super0.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@cse.unsw.edu.au>
  23  *    Paper: Neil Brown
  24  *           School of Computer Science and Engineering
  25  *           The University of New South Wales
  26  *           Sydney, 2052
  27  *           Australia
  28  */
  29
  30 #define HAVE_STDINT_H 1
  31 #include "mdadm.h"
  32 #include "sha1.h"
  33 /*
  34  * All handling for the 0.90.0 version superblock is in
  35  * this file.
  36  * This includes:
  37  *   - finding, loading, and writing the superblock.
  38  *   - initialising a new superblock
  39  *   - printing the superblock for --examine
  40  *   - printing part of the superblock for --detail
  41  * .. other stuff
  42  */
  43
  44
  45 static unsigned long calc_sb0_csum(mdp_super_t *super)
  46 {
  47         unsigned long csum = super->sb_csum;
  48         unsigned long newcsum;
  49         super->sb_csum= 0 ;
  50         newcsum = calc_csum(super, MD_SB_BYTES);
  51         super->sb_csum = csum;
  52         return newcsum;
  53 }
  54
  55
  56 void super0_swap_endian(struct mdp_superblock_s *sb)
  57 {
  58         /* as super0 superblocks are host-endian, it is sometimes
  59          * useful to be able to swap the endianness
  60          * as (almost) everything is u32's we byte-swap every 4byte
  61          * number.
  62          * We then also have to swap the events_hi and events_lo
  63          */
  64         char *sbc = (char *)sb;
  65         __u32 t32;
  66         int i;
  67
  68         for (i=0; i < MD_SB_BYTES ; i+=4) {
  69                 char t = sbc[i];
  70                 sbc[i] = sbc[i+3];
  71                 sbc[i+3] = t;
  72                 t=sbc[i+1];
  73                 sbc[i+1]=sbc[i+2];
  74                 sbc[i+2]=t;
  75         }
  76         t32 = sb->events_hi;
  77         sb->events_hi = sb->events_lo;
  78         sb->events_lo = t32;
  79
  80         t32 = sb->cp_events_hi;
  81         sb->cp_events_hi = sb->cp_events_lo;
  82         sb->cp_events_lo = t32;
  83
  84 }
  85
  86 #ifndef MDASSEMBLE
  87
  88 static void examine_super0(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
 305 static void export_detail_super0(struct supertype *st)
 306 {
 307         mdp_super_t *sb = st->sb;
 308         printf("MD_UUID=");
 309         if (sb->minor_version >= 90)
 310                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 311                        sb->set_uuid2, sb->set_uuid3);
 312         else
 313                 printf("%08x", sb->set_uuid0);
 314         printf("\n");
 315 }
 316 #endif
 317
 318 static int match_home0(struct supertype *st, char *homehost)
 319 {
 320         mdp_super_t *sb = st->sb;
 321         char buf[20];
 322         char *hash = sha1_buffer(homehost,
 323                                  strlen(homehost),
 324                                  buf);
 325
 326         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 327 }
 328
 329 static void uuid_from_super0(struct supertype *st, int uuid[4])
 330 {
 331         mdp_super_t *super = st->sb;
 332         uuid[0] = super->set_uuid0;
 333         if (super->minor_version >= 90) {
 334                 uuid[1] = super->set_uuid1;
 335                 uuid[2] = super->set_uuid2;
 336                 uuid[3] = super->set_uuid3;
 337         } else {
 338                 uuid[1] = 0;
 339                 uuid[2] = 0;
 340                 uuid[3] = 0;
 341         }
 342 }
 343
 344 static void getinfo_super0(struct supertype *st, struct mdinfo *info)
 345 {
 346         mdp_super_t *sb = st->sb;
 347         int working = 0;
 348         int i;
 349
 350         info->array.major_version = sb->major_version;
 351         info->array.minor_version = sb->minor_version;
 352         info->array.patch_version = sb->patch_version;
 353         info->array.raid_disks = sb->raid_disks;
 354         info->array.level = sb->level;
 355         info->array.layout = sb->layout;
 356         info->array.md_minor = sb->md_minor;
 357         info->array.ctime = sb->ctime;
 358         info->array.utime = sb->utime;
 359         info->array.chunk_size = sb->chunk_size;
 360         info->array.state = sb->state;
 361         info->component_size = sb->size*2;
 362
 363         info->disk.state = sb->this_disk.state;
 364         info->disk.major = sb->this_disk.major;
 365         info->disk.minor = sb->this_disk.minor;
 366         info->disk.raid_disk = sb->this_disk.raid_disk;
