]> git.ipfire.org Git - thirdparty/mdadm.git/blob - super1.c
projects / thirdparty / mdadm.git / blob
? search:


167f2cafd1cad9c6f9c46d69cf6cf937710ba6ab
[thirdparty/mdadm.git] / super1.c
1 /*
2 * mdadm - manage Linux "md" devices aka RAID arrays.
3 *
4 * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
5 *
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
11 *
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
16 *
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 *
21 * Author: Neil Brown
22 * Email: <neilb@suse.de>
23 */
24
25 #include <stddef.h>
26 #include "mdadm.h"
27 /*
28 * The version-1 superblock :
29 * All numeric fields are little-endian.
30 *
31 * total size: 256 bytes plus 2 per device.
32 * 1K allows 384 devices.
33 */
34 struct mdp_superblock_1 {
35 /* constant array information - 128 bytes */
36 __u32 magic; /* MD_SB_MAGIC: 0xa92b4efc - little endian */
37 __u32 major_version; /* 1 */
38 __u32 feature_map; /* 0 for now */
39 __u32 pad0; /* always set to 0 when writing */
40
41 __u8 set_uuid[16]; /* user-space generated. */
42 char set_name[32]; /* set and interpreted by user-space */
43
44 __u64 ctime; /* lo 40 bits are seconds, top 24 are microseconds or 0*/
45 __u32 level; /* -4 (multipath), -1 (linear), 0,1,4,5 */
46 __u32 layout; /* only for raid5 currently */
47 __u64 size; /* used size of component devices, in 512byte sectors */
48
49 __u32 chunksize; /* in 512byte sectors */
50 __u32 raid_disks;
51 __u32 bitmap_offset; /* sectors after start of superblock that bitmap starts
52 * NOTE: signed, so bitmap can be before superblock
53 * only meaningful of feature_map[0] is set.
54 */
55
56 /* These are only valid with feature bit '4' */
57 __u32 new_level; /* new level we are reshaping to */
58 __u64 reshape_position; /* next address in array-space for reshape */
59 __u32 delta_disks; /* change in number of raid_disks */
60 __u32 new_layout; /* new layout */
61 __u32 new_chunk; /* new chunk size (sectors) */
62 __u32 new_offset; /* signed number to add to data_offset in new
63 * layout. 0 == no-change. This can be
64 * different on each device in the array.
65 */
66
67 /* constant this-device information - 64 bytes */
68 __u64 data_offset; /* sector start of data, often 0 */
69 __u64 data_size; /* sectors in this device that can be used for data */
70 __u64 super_offset; /* sector start of this superblock */
71 __u64 recovery_offset;/* sectors before this offset (from data_offset) have been recovered */
72 __u32 dev_number; /* permanent identifier of this device - not role in raid */
73 __u32 cnt_corrected_read; /* number of read errors that were corrected by re-writing */
74 __u8 device_uuid[16]; /* user-space setable, ignored by kernel */
75 __u8 devflags; /* per-device flags. Only one defined...*/
76 #define WriteMostly1 1 /* mask for writemostly flag in above */
77 /* bad block log. If there are any bad blocks the feature flag is set.
78 * if offset and size are non-zero, that space is reserved and available.
79 */
80 __u8 bblog_shift; /* shift from sectors to block size for badblocklist */
81 __u16 bblog_size; /* number of sectors reserved for badblocklist */
82 __u32 bblog_offset; /* sector offset from superblock to bblog, signed */
83
84 /* array state information - 64 bytes */
85 __u64 utime; /* 40 bits second, 24 btes microseconds */
86 __u64 events; /* incremented when superblock updated */
87 __u64 resync_offset; /* data before this offset (from data_offset) known to be in sync */
88 __u32 sb_csum; /* checksum upto dev_roles[max_dev] */
89 __u32 max_dev; /* size of dev_roles[] array to consider */
90 __u8 pad3[64-32]; /* set to 0 when writing */
91
92 /* device state information. Indexed by dev_number.
93 * 2 bytes per device
94 * Note there are no per-device state flags. State information is rolled
95 * into the 'roles' value. If a device is spare or faulty, then it doesn't
96 * have a meaningful role.
97 */
98 __u16 dev_roles[0]; /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
99 };
100
101 #define MAX_SB_SIZE 4096
102 /* bitmap super size is 256, but we round up to a sector for alignment */
103 #define BM_SUPER_SIZE 512
104 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
105 #define SUPER1_SIZE (MAX_SB_SIZE + BM_SUPER_SIZE \
106 + sizeof(struct misc_dev_info))
107
108 struct misc_dev_info {
109 __u64 device_size;
110 };
111
112 /* feature_map bits */
113 #define MD_FEATURE_BITMAP_OFFSET 1
114 #define MD_FEATURE_RECOVERY_OFFSET 2 /* recovery_offset is present and
115 * must be honoured
116 */
117 #define MD_FEATURE_RESHAPE_ACTIVE 4
118 #define MD_FEATURE_BAD_BLOCKS 8 /* badblock list is not empty */
119 #define MD_FEATURE_REPLACEMENT 16 /* This device is replacing an
120 * active device with same 'role'.
121 * 'recovery_offset' is also set.
122 */
123 #define MD_FEATURE_RESHAPE_BACKWARDS 32 /* Reshape doesn't change number
124 * of devices, but is going
125 * backwards anyway.
126 */
127 #define MD_FEATURE_NEW_OFFSET 64 /* new_offset must be honoured */
128 #define MD_FEATURE_ALL (MD_FEATURE_BITMAP_OFFSET \
129 |MD_FEATURE_RECOVERY_OFFSET \
130 |MD_FEATURE_RESHAPE_ACTIVE \
131 |MD_FEATURE_BAD_BLOCKS \
132 |MD_FEATURE_REPLACEMENT \
133 |MD_FEATURE_RESHAPE_BACKWARDS \
134 |MD_FEATURE_NEW_OFFSET \
135 )
136
137 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
138 {
139 unsigned int disk_csum, csum;
140 unsigned long long newcsum;
141 int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2;
142 unsigned int *isuper = (unsigned int*)sb;
143
144 /* make sure I can count... */
145 if (offsetof(struct mdp_superblock_1,data_offset) != 128 ||
146 offsetof(struct mdp_superblock_1, utime) != 192 ||
147 sizeof(struct mdp_superblock_1) != 256) {
148 fprintf(stderr, "WARNING - superblock isn't sized correctly\n");
149 }
150
151 disk_csum = sb->sb_csum;
152 sb->sb_csum = 0;
153 newcsum = 0;
154 for (; size>=4; size -= 4 ) {
155 newcsum += __le32_to_cpu(*isuper);
156 isuper++;
157 }
158
159 if (size == 2)
160 newcsum += __le16_to_cpu(*(unsigned short*) isuper);
161
162 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
163 sb->sb_csum = disk_csum;
164 return __cpu_to_le32(csum);
165 }
166
167 /*
168 * Information related to file descriptor used for aligned reads/writes.
169 * Cache the block size.
170 */
171 struct align_fd {
172 int fd;
173 int blk_sz;
174 };
175
176 static void init_afd(struct align_fd *afd, int fd)
177 {
178 afd->fd = fd;
179
180 if (ioctl(afd->fd, BLKSSZGET, &afd->blk_sz) != 0)
181 afd->blk_sz = 512;
182 }
183
184 static char abuf[4096+4096];
185 static int aread(struct align_fd *afd, void *buf, int len)
186 {
187 /* aligned read.
188 * On devices with a 4K sector size, we need to read
189 * the full sector and copy relevant bits into
190 * the buffer
191 */
192 int bsize, iosize;
193 char *b;
194 int n;
195
196 bsize = afd->blk_sz;
197
198 if (!bsize || bsize > 4096 || len > 4096) {
199 if (!bsize)
200 fprintf(stderr, "WARNING - aread() called with invalid block size\n");
201 return -1;
202 }
203 b = ROUND_UP_PTR((char *)abuf, 4096);
204
205 for (iosize = 0; iosize < len; iosize += bsize)
206 ;
207 n = read(afd->fd, b, iosize);
208 if (n <= 0)
209 return n;
210 lseek(afd->fd, len - n, 1);
211 if (n > len)
212 n = len;
213 memcpy(buf, b, n);
214 return n;
215 }
216
217 static int awrite(struct align_fd *afd, void *buf, int len)
218 {
219 /* aligned write.
220 * On devices with a 4K sector size, we need to write
221 * the full sector. We pre-read if the sector is larger
222 * than the write.
223 * The address must be sector-aligned.
