Monitor.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@suse.de>
  23  */
  24
  25 #include        "mdadm.h"
  26 #include        "md_p.h"
  27 #include        "md_u.h"
  28 #include        <sys/wait.h>
  29 #include        <signal.h>
  30 #include        <limits.h>
  31 #include        <syslog.h>
  32
  33 struct state {
  34         char *devname;
  35         char devnm[32]; /* to sync with mdstat info */
  36         unsigned int utime;
  37         int err;
  38         char *spare_group;
  39         int active, working, failed, spare, raid;
  40         int from_config;
  41         int from_auto;
  42         int expected_spares;
  43         int devstate[MAX_DISKS];
  44         dev_t devid[MAX_DISKS];
  45         int percent;
  46         char parent_devnm[32]; /* For subarray, devnm of parent.
  47                                 * For others, ""
  48                                 */
  49         struct supertype *metadata;
  50         struct state *subarray;/* for a container it is a link to first subarray
  51                                 * for a subarray it is a link to next subarray
  52                                 * in the same container */
  53         struct state *parent;  /* for a subarray it is a link to its container
  54                                 */
  55         struct state *next;
  56 };
  57
  58 struct alert_info {
  59         char *mailaddr;
  60         char *mailfrom;
  61         char *alert_cmd;
  62         int dosyslog;
  63 };
  64 static int make_daemon(char *pidfile);
  65 static int check_one_sharer(int scan);
  66 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
  67 static int check_array(struct state *st, struct mdstat_ent *mdstat,
  68                        int test, struct alert_info *info,
  69                        int increments, char *prefer);
  70 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
  71                           int test, struct alert_info *info);
  72 static void try_spare_migration(struct state *statelist, struct alert_info *info);
  73 static void link_containers_with_subarrays(struct state *list);
  74
  75 int Monitor(struct mddev_dev *devlist,
  76             char *mailaddr, char *alert_cmd,
  77             struct context *c,
  78             int daemonise, int oneshot,
  79             int dosyslog, char *pidfile, int increments,
  80             int share)
  81 {
  82         /*
  83          * Every few seconds, scan every md device looking for changes
  84          * When a change is found, log it, possibly run the alert command,
  85          * and possibly send Email
  86          *
  87          * For each array, we record:
  88          *   Update time
  89          *   active/working/failed/spare drives
  90          *   State of each device.
  91          *   %rebuilt if rebuilding
  92          *
  93          * If the update time changes, check out all the data again
  94          * It is possible that we cannot get the state of each device
  95          * due to bugs in the md kernel module.
  96          * We also read /proc/mdstat to get rebuild percent,
  97          * and to get state on all active devices incase of kernel bug.
  98          *
  99          * Events are:
 100          *    Fail
 101          *      An active device had Faulty set or Active/Sync removed
 102          *    FailSpare
 103          *      A spare device had Faulty set
 104          *    SpareActive
 105          *      An active device had a reverse transition
 106          *    RebuildStarted
 107          *      percent went from -1 to +ve
 108          *    RebuildNN
 109          *      percent went from below to not-below NN%
 110          *    DeviceDisappeared
 111          *      Couldn't access a device which was previously visible
 112          *
 113          * if we detect an array with active<raid and spare==0
 114          * we look at other arrays that have same spare-group
 115          * If we find one with active==raid and spare>0,
 116          *  and if we can get_disk_info and find a name
 117          *  Then we hot-remove and hot-add to the other array
 118          *
 119          * If devlist is NULL, then we can monitor everything because --scan
 120          * was given.  We get an initial list from config file and add anything
 121          * that appears in /proc/mdstat
 122          */
 123
 124         struct state *statelist = NULL;
 125         struct state *st2;
 126         int finished = 0;
 127         struct mdstat_ent *mdstat = NULL;
 128         char *mailfrom = NULL;
 129         struct alert_info info;
 130
 131         if (!mailaddr) {
 132                 mailaddr = conf_get_mailaddr();
 133                 if (mailaddr && ! c->scan)
 134                         pr_err("Monitor using email address \"%s\" from config file\n",
 135                                mailaddr);
 136         }
 137         mailfrom = conf_get_mailfrom();
 138
 139         if (!alert_cmd) {
 140                 alert_cmd = conf_get_program();
 141                 if (alert_cmd && ! c->scan)
 142                         pr_err("Monitor using program \"%s\" from config file\n",
 143                                alert_cmd);
 144         }
 145         if (c->scan && !mailaddr && !alert_cmd && !dosyslog) {
 146                 pr_err("No mail address or alert command - not monitoring.\n");
