Assemble.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
   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
  32 int Assemble(char *mddev, int mdfd,
  33              mddev_ident_t ident, char *conffile,
  34              mddev_dev_t devlist,
  35              int readonly, int runstop,
  36              char *update,
  37              int verbose, int force)
  38 {
  39         /*
  40          * The task of Assemble is to find a collection of
  41          * devices that should (according to their superblocks)
  42          * form an array, and to give this collection to the MD driver.
  43          * In Linux-2.4 and later, this involves submitting a
  44          * SET_ARRAY_INFO ioctl with no arg - to prepare
  45          * the array - and then submit a number of
  46          * ADD_NEW_DISK ioctls to add disks into
  47          * the array.  Finally RUN_ARRAY might
  48          * be submitted to start the array.
  49          *
  50          * Much of the work of Assemble is in finding and/or
  51          * checking the disks to make sure they look right.
  52          *
  53          * If mddev is not set, then scan must be set and we
  54          *  read through the config file for dev+uuid mapping
  55          *  We recurse, setting mddev, for each device that
  56          *    - isn't running
  57          *    - has a valid uuid (or any uuid if !uuidset)
  58          *
  59          * If mddev is set, we try to determine state of md.
  60          *   check version - must be at least 0.90.0
  61          *   check kernel version.  must be at least 2.4.
  62          *    If not, we can possibly fall back on START_ARRAY
  63          *   Try to GET_ARRAY_INFO.
  64          *     If possible, give up
  65          *     If not, try to STOP_ARRAY just to make sure
  66          *
  67          * If !uuidset and scan, look in conf-file for uuid
  68          *       If not found, give up
  69          * If !devlist and scan and uuidset, get list of devs from conf-file
  70          *
  71          * For each device:
  72          *   Check superblock - discard if bad
  73          *   Check uuid (set if we don't have one) - discard if no match
  74          *   Check superblock similarity if we have a superblock - discard if different
  75          *   Record events, devicenum
  76          * This should give us a list of devices for the array
  77          * We should collect the most recent event number
  78          *
  79          * Count disks with recent enough event count
  80          * While force && !enough disks
  81          *    Choose newest rejected disks, update event count
  82          *     mark clean and rewrite superblock
  83          * If recent kernel:
  84          *    SET_ARRAY_INFO
  85          *    foreach device with recent events : ADD_NEW_DISK
  86          *    if runstop == 1 || "enough" disks and runstop==0 -> RUN_ARRAY
  87          * If old kernel:
  88          *    Check the device numbers in superblock are right
  89          *    update superblock if any changes
  90          *    START_ARRAY
  91          *
  92          */
  93         int old_linux = 0;
  94         int vers;
  95         void *first_super = NULL, *super = NULL;
  96         struct {
  97                 char *devname;
  98                 unsigned int major, minor;
  99                 unsigned int oldmajor, oldminor;
 100                 long long events;
 101                 int uptodate;
 102                 int state;
 103                 int raid_disk;
 104         } *devices;
 105         int *best = NULL; /* indexed by raid_disk */
 106         unsigned int bestcnt = 0;
 107         int devcnt = 0;
 108         unsigned int okcnt, sparecnt;
 109         unsigned int req_cnt;
 110         unsigned int i;
 111         int most_recent = 0;
 112         int chosen_drive;
 113         int change = 0;
 114         int inargv = 0;
 115         int start_partial_ok = force || devlist==NULL;
 116         unsigned int num_devs;
 117         mddev_dev_t tmpdev;
 118         struct mdinfo info;
 119
 120         vers = md_get_version(mdfd);
 121         if (vers <= 0) {
 122                 fprintf(stderr, Name ": %s appears not to be an md device.\n", mddev);
 123                 return 1;
 124         }
 125         if (vers < 9000) {
 126                 fprintf(stderr, Name ": Assemble requires driver version 0.90.0 or later.\n"
 127                         "    Upgrade your kernel or try --build\n");
 128                 return 1;
 129         }
 130         if (get_linux_version() < 2004000)
 131                 old_linux = 1;
 132
 133         if (ioctl(mdfd, GET_ARRAY_INFO, &info.array)>=0) {
 134                 fprintf(stderr, Name ": device %s already active - cannot assemble it\n",
 135                         mddev);
 136                 return 1;
 137         }
 138         ioctl(mdfd, STOP_ARRAY, NULL); /* just incase it was started but has no content */
 139
 140         /*
 141          * If any subdevs are listed, then any that don't
 142          * match ident are discarded.  Remainder must all match and
 143          * become the array.
