Create.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2009 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@suse.de>
  23  */
  24
  25 #include "mdadm.h"
  26 #include        "md_u.h"
  27 #include        "md_p.h"
  28 #include        <ctype.h>
  29
  30 static int default_layout(struct supertype *st, int level, int verbose)
  31 {
  32         int layout = UnSet;
  33
  34         if (st && st->ss->default_geometry)
  35                 st->ss->default_geometry(st, &level, &layout, NULL);
  36
  37         if (layout == UnSet)
  38                 switch(level) {
  39                 default: /* no layout */
  40                         layout = 0;
  41                         break;
  42                 case 10:
  43                         layout = 0x102; /* near=2, far=1 */
  44                         if (verbose > 0)
  45                                 fprintf(stderr,
  46                                         Name ": layout defaults to n2\n");
  47                         break;
  48                 case 5:
  49                 case 6:
  50                         layout = map_name(r5layout, "default");
  51                         if (verbose > 0)
  52                                 fprintf(stderr,
  53                                         Name ": layout defaults to %s\n", map_num(r5layout, layout));
  54                         break;
  55                 case LEVEL_FAULTY:
  56                         layout = map_name(faultylayout, "default");
  57
  58                         if (verbose > 0)
  59                                 fprintf(stderr,
  60                                         Name ": layout defaults to %s\n", map_num(faultylayout, layout));
  61                         break;
  62                 }
  63
  64         return layout;
  65 }
  66
  67
  68 int Create(struct supertype *st, char *mddev,
  69            int chunk, int level, int layout, unsigned long long size,
  70            int raiddisks, int sparedisks,
  71            char *name, char *homehost, int *uuid,
  72            int subdevs, struct mddev_dev *devlist,
  73            int runstop, int verbose, int force, int assume_clean,
  74            char *bitmap_file, int bitmap_chunk, int write_behind,
  75            int delay, int autof)
  76 {
  77         /*
  78          * Create a new raid array.
  79          *
  80          * First check that necessary details are available
  81          * (i.e. level, raid-disks)
  82          *
  83          * Then check each disk to see what might be on it
  84          * and report anything interesting.
  85          *
  86          * If anything looks odd, and runstop not set,
  87          * abort.
  88          *
  89          * SET_ARRAY_INFO and ADD_NEW_DISK, and
  90          * if runstop==run, or raiddisks disks were used,
  91          * RUN_ARRAY
  92          */
  93         int mdfd;
  94         unsigned long long minsize=0, maxsize=0;
  95         char *mindisc = NULL;
  96         char *maxdisc = NULL;
  97         int dnum;
  98         struct mddev_dev *dv;
  99         int fail=0, warn=0;
 100         struct stat stb;
 101         int first_missing = subdevs * 2;
 102         int second_missing = subdevs * 2;
 103         int missing_disks = 0;
 104         int insert_point = subdevs * 2; /* where to insert a missing drive */
 105         int total_slots;
 106         int pass;
 107         int vers;
 108         int rv;
 109         int bitmap_fd;
 110         int have_container = 0;
 111         int container_fd = -1;
 112         int need_mdmon = 0;
 113         unsigned long long bitmapsize;
 114         struct mdinfo info, *infos;
 115         int did_default = 0;
 116         int do_default_layout = 0;
 117         unsigned long safe_mode_delay = 0;
 118         char chosen_name[1024];
 119         struct map_ent *map = NULL;
 120         unsigned long long newsize;
 121
 122         int major_num = BITMAP_MAJOR_HI;
 123
 124         memset(&info, 0, sizeof(info));
 125         if (level == UnSet && st && st->ss->default_geometry)
 126                 st->ss->default_geometry(st, &level, NULL, NULL);
 127         if (level == UnSet) {
 128                 fprintf(stderr,
 129                         Name ": a RAID level is needed to create an array.\n");
 130                 return 1;
 131         }
 132         if (raiddisks < 4 && level == 6) {
 133                 fprintf(stderr,
 134                         Name ": at least 4 raid-devices needed for level 6\n");
 135                 return 1;
 136         }
 137         if (raiddisks > 256 && level == 6) {
 138                 fprintf(stderr,
 139                         Name ": no more than 256 raid-devices supported for level 6\n");
 140                 return 1;
 141         }
 142         if (raiddisks < 2 && level >= 4) {
 143                 fprintf(stderr,
 144                         Name ": at least 2 raid-devices needed for level 4 or 5\n");
