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.\n");
 300                         }
 301                 }
 302                 fclose(fp);
 303         }
 304         if (scan) {
 305                 if (mkdir("/var/run/mdadm", S_IRWXU) < 0 &&
 306                     errno != EEXIST) {
 307                         fprintf(stderr, Name ": Can't create "
 308                                 "autorebuild.pid file\n");
 309                 } else {
 310                         fp = fopen("/var/run/mdadm/autorebuild.pid", "w");
 311                         if (!fp)
 312                                 fprintf(stderr, Name ": Cannot create"
 313                                         " autorebuild.pid"
 314                                         "file\n");
 315                         else {
 316                                 pid = getpid();
 317                                 fprintf(fp, "%d\n", pid);
 318                                 fclose(fp);
 319                         }
 320                 }
 321         }
 322         return 0;
 323 }
 324
 325 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
 326 {
 327         int priority;
 328
 329         if (!info->alert_cmd && !info->mailaddr) {
 330                 time_t now = time(0);
 331
 332                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 333         }
 334         if (info->alert_cmd) {
 335                 int pid = fork();
 336                 switch(pid) {
 337                 default:
 338                         waitpid(pid, NULL, 0);
 339                         break;
 340                 case -1:
 341                         break;
 342                 case 0:
 343                         execl(info->alert_cmd, info->alert_cmd,
 344                               event, dev, disc, NULL);
 345                         exit(2);
 346                 }
 347         }
 348         if (info->mailaddr &&
 349             (strncmp(event, "Fail", 4)==0 ||
 350              strncmp(event, "Test", 4)==0 ||
 351              strncmp(event, "Spares", 6)==0 ||
 352              strncmp(event, "Degrade", 7)==0)) {
 353                 FILE *mp = popen(Sendmail, "w");
 354                 if (mp) {
 355                         FILE *mdstat;
 356                         char hname[256];
 357                         gethostname(hname, sizeof(hname));
 358                         signal(SIGPIPE, SIG_IGN);
 359                         if (info->mailfrom)
 360                                 fprintf(mp, "From: %s\n", info->mailfrom);
 361                         else
 362                                 fprintf(mp, "From: " Name " monitoring <root>\n");
 363                         fprintf(mp, "To: %s\n", info->mailaddr);
 364                         fprintf(mp, "Subject: %s event on %s:%s\n\n",
 365                                 event, dev, hname);
 366
 367                         fprintf(mp,
 368                                 "This is an automatically generated"
 369                                 " mail message from " Name "\n");
 370                         fprintf(mp, "running on %s\n\n", hname);
 371
 372                         fprintf(mp,
 373                                 "A %s event had been detected on"
 374                                 " md device %s.\n\n", event, dev);
 375
 376                         if (disc && disc[0] != ' ')
 377                                 fprintf(mp,
 378                                         "It could be related to"
 379                                         " component device %s.\n\n", disc);
 380                         if (disc && disc[0] == ' ')
 381                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 382
 383                         fprintf(mp, "Faithfully yours, etc.\n");
 384
 385                         mdstat = fopen("/proc/mdstat", "r");
 386                         if (mdstat) {
 387                                 char buf[8192];
 388                                 int n;
 389                                 fprintf(mp,
 390                                         "\nP.S. The /proc/mdstat file"
 391                                         " currently contains the following:\n\n");
 392                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 393                                         n=fwrite(buf, 1, n, mp);
 394                                 fclose(mdstat);
 395                         }
 396                         pclose(mp);
 397                 }
 398         }
 399
 400         /* log the event to syslog maybe */
 401         if (info->dosyslog) {
 402                 /* Log at a different severity depending on the event.
