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         unsigned int utime;
  37         int err;
  38         char *spare_group;
  39         int active, working, failed, spare, raid;
  40         int from_config;
  41         int from_auto;
  42         int expected_spares;
  43         int devstate[MAX_DISKS];
  44         dev_t devid[MAX_DISKS];
  45         int percent;
  46         char parent_devnm[32]; /* For subarray, devnm of parent.
  47                                 * For others, ""
  48                                 */
  49         struct supertype *metadata;
  50         struct state *subarray;/* for a container it is a link to first subarray
  51                                 * for a subarray it is a link to next subarray
  52                                 * in the same container */
  53         struct state *parent;  /* for a subarray it is a link to its container
  54                                 */
  55         struct state *next;
  56 };
  57
  58 struct alert_info {
  59         char *mailaddr;
  60         char *mailfrom;
  61         char *alert_cmd;
  62         int dosyslog;
  63 };
  64 static int make_daemon(char *pidfile);
  65 static int check_one_sharer(int scan);
  66 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
  67 static int check_array(struct state *st, struct mdstat_ent *mdstat,
  68                        int test, struct alert_info *info,
  69                        int increments, char *prefer);
  70 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
  71                           int test, struct alert_info *info);
  72 static void try_spare_migration(struct state *statelist, struct alert_info *info);
  73 static void link_containers_with_subarrays(struct state *list);
  74
  75 int Monitor(struct mddev_dev *devlist,
  76             char *mailaddr, char *alert_cmd,
  77             struct context *c,
  78             int daemonise, int oneshot,
  79             int dosyslog, char *pidfile, int increments,
  80             int share)
  81 {
  82         /*
  83          * Every few seconds, scan every md device looking for changes
  84          * When a change is found, log it, possibly run the alert command,
  85          * and possibly send Email
  86          *
  87          * For each array, we record:
  88          *   Update time
  89          *   active/working/failed/spare drives
  90          *   State of each device.
  91          *   %rebuilt if rebuilding
  92          *
  93          * If the update time changes, check out all the data again
  94          * It is possible that we cannot get the state of each device
  95          * due to bugs in the md kernel module.
  96          * We also read /proc/mdstat to get rebuild percent,
  97          * and to get state on all active devices incase of kernel bug.
  98          *
  99          * Events are:
 100          *    Fail
 101          *      An active device had Faulty set or Active/Sync removed
 102          *    FailSpare
 103          *      A spare device had Faulty set
 104          *    SpareActive
 105          *      An active device had a reverse transition
 106          *    RebuildStarted
 107          *      percent went from -1 to +ve
 108          *    RebuildNN
 109          *      percent went from below to not-below NN%
 110          *    DeviceDisappeared
 111          *      Couldn't access a device which was previously visible
 112          *
 113          * if we detect an array with active<raid and spare==0
 114          * we look at other arrays that have same spare-group
 115          * If we find one with active==raid and spare>0,
 116          *  and if we can get_disk_info and find a name
 117          *  Then we hot-remove and hot-add to the other array
 118          *
 119          * If devlist is NULL, then we can monitor everything because --scan
 120          * was given.  We get an initial list from config file and add anything
 121          * that appears in /proc/mdstat
 122          */
 123
 124         struct state *statelist = NULL;
 125         struct state *st2;
 126         int finished = 0;
 127         struct mdstat_ent *mdstat = NULL;
 128         char *mailfrom = NULL;
 129         struct alert_info info;
 130
 131         if (!mailaddr) {
 132                 mailaddr = conf_get_mailaddr();
 133                 if (mailaddr && ! c->scan)
 134                         pr_err("Monitor using email address \"%s\" from config file\n",
 135                                mailaddr);
 136         }
 137         mailfrom = conf_get_mailfrom();
 138
 139         if (!alert_cmd) {
 140                 alert_cmd = conf_get_program();
 141                 if (alert_cmd && ! c->scan)
 142                         pr_err("Monitor using program \"%s\" from config file\n",
 143                                alert_cmd);
 144         }
 145         if (c->scan && !mailaddr && !alert_cmd && !dosyslog) {
 146                 pr_err("No mail address or alert command - not monitoring.\n");
 147                 return 1;
 148         }
 149         info.alert_cmd = alert_cmd;
 150         info.mailaddr = mailaddr;
 151         info.mailfrom = mailfrom;
 152         info.dosyslog = dosyslog;
