super1.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2016 Neil Brown <neilb@suse.com>
   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 #include <stddef.h>
  26 #include "mdadm.h"
  27 /*
  28  * The version-1 superblock :
  29  * All numeric fields are little-endian.
  30  *
  31  * total size: 256 bytes plus 2 per device.
  32  *  1K allows 384 devices.
  33  */
  34 struct mdp_superblock_1 {
  35         /* constant array information - 128 bytes */
  36         __u32   magic;          /* MD_SB_MAGIC: 0xa92b4efc - little endian */
  37         __u32   major_version;  /* 1 */
  38         __u32   feature_map;    /* 0 for now */
  39         __u32   pad0;           /* always set to 0 when writing */
  40
  41         __u8    set_uuid[16];   /* user-space generated. */
  42         char    set_name[32];   /* set and interpreted by user-space */
  43
  44         __u64   ctime;          /* lo 40 bits are seconds, top 24 are microseconds or 0*/
  45         __u32   level;          /* -4 (multipath), -1 (linear), 0,1,4,5 */
  46         __u32   layout;         /* only for raid5 currently */
  47         __u64   size;           /* used size of component devices, in 512byte sectors */
  48
  49         __u32   chunksize;      /* in 512byte sectors */
  50         __u32   raid_disks;
  51         __u32   bitmap_offset;  /* sectors after start of superblock that bitmap starts
  52                                  * NOTE: signed, so bitmap can be before superblock
  53                                  * only meaningful of feature_map[0] is set.
  54                                  */
  55
  56         /* These are only valid with feature bit '4' */
  57         __u32   new_level;      /* new level we are reshaping to                */
  58         __u64   reshape_position;       /* next address in array-space for reshape */
  59         __u32   delta_disks;    /* change in number of raid_disks               */
  60         __u32   new_layout;     /* new layout                                   */
  61         __u32   new_chunk;      /* new chunk size (sectors)                     */
  62         __u32   new_offset;     /* signed number to add to data_offset in new
  63                                  * layout.  0 == no-change.  This can be
  64                                  * different on each device in the array.
  65                                  */
  66
  67         /* constant this-device information - 64 bytes */
  68         __u64   data_offset;    /* sector start of data, often 0 */
  69         __u64   data_size;      /* sectors in this device that can be used for data */
  70         __u64   super_offset;   /* sector start of this superblock */
  71         union {
  72                 __u64   recovery_offset;/* sectors before this offset (from data_offset) have been recovered */
  73                 __u64   journal_tail;/* journal tail of journal device (from data_offset) */
  74         };
  75         __u32   dev_number;     /* permanent identifier of this  device - not role in raid */
  76         __u32   cnt_corrected_read; /* number of read errors that were corrected by re-writing */
  77         __u8    device_uuid[16]; /* user-space setable, ignored by kernel */
  78         __u8    devflags;        /* per-device flags.  Only one defined...*/
  79 #define WriteMostly1    1        /* mask for writemostly flag in above */
  80 #define FailFast1       2        /* Device should get FailFast requests */
  81         /* bad block log.  If there are any bad blocks the feature flag is set.
  82          * if offset and size are non-zero, that space is reserved and available.
  83          */
  84         __u8    bblog_shift;    /* shift from sectors to block size for badblocklist */
  85         __u16   bblog_size;     /* number of sectors reserved for badblocklist */
  86         __u32   bblog_offset;   /* sector offset from superblock to bblog, signed */
  87
  88         /* array state information - 64 bytes */
  89         __u64   utime;          /* 40 bits second, 24 btes microseconds */
  90         __u64   events;         /* incremented when superblock updated */
  91         __u64   resync_offset;  /* data before this offset (from data_offset) known to be in sync */
  92         __u32   sb_csum;        /* checksum upto dev_roles[max_dev] */
  93         __u32   max_dev;        /* size of dev_roles[] array to consider */
  94         __u8    pad3[64-32];    /* set to 0 when writing */
  95
  96         /* device state information. Indexed by dev_number.
  97          * 2 bytes per device
  98          * Note there are no per-device state flags. State information is rolled
  99          * into the 'roles' value.  If a device is spare or faulty, then it doesn't
 100          * have a meaningful role.
 101          */
 102         __u16   dev_roles[0];   /* role in array, or 0xffff for a spare, or 0xfffe for faulty */
 103 };
 104
 105 #define MAX_SB_SIZE 4096
 106 /* bitmap super size is 256, but we round up to a sector for alignment */
 107 #define BM_SUPER_SIZE 512
 108 #define MAX_DEVS ((int)(MAX_SB_SIZE - sizeof(struct mdp_superblock_1)) / 2)
 109 #define SUPER1_SIZE     (MAX_SB_SIZE + BM_SUPER_SIZE \
 110                          + sizeof(struct misc_dev_info))
 111
 112 struct misc_dev_info {
 113         __u64 device_size;
 114 };
 115
 116 /* feature_map bits */
 117 #define MD_FEATURE_BITMAP_OFFSET        1
 118 #define MD_FEATURE_RECOVERY_OFFSET      2 /* recovery_offset is present and
 119                                            * must be honoured
 120                                            */
 121 #define MD_FEATURE_RESHAPE_ACTIVE       4
 122 #define MD_FEATURE_BAD_BLOCKS           8 /* badblock list is not empty */
 123 #define MD_FEATURE_REPLACEMENT          16 /* This device is replacing an
 124                                             * active device with same 'role'.
 125                                             * 'recovery_offset' is also set.
 126                                             */
 127 #define MD_FEATURE_RESHAPE_BACKWARDS    32 /* Reshape doesn't change number
 128                                             * of devices, but is going
 129                                             * backwards anyway.
 130                                             */
 131 #define MD_FEATURE_NEW_OFFSET           64 /* new_offset must be honoured */
 132 #define MD_FEATURE_BITMAP_VERSIONED     256 /* bitmap version number checked properly */
 133 #define MD_FEATURE_JOURNAL              512 /* support write journal */
 134 #define MD_FEATURE_ALL                  (MD_FEATURE_BITMAP_OFFSET       \
 135                                         |MD_FEATURE_RECOVERY_OFFSET     \
 136                                         |MD_FEATURE_RESHAPE_ACTIVE      \
 137                                         |MD_FEATURE_BAD_BLOCKS          \
 138                                         |MD_FEATURE_REPLACEMENT         \
 139                                         |MD_FEATURE_RESHAPE_BACKWARDS   \
 140                                         |MD_FEATURE_NEW_OFFSET          \
 141                                         |MD_FEATURE_BITMAP_VERSIONED    \
 142                                         |MD_FEATURE_JOURNAL             \
 143                                         )
 144
 145 #ifndef MDASSEMBLE
 146 static int role_from_sb(struct mdp_superblock_1 *sb)
 147 {
 148         unsigned int d;
 149         int role;
 150
 151         d = __le32_to_cpu(sb->dev_number);
 152         if (d < __le32_to_cpu(sb->max_dev))
 153                 role = __le16_to_cpu(sb->dev_roles[d]);
 154         else
 155                 role = MD_DISK_ROLE_SPARE;
 156         return role;
 157 }
 158 #endif
 159
 160 /* return how many bytes are needed for bitmap, for cluster-md each node
 161  * should have it's own bitmap */
 162 static unsigned int calc_bitmap_size(bitmap_super_t *bms, unsigned int boundary)
 163 {
 164         unsigned long long bits, bytes;
 165
 166         bits = bitmap_bits(__le64_to_cpu(bms->sync_size),
 167                            __le32_to_cpu(bms->chunksize));
 168         bytes = (bits+7) >> 3;
 169         bytes += sizeof(bitmap_super_t);
 170         bytes = ROUND_UP(bytes, boundary);
 171
 172         return bytes;
 173 }
 174
 175 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 176 {
 177         unsigned int disk_csum, csum;
 178         unsigned long long newcsum;
 179         int size = sizeof(*sb) + __le32_to_cpu(sb->max_dev)*2;
 180         unsigned int *isuper = (unsigned int*)sb;
 181
 182 /* make sure I can count... */
 183         if (offsetof(struct mdp_superblock_1,data_offset) != 128 ||
 184             offsetof(struct mdp_superblock_1, utime) != 192 ||
 185             sizeof(struct mdp_superblock_1) != 256) {
 186                 fprintf(stderr, "WARNING - superblock isn't sized correctly\n");
 187         }
 188
 189         disk_csum = sb->sb_csum;
 190         sb->sb_csum = 0;
 191         newcsum = 0;
 192         for (; size>=4; size -= 4 ) {
 193                 newcsum += __le32_to_cpu(*isuper);
 194                 isuper++;
 195         }
 196
 197         if (size == 2)
 198                 newcsum += __le16_to_cpu(*(unsigned short*) isuper);
 199
 200         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 201         sb->sb_csum = disk_csum;
 202         return __cpu_to_le32(csum);
 203 }
 204
 205 /*
 206  * Information related to file descriptor used for aligned reads/writes.
 207  * Cache the block size.
 208  */
 209 struct align_fd {
 210         int fd;
 211         int blk_sz;
 212 };
 213
 214 static void init_afd(struct align_fd *afd, int fd)
 215 {
 216         afd->fd = fd;
 217         if (!get_dev_sector_size(afd->fd, NULL, (unsigned int *)&afd->blk_sz))
 218                 afd->blk_sz = 512;
 219 }
 220
 221 static char abuf[4096+4096];
 222 static int aread(struct align_fd *afd, void *buf, int len)
 223 {
 224         /* aligned read.
 225          * On devices with a 4K sector size, we need to read
 226          * the full sector and copy relevant bits into
 227          * the buffer
 228          */
 229         int bsize, iosize;
 230         char *b;
 231         int n;
 232
 233         bsize = afd->blk_sz;
 234
 235         if (!bsize || bsize > 4096 || len > 4096) {
 236                 if (!bsize)
 237                         fprintf(stderr, "WARNING - aread() called with invalid block size\n");
 238                 return -1;
 239         }
 240         b = ROUND_UP_PTR((char *)abuf, 4096);
 241
 242         for (iosize = 0; iosize < len; iosize += bsize)
 243                 ;
 244         n = read(afd->fd, b, iosize);
 245         if (n <= 0)
 246                 return n;
 247         lseek(afd->fd, len - n, 1);
 248         if (n > len)
 249                 n = len;
 250         memcpy(buf, b, n);
 251         return n;
 252 }
 253
 254 static int awrite(struct align_fd *afd, void *buf, int len)
 255 {
 256         /* aligned write.
 257          * On devices with a 4K sector size, we need to write
 258          * the full sector.  We pre-read if the sector is larger
 259          * than the write.
 260          * The address must be sector-aligned.
