Monitor.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@suse.de>
  23  */
  24
  25 #include        "mdadm.h"
  26 #include        "md_p.h"
  27 #include        "md_u.h"
  28 #include        <sys/wait.h>
  29 #include        <signal.h>
  30 #include        <limits.h>
  31 #include        <syslog.h>
  32
  33 /* The largest number of disks current arrays can manage is 384
  34  * This really should be dynamically, but that will have to wait
  35  * At least it isn't MD_SB_DISKS.
  36  */
  37 #define MaxDisks 384
  38 struct state {
  39         char *devname;
  40         int devnum;     /* to sync with mdstat info */
  41         long utime;
  42         int err;
  43         char *spare_group;
  44         int active, working, failed, spare, raid;
  45         int expected_spares;
  46         int devstate[MaxDisks];
  47         dev_t devid[MaxDisks];
  48         int percent;
  49         int parent_dev; /* For subarray, devnum of parent.
  50                          * For others, NoMdDev
  51                          */
  52         struct supertype *metadata;
  53         struct state *subarray;/* for a container it is a link to first subarray
  54                                 * for a subarray it is a link to next subarray
  55                                 * in the same container */
  56         struct state *parent;  /* for a subarray it is a link to its container
  57                                 */
  58         struct state *next;
  59 };
  60
  61 struct alert_info {
  62         char *mailaddr;
  63         char *mailfrom;
  64         char *alert_cmd;
  65         int dosyslog;
  66 };
  67 static int make_daemon(char *pidfile);
  68 static int check_one_sharer(int scan);
  69 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
  70 static int check_array(struct state *st, struct mdstat_ent *mdstat,
  71                        int test, struct alert_info *info,
  72                        int increments);
  73 static int add_new_arrays(struct mdstat_ent *mdstat, struct state *statelist,
  74                           int test, struct alert_info *info);
  75 static void try_spare_migration(struct state *statelist, struct alert_info *info);
  76 static void link_containers_with_subarrays(struct state *list);
  77
  78 int Monitor(struct mddev_dev *devlist,
  79             char *mailaddr, char *alert_cmd,
  80             int period, int daemonise, int scan, int oneshot,
  81             int dosyslog, int test, char *pidfile, int increments,
  82             int share)
  83 {
  84         /*
  85          * Every few seconds, scan every md device looking for changes
  86          * When a change is found, log it, possibly run the alert command,
  87          * and possibly send Email
  88          *
  89          * For each array, we record:
  90          *   Update time
  91          *   active/working/failed/spare drives
  92          *   State of each device.
  93          *   %rebuilt if rebuilding
  94          *
  95          * If the update time changes, check out all the data again
  96          * It is possible that we cannot get the state of each device
  97          * due to bugs in the md kernel module.
  98          * We also read /proc/mdstat to get rebuild percent,
  99          * and to get state on all active devices incase of kernel bug.
 100          *
 101          * Events are:
 102          *    Fail
 103          *      An active device had Faulty set or Active/Sync removed
 104          *    FailSpare
 105          *      A spare device had Faulty set
 106          *    SpareActive
 107          *      An active device had a reverse transition
 108          *    RebuildStarted
 109          *      percent went from -1 to +ve
 110          *    RebuildNN
 111          *      percent went from below to not-below NN%
 112          *    DeviceDisappeared
 113          *      Couldn't access a device which was previously visible
 114          *
 115          * if we detect an array with active<raid and spare==0
 116          * we look at other arrays that have same spare-group
 117          * If we find one with active==raid and spare>0,
 118          *  and if we can get_disk_info and find a name
 119          *  Then we hot-remove and hot-add to the other array
 120          *
 121          * If devlist is NULL, then we can monitor everything because --scan
 122          * was given.  We get an initial list from config file and add anything
 123          * that appears in /proc/mdstat
 124          */
 125
 126         struct state *statelist = NULL;
 127         int finished = 0;
 128         struct mdstat_ent *mdstat = NULL;
 129         char *mailfrom = NULL;
 130         struct alert_info info;
 131
 132         if (!mailaddr) {
 133                 mailaddr = conf_get_mailaddr();
 134                 if (mailaddr && ! scan)
 135                         fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n",
 136                                mailaddr);
 137         }
 138         mailfrom = conf_get_mailfrom();
 139
 140         if (!alert_cmd) {
 141                 alert_cmd = conf_get_program();
 142                 if (alert_cmd && ! scan)
 143                         fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n",
 144                                alert_cmd);
 145         }
 146         if (scan && !mailaddr && !alert_cmd) {
