super0.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2009 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@suse.de>
  23  */
  24
  25 #define HAVE_STDINT_H 1
  26 #include "mdadm.h"
  27 #include "sha1.h"
  28 /*
  29  * All handling for the 0.90.0 version superblock is in
  30  * this file.
  31  * This includes:
  32  *   - finding, loading, and writing the superblock.
  33  *   - initialising a new superblock
  34  *   - printing the superblock for --examine
  35  *   - printing part of the superblock for --detail
  36  * .. other stuff
  37  */
  38
  39
  40 static unsigned long calc_sb0_csum(mdp_super_t *super)
  41 {
  42         unsigned long csum = super->sb_csum;
  43         unsigned long newcsum;
  44         super->sb_csum= 0 ;
  45         newcsum = calc_csum(super, MD_SB_BYTES);
  46         super->sb_csum = csum;
  47         return newcsum;
  48 }
  49
  50
  51 static void super0_swap_endian(struct mdp_superblock_s *sb)
  52 {
  53         /* as super0 superblocks are host-endian, it is sometimes
  54          * useful to be able to swap the endianness
  55          * as (almost) everything is u32's we byte-swap every 4byte
  56          * number.
  57          * We then also have to swap the events_hi and events_lo
  58          */
  59         char *sbc = (char *)sb;
  60         __u32 t32;
  61         int i;
  62
  63         for (i=0; i < MD_SB_BYTES ; i+=4) {
  64                 char t = sbc[i];
  65                 sbc[i] = sbc[i+3];
  66                 sbc[i+3] = t;
  67                 t=sbc[i+1];
  68                 sbc[i+1]=sbc[i+2];
  69                 sbc[i+2]=t;
  70         }
  71         t32 = sb->events_hi;
  72         sb->events_hi = sb->events_lo;
  73         sb->events_lo = t32;
  74
  75         t32 = sb->cp_events_hi;
  76         sb->cp_events_hi = sb->cp_events_lo;
  77         sb->cp_events_lo = t32;
  78
  79 }
  80
  81 #ifndef MDASSEMBLE
  82
  83 static void examine_super0(struct supertype *st, char *homehost)
  84 {
  85         mdp_super_t *sb = st->sb;
  86         time_t atime;
  87         int d;
  88         int delta_extra = 0;
  89         char *c;
  90
  91         printf("          Magic : %08x\n", sb->md_magic);
  92         printf("        Version : %d.%02d.%02d\n", sb->major_version, sb->minor_version,
  93                sb->patch_version);
  94         if (sb->minor_version >= 90) {
  95                 printf("           UUID : %08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
  96                        sb->set_uuid2, sb->set_uuid3);
  97                 if (homehost) {
  98                         char buf[20];
  99                         void *hash = sha1_buffer(homehost,
 100                                                  strlen(homehost),
 101                                                  buf);
 102                         if (memcmp(&sb->set_uuid2, hash, 8)==0)
 103                                 printf(" (local to host %s)", homehost);
 104                 }
 105                 printf("\n");
 106         } else
 107                 printf("           UUID : %08x\n", sb->set_uuid0);
 108
 109         if (sb->not_persistent)
 110                 printf("           Eedk : not persistent\n");
 111
 112         atime = sb->ctime;
 113         printf("  Creation Time : %.24s\n", ctime(&atime));
 114         c=map_num(pers, sb->level);
 115         printf("     Raid Level : %s\n", c?c:"-unknown-");
 116         if ((int)sb->level >= 0) {
 117                 int ddsks=0;
 118                 printf("  Used Dev Size : %d%s\n", sb->size,
 119                        human_size((long long)sb->size<<10));
 120                 switch(sb->level) {
 121                 case 1: ddsks=1;break;
 122                 case 4:
 123                 case 5: ddsks = sb->raid_disks-1; break;
 124                 case 6: ddsks = sb->raid_disks-2; break;
 125                 case 10: ddsks = sb->raid_disks / (sb->layout&255) / ((sb->layout>>8)&255);
 126                 }
 127                 if (ddsks)
 128                         printf("     Array Size : %llu%s\n", (unsigned long long)ddsks * sb->size,
 129                                human_size(ddsks*(long long)sb->size<<10));
 130         }
 131         printf("   Raid Devices : %d\n", sb->raid_disks);
 132         printf("  Total Devices : %d\n", sb->nr_disks);
 133         printf("Preferred Minor : %d\n", sb->md_minor);
 134         printf("\n");
 135         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 136                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)sb->reshape_position/2, human_size((long long)sb->reshape_position<<9));
 137                 if (sb->delta_disks) {
 138                         printf("  Delta Devices : %d", sb->delta_disks);
 139                         printf(" (%d->%d)\n", sb->raid_disks-sb->delta_disks, sb->raid_disks);
 140                         if (((int)sb->delta_disks) < 0)
 141                                 delta_extra = - sb->delta_disks;
 142                 }
 143                 if (sb->new_level != sb->level) {
 144                         c = map_num(pers, sb->new_level);
 145                         printf("      New Level : %s\n", c?c:"-unknown-");
 146                 }
 147                 if (sb->new_layout != sb->layout) {
 148                         if (sb->level == 5) {
 149                                 c = map_num(r5layout, sb->new_layout);
