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 struct state {
  34         char *devname;
  35         char devnm[32]; /* to sync with mdstat info */
  36         long utime;
  37         int err;
  38         char *spare_group;
  39         int active, working, failed, spare, raid;
  40         int from_config;
  41         int expected_spares;
  42         int devstate[MAX_DISKS];
  43         dev_t devid[MAX_DISKS];
  44         int percent;
  45         char parent_devnm[32]; /* For subarray, devnm of parent.
  46                                 * For others, ""
  47                                 */
  48         struct supertype *metadata;
  49         struct state *subarray;/* for a container it is a link to first subarray
  50                                 * for a subarray it is a link to next subarray
  51                                 * in the same container */
  52         struct state *parent;  /* for a subarray it is a link to its container
  53                                 */
  54         struct state *next;
  55 };
  56
  57 struct alert_info {
  58         char *mailaddr;
  59         char *mailfrom;
  60         char *alert_cmd;
  61         int dosyslog;
  62 };
  63 static int make_daemon(char *pidfile);
  64 static int check_one_sharer(int scan);
  65 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
  66 static int check_array(struct state *st, struct mdstat_ent *mdstat,
  67                        int test, struct alert_info *info,
  68                        int increments, char *prefer);
  69 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
  70                           int test, struct alert_info *info);
  71 static void try_spare_migration(struct state *statelist, struct alert_info *info);
  72 static void link_containers_with_subarrays(struct state *list);
  73
  74 int Monitor(struct mddev_dev *devlist,
  75             char *mailaddr, char *alert_cmd,
  76             struct context *c,
  77             int daemonise, int oneshot,
  78             int dosyslog, char *pidfile, int increments,
  79             int share)
  80 {
  81         /*
  82          * Every few seconds, scan every md device looking for changes
  83          * When a change is found, log it, possibly run the alert command,
  84          * and possibly send Email
  85          *
  86          * For each array, we record:
  87          *   Update time
  88          *   active/working/failed/spare drives
  89          *   State of each device.
  90          *   %rebuilt if rebuilding
  91          *
  92          * If the update time changes, check out all the data again
  93          * It is possible that we cannot get the state of each device
  94          * due to bugs in the md kernel module.
  95          * We also read /proc/mdstat to get rebuild percent,
  96          * and to get state on all active devices incase of kernel bug.
  97          *
  98          * Events are:
  99          *    Fail
 100          *      An active device had Faulty set or Active/Sync removed
 101          *    FailSpare
 102          *      A spare device had Faulty set
 103          *    SpareActive
 104          *      An active device had a reverse transition
 105          *    RebuildStarted
 106          *      percent went from -1 to +ve
 107          *    RebuildNN
 108          *      percent went from below to not-below NN%
 109          *    DeviceDisappeared
 110          *      Couldn't access a device which was previously visible
 111          *
 112          * if we detect an array with active<raid and spare==0
 113          * we look at other arrays that have same spare-group
 114          * If we find one with active==raid and spare>0,
 115          *  and if we can get_disk_info and find a name
 116          *  Then we hot-remove and hot-add to the other array
 117          *
 118          * If devlist is NULL, then we can monitor everything because --scan
 119          * was given.  We get an initial list from config file and add anything
 120          * that appears in /proc/mdstat
 121          */
 122
 123         struct state *statelist = NULL;
 124         struct state *st2;
 125         int finished = 0;
 126         struct mdstat_ent *mdstat = NULL;
 127         char *mailfrom = NULL;
 128         struct alert_info info;
 129
 130         if (!mailaddr) {
 131                 mailaddr = conf_get_mailaddr();
 132                 if (mailaddr && ! c->scan)
 133                         pr_err("Monitor using email address \"%s\" from config file\n",
 134                                mailaddr);
 135         }
 136         mailfrom = conf_get_mailfrom();
 137
 138         if (!alert_cmd) {
 139                 alert_cmd = conf_get_program();