 145                 return 1;
 146         }
 147         if (level <= 0 && sparedisks) {
 148                 fprintf(stderr,
 149                         Name ": This level does not support spare devices\n");
 150                 return 1;
 151         }
 152
 153         if (subdevs == 1 && strcmp(devlist->devname, "missing") != 0) {
 154                 /* If given a single device, it might be a container, and we can
 155                  * extract a device list from there
 156                  */
 157                 mdu_array_info_t inf;
 158                 int fd;
 159
 160                 memset(&inf, 0, sizeof(inf));
 161                 fd = open(devlist->devname, O_RDONLY);
 162                 if (fd >= 0 &&
 163                     ioctl(fd, GET_ARRAY_INFO, &inf) == 0 &&
 164                     inf.raid_disks == 0) {
 165                         /* yep, looks like a container */
 166                         if (st) {
 167                                 rv = st->ss->load_container(st, fd,
 168                                                             devlist->devname);
 169                                 if (rv == 0)
 170                                         have_container = 1;
 171                         } else {
 172                                 st = super_by_fd(fd, NULL);
 173                                 if (st && !(rv = st->ss->
 174                                             load_container(st, fd,
 175                                                            devlist->devname)))
 176                                         have_container = 1;
 177                                 else
 178                                         st = NULL;
 179                         }
 180                         if (have_container) {
 181                                 subdevs = raiddisks;
 182                                 first_missing = subdevs * 2;
 183                                 second_missing = subdevs * 2;
 184                                 insert_point = subdevs * 2;
 185                         }
 186                 }
 187                 if (fd >= 0)
 188                         close(fd);
 189         }
 190         if (st && st->ss->external && sparedisks) {
 191                 fprintf(stderr,
 192                         Name ": This metadata type does not support "
 193                         "spare disks at create time\n");
 194                 return 1;
 195         }
 196         if (subdevs > raiddisks+sparedisks) {
 197                 fprintf(stderr, Name ": You have listed more devices (%d) than are in the array(%d)!\n", subdevs, raiddisks+sparedisks);
 198                 return 1;
 199         }
 200         if (!have_container && subdevs < raiddisks+sparedisks) {
 201                 fprintf(stderr, Name ": You haven't given enough devices (real or missing) to create this array\n");
 202                 return 1;
 203         }
 204         if (bitmap_file && level <= 0) {
 205                 fprintf(stderr, Name ": bitmaps not meaningful with level %s\n",
 206                         map_num(pers, level)?:"given");
 207                 return 1;
 208         }
 209
 210         /* now set some defaults */
 211
 212
 213         if (layout == UnSet) {
 214                 do_default_layout = 1;
 215                 layout = default_layout(st, level, verbose);
 216         }
 217
 218         if (level == 10)
 219                 /* check layout fits in array*/
 220                 if ((layout&255) * ((layout>>8)&255) > raiddisks) {
 221                         fprintf(stderr, Name ": that layout requires at least %d devices\n",
 222                                 (layout&255) * ((layout>>8)&255));
 223                         return 1;
 224                 }
 225
 226         switch(level) {
 227         case 4:
 228         case 5:
 229         case 10:
 230         case 6:
 231         case 0:
 232                 if (chunk == 0) {
 233                         if (st && st->ss->default_geometry)
 234                                 st->ss->default_geometry(st, NULL, NULL, &chunk);
 235                         chunk = chunk ? : 512;
 236                         if (verbose > 0)
 237                                 fprintf(stderr, Name ": chunk size defaults to %dK\n", chunk);
 238                 }
 239                 break;
 240         case LEVEL_LINEAR:
 241                 /* a chunksize of zero 0s perfectly valid (and preferred) since 2.6.16 */
 242                 if (get_linux_version() < 2006016 && chunk == 0) {
 243                         chunk = 64;
 244                         if (verbose > 0)
 245                                 fprintf(stderr, Name ": chunk size defaults to 64K\n");
 246                 }
 247                 break;
 248         case 1:
 249         case LEVEL_FAULTY:
 250         case LEVEL_MULTIPATH:
 251         case LEVEL_CONTAINER:
 252                 if (chunk) {
