Incremental.c

   1 /*
   2  * Incremental.c - support --incremental.  Part of:
   3  * mdadm - manage Linux "md" devices aka RAID arrays.
   4  *
   5  * Copyright (C) 2006-2013 Neil Brown <neilb@suse.de>
   6  *
   7  *
   8  *    This program is free software; you can redistribute it and/or modify
   9  *    it under the terms of the GNU General Public License as published by
  10  *    the Free Software Foundation; either version 2 of the License, or
  11  *    (at your option) any later version.
  12  *
  13  *    This program is distributed in the hope that it will be useful,
  14  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  *    GNU General Public License for more details.
  17  *
  18  *    You should have received a copy of the GNU General Public License
  19  *    along with this program; if not, write to the Free Software
  20  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  21  *
  22  *    Author: Neil Brown
  23  *    Email: <neilb@suse.de>
  24  *    Paper: Neil Brown
  25  *           Novell Inc
  26  *           GPO Box Q1283
  27  *           QVB Post Office, NSW 1230
  28  *           Australia
  29  */
  30
  31 #include        "mdadm.h"
  32 #include        <sys/wait.h>
  33 #include        <dirent.h>
  34 #include        <ctype.h>
  35
  36 static int count_active(struct supertype *st, struct mdinfo *sra,
  37                         int mdfd, char **availp,
  38                         struct mdinfo *info);
  39 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
  40                         int number, __u64 events, int verbose,
  41                         char *array_name);
  42 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
  43                      struct map_ent *target,
  44                      struct supertype *st, int verbose);
  45
  46 static int Incremental_container(struct supertype *st, char *devname,
  47                                  struct context *c, char *only);
  48
  49 int Incremental(struct mddev_dev *devlist, struct context *c,
  50                 struct supertype *st)
  51 {
  52         /* Add this device to an array, creating the array if necessary
  53          * and starting the array if sensible or - if runstop>0 - if possible.
  54          *
  55          * This has several steps:
  56          *
  57          * 1/ Check if device is permitted by mdadm.conf, reject if not.
  58          * 2/ Find metadata, reject if none appropriate (check
  59          *       version/name from args)
  60          * 3/ Check if there is a match in mdadm.conf
  61          * 3a/ if not, check for homehost match.  If no match, assemble as
  62          *    a 'foreign' array.
  63          * 4/ Determine device number.
  64          * - If in mdadm.conf with std name, use that
  65          * - UUID in /var/run/mdadm.map  use that
  66          * - If name is suggestive, use that. unless in use with different uuid.
  67          * - Choose a free, high number.
  68          * - Use a partitioned device unless strong suggestion not to.
  69          *         e.g. auto=md
  70          *   Don't choose partitioned for containers.
  71          * 5/ Find out if array already exists
  72          * 5a/ if it does not
  73          * - choose a name, from mdadm.conf or 'name' field in array.
  74          * - create the array
  75          * - add the device
  76          * 5b/ if it does
  77          * - check one drive in array to make sure metadata is a reasonably
  78          *       close match.  Reject if not (e.g. different type)
  79          * - add the device
  80          * 6/ Make sure /var/run/mdadm.map contains this array.
  81          * 7/ Is there enough devices to possibly start the array?
  82          *     For a container, this means running Incremental_container.
  83          * 7a/ if not, finish with success.
  84          * 7b/ if yes,
  85          * - read all metadata and arrange devices like -A does
  86          * - if number of OK devices match expected, or -R and there are enough,
  87          *   start the array (auto-readonly).
  88          */
  89         dev_t rdev, rdev2;
  90         struct mdinfo info, dinfo;
  91         struct mdinfo *sra = NULL, *d;
  92         struct mddev_ident *match;
  93         char chosen_name[1024];
  94         char *md_devname;
  95         int rv = 1;
  96         struct map_ent *mp, *map = NULL;
  97         int dfd = -1, mdfd = -1;
  98         char *avail = NULL;
  99         int active_disks;
 100         int trustworthy;
 101         char *name_to_use;
 102         struct dev_policy *policy = NULL;
 103         struct map_ent target_array;
 104         int have_target;
 105         char *devname = devlist->devname;
 106         int journal_device_missing = 0;
 107
 108         struct createinfo *ci = conf_get_create_info();
 109
 110         if (!stat_is_blkdev(devname, &rdev))
 111                 return rv;
 112         dfd = dev_open(devname, O_RDONLY);
 113         if (dfd < 0) {
 114                 if (c->verbose >= 0)
 115                         pr_err("cannot open %s: %s.\n",
 116                                 devname, strerror(errno));
 117                 return rv;
 118         }
 119         /* If the device is a container, we do something very different */
 120         if (must_be_container(dfd)) {
 121                 if (!st)
 122                         st = super_by_fd(dfd, NULL);
 123                 if (st && st->ss->load_container)
 124                         rv = st->ss->load_container(st, dfd, NULL);
 125
 126                 close(dfd);
 127                 if (!rv && st->ss->container_content) {
 128                         if (map_lock(&map))
 129                                 pr_err("failed to get exclusive lock on mapfile\n");
 130                         if (c->export)
 131                                 printf("MD_DEVNAME=%s\n", devname);
 132                         rv = Incremental_container(st, devname, c, NULL);
 133                         map_unlock(&map);
 134                         return rv;
 135                 }
 136
 137                 pr_err("%s is not part of an md array.\n",
 138                         devname);
 139                 return rv;
 140         }
 141
 142         /* 1/ Check if device is permitted by mdadm.conf */
 143
 144         for (;devlist; devlist = devlist->next)
 145                 if (conf_test_dev(devlist->devname))
 146                         break;
 147         if (!devlist) {
 148                 devlist = conf_get_devs();
 149                 for (;devlist; devlist = devlist->next) {
 150                         if (stat_is_blkdev(devlist->devname, &rdev2) &&
 151                             rdev2 == rdev)
 152                                 break;
 153                 }
 154         }
 155         if (!devlist) {
 156                 if (c->verbose >= 0)
 157                         pr_err("%s not permitted by mdadm.conf.\n",
 158                                devname);
 159                 goto out;
 160         }
 161
 162         /* 2/ Find metadata, reject if none appropriate (check
 163          *            version/name from args) */
 164
 165         if (!fstat_is_blkdev(dfd, devname, &rdev))
 166                 goto out;
 167
 168         dinfo.disk.major = major(rdev);
 169         dinfo.disk.minor = minor(rdev);
 170
 171         policy = disk_policy(&dinfo);
 172         have_target = policy_check_path(&dinfo, &target_array);
 173
 174         if (st == NULL && (st = guess_super_type(dfd, guess_array)) == NULL) {
 175                 if (c->verbose >= 0)
 176                         pr_err("no recognisable superblock on %s.\n",
 177                                devname);
 178                 rv = try_spare(devname, &dfd, policy,
 179                                have_target ? &target_array : NULL,
 180                                NULL, c->verbose);
 181                 goto out;
 182         }
 183         st->ignore_hw_compat = 0;
 184
 185         if (st->ss->compare_super == NULL ||
 186             st->ss->load_super(st, dfd, c->verbose >= 0 ? devname : NULL)) {
 187                 if (c->verbose >= 0)
 188                         pr_err("no RAID superblock on %s.\n",
 189                                 devname);
 190                 rv = try_spare(devname, &dfd, policy,
 191                                have_target ? &target_array : NULL,
 192                                st, c->verbose);
 193                 free(st);
 194                 goto out;
 195         }
 196         close (dfd); dfd = -1;
 197
 198         st->ss->getinfo_super(st, &info, NULL);
 199
 200         /* 3/ Check if there is a match in mdadm.conf */
 201         match = conf_match(st, &info, devname, c->verbose, &rv);
 202         if (!match && rv == 2)
 203                 goto out;
 204
 205         if (match && match->devname &&
 206             strcasecmp(match->devname, "<ignore>") == 0) {
 207                 if (c->verbose >= 0)
 208                         pr_err("array containing %s is explicitly ignored by mdadm.conf\n",
 209                                 devname);
 210                 goto out;
 211         }
 212
 213         /* 3a/ if not, check for homehost match.  If no match, continue
 214          * but don't trust the 'name' in the array. Thus a 'random' minor
 215          * number will be assigned, and the device name will be based
 216          * on that. */
 217         if (match)
 218                 trustworthy = LOCAL;
 219         else if (st->ss->match_home(st, c->homehost) == 1)
 220                 trustworthy = LOCAL;
 221         else if (st->ss->match_home(st, "any") == 1)
 222                 trustworthy = LOCAL_ANY;
 223         else
 224                 trustworthy = FOREIGN;
 225
 226         if (!match && !conf_test_metadata(st->ss->name, policy,
 227                                           (trustworthy == LOCAL))) {
 228                 if (c->verbose >= 1)
 229                         pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
 230                                devname, st->ss->name);
 231                 goto out;
 232         }
 233         if (trustworthy == LOCAL_ANY)
 234                 trustworthy = LOCAL;
 235
 236         /* There are three possible sources for 'autof':  command line,
 237          * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
 238          * ARRAY takes precedence, then command line, then
 239          * CREATE.