 140                 if (alert_cmd && ! c->scan)
 141                         pr_err("Monitor using program \"%s\" from config file\n",
 142                                alert_cmd);
 143         }
 144         if (c->scan && !mailaddr && !alert_cmd && !dosyslog) {
 145                 pr_err("No mail address or alert command - not monitoring.\n");
 146                 return 1;
 147         }
 148         info.alert_cmd = alert_cmd;
 149         info.mailaddr = mailaddr;
 150         info.mailfrom = mailfrom;
 151         info.dosyslog = dosyslog;
 152
 153         if (daemonise) {
 154                 int rv = make_daemon(pidfile);
 155                 if (rv >= 0)
 156                         return rv;
 157         }
 158
 159         if (share)
 160                 if (check_one_sharer(c->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 = xcalloc(1, sizeof *st);
 172                         if (mdlist->devname[0] == '/')
 173                                 st->devname = xstrdup(mdlist->devname);
 174                         else {
 175                                 st->devname = xmalloc(8+strlen(mdlist->devname)+1);
 176                                 strcpy(strcpy(st->devname, "/dev/md/"),
 177                                        mdlist->devname);
 178                         }
 179                         st->next = statelist;
 180                         st->devnm[0] = 0;
 181                         st->percent = RESYNC_UNKNOWN;
 182                         st->from_config = 1;
 183                         st->expected_spares = mdlist->spare_disks;
 184                         if (mdlist->spare_group)
 185                                 st->spare_group = xstrdup(mdlist->spare_group);
 186                         statelist = st;
 187                 }
 188         } else {
 189                 struct mddev_dev *dv;
 190                 for (dv=devlist ; dv; dv=dv->next) {
 191                         struct mddev_ident *mdlist = conf_get_ident(dv->devname);
 192                         struct state *st = xcalloc(1, sizeof *st);
 193                         st->devname = xstrdup(dv->devname);
 194                         st->next = statelist;
 195                         st->devnm[0] = 0;
 196                         st->percent = RESYNC_UNKNOWN;
 197                         st->expected_spares = -1;
 198                         if (mdlist) {
 199                                 st->expected_spares = mdlist->spare_disks;
 200                                 if (mdlist->spare_group)
 201                                         st->spare_group = xstrdup(mdlist->spare_group);
 202                         }
 203                         statelist = st;
 204                 }
 205         }
 206
 207         while (! finished) {
 208                 int new_found = 0;
 209                 struct state *st;
 210                 int anydegraded = 0;
 211
 212                 if (mdstat)
 213                         free_mdstat(mdstat);
 214                 mdstat = mdstat_read(oneshot?0:1, 0);
 215
 216                 for (st=statelist; st; st=st->next)
 217                         if (check_array(st, mdstat, c->test, &info,
 218                                         increments, c->prefer))
 219                                 anydegraded = 1;
 220
 221                 /* now check if there are any new devices found in mdstat */
 222                 if (c->scan)
 223                         new_found = add_new_arrays(mdstat, &statelist, c->test,
 224                                                    &info);
 225
 226                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 227                  * Look for another array with spare > 0 and active == raid and same spare_group
 228                  *  if found, choose a device and hotremove/hotadd
 229                  */
 230                 if (share && anydegraded)
 231                         try_spare_migration(statelist, &info);
 232                 if (!new_found) {
 233                         if (oneshot)
 234                                 break;
 235                         else
 236                                 mdstat_wait(c->delay);
 237                 }
 238                 c->test = 0;
 239         }
 240         for (st2 = statelist; st2; st2 = statelist) {
 241                 statelist = st2->next;
 242                 free(st2);
 243         }
 244
 245         if (pidfile)
 246                 unlink(pidfile);
 247         return 0;
 248 }
 249
 250 static int make_daemon(char *pidfile)
 251 {
 252         /* Return:
 253          * -1 in the forked daemon
 254          *  0 in the parent
 255          *  1 on error
 256          * so a none-negative becomes the exit code.
