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, char *map)
 343 {
 344         mdp_super_t *sb = st->sb;
 345         int working = 0;
 346         int i;
 347         int map_disks = info->array.raid_disks;
 348
 349         info->array.major_version = sb->major_version;
 350         info->array.minor_version = sb->minor_version;
 351         info->array.patch_version = sb->patch_version;
 352         info->array.raid_disks = sb->raid_disks;
 353         info->array.level = sb->level;
 354         info->array.layout = sb->layout;
 355         info->array.md_minor = sb->md_minor;
 356         info->array.ctime = sb->ctime;
 357         info->array.utime = sb->utime;
 358         info->array.chunk_size = sb->chunk_size;
 359         info->array.state = sb->state;
 360         info->component_size = sb->size*2;
 361
 362         info->disk.state = sb->this_disk.state;
 363         info->disk.major = sb->this_disk.major;
 364         info->disk.minor = sb->this_disk.minor;
 365         info->disk.raid_disk = sb->this_disk.raid_disk;
 366         info->disk.number = sb->this_disk.number;
 367
 368         info->events = md_event(sb);
 369         info->data_offset = 0;
 370
 371         sprintf(info->text_version, "0.%d", sb->minor_version);
 372         info->safe_mode_delay = 200;
 373
 374         uuid_from_super0(st, info->uuid);
 375
 376         info->recovery_start = MaxSector;
 377         if (sb->minor_version > 90 && (sb->reshape_position+1) != 0) {
 378                 info->reshape_active = 1;
 379                 info->reshape_progress = sb->reshape_position;
 380                 info->new_level = sb->new_level;
 381                 info->delta_disks = sb->delta_disks;
 382                 info->new_layout = sb->new_layout;
 383                 info->new_chunk = sb->new_chunk;
 384                 if (info->delta_disks < 0)
 385                         info->array.raid_disks -= info->delta_disks;
 386         } else
 387                 info->reshape_active = 0;
 388
 389         sprintf(info->name, "%d", sb->md_minor);
 390         /* work_disks is calculated rather than read directly */
 391         for (i=0; i < MD_SB_DISKS; i++)
 392                 if ((sb->disks[i].state & (1<<MD_DISK_SYNC)) &&
 393                     (sb->disks[i].raid_disk < (unsigned)info->array.raid_disks) &&
 394                     (sb->disks[i].state & (1<<MD_DISK_ACTIVE)) &&
 395                     !(sb->disks[i].state & (1<<MD_DISK_FAULTY))) {
 396                         working ++;
 397                         if (map && i < map_disks)
 398                                 map[i] = 1;
 399                 } else if (map && i < map_disks)
 400                         map[i] = 0;
 401         info->array.working_disks = working;
 402 }
 403
 404 static struct mdinfo *container_content0(struct supertype *st, char *subarray)
 405 {
 406         struct mdinfo *info;
 407
 408         if (subarray)
 409                 return NULL;
 410
 411         info = malloc(sizeof(*info));
 412         getinfo_super0(st, info, NULL);
 413         return info;
 414 }
 415
 416 static int update_super0(struct supertype *st, struct mdinfo *info,
 417                          char *update,
 418                          char *devname, int verbose,
 419                          int uuid_set, char *homehost)
 420 {
 421         /* NOTE: for 'assemble' and 'force' we need to return non-zero if any change was made.
 422          * For others, the return value is ignored.
