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         if (sb->not_persistent)
 114                 printf("           Eedk : not persistent\n");
 115
 116         atime = sb->ctime;
 117         printf("  Creation Time : %.24s\n", ctime(&atime));
 118         c=map_num(pers, sb->level);
 119         printf("     Raid Level : %s\n", c?c:"-unknown-");
 120         if ((int)sb->level >= 0) {
 121                 int ddsks=0;
 122                 printf("  Used Dev Size : %d%s\n", sb->size,
 123                        human_size((long long)sb->size<<10));
 124                 switch(sb->level) {
 125                 case 1: ddsks=1;break;
 126                 case 4:
 127                 case 5: ddsks = sb->raid_disks-1; break;
 128                 case 6: ddsks = sb->raid_disks-2; break;
 129                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 130                 }
 131                 if (ddsks)
 132                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 133                                human_size(ddsks*(long long)sb->size<<10));
 134         }
 135         printf("   Raid Devices : %d\n", sb->raid_disks);
 136         printf("  Total Devices : %d\n", sb->nr_disks);
 137         printf("Preferred Minor : %d\n", sb->md_minor);
 138         printf("\n");
 139         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 140                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 141                 if (sb->delta_disks) {
 142                         printf("  Delta Devices : %d", sb->delta_disks);
 143                         if (sb->delta_disks)
 144                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 145                         else
 146                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 147                 }
 148                 if (sb->new_level != sb->level) {
 149                         c = map_num(pers, sb->new_level);
 150                         printf("      New Level : %s\n", c?c:"-unknown-");
 151                 }
 152                 if (sb->new_layout != sb->layout) {
 153                         if (sb->level == 5) {
 154                                 c = map_num(r5layout, sb->new_layout);
 155                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 156                         }
 157                         if (sb->level == 10) {
 158                                 printf("     New Layout : near=%d, %s=%d\n",
 159                                        sb->new_layout&255,
 160                                        (sb->new_layout&0x10000)?"offset":"far",
 161                                        (sb->new_layout>>8)&255);
 162                         }
 163                 }
 164                 if (sb->new_chunk != sb->chunk_size)
 165                         printf("  New Chunksize : %d\n", sb->new_chunk);
 166                 printf("\n");
 167         }
 168         atime = sb->utime;
 169         printf("    Update Time : %.24s\n", ctime(&atime));
 170         printf("          State : %s\n",
 171                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 172         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 173                 printf("Internal Bitmap : present\n");
 174         printf(" Active Devices : %d\n", sb->active_disks);
 175         printf("Working Devices : %d\n", sb->working_disks);
 176         printf(" Failed Devices : %d\n", sb->failed_disks);
 177         printf("  Spare Devices : %d\n", sb->spare_disks);
 178         if (calc_sb0_csum(sb) == sb->sb_csum)
 179                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 180         else
 181                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 182         printf("         Events : %d.%d\n", sb->events_hi, sb->events_lo);
 183         printf("\n");
 184         if (sb->level == 5) {
 185                 c = map_num(r5layout, sb->layout);
 186                 printf("         Layout : %s\n", c?c:"-unknown-");
 187         }
 188         if (sb->level == 10) {
 189                 printf("         Layout : near=%d, %s=%d\n",
 190                        sb->layout&255,
 191                        (sb->layout&0x10000)?"offset":"far",
 192                        (sb->layout>>8)&255);
 193         }
 194         switch(sb->level) {
 195         case 0:
 196         case 4:
 197         case 5:
 198         case 6:
 199         case 10:
 200                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 201                 break;
 202         case -1:
 203                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 204                 break;
 205         default: break;
 206         }
 207         printf("\n");
 208         printf("      Number   Major   Minor   RaidDevice State\n");
 209         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
 210                 mdp_disk_t *dp;
 211                 char *dv;
 212                 char nb[5];
 213                 int wonly;
 214                 if (d>=0) dp = &sb->disks[d];
 215                 else dp = &sb->this_disk;
 216                 snprintf(nb, sizeof(nb), "%4d", d);
 217                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 218                        dp->number, dp->major, dp->minor, dp->raid_disk);
 219                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 220                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 221                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 222                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 223                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 224                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 225                 if (wonly) printf(" write-mostly");
 226                 if (dp->state == 0) printf(" spare");
 227                 if ((dv=map_dev(dp->major, dp->minor, 0)))
