super0.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
  23  *    Paper: Neil Brown
  24  *           School of Computer Science and Engineering
  25  *           The University of New South Wales
  26  *           Sydney, 2052
  27  *           Australia
  28  */
  29
  30 #define HAVE_STDINT_H 1
  31 #include "mdadm.h"
  32 #include "sha1.h"
  33 /*
  34  * All handling for the 0.90.0 version superblock is in
  35  * this file.
  36  * This includes:
  37  *   - finding, loading, and writing the superblock.
  38  *   - initialising a new superblock
  39  *   - printing the superblock for --examine
  40  *   - printing part of the superblock for --detail
  41  * .. other stuff
  42  */
  43
  44
  45 static unsigned long calc_sb0_csum(mdp_super_t *super)
  46 {
  47         unsigned long csum = super->sb_csum;
  48         unsigned long newcsum;
  49         super->sb_csum= 0 ;
  50         newcsum = calc_csum(super, MD_SB_BYTES);
  51         super->sb_csum = csum;
  52         return newcsum;
  53 }
  54
  55
  56 void super0_swap_endian(struct mdp_superblock_s *sb)
  57 {
  58         /* as super0 superblocks are host-endian, it is sometimes
  59          * useful to be able to swap the endianness
  60          * as (almost) everything is u32's we byte-swap every 4byte
  61          * number.
  62          * We then also have to swap the events_hi and events_lo
  63          */
  64         char *sbc = (char *)sb;
  65         __u32 t32;
  66         int i;
  67
  68         for (i=0; i < MD_SB_BYTES ; i+=4) {
  69                 char t = sbc[i];
  70                 sbc[i] = sbc[i+3];
  71                 sbc[i+3] = t;
  72                 t=sbc[i+1];
  73                 sbc[i+1]=sbc[i+2];
  74                 sbc[i+2]=t;
  75         }
  76         t32 = sb->events_hi;
  77         sb->events_hi = sb->events_lo;
  78         sb->events_lo = t32;
  79
  80         t32 = sb->cp_events_hi;
  81         sb->cp_events_hi = sb->cp_events_lo;
  82         sb->cp_events_lo = t32;
  83
  84 }
  85
  86 #ifndef MDASSEMBLE
  87
  88 static void examine_super0(void *sbv, char *homehost)
  89 {
  90         mdp_super_t *sb = sbv;
  91         time_t atime;
  92         int d;
  93         char *c;
  94
  95         printf("          Magic : %08x\n", sb->md_magic);
  96         printf("        Version : %02d.%02d.%02d\n", sb->major_version, sb->minor_version,
  97                sb->patch_version);
  98         if (sb->minor_version >= 90) {
  99                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 100                        sb->set_uuid2, sb->set_uuid3);
 101                 if (homehost) {
 102                         char buf[20];
 103                         void *hash = sha1_buffer(homehost,
 104                                                  strlen(homehost),
 105                                                  buf);
 106                         if (memcmp(&sb->set_uuid2, hash, 8)==0)
 107                                 printf(" (local to host %s)", homehost);
 108                 }
 109                 printf("\n");
 110         } else
 111                 printf("           UUID : %08x\n", sb->set_uuid0);
 112
 113         atime = sb->ctime;
 114         printf("  Creation Time : %.24s\n", ctime(&atime));
 115         c=map_num(pers, sb->level);
 116         printf("     Raid Level : %s\n", c?c:"-unknown-");
 117         if ((int)sb->level >= 0) {
 118                 int ddsks=0;
 119                 printf("    Device Size : %d%s\n", sb->size, human_size((long long)sb->size<<10));
 120                 switch(sb->level) {
 121                 case 1: ddsks=1;break;
 122                 case 4:
 123                 case 5: ddsks = sb->raid_disks-1; break;
 124                 case 6: ddsks = sb->raid_disks-2; break;
 125                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 126                 }
 127                 if (ddsks)
 128                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 129                                human_size(ddsks*(long long)sb->size<<10));
 130         }
 131         printf("   Raid Devices : %d\n", sb->raid_disks);
 132         printf("  Total Devices : %d\n", sb->nr_disks);
 133         printf("Preferred Minor : %d\n", sb->md_minor);
 134         printf("\n");
 135         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 136                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 137                 if (sb->delta_disks) {
 138                         printf("  Delta Devices : %d", sb->delta_disks);
 139                         if (sb->delta_disks)
 140                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 141                         else