 403                  *
 404                  * These are the critical events:  */
 405                 if (strncmp(event, "Fail", 4)==0 ||
 406                     strncmp(event, "Degrade", 7)==0 ||
 407                     strncmp(event, "DeviceDisappeared", 17)==0)
 408                         priority = LOG_CRIT;
 409                 /* Good to know about, but are not failures: */
 410                 else if (strncmp(event, "Rebuild", 7)==0 ||
 411                          strncmp(event, "MoveSpare", 9)==0 ||
 412                          strncmp(event, "Spares", 6) != 0)
 413                         priority = LOG_WARNING;
 414                 /* Everything else: */
 415                 else
 416                         priority = LOG_INFO;
 417
 418                 if (disc)
 419                         syslog(priority,
 420                                "%s event detected on md device %s,"
 421                                " component device %s", event, dev, disc);
 422                 else
 423                         syslog(priority,
 424                                "%s event detected on md device %s",
 425                                event, dev);
 426         }
 427 }
 428
 429 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 430                        int test, struct alert_info *ainfo,
 431                        int increments)
 432 {
 433         /* Update the state 'st' to reflect any changes shown in mdstat,
 434          * or found by directly examining the array, and return
 435          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 436          */
 437         struct { int state, major, minor; } info[MaxDisks];
 438         mdu_array_info_t array;
 439         struct mdstat_ent *mse = NULL, *mse2;
 440         char *dev = st->devname;
 441         int fd;
 442         int i;
 443
 444         if (test)
 445                 alert("TestMessage", dev, NULL, ainfo);
 446         fd = open(dev, O_RDONLY);
 447         if (fd < 0) {
 448                 if (!st->err)
 449                         alert("DeviceDisappeared", dev, NULL, ainfo);
 450                 st->err=1;
 451                 return 0;
 452         }
 453         fcntl(fd, F_SETFD, FD_CLOEXEC);
 454         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 455                 if (!st->err)
 456                         alert("DeviceDisappeared", dev, NULL, ainfo);
 457                 st->err=1;
 458                 close(fd);
 459                 return 0;
 460         }
 461         /* It's much easier to list what array levels can't
 462          * have a device disappear than all of them that can
 463          */
 464         if (array.level == 0 || array.level == -1) {
 465                 if (!st->err)
 466                         alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
 467                 st->err = 1;
 468                 close(fd);
 469                 return 0;
 470         }
 471         if (st->devnum == INT_MAX) {
 472                 struct stat stb;
 473                 if (fstat(fd, &stb) == 0 &&
 474                     (S_IFMT&stb.st_mode)==S_IFBLK) {
 475                         if (major(stb.st_rdev) == MD_MAJOR)
 476                                 st->devnum = minor(stb.st_rdev);
 477                         else
 478                                 st->devnum = -1- (minor(stb.st_rdev)>>6);
 479                 }
 480         }
 481
 482         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 483                 if (mse2->devnum == st->devnum) {
 484                         mse2->devnum = INT_MAX; /* flag it as "used" */
 485                         mse = mse2;
 486                 }
 487
 488         if (!mse) {
 489                 /* duplicated array in statelist
 490                  * or re-created after reading mdstat*/
 491                 st->err = 1;
 492                 close(fd);
 493                 return 0;
 494         }
 495         /* this array is in /proc/mdstat */
 496         if (array.utime == 0)
 497                 /* external arrays don't update utime, so
 498                  * just make sure it is always different. */
 499                 array.utime = st->utime + 1;;
 500
 501         if (st->utime == array.utime &&
 502             st->failed == array.failed_disks &&
 503             st->working == array.working_disks &&
 504             st->spare == array.spare_disks &&
 505             (mse == NULL  || (
 506                     mse->percent == st->percent
 507                     ))) {
 508                 close(fd);
 509                 st->err = 0;
 510                 if ((st->active < st->raid) && st->spare == 0)
 511                         return 1;
 512                 else
 513                         return 0;
 514         }
 515         if (st->utime == 0 && /* new array */
 516             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 517                 )
 518                 alert("DegradedArray", dev, NULL, ainfo);
 519
 520         if (st->utime == 0 && /* new array */
 521             st->expected_spares > 0 &&
 522             array.spare_disks < st->expected_spares)
 523                 alert("SparesMissing", dev, NULL, ainfo);
 524         if (st->percent == -1 &&
 525             mse->percent >= 0)
 526                 alert("RebuildStarted", dev, NULL, ainfo);
 527         if (st->percent >= 0 &&
 528             mse->percent >= 0 &&
 529             (mse->percent / increments) > (st->percent / increments)) {
 530                 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 531
 532                 if((mse->percent / increments) == 0)
 533                         snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
 534                 else
 535                         snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
 536
 537                 alert(percentalert, dev, NULL, ainfo);
 538         }
 539
 540         if (mse->percent == -1 &&
 541             st->percent >= 0) {
 542                 /* Rebuild/sync/whatever just finished.