 367         info->disk.number = sb->this_disk.number;
 368
 369         info->events = md_event(sb);
 370         info->data_offset = 0;
 371
 372         sprintf(info->text_version, "0.%d", sb->minor_version);
 373
 374         uuid_from_super0(st, info->uuid);
 375
 376         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 377                 info->reshape_active = 1;
 378                 info->reshape_progress = sb->reshape_position;
 379                 info->new_level = sb->new_level;
 380                 info->delta_disks = sb->delta_disks;
 381                 info->new_layout = sb->new_layout;
 382                 info->new_chunk = sb->new_chunk;
 383         } else
 384                 info->reshape_active = 0;
 385
 386         sprintf(info->name, "%d", sb->md_minor);
 387         /* work_disks is calculated rather than read directly */
 388         for (i=0; i < MD_SB_DISKS; i++)
 389                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 390                     (sb->disks[i].raid_disk < info->array.raid_disks) &&
 391                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 392                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 393                         working ++;
 394         info->array.working_disks = working;
 395 }
 396
 397
 398 static int update_super0(struct supertype *st, struct mdinfo *info,
 399                          char *update,
 400                          char *devname, int verbose,
 401                          int uuid_set, char *homehost)
 402 {
 403         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 404          * For others, the return value is ignored.
 405          */
 406         int rv = 0;
 407         mdp_super_t *sb = st->sb;
 408         if (strcmp(update, "sparc2.2")==0 ) {
 409                 /* 2.2 sparc put the events in the wrong place
 410                  * So we copy the tail of the superblock
 411                  * up 4 bytes before continuing
 412                  */
 413                 __u32 *sb32 = (__u32*)sb;
 414                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 415                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 416                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 417                 if (verbose >= 0)
 418                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 419                                  devname);
 420         }
 421         if (strcmp(update, "super-minor") ==0) {
 422                 sb->md_minor = info->array.md_minor;
 423                 if (verbose > 0)
 424                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 425                                 devname, info->array.md_minor);
 426         }
 427         if (strcmp(update, "summaries") == 0) {
 428                 int i;
 429                 /* set nr_disks, active_disks, working_disks,
 430                  * failed_disks, spare_disks based on disks[]
 431                  * array in superblock.
 432                  * Also make sure extra slots aren't 'failed'
 433                  */
 434                 sb->nr_disks = sb->active_disks =
 435                         sb->working_disks = sb->failed_disks =
 436                         sb->spare_disks = 0;
 437                 for (i=0; i < MD_SB_DISKS ; i++)
 438                         if (sb->disks[i].major ||
 439                             sb->disks[i].minor) {
 440                                 int state = sb->disks[i].state;
 441                                 if (state & (1<<MD_DISK_REMOVED))
 442                                         continue;
 443                                 sb->nr_disks++;
 444                                 if (state & (1<<MD_DISK_ACTIVE))
 445                                         sb->active_disks++;
 446                                 if (state & (1<<MD_DISK_FAULTY))
 447                                         sb->failed_disks++;
 448                                 else
 449                                         sb->working_disks++;
 450                                 if (state == 0)
 451                                         sb->spare_disks++;
 452                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 453                                 sb->disks[i].state = 0;
 454         }
 455         if (strcmp(update, "force-one")==0) {
 456                 /* Not enough devices for a working array, so
 457                  * bring this one up-to-date.
