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                 int rv = make_daemon(pidfile);
 157                 if (rv >= 0)
 158                         return rv;
 159         }
 160
 161         if (share)
 162                 if (check_one_sharer(scan))
 163                         return 1;
 164
 165         if (devlist == NULL) {
 166                 struct mddev_ident *mdlist = conf_get_ident(NULL);
 167                 for (; mdlist; mdlist=mdlist->next) {
 168                         struct state *st;
 169                         if (mdlist->devname == NULL)
 170                                 continue;
 171                         if (strcasecmp(mdlist->devname, "<ignore>") == 0)
 172                                 continue;
 173                         st = calloc(1, sizeof *st);
 174                         if (st == NULL)
 175                                 continue;
 176                         if (mdlist->devname[0] == '/')
 177                                 st->devname = strdup(mdlist->devname);
 178                         else {
 179                                 st->devname = malloc(8+strlen(mdlist->devname)+1);
 180                                 strcpy(strcpy(st->devname, "/dev/md/"),
 181                                        mdlist->devname);
 182                         }
 183                         st->next = statelist;
 184                         st->devnum = INT_MAX;
 185                         st->percent = -2;
 186                         st->expected_spares = mdlist->spare_disks;
 187                         if (mdlist->spare_group)
 188                                 st->spare_group = strdup(mdlist->spare_group);
 189                         statelist = st;
 190                 }
 191         } else {
 192                 struct mddev_dev *dv;
 193                 for (dv=devlist ; dv; dv=dv->next) {
 194                         struct mddev_ident *mdlist = conf_get_ident(dv->devname);
 195                         struct state *st = calloc(1, sizeof *st);
 196                         if (st == NULL)
 197                                 continue;
 198                         st->devname = strdup(dv->devname);
 199                         st->next = statelist;
 200                         st->devnum = INT_MAX;
 201                         st->percent = -2;
 202                         st->expected_spares = -1;
 203                         if (mdlist) {
 204                                 st->expected_spares = mdlist->spare_disks;
 205                                 if (mdlist->spare_group)
 206                                         st->spare_group = strdup(mdlist->spare_group);
 207                         }
 208                         statelist = st;
 209                 }
 210         }
 211
 212
 213         while (! finished) {
 214                 int new_found = 0;
 215                 struct state *st;
 216                 int anydegraded = 0;
 217
 218                 if (mdstat)
 219                         free_mdstat(mdstat);
 220                 mdstat = mdstat_read(oneshot?0:1, 0);
 221
 222                 for (st=statelist; st; st=st->next)
 223                         if (check_array(st, mdstat, test, &info, increments))
 224                                 anydegraded = 1;
 225
 226                 /* now check if there are any new devices found in mdstat */
 227                 if (scan)
 228                         new_found = add_new_arrays(mdstat, &statelist, test,
 229                                                    &info);
 230
 231                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 232                  * Look for another array with spare > 0 and active == raid and same spare_group
 233                  *  if found, choose a device and hotremove/hotadd
 234                  */
 235                 if (share && anydegraded)
 236                         try_spare_migration(statelist, &info);
 237                 if (!new_found) {
 238                         if (oneshot)
 239                                 break;
 240                         else
 241                                 mdstat_wait(period);
 242                 }
 243                 test = 0;
 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         struct stat buf;
 292         fp = fopen("/var/run/mdadm/autorebuild.pid", "r");
 293         if (fp) {
 294                 fscanf(fp, "%d", &pid);
 295                 sprintf(dir, "/proc/%d", pid);
 296                 rv = stat(dir, &buf);
 297                 if (rv != -1) {
 298                         if (scan) {
 299                                 fprintf(stderr, Name ": Only one "
 300                                         "autorebuild process allowed"
 301                                         " in scan mode, aborting\n");
 302                                 fclose(fp);
