super0.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 #define HAVE_STDINT_H 1
  26 #include "mdadm.h"
  27 #include "sha1.h"
  28 /*
  29  * All handling for the 0.90.0 version superblock is in
  30  * this file.
  31  * This includes:
  32  *   - finding, loading, and writing the superblock.
  33  *   - initialising a new superblock
  34  *   - printing the superblock for --examine
  35  *   - printing part of the superblock for --detail
  36  * .. other stuff
  37  */
  38
  39 static unsigned long calc_sb0_csum(mdp_super_t *super)
  40 {
  41         unsigned long csum = super->sb_csum;
  42         unsigned long newcsum;
  43         super->sb_csum= 0 ;
  44         newcsum = calc_csum(super, MD_SB_BYTES);
  45         super->sb_csum = csum;
  46         return newcsum;
  47 }
  48
  49 static void super0_swap_endian(struct mdp_superblock_s *sb)
  50 {
  51         /* as super0 superblocks are host-endian, it is sometimes
  52          * useful to be able to swap the endianness
  53          * as (almost) everything is u32's we byte-swap every 4byte
  54          * number.
  55          * We then also have to swap the events_hi and events_lo
  56          */
  57         char *sbc = (char *)sb;
  58         __u32 t32;
  59         int i;
  60
  61         for (i=0; i < MD_SB_BYTES ; i+=4) {
  62                 char t = sbc[i];
  63                 sbc[i] = sbc[i+3];
  64                 sbc[i+3] = t;
  65                 t=sbc[i+1];
  66                 sbc[i+1]=sbc[i+2];
  67                 sbc[i+2]=t;
  68         }
  69         t32 = sb->events_hi;
  70         sb->events_hi = sb->events_lo;
  71         sb->events_lo = t32;
  72
  73         t32 = sb->cp_events_hi;
  74         sb->cp_events_hi = sb->cp_events_lo;
  75         sb->cp_events_lo = t32;
  76
  77 }
  78
  79 #ifndef MDASSEMBLE
  80
  81 static void examine_super0(struct supertype *st, char *homehost)
  82 {
  83         mdp_super_t *sb = st->sb;
  84         time_t atime;
  85         int d;
  86         int delta_extra = 0;
  87         char *c;
  88
  89         printf("          Magic : %08x\n", sb->md_magic);
  90         printf("        Version : %d.%02d.%02d\n",
  91                sb->major_version, sb->minor_version, sb->patch_version);
  92         if (sb->minor_version >= 90) {
  93                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0,
  94                        sb->set_uuid1, sb->set_uuid2, sb->set_uuid3);
  95                 if (homehost) {
  96                         char buf[20];
  97                         void *hash;
  98
  99                         hash = sha1_buffer(homehost, strlen(homehost), buf);
 100                         if (memcmp(&sb->set_uuid2, hash, 8) == 0)
 101                                 printf(" (local to host %s)", homehost);
 102                 }
 103                 printf("\n");
 104         } else
 105                 printf("           UUID : %08x\n", sb->set_uuid0);
 106
 107         if (sb->not_persistent)
 108                 printf("           Eedk : not persistent\n");
 109
 110         atime = sb->ctime;
 111         printf("  Creation Time : %.24s\n", ctime(&atime));
 112         c = map_num(pers, sb->level);
 113         printf("     Raid Level : %s\n", c?c:"-unknown-");
 114         if ((int)sb->level > 0) {
 115                 int ddsks = 0, ddsks_denom = 1;
 116                 printf("  Used Dev Size : %d%s\n", sb->size,
 117                        human_size((long long)sb->size<<10));
 118                 switch(sb->level) {
 119                 case 1:
 120                         ddsks=1;
 121                         break;
 122                 case 4:
 123                 case 5:
 124                         ddsks = sb->raid_disks - 1;
 125                         break;
 126                 case 6:
 127                         ddsks = sb->raid_disks - 2;
 128                         break;
 129                 case 10:
 130                         ddsks = sb->raid_disks;
 131                         ddsks_denom =
 132                                 (sb->layout & 255) * ((sb->layout >> 8) & 255);
 133                 }
 134                 if (ddsks) {
 135                         long long asize = sb->size;
 136                         asize = (asize << 10) * ddsks / ddsks_denom;
 137                         printf("     Array Size : %llu%s\n",
 138                                asize >> 10,  human_size(asize));
 139                 }
 140         }
 141         printf("   Raid Devices : %d\n", sb->raid_disks);
 142         printf("  Total Devices : %d\n", sb->nr_disks);
 143         printf("Preferred Minor : %d\n", sb->md_minor);
 144         printf("\n");
 145         if (sb->minor_version > 90 && (sb->reshape_position + 1) != 0) {
 146                 printf("  Reshape pos'n : %llu%s\n",
 147                        (unsigned long long)sb->reshape_position / 2,
 148                        human_size((long long)sb->reshape_position << 9));
 149                 if (sb->delta_disks) {
 150                         printf("  Delta Devices : %d", sb->delta_disks);
 151                         printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks,
 152                                sb->raid_disks);
 153                         if (((int)sb->delta_disks) < 0)
 154                                 delta_extra = - sb->delta_disks;
 155                 }
 156                 if (sb->new_level != sb->level) {
 157                         c = map_num(pers, sb->new_level);
 158                         printf("      New Level : %s\n", c?c:"-unknown-");
 159                 }
 160                 if (sb->new_layout != sb->layout) {
 161                         if (sb->level == 5) {
 162                                 c = map_num(r5layout, sb->new_layout);
 163                                 printf("     New Layout : %s\n",
 164                                        c?c:"-unknown-");
 165                         }
 166                         if (sb->level == 6) {
 167                                 c = map_num(r6layout, sb->new_layout);
 168                                 printf("     New Layout : %s\n",
 169                                        c?c:"-unknown-");
 170                         }
 171                         if (sb->level == 10) {
 172                                 printf("     New Layout : near=%d, %s=%d\n",
 173                                        sb->new_layout&255,
 174                                        (sb->new_layout&0x10000)?"offset":"far",
 175                                        (sb->new_layout>>8)&255);
 176                         }
