super1.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2004 Neil Brown <neilb@cse.unsw.edu.au>
   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 <endian.h>
  32 #include "asm/byteorder.h"
  33 /*
  34  * The version-1 superblock :
  35  * All numeric fields are little-endian.
  36  *
  37  * total size: 256 bytes plus 2 per device.
  38  *  1K allows 384 devices.
  39  */
  40 struct mdp_superblock_1 {
  41         /* constant array information - 128 bytes */
  42         __u32   magic;          /* MD_SB_MAGIC: 0xa92b4efc - little endian */
  43         __u32   major_version;  /* 1 */
  44         __u32   feature_map;    /* 0 for now */
  45         __u32   pad0;           /* always set to 0 when writing */
  46
  47         __u8    set_uuid[16];   /* user-space generated. */
  48         char    set_name[32];   /* set and interpreted by user-space */
  49
  50         __u64   ctime;          /* lo 40 bits are seconds, top 24 are microseconds or 0*/
  51         __u32   level;          /* -4 (multipath), -1 (linear), 0,1,4,5 */
  52         __u32   layout;         /* only for raid5 currently */
  53         __u64   size;           /* used size of component devices, in 512byte sectors */
  54
  55         __u32   chunksize;      /* in 512byte sectors */
  56         __u32   raid_disks;
  57         __u32   bitmap_offset;  /* sectors after start of superblock that bitmap starts
  58                                  * NOTE: signed, so bitmap can be before superblock
  59                                  * only meaningful of feature_map[0] is set.
  60                                  */
  61         __u8    pad1[128-100];  /* set to 0 when written */
  62
  63         /* constant this-device information - 64 bytes */
  64         __u64   data_offset;    /* sector start of data, often 0 */
  65         __u64   data_size;      /* sectors in this device that can be used for data */
  66         __u64   super_offset;   /* sector start of this superblock */
  67         __u64   recovery_offset;/* sectors before this offset (from data_offset) have been recovered */
  68         __u32   dev_number;     /* permanent identifier of this  device - not role in raid */
  69         __u32   cnt_corrected_read; /* number of read errors that were corrected by re-writing */
  70         __u8    device_uuid[16]; /* user-space setable, ignored by kernel */
  71         __u8    devflags;        /* per-device flags.  Only one defined...*/
  72 #define WriteMostly1    1        /* mask for writemostly flag in above */
  73         __u8    pad2[64-57];    /* set to 0 when writing */
  74
  75         /* array state information - 64 bytes */
  76         __u64   utime;          /* 40 bits second, 24 btes microseconds */
  77         __u64   events;         /* incremented when superblock updated */
  78         __u64   resync_offset;  /* data before this offset (from data_offset) known to be in sync */
  79         __u32   sb_csum;        /* checksum upto devs[max_dev] */
  80         __u32   max_dev;        /* size of devs[] array to consider */
  81         __u8    pad3[64-32];    /* set to 0 when writing */
  82
  83         /* device state information. Indexed by dev_number.
  84          * 2 bytes per device
  85          * Note there are no per-device state flags. State information is rolled
  86          * into the 'roles' value.  If a device is spare or faulty, then it doesn't
  87          * have a meaningful role.