 147                 fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n");
 148                 return 1;
 149         }
 150         info.alert_cmd = alert_cmd;
 151         info.mailaddr = mailaddr;
 152         info.mailfrom = mailfrom;
 153         info.dosyslog = dosyslog;
 154
 155         if (daemonise)
 156                 if (make_daemon(pidfile))
 157                         return 1;
 158
 159         if (share)
 160                 if (check_one_sharer(scan))
 161                         return 1;
 162
 163         if (devlist == NULL) {
 164                 struct mddev_ident *mdlist = conf_get_ident(NULL);
 165                 for (; mdlist; mdlist=mdlist->next) {
 166                         struct state *st;
 167                         if (mdlist->devname == NULL)
 168                                 continue;
 169                         if (strcasecmp(mdlist->devname, "<ignore>") == 0)
 170                                 continue;
 171                         st = calloc(1, sizeof *st);
 172                         if (st == NULL)
 173                                 continue;
 174                         if (mdlist->devname[0] == '/')
 175                                 st->devname = strdup(mdlist->devname);
 176                         else {
 177                                 st->devname = malloc(8+strlen(mdlist->devname)+1);
 178                                 strcpy(strcpy(st->devname, "/dev/md/"),
 179                                        mdlist->devname);
 180                         }
 181                         st->next = statelist;
 182                         st->devnum = INT_MAX;
 183                         st->percent = -2;
 184                         st->expected_spares = mdlist->spare_disks;
 185                         if (mdlist->spare_group)
 186                                 st->spare_group = strdup(mdlist->spare_group);
 187                         statelist = st;
 188                 }
 189         } else {
 190                 struct mddev_dev *dv;
 191                 for (dv=devlist ; dv; dv=dv->next) {
 192                         struct mddev_ident *mdlist = conf_get_ident(dv->devname);
 193                         struct state *st = calloc(1, sizeof *st);
 194                         if (st == NULL)
 195                                 continue;
 196                         st->devname = strdup(dv->devname);
 197                         st->next = statelist;
 198                         st->devnum = INT_MAX;
 199                         st->percent = -2;
 200                         st->expected_spares = -1;
 201                         if (mdlist) {
 202                                 st->expected_spares = mdlist->spare_disks;
 203                                 if (mdlist->spare_group)
 204                                         st->spare_group = strdup(mdlist->spare_group);
 205                         }
 206                         statelist = st;
 207                 }
 208         }
 209
 210
 211         while (! finished) {
 212                 int new_found = 0;
 213                 struct state *st;
 214                 int anydegraded = 0;
 215
 216                 if (mdstat)
 217                         free_mdstat(mdstat);
 218                 mdstat = mdstat_read(oneshot?0:1, 0);
 219
 220                 for (st=statelist; st; st=st->next)
 221                         if (check_array(st, mdstat, test, &info, increments))
 222                                 anydegraded = 1;
 223
 224                 /* now check if there are any new devices found in mdstat */
 225                 if (scan)
 226                         new_found = add_new_arrays(mdstat, statelist, test,
 227                                                    &info);
 228
 229                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 230                  * Look for another array with spare > 0 and active == raid and same spare_group
 231                  *  if found, choose a device and hotremove/hotadd
 232                  */
 233                 if (share && anydegraded)
 234                         try_spare_migration(statelist, &info);
 235                 if (!new_found) {
 236                         if (oneshot)
 237                                 break;
 238                         else
 239                                 mdstat_wait(period);
 240                 }
 241                 test = 0;
 242         }
 243         if (pidfile)
 244                 unlink(pidfile);
 245         return 0;
 246 }
 247
 248 static int make_daemon(char *pidfile)
 249 {
 250         int pid = fork();
 251         if (pid > 0) {
 252                 if (!pidfile)
 253                         printf("%d\n", pid);
 254                 else {
 255                         FILE *pid_file;
 256                         pid_file=fopen(pidfile, "w");
 257                         if (!pid_file)
 258                                 perror("cannot create pid file");
 259                         else {
 260                                 fprintf(pid_file,"%d\n", pid);
 261                                 fclose(pid_file);
 262                         }
 263                 }
 264                 return 0;
 265         }
 266         if (pid < 0) {
 267                 perror("daemonise");
 268                 return 1;
 269         }
 270         close(0);
 271         open("/dev/null", O_RDWR);
