super0.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@cse.unsw.edu.au>
  23  *    Paper: Neil Brown
  24  *           School of Computer Science and Engineering
  25  *           The University of New South Wales
  26  *           Sydney, 2052
  27  *           Australia
  28  */
  29
  30 #define HAVE_STDINT_H 1
  31 #include "mdadm.h"
  32 #include "sha1.h"
  33 /*
  34  * All handling for the 0.90.0 version superblock is in
  35  * this file.
  36  * This includes:
  37  *   - finding, loading, and writing the superblock.
  38  *   - initialising a new superblock
  39  *   - printing the superblock for --examine
  40  *   - printing part of the superblock for --detail
  41  * .. other stuff
  42  */
  43
  44
  45 static unsigned long calc_sb0_csum(mdp_super_t *super)
  46 {
  47         unsigned long csum = super->sb_csum;
  48         unsigned long newcsum;
  49         super->sb_csum= 0 ;
  50         newcsum = calc_csum(super, MD_SB_BYTES);
  51         super->sb_csum = csum;
  52         return newcsum;
  53 }
  54
  55
  56 void super0_swap_endian(struct mdp_superblock_s *sb)
  57 {
  58         /* as super0 superblocks are host-endian, it is sometimes
  59          * useful to be able to swap the endianness
  60          * as (almost) everything is u32's we byte-swap every 4byte
  61          * number.
  62          * We then also have to swap the events_hi and events_lo
  63          */
  64         char *sbc = (char *)sb;
  65         __u32 t32;
  66         int i;
  67
  68         for (i=0; i < MD_SB_BYTES ; i+=4) {
  69                 char t = sbc[i];
  70                 sbc[i] = sbc[i+3];
  71                 sbc[i+3] = t;
  72                 t=sbc[i+1];
  73                 sbc[i+1]=sbc[i+2];
  74                 sbc[i+2]=t;
  75         }
  76         t32 = sb->events_hi;
  77         sb->events_hi = sb->events_lo;
  78         sb->events_lo = t32;
  79
  80         t32 = sb->cp_events_hi;
  81         sb->cp_events_hi = sb->cp_events_lo;
  82         sb->cp_events_lo = t32;
  83
  84 }
  85
  86 #ifndef MDASSEMBLE
  87
  88 static void examine_super0(void *sbv, char *homehost)
  89 {
  90         mdp_super_t *sb = sbv;
  91         time_t atime;
  92         int d;
  93         char *c;
  94
  95         printf("          Magic : %08x\n", sb->md_magic);
  96         printf("        Version : %02d.%02d.%02d\n", sb->major_version, sb->minor_version,
  97                sb->patch_version);
  98         if (sb->minor_version >= 90) {
  99                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 100                        sb->set_uuid2, sb->set_uuid3);
 101                 if (homehost) {
 102                         char buf[20];
 103                         void *hash = sha1_buffer(homehost,
 104                                                  strlen(homehost),
 105                                                  buf);
 106                         if (memcmp(&sb->set_uuid2, hash, 8)==0)
 107                                 printf(" (local to host %s)", homehost);
 108                 }
 109                 printf("\n");
 110         } else
 111                 printf("           UUID : %08x\n", sb->set_uuid0);
 112
 113         atime = sb->ctime;
 114         printf("  Creation Time : %.24s\n", ctime(&atime));
 115         c=map_num(pers, sb->level);
 116         printf("     Raid Level : %s\n", c?c:"-unknown-");
 117         if ((int)sb->level >= 0) {
 118                 int ddsks=0;
 119                 printf("    Device Size : %d%s\n", sb->size, human_size((long long)sb->size<<10));
 120                 switch(sb->level) {
 121                 case 1: ddsks=1;break;
 122                 case 4:
 123                 case 5: ddsks = sb->raid_disks-1; break;
 124                 case 6: ddsks = sb->raid_disks-2; break;
 125                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 126                 }
 127                 if (ddsks)
 128                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 129                                human_size(ddsks*(long long)sb->size<<10));
 130         }
 131         printf("   Raid Devices : %d\n", sb->raid_disks);
 132         printf("  Total Devices : %d\n", sb->nr_disks);
 133         printf("Preferred Minor : %d\n", sb->md_minor);
 134         printf("\n");
 135         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 136                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 137                 if (sb->delta_disks) {
 138                         printf("  Delta Devices : %d", sb->delta_disks);
 139                         if (sb->delta_disks)
 140                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 141                         else
