super0.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2009 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@suse.de>
  23  */
  24
  25 #define HAVE_STDINT_H 1
  26 #include "mdadm.h"
  27 #include "sha1.h"
  28 /*
  29  * All handling for the 0.90.0 version superblock is in
  30  * this file.
  31  * This includes:
  32  *   - finding, loading, and writing the superblock.
  33  *   - initialising a new superblock
  34  *   - printing the superblock for --examine
  35  *   - printing part of the superblock for --detail
  36  * .. other stuff
  37  */
  38
  39
  40 static unsigned long calc_sb0_csum(mdp_super_t *super)
  41 {
  42         unsigned long csum = super->sb_csum;
  43         unsigned long newcsum;
  44         super->sb_csum= 0 ;
  45         newcsum = calc_csum(super, MD_SB_BYTES);
  46         super->sb_csum = csum;
  47         return newcsum;
  48 }
  49
  50
  51 static void super0_swap_endian(struct mdp_superblock_s *sb)
  52 {
  53         /* as super0 superblocks are host-endian, it is sometimes
  54          * useful to be able to swap the endianness
  55          * as (almost) everything is u32's we byte-swap every 4byte
  56          * number.
  57          * We then also have to swap the events_hi and events_lo
  58          */
  59         char *sbc = (char *)sb;
  60         __u32 t32;
  61         int i;
  62
  63         for (i=0; i < MD_SB_BYTES ; i+=4) {
  64                 char t = sbc[i];
  65                 sbc[i] = sbc[i+3];
  66                 sbc[i+3] = t;
  67                 t=sbc[i+1];
  68                 sbc[i+1]=sbc[i+2];
  69                 sbc[i+2]=t;
  70         }
  71         t32 = sb->events_hi;
  72         sb->events_hi = sb->events_lo;
  73         sb->events_lo = t32;
  74
  75         t32 = sb->cp_events_hi;
  76         sb->cp_events_hi = sb->cp_events_lo;
  77         sb->cp_events_lo = t32;
  78
  79 }
  80
  81 #ifndef MDASSEMBLE
  82
  83 static void examine_super0(struct supertype *st, char *homehost)
  84 {
  85         mdp_super_t *sb = st->sb;
  86         time_t atime;
  87         int d;
  88         char *c;
  89
  90         printf("          Magic : %08x\n", sb->md_magic);
  91         printf("        Version : %d.%02d.%02d\n", sb->major_version, sb->minor_version,
  92                sb->patch_version);
  93         if (sb->minor_version >= 90) {
  94                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
  95                        sb->set_uuid2, sb->set_uuid3);
  96                 if (homehost) {
  97                         char buf[20];
  98                         void *hash = sha1_buffer(homehost,
  99                                                  strlen(homehost),
 100                                                  buf);
 101                         if (memcmp(&sb->set_uuid2, hash, 8)==0)
 102                                 printf(" (local to host %s)", homehost);
 103                 }
 104                 printf("\n");
 105         } else
 106                 printf("           UUID : %08x\n", sb->set_uuid0);
 107
 108         if (sb->not_persistent)
 109                 printf("           Eedk : not persistent\n");
 110
 111         atime = sb->ctime;
 112         printf("  Creation Time : %.24s\n", ctime(&atime));
 113         c=map_num(pers, sb->level);
 114         printf("     Raid Level : %s\n", c?c:"-unknown-");
 115         if ((int)sb->level >= 0) {
 116                 int ddsks=0;
 117                 printf("  Used Dev Size : %d%s\n", sb->size,
 118                        human_size((long long)sb->size<<10));
 119                 switch(sb->level) {
 120                 case 1: ddsks=1;break;
 121                 case 4:
 122                 case 5: ddsks = sb->raid_disks-1; break;
 123                 case 6: ddsks = sb->raid_disks-2; break;
 124                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 125                 }
 126                 if (ddsks)
 127                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 128                                human_size(ddsks*(long long)sb->size<<10));
 129         }
 130         printf("   Raid Devices : %d\n", sb->raid_disks);
 131         printf("  Total Devices : %d\n", sb->nr_disks);
 132         printf("Preferred Minor : %d\n", sb->md_minor);
 133         printf("\n");
 134         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 135                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 136                 if (sb->delta_disks) {
 137                         printf("  Delta Devices : %d", sb->delta_disks);
 138                         if (sb->delta_disks > 0)
 139                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 140                         else
 141                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 142                 }
 143                 if (sb->new_level != sb->level) {
 144                         c = map_num(pers, sb->new_level);
 145                         printf("      New Level : %s\n", c?c:"-unknown-");
 146                 }
 147                 if (sb->new_layout != sb->layout) {
 148                         if (sb->level == 5) {
 149                                 c = map_num(r5layout, sb->new_layout);
 150                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 151                         }
 152                         if (sb->level == 6) {
 153                                 c = map_num(r6layout, sb->new_layout);
 154                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 155                         }
 156                         if (sb->level == 10) {
 157                                 printf("     New Layout : near=%d, %s=%d\n",
 158                                        sb->new_layout&255,