 261          */
 262         int bsize, iosize;
 263         char *b;
 264         int n;
 265
 266         bsize = afd->blk_sz;
 267         if (!bsize || bsize > 4096 || len > 4096) {
 268                 if (!bsize)
 269                         fprintf(stderr, "WARNING - awrite() called with invalid block size\n");
 270                 return -1;
 271         }
 272         b = ROUND_UP_PTR((char *)abuf, 4096);
 273
 274         for (iosize = 0; iosize < len ; iosize += bsize)
 275                 ;
 276
 277         if (len != iosize) {
 278                 n = read(afd->fd, b, iosize);
 279                 if (n <= 0)
 280                         return n;
 281                 lseek(afd->fd, -n, 1);
 282         }
 283
 284         memcpy(b, buf, len);
 285         n = write(afd->fd, b, iosize);
 286         if (n <= 0)
 287                 return n;
 288         lseek(afd->fd, len - n, 1);
 289         return len;
 290 }
 291
 292 #ifndef MDASSEMBLE
 293 static void examine_super1(struct supertype *st, char *homehost)
 294 {
 295         struct mdp_superblock_1 *sb = st->sb;
 296         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
 297         time_t atime;
 298         unsigned int d;
 299         int role;
 300         int delta_extra = 0;
 301         int i;
 302         char *c;
 303         int l = homehost ? strlen(homehost) : 0;
 304         int layout;
 305         unsigned long long sb_offset;
 306         struct mdinfo info;
 307
 308         printf("          Magic : %08x\n", __le32_to_cpu(sb->magic));
 309         printf("        Version : 1");
 310         sb_offset = __le64_to_cpu(sb->super_offset);
 311         if (sb_offset <= 4)
 312                 printf(".1\n");
 313         else if (sb_offset <= 8)
 314                 printf(".2\n");
 315         else
 316                 printf(".0\n");
 317         printf("    Feature Map : 0x%x\n", __le32_to_cpu(sb->feature_map));
 318         printf("     Array UUID : ");
 319         for (i=0; i<16; i++) {
 320                 if ((i&3)==0 && i != 0) printf(":");
 321                 printf("%02x", sb->set_uuid[i]);
 322         }
 323         printf("\n");
 324         printf("           Name : %.32s", sb->set_name);
 325         if (l > 0 && l < 32 &&
 326             sb->set_name[l] == ':' &&
 327             strncmp(sb->set_name, homehost, l) == 0)
 328                 printf("  (local to host %s)", homehost);
 329         printf("\n");
 330         if (bms->nodes > 0 && (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
 331                 printf("   Cluster Name : %-64s\n", bms->cluster_name);
 332         atime = __le64_to_cpu(sb->ctime) & 0xFFFFFFFFFFULL;
 333         printf("  Creation Time : %.24s\n", ctime(&atime));
 334         c=map_num(pers, __le32_to_cpu(sb->level));
 335         printf("     Raid Level : %s\n", c?c:"-unknown-");
 336         printf("   Raid Devices : %d\n", __le32_to_cpu(sb->raid_disks));
 337         printf("\n");
 338         printf(" Avail Dev Size : %llu%s\n",
 339                (unsigned long long)__le64_to_cpu(sb->data_size),
 340                human_size(__le64_to_cpu(sb->data_size)<<9));
 341         if (__le32_to_cpu(sb->level) > 0) {
 342                 int ddsks = 0, ddsks_denom = 1;
 343                 switch(__le32_to_cpu(sb->level)) {
 344                 case 1: ddsks=1;break;
 345                 case 4:
 346                 case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
 347                 case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
 348                 case 10:
 349                         layout = __le32_to_cpu(sb->layout);
 350                         ddsks = __le32_to_cpu(sb->raid_disks);
 351                         ddsks_denom = (layout&255) * ((layout>>8)&255);
 352                 }
 353                 if (ddsks) {
 354                         long long asize = __le64_to_cpu(sb->size);
 355                         asize = (asize << 9) * ddsks / ddsks_denom;
 356                         printf("     Array Size : %llu%s\n",
 357                                asize >> 10,  human_size(asize));
 358                 }
 359                 if (sb->size != sb->data_size)
 360                         printf("  Used Dev Size : %llu%s\n",
 361                                (unsigned long long)__le64_to_cpu(sb->size),
 362                                human_size(__le64_to_cpu(sb->size)<<9));
 363         }
 364         if (sb->data_offset)
 365                 printf("    Data Offset : %llu sectors\n",
 366                        (unsigned long long)__le64_to_cpu(sb->data_offset));
 367         if (sb->new_offset &&
 368             (__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET)) {
 369                 unsigned long long offset = __le64_to_cpu(sb->data_offset);
 370                 offset += (signed)(int32_t)__le32_to_cpu(sb->new_offset);
 371                 printf("     New Offset : %llu sectors\n", offset);
 372         }
 373         printf("   Super Offset : %llu sectors\n",
 374                (unsigned long long)__le64_to_cpu(sb->super_offset));
 375         if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET)
 376                 printf("Recovery Offset : %llu sectors\n", (unsigned long long)__le64_to_cpu(sb->recovery_offset));
 377
 378         st->ss->getinfo_super(st, &info, NULL);
 379         if (info.space_after != 1 &&
 380             !(__le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
 381                 printf("   Unused Space : before=%llu sectors, after=%llu sectors\n",
 382                        info.space_before, info.space_after);
 383
 384         printf("          State : %s\n", (__le64_to_cpu(sb->resync_offset)+1)? "active":"clean");
 385         printf("    Device UUID : ");
 386         for (i=0; i<16; i++) {
 387                 if ((i&3)==0 && i != 0) printf(":");
 388                 printf("%02x", sb->device_uuid[i]);
 389         }
 390         printf("\n");
 391         printf("\n");
 392         if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
 393                 printf("Internal Bitmap : %ld sectors from superblock\n",
 394                        (long)(int32_t)__le32_to_cpu(sb->bitmap_offset));
 395         }
 396         if (sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)) {
 397                 printf("  Reshape pos'n : %llu%s\n", (unsigned long long)__le64_to_cpu(sb->reshape_position)/2,
 398                        human_size(__le64_to_cpu(sb->reshape_position)<<9));
 399                 if (__le32_to_cpu(sb->delta_disks)) {
 400                         printf("  Delta Devices : %d", __le32_to_cpu(sb->delta_disks));
 401                         printf(" (%d->%d)\n",
 402                                __le32_to_cpu(sb->raid_disks)-__le32_to_cpu(sb->delta_disks),
 403                                __le32_to_cpu(sb->raid_disks));
 404                         if ((int)__le32_to_cpu(sb->delta_disks) < 0)
 405                                 delta_extra = -__le32_to_cpu(sb->delta_disks);
 406                 }
 407                 if (__le32_to_cpu(sb->new_level) != __le32_to_cpu(sb->level)) {
 408                         c = map_num(pers, __le32_to_cpu(sb->new_level));
 409                         printf("      New Level : %s\n", c?c:"-unknown-");
 410                 }
 411                 if (__le32_to_cpu(sb->new_layout) != __le32_to_cpu(sb->layout)) {
 412                         if (__le32_to_cpu(sb->level) == 5) {
 413                                 c = map_num(r5layout, __le32_to_cpu(sb->new_layout));
 414                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 415                         }
 416                         if (__le32_to_cpu(sb->level) == 6) {
 417                                 c = map_num(r6layout, __le32_to_cpu(sb->new_layout));
 418                                 printf("     New Layout : %s\n", c?c:"-unknown-");
 419                         }
 420                         if (__le32_to_cpu(sb->level) == 10) {
 421                                 printf("     New Layout :");
 422                                 print_r10_layout(__le32_to_cpu(sb->new_layout));
 423                                 printf("\n");
 424                         }
 425                 }
 426                 if (__le32_to_cpu(sb->new_chunk) != __le32_to_cpu(sb->chunksize))
 427                         printf("  New Chunksize : %dK\n", __le32_to_cpu(sb->new_chunk)/2);
 428                 printf("\n");
 429         }
 430         if (sb->devflags) {
 431                 printf("          Flags :");
 432                 if (sb->devflags & WriteMostly1)
 433                         printf(" write-mostly");
 434                 if (sb->devflags & FailFast1)
 435                         printf(" failfast");
 436                 printf("\n");
 437         }
 438
 439         atime = __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL;
 440         printf("    Update Time : %.24s\n", ctime(&atime));
 441
 442         if (sb->bblog_size && sb->bblog_offset) {
 443                 printf("  Bad Block Log : %d entries available at offset %ld sectors",
 444                        __le16_to_cpu(sb->bblog_size)*512/8,
 445                        (long)(int32_t)__le32_to_cpu(sb->bblog_offset));
 446                 if (sb->feature_map &
 447                     __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
 448                         printf(" - bad blocks present.");
 449                 printf("\n");
 450         }
 451
 452         if (calc_sb_1_csum(sb) == sb->sb_csum)
 453                 printf("       Checksum : %x - correct\n", __le32_to_cpu(sb->sb_csum));
 454         else
 455                 printf("       Checksum : %x - expected %x\n", __le32_to_cpu(sb->sb_csum),
 456                        __le32_to_cpu(calc_sb_1_csum(sb)));
 457         printf("         Events : %llu\n", (unsigned long long)__le64_to_cpu(sb->events));
 458         printf("\n");
 459         if (__le32_to_cpu(sb->level) == 5) {
 460                 c = map_num(r5layout, __le32_to_cpu(sb->layout));
 461                 printf("         Layout : %s\n", c?c:"-unknown-");
 462         }
 463         if (__le32_to_cpu(sb->level) == 6) {
 464                 c = map_num(r6layout, __le32_to_cpu(sb->layout));
 465                 printf("         Layout : %s\n", c?c:"-unknown-");
 466         }
 467         if (__le32_to_cpu(sb->level) == 10) {
 468                 int lo = __le32_to_cpu(sb->layout);
 469                 printf("         Layout :");
 470                 print_r10_layout(lo);
 471                 printf("\n");
 472         }
 473         switch(__le32_to_cpu(sb->level)) {
 474         case 0:
 475         case 4:
 476         case 5:
 477         case 6:
 478         case 10:
 479                 printf("     Chunk Size : %dK\n", __le32_to_cpu(sb->chunksize)/2);
 480                 break;
 481         case -1:
 482                 printf("       Rounding : %dK\n", __le32_to_cpu(sb->chunksize)/2);
 483                 break;
 484         default: break;
 485         }
 486         printf("\n");
 487 #if 0
 488         /* This turns out to just be confusing */
 489         printf("    Array Slot : %d (", __le32_to_cpu(sb->dev_number));
 490         for (i= __le32_to_cpu(sb->max_dev); i> 0 ; i--)
 491                 if (__le16_to_cpu(sb->dev_roles[i-1]) != MD_DISK_ROLE_SPARE)
 492                         break;
 493         for (d=0; d < i; d++) {
 494                 int role = __le16_to_cpu(sb->dev_roles[d]);
 495                 if (d) printf(", ");
 496                 if (role == MD_DISK_ROLE_SPARE) printf("empty");
 497                 else if(role == MD_DISK_ROLE_FAULTY) printf("failed");
 498                 else printf("%d", role);
 499         }
 500         printf(")\n");
 501 #endif
 502         printf("   Device Role : ");
 503         role = role_from_sb(sb);
 504         if (role >= MD_DISK_ROLE_FAULTY)
 505                 printf("spare\n");
 506         else if (role == MD_DISK_ROLE_JOURNAL)
 507                 printf("Journal\n");
 508         else if (sb->feature_map & __cpu_to_le32(MD_FEATURE_REPLACEMENT))
 509                 printf("Replacement device %d\n", role);
 510         else
 511                 printf("Active device %d\n", role);
 512
 513         printf("   Array State : ");
 514         for (d=0; d<__le32_to_cpu(sb->raid_disks) + delta_extra; d++) {
 515                 int cnt = 0;
 516                 unsigned int i;
 517                 for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
 518                         unsigned int role = __le16_to_cpu(sb->dev_roles[i]);
 519                         if (role == d)
 520                                 cnt++;
 521                 }
 522                 if (cnt == 2)
 523                         printf("R");
 524                 else if (cnt == 1)
 525                         printf("A");
 526                 else if (cnt == 0)
 527                         printf(".");
 528                 else
 529                         printf("?");
 530         }
 531 #if 0
 532         /* This is confusing too */
 533         faulty = 0;
 534         for (i=0; i< __le32_to_cpu(sb->max_dev); i++) {
 535                 int role = __le16_to_cpu(sb->dev_roles[i]);
 536                 if (role == MD_DISK_ROLE_FAULTY)
 537                         faulty++;
 538         }
 539         if (faulty) printf(" %d failed", faulty);
 540 #endif
 541         printf(" ('A' == active, '.' == missing, 'R' == replacing)");
 542         printf("\n");
 543 }
 544
 545 static void brief_examine_super1(struct supertype *st, int verbose)
 546 {
 547         struct mdp_superblock_1 *sb = st->sb;
 548         int i;
 549         unsigned long long sb_offset;
 550         char *nm;
 551         char *c=map_num(pers, __le32_to_cpu(sb->level));
 552
 553         nm = strchr(sb->set_name, ':');
 554         if (nm)
 555                 nm++;
 556         else if (sb->set_name[0])
 557                 nm = sb->set_name;
 558         else
 559                 nm = NULL;
 560
 561         printf("ARRAY ");
 562         if (nm) {
 563                 printf("/dev/md/");
 564                 print_escape(nm);
 565                 putchar(' ');
 566         }
 567         if (verbose && c)
 568                 printf(" level=%s", c);
 569         sb_offset = __le64_to_cpu(sb->super_offset);
 570         if (sb_offset <= 4)
 571                 printf(" metadata=1.1 ");
 572         else if (sb_offset <= 8)
 573                 printf(" metadata=1.2 ");
 574         else
 575                 printf(" metadata=1.0 ");
 576         if (verbose)
 577                 printf("num-devices=%d ", __le32_to_cpu(sb->raid_disks));
 578         printf("UUID=");
 579         for (i=0; i<16; i++) {
 580                 if ((i&3)==0 && i != 0) printf(":");
 581                 printf("%02x", sb->set_uuid[i]);
 582         }
 583         if (sb->set_name[0]) {
 584                 printf(" name=");
 585                 print_quoted(sb->set_name);
 586         }
 587         printf("\n");
 588 }
 589
 590 static void export_examine_super1(struct supertype *st)
 591 {
 592         struct mdp_superblock_1 *sb = st->sb;
 593         int i;
 594         int len = 32;
 595         int layout;
 596
 597         printf("MD_LEVEL=%s\n", map_num(pers, __le32_to_cpu(sb->level)));
 598         printf("MD_DEVICES=%d\n", __le32_to_cpu(sb->raid_disks));
 599         for (i=0; i<32; i++)
 600                 if (sb->set_name[i] == '\n' ||
 601                     sb->set_name[i] == '\0') {
 602                         len = i;
 603                         break;
 604                 }
 605         if (len)
 606                 printf("MD_NAME=%.*s\n", len, sb->set_name);
 607         if (__le32_to_cpu(sb->level) > 0) {
 608                 int ddsks = 0, ddsks_denom = 1;
 609                 switch(__le32_to_cpu(sb->level)) {
 610                         case 1: ddsks=1;break;
 611                         case 4:
 612                         case 5: ddsks = __le32_to_cpu(sb->raid_disks)-1; break;
 613                         case 6: ddsks = __le32_to_cpu(sb->raid_disks)-2; break;
 614                         case 10:
 615                                 layout = __le32_to_cpu(sb->layout);
 616                                 ddsks = __le32_to_cpu(sb->raid_disks);
 617                                 ddsks_denom = (layout&255) * ((layout>>8)&255);
 618                         }
 619                 if (ddsks) {
 620                         long long asize = __le64_to_cpu(sb->size);
 621                         asize = (asize << 9) * ddsks / ddsks_denom;
 622                         printf("MD_ARRAY_SIZE=%s\n",human_size_brief(asize,JEDEC));
 623                 }
 624         }
 625         printf("MD_UUID=");
 626         for (i=0; i<16; i++) {
 627                 if ((i&3)==0 && i != 0) printf(":");
 628                 printf("%02x", sb->set_uuid[i]);
 629         }
 630         printf("\n");
 631         printf("MD_UPDATE_TIME=%llu\n",
 632                __le64_to_cpu(sb->utime) & 0xFFFFFFFFFFULL);
 633         printf("MD_DEV_UUID=");
 634         for (i=0; i<16; i++) {
 635                 if ((i&3)==0 && i != 0) printf(":");
 636                 printf("%02x", sb->device_uuid[i]);
 637         }
 638         printf("\n");
 639         printf("MD_EVENTS=%llu\n",
 640                (unsigned long long)__le64_to_cpu(sb->events));
 641 }
 642
 643 static int copy_metadata1(struct supertype *st, int from, int to)
 644 {
 645         /* Read superblock.  If it looks good, write it out.