 144          * If no subdevs, then we scan all devices in the config file, but
 145          * there must be something in the identity
 146          */
 147
 148         if (!devlist &&
 149             ident->uuid_set == 0 &&
 150             ident->super_minor < 0 &&
 151             ident->devices == NULL) {
 152                 fprintf(stderr, Name ": No identity information available for %s - cannot assemble.\n",
 153                         mddev);
 154                 return 1;
 155         }
 156         if (devlist == NULL)
 157                 devlist = conf_get_devs(conffile);
 158         else inargv = 1;
 159
 160         tmpdev = devlist; num_devs = 0;
 161         while (tmpdev) {
 162                 num_devs++;
 163                 tmpdev = tmpdev->next;
 164         }
 165         devices = malloc(num_devs * sizeof(*devices));
 166
 167
 168         if (verbose)
 169             fprintf(stderr, Name ": looking for devices for %s\n",
 170                     mddev);
 171
 172         while ( devlist) {
 173                 char *devname;
 174                 int dfd;
 175                 struct stat stb;
 176
 177                 devname = devlist->devname;
 178                 devlist = devlist->next;
 179
 180                 if (ident->devices &&
 181                     !match_oneof(ident->devices, devname)) {
 182                         if (inargv || verbose)
 183                                 fprintf(stderr, Name ": %s is not one of %s\n", devname, ident->devices);
 184                         continue;
 185                 }
 186
 187                 if (super) {
 188                         free(super);
 189                         super = NULL;
 190                 }
 191
 192                 dfd = open(devname, O_RDONLY|O_EXCL, 0);
 193                 if (dfd < 0) {
 194                         if (inargv || verbose)
 195                                 fprintf(stderr, Name ": cannot open device %s: %s\n",
 196                                         devname, strerror(errno));
 197                 } else if (fstat(dfd, &stb)< 0) {
 198                         /* Impossible! */
 199                         fprintf(stderr, Name ": fstat failed for %s: %s\n",
 200                                 devname, strerror(errno));
 201                         close(dfd);
 202                 } else if ((stb.st_mode & S_IFMT) != S_IFBLK) {
 203                         fprintf(stderr, Name ": %s is not a block device.\n",
 204                                 devname);
 205                         close(dfd);
 206                 } else if (load_super0(dfd, &super, NULL)) {
 207                         if (inargv || verbose)
 208                                 fprintf( stderr, Name ": no RAID superblock on %s\n",
 209                                          devname);
 210                         close(dfd);
 211                 } else {
 212                         getinfo_super0(&info, super);
 213                         close(dfd);
 214                 }
 215
 216                 if (ident->uuid_set &&
 217                     (!super || same_uuid(info.uuid, ident->uuid)==0)) {
 218                         if (inargv || verbose)
 219                                 fprintf(stderr, Name ": %s has wrong uuid.\n",
 220                                         devname);
 221                         continue;
 222                 }
 223                 if (ident->super_minor != UnSet &&
 224                     (!super || ident->super_minor != info.array.md_minor)) {
 225                         if (inargv || verbose)
 226                                 fprintf(stderr, Name ": %s has wrong super-minor.\n",
 227                                         devname);
 228                         continue;
 229                 }
 230                 if (ident->level != UnSet &&
 231                     (!super|| ident->level != info.array.level)) {
 232                         if (inargv || verbose)
 233                                 fprintf(stderr, Name ": %s has wrong raid level.\n",
 234                                         devname);
 235                         continue;
 236                 }
 237                 if (ident->raid_disks != UnSet &&
 238                     (!super || ident->raid_disks!= info.array.raid_disks)) {
 239                         if (inargv || verbose)
 240                                 fprintf(stderr, Name ": %s requires wrong number of drives.\n",
 241                                         devname);
 242                         continue;
 243                 }
 244
 245                 /* If we are this far, then we are commited to this device.