 543                  * If there is a number in /mismatch_cnt,
 544                  * we should report that.
 545                  */
 546                 struct mdinfo *sra =
 547                         sysfs_read(-1, st->devnum, GET_MISMATCH);
 548                 if (sra && sra->mismatch_cnt > 0) {
 549                         char cnt[40];
 550                         sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt);
 551                         alert("RebuildFinished", dev, cnt, ainfo);
 552                 } else
 553                         alert("RebuildFinished", dev, NULL, ainfo);
 554                 if (sra)
 555                         free(sra);
 556         }
 557         st->percent = mse->percent;
 558
 559         for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks;
 560              i++) {
 561                 mdu_disk_info_t disc;
 562                 disc.number = i;
 563                 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
 564                         info[i].state = disc.state;
 565                         info[i].major = disc.major;
 566                         info[i].minor = disc.minor;
 567                 } else
 568                         info[i].major = info[i].minor = 0;
 569         }
 570
 571         if (strncmp(mse->metadata_version, "external:", 9) == 0 &&
 572             is_subarray(mse->metadata_version+9))
 573                 st->parent_dev =
 574                         devname2devnum(mse->metadata_version+10);
 575         else
 576                 st->parent_dev = NoMdDev;
 577         if (st->metadata == NULL &&
 578             st->parent_dev == NoMdDev)
 579                 st->metadata = super_by_fd(fd, NULL);
 580
 581         close(fd);
 582
 583         for (i=0; i<MaxDisks; i++) {
 584                 mdu_disk_info_t disc = {0,0,0,0,0};
 585                 int newstate=0;
 586                 int change;
 587                 char *dv = NULL;
 588                 disc.number = i;
 589                 if (i > array.raid_disks + array.nr_disks) {
 590                         newstate = 0;
 591                         disc.major = disc.minor = 0;
 592                 } else if (info[i].major || info[i].minor) {
 593                         newstate = info[i].state;
 594                         dv = map_dev(info[i].major, info[i].minor, 1);
 595                         disc.state = newstate;
 596                         disc.major = info[i].major;
 597                         disc.minor = info[i].minor;
 598                 } else if (mse &&  mse->pattern && i < (int)strlen(mse->pattern)) {
 599                         switch(mse->pattern[i]) {
 600                         case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
 601                         case '_': newstate = 0; break;
 602                         }
 603                         disc.major = disc.minor = 0;
 604                 }
 605                 if (dv == NULL && st->devid[i])
 606                         dv = map_dev(major(st->devid[i]),
 607                                      minor(st->devid[i]), 1);
 608                 change = newstate ^ st->devstate[i];
 609                 if (st->utime && change && !st->err) {
 610                         if (i < array.raid_disks &&
 611                             (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
 612                              ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
 613                              ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
 614                                 )
 615                                 alert("Fail", dev, dv, ainfo);
 616                         else if (i >= array.raid_disks &&
 617                                  (disc.major || disc.minor) &&
 618                                  st->devid[i] == makedev(disc.major, disc.minor) &&
 619                                  ((newstate&change)&(1<<MD_DISK_FAULTY))
 620                                 )
 621                                 alert("FailSpare", dev, dv, ainfo);
 622                         else if (i < array.raid_disks &&
 623                                  ! (newstate & (1<<MD_DISK_REMOVED)) &&
 624                                  (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
 625                                   ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
 626                                   ((newstate&change)&(1<<MD_DISK_SYNC)))
 627                                 )
 628                                 alert("SpareActive", dev, dv, ainfo);
 629                 }
 630                 st->devstate[i] = newstate;
 631                 st->devid[i] = makedev(disc.major, disc.minor);
 632         }
 633         st->active = array.active_disks;