 303                                 return 1;
 304                         } else {
 305                                 fprintf(stderr, Name ": Warning: One"
 306                                         " autorebuild process already"
 307                                         " running.\n");
 308                         }
 309                 }
 310                 fclose(fp);
 311         }
 312         if (scan) {
 313                 if (mkdir("/var/run/mdadm", S_IRWXU) < 0 &&
 314                     errno != EEXIST) {
 315                         fprintf(stderr, Name ": Can't create "
 316                                 "autorebuild.pid file\n");
 317                 } else {
 318                         fp = fopen("/var/run/mdadm/autorebuild.pid", "w");
 319                         if (!fp)
 320                                 fprintf(stderr, Name ": Cannot create"
 321                                         " autorebuild.pid"
 322                                         "file\n");
 323                         else {
 324                                 pid = getpid();
 325                                 fprintf(fp, "%d\n", pid);
 326                                 fclose(fp);
 327                         }
 328                 }
 329         }
 330         return 0;
 331 }
 332
 333 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
 334 {
 335         int priority;
 336
 337         if (!info->alert_cmd && !info->mailaddr) {
 338                 time_t now = time(0);
 339
 340                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 341         }
 342         if (info->alert_cmd) {
 343                 int pid = fork();
 344                 switch(pid) {
 345                 default:
 346                         waitpid(pid, NULL, 0);
 347                         break;
 348                 case -1:
 349                         break;
 350                 case 0:
 351                         execl(info->alert_cmd, info->alert_cmd,
 352                               event, dev, disc, NULL);
 353                         exit(2);
 354                 }
 355         }
 356         if (info->mailaddr &&
 357             (strncmp(event, "Fail", 4)==0 ||
 358              strncmp(event, "Test", 4)==0 ||
 359              strncmp(event, "Spares", 6)==0 ||
 360              strncmp(event, "Degrade", 7)==0)) {
 361                 FILE *mp = popen(Sendmail, "w");
 362                 if (mp) {
 363                         FILE *mdstat;
 364                         char hname[256];
 365                         gethostname(hname, sizeof(hname));
 366                         signal(SIGPIPE, SIG_IGN);
 367                         if (info->mailfrom)
 368                                 fprintf(mp, "From: %s\n", info->mailfrom);
 369                         else
 370                                 fprintf(mp, "From: " Name " monitoring <root>\n");
 371                         fprintf(mp, "To: %s\n", info->mailaddr);
 372                         fprintf(mp, "Subject: %s event on %s:%s\n\n",
 373                                 event, dev, hname);
 374
 375                         fprintf(mp,
 376                                 "This is an automatically generated"
 377                                 " mail message from " Name "\n");
 378                         fprintf(mp, "running on %s\n\n", hname);
 379
 380                         fprintf(mp,
 381                                 "A %s event had been detected on"
 382                                 " md device %s.\n\n", event, dev);
 383
 384                         if (disc && disc[0] != ' ')
 385                                 fprintf(mp,
 386                                         "It could be related to"
 387                                         " component device %s.\n\n", disc);
 388                         if (disc && disc[0] == ' ')
 389                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 390
 391                         fprintf(mp, "Faithfully yours, etc.\n");
 392
 393                         mdstat = fopen("/proc/mdstat", "r");
 394                         if (mdstat) {
 395                                 char buf[8192];
 396                                 int n;
 397                                 fprintf(mp,
 398                                         "\nP.S. The /proc/mdstat file"
 399                                         " currently contains the following:\n\n");
 400                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 401                                         n=fwrite(buf, 1, n, mp);
 402                                 fclose(mdstat);
 403                         }
 404                         pclose(mp);
 405                 }
 406         }
 407
 408         /* log the event to syslog maybe */
 409         if (info->dosyslog) {
 410                 /* Log at a different severity depending on the event.