 177                 }
 178                 if (sb->new_chunk != sb->chunk_size)
 179                         printf("  New Chunksize : %d\n", sb->new_chunk);
 180                 printf("\n");
 181         }
 182         atime = sb->utime;
 183         printf("    Update Time : %.24s\n", ctime(&atime));
 184         printf("          State : %s\n",
 185                (sb->state&(1 << MD_SB_CLEAN)) ? "clean":"active");
 186         if (sb->state & (1 << MD_SB_BITMAP_PRESENT))
 187                 printf("Internal Bitmap : present\n");
 188         printf(" Active Devices : %d\n", sb->active_disks);
 189         printf("Working Devices : %d\n", sb->working_disks);
 190         printf(" Failed Devices : %d\n", sb->failed_disks);
 191         printf("  Spare Devices : %d\n", sb->spare_disks);
 192         if (calc_sb0_csum(sb) == sb->sb_csum)
 193                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 194         else
 195                 printf("       Checksum : %x - expected %lx\n",
 196                        sb->sb_csum, calc_sb0_csum(sb));
 197         printf("         Events : %llu\n",
 198                ((unsigned long long)sb->events_hi << 32) + sb->events_lo);
 199         printf("\n");
 200         if (sb->level == 5) {
 201                 c = map_num(r5layout, sb->layout);
 202                 printf("         Layout : %s\n", c?c:"-unknown-");
 203         }
 204         if (sb->level == 6) {
 205                 c = map_num(r6layout, sb->layout);
 206                 printf("         Layout : %s\n", c?c:"-unknown-");
 207         }
 208         if (sb->level == 10) {
 209                 printf("         Layout :");
 210                 print_r10_layout(sb->layout);
 211                 printf("\n");
 212         }
 213         switch(sb->level) {
 214         case 0:
 215         case 4:
 216         case 5:
 217         case 6:
 218         case 10:
 219                 printf("     Chunk Size : %dK\n", sb->chunk_size / 1024);
 220                 break;
 221         case -1:
 222                 printf("       Rounding : %dK\n", sb->chunk_size / 1024);
 223                 break;
 224         default:
 225                 break;
 226         }
 227         printf("\n");
 228         printf("      Number   Major   Minor   RaidDevice State\n");
 229         for (d = -1;
 230              d < (signed int)(sb->raid_disks + delta_extra + sb->spare_disks);
 231              d++) {
 232                 mdp_disk_t *dp;
 233                 char *dv;
 234                 char nb[5];
 235                 int wonly;
 236                 if (d>=0) dp = &sb->disks[d];
 237                 else dp = &sb->this_disk;
 238                 snprintf(nb, sizeof(nb), "%4d", d);
 239                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" : nb,
 240                        dp->number, dp->major, dp->minor, dp->raid_disk);
 241                 wonly = dp->state & (1 << MD_DISK_WRITEMOSTLY);
 242                 dp->state &= ~(1 << MD_DISK_WRITEMOSTLY);
 243                 if (dp->state & (1 << MD_DISK_FAULTY))
 244                         printf(" faulty");
 245                 if (dp->state & (1 << MD_DISK_ACTIVE))
 246                         printf(" active");
 247                 if (dp->state & (1 << MD_DISK_SYNC))
 248                         printf(" sync");
 249                 if (dp->state & (1 << MD_DISK_REMOVED))
 250                         printf(" removed");
 251                 if (wonly)
 252                         printf(" write-mostly");
 253                 if (dp->state == 0)
 254                         printf(" spare");
 255                 if ((dv = map_dev(dp->major, dp->minor, 0)))
 256                         printf("   %s", dv);
 257                 printf("\n");
 258                 if (d == -1)
 259                         printf("\n");
 260         }
 261 }
 262
 263 static void brief_examine_super0(struct supertype *st, int verbose)
 264 {
 265         mdp_super_t *sb = st->sb;
 266         char *c=map_num(pers, sb->level);
 267         char devname[20];
 268
 269         sprintf(devname, "/dev/md%d", sb->md_minor);
 270
 271         if (verbose) {
 272                 printf("ARRAY %s level=%s num-devices=%d",
 273                        devname,
 274                        c?c:"-unknown-", sb->raid_disks);
 275         } else
 276                 printf("ARRAY %s", devname);
 277
 278         if (sb->minor_version >= 90)
 279                 printf(" UUID=%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 280                        sb->set_uuid2, sb->set_uuid3);
 281         else
 282                 printf(" UUID=%08x", sb->set_uuid0);
 283         printf("\n");
 284 }
 285
 286 static void export_examine_super0(struct supertype *st)
 287 {
 288         mdp_super_t *sb = st->sb;
 289
 290         printf("MD_LEVEL=%s\n", map_num(pers, sb->level));
 291         printf("MD_DEVICES=%d\n", sb->raid_disks);
 292         if (sb->minor_version >= 90)
 293                 printf("MD_UUID=%08x:%08x:%08x:%08x\n",
 294                        sb->set_uuid0, sb->set_uuid1,
 295                        sb->set_uuid2, sb->set_uuid3);
 296         else
 297                 printf("MD_UUID=%08x\n", sb->set_uuid0);
 298         printf("MD_UPDATE_TIME=%llu\n",
 299                __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL);
 300         printf("MD_EVENTS=%llu\n",
 301                ((unsigned long long)sb->events_hi << 32)
 302                + sb->events_lo);
 303 }
 304
 305 static int copy_metadata0(struct supertype *st, int from, int to)
 306 {
 307         /* Read 64K from the appropriate offset of 'from'
 308          * and if it looks a little like a 0.90 superblock,
 309          * write it to the same offset of 'to'
 310          */
 311         void *buf;
 312         unsigned long long dsize, offset;
 313         const int bufsize = 64*1024;
 314         mdp_super_t *super;
 315
 316         if (posix_memalign(&buf, 4096, bufsize) != 0)
 317                 return 1;
 318
 319         if (!get_dev_size(from, NULL, &dsize))
 320                 goto err;
 321
 322         if (dsize < MD_RESERVED_SECTORS*512)
 323                 goto err;
 324
 325         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 326
 327         offset *= 512;
 328
 329         if (lseek64(from, offset, 0) < 0LL)
 330                 goto err;
 331         if (read(from, buf, bufsize) != bufsize)
 332                 goto err;
 333
 334         if (lseek64(to, offset, 0) < 0LL)
 335                 goto err;
 336         super = buf;
 337         if (super->md_magic != MD_SB_MAGIC ||
