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         __u64 max_journal_events = 0;
 688         char *avail = NULL;
 689         int *best = NULL;
 690         char *devmap = NULL;
 691         int numdevs = 0;
 692         int devnum;
 693         int b, i;
 694         int raid_disks = 0;
 695
 696         if (!sra)
 697                 return 0;
 698
 699         for (d = sra->devs ; d ; d = d->next)
 700                 numdevs++;
 701         for (d = sra->devs, devnum = 0 ; d ; d = d->next, devnum++) {
 702                 char dn[30];
 703                 int dfd;
 704                 int ok;
 705                 struct mdinfo info;
 706
 707                 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
 708                 dfd = dev_open(dn, O_RDONLY);
 709                 if (dfd < 0)
 710                         continue;
 711                 ok =  st->ss->load_super(st, dfd, NULL);
 712                 close(dfd);
 713                 if (ok != 0)
 714                         continue;
 715
 716                 info.array.raid_disks = raid_disks;
 717                 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
 718                 if (info.disk.raid_disk == MD_DISK_ROLE_JOURNAL &&
 719                     info.events > max_journal_events)
 720                         max_journal_events = info.events;
 721                 if (!avail) {
 722                         raid_disks = info.array.raid_disks;
 723                         avail = xcalloc(raid_disks, 1);
 724                         *availp = avail;
 725
 726                         best = xcalloc(raid_disks, sizeof(int));
 727                         devmap = xcalloc(raid_disks, numdevs);
 728
 729                         st->ss->getinfo_super(st, &info, devmap);
 730                 }
 731
 732                 if (info.disk.state & (1<<MD_DISK_SYNC))
 733                 {
 734                         if (cnt == 0) {
 735                                 cnt++;
 736                                 max_events = info.events;
 737                                 avail[info.disk.raid_disk] = 2;
 738                                 best[info.disk.raid_disk] = devnum;
 739                                 st->ss->getinfo_super(st, bestinfo, NULL);
 740                         } else if (info.events == max_events) {
 741                                 avail[info.disk.raid_disk] = 2;
 742                                 best[info.disk.raid_disk] = devnum;
 743                         } else if (info.events == max_events-1) {
 744                                 if (avail[info.disk.raid_disk] == 0) {
 745                                         avail[info.disk.raid_disk] = 1;
 746                                         best[info.disk.raid_disk] = devnum;
 747                                 }
 748                         } else if (info.events < max_events - 1)
 749                                 ;
 750                         else if (info.events == max_events+1) {
 751                                 int i;
 752                                 max_events = info.events;
 753                                 for (i = 0; i < raid_disks; i++)
 754                                         if (avail[i])
 755                                                 avail[i]--;
 756                                 avail[info.disk.raid_disk] = 2;
 757                                 best[info.disk.raid_disk] = devnum;
 758                                 st->ss->getinfo_super(st, bestinfo, NULL);
 759                         } else { /* info.events much bigger */
 760                                 memset(avail, 0, raid_disks);
 761                                 max_events = info.events;
 762                                 avail[info.disk.raid_disk] = 2;
 763                                 best[info.disk.raid_disk] = devnum;
 764                                 st->ss->getinfo_super(st, bestinfo, NULL);
 765                         }
 766                 } else if (info.disk.state & (1<<MD_DISK_REPLACEMENT))
 767                         replcnt++;
 768                 st->ss->free_super(st);
 769         }
 770         if (max_journal_events >= max_events - 1)
 771                 bestinfo->journal_clean = 1;
 772
 773         if (!avail)
 774                 return 0;
 775         /* We need to reject any device that thinks the best device is
 776          * failed or missing */
 777         for (b = 0; b < raid_disks; b++)
 778                 if (avail[b] == 2)
 779                         break;
 780         cnt = 0;
 781         for (i = 0 ; i < raid_disks ; i++) {
 782                 if (i != b && avail[i])
 783                         if (devmap[raid_disks * best[i] + b] == 0) {
 784                                 /* This device thinks 'b' is failed -
 785                                  * don't use it */
 786                                 devnum = best[i];
 787                                 for (d=sra->devs ; devnum; d = d->next)
 788                                         devnum--;
 789                                 d->disk.state |= (1 << MD_DISK_REMOVED);
 790                                 avail[i] = 0;
 791                         }
 792                 if (avail[i])
 793                         cnt++;
 794         }
 795         /* Also need to reject any spare device with an event count that
 796          * is too high
 797          */
 798         for (d = sra->devs; d; d = d->next) {
 799                 if (!(d->disk.state & (1<<MD_DISK_SYNC)) &&
 800                     d->events > max_events)
 801                         d->disk.state |= (1 << MD_DISK_REMOVED);
 802         }
 803         free(best);
 804         free(devmap);
 805         return cnt + replcnt;
 806 }
 807
 808 /* test if container has degraded member(s) */
 809 static int
 810 container_members_max_degradation(struct map_ent *map, struct map_ent *me)
 811 {
 812         struct mdinfo *sra;
 813         int degraded, max_degraded = 0;
 814
 815         for(; map; map = map->next) {
 816                 if (!metadata_container_matches(map->metadata, me->devnm))
 817                         continue;
 818                 /* most accurate information regarding array degradation */
 819                 sra = sysfs_read(-1, map->devnm,
 820                                  GET_DISKS | GET_DEVS | GET_STATE);
 821                 if (!sra)
 822                         continue;
 823                 degraded = sra->array.raid_disks - sra->array.active_disks -
 824                         sra->array.spare_disks;
 825                 if (degraded > max_degraded)
 826                         max_degraded = degraded;
 827                 sysfs_free(sra);
 828         }
 829
 830         return max_degraded;
 831 }
 832
 833 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
 834                            struct map_ent *target, int bare,
 835                            struct supertype *st, int verbose)
 836 {
 837         /* This device doesn't have any md metadata
 838          * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
 839          * If 'st' is not set, then we only know that some metadata allows this,
 840          * others possibly don't.