 272         dup2(0,1);
 273         dup2(0,2);
 274         setsid();
 275         return 0;
 276 }
 277
 278 static int check_one_sharer(int scan)
 279 {
 280         int pid, rv;
 281         FILE *fp;
 282         char dir[20];
 283         struct stat buf;
 284         fp = fopen("/var/run/mdadm/autorebuild.pid", "r");
 285         if (fp) {
 286                 fscanf(fp, "%d", &pid);
 287                 sprintf(dir, "/proc/%d", pid);
 288                 rv = stat(dir, &buf);
 289                 if (rv != -1) {
 290                         if (scan) {
 291                                 fprintf(stderr, Name ": Only one "
 292                                         "autorebuild process allowed"
 293                                         " in scan mode, aborting\n");
 294                                 fclose(fp);
 295                                 return 1;
 296                         } else {
 297                                 fprintf(stderr, Name ": Warning: One"
 298                                         " autorebuild process already"
 299                                         " running.");
 300                         }
 301                 }
 302                 fclose(fp);
 303         }
 304         if (scan) {
 305                 fp = fopen("/var/run/mdadm/autorebuild.pid", "w");
 306                 if (!fp)
 307                         fprintf(stderr, Name ": Cannot create"
 308                                 " autorebuild.pid "
 309                                 "file\n");
 310                 else {
 311                         pid = getpid();
 312                         fprintf(fp, "%d\n", pid);
 313                         fclose(fp);
 314                 }
 315         }
 316         return 0;
 317 }
 318
 319 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
 320 {
 321         int priority;
 322
 323         if (!info->alert_cmd && !info->mailaddr) {
 324                 time_t now = time(0);
 325
 326                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 327         }
 328         if (info->alert_cmd) {
 329                 int pid = fork();
 330                 switch(pid) {
 331                 default:
 332                         waitpid(pid, NULL, 0);
 333                         break;
 334                 case -1:
 335                         break;
 336                 case 0:
 337                         execl(info->alert_cmd, info->alert_cmd,
 338                               event, dev, disc, NULL);
 339                         exit(2);
 340                 }
 341         }
 342         if (info->mailaddr &&
 343             (strncmp(event, "Fail", 4)==0 ||
 344              strncmp(event, "Test", 4)==0 ||
 345              strncmp(event, "Spares", 6)==0 ||
 346              strncmp(event, "Degrade", 7)==0)) {
 347                 FILE *mp = popen(Sendmail, "w");
 348                 if (mp) {
 349                         FILE *mdstat;
 350                         char hname[256];
 351                         gethostname(hname, sizeof(hname));
 352                         signal(SIGPIPE, SIG_IGN);
 353                         if (info->mailfrom)
 354                                 fprintf(mp, "From: %s\n", info->mailfrom);
 355                         else
 356                                 fprintf(mp, "From: " Name " monitoring <root>\n");
 357                         fprintf(mp, "To: %s\n", info->mailaddr);
 358                         fprintf(mp, "Subject: %s event on %s:%s\n\n",
 359                                 event, dev, hname);
 360
 361                         fprintf(mp,
 362                                 "This is an automatically generated"
 363                                 " mail message from " Name "\n");
 364                         fprintf(mp, "running on %s\n\n", hname);
 365
 366                         fprintf(mp,
 367                                 "A %s event had been detected on"
 368                                 " md device %s.\n\n", event, dev);
 369
 370                         if (disc && disc[0] != ' ')
 371                                 fprintf(mp,
 372                                         "It could be related to"
 373                                         " component device %s.\n\n", disc);
 374                         if (disc && disc[0] == ' ')
 375                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 376
 377                         fprintf(mp, "Faithfully yours, etc.\n");
 378
 379                         mdstat = fopen("/proc/mdstat", "r");
 380                         if (mdstat) {
 381                                 char buf[8192];
 382                                 int n;
 383                                 fprintf(mp,
 384                                         "\nP.S. The /proc/mdstat file"
 385                                         " currently contains the following:\n\n");
 386                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 387                                         n=fwrite(buf, 1, n, mp);
 388                                 fclose(mdstat);
 389                         }
 390                         pclose(mp);
 391                 }
 392         }
 393
 394         /* log the event to syslog maybe */
 395         if (info->dosyslog) {
 396                 /* Log at a different severity depending on the event.