 142                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 143                 }
 144                 if (sb->new_level != sb->level) {
 145                         c = map_num(pers, sb->new_level);
 146                         printf("      New Level : %s\n", c?c:"-unknown-");
 147                 }
 148                 if (sb->new_layout != sb->layout) {
 149                         if (sb->level == 5) {
 150                                 c = map_num(r5layout, sb->new_layout);
 151                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 152                         }
 153                         if (sb->level == 10) {
 154                                 printf("     New Layout : near=%d, %s=%d\n",
 155                                        sb->new_layout&255,
 156                                        (sb->new_layout&0x10000)?"offset":"far",
 157                                        (sb->new_layout>>8)&255);
 158                         }
 159                 }
 160                 if (sb->new_chunk != sb->chunk_size)
 161                         printf("  New Chunksize : %d\n", sb->new_chunk);
 162                 printf("\n");
 163         }
 164         atime = sb->utime;
 165         printf("    Update Time : %.24s\n", ctime(&atime));
 166         printf("          State : %s\n",
 167                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 168         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 169                 printf("Internal Bitmap : present\n");
 170         printf(" Active Devices : %d\n", sb->active_disks);
 171         printf("Working Devices : %d\n", sb->working_disks);
 172         printf(" Failed Devices : %d\n", sb->failed_disks);
 173         printf("  Spare Devices : %d\n", sb->spare_disks);
 174         if (calc_sb0_csum(sb) == sb->sb_csum)
 175                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 176         else
 177                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 178         printf("         Events : %d.%d\n", sb->events_hi, sb->events_lo);
 179         printf("\n");
 180         if (sb->level == 5) {
 181                 c = map_num(r5layout, sb->layout);
 182                 printf("         Layout : %s\n", c?c:"-unknown-");
 183         }
 184         if (sb->level == 10) {
 185                 printf("         Layout : near=%d, %s=%d\n",
 186                        sb->layout&255,
 187                        (sb->layout&0x10000)?"offset":"far",
 188                        (sb->layout>>8)&255);
 189         }
 190         switch(sb->level) {
 191         case 0:
 192         case 4:
 193         case 5:
 194         case 6:
 195         case 10:
 196                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 197                 break;
 198         case -1:
 199                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 200                 break;
 201         default: break;
 202         }
 203         printf("\n");
 204         printf("      Number   Major   Minor   RaidDevice State\n");
 205         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
 206                 mdp_disk_t *dp;
 207                 char *dv;
 208                 char nb[5];
 209                 int wonly;
 210                 if (d>=0) dp = &sb->disks[d];
 211                 else dp = &sb->this_disk;
 212                 snprintf(nb, sizeof(nb), "%4d", d);
 213                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 214                        dp->number, dp->major, dp->minor, dp->raid_disk);
 215                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 216                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 217                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 218                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 219                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 220                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 221                 if (wonly) printf(" write-mostly");
 222                 if (dp->state == 0) printf(" spare");
 223                 if ((dv=map_dev(dp->major, dp->minor, 0)))
 224                         printf("   %s", dv);
 225                 printf("\n");
 226                 if (d == -1) printf("\n");
 227         }
 228 }
 229
 230 static void brief_examine_super0(void *sbv)
 231 {
 232         mdp_super_t *sb = sbv;
 233         char *c=map_num(pers, sb->level);
 234         char devname[20];
 235
 236         sprintf(devname, "/dev/md%d", sb->md_minor);
 237
 238         printf("ARRAY %s level=%s num-devices=%d UUID=",
 239                devname,
 240                c?c:"-unknown-", sb->raid_disks);
 241         if (sb->minor_version >= 90)