224 */
225 int bsize, iosize;
226 char *b;
227 int n;
228
229 bsize = afd->blk_sz;
230 if (!bsize || bsize > 4096 || len > 4096) {
231 if (!bsize)
232 fprintf(stderr, "WARNING - awrite() called with invalid block size\n");
233 return -1;
234 }
235 b = ROUND_UP_PTR((char *)abuf, 4096);
236
237 for (iosize = 0; iosize < len ; iosize += bsize)
238 ;
239
240 if (len != iosize) {
241 n = read(afd->fd, b, iosize);
242 if (n <= 0)
243 return n;
244 lseek(afd->fd, -n, 1);
245 }
246
247 memcpy(b, buf, len);
248 n = write(afd->fd, b, iosize);
249 if (n <= 0)
250 return n;
251 lseek(afd->fd, len - n, 1);
252 return len;
253 }
254
255 #ifndef MDASSEMBLE
256 static void examine_super1(struct supertype *st, char *homehost)
257 {
258 struct mdp_superblock_1 *sb = st->sb;
259 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
260 time_t atime;
261 unsigned int d;
262 int role;
263 int delta_extra = 0;
264 int i;
265 char *c;
266 int l = homehost ? strlen(homehost) : 0;
267 int layout;
268 unsigned long long sb_offset;
269 struct mdinfo info;
270
271 printf(" Magic : %08x\n", __le32_to_cpu(sb->magic));
272 printf(" Version : 1");
273 sb_offset = __le64_to_cpu(sb->super_offset);
274 if (sb_offset <= 4)
275 printf(".1\n");
276 else if (sb_offset <= 8)
277 printf(".2\n");
278 else
279 printf(".0\n");
280 printf(" Feature Map : 0x%x\n", __le32_to_cpu(sb->feature_map));
281 printf(" Array UUID : ");
282 for (i=0; i<16; i++) {
283 if ((i&3)==0 && i != 0) printf(":");
284 printf("%02x", sb->set_uuid[i]);
285 }
286 printf("\n");
287 printf(" Name : %.32s", sb->set_name);
288 if (l > 0 && l < 32 &&
289 sb->set_name[l] == ':' &&
290 strncmp(sb->set_name, homehost, l) == 0)
291 printf(" (local to host %s)", homehost);
292 printf("\n");
293 if (bms->nodes > 0)
294 printf("Cluster Name : %s", bms->cluster_name);
295 atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL;
296 printf(" Creation Time : %.24s\n", ctime(&atime));
297 c=map_num(pers, __le32_to_cpu(sb->level));
298 printf(" Raid Level : %s\n", c?c:"-unknown-");
299 printf(" Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks));
300 printf("\n");
301 printf(" Avail Dev Size : %llu%s\n",
302 (unsigned long long)__le64_to_cpu(sb->data_size),
303 human_size(__le64_to_cpu(sb->data_size)<<9));
304 if (__le32_to_cpu(sb->level) > 0) {
305 int ddsks = 0, ddsks_denom = 1;
306 switch(__le32_to_cpu(sb->level)) {
307 case 1: ddsks=1;break;
308 case 4:
309 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
310 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
311 case 10:
312 layout = __le32_to_cpu(sb->layout);
313 ddsks = __le32_to_cpu(sb->raid_disks);
314 ddsks_denom = (layout&255) * ((layout>>8)&255);
315 }
316 if (ddsks) {
317 long long asize = __le64_to_cpu(sb->size);
318 asize = (asize << 9) * ddsks / ddsks_denom;
319 printf(" Array Size : %llu%s\n",
320 asize >> 10, human_size(asize));
321 }
322 if (sb->size != sb->data_size)
323 printf(" Used Dev Size : %llu%s\n",
324 (unsigned long long)__le64_to_cpu(sb->size),
325 human_size(__le64_to_cpu(sb->size)<<9));
326 }
327 if (sb->data_offset)
328 printf(" Data Offset : %llu sectors\n",
329 (unsigned long long)__le64_to_cpu(sb->data_offset));
330 if (sb->new_offset &&
331 (__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) {
332 unsigned long long offset = __le64_to_cpu(sb->data_offset);
333 offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset);
334 printf(" New Offset : %llu sectors\n", offset);
335 }
336 printf(" Super Offset : %llu sectors\n",
337 (unsigned long long)__le64_to_cpu(sb->super_offset));
338 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)
339 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb->recovery_offset));
340
341 st->ss->getinfo_super(st, &info, NULL);
342 if (info.space_after != 1 &&
343 !(__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
344 printf(" Unused Space : before=%llu sectors, after=%llu sectors\n",
345 info.space_before, info.space_after);
346
347 printf(" State : %s\n", (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean");
348 printf(" Device UUID : ");
349 for (i=0; i<16; i++) {
350 if ((i&3)==0 && i != 0) printf(":");
351 printf("%02x", sb->device_uuid[i]);
352 }
353 printf("\n");
354 printf("\n");
355 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
356 printf("Internal Bitmap : %ld sectors from superblock\n",
357 (long)(int32_t)__le32_to_cpu(sb->bitmap_offset));
358 }
359 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)) {
360 printf(" Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb->reshape_position)/2,
361 human_size(__le64_to_cpu(sb->reshape_position)<<9));
362 if (__le32_to_cpu(sb->delta_disks)) {
363 printf(" Delta Devices : %d", __le32_to_cpu(sb->delta_disks));
364 printf(" (%d->%d)\n",
365 __le32_to_cpu(sb->raid_disks)-__le32_to_cpu(sb->delta_disks),
366 __le32_to_cpu(sb->raid_disks));
367 if ((int)__le32_to_cpu(sb->delta_disks) < 0)
368 delta_extra = -__le32_to_cpu(sb->delta_disks);
369 }
370 if (__le32_to_cpu(sb->new_level) != __le32_to_cpu(sb->level)) {
371 c = map_num(pers, __le32_to_cpu(sb->new_level));
372 printf(" New Level : %s\n", c?c:"-unknown-");
373 }
374 if (__le32_to_cpu(sb->new_layout) != __le32_to_cpu(sb->layout)) {
375 if (__le32_to_cpu(sb->level) == 5) {
376 c = map_num(r5layout, __le32_to_cpu(sb->new_layout));
377 printf(" New Layout : %s\n", c?c:"-unknown-");
378 }
379 if (__le32_to_cpu(sb->level) == 6) {
380 c = map_num(r6layout, __le32_to_cpu(sb->new_layout));
381 printf(" New Layout : %s\n", c?c:"-unknown-");
382 }
383 if (__le32_to_cpu(sb->level) == 10) {
384 printf(" New Layout :");
385 print_r10_layout(__le32_to_cpu(sb->new_layout));
386 printf("\n");
387 }
388 }
389 if (__le32_to_cpu(sb->new_chunk) != __le32_to_cpu(sb->chunksize))
390 printf(" New Chunksize : %dK\n", __le32_to_cpu(sb->new_chunk)/2);
391 printf("\n");
392 }
393 if (sb->devflags) {
394 printf(" Flags :");
395 if (sb->devflags & WriteMostly1)
396 printf(" write-mostly");
397 printf("\n");
398 }
399
400 atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL;
401 printf(" Update Time : %.24s\n", ctime(&atime));
402
403 if (sb->bblog_size && sb->bblog_offset) {
404 printf(" Bad Block Log : %d entries available at offset %ld sectors",
405 __le16_to_cpu(sb->bblog_size)*512/8,
406 (long)(int32_t)__le32_to_cpu(sb->bblog_offset));
407 if (sb->feature_map &
408 __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
409 printf(" - bad blocks present.");
410 printf("\n");
411 }
412
413 if (calc_sb_1_csum(sb) == sb->sb_csum)
414 printf(" Checksum : %x - correct\n", __le32_to_cpu(sb->sb_csum));
415 else
416 printf(" Checksum : %x - expected %x\n", __le32_to_cpu(sb->sb_csum),
417 __le32_to_cpu(calc_sb_1_csum(sb)));
418 printf(" Events : %llu\n", (unsigned long long)__le64_to_cpu(sb->events));
419 printf("\n");
420 if (__le32_to_cpu(sb->level) == 5) {
421 c = map_num(r5layout, __le32_to_cpu(sb->layout));
422 printf(" Layout : %s\n", c?c:"-unknown-");
423 }
424 if (__le32_to_cpu(sb->level) == 6) {
425 c = map_num(r6layout, __le32_to_cpu(sb->layout));
426 printf(" Layout : %s\n", c?c:"-unknown-");
427 }
428 if (__le32_to_cpu(sb->level) == 10) {
429 int lo = __le32_to_cpu(sb->layout);
430 printf(" Layout :");
431 print_r10_layout(lo);
432 printf("\n");
433 }
434 switch(__le32_to_cpu(sb->level)) {
435 case 0:
436 case 4:
437 case 5:
438 case 6:
439 case 10:
440 printf(" Chunk Size : %dK\n", __le32_to_cpu(sb->chunksize)/2);
441 break;
442 case -1:
443 printf(" Rounding : %dK\n", __le32_to_cpu(sb->chunksize)/2);
444 break;
445 default: break;
446 }
447 printf("\n");
448 #if 0
449 /* This turns out to just be confusing */
450 printf(" Array Slot : %d (", __le32_to_cpu(sb->dev_number));
451 for (i= __le32_to_cpu(sb->max_dev); i> 0 ; i--)
452 if (__le16_to_cpu(sb->dev_roles[i-1]) != 0xffff)
453 break;
454 for (d=0; d < i; d++) {
455 int role = __le16_to_cpu(sb->dev_roles[d]);
456 if (d) printf(", ");
457 if (role == 0xffff) printf("empty");
458 else if(role == 0xfffe) printf("failed");
459 else printf("%d", role);
460 }
461 printf(")\n");
462 #endif
463 printf(" Device Role : ");
464 d = __le32_to_cpu(sb->dev_number);
465 if (d < __le32_to_cpu(sb->max_dev))
466 role = __le16_to_cpu(sb->dev_roles[d]);
467 else
468 role = 0xFFFF;
469 if (role >= 0xFFFE)
470 printf("spare\n");
471 else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_REPLACEMENT))
472 printf("Replacement device %d\n", role);
473 else
474 printf("Active device %d\n", role);
475
476 printf(" Array State : ");
477 for (d=0; d<__le32_to_cpu(sb->raid_disks) + delta_extra; d++) {
478 int cnt = 0;
479 unsigned int i;
480 for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
481 unsigned int role = __le16_to_cpu(sb->dev_roles[i]);
482 if (role == d)
483 cnt++;
484 }
485 if (cnt == 2)
486 printf("R");
487 else if (cnt == 1)
488 printf("A");
489 else if (cnt == 0)
490 printf(".");
491 else
492 printf("?");
493 }
494 #if 0
495 /* This is confusing too */
496 faulty = 0;
497 for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
498 int role = __le16_to_cpu(sb->dev_roles[i]);
499 if (role == 0xFFFE)
500 faulty++;
501 }
502 if (faulty) printf(" %d failed", faulty);
503 #endif
504 printf(" ('A' == active, '.' == missing, 'R' == replacing)");
505 printf("\n");
506 }
507
508 static void brief_examine_super1(struct supertype *st, int verbose)
509 {
510 struct mdp_superblock_1 *sb = st->sb;
511 int i;
512 unsigned long long sb_offset;
513 char *nm;
514 char *c=map_num(pers, __le32_to_cpu(sb->level));
515
516 nm = strchr(sb->set_name, ':');
517 if (nm)
518 nm++;
519 else if (sb->set_name[0])
520 nm = sb->set_name;
521 else
522 nm = NULL;
523
524 printf("ARRAY ");
525 if (nm) {
526 printf("/dev/md/");
527 print_escape(nm);
528 putchar(' ');
529 }
530 if (verbose && c)
531 printf(" level=%s", c);
532 sb_offset = __le64_to_cpu(sb->super_offset);
533 if (sb_offset <= 4)
534 printf(" metadata=1.1 ");
535 else if (sb_offset <= 8)
536 printf(" metadata=1.2 ");
537 else
538 printf(" metadata=1.0 ");
539 if (verbose)
540 printf("num-devices=%d ", __le32_to_cpu(sb->raid_disks));
541 printf("UUID=");
542 for (i=0; i<16; i++) {
543 if ((i&3)==0 && i != 0) printf(":");
544 printf("%02x", sb->set_uuid[i]);
545 }
546 if (sb->set_name[0]) {
547 printf(" name=");
548 print_quoted(sb->set_name);
549 }
550 printf("\n");
551 }
552
553 static void export_examine_super1(struct supertype *st)
554 {
555 struct mdp_superblock_1 *sb = st->sb;
556 int i;
557 int len = 32;
558 int layout;
559
560 printf("MD_LEVEL=%s\n", map_num(pers, __le32_to_cpu(sb->level)));
561 printf("MD_DEVICES=%d\n", __le32_to_cpu(sb->raid_disks));
562 for (i=0; i<32; i++)
563 if (sb->set_name[i] == '\n' ||
564 sb->set_name[i] == '\0') {
565 len = i;
566 break;
567 }
568 if (len)
569 printf("MD_NAME=%.*s\n", len, sb->set_name);
570 if (__le32_to_cpu(sb->level) > 0) {
571 int ddsks = 0, ddsks_denom = 1;
572 switch(__le32_to_cpu(sb->level)) {
573 case 1: ddsks=1;break;
574 case 4:
575 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
576 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
577 case 10:
578 layout = __le32_to_cpu(sb->layout);
579 ddsks = __le32_to_cpu(sb->raid_disks);
580 ddsks_denom = (layout&255) * ((layout>>8)&255);
581 }
582 if (ddsks) {
583 long long asize = __le64_to_cpu(sb->size);
584 asize = (asize << 9) * ddsks / ddsks_denom;
585 printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize,JEDEC));
586 }
587 }
588 printf("MD_UUID=");
589 for (i=0; i<16; i++) {
590 if ((i&3)==0 && i != 0) printf(":");
591 printf("%02x", sb->set_uuid[i]);
592 }
593 printf("\n");
594 printf("MD_UPDATE_TIME=%llu\n",
595 __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL);
596 printf("MD_DEV_UUID=");
597 for (i=0; i<16; i++) {
598 if ((i&3)==0 && i != 0) printf(":");
599 printf("%02x", sb->device_uuid[i]);
600 }
601 printf("\n");
602 printf("MD_EVENTS=%llu\n",
603 (unsigned long long)__le64_to_cpu(sb->events));
604 }
605
606 static int copy_metadata1(struct supertype *st, int from, int to)
607 {
608 /* Read superblock. If it looks good, write it out.