 228                         printf("   %s", dv);
 229                 printf("\n");
 230                 if (d == -1) printf("\n");
 231         }
 232 }
 233
 234 static void brief_examine_super0(void *sbv)
 235 {
 236         mdp_super_t *sb = sbv;
 237         char *c=map_num(pers, sb->level);
 238         char devname[20];
 239
 240         sprintf(devname, "/dev/md%d", sb->md_minor);
 241
 242         printf("ARRAY %s level=%s num-devices=%d UUID=",
 243                devname,
 244                c?c:"-unknown-", sb->raid_disks);
 245         if (sb->minor_version >= 90)
 246                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 247                        sb->set_uuid2, sb->set_uuid3);
 248         else
 249                 printf("%08x", sb->set_uuid0);
 250         printf("\n");
 251 }
 252
 253 static void detail_super0(void *sbv, char *homehost)
 254 {
 255         mdp_super_t *sb = sbv;
 256         printf("           UUID : ");
 257         if (sb->minor_version >= 90)
 258                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 259                        sb->set_uuid2, sb->set_uuid3);
 260         else
 261                 printf("%08x", sb->set_uuid0);
 262         if (homehost) {
 263                 char buf[20];
 264                 void *hash = sha1_buffer(homehost,
 265                                          strlen(homehost),
 266                                          buf);
 267                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 268                         printf(" (local to host %s)", homehost);
 269         }
 270         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 271 }
 272
 273 static void brief_detail_super0(void *sbv)
 274 {
 275         mdp_super_t *sb = sbv;
 276         printf(" UUID=");
 277         if (sb->minor_version >= 90)
 278                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 279                        sb->set_uuid2, sb->set_uuid3);
 280         else
 281                 printf("%08x", sb->set_uuid0);
 282 }
 283
 284 static void export_super0(void *sbv)
 285 {
 286         mdp_super_t *sb = sbv;
 287         printf("MD_UUID=");
 288         if (sb->minor_version >= 90)
 289                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 290                        sb->set_uuid2, sb->set_uuid3);
 291         else
 292                 printf("%08x", sb->set_uuid0);
 293         printf("\n");
 294 }
 295 #endif
 296
 297 static int match_home0(void *sbv, char *homehost)
 298 {
 299         mdp_super_t *sb = sbv;
 300         char buf[20];
 301         char *hash = sha1_buffer(homehost,
 302                                  strlen(homehost),
 303                                  buf);
 304
 305         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 306 }
 307
 308 static void uuid_from_super0(int uuid[4], void * sbv)
 309 {
 310         mdp_super_t *super = sbv;
 311         uuid[0] = super->set_uuid0;
 312         if (super->minor_version >= 90) {
 313                 uuid[1] = super->set_uuid1;
 314                 uuid[2] = super->set_uuid2;
 315                 uuid[3] = super->set_uuid3;
 316         } else {
 317                 uuid[1] = 0;
 318                 uuid[2] = 0;
 319                 uuid[3] = 0;
 320         }
 321 }
 322
 323 static void getinfo_super0(struct mdinfo *info, void *sbv)
 324 {
 325         mdp_super_t *sb = sbv;
 326         int working = 0;
 327         int i;
 328
 329         info->array.major_version = sb->major_version;
 330         info->array.minor_version = sb->minor_version;
 331         info->array.patch_version = sb->patch_version;
 332         info->array.raid_disks = sb->raid_disks;
 333         info->array.level = sb->level;
 334         info->array.layout = sb->layout;
 335         info->array.md_minor = sb->md_minor;
 336         info->array.ctime = sb->ctime;
 337         info->array.utime = sb->utime;
 338         info->array.chunk_size = sb->chunk_size;
 339         info->array.state = sb->state;
 340         info->component_size = sb->size*2;
 341
 342         info->disk.state = sb->this_disk.state;
 343         info->disk.major = sb->this_disk.major;
 344         info->disk.minor = sb->this_disk.minor;
 345         info->disk.raid_disk = sb->this_disk.raid_disk;
 346         info->disk.number = sb->this_disk.number;
 347
 348         info->events = md_event(sb);
 349         info->data_offset = 0;
 350
 351         uuid_from_super0(info->uuid, sbv);
 352
 353         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 354                 info->reshape_active = 1;
 355                 info->reshape_progress = sb->reshape_position;
 356                 info->new_level = sb->new_level;
 357                 info->delta_disks = sb->delta_disks;
 358                 info->new_layout = sb->new_layout;
 359                 info->new_chunk = sb->new_chunk;
 360         } else
 361                 info->reshape_active = 0;
 362
 363         sprintf(info->name, "%d", sb->md_minor);
 364         /* work_disks is calculated rather than read directly */
 365         for (i=0; i < MD_SB_DISKS; i++)
 366                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 367                     (sb->disks[i].raid_disk < info->array.raid_disks) &&
 368                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 369                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 370                         working ++;
 371         info->array.working_disks = working;
 372 }
 373
 374
 375 static int update_super0(struct mdinfo *info, void *sbv, char *update,
 376                          char *devname, int verbose,
 377                          int uuid_set, char *homehost)
 378 {
 379         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 380          * For others, the return value is ignored.