 153
 154         if (daemonise) {
 155                 int rv = make_daemon(pidfile);
 156                 if (rv >= 0)
 157                         return rv;
 158         }
 159
 160         if (share)
 161                 if (check_one_sharer(c->scan))
 162                         return 1;
 163
 164         if (devlist == NULL) {
 165                 struct mddev_ident *mdlist = conf_get_ident(NULL);
 166                 for (; mdlist; mdlist=mdlist->next) {
 167                         struct state *st;
 168                         if (mdlist->devname == NULL)
 169                                 continue;
 170                         if (strcasecmp(mdlist->devname, "<ignore>") == 0)
 171                                 continue;
 172                         st = xcalloc(1, sizeof *st);
 173                         if (mdlist->devname[0] == '/')
 174                                 st->devname = xstrdup(mdlist->devname);
 175                         else {
 176                                 st->devname = xmalloc(8+strlen(mdlist->devname)+1);
 177                                 strcpy(strcpy(st->devname, "/dev/md/"),
 178                                        mdlist->devname);
 179                         }
 180                         st->next = statelist;
 181                         st->devnm[0] = 0;
 182                         st->percent = RESYNC_UNKNOWN;
 183                         st->from_config = 1;
 184                         st->expected_spares = mdlist->spare_disks;
 185                         if (mdlist->spare_group)
 186                                 st->spare_group = xstrdup(mdlist->spare_group);
 187                         statelist = st;
 188                 }
 189         } else {
 190                 struct mddev_dev *dv;
 191                 for (dv=devlist ; dv; dv=dv->next) {
 192                         struct mddev_ident *mdlist = conf_get_ident(dv->devname);
 193                         struct state *st = xcalloc(1, sizeof *st);
 194                         st->devname = xstrdup(dv->devname);
 195                         st->next = statelist;
 196                         st->devnm[0] = 0;
 197                         st->percent = RESYNC_UNKNOWN;
 198                         st->expected_spares = -1;
 199                         if (mdlist) {
 200                                 st->expected_spares = mdlist->spare_disks;
 201                                 if (mdlist->spare_group)
 202                                         st->spare_group = xstrdup(mdlist->spare_group);
 203                         }
 204                         statelist = st;
 205                 }
 206         }
 207
 208         while (! finished) {
 209                 int new_found = 0;
 210                 struct state *st, **stp;
 211                 int anydegraded = 0;
 212
 213                 if (mdstat)
 214                         free_mdstat(mdstat);
 215                 mdstat = mdstat_read(oneshot?0:1, 0);
 216                 if (!mdstat)
 217                         mdstat_close();
 218
 219                 for (st=statelist; st; st=st->next)
 220                         if (check_array(st, mdstat, c->test, &info,
 221                                         increments, c->prefer))
 222                                 anydegraded = 1;
 223
 224                 /* now check if there are any new devices found in mdstat */
 225                 if (c->scan)
 226                         new_found = add_new_arrays(mdstat, &statelist, c->test,
 227                                                    &info);
 228
 229                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 230                  * Look for another array with spare > 0 and active == raid and same spare_group
 231                  *  if found, choose a device and hotremove/hotadd
 232                  */
 233                 if (share && anydegraded)
 234                         try_spare_migration(statelist, &info);
 235                 if (!new_found) {
 236                         if (oneshot)
 237                                 break;
 238                         else
 239                                 mdstat_wait(c->delay);
 240                 }
 241                 c->test = 0;
 242
 243                 for (stp = &statelist; (st = *stp) != NULL; ) {
 244                         if (st->from_auto && st->err > 5) {
 245                                 *stp = st->next;
 246                                 free(st->devname);
 247                                 free(st->spare_group);
 248                                 free(st);
 249                         } else
 250                                 stp = &st->next;
 251                 }
 252         }
 253         for (st2 = statelist; st2; st2 = statelist) {
 254                 statelist = st2->next;
 255                 free(st2);
 256         }
 257
 258         if (pidfile)
 259                 unlink(pidfile);
 260         return 0;
 261 }
 262
 263 static int make_daemon(char *pidfile)
 264 {
 265         /* Return:
 266          * -1 in the forked daemon
 267          *  0 in the parent
 268          *  1 on error
 269          * so a none-negative becomes the exit code.