 142                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 143                 }
 144                 if (sb->new_level != sb->level) {
 145                         c = map_num(pers, sb->new_level);
 146                         printf("      New Level : %s\n", c?c:"-unknown-");
 147                 }
 148                 if (sb->new_layout != sb->layout) {
 149                         if (sb->level == 5) {
 150                                 c = map_num(r5layout, sb->new_layout);
 151                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 152                         }
 153                         if (sb->level == 10) {
 154                                 printf("     New Layout : near=%d, %s=%d\n",
 155                                        sb->new_layout&255,
 156                                        (sb->new_layout&0x10000)?"offset":"far",
 157                                        (sb->new_layout>>8)&255);
 158                         }
 159                 }
 160                 if (sb->new_chunk != sb->chunk_size)
 161                         printf("  New Chunksize : %d\n", sb->new_chunk);
 162                 printf("\n");
 163         }
 164         atime = sb->utime;
 165         printf("    Update Time : %.24s\n", ctime(&atime));
 166         printf("          State : %s\n",
 167                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 168         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 169                 printf("Internal Bitmap : present\n");
 170         printf(" Active Devices : %d\n", sb->active_disks);
 171         printf("Working Devices : %d\n", sb->working_disks);
 172         printf(" Failed Devices : %d\n", sb->failed_disks);
 173         printf("  Spare Devices : %d\n", sb->spare_disks);
 174         if (calc_sb0_csum(sb) == sb->sb_csum)
 175                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 176         else
 177                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 178         printf("         Events : %d.%d\n", sb->events_hi, sb->events_lo);
 179         printf("\n");
 180         if (sb->level == 5) {
 181                 c = map_num(r5layout, sb->layout);
 182                 printf("         Layout : %s\n", c?c:"-unknown-");
 183         }
 184         if (sb->level == 10) {
 185                 printf("         Layout : near=%d, %s=%d\n",
 186                        sb->layout&255,
 187                        (sb->layout&0x10000)?"offset":"far",
 188                        (sb->layout>>8)&255);
 189         }
 190         switch(sb->level) {
 191         case 0:
 192         case 4:
 193         case 5:
 194                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 195                 break;
 196         case -1:
 197                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 198                 break;
 199         default: break;
 200         }
 201         printf("\n");
 202         printf("      Number   Major   Minor   RaidDevice State\n");
 203         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
 204                 mdp_disk_t *dp;
 205                 char *dv;
 206                 char nb[5];
 207                 int wonly;
 208                 if (d>=0) dp = &sb->disks[d];
 209                 else dp = &sb->this_disk;
 210                 snprintf(nb, sizeof(nb), "%4d", d);
 211                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 212                        dp->number, dp->major, dp->minor, dp->raid_disk);
 213                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 214                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 215                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 216                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 217                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 218                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 219                 if (wonly) printf(" write-mostly");
 220                 if (dp->state == 0) printf(" spare");
 221                 if ((dv=map_dev(dp->major, dp->minor, 0)))
 222                         printf("   %s", dv);
 223                 printf("\n");
 224                 if (d == -1) printf("\n");
 225         }
 226 }
 227
 228 static void brief_examine_super0(void *sbv)
 229 {
 230         mdp_super_t *sb = sbv;
 231         char *c=map_num(pers, sb->level);
 232         char devname[20];
 233
 234         sprintf(devname, "/dev/md%d", sb->md_minor);
 235
 236         printf("ARRAY %s level=%s num-devices=%d UUID=",
 237                devname,
 238                c?c:"-unknown-", sb->raid_disks);
 239         if (sb->minor_version >= 90)
 240                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 241                        sb->set_uuid2, sb->set_uuid3);