 411                  *
 412                  * These are the critical events:  */
 413                 if (strncmp(event, "Fail", 4)==0 ||
 414                     strncmp(event, "Degrade", 7)==0 ||
 415                     strncmp(event, "DeviceDisappeared", 17)==0)
 416                         priority = LOG_CRIT;
 417                 /* Good to know about, but are not failures: */
 418                 else if (strncmp(event, "Rebuild", 7)==0 ||
 419                          strncmp(event, "MoveSpare", 9)==0 ||
 420                          strncmp(event, "Spares", 6) != 0)
 421                         priority = LOG_WARNING;
 422                 /* Everything else: */
 423                 else
 424                         priority = LOG_INFO;
 425
 426                 if (disc)
 427                         syslog(priority,
 428                                "%s event detected on md device %s,"
 429                                " component device %s", event, dev, disc);
 430                 else
 431                         syslog(priority,
 432                                "%s event detected on md device %s",
 433                                event, dev);
 434         }
 435 }
 436
 437 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 438                        int test, struct alert_info *ainfo,
 439                        int increments)
 440 {
 441         /* Update the state 'st' to reflect any changes shown in mdstat,
 442          * or found by directly examining the array, and return
 443          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 444          */
 445         struct { int state, major, minor; } info[MaxDisks];
 446         mdu_array_info_t array;
 447         struct mdstat_ent *mse = NULL, *mse2;
 448         char *dev = st->devname;
 449         int fd;
 450         int i;
 451
 452         if (test)
 453                 alert("TestMessage", dev, NULL, ainfo);
 454         fd = open(dev, O_RDONLY);
 455         if (fd < 0) {
 456                 if (!st->err)
 457                         alert("DeviceDisappeared", dev, NULL, ainfo);
 458                 st->err=1;
 459                 return 0;
 460         }
 461         fcntl(fd, F_SETFD, FD_CLOEXEC);
 462         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 463                 if (!st->err)
 464                         alert("DeviceDisappeared", dev, NULL, ainfo);
 465                 st->err=1;
 466                 close(fd);
 467                 return 0;
 468         }
 469         /* It's much easier to list what array levels can't
 470          * have a device disappear than all of them that can
 471          */
 472         if (array.level == 0 || array.level == -1) {
 473                 if (!st->err)
 474                         alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
 475                 st->err = 1;
 476                 close(fd);
 477                 return 0;
 478         }
 479         if (st->devnum == INT_MAX) {
 480                 struct stat stb;
 481                 if (fstat(fd, &stb) == 0 &&
 482                     (S_IFMT&stb.st_mode)==S_IFBLK) {
 483                         if (major(stb.st_rdev) == MD_MAJOR)
 484                                 st->devnum = minor(stb.st_rdev);
 485                         else
 486                                 st->devnum = -1- (minor(stb.st_rdev)>>6);
 487                 }
 488         }
 489
 490         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 491                 if (mse2->devnum == st->devnum) {
 492                         mse2->devnum = INT_MAX; /* flag it as "used" */
 493                         mse = mse2;
 494                 }
 495
 496         if (!mse) {
 497                 /* duplicated array in statelist
 498                  * or re-created after reading mdstat*/
 499                 st->err = 1;
 500                 close(fd);
 501                 return 0;
 502         }
 503         /* this array is in /proc/mdstat */
 504         if (array.utime == 0)
 505                 /* external arrays don't update utime, so
 506                  * just make sure it is always different. */
 507                 array.utime = st->utime + 1;;
 508
 509         if (st->utime == array.utime &&
 510             st->failed == array.failed_disks &&
 511             st->working == array.working_disks &&
 512             st->spare == array.spare_disks &&
 513             (mse == NULL  || (
 514                     mse->percent == st->percent
 515                     ))) {
 516                 close(fd);
 517                 st->err = 0;
 518                 if ((st->active < st->raid) && st->spare == 0)
 519                         return 1;
 520                 else
 521                         return 0;
 522         }
 523         if (st->utime == 0 && /* new array */
 524             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 525                 )
 526                 alert("DegradedArray", dev, NULL, ainfo);
 527
 528         if (st->utime == 0 && /* new array */
 529             st->expected_spares > 0 &&
 530             array.spare_disks < st->expected_spares)
 531                 alert("SparesMissing", dev, NULL, ainfo);
 532         if (st->percent == -1 &&
 533             mse->percent >= 0)
 534                 alert("RebuildStarted", dev, NULL, ainfo);
 535         if (st->percent >= 0 &&
 536             mse->percent >= 0 &&
 537             (mse->percent / increments) > (st->percent / increments)) {
 538                 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 539
 540                 if((mse->percent / increments) == 0)
 541                         snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
 542                 else
 543                         snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
 544
 545                 alert(percentalert, dev, NULL, ainfo);
 546         }
 547
 548         if (mse->percent == -1 &&
 549             st->percent >= 0) {
 550                 /* Rebuild/sync/whatever just finished.
