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         struct mdinfo *sra = NULL;
 455         mdu_array_info_t array;
 456         struct mdstat_ent *mse = NULL, *mse2;
 457         char *dev = st->devname;
 458         int fd;
 459         int i;
 460         int remaining_disks;
 461         int last_disk;
 462         int new_array = 0;
 463         int retval;
 464
 465         if (test)
 466                 alert("TestMessage", dev, NULL, ainfo);
 467
 468         retval = 0;
 469
 470         fd = open(dev, O_RDONLY);
 471         if (fd < 0)
 472                 goto disappeared;
 473
 474         if (!md_array_active(fd))
 475                 goto disappeared;
 476
 477         fcntl(fd, F_SETFD, FD_CLOEXEC);
 478         if (md_get_array_info(fd, &array) < 0)
 479                 goto disappeared;
 480
 481         if (st->devnm[0] == 0)
 482                 strcpy(st->devnm, fd2devnm(fd));
 483
 484         sra = sysfs_read(-1, st->devnm, GET_LEVEL | GET_DISKS | GET_DEGRADED |
 485                          GET_MISMATCH | GET_DEVS | GET_STATE);
 486         if (!sra)
 487                 goto disappeared;
 488
 489         /* It's much easier to list what array levels can't
 490          * have a device disappear than all of them that can
 491          */
 492         if (sra->array.level == 0 || sra->array.level == -1) {
 493                 if (!st->err && !st->from_config)
 494                         alert("DeviceDisappeared", dev, " Wrong-Level", ainfo);
 495                 st->err++;
 496                 goto out;
 497         }
 498
 499         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 500                 if (strcmp(mse2->devnm, st->devnm) == 0) {
 501                         mse2->devnm[0] = 0; /* flag it as "used" */
 502                         mse = mse2;
 503                 }
 504
 505         if (!mse) {
 506                 /* duplicated array in statelist
 507                  * or re-created after reading mdstat*/
 508                 st->err++;
 509                 close(fd);
 510                 goto out;
 511         }
 512         /* this array is in /proc/mdstat */
 513         if (array.utime == 0)
 514                 /* external arrays don't update utime, so
 515                  * just make sure it is always different. */
 516                 array.utime = st->utime + 1;;
 517
 518         if (st->err) {
 519                 /* New array appeared where previously had an error */
 520                 st->err = 0;
 521                 st->percent = RESYNC_NONE;
 522                 new_array = 1;
 523                 alert("NewArray", st->devname, NULL, ainfo);
 524         }
 525
 526         if (st->utime == array.utime && st->failed == sra->array.failed_disks &&
 527             st->working == sra->array.working_disks &&
 528             st->spare == sra->array.spare_disks &&
 529             (mse == NULL  || (mse->percent == st->percent))) {
 530                 if ((st->active < st->raid) && st->spare == 0)
 531                         retval = 1;
 532                 goto out;
 533         }
 534         if (st->utime == 0 && /* new array */
 535             mse->pattern && strchr(mse->pattern, '_') /* degraded */)
 536                 alert("DegradedArray", dev, NULL, ainfo);
 537
 538         if (st->utime == 0 && /* new array */ st->expected_spares > 0 &&
 539             sra->array.spare_disks < st->expected_spares)
 540                 alert("SparesMissing", dev, NULL, ainfo);
 541         if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
 542             mse->percent >= 0)
 543                 alert("RebuildStarted", dev, NULL, ainfo);
 544         if (st->percent >= 0 && mse->percent >= 0 &&
 545             (mse->percent / increments) > (st->percent / increments)) {
 546                 char percentalert[15];
 547                 /*
 548                  * "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 549                  */
 550
 551                 if((mse->percent / increments) == 0)
 552                         snprintf(percentalert, sizeof(percentalert),
 553                                  "RebuildStarted");
 554                 else
 555                         snprintf(percentalert, sizeof(percentalert),
 556                                  "Rebuild%02d", mse->percent);
 557
 558                 alert(percentalert, dev, NULL, ainfo);
 559         }
 560
 561         if (mse->percent == RESYNC_NONE && st->percent >= 0) {
 562                 /* Rebuild/sync/whatever just finished.