 253                         chunk = 0;
 254                         if (verbose > 0)
 255                                 fprintf(stderr, Name ": chunk size ignored for this level\n");
 256                 }
 257                 break;
 258         default:
 259                 fprintf(stderr, Name ": unknown level %d\n", level);
 260                 return 1;
 261         }
 262
 263         if (size && chunk)
 264                 size &= ~(unsigned long long)(chunk - 1);
 265         newsize = size * 2;
 266         if (st && ! st->ss->validate_geometry(st, level, layout, raiddisks,
 267                                               chunk, size*2, NULL, &newsize, verbose>=0))
 268                 return 1;
 269         if (size == 0) {
 270                 size = newsize / 2;
 271                 if (size && verbose > 0)
 272                         fprintf(stderr, Name ": setting size to %lluK\n",
 273                                 (unsigned long long)size);
 274         }
 275
 276         /* now look at the subdevs */
 277         info.array.active_disks = 0;
 278         info.array.working_disks = 0;
 279         dnum = 0;
 280         for (dv=devlist; dv && !have_container; dv=dv->next, dnum++) {
 281                 char *dname = dv->devname;
 282                 unsigned long long freesize;
 283                 if (strcasecmp(dname, "missing")==0) {
 284                         if (first_missing > dnum)
 285                                 first_missing = dnum;
 286                         if (second_missing > dnum && dnum > first_missing)
 287                                 second_missing = dnum;
 288                         missing_disks ++;
 289                         continue;
 290                 }
 291                 info.array.working_disks++;
 292                 if (dnum < raiddisks)
 293                         info.array.active_disks++;
 294                 if (st == NULL) {
 295                         struct createinfo *ci = conf_get_create_info();
 296                         if (ci)
 297                                 st = ci->supertype;
 298                 }
 299                 if (st == NULL) {
 300                         /* Need to choose a default metadata, which is different
 301                          * depending on geometry of array.
 302                          */
 303                         int i;
 304                         char *name = "default";
 305                         for(i=0; !st && superlist[i]; i++) {
 306                                 st = superlist[i]->match_metadata_desc(name);
 307                                 if (do_default_layout)
 308                                         layout = default_layout(st, level, verbose);
 309                                 if (st && !st->ss->validate_geometry
 310                                                 (st, level, layout, raiddisks,
 311                                                  chunk, size*2, dname, &freesize,
 312                                                  verbose > 0))
 313                                         st = NULL;
 314                         }
 315
 316                         if (!st) {
 317                                 fprintf(stderr, Name ": device %s not suitable "
 318                                         "for any style of array\n",
 319                                         dname);
 320                                 exit(2);
 321                         }
 322                         if (st->ss != &super0 ||
 323                             st->minor_version != 90)
 324                                 did_default = 1;
 325                 } else {
 326                         if (do_default_layout)
 327                                 layout = default_layout(st, level, verbose);
 328                         if (!st->ss->validate_geometry(st, level, layout,
 329                                                        raiddisks,
 330                                                        chunk, size*2, dname,
 331                                                        &freesize,
 332                                                        verbose >= 0)) {
 333
 334                                 fprintf(stderr,
 335                                         Name ": %s is not suitable for "
 336                                         "this array.\n",
 337                                         dname);
 338                                 fail = 1;
 339                                 continue;
 340                         }
 341                 }
 342
 343                 freesize /= 2; /* convert to K */
 344                 if (chunk) {
 345                         /* round to chunk size */
 346                         freesize = freesize & ~(chunk-1);
 347                 }
 348
 349                 if (size && freesize < size) {
 350                         fprintf(stderr, Name ": %s is smaller than given size."
