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)
  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\n", sb->set_uuid0, sb->set_uuid1,
  99                        sb->set_uuid2, sb->set_uuid3);
 100         else
 101                 printf("           UUID : %08x\n", sb->set_uuid0);
 102
 103         atime = sb->ctime;
 104         printf("  Creation Time : %.24s\n", ctime(&atime));
 105         c=map_num(pers, sb->level);
 106         printf("     Raid Level : %s\n", c?c:"-unknown-");
 107         if ((int)sb->level >= 0) {
 108                 int ddsks=0;
 109                 printf("    Device Size : %d%s\n", sb->size, human_size((long long)sb->size<<10));
 110                 switch(sb->level) {
 111                 case 1: ddsks=1;break;
 112                 case 4:
 113                 case 5: ddsks = sb->raid_disks-1; break;
 114                 case 6: ddsks = sb->raid_disks-2; break;
 115                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 116                 }
 117                 if (ddsks)
 118                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 119                                human_size(ddsks*(long long)sb->size<<10));
 120         }
 121         printf("   Raid Devices : %d\n", sb->raid_disks);
 122         printf("  Total Devices : %d\n", sb->nr_disks);
 123         printf("Preferred Minor : %d\n", sb->md_minor);
 124         printf("\n");
 125         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 126                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 127                 if (sb->delta_disks) {
 128                         printf("  Delta Devices : %d", sb->delta_disks);
 129                         if (sb->delta_disks)
 130                                 printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 131                         else
 132                                 printf(" (%d->%d)\n", sb->raid_disks, sb->raid_disks+sb->delta_disks);
 133                 }
 134                 if (sb->new_level != sb->level) {
 135                         c = map_num(pers, sb->new_level);
 136                         printf("      New Level : %s\n", c?c:"-unknown-");
 137                 }
 138                 if (sb->new_layout != sb->layout) {
 139                         if (sb->level == 5) {
 140                                 c = map_num(r5layout, sb->new_layout);
 141                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 142                         }
 143                         if (sb->level == 10) {
 144                                 printf("     New Layout : near=%d, %s=%d\n",
 145                                        sb->new_layout&255,
 146                                        (sb->new_layout&0x10000)?"offset":"far",
 147                                        (sb->new_layout>>8)&255);
 148                         }
 149                 }
 150                 if (sb->new_chunk != sb->chunk_size)
 151                         printf("  New Chunksize : %d\n", sb->new_chunk);
 152                 printf("\n");
 153         }
 154         atime = sb->utime;
 155         printf("    Update Time : %.24s\n", ctime(&atime));
 156         printf("          State : %s\n",
 157                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 158         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 159                 printf("Internal Bitmap : present\n");
 160         printf(" Active Devices : %d\n", sb->active_disks);
 161         printf("Working Devices : %d\n", sb->working_disks);
 162         printf(" Failed Devices : %d\n", sb->failed_disks);
 163         printf("  Spare Devices : %d\n", sb->spare_disks);
 164         if (calc_sb0_csum(sb) == sb->sb_csum)
 165                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 166         else
 167                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 168         printf("         Events : %d.%d\n", sb->events_hi, sb->events_lo);
 169         printf("\n");
 170         if (sb->level == 5) {
 171                 c = map_num(r5layout, sb->layout);
 172                 printf("         Layout : %s\n", c?c:"-unknown-");
 173         }
 174         if (sb->level == 10) {
 175                 printf("         Layout : near=%d, %s=%d\n",
 176                        sb->layout&255,
 177                        (sb->layout&0x10000)?"offset":"far",
 178                        (sb->layout>>8)&255);
 179         }
 180         switch(sb->level) {
 181         case 0:
 182         case 4:
 183         case 5:
 184                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 185                 break;
 186         case -1:
 187                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 188                 break;
 189         default: break;
 190         }
 191         printf("\n");
 192         printf("      Number   Major   Minor   RaidDevice State\n");
 193         for (d= -1; d<(signed int)(sb->raid_disks+sb->spare_disks); d++) {
 194                 mdp_disk_t *dp;
 195                 char *dv;
 196                 char nb[5];
 197                 int wonly;
 198                 if (d>=0) dp = &sb->disks[d];
 199                 else dp = &sb->this_disk;
 200                 snprintf(nb, sizeof(nb), "%4d", d);
 201                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 202                        dp->number, dp->major, dp->minor, dp->raid_disk);
