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         struct stat stb;
  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         mdu_array_info_t ainf;
 103         struct dev_policy *policy = NULL;
 104         struct map_ent target_array;
 105         int have_target;
 106         char *devname = devlist->devname;
 107         int journal_device_missing = 0;
 108
 109         struct createinfo *ci = conf_get_create_info();
 110
 111         if (stat(devname, &stb) < 0) {
 112                 if (c->verbose >= 0)
 113                         pr_err("stat failed for %s: %s.\n",
 114                                 devname, strerror(errno));
 115                 return rv;
 116         }
 117         if ((stb.st_mode & S_IFMT) != S_IFBLK) {
 118                 if (c->verbose >= 0)
 119                         pr_err("%s is not a block device.\n",
 120                                 devname);
 121                 return rv;
 122         }
 123         dfd = dev_open(devname, O_RDONLY);
 124         if (dfd < 0) {
 125                 if (c->verbose >= 0)
 126                         pr_err("cannot open %s: %s.\n",
 127                                 devname, strerror(errno));
 128                 return rv;
 129         }
 130         /* If the device is a container, we do something very different */
 131         if (must_be_container(dfd)) {
 132                 if (!st)
 133                         st = super_by_fd(dfd, NULL);
 134                 if (st && st->ss->load_container)
 135                         rv = st->ss->load_container(st, dfd, NULL);
 136
 137                 close(dfd);
 138                 if (!rv && st->ss->container_content) {
 139                         if (map_lock(&map))
 140                                 pr_err("failed to get exclusive lock on mapfile\n");
 141                         if (c->export)
 142                                 printf("MD_DEVNAME=%s\n", devname);
 143                         rv = Incremental_container(st, devname, c, NULL);
 144                         map_unlock(&map);
 145                         return rv;
 146                 }
 147
 148                 pr_err("%s is not part of an md array.\n",
 149                         devname);
 150                 return rv;
 151         }
 152
 153         /* 1/ Check if device is permitted by mdadm.conf */
 154
 155         for (;devlist; devlist = devlist->next)
 156                 if (conf_test_dev(devlist->devname))
 157                         break;
 158         if (!devlist) {
 159                 devlist = conf_get_devs();
 160                 for (;devlist; devlist = devlist->next) {
 161                         struct stat st2;
 162                         if (stat(devlist->devname, &st2) == 0 &&
 163                             (st2.st_mode & S_IFMT) == S_IFBLK &&
 164                             st2.st_rdev == stb.st_rdev)
 165                                 break;
 166                 }
 167         }
 168         if (!devlist) {
 169                 if (c->verbose >= 0)
 170                         pr_err("%s not permitted by mdadm.conf.\n",
 171                                devname);
 172                 goto out;
 173         }
 174
 175         /* 2/ Find metadata, reject if none appropriate (check
 176          *            version/name from args) */
 177
 178         if (fstat(dfd, &stb) < 0) {
 179                 if (c->verbose >= 0)
 180                         pr_err("fstat failed for %s: %s.\n",
 181                                 devname, strerror(errno));
 182                 goto out;
 183         }
 184         if ((stb.st_mode & S_IFMT) != S_IFBLK) {
 185                 if (c->verbose >= 0)
 186                         pr_err("%s is not a block device.\n",
 187                                 devname);
 188                 goto out;
 189         }
 190
 191         dinfo.disk.major = major(stb.st_rdev);
 192         dinfo.disk.minor = minor(stb.st_rdev);
 193
 194         policy = disk_policy(&dinfo);
 195         have_target = policy_check_path(&dinfo, &target_array);
 196
 197         if (st == NULL && (st = guess_super_type(dfd, guess_array)) == NULL) {
 198                 if (c->verbose >= 0)
 199                         pr_err("no recognisable superblock on %s.\n",
 200                                devname);
 201                 rv = try_spare(devname, &dfd, policy,
 202                                have_target ? &target_array : NULL,
 203                                NULL, c->verbose);
 204                 goto out;
 205         }
 206         st->ignore_hw_compat = 0;
 207
 208         if (st->ss->compare_super == NULL ||
 209             st->ss->load_super(st, dfd, c->verbose >= 0 ? devname : NULL)) {
 210                 if (c->verbose >= 0)
 211                         pr_err("no RAID superblock on %s.\n",
 212                                 devname);
 213                 rv = try_spare(devname, &dfd, policy,
 214                                have_target ? &target_array : NULL,
 215                                st, c->verbose);
 216                 free(st);
 217                 goto out;
 218         }
 219         close (dfd); dfd = -1;
 220
 221         st->ss->getinfo_super(st, &info, NULL);
 222
 223         /* 3/ Check if there is a match in mdadm.conf */
 224         match = conf_match(st, &info, devname, c->verbose, &rv);
 225         if (!match && rv == 2)
 226                 goto out;
 227
 228         if (match && match->devname
 229             && strcasecmp(match->devname, "<ignore>") == 0) {
 230                 if (c->verbose >= 0)
 231                         pr_err("array containing %s is explicitly ignored by mdadm.conf\n",
 232                                 devname);
 233                 goto out;
 234         }
 235
 236         /* 3a/ if not, check for homehost match.  If no match, continue
 237          * but don't trust the 'name' in the array. Thus a 'random' minor
 238          * number will be assigned, and the device name will be based
 239          * on that. */
 240         if (match)
 241                 trustworthy = LOCAL;
 242         else if (st->ss->match_home(st, c->homehost) == 1)
 243                 trustworthy = LOCAL;
 244         else if (st->ss->match_home(st, "any") == 1)
 245                 trustworthy = LOCAL_ANY;
 246         else
 247                 trustworthy = FOREIGN;
 248
 249         if (!match && !conf_test_metadata(st->ss->name, policy,
 250                                           (trustworthy == LOCAL))) {
 251                 if (c->verbose >= 1)
 252                         pr_err("%s has metadata type %s for which auto-assembly is disabled\n",
 253                                devname, st->ss->name);
 254                 goto out;
 255         }
 256         if (trustworthy == LOCAL_ANY)
 257                 trustworthy = LOCAL;
 258
 259         /* There are three possible sources for 'autof':  command line,
 260          * ARRAY line in mdadm.conf, or CREATE line in mdadm.conf.
 261          * ARRAY takes precedence, then command line, then
 262          * CREATE.
 263          */
 264         if (match && match->autof)
 265                 c->autof = match->autof;
 266         if (c->autof == 0)
 267                 c->autof = ci->autof;
 268
 269         name_to_use = info.name;
 270         if (name_to_use[0] == 0 &&
 271             info.array.level == LEVEL_CONTAINER) {
 272                 name_to_use = info.text_version;
 273                 trustworthy = METADATA;
 274         }
 275         if (name_to_use[0] && trustworthy != LOCAL &&
 276             ! c->require_homehost &&
 277             conf_name_is_free(name_to_use))
 278                 trustworthy = LOCAL;
 279
 280         /* strip "hostname:" prefix from name if we have decided
 281          * to treat it as LOCAL
 282          */
 283         if (trustworthy == LOCAL && strchr(name_to_use, ':') != NULL)
 284                 name_to_use = strchr(name_to_use, ':')+1;
 285
 286         /* 4/ Check if array exists.
 287          */
 288         if (map_lock(&map))
 289                 pr_err("failed to get exclusive lock on mapfile\n");
 290         /* Now check we can get O_EXCL.  If not, probably "mdadm -A" has
 291          * taken over
 292          */
 293         dfd = dev_open(devname, O_RDONLY|O_EXCL);
 294         if (dfd < 0) {
 295                 if (c->verbose >= 0)
 296                         pr_err("cannot reopen %s: %s.\n",
 297                                 devname, strerror(errno));
 298                 goto out_unlock;
 299         }
 300         /* Cannot hold it open while we add the device to the array,
 301          * so we must release the O_EXCL and depend on the map_lock()
 302          * So now is the best time to remove any partitions.