 242         else
 243                 printf("%08x", sb->set_uuid0);
 244         printf("\n");
 245 }
 246
 247 static void detail_super0(void *sbv, char *homehost)
 248 {
 249         mdp_super_t *sb = sbv;
 250         printf("           UUID : ");
 251         if (sb->minor_version >= 90)
 252                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 253                        sb->set_uuid2, sb->set_uuid3);
 254         else
 255                 printf("%08x", sb->set_uuid0);
 256         if (homehost) {
 257                 char buf[20];
 258                 void *hash = sha1_buffer(homehost,
 259                                          strlen(homehost),
 260                                          buf);
 261                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 262                         printf(" (local to host %s)", homehost);
 263         }
 264         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 265 }
 266
 267 static void brief_detail_super0(void *sbv)
 268 {
 269         mdp_super_t *sb = sbv;
 270         printf(" UUID=");
 271         if (sb->minor_version >= 90)
 272                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 273                        sb->set_uuid2, sb->set_uuid3);
 274         else
 275                 printf("%08x", sb->set_uuid0);
 276 }
 277 #endif
 278
 279 static int match_home0(void *sbv, char *homehost)
 280 {
 281         mdp_super_t *sb = sbv;
 282         char buf[20];
 283         char *hash = sha1_buffer(homehost,
 284                                  strlen(homehost),
 285                                  buf);
 286
 287         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 288 }
 289
 290 static void uuid_from_super0(int uuid[4], void * sbv)
 291 {
 292         mdp_super_t *super = sbv;
 293         uuid[0] = super->set_uuid0;
 294         if (super->minor_version >= 90) {
 295                 uuid[1] = super->set_uuid1;
 296                 uuid[2] = super->set_uuid2;
 297                 uuid[3] = super->set_uuid3;
 298         } else {
 299                 uuid[1] = 0;
 300                 uuid[2] = 0;
 301                 uuid[3] = 0;
 302         }
 303 }
 304
 305 static void getinfo_super0(struct mdinfo *info, void *sbv)
 306 {
 307         mdp_super_t *sb = sbv;
 308         int working = 0;
 309         int i;
 310
 311         info->array.major_version = sb->major_version;
 312         info->array.minor_version = sb->minor_version;
 313         info->array.patch_version = sb->patch_version;
 314         info->array.raid_disks = sb->raid_disks;
 315         info->array.level = sb->level;
 316         info->array.layout = sb->layout;
 317         info->array.md_minor = sb->md_minor;
 318         info->array.ctime = sb->ctime;
 319         info->array.utime = sb->utime;
 320         info->array.chunk_size = sb->chunk_size;
 321         info->component_size = sb->size*2;
 322
 323         info->disk.state = sb->this_disk.state;
 324         info->disk.major = sb->this_disk.major;
 325         info->disk.minor = sb->this_disk.minor;
 326         info->disk.raid_disk = sb->this_disk.raid_disk;
 327         info->disk.number = sb->this_disk.number;
 328
 329         info->events = md_event(sb);
 330         info->data_offset = 0;
 331
 332         uuid_from_super0(info->uuid, sbv);
 333
 334         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 335                 info->reshape_active = 1;
 336                 info->reshape_progress = sb->reshape_position;
 337                 info->new_level = sb->new_level;
 338                 info->delta_disks = sb->delta_disks;
 339                 info->new_layout = sb->new_layout;
 340                 info->new_chunk = sb->new_chunk;
 341         } else
 342                 info->reshape_active = 0;
 343
 344         sprintf(info->name, "%d", sb->md_minor);
 345         /* work_disks is calculated rather than read directly */
 346         for (i=0; i < MD_SB_DISKS; i++)
 347                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 348                     (sb->disks[i].raid_disk < info->array.raid_disks) &&
 349                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 350                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 351                         working ++;
 352         info->array.working_disks = working;
 353 }
 354
 355
 356 static int update_super0(struct mdinfo *info, void *sbv, char *update,
 357                          char *devname, int verbose,
 358                          int uuid_set, char *homehost)
 359 {
 360         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 361          * For others, the return value is ignored.
