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 static 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(struct supertype *st, char *homehost)
  89 {
  90         mdp_super_t *sb = st->sb;
  91         time_t atime;
  92         int d;
  93         char *c;
  94
  95         printf("          Magic : %08x\n", sb->md_magic);
  96         printf("        Version : %02d.%02d.%02d\n", sb->major_version, sb->minor_version,
  97                sb->patch_version);
  98         if (sb->minor_version >= 90) {
  99                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 100                        sb->set_uuid2, sb->set_uuid3);
 101                 if (homehost) {
 102                         char buf[20];
 103                         void *hash = sha1_buffer(homehost,
 104                                                  strlen(homehost),
 105                                                  buf);
 106                         if (memcmp(&sb->set_uuid2, hash, 8)==0)
 107                                 printf(" (local to host %s)", homehost);
 108                 }
 109                 printf("\n");
 110         } else
 111                 printf("           UUID : %08x\n", sb->set_uuid0);
 112
 113         if (sb->not_persistent)
 114                 printf("           Eedk : not persistent\n");
 115
 116         atime = sb->ctime;
 117         printf("  Creation Time : %.24s\n", ctime(&atime));
 118         c=map_num(pers, sb->level);
 119         printf("     Raid Level : %s\n", c?c:"-unknown-");
 120         if ((int)sb->level >= 0) {
 121                 int ddsks=0;
 122                 printf("  Used Dev Size : %d%s\n", sb->size,
 123                        human_size((long long)sb->size<<10));
 124                 switch(sb->level) {
 125                 case 1: ddsks=1;break;
 126                 case 4:
 127                 case 5: ddsks = sb->raid_disks-1; break;
 128                 case 6: ddsks = sb->raid_disks-2; break;
 129                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 130                 }
 131                 if (ddsks)
 132                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 133                                human_size(ddsks*(long long)sb->size<<10));
 134         }
 135         printf("   Raid Devices : %d\n", sb->raid_disks);
 136         printf("  Total Devices : %d\n", sb->nr_disks);
 137         printf("Preferred Minor : %d\n", sb->md_minor);
 138         printf("\n");
 139         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 140                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 141                 if (sb->delta_disks) {
 142                         printf("  Delta Devices : %d", sb->delta_disks);
 143                         if (sb->delta_disks)
 144                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 145                         else
 146                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 147                 }
 148                 if (sb->new_level != sb->level) {
 149                         c = map_num(pers, sb->new_level);
 150                         printf("      New Level : %s\n", c?c:"-unknown-");
 151                 }
 152                 if (sb->new_layout != sb->layout) {
 153                         if (sb->level == 5) {
 154                                 c = map_num(r5layout, sb->new_layout);
 155                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 156                         }
 157                         if (sb->level == 10) {
 158                                 printf("     New Layout : near=%d, %s=%d\n",
 159                                        sb->new_layout&255,
 160                                        (sb->new_layout&0x10000)?"offset":"far",
 161                                        (sb->new_layout>>8)&255);
 162                         }
 163                 }
 164                 if (sb->new_chunk != sb->chunk_size)
 165                         printf("  New Chunksize : %d\n", sb->new_chunk);
 166                 printf("\n");
 167         }
 168         atime = sb->utime;
 169         printf("    Update Time : %.24s\n", ctime(&atime));
 170         printf("          State : %s\n",
 171                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 172         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 173                 printf("Internal Bitmap : present\n");
 174         printf(" Active Devices : %d\n", sb->active_disks);
 175         printf("Working Devices : %d\n", sb->working_disks);
 176         printf(" Failed Devices : %d\n", sb->failed_disks);
 177         printf("  Spare Devices : %d\n", sb->spare_disks);
 178         if (calc_sb0_csum(sb) == sb->sb_csum)
 179                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 180         else
 181                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 182         printf("         Events : %llu\n",
 183                ((unsigned long long)sb->events_hi << 32)
 184                + sb->events_lo);
 185         printf("\n");
 186         if (sb->level == 5) {
 187                 c = map_num(r5layout, sb->layout);
 188                 printf("         Layout : %s\n", c?c:"-unknown-");
 189         }
 190         if (sb->level == 10) {
 191                 printf("         Layout : near=%d, %s=%d\n",
 192                        sb->layout&255,
 193                        (sb->layout&0x10000)?"offset":"far",