 159                                        (sb->new_layout&0x10000)?"offset":"far",
 160                                        (sb->new_layout>>8)&255);
 161                         }
 162                 }
 163                 if (sb->new_chunk != sb->chunk_size)
 164                         printf("  New Chunksize : %d\n", sb->new_chunk);
 165                 printf("\n");
 166         }
 167         atime = sb->utime;
 168         printf("    Update Time : %.24s\n", ctime(&atime));
 169         printf("          State : %s\n",
 170                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 171         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 172                 printf("Internal Bitmap : present\n");
 173         printf(" Active Devices : %d\n", sb->active_disks);
 174         printf("Working Devices : %d\n", sb->working_disks);
 175         printf(" Failed Devices : %d\n", sb->failed_disks);
 176         printf("  Spare Devices : %d\n", sb->spare_disks);
 177         if (calc_sb0_csum(sb) == sb->sb_csum)
 178                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 179         else
 180                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 181         printf("         Events : %llu\n",
 182                ((unsigned long long)sb->events_hi << 32)
 183                + sb->events_lo);
 184         printf("\n");
 185         if (sb->level == 5) {
 186                 c = map_num(r5layout, sb->layout);
 187                 printf("         Layout : %s\n", c?c:"-unknown-");
 188         }
 189         if (sb->level == 6) {
 190                 c = map_num(r6layout, sb->layout);
 191                 printf("         Layout : %s\n", c?c:"-unknown-");
 192         }
 193         if (sb->level == 10) {
 194                 printf("         Layout :");
 195                 print_r10_layout(sb->layout);
 196                 printf("\n");
 197         }
 198         switch(sb->level) {
 199         case 0:
 200         case 4:
 201         case 5:
 202         case 6:
 203         case 10:
 204                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 205                 break;
 206         case -1:
 207                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 208                 break;
 209         default: break;
 210         }
 211         printf("\n");
 212         printf("      Number   Major   Minor   RaidDevice State\n");
 213         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
 214                 mdp_disk_t *dp;
 215                 char *dv;
 216                 char nb[5];
 217                 int wonly;
 218                 if (d>=0) dp = &sb->disks[d];
 219                 else dp = &sb->this_disk;
 220                 snprintf(nb, sizeof(nb), "%4d", d);
 221                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 222                        dp->number, dp->major, dp->minor, dp->raid_disk);
 223                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 224                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 225                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 226                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 227                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 228                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 229                 if (wonly) printf(" write-mostly");
 230                 if (dp->state == 0) printf(" spare");
 231                 if ((dv=map_dev(dp->major, dp->minor, 0)))
 232                         printf("   %s", dv);
 233                 printf("\n");
 234                 if (d == -1) printf("\n");
 235         }
 236 }
 237
 238 static void brief_examine_super0(struct supertype *st, int verbose)
 239 {
 240         mdp_super_t *sb = st->sb;
 241         char *c=map_num(pers, sb->level);
 242         char devname[20];
 243
 244         sprintf(devname, "/dev/md%d", sb->md_minor);
 245
 246         if (verbose) {
 247                 printf("ARRAY %s level=%s num-devices=%d",
 248                        devname,
 249                        c?c:"-unknown-", sb->raid_disks);
 250         } else
 251                 printf("ARRAY %s", devname);
 252
 253         if (sb->minor_version >= 90)
 254                 printf(" UUID=%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 255                        sb->set_uuid2, sb->set_uuid3);
 256         else
 257                 printf(" UUID=%08x", sb->set_uuid0);
 258         printf("\n");
 259 }
 260
 261 static void export_examine_super0(struct supertype *st)
 262 {
 263         mdp_super_t *sb = st->sb;
 264
 265         printf("MD_LEVEL=%s\n", map_num(pers, sb->level));
 266         printf("MD_DEVICES=%d\n", sb->raid_disks);
 267         if (sb->minor_version >= 90)
 268                 printf("MD_UUID=%08x:%08x:%08x:%08x\n",
 269                        sb->set_uuid0, sb->set_uuid1,
 270                        sb->set_uuid2, sb->set_uuid3);
 271         else
 272                 printf("MD_UUID=%08x\n", sb->set_uuid0);
 273         printf("MD_UPDATE_TIME=%llu\n",
 274                __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL);
 275         printf("MD_EVENTS=%llu\n",
 276                ((unsigned long long)sb->events_hi << 32)
 277                + sb->events_lo);
 278 }
 279
 280 static void detail_super0(struct supertype *st, char *homehost)
 281 {
 282         mdp_super_t *sb = st->sb;
 283         printf("           UUID : ");
 284         if (sb->minor_version >= 90)
 285                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 286                        sb->set_uuid2, sb->set_uuid3);
 287         else
 288                 printf("%08x", sb->set_uuid0);
 289         if (homehost) {
 290                 char buf[20];