 246                  * If the super_block doesn't exist, or doesn't match others,
 247                  * then we cannot continue
 248                  */
 249
 250                 if (!super) {
 251                         fprintf(stderr, Name ": %s has no superblock - assembly aborted\n",
 252                                 devname);
 253                         free(first_super);
 254                         return 1;
 255                 }
 256                 if (compare_super0(&first_super, super)) {
 257                         fprintf(stderr, Name ": superblock on %s doesn't match others - assembly aborted\n",
 258                                 devname);
 259                         free(super);
 260                         free(first_super);
 261                         return 1;
 262                 }
 263
 264                 /* looks like a good enough match to update the super block if needed */
 265                 if (update) {
 266                         /* prepare useful information in info structures */
 267                         struct stat stb2;
 268                         fstat(mdfd, &stb2);
 269                         info.array.md_minor = minor(stb2.st_rdev);
 270
 271                         update_super0(&info, super, update, devname, verbose);
 272
 273                         dfd = open(devname, O_RDWR|O_EXCL, 0);
 274                         if (dfd < 0)
 275                                 fprintf(stderr, Name ": Cannot open %s for superblock update\n",
 276                                         devname);
 277                         else if (store_super0(dfd, super))
 278                                 fprintf(stderr, Name ": Could not re-write superblock on %s.\n",
 279                                         devname);
 280                         if (dfd >= 0)
 281                                 close(dfd);
 282                 }
 283
 284                 if (verbose)
 285                         fprintf(stderr, Name ": %s is identified as a member of %s, slot %d.\n",
 286                                 devname, mddev, info.disk.raid_disk);
 287                 devices[devcnt].devname = devname;
 288                 devices[devcnt].major = major(stb.st_rdev);
 289                 devices[devcnt].minor = minor(stb.st_rdev);
 290                 devices[devcnt].oldmajor = info.disk.major;
 291                 devices[devcnt].oldminor = info.disk.minor;
 292                 devices[devcnt].events = info.events;
 293                 devices[devcnt].raid_disk = info.disk.raid_disk;
 294                 devices[devcnt].uptodate = 0;
 295                 devices[devcnt].state = info.disk.state;
 296                 if (most_recent < devcnt) {
 297                         if (devices[devcnt].events
 298                             > devices[most_recent].events)
 299                                 most_recent = devcnt;
 300                 }
 301                 if (info.array.level == -4)
 302                         /* with multipath, the raid_disk from the superblock is meaningless */
 303                         i = devcnt;
 304                 else
 305                         i = devices[devcnt].raid_disk;
 306                 if (i < 10000) {
 307                         if (i >= bestcnt) {
 308                                 unsigned int newbestcnt = i+10;
 309                                 int *newbest = malloc(sizeof(int)*newbestcnt);
 310                                 unsigned int c;
 311                                 for (c=0; c < newbestcnt; c++)
 312                                         if (c < bestcnt)
 313                                                 newbest[c] = best[c];
 314                                         else
 315                                                 newbest[c] = -1;
 316                                 if (best)free(best);
 317                                 best = newbest;
 318                                 bestcnt = newbestcnt;
 319                         }
 320                         if (best[i] == -1
 321                             || devices[best[i]].events < devices[devcnt].events)
 322                                 best[i] = devcnt;
 323                 }
 324                 devcnt++;
 325         }
 326
 327         if (super)
 328                 free(super);
 329         super = NULL;
 330
 331         if (devcnt == 0) {
 332                 fprintf(stderr, Name ": no devices found for %s\n",
 333                         mddev);
 334                 free(first_super);
 335                 return 1;
 336         }
 337
 338         getinfo_super0(&info, first_super);
 339
 340         /* now we have some devices that might be suitable.