 150                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 151                         }
 152                         if (sb->level == 6) {
 153                                 c = map_num(r6layout, sb->new_layout);
 154                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 155                         }
 156                         if (sb->level == 10) {
 157                                 printf("     New Layout : near=%d, %s=%d\n",
 158                                        sb->new_layout&255,
 159                                        (sb->new_layout&0x10000)?"offset":"far",
 160                                        (sb->new_layout>>8)&255);
 161                         }
 162                 }
 163                 if (sb->new_chunk != sb->chunk_size)
 164                         printf("  New Chunksize : %d\n", sb->new_chunk);
 165                 printf("\n");
 166         }
 167         atime = sb->utime;
 168         printf("    Update Time : %.24s\n", ctime(&atime));
 169         printf("          State : %s\n",
 170                (sb->state&(1<<MD_SB_CLEAN))?"clean":"active");
 171         if (sb->state & (1<<MD_SB_BITMAP_PRESENT))
 172                 printf("Internal Bitmap : present\n");
 173         printf(" Active Devices : %d\n", sb->active_disks);
 174         printf("Working Devices : %d\n", sb->working_disks);
 175         printf(" Failed Devices : %d\n", sb->failed_disks);
 176         printf("  Spare Devices : %d\n", sb->spare_disks);
 177         if (calc_sb0_csum(sb) == sb->sb_csum)
 178                 printf("       Checksum : %x - correct\n", sb->sb_csum);
 179         else
 180                 printf("       Checksum : %x - expected %lx\n", sb->sb_csum, calc_sb0_csum(sb));
 181         printf("         Events : %llu\n",
 182                ((unsigned long long)sb->events_hi << 32)
 183                + sb->events_lo);
 184         printf("\n");
 185         if (sb->level == 5) {
 186                 c = map_num(r5layout, sb->layout);
 187                 printf("         Layout : %s\n", c?c:"-unknown-");
 188         }
 189         if (sb->level == 6) {
 190                 c = map_num(r6layout, sb->layout);
 191                 printf("         Layout : %s\n", c?c:"-unknown-");
 192         }
 193         if (sb->level == 10) {
 194                 printf("         Layout :");
 195                 print_r10_layout(sb->layout);
 196                 printf("\n");
 197         }
 198         switch(sb->level) {
 199         case 0:
 200         case 4:
 201         case 5:
 202         case 6:
 203         case 10:
 204                 printf("     Chunk Size : %dK\n", sb->chunk_size/1024);
 205                 break;
 206         case -1:
 207                 printf("       Rounding : %dK\n", sb->chunk_size/1024);
 208                 break;
 209         default: break;
 210         }
 211         printf("\n");
 212         printf("      Number   Major   Minor   RaidDevice State\n");
 213         for (d= -1; d<(signed int)(sb->raid_disks+delta_extra + sb->spare_disks); d++) {
 214                 mdp_disk_t *dp;
 215                 char *dv;
 216                 char nb[5];
 217                 int wonly;
 218                 if (d>=0) dp = &sb->disks[d];
 219                 else dp = &sb->this_disk;
 220                 snprintf(nb, sizeof(nb), "%4d", d);
 221                 printf("%4s %5d   %5d    %5d    %5d     ", d < 0 ? "this" :  nb,
 222                        dp->number, dp->major, dp->minor, dp->raid_disk);
 223                 wonly = dp->state & (1<<MD_DISK_WRITEMOSTLY);
 224                 dp->state &= ~(1<<MD_DISK_WRITEMOSTLY);
 225                 if (dp->state & (1<<MD_DISK_FAULTY)) printf(" faulty");
 226                 if (dp->state & (1<<MD_DISK_ACTIVE)) printf(" active");
 227                 if (dp->state & (1<<MD_DISK_SYNC)) printf(" sync");
 228                 if (dp->state & (1<<MD_DISK_REMOVED)) printf(" removed");
 229                 if (wonly) printf(" write-mostly");
 230                 if (dp->state == 0) printf(" spare");
 231                 if ((dv=map_dev(dp->major, dp->minor, 0)))
 232                         printf("   %s", dv);
 233                 printf("\n");
 234                 if (d == -1) printf("\n");
 235         }
 236 }
 237
 238 static void brief_examine_super0(struct supertype *st, int verbose)
 239 {
 240         mdp_super_t *sb = st->sb;
 241         char *c=map_num(pers, sb->level);
 242         char devname[20];
 243
 244         sprintf(devname, "/dev/md%d", sb->md_minor);
 245
 246         if (verbose) {
 247                 printf("ARRAY %s level=%s num-devices=%d",
 248                        devname,
 249                        c?c:"-unknown-", sb->raid_disks);
 250         } else
 251                 printf("ARRAY %s", devname);
 252
 253         if (sb->minor_version >= 90)
 254                 printf(" UUID=%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 255                        sb->set_uuid2, sb->set_uuid3);
 256         else
 257                 printf(" UUID=%08x", sb->set_uuid0);
 258         printf("\n");
 259 }
 260
 261 static void export_examine_super0(struct supertype *st)
 262 {
 263         mdp_super_t *sb = st->sb;
 264
 265         printf("MD_LEVEL=%s\n", map_num(pers, sb->level));
 266         printf("MD_DEVICES=%d\n", sb->raid_disks);
 267         if (sb->minor_version >= 90)
 268                 printf("MD_UUID=%08x:%08x:%08x:%08x\n",
 269                        sb->set_uuid0, sb->set_uuid1,
