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, int *uuid,
  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_num = 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         } else {
  85                 mdu_array_info_t inf;
  86                 memset(&inf, 0, sizeof(inf));
  87                 ioctl(mdfd, GET_ARRAY_INFO, &inf);
  88                 if (inf.working_disks != 0) {
  89                         fprintf(stderr, Name ": another array by this name"
  90                                 " is already running.\n");
  91                         return 1;
  92                 }
  93         }
  94         if (level == UnSet) {
  95                 fprintf(stderr,
  96                         Name ": a RAID level is needed to create an array.\n");
  97                 return 1;
  98         }
  99         if (raiddisks < 4 && level == 6) {
 100                 fprintf(stderr,
 101                         Name ": at least 4 raid-devices needed for level 6\n");
 102                 return 1;
 103         }
 104         if (raiddisks > 256 && level == 6) {
 105                 fprintf(stderr,
 106                         Name ": no more than 256 raid-devices supported for level 6\n");
 107                 return 1;
 108         }
 109         if (raiddisks < 2 && level >= 4) {
 110                 fprintf(stderr,
 111                         Name ": at least 2 raid-devices needed for level 4 or 5\n");
 112                 return 1;
 113         }
 114         if (level <= 0 && sparedisks) {
 115                 fprintf(stderr,
 116                         Name ": This level does not support spare devices\n");
 117                 return 1;
 118         }
 119         if (subdevs > raiddisks+sparedisks) {
 120                 fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
 121                 return 1;
 122         }
 123         if (subdevs < raiddisks+sparedisks) {
 124                 fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
 125                 return 1;
 126         }
 127         if (bitmap_file && level <= 0) {
 128                 fprintf(stderr, Name ": bitmaps not meaningful with level %s\n",
 129                         map_num(pers, level)?:"given");
 130                 return 1;
 131         }
 132
 133         /* now set some defaults */
 134         if (layout == UnSet)
 135                 switch(level) {
 136                 default: /* no layout */
 137                         layout = 0;
 138                         break;
 139                 case 10:
 140                         layout = 0x102; /* near=2, far=1 */
 141                         if (verbose > 0)
 142                                 fprintf(stderr,
 143                                         Name ": layout defaults to n1\n");
 144                         break;
 145                 case 5:
 146                 case 6:
 147                         layout = map_name(r5layout, "default");
 148                         if (verbose > 0)
 149                                 fprintf(stderr,
 150                                         Name ": layout defaults to %s\n", map_num(r5layout, layout));
 151                         break;
 152                 case LEVEL_FAULTY:
 153                         layout = map_name(faultylayout, "default");
 154
 155                         if (verbose > 0)
 156                                 fprintf(stderr,
 157                                         Name ": layout defaults to %s\n", map_num(faultylayout, layout));
 158                         break;
 159                 }
 160
 161         if (level == 10)
 162                 /* check layout fits in array*/
 163                 if ((layout&255) * ((layout>>8)&255) > raiddisks) {
 164                         fprintf(stderr, Name ": that layout requires at least %d devices\n",
 165                                 (layout&255) * ((layout>>8)&255));
 166                         return 1;
 167                 }
 168
 169         switch(level) {
 170         case 4:
 171         case 5:
 172         case 10:
 173         case 6:
 174         case 0:
 175         case LEVEL_LINEAR: /* linear */
 176                 if (chunk == 0) {
 177                         chunk = 64;
 178                         if (verbose > 0)
 179                                 fprintf(stderr, Name ": chunk size defaults to 64K\n");
 180                 }
 181                 break;
 182         case 1:
 183         case LEVEL_MULTIPATH:
 184                 if (chunk) {
 185                         chunk = 0;
 186                         if (verbose > 0)
 187                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
 188                 }
 189                 break;
 190         default:
 191                 fprintf(stderr, Name ": unknown level %d\n", level);
 192                 return 1;
 193         }
 194
 195         /* now look at the subdevs */
 196         array.active_disks = 0;
 197         array.working_disks = 0;
 198         dnum = 0;
 199         for (dv=devlist; dv; dv=dv->next, dnum++) {
 200                 char *dname = dv->devname;
 201                 unsigned long long ldsize, freesize;
 202                 int fd;
 203                 if (strcasecmp(dname, "missing")==0) {
 204                         if (first_missing > dnum)
 205                                 first_missing = dnum;
 206                         missing_disks ++;
 207                         continue;
 208                 }
 209                 array.working_disks++;
 210                 if (dnum < raiddisks)
 211                         array.active_disks++;