 303          */
 304         remove_partitions(dfd);
 305         close(dfd);
 306         dfd = -1;
 307
 308         mp = map_by_uuid(&map, info.uuid);
 309         if (mp)
 310                 mdfd = open_dev(mp->devnm);
 311         else
 312                 mdfd = -1;
 313
 314         if (mdfd < 0) {
 315
 316                 /* Skip the clustered ones. This should be started by
 317                  * clustering resource agents
 318                  */
 319                 if (info.array.state & (1 << MD_SB_CLUSTERED))
 320                         goto out;
 321
 322                 /* Couldn't find an existing array, maybe make a new one */
 323                 mdfd = create_mddev(match ? match->devname : NULL,
 324                                     name_to_use, c->autof, trustworthy, chosen_name);
 325
 326                 if (mdfd < 0)
 327                         goto out_unlock;
 328
 329                 sysfs_init(&info, mdfd, NULL);
 330
 331                 if (set_array_info(mdfd, st, &info) != 0) {
 332                         pr_err("failed to set array info for %s: %s\n",
 333                                 chosen_name, strerror(errno));
 334                         rv = 2;
 335                         goto out_unlock;
 336                 }
 337
 338                 dinfo = info;
 339                 dinfo.disk.major = major(stb.st_rdev);
 340                 dinfo.disk.minor = minor(stb.st_rdev);
 341                 if (add_disk(mdfd, st, &info, &dinfo) != 0) {
 342                         pr_err("failed to add %s to new array %s: %s.\n",
 343                                 devname, chosen_name, strerror(errno));
 344                         ioctl(mdfd, STOP_ARRAY, 0);
 345                         rv = 2;
 346                         goto out_unlock;
 347                 }
 348                 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
 349                                               GET_OFFSET | GET_SIZE));
 350
 351                 if (!sra || !sra->devs || sra->devs->disk.raid_disk >= 0) {
 352                         /* It really should be 'none' - must be old buggy
 353                          * kernel, and mdadm -I may not be able to complete.
 354                          * So reject it.
 355                          */
 356                         ioctl(mdfd, STOP_ARRAY, NULL);
 357                         pr_err("You have an old buggy kernel which cannot support\n      --incremental reliably.  Aborting.\n");
 358                         rv = 2;
 359                         goto out_unlock;
 360                 }
 361                 info.array.working_disks = 1;
 362                 /* 6/ Make sure /var/run/mdadm.map contains this array. */
 363                 map_update(&map, fd2devnm(mdfd),
 364                            info.text_version,
 365                            info.uuid, chosen_name);
 366         } else {
 367         /* 5b/ if it does */
 368         /* - check one drive in array to make sure metadata is a reasonably */
 369         /*        close match.  Reject if not (e.g. different type) */
 370         /* - add the device */
 371                 char dn[20];
 372                 int dfd2;
 373                 int err;
 374                 struct supertype *st2;
 375                 struct mdinfo info2, *d;
 376
 377                 sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
 378                                             GET_OFFSET | GET_SIZE));
 379
 380                 if (mp->path)
 381                         strcpy(chosen_name, mp->path);
 382                 else
 383                         strcpy(chosen_name, mp->devnm);
 384
 385                 /* It is generally not OK to add non-spare drives to a
 386                  * running array as they are probably missing because
 387                  * they failed.  However if runstop is 1, then the
 388                  * array was possibly started early and our best bet is
 389                  * to add this anyway.
 390                  * Also if action policy is re-add or better we allow
 391                  * re-add.
 392                  * This doesn't apply to containers as the 'non-spare'
 393                  * flag has a different meaning.  The test has to happen
 394                  * at the device level there
 395                  */
 396                 if (!st->ss->external
 397                     && (info.disk.state & (1<<MD_DISK_SYNC)) != 0
 398                     && ! policy_action_allows(policy, st->ss->name,
 399                                               act_re_add)
 400                     && c->runstop < 1) {
 401                         if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
 402                                 pr_err("not adding %s to active array (without --run) %s\n",
 403                                        devname, chosen_name);
 404                                 rv = 2;
 405                                 goto out_unlock;
 406                         }
 407                 }
 408                 if (!sra) {
 409                         rv = 2;
 410                         goto out_unlock;
 411                 }
 412                 if (sra->devs) {
 413                         sprintf(dn, "%d:%d", sra->devs->disk.major,
 414                                 sra->devs->disk.minor);
 415                         dfd2 = dev_open(dn, O_RDONLY);
 416                         if (dfd2 < 0) {
 417                                 pr_err("unable to open %s\n", devname);
 418                                 rv = 2;
 419                                 goto out_unlock;
 420                         }
 421                         st2 = dup_super(st);
 422                         if (st2->ss->load_super(st2, dfd2, NULL) ||
 423                             st->ss->compare_super(st, st2) != 0) {
 424                                 pr_err("metadata mismatch between %s and chosen array %s\n",
 425                                        devname, chosen_name);
 426                                 close(dfd2);
 427                                 rv = 2;
 428                                 goto out_unlock;
 429                         }
 430                         close(dfd2);
 431                         st2->ss->getinfo_super(st2, &info2, NULL);
 432                         st2->ss->free_super(st2);
 433                         if (info.array.level != info2.array.level ||
 434                             memcmp(info.uuid, info2.uuid, 16) != 0 ||
 435                             info.array.raid_disks != info2.array.raid_disks) {
 436                                 pr_err("unexpected difference between %s and %s.\n",
 437                                        chosen_name, devname);
 438                                 rv = 2;
 439                                 goto out_unlock;
 440                         }
 441                 }
 442                 info.disk.major = major(stb.st_rdev);
 443                 info.disk.minor = minor(stb.st_rdev);
 444                 /* add disk needs to know about containers */
 445                 if (st->ss->external)
 446                         sra->array.level = LEVEL_CONTAINER;
 447
 448                 if (info.array.state & (1 << MD_SB_CLUSTERED))
 449                         info.disk.state |= (1 << MD_DISK_CLUSTER_ADD);
 450
 451                 err = add_disk(mdfd, st, sra, &info);
 452                 if (err < 0 && errno == EBUSY) {
 453                         /* could be another device present with the same
 454                          * disk.number. Find and reject any such
 455                          */
 456                         find_reject(mdfd, st, sra, info.disk.number,
 457                                     info.events, c->verbose, chosen_name);
 458                         err = add_disk(mdfd, st, sra, &info);
 459                 }
 460                 if (err < 0 && errno == EINVAL &&
 461                     info.disk.state & (1<<MD_DISK_SYNC)) {
 462                         /* Maybe it needs to be added as a spare */
 463                         if (policy_action_allows(policy, st->ss->name,
 464                                                  act_force_spare)) {
 465                                 info.disk.state &= ~(1<<MD_DISK_SYNC);
 466                                 err = add_disk(mdfd, st, sra, &info);
 467                         } else
 468                                 if (c->verbose >= 0)
 469                                         pr_err("can only add %s to %s as a spare, and force-spare is not set.\n",
 470                                                devname, chosen_name);
 471                 }
 472                 if (err < 0) {
 473                         pr_err("failed to add %s to existing array %s: %s.\n",
 474                                 devname, chosen_name, strerror(errno));
 475                         rv = 2;
 476                         goto out_unlock;
 477                 }
 478                 info.array.working_disks = 0;
 479                 for (d = sra->devs; d; d=d->next)
 480                         info.array.working_disks ++;
 481
 482         }
 483         if (strncmp(chosen_name, "/dev/md/", 8) == 0)
 484                 md_devname = chosen_name+8;
 485         else
 486                 md_devname = chosen_name;
 487         if (c->export) {
 488                 printf("MD_DEVICE=%s\n", fd2devnm(mdfd));
 489                 printf("MD_DEVNAME=%s\n", md_devname);
 490                 printf("MD_FOREIGN=%s\n", trustworthy == FOREIGN ? "yes" : "no");
 491         }
 492
 493         /* 7/ Is there enough devices to possibly start the array? */
 494         /* 7a/ if not, finish with success. */
 495         if (info.array.level == LEVEL_CONTAINER) {
 496                 char devnm[32];
 497                 /* Try to assemble within the container */
 498                 sysfs_uevent(sra, "change");
 499                 if (!c->export && c->verbose >= 0)
 500                         pr_err("container %s now has %d device%s\n",
 501                                chosen_name, info.array.working_disks,
 502                                info.array.working_disks == 1?"":"s");
 503                 wait_for(chosen_name, mdfd);
 504                 if (st->ss->external)
 505                         strcpy(devnm, fd2devnm(mdfd));
 506                 if (st->ss->load_container)
 507                         rv = st->ss->load_container(st, mdfd, NULL);
 508                 close(mdfd);
 509                 sysfs_free(sra);
 510                 if (!rv)
 511                         rv = Incremental_container(st, chosen_name, c, NULL);
 512                 map_unlock(&map);
 513                 /* after spare is added, ping monitor for external metadata
 514                  * so that it can eg. try to rebuild degraded array */
 515                 if (st->ss->external)
 516                         ping_monitor(devnm);
 517                 return rv;
 518         }
 519
 520         /* We have added something to the array, so need to re-read the
 521          * state.  Eventually this state should be kept up-to-date as
 522          * things change.
