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 #include "mdadm.h"
  31 #include <openssl/sha.h> /* for SHA1 */
  32 /*
  33  * All handling for the 0.90.0 version superblock is in
  34  * this file.
  35  * This includes:
  36  *   - finding, loading, and writing the superblock.
  37  *   - initialising a new superblock
  38  *   - printing the superblock for --examine
  39  *   - printing part of the superblock for --detail
  40  * .. other stuff
  41  */
  42
  43
  44 static unsigned long calc_sb0_csum(mdp_super_t *super)
  45 {
  46         unsigned long csum = super->sb_csum;
  47         unsigned long newcsum;
  48         super->sb_csum= 0 ;
  49         newcsum = calc_csum(super, MD_SB_BYTES);
  50         super->sb_csum = csum;
  51         return newcsum;
  52 }
  53
  54
  55 void super0_swap_endian(struct mdp_superblock_s *sb)
  56 {
  57         /* as super0 superblocks are host-endian, it is sometimes
  58          * useful to be able to swap the endianness
  59          * as (almost) everything is u32's we byte-swap every 4byte
  60          * number.
  61          * We then also have to swap the events_hi and events_lo
  62          */
  63         char *sbc = (char *)sb;
  64         __u32 t32;
  65         int i;
  66
  67         for (i=0; i < MD_SB_BYTES ; i+=4) {
  68                 char t = sbc[i];
  69                 sbc[i] = sbc[i+3];
  70                 sbc[i+3] = t;
  71                 t=sbc[i+1];
  72                 sbc[i+1]=sbc[i+2];
  73                 sbc[i+2]=t;
  74         }
  75         t32 = sb->events_hi;
  76         sb->events_hi = sb->events_lo;
  77         sb->events_lo = t32;
  78
  79         t32 = sb->cp_events_hi;
  80         sb->cp_events_hi = sb->cp_events_lo;
  81         sb->cp_events_lo = t32;
  82
  83 }
  84
  85 #ifndef MDASSEMBLE
  86
  87 static void examine_super0(void *sbv, char *homehost)
  88 {
  89         mdp_super_t *sb = sbv;
  90         time_t atime;
  91         int d;
  92         char *c;
  93
  94         printf("          Magic : %08x\n", sb->md_magic);
  95         printf("        Version : %02d.%02d.%02d\n", sb->major_version, sb->minor_version,
  96                sb->patch_version);
  97         if (sb->minor_version >= 90) {
  98                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
  99                        sb->set_uuid2, sb->set_uuid3);
 100                 if (homehost) {
 101                         unsigned char *hash = SHA1((unsigned char *)homehost,
 102                                                    strlen(homehost),
 103                                                    NULL);
 104                         if (memcmp(&sb->set_uuid2, hash, 8)==0)
 105                                 printf(" (local to host %s)", homehost);
 106                 }
 107                 printf("\n");
 108         } else
 109                 printf("           UUID : %08x\n", sb->set_uuid0);
 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("    Device Size : %d%s\n", sb->size, human_size((long long)sb->size<<10));
 118                 switch(sb->level) {
 119                 case 1: ddsks=1;break;
 120                 case 4:
 121                 case 5: ddsks = sb->raid_disks-1; break;
 122                 case 6: ddsks = sb->raid_disks-2; break;
 123                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 124                 }
 125                 if (ddsks)
 126                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 127                                human_size(ddsks*(long long)sb->size<<10));
 128         }
 129         printf("   Raid Devices : %d\n", sb->raid_disks);
 130         printf("  Total Devices : %d\n", sb->nr_disks);
 131         printf("Preferred Minor : %d\n", sb->md_minor);
 132         printf("\n");
 133         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 134                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 135                 if (sb->delta_disks) {
 136                         printf("  Delta Devices : %d", sb->delta_disks);
 137                         if (sb->delta_disks)
 138                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 139                         else
 140                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 141                 }
 142                 if (sb->new_level != sb->level) {
 143                         c = map_num(pers, sb->new_level);
 144                         printf("      New Level : %s\n", c?c:"-unknown-");
 145                 }
 146                 if (sb->new_layout != sb->layout) {
 147                         if (sb->level == 5) {