 423          */
 424         int rv = 0;
 425         mdp_super_t *sb = st->sb;
 426         if (strcmp(update, "sparc2.2")==0 ) {
 427                 /* 2.2 sparc put the events in the wrong place
 428                  * So we copy the tail of the superblock
 429                  * up 4 bytes before continuing
 430                  */
 431                 __u32 *sb32 = (__u32*)sb;
 432                 memcpy(sb32+MD_SB_GENERIC_CONSTANT_WORDS+7,
 433                        sb32+MD_SB_GENERIC_CONSTANT_WORDS+7+1,
 434                        (MD_SB_WORDS - (MD_SB_GENERIC_CONSTANT_WORDS+7+1))*4);
 435                 if (verbose >= 0)
 436                         fprintf (stderr, Name ": adjusting superblock of %s for 2.2/sparc compatability.\n",
 437                                  devname);
 438         } else if (strcmp(update, "super-minor") ==0) {
 439                 sb->md_minor = info->array.md_minor;
 440                 if (verbose > 0)
 441                         fprintf(stderr, Name ": updating superblock of %s with minor number %d\n",
 442                                 devname, info->array.md_minor);
 443         } else if (strcmp(update, "summaries") == 0) {
 444                 unsigned int i;
 445                 /* set nr_disks, active_disks, working_disks,
 446                  * failed_disks, spare_disks based on disks[]
 447                  * array in superblock.
 448                  * Also make sure extra slots aren't 'failed'
 449                  */
 450                 sb->nr_disks = sb->active_disks =
 451                         sb->working_disks = sb->failed_disks =
 452                         sb->spare_disks = 0;
 453                 for (i=0; i < MD_SB_DISKS ; i++)
 454                         if (sb->disks[i].major ||
 455                             sb->disks[i].minor) {
 456                                 int state = sb->disks[i].state;
 457                                 if (state & (1<<MD_DISK_REMOVED))
 458                                         continue;
 459                                 sb->nr_disks++;
 460                                 if (state & (1<<MD_DISK_ACTIVE))
 461                                         sb->active_disks++;
 462                                 if (state & (1<<MD_DISK_FAULTY))
 463                                         sb->failed_disks++;
 464                                 else
 465                                         sb->working_disks++;
 466                                 if (state == 0)
 467                                         sb->spare_disks++;
 468                         } else if (i >= sb->raid_disks && sb->disks[i].number == 0)
 469                                 sb->disks[i].state = 0;
 470         } else if (strcmp(update, "force-one")==0) {
 471                 /* Not enough devices for a working array, so
 472                  * bring this one up-to-date.
 473                  */
 474                 __u32 ehi = sb->events_hi, elo = sb->events_lo;
 475                 sb->events_hi = (info->events>>32) & 0xFFFFFFFF;
 476                 sb->events_lo = (info->events) & 0xFFFFFFFF;
 477                 if (sb->events_hi != ehi ||
 478                     sb->events_lo != elo)
 479                         rv = 1;
 480         } else if (strcmp(update, "force-array")==0) {
 481                 /* degraded array and 'force' requested, so
 482                  * maybe need to mark it 'clean'
 483                  */
 484                 if ((sb->level == 5 || sb->level == 4 || sb->level == 6) &&
 485                     (sb->state & (1 << MD_SB_CLEAN)) == 0) {
 486                         /* need to force clean */
 487                         sb->state |= (1 << MD_SB_CLEAN);
 488                         rv = 1;
 489                 }
 490         } else if (strcmp(update, "assemble")==0) {
 491                 int d = info->disk.number;
 492                 int wonly = sb->disks[d].state & (1<<MD_DISK_WRITEMOSTLY);
 493                 int mask = (1<<MD_DISK_WRITEMOSTLY);
 494                 int add = 0;
 495                 if (sb->minor_version >= 91)
 496                         /* During reshape we don't insist on everything
 497                          * being marked 'sync'