 203                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 204                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 205                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 206                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 207                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 208                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 209                 if (wonly) printf(" write-mostly");
 210                 if (dp->state == 0) printf(" spare");
 211                 if ((dv=map_dev(dp->major, dp->minor, 0)))
 212                         printf("   %s", dv);
 213                 printf("\n");
 214                 if (d == -1) printf("\n");
 215         }
 216 }
 217
 218 static void brief_examine_super0(void *sbv)
 219 {
 220         mdp_super_t *sb = sbv;
 221         char *c=map_num(pers, sb->level);
 222
 223         printf("ARRAY %s level=%s num-devices=%d UUID=",
 224                get_md_name(sb->md_minor),
 225                c?c:"-unknown-", sb->raid_disks);
 226         if (sb->minor_version >= 90)
 227                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 228                        sb->set_uuid2, sb->set_uuid3);
 229         else
 230                 printf("%08x", sb->set_uuid0);
 231         printf("\n");
 232 }
 233
 234 static void detail_super0(void *sbv, char *homehost)
 235 {
 236         mdp_super_t *sb = sbv;
 237         printf("           UUID : ");
 238         if (sb->minor_version >= 90)
 239                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 240                        sb->set_uuid2, sb->set_uuid3);
 241         else
 242                 printf("%08x", sb->set_uuid0);
 243         if (homehost) {
 244                 unsigned char *hash = SHA1((unsigned char *)homehost,
 245                                            strlen(homehost),
 246                                            NULL);
 247                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 248                         printf(" (local to host %s)", homehost);
 249         }
 250         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 251 }
 252
 253 static void brief_detail_super0(void *sbv)
 254 {
 255         mdp_super_t *sb = sbv;
 256         printf(" UUID=");
 257         if (sb->minor_version >= 90)
 258                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 259                        sb->set_uuid2, sb->set_uuid3);
 260         else
 261                 printf("%08x", sb->set_uuid0);
 262 }
 263 #endif
 264 static void uuid_from_super0(int uuid[4], void * sbv)
 265 {
 266         mdp_super_t *super = sbv;
 267         uuid[0] = super->set_uuid0;
 268         if (super->minor_version >= 90) {
 269                 uuid[1] = super->set_uuid1;
 270                 uuid[2] = super->set_uuid2;
 271                 uuid[3] = super->set_uuid3;
 272         } else {
 273                 uuid[1] = 0;
 274                 uuid[2] = 0;
 275                 uuid[3] = 0;
 276         }
 277 }
 278
 279 static void getinfo_super0(struct mdinfo *info, void *sbv)
 280 {
 281         mdp_super_t *sb = sbv;
 282         int working = 0;
 283         int i;
 284
 285         info->array.major_version = sb->major_version;
 286         info->array.minor_version = sb->minor_version;
 287         info->array.patch_version = sb->patch_version;
 288         info->array.raid_disks = sb->raid_disks;
 289         info->array.level = sb->level;
 290         info->array.layout = sb->layout;
 291         info->array.md_minor = sb->md_minor;
 292         info->array.ctime = sb->ctime;
 293         info->array.utime = sb->utime;
 294         info->array.chunk_size = sb->chunk_size;
 295         info->component_size = sb->size*2;
 296
 297         info->disk.state = sb->this_disk.state;
 298         info->disk.major = sb->this_disk.major;
 299         info->disk.minor = sb->this_disk.minor;
 300         info->disk.raid_disk = sb->this_disk.raid_disk;
 301         info->disk.number = sb->this_disk.number;
 302
 303         info->events = md_event(sb);
 304         info->data_offset = 0;
 305
 306         uuid_from_super0(info->uuid, sbv);
 307
 308         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 309                 info->reshape_active = 1;
 310                 info->reshape_progress = sb->reshape_position;
 311                 info->new_level = sb->new_level;
 312                 info->delta_disks = sb->delta_disks;
 313                 info->new_layout = sb->new_layout;
 314                 info->new_chunk = sb->new_chunk;
 315         } else
 316                 info->reshape_active = 0;
 317
 318         info->name[0] = 0;
 319         /* work_disks is calculated rather than read directly */
 320         for (i=0; i < MD_SB_DISKS; i++)
 321                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 322                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 323                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 324                         working ++;
 325         info->array.working_disks = working;
 326 }
 327
 328
 329 static int update_super0(struct mdinfo *info, void *sbv, char *update, char *devname, int verbose)
 330 {
 331         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 332          * For others, the return value is ignored.
