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 #ifndef UCLIBC
  32 #include <openssl/sha.h> /* for SHA1 */
  33 #else
  34 extern unsigned char *SHA1(unsigned char *buf, int len, unsigned char *dest);
  35 #endif
  36 /*
  37  * All handling for the 0.90.0 version superblock is in
  38  * this file.
  39  * This includes:
  40  *   - finding, loading, and writing the superblock.
  41  *   - initialising a new superblock
  42  *   - printing the superblock for --examine
  43  *   - printing part of the superblock for --detail
  44  * .. other stuff
  45  */
  46
  47
  48 static unsigned long calc_sb0_csum(mdp_super_t *super)
  49 {
  50         unsigned long csum = super->sb_csum;
  51         unsigned long newcsum;
  52         super->sb_csum= 0 ;
  53         newcsum = calc_csum(super, MD_SB_BYTES);
  54         super->sb_csum = csum;
  55         return newcsum;
  56 }
  57
  58
  59 void super0_swap_endian(struct mdp_superblock_s *sb)
  60 {
  61         /* as super0 superblocks are host-endian, it is sometimes
  62          * useful to be able to swap the endianness
  63          * as (almost) everything is u32's we byte-swap every 4byte
  64          * number.
  65          * We then also have to swap the events_hi and events_lo
  66          */
  67         char *sbc = (char *)sb;
  68         __u32 t32;
  69         int i;
  70
  71         for (i=0; i < MD_SB_BYTES ; i+=4) {
  72                 char t = sbc[i];
  73                 sbc[i] = sbc[i+3];
  74                 sbc[i+3] = t;
  75                 t=sbc[i+1];
  76                 sbc[i+1]=sbc[i+2];
  77                 sbc[i+2]=t;
  78         }
  79         t32 = sb->events_hi;
  80         sb->events_hi = sb->events_lo;
  81         sb->events_lo = t32;
  82
  83         t32 = sb->cp_events_hi;
  84         sb->cp_events_hi = sb->cp_events_lo;
  85         sb->cp_events_lo = t32;
  86
  87 }
  88
  89 #ifndef MDASSEMBLE
  90
  91 static void examine_super0(void *sbv, char *homehost)
  92 {
  93         mdp_super_t *sb = sbv;
  94         time_t atime;
  95         int d;
  96         char *c;
  97
  98         printf("          Magic : %08x\n", sb->md_magic);
  99         printf("        Version : %02d.%02d.%02d\n", sb->major_version, sb->minor_version,
 100                sb->patch_version);
 101         if (sb->minor_version >= 90) {
 102                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 103                        sb->set_uuid2, sb->set_uuid3);
 104                 if (homehost) {
 105                         unsigned char *hash = SHA1((unsigned char *)homehost,
 106                                                    strlen(homehost),
 107                                                    NULL);
 108                         if (memcmp(&sb->set_uuid2, hash, 8)==0)
 109                                 printf(" (local to host %s)", homehost);
 110                 }
 111                 printf("\n");
 112         } else
 113                 printf("           UUID : %08x\n", sb->set_uuid0);
 114
 115         atime = sb->ctime;
 116         printf("  Creation Time : %.24s\n", ctime(&atime));
 117         c=map_num(pers, sb->level);
 118         printf("     Raid Level : %s\n", c?c:"-unknown-");
 119         if ((int)sb->level >= 0) {
 120                 int ddsks=0;
 121                 printf("    Device Size : %d%s\n", sb->size, human_size((long long)sb->size<<10));
 122                 switch(sb->level) {
 123                 case 1: ddsks=1;break;
 124                 case 4:
 125                 case 5: ddsks = sb->raid_disks-1; break;
 126                 case 6: ddsks = sb->raid_disks-2; break;
 127                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 128                 }
 129                 if (ddsks)
 130                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 131                                human_size(ddsks*(long long)sb->size<<10));
 132         }
 133         printf("   Raid Devices : %d\n", sb->raid_disks);
 134         printf("  Total Devices : %d\n", sb->nr_disks);
 135         printf("Preferred Minor : %d\n", sb->md_minor);
 136         printf("\n");
 137         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 138                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 139                 if (sb->delta_disks) {
 140                         printf("  Delta Devices : %d", sb->delta_disks);
 141                         if (sb->delta_disks)
 142                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 143                         else
 144                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 145                 }