 291                 void *hash = sha1_buffer(homehost,
 292                                          strlen(homehost),
 293                                          buf);
 294                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 295                         printf(" (local to host %s)", homehost);
 296         }
 297         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 298 }
 299
 300 static void brief_detail_super0(struct supertype *st)
 301 {
 302         mdp_super_t *sb = st->sb;
 303         printf(" UUID=");
 304         if (sb->minor_version >= 90)
 305                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 306                        sb->set_uuid2, sb->set_uuid3);
 307         else
 308                 printf("%08x", sb->set_uuid0);
 309 }
 310 #endif
 311
 312 static int match_home0(struct supertype *st, char *homehost)
 313 {
 314         mdp_super_t *sb = st->sb;
 315         char buf[20];
 316         char *hash;
 317
 318         if (!homehost)
 319                 return 0;
 320         hash = sha1_buffer(homehost,
 321                            strlen(homehost),
 322                            buf);
 323
 324         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 325 }
 326
 327 static void uuid_from_super0(struct supertype *st, int uuid[4])
 328 {
 329         mdp_super_t *super = st->sb;
 330         uuid[0] = super->set_uuid0;
 331         if (super->minor_version >= 90) {
 332                 uuid[1] = super->set_uuid1;
 333                 uuid[2] = super->set_uuid2;
 334                 uuid[3] = super->set_uuid3;
 335         } else {
 336                 uuid[1] = 0;
 337                 uuid[2] = 0;
 338                 uuid[3] = 0;
 339         }
 340 }
 341
 342 static void getinfo_super0(struct supertype *st, struct mdinfo *info)
 343 {
 344         mdp_super_t *sb = st->sb;
 345         int working = 0;
 346         int i;
 347
 348         info->array.major_version = sb->major_version;
 349         info->array.minor_version = sb->minor_version;
 350         info->array.patch_version = sb->patch_version;
 351         info->array.raid_disks = sb->raid_disks;
 352         info->array.level = sb->level;
 353         info->array.layout = sb->layout;
 354         info->array.md_minor = sb->md_minor;
 355         info->array.ctime = sb->ctime;
 356         info->array.utime = sb->utime;
 357         info->array.chunk_size = sb->chunk_size;
 358         info->array.state = sb->state;
 359         info->component_size = sb->size*2;
 360
 361         info->disk.state = sb->this_disk.state;
 362         info->disk.major = sb->this_disk.major;
 363         info->disk.minor = sb->this_disk.minor;
 364         info->disk.raid_disk = sb->this_disk.raid_disk;
 365         info->disk.number = sb->this_disk.number;
 366
 367         info->events = md_event(sb);
 368         info->data_offset = 0;
 369
 370         sprintf(info->text_version, "0.%d", sb->minor_version);
 371         info->safe_mode_delay = 200;
 372
 373         uuid_from_super0(st, info->uuid);
 374
 375         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 376                 info->reshape_active = 1;
 377                 info->reshape_progress = sb->reshape_position;
 378                 info->new_level = sb->new_level;
 379                 info->delta_disks = sb->delta_disks;
 380                 info->new_layout = sb->new_layout;
 381                 info->new_chunk = sb->new_chunk;
 382         } else
 383                 info->reshape_active = 0;
 384
 385         sprintf(info->name, "%d", sb->md_minor);
 386         /* work_disks is calculated rather than read directly */
 387         for (i=0; i < MD_SB_DISKS; i++)
 388                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 389                     (sb->disks[i].raid_disk < info->array.raid_disks) &&
 390                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 391                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 392                         working ++;
 393         info->array.working_disks = working;
 394 }
 395
 396
 397 static int update_super0(struct supertype *st, struct mdinfo *info,
 398                          char *update,
 399                          char *devname, int verbose,
 400                          int uuid_set, char *homehost)
 401 {
 402         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 403          * For others, the return value is ignored.
 404          */
 405         int rv = 0;
 406         mdp_super_t *sb = st->sb;
 407         if (strcmp(update, "sparc2.2")==0 ) {
 408                 /* 2.2 sparc put the events in the wrong place
 409                  * So we copy the tail of the superblock
 410                  * up 4 bytes before continuing
 411                  */
 412                 __u32 *sb32 = (__u32*)sb;
 413                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 414                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 415                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 416                 if (verbose >= 0)
 417                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 418                                  devname);
 419         }
 420         if (strcmp(update, "super-minor") ==0) {
 421                 sb->md_minor = info->array.md_minor;
 422                 if (verbose > 0)
 423                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 424                                 devname, info->array.md_minor);
 425         }
 426         if (strcmp(update, "summaries") == 0) {
 427                 int i;
 428                 /* set nr_disks, active_disks, working_disks,
 429                  * failed_disks, spare_disks based on disks[]
 430                  * array in superblock.