 212                 fd = open(dname, O_RDONLY|O_EXCL, 0);
 213                 if (fd <0 ) {
 214                         fprintf(stderr, Name ": Cannot open %s: %s\n",
 215                                 dname, strerror(errno));
 216                         fail=1;
 217                         continue;
 218                 }
 219                 if (!get_dev_size(fd, dname, &ldsize)) {
 220                         fail = 1;
 221                         close(fd);
 222                         continue;
 223                 }
 224                 if (st == NULL) {
 225                         struct createinfo *ci = conf_get_create_info();
 226                         if (ci)
 227                                 st = ci->supertype;
 228                 }
 229                 if (st == NULL) {
 230                         /* Need to choose a default metadata, which is different
 231                          * depending on the sizes of devices
 232                          */
 233                         int i;
 234                         char *name = "default";
 235                         if (level >= 1 && ldsize > (0x7fffffffULL<<10))
 236                                 name = "default/large";
 237                         for(i=0; !st && superlist[i]; i++)
 238                                 st = superlist[i]->match_metadata_desc(name);
 239
 240                         if (!st) {
 241                                 fprintf(stderr, Name ": internal error - no default metadata style\n");
 242                                 exit(2);
 243                         }
 244                         if (st->ss->major != 0 ||
 245                             st->minor_version != 90)
 246                                 fprintf(stderr, Name ": Defaulting to version"
 247                                         " %d.%d metadata\n",
 248                                         st->ss->major,
 249                                         st->minor_version);
 250                 }
 251                 freesize = st->ss->avail_size(st, ldsize >> 9);
 252                 if (freesize == 0) {
 253                         fprintf(stderr, Name ": %s is too small: %luK\n",
 254                                 dname, (unsigned long)(ldsize>>10));
 255                         fail = 1;
 256                         close(fd);
 257                         continue;
 258                 }
 259
 260                 freesize /= 2; /* convert to K */
 261                 if (chunk) {
 262                         /* round to chunk size */
 263                         freesize = freesize & ~(chunk-1);
 264                 }
 265
 266                 if (size && freesize < size) {
 267                         fprintf(stderr, Name ": %s is smaller that given size."
 268                                 " %lluK < %lluK + superblock\n", dname, freesize, size);
 269                         fail = 1;
 270                         close(fd);
 271                         continue;
 272                 }
 273                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
 274                         maxdisc = dname;
 275                         maxsize = freesize;
 276                 }
 277                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
 278                         mindisc = dname;
 279                         minsize = freesize;
 280                 }
 281                 if (runstop != 1 || verbose >= 0) {
 282                         warn |= check_ext2(fd, dname);
 283                         warn |= check_reiser(fd, dname);
 284                         warn |= check_raid(fd, dname);
 285                 }
 286                 close(fd);
 287         }
 288         if (fail) {
 289                 fprintf(stderr, Name ": create aborted\n");
 290                 return 1;
 291         }
 292         if (size == 0) {
 293                 if (mindisc == NULL) {
 294                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 295                         return 1;
 296                 }
 297                 if (level > 0 || level == LEVEL_MULTIPATH || level == LEVEL_FAULTY) {
 298                         /* size is meaningful */
 299                         if (minsize > 0x100000000ULL && st->ss->major == 0) {
 300                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 301                                 return 1;
 302                         }
 303                         size = minsize;
 304                         if (verbose > 0)
 305                                 fprintf(stderr, Name ": size set to %lluK\n", size);
 306                 }
 307         }
 308         if (level > 0 && ((maxsize-size)*100 > maxsize)) {
 309                 if (runstop != 1 || verbose >= 0)
 310                         fprintf(stderr, Name ": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
 311                                 maxdisc, size);
 312                 warn = 1;
 313         }
 314
 315         if (warn) {
 316                 if (runstop!= 1) {
 317                         if (!ask("Continue creating array? ")) {
 318                                 fprintf(stderr, Name ": create aborted.\n");
 319                                 return 1;
 320                         }
 321                 } else {
 322                         if (verbose > 0)
 323                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 324                 }
 325         }
 326
 327         /* If this is raid4/5, we want to configure the last active slot
 328          * as missing, so that a reconstruct happens (faster than re-parity)
 329          * FIX: Can we do this for raid6 as well?