 341          * I wonder how many
 342          */
 343         okcnt = 0;
 344         sparecnt=0;
 345         for (i=0; i< bestcnt ;i++) {
 346                 int j = best[i];
 347                 int event_margin = !force;
 348                 if (j < 0) continue;
 349                 /* note: we ignore error flags in multipath arrays
 350                  * as they don't make sense
 351                  */
 352                 if (info.array.level != -4)
 353                         if (!(devices[j].state & (1<<MD_DISK_SYNC))) {
 354                                 if (!(devices[j].state & (1<<MD_DISK_FAULTY)))
 355                                         sparecnt++;
 356                                 continue;
 357                         }
 358                 if (devices[j].events+event_margin >=
 359                     devices[most_recent].events) {
 360                         devices[j].uptodate = 1;
 361                         if (i < info.array.raid_disks)
 362                                 okcnt++;
 363                         else
 364                                 sparecnt++;
 365                 }
 366         }
 367         while (force && !enough(info.array.level, info.array.raid_disks, okcnt)) {
 368                 /* Choose the newest best drive which is
 369                  * not up-to-date, update the superblock
 370                  * and add it.
 371                  */
 372                 int fd;
 373                 chosen_drive = -1;
 374                 for (i=0; i<info.array.raid_disks && i < bestcnt; i++) {
 375                         int j = best[i];
 376                         if (j>=0 &&
 377                             !devices[j].uptodate &&
 378                             devices[j].events > 0 &&
 379                             (chosen_drive < 0 ||
 380                              devices[j].events > devices[chosen_drive].events))
 381                                 chosen_drive = j;
 382                 }
 383                 if (chosen_drive < 0)
 384                         break;
 385                 fprintf(stderr, Name ": forcing event count in %s(%d) from %d upto %d\n",
 386                         devices[chosen_drive].devname, devices[chosen_drive].raid_disk,
 387                         (int)(devices[chosen_drive].events),
 388                         (int)(devices[most_recent].events));
 389                 fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
 390                 if (fd < 0) {
 391                         fprintf(stderr, Name ": Couldn't open %s for write - not updating\n",
 392                                 devices[chosen_drive].devname);
 393                         devices[chosen_drive].events = 0;
 394                         continue;
 395                 }
 396                 if (load_super0(fd, &super, NULL)) {
 397                         close(fd);
 398                         fprintf(stderr, Name ": RAID superblock disappeared from %s - not updating.\n",
 399                                 devices[chosen_drive].devname);
 400                         devices[chosen_drive].events = 0;
 401                         continue;
 402                 }
 403                 info.events = devices[most_recent].events;
 404                 update_super0(&info, super, "force", devices[chosen_drive].devname, verbose);
 405
 406                 if (store_super0(fd, super)) {
 407                         close(fd);
 408                         fprintf(stderr, Name ": Could not re-write superblock on %s\n",
 409                                 devices[chosen_drive].devname);
 410                         devices[chosen_drive].events = 0;
 411                         free(super);
 412                         continue;
 413                 }
 414                 close(fd);
 415                 devices[chosen_drive].events = devices[most_recent].events;
 416                 devices[chosen_drive].uptodate = 1;
 417                 okcnt++;
 418                 free(super);
 419         }
 420
 421         /* Now we want to look at the superblock which the kernel will base things on
 422          * and compare the devices that we think are working with the devices that the
 423          * superblock thinks are working.
 424          * If there are differences and --force is given, then update this chosen
 425          * superblock.