 498                          */
 499                         add = (1<<MD_DISK_SYNC);
 500                 if (((sb->disks[d].state & ~mask) | add)
 501                     != (unsigned)info->disk.state) {
 502                         sb->disks[d].state = info->disk.state | wonly;
 503                         rv = 1;
 504                 }
 505         } else if (strcmp(update, "linear-grow-new") == 0) {
 506                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 507                 sb->disks[info->disk.number].number = info->disk.number;
 508                 sb->disks[info->disk.number].major = info->disk.major;
 509                 sb->disks[info->disk.number].minor = info->disk.minor;
 510                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 511                 sb->disks[info->disk.number].state = info->disk.state;
 512                 sb->this_disk = sb->disks[info->disk.number];
 513         } else if (strcmp(update, "linear-grow-update") == 0) {
 514                 sb->raid_disks = info->array.raid_disks;
 515                 sb->nr_disks = info->array.nr_disks;
 516                 sb->active_disks = info->array.active_disks;
 517                 sb->working_disks = info->array.working_disks;
 518                 memset(&sb->disks[info->disk.number], 0, sizeof(sb->disks[0]));
 519                 sb->disks[info->disk.number].number = info->disk.number;
 520                 sb->disks[info->disk.number].major = info->disk.major;
 521                 sb->disks[info->disk.number].minor = info->disk.minor;
 522                 sb->disks[info->disk.number].raid_disk = info->disk.raid_disk;
 523                 sb->disks[info->disk.number].state = info->disk.state;
 524         } else if (strcmp(update, "resync") == 0) {
 525                 /* make sure resync happens */
 526                 sb->state &= ~(1<<MD_SB_CLEAN);
 527                 sb->recovery_cp = 0;
 528         } else if (strcmp(update, "homehost") == 0 &&
 529             homehost) {
 530                 uuid_set = 0;
 531                 update = "uuid";
 532                 info->uuid[0] = sb->set_uuid0;
 533                 info->uuid[1] = sb->set_uuid1;
 534         } else if (strcmp(update, "uuid") == 0) {
 535                 if (!uuid_set && homehost) {
 536                         char buf[20];
 537                         char *hash = sha1_buffer(homehost,
 538                                                  strlen(homehost),
 539                                                  buf);
 540                         memcpy(info->uuid+2, hash, 8);
 541                 }
 542                 sb->set_uuid0 = info->uuid[0];
 543                 sb->set_uuid1 = info->uuid[1];
 544                 sb->set_uuid2 = info->uuid[2];
 545                 sb->set_uuid3 = info->uuid[3];
 546                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
 547                         struct bitmap_super_s *bm;
 548                         bm = (struct bitmap_super_s*)(sb+1);
 549                         uuid_from_super0(st, (int*)bm->uuid);
 550                 }
 551         } else if (strcmp(update, "_reshape_progress")==0)
 552                 sb->reshape_position = info->reshape_progress;
 553         else
 554                 rv = -1;
 555
 556         sb->sb_csum = calc_sb0_csum(sb);
 557         return rv;
 558 }
 559
 560 /*
 561  * For verion-0 superblock, the homehost is 'stored' in the
 562  * uuid.  8 bytes for a hash of the host leaving 8 bytes
 563  * of random material.
 564  * We use the first 8 bytes (64bits) of the sha1 of the
 565  * host name
 566  */
 567
 568
 569 static int init_super0(struct supertype *st, mdu_array_info_t *info,
 570                        unsigned long long size, char *ignored_name, char *homehost,
 571                        int *uuid)
 572 {
 573         mdp_super_t *sb;
 574         int spares;
 575
 576         if (posix_memalign((void**)&sb, 4096,
 577                            MD_SB_BYTES + ROUND_UP(sizeof(bitmap_super_t), 4096)) != 0) {
 578                 fprintf(stderr, Name ": %s could not allocate superblock\n", __func__);
 579                 return 0;
 580         }
 581         memset(sb, 0, MD_SB_BYTES + sizeof(bitmap_super_t));
