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;
 129         struct alert_info info;
 130         struct mddev_ident *mdlist;
 131         int delay_for_event = c->delay;
 132
 133         if (!mailaddr) {
 134                 mailaddr = conf_get_mailaddr();
 135                 if (mailaddr && ! c->scan)
 136                         pr_err("Monitor using email address \"%s\" from config file\n",
 137                                mailaddr);
 138         }
 139         mailfrom = conf_get_mailfrom();
 140
 141         if (!alert_cmd) {
 142                 alert_cmd = conf_get_program();
 143                 if (alert_cmd && !c->scan)
 144                         pr_err("Monitor using program \"%s\" from config file\n",
 145                                alert_cmd);
 146         }
 147         if (c->scan && !mailaddr && !alert_cmd && !dosyslog) {
 148                 pr_err("No mail address or alert command - not monitoring.\n");
 149                 return 1;
 150         }
 151         info.alert_cmd = alert_cmd;
 152         info.mailaddr = mailaddr;
 153         info.mailfrom = mailfrom;
 154         info.dosyslog = dosyslog;
 155
 156         if (daemonise) {
 157                 int rv = make_daemon(pidfile);
 158                 if (rv >= 0)
 159                         return rv;
 160         }
 161
 162         if (share)
 163                 if (check_one_sharer(c->scan))
 164                         return 1;
 165
 166         if (devlist == NULL) {
 167                 mdlist = conf_get_ident(NULL);
 168                 for (; mdlist; mdlist = mdlist->next) {
 169                         struct state *st;
 170
 171                         if (mdlist->devname == NULL)
 172                                 continue;
 173                         if (strcasecmp(mdlist->devname, "<ignore>") == 0)
 174                                 continue;
 175                         st = xcalloc(1, sizeof *st);
 176                         if (mdlist->devname[0] == '/')
 177                                 st->devname = xstrdup(mdlist->devname);
 178                         else {
 179                                 st->devname = xmalloc(8+strlen(mdlist->devname)+1);
 180                                 strcpy(strcpy(st->devname, "/dev/md/"),
 181                                        mdlist->devname);
 182                         }
 183                         st->next = statelist;
 184                         st->devnm[0] = 0;
 185                         st->percent = RESYNC_UNKNOWN;
 186                         st->from_config = 1;
 187                         st->expected_spares = mdlist->spare_disks;
 188                         if (mdlist->spare_group)
 189                                 st->spare_group = xstrdup(mdlist->spare_group);
 190                         statelist = st;
 191                 }
 192         } else {
 193                 struct mddev_dev *dv;
 194
 195                 for (dv = devlist; dv; dv = dv->next) {
 196                         struct state *st = xcalloc(1, sizeof *st);
 197                         mdlist = conf_get_ident(dv->devname);
 198                         st->devname = xstrdup(dv->devname);
 199                         st->next = statelist;
 200                         st->devnm[0] = 0;
 201                         st->percent = RESYNC_UNKNOWN;
 202                         st->expected_spares = -1;
 203                         if (mdlist) {
 204                                 st->expected_spares = mdlist->spare_disks;
 205                                 if (mdlist->spare_group)
 206                                         st->spare_group = xstrdup(mdlist->spare_group);
 207                         }
 208                         statelist = st;
 209                 }
 210         }
 211
 212         while (!finished) {
 213                 int new_found = 0;
 214                 struct state *st, **stp;
 215                 int anydegraded = 0;
 216                 int anyredundant = 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, c->test, &info,
 224                                         increments, c->prefer))
 225                                 anydegraded = 1;
 226                         /* for external arrays, metadata is filled for
 227                          * containers only
 228                          */
 229                         if (st->metadata && st->metadata->ss->external)
 230                                 continue;
 231                         if (st->err == 0 && !anyredundant)
 232                                 anyredundant = 1;
 233                 }
 234
 235                 /* now check if there are any new devices found in mdstat */
 236                 if (c->scan)
 237                         new_found = add_new_arrays(mdstat, &statelist, c->test,
 238                                                    &info);
 239
 240                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 241                  * Look for another array with spare > 0 and active == raid and same spare_group
 242                  *  if found, choose a device and hotremove/hotadd
 243                  */
 244                 if (share && anydegraded)
 245                         try_spare_migration(statelist, &info);
 246                 if (!new_found) {
 247                         if (oneshot)
 248                                 break;
 249                         else if (!anyredundant) {
 250                                 break;
 251                         }
 252                         else {
 253                                 int wait_result = mdstat_wait(delay_for_event);
 254
 255                                 /*
 256                                  * If mdmonitor is awaken by event, set small delay once
 257                                  * to deal with udev and mdadm.
