Create.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@cse.unsw.edu.au>
  23  *    Paper: Neil Brown
  24  *           School of Computer Science and Engineering
  25  *           The University of New South Wales
  26  *           Sydney, 2052
  27  *           Australia
  28  */
  29
  30 #include "mdadm.h"
  31 #include        "md_u.h"
  32 #include        "md_p.h"
  33
  34 int Create(struct supertype *st, char *mddev, int mdfd,
  35            int chunk, int level, int layout, unsigned long long size, int raiddisks, int sparedisks,
  36            char *name,
  37            int subdevs, mddev_dev_t devlist,
  38            int runstop, int verbose, int force, int assume_clean,
  39            char *bitmap_file, int bitmap_chunk, int write_behind, int delay)
  40 {
  41         /*
  42          * Create a new raid array.
  43          *
  44          * First check that necessary details are available
  45          * (i.e. level, raid-disks)
  46          *
  47          * Then check each disk to see what might be on it
  48          * and report anything interesting.
  49          *
  50          * If anything looks odd, and runstop not set,
  51          * abort.
  52          *
  53          * SET_ARRAY_INFO and ADD_NEW_DISK, and
  54          * if runstop==run, or raiddisks disks were used,
  55          * RUN_ARRAY
  56          */
  57         unsigned long long minsize=0, maxsize=0;
  58         char *mindisc = NULL;
  59         char *maxdisc = NULL;
  60         int dnum;
  61         mddev_dev_t dv;
  62         int fail=0, warn=0;
  63         struct stat stb;
  64         int first_missing = subdevs * 2;
  65         int missing_disks = 0;
  66         int insert_point = subdevs * 2; /* where to insert a missing drive */
  67         void *super;
  68         int pass;
  69         int vers;
  70         int rv;
  71         int bitmap_fd;
  72         unsigned long long bitmapsize;
  73
  74         mdu_array_info_t array;
  75         int major = BITMAP_MAJOR_HI;
  76
  77         memset(&array, 0, sizeof(array));
  78
  79         vers = md_get_version(mdfd);
  80         if (vers < 9000) {
  81                 fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n");
  82                 return 1;
  83         }
  84         if (level == UnSet) {
  85                 fprintf(stderr,
  86                         Name ": a RAID level is needed to create an array.\n");
  87                 return 1;
  88         }
  89         if (raiddisks < 1) {
  90                 fprintf(stderr,
  91                         Name ": a number of --raid-devices must be given to create an array\n");
  92                 return 1;
  93         }
  94         if (raiddisks < 4 && level == 6) {
  95                 fprintf(stderr,
  96                         Name ": at least 4 raid-devices needed for level 6\n");
  97                 return 1;
  98         }
  99         if (raiddisks > 256 && level == 6) {
 100                 fprintf(stderr,
 101                         Name ": no more than 256 raid-devices supported for level 6\n");
 102                 return 1;
 103         }
 104         if (raiddisks < 2 && level >= 4) {
 105                 fprintf(stderr,
 106                         Name ": at least 2 raid-devices needed for level 4 or 5\n");
 107                 return 1;
 108         }
 109         if (subdevs > raiddisks+sparedisks) {
 110                 fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
 111                 return 1;
 112         }
 113         if (subdevs < raiddisks+sparedisks) {
 114                 fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
 115                 return 1;
 116         }
 117
 118         /* now set some defaults */
 119         if (layout == UnSet)
 120                 switch(level) {
 121                 default: /* no layout */
 122                         layout = 0;
 123                         break;
 124                 case 10:
 125                         layout = 0x102; /* near=2, far=1 */
 126                         if (verbose > 0)
 127                                 fprintf(stderr,
 128                                         Name ": layout defaults to n1\n");
 129                         break;
 130                 case 5:
 131                 case 6:
 132                         layout = map_name(r5layout, "default");