  88          */
  89         __u16   dev_roles[0];   /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
  90 };
  91
  92 #ifndef offsetof
  93 #define offsetof(t,f) ((int)&(((t*)0)->f))
  94 #endif
  95 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
  96 {
  97         unsigned int disk_csum, csum;
  98         unsigned long long newcsum;
  99         int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2;
 100         unsigned int *isuper = (unsigned int*)sb;
 101         int i;
 102
 103 /* make sure I can count... */
 104         if (offsetof(struct mdp_superblock_1,data_offset) != 128 ||
 105             offsetof(struct mdp_superblock_1, utime) != 192 ||
 106             sizeof(struct mdp_superblock_1) != 256) {
 107                 fprintf(stderr, "WARNING - superblock isn't sized correctly\n");
 108         }
 109
 110         disk_csum = sb->sb_csum;
 111         sb->sb_csum = 0;
 112         newcsum = 0;
 113         for (i=0; size>=4; size -= 4 )
 114                 newcsum += __le32_to_cpu(*isuper++);
 115
 116         if (size == 2)
 117                 newcsum += __le16_to_cpu(*(unsigned short*) isuper);
 118
 119         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 120         sb->sb_csum = disk_csum;
 121         return csum;
 122 }
 123
 124 #ifndef MDASSEMBLE
 125 static void examine_super1(void *sbv)
 126 {
 127         struct mdp_superblock_1 *sb = sbv;
 128         time_t atime;
 129         int d;
 130         int faulty;
 131         int i;
 132         char *c;
 133
 134         printf("          Magic : %08x\n", __le32_to_cpu(sb->magic));
 135         printf("        Version : %02d.%02d\n", 1, __le32_to_cpu(sb->feature_map));
 136         printf("     Array UUID : ");
 137         for (i=0; i<16; i++) {
 138                 if ((i&3)==0 && i != 0) printf(":");
 139                 printf("%02x", sb->set_uuid[i]);
 140         }
 141         printf("\n");
 142         printf("           Name : %.32s\n", sb->set_name);
 143
 144         atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL;
 145         printf("  Creation Time : %.24s\n", ctime(&atime));
 146         c=map_num(pers, __le32_to_cpu(sb->level));
 147         printf("     Raid Level : %s\n", c?c:"-unknown-");
 148         printf("   Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks));
 149         printf("\n");
 150         printf("    Device Size : %llu%s\n", (unsigned long long)sb->data_size, human_size(sb->data_size<<9));
 151         if (sb->data_offset)
 152                 printf("    Data Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb->data_offset));
 153         if (sb->super_offset)
 154                 printf("   Super Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb->super_offset));
 155         printf("          State : %s\n", (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean");
 156         printf("    Device UUID : ");
 157         for (i=0; i<16; i++) {
 158                 if ((i&3)==0 && i != 0) printf(":");
 159                 printf("%02x", sb->device_uuid[i]);
 160         }
 161         printf("\n");
 162         if (sb->devflags) {
 163                 printf("      Flags :");
 164                 if (sb->devflags & WriteMostly1)
 165                         printf(" write-mostly");
 166                 printf("\n");
 167         }
 168
 169         atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL;
 170         printf("    Update Time : %.24s\n", ctime(&atime));
 171
 172         if (calc_sb_1_csum(sb) == sb->sb_csum)
 173                 printf("       Checksum : %x - correct\n", __le32_to_cpu(sb->sb_csum));
 174         else
 175                 printf("       Checksum : %x - expected %x\n", __le32_to_cpu(sb->sb_csum),
 176                        __le32_to_cpu(calc_sb_1_csum(sb)));
 177         printf("         Events : %llu\n", (unsigned long long)__le64_to_cpu(sb->events));
 178         printf("\n");
 179         if (__le32_to_cpu(sb->level) == 5) {
 180                 c = map_num(r5layout, __le32_to_cpu(sb->layout));
 181                 printf("         Layout : %s\n", c?c:"-unknown-");
 182         }
 183         if (__le32_to_cpu(sb->level) == 10) {
 184                 int lo = __le32_to_cpu(sb->layout);
 185                 printf("         Layout : near=%d, far=%d\n",
 186                        lo&255, (lo>>8)&255);
 187         }
 188         switch(__le32_to_cpu(sb->level)) {
 189         case 0:
 190         case 4:
 191         case 5:
 192                 printf("     Chunk Size : %dK\n", __le32_to_cpu(sb->chunksize/2));
 193                 break;
 194         case -1:
 195                 printf("       Rounding : %dK\n", __le32_to_cpu(sb->chunksize/2));
 196                 break;
 197         default: break;
 198         }
 199         printf("\n");
 200         printf("   Array State : ");
 201         for (d=0; d<__le32_to_cpu(sb->raid_disks); d++) {
 202                 int cnt = 0;
 203                 int me = 0;
 204                 int i;
 205                 for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
 206                         int role = __le16_to_cpu(sb->dev_roles[i]);
 207                         if (role == d) {
