Monitor.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2009 Neil Brown <neilb@suse.de>
   5  *
   6  *
   7  *    This program is free software; you can redistribute it and/or modify
   8  *    it under the terms of the GNU General Public License as published by
   9  *    the Free Software Foundation; either version 2 of the License, or
  10  *    (at your option) any later version.
  11  *
  12  *    This program is distributed in the hope that it will be useful,
  13  *    but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *    GNU General Public License for more details.
  16  *
  17  *    You should have received a copy of the GNU General Public License
  18  *    along with this program; if not, write to the Free Software
  19  *    Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  *    Author: Neil Brown
  22  *    Email: <neilb@suse.de>
  23  */
  24
  25 #include        "mdadm.h"
  26 #include        "md_p.h"
  27 #include        "md_u.h"
  28 #include        <sys/wait.h>
  29 #include        <signal.h>
  30 #include        <limits.h>
  31 #include        <syslog.h>
  32
  33 /* The largest number of disks current arrays can manage is 384
  34  * This really should be dynamically, but that will have to wait
  35  * At least it isn't MD_SB_DISKS.
  36  */
  37 #define MaxDisks 384
  38 struct state {
  39         char *devname;
  40         int devnum;     /* to sync with mdstat info */
  41         long utime;
  42         int err;
  43         char *spare_group;
  44         int active, working, failed, spare, raid;
  45         int expected_spares;
  46         int devstate[MaxDisks];
  47         dev_t devid[MaxDisks];
  48         int percent;
  49         int parent_dev; /* For subarray, devnum of parent.
  50                          * For others, NoMdDev
  51                          */
  52         struct supertype *metadata;
  53         struct state *subarray;/* for a container it is a link to first subarray
  54                                 * for a subarray it is a link to next subarray
  55                                 * in the same container */
  56         struct state *parent;  /* for a subarray it is a link to its container
  57                                 */
  58         struct state *next;
  59 };
  60
  61 struct alert_info {
  62         char *mailaddr;
  63         char *mailfrom;
  64         char *alert_cmd;
  65         int dosyslog;
  66 };
  67 static int make_daemon(char *pidfile);
  68 static int check_one_sharer(int scan);
  69 static void alert(char *event, char *dev, char *disc, struct alert_info *info);
  70 static int check_array(struct state *st, struct mdstat_ent *mdstat,
  71                        int test, struct alert_info *info,
  72                        int increments);
  73 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
  74                           int test, struct alert_info *info);
  75 static void try_spare_migration(struct state *statelist, struct alert_info *info);
  76 static void link_containers_with_subarrays(struct state *list);
  77
  78 int Monitor(struct mddev_dev *devlist,
  79             char *mailaddr, char *alert_cmd,
  80             int period, int daemonise, int scan, int oneshot,
  81             int dosyslog, int test, char *pidfile, int increments,
  82             int share)
  83 {
  84         /*
  85          * Every few seconds, scan every md device looking for changes
  86          * When a change is found, log it, possibly run the alert command,
  87          * and possibly send Email
  88          *
  89          * For each array, we record:
  90          *   Update time
  91          *   active/working/failed/spare drives
  92          *   State of each device.
  93          *   %rebuilt if rebuilding
  94          *
  95          * If the update time changes, check out all the data again
  96          * It is possible that we cannot get the state of each device
  97          * due to bugs in the md kernel module.
  98          * We also read /proc/mdstat to get rebuild percent,
  99          * and to get state on all active devices incase of kernel bug.