 563                  * If there is a number in /mismatch_cnt,
 564                  * we should report that.
 565                  */
 566                 if (sra && sra->mismatch_cnt > 0) {
 567                         char cnt[80];
 568                         snprintf(cnt, sizeof(cnt),
 569                                  " mismatches found: %d (on raid level %d)",
 570                                 sra->mismatch_cnt, sra->array.level);
 571                         alert("RebuildFinished", dev, cnt, ainfo);
 572                 } else
 573                         alert("RebuildFinished", dev, NULL, ainfo);
 574         }
 575         st->percent = mse->percent;
 576
 577         remaining_disks = sra->array.nr_disks;
 578         for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) {
 579                 mdu_disk_info_t disc;
 580                 disc.number = i;
 581                 if (md_get_disk_info(fd, &disc) >= 0) {
 582                         info[i].state = disc.state;
 583                         info[i].major = disc.major;
 584                         info[i].minor = disc.minor;
 585                         if (disc.major || disc.minor)
 586                                 remaining_disks --;
 587                 } else
 588                         info[i].major = info[i].minor = 0;
 589         }
 590         last_disk = i;
 591
 592         if (mse->metadata_version &&
 593             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 594             is_subarray(mse->metadata_version+9)) {
 595                 char *sl;
 596                 strcpy(st->parent_devnm, mse->metadata_version+10);
 597                 sl = strchr(st->parent_devnm, '/');
 598                 if (sl)
 599                         *sl = 0;
 600         } else
 601                 st->parent_devnm[0] = 0;
 602         if (st->metadata == NULL && st->parent_devnm[0] == 0)
 603                 st->metadata = super_by_fd(fd, NULL);
 604
 605         for (i=0; i<MAX_DISKS; i++) {
 606                 mdu_disk_info_t disc = {0,0,0,0,0};
 607                 int newstate=0;
 608                 int change;
 609                 char *dv = NULL;
 610                 disc.number = i;
 611                 if (i < last_disk && (info[i].major || info[i].minor)) {
 612                         newstate = info[i].state;
 613                         dv = map_dev_preferred(info[i].major, info[i].minor, 1,
 614                                                prefer);
 615                         disc.state = newstate;
 616                         disc.major = info[i].major;
 617                         disc.minor = info[i].minor;
 618                 } else
 619                         newstate = (1 << MD_DISK_REMOVED);
 620
 621                 if (dv == NULL && st->devid[i])
 622                         dv = map_dev_preferred(major(st->devid[i]),
 623                                                minor(st->devid[i]), 1, prefer);
 624                 change = newstate ^ st->devstate[i];
 625                 if (st->utime && change && !st->err && !new_array) {
 626                         if ((st->devstate[i]&change) & (1 << MD_DISK_SYNC))
 627                                 alert("Fail", dev, dv, ainfo);
 628                         else if ((newstate & (1 << MD_DISK_FAULTY)) &&
 629                                  (disc.major || disc.minor) &&
 630                                  st->devid[i] == makedev(disc.major,
 631                                                          disc.minor))
 632                                 alert("FailSpare", dev, dv, ainfo);
 633                         else if ((newstate&change) & (1 << MD_DISK_SYNC))
 634                                 alert("SpareActive", dev, dv, ainfo);
 635                 }
 636                 st->devstate[i] = newstate;
 637                 st->devid[i] = makedev(disc.major, disc.minor);
 638         }
 639         st->active = sra->array.active_disks;
 640         st->working = sra->array.working_disks;
 641         st->spare = sra->array.spare_disks;
 642         st->failed = sra->array.failed_disks;
 643         st->utime = array.utime;
 644         st->raid = sra->array.raid_disks;
 645         st->err = 0;
 646         if ((st->active < st->raid) && st->spare == 0)
 647                 retval = 1;
 648
 649  out:
 650         if (sra)
 651                 sysfs_free(sra);
 652         if (fd > 0)
 653                 close(fd);
 654         return retval;
 655
 656  disappeared:
 657         if (!st->err)
 658                 alert("DeviceDisappeared", dev, NULL, ainfo);
 659         st->err++;
 660         goto out;
 661 }
 662
 663 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
 664                           int test, struct alert_info *info)
 665 {
 666         struct mdstat_ent *mse;
 667         int new_found = 0;
 668         char *name;
 669
 670         for (mse=mdstat; mse; mse=mse->next)
 671                 if (mse->devnm[0] &&
 672                     (!mse->level  || /* retrieve containers */
 673                      (strcmp(mse->level, "raid0") != 0 &&