 458                  */
 459                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 460                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 461                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 462                 if (sb->events_hi != ehi ||
 463                     sb->events_lo != elo)
 464                         rv = 1;
 465         }
 466         if (strcmp(update, "force-array")==0) {
 467                 /* degraded array and 'force' requested, so
 468                  * maybe need to mark it 'clean'
 469                  */
 470                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 471                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 472                         /* need to force clean */
 473                         sb->state |= (1 << MD_SB_CLEAN);
 474                         rv = 1;
 475                 }
 476         }
 477         if (strcmp(update, "assemble")==0) {
 478                 int d = info->disk.number;
 479                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 480                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 481                     != info->disk.state) {
 482                         sb->disks[d].state = info->disk.state | wonly;
 483                         rv = 1;
 484                 }
 485         }
 486         if (strcmp(update, "linear-grow-new") == 0) {
 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                 sb->this_disk = sb->disks[info->disk.number];
 494         }
 495         if (strcmp(update, "linear-grow-update") == 0) {
 496                 sb->raid_disks = info->array.raid_disks;
 497                 sb->nr_disks = info->array.nr_disks;
 498                 sb->active_disks = info->array.active_disks;
 499                 sb->working_disks = info->array.working_disks;
 500                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 501                 sb->disks[info->disk.number].number = info->disk.number;
 502                 sb->disks[info->disk.number].major = info->disk.major;
 503                 sb->disks[info->disk.number].minor = info->disk.minor;
 504                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 505                 sb->disks[info->disk.number].state = info->disk.state;
 506         }
 507         if (strcmp(update, "resync") == 0) {
 508                 /* make sure resync happens */
 509                 sb->state &= ~(1<<MD_SB_CLEAN);
 510                 sb->recovery_cp = 0;
 511         }
 512         if (strcmp(update, "homehost") == 0 &&
 513             homehost) {
 514                 uuid_set = 0;
 515                 update = "uuid";
 516                 info->uuid[0] = sb->set_uuid0;
 517                 info->uuid[1] = sb->set_uuid1;
 518         }
 519         if (strcmp(update, "uuid") == 0) {
 520                 if (!uuid_set && homehost) {
 521                         char buf[20];
 522                         char *hash = sha1_buffer(homehost,
 523                                                  strlen(homehost),
 524                                                  buf);
 525                         memcpy(info->uuid+2, hash, 8);
 526                 }
 527                 sb->set_uuid0 = info->uuid[0];
 528                 sb->set_uuid1 = info->uuid[1];
 529                 sb->set_uuid2 = info->uuid[2];
 530                 sb->set_uuid3 = info->uuid[3];
 531                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 532                         struct bitmap_super_s *bm;
 533                         bm = (struct bitmap_super_s*)(sb+1);
 534                         uuid_from_super0(st, (int*)bm->uuid);
 535                 }
 536         }
 537         if (strcmp(update, "_reshape_progress")==0)
 538                 sb->reshape_position = info->reshape_progress;
 539
 540         sb->sb_csum = calc_sb0_csum(sb);
 541         return rv;
 542 }
 543
 544 /*
 545  * For verion-0 superblock, the homehost is 'stored' in the
 546  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 547  * of random material.