 148                                 c = map_num(r5layout, sb->new_layout);
 149                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 150                         }
 151                         if (sb->level == 10) {
 152                                 printf("     New Layout : near=%d, %s=%d\n",
 153                                        sb->new_layout&255,
 154                                        (sb->new_layout&0x10000)?"offset":"far",
 155                                        (sb->new_layout>>8)&255);
 156                         }
 157                 }
 158                 if (sb->new_chunk != sb->chunk_size)
 159                         printf("  New Chunksize : %d\n", sb->new_chunk);
 160                 printf("\n");
 161         }
 162         atime = sb->utime;
 163         printf("    Update Time : %.24s\n", ctime(&atime));
 164         printf("          State : %s\n",
 165                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 166         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 167                 printf("Internal Bitmap : present\n");
 168         printf(" Active Devices : %d\n", sb->active_disks);
 169         printf("Working Devices : %d\n", sb->working_disks);
 170         printf(" Failed Devices : %d\n", sb->failed_disks);
 171         printf("  Spare Devices : %d\n", sb->spare_disks);
 172         if (calc_sb0_csum(sb) == sb->sb_csum)
 173                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 174         else
 175                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 176         printf("         Events : %d.%d\n", sb->events_hi, sb->events_lo);
 177         printf("\n");
 178         if (sb->level == 5) {
 179                 c = map_num(r5layout, sb->layout);
 180                 printf("         Layout : %s\n", c?c:"-unknown-");
 181         }
 182         if (sb->level == 10) {
 183                 printf("         Layout : near=%d, %s=%d\n",
 184                        sb->layout&255,
 185                        (sb->layout&0x10000)?"offset":"far",
 186                        (sb->layout>>8)&255);
 187         }
 188         switch(sb->level) {
 189         case 0:
 190         case 4:
 191         case 5:
 192                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 193                 break;
 194         case -1:
 195                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 196                 break;
 197         default: break;
 198         }
 199         printf("\n");
 200         printf("      Number   Major   Minor   RaidDevice State\n");
 201         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
 202                 mdp_disk_t *dp;
 203                 char *dv;
 204                 char nb[5];
 205                 int wonly;
 206                 if (d>=0) dp = &sb->disks[d];
 207                 else dp = &sb->this_disk;
 208                 snprintf(nb, sizeof(nb), "%4d", d);
 209                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 210                        dp->number, dp->major, dp->minor, dp->raid_disk);
 211                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 212                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 213                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 214                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 215                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 216                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 217                 if (wonly) printf(" write-mostly");
 218                 if (dp->state == 0) printf(" spare");
 219                 if ((dv=map_dev(dp->major, dp->minor, 0)))
 220                         printf("   %s", dv);
 221                 printf("\n");
 222                 if (d == -1) printf("\n");
 223         }
 224 }
 225
 226 static void brief_examine_super0(void *sbv)
 227 {
 228         mdp_super_t *sb = sbv;
 229         char *c=map_num(pers, sb->level);
 230         char devname[20];
 231         struct stat stb;
 232
 233         sprintf(devname, "/dev/md%d", sb->md_minor);
 234         if (stat(devname, &stb) != 0) {
 235                 /* /dev/mdX doesn't exist, so use /dev/md/X */
 236                 sprintf(devname, "/dev/md/%d", sb->md_minor);
 237         }
 238
 239         printf("ARRAY %s level=%s num-devices=%d UUID=",
 240                devname,
 241                c?c:"-unknown-", sb->raid_disks);
 242         if (sb->minor_version >= 90)
 243                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 244                        sb->set_uuid2, sb->set_uuid3);
