super0.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2004 Neil Brown <neilb@cse.unsw.edu.au>
   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 #include "mdadm.h"
  31
  32 /*
  33  * All handling for the 0.90.0 version superblock is in
  34  * this file.
  35  * This includes:
  36  *   - finding, loading, and writing the superblock.
  37  *   - initialising a new superblock
  38  *   - printing the superblock for --examine
  39  *   - printing part of the superblock for --detail
  40  * .. other stuff
  41  */
  42
  43
  44 static unsigned long calc_sb0_csum(mdp_super_t *super)
  45 {
  46         unsigned long csum = super->sb_csum;
  47         unsigned long newcsum;
  48         super->sb_csum= 0 ;
  49         newcsum = calc_csum(super, MD_SB_BYTES);
  50         super->sb_csum = csum;
  51         return newcsum;
  52 }
  53
  54
  55 void super0_swap_endian(struct mdp_superblock_s *sb)
  56 {
  57         /* as super0 superblocks are host-endian, it is sometimes
  58          * useful to be able to swap the endianness
  59          * as (almost) everything is u32's we byte-swap every 4byte
  60          * number.
  61          * We then also have to swap the events_hi and events_lo
  62          */
  63         char *sbc = (char *)sb;
  64         __u32 t32;
  65         int i;
  66
  67         for (i=0; i < MD_SB_BYTES ; i+=4) {
  68                 char t = sbc[i];
  69                 sbc[i] = sbc[i+3];
  70                 sbc[i+3] = t;
  71                 t=sbc[i+1];
  72                 sbc[i+1]=sbc[i+2];
  73                 sbc[i+2]=t;
  74         }
  75         t32 = sb->events_hi;
  76         sb->events_hi = sb->events_lo;
  77         sb->events_lo = t32;
  78
  79         t32 = sb->cp_events_hi;
  80         sb->cp_events_hi = sb->cp_events_lo;
  81         sb->cp_events_lo = t32;
  82
  83 }
  84
  85 #ifndef MDASSEMBLE
  86
  87 static void examine_super0(void *sbv)
  88 {
  89         mdp_super_t *sb = sbv;
  90         time_t atime;
  91         int d;
  92         char *c;
  93
  94         printf("          Magic : %08x\n", sb->md_magic);
  95         printf("        Version : %02d.%02d.%02d\n", sb->major_version, sb->minor_version,
  96                sb->patch_version);
  97         if (sb->minor_version >= 90)
  98                 printf("           UUID : %08x:%08x:%08x:%08x\n", sb->set_uuid0, sb->set_uuid1,
  99                        sb->set_uuid2, sb->set_uuid3);
 100         else
 101                 printf("           UUID : %08x\n", sb->set_uuid0);
 102
 103         atime = sb->ctime;
 104         printf("  Creation Time : %.24s\n", ctime(&atime));
 105         c=map_num(pers, sb->level);
 106         printf("     Raid Level : %s\n", c?c:"-unknown-");
 107         if ((int)sb->level >= 0) {
 108                 int ddsks=0;
 109                 printf("    Device Size : %d%s\n", sb->size, human_size((long long)sb->size<<10));
 110                 switch(sb->level) {
 111                 case 1: ddsks=1;break;
 112                 case 4:
 113                 case 5: ddsks = sb->raid_disks-1; break;
 114                 case 6: ddsks = sb->raid_disks-2; break;
 115                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 116                 }
 117                 if (ddsks)
 118                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 119                                human_size(ddsks*(long long)sb->size<<10));
 120         }
 121         printf("   Raid Devices : %d\n", sb->raid_disks);
 122         printf("  Total Devices : %d\n", sb->nr_disks);
 123         printf("Preferred Minor : %d\n", sb->md_minor);
 124         printf("\n");
 125         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 126                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 127                 if (sb->delta_disks) {
 128                         printf("  Delta Devices : %d", sb->delta_disks);
 129                         if (sb->delta_disks)
 130                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 131                         else
 132                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 133                 }
 134                 if (sb->new_level != sb->level) {
 135                         c = map_num(pers, sb->new_level);