 194                        (sb->layout>>8)&255);
 195         }
 196         switch(sb->level) {
 197         case 0:
 198         case 4:
 199         case 5:
 200         case 6:
 201         case 10:
 202                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 203                 break;
 204         case -1:
 205                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 206                 break;
 207         default: break;
 208         }
 209         printf("\n");
 210         printf("      Number   Major   Minor   RaidDevice State\n");
 211         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
 212                 mdp_disk_t *dp;
 213                 char *dv;
 214                 char nb[5];
 215                 int wonly;
 216                 if (d>=0) dp = &sb->disks[d];
 217                 else dp = &sb->this_disk;
 218                 snprintf(nb, sizeof(nb), "%4d", d);
 219                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 220                        dp->number, dp->major, dp->minor, dp->raid_disk);
 221                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 222                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 223                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 224                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 225                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 226                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 227                 if (wonly) printf(" write-mostly");
 228                 if (dp->state == 0) printf(" spare");
 229                 if ((dv=map_dev(dp->major, dp->minor, 0)))
 230                         printf("   %s", dv);
 231                 printf("\n");
 232                 if (d == -1) printf("\n");
 233         }
 234 }
 235
 236 static void brief_examine_super0(struct supertype *st)
 237 {
 238         mdp_super_t *sb = st->sb;
 239         char *c=map_num(pers, sb->level);
 240         char devname[20];
 241
 242         sprintf(devname, "/dev/md%d", sb->md_minor);
 243
 244         printf("ARRAY %s level=%s num-devices=%d UUID=",
 245                devname,
 246                c?c:"-unknown-", sb->raid_disks);
 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         printf("\n");
 253 }
 254
 255 static void export_examine_super0(struct supertype *st)
 256 {
 257         mdp_super_t *sb = st->sb;
 258
 259         printf("MD_LEVEL=%s\n", map_num(pers, sb->level));
 260         printf("MD_DEVICES=%d\n", sb->raid_disks);
 261         if (sb->minor_version >= 90)
 262                 printf("MD_UUID=%08x:%08x:%08x:%08x\n",
 263                        sb->set_uuid0, sb->set_uuid1,
 264                        sb->set_uuid2, sb->set_uuid3);
 265         else
 266                 printf("MD_UUID=%08x\n", sb->set_uuid0);
 267         printf("MD_UPDATE_TIME=%llu\n",
 268                __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL);
 269         printf("MD_EVENTS=%llu\n",
 270                ((unsigned long long)sb->events_hi << 32)
 271                + sb->events_lo);
 272 }
 273
 274 static void detail_super0(struct supertype *st, char *homehost)
 275 {
 276         mdp_super_t *sb = st->sb;
 277         printf("           UUID : ");
 278         if (sb->minor_version >= 90)
 279                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 280                        sb->set_uuid2, sb->set_uuid3);
 281         else
 282                 printf("%08x", sb->set_uuid0);
 283         if (homehost) {
 284                 char buf[20];
 285                 void *hash = sha1_buffer(homehost,
 286                                          strlen(homehost),
 287                                          buf);
 288                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 289                         printf(" (local to host %s)", homehost);
 290         }
 291         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 292 }
 293
 294 static void brief_detail_super0(struct supertype *st)
 295 {
 296         mdp_super_t *sb = st->sb;
 297         printf(" UUID=");
 298         if (sb->minor_version >= 90)
 299                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 300                        sb->set_uuid2, sb->set_uuid3);
 301         else
 302                 printf("%08x", sb->set_uuid0);
 303 }
 304 #endif
 305
 306 static int match_home0(struct supertype *st, char *homehost)
 307 {
 308         mdp_super_t *sb = st->sb;
 309         char buf[20];
 310         char *hash = sha1_buffer(homehost,
 311                                  strlen(homehost),
 312                                  buf);
 313
 314         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 315 }
 316
 317 static void uuid_from_super0(struct supertype *st, int uuid[4])
 318 {
 319         mdp_super_t *super = st->sb;
 320         uuid[0] = super->set_uuid0;
 321         if (super->minor_version >= 90) {
 322                 uuid[1] = super->set_uuid1;
 323                 uuid[2] = super->set_uuid2;
 324                 uuid[3] = super->set_uuid3;
 325         } else {
 326                 uuid[1] = 0;
 327                 uuid[2] = 0;
 328                 uuid[3] = 0;
 329         }
 330 }
 331
 332 static void getinfo_super0(struct supertype *st, struct mdinfo *info)
 333 {
 334         mdp_super_t *sb = st->sb;