 551                  * If there is a number in /mismatch_cnt,
 552                  * we should report that.
 553                  */
 554                 struct mdinfo *sra =
 555                         sysfs_read(-1, st->devnum, GET_MISMATCH);
 556                 if (sra && sra->mismatch_cnt > 0) {
 557                         char cnt[40];
 558                         sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt);
 559                         alert("RebuildFinished", dev, cnt, ainfo);
 560                 } else
 561                         alert("RebuildFinished", dev, NULL, ainfo);
 562                 if (sra)
 563                         free(sra);
 564         }
 565         st->percent = mse->percent;
 566
 567         for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks;
 568              i++) {
 569                 mdu_disk_info_t disc;
 570                 disc.number = i;
 571                 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
 572                         info[i].state = disc.state;
 573                         info[i].major = disc.major;
 574                         info[i].minor = disc.minor;
 575                 } else
 576                         info[i].major = info[i].minor = 0;
 577         }
 578
 579         if (strncmp(mse->metadata_version, "external:", 9) == 0 &&
 580             is_subarray(mse->metadata_version+9))
 581                 st->parent_dev =
 582                         devname2devnum(mse->metadata_version+10);
 583         else
 584                 st->parent_dev = NoMdDev;
 585         if (st->metadata == NULL &&
 586             st->parent_dev == NoMdDev)
 587                 st->metadata = super_by_fd(fd, NULL);
 588
 589         close(fd);
 590
 591         for (i=0; i<MaxDisks; i++) {
 592                 mdu_disk_info_t disc = {0,0,0,0,0};
 593                 int newstate=0;
 594                 int change;
 595                 char *dv = NULL;
 596                 disc.number = i;
 597                 if (i > array.raid_disks + array.nr_disks) {
 598                         newstate = 0;
 599                         disc.major = disc.minor = 0;
 600                 } else if (info[i].major || info[i].minor) {
 601                         newstate = info[i].state;
 602                         dv = map_dev(info[i].major, info[i].minor, 1);
 603                         disc.state = newstate;
 604                         disc.major = info[i].major;
 605                         disc.minor = info[i].minor;
 606                 } else if (mse &&  mse->pattern && i < (int)strlen(mse->pattern)) {
 607                         switch(mse->pattern[i]) {
 608                         case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
 609                         case '_': newstate = 0; break;
 610                         }
 611                         disc.major = disc.minor = 0;
 612                 }
 613                 if (dv == NULL && st->devid[i])
 614                         dv = map_dev(major(st->devid[i]),
 615                                      minor(st->devid[i]), 1);
 616                 change = newstate ^ st->devstate[i];
 617                 if (st->utime && change && !st->err) {
 618                         if (i < array.raid_disks &&
 619                             (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
 620                              ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
 621                              ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
 622                                 )
 623                                 alert("Fail", dev, dv, ainfo);
 624                         else if (i >= array.raid_disks &&
 625                                  (disc.major || disc.minor) &&
 626                                  st->devid[i] == makedev(disc.major, disc.minor) &&
 627                                  ((newstate&change)&(1<<MD_DISK_FAULTY))
 628                                 )
 629                                 alert("FailSpare", dev, dv, ainfo);
 630                         else if (i < array.raid_disks &&
 631                                  ! (newstate & (1<<MD_DISK_REMOVED)) &&
 632                                  (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
 633                                   ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
 634                                   ((newstate&change)&(1<<MD_DISK_SYNC)))
 635                                 )
 636                                 alert("SpareActive", dev, dv, ainfo);
 637                 }
 638                 st->devstate[i] = newstate;
 639                 st->devid[i] = makedev(disc.major, disc.minor);