 257          */
 258         int pid = fork();
 259         if (pid > 0) {
 260                 if (!pidfile)
 261                         printf("%d\n", pid);
 262                 else {
 263                         FILE *pid_file;
 264                         pid_file=fopen(pidfile, "w");
 265                         if (!pid_file)
 266                                 perror("cannot create pid file");
 267                         else {
 268                                 fprintf(pid_file,"%d\n", pid);
 269                                 fclose(pid_file);
 270                         }
 271                 }
 272                 return 0;
 273         }
 274         if (pid < 0) {
 275                 perror("daemonise");
 276                 return 1;
 277         }
 278         close(0);
 279         open("/dev/null", O_RDWR);
 280         dup2(0,1);
 281         dup2(0,2);
 282         setsid();
 283         return -1;
 284 }
 285
 286 static int check_one_sharer(int scan)
 287 {
 288         int pid, rv;
 289         FILE *fp;
 290         char dir[20];
 291         char path[100];
 292         struct stat buf;
 293         sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
 294         fp = fopen(path, "r");
 295         if (fp) {
 296                 if (fscanf(fp, "%d", &pid) != 1)
 297                         pid = -1;
 298                 sprintf(dir, "/proc/%d", pid);
 299                 rv = stat(dir, &buf);
 300                 if (rv != -1) {
 301                         if (scan) {
 302                                 pr_err("Only one "
 303                                         "autorebuild process allowed"
 304                                         " in scan mode, aborting\n");
 305                                 fclose(fp);
 306                                 return 1;
 307                         } else {
 308                                 pr_err("Warning: One"
 309                                         " autorebuild process already"
 310                                         " running.\n");
 311                         }
 312                 }
 313                 fclose(fp);
 314         }
 315         if (scan) {
 316                 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
 317                     errno != EEXIST) {
 318                         pr_err("Can't create "
 319                                 "autorebuild.pid file\n");
 320                 } else {
 321                         fp = fopen(path, "w");
 322                         if (!fp)
 323                                 pr_err("Cannot create"
 324                                         " autorebuild.pid"
 325                                         "file\n");
 326                         else {
 327                                 pid = getpid();
 328                                 fprintf(fp, "%d\n", pid);
 329                                 fclose(fp);
 330                         }
 331                 }
 332         }
 333         return 0;
 334 }
 335
 336 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
 337 {
 338         int priority;
 339
 340         if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
 341                 time_t now = time(0);
 342
 343                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 344         }
 345         if (info->alert_cmd) {
 346                 int pid = fork();
 347                 switch(pid) {
 348                 default:
 349                         waitpid(pid, NULL, 0);
 350                         break;
 351                 case -1:
 352                         break;
 353                 case 0:
 354                         execl(info->alert_cmd, info->alert_cmd,
 355                               event, dev, disc, NULL);
 356                         exit(2);
 357                 }
 358         }
 359         if (info->mailaddr &&
 360             (strncmp(event, "Fail", 4)==0 ||
 361              strncmp(event, "Test", 4)==0 ||
 362              strncmp(event, "Spares", 6)==0 ||
 363              strncmp(event, "Degrade", 7)==0)) {
 364                 FILE *mp = popen(Sendmail, "w");
 365                 if (mp) {
 366                         FILE *mdstat;
 367                         char hname[256];
 368                         gethostname(hname, sizeof(hname));
 369                         signal(SIGPIPE, SIG_IGN);
 370                         if (info->mailfrom)
 371                                 fprintf(mp, "From: %s\n", info->mailfrom);
 372                         else
 373                                 fprintf(mp, "From: " Name " monitoring <root>\n");
 374                         fprintf(mp, "To: %s\n", info->mailaddr);
 375                         fprintf(mp, "Subject: %s event on %s:%s\n\n",
 376                                 event, dev, hname);
 377
 378                         fprintf(mp,
 379                                 "This is an automatically generated"
 380                                 " mail message from " Name "\n");
 381                         fprintf(mp, "running on %s\n\n", hname);
 382
 383                         fprintf(mp,
 384                                 "A %s event had been detected on"
 385                                 " md device %s.\n\n", event, dev);
 386
 387                         if (disc && disc[0] != ' ')
 388                                 fprintf(mp,
 389                                         "It could be related to"
 390                                         " component device %s.\n\n", disc);
 391                         if (disc && disc[0] == ' ')
 392                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 393
 394                         fprintf(mp, "Faithfully yours, etc.\n");
 395
 396                         mdstat = fopen("/proc/mdstat", "r");
 397                         if (mdstat) {
 398                                 char buf[8192];
 399                                 int n;
 400                                 fprintf(mp,
 401                                         "\nP.S. The /proc/mdstat file"
 402                                         " currently contains the following:\n\n");
 403                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 404                                         n=fwrite(buf, 1, n, mp);
 405                                 fclose(mdstat);
 406                         }
 407                         pclose(mp);
 408                 }
 409         }
 410
 411         /* log the event to syslog maybe */
 412         if (info->dosyslog) {
 413                 /* Log at a different severity depending on the event.
