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 second_missing = subdevs * 2;
  67         int missing_disks = 0;
  68         int insert_point = subdevs * 2; /* where to insert a missing drive */
  69         int pass;
  70         int vers;
  71         int rv;
  72         int bitmap_fd;
  73         unsigned long long bitmapsize;
  74         struct mdinfo info;
  75
  76         int major_num = BITMAP_MAJOR_HI;
  77
  78         memset(&info, 0, sizeof(info));
  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_FAULTY:
 184         case LEVEL_MULTIPATH:
 185         case LEVEL_CONTAINER:
 186                 if (chunk) {
 187                         chunk = 0;
 188                         if (verbose > 0)
 189                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
 190                 }
 191                 break;
 192         default:
 193                 fprintf(stderr, Name ": unknown level %d\n", level);
 194                 return 1;
 195         }
 196
 197         if (st && ! st->ss->validate_geometry(st, level, layout, raiddisks,
 198                                               chunk, size, NULL, NULL))
 199                 return 1;
 200
 201         /* now look at the subdevs */
 202         info.array.active_disks = 0;
 203         info.array.working_disks = 0;
 204         dnum = 0;
 205         for (dv=devlist; dv; dv=dv->next, dnum++) {
 206                 char *dname = dv->devname;
 207                 unsigned long long freesize;
 208                 int fd;
 209                 if (strcasecmp(dname, "missing")==0) {
 210                         if (first_missing > dnum)
 211                                 first_missing = dnum;
 212                         if (second_missing > dnum && dnum > first_missing)
 213                                 second_missing = dnum;
 214                         missing_disks ++;
 215                         continue;
 216                 }
 217                 info.array.working_disks++;
 218                 if (dnum < raiddisks)
 219                         info.array.active_disks++;
 220                 fd = open(dname, O_RDONLY|O_EXCL, 0);
 221                 if (fd <0 ) {
 222                         fprintf(stderr, Name ": Cannot open %s: %s\n",
 223                                 dname, strerror(errno));
 224                         fail=1;
 225                         continue;
 226                 }
 227                 if (st == NULL) {
 228                         struct createinfo *ci = conf_get_create_info();
 229                         if (ci)
 230                                 st = ci->supertype;
 231                 }
 232                 if (st == NULL) {
 233                         /* Need to choose a default metadata, which is different
 234                          * depending on geometry of array.
 235                          */
 236                         int i;
 237                         char *name = "default";
 238                         for(i=0; !st && superlist[i]; i++) {
 239                                 st = superlist[i]->match_metadata_desc(name);
 240                                 if (st && !st->ss->validate_geometry
 241                                                 (st, level, layout, raiddisks,
 242                                                  chunk, size, dname, &freesize))
 243                                         st = NULL;
 244                         }
 245
 246                         if (!st) {
 247                                 fprintf(stderr, Name ": device %s not suitable "
 248                                         "for any style of array\n",
 249                                         dname);
 250                                 exit(2);
 251                         }
 252                         if (st->ss->major != 0 ||
 253                             st->minor_version != 90)
 254                                 fprintf(stderr, Name ": Defaulting to version"
 255                                         " %d.%d metadata\n",
 256                                         st->ss->major,
 257                                         st->minor_version);
 258                 } else {
 259                         if (!st->ss->validate_geometry(st, level, layout,
 260                                                        raiddisks,
 261                                                        chunk, size, dname,
 262                                                        &freesize)) {
 263
 264                                 fprintf(stderr,
 265                                         Name ": %s is not suitable for "
 266                                         "this array.\n",
 267                                         dname);
 268                                 fail = 1;
 269                                 continue;
 270                         }
 271                 }
 272
 273                 freesize /= 2; /* convert to K */
 274                 if (chunk) {
 275                         /* round to chunk size */
 276                         freesize = freesize & ~(chunk-1);
 277                 }
 278
 279                 if (size && freesize < size) {
 280                         fprintf(stderr, Name ": %s is smaller that given size."
