Create.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2006 Neil Brown <neilb@suse.de>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@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 #include        <ctype.h>
  34
  35 int Create(struct supertype *st, char *mddev, int mdfd,
  36            int chunk, int level, int layout, unsigned long long size, int raiddisks, int sparedisks,
  37            char *name, char *homehost,
  38            int subdevs, mddev_dev_t devlist,
  39            int runstop, int verbose, int force, int assume_clean,
  40            char *bitmap_file, int bitmap_chunk, int write_behind, int delay)
  41 {
  42         /*
  43          * Create a new raid array.
  44          *
  45          * First check that necessary details are available
  46          * (i.e. level, raid-disks)
  47          *
  48          * Then check each disk to see what might be on it
  49          * and report anything interesting.
  50          *
  51          * If anything looks odd, and runstop not set,
  52          * abort.
  53          *
  54          * SET_ARRAY_INFO and ADD_NEW_DISK, and
  55          * if runstop==run, or raiddisks disks were used,
  56          * RUN_ARRAY
  57          */
  58         unsigned long long minsize=0, maxsize=0;
  59         char *mindisc = NULL;
  60         char *maxdisc = NULL;
  61         int dnum;
  62         mddev_dev_t dv;
  63         int fail=0, warn=0;
  64         struct stat stb;
  65         int first_missing = subdevs * 2;
  66         int missing_disks = 0;
  67         int insert_point = subdevs * 2; /* where to insert a missing drive */
  68         void *super;
  69         int pass;
  70         int vers;
  71         int rv;
  72         int bitmap_fd;
  73         unsigned long long bitmapsize;
  74
  75         mdu_array_info_t array;
  76         int major = BITMAP_MAJOR_HI;
  77
  78         memset(&array, 0, sizeof(array));
  79
  80         vers = md_get_version(mdfd);
  81         if (vers < 9000) {
  82                 fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n");
  83                 return 1;
  84         }
  85         if (level == UnSet) {
  86                 fprintf(stderr,
  87                         Name ": a RAID level is needed to create an array.\n");
  88                 return 1;
  89         }
  90         if (raiddisks < 1) {
  91                 fprintf(stderr,
  92                         Name ": a number of --raid-devices must be given to create an array\n");
  93                 return 1;
  94         }
  95         if (raiddisks < 4 && level == 6) {
  96                 fprintf(stderr,
  97                         Name ": at least 4 raid-devices needed for level 6\n");
  98                 return 1;
  99         }
 100         if (raiddisks > 256 && level == 6) {
 101                 fprintf(stderr,
 102                         Name ": no more than 256 raid-devices supported for level 6\n");
 103                 return 1;
 104         }
 105         if (raiddisks < 2 && level >= 4) {
 106                 fprintf(stderr,
 107                         Name ": at least 2 raid-devices needed for level 4 or 5\n");
 108                 return 1;
 109         }
 110         if (subdevs > raiddisks+sparedisks) {
 111                 fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
 112                 return 1;
 113         }
 114         if (subdevs < raiddisks+sparedisks) {
 115                 fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
 116                 return 1;
 117         }
 118
 119         /* now set some defaults */
 120         if (layout == UnSet)
 121                 switch(level) {
 122                 default: /* no layout */
 123                         layout = 0;
 124                         break;
 125                 case 10:
 126                         layout = 0x102; /* near=2, far=1 */
 127                         if (verbose > 0)
 128                                 fprintf(stderr,
 129                                         Name ": layout defaults to n1\n");
 130                         break;
 131                 case 5:
 132                 case 6:
 133                         layout = map_name(r5layout, "default");
 134                         if (verbose > 0)
 135                                 fprintf(stderr,
 136                                         Name ": layout defaults to %s\n", map_num(r5layout, layout));
 137                         break;
 138                 case LEVEL_FAULTY:
 139                         layout = map_name(faultylayout, "default");
 140
 141                         if (verbose > 0)
 142                                 fprintf(stderr,
 143                                         Name ": layout defaults to %s\n", map_num(faultylayout, layout));
 144                         break;
 145                 }
 146
 147         if (level == 10)
 148                 /* check layout fits in array*/
 149                 if ((layout&255) * ((layout>>8)&255) > raiddisks) {
 150                         fprintf(stderr, Name ": that layout requires at least %d devices\n",
 151                                 (layout&255) * ((layout>>8)&255));
 152                         return 1;
 153                 }
 154
 155         switch(level) {
 156         case 4:
 157         case 5:
 158         case 10:
 159         case 6:
 160         case 0:
 161         case -1: /* linear */
 162                 if (chunk == 0) {
 163                         chunk = 64;
 164                         if (verbose > 0)
 165                                 fprintf(stderr, Name ": chunk size defaults to 64K\n");
 166                 }
 167                 break;
 168         default: /* raid1, multipath */
 169                 if (chunk) {
 170                         chunk = 0;
 171                         if (verbose > 0)
 172                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
 173                 }
 174                 break;
 175         }
 176
 177         /* now look at the subdevs */
 178         array.active_disks = 0;
 179         array.working_disks = 0;
 180         dnum = 0;
 181         for (dv=devlist; dv; dv=dv->next, dnum++) {
 182                 char *dname = dv->devname;
 183                 unsigned long dsize;
 184                 unsigned long long ldsize, freesize;
 185                 int fd;
 186                 if (strcasecmp(dname, "missing")==0) {