 100          *
 101          * Events are:
 102          *    Fail
 103          *      An active device had Faulty set or Active/Sync removed
 104          *    FailSpare
 105          *      A spare device had Faulty set
 106          *    SpareActive
 107          *      An active device had a reverse transition
 108          *    RebuildStarted
 109          *      percent went from -1 to +ve
 110          *    RebuildNN
 111          *      percent went from below to not-below NN%
 112          *    DeviceDisappeared
 113          *      Couldn't access a device which was previously visible
 114          *
 115          * if we detect an array with active<raid and spare==0
 116          * we look at other arrays that have same spare-group
 117          * If we find one with active==raid and spare>0,
 118          *  and if we can get_disk_info and find a name
 119          *  Then we hot-remove and hot-add to the other array
 120          *
 121          * If devlist is NULL, then we can monitor everything because --scan
 122          * was given.  We get an initial list from config file and add anything
 123          * that appears in /proc/mdstat
 124          */
 125
 126         struct state *statelist = NULL;
 127         struct state *st2;
 128         int finished = 0;
 129         struct mdstat_ent *mdstat = NULL;
 130         char *mailfrom = NULL;
 131         struct alert_info info;
 132
 133         if (!mailaddr) {
 134                 mailaddr = conf_get_mailaddr();
 135                 if (mailaddr && ! scan)
 136                         fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n",
 137                                mailaddr);
 138         }
 139         mailfrom = conf_get_mailfrom();
 140
 141         if (!alert_cmd) {
 142                 alert_cmd = conf_get_program();
 143                 if (alert_cmd && ! scan)
 144                         fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n",
 145                                alert_cmd);
 146         }
 147         if (scan && !mailaddr && !alert_cmd && !dosyslog) {
 148                 fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n");
 149                 return 1;
 150         }
 151         info.alert_cmd = alert_cmd;
 152         info.mailaddr = mailaddr;
 153         info.mailfrom = mailfrom;
 154         info.dosyslog = dosyslog;
 155
 156         if (daemonise) {
 157                 int rv = make_daemon(pidfile);
 158                 if (rv >= 0)
 159                         return rv;
 160         }
 161
 162         if (share)
 163                 if (check_one_sharer(scan))
 164                         return 1;
 165
 166         if (devlist == NULL) {
 167                 struct mddev_ident *mdlist = conf_get_ident(NULL);
 168                 for (; mdlist; mdlist=mdlist->next) {
 169                         struct state *st;
 170                         if (mdlist->devname == NULL)
 171                                 continue;
 172                         if (strcasecmp(mdlist->devname, "<ignore>") == 0)
 173                                 continue;
 174                         st = calloc(1, sizeof *st);
 175                         if (st == NULL)
 176                                 continue;
 177                         if (mdlist->devname[0] == '/')
 178                                 st->devname = strdup(mdlist->devname);
 179                         else {
 180                                 st->devname = malloc(8+strlen(mdlist->devname)+1);
 181                                 strcpy(strcpy(st->devname, "/dev/md/"),
 182                                        mdlist->devname);
 183                         }
 184                         st->next = statelist;
 185                         st->devnum = INT_MAX;
 186                         st->percent = -2;
 187                         st->expected_spares = mdlist->spare_disks;
 188                         if (mdlist->spare_group)
 189                                 st->spare_group = strdup(mdlist->spare_group);
 190                         statelist = st;
 191                 }
 192         } else {
 193                 struct mddev_dev *dv;
 194                 for (dv=devlist ; dv; dv=dv->next) {
 195                         struct mddev_ident *mdlist = conf_get_ident(dv->devname);
 196                         struct state *st = calloc(1, sizeof *st);
 197                         if (st == NULL)
 198                                 continue;
 199                         st->devname = strdup(dv->devname);
 200                         st->next = statelist;
 201                         st->devnum = INT_MAX;
 202                         st->percent = -2;
 203                         st->expected_spares = -1;
 204                         if (mdlist) {
 205                                 st->expected_spares = mdlist->spare_disks;
 206                                 if (mdlist->spare_group)
 207                                         st->spare_group = strdup(mdlist->spare_group);
 208                         }
 209                         statelist = st;
 210                 }
 211         }
 212
 213
 214         while (! finished) {
 215                 int new_found = 0;
 216                 struct state *st;
 217                 int anydegraded = 0;
 218
 219                 if (mdstat)
 220                         free_mdstat(mdstat);
 221                 mdstat = mdstat_read(oneshot?0:1, 0);
 222
 223                 for (st=statelist; st; st=st->next)
 224                         if (check_array(st, mdstat, test, &info, increments))
 225                                 anydegraded = 1;
 226
 227                 /* now check if there are any new devices found in mdstat */
 228                 if (scan)
 229                         new_found = add_new_arrays(mdstat, &statelist, test,
 230                                                    &info);
 231
 232                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 233                  * Look for another array with spare > 0 and active == raid and same spare_group
 234                  *  if found, choose a device and hotremove/hotadd
 235                  */
 236                 if (share && anydegraded)
 237                         try_spare_migration(statelist, &info);
 238                 if (!new_found) {
 239                         if (oneshot)
 240                                 break;
 241                         else
 242                                 mdstat_wait(period);
 243                 }
 244                 test = 0;
 245         }
 246         for (st2 = statelist; st2; st2 = statelist) {
 247                 statelist = st2->next;
 248                 free(st2);
 249         }
 250
 251         if (pidfile)
 252                 unlink(pidfile);
 253         return 0;
 254 }
 255
 256 static int make_daemon(char *pidfile)
 257 {
 258         /* Return:
 259          * -1 in the forked daemon
 260          *  0 in the parent
 261          *  1 on error
 262          * so a none-negative becomes the exit code.