 240          */
 241         if (match && match->autof)
 242                 c->autof = match->autof;
 243         if (c->autof == 0)
 244                 c->autof = ci->autof;
 245
 246         name_to_use = info.name;
 247         if (name_to_use[0] == 0 && info.array.level == LEVEL_CONTAINER) {
 248                 name_to_use = info.text_version;
 249                 trustworthy = METADATA;
 250         }
 251         if (name_to_use[0] && trustworthy != LOCAL &&
 252             ! c->require_homehost &&
 253             conf_name_is_free(name_to_use))
 254                 trustworthy = LOCAL;
 255
 256         /* strip "hostname:" prefix from name if we have decided
 257          * to treat it as LOCAL
 258          */
 259         if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
 260                 name_to_use = strchr(name_to_use, ':')+1;
 261
 262         /* 4/ Check if array exists.
 263          */
 264         if (map_lock(&map))
 265                 pr_err("failed to get exclusive lock on mapfile\n");
 266         /* Now check we can get O_EXCL.  If not, probably "mdadm -A" has
 267          * taken over
 268          */
 269         dfd = dev_open(devname, O_RDONLY|O_EXCL);
 270         if (dfd < 0) {
 271                 if (c->verbose >= 0)
 272                         pr_err("cannot reopen %s: %s.\n",
 273                                 devname, strerror(errno));
 274                 goto out_unlock;
 275         }
 276         /* Cannot hold it open while we add the device to the array,
 277          * so we must release the O_EXCL and depend on the map_lock()
 278          * So now is the best time to remove any partitions.
 279          */
 280         remove_partitions(dfd);
 281         close(dfd);
 282         dfd = -1;
 283
 284         mp = map_by_uuid(&map, info.uuid);
 285         if (mp)
 286                 mdfd = open_dev(mp->devnm);
 287         else
 288                 mdfd = -1;
 289
 290         if (mdfd < 0) {
 291
 292                 /* Skip the clustered ones. This should be started by
 293                  * clustering resource agents
 294                  */
 295                 if (info.array.state & (1 << MD_SB_CLUSTERED))
 296                         goto out;
 297
 298                 /* Couldn't find an existing array, maybe make a new one */
 299                 mdfd = create_mddev(match ? match->devname : NULL,
 300                                     name_to_use, c->autof, trustworthy, chosen_name, 0);
 301
 302                 if (mdfd < 0)
 303                         goto out_unlock;
 304
 305                 if (sysfs_init(&info, mdfd, NULL)) {
 306                         pr_err("unable to initialize sysfs for %s\n",
 307                                chosen_name);
 308                         rv = 2;
 309                         goto out_unlock;
 310                 }
 311
 312                 if (set_array_info(mdfd, st, &info) != 0) {
 313                         pr_err("failed to set array info for %s: %s\n",
 314                                 chosen_name, strerror(errno));
 315                         rv = 2;
 316                         goto out_unlock;
 317                 }
 318
 319                 dinfo = info;
 320                 dinfo.disk.major = major(rdev);
 321                 dinfo.disk.minor = minor(rdev);
 322                 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
 323                         pr_err("failed to add %s to new array %s: %s.\n",
 324                                 devname, chosen_name, strerror(errno));
 325                         ioctl(mdfd, STOP_ARRAY, 0);
 326                         rv = 2;
 327                         goto out_unlock;
 328                 }
 329                 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
 330                                               GET_OFFSET | GET_SIZE));
 331
 332                 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
 333                         /* It really should be 'none' - must be old buggy
 334                          * kernel, and mdadm -I may not be able to complete.
 335                          * So reject it.
 336                          */
 337                         ioctl(mdfd, STOP_ARRAY, NULL);
 338                         pr_err("You have an old buggy kernel which cannot support\n      --incremental reliably.  Aborting.\n");
 339                         rv = 2;
 340                         goto out_unlock;
 341                 }
 342                 info.array.working_disks = 1;
 343                 /* 6/ Make sure /var/run/mdadm.map contains this array. */
 344                 map_update(&map, fd2devnm(mdfd),
 345                            info.text_version,
 346                            info.uuid, chosen_name);
 347         } else {
 348         /* 5b/ if it does */
 349         /* - check one drive in array to make sure metadata is a reasonably */
 350         /*        close match.  Reject if not (e.g. different type) */
 351         /* - add the device */
 352                 char dn[20];
 353                 int dfd2;
 354                 int err;
 355                 struct supertype *st2;
 356                 struct mdinfo info2, *d;
 357
 358                 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
 359                                             GET_OFFSET | GET_SIZE));
 360
 361                 if (mp->path)
 362                         strcpy(chosen_name, mp->path);
 363                 else
 364                         strcpy(chosen_name, mp->devnm);
 365
 366                 /* It is generally not OK to add non-spare drives to a
 367                  * running array as they are probably missing because
 368                  * they failed.  However if runstop is 1, then the
 369                  * array was possibly started early and our best bet is
 370                  * to add this anyway.
 371                  * Also if action policy is re-add or better we allow
 372                  * re-add.
 373                  * This doesn't apply to containers as the 'non-spare'
 374                  * flag has a different meaning.  The test has to happen
 375                  * at the device level there
 376                  */
 377                 if (!st->ss->external &&
 378                     (info.disk.state & (1 << MD_DISK_SYNC)) != 0 &&
 379                     !policy_action_allows(policy, st->ss->name, act_re_add) &&
 380                     c->runstop < 1) {
 381                         if (md_array_active(mdfd)) {
 382                                 pr_err("not adding %s to active array (without --run) %s\n",
 383                                        devname, chosen_name);
 384                                 rv = 2;
 385                                 goto out_unlock;
 386                         }
 387                 }
 388                 if (!sra) {
 389                         rv = 2;
 390                         goto out_unlock;
 391                 }
 392                 if (sra->devs) {
 393                         sprintf(dn, "%d:%d", sra->devs->disk.major,
 394                                 sra->devs->disk.minor);
 395                         dfd2 = dev_open(dn, O_RDONLY);
 396                         if (dfd2 < 0) {
 397                                 pr_err("unable to open %s\n", devname);
 398                                 rv = 2;
 399                                 goto out_unlock;
 400                         }
 401                         st2 = dup_super(st);
 402                         if (st2->ss->load_super(st2, dfd2, NULL) ||
 403                             st->ss->compare_super(st, st2) != 0) {
 404                                 pr_err("metadata mismatch between %s and chosen array %s\n",
 405                                        devname, chosen_name);
 406                                 close(dfd2);
 407                                 rv = 2;
 408                                 goto out_unlock;
 409                         }
 410                         close(dfd2);
 411                         st2->ss->getinfo_super(st2, &info2, NULL);
 412                         st2->ss->free_super(st2);
 413                         if (info.array.level != info2.array.level ||
 414                             memcmp(info.uuid, info2.uuid, 16) != 0 ||
 415                             info.array.raid_disks != info2.array.raid_disks) {
 416                                 pr_err("unexpected difference between %s and %s.\n",
 417                                        chosen_name, devname);
 418                                 rv = 2;
 419                                 goto out_unlock;
 420                         }
 421                 }
 422                 info.disk.major = major(rdev);
 423                 info.disk.minor = minor(rdev);
 424                 /* add disk needs to know about containers */
 425                 if (st->ss->external)
 426                         sra->array.level = LEVEL_CONTAINER;
 427
 428                 if (info.array.state & (1 << MD_SB_CLUSTERED))
 429                         info.disk.state |= (1 << MD_DISK_CLUSTER_ADD);
 430
 431                 err = add_disk(mdfd, st, sra, &info);
 432                 if (err < 0 && errno == EBUSY) {
 433                         /* could be another device present with the same
 434                          * disk.number. Find and reject any such
 435                          */
 436                         find_reject(mdfd, st, sra, info.disk.number,
 437                                     info.events, c->verbose, chosen_name);
 438                         err = add_disk(mdfd, st, sra, &info);
 439                 }
 440                 if (err < 0 && errno == EINVAL &&
 441                     info.disk.state & (1<<MD_DISK_SYNC)) {
 442                         /* Maybe it needs to be added as a spare */
 443                         if (policy_action_allows(policy, st->ss->name,
 444                                                  act_force_spare)) {