 338             super->major_version != 0 ||
 339             calc_sb0_csum(super) != super->sb_csum)
 340                 goto err;
 341         if (write(to, buf, bufsize) != bufsize)
 342                 goto err;
 343         free(buf);
 344         return 0;
 345 err:
 346         free(buf);
 347         return 1;
 348 }
 349
 350 static void detail_super0(struct supertype *st, char *homehost)
 351 {
 352         mdp_super_t *sb = st->sb;
 353         printf("           UUID : ");
 354         if (sb->minor_version >= 90)
 355                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 356                        sb->set_uuid2, sb->set_uuid3);
 357         else
 358                 printf("%08x", sb->set_uuid0);
 359         if (homehost) {
 360                 char buf[20];
 361                 void *hash = sha1_buffer(homehost,
 362                                          strlen(homehost),
 363                                          buf);
 364                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 365                         printf(" (local to host %s)", homehost);
 366         }
 367         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 368 }
 369
 370 static void brief_detail_super0(struct supertype *st)
 371 {
 372         mdp_super_t *sb = st->sb;
 373         printf(" UUID=");
 374         if (sb->minor_version >= 90)
 375                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 376                        sb->set_uuid2, sb->set_uuid3);
 377         else
 378                 printf("%08x", sb->set_uuid0);
 379 }
 380 #endif
 381
 382 static int match_home0(struct supertype *st, char *homehost)
 383 {
 384         mdp_super_t *sb = st->sb;
 385         char buf[20];
 386         char *hash;
 387
 388         if (!homehost)
 389                 return 0;
 390         hash = sha1_buffer(homehost,
 391                            strlen(homehost),
 392                            buf);
 393
 394         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 395 }
 396
 397 static void uuid_from_super0(struct supertype *st, int uuid[4])
 398 {
 399         mdp_super_t *super = st->sb;
 400         uuid[0] = super->set_uuid0;
 401         if (super->minor_version >= 90) {
 402                 uuid[1] = super->set_uuid1;
 403                 uuid[2] = super->set_uuid2;
 404                 uuid[3] = super->set_uuid3;
 405         } else {
 406                 uuid[1] = 0;
 407                 uuid[2] = 0;
 408                 uuid[3] = 0;
 409         }
 410 }
 411
 412 static void getinfo_super0(struct supertype *st, struct mdinfo *info, char *map)
 413 {
 414         mdp_super_t *sb = st->sb;
 415         int working = 0;
 416         int i;
 417         int map_disks = info->array.raid_disks;
 418
 419         memset(info, 0, sizeof(*info));
 420         info->array.major_version = sb->major_version;
 421         info->array.minor_version = sb->minor_version;
 422         info->array.patch_version = sb->patch_version;
 423         info->array.raid_disks = sb->raid_disks;
 424         info->array.level = sb->level;
 425         info->array.layout = sb->layout;
 426         info->array.md_minor = sb->md_minor;
 427         info->array.ctime = sb->ctime;
 428         info->array.utime = sb->utime;
 429         info->array.chunk_size = sb->chunk_size;
 430         info->array.state = sb->state;
 431         info->component_size = sb->size;
 432         info->component_size *= 2;
 433
 434         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 435                 info->bitmap_offset = 8;
 436
 437         info->disk.state = sb->this_disk.state;
 438         info->disk.major = sb->this_disk.major;
 439         info->disk.minor = sb->this_disk.minor;
 440         info->disk.raid_disk = sb->this_disk.raid_disk;
 441         info->disk.number = sb->this_disk.number;
 442
 443         info->events = md_event(sb);
 444         info->data_offset = 0;
 445
 446         sprintf(info->text_version, "0.%d", sb->minor_version);
 447         info->safe_mode_delay = 200;
 448
 449         uuid_from_super0(st, info->uuid);
 450
 451         info->recovery_start = MaxSector;
 452         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 453                 info->reshape_active = 1;
 454                 info->reshape_progress = sb->reshape_position;
 455                 info->new_level = sb->new_level;
 456                 info->delta_disks = sb->delta_disks;
 457                 info->new_layout = sb->new_layout;
 458                 info->new_chunk = sb->new_chunk;
 459                 if (info->delta_disks < 0)
 460                         info->array.raid_disks -= info->delta_disks;
 461         } else
 462                 info->reshape_active = 0;
 463
 464         info->recovery_blocked = info->reshape_active;
 465
 466         sprintf(info->name, "%d", sb->md_minor);
 467         /* work_disks is calculated rather than read directly */
 468         for (i=0; i < MD_SB_DISKS; i++)
 469                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 470                     (sb->disks[i].raid_disk < (unsigned)info->array.raid_disks) &&
 471                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 472                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY))) {
 473                         working ++;
 474                         if (map && i < map_disks)
 475                                 map[i] = 1;
 476                 } else if (map && i < map_disks)
 477                         map[i] = 0;
 478         info->array.working_disks = working;
 479 }
 480
 481 static struct mdinfo *container_content0(struct supertype *st, char *subarray)
 482 {
 483         struct mdinfo *info;
 484
 485         if (subarray)
 486                 return NULL;
 487
 488         info = xmalloc(sizeof(*info));
 489         getinfo_super0(st, info, NULL);
 490         return info;
 491 }
 492
 493 static int update_super0(struct supertype *st, struct mdinfo *info,
 494                          char *update,
 495                          char *devname, int verbose,
 496                          int uuid_set, char *homehost)
 497 {
 498         /* NOTE: for 'assemble' and 'force' we need to return non-zero
 499          * if any change was made.  For others, the return value is
 500          * ignored.
 501          */
 502         int rv = 0;
 503         int uuid[4];
 504         mdp_super_t *sb = st->sb;
 505
 506         if (strcmp(update, "homehost") == 0 &&
 507             homehost) {
 508                 /* note that 'homehost' is special as it is really
 509                  * a "uuid" update.