 263          */
 264         int pid = fork();
 265         if (pid > 0) {
 266                 if (!pidfile)
 267                         printf("%d\n", pid);
 268                 else {
 269                         FILE *pid_file;
 270                         pid_file=fopen(pidfile, "w");
 271                         if (!pid_file)
 272                                 perror("cannot create pid file");
 273                         else {
 274                                 fprintf(pid_file,"%d\n", pid);
 275                                 fclose(pid_file);
 276                         }
 277                 }
 278                 return 0;
 279         }
 280         if (pid < 0) {
 281                 perror("daemonise");
 282                 return 1;
 283         }
 284         close(0);
 285         open("/dev/null", O_RDWR);
 286         dup2(0,1);
 287         dup2(0,2);
 288         setsid();
 289         return -1;
 290 }
 291
 292 static int check_one_sharer(int scan)
 293 {
 294         int pid, rv;
 295         FILE *fp;
 296         char dir[20];
 297         char path[100];
 298         struct stat buf;
 299         sprintf(path, "%s/autorebuild.pid", MDMON_DIR);
 300         fp = fopen(path, "r");
 301         if (fp) {
 302                 if (fscanf(fp, "%d", &pid) != 1)
 303                         pid = -1;
 304                 sprintf(dir, "/proc/%d", pid);
 305                 rv = stat(dir, &buf);
 306                 if (rv != -1) {
 307                         if (scan) {
 308                                 fprintf(stderr, Name ": Only one "
 309                                         "autorebuild process allowed"
 310                                         " in scan mode, aborting\n");
 311                                 fclose(fp);
 312                                 return 1;
 313                         } else {
 314                                 fprintf(stderr, Name ": Warning: One"
 315                                         " autorebuild process already"
 316                                         " running.\n");
 317                         }
 318                 }
 319                 fclose(fp);
 320         }
 321         if (scan) {
 322                 if (mkdir(MDMON_DIR, S_IRWXU) < 0 &&
 323                     errno != EEXIST) {
 324                         fprintf(stderr, Name ": Can't create "
 325                                 "autorebuild.pid file\n");
 326                 } else {
 327                         fp = fopen(path, "w");
 328                         if (!fp)
 329                                 fprintf(stderr, Name ": Cannot create"
 330                                         " autorebuild.pid"
 331                                         "file\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: " Name " monitoring <root>\n");
 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"
 386                                 " mail message from " Name "\n");
 387                         fprintf(mp, "running on %s\n\n", hname);
 388
 389                         fprintf(mp,
 390                                 "A %s event had been detected on"
 391                                 " md device %s.\n\n", event, dev);
 392
 393                         if (disc && disc[0] != ' ')
 394                                 fprintf(mp,
 395                                         "It could be related to"
 396                                         " component device %s.\n\n", disc);
 397                         if (disc && disc[0] == ' ')
 398                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 399
 400                         fprintf(mp, "Faithfully yours, etc.\n");
 401
 402                         mdstat = fopen("/proc/mdstat", "r");
 403                         if (mdstat) {
 404                                 char buf[8192];
 405                                 int n;
 406                                 fprintf(mp,
 407                                         "\nP.S. The /proc/mdstat file"
 408                                         " currently contains the following:\n\n");
 409                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 410                                         n=fwrite(buf, 1, n, mp);
 411                                 fclose(mdstat);
 412                         }
 413                         pclose(mp);
 414                 }
 415         }
 416
 417         /* log the event to syslog maybe */
 418         if (info->dosyslog) {
 419                 /* Log at a different severity depending on the event.