 397                  *
 398                  * These are the critical events:  */
 399                 if (strncmp(event, "Fail", 4)==0 ||
 400                     strncmp(event, "Degrade", 7)==0 ||
 401                     strncmp(event, "DeviceDisappeared", 17)==0)
 402                         priority = LOG_CRIT;
 403                 /* Good to know about, but are not failures: */
 404                 else if (strncmp(event, "Rebuild", 7)==0 ||
 405                          strncmp(event, "MoveSpare", 9)==0 ||
 406                          strncmp(event, "Spares", 6) != 0)
 407                         priority = LOG_WARNING;
 408                 /* Everything else: */
 409                 else
 410                         priority = LOG_INFO;
 411
 412                 if (disc)
 413                         syslog(priority,
 414                                "%s event detected on md device %s,"
 415                                " component device %s", event, dev, disc);
 416                 else
 417                         syslog(priority,
 418                                "%s event detected on md device %s",
 419                                event, dev);
 420         }
 421 }
 422
 423 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 424                        int test, struct alert_info *ainfo,
 425                        int increments)
 426 {
 427         /* Update the state 'st' to reflect any changes shown in mdstat,
 428          * or found by directly examining the array, and return
 429          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 430          */
 431         struct { int state, major, minor; } info[MaxDisks];
 432         mdu_array_info_t array;
 433         struct mdstat_ent *mse = NULL, *mse2;
 434         char *dev = st->devname;
 435         int fd;
 436         int i;
 437
 438         if (test)
 439                 alert("TestMessage", dev, NULL, ainfo);
 440         fd = open(dev, O_RDONLY);
 441         if (fd < 0) {
 442                 if (!st->err)
 443                         alert("DeviceDisappeared", dev, NULL, ainfo);
 444                 st->err=1;
 445                 return 0;
 446         }
 447         fcntl(fd, F_SETFD, FD_CLOEXEC);
 448         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 449                 if (!st->err)
 450                         alert("DeviceDisappeared", dev, NULL, ainfo);
 451                 st->err=1;
 452                 close(fd);
 453                 return 0;
 454         }
 455         /* It's much easier to list what array levels can't
 456          * have a device disappear than all of them that can
 457          */
 458         if (array.level == 0 || array.level == -1) {
 459                 if (!st->err)
 460                         alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
 461                 st->err = 1;
 462                 close(fd);
 463                 return 0;
 464         }
 465         if (st->devnum == INT_MAX) {
 466                 struct stat stb;
 467                 if (fstat(fd, &stb) == 0 &&
 468                     (S_IFMT&stb.st_mode)==S_IFBLK) {
 469                         if (major(stb.st_rdev) == MD_MAJOR)
 470                                 st->devnum = minor(stb.st_rdev);
 471                         else
 472                                 st->devnum = -1- (minor(stb.st_rdev)>>6);
 473                 }
 474         }
 475
 476         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 477                 if (mse2->devnum == st->devnum) {
 478                         mse2->devnum = INT_MAX; /* flag it as "used" */
 479                         mse = mse2;
 480                 }
 481
 482         if (!mse) {
 483                 /* duplicated array in statelist
 484                  * or re-created after reading mdstat*/
 485                 st->err = 1;
 486                 close(fd);
 487                 return 0;
 488         }
 489         /* this array is in /proc/mdstat */
 490         if (array.utime == 0)
 491                 /* external arrays don't update utime, so
 492                  * just make sure it is always different. */
 493                 array.utime = st->utime + 1;;
 494
 495         if (st->utime == array.utime &&
 496             st->failed == array.failed_disks &&
 497             st->working == array.working_disks &&
 498             st->spare == array.spare_disks &&
 499             (mse == NULL  || (
 500                     mse->percent == st->percent
 501                     ))) {
 502                 close(fd);
 503                 st->err = 0;
 504                 if ((st->active < st->raid) && st->spare == 0)
 505                         return 1;
 506                 else
 507                         return 0;
 508         }
 509         if (st->utime == 0 && /* new array */
 510             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 511                 )
 512                 alert("DegradedArray", dev, NULL, ainfo);
 513