 136                         printf("      New Level : %s\n", c?c:"-unknown-");
 137                 }
 138                 if (sb->new_layout != sb->layout) {
 139                         if (sb->level == 5) {
 140                                 c = map_num(r5layout, sb->new_layout);
 141                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 142                         }
 143                         if (sb->level == 10) {
 144                                 printf("     New Layout : near=%d, far=%d\n",
 145                                        sb->new_layout&255, (sb->new_layout>>8)&255);
 146                         }
 147                 }
 148                 if (sb->new_chunk != sb->chunk_size)
 149                         printf("  New Chunksize : %d\n", sb->new_chunk);
 150                 printf("\n");
 151         }
 152         atime = sb->utime;
 153         printf("    Update Time : %.24s\n", ctime(&atime));
 154         printf("          State : %s\n",
 155                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 156         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 157                 printf("Internal Bitmap : present\n");
 158         printf(" Active Devices : %d\n", sb->active_disks);
 159         printf("Working Devices : %d\n", sb->working_disks);
 160         printf(" Failed Devices : %d\n", sb->failed_disks);
 161         printf("  Spare Devices : %d\n", sb->spare_disks);
 162         if (calc_sb0_csum(sb) == sb->sb_csum)
 163                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 164         else
 165                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 166         printf("         Events : %d.%d\n", sb->events_hi, sb->events_lo);
 167         printf("\n");
 168         if (sb->level == 5) {
 169                 c = map_num(r5layout, sb->layout);
 170                 printf("         Layout : %s\n", c?c:"-unknown-");
 171         }
 172         if (sb->level == 10) {
 173                 printf("         Layout : near=%d, far=%d\n",
 174                        sb->layout&255, (sb->layout>>8)&255);
 175         }
 176         switch(sb->level) {
 177         case 0:
 178         case 4:
 179         case 5:
 180                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 181                 break;
 182         case -1:
 183                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 184                 break;
 185         default: break;
 186         }
 187         printf("\n");
 188         printf("      Number   Major   Minor   RaidDevice State\n");
 189         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
 190                 mdp_disk_t *dp;
 191                 char *dv;
 192                 char nb[5];
 193                 int wonly;
 194                 if (d>=0) dp = &sb->disks[d];
 195                 else dp = &sb->this_disk;
 196                 snprintf(nb, sizeof(nb), "%4d", d);
 197                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 198                        dp->number, dp->major, dp->minor, dp->raid_disk);
 199                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 200                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 201                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 202                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 203                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 204                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 205                 if (wonly) printf(" write-mostly");
 206                 if (dp->state == 0) printf(" spare");
 207                 if ((dv=map_dev(dp->major, dp->minor)))
 208                         printf("   %s", dv);
 209                 printf("\n");
 210                 if (d == -1) printf("\n");
 211         }
 212 }
 213
 214 static void brief_examine_super0(void *sbv)
 215 {
 216         mdp_super_t *sb = sbv;
 217         char *c=map_num(pers, sb->level);
 218
 219         printf("ARRAY %s level=%s num-devices=%d UUID=",
 220                get_md_name(sb->md_minor),
 221                c?c:"-unknown-", sb->raid_disks);
 222         if (sb->minor_version >= 90)
 223                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 224                        sb->set_uuid2, sb->set_uuid3);