 351                                 " %lluK < %lluK + metadata\n",
 352                                 dname, freesize, size);
 353                         fail = 1;
 354                         continue;
 355                 }
 356                 if (maxdisc == NULL || (maxdisc && freesize > maxsize)) {
 357                         maxdisc = dname;
 358                         maxsize = freesize;
 359                 }
 360                 if (mindisc ==NULL || (mindisc && freesize < minsize)) {
 361                         mindisc = dname;
 362                         minsize = freesize;
 363                 }
 364                 if (runstop != 1 || verbose >= 0) {
 365                         int fd = open(dname, O_RDONLY);
 366                         if (fd <0 ) {
 367                                 fprintf(stderr, Name ": Cannot open %s: %s\n",
 368                                         dname, strerror(errno));
 369                                 fail=1;
 370                                 continue;
 371                         }
 372                         warn |= check_ext2(fd, dname);
 373                         warn |= check_reiser(fd, dname);
 374                         warn |= check_raid(fd, dname);
 375                         if (strcmp(st->ss->name, "1.x") == 0 &&
 376                             st->minor_version >= 1)
 377                                 /* metadata at front */
 378                                 warn |= check_partitions(fd, dname, 0);
 379                         else if (level == 1 || level == LEVEL_CONTAINER)
 380                                 /* partitions could be meaningful */
 381                                 warn |= check_partitions(fd, dname, freesize*2);
 382                         else
 383                                 /* partitions cannot be meaningful */
 384                                 warn |= check_partitions(fd, dname, 0);
 385                         if (strcmp(st->ss->name, "1.x") == 0 &&
 386                             st->minor_version >= 1 &&
 387                             did_default &&
 388                             level == 1 &&
 389                             (warn & 1024) == 0) {
 390                                 warn |= 1024;
 391                                 fprintf(stderr, Name ": Note: this array has metadata at the start and\n"
 392                                         "    may not be suitable as a boot device.  If you plan to\n"
 393                                         "    store '/boot' on this device please ensure that\n"
 394                                         "    your boot-loader understands md/v1.x metadata, or use\n"
 395                                         "    --metadata=0.90\n");
 396                         }
 397                         close(fd);
 398                 }
 399         }
 400         if (raiddisks + sparedisks > st->max_devs) {
 401                 fprintf(stderr, Name ": Too many devices:"
 402                         " %s metadata only supports %d\n",
 403                         st->ss->name, st->max_devs);
 404                 return 1;
 405         }
 406         if (have_container)
 407                 info.array.working_disks = raiddisks;
 408         if (fail) {
 409                 fprintf(stderr, Name ": create aborted\n");
 410                 return 1;
 411         }
 412         if (size == 0) {
 413                 if (mindisc == NULL && !have_container) {
 414                         fprintf(stderr, Name ": no size and no drives given - aborting create.\n");
 415                         return 1;
 416                 }
 417                 if (level > 0 || level == LEVEL_MULTIPATH
 418                     || level == LEVEL_FAULTY
 419                     || st->ss->external ) {
 420                         /* size is meaningful */
 421                         if (!st->ss->validate_geometry(st, level, layout,
 422                                                        raiddisks,
 423                                                        chunk, minsize*2,
 424                                                        NULL, NULL, 0)) {
 425                                 fprintf(stderr, Name ": devices too large for RAID level %d\n", level);
 426                                 return 1;
 427                         }
 428                         size = minsize;
 429                         if (verbose > 0)
 430                                 fprintf(stderr, Name ": size set to %lluK\n", size);
 431                 }
 432         }
 433         if (!have_container && level > 0 && ((maxsize-size)*100 > maxsize)) {
 434                 if (runstop != 1 || verbose >= 0)
 435                         fprintf(stderr, Name ": largest drive (%s) exceeds size (%lluK) by more than 1%%\n",
 436                                 maxdisc, size);
 437                 warn = 1;
 438         }
 439
 440         if (st->ss->detail_platform && st->ss->detail_platform(0, 1) != 0) {
 441                 if (runstop != 1 || verbose >= 0)
 442                         fprintf(stderr, Name ": %s unable to enumerate platform support\n"
 443                                 "    array may not be compatible with hardware/firmware\n",
 444                                 st->ss->name);
 445                 warn = 1;
 446         }
 447
 448         if (warn) {
 449                 if (runstop!= 1) {
 450                         if (!ask("Continue creating array? ")) {
 451                                 fprintf(stderr, Name ": create aborted.\n");
 452                                 return 1;
 453                         }
 454                 } else {
 455                         if (verbose > 0)
 456                                 fprintf(stderr, Name ": creation continuing despite oddities due to --run\n");
 457                 }
 458         }
 459
 460         /* If this is raid4/5, we want to configure the last active slot
 461          * as missing, so that a reconstruct happens (faster than re-parity)
 462          * FIX: Can we do this for raid6 as well?