 270          */
 271         int pid = fork();
 272         if (pid > 0) {
 273                 if (!pidfile)
 274                         printf("%d\n", pid);
 275                 else {
 276                         FILE *pid_file;
 277                         pid_file=fopen(pidfile, "w");
 278                         if (!pid_file)
 279                                 perror("cannot create pid file");
 280                         else {
 281                                 fprintf(pid_file,"%d\n", pid);
 282                                 fclose(pid_file);
 283                         }
 284                 }
 285                 return 0;
 286         }
 287         if (pid < 0) {
 288                 perror("daemonise");
 289                 return 1;
 290         }
 291         close(0);
 292         open("/dev/null", O_RDWR);
 293         dup2(0,1);
 294         dup2(0,2);
 295         setsid();
 296         return -1;
 297 }
 298
 299 static int check_one_sharer(int scan)
 300 {
 301         int pid, rv;
 302         FILE *fp;
 303         char dir[20];
 304         char path[100];
 305         struct stat buf;
 306         sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
 307         fp = fopen(path, "r");
 308         if (fp) {
 309                 if (fscanf(fp, "%d", &pid) != 1)
 310                         pid = -1;
 311                 sprintf(dir, "/proc/%d", pid);
 312                 rv = stat(dir, &buf);
 313                 if (rv != -1) {
 314                         if (scan) {
 315                                 pr_err("Only one autorebuild process allowed in scan mode, aborting\n");
 316                                 fclose(fp);
 317                                 return 1;
 318                         } else {
 319                                 pr_err("Warning: One autorebuild process already running.\n");
 320                         }
 321                 }
 322                 fclose(fp);
 323         }
 324         if (scan) {
 325                 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
 326                     errno != EEXIST) {
 327                         pr_err("Can't create autorebuild.pid file\n");
 328                 } else {
 329                         fp = fopen(path, "w");
 330                         if (!fp)
 331                                 pr_err("Cannot create autorebuild.pidfile\n");
 332                         else {
 333                                 pid = getpid();
 334                                 fprintf(fp, "%d\n", pid);
 335                                 fclose(fp);
 336                         }
 337                 }
 338         }
 339         return 0;
 340 }
 341
 342 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
 343 {
 344         int priority;
 345
 346         if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
 347                 time_t now = time(0);
 348
 349                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 350         }
 351         if (info->alert_cmd) {
 352                 int pid = fork();
 353                 switch(pid) {
 354                 default:
 355                         waitpid(pid, NULL, 0);
 356                         break;
 357                 case -1:
 358                         break;
 359                 case 0:
 360                         execl(info->alert_cmd, info->alert_cmd,
 361                               event, dev, disc, NULL);
 362                         exit(2);
 363                 }
 364         }
 365         if (info->mailaddr &&
 366             (strncmp(event, "Fail", 4)==0 ||
 367              strncmp(event, "Test", 4)==0 ||
 368              strncmp(event, "Spares", 6)==0 ||
 369              strncmp(event, "Degrade", 7)==0)) {
 370                 FILE *mp = popen(Sendmail, "w");
 371                 if (mp) {
 372                         FILE *mdstat;
 373                         char hname[256];
 374                         gethostname(hname, sizeof(hname));
 375                         signal(SIGPIPE, SIG_IGN);
 376                         if (info->mailfrom)
 377                                 fprintf(mp, "From: %s\n", info->mailfrom);
 378                         else
 379                                 fprintf(mp, "From: %s monitoring <root>\n", Name);
 380                         fprintf(mp, "To: %s\n", info->mailaddr);
 381                         fprintf(mp, "Subject: %s event on %s:%s\n\n",
 382                                 event, dev, hname);
 383
 384                         fprintf(mp,
 385                                 "This is an automatically generated mail message from %s\n", Name);
 386                         fprintf(mp, "running on %s\n\n", hname);
 387
 388                         fprintf(mp,
 389                                 "A %s event had been detected on md device %s.\n\n", event, dev);
 390
 391                         if (disc && disc[0] != ' ')
 392                                 fprintf(mp,
 393                                         "It could be related to component device %s.\n\n", disc);
 394                         if (disc && disc[0] == ' ')
 395                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 396
 397                         fprintf(mp, "Faithfully yours, etc.\n");
 398
 399                         mdstat = fopen("/proc/mdstat", "r");
 400                         if (mdstat) {
 401                                 char buf[8192];
 402                                 int n;
 403                                 fprintf(mp,
 404                                         "\nP.S. The /proc/mdstat file currently contains the following:\n\n");
 405                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 406                                         n=fwrite(buf, 1, n, mp);
 407                                 fclose(mdstat);
 408                         }
 409                         pclose(mp);
 410                 }
 411         }
 412
 413         /* log the event to syslog maybe */
 414         if (info->dosyslog) {
 415                 /* Log at a different severity depending on the event.
 416                  *
 417                  * These are the critical events:  */
 418                 if (strncmp(event, "Fail", 4)==0 ||
 419                     strncmp(event, "Degrade", 7)==0 ||
 420                     strncmp(event, "DeviceDisappeared", 17)==0)
 421                         priority = LOG_CRIT;
 422                 /* Good to know about, but are not failures: */
 423                 else if (strncmp(event, "Rebuild", 7)==0 ||
 424                          strncmp(event, "MoveSpare", 9)==0 ||
 425                          strncmp(event, "Spares", 6) != 0)
 426                         priority = LOG_WARNING;
 427                 /* Everything else: */
 428                 else
 429                         priority = LOG_INFO;
 430
 431                 if (disc && disc[0] != ' ')
 432                         syslog(priority,
 433                                "%s event detected on md device %s, component device %s", event, dev, disc);
 434                 else if (disc)
 435                         syslog(priority,
 436                                "%s event detected on md device %s: %s",
 437                                event, dev, disc);
 438                 else
 439                         syslog(priority,
 440                                "%s event detected on md device %s",
 441                                event, dev);
 442         }
 443 }
 444
 445 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 446                        int test, struct alert_info *ainfo,
 447                        int increments, char *prefer)
 448 {
 449         /* Update the state 'st' to reflect any changes shown in mdstat,
 450          * or found by directly examining the array, and return
 451          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 452          */
 453         struct { int state, major, minor; } info[MAX_DISKS];
 454         mdu_array_info_t array;
 455         struct mdstat_ent *mse = NULL, *mse2;
 456         char *dev = st->devname;
 457         int fd = -1;
 458         int i;
 459         int remaining_disks;
 460         int last_disk;
 461         int new_array = 0;
 462
 463         if (test)
 464                 alert("TestMessage", dev, NULL, ainfo);
 465         if (st->devnm[0])
 466                 fd = open("/sys/block", O_RDONLY|O_DIRECTORY);
 467         if (fd >= 0) {
 468                 /* Don't open the device unless it is present and
 469                  * active in sysfs.