 333          */
 334         int rv = 0;
 335         mdp_super_t *sb = sbv;
 336         if (strcmp(update, "sparc2.2")==0 ) {
 337                 /* 2.2 sparc put the events in the wrong place
 338                  * So we copy the tail of the superblock
 339                  * up 4 bytes before continuing
 340                  */
 341                 __u32 *sb32 = (__u32*)sb;
 342                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 343                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 344                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 345                 if (verbose >= 0)
 346                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 347                                  devname);
 348         }
 349         if (strcmp(update, "super-minor") ==0) {
 350                 sb->md_minor = info->array.md_minor;
 351                 if (verbose > 0)
 352                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 353                                 devname, info->array.md_minor);
 354         }
 355         if (strcmp(update, "summaries") == 0) {
 356                 int i;
 357                 /* set nr_disks, active_disks, working_disks,
 358                  * failed_disks, spare_disks based on disks[]
 359                  * array in superblock.
 360                  * Also make sure extra slots aren't 'failed'
 361                  */
 362                 sb->nr_disks = sb->active_disks =
 363                         sb->working_disks = sb->failed_disks =
 364                         sb->spare_disks = 0;
 365                 for (i=0; i < MD_SB_DISKS ; i++)
 366                         if (sb->disks[i].major ||
 367                             sb->disks[i].minor) {
 368                                 int state = sb->disks[i].state;
 369                                 if (state & (1<<MD_DISK_REMOVED))
 370                                         continue;
 371                                 sb->nr_disks++;
 372                                 if (state & (1<<MD_DISK_ACTIVE))
 373                                         sb->active_disks++;
 374                                 if (state & (1<<MD_DISK_FAULTY))
 375                                         sb->failed_disks++;
 376                                 else
 377                                         sb->working_disks++;
 378                                 if (state == 0)
 379                                         sb->spare_disks++;
 380                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 381                                 sb->disks[i].state = 0;
 382         }
 383         if (strcmp(update, "force")==0) {
 384                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 385                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 386                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 387                 if (sb->events_hi != ehi ||
 388                     sb->events_lo != elo)
 389                         rv = 1;
 390                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 391                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 392                         /* need to force clean */
 393                         sb->state |= (1 << MD_SB_CLEAN);
 394                         rv = 1;
 395                 }
 396         }
 397         if (strcmp(update, "assemble")==0) {
 398                 int d = info->disk.number;
 399                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 400                 if ((sb->disks[d].state & ~(1<<MD_DISK_WRITEMOSTLY))
 401                     != info->disk.state) {
 402                         sb->disks[d].state = info->disk.state | wonly;
 403                         rv = 1;
 404                 }
 405         }
 406         if (strcmp(update, "newdev") == 0) {
 407                 int d = info->disk.number;
 408                 memset(&sb->disks[d], 0, sizeof(sb->disks[d]));
 409                 sb->disks[d].number = d;
 410                 sb->disks[d].major = info->disk.major;
 411                 sb->disks[d].minor = info->disk.minor;
 412                 sb->disks[d].raid_disk = info->disk.raid_disk;
 413                 sb->disks[d].state = info->disk.state;
 414                 sb->this_disk = sb->disks[d];
 415         }
 416         if (strcmp(update, "grow") == 0) {
 417                 sb->raid_disks = info->array.raid_disks;
 418                 sb->nr_disks = info->array.nr_disks;
 419                 sb->active_disks = info->array.active_disks;
 420                 sb->working_disks = info->array.working_disks;