 242                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 243                        sb->set_uuid2, sb->set_uuid3);
 244         else
 245                 printf("%08x", sb->set_uuid0);
 246         printf("\n");
 247 }
 248
 249 static void detail_super0(void *sbv, char *homehost)
 250 {
 251         mdp_super_t *sb = sbv;
 252         printf("           UUID : ");
 253         if (sb->minor_version >= 90)
 254                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 255                        sb->set_uuid2, sb->set_uuid3);
 256         else
 257                 printf("%08x", sb->set_uuid0);
 258         if (homehost) {
 259                 char buf[20];
 260                 void *hash = sha1_buffer(homehost,
 261                                          strlen(homehost),
 262                                          buf);
 263                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 264                         printf(" (local to host %s)", homehost);
 265         }
 266         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 267 }
 268
 269 static void brief_detail_super0(void *sbv)
 270 {
 271         mdp_super_t *sb = sbv;
 272         printf(" UUID=");
 273         if (sb->minor_version >= 90)
 274                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 275                        sb->set_uuid2, sb->set_uuid3);
 276         else
 277                 printf("%08x", sb->set_uuid0);
 278 }
 279 #endif
 280
 281 static int match_home0(void *sbv, char *homehost)
 282 {
 283         mdp_super_t *sb = sbv;
 284         char buf[20];
 285         char *hash = sha1_buffer(homehost,
 286                                  strlen(homehost),
 287                                  buf);
 288
 289         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 290 }
 291
 292 static void uuid_from_super0(int uuid[4], void * sbv)
 293 {
 294         mdp_super_t *super = sbv;
 295         uuid[0] = super->set_uuid0;
 296         if (super->minor_version >= 90) {
 297                 uuid[1] = super->set_uuid1;
 298                 uuid[2] = super->set_uuid2;
 299                 uuid[3] = super->set_uuid3;
 300         } else {
 301                 uuid[1] = 0;
 302                 uuid[2] = 0;
 303                 uuid[3] = 0;
 304         }
 305 }
 306
 307 static void getinfo_super0(struct mdinfo *info, void *sbv)
 308 {
 309         mdp_super_t *sb = sbv;
 310         int working = 0;
 311         int i;
 312
 313         info->array.major_version = sb->major_version;
 314         info->array.minor_version = sb->minor_version;
 315         info->array.patch_version = sb->patch_version;
 316         info->array.raid_disks = sb->raid_disks;
 317         info->array.level = sb->level;
 318         info->array.layout = sb->layout;
 319         info->array.md_minor = sb->md_minor;
 320         info->array.ctime = sb->ctime;
 321         info->array.utime = sb->utime;
 322         info->array.chunk_size = sb->chunk_size;
 323         info->component_size = sb->size*2;
 324
 325         info->disk.state = sb->this_disk.state;
 326         info->disk.major = sb->this_disk.major;
 327         info->disk.minor = sb->this_disk.minor;
 328         info->disk.raid_disk = sb->this_disk.raid_disk;
 329         info->disk.number = sb->this_disk.number;
 330
 331         info->events = md_event(sb);
 332         info->data_offset = 0;
 333
 334         uuid_from_super0(info->uuid, sbv);
 335
 336         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 337                 info->reshape_active = 1;
 338                 info->reshape_progress = sb->reshape_position;
 339                 info->new_level = sb->new_level;
 340                 info->delta_disks = sb->delta_disks;
 341                 info->new_layout = sb->new_layout;
 342                 info->new_chunk = sb->new_chunk;
 343         } else
 344                 info->reshape_active = 0;
 345
 346         sprintf(info->name, "%d", sb->md_minor);
 347         /* work_disks is calculated rather than read directly */
 348         for (i=0; i < MD_SB_DISKS; i++)
 349                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 350                     (sb->disks[i].raid_disk < info->array.raid_disks) &&
 351                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 352                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 353                         working ++;
 354         info->array.working_disks = working;
 355 }
 356
 357
 358 static int update_super0(struct mdinfo *info, void *sbv, char *update,
 359                          char *devname, int verbose,
 360                          int uuid_set, char *homehost)
 361 {
 362         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 363          * For others, the return value is ignored.