 245         else
 246                 printf("%08x", sb->set_uuid0);
 247         printf("\n");
 248 }
 249
 250 static void detail_super0(void *sbv, char *homehost)
 251 {
 252         mdp_super_t *sb = sbv;
 253         printf("           UUID : ");
 254         if (sb->minor_version >= 90)
 255                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 256                        sb->set_uuid2, sb->set_uuid3);
 257         else
 258                 printf("%08x", sb->set_uuid0);
 259         if (homehost) {
 260                 unsigned char *hash = SHA1((unsigned char *)homehost,
 261                                            strlen(homehost),
 262                                            NULL);
 263                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 264                         printf(" (local to host %s)", homehost);
 265         }
 266         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 267 }
 268
 269 static void brief_detail_super0(void *sbv)
 270 {
 271         mdp_super_t *sb = sbv;
 272         printf(" UUID=");
 273         if (sb->minor_version >= 90)
 274                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 275                        sb->set_uuid2, sb->set_uuid3);
 276         else
 277                 printf("%08x", sb->set_uuid0);
 278 }
 279 #endif
 280
 281 static int match_home0(void *sbv, char *homehost)
 282 {
 283         mdp_super_t *sb = sbv;
 284         unsigned char *hash = SHA1((unsigned char *)homehost,
 285                                    strlen(homehost),
 286                                    NULL);
 287
 288         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 289 }
 290
 291 static void uuid_from_super0(int uuid[4], void * sbv)
 292 {
 293         mdp_super_t *super = sbv;
 294         uuid[0] = super->set_uuid0;
 295         if (super->minor_version >= 90) {
 296                 uuid[1] = super->set_uuid1;
 297                 uuid[2] = super->set_uuid2;
 298                 uuid[3] = super->set_uuid3;
 299         } else {
 300                 uuid[1] = 0;
 301                 uuid[2] = 0;
 302                 uuid[3] = 0;
 303         }
 304 }
 305
 306 static void getinfo_super0(struct mdinfo *info, void *sbv)
 307 {
 308         mdp_super_t *sb = sbv;
 309         int working = 0;
 310         int i;
 311
 312         info->array.major_version = sb->major_version;
 313         info->array.minor_version = sb->minor_version;
 314         info->array.patch_version = sb->patch_version;
 315         info->array.raid_disks = sb->raid_disks;
 316         info->array.level = sb->level;
 317         info->array.layout = sb->layout;
 318         info->array.md_minor = sb->md_minor;
 319         info->array.ctime = sb->ctime;
 320         info->array.utime = sb->utime;
 321         info->array.chunk_size = sb->chunk_size;
 322         info->component_size = sb->size*2;
 323
 324         info->disk.state = sb->this_disk.state;
 325         info->disk.major = sb->this_disk.major;
 326         info->disk.minor = sb->this_disk.minor;
 327         info->disk.raid_disk = sb->this_disk.raid_disk;
 328         info->disk.number = sb->this_disk.number;
 329
 330         info->events = md_event(sb);
 331         info->data_offset = 0;
 332
 333         uuid_from_super0(info->uuid, sbv);
 334
 335         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 336                 info->reshape_active = 1;
 337                 info->reshape_progress = sb->reshape_position;
 338                 info->new_level = sb->new_level;
 339                 info->delta_disks = sb->delta_disks;
 340                 info->new_layout = sb->new_layout;
 341                 info->new_chunk = sb->new_chunk;
 342         } else
 343                 info->reshape_active = 0;
 344
 345         sprintf(info->name, "%d", sb->md_minor);
 346         /* work_disks is calculated rather than read directly */
 347         for (i=0; i < MD_SB_DISKS; i++)
 348                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 349                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 350                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 351                         working ++;
 352         info->array.working_disks = working;
 353 }
 354
 355
 356 static int update_super0(struct mdinfo *info, void *sbv, char *update,
 357                          char *devname, int verbose,
 358                          int uuid_set, char *homehost)
 359 {
 360         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 361          * For others, the return value is ignored.