 674                       strcmp(mse->level, "linear") != 0))
 675                         ) {
 676                         struct state *st = xcalloc(1, sizeof *st);
 677                         mdu_array_info_t array;
 678                         int fd;
 679
 680                         name = get_md_name(mse->devnm);
 681                         if (!name) {
 682                                 free(st);
 683                                 continue;
 684                         }
 685
 686                         st->devname = xstrdup(name);
 687                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 688                             md_get_array_info(fd, &array) < 0) {
 689                                 /* no such array */
 690                                 if (fd >= 0)
 691                                         close(fd);
 692                                 put_md_name(st->devname);
 693                                 free(st->devname);
 694                                 if (st->metadata) {
 695                                         st->metadata->ss->free_super(st->metadata);
 696                                         free(st->metadata);
 697                                 }
 698                                 free(st);
 699                                 continue;
 700                         }
 701                         close(fd);
 702                         st->next = *statelist;
 703                         st->err = 1;
 704                         st->from_auto = 1;
 705                         strcpy(st->devnm, mse->devnm);
 706                         st->percent = RESYNC_UNKNOWN;
 707                         st->expected_spares = -1;
 708                         if (mse->metadata_version &&
 709                             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 710                             is_subarray(mse->metadata_version+9)) {
 711                                 char *sl;
 712                                 strcpy(st->parent_devnm,
 713                                         mse->metadata_version+10);
 714                                 sl = strchr(st->parent_devnm, '/');
 715                                 *sl = 0;
 716                         } else
 717                                 st->parent_devnm[0] = 0;
 718                         *statelist = st;
 719                         if (test)
 720                                 alert("TestMessage", st->devname, NULL, info);
 721                         new_found = 1;
 722                 }
 723         return new_found;
 724 }
 725
 726 static int get_required_spare_criteria(struct state *st,
 727                                        struct spare_criteria *sc)
 728 {
 729         int fd;
 730
 731         if (!st->metadata ||
 732             !st->metadata->ss->get_spare_criteria) {
 733                 sc->min_size = 0;
 734                 sc->sector_size = 0;
 735                 return 0;
 736         }
 737
 738         fd = open(st->devname, O_RDONLY);
 739         if (fd < 0)
 740                 return 1;
 741         if (st->metadata->ss->external)
 742                 st->metadata->ss->load_container(st->metadata, fd, st->devname);
 743         else
 744                 st->metadata->ss->load_super(st->metadata, fd, st->devname);
 745         close(fd);
 746         if (!st->metadata->sb)
 747                 return 1;
 748
 749         st->metadata->ss->get_spare_criteria(st->metadata, sc);
 750         st->metadata->ss->free_super(st->metadata);
 751
 752         return 0;
 753 }
 754
 755 static int check_donor(struct state *from, struct state *to)
 756 {
 757         struct state *sub;
 758
 759         if (from == to)
 760                 return 0;
 761         if (from->parent)
 762                 /* Cannot move from a member */
 763                 return 0;
 764         if (from->err)
 765                 return 0;
 766         for (sub = from->subarray; sub; sub = sub->subarray)
 767                 /* If source array has degraded subarrays, don't
 768                  * remove anything
 769                  */
 770                 if (sub->active < sub->raid)
 771                         return 0;
 772         if (from->metadata->ss->external == 0)
 773                 if (from->active < from->raid)
 774                         return 0;
 775         if (from->spare <= 0)
 776                 return 0;
 777         return 1;
 778 }
 779
 780 static dev_t choose_spare(struct state *from, struct state *to,
 781                         struct domainlist *domlist, struct spare_criteria *sc)
 782 {
 783         int d;
 784         dev_t dev = 0;
 785
 786         for (d = from->raid; !dev && d < MAX_DISKS; d++) {
 787                 if (from->devid[d] > 0 &&
 788                     from->devstate[d] == 0) {
 789                         struct dev_policy *pol;
 790                         unsigned long long dev_size;
 791                         unsigned int dev_sector_size;
 792
 793                         if (to->metadata->ss->external &&
 794                             test_partition_from_id(from->devid[d]))