 133                         if (verbose > 0)
 134                                 fprintf(stderr,
 135                                         Name ": layout defaults to %s\n", map_num(r5layout, layout));
 136                         break;
 137                 case LEVEL_FAULTY:
 138                         layout = map_name(faultylayout, "default");
 139
 140                         if (verbose > 0)
 141                                 fprintf(stderr,
 142                                         Name ": layout defaults to %s\n", map_num(faultylayout, layout));
 143                         break;
 144                 }
 145
 146         if (level == 10)
 147                 /* check layout fits in array*/
 148                 if ((layout&255) * ((layout>>8)&255) > raiddisks) {
 149                         fprintf(stderr, Name ": that layout requires at least %d devices\n",
 150                                 (layout&255) * ((layout>>8)&255));
 151                         return 1;
 152                 }
 153
 154         switch(level) {
 155         case 4:
 156         case 5:
 157         case 10:
 158         case 6:
 159         case 0:
 160         case -1: /* linear */
 161                 if (chunk == 0) {
 162                         chunk = 64;
 163                         if (verbose > 0)
 164                                 fprintf(stderr, Name ": chunk size defaults to 64K\n");
 165                 }
 166                 break;
 167         default: /* raid1, multipath */
 168                 if (chunk) {
 169                         chunk = 0;
 170                         if (verbose > 0)
 171                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
 172                 }
 173                 break;
 174         }
 175
 176         /* now look at the subdevs */
 177         array.active_disks = 0;
 178         array.working_disks = 0;
 179         dnum = 0;
 180         for (dv=devlist; dv; dv=dv->next, dnum++) {
 181                 char *dname = dv->devname;
 182                 unsigned long dsize;
 183                 unsigned long long ldsize, freesize;
 184                 int fd;
 185                 if (strcasecmp(dname, "missing")==0) {
 186                         if (first_missing > dnum)
 187                                 first_missing = dnum;
 188                         missing_disks ++;
 189                         continue;
 190                 }
 191                 array.working_disks++;
 192                 if (dnum < raiddisks)
 193                         array.active_disks++;
 194                 fd = open(dname, O_RDONLY|O_EXCL, 0);
 195                 if (fd <0 ) {
 196                         fprintf(stderr, Name ": Cannot open %s: %s\n",
 197                                 dname, strerror(errno));
 198                         fail=1;
 199                         continue;
 200                 }
 201 #ifdef BLKGETSIZE64
 202                 if (ioctl(fd, BLKGETSIZE64, &ldsize)==0)
 203                         ;
 204                 else
 205 #endif
 206                 if (ioctl(fd, BLKGETSIZE, &dsize)) {
 207                         fprintf(stderr, Name ": Cannot get size of %s: %s\n",
 208                                 dname, strerror(errno));
 209                         fail = 1;
 210                         close(fd);
 211                         continue;
 212                 }
 213                 else {
 214                         ldsize = dsize;
 215                         ldsize <<= 9;
 216                 }
 217                 freesize = st->ss->avail_size(st, ldsize >> 9);
 218                 if (freesize == 0) {
 219                         fprintf(stderr, Name ": %s is too small: %luK\n",
 220                                 dname, (unsigned long)(ldsize>>10));
 221                         fail = 1;
 222                         close(fd);
 223                         continue;
 224                 }
 225
 226                 freesize /= 2; /* convert to K */
 227                 if (chunk) {
 228                         /* round to chunk size */
 229                         freesize = freesize & ~(chunk-1);
 230                 }
 231
 232                 if (size && freesize < size) {
 233                         fprintf(stderr, Name ": %s is smaller that given size."