 523          */
 524         sysfs_free(sra);
 525         sra = sysfs_read(mdfd, NULL, (GET_DEVS | GET_STATE |
 526                                     GET_OFFSET | GET_SIZE));
 527         active_disks = count_active(st, sra, mdfd, &avail, &info);
 528
 529         journal_device_missing = (info.journal_device_required) && (info.journal_clean == 0);
 530
 531         if (enough(info.array.level, info.array.raid_disks,
 532                    info.array.layout, info.array.state & 1,
 533                    avail) == 0) {
 534                 if (c->export) {
 535                         printf("MD_STARTED=no\n");
 536                 } else if (c->verbose >= 0)
 537                         pr_err("%s attached to %s, not enough to start (%d).\n",
 538                                devname, chosen_name, active_disks);
 539                 rv = 0;
 540                 goto out_unlock;
 541         }
 542
 543         /* 7b/ if yes, */
 544         /* - if number of OK devices match expected, or -R and there */
 545         /*             are enough, */
 546         /*   + add any bitmap file  */
 547         /*   + start the array (auto-readonly). */
 548
 549         if (ioctl(mdfd, GET_ARRAY_INFO, &ainf) == 0) {
 550                 if (c->export) {
 551                         printf("MD_STARTED=already\n");
 552                 } else if (c->verbose >= 0)
 553                         pr_err("%s attached to %s which is already active.\n",
 554                                devname, chosen_name);
 555                 rv = 0;
 556                 goto out_unlock;
 557         }
 558
 559         map_unlock(&map);
 560         if (c->runstop > 0 || (!journal_device_missing && active_disks >= info.array.working_disks)) {
 561                 struct mdinfo *dsk;
 562                 /* Let's try to start it */
 563
 564                 if (journal_device_missing)
 565                         pr_err("Trying to run with missing journal device\n");
 566                 if (info.reshape_active && !(info.reshape_active & RESHAPE_NO_BACKUP)) {
 567                         pr_err("%s: This array is being reshaped and cannot be started\n",
 568                                chosen_name);
 569                         cont_err("by --incremental.  Please use --assemble\n");
 570                         goto out;
 571                 }
 572                 if (match && match->bitmap_file) {
 573                         int bmfd = open(match->bitmap_file, O_RDWR);
 574                         if (bmfd < 0) {
 575                                 pr_err("Could not open bitmap file %s.\n",
 576                                        match->bitmap_file);
 577                                 goto out;
 578                         }
 579                         if (ioctl(mdfd, SET_BITMAP_FILE, bmfd) != 0) {
 580                                 close(bmfd);
 581                                 pr_err("Failed to set bitmapfile for %s.\n",
 582                                        chosen_name);
 583                                 goto out;
 584                         }
 585                         close(bmfd);
 586                 }
 587                 /* Need to remove from the array any devices which
 588                  * 'count_active' discerned were too old or inappropriate
 589                  */
 590                 for (d = sra ? sra->devs : NULL ; d ; d = d->next)
 591                         if (d->disk.state & (1<<MD_DISK_REMOVED))
 592                                 remove_disk(mdfd, st, sra, d);
 593
 594                 if ((sra == NULL || active_disks >= info.array.working_disks)
 595                     && trustworthy != FOREIGN)
 596                         rv = ioctl(mdfd, RUN_ARRAY, NULL);
 597                 else
 598                         rv = sysfs_set_str(sra, NULL,
 599                                            "array_state", "read-auto");
 600                 /* Array might be O_EXCL which  will interfere with
 601                  * fsck and mount.  So re-open without O_EXCL.
 602                  */
 603                 reopen_mddev(mdfd);
 604                 if (rv == 0) {
 605                         if (c->export) {
 606                                 printf("MD_STARTED=yes\n");
 607                         } else if (c->verbose >= 0)
 608                                 pr_err("%s attached to %s, which has been started.\n",
 609                                        devname, chosen_name);
 610                         rv = 0;
 611                         wait_for(chosen_name, mdfd);
 612                         /* We just started the array, so some devices
 613                          * might have been evicted from the array
 614                          * because their event counts were too old.
 615                          * If the action=re-add policy is in-force for
 616                          * those devices we should re-add them now.
 617                          */
 618                         for (dsk = sra->devs; dsk ; dsk = dsk->next) {
 619                                 if (disk_action_allows(dsk, st->ss->name, act_re_add) &&
 620                                     add_disk(mdfd, st, sra, dsk) == 0)
 621                                         pr_err("%s re-added to %s\n",
 622                                                dsk->sys_name, chosen_name);
 623                         }
 624                 } else {
 625                         pr_err("%s attached to %s, but failed to start: %s.\n",
 626                                devname, chosen_name, strerror(errno));
 627                         rv = 1;
 628                 }
 629         } else {
 630                 if (c->export) {
 631                         printf("MD_STARTED=unsafe\n");
 632                 } else if (journal_device_missing) {
 633                         pr_err("Journal device is missing, not safe to start yet.\n");
 634                 } else if (c->verbose >= 0)
 635                         pr_err("%s attached to %s, not enough to start safely.\n",
 636                                devname, chosen_name);
 637                 rv = 0;
 638         }
 639 out:
 640         free(avail);
 641         if (dfd >= 0)
 642                 close(dfd);
 643         if (mdfd >= 0)
 644                 close(mdfd);
 645         if (policy)
 646                 dev_policy_free(policy);
 647         sysfs_free(sra);
 648         return rv;
 649 out_unlock:
 650         map_unlock(&map);
 651         goto out;
 652 }
 653
 654 static void find_reject(int mdfd, struct supertype *st, struct mdinfo *sra,
 655                         int number, __u64 events, int verbose,
 656                         char *array_name)
 657 {
 658         /* Find a device attached to this array with a disk.number of number
 659          * and events less than the passed events, and remove the device.
