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