 841          * So look for a container or array to attach the device to.
 842          * Prefer 'target' if that is set and the array is found.
 843          *
 844          * If st is set, then only arrays of that type are considered
 845          * Return 0 on success, or some exit code on failure, probably 1.
 846          */
 847         int rv = 1;
 848         dev_t rdev;
 849         struct map_ent *mp, *map = NULL;
 850         struct mdinfo *chosen = NULL;
 851         int dfd = *dfdp;
 852
 853         if (!fstat_is_blkdev(dfd, devname, &rdev))
 854                 return 1;
 855
 856         /*
 857          * Now we need to find a suitable array to add this to.
 858          * We only accept arrays that:
 859          *  - match 'st'
 860          *  - are in the same domains as the device
 861          *  - are of an size for which the device will be useful
 862          * and we choose the one that is the most degraded
 863          */
 864
 865         if (map_lock(&map)) {
 866                 pr_err("failed to get exclusive lock on mapfile\n");
 867                 return 1;
 868         }
 869         for (mp = map ; mp ; mp = mp->next) {
 870                 struct supertype *st2;
 871                 struct domainlist *dl = NULL;
 872                 struct mdinfo *sra;
 873                 unsigned long long devsize;
 874                 struct spare_criteria sc = {0, 0};
 875
 876                 if (is_subarray(mp->metadata))
 877                         continue;
 878                 if (st) {
 879                         st2 = st->ss->match_metadata_desc(mp->metadata);
 880                         if (!st2 ||
 881                             (st->minor_version >= 0 &&
 882                              st->minor_version != st2->minor_version)) {
 883                                 if (verbose > 1)
 884                                         pr_err("not adding %s to %s as metadata type doesn't match\n",
 885                                                 devname, mp->path);
 886                                 free(st2);
 887                                 continue;
 888                         }
 889                         free(st2);
 890                 }
 891                 sra = sysfs_read(-1, mp->devnm,
 892                                  GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
 893                                  GET_COMPONENT|GET_VERSION);
 894                 if (sra)
 895                         sra->array.failed_disks = -1;
 896                 else
 897                         continue;
 898                 if (st == NULL) {
 899                         int i;
 900                         st2 = NULL;
 901                         for(i = 0; !st2 && superlist[i]; i++)
 902                                 st2 = superlist[i]->match_metadata_desc(
 903                                         sra->text_version);
 904                         if (!st2) {
 905                                 if (verbose > 1)
 906                                         pr_err("not adding %s to %s as metadata not recognised.\n",
 907                                                 devname, mp->path);
 908                                 goto next;
 909                         }
 910                         /* Need to double check the 'act_spare' permissions applies
 911                          * to this metadata.
 912                          */
 913                         if (!policy_action_allows(pol, st2->ss->name, act_spare))
 914                                 goto next;
 915                         if (!bare && !policy_action_allows(pol, st2->ss->name,
 916                                                            act_spare_same_slot))
 917                                 goto next;
 918                 } else
 919                         st2 = st;
 920                 /* update number of failed disks for mostly degraded
 921                  * container member */
 922                 if (sra->array.failed_disks == -1)
 923                         sra->array.failed_disks = container_members_max_degradation(map, mp);
 924
 925                 get_dev_size(dfd, NULL, &devsize);
 926                 if (sra->component_size == 0) {
 927                         /* true for containers, here we must read superblock
 928                          * to obtain minimum spare size */
 929                         struct supertype *st3 = dup_super(st2);
 930                         int mdfd = open_dev(mp->devnm);
 931                         if (mdfd < 0) {
 932                                 free(st3);
 933                                 goto next;
 934                         }
 935                         if (st3->ss->load_container &&
 936                             !st3->ss->load_container(st3, mdfd, mp->path)) {
 937                                 if (st3->ss->get_spare_criteria)
 938                                         st3->ss->get_spare_criteria(st3, &sc);
 939                                 st3->ss->free_super(st3);
 940                         }
 941                         free(st3);
 942                         close(mdfd);
 943                 }
 944                 if ((sra->component_size > 0 &&
 945                      st2->ss->avail_size(st2, devsize,
 946                                          sra->devs ? sra->devs->data_offset :
 947                                          INVALID_SECTORS) <
 948                      sra->component_size) ||
 949                     (sra->component_size == 0 && devsize < sc.min_size)) {
 950                         if (verbose > 1)
 951                                 pr_err("not adding %s to %s as it is too small\n",
 952                                         devname, mp->path);
 953                         goto next;
 954                 }
 955                 /* test against target.