 445                                 info.disk.state &= ~(1<<MD_DISK_SYNC);
 446                                 err = add_disk(mdfd, st, sra, &info);
 447                         } else
 448                                 if (c->verbose >= 0)
 449                                         pr_err("can only add %s to %s as a spare, and force-spare is not set.\n",
 450                                                devname, chosen_name);
 451                 }
 452                 if (err < 0) {
 453                         pr_err("failed to add %s to existing array %s: %s.\n",
 454                                 devname, chosen_name, strerror(errno));
 455                         rv = 2;
 456                         goto out_unlock;
 457                 }
 458                 info.array.working_disks = 0;
 459                 for (d = sra->devs; d; d=d->next)
 460                         info.array.working_disks ++;
 461
 462         }
 463         if (strncmp(chosen_name, "/dev/md/", 8) == 0)
 464                 md_devname = chosen_name+8;
 465         else
 466                 md_devname = chosen_name;
 467         if (c->export) {
 468                 printf("MD_DEVICE=%s\n", fd2devnm(mdfd));
 469                 printf("MD_DEVNAME=%s\n", md_devname);
 470                 printf("MD_FOREIGN=%s\n", trustworthy == FOREIGN ? "yes" : "no");
 471         }
 472
 473         /* 7/ Is there enough devices to possibly start the array? */
 474         /* 7a/ if not, finish with success. */
 475         if (info.array.level == LEVEL_CONTAINER) {
 476                 char devnm[32];
 477                 /* Try to assemble within the container */
 478                 sysfs_uevent(sra, "change");
 479                 if (!c->export && c->verbose >= 0)
 480                         pr_err("container %s now has %d device%s\n",
 481                                chosen_name, info.array.working_disks,
 482                                info.array.working_disks == 1?"":"s");
 483                 wait_for(chosen_name, mdfd);
 484                 if (st->ss->external)
 485                         strcpy(devnm, fd2devnm(mdfd));
 486                 if (st->ss->load_container)
 487                         rv = st->ss->load_container(st, mdfd, NULL);
 488                 close(mdfd);
 489                 sysfs_free(sra);
 490                 if (!rv)
 491                         rv = Incremental_container(st, chosen_name, c, NULL);
 492                 map_unlock(&map);
 493                 /* after spare is added, ping monitor for external metadata
 494                  * so that it can eg. try to rebuild degraded array */
 495                 if (st->ss->external)
 496                         ping_monitor(devnm);
 497                 return rv;
 498         }
 499
 500         /* We have added something to the array, so need to re-read the
 501          * state.  Eventually this state should be kept up-to-date as
 502          * things change.
 503          */
 504         sysfs_free(sra);
 505         sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
 506                                     GET_OFFSET | GET_SIZE));
 507         active_disks = count_active(st, sra, mdfd, &avail, &info);
 508
 509         journal_device_missing = (info.journal_device_required) && (info.journal_clean == 0);
 510
 511         if (info.consistency_policy == CONSISTENCY_POLICY_PPL)
 512                 info.array.state |= 1;
 513
 514         if (enough(info.array.level, info.array.raid_disks,
 515                    info.array.layout, info.array.state & 1, avail) == 0) {
 516                 if (c->export) {
 517                         printf("MD_STARTED=no\n");
 518                 } else if (c->verbose >= 0)
 519                         pr_err("%s attached to %s, not enough to start (%d).\n",
 520                                devname, chosen_name, active_disks);
 521                 rv = 0;
 522                 goto out_unlock;
 523         }
 524
 525         /* 7b/ if yes, */
 526         /* - if number of OK devices match expected, or -R and there */
 527         /*             are enough, */
 528         /*   + add any bitmap file  */
 529         /*   + start the array (auto-readonly). */
 530
 531         if (md_array_active(mdfd)) {
 532                 if (c->export) {
 533                         printf("MD_STARTED=already\n");
 534                 } else if (c->verbose >= 0)
 535                         pr_err("%s attached to %s which is already active.\n",
 536                                devname, chosen_name);
 537                 rv = 0;
 538                 goto out_unlock;
 539         }
 540
 541         map_unlock(&map);
 542         if (c->runstop > 0 || (!journal_device_missing && active_disks >= info.array.working_disks)) {
 543                 struct mdinfo *dsk;
 544                 /* Let's try to start it */
 545
 546                 if (journal_device_missing)
 547                         pr_err("Trying to run with missing journal device\n");
 548                 if (info.reshape_active && !(info.reshape_active & RESHAPE_NO_BACKUP)) {
 549                         pr_err("%s: This array is being reshaped and cannot be started\n",
 550                                chosen_name);
 551                         cont_err("by --incremental.  Please use --assemble\n");
 552                         goto out;
 553                 }
 554                 if (match && match->bitmap_file) {
 555                         int bmfd = open(match->bitmap_file, O_RDWR);
 556                         if (bmfd < 0) {
 557                                 pr_err("Could not open bitmap file %s.\n",
 558                                        match->bitmap_file);
 559                                 goto out;
 560                         }
 561                         if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
 562                                 close(bmfd);
 563                                 pr_err("Failed to set bitmapfile for %s.\n",
 564                                        chosen_name);
 565                                 goto out;
 566                         }
 567                         close(bmfd);
 568                 }
 569                 /* Need to remove from the array any devices which
 570                  * 'count_active' discerned were too old or inappropriate
 571                  */
 572                 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
 573                         if (d->disk.state & (1<<MD_DISK_REMOVED))
 574                                 remove_disk(mdfd, st, sra, d);
 575
 576                 if ((sra == NULL || active_disks >= info.array.working_disks) &&
 577                     trustworthy != FOREIGN)
 578                         rv = ioctl(mdfd, RUN_ARRAY, NULL);
 579                 else
 580                         rv = sysfs_set_str(sra, NULL,
 581                                            "array_state", "read-auto");
 582                 /* Array might be O_EXCL which  will interfere with
 583                  * fsck and mount.  So re-open without O_EXCL.
 584                  */
 585                 reopen_mddev(mdfd);
 586                 if (rv == 0) {
 587                         if (c->export) {
 588                                 printf("MD_STARTED=yes\n");
 589                         } else if (c->verbose >= 0)
 590                                 pr_err("%s attached to %s, which has been started.\n",
 591                                        devname, chosen_name);
 592                         rv = 0;
 593                         wait_for(chosen_name, mdfd);
 594                         /* We just started the array, so some devices
 595                          * might have been evicted from the array
 596                          * because their event counts were too old.
 597                          * If the action=re-add policy is in-force for
 598                          * those devices we should re-add them now.
 599                          */
 600                         for (dsk = sra->devs; dsk ; dsk = dsk->next) {
 601                                 if (disk_action_allows(dsk, st->ss->name,
 602                                                        act_re_add) &&
 603                                     add_disk(mdfd, st, sra, dsk) == 0)
 604                                         pr_err("%s re-added to %s\n",
 605                                                dsk->sys_name, chosen_name);
 606                         }
 607                 } else {
 608                         pr_err("%s attached to %s, but failed to start: %s.\n",
 609                                devname, chosen_name, strerror(errno));
 610                         rv = 1;
 611                 }
 612         } else {
 613                 if (c->export) {
 614                         printf("MD_STARTED=unsafe\n");
 615                 } else if (journal_device_missing) {
 616                         pr_err("Journal device is missing, not safe to start yet.\n");
 617                 } else if (c->verbose >= 0)
 618                         pr_err("%s attached to %s, not enough to start safely.\n",
 619                                devname, chosen_name);
 620                 rv = 0;
 621         }
 622 out:
 623         free(avail);
 624         if (dfd >= 0)
 625                 close(dfd);
 626         if (mdfd >= 0)
 627                 close(mdfd);
 628         if (policy)
 629                 dev_policy_free(policy);
 630         sysfs_free(sra);
 631         return rv;
 632 out_unlock:
 633         map_unlock(&map);
 634         goto out;
 635 }
 636
 637 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
 638                         int number, __u64 events, int verbose,
 639                         char *array_name)
 640 {
 641         /* Find a device attached to this array with a disk.number of number
 642          * and events less than the passed events, and remove the device.