 646          * Then if a bitmap is present, copy that.
 647          * And if a bad-block-list is present, copy that too.
 648          */
 649         void *buf;
 650         unsigned long long dsize, sb_offset;
 651         const int bufsize = 4*1024;
 652         struct mdp_superblock_1 super, *sb;
 653
 654         if (posix_memalign(&buf, 4096, bufsize) != 0)
 655                 return 1;
 656
 657         if (!get_dev_size(from, NULL, &dsize))
 658                 goto err;
 659
 660         dsize >>= 9;
 661         if (dsize < 24)
 662                 goto err;
 663         switch(st->minor_version) {
 664         case 0:
 665                 sb_offset = dsize;
 666                 sb_offset -= 8*2;
 667                 sb_offset &= ~(4*2-1);
 668                 break;
 669         case 1:
 670                 sb_offset = 0;
 671                 break;
 672         case 2:
 673                 sb_offset = 4*2;
 674                 break;
 675         default:
 676                 goto err;
 677         }
 678
 679         if (lseek64(from, sb_offset << 9, 0) < 0LL)
 680                 goto err;
 681         if (read(from, buf, bufsize) != bufsize)
 682                 goto err;
 683
 684         sb = buf;
 685         super = *sb; // save most of sb for when we reuse buf
 686
 687         if (__le32_to_cpu(super.magic) != MD_SB_MAGIC ||
 688             __le32_to_cpu(super.major_version) != 1 ||
 689             __le64_to_cpu(super.super_offset) != sb_offset ||
 690             calc_sb_1_csum(sb) != super.sb_csum)
 691                 goto err;
 692
 693         if (lseek64(to, sb_offset << 9, 0) < 0LL)
 694                 goto err;
 695         if (write(to, buf, bufsize) != bufsize)
 696                 goto err;
 697
 698         if (super.feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET)) {
 699                 unsigned long long bitmap_offset = sb_offset;
 700                 int bytes = 4096; // just an estimate.
 701                 int written = 0;
 702                 struct align_fd afrom, ato;
 703
 704                 init_afd(&afrom, from);
 705                 init_afd(&ato, to);
 706
 707                 bitmap_offset += (int32_t)__le32_to_cpu(super.bitmap_offset);
 708
 709                 if (lseek64(from, bitmap_offset<<9, 0) < 0)
 710                         goto err;
 711                 if (lseek64(to, bitmap_offset<<9, 0) < 0)
 712                         goto err;
 713
 714                 for (written = 0; written < bytes ; ) {
 715                         int n = bytes - written;
 716                         if (n > 4096)
 717                                 n = 4096;
 718                         if (aread(&afrom, buf, n) != n)
 719                                 goto err;
 720                         if (written == 0) {
 721                                 /* have the header, can calculate
 722                                  * correct bitmap bytes */
 723                                 bitmap_super_t *bms;
 724                                 bms = (void*)buf;
 725                                 bytes = calc_bitmap_size(bms, 512);
 726                                 if (n > bytes)
 727                                         n =  bytes;
 728                         }
 729                         if (awrite(&ato, buf, n) != n)
 730                                 goto err;
 731                         written += n;
 732                 }
 733         }
 734
 735         if (super.bblog_size != 0 &&
 736             __le16_to_cpu(super.bblog_size) <= 100 &&
 737             super.bblog_offset != 0 &&
 738             (super.feature_map & __le32_to_cpu(MD_FEATURE_BAD_BLOCKS))) {
 739                 /* There is a bad block log */
 740                 unsigned long long bb_offset = sb_offset;
 741                 int bytes = __le16_to_cpu(super.bblog_size) * 512;
 742                 int written = 0;
 743                 struct align_fd afrom, ato;
 744
 745                 init_afd(&afrom, from);
 746                 init_afd(&ato, to);
 747
 748                 bb_offset += (int32_t)__le32_to_cpu(super.bblog_offset);
 749
 750                 if (lseek64(from, bb_offset<<9, 0) < 0)
 751                         goto err;
 752                 if (lseek64(to, bb_offset<<9, 0) < 0)
 753                         goto err;
 754
 755                 for (written = 0; written < bytes ; ) {
 756                         int n = bytes - written;
 757                         if (n > 4096)
 758                                 n = 4096;
 759                         if (aread(&afrom, buf, n) != n)
 760                                 goto err;
 761
 762                         if (awrite(&ato, buf, n) != n)
 763                                 goto err;
 764                         written += n;
 765                 }
 766         }
 767
 768         free(buf);
 769         return 0;
 770
 771 err:
 772         free(buf);
 773         return 1;
 774 }
 775
 776 static void detail_super1(struct supertype *st, char *homehost)
 777 {
 778         struct mdp_superblock_1 *sb = st->sb;
 779         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
 780         int i;
 781         int l = homehost ? strlen(homehost) : 0;
 782
 783         printf("           Name : %.32s", sb->set_name);
 784         if (l > 0 && l < 32 &&
 785             sb->set_name[l] == ':' &&
 786             strncmp(sb->set_name, homehost, l) == 0)
 787                 printf("  (local to host %s)", homehost);
 788         if (bms->nodes > 0 && (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
 789             printf("\n   Cluster Name : %-64s", bms->cluster_name);
 790         printf("\n           UUID : ");
 791         for (i=0; i<16; i++) {
 792                 if ((i&3)==0 && i != 0) printf(":");
 793                 printf("%02x", sb->set_uuid[i]);
 794         }
 795         printf("\n         Events : %llu\n\n", (unsigned long long)__le64_to_cpu(sb->events));
 796 }
 797
 798 static void brief_detail_super1(struct supertype *st)
 799 {
 800         struct mdp_superblock_1 *sb = st->sb;
 801         int i;
 802
 803         if (sb->set_name[0]) {
 804                 printf(" name=");
 805                 print_quoted(sb->set_name);
 806         }
 807         printf(" UUID=");
 808         for (i=0; i<16; i++) {
 809                 if ((i&3)==0 && i != 0) printf(":");
 810                 printf("%02x", sb->set_uuid[i]);
 811         }
 812 }
 813
 814 static void export_detail_super1(struct supertype *st)
 815 {
 816         struct mdp_superblock_1 *sb = st->sb;
 817         int i;
 818         int len = 32;
 819
 820         for (i=0; i<32; i++)
 821                 if (sb->set_name[i] == '\n' ||
 822                     sb->set_name[i] == '\0') {
 823                         len = i;
 824                         break;
 825                 }
 826         if (len)
 827                 printf("MD_NAME=%.*s\n", len, sb->set_name);
 828 }
 829
 830 static int examine_badblocks_super1(struct supertype *st, int fd, char *devname)
 831 {
 832         struct mdp_superblock_1 *sb = st->sb;
 833         unsigned long long offset;
 834         int size;
 835         __u64 *bbl, *bbp;
 836         int i;
 837
 838         if  (!sb->bblog_size || __le16_to_cpu(sb->bblog_size) > 100
 839              || !sb->bblog_offset){
 840                 printf("No bad-blocks list configured on %s\n", devname);
 841                 return 0;
 842         }
 843         if ((sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
 844             == 0) {
 845                 printf("Bad-blocks list is empty in %s\n", devname);
 846                 return 0;
 847         }
 848
 849         size = __le16_to_cpu(sb->bblog_size)* 512;
 850         if (posix_memalign((void**)&bbl, 4096, size) != 0) {
 851                 pr_err("could not allocate badblocks list\n");
 852                 return 0;
 853         }
 854         offset = __le64_to_cpu(sb->super_offset) +
 855                 (int)__le32_to_cpu(sb->bblog_offset);
 856         offset <<= 9;
 857         if (lseek64(fd, offset, 0) < 0) {
 858                 pr_err("Cannot seek to bad-blocks list\n");
 859                 return 1;
 860         }
 861         if (read(fd, bbl, size) != size) {
 862                 pr_err("Cannot read bad-blocks list\n");
 863                 return 1;
 864         }
 865         /* 64bits per entry. 10 bits is block-count, 54 bits is block
 866          * offset.  Blocks are sectors unless bblog->shift makes them bigger
 867          */
 868         bbp = (__u64*)bbl;
 869         printf("Bad-blocks on %s:\n", devname);
 870         for (i = 0; i < size/8; i++, bbp++) {
 871                 __u64 bb = __le64_to_cpu(*bbp);
 872                 int count = bb & 0x3ff;
 873                 unsigned long long sector = bb >> 10;
 874
 875                 if (bb + 1 == 0)
 876                         break;
 877
 878                 sector <<= sb->bblog_shift;
 879                 count <<= sb->bblog_shift;
 880
 881                 printf("%20llu for %d sectors\n", sector, count);
 882         }
 883         return 0;
 884 }
 885
 886 #endif
 887
 888 static int match_home1(struct supertype *st, char *homehost)
 889 {
 890         struct mdp_superblock_1 *sb = st->sb;
 891         int l = homehost ? strlen(homehost) : 0;
 892
 893         return (l > 0 && l < 32 &&
 894                 sb->set_name[l] == ':' &&
 895                 strncmp(sb->set_name, homehost, l) == 0);
 896 }
 897
 898 static void uuid_from_super1(struct supertype *st, int uuid[4])
 899 {
 900         struct mdp_superblock_1 *super = st->sb;
 901         char *cuuid = (char*)uuid;
 902         int i;
 903         for (i=0; i<16; i++)
 904                 cuuid[i] = super->set_uuid[i];
 905 }
 906
 907 static void getinfo_super1(struct supertype *st, struct mdinfo *info, char *map)
 908 {
 909         struct mdp_superblock_1 *sb = st->sb;
 910         struct bitmap_super_s *bsb = (void*)(((char*)sb)+MAX_SB_SIZE);
 911         struct misc_dev_info *misc = (void*)(((char*)sb)+MAX_SB_SIZE+BM_SUPER_SIZE);
 912         int working = 0;
 913         unsigned int i;
 914         unsigned int role;
 915         unsigned int map_disks = info->array.raid_disks;
 916         unsigned long long super_offset;
 917         unsigned long long data_size;
 918
 919         memset(info, 0, sizeof(*info));
 920         info->array.major_version = 1;
 921         info->array.minor_version = st->minor_version;
 922         info->array.patch_version = 0;
 923         info->array.raid_disks = __le32_to_cpu(sb->raid_disks);
 924         info->array.level = __le32_to_cpu(sb->level);
 925         info->array.layout = __le32_to_cpu(sb->layout);
 926         info->array.md_minor = -1;
 927         info->array.ctime = __le64_to_cpu(sb->ctime);
 928         info->array.utime = __le64_to_cpu(sb->utime);
 929         info->array.chunk_size = __le32_to_cpu(sb->chunksize)*512;
 930         info->array.state =
 931                 (__le64_to_cpu(sb->resync_offset) == MaxSector)
 932                 ? 1 : 0;
 933         if (__le32_to_cpu(bsb->nodes) > 1)
 934                 info->array.state |= (1 << MD_SB_CLUSTERED);
 935
 936         info->data_offset = __le64_to_cpu(sb->data_offset);
 937         info->component_size = __le64_to_cpu(sb->size);
 938         if (sb->feature_map & __le32_to_cpu(MD_FEATURE_BITMAP_OFFSET))
 939                 info->bitmap_offset = (int32_t)__le32_to_cpu(sb->bitmap_offset);
 940
 941         info->disk.major = 0;
 942         info->disk.minor = 0;
 943         info->disk.number = __le32_to_cpu(sb->dev_number);
 944         if (__le32_to_cpu(sb->dev_number) >= __le32_to_cpu(sb->max_dev) ||
 945             __le32_to_cpu(sb->dev_number) >= MAX_DEVS)
 946                 role = MD_DISK_ROLE_FAULTY;
 947         else
 948                 role = __le16_to_cpu(sb->dev_roles[__le32_to_cpu(sb->dev_number)]);
 949
 950         super_offset = __le64_to_cpu(sb->super_offset);
 951         if (info->array.level <= 0)
 952                 data_size = __le64_to_cpu(sb->data_size);
 953         else
 954                 data_size = __le64_to_cpu(sb->size);
 955         if (info->data_offset < super_offset) {
 956                 unsigned long long end;
 957                 info->space_before = info->data_offset;
 958                 end = super_offset;
 959
 960                 if (sb->bblog_offset && sb->bblog_size) {
 961                         unsigned long long bboffset = super_offset;
 962                         bboffset += (int32_t)__le32_to_cpu(sb->bblog_offset);
 963                         if (bboffset < end)
 964                                 end = bboffset;
 965                 }
 966
 967                 if (super_offset + info->bitmap_offset < end)
 968                         end = super_offset + info->bitmap_offset;
 969
 970                 if (info->data_offset + data_size < end)
 971                         info->space_after = end - data_size - info->data_offset;
 972                 else
 973                         info->space_after = 0;
 974         } else {
 975                 unsigned long long earliest;
 976                 earliest = super_offset + (32+4)*2; /* match kernel */
 977                 if (info->bitmap_offset > 0) {
 978                         unsigned long long bmend = info->bitmap_offset;
 979                         unsigned long long size = calc_bitmap_size(bsb, 4096);
 980                         size /= 512;
 981                         bmend += size;
 982                         if (bmend > earliest)
 983                                 earliest = bmend;
 984                 }
 985                 if (sb->bblog_offset && sb->bblog_size) {
 986                         unsigned long long bbend = super_offset;
 987                         bbend += (int32_t)__le32_to_cpu(sb->bblog_offset);
 988                         bbend += __le16_to_cpu(sb->bblog_size);
 989                         if (bbend > earliest)
 990                                 earliest = bbend;
 991                 }
 992                 if (earliest < info->data_offset)
 993                         info->space_before = info->data_offset - earliest;
 994                 else
 995                         info->space_before = 0;
 996                 info->space_after = misc->device_size - data_size - info->data_offset;
 997         }
 998         if (info->space_before == 0 && info->space_after == 0) {
 999                 /* It will look like we don't support data_offset changes,
1000                  * be we do - it's just that there is no room.