 330          */
 331         if (assume_clean==0 && force == 0 && first_missing >= raiddisks) {
 332                 switch ( level ) {
 333                 case 4:
 334                 case 5:
 335                         insert_point = raiddisks-1;
 336                         sparedisks++;
 337                         array.active_disks--;
 338                         missing_disks++;
 339                         break;
 340                 default:
 341                         break;
 342                 }
 343         }
 344
 345         if (level <= 0 && first_missing != subdevs * 2) {
 346                 fprintf(stderr,
 347                         Name ": This level does not support missing devices\n");
 348                 return 1;
 349         }
 350
 351         /* Ok, lets try some ioctls */
 352
 353         array.level = level;
 354         array.size = size;
 355         array.raid_disks = raiddisks;
 356         /* The kernel should *know* what md_minor we are dealing
 357          * with, but it chooses to trust me instead. Sigh
 358          */
 359         array.md_minor = 0;
 360         if (fstat(mdfd, &stb)==0)
 361                 array.md_minor = minor(stb.st_rdev);
 362         array.not_persistent = 0;
 363         /*** FIX: Need to do something about RAID-6 here ***/
 364         if ( ( (level == 4 || level == 5) &&
 365                (insert_point < raiddisks || first_missing < raiddisks) )
 366              ||
 367              ( level == 6 && missing_disks == 2)
 368              ||
 369              assume_clean
 370                 )
 371                 array.state = 1; /* clean, but one+ drive will be missing */
 372         else
 373                 array.state = 0; /* not clean, but no errors */
 374
 375         if (level == 10) {
 376                 /* for raid10, the bitmap size is the capacity of the array,
 377                  * which is array.size * raid_disks / ncopies;
 378                  * .. but convert to sectors.
 379                  */
 380                 int ncopies = ((layout>>8) & 255) * (layout & 255);
 381                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
 382 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
 383         } else
 384                 bitmapsize = (unsigned long long)size * 2;
 385
 386         /* There is lots of redundancy in these disk counts,
 387          * raid_disks is the most meaningful value
 388          *          it describes the geometry of the array
 389          *          it is constant
 390          * nr_disks is total number of used slots.
 391          *          it should be raid_disks+spare_disks
 392          * spare_disks is the number of extra disks present
 393          *          see above
 394          * active_disks is the number of working disks in
 395          *          active slots. (With raid_disks)
 396          * working_disks is the total number of working disks,
 397          *          including spares
 398          * failed_disks is the number of disks marked failed
 399          *
 400          * Ideally, the kernel would keep these (except raid_disks)
 401          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
 402          * So for now, we assume that all raid and spare
 403          * devices will be given.
 404          */
 405         array.spare_disks=sparedisks;
 406         array.failed_disks=missing_disks;
 407         array.nr_disks = array.working_disks + array.failed_disks;
 408         array.layout = layout;
 409         array.chunk_size = chunk*1024;
 410         array.major_version = st->ss->major;
 411
 412         if (name == NULL || *name == 0) {
 413                 /* base name on mddev */
 414                 /*  /dev/md0 -> 0
 415                  *  /dev/md_d0 -> d0
 416                  *  /dev/md/1 -> 1
 417                  *  /dev/md/d1 -> d1
 418                  *  /dev/md/home -> home
 419                  *  /dev/mdhome -> home
 420                  */
 421                 name = strrchr(mddev, '/');
 422                 if (name) {
 423                         name++;
 424                         if (strncmp(name, "md_d", 4)==0 &&
 425                             strlen(name) > 4 &&
 426                             isdigit(name[4]) &&
 427                             (name-mddev) == 5 /* /dev/ */)
 428                                 name += 3;
 429                         else if (strncmp(name, "md", 2)==0 &&
 430                                  strlen(name) > 2 &&
 431                                  isdigit(name[2]) &&
 432                                  (name-mddev) == 5 /* /dev/ */)
 433                                 name += 2;
 434                 }
 435         }
 436         if (!st->ss->init_super(st, &super, &array, size, name, homehost, uuid))
 437                 return 1;
 438
 439         if (bitmap_file && vers < 9003) {
 440                 major_num = BITMAP_MAJOR_HOSTENDIAN;
 441 #ifdef __BIG_ENDIAN
 442                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
 443                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
 444 #endif
 445         }
 446
 447         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
 448                 if ((vers%100) < 2) {
 449                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
 450                         return 1;
 451                 }
 452                 if (!st->ss->add_internal_bitmap(st, super, &bitmap_chunk,
 453                                                  delay, write_behind,
 454                                                  bitmapsize, 1, major_num)) {
 455                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
 456                         return 1;
 457                 }
 458                 bitmap_file = NULL;