 795                                 continue;
 796
 797                         if (sc->min_size &&
 798                             dev_size_from_id(from->devid[d], &dev_size) &&
 799                             dev_size < sc->min_size)
 800                                 continue;
 801
 802                         if (sc->sector_size &&
 803                             dev_sector_size_from_id(from->devid[d],
 804                                                     &dev_sector_size) &&
 805                             sc->sector_size != dev_sector_size)
 806                                 continue;
 807
 808                         pol = devid_policy(from->devid[d]);
 809                         if (from->spare_group)
 810                                 pol_add(&pol, pol_domain,
 811                                         from->spare_group, NULL);
 812                         if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
 813                             dev = from->devid[d];
 814                         dev_policy_free(pol);
 815                 }
 816         }
 817         return dev;
 818 }
 819
 820 static dev_t container_choose_spare(struct state *from, struct state *to,
 821                                     struct domainlist *domlist,
 822                                     struct spare_criteria *sc, int active)
 823 {
 824         /* This is similar to choose_spare, but we cannot trust devstate,
 825          * so we need to read the metadata instead
 826          */
 827         struct mdinfo *list;
 828         struct supertype *st = from->metadata;
 829         int fd = open(from->devname, O_RDONLY);
 830         int err;
 831         dev_t dev = 0;
 832
 833         if (fd < 0)
 834                 return 0;
 835         if (!st->ss->getinfo_super_disks) {
 836                 close(fd);
 837                 return 0;
 838         }
 839
 840         err = st->ss->load_container(st, fd, NULL);
 841         close(fd);
 842         if (err)
 843                 return 0;
 844
 845         if (from == to) {
 846                 /* We must check if number of active disks has not increased
 847                  * since ioctl in main loop. mdmon may have added spare
 848                  * to subarray. If so we do not need to look for more spares
 849                  * so return non zero value */
 850                 int active_cnt = 0;
 851                 struct mdinfo *dp;
 852                 list = st->ss->getinfo_super_disks(st);
 853                 if (!list) {
 854                         st->ss->free_super(st);
 855                         return 1;
 856                 }
 857                 dp = list->devs;
 858                 while (dp) {
 859                         if (dp->disk.state & (1<<MD_DISK_SYNC) &&
 860                             !(dp->disk.state & (1<<MD_DISK_FAULTY)))
 861                                 active_cnt++;
 862                         dp = dp->next;
 863                 }
 864                 sysfs_free(list);
 865                 if (active < active_cnt) {
 866                         /* Spare just activated.*/
 867                         st->ss->free_super(st);
 868                         return 1;
 869                 }
 870         }
 871
 872         /* We only need one spare so full list not needed */
 873         list = container_choose_spares(st, sc, domlist, from->spare_group,
 874                                        to->metadata->ss->name, 1);
 875         if (list) {
 876                 struct mdinfo *disks = list->devs;
 877                 if (disks)
 878                         dev = makedev(disks->disk.major, disks->disk.minor);
 879                 sysfs_free(list);
 880         }
 881         st->ss->free_super(st);
 882         return dev;
 883 }
 884
 885 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 886 {
 887         struct state *from;
 888         struct state *st;
 889         struct spare_criteria sc;
 890
 891         link_containers_with_subarrays(statelist);
 892         for (st = statelist; st; st = st->next)
 893                 if (st->active < st->raid &&
 894                     st->spare == 0 && !st->err) {
 895                         struct domainlist *domlist = NULL;
 896                         int d;
 897                         struct state *to = st;
 898
 899                         if (to->parent_devnm[0] && !to->parent)
 900                                 /* subarray monitored without parent container
 901                                  * we can't move spares here */
 902                                 continue;
 903
 904                         if (to->parent)
 905                                 /* member of a container */
 906                                 to = to->parent;
 907
 908                         if (get_required_spare_criteria(to, &sc))
 909                                 continue;
 910                         if (to->metadata->ss->external) {
 911                                 /* We must make sure there is
 912                                  * no suitable spare in container already.