 420                  *
 421                  * These are the critical events:  */
 422                 if (strncmp(event, "Fail", 4)==0 ||
 423                     strncmp(event, "Degrade", 7)==0 ||
 424                     strncmp(event, "DeviceDisappeared", 17)==0)
 425                         priority = LOG_CRIT;
 426                 /* Good to know about, but are not failures: */
 427                 else if (strncmp(event, "Rebuild", 7)==0 ||
 428                          strncmp(event, "MoveSpare", 9)==0 ||
 429                          strncmp(event, "Spares", 6) != 0)
 430                         priority = LOG_WARNING;
 431                 /* Everything else: */
 432                 else
 433                         priority = LOG_INFO;
 434
 435                 if (disc)
 436                         syslog(priority,
 437                                "%s event detected on md device %s,"
 438                                " component device %s", event, dev, disc);
 439                 else
 440                         syslog(priority,
 441                                "%s event detected on md device %s",
 442                                event, dev);
 443         }
 444 }
 445
 446 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 447                        int test, struct alert_info *ainfo,
 448                        int increments)
 449 {
 450         /* Update the state 'st' to reflect any changes shown in mdstat,
 451          * or found by directly examining the array, and return
 452          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 453          */
 454         struct { int state, major, minor; } info[MaxDisks];
 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
 463         if (test)
 464                 alert("TestMessage", dev, NULL, ainfo);
 465         fd = open(dev, O_RDONLY);
 466         if (fd < 0) {
 467                 if (!st->err)
 468                         alert("DeviceDisappeared", dev, NULL, ainfo);
 469                 st->err=1;
 470                 return 0;
 471         }
 472         fcntl(fd, F_SETFD, FD_CLOEXEC);
 473         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 474                 if (!st->err)
 475                         alert("DeviceDisappeared", dev, NULL, ainfo);
 476                 st->err=1;
 477                 close(fd);
 478                 return 0;
 479         }
 480         /* It's much easier to list what array levels can't
 481          * have a device disappear than all of them that can
 482          */
 483         if (array.level == 0 || array.level == -1) {
 484                 if (!st->err)
 485                         alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
 486                 st->err = 1;
 487                 close(fd);
 488                 return 0;
 489         }
 490         if (st->devnum == INT_MAX) {
 491                 struct stat stb;
 492                 if (fstat(fd, &stb) == 0 &&
 493                     (S_IFMT&stb.st_mode)==S_IFBLK) {
 494                         if (major(stb.st_rdev) == MD_MAJOR)
 495                                 st->devnum = minor(stb.st_rdev);
 496                         else
 497                                 st->devnum = -1- (minor(stb.st_rdev)>>6);
 498                 }
 499         }
 500
 501         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 502                 if (mse2->devnum == st->devnum) {
 503                         mse2->devnum = INT_MAX; /* flag it as "used" */
 504                         mse = mse2;
 505                 }
 506
 507         if (!mse) {
 508                 /* duplicated array in statelist
 509                  * or re-created after reading mdstat*/
 510                 st->err = 1;
 511                 close(fd);
 512                 return 0;
 513         }
 514         /* this array is in /proc/mdstat */
 515         if (array.utime == 0)
 516                 /* external arrays don't update utime, so
 517                  * just make sure it is always different. */
 518                 array.utime = st->utime + 1;;
 519
 520         if (st->utime == array.utime &&
 521             st->failed == array.failed_disks &&
 522             st->working == array.working_disks &&
 523             st->spare == array.spare_disks &&
 524             (mse == NULL  || (
 525                     mse->percent == st->percent
 526                     ))) {
 527                 close(fd);
 528                 st->err = 0;
 529                 if ((st->active < st->raid) && st->spare == 0)
 530                         return 1;
 531                 else
 532                         return 0;
 533         }
 534         if (st->utime == 0 && /* new array */
 535             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 536                 )
 537                 alert("DegradedArray", dev, NULL, ainfo);
 538
 539         if (st->utime == 0 && /* new array */
 540             st->expected_spares > 0 &&
 541             array.spare_disks < st->expected_spares)
 542                 alert("SparesMissing", dev, NULL, ainfo);
 543         if (st->percent == -1 &&
 544             mse->percent >= 0)
 545                 alert("RebuildStarted", dev, NULL, ainfo);
 546         if (st->percent >= 0 &&
