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_FAULTY:
 184         case LEVEL_MULTIPATH:
 185                 if (chunk) {
 186                         chunk = 0;
 187                         if (verbose > 0)
 188                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
 189                 }
 190                 break;
 191         default:
 192                 fprintf(stderr, Name ": unknown level %d\n", level);
 193                 return 1;
 194         }
 195
 196         /* now look at the subdevs */
 197         array.active_disks = 0;
 198         array.working_disks = 0;
 199         dnum = 0;
 200         for (dv=devlist; dv; dv=dv->next, dnum++) {
 201                 char *dname = dv->devname;
 202                 unsigned long long ldsize, freesize;
 203                 int fd;
 204                 if (strcasecmp(dname, "missing")==0) {
 205                         if (first_missing > dnum)
 206                                 first_missing = dnum;
 207                         missing_disks ++;
 208                         continue;
 209                 }
 210                 array.working_disks++;
 211                 if (dnum < raiddisks)
 212                         array.active_disks++;
 213                 fd = open(dname, O_RDONLY|O_EXCL, 0);
 214                 if (fd <0 ) {
 215                         fprintf(stderr, Name ": Cannot open %s: %s\n",
 216                                 dname, strerror(errno));
 217                         fail=1;
 218                         continue;
 219                 }
 220                 if (!get_dev_size(fd, dname, &ldsize)) {
 221                         fail = 1;
 222                         close(fd);
 223                         continue;
 224                 }
 225                 if (st == NULL) {
 226                         struct createinfo *ci = conf_get_create_info();
 227                         if (ci)
 228                                 st = ci->supertype;
 229                 }
 230                 if (st == NULL) {
 231                         /* Need to choose a default metadata, which is different
 232                          * depending on the sizes of devices
 233                          */
 234                         int i;
 235                         char *name = "default";
 236                         if (level >= 1 && ldsize > (0x7fffffffULL<<10))
 237                                 name = "default/large";
 238                         for(i=0; !st && superlist[i]; i++)
 239                                 st = superlist[i]->match_metadata_desc(name);
 240
 241                         if (!st) {
 242                                 fprintf(stderr, Name ": internal error - no default metadata style\n");
 243                                 exit(2);
 244                         }
 245                         if (st->ss->major != 0 ||
 246                             st->minor_version != 90)
 247                                 fprintf(stderr, Name ": Defaulting to version"
 248                                         " %d.%d metadata\n",
 249                                         st->ss->major,
 250                                         st->minor_version);
 251                 }
 252                 freesize = st->ss->avail_size(st, ldsize >> 9);
 253                 if (freesize == 0) {
 254                         fprintf(stderr, Name ": %s is too small: %luK\n",
 255                                 dname, (unsigned long)(ldsize>>10));
 256                         fail = 1;
 257                         close(fd);
 258                         continue;
 259                 }
 260
 261                 freesize /= 2; /* convert to K */
 262                 if (chunk) {
 263                         /* round to chunk size */
 264                         freesize = freesize & ~(chunk-1);
 265                 }
 266
 267                 if (size && freesize < size) {
 268                         fprintf(stderr, Name ": %s is smaller that given size."
 269                                 " %lluK < %lluK + superblock\n", dname, freesize, size);
 270                         fail = 1;
 271                         close(fd);
 272                         continue;
 273                 }
 274                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
 275                         maxdisc = dname;
 276                         maxsize = freesize;
 277                 }
 278                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
 279                         mindisc = dname;
 280                         minsize = freesize;
 281                 }
 282                 if (runstop != 1 || verbose >= 0) {
 283                         warn |= check_ext2(fd, dname);
 284                         warn |= check_reiser(fd, dname);
 285                         warn |= check_raid(fd, dname);
 286                 }
 287                 close(fd);
 288         }
 289         if (fail) {
 290                 fprintf(stderr, Name ": create aborted\n");
 291                 return 1;
 292         }
 293         if (size == 0) {
 294                 if (mindisc == NULL) {
 295                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 296                         return 1;
 297                 }
 298                 if (level > 0 || level == LEVEL_MULTIPATH || level == LEVEL_FAULTY) {
 299                         /* size is meaningful */
 300                         if (minsize > 0x100000000ULL && st->ss->major == 0) {
 301                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 302                                 return 1;
 303                         }
 304                         size = minsize;
 305                         if (verbose > 0)
 306                                 fprintf(stderr, Name ": size set to %lluK\n", size);
 307                 }
 308         }
 309         if (level > 0 && ((maxsize-size)*100 > maxsize)) {
 310                 if (runstop != 1 || verbose >= 0)
 311                         fprintf(stderr, Name ": largest drive (%s) exceed size (%lluK) by more than 1%%\n",
 312                                 maxdisc, size);
 313                 warn = 1;
 314         }
 315
 316         if (warn) {
 317                 if (runstop!= 1) {
 318                         if (!ask("Continue creating array? ")) {
 319                                 fprintf(stderr, Name ": create aborted.\n");
 320                                 return 1;
 321                         }
 322                 } else {
 323                         if (verbose > 0)
 324                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 325                 }
 326         }
 327
 328         /* If this is raid4/5, we want to configure the last active slot
 329          * as missing, so that a reconstruct happens (faster than re-parity)
 330          * FIX: Can we do this for raid6 as well?