 281                                 " %lluK < %lluK + metadata\n",
 282                                 dname, freesize, size);
 283                         fail = 1;
 284                         close(fd);
 285                         continue;
 286                 }
 287                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
 288                         maxdisc = dname;
 289                         maxsize = freesize;
 290                 }
 291                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
 292                         mindisc = dname;
 293                         minsize = freesize;
 294                 }
 295                 if (runstop != 1 || verbose >= 0) {
 296                         warn |= check_ext2(fd, dname);
 297                         warn |= check_reiser(fd, dname);
 298                         warn |= check_raid(fd, dname);
 299                 }
 300                 close(fd);
 301         }
 302         if (fail) {
 303                 fprintf(stderr, Name ": create aborted\n");
 304                 return 1;
 305         }
 306         if (size == 0) {
 307                 if (mindisc == NULL) {
 308                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 309                         return 1;
 310                 }
 311                 if (level > 0 || level == LEVEL_MULTIPATH || level == LEVEL_FAULTY) {
 312                         /* size is meaningful */
 313                         if (minsize > 0x100000000ULL && st->ss->major == 0) {
 314                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 315                                 return 1;
 316                         }
 317                         size = minsize;
 318                         if (verbose > 0)
 319                                 fprintf(stderr, Name ": size set to %lluK\n", size);
 320                 }
 321         }
 322         if (level > 0 && ((maxsize-size)*100 > maxsize)) {
 323                 if (runstop != 1 || verbose >= 0)
 324                         fprintf(stderr, Name ": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
 325                                 maxdisc, size);
 326                 warn = 1;
 327         }
 328
 329         if (warn) {
 330                 if (runstop!= 1) {
 331                         if (!ask("Continue creating array? ")) {
 332                                 fprintf(stderr, Name ": create aborted.\n");
 333                                 return 1;
 334                         }
 335                 } else {
 336                         if (verbose > 0)
 337                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 338                 }
 339         }
 340
 341         /* If this is raid4/5, we want to configure the last active slot
 342          * as missing, so that a reconstruct happens (faster than re-parity)
 343          * FIX: Can we do this for raid6 as well?
 344          */
 345         if (assume_clean==0 && force == 0 && first_missing >= raiddisks) {
 346                 switch ( level ) {
 347                 case 4:
 348                 case 5:
 349                         insert_point = raiddisks-1;
 350                         sparedisks++;
 351                         info.array.active_disks--;
 352                         missing_disks++;
 353                         break;
 354                 default:
 355                         break;
 356                 }
 357         }
 358         /* For raid6, if creating with 1 missing drive, make a good drive
 359          * into a spare, else the create will fail
 360          */
 361         if (assume_clean == 0 && force == 0 && first_missing < raiddisks &&
 362             second_missing >= raiddisks && level == 6) {
 363                 insert_point = raiddisks - 1;
 364                 if (insert_point == first_missing)
 365                         insert_point--;
 366                 sparedisks ++;
 367                 info.array.active_disks--;
 368                 missing_disks++;
 369         }
 370
 371         if (level <= 0 && first_missing != subdevs * 2) {
 372                 fprintf(stderr,
 373                         Name ": This level does not support missing devices\n");
 374                 return 1;
 375         }
 376
 377         /* Ok, lets try some ioctls */
 378
 379         info.array.level = level;
 380         info.array.size = size;
 381         info.array.raid_disks = raiddisks;
 382         /* The kernel should *know* what md_minor we are dealing
 383          * with, but it chooses to trust me instead. Sigh
 384          */
 385         info.array.md_minor = 0;
 386         if (fstat(mdfd, &stb)==0)
 387                 info.array.md_minor = minor(stb.st_rdev);
 388         info.array.not_persistent = 0;
 389
 390         if ( ( (level == 4 || level == 5) &&
 391                (insert_point < raiddisks || first_missing < raiddisks) )
 392              ||
 393              ( level == 6 && (insert_point < raiddisks
 394                               || second_missing < raiddisks))
 395              ||
 396              assume_clean
 397                 )
 398                 info.array.state = 1; /* clean, but one+ drive will be missing*/
 399         else
 400                 info.array.state = 0; /* not clean, but no errors */
 401
 402         if (level == 10) {
 403                 /* for raid10, the bitmap size is the capacity of the array,
 404                  * which is array.size * raid_disks / ncopies;
 405                  * .. but convert to sectors.
