uuid
2098 * 3 wrong other info
2099 */
2100 struct mdp_superblock_1 *first = st->sb;
2101 struct mdp_superblock_1 *second = tst->sb;
2102
2103 if (second->magic != __cpu_to_le32(MD_SB_MAGIC))
2104 return 1;
2105 if (second->major_version != __cpu_to_le32(1))
2106 return 1;
2107
2108 if (!first) {
2109 if (posix_memalign((void**)&first, 4096, SUPER1_SIZE) != 0) {
2110 pr_err("could not allocate superblock\n");
2111 return 1;
2112 }
2113 memcpy(first, second, SUPER1_SIZE);
2114 st->sb = first;
2115 return 0;
2116 }
2117 if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0)
2118 return 2;
2119
2120 if (first->ctime != second->ctime ||
2121 first->level != second->level ||
2122 first->layout != second->layout ||
2123 first->size != second->size ||
2124 first->chunksize != second->chunksize ||
2125 first->raid_disks != second->raid_disks)
2126 return 3;
2127 return 0;
2128 }
2129
2130 static int load_super1(struct supertype *st, int fd, char *devname)
2131 {
2132 unsigned long long dsize;
2133 unsigned long long sb_offset;
2134 struct mdp_superblock_1 *super;
2135 int uuid[4];
2136 struct bitmap_super_s *bsb;
2137 struct misc_dev_info *misc;
2138 struct align_fd afd;
2139
2140 free_super1(st);
2141
2142 init_afd(&afd, fd);
2143
2144 if (st->ss == NULL || st->minor_version == -1) {
2145 int bestvers = -1;
2146 struct supertype tst;
2147 __u64 bestctime = 0;
2148 /* guess... choose latest ctime */
2149 memset(&tst, 0, sizeof(tst));
2150 tst.ss = &super1;
2151 for (tst.minor_version = 0; tst.minor_version <= 2;
2152 tst.minor_version++) {
2153 switch(load_super1(&tst, fd, devname)) {
2154 case 0: super = tst.sb;
2155 if (bestvers == -1 ||
2156 bestctime < __le64_to_cpu(super->ctime)) {
2157 bestvers = tst.minor_version;
2158 bestctime = __le64_to_cpu(super->ctime);
2159 }
2160 free(super);
2161 tst.sb = NULL;
2162 break;
2163 case 1: return 1; /*bad device */
2164 case 2: break; /* bad, try next */
2165 }
2166 }
2167 if (bestvers != -1) {
2168 int rv;
2169 tst.minor_version = bestvers;
2170 tst.ss = &super1;
2171 tst.max_devs = MAX_DEVS;
2172 rv = load_super1(&tst, fd, devname);
2173 if (rv == 0)
2174 *st = tst;
2175 return rv;
2176 }
2177 return 2;
2178 }
2179 if (!get_dev_size(fd, devname, &dsize))
2180 return 1;
2181 dsize >>= 9;
2182
2183 if (dsize < 24) {
2184 if (devname)
2185 pr_err("%s is too small for md: size is %llu sectors.\n",
2186 devname, dsize);
2187 return 1;
2188 }
2189
2190 /*
2191 * Calculate the position of the superblock.
2192 * It is always aligned to a 4K boundary and
2193 * depending on minor_version, it can be:
2194 * 0: At least 8K, but less than 12K, from end of device
2195 * 1: At start of device
2196 * 2: 4K from start of device.