 381          */
 382         int rv = 0;
 383         mdp_super_t *sb = sbv;
 384         if (strcmp(update, "sparc2.2")==0 ) {
 385                 /* 2.2 sparc put the events in the wrong place
 386                  * So we copy the tail of the superblock
 387                  * up 4 bytes before continuing
 388                  */
 389                 __u32 *sb32 = (__u32*)sb;
 390                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 391                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 392                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 393                 if (verbose >= 0)
 394                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 395                                  devname);
 396         }
 397         if (strcmp(update, "super-minor") ==0) {
 398                 sb->md_minor = info->array.md_minor;
 399                 if (verbose > 0)
 400                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 401                                 devname, info->array.md_minor);
 402         }
 403         if (strcmp(update, "summaries") == 0) {
 404                 int i;
 405                 /* set nr_disks, active_disks, working_disks,
 406                  * failed_disks, spare_disks based on disks[]
 407                  * array in superblock.
 408                  * Also make sure extra slots aren't 'failed'
 409                  */
 410                 sb->nr_disks = sb->active_disks =
 411                         sb->working_disks = sb->failed_disks =
 412                         sb->spare_disks = 0;
 413                 for (i=0; i < MD_SB_DISKS ; i++)
 414                         if (sb->disks[i].major ||
 415                             sb->disks[i].minor) {
 416                                 int state = sb->disks[i].state;
 417                                 if (state & (1<<MD_DISK_REMOVED))
 418                                         continue;
 419                                 sb->nr_disks++;
 420                                 if (state & (1<<MD_DISK_ACTIVE))
 421                                         sb->active_disks++;
 422                                 if (state & (1<<MD_DISK_FAULTY))
 423                                         sb->failed_disks++;
 424                                 else
 425                                         sb->working_disks++;
 426                                 if (state == 0)
 427                                         sb->spare_disks++;
 428                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 429                                 sb->disks[i].state = 0;
 430         }
 431         if (strcmp(update, "force-one")==0) {
 432                 /* Not enough devices for a working array, so
 433                  * bring this one up-to-date.
 434                  */
 435                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 436                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 437                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 438                 if (sb->events_hi != ehi ||
 439                     sb->events_lo != elo)
 440                         rv = 1;
 441         }
 442         if (strcmp(update, "force-array")==0) {
 443                 /* degraded array and 'force' requested, so
 444                  * maybe need to mark it 'clean'
 445                  */
 446                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 447                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 448                         /* need to force clean */
 449                         sb->state |= (1 << MD_SB_CLEAN);
 450                         rv = 1;
 451                 }
 452         }
 453         if (strcmp(update, "assemble")==0) {
 454                 int d = info->disk.number;
 455                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 456                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 457                     != info->disk.state) {
 458                         sb->disks[d].state = info->disk.state | wonly;
 459                         rv = 1;
 460                 }
 461         }
 462         if (strcmp(update, "grow") == 0) {
 463                 sb->raid_disks = info->array.raid_disks;
 464                 sb->nr_disks = info->array.nr_disks;
 465                 sb->active_disks = info->array.active_disks;
 466                 sb->working_disks = info->array.working_disks;
 467                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 468                 sb->disks[info->disk.number].number = info->disk.number;
 469                 sb->disks[info->disk.number].major = info->disk.major;
 470                 sb->disks[info->disk.number].minor = info->disk.minor;
 471                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 472                 sb->disks[info->disk.number].state = info->disk.state;
 473                 if (sb->this_disk.number == info->disk.number)
 474                         sb->this_disk = sb->disks[info->disk.number];
 475         }