 431                  * Also make sure extra slots aren't 'failed'
 432                  */
 433                 sb->nr_disks = sb->active_disks =
 434                         sb->working_disks = sb->failed_disks =
 435                         sb->spare_disks = 0;
 436                 for (i=0; i < MD_SB_DISKS ; i++)
 437                         if (sb->disks[i].major ||
 438                             sb->disks[i].minor) {
 439                                 int state = sb->disks[i].state;
 440                                 if (state & (1<<MD_DISK_REMOVED))
 441                                         continue;
 442                                 sb->nr_disks++;
 443                                 if (state & (1<<MD_DISK_ACTIVE))
 444                                         sb->active_disks++;
 445                                 if (state & (1<<MD_DISK_FAULTY))
 446                                         sb->failed_disks++;
 447                                 else
 448                                         sb->working_disks++;
 449                                 if (state == 0)
 450                                         sb->spare_disks++;
 451                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 452                                 sb->disks[i].state = 0;
 453         }
 454         if (strcmp(update, "force-one")==0) {
 455                 /* Not enough devices for a working array, so
 456                  * bring this one up-to-date.
 457                  */
 458                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 459                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 460                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 461                 if (sb->events_hi != ehi ||
 462                     sb->events_lo != elo)
 463                         rv = 1;
 464         }
 465         if (strcmp(update, "force-array")==0) {
 466                 /* degraded array and 'force' requested, so
 467                  * maybe need to mark it 'clean'
 468                  */
 469                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 470                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 471                         /* need to force clean */
 472                         sb->state |= (1 << MD_SB_CLEAN);
 473                         rv = 1;
 474                 }
 475         }
 476         if (strcmp(update, "assemble")==0) {
 477                 int d = info->disk.number;
 478                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 479                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 480                     != info->disk.state) {
 481                         sb->disks[d].state = info->disk.state | wonly;
 482                         rv = 1;
 483                 }
 484         }
 485         if (strcmp(update, "linear-grow-new") == 0) {
 486                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 487                 sb->disks[info->disk.number].number = info->disk.number;
 488                 sb->disks[info->disk.number].major = info->disk.major;
 489                 sb->disks[info->disk.number].minor = info->disk.minor;
 490                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 491                 sb->disks[info->disk.number].state = info->disk.state;
 492                 sb->this_disk = sb->disks[info->disk.number];
 493         }
 494         if (strcmp(update, "linear-grow-update") == 0) {
 495                 sb->raid_disks = info->array.raid_disks;
 496                 sb->nr_disks = info->array.nr_disks;
 497                 sb->active_disks = info->array.active_disks;
 498                 sb->working_disks = info->array.working_disks;
 499                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 500                 sb->disks[info->disk.number].number = info->disk.number;
 501                 sb->disks[info->disk.number].major = info->disk.major;
 502                 sb->disks[info->disk.number].minor = info->disk.minor;
 503                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 504                 sb->disks[info->disk.number].state = info->disk.state;
 505         }
 506         if (strcmp(update, "resync") == 0) {
 507                 /* make sure resync happens */
 508                 sb->state &= ~(1<<MD_SB_CLEAN);
 509                 sb->recovery_cp = 0;
 510         }
 511         if (strcmp(update, "homehost") == 0 &&
 512             homehost) {
 513                 uuid_set = 0;
 514                 update = "uuid";
 515                 info->uuid[0] = sb->set_uuid0;
 516                 info->uuid[1] = sb->set_uuid1;
 517         }
 518         if (strcmp(update, "uuid") == 0) {
 519                 if (!uuid_set && homehost) {
 520                         char buf[20];
 521                         char *hash = sha1_buffer(homehost,
 522                                                  strlen(homehost),
 523                                                  buf);
 524                         memcpy(info->uuid+2, hash, 8);
 525                 }
 526                 sb->set_uuid0 = info->uuid[0];
 527                 sb->set_uuid1 = info->uuid[1];
 528                 sb->set_uuid2 = info->uuid[2];
 529                 sb->set_uuid3 = info->uuid[3];
 530                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 531                         struct bitmap_super_s *bm;
 532                         bm = (struct bitmap_super_s*)(sb+1);
 533                         uuid_from_super0(st, (int*)bm->uuid);
 534                 }
 535         }
 536         if (strcmp(update, "_reshape_progress")==0)
 537                 sb->reshape_position = info->reshape_progress;
 538
 539         sb->sb_csum = calc_sb0_csum(sb);
 540         return rv;
 541 }
 542
 543 /*
 544  * For verion-0 superblock, the homehost is 'stored' in the
 545  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 546  * of random material.