 362          */
 363         int rv = 0;
 364         mdp_super_t *sb = sbv;
 365         if (strcmp(update, "sparc2.2")==0 ) {
 366                 /* 2.2 sparc put the events in the wrong place
 367                  * So we copy the tail of the superblock
 368                  * up 4 bytes before continuing
 369                  */
 370                 __u32 *sb32 = (__u32*)sb;
 371                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 372                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 373                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 374                 if (verbose >= 0)
 375                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 376                                  devname);
 377         }
 378         if (strcmp(update, "super-minor") ==0) {
 379                 sb->md_minor = info->array.md_minor;
 380                 if (verbose > 0)
 381                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 382                                 devname, info->array.md_minor);
 383         }
 384         if (strcmp(update, "summaries") == 0) {
 385                 int i;
 386                 /* set nr_disks, active_disks, working_disks,
 387                  * failed_disks, spare_disks based on disks[]
 388                  * array in superblock.
 389                  * Also make sure extra slots aren't 'failed'
 390                  */
 391                 sb->nr_disks = sb->active_disks =
 392                         sb->working_disks = sb->failed_disks =
 393                         sb->spare_disks = 0;
 394                 for (i=0; i < MD_SB_DISKS ; i++)
 395                         if (sb->disks[i].major ||
 396                             sb->disks[i].minor) {
 397                                 int state = sb->disks[i].state;
 398                                 if (state & (1<<MD_DISK_REMOVED))
 399                                         continue;
 400                                 sb->nr_disks++;
 401                                 if (state & (1<<MD_DISK_ACTIVE))
 402                                         sb->active_disks++;
 403                                 if (state & (1<<MD_DISK_FAULTY))
 404                                         sb->failed_disks++;
 405                                 else
 406                                         sb->working_disks++;
 407                                 if (state == 0)
 408                                         sb->spare_disks++;
 409                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 410                                 sb->disks[i].state = 0;
 411         }
 412         if (strcmp(update, "force")==0) {
 413                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 414                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 415                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 416                 if (sb->events_hi != ehi ||
 417                     sb->events_lo != elo)
 418                         rv = 1;
 419                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 420                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 421                         /* need to force clean */
 422                         sb->state |= (1 << MD_SB_CLEAN);
 423                         rv = 1;
 424                 }
 425         }
 426         if (strcmp(update, "assemble")==0) {
 427                 int d = info->disk.number;
 428                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 429                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 430                     != info->disk.state) {
 431                         sb->disks[d].state = info->disk.state | wonly;
 432                         rv = 1;
 433                 }
 434         }
 435         if (strcmp(update, "newdev") == 0) {
 436                 int d = info->disk.number;
 437                 memset(&sb->disks[d], 0, sizeof(sb->disks[d]));
 438                 sb->disks[d].number = d;
 439                 sb->disks[d].major = info->disk.major;
 440                 sb->disks[d].minor = info->disk.minor;
 441                 sb->disks[d].raid_disk = info->disk.raid_disk;
 442                 sb->disks[d].state = info->disk.state;
 443                 sb->this_disk = sb->disks[d];
 444         }
 445         if (strcmp(update, "grow") == 0) {
 446                 sb->raid_disks = info->array.raid_disks;
 447                 sb->nr_disks = info->array.nr_disks;
 448                 sb->active_disks = info->array.active_disks;
 449                 sb->working_disks = info->array.working_disks;
 450                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 451                 sb->disks[info->disk.number].number = info->disk.number;
 452                 sb->disks[info->disk.number].major = info->disk.major;
 453                 sb->disks[info->disk.number].minor = info->disk.minor;
 454                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 455                 sb->disks[info->disk.number].state = info->disk.state;
 456                 if (sb->this_disk.number == info->disk.number)
 457                         sb->this_disk = sb->disks[info->disk.number];
 458         }
 459         if (strcmp(update, "resync") == 0) {
 460                 /* make sure resync happens */
 461                 sb->state &= ~(1<<MD_SB_CLEAN);
 462                 sb->recovery_cp = 0;
 463         }
 464         if (strcmp(update, "homehost") == 0 &&
 465             homehost) {
 466                 uuid_set = 0;
 467                 update = "uuid";
 468                 info->uuid[0] = sb->set_uuid0;
 469                 info->uuid[1] = sb->set_uuid1;
 470         }
 471         if (strcmp(update, "uuid") == 0) {
 472                 if (!uuid_set && homehost) {
 473                         unsigned char *hash = SHA1((unsigned char*)homehost,
 474                                                    strlen(homehost),
 475                                                    NULL);
 476                         memcpy(info->uuid+2, hash, 8);
 477                 }
 478                 sb->set_uuid0 = info->uuid[0];
 479                 sb->set_uuid1 = info->uuid[1];
 480                 sb->set_uuid2 = info->uuid[2];
 481                 sb->set_uuid3 = info->uuid[3];
 482                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 483                         struct bitmap_super_s *bm;
 484                         bm = (struct bitmap_super_s*)(sb+1);
 485                         uuid_from_super0((int*)bm->uuid, sbv);
 486                 }
 487         }
 488         if (strcmp(update, "_reshape_progress")==0)
 489                 sb->reshape_position = info->reshape_progress;
 490
 491         sb->sb_csum = calc_sb0_csum(sb);
 492         return rv;
 493 }
 494
 495 static __u64 event_super0(void *sbv)
 496 {
 497         mdp_super_t *sb = sbv;
 498         return md_event(sb);
 499 }
 500
 501 /*
 502  * For verion-0 superblock, the homehost is 'stored' in the
 503  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 504  * of random material.