 956                  * If 'target' is set and 'bare' is false, we only accept
 957                  * arrays/containers that match 'target'.
 958                  * If 'target' is set and 'bare' is true, we prefer the
 959                  * array which matches 'target'.
 960                  * target is considered only if we deal with degraded array
 961                  */
 962                 if (target && policy_action_allows(pol, st2->ss->name,
 963                                                    act_spare_same_slot)) {
 964                         if (strcmp(target->metadata, mp->metadata) == 0 &&
 965                             memcmp(target->uuid, mp->uuid,
 966                                    sizeof(target->uuid)) == 0 &&
 967                             sra->array.failed_disks > 0) {
 968                                 /* This is our target!! */
 969                                 sysfs_free(chosen);
 970                                 chosen = sra;
 971                                 sra = NULL;
 972                                 /* skip to end so we don't check any more */
 973                                 while (mp->next)
 974                                         mp = mp->next;
 975                                 goto next;
 976                         }
 977                         /* not our target */
 978                         if (!bare)
 979                                 goto next;
 980                 }
 981
 982                 dl = domain_from_array(sra, st2->ss->name);
 983                 if (domain_test(dl, pol, st2->ss->name) != 1) {
 984                         /* domain test fails */
 985                         if (verbose > 1)
 986                                 pr_err("not adding %s to %s as it is not in a compatible domain\n",
 987                                         devname, mp->path);
 988
 989                         goto next;
 990                 }
 991                 /* all tests passed, OK to add to this array */
 992                 if (!chosen) {
 993                         chosen = sra;
 994                         sra = NULL;
 995                 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
 996                         sysfs_free(chosen);
 997                         chosen = sra;
 998                         sra = NULL;
 999                 }
1000         next:
1001                 sysfs_free(sra);
1002                 if (st != st2)
1003                         free(st2);
1004                 if (dl)
1005                         domain_free(dl);
1006         }
1007         if (chosen) {
1008                 /* add current device to chosen array as a spare */
1009                 int mdfd = open_dev(chosen->sys_name);
1010                 if (mdfd >= 0) {
1011                         struct mddev_dev devlist;
1012                         char chosen_devname[24]; // 2*11 for int (including signs) + colon + null
1013                         devlist.next = NULL;
1014                         devlist.used = 0;
1015                         devlist.writemostly = FlagDefault;
1016                         devlist.failfast = FlagDefault;
1017                         devlist.devname = chosen_devname;
1018                         sprintf(chosen_devname, "%d:%d", major(rdev),
1019                                 minor(rdev));
1020                         devlist.disposition = 'a';
1021                         close(dfd);
1022                         *dfdp = -1;
1023                         rv =  Manage_subdevs(chosen->sys_name, mdfd, &devlist,
1024                                              -1, 0, NULL, 0);
1025                         close(mdfd);
1026                 }
1027                 if (verbose > 0) {
1028                         if (rv == 0)
1029                                 pr_err("added %s as spare for %s\n",
1030                                         devname, chosen->sys_name);
1031                         else
1032                                 pr_err("failed to add %s as spare for %s\n",
1033                                         devname, chosen->sys_name);
1034                 }
1035                 sysfs_free(chosen);
1036         }
1037         map_unlock(&map);
1038         return rv;
1039 }
1040
1041 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1042                                struct supertype *st, int verbose)
1043 {
1044         /* we know that at least one partition virtual-metadata is
1045          * allowed to incorporate spares like this device.  We need to
1046          * find a suitable device to copy partition information from.
1047          *
1048          * Getting a list of all disk (not partition) devices is
1049          * slightly non-trivial.  We could look at /sys/block, but
1050          * that is theoretically due to be removed.  Maybe best to use
1051          * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1052          * we depend on this directory of 'path' info.  But that fails
1053          * to find loop devices and probably others.  Maybe don't
1054          * worry about that, they aren't the real target.
1055          *
1056          * So: check things in /dev/disk/by-path to see if they are in
1057          * a compatible domain, then load the partition table and see
1058          * if it is OK for the new device, and choose the largest
1059          * partition table that fits.