 547  * We use the first 8 bytes (64bits) of the sha1 of the
 548  * host name
 549  */
 550
 551
 552 static int init_super0(struct supertype *st, mdu_array_info_t *info,
 553                        unsigned long long size, char *ignored_name, char *homehost,
 554                        int *uuid)
 555 {
 556         mdp_super_t *sb;
 557         int spares;
 558
 559         if (posix_memalign((void**)&sb, 4096,
 560                            MD_SB_BYTES + ROUND_UP(sizeof(bitmap_super_t), 4096)) != 0) {
 561                 fprintf(stderr, Name ": %s could not allocate superblock\n", __func__);
 562                 return 0;
 563         }
 564         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 565
 566         st->sb = sb;
 567         if (info == NULL) {
 568                 /* zeroing the superblock */
 569                 return 0;
 570         }
 571
 572         spares = info->working_disks - info->active_disks;
 573         if (info->raid_disks + spares  > MD_SB_DISKS) {
 574                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 575                         info->raid_disks , spares, MD_SB_DISKS);
 576                 return 0;
 577         }
 578
 579         sb->md_magic = MD_SB_MAGIC;
 580         sb->major_version = 0;
 581         sb->minor_version = 90;
 582         sb->patch_version = 0;
 583         sb->gvalid_words = 0; /* ignored */
 584         sb->ctime = time(0);
 585         sb->level = info->level;
 586         if (size != info->size)
 587                 return 0;
 588         sb->size = info->size;
 589         sb->nr_disks = info->nr_disks;
 590         sb->raid_disks = info->raid_disks;
 591         sb->md_minor = info->md_minor;
 592         sb->not_persistent = 0;
 593         if (uuid) {
 594                 sb->set_uuid0 = uuid[0];
 595                 sb->set_uuid1 = uuid[1];
 596                 sb->set_uuid2 = uuid[2];
 597                 sb->set_uuid3 = uuid[3];
 598         } else {
 599                 int rfd = open("/dev/urandom", O_RDONLY);
 600                 if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 601                         sb->set_uuid0 = random();
 602                 if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 603                         sb->set_uuid1 = random();
 604                         sb->set_uuid2 = random();
 605                         sb->set_uuid3 = random();
 606                 }
 607                 if (rfd >= 0)
 608                         close(rfd);
 609         }
 610         if (homehost) {
 611                 char buf[20];
 612                 char *hash = sha1_buffer(homehost,
 613                                          strlen(homehost),
 614                                          buf);
 615                 memcpy(&sb->set_uuid2, hash, 8);
 616         }
 617
 618         sb->utime = sb->ctime;
 619         sb->state = info->state;
 620         sb->active_disks = info->active_disks;
 621         sb->working_disks = info->working_disks;
 622         sb->failed_disks = info->failed_disks;
 623         sb->spare_disks = info->spare_disks;
 624         sb->events_hi = 0;
 625         sb->events_lo = 1;
 626
 627         sb->layout = info->layout;
 628         sb->chunk_size = info->chunk_size;
 629
 630         return 1;
 631 }
 632
 633 struct devinfo {
 634         int fd;
 635         char *devname;
 636         mdu_disk_info_t disk;
 637         struct devinfo *next;
 638 };
 639
 640 #ifndef MDASSEMBLE
 641 /* Add a device to the superblock being created */
 642 static int add_to_super0(struct supertype *st, mdu_disk_info_t *dinfo,
 643                           int fd, char *devname)
 644 {
 645         mdp_super_t *sb = st->sb;
 646         mdp_disk_t *dk = &sb->disks[dinfo->number];
 647         struct devinfo *di, **dip;
 648
 649         dk->number = dinfo->number;
 650         dk->major = dinfo->major;
 651         dk->minor = dinfo->minor;
 652         dk->raid_disk = dinfo->raid_disk;
 653         dk->state = dinfo->state;
 654
 655         sb->this_disk = sb->disks[dinfo->number];
 656         sb->sb_csum = calc_sb0_csum(sb);
 657
 658         dip = (struct devinfo **)&st->info;
 659         while (*dip)
 660                 dip = &(*dip)->next;
 661         di = malloc(sizeof(struct devinfo));
 662         di->fd = fd;
 663         di->devname = devname;
 664         di->disk = *dinfo;
 665         di->next = NULL;
 666         *dip = di;
 667
 668         return 0;
 669 }
 670 #endif
 671
 672 static int store_super0(struct supertype *st, int fd)
 673 {
 674         unsigned long long dsize;
 675         unsigned long long offset;
 676         mdp_super_t *super = st->sb;
 677
 678         if (!get_dev_size(fd, NULL, &dsize))
 679                 return 1;
 680
 681         if (dsize < MD_RESERVED_SECTORS*512)
 682                 return 2;
 683
 684         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 685
 686         offset *= 512;
 687
 688         if (lseek64(fd, offset, 0)< 0LL)
 689                 return 3;
 690
 691         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 692                 return 4;
 693
 694         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 695                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 696                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 697                         if (write(fd, bm, ROUND_UP(sizeof(*bm),4096)) !=
 698                             ROUND_UP(sizeof(*bm),4096))
 699                             return 5;
 700         }
 701
 702         fsync(fd);
 703         return 0;
 704 }
 705
 706 #ifndef MDASSEMBLE
 707 static int write_init_super0(struct supertype *st)
 708 {
 709         mdp_super_t *sb = st->sb;
 710         int rv = 0;
 711         struct devinfo *di;
 712
 713         for (di = st->info ; di && ! rv ; di = di->next) {
 714
 715                 if (di->disk.state == 1)
 716                         continue;
 717                 if (di->fd == -1)
 718                         continue;
 719                 Kill(di->devname, 0, 1, 1);
 720                 Kill(di->devname, 0, 1, 1);