 208                                 if (i == __le32_to_cpu(sb->dev_number))
 209                                         me = 1;
 210                                 cnt++;
 211                         }
 212                 }
 213                 if (cnt > 1) printf("?");
 214                 else if (cnt == 1 && me) printf("U");
 215                 else if (cnt == 1) printf("u");
 216                 else printf ("_");
 217         }
 218         faulty = 0;
 219         for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
 220                 int role = __le16_to_cpu(sb->dev_roles[i]);
 221                 if (role == 0xFFFE)
 222                         faulty++;
 223         }
 224         if (faulty) printf(" %d failed", faulty);
 225         printf("\n");
 226 }
 227
 228
 229 static void brief_examine_super1(void *sbv)
 230 {
 231         struct mdp_superblock_1 *sb = sbv;
 232         int i;
 233
 234         char *c=map_num(pers, __le32_to_cpu(sb->level));
 235
 236         printf("ARRAY /dev/?? level=%s metadata=1 num-devices=%d UUID=",
 237                c?c:"-unknown-", sb->raid_disks);
 238         for (i=0; i<16; i++) {
 239                 printf("%02x", sb->set_uuid[i]);
 240                 if ((i&3)==0 && i != 0) printf(":");
 241         }
 242         if (sb->set_name[0])
 243                 printf(" name=%.32s", sb->set_name);
 244         printf("\n");
 245 }
 246
 247 static void detail_super1(void *sbv)
 248 {
 249         struct mdp_superblock_1 *sb = sbv;
 250         int i;
 251
 252         printf("           Name : %.32s\n", sb->set_name);
 253         printf("           UUID : ");
 254         for (i=0; i<16; i++) {
 255                 if ((i&3)==0 && i != 0) printf(":");
 256                 printf("%02x", sb->set_uuid[i]);
 257         }
 258         printf("\n         Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb->events));
 259 }
 260
 261 static void brief_detail_super1(void *sbv)
 262 {
 263         struct mdp_superblock_1 *sb = sbv;
 264         int i;
 265
 266         if (sb->set_name[0])
 267                 printf(" name=%.32s", sb->set_name);
 268         printf(" UUID=");
 269         for (i=0; i<16; i++) {
 270                 if ((i&3)==0 && i != 0) printf(":");
 271                 printf("%02x", sb->set_uuid[i]);
 272         }
 273 }
 274
 275 #endif
 276
 277 static void uuid_from_super1(int uuid[4], void * sbv)
 278 {
 279         struct mdp_superblock_1 *super = sbv;
 280         char *cuuid = (char*)uuid;
 281         int i;
 282         for (i=0; i<16; i++)
 283                 cuuid[i] = super->set_uuid[i];
 284 }
 285
 286 static void getinfo_super1(struct mdinfo *info, mddev_ident_t ident, void *sbv)
 287 {
 288         struct mdp_superblock_1 *sb = sbv;
 289         int working = 0;
 290         int i;
 291         int role;
 292
 293         info->array.major_version = 1;
 294         info->array.minor_version = __le32_to_cpu(sb->feature_map);
 295         info->array.patch_version = 0;
 296         info->array.raid_disks = __le32_to_cpu(sb->raid_disks);
 297         info->array.level = __le32_to_cpu(sb->level);
 298         info->array.layout = __le32_to_cpu(sb->layout);
 299         info->array.md_minor = -1;
 300         info->array.ctime = __le64_to_cpu(sb->ctime);
 301
 302         info->disk.major = 0;
 303         info->disk.minor = 0;
 304         info->disk.number = __le32_to_cpu(sb->dev_number);
 305         if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) ||
 306             __le32_to_cpu(sb->max_dev) > 512)
 307                 role = 0xfffe;
 308         else
 309                 role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]);
 310
 311         info->disk.raid_disk = -1;
 312         switch(role) {
 313         case 0xFFFF:
 314                 info->disk.state = 2; /* spare: ACTIVE, not sync, not faulty */
 315                 break;
 316         case 0xFFFE:
 317                 info->disk.state = 1; /* faulty */
 318                 break;
 319         default:
 320                 info->disk.state = 6; /* active and in sync */
 321                 info->disk.raid_disk = role;
 322         }
 323         info->events = __le64_to_cpu(sb->events);
 324
 325         memcpy(info->uuid, sb->set_uuid, 16);
 326
 327         strncpy(ident->name, sb->set_name, 32);
 328         ident->name[32] = 0;
 329
 330         for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
 331                 role = __le16_to_cpu(sb->dev_roles[i]);
 332                 if (/*role == 0xFFFF || */role < info->array.raid_disks)
 333                         working++;
 334         }
 335
 336         info->array.working_disks = working;
 337 }
 338
 339 static int update_super1(struct mdinfo *info, void *sbv, char *update, char *devname, int verbose)
 340 {
 341         int rv = 0;
 342         struct mdp_superblock_1 *sb = sbv;
 343
 344         if (strcmp(update, "force")==0) {
 345                 sb->events = __cpu_to_le64(info->events);
 346                 switch(__le32_to_cpu(sb->level)) {
 347                 case 5: case 4: case 6:
 348                         /* need to force clean */