1060          */
1061         DIR *dir;
1062         struct dirent *de;
1063         char *chosen = NULL;
1064         unsigned long long chosen_size = 0;
1065         struct supertype *chosen_st = NULL;
1066         int fd;
1067
1068         dir = opendir("/dev/disk/by-path");
1069         if (!dir)
1070                 return 1;
1071         while ((de = readdir(dir)) != NULL) {
1072                 char *ep;
1073                 struct dev_policy *pol2 = NULL;
1074                 struct domainlist *domlist = NULL;
1075                 int fd = -1;
1076                 struct mdinfo info;
1077                 struct supertype *st2 = NULL;
1078                 char *devname = NULL;
1079                 unsigned long long devsectors;
1080
1081                 if (de->d_ino == 0 || de->d_name[0] == '.' ||
1082                     (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1083                         goto next;
1084
1085                 ep = de->d_name + strlen(de->d_name);
1086                 while (ep > de->d_name &&
1087                        isdigit(ep[-1]))
1088                         ep--;
1089                 if (ep > de->d_name + 5 &&
1090                     strncmp(ep-5, "-part", 5) == 0)
1091                         /* This is a partition - skip it */
1092                         goto next;
1093
1094                 pol2 = path_policy(de->d_name, type_disk);
1095
1096                 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1097                 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1098                         /* new device is incompatible with this device. */
1099                         goto next;
1100
1101                 domain_free(domlist);
1102                 domlist = NULL;
1103
1104                 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1105                         devname = NULL;
1106                         goto next;
1107                 }
1108                 fd = open(devname, O_RDONLY);
1109                 if (fd < 0)
1110                         goto next;
1111                 if (get_dev_size(fd, devname, &devsectors) == 0)
1112                         goto next;
1113                 devsectors >>= 9;
1114
1115                 if (st)
1116                         st2 = dup_super(st);
1117                 else
1118                         st2 = guess_super_type(fd, guess_partitions);
1119                 if (st2 == NULL || st2->ss->load_super(st2, fd, NULL) < 0)
1120                         goto next;
1121                 st2->ignore_hw_compat = 0;
1122
1123                 if (!st) {
1124                         /* Check domain policy again, this time referring to metadata */
1125                         domain_merge(&domlist, pol2, st2->ss->name);
1126                         if (domain_test(domlist, pol, st2->ss->name) != 1)
1127                                 /* Incompatible devices for this metadata type */
1128                                 goto next;
1129                         if (!policy_action_allows(pol, st2->ss->name, act_spare))
1130                                 /* Some partition types allow sparing, but not
1131                                  * this one.
1132                                  */
1133                                 goto next;
1134                 }
1135
1136                 st2->ss->getinfo_super(st2, &info, NULL);
1137                 if (info.component_size > devsectors)
1138                         /* This partitioning doesn't fit in the device */
1139                         goto next;
1140
1141                 /* This is an acceptable device to copy partition
1142                  * metadata from.  We could just stop here, but I
1143                  * think I want to keep looking incase a larger
1144                  * metadata which makes better use of the device can
1145                  * be found.
1146                  */
1147                 if (chosen == NULL || chosen_size < info.component_size) {
1148                         chosen_size = info.component_size;
1149                         free(chosen);
1150                         chosen = devname;
1151                         devname = NULL;
1152                         if (chosen_st) {
1153                                 chosen_st->ss->free_super(chosen_st);
1154                                 free(chosen_st);
1155                         }
1156                         chosen_st = st2;
1157                         st2 = NULL;
1158                 }
1159
1160         next:
1161                 free(devname);
1162                 domain_free(domlist);
1163                 dev_policy_free(pol2);
1164                 if (st2)
1165                         st2->ss->free_super(st2);
1166                 free(st2);
1167
1168                 if (fd >= 0)
1169                         close(fd);
1170         }
1171
1172         closedir(dir);
1173
1174         if (!chosen)
1175                 return 1;
1176
1177         /* 'chosen' is the best device we can find.  Let's write its
1178          * metadata to devname dfd is read-only so don't use that
1179          */
1180         fd = open(devname, O_RDWR);
1181         if (fd >= 0) {
1182                 chosen_st->ss->store_super(chosen_st, fd);
1183                 close(fd);
1184         }
1185         free(chosen);
1186         chosen_st->ss->free_super(chosen_st);
1187         free(chosen_st);
1188         return 0;
1189 }
1190
1191 static int is_bare(int dfd)
1192 {
1193         unsigned long long size = 0;
1194         char bufpad[4096 + 4096];
1195         char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1196
1197         if (lseek(dfd, 0, SEEK_SET) != 0 ||
1198             read(dfd, buf, 4096) != 4096)
1199                 return 0;
1200
1201         if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1202                 return 0;
1203         if (memcmp(buf, buf+1, 4095) != 0)
1204                 return 0;
1205
1206         /* OK, first 4K appear blank, try the end. */
1207         get_dev_size(dfd, NULL, &size);
1208         if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1209             read(dfd, buf, 4096) != 4096)
1210                 return 0;
1211
1212         if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1213                 return 0;
1214         if (memcmp(buf, buf+1, 4095) != 0)
1215                 return 0;
1216
1217         return 1;
1218 }
1219
1220 /* adding a spare to a regular array is quite different from adding one to
1221  * a set-of-partitions virtual array.
1222  * This function determines which is worth trying and tries as appropriate.
1223  * Arrays are given priority over partitions.