 270                        sb->set_uuid2, sb->set_uuid3);
 271         else
 272                 printf("MD_UUID=%08x\n", sb->set_uuid0);
 273         printf("MD_UPDATE_TIME=%llu\n",
 274                __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL);
 275         printf("MD_EVENTS=%llu\n",
 276                ((unsigned long long)sb->events_hi << 32)
 277                + sb->events_lo);
 278 }
 279
 280 static void detail_super0(struct supertype *st, char *homehost)
 281 {
 282         mdp_super_t *sb = st->sb;
 283         printf("           UUID : ");
 284         if (sb->minor_version >= 90)
 285                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 286                        sb->set_uuid2, sb->set_uuid3);
 287         else
 288                 printf("%08x", sb->set_uuid0);
 289         if (homehost) {
 290                 char buf[20];
 291                 void *hash = sha1_buffer(homehost,
 292                                          strlen(homehost),
 293                                          buf);
 294                 if (memcmp(&sb->set_uuid2, hash, 8)==0)
 295                         printf(" (local to host %s)", homehost);
 296         }
 297         printf("\n         Events : %d.%d\n\n", sb->events_hi, sb->events_lo);
 298 }
 299
 300 static void brief_detail_super0(struct supertype *st)
 301 {
 302         mdp_super_t *sb = st->sb;
 303         printf(" UUID=");
 304         if (sb->minor_version >= 90)
 305                 printf("%08x:%08x:%08x:%08x", sb->set_uuid0, sb->set_uuid1,
 306                        sb->set_uuid2, sb->set_uuid3);
 307         else
 308                 printf("%08x", sb->set_uuid0);
 309 }
 310 #endif
 311
 312 static int match_home0(struct supertype *st, char *homehost)
 313 {
 314         mdp_super_t *sb = st->sb;
 315         char buf[20];
 316         char *hash;
 317
 318         if (!homehost)
 319                 return 0;
 320         hash = sha1_buffer(homehost,
 321                            strlen(homehost),
 322                            buf);
 323
 324         return (memcmp(&sb->set_uuid2, hash, 8)==0);
 325 }
 326
 327 static void uuid_from_super0(struct supertype *st, int uuid[4])
 328 {
 329         mdp_super_t *super = st->sb;
 330         uuid[0] = super->set_uuid0;
 331         if (super->minor_version >= 90) {
 332                 uuid[1] = super->set_uuid1;
 333                 uuid[2] = super->set_uuid2;
 334                 uuid[3] = super->set_uuid3;
 335         } else {
 336                 uuid[1] = 0;
 337                 uuid[2] = 0;
 338                 uuid[3] = 0;
 339         }
 340 }
 341
 342 static void getinfo_super0(struct supertype *st, struct mdinfo *info)
 343 {
 344         mdp_super_t *sb = st->sb;
 345         int working = 0;
 346         int i;
 347
 348         info->array.major_version = sb->major_version;
 349         info->array.minor_version = sb->minor_version;
 350         info->array.patch_version = sb->patch_version;
 351         info->array.raid_disks = sb->raid_disks;
 352         info->array.level = sb->level;
 353         info->array.layout = sb->layout;
 354         info->array.md_minor = sb->md_minor;
 355         info->array.ctime = sb->ctime;
 356         info->array.utime = sb->utime;
 357         info->array.chunk_size = sb->chunk_size;
 358         info->array.state = sb->state;
 359         info->component_size = sb->size*2;
 360
 361         info->disk.state = sb->this_disk.state;
 362         info->disk.major = sb->this_disk.major;
 363         info->disk.minor = sb->this_disk.minor;
 364         info->disk.raid_disk = sb->this_disk.raid_disk;
 365         info->disk.number = sb->this_disk.number;
 366
 367         info->events = md_event(sb);
 368         info->data_offset = 0;
 369
 370         sprintf(info->text_version, "0.%d", sb->minor_version);
 371         info->safe_mode_delay = 200;
 372
 373         uuid_from_super0(st, info->uuid);
 374
 375         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 376                 info->reshape_active = 1;
 377                 info->reshape_progress = sb->reshape_position;
 378                 info->new_level = sb->new_level;
 379                 info->delta_disks = sb->delta_disks;
 380                 info->new_layout = sb->new_layout;
 381                 info->new_chunk = sb->new_chunk;
 382                 if (info->delta_disks < 0)
 383                         info->array.raid_disks -= info->delta_disks;
 384         } else
 385                 info->reshape_active = 0;
 386
 387         sprintf(info->name, "%d", sb->md_minor);
 388         /* work_disks is calculated rather than read directly */
 389         for (i=0; i < MD_SB_DISKS; i++)
 390                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 391                     (sb->disks[i].raid_disk < info->array.raid_disks) &&
 392                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 393                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY)))
 394                         working ++;
 395         info->array.working_disks = working;
 396 }
 397
 398
 399 static int update_super0(struct supertype *st, struct mdinfo *info,
 400                          char *update,
 401                          char *devname, int verbose,
 402                          int uuid_set, char *homehost)
 403 {
 404         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 405          * For others, the return value is ignored.