 349                         sb->resync_offset = ~0ULL;
 350                 }
 351         }
 352         if (strcmp(update, "assemble")==0) {
 353                 int d = info->disk.number;
 354                 int want;
 355                 if (info->disk.state == 6)
 356                         want = __cpu_to_le32(info->disk.raid_disk);
 357                 else
 358                         want = 0xFFFF;
 359                 if (sb->dev_roles[d] != want) {
 360                         sb->dev_roles[d] = want;
 361                         rv = 1;
 362                 }
 363         }
 364 #if 0
 365         if (strcmp(update, "newdev") == 0) {
 366                 int d = info->disk.number;
 367                 memset(&sb->disks[d], 0, sizeof(sb->disks[d]));
 368                 sb->disks[d].number = d;
 369                 sb->disks[d].major = info->disk.major;
 370                 sb->disks[d].minor = info->disk.minor;
 371                 sb->disks[d].raid_disk = info->disk.raid_disk;
 372                 sb->disks[d].state = info->disk.state;
 373                 sb->this_disk = sb->disks[d];
 374         }
 375 #endif
 376         if (strcmp(update, "grow") == 0) {
 377                 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
 378                 /* FIXME */
 379         }
 380         if (strcmp(update, "resync") == 0) {
 381                 /* make sure resync happens */
 382                 sb->resync_offset = ~0ULL;
 383         }
 384
 385         sb->sb_csum = calc_sb_1_csum(sb);
 386         return rv;
 387 }
 388
 389
 390 static __u64 event_super1(void *sbv)
 391 {
 392         struct mdp_superblock_1 *sb = sbv;
 393         return __le64_to_cpu(sb->events);
 394 }
 395
 396 static int init_super1(struct supertype *st, void **sbp, mdu_array_info_t *info, char *name)
 397 {
 398         struct mdp_superblock_1 *sb = malloc(1024 + sizeof(bitmap_super_t));
 399         int spares;
 400         int rfd;
 401         memset(sb, 0, 1024);
 402
 403         if (info->major_version == -1)
 404                 /* zeroing superblock */
 405                 return 0;
 406
 407         spares = info->working_disks - info->active_disks;
 408         if (info->raid_disks + spares  > 384) {
 409                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 410                         info->raid_disks , spares, 384);
 411                 return 0;
 412         }
 413
 414
 415         sb->magic = __cpu_to_le32(MD_SB_MAGIC);
 416         sb->major_version = __cpu_to_le32(1);
 417         sb->feature_map = 0;
 418         sb->pad0 = 0;
 419
 420         if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
 421             read(rfd, sb->set_uuid, 16) != 16) {
 422                 *(__u32*)(sb->set_uuid) = random();
 423                 *(__u32*)(sb->set_uuid+4) = random();
 424                 *(__u32*)(sb->set_uuid+8) = random();
 425                 *(__u32*)(sb->set_uuid+12) = random();
 426         }
 427         if (rfd >= 0) close(rfd);
 428
 429         memset(sb->set_name, 0, 32);
 430         strcpy(sb->set_name, name);
 431
 432         sb->ctime = __cpu_to_le64((unsigned long long)time(0));
 433         sb->level = __cpu_to_le32(info->level);
 434         sb->layout = __cpu_to_le32(info->layout);
 435         sb->size = __cpu_to_le64(info->size*2ULL);
 436         sb->chunksize = __cpu_to_le32(info->chunk_size>>9);
 437         sb->raid_disks = __cpu_to_le32(info->raid_disks);
 438
 439         sb->data_offset = __cpu_to_le64(0);
 440         sb->data_size = __cpu_to_le64(0);
 441         sb->super_offset = __cpu_to_le64(0);
 442         sb->recovery_offset = __cpu_to_le64(0);
 443
 444         sb->utime = sb->ctime;
 445         sb->events = __cpu_to_le64(1);
 446         if (info->state & (1<<MD_SB_CLEAN))
 447                 sb->resync_offset = ~0ULL;
 448         else
 449                 sb->resync_offset = 0;
 450         sb->max_dev = __cpu_to_le32((1024- sizeof(struct mdp_superblock_1))/
 451                                     sizeof(sb->dev_roles[0]));
 452         memset(sb->pad3, 0, sizeof(sb->pad3));
 453
 454         memset(sb->dev_roles, 0xff, 1024 - sizeof(struct mdp_superblock_1));
 455
 456         *sbp = sb;
 457         return 1;
 458 }
 459
 460 /* Add a device to the superblock being created */
 461 static void add_to_super1(void *sbv, mdu_disk_info_t *dk)
 462 {
 463         struct mdp_superblock_1 *sb = sbv;
 464         __u16 *rp = sb->dev_roles + dk->number;
 465         if ((dk->state & 6) == 6) /* active, sync */
 466                 *rp = __cpu_to_le16(dk->raid_disk);
 467         else if ((dk->state & ~2) == 0) /* active or idle -> spare */
 468                 *rp = 0xffff;
 469         else
 470                 *rp = 0xfffe;
 471 }
 472
 473 static int store_super1(struct supertype *st, int fd, void *sbv)
 474 {
 475         struct mdp_superblock_1 *sb = sbv;
 476         unsigned long long sb_offset;
 477         int sbsize;
 478         long size;
 479
 480         if (ioctl(fd, BLKGETSIZE, &size))
 481                 return 1;
 482
 483
 484         if (size < 24)
 485                 return 2;
 486
 487         /*
 488          * Calculate the position of the superblock.