 582
 583         st->sb = sb;
 584         if (info == NULL) {
 585                 /* zeroing the superblock */
 586                 return 0;
 587         }
 588
 589         spares = info->working_disks - info->active_disks;
 590         if (info->raid_disks + spares  > MD_SB_DISKS) {
 591                 fprintf(stderr, Name ": too many devices requested: %d+%d > %d\n",
 592                         info->raid_disks , spares, MD_SB_DISKS);
 593                 return 0;
 594         }
 595
 596         sb->md_magic = MD_SB_MAGIC;
 597         sb->major_version = 0;
 598         sb->minor_version = 90;
 599         sb->patch_version = 0;
 600         sb->gvalid_words = 0; /* ignored */
 601         sb->ctime = time(0);
 602         sb->level = info->level;
 603         if (size != (unsigned long long)info->size)
 604                 return 0;
 605         sb->size = info->size;
 606         sb->nr_disks = info->nr_disks;
 607         sb->raid_disks = info->raid_disks;
 608         sb->md_minor = info->md_minor;
 609         sb->not_persistent = 0;
 610         if (uuid) {
 611                 sb->set_uuid0 = uuid[0];
 612                 sb->set_uuid1 = uuid[1];
 613                 sb->set_uuid2 = uuid[2];
 614                 sb->set_uuid3 = uuid[3];
 615         } else {
 616                 int rfd = open("/dev/urandom", O_RDONLY);
 617                 if (rfd < 0 || read(rfd, &sb->set_uuid0, 4) != 4)
 618                         sb->set_uuid0 = random();
 619                 if (rfd < 0 || read(rfd, &sb->set_uuid1, 12) != 12) {
 620                         sb->set_uuid1 = random();
 621                         sb->set_uuid2 = random();
 622                         sb->set_uuid3 = random();
 623                 }
 624                 if (rfd >= 0)
 625                         close(rfd);
 626         }
 627         if (homehost) {
 628                 char buf[20];
 629                 char *hash = sha1_buffer(homehost,
 630                                          strlen(homehost),
 631                                          buf);
 632                 memcpy(&sb->set_uuid2, hash, 8);
 633         }
 634
 635         sb->utime = sb->ctime;
 636         sb->state = info->state;
 637         sb->active_disks = info->active_disks;
 638         sb->working_disks = info->working_disks;
 639         sb->failed_disks = info->failed_disks;
 640         sb->spare_disks = info->spare_disks;
 641         sb->events_hi = 0;
 642         sb->events_lo = 1;
 643
 644         sb->layout = info->layout;
 645         sb->chunk_size = info->chunk_size;
 646
 647         return 1;
 648 }
 649
 650 struct devinfo {
 651         int fd;
 652         char *devname;
 653         mdu_disk_info_t disk;
 654         struct devinfo *next;
 655 };
 656
 657 #ifndef MDASSEMBLE
 658 /* Add a device to the superblock being created */
 659 static int add_to_super0(struct supertype *st, mdu_disk_info_t *dinfo,
 660                           int fd, char *devname)
 661 {
 662         mdp_super_t *sb = st->sb;
 663         mdp_disk_t *dk = &sb->disks[dinfo->number];
 664         struct devinfo *di, **dip;
 665
 666         dk->number = dinfo->number;
 667         dk->major = dinfo->major;
 668         dk->minor = dinfo->minor;
 669         dk->raid_disk = dinfo->raid_disk;
 670         dk->state = dinfo->state;
 671
 672         sb->this_disk = sb->disks[dinfo->number];
 673         sb->sb_csum = calc_sb0_csum(sb);
 674
 675         dip = (struct devinfo **)&st->info;
 676         while (*dip)
 677                 dip = &(*dip)->next;
 678         di = malloc(sizeof(struct devinfo));
 679         di->fd = fd;
 680         di->devname = devname;
 681         di->disk = *dinfo;
 682         di->next = NULL;
 683         *dip = di;
 684
 685         return 0;
 686 }
 687 #endif
 688
 689 static int store_super0(struct supertype *st, int fd)
 690 {
 691         unsigned long long dsize;
 692         unsigned long long offset;
 693         mdp_super_t *super = st->sb;
 694
 695         if (!get_dev_size(fd, NULL, &dsize))
 696                 return 1;
 697
 698         if (dsize < MD_RESERVED_SECTORS*512)
 699                 return 2;
 700
 701         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 702