 643          */
 644         struct mdinfo *d;
 645
 646         if (md_array_active(mdfd))
 647                 return; /* not safe to remove from active arrays
 648                          * without thinking more */
 649
 650         for (d = sra->devs; d ; d = d->next) {
 651                 char dn[24]; // 2*11 bytes for ints (including sign) + colon + null byte
 652                 int dfd;
 653                 struct mdinfo info;
 654                 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
 655                 dfd = dev_open(dn, O_RDONLY);
 656                 if (dfd < 0)
 657                         continue;
 658                 if (st->ss->load_super(st, dfd, NULL)) {
 659                         close(dfd);
 660                         continue;
 661                 }
 662                 st->ss->getinfo_super(st, &info, NULL);
 663                 st->ss->free_super(st);
 664                 close(dfd);
 665
 666                 if (info.disk.number != number || info.events >= events)
 667                         continue;
 668
 669                 if (d->disk.raid_disk > -1)
 670                         sysfs_set_str(sra, d, "slot", "none");
 671                 if (sysfs_set_str(sra, d, "state", "remove") == 0)
 672                         if (verbose >= 0)
 673                                 pr_err("removing old device %s from %s\n",
 674                                        d->sys_name+4, array_name);
 675         }
 676 }
 677
 678 static int count_active(struct supertype *st, struct mdinfo *sra,
 679                         int mdfd, char **availp,
 680                         struct mdinfo *bestinfo)
 681 {
 682         /* count how many devices in sra think they are active */
 683         struct mdinfo *d;
 684         int cnt = 0;
 685         int replcnt = 0;
 686         __u64 max_events = 0;
 687         char *avail = NULL;
 688         int *best = NULL;
 689         char *devmap = NULL;
 690         int numdevs = 0;
 691         int devnum;
 692         int b, i;
 693         int raid_disks = 0;
 694
 695         if (!sra)
 696                 return 0;
 697
 698         for (d = sra->devs ; d ; d = d->next)
 699                 numdevs++;
 700         for (d = sra->devs, devnum = 0 ; d ; d = d->next, devnum++) {
 701                 char dn[30];
 702                 int dfd;
 703                 int ok;
 704                 struct mdinfo info;
 705
 706                 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
 707                 dfd = dev_open(dn, O_RDONLY);
 708                 if (dfd < 0)
 709                         continue;
 710                 ok =  st->ss->load_super(st, dfd, NULL);
 711                 close(dfd);
 712                 if (ok != 0)
 713                         continue;
 714
 715                 info.array.raid_disks = raid_disks;
 716                 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
 717                 if (info.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
 718                         bestinfo->journal_clean = 1;
 719                 if (!avail) {
 720                         raid_disks = info.array.raid_disks;
 721                         avail = xcalloc(raid_disks, 1);
 722                         *availp = avail;
 723
 724                         best = xcalloc(raid_disks, sizeof(int));
 725                         devmap = xcalloc(raid_disks, numdevs);
 726
 727                         st->ss->getinfo_super(st, &info, devmap);
 728                 }
 729
 730                 if (info.disk.state & (1<<MD_DISK_SYNC))
 731                 {
 732                         if (cnt == 0) {
 733                                 cnt++;
 734                                 max_events = info.events;
 735                                 avail[info.disk.raid_disk] = 2;
 736                                 best[info.disk.raid_disk] = devnum;
 737                                 st->ss->getinfo_super(st, bestinfo, NULL);
 738                         } else if (info.events == max_events) {
 739                                 avail[info.disk.raid_disk] = 2;
 740                                 best[info.disk.raid_disk] = devnum;
 741                         } else if (info.events == max_events-1) {
 742                                 if (avail[info.disk.raid_disk] == 0) {
 743                                         avail[info.disk.raid_disk] = 1;
 744                                         best[info.disk.raid_disk] = devnum;
 745                                 }
 746                         } else if (info.events < max_events - 1)
 747                                 ;
 748                         else if (info.events == max_events+1) {
 749                                 int i;
 750                                 max_events = info.events;
 751                                 for (i = 0; i < raid_disks; i++)
 752                                         if (avail[i])
 753                                                 avail[i]--;
 754                                 avail[info.disk.raid_disk] = 2;
 755                                 best[info.disk.raid_disk] = devnum;
 756                                 st->ss->getinfo_super(st, bestinfo, NULL);
 757                         } else { /* info.events much bigger */
 758                                 memset(avail, 0, raid_disks);
 759                                 max_events = info.events;
 760                                 avail[info.disk.raid_disk] = 2;
 761                                 best[info.disk.raid_disk] = devnum;
 762                                 st->ss->getinfo_super(st, bestinfo, NULL);
 763                         }
 764                 } else if (info.disk.state & (1<<MD_DISK_REPLACEMENT))
 765                         replcnt++;
 766                 st->ss->free_super(st);
 767         }
 768
 769         if (!avail)
 770                 return 0;
 771         /* We need to reject any device that thinks the best device is
 772          * failed or missing */
 773         for (b = 0; b < raid_disks; b++)
 774                 if (avail[b] == 2)
 775                         break;
 776         cnt = 0;
 777         for (i = 0 ; i < raid_disks ; i++) {
 778                 if (i != b && avail[i])
 779                         if (devmap[raid_disks * best[i] + b] == 0) {
 780                                 /* This device thinks 'b' is failed -
 781                                  * don't use it */
 782                                 devnum = best[i];
 783                                 for (d=sra->devs ; devnum; d = d->next)
 784                                         devnum--;
 785                                 d->disk.state |= (1 << MD_DISK_REMOVED);
 786                                 avail[i] = 0;
 787                         }
 788                 if (avail[i])
 789                         cnt++;
 790         }
 791         /* Also need to reject any spare device with an event count that
 792          * is too high
 793          */
 794         for (d = sra->devs; d; d = d->next) {
 795                 if (!(d->disk.state & (1<<MD_DISK_SYNC)) &&
 796                     d->events > max_events)
 797                         d->disk.state |= (1 << MD_DISK_REMOVED);
 798         }
 799         free(best);
 800         free(devmap);
 801         return cnt + replcnt;
 802 }
 803
 804 /* test if container has degraded member(s) */
 805 static int
 806 container_members_max_degradation(struct map_ent *map, struct map_ent *me)
 807 {
 808         struct mdinfo *sra;
 809         int degraded, max_degraded = 0;
 810
 811         for(; map; map = map->next) {
 812                 if (!metadata_container_matches(map->metadata, me->devnm))
 813                         continue;
 814                 /* most accurate information regarding array degradation */
 815                 sra = sysfs_read(-1, map->devnm,
 816                                  GET_DISKS | GET_DEVS | GET_STATE);
 817                 if (!sra)
 818                         continue;
 819                 degraded = sra->array.raid_disks - sra->array.active_disks -
 820                         sra->array.spare_disks;
 821                 if (degraded > max_degraded)
 822                         max_degraded = degraded;
 823                 sysfs_free(sra);
 824         }
 825
 826         return max_degraded;
 827 }
 828
 829 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
 830                            struct map_ent *target, int bare,
 831                            struct supertype *st, int verbose)
 832 {
 833         /* This device doesn't have any md metadata
 834          * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
 835          * If 'st' is not set, then we only know that some metadata allows this,
 836          * others possibly don't.
 837          * So look for a container or array to attach the device to.
 838          * Prefer 'target' if that is set and the array is found.
 839          *
 840          * If st is set, then only arrays of that type are considered
 841          * Return 0 on success, or some exit code on failure, probably 1.
 842          */
 843         int rv = 1;
 844         dev_t rdev;
 845         struct map_ent *mp, *map = NULL;
 846         struct mdinfo *chosen = NULL;
 847         int dfd = *dfdp;
 848
 849         if (!fstat_is_blkdev(dfd, devname, &rdev))
 850                 return 1;
 851
 852         /*
 853          * Now we need to find a suitable array to add this to.