 362          */
 363         int rv = 0;
 364         mdp_super_t *sb = sbv;
 365         if (strcmp(update, "sparc2.2")==0 ) {
 366                 /* 2.2 sparc put the events in the wrong place
 367                  * So we copy the tail of the superblock
 368                  * up 4 bytes before continuing
 369                  */
 370                 __u32 *sb32 = (__u32*)sb;
 371                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 372                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 373                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 374                 if (verbose >= 0)
 375                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 376                                  devname);
 377         }
 378         if (strcmp(update, "super-minor") ==0) {
 379                 sb->md_minor = info->array.md_minor;
 380                 if (verbose > 0)
 381                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 382                                 devname, info->array.md_minor);
 383         }
 384         if (strcmp(update, "summaries") == 0) {
 385                 int i;
 386                 /* set nr_disks, active_disks, working_disks,
 387                  * failed_disks, spare_disks based on disks[]
 388                  * array in superblock.
 389                  * Also make sure extra slots aren't 'failed'
 390                  */
 391                 sb->nr_disks = sb->active_disks =
 392                         sb->working_disks = sb->failed_disks =
 393                         sb->spare_disks = 0;
 394                 for (i=0; i < MD_SB_DISKS ; i++)
 395                         if (sb->disks[i].major ||
 396                             sb->disks[i].minor) {
 397                                 int state = sb->disks[i].state;
 398                                 if (state & (1<<MD_DISK_REMOVED))
 399                                         continue;
 400                                 sb->nr_disks++;
 401                                 if (state & (1<<MD_DISK_ACTIVE))
 402                                         sb->active_disks++;
 403                                 if (state & (1<<MD_DISK_FAULTY))
 404                                         sb->failed_disks++;
 405                                 else
 406                                         sb->working_disks++;
 407                                 if (state == 0)
 408                                         sb->spare_disks++;
 409                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 410                                 sb->disks[i].state = 0;
 411         }
 412         if (strcmp(update, "force")==0) {
 413                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 414                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 415                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 416                 if (sb->events_hi != ehi ||
 417                     sb->events_lo != elo)
 418                         rv = 1;
 419                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 420                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 421                         /* need to force clean */
 422                         sb->state |= (1 << MD_SB_CLEAN);
 423                         rv = 1;
 424                 }
 425         }
 426         if (strcmp(update, "assemble")==0) {
 427                 int d = info->disk.number;
 428                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 429                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 430                     != info->disk.state) {
 431                         sb->disks[d].state = info->disk.state | wonly;
 432                         rv = 1;
 433                 }
 434         }
 435         if (strcmp(update, "newdev") == 0) {
 436                 int d = info->disk.number;
 437                 memset(&sb->disks[d], 0, sizeof(sb->disks[d]));
 438                 sb->disks[d].number = d;
 439                 sb->disks[d].major = info->disk.major;
 440                 sb->disks[d].minor = info->disk.minor;
 441                 sb->disks[d].raid_disk = info->disk.raid_disk;
 442                 sb->disks[d].state = info->disk.state;
 443                 sb->this_disk = sb->disks[d];
 444         }
 445         if (strcmp(update, "grow") == 0) {
 446                 sb->raid_disks = info->array.raid_disks;
 447                 sb->nr_disks = info->array.nr_disks;
 448                 sb->active_disks = info->array.active_disks;
 449                 sb->working_disks = info->array.working_disks;
 450                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 451                 sb->disks[info->disk.number].number = info->disk.number;
 452                 sb->disks[info->disk.number].major = info->disk.major;