 463          */
 464         if (st->ss->external == 0 &&
 465             assume_clean==0 && force == 0 && first_missing >= raiddisks) {
 466                 switch ( level ) {
 467                 case 4:
 468                 case 5:
 469                         insert_point = raiddisks-1;
 470                         sparedisks++;
 471                         info.array.active_disks--;
 472                         missing_disks++;
 473                         break;
 474                 default:
 475                         break;
 476                 }
 477         }
 478         /* For raid6, if creating with 1 missing drive, make a good drive
 479          * into a spare, else the create will fail
 480          */
 481         if (assume_clean == 0 && force == 0 && first_missing < raiddisks &&
 482             st->ss->external == 0 &&
 483             second_missing >= raiddisks && level == 6) {
 484                 insert_point = raiddisks - 1;
 485                 if (insert_point == first_missing)
 486                         insert_point--;
 487                 sparedisks ++;
 488                 info.array.active_disks--;
 489                 missing_disks++;
 490         }
 491
 492         if (level <= 0 && first_missing < subdevs * 2) {
 493                 fprintf(stderr,
 494                         Name ": This level does not support missing devices\n");
 495                 return 1;
 496         }
 497
 498         /* We need to create the device */
 499         map_lock(&map);
 500         mdfd = create_mddev(mddev, name, autof, LOCAL, chosen_name);
 501         if (mdfd < 0)
 502                 return 1;
 503         mddev = chosen_name;
 504
 505         vers = md_get_version(mdfd);
 506         if (vers < 9000) {
 507                 fprintf(stderr, Name ": Create requires md driver version 0.90.0 or later\n");
 508                 goto abort;
 509         } else {
 510                 mdu_array_info_t inf;
 511                 memset(&inf, 0, sizeof(inf));
 512                 ioctl(mdfd, GET_ARRAY_INFO, &inf);
 513                 if (inf.working_disks != 0) {
 514                         fprintf(stderr, Name ": another array by this name"
 515                                 " is already running.\n");
 516                         goto abort;
 517                 }
 518         }
 519
 520         /* Ok, lets try some ioctls */
 521
 522         info.array.level = level;
 523         info.array.size = size;
 524         info.array.raid_disks = raiddisks;
 525         /* The kernel should *know* what md_minor we are dealing
 526          * with, but it chooses to trust me instead. Sigh
 527          */
 528         info.array.md_minor = 0;
 529         if (fstat(mdfd, &stb)==0)
 530                 info.array.md_minor = minor(stb.st_rdev);
 531         info.array.not_persistent = 0;
 532
 533         if ( ( (level == 4 || level == 5) &&
 534                (insert_point < raiddisks || first_missing < raiddisks) )
 535              ||
 536              ( level == 6 && (insert_point < raiddisks
 537                               || second_missing < raiddisks))
 538              ||
 539              ( level <= 0 )
 540              ||
 541              assume_clean
 542                 ) {
 543                 info.array.state = 1; /* clean, but one+ drive will be missing*/
 544                 info.resync_start = MaxSector;
 545         } else {
 546                 info.array.state = 0; /* not clean, but no errors */
 547                 info.resync_start = 0;
 548         }
 549         if (level == 10) {
 550                 /* for raid10, the bitmap size is the capacity of the array,
 551                  * which is array.size * raid_disks / ncopies;
 552                  * .. but convert to sectors.
 553                  */
 554                 int ncopies = ((layout>>8) & 255) * (layout & 255);
 555                 bitmapsize = (unsigned long long)size * raiddisks / ncopies * 2;
 556 /*              printf("bms=%llu as=%d rd=%d nc=%d\n", bitmapsize, size, raiddisks, ncopies);*/
 557         } else
 558                 bitmapsize = (unsigned long long)size * 2;
 559
 560         /* There is lots of redundancy in these disk counts,
 561          * raid_disks is the most meaningful value
 562          *          it describes the geometry of the array
 563          *          it is constant
 564          * nr_disks is total number of used slots.
 565          *          it should be raid_disks+spare_disks
 566          * spare_disks is the number of extra disks present
 567          *          see above
 568          * active_disks is the number of working disks in
 569          *          active slots. (With raid_disks)
 570          * working_disks is the total number of working disks,
 571          *          including spares
 572          * failed_disks is the number of disks marked failed
 573          *
 574          * Ideally, the kernel would keep these (except raid_disks)
 575          * up-to-date as we ADD_NEW_DISK, but it doesn't (yet).
 576          * So for now, we assume that all raid and spare
 577          * devices will be given.