1001                  * A change that reduced the number of devices should
1002                  * still be allowed, so set the otherwise useless value of '1'
1003                  */
1004                 info->space_after = 1;
1005         }
1006
1007         info->disk.raid_disk = -1;
1008         switch(role) {
1009         case MD_DISK_ROLE_SPARE:
1010                 info->disk.state = 0; /* spare: not active, not sync, not faulty */
1011                 break;
1012         case MD_DISK_ROLE_FAULTY:
1013                 info->disk.state = 1; /* faulty */
1014                 break;
1015         case MD_DISK_ROLE_JOURNAL:
1016                 info->disk.state = (1 << MD_DISK_JOURNAL);
1017                 info->disk.raid_disk = role;
1018                 info->space_after = (misc->device_size - info->data_offset) % 8; /* journal uses all 4kB blocks*/
1019                 break;
1020         default:
1021                 info->disk.state = 6; /* active and in sync */
1022                 info->disk.raid_disk = role;
1023         }
1024         if (sb->devflags & WriteMostly1)
1025                 info->disk.state |= (1 << MD_DISK_WRITEMOSTLY);
1026         if (sb->devflags & FailFast1)
1027                 info->disk.state |= (1 << MD_DISK_FAILFAST);
1028         info->events = __le64_to_cpu(sb->events);
1029         sprintf(info->text_version, "1.%d", st->minor_version);
1030         info->safe_mode_delay = 200;
1031
1032         memcpy(info->uuid, sb->set_uuid, 16);
1033
1034         strncpy(info->name, sb->set_name, 32);
1035         info->name[32] = 0;
1036
1037         if ((__le32_to_cpu(sb->feature_map)&MD_FEATURE_REPLACEMENT)) {
1038                 info->disk.state &= ~(1 << MD_DISK_SYNC);
1039                 info->disk.state |=  1 << MD_DISK_REPLACEMENT;
1040         }
1041
1042         if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RECOVERY_OFFSET))
1043                 info->recovery_start = __le32_to_cpu(sb->recovery_offset);
1044         else
1045                 info->recovery_start = MaxSector;
1046
1047         if (sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE)) {
1048                 info->reshape_active = 1;
1049                 if ((sb->feature_map & __le32_to_cpu(MD_FEATURE_NEW_OFFSET)) &&
1050                     sb->new_offset != 0)
1051                         info->reshape_active |= RESHAPE_NO_BACKUP;
1052                 info->reshape_progress = __le64_to_cpu(sb->reshape_position);
1053                 info->new_level = __le32_to_cpu(sb->new_level);
1054                 info->delta_disks = __le32_to_cpu(sb->delta_disks);
1055                 info->new_layout = __le32_to_cpu(sb->new_layout);
1056                 info->new_chunk = __le32_to_cpu(sb->new_chunk)<<9;
1057                 if (info->delta_disks < 0)
1058                         info->array.raid_disks -= info->delta_disks;
1059         } else
1060                 info->reshape_active = 0;
1061
1062         info->recovery_blocked = info->reshape_active;
1063
1064         if (map)
1065                 for (i=0; i<map_disks; i++)
1066                         map[i] = 0;
1067         for (i = 0; i < __le32_to_cpu(sb->max_dev); i++) {
1068                 role = __le16_to_cpu(sb->dev_roles[i]);
1069                 if (/*role == MD_DISK_ROLE_SPARE || */role < (unsigned) info->array.raid_disks) {
1070                         working++;
1071                         if (map && role < map_disks)
1072                                 map[role] = 1;
1073                 }
1074         }
1075
1076         info->array.working_disks = working;
1077         if (sb->feature_map & __le32_to_cpu(MD_FEATURE_JOURNAL))
1078                 info->journal_device_required = 1;
1079         info->journal_clean = 0;
1080 }
1081
1082 static struct mdinfo *container_content1(struct supertype *st, char *subarray)
1083 {
1084         struct mdinfo *info;
1085
1086         if (subarray)
1087                 return NULL;
1088
1089         info = xmalloc(sizeof(*info));
1090         getinfo_super1(st, info, NULL);
1091         return info;
1092 }
1093
1094 static int update_super1(struct supertype *st, struct mdinfo *info,
1095                          char *update,
1096                          char *devname, int verbose,
1097                          int uuid_set, char *homehost)
1098 {
1099         /* NOTE: for 'assemble' and 'force' we need to return non-zero
1100          * if any change was made.  For others, the return value is
1101          * ignored.
1102          */
1103         int rv = 0;
1104         int lockid;
1105         struct mdp_superblock_1 *sb = st->sb;
1106         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1107
1108         if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1109                 rv = cluster_get_dlmlock(&lockid);
1110                 if (rv) {
1111                         pr_err("Cannot get dlmlock in %s return %d\n", __func__, rv);
1112                         cluster_release_dlmlock(lockid);
1113                         return rv;
1114                 }
1115         }
1116
1117         if (strcmp(update, "homehost") == 0 &&
1118             homehost) {
1119                 /* Note that 'homehost' is special as it is really
1120                  * a "name" update.
1121                  */
1122                 char *c;
1123                 update = "name";
1124                 c = strchr(sb->set_name, ':');
1125                 if (c)
1126                         strncpy(info->name, c+1, 31 - (c-sb->set_name));
1127                 else
1128                         strncpy(info->name, sb->set_name, 32);
1129                 info->name[32] = 0;
1130         }
1131
1132         if (strcmp(update, "force-one")==0) {
1133                 /* Not enough devices for a working array,
1134                  * so bring this one up-to-date
1135                  */
1136                 if (sb->events != __cpu_to_le64(info->events))
1137                         rv = 1;
1138                 sb->events = __cpu_to_le64(info->events);
1139         } else if (strcmp(update, "force-array")==0) {
1140                 /* Degraded array and 'force' requests to
1141                  * maybe need to mark it 'clean'.
1142                  */
1143                 switch(__le32_to_cpu(sb->level)) {
1144                 case 5: case 4: case 6:
1145                         /* need to force clean */
1146                         if (sb->resync_offset != MaxSector)
1147                                 rv = 1;
1148                         sb->resync_offset = MaxSector;
1149                 }
1150         } else if (strcmp(update, "assemble")==0) {
1151                 int d = info->disk.number;
1152                 int want;
1153                 if (info->disk.state & (1<<MD_DISK_ACTIVE))
1154                         want = info->disk.raid_disk;
1155                 else if (info->disk.state & (1<<MD_DISK_JOURNAL))
1156                         want = MD_DISK_ROLE_JOURNAL;
1157                 else
1158                         want = MD_DISK_ROLE_SPARE;
1159                 if (sb->dev_roles[d] != __cpu_to_le16(want)) {
1160                         sb->dev_roles[d] = __cpu_to_le16(want);
1161                         rv = 1;
1162                 }
1163                 if (info->reshape_active &&
1164                     sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1165                     info->delta_disks >= 0 &&
1166                     info->reshape_progress < __le64_to_cpu(sb->reshape_position)) {
1167                         sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1168                         rv = 1;
1169                 }
1170                 if (info->reshape_active &&
1171                     sb->feature_map & __le32_to_cpu(MD_FEATURE_RESHAPE_ACTIVE) &&
1172                     info->delta_disks < 0 &&
1173                     info->reshape_progress > __le64_to_cpu(sb->reshape_position)) {
1174                         sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1175                         rv = 1;
1176                 }
1177         } else if (strcmp(update, "linear-grow-new") == 0) {
1178                 unsigned int i;
1179                 int fd;
1180                 unsigned int max = __le32_to_cpu(sb->max_dev);
1181
1182                 for (i=0 ; i < max ; i++)
1183                         if (__le16_to_cpu(sb->dev_roles[i]) >= MD_DISK_ROLE_FAULTY)
1184                                 break;
1185                 sb->dev_number = __cpu_to_le32(i);
1186                 info->disk.number = i;
1187                 if (max >= __le32_to_cpu(sb->max_dev))
1188                         sb->max_dev = __cpu_to_le32(max+1);
1189
1190                 random_uuid(sb->device_uuid);
1191
1192                 sb->dev_roles[i] =
1193                         __cpu_to_le16(info->disk.raid_disk);
1194
1195                 fd = open(devname, O_RDONLY);
1196                 if (fd >= 0) {
1197                         unsigned long long ds;
1198                         get_dev_size(fd, devname, &ds);
1199                         close(fd);
1200                         ds >>= 9;
1201                         if (__le64_to_cpu(sb->super_offset) <
1202                             __le64_to_cpu(sb->data_offset)) {
1203                                 sb->data_size = __cpu_to_le64(
1204                                         ds - __le64_to_cpu(sb->data_offset));
1205                         } else {
1206                                 ds -= 8*2;
1207                                 ds &= ~(unsigned long long)(4*2-1);
1208                                 sb->super_offset = __cpu_to_le64(ds);
1209                                 sb->data_size = __cpu_to_le64(
1210                                         ds - __le64_to_cpu(sb->data_offset));
1211                         }
1212                 }
1213         } else if (strcmp(update, "linear-grow-update") == 0) {
1214                 sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
1215                 sb->dev_roles[info->disk.number] =
1216                         __cpu_to_le16(info->disk.raid_disk);
1217         } else if (strcmp(update, "resync") == 0) {
1218                 /* make sure resync happens */
1219                 sb->resync_offset = 0ULL;
1220         } else if (strcmp(update, "uuid") == 0) {
1221                 copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
1222
1223                 if (__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)
1224                         memcpy(bms->uuid, sb->set_uuid, 16);
1225         } else if (strcmp(update, "no-bitmap") == 0) {
1226                 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1227         } else if (strcmp(update, "bbl") == 0) {
1228                 /* only possible if there is room after the bitmap, or if
1229                  * there is no bitmap
1230                  */
1231                 unsigned long long sb_offset = __le64_to_cpu(sb->super_offset);
1232                 unsigned long long data_offset = __le64_to_cpu(sb->data_offset);
1233                 long bitmap_offset = 0;
1234                 long bm_sectors = 0;
1235                 long space;
1236
1237 #ifndef MDASSEMBLE
1238                 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1239                         bitmap_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1240                         bm_sectors = calc_bitmap_size(bms, 4096) >> 9;
1241                 }
1242 #endif
1243                 if (sb_offset < data_offset) {
1244                         /* 1.1 or 1.2.  Put bbl after bitmap leaving at least 32K
1245                          */
1246                         long bb_offset;
1247                         bb_offset = sb_offset + 8;
1248                         if (bm_sectors && bitmap_offset > 0)
1249                                 bb_offset = bitmap_offset + bm_sectors;
1250                         while (bb_offset < (long)sb_offset + 8 + 32*2
1251                                && bb_offset + 8+8 <= (long)data_offset)
1252                                 /* too close to bitmap, and room to grow */
1253                                 bb_offset += 8;
1254                         if (bb_offset + 8 <= (long)data_offset) {
1255                                 sb->bblog_size = __cpu_to_le16(8);
1256                                 sb->bblog_offset = __cpu_to_le32(bb_offset);
1257                         }
1258                 } else {
1259                         /* 1.0 - Put bbl just before super block */
1260                         if (bm_sectors && bitmap_offset < 0)
1261                                 space = -bitmap_offset - bm_sectors;
1262                         else
1263                                 space = sb_offset - data_offset -
1264                                         __le64_to_cpu(sb->data_size);
1265                         if (space >= 8) {
1266                                 sb->bblog_size = __cpu_to_le16(8);
1267                                 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1268                         }
1269                 }
1270         } else if (strcmp(update, "no-bbl") == 0) {
1271                 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BAD_BLOCKS))
1272                         pr_err("Cannot remove active bbl from %s\n",devname);
1273                 else {
1274                         sb->bblog_size = 0;
1275                         sb->bblog_shift = 0;
1276                         sb->bblog_offset = 0;
1277                 }
1278         } else if (strcmp(update, "force-no-bbl") == 0) {
1279                 sb->feature_map &= ~ __cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1280                 sb->bblog_size = 0;
1281                 sb->bblog_shift = 0;
1282                 sb->bblog_offset = 0;
1283         } else if (strcmp(update, "name") == 0) {
1284                 if (info->name[0] == 0)
1285                         sprintf(info->name, "%d", info->array.md_minor);
1286                 memset(sb->set_name, 0, sizeof(sb->set_name));
1287                 if (homehost &&
1288                     strchr(info->name, ':') == NULL &&
1289                     strlen(homehost)+1+strlen(info->name) < 32) {
1290                         strcpy(sb->set_name, homehost);
1291                         strcat(sb->set_name, ":");
1292                         strcat(sb->set_name, info->name);
1293                 } else
1294                         strncpy(sb->set_name, info->name, sizeof(sb->set_name));
1295         } else if (strcmp(update, "devicesize") == 0 &&
1296             __le64_to_cpu(sb->super_offset) <
1297             __le64_to_cpu(sb->data_offset)) {
1298                 /* set data_size to device size less data_offset */
1299                 struct misc_dev_info *misc = (struct misc_dev_info*)
1300                         (st->sb + MAX_SB_SIZE + BM_SUPER_SIZE);
1301                 sb->data_size = __cpu_to_le64(
1302                         misc->device_size - __le64_to_cpu(sb->data_offset));
1303         } else if (strncmp(update, "revert-reshape", 14) == 0) {
1304                 rv = -2;
1305                 if (!(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE)))
1306                         pr_err("No active reshape to revert on %s\n",
1307                                devname);
1308                 else {
1309                         __u32 temp;
1310                         unsigned long long reshape_sectors;
1311                         long reshape_chunk;
1312                         rv = 0;
1313                         /* If the reshape hasn't started, just stop it.