 548  * We use the first 8 bytes (64bits) of the sha1 of the
 549  * host name
 550  */
 551
 552
 553 static int init_super0(struct supertype *st, mdu_array_info_t *info,
 554                        unsigned long long size, char *ignored_name, char *homehost,
 555                        int *uuid)
 556 {
 557         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 558         int spares;
 559         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 560
 561         st->sb = sb;
 562         if (info->major_version == -1) {
 563                 /* zeroing the superblock */
 564                 return 0;
 565         }
 566
 567         spares = info->working_disks - info->active_disks;
 568         if (info->raid_disks + spares  > MD_SB_DISKS) {
 569                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 570                         info->raid_disks , spares, MD_SB_DISKS);
 571                 return 0;
 572         }
 573
 574         sb->md_magic = MD_SB_MAGIC;
 575         sb->major_version = 0;
 576         sb->minor_version = 90;
 577         sb->patch_version = 0;
 578         sb->gvalid_words = 0; /* ignored */
 579         sb->ctime = time(0);
 580         sb->level = info->level;
 581         if (size != info->size)
 582                 return 0;
 583         sb->size = info->size;
 584         sb->nr_disks = info->nr_disks;
 585         sb->raid_disks = info->raid_disks;
 586         sb->md_minor = info->md_minor;
 587         sb->not_persistent = 0;
 588         if (uuid) {
 589                 sb->set_uuid0 = uuid[0];
 590                 sb->set_uuid1 = uuid[1];
 591                 sb->set_uuid2 = uuid[2];
 592                 sb->set_uuid3 = uuid[3];
 593         } else {
 594                 int rfd = open("/dev/urandom", O_RDONLY);
 595                 if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 596                         sb->set_uuid0 = random();
 597                 if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 598                         sb->set_uuid1 = random();
 599                         sb->set_uuid2 = random();
 600                         sb->set_uuid3 = random();
 601                 }
 602                 if (rfd >= 0)
 603                         close(rfd);
 604         }
 605         if (homehost) {
 606                 char buf[20];
 607                 char *hash = sha1_buffer(homehost,
 608                                          strlen(homehost),
 609                                          buf);
 610                 memcpy(&sb->set_uuid2, hash, 8);
 611         }
 612
 613         sb->utime = sb->ctime;
 614         sb->state = info->state;
 615         sb->active_disks = info->active_disks;
 616         sb->working_disks = info->working_disks;
 617         sb->failed_disks = info->failed_disks;
 618         sb->spare_disks = info->spare_disks;
 619         sb->events_hi = 0;
 620         sb->events_lo = 1;
 621
 622         sb->layout = info->layout;
 623         sb->chunk_size = info->chunk_size;
 624
 625         return 1;
 626 }
 627
 628 struct devinfo {
 629         int fd;
 630         char *devname;
 631         mdu_disk_info_t disk;
 632         struct devinfo *next;
 633 };
 634 /* Add a device to the superblock being created */
 635 static void add_to_super0(struct supertype *st, mdu_disk_info_t *dinfo,
 636                           int fd, char *devname)
 637 {
 638         mdp_super_t *sb = st->sb;
 639         mdp_disk_t *dk = &sb->disks[dinfo->number];
 640         struct devinfo *di, **dip;
 641
 642         dk->number = dinfo->number;
 643         dk->major = dinfo->major;
 644         dk->minor = dinfo->minor;
 645         dk->raid_disk = dinfo->raid_disk;
 646         dk->state = dinfo->state;
 647
 648         sb->this_disk = sb->disks[dinfo->number];
 649         sb->sb_csum = calc_sb0_csum(sb);
 650
 651         dip = (struct devinfo **)&st->info;
 652         while (*dip)
 653                 dip = &(*dip)->next;
 654         di = malloc(sizeof(struct devinfo));
 655         di->fd = fd;
 656         di->devname = devname;
 657         di->disk = *dinfo;
 658         di->next = NULL;
 659         *dip = di;
 660 }
 661
 662 static int store_super0(struct supertype *st, int fd)
 663 {
 664         unsigned long long dsize;
 665         unsigned long long offset;
 666         mdp_super_t *super = st->sb;
 667
 668         if (!get_dev_size(fd, NULL, &dsize))
 669                 return 1;
 670
 671         if (dsize < MD_RESERVED_SECTORS*512)
 672                 return 2;
 673
 674         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 675
 676         offset *= 512;
 677
 678         if (lseek64(fd, offset, 0)< 0LL)
 679                 return 3;
 680
 681         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 682                 return 4;
 683
 684         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 685                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 686                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 687                         if (write(fd, bm, sizeof(*bm)) != sizeof(*bm))
 688                             return 5;
 689         }
 690
 691         fsync(fd);
 692         return 0;
 693 }
 694
 695 #ifndef MDASSEMBLE
 696 static int write_init_super0(struct supertype *st)
 697 {
 698         mdp_super_t *sb = st->sb;
 699         int rv = 0;
 700         struct devinfo *di;
 701
 702         for (di = st->info ; di && ! rv ; di = di->next) {
 703
 704                 if (di->disk.state == 1)
 705                         continue;
 706                 if (di->fd == -1)