 258                                  */
 259                                 if (wait_result != 0) {
 260                                         if (c->delay > 5)
 261                                                 delay_for_event = 5;
 262                                 } else
 263                                         delay_for_event = c->delay;
 264
 265                                 mdstat_close();
 266                         }
 267                 }
 268                 c->test = 0;
 269
 270                 for (stp = &statelist; (st = *stp) != NULL; ) {
 271                         if (st->from_auto && st->err > 5) {
 272                                 *stp = st->next;
 273                                 free(st->devname);
 274                                 free(st->spare_group);
 275                                 free(st);
 276                         } else
 277                                 stp = &st->next;
 278                 }
 279         }
 280         for (st2 = statelist; st2; st2 = statelist) {
 281                 statelist = st2->next;
 282                 free(st2);
 283         }
 284
 285         if (pidfile)
 286                 unlink(pidfile);
 287         return 0;
 288 }
 289
 290 static int make_daemon(char *pidfile)
 291 {
 292         /* Return:
 293          * -1 in the forked daemon
 294          *  0 in the parent
 295          *  1 on error
 296          * so a none-negative becomes the exit code.
 297          */
 298         int pid = fork();
 299         if (pid > 0) {
 300                 if (!pidfile)
 301                         printf("%d\n", pid);
 302                 else {
 303                         FILE *pid_file;
 304                         pid_file=fopen(pidfile, "w");
 305                         if (!pid_file)
 306                                 perror("cannot create pid file");
 307                         else {
 308                                 fprintf(pid_file,"%d\n", pid);
 309                                 fclose(pid_file);
 310                         }
 311                 }
 312                 return 0;
 313         }
 314         if (pid < 0) {
 315                 perror("daemonise");
 316                 return 1;
 317         }
 318         close(0);
 319         open("/dev/null", O_RDWR);
 320         dup2(0, 1);
 321         dup2(0, 2);
 322         setsid();
 323         return -1;
 324 }
 325
 326 static int check_one_sharer(int scan)
 327 {
 328         int pid;
 329         FILE *comm_fp;
 330         FILE *fp;
 331         char comm_path[100];
 332         char path[100];
 333         char comm[20];
 334
 335         sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
 336         fp = fopen(path, "r");
 337         if (fp) {
 338                 if (fscanf(fp, "%d", &pid) != 1)
 339                         pid = -1;
 340                 snprintf(comm_path, sizeof(comm_path),
 341                          "/proc/%d/comm", pid);
 342                 comm_fp = fopen(comm_path, "r");
 343                 if (comm_fp) {
 344                         if (fscanf(comm_fp, "%s", comm) &&
 345                             strncmp(basename(comm), Name, strlen(Name)) == 0) {
 346                                 if (scan) {
 347                                         pr_err("Only one autorebuild process allowed in scan mode, aborting\n");
 348                                         fclose(comm_fp);
 349                                         fclose(fp);
 350                                         return 1;
 351                                 } else {
 352                                         pr_err("Warning: One autorebuild process already running.\n");
 353                                 }
 354                         }
 355                         fclose(comm_fp);
 356                 }
 357                 fclose(fp);
 358         }
 359         if (scan) {
 360                 if (mkdir(MDMON_DIR, S_IRWXU) < 0 && errno != EEXIST) {
 361                         pr_err("Can't create autorebuild.pid file\n");
 362                 } else {
 363                         fp = fopen(path, "w");
 364                         if (!fp)
 365                                 pr_err("Cannot create autorebuild.pidfile\n");
 366                         else {
 367                                 pid = getpid();
 368                                 fprintf(fp, "%d\n", pid);
 369                                 fclose(fp);
 370                         }
 371                 }
 372         }
 373         return 0;
 374 }
 375
 376 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
 377 {
 378         int priority;
 379
 380         if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
 381                 time_t now = time(0);
 382
 383                 printf("%1.15s: %s on %s %s\n", ctime(&now) + 4,
 384                        event, dev, disc?disc:"unknown device");
 385         }
 386         if (info->alert_cmd) {
 387                 int pid = fork();
 388                 switch(pid) {
 389                 default:
 390                         waitpid(pid, NULL, 0);
 391                         break;
 392                 case -1:
 393                         break;
 394                 case 0:
 395                         execl(info->alert_cmd, info->alert_cmd,
 396                               event, dev, disc, NULL);
 397                         exit(2);
 398                 }
 399         }
 400         if (info->mailaddr && (strncmp(event, "Fail", 4) == 0 ||
 401                                strncmp(event, "Test", 4) == 0 ||
 402                                strncmp(event, "Spares", 6) == 0 ||
 403                                strncmp(event, "Degrade", 7) == 0)) {
 404                 FILE *mp = popen(Sendmail, "w");
 405                 if (mp) {