 225         else
 226                 printf("%08x", sb->set_uuid0);
 227         printf("\n");
 228 }
 229
 230 static void detail_super0(void *sbv)
 231 {
 232         mdp_super_t *sb = sbv;
 233         printf("           UUID : ");
 234         if (sb->minor_version >= 90)
 235                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 236                        sb->set_uuid2, sb->set_uuid3);
 237         else
 238                 printf("%08x", sb->set_uuid0);
 239         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 240 }
 241
 242 static void brief_detail_super0(void *sbv)
 243 {
 244         mdp_super_t *sb = sbv;
 245         printf(" UUID=");
 246         if (sb->minor_version >= 90)
 247                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 248                        sb->set_uuid2, sb->set_uuid3);
 249         else
 250                 printf("%08x", sb->set_uuid0);
 251 }
 252 #endif
 253 static void uuid_from_super0(int uuid[4], void * sbv)
 254 {
 255         mdp_super_t *super = sbv;
 256         uuid[0] = super->set_uuid0;
 257         if (super->minor_version >= 90) {
 258                 uuid[1] = super->set_uuid1;
 259                 uuid[2] = super->set_uuid2;
 260                 uuid[3] = super->set_uuid3;
 261         } else {
 262                 uuid[1] = 0;
 263                 uuid[2] = 0;
 264                 uuid[3] = 0;
 265         }
 266 }
 267
 268 static void getinfo_super0(struct mdinfo *info, mddev_ident_t ident, void *sbv)
 269 {
 270         mdp_super_t *sb = sbv;
 271         int working = 0;
 272         int i;
 273
 274         info->array.major_version = sb->major_version;
 275         info->array.minor_version = sb->minor_version;
 276         info->array.patch_version = sb->patch_version;
 277         info->array.raid_disks = sb->raid_disks;
 278         info->array.level = sb->level;
 279         info->array.layout = sb->layout;
 280         info->array.md_minor = sb->md_minor;
 281         info->array.ctime = sb->ctime;
 282
 283         info->disk.state = sb->this_disk.state;
 284         info->disk.major = sb->this_disk.major;
 285         info->disk.minor = sb->this_disk.minor;
 286         info->disk.raid_disk = sb->this_disk.raid_disk;
 287         info->disk.number = sb->this_disk.number;
 288
 289         info->events = md_event(sb);
 290
 291         uuid_from_super0(info->uuid, sbv);
 292
 293         ident->name[0] = 0;
 294         /* work_disks is calculated rather than read directly */
 295         for (i=0; i < MD_SB_DISKS; i++)
 296                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 297                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 298                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 299                         working ++;
 300         info->array.working_disks = working;
 301 }
 302
 303
 304 static int update_super0(struct mdinfo *info, void *sbv, char *update, char *devname, int verbose)
 305 {
 306         int rv = 0;
 307         mdp_super_t *sb = sbv;
 308         if (strcmp(update, "sparc2.2")==0 ) {
 309                 /* 2.2 sparc put the events in the wrong place
 310                  * So we copy the tail of the superblock
 311                  * up 4 bytes before continuing
 312                  */
 313                 __u32 *sb32 = (__u32*)sb;
 314                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 315                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 316                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 317                 if (verbose >= 0)
 318                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 319                                  devname);
 320         }
 321         if (strcmp(update, "super-minor") ==0) {
 322                 sb->md_minor = info->array.md_minor;
 323                 if (verbose > 0)
 324                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 325                                 devname, info->array.md_minor);
 326         }
 327         if (strcmp(update, "summaries") == 0) {
 328                 int i;
 329                 /* set nr_disks, active_disks, working_disks,
 330                  * failed_disks, spare_disks based on disks[]
 331                  * array in superblock.