 476         if (strcmp(update, "resync") == 0) {
 477                 /* make sure resync happens */
 478                 sb->state &= ~(1<<MD_SB_CLEAN);
 479                 sb->recovery_cp = 0;
 480         }
 481         if (strcmp(update, "homehost") == 0 &&
 482             homehost) {
 483                 uuid_set = 0;
 484                 update = "uuid";
 485                 info->uuid[0] = sb->set_uuid0;
 486                 info->uuid[1] = sb->set_uuid1;
 487         }
 488         if (strcmp(update, "uuid") == 0) {
 489                 if (!uuid_set && homehost) {
 490                         char buf[20];
 491                         char *hash = sha1_buffer(homehost,
 492                                                  strlen(homehost),
 493                                                  buf);
 494                         memcpy(info->uuid+2, hash, 8);
 495                 }
 496                 sb->set_uuid0 = info->uuid[0];
 497                 sb->set_uuid1 = info->uuid[1];
 498                 sb->set_uuid2 = info->uuid[2];
 499                 sb->set_uuid3 = info->uuid[3];
 500                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 501                         struct bitmap_super_s *bm;
 502                         bm = (struct bitmap_super_s*)(sb+1);
 503                         uuid_from_super0((int*)bm->uuid, sbv);
 504                 }
 505         }
 506         if (strcmp(update, "_reshape_progress")==0)
 507                 sb->reshape_position = info->reshape_progress;
 508
 509         sb->sb_csum = calc_sb0_csum(sb);
 510         return rv;
 511 }
 512
 513 /*
 514  * For verion-0 superblock, the homehost is 'stored' in the
 515  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 516  * of random material.
 517  * We use the first 8 bytes (64bits) of the sha1 of the
 518  * host name
 519  */
 520
 521
 522 static int init_super0(struct supertype *st, void **sbp, mdu_array_info_t *info,
 523                        unsigned long long size, char *ignored_name, char *homehost,
 524                        int *uuid)
 525 {
 526         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 527         int spares;
 528         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 529
 530         if (info->major_version == -1) {
 531                 /* zeroing the superblock */
 532                 *sbp = sb;
 533                 return 0;
 534         }
 535
 536         spares = info->working_disks - info->active_disks;
 537         if (info->raid_disks + spares  > MD_SB_DISKS) {
 538                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 539                         info->raid_disks , spares, MD_SB_DISKS);
 540                 return 0;
 541         }
 542
 543         sb->md_magic = MD_SB_MAGIC;
 544         sb->major_version = 0;
 545         sb->minor_version = 90;
 546         sb->patch_version = 0;
 547         sb->gvalid_words = 0; /* ignored */
 548         sb->ctime = time(0);
 549         sb->level = info->level;
 550         if (size != info->size)
 551                 return 0;
 552         sb->size = info->size;
 553         sb->nr_disks = info->nr_disks;
 554         sb->raid_disks = info->raid_disks;
 555         sb->md_minor = info->md_minor;
 556         sb->not_persistent = 0;
 557         if (uuid) {
 558                 sb->set_uuid0 = uuid[0];
 559                 sb->set_uuid1 = uuid[1];
 560                 sb->set_uuid2 = uuid[2];
 561                 sb->set_uuid3 = uuid[3];
 562         } else {
 563                 int rfd = open("/dev/urandom", O_RDONLY);
 564                 if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 565                         sb->set_uuid0 = random();
 566                 if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 567                         sb->set_uuid1 = random();
 568                         sb->set_uuid2 = random();
 569                         sb->set_uuid3 = random();
 570                 }
 571                 if (rfd >= 0)
 572                         close(rfd);
 573         }
 574         if (homehost) {
 575                 char buf[20];
 576                 char *hash = sha1_buffer(homehost,
 577                                          strlen(homehost),
 578                                          buf);
 579                 memcpy(&sb->set_uuid2, hash, 8);
 580         }
 581
 582         sb->utime = sb->ctime;
 583         sb->state = info->state;
 584         sb->active_disks = info->active_disks;
 585         sb->working_disks = info->working_disks;
 586         sb->failed_disks = info->failed_disks;
 587         sb->spare_disks = info->spare_disks;
 588         sb->events_hi = 0;
 589         sb->events_lo = 1;
 590
 591         sb->layout = info->layout;
 592         sb->chunk_size = info->chunk_size;
 593
 594         *sbp = sb;
 595         return 1;
 596 }
 597
 598 /* Add a device to the superblock being created */
 599 static void add_to_super0(void *sbv, mdu_disk_info_t *dinfo)
 600 {
 601         mdp_super_t *sb = sbv;
 602         mdp_disk_t *dk = &sb->disks[dinfo->number];