 406                         FILE *mdstat;
 407                         char hname[256];
 408                         gethostname(hname, sizeof(hname));
 409                         signal(SIGPIPE, SIG_IGN);
 410                         if (info->mailfrom)
 411                                 fprintf(mp, "From: %s\n", info->mailfrom);
 412                         else
 413                                 fprintf(mp, "From: %s monitoring <root>\n",
 414                                         Name);
 415                         fprintf(mp, "To: %s\n", info->mailaddr);
 416                         fprintf(mp, "Subject: %s event on %s:%s\n\n",
 417                                 event, dev, hname);
 418
 419                         fprintf(mp,
 420                                 "This is an automatically generated mail message from %s\n", Name);
 421                         fprintf(mp, "running on %s\n\n", hname);
 422
 423                         fprintf(mp,
 424                                 "A %s event had been detected on md device %s.\n\n", event, dev);
 425
 426                         if (disc && disc[0] != ' ')
 427                                 fprintf(mp,
 428                                         "It could be related to component device %s.\n\n", disc);
 429                         if (disc && disc[0] == ' ')
 430                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 431
 432                         fprintf(mp, "Faithfully yours, etc.\n");
 433
 434                         mdstat = fopen("/proc/mdstat", "r");
 435                         if (mdstat) {
 436                                 char buf[8192];
 437                                 int n;
 438                                 fprintf(mp,
 439                                         "\nP.S. The /proc/mdstat file currently contains the following:\n\n");
 440                                 while ((n = fread(buf, 1, sizeof(buf),
 441                                                   mdstat)) > 0)
 442                                         n = fwrite(buf, 1, n, mp);
 443                                 fclose(mdstat);
 444                         }
 445                         pclose(mp);
 446                 }
 447         }
 448
 449         /* log the event to syslog maybe */
 450         if (info->dosyslog) {
 451                 /* Log at a different severity depending on the event.
 452                  *
 453                  * These are the critical events:  */
 454                 if (strncmp(event, "Fail", 4) == 0 ||
 455                     strncmp(event, "Degrade", 7) == 0 ||
 456                     strncmp(event, "DeviceDisappeared", 17) == 0)
 457                         priority = LOG_CRIT;
 458                 /* Good to know about, but are not failures: */
 459                 else if (strncmp(event, "Rebuild", 7) == 0 ||
 460                          strncmp(event, "MoveSpare", 9) == 0 ||
 461                          strncmp(event, "Spares", 6) != 0)
 462                         priority = LOG_WARNING;
 463                 /* Everything else: */
 464                 else
 465                         priority = LOG_INFO;
 466
 467                 if (disc && disc[0] != ' ')
 468                         syslog(priority,
 469                                "%s event detected on md device %s, component device %s", event, dev, disc);
 470                 else if (disc)
 471                         syslog(priority,
 472                                "%s event detected on md device %s: %s",
 473                                event, dev, disc);
 474                 else
 475                         syslog(priority,
 476                                "%s event detected on md device %s",
 477                                event, dev);
 478         }
 479 }
 480
 481 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 482                        int test, struct alert_info *ainfo,
 483                        int increments, char *prefer)
 484 {
 485         /* Update the state 'st' to reflect any changes shown in mdstat,
 486          * or found by directly examining the array, and return
 487          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 488          */
 489         struct { int state, major, minor; } info[MAX_DISKS];
 490         struct mdinfo *sra = NULL;
 491         mdu_array_info_t array;
 492         struct mdstat_ent *mse = NULL, *mse2;
 493         char *dev = st->devname;
 494         int fd;
 495         int i;
 496         int remaining_disks;
 497         int last_disk;
 498         int new_array = 0;
 499         int retval;
 500         int is_container = 0;
 501         unsigned long redundancy_only_flags = 0;
 502
 503         if (test)
 504                 alert("TestMessage", dev, NULL, ainfo);
 505
 506         retval = 0;
 507
 508         fd = open(dev, O_RDONLY);
 509         if (fd < 0)
 510                 goto disappeared;
 511
 512         if (st->devnm[0] == 0)
 513                 strcpy(st->devnm, fd2devnm(fd));
 514
 515         for (mse2 = mdstat; mse2; mse2 = mse2->next)
 516                 if (strcmp(mse2->devnm, st->devnm) == 0) {
 517                         mse2->devnm[0] = 0; /* flag it as "used" */
 518                         mse = mse2;
 519                 }
 520
 521         if (!mse) {
 522                 /* duplicated array in statelist
 523                  * or re-created after reading mdstat
 524                  */
 525                 st->err++;
 526                 goto out;
 527         }
 528
 529         if (mse->level == NULL)
 530                 is_container = 1;
 531
 532         if (!is_container && !md_array_active(fd))
 533                 goto disappeared;
 534
 535         fcntl(fd, F_SETFD, FD_CLOEXEC);
 536         if (md_get_array_info(fd, &array) < 0)