 510                  */
 511                 uuid_set = 0;
 512                 update = "uuid";
 513                 info->uuid[0] = sb->set_uuid0;
 514                 info->uuid[1] = sb->set_uuid1;
 515         }
 516
 517         if (strcmp(update, "sparc2.2")==0 ) {
 518                 /* 2.2 sparc put the events in the wrong place
 519                  * So we copy the tail of the superblock
 520                  * up 4 bytes before continuing
 521                  */
 522                 __u32 *sb32 = (__u32*)sb;
 523                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 524                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 525                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 526                 if (verbose >= 0)
 527                         pr_err("adjusting superblock of %s for 2.2/sparc compatibility.\n",
 528                                devname);
 529         } else if (strcmp(update, "super-minor") ==0) {
 530                 sb->md_minor = info->array.md_minor;
 531                 if (verbose > 0)
 532                         pr_err("updating superblock of %s with minor number %d\n",
 533                                 devname, info->array.md_minor);
 534         } else if (strcmp(update, "summaries") == 0) {
 535                 unsigned int i;
 536                 /* set nr_disks, active_disks, working_disks,
 537                  * failed_disks, spare_disks based on disks[]
 538                  * array in superblock.
 539                  * Also make sure extra slots aren't 'failed'
 540                  */
 541                 sb->nr_disks = sb->active_disks =
 542                         sb->working_disks = sb->failed_disks =
 543                         sb->spare_disks = 0;
 544                 for (i=0; i < MD_SB_DISKS ; i++)
 545                         if (sb->disks[i].major ||
 546                             sb->disks[i].minor) {
 547                                 int state = sb->disks[i].state;
 548                                 if (state & (1<<MD_DISK_REMOVED))
 549                                         continue;
 550                                 sb->nr_disks++;
 551                                 if (state & (1<<MD_DISK_ACTIVE))
 552                                         sb->active_disks++;
 553                                 if (state & (1<<MD_DISK_FAULTY))
 554                                         sb->failed_disks++;
 555                                 else
 556                                         sb->working_disks++;
 557                                 if (state == 0)
 558                                         sb->spare_disks++;
 559                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 560                                 sb->disks[i].state = 0;
 561         } else if (strcmp(update, "force-one")==0) {
 562                 /* Not enough devices for a working array, so
 563                  * bring this one up-to-date.
 564                  */
 565                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 566                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 567                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 568                 if (sb->events_hi != ehi ||
 569                     sb->events_lo != elo)
 570                         rv = 1;
 571         } else if (strcmp(update, "force-array")==0) {
 572                 /* degraded array and 'force' requested, so
 573                  * maybe need to mark it 'clean'
 574                  */
 575                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 576                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 577                         /* need to force clean */
 578                         sb->state |= (1 << MD_SB_CLEAN);
 579                         rv = 1;
 580                 }
 581         } else if (strcmp(update, "assemble")==0) {
 582                 int d = info->disk.number;
 583                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 584                 int mask = (1<<MD_DISK_WRITEMOSTLY);
 585                 int add = 0;
 586                 if (sb->minor_version >= 91)
 587                         /* During reshape we don't insist on everything
 588                          * being marked 'sync'
 589                          */
 590                         add = (1<<MD_DISK_SYNC);
 591                 if (((sb->disks[d].state & ~mask) | add)
 592                     != (unsigned)info->disk.state) {
 593                         sb->disks[d].state = info->disk.state | wonly;
 594                         rv = 1;
 595                 }
 596                 if (info->reshape_active &&
 597                     sb->minor_version > 90 && (sb->reshape_position+1) != 0 &&
 598                     info->delta_disks >= 0 &&
 599                     info->reshape_progress < sb->reshape_position) {
 600                         sb->reshape_position = info->reshape_progress;
 601                         rv = 1;
 602                 }
 603                 if (info->reshape_active &&
 604                     sb->minor_version > 90 && (sb->reshape_position+1) != 0 &&
 605                     info->delta_disks < 0 &&
 606                     info->reshape_progress > sb->reshape_position) {
 607                         sb->reshape_position = info->reshape_progress;
 608                         rv = 1;
 609                 }
 610         } else if (strcmp(update, "linear-grow-new") == 0) {
 611                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 612                 sb->disks[info->disk.number].number = info->disk.number;
 613                 sb->disks[info->disk.number].major = info->disk.major;
 614                 sb->disks[info->disk.number].minor = info->disk.minor;
 615                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 616                 sb->disks[info->disk.number].state = info->disk.state;
 617                 sb->this_disk = sb->disks[info->disk.number];
 618         } else if (strcmp(update, "linear-grow-update") == 0) {
 619                 sb->raid_disks = info->array.raid_disks;
 620                 sb->nr_disks = info->array.nr_disks;
 621                 sb->active_disks = info->array.active_disks;
 622                 sb->working_disks = info->array.working_disks;
 623                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 624                 sb->disks[info->disk.number].number = info->disk.number;
 625                 sb->disks[info->disk.number].major = info->disk.major;
 626                 sb->disks[info->disk.number].minor = info->disk.minor;
 627                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 628                 sb->disks[info->disk.number].state = info->disk.state;
 629         } else if (strcmp(update, "resync") == 0) {
 630                 /* make sure resync happens */
 631                 sb->state &= ~(1<<MD_SB_CLEAN);
 632                 sb->recovery_cp = 0;
 633         } else if (strcmp(update, "uuid") == 0) {
 634                 if (!uuid_set && homehost) {
 635                         char buf[20];
 636                         char *hash = sha1_buffer(homehost,
 637                                                  strlen(homehost),
 638                                                  buf);
 639                         memcpy(info->uuid+2, hash, 8);
 640                 }
 641                 sb->set_uuid0 = info->uuid[0];
 642                 sb->set_uuid1 = info->uuid[1];
 643                 sb->set_uuid2 = info->uuid[2];
 644                 sb->set_uuid3 = info->uuid[3];
 645                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 646                         struct bitmap_super_s *bm;
 647                         bm = (struct bitmap_super_s*)(sb+1);
 648                         uuid_from_super0(st, uuid);
 649                         memcpy(bm->uuid, uuid, 16);
 650                 }
 651         } else if (strcmp(update, "metadata") == 0) {
 652                 /* Create some v1.0 metadata to match ours but make the
 653                  * ctime bigger.  Also update info->array.*_version.