 335         int working = 0;
 336         int i;
 337
 338         info->array.major_version = sb->major_version;
 339         info->array.minor_version = sb->minor_version;
 340         info->array.patch_version = sb->patch_version;
 341         info->array.raid_disks = sb->raid_disks;
 342         info->array.level = sb->level;
 343         info->array.layout = sb->layout;
 344         info->array.md_minor = sb->md_minor;
 345         info->array.ctime = sb->ctime;
 346         info->array.utime = sb->utime;
 347         info->array.chunk_size = sb->chunk_size;
 348         info->array.state = sb->state;
 349         info->component_size = sb->size*2;
 350
 351         info->disk.state = sb->this_disk.state;
 352         info->disk.major = sb->this_disk.major;
 353         info->disk.minor = sb->this_disk.minor;
 354         info->disk.raid_disk = sb->this_disk.raid_disk;
 355         info->disk.number = sb->this_disk.number;
 356
 357         info->events = md_event(sb);
 358         info->data_offset = 0;
 359
 360         sprintf(info->text_version, "0.%d", sb->minor_version);
 361         info->safe_mode_delay = 200;
 362
 363         uuid_from_super0(st, info->uuid);
 364
 365         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 366                 info->reshape_active = 1;
 367                 info->reshape_progress = sb->reshape_position;
 368                 info->new_level = sb->new_level;
 369                 info->delta_disks = sb->delta_disks;
 370                 info->new_layout = sb->new_layout;
 371                 info->new_chunk = sb->new_chunk;
 372         } else
 373                 info->reshape_active = 0;
 374
 375         sprintf(info->name, "%d", sb->md_minor);
 376         /* work_disks is calculated rather than read directly */
 377         for (i=0; i < MD_SB_DISKS; i++)
 378                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 379                     (sb->disks[i].raid_disk < info->array.raid_disks) &&
 380                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 381                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 382                         working ++;
 383         info->array.working_disks = working;
 384 }
 385
 386
 387 static int update_super0(struct supertype *st, struct mdinfo *info,
 388                          char *update,
 389                          char *devname, int verbose,
 390                          int uuid_set, char *homehost)
 391 {
 392         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 393          * For others, the return value is ignored.
 394          */
 395         int rv = 0;
 396         mdp_super_t *sb = st->sb;
 397         if (strcmp(update, "sparc2.2")==0 ) {
 398                 /* 2.2 sparc put the events in the wrong place
 399                  * So we copy the tail of the superblock
 400                  * up 4 bytes before continuing
 401                  */
 402                 __u32 *sb32 = (__u32*)sb;
 403                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 404                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 405                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 406                 if (verbose >= 0)
 407                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 408                                  devname);
 409         }
 410         if (strcmp(update, "super-minor") ==0) {
 411                 sb->md_minor = info->array.md_minor;
 412                 if (verbose > 0)
 413                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 414                                 devname, info->array.md_minor);
 415         }
 416         if (strcmp(update, "summaries") == 0) {
 417                 int i;
 418                 /* set nr_disks, active_disks, working_disks,
 419                  * failed_disks, spare_disks based on disks[]
 420                  * array in superblock.
 421                  * Also make sure extra slots aren't 'failed'
 422                  */
 423                 sb->nr_disks = sb->active_disks =
 424                         sb->working_disks = sb->failed_disks =
 425                         sb->spare_disks = 0;
 426                 for (i=0; i < MD_SB_DISKS ; i++)
 427                         if (sb->disks[i].major ||
 428                             sb->disks[i].minor) {
 429                                 int state = sb->disks[i].state;
 430                                 if (state & (1<<MD_DISK_REMOVED))
 431                                         continue;
 432                                 sb->nr_disks++;
 433                                 if (state & (1<<MD_DISK_ACTIVE))
 434                                         sb->active_disks++;
 435                                 if (state & (1<<MD_DISK_FAULTY))
 436                                         sb->failed_disks++;
 437                                 else
 438                                         sb->working_disks++;
 439                                 if (state == 0)
 440                                         sb->spare_disks++;
 441                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 442                                 sb->disks[i].state = 0;
 443         }
 444         if (strcmp(update, "force-one")==0) {
 445                 /* Not enough devices for a working array, so
 446                  * bring this one up-to-date.