 146                 if (sb->new_level != sb->level) {
 147                         c = map_num(pers, sb->new_level);
 148                         printf("      New Level : %s\n", c?c:"-unknown-");
 149                 }
 150                 if (sb->new_layout != sb->layout) {
 151                         if (sb->level == 5) {
 152                                 c = map_num(r5layout, sb->new_layout);
 153                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 154                         }
 155                         if (sb->level == 10) {
 156                                 printf("     New Layout : near=%d, %s=%d\n",
 157                                        sb->new_layout&255,
 158                                        (sb->new_layout&0x10000)?"offset":"far",
 159                                        (sb->new_layout>>8)&255);
 160                         }
 161                 }
 162                 if (sb->new_chunk != sb->chunk_size)
 163                         printf("  New Chunksize : %d\n", sb->new_chunk);
 164                 printf("\n");
 165         }
 166         atime = sb->utime;
 167         printf("    Update Time : %.24s\n", ctime(&atime));
 168         printf("          State : %s\n",
 169                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 170         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 171                 printf("Internal Bitmap : present\n");
 172         printf(" Active Devices : %d\n", sb->active_disks);
 173         printf("Working Devices : %d\n", sb->working_disks);
 174         printf(" Failed Devices : %d\n", sb->failed_disks);
 175         printf("  Spare Devices : %d\n", sb->spare_disks);
 176         if (calc_sb0_csum(sb) == sb->sb_csum)
 177                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 178         else
 179                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 180         printf("         Events : %d.%d\n", sb->events_hi, sb->events_lo);
 181         printf("\n");
 182         if (sb->level == 5) {
 183                 c = map_num(r5layout, sb->layout);
 184                 printf("         Layout : %s\n", c?c:"-unknown-");
 185         }
 186         if (sb->level == 10) {
 187                 printf("         Layout : near=%d, %s=%d\n",
 188                        sb->layout&255,
 189                        (sb->layout&0x10000)?"offset":"far",
 190                        (sb->layout>>8)&255);
 191         }
 192         switch(sb->level) {
 193         case 0:
 194         case 4:
 195         case 5:
 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(void *sbv)
 231 {
 232         mdp_super_t *sb = sbv;
 233         char *c=map_num(pers, sb->level);
 234         char devname[20];
 235         struct stat stb;
 236
 237         sprintf(devname, "/dev/md%d", sb->md_minor);
 238         if (stat(devname, &stb) != 0) {
 239                 /* /dev/mdX doesn't exist, so use /dev/md/X */
 240                 sprintf(devname, "/dev/md/%d", sb->md_minor);
 241         }
 242
 243         printf("ARRAY %s level=%s num-devices=%d UUID=",
 244                devname,
 245                c?c:"-unknown-", sb->raid_disks);
 246         if (sb->minor_version >= 90)
 247                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 248                        sb->set_uuid2, sb->set_uuid3);
 249         else
 250                 printf("%08x", sb->set_uuid0);
 251         printf("\n");
 252 }
 253
 254 static void detail_super0(void *sbv, char *homehost)
 255 {
 256         mdp_super_t *sb = sbv;
 257         printf("           UUID : ");
 258         if (sb->minor_version >= 90)
 259                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 260                        sb->set_uuid2, sb->set_uuid3);
 261         else
 262                 printf("%08x", sb->set_uuid0);
 263         if (homehost) {
 264                 unsigned char *hash = SHA1((unsigned char *)homehost,
 265                                            strlen(homehost),
 266                                            NULL);
 267                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 268                         printf(" (local to host %s)", homehost);
 269         }
 270         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 271 }
 272
 273 static void brief_detail_super0(void *sbv)
 274 {
 275         mdp_super_t *sb = sbv;
 276         printf(" UUID=");
 277         if (sb->minor_version >= 90)
 278                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 279                        sb->set_uuid2, sb->set_uuid3);
 280         else
 281                 printf("%08x", sb->set_uuid0);