 505  * We use the first 8 bytes (64bits) of the sha1 of the
 506  * host name
 507  */
 508
 509
 510 static int init_super0(struct supertype *st, void **sbp, mdu_array_info_t *info,
 511                        unsigned long long size, char *ignored_name, char *homehost)
 512 {
 513         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 514         int spares;
 515         int rfd;
 516         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 517
 518         if (info->major_version == -1) {
 519                 /* zeroing the superblock */
 520                 *sbp = sb;
 521                 return 0;
 522         }
 523
 524         spares = info->working_disks - info->active_disks;
 525         if (info->raid_disks + spares  > MD_SB_DISKS) {
 526                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 527                         info->raid_disks , spares, MD_SB_DISKS);
 528                 return 0;
 529         }
 530
 531         rfd = open("/dev/urandom", O_RDONLY);
 532         sb->md_magic = MD_SB_MAGIC;
 533         sb->major_version = 0;
 534         sb->minor_version = 90;
 535         sb->patch_version = 0;
 536         sb->gvalid_words = 0; /* ignored */
 537         if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 538                 sb->set_uuid0 = random();
 539         sb->ctime = time(0);
 540         sb->level = info->level;
 541         if (size != info->size)
 542                 return 0;
 543         sb->size = info->size;
 544         sb->nr_disks = info->nr_disks;
 545         sb->raid_disks = info->raid_disks;
 546         sb->md_minor = info->md_minor;
 547         sb->not_persistent = 0;
 548         if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 549                 sb->set_uuid1 = random();
 550                 sb->set_uuid2 = random();
 551                 sb->set_uuid3 = random();
 552         }
 553         if (rfd >= 0)
 554                 close(rfd);
 555         if (homehost) {
 556                 unsigned char *hash = SHA1((unsigned char*)homehost,
 557                                            strlen(homehost),
 558                                            NULL);
 559                 memcpy(&sb->set_uuid2, hash, 8);
 560         }
 561
 562         sb->utime = sb->ctime;
 563         sb->state = info->state;
 564         sb->active_disks = info->active_disks;
 565         sb->working_disks = info->working_disks;
 566         sb->failed_disks = info->failed_disks;
 567         sb->spare_disks = info->spare_disks;
 568         sb->events_hi = 0;
 569         sb->events_lo = 1;
 570
 571         sb->layout = info->layout;
 572         sb->chunk_size = info->chunk_size;
 573
 574         *sbp = sb;
 575         return 1;
 576 }
 577
 578 /* Add a device to the superblock being created */
 579 static void add_to_super0(void *sbv, mdu_disk_info_t *dinfo)
 580 {
 581         mdp_super_t *sb = sbv;
 582         mdp_disk_t *dk = &sb->disks[dinfo->number];
 583
 584         dk->number = dinfo->number;
 585         dk->major = dinfo->major;
 586         dk->minor = dinfo->minor;
 587         dk->raid_disk = dinfo->raid_disk;
 588         dk->state = dinfo->state;
 589 }
 590
 591 static int store_super0(struct supertype *st, int fd, void *sbv)
 592 {
 593         unsigned long size;
 594         unsigned long long dsize;
 595         unsigned long long offset;
 596         mdp_super_t *super = sbv;
 597
 598 #ifdef BLKGETSIZE64
 599         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 600 #endif
 601         {
 602                 if (ioctl(fd, BLKGETSIZE, &size))
 603                         return 1;
 604                 else
 605                         dsize = ((unsigned long long)size)<<9;
 606         }
 607
 608         if (dsize < MD_RESERVED_SECTORS*2*512)
 609                 return 2;
 610
 611         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 612
 613         offset *= 512;
 614
 615         if (lseek64(fd, offset, 0)< 0LL)
 616                 return 3;
 617
 618         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 619                 return 4;