 406          */
 407         int rv = 0;
 408         mdp_super_t *sb = st->sb;
 409         if (strcmp(update, "sparc2.2")==0 ) {
 410                 /* 2.2 sparc put the events in the wrong place
 411                  * So we copy the tail of the superblock
 412                  * up 4 bytes before continuing
 413                  */
 414                 __u32 *sb32 = (__u32*)sb;
 415                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 416                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 417                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 418                 if (verbose >= 0)
 419                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 420                                  devname);
 421         }
 422         if (strcmp(update, "super-minor") ==0) {
 423                 sb->md_minor = info->array.md_minor;
 424                 if (verbose > 0)
 425                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 426                                 devname, info->array.md_minor);
 427         }
 428         if (strcmp(update, "summaries") == 0) {
 429                 int i;
 430                 /* set nr_disks, active_disks, working_disks,
 431                  * failed_disks, spare_disks based on disks[]
 432                  * array in superblock.
 433                  * Also make sure extra slots aren't 'failed'
 434                  */
 435                 sb->nr_disks = sb->active_disks =
 436                         sb->working_disks = sb->failed_disks =
 437                         sb->spare_disks = 0;
 438                 for (i=0; i < MD_SB_DISKS ; i++)
 439                         if (sb->disks[i].major ||
 440                             sb->disks[i].minor) {
 441                                 int state = sb->disks[i].state;
 442                                 if (state & (1<<MD_DISK_REMOVED))
 443                                         continue;
 444                                 sb->nr_disks++;
 445                                 if (state & (1<<MD_DISK_ACTIVE))
 446                                         sb->active_disks++;
 447                                 if (state & (1<<MD_DISK_FAULTY))
 448                                         sb->failed_disks++;
 449                                 else
 450                                         sb->working_disks++;
 451                                 if (state == 0)
 452                                         sb->spare_disks++;
 453                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 454                                 sb->disks[i].state = 0;
 455         }
 456         if (strcmp(update, "force-one")==0) {
 457                 /* Not enough devices for a working array, so
 458                  * bring this one up-to-date.
 459                  */
 460                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 461                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 462                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 463                 if (sb->events_hi != ehi ||
 464                     sb->events_lo != elo)
 465                         rv = 1;
 466         }
 467         if (strcmp(update, "force-array")==0) {
 468                 /* degraded array and 'force' requested, so
 469                  * maybe need to mark it 'clean'
 470                  */
 471                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 472                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 473                         /* need to force clean */
 474                         sb->state |= (1 << MD_SB_CLEAN);
 475                         rv = 1;
 476                 }
 477         }
 478         if (strcmp(update, "assemble")==0) {
 479                 int d = info->disk.number;
 480                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 481                 int mask = (1<<MD_DISK_WRITEMOSTLY);
 482                 int add = 0;
 483                 if (sb->minor_version >= 91)
 484                         /* During reshape we don't insist on everything
 485                          * being marked 'sync'
 486                          */
 487                         add = (1<<MD_DISK_SYNC);
 488                 if (((sb->disks[d].state & ~mask) | add)
 489                     != info->disk.state) {
 490                         sb->disks[d].state = info->disk.state | wonly;
 491                         rv = 1;
 492                 }
 493         }
 494         if (strcmp(update, "linear-grow-new") == 0) {
 495                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 496                 sb->disks[info->disk.number].number = info->disk.number;
 497                 sb->disks[info->disk.number].major = info->disk.major;
 498                 sb->disks[info->disk.number].minor = info->disk.minor;
 499                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 500                 sb->disks[info->disk.number].state = info->disk.state;
 501                 sb->this_disk = sb->disks[info->disk.number];
 502         }
 503         if (strcmp(update, "linear-grow-update") == 0) {
 504                 sb->raid_disks = info->array.raid_disks;
 505                 sb->nr_disks = info->array.nr_disks;
 506                 sb->active_disks = info->array.active_disks;
 507                 sb->working_disks = info->array.working_disks;
 508                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 509                 sb->disks[info->disk.number].number = info->disk.number;
 510                 sb->disks[info->disk.number].major = info->disk.major;
 511                 sb->disks[info->disk.number].minor = info->disk.minor;
 512                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 513                 sb->disks[info->disk.number].state = info->disk.state;
 514         }
 515         if (strcmp(update, "resync") == 0) {
 516                 /* make sure resync happens */
 517                 sb->state &= ~(1<<MD_SB_CLEAN);
 518                 sb->recovery_cp = 0;
 519         }
 520         if (strcmp(update, "homehost") == 0 &&
 521             homehost) {
 522                 uuid_set = 0;
 523                 update = "uuid";
 524                 info->uuid[0] = sb->set_uuid0;
 525                 info->uuid[1] = sb->set_uuid1;
 526         }
 527         if (strcmp(update, "uuid") == 0) {
 528                 if (!uuid_set && homehost) {
 529                         char buf[20];
 530                         char *hash = sha1_buffer(homehost,
 531                                                  strlen(homehost),
 532                                                  buf);
 533                         memcpy(info->uuid+2, hash, 8);
 534                 }
 535                 sb->set_uuid0 = info->uuid[0];
 536                 sb->set_uuid1 = info->uuid[1];
 537                 sb->set_uuid2 = info->uuid[2];
 538                 sb->set_uuid3 = info->uuid[3];
 539                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 540                         struct bitmap_super_s *bm;
 541                         bm = (struct bitmap_super_s*)(sb+1);
 542                         uuid_from_super0(st, (int*)bm->uuid);
 543                 }
 544         }
 545         if (strcmp(update, "_reshape_progress")==0)
 546                 sb->reshape_position = info->reshape_progress;
 547
 548         sb->sb_csum = calc_sb0_csum(sb);
 549         return rv;
 550 }
 551
 552 /*
 553  * For verion-0 superblock, the homehost is 'stored' in the
 554  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 555  * of random material.