 578          */
 579         info.array.spare_disks=sparedisks;
 580         info.array.failed_disks=missing_disks;
 581         info.array.nr_disks = info.array.working_disks
 582                 + info.array.failed_disks;
 583         info.array.layout = layout;
 584         info.array.chunk_size = chunk*1024;
 585
 586         if (name == NULL || *name == 0) {
 587                 /* base name on mddev */
 588                 /*  /dev/md0 -> 0
 589                  *  /dev/md_d0 -> d0
 590                  *  /dev/md/1 -> 1
 591                  *  /dev/md/d1 -> d1
 592                  *  /dev/md/home -> home
 593                  *  /dev/mdhome -> home
 594                  */
 595                 /* FIXME compare this with rules in create_mddev */
 596                 name = strrchr(mddev, '/');
 597                 if (name) {
 598                         name++;
 599                         if (strncmp(name, "md_d", 4)==0 &&
 600                             strlen(name) > 4 &&
 601                             isdigit(name[4]) &&
 602                             (name-mddev) == 5 /* /dev/ */)
 603                                 name += 3;
 604                         else if (strncmp(name, "md", 2)==0 &&
 605                                  strlen(name) > 2 &&
 606                                  isdigit(name[2]) &&
 607                                  (name-mddev) == 5 /* /dev/ */)
 608                                 name += 2;
 609                 }
 610         }
 611         if (!st->ss->init_super(st, &info.array, size, name, homehost, uuid))
 612                 goto abort;
 613
 614         total_slots = info.array.nr_disks;
 615         sysfs_init(&info, mdfd, 0);
 616         st->ss->getinfo_super(st, &info, NULL);
 617
 618         if (did_default && verbose >= 0) {
 619                 if (is_subarray(info.text_version)) {
 620                         int dnum = devname2devnum(info.text_version+1);
 621                         char *path;
 622                         int mdp = get_mdp_major();
 623                         struct mdinfo *mdi;
 624                         if (dnum > 0)
 625                                 path = map_dev(MD_MAJOR, dnum, 1);
 626                         else
 627                                 path = map_dev(mdp, (-1-dnum)<< 6, 1);
 628
 629                         mdi = sysfs_read(-1, dnum, GET_VERSION);
 630
 631                         fprintf(stderr, Name ": Creating array inside "
 632                                 "%s container %s\n",
 633                                 mdi?mdi->text_version:"managed", path);
 634                         sysfs_free(mdi);
 635                 } else
 636                         fprintf(stderr, Name ": Defaulting to version"
 637                                 " %s metadata\n", info.text_version);
 638         }
 639
 640         map_update(&map, fd2devnum(mdfd), info.text_version,
 641                    info.uuid, chosen_name);
 642         map_unlock(&map);
 643
 644         if (bitmap_file && vers < 9003) {
 645                 major_num = BITMAP_MAJOR_HOSTENDIAN;
 646 #ifdef __BIG_ENDIAN
 647                 fprintf(stderr, Name ": Warning - bitmaps created on this kernel are not portable\n"
 648                         "  between different architectured.  Consider upgrading the Linux kernel.\n");
 649 #endif
 650         }
 651
 652         if (bitmap_file && strcmp(bitmap_file, "internal")==0) {
 653                 if ((vers%100) < 2) {
 654                         fprintf(stderr, Name ": internal bitmaps not supported by this kernel.\n");
 655                         goto abort;
 656                 }
 657                 if (!st->ss->add_internal_bitmap) {
 658                         fprintf(stderr, Name ": internal bitmaps not supported with %s metadata\n",
 659                                 st->ss->name);
 660                         goto abort;
 661                 }
 662                 if (!st->ss->add_internal_bitmap(st, &bitmap_chunk,
 663                                                  delay, write_behind,
 664                                                  bitmapsize, 1, major_num)) {
 665                         fprintf(stderr, Name ": Given bitmap chunk size not supported.\n");
 666                         goto abort;
 667                 }
 668                 bitmap_file = NULL;
 669         }
 670
 671
 672         sysfs_init(&info, mdfd, 0);
 673
 674         if (st->ss->external && st->container_dev != NoMdDev) {
 675                 /* member */
 676
 677                 /* When creating a member, we need to be careful
 678                  * to negotiate with mdmon properly.
 679                  * If it is already running, we cannot write to
 680                  * the devices and must ask it to do that part.
 681                  * If it isn't running, we write to the devices,
 682                  * and then start it.
 683                  * We hold an exclusive open on the container
 684                  * device to make sure mdmon doesn't exit after
 685                  * we checked that it is running.
 686                  *
 687                  * For now, fail if it is already running.