609 * Then if a bitmap is present, copy that.
610 * And if a bad-block-list is present, copy that too.
611 */
612 void *buf;
613 unsigned long long dsize, sb_offset;
614 const int bufsize = 4*1024;
615 struct mdp_superblock_1 super, *sb;
616
617 if (posix_memalign(&buf, 4096, bufsize) != 0)
618 return 1;
619
620 if (!get_dev_size(from, NULL, &dsize))
621 goto err;
622
623 dsize >>= 9;
624 if (dsize < 24)
625 goto err;
626 switch(st->minor_version) {
627 case 0:
628 sb_offset = dsize;
629 sb_offset -= 8*2;
630 sb_offset &= ~(4*2-1);
631 break;
632 case 1:
633 sb_offset = 0;
634 break;
635 case 2:
636 sb_offset = 4*2;
637 break;
638 default:
639 goto err;
640 }
641
642 if (lseek64(from, sb_offset << 9, 0) < 0LL)
643 goto err;
644 if (read(from, buf, bufsize) != bufsize)
645 goto err;
646
647 sb = buf;
648 super = *sb; // save most of sb for when we reuse buf
649
650 if (__le32_to_cpu(super.magic) != MD_SB_MAGIC ||
651 __le32_to_cpu(super.major_version) != 1 ||
652 __le64_to_cpu(super.super_offset) != sb_offset ||
653 calc_sb_1_csum(sb) != super.sb_csum)
654 goto err;
655
656 if (lseek64(to, sb_offset << 9, 0) < 0LL)
657 goto err;
658 if (write(to, buf, bufsize) != bufsize)
659 goto err;
660
661 if (super.feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
662 unsigned long long bitmap_offset = sb_offset;
663 int bytes = 4096; // just an estimate.
664 int written = 0;
665 struct align_fd afrom, ato;
666
667 init_afd(&afrom, from);
668 init_afd(&ato, to);
669
670 bitmap_offset += (int32_t)__le32_to_cpu(super.bitmap_offset);
671
672 if (lseek64(from, bitmap_offset<<9, 0) < 0)
673 goto err;
674 if (lseek64(to, bitmap_offset<<9, 0) < 0)
675 goto err;
676
677 for (written = 0; written < bytes ; ) {
678 int n = bytes - written;
679 if (n > 4096)
680 n = 4096;
681 if (aread(&afrom, buf, n) != n)
682 goto err;
683 if (written == 0) {
684 /* have the header, can calculate
685 * correct bitmap bytes */
686 bitmap_super_t *bms;
687 int bits;
688 bms = (void*)buf;
689 bits = __le64_to_cpu(bms->sync_size) / (__le32_to_cpu(bms->chunksize)>>9);
690 bytes = (bits+7) >> 3;
691 bytes += sizeof(bitmap_super_t);
692 bytes = ROUND_UP(bytes, 512);
693 if (n > bytes)
694 n = bytes;
695 }
696 if (awrite(&ato, buf, n) != n)
697 goto err;
698 written += n;
699 }
700 }
701
702 if (super.bblog_size != 0 &&
703 __le32_to_cpu(super.bblog_size) <= 100 &&
704 super.bblog_offset != 0 &&
705 (super.feature_map & __le32_to_cpu(MD_FEATURE_BAD_BLOCKS))) {
706 /* There is a bad block log */
707 unsigned long long bb_offset = sb_offset;
708 int bytes = __le32_to_cpu(super.bblog_size) * 512;
709 int written = 0;
710 struct align_fd afrom, ato;
711
712 init_afd(&afrom, from);
713 init_afd(&ato, to);
714
715 bb_offset += (int32_t)__le32_to_cpu(super.bblog_offset);
716
717 if (lseek64(from, bb_offset<<9, 0) < 0)
718 goto err;
719 if (lseek64(to, bb_offset<<9, 0) < 0)
720 goto err;
721
722 for (written = 0; written < bytes ; ) {
723 int n = bytes - written;
724 if (n > 4096)
725 n = 4096;
726 if (aread(&afrom, buf, n) != n)
727 goto err;
728
729 if (awrite(&ato, buf, n) != n)
730 goto err;
731 written += n;
732 }
733 }
734
735 free(buf);
736 return 0;
737
738 err:
739 free(buf);
740 return 1;
741 }
742
743 static void detail_super1(struct supertype *st, char *homehost)
744 {
745 struct mdp_superblock_1 *sb = st->sb;
746 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
747 int i;
748 int l = homehost ? strlen(homehost) : 0;
749
750 printf(" Name : %.32s", sb->set_name);
751 if (l > 0 && l < 32 &&
752 sb->set_name[l] == ':' &&
753 strncmp(sb->set_name, homehost, l) == 0)
754 printf(" (local to host %s)", homehost);
755 if (bms->nodes > 0)
756 printf("Cluster Name : %64s", bms->cluster_name);
757 printf("\n UUID : ");
758 for (i=0; i<16; i++) {
759 if ((i&3)==0 && i != 0) printf(":");
760 printf("%02x", sb->set_uuid[i]);
761 }
762 printf("\n Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb->events));
763 }
764
765 static void brief_detail_super1(struct supertype *st)
766 {
767 struct mdp_superblock_1 *sb = st->sb;
768 int i;
769
770 if (sb->set_name[0]) {
771 printf(" name=");
772 print_quoted(sb->set_name);
773 }
774 printf(" UUID=");
775 for (i=0; i<16; i++) {
776 if ((i&3)==0 && i != 0) printf(":");
777 printf("%02x", sb->set_uuid[i]);
778 }
779 }
780
781 static void export_detail_super1(struct supertype *st)
782 {
783 struct mdp_superblock_1 *sb = st->sb;
784 int i;
785 int len = 32;
786
787 for (i=0; i<32; i++)
788 if (sb->set_name[i] == '\n' ||
789 sb->set_name[i] == '\0') {
790 len = i;
791 break;
792 }
793 if (len)
794 printf("MD_NAME=%.*s\n", len, sb->set_name);
795 }
796
797 static int examine_badblocks_super1(struct supertype *st, int fd, char *devname)
798 {
799 struct mdp_superblock_1 *sb = st->sb;
800 unsigned long long offset;
801 int size;
802 __u64 *bbl, *bbp;
803 int i;
804
805 if (!sb->bblog_size || __le32_to_cpu(sb->bblog_size) > 100
806 || !sb->bblog_offset){
807 printf("No bad-blocks list configured on %s\n", devname);
808 return 0;
809 }
810 if ((sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
811 == 0) {
812 printf("Bad-blocks list is empty in %s\n", devname);
813 return 0;
814 }
815
816 size = __le32_to_cpu(sb->bblog_size)* 512;
817 if (posix_memalign((void**)&bbl, 4096, size) != 0) {
818 pr_err("could not allocate badblocks list\n");
819 return 0;
820 }
821 offset = __le64_to_cpu(sb->super_offset) +
822 (int)__le32_to_cpu(sb->bblog_offset);
823 offset <<= 9;
824 if (lseek64(fd, offset, 0) < 0) {
825 pr_err("Cannot seek to bad-blocks list\n");
826 return 1;
827 }
828 if (read(fd, bbl, size) != size) {
829 pr_err("Cannot read bad-blocks list\n");
830 return 1;
831 }
832 /* 64bits per entry. 10 bits is block-count, 54 bits is block
833 * offset. Blocks are sectors unless bblog->shift makes them bigger
834 */
835 bbp = (__u64*)bbl;
836 printf("Bad-blocks on %s:\n", devname);
837 for (i = 0; i < size/8; i++, bbp++) {
838 __u64 bb = __le64_to_cpu(*bbp);
839 int count = bb & 0x3ff;
840 unsigned long long sector = bb >> 10;
841
842 if (bb + 1 == 0)
843 break;
844
845 sector <<= sb->bblog_shift;
846 count <<= sb->bblog_shift;
847
848 printf("%20llu for %d sectors\n", sector, count);
849 }
850 return 0;
851 }
852
853 #endif
854
855 static int match_home1(struct supertype *st, char *homehost)
856 {
857 struct mdp_superblock_1 *sb = st->sb;
858 int l = homehost ? strlen(homehost) : 0;
859
860 return (l > 0 && l < 32 &&
861 sb->set_name[l] == ':' &&
862 strncmp(sb->set_name, homehost, l) == 0);
863 }
864
865 static void uuid_from_super1(struct supertype *st, int uuid[4])
866 {
867 struct mdp_superblock_1 *super = st->sb;
868 char *cuuid = (char*)uuid;
869 int i;
870 for (i=0; i<16; i++)
871 cuuid[i] = super->set_uuid[i];
872 }
873
874 static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
875 {
876 struct mdp_superblock_1 *sb = st->sb;
877 struct bitmap_super_s *bsb = (void*)(((char*)sb)+MAX_SB_SIZE);
878 struct misc_dev_info *misc = (void*)(((char*)sb)+MAX_SB_SIZE+BM_SUPER_SIZE);
879 int working = 0;
880 unsigned int i;
881 unsigned int role;
882 unsigned int map_disks = info->array.raid_disks;
883 unsigned long long super_offset;
884 unsigned long long data_size;
885
886 memset(info, 0, sizeof(*info));
887 info->array.major_version = 1;
888 info->array.minor_version = st->minor_version;
889 info->array.patch_version = 0;
890 info->array.raid_disks = __le32_to_cpu(sb->raid_disks);
891 info->array.level = __le32_to_cpu(sb->level);
892 info->array.layout = __le32_to_cpu(sb->layout);
893 info->array.md_minor = -1;
894 info->array.ctime = __le64_to_cpu(sb->ctime);
895 info->array.utime = __le64_to_cpu(sb->utime);
896 info->array.chunk_size = __le32_to_cpu(sb->chunksize)*512;
897 info->array.state =
898 (__le64_to_cpu(sb->resync_offset) == MaxSector)
899 ? 1 : 0;
900 if (__le32_to_cpu(bsb->nodes) > 1)
901 info->array.state |= (1 << MD_SB_CLUSTERED);
902
903 info->data_offset = __le64_to_cpu(sb->data_offset);
904 info->component_size = __le64_to_cpu(sb->size);
905 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET))
906 info->bitmap_offset = (int32_t)__le32_to_cpu(sb->bitmap_offset);
907
908 info->disk.major = 0;
909 info->disk.minor = 0;
910 info->disk.number = __le32_to_cpu(sb->dev_number);
911 if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) ||
912 __le32_to_cpu(sb->dev_number) >= MAX_DEVS)
913 role = 0xfffe;
914 else
915 role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]);
916
917 super_offset = __le64_to_cpu(sb->super_offset);
918 if (info->array.level <= 0)
919 data_size = __le64_to_cpu(sb->data_size);
920 else
921 data_size = __le64_to_cpu(sb->size);
922 if (info->data_offset < super_offset) {
923 unsigned long long end;
924 info->space_before = info->data_offset;
925 end = super_offset;
926
927 if (sb->bblog_offset && sb->bblog_size) {
928 unsigned long long bboffset = super_offset;
929 bboffset += (int32_t)__le32_to_cpu(sb->bblog_offset);
930 if (bboffset < end)
931 end = bboffset;
932 }
933
934 if (super_offset + info->bitmap_offset < end)
935 end = super_offset + info->bitmap_offset;
936
937 if (info->data_offset + data_size < end)
938 info->space_after = end - data_size - info->data_offset;
939 else
940 info->space_after = 0;
941 } else {
942 unsigned long long earliest;
943 earliest = super_offset + (32+4)*2; /* match kernel */
944 if (info->bitmap_offset > 0) {
945 unsigned long long bmend = info->bitmap_offset;
946 unsigned long long size = __le64_to_cpu(bsb->sync_size);
947 size /= __le32_to_cpu(bsb->chunksize) >> 9;
948 size = (size + 7) >> 3;
949 size += sizeof(bitmap_super_t);
950 size = ROUND_UP(size, 4096);
951 size /= 512;
952 bmend += size;
953 if (bmend > earliest)
954 bmend = earliest;
955 }
956 if (sb->bblog_offset && sb->bblog_size) {
957 unsigned long long bbend = super_offset;
958 bbend += (int32_t)__le32_to_cpu(sb->bblog_offset);
959 bbend += __le32_to_cpu(sb->bblog_size);
960 if (bbend > earliest)
961 earliest = bbend;
962 }
963 if (earliest < info->data_offset)
964 info->space_before = info->data_offset - earliest;
965 else
966 info->space_before = 0;
967 info->space_after = misc->device_size - data_size - info->data_offset;
968 }
969 if (info->space_before == 0 && info->space_after == 0) {
970 /* It will look like we don't support data_offset changes,
971 * be we do - it's just that there is no room.