 547             mse->percent >= 0 &&
 548             (mse->percent / increments) > (st->percent / increments)) {
 549                 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 550
 551                 if((mse->percent / increments) == 0)
 552                         snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
 553                 else
 554                         snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
 555
 556                 alert(percentalert, dev, NULL, ainfo);
 557         }
 558
 559         if (mse->percent == -1 &&
 560             st->percent >= 0) {
 561                 /* Rebuild/sync/whatever just finished.
 562                  * If there is a number in /mismatch_cnt,
 563                  * we should report that.
 564                  */
 565                 struct mdinfo *sra =
 566                         sysfs_read(-1, st->devnum, GET_MISMATCH);
 567                 if (sra && sra->mismatch_cnt > 0) {
 568                         char cnt[80];
 569                         snprintf(cnt, sizeof(cnt),
 570                                  " mismatches found: %d (on raid level %d)",
 571                                 sra->mismatch_cnt, array.level);
 572                         alert("RebuildFinished", dev, cnt, ainfo);
 573                 } else
 574                         alert("RebuildFinished", dev, NULL, ainfo);
 575                 if (sra)
 576                         free(sra);
 577         }
 578         st->percent = mse->percent;
 579
 580         remaining_disks = array.nr_disks;
 581         for (i=0; i<MaxDisks && remaining_disks > 0;
 582              i++) {
 583                 mdu_disk_info_t disc;
 584                 disc.number = i;
 585                 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
 586                         info[i].state = disc.state;
 587                         info[i].major = disc.major;
 588                         info[i].minor = disc.minor;
 589                         if (disc.major || disc.minor)
 590                                 remaining_disks --;
 591                 } else
 592                         info[i].major = info[i].minor = 0;
 593         }
 594         last_disk = i;
 595
 596         if (mse->metadata_version &&
 597             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 598             is_subarray(mse->metadata_version+9))
 599                 st->parent_dev =
 600                         devname2devnum(mse->metadata_version+10);
 601         else
 602                 st->parent_dev = NoMdDev;
 603         if (st->metadata == NULL &&
 604             st->parent_dev == NoMdDev)
 605                 st->metadata = super_by_fd(fd, NULL);
 606
 607         close(fd);
 608
 609         for (i=0; i<MaxDisks; i++) {
 610                 mdu_disk_info_t disc = {0,0,0,0,0};
 611                 int newstate=0;
 612                 int change;
 613                 char *dv = NULL;
 614                 disc.number = i;
 615                 if (i >= last_disk) {
 616                         newstate = 0;
 617                         disc.major = disc.minor = 0;
 618                 } else if (info[i].major || info[i].minor) {
 619                         newstate = info[i].state;
 620                         dv = map_dev(info[i].major, info[i].minor, 1);
 621                         disc.state = newstate;
 622                         disc.major = info[i].major;
 623                         disc.minor = info[i].minor;
 624                 } else if (mse &&  mse->pattern && i < (int)strlen(mse->pattern)) {
 625                         switch(mse->pattern[i]) {
 626                         case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
 627                         case '_': newstate = 8 /* REMOVED */; break;
 628                         }
 629                         disc.major = disc.minor = 0;
 630                 }
 631                 if (dv == NULL && st->devid[i])
 632                         dv = map_dev(major(st->devid[i]),
 633                                      minor(st->devid[i]), 1);
 634                 change = newstate ^ st->devstate[i];
 635                 if (st->utime && change && !st->err) {
 636                         if (i < array.raid_disks &&
 637                             (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
 638                              ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
 639                              ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
 640                                 )
 641                                 alert("Fail", dev, dv, ainfo);
 642                         else if (i >= array.raid_disks &&
 643                                  (disc.major || disc.minor) &&
 644                                  st->devid[i] == makedev(disc.major, disc.minor) &&
 645                                  ((newstate&change)&(1<<MD_DISK_FAULTY))
 646                                 )
 647                                 alert("FailSpare", dev, dv, ainfo);
 648                         else if (i < array.raid_disks &&
 649                                  ! (newstate & (1<<MD_DISK_REMOVED)) &&
 650                                  (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