 703         offset *= 512;
 704
 705         if (lseek64(fd, offset, 0)< 0LL)
 706                 return 3;
 707
 708         if (write(fd, super, sizeof(*super)) != sizeof(*super))
 709                 return 4;
 710
 711         if (super->state & (1<<MD_SB_BITMAP_PRESENT)) {
 712                 struct bitmap_super_s * bm = (struct bitmap_super_s*)(super+1);
 713                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC)
 714                         if (write(fd, bm, ROUND_UP(sizeof(*bm),4096)) !=
 715                             ROUND_UP(sizeof(*bm),4096))
 716                             return 5;
 717         }
 718
 719         fsync(fd);
 720         return 0;
 721 }
 722
 723 #ifndef MDASSEMBLE
 724 static int write_init_super0(struct supertype *st)
 725 {
 726         mdp_super_t *sb = st->sb;
 727         int rv = 0;
 728         struct devinfo *di;
 729
 730         for (di = st->info ; di && ! rv ; di = di->next) {
 731
 732                 if (di->disk.state == 1)
 733                         continue;
 734                 if (di->fd == -1)
 735                         continue;
 736                 while (Kill(di->devname, NULL, 0, 1, 1) == 0)
 737                         ;
 738
 739                 sb->disks[di->disk.number].state &= ~(1<<MD_DISK_FAULTY);
 740
 741                 sb->this_disk = sb->disks[di->disk.number];
 742                 sb->sb_csum = calc_sb0_csum(sb);
 743                 rv = store_super0(st, di->fd);
 744
 745                 if (rv == 0 && (sb->state & (1<<MD_SB_BITMAP_PRESENT)))
 746                         rv = st->ss->write_bitmap(st, di->fd);
 747
 748                 if (rv)
 749                         fprintf(stderr,
 750                                 Name ": failed to write superblock to %s\n",
 751                                 di->devname);
 752                 close(di->fd);
 753                 di->fd = -1;
 754         }
 755         return rv;
 756 }
 757 #endif
 758
 759 static int compare_super0(struct supertype *st, struct supertype *tst)
 760 {
 761         /*
 762          * return:
 763          *  0 same, or first was empty, and second was copied
 764          *  1 second had wrong number
 765          *  2 wrong uuid
 766          *  3 wrong other info
 767          */
 768         mdp_super_t *first = st->sb;
 769         mdp_super_t *second = tst->sb;
 770         int uuid1[4], uuid2[4];
 771
 772         if (second->md_magic != MD_SB_MAGIC)
 773                 return 1;
 774         if (!first) {
 775                 if (posix_memalign((void**)&first, 4096,
 776                              MD_SB_BYTES +
 777                              ROUND_UP(sizeof(struct bitmap_super_s), 4096)) != 0) {
 778                         fprintf(stderr, Name
 779                                 ": %s could not allocate superblock\n", __func__);
 780                         return 1;
 781                 }
 782                 memcpy(first, second, MD_SB_BYTES + sizeof(struct bitmap_super_s));
 783                 st->sb = first;
 784                 return 0;
 785         }
 786
 787         uuid_from_super0(st, uuid1);
 788         uuid_from_super0(tst, uuid2);
 789         if (!same_uuid(uuid1, uuid2, 0))
 790                 return 2;
 791         if (first->major_version != second->major_version ||
 792             first->minor_version != second->minor_version ||
 793             first->patch_version != second->patch_version ||
 794             first->gvalid_words  != second->gvalid_words  ||
 795             first->ctime         != second->ctime         ||
 796             first->level         != second->level         ||
 797             first->size          != second->size          ||
 798             first->raid_disks    != second->raid_disks    )
 799                 return 3;
 800
 801         return 0;
 802 }
 803
 804
 805 static void free_super0(struct supertype *st);
 806
 807 static int load_super0(struct supertype *st, int fd, char *devname)
 808 {
 809         /* try to read in the superblock
 810          * Return:
 811          *  0 on success
 812          *  1 on cannot get superblock
 813          *  2 on superblock meaningless
 814          */