 331          */
 332         if (assume_clean==0 && force == 0 && first_missing >= raiddisks) {
 333                 switch ( level ) {
 334                 case 4:
 335                 case 5:
 336                         insert_point = raiddisks-1;
 337                         sparedisks++;
 338                         array.active_disks--;
 339                         missing_disks++;
 340                         break;
 341                 default:
 342                         break;
 343                 }
 344         }
 345
 346         if (level <= 0 && first_missing != subdevs * 2) {
 347                 fprintf(stderr,
 348                         Name ": This level does not support missing devices\n");
 349                 return 1;
 350         }
 351
 352         /* Ok, lets try some ioctls */
 353
 354         array.level = level;
 355         array.size = size;
 356         array.raid_disks = raiddisks;
 357         /* The kernel should *know* what md_minor we are dealing
 358          * with, but it chooses to trust me instead. Sigh
 359          */
 360         array.md_minor = 0;
 361         if (fstat(mdfd, &stb)==0)
 362                 array.md_minor = minor(stb.st_rdev);
 363         array.not_persistent = 0;
 364         /*** FIX: Need to do something about RAID-6 here ***/
 365         if ( ( (level == 4 || level == 5) &&
 366                (insert_point < raiddisks || first_missing < raiddisks) )
 367              ||
 368              ( level == 6 && missing_disks == 2)
 369              ||
 370              assume_clean
 371                 )
 372                 array.state = 1; /* clean, but one+ drive will be missing */
 373         else
 374                 array.state = 0; /* not clean, but no errors */
 375
 376         if (level == 10) {
 377                 /* for raid10, the bitmap size is the capacity of the array,
 378                  * which is array.size * raid_disks / ncopies;
 379                  * .. but convert to sectors.
 380                  */
 381                 int ncopies = ((layout>>8) & 255) * (layout & 255);
 382                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
 383 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
 384         } else
 385                 bitmapsize = (unsigned long long)size * 2;
 386
 387         /* There is lots of redundancy in these disk counts,
 388          * raid_disks is the most meaningful value
 389          *          it describes the geometry of the array
 390          *          it is constant
 391          * nr_disks is total number of used slots.
 392          *          it should be raid_disks+spare_disks
 393          * spare_disks is the number of extra disks present
 394          *          see above
 395          * active_disks is the number of working disks in
 396          *          active slots. (With raid_disks)
 397          * working_disks is the total number of working disks,
 398          *          including spares
 399          * failed_disks is the number of disks marked failed
 400          *
 401          * Ideally, the kernel would keep these (except raid_disks)
 402          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
 403          * So for now, we assume that all raid and spare
 404          * devices will be given.
 405          */
 406         array.spare_disks=sparedisks;
 407         array.failed_disks=missing_disks;
 408         array.nr_disks = array.working_disks + array.failed_disks;
 409         array.layout = layout;
 410         array.chunk_size = chunk*1024;
 411         array.major_version = st->ss->major;
 412
 413         if (name == NULL || *name == 0) {
 414                 /* base name on mddev */
 415                 /*  /dev/md0 -> 0
 416                  *  /dev/md_d0 -> d0
 417                  *  /dev/md/1 -> 1
 418                  *  /dev/md/d1 -> d1
 419                  *  /dev/md/home -> home
 420                  *  /dev/mdhome -> home
 421                  */
 422                 name = strrchr(mddev, '/');
 423                 if (name) {
 424                         name++;
 425                         if (strncmp(name, "md_d", 4)==0 &&
 426                             strlen(name) > 4 &&
 427                             isdigit(name[4]) &&
 428                             (name-mddev) == 5 /* /dev/ */)
 429                                 name += 3;
 430                         else if (strncmp(name, "md", 2)==0 &&
 431                                  strlen(name) > 2 &&
 432                                  isdigit(name[2]) &&
 433                                  (name-mddev) == 5 /* /dev/ */)
 434                                 name += 2;
 435                 }
 436         }
 437         if (!st->ss->init_super(st, &super, &array, size, name, homehost, uuid))
 438                 return 1;
 439
 440         if (bitmap_file && vers < 9003) {
 441                 major_num = BITMAP_MAJOR_HOSTENDIAN;
 442 #ifdef __BIG_ENDIAN
 443                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
 444                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
 445 #endif
 446         }
 447
 448         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
 449                 if ((vers%100) < 2) {
 450                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
 451                         return 1;
 452                 }
 453                 if (!st->ss->add_internal_bitmap(st, super, &bitmap_chunk,
 454                                                  delay, write_behind,
 455                                                  bitmapsize, 1, major_num)) {
 456                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
 457                         return 1;
 458                 }
 459                 bitmap_file = NULL;
 460         }
 461
 462