 421                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 422                 sb->disks[info->disk.number].number = info->disk.number;
 423                 sb->disks[info->disk.number].major = info->disk.major;
 424                 sb->disks[info->disk.number].minor = info->disk.minor;
 425                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 426                 sb->disks[info->disk.number].state = info->disk.state;
 427                 if (sb->this_disk.number == info->disk.number)
 428                         sb->this_disk = sb->disks[info->disk.number];
 429         }
 430         if (strcmp(update, "resync") == 0) {
 431                 /* make sure resync happens */
 432                 sb->state &= ~(1<<MD_SB_CLEAN);
 433                 sb->recovery_cp = 0;
 434         }
 435         if (strcmp(update, "uuid") == 0) {
 436                 sb->set_uuid0 = info->uuid[0];
 437                 sb->set_uuid1 = info->uuid[1];
 438                 sb->set_uuid2 = info->uuid[2];
 439                 sb->set_uuid3 = info->uuid[3];
 440                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 441                         struct bitmap_super_s *bm;
 442                         bm = (struct bitmap_super_s*)(sb+1);
 443                         uuid_from_super0((int*)bm->uuid, sbv);
 444                 }
 445         }
 446         if (strcmp(update, "_reshape_progress")==0)
 447                 sb->reshape_position = info->reshape_progress;
 448
 449         sb->sb_csum = calc_sb0_csum(sb);
 450         return rv;
 451 }
 452
 453 static __u64 event_super0(void *sbv)
 454 {
 455         mdp_super_t *sb = sbv;
 456         return md_event(sb);
 457 }
 458
 459 /*
 460  * For verion-0 superblock, the homehost is 'stored' in the
 461  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 462  * of random material.
 463  * We use the first 8 bytes (64bits) of the sha1 of the
 464  * host name
 465  */
 466
 467
 468 static int init_super0(struct supertype *st, void **sbp, mdu_array_info_t *info,
 469                        unsigned long long size, char *ignored_name, char *homehost)
 470 {
 471         mdp_super_t *sb = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 472         int spares;
 473         int rfd;
 474         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 475
 476         if (info->major_version == -1) {
 477                 /* zeroing the superblock */
 478                 *sbp = sb;
 479                 return 0;
 480         }
 481
 482         spares = info->working_disks - info->active_disks;
 483         if (info->raid_disks + spares  > MD_SB_DISKS) {
 484                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 485                         info->raid_disks , spares, MD_SB_DISKS);
 486                 return 0;
 487         }
 488
 489         rfd = open("/dev/urandom", O_RDONLY);
 490         sb->md_magic = MD_SB_MAGIC;
 491         sb->major_version = 0;
 492         sb->minor_version = 90;
 493         sb->patch_version = 0;
 494         sb->gvalid_words = 0; /* ignored */
 495         if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 496                 sb->set_uuid0 = random();
 497         sb->ctime = time(0);
 498         sb->level = info->level;
 499         if (size != info->size)
 500                 return 0;
 501         sb->size = info->size;
 502         sb->nr_disks = info->nr_disks;
 503         sb->raid_disks = info->raid_disks;
 504         sb->md_minor = info->md_minor;
 505         sb->not_persistent = 0;
 506         if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 507                 sb->set_uuid1 = random();
 508                 sb->set_uuid2 = random();
 509                 sb->set_uuid3 = random();
 510         }
 511         if (rfd >= 0)
 512                 close(rfd);
 513         if (homehost) {
 514                 unsigned char *hash = SHA1((unsigned char*)homehost,
 515                                            strlen(homehost),
 516                                            NULL);
 517                 memcpy(&sb->set_uuid2, hash, 8);
 518         }
 519
 520         sb->utime = sb->ctime;
 521         sb->state = info->state;
 522         sb->active_disks = info->active_disks;
 523         sb->working_disks = info->working_disks;
 524         sb->failed_disks = info->failed_disks;
 525         sb->spare_disks = info->spare_disks;
 526         sb->events_hi = 0;
 527         sb->events_lo = 1;
 528
 529         sb->layout = info->layout;
 530         sb->chunk_size = info->chunk_size;
 531
 532         *sbp = sb;
 533         return 1;
 534 }
 535
 536 /* Add a device to the superblock being created */