 815         unsigned long long dsize;
 816         unsigned long long offset;
 817         mdp_super_t *super;
 818         int uuid[4];
 819         struct bitmap_super_s *bsb;
 820
 821         free_super0(st);
 822
 823         if (!get_dev_size(fd, devname, &dsize))
 824                 return 1;
 825
 826         if (dsize < MD_RESERVED_SECTORS*512) {
 827                 if (devname)
 828                         fprintf(stderr, Name
 829                             ": %s is too small for md: size is %llu sectors.\n",
 830                                 devname, dsize);
 831                 return 1;
 832         }
 833
 834         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
 835
 836         offset *= 512;
 837
 838         ioctl(fd, BLKFLSBUF, 0); /* make sure we read current data */
 839
 840         if (lseek64(fd, offset, 0)< 0LL) {
 841                 if (devname)
 842                         fprintf(stderr, Name ": Cannot seek to superblock on %s: %s\n",
 843                                 devname, strerror(errno));
 844                 return 1;
 845         }
 846
 847         if (posix_memalign((void**)&super, 4096,
 848                            MD_SB_BYTES +
 849                            ROUND_UP(sizeof(bitmap_super_t), 4096)) != 0) {
 850                 fprintf(stderr, Name
 851                         ": %s could not allocate superblock\n", __func__);
 852                 return 1;
 853         }
 854
 855         if (read(fd, super, sizeof(*super)) != MD_SB_BYTES) {
 856                 if (devname)
 857                         fprintf(stderr, Name ": Cannot read superblock on %s\n",
 858                                 devname);
 859                 free(super);
 860                 return 1;
 861         }
 862
 863         if (st->ss && st->minor_version == 9)
 864                 super0_swap_endian(super);
 865
 866         if (super->md_magic != MD_SB_MAGIC) {
 867                 if (devname)
 868                         fprintf(stderr, Name ": No super block found on %s (Expected magic %08x, got %08x)\n",
 869                                 devname, MD_SB_MAGIC, super->md_magic);
 870                 free(super);
 871                 return 2;
 872         }
 873
 874         if (super->major_version != 0) {
 875                 if (devname)
 876                         fprintf(stderr, Name ": Cannot interpret superblock on %s - version is %d\n",
 877                                 devname, super->major_version);
 878                 free(super);
 879                 return 2;
 880         }
 881         st->sb = super;
 882
 883         if (st->ss == NULL) {
 884                 st->ss = &super0;
 885                 st->minor_version = super->minor_version;
 886                 st->max_devs = MD_SB_DISKS;
 887                 st->info = NULL;
 888         }
 889
 890         /* Now check on the bitmap superblock */
 891         if ((super->state & (1<<MD_SB_BITMAP_PRESENT)) == 0)
 892                 return 0;
 893         /* Read the bitmap superblock and make sure it looks
 894          * valid.  If it doesn't clear the bit.  An --assemble --force
 895          * should get that written out.
 896          */
 897         if (read(fd, super+1, ROUND_UP(sizeof(struct bitmap_super_s),4096))
 898             != ROUND_UP(sizeof(struct bitmap_super_s),4096))
 899                 goto no_bitmap;
 900
 901         uuid_from_super0(st, uuid);
 902         bsb = (struct bitmap_super_s *)(super+1);
 903         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
 904             memcmp(bsb->uuid, uuid, 16) != 0)
 905                 goto no_bitmap;
 906         return 0;
 907
 908  no_bitmap:
 909         super->state &= ~(1<<MD_SB_BITMAP_PRESENT);
 910
 911         return 0;
 912 }
 913
 914 static struct supertype *match_metadata_desc0(char *arg)
 915 {
 916         struct supertype *st = malloc(sizeof(*st));
 917         if (!st) return st;
 918
 919         memset(st, 0, sizeof(*st));
 920         st->container_dev = NoMdDev;
 921         st->ss = &super0;
 922         st->info = NULL;
 923         st->minor_version = 90;
 924         st->max_devs = MD_SB_DISKS;
 925         st->sb = NULL;
 926         /* we sometimes get 00.90 */