 688                  */
 689                 container_fd = open_dev_excl(st->container_dev);
 690                 if (container_fd < 0) {
 691                         fprintf(stderr, Name ": Cannot get exclusive "
 692                                 "open on container - weird.\n");
 693                         goto abort;
 694                 }
 695                 if (mdmon_running(st->container_dev)) {
 696                         if (verbose)
 697                                 fprintf(stderr, Name ": reusing mdmon "
 698                                         "for %s.\n",
 699                                         devnum2devname(st->container_dev));
 700                         st->update_tail = &st->updates;
 701                 } else
 702                         need_mdmon = 1;
 703         }
 704         rv = set_array_info(mdfd, st, &info);
 705         if (rv) {
 706                 fprintf(stderr, Name ": failed to set array info for %s: %s\n",
 707                         mddev, strerror(errno));
 708                 goto abort;
 709         }
 710
 711         if (bitmap_file) {
 712                 int uuid[4];
 713
 714                 st->ss->uuid_from_super(st, uuid);
 715                 if (CreateBitmap(bitmap_file, force, (char*)uuid, bitmap_chunk,
 716                                  delay, write_behind,
 717                                  bitmapsize,
 718                                  major_num)) {
 719                         goto abort;
 720                 }
 721                 bitmap_fd = open(bitmap_file, O_RDWR);
 722                 if (bitmap_fd < 0) {
 723                         fprintf(stderr, Name ": weird: %s cannot be openned\n",
 724                                 bitmap_file);
 725                         goto abort;
 726                 }
 727                 if (ioctl(mdfd, SET_BITMAP_FILE, bitmap_fd) < 0) {
 728                         fprintf(stderr, Name ": Cannot set bitmap file for %s: %s\n",
 729                                 mddev, strerror(errno));
 730                         goto abort;
 731                 }
 732         }
 733
 734         infos = malloc(sizeof(*infos) * total_slots);
 735
 736         for (pass=1; pass <=2 ; pass++) {
 737                 struct mddev_dev *moved_disk = NULL; /* the disk that was moved out of the insert point */
 738
 739                 for (dnum=0, dv = devlist ; dv ;
 740                      dv=(dv->next)?(dv->next):moved_disk, dnum++) {
 741                         int fd;
 742                         struct stat stb;
 743                         struct mdinfo *inf = &infos[dnum];
 744
 745                         if (dnum >= total_slots)
 746                                 abort();
 747                         if (dnum == insert_point) {
 748                                 moved_disk = dv;
 749                                 continue;
 750                         }
 751                         if (strcasecmp(dv->devname, "missing")==0)
 752                                 continue;
 753                         if (have_container)
 754                                 moved_disk = NULL;
 755                         if (have_container && dnum < info.array.raid_disks - 1)
 756                                 /* repeatedly use the container */
 757                                 moved_disk = dv;
 758
 759                         switch(pass) {
 760                         case 1:
 761                                 *inf = info;
 762
 763                                 inf->disk.number = dnum;
 764                                 inf->disk.raid_disk = dnum;
 765                                 if (inf->disk.raid_disk < raiddisks)
 766                                         inf->disk.state = (1<<MD_DISK_ACTIVE) |
 767                                                 (1<<MD_DISK_SYNC);
 768                                 else
 769                                         inf->disk.state = 0;
 770
 771                                 if (dv->writemostly == 1)
 772                                         inf->disk.state |= (1<<MD_DISK_WRITEMOSTLY);
 773
 774                                 if (have_container)
 775                                         fd = -1;
 776                                 else {
 777                                         if (st->ss->external &&
 778                                             st->container_dev != NoMdDev)
 779                                                 fd = open(dv->devname, O_RDWR);
 780                                         else
 781                                                 fd = open(dv->devname, O_RDWR|O_EXCL);
 782
 783                                         if (fd < 0) {
 784                                                 fprintf(stderr, Name ": failed to open %s "
 785                                                         "after earlier success - aborting\n",
 786                                                         dv->devname);
 787                                                 goto abort;
 788                                         }
 789                                         fstat(fd, &stb);
 790                                         inf->disk.major = major(stb.st_rdev);
 791                                         inf->disk.minor = minor(stb.st_rdev);
 792                                 }
 793                                 if (fd >= 0)
 794                                         remove_partitions(fd);
 795                                 if (st->ss->add_to_super(st, &inf->disk,
 796                                                          fd, dv->devname)) {
 797                                         ioctl(mdfd, STOP_ARRAY, NULL);
 798                                         goto abort;
 799                                 }
 800                                 st->ss->getinfo_super(st, inf, NULL);
 801                                 safe_mode_delay = inf->safe_mode_delay;
 802
 803                                 if (have_container && verbose > 0)
 804                                         fprintf(stderr, Name ": Using %s for device %d\n",
 805                                                 map_dev(inf->disk.major,
 806                                                         inf->disk.minor,
 807                                                         0), dnum);
 808
 809                                 if (!have_container) {
 810                                         /* getinfo_super might have lost these ... */
 811                                         inf->disk.major = major(stb.st_rdev);
 812                                         inf->disk.minor = minor(stb.st_rdev);
 813                                 }
 814                                 break;
 815                         case 2:
 816                                 inf->errors = 0;
 817                                 rv = 0;
 818
 819                                 rv = add_disk(mdfd, st, &info, inf);
 820
 821                                 if (rv) {
 822                                         fprintf(stderr,
 823                                                 Name ": ADD_NEW_DISK for %s "
 824                                                 "failed: %s\n",
 825                                                 dv->devname, strerror(errno));
 826                                         goto abort;
 827                                 }
 828                                 break;
 829                         }
 830                         if (!have_container &&
 831                             dv == moved_disk && dnum != insert_point) break;
 832                 }
 833                 if (pass == 1) {
 834                         struct mdinfo info_new;
 835                         struct map_ent *me = NULL;
 836
 837                         /* check to see if the uuid has changed due to these
 838                          * metadata changes, and if so update the member array
 839                          * and container uuid.  Note ->write_init_super clears
 840                          * the subarray cursor such that ->getinfo_super once
 841                          * again returns container info.