 603
 604         dk->number = dinfo->number;
 605         dk->major = dinfo->major;
 606         dk->minor = dinfo->minor;
 607         dk->raid_disk = dinfo->raid_disk;
 608         dk->state = dinfo->state;
 609 }
 610
 611 static int store_super0(struct supertype *st, int fd, void *sbv)
 612 {
 613         unsigned long long dsize;
 614         unsigned long long offset;
 615         mdp_super_t *super = sbv;
 616
 617         if (!get_dev_size(fd, NULL, &dsize))
 618                 return 1;
 619
 620         if (dsize < MD_RESERVED_SECTORS*2*512)
 621                 return 2;
 622
 623         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 624
 625         offset *= 512;
 626
 627         if (lseek64(fd, offset, 0)< 0LL)
 628                 return 3;
 629
 630         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 631                 return 4;
 632
 633         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 634                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 635                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 636                         if (write(fd, bm, sizeof(*bm)) != sizeof(*bm))
 637                             return 5;
 638         }
 639
 640         fsync(fd);
 641         return 0;
 642 }
 643
 644 static int write_init_super0(struct supertype *st, void *sbv, mdu_disk_info_t *dinfo, char *devname)
 645 {
 646         mdp_super_t *sb = sbv;
 647         int fd = open(devname, O_RDWR|O_EXCL);
 648         int rv;
 649
 650         if (fd < 0) {
 651                 fprintf(stderr, Name ": Failed to open %s to write superblock\n", devname);
 652                 return -1;
 653         }
 654
 655         sb->disks[dinfo->number].state &= ~(1<<MD_DISK_FAULTY);
 656
 657         sb->this_disk = sb->disks[dinfo->number];
 658         sb->sb_csum = calc_sb0_csum(sb);
 659         rv = store_super0(st, fd, sb);
 660
 661         if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 662                 rv = st->ss->write_bitmap(st, fd, sbv);
 663
 664         close(fd);
 665         if (rv)
 666                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
 667         return rv;
 668 }
 669
 670 static int compare_super0(void **firstp, void *secondv)
 671 {
 672         /*
 673          * return:
 674          *  0 same, or first was empty, and second was copied
 675          *  1 second had wrong number
 676          *  2 wrong uuid
 677          *  3 wrong other info
 678          */
 679         mdp_super_t *first = *firstp;
 680         mdp_super_t *second = secondv;
 681
 682         int uuid1[4], uuid2[4];
 683         if (second->md_magic != MD_SB_MAGIC)
 684                 return 1;
 685         if (!first) {
 686                 first = malloc(MD_SB_BYTES + sizeof(struct bitmap_super_s));
 687                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 688                 *firstp = first;
 689                 return 0;
 690         }
 691
 692         uuid_from_super0(uuid1, first);
 693         uuid_from_super0(uuid2, second);
 694         if (!same_uuid(uuid1, uuid2, 0))
 695                 return 2;
 696         if (first->major_version != second->major_version ||
 697             first->minor_version != second->minor_version ||
 698             first->patch_version != second->patch_version ||
 699             first->gvalid_words  != second->gvalid_words  ||
 700             first->ctime         != second->ctime         ||
 701             first->level         != second->level         ||
 702             first->size          != second->size          ||
 703             first->raid_disks    != second->raid_disks    )
 704                 return 3;
 705
 706         return 0;
 707 }
 708
 709
 710 static int load_super0(struct supertype *st, int fd, void **sbp, char *devname)
 711 {
 712         /* try to read in the superblock
 713          * Return:
 714          *  0 on success
 715          *  1 on cannot get superblock
 716          *  2 on superblock meaningless
 717          */
 718         unsigned long long dsize;
 719         unsigned long long offset;
 720         mdp_super_t *super;
 721         int uuid[4];
 722         struct bitmap_super_s *bsb;
 723
 724         if (!get_dev_size(fd, devname, &dsize))
 725                 return 1;
 726
 727         if (dsize < MD_RESERVED_SECTORS*512 * 2) {
 728                 if (devname)
 729                         fprintf(stderr, Name
 730                             ": %s is too small for md: size is %llu sectors.\n",
 731                                 devname, dsize);
 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 *chunkp,
 837                                 int delay, int write_behind,
 838                                 unsigned long long size, int may_change,
 839                                 int major)
 840 {
 841         /*
 842          * The bitmap comes immediately after the superblock and must be 60K in size
 843          * at most.  The default size is between 30K and 60K
 844          *
 845          * size is in sectors,  chunk is in bytes !!!