 364          */
 365         int rv = 0;
 366         mdp_super_t *sb = sbv;
 367         if (strcmp(update, "sparc2.2")==0 ) {
 368                 /* 2.2 sparc put the events in the wrong place
 369                  * So we copy the tail of the superblock
 370                  * up 4 bytes before continuing
 371                  */
 372                 __u32 *sb32 = (__u32*)sb;
 373                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 374                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 375                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 376                 if (verbose >= 0)
 377                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 378                                  devname);
 379         }
 380         if (strcmp(update, "super-minor") ==0) {
 381                 sb->md_minor = info->array.md_minor;
 382                 if (verbose > 0)
 383                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 384                                 devname, info->array.md_minor);
 385         }
 386         if (strcmp(update, "summaries") == 0) {
 387                 int i;
 388                 /* set nr_disks, active_disks, working_disks,
 389                  * failed_disks, spare_disks based on disks[]
 390                  * array in superblock.
 391                  * Also make sure extra slots aren't 'failed'
 392                  */
 393                 sb->nr_disks = sb->active_disks =
 394                         sb->working_disks = sb->failed_disks =
 395                         sb->spare_disks = 0;
 396                 for (i=0; i < MD_SB_DISKS ; i++)
 397                         if (sb->disks[i].major ||
 398                             sb->disks[i].minor) {
 399                                 int state = sb->disks[i].state;
 400                                 if (state & (1<<MD_DISK_REMOVED))
 401                                         continue;
 402                                 sb->nr_disks++;
 403                                 if (state & (1<<MD_DISK_ACTIVE))
 404                                         sb->active_disks++;
 405                                 if (state & (1<<MD_DISK_FAULTY))
 406                                         sb->failed_disks++;
 407                                 else
 408                                         sb->working_disks++;
 409                                 if (state == 0)
 410                                         sb->spare_disks++;
 411                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 412                                 sb->disks[i].state = 0;
 413         }
 414         if (strcmp(update, "force")==0) {
 415                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 416                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 417                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 418                 if (sb->events_hi != ehi ||
 419                     sb->events_lo != elo)
 420                         rv = 1;
 421                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 422                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 423                         /* need to force clean */
 424                         sb->state |= (1 << MD_SB_CLEAN);
 425                         rv = 1;
 426                 }
 427         }
 428         if (strcmp(update, "assemble")==0) {
 429                 int d = info->disk.number;
 430                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 431                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 432                     != info->disk.state) {
 433                         sb->disks[d].state = info->disk.state | wonly;
 434                         rv = 1;
 435                 }
 436         }
 437         if (strcmp(update, "newdev") == 0) {
 438                 int d = info->disk.number;
 439                 memset(&sb->disks[d], 0, sizeof(sb->disks[d]));
 440                 sb->disks[d].number = d;
 441                 sb->disks[d].major = info->disk.major;
 442                 sb->disks[d].minor = info->disk.minor;
 443                 sb->disks[d].raid_disk = info->disk.raid_disk;
 444                 sb->disks[d].state = info->disk.state;
 445                 sb->this_disk = sb->disks[d];
 446         }
 447         if (strcmp(update, "grow") == 0) {
 448                 sb->raid_disks = info->array.raid_disks;
 449                 sb->nr_disks = info->array.nr_disks;
 450                 sb->active_disks = info->array.active_disks;
 451                 sb->working_disks = info->array.working_disks;
 452                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 453                 sb->disks[info->disk.number].number = info->disk.number;
