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