 147                 return 1;
 148         }
 149         info.alert_cmd = alert_cmd;
 150         info.mailaddr = mailaddr;
 151         info.mailfrom = mailfrom;
 152         info.dosyslog = dosyslog;
 153
 154         if (daemonise) {
 155                 int rv = make_daemon(pidfile);
 156                 if (rv >= 0)
 157                         return rv;
 158         }
 159
 160         if (share)
 161                 if (check_one_sharer(c->scan))
 162                         return 1;
 163
 164         if (devlist == NULL) {
 165                 struct mddev_ident *mdlist = conf_get_ident(NULL);
 166                 for (; mdlist; mdlist=mdlist->next) {
 167                         struct state *st;
 168                         if (mdlist->devname == NULL)
 169                                 continue;
 170                         if (strcasecmp(mdlist->devname, "<ignore>") == 0)
 171                                 continue;
 172                         st = xcalloc(1, sizeof *st);
 173                         if (mdlist->devname[0] == '/')
 174                                 st->devname = xstrdup(mdlist->devname);
 175                         else {
 176                                 st->devname = xmalloc(8+strlen(mdlist->devname)+1);
 177                                 strcpy(strcpy(st->devname, "/dev/md/"),
 178                                        mdlist->devname);
 179                         }
 180                         st->next = statelist;
 181                         st->devnm[0] = 0;
 182                         st->percent = RESYNC_UNKNOWN;
 183                         st->from_config = 1;
 184                         st->expected_spares = mdlist->spare_disks;
 185                         if (mdlist->spare_group)
 186                                 st->spare_group = xstrdup(mdlist->spare_group);
 187                         statelist = st;
 188                 }
 189         } else {
 190                 struct mddev_dev *dv;
 191                 for (dv=devlist ; dv; dv=dv->next) {
 192                         struct mddev_ident *mdlist = conf_get_ident(dv->devname);
 193                         struct state *st = xcalloc(1, sizeof *st);
 194                         st->devname = xstrdup(dv->devname);
 195                         st->next = statelist;
 196                         st->devnm[0] = 0;
 197                         st->percent = RESYNC_UNKNOWN;
 198                         st->expected_spares = -1;
 199                         if (mdlist) {
 200                                 st->expected_spares = mdlist->spare_disks;
 201                                 if (mdlist->spare_group)
 202                                         st->spare_group = xstrdup(mdlist->spare_group);
 203                         }
 204                         statelist = st;
 205                 }
 206         }
 207
 208         while (! finished) {
 209                 int new_found = 0;
 210                 struct state *st, **stp;
 211                 int anydegraded = 0;
 212
 213                 if (mdstat)
 214                         free_mdstat(mdstat);
 215                 mdstat = mdstat_read(oneshot?0:1, 0);
 216                 if (!mdstat)
 217                         mdstat_close();
 218
 219                 for (st=statelist; st; st=st->next)
 220                         if (check_array(st, mdstat, c->test, &info,
 221                                         increments, c->prefer))
 222                                 anydegraded = 1;
 223
 224                 /* now check if there are any new devices found in mdstat */
 225                 if (c->scan)
 226                         new_found = add_new_arrays(mdstat, &statelist, c->test,
 227                                                    &info);
 228
 229                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 230                  * Look for another array with spare > 0 and active == raid and same spare_group
 231                  *  if found, choose a device and hotremove/hotadd
 232                  */
 233                 if (share && anydegraded)
 234                         try_spare_migration(statelist, &info);
 235                 if (!new_found) {
 236                         if (oneshot)
 237                                 break;
 238                         else
 239                                 mdstat_wait(c->delay);
 240                 }
 241                 c->test = 0;
 242
 243                 for (stp = &statelist; (st = *stp) != NULL; ) {
 244                         if (st->from_auto && st->err > 5) {
 245                                 *stp = st->next;
 246                                 free(st->devname);
 247                                 free(st->spare_group);
 248                                 free(st);
 249                         } else
 250                                 stp = &st->next;
 251                 }
 252         }
 253         for (st2 = statelist; st2; st2 = statelist) {
 254                 statelist = st2->next;
 255                 free(st2);
 256         }
 257
 258         if (pidfile)
 259                 unlink(pidfile);
 260         return 0;
 261 }
 262
 263 static int make_daemon(char *pidfile)
 264 {
 265         /* Return:
 266          * -1 in the forked daemon
 267          *  0 in the parent
 268          *  1 on error
 269          * so a none-negative becomes the exit code.