 927         while (arg[0] == '0' && arg[1] == '0')
 928                 arg++;
 929         if (strcmp(arg, "0") == 0 ||
 930 #ifdef DEFAULT_OLD_METADATA /* ifndef in super1.c */
 931             strcmp(arg, "default") == 0 ||
 932 #endif /* DEFAULT_OLD_METADATA */
 933             strcmp(arg, "0.90") == 0 ||
 934             strcmp(arg, "") == 0 /* no metadata  - i.e. non_persistent */
 935                 )
 936                 return st;
 937
 938         st->minor_version = 91; /* reshape in progress */
 939         if (strcmp(arg, "0.91") == 0) /* For dup_super support */
 940                 return st;
 941
 942         st->minor_version = 9; /* flag for 'byte-swapped' */
 943         if (strcmp(arg, "0.swap")==0 ||
 944             strcmp(arg, "0.9") == 0) /* For dup_super support */
 945                 return st;
 946
 947         free(st);
 948         return NULL;
 949 }
 950
 951 static __u64 avail_size0(struct supertype *st, __u64 devsize)
 952 {
 953         if (devsize < MD_RESERVED_SECTORS)
 954                 return 0ULL;
 955         return MD_NEW_SIZE_SECTORS(devsize);
 956 }
 957
 958 static int add_internal_bitmap0(struct supertype *st, int *chunkp,
 959                                 int delay, int write_behind,
 960                                 unsigned long long size, int may_change,
 961                                 int major)
 962 {
 963         /*
 964          * The bitmap comes immediately after the superblock and must be 60K in size
 965          * at most.  The default size is between 30K and 60K
 966          *
 967          * size is in sectors,  chunk is in bytes !!!
 968          */
 969         unsigned long long bits;
 970         unsigned long long max_bits = (60*1024 - sizeof(bitmap_super_t))*8;
 971         unsigned long long min_chunk;
 972         int chunk = *chunkp;
 973         mdp_super_t *sb = st->sb;
 974         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MD_SB_BYTES);
 975
 976
 977         min_chunk = 4096; /* sub-page chunks don't work yet.. */
 978         bits = (size * 512) / min_chunk + 1;
 979         while (bits > max_bits) {
 980                 min_chunk *= 2;
 981                 bits = (bits+1)/2;
 982         }
 983         if (chunk == UnSet) {
 984                 /* A chunk size less than a few Megabytes gives poor
 985                  * performance without increasing resync noticeably
 986                  */
 987                 chunk = min_chunk;
 988                 if (chunk < 64*1024*1024)
 989                         chunk = 64*1024*1024;
 990         } else if ((unsigned long long)chunk < min_chunk)
 991                 return 0; /* chunk size too small */
 992
 993         sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 994
 995         memset(bms, 0, sizeof(*bms));
 996         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
 997         bms->version = __cpu_to_le32(major);
 998         uuid_from_super0(st, (int*)bms->uuid);
 999         bms->chunksize = __cpu_to_le32(chunk);
1000         bms->daemon_sleep = __cpu_to_le32(delay);
1001         bms->sync_size = __cpu_to_le64(size);
1002         bms->write_behind = __cpu_to_le32(write_behind);
1003         *chunkp = chunk;
1004         return 1;
1005 }
1006
1007
1008 static void locate_bitmap0(struct supertype *st, int fd)
1009 {
1010         unsigned long long dsize;
1011         unsigned long long offset;
1012
1013         if (!get_dev_size(fd, NULL, &dsize))
1014                 return;
1015
1016         if (dsize < MD_RESERVED_SECTORS*512)
1017                 return;
1018
1019         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
1020
1021         offset *= 512;
1022
1023         offset += MD_SB_BYTES;
1024
1025         lseek64(fd, offset, 0);
1026 }
1027
1028 static int write_bitmap0(struct supertype *st, int fd)
1029 {
1030         unsigned long long dsize;
1031         unsigned long long offset;
1032         mdp_super_t *sb = st->sb;
1033
1034         int rv = 0;
1035
1036         int towrite, n;
1037         char abuf[4096+4096];
1038         char *buf = (char*)(((long)(abuf+4096))&~4095L);
1039
1040         if (!get_dev_size(fd, NULL, &dsize))
1041                 return 1;
1042
1043
1044         if (dsize < MD_RESERVED_SECTORS*512)