972 * A change that reduced the number of devices should
973 * still be allowed, so set the otherwise useless value of '1'
974 */
975 info->space_after = 1;
976 }
977
978 info->disk.raid_disk = -1;
979 switch(role) {
980 case 0xFFFF:
981 info->disk.state = 0; /* spare: not active, not sync, not faulty */
982 break;
983 case 0xFFFE:
984 info->disk.state = 1; /* faulty */
985 break;
986 default:
987 info->disk.state = 6; /* active and in sync */
988 info->disk.raid_disk = role;
989 }
990 if (sb->devflags & WriteMostly1)
991 info->disk.state |= (1 << MD_DISK_WRITEMOSTLY);
992 info->events = __le64_to_cpu(sb->events);
993 sprintf(info->text_version, "1.%d", st->minor_version);
994 info->safe_mode_delay = 200;
995
996 memcpy(info->uuid, sb->set_uuid, 16);
997
998 strncpy(info->name, sb->set_name, 32);
999 info->name[32] = 0;
1000
1001 if ((__le32_to_cpu(sb->feature_map)&MD_FEATURE_REPLACEMENT)) {
1002 info->disk.state &= ~(1 << MD_DISK_SYNC);
1003 info->disk.state |= 1 << MD_DISK_REPLACEMENT;
1004 }
1005
1006 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET))
1007 info->recovery_start = __le32_to_cpu(sb->recovery_offset);
1008 else
1009 info->recovery_start = MaxSector;
1010
1011 if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) {
1012 info->reshape_active = 1;
1013 if ((sb->feature_map & __le32_to_cpu(MD_FEATURE_NEW_OFFSET)) &&
1014 sb->new_offset != 0)
1015 info->reshape_active |= RESHAPE_NO_BACKUP;
1016 info->reshape_progress = __le64_to_cpu(sb->reshape_position);
1017 info->new_level = __le32_to_cpu(sb->new_level);
1018 info->delta_disks = __le32_to_cpu(sb->delta_disks);
1019 info->new_layout = __le32_to_cpu(sb->new_layout);
1020 info->new_chunk = __le32_to_cpu(sb->new_chunk)<<9;
1021 if (info->delta_disks < 0)
1022 info->array.raid_disks -= info->delta_disks;
1023 } else
1024 info->reshape_active = 0;
1025
1026 info->recovery_blocked = info->reshape_active;
1027
1028 if (map)
1029 for (i=0; i<map_disks; i++)
1030 map[i] = 0;
1031 for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
1032 role = __le16_to_cpu(sb->dev_roles[i]);
1033 if (/*role == 0xFFFF || */role < (unsigned) info->array.raid_disks) {
1034 working++;
1035 if (map && role < map_disks)
1036 map[role] = 1;
1037 }
1038 }
1039
1040 info->array.working_disks = working;
1041 }
1042
1043 static struct mdinfo *container_content1(struct supertype *st, char *subarray)
1044 {
1045 struct mdinfo *info;
1046
1047 if (subarray)
1048 return NULL;
1049
1050 info = xmalloc(sizeof(*info));
1051 getinfo_super1(st, info, NULL);
1052 return info;
1053 }
1054
1055 static int update_super1(struct supertype *st, struct mdinfo *info,
1056 char *update,
1057 char *devname, int verbose,
1058 int uuid_set, char *homehost)
1059 {
1060 /* NOTE: for 'assemble' and 'force' we need to return non-zero
1061 * if any change was made. For others, the return value is
1062 * ignored.
1063 */
1064 int rv = 0;
1065 struct mdp_superblock_1 *sb = st->sb;
1066
1067 if (strcmp(update, "homehost") == 0 &&
1068 homehost) {
1069 /* Note that 'homehost' is special as it is really
1070 * a "name" update.
1071 */
1072 char *c;
1073 update = "name";
1074 c = strchr(sb->set_name, ':');
1075 if (c)
1076 strncpy(info->name, c+1, 31 - (c-sb->set_name));
1077 else
1078 strncpy(info->name, sb->set_name, 32);
1079 info->name[32] = 0;
1080 }
1081
1082 if (strcmp(update, "force-one")==0) {
1083 /* Not enough devices for a working array,
1084 * so bring this one up-to-date
1085 */
1086 if (sb->events != __cpu_to_le64(info->events))
1087 rv = 1;
1088 sb->events = __cpu_to_le64(info->events);
1089 } else if (strcmp(update, "force-array")==0) {
1090 /* Degraded array and 'force' requests to
1091 * maybe need to mark it 'clean'.
1092 */
1093 switch(__le32_to_cpu(sb->level)) {
1094 case 5: case 4: case 6:
1095 /* need to force clean */
1096 if (sb->resync_offset != MaxSector)
1097 rv = 1;
1098 sb->resync_offset = MaxSector;
1099 }
1100 } else if (strcmp(update, "assemble")==0) {
1101 int d = info->disk.number;
1102 int want;
1103 if (info->disk.state & (1<<MD_DISK_ACTIVE))
1104 want = info->disk.raid_disk;
1105 else
1106 want = 0xFFFF;
1107 if (sb->dev_roles[d] != __cpu_to_le16(want)) {
1108 sb->dev_roles[d] = __cpu_to_le16(want);
1109 rv = 1;
1110 }
1111 if (info->reshape_active &&
1112 sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1113 info->delta_disks >= 0 &&
1114 info->reshape_progress < __le64_to_cpu(sb->reshape_position)) {
1115 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1116 rv = 1;
1117 }
1118 if (info->reshape_active &&
1119 sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1120 info->delta_disks < 0 &&
1121 info->reshape_progress > __le64_to_cpu(sb->reshape_position)) {
1122 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1123 rv = 1;
1124 }
1125 } else if (strcmp(update, "linear-grow-new") == 0) {
1126 unsigned int i;
1127 int rfd, fd;
1128 unsigned int max = __le32_to_cpu(sb->max_dev);
1129
1130 for (i=0 ; i < max ; i++)
1131 if (__le16_to_cpu(sb->dev_roles[i]) >= 0xfffe)
1132 break;
1133 sb->dev_number = __cpu_to_le32(i);
1134 info->disk.number = i;
1135 if (max >= __le32_to_cpu(sb->max_dev))
1136 sb->max_dev = __cpu_to_le32(max+1);
1137
1138 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
1139 read(rfd, sb->device_uuid, 16) != 16) {
1140 __u32 r[4] = {random(), random(), random(), random()};
1141 memcpy(sb->device_uuid, r, 16);
1142 }
1143 if (rfd >= 0)
1144 close(rfd);
1145
1146 sb->dev_roles[i] =
1147 __cpu_to_le16(info->disk.raid_disk);
1148
1149 fd = open(devname, O_RDONLY);
1150 if (fd >= 0) {
1151 unsigned long long ds;
1152 get_dev_size(fd, devname, &ds);
1153 close(fd);
1154 ds >>= 9;
1155 if (__le64_to_cpu(sb->super_offset) <
1156 __le64_to_cpu(sb->data_offset)) {
1157 sb->data_size = __cpu_to_le64(
1158 ds - __le64_to_cpu(sb->data_offset));
1159 } else {
1160 ds -= 8*2;
1161 ds &= ~(unsigned long long)(4*2-1);
1162 sb->super_offset = __cpu_to_le64(ds);
1163 sb->data_size = __cpu_to_le64(
1164 ds - __le64_to_cpu(sb->data_offset));
1165 }
1166 }
1167 } else if (strcmp(update, "linear-grow-update") == 0) {
1168 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
1169 sb->dev_roles[info->disk.number] =
1170 __cpu_to_le16(info->disk.raid_disk);
1171 } else if (strcmp(update, "resync") == 0) {
1172 /* make sure resync happens */
1173 sb->resync_offset = 0ULL;
1174 } else if (strcmp(update, "uuid") == 0) {
1175 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
1176
1177 if (__le32_to_cpu(sb->feature_map)&MD_FEATURE_BITMAP_OFFSET) {
1178 struct bitmap_super_s *bm;
1179 bm = (struct bitmap_super_s*)(st->sb+MAX_SB_SIZE);
1180 memcpy(bm->uuid, sb->set_uuid, 16);
1181 }
1182 } else if (strcmp(update, "no-bitmap") == 0) {
1183 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1184 } else if (strcmp(update, "bbl") == 0) {
1185 /* only possible if there is room after the bitmap, or if
1186 * there is no bitmap
1187 */
1188 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1189 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1190 long bitmap_offset = (long)(int32_t)__le32_to_cpu(sb->bitmap_offset);
1191 long bm_sectors = 0;
1192 long space;
1193
1194 #ifndef MDASSEMBLE
1195 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1196 struct bitmap_super_s *bsb;
1197 bsb = (struct bitmap_super_s *)(((char*)sb)+MAX_SB_SIZE);
1198 bm_sectors = bitmap_sectors(bsb);
1199 }
1200 #endif
1201 if (sb_offset < data_offset) {
1202 /* 1.1 or 1.2. Put bbl after bitmap leaving at least 32K
1203 */
1204 long bb_offset;
1205 bb_offset = sb_offset + 8;
1206 if (bm_sectors && bitmap_offset > 0)
1207 bb_offset = bitmap_offset + bm_sectors;
1208 while (bb_offset < (long)sb_offset + 8 + 32*2
1209 && bb_offset + 8+8 <= (long)data_offset)
1210 /* too close to bitmap, and room to grow */
1211 bb_offset += 8;
1212 if (bb_offset + 8 <= (long)data_offset) {
1213 sb->bblog_size = __cpu_to_le16(8);
1214 sb->bblog_offset = __cpu_to_le32(bb_offset);
1215 }
1216 } else {
1217 /* 1.0 - Put bbl just before super block */
1218 if (bm_sectors && bitmap_offset < 0)
1219 space = -bitmap_offset - bm_sectors;
1220 else
1221 space = sb_offset - data_offset -
1222 __le64_to_cpu(sb->data_size);
1223 if (space >= 8) {
1224 sb->bblog_size = __cpu_to_le16(8);
1225 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1226 }
1227 }
1228 } else if (strcmp(update, "no-bbl") == 0) {
1229 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
1230 pr_err("Cannot remove active bbl from %s\n",devname);
1231 else {
1232 sb->bblog_size = 0;
1233 sb->bblog_shift = 0;
1234 sb->bblog_offset = 0;
1235 }
1236 } else if (strcmp(update, "name") == 0) {
1237 if (info->name[0] == 0)
1238 sprintf(info->name, "%d", info->array.md_minor);
1239 memset(sb->set_name, 0, sizeof(sb->set_name));
1240 if (homehost &&
1241 strchr(info->name, ':') == NULL &&
1242 strlen(homehost)+1+strlen(info->name) < 32) {
1243 strcpy(sb->set_name, homehost);
1244 strcat(sb->set_name, ":");
1245 strcat(sb->set_name, info->name);
1246 } else
1247 strcpy(sb->set_name, info->name);
1248 } else if (strcmp(update, "devicesize") == 0 &&
1249 __le64_to_cpu(sb->super_offset) <
1250 __le64_to_cpu(sb->data_offset)) {
1251 /* set data_size to device size less data_offset */
1252 struct misc_dev_info *misc = (struct misc_dev_info*)
1253 (st->sb + MAX_SB_SIZE + BM_SUPER_SIZE);
1254 sb->data_size = __cpu_to_le64(
1255 misc->device_size - __le64_to_cpu(sb->data_offset));
1256 } else if (strcmp(update, "revert-reshape") == 0) {
1257 rv = -2;
1258 if (!(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)))
1259 pr_err("No active reshape to revert on %s\n",
1260 devname);
1261 else {
1262 __u32 temp;
1263 unsigned long long reshape_sectors;
1264 long reshape_chunk;
1265 rv = 0;
1266 /* reshape_position is a little messy.