 447                  */
 448                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 449                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 450                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 451                 if (sb->events_hi != ehi ||
 452                     sb->events_lo != elo)
 453                         rv = 1;
 454         }
 455         if (strcmp(update, "force-array")==0) {
 456                 /* degraded array and 'force' requested, so
 457                  * maybe need to mark it 'clean'
 458                  */
 459                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 460                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 461                         /* need to force clean */
 462                         sb->state |= (1 << MD_SB_CLEAN);
 463                         rv = 1;
 464                 }
 465         }
 466         if (strcmp(update, "assemble")==0) {
 467                 int d = info->disk.number;
 468                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 469                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 470                     != info->disk.state) {
 471                         sb->disks[d].state = info->disk.state | wonly;
 472                         rv = 1;
 473                 }
 474         }
 475         if (strcmp(update, "linear-grow-new") == 0) {
 476                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 477                 sb->disks[info->disk.number].number = info->disk.number;
 478                 sb->disks[info->disk.number].major = info->disk.major;
 479                 sb->disks[info->disk.number].minor = info->disk.minor;
 480                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 481                 sb->disks[info->disk.number].state = info->disk.state;
 482                 sb->this_disk = sb->disks[info->disk.number];
 483         }
 484         if (strcmp(update, "linear-grow-update") == 0) {
 485                 sb->raid_disks = info->array.raid_disks;
 486                 sb->nr_disks = info->array.nr_disks;
 487                 sb->active_disks = info->array.active_disks;
 488                 sb->working_disks = info->array.working_disks;
 489                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 490                 sb->disks[info->disk.number].number = info->disk.number;
 491                 sb->disks[info->disk.number].major = info->disk.major;
 492                 sb->disks[info->disk.number].minor = info->disk.minor;
 493                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 494                 sb->disks[info->disk.number].state = info->disk.state;
 495         }
 496         if (strcmp(update, "resync") == 0) {
 497                 /* make sure resync happens */
 498                 sb->state &= ~(1<<MD_SB_CLEAN);
 499                 sb->recovery_cp = 0;
 500         }
 501         if (strcmp(update, "homehost") == 0 &&
 502             homehost) {
 503                 uuid_set = 0;
 504                 update = "uuid";
 505                 info->uuid[0] = sb->set_uuid0;
 506                 info->uuid[1] = sb->set_uuid1;
 507         }
 508         if (strcmp(update, "uuid") == 0) {
 509                 if (!uuid_set && homehost) {
 510                         char buf[20];
 511                         char *hash = sha1_buffer(homehost,
 512                                                  strlen(homehost),
 513                                                  buf);
 514                         memcpy(info->uuid+2, hash, 8);
 515                 }
 516                 sb->set_uuid0 = info->uuid[0];
 517                 sb->set_uuid1 = info->uuid[1];
 518                 sb->set_uuid2 = info->uuid[2];
 519                 sb->set_uuid3 = info->uuid[3];
 520                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 521                         struct bitmap_super_s *bm;
 522                         bm = (struct bitmap_super_s*)(sb+1);
 523                         uuid_from_super0(st, (int*)bm->uuid);
 524                 }
 525         }
 526         if (strcmp(update, "_reshape_progress")==0)
 527                 sb->reshape_position = info->reshape_progress;
 528
 529         sb->sb_csum = calc_sb0_csum(sb);
 530         return rv;
 531 }
 532
 533 /*
 534  * For verion-0 superblock, the homehost is 'stored' in the
 535  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 536  * of random material.