1224  */
1225 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1226                      struct map_ent *target,
1227                      struct supertype *st, int verbose)
1228 {
1229         int i;
1230         int rv;
1231         int arrays_ok = 0;
1232         int partitions_ok = 0;
1233         int dfd = *dfdp;
1234         int bare;
1235
1236         /* Can only add a spare if device has at least one domain */
1237         if (pol_find(pol, pol_domain) == NULL)
1238                 return 1;
1239         /* And only if some action allows spares */
1240         if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1241                 return 1;
1242
1243         /* Now check if the device is bare.
1244          * bare devices can always be added as a spare
1245          * non-bare devices can only be added if spare-same-slot is permitted,
1246          * and this device is replacing a previous device - in which case 'target'
1247          * will be set.
1248          */
1249         if (!is_bare(dfd)) {
1250                 /* Must have a target and allow same_slot */
1251                 /* Later - may allow force_spare without target */
1252                 if (!target ||
1253                     !policy_action_allows(pol, st?st->ss->name:NULL,
1254                                           act_spare_same_slot)) {
1255                         if (verbose > 1)
1256                                 pr_err("%s is not bare, so not considering as a spare\n",
1257                                         devname);
1258                         return 1;
1259                 }
1260                 bare = 0;
1261         } else
1262                 bare = 1;
1263
1264         /* It might be OK to add this device to an array - need to see
1265          * what arrays might be candidates.
1266          */
1267         if (st) {
1268                 /* just try try 'array' or 'partition' based on this metadata */
1269                 if (st->ss->add_to_super)
1270                         return array_try_spare(devname, dfdp, pol, target, bare,
1271                                                st, verbose);
1272                 else
1273                         return partition_try_spare(devname, dfdp, pol,
1274                                                    st, verbose);
1275         }
1276         /* No metadata was specified or found so options are open.
1277          * Check for whether any array metadata, or any partition metadata
1278          * might allow adding the spare.  This check is just help to avoid
1279          * a more costly scan of all arrays when we can be sure that will
1280          * fail.
1281          */
1282         for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1283                 if (superlist[i]->add_to_super && !arrays_ok &&
1284                     policy_action_allows(pol, superlist[i]->name, act_spare))
1285                         arrays_ok = 1;
1286                 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1287                     policy_action_allows(pol, superlist[i]->name, act_spare))
1288                         partitions_ok = 1;
1289         }
1290         rv = 1;
1291         if (arrays_ok)
1292                 rv = array_try_spare(devname, dfdp, pol, target, bare,
1293                                      st, verbose);
1294         if (rv != 0 && partitions_ok)
1295                 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1296         return rv;
1297 }
1298
1299 int IncrementalScan(struct context *c, char *devnm)
1300 {
1301         /* look at every device listed in the 'map' file.
1302          * If one is found that is not running then:
1303          *  look in mdadm.conf for bitmap file.
1304          *   if one exists, but array has none, add it.
1305          *  try to start array in auto-readonly mode
1306          */
1307         struct map_ent *mapl = NULL;
1308         struct map_ent *me;
1309         struct mddev_ident *devs, *mddev;
1310         int rv = 0;
1311         char container[32];
1312         char *only = NULL;
1313
1314         map_read(&mapl);
1315         devs = conf_get_ident(NULL);
1316
1317 restart:
1318         for (me = mapl ; me ; me = me->next) {
1319                 struct mdinfo *sra;
1320                 int mdfd;
1321
1322                 if (devnm && strcmp(devnm, me->devnm) != 0)
1323                         continue;
1324                 if (me->metadata[0] == '/') {
1325                         char *sl;
1326
1327                         if (!devnm)
1328                                 continue;
1329
1330                         /* member array, need to work on container */
1331                         strncpy(container, me->metadata+1, 32);
1332                         container[31] = 0;
1333                         sl = strchr(container, '/');
1334                         if (sl)
1335                                 *sl = 0;
1336                         only = devnm;
1337                         devnm = container;
1338                         goto restart;
1339                 }
1340                 mdfd = open_dev(me->devnm);
1341
1342                 if (mdfd < 0)
1343                         continue;
1344                 if (!isdigit(me->metadata[0])) {
1345                         /* must be a container */
1346                         struct supertype *st = super_by_fd(mdfd, NULL);
1347                         int ret = 0;
1348                         struct map_ent *map = NULL;
1349
1350                         if (st && st->ss->load_container)
1351                                 ret = st->ss->load_container(st, mdfd, NULL);
1352                         close(mdfd);
1353                         if (!ret && st && st->ss->container_content) {
1354                                 if (map_lock(&map))
1355                                         pr_err("failed to get exclusive lock on mapfile\n");
1356                                 ret = Incremental_container(st, me->path, c, only);
1357                                 map_unlock(&map);
1358                         }
1359                         if (ret)
1360                                 rv = 1;
1361                         continue;
1362                 }
1363                 if (md_array_active(mdfd)) {
1364                         close(mdfd);
1365                         continue;
1366                 }
1367                 /* Ok, we can try this one.   Maybe it needs a bitmap */
1368                 for (mddev = devs ; mddev ; mddev = mddev->next)
1369                         if (mddev->devname && me->path &&
1370                             devname_matches(mddev->devname, me->path))
1371                                 break;
1372                 if (mddev && mddev->bitmap_file) {
1373                         /*
1374                          * Note: early kernels will wrongly fail this, so it
1375                          * is a hint only
1376                          */
1377                         int added = -1;
1378                         int bmfd;
1379
1380                         bmfd = open(mddev->bitmap_file, O_RDWR);
1381                         if (bmfd >= 0) {
1382                                 added = ioctl(mdfd, SET_BITMAP_FILE, bmfd);
1383                                 close(bmfd);
1384                         }
1385                         if (c->verbose >= 0) {
1386                                 if (added == 0)
1387                                         pr_err("Added bitmap %s to %s\n",
1388                                                mddev->bitmap_file, me->path);
1389                                 else if (errno != EEXIST)
1390                                         pr_err("Failed to add bitmap to %s: %s\n",
1391                                                me->path, strerror(errno));
1392                         }
1393                 }
1394                 /* FIXME check for reshape_active and consider not
1395                  * starting array.