 270          */
 271         int pid = fork();
 272         if (pid > 0) {
 273                 if (!pidfile)
 274                         printf("%d\n", pid);
 275                 else {
 276                         FILE *pid_file;
 277                         pid_file=fopen(pidfile, "w");
 278                         if (!pid_file)
 279                                 perror("cannot create pid file");
 280                         else {
 281                                 fprintf(pid_file,"%d\n", pid);
 282                                 fclose(pid_file);
 283                         }
 284                 }
 285                 return 0;
 286         }
 287         if (pid < 0) {
 288                 perror("daemonise");
 289                 return 1;
 290         }
 291         close(0);
 292         open("/dev/null", O_RDWR);
 293         dup2(0,1);
 294         dup2(0,2);
 295         setsid();
 296         return -1;
 297 }
 298
 299 static int check_one_sharer(int scan)
 300 {
 301         int pid, rv;
 302         FILE *fp;
 303         char dir[20];
 304         char path[100];
 305         struct stat buf;
 306         sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
 307         fp = fopen(path, "r");
 308         if (fp) {
 309                 if (fscanf(fp, "%d", &pid) != 1)
 310                         pid = -1;
 311                 sprintf(dir, "/proc/%d", pid);
 312                 rv = stat(dir, &buf);
 313                 if (rv != -1) {
 314                         if (scan) {
 315                                 pr_err("Only one autorebuild process allowed in scan mode, aborting\n");
 316                                 fclose(fp);
 317                                 return 1;
 318                         } else {
 319                                 pr_err("Warning: One autorebuild process already running.\n");
 320                         }
 321                 }
 322                 fclose(fp);
 323         }
 324         if (scan) {
 325                 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
 326                     errno != EEXIST) {
 327                         pr_err("Can't create autorebuild.pid file\n");
 328                 } else {
 329                         fp = fopen(path, "w");
 330                         if (!fp)
 331                                 pr_err("Cannot create autorebuild.pidfile\n");
 332                         else {
 333                                 pid = getpid();
 334                                 fprintf(fp, "%d\n", pid);
 335                                 fclose(fp);
 336                         }
 337                 }
 338         }
 339         return 0;
 340 }
 341
 342 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
 343 {
 344         int priority;
 345
 346         if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
 347                 time_t now = time(0);
 348
 349                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 350         }
 351         if (info->alert_cmd) {
 352                 int pid = fork();
 353                 switch(pid) {
 354                 default:
 355                         waitpid(pid, NULL, 0);
 356                         break;
 357                 case -1:
 358                         break;
 359                 case 0:
 360                         execl(info->alert_cmd, info->alert_cmd,
 361                               event, dev, disc, NULL);
 362                         exit(2);
 363                 }
 364         }
 365         if (info->mailaddr &&
 366             (strncmp(event, "Fail", 4)==0 ||
 367              strncmp(event, "Test", 4)==0 ||
 368              strncmp(event, "Spares", 6)==0 ||
 369              strncmp(event, "Degrade", 7)==0)) {
 370                 FILE *mp = popen(Sendmail, "w");
 371                 if (mp) {
 372                         FILE *mdstat;
 373                         char hname[256];
 374                         gethostname(hname, sizeof(hname));
 375                         signal(SIGPIPE, SIG_IGN);
 376                         if (info->mailfrom)
 377                                 fprintf(mp, "From: %s\n", info->mailfrom);
 378                         else
 379                                 fprintf(mp, "From: %s monitoring <root>\n", Name);
 380                         fprintf(mp, "To: %s\n", info->mailaddr);
 381                         fprintf(mp, "Subject: %s event on %s:%s\n\n",
 382                                 event, dev, hname);
 383
 384                         fprintf(mp,
 385                                 "This is an automatically generated mail message from %s\n", Name);
 386                         fprintf(mp, "running on %s\n\n", hname);
 387
 388                         fprintf(mp,
 389                                 "A %s event had been detected on md device %s.\n\n", event, dev);
 390
 391                         if (disc && disc[0] != ' ')
 392                                 fprintf(mp,
 393                                         "It could be related to component device %s.\n\n", disc);
 394                         if (disc && disc[0] == ' ')
 395                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 396
 397                         fprintf(mp, "Faithfully yours, etc.\n");
 398
 399                         mdstat = fopen("/proc/mdstat", "r");
 400                         if (mdstat) {
 401                                 char buf[8192];
 402                                 int n;
 403                                 fprintf(mp,
 404                                         "\nP.S. The /proc/mdstat file currently contains the following:\n\n");
 405                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 406                                         n=fwrite(buf, 1, n, mp);
 407                                 fclose(mdstat);
 408                         }
 409                         pclose(mp);
 410                 }
 411         }
 412
 413         /* log the event to syslog maybe */
 414         if (info->dosyslog) {
 415                 /* Log at a different severity depending on the event.