 660          */
 661         struct mdinfo *d;
 662         mdu_array_info_t ra;
 663
 664         if (ioctl(mdfd, GET_ARRAY_INFO, &ra) == 0)
 665                 return; /* not safe to remove from active arrays
 666                          * without thinking more */
 667
 668         for (d = sra->devs; d ; d = d->next) {
 669                 char dn[24]; // 2*11 bytes for ints (including sign) + colon + null byte
 670                 int dfd;
 671                 struct mdinfo info;
 672                 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
 673                 dfd = dev_open(dn, O_RDONLY);
 674                 if (dfd < 0)
 675                         continue;
 676                 if (st->ss->load_super(st, dfd, NULL)) {
 677                         close(dfd);
 678                         continue;
 679                 }
 680                 st->ss->getinfo_super(st, &info, NULL);
 681                 st->ss->free_super(st);
 682                 close(dfd);
 683
 684                 if (info.disk.number != number ||
 685                     info.events >= events)
 686                         continue;
 687
 688                 if (d->disk.raid_disk > -1)
 689                         sysfs_set_str(sra, d, "slot", "none");
 690                 if (sysfs_set_str(sra, d, "state", "remove") == 0)
 691                         if (verbose >= 0)
 692                                 pr_err("removing old device %s from %s\n",
 693                                        d->sys_name+4, array_name);
 694         }
 695 }
 696
 697 static int count_active(struct supertype *st, struct mdinfo *sra,
 698                         int mdfd, char **availp,
 699                         struct mdinfo *bestinfo)
 700 {
 701         /* count how many devices in sra think they are active */
 702         struct mdinfo *d;
 703         int cnt = 0;
 704         int replcnt = 0;
 705         __u64 max_events = 0;
 706         char *avail = NULL;
 707         int *best = NULL;
 708         char *devmap = NULL;
 709         int numdevs = 0;
 710         int devnum;
 711         int b, i;
 712         int raid_disks = 0;
 713
 714         if (!sra)
 715                 return 0;
 716
 717         for (d = sra->devs ; d ; d = d->next)
 718                 numdevs++;
 719         for (d = sra->devs, devnum = 0 ; d ; d = d->next, devnum++) {
 720                 char dn[30];
 721                 int dfd;
 722                 int ok;
 723                 struct mdinfo info;
 724
 725                 sprintf(dn, "%d:%d", d->disk.major, d->disk.minor);
 726                 dfd = dev_open(dn, O_RDONLY);
 727                 if (dfd < 0)
 728                         continue;
 729                 ok =  st->ss->load_super(st, dfd, NULL);
 730                 close(dfd);
 731                 if (ok != 0)
 732                         continue;
 733
 734                 info.array.raid_disks = raid_disks;
 735                 st->ss->getinfo_super(st, &info, devmap + raid_disks * devnum);
 736                 if (info.disk.raid_disk == MD_DISK_ROLE_JOURNAL)
 737                         bestinfo->journal_clean = 1;
 738                 if (!avail) {
 739                         raid_disks = info.array.raid_disks;
 740                         avail = xcalloc(raid_disks, 1);
 741                         *availp = avail;
 742
 743                         best = xcalloc(raid_disks, sizeof(int));
 744                         devmap = xcalloc(raid_disks, numdevs);
 745
 746                         st->ss->getinfo_super(st, &info, devmap);
 747                 }
 748
 749                 if (info.disk.state & (1<<MD_DISK_SYNC))
 750                 {
 751                         if (cnt == 0) {
 752                                 cnt++;
 753                                 max_events = info.events;
 754                                 avail[info.disk.raid_disk] = 2;
 755                                 best[info.disk.raid_disk] = devnum;
 756                                 st->ss->getinfo_super(st, bestinfo, NULL);
 757                         } else if (info.events == max_events) {
 758                                 avail[info.disk.raid_disk] = 2;
 759                                 best[info.disk.raid_disk] = devnum;
 760                         } else if (info.events == max_events-1) {
 761                                 if (avail[info.disk.raid_disk] == 0) {
 762                                         avail[info.disk.raid_disk] = 1;
 763                                         best[info.disk.raid_disk] = devnum;
 764                                 }
 765                         } else if (info.events < max_events - 1)
 766                                 ;
 767                         else if (info.events == max_events+1) {
 768                                 int i;
 769                                 max_events = info.events;
 770                                 for (i = 0; i < raid_disks; i++)
 771                                         if (avail[i])
 772                                                 avail[i]--;
 773                                 avail[info.disk.raid_disk] = 2;
 774                                 best[info.disk.raid_disk] = devnum;
 775                                 st->ss->getinfo_super(st, bestinfo, NULL);
 776                         } else { /* info.events much bigger */
 777                                 memset(avail, 0, raid_disks);
 778                                 max_events = info.events;
 779                                 avail[info.disk.raid_disk] = 2;
 780                                 best[info.disk.raid_disk] = devnum;
 781                                 st->ss->getinfo_super(st, bestinfo, NULL);
 782                         }
 783                 } else if (info.disk.state & (1<<MD_DISK_REPLACEMENT))
 784                         replcnt++;
 785                 st->ss->free_super(st);
 786         }
 787
 788         if (!avail)
 789                 return 0;
 790         /* We need to reject any device that thinks the best device is
 791          * failed or missing */
 792         for (b = 0; b < raid_disks; b++)
 793                 if (avail[b] == 2)
 794                         break;
 795         cnt = 0;
 796         for (i = 0 ; i < raid_disks ; i++) {
 797                 if (i != b && avail[i])
 798                         if (devmap[raid_disks * best[i] + b] == 0) {
 799                                 /* This device thinks 'b' is failed -
 800                                  * don't use it */
 801                                 devnum = best[i];
 802                                 for (d=sra->devs ; devnum; d = d->next)
 803                                         devnum--;
 804                                 d->disk.state |= (1 << MD_DISK_REMOVED);
 805                                 avail[i] = 0;
 806                         }
 807                 if (avail[i])
 808                         cnt++;
 809         }
 810         /* Also need to reject any spare device with an event count that
 811          * is too high
 812          */
 813         for (d = sra->devs; d; d = d->next) {
 814                 if (!(d->disk.state & (1<<MD_DISK_SYNC)) &&
 815                     d->events > max_events)
 816                         d->disk.state |= (1 << MD_DISK_REMOVED);
 817         }
 818         free(best);
 819         free(devmap);
 820         return cnt + replcnt;
 821 }
 822
 823 /* test if container has degraded member(s) */
 824 static int container_members_max_degradation(struct map_ent *map, struct map_ent *me)
 825 {
 826         mdu_array_info_t array;
 827         int afd;
 828         int max_degraded = 0;
 829
 830         for(; map; map = map->next) {
 831                 if (!metadata_container_matches(map->metadata, me->devnm))
 832                         continue;
 833                 afd = open_dev(map->devnm);
 834                 if (afd < 0)
 835                         continue;
 836                 /* most accurate information regarding array degradation */
 837                 if (ioctl(afd, GET_ARRAY_INFO, &array) >= 0) {
 838                         int degraded = array.raid_disks - array.active_disks -
 839                                        array.spare_disks;
 840                         if (degraded > max_degraded)
 841                                 max_degraded = degraded;
 842                 }
 843                 close(afd);
 844         }
 845         return (max_degraded);
 846 }
 847
 848 static int array_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
 849                            struct map_ent *target, int bare,
 850                            struct supertype *st, int verbose)
 851 {
 852         /* This device doesn't have any md metadata
 853          * The device policy allows 'spare' and if !bare, it allows spare-same-slot.
 854          * If 'st' is not set, then we only know that some metadata allows this,
 855          * others possibly don't.
 856          * So look for a container or array to attach the device to.
 857          * Prefer 'target' if that is set and the array is found.
 858          *
 859          * If st is set, then only arrays of that type are considered
 860          * Return 0 on success, or some exit code on failure, probably 1.
 861          */
 862         int rv = 1;
 863         struct stat stb;
 864         struct map_ent *mp, *map = NULL;
 865         struct mdinfo *chosen = NULL;
 866         int dfd = *dfdp;
 867
 868         if (fstat(dfd, &stb) != 0)
 869                 return 1;
 870
 871         /*
 872          * Now we need to find a suitable array to add this to.
 873          * We only accept arrays that:
 874          *  - match 'st'
 875          *  - are in the same domains as the device
 876          *  - are of an size for which the device will be useful
 877          * and we choose the one that is the most degraded
 878          */
 879
 880         if (map_lock(&map)) {
 881                 pr_err("failed to get exclusive lock on mapfile\n");
 882                 return 1;
 883         }
 884         for (mp = map ; mp ; mp = mp->next) {
 885                 struct supertype *st2;
 886                 struct domainlist *dl = NULL;
 887                 struct mdinfo *sra;
 888                 unsigned long long devsize;
 889                 unsigned long long component_size = 0;
 890
 891                 if (is_subarray(mp->metadata))
 892                         continue;
 893                 if (st) {
 894                         st2 = st->ss->match_metadata_desc(mp->metadata);
 895                         if (!st2 ||
 896                             (st->minor_version >= 0 &&
 897                              st->minor_version != st2->minor_version)) {
 898                                 if (verbose > 1)
 899                                         pr_err("not adding %s to %s as metadata type doesn't match\n",
 900                                                 devname, mp->path);
 901                                 free(st2);
 902                                 continue;
 903                         }
 904                         free(st2);
 905                 }
 906                 sra = sysfs_read(-1, mp->devnm,
 907                                  GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
 908                                  GET_DEGRADED|GET_COMPONENT|GET_VERSION);
 909                 if (!sra) {
 910                         /* Probably a container - no degraded info */
 911                         sra = sysfs_read(-1, mp->devnm,
 912                                          GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE|
 913                                          GET_COMPONENT|GET_VERSION);
 914                         if (sra)
 915                                 sra->array.failed_disks = -1;
 916                 }
 917                 if (!sra)
 918                         continue;
 919                 if (st == NULL) {
 920                         int i;
 921                         st2 = NULL;
 922                         for(i = 0; !st2 && superlist[i]; i++)
 923                                 st2 = superlist[i]->match_metadata_desc(
 924                                         sra->text_version);
 925                         if (!st2) {
 926                                 if (verbose > 1)
 927                                         pr_err("not adding %s to %s as metadata not recognised.\n",
 928                                                 devname, mp->path);
 929                                 goto next;
 930                         }
 931                         /* Need to double check the 'act_spare' permissions applies
 932                          * to this metadata.