 453                 sb->disks[info->disk.number].minor = info->disk.minor;
 454                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 455                 sb->disks[info->disk.number].state = info->disk.state;
 456                 if (sb->this_disk.number == info->disk.number)
 457                         sb->this_disk = sb->disks[info->disk.number];
 458         }
 459         if (strcmp(update, "resync") == 0) {
 460                 /* make sure resync happens */
 461                 sb->state &= ~(1<<MD_SB_CLEAN);
 462                 sb->recovery_cp = 0;
 463         }
 464         if (strcmp(update, "homehost") == 0 &&
 465             homehost) {
 466                 uuid_set = 0;
 467                 update = "uuid";
 468                 info->uuid[0] = sb->set_uuid0;
 469                 info->uuid[1] = sb->set_uuid1;
 470         }
 471         if (strcmp(update, "uuid") == 0) {
 472                 if (!uuid_set && homehost) {
 473                         char buf[20];
 474                         char *hash = sha1_buffer(homehost,
 475                                                  strlen(homehost),
 476                                                  buf);
 477                         memcpy(info->uuid+2, hash, 8);
 478                 }
 479                 sb->set_uuid0 = info->uuid[0];
 480                 sb->set_uuid1 = info->uuid[1];
 481                 sb->set_uuid2 = info->uuid[2];
 482                 sb->set_uuid3 = info->uuid[3];
 483                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 484                         struct bitmap_super_s *bm;
 485                         bm = (struct bitmap_super_s*)(sb+1);
 486                         uuid_from_super0((int*)bm->uuid, sbv);
 487                 }
 488         }
 489         if (strcmp(update, "_reshape_progress")==0)
 490                 sb->reshape_position = info->reshape_progress;
 491
 492         sb->sb_csum = calc_sb0_csum(sb);
 493         return rv;
 494 }
 495
 496 static __u64 event_super0(void *sbv)
 497 {
 498         mdp_super_t *sb = sbv;
 499         return md_event(sb);
 500 }
 501
 502 /*
 503  * For verion-0 superblock, the homehost is 'stored' in the
 504  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 505  * of random material.
 506  * We use the first 8 bytes (64bits) of the sha1 of the
 507  * host name
 508  */
 509
 510
 511 static int init_super0(struct supertype *st, void **sbp, mdu_array_info_t *info,
 512                        unsigned long long size, char *ignored_name, char *homehost)
 513 {
 514         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 515         int spares;
 516         int rfd;
 517         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 518
 519         if (info->major_version == -1) {
 520                 /* zeroing the superblock */
 521                 *sbp = sb;
 522                 return 0;
 523         }
 524
 525         spares = info->working_disks - info->active_disks;
 526         if (info->raid_disks + spares  > MD_SB_DISKS) {
 527                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 528                         info->raid_disks , spares, MD_SB_DISKS);
 529                 return 0;
 530         }
 531
 532         rfd = open("/dev/urandom", O_RDONLY);
 533         sb->md_magic = MD_SB_MAGIC;
 534         sb->major_version = 0;
 535         sb->minor_version = 90;
 536         sb->patch_version = 0;
 537         sb->gvalid_words = 0; /* ignored */
 538         if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 539                 sb->set_uuid0 = random();
 540         sb->ctime = time(0);
 541         sb->level = info->level;
 542         if (size != info->size)
 543                 return 0;
 544         sb->size = info->size;
 545         sb->nr_disks = info->nr_disks;
 546         sb->raid_disks = info->raid_disks;
 547         sb->md_minor = info->md_minor;
 548         sb->not_persistent = 0;
 549         if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 550                 sb->set_uuid1 = random();
 551                 sb->set_uuid2 = random();
 552                 sb->set_uuid3 = random();
 553         }
 554         if (rfd >= 0)
 555                 close(rfd);
 556         if (homehost) {
 557                 char buf[20];
 558                 char *hash = sha1_buffer(homehost,
 559                                          strlen(homehost),
 560                                          buf);
 561                 memcpy(&sb->set_uuid2, hash, 8);
 562         }
 563
 564         sb->utime = sb->ctime;
 565         sb->state = info->state;
 566         sb->active_disks = info->active_disks;
 567         sb->working_disks = info->working_disks;
 568         sb->failed_disks = info->failed_disks;
 569         sb->spare_disks = info->spare_disks;
 570         sb->events_hi = 0;