 414                  *
 415                  * These are the critical events:  */
 416                 if (strncmp(event, "Fail", 4)==0 ||
 417                     strncmp(event, "Degrade", 7)==0 ||
 418                     strncmp(event, "DeviceDisappeared", 17)==0)
 419                         priority = LOG_CRIT;
 420                 /* Good to know about, but are not failures: */
 421                 else if (strncmp(event, "Rebuild", 7)==0 ||
 422                          strncmp(event, "MoveSpare", 9)==0 ||
 423                          strncmp(event, "Spares", 6) != 0)
 424                         priority = LOG_WARNING;
 425                 /* Everything else: */
 426                 else
 427                         priority = LOG_INFO;
 428
 429                 if (disc)
 430                         syslog(priority,
 431                                "%s event detected on md device %s,"
 432                                " component device %s", event, dev, disc);
 433                 else
 434                         syslog(priority,
 435                                "%s event detected on md device %s",
 436                                event, dev);
 437         }
 438 }
 439
 440 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 441                        int test, struct alert_info *ainfo,
 442                        int increments, char *prefer)
 443 {
 444         /* Update the state 'st' to reflect any changes shown in mdstat,
 445          * or found by directly examining the array, and return
 446          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 447          */
 448         struct { int state, major, minor; } info[MAX_DISKS];
 449         mdu_array_info_t array;
 450         struct mdstat_ent *mse = NULL, *mse2;
 451         char *dev = st->devname;
 452         int fd;
 453         int i;
 454         int remaining_disks;
 455         int last_disk;
 456         int new_array = 0;
 457
 458         if (test)
 459                 alert("TestMessage", dev, NULL, ainfo);
 460         fd = open(dev, O_RDONLY);
 461         if (fd < 0) {
 462                 if (!st->err)
 463                         alert("DeviceDisappeared", dev, NULL, ainfo);
 464                 st->err=1;
 465                 return 0;
 466         }
 467         fcntl(fd, F_SETFD, FD_CLOEXEC);
 468         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 469                 if (!st->err)
 470                         alert("DeviceDisappeared", dev, NULL, ainfo);
 471                 st->err=1;
 472                 close(fd);
 473                 return 0;
 474         }
 475         /* It's much easier to list what array levels can't
 476          * have a device disappear than all of them that can
 477          */
 478         if (array.level == 0 || array.level == -1) {
 479                 if (!st->err && !st->from_config)
 480                         alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
 481                 st->err = 1;
 482                 close(fd);
 483                 return 0;
 484         }
 485         if (st->devnm[0] == 0)
 486                 strcpy(st->devnm, fd2devnm(fd));
 487
 488         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 489                 if (strcmp(mse2->devnm, st->devnm) == 0) {
 490                         mse2->devnm[0] = 0; /* flag it as "used" */
 491                         mse = mse2;
 492                 }
 493
 494         if (!mse) {
 495                 /* duplicated array in statelist
 496                  * or re-created after reading mdstat*/
 497                 st->err = 1;
 498                 close(fd);
 499                 return 0;
 500         }
 501         /* this array is in /proc/mdstat */
 502         if (array.utime == 0)
 503                 /* external arrays don't update utime, so
 504                  * just make sure it is always different. */
 505                 array.utime = st->utime + 1;;
 506
 507         if (st->err) {
 508                 /* New array appeared where previously had and error */
 509                 st->err = 0;
 510                 st->percent = RESYNC_NONE;
 511                 new_array = 1;
 512                 alert("NewArray", st->devname, NULL, ainfo);
 513         }
 514
 515         if (st->utime == array.utime &&
 516             st->failed == array.failed_disks &&
 517             st->working == array.working_disks &&
 518             st->spare == array.spare_disks &&
 519             (mse == NULL  || (
 520                     mse->percent == st->percent
 521                     ))) {
 522                 close(fd);
 523                 if ((st->active < st->raid) && st->spare == 0)
 524                         return 1;
 525                 else
 526                         return 0;
 527         }
 528         if (st->utime == 0 && /* new array */
 529             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 530                 )
 531                 alert("DegradedArray", dev, NULL, ainfo);
 532
 533         if (st->utime == 0 && /* new array */
 534             st->expected_spares > 0 &&
 535             array.spare_disks < st->expected_spares)
 536                 alert("SparesMissing", dev, NULL, ainfo);
 537         if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
 538             mse->percent >= 0)
 539                 alert("RebuildStarted", dev, NULL, ainfo);
 540         if (st->percent >= 0 &&
 541             mse->percent >= 0 &&
 542             (mse->percent / increments) > (st->percent / increments)) {
 543                 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 544
 545                 if((mse->percent / increments) == 0)
 546                         snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
 547                 else
 548                         snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
 549
 550                 alert(percentalert, dev, NULL, ainfo);
 551         }
 552
 553         if (mse->percent == RESYNC_NONE &&
 554             st->percent >= 0) {
 555                 /* Rebuild/sync/whatever just finished.
 556                  * If there is a number in /mismatch_cnt,
 557                  * we should report that.