 651                                   ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
 652                                   ((newstate&change)&(1<<MD_DISK_SYNC)))
 653                                 )
 654                                 alert("SpareActive", dev, dv, ainfo);
 655                 }
 656                 st->devstate[i] = newstate;
 657                 st->devid[i] = makedev(disc.major, disc.minor);
 658         }
 659         st->active = array.active_disks;
 660         st->working = array.working_disks;
 661         st->spare = array.spare_disks;
 662         st->failed = array.failed_disks;
 663         st->utime = array.utime;
 664         st->raid = array.raid_disks;
 665         st->err = 0;
 666         if ((st->active < st->raid) && st->spare == 0)
 667                 return 1;
 668         return 0;
 669 }
 670
 671 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
 672                           int test, struct alert_info *info)
 673 {
 674         struct mdstat_ent *mse;
 675         int new_found = 0;
 676
 677         for (mse=mdstat; mse; mse=mse->next)
 678                 if (mse->devnum != INT_MAX &&
 679                     (!mse->level  || /* retrieve containers */
 680                      (strcmp(mse->level, "raid0") != 0 &&
 681                       strcmp(mse->level, "linear") != 0))
 682                         ) {
 683                         struct state *st = calloc(1, sizeof *st);
 684                         mdu_array_info_t array;
 685                         int fd;
 686                         if (st == NULL)
 687                                 continue;
 688                         st->devname = strdup(get_md_name(mse->devnum));
 689                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 690                             ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
 691                                 /* no such array */
 692                                 if (fd >=0) close(fd);
 693                                 put_md_name(st->devname);
 694                                 free(st->devname);
 695                                 if (st->metadata) {
 696                                         st->metadata->ss->free_super(st->metadata);
 697                                         free(st->metadata);
 698                                 }
 699                                 free(st);
 700                                 continue;
 701                         }
 702                         close(fd);
 703                         st->next = *statelist;
 704                         st->err = 1;
 705                         st->devnum = mse->devnum;
 706                         st->percent = -2;
 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                                 st->parent_dev =
 712                                         devname2devnum(mse->metadata_version+10);
 713                         else
 714                                 st->parent_dev = NoMdDev;
 715                         *statelist = st;
 716                         if (test)
 717                                 alert("TestMessage", st->devname, NULL, info);
 718                         alert("NewArray", st->devname, NULL, info);
 719                         new_found = 1;
 720                 }
 721         return new_found;
 722 }
 723
 724 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
 725 {
 726         int fd;
 727
 728         if (!st->metadata ||
 729             !st->metadata->ss->min_acceptable_spare_size) {
 730                 *sizep = 0;
 731                 return 0;
 732         }
 733
 734         fd = open(st->devname, O_RDONLY);
 735         if (fd < 0)
 736                 return 1;
 737         if (st->metadata->ss->external)
 738                 st->metadata->ss->load_container(st->metadata, fd, st->devname);
 739         else
 740                 st->metadata->ss->load_super(st->metadata, fd, st->devname);
 741         close(fd);
 742         if (!st->metadata->sb)
 743                 return 1;
 744         *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
 745         st->metadata->ss->free_super(st->metadata);
 746
 747         return 0;
 748 }
 749
 750 static int check_donor(struct state *from, struct state *to)
 751 {
 752         struct state *sub;
 753
 754         if (from == to)
 755                 return 0;
 756         if (from->parent)
 757                 /* Cannot move from a member */
 758                 return 0;
 759         if (from->err)
 760                 return 0;
 761         for (sub = from->subarray; sub; sub = sub->subarray)
 762                 /* If source array has degraded subarrays, don't
 763                  * remove anything
 764                  */
 765                 if (sub->active < sub->raid)
 766                         return 0;
 767         if (from->metadata->ss->external == 0)
 768                 if (from->active < from->raid)
 769                         return 0;
 770         if (from->spare <= 0)
 771                 return 0;
 772         return 1;
 773 }
 774
 775 static dev_t choose_spare(struct state *from, struct state *to,
 776                         struct domainlist *domlist, unsigned long long min_size)
 777 {
 778         int d;