 721
 722                 sb->disks[di->disk.number].state &= ~(1<<MD_DISK_FAULTY);
 723
 724                 sb->this_disk = sb->disks[di->disk.number];
 725                 sb->sb_csum = calc_sb0_csum(sb);
 726                 rv = store_super0(st, di->fd);
 727
 728                 if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 729                         rv = st->ss->write_bitmap(st, di->fd);
 730
 731                 if (rv)
 732                         fprintf(stderr,
 733                                 Name ": failed to write superblock to %s\n",
 734                                 di->devname);
 735                 close(di->fd);
 736                 di->fd = -1;
 737         }
 738         return rv;
 739 }
 740 #endif
 741
 742 static int compare_super0(struct supertype *st, struct supertype *tst)
 743 {
 744         /*
 745          * return:
 746          *  0 same, or first was empty, and second was copied
 747          *  1 second had wrong number
 748          *  2 wrong uuid
 749          *  3 wrong other info
 750          */
 751         mdp_super_t *first = st->sb;
 752         mdp_super_t *second = tst->sb;
 753         int uuid1[4], uuid2[4];
 754
 755         if (second->md_magic != MD_SB_MAGIC)
 756                 return 1;
 757         if (!first) {
 758                 if (posix_memalign((void**)&first, 4096,
 759                              MD_SB_BYTES +
 760                              ROUND_UP(sizeof(struct bitmap_super_s), 4096)) != 0) {
 761                         fprintf(stderr, Name
 762                                 ": %s could not allocate superblock\n", __func__);
 763                         return 1;
 764                 }
 765                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 766                 st->sb = first;
 767                 return 0;
 768         }
 769
 770         uuid_from_super0(st, uuid1);
 771         uuid_from_super0(tst, uuid2);
 772         if (!same_uuid(uuid1, uuid2, 0))
 773                 return 2;
 774         if (first->major_version != second->major_version ||
 775             first->minor_version != second->minor_version ||
 776             first->patch_version != second->patch_version ||
 777             first->gvalid_words  != second->gvalid_words  ||
 778             first->ctime         != second->ctime         ||
 779             first->level         != second->level         ||
 780             first->size          != second->size          ||
 781             first->raid_disks    != second->raid_disks    )
 782                 return 3;
 783
 784         return 0;
 785 }
 786
 787
 788 static void free_super0(struct supertype *st);
 789
 790 static int load_super0(struct supertype *st, int fd, char *devname)
 791 {
 792         /* try to read in the superblock
 793          * Return:
 794          *  0 on success
 795          *  1 on cannot get superblock
 796          *  2 on superblock meaningless
 797          */
 798         unsigned long long dsize;
 799         unsigned long long offset;
 800         mdp_super_t *super;
 801         int uuid[4];
 802         struct bitmap_super_s *bsb;
 803
 804         free_super0(st);
 805
 806         if (st->subarray[0])
 807                 return 1;
 808
 809         if (!get_dev_size(fd, devname, &dsize))
 810                 return 1;
 811
 812         if (dsize < MD_RESERVED_SECTORS*512) {
 813                 if (devname)
 814                         fprintf(stderr, Name
 815                             ": %s is too small for md: size is %llu sectors.\n",
 816                                 devname, dsize);
 817                 return 1;
 818         }
 819
 820         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 821
 822         offset *= 512;
 823
 824         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 825
 826         if (lseek64(fd, offset, 0)< 0LL) {
 827                 if (devname)
 828                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 829                                 devname, strerror(errno));
 830                 return 1;
 831         }
 832
 833         if (posix_memalign((void**)&super, 4096,
 834                            MD_SB_BYTES +
 835                            ROUND_UP(sizeof(bitmap_super_t), 4096)) != 0) {
 836                 fprintf(stderr, Name
 837                         ": %s could not allocate superblock\n", __func__);
 838                 return 1;
 839         }
 840
 841         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 842                 if (devname)
 843                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 844                                 devname);
 845                 free(super);
 846                 return 1;
 847         }
 848
 849         if (st->ss && st->minor_version == 9)
 850                 super0_swap_endian(super);
 851
 852         if (super->md_magic != MD_SB_MAGIC) {
 853                 if (devname)
 854                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 855                                 devname, MD_SB_MAGIC, super->md_magic);
 856                 free(super);
 857                 return 2;
 858         }
 859
 860         if (super->major_version != 0) {
 861                 if (devname)
 862                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 863                                 devname, super->major_version);
 864                 free(super);
 865                 return 2;
 866         }
 867         st->sb = super;
 868
 869         if (st->ss == NULL) {
 870                 st->ss = &super0;
 871                 st->minor_version = super->minor_version;
 872                 st->max_devs = MD_SB_DISKS;
 873                 st->info = NULL;
 874         }
 875
 876         /* Now check on the bitmap superblock */
 877         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 878                 return 0;
 879         /* Read the bitmap superblock and make sure it looks
 880          * valid.  If it doesn't clear the bit.  An --assemble --force
 881          * should get that written out.