 470                  */
 471                 char buf[10];
 472                 close(fd);
 473                 fd = sysfs_open(st->devnm, NULL, "array_state");
 474                 if (fd < 0 ||
 475                     read(fd, buf, 10) < 5 ||
 476                     strncmp(buf,"clear",5) == 0 ||
 477                     strncmp(buf,"inact",5) == 0) {
 478                         if (fd >= 0)
 479                                 close(fd);
 480                         fd = sysfs_open(st->devnm, NULL, "level");
 481                         if (fd < 0 || read(fd, buf, 10) != 0) {
 482                                 if (fd >= 0)
 483                                         close(fd);
 484                                 if (!st->err)
 485                                         alert("DeviceDisappeared", dev, NULL, ainfo);
 486                                 st->err++;
 487                                 return 0;
 488                         }
 489                 }
 490                 close(fd);
 491         }
 492         fd = open(dev, O_RDONLY);
 493         if (fd < 0) {
 494                 if (!st->err)
 495                         alert("DeviceDisappeared", dev, NULL, ainfo);
 496                 st->err++;
 497                 return 0;
 498         }
 499         fcntl(fd, F_SETFD, FD_CLOEXEC);
 500         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 501                 if (!st->err)
 502                         alert("DeviceDisappeared", dev, NULL, ainfo);
 503                 st->err++;
 504                 close(fd);
 505                 return 0;
 506         }
 507         /* It's much easier to list what array levels can't
 508          * have a device disappear than all of them that can
 509          */
 510         if (array.level == 0 || array.level == -1) {
 511                 if (!st->err && !st->from_config)
 512                         alert("DeviceDisappeared", dev, " Wrong-Level", ainfo);
 513                 st->err++;
 514                 close(fd);
 515                 return 0;
 516         }
 517         if (st->devnm[0] == 0)
 518                 strcpy(st->devnm, fd2devnm(fd));
 519
 520         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 521                 if (strcmp(mse2->devnm, st->devnm) == 0) {
 522                         mse2->devnm[0] = 0; /* flag it as "used" */
 523                         mse = mse2;
 524                 }
 525
 526         if (!mse) {
 527                 /* duplicated array in statelist
 528                  * or re-created after reading mdstat*/
 529                 st->err++;
 530                 close(fd);
 531                 return 0;
 532         }
 533         /* this array is in /proc/mdstat */
 534         if (array.utime == 0)
 535                 /* external arrays don't update utime, so
 536                  * just make sure it is always different. */
 537                 array.utime = st->utime + 1;;
 538
 539         if (st->err) {
 540                 /* New array appeared where previously had an error */
 541                 st->err = 0;
 542                 st->percent = RESYNC_NONE;
 543                 new_array = 1;
 544                 alert("NewArray", st->devname, NULL, ainfo);
 545         }
 546
 547         if (st->utime == array.utime &&
 548             st->failed == array.failed_disks &&
 549             st->working == array.working_disks &&
 550             st->spare == array.spare_disks &&
 551             (mse == NULL  || (
 552                     mse->percent == st->percent
 553                     ))) {
 554                 close(fd);
 555                 if ((st->active < st->raid) && st->spare == 0)
 556                         return 1;
 557                 else
 558                         return 0;
 559         }
 560         if (st->utime == 0 && /* new array */
 561             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 562                 )
 563                 alert("DegradedArray", dev, NULL, ainfo);
 564
 565         if (st->utime == 0 && /* new array */
 566             st->expected_spares > 0 &&
 567             array.spare_disks < st->expected_spares)
 568                 alert("SparesMissing", dev, NULL, ainfo);
 569         if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
 570             mse->percent >= 0)
 571                 alert("RebuildStarted", dev, NULL, ainfo);
 572         if (st->percent >= 0 &&
 573             mse->percent >= 0 &&
 574             (mse->percent / increments) > (st->percent / increments)) {
 575                 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 576
 577                 if((mse->percent / increments) == 0)
 578                         snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
 579                 else
 580                         snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
 581
 582                 alert(percentalert, dev, NULL, ainfo);
 583         }
 584
 585         if (mse->percent == RESYNC_NONE &&
 586             st->percent >= 0) {
 587                 /* Rebuild/sync/whatever just finished.