 537 static void add_to_super0(void *sbv, mdu_disk_info_t *dinfo)
 538 {
 539         mdp_super_t *sb = sbv;
 540         mdp_disk_t *dk = &sb->disks[dinfo->number];
 541
 542         dk->number = dinfo->number;
 543         dk->major = dinfo->major;
 544         dk->minor = dinfo->minor;
 545         dk->raid_disk = dinfo->raid_disk;
 546         dk->state = dinfo->state;
 547 }
 548
 549 static int store_super0(struct supertype *st, int fd, void *sbv)
 550 {
 551         unsigned long size;
 552         unsigned long long dsize;
 553         unsigned long long offset;
 554         mdp_super_t *super = sbv;
 555
 556 #ifdef BLKGETSIZE64
 557         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 558 #endif
 559         {
 560                 if (ioctl(fd, BLKGETSIZE, &size))
 561                         return 1;
 562                 else
 563                         dsize = ((unsigned long long)size)<<9;
 564         }
 565
 566         if (dsize < MD_RESERVED_SECTORS*2*512)
 567                 return 2;
 568
 569         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 570
 571         offset *= 512;
 572
 573         if (lseek64(fd, offset, 0)< 0LL)
 574                 return 3;
 575
 576         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 577                 return 4;
 578
 579         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 580                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 581                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 582                         write(fd, bm, sizeof(*bm));
 583         }
 584
 585         fsync(fd);
 586         return 0;
 587 }
 588
 589 static int write_init_super0(struct supertype *st, void *sbv, mdu_disk_info_t *dinfo, char *devname)
 590 {
 591         mdp_super_t *sb = sbv;
 592         int fd = open(devname, O_RDWR|O_EXCL);
 593         int rv;
 594
 595         if (fd < 0) {
 596                 fprintf(stderr, Name ": Failed to open %s to write superblock\n", devname);
 597                 return -1;
 598         }
 599
 600         sb->disks[dinfo->number].state &= ~(1<<MD_DISK_FAULTY);
 601         sb->disks[dinfo->number].state |= (1<<MD_DISK_SYNC);
 602
 603         sb->this_disk = sb->disks[dinfo->number];
 604         sb->sb_csum = calc_sb0_csum(sb);
 605         rv = store_super0(st, fd, sb);
 606
 607         if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 608                 rv = st->ss->write_bitmap(st, fd, sbv);
 609
 610         close(fd);
 611         if (rv)
 612                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
 613         return rv;
 614 }
 615
 616 static int compare_super0(void **firstp, void *secondv)
 617 {
 618         /*
 619          * return:
 620          *  0 same, or first was empty, and second was copied
 621          *  1 second had wrong number
 622          *  2 wrong uuid
 623          *  3 wrong other info
 624          */
 625         mdp_super_t *first = *firstp;
 626         mdp_super_t *second = secondv;
 627
 628         int uuid1[4], uuid2[4];
 629         if (second->md_magic != MD_SB_MAGIC)
 630                 return 1;
 631         if (!first) {
 632                 first = malloc(MD_SB_BYTES + sizeof(struct bitmap_super_s));
 633                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 634                 *firstp = first;
 635                 return 0;
 636         }
 637
 638         uuid_from_super0(uuid1, first);
 639         uuid_from_super0(uuid2, second);
 640         if (!same_uuid(uuid1, uuid2, 0))
 641                 return 2;
 642         if (first->major_version != second->major_version ||
 643             first->minor_version != second->minor_version ||
 644             first->patch_version != second->patch_version ||
 645             first->gvalid_words  != second->gvalid_words  ||
 646             first->ctime         != second->ctime         ||
 647             first->level         != second->level         ||
 648             first->size          != second->size          ||
 649             first->raid_disks    != second->raid_disks    )
 650                 return 3;
 651
 652         return 0;
 653 }
 654
 655
 656 static int load_super0(struct supertype *st, int fd, void **sbp, char *devname)
 657 {
 658         /* try to read in the superblock
 659          * Return:
 660          *  0 on success
 661          *  1 on cannot get superblock
 662          *  2 on superblock meaningless
 663          */
 664         unsigned long size;
 665         unsigned long long dsize;
 666         unsigned long long offset;
 667         mdp_super_t *super;
 668         int uuid[4];
 669         struct bitmap_super_s *bsb;