 537  * We use the first 8 bytes (64bits) of the sha1 of the
 538  * host name
 539  */
 540
 541
 542 static int init_super0(struct supertype *st, mdu_array_info_t *info,
 543                        unsigned long long size, char *ignored_name, char *homehost,
 544                        int *uuid)
 545 {
 546         mdp_super_t *sb;
 547         int spares;
 548
 549         posix_memalign((void**)&sb, 512, MD_SB_BYTES + sizeof(bitmap_super_t));
 550         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 551
 552         st->sb = sb;
 553         if (info == NULL) {
 554                 /* zeroing the superblock */
 555                 return 0;
 556         }
 557
 558         spares = info->working_disks - info->active_disks;
 559         if (info->raid_disks + spares  > MD_SB_DISKS) {
 560                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 561                         info->raid_disks , spares, MD_SB_DISKS);
 562                 return 0;
 563         }
 564
 565         sb->md_magic = MD_SB_MAGIC;
 566         sb->major_version = 0;
 567         sb->minor_version = 90;
 568         sb->patch_version = 0;
 569         sb->gvalid_words = 0; /* ignored */
 570         sb->ctime = time(0);
 571         sb->level = info->level;
 572         if (size != info->size)
 573                 return 0;
 574         sb->size = info->size;
 575         sb->nr_disks = info->nr_disks;
 576         sb->raid_disks = info->raid_disks;
 577         sb->md_minor = info->md_minor;
 578         sb->not_persistent = 0;
 579         if (uuid) {
 580                 sb->set_uuid0 = uuid[0];
 581                 sb->set_uuid1 = uuid[1];
 582                 sb->set_uuid2 = uuid[2];
 583                 sb->set_uuid3 = uuid[3];
 584         } else {
 585                 int rfd = open("/dev/urandom", O_RDONLY);
 586                 if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 587                         sb->set_uuid0 = random();
 588                 if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 589                         sb->set_uuid1 = random();
 590                         sb->set_uuid2 = random();
 591                         sb->set_uuid3 = random();
 592                 }
 593                 if (rfd >= 0)
 594                         close(rfd);
 595         }
 596         if (homehost) {
 597                 char buf[20];
 598                 char *hash = sha1_buffer(homehost,
 599                                          strlen(homehost),
 600                                          buf);
 601                 memcpy(&sb->set_uuid2, hash, 8);
 602         }
 603
 604         sb->utime = sb->ctime;
 605         sb->state = info->state;
 606         sb->active_disks = info->active_disks;
 607         sb->working_disks = info->working_disks;
 608         sb->failed_disks = info->failed_disks;
 609         sb->spare_disks = info->spare_disks;
 610         sb->events_hi = 0;
 611         sb->events_lo = 1;
 612
 613         sb->layout = info->layout;
 614         sb->chunk_size = info->chunk_size;
 615
 616         return 1;
 617 }
 618
 619 struct devinfo {
 620         int fd;
 621         char *devname;
 622         mdu_disk_info_t disk;
 623         struct devinfo *next;
 624 };
 625
 626 #ifndef MDASSEMBLE
 627 /* Add a device to the superblock being created */
 628 static void add_to_super0(struct supertype *st, mdu_disk_info_t *dinfo,
 629                           int fd, char *devname)
 630 {
 631         mdp_super_t *sb = st->sb;
 632         mdp_disk_t *dk = &sb->disks[dinfo->number];
 633         struct devinfo *di, **dip;
 634
 635         dk->number = dinfo->number;
 636         dk->major = dinfo->major;
 637         dk->minor = dinfo->minor;
 638         dk->raid_disk = dinfo->raid_disk;
 639         dk->state = dinfo->state;
 640
 641         sb->this_disk = sb->disks[dinfo->number];
 642         sb->sb_csum = calc_sb0_csum(sb);
 643
 644         dip = (struct devinfo **)&st->info;
 645         while (*dip)
 646                 dip = &(*dip)->next;
 647         di = malloc(sizeof(struct devinfo));
 648         di->fd = fd;
 649         di->devname = devname;
 650         di->disk = *dinfo;
 651         di->next = NULL;
 652         *dip = di;
 653 }
 654 #endif
 655
 656 static int store_super0(struct supertype *st, int fd)
 657 {
 658         unsigned long long dsize;
 659         unsigned long long offset;
 660         mdp_super_t *super = st->sb;
 661
 662         if (!get_dev_size(fd, NULL, &dsize))
 663                 return 1;
 664
 665         if (dsize < MD_RESERVED_SECTORS*512)
 666                 return 2;
 667
 668         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 669
 670         offset *= 512;
 671
 672         if (lseek64(fd, offset, 0)< 0LL)
 673                 return 3;
 674
 675         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 676                 return 4;
 677
 678         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 679                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 680                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 681                         if (write(fd, bm, ROUND_UP(sizeof(*bm),512)) !=
 682                             ROUND_UP(sizeof(*bm),512))
 683                             return 5;
 684         }
 685
 686         fsync(fd);
 687         return 0;
 688 }
 689
 690 #ifndef MDASSEMBLE
 691 static int write_init_super0(struct supertype *st)
 692 {
 693         mdp_super_t *sb = st->sb;
 694         int rv = 0;
 695         struct devinfo *di;
 696
 697         for (di = st->info ; di && ! rv ; di = di->next) {
 698
 699                 if (di->disk.state == 1)
 700                         continue;