 854          * We only accept arrays that:
 855          *  - match 'st'
 856          *  - are in the same domains as the device
 857          *  - are of an size for which the device will be useful
 858          * and we choose the one that is the most degraded
 859          */
 860
 861         if (map_lock(&map)) {
 862                 pr_err("failed to get exclusive lock on mapfile\n");
 863                 return 1;
 864         }
 865         for (mp = map ; mp ; mp = mp->next) {
 866                 struct supertype *st2;
 867                 struct domainlist *dl = NULL;
 868                 struct mdinfo *sra;
 869                 unsigned long long devsize;
 870                 unsigned long long component_size = 0;
 871
 872                 if (is_subarray(mp->metadata))
 873                         continue;
 874                 if (st) {
 875                         st2 = st->ss->match_metadata_desc(mp->metadata);
 876                         if (!st2 ||
 877                             (st->minor_version >= 0 &&
 878                              st->minor_version != st2->minor_version)) {
 879                                 if (verbose > 1)
 880                                         pr_err("not adding %s to %s as metadata type doesn't match\n",
 881                                                 devname, mp->path);
 882                                 free(st2);
 883                                 continue;
 884                         }
 885                         free(st2);
 886                 }
 887                 sra = sysfs_read(-1, mp->devnm,
 888                                  GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
 889                                  GET_DEGRADED|GET_COMPONENT|GET_VERSION);
 890                 if (!sra) {
 891                         /* Probably a container - no degraded info */
 892                         sra = sysfs_read(-1, mp->devnm,
 893                                          GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
 894                                          GET_COMPONENT|GET_VERSION);
 895                         if (sra)
 896                                 sra->array.failed_disks = -1;
 897                 }
 898                 if (!sra)
 899                         continue;
 900                 if (st == NULL) {
 901                         int i;
 902                         st2 = NULL;
 903                         for(i = 0; !st2 && superlist[i]; i++)
 904                                 st2 = superlist[i]->match_metadata_desc(
 905                                         sra->text_version);
 906                         if (!st2) {
 907                                 if (verbose > 1)
 908                                         pr_err("not adding %s to %s as metadata not recognised.\n",
 909                                                 devname, mp->path);
 910                                 goto next;
 911                         }
 912                         /* Need to double check the 'act_spare' permissions applies
 913                          * to this metadata.
 914                          */
 915                         if (!policy_action_allows(pol, st2->ss->name, act_spare))
 916                                 goto next;
 917                         if (!bare && !policy_action_allows(pol, st2->ss->name,
 918                                                            act_spare_same_slot))
 919                                 goto next;
 920                 } else
 921                         st2 = st;
 922                 /* update number of failed disks for mostly degraded
 923                  * container member */
 924                 if (sra->array.failed_disks == -1)
 925                         sra->array.failed_disks = container_members_max_degradation(map, mp);
 926
 927                 get_dev_size(dfd, NULL, &devsize);
 928                 if (sra->component_size == 0) {
 929                         /* true for containers, here we must read superblock
 930                          * to obtain minimum spare size */
 931                         struct supertype *st3 = dup_super(st2);
 932                         int mdfd = open_dev(mp->devnm);
 933                         if (mdfd < 0) {
 934                                 free(st3);
 935                                 goto next;
 936                         }
 937                         if (st3->ss->load_container &&
 938                             !st3->ss->load_container(st3, mdfd, mp->path)) {
 939                                 component_size = st3->ss->min_acceptable_spare_size(st3);
 940                                 st3->ss->free_super(st3);
 941                         }
 942                         free(st3);
 943                         close(mdfd);
 944                 }
 945                 if ((sra->component_size > 0 &&
 946                      st2->ss->avail_size(st2, devsize,
 947                                          sra->devs ? sra->devs->data_offset :
 948                                          INVALID_SECTORS) <
 949                      sra->component_size) ||
 950                     (sra->component_size == 0 && devsize < component_size)) {
 951                         if (verbose > 1)
 952                                 pr_err("not adding %s to %s as it is too small\n",
 953                                         devname, mp->path);
 954                         goto next;
 955                 }
 956                 /* test against target.
 957                  * If 'target' is set and 'bare' is false, we only accept
 958                  * arrays/containers that match 'target'.
 959                  * If 'target' is set and 'bare' is true, we prefer the
 960                  * array which matches 'target'.
 961                  * target is considered only if we deal with degraded array
 962                  */
 963                 if (target && policy_action_allows(pol, st2->ss->name,
 964                                                    act_spare_same_slot)) {
 965                         if (strcmp(target->metadata, mp->metadata) == 0 &&
 966                             memcmp(target->uuid, mp->uuid,
 967                                    sizeof(target->uuid)) == 0 &&
 968                             sra->array.failed_disks > 0) {
 969                                 /* This is our target!! */
 970                                 sysfs_free(chosen);
 971                                 chosen = sra;
 972                                 sra = NULL;
 973                                 /* skip to end so we don't check any more */
 974                                 while (mp->next)
 975                                         mp = mp->next;
 976                                 goto next;
 977                         }
 978                         /* not our target */
 979                         if (!bare)
 980                                 goto next;
 981                 }
 982
 983                 dl = domain_from_array(sra, st2->ss->name);
 984                 if (domain_test(dl, pol, st2->ss->name) != 1) {
 985                         /* domain test fails */
 986                         if (verbose > 1)
 987                                 pr_err("not adding %s to %s as it is not in a compatible domain\n",
 988                                         devname, mp->path);
 989
 990                         goto next;
 991                 }
 992                 /* all tests passed, OK to add to this array */
 993                 if (!chosen) {
 994                         chosen = sra;
 995                         sra = NULL;
 996                 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
 997                         sysfs_free(chosen);
 998                         chosen = sra;
 999                         sra = NULL;
1000                 }
1001         next:
1002                 sysfs_free(sra);
1003                 if (st != st2)
1004                         free(st2);
1005                 if (dl)
1006                         domain_free(dl);
1007         }
1008         if (chosen) {
1009                 /* add current device to chosen array as a spare */
1010                 int mdfd = open_dev(chosen->sys_name);
1011                 if (mdfd >= 0) {
1012                         struct mddev_dev devlist;
1013                         char chosen_devname[24]; // 2*11 for int (including signs) + colon + null
1014                         devlist.next = NULL;
1015                         devlist.used = 0;
1016                         devlist.writemostly = FlagDefault;
1017                         devlist.failfast = FlagDefault;
1018                         devlist.devname = chosen_devname;
1019                         sprintf(chosen_devname, "%d:%d", major(rdev),
1020                                 minor(rdev));
1021                         devlist.disposition = 'a';
1022                         close(dfd);
1023                         *dfdp = -1;
1024                         rv =  Manage_subdevs(chosen->sys_name, mdfd, &devlist,
1025                                              -1, 0, NULL, 0);
1026                         close(mdfd);
1027                 }
1028                 if (verbose > 0) {
1029                         if (rv == 0)
1030                                 pr_err("added %s as spare for %s\n",
1031                                         devname, chosen->sys_name);
1032                         else
1033                                 pr_err("failed to add %s as spare for %s\n",
1034                                         devname, chosen->sys_name);
1035                 }
1036                 sysfs_free(chosen);
1037         }
1038         map_unlock(&map);
1039         return rv;
1040 }
1041
1042 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1043                                struct supertype *st, int verbose)
1044 {
1045         /* we know that at least one partition virtual-metadata is
1046          * allowed to incorporate spares like this device.  We need to
1047          * find a suitable device to copy partition information from.
1048          *
1049          * Getting a list of all disk (not partition) devices is
1050          * slightly non-trivial.  We could look at /sys/block, but
1051          * that is theoretically due to be removed.  Maybe best to use
1052          * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1053          * we depend on this directory of 'path' info.  But that fails
1054          * to find loop devices and probably others.  Maybe don't
1055          * worry about that, they aren't the real target.
1056          *
1057          * So: check things in /dev/disk/by-path to see if they are in
1058          * a compatible domain, then load the partition table and see
1059          * if it is OK for the new device, and choose the largest
1060          * partition table that fits.