 654                  * We need to arrange that store_super writes out
 655                  * the v1.0 metadata.
 656                  * Not permitted for unclean array, or array with
 657                  * bitmap.
 658                  */
 659                 if (info->bitmap_offset) {
 660                         pr_err("Cannot update metadata when bitmap is present\n");
 661                         rv = -2;
 662                 } else if (info->array.state != 1) {
 663                         pr_err("Cannot update metadata on unclean array\n");
 664                         rv = -2;
 665                 } else {
 666                         info->array.major_version = 1;
 667                         info->array.minor_version = 0;
 668                         uuid_from_super0(st, info->uuid);
 669                         st->other = super1_make_v0(st, info, st->sb);
 670                 }
 671         } else if (strcmp(update, "revert-reshape") == 0) {
 672                 rv = -2;
 673                 if (sb->minor_version <= 90)
 674                         pr_err("No active reshape to revert on %s\n",
 675                                devname);
 676                 else if (sb->delta_disks == 0)
 677                         pr_err("%s: Can only revert reshape which changes number of devices\n",
 678                                devname);
 679                 else {
 680                         int tmp;
 681                         int parity = sb->level == 6 ? 2 : 1;
 682                         rv = 0;
 683
 684                         if (sb->level >= 4 && sb->level <= 6 &&
 685                             sb->reshape_position % (
 686                                     sb->new_chunk/512 *
 687                                     (sb->raid_disks - sb->delta_disks - parity))) {
 688                                 pr_err("Reshape position is not suitably aligned.\n");
 689                                 pr_err("Try normal assembly and stop again\n");
 690                                 return -2;
 691                         }
 692                         sb->raid_disks -= sb->delta_disks;
 693                         sb->delta_disks = -sb->delta_disks;
 694
 695                         tmp = sb->new_layout;
 696                         sb->new_layout = sb->layout;
 697                         sb->layout = tmp;
 698
 699                         tmp = sb->new_chunk;
 700                         sb->new_chunk = sb->chunk_size;
 701                         sb->chunk_size = tmp;
 702                 }
 703         } else if (strcmp(update, "no-bitmap") == 0) {
 704                 sb->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 705         } else if (strcmp(update, "_reshape_progress")==0)
 706                 sb->reshape_position = info->reshape_progress;
 707         else if (strcmp(update, "writemostly")==0)
 708                 sb->state |= (1<<MD_DISK_WRITEMOSTLY);
 709         else if (strcmp(update, "readwrite")==0)
 710                 sb->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 711         else
 712                 rv = -1;
 713
 714         sb->sb_csum = calc_sb0_csum(sb);
 715         return rv;
 716 }
 717
 718 /*
 719  * For version-0 superblock, the homehost is 'stored' in the uuid.
 720  * 8 bytes for a hash of the host leaving 8 bytes of random material.
 721  * We use the first 8 bytes (64bits) of the sha1 of the host name
 722  */
 723 static int init_super0(struct supertype *st, mdu_array_info_t *info,
 724                        unsigned long long size, char *ignored_name,
 725                        char *homehost, int *uuid,
 726                        unsigned long long data_offset)
 727 {
 728         mdp_super_t *sb;
 729         int spares;
 730
 731         if (data_offset != INVALID_SECTORS) {
 732                 pr_err("data-offset not support for 0.90\n");
 733                 return 0;
 734         }
 735
 736         if (posix_memalign((void**)&sb, 4096,
 737                            MD_SB_BYTES + ROUND_UP(sizeof(bitmap_super_t), 4096)) != 0) {
 738                 pr_err("could not allocate superblock\n");
 739                 return 0;
 740         }
 741         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 742
 743         st->sb = sb;
 744         if (info == NULL) {
 745                 /* zeroing the superblock */
 746                 return 0;
 747         }
 748
 749         spares = info->working_disks - info->active_disks;
 750         if (info->raid_disks + spares  > MD_SB_DISKS) {
 751                 pr_err("too many devices requested: %d+%d > %d\n",
 752                         info->raid_disks , spares, MD_SB_DISKS);
 753                 return 0;
 754         }
 755
 756         sb->md_magic = MD_SB_MAGIC;
 757         sb->major_version = 0;
 758         sb->minor_version = 90;
 759         sb->patch_version = 0;
 760         sb->gvalid_words = 0; /* ignored */
 761         sb->ctime = time(0);
 762         sb->level = info->level;
 763         sb->size = size;
 764         if (size != (unsigned long long)sb->size)
 765                 return 0;
 766         sb->nr_disks = info->nr_disks;
 767         sb->raid_disks = info->raid_disks;
 768         sb->md_minor = info->md_minor;
 769         sb->not_persistent = 0;
 770         if (uuid) {
 771                 sb->set_uuid0 = uuid[0];
 772                 sb->set_uuid1 = uuid[1];
 773                 sb->set_uuid2 = uuid[2];
 774                 sb->set_uuid3 = uuid[3];
 775         } else {
 776                 __u32 r[4];
 777                 random_uuid((__u8 *)r);
 778                 sb->set_uuid0 = r[0];
 779                 sb->set_uuid1 = r[1];
 780                 sb->set_uuid2 = r[2];
 781                 sb->set_uuid3 = r[3];
 782         }
 783         if (homehost && !uuid) {
 784                 char buf[20];
 785                 char *hash = sha1_buffer(homehost,
 786                                          strlen(homehost),
 787                                          buf);
 788                 memcpy(&sb->set_uuid2, hash, 8);
 789         }
 790
 791         sb->utime = sb->ctime;
 792         sb->state = info->state;
 793         sb->active_disks = info->active_disks;