 588                  * If there is a number in /mismatch_cnt,
 589                  * we should report that.
 590                  */
 591                 struct mdinfo *sra =
 592                         sysfs_read(-1, st->devnm, GET_MISMATCH);
 593                 if (sra && sra->mismatch_cnt > 0) {
 594                         char cnt[80];
 595                         snprintf(cnt, sizeof(cnt),
 596                                  " mismatches found: %d (on raid level %d)",
 597                                 sra->mismatch_cnt, array.level);
 598                         alert("RebuildFinished", dev, cnt, ainfo);
 599                 } else
 600                         alert("RebuildFinished", dev, NULL, ainfo);
 601                 if (sra)
 602                         sysfs_free(sra);
 603         }
 604         st->percent = mse->percent;
 605
 606         remaining_disks = array.nr_disks;
 607         for (i=0; i<MAX_DISKS && remaining_disks > 0;
 608              i++) {
 609                 mdu_disk_info_t disc;
 610                 disc.number = i;
 611                 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
 612                         info[i].state = disc.state;
 613                         info[i].major = disc.major;
 614                         info[i].minor = disc.minor;
 615                         if (disc.major || disc.minor)
 616                                 remaining_disks --;
 617                 } else
 618                         info[i].major = info[i].minor = 0;
 619         }
 620         last_disk = i;
 621
 622         if (mse->metadata_version &&
 623             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 624             is_subarray(mse->metadata_version+9)) {
 625                 char *sl;
 626                 strcpy(st->parent_devnm,
 627                        mse->metadata_version+10);
 628                 sl = strchr(st->parent_devnm, '/');
 629                 if (sl)
 630                         *sl = 0;
 631         } else
 632                 st->parent_devnm[0] = 0;
 633         if (st->metadata == NULL &&
 634             st->parent_devnm[0] == 0)
 635                 st->metadata = super_by_fd(fd, NULL);
 636
 637         close(fd);
 638
 639         for (i=0; i<MAX_DISKS; i++) {
 640                 mdu_disk_info_t disc = {0,0,0,0,0};
 641                 int newstate=0;
 642                 int change;
 643                 char *dv = NULL;
 644                 disc.number = i;
 645                 if (i < last_disk &&
 646                     (info[i].major || info[i].minor)) {
 647                         newstate = info[i].state;
 648                         dv = map_dev_preferred(
 649                                 info[i].major, info[i].minor, 1,
 650                                 prefer);
 651                         disc.state = newstate;
 652                         disc.major = info[i].major;
 653                         disc.minor = info[i].minor;
 654                 } else
 655                         newstate = (1 << MD_DISK_REMOVED);
 656
 657                 if (dv == NULL && st->devid[i])
 658                         dv = map_dev_preferred(
 659                                 major(st->devid[i]),
 660                                 minor(st->devid[i]), 1, prefer);
 661                 change = newstate ^ st->devstate[i];
 662                 if (st->utime && change && !st->err && !new_array) {
 663                         if ((st->devstate[i]&change)&(1<<MD_DISK_SYNC))
 664                                 alert("Fail", dev, dv, ainfo);
 665                         else if ((newstate & (1<<MD_DISK_FAULTY)) &&
 666                                  (disc.major || disc.minor) &&
 667                                  st->devid[i] == makedev(disc.major, disc.minor))
 668                                 alert("FailSpare", dev, dv, ainfo);
 669                         else if ((newstate&change)&(1<<MD_DISK_SYNC))
 670                                 alert("SpareActive", dev, dv, ainfo);
 671                 }
 672                 st->devstate[i] = newstate;
 673                 st->devid[i] = makedev(disc.major, disc.minor);
 674         }
 675         st->active = array.active_disks;
 676         st->working = array.working_disks;
 677         st->spare = array.spare_disks;
 678         st->failed = array.failed_disks;
 679         st->utime = array.utime;
 680         st->raid = array.raid_disks;
 681         st->err = 0;
 682         if ((st->active < st->raid) && st->spare == 0)
 683                 return 1;
 684         return 0;
 685 }
 686
 687 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
 688                           int test, struct alert_info *info)
 689 {
 690         struct mdstat_ent *mse;
 691         int new_found = 0;
 692         char *name;
 693
 694         for (mse=mdstat; mse; mse=mse->next)
 695                 if (mse->devnm[0] &&
 696                     (!mse->level  || /* retrieve containers */
 697                      (strcmp(mse->level, "raid0") != 0 &&
 698                       strcmp(mse->level, "linear") != 0))
 699                         ) {
 700                         struct state *st = xcalloc(1, sizeof *st);
 701                         mdu_array_info_t array;
 702                         int fd;
 703
 704                         name = get_md_name(mse->devnm);
 705                         if (!name) {
 706                                 free(st);
 707                                 continue;
 708                         }
 709
 710                         st->devname = xstrdup(name);
 711                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 712                             ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
 713                                 /* no such array */
 714                                 if (fd >=0) close(fd);
 715                                 put_md_name(st->devname);
 716                                 free(st->devname);
 717                                 if (st->metadata) {
 718                                         st->metadata->ss->free_super(st->metadata);
 719                                         free(st->metadata);
 720                                 }
 721                                 free(st);
 722                                 continue;
 723                         }
 724                         close(fd);