 779         dev_t dev = 0;
 780
 781         for (d = from->raid; !dev && d < MaxDisks; d++) {
 782                 if (from->devid[d] > 0 &&
 783                     from->devstate[d] == 0) {
 784                         struct dev_policy *pol;
 785                         unsigned long long dev_size;
 786
 787                         if (to->metadata->ss->external &&
 788                             test_partition_from_id(from->devid[d]))
 789                                 continue;
 790
 791                         if (min_size &&
 792                             dev_size_from_id(from->devid[d], &dev_size) &&
 793                             dev_size < min_size)
 794                                 continue;
 795
 796                         pol = devnum_policy(from->devid[d]);
 797                         if (from->spare_group)
 798                                 pol_add(&pol, pol_domain,
 799                                         from->spare_group, NULL);
 800                         if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
 801                             dev = from->devid[d];
 802                         dev_policy_free(pol);
 803                 }
 804         }
 805         return dev;
 806 }
 807
 808 static dev_t container_choose_spare(struct state *from, struct state *to,
 809                                     struct domainlist *domlist,
 810                                     unsigned long long min_size, int active)
 811 {
 812         /* This is similar to choose_spare, but we cannot trust devstate,
 813          * so we need to read the metadata instead
 814          */
 815         struct mdinfo *list;
 816         struct supertype *st = from->metadata;
 817         int fd = open(from->devname, O_RDONLY);
 818         int err;
 819         dev_t dev = 0;
 820
 821         if (fd < 0)
 822                 return 0;
 823         if (!st->ss->getinfo_super_disks) {
 824                 close(fd);
 825                 return 0;
 826         }
 827
 828         err = st->ss->load_container(st, fd, NULL);
 829         close(fd);
 830         if (err)
 831                 return 0;
 832
 833         if (from == to) {
 834                 /* We must check if number of active disks has not increased
 835                  * since ioctl in main loop. mdmon may have added spare
 836                  * to subarray. If so we do not need to look for more spares
 837                  * so return non zero value */
 838                 int active_cnt = 0;
 839                 struct mdinfo *dp;
 840                 list = st->ss->getinfo_super_disks(st);
 841                 if (!list) {
 842                         st->ss->free_super(st);
 843                         return 1;
 844                 }
 845                 dp = list->devs;
 846                 while (dp) {
 847                         if (dp->disk.state & (1<<MD_DISK_SYNC) &&
 848                             !(dp->disk.state & (1<<MD_DISK_FAULTY)))
 849                                 active_cnt++;
 850                         dp = dp->next;
 851                 }
 852                 sysfs_free(list);
 853                 if (active < active_cnt) {
 854                         /* Spare just activated.*/
 855                         st->ss->free_super(st);
 856                         return 1;
 857                 }
 858         }
 859
 860         /* We only need one spare so full list not needed */
 861         list = container_choose_spares(st, min_size, domlist, from->spare_group,
 862                                        to->metadata->ss->name, 1);
 863         if (list) {
 864                 struct mdinfo *disks = list->devs;
 865                 if (disks)
 866                         dev = makedev(disks->disk.major, disks->disk.minor);
 867                 sysfs_free(list);
 868         }
 869         st->ss->free_super(st);
 870         return dev;
 871 }
 872
 873
 874 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 875 {
 876         struct state *from;
 877         struct state *st;
 878
 879         link_containers_with_subarrays(statelist);
 880         for (st = statelist; st; st = st->next)
 881                 if (st->active < st->raid &&
 882                     st->spare == 0 && !st->err) {
 883                         struct domainlist *domlist = NULL;
 884                         int d;
 885                         struct state *to = st;
 886                         unsigned long long min_size;
 887
 888                         if (to->parent_dev != NoMdDev && !to->parent)
 889                                 /* subarray monitored without parent container
 890                                  * we can't move spares here */
 891                                 continue;
 892
 893                         if (to->parent)
 894                                 /* member of a container */
 895                                 to = to->parent;
 896
 897                         if (get_min_spare_size_required(to, &min_size))
 898                                 continue;
 899                         if (to->metadata->ss->external) {
 900                                 /* We must make sure there is
 901                                  * no suitable spare in container already.