 882          */
 883         if (read(fd, super+1, ROUND_UP(sizeof(struct bitmap_super_s),4096))
 884             != ROUND_UP(sizeof(struct bitmap_super_s),4096))
 885                 goto no_bitmap;
 886
 887         uuid_from_super0(st, uuid);
 888         bsb = (struct bitmap_super_s *)(super+1);
 889         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 890             memcmp(bsb->uuid, uuid, 16) != 0)
 891                 goto no_bitmap;
 892         return 0;
 893
 894  no_bitmap:
 895         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 896
 897         return 0;
 898 }
 899
 900 static struct supertype *match_metadata_desc0(char *arg)
 901 {
 902         struct supertype *st = malloc(sizeof(*st));
 903         if (!st) return st;
 904
 905         memset(st, 0, sizeof(*st));
 906         st->ss = &super0;
 907         st->info = NULL;
 908         st->minor_version = 90;
 909         st->max_devs = MD_SB_DISKS;
 910         st->sb = NULL;
 911         /* we sometimes get 00.90 */
 912         while (arg[0] == '0' && arg[1] == '0')
 913                 arg++;
 914         if (strcmp(arg, "0") == 0 ||
 915             strcmp(arg, "0.90") == 0 ||
 916             strcmp(arg, "default") == 0 ||
 917             strcmp(arg, "") == 0 /* no metadata */
 918                 )
 919                 return st;
 920
 921         st->minor_version = 91; /* reshape in progress */
 922         if (strcmp(arg, "0.91") == 0) /* For dup_super support */
 923                 return st;
 924
 925         st->minor_version = 9; /* flag for 'byte-swapped' */
 926         if (strcmp(arg, "0.swap")==0 ||
 927             strcmp(arg, "0.9") == 0) /* For dup_super support */
 928                 return st;
 929
 930         free(st);
 931         return NULL;
 932 }
 933
 934 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 935 {
 936         if (devsize < MD_RESERVED_SECTORS)
 937                 return 0ULL;
 938         return MD_NEW_SIZE_SECTORS(devsize);
 939 }
 940
 941 static int add_internal_bitmap0(struct supertype *st, int *chunkp,
 942                                 int delay, int write_behind,
 943                                 unsigned long long size, int may_change,
 944                                 int major)
 945 {
 946         /*
 947          * The bitmap comes immediately after the superblock and must be 60K in size
 948          * at most.  The default size is between 30K and 60K
 949          *
 950          * size is in sectors,  chunk is in bytes !!!