 282 }
 283 #endif
 284
 285 static int match_home0(void *sbv, char *homehost)
 286 {
 287         mdp_super_t *sb = sbv;
 288         unsigned char *hash = SHA1((unsigned char *)homehost,
 289                                    strlen(homehost),
 290                                    NULL);
 291
 292         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 293 }
 294
 295 static void uuid_from_super0(int uuid[4], void * sbv)
 296 {
 297         mdp_super_t *super = sbv;
 298         uuid[0] = super->set_uuid0;
 299         if (super->minor_version >= 90) {
 300                 uuid[1] = super->set_uuid1;
 301                 uuid[2] = super->set_uuid2;
 302                 uuid[3] = super->set_uuid3;
 303         } else {
 304                 uuid[1] = 0;
 305                 uuid[2] = 0;
 306                 uuid[3] = 0;
 307         }
 308 }
 309
 310 static void getinfo_super0(struct mdinfo *info, void *sbv)
 311 {
 312         mdp_super_t *sb = sbv;
 313         int working = 0;
 314         int i;
 315
 316         info->array.major_version = sb->major_version;
 317         info->array.minor_version = sb->minor_version;
 318         info->array.patch_version = sb->patch_version;
 319         info->array.raid_disks = sb->raid_disks;
 320         info->array.level = sb->level;
 321         info->array.layout = sb->layout;
 322         info->array.md_minor = sb->md_minor;
 323         info->array.ctime = sb->ctime;
 324         info->array.utime = sb->utime;
 325         info->array.chunk_size = sb->chunk_size;
 326         info->component_size = sb->size*2;
 327
 328         info->disk.state = sb->this_disk.state;
 329         info->disk.major = sb->this_disk.major;
 330         info->disk.minor = sb->this_disk.minor;
 331         info->disk.raid_disk = sb->this_disk.raid_disk;
 332         info->disk.number = sb->this_disk.number;
 333
 334         info->events = md_event(sb);
 335         info->data_offset = 0;
 336
 337         uuid_from_super0(info->uuid, sbv);
 338
 339         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 340                 info->reshape_active = 1;
 341                 info->reshape_progress = sb->reshape_position;
 342                 info->new_level = sb->new_level;
 343                 info->delta_disks = sb->delta_disks;
 344                 info->new_layout = sb->new_layout;
 345                 info->new_chunk = sb->new_chunk;
 346         } else
 347                 info->reshape_active = 0;
 348
 349         sprintf(info->name, "%d", sb->md_minor);
 350         /* work_disks is calculated rather than read directly */
 351         for (i=0; i < MD_SB_DISKS; i++)
 352                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 353                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 354                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 355                         working ++;
 356         info->array.working_disks = working;
 357 }
 358
 359
 360 static int update_super0(struct mdinfo *info, void *sbv, char *update,
 361                          char *devname, int verbose,
 362                          int uuid_set, char *homehost)
 363 {
 364         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 365          * For others, the return value is ignored.
 366          */
 367         int rv = 0;
 368         mdp_super_t *sb = sbv;
 369         if (strcmp(update, "sparc2.2")==0 ) {
 370                 /* 2.2 sparc put the events in the wrong place
 371                  * So we copy the tail of the superblock
 372                  * up 4 bytes before continuing
 373                  */
 374                 __u32 *sb32 = (__u32*)sb;
 375                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 376                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 377                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 378                 if (verbose >= 0)
 379                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 380                                  devname);
 381         }
 382         if (strcmp(update, "super-minor") ==0) {
 383                 sb->md_minor = info->array.md_minor;
 384                 if (verbose > 0)
 385                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 386                                 devname, info->array.md_minor);
 387         }
 388         if (strcmp(update, "summaries") == 0) {
 389                 int i;
 390                 /* set nr_disks, active_disks, working_disks,
 391                  * failed_disks, spare_disks based on disks[]
 392                  * array in superblock.