1267 * Its value must be a multiple of the larger
1268 * chunk size, and of the "after" data disks.
1269 * So when reverting we need to change it to
1270 * be a multiple of the new "after" data disks,
1271 * which is the old "before".
1272 * If it isn't already a multiple of 'before',
1273 * the only thing we could do would be
1274 * copy some block around on the disks, which
1275 * is easy to get wrong.
1276 * So we reject a revert-reshape unless the
1277 * alignment is good.
1278 */
1279 if (__le32_to_cpu(sb->level) >= 4 &&
1280 __le32_to_cpu(sb->level) <= 6) {
1281 reshape_sectors = __le64_to_cpu(sb->reshape_position);
1282 reshape_chunk = __le32_to_cpu(sb->new_chunk);
1283 reshape_chunk *= __le32_to_cpu(sb->raid_disks) - __le32_to_cpu(sb->delta_disks) -
1284 (__le32_to_cpu(sb->level)==6 ? 2 : 1);
1285 if (reshape_sectors % reshape_chunk) {
1286 pr_err("Reshape position is not suitably aligned.\n");
1287 pr_err("Try normal assembly and stop again\n");
1288 return -2;
1289 }
1290 }
1291 sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1292 __le32_to_cpu(sb->delta_disks));
1293 if (sb->delta_disks == 0)
1294 sb->feature_map ^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1295 else
1296 sb->delta_disks = __cpu_to_le32(-__le32_to_cpu(sb->delta_disks));
1297
1298 temp = sb->new_layout;
1299 sb->new_layout = sb->layout;
1300 sb->layout = temp;
1301
1302 temp = sb->new_chunk;
1303 sb->new_chunk = sb->chunksize;
1304 sb->chunksize = temp;
1305
1306 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_NEW_OFFSET)) {
1307 long offset_delta = (int32_t)__le32_to_cpu(sb->new_offset);
1308 sb->data_offset = __cpu_to_le64(__le64_to_cpu(sb->data_offset) + offset_delta);
1309 sb->new_offset = __cpu_to_le32(-offset_delta);
1310 sb->data_size = __cpu_to_le64(__le64_to_cpu(sb->data_size) - offset_delta);
1311 }
1312 }
1313 } else if (strcmp(update, "_reshape_progress")==0)
1314 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1315 else if (strcmp(update, "writemostly")==0)
1316 sb->devflags |= WriteMostly1;
1317 else if (strcmp(update, "readwrite")==0)
1318 sb->devflags &= ~WriteMostly1;
1319 else
1320 rv = -1;
1321
1322 sb->sb_csum = calc_sb_1_csum(sb);
1323 return rv;
1324 }
1325
1326 static int init_super1(struct supertype *st, mdu_array_info_t *info,
1327 unsigned long long size, char *name, char *homehost,
1328 int *uuid, unsigned long long data_offset)
1329 {
1330 struct mdp_superblock_1 *sb;
1331 int spares;
1332 int rfd;
1333 char defname[10];
1334 int sbsize;
1335
1336 if (posix_memalign((void**)&sb, 4096, SUPER1_SIZE) != 0) {
1337 pr_err("could not allocate superblock\n");
1338 return 0;
1339 }
1340 memset(sb, 0, SUPER1_SIZE);
1341
1342 st->sb = sb;
1343 if (info == NULL) {
1344 /* zeroing superblock */
1345 return 0;
1346 }
1347
1348 spares = info->working_disks - info->active_disks;
1349 if (info->raid_disks + spares > MAX_DEVS) {
1350 pr_err("too many devices requested: %d+%d > %d\n",
1351 info->raid_disks , spares, MAX_DEVS);
1352 return 0;
1353 }
1354
1355 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
1356 sb->major_version = __cpu_to_le32(1);
1357 sb->feature_map = 0;
1358 sb->pad0 = 0;
1359
1360 if (uuid)
1361 copy_uuid(sb->set_uuid, uuid, super1.swapuuid);
1362 else {
1363 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
1364 read(rfd, sb->set_uuid, 16) != 16) {
1365 __u32 r[4] = {random(), random(), random(), random()};
1366 memcpy(sb->set_uuid, r, 16);
1367 }
1368 if (rfd >= 0) close(rfd);
1369 }
1370
1371 if (name == NULL || *name == 0) {
1372 sprintf(defname, "%d", info->md_minor);
1373 name = defname;
1374 }
1375 if (homehost &&
1376 strchr(name, ':')== NULL &&
1377 strlen(homehost)+1+strlen(name) < 32) {
1378 strcpy(sb->set_name, homehost);
1379 strcat(sb->set_name, ":");
1380 strcat(sb->set_name, name);
1381 } else
1382 strcpy(sb->set_name, name);
1383
1384 sb->ctime = __cpu_to_le64((unsigned long long)time(0));
1385 sb->level = __cpu_to_le32(info->level);
1386 sb->layout = __cpu_to_le32(info->layout);
1387 sb->size = __cpu_to_le64(size*2ULL);
1388 sb->chunksize = __cpu_to_le32(info->chunk_size>>9);
1389 sb->raid_disks = __cpu_to_le32(info->raid_disks);
1390
1391 sb->data_offset = __cpu_to_le64(data_offset);
1392 sb->data_size = __cpu_to_le64(0);
1393 sb->super_offset = __cpu_to_le64(0);
1394 sb->recovery_offset = __cpu_to_le64(0);
1395
1396 sb->utime = sb->ctime;
1397 sb->events = __cpu_to_le64(1);
1398 if (info->state & (1<<MD_SB_CLEAN))
1399 sb->resync_offset = MaxSector;
1400 else
1401 sb->resync_offset = 0;
1402 sbsize = sizeof(struct mdp_superblock_1) + 2 * (info->raid_disks + spares);
1403 sbsize = ROUND_UP(sbsize, 512);
1404 sb->max_dev = __cpu_to_le32((sbsize - sizeof(struct mdp_superblock_1)) / 2);
1405
1406 memset(sb->dev_roles, 0xff, MAX_SB_SIZE - sizeof(struct mdp_superblock_1));
1407
1408 return 1;
1409 }
1410
1411 struct devinfo {
1412 int fd;
1413 char *devname;
1414 long long data_offset;
1415 mdu_disk_info_t disk;
1416 struct devinfo *next;
1417 };
1418 #ifndef MDASSEMBLE
1419 /* Add a device to the superblock being created */
1420 static int add_to_super1(struct supertype *st, mdu_disk_info_t *dk,
1421 int fd, char *devname, unsigned long long data_offset)
1422 {
1423 struct mdp_superblock_1 *sb = st->sb;
1424 __u16 *rp = sb->dev_roles + dk->number;
1425 struct devinfo *di, **dip;
1426
1427 if ((dk->state & 6) == 6) /* active, sync */
1428 *rp = __cpu_to_le16(dk->raid_disk);
1429 else if ((dk->state & ~2) == 0) /* active or idle -> spare */
1430 *rp = 0xffff;
1431 else
1432 *rp = 0xfffe;
1433
1434 if (dk->number >= (int)__le32_to_cpu(sb->max_dev) &&
1435 __le32_to_cpu(sb->max_dev) < MAX_DEVS)
1436 sb->max_dev = __cpu_to_le32(dk->number+1);
1437
1438 sb->dev_number = __cpu_to_le32(dk->number);
1439 sb->devflags = 0; /* don't copy another disks flags */
1440 sb->sb_csum = calc_sb_1_csum(sb);
1441
1442 dip = (struct devinfo **)&st->info;
1443 while (*dip)
1444 dip = &(*dip)->next;
1445 di = xmalloc(sizeof(struct devinfo));
1446 di->fd = fd;
1447 di->devname = devname;
1448 di->disk = *dk;
1449 di->data_offset = data_offset;
1450 di->next = NULL;
1451 *dip = di;
1452
1453 return 0;
1454 }
1455 #endif
1456
1457 static void locate_bitmap1(struct supertype *st, int fd);
1458
1459 static int store_super1(struct supertype *st, int fd)
1460 {
1461 struct mdp_superblock_1 *sb = st->sb;
1462 unsigned long long sb_offset;
1463 struct align_fd afd;
1464 int sbsize;
1465 unsigned long long dsize;
1466
1467 if (!get_dev_size(fd, NULL, &dsize))
1468 return 1;
1469
1470 dsize >>= 9;
1471
1472 if (dsize < 24)
1473 return 2;
1474
1475 init_afd(&afd, fd);
1476
1477 /*
1478 * Calculate the position of the superblock.
1479 * It is always aligned to a 4K boundary and
1480 * depending on minor_version, it can be:
1481 * 0: At least 8K, but less than 12K, from end of device
1482 * 1: At start of device
1483 * 2: 4K from start of device.
