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