2197 */
2198 switch(st->minor_version) {
2199 case 0:
2200 sb_offset = dsize;
2201 sb_offset -= 8*2;
2202 sb_offset &= ~(4*2-1);
2203 break;
2204 case 1:
2205 sb_offset = 0;
2206 break;
2207 case 2:
2208 sb_offset = 4*2;
2209 break;
2210 default:
2211 return -EINVAL;
2212 }
2213
2214 if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
2215 if (devname)
2216 pr_err("Cannot seek to superblock on %s: %s\n",
2217 devname, strerror(errno));
2218 return 1;
2219 }
2220
2221 if (posix_memalign((void**)&super, 4096, SUPER1_SIZE) != 0) {
2222 pr_err("could not allocate superblock\n");
2223 return 1;
2224 }
2225
2226 memset(super, 0, SUPER1_SIZE);
2227
2228 if (aread(&afd, super, MAX_SB_SIZE) != MAX_SB_SIZE) {
2229 if (devname)
2230 pr_err("Cannot read superblock on %s\n",
2231 devname);
2232 free(super);
2233 return 1;
2234 }
2235
2236 if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) {
2237 if (devname)
2238 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2239 devname, MD_SB_MAGIC,
2240 __le32_to_cpu(super->magic));
2241 free(super);
2242 return 2;
2243 }
2244
2245 if (__le32_to_cpu(super->major_version) != 1) {
2246 if (devname)
2247 pr_err("Cannot interpret superblock on %s - version is %d\n",
2248 devname, __le32_to_cpu(super->major_version));
2249 free(super);
2250 return 2;
2251 }
2252 if (__le64_to_cpu(super->super_offset) != sb_offset) {
2253 if (devname)
2254 pr_err("No superblock found on %s (super_offset is wrong)\n",
2255 devname);
2256 free(super);
2257 return 2;
2258 }
2259 st->sb = super;
2260
2261 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
2262
2263 misc = (struct misc_dev_info*)
2264 (((char*)super)+MAX_SB_SIZE+BM_SUPER_SIZE);
2265 misc->device_size = dsize;
2266 if (st->data_offset == INVALID_SECTORS)
2267 st->data_offset = __le64_to_cpu(super->data_offset);
2268
2269 /* Now check on the bitmap superblock */
2270 if ((__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) == 0)
2271 return 0;
2272 /* Read the bitmap superblock and make sure it looks
2273 * valid. If it doesn't clear the bit. An --assemble --force
2274 * should get that written out.
2275 */
2276 locate_bitmap1(st, fd, 0);
2277 if (aread(&afd, bsb, 512) != 512)
2278 goto no_bitmap;
2279
2280 uuid_from_super1(st, uuid);
2281 if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
2282 memcmp(bsb->uuid, uuid, 16) != 0)
2283 goto no_bitmap;
2284 return 0;
2285
2286 no_bitmap:
2287 super->feature_map = __cpu_to_le32(__le32_to_cpu(super->feature_map)
2288 & ~MD_FEATURE_BITMAP_OFFSET);
2289 return 0;
2290 }
2291
2292 static struct supertype *match_metadata_desc1(char *arg)
2293 {
2294 struct supertype *st = xcalloc(1, sizeof(*st));
2295
2296 st->container_devnm[0] = 0;
2297 st->ss = &super1;
2298 st->max_devs = MAX_DEVS;
2299 st->sb = NULL;
2300 st->data_offset = INVALID_SECTORS;
2301 /* leading zeros can be safely ignored. --detail generates them. */
2302 while (*arg == '0')
2303 arg++;
2304 if (strcmp(arg, "1.0") == 0 || strcmp(arg, "1.00") == 0) {
2305 st->minor_version = 0;
2306 return st;
2307 }
2308 if (strcmp(arg, "1.1") == 0 || strcmp(arg, "1.01") == 0
2309 ) {
2310 st->minor_version = 1;
2311 return st;
2312 }
2313 if (strcmp(arg, "1.2") == 0 ||
2314 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2315 strcmp(arg, "default") == 0 ||
2316 #endif /* DEFAULT_OLD_METADATA */