 640         }
 641         st->active = array.active_disks;
 642         st->working = array.working_disks;
 643         st->spare = array.spare_disks;
 644         st->failed = array.failed_disks;
 645         st->utime = array.utime;
 646         st->raid = array.raid_disks;
 647         st->err = 0;
 648         if ((st->active < st->raid) && st->spare == 0)
 649                 return 1;
 650         return 0;
 651 }
 652
 653 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
 654                           int test, struct alert_info *info)
 655 {
 656         struct mdstat_ent *mse;
 657         int new_found = 0;
 658
 659         for (mse=mdstat; mse; mse=mse->next)
 660                 if (mse->devnum != INT_MAX &&
 661                     (!mse->level  || /* retrieve containers */
 662                      (strcmp(mse->level, "raid0") != 0 &&
 663                       strcmp(mse->level, "linear") != 0))
 664                         ) {
 665                         struct state *st = calloc(1, sizeof *st);
 666                         mdu_array_info_t array;
 667                         int fd;
 668                         if (st == NULL)
 669                                 continue;
 670                         st->devname = strdup(get_md_name(mse->devnum));
 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                         st->devnum = mse->devnum;
 688                         st->percent = -2;
 689                         st->expected_spares = -1;
 690                         if (strncmp(mse->metadata_version, "external:", 9) == 0 &&
 691                             is_subarray(mse->metadata_version+9))
 692                                 st->parent_dev =
 693                                         devname2devnum(mse->metadata_version+10);
 694                         else
 695                                 st->parent_dev = NoMdDev;
 696                         *statelist = st;
 697                         if (test)
 698                                 alert("TestMessage", st->devname, NULL, info);
 699                         alert("NewArray", st->devname, NULL, info);
 700                         new_found = 1;
 701                 }
 702         return new_found;
 703 }
 704
 705 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
 706 {
 707         int fd;
 708
 709         if (!st->metadata ||
 710             !st->metadata->ss->min_acceptable_spare_size) {
 711                 *sizep = 0;
 712                 return 0;
 713         }
 714
 715         fd = open(st->devname, O_RDONLY);
 716         if (fd < 0)
 717                 return 1;
 718         if (st->metadata->ss->external)
 719                 st->metadata->ss->load_container(st->metadata, fd, st->devname);
 720         else
 721                 st->metadata->ss->load_super(st->metadata, fd, st->devname);
 722         close(fd);
 723         if (!st->metadata->sb)
 724                 return 1;
 725         *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
 726         st->metadata->ss->free_super(st->metadata);
 727
 728         return 0;
 729 }
 730
 731 static int check_donor(struct state *from, struct state *to)
 732 {
 733         struct state *sub;
 734
 735         if (from == to)
 736                 return 0;
 737         if (from->parent)
 738                 /* Cannot move from a member */
 739                 return 0;
 740         if (from->err)
 741                 return 0;
 742         for (sub = from->subarray; sub; sub = sub->subarray)
 743                 /* If source array has degraded subarrays, don't
 744                  * remove anything
 745                  */
 746                 if (sub->active < sub->raid)
 747                         return 0;
 748         if (from->metadata->ss->external == 0)
 749                 if (from->active < from->raid)
 750                         return 0;
 751         if (from->spare <= 0)
 752                 return 0;
 753         return 1;
 754 }
 755
 756 static dev_t choose_spare(struct state *from, struct state *to,
 757                         struct domainlist *domlist, unsigned long long min_size)
 758 {
 759         int d;
 760         dev_t dev = 0;
 761
 762         for (d = from->raid; !dev && d < MaxDisks; d++) {
 763                 if (from->devid[d] > 0 &&
 764                     from->devstate[d] == 0) {
 765                         struct dev_policy *pol;
 766                         unsigned long long dev_size;
 767
 768                         if (to->metadata->ss->external &&
 769                             test_partition_from_id(from->devid[d]))
 770                                 continue;
 771
 772                         if (min_size &&