 187                         if (first_missing > dnum)
 188                                 first_missing = dnum;
 189                         missing_disks ++;
 190                         continue;
 191                 }
 192                 array.working_disks++;
 193                 if (dnum < raiddisks)
 194                         array.active_disks++;
 195                 fd = open(dname, O_RDONLY|O_EXCL, 0);
 196                 if (fd <0 ) {
 197                         fprintf(stderr, Name ": Cannot open %s: %s\n",
 198                                 dname, strerror(errno));
 199                         fail=1;
 200                         continue;
 201                 }
 202 #ifdef BLKGETSIZE64
 203                 if (ioctl(fd, BLKGETSIZE64, &ldsize)==0)
 204                         ;
 205                 else
 206 #endif
 207                 if (ioctl(fd, BLKGETSIZE, &dsize)) {
 208                         fprintf(stderr, Name ": Cannot get size of %s: %s\n",
 209                                 dname, strerror(errno));
 210                         fail = 1;
 211                         close(fd);
 212                         continue;
 213                 }
 214                 else {
 215                         ldsize = dsize;
 216                         ldsize <<= 9;
 217                 }
 218                 if (st == NULL) {
 219                         struct createinfo *ci = conf_get_create_info();
 220                         if (ci)
 221                                 st = ci->supertype;
 222                 }
 223                 if (st == NULL) {
 224                         /* Need to choose a default metadata, which is different
 225                          * depending on the sizes of devices
 226                          */
 227                         int i;
 228                         char *name = "default";
 229                         if (level >= 1 && ldsize > (0x7fffffffULL<<10))
 230                                 name = "default/large";
 231                         for(i=0; !st && superlist[i]; i++)
 232                                 st = superlist[i]->match_metadata_desc(name);
 233
 234                         if (!st) {
 235                                 fprintf(stderr, Name ": internal error - no default metadata style\n");
 236                                 exit(2);
 237                         }
 238                         if (st->ss->major != 0 ||
 239                             st->minor_version != 90)
 240                                 fprintf(stderr, Name ": Defaulting to verion %d.%d metadata\n",
 241                                         st->ss->major,
 242                                         st->minor_version);
 243                 }
 244                 freesize = st->ss->avail_size(st, ldsize >> 9);
 245                 if (freesize == 0) {
 246                         fprintf(stderr, Name ": %s is too small: %luK\n",
 247                                 dname, (unsigned long)(ldsize>>10));
 248                         fail = 1;
 249                         close(fd);
 250                         continue;
 251                 }
 252
 253                 freesize /= 2; /* convert to K */
 254                 if (chunk) {
 255                         /* round to chunk size */
 256                         freesize = freesize & ~(chunk-1);
 257                 }
 258
 259                 if (size && freesize < size) {
 260                         fprintf(stderr, Name ": %s is smaller that given size."
 261                                 " %lluK < %lluK + superblock\n", dname, freesize, size);
 262                         fail = 1;
 263                         close(fd);
 264                         continue;
 265                 }
 266                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
 267                         maxdisc = dname;
 268                         maxsize = freesize;
 269                 }
 270                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
 271                         mindisc = dname;
 272                         minsize = freesize;
 273                 }
 274                 if (runstop != 1 || verbose >= 0) {
 275                         warn |= check_ext2(fd, dname);
 276                         warn |= check_reiser(fd, dname);
 277                         warn |= check_raid(fd, dname);
 278                 }
 279                 close(fd);
 280         }
 281         if (fail) {
 282                 fprintf(stderr, Name ": create aborted\n");
 283                 return 1;
 284         }
 285         if (size == 0) {
 286                 if (mindisc == NULL) {
 287                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 288                         return 1;
 289                 }
 290                 if (level > 0 || level == LEVEL_MULTIPATH || level == LEVEL_FAULTY) {
 291                         /* size is meaningful */
 292                         if (minsize > 0x100000000ULL && st->ss->major == 0) {
 293                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 294                                 return 1;
 295                         }
 296                         size = minsize;
 297                         if (verbose > 0)
 298                                 fprintf(stderr, Name ": size set to %lluK\n", size);
 299                 }
 300         }
 301         if (level > 0 && ((maxsize-size)*100 > maxsize)) {
 302                 if (runstop != 1 || verbose >= 0)
 303                         fprintf(stderr, Name ": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
 304                                 maxdisc, size);
 305                 warn = 1;
 306         }
 307
 308         if (warn) {
 309                 if (runstop!= 1) {
 310                         if (!ask("Continue creating array? ")) {
 311                                 fprintf(stderr, Name ": create aborted.\n");
 312                                 return 1;
 313                         }
 314                 } else {
 315                         if (verbose > 0)
 316                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 317                 }
 318         }
 319
 320         /* If this is  raid5, we want to configure the last active slot
 321          * as missing, so that a reconstruct happens (faster than re-parity)
 322          * FIX: Can we do this for raid6 as well?