 571         sb->events_lo = 1;
 572
 573         sb->layout = info->layout;
 574         sb->chunk_size = info->chunk_size;
 575
 576         *sbp = sb;
 577         return 1;
 578 }
 579
 580 /* Add a device to the superblock being created */
 581 static void add_to_super0(void *sbv, mdu_disk_info_t *dinfo)
 582 {
 583         mdp_super_t *sb = sbv;
 584         mdp_disk_t *dk = &sb->disks[dinfo->number];
 585
 586         dk->number = dinfo->number;
 587         dk->major = dinfo->major;
 588         dk->minor = dinfo->minor;
 589         dk->raid_disk = dinfo->raid_disk;
 590         dk->state = dinfo->state;
 591 }
 592
 593 static int store_super0(struct supertype *st, int fd, void *sbv)
 594 {
 595         unsigned long size;
 596         unsigned long long dsize;
 597         unsigned long long offset;
 598         mdp_super_t *super = sbv;
 599
 600 #ifdef BLKGETSIZE64
 601         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 602 #endif
 603         {
 604                 if (ioctl(fd, BLKGETSIZE, &size))
 605                         return 1;
 606                 else
 607                         dsize = ((unsigned long long)size)<<9;
 608         }
 609
 610         if (dsize < MD_RESERVED_SECTORS*2*512)
 611                 return 2;
 612
 613         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 614
 615         offset *= 512;
 616
 617         if (lseek64(fd, offset, 0)< 0LL)
 618                 return 3;
 619
 620         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 621                 return 4;
 622
 623         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 624                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 625                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 626                         if (write(fd, bm, sizeof(*bm)) != sizeof(*bm))
 627                             return 5;
 628         }
 629
 630         fsync(fd);
 631         return 0;
 632 }
 633
 634 static int write_init_super0(struct supertype *st, void *sbv, mdu_disk_info_t *dinfo, char *devname)
 635 {
 636         mdp_super_t *sb = sbv;
 637         int fd = open(devname, O_RDWR|O_EXCL);
 638         int rv;
 639
 640         if (fd < 0) {
 641                 fprintf(stderr, Name ": Failed to open %s to write superblock\n", devname);
 642                 return -1;
 643         }
 644
 645         sb->disks[dinfo->number].state &= ~(1<<MD_DISK_FAULTY);
 646
 647         sb->this_disk = sb->disks[dinfo->number];
 648         sb->sb_csum = calc_sb0_csum(sb);
 649         rv = store_super0(st, fd, sb);
 650
 651         if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 652                 rv = st->ss->write_bitmap(st, fd, sbv);
 653
 654         close(fd);
 655         if (rv)
 656                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
 657         return rv;
 658 }
 659
 660 static int compare_super0(void **firstp, void *secondv)
 661 {
 662         /*
 663          * return:
 664          *  0 same, or first was empty, and second was copied
 665          *  1 second had wrong number
 666          *  2 wrong uuid
 667          *  3 wrong other info
 668          */
 669         mdp_super_t *first = *firstp;
 670         mdp_super_t *second = secondv;
 671
 672         int uuid1[4], uuid2[4];
 673         if (second->md_magic != MD_SB_MAGIC)
 674                 return 1;
 675         if (!first) {
 676                 first = malloc(MD_SB_BYTES + sizeof(struct bitmap_super_s));
 677                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 678                 *firstp = first;
 679                 return 0;
 680         }
 681
 682         uuid_from_super0(uuid1, first);
 683         uuid_from_super0(uuid2, second);
 684         if (!same_uuid(uuid1, uuid2, 0))
 685                 return 2;
 686         if (first->major_version != second->major_version ||
 687             first->minor_version != second->minor_version ||
 688             first->patch_version != second->patch_version ||
 689             first->gvalid_words  != second->gvalid_words  ||
 690             first->ctime         != second->ctime         ||
 691             first->level         != second->level         ||
 692             first->size          != second->size          ||
 693             first->raid_disks    != second->raid_disks    )
 694                 return 3;
 695
 696         return 0;
 697 }
 698
 699
 700 static int load_super0(struct supertype *st, int fd, void **sbp, char *devname)
 701 {
 702         /* try to read in the superblock
 703          * Return:
 704          *  0 on success
 705          *  1 on cannot get superblock
 706          *  2 on superblock meaningless
 707          */
 708         unsigned long size;
 709         unsigned long long dsize;
 710         unsigned long long offset;
 711         mdp_super_t *super;