 558                  */
 559                 struct mdinfo *sra =
 560                         sysfs_read(-1, st->devnm, GET_MISMATCH);
 561                 if (sra && sra->mismatch_cnt > 0) {
 562                         char cnt[80];
 563                         snprintf(cnt, sizeof(cnt),
 564                                  " mismatches found: %d (on raid level %d)",
 565                                 sra->mismatch_cnt, array.level);
 566                         alert("RebuildFinished", dev, cnt, ainfo);
 567                 } else
 568                         alert("RebuildFinished", dev, NULL, ainfo);
 569                 if (sra)
 570                         free(sra);
 571         }
 572         st->percent = mse->percent;
 573
 574         remaining_disks = array.nr_disks;
 575         for (i=0; i<MAX_DISKS && remaining_disks > 0;
 576              i++) {
 577                 mdu_disk_info_t disc;
 578                 disc.number = i;
 579                 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
 580                         info[i].state = disc.state;
 581                         info[i].major = disc.major;
 582                         info[i].minor = disc.minor;
 583                         if (disc.major || disc.minor)
 584                                 remaining_disks --;
 585                 } else
 586                         info[i].major = info[i].minor = 0;
 587         }
 588         last_disk = i;
 589
 590         if (mse->metadata_version &&
 591             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 592             is_subarray(mse->metadata_version+9)) {
 593                 char *sl;
 594                 strcpy(st->parent_devnm,
 595                        mse->metadata_version+10);
 596                 sl = strchr(st->parent_devnm, '/');
 597                 if (sl)
 598                         *sl = 0;
 599         } else
 600                 st->parent_devnm[0] = 0;
 601         if (st->metadata == NULL &&
 602             st->parent_devnm[0] == 0)
 603                 st->metadata = super_by_fd(fd, NULL);
 604
 605         close(fd);
 606
 607         for (i=0; i<MAX_DISKS; i++) {
 608                 mdu_disk_info_t disc = {0,0,0,0,0};
 609                 int newstate=0;
 610                 int change;
 611                 char *dv = NULL;
 612                 disc.number = i;
 613                 if (i < last_disk &&
 614                     (info[i].major || info[i].minor)) {
 615                         newstate = info[i].state;
 616                         dv = map_dev_preferred(
 617                                 info[i].major, info[i].minor, 1,
 618                                 prefer);
 619                         disc.state = newstate;
 620                         disc.major = info[i].major;
 621                         disc.minor = info[i].minor;
 622                 } else
 623                         newstate = (1 << MD_DISK_REMOVED);
 624
 625                 if (dv == NULL && st->devid[i])
 626                         dv = map_dev_preferred(
 627                                 major(st->devid[i]),
 628                                 minor(st->devid[i]), 1, prefer);
 629                 change = newstate ^ st->devstate[i];
 630                 if (st->utime && change && !st->err && !new_array) {
 631                         if ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))
 632                                 alert("Fail", dev, dv, ainfo);
 633                         else if ((newstate & (1<<MD_DISK_FAULTY)) &&
 634                                  (disc.major || disc.minor) &&
 635                                  st->devid[i] == makedev(disc.major, disc.minor))
 636                                 alert("FailSpare", dev, dv, ainfo);
 637                         else if ((newstate&change)&(1<<MD_DISK_SYNC))
 638                                 alert("SpareActive", dev, dv, ainfo);
 639                 }
 640                 st->devstate[i] = newstate;
 641                 st->devid[i] = makedev(disc.major, disc.minor);
 642         }
 643         st->active = array.active_disks;
 644         st->working = array.working_disks;
 645         st->spare = array.spare_disks;
 646         st->failed = array.failed_disks;
 647         st->utime = array.utime;
 648         st->raid = array.raid_disks;
 649         st->err = 0;
 650         if ((st->active < st->raid) && st->spare == 0)
 651                 return 1;
 652         return 0;
 653 }
 654
 655 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
 656                           int test, struct alert_info *info)
 657 {
 658         struct mdstat_ent *mse;
 659         int new_found = 0;
 660
 661         for (mse=mdstat; mse; mse=mse->next)
 662                 if (mse->devnm[0] &&
 663                     (!mse->level  || /* retrieve containers */
 664                      (strcmp(mse->level, "raid0") != 0 &&
 665                       strcmp(mse->level, "linear") != 0))
 666                         ) {
 667                         struct state *st = xcalloc(1, sizeof *st);
 668                         mdu_array_info_t array;
 669                         int fd;
 670                         st->devname = xstrdup(get_md_name(mse->devnm));
 671                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 672                             ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
 673                                 /* no such array */
 674                                 if (fd >=0) close(fd);
 675                                 put_md_name(st->devname);
 676                                 free(st->devname);
 677                                 if (st->metadata) {
 678                                         st->metadata->ss->free_super(st->metadata);
 679                                         free(st->metadata);
 680                                 }
 681                                 free(st);
 682                                 continue;
 683                         }
 684                         close(fd);
 685                         st->next = *statelist;
 686                         st->err = 1;