 842                          */
 843                         map_lock(&map);
 844                         st->ss->getinfo_super(st, &info_new, NULL);
 845                         if (st->ss->external && level != LEVEL_CONTAINER &&
 846                             !same_uuid(info_new.uuid, info.uuid, 0)) {
 847                                 map_update(&map, fd2devnum(mdfd),
 848                                            info_new.text_version,
 849                                            info_new.uuid, chosen_name);
 850                                 me = map_by_devnum(&map, st->container_dev);
 851                         }
 852
 853                         st->ss->write_init_super(st);
 854
 855                         /* update parent container uuid */
 856                         if (me) {
 857                                 char *path = strdup(me->path);
 858
 859                                 st->ss->getinfo_super(st, &info_new, NULL);
 860                                 map_update(&map, st->container_dev,
 861                                            info_new.text_version,
 862                                            info_new.uuid, path);
 863                                 free(path);
 864                         }
 865                         map_unlock(&map);
 866
 867                         flush_metadata_updates(st);
 868                         st->ss->free_super(st);
 869                 }
 870         }
 871         free(infos);
 872
 873         if (level == LEVEL_CONTAINER) {
 874                 /* No need to start.  But we should signal udev to
 875                  * create links */
 876                 sysfs_uevent(&info, "change");
 877                 if (verbose >= 0)
 878                         fprintf(stderr, Name ": container %s prepared.\n", mddev);
 879                 wait_for(chosen_name, mdfd);
 880         } else if (runstop == 1 || subdevs >= raiddisks) {
 881                 if (st->ss->external) {
 882                         int err;
 883                         switch(level) {
 884                         case LEVEL_LINEAR:
 885                         case LEVEL_MULTIPATH:
 886                         case 0:
 887                                 err = sysfs_set_str(&info, NULL, "array_state",
 888                                                     "active");
 889                                 need_mdmon = 0;
 890                                 break;
 891                         default:
 892                                 err = sysfs_set_str(&info, NULL, "array_state",
 893                                                     "readonly");
 894                                 break;
 895                         }
 896                         sysfs_set_safemode(&info, safe_mode_delay);
 897                         if (err) {
 898                                 fprintf(stderr, Name ": failed to"
 899                                         " activate array.\n");
 900                                 ioctl(mdfd, STOP_ARRAY, NULL);
 901                                 goto abort;
 902                         }
 903                 } else {
 904                         /* param is not actually used */
 905                         mdu_param_t param;
 906                         if (ioctl(mdfd, RUN_ARRAY, &param)) {
 907                                 fprintf(stderr, Name ": RUN_ARRAY failed: %s\n",
 908                                         strerror(errno));
 909                                 ioctl(mdfd, STOP_ARRAY, NULL);
 910                                 goto abort;
 911                         }
 912                 }
 913                 if (verbose >= 0)
 914                         fprintf(stderr, Name ": array %s started.\n", mddev);
 915                 if (st->ss->external && st->container_dev != NoMdDev) {
 916                         if (need_mdmon)
 917                                 start_mdmon(st->container_dev);
 918
 919                         ping_monitor(devnum2devname(st->container_dev));
 920                         close(container_fd);
 921                 }
 922                 wait_for(chosen_name, mdfd);
 923         } else {
 924                 fprintf(stderr, Name ": not starting array - not enough devices.\n");
 925         }
 926         close(mdfd);
 927         return 0;
 928
 929  abort:
 930         map_lock(&map);
 931         map_remove(&map, fd2devnum(mdfd));
 932         map_unlock(&map);
 933
 934         if (mdfd >= 0)
 935                 close(mdfd);
 936         return 1;
 937 }