1314                          * It is conceivable that a stripe was modified but
1315                          * the metadata not updated.  In that case the backup
1316                          * should have been used to get passed the critical stage.
1317                          * If that couldn't happen, the "-nobackup" version
1318                          * will be used.
1319                          */
1320                         if (strcmp(update, "revert-reshape-nobackup") == 0 &&
1321                             sb->reshape_position == 0 &&
1322                             (__le32_to_cpu(sb->delta_disks) > 0 ||
1323                              (__le32_to_cpu(sb->delta_disks) == 0 &&
1324                               !(sb->feature_map & __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS))))) {
1325                                 sb->feature_map &= ~__cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1326                                 sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1327                                                                __le32_to_cpu(sb->delta_disks));
1328                                 sb->delta_disks = 0;
1329                                 goto done;
1330                         }
1331                         /* reshape_position is a little messy.
1332                          * Its value must be a multiple of the larger
1333                          * chunk size, and of the "after" data disks.
1334                          * So when reverting we need to change it to
1335                          * be a multiple of the new "after" data disks,
1336                          * which is the old "before".
1337                          * If it isn't already a multiple of 'before',
1338                          * the only thing we could do would be
1339                          * copy some block around on the disks, which
1340                          * is easy to get wrong.
1341                          * So we reject a revert-reshape unless the
1342                          * alignment is good.
1343                          */
1344                         if (__le32_to_cpu(sb->level) >= 4 &&
1345                             __le32_to_cpu(sb->level) <= 6) {
1346                                 reshape_sectors = __le64_to_cpu(sb->reshape_position);
1347                                 reshape_chunk = __le32_to_cpu(sb->new_chunk);
1348                                 reshape_chunk *= __le32_to_cpu(sb->raid_disks) - __le32_to_cpu(sb->delta_disks) -
1349                                         (__le32_to_cpu(sb->level)==6 ? 2 : 1);
1350                                 if (reshape_sectors % reshape_chunk) {
1351                                         pr_err("Reshape position is not suitably aligned.\n");
1352                                         pr_err("Try normal assembly and stop again\n");
1353                                         return -2;
1354                                 }
1355                         }
1356                         sb->raid_disks = __cpu_to_le32(__le32_to_cpu(sb->raid_disks) -
1357                                                        __le32_to_cpu(sb->delta_disks));
1358                         if (sb->delta_disks == 0)
1359                                 sb->feature_map ^= __cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1360                         else
1361                                 sb->delta_disks = __cpu_to_le32(-__le32_to_cpu(sb->delta_disks));
1362
1363                         temp = sb->new_layout;
1364                         sb->new_layout = sb->layout;
1365                         sb->layout = temp;
1366
1367                         temp = sb->new_chunk;
1368                         sb->new_chunk = sb->chunksize;
1369                         sb->chunksize = temp;
1370
1371                         if (sb->feature_map & __cpu_to_le32(MD_FEATURE_NEW_OFFSET)) {
1372                                 long offset_delta = (int32_t)__le32_to_cpu(sb->new_offset);
1373                                 sb->data_offset = __cpu_to_le64(__le64_to_cpu(sb->data_offset) + offset_delta);
1374                                 sb->new_offset = __cpu_to_le32(-offset_delta);
1375                                 sb->data_size = __cpu_to_le64(__le64_to_cpu(sb->data_size) - offset_delta);
1376                         }
1377                 done:;
1378                 }
1379         } else if (strcmp(update, "_reshape_progress")==0)
1380                 sb->reshape_position = __cpu_to_le64(info->reshape_progress);
1381         else if (strcmp(update, "writemostly")==0)
1382                 sb->devflags |= WriteMostly1;
1383         else if (strcmp(update, "readwrite")==0)
1384                 sb->devflags &= ~WriteMostly1;
1385         else if (strcmp(update, "failfast") == 0)
1386                 sb->devflags |= FailFast1;
1387         else if (strcmp(update, "nofailfast") == 0)
1388                 sb->devflags &= ~FailFast1;
1389         else
1390                 rv = -1;
1391
1392         sb->sb_csum = calc_sb_1_csum(sb);
1393         if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1394                 cluster_release_dlmlock(lockid);
1395
1396         return rv;
1397 }
1398
1399 static int init_super1(struct supertype *st, mdu_array_info_t *info,
1400                        unsigned long long size, char *name, char *homehost,
1401                        int *uuid, unsigned long long data_offset)
1402 {
1403         struct mdp_superblock_1 *sb;
1404         int spares;
1405         char defname[10];
1406         int sbsize;
1407
1408         if (posix_memalign((void**)&sb, 4096, SUPER1_SIZE) != 0) {
1409                 pr_err("could not allocate superblock\n");
1410                 return 0;
1411         }
1412         memset(sb, 0, SUPER1_SIZE);
1413
1414         st->sb = sb;
1415         if (info == NULL) {
1416                 /* zeroing superblock */
1417                 return 0;
1418         }
1419
1420         spares = info->working_disks - info->active_disks;
1421         if (info->raid_disks + spares  > MAX_DEVS) {
1422                 pr_err("too many devices requested: %d+%d > %d\n",
1423                         info->raid_disks , spares, MAX_DEVS);
1424                 return 0;
1425         }
1426
1427         sb->magic = __cpu_to_le32(MD_SB_MAGIC);
1428         sb->major_version = __cpu_to_le32(1);
1429         sb->feature_map = 0;
1430         sb->pad0 = 0;
1431
1432         if (uuid)
1433                 copy_uuid(sb->set_uuid, uuid, super1.swapuuid);
1434         else
1435                 random_uuid(sb->set_uuid);;
1436
1437         if (name == NULL || *name == 0) {
1438                 sprintf(defname, "%d", info->md_minor);
1439                 name = defname;
1440         }
1441         if (homehost &&
1442             strchr(name, ':')== NULL &&
1443             strlen(homehost)+1+strlen(name) < 32) {
1444                 strcpy(sb->set_name, homehost);
1445                 strcat(sb->set_name, ":");
1446                 strcat(sb->set_name, name);
1447         } else
1448                 strncpy(sb->set_name, name, sizeof(sb->set_name));
1449
1450         sb->ctime = __cpu_to_le64((unsigned long long)time(0));
1451         sb->level = __cpu_to_le32(info->level);
1452         sb->layout = __cpu_to_le32(info->layout);
1453         sb->size = __cpu_to_le64(size*2ULL);
1454         sb->chunksize = __cpu_to_le32(info->chunk_size>>9);
1455         sb->raid_disks = __cpu_to_le32(info->raid_disks);
1456
1457         sb->data_offset = __cpu_to_le64(data_offset);
1458         sb->data_size = __cpu_to_le64(0);
1459         sb->super_offset = __cpu_to_le64(0);
1460         sb->recovery_offset = __cpu_to_le64(0);
1461
1462         sb->utime = sb->ctime;
1463         sb->events = __cpu_to_le64(1);
1464         if (info->state & (1<<MD_SB_CLEAN))
1465                 sb->resync_offset = MaxSector;
1466         else
1467                 sb->resync_offset = 0;
1468         sbsize = sizeof(struct mdp_superblock_1) + 2 * (info->raid_disks + spares);
1469         sbsize = ROUND_UP(sbsize, 512);
1470         sb->max_dev = __cpu_to_le32((sbsize - sizeof(struct mdp_superblock_1)) / 2);
1471
1472         memset(sb->dev_roles, 0xff, MAX_SB_SIZE - sizeof(struct mdp_superblock_1));
1473
1474         return 1;
1475 }
1476
1477 struct devinfo {
1478         int fd;
1479         char *devname;
1480         long long data_offset;
1481         mdu_disk_info_t disk;
1482         struct devinfo *next;
1483 };
1484 #ifndef MDASSEMBLE
1485 /* Add a device to the superblock being created */
1486 static int add_to_super1(struct supertype *st, mdu_disk_info_t *dk,
1487                          int fd, char *devname, unsigned long long data_offset)
1488 {
1489         struct mdp_superblock_1 *sb = st->sb;
1490         __u16 *rp = sb->dev_roles + dk->number;
1491         struct devinfo *di, **dip;
1492         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1493         int rv, lockid;
1494
1495         if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1496                 rv = cluster_get_dlmlock(&lockid);
1497                 if (rv) {
1498                         pr_err("Cannot get dlmlock in %s return %d\n", __func__, rv);
1499                         cluster_release_dlmlock(lockid);
1500                         return rv;
1501                 }
1502         }
1503
1504         if ((dk->state & 6) == 6) /* active, sync */
1505                 *rp = __cpu_to_le16(dk->raid_disk);
1506         else if (dk->state & (1<<MD_DISK_JOURNAL))
1507                 *rp = MD_DISK_ROLE_JOURNAL;
1508         else if ((dk->state & ~2) == 0) /* active or idle -> spare */
1509                 *rp = MD_DISK_ROLE_SPARE;
1510         else
1511                 *rp = MD_DISK_ROLE_FAULTY;
1512
1513         if (dk->number >= (int)__le32_to_cpu(sb->max_dev) &&
1514             __le32_to_cpu(sb->max_dev) < MAX_DEVS)
1515                 sb->max_dev = __cpu_to_le32(dk->number+1);
1516
1517         sb->dev_number = __cpu_to_le32(dk->number);
1518         sb->devflags = 0; /* don't copy another disks flags */
1519         sb->sb_csum = calc_sb_1_csum(sb);
1520
1521         dip = (struct devinfo **)&st->info;
1522         while (*dip)
1523                 dip = &(*dip)->next;
1524         di = xmalloc(sizeof(struct devinfo));
1525         di->fd = fd;
1526         di->devname = devname;
1527         di->disk = *dk;
1528         di->data_offset = data_offset;
1529         di->next = NULL;
1530         *dip = di;
1531
1532         if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1533                 cluster_release_dlmlock(lockid);
1534
1535         return 0;
1536 }
1537 #endif
1538
1539 static int locate_bitmap1(struct supertype *st, int fd, int node_num);
1540
1541 static int store_super1(struct supertype *st, int fd)
1542 {
1543         struct mdp_superblock_1 *sb = st->sb;
1544         unsigned long long sb_offset;
1545         struct align_fd afd;
1546         int sbsize;
1547         unsigned long long dsize;
1548         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
1549         int rv, lockid;
1550
1551         if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready()) {
1552                 rv = cluster_get_dlmlock(&lockid);
1553                 if (rv) {
1554                         pr_err("Cannot get dlmlock in %s return %d\n", __func__, rv);
1555                         cluster_release_dlmlock(lockid);
1556                         return rv;
1557                 }
1558         }
1559
1560         if (!get_dev_size(fd, NULL, &dsize))
1561                 return 1;
1562
1563         dsize >>= 9;
1564
1565         if (dsize < 24)
1566                 return 2;
1567
1568         init_afd(&afd, fd);
1569
1570         /*
1571          * Calculate the position of the superblock.
1572          * It is always aligned to a 4K boundary and
1573          * depending on minor_version, it can be:
1574          * 0: At least 8K, but less than 12K, from end of device
1575          * 1: At start of device
1576          * 2: 4K from start of device.
1577          */
1578         switch(st->minor_version) {
1579         case 0:
1580                 sb_offset = dsize;
1581                 sb_offset -= 8*2;
1582                 sb_offset &= ~(4*2-1);
1583                 break;
1584         case 1:
1585                 sb_offset = 0;
1586                 break;
1587         case 2:
1588                 sb_offset = 4*2;
1589                 break;
1590         default:
1591                 return -EINVAL;
1592         }
1593
1594         if (sb_offset != __le64_to_cpu(sb->super_offset) &&
1595             0 != __le64_to_cpu(sb->super_offset)
1596                 ) {
1597                 pr_err("internal error - sb_offset is wrong\n");
1598                 abort();
1599         }
1600
1601         if (lseek64(fd, sb_offset << 9, 0)< 0LL)
1602                 return 3;
1603
1604         sbsize = ROUND_UP(sizeof(*sb) + 2 * __le32_to_cpu(sb->max_dev), 512);
1605
1606         if (awrite(&afd, sb, sbsize) != sbsize)
1607                 return 4;
1608
1609         if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1610                 struct bitmap_super_s *bm = (struct bitmap_super_s*)
1611                         (((char*)sb)+MAX_SB_SIZE);
1612                 if (__le32_to_cpu(bm->magic) == BITMAP_MAGIC) {
1613                         locate_bitmap1(st, fd, 0);
1614                         if (awrite(&afd, bm, sizeof(*bm)) != sizeof(*bm))
1615                                 return 5;
1616                 }
1617         }
1618         fsync(fd);
1619         if (bms->version == BITMAP_MAJOR_CLUSTERED && dlm_funs_ready())
1620                 cluster_release_dlmlock(lockid);
1621
1622         return 0;
1623 }
1624
1625 static int load_super1(struct supertype *st, int fd, char *devname);
1626
1627 static unsigned long choose_bm_space(unsigned long devsize)
1628 {
1629         /* if the device is bigger than 8Gig, save 64k for bitmap usage,
1630          * if bigger than 200Gig, save 128k
1631          * NOTE: result must be multiple of 4K else bad things happen
1632          * on 4K-sector devices.