 687                         strcpy(st->devnm, mse->devnm);
 688                         st->percent = RESYNC_UNKNOWN;
 689                         st->expected_spares = -1;
 690                         if (mse->metadata_version &&
 691                             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 692                             is_subarray(mse->metadata_version+9)) {
 693                                 char *sl;
 694                                 strcpy(st->parent_devnm,
 695                                         mse->metadata_version+10);
 696                                 sl = strchr(st->parent_devnm, '/');
 697                                 *sl = 0;
 698                         } else
 699                                 st->parent_devnm[0] = 0;
 700                         *statelist = st;
 701                         if (test)
 702                                 alert("TestMessage", st->devname, NULL, info);
 703                         new_found = 1;
 704                 }
 705         return new_found;
 706 }
 707
 708 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
 709 {
 710         int fd;
 711
 712         if (!st->metadata ||
 713             !st->metadata->ss->min_acceptable_spare_size) {
 714                 *sizep = 0;
 715                 return 0;
 716         }
 717
 718         fd = open(st->devname, O_RDONLY);
 719         if (fd < 0)
 720                 return 1;
 721         if (st->metadata->ss->external)
 722                 st->metadata->ss->load_container(st->metadata, fd, st->devname);
 723         else
 724                 st->metadata->ss->load_super(st->metadata, fd, st->devname);
 725         close(fd);
 726         if (!st->metadata->sb)
 727                 return 1;
 728         *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
 729         st->metadata->ss->free_super(st->metadata);
 730
 731         return 0;
 732 }
 733
 734 static int check_donor(struct state *from, struct state *to)
 735 {
 736         struct state *sub;
 737
 738         if (from == to)
 739                 return 0;
 740         if (from->parent)
 741                 /* Cannot move from a member */
 742                 return 0;
 743         if (from->err)
 744                 return 0;
 745         for (sub = from->subarray; sub; sub = sub->subarray)
 746                 /* If source array has degraded subarrays, don't
 747                  * remove anything
 748                  */
 749                 if (sub->active < sub->raid)
 750                         return 0;
 751         if (from->metadata->ss->external == 0)
 752                 if (from->active < from->raid)
 753                         return 0;
 754         if (from->spare <= 0)
 755                 return 0;
 756         return 1;
 757 }
 758
 759 static dev_t choose_spare(struct state *from, struct state *to,
 760                         struct domainlist *domlist, unsigned long long min_size)
 761 {
 762         int d;
 763         dev_t dev = 0;
 764
 765         for (d = from->raid; !dev && d < MAX_DISKS; d++) {
 766                 if (from->devid[d] > 0 &&
 767                     from->devstate[d] == 0) {
 768                         struct dev_policy *pol;
 769                         unsigned long long dev_size;
 770
 771                         if (to->metadata->ss->external &&
 772                             test_partition_from_id(from->devid[d]))
 773                                 continue;
 774
 775                         if (min_size &&
 776                             dev_size_from_id(from->devid[d], &dev_size) &&
 777                             dev_size < min_size)
 778                                 continue;
 779
 780                         pol = devid_policy(from->devid[d]);
 781                         if (from->spare_group)
 782                                 pol_add(&pol, pol_domain,
 783                                         from->spare_group, NULL);
 784                         if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
 785                             dev = from->devid[d];
 786                         dev_policy_free(pol);
 787                 }
 788         }
 789         return dev;
 790 }
 791
 792 static dev_t container_choose_spare(struct state *from, struct state *to,
 793                                     struct domainlist *domlist,
 794                                     unsigned long long min_size, int active)
 795 {
 796         /* This is similar to choose_spare, but we cannot trust devstate,
 797          * so we need to read the metadata instead
 798          */
 799         struct mdinfo *list;
 800         struct supertype *st = from->metadata;
 801         int fd = open(from->devname, O_RDONLY);
 802         int err;
 803         dev_t dev = 0;
 804
 805         if (fd < 0)
 806                 return 0;
 807         if (!st->ss->getinfo_super_disks) {
 808                 close(fd);
 809                 return 0;
 810         }
 811
 812         err = st->ss->load_container(st, fd, NULL);
 813         close(fd);
 814         if (err)
 815                 return 0;
 816
 817         if (from == to) {
 818                 /* We must check if number of active disks has not increased
 819                  * since ioctl in main loop. mdmon may have added spare
 820                  * to subarray. If so we do not need to look for more spares
 821                  * so return non zero value */
 822                 int active_cnt = 0;
 823                 struct mdinfo *dp;
 824                 list = st->ss->getinfo_super_disks(st);
 825                 if (!list) {
 826                         st->ss->free_super(st);
 827                         return 1;
 828                 }
 829                 dp = list->devs;
 830                 while (dp) {
 831                         if (dp->disk.state & (1<<MD_DISK_SYNC) &&
 832                             !(dp->disk.state & (1<<MD_DISK_FAULTY)))
 833                                 active_cnt++;
 834                         dp = dp->next;
 835                 }
 836                 sysfs_free(list);