 454                 sb->disks[info->disk.number].major = info->disk.major;
 455                 sb->disks[info->disk.number].minor = info->disk.minor;
 456                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 457                 sb->disks[info->disk.number].state = info->disk.state;
 458                 if (sb->this_disk.number == info->disk.number)
 459                         sb->this_disk = sb->disks[info->disk.number];
 460         }
 461         if (strcmp(update, "resync") == 0) {
 462                 /* make sure resync happens */
 463                 sb->state &= ~(1<<MD_SB_CLEAN);
 464                 sb->recovery_cp = 0;
 465         }
 466         if (strcmp(update, "homehost") == 0 &&
 467             homehost) {
 468                 uuid_set = 0;
 469                 update = "uuid";
 470                 info->uuid[0] = sb->set_uuid0;
 471                 info->uuid[1] = sb->set_uuid1;
 472         }
 473         if (strcmp(update, "uuid") == 0) {
 474                 if (!uuid_set && homehost) {
 475                         char buf[20];
 476                         char *hash = sha1_buffer(homehost,
 477                                                  strlen(homehost),
 478                                                  buf);
 479                         memcpy(info->uuid+2, hash, 8);
 480                 }
 481                 sb->set_uuid0 = info->uuid[0];
 482                 sb->set_uuid1 = info->uuid[1];
 483                 sb->set_uuid2 = info->uuid[2];
 484                 sb->set_uuid3 = info->uuid[3];
 485                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 486                         struct bitmap_super_s *bm;
 487                         bm = (struct bitmap_super_s*)(sb+1);
 488                         uuid_from_super0((int*)bm->uuid, sbv);
 489                 }
 490         }
 491         if (strcmp(update, "_reshape_progress")==0)
 492                 sb->reshape_position = info->reshape_progress;
 493
 494         sb->sb_csum = calc_sb0_csum(sb);
 495         return rv;
 496 }
 497
 498 static __u64 event_super0(void *sbv)
 499 {
 500         mdp_super_t *sb = sbv;
 501         return md_event(sb);
 502 }
 503
 504 /*
 505  * For verion-0 superblock, the homehost is 'stored' in the
 506  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 507  * of random material.
 508  * We use the first 8 bytes (64bits) of the sha1 of the
 509  * host name
 510  */
 511
 512
 513 static int init_super0(struct supertype *st, void **sbp, mdu_array_info_t *info,
 514                        unsigned long long size, char *ignored_name, char *homehost)
 515 {
 516         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 517         int spares;
 518         int rfd;
 519         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 520
 521         if (info->major_version == -1) {
 522                 /* zeroing the superblock */
 523                 *sbp = sb;
 524                 return 0;
 525         }
 526
 527         spares = info->working_disks - info->active_disks;
 528         if (info->raid_disks + spares  > MD_SB_DISKS) {
 529                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 530                         info->raid_disks , spares, MD_SB_DISKS);
 531                 return 0;
 532         }
 533
 534         rfd = open("/dev/urandom", O_RDONLY);
 535         sb->md_magic = MD_SB_MAGIC;
 536         sb->major_version = 0;
 537         sb->minor_version = 90;
 538         sb->patch_version = 0;
 539         sb->gvalid_words = 0; /* ignored */
 540         if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 541                 sb->set_uuid0 = random();
 542         sb->ctime = time(0);
 543         sb->level = info->level;
 544         if (size != info->size)
 545                 return 0;
 546         sb->size = info->size;
 547         sb->nr_disks = info->nr_disks;
 548         sb->raid_disks = info->raid_disks;
 549         sb->md_minor = info->md_minor;
 550         sb->not_persistent = 0;
 551         if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 552                 sb->set_uuid1 = random();
 553                 sb->set_uuid2 = random();
 554                 sb->set_uuid3 = random();
 555         }
 556         if (rfd >= 0)
 557                 close(rfd);
 558         if (homehost) {
 559                 char buf[20];
 560                 char *hash = sha1_buffer(homehost,
 561                                          strlen(homehost),
 562                                          buf);
 563                 memcpy(&sb->set_uuid2, hash, 8);
 564         }
 565
 566         sb->utime = sb->ctime;
 567         sb->state = info->state;
 568         sb->active_disks = info->active_disks;
 569         sb->working_disks = info->working_disks;
 570         sb->failed_disks = info->failed_disks;