 416                  *
 417                  * These are the critical events:  */
 418                 if (strncmp(event, "Fail", 4)==0 ||
 419                     strncmp(event, "Degrade", 7)==0 ||
 420                     strncmp(event, "DeviceDisappeared", 17)==0)
 421                         priority = LOG_CRIT;
 422                 /* Good to know about, but are not failures: */
 423                 else if (strncmp(event, "Rebuild", 7)==0 ||
 424                          strncmp(event, "MoveSpare", 9)==0 ||
 425                          strncmp(event, "Spares", 6) != 0)
 426                         priority = LOG_WARNING;
 427                 /* Everything else: */
 428                 else
 429                         priority = LOG_INFO;
 430
 431                 if (disc && disc[0] != ' ')
 432                         syslog(priority,
 433                                "%s event detected on md device %s, component device %s", event, dev, disc);
 434                 else if (disc)
 435                         syslog(priority,
 436                                "%s event detected on md device %s: %s",
 437                                event, dev, disc);
 438                 else
 439                         syslog(priority,
 440                                "%s event detected on md device %s",
 441                                event, dev);
 442         }
 443 }
 444
 445 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 446                        int test, struct alert_info *ainfo,
 447                        int increments, char *prefer)
 448 {
 449         /* Update the state 'st' to reflect any changes shown in mdstat,
 450          * or found by directly examining the array, and return
 451          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 452          */
 453         struct { int state, major, minor; } info[MAX_DISKS];
 454         mdu_array_info_t array;
 455         struct mdstat_ent *mse = NULL, *mse2;
 456         char *dev = st->devname;
 457         int fd;
 458         int i;
 459         int remaining_disks;
 460         int last_disk;
 461         int new_array = 0;
 462         int retval;
 463
 464         if (test)
 465                 alert("TestMessage", dev, NULL, ainfo);
 466
 467         retval = 0;
 468
 469         fd = open(dev, O_RDONLY);
 470         if (fd < 0) {
 471                 if (!st->err)
 472                         alert("DeviceDisappeared", dev, NULL, ainfo);
 473                 st->err++;
 474                 goto out;
 475         }
 476
 477         if (!md_array_active(fd)) {
 478                 close(fd);
 479                 if (!st->err)
 480                         alert("DeviceDisappeared", dev, NULL, ainfo);
 481                 st->err++;
 482                 goto out;
 483         }
 484
 485         fcntl(fd, F_SETFD, FD_CLOEXEC);
 486         if (md_get_array_info(fd, &array) < 0) {
 487                 if (!st->err)
 488                         alert("DeviceDisappeared", dev, NULL, ainfo);
 489                 st->err++;
 490                 close(fd);
 491                 goto out;
 492         }
 493         /* It's much easier to list what array levels can't
 494          * have a device disappear than all of them that can
 495          */
 496         if (array.level == 0 || array.level == -1) {
 497                 if (!st->err && !st->from_config)
 498                         alert("DeviceDisappeared", dev, " Wrong-Level", ainfo);
 499                 st->err++;
 500                 close(fd);
 501                 goto out;
 502         }
 503         if (st->devnm[0] == 0)
 504                 strcpy(st->devnm, fd2devnm(fd));
 505
 506         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 507                 if (strcmp(mse2->devnm, st->devnm) == 0) {
 508                         mse2->devnm[0] = 0; /* flag it as "used" */
 509                         mse = mse2;
 510                 }
 511
 512         if (!mse) {
 513                 /* duplicated array in statelist
 514                  * or re-created after reading mdstat*/
 515                 st->err++;
 516                 close(fd);
 517                 goto out;
 518         }
 519         /* this array is in /proc/mdstat */
 520         if (array.utime == 0)
 521                 /* external arrays don't update utime, so
 522                  * just make sure it is always different. */
 523                 array.utime = st->utime + 1;;
 524
 525         if (st->err) {
 526                 /* New array appeared where previously had an error */
 527                 st->err = 0;
 528                 st->percent = RESYNC_NONE;
 529                 new_array = 1;
 530                 alert("NewArray", st->devname, NULL, ainfo);
 531         }
 532
 533         if (st->utime == array.utime && st->failed == array.failed_disks &&
 534             st->working == array.working_disks &&
 535             st->spare == array.spare_disks &&
 536             (mse == NULL  || (mse->percent == st->percent))) {
 537                 close(fd);
 538                 if ((st->active < st->raid) && st->spare == 0)
 539                         retval = 1;
 540                 goto out;
 541         }
 542         if (st->utime == 0 && /* new array */
 543             mse->pattern && strchr(mse->pattern, '_') /* degraded */)
 544                 alert("DegradedArray", dev, NULL, ainfo);
 545
 546         if (st->utime == 0 && /* new array */
 547             st->expected_spares > 0 && array.spare_disks < st->expected_spares)
 548                 alert("SparesMissing", dev, NULL, ainfo);
 549         if (st->percent < 0 && st->percent != RESYNC_UNKNOWN &&
 550             mse->percent >= 0)
 551                 alert("RebuildStarted", dev, NULL, ainfo);
 552         if (st->percent >= 0 && mse->percent >= 0 &&
 553             (mse->percent / increments) > (st->percent / increments)) {
 554                 char percentalert[15];
 555                 /*
 556                  * "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 557                  */
 558
 559                 if((mse->percent / increments) == 0)
 560                         snprintf(percentalert, sizeof(percentalert),
 561                                  "RebuildStarted");
 562                 else
 563                         snprintf(percentalert, sizeof(percentalert),
 564                                  "Rebuild%02d", mse->percent);
 565
 566                 alert(percentalert, dev, NULL, ainfo);
 567         }
 568
 569         if (mse->percent == RESYNC_NONE && st->percent >= 0) {
 570                 /* Rebuild/sync/whatever just finished.
 571                  * If there is a number in /mismatch_cnt,
 572                  * we should report that.