 707                         continue;
 708                 Kill(di->devname, 0, 1, 1);
 709                 Kill(di->devname, 0, 1, 1);
 710
 711                 sb->disks[di->disk.number].state &= ~(1<<MD_DISK_FAULTY);
 712
 713                 sb->this_disk = sb->disks[di->disk.number];
 714                 sb->sb_csum = calc_sb0_csum(sb);
 715                 rv = store_super0(st, di->fd);
 716
 717                 if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 718                         rv = st->ss->write_bitmap(st, di->fd);
 719
 720                 if (rv)
 721                         fprintf(stderr,
 722                                 Name ": failed to write superblock to %s\n",
 723                                 di->devname);
 724                 close(di->fd);
 725                 di->fd = -1;
 726         }
 727         return rv;
 728 }
 729 #endif
 730
 731 static int compare_super0(struct supertype *st, struct supertype *tst)
 732 {
 733         /*
 734          * return:
 735          *  0 same, or first was empty, and second was copied
 736          *  1 second had wrong number
 737          *  2 wrong uuid
 738          *  3 wrong other info
 739          */
 740         mdp_super_t *first = st->sb;
 741         mdp_super_t *second = tst->sb;
 742         int uuid1[4], uuid2[4];
 743
 744         if (second->md_magic != MD_SB_MAGIC)
 745                 return 1;
 746         if (!first) {
 747                 first = malloc(MD_SB_BYTES + sizeof(struct bitmap_super_s));
 748                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 749                 st->sb = first;
 750                 return 0;
 751         }
 752
 753         uuid_from_super0(st, uuid1);
 754         uuid_from_super0(tst, uuid2);
 755         if (!same_uuid(uuid1, uuid2, 0))
 756                 return 2;
 757         if (first->major_version != second->major_version ||
 758             first->minor_version != second->minor_version ||
 759             first->patch_version != second->patch_version ||
 760             first->gvalid_words  != second->gvalid_words  ||
 761             first->ctime         != second->ctime         ||
 762             first->level         != second->level         ||
 763             first->size          != second->size          ||
 764             first->raid_disks    != second->raid_disks    )
 765                 return 3;
 766
 767         return 0;
 768 }
 769
 770
 771 static void free_super0(struct supertype *st);
 772
 773 static int load_super0(struct supertype *st, int fd, char *devname)
 774 {
 775         /* try to read in the superblock
 776          * Return:
 777          *  0 on success
 778          *  1 on cannot get superblock
 779          *  2 on superblock meaningless
 780          */
 781         unsigned long long dsize;
 782         unsigned long long offset;
 783         mdp_super_t *super;
 784         int uuid[4];
 785         struct bitmap_super_s *bsb;
 786
 787         free_super0(st);
 788
 789         if (st->subarray[0])
 790                 return 1;
 791
 792         if (!get_dev_size(fd, devname, &dsize))
 793                 return 1;
 794
 795         if (dsize < MD_RESERVED_SECTORS*512) {
 796                 if (devname)
 797                         fprintf(stderr, Name
 798                             ": %s is too small for md: size is %llu sectors.\n",
 799                                 devname, dsize);
 800                 return 1;
 801         }
 802
 803         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 804
 805         offset *= 512;
 806
 807         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 808
 809         if (lseek64(fd, offset, 0)< 0LL) {
 810                 if (devname)
 811                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 812                                 devname, strerror(errno));
 813                 return 1;
 814         }
 815
 816         super = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 817
 818         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 819                 if (devname)
 820                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 821                                 devname);
 822                 free(super);
 823                 return 1;
 824         }
 825
 826         if (st->ss && st->minor_version == 9)
 827                 super0_swap_endian(super);
 828
 829         if (super->md_magic != MD_SB_MAGIC) {
 830                 if (devname)
 831                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 832                                 devname, MD_SB_MAGIC, super->md_magic);
 833                 free(super);
 834                 return 2;
 835         }
 836
 837         if (super->major_version != 0) {
 838                 if (devname)
 839                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 840                                 devname, super->major_version);
 841                 free(super);
 842                 return 2;
 843         }
 844         st->sb = super;
 845
 846         if (st->ss == NULL) {
 847                 st->ss = &super0;
 848                 st->minor_version = super->minor_version;
 849                 st->max_devs = MD_SB_DISKS;
 850                 st->info = NULL;
 851         }
 852
 853         /* Now check on the bitmap superblock */
 854         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 855                 return 0;
 856         /* Read the bitmap superblock and make sure it looks
 857          * valid.  If it doesn't clear the bit.  An --assemble --force
 858          * should get that written out.