2317 strcmp(arg, "1.02") == 0) {
2318 st->minor_version = 2;
2319 return st;
2320 }
2321 if (strcmp(arg, "1") == 0 || strcmp(arg, "default") == 0) {
2322 st->minor_version = -1;
2323 return st;
2324 }
2325
2326 free(st);
2327 return NULL;
2328 }
2329
2330 /* find available size on device with this devsize, using
2331 * superblock type st, and reserving 'reserve' sectors for
2332 * a possible bitmap
2333 */
2334 static __u64 avail_size1(struct supertype *st, __u64 devsize,
2335 unsigned long long data_offset)
2336 {
2337 struct mdp_superblock_1 *super = st->sb;
2338 int bmspace = 0;
2339 int bbspace = 0;
2340 if (devsize < 24)
2341 return 0;
2342
2343 if (__le32_to_cpu(super->feature_map) & MD_FEATURE_BITMAP_OFFSET) {
2344 /* hot-add. allow for actual size of bitmap */
2345 struct bitmap_super_s *bsb;
2346 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
2347 bmspace = calc_bitmap_size(bsb, 4096) >> 9;
2348 } else if (__le32_to_cpu(super->feature_map) & MD_FEATURE_PPL) {
2349 bmspace = __le16_to_cpu(super->ppl.size);
2350 }
2351
2352 /* Allow space for bad block log */
2353 if (super->bblog_size)
2354 bbspace = __le16_to_cpu(super->bblog_size);
2355
2356 if (st->minor_version < 0)
2357 /* not specified, so time to set default */
2358 st->minor_version = 2;
2359
2360 if (data_offset == INVALID_SECTORS)
2361 data_offset = st->data_offset;
2362
2363 if (data_offset != INVALID_SECTORS)
2364 switch(st->minor_version) {
2365 case 0:
2366 return devsize - data_offset - 8*2 - bbspace;
2367 case 1:
2368 case 2:
2369 return devsize - data_offset;
2370 default:
2371 return 0;
2372 }
2373
2374 devsize -= bmspace;
2375
2376 switch(st->minor_version) {
2377 case 0:
2378 /* at end */
2379 return ((devsize - 8*2 - bbspace ) & ~(4*2-1));
2380 case 1:
2381 /* at start, 4K for superblock and possible bitmap */
2382 return devsize - 4*2 - bbspace;
2383 case 2:
2384 /* 4k from start, 4K for superblock and possible bitmap */
2385 return devsize - (4+4)*2 - bbspace;
2386 }
2387 return 0;
2388 }
2389
2390 static int
2391 add_internal_bitmap1(struct supertype *st,
2392 int *chunkp, int delay, int write_behind,
2393 unsigned long long size,
2394 int may_change, int major)
2395 {
2396 /*
2397 * If not may_change, then this is a 'Grow' without sysfs support for
2398 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2399 * If may_change, then this is create or a Grow with sysfs support,
2400 * and we can put the bitmap wherever we like.
2401 *
2402 * size is in sectors, chunk is in bytes !!!
2403 */
2404
2405 unsigned long long bits;
2406 unsigned long long max_bits;
2407 unsigned long long min_chunk;
2408 long offset;
2409 long bbl_offset, bbl_size;
2410 unsigned long long chunk = *chunkp;
2411 int room = 0;
2412 int creating = 0;
2413 int len;
2414 struct mdp_superblock_1 *sb = st->sb;
2415 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
2416 int uuid[4];
2417
2418 if (__le64_to_cpu(sb->data_size) == 0)
2419 /*
2420 * Must be creating the array, else data_size
2421 * would be non-zero
2422 */
2423 creating = 1;
2424 switch(st->minor_version) {
2425 case 0:
2426 /*
2427 * either 3K after the superblock (when hot-add),
2428 * or some amount of space before.
2429 */
2430 if (creating) {
2431 /*
2432 * We are creating array, so we *know* how much room has
2433 * been left.