 463
 464         if ((vers % 100) >= 1) { /* can use different versions */
 465                 mdu_array_info_t inf;
 466                 memset(&inf, 0, sizeof(inf));
 467                 inf.major_version = st->ss->major;
 468                 inf.minor_version = st->minor_version;
 469                 rv = ioctl(mdfd, SET_ARRAY_INFO, &inf);
 470         } else
 471                 rv = ioctl(mdfd, SET_ARRAY_INFO, NULL);
 472         if (rv) {
 473                 fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
 474                         mddev, strerror(errno));
 475                 return 1;
 476         }
 477
 478         if (bitmap_file) {
 479                 int uuid[4];
 480
 481                 st->ss->uuid_from_super(st, uuid, super);
 482                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
 483                                  delay, write_behind,
 484                                  bitmapsize,
 485                                  major_num)) {
 486                         return 1;
 487                 }
 488                 bitmap_fd = open(bitmap_file, O_RDWR);
 489                 if (bitmap_fd < 0) {
 490                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
 491                                 bitmap_file);
 492                         return 1;
 493                 }
 494                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
 495                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
 496                                 mddev, strerror(errno));
 497                         return 1;
 498                 }
 499         }
 500
 501
 502
 503         for (pass=1; pass <=2 ; pass++) {
 504                 mddev_dev_t moved_disk = NULL; /* the disk that was moved out of the insert point */
 505
 506                 for (dnum=0, dv = devlist ; dv ;
 507                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 508                         int fd;
 509                         struct stat stb;
 510                         mdu_disk_info_t disk;
 511
 512                         disk.number = dnum;
 513                         if (dnum == insert_point) {
 514                                 moved_disk = dv;
 515                         }
 516                         disk.raid_disk = disk.number;
 517                         if (disk.raid_disk < raiddisks)
 518                                 disk.state = (1<<MD_DISK_ACTIVE) |
 519                                                 (1<<MD_DISK_SYNC);
 520                         else
 521                                 disk.state = 0;
 522                         if (dv->writemostly)
 523                                 disk.state |= (1<<MD_DISK_WRITEMOSTLY);
 524
 525                         if (dnum == insert_point ||
 526                             strcasecmp(dv->devname, "missing")==0) {
 527                                 disk.major = 0;
 528                                 disk.minor = 0;
 529                                 disk.state = (1<<MD_DISK_FAULTY);
 530                         } else {
 531                                 fd = open(dv->devname, O_RDONLY|O_EXCL, 0);
 532                                 if (fd < 0) {
 533                                         fprintf(stderr, Name ": failed to open %s after earlier success - aborting\n",
 534                                                 dv->devname);
 535                                         return 1;
 536                                 }
 537                                 fstat(fd, &stb);
 538                                 disk.major = major(stb.st_rdev);
 539                                 disk.minor = minor(stb.st_rdev);
 540                                 remove_partitions(fd);
 541                                 close(fd);
 542                         }
 543                         switch(pass){
 544                         case 1:
 545                                 st->ss->add_to_super(st, super, &disk);
 546                                 break;
 547                         case 2:
 548                                 if (disk.state == 1) break;
 549                                 Kill(dv->devname, 0, 1); /* Just be sure it is clean */
 550                                 Kill(dv->devname, 0, 1); /* and again, there could be two superblocks */
 551                                 st->ss->write_init_super(st, super, &disk, dv->devname);
 552
 553                                 if (ioctl(mdfd, ADD_NEW_DISK, &disk)) {
 554                                         fprintf(stderr, Name ": ADD_NEW_DISK for %s failed: %s\n",
 555                                                 dv->devname, strerror(errno));
 556                                         st->ss->free_super(st, super);
 557                                         return 1;
 558                                 }
 559
 560                                 break;
 561                         }
 562                         if (dv == moved_disk && dnum != insert_point) break;
 563                 }
 564         }
 565         st->ss->free_super(st, super);
 566
 567         /* param is not actually used */
 568         if (runstop == 1 || subdevs >= raiddisks) {
 569                 mdu_param_t param;
 570                 if (ioctl(mdfd, RUN_ARRAY, &param)) {
 571                         fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
 572                                 strerror(errno));
 573                         Manage_runstop(mddev, mdfd, -1, 0);
 574                         return 1;
 575                 }
 576                 if (verbose >= 0)
 577                         fprintf(stderr, Name ": array %s started.\n", mddev);
 578         } else {
 579                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
 580         }
 581         return 0;
 582 }