 573                  */
 574                 struct mdinfo *sra =
 575                         sysfs_read(-1, st->devnm, GET_MISMATCH);
 576                 if (sra && sra->mismatch_cnt > 0) {
 577                         char cnt[80];
 578                         snprintf(cnt, sizeof(cnt),
 579                                  " mismatches found: %d (on raid level %d)",
 580                                 sra->mismatch_cnt, array.level);
 581                         alert("RebuildFinished", dev, cnt, ainfo);
 582                 } else
 583                         alert("RebuildFinished", dev, NULL, ainfo);
 584                 if (sra)
 585                         sysfs_free(sra);
 586         }
 587         st->percent = mse->percent;
 588
 589         remaining_disks = array.nr_disks;
 590         for (i = 0; i < MAX_DISKS && remaining_disks > 0; i++) {
 591                 mdu_disk_info_t disc;
 592                 disc.number = i;
 593                 if (md_get_disk_info(fd, &disc) >= 0) {
 594                         info[i].state = disc.state;
 595                         info[i].major = disc.major;
 596                         info[i].minor = disc.minor;
 597                         if (disc.major || disc.minor)
 598                                 remaining_disks --;
 599                 } else
 600                         info[i].major = info[i].minor = 0;
 601         }
 602         last_disk = i;
 603
 604         if (mse->metadata_version &&
 605             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 606             is_subarray(mse->metadata_version+9)) {
 607                 char *sl;
 608                 strcpy(st->parent_devnm, mse->metadata_version+10);
 609                 sl = strchr(st->parent_devnm, '/');
 610                 if (sl)
 611                         *sl = 0;
 612         } else
 613                 st->parent_devnm[0] = 0;
 614         if (st->metadata == NULL && st->parent_devnm[0] == 0)
 615                 st->metadata = super_by_fd(fd, NULL);
 616
 617         close(fd);
 618
 619         for (i=0; i<MAX_DISKS; i++) {
 620                 mdu_disk_info_t disc = {0,0,0,0,0};
 621                 int newstate=0;
 622                 int change;
 623                 char *dv = NULL;
 624                 disc.number = i;
 625                 if (i < last_disk && (info[i].major || info[i].minor)) {
 626                         newstate = info[i].state;
 627                         dv = map_dev_preferred(info[i].major, info[i].minor, 1,
 628                                                prefer);
 629                         disc.state = newstate;
 630                         disc.major = info[i].major;
 631                         disc.minor = info[i].minor;
 632                 } else
 633                         newstate = (1 << MD_DISK_REMOVED);
 634
 635                 if (dv == NULL && st->devid[i])
 636                         dv = map_dev_preferred(major(st->devid[i]),
 637                                                minor(st->devid[i]), 1, prefer);
 638                 change = newstate ^ st->devstate[i];
 639                 if (st->utime && change && !st->err && !new_array) {
 640                         if ((st->devstate[i]&change) & (1 << MD_DISK_SYNC))
 641                                 alert("Fail", dev, dv, ainfo);
 642                         else if ((newstate & (1 << MD_DISK_FAULTY)) &&
 643                                  (disc.major || disc.minor) &&
 644                                  st->devid[i] == makedev(disc.major,
 645                                                          disc.minor))
 646                                 alert("FailSpare", dev, dv, ainfo);
 647                         else if ((newstate&change) & (1 << MD_DISK_SYNC))
 648                                 alert("SpareActive", dev, dv, ainfo);
 649                 }
 650                 st->devstate[i] = newstate;
 651                 st->devid[i] = makedev(disc.major, disc.minor);
 652         }
 653         st->active = array.active_disks;
 654         st->working = array.working_disks;
 655         st->spare = array.spare_disks;
 656         st->failed = array.failed_disks;
 657         st->utime = array.utime;
 658         st->raid = array.raid_disks;
 659         st->err = 0;
 660         if ((st->active < st->raid) && st->spare == 0)
 661                 retval = 1;
 662
 663  out:
 664         return retval;
 665 }
 666
 667 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
 668                           int test, struct alert_info *info)
 669 {
 670         struct mdstat_ent *mse;
 671         int new_found = 0;
 672         char *name;
 673
 674         for (mse=mdstat; mse; mse=mse->next)
 675                 if (mse->devnm[0] &&
 676                     (!mse->level  || /* retrieve containers */
 677                      (strcmp(mse->level, "raid0") != 0 &&
 678                       strcmp(mse->level, "linear") != 0))
 679                         ) {
 680                         struct state *st = xcalloc(1, sizeof *st);
 681                         mdu_array_info_t array;
 682                         int fd;
 683
 684                         name = get_md_name(mse->devnm);
 685                         if (!name) {
 686                                 free(st);
 687                                 continue;
 688                         }
 689
 690                         st->devname = xstrdup(name);
 691                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 692                             md_get_array_info(fd, &array) < 0) {
 693                                 /* no such array */
 694                                 if (fd >= 0)
 695                                         close(fd);
 696                                 put_md_name(st->devname);
 697                                 free(st->devname);
 698                                 if (st->metadata) {
 699                                         st->metadata->ss->free_super(st->metadata);
 700                                         free(st->metadata);
 701                                 }
 702                                 free(st);
 703                                 continue;
 704                         }
 705                         close(fd);
 706                         st->next = *statelist;
 707                         st->err = 1;
 708                         st->from_auto = 1;
 709                         strcpy(st->devnm, mse->devnm);