 933                          */
 934                         if (!policy_action_allows(pol, st2->ss->name, act_spare))
 935                                 goto next;
 936                         if (!bare && !policy_action_allows(pol, st2->ss->name,
 937                                                            act_spare_same_slot))
 938                                 goto next;
 939                 } else
 940                         st2 = st;
 941                 /* update number of failed disks for mostly degraded
 942                  * container member */
 943                 if (sra->array.failed_disks == -1)
 944                         sra->array.failed_disks = container_members_max_degradation(map, mp);
 945
 946                 get_dev_size(dfd, NULL, &devsize);
 947                 if (sra->component_size == 0) {
 948                         /* true for containers, here we must read superblock
 949                          * to obtain minimum spare size */
 950                         struct supertype *st3 = dup_super(st2);
 951                         int mdfd = open_dev(mp->devnm);
 952                         if (mdfd < 0) {
 953                                 free(st3);
 954                                 goto next;
 955                         }
 956                         if (st3->ss->load_container &&
 957                             !st3->ss->load_container(st3, mdfd, mp->path)) {
 958                                 component_size = st3->ss->min_acceptable_spare_size(st3);
 959                                 st3->ss->free_super(st3);
 960                         }
 961                         free(st3);
 962                         close(mdfd);
 963                 }
 964                 if ((sra->component_size > 0 &&
 965                      st2->ss->avail_size(st2, devsize,
 966                                          sra->devs
 967                                          ? sra->devs->data_offset
 968                                          : INVALID_SECTORS)
 969                      < sra->component_size)
 970                     ||
 971                     (sra->component_size == 0 && devsize < component_size)) {
 972                         if (verbose > 1)
 973                                 pr_err("not adding %s to %s as it is too small\n",
 974                                         devname, mp->path);
 975                         goto next;
 976                 }
 977                 /* test against target.
 978                  * If 'target' is set and 'bare' is false, we only accept
 979                  * arrays/containers that match 'target'.
 980                  * If 'target' is set and 'bare' is true, we prefer the
 981                  * array which matches 'target'.
 982                  * target is considered only if we deal with degraded array
 983                  */
 984                 if (target && policy_action_allows(pol, st2->ss->name,
 985                                                    act_spare_same_slot)) {
 986                         if (strcmp(target->metadata, mp->metadata) == 0 &&
 987                             memcmp(target->uuid, mp->uuid,
 988                                    sizeof(target->uuid)) == 0 &&
 989                             sra->array.failed_disks > 0) {
 990                                 /* This is our target!! */
 991                                 sysfs_free(chosen);
 992                                 chosen = sra;
 993                                 sra = NULL;
 994                                 /* skip to end so we don't check any more */
 995                                 while (mp->next)
 996                                         mp = mp->next;
 997                                 goto next;
 998                         }
 999                         /* not our target */
1000                         if (!bare)
1001                                 goto next;
1002                 }
1003
1004                 dl = domain_from_array(sra, st2->ss->name);
1005                 if (domain_test(dl, pol, st2->ss->name) != 1) {
1006                         /* domain test fails */
1007                         if (verbose > 1)
1008                                 pr_err("not adding %s to %s as it is not in a compatible domain\n",
1009                                         devname, mp->path);
1010
1011                         goto next;
1012                 }
1013                 /* all tests passed, OK to add to this array */
1014                 if (!chosen) {
1015                         chosen = sra;
1016                         sra = NULL;
1017                 } else if (chosen->array.failed_disks < sra->array.failed_disks) {
1018                         sysfs_free(chosen);
1019                         chosen = sra;
1020                         sra = NULL;
1021                 }
1022         next:
1023                 sysfs_free(sra);
1024                 if (st != st2)
1025                         free(st2);
1026                 if (dl)
1027                         domain_free(dl);
1028         }
1029         if (chosen) {
1030                 /* add current device to chosen array as a spare */
1031                 int mdfd = open_dev(chosen->sys_name);
1032                 if (mdfd >= 0) {
1033                         struct mddev_dev devlist;
1034                         char chosen_devname[24]; // 2*11 for int (including signs) + colon + null
1035                         devlist.next = NULL;
1036                         devlist.used = 0;
1037                         devlist.writemostly = 0;
1038                         devlist.devname = chosen_devname;
1039                         sprintf(chosen_devname, "%d:%d", major(stb.st_rdev),
1040                                 minor(stb.st_rdev));
1041                         devlist.disposition = 'a';
1042                         close(dfd);
1043                         *dfdp = -1;
1044                         rv =  Manage_subdevs(chosen->sys_name, mdfd, &devlist,
1045                                              -1, 0, NULL, 0);
1046                         close(mdfd);
1047                 }
1048                 if (verbose > 0) {
1049                         if (rv == 0)
1050                                 pr_err("added %s as spare for %s\n",
1051                                         devname, chosen->sys_name);
1052                         else
1053                                 pr_err("failed to add %s as spare for %s\n",
1054                                         devname, chosen->sys_name);
1055                 }
1056                 sysfs_free(chosen);
1057         }
1058         map_unlock(&map);
1059         return rv;
1060 }
1061
1062 static int partition_try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1063                                struct supertype *st, int verbose)
1064 {
1065         /* we know that at least one partition virtual-metadata is
1066          * allowed to incorporate spares like this device.  We need to
1067          * find a suitable device to copy partition information from.
1068          *
1069          * Getting a list of all disk (not partition) devices is
1070          * slightly non-trivial.  We could look at /sys/block, but
1071          * that is theoretically due to be removed.  Maybe best to use
1072          * /dev/disk/by-path/?* and ignore names ending '-partNN' as
1073          * we depend on this directory of 'path' info.  But that fails
1074          * to find loop devices and probably others.  Maybe don't
1075          * worry about that, they aren't the real target.
1076          *
1077          * So: check things in /dev/disk/by-path to see if they are in
1078          * a compatible domain, then load the partition table and see
1079          * if it is OK for the new device, and choose the largest
1080          * partition table that fits.
1081          */
1082         DIR *dir;
1083         struct dirent *de;
1084         char *chosen = NULL;
1085         unsigned long long chosen_size = 0;
1086         struct supertype *chosen_st = NULL;
1087         int fd;
1088
1089         dir = opendir("/dev/disk/by-path");
1090         if (!dir)
1091                 return 1;
1092         while ((de = readdir(dir)) != NULL) {
1093                 char *ep;
1094                 struct dev_policy *pol2 = NULL;
1095                 struct domainlist *domlist = NULL;
1096                 int fd = -1;
1097                 struct mdinfo info;
1098                 struct supertype *st2 = NULL;
1099                 char *devname = NULL;
1100                 unsigned long long devsectors;
1101
1102                 if (de->d_ino == 0 ||
1103                     de->d_name[0] == '.' ||
1104                     (de->d_type != DT_LNK && de->d_type != DT_UNKNOWN))
1105                         goto next;
1106
1107                 ep = de->d_name + strlen(de->d_name);
1108                 while (ep > de->d_name &&
1109                        isdigit(ep[-1]))
1110                         ep--;
1111                 if (ep > de->d_name + 5 &&
1112                     strncmp(ep-5, "-part", 5) == 0)
1113                         /* This is a partition - skip it */
1114                         goto next;
1115
1116                 pol2 = path_policy(de->d_name, type_disk);
1117
1118                 domain_merge(&domlist, pol2, st ? st->ss->name : NULL);
1119                 if (domain_test(domlist, pol, st ? st->ss->name : NULL) != 1)
1120                         /* new device is incompatible with this device. */
1121                         goto next;
1122
1123                 domain_free(domlist);
1124                 domlist = NULL;
1125
1126                 if (asprintf(&devname, "/dev/disk/by-path/%s", de->d_name) != 1) {
1127                         devname = NULL;
1128                         goto next;
1129                 }
1130                 fd = open(devname, O_RDONLY);
1131                 if (fd < 0)
1132                         goto next;
1133                 if (get_dev_size(fd, devname, &devsectors) == 0)
1134                         goto next;
1135                 devsectors >>= 9;
1136
1137                 if (st)
1138                         st2 = dup_super(st);
1139                 else
1140                         st2 = guess_super_type(fd, guess_partitions);
1141                 if (st2 == NULL ||
1142                     st2->ss->load_super(st2, fd, NULL) < 0)
1143                         goto next;
1144                 st2->ignore_hw_compat = 0;
1145
1146                 if (!st) {
1147                         /* Check domain policy again, this time referring to metadata */
1148                         domain_merge(&domlist, pol2, st2->ss->name);
1149                         if (domain_test(domlist, pol, st2->ss->name) != 1)
1150                                 /* Incompatible devices for this metadata type */
1151                                 goto next;
1152                         if (!policy_action_allows(pol, st2->ss->name, act_spare))
1153                                 /* Some partition types allow sparing, but not
1154                                  * this one.