 913                                  * If there is we don't add more */
 914                                 dev_t devid = container_choose_spare(
 915                                         to, to, NULL, &sc, st->active);
 916                                 if (devid > 0)
 917                                         continue;
 918                         }
 919                         for (d = 0; d < MAX_DISKS; d++)
 920                                 if (to->devid[d])
 921                                         domainlist_add_dev(&domlist,
 922                                                            to->devid[d],
 923                                                            to->metadata->ss->name);
 924                         if (to->spare_group)
 925                                 domain_add(&domlist, to->spare_group);
 926                         /*
 927                          * No spare migration if the destination
 928                          * has no domain. Skip this array.
 929                          */
 930                         if (!domlist)
 931                                 continue;
 932                         for (from=statelist ; from ; from=from->next) {
 933                                 dev_t devid;
 934                                 if (!check_donor(from, to))
 935                                         continue;
 936                                 if (from->metadata->ss->external)
 937                                         devid = container_choose_spare(
 938                                                 from, to, domlist, &sc, 0);
 939                                 else
 940                                         devid = choose_spare(from, to, domlist,
 941                                                              &sc);
 942                                 if (devid > 0
 943                                     && move_spare(from->devname, to->devname, devid)) {
 944                                         alert("MoveSpare", to->devname, from->devname, info);
 945                                         break;
 946                                 }
 947                         }
 948                         domain_free(domlist);
 949                 }
 950 }
 951
 952 /* search the statelist to connect external
 953  * metadata subarrays with their containers
 954  * We always completely rebuild the tree from scratch as
 955  * that is safest considering the possibility of entries
 956  * disappearing or changing.
 957  */
 958 static void link_containers_with_subarrays(struct state *list)
 959 {
 960         struct state *st;
 961         struct state *cont;
 962         for (st = list; st; st = st->next) {
 963                 st->parent = NULL;
 964                 st->subarray = NULL;
 965         }
 966         for (st = list; st; st = st->next)
 967                 if (st->parent_devnm[0])
 968                         for (cont = list; cont; cont = cont->next)
 969                                 if (!cont->err &&
 970                                     cont->parent_devnm[0] == 0 &&
 971                                     strcmp(cont->devnm, st->parent_devnm) == 0) {
 972                                         st->parent = cont;
 973                                         st->subarray = cont->subarray;
 974                                         cont->subarray = st;
 975                                         break;
 976                                 }
 977 }
 978
 979 /* Not really Monitor but ... */
 980 int Wait(char *dev)
 981 {
 982         char devnm[32];
 983         int rv = 1;
 984         int frozen_remaining = 3;
 985
 986         if (!stat_is_blkdev(dev, NULL))
 987                 return 2;
 988         strcpy(devnm, dev);
 989
 990         while(1) {
 991                 struct mdstat_ent *ms = mdstat_read(1, 0);
 992                 struct mdstat_ent *e;
 993
 994                 for (e=ms ; e; e=e->next)
 995                         if (strcmp(e->devnm, devnm) == 0)
 996                                 break;
 997
 998                 if (e && e->percent == RESYNC_NONE) {
 999                         /* We could be in the brief pause before something
1000                          * starts. /proc/mdstat doesn't show that, but
1001                          * sync_action does.
