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