 393                  * Also make sure extra slots aren't 'failed'
 394                  */
 395                 sb->nr_disks = sb->active_disks =
 396                         sb->working_disks = sb->failed_disks =
 397                         sb->spare_disks = 0;
 398                 for (i=0; i < MD_SB_DISKS ; i++)
 399                         if (sb->disks[i].major ||
 400                             sb->disks[i].minor) {
 401                                 int state = sb->disks[i].state;
 402                                 if (state & (1<<MD_DISK_REMOVED))
 403                                         continue;
 404                                 sb->nr_disks++;
 405                                 if (state & (1<<MD_DISK_ACTIVE))
 406                                         sb->active_disks++;
 407                                 if (state & (1<<MD_DISK_FAULTY))
 408                                         sb->failed_disks++;
 409                                 else
 410                                         sb->working_disks++;
 411                                 if (state == 0)
 412                                         sb->spare_disks++;
 413                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 414                                 sb->disks[i].state = 0;
 415         }
 416         if (strcmp(update, "force")==0) {
 417                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 418                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 419                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 420                 if (sb->events_hi != ehi ||
 421                     sb->events_lo != elo)
 422                         rv = 1;
 423                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 424                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 425                         /* need to force clean */
 426                         sb->state |= (1 << MD_SB_CLEAN);
 427                         rv = 1;
 428                 }
 429         }
 430         if (strcmp(update, "assemble")==0) {
 431                 int d = info->disk.number;
 432                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 433                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 434                     != info->disk.state) {
 435                         sb->disks[d].state = info->disk.state | wonly;
 436                         rv = 1;
 437                 }
 438         }
 439         if (strcmp(update, "newdev") == 0) {
 440                 int d = info->disk.number;
 441                 memset(&sb->disks[d], 0, sizeof(sb->disks[d]));
 442                 sb->disks[d].number = d;
 443                 sb->disks[d].major = info->disk.major;
 444                 sb->disks[d].minor = info->disk.minor;
 445                 sb->disks[d].raid_disk = info->disk.raid_disk;
 446                 sb->disks[d].state = info->disk.state;
 447                 sb->this_disk = sb->disks[d];
 448         }
 449         if (strcmp(update, "grow") == 0) {
 450                 sb->raid_disks = info->array.raid_disks;
 451                 sb->nr_disks = info->array.nr_disks;
 452                 sb->active_disks = info->array.active_disks;
 453                 sb->working_disks = info->array.working_disks;
 454                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 455                 sb->disks[info->disk.number].number = info->disk.number;
 456                 sb->disks[info->disk.number].major = info->disk.major;
 457                 sb->disks[info->disk.number].minor = info->disk.minor;
 458                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 459                 sb->disks[info->disk.number].state = info->disk.state;
 460                 if (sb->this_disk.number == info->disk.number)
 461                         sb->this_disk = sb->disks[info->disk.number];
 462         }
 463         if (strcmp(update, "resync") == 0) {
 464                 /* make sure resync happens */
 465                 sb->state &= ~(1<<MD_SB_CLEAN);
 466                 sb->recovery_cp = 0;
 467         }
 468         if (strcmp(update, "homehost") == 0 &&
 469             homehost) {
 470                 uuid_set = 0;
 471                 update = "uuid";
 472                 info->uuid[0] = sb->set_uuid0;
 473                 info->uuid[1] = sb->set_uuid1;
 474         }
 475         if (strcmp(update, "uuid") == 0) {
 476                 if (!uuid_set && homehost) {
 477                         unsigned char *hash = SHA1((unsigned char*)homehost,
 478                                                    strlen(homehost),
 479                                                    NULL);
 480                         memcpy(info->uuid+2, hash, 8);
 481                 }
 482                 sb->set_uuid0 = info->uuid[0];
 483                 sb->set_uuid1 = info->uuid[1];
 484                 sb->set_uuid2 = info->uuid[2];
 485                 sb->set_uuid3 = info->uuid[3];
 486                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 487                         struct bitmap_super_s *bm;
 488                         bm = (struct bitmap_super_s*)(sb+1);
 489                         uuid_from_super0((int*)bm->uuid, sbv);
 490                 }
 491         }
 492         if (strcmp(update, "_reshape_progress")==0)
 493                 sb->reshape_position = info->reshape_progress;
 494
 495         sb->sb_csum = calc_sb0_csum(sb);
 496         return rv;
 497 }
 498
 499 static __u64 event_super0(void *sbv)
 500 {
 501         mdp_super_t *sb = sbv;
 502         return md_event(sb);
 503 }
 504
 505 /*
 506  * For verion-0 superblock, the homehost is 'stored' in the
 507  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 508  * of random material.