 846          */
 847         unsigned long long bits;
 848         unsigned long long max_bits = 60*1024*8;
 849         unsigned long long min_chunk;
 850         int chunk = *chunkp;
 851         mdp_super_t *sb = sbv;
 852         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 853
 854
 855         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 856         bits = (size * 512) / min_chunk + 1;
 857         while (bits > max_bits) {
 858                 min_chunk *= 2;
 859                 bits = (bits+1)/2;
 860         }
 861         if (chunk == UnSet)
 862                 chunk = min_chunk;
 863         else if (chunk < min_chunk)
 864                 return 0; /* chunk size too small */
 865
 866         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 867
 868         memset(bms, 0, sizeof(*bms));
 869         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 870         bms->version = __cpu_to_le32(major);
 871         uuid_from_super0((int*)bms->uuid, sb);
 872         bms->chunksize = __cpu_to_le32(chunk);
 873         bms->daemon_sleep = __cpu_to_le32(delay);
 874         bms->sync_size = __cpu_to_le64(size);
 875         bms->write_behind = __cpu_to_le32(write_behind);
 876         *chunkp = chunk;
 877         return 1;
 878 }
 879
 880
 881 void locate_bitmap0(struct supertype *st, int fd, void *sbv)
 882 {
 883         unsigned long long dsize;
 884         unsigned long long offset;
 885
 886         if (!get_dev_size(fd, NULL, &dsize))
 887                 return;
 888
 889         if (dsize < MD_RESERVED_SECTORS*512 * 2)
 890                 return;
 891
 892         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 893
 894         offset *= 512;
 895
 896         offset += MD_SB_BYTES;
 897
 898         lseek64(fd, offset, 0);
 899 }
 900
 901 int write_bitmap0(struct supertype *st, int fd, void *sbv)
 902 {
 903         unsigned long long dsize;
 904         unsigned long long offset;
 905         mdp_super_t *sb = sbv;
 906
 907         int rv = 0;
 908
 909         int towrite, n;
 910         char buf[4096];
 911
 912         if (!get_dev_size(fd, NULL, &dsize))
 913                 return 1;
 914
 915
 916         if (dsize < MD_RESERVED_SECTORS*512 * 2)
 917         return -1;
 918
 919         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 920
 921         offset *= 512;
 922
 923         if (lseek64(fd, offset + 4096, 0)< 0LL)
 924                 return 3;
 925
 926
 927         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
 928             sizeof(bitmap_super_t))
 929                 return -2;
 930         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
 931         memset(buf, 0xff, sizeof(buf));
 932         while (towrite > 0) {
 933                 n = towrite;
 934                 if (n > sizeof(buf))
 935                         n = sizeof(buf);
 936                 n = write(fd, buf, n);
 937                 if (n > 0)
 938                         towrite -= n;
 939                 else
 940                         break;
 941         }
 942         fsync(fd);
 943         if (towrite)
 944                 rv = -2;
 945
 946         return rv;
 947 }
 948
 949 struct superswitch super0 = {
 950 #ifndef MDASSEMBLE
 951         .examine_super = examine_super0,
 952         .brief_examine_super = brief_examine_super0,
 953         .detail_super = detail_super0,
 954         .brief_detail_super = brief_detail_super0,
 955         .export_super = export_super0,
 956 #endif
 957         .match_home = match_home0,
 958         .uuid_from_super = uuid_from_super0,
 959         .getinfo_super = getinfo_super0,
 960         .update_super = update_super0,
 961         .init_super = init_super0,
 962         .add_to_super = add_to_super0,
 963         .store_super = store_super0,
 964         .write_init_super = write_init_super0,
 965         .compare_super = compare_super0,
 966         .load_super = load_super0,
 967         .match_metadata_desc = match_metadata_desc0,
 968         .avail_size = avail_size0,
 969         .add_internal_bitmap = add_internal_bitmap0,
 970         .locate_bitmap = locate_bitmap0,
 971         .write_bitmap = write_bitmap0,
 972         .major = 0,
 973         .swapuuid = 0,
 974 };