1155                                  */
1156                                 goto next;
1157                 }
1158
1159                 st2->ss->getinfo_super(st2, &info, NULL);
1160                 if (info.component_size > devsectors)
1161                         /* This partitioning doesn't fit in the device */
1162                         goto next;
1163
1164                 /* This is an acceptable device to copy partition
1165                  * metadata from.  We could just stop here, but I
1166                  * think I want to keep looking incase a larger
1167                  * metadata which makes better use of the device can
1168                  * be found.
1169                  */
1170                 if (chosen == NULL ||
1171                     chosen_size < info.component_size) {
1172                         chosen_size = info.component_size;
1173                         free(chosen);
1174                         chosen = devname;
1175                         devname = NULL;
1176                         if (chosen_st) {
1177                                 chosen_st->ss->free_super(chosen_st);
1178                                 free(chosen_st);
1179                         }
1180                         chosen_st = st2;
1181                         st2 = NULL;
1182                 }
1183
1184         next:
1185                 free(devname);
1186                 domain_free(domlist);
1187                 dev_policy_free(pol2);
1188                 if (st2)
1189                         st2->ss->free_super(st2);
1190                 free(st2);
1191
1192                 if (fd >= 0)
1193                         close(fd);
1194         }
1195
1196         closedir(dir);
1197
1198         if (!chosen)
1199                 return 1;
1200
1201         /* 'chosen' is the best device we can find.  Let's write its
1202          * metadata to devname dfd is read-only so don't use that
1203          */
1204         fd = open(devname, O_RDWR);
1205         if (fd >= 0) {
1206                 chosen_st->ss->store_super(chosen_st, fd);
1207                 close(fd);
1208         }
1209         free(chosen);
1210         chosen_st->ss->free_super(chosen_st);
1211         free(chosen_st);
1212         return 0;
1213 }
1214
1215 static int is_bare(int dfd)
1216 {
1217         unsigned long long size = 0;
1218         char bufpad[4096 + 4096];
1219         char *buf = (char*)(((long)bufpad + 4096) & ~4095);
1220
1221         if (lseek(dfd, 0, SEEK_SET) != 0 ||
1222             read(dfd, buf, 4096) != 4096)
1223                 return 0;
1224
1225         if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1226                 return 0;
1227         if (memcmp(buf, buf+1, 4095) != 0)
1228                 return 0;
1229
1230         /* OK, first 4K appear blank, try the end. */
1231         get_dev_size(dfd, NULL, &size);
1232         if (lseek(dfd, size-4096, SEEK_SET) < 0 ||
1233             read(dfd, buf, 4096) != 4096)
1234                 return 0;
1235
1236         if (buf[0] != '\0' && buf[0] != '\x5a' && buf[0] != '\xff')
1237                 return 0;
1238         if (memcmp(buf, buf+1, 4095) != 0)
1239                 return 0;
1240
1241         return 1;
1242 }
1243
1244 /* adding a spare to a regular array is quite different from adding one to
1245  * a set-of-partitions virtual array.
1246  * This function determines which is worth trying and tries as appropriate.
1247  * Arrays are given priority over partitions.
1248  */
1249 static int try_spare(char *devname, int *dfdp, struct dev_policy *pol,
1250                      struct map_ent *target,
1251                      struct supertype *st, int verbose)
1252 {
1253         int i;
1254         int rv;
1255         int arrays_ok = 0;
1256         int partitions_ok = 0;
1257         int dfd = *dfdp;
1258         int bare;
1259
1260         /* Can only add a spare if device has at least one domain */
1261         if (pol_find(pol, pol_domain) == NULL)
1262                 return 1;
1263         /* And only if some action allows spares */
1264         if (!policy_action_allows(pol, st?st->ss->name:NULL, act_spare))
1265                 return 1;
1266
1267         /* Now check if the device is bare.
1268          * bare devices can always be added as a spare
1269          * non-bare devices can only be added if spare-same-slot is permitted,
1270          * and this device is replacing a previous device - in which case 'target'
1271          * will be set.
1272          */
1273         if (!is_bare(dfd)) {
1274                 /* Must have a target and allow same_slot */
1275                 /* Later - may allow force_spare without target */
1276                 if (!target ||
1277                     !policy_action_allows(pol, st?st->ss->name:NULL,
1278                                           act_spare_same_slot)) {
1279                         if (verbose > 1)
1280                                 pr_err("%s is not bare, so not considering as a spare\n",
1281                                         devname);
1282                         return 1;
1283                 }
1284                 bare = 0;
1285         } else
1286                 bare = 1;
1287
1288         /* It might be OK to add this device to an array - need to see
1289          * what arrays might be candidates.
1290          */
1291         if (st) {
1292                 /* just try try 'array' or 'partition' based on this metadata */
1293                 if (st->ss->add_to_super)
1294                         return array_try_spare(devname, dfdp, pol, target, bare,
1295                                                st, verbose);
1296                 else
1297                         return partition_try_spare(devname, dfdp, pol,
1298                                                    st, verbose);
1299         }
1300         /* No metadata was specified or found so options are open.
1301          * Check for whether any array metadata, or any partition metadata
1302          * might allow adding the spare.  This check is just help to avoid
1303          * a more costly scan of all arrays when we can be sure that will
1304          * fail.
1305          */
1306         for (i = 0; (!arrays_ok || !partitions_ok) && superlist[i] ; i++) {
1307                 if (superlist[i]->add_to_super && !arrays_ok &&
1308                     policy_action_allows(pol, superlist[i]->name, act_spare))
1309                         arrays_ok = 1;
1310                 if (superlist[i]->add_to_super == NULL && !partitions_ok &&
1311                     policy_action_allows(pol, superlist[i]->name, act_spare))
1312                         partitions_ok = 1;
1313         }
1314         rv = 1;
1315         if (arrays_ok)
1316                 rv = array_try_spare(devname, dfdp, pol, target, bare,
1317                                      st, verbose);
1318         if (rv != 0 && partitions_ok)
1319                 rv = partition_try_spare(devname, dfdp, pol, st, verbose);
1320         return rv;
1321 }
1322
1323 int IncrementalScan(struct context *c, char *devnm)
1324 {
1325         /* look at every device listed in the 'map' file.
1326          * If one is found that is not running then:
1327          *  look in mdadm.conf for bitmap file.
1328          *   if one exists, but array has none, add it.