 794         sb->working_disks = info->working_disks;
 795         sb->failed_disks = info->failed_disks;
 796         sb->spare_disks = info->spare_disks;
 797         sb->events_hi = 0;
 798         sb->events_lo = 1;
 799
 800         sb->layout = info->layout;
 801         sb->chunk_size = info->chunk_size;
 802
 803         return 1;
 804 }
 805
 806 struct devinfo {
 807         int fd;
 808         char *devname;
 809         mdu_disk_info_t disk;
 810         struct devinfo *next;
 811 };
 812
 813 #ifndef MDASSEMBLE
 814 /* Add a device to the superblock being created */
 815 static int add_to_super0(struct supertype *st, mdu_disk_info_t *dinfo,
 816                          int fd, char *devname, unsigned long long data_offset)
 817 {
 818         mdp_super_t *sb = st->sb;
 819         mdp_disk_t *dk = &sb->disks[dinfo->number];
 820         struct devinfo *di, **dip;
 821
 822         dk->number = dinfo->number;
 823         dk->major = dinfo->major;
 824         dk->minor = dinfo->minor;
 825         dk->raid_disk = dinfo->raid_disk;
 826         dk->state = dinfo->state & ((1<<MD_DISK_ACTIVE) |
 827                                     (1<<MD_DISK_SYNC));
 828
 829         sb->this_disk = sb->disks[dinfo->number];
 830         sb->sb_csum = calc_sb0_csum(sb);
 831
 832         dip = (struct devinfo **)&st->info;
 833         while (*dip)
 834                 dip = &(*dip)->next;
 835         di = xmalloc(sizeof(struct devinfo));
 836         di->fd = fd;
 837         di->devname = devname;
 838         di->disk = *dinfo;
 839         di->next = NULL;
 840         *dip = di;
 841
 842         return 0;
 843 }
 844 #endif
 845
 846 static int store_super0(struct supertype *st, int fd)
 847 {
 848         unsigned long long dsize;
 849         unsigned long long offset;
 850         mdp_super_t *super = st->sb;
 851
 852         if (!get_dev_size(fd, NULL, &dsize))
 853                 return 1;
 854
 855         if (dsize < MD_RESERVED_SECTORS*512)
 856                 return 2;
 857
 858         if (st->other) {
 859                 /* Writing out v1.0 metadata for --update=metadata */
 860                 int ret = 0;
 861
 862                 offset = dsize/512 - 8*2;
 863                 offset &= ~(4*2-1);
 864                 offset *= 512;
 865                 if (lseek64(fd, offset, 0)< 0LL)
 866                         ret = 3;
 867                 else if (write(fd, st->other, 1024) != 1024)
 868                         ret = 4;
 869                 else
 870                         fsync(fd);
 871                 free(st->other);
 872                 st->other = NULL;
 873                 return ret;
 874         }
 875
 876         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 877
 878         offset *= 512;
 879
 880         if (lseek64(fd, offset, 0)< 0LL)
 881                 return 3;
 882
 883         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 884                 return 4;
 885
 886         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 887                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 888                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 889                         if (write(fd, bm, ROUND_UP(sizeof(*bm),4096)) !=
 890                             ROUND_UP(sizeof(*bm),4096))
 891                             return 5;
 892         }
 893
 894         fsync(fd);
 895         return 0;
 896 }
 897
 898 #ifndef MDASSEMBLE
 899 static int write_init_super0(struct supertype *st)
 900 {
 901         mdp_super_t *sb = st->sb;
 902         int rv = 0;
 903         struct devinfo *di;
 904
 905         for (di = st->info ; di && ! rv ; di = di->next) {
 906
 907                 if (di->disk.state & (1 << MD_DISK_FAULTY))
 908                         continue;
 909                 if (di->fd == -1)
 910                         continue;
 911                 while (Kill(di->devname, NULL, 0, -1, 1) == 0)
 912                         ;
 913
 914                 sb->disks[di->disk.number].state &= ~(1<<MD_DISK_FAULTY);
 915
 916                 sb->this_disk = sb->disks[di->disk.number];
 917                 sb->sb_csum = calc_sb0_csum(sb);
 918                 rv = store_super0(st, di->fd);
 919
 920                 if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 921                         rv = st->ss->write_bitmap(st, di->fd, NoUpdate);
 922
 923                 if (rv)
 924                         pr_err("failed to write superblock to %s\n",
 925                                di->devname);
 926         }
 927         return rv;
 928 }
 929 #endif
 930
 931 static int compare_super0(struct supertype *st, struct supertype *tst)
 932 {
 933         /*
 934          * return:
 935          *  0 same, or first was empty, and second was copied
 936          *  1 second had wrong number
 937          *  2 wrong uuid
 938          *  3 wrong other info
 939          */
 940         mdp_super_t *first = st->sb;
 941         mdp_super_t *second = tst->sb;
 942         int uuid1[4], uuid2[4];
 943
 944         if (second->md_magic != MD_SB_MAGIC)
 945                 return 1;
 946         if (!first) {
 947                 if (posix_memalign((void**)&first, 4096,
 948                              MD_SB_BYTES +
 949                              ROUND_UP(sizeof(struct bitmap_super_s), 4096)) != 0) {
 950                         pr_err("could not allocate superblock\n");
 951                         return 1;
 952                 }
 953                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 954                 st->sb = first;
 955                 return 0;
 956         }
 957
 958         uuid_from_super0(st, uuid1);
 959         uuid_from_super0(tst, uuid2);
 960         if (!same_uuid(uuid1, uuid2, 0))
 961                 return 2;
 962         if (first->major_version != second->major_version ||
 963             first->minor_version != second->minor_version ||
 964             first->patch_version != second->patch_version ||
 965             first->gvalid_words  != second->gvalid_words  ||
 966             first->ctime         != second->ctime         ||
 967             first->level         != second->level         ||
 968             first->size          != second->size          ||