 537                 goto disappeared;
 538
 539         if (!is_container && map_name(pers, mse->level) > 0)
 540                 redundancy_only_flags |= GET_MISMATCH;
 541
 542         sra = sysfs_read(-1, st->devnm, GET_LEVEL | GET_DISKS | GET_DEVS |
 543                         GET_STATE | redundancy_only_flags);
 544
 545         if (!sra)
 546                 goto disappeared;
 547
 548         /* It's much easier to list what array levels can't
 549          * have a device disappear than all of them that can
 550          */
 551         if (sra->array.level == 0 || sra->array.level == -1) {
 552                 if (!st->err && !st->from_config)
 553                         alert("DeviceDisappeared", dev, " Wrong-Level", ainfo);
 554                 st->err++;
 555                 goto out;
 556         }
 557
 558         /* this array is in /proc/mdstat */
 559         if (array.utime == 0)
 560                 /* external arrays don't update utime, so
 561                  * just make sure it is always different. */
 562                 array.utime = st->utime + 1;;
 563
 564         if (st->err) {
 565                 /* New array appeared where previously had an error */
 566                 st->err = 0;
 567                 st->percent = RESYNC_NONE;
 568                 new_array = 1;
 569                 if (!is_container)
 570                         alert("NewArray", st->devname, NULL, ainfo);
 571         }
 572
 573         if (st->utime == array.utime && st->failed == sra->array.failed_disks &&
 574             st->working == sra->array.working_disks &&
 575             st->spare == sra->array.spare_disks &&
 576             (mse == NULL || (mse->percent == st->percent))) {
 577                 if ((st->active < st->raid) && st->spare == 0)
 578                         retval = 1;
 579                 goto out;
 580         }
 581         if (st->utime == 0 && /* new array */
 582             mse->pattern && strchr(mse->pattern, '_') /* degraded */)
 583                 alert("DegradedArray", dev, NULL, ainfo);
 584
 585         if (st->utime == 0 && /* new array */ st->expected_spares > 0 &&
 586             sra->array.spare_disks < st->expected_spares)
 587                 alert("SparesMissing", dev, NULL, ainfo);
 588         if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
 589             mse->percent >= 0)
 590                 alert("RebuildStarted", dev, NULL, ainfo);
 591         if (st->percent >= 0 && mse->percent >= 0 &&
 592             (mse->percent / increments) > (st->percent / increments)) {
 593                 char percentalert[18];
 594                 /*
 595                  * "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 596                  */
 597
 598                 if((mse->percent / increments) == 0)
 599                         snprintf(percentalert, sizeof(percentalert),
 600                                  "RebuildStarted");
 601                 else
 602                         snprintf(percentalert, sizeof(percentalert),
 603                                  "Rebuild%02d", mse->percent);
 604
 605                 alert(percentalert, dev, NULL, ainfo);
 606         }
 607
 608         if (mse->percent == RESYNC_NONE && st->percent >= 0) {
 609                 /* Rebuild/sync/whatever just finished.
 610                  * If there is a number in /mismatch_cnt,
 611                  * we should report that.
 612                  */
 613                 if (sra && sra->mismatch_cnt > 0) {
 614                         char cnt[80];
 615                         snprintf(cnt, sizeof(cnt),
 616                                  " mismatches found: %d (on raid level %d)",
 617                                  sra->mismatch_cnt, sra->array.level);
 618                         alert("RebuildFinished", dev, cnt, ainfo);
 619                 } else
 620                         alert("RebuildFinished", dev, NULL, ainfo);
 621         }
 622         st->percent = mse->percent;
 623
 624         remaining_disks = sra->array.nr_disks;
 625         for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) {
 626                 mdu_disk_info_t disc;
 627                 disc.number = i;
 628                 if (md_get_disk_info(fd, &disc) >= 0) {
 629                         info[i].state = disc.state;
 630                         info[i].major = disc.major;
 631                         info[i].minor = disc.minor;
 632                         if (disc.major || disc.minor)
 633                                 remaining_disks --;
 634                 } else
 635                         info[i].major = info[i].minor = 0;
 636         }
 637         last_disk = i;
 638
 639         if (mse->metadata_version &&
 640             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 641             is_subarray(mse->metadata_version+9)) {
 642                 char *sl;
 643                 strcpy(st->parent_devnm, mse->metadata_version + 10);
 644                 sl = strchr(st->parent_devnm, '/');
 645                 if (sl)
 646                         *sl = 0;
 647         } else
 648                 st->parent_devnm[0] = 0;
 649         if (st->metadata == NULL && st->parent_devnm[0] == 0)
 650                 st->metadata = super_by_fd(fd, NULL);
 651
 652         for (i = 0; i < MAX_DISKS; i++) {
 653                 mdu_disk_info_t disc = {0, 0, 0, 0, 0};
 654                 int newstate = 0;
 655                 int change;
 656                 char *dv = NULL;
 657                 disc.number = i;
 658                 if (i < last_disk && (info[i].major || info[i].minor)) {
 659                         newstate = info[i].state;
 660                         dv = map_dev_preferred(info[i].major, info[i].minor, 1,
 661                                                prefer);