 332                  * Also make sure extra slots aren't 'failed'
 333                  */
 334                 sb->nr_disks = sb->active_disks =
 335                         sb->working_disks = sb->failed_disks =
 336                         sb->spare_disks = 0;
 337                 for (i=0; i < MD_SB_DISKS ; i++)
 338                         if (sb->disks[i].major ||
 339                             sb->disks[i].minor) {
 340                                 int state = sb->disks[i].state;
 341                                 if (state & (1<<MD_DISK_REMOVED))
 342                                         continue;
 343                                 sb->nr_disks++;
 344                                 if (state & (1<<MD_DISK_ACTIVE))
 345                                         sb->active_disks++;
 346                                 if (state & (1<<MD_DISK_FAULTY))
 347                                         sb->failed_disks++;
 348                                 else
 349                                         sb->working_disks++;
 350                                 if (state == 0)
 351                                         sb->spare_disks++;
 352                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 353                                 sb->disks[i].state = 0;
 354         }
 355         if (strcmp(update, "force")==0) {
 356                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 357                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 358                 if (sb->level == 5 || sb->level == 4 || sb->level == 6)
 359                         /* need to force clean */
 360                         sb->state |= (1 << MD_SB_CLEAN);
 361         }
 362         if (strcmp(update, "assemble")==0) {
 363                 int d = info->disk.number;
 364                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 365                 sb->disks[d].state &= ~(1<<MD_DISK_WRITEMOSTLY);
 366                 if (sb->disks[d].state != info->disk.state) {
 367                         sb->disks[d].state = info->disk.state & wonly;
 368                         rv = 1;
 369                 }
 370         }
 371         if (strcmp(update, "newdev") == 0) {
 372                 int d = info->disk.number;
 373                 memset(&sb->disks[d], 0, sizeof(sb->disks[d]));
 374                 sb->disks[d].number = d;
 375                 sb->disks[d].major = info->disk.major;
 376                 sb->disks[d].minor = info->disk.minor;
 377                 sb->disks[d].raid_disk = info->disk.raid_disk;
 378                 sb->disks[d].state = info->disk.state;
 379                 sb->this_disk = sb->disks[d];
 380         }
 381         if (strcmp(update, "grow") == 0) {
 382                 sb->raid_disks = info->array.raid_disks;
 383                 sb->nr_disks = info->array.nr_disks;
 384                 sb->active_disks = info->array.active_disks;
 385                 sb->working_disks = info->array.working_disks;
 386                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 387                 sb->disks[info->disk.number].number = info->disk.number;
 388                 sb->disks[info->disk.number].major = info->disk.major;
 389                 sb->disks[info->disk.number].minor = info->disk.minor;
 390                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 391                 sb->disks[info->disk.number].state = info->disk.state;
 392                 if (sb->this_disk.number == info->disk.number)
 393                         sb->this_disk = sb->disks[info->disk.number];
 394         }
 395         if (strcmp(update, "resync") == 0) {
 396                 /* make sure resync happens */
 397                 sb->state &= ~(1<<MD_SB_CLEAN);
 398                 sb->recovery_cp = 0;
 399         }
 400         if (strcmp(update, "uuid") == 0) {
 401                 sb->set_uuid0 = info->uuid[0];
 402                 sb->set_uuid1 = info->uuid[1];
 403                 sb->set_uuid2 = info->uuid[2];
 404                 sb->set_uuid3 = info->uuid[3];
 405         }
 406
 407         sb->sb_csum = calc_sb0_csum(sb);
 408         return rv;
 409 }
 410
 411 static __u64 event_super0(void *sbv)
 412 {
 413         mdp_super_t *sb = sbv;
 414         return md_event(sb);
 415 }
 416
 417
 418
 419 static int init_super0(struct supertype *st, void **sbp, mdu_array_info_t *info, unsigned long long size, char *ignored_name)
 420 {
 421         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 422         int spares;
 423         int rfd;
 424         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 425
 426         if (info->major_version == -1) {
 427                 /* zeroing the superblock */
 428                 return 0;
 429         }
 430
 431         spares = info->working_disks - info->active_disks;
 432         if (info->raid_disks + spares  > MD_SB_DISKS) {