 859          */
 860         if (read(fd, super+1, sizeof(struct bitmap_super_s))
 861             != sizeof(struct bitmap_super_s))
 862                 goto no_bitmap;
 863
 864         uuid_from_super0(st, uuid);
 865         bsb = (struct bitmap_super_s *)(super+1);
 866         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 867             memcmp(bsb->uuid, uuid, 16) != 0)
 868                 goto no_bitmap;
 869         return 0;
 870
 871  no_bitmap:
 872         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 873
 874         return 0;
 875 }
 876
 877 static struct supertype *match_metadata_desc0(char *arg)
 878 {
 879         struct supertype *st = malloc(sizeof(*st));
 880         if (!st) return st;
 881
 882         memset(st, 0, sizeof(*st));
 883         st->ss = &super0;
 884         st->info = NULL;
 885         st->minor_version = 90;
 886         st->max_devs = MD_SB_DISKS;
 887         st->sb = NULL;
 888         if (strcmp(arg, "0") == 0 ||
 889             strcmp(arg, "0.90") == 0 ||
 890             strcmp(arg, "default") == 0 ||
 891             strcmp(arg, "") == 0 /* no metadata */
 892                 )
 893                 return st;
 894
 895         st->minor_version = 91; /* reshape in progress */
 896         if (strcmp(arg, "0.91") == 0) /* For dup_super support */
 897                 return st;
 898
 899         st->minor_version = 9; /* flag for 'byte-swapped' */
 900         if (strcmp(arg, "0.swap")==0 ||
 901             strcmp(arg, "0.9") == 0) /* For dup_super support */
 902                 return st;
 903
 904         free(st);
 905         return NULL;
 906 }
 907
 908 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 909 {
 910         if (devsize < MD_RESERVED_SECTORS)
 911                 return 0ULL;
 912         return MD_NEW_SIZE_SECTORS(devsize);
 913 }
 914
 915 static int add_internal_bitmap0(struct supertype *st, int *chunkp,
 916                                 int delay, int write_behind,
 917                                 unsigned long long size, int may_change,
 918                                 int major)
 919 {
 920         /*
 921          * The bitmap comes immediately after the superblock and must be 60K in size
 922          * at most.  The default size is between 30K and 60K
 923          *
 924          * size is in sectors,  chunk is in bytes !!!