 837                 if (active < active_cnt) {
 838                         /* Spare just activated.*/
 839                         st->ss->free_super(st);
 840                         return 1;
 841                 }
 842         }
 843
 844         /* We only need one spare so full list not needed */
 845         list = container_choose_spares(st, min_size, domlist, from->spare_group,
 846                                        to->metadata->ss->name, 1);
 847         if (list) {
 848                 struct mdinfo *disks = list->devs;
 849                 if (disks)
 850                         dev = makedev(disks->disk.major, disks->disk.minor);
 851                 sysfs_free(list);
 852         }
 853         st->ss->free_super(st);
 854         return dev;
 855 }
 856
 857 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 858 {
 859         struct state *from;
 860         struct state *st;
 861
 862         link_containers_with_subarrays(statelist);
 863         for (st = statelist; st; st = st->next)
 864                 if (st->active < st->raid &&
 865                     st->spare == 0 && !st->err) {
 866                         struct domainlist *domlist = NULL;
 867                         int d;
 868                         struct state *to = st;
 869                         unsigned long long min_size;
 870
 871                         if (to->parent_devnm[0] && !to->parent)
 872                                 /* subarray monitored without parent container
 873                                  * we can't move spares here */
 874                                 continue;
 875
 876                         if (to->parent)
 877                                 /* member of a container */
 878                                 to = to->parent;
 879
 880                         if (get_min_spare_size_required(to, &min_size))
 881                                 continue;
 882                         if (to->metadata->ss->external) {
 883                                 /* We must make sure there is
 884                                  * no suitable spare in container already.
 885                                  * If there is we don't add more */
 886                                 dev_t devid = container_choose_spare(
 887                                         to, to, NULL, min_size, st->active);
 888                                 if (devid > 0)
 889                                         continue;
 890                         }
 891                         for (d = 0; d < MAX_DISKS; 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                          * No spare migration if the destination
 900                          * has no domain. Skip this array.
 901                          */
 902                         if (!domlist)
 903                                 continue;
 904                         for (from=statelist ; from ; from=from->next) {
 905                                 dev_t devid;
 906                                 if (!check_donor(from, to))
 907                                         continue;
 908                                 if (from->metadata->ss->external)
 909                                         devid = container_choose_spare(
 910                                                 from, to, domlist, min_size, 0);
 911                                 else
 912                                         devid = choose_spare(from, to, domlist,
 913                                                              min_size);
 914                                 if (devid > 0
 915                                     && move_spare(from->devname, to->devname, devid)) {
 916                                         alert("MoveSpare", to->devname, from->devname, info);
 917                                         break;
 918                                 }
 919                         }
 920                         domain_free(domlist);
 921                 }
 922 }
 923
 924 /* search the statelist to connect external
 925  * metadata subarrays with their containers
 926  * We always completely rebuild the tree from scratch as
 927  * that is safest considering the possibility of entries
 928  * disappearing or changing.
 929  */
 930 static void link_containers_with_subarrays(struct state *list)
 931 {
 932         struct state *st;
 933         struct state *cont;
 934         for (st = list; st; st = st->next) {
 935                 st->parent = NULL;
 936                 st->subarray = NULL;
 937         }
 938         for (st = list; st; st = st->next)
 939                 if (st->parent_devnm[0])
 940                         for (cont = list; cont; cont = cont->next)
 941                                 if (!cont->err &&
 942                                     cont->parent_devnm[0] == 0 &&
 943                                     strcmp(cont->devnm, st->parent_devnm) == 0) {
 944                                         st->parent = cont;
 945                                         st->subarray = cont->subarray;
 946                                         cont->subarray = st;
 947                                         break;
 948                                 }
 949 }
 950
 951 /* Not really Monitor but ... */
 952 int Wait(char *dev)
 953 {
 954         struct stat stb;
 955         char devnm[32];
 956         int rv = 1;
 957
 958         if (stat(dev, &stb) != 0) {
 959                 pr_err("Cannot find %s: %s\n", dev,
 960                         strerror(errno));
 961                 return 2;
 962         }
 963         strcpy(devnm, stat2devnm(&stb));
 964
 965         while(1) {
 966                 struct mdstat_ent *ms = mdstat_read(1, 0);
 967                 struct mdstat_ent *e;
 968
 969                 for (e=ms ; e; e=e->next)
 970                         if (strcmp(e->devnm, devnm) == 0)
 971                                 break;
 972
 973                 if (e->percent == RESYNC_NONE) {
 974                         /* We could be in the brief pause before something
 975                          * starts. /proc/mdstat doesn't show that, but
 976                          * sync_action does.