 634         st->working = array.working_disks;
 635         st->spare = array.spare_disks;
 636         st->failed = array.failed_disks;
 637         st->utime = array.utime;
 638         st->raid = array.raid_disks;
 639         st->err = 0;
 640         if ((st->active < st->raid) && st->spare == 0)
 641                 return 1;
 642         return 0;
 643 }
 644
 645 static int add_new_arrays(struct mdstat_ent *mdstat, struct state *statelist,
 646                           int test, struct alert_info *info)
 647 {
 648         struct mdstat_ent *mse;
 649         int new_found = 0;
 650
 651         for (mse=mdstat; mse; mse=mse->next)
 652                 if (mse->devnum != INT_MAX &&
 653                     (!mse->level  || /* retrieve containers */
 654                      (strcmp(mse->level, "raid0") != 0 &&
 655                       strcmp(mse->level, "linear") != 0))
 656                         ) {
 657                         struct state *st = calloc(1, sizeof *st);
 658                         mdu_array_info_t array;
 659                         int fd;
 660                         if (st == NULL)
 661                                 continue;
 662                         st->devname = strdup(get_md_name(mse->devnum));
 663                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 664                             ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
 665                                 /* no such array */
 666                                 if (fd >=0) close(fd);
 667                                 put_md_name(st->devname);
 668                                 free(st->devname);
 669                                 if (st->metadata) {
 670                                         st->metadata->ss->free_super(st->metadata);
 671                                         free(st->metadata);
 672                                 }
 673                                 free(st);
 674                                 continue;
 675                         }
 676                         close(fd);
 677                         st->next = statelist;
 678                         st->err = 1;
 679                         st->devnum = mse->devnum;
 680                         st->percent = -2;
 681                         st->expected_spares = -1;
 682                         if (strncmp(mse->metadata_version, "external:", 9) == 0 &&
 683                             is_subarray(mse->metadata_version+9))
 684                                 st->parent_dev =
 685                                         devname2devnum(mse->metadata_version+10);
 686                         else
 687                                 st->parent_dev = NoMdDev;
 688                         statelist = st;
 689                         if (test)
 690                                 alert("TestMessage", st->devname, NULL, info);
 691                         alert("NewArray", st->devname, NULL, info);
 692                         new_found = 1;
 693                 }
 694         return new_found;
 695 }
 696
 697 unsigned long long min_spare_size_required(struct state *st)
 698 {
 699         int fd;
 700         unsigned long long rv = 0;
 701
 702         if (!st->metadata ||
 703             !st->metadata->ss->min_acceptable_spare_size)
 704                 return rv;
 705
 706         fd = open(st->devname, O_RDONLY);
 707         if (fd < 0)
 708                 return 0;
 709         st->metadata->ss->load_super(st->metadata, fd, st->devname);
 710         close(fd);
 711         rv = st->metadata->ss->min_acceptable_spare_size(st->metadata);
 712         st->metadata->ss->free_super(st->metadata);
 713
 714         return rv;
 715 }
 716
 717 static int move_spare(struct state *from, struct state *to,
 718                       dev_t devid,
 719                       struct alert_info *info)
 720 {
 721         struct mddev_dev devlist;
 722         char devname[20];
 723
 724         /* try to remove and add */
 725         int fd1 = open(to->devname, O_RDONLY);
 726         int fd2 = open(from->devname, O_RDONLY);
 727
 728         if (fd1 < 0 || fd2 < 0) {
 729                 if (fd1>=0) close(fd1);
 730                 if (fd2>=0) close(fd2);
 731                 return 0;
 732         }
 733
 734         devlist.next = NULL;
 735         devlist.used = 0;
 736         devlist.re_add = 0;
 737         devlist.writemostly = 0;
 738         devlist.devname = devname;
 739         sprintf(devname, "%d:%d", major(devid), minor(devid));
 740
 741         devlist.disposition = 'r';
 742         if (Manage_subdevs(from->devname, fd2, &devlist, -1, 0) == 0) {
 743                 devlist.disposition = 'a';
 744                 if (Manage_subdevs(to->devname, fd1, &devlist, -1, 0) == 0) {