 773                             dev_size_from_id(from->devid[d], &dev_size) &&
 774                             dev_size < min_size)
 775                                 continue;
 776
 777                         pol = devnum_policy(from->devid[d]);
 778                         if (from->spare_group)
 779                                 pol_add(&pol, pol_domain,
 780                                         from->spare_group, NULL);
 781                         if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
 782                             dev = from->devid[d];
 783                         dev_policy_free(pol);
 784                 }
 785         }
 786         return dev;
 787 }
 788
 789 static dev_t container_choose_spare(struct state *from, struct state *to,
 790                                     struct domainlist *domlist,
 791                                     unsigned long long min_size, int active)
 792 {
 793         /* This is similar to choose_spare, but we cannot trust devstate,
 794          * so we need to read the metadata instead
 795          */
 796         struct mdinfo *list;
 797         struct supertype *st = from->metadata;
 798         int fd = open(from->devname, O_RDONLY);
 799         int err;
 800         dev_t dev = 0;
 801
 802         if (fd < 0)
 803                 return 0;
 804         if (!st->ss->getinfo_super_disks) {
 805                 close(fd);
 806                 return 0;
 807         }
 808
 809         err = st->ss->load_container(st, fd, NULL);
 810         close(fd);
 811         if (err)
 812                 return 0;
 813
 814         if (from == to) {
 815                 /* We must check if number of active disks has not increased
 816                  * since ioctl in main loop. mdmon may have added spare
 817                  * to subarray. If so we do not need to look for more spares
 818                  * so return non zero value */
 819                 int active_cnt = 0;
 820                 struct mdinfo *dp;
 821                 list = st->ss->getinfo_super_disks(st);
 822                 if (!list) {
 823                         st->ss->free_super(st);
 824                         return 1;
 825                 }
 826                 dp = list->devs;
 827                 while (dp) {
 828                         if (dp->disk.state & (1<<MD_DISK_SYNC) &&
 829                             !(dp->disk.state & (1<<MD_DISK_FAULTY)))
 830                                 active_cnt++;
 831                         dp = dp->next;
 832                 }
 833                 sysfs_free(list);
 834                 if (active < active_cnt) {
 835                         /* Spare just activated.*/
 836                         st->ss->free_super(st);
 837                         return 1;
 838                 }
 839         }
 840
 841         /* We only need one spare so full list not needed */
 842         list = container_choose_spares(st, min_size, domlist, from->spare_group,
 843                                        to->metadata->ss->name, 1);
 844         if (list) {
 845                 struct mdinfo *disks = list->devs;
 846                 if (disks)
 847                         dev = makedev(disks->disk.major, disks->disk.minor);
 848                 sysfs_free(list);
 849         }
 850         st->ss->free_super(st);
 851         return dev;
 852 }
 853
 854
 855 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 856 {
 857         struct state *from;
 858         struct state *st;
 859
 860         link_containers_with_subarrays(statelist);
 861         for (st = statelist; st; st = st->next)
 862                 if (st->active < st->raid &&
 863                     st->spare == 0 && !st->err) {
 864                         struct domainlist *domlist = NULL;
 865                         int d;
 866                         struct state *to = st;
 867                         unsigned long long min_size;
 868
 869                         if (to->parent_dev != NoMdDev && !to->parent)
 870                                 /* subarray monitored without parent container
 871                                  * we can't move spares here */
 872                                 continue;
 873
 874                         if (to->parent)
 875                                 /* member of a container */
 876                                 to = to->parent;
 877
 878                         if (get_min_spare_size_required(to, &min_size))
 879                                 continue;
 880                         if (to->metadata->ss->external) {
 881                                 /* We must make sure there is
 882                                  * no suitable spare in container already.