 426          */
 427         chosen_drive = -1;
 428         super = NULL;
 429         for (i=0; chosen_drive < 0 && i<bestcnt; i++) {
 430                 int j = best[i];
 431                 int fd;
 432
 433                 if (j<0)
 434                         continue;
 435                 if (!devices[j].uptodate)
 436                         continue;
 437                 chosen_drive = j;
 438                 if ((fd=open(devices[j].devname, O_RDONLY|O_EXCL))< 0) {
 439                         fprintf(stderr, Name ": Cannot open %s: %s\n",
 440                                 devices[j].devname, strerror(errno));
 441                         return 1;
 442                 }
 443                 if (load_super0(fd, &super, NULL)) {
 444                         close(fd);
 445                         fprintf(stderr, Name ": RAID superblock has disappeared from %s\n",
 446                                 devices[j].devname);
 447                         return 1;
 448                 }
 449                 close(fd);
 450         }
 451         if (super == NULL) {
 452                 fprintf(stderr, Name ": No suitable drives found for %s\n", mddev);
 453                 return 1;
 454         }
 455         getinfo_super0(&info, super);
 456         for (i=0; i<bestcnt; i++) {
 457                 int j = best[i];
 458                 unsigned int desired_state;
 459
 460                 if (i < info.array.raid_disks)
 461                         desired_state = (1<<MD_DISK_ACTIVE) | (1<<MD_DISK_SYNC);
 462                 else
 463                         desired_state = 0;
 464
 465                 if (j<0)
 466                         continue;
 467                 if (!devices[j].uptodate)
 468                         continue;
 469                 info.disk.number = i;
 470                 info.disk.state = desired_state;
 471
 472                 if (devices[j].uptodate &&
 473                     update_super0(&info, super, "assemble", NULL, 0)) {
 474                         if (force) {
 475                                 fprintf(stderr, Name ": "
 476                                         "clearing FAULTY flag for device %d in %s for %s\n",
 477                                         j, mddev, devices[j].devname);
 478                                 change = 1;
 479                         } else {
 480                                 fprintf(stderr, Name ": "
 481                                         "device %d in %s has wrong state in superblock, but %s seems ok\n",
 482                                         i, mddev, devices[j].devname);
 483                         }
 484                 }
 485 #if 0
 486                 if (!devices[j].uptodate &&
 487                     !(super.disks[i].state & (1 << MD_DISK_FAULTY))) {
 488                         fprintf(stderr, Name ": devices %d of %s is not marked FAULTY in superblock, but cannot be found\n",
 489                                 i, mddev);
 490                 }
 491 #endif
 492         }
 493         if (force && okcnt == info.array.raid_disks-1) {
 494                 /* FIXME check event count */
 495                 change += update_super0(&info, super, "force",
 496                                         devices[chosen_drive].devname, 0);
 497         }
 498
 499         if (change) {
 500                 int fd;
 501                 fd = open(devices[chosen_drive].devname, O_RDWR|O_EXCL);
 502                 if (fd < 0) {
 503                         fprintf(stderr, Name ": Could open %s for write - cannot Assemble array.\n",
 504                                 devices[chosen_drive].devname);
 505                         return 1;
 506                 }
 507                 if (store_super0(fd, super)) {
 508                         close(fd);
 509                         fprintf(stderr, Name ": Could not re-write superblock on %s\n",
 510                                 devices[chosen_drive].devname);
 511                         return 1;
 512                 }
 513                 close(fd);
 514         }
 515
 516         /* count number of in-sync devices according to the superblock.