 620
 621         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 622                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 623                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 624                         write(fd, bm, sizeof(*bm));
 625         }
 626
 627         fsync(fd);
 628         return 0;
 629 }
 630
 631 static int write_init_super0(struct supertype *st, void *sbv, mdu_disk_info_t *dinfo, char *devname)
 632 {
 633         mdp_super_t *sb = sbv;
 634         int fd = open(devname, O_RDWR|O_EXCL);
 635         int rv;
 636
 637         if (fd < 0) {
 638                 fprintf(stderr, Name ": Failed to open %s to write superblock\n", devname);
 639                 return -1;
 640         }
 641
 642         sb->disks[dinfo->number].state &= ~(1<<MD_DISK_FAULTY);
 643         sb->disks[dinfo->number].state |= (1<<MD_DISK_SYNC);
 644
 645         sb->this_disk = sb->disks[dinfo->number];
 646         sb->sb_csum = calc_sb0_csum(sb);
 647         rv = store_super0(st, fd, sb);
 648
 649         if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 650                 rv = st->ss->write_bitmap(st, fd, sbv);
 651
 652         close(fd);
 653         if (rv)
 654                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
 655         return rv;
 656 }
 657
 658 static int compare_super0(void **firstp, void *secondv)
 659 {
 660         /*
 661          * return:
 662          *  0 same, or first was empty, and second was copied
 663          *  1 second had wrong number
 664          *  2 wrong uuid
 665          *  3 wrong other info
 666          */
 667         mdp_super_t *first = *firstp;
 668         mdp_super_t *second = secondv;
 669
 670         int uuid1[4], uuid2[4];
 671         if (second->md_magic != MD_SB_MAGIC)
 672                 return 1;
 673         if (!first) {
 674                 first = malloc(MD_SB_BYTES + sizeof(struct bitmap_super_s));
 675                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 676                 *firstp = first;
 677                 return 0;
 678         }
 679
 680         uuid_from_super0(uuid1, first);
 681         uuid_from_super0(uuid2, second);
 682         if (!same_uuid(uuid1, uuid2, 0))
 683                 return 2;
 684         if (first->major_version != second->major_version ||
 685             first->minor_version != second->minor_version ||
 686             first->patch_version != second->patch_version ||
 687             first->gvalid_words  != second->gvalid_words  ||
 688             first->ctime         != second->ctime         ||
 689             first->level         != second->level         ||
 690             first->size          != second->size          ||
 691             first->raid_disks    != second->raid_disks    )
 692                 return 3;
 693
 694         return 0;
 695 }
 696
 697
 698 static int load_super0(struct supertype *st, int fd, void **sbp, char *devname)
 699 {
 700         /* try to read in the superblock
 701          * Return:
 702          *  0 on success
 703          *  1 on cannot get superblock
 704          *  2 on superblock meaningless
 705          */
 706         unsigned long size;
 707         unsigned long long dsize;
 708         unsigned long long offset;
 709         mdp_super_t *super;
 710         int uuid[4];
 711         struct bitmap_super_s *bsb;
 712
 713 #ifdef BLKGETSIZE64
 714         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 715 #endif
 716         {
 717                 if (ioctl(fd, BLKGETSIZE, &size)) {
 718                         if (devname)
 719                                 fprintf(stderr, Name ": cannot find device size for %s: %s\n",
 720                                         devname, strerror(errno));
 721                         return 1;
 722                 } else
 723                         dsize = size << 9;
 724         }
 725
 726         if (dsize < MD_RESERVED_SECTORS*2) {
 727                 if (devname)
 728                         fprintf(stderr, Name ": %s is too small for md: size is %ld sectors.\n",