1061          */
1062         DIR *dir;
1063         struct dirent *de;
1064         char *chosen = NULL;
1065         unsigned long long chosen_size = 0;
1066         struct supertype *chosen_st = NULL;
1067         int fd;
1068
1069         dir = opendir("/dev/disk/by-path");
1070         if (!dir)
1071                 return 1;
1072         while ((de = readdir(dir)) != NULL) {
1073                 char *ep;
1074                 struct dev_policy *pol2 = NULL;
1075                 struct domainlist *domlist = NULL;
1076                 int fd = -1;
1077                 struct mdinfo info;
1078                 struct supertype *st2 = NULL;
1079                 char *devname = NULL;
1080                 unsigned long long devsectors;
1081
1082                 if (de->d_ino == 0 || de->d_name[0] == '.' ||
1083                     (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1084                         goto next;
1085
1086                 ep = de->d_name + strlen(de->d_name);
1087                 while (ep > de->d_name &&
1088                        isdigit(ep[-1]))
1089                         ep--;
1090                 if (ep > de->d_name + 5 &&
1091                     strncmp(ep-5, "-part", 5) == 0)
1092                         /* This is a partition - skip it */
1093                         goto next;
1094
1095                 pol2 = path_policy(de->d_name, type_disk);
1096
1097                 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1098                 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1099                         /* new device is incompatible with this device. */
1100                         goto next;
1101
1102                 domain_free(domlist);
1103                 domlist = NULL;
1104
1105                 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1106                         devname = NULL;
1107                         goto next;
1108                 }
1109                 fd = open(devname, O_RDONLY);
1110                 if (fd < 0)
1111                         goto next;
1112                 if (get_dev_size(fd, devname, &devsectors) == 0)
1113                         goto next;
1114                 devsectors >>= 9;
1115
1116                 if (st)
1117                         st2 = dup_super(st);
1118                 else
1119                         st2 = guess_super_type(fd, guess_partitions);
1120                 if (st2 == NULL || st2->ss->load_super(st2, fd, NULL) < 0)
1121                         goto next;
1122                 st2->ignore_hw_compat = 0;
1123
1124                 if (!st) {
1125                         /* Check domain policy again, this time referring to metadata */
1126                         domain_merge(&domlist, pol2, st2->ss->name);
1127                         if (domain_test(domlist, pol, st2->ss->name) != 1)
1128                                 /* Incompatible devices for this metadata type */
1129                                 goto next;
1130                         if (!policy_action_allows(pol, st2->ss->name, act_spare))
1131                                 /* Some partition types allow sparing, but not
1132                                  * this one.
1133                                  */
1134                                 goto next;
1135                 }
1136
1137                 st2->ss->getinfo_super(st2, &info, NULL);
1138                 if (info.component_size > devsectors)
1139                         /* This partitioning doesn't fit in the device */
1140                         goto next;
1141
1142                 /* This is an acceptable device to copy partition
1143                  * metadata from.  We could just stop here, but I
1144                  * think I want to keep looking incase a larger
1145                  * metadata which makes better use of the device can
1146                  * be found.
1147                  */
1148                 if (chosen == NULL || chosen_size < info.component_size) {
1149                         chosen_size = info.component_size;
1150                         free(chosen);
1151                         chosen = devname;
1152                         devname = NULL;
1153                         if (chosen_st) {
1154                                 chosen_st->ss->free_super(chosen_st);
1155                                 free(chosen_st);
1156                         }
1157                         chosen_st = st2;
1158                         st2 = NULL;
1159                 }
1160
1161         next:
1162                 free(devname);
1163                 domain_free(domlist);
1164                 dev_policy_free(pol2);
1165                 if (st2)
1166                         st2->ss->free_super(st2);
1167                 free(st2);
1168
1169                 if (fd >= 0)
1170                         close(fd);
1171         }
1172
1173         closedir(dir);
1174
1175         if (!chosen)
1176                 return 1;
1177
1178         /* 'chosen' is the best device we can find.  Let's write its
1179          * metadata to devname dfd is read-only so don't use that
1180          */
1181         fd = open(devname, O_RDWR);
1182         if (fd >= 0) {
1183                 chosen_st->ss->store_super(chosen_st, fd);
1184                 close(fd);
1185         }
1186         free(chosen);
1187         chosen_st->ss->free_super(chosen_st);
1188         free(chosen_st);
1189         return 0;
1190 }
1191
1192 static int is_bare(int dfd)
1193 {
1194         unsigned long long size = 0;
1195         char bufpad[4096 + 4096];
1196         char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1197
1198         if (lseek(dfd, 0, SEEK_SET) != 0 ||
1199             read(dfd, buf, 4096) != 4096)
1200                 return 0;
1201
1202         if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1203                 return 0;
1204         if (memcmp(buf, buf+1, 4095) != 0)
1205                 return 0;
1206
1207         /* OK, first 4K appear blank, try the end. */
1208         get_dev_size(dfd, NULL, &size);
1209         if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1210             read(dfd, buf, 4096) != 4096)
1211                 return 0;
1212
1213         if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1214                 return 0;
1215         if (memcmp(buf, buf+1, 4095) != 0)
1216                 return 0;
1217
1218         return 1;
1219 }
1220
1221 /* adding a spare to a regular array is quite different from adding one to
1222  * a set-of-partitions virtual array.
1223  * This function determines which is worth trying and tries as appropriate.
1224  * Arrays are given priority over partitions.
1225  */
1226 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1227                      struct map_ent *target,
1228                      struct supertype *st, int verbose)
1229 {
1230         int i;
1231         int rv;
1232         int arrays_ok = 0;
1233         int partitions_ok = 0;
1234         int dfd = *dfdp;
1235         int bare;
1236
1237         /* Can only add a spare if device has at least one domain */
1238         if (pol_find(pol, pol_domain) == NULL)
1239                 return 1;
1240         /* And only if some action allows spares */
1241         if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1242                 return 1;
1243
1244         /* Now check if the device is bare.
1245          * bare devices can always be added as a spare
1246          * non-bare devices can only be added if spare-same-slot is permitted,
1247          * and this device is replacing a previous device - in which case 'target'
1248          * will be set.
1249          */
1250         if (!is_bare(dfd)) {
1251                 /* Must have a target and allow same_slot */
1252                 /* Later - may allow force_spare without target */
1253                 if (!target ||
1254                     !policy_action_allows(pol, st?st->ss->name:NULL,
1255                                           act_spare_same_slot)) {
1256                         if (verbose > 1)
1257                                 pr_err("%s is not bare, so not considering as a spare\n",
1258                                         devname);
1259                         return 1;
1260                 }
1261                 bare = 0;
1262         } else
1263                 bare = 1;
1264
1265         /* It might be OK to add this device to an array - need to see
1266          * what arrays might be candidates.
1267          */
1268         if (st) {
1269                 /* just try try 'array' or 'partition' based on this metadata */
1270                 if (st->ss->add_to_super)
1271                         return array_try_spare(devname, dfdp, pol, target, bare,
1272                                                st, verbose);
1273                 else
1274                         return partition_try_spare(devname, dfdp, pol,
1275                                                    st, verbose);
1276         }
1277         /* No metadata was specified or found so options are open.
1278          * Check for whether any array metadata, or any partition metadata
1279          * might allow adding the spare.  This check is just help to avoid
1280          * a more costly scan of all arrays when we can be sure that will
1281          * fail.
1282          */
1283         for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1284                 if (superlist[i]->add_to_super && !arrays_ok &&
1285                     policy_action_allows(pol, superlist[i]->name, act_spare))
1286                         arrays_ok = 1;
1287                 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1288                     policy_action_allows(pol, superlist[i]->name, act_spare))
1289                         partitions_ok = 1;
1290         }
1291         rv = 1;
1292         if (arrays_ok)
1293                 rv = array_try_spare(devname, dfdp, pol, target, bare,
1294                                      st, verbose);
1295         if (rv != 0 && partitions_ok)
1296                 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1297         return rv;
1298 }
1299
1300 int IncrementalScan(struct context *c, char *devnm)
1301 {
1302         /* look at every device listed in the 'map' file.
1303          * If one is found that is not running then:
1304          *  look in mdadm.conf for bitmap file.
1305          *   if one exists, but array has none, add it.