 571         sb->spare_disks = info->spare_disks;
 572         sb->events_hi = 0;
 573         sb->events_lo = 1;
 574
 575         sb->layout = info->layout;
 576         sb->chunk_size = info->chunk_size;
 577
 578         *sbp = sb;
 579         return 1;
 580 }
 581
 582 /* Add a device to the superblock being created */
 583 static void add_to_super0(void *sbv, mdu_disk_info_t *dinfo)
 584 {
 585         mdp_super_t *sb = sbv;
 586         mdp_disk_t *dk = &sb->disks[dinfo->number];
 587
 588         dk->number = dinfo->number;
 589         dk->major = dinfo->major;
 590         dk->minor = dinfo->minor;
 591         dk->raid_disk = dinfo->raid_disk;
 592         dk->state = dinfo->state;
 593 }
 594
 595 static int store_super0(struct supertype *st, int fd, void *sbv)
 596 {
 597         unsigned long size;
 598         unsigned long long dsize;
 599         unsigned long long offset;
 600         mdp_super_t *super = sbv;
 601
 602 #ifdef BLKGETSIZE64
 603         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 604 #endif
 605         {
 606                 if (ioctl(fd, BLKGETSIZE, &size))
 607                         return 1;
 608                 else
 609                         dsize = ((unsigned long long)size)<<9;
 610         }
 611
 612         if (dsize < MD_RESERVED_SECTORS*2*512)
 613                 return 2;
 614
 615         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 616
 617         offset *= 512;
 618
 619         if (lseek64(fd, offset, 0)< 0LL)
 620                 return 3;
 621
 622         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 623                 return 4;
 624
 625         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 626                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 627                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 628                         if (write(fd, bm, sizeof(*bm)) != sizeof(*bm))
 629                             return 5;
 630         }
 631
 632         fsync(fd);
 633         return 0;
 634 }
 635
 636 static int write_init_super0(struct supertype *st, void *sbv, mdu_disk_info_t *dinfo, char *devname)
 637 {
 638         mdp_super_t *sb = sbv;
 639         int fd = open(devname, O_RDWR|O_EXCL);
 640         int rv;
 641
 642         if (fd < 0) {
 643                 fprintf(stderr, Name ": Failed to open %s to write superblock\n", devname);
 644                 return -1;
 645         }
 646
 647         sb->disks[dinfo->number].state &= ~(1<<MD_DISK_FAULTY);
 648
 649         sb->this_disk = sb->disks[dinfo->number];
 650         sb->sb_csum = calc_sb0_csum(sb);
 651         rv = store_super0(st, fd, sb);
 652
 653         if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 654                 rv = st->ss->write_bitmap(st, fd, sbv);
 655
 656         close(fd);
 657         if (rv)
 658                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
 659         return rv;
 660 }
 661
 662 static int compare_super0(void **firstp, void *secondv)
 663 {
 664         /*
 665          * return:
 666          *  0 same, or first was empty, and second was copied
 667          *  1 second had wrong number
 668          *  2 wrong uuid
 669          *  3 wrong other info
 670          */
 671         mdp_super_t *first = *firstp;
 672         mdp_super_t *second = secondv;
 673
 674         int uuid1[4], uuid2[4];
 675         if (second->md_magic != MD_SB_MAGIC)
 676                 return 1;
 677         if (!first) {
 678                 first = malloc(MD_SB_BYTES + sizeof(struct bitmap_super_s));
 679                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 680                 *firstp = first;
 681                 return 0;
 682         }
 683
 684         uuid_from_super0(uuid1, first);
 685         uuid_from_super0(uuid2, second);
 686         if (!same_uuid(uuid1, uuid2, 0))
 687                 return 2;
 688         if (first->major_version != second->major_version ||
 689             first->minor_version != second->minor_version ||
 690             first->patch_version != second->patch_version ||
 691             first->gvalid_words  != second->gvalid_words  ||
 692             first->ctime         != second->ctime         ||
 693             first->level         != second->level         ||
 694             first->size          != second->size          ||
 695             first->raid_disks    != second->raid_disks    )
 696                 return 3;
 697
 698         return 0;
 699 }
 700
 701
 702 static int load_super0(struct supertype *st, int fd, void **sbp, char *devname)
 703 {
 704         /* try to read in the superblock
 705          * Return:
 706          *  0 on success
 707          *  1 on cannot get superblock
 708          *  2 on superblock meaningless
 709          */
 710         unsigned long size;
 711         unsigned long long dsize;