 433                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 434                         info->raid_disks , spares, MD_SB_DISKS);
 435                 return 0;
 436         }
 437
 438         rfd = open("/dev/urandom", O_RDONLY);
 439         sb->md_magic = MD_SB_MAGIC;
 440         sb->major_version = 0;
 441         sb->minor_version = 90;
 442         sb->patch_version = 0;
 443         sb->gvalid_words = 0; /* ignored */
 444         if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 445                 sb->set_uuid0 = random();
 446         sb->ctime = time(0);
 447         sb->level = info->level;
 448         if (size != info->size)
 449                 return 0;
 450         sb->size = info->size;
 451         sb->nr_disks = info->nr_disks;
 452         sb->raid_disks = info->raid_disks;
 453         sb->md_minor = info->md_minor;
 454         sb->not_persistent = 0;
 455         if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 456                 sb->set_uuid1 = random();
 457                 sb->set_uuid2 = random();
 458                 sb->set_uuid3 = random();
 459         }
 460         if (rfd >= 0)
 461                 close(rfd);
 462
 463         sb->utime = sb->ctime;
 464         sb->state = info->state;
 465         sb->active_disks = info->active_disks;
 466         sb->working_disks = info->working_disks;
 467         sb->failed_disks = info->failed_disks;
 468         sb->spare_disks = info->spare_disks;
 469         sb->events_hi = 0;
 470         sb->events_lo = 1;
 471
 472         sb->layout = info->layout;
 473         sb->chunk_size = info->chunk_size;
 474
 475         *sbp = sb;
 476         return 1;
 477 }
 478
 479 /* Add a device to the superblock being created */
 480 static void add_to_super0(void *sbv, mdu_disk_info_t *dinfo)
 481 {
 482         mdp_super_t *sb = sbv;
 483         mdp_disk_t *dk = &sb->disks[dinfo->number];
 484
 485         dk->number = dinfo->number;
 486         dk->major = dinfo->major;
 487         dk->minor = dinfo->minor;
 488         dk->raid_disk = dinfo->raid_disk;
 489         dk->state = dinfo->state;
 490 }
 491
 492 static int store_super0(struct supertype *st, int fd, void *sbv)
 493 {
 494         unsigned long size;
 495         unsigned long long dsize;
 496         unsigned long long offset;
 497         mdp_super_t *super = sbv;
 498
 499 #ifdef BLKGETSIZE64
 500         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 501 #endif
 502         {
 503                 if (ioctl(fd, BLKGETSIZE, &size))
 504                         return 1;
 505                 else
 506                         dsize = ((unsigned long long)size)<<9;
 507         }
 508
 509         if (dsize < MD_RESERVED_SECTORS*2*512)
 510                 return 2;
 511
 512         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 513
 514         offset *= 512;
 515
 516         if (lseek64(fd, offset, 0)< 0LL)
 517                 return 3;
 518
 519         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 520                 return 4;
 521
 522         fsync(fd);
 523         return 0;
 524 }
 525
 526 static int write_init_super0(struct supertype *st, void *sbv, mdu_disk_info_t *dinfo, char *devname)
 527 {
 528         mdp_super_t *sb = sbv;
 529         int fd = open(devname, O_RDWR|O_EXCL);
 530         int rv;
 531
 532         if (fd < 0) {
 533                 fprintf(stderr, Name ": Failed to open %s to write superblock\n", devname);
 534                 return -1;
 535         }
 536
 537         sb->disks[dinfo->number].state &= ~(1<<MD_DISK_FAULTY);
 538         sb->disks[dinfo->number].state |= (1<<MD_DISK_SYNC);
 539
 540         sb->this_disk = sb->disks[dinfo->number];
 541         sb->sb_csum = calc_sb0_csum(sb);
 542         rv = store_super0(st, fd, sb);
 543
 544         if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 545                 rv = st->ss->write_bitmap(st, fd, sbv);
 546
 547         close(fd);
 548         if (rv)
 549                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
 550         return rv;
 551 }
 552
 553 static int compare_super0(void **firstp, void *secondv)
 554 {
 555         /*
 556          * return:
 557          *  0 same, or first was empty, and second was copied
 558          *  1 second had wrong number
 559          *  2 wrong uuid
 560          *  3 wrong other info
 561          */
 562         mdp_super_t *first = *firstp;
 563         mdp_super_t *second = secondv;
 564
 565         int uuid1[4], uuid2[4];
 566         if (second->md_magic != MD_SB_MAGIC)
 567                 return 1;
 568         if (!first) {
 569                 first = malloc(MD_SB_BYTES);
 570                 memcpy(first, second, MD_SB_BYTES);
 571                 *firstp = first;
 572                 return 0;
 573         }
 574
 575         uuid_from_super0(uuid1, first);
 576         uuid_from_super0(uuid2, second);