 509  * We use the first 8 bytes (64bits) of the sha1 of the
 510  * host name
 511  */
 512
 513
 514 static int init_super0(struct supertype *st, void **sbp, mdu_array_info_t *info,
 515                        unsigned long long size, char *ignored_name, char *homehost)
 516 {
 517         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 518         int spares;
 519         int rfd;
 520         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 521
 522         if (info->major_version == -1) {
 523                 /* zeroing the superblock */
 524                 *sbp = sb;
 525                 return 0;
 526         }
 527
 528         spares = info->working_disks - info->active_disks;
 529         if (info->raid_disks + spares  > MD_SB_DISKS) {
 530                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 531                         info->raid_disks , spares, MD_SB_DISKS);
 532                 return 0;
 533         }
 534
 535         rfd = open("/dev/urandom", O_RDONLY);
 536         sb->md_magic = MD_SB_MAGIC;
 537         sb->major_version = 0;
 538         sb->minor_version = 90;
 539         sb->patch_version = 0;
 540         sb->gvalid_words = 0; /* ignored */
 541         if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 542                 sb->set_uuid0 = random();
 543         sb->ctime = time(0);
 544         sb->level = info->level;
 545         if (size != info->size)
 546                 return 0;
 547         sb->size = info->size;
 548         sb->nr_disks = info->nr_disks;
 549         sb->raid_disks = info->raid_disks;
 550         sb->md_minor = info->md_minor;
 551         sb->not_persistent = 0;
 552         if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 553                 sb->set_uuid1 = random();
 554                 sb->set_uuid2 = random();
 555                 sb->set_uuid3 = random();
 556         }
 557         if (rfd >= 0)
 558                 close(rfd);
 559         if (homehost) {
 560                 unsigned char *hash = SHA1((unsigned char*)homehost,
 561                                            strlen(homehost),
 562                                            NULL);
 563                 memcpy(&sb->set_uuid2, hash, 8);
 564         }
 565
 566         sb->utime = sb->ctime;
 567         sb->state = info->state;
 568         sb->active_disks = info->active_disks;
 569         sb->working_disks = info->working_disks;
 570         sb->failed_disks = info->failed_disks;
 571         sb->spare_disks = info->spare_disks;
 572         sb->events_hi = 0;
 573         sb->events_lo = 1;
 574
 575         sb->layout = info->layout;
 576         sb->chunk_size = info->chunk_size;
 577
 578         *sbp = sb;
 579         return 1;
 580 }
 581
 582 /* Add a device to the superblock being created */
 583 static void add_to_super0(void *sbv, mdu_disk_info_t *dinfo)
 584 {
 585         mdp_super_t *sb = sbv;
 586         mdp_disk_t *dk = &sb->disks[dinfo->number];
 587
 588         dk->number = dinfo->number;
 589         dk->major = dinfo->major;
 590         dk->minor = dinfo->minor;
 591         dk->raid_disk = dinfo->raid_disk;
 592         dk->state = dinfo->state;
 593 }
 594
 595 static int store_super0(struct supertype *st, int fd, void *sbv)
 596 {
 597         unsigned long size;
 598         unsigned long long dsize;
 599         unsigned long long offset;
 600         mdp_super_t *super = sbv;
 601
 602 #ifdef BLKGETSIZE64
 603         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 604 #endif
 605         {
 606                 if (ioctl(fd, BLKGETSIZE, &size))
 607                         return 1;
 608                 else
 609                         dsize = ((unsigned long long)size)<<9;
 610         }
 611
 612         if (dsize < MD_RESERVED_SECTORS*2*512)
 613                 return 2;
 614
 615         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 616
 617         offset *= 512;
 618
 619         if (lseek64(fd, offset, 0)< 0LL)
 620                 return 3;
 621
 622         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 623                 return 4;