1633          */
1634         if (devsize < 64*2)
1635                 return 0;
1636         if (devsize - 64*2 >= 200*1024*1024*2)
1637                 return 128*2;
1638         if (devsize - 4*2 > 8*1024*1024*2)
1639                 return 64*2;
1640         return 4*2;
1641 }
1642
1643 static void free_super1(struct supertype *st);
1644
1645 #define META_BLOCK_SIZE 4096
1646 __u32 crc32c_le(__u32 crc, unsigned char const *p, size_t len);
1647
1648 #ifndef MDASSEMBLE
1649 static int write_empty_r5l_meta_block(struct supertype *st, int fd)
1650 {
1651         struct r5l_meta_block *mb;
1652         struct mdp_superblock_1 *sb = st->sb;
1653         struct align_fd afd;
1654         __u32 crc;
1655
1656         init_afd(&afd, fd);
1657
1658         if (posix_memalign((void**)&mb, 4096, META_BLOCK_SIZE) != 0) {
1659                 pr_err("Could not allocate memory for the meta block.\n");
1660                 return 1;
1661         }
1662
1663         memset(mb, 0, META_BLOCK_SIZE);
1664
1665         mb->magic = __cpu_to_le32(R5LOG_MAGIC);
1666         mb->version = R5LOG_VERSION;
1667         mb->meta_size = __cpu_to_le32(sizeof(struct r5l_meta_block));
1668         mb->seq = __cpu_to_le64(random32());
1669         mb->position = __cpu_to_le64(0);
1670
1671         crc = crc32c_le(0xffffffff, sb->set_uuid, sizeof(sb->set_uuid));
1672         crc = crc32c_le(crc, (void *)mb, META_BLOCK_SIZE);
1673         mb->checksum = crc;
1674
1675         if (lseek64(fd, (sb->data_offset) * 512, 0) < 0LL) {
1676                 pr_err("cannot seek to offset of the meta block\n");
1677                 goto fail_to_write;
1678         }
1679
1680         if (awrite(&afd, mb, META_BLOCK_SIZE) != META_BLOCK_SIZE) {
1681                 pr_err("failed to store write the meta block \n");
1682                 goto fail_to_write;
1683         }
1684         fsync(fd);
1685
1686         free(mb);
1687         return 0;
1688
1689 fail_to_write:
1690         free(mb);
1691         return 1;
1692 }
1693
1694 static int write_init_super1(struct supertype *st)
1695 {
1696         struct mdp_superblock_1 *sb = st->sb;
1697         struct supertype *refst;
1698         int rv = 0;
1699         unsigned long long bm_space;
1700         struct devinfo *di;
1701         unsigned long long dsize, array_size;
1702         unsigned long long sb_offset;
1703         unsigned long long data_offset;
1704         long bm_offset;
1705
1706         for (di = st->info; di; di = di->next) {
1707                 if (di->disk.state & (1 << MD_DISK_JOURNAL))
1708                         sb->feature_map |= MD_FEATURE_JOURNAL;
1709         }
1710
1711         for (di = st->info; di; di = di->next) {
1712                 if (di->disk.state & (1 << MD_DISK_FAULTY))
1713                         continue;
1714                 if (di->fd < 0)
1715                         continue;
1716
1717                 while (Kill(di->devname, NULL, 0, -1, 1) == 0)
1718                         ;
1719
1720                 sb->dev_number = __cpu_to_le32(di->disk.number);
1721                 if (di->disk.state & (1<<MD_DISK_WRITEMOSTLY))
1722                         sb->devflags |= WriteMostly1;
1723                 else
1724                         sb->devflags &= ~WriteMostly1;
1725                 if (di->disk.state & (1<<MD_DISK_FAILFAST))
1726                         sb->devflags |= FailFast1;
1727                 else
1728                         sb->devflags &= ~FailFast1;
1729
1730                 random_uuid(sb->device_uuid);
1731
1732                 if (!(di->disk.state & (1<<MD_DISK_JOURNAL)))
1733                         sb->events = 0;
1734
1735                 refst = dup_super(st);
1736                 if (load_super1(refst, di->fd, NULL)==0) {
1737                         struct mdp_superblock_1 *refsb = refst->sb;
1738
1739                         memcpy(sb->device_uuid, refsb->device_uuid, 16);
1740                         if (memcmp(sb->set_uuid, refsb->set_uuid, 16)==0) {
1741                                 /* same array, so preserve events and
1742                                  * dev_number */
1743                                 sb->events = refsb->events;
1744                                 /* bugs in 2.6.17 and earlier mean the
1745                                  * dev_number chosen in Manage must be preserved
1746                                  */
1747                                 if (get_linux_version() >= 2006018)
1748                                         sb->dev_number = refsb->dev_number;
1749                         }
1750                         free_super1(refst);
1751                 }
1752                 free(refst);
1753
1754                 if (!get_dev_size(di->fd, NULL, &dsize)) {
1755                         rv = 1;
1756                         goto error_out;
1757                 }
1758                 dsize >>= 9;
1759
1760                 if (dsize < 24) {
1761                         close(di->fd);
1762                         rv = 2;
1763                         goto error_out;
1764                 }
1765
1766                 /*
1767                  * Calculate the position of the superblock.
1768                  * It is always aligned to a 4K boundary and
1769                  * depending on minor_version, it can be:
1770                  * 0: At least 8K, but less than 12K, from end of device
1771                  * 1: At start of device
1772                  * 2: 4K from start of device.
1773                  * data_offset has already been set.
1774                  */
1775                 array_size = __le64_to_cpu(sb->size);
1776
1777                 /* work out how much space we left for a bitmap */
1778                 if (sb->feature_map & __cpu_to_le32(MD_FEATURE_BITMAP_OFFSET)) {
1779                         bitmap_super_t *bms = (bitmap_super_t *)
1780                                         (((char *)sb) + MAX_SB_SIZE);
1781                         bm_space = calc_bitmap_size(bms, 4096) >> 9;
1782                         bm_offset = (long)__le32_to_cpu(sb->bitmap_offset);
1783                 } else {
1784                         bm_space = choose_bm_space(array_size);
1785                         bm_offset = 8;
1786                 }
1787
1788                 data_offset = di->data_offset;
1789                 if (data_offset == INVALID_SECTORS)
1790                         data_offset = st->data_offset;
1791                 switch(st->minor_version) {
1792                 case 0:
1793                         /* Add 8 sectors for bad block log */
1794                         bm_space += 8;
1795                         if (data_offset == INVALID_SECTORS)
1796                                 data_offset = 0;
1797                         sb_offset = dsize;
1798                         sb_offset -= 8*2;
1799                         sb_offset &= ~(4*2-1);
1800                         sb->data_offset = __cpu_to_le64(data_offset);
1801                         sb->super_offset = __cpu_to_le64(sb_offset);
1802                         if (sb_offset < array_size + bm_space)
1803                                 bm_space = sb_offset - array_size;
1804                         sb->data_size = __cpu_to_le64(sb_offset - bm_space);
1805                         if (bm_space >= 8) {
1806                                 sb->bblog_size = __cpu_to_le16(8);
1807                                 sb->bblog_offset = __cpu_to_le32((unsigned)-8);
1808                         }
1809                         break;
1810                 case 1:
1811                 case 2:
1812                         sb_offset = st->minor_version == 2 ? 8 : 0;
1813                         sb->super_offset = __cpu_to_le64(sb_offset);
1814                         if (data_offset == INVALID_SECTORS)
1815                                 data_offset = sb_offset + 16;
1816
1817                         sb->data_offset = __cpu_to_le64(data_offset);
1818                         sb->data_size = __cpu_to_le64(dsize - data_offset);
1819                         if (data_offset >= sb_offset+bm_offset+bm_space+8) {
1820                                 sb->bblog_size = __cpu_to_le16(8);
1821                                 sb->bblog_offset = __cpu_to_le32(bm_offset +
1822                                                                  bm_space);
1823                         } else if (data_offset >= sb_offset + 16) {
1824                                 sb->bblog_size = __cpu_to_le16(8);
1825                                 /* '8' sectors for the bblog, and 'sb_offset'
1826                                  * because we want offset from superblock, not
1827                                  * start of device.
1828                                  */
1829                                 sb->bblog_offset = __cpu_to_le32(data_offset -
1830                                                                  8 - sb_offset);
1831                         }
1832                         break;
1833                 default:
1834                         pr_err("Failed to write invalid metadata format 1.%i to %s\n",
1835                                st->minor_version, di->devname);
1836                         rv = -EINVAL;
1837                         goto out;
1838                 }
1839                 /* Disable badblock log on clusters, or when explicitly requested */
1840                 if (st->nodes > 0 || conf_get_create_info()->bblist == 0) {
1841                         sb->bblog_size = 0;
1842                         sb->bblog_offset = 0;
1843                 }
1844
1845                 sb->sb_csum = calc_sb_1_csum(sb);
1846                 rv = store_super1(st, di->fd);
1847
1848                 if (rv == 0 && (di->disk.state & (1 << MD_DISK_JOURNAL))) {
1849                         rv = write_empty_r5l_meta_block(st, di->fd);
1850                         if (rv)
1851                                 goto error_out;
1852                 }
1853
1854                 if (rv == 0 && (__le32_to_cpu(sb->feature_map) & 1))
1855                         rv = st->ss->write_bitmap(st, di->fd, NodeNumUpdate);
1856                 close(di->fd);
1857                 di->fd = -1;
1858                 if (rv)
1859                         goto error_out;
1860         }
1861 error_out:
1862         if (rv)
1863                 pr_err("Failed to write metadata to %s\n",
1864                        di->devname);
1865 out:
1866         return rv;
1867 }
1868 #endif
1869
1870 static int compare_super1(struct supertype *st, struct supertype *tst)
1871 {
1872         /*
1873          * return:
1874          *  0 same, or first was empty, and second was copied
1875          *  1 second had wrong number
1876          *  2 wrong uuid
1877          *  3 wrong other info
1878          */
1879         struct mdp_superblock_1 *first = st->sb;
1880         struct mdp_superblock_1 *second = tst->sb;
1881
1882         if (second->magic != __cpu_to_le32(MD_SB_MAGIC))
1883                 return 1;
1884         if (second->major_version != __cpu_to_le32(1))
1885                 return 1;
1886
1887         if (!first) {
1888                 if (posix_memalign((void**)&first, 4096, SUPER1_SIZE) != 0) {
1889                         pr_err("could not allocate superblock\n");
1890                         return 1;
1891                 }
1892                 memcpy(first, second, SUPER1_SIZE);
1893                 st->sb = first;
1894                 return 0;
1895         }
1896         if (memcmp(first->set_uuid, second->set_uuid, 16)!= 0)
1897                 return 2;
1898
1899         if (first->ctime      != second->ctime     ||
1900             first->level      != second->level     ||
1901             first->layout     != second->layout    ||
1902             first->size       != second->size      ||
1903             first->chunksize  != second->chunksize ||
1904             first->raid_disks != second->raid_disks)
1905                 return 3;
1906         return 0;
1907 }
1908
1909 static int load_super1(struct supertype *st, int fd, char *devname)
1910 {
1911         unsigned long long dsize;
1912         unsigned long long sb_offset;
1913         struct mdp_superblock_1 *super;
1914         int uuid[4];
1915         struct bitmap_super_s *bsb;
1916         struct misc_dev_info *misc;
1917         struct align_fd afd;
1918
1919         free_super1(st);
1920
1921         init_afd(&afd, fd);
1922
1923         if (st->ss == NULL || st->minor_version == -1) {
1924                 int bestvers = -1;
1925                 struct supertype tst;
1926                 __u64 bestctime = 0;
1927                 /* guess... choose latest ctime */
1928                 memset(&tst, 0, sizeof(tst));
1929                 tst.ss = &super1;
1930                 for (tst.minor_version = 0; tst.minor_version <= 2 ; tst.minor_version++) {
1931                         switch(load_super1(&tst, fd, devname)) {
1932                         case 0: super = tst.sb;
1933                                 if (bestvers == -1 ||
1934                                     bestctime < __le64_to_cpu(super->ctime)) {
1935                                         bestvers = tst.minor_version;
1936                                         bestctime = __le64_to_cpu(super->ctime);
1937                                 }
1938                                 free(super);
1939                                 tst.sb = NULL;
1940                                 break;
1941                         case 1: return 1; /*bad device */
1942                         case 2: break; /* bad, try next */
1943                         }
1944                 }
1945                 if (bestvers != -1) {
1946                         int rv;
1947                         tst.minor_version = bestvers;
1948                         tst.ss = &super1;
1949                         tst.max_devs = MAX_DEVS;
1950                         rv = load_super1(&tst, fd, devname);
1951                         if (rv == 0)
1952                                 *st = tst;
1953                         return rv;
1954                 }
1955                 return 2;
1956         }
1957         if (!get_dev_size(fd, devname, &dsize))
1958                 return 1;
1959         dsize >>= 9;
1960
1961         if (dsize < 24) {
1962                 if (devname)
1963                         pr_err("%s is too small for md: size is %llu sectors.\n",
1964                                 devname, dsize);
1965                 return 1;
1966         }
1967
1968         /*
1969          * Calculate the position of the superblock.