 514         if (st->utime == 0 && /* new array */
 515             st->expected_spares > 0 &&
 516             array.spare_disks < st->expected_spares)
 517                 alert("SparesMissing", dev, NULL, ainfo);
 518         if (st->percent == -1 &&
 519             mse->percent >= 0)
 520                 alert("RebuildStarted", dev, NULL, ainfo);
 521         if (st->percent >= 0 &&
 522             mse->percent >= 0 &&
 523             (mse->percent / increments) > (st->percent / increments)) {
 524                 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 525
 526                 if((mse->percent / increments) == 0)
 527                         snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
 528                 else
 529                         snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
 530
 531                 alert(percentalert, dev, NULL, ainfo);
 532         }
 533
 534         if (mse->percent == -1 &&
 535             st->percent >= 0) {
 536                 /* Rebuild/sync/whatever just finished.
 537                  * If there is a number in /mismatch_cnt,
 538                  * we should report that.
 539                  */
 540                 struct mdinfo *sra =
 541                         sysfs_read(-1, st->devnum, GET_MISMATCH);
 542                 if (sra && sra->mismatch_cnt > 0) {
 543                         char cnt[40];
 544                         sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt);
 545                         alert("RebuildFinished", dev, cnt, ainfo);
 546                 } else
 547                         alert("RebuildFinished", dev, NULL, ainfo);
 548                 if (sra)
 549                         free(sra);
 550         }
 551         st->percent = mse->percent;
 552
 553         for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks;
 554              i++) {
 555                 mdu_disk_info_t disc;
 556                 disc.number = i;
 557                 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
 558                         info[i].state = disc.state;
 559                         info[i].major = disc.major;
 560                         info[i].minor = disc.minor;
 561                 } else
 562                         info[i].major = info[i].minor = 0;
 563         }
 564
 565         if (strncmp(mse->metadata_version, "external:", 9) == 0 &&
 566             is_subarray(mse->metadata_version+9))
 567                 st->parent_dev =
 568                         devname2devnum(mse->metadata_version+10);
 569         else
 570                 st->parent_dev = NoMdDev;
 571         if (st->metadata == NULL &&
 572             st->parent_dev == NoMdDev)
 573                 st->metadata = super_by_fd(fd, NULL);
 574
 575         close(fd);
 576
 577         for (i=0; i<MaxDisks; i++) {
 578                 mdu_disk_info_t disc = {0,0,0,0,0};
 579                 int newstate=0;
 580                 int change;
 581                 char *dv = NULL;
 582                 disc.number = i;
 583                 if (i > array.raid_disks + array.nr_disks) {
 584                         newstate = 0;
 585                         disc.major = disc.minor = 0;
 586                 } else if (info[i].major || info[i].minor) {
 587                         newstate = info[i].state;
 588                         dv = map_dev(info[i].major, info[i].minor, 1);
 589                         disc.state = newstate;
 590                         disc.major = info[i].major;
 591                         disc.minor = info[i].minor;
 592                 } else if (mse &&  mse->pattern && i < (int)strlen(mse->pattern)) {
 593                         switch(mse->pattern[i]) {
 594                         case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
 595                         case '_': newstate = 0; break;
 596                         }
 597                         disc.major = disc.minor = 0;
 598                 }
 599                 if (dv == NULL && st->devid[i])
 600                         dv = map_dev(major(st->devid[i]),
 601                                      minor(st->devid[i]), 1);
 602                 change = newstate ^ st->devstate[i];
 603                 if (st->utime && change && !st->err) {
 604                         if (i < array.raid_disks &&
 605                             (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
 606                              ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
 607                              ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
 608                                 )
 609                                 alert("Fail", dev, dv, ainfo);
 610                         else if (i >= array.raid_disks &&
 611                                  (disc.major || disc.minor) &&
 612                                  st->devid[i] == makedev(disc.major, disc.minor) &&
 613                                  ((newstate&change)&(1<<MD_DISK_FAULTY))
 614                                 )