 624
 625         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 626                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 627                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 628                         if (write(fd, bm, sizeof(*bm)) != sizeof(*bm))
 629                             return 5;
 630         }
 631
 632         fsync(fd);
 633         return 0;
 634 }
 635
 636 static int write_init_super0(struct supertype *st, void *sbv, mdu_disk_info_t *dinfo, char *devname)
 637 {
 638         mdp_super_t *sb = sbv;
 639         int fd = open(devname, O_RDWR|O_EXCL);
 640         int rv;
 641
 642         if (fd < 0) {
 643                 fprintf(stderr, Name ": Failed to open %s to write superblock\n", devname);
 644                 return -1;
 645         }
 646
 647         sb->disks[dinfo->number].state &= ~(1<<MD_DISK_FAULTY);
 648         sb->disks[dinfo->number].state |= (1<<MD_DISK_SYNC);
 649
 650         sb->this_disk = sb->disks[dinfo->number];
 651         sb->sb_csum = calc_sb0_csum(sb);
 652         rv = store_super0(st, fd, sb);
 653
 654         if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 655                 rv = st->ss->write_bitmap(st, fd, sbv);
 656
 657         close(fd);
 658         if (rv)
 659                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
 660         return rv;
 661 }
 662
 663 static int compare_super0(void **firstp, void *secondv)
 664 {
 665         /*
 666          * return:
 667          *  0 same, or first was empty, and second was copied
 668          *  1 second had wrong number
 669          *  2 wrong uuid
 670          *  3 wrong other info
 671          */
 672         mdp_super_t *first = *firstp;
 673         mdp_super_t *second = secondv;
 674
 675         int uuid1[4], uuid2[4];
 676         if (second->md_magic != MD_SB_MAGIC)
 677                 return 1;
 678         if (!first) {
 679                 first = malloc(MD_SB_BYTES + sizeof(struct bitmap_super_s));
 680                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 681                 *firstp = first;
 682                 return 0;
 683         }
 684
 685         uuid_from_super0(uuid1, first);
 686         uuid_from_super0(uuid2, second);
 687         if (!same_uuid(uuid1, uuid2, 0))
 688                 return 2;
 689         if (first->major_version != second->major_version ||
 690             first->minor_version != second->minor_version ||
 691             first->patch_version != second->patch_version ||
 692             first->gvalid_words  != second->gvalid_words  ||
 693             first->ctime         != second->ctime         ||
 694             first->level         != second->level         ||
 695             first->size          != second->size          ||
 696             first->raid_disks    != second->raid_disks    )
 697                 return 3;
 698
 699         return 0;
 700 }
 701
 702
 703 static int load_super0(struct supertype *st, int fd, void **sbp, char *devname)
 704 {
 705         /* try to read in the superblock
 706          * Return:
 707          *  0 on success
 708          *  1 on cannot get superblock
 709          *  2 on superblock meaningless
 710          */
 711         unsigned long size;
 712         unsigned long long dsize;
 713         unsigned long long offset;
 714         mdp_super_t *super;
 715         int uuid[4];
 716         struct bitmap_super_s *bsb;
 717
 718 #ifdef BLKGETSIZE64
 719         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 720 #endif
 721         {
 722                 if (ioctl(fd, BLKGETSIZE, &size)) {
 723                         if (devname)
 724                                 fprintf(stderr, Name ": cannot find device size for %s: %s\n",
 725                                         devname, strerror(errno));
 726                         return 1;
 727                 } else
 728                         dsize = size << 9;
 729         }
 730
 731         if (dsize < MD_RESERVED_SECTORS*2) {
 732                 if (devname)