 701                 if (di->fd == -1)
 702                         continue;
 703                 Kill(di->devname, 0, 1, 1);
 704                 Kill(di->devname, 0, 1, 1);
 705
 706                 sb->disks[di->disk.number].state &= ~(1<<MD_DISK_FAULTY);
 707
 708                 sb->this_disk = sb->disks[di->disk.number];
 709                 sb->sb_csum = calc_sb0_csum(sb);
 710                 rv = store_super0(st, di->fd);
 711
 712                 if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 713                         rv = st->ss->write_bitmap(st, di->fd);
 714
 715                 if (rv)
 716                         fprintf(stderr,
 717                                 Name ": failed to write superblock to %s\n",
 718                                 di->devname);
 719                 close(di->fd);
 720                 di->fd = -1;
 721         }
 722         return rv;
 723 }
 724 #endif
 725
 726 static int compare_super0(struct supertype *st, struct supertype *tst)
 727 {
 728         /*
 729          * return:
 730          *  0 same, or first was empty, and second was copied
 731          *  1 second had wrong number
 732          *  2 wrong uuid
 733          *  3 wrong other info
 734          */
 735         mdp_super_t *first = st->sb;
 736         mdp_super_t *second = tst->sb;
 737         int uuid1[4], uuid2[4];
 738
 739         if (second->md_magic != MD_SB_MAGIC)
 740                 return 1;
 741         if (!first) {
 742                 posix_memalign((void**)&first, 512,
 743                                MD_SB_BYTES + sizeof(struct bitmap_super_s));
 744                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 745                 st->sb = first;
 746                 return 0;
 747         }
 748
 749         uuid_from_super0(st, uuid1);
 750         uuid_from_super0(tst, uuid2);
 751         if (!same_uuid(uuid1, uuid2, 0))
 752                 return 2;
 753         if (first->major_version != second->major_version ||
 754             first->minor_version != second->minor_version ||
 755             first->patch_version != second->patch_version ||
 756             first->gvalid_words  != second->gvalid_words  ||
 757             first->ctime         != second->ctime         ||
 758             first->level         != second->level         ||
 759             first->size          != second->size          ||
 760             first->raid_disks    != second->raid_disks    )
 761                 return 3;
 762
 763         return 0;
 764 }
 765
 766
 767 static void free_super0(struct supertype *st);
 768
 769 static int load_super0(struct supertype *st, int fd, char *devname)
 770 {
 771         /* try to read in the superblock
 772          * Return:
 773          *  0 on success
 774          *  1 on cannot get superblock
 775          *  2 on superblock meaningless
 776          */
 777         unsigned long long dsize;
 778         unsigned long long offset;
 779         mdp_super_t *super;
 780         int uuid[4];
 781         struct bitmap_super_s *bsb;
 782
 783         free_super0(st);
 784
 785         if (st->subarray[0])
 786                 return 1;
 787
 788         if (!get_dev_size(fd, devname, &dsize))
 789                 return 1;
 790
 791         if (dsize < MD_RESERVED_SECTORS*512) {
 792                 if (devname)
 793                         fprintf(stderr, Name
 794                             ": %s is too small for md: size is %llu sectors.\n",
 795                                 devname, dsize);
 796                 return 1;
 797         }
 798
 799         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 800
 801         offset *= 512;
 802
 803         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 804
 805         if (lseek64(fd, offset, 0)< 0LL) {
 806                 if (devname)
 807                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 808                                 devname, strerror(errno));
 809                 return 1;
 810         }
 811
 812         posix_memalign((void**)&super, 512, MD_SB_BYTES + sizeof(bitmap_super_t)+512);
 813
 814         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 815                 if (devname)
 816                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 817                                 devname);
 818                 free(super);
 819                 return 1;
 820         }
 821
 822         if (st->ss && st->minor_version == 9)
 823                 super0_swap_endian(super);
 824
 825         if (super->md_magic != MD_SB_MAGIC) {
 826                 if (devname)
 827                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 828                                 devname, MD_SB_MAGIC, super->md_magic);
 829                 free(super);
 830                 return 2;
 831         }
 832
 833         if (super->major_version != 0) {
 834                 if (devname)
 835                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 836                                 devname, super->major_version);
 837                 free(super);
 838                 return 2;
 839         }
 840         st->sb = super;
 841
 842         if (st->ss == NULL) {
 843                 st->ss = &super0;
 844                 st->minor_version = super->minor_version;
 845                 st->max_devs = MD_SB_DISKS;
 846                 st->info = NULL;
 847         }
 848
 849         /* Now check on the bitmap superblock */
 850         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 851                 return 0;
 852         /* Read the bitmap superblock and make sure it looks
 853          * valid.  If it doesn't clear the bit.  An --assemble --force
 854          * should get that written out.