 670
 671 #ifdef BLKGETSIZE64
 672         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 673 #endif
 674         {
 675                 if (ioctl(fd, BLKGETSIZE, &size)) {
 676                         if (devname)
 677                                 fprintf(stderr, Name ": cannot find device size for %s: %s\n",
 678                                         devname, strerror(errno));
 679                         return 1;
 680                 } else
 681                         dsize = size << 9;
 682         }
 683
 684         if (dsize < MD_RESERVED_SECTORS*2) {
 685                 if (devname)
 686                         fprintf(stderr, Name ": %s is too small for md: size is %ld sectors.\n",
 687                                 devname, size);
 688                 return 1;
 689         }
 690
 691         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 692
 693         offset *= 512;
 694
 695         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 696
 697         if (lseek64(fd, offset, 0)< 0LL) {
 698                 if (devname)
 699                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 700                                 devname, strerror(errno));
 701                 return 1;
 702         }
 703
 704         super = malloc(MD_SB_BYTES + sizeof(bitmap_super_t));
 705
 706         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 707                 if (devname)
 708                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 709                                 devname);
 710                 free(super);
 711                 return 1;
 712         }
 713
 714         if (st->ss && st->minor_version == 9)
 715                 super0_swap_endian(super);
 716
 717         if (super->md_magic != MD_SB_MAGIC) {
 718                 if (devname)
 719                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 720                                 devname, MD_SB_MAGIC, super->md_magic);
 721                 free(super);
 722                 return 2;
 723         }
 724
 725         if (super->major_version != 0) {
 726                 if (devname)
 727                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 728                                 devname, super->major_version);
 729                 free(super);
 730                 return 2;
 731         }
 732         *sbp = super;
 733         if (st->ss == NULL) {
 734                 st->ss = &super0;
 735                 st->minor_version = 90;
 736                 st->max_devs = MD_SB_DISKS;
 737         }
 738
 739         /* Now check on the bitmap superblock */
 740         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 741                 return 0;
 742         /* Read the bitmap superblock and make sure it looks
 743          * valid.  If it doesn't clear the bit.  An --assemble --force
 744          * should get that written out.
 745          */
 746         if (read(fd, super+1, sizeof(struct bitmap_super_s))
 747             != sizeof(struct bitmap_super_s))
 748                 goto no_bitmap;
 749
 750         uuid_from_super0(uuid, super);
 751         bsb = (struct bitmap_super_s *)(super+1);
 752         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 753             memcmp(bsb->uuid, uuid, 16) != 0)
 754                 goto no_bitmap;
 755         return 0;
 756
 757  no_bitmap:
 758         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 759
 760         return 0;
 761 }
 762
 763 static struct supertype *match_metadata_desc0(char *arg)
 764 {
 765         struct supertype *st = malloc(sizeof(*st));
 766         if (!st) return st;
 767
 768         st->ss = &super0;
 769         st->minor_version = 90;
 770         st->max_devs = MD_SB_DISKS;
 771         if (strcmp(arg, "0") == 0 ||
 772             strcmp(arg, "0.90") == 0 ||
 773             strcmp(arg, "default") == 0
 774                 )
 775                 return st;
 776
 777         st->minor_version = 9; /* flag for 'byte-swapped' */
 778         if (strcmp(arg, "0.swap")==0)
 779                 return st;
 780
 781         free(st);
 782         return NULL;
 783 }
 784
 785 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 786 {
 787         if (devsize < MD_RESERVED_SECTORS*2)
 788                 return 0ULL;
 789         return MD_NEW_SIZE_SECTORS(devsize);
 790 }
 791
 792 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)
 793 {
 794         /*
 795          * The bitmap comes immediately after the superblock and must be 60K in size
 796          * at most.  The default size is between 30K and 60K
 797          *
 798          * size is in sectors,  chunk is in bytes !!!