2434 */
2435 offset = 0;
2436 bbl_size = 8;
2437 room =
2438 choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2439 } else {
2440 room = __le64_to_cpu(sb->super_offset)
2441 - __le64_to_cpu(sb->data_offset)
2442 - __le64_to_cpu(sb->data_size);
2443 bbl_size = __le16_to_cpu(sb->bblog_size);
2444 if (bbl_size < 8)
2445 bbl_size = 8;
2446 bbl_offset = (__s32)__le32_to_cpu(sb->bblog_offset);
2447 if (bbl_size < -bbl_offset)
2448 bbl_size = -bbl_offset;
2449
2450 if (!may_change ||
2451 (room < 3*2 && __le32_to_cpu(sb->max_dev) <= 384)) {
2452 room = 3*2;
2453 offset = 1*2;
2454 bbl_size = 0;
2455 } else {
2456 offset = 0; /* means movable offset */
2457 }
2458 }
2459 break;
2460 case 1:
2461 case 2: /* between superblock and data */
2462 if (creating) {
2463 offset = 4*2;
2464 bbl_size = 8;
2465 room =
2466 choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2467 } else {
2468 room = __le64_to_cpu(sb->data_offset)
2469 - __le64_to_cpu(sb->super_offset);
2470 bbl_size = __le16_to_cpu(sb->bblog_size);
2471 if (bbl_size)
2472 room =
2473 __le32_to_cpu(sb->bblog_offset) + bbl_size;
2474 else
2475 bbl_size = 8;
2476
2477 if (!may_change) {
2478 room -= 2; /* Leave 1K for superblock */
2479 offset = 2;
2480 bbl_size = 0;
2481 } else {
2482 room -= 4*2; /* leave 4K for superblock */
2483 offset = 4*2;
2484 }
2485 }
2486 break;
2487 default:
2488 return -ENOSPC;
2489 }
2490
2491 room -= bbl_size;
2492 if (chunk == UnSet && room > 128*2)
2493 /* Limit to 128K of bitmap when chunk size not requested */
2494 room = 128*2;
2495
2496 if (room <= 1)
2497 /* No room for a bitmap */
2498 return -ENOSPC;
2499
2500 max_bits = (room * 512 - sizeof(bitmap_super_t)) * 8;
2501
2502 min_chunk = 4096; /* sub-page chunks don't work yet.. */
2503 bits = (size*512)/min_chunk +1;
2504 while (bits > max_bits) {
2505 min_chunk *= 2;
2506 bits = (bits+1)/2;
2507 }
2508 if (chunk == UnSet) {
2509 /* For practical purpose, 64Meg is a good
2510 * default chunk size for internal bitmaps.
2511 */
2512 chunk = min_chunk;
2513 if (chunk < 64*1024*1024)
2514 chunk = 64*1024*1024;
2515 } else if (chunk < min_chunk)
2516 return -EINVAL; /* chunk size too small */
2517 if (chunk == 0) /* rounding problem */
2518 return -EINVAL;
2519
2520 if (offset == 0) {
2521 /* start bitmap on a 4K boundary with enough space for
2522 * the bitmap
2523 */
2524 bits = (size*512) / chunk + 1;
2525 room = ((bits+7)/8 + sizeof(bitmap_super_t) +4095)/4096;
2526 room *= 8; /* convert 4K blocks to sectors */
2527 offset = -room - bbl_size;
2528 }
2529
2530 sb->bitmap_offset = (int32_t)__cpu_to_le32(offset);
2531
2532 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map) |
2533 MD_FEATURE_BITMAP_OFFSET);
2534 memset(bms, 0, sizeof(*bms));
2535 bms->magic = __cpu_to_le32(BITMAP_MAGIC);
2536 bms->version = __cpu_to_le32(major);
2537 uuid_from_super1(st, uuid);
2538 memcpy(bms->uuid, uuid, 16);
2539 bms->chunksize = __cpu_to_le32(chunk);
2540 bms->daemon_sleep = __cpu_to_le32(delay);
2541 bms->sync_size = __cpu_to_le64(size);
2542 bms->write_behind = __cpu_to_le32(write_behind);
2543 bms->nodes = __cpu_to_le32(st->nodes);
2544 if (st->nodes)
2545 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map) |
2546 MD_FEATURE_BITMAP_VERSIONED);
2547 if (st->cluster_name) {
2548 len = sizeof(bms->cluster_name);
2549 strncpy((char *)bms->cluster_name, st->cluster_name, len);
2550 bms->cluster_name[len - 1] = '\0';
2551 }
2552