1484 */
1485 switch(st->minor_version) {
1486 case 0:
1487 sb_offset = dsize;
1488 sb_offset -= 8*2;
1489 sb_offset &= ~(4*2-1);
1490 break;
1491 case 1:
1492 sb_offset = 0;
1493 break;
1494 case 2:
1495 sb_offset = 4*2;
1496 break;
1497 default:
1498 return -EINVAL;
1499 }
1500
1501 if (sb_offset != __le64_to_cpu(sb->super_offset) &&
1502 0 != __le64_to_cpu(sb->super_offset)
1503 ) {
1504 pr_err("internal error - sb_offset is wrong\n");
1505 abort();
1506 }
1507
1508 if (lseek64(fd, sb_offset << 9, 0)< 0LL)
1509 return 3;
1510
1511 sbsize = ROUND_UP(sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev), 512);
1512
1513 if (awrite(&afd, sb, sbsize) != sbsize)
1514 return 4;
1515
1516 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1517 struct bitmap_super_s *bm = (struct bitmap_super_s*)
1518 (((char*)sb)+MAX_SB_SIZE);
1519 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC) {
1520 locate_bitmap1(st, fd);
1521 if (awrite(&afd, bm, sizeof(*bm)) != sizeof(*bm))
1522 return 5;
1523 }
1524 }
1525 fsync(fd);
1526 return 0;
1527 }
1528
1529 static int load_super1(struct supertype *st, int fd, char *devname);
1530
1531 static unsigned long choose_bm_space(unsigned long devsize)
1532 {
1533 /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1534 * if bigger than 200Gig, save 128k
1535 * NOTE: result must be multiple of 4K else bad things happen
1536 * on 4K-sector devices.
1537 */
1538 if (devsize < 64*2) return 0;
1539 if (devsize - 64*2 >= 200*1024*1024*2)
1540 return 128*2;
1541 if (devsize - 4*2 > 8*1024*1024*2)
1542 return 64*2;
1543 return 4*2;
1544 }
1545
1546 static void free_super1(struct supertype *st);
1547
1548 #ifndef MDASSEMBLE
1549 static int write_init_super1(struct supertype *st)
1550 {
1551 struct mdp_superblock_1 *sb = st->sb;
1552 struct supertype *refst;
1553 int rfd;
1554 int rv = 0;
1555 unsigned long long bm_space;
1556 struct devinfo *di;
1557 unsigned long long dsize, array_size;
1558 unsigned long long sb_offset;
1559 unsigned long long data_offset;
1560
1561 for (di = st->info; di; di = di->next) {
1562 if (di->disk.state & (1 << MD_DISK_FAULTY))
1563 continue;
1564 if (di->fd < 0)
1565 continue;
1566
1567 while (Kill(di->devname, NULL, 0, -1, 1) == 0)
1568 ;
1569
1570 sb->dev_number = __cpu_to_le32(di->disk.number);
1571 if (di->disk.state & (1<<MD_DISK_WRITEMOSTLY))
1572 sb->devflags |= WriteMostly1;
1573 else
1574 sb->devflags &= ~WriteMostly1;
1575
1576 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
1577 read(rfd, sb->device_uuid, 16) != 16) {
1578 __u32 r[4] = {random(), random(), random(), random()};
1579 memcpy(sb->device_uuid, r, 16);
1580 }
1581 if (rfd >= 0)
1582 close(rfd);
1583
1584 sb->events = 0;
1585
1586 refst = dup_super(st);
1587 if (load_super1(refst, di->fd, NULL)==0) {
1588 struct mdp_superblock_1 *refsb = refst->sb;
1589
1590 memcpy(sb->device_uuid, refsb->device_uuid, 16);
1591 if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) {
1592 /* same array, so preserve events and
1593 * dev_number */
1594 sb->events = refsb->events;
1595 /* bugs in 2.6.17 and earlier mean the
1596 * dev_number chosen in Manage must be preserved
1597 */
1598 if (get_linux_version() >= 2006018)
1599 sb->dev_number = refsb->dev_number;
1600 }
1601 free_super1(refst);
1602 }
1603 free(refst);
1604
1605 if (!get_dev_size(di->fd, NULL, &dsize)) {
1606 rv = 1;
1607 goto error_out;
1608 }
1609 dsize >>= 9;
1610
1611 if (dsize < 24) {
1612 close(di->fd);
1613 rv = 2;
1614 goto error_out;
1615 }
1616
1617 /*
1618 * Calculate the position of the superblock.
1619 * It is always aligned to a 4K boundary and
1620 * depending on minor_version, it can be:
1621 * 0: At least 8K, but less than 12K, from end of device
1622 * 1: At start of device
1623 * 2: 4K from start of device.
1624 * data_offset has already been set.
1625 */
1626 array_size = __le64_to_cpu(sb->size);
1627 /* work out how much space we left for a bitmap,
1628 * Add 8 sectors for bad block log */
1629 bm_space = choose_bm_space(array_size) + 8;
1630
1631 data_offset = di->data_offset;
1632 if (data_offset == INVALID_SECTORS)
1633 data_offset = st->data_offset;
1634 switch(st->minor_version) {
1635 case 0:
1636 if (data_offset == INVALID_SECTORS)
1637 data_offset = 0;
1638 sb_offset = dsize;
1639 sb_offset -= 8*2;
1640 sb_offset &= ~(4*2-1);
1641 sb->data_offset = __cpu_to_le64(data_offset);
1642 sb->super_offset = __cpu_to_le64(sb_offset);
1643 if (sb_offset < array_size + bm_space)
1644 bm_space = sb_offset - array_size;
1645 sb->data_size = __cpu_to_le64(sb_offset - bm_space);
1646 if (bm_space >= 8) {
1647 sb->bblog_size = __cpu_to_le16(8);
1648 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1649 }
1650 break;
1651 case 1:
1652 sb->super_offset = __cpu_to_le64(0);
1653 if (data_offset == INVALID_SECTORS)
1654 data_offset = 16;
1655
1656 sb->data_offset = __cpu_to_le64(data_offset);
1657 sb->data_size = __cpu_to_le64(dsize - data_offset);
1658 if (data_offset >= 8 + 32*2 + 8) {
1659 sb->bblog_size = __cpu_to_le16(8);
1660 sb->bblog_offset = __cpu_to_le32(8 + 32*2);
1661 } else if (data_offset >= 16) {
1662 sb->bblog_size = __cpu_to_le16(8);
1663 sb->bblog_offset = __cpu_to_le32(data_offset-8);
1664 }
1665 break;
1666 case 2:
1667 sb_offset = 4*2;
1668 sb->super_offset = __cpu_to_le64(sb_offset);
1669 if (data_offset == INVALID_SECTORS)
1670 data_offset = 24;
1671
1672 sb->data_offset = __cpu_to_le64(data_offset);
1673 sb->data_size = __cpu_to_le64(dsize - data_offset);
1674 if (data_offset >= 16 + 32*2 + 8) {
1675 sb->bblog_size = __cpu_to_le16(8);
1676 sb->bblog_offset = __cpu_to_le32(8 + 32*2);
1677 } else if (data_offset >= 16+16) {
1678 sb->bblog_size = __cpu_to_le16(8);
1679 /* '8' sectors for the bblog, and another '8'
1680 * because we want offset from superblock, not
1681 * start of device.
1682 */
1683 sb->bblog_offset = __cpu_to_le32(data_offset-8-8);
1684 }
1685 break;
1686 default:
1687 pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1688 st->minor_version, di->devname);
1689 rv = -EINVAL;
1690 goto out;
1691 }
1692 if (conf_get_create_info()->bblist == 0) {
1693 sb->bblog_size = 0;
1694 sb->bblog_offset = 0;
1695 }
1696
1697 sb->sb_csum = calc_sb_1_csum(sb);
1698 rv = store_super1(st, di->fd);
1699 if (rv == 0 && (__le32_to_cpu(sb->feature_map) & 1))
1700 rv = st->ss->write_bitmap(st, di->fd, NoUpdate);
1701 close(di->fd);
1702 di->fd = -1;
1703 if (rv)
1704 goto error_out;
1705 }
1706 error_out:
1707 if (rv)
1708 pr_err("Failed to write metadata to %s\n",
1709 di->devname);
1710 out:
1711 return rv;
1712 }
1713 #endif
1714
1715 static int compare_super1(struct supertype *st, struct supertype *tst)
1716 {
1717 /*
1718 * return:
1719 * 0 same, or first was empty, and second was copied
1720 * 1 second had wrong number
1721 * 2 wrong uuid
1722 * 3 wrong other info
1723 */
1724 struct mdp_superblock_1 *first = st->sb;
1725 struct mdp_superblock_1 *second = tst->sb;
1726
1727 if (second->magic != __cpu_to_le32(MD_SB_MAGIC))
1728 return 1;
1729 if (second->major_version != __cpu_to_le32(1))
1730 return 1;
1731
1732 if (!first) {
1733 if (posix_memalign((void**)&first, 4096, SUPER1_SIZE) != 0) {
1734 pr_err("could not allocate superblock\n");
1735 return 1;
1736 }
1737 memcpy(first, second, SUPER1_SIZE);
1738 st->sb = first;
1739 return 0;
1740 }
1741 if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0)
1742 return 2;
1743
1744 if (first->ctime != second->ctime ||
1745 first->level != second->level ||
1746 first->layout != second->layout ||
1747 first->size != second->size ||
1748 first->chunksize != second->chunksize ||
1749 first->raid_disks != second->raid_disks)
1750 return 3;
1751 return 0;
1752 }
1753
1754 static int load_super1(struct supertype *st, int fd, char *devname)
1755 {
1756 unsigned long long dsize;
1757 unsigned long long sb_offset;
1758 struct mdp_superblock_1 *super;
1759 int uuid[4];
1760 struct bitmap_super_s *bsb;
1761 struct misc_dev_info *misc;
1762 struct align_fd afd;
1763
1764 free_super1(st);
1765
1766 init_afd(&afd, fd);
1767
1768 if (st->ss == NULL || st->minor_version == -1) {
1769 int bestvers = -1;
1770 struct supertype tst;
1771 __u64 bestctime = 0;
1772 /* guess... choose latest ctime */
1773 memset(&tst, 0, sizeof(tst));
1774 tst.ss = &super1;
1775 for (tst.minor_version = 0; tst.minor_version <= 2 ; tst.minor_version++) {
1776 switch(load_super1(&tst, fd, devname)) {
1777 case 0: super = tst.sb;
1778 if (bestvers == -1 ||
1779 bestctime < __le64_to_cpu(super->ctime)) {
1780 bestvers = tst.minor_version;
1781 bestctime = __le64_to_cpu(super->ctime);
1782 }
1783 free(super);
1784 tst.sb = NULL;
1785 break;
1786 case 1: return 1; /*bad device */
1787 case 2: break; /* bad, try next */
1788 }
1789 }
1790 if (bestvers != -1) {
1791 int rv;
1792 tst.minor_version = bestvers;
1793 tst.ss = &super1;
1794 tst.max_devs = MAX_DEVS;
1795 rv = load_super1(&tst, fd, devname);
1796 if (rv == 0)
1797 *st = tst;
1798 return rv;
1799 }
1800 return 2;
1801 }
1802 if (!get_dev_size(fd, devname, &dsize))
1803 return 1;
1804 dsize >>= 9;
1805
1806 if (dsize < 24) {
1807 if (devname)
1808 pr_err("%s is too small for md: size is %llu sectors.\n",
1809 devname, dsize);
1810 return 1;
1811 }
1812
1813 /*
1814 * Calculate the position of the superblock.