 733                         fprintf(stderr, Name ": %s is too small for md: size is %ld sectors.\n",
 734                                 devname, size);
 735                 return 1;
 736         }
 737
 738         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 739
 740         offset *= 512;
 741
 742         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 743
 744         if (lseek64(fd, offset, 0)< 0LL) {
 745                 if (devname)
 746                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 747                                 devname, strerror(errno));
 748                 return 1;
 749         }
 750
 751         super = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 752
 753         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 754                 if (devname)
 755                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 756                                 devname);
 757                 free(super);
 758                 return 1;
 759         }
 760
 761         if (st->ss && st->minor_version == 9)
 762                 super0_swap_endian(super);
 763
 764         if (super->md_magic != MD_SB_MAGIC) {
 765                 if (devname)
 766                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 767                                 devname, MD_SB_MAGIC, super->md_magic);
 768                 free(super);
 769                 return 2;
 770         }
 771
 772         if (super->major_version != 0) {
 773                 if (devname)
 774                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 775                                 devname, super->major_version);
 776                 free(super);
 777                 return 2;
 778         }
 779         *sbp = super;
 780         if (st->ss == NULL) {
 781                 st->ss = &super0;
 782                 st->minor_version = 90;
 783                 st->max_devs = MD_SB_DISKS;
 784         }
 785
 786         /* Now check on the bitmap superblock */
 787         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 788                 return 0;
 789         /* Read the bitmap superblock and make sure it looks
 790          * valid.  If it doesn't clear the bit.  An --assemble --force
 791          * should get that written out.
 792          */
 793         if (read(fd, super+1, sizeof(struct bitmap_super_s))
 794             != sizeof(struct bitmap_super_s))
 795                 goto no_bitmap;
 796
 797         uuid_from_super0(uuid, super);
 798         bsb = (struct bitmap_super_s *)(super+1);
 799         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 800             memcmp(bsb->uuid, uuid, 16) != 0)
 801                 goto no_bitmap;
 802         return 0;
 803
 804  no_bitmap:
 805         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 806
 807         return 0;
 808 }
 809
 810 static struct supertype *match_metadata_desc0(char *arg)
 811 {
 812         struct supertype *st = malloc(sizeof(*st));
 813         if (!st) return st;
 814
 815         st->ss = &super0;
 816         st->minor_version = 90;
 817         st->max_devs = MD_SB_DISKS;
 818         if (strcmp(arg, "0") == 0 ||
 819             strcmp(arg, "0.90") == 0 ||
 820             strcmp(arg, "default") == 0
 821                 )
 822                 return st;
 823
 824         st->minor_version = 9; /* flag for 'byte-swapped' */
 825         if (strcmp(arg, "0.swap")==0)
 826                 return st;
 827
 828         free(st);
 829         return NULL;
 830 }
 831
 832 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 833 {
 834         if (devsize < MD_RESERVED_SECTORS*2)
 835                 return 0ULL;
 836         return MD_NEW_SIZE_SECTORS(devsize);
 837 }
 838
 839 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)
 840 {
 841         /*
 842          * The bitmap comes immediately after the superblock and must be 60K in size
 843          * at most.  The default size is between 30K and 60K
 844          *
 845          * size is in sectors,  chunk is in bytes !!!