 712         unsigned long long offset;
 713         mdp_super_t *super;
 714         int uuid[4];
 715         struct bitmap_super_s *bsb;
 716
 717 #ifdef BLKGETSIZE64
 718         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 719 #endif
 720         {
 721                 if (ioctl(fd, BLKGETSIZE, &size)) {
 722                         if (devname)
 723                                 fprintf(stderr, Name ": cannot find device size for %s: %s\n",
 724                                         devname, strerror(errno));
 725                         return 1;
 726                 } else
 727                         dsize = size << 9;
 728         }
 729
 730         if (dsize < MD_RESERVED_SECTORS*2) {
 731                 if (devname)
 732                         fprintf(stderr, Name ": %s is too small for md: size is %ld sectors.\n",
 733                                 devname, size);
 734                 return 1;
 735         }
 736
 737         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 738
 739         offset *= 512;
 740
 741         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 742
 743         if (lseek64(fd, offset, 0)< 0LL) {
 744                 if (devname)
 745                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 746                                 devname, strerror(errno));
 747                 return 1;
 748         }
 749
 750         super = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 751
 752         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 753                 if (devname)
 754                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 755                                 devname);
 756                 free(super);
 757                 return 1;
 758         }
 759
 760         if (st->ss && st->minor_version == 9)
 761                 super0_swap_endian(super);
 762
 763         if (super->md_magic != MD_SB_MAGIC) {
 764                 if (devname)
 765                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 766                                 devname, MD_SB_MAGIC, super->md_magic);
 767                 free(super);
 768                 return 2;
 769         }
 770
 771         if (super->major_version != 0) {
 772                 if (devname)
 773                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 774                                 devname, super->major_version);
 775                 free(super);
 776                 return 2;
 777         }
 778         *sbp = super;
 779         if (st->ss == NULL) {
 780                 st->ss = &super0;
 781                 st->minor_version = 90;
 782                 st->max_devs = MD_SB_DISKS;
 783         }
 784
 785         /* Now check on the bitmap superblock */
 786         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 787                 return 0;
 788         /* Read the bitmap superblock and make sure it looks
 789          * valid.  If it doesn't clear the bit.  An --assemble --force
 790          * should get that written out.
 791          */
 792         if (read(fd, super+1, sizeof(struct bitmap_super_s))
 793             != sizeof(struct bitmap_super_s))
 794                 goto no_bitmap;
 795
 796         uuid_from_super0(uuid, super);
 797         bsb = (struct bitmap_super_s *)(super+1);
 798         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 799             memcmp(bsb->uuid, uuid, 16) != 0)
 800                 goto no_bitmap;
 801         return 0;
 802
 803  no_bitmap:
 804         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 805
 806         return 0;
 807 }
 808
 809 static struct supertype *match_metadata_desc0(char *arg)
 810 {
 811         struct supertype *st = malloc(sizeof(*st));
 812         if (!st) return st;
 813
 814         st->ss = &super0;
 815         st->minor_version = 90;
 816         st->max_devs = MD_SB_DISKS;
 817         if (strcmp(arg, "0") == 0 ||
 818             strcmp(arg, "0.90") == 0 ||
 819             strcmp(arg, "default") == 0
 820                 )
 821                 return st;
 822
 823         st->minor_version = 9; /* flag for 'byte-swapped' */
 824         if (strcmp(arg, "0.swap")==0)
 825                 return st;
 826
 827         free(st);
 828         return NULL;
 829 }
 830
 831 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 832 {
 833         if (devsize < MD_RESERVED_SECTORS*2)
 834                 return 0ULL;
 835         return MD_NEW_SIZE_SECTORS(devsize);
 836 }
 837
 838 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)
 839 {
 840         /*
 841          * The bitmap comes immediately after the superblock and must be 60K in size
 842          * at most.  The default size is between 30K and 60K
 843          *
 844          * size is in sectors,  chunk is in bytes !!!