 712         int uuid[4];
 713         struct bitmap_super_s *bsb;
 714
 715 #ifdef BLKGETSIZE64
 716         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 717 #endif
 718         {
 719                 if (ioctl(fd, BLKGETSIZE, &size)) {
 720                         if (devname)
 721                                 fprintf(stderr, Name ": cannot find device size for %s: %s\n",
 722                                         devname, strerror(errno));
 723                         return 1;
 724                 } else
 725                         dsize = size << 9;
 726         }
 727
 728         if (dsize < MD_RESERVED_SECTORS*2) {
 729                 if (devname)
 730                         fprintf(stderr, Name ": %s is too small for md: size is %ld sectors.\n",
 731                                 devname, size);
 732                 return 1;
 733         }
 734
 735         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 736
 737         offset *= 512;
 738
 739         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 740
 741         if (lseek64(fd, offset, 0)< 0LL) {
 742                 if (devname)
 743                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 744                                 devname, strerror(errno));
 745                 return 1;
 746         }
 747
 748         super = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 749
 750         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 751                 if (devname)
 752                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 753                                 devname);
 754                 free(super);
 755                 return 1;
 756         }
 757
 758         if (st->ss && st->minor_version == 9)
 759                 super0_swap_endian(super);
 760
 761         if (super->md_magic != MD_SB_MAGIC) {
 762                 if (devname)
 763                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 764                                 devname, MD_SB_MAGIC, super->md_magic);
 765                 free(super);
 766                 return 2;
 767         }
 768
 769         if (super->major_version != 0) {
 770                 if (devname)
 771                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 772                                 devname, super->major_version);
 773                 free(super);
 774                 return 2;
 775         }
 776         *sbp = super;
 777         if (st->ss == NULL) {
 778                 st->ss = &super0;
 779                 st->minor_version = 90;
 780                 st->max_devs = MD_SB_DISKS;
 781         }
 782
 783         /* Now check on the bitmap superblock */
 784         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 785                 return 0;
 786         /* Read the bitmap superblock and make sure it looks
 787          * valid.  If it doesn't clear the bit.  An --assemble --force
 788          * should get that written out.
 789          */
 790         if (read(fd, super+1, sizeof(struct bitmap_super_s))
 791             != sizeof(struct bitmap_super_s))
 792                 goto no_bitmap;
 793
 794         uuid_from_super0(uuid, super);
 795         bsb = (struct bitmap_super_s *)(super+1);
 796         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 797             memcmp(bsb->uuid, uuid, 16) != 0)
 798                 goto no_bitmap;
 799         return 0;
 800
 801  no_bitmap:
 802         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 803
 804         return 0;
 805 }
 806
 807 static struct supertype *match_metadata_desc0(char *arg)
 808 {
 809         struct supertype *st = malloc(sizeof(*st));
 810         if (!st) return st;
 811
 812         st->ss = &super0;
 813         st->minor_version = 90;
 814         st->max_devs = MD_SB_DISKS;
 815         if (strcmp(arg, "0") == 0 ||
 816             strcmp(arg, "0.90") == 0 ||
 817             strcmp(arg, "default") == 0
 818                 )
 819                 return st;
 820
 821         st->minor_version = 9; /* flag for 'byte-swapped' */
 822         if (strcmp(arg, "0.swap")==0)
 823                 return st;
 824
 825         free(st);
 826         return NULL;
 827 }
 828
 829 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 830 {
 831         if (devsize < MD_RESERVED_SECTORS*2)
 832                 return 0ULL;
 833         return MD_NEW_SIZE_SECTORS(devsize);
 834 }
 835
 836 static int add_internal_bitmap0(struct supertype *st, void *sbv, int chunk, int delay, int write_behind, unsigned long long size, int may_change, int major)
 837 {
 838         /*
 839          * The bitmap comes immediately after the superblock and must be 60K in size
 840          * at most.  The default size is between 30K and 60K
 841          *
 842          * size is in sectors,  chunk is in bytes !!!