1045                 return -1;
1046
1047         offset = MD_NEW_SIZE_SECTORS(dsize>>9);
1048
1049         offset *= 512;
1050
1051         if (lseek64(fd, offset + 4096, 0)< 0LL)
1052                 return 3;
1053
1054         memset(buf, 0xff, 4096);
1055         memcpy(buf,  ((char*)sb)+MD_SB_BYTES, sizeof(bitmap_super_t));
1056         towrite = 60*1024;
1057         while (towrite > 0) {
1058                 n = towrite;
1059                 if (n > 4096)
1060                         n = 4096;
1061                 n = write(fd, buf, n);
1062                 if (n > 0)
1063                         towrite -= n;
1064                 else
1065                         break;
1066                 memset(buf, 0xff, 4096);
1067         }
1068         fsync(fd);
1069         if (towrite)
1070                 rv = -2;
1071
1072         return rv;
1073 }
1074
1075 static void free_super0(struct supertype *st)
1076 {
1077         if (st->sb)
1078                 free(st->sb);
1079         st->sb = NULL;
1080 }
1081
1082 #ifndef MDASSEMBLE
1083 static int validate_geometry0(struct supertype *st, int level,
1084                               int layout, int raiddisks,
1085                               int chunk, unsigned long long size,
1086                               char *subdev, unsigned long long *freesize,
1087                               int verbose)
1088 {
1089         unsigned long long ldsize;
1090         int fd;
1091
1092         if (level == LEVEL_CONTAINER) {
1093                 if (verbose)
1094                         fprintf(stderr, Name ": 0.90 metadata does not support containers\n");
1095                 return 0;
1096         }
1097         if (raiddisks > MD_SB_DISKS) {
1098                 if (verbose)
1099                         fprintf(stderr, Name ": 0.90 metadata supports at most %d devices per array\n",
1100                                 MD_SB_DISKS);
1101                 return 0;
1102         }
1103         if (size > (0x7fffffffULL<<9)) {
1104                 if (verbose)
1105                         fprintf(stderr, Name ": 0.90 metadata supports at most 2 terrabytes per device\n");
1106                 return 0;
1107         }
1108         if (!subdev)
1109                 return 1;
1110
1111         fd = open(subdev, O_RDONLY|O_EXCL, 0);
1112         if (fd < 0) {
1113                 if (verbose)
1114                         fprintf(stderr, Name ": super0.90 cannot open %s: %s\n",
1115                                 subdev, strerror(errno));
1116                 return 0;
1117         }
1118
1119         if (!get_dev_size(fd, subdev, &ldsize)) {
1120                 close(fd);
1121                 return 0;
1122         }
1123         close(fd);
1124
1125         if (ldsize < MD_RESERVED_SECTORS * 512)
1126                 return 0;
1127         if (size > (0x7fffffffULL<<9))
1128                 return 0;
1129         *freesize = MD_NEW_SIZE_SECTORS(ldsize >> 9);
1130         return 1;
1131 }
1132 #endif /* MDASSEMBLE */
1133
1134 struct superswitch super0 = {
1135 #ifndef MDASSEMBLE
1136         .examine_super = examine_super0,
1137         .brief_examine_super = brief_examine_super0,
1138         .export_examine_super = export_examine_super0,
1139         .detail_super = detail_super0,
1140         .brief_detail_super = brief_detail_super0,
1141         .write_init_super = write_init_super0,
1142         .validate_geometry = validate_geometry0,
1143         .add_to_super = add_to_super0,
1144 #endif
1145         .match_home = match_home0,
1146         .uuid_from_super = uuid_from_super0,
1147         .getinfo_super = getinfo_super0,
1148         .container_content = container_content0,
1149         .update_super = update_super0,
1150         .init_super = init_super0,
1151         .store_super = store_super0,
1152         .compare_super = compare_super0,
1153         .load_super = load_super0,
1154         .match_metadata_desc = match_metadata_desc0,
1155         .avail_size = avail_size0,
1156         .add_internal_bitmap = add_internal_bitmap0,
1157         .locate_bitmap = locate_bitmap0,
1158         .write_bitmap = write_bitmap0,
1159         .free_super = free_super0,
1160         .name = "0.90",
1161 };