1970          * It is always aligned to a 4K boundary and
1971          * depending on minor_version, it can be:
1972          * 0: At least 8K, but less than 12K, from end of device
1973          * 1: At start of device
1974          * 2: 4K from start of device.
1975          */
1976         switch(st->minor_version) {
1977         case 0:
1978                 sb_offset = dsize;
1979                 sb_offset -= 8*2;
1980                 sb_offset &= ~(4*2-1);
1981                 break;
1982         case 1:
1983                 sb_offset = 0;
1984                 break;
1985         case 2:
1986                 sb_offset = 4*2;
1987                 break;
1988         default:
1989                 return -EINVAL;
1990         }
1991
1992         if (lseek64(fd, sb_offset << 9, 0)< 0LL) {
1993                 if (devname)
1994                         pr_err("Cannot seek to superblock on %s: %s\n",
1995                                 devname, strerror(errno));
1996                 return 1;
1997         }
1998
1999         if (posix_memalign((void**)&super, 4096, SUPER1_SIZE) != 0) {
2000                 pr_err("could not allocate superblock\n");
2001                 return 1;
2002         }
2003
2004         memset(super, 0, SUPER1_SIZE);
2005
2006         if (aread(&afd, super, MAX_SB_SIZE) != MAX_SB_SIZE) {
2007                 if (devname)
2008                         pr_err("Cannot read superblock on %s\n",
2009                                 devname);
2010                 free(super);
2011                 return 1;
2012         }
2013
2014         if (__le32_to_cpu(super->magic) != MD_SB_MAGIC) {
2015                 if (devname)
2016                         pr_err("No super block found on %s (Expected magic %08x, got %08x)\n",
2017                                 devname, MD_SB_MAGIC, __le32_to_cpu(super->magic));
2018                 free(super);
2019                 return 2;
2020         }
2021
2022         if (__le32_to_cpu(super->major_version) != 1) {
2023                 if (devname)
2024                         pr_err("Cannot interpret superblock on %s - version is %d\n",
2025                                 devname, __le32_to_cpu(super->major_version));
2026                 free(super);
2027                 return 2;
2028         }
2029         if (__le64_to_cpu(super->super_offset) != sb_offset) {
2030                 if (devname)
2031                         pr_err("No superblock found on %s (super_offset is wrong)\n",
2032                                 devname);
2033                 free(super);
2034                 return 2;
2035         }
2036         st->sb = super;
2037
2038         bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
2039
2040         misc = (struct misc_dev_info*) (((char*)super)+MAX_SB_SIZE+BM_SUPER_SIZE);
2041         misc->device_size = dsize;
2042         if (st->data_offset == INVALID_SECTORS)
2043                 st->data_offset = __le64_to_cpu(super->data_offset);
2044
2045         /* Now check on the bitmap superblock */
2046         if ((__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) == 0)
2047                 return 0;
2048         /* Read the bitmap superblock and make sure it looks
2049          * valid.  If it doesn't clear the bit.  An --assemble --force
2050          * should get that written out.
2051          */
2052         locate_bitmap1(st, fd, 0);
2053         if (aread(&afd, bsb, 512) != 512)
2054                 goto no_bitmap;
2055
2056         uuid_from_super1(st, uuid);
2057         if (__le32_to_cpu(bsb->magic) != BITMAP_MAGIC ||
2058             memcmp(bsb->uuid, uuid, 16) != 0)
2059                 goto no_bitmap;
2060         return 0;
2061
2062  no_bitmap:
2063         super->feature_map = __cpu_to_le32(__le32_to_cpu(super->feature_map)
2064                                            & ~MD_FEATURE_BITMAP_OFFSET);
2065         return 0;
2066 }
2067
2068 static struct supertype *match_metadata_desc1(char *arg)
2069 {
2070         struct supertype *st = xcalloc(1, sizeof(*st));
2071
2072         st->container_devnm[0] = 0;
2073         st->ss = &super1;
2074         st->max_devs = MAX_DEVS;
2075         st->sb = NULL;
2076         st->data_offset = INVALID_SECTORS;
2077         /* leading zeros can be safely ignored.  --detail generates them. */
2078         while (*arg == '0')
2079                 arg++;
2080         if (strcmp(arg, "1.0") == 0 ||
2081             strcmp(arg, "1.00") == 0) {
2082                 st->minor_version = 0;
2083                 return st;
2084         }
2085         if (strcmp(arg, "1.1") == 0 ||
2086             strcmp(arg, "1.01") == 0
2087                 ) {
2088                 st->minor_version = 1;
2089                 return st;
2090         }
2091         if (strcmp(arg, "1.2") == 0 ||
2092 #ifndef DEFAULT_OLD_METADATA /* ifdef in super0.c */
2093             strcmp(arg, "default") == 0 ||
2094 #endif /* DEFAULT_OLD_METADATA */
2095             strcmp(arg, "1.02") == 0) {
2096                 st->minor_version = 2;
2097                 return st;
2098         }
2099         if (strcmp(arg, "1") == 0 ||
2100             strcmp(arg, "default") == 0) {
2101                 st->minor_version = -1;
2102                 return st;
2103         }
2104
2105         free(st);
2106         return NULL;
2107 }
2108
2109 /* find available size on device with this devsize, using
2110  * superblock type st, and reserving 'reserve' sectors for
2111  * a possible bitmap
2112  */
2113 static __u64 avail_size1(struct supertype *st, __u64 devsize,
2114                          unsigned long long data_offset)
2115 {
2116         struct mdp_superblock_1 *super = st->sb;
2117         int bmspace = 0;
2118         int bbspace = 0;
2119         if (devsize < 24)
2120                 return 0;
2121
2122 #ifndef MDASSEMBLE
2123         if (__le32_to_cpu(super->feature_map)&MD_FEATURE_BITMAP_OFFSET) {
2124                 /* hot-add. allow for actual size of bitmap */
2125                 struct bitmap_super_s *bsb;
2126                 bsb = (struct bitmap_super_s *)(((char*)super)+MAX_SB_SIZE);
2127                 bmspace = calc_bitmap_size(bsb, 4096) >> 9;
2128         }
2129 #endif
2130         /* Allow space for bad block log */
2131         if (super->bblog_size)
2132                 bbspace = __le16_to_cpu(super->bblog_size);
2133
2134         if (st->minor_version < 0)
2135                 /* not specified, so time to set default */
2136                 st->minor_version = 2;
2137
2138         if (data_offset == INVALID_SECTORS)
2139                 data_offset = st->data_offset;
2140
2141         if (data_offset != INVALID_SECTORS)
2142                 switch(st->minor_version) {
2143                 case 0:
2144                         return devsize - data_offset - 8*2 - bbspace;
2145                 case 1:
2146                 case 2:
2147                         return devsize - data_offset;
2148                 default:
2149                         return 0;
2150                 }
2151
2152         devsize -= bmspace;
2153
2154         switch(st->minor_version) {
2155         case 0:
2156                 /* at end */
2157                 return ((devsize - 8*2 - bbspace ) & ~(4*2-1));
2158         case 1:
2159                 /* at start, 4K for superblock and possible bitmap */
2160                 return devsize - 4*2 - bbspace;
2161         case 2:
2162                 /* 4k from start, 4K for superblock and possible bitmap */
2163                 return devsize - (4+4)*2 - bbspace;
2164         }
2165         return 0;
2166 }
2167
2168 static int
2169 add_internal_bitmap1(struct supertype *st,
2170                      int *chunkp, int delay, int write_behind,
2171                      unsigned long long size,
2172                      int may_change, int major)
2173 {
2174         /*
2175          * If not may_change, then this is a 'Grow' without sysfs support for
2176          * bitmaps, and the bitmap must fit after the superblock at 1K offset.
2177          * If may_change, then this is create or a Grow with sysfs syupport,
2178          * and we can put the bitmap wherever we like.
2179          *
2180          * size is in sectors,  chunk is in bytes !!!
2181          */
2182
2183         unsigned long long bits;
2184         unsigned long long max_bits;
2185         unsigned long long min_chunk;
2186         long offset;
2187         long bbl_offset, bbl_size;
2188         unsigned long long chunk = *chunkp;
2189         int room = 0;
2190         int creating = 0;
2191         int len;
2192         struct mdp_superblock_1 *sb = st->sb;
2193         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb) + MAX_SB_SIZE);
2194         int uuid[4];
2195
2196         if (__le64_to_cpu(sb->data_size) == 0)
2197                 /* Must be creating the array, else data_size would be non-zero */
2198                 creating = 1;
2199         switch(st->minor_version) {
2200         case 0:
2201                 /* either 3K after the superblock (when hot-add),
2202                  * or some amount of space before.
2203                  */
2204                 if (creating) {
2205                         /* We are creating array, so we *know* how much room has
2206                          * been left.
2207                          */
2208                         offset = 0;
2209                         bbl_size = 8;
2210                         room = choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2211                 } else {
2212                         room = __le64_to_cpu(sb->super_offset)
2213                                 - __le64_to_cpu(sb->data_offset)
2214                                 - __le64_to_cpu(sb->data_size);
2215                         bbl_size = __le16_to_cpu(sb->bblog_size);
2216                         if (bbl_size < 8)
2217                                 bbl_size = 8;
2218                         bbl_offset = (__s32)__le32_to_cpu(sb->bblog_offset);
2219                         if (bbl_size < -bbl_offset)
2220                                 bbl_size = -bbl_offset;
2221
2222                         if (!may_change || (room < 3*2 &&
2223                                             __le32_to_cpu(sb->max_dev) <= 384)) {
2224                                 room = 3*2;
2225                                 offset = 1*2;
2226                                 bbl_size = 0;
2227                         } else {
2228                                 offset = 0; /* means movable offset */
2229                         }
2230                 }
2231                 break;
2232         case 1:
2233         case 2: /* between superblock and data */
2234                 if (creating) {
2235                         offset = 4*2;
2236                         bbl_size = 8;
2237                         room = choose_bm_space(__le64_to_cpu(sb->size)) + bbl_size;
2238                 } else {
2239                         room = __le64_to_cpu(sb->data_offset)
2240                                 - __le64_to_cpu(sb->super_offset);
2241                         bbl_size = __le16_to_cpu(sb->bblog_size);
2242                         if (bbl_size)
2243                                 room = __le32_to_cpu(sb->bblog_offset) + bbl_size;
2244                         else
2245                                 bbl_size = 8;
2246
2247                         if (!may_change) {
2248                                 room -= 2; /* Leave 1K for superblock */
2249                                 offset = 2;
2250                                 bbl_size = 0;
2251                         } else {
2252                                 room -= 4*2; /* leave 4K for superblock */
2253                                 offset = 4*2;
2254                         }
2255                 }
2256                 break;
2257         default:
2258                 return -ENOSPC;
2259         }
2260
2261         room -= bbl_size;
2262         if (chunk == UnSet && room > 128*2)
2263                 /* Limit to 128K of bitmap when chunk size not requested */
2264                 room = 128*2;
2265
2266         if (room <= 1)
2267                 /* No room for a bitmap */
2268                 return -ENOSPC;
2269
2270         max_bits = (room * 512 - sizeof(bitmap_super_t)) * 8;
2271
2272         min_chunk = 4096; /* sub-page chunks don't work yet.. */
2273         bits = (size*512)/min_chunk +1;
2274         while (bits > max_bits) {
2275                 min_chunk *= 2;
2276                 bits = (bits+1)/2;
2277         }
2278         if (chunk == UnSet) {
2279                 /* For practical purpose, 64Meg is a good
2280                  * default chunk size for internal bitmaps.