 710                         st->percent = RESYNC_UNKNOWN;
 711                         st->expected_spares = -1;
 712                         if (mse->metadata_version &&
 713                             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 714                             is_subarray(mse->metadata_version+9)) {
 715                                 char *sl;
 716                                 strcpy(st->parent_devnm,
 717                                         mse->metadata_version+10);
 718                                 sl = strchr(st->parent_devnm, '/');
 719                                 *sl = 0;
 720                         } else
 721                                 st->parent_devnm[0] = 0;
 722                         *statelist = st;
 723                         if (test)
 724                                 alert("TestMessage", st->devname, NULL, info);
 725                         new_found = 1;
 726                 }
 727         return new_found;
 728 }
 729
 730 static int get_required_spare_criteria(struct state *st,
 731                                        struct spare_criteria *sc)
 732 {
 733         int fd;
 734
 735         if (!st->metadata ||
 736             !st->metadata->ss->get_spare_criteria) {
 737                 sc->min_size = 0;
 738                 sc->sector_size = 0;
 739                 return 0;
 740         }
 741
 742         fd = open(st->devname, O_RDONLY);
 743         if (fd < 0)
 744                 return 1;
 745         if (st->metadata->ss->external)
 746                 st->metadata->ss->load_container(st->metadata, fd, st->devname);
 747         else
 748                 st->metadata->ss->load_super(st->metadata, fd, st->devname);
 749         close(fd);
 750         if (!st->metadata->sb)
 751                 return 1;
 752
 753         st->metadata->ss->get_spare_criteria(st->metadata, sc);
 754         st->metadata->ss->free_super(st->metadata);
 755
 756         return 0;
 757 }
 758
 759 static int check_donor(struct state *from, struct state *to)
 760 {
 761         struct state *sub;
 762
 763         if (from == to)
 764                 return 0;
 765         if (from->parent)
 766                 /* Cannot move from a member */
 767                 return 0;
 768         if (from->err)
 769                 return 0;
 770         for (sub = from->subarray; sub; sub = sub->subarray)
 771                 /* If source array has degraded subarrays, don't
 772                  * remove anything
 773                  */
 774                 if (sub->active < sub->raid)
 775                         return 0;
 776         if (from->metadata->ss->external == 0)
 777                 if (from->active < from->raid)
 778                         return 0;
 779         if (from->spare <= 0)
 780                 return 0;
 781         return 1;
 782 }
 783
 784 static dev_t choose_spare(struct state *from, struct state *to,
 785                         struct domainlist *domlist, struct spare_criteria *sc)
 786 {
 787         int d;
 788         dev_t dev = 0;
 789
 790         for (d = from->raid; !dev && d < MAX_DISKS; d++) {
 791                 if (from->devid[d] > 0 &&
 792                     from->devstate[d] == 0) {
 793                         struct dev_policy *pol;
 794                         unsigned long long dev_size;
 795                         unsigned int dev_sector_size;
 796
 797                         if (to->metadata->ss->external &&
 798                             test_partition_from_id(from->devid[d]))
 799                                 continue;
 800
 801                         if (sc->min_size &&
 802                             dev_size_from_id(from->devid[d], &dev_size) &&
 803                             dev_size < sc->min_size)
 804                                 continue;
 805
 806                         if (sc->sector_size &&
 807                             dev_sector_size_from_id(from->devid[d],
 808                                                     &dev_sector_size) &&
 809                             sc->sector_size != dev_sector_size)
 810                                 continue;
 811
 812                         pol = devid_policy(from->devid[d]);
 813                         if (from->spare_group)
 814                                 pol_add(&pol, pol_domain,
 815                                         from->spare_group, NULL);
 816                         if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
 817                             dev = from->devid[d];
 818                         dev_policy_free(pol);
 819                 }
 820         }
 821         return dev;
 822 }
 823
 824 static dev_t container_choose_spare(struct state *from, struct state *to,
 825                                     struct domainlist *domlist,
 826                                     struct spare_criteria *sc, int active)
 827 {
 828         /* This is similar to choose_spare, but we cannot trust devstate,
 829          * so we need to read the metadata instead
 830          */
 831         struct mdinfo *list;
 832         struct supertype *st = from->metadata;
 833         int fd = open(from->devname, O_RDONLY);
 834         int err;
 835         dev_t dev = 0;
 836
 837         if (fd < 0)
 838                 return 0;
 839         if (!st->ss->getinfo_super_disks) {
 840                 close(fd);
 841                 return 0;
 842         }
 843
 844         err = st->ss->load_container(st, fd, NULL);
 845         close(fd);
 846         if (err)
 847                 return 0;
 848
 849         if (from == to) {
 850                 /* We must check if number of active disks has not increased
 851                  * since ioctl in main loop. mdmon may have added spare
 852                  * to subarray. If so we do not need to look for more spares
 853                  * so return non zero value */
 854                 int active_cnt = 0;
 855                 struct mdinfo *dp;
 856                 list = st->ss->getinfo_super_disks(st);
 857                 if (!list) {
 858                         st->ss->free_super(st);
 859                         return 1;
 860                 }
 861                 dp = list->devs;
 862                 while (dp) {
 863                         if (dp->disk.state & (1<<MD_DISK_SYNC) &&
 864                             !(dp->disk.state & (1<<MD_DISK_FAULTY)))
 865                                 active_cnt++;
 866                         dp = dp->next;