 459         }
 460
 461
 462
 463         if ((vers % 100) >= 1) { /* can use different versions */
 464                 mdu_array_info_t inf;
 465                 memset(&inf, 0, sizeof(inf));
 466                 inf.major_version = st->ss->major;
 467                 inf.minor_version = st->minor_version;
 468                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
 469         } else
 470                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
 471         if (rv) {
 472                 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
 473                         mddev, strerror(errno));
 474                 return 1;
 475         }
 476
 477         if (bitmap_file) {
 478                 int uuid[4];
 479
 480                 st->ss->uuid_from_super(uuid, super);
 481                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
 482                                  delay, write_behind,
 483                                  bitmapsize,
 484                                  major_num)) {
 485                         return 1;
 486                 }
 487                 bitmap_fd = open(bitmap_file, O_RDWR);
 488                 if (bitmap_fd < 0) {
 489                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
 490                                 bitmap_file);
 491                         return 1;
 492                 }
 493                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
 494                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
 495                                 mddev, strerror(errno));
 496                         return 1;
 497                 }
 498         }
 499
 500
 501
 502         for (pass=1; pass <=2 ; pass++) {
 503                 mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
 504
 505                 for (dnum=0, dv = devlist ; dv ;
 506                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 507                         int fd;
 508                         struct stat stb;
 509                         mdu_disk_info_t disk;
 510
 511                         disk.number = dnum;
 512                         if (dnum == insert_point) {
 513                                 moved_disk = dv;
 514                         }
 515                         disk.raid_disk = disk.number;
 516                         if (disk.raid_disk < raiddisks)
 517                                 disk.state = (1<<MD_DISK_ACTIVE) |
 518                                                 (1<<MD_DISK_SYNC);
 519                         else
 520                                 disk.state = 0;
 521                         if (dv->writemostly)
 522                                 disk.state |= (1<<MD_DISK_WRITEMOSTLY);
 523
 524                         if (dnum == insert_point ||
 525                             strcasecmp(dv->devname, "missing")==0) {
 526                                 disk.major = 0;
 527                                 disk.minor = 0;
 528                                 disk.state = (1<<MD_DISK_FAULTY);
 529                         } else {
 530                                 fd = open(dv->devname, O_RDONLY|O_EXCL, 0);
 531                                 if (fd < 0) {
 532                                         fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
 533                                                 dv->devname);
 534                                         return 1;
 535                                 }
 536                                 fstat(fd, &stb);
 537                                 disk.major = major(stb.st_rdev);
 538                                 disk.minor = minor(stb.st_rdev);
 539                                 remove_partitions(fd);
 540                                 close(fd);
 541                         }
 542                         switch(pass){
 543                         case 1:
 544                                 st->ss->add_to_super(super, &disk);
 545                                 break;
 546                         case 2:
 547                                 if (disk.state == 1) break;
 548                                 Kill(dv->devname, 0, 1); /* Just be sure it is clean */
 549                                 Kill(dv->devname, 0, 1); /* and again, there could be two superblocks */
 550                                 st->ss->write_init_super(st, super, &disk, dv->devname);
 551
 552                                 if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
 553                                         fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
 554                                                 dv->devname, strerror(errno));
 555                                         free(super);
 556                                         return 1;
 557                                 }
 558
 559                                 break;
 560                         }
 561                         if (dv == moved_disk && dnum != insert_point) break;
 562                 }
 563         }
 564         free(super);
 565
 566         /* param is not actually used */
 567         if (runstop == 1 || subdevs >= raiddisks) {
 568                 mdu_param_t param;
 569                 if (ioctl(mdfd, RUN_ARRAY, &param)) {
 570                         fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
 571                                 strerror(errno));
 572                         Manage_runstop(mddev, mdfd, -1, 0);
 573                         return 1;
 574                 }
 575                 if (verbose >= 0)
 576                         fprintf(stderr, Name ": array %s started.\n", mddev);
 577         } else {
 578                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
 579         }
 580         return 0;
 581 }