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