 406                  */
 407                 int ncopies = ((layout>>8) & 255) * (layout & 255);
 408                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
 409 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
 410         } else
 411                 bitmapsize = (unsigned long long)size * 2;
 412
 413         /* There is lots of redundancy in these disk counts,
 414          * raid_disks is the most meaningful value
 415          *          it describes the geometry of the array
 416          *          it is constant
 417          * nr_disks is total number of used slots.
 418          *          it should be raid_disks+spare_disks
 419          * spare_disks is the number of extra disks present
 420          *          see above
 421          * active_disks is the number of working disks in
 422          *          active slots. (With raid_disks)
 423          * working_disks is the total number of working disks,
 424          *          including spares
 425          * failed_disks is the number of disks marked failed
 426          *
 427          * Ideally, the kernel would keep these (except raid_disks)
 428          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
 429          * So for now, we assume that all raid and spare
 430          * devices will be given.
 431          */
 432         info.array.spare_disks=sparedisks;
 433         info.array.failed_disks=missing_disks;
 434         info.array.nr_disks = info.array.working_disks
 435                 + info.array.failed_disks;
 436         info.array.layout = layout;
 437         info.array.chunk_size = chunk*1024;
 438         info.array.major_version = st->ss->major;
 439
 440         if (name == NULL || *name == 0) {
 441                 /* base name on mddev */
 442                 /*  /dev/md0 -> 0
 443                  *  /dev/md_d0 -> d0
 444                  *  /dev/md/1 -> 1
 445                  *  /dev/md/d1 -> d1
 446                  *  /dev/md/home -> home
 447                  *  /dev/mdhome -> home
 448                  */
 449                 name = strrchr(mddev, '/');
 450                 if (name) {
 451                         name++;
 452                         if (strncmp(name, "md_d", 4)==0 &&
 453                             strlen(name) > 4 &&
 454                             isdigit(name[4]) &&
 455                             (name-mddev) == 5 /* /dev/ */)
 456                                 name += 3;
 457                         else if (strncmp(name, "md", 2)==0 &&
 458                                  strlen(name) > 2 &&
 459                                  isdigit(name[2]) &&
 460                                  (name-mddev) == 5 /* /dev/ */)
 461                                 name += 2;
 462                 }
 463         }
 464         if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid))
 465                 return 1;
 466
 467         if (bitmap_file && vers < 9003) {
 468                 major_num = BITMAP_MAJOR_HOSTENDIAN;
 469 #ifdef __BIG_ENDIAN
 470                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
 471                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
 472 #endif
 473         }
 474
 475         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
 476                 if ((vers%100) < 2) {
 477                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
 478                         return 1;
 479                 }
 480                 if (!st->ss->add_internal_bitmap(st, &bitmap_chunk,
 481                                                  delay, write_behind,
 482                                                  bitmapsize, 1, major_num)) {
 483                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
 484                         return 1;
 485                 }
 486                 bitmap_file = NULL;
 487         }
 488
 489
 490
 491         if ((vers % 100) >= 1) { /* can use different versions */
 492                 mdu_array_info_t inf;
 493                 memset(&inf, 0, sizeof(inf));
 494                 inf.major_version = st->ss->major;
 495                 inf.minor_version = st->minor_version;
 496                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
 497         } else
 498                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
 499         if (rv) {
 500                 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
 501                         mddev, strerror(errno));
 502                 return 1;
 503         }
 504
 505         if (bitmap_file) {
 506                 int uuid[4];
 507
 508                 st->ss->uuid_from_super(st, uuid);
 509                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
 510                                  delay, write_behind,
 511                                  bitmapsize,
 512                                  major_num)) {
 513                         return 1;