 725                         st->next = *statelist;
 726                         st->err = 1;
 727                         st->from_auto = 1;
 728                         strcpy(st->devnm, mse->devnm);
 729                         st->percent = RESYNC_UNKNOWN;
 730                         st->expected_spares = -1;
 731                         if (mse->metadata_version &&
 732                             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 733                             is_subarray(mse->metadata_version+9)) {
 734                                 char *sl;
 735                                 strcpy(st->parent_devnm,
 736                                         mse->metadata_version+10);
 737                                 sl = strchr(st->parent_devnm, '/');
 738                                 *sl = 0;
 739                         } else
 740                                 st->parent_devnm[0] = 0;
 741                         *statelist = st;
 742                         if (test)
 743                                 alert("TestMessage", st->devname, NULL, info);
 744                         new_found = 1;
 745                 }
 746         return new_found;
 747 }
 748
 749 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
 750 {
 751         int fd;
 752
 753         if (!st->metadata ||
 754             !st->metadata->ss->min_acceptable_spare_size) {
 755                 *sizep = 0;
 756                 return 0;
 757         }
 758
 759         fd = open(st->devname, O_RDONLY);
 760         if (fd < 0)
 761                 return 1;
 762         if (st->metadata->ss->external)
 763                 st->metadata->ss->load_container(st->metadata, fd, st->devname);
 764         else
 765                 st->metadata->ss->load_super(st->metadata, fd, st->devname);
 766         close(fd);
 767         if (!st->metadata->sb)
 768                 return 1;
 769         *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
 770         st->metadata->ss->free_super(st->metadata);
 771
 772         return 0;
 773 }
 774
 775 static int check_donor(struct state *from, struct state *to)
 776 {
 777         struct state *sub;
 778
 779         if (from == to)
 780                 return 0;
 781         if (from->parent)
 782                 /* Cannot move from a member */
 783                 return 0;
 784         if (from->err)
 785                 return 0;
 786         for (sub = from->subarray; sub; sub = sub->subarray)
 787                 /* If source array has degraded subarrays, don't
 788                  * remove anything
 789                  */
 790                 if (sub->active < sub->raid)
 791                         return 0;
 792         if (from->metadata->ss->external == 0)
 793                 if (from->active < from->raid)
 794                         return 0;
 795         if (from->spare <= 0)
 796                 return 0;
 797         return 1;
 798 }
 799
 800 static dev_t choose_spare(struct state *from, struct state *to,
 801                         struct domainlist *domlist, unsigned long long min_size)
 802 {
 803         int d;
 804         dev_t dev = 0;
 805
 806         for (d = from->raid; !dev && d < MAX_DISKS; d++) {
 807                 if (from->devid[d] > 0 &&
 808                     from->devstate[d] == 0) {
 809                         struct dev_policy *pol;
 810                         unsigned long long dev_size;
 811
 812                         if (to->metadata->ss->external &&
 813                             test_partition_from_id(from->devid[d]))
 814                                 continue;
 815
 816                         if (min_size &&
 817                             dev_size_from_id(from->devid[d], &dev_size) &&
 818                             dev_size < min_size)
 819                                 continue;
 820
 821                         pol = devid_policy(from->devid[d]);
 822                         if (from->spare_group)
 823                                 pol_add(&pol, pol_domain,
 824                                         from->spare_group, NULL);
 825                         if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
 826                             dev = from->devid[d];
 827                         dev_policy_free(pol);
 828                 }
 829         }
 830         return dev;
 831 }
 832
 833 static dev_t container_choose_spare(struct state *from, struct state *to,
 834                                     struct domainlist *domlist,
 835                                     unsigned long long min_size, int active)
 836 {
 837         /* This is similar to choose_spare, but we cannot trust devstate,
 838          * so we need to read the metadata instead
 839          */
 840         struct mdinfo *list;
 841         struct supertype *st = from->metadata;
 842         int fd = open(from->devname, O_RDONLY);
 843         int err;
 844         dev_t dev = 0;
 845
 846         if (fd < 0)
 847                 return 0;
 848         if (!st->ss->getinfo_super_disks) {
 849                 close(fd);
 850                 return 0;
 851         }
 852
 853         err = st->ss->load_container(st, fd, NULL);
 854         close(fd);
 855         if (err)
 856                 return 0;
 857
 858         if (from == to) {
 859                 /* We must check if number of active disks has not increased
 860                  * since ioctl in main loop. mdmon may have added spare
 861                  * to subarray. If so we do not need to look for more spares
 862                  * so return non zero value */
 863                 int active_cnt = 0;
 864                 struct mdinfo *dp;
 865                 list = st->ss->getinfo_super_disks(st);
 866                 if (!list) {
 867                         st->ss->free_super(st);
 868                         return 1;
 869                 }
 870                 dp = list->devs;
 871                 while (dp) {
 872                         if (dp->disk.state & (1<<MD_DISK_SYNC) &&
 873                             !(dp->disk.state & (1<<MD_DISK_FAULTY)))
 874                                 active_cnt++;
 875                         dp = dp->next;
 876                 }
 877                 sysfs_free(list);
 878                 if (active < active_cnt) {
 879                         /* Spare just activated.*/
 880                         st->ss->free_super(st);
 881                         return 1;