 867                 }
 868                 sysfs_free(list);
 869                 if (active < active_cnt) {
 870                         /* Spare just activated.*/
 871                         st->ss->free_super(st);
 872                         return 1;
 873                 }
 874         }
 875
 876         /* We only need one spare so full list not needed */
 877         list = container_choose_spares(st, sc, domlist, from->spare_group,
 878                                        to->metadata->ss->name, 1);
 879         if (list) {
 880                 struct mdinfo *disks = list->devs;
 881                 if (disks)
 882                         dev = makedev(disks->disk.major, disks->disk.minor);
 883                 sysfs_free(list);
 884         }
 885         st->ss->free_super(st);
 886         return dev;
 887 }
 888
 889 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 890 {
 891         struct state *from;
 892         struct state *st;
 893         struct spare_criteria sc;
 894
 895         link_containers_with_subarrays(statelist);
 896         for (st = statelist; st; st = st->next)
 897                 if (st->active < st->raid &&
 898                     st->spare == 0 && !st->err) {
 899                         struct domainlist *domlist = NULL;
 900                         int d;
 901                         struct state *to = st;
 902
 903                         if (to->parent_devnm[0] && !to->parent)
 904                                 /* subarray monitored without parent container
 905                                  * we can't move spares here */
 906                                 continue;
 907
 908                         if (to->parent)
 909                                 /* member of a container */
 910                                 to = to->parent;
 911
 912                         if (get_required_spare_criteria(to, &sc))
 913                                 continue;
 914                         if (to->metadata->ss->external) {
 915                                 /* We must make sure there is
 916                                  * no suitable spare in container already.
 917                                  * If there is we don't add more */
 918                                 dev_t devid = container_choose_spare(
 919                                         to, to, NULL, &sc, st->active);
 920                                 if (devid > 0)
 921                                         continue;
 922                         }
 923                         for (d = 0; d < MAX_DISKS; d++)
 924                                 if (to->devid[d])
 925                                         domainlist_add_dev(&domlist,
 926                                                            to->devid[d],
 927                                                            to->metadata->ss->name);
 928                         if (to->spare_group)
 929                                 domain_add(&domlist, to->spare_group);
 930                         /*
 931                          * No spare migration if the destination
 932                          * has no domain. Skip this array.
 933                          */
 934                         if (!domlist)
 935                                 continue;
 936                         for (from=statelist ; from ; from=from->next) {
 937                                 dev_t devid;
 938                                 if (!check_donor(from, to))
 939                                         continue;
 940                                 if (from->metadata->ss->external)
 941                                         devid = container_choose_spare(
 942                                                 from, to, domlist, &sc, 0);
 943                                 else
 944                                         devid = choose_spare(from, to, domlist,
 945                                                              &sc);
 946                                 if (devid > 0
 947                                     && move_spare(from->devname, to->devname, devid)) {
 948                                         alert("MoveSpare", to->devname, from->devname, info);
 949                                         break;
 950                                 }
 951                         }
 952                         domain_free(domlist);
 953                 }
 954 }
 955
 956 /* search the statelist to connect external
 957  * metadata subarrays with their containers
 958  * We always completely rebuild the tree from scratch as
 959  * that is safest considering the possibility of entries
 960  * disappearing or changing.
 961  */
 962 static void link_containers_with_subarrays(struct state *list)
 963 {
 964         struct state *st;
 965         struct state *cont;
 966         for (st = list; st; st = st->next) {
 967                 st->parent = NULL;
 968                 st->subarray = NULL;
 969         }
 970         for (st = list; st; st = st->next)
 971                 if (st->parent_devnm[0])
 972                         for (cont = list; cont; cont = cont->next)
 973                                 if (!cont->err &&
 974                                     cont->parent_devnm[0] == 0 &&
 975                                     strcmp(cont->devnm, st->parent_devnm) == 0) {
 976                                         st->parent = cont;
 977                                         st->subarray = cont->subarray;
 978                                         cont->subarray = st;
 979                                         break;
 980                                 }
 981 }
 982
 983 /* Not really Monitor but ... */
 984 int Wait(char *dev)
 985 {
 986         char devnm[32];
 987         int rv = 1;
 988         int frozen_remaining = 3;
 989
 990         if (!stat_is_blkdev(dev, NULL))
 991                 return 2;
 992         strcpy(devnm, dev);
 993
 994         while(1) {
 995                 struct mdstat_ent *ms = mdstat_read(1, 0);
 996                 struct mdstat_ent *e;
 997
 998                 for (e=ms ; e; e=e->next)
 999                         if (strcmp(e->devnm, devnm) == 0)
1000                                 break;
1001
1002                 if (e && e->percent == RESYNC_NONE) {
1003                         /* We could be in the brief pause before something
1004                          * starts. /proc/mdstat doesn't show that, but
1005                          * sync_action does.