2553 *chunkp = chunk;
2554 return 0;
2555 }
2556
2557 static int locate_bitmap1(struct supertype *st, int fd, int node_num)
2558 {
2559 unsigned long long offset;
2560 struct mdp_superblock_1 *sb;
2561 int mustfree = 0;
2562 int ret;
2563
2564 if (!st->sb) {
2565 if (st->ss->load_super(st, fd, NULL))
2566 return -1; /* no error I hope... */
2567 mustfree = 1;
2568 }
2569 sb = st->sb;
2570
2571 if ((__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
2572 ret = 0;
2573 else
2574 ret = -1;
2575 offset = __le64_to_cpu(sb->super_offset);
2576 offset += (int32_t) __le32_to_cpu(sb->bitmap_offset) * (node_num + 1);
2577 if (mustfree)
2578 free(sb);
2579 lseek64(fd, offset<<9, 0);
2580 return ret;
2581 }
2582
2583 static int write_bitmap1(struct supertype *st, int fd, enum bitmap_update update)
2584 {
2585 struct mdp_superblock_1 *sb = st->sb;
2586 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
2587 int rv = 0;
2588 void *buf;
2589 int towrite, n, len;
2590 struct align_fd afd;
2591 unsigned int i = 0;
2592 unsigned long long total_bm_space, bm_space_per_node;
2593
2594 switch (update) {
2595 case NameUpdate:
2596 /* update cluster name */
2597 if (st->cluster_name) {
2598 len = sizeof(bms->cluster_name);
2599 memset((char *)bms->cluster_name, 0, len);
2600 strncpy((char *)bms->cluster_name,
2601 st->cluster_name, len);
2602 bms->cluster_name[len - 1] = '\0';
2603 }
2604 break;
2605 case NodeNumUpdate:
2606 /* cluster md only supports superblock 1.2 now */
2607 if (st->minor_version != 2 &&
2608 bms->version == BITMAP_MAJOR_CLUSTERED) {
2609 pr_err("Warning: cluster md only works with superblock 1.2\n");
2610 return -EINVAL;
2611 }
2612
2613 if (bms->version == BITMAP_MAJOR_CLUSTERED) {
2614 if (__cpu_to_le32(st->nodes) < bms->nodes) {
2615 /*
2616 * Since the nodes num is not increased, no
2617 * need to check the space enough or not,
2618 * just update bms->nodes
2619 */
2620 bms->nodes = __cpu_to_le32(st->nodes);
2621 break;
2622 }
2623 } else {
2624 /*
2625 * no need to change bms->nodes for other
2626 * bitmap types
2627 */
2628 if (st->nodes)
2629 pr_err("Warning: --nodes option is only suitable for clustered bitmap\n");
2630 break;
2631 }
2632
2633 /*
2634 * Each node has an independent bitmap, it is necessary to
2635 * calculate the space is enough or not, first get how many
2636 * bytes for the total bitmap
2637 */
2638 bm_space_per_node = calc_bitmap_size(bms, 4096);
2639
2640 total_bm_space = 512 * (__le64_to_cpu(sb->data_offset) -
2641 __le64_to_cpu(sb->super_offset));
2642 /* leave another 4k for superblock */
2643 total_bm_space = total_bm_space - 4096;
2644
2645 if (bm_space_per_node * st->nodes > total_bm_space) {
2646 pr_err("Warning: The max num of nodes can't exceed %llu\n",
2647 total_bm_space / bm_space_per_node);
2648 return -ENOMEM;
2649 }
2650
2651 bms->nodes = __cpu_to_le32(st->nodes);
2652 break;
2653 case NoUpdate:
2654 default:
2655 break;
2656 }
2657
2658 init_afd(&afd, fd);
2659
2660 locate_bitmap1(st, fd, 0);
2661
2662 if (posix_memalign(&buf, 4096, 4096))
2663 return -ENOMEM;
2664
2665 do {
2666 /* Only the bitmap[0] should resync
2667 * whole device on initial assembly
2668 */
2669 if (i)
2670 memset(buf, 0x00, 4096);
2671 else
2672 memset(buf, 0xff, 4096);
2673 memcpy(buf, (char *)bms, sizeof(bitmap_super_t));
2674
2675 /*
2676 * use 4096 boundary if bitmap_offset is aligned
2677 * with 8 sectors, then it should compatible with
2678 * older mdadm.