1396                  */
1397                 sra = sysfs_read(mdfd, NULL, 0);
1398                 if (sra) {
1399                         if (sysfs_set_str(sra, NULL,
1400                                           "array_state", "read-auto") == 0) {
1401                                 if (c->verbose >= 0)
1402                                         pr_err("started array %s\n",
1403                                                me->path ?: me->devnm);
1404                         } else {
1405                                 pr_err("failed to start array %s: %s\n",
1406                                        me->path ?: me->devnm,
1407                                        strerror(errno));
1408                                 rv = 1;
1409                         }
1410                         sysfs_free(sra);
1411                 }
1412         }
1413         return rv;
1414 }
1415
1416 static char *container2devname(char *devname)
1417 {
1418         char *mdname = NULL;
1419
1420         if (devname[0] == '/') {
1421                 int fd = open(devname, O_RDONLY);
1422                 if (fd >= 0) {
1423                         mdname = xstrdup(fd2devnm(fd));
1424                         close(fd);
1425                 }
1426         } else {
1427                 int uuid[4];
1428                 struct map_ent *mp, *map = NULL;
1429
1430                 if (!parse_uuid(devname, uuid))
1431                         return mdname;
1432                 mp = map_by_uuid(&map, uuid);
1433                 if (mp)
1434                         mdname = xstrdup(mp->devnm);
1435                 map_free(map);
1436         }
1437
1438         return mdname;
1439 }
1440
1441 static int Incremental_container(struct supertype *st, char *devname,
1442                                  struct context *c, char *only)
1443 {
1444         /* Collect the contents of this container and for each
1445          * array, choose a device name and assemble the array.
1446          */
1447
1448         struct mdinfo *list;
1449         struct mdinfo *ra;
1450         struct map_ent *map = NULL;
1451         struct mdinfo info;
1452         int trustworthy;
1453         struct mddev_ident *match;
1454         int rv = 0;
1455         struct domainlist *domains;
1456         struct map_ent *smp;
1457         int suuid[4];
1458         int sfd;
1459         int ra_blocked = 0;
1460         int ra_all = 0;
1461         int result = 0;
1462
1463         st->ss->getinfo_super(st, &info, NULL);
1464
1465         if ((c->runstop > 0 && info.container_enough >= 0) ||
1466             info.container_enough > 0)
1467                 /* pass */;
1468         else {
1469                 if (c->export) {
1470                         printf("MD_STARTED=no\n");
1471                 } else if (c->verbose)
1472                         pr_err("not enough devices to start the container\n");
1473                 return 0;
1474         }
1475
1476         match = conf_match(st, &info, devname, c->verbose, &rv);
1477         if (match == NULL && rv == 2)
1478                 return rv;
1479
1480         /* Need to compute 'trustworthy' */
1481         if (match)
1482                 trustworthy = LOCAL;
1483         else if (st->ss->match_home(st, c->homehost) == 1)
1484                 trustworthy = LOCAL;
1485         else if (st->ss->match_home(st, "any") == 1)
1486                 trustworthy = LOCAL;
1487         else
1488                 trustworthy = FOREIGN;
1489
1490         list = st->ss->container_content(st, NULL);
1491         /* when nothing to activate - quit */
1492         if (list == NULL) {
1493                 if (c->export) {
1494                         printf("MD_STARTED=nothing\n");
1495                 }
1496                 return 0;
1497         }
1498         for (ra = list ; ra ; ra = ra->next) {
1499                 int mdfd;
1500                 char chosen_name[1024];
1501                 struct map_ent *mp;
1502                 struct mddev_ident *match = NULL;
1503
1504                 ra_all++;
1505                 /* do not activate arrays blocked by metadata handler */
1506                 if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) {
1507                         pr_err("Cannot activate array %s in %s.\n",
1508                                 ra->text_version, devname);
1509                         ra_blocked++;
1510                         continue;
1511                 }
1512                 mp = map_by_uuid(&map, ra->uuid);
1513
1514                 if (mp) {
1515                         mdfd = open_dev(mp->devnm);
1516                         if (mp->path)
1517                                 strcpy(chosen_name, mp->path);
1518                         else
1519                                 strcpy(chosen_name, mp->devnm);
1520                 } else if (!only) {
1521
1522                         /* Check in mdadm.conf for container == devname and
1523                          * member == ra->text_version after second slash.