 577         if (!same_uuid(uuid1, uuid2, 0))
 578                 return 2;
 579         if (first->major_version != second->major_version ||
 580             first->minor_version != second->minor_version ||
 581             first->patch_version != second->patch_version ||
 582             first->gvalid_words  != second->gvalid_words  ||
 583             first->ctime         != second->ctime         ||
 584             first->level         != second->level         ||
 585             first->size          != second->size          ||
 586             first->raid_disks    != second->raid_disks    )
 587                 return 3;
 588
 589         return 0;
 590 }
 591
 592
 593 static int load_super0(struct supertype *st, int fd, void **sbp, char *devname)
 594 {
 595         /* try to read in the superblock
 596          * Return:
 597          *  0 on success
 598          *  1 on cannot get superblock
 599          *  2 on superblock meaningless
 600          */
 601         unsigned long size;
 602         unsigned long long dsize;
 603         unsigned long long offset;
 604         mdp_super_t *super;
 605
 606 #ifdef BLKGETSIZE64
 607         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 608 #endif
 609         {
 610                 if (ioctl(fd, BLKGETSIZE, &size)) {
 611                         if (devname)
 612                                 fprintf(stderr, Name ": cannot find device size for %s: %s\n",
 613                                         devname, strerror(errno));
 614                         return 1;
 615                 } else
 616                         dsize = size << 9;
 617         }
 618
 619         if (dsize < MD_RESERVED_SECTORS*2) {
 620                 if (devname)
 621                         fprintf(stderr, Name ": %s is too small for md: size is %ld sectors.\n",
 622                                 devname, size);
 623                 return 1;
 624         }
 625
 626         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 627
 628         offset *= 512;
 629
 630         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 631
 632         if (lseek64(fd, offset, 0)< 0LL) {
 633                 if (devname)
 634                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 635                                 devname, strerror(errno));
 636                 return 1;
 637         }
 638
 639         super = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 640
 641         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 642                 if (devname)
 643                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 644                                 devname);
 645                 free(super);
 646                 return 1;
 647         }
 648
 649         if (st->ss && st->minor_version == 9)
 650                 super0_swap_endian(super);
 651
 652         if (super->md_magic != MD_SB_MAGIC) {
 653                 if (devname)
 654                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 655                                 devname, MD_SB_MAGIC, super->md_magic);
 656                 free(super);
 657                 return 2;
 658         }
 659
 660         if (super->major_version != 0) {
 661                 if (devname)
 662                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 663                                 devname, super->major_version);
 664                 free(super);
 665                 return 2;
 666         }
 667         *sbp = super;
 668         if (st->ss == NULL) {
 669                 st->ss = &super0;
 670                 st->minor_version = 90;
 671                 st->max_devs = MD_SB_DISKS;
 672         }
 673
 674         return 0;
 675 }
 676
 677 static struct supertype *match_metadata_desc0(char *arg)
 678 {
 679         struct supertype *st = malloc(sizeof(*st));
 680         if (!st) return st;
 681
 682         st->ss = &super0;
 683         st->minor_version = 90;
 684         st->max_devs = MD_SB_DISKS;
 685         if (strcmp(arg, "0") == 0 ||
 686             strcmp(arg, "0.90") == 0 ||
 687             strcmp(arg, "default") == 0
 688                 )
 689                 return st;
 690
 691         st->minor_version = 9; /* flag for 'byte-swapped' */
 692         if (strcmp(arg, "0.swap")==0)
 693                 return st;
 694
 695         free(st);
 696         return NULL;
 697 }
 698
 699 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 700 {
 701         if (devsize < MD_RESERVED_SECTORS*2)
 702                 return 0ULL;
 703         return MD_NEW_SIZE_SECTORS(devsize);
 704 }
 705
 706 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)
 707 {
 708         /*
 709          * The bitmap comes immediately after the superblock and must be 60K in size
 710          * at most.  The default size is between 30K and 60K
 711          *
 712          * size is in K,  chunk is in bytes !!!