2679 */
2680 if (__le32_to_cpu(sb->bitmap_offset) & 7)
2681 towrite = calc_bitmap_size(bms, 512);
2682 else
2683 towrite = calc_bitmap_size(bms, 4096);
2684 while (towrite > 0) {
2685 n = towrite;
2686 if (n > 4096)
2687 n = 4096;
2688 n = awrite(&afd, buf, n);
2689 if (n > 0)
2690 towrite -= n;
2691 else
2692 break;
2693 if (i)
2694 memset(buf, 0x00, 4096);
2695 else
2696 memset(buf, 0xff, 4096);
2697 }
2698 fsync(fd);
2699 if (towrite) {
2700 rv = -2;
2701 break;
2702 }
2703 } while (++i < __le32_to_cpu(bms->nodes));
2704
2705 free(buf);
2706 return rv;
2707 }
2708
2709 static void free_super1(struct supertype *st)
2710 {
2711
2712 if (st->sb)
2713 free(st->sb);
2714 while (st->info) {
2715 struct devinfo *di = st->info;
2716 st->info = di->next;
2717 if (di->fd >= 0)
2718 close(di->fd);
2719 free(di);
2720 }
2721 st->sb = NULL;
2722 }
2723
2724 static int validate_geometry1(struct supertype *st, int level,
2725 int layout, int raiddisks,
2726 int *chunk, unsigned long long size,
2727 unsigned long long data_offset,
2728 char *subdev, unsigned long long *freesize,
2729 int consistency_policy, int verbose)
2730 {
2731 unsigned long long ldsize, devsize;
2732 int bmspace;
2733 unsigned long long headroom;
2734 int fd;
2735
2736 if (level == LEVEL_CONTAINER) {
2737 if (verbose)
2738 pr_err("1.x metadata does not support containers\n");
2739 return 0;
2740 }
2741 if (*chunk == UnSet)
2742 *chunk = DEFAULT_CHUNK;
2743
2744 if (!subdev)
2745 return 1;
2746
2747 if (st->minor_version < 0)
2748 /* not specified, so time to set default */
2749 st->minor_version = 2;
2750
2751 fd = open(subdev, O_RDONLY|O_EXCL, 0);
2752 if (fd < 0) {
2753 if (verbose)
2754 pr_err("super1.x cannot open %s: %s\n",
2755 subdev, strerror(errno));
2756 return 0;
2757 }
2758
2759 if (!get_dev_size(fd, subdev, &ldsize)) {
2760 close(fd);
2761 return 0;
2762 }
2763 close(fd);
2764
2765 devsize = ldsize >> 9;
2766 if (devsize < 24) {
2767 *freesize = 0;
2768 return 0;
2769 }
2770
2771 /* creating: allow suitable space for bitmap or PPL */
2772 bmspace = consistency_policy == CONSISTENCY_POLICY_PPL ?
2773 choose_ppl_space((*chunk)*2) : choose_bm_space(devsize);
2774
2775 if (data_offset == INVALID_SECTORS)
2776 data_offset = st->data_offset;
2777 if (data_offset == INVALID_SECTORS)
2778 switch (st->minor_version) {
2779 case 0:
2780 data_offset = 0;
2781 break;
2782 case 1:
2783 case 2:
2784 /* Choose data offset appropriate for this device
2785 * and use as default for whole array.
2786 * The data_offset must allow for bitmap space
2787 * and base metadata, should allow for some headroom
2788 * for reshape, and should be rounded to multiple
2789 * of 1M.