 489          * It is always aligned to a 4K boundary and
 490          * depending on minor_version, it can be:
 491          * 0: At least 8K, but less than 12K, from end of device
 492          * 1: At start of device
 493          * 2: 4K from start of device.
 494          */
 495         switch(st->minor_version) {
 496         case 0:
 497                 sb_offset = size;
 498                 sb_offset -= 8*2;
 499                 sb_offset &= ~(4*2-1);
 500                 break;
 501         case 1:
 502                 sb_offset = 0;
 503                 break;
 504         case 2:
 505                 sb_offset = 4*2;
 506                 break;
 507         default:
 508                 return -EINVAL;
 509         }
 510
 511
 512
 513         if (sb_offset != __le64_to_cpu(sb->super_offset) &&
 514             0 != __le64_to_cpu(sb->super_offset)
 515                 ) {
 516                 fprintf(stderr, Name ": internal error - sb_offset is wrong\n");
 517                 abort();
 518         }
 519
 520         if (lseek64(fd, sb_offset << 9, 0)< 0LL)
 521                 return 3;
 522
 523         sbsize = sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev);
 524
 525         if (write(fd, sb, sbsize) != sbsize)
 526                 return 4;
 527
 528         fsync(fd);
 529         return 0;
 530 }
 531
 532 static int load_super1(struct supertype *st, int fd, void **sbp, char *devname);
 533
 534 static int write_init_super1(struct supertype *st, void *sbv,
 535                              mdu_disk_info_t *dinfo, char *devname)
 536 {
 537         struct mdp_superblock_1 *sb = sbv;
 538         void *refsbv = NULL;
 539         int fd = open(devname, O_RDWR | O_EXCL);
 540         int rfd;
 541         int rv;
 542
 543         long size;
 544         long long sb_offset;
 545
 546
 547         if (fd < 0) {
 548                 fprintf(stderr, Name ": Failed to open %s to write superblock\n",
 549                         devname);
 550                 return -1;
 551         }
 552
 553         sb->dev_number = __cpu_to_le32(dinfo->number);
 554         if (dinfo->state & (1<<MD_DISK_WRITEMOSTLY))
 555                 sb->devflags |= WriteMostly1;
 556
 557         if ((rfd = open("/dev/urandom", O_RDONLY)) < 0 ||
 558             read(rfd, sb->device_uuid, 16) != 16) {
 559                 *(__u32*)(sb->device_uuid) = random();
 560                 *(__u32*)(sb->device_uuid+4) = random();
 561                 *(__u32*)(sb->device_uuid+8) = random();
 562                 *(__u32*)(sb->device_uuid+12) = random();
 563         }
 564         if (rfd >= 0) close(rfd);
 565         sb->events = 0;
 566
 567         if (load_super1(st, fd, &refsbv, NULL)==0) {
 568                 struct mdp_superblock_1 *refsb = refsbv;
 569
 570                 memcpy(sb->device_uuid, refsb->device_uuid, 16);
 571                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) {
 572                         /* same array, so preserve events and dev_number */
 573                         sb->events = refsb->events;
 574                         sb->dev_number = refsb->dev_number;
 575                 }
 576                 free(refsb);
 577         }
 578
 579         if (ioctl(fd, BLKGETSIZE, &size)) {
 580                 close(fd);
 581                 return 1;
 582         }
 583
 584         if (size < 24) {
 585                 close(fd);
 586                 return 2;
 587         }
 588
 589
 590         /*
 591          * Calculate the position of the superblock.