 729                                 devname, size);
 730                 return 1;
 731         }
 732
 733         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 734
 735         offset *= 512;
 736
 737         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 738
 739         if (lseek64(fd, offset, 0)< 0LL) {
 740                 if (devname)
 741                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 742                                 devname, strerror(errno));
 743                 return 1;
 744         }
 745
 746         super = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 747
 748         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 749                 if (devname)
 750                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 751                                 devname);
 752                 free(super);
 753                 return 1;
 754         }
 755
 756         if (st->ss && st->minor_version == 9)
 757                 super0_swap_endian(super);
 758
 759         if (super->md_magic != MD_SB_MAGIC) {
 760                 if (devname)
 761                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 762                                 devname, MD_SB_MAGIC, super->md_magic);
 763                 free(super);
 764                 return 2;
 765         }
 766
 767         if (super->major_version != 0) {
 768                 if (devname)
 769                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 770                                 devname, super->major_version);
 771                 free(super);
 772                 return 2;
 773         }
 774         *sbp = super;
 775         if (st->ss == NULL) {
 776                 st->ss = &super0;
 777                 st->minor_version = 90;
 778                 st->max_devs = MD_SB_DISKS;
 779         }
 780
 781         /* Now check on the bitmap superblock */
 782         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 783                 return 0;
 784         /* Read the bitmap superblock and make sure it looks
 785          * valid.  If it doesn't clear the bit.  An --assemble --force
 786          * should get that written out.
 787          */
 788         if (read(fd, super+1, sizeof(struct bitmap_super_s))
 789             != sizeof(struct bitmap_super_s))
 790                 goto no_bitmap;
 791
 792         uuid_from_super0(uuid, super);
 793         bsb = (struct bitmap_super_s *)(super+1);
 794         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 795             memcmp(bsb->uuid, uuid, 16) != 0)
 796                 goto no_bitmap;
 797         return 0;
 798
 799  no_bitmap:
 800         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 801
 802         return 0;
 803 }
 804
 805 static struct supertype *match_metadata_desc0(char *arg)
 806 {
 807         struct supertype *st = malloc(sizeof(*st));
 808         if (!st) return st;
 809
 810         st->ss = &super0;
 811         st->minor_version = 90;
 812         st->max_devs = MD_SB_DISKS;
 813         if (strcmp(arg, "0") == 0 ||
 814             strcmp(arg, "0.90") == 0 ||
 815             strcmp(arg, "default") == 0
 816                 )
 817                 return st;
 818
 819         st->minor_version = 9; /* flag for 'byte-swapped' */
 820         if (strcmp(arg, "0.swap")==0)
 821                 return st;
 822
 823         free(st);
 824         return NULL;
 825 }
 826
 827 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 828 {
 829         if (devsize < MD_RESERVED_SECTORS*2)
 830                 return 0ULL;
 831         return MD_NEW_SIZE_SECTORS(devsize);
 832 }
 833
 834 static int add_internal_bitmap0(struct supertype *st, void *sbv, int chunk, int delay, int write_behind, unsigned long long size, int may_change, int major)
 835 {
 836         /*
 837          * The bitmap comes immediately after the superblock and must be 60K in size
 838          * at most.  The default size is between 30K and 60K
 839          *
 840          * size is in sectors,  chunk is in bytes !!!