1329          *  try to start array in auto-readonly mode
1330          */
1331         struct map_ent *mapl = NULL;
1332         struct map_ent *me;
1333         struct mddev_ident *devs, *mddev;
1334         int rv = 0;
1335         char container[32];
1336         char *only = NULL;
1337
1338         map_read(&mapl);
1339         devs = conf_get_ident(NULL);
1340
1341 restart:
1342         for (me = mapl ; me ; me = me->next) {
1343                 mdu_array_info_t array;
1344                 mdu_bitmap_file_t bmf;
1345                 struct mdinfo *sra;
1346                 int mdfd;
1347
1348                 if (devnm && strcmp(devnm, me->devnm) != 0)
1349                         continue;
1350                 if (me->metadata[0] == '/') {
1351                         char *sl;
1352
1353                         if (!devnm)
1354                                 continue;
1355
1356                         /* member array, need to work on container */
1357                         strncpy(container, me->metadata+1, 32);
1358                         container[31] = 0;
1359                         sl = strchr(container, '/');
1360                         if (sl)
1361                                 *sl = 0;
1362                         only = devnm;
1363                         devnm = container;
1364                         goto restart;
1365                 }
1366                 mdfd = open_dev(me->devnm);
1367
1368                 if (mdfd < 0)
1369                         continue;
1370                 if (!isdigit(me->metadata[0])) {
1371                         /* must be a container */
1372                         struct supertype *st = super_by_fd(mdfd, NULL);
1373                         int ret = 0;
1374                         struct map_ent *map = NULL;
1375
1376                         if (st && st->ss->load_container)
1377                                 ret = st->ss->load_container(st, mdfd, NULL);
1378                         close(mdfd);
1379                         if (!ret && st && st->ss->container_content) {
1380                                 if (map_lock(&map))
1381                                         pr_err("failed to get exclusive lock on mapfile\n");
1382                                 ret = Incremental_container(st, me->path, c, only);
1383                                 map_unlock(&map);
1384                         }
1385                         if (ret)
1386                                 rv = 1;
1387                         continue;
1388                 }
1389                 if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
1390                     errno != ENODEV) {
1391                         close(mdfd);
1392                         continue;
1393                 }
1394                 /* Ok, we can try this one.   Maybe it needs a bitmap */
1395                 for (mddev = devs ; mddev ; mddev = mddev->next)
1396                         if (mddev->devname && me->path
1397                             && devname_matches(mddev->devname, me->path))
1398                                 break;
1399                 if (mddev && mddev->bitmap_file) {
1400                         /*
1401                          * Note: early kernels will wrongly fail this, so it
1402                          * is a hint only
1403                          */
1404                         int added = -1;
1405                         if (ioctl(mdfd, GET_ARRAY_INFO, &bmf) < 0) {
1406                                 int bmfd = open(mddev->bitmap_file, O_RDWR);
1407                                 if (bmfd >= 0) {
1408                                         added = ioctl(mdfd, SET_BITMAP_FILE,
1409                                                       bmfd);
1410                                         close(bmfd);
1411                                 }
1412                         }
1413                         if (c->verbose >= 0) {
1414                                 if (added == 0)
1415                                         pr_err("Added bitmap %s to %s\n",
1416                                                mddev->bitmap_file, me->path);
1417                                 else if (errno != EEXIST)
1418                                         pr_err("Failed to add bitmap to %s: %s\n",
1419                                                me->path, strerror(errno));
1420                         }
1421                 }
1422                 /* FIXME check for reshape_active and consider not
1423                  * starting array.
1424                  */
1425                 sra = sysfs_read(mdfd, NULL, 0);
1426                 if (sra) {
1427                         if (sysfs_set_str(sra, NULL,
1428                                           "array_state", "read-auto") == 0) {
1429                                 if (c->verbose >= 0)
1430                                         pr_err("started array %s\n",
1431                                                me->path ?: me->devnm);
1432                         } else {
1433                                 pr_err("failed to start array %s: %s\n",
1434                                        me->path ?: me->devnm,
1435                                        strerror(errno));
1436                                 rv = 1;
1437                         }
1438                         sysfs_free(sra);
1439                 }
1440         }
1441         return rv;
1442 }
1443
1444 static char *container2devname(char *devname)
1445 {
1446         char *mdname = NULL;
1447
1448         if (devname[0] == '/') {
1449                 int fd = open(devname, O_RDONLY);
1450                 if (fd >= 0) {
1451                         mdname = xstrdup(fd2devnm(fd));
1452                         close(fd);
1453                 }
1454         } else {
1455                 int uuid[4];
1456                 struct map_ent *mp, *map = NULL;
1457
1458                 if (!parse_uuid(devname, uuid))
1459                         return mdname;
1460                 mp = map_by_uuid(&map, uuid);
1461                 if (mp)
1462                         mdname = xstrdup(mp->devnm);
1463                 map_free(map);
1464         }
1465
1466         return mdname;
1467 }
1468
1469 static int Incremental_container(struct supertype *st, char *devname,
1470                                  struct context *c, char *only)
1471 {
1472         /* Collect the contents of this container and for each
1473          * array, choose a device name and assemble the array.
1474          */
1475
1476         struct mdinfo *list;
1477         struct mdinfo *ra;
1478         struct map_ent *map = NULL;
1479         struct mdinfo info;
1480         int trustworthy;
1481         struct mddev_ident *match;
1482         int rv = 0;
1483         struct domainlist *domains;
1484         struct map_ent *smp;
1485         int suuid[4];
1486         int sfd;
1487         int ra_blocked = 0;
1488         int ra_all = 0;
1489         int result = 0;
1490
1491         st->ss->getinfo_super(st, &info, NULL);
1492
1493         if ((c->runstop > 0 && info.container_enough >= 0) ||
1494             info.container_enough > 0)
1495                 /* pass */;
1496         else {
1497                 if (c->export) {
1498                         printf("MD_STARTED=no\n");
1499                 } else if (c->verbose)
1500                         pr_err("not enough devices to start the container\n");
1501                 return 0;
1502         }
1503
1504         match = conf_match(st, &info, devname, c->verbose, &rv);
1505         if (match == NULL && rv == 2)
1506                 return rv;
1507
1508         /* Need to compute 'trustworthy' */
1509         if (match)
1510                 trustworthy = LOCAL;
1511         else if (st->ss->match_home(st, c->homehost) == 1)
1512                 trustworthy = LOCAL;
1513         else if (st->ss->match_home(st, "any") == 1)
1514                 trustworthy = LOCAL;
1515         else
1516                 trustworthy = FOREIGN;
1517
1518         list = st->ss->container_content(st, NULL);
1519         /* when nothing to activate - quit */
1520         if (list == NULL) {
1521                 if (c->export) {
1522                         printf("MD_STARTED=nothing\n");
1523                 }
1524                 return 0;
1525         }
1526         for (ra = list ; ra ; ra = ra->next) {
1527                 int mdfd;
1528                 char chosen_name[1024];
1529                 struct map_ent *mp;
1530                 struct mddev_ident *match = NULL;
1531
1532                 ra_all++;
1533                 /* do not activate arrays blocked by metadata handler */
1534                 if (ra->array.state & (1 << MD_SB_BLOCK_VOLUME)) {
1535                         pr_err("Cannot activate array %s in %s.\n",
1536                                 ra->text_version, devname);
1537                         ra_blocked++;
1538                         continue;
1539                 }
1540                 mp = map_by_uuid(&map, ra->uuid);
1541
1542                 if (mp) {
1543                         mdfd = open_dev(mp->devnm);
1544                         if (mp->path)
1545                                 strcpy(chosen_name, mp->path);
1546                         else
1547                                 strcpy(chosen_name, mp->devnm);
1548                 } else if (!only) {
1549
1550                         /* Check in mdadm.conf for container == devname and
1551                          * member == ra->text_version after second slash.