 514                 }
 515                 bitmap_fd = open(bitmap_file, O_RDWR);
 516                 if (bitmap_fd < 0) {
 517                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
 518                                 bitmap_file);
 519                         return 1;
 520                 }
 521                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
 522                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
 523                                 mddev, strerror(errno));
 524                         return 1;
 525                 }
 526         }
 527
 528
 529
 530         for (pass=1; pass <=2 ; pass++) {
 531                 mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
 532
 533                 for (dnum=0, dv = devlist ; dv ;
 534                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 535                         int fd;
 536                         struct stat stb;
 537
 538                         info.disk.number = dnum;
 539                         if (dnum == insert_point) {
 540                                 moved_disk = dv;
 541                         }
 542                         info.disk.raid_disk = info.disk.number;
 543                         if (info.disk.raid_disk < raiddisks)
 544                                 info.disk.state = (1<<MD_DISK_ACTIVE) |
 545                                                 (1<<MD_DISK_SYNC);
 546                         else
 547                                 info.disk.state = 0;
 548                         if (dv->writemostly)
 549                                 info.disk.state |= (1<<MD_DISK_WRITEMOSTLY);
 550
 551                         if (dnum == insert_point ||
 552                             strcasecmp(dv->devname, "missing")==0) {
 553                                 info.disk.major = 0;
 554                                 info.disk.minor = 0;
 555                                 info.disk.state = (1<<MD_DISK_FAULTY);
 556                         } else {
 557                                 fd = open(dv->devname, O_RDONLY|O_EXCL, 0);
 558                                 if (fd < 0) {
 559                                         fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
 560                                                 dv->devname);
 561                                         return 1;
 562                                 }
 563                                 fstat(fd, &stb);
 564                                 info.disk.major = major(stb.st_rdev);
 565                                 info.disk.minor = minor(stb.st_rdev);
 566                                 remove_partitions(fd);
 567                                 close(fd);
 568                         }
 569                         switch(pass){
 570                         case 1:
 571                                 st->ss->add_to_super(st, &info.disk);
 572                                 break;
 573                         case 2:
 574                                 if (info.disk.state == 1) break;
 575                                 Kill(dv->devname, 0, 1); /* Just be sure it is clean */
 576                                 Kill(dv->devname, 0, 1); /* and again, there could be two superblocks */
 577                                 st->ss->write_init_super(st, &info.disk,
 578                                                          dv->devname);
 579
 580                                 if (ioctl(mdfd, ADD_NEW_DISK, &info.disk)) {
 581                                         fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
 582                                                 dv->devname, strerror(errno));
 583                                         st->ss->free_super(st);
 584                                         return 1;
 585                                 }
 586
 587                                 break;
 588                         }
 589                         if (dv == moved_disk && dnum != insert_point) break;
 590                 }
 591         }
 592         st->ss->free_super(st);
 593
 594         /* param is not actually used */
 595         if (level == LEVEL_CONTAINER)
 596                 /* No need to start */
 597                 ;
 598         else if (runstop == 1 || subdevs >= raiddisks) {
 599                 mdu_param_t param;
 600                 if (ioctl(mdfd, RUN_ARRAY, &param)) {
 601                         fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
 602                                 strerror(errno));
 603                         Manage_runstop(mddev, mdfd, -1, 0);
 604                         return 1;
 605                 }
 606                 if (verbose >= 0)
 607                         fprintf(stderr, Name ": array %s started.\n", mddev);
 608         } else {
 609                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
 610         }
 611         return 0;
 612 }