 883                                  * If there is we don't add more */
 884                                 dev_t devid = container_choose_spare(
 885                                         to, to, NULL, min_size, st->active);
 886                                 if (devid > 0)
 887                                         continue;
 888                         }
 889                         for (d = 0; d < MaxDisks; d++)
 890                                 if (to->devid[d])
 891                                         domainlist_add_dev(&domlist,
 892                                                            to->devid[d],
 893                                                            to->metadata->ss->name);
 894                         if (to->spare_group)
 895                                 domain_add(&domlist, to->spare_group);
 896                         /*
 897                          * No spare migration if the destination
 898                          * has no domain. Skip this array.
 899                          */
 900                         if (!domlist)
 901                                 continue;
 902                         for (from=statelist ; from ; from=from->next) {
 903                                 dev_t devid;
 904                                 if (!check_donor(from, to))
 905                                         continue;
 906                                 if (from->metadata->ss->external)
 907                                         devid = container_choose_spare(
 908                                                 from, to, domlist, min_size, 0);
 909                                 else
 910                                         devid = choose_spare(from, to, domlist,
 911                                                              min_size);
 912                                 if (devid > 0
 913                                     && move_spare(from->devname, to->devname, devid)) {
 914                                         alert("MoveSpare", to->devname, from->devname, info);
 915                                         break;
 916                                 }
 917                         }
 918                         domain_free(domlist);
 919                 }
 920 }
 921
 922 /* search the statelist to connect external
 923  * metadata subarrays with their containers
 924  * We always completely rebuild the tree from scratch as
 925  * that is safest considering the possibility of entries
 926  * disappearing or changing.
 927  */
 928 static void link_containers_with_subarrays(struct state *list)
 929 {
 930         struct state *st;
 931         struct state *cont;
 932         for (st = list; st; st = st->next) {
 933                 st->parent = NULL;
 934                 st->subarray = NULL;
 935         }
 936         for (st = list; st; st = st->next)
 937                 if (st->parent_dev != NoMdDev)
 938                         for (cont = list; cont; cont = cont->next)
 939                                 if (!cont->err &&
 940                                     cont->parent_dev == NoMdDev &&
 941                                     cont->devnum == st->parent_dev) {
 942                                         st->parent = cont;
 943                                         st->subarray = cont->subarray;
 944                                         cont->subarray = st;
 945                                         break;
 946                                 }
 947 }
 948
 949 /* Not really Monitor but ... */
 950 int Wait(char *dev)
 951 {
 952         struct stat stb;
 953         int devnum;
 954         int rv = 1;
 955
 956         if (stat(dev, &stb) != 0) {
 957                 fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
 958                         strerror(errno));
 959                 return 2;
 960         }
 961         devnum = stat2devnum(&stb);
 962
 963         while(1) {
 964                 struct mdstat_ent *ms = mdstat_read(1, 0);
 965                 struct mdstat_ent *e;
 966
 967                 for (e=ms ; e; e=e->next)
 968                         if (e->devnum == devnum)
 969                                 break;
 970
 971                 if (!e || e->percent < 0) {
 972                         if (e && e->metadata_version &&
 973                             strncmp(e->metadata_version, "external:", 9) == 0) {
 974                                 if (is_subarray(&e->metadata_version[9]))
 975                                         ping_monitor(&e->metadata_version[9]);
 976                                 else
 977                                         ping_monitor(devnum2devname(devnum));
 978                         }
 979                         free_mdstat(ms);
 980                         return rv;
 981                 }
 982                 free_mdstat(ms);
 983                 rv = 0;
 984                 mdstat_wait(5);
 985         }
 986 }