 745                         alert("MoveSpare", to->devname, from->devname, info);
 746                         close(fd1);
 747                         close(fd2);
 748                         return 1;
 749                 }
 750                 else Manage_subdevs(from->devname, fd2, &devlist, -1, 0);
 751         }
 752         close(fd1);
 753         close(fd2);
 754         return 0;
 755 }
 756
 757 static int check_donor(struct state *from, struct state *to,
 758                        struct domainlist *domlist)
 759 {
 760         struct state *sub;
 761
 762         if (from == to)
 763                 return 0;
 764         if (from->parent)
 765                 /* Cannot move from a member */
 766                 return 0;
 767         if (from->err)
 768                 return 0;
 769         for (sub = from->subarray; sub; sub = sub->subarray)
 770                 /* If source array has degraded subarrays, don't
 771                  * remove anything
 772                  */
 773                 if (sub->active < sub->raid)
 774                         return 0;
 775         if (from->metadata->ss->external == 0)
 776                 if (from->active < from->raid)
 777                         return 0;
 778         if (from->spare <= 0)
 779                 return 0;
 780         if (domlist == NULL)
 781                 return 0;
 782         return 1;
 783 }
 784
 785 static dev_t choose_spare(struct state *from, struct state *to,
 786                         struct domainlist *domlist)
 787 {
 788         int d;
 789         dev_t dev = 0;
 790         unsigned long long min_size
 791                 = min_spare_size_required(to);
 792
 793         for (d = from->raid; !dev && d < MaxDisks; d++) {
 794                 if (from->devid[d] > 0 &&
 795                     from->devstate[d] == 0) {
 796                         struct dev_policy *pol;
 797                         unsigned long long dev_size;
 798
 799                         if (min_size &&
 800                             dev_size_from_id(from->devid[d], &dev_size) &&
 801                             dev_size < min_size)
 802                                 continue;
 803
 804                         pol = devnum_policy(from->devid[d]);
 805                         if (from->spare_group)
 806                                 pol_add(&pol, pol_domain,
 807                                         from->spare_group, NULL);
 808                         if (domain_test(domlist, pol, to->metadata->ss->name))
 809                             dev = from->devid[d];
 810                         dev_policy_free(pol);
 811                 }
 812         }
 813         return dev;
 814 }
 815
 816 static dev_t container_choose_spare(struct state *from, struct state *to,
 817                                   struct domainlist *domlist)
 818 {
 819         /* This is similar to choose_spare, but we cannot trust devstate,
 820          * so we need to read the metadata instead
 821          */
 822
 823         struct supertype *st = from->metadata;
 824         int fd = open(from->devname, O_RDONLY);
 825         int err;
 826         struct mdinfo *disks, *d;
 827         unsigned long long min_size
 828                 = min_spare_size_required(to);
 829         dev_t dev = 0;
 830
 831         if (fd < 0)
 832                 return 0;
 833         if (!st->ss->getinfo_super_disks)
 834                 return 0;
 835
 836         err = st->ss->load_container(st, fd, NULL);
 837         close(fd);
 838         if (err)
 839                 return 0;
 840
 841         disks = st->ss->getinfo_super_disks(st);
 842         st->ss->free_super(st);
 843
 844         if (!disks)
 845                 return 0;
 846
 847         for (d = disks->devs ; d && !dev ; d = d->next) {
 848                 if (d->disk.state == 0) {
 849                         struct dev_policy *pol;
 850                         unsigned long long dev_size;
 851                         dev = makedev(d->disk.major,d->disk.minor);
 852
 853                         if (min_size &&
 854                             dev_size_from_id(dev,  &dev_size) &&
 855                             dev_size < min_size) {
 856                                 dev = 0;
 857                                 continue;
 858                         }
 859                         pol = devnum_policy(dev);
 860                         if (from->spare_group)
 861                                 pol_add(&pol, pol_domain,
 862                                         from->spare_group, NULL);
 863                         if (!domain_test(domlist, pol, to->metadata->ss->name))
 864                                 dev = 0;
 865
 866                         dev_policy_free(pol);
 867                 }
 868         }
 869         sysfs_free(disks);