 855          */
 856         if (read(fd, super+1, ROUND_UP(sizeof(struct bitmap_super_s),512))
 857             != ROUND_UP(sizeof(struct bitmap_super_s),512))
 858                 goto no_bitmap;
 859
 860         uuid_from_super0(st, uuid);
 861         bsb = (struct bitmap_super_s *)(super+1);
 862         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 863             memcmp(bsb->uuid, uuid, 16) != 0)
 864                 goto no_bitmap;
 865         return 0;
 866
 867  no_bitmap:
 868         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 869
 870         return 0;
 871 }
 872
 873 static struct supertype *match_metadata_desc0(char *arg)
 874 {
 875         struct supertype *st = malloc(sizeof(*st));
 876         if (!st) return st;
 877
 878         memset(st, 0, sizeof(*st));
 879         st->ss = &super0;
 880         st->info = NULL;
 881         st->minor_version = 90;
 882         st->max_devs = MD_SB_DISKS;
 883         st->sb = NULL;
 884         /* we sometimes get 00.90 */
 885         while (arg[0] == '0' && arg[1] == '0')
 886                 arg++;
 887         if (strcmp(arg, "0") == 0 ||
 888             strcmp(arg, "0.90") == 0 ||
 889             strcmp(arg, "default") == 0 ||
 890             strcmp(arg, "") == 0 /* no metadata */
 891                 )
 892                 return st;
 893
 894         st->minor_version = 91; /* reshape in progress */
 895         if (strcmp(arg, "0.91") == 0) /* For dup_super support */
 896                 return st;
 897
 898         st->minor_version = 9; /* flag for 'byte-swapped' */
 899         if (strcmp(arg, "0.swap")==0 ||
 900             strcmp(arg, "0.9") == 0) /* For dup_super support */
 901                 return st;
 902
 903         free(st);
 904         return NULL;
 905 }
 906
 907 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 908 {
 909         if (devsize < MD_RESERVED_SECTORS)
 910                 return 0ULL;
 911         return MD_NEW_SIZE_SECTORS(devsize);
 912 }
 913
 914 static int add_internal_bitmap0(struct supertype *st, int *chunkp,
 915                                 int delay, int write_behind,
 916                                 unsigned long long size, int may_change,
 917                                 int major)
 918 {
 919         /*
 920          * The bitmap comes immediately after the superblock and must be 60K in size
 921          * at most.  The default size is between 30K and 60K
 922          *
 923          * size is in sectors,  chunk is in bytes !!!
 924          */
 925         unsigned long long bits;
 926         unsigned long long max_bits = 60*1024*8;
 927         unsigned long long min_chunk;
 928         int chunk = *chunkp;
 929         mdp_super_t *sb = st->sb;
 930         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 931
 932
 933         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 934         bits = (size * 512) / min_chunk + 1;
 935         while (bits > max_bits) {
 936                 min_chunk *= 2;
 937                 bits = (bits+1)/2;
 938         }
 939         if (chunk == UnSet)
 940                 chunk = min_chunk;
 941         else if (chunk < min_chunk)
 942                 return 0; /* chunk size too small */
 943
 944         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 945
 946         memset(bms, 0, sizeof(*bms));
 947         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 948         bms->version = __cpu_to_le32(major);
 949         uuid_from_super0(st, (int*)bms->uuid);
 950         bms->chunksize = __cpu_to_le32(chunk);
 951         bms->daemon_sleep = __cpu_to_le32(delay);
 952         bms->sync_size = __cpu_to_le64(size);
 953         bms->write_behind = __cpu_to_le32(write_behind);
 954         *chunkp = chunk;
 955         return 1;
 956 }
 957
 958
 959 static void locate_bitmap0(struct supertype *st, int fd)
 960 {
 961         unsigned long long dsize;
 962         unsigned long long offset;
 963
 964         if (!get_dev_size(fd, NULL, &dsize))
 965                 return;
 966
 967         if (dsize < MD_RESERVED_SECTORS*512)
 968                 return;
 969
 970         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 971
 972         offset *= 512;