 662                         disc.state = newstate;
 663                         disc.major = info[i].major;
 664                         disc.minor = info[i].minor;
 665                 } else
 666                         newstate = (1 << MD_DISK_REMOVED);
 667
 668                 if (dv == NULL && st->devid[i])
 669                         dv = map_dev_preferred(major(st->devid[i]),
 670                                                minor(st->devid[i]), 1, prefer);
 671                 change = newstate ^ st->devstate[i];
 672                 if (st->utime && change && !st->err && !new_array) {
 673                         if ((st->devstate[i]&change) & (1 << MD_DISK_SYNC))
 674                                 alert("Fail", dev, dv, ainfo);
 675                         else if ((newstate & (1 << MD_DISK_FAULTY)) &&
 676                                  (disc.major || disc.minor) &&
 677                                  st->devid[i] == makedev(disc.major,
 678                                                          disc.minor))
 679                                 alert("FailSpare", dev, dv, ainfo);
 680                         else if ((newstate&change) & (1 << MD_DISK_SYNC))
 681                                 alert("SpareActive", dev, dv, ainfo);
 682                 }
 683                 st->devstate[i] = newstate;
 684                 st->devid[i] = makedev(disc.major, disc.minor);
 685         }
 686         st->active = sra->array.active_disks;
 687         st->working = sra->array.working_disks;
 688         st->spare = sra->array.spare_disks;
 689         st->failed = sra->array.failed_disks;
 690         st->utime = array.utime;
 691         st->raid = sra->array.raid_disks;
 692         st->err = 0;
 693         if ((st->active < st->raid) && st->spare == 0)
 694                 retval = 1;
 695
 696  out:
 697         if (sra)
 698                 sysfs_free(sra);
 699         if (fd >= 0)
 700                 close(fd);
 701         return retval;
 702
 703  disappeared:
 704         if (!st->err && !is_container)
 705                 alert("DeviceDisappeared", dev, NULL, ainfo);
 706         st->err++;
 707         goto out;
 708 }
 709
 710 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
 711                           int test, struct alert_info *info)
 712 {
 713         struct mdstat_ent *mse;
 714         int new_found = 0;
 715         char *name;
 716
 717         for (mse = mdstat; mse; mse = mse->next)
 718                 if (mse->devnm[0] && (!mse->level || /* retrieve containers */
 719                                       (strcmp(mse->level, "raid0") != 0 &&
 720                                        strcmp(mse->level, "linear") != 0))) {
 721                         struct state *st = xcalloc(1, sizeof *st);
 722                         mdu_array_info_t array;
 723                         int fd;
 724
 725                         name = get_md_name(mse->devnm);
 726                         if (!name) {
 727                                 free(st);
 728                                 continue;
 729                         }
 730
 731                         st->devname = xstrdup(name);
 732                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 733                             md_get_array_info(fd, &array) < 0) {
 734                                 /* no such array */
 735                                 if (fd >= 0)
 736                                         close(fd);
 737                                 put_md_name(st->devname);
 738                                 free(st->devname);
 739                                 if (st->metadata) {
 740                                         st->metadata->ss->free_super(st->metadata);
 741                                         free(st->metadata);
 742                                 }
 743                                 free(st);
 744                                 continue;
 745                         }
 746                         close(fd);
 747                         st->next = *statelist;
 748                         st->err = 1;
 749                         st->from_auto = 1;
 750                         strcpy(st->devnm, mse->devnm);
 751                         st->percent = RESYNC_UNKNOWN;
 752                         st->expected_spares = -1;
 753                         if (mse->metadata_version &&
 754                             strncmp(mse->metadata_version,
 755                                     "external:", 9) == 0 &&
 756                             is_subarray(mse->metadata_version+9)) {
 757                                 char *sl;
 758                                 strcpy(st->parent_devnm,
 759                                         mse->metadata_version+10);
 760                                 sl = strchr(st->parent_devnm, '/');
 761                                 *sl = 0;
 762                         } else
 763                                 st->parent_devnm[0] = 0;
 764                         *statelist = st;
 765                         if (test)
 766                                 alert("TestMessage", st->devname, NULL, info);
 767                         new_found = 1;
 768                 }
 769         return new_found;
 770 }
 771
 772 static int get_required_spare_criteria(struct state *st,
 773                                        struct spare_criteria *sc)
 774 {
 775         int fd;
 776
 777         if (!st->metadata || !st->metadata->ss->get_spare_criteria) {
 778                 sc->min_size = 0;
 779                 sc->sector_size = 0;
 780                 return 0;
 781         }
 782
 783         fd = open(st->devname, O_RDONLY);
 784         if (fd < 0)
 785                 return 1;
 786         if (st->metadata->ss->external)
 787                 st->metadata->ss->load_container(st->metadata, fd, st->devname);