 882                 }
 883         }
 884
 885         /* We only need one spare so full list not needed */
 886         list = container_choose_spares(st, min_size, domlist, from->spare_group,
 887                                        to->metadata->ss->name, 1);
 888         if (list) {
 889                 struct mdinfo *disks = list->devs;
 890                 if (disks)
 891                         dev = makedev(disks->disk.major, disks->disk.minor);
 892                 sysfs_free(list);
 893         }
 894         st->ss->free_super(st);
 895         return dev;
 896 }
 897
 898 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 899 {
 900         struct state *from;
 901         struct state *st;
 902
 903         link_containers_with_subarrays(statelist);
 904         for (st = statelist; st; st = st->next)
 905                 if (st->active < st->raid &&
 906                     st->spare == 0 && !st->err) {
 907                         struct domainlist *domlist = NULL;
 908                         int d;
 909                         struct state *to = st;
 910                         unsigned long long min_size;
 911
 912                         if (to->parent_devnm[0] && !to->parent)
 913                                 /* subarray monitored without parent container
 914                                  * we can't move spares here */
 915                                 continue;
 916
 917                         if (to->parent)
 918                                 /* member of a container */
 919                                 to = to->parent;
 920
 921                         if (get_min_spare_size_required(to, &min_size))
 922                                 continue;
 923                         if (to->metadata->ss->external) {
 924                                 /* We must make sure there is
 925                                  * no suitable spare in container already.
 926                                  * If there is we don't add more */
 927                                 dev_t devid = container_choose_spare(
 928                                         to, to, NULL, min_size, st->active);
 929                                 if (devid > 0)
 930                                         continue;
 931                         }
 932                         for (d = 0; d < MAX_DISKS; d++)
 933                                 if (to->devid[d])
 934                                         domainlist_add_dev(&domlist,
 935                                                            to->devid[d],
 936                                                            to->metadata->ss->name);
 937                         if (to->spare_group)
 938                                 domain_add(&domlist, to->spare_group);
 939                         /*
 940                          * No spare migration if the destination
 941                          * has no domain. Skip this array.
 942                          */
 943                         if (!domlist)
 944                                 continue;
 945                         for (from=statelist ; from ; from=from->next) {
 946                                 dev_t devid;
 947                                 if (!check_donor(from, to))
 948                                         continue;
 949                                 if (from->metadata->ss->external)
 950                                         devid = container_choose_spare(
 951                                                 from, to, domlist, min_size, 0);
 952                                 else
 953                                         devid = choose_spare(from, to, domlist,
 954                                                              min_size);
 955                                 if (devid > 0
 956                                     && move_spare(from->devname, to->devname, devid)) {
 957                                         alert("MoveSpare", to->devname, from->devname, info);
 958                                         break;
 959                                 }
 960                         }
 961                         domain_free(domlist);
 962                 }
 963 }
 964
 965 /* search the statelist to connect external
 966  * metadata subarrays with their containers
 967  * We always completely rebuild the tree from scratch as
 968  * that is safest considering the possibility of entries
 969  * disappearing or changing.
 970  */
 971 static void link_containers_with_subarrays(struct state *list)
 972 {
 973         struct state *st;
 974         struct state *cont;
 975         for (st = list; st; st = st->next) {
 976                 st->parent = NULL;
 977                 st->subarray = NULL;
 978         }
 979         for (st = list; st; st = st->next)
 980                 if (st->parent_devnm[0])
 981                         for (cont = list; cont; cont = cont->next)
 982                                 if (!cont->err &&
 983                                     cont->parent_devnm[0] == 0 &&
 984                                     strcmp(cont->devnm, st->parent_devnm) == 0) {
 985                                         st->parent = cont;
 986                                         st->subarray = cont->subarray;
 987                                         cont->subarray = st;
 988                                         break;
 989                                 }
 990 }
 991
 992 /* Not really Monitor but ... */
 993 int Wait(char *dev)
 994 {
 995         struct stat stb;
 996         char devnm[32];
 997         int rv = 1;
 998         int frozen_remaining = 3;
 999
1000         if (stat(dev, &stb) != 0) {
1001                 pr_err("Cannot find %s: %s\n", dev,
1002                         strerror(errno));
1003                 return 2;
1004         }
1005         strcpy(devnm, stat2devnm(&stb));
1006
1007         while(1) {
1008                 struct mdstat_ent *ms = mdstat_read(1, 0);
1009                 struct mdstat_ent *e;
1010
1011                 for (e=ms ; e; e=e->next)
1012                         if (strcmp(e->devnm, devnm) == 0)
1013                                 break;
1014
1015                 if (e && e->percent == RESYNC_NONE) {
1016                         /* We could be in the brief pause before something
1017                          * starts. /proc/mdstat doesn't show that, but
1018                          * sync_action does.