 973
 974         offset += MD_SB_BYTES;
 975
 976         lseek64(fd, offset, 0);
 977 }
 978
 979 static int write_bitmap0(struct supertype *st, int fd)
 980 {
 981         unsigned long long dsize;
 982         unsigned long long offset;
 983         mdp_super_t *sb = st->sb;
 984
 985         int rv = 0;
 986
 987         int towrite, n;
 988         char abuf[4096+512];
 989         char *buf = (char*)(((long)(abuf+512))&~511UL);
 990
 991         if (!get_dev_size(fd, NULL, &dsize))
 992                 return 1;
 993
 994
 995         if (dsize < MD_RESERVED_SECTORS*512)
 996                 return -1;
 997
 998         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 999
1000         offset *= 512;
1001
1002         if (lseek64(fd, offset + 4096, 0)< 0LL)
1003                 return 3;
1004
1005         memset(buf, 0xff, 4096);
1006         memcpy(buf,  ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t));
1007         towrite = 64*1024;
1008         while (towrite > 0) {
1009                 n = towrite;
1010                 if (n > 4096)
1011                         n = 4096;
1012                 n = write(fd, buf, n);
1013                 if (n > 0)
1014                         towrite -= n;
1015                 else
1016                         break;
1017                 memset(buf, 0xff, 4096);
1018         }
1019         fsync(fd);
1020         if (towrite)
1021                 rv = -2;
1022
1023         return rv;
1024 }
1025
1026 static void free_super0(struct supertype *st)
1027 {
1028         if (st->sb)
1029                 free(st->sb);
1030         st->sb = NULL;
1031 }
1032
1033 #ifndef MDASSEMBLE
1034 static int validate_geometry0(struct supertype *st, int level,
1035                               int layout, int raiddisks,
1036                               int chunk, unsigned long long size,
1037                               char *subdev, unsigned long long *freesize,
1038                               int verbose)
1039 {
1040         unsigned long long ldsize;
1041         int fd;
1042
1043         if (level == LEVEL_CONTAINER)
1044                 return 0;
1045         if (raiddisks > MD_SB_DISKS)
1046                 return 0;
1047         if (size > (0x7fffffffULL<<10))
1048                 return 0;
1049         if (!subdev)
1050                 return 1;
1051
1052         fd = open(subdev, O_RDONLY|O_EXCL, 0);
1053         if (fd < 0) {
1054                 if (verbose)
1055                         fprintf(stderr, Name ": super0.90 cannot open %s: %s\n",
1056                                 subdev, strerror(errno));
1057                 return 0;
1058         }
1059
1060         if (!get_dev_size(fd, subdev, &ldsize)) {
1061                 close(fd);
1062                 return 0;
1063         }
1064         close(fd);
1065
1066         if (ldsize < MD_RESERVED_SECTORS * 512)
1067                 return 0;
1068         if (size > (0x7fffffffULL<<10))
1069                 return 0;
1070         *freesize = MD_NEW_SIZE_SECTORS(ldsize >> 9);
1071         return 1;
1072 }
1073 #endif /* MDASSEMBLE */
1074
1075 struct superswitch super0 = {
1076 #ifndef MDASSEMBLE
1077         .examine_super = examine_super0,
1078         .brief_examine_super = brief_examine_super0,
1079         .export_examine_super = export_examine_super0,
1080         .detail_super = detail_super0,
1081         .brief_detail_super = brief_detail_super0,
1082         .write_init_super = write_init_super0,
1083         .validate_geometry = validate_geometry0,
1084         .add_to_super = add_to_super0,
1085 #endif
1086         .match_home = match_home0,
1087         .uuid_from_super = uuid_from_super0,
1088         .getinfo_super = getinfo_super0,
1089         .update_super = update_super0,
1090         .init_super = init_super0,
1091         .store_super = store_super0,
1092         .compare_super = compare_super0,
1093         .load_super = load_super0,
1094         .match_metadata_desc = match_metadata_desc0,
1095         .avail_size = avail_size0,
1096         .add_internal_bitmap = add_internal_bitmap0,
1097         .locate_bitmap = locate_bitmap0,
1098         .write_bitmap = write_bitmap0,
1099         .free_super = free_super0,
1100 };