 969             first->raid_disks    != second->raid_disks    )
 970                 return 3;
 971
 972         return 0;
 973 }
 974
 975 static void free_super0(struct supertype *st);
 976
 977 static int load_super0(struct supertype *st, int fd, char *devname)
 978 {
 979         /* try to read in the superblock
 980          * Return:
 981          *  0 on success
 982          *  1 on cannot get superblock
 983          *  2 on superblock meaningless
 984          */
 985         unsigned long long dsize;
 986         unsigned long long offset;
 987         mdp_super_t *super;
 988         int uuid[4];
 989         struct bitmap_super_s *bsb;
 990
 991         free_super0(st);
 992
 993         if (!get_dev_size(fd, devname, &dsize))
 994                 return 1;
 995
 996         if (dsize < MD_RESERVED_SECTORS*512) {
 997                 if (devname)
 998                         pr_err("%s is too small for md: size is %llu sectors.\n",
 999                                devname, dsize);
1000                 return 1;
1001         }
1002         st->devsize = dsize;
1003
1004         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
1005
1006         offset *= 512;
1007
1008         if (lseek64(fd, offset, 0)< 0LL) {
1009                 if (devname)
1010                         pr_err("Cannot seek to superblock on %s: %s\n",
1011                                 devname, strerror(errno));
1012                 return 1;
1013         }
1014
1015         if (posix_memalign((void**)&super, 4096,
1016                            MD_SB_BYTES +
1017                            ROUND_UP(sizeof(bitmap_super_t), 4096)) != 0) {
1018                 pr_err("could not allocate superblock\n");
1019                 return 1;
1020         }
1021
1022         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
1023                 if (devname)
1024                         pr_err("Cannot read superblock on %s\n",
1025                                 devname);
1026                 free(super);
1027                 return 1;
1028         }
1029
1030         if (st->ss && st->minor_version == 9)
1031                 super0_swap_endian(super);
1032
1033         if (super->md_magic != MD_SB_MAGIC) {
1034                 if (devname)
1035                         pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
1036                                 devname, MD_SB_MAGIC, super->md_magic);
1037                 free(super);
1038                 return 2;
1039         }
1040
1041         if (super->major_version != 0) {
1042                 if (devname)
1043                         pr_err("Cannot interpret superblock on %s - version is %d\n",
1044                                 devname, super->major_version);
1045                 free(super);
1046                 return 2;
1047         }
1048         st->sb = super;
1049
1050         if (st->ss == NULL) {
1051                 st->ss = &super0;
1052                 st->minor_version = super->minor_version;
1053                 st->max_devs = MD_SB_DISKS;
1054                 st->info = NULL;
1055         }
1056
1057         /* Now check on the bitmap superblock */
1058         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
1059                 return 0;
1060         /* Read the bitmap superblock and make sure it looks
1061          * valid.  If it doesn't clear the bit.  An --assemble --force
1062          * should get that written out.
1063          */
1064         if (read(fd, super+1, ROUND_UP(sizeof(struct bitmap_super_s),4096))
1065             != ROUND_UP(sizeof(struct bitmap_super_s),4096))
1066                 goto no_bitmap;
1067
1068         uuid_from_super0(st, uuid);
1069         bsb = (struct bitmap_super_s *)(super+1);
1070         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
1071             memcmp(bsb->uuid, uuid, 16) != 0)
1072                 goto no_bitmap;
1073         return 0;
1074
1075  no_bitmap:
1076         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
1077
1078         return 0;
1079 }
1080
1081 static struct supertype *match_metadata_desc0(char *arg)
1082 {
1083         struct supertype *st = xcalloc(1, sizeof(*st));
1084
1085         st->container_devnm[0] = 0;
1086         st->ss = &super0;
1087         st->info = NULL;
1088         st->minor_version = 90;
1089         st->max_devs = MD_SB_DISKS;
1090         st->sb = NULL;
1091         /* we sometimes get 00.90 */
1092         while (arg[0] == '0' && arg[1] == '0')
1093                 arg++;
1094         if (strcmp(arg, "0") == 0 ||
1095 #ifdef DEFAULT_OLD_METADATA /* ifndef in super1.c */
1096             strcmp(arg, "default") == 0 ||
1097 #endif /* DEFAULT_OLD_METADATA */
1098             strcmp(arg, "0.90") == 0 ||
1099             strcmp(arg, "") == 0 /* no metadata  - i.e. non_persistent */
1100                 )
1101                 return st;
1102
1103         st->minor_version = 91; /* reshape in progress */
1104         if (strcmp(arg, "0.91") == 0) /* For dup_super support */
1105                 return st;
1106
1107         st->minor_version = 9; /* flag for 'byte-swapped' */
1108         if (strcmp(arg, "0.swap")==0 ||
1109             strcmp(arg, "0.9") == 0) /* For dup_super support */
1110                 return st;
1111
1112         free(st);
1113         return NULL;
1114 }
1115
1116 static __u64 avail_size0(struct supertype *st, __u64 devsize,
1117                          unsigned long long data_offset)
1118 {
1119         if (data_offset != 0 && data_offset != INVALID_SECTORS)
1120                 return 0ULL;
1121         if (devsize < MD_RESERVED_SECTORS)
1122                 return 0ULL;
1123         return MD_NEW_SIZE_SECTORS(devsize);
1124 }
1125
1126 static int add_internal_bitmap0(struct supertype *st, int *chunkp,
1127                                 int delay, int write_behind,
1128                                 unsigned long long size, int may_change,
1129                                 int major)
1130 {
1131         /*
1132          * The bitmap comes immediately after the superblock and must be 60K in size
1133          * at most.  The default size is between 30K and 60K
1134          *
1135          * size is in sectors,  chunk is in bytes !!!