2281                  */
2282                 chunk = min_chunk;
2283                 if (chunk < 64*1024*1024)
2284                         chunk = 64*1024*1024;
2285         } else if (chunk < min_chunk)
2286                 return -EINVAL; /* chunk size too small */
2287         if (chunk == 0) /* rounding problem */
2288                 return -EINVAL;
2289
2290         if (offset == 0) {
2291                 /* start bitmap on a 4K boundary with enough space for
2292                  * the bitmap
2293                  */
2294                 bits = (size*512) / chunk + 1;
2295                 room = ((bits+7)/8 + sizeof(bitmap_super_t) +4095)/4096;
2296                 room *= 8; /* convert 4K blocks to sectors */
2297                 offset = -room - bbl_size;
2298         }
2299
2300         sb->bitmap_offset = (int32_t)__cpu_to_le32(offset);
2301
2302         sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map)
2303                                         | MD_FEATURE_BITMAP_OFFSET);
2304         memset(bms, 0, sizeof(*bms));
2305         bms->magic = __cpu_to_le32(BITMAP_MAGIC);
2306         bms->version = __cpu_to_le32(major);
2307         uuid_from_super1(st, uuid);
2308         memcpy(bms->uuid, uuid, 16);
2309         bms->chunksize = __cpu_to_le32(chunk);
2310         bms->daemon_sleep = __cpu_to_le32(delay);
2311         bms->sync_size = __cpu_to_le64(size);
2312         bms->write_behind = __cpu_to_le32(write_behind);
2313         bms->nodes = __cpu_to_le32(st->nodes);
2314         if (st->nodes)
2315                 sb->feature_map = __cpu_to_le32(__le32_to_cpu(sb->feature_map)
2316                                                 | MD_FEATURE_BITMAP_VERSIONED);
2317         if (st->cluster_name) {
2318                 len = sizeof(bms->cluster_name);
2319                 strncpy((char *)bms->cluster_name, st->cluster_name, len);
2320                 bms->cluster_name[len - 1] = '\0';
2321         }
2322
2323         *chunkp = chunk;
2324         return 0;
2325 }
2326
2327 static int locate_bitmap1(struct supertype *st, int fd, int node_num)
2328 {
2329         unsigned long long offset;
2330         struct mdp_superblock_1 *sb;
2331         int mustfree = 0;
2332         int ret;
2333
2334         if (!st->sb) {
2335                 if (st->ss->load_super(st, fd, NULL))
2336                         return -1; /* no error I hope... */
2337                 mustfree = 1;
2338         }
2339         sb = st->sb;
2340
2341         if ((__le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET))
2342                 ret = 0;
2343         else
2344                 ret = -1;
2345         offset = __le64_to_cpu(sb->super_offset);
2346         offset += (int32_t) __le32_to_cpu(sb->bitmap_offset) * (node_num + 1);
2347         if (mustfree)
2348                 free(sb);
2349         lseek64(fd, offset<<9, 0);
2350         return ret;
2351 }
2352
2353 static int write_bitmap1(struct supertype *st, int fd, enum bitmap_update update)
2354 {
2355         struct mdp_superblock_1 *sb = st->sb;
2356         bitmap_super_t *bms = (bitmap_super_t*)(((char*)sb)+MAX_SB_SIZE);
2357         int rv = 0;
2358         void *buf;
2359         int towrite, n, len;
2360         struct align_fd afd;
2361         unsigned int i = 0;
2362         unsigned long long total_bm_space, bm_space_per_node;
2363
2364         switch (update) {
2365         case NameUpdate:
2366                 /* update cluster name */
2367                 if (st->cluster_name) {
2368                         len = sizeof(bms->cluster_name);
2369                         memset((char *)bms->cluster_name, 0, len);
2370                         strncpy((char *)bms->cluster_name,
2371                                 st->cluster_name, len);
2372                         bms->cluster_name[len - 1] = '\0';
2373                 }
2374                 break;
2375         case NodeNumUpdate:
2376                 /* cluster md only supports superblock 1.2 now */
2377                 if (st->minor_version != 2 && bms->version == BITMAP_MAJOR_CLUSTERED) {
2378                         pr_err("Warning: cluster md only works with superblock 1.2\n");
2379                         return -EINVAL;
2380                 }
2381
2382                 if (bms->version == BITMAP_MAJOR_CLUSTERED) {
2383                         if (st->nodes == 1) {
2384                                 /* the parameter for nodes is not valid */
2385                                 pr_err("Warning: cluster-md at least needs two nodes\n");
2386                                 return -EINVAL;
2387                         } else if (st->nodes == 0)
2388                                 /* --nodes is not specified */
2389                                 break;
2390                         else if (__cpu_to_le32(st->nodes) < bms->nodes) {
2391                                 /* Since the nodes num is not increased, no need to check the space
2392                                  * is enough or not, just update bms->nodes */
2393                                 bms->nodes = __cpu_to_le32(st->nodes);
2394                                 break;
2395                         }
2396                 } else {
2397                         /* no need to change bms->nodes for other bitmap types */
2398                         if (st->nodes)
2399                                 pr_err("Warning: --nodes option is only suitable for clustered bitmap\n");
2400                         break;
2401                 }
2402
2403                 /* Each node has an independent bitmap, it is necessary to calculate the
2404                  * space is enough or not, first get how many bytes for the total bitmap */
2405                 bm_space_per_node = calc_bitmap_size(bms, 4096);
2406
2407                 total_bm_space = 512 * (__le64_to_cpu(sb->data_offset) - __le64_to_cpu(sb->super_offset));
2408                 total_bm_space = total_bm_space - 4096; /* leave another 4k for superblock */
2409
2410                 if (bm_space_per_node * st->nodes > total_bm_space) {
2411                         pr_err("Warning: The max num of nodes can't exceed %llu\n",
2412                                 total_bm_space / bm_space_per_node);
2413                         return -ENOMEM;
2414                 }
2415
2416                 bms->nodes = __cpu_to_le32(st->nodes);
2417                 break;
2418         case NoUpdate:
2419         default:
2420                 break;
2421         }
2422
2423         init_afd(&afd, fd);
2424
2425         locate_bitmap1(st, fd, 0);
2426
2427         if (posix_memalign(&buf, 4096, 4096))
2428                 return -ENOMEM;
2429
2430         do {
2431                 /* Only the bitmap[0] should resync
2432                  * whole device on initial assembly
2433                  */
2434                 if (i)
2435                         memset(buf, 0x00, 4096);
2436                 else
2437                         memset(buf, 0xff, 4096);
2438                 memcpy(buf, (char *)bms, sizeof(bitmap_super_t));
2439
2440                 /*
2441                  * use 4096 boundary if bitmap_offset is aligned
2442                  * with 8 sectors, then it should compatible with
2443                  * older mdadm.
2444                  */
2445                 if (__le32_to_cpu(sb->bitmap_offset) & 7)
2446                         towrite = calc_bitmap_size(bms, 512);
2447                 else
2448                         towrite = calc_bitmap_size(bms, 4096);
2449                 while (towrite > 0) {
2450                         n = towrite;
2451                         if (n > 4096)
2452                                 n = 4096;
2453                         n = awrite(&afd, buf, n);
2454                         if (n > 0)
2455                                 towrite -= n;
2456                         else
2457                                 break;
2458                         if (i)
2459                                 memset(buf, 0x00, 4096);
2460                         else
2461                                 memset(buf, 0xff, 4096);
2462                 }
2463                 fsync(fd);
2464                 if (towrite) {
2465                         rv = -2;
2466                         break;
2467                 }
2468         } while (++i < __le32_to_cpu(bms->nodes));
2469
2470         free(buf);
2471         return rv;
2472 }
2473
2474 static void free_super1(struct supertype *st)
2475 {
2476
2477         if (st->sb)
2478                 free(st->sb);
2479         while (st->info) {
2480                 struct devinfo *di = st->info;
2481                 st->info = di->next;
2482                 if (di->fd >= 0)
2483                         close(di->fd);
2484                 free(di);
2485         }
2486         st->sb = NULL;
2487 }
2488
2489 #ifndef MDASSEMBLE
2490 static int validate_geometry1(struct supertype *st, int level,
2491                               int layout, int raiddisks,
2492                               int *chunk, unsigned long long size,
2493                               unsigned long long data_offset,
2494                               char *subdev, unsigned long long *freesize,
2495                               int verbose)
2496 {
2497         unsigned long long ldsize, devsize;
2498         int bmspace;
2499         unsigned long long headroom;
2500         int fd;
2501
2502         if (level == LEVEL_CONTAINER) {
2503                 if (verbose)
2504                         pr_err("1.x metadata does not support containers\n");
2505                 return 0;
2506         }
2507         if (*chunk == UnSet)
2508                 *chunk = DEFAULT_CHUNK;
2509
2510         if (!subdev)
2511                 return 1;
2512
2513         if (st->minor_version < 0)
2514                 /* not specified, so time to set default */
2515                 st->minor_version = 2;
2516
2517         fd = open(subdev, O_RDONLY|O_EXCL, 0);
2518         if (fd < 0) {
2519                 if (verbose)
2520                         pr_err("super1.x cannot open %s: %s\n",
2521                                 subdev, strerror(errno));
2522                 return 0;
2523         }
2524
2525         if (!get_dev_size(fd, subdev, &ldsize)) {
2526                 close(fd);
2527                 return 0;
2528         }
2529         close(fd);
2530
2531         devsize = ldsize >> 9;
2532         if (devsize < 24) {
2533                 *freesize = 0;
2534                 return 0;
2535         }
2536
2537         /* creating:  allow suitable space for bitmap */
2538         bmspace = choose_bm_space(devsize);
2539
2540         if (data_offset == INVALID_SECTORS)
2541                 data_offset = st->data_offset;
2542         if (data_offset == INVALID_SECTORS)
2543                 switch (st->minor_version) {
2544                 case 0:
2545                         data_offset = 0;
2546                         break;
2547                 case 1:
2548                 case 2:
2549                         /* Choose data offset appropriate for this device
2550                          * and use as default for whole array.
2551                          * The data_offset must allow for bitmap space
2552                          * and base metadata, should allow for some headroom
2553                          * for reshape, and should be rounded to multiple
2554                          * of 1M.
2555                          * Headroom is limited to 128M, but aim for about 0.1%
2556                          */
2557                         headroom = 128*1024*2;
2558                         while ((headroom << 10) > devsize &&
2559                                (*chunk == 0 ||
2560                                 headroom / 2 >= ((unsigned)(*chunk)*2)*2))
2561                                 headroom >>= 1;
2562                         data_offset = 12*2 + bmspace + headroom;
2563                         #define ONE_MEG (2*1024)
2564                         if (data_offset > ONE_MEG)
2565                                 data_offset = (data_offset / ONE_MEG) * ONE_MEG;
2566                         break;
2567                 }
2568         if (st->data_offset == INVALID_SECTORS)
2569                 st->data_offset = data_offset;
2570         switch(st->minor_version) {
2571         case 0: /* metadata at end.  Round down and subtract space to reserve */
2572                 devsize = (devsize & ~(4ULL*2-1));
2573                 /* space for metadata, bblog, bitmap */
2574                 devsize -= 8*2 + 8 + bmspace;
2575                 break;
2576         case 1:
2577         case 2:
2578                 devsize -= data_offset;
2579                 break;
2580         }
2581         *freesize = devsize;
2582         return 1;
2583 }
2584 #endif /* MDASSEMBLE */
2585
2586 void *super1_make_v0(struct supertype *st, struct mdinfo *info, mdp_super_t *sb0)
2587 {
2588         /* Create a v1.0 superblock based on 'info'*/
2589         void *ret;
2590         struct mdp_superblock_1 *sb;
2591         int i;
2592         unsigned long long offset;
2593
2594         if (posix_memalign(&ret, 4096, 1024) != 0)
2595                 return NULL;
2596         sb = ret;
2597         memset(ret, 0, 1024);
2598         sb->magic = __cpu_to_le32(MD_SB_MAGIC);
2599         sb->major_version = __cpu_to_le32(1);
2600
2601         copy_uuid(sb->set_uuid, info->uuid, super1.swapuuid);
2602         sprintf(sb->set_name, "%d", sb0->md_minor);
2603         sb->ctime = __cpu_to_le32(info->array.ctime+1);
2604         sb->level = __cpu_to_le32(info->array.level);
2605         sb->layout = __cpu_to_le32(info->array.layout);
2606         sb->size = __cpu_to_le64(info->component_size);
2607         sb->chunksize = __cpu_to_le32(info->array.chunk_size/512);
2608         sb->raid_disks = __cpu_to_le32(info->array.raid_disks);
2609         if (info->array.level > 0)
2610                 sb->data_size = sb->size;
2611         else
2612                 sb->data_size = st->ss->avail_size(st, st->devsize/512, 0);
2613         sb->resync_offset = MaxSector;
2614         sb->max_dev = __cpu_to_le32(MD_SB_DISKS);
2615         sb->dev_number = __cpu_to_le32(info->disk.number);
2616         sb->utime = __cpu_to_le64(info->array.utime);
2617
2618         offset = st->devsize/512 - 8*2;
2619         offset &= ~(4*2-1);
2620         sb->super_offset = __cpu_to_le64(offset);
2621         //*(__u64*)(st->other + 128 + 8 + 8) = __cpu_to_le64(offset);
2622
2623         random_uuid(sb->device_uuid);
2624
2625         for (i = 0; i < MD_SB_DISKS; i++) {
2626                 int state = sb0->disks[i].state;
2627                 sb->dev_roles[i] = MD_DISK_ROLE_SPARE;
2628                 if ((state & (1<<MD_DISK_SYNC)) &&
2629                     !(state & (1<<MD_DISK_FAULTY)))
2630                         sb->dev_roles[i] = __cpu_to_le16(sb0->disks[i].raid_disk);
2631         }
2632         sb->sb_csum = calc_sb_1_csum(sb);
2633         return ret;
2634 }
2635
2636 struct superswitch super1 = {
2637 #ifndef MDASSEMBLE
2638         .examine_super = examine_super1,
2639         .brief_examine_super = brief_examine_super1,
2640         .export_examine_super = export_examine_super1,
2641         .detail_super = detail_super1,
2642         .brief_detail_super = brief_detail_super1,
2643         .export_detail_super = export_detail_super1,
2644         .write_init_super = write_init_super1,
2645         .validate_geometry = validate_geometry1,
2646         .add_to_super = add_to_super1,
2647         .examine_badblocks = examine_badblocks_super1,
2648         .copy_metadata = copy_metadata1,
2649 #endif
2650         .match_home = match_home1,
2651         .uuid_from_super = uuid_from_super1,
2652         .getinfo_super = getinfo_super1,
2653         .container_content = container_content1,
2654         .update_super = update_super1,
2655         .init_super = init_super1,
2656         .store_super = store_super1,
2657         .compare_super = compare_super1,
2658         .load_super = load_super1,
2659         .match_metadata_desc = match_metadata_desc1,
2660         .avail_size = avail_size1,
2661         .add_internal_bitmap = add_internal_bitmap1,
2662         .locate_bitmap = locate_bitmap1,
2663         .write_bitmap = write_bitmap1,
2664         .free_super = free_super1,
2665 #if __BYTE_ORDER == BIG_ENDIAN
2666         .swapuuid = 0,
2667 #else
2668         .swapuuid = 1,
2669 #endif
2670         .name = "1.x",
2671 };