 615                                 alert("FailSpare", dev, dv, ainfo);
 616                         else if (i < array.raid_disks &&
 617                                  ! (newstate & (1<<MD_DISK_REMOVED)) &&
 618                                  (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
 619                                   ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
 620                                   ((newstate&change)&(1<<MD_DISK_SYNC)))
 621                                 )
 622                                 alert("SpareActive", dev, dv, ainfo);
 623                 }
 624                 st->devstate[i] = newstate;
 625                 st->devid[i] = makedev(disc.major, disc.minor);
 626         }
 627         st->active = array.active_disks;
 628         st->working = array.working_disks;
 629         st->spare = array.spare_disks;
 630         st->failed = array.failed_disks;
 631         st->utime = array.utime;
 632         st->raid = array.raid_disks;
 633         st->err = 0;
 634         if ((st->active < st->raid) && st->spare == 0)
 635                 return 1;
 636         return 0;
 637 }
 638
 639 static int add_new_arrays(struct mdstat_ent *mdstat, struct state *statelist,
 640                           int test, struct alert_info *info)
 641 {
 642         struct mdstat_ent *mse;
 643         int new_found = 0;
 644
 645         for (mse=mdstat; mse; mse=mse->next)
 646                 if (mse->devnum != INT_MAX &&
 647                     (!mse->level  || /* retrieve containers */
 648                      (strcmp(mse->level, "raid0") != 0 &&
 649                       strcmp(mse->level, "linear") != 0))
 650                         ) {
 651                         struct state *st = calloc(1, sizeof *st);
 652                         mdu_array_info_t array;
 653                         int fd;
 654                         if (st == NULL)
 655                                 continue;
 656                         st->devname = strdup(get_md_name(mse->devnum));
 657                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 658                             ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
 659                                 /* no such array */
 660                                 if (fd >=0) close(fd);
 661                                 put_md_name(st->devname);
 662                                 free(st->devname);
 663                                 if (st->metadata) {
 664                                         st->metadata->ss->free_super(st->metadata);
 665                                         free(st->metadata);
 666                                 }
 667                                 free(st);
 668                                 continue;
 669                         }
 670                         close(fd);
 671                         st->next = statelist;
 672                         st->err = 1;
 673                         st->devnum = mse->devnum;
 674                         st->percent = -2;
 675                         st->expected_spares = -1;
 676                         if (strncmp(mse->metadata_version, "external:", 9) == 0 &&
 677                             is_subarray(mse->metadata_version+9))
 678                                 st->parent_dev =
 679                                         devname2devnum(mse->metadata_version+10);
 680                         else
 681                                 st->parent_dev = NoMdDev;
 682                         statelist = st;
 683                         if (test)
 684                                 alert("TestMessage", st->devname, NULL, info);
 685                         alert("NewArray", st->devname, NULL, info);
 686                         new_found = 1;
 687                 }
 688         return new_found;
 689 }
 690
 691 unsigned long long min_spare_size_required(struct state *st)
 692 {
 693         int fd;
 694         unsigned long long rv = 0;
 695
 696         if (!st->metadata ||
 697             !st->metadata->ss->min_acceptable_spare_size)
 698                 return rv;
 699
 700         fd = open(st->devname, O_RDONLY);
 701         if (fd < 0)
 702                 return 0;
 703         st->metadata->ss->load_super(st->metadata, fd, st->devname);
 704         close(fd);
 705         rv = st->metadata->ss->min_acceptable_spare_size(st->metadata);
 706         st->metadata->ss->free_super(st->metadata);
 707
 708         return rv;
 709 }
 710
 711 static int move_spare(struct state *from, struct state *to,
 712                       dev_t devid,
 713                       struct alert_info *info)
 714 {
 715         struct mddev_dev devlist;
 716         char devname[20];
 717
 718         /* try to remove and add */
 719         int fd1 = open(to->devname, O_RDONLY);
 720         int fd2 = open(from->devname, O_RDONLY);
 721
 722         if (fd1 < 0 || fd2 < 0) {
 723                 if (fd1>=0) close(fd1);
 724                 if (fd2>=0) close(fd2);
 725                 return 0;