 592          * It is always aligned to a 4K boundary and
 593          * depending on minor_version, it can be:
 594          * 0: At least 8K, but less than 12K, from end of device
 595          * 1: At start of device
 596          * 2: 4K from start of device.
 597          */
 598         switch(st->minor_version) {
 599         case 0:
 600                 sb_offset = size;
 601                 sb_offset -= 8*2;
 602                 sb_offset &= ~(4*2-1);
 603                 sb->super_offset = __cpu_to_le64(sb_offset);
 604                 sb->data_offset = __cpu_to_le64(0);
 605                 sb->data_size = __cpu_to_le64(sb_offset);
 606                 break;
 607         case 1:
 608                 sb->super_offset = __cpu_to_le64(0);
 609                 sb->data_offset = __cpu_to_le64(4*2); /* leave 4k for super and bitmap */
 610                 sb->data_size = __cpu_to_le64(size - 4*2);
 611                 break;
 612         case 2:
 613                 sb_offset = 4*2;
 614                 sb->super_offset = __cpu_to_le64(sb_offset);
 615                 sb->data_offset = __cpu_to_le64(sb_offset+4*2);
 616                 sb->data_size = __cpu_to_le64(size - 4*2 - 4*2);
 617                 break;
 618         default:
 619                 return -EINVAL;
 620         }
 621
 622
 623         sb->sb_csum = calc_sb_1_csum(sb);
 624         rv = store_super1(st, fd, sb);
 625         if (rv)
 626                 fprintf(stderr, Name ": failed to write superblock to %s\n", devname);
 627
 628         if (rv == 0 && (__le32_to_cpu(sb->feature_map) & 1))
 629                 rv = st->ss->write_bitmap(st, fd, sbv);
 630         close(fd);
 631         return rv;
 632 }
 633
 634 static int compare_super1(void **firstp, void *secondv)
 635 {
 636         /*
 637          * return:
 638          *  0 same, or first was empty, and second was copied
 639          *  1 second had wrong number
 640          *  2 wrong uuid
 641          *  3 wrong other info
 642          */
 643         struct mdp_superblock_1 *first = *firstp;
 644         struct mdp_superblock_1 *second = secondv;
 645
 646         if (second->magic != __cpu_to_le32(MD_SB_MAGIC))
 647                 return 1;
 648         if (second->major_version != __cpu_to_le32(1))
 649                 return 1;
 650
 651         if (!first) {
 652                 first = malloc(1024);
 653                 memcpy(first, second, 1024);
 654                 *firstp = first;
 655                 return 0;
 656         }
 657         if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0)
 658                 return 2;
 659
 660         if (first->ctime      != second->ctime     ||
 661             first->level      != second->level     ||
 662             first->layout     != second->layout    ||
 663             first->size       != second->size      ||
 664             first->chunksize  != second->chunksize ||
 665             first->raid_disks != second->raid_disks)
 666                 return 3;
 667         return 0;
 668 }
 669
 670 static int load_super1(struct supertype *st, int fd, void **sbp, char *devname)
 671 {
 672         unsigned long size;
 673         unsigned long long sb_offset;
 674         struct mdp_superblock_1 *super;
 675
 676
 677
 678         if (st->ss == NULL) {
 679                 int bestvers = -1;
 680                 __u64 bestctime = 0;
 681                 /* guess... choose latest ctime */
 682                 st->ss = &super1;
 683                 for (st->minor_version = 0; st->minor_version <= 2 ; st->minor_version++) {
 684                         switch(load_super1(st, fd, sbp, devname)) {
 685                         case 0: super = *sbp;
 686                                 if (bestvers == -1 ||
 687                                     bestctime < __le64_to_cpu(super->ctime)) {
 688                                         bestvers = st->minor_version;
 689                                         bestctime = __le64_to_cpu(super->ctime);
 690                                 }
 691                                 free(super);
 692                                 *sbp = NULL;
 693                                 break;
 694                         case 1: st->ss = NULL; return 1; /*bad device */
 695                         case 2: break; /* bad, try next */
 696                         }
 697                 }
 698                 if (bestvers != -1) {
 699                         int rv;
 700                         st->minor_version = bestvers;
 701                         st->ss = &super1;
 702                         st->max_devs = 384;
 703                         rv = load_super1(st, fd, sbp, devname);
 704                         if (rv) st->ss = NULL;
 705                         return rv;
 706                 }
 707                 st->ss = NULL;
 708                 return 2;
 709         }
 710         if (ioctl(fd, BLKGETSIZE, &size)) {
 711                 if (devname)
 712                         fprintf(stderr, Name ": cannot find device size for %s: %s\n",
 713                                 devname, strerror(errno));
 714                 return 1;
 715         }
 716
 717         if (size < 24) {
 718                 if (devname)
 719                         fprintf(stderr, Name ": %s is too small for md: size is %lu sectors.\n",
 720                                 devname, size);
 721                 return 1;
 722         }
 723
 724         /*
 725          * Calculate the position of the superblock.