 323          */
 324         if (assume_clean==0 && force == 0 && first_missing >= raiddisks) {
 325                 switch ( level ) {
 326                 case 5:
 327                         insert_point = raiddisks-1;
 328                         sparedisks++;
 329                         array.active_disks--;
 330                         missing_disks++;
 331                         break;
 332                 default:
 333                         break;
 334                 }
 335         }
 336
 337         /* Ok, lets try some ioctls */
 338
 339         array.level = level;
 340         array.size = size;
 341         array.raid_disks = raiddisks;
 342         /* The kernel should *know* what md_minor we are dealing
 343          * with, but it chooses to trust me instead. Sigh
 344          */
 345         array.md_minor = 0;
 346         if (fstat(mdfd, &stb)==0)
 347                 array.md_minor = minor(stb.st_rdev);
 348         array.not_persistent = 0;
 349         /*** FIX: Need to do something about RAID-6 here ***/
 350         if ( ( (level == 5) &&
 351                (insert_point < raiddisks || first_missing < raiddisks) )
 352              ||
 353              ( level == 6 && missing_disks == 2)
 354              ||
 355              assume_clean
 356                 )
 357                 array.state = 1; /* clean, but one+ drive will be missing */
 358         else
 359                 array.state = 0; /* not clean, but no errors */
 360
 361         if (level == 10) {
 362                 /* for raid10, the bitmap size is the capacity of the array,
 363                  * which is array.size * raid_disks / ncopies;
 364                  * .. but convert to sectors.
 365                  */
 366                 int ncopies = (layout>>8) * (layout & 255);
 367                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
 368 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
 369         } else
 370                 bitmapsize = (unsigned long long)size * 2;
 371
 372         /* There is lots of redundancy in these disk counts,
 373          * raid_disks is the most meaningful value
 374          *          it describes the geometry of the array
 375          *          it is constant
 376          * nr_disks is total number of used slots.
 377          *          it should be raid_disks+spare_disks
 378          * spare_disks is the number of extra disks present
 379          *          see above
 380          * active_disks is the number of working disks in
 381          *          active slots. (With raid_disks)
 382          * working_disks is the total number of working disks,
 383          *          including spares
 384          * failed_disks is the number of disks marked failed
 385          *
 386          * Ideally, the kernel would keep these (except raid_disks)
 387          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
 388          * So for now, we assume that all raid and spare
 389          * devices will be given.