 234                                 " %lluK < %lluK + superblock\n", dname, freesize, size);
 235                         fail = 1;
 236                         close(fd);
 237                         continue;
 238                 }
 239                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
 240                         maxdisc = dname;
 241                         maxsize = freesize;
 242                 }
 243                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
 244                         mindisc = dname;
 245                         minsize = freesize;
 246                 }
 247                 if (runstop != 1 || verbose >= 0) {
 248                         warn |= check_ext2(fd, dname);
 249                         warn |= check_reiser(fd, dname);
 250                         warn |= check_raid(fd, dname);
 251                 }
 252                 close(fd);
 253         }
 254         if (fail) {
 255                 fprintf(stderr, Name ": create aborted\n");
 256                 return 1;
 257         }
 258         if (size == 0) {
 259                 if (mindisc == NULL) {
 260                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 261                         return 1;
 262                 }
 263                 if (level > 0 || level == LEVEL_MULTIPATH || level == LEVEL_FAULTY) {
 264                         /* size is meaningful */
 265                         if (minsize > 0x100000000ULL && st->ss->major == 0) {
 266                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 267                                 return 1;
 268                         }
 269                         size = minsize;
 270                         if (verbose > 0)
 271                                 fprintf(stderr, Name ": size set to %lluK\n", size);
 272                 }
 273         }
 274         if (level > 0 && ((maxsize-size)*100 > maxsize)) {
 275                 if (runstop != 1 || verbose >= 0)
 276                         fprintf(stderr, Name ": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
 277                                 maxdisc, size);
 278                 warn = 1;
 279         }
 280
 281         if (warn) {
 282                 if (runstop!= 1) {
 283                         if (!ask("Continue creating array? ")) {
 284                                 fprintf(stderr, Name ": create aborted.\n");
 285                                 return 1;
 286                         }
 287                 } else {
 288                         if (verbose > 0)
 289                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 290                 }
 291         }
 292
 293         /* If this is  raid5, we want to configure the last active slot
 294          * as missing, so that a reconstruct happens (faster than re-parity)
 295          * FIX: Can we do this for raid6 as well?
 296          */
 297         if (assume_clean==0 && force == 0 && first_missing >= raiddisks) {
 298                 switch ( level ) {
 299                 case 5:
 300                         insert_point = raiddisks-1;
 301                         sparedisks++;
 302                         array.active_disks--;
 303                         missing_disks++;
 304                         break;
 305                 default:
 306                         break;
 307                 }
 308         }
 309
 310         /* Ok, lets try some ioctls */
 311
 312         array.level = level;
 313         array.size = size;
 314         array.raid_disks = raiddisks;
 315         /* The kernel should *know* what md_minor we are dealing
 316          * with, but it chooses to trust me instead. Sigh
 317          */
 318         array.md_minor = 0;
 319         if (fstat(mdfd, &stb)==0)
 320                 array.md_minor = minor(stb.st_rdev);
 321         array.not_persistent = 0;
 322         /*** FIX: Need to do something about RAID-6 here ***/
 323         if ( ( (level == 5) &&
 324                (insert_point < raiddisks || first_missing < raiddisks) )
 325              ||
 326              ( level == 6 && missing_disks == 2)
 327              ||
 328              assume_clean
 329                 )
 330                 array.state = 1; /* clean, but one+ drive will be missing */
 331         else
 332                 array.state = 0; /* not clean, but no errors */
 333
 334         if (level == 10) {
 335                 /* for raid10, the bitmap size is the capacity of the array,
 336                  * which is array.size * raid_disks / ncopies;
 337                  * .. but convert to sectors.
 338                  */
 339                 int ncopies = (layout>>8) * (layout & 255);
 340                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
 341 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
 342         } else
 343                 bitmapsize = (unsigned long long)size * 2;
 344
 345         /* There is lots of redundancy in these disk counts,
 346          * raid_disks is the most meaningful value
 347          *          it describes the geometry of the array
 348          *          it is constant
 349          * nr_disks is total number of used slots.
 350          *          it should be raid_disks+spare_disks
 351          * spare_disks is the number of extra disks present
 352          *          see above
 353          * active_disks is the number of working disks in
 354          *          active slots. (With raid_disks)
 355          * working_disks is the total number of working disks,
 356          *          including spares
 357          * failed_disks is the number of disks marked failed
 358          *
 359          * Ideally, the kernel would keep these (except raid_disks)
 360          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
 361          * So for now, we assume that all raid and spare
 362          * devices will be given.