 726         }
 727
 728         devlist.next = NULL;
 729         devlist.used = 0;
 730         devlist.re_add = 0;
 731         devlist.writemostly = 0;
 732         devlist.devname = devname;
 733         sprintf(devname, "%d:%d", major(devid), minor(devid));
 734
 735         devlist.disposition = 'r';
 736         if (Manage_subdevs(from->devname, fd2, &devlist, -1, 0) == 0) {
 737                 devlist.disposition = 'a';
 738                 if (Manage_subdevs(to->devname, fd1, &devlist, -1, 0) == 0) {
 739                         alert("MoveSpare", to->devname, from->devname, info);
 740                         close(fd1);
 741                         close(fd2);
 742                         return 1;
 743                 }
 744                 else Manage_subdevs(from->devname, fd2, &devlist, -1, 0);
 745         }
 746         close(fd1);
 747         close(fd2);
 748         return 0;
 749 }
 750
 751 static int check_donor(struct state *from, struct state *to,
 752                        struct domainlist *domlist)
 753 {
 754         struct state *sub;
 755
 756         if (from == to)
 757                 return 0;
 758         if (from->parent)
 759                 /* Cannot move from a member */
 760                 return 0;
 761         if (from->err)
 762                 return 0;
 763         for (sub = from->subarray; sub; sub = sub->subarray)
 764                 /* If source array has degraded subarrays, don't
 765                  * remove anything
 766                  */
 767                 if (sub->active < sub->raid)
 768                         return 0;
 769         if (from->metadata->ss->external == 0)
 770                 if (from->active < from->raid)
 771                         return 0;
 772         if (from->spare <= 0)
 773                 return 0;
 774         if (domlist == NULL)
 775                 return 0;
 776         return 1;
 777 }
 778
 779 static dev_t choose_spare(struct state *from, struct state *to,
 780                         struct domainlist *domlist)
 781 {
 782         int d;
 783         dev_t dev = 0;
 784         unsigned long long min_size
 785                 = min_spare_size_required(to);
 786
 787         for (d = from->raid; !dev && d < MaxDisks; d++) {
 788                 if (from->devid[d] > 0 &&
 789                     from->devstate[d] == 0) {
 790                         struct dev_policy *pol;
 791                         unsigned long long dev_size;
 792
 793                         if (min_size &&
 794                             dev_size_from_id(from->devid[d], &dev_size) &&
 795                             dev_size < min_size)
 796                                 continue;
 797
 798                         pol = devnum_policy(from->devid[d]);
 799                         if (from->spare_group)
 800                                 pol_add(&pol, pol_domain,
 801                                         from->spare_group, NULL);
 802                         if (domain_test(domlist, pol, to->metadata->ss->name))
 803                             dev = from->devid[d];
 804                         dev_policy_free(pol);
 805                 }
 806         }
 807         return dev;
 808 }
 809
 810 static dev_t container_choose_spare(struct state *from, struct state *to,
 811                                   struct domainlist *domlist)
 812 {
 813         /* This is similar to choose_spare, but we cannot trust devstate,
 814          * so we need to read the metadata instead
 815          */
 816
 817         struct supertype *st = from->metadata;
 818         int fd = open(from->devname, O_RDONLY);
 819         int err;
 820         struct mdinfo *disks, *d;
 821         unsigned long long min_size
 822                 = min_spare_size_required(to);
 823         dev_t dev = 0;
 824
 825         if (fd < 0)
 826                 return 0;
 827         if (!st->ss->getinfo_super_disks)
 828                 return 0;
 829
 830         err = st->ss->load_container(st, fd, NULL);
 831         close(fd);
 832         if (err)
 833                 return 0;
 834
 835         disks = st->ss->getinfo_super_disks(st);
 836         st->ss->free_super(st);
 837
 838         if (!disks)
 839                 return 0;
 840
 841         for (d = disks->devs ; d && !dev ; d = d->next) {
 842                 if (d->disk.state == 0) {
 843                         struct dev_policy *pol;
 844                         unsigned long long dev_size;
 845                         dev = makedev(d->disk.major,d->disk.minor);
 846
 847                         if (min_size &&
 848                             dev_size_from_id(dev,  &dev_size) &&
 849                             dev_size < min_size)
 850                                 continue;
 851
 852                         pol = devnum_policy(dev);
 853                         if (from->spare_group)
 854                                 pol_add(&pol, pol_domain,
 855                                         from->spare_group, NULL);