 788         else
 789                 st->metadata->ss->load_super(st->metadata, fd, st->devname);
 790         close(fd);
 791         if (!st->metadata->sb)
 792                 return 1;
 793
 794         st->metadata->ss->get_spare_criteria(st->metadata, sc);
 795         st->metadata->ss->free_super(st->metadata);
 796
 797         return 0;
 798 }
 799
 800 static int check_donor(struct state *from, struct state *to)
 801 {
 802         struct state *sub;
 803
 804         if (from == to)
 805                 return 0;
 806         if (from->parent)
 807                 /* Cannot move from a member */
 808                 return 0;
 809         if (from->err)
 810                 return 0;
 811         for (sub = from->subarray; sub; sub = sub->subarray)
 812                 /* If source array has degraded subarrays, don't
 813                  * remove anything
 814                  */
 815                 if (sub->active < sub->raid)
 816                         return 0;
 817         if (from->metadata->ss->external == 0)
 818                 if (from->active < from->raid)
 819                         return 0;
 820         if (from->spare <= 0)
 821                 return 0;
 822         return 1;
 823 }
 824
 825 static dev_t choose_spare(struct state *from, struct state *to,
 826                           struct domainlist *domlist, struct spare_criteria *sc)
 827 {
 828         int d;
 829         dev_t dev = 0;
 830
 831         for (d = from->raid; !dev && d < MAX_DISKS; d++) {
 832                 if (from->devid[d] > 0 && from->devstate[d] == 0) {
 833                         struct dev_policy *pol;
 834                         unsigned long long dev_size;
 835                         unsigned int dev_sector_size;
 836
 837                         if (to->metadata->ss->external &&
 838                             test_partition_from_id(from->devid[d]))
 839                                 continue;
 840
 841                         if (sc->min_size &&
 842                             dev_size_from_id(from->devid[d], &dev_size) &&
 843                             dev_size < sc->min_size)
 844                                 continue;
 845
 846                         if (sc->sector_size &&
 847                             dev_sector_size_from_id(from->devid[d],
 848                                                     &dev_sector_size) &&
 849                             sc->sector_size != dev_sector_size)
 850                                 continue;
 851
 852                         pol = devid_policy(from->devid[d]);
 853                         if (from->spare_group)
 854                                 pol_add(&pol, pol_domain,
 855                                         from->spare_group, NULL);
 856                         if (domain_test(domlist, pol,
 857                                         to->metadata->ss->name) == 1)
 858                             dev = from->devid[d];
 859                         dev_policy_free(pol);
 860                 }
 861         }
 862         return dev;
 863 }
 864
 865 static dev_t container_choose_spare(struct state *from, struct state *to,
 866                                     struct domainlist *domlist,
 867                                     struct spare_criteria *sc, int active)
 868 {
 869         /* This is similar to choose_spare, but we cannot trust devstate,
 870          * so we need to read the metadata instead
 871          */
 872         struct mdinfo *list;
 873         struct supertype *st = from->metadata;
 874         int fd = open(from->devname, O_RDONLY);
 875         int err;
 876         dev_t dev = 0;
 877
 878         if (fd < 0)
 879                 return 0;
 880         if (!st->ss->getinfo_super_disks) {
 881                 close(fd);
 882                 return 0;
 883         }
 884
 885         err = st->ss->load_container(st, fd, NULL);
 886         close(fd);
 887         if (err)
 888                 return 0;
 889
 890         if (from == to) {
 891                 /* We must check if number of active disks has not increased
 892                  * since ioctl in main loop. mdmon may have added spare
 893                  * to subarray. If so we do not need to look for more spares
 894                  * so return non zero value */
 895                 int active_cnt = 0;
 896                 struct mdinfo *dp;
 897                 list = st->ss->getinfo_super_disks(st);
 898                 if (!list) {
 899                         st->ss->free_super(st);
 900                         return 1;
 901                 }
 902                 dp = list->devs;
 903                 while (dp) {
 904                         if (dp->disk.state & (1 << MD_DISK_SYNC) &&
 905                             !(dp->disk.state & (1 << MD_DISK_FAULTY)))
 906                                 active_cnt++;
 907                         dp = dp->next;
 908                 }
 909                 sysfs_free(list);
 910                 if (active < active_cnt) {
 911                         /* Spare just activated.*/
 912                         st->ss->free_super(st);
 913                         return 1;
 914                 }
 915         }
 916
 917         /* We only need one spare so full list not needed */
 918         list = container_choose_spares(st, sc, domlist, from->spare_group,
 919                                        to->metadata->ss->name, 1);
 920         if (list) {
 921                 struct mdinfo *disks = list->devs;
 922                 if (disks)
 923                         dev = makedev(disks->disk.major, disks->disk.minor);
 924                 sysfs_free(list);
 925         }
 926         st->ss->free_super(st);
 927         return dev;
 928 }
 929
 930 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 931 {