1815 * It is always aligned to a 4K boundary and
1816 * depending on minor_version, it can be:
1817 * 0: At least 8K, but less than 12K, from end of device
1818 * 1: At start of device
1819 * 2: 4K from start of device.
1820 */
1821 switch(st->minor_version) {
1822 case 0:
1823 sb_offset = dsize;
1824 sb_offset -= 8*2;
1825 sb_offset &= ~(4*2-1);
1826 break;
1827 case 1:
1828 sb_offset = 0;
1829 break;
1830 case 2:
1831 sb_offset = 4*2;
1832 break;
1833 default:
1834 return -EINVAL;
1835 }
1836
1837 if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
1838 if (devname)
1839 pr_err("Cannot seek to superblock on %s: %s\n",
1840 devname, strerror(errno));
1841 return 1;
1842 }
1843
1844 if (posix_memalign((void**)&super, 4096, SUPER1_SIZE) != 0) {
1845 pr_err("could not allocate superblock\n");
1846 return 1;
1847 }
1848
1849 if (aread(&afd, super, MAX_SB_SIZE) != MAX_SB_SIZE) {
1850 if (devname)
1851 pr_err("Cannot read superblock on %s\n",
1852 devname);
1853 free(super);
1854 return 1;
1855 }
1856
1857 if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) {
1858 if (devname)
1859 pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
1860 devname, MD_SB_MAGIC, __le32_to_cpu(super->magic));
1861 free(super);
1862 return 2;
1863 }
1864
1865 if (__le32_to_cpu(super->major_version) != 1) {
1866 if (devname)
1867 pr_err("Cannot interpret superblock on %s - version is %d\n",
1868 devname, __le32_to_cpu(super->major_version));
1869 free(super);
1870 return 2;
1871 }
1872 if (__le64_to_cpu(super->super_offset) != sb_offset) {
1873 if (devname)
1874 pr_err("No superblock found on %s (super_offset is wrong)\n",
1875 devname);
1876 free(super);
1877 return 2;
1878 }
1879 st->sb = super;
1880
1881 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
1882
1883 misc = (struct misc_dev_info*) (((char*)super)+MAX_SB_SIZE+BM_SUPER_SIZE);
1884 misc->device_size = dsize;
1885 if (st->data_offset == INVALID_SECTORS)
1886 st->data_offset = __le64_to_cpu(super->data_offset);
1887
1888 /* Now check on the bitmap superblock */
1889 if ((__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) == 0)
1890 return 0;
1891 /* Read the bitmap superblock and make sure it looks
1892 * valid. If it doesn't clear the bit. An --assemble --force
1893 * should get that written out.
1894 */
1895 locate_bitmap1(st, fd);
1896 if (aread(&afd, bsb, 512) != 512)
1897 goto no_bitmap;
1898
1899 uuid_from_super1(st, uuid);
1900 if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
1901 memcmp(bsb->uuid, uuid, 16) != 0)
1902 goto no_bitmap;
1903 return 0;
1904
1905 no_bitmap:
1906 super->feature_map = __cpu_to_le32(__le32_to_cpu(super->feature_map)
1907 & ~MD_FEATURE_BITMAP_OFFSET);
1908 return 0;
1909 }
1910
1911 static struct supertype *match_metadata_desc1(char *arg)
1912 {
1913 struct supertype *st = xcalloc(1, sizeof(*st));
1914
1915 st->container_devnm[0] = 0;
1916 st->ss = &super1;
1917 st->max_devs = MAX_DEVS;
1918 st->sb = NULL;
1919 st->data_offset = INVALID_SECTORS;
1920 /* leading zeros can be safely ignored. --detail generates them. */
1921 while (*arg == '0')
1922 arg++;
1923 if (strcmp(arg, "1.0") == 0 ||
1924 strcmp(arg, "1.00") == 0) {
1925 st->minor_version = 0;
1926 return st;
1927 }
1928 if (strcmp(arg, "1.1") == 0 ||
1929 strcmp(arg, "1.01") == 0
1930 ) {
1931 st->minor_version = 1;
1932 return st;
1933 }
1934 if (strcmp(arg, "1.2") == 0 ||
1935 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
1936 strcmp(arg, "default") == 0 ||
1937 #endif /* DEFAULT_OLD_METADATA */
1938 strcmp(arg, "1.02") == 0) {
1939 st->minor_version = 2;
1940 return st;
1941 }
1942 if (strcmp(arg, "1") == 0 ||
1943 strcmp(arg, "default") == 0) {
1944 st->minor_version = -1;
1945 return st;
1946 }
1947
1948 free(st);
1949 return NULL;
1950 }
1951
1952 /* find available size on device with this devsize, using
1953 * superblock type st, and reserving 'reserve' sectors for
1954 * a possible bitmap
1955 */
1956 static __u64 avail_size1(struct supertype *st, __u64 devsize,
1957 unsigned long long data_offset)
1958 {
1959 struct mdp_superblock_1 *super = st->sb;
1960 int bmspace = 0;
1961 int bbspace = 0;
1962 if (devsize < 24)
1963 return 0;
1964
1965 #ifndef MDASSEMBLE
1966 if (__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) {
1967 /* hot-add. allow for actual size of bitmap */
1968 struct bitmap_super_s *bsb;
1969 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
1970 bmspace = bitmap_sectors(bsb);
1971 }
1972 #endif
1973 /* Allow space for bad block log */
1974 if (super->bblog_size)
1975 bbspace = __le16_to_cpu(super->bblog_size);
1976
1977 if (st->minor_version < 0)
1978 /* not specified, so time to set default */
1979 st->minor_version = 2;
1980
1981 if (data_offset == INVALID_SECTORS)
1982 data_offset = st->data_offset;
1983
1984 if (data_offset != INVALID_SECTORS)
1985 switch(st->minor_version) {
1986 case 0:
1987 return devsize - data_offset - 8*2 - bbspace;
1988 case 1:
1989 case 2:
1990 return devsize - data_offset;
1991 default:
1992 return 0;
1993 }
1994
1995 devsize -= bmspace;
1996
1997 switch(st->minor_version) {
1998 case 0:
1999 /* at end */
2000 return ((devsize - 8*2 - bbspace ) & ~(4*2-1));
2001 case 1:
2002 /* at start, 4K for superblock and possible bitmap */
2003 return devsize - 4*2 - bbspace;
2004 case 2:
2005 /* 4k from start, 4K for superblock and possible bitmap */
2006 return devsize - (4+4)*2 - bbspace;
2007 }
2008 return 0;
2009 }
2010
2011 static int
2012 add_internal_bitmap1(struct supertype *st,
2013 int *chunkp, int delay, int write_behind,
2014 unsigned long long size,
2015 int may_change, int major)
2016 {
2017 /*
2018 * If not may_change, then this is a 'Grow' without sysfs support for
2019 * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2020 * If may_change, then this is create or a Grow with sysfs syupport,
2021 * and we can put the bitmap wherever we like.
2022 *
2023 * size is in sectors, chunk is in bytes !!!
2024 */
2025
2026 unsigned long long bits;
2027 unsigned long long max_bits;
2028 unsigned long long min_chunk;
2029 long offset;
2030 long bbl_offset, bbl_size;
2031 unsigned long long chunk = *chunkp;
2032 int room = 0;
2033 int creating = 0;
2034 struct mdp_superblock_1 *sb = st->sb;
2035 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
2036 int uuid[4];
2037
2038 if (__le64_to_cpu(sb->data_size) == 0)
2039 /* Must be creating the array, else data_size would be non-zero */
2040 creating = 1;
2041 switch(st->minor_version) {
2042 case 0:
2043 /* either 3K after the superblock (when hot-add),
2044 * or some amount of space before.
2045 */
2046 if (creating) {
2047 /* We are creating array, so we *know* how much room has
2048 * been left.
2049 */
2050 offset = 0;
2051 bbl_size = 8;
2052 room = choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2053 } else {
2054 room = __le64_to_cpu(sb->super_offset)
2055 - __le64_to_cpu(sb->data_offset)
2056 - __le64_to_cpu(sb->data_size);
2057 bbl_size = __le16_to_cpu(sb->bblog_size);
2058 if (bbl_size < 8)
2059 bbl_size = 8;
2060 bbl_offset = (__s32)__le32_to_cpu(sb->bblog_offset);
2061 if (bbl_size < -bbl_offset)
2062 bbl_size = -bbl_offset;
2063
2064 if (!may_change || (room < 3*2 &&
2065 __le32_to_cpu(sb->max_dev) <= 384)) {
2066 room = 3*2;
2067 offset = 1*2;
2068 bbl_size = 0;
2069 } else {
2070 offset = 0; /* means movable offset */
2071 }
2072 }
2073 break;
2074 case 1:
2075 case 2: /* between superblock and data */
2076 if (creating) {
2077 offset = 4*2;
2078 bbl_size = 8;
2079 room = choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2080 } else {
2081 room = __le64_to_cpu(sb->data_offset)
2082 - __le64_to_cpu(sb->super_offset);
2083 bbl_size = __le16_to_cpu(sb->bblog_size);
2084 if (bbl_size)
2085 room = __le32_to_cpu(sb->bblog_offset) + bbl_size;
2086 else
2087 bbl_size = 8;
2088
2089 if (!may_change) {
2090 room -= 2; /* Leave 1K for superblock */
2091 offset = 2;
2092 bbl_size = 0;
2093 } else {
2094 room -= 4*2; /* leave 4K for superblock */
2095 offset = 4*2;
2096 }
2097 }
2098 break;
2099 default:
2100 return 0;
2101 }
2102
2103 room -= bbl_size;
2104 if (chunk == UnSet && room > 128*2)
2105 /* Limit to 128K of bitmap when chunk size not requested */
2106 room = 128*2;
2107
2108 if (room <= 1)
2109 /* No room for a bitmap */
2110 return 0;
2111
2112 max_bits = (room * 512 - sizeof(bitmap_super_t)) * 8;
2113
2114 min_chunk = 4096; /* sub-page chunks don't work yet.. */
2115 bits = (size*512)/min_chunk +1;
2116 while (bits > max_bits) {
2117 min_chunk *= 2;
2118 bits = (bits+1)/2;
2119 }
2120 if (chunk == UnSet) {
2121 /* For practical purpose, 64Meg is a good
2122 * default chunk size for internal bitmaps.