1524                          */
1525                         char *sub = strchr(ra->text_version+1, '/');
1526                         struct mddev_ident *array_list;
1527                         if (sub) {
1528                                 sub++;
1529                                 array_list = conf_get_ident(NULL);
1530                         } else
1531                                 array_list = NULL;
1532                         for(; array_list ; array_list = array_list->next) {
1533                                 char *dn;
1534                                 if (array_list->member == NULL ||
1535                                     array_list->container == NULL)
1536                                         continue;
1537                                 if (strcmp(array_list->member, sub) != 0)
1538                                         continue;
1539                                 if (array_list->uuid_set &&
1540                                     !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1541                                         continue;
1542                                 dn = container2devname(array_list->container);
1543                                 if (dn == NULL)
1544                                         continue;
1545                                 if (strncmp(dn, ra->text_version+1,
1546                                             strlen(dn)) != 0 ||
1547                                     ra->text_version[strlen(dn)+1] != '/') {
1548                                         free(dn);
1549                                         continue;
1550                                 }
1551                                 free(dn);
1552                                 /* we have a match */
1553                                 match = array_list;
1554                                 if (c->verbose>0)
1555                                         pr_err("match found for member %s\n",
1556                                                 array_list->member);
1557                                 break;
1558                         }
1559
1560                         if (match && match->devname &&
1561                             strcasecmp(match->devname, "<ignore>") == 0) {
1562                                 if (c->verbose > 0)
1563                                         pr_err("array %s/%s is explicitly ignored by mdadm.conf\n",
1564                                                match->container, match->member);
1565                                 continue;
1566                         }
1567                         if (match)
1568                                 trustworthy = LOCAL;
1569
1570                         mdfd = create_mddev(match ? match->devname : NULL,
1571                                             ra->name,
1572                                             c->autof,
1573                                             trustworthy,
1574                                             chosen_name, 0);
1575                 }
1576                 if (only && (!mp || strcmp(mp->devnm, only) != 0))
1577                         continue;
1578
1579                 if (mdfd < 0) {
1580                         pr_err("failed to open %s: %s.\n",
1581                                 chosen_name, strerror(errno));
1582                         return 2;
1583                 }
1584
1585                 assemble_container_content(st, mdfd, ra, c,
1586                                            chosen_name, &result);
1587                 close(mdfd);
1588         }
1589         if (c->export && result) {
1590                 char sep = '=';
1591                 printf("MD_STARTED");
1592                 if (result & INCR_NO) {
1593                         printf("%cno", sep);
1594                         sep = ',';
1595                 }
1596                 if (result & INCR_UNSAFE) {
1597                         printf("%cunsafe", sep);
1598                         sep = ',';
1599                 }
1600                 if (result & INCR_ALREADY) {
1601                         printf("%calready", sep);
1602                         sep = ',';
1603                 }
1604                 if (result & INCR_YES) {
1605                         printf("%cyes", sep);
1606                         sep = ',';
1607                 }
1608                 printf("\n");
1609         }
1610
1611         /* don't move spares to container with volume being activated
1612            when all volumes are blocked */
1613         if (ra_all == ra_blocked)
1614                 return 0;
1615
1616         /* Now move all suitable spares from spare container */
1617         domains = domain_from_array(list, st->ss->name);
1618         memcpy(suuid, uuid_zero, sizeof(int[4]));
1619         if (domains &&
1620             (smp = map_by_uuid(&map, suuid)) != NULL &&
1621             (sfd = open(smp->path, O_RDONLY)) >= 0) {
1622                 /* spare container found */
1623                 struct supertype *sst =
1624                         super_imsm.match_metadata_desc("imsm");
1625                 struct mdinfo *sinfo;
1626
1627                 if (!sst->ss->load_container(sst, sfd, NULL)) {
1628                         struct spare_criteria sc = {0, 0};
1629
1630                         if (st->ss->get_spare_criteria)
1631                                 st->ss->get_spare_criteria(st, &sc);
1632
1633                         close(sfd);
1634                         sinfo = container_choose_spares(sst, &sc,
1635                                                         domains, NULL,
1636                                                         st->ss->name, 0);
1637                         sst->ss->free_super(sst);
1638                         if (sinfo){
1639                                 int count = 0;
1640                                 struct mdinfo *disks = sinfo->devs;
1641                                 while (disks) {
1642                                         /* move spare from spare
1643                                          * container to currently
1644                                          * assembled one
1645                                          */
1646                                         if (move_spare(
1647                                                     smp->path,
1648                                                     devname,
1649                                                     makedev(disks->disk.major,
1650                                                             disks->disk.minor)))
1651                                                 count++;
1652                                         disks = disks->next;
1653                                 }
1654                                 if (count)
1655                                         pr_err("Added %d spare%s to %s\n",
1656                                                count, count>1?"s":"", devname);
1657                         }
1658                         sysfs_free(sinfo);
1659                 } else
1660                         close(sfd);
1661         }
1662         domain_free(domains);
1663         return 0;
1664 }
1665
1666 static void run_udisks(char *arg1, char *arg2)
1667 {
1668         int pid = fork();
1669         int status;
1670         if (pid == 0) {
1671                 execl("/usr/bin/udisks", "udisks", arg1, arg2, NULL);
1672                 execl("/bin/udisks", "udisks", arg1, arg2, NULL);
1673                 exit(1);
1674         }
1675         while (pid > 0 && wait(&status) != pid)
1676                 ;
1677 }
1678
1679 /*
1680  * IncrementalRemove - Attempt to see if the passed in device belongs to any
1681  * raid arrays, and if so first fail (if needed) and then remove the device.