 932         struct state *from;
 933         struct state *st;
 934         struct spare_criteria sc;
 935
 936         link_containers_with_subarrays(statelist);
 937         for (st = statelist; st; st = st->next)
 938                 if (st->active < st->raid && st->spare == 0 && !st->err) {
 939                         struct domainlist *domlist = NULL;
 940                         int d;
 941                         struct state *to = st;
 942
 943                         if (to->parent_devnm[0] && !to->parent)
 944                                 /* subarray monitored without parent container
 945                                  * we can't move spares here */
 946                                 continue;
 947
 948                         if (to->parent)
 949                                 /* member of a container */
 950                                 to = to->parent;
 951
 952                         if (get_required_spare_criteria(to, &sc))
 953                                 continue;
 954                         if (to->metadata->ss->external) {
 955                                 /* We must make sure there is
 956                                  * no suitable spare in container already.
 957                                  * If there is we don't add more */
 958                                 dev_t devid = container_choose_spare(
 959                                         to, to, NULL, &sc, st->active);
 960                                 if (devid > 0)
 961                                         continue;
 962                         }
 963                         for (d = 0; d < MAX_DISKS; d++)
 964                                 if (to->devid[d])
 965                                         domainlist_add_dev(&domlist,
 966                                                            to->devid[d],
 967                                                            to->metadata->ss->name);
 968                         if (to->spare_group)
 969                                 domain_add(&domlist, to->spare_group);
 970                         /*
 971                          * No spare migration if the destination
 972                          * has no domain. Skip this array.
 973                          */
 974                         if (!domlist)
 975                                 continue;
 976                         for (from=statelist ; from ; from=from->next) {
 977                                 dev_t devid;
 978                                 if (!check_donor(from, to))
 979                                         continue;
 980                                 if (from->metadata->ss->external)
 981                                         devid = container_choose_spare(
 982                                                 from, to, domlist, &sc, 0);
 983                                 else
 984                                         devid = choose_spare(from, to, domlist,
 985                                                              &sc);
 986                                 if (devid > 0 &&
 987                                     move_spare(from->devname, to->devname,
 988                                                devid)) {
 989                                         alert("MoveSpare", to->devname,
 990                                               from->devname, info);
 991                                         break;
 992                                 }
 993                         }
 994                         domain_free(domlist);
 995                 }
 996 }
 997
 998 /* search the statelist to connect external
 999  * metadata subarrays with their containers
1000  * We always completely rebuild the tree from scratch as
1001  * that is safest considering the possibility of entries
1002  * disappearing or changing.
1003  */
1004 static void link_containers_with_subarrays(struct state *list)
1005 {
1006         struct state *st;
1007         struct state *cont;
1008         for (st = list; st; st = st->next) {
1009                 st->parent = NULL;
1010                 st->subarray = NULL;
1011         }
1012         for (st = list; st; st = st->next)
1013                 if (st->parent_devnm[0])
1014                         for (cont = list; cont; cont = cont->next)
1015                                 if (!cont->err && cont->parent_devnm[0] == 0 &&
1016                                     strcmp(cont->devnm, st->parent_devnm) == 0) {
1017                                         st->parent = cont;
1018                                         st->subarray = cont->subarray;
1019                                         cont->subarray = st;
1020                                         break;
1021                                 }
1022 }
1023
1024 /* Not really Monitor but ... */
1025 int Wait(char *dev)
1026 {
1027         char devnm[32];
1028         dev_t rdev;
1029         char *tmp;
1030         int rv = 1;
1031         int frozen_remaining = 3;
1032
1033         if (!stat_is_blkdev(dev, &rdev))
1034                 return 2;
1035
1036         tmp = devid2devnm(rdev);
1037         if (!tmp) {
1038                 pr_err("Cannot get md device name.\n");
1039                 return 2;
1040         }
1041
1042         strcpy(devnm, tmp);
1043
1044         while(1) {
1045                 struct mdstat_ent *ms = mdstat_read(1, 0);
1046                 struct mdstat_ent *e;
1047
1048                 for (e = ms; e; e = e->next)
1049                         if (strcmp(e->devnm, devnm) == 0)
1050                                 break;
1051
1052                 if (e && e->percent == RESYNC_NONE) {
1053                         /* We could be in the brief pause before something
1054                          * starts. /proc/mdstat doesn't show that, but
1055                          * sync_action does.