2123 */
2124 chunk = min_chunk;
2125 if (chunk < 64*1024*1024)
2126 chunk = 64*1024*1024;
2127 } else if (chunk < min_chunk)
2128 return 0; /* chunk size too small */
2129 if (chunk == 0) /* rounding problem */
2130 return 0;
2131
2132 if (offset == 0) {
2133 /* start bitmap on a 4K boundary with enough space for
2134 * the bitmap
2135 */
2136 bits = (size*512) / chunk + 1;
2137 room = ((bits+7)/8 + sizeof(bitmap_super_t) +4095)/4096;
2138 room *= 8; /* convert 4K blocks to sectors */
2139 offset = -room - bbl_size;
2140 }
2141
2142 sb->bitmap_offset = (int32_t)__cpu_to_le32(offset);
2143
2144 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map)
2145 | MD_FEATURE_BITMAP_OFFSET);
2146 memset(bms, 0, sizeof(*bms));
2147 bms->magic = __cpu_to_le32(BITMAP_MAGIC);
2148 bms->version = __cpu_to_le32(major);
2149 uuid_from_super1(st, uuid);
2150 memcpy(bms->uuid, uuid, 16);
2151 bms->chunksize = __cpu_to_le32(chunk);
2152 bms->daemon_sleep = __cpu_to_le32(delay);
2153 bms->sync_size = __cpu_to_le64(size);
2154 bms->write_behind = __cpu_to_le32(write_behind);
2155 bms->nodes = __cpu_to_le32(st->nodes);
2156 if (st->cluster_name)
2157 strncpy((char *)bms->cluster_name,
2158 st->cluster_name, strlen(st->cluster_name));
2159
2160 *chunkp = chunk;
2161 return 1;
2162 }
2163
2164 static void locate_bitmap1(struct supertype *st, int fd)
2165 {
2166 unsigned long long offset;
2167 struct mdp_superblock_1 *sb;
2168 int mustfree = 0;
2169
2170 if (!st->sb) {
2171 if (st->ss->load_super(st, fd, NULL))
2172 return; /* no error I hope... */
2173 mustfree = 1;
2174 }
2175 sb = st->sb;
2176
2177 offset = __le64_to_cpu(sb->super_offset);
2178 offset += (int32_t) __le32_to_cpu(sb->bitmap_offset);
2179 if (mustfree)
2180 free(sb);
2181 lseek64(fd, offset<<9, 0);
2182 }
2183
2184 static int write_bitmap1(struct supertype *st, int fd, enum bitmap_update update)
2185 {
2186 struct mdp_superblock_1 *sb = st->sb;
2187 bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
2188 int rv = 0;
2189 void *buf;
2190 int towrite, n;
2191 struct align_fd afd;
2192 unsigned int i = 0;
2193
2194 switch (update) {
2195 case NameUpdate:
2196 /* update cluster name */
2197 if (st->cluster_name) {
2198 memset((char *)bms->cluster_name, 0, sizeof(bms->cluster_name));
2199 strncpy((char *)bms->cluster_name, st->cluster_name, 64);
2200 }
2201 break;
2202 case NoUpdate:
2203 default:
2204 break;
2205 }
2206
2207 init_afd(&afd, fd);
2208
2209 locate_bitmap1(st, fd);
2210
2211 if (posix_memalign(&buf, 4096, 4096))
2212 return -ENOMEM;
2213
2214 do {
2215 /* Only the bitmap[0] should resync
2216 * whole device on initial assembly
2217 */
2218 if (i)
2219 memset(buf, 0x00, 4096);
2220 else
2221 memset(buf, 0xff, 4096);
2222 memcpy(buf, (char *)bms, sizeof(bitmap_super_t));
2223
2224 towrite = __le64_to_cpu(bms->sync_size) / (__le32_to_cpu(bms->chunksize)>>9);
2225 towrite = (towrite+7) >> 3; /* bits to bytes */
2226 towrite += sizeof(bitmap_super_t);
2227 /* we need the bitmaps to be at 4k boundary */
2228 towrite = ROUND_UP(towrite, 4096);
2229 while (towrite > 0) {
2230 n = towrite;
2231 if (n > 4096)
2232 n = 4096;
2233 n = awrite(&afd, buf, n);
2234 if (n > 0)
2235 towrite -= n;
2236 else
2237 break;
2238 if (i)
2239 memset(buf, 0x00, 4096);
2240 else
2241 memset(buf, 0xff, 4096);
2242 }
2243 fsync(fd);
2244 if (towrite) {
2245 rv = -2;
2246 break;
2247 }
2248 } while (++i < __le32_to_cpu(bms->nodes));
2249
2250 free(buf);
2251 return rv;
2252 }
2253
2254 static void free_super1(struct supertype *st)
2255 {
2256 if (st->sb)
2257 free(st->sb);
2258 while (st->info) {
2259 struct devinfo *di = st->info;
2260 st->info = di->next;
2261 if (di->fd >= 0)
2262 close(di->fd);
2263 free(di);
2264 }
2265 st->sb = NULL;
2266 }
2267
2268 #ifndef MDASSEMBLE
2269 static int validate_geometry1(struct supertype *st, int level,
2270 int layout, int raiddisks,
2271 int *chunk, unsigned long long size,
2272 unsigned long long data_offset,
2273 char *subdev, unsigned long long *freesize,
2274 int verbose)
2275 {
2276 unsigned long long ldsize, devsize;
2277 int bmspace;
2278 unsigned long long headroom;
2279 int fd;
2280
2281 if (level == LEVEL_CONTAINER) {
2282 if (verbose)
2283 pr_err("1.x metadata does not support containers\n");
2284 return 0;
2285 }
2286 if (*chunk == UnSet)
2287 *chunk = DEFAULT_CHUNK;
2288
2289 if (!subdev)
2290 return 1;
2291
2292 if (st->minor_version < 0)
2293 /* not specified, so time to set default */
2294 st->minor_version = 2;
2295
2296 fd = open(subdev, O_RDONLY|O_EXCL, 0);
2297 if (fd < 0) {
2298 if (verbose)
2299 pr_err("super1.x cannot open %s: %s\n",
2300 subdev, strerror(errno));
2301 return 0;
2302 }
2303
2304 if (!get_dev_size(fd, subdev, &ldsize)) {
2305 close(fd);
2306 return 0;
2307 }
2308 close(fd);
2309
2310 devsize = ldsize >> 9;
2311 if (devsize < 24) {
2312 *freesize = 0;
2313 return 0;
2314 }
2315
2316 /* creating: allow suitable space for bitmap */
2317 bmspace = choose_bm_space(devsize);
2318
2319 if (data_offset == INVALID_SECTORS)
2320 data_offset = st->data_offset;
2321 if (data_offset == INVALID_SECTORS)
2322 switch (st->minor_version) {
2323 case 0:
2324 data_offset = 0;
2325 break;
2326 case 1:
2327 case 2:
2328 /* Choose data offset appropriate for this device
2329 * and use as default for whole array.
2330 * The data_offset must allow for bitmap space
2331 * and base metadata, should allow for some headroom
2332 * for reshape, and should be rounded to multiple
2333 * of 1M.
2334 * Headroom is limited to 128M, but aim for about 0.1%
2335 */
2336 headroom = 128*1024*2;
2337 while ((headroom << 10) > devsize &&
2338 (*chunk == 0 ||
2339 headroom / 2 >= ((unsigned)(*chunk)*2)*2))
2340 headroom >>= 1;
2341 data_offset = 12*2 + bmspace + headroom;
2342 #define ONE_MEG (2*1024)
2343 if (data_offset > ONE_MEG)
2344 data_offset = (data_offset / ONE_MEG) * ONE_MEG;
2345 break;
2346 }
2347 if (st->data_offset == INVALID_SECTORS)
2348 st->data_offset = data_offset;
2349 switch(st->minor_version) {
2350 case 0: /* metadata at end. Round down and subtract space to reserve */
2351 devsize = (devsize & ~(4ULL*2-1));
2352 /* space for metadata, bblog, bitmap */
2353 devsize -= 8*2 + 8 + bmspace;
2354 break;
2355 case 1:
2356 case 2:
2357 devsize -= data_offset;
2358 break;
2359 }
2360 *freesize = devsize;
2361 return 1;
2362 }
2363 #endif /* MDASSEMBLE */
2364
2365 void *super1_make_v0(struct supertype *st, struct mdinfo *info, mdp_super_t *sb0)
2366 {
2367 /* Create a v1.0 superblock based on 'info'*/
2368 void *ret;
2369 struct mdp_superblock_1 *sb;
2370 int i;
2371 int rfd;
2372 unsigned long long offset;
2373
2374 if (posix_memalign(&ret, 4096, 1024) != 0)
2375 return NULL;
2376 sb = ret;
2377 memset(ret, 0, 1024);
2378 sb->magic = __cpu_to_le32(MD_SB_MAGIC);
2379 sb->major_version = __cpu_to_le32(1);
2380
2381 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
2382 sprintf(sb->set_name, "%d", sb0->md_minor);
2383 sb->ctime = __cpu_to_le32(info->array.ctime+1);
2384 sb->level = __cpu_to_le32(info->array.level);
2385 sb->layout = __cpu_to_le32(info->array.layout);
2386 sb->size = __cpu_to_le64(info->component_size);
2387 sb->chunksize = __cpu_to_le32(info->array.chunk_size/512);
2388 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
2389 if (info->array.level > 0)
2390 sb->data_size = sb->size;
2391 else
2392 sb->data_size = st->ss->avail_size(st, st->devsize/512, 0);
2393 sb->resync_offset = MaxSector;
2394 sb->max_dev = __cpu_to_le32(MD_SB_DISKS);
2395 sb->dev_number = __cpu_to_le32(info->disk.number);
2396 sb->utime = __cpu_to_le64(info->array.utime);
2397
2398 offset = st->devsize/512 - 8*2;
2399 offset &= ~(4*2-1);
2400 sb->super_offset = __cpu_to_le64(offset);
2401 //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2402
2403 if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
2404 read(rfd, sb->device_uuid, 16) != 16) {
2405 __u32 r[4] = {random(), random(), random(), random()};
2406 memcpy(sb->device_uuid, r, 16);
2407 }
2408 if (rfd >= 0)
2409 close(rfd);
2410
2411 for (i = 0; i < MD_SB_DISKS; i++) {
2412 int state = sb0->disks[i].state;
2413 sb->dev_roles[i] = 0xFFFF;
2414 if ((state & (1<<MD_DISK_SYNC)) &&
2415 !(state & (1<<MD_DISK_FAULTY)))
2416 sb->dev_roles[i] = __cpu_to_le16(sb0->disks[i].raid_disk);
2417 }
2418 sb->sb_csum = calc_sb_1_csum(sb);
2419 return ret;
2420 }
2421
2422 struct superswitch super1 = {
2423 #ifndef MDASSEMBLE
2424 .examine_super = examine_super1,
2425 .brief_examine_super = brief_examine_super1,
2426 .export_examine_super = export_examine_super1,
2427 .detail_super = detail_super1,
2428 .brief_detail_super = brief_detail_super1,
2429 .export_detail_super = export_detail_super1,
2430 .write_init_super = write_init_super1,
2431 .validate_geometry = validate_geometry1,
2432 .add_to_super = add_to_super1,
2433 .examine_badblocks = examine_badblocks_super1,
2434 .copy_metadata = copy_metadata1,
2435 #endif
2436 .match_home = match_home1,
2437 .uuid_from_super = uuid_from_super1,
2438 .getinfo_super = getinfo_super1,
2439 .container_content = container_content1,
2440 .update_super = update_super1,
2441 .init_super = init_super1,
2442 .store_super = store_super1,
2443 .compare_super = compare_super1,
2444 .load_super = load_super1,
2445 .match_metadata_desc = match_metadata_desc1,
2446 .avail_size = avail_size1,
2447 .add_internal_bitmap = add_internal_bitmap1,
2448 .locate_bitmap = locate_bitmap1,
2449 .write_bitmap = write_bitmap1,
2450 .free_super = free_super1,
2451 #if __BYTE_ORDER == BIG_ENDIAN
2452 .swapuuid = 0,
2453 #else
2454 .swapuuid = 1,
2455 #endif
2456 .name = "1.x",
2457 };
Unnamed repository; edit this file 'description' to name the repository.