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