 726          * It is always aligned to a 4K boundary and
 727          * depeding on minor_version, it can be:
 728          * 0: At least 8K, but less than 12K, from end of device
 729          * 1: At start of device
 730          * 2: 4K from start of device.
 731          */
 732         switch(st->minor_version) {
 733         case 0:
 734                 sb_offset = size;
 735                 sb_offset -= 8*2;
 736                 sb_offset &= ~(4*2-1);
 737                 break;
 738         case 1:
 739                 sb_offset = 0;
 740                 break;
 741         case 2:
 742                 sb_offset = 4*2;
 743                 break;
 744         default:
 745                 return -EINVAL;
 746         }
 747
 748         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 749
 750
 751         if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
 752                 if (devname)
 753                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 754                                 devname, strerror(errno));
 755                 return 1;
 756         }
 757
 758         super = malloc(1024 + sizeof(bitmap_super_t));
 759
 760         if (read(fd, super, 1024) != 1024) {
 761                 if (devname)
 762                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 763                                 devname);
 764                 free(super);
 765                 return 1;
 766         }
 767
 768         if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) {
 769                 if (devname)
 770                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 771                                 devname, MD_SB_MAGIC, __le32_to_cpu(super->magic));
 772                 free(super);
 773                 return 2;
 774         }
 775
 776         if (__le32_to_cpu(super->major_version) != 1) {
 777                 if (devname)
 778                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 779                                 devname, __le32_to_cpu(super->major_version));
 780                 free(super);
 781                 return 2;
 782         }
 783         if (__le64_to_cpu(super->super_offset) != sb_offset) {
 784                 if (devname)
 785                         fprintf(stderr, Name ": No superblock found on %s (super_offset is wrong)\n",
 786                                 devname);
 787                 free(super);
 788                 return 2;
 789         }
 790         *sbp = super;
 791         return 0;
 792 }
 793
 794
 795 static struct supertype *match_metadata_desc1(char *arg)
 796 {
 797         struct supertype *st = malloc(sizeof(*st));
 798         if (!st) return st;
 799
 800         st->ss = &super1;
 801         st->max_devs = 384;
 802         if (strcmp(arg, "1") == 0 ||
 803             strcmp(arg, "1.0") == 0) {
 804                 st->minor_version = 0;
 805                 return st;
 806         }
 807         if (strcmp(arg, "1.1") == 0) {
 808                 st->minor_version = 1;
 809                 return st;
 810         }
 811         if (strcmp(arg, "1.2") == 0) {
 812                 st->minor_version = 2;
 813                 return st;
 814         }
 815
 816         free(st);
 817         return NULL;
 818 }
 819
 820 /* find available size on device with this devsize, using
 821  * superblock type st, and reserving 'reserve' sectors for
 822  * a possible bitmap
 823  */
 824 static __u64 avail_size1(struct supertype *st, __u64 devsize)
 825 {
 826         if (devsize < 24)
 827                 return 0;
 828
 829         switch(st->minor_version) {
 830         case 0:
 831                 /* at end */
 832                 return ((devsize - 8*2 ) & ~(4*2-1));
 833         case 1:
 834                 /* at start, 4K for superblock and possible bitmap */
 835                 return devsize - 4*2;
 836         case 2:
 837                 /* 4k from start, 4K for superblock and possible bitmap */
 838                 return devsize - (4+4)*2;
 839         }
 840         return 0;
 841 }
 842
 843 static int
 844 add_internal_bitmap1(struct supertype *st, void *sbv,
 845                      int chunk, int delay, int write_behind, int *sizep, int may_change, int major)
 846 {
 847         /*
 848          * If not may_change, then this is a 'Grow', and the bitmap
 849          * must fit after the superblock.
 850          * If may_change, then this is create, and we can put the bitmap
 851          * before the superblock if we like, or may move the start.
 852          * For now, just squeeze the bitmap into 3k and don't change anything.
 853          *
 854          * size is in K,  chunk is in bytes !!!