 713          */
 714         unsigned long long bits;
 715         unsigned long long max_bits = 60*1024*8;
 716         unsigned long long min_chunk;
 717         mdp_super_t *sb = sbv;
 718         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 719
 720
 721         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 722         bits = (size * 512)/ min_chunk +1;
 723         while (bits > max_bits) {
 724                 min_chunk *= 2;
 725                 bits = (bits+1)/2;
 726         }
 727         if (chunk == UnSet)
 728                 chunk = min_chunk;
 729         else if (chunk < min_chunk)
 730                 return 0; /* chunk size too small */
 731
 732         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 733
 734         memset(bms, 0, sizeof(*bms));
 735         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 736         bms->version = __cpu_to_le32(major);
 737         uuid_from_super0((int*)bms->uuid, sb);
 738         bms->chunksize = __cpu_to_le32(chunk);
 739         bms->daemon_sleep = __cpu_to_le32(delay);
 740         bms->sync_size = __cpu_to_le64(size);
 741         bms->write_behind = __cpu_to_le32(write_behind);
 742
 743
 744
 745         return 1;
 746 }
 747
 748
 749 void locate_bitmap0(struct supertype *st, int fd, void *sbv)
 750 {
 751         unsigned long long dsize;
 752         unsigned long size;
 753         unsigned long long offset;
 754 #ifdef BLKGETSIZE64
 755         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 756 #endif
 757         {
 758                 if (ioctl(fd, BLKGETSIZE, &size))
 759                         return;
 760                 else
 761                         dsize = ((unsigned long long)size)<<9;
 762         }
 763
 764         if (dsize < MD_RESERVED_SECTORS*2)
 765                 return;
 766
 767         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 768
 769         offset *= 512;
 770
 771         offset += MD_SB_BYTES;
 772
 773         lseek64(fd, offset, 0);
 774 }
 775
 776 int write_bitmap0(struct supertype *st, int fd, void *sbv)
 777 {
 778         unsigned long size;
 779         unsigned long long dsize;
 780         unsigned long long offset;
 781         mdp_super_t *sb = sbv;
 782
 783         int rv = 0;
 784
 785         int towrite, n;
 786         char buf[4096];
 787
 788 #ifdef BLKGETSIZE64
 789         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 790 #endif
 791         {
 792                 if (ioctl(fd, BLKGETSIZE, &size))
 793                         return 1;
 794                 else
 795                         dsize = ((unsigned long long)size)<<9;
 796         }
 797
 798         if (dsize < MD_RESERVED_SECTORS*2)
 799                 return -1;
 800
 801         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 802
 803         offset *= 512;
 804
 805         if (lseek64(fd, offset + 4096, 0)< 0LL)
 806                 return 3;
 807
 808
 809         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
 810             sizeof(bitmap_super_t))
 811                 return -2;
 812         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
 813         memset(buf, 0xff, sizeof(buf));
 814         while (towrite > 0) {
 815                 n = towrite;
 816                 if (n > sizeof(buf))
 817                         n = sizeof(buf);
 818                 n = write(fd, buf, n);
 819                 if (n > 0)
 820                         towrite -= n;
 821                 else
 822                         break;
 823         }
 824         fsync(fd);
 825         if (towrite)
 826                 rv = -2;
 827
 828         return rv;
 829 }
 830
 831 struct superswitch super0 = {
 832 #ifndef MDASSEMBLE
 833         .examine_super = examine_super0,
 834         .brief_examine_super = brief_examine_super0,
 835         .detail_super = detail_super0,
 836         .brief_detail_super = brief_detail_super0,
 837 #endif
 838         .uuid_from_super = uuid_from_super0,
 839         .getinfo_super = getinfo_super0,
 840         .update_super = update_super0,
 841         .event_super = event_super0,
 842         .init_super = init_super0,
 843         .add_to_super = add_to_super0,
 844         .store_super = store_super0,
 845         .write_init_super = write_init_super0,
 846         .compare_super = compare_super0,
 847         .load_super = load_super0,
 848         .match_metadata_desc = match_metadata_desc0,
 849         .avail_size = avail_size0,
 850         .add_internal_bitmap = add_internal_bitmap0,
 851         .locate_bitmap = locate_bitmap0,
 852         .write_bitmap = write_bitmap0,
 853         .major = 0,
 854         .swapuuid = 0,
 855 };