1306          *  try to start array in auto-readonly mode
1307          */
1308         struct map_ent *mapl = NULL;
1309         struct map_ent *me;
1310         struct mddev_ident *devs, *mddev;
1311         int rv = 0;
1312         char container[32];
1313         char *only = NULL;
1314
1315         map_read(&mapl);
1316         devs = conf_get_ident(NULL);
1317
1318 restart:
1319         for (me = mapl ; me ; me = me->next) {
1320                 struct mdinfo *sra;
1321                 int mdfd;
1322
1323                 if (devnm && strcmp(devnm, me->devnm) != 0)
1324                         continue;
1325                 if (me->metadata[0] == '/') {
1326                         char *sl;
1327
1328                         if (!devnm)
1329                                 continue;
1330
1331                         /* member array, need to work on container */
1332                         strncpy(container, me->metadata+1, 32);
1333                         container[31] = 0;
1334                         sl = strchr(container, '/');
1335                         if (sl)
1336                                 *sl = 0;
1337                         only = devnm;
1338                         devnm = container;
1339                         goto restart;
1340                 }
1341                 mdfd = open_dev(me->devnm);
1342
1343                 if (mdfd < 0)
1344                         continue;
1345                 if (!isdigit(me->metadata[0])) {
1346                         /* must be a container */
1347                         struct supertype *st = super_by_fd(mdfd, NULL);
1348                         int ret = 0;
1349                         struct map_ent *map = NULL;
1350
1351                         if (st && st->ss->load_container)
1352                                 ret = st->ss->load_container(st, mdfd, NULL);
1353                         close(mdfd);
1354                         if (!ret && st && st->ss->container_content) {
1355                                 if (map_lock(&map))
1356                                         pr_err("failed to get exclusive lock on mapfile\n");
1357                                 ret = Incremental_container(st, me->path, c, only);
1358                                 map_unlock(&map);
1359                         }
1360                         if (ret)
1361                                 rv = 1;
1362                         continue;
1363                 }
1364                 if (md_array_active(mdfd)) {
1365                         close(mdfd);
1366                         continue;
1367                 }
1368                 /* Ok, we can try this one.   Maybe it needs a bitmap */
1369                 for (mddev = devs ; mddev ; mddev = mddev->next)
1370                         if (mddev->devname && me->path &&
1371                             devname_matches(mddev->devname, me->path))
1372                                 break;
1373                 if (mddev && mddev->bitmap_file) {
1374                         /*
1375                          * Note: early kernels will wrongly fail this, so it
1376                          * is a hint only
1377                          */
1378                         int added = -1;
1379                         int bmfd;
1380
1381                         bmfd = open(mddev->bitmap_file, O_RDWR);
1382                         if (bmfd >= 0) {
1383                                 added = ioctl(mdfd, SET_BITMAP_FILE, bmfd);
1384                                 close(bmfd);
1385                         }
1386                         if (c->verbose >= 0) {
1387                                 if (added == 0)
1388                                         pr_err("Added bitmap %s to %s\n",
1389                                                mddev->bitmap_file, me->path);
1390                                 else if (errno != EEXIST)
1391                                         pr_err("Failed to add bitmap to %s: %s\n",
1392                                                me->path, strerror(errno));
1393                         }
1394                 }
1395                 /* FIXME check for reshape_active and consider not
1396                  * starting array.
1397                  */
1398                 sra = sysfs_read(mdfd, NULL, 0);
1399                 if (sra) {
1400                         if (sysfs_set_str(sra, NULL,
1401                                           "array_state", "read-auto") == 0) {
1402                                 if (c->verbose >= 0)
1403                                         pr_err("started array %s\n",
1404                                                me->path ?: me->devnm);
1405                         } else {
1406                                 pr_err("failed to start array %s: %s\n",
1407                                        me->path ?: me->devnm,
1408                                        strerror(errno));
1409                                 rv = 1;
1410                         }
1411                         sysfs_free(sra);
1412                 }
1413         }
1414         return rv;
1415 }
1416
1417 static char *container2devname(char *devname)
1418 {
1419         char *mdname = NULL;
1420
1421         if (devname[0] == '/') {
1422                 int fd = open(devname, O_RDONLY);
1423                 if (fd >= 0) {
1424                         mdname = xstrdup(fd2devnm(fd));
1425                         close(fd);
1426                 }
1427         } else {
1428                 int uuid[4];
1429                 struct map_ent *mp, *map = NULL;
1430
1431                 if (!parse_uuid(devname, uuid))
1432                         return mdname;
1433                 mp = map_by_uuid(&map, uuid);
1434                 if (mp)
1435                         mdname = xstrdup(mp->devnm);
1436                 map_free(map);
1437         }
1438
1439         return mdname;
1440 }
1441
1442 static int Incremental_container(struct supertype *st, char *devname,
1443                                  struct context *c, char *only)
1444 {
1445         /* Collect the contents of this container and for each
1446          * array, choose a device name and assemble the array.
1447          */
1448
1449         struct mdinfo *list;
1450         struct mdinfo *ra;
1451         struct map_ent *map = NULL;
1452         struct mdinfo info;
1453         int trustworthy;
1454         struct mddev_ident *match;
1455         int rv = 0;
1456         struct domainlist *domains;
1457         struct map_ent *smp;
1458         int suuid[4];
1459         int sfd;
1460         int ra_blocked = 0;
1461         int ra_all = 0;
1462         int result = 0;
1463
1464         st->ss->getinfo_super(st, &info, NULL);
1465
1466         if ((c->runstop > 0 && info.container_enough >= 0) ||
1467             info.container_enough > 0)
1468                 /* pass */;
1469         else {
1470                 if (c->export) {
1471                         printf("MD_STARTED=no\n");
1472                 } else if (c->verbose)
1473                         pr_err("not enough devices to start the container\n");
1474                 return 0;
1475         }
1476
1477         match = conf_match(st, &info, devname, c->verbose, &rv);
1478         if (match == NULL && rv == 2)
1479                 return rv;
1480
1481         /* Need to compute 'trustworthy' */
1482         if (match)
1483                 trustworthy = LOCAL;
1484         else if (st->ss->match_home(st, c->homehost) == 1)
1485                 trustworthy = LOCAL;
1486         else if (st->ss->match_home(st, "any") == 1)
1487                 trustworthy = LOCAL;
1488         else
1489                 trustworthy = FOREIGN;
1490
1491         list = st->ss->container_content(st, NULL);
1492         /* when nothing to activate - quit */
1493         if (list == NULL) {
1494                 if (c->export) {
1495                         printf("MD_STARTED=nothing\n");
1496                 }
1497                 return 0;
1498         }
1499         for (ra = list ; ra ; ra = ra->next) {
1500                 int mdfd;
1501                 char chosen_name[1024];
1502                 struct map_ent *mp;
1503                 struct mddev_ident *match = NULL;
1504
1505                 ra_all++;
1506                 /* do not activate arrays blocked by metadata handler */
1507                 if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) {
1508                         pr_err("Cannot activate array %s in %s.\n",
1509                                 ra->text_version, devname);
1510                         ra_blocked++;
1511                         continue;
1512                 }
1513                 mp = map_by_uuid(&map, ra->uuid);
1514
1515                 if (mp) {
1516                         mdfd = open_dev(mp->devnm);
1517                         if (mp->path)
1518                                 strcpy(chosen_name, mp->path);
1519                         else
1520                                 strcpy(chosen_name, mp->devnm);
1521                 } else if (!only) {
1522
1523                         /* Check in mdadm.conf for container == devname and
1524                          * member == ra->text_version after second slash.