1682  *
1683  * @devname - The device we want to remove
1684  * @id_path - name as found in /dev/disk/by-path for this device
1685  *
1686  * Note: the device name must be a kernel name like "sda", so
1687  * that we can find it in /proc/mdstat
1688  */
1689 int IncrementalRemove(char *devname, char *id_path, int verbose)
1690 {
1691         int mdfd;
1692         int rv = 0;
1693         struct mdstat_ent *ent;
1694         struct mddev_dev devlist;
1695         struct mdinfo mdi;
1696         char buf[32];
1697
1698         if (!id_path)
1699                 dprintf("incremental removal without --path <id_path> lacks the possibility to re-add new device in this port\n");
1700
1701         if (strchr(devname, '/')) {
1702                 pr_err("incremental removal requires a kernel device name, not a file: %s\n", devname);
1703                 return 1;
1704         }
1705         ent = mdstat_by_component(devname);
1706         if (!ent) {
1707                 if (verbose >= 0)
1708                         pr_err("%s does not appear to be a component of any array\n", devname);
1709                 return 1;
1710         }
1711         if (sysfs_init(&mdi, -1, ent->devnm)) {
1712                 pr_err("unable to initialize sysfs for: %s\n", devname);
1713                 return 1;
1714         }
1715         mdfd = open_dev_excl(ent->devnm);
1716         if (mdfd > 0) {
1717                 close(mdfd);
1718                 if (sysfs_get_str(&mdi, NULL, "array_state",
1719                                   buf, sizeof(buf)) > 0) {
1720                         if (strncmp(buf, "active", 6) == 0 ||
1721                             strncmp(buf, "clean", 5) == 0)
1722                                 sysfs_set_str(&mdi, NULL,
1723                                               "array_state", "read-auto");
1724                 }
1725         }
1726         mdfd = open_dev(ent->devnm);
1727         if (mdfd < 0) {
1728                 if (verbose >= 0)
1729                         pr_err("Cannot open array %s!!\n", ent->devnm);
1730                 free_mdstat(ent);
1731                 return 1;
1732         }
1733
1734         if (id_path) {
1735                 struct map_ent *map = NULL, *me;
1736                 me = map_by_devnm(&map, ent->devnm);
1737                 if (me)
1738                         policy_save_path(id_path, me);
1739                 map_free(map);
1740         }
1741
1742         memset(&devlist, 0, sizeof(devlist));
1743         devlist.devname = devname;
1744         devlist.disposition = 'f';
1745         /* for a container, we must fail each member array */
1746         if (ent->metadata_version &&
1747             strncmp(ent->metadata_version, "external:", 9) == 0) {
1748                 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1749                 struct mdstat_ent *memb;
1750                 for (memb = mdstat ; memb ; memb = memb->next)
1751                         if (is_container_member(memb, ent->devnm)) {
1752                                 int subfd = open_dev(memb->devnm);
1753                                 if (subfd >= 0) {
1754                                         rv |= Manage_subdevs(
1755                                                 memb->devnm, subfd,
1756                                                 &devlist, verbose, 0,
1757                                                 NULL, 0);
1758                                         close(subfd);
1759                                 }
1760                         }
1761                 free_mdstat(mdstat);
1762         } else
1763                 rv |= Manage_subdevs(ent->devnm, mdfd, &devlist,
1764                                     verbose, 0, NULL, 0);
1765         if (rv & 2) {
1766                 /* Failed due to EBUSY, try to stop the array.
1767                  * Give udisks a chance to unmount it first.
1768                  */
1769                 int devid = devnm2devid(ent->devnm);
1770                 run_udisks("--unmount", map_dev(major(devid),minor(devid), 0));
1771                 rv = Manage_stop(ent->devnm, mdfd, verbose, 1);
1772                 if (rv)
1773                         /* At least we can try to trigger a 'remove' */
1774                         sysfs_uevent(&mdi, "remove");
1775                 if (verbose) {
1776                         if (rv)
1777                                 pr_err("Fail to stop %s too.\n", ent->devnm);
1778                 }
1779         } else {
1780                 devlist.disposition = 'r';
1781                 rv = Manage_subdevs(ent->devnm, mdfd, &devlist,
1782                                     verbose, 0, NULL, 0);
1783         }
1784         close(mdfd);
1785         free_mdstat(ent);
1786         return rv;
1787 }