 390          */
 391         array.spare_disks=sparedisks;
 392         array.failed_disks=missing_disks;
 393         array.nr_disks = array.working_disks + array.failed_disks;
 394         array.layout = layout;
 395         array.chunk_size = chunk*1024;
 396         array.major_version = st->ss->major;
 397
 398         if (name == NULL || *name == 0) {
 399                 /* base name on mddev */
 400                 /*  /dev/md0 -> 0
 401                  *  /dev/md_d0 -> d0
 402                  *  /dev/md/1 -> 1
 403                  *  /dev/md/d1 -> d1
 404                  *  /dev/md/home -> home
 405                  *  /dev/mdhome -> home
 406                  */
 407                 name = strrchr(mddev, '/');
 408                 if (name) {
 409                         name++;
 410                         if (strncmp(name, "md_d", 4)==0 &&
 411                             strlen(name) > 4 &&
 412                             isdigit(name[4]) &&
 413                             (name-mddev) == 5 /* /dev/ */)
 414                                 name += 3;
 415                         else if (strncmp(name, "md", 2)==0 &&
 416                                  strlen(name) > 2 &&
 417                                  isdigit(name[2]) &&
 418                                  (name-mddev) == 5 /* /dev/ */)
 419                                 name += 2;
 420                 }
 421         }
 422         if (!st->ss->init_super(st, &super, &array, size, name, homehost))
 423                 return 1;
 424
 425         if (bitmap_file && vers < 9003) {
 426                 major = BITMAP_MAJOR_HOSTENDIAN;
 427 #ifdef __BIG_ENDIAN
 428                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
 429                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
 430 #endif
 431         }
 432
 433         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
 434                 if ((vers%100) < 2) {
 435                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
 436                         return 1;
 437                 }
 438                 if (!st->ss->add_internal_bitmap(st, super, bitmap_chunk, delay, write_behind,
 439                                                  bitmapsize, 1, major)) {
 440                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
 441                         return 1;
 442                 }
 443                 bitmap_file = NULL;
 444         }
 445
 446
 447
 448         if ((vers % 100) >= 1) { /* can use different versions */
 449                 mdu_array_info_t inf;
 450                 memset(&inf, 0, sizeof(inf));
 451                 inf.major_version = st->ss->major;
 452                 inf.minor_version = st->minor_version;
 453                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
 454         } else
 455                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
 456         if (rv) {
 457                 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
 458                         mddev, strerror(errno));
 459                 return 1;
 460         }
 461
 462         if (bitmap_file) {
 463                 int uuid[4];
 464
 465                 st->ss->uuid_from_super(uuid, super);
 466                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
 467                                  delay, write_behind,
 468                                  bitmapsize,
 469                                  major)) {
 470                         return 1;
 471                 }
 472                 bitmap_fd = open(bitmap_file, O_RDWR);
 473                 if (bitmap_fd < 0) {
 474                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
 475                                 bitmap_file);
 476                         return 1;
 477                 }
 478                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
 479                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
 480                                 mddev, strerror(errno));
 481                         return 1;
 482                 }
 483         }
 484
 485
 486
 487         for (pass=1; pass <=2 ; pass++) {
 488                 mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
 489
 490                 for (dnum=0, dv = devlist ; dv ;
 491                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 492                         int fd;
 493                         struct stat stb;
 494                         mdu_disk_info_t disk;
 495
 496                         disk.number = dnum;
 497                         if (dnum == insert_point) {
 498                                 moved_disk = dv;
 499                         }
 500                         disk.raid_disk = disk.number;
 501                         if (disk.raid_disk < raiddisks)
 502                                 disk.state = (1<<MD_DISK_ACTIVE) |
 503                                                 (1<<MD_DISK_SYNC);
 504                         else
 505                                 disk.state = 0;
 506                         if (dv->writemostly)
 507                                 disk.state |= (1<<MD_DISK_WRITEMOSTLY);
 508
 509                         if (dnum == insert_point ||
 510                             strcasecmp(dv->devname, "missing")==0) {
 511                                 disk.major = 0;
 512                                 disk.minor = 0;
 513                                 disk.state = (1<<MD_DISK_FAULTY);
 514                         } else {
 515                                 fd = open(dv->devname, O_RDONLY|O_EXCL, 0);
 516                                 if (fd < 0) {
 517                                         fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
 518                                                 dv->devname);
 519                                         return 1;
 520                                 }
 521                                 fstat(fd, &stb);
 522                                 disk.major = major(stb.st_rdev);
 523                                 disk.minor = minor(stb.st_rdev);
 524                                 remove_partitions(fd);
 525                                 close(fd);
 526                         }
 527                         switch(pass){
 528                         case 1:
 529                                 st->ss->add_to_super(super, &disk);
 530                                 break;
 531                         case 2:
 532                                 if (disk.state == 1) break;
 533                                 Kill(dv->devname, 0, 1); /* Just be sure it is clean */
 534                                 Kill(dv->devname, 0, 1); /* and again, there could be two superblocks */
 535                                 st->ss->write_init_super(st, super, &disk, dv->devname);
 536
 537                                 if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
 538                                         fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
 539                                                 dv->devname, strerror(errno));
 540                                         free(super);
 541                                         return 1;
 542                                 }
 543
 544                                 break;
 545                         }
 546                         if (dv == moved_disk && dnum != insert_point) break;
 547                 }
 548         }
 549         free(super);
 550
 551         /* param is not actually used */
 552         if (runstop == 1 || subdevs >= raiddisks) {
 553                 mdu_param_t param;
 554                 if (ioctl(mdfd, RUN_ARRAY, &param)) {
 555                         fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
 556                                 strerror(errno));
 557                         Manage_runstop(mddev, mdfd, -1, 0);
 558                         return 1;
 559                 }
 560                 if (verbose >= 0)
 561                         fprintf(stderr, Name ": array %s started.\n", mddev);
 562         } else {
 563                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
 564         }
 565         return 0;
 566 }