1056                          */
1057                         struct mdinfo mdi;
1058                         char buf[21];
1059
1060                         if (sysfs_init(&mdi, -1, devnm))
1061                                 return 2;
1062                         if (sysfs_get_str(&mdi, NULL, "sync_action",
1063                                           buf, 20) > 0 &&
1064                             strcmp(buf,"idle\n") != 0) {
1065                                 e->percent = RESYNC_UNKNOWN;
1066                                 if (strcmp(buf, "frozen\n") == 0) {
1067                                         if (frozen_remaining == 0)
1068                                                 e->percent = RESYNC_NONE;
1069                                         else
1070                                                 frozen_remaining -= 1;
1071                                 }
1072                         }
1073                 }
1074                 if (!e || e->percent == RESYNC_NONE) {
1075                         if (e && e->metadata_version &&
1076                             strncmp(e->metadata_version, "external:", 9) == 0) {
1077                                 if (is_subarray(&e->metadata_version[9]))
1078                                         ping_monitor(&e->metadata_version[9]);
1079                                 else
1080                                         ping_monitor(devnm);
1081                         }
1082                         free_mdstat(ms);
1083                         return rv;
1084                 }
1085                 free_mdstat(ms);
1086                 rv = 0;
1087                 mdstat_wait(5);
1088         }
1089 }
1090
1091 /* The state "broken" is used only for RAID0/LINEAR - it's the same as
1092  * "clean", but used in case the array has one or more members missing.
1093  */
1094 static char *clean_states[] = {
1095         "clear", "inactive", "readonly", "read-auto", "clean", "broken", NULL };
1096
1097 int WaitClean(char *dev, int verbose)
1098 {
1099         int fd;
1100         struct mdinfo *mdi;
1101         int rv = 1;
1102         char devnm[32];
1103
1104         if (!stat_is_blkdev(dev, NULL))
1105                 return 2;
1106         fd = open(dev, O_RDONLY);
1107         if (fd < 0) {
1108                 if (verbose)
1109                         pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1110                 return 1;
1111         }
1112
1113         strcpy(devnm, fd2devnm(fd));
1114         mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1115         if (!mdi) {
1116                 if (verbose)
1117                         pr_err("Failed to read sysfs attributes for %s\n", dev);
1118                 close(fd);
1119                 return 0;
1120         }
1121
1122         switch(mdi->array.level) {
1123         case LEVEL_LINEAR:
1124         case LEVEL_MULTIPATH:
1125         case 0:
1126                 /* safemode delay is irrelevant for these levels */
1127                 rv = 0;
1128         }
1129
1130         /* for internal metadata the kernel handles the final clean
1131          * transition, containers can never be dirty
1132          */
1133         if (!is_subarray(mdi->text_version))
1134                 rv = 0;
1135
1136         /* safemode disabled ? */
1137         if (mdi->safe_mode_delay == 0)
1138                 rv = 0;
1139
1140         if (rv) {
1141                 int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state");
1142                 char buf[20];
1143                 int delay = 5000;
1144
1145                 /* minimize the safe_mode_delay and prepare to wait up to 5s
1146                  * for writes to quiesce
1147                  */
1148                 sysfs_set_safemode(mdi, 1);
1149
1150                 /* wait for array_state to be clean */
1151                 while (1) {
1152                         rv = read(state_fd, buf, sizeof(buf));
1153                         if (rv < 0)
1154                                 break;
1155                         if (sysfs_match_word(buf, clean_states) <
1156                             (int)ARRAY_SIZE(clean_states) - 1)
1157                                 break;
1158                         rv = sysfs_wait(state_fd, &delay);
1159                         if (rv < 0 && errno != EINTR)
1160                                 break;
1161                         lseek(state_fd, 0, SEEK_SET);
1162                 }
1163                 if (rv < 0)
1164                         rv = 1;
1165                 else if (ping_monitor(mdi->text_version) == 0) {
1166                         /* we need to ping to close the window between array
1167                          * state transitioning to clean and the metadata being
1168                          * marked clean
1169                          */
1170                         rv = 0;
1171                 } else {
1172                         rv = 1;
1173                         pr_err("Error connecting monitor with %s\n", dev);
1174                 }
1175                 if (rv && verbose)
1176                         pr_err("Error waiting for %s to be clean\n", dev);
1177
1178                 /* restore the original safe_mode_delay */
1179                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1180                 close(state_fd);
1181         }
1182
1183         sysfs_free(mdi);
1184         close(fd);
1185
1186         return rv;
1187 }