 855          */
 856
 857         unsigned long long size = *sizep;
 858         unsigned long long bits;
 859         unsigned long long max_bits = (3*512 - sizeof(bitmap_super_t)) * 8;
 860         unsigned long long min_chunk;
 861         struct mdp_superblock_1 *sb = sbv;
 862         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + 1024);
 863
 864         if (st->minor_version && !may_change &&
 865             __le64_to_cpu(sb->data_offset) - __le64_to_cpu(sb->super_offset) < 8)
 866                 return 0; /* doesn't fit */
 867
 868
 869
 870         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 871         bits = (size*1024)/min_chunk +1;
 872         while (bits > max_bits) {
 873                 min_chunk *= 2;
 874                 bits = (bits+1)/2;
 875         }
 876         if (chunk == UnSet)
 877                 chunk = min_chunk;
 878         else if (chunk < min_chunk)
 879                 return 0; /* chunk size too small */
 880
 881         sb->bitmap_offset = __cpu_to_le32(2);
 882
 883         sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map) | 1);
 884         memset(bms, sizeof(*bms), 0);
 885         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 886         bms->version = __cpu_to_le32(major);
 887         uuid_from_super1((int*)bms->uuid, sb);
 888         bms->chunksize = __cpu_to_le32(chunk);
 889         bms->daemon_sleep = __cpu_to_le32(delay);
 890         bms->sync_size = __cpu_to_le64(size<<1);
 891         bms->write_behind = __cpu_to_le32(write_behind);
 892
 893         return 1;
 894 }
 895
 896
 897 void locate_bitmap1(struct supertype *st, int fd, void *sbv)
 898 {
 899         unsigned long long offset;
 900         struct mdp_superblock_1 *sb;
 901
 902         if (!sbv)
 903                 if (st->ss->load_super(st, fd, sbv, NULL))
 904                         return; /* no error I hope... */
 905
 906         sb = sbv;
 907
 908         offset = __le64_to_cpu(sb->super_offset);
 909         offset += (long) __le32_to_cpu(sb->bitmap_offset);
 910         if (!sbv)
 911                 free(sb);
 912         lseek64(fd, offset<<9, 0);
 913 }
 914
 915 int write_bitmap1(struct supertype *st, int fd, void *sbv)
 916 {
 917         struct mdp_superblock_1 *sb = sbv;
 918         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+1024);
 919         int rv = 0;
 920
 921         int towrite, n;
 922         char buf[4096];
 923
 924         locate_bitmap1(st, fd, sbv);
 925
 926         write(fd, ((char*)sb)+1024, sizeof(bitmap_super_t));
 927         towrite = __le64_to_cpu(bms->sync_size) / (__le32_to_cpu(bms->chunksize)>>9);
 928         towrite = (towrite+7) >> 3; /* bits to bytes */
 929         memset(buf, 0xff, sizeof(buf));
 930         while (towrite > 0) {
 931                 n = towrite;
 932                 if (n > sizeof(buf))
 933                         n = sizeof(buf);
 934                 n = write(fd, buf, n);
 935                 if (n > 0)
 936                         towrite -= n;
 937                 else
 938                         break;
 939         }
 940         fsync(fd);
 941         if (towrite)
 942                 rv = -2;
 943
 944         return rv;
 945 }
 946
 947 struct superswitch super1 = {
 948 #ifndef MDASSEMBLE
 949         .examine_super = examine_super1,
 950         .brief_examine_super = brief_examine_super1,
 951         .detail_super = detail_super1,
 952         .brief_detail_super = brief_detail_super1,
 953 #endif
 954         .uuid_from_super = uuid_from_super1,
 955         .getinfo_super = getinfo_super1,
 956         .update_super = update_super1,
 957         .event_super = event_super1,
 958         .init_super = init_super1,
 959         .add_to_super = add_to_super1,
 960         .store_super = store_super1,
 961         .write_init_super = write_init_super1,
 962         .compare_super = compare_super1,
 963         .load_super = load_super1,
 964         .match_metadata_desc = match_metadata_desc1,
 965         .avail_size = avail_size1,
 966         .add_internal_bitmap = add_internal_bitmap1,
 967         .locate_bitmap = locate_bitmap1,
 968         .write_bitmap = write_bitmap1,
 969         .major = 1,
 970 #if __BYTE_ORDER == BIG_ENDIAN
 971         .swapuuid = 0,
 972 #else
 973         .swapuuid = 1,
 974 #endif
 975 };