 925          */
 926         unsigned long long bits;
 927         unsigned long long max_bits = 60*1024*8;
 928         unsigned long long min_chunk;
 929         int chunk = *chunkp;
 930         mdp_super_t *sb = st->sb;
 931         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 932
 933
 934         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 935         bits = (size * 512) / min_chunk + 1;
 936         while (bits > max_bits) {
 937                 min_chunk *= 2;
 938                 bits = (bits+1)/2;
 939         }
 940         if (chunk == UnSet)
 941                 chunk = min_chunk;
 942         else if (chunk < min_chunk)
 943                 return 0; /* chunk size too small */
 944
 945         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 946
 947         memset(bms, 0, sizeof(*bms));
 948         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 949         bms->version = __cpu_to_le32(major);
 950         uuid_from_super0(st, (int*)bms->uuid);
 951         bms->chunksize = __cpu_to_le32(chunk);
 952         bms->daemon_sleep = __cpu_to_le32(delay);
 953         bms->sync_size = __cpu_to_le64(size);
 954         bms->write_behind = __cpu_to_le32(write_behind);
 955         *chunkp = chunk;
 956         return 1;
 957 }
 958
 959
 960 void locate_bitmap0(struct supertype *st, int fd)
 961 {
 962         unsigned long long dsize;
 963         unsigned long long offset;
 964
 965         if (!get_dev_size(fd, NULL, &dsize))
 966                 return;
 967
 968         if (dsize < MD_RESERVED_SECTORS*512)
 969                 return;
 970
 971         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 972
 973         offset *= 512;
 974
 975         offset += MD_SB_BYTES;
 976
 977         lseek64(fd, offset, 0);
 978 }
 979
 980 int write_bitmap0(struct supertype *st, int fd)
 981 {
 982         unsigned long long dsize;
 983         unsigned long long offset;
 984         mdp_super_t *sb = st->sb;
 985
 986         int rv = 0;
 987
 988         int towrite, n;
 989         char buf[4096];
 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
1006         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
1007             sizeof(bitmap_super_t))
1008                 return -2;
1009         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
1010         memset(buf, 0xff, sizeof(buf));
1011         while (towrite > 0) {
1012                 n = towrite;
1013                 if (n > sizeof(buf))
1014                         n = sizeof(buf);
1015                 n = write(fd, buf, n);
1016                 if (n > 0)
1017                         towrite -= n;
1018                 else
1019                         break;
1020         }
1021         fsync(fd);
1022         if (towrite)
1023                 rv = -2;
1024
1025         return rv;
1026 }
1027
1028 static void free_super0(struct supertype *st)
1029 {
1030         if (st->sb)
1031                 free(st->sb);
1032         st->sb = NULL;
1033 }
1034
1035 static int validate_geometry0(struct supertype *st, int level,
1036                               int layout, int raiddisks,
1037                               int chunk, unsigned long long size,
1038                               char *subdev, unsigned long long *freesize)
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                 fprintf(stderr, Name ": Cannot open %s: %s\n",
1055                         subdev, strerror(errno));
1056                 return 0;
1057         }
1058         if (!get_dev_size(fd, subdev, &ldsize)) {
1059                 close(fd);
1060                 return 0;
1061         }
1062         close(fd);
1063
1064         if (ldsize < MD_RESERVED_SECTORS * 512)
1065                 return 0;
1066         if (size > (0x7fffffffULL<<10))
1067                 return 0;
1068         *freesize = MD_NEW_SIZE_SECTORS(ldsize >> 9);
1069         return 1;
1070 }
1071
1072 struct superswitch super0 = {
1073 #ifndef MDASSEMBLE
1074         .examine_super = examine_super0,
1075         .brief_examine_super = brief_examine_super0,
1076         .export_examine_super = export_examine_super0,
1077         .detail_super = detail_super0,
1078         .brief_detail_super = brief_detail_super0,
1079         .export_detail_super = export_detail_super0,
1080         .write_init_super = write_init_super0,
1081 #endif
1082         .match_home = match_home0,
1083         .uuid_from_super = uuid_from_super0,
1084         .getinfo_super = getinfo_super0,
1085         .update_super = update_super0,
1086         .init_super = init_super0,
1087         .add_to_super = add_to_super0,
1088         .store_super = store_super0,
1089         .compare_super = compare_super0,
1090         .load_super = load_super0,
1091         .match_metadata_desc = match_metadata_desc0,
1092         .avail_size = avail_size0,
1093         .add_internal_bitmap = add_internal_bitmap0,
1094         .locate_bitmap = locate_bitmap0,
1095         .write_bitmap = write_bitmap0,
1096         .free_super = free_super0,
1097         .validate_geometry = validate_geometry0,
1098 };