 556  * We use the first 8 bytes (64bits) of the sha1 of the
 557  * host name
 558  */
 559
 560
 561 static int init_super0(struct supertype *st, mdu_array_info_t *info,
 562                        unsigned long long size, char *ignored_name, char *homehost,
 563                        int *uuid)
 564 {
 565         mdp_super_t *sb;
 566         int spares;
 567
 568         if (posix_memalign((void**)&sb, 4096,
 569                            MD_SB_BYTES + ROUND_UP(sizeof(bitmap_super_t), 4096)) != 0) {
 570                 fprintf(stderr, Name ": %s could not allocate superblock\n", __func__);
 571                 return 0;
 572         }
 573         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 574
 575         st->sb = sb;
 576         if (info == NULL) {
 577                 /* zeroing the superblock */
 578                 return 0;
 579         }
 580
 581         spares = info->working_disks - info->active_disks;
 582         if (info->raid_disks + spares  > MD_SB_DISKS) {
 583                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 584                         info->raid_disks , spares, MD_SB_DISKS);
 585                 return 0;
 586         }
 587
 588         sb->md_magic = MD_SB_MAGIC;
 589         sb->major_version = 0;
 590         sb->minor_version = 90;
 591         sb->patch_version = 0;
 592         sb->gvalid_words = 0; /* ignored */
 593         sb->ctime = time(0);
 594         sb->level = info->level;
 595         if (size != info->size)
 596                 return 0;
 597         sb->size = info->size;
 598         sb->nr_disks = info->nr_disks;
 599         sb->raid_disks = info->raid_disks;
 600         sb->md_minor = info->md_minor;
 601         sb->not_persistent = 0;
 602         if (uuid) {
 603                 sb->set_uuid0 = uuid[0];
 604                 sb->set_uuid1 = uuid[1];
 605                 sb->set_uuid2 = uuid[2];
 606                 sb->set_uuid3 = uuid[3];
 607         } else {
 608                 int rfd = open("/dev/urandom", O_RDONLY);
 609                 if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 610                         sb->set_uuid0 = random();
 611                 if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 612                         sb->set_uuid1 = random();
 613                         sb->set_uuid2 = random();
 614                         sb->set_uuid3 = random();
 615                 }
 616                 if (rfd >= 0)
 617                         close(rfd);
 618         }
 619         if (homehost) {
 620                 char buf[20];
 621                 char *hash = sha1_buffer(homehost,
 622                                          strlen(homehost),
 623                                          buf);
 624                 memcpy(&sb->set_uuid2, hash, 8);
 625         }
 626
 627         sb->utime = sb->ctime;
 628         sb->state = info->state;
 629         sb->active_disks = info->active_disks;
 630         sb->working_disks = info->working_disks;
 631         sb->failed_disks = info->failed_disks;
 632         sb->spare_disks = info->spare_disks;
 633         sb->events_hi = 0;
 634         sb->events_lo = 1;
 635
 636         sb->layout = info->layout;
 637         sb->chunk_size = info->chunk_size;
 638
 639         return 1;
 640 }
 641
 642 struct devinfo {
 643         int fd;
 644         char *devname;
 645         mdu_disk_info_t disk;
 646         struct devinfo *next;
 647 };
 648
 649 #ifndef MDASSEMBLE
 650 /* Add a device to the superblock being created */
 651 static int add_to_super0(struct supertype *st, mdu_disk_info_t *dinfo,
 652                           int fd, char *devname)
 653 {
 654         mdp_super_t *sb = st->sb;
 655         mdp_disk_t *dk = &sb->disks[dinfo->number];
 656         struct devinfo *di, **dip;
 657
 658         dk->number = dinfo->number;
 659         dk->major = dinfo->major;
 660         dk->minor = dinfo->minor;
 661         dk->raid_disk = dinfo->raid_disk;
 662         dk->state = dinfo->state;
 663
 664         sb->this_disk = sb->disks[dinfo->number];
 665         sb->sb_csum = calc_sb0_csum(sb);
 666
 667         dip = (struct devinfo **)&st->info;
 668         while (*dip)
 669                 dip = &(*dip)->next;
 670         di = malloc(sizeof(struct devinfo));
 671         di->fd = fd;
 672         di->devname = devname;
 673         di->disk = *dinfo;
 674         di->next = NULL;
 675         *dip = di;
 676
 677         return 0;
 678 }
 679 #endif
 680
 681 static int store_super0(struct supertype *st, int fd)
 682 {
 683         unsigned long long dsize;
 684         unsigned long long offset;
 685         mdp_super_t *super = st->sb;
 686
 687         if (!get_dev_size(fd, NULL, &dsize))
 688                 return 1;
 689
 690         if (dsize < MD_RESERVED_SECTORS*512)
 691                 return 2;
 692
 693         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 694