 870         return dev;
 871 }
 872
 873
 874 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 875 {
 876         struct state *from;
 877         struct state *st;
 878
 879         link_containers_with_subarrays(statelist);
 880         for (st = statelist; st; st = st->next)
 881                 if (st->active < st->raid &&
 882                     st->spare == 0 && !st->err) {
 883                         struct domainlist *domlist = NULL;
 884                         int d;
 885                         struct state *to = st;
 886
 887                         if (to->parent)
 888                                 /* member of a container */
 889                                 to = to->parent;
 890
 891                         for (d = 0; d < MaxDisks; d++)
 892                                 if (to->devid[d])
 893                                         domainlist_add_dev(&domlist,
 894                                                            to->devid[d],
 895                                                            to->metadata->ss->name);
 896                         if (to->spare_group)
 897                                 domain_add(&domlist, to->spare_group);
 898
 899                         for (from=statelist ; from ; from=from->next) {
 900                                 dev_t devid;
 901                                 if (!check_donor(from, to, domlist))
 902                                         continue;
 903                                 if (from->metadata->ss->external)
 904                                         devid = container_choose_spare(
 905                                                 from, to, domlist);
 906                                 else
 907                                         devid = choose_spare(from, to, domlist);
 908                                 if (devid > 0
 909                                     && move_spare(from, to, devid, info))
 910                                                 break;
 911                         }
 912                         domain_free(domlist);
 913                 }
 914 }
 915
 916 /* search the statelist to connect external
 917  * metadata subarrays with their containers
 918  * We always completely rebuild the tree from scratch as
 919  * that is safest considering the possibility of entries
 920  * disappearing or changing.
 921  */
 922 static void link_containers_with_subarrays(struct state *list)
 923 {
 924         struct state *st;
 925         struct state *cont;
 926         for (st = list; st; st = st->next) {
 927                 st->parent = NULL;
 928                 st->subarray = NULL;
 929         }
 930         for (st = list; st; st = st->next)
 931                 if (st->parent_dev != NoMdDev)
 932                         for (cont = list; cont; cont = cont->next)
 933                                 if (!cont->err &&
 934                                     cont->parent_dev == NoMdDev &&
 935                                     cont->devnum == st->parent_dev) {
 936                                         st->parent = cont;
 937                                         st->subarray = cont->subarray;
 938                                         cont->subarray = st;
 939                                         break;
 940                                 }
 941 }
 942
 943 /* Not really Monitor but ... */
 944 int Wait(char *dev)
 945 {
 946         struct stat stb;
 947         int devnum;
 948         int rv = 1;
 949
 950         if (stat(dev, &stb) != 0) {
 951                 fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
 952                         strerror(errno));
 953                 return 2;
 954         }
 955         devnum = stat2devnum(&stb);
 956
 957         while(1) {
 958                 struct mdstat_ent *ms = mdstat_read(1, 0);
 959                 struct mdstat_ent *e;
 960
 961                 for (e=ms ; e; e=e->next)
 962                         if (e->devnum == devnum)
 963                                 break;
 964
 965                 if (!e || e->percent < 0) {
 966                         if (e && e->metadata_version &&
 967                             strncmp(e->metadata_version, "external:", 9) == 0) {
 968                                 if (is_subarray(&e->metadata_version[9]))
 969                                         ping_monitor(&e->metadata_version[9]);
 970                                 else
 971                                         ping_monitor(devnum2devname(devnum));
 972                         }
 973                         free_mdstat(ms);
 974                         return rv;
 975                 }
 976                 free_mdstat(ms);
 977                 rv = 0;
 978                 mdstat_wait(5);
 979         }
 980 }