1002                          */
1003                         struct mdinfo mdi;
1004                         char buf[21];
1005
1006                         if (sysfs_init(&mdi, -1, devnm))
1007                                 return 2;
1008                         if (sysfs_get_str(&mdi, NULL, "sync_action",
1009                                           buf, 20) > 0 &&
1010                             strcmp(buf,"idle\n") != 0) {
1011                                 e->percent = RESYNC_UNKNOWN;
1012                                 if (strcmp(buf, "frozen\n") == 0) {
1013                                         if (frozen_remaining == 0)
1014                                                 e->percent = RESYNC_NONE;
1015                                         else
1016                                                 frozen_remaining -= 1;
1017                                 }
1018                         }
1019                 }
1020                 if (!e || e->percent == RESYNC_NONE) {
1021                         if (e && e->metadata_version &&
1022                             strncmp(e->metadata_version, "external:", 9) == 0) {
1023                                 if (is_subarray(&e->metadata_version[9]))
1024                                         ping_monitor(&e->metadata_version[9]);
1025                                 else
1026                                         ping_monitor(devnm);
1027                         }
1028                         free_mdstat(ms);
1029                         return rv;
1030                 }
1031                 free_mdstat(ms);
1032                 rv = 0;
1033                 mdstat_wait(5);
1034         }
1035 }
1036
1037 static char *clean_states[] = {
1038         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1039
1040 int WaitClean(char *dev, int sock, int verbose)
1041 {
1042         int fd;
1043         struct mdinfo *mdi;
1044         int rv = 1;
1045         char devnm[32];
1046
1047         if (!stat_is_blkdev(dev, NULL))
1048                 return 2;
1049         fd = open(dev, O_RDONLY);
1050         if (fd < 0) {
1051                 if (verbose)
1052                         pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1053                 return 1;
1054         }
1055
1056         strcpy(devnm, fd2devnm(fd));
1057         mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1058         if (!mdi) {
1059                 if (verbose)
1060                         pr_err("Failed to read sysfs attributes for %s\n", dev);
1061                 close(fd);
1062                 return 0;
1063         }
1064
1065         switch(mdi->array.level) {
1066         case LEVEL_LINEAR:
1067         case LEVEL_MULTIPATH:
1068         case 0:
1069                 /* safemode delay is irrelevant for these levels */
1070                 rv = 0;
1071         }
1072
1073         /* for internal metadata the kernel handles the final clean
1074          * transition, containers can never be dirty
1075          */
1076         if (!is_subarray(mdi->text_version))
1077                 rv = 0;
1078
1079         /* safemode disabled ? */
1080         if (mdi->safe_mode_delay == 0)
1081                 rv = 0;
1082
1083         if (rv) {
1084                 int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state");
1085                 char buf[20];
1086                 int delay = 5000;
1087
1088                 /* minimize the safe_mode_delay and prepare to wait up to 5s
1089                  * for writes to quiesce
1090                  */
1091                 sysfs_set_safemode(mdi, 1);
1092
1093                 /* wait for array_state to be clean */
1094                 while (1) {
1095                         rv = read(state_fd, buf, sizeof(buf));
1096                         if (rv < 0)
1097                                 break;
1098                         if (sysfs_match_word(buf, clean_states) <= 4)
1099                                 break;
1100                         rv = sysfs_wait(state_fd, &delay);
1101                         if (rv < 0 && errno != EINTR)
1102                                 break;
1103                         lseek(state_fd, 0, SEEK_SET);
1104                 }
1105                 if (rv < 0)
1106                         rv = 1;
1107                 else if (fping_monitor(sock) == 0 ||
1108                          ping_monitor(mdi->text_version) == 0) {
1109                         /* we need to ping to close the window between array
1110                          * state transitioning to clean and the metadata being
1111                          * marked clean
1112                          */
1113                         rv = 0;
1114                 } else
1115                         rv = 1;
1116                 if (rv && verbose)
1117                         pr_err("Error waiting for %s to be clean\n",
1118                                 dev);
1119
1120                 /* restore the original safe_mode_delay */
1121                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1122                 close(state_fd);
1123         }
1124
1125         sysfs_free(mdi);
1126         close(fd);
1127
1128         return rv;
1129 }