 695         offset *= 512;
 696
 697         if (lseek64(fd, offset, 0)< 0LL)
 698                 return 3;
 699
 700         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 701                 return 4;
 702
 703         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 704                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 705                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 706                         if (write(fd, bm, ROUND_UP(sizeof(*bm),4096)) !=
 707                             ROUND_UP(sizeof(*bm),4096))
 708                             return 5;
 709         }
 710
 711         fsync(fd);
 712         return 0;
 713 }
 714
 715 #ifndef MDASSEMBLE
 716 static int write_init_super0(struct supertype *st)
 717 {
 718         mdp_super_t *sb = st->sb;
 719         int rv = 0;
 720         struct devinfo *di;
 721
 722         for (di = st->info ; di && ! rv ; di = di->next) {
 723
 724                 if (di->disk.state == 1)
 725                         continue;
 726                 if (di->fd == -1)
 727                         continue;
 728                 Kill(di->devname, 0, 1, 1);
 729                 Kill(di->devname, 0, 1, 1);
 730
 731                 sb->disks[di->disk.number].state &= ~(1<<MD_DISK_FAULTY);
 732
 733                 sb->this_disk = sb->disks[di->disk.number];
 734                 sb->sb_csum = calc_sb0_csum(sb);
 735                 rv = store_super0(st, di->fd);
 736
 737                 if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 738                         rv = st->ss->write_bitmap(st, di->fd);
 739
 740                 if (rv)
 741                         fprintf(stderr,
 742                                 Name ": failed to write superblock to %s\n",
 743                                 di->devname);
 744                 close(di->fd);
 745                 di->fd = -1;
 746         }
 747         return rv;
 748 }
 749 #endif
 750
 751 static int compare_super0(struct supertype *st, struct supertype *tst)
 752 {
 753         /*
 754          * return:
 755          *  0 same, or first was empty, and second was copied
 756          *  1 second had wrong number
 757          *  2 wrong uuid
 758          *  3 wrong other info
 759          */
 760         mdp_super_t *first = st->sb;
 761         mdp_super_t *second = tst->sb;
 762         int uuid1[4], uuid2[4];
 763
 764         if (second->md_magic != MD_SB_MAGIC)
 765                 return 1;
 766         if (!first) {
 767                 if (posix_memalign((void**)&first, 4096,
 768                              MD_SB_BYTES +
 769                              ROUND_UP(sizeof(struct bitmap_super_s), 4096)) != 0) {
 770                         fprintf(stderr, Name
 771                                 ": %s could not allocate superblock\n", __func__);
 772                         return 1;
 773                 }
 774                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 775                 st->sb = first;
 776                 return 0;
 777         }
 778
 779         uuid_from_super0(st, uuid1);
 780         uuid_from_super0(tst, uuid2);
 781         if (!same_uuid(uuid1, uuid2, 0))
 782                 return 2;
 783         if (first->major_version != second->major_version ||
 784             first->minor_version != second->minor_version ||
 785             first->patch_version != second->patch_version ||
 786             first->gvalid_words  != second->gvalid_words  ||
 787             first->ctime         != second->ctime         ||
 788             first->level         != second->level         ||
 789             first->size          != second->size          ||
 790             first->raid_disks    != second->raid_disks    )
 791                 return 3;
 792
 793         return 0;
 794 }
 795
 796
 797 static void free_super0(struct supertype *st);
 798
 799 static int load_super0(struct supertype *st, int fd, char *devname)
 800 {
 801         /* try to read in the superblock
 802          * Return:
 803          *  0 on success
 804          *  1 on cannot get superblock
 805          *  2 on superblock meaningless
 806          */
 807         unsigned long long dsize;
 808         unsigned long long offset;
 809         mdp_super_t *super;
 810         int uuid[4];
 811         struct bitmap_super_s *bsb;
 812
 813         free_super0(st);
 814
 815         if (st->subarray[0])
 816                 return 1;
 817
 818         if (!get_dev_size(fd, devname, &dsize))
 819                 return 1;
 820
 821         if (dsize < MD_RESERVED_SECTORS*512) {
 822                 if (devname)
 823                         fprintf(stderr, Name
 824                             ": %s is too small for md: size is %llu sectors.\n",
 825                                 devname, dsize);
 826                 return 1;
 827         }
 828
 829         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 830