 363          */
 364         array.spare_disks=sparedisks;
 365         array.failed_disks=missing_disks;
 366         array.nr_disks = array.working_disks + array.failed_disks;
 367         array.layout = layout;
 368         array.chunk_size = chunk*1024;
 369         array.major_version = st->ss->major;
 370
 371         if (!st->ss->init_super(st, &super, &array, size, name))
 372                 return 1;
 373
 374         if (bitmap_file && vers < 9003) {
 375                 major = BITMAP_MAJOR_HOSTENDIAN;
 376 #ifdef __BIG_ENDIAN
 377                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
 378                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
 379 #endif
 380         }
 381
 382         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
 383                 if ((vers%100) < 2) {
 384                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
 385                         return 1;
 386                 }
 387                 if (!st->ss->add_internal_bitmap(st, super, bitmap_chunk, delay, write_behind,
 388                                                  bitmapsize, 1, major)) {
 389                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
 390                         return 1;
 391                 }
 392                 bitmap_file = NULL;
 393         }
 394
 395
 396
 397         if ((vers % 100) >= 1) { /* can use different versions */
 398                 mdu_array_info_t inf;
 399                 memset(&inf, 0, sizeof(inf));
 400                 inf.major_version = st->ss->major;
 401                 inf.minor_version = st->minor_version;
 402                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
 403         } else
 404                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
 405         if (rv) {
 406                 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
 407                         mddev, strerror(errno));
 408                 return 1;
 409         }
 410
 411         if (bitmap_file) {
 412                 int uuid[4];
 413
 414                 if (bitmap_chunk == UnSet)
 415                         bitmap_chunk = DEFAULT_BITMAP_CHUNK;
 416
 417                 st->ss->uuid_from_super(uuid, super);
 418                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
 419                                  delay, write_behind,
 420                                  bitmapsize,
 421                                  major)) {
 422                         return 1;
 423                 }
 424                 bitmap_fd = open(bitmap_file, O_RDWR);
 425                 if (bitmap_fd < 0) {
 426                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
 427                                 bitmap_file);
 428                         return 1;
 429                 }
 430                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
 431                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
 432                                 mddev, strerror(errno));
 433                         return 1;
 434                 }
 435         }
 436
 437
 438
 439         for (pass=1; pass <=2 ; pass++) {
 440                 mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
 441
 442                 for (dnum=0, dv = devlist ; dv ;
 443                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 444                         int fd;
 445                         struct stat stb;
 446                         mdu_disk_info_t disk;
 447
 448                         disk.number = dnum;
 449                         if (dnum == insert_point) {
 450                                 moved_disk = dv;
 451                         }
 452                         disk.raid_disk = disk.number;
 453                         if (disk.raid_disk < raiddisks)
 454                                 disk.state = (1<<MD_DISK_ACTIVE) |
 455                                                 (1<<MD_DISK_SYNC);
 456                         else
 457                                 disk.state = 0;
 458                         if (dv->writemostly)
 459                                 disk.state |= (1<<MD_DISK_WRITEMOSTLY);
 460
 461                         if (dnum == insert_point ||
 462                             strcasecmp(dv->devname, "missing")==0) {
 463                                 disk.major = 0;
 464                                 disk.minor = 0;
 465                                 disk.state = (1<<MD_DISK_FAULTY);
 466                         } else {
 467                                 fd = open(dv->devname, O_RDONLY|O_EXCL, 0);
 468                                 if (fd < 0) {
 469                                         fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
 470                                                 dv->devname);
 471                                         return 1;
 472                                 }
 473                                 fstat(fd, &stb);
 474                                 disk.major = major(stb.st_rdev);
 475                                 disk.minor = minor(stb.st_rdev);
 476                                 close(fd);
 477                         }
 478                         switch(pass){
 479                         case 1:
 480                                 st->ss->add_to_super(super, &disk);
 481                                 break;
 482                         case 2:
 483                                 if (disk.state == 1) break;
 484                                 st->ss->write_init_super(st, super, &disk, dv->devname);
 485
 486                                 if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
 487                                         fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
 488                                                 dv->devname, strerror(errno));
 489                                         free(super);
 490                                         return 1;
 491                                 }
 492
 493                                 break;
 494                         }
 495                         if (dv == moved_disk && dnum != insert_point) break;
 496                 }
 497         }
 498         free(super);
 499
 500         /* param is not actually used */
 501         if (runstop == 1 || subdevs >= raiddisks) {
 502                 mdu_param_t param;
 503                 if (ioctl(mdfd, RUN_ARRAY, &param)) {
 504                         fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
 505                                 strerror(errno));
 506                         Manage_runstop(mddev, mdfd, -1, 0);
 507                         return 1;
 508                 }
 509                 if (verbose >= 0)
 510                         fprintf(stderr, Name ": array %s started.\n", mddev);
 511         } else {
 512                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
 513         }
 514         return 0;
 515 }