 841          */
 842         unsigned long long bits;
 843         unsigned long long max_bits = 60*1024*8;
 844         unsigned long long min_chunk;
 845         mdp_super_t *sb = sbv;
 846         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 847
 848
 849         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 850         bits = (size * 512) / min_chunk + 1;
 851         while (bits > max_bits) {
 852                 min_chunk *= 2;
 853                 bits = (bits+1)/2;
 854         }
 855         if (chunk == UnSet)
 856                 chunk = min_chunk;
 857         else if (chunk < min_chunk)
 858                 return 0; /* chunk size too small */
 859
 860         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 861
 862         memset(bms, 0, sizeof(*bms));
 863         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 864         bms->version = __cpu_to_le32(major);
 865         uuid_from_super0((int*)bms->uuid, sb);
 866         bms->chunksize = __cpu_to_le32(chunk);
 867         bms->daemon_sleep = __cpu_to_le32(delay);
 868         bms->sync_size = __cpu_to_le64(size);
 869         bms->write_behind = __cpu_to_le32(write_behind);
 870
 871         return 1;
 872 }
 873
 874
 875 void locate_bitmap0(struct supertype *st, int fd, void *sbv)
 876 {
 877         unsigned long long dsize;
 878         unsigned long size;
 879         unsigned long long offset;
 880 #ifdef BLKGETSIZE64
 881         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 882 #endif
 883         {
 884                 if (ioctl(fd, BLKGETSIZE, &size))
 885                         return;
 886                 else
 887                         dsize = ((unsigned long long)size)<<9;
 888         }
 889
 890         if (dsize < MD_RESERVED_SECTORS*2)
 891                 return;
 892
 893         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 894
 895         offset *= 512;
 896
 897         offset += MD_SB_BYTES;
 898
 899         lseek64(fd, offset, 0);
 900 }
 901
 902 int write_bitmap0(struct supertype *st, int fd, void *sbv)
 903 {
 904         unsigned long size;
 905         unsigned long long dsize;
 906         unsigned long long offset;
 907         mdp_super_t *sb = sbv;
 908
 909         int rv = 0;
 910
 911         int towrite, n;
 912         char buf[4096];
 913
 914 #ifdef BLKGETSIZE64
 915         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 916 #endif
 917         {
 918                 if (ioctl(fd, BLKGETSIZE, &size))
 919                         return 1;
 920                 else
 921                         dsize = ((unsigned long long)size)<<9;
 922         }
 923
 924         if (dsize < MD_RESERVED_SECTORS*2)
 925         return -1;
 926
 927         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 928
 929         offset *= 512;
 930
 931         if (lseek64(fd, offset + 4096, 0)< 0LL)
 932                 return 3;
 933
 934
 935         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
 936             sizeof(bitmap_super_t))
 937                 return -2;
 938         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
 939         memset(buf, 0xff, sizeof(buf));
 940         while (towrite > 0) {
 941                 n = towrite;
 942                 if (n > sizeof(buf))
 943                         n = sizeof(buf);
 944                 n = write(fd, buf, n);
 945                 if (n > 0)
 946                         towrite -= n;
 947                 else
 948                         break;
 949         }
 950         fsync(fd);
 951         if (towrite)
 952                 rv = -2;
 953
 954         return rv;
 955 }
 956
 957 struct superswitch super0 = {
 958 #ifndef MDASSEMBLE
 959         .examine_super = examine_super0,
 960         .brief_examine_super = brief_examine_super0,
 961         .detail_super = detail_super0,
 962         .brief_detail_super = brief_detail_super0,
 963 #endif
 964         .match_home = match_home0,
 965         .uuid_from_super = uuid_from_super0,
 966         .getinfo_super = getinfo_super0,
 967         .update_super = update_super0,
 968         .event_super = event_super0,
 969         .init_super = init_super0,
 970         .add_to_super = add_to_super0,
 971         .store_super = store_super0,
 972         .write_init_super = write_init_super0,
 973         .compare_super = compare_super0,
 974         .load_super = load_super0,
 975         .match_metadata_desc = match_metadata_desc0,
 976         .avail_size = avail_size0,
 977         .add_internal_bitmap = add_internal_bitmap0,
 978         .locate_bitmap = locate_bitmap0,
 979         .write_bitmap = write_bitmap0,
 980         .major = 0,
 981         .swapuuid = 0,
 982 };