 831         offset *= 512;
 832
 833         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 834
 835         if (lseek64(fd, offset, 0)< 0LL) {
 836                 if (devname)
 837                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 838                                 devname, strerror(errno));
 839                 return 1;
 840         }
 841
 842         if (posix_memalign((void**)&super, 4096,
 843                            MD_SB_BYTES +
 844                            ROUND_UP(sizeof(bitmap_super_t), 4096)) != 0) {
 845                 fprintf(stderr, Name
 846                         ": %s could not allocate superblock\n", __func__);
 847                 return 1;
 848         }
 849
 850         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 851                 if (devname)
 852                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 853                                 devname);
 854                 free(super);
 855                 return 1;
 856         }
 857
 858         if (st->ss && st->minor_version == 9)
 859                 super0_swap_endian(super);
 860
 861         if (super->md_magic != MD_SB_MAGIC) {
 862                 if (devname)
 863                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 864                                 devname, MD_SB_MAGIC, super->md_magic);
 865                 free(super);
 866                 return 2;
 867         }
 868
 869         if (super->major_version != 0) {
 870                 if (devname)
 871                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 872                                 devname, super->major_version);
 873                 free(super);
 874                 return 2;
 875         }
 876         st->sb = super;
 877
 878         if (st->ss == NULL) {
 879                 st->ss = &super0;
 880                 st->minor_version = super->minor_version;
 881                 st->max_devs = MD_SB_DISKS;
 882                 st->info = NULL;
 883         }
 884
 885         /* Now check on the bitmap superblock */
 886         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 887                 return 0;
 888         /* Read the bitmap superblock and make sure it looks
 889          * valid.  If it doesn't clear the bit.  An --assemble --force
 890          * should get that written out.
 891          */
 892         if (read(fd, super+1, ROUND_UP(sizeof(struct bitmap_super_s),4096))
 893             != ROUND_UP(sizeof(struct bitmap_super_s),4096))
 894                 goto no_bitmap;
 895
 896         uuid_from_super0(st, uuid);
 897         bsb = (struct bitmap_super_s *)(super+1);
 898         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 899             memcmp(bsb->uuid, uuid, 16) != 0)
 900                 goto no_bitmap;
 901         return 0;
 902
 903  no_bitmap:
 904         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 905
 906         return 0;
 907 }
 908
 909 static struct supertype *match_metadata_desc0(char *arg)
 910 {
 911         struct supertype *st = malloc(sizeof(*st));
 912         if (!st) return st;
 913
 914         memset(st, 0, sizeof(*st));
 915         st->ss = &super0;
 916         st->info = NULL;
 917         st->minor_version = 90;
 918         st->max_devs = MD_SB_DISKS;
 919         st->sb = NULL;
 920         /* we sometimes get 00.90 */
 921         while (arg[0] == '0' && arg[1] == '0')
 922                 arg++;
 923         if (strcmp(arg, "0") == 0 ||
 924             strcmp(arg, "0.90") == 0 ||
 925             strcmp(arg, "default") == 0 ||
 926             strcmp(arg, "") == 0 /* no metadata */
 927                 )
 928                 return st;
 929
 930         st->minor_version = 91; /* reshape in progress */
 931         if (strcmp(arg, "0.91") == 0) /* For dup_super support */
 932                 return st;
 933
 934         st->minor_version = 9; /* flag for 'byte-swapped' */
 935         if (strcmp(arg, "0.swap")==0 ||
 936             strcmp(arg, "0.9") == 0) /* For dup_super support */
 937                 return st;
 938
 939         free(st);
 940         return NULL;
 941 }
 942
 943 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 944 {
 945         if (devsize < MD_RESERVED_SECTORS)
 946                 return 0ULL;
 947         return MD_NEW_SIZE_SECTORS(devsize);
 948 }
 949
 950 static int add_internal_bitmap0(struct supertype *st, int *chunkp,
 951                                 int delay, int write_behind,
 952                                 unsigned long long size, int may_change,
 953                                 int major)
 954 {
 955         /*
 956          * The bitmap comes immediately after the superblock and must be 60K in size
 957          * at most.  The default size is between 30K and 60K
 958          *
 959          * size is in sectors,  chunk is in bytes !!!