 856                         if (!domain_test(domlist, pol, to->metadata->ss->name))
 857                                 dev = 0;
 858
 859                         dev_policy_free(pol);
 860                 }
 861         }
 862         sysfs_free(disks);
 863         return dev;
 864 }
 865
 866
 867 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 868 {
 869         struct state *from;
 870         struct state *st;
 871
 872         link_containers_with_subarrays(statelist);
 873         for (st = statelist; st; st = st->next)
 874                 if (st->active < st->raid &&
 875                     st->spare == 0) {
 876                         struct domainlist *domlist = NULL;
 877                         int d;
 878                         struct state *to = st;
 879
 880                         if (to->parent)
 881                                 /* member of a container */
 882                                 to = to->parent;
 883
 884                         for (d = 0; d < MaxDisks; d++)
 885                                 if (to->devid[d])
 886                                         domainlist_add_dev(&domlist,
 887                                                            to->devid[d],
 888                                                            to->metadata->ss->name);
 889                         if (to->spare_group)
 890                                 domain_add(&domlist, to->spare_group);
 891
 892                         for (from=statelist ; from ; from=from->next) {
 893                                 dev_t devid;
 894                                 if (!check_donor(from, to, domlist))
 895                                         continue;
 896                                 if (from->metadata->ss->external)
 897                                         devid = container_choose_spare(
 898                                                 from, to, domlist);
 899                                 else
 900                                         devid = choose_spare(from, to, domlist);
 901                                 if (devid > 0
 902                                     && move_spare(from, to, devid, info))
 903                                                 break;
 904                         }
 905                         domain_free(domlist);
 906                 }
 907 }
 908
 909 /* search the statelist to connect external
 910  * metadata subarrays with their containers
 911  * We always completely rebuild the tree from scratch as
 912  * that is safest considering the possibility of entries
 913  * disappearing or changing.
 914  */
 915 static void link_containers_with_subarrays(struct state *list)
 916 {
 917         struct state *st;
 918         struct state *cont;
 919         for (st = list; st; st = st->next) {
 920                 st->parent = NULL;
 921                 st->subarray = NULL;
 922         }
 923         for (st = list; st; st = st->next)
 924                 if (st->parent_dev != NoMdDev)
 925                         for (cont = list; cont; cont = cont->next)
 926                                 if (!cont->err &&
 927                                     cont->parent_dev == NoMdDev &&
 928                                     cont->devnum == st->parent_dev) {
 929                                         st->parent = cont;
 930                                         st->subarray = cont->subarray;
 931                                         cont->subarray = st;
 932                                         break;
 933                                 }
 934 }
 935
 936 /* Not really Monitor but ... */
 937 int Wait(char *dev)
 938 {
 939         struct stat stb;
 940         int devnum;
 941         int rv = 1;
 942
 943         if (stat(dev, &stb) != 0) {
 944                 fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
 945                         strerror(errno));
 946                 return 2;
 947         }
 948         devnum = stat2devnum(&stb);
 949
 950         while(1) {
 951                 struct mdstat_ent *ms = mdstat_read(1, 0);
 952                 struct mdstat_ent *e;
 953
 954                 for (e=ms ; e; e=e->next)
 955                         if (e->devnum == devnum)
 956                                 break;
 957
 958                 if (!e || e->percent < 0) {
 959                         if (e && e->metadata_version &&
 960                             strncmp(e->metadata_version, "external:", 9) == 0) {
 961                                 if (is_subarray(&e->metadata_version[9]))
 962                                         ping_monitor(&e->metadata_version[9]);
 963                                 else
 964                                         ping_monitor(devnum2devname(devnum));
 965                         }
 966                         free_mdstat(ms);
 967                         return rv;
 968                 }
 969                 free_mdstat(ms);
 970                 rv = 0;
 971                 mdstat_wait(5);
 972         }
 973 }