 951          */
 952         unsigned long long bits;
 953         unsigned long long max_bits = 60*1024*8;
 954         unsigned long long min_chunk;
 955         int chunk = *chunkp;
 956         mdp_super_t *sb = st->sb;
 957         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 958
 959
 960         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 961         bits = (size * 512) / min_chunk + 1;
 962         while (bits > max_bits) {
 963                 min_chunk *= 2;
 964                 bits = (bits+1)/2;
 965         }
 966         if (chunk == UnSet)
 967                 chunk = min_chunk;
 968         else if (chunk < min_chunk)
 969                 return 0; /* chunk size too small */
 970
 971         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 972
 973         memset(bms, 0, sizeof(*bms));
 974         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 975         bms->version = __cpu_to_le32(major);
 976         uuid_from_super0(st, (int*)bms->uuid);
 977         bms->chunksize = __cpu_to_le32(chunk);
 978         bms->daemon_sleep = __cpu_to_le32(delay);
 979         bms->sync_size = __cpu_to_le64(size);
 980         bms->write_behind = __cpu_to_le32(write_behind);
 981         *chunkp = chunk;
 982         return 1;
 983 }
 984
 985
 986 static void locate_bitmap0(struct supertype *st, int fd)
 987 {
 988         unsigned long long dsize;
 989         unsigned long long offset;
 990
 991         if (!get_dev_size(fd, NULL, &dsize))
 992                 return;
 993
 994         if (dsize < MD_RESERVED_SECTORS*512)
 995                 return;
 996
 997         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 998
 999         offset *= 512;
1000
1001         offset += MD_SB_BYTES;
1002
1003         lseek64(fd, offset, 0);
1004 }
1005
1006 static int write_bitmap0(struct supertype *st, int fd)
1007 {
1008         unsigned long long dsize;
1009         unsigned long long offset;
1010         mdp_super_t *sb = st->sb;
1011
1012         int rv = 0;
1013
1014         int towrite, n;
1015         char abuf[4096+4096];
1016         char *buf = (char*)(((long)(abuf+4096))&~4095L);
1017
1018         if (!get_dev_size(fd, NULL, &dsize))
1019                 return 1;
1020
1021
1022         if (dsize < MD_RESERVED_SECTORS*512)
1023                 return -1;
1024
1025         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
1026
1027         offset *= 512;
1028
1029         if (lseek64(fd, offset + 4096, 0)< 0LL)
1030                 return 3;
1031
1032         memset(buf, 0xff, 4096);
1033         memcpy(buf,  ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t));
1034         towrite = 64*1024;
1035         while (towrite > 0) {
1036                 n = towrite;
1037                 if (n > 4096)
1038                         n = 4096;
1039                 n = write(fd, buf, n);
1040                 if (n > 0)
1041                         towrite -= n;
1042                 else
1043                         break;
1044                 memset(buf, 0xff, 4096);
1045         }
1046         fsync(fd);
1047         if (towrite)
1048                 rv = -2;
1049
1050         return rv;
1051 }
1052
1053 static void free_super0(struct supertype *st)
1054 {
1055         if (st->sb)
1056                 free(st->sb);
1057         st->sb = NULL;
1058 }
1059
1060 #ifndef MDASSEMBLE
1061 static int validate_geometry0(struct supertype *st, int level,
1062                               int layout, int raiddisks,
1063                               int chunk, unsigned long long size,
1064                               char *subdev, unsigned long long *freesize,
1065                               int verbose)
1066 {
1067         unsigned long long ldsize;
1068         int fd;
1069
1070         if (level == LEVEL_CONTAINER)
1071                 return 0;
1072         if (raiddisks > MD_SB_DISKS)
1073                 return 0;
1074         if (size > (0x7fffffffULL<<9))
1075                 return 0;
1076         if (!subdev)
1077                 return 1;
1078
1079         fd = open(subdev, O_RDONLY|O_EXCL, 0);
1080         if (fd < 0) {
1081                 if (verbose)
1082                         fprintf(stderr, Name ": super0.90 cannot open %s: %s\n",
1083                                 subdev, strerror(errno));
1084                 return 0;
1085         }
1086
1087         if (!get_dev_size(fd, subdev, &ldsize)) {
1088                 close(fd);
1089                 return 0;
1090         }
1091         close(fd);
1092
1093         if (ldsize < MD_RESERVED_SECTORS * 512)
1094                 return 0;
1095         if (size > (0x7fffffffULL<<9))
1096                 return 0;
1097         *freesize = MD_NEW_SIZE_SECTORS(ldsize >> 9);
1098         return 1;
1099 }
1100 #endif /* MDASSEMBLE */
1101
1102 struct superswitch super0 = {
1103 #ifndef MDASSEMBLE
1104         .examine_super = examine_super0,
1105         .brief_examine_super = brief_examine_super0,
1106         .export_examine_super = export_examine_super0,
1107         .detail_super = detail_super0,
1108         .brief_detail_super = brief_detail_super0,
1109         .write_init_super = write_init_super0,
1110         .validate_geometry = validate_geometry0,
1111         .add_to_super = add_to_super0,
1112 #endif
1113         .match_home = match_home0,
1114         .uuid_from_super = uuid_from_super0,
1115         .getinfo_super = getinfo_super0,
1116         .update_super = update_super0,
1117         .init_super = init_super0,
1118         .store_super = store_super0,
1119         .compare_super = compare_super0,
1120         .load_super = load_super0,
1121         .match_metadata_desc = match_metadata_desc0,
1122         .avail_size = avail_size0,
1123         .add_internal_bitmap = add_internal_bitmap0,
1124         .locate_bitmap = locate_bitmap0,
1125         .write_bitmap = write_bitmap0,
1126         .free_super = free_super0,
1127         .name = "0.90",
1128 };