1006                          */
1007                         struct mdinfo mdi;
1008                         char buf[21];
1009
1010                         if (sysfs_init(&mdi, -1, devnm))
1011                                 return 2;
1012                         if (sysfs_get_str(&mdi, NULL, "sync_action",
1013                                           buf, 20) > 0 &&
1014                             strcmp(buf,"idle\n") != 0) {
1015                                 e->percent = RESYNC_UNKNOWN;
1016                                 if (strcmp(buf, "frozen\n") == 0) {
1017                                         if (frozen_remaining == 0)
1018                                                 e->percent = RESYNC_NONE;
1019                                         else
1020                                                 frozen_remaining -= 1;
1021                                 }
1022                         }
1023                 }
1024                 if (!e || e->percent == RESYNC_NONE) {
1025                         if (e && e->metadata_version &&
1026                             strncmp(e->metadata_version, "external:", 9) == 0) {
1027                                 if (is_subarray(&e->metadata_version[9]))
1028                                         ping_monitor(&e->metadata_version[9]);
1029                                 else
1030                                         ping_monitor(devnm);
1031                         }
1032                         free_mdstat(ms);
1033                         return rv;
1034                 }
1035                 free_mdstat(ms);
1036                 rv = 0;
1037                 mdstat_wait(5);
1038         }
1039 }
1040
1041 static char *clean_states[] = {
1042         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1043
1044 int WaitClean(char *dev, int sock, int verbose)
1045 {
1046         int fd;
1047         struct mdinfo *mdi;
1048         int rv = 1;
1049         char devnm[32];
1050
1051         if (!stat_is_blkdev(dev, NULL))
1052                 return 2;
1053         fd = open(dev, O_RDONLY);
1054         if (fd < 0) {
1055                 if (verbose)
1056                         pr_err("Couldn't open %s: %s\n", dev, strerror(errno));
1057                 return 1;
1058         }
1059
1060         strcpy(devnm, fd2devnm(fd));
1061         mdi = sysfs_read(fd, devnm, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1062         if (!mdi) {
1063                 if (verbose)
1064                         pr_err("Failed to read sysfs attributes for %s\n", dev);
1065                 close(fd);
1066                 return 0;
1067         }
1068
1069         switch(mdi->array.level) {
1070         case LEVEL_LINEAR:
1071         case LEVEL_MULTIPATH:
1072         case 0:
1073                 /* safemode delay is irrelevant for these levels */
1074                 rv = 0;
1075         }
1076
1077         /* for internal metadata the kernel handles the final clean
1078          * transition, containers can never be dirty
1079          */
1080         if (!is_subarray(mdi->text_version))
1081                 rv = 0;
1082
1083         /* safemode disabled ? */
1084         if (mdi->safe_mode_delay == 0)
1085                 rv = 0;
1086
1087         if (rv) {
1088                 int state_fd = sysfs_open(fd2devnm(fd), NULL, "array_state");
1089                 char buf[20];
1090                 int delay = 5000;
1091
1092                 /* minimize the safe_mode_delay and prepare to wait up to 5s
1093                  * for writes to quiesce
1094                  */
1095                 sysfs_set_safemode(mdi, 1);
1096
1097                 /* wait for array_state to be clean */
1098                 while (1) {
1099                         rv = read(state_fd, buf, sizeof(buf));
1100                         if (rv < 0)
1101                                 break;
1102                         if (sysfs_match_word(buf, clean_states) <= 4)
1103                                 break;
1104                         rv = sysfs_wait(state_fd, &delay);
1105                         if (rv < 0 && errno != EINTR)
1106                                 break;
1107                         lseek(state_fd, 0, SEEK_SET);
1108                 }
1109                 if (rv < 0)
1110                         rv = 1;
1111                 else if (fping_monitor(sock) == 0 ||
1112                          ping_monitor(mdi->text_version) == 0) {
1113                         /* we need to ping to close the window between array
1114                          * state transitioning to clean and the metadata being
1115                          * marked clean
1116                          */
1117                         rv = 0;
1118                 } else
1119                         rv = 1;
1120                 if (rv && verbose)
1121                         pr_err("Error waiting for %s to be clean\n",
1122                                 dev);
1123
1124                 /* restore the original safe_mode_delay */
1125                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1126                 close(state_fd);
1127         }
1128
1129         sysfs_free(mdi);
1130         close(fd);
1131
1132         return rv;
1133 }