 843          */
 844         unsigned long long bits;
 845         unsigned long long max_bits = 60*1024*8;
 846         unsigned long long min_chunk;
 847         mdp_super_t *sb = sbv;
 848         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 849
 850
 851         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 852         bits = (size * 512) / min_chunk + 1;
 853         while (bits > max_bits) {
 854                 min_chunk *= 2;
 855                 bits = (bits+1)/2;
 856         }
 857         if (chunk == UnSet)
 858                 chunk = min_chunk;
 859         else if (chunk < min_chunk)
 860                 return 0; /* chunk size too small */
 861
 862         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 863
 864         memset(bms, 0, sizeof(*bms));
 865         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 866         bms->version = __cpu_to_le32(major);
 867         uuid_from_super0((int*)bms->uuid, sb);
 868         bms->chunksize = __cpu_to_le32(chunk);
 869         bms->daemon_sleep = __cpu_to_le32(delay);
 870         bms->sync_size = __cpu_to_le64(size);
 871         bms->write_behind = __cpu_to_le32(write_behind);
 872
 873         return 1;
 874 }
 875
 876
 877 void locate_bitmap0(struct supertype *st, int fd, void *sbv)
 878 {
 879         unsigned long long dsize;
 880         unsigned long size;
 881         unsigned long long offset;
 882 #ifdef BLKGETSIZE64
 883         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 884 #endif
 885         {
 886                 if (ioctl(fd, BLKGETSIZE, &size))
 887                         return;
 888                 else
 889                         dsize = ((unsigned long long)size)<<9;
 890         }
 891
 892         if (dsize < MD_RESERVED_SECTORS*2)
 893                 return;
 894
 895         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 896
 897         offset *= 512;
 898
 899         offset += MD_SB_BYTES;
 900
 901         lseek64(fd, offset, 0);
 902 }
 903
 904 int write_bitmap0(struct supertype *st, int fd, void *sbv)
 905 {
 906         unsigned long size;
 907         unsigned long long dsize;
 908         unsigned long long offset;
 909         mdp_super_t *sb = sbv;
 910
 911         int rv = 0;
 912
 913         int towrite, n;
 914         char buf[4096];
 915
 916 #ifdef BLKGETSIZE64
 917         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 918 #endif
 919         {
 920                 if (ioctl(fd, BLKGETSIZE, &size))
 921                         return 1;
 922                 else
 923                         dsize = ((unsigned long long)size)<<9;
 924         }
 925
 926         if (dsize < MD_RESERVED_SECTORS*2)
 927         return -1;
 928
 929         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 930
 931         offset *= 512;
 932
 933         if (lseek64(fd, offset + 4096, 0)< 0LL)
 934                 return 3;
 935
 936
 937         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
 938             sizeof(bitmap_super_t))
 939                 return -2;
 940         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
 941         memset(buf, 0xff, sizeof(buf));
 942         while (towrite > 0) {
 943                 n = towrite;
 944                 if (n > sizeof(buf))
 945                         n = sizeof(buf);
 946                 n = write(fd, buf, n);
 947                 if (n > 0)
 948                         towrite -= n;
 949                 else
 950                         break;
 951         }
 952         fsync(fd);
 953         if (towrite)
 954                 rv = -2;
 955
 956         return rv;
 957 }
 958
 959 struct superswitch super0 = {
 960 #ifndef MDASSEMBLE
 961         .examine_super = examine_super0,
 962         .brief_examine_super = brief_examine_super0,
 963         .detail_super = detail_super0,
 964         .brief_detail_super = brief_detail_super0,
 965 #endif
 966         .match_home = match_home0,
 967         .uuid_from_super = uuid_from_super0,
 968         .getinfo_super = getinfo_super0,
 969         .update_super = update_super0,
 970         .event_super = event_super0,
 971         .init_super = init_super0,
 972         .add_to_super = add_to_super0,
 973         .store_super = store_super0,
 974         .write_init_super = write_init_super0,
 975         .compare_super = compare_super0,
 976         .load_super = load_super0,
 977         .match_metadata_desc = match_metadata_desc0,
 978         .avail_size = avail_size0,
 979         .add_internal_bitmap = add_internal_bitmap0,
 980         .locate_bitmap = locate_bitmap0,
 981         .write_bitmap = write_bitmap0,
 982         .major = 0,
 983         .swapuuid = 0,
 984 };