1525                          */
1526                         char *sub = strchr(ra->text_version+1, '/');
1527                         struct mddev_ident *array_list;
1528                         if (sub) {
1529                                 sub++;
1530                                 array_list = conf_get_ident(NULL);
1531                         } else
1532                                 array_list = NULL;
1533                         for(; array_list ; array_list = array_list->next) {
1534                                 char *dn;
1535                                 if (array_list->member == NULL ||
1536                                     array_list->container == NULL)
1537                                         continue;
1538                                 if (strcmp(array_list->member, sub) != 0)
1539                                         continue;
1540                                 if (array_list->uuid_set &&
1541                                     !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1542                                         continue;
1543                                 dn = container2devname(array_list->container);
1544                                 if (dn == NULL)
1545                                         continue;
1546                                 if (strncmp(dn, ra->text_version+1,
1547                                             strlen(dn)) != 0 ||
1548                                     ra->text_version[strlen(dn)+1] != '/') {
1549                                         free(dn);
1550                                         continue;
1551                                 }
1552                                 free(dn);
1553                                 /* we have a match */
1554                                 match = array_list;
1555                                 if (c->verbose>0)
1556                                         pr_err("match found for member %s\n",
1557                                                 array_list->member);
1558                                 break;
1559                         }
1560
1561                         if (match && match->devname &&
1562                             strcasecmp(match->devname, "<ignore>") == 0) {
1563                                 if (c->verbose > 0)
1564                                         pr_err("array %s/%s is explicitly ignored by mdadm.conf\n",
1565                                                match->container, match->member);
1566                                 continue;
1567                         }
1568                         if (match)
1569                                 trustworthy = LOCAL;
1570
1571                         mdfd = create_mddev(match ? match->devname : NULL,
1572                                             ra->name,
1573                                             c->autof,
1574                                             trustworthy,
1575                                             chosen_name, 0);
1576                 }
1577                 if (only && (!mp || strcmp(mp->devnm, only) != 0))
1578                         continue;
1579
1580                 if (mdfd < 0) {
1581                         pr_err("failed to open %s: %s.\n",
1582                                 chosen_name, strerror(errno));
1583                         return 2;
1584                 }
1585
1586                 assemble_container_content(st, mdfd, ra, c,
1587                                            chosen_name, &result);
1588                 close(mdfd);
1589         }
1590         if (c->export && result) {
1591                 char sep = '=';
1592                 printf("MD_STARTED");
1593                 if (result & INCR_NO) {
1594                         printf("%cno", sep);
1595                         sep = ',';
1596                 }
1597                 if (result & INCR_UNSAFE) {
1598                         printf("%cunsafe", sep);
1599                         sep = ',';
1600                 }
1601                 if (result & INCR_ALREADY) {
1602                         printf("%calready", sep);
1603                         sep = ',';
1604                 }
1605                 if (result & INCR_YES) {
1606                         printf("%cyes", sep);
1607                         sep = ',';
1608                 }
1609                 printf("\n");
1610         }
1611
1612         /* don't move spares to container with volume being activated
1613            when all volumes are blocked */
1614         if (ra_all == ra_blocked)
1615                 return 0;
1616
1617         /* Now move all suitable spares from spare container */
1618         domains = domain_from_array(list, st->ss->name);
1619         memcpy(suuid, uuid_zero, sizeof(int[4]));
1620         if (domains &&
1621             (smp = map_by_uuid(&map, suuid)) != NULL &&
1622             (sfd = open(smp->path, O_RDONLY)) >= 0) {
1623                 /* spare container found */
1624                 struct supertype *sst =
1625                         super_imsm.match_metadata_desc("imsm");
1626                 struct mdinfo *sinfo;
1627                 unsigned long long min_size = 0;
1628                 if (st->ss->min_acceptable_spare_size)
1629                         min_size = st->ss->min_acceptable_spare_size(st);
1630                 if (!sst->ss->load_container(sst, sfd, NULL)) {
1631                         close(sfd);
1632                         sinfo = container_choose_spares(sst, min_size,
1633                                                         domains, NULL,
1634                                                         st->ss->name, 0);
1635                         sst->ss->free_super(sst);
1636                         if (sinfo){
1637                                 int count = 0;
1638                                 struct mdinfo *disks = sinfo->devs;
1639                                 while (disks) {
1640                                         /* move spare from spare
1641                                          * container to currently
1642                                          * assembled one
1643                                          */
1644                                         if (move_spare(
1645                                                     smp->path,
1646                                                     devname,
1647                                                     makedev(disks->disk.major,
1648                                                             disks->disk.minor)))
1649                                                 count++;
1650                                         disks = disks->next;
1651                                 }
1652                                 if (count)
1653                                         pr_err("Added %d spare%s to %s\n",
1654                                                count, count>1?"s":"", devname);
1655                         }
1656                         sysfs_free(sinfo);
1657                 } else
1658                         close(sfd);
1659         }
1660         domain_free(domains);
1661         return 0;
1662 }
1663
1664 static void run_udisks(char *arg1, char *arg2)
1665 {
1666         int pid = fork();
1667         int status;
1668         if (pid == 0) {
1669                 execl("/usr/bin/udisks", "udisks", arg1, arg2, NULL);
1670                 execl("/bin/udisks", "udisks", arg1, arg2, NULL);
1671                 exit(1);
1672         }
1673         while (pid > 0 && wait(&status) != pid)
1674                 ;
1675 }
1676
1677 /*
1678  * IncrementalRemove - Attempt to see if the passed in device belongs to any
1679  * raid arrays, and if so first fail (if needed) and then remove the device.
1680  *
1681  * @devname - The device we want to remove
1682  * @id_path - name as found in /dev/disk/by-path for this device
1683  *
1684  * Note: the device name must be a kernel name like "sda", so
1685  * that we can find it in /proc/mdstat
1686  */
1687 int IncrementalRemove(char *devname, char *id_path, int verbose)
1688 {
1689         int mdfd;
1690         int rv = 0;
1691         struct mdstat_ent *ent;
1692         struct mddev_dev devlist;
1693         struct mdinfo mdi;
1694         char buf[32];
1695
1696         if (!id_path)
1697                 dprintf("incremental removal without --path <id_path> lacks the possibility to re-add new device in this port\n");
1698
1699         if (strchr(devname, '/')) {
1700                 pr_err("incremental removal requires a kernel device name, not a file: %s\n", devname);
1701                 return 1;
1702         }
1703         ent = mdstat_by_component(devname);
1704         if (!ent) {
1705                 if (verbose >= 0)
1706                         pr_err("%s does not appear to be a component of any array\n", devname);
1707                 return 1;
1708         }
1709         if (sysfs_init(&mdi, -1, ent->devnm)) {
1710                 pr_err("unable to initialize sysfs for: %s\n", devname);
1711                 return 1;
1712         }
1713         mdfd = open_dev_excl(ent->devnm);
1714         if (mdfd > 0) {
1715                 close(mdfd);
1716                 if (sysfs_get_str(&mdi, NULL, "array_state",
1717                                   buf, sizeof(buf)) > 0) {
1718                         if (strncmp(buf, "active", 6) == 0 ||
1719                             strncmp(buf, "clean", 5) == 0)
1720                                 sysfs_set_str(&mdi, NULL,
1721                                               "array_state", "read-auto");
1722                 }
1723         }
1724         mdfd = open_dev(ent->devnm);
1725         if (mdfd < 0) {
1726                 if (verbose >= 0)
1727                         pr_err("Cannot open array %s!!\n", ent->devnm);
1728                 free_mdstat(ent);
1729                 return 1;
1730         }
1731
1732         if (id_path) {
1733                 struct map_ent *map = NULL, *me;
1734                 me = map_by_devnm(&map, ent->devnm);
1735                 if (me)
1736                         policy_save_path(id_path, me);
1737                 map_free(map);
1738         }
1739
1740         memset(&devlist, 0, sizeof(devlist));
1741         devlist.devname = devname;
1742         devlist.disposition = 'f';
1743         /* for a container, we must fail each member array */
1744         if (ent->metadata_version &&
1745             strncmp(ent->metadata_version, "external:", 9) == 0) {
1746                 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1747                 struct mdstat_ent *memb;
1748                 for (memb = mdstat ; memb ; memb = memb->next)
1749                         if (is_container_member(memb, ent->devnm)) {
1750                                 int subfd = open_dev(memb->devnm);
1751                                 if (subfd >= 0) {
1752                                         rv |= Manage_subdevs(
1753                                                 memb->devnm, subfd,
1754                                                 &devlist, verbose, 0,
1755                                                 NULL, 0);
1756                                         close(subfd);
1757                                 }
1758                         }
1759                 free_mdstat(mdstat);
1760         } else
1761                 rv |= Manage_subdevs(ent->devnm, mdfd, &devlist,
1762                                     verbose, 0, NULL, 0);
1763         if (rv & 2) {
1764                 /* Failed due to EBUSY, try to stop the array.
1765                  * Give udisks a chance to unmount it first.
1766                  */
1767                 int devid = devnm2devid(ent->devnm);
1768                 run_udisks("--unmount", map_dev(major(devid),minor(devid), 0));
1769                 rv = Manage_stop(ent->devnm, mdfd, verbose, 1);
1770                 if (rv)
1771                         /* At least we can try to trigger a 'remove' */
1772                         sysfs_uevent(&mdi, "remove");
1773                 if (verbose) {
1774                         if (rv)
1775                                 pr_err("Fail to stop %s too.\n", ent->devnm);
1776                 }
1777         } else {
1778                 devlist.disposition = 'r';
1779                 rv = Manage_subdevs(ent->devnm, mdfd, &devlist,
1780                                     verbose, 0, NULL, 0);
1781         }
1782         close(mdfd);
1783         free_mdstat(ent);
1784         return rv;
1785 }