 846          */
 847         unsigned long long bits;
 848         unsigned long long max_bits = 60*1024*8;
 849         unsigned long long min_chunk;
 850         mdp_super_t *sb = sbv;
 851         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 852
 853
 854         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 855         bits = (size * 512) / min_chunk + 1;
 856         while (bits > max_bits) {
 857                 min_chunk *= 2;
 858                 bits = (bits+1)/2;
 859         }
 860         if (chunk == UnSet)
 861                 chunk = min_chunk;
 862         else if (chunk < min_chunk)
 863                 return 0; /* chunk size too small */
 864
 865         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 866
 867         memset(bms, 0, sizeof(*bms));
 868         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 869         bms->version = __cpu_to_le32(major);
 870         uuid_from_super0((int*)bms->uuid, sb);
 871         bms->chunksize = __cpu_to_le32(chunk);
 872         bms->daemon_sleep = __cpu_to_le32(delay);
 873         bms->sync_size = __cpu_to_le64(size);
 874         bms->write_behind = __cpu_to_le32(write_behind);
 875
 876         return 1;
 877 }
 878
 879
 880 void locate_bitmap0(struct supertype *st, int fd, void *sbv)
 881 {
 882         unsigned long long dsize;
 883         unsigned long size;
 884         unsigned long long offset;
 885 #ifdef BLKGETSIZE64
 886         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 887 #endif
 888         {
 889                 if (ioctl(fd, BLKGETSIZE, &size))
 890                         return;
 891                 else
 892                         dsize = ((unsigned long long)size)<<9;
 893         }
 894
 895         if (dsize < MD_RESERVED_SECTORS*2)
 896                 return;
 897
 898         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 899
 900         offset *= 512;
 901
 902         offset += MD_SB_BYTES;
 903
 904         lseek64(fd, offset, 0);
 905 }
 906
 907 int write_bitmap0(struct supertype *st, int fd, void *sbv)
 908 {
 909         unsigned long size;
 910         unsigned long long dsize;
 911         unsigned long long offset;
 912         mdp_super_t *sb = sbv;
 913
 914         int rv = 0;
 915
 916         int towrite, n;
 917         char buf[4096];
 918
 919 #ifdef BLKGETSIZE64
 920         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 921 #endif
 922         {
 923                 if (ioctl(fd, BLKGETSIZE, &size))
 924                         return 1;
 925                 else
 926                         dsize = ((unsigned long long)size)<<9;
 927         }
 928
 929         if (dsize < MD_RESERVED_SECTORS*2)
 930         return -1;
 931
 932         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 933
 934         offset *= 512;
 935
 936         if (lseek64(fd, offset + 4096, 0)< 0LL)
 937                 return 3;
 938
 939
 940         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
 941             sizeof(bitmap_super_t))
 942                 return -2;
 943         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
 944         memset(buf, 0xff, sizeof(buf));
 945         while (towrite > 0) {
 946                 n = towrite;
 947                 if (n > sizeof(buf))
 948                         n = sizeof(buf);
 949                 n = write(fd, buf, n);
 950                 if (n > 0)
 951                         towrite -= n;
 952                 else
 953                         break;
 954         }
 955         fsync(fd);
 956         if (towrite)
 957                 rv = -2;
 958
 959         return rv;
 960 }
 961
 962 struct superswitch super0 = {
 963 #ifndef MDASSEMBLE
 964         .examine_super = examine_super0,
 965         .brief_examine_super = brief_examine_super0,
 966         .detail_super = detail_super0,
 967         .brief_detail_super = brief_detail_super0,
 968 #endif
 969         .match_home = match_home0,
 970         .uuid_from_super = uuid_from_super0,
 971         .getinfo_super = getinfo_super0,
 972         .update_super = update_super0,
 973         .event_super = event_super0,
 974         .init_super = init_super0,
 975         .add_to_super = add_to_super0,
 976         .store_super = store_super0,
 977         .write_init_super = write_init_super0,
 978         .compare_super = compare_super0,
 979         .load_super = load_super0,
 980         .match_metadata_desc = match_metadata_desc0,
 981         .avail_size = avail_size0,
 982         .add_internal_bitmap = add_internal_bitmap0,
 983         .locate_bitmap = locate_bitmap0,
 984         .write_bitmap = write_bitmap0,
 985         .major = 0,
 986         .swapuuid = 0,
 987 };