1552                          */
1553                         char *sub = strchr(ra->text_version+1, '/');
1554                         struct mddev_ident *array_list;
1555                         if (sub) {
1556                                 sub++;
1557                                 array_list = conf_get_ident(NULL);
1558                         } else
1559                                 array_list = NULL;
1560                         for(; array_list ; array_list = array_list->next) {
1561                                 char *dn;
1562                                 if (array_list->member == NULL ||
1563                                     array_list->container == NULL)
1564                                         continue;
1565                                 if (strcmp(array_list->member, sub) != 0)
1566                                         continue;
1567                                 if (array_list->uuid_set &&
1568                                     !same_uuid(ra->uuid, array_list->uuid, st->ss->swapuuid))
1569                                         continue;
1570                                 dn = container2devname(array_list->container);
1571                                 if (dn == NULL)
1572                                         continue;
1573                                 if (strncmp(dn, ra->text_version+1,
1574                                             strlen(dn)) != 0 ||
1575                                     ra->text_version[strlen(dn)+1] != '/') {
1576                                         free(dn);
1577                                         continue;
1578                                 }
1579                                 free(dn);
1580                                 /* we have a match */
1581                                 match = array_list;
1582                                 if (c->verbose>0)
1583                                         pr_err("match found for member %s\n",
1584                                                 array_list->member);
1585                                 break;
1586                         }
1587
1588                         if (match && match->devname &&
1589                             strcasecmp(match->devname, "<ignore>") == 0) {
1590                                 if (c->verbose > 0)
1591                                         pr_err("array %s/%s is explicitly ignored by mdadm.conf\n",
1592                                                match->container, match->member);
1593                                 continue;
1594                         }
1595                         if (match)
1596                                 trustworthy = LOCAL;
1597
1598                         mdfd = create_mddev(match ? match->devname : NULL,
1599                                             ra->name,
1600                                             c->autof,
1601                                             trustworthy,
1602                                             chosen_name);
1603                 }
1604                 if (only && (!mp || strcmp(mp->devnm, only) != 0))
1605                         continue;
1606
1607                 if (mdfd < 0) {
1608                         pr_err("failed to open %s: %s.\n",
1609                                 chosen_name, strerror(errno));
1610                         return 2;
1611                 }
1612
1613                 assemble_container_content(st, mdfd, ra, c,
1614                                            chosen_name, &result);
1615                 close(mdfd);
1616         }
1617         if (c->export && result) {
1618                 char sep = '=';
1619                 printf("MD_STARTED");
1620                 if (result & INCR_NO) {
1621                         printf("%cno", sep);
1622                         sep = ',';
1623                 }
1624                 if (result & INCR_UNSAFE) {
1625                         printf("%cunsafe", sep);
1626                         sep = ',';
1627                 }
1628                 if (result & INCR_ALREADY) {
1629                         printf("%calready", sep);
1630                         sep = ',';
1631                 }
1632                 if (result & INCR_YES) {
1633                         printf("%cyes", sep);
1634                         sep = ',';
1635                 }
1636                 printf("\n");
1637         }
1638
1639         /* don't move spares to container with volume being activated
1640            when all volumes are blocked */
1641         if (ra_all == ra_blocked)
1642                 return 0;
1643
1644         /* Now move all suitable spares from spare container */
1645         domains = domain_from_array(list, st->ss->name);
1646         memcpy(suuid, uuid_zero, sizeof(int[4]));
1647         if (domains &&
1648             (smp = map_by_uuid(&map, suuid)) != NULL &&
1649             (sfd = open(smp->path, O_RDONLY)) >= 0) {
1650                 /* spare container found */
1651                 struct supertype *sst =
1652                         super_imsm.match_metadata_desc("imsm");
1653                 struct mdinfo *sinfo;
1654                 unsigned long long min_size = 0;
1655                 if (st->ss->min_acceptable_spare_size)
1656                         min_size = st->ss->min_acceptable_spare_size(st);
1657                 if (!sst->ss->load_container(sst, sfd, NULL)) {
1658                         close(sfd);
1659                         sinfo = container_choose_spares(sst, min_size,
1660                                                         domains, NULL,
1661                                                         st->ss->name, 0);
1662                         sst->ss->free_super(sst);
1663                         if (sinfo){
1664                                 int count = 0;
1665                                 struct mdinfo *disks = sinfo->devs;
1666                                 while (disks) {
1667                                         /* move spare from spare
1668                                          * container to currently
1669                                          * assembled one
1670                                          */
1671                                         if (move_spare(
1672                                                     smp->path,
1673                                                     devname,
1674                                                     makedev(disks->disk.major,
1675                                                             disks->disk.minor)))
1676                                                 count++;
1677                                         disks = disks->next;
1678                                 }
1679                                 if (count)
1680                                         pr_err("Added %d spare%s to %s\n",
1681                                                count, count>1?"s":"", devname);
1682                         }
1683                         sysfs_free(sinfo);
1684                 } else
1685                         close(sfd);
1686         }
1687         domain_free(domains);
1688         return 0;
1689 }
1690
1691 static void run_udisks(char *arg1, char *arg2)
1692 {
1693         int pid = fork();
1694         int status;
1695         if (pid == 0) {
1696                 execl("/usr/bin/udisks", "udisks", arg1, arg2, NULL);
1697                 execl("/bin/udisks", "udisks", arg1, arg2, NULL);
1698                 exit(1);
1699         }
1700         while (pid > 0 && wait(&status) != pid)
1701                 ;
1702 }
1703
1704 /*
1705  * IncrementalRemove - Attempt to see if the passed in device belongs to any
1706  * raid arrays, and if so first fail (if needed) and then remove the device.
1707  *
1708  * @devname - The device we want to remove
1709  * @id_path - name as found in /dev/disk/by-path for this device
1710  *
1711  * Note: the device name must be a kernel name like "sda", so
1712  * that we can find it in /proc/mdstat
1713  */
1714 int IncrementalRemove(char *devname, char *id_path, int verbose)
1715 {
1716         int mdfd;
1717         int rv = 0;
1718         struct mdstat_ent *ent;
1719         struct mddev_dev devlist;
1720         struct mdinfo mdi;
1721         char buf[32];
1722
1723         if (!id_path)
1724                 dprintf("incremental removal without --path <id_path> lacks the possibility to re-add new device in this port\n");
1725
1726         if (strchr(devname, '/')) {
1727                 pr_err("incremental removal requires a kernel device name, not a file: %s\n", devname);
1728                 return 1;
1729         }
1730         ent = mdstat_by_component(devname);
1731         if (!ent) {
1732                 if (verbose >= 0)
1733                         pr_err("%s does not appear to be a component of any array\n", devname);
1734                 return 1;
1735         }
1736         sysfs_init(&mdi, -1, ent->devnm);
1737         mdfd = open_dev_excl(ent->devnm);
1738         if (mdfd > 0) {
1739                 close(mdfd);
1740                 if (sysfs_get_str(&mdi, NULL, "array_state",
1741                                   buf, sizeof(buf)) > 0) {
1742                         if (strncmp(buf, "active", 6) == 0 ||
1743                             strncmp(buf, "clean", 5) == 0)
1744                                 sysfs_set_str(&mdi, NULL,
1745                                               "array_state", "read-auto");
1746                 }
1747         }
1748         mdfd = open_dev(ent->devnm);
1749         if (mdfd < 0) {
1750                 if (verbose >= 0)
1751                         pr_err("Cannot open array %s!!\n", ent->devnm);
1752                 free_mdstat(ent);
1753                 return 1;
1754         }
1755
1756         if (id_path) {
1757                 struct map_ent *map = NULL, *me;
1758                 me = map_by_devnm(&map, ent->devnm);
1759                 if (me)
1760                         policy_save_path(id_path, me);
1761                 map_free(map);
1762         }
1763
1764         memset(&devlist, 0, sizeof(devlist));
1765         devlist.devname = devname;
1766         devlist.disposition = 'f';
1767         /* for a container, we must fail each member array */
1768         if (ent->metadata_version &&
1769             strncmp(ent->metadata_version, "external:", 9) == 0) {
1770                 struct mdstat_ent *mdstat = mdstat_read(0, 0);
1771                 struct mdstat_ent *memb;
1772                 for (memb = mdstat ; memb ; memb = memb->next)
1773                         if (is_container_member(memb, ent->devnm)) {
1774                                 int subfd = open_dev(memb->devnm);
1775                                 if (subfd >= 0) {
1776                                         rv |= Manage_subdevs(
1777                                                 memb->devnm, subfd,
1778                                                 &devlist, verbose, 0,
1779                                                 NULL, 0);
1780                                         close(subfd);
1781                                 }
1782                         }
1783                 free_mdstat(mdstat);
1784         } else
1785                 rv |= Manage_subdevs(ent->devnm, mdfd, &devlist,
1786                                     verbose, 0, NULL, 0);
1787         if (rv & 2) {
1788                 /* Failed due to EBUSY, try to stop the array.
1789                  * Give udisks a chance to unmount it first.
1790                  */
1791                 int devid = devnm2devid(ent->devnm);
1792                 run_udisks("--unmount", map_dev(major(devid),minor(devid), 0));
1793                 rv = Manage_stop(ent->devnm, mdfd, verbose, 1);
1794                 if (rv)
1795                         /* At least we can try to trigger a 'remove' */
1796                         sysfs_uevent(&mdi, "remove");
1797                 if (verbose) {
1798                         if (rv)
1799                                 pr_err("Fail to stop %s too.\n", ent->devnm);
1800                 }
1801         } else {
1802                 devlist.disposition = 'r';
1803                 rv = Manage_subdevs(ent->devnm, mdfd, &devlist,
1804                                     verbose, 0, NULL, 0);
1805         }
1806         close(mdfd);
1807         free_mdstat(ent);
1808         return rv;
1809 }