1019                          */
1020                         struct mdinfo mdi;
1021                         char buf[21];
1022                         sysfs_init(&mdi, -1, devnm);
1023                         if (sysfs_get_str(&mdi, NULL, "sync_action",
1024                                           buf, 20) > 0 &&
1025                             strcmp(buf,"idle\n") != 0) {
1026                                 e->percent = RESYNC_UNKNOWN;
1027                                 if (strcmp(buf, "frozen\n") == 0) {
1028                                         if (frozen_remaining == 0)
1029                                                 e->percent = RESYNC_NONE;
1030                                         else
1031                                                 frozen_remaining -= 1;
1032                                 }
1033                         }
1034                 }
1035                 if (!e || e->percent == RESYNC_NONE) {
1036                         if (e && e->metadata_version &&
1037                             strncmp(e->metadata_version, "external:", 9) == 0) {
1038                                 if (is_subarray(&e->metadata_version[9]))
1039                                         ping_monitor(&e->metadata_version[9]);
1040                                 else
1041                                         ping_monitor(devnm);
1042                         }
1043                         free_mdstat(ms);
1044                         return rv;
1045                 }
1046                 free_mdstat(ms);
1047                 rv = 0;
1048                 mdstat_wait(5);
1049         }
1050 }
1051
1052 #ifndef MDASSEMBLE
1053
1054 static char *clean_states[] = {
1055         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1056
1057 int WaitClean(char *dev, int sock, int verbose)
1058 {
1059         int fd;
1060         struct mdinfo *mdi;
1061         int rv = 1;
1062         char devnm[32];
1063
1064         fd = open(dev, O_RDONLY);
1065         if (fd < 0) {
1066                 if (verbose)
1067                         pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1068                 return 1;
1069         }
1070
1071         strcpy(devnm, fd2devnm(fd));
1072         mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1073         if (!mdi) {
1074                 if (verbose)
1075                         pr_err("Failed to read sysfs attributes for %s\n", dev);
1076                 close(fd);
1077                 return 0;
1078         }
1079
1080         switch(mdi->array.level) {
1081         case LEVEL_LINEAR:
1082         case LEVEL_MULTIPATH:
1083         case 0:
1084                 /* safemode delay is irrelevant for these levels */
1085                 rv = 0;
1086         }
1087
1088         /* for internal metadata the kernel handles the final clean
1089          * transition, containers can never be dirty
1090          */
1091         if (!is_subarray(mdi->text_version))
1092                 rv = 0;
1093
1094         /* safemode disabled ? */
1095         if (mdi->safe_mode_delay == 0)
1096                 rv = 0;
1097
1098         if (rv) {
1099                 int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state");
1100                 char buf[20];
1101                 int delay = 5000;
1102
1103                 /* minimize the safe_mode_delay and prepare to wait up to 5s
1104                  * for writes to quiesce
1105                  */
1106                 sysfs_set_safemode(mdi, 1);
1107
1108                 /* wait for array_state to be clean */
1109                 while (1) {
1110                         rv = read(state_fd, buf, sizeof(buf));
1111                         if (rv < 0)
1112                                 break;
1113                         if (sysfs_match_word(buf, clean_states) <= 4)
1114                                 break;
1115                         rv = sysfs_wait(state_fd, &delay);
1116                         if (rv < 0 && errno != EINTR)
1117                                 break;
1118                         lseek(state_fd, 0, SEEK_SET);
1119                 }
1120                 if (rv < 0)
1121                         rv = 1;
1122                 else if (fping_monitor(sock) == 0 ||
1123                          ping_monitor(mdi->text_version) == 0) {
1124                         /* we need to ping to close the window between array
1125                          * state transitioning to clean and the metadata being
1126                          * marked clean
1127                          */
1128                         rv = 0;
1129                 } else
1130                         rv = 1;
1131                 if (rv && verbose)
1132                         pr_err("Error waiting for %s to be clean\n",
1133                                 dev);
1134
1135                 /* restore the original safe_mode_delay */
1136                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1137                 close(state_fd);
1138         }
1139
1140         sysfs_free(mdi);
1141         close(fd);
1142
1143         return rv;
1144 }
1145 #endif /* MDASSEMBLE */