2790 * Headroom is limited to 128M, but aim for about 0.1%
2791 */
2792 headroom = 128*1024*2;
2793 while ((headroom << 10) > devsize &&
2794 (*chunk == 0 ||
2795 headroom / 2 >= ((unsigned)(*chunk)*2)*2))
2796 headroom >>= 1;
2797 data_offset = 12*2 + bmspace + headroom;
2798 #define ONE_MEG (2*1024)
2799 data_offset = ROUND_UP(data_offset, ONE_MEG);
2800 break;
2801 }
2802 if (st->data_offset == INVALID_SECTORS)
2803 st->data_offset = data_offset;
2804 switch(st->minor_version) {
2805 case 0: /* metadata at end. Round down and subtract space to reserve */
2806 devsize = (devsize & ~(4ULL*2-1));
2807 /* space for metadata, bblog, bitmap/ppl */
2808 devsize -= 8*2 + 8 + bmspace;
2809 break;
2810 case 1:
2811 case 2:
2812 devsize -= data_offset;
2813 break;
2814 }
2815 *freesize = devsize;
2816 return 1;
2817 }
2818
2819 void *super1_make_v0(struct supertype *st, struct mdinfo *info, mdp_super_t *sb0)
2820 {
2821 /* Create a v1.0 superblock based on 'info'*/
2822 void *ret;
2823 struct mdp_superblock_1 *sb;
2824 int i;
2825 unsigned long long offset;
2826
2827 if (posix_memalign(&ret, 4096, 1024) != 0)
2828 return NULL;
2829 sb = ret;
2830 memset(ret, 0, 1024);
2831 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
2832 sb->major_version = __cpu_to_le32(1);
2833
2834 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
2835 sprintf(sb->set_name, "%d", sb0->md_minor);
2836 sb->ctime = __cpu_to_le32(info->array.ctime+1);
2837 sb->level = __cpu_to_le32(info->array.level);
2838 sb->layout = __cpu_to_le32(info->array.layout);
2839 sb->size = __cpu_to_le64(info->component_size);
2840 sb->chunksize = __cpu_to_le32(info->array.chunk_size/512);
2841 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
2842 if (info->array.level > 0)
2843 sb->data_size = sb->size;
2844 else
2845 sb->data_size = st->ss->avail_size(st, st->devsize/512, 0);
2846 sb->resync_offset = MaxSector;
2847 sb->max_dev = __cpu_to_le32(MD_SB_DISKS);
2848 sb->dev_number = __cpu_to_le32(info->disk.number);
2849 sb->utime = __cpu_to_le64(info->array.utime);
2850
2851 offset = st->devsize/512 - 8*2;
2852 offset &= ~(4*2-1);
2853 sb->super_offset = __cpu_to_le64(offset);
2854 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2855
2856 random_uuid(sb->device_uuid);
2857
2858 for (i = 0; i < MD_SB_DISKS; i++) {
2859 int state = sb0->disks[i].state;
2860 sb->dev_roles[i] = MD_DISK_ROLE_SPARE;
2861 if ((state & (1<<MD_DISK_SYNC)) &&
2862 !(state & (1<<MD_DISK_FAULTY)))
2863 sb->dev_roles[i] = __cpu_to_le16(sb0->disks[i].raid_disk);
2864 }
2865 sb->sb_csum = calc_sb_1_csum(sb);
2866 return ret;
2867 }
2868
2869 struct superswitch super1 = {
2870 .examine_super = examine_super1,
2871 .brief_examine_super = brief_examine_super1,
2872 .export_examine_super = export_examine_super1,
2873 .detail_super = detail_super1,
2874 .brief_detail_super = brief_detail_super1,
2875 .export_detail_super = export_detail_super1,
2876 .write_init_super = write_init_super1,
2877 .validate_geometry = validate_geometry1,
2878 .add_to_super = add_to_super1,
2879 .examine_badblocks = examine_badblocks_super1,
2880 .copy_metadata = copy_metadata1,
2881 .write_init_ppl = write_init_ppl1,
2882 .match_home = match_home1,
2883 .uuid_from_super = uuid_from_super1,
2884 .getinfo_super = getinfo_super1,
2885 .container_content = container_content1,
2886 .update_super = update_super1,
2887 .init_super = init_super1,
2888 .store_super = store_super1,
2889 .compare_super = compare_super1,
2890 .load_super = load_super1,
2891 .match_metadata_desc = match_metadata_desc1,
2892 .avail_size = avail_size1,
2893 .add_internal_bitmap = add_internal_bitmap1,
2894 .locate_bitmap = locate_bitmap1,
2895 .write_bitmap = write_bitmap1,
2896 .free_super = free_super1,
2897 #if __BYTE_ORDER == BIG_ENDIAN
2898 .swapuuid = 0,
2899 #else
2900 .swapuuid = 1,
2901 #endif
2902 .name = "1.x",
2903 };
Unnamed repository; edit this file 'description' to name the repository.