 845          */
 846         unsigned long long bits;
 847         unsigned long long max_bits = 60*1024*8;
 848         unsigned long long min_chunk;
 849         mdp_super_t *sb = sbv;
 850         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 851
 852
 853         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 854         bits = (size * 512) / min_chunk + 1;
 855         while (bits > max_bits) {
 856                 min_chunk *= 2;
 857                 bits = (bits+1)/2;
 858         }
 859         if (chunk == UnSet)
 860                 chunk = min_chunk;
 861         else if (chunk < min_chunk)
 862                 return 0; /* chunk size too small */
 863
 864         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 865
 866         memset(bms, 0, sizeof(*bms));
 867         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 868         bms->version = __cpu_to_le32(major);
 869         uuid_from_super0((int*)bms->uuid, sb);
 870         bms->chunksize = __cpu_to_le32(chunk);
 871         bms->daemon_sleep = __cpu_to_le32(delay);
 872         bms->sync_size = __cpu_to_le64(size);
 873         bms->write_behind = __cpu_to_le32(write_behind);
 874
 875         return 1;
 876 }
 877
 878
 879 void locate_bitmap0(struct supertype *st, int fd, void *sbv)
 880 {
 881         unsigned long long dsize;
 882         unsigned long size;
 883         unsigned long long offset;
 884 #ifdef BLKGETSIZE64
 885         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 886 #endif
 887         {
 888                 if (ioctl(fd, BLKGETSIZE, &size))
 889                         return;
 890                 else
 891                         dsize = ((unsigned long long)size)<<9;
 892         }
 893
 894         if (dsize < MD_RESERVED_SECTORS*2)
 895                 return;
 896
 897         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 898
 899         offset *= 512;
 900
 901         offset += MD_SB_BYTES;
 902
 903         lseek64(fd, offset, 0);
 904 }
 905
 906 int write_bitmap0(struct supertype *st, int fd, void *sbv)
 907 {
 908         unsigned long size;
 909         unsigned long long dsize;
 910         unsigned long long offset;
 911         mdp_super_t *sb = sbv;
 912
 913         int rv = 0;
 914
 915         int towrite, n;
 916         char buf[4096];
 917
 918 #ifdef BLKGETSIZE64
 919         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 920 #endif
 921         {
 922                 if (ioctl(fd, BLKGETSIZE, &size))
 923                         return 1;
 924                 else
 925                         dsize = ((unsigned long long)size)<<9;
 926         }
 927
 928         if (dsize < MD_RESERVED_SECTORS*2)
 929         return -1;
 930
 931         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 932
 933         offset *= 512;
 934
 935         if (lseek64(fd, offset + 4096, 0)< 0LL)
 936                 return 3;
 937
 938
 939         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
 940             sizeof(bitmap_super_t))
 941                 return -2;
 942         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
 943         memset(buf, 0xff, sizeof(buf));
 944         while (towrite > 0) {
 945                 n = towrite;
 946                 if (n > sizeof(buf))
 947                         n = sizeof(buf);
 948                 n = write(fd, buf, n);
 949                 if (n > 0)
 950                         towrite -= n;
 951                 else
 952                         break;
 953         }
 954         fsync(fd);
 955         if (towrite)
 956                 rv = -2;
 957
 958         return rv;
 959 }
 960
 961 struct superswitch super0 = {
 962 #ifndef MDASSEMBLE
 963         .examine_super = examine_super0,
 964         .brief_examine_super = brief_examine_super0,
 965         .detail_super = detail_super0,
 966         .brief_detail_super = brief_detail_super0,
 967 #endif
 968         .match_home = match_home0,
 969         .uuid_from_super = uuid_from_super0,
 970         .getinfo_super = getinfo_super0,
 971         .update_super = update_super0,
 972         .event_super = event_super0,
 973         .init_super = init_super0,
 974         .add_to_super = add_to_super0,
 975         .store_super = store_super0,
 976         .write_init_super = write_init_super0,
 977         .compare_super = compare_super0,
 978         .load_super = load_super0,
 979         .match_metadata_desc = match_metadata_desc0,
 980         .avail_size = avail_size0,
 981         .add_internal_bitmap = add_internal_bitmap0,
 982         .locate_bitmap = locate_bitmap0,
 983         .write_bitmap = write_bitmap0,
 984         .major = 0,
 985         .swapuuid = 0,
 986 };