 977                          */
 978                         struct mdinfo mdi;
 979                         char buf[21];
 980                         sysfs_init(&mdi, -1, devnm);
 981                         if (sysfs_get_str(&mdi, NULL, "sync_action",
 982                                           buf, 20) > 0 &&
 983                             strcmp(buf,"idle\n") != 0)
 984                                 e->percent = RESYNC_UNKNOWN;
 985                 }
 986                 if (!e || e->percent == RESYNC_NONE) {
 987                         if (e && e->metadata_version &&
 988                             strncmp(e->metadata_version, "external:", 9) == 0) {
 989                                 if (is_subarray(&e->metadata_version[9]))
 990                                         ping_monitor(&e->metadata_version[9]);
 991                                 else
 992                                         ping_monitor(devnm);
 993                         }
 994                         free_mdstat(ms);
 995                         return rv;
 996                 }
 997                 free_mdstat(ms);
 998                 rv = 0;
 999                 mdstat_wait(5);
1000         }
1001 }
1002
1003 #ifndef MDASSEMBLE
1004
1005 static char *clean_states[] = {
1006         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1007
1008 int WaitClean(char *dev, int sock, int verbose)
1009 {
1010         int fd;
1011         struct mdinfo *mdi;
1012         int rv = 1;
1013         char devnm[32];
1014
1015         fd = open(dev, O_RDONLY);
1016         if (fd < 0) {
1017                 if (verbose)
1018                         pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1019                 return 1;
1020         }
1021
1022         strcpy(devnm, fd2devnm(fd));
1023         mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1024         if (!mdi) {
1025                 if (verbose)
1026                         pr_err("Failed to read sysfs attributes for "
1027                                "%s\n", dev);
1028                 close(fd);
1029                 return 0;
1030         }
1031
1032         switch(mdi->array.level) {
1033         case LEVEL_LINEAR:
1034         case LEVEL_MULTIPATH:
1035         case 0:
1036                 /* safemode delay is irrelevant for these levels */
1037                 rv = 0;
1038         }
1039
1040         /* for internal metadata the kernel handles the final clean
1041          * transition, containers can never be dirty
1042          */
1043         if (!is_subarray(mdi->text_version))
1044                 rv = 0;
1045
1046         /* safemode disabled ? */
1047         if (mdi->safe_mode_delay == 0)
1048                 rv = 0;
1049
1050         if (rv) {
1051                 int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state");
1052                 char buf[20];
1053                 int delay = 5000;
1054
1055                 /* minimize the safe_mode_delay and prepare to wait up to 5s
1056                  * for writes to quiesce
1057                  */
1058                 sysfs_set_safemode(mdi, 1);
1059
1060                 /* wait for array_state to be clean */
1061                 while (1) {
1062                         rv = read(state_fd, buf, sizeof(buf));
1063                         if (rv < 0)
1064                                 break;
1065                         if (sysfs_match_word(buf, clean_states) <= 4)
1066                                 break;
1067                         rv = sysfs_wait(state_fd, &delay);
1068                         if (rv < 0 && errno != EINTR)
1069                                 break;
1070                         lseek(state_fd, 0, SEEK_SET);
1071                 }
1072                 if (rv < 0)
1073                         rv = 1;
1074                 else if (fping_monitor(sock) == 0 ||
1075                          ping_monitor(mdi->text_version) == 0) {
1076                         /* we need to ping to close the window between array
1077                          * state transitioning to clean and the metadata being
1078                          * marked clean
1079                          */
1080                         rv = 0;
1081                 } else
1082                         rv = 1;
1083                 if (rv && verbose)
1084                         pr_err("Error waiting for %s to be clean\n",
1085                                 dev);
1086
1087                 /* restore the original safe_mode_delay */
1088                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1089                 close(state_fd);
1090         }
1091
1092         sysfs_free(mdi);
1093         close(fd);
1094
1095         return rv;
1096 }
1097 #endif /* MDASSEMBLE */