 517          * We must have this number to start the array without -s or -R
 518          */
 519         req_cnt = info.array.working_disks;
 520
 521         /* Almost ready to actually *do* something */
 522         if (!old_linux) {
 523                 if (ioctl(mdfd, SET_ARRAY_INFO, NULL) != 0) {
 524                         fprintf(stderr, Name ": SET_ARRAY_INFO failed for %s: %s\n",
 525                                 mddev, strerror(errno));
 526                         return 1;
 527                 }
 528                 /* First, add the raid disks, but add the chosen one last */
 529                 for (i=0; i<= bestcnt; i++) {
 530                         int j;
 531                         if (i < bestcnt) {
 532                                 j = best[i];
 533                                 if (j == chosen_drive)
 534                                         continue;
 535                         } else
 536                                 j = chosen_drive;
 537
 538                         if (j >= 0 /* && devices[j].uptodate */) {
 539                                 mdu_disk_info_t disk;
 540                                 memset(&disk, 0, sizeof(disk));
 541                                 disk.major = devices[j].major;
 542                                 disk.minor = devices[j].minor;
 543                                 if (ioctl(mdfd, ADD_NEW_DISK, &disk)!=0) {
 544                                         fprintf(stderr, Name ": failed to add %s to %s: %s\n",
 545                                                 devices[j].devname,
 546                                                 mddev,
 547                                                 strerror(errno));
 548                                         if (i < info.array.raid_disks || i == bestcnt)
 549                                                 okcnt--;
 550                                         else
 551                                                 sparecnt--;
 552                                 } else if (verbose)
 553                                         fprintf(stderr, Name ": added %s to %s as %d\n",
 554                                                 devices[j].devname, mddev, devices[j].raid_disk);
 555                         } else if (verbose && i < info.array.raid_disks)
 556                                 fprintf(stderr, Name ": no uptodate device for slot %d of %s\n",
 557                                         i, mddev);
 558                 }
 559
 560                 if (runstop == 1 ||
 561                     (runstop == 0 &&
 562                      ( enough(info.array.level, info.array.raid_disks, okcnt) &&
 563                        (okcnt >= req_cnt || start_partial_ok)
 564                              ))) {
 565                         if (ioctl(mdfd, RUN_ARRAY, NULL)==0) {
 566                                 fprintf(stderr, Name ": %s has been started with %d drive%s",
 567                                         mddev, okcnt, okcnt==1?"":"s");
 568                                 if (okcnt < info.array.raid_disks)
 569                                         fprintf(stderr, " (out of %d)", info.array.raid_disks);
 570                                 if (sparecnt)
 571                                         fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
 572                                 fprintf(stderr, ".\n");
 573                                 return 0;
 574                         }
 575                         fprintf(stderr, Name ": failed to RUN_ARRAY %s: %s\n",
 576                                 mddev, strerror(errno));
 577                         return 1;
 578                 }
 579                 if (runstop == -1) {
 580                         fprintf(stderr, Name ": %s assembled from %d drive%s, but not started.\n",
 581                                 mddev, okcnt, okcnt==1?"":"s");
 582                         return 0;
 583                 }
 584                 fprintf(stderr, Name ": %s assembled from %d drive%s", mddev, okcnt, okcnt==1?"":"s");
 585                 if (sparecnt)
 586                         fprintf(stderr, " and %d spare%s", sparecnt, sparecnt==1?"":"s");
 587                 if (!enough(info.array.level, info.array.raid_disks, okcnt))
 588                         fprintf(stderr, " - not enough to start the array.\n");
 589                 else {
 590                         if (req_cnt == info.array.raid_disks)
 591                                 fprintf(stderr, " - need all %d to start it", req_cnt);
 592                         else
 593                                 fprintf(stderr, " - need %d of %d to start", req_cnt, info.array.raid_disks);
 594                         fprintf(stderr, " (use --run to insist).\n");
 595                 }
 596                 return 1;
 597         } else {
 598                 /* The "chosen_drive" is a good choice, and if necessary, the superblock has
 599                  * been updated to point to the current locations of devices.
 600                  * so we can just start the array
 601                  */
 602                 unsigned long dev;
 603                 dev = makedev(devices[chosen_drive].major,
 604                             devices[chosen_drive].minor);
 605                 if (ioctl(mdfd, START_ARRAY, dev)) {
 606                     fprintf(stderr, Name ": Cannot start array: %s\n",
 607                             strerror(errno));
 608                 }
 609
 610         }
 611         return 0;
 612 }