 799          */
 800         unsigned long long bits;
 801         unsigned long long max_bits = 60*1024*8;
 802         unsigned long long min_chunk;
 803         mdp_super_t *sb = sbv;
 804         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 805
 806
 807         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 808         bits = (size * 512) / min_chunk + 1;
 809         while (bits > max_bits) {
 810                 min_chunk *= 2;
 811                 bits = (bits+1)/2;
 812         }
 813         if (chunk == UnSet)
 814                 chunk = min_chunk;
 815         else if (chunk < min_chunk)
 816                 return 0; /* chunk size too small */
 817
 818         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 819
 820         memset(bms, 0, sizeof(*bms));
 821         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 822         bms->version = __cpu_to_le32(major);
 823         uuid_from_super0((int*)bms->uuid, sb);
 824         bms->chunksize = __cpu_to_le32(chunk);
 825         bms->daemon_sleep = __cpu_to_le32(delay);
 826         bms->sync_size = __cpu_to_le64(size);
 827         bms->write_behind = __cpu_to_le32(write_behind);
 828
 829         return 1;
 830 }
 831
 832
 833 void locate_bitmap0(struct supertype *st, int fd, void *sbv)
 834 {
 835         unsigned long long dsize;
 836         unsigned long size;
 837         unsigned long long offset;
 838 #ifdef BLKGETSIZE64
 839         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 840 #endif
 841         {
 842                 if (ioctl(fd, BLKGETSIZE, &size))
 843                         return;
 844                 else
 845                         dsize = ((unsigned long long)size)<<9;
 846         }
 847
 848         if (dsize < MD_RESERVED_SECTORS*2)
 849                 return;
 850
 851         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 852
 853         offset *= 512;
 854
 855         offset += MD_SB_BYTES;
 856
 857         lseek64(fd, offset, 0);
 858 }
 859
 860 int write_bitmap0(struct supertype *st, int fd, void *sbv)
 861 {
 862         unsigned long size;
 863         unsigned long long dsize;
 864         unsigned long long offset;
 865         mdp_super_t *sb = sbv;
 866
 867         int rv = 0;
 868
 869         int towrite, n;
 870         char buf[4096];
 871
 872 #ifdef BLKGETSIZE64
 873         if (ioctl(fd, BLKGETSIZE64, &dsize) != 0)
 874 #endif
 875         {
 876                 if (ioctl(fd, BLKGETSIZE, &size))
 877                         return 1;
 878                 else
 879                         dsize = ((unsigned long long)size)<<9;
 880         }
 881
 882         if (dsize < MD_RESERVED_SECTORS*2)
 883         return -1;
 884
 885         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 886
 887         offset *= 512;
 888
 889         if (lseek64(fd, offset + 4096, 0)< 0LL)
 890                 return 3;
 891
 892
 893         if (write(fd, ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t)) !=
 894             sizeof(bitmap_super_t))
 895                 return -2;
 896         towrite = 64*1024 - MD_SB_BYTES - sizeof(bitmap_super_t);
 897         memset(buf, 0xff, sizeof(buf));
 898         while (towrite > 0) {
 899                 n = towrite;
 900                 if (n > sizeof(buf))
 901                         n = sizeof(buf);
 902                 n = write(fd, buf, n);
 903                 if (n > 0)
 904                         towrite -= n;
 905                 else
 906                         break;
 907         }
 908         fsync(fd);
 909         if (towrite)
 910                 rv = -2;
 911
 912         return rv;
 913 }
 914
 915 struct superswitch super0 = {
 916 #ifndef MDASSEMBLE
 917         .examine_super = examine_super0,
 918         .brief_examine_super = brief_examine_super0,
 919         .detail_super = detail_super0,
 920         .brief_detail_super = brief_detail_super0,
 921 #endif
 922         .uuid_from_super = uuid_from_super0,
 923         .getinfo_super = getinfo_super0,
 924         .update_super = update_super0,
 925         .event_super = event_super0,
 926         .init_super = init_super0,
 927         .add_to_super = add_to_super0,
 928         .store_super = store_super0,
 929         .write_init_super = write_init_super0,
 930         .compare_super = compare_super0,
 931         .load_super = load_super0,
 932         .match_metadata_desc = match_metadata_desc0,
 933         .avail_size = avail_size0,
 934         .add_internal_bitmap = add_internal_bitmap0,
 935         .locate_bitmap = locate_bitmap0,
 936         .write_bitmap = write_bitmap0,
 937         .major = 0,
 938         .swapuuid = 0,
 939 };