1136          */
1137         unsigned long long bits;
1138         unsigned long long max_bits = (60*1024 - sizeof(bitmap_super_t))*8;
1139         unsigned long long min_chunk;
1140         int chunk = *chunkp;
1141         mdp_super_t *sb = st->sb;
1142         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
1143         int uuid[4];
1144
1145         min_chunk = 4096; /* sub-page chunks don't work yet.. */
1146         bits = (size * 512) / min_chunk + 1;
1147         while (bits > max_bits) {
1148                 min_chunk *= 2;
1149                 bits = (bits+1)/2;
1150         }
1151         if (chunk == UnSet) {
1152                 /* A chunk size less than a few Megabytes gives poor
1153                  * performance without increasing resync noticeably
1154                  */
1155                 chunk = min_chunk;
1156                 if (chunk < 64*1024*1024)
1157                         chunk = 64*1024*1024;
1158         } else if ((unsigned long long)chunk < min_chunk)
1159                 return -EINVAL; /* chunk size too small */
1160
1161         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1162
1163         memset(bms, 0, sizeof(*bms));
1164         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
1165         bms->version = __cpu_to_le32(major);
1166         uuid_from_super0(st, uuid);
1167         memcpy(bms->uuid, uuid, 16);
1168         bms->chunksize = __cpu_to_le32(chunk);
1169         bms->daemon_sleep = __cpu_to_le32(delay);
1170         bms->sync_size = __cpu_to_le64(size);
1171         bms->write_behind = __cpu_to_le32(write_behind);
1172         *chunkp = chunk;
1173         return 0;
1174 }
1175
1176 static int locate_bitmap0(struct supertype *st, int fd, int node_num)
1177 {
1178         unsigned long long dsize;
1179         unsigned long long offset;
1180
1181         if (!get_dev_size(fd, NULL, &dsize))
1182                 return -1;
1183
1184         if (dsize < MD_RESERVED_SECTORS*512)
1185                 return -1;
1186
1187         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
1188
1189         offset *= 512;
1190
1191         offset += MD_SB_BYTES;
1192
1193         lseek64(fd, offset, 0);
1194         return 0;
1195 }
1196
1197 static int write_bitmap0(struct supertype *st, int fd, enum bitmap_update update)
1198 {
1199         unsigned long long dsize;
1200         unsigned long long offset;
1201         mdp_super_t *sb = st->sb;
1202
1203         int rv = 0;
1204
1205         int towrite, n;
1206         void *buf;
1207
1208         if (!get_dev_size(fd, NULL, &dsize))
1209                 return 1;
1210
1211         if (dsize < MD_RESERVED_SECTORS*512)
1212                 return -1;
1213
1214         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
1215
1216         offset *= 512;
1217
1218         if (lseek64(fd, offset + 4096, 0)< 0LL)
1219                 return 3;
1220
1221         if (posix_memalign(&buf, 4096, 4096))
1222                 return -ENOMEM;
1223
1224         memset(buf, 0xff, 4096);
1225         memcpy(buf,  ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t));
1226         towrite = 60*1024;
1227         while (towrite > 0) {
1228                 n = towrite;
1229                 if (n > 4096)
1230                         n = 4096;
1231                 n = write(fd, buf, n);
1232                 if (n > 0)
1233                         towrite -= n;
1234                 else
1235                         break;
1236                 memset(buf, 0xff, 4096);
1237         }
1238         fsync(fd);
1239         if (towrite)
1240                 rv = -2;
1241
1242         free(buf);
1243         return rv;
1244 }
1245
1246 static void free_super0(struct supertype *st)
1247 {
1248         if (st->sb)
1249                 free(st->sb);
1250         while (st->info) {
1251                 struct devinfo *di = st->info;
1252                 st->info = di->next;
1253                 if (di->fd >= 0)
1254                         close(di->fd);
1255                 free(di);
1256         }
1257         st->sb = NULL;
1258 }
1259
1260 #ifndef MDASSEMBLE
1261 static int validate_geometry0(struct supertype *st, int level,
1262                               int layout, int raiddisks,
1263                               int *chunk, unsigned long long size,
1264                               unsigned long long data_offset,
1265                               char *subdev, unsigned long long *freesize,
1266                               int verbose)
1267 {
1268         unsigned long long ldsize;
1269         int fd;
1270         unsigned int tbmax = 4;
1271
1272         /* prior to linux 3.1, a but limits usable device size to 2TB.
1273          * It was introduced in 2.6.29, but we won't worry about that detail
1274          */
1275         if (get_linux_version() < 3001000)
1276                 tbmax = 2;
1277
1278         if (level == LEVEL_CONTAINER) {
1279                 if (verbose)
1280                         pr_err("0.90 metadata does not support containers\n");
1281                 return 0;
1282         }
1283         if (raiddisks > MD_SB_DISKS) {
1284                 if (verbose)
1285                         pr_err("0.90 metadata supports at most %d devices per array\n",
1286                                 MD_SB_DISKS);
1287                 return 0;
1288         }
1289         if (size >= tbmax * 2ULL*1024*1024*1024) {
1290                 if (verbose)
1291                         pr_err("0.90 metadata supports at most %d terabytes per device\n", tbmax);
1292                 return 0;
1293         }
1294         if (*chunk == UnSet)
1295                 *chunk = DEFAULT_CHUNK;
1296
1297         if (!subdev)
1298                 return 1;
1299
1300         fd = open(subdev, O_RDONLY|O_EXCL, 0);
1301         if (fd < 0) {
1302                 if (verbose)
1303                         pr_err("super0.90 cannot open %s: %s\n",
1304                                 subdev, strerror(errno));
1305                 return 0;
1306         }
1307
1308         if (!get_dev_size(fd, subdev, &ldsize)) {
1309                 close(fd);
1310                 return 0;
1311         }
1312         close(fd);
1313
1314         if (ldsize < MD_RESERVED_SECTORS * 512)
1315                 return 0;
1316         *freesize = MD_NEW_SIZE_SECTORS(ldsize >> 9);
1317         return 1;
1318 }
1319 #endif /* MDASSEMBLE */
1320
1321 struct superswitch super0 = {
1322 #ifndef MDASSEMBLE
1323         .examine_super = examine_super0,
1324         .brief_examine_super = brief_examine_super0,
1325         .export_examine_super = export_examine_super0,
1326         .detail_super = detail_super0,
1327         .brief_detail_super = brief_detail_super0,
1328         .write_init_super = write_init_super0,
1329         .validate_geometry = validate_geometry0,
1330         .add_to_super = add_to_super0,
1331         .copy_metadata = copy_metadata0,
1332 #endif
1333         .match_home = match_home0,
1334         .uuid_from_super = uuid_from_super0,
1335         .getinfo_super = getinfo_super0,
1336         .container_content = container_content0,
1337         .update_super = update_super0,
1338         .init_super = init_super0,
1339         .store_super = store_super0,
1340         .compare_super = compare_super0,
1341         .load_super = load_super0,
1342         .match_metadata_desc = match_metadata_desc0,
1343         .avail_size = avail_size0,
1344         .add_internal_bitmap = add_internal_bitmap0,
1345         .locate_bitmap = locate_bitmap0,
1346         .write_bitmap = write_bitmap0,
1347         .free_super = free_super0,
1348         .name = "0.90",
1349 };