 960          */
 961         unsigned long long bits;
 962         unsigned long long max_bits = 60*1024*8;
 963         unsigned long long min_chunk;
 964         int chunk = *chunkp;
 965         mdp_super_t *sb = st->sb;
 966         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 967
 968
 969         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 970         bits = (size * 512) / min_chunk + 1;
 971         while (bits > max_bits) {
 972                 min_chunk *= 2;
 973                 bits = (bits+1)/2;
 974         }
 975         if (chunk == UnSet)
 976                 chunk = min_chunk;
 977         else if (chunk < min_chunk)
 978                 return 0; /* chunk size too small */
 979
 980         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 981
 982         memset(bms, 0, sizeof(*bms));
 983         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 984         bms->version = __cpu_to_le32(major);
 985         uuid_from_super0(st, (int*)bms->uuid);
 986         bms->chunksize = __cpu_to_le32(chunk);
 987         bms->daemon_sleep = __cpu_to_le32(delay);
 988         bms->sync_size = __cpu_to_le64(size);
 989         bms->write_behind = __cpu_to_le32(write_behind);
 990         *chunkp = chunk;
 991         return 1;
 992 }
 993
 994
 995 static void locate_bitmap0(struct supertype *st, int fd)
 996 {
 997         unsigned long long dsize;
 998         unsigned long long offset;
 999
1000         if (!get_dev_size(fd, NULL, &dsize))
1001                 return;
1002
1003         if (dsize < MD_RESERVED_SECTORS*512)
1004                 return;
1005
1006         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
1007
1008         offset *= 512;
1009
1010         offset += MD_SB_BYTES;
1011
1012         lseek64(fd, offset, 0);
1013 }
1014
1015 static int write_bitmap0(struct supertype *st, int fd)
1016 {
1017         unsigned long long dsize;
1018         unsigned long long offset;
1019         mdp_super_t *sb = st->sb;
1020
1021         int rv = 0;
1022
1023         int towrite, n;
1024         char abuf[4096+4096];
1025         char *buf = (char*)(((long)(abuf+4096))&~4095L);
1026
1027         if (!get_dev_size(fd, NULL, &dsize))
1028                 return 1;
1029
1030
1031         if (dsize < MD_RESERVED_SECTORS*512)
1032                 return -1;
1033
1034         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
1035
1036         offset *= 512;
1037
1038         if (lseek64(fd, offset + 4096, 0)< 0LL)
1039                 return 3;
1040
1041         memset(buf, 0xff, 4096);
1042         memcpy(buf,  ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t));
1043         towrite = 64*1024;
1044         while (towrite > 0) {
1045                 n = towrite;
1046                 if (n > 4096)
1047                         n = 4096;
1048                 n = write(fd, buf, n);
1049                 if (n > 0)
1050                         towrite -= n;
1051                 else
1052                         break;
1053                 memset(buf, 0xff, 4096);
1054         }
1055         fsync(fd);
1056         if (towrite)
1057                 rv = -2;
1058
1059         return rv;
1060 }
1061
1062 static void free_super0(struct supertype *st)
1063 {
1064         if (st->sb)
1065                 free(st->sb);
1066         st->sb = NULL;
1067 }
1068
1069 #ifndef MDASSEMBLE
1070 static int validate_geometry0(struct supertype *st, int level,
1071                               int layout, int raiddisks,
1072                               int chunk, unsigned long long size,
1073                               char *subdev, unsigned long long *freesize,
1074                               int verbose)
1075 {
1076         unsigned long long ldsize;
1077         int fd;
1078
1079         if (level == LEVEL_CONTAINER)
1080                 return 0;
1081         if (raiddisks > MD_SB_DISKS)
1082                 return 0;
1083         if (size > (0x7fffffffULL<<9))
1084                 return 0;
1085         if (!subdev)
1086                 return 1;
1087
1088         fd = open(subdev, O_RDONLY|O_EXCL, 0);
1089         if (fd < 0) {
1090                 if (verbose)
1091                         fprintf(stderr, Name ": super0.90 cannot open %s: %s\n",
1092                                 subdev, strerror(errno));
1093                 return 0;
1094         }
1095
1096         if (!get_dev_size(fd, subdev, &ldsize)) {
1097                 close(fd);
1098                 return 0;
1099         }
1100         close(fd);
1101
1102         if (ldsize < MD_RESERVED_SECTORS * 512)
1103                 return 0;
1104         if (size > (0x7fffffffULL<<9))
1105                 return 0;
1106         *freesize = MD_NEW_SIZE_SECTORS(ldsize >> 9);
1107         return 1;
1108 }
1109 #endif /* MDASSEMBLE */
1110
1111 struct superswitch super0 = {
1112 #ifndef MDASSEMBLE
1113         .examine_super = examine_super0,
1114         .brief_examine_super = brief_examine_super0,
1115         .export_examine_super = export_examine_super0,
1116         .detail_super = detail_super0,
1117         .brief_detail_super = brief_detail_super0,
1118         .write_init_super = write_init_super0,
1119         .validate_geometry = validate_geometry0,
1120         .add_to_super = add_to_super0,
1121 #endif
1122         .match_home = match_home0,
1123         .uuid_from_super = uuid_from_super0,
1124         .getinfo_super = getinfo_super0,
1125         .update_super = update_super0,
1126         .init_super = init_super0,
1127         .store_super = store_super0,
1128         .compare_super = compare_super0,
1129         .load_super = load_super0,
1130         .match_metadata_desc = match_metadata_desc0,
1131         .avail_size = avail_size0,
1132         .add_internal_bitmap = add_internal_bitmap0,
1133         .locate_bitmap = locate_bitmap0,
1134         .write_bitmap = write_bitmap0,
1135         .free_super = free_super0,
1136         .name = "0.90",
1137 };