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