 902                                  * If there is we don't add more */
 903                                 dev_t devid = container_choose_spare(
 904                                         to, to, NULL, min_size, st->active);
 905                                 if (devid > 0)
 906                                         continue;
 907                         }
 908                         for (d = 0; d < MaxDisks; d++)
 909                                 if (to->devid[d])
 910                                         domainlist_add_dev(&domlist,
 911                                                            to->devid[d],
 912                                                            to->metadata->ss->name);
 913                         if (to->spare_group)
 914                                 domain_add(&domlist, to->spare_group);
 915                         /*
 916                          * No spare migration if the destination
 917                          * has no domain. Skip this array.
 918                          */
 919                         if (!domlist)
 920                                 continue;
 921                         for (from=statelist ; from ; from=from->next) {
 922                                 dev_t devid;
 923                                 if (!check_donor(from, to))
 924                                         continue;
 925                                 if (from->metadata->ss->external)
 926                                         devid = container_choose_spare(
 927                                                 from, to, domlist, min_size, 0);
 928                                 else
 929                                         devid = choose_spare(from, to, domlist,
 930                                                              min_size);
 931                                 if (devid > 0
 932                                     && move_spare(from->devname, to->devname, devid)) {
 933                                         alert("MoveSpare", to->devname, from->devname, info);
 934                                         break;
 935                                 }
 936                         }
 937                         domain_free(domlist);
 938                 }
 939 }
 940
 941 /* search the statelist to connect external
 942  * metadata subarrays with their containers
 943  * We always completely rebuild the tree from scratch as
 944  * that is safest considering the possibility of entries
 945  * disappearing or changing.
 946  */
 947 static void link_containers_with_subarrays(struct state *list)
 948 {
 949         struct state *st;
 950         struct state *cont;
 951         for (st = list; st; st = st->next) {
 952                 st->parent = NULL;
 953                 st->subarray = NULL;
 954         }
 955         for (st = list; st; st = st->next)
 956                 if (st->parent_dev != NoMdDev)
 957                         for (cont = list; cont; cont = cont->next)
 958                                 if (!cont->err &&
 959                                     cont->parent_dev == NoMdDev &&
 960                                     cont->devnum == st->parent_dev) {
 961                                         st->parent = cont;
 962                                         st->subarray = cont->subarray;
 963                                         cont->subarray = st;
 964                                         break;
 965                                 }
 966 }
 967
 968 /* Not really Monitor but ... */
 969 int Wait(char *dev)
 970 {
 971         struct stat stb;
 972         int devnum;
 973         int rv = 1;
 974
 975         if (stat(dev, &stb) != 0) {
 976                 fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
 977                         strerror(errno));
 978                 return 2;
 979         }
 980         devnum = stat2devnum(&stb);
 981
 982         while(1) {
 983                 struct mdstat_ent *ms = mdstat_read(1, 0);
 984                 struct mdstat_ent *e;
 985
 986                 for (e=ms ; e; e=e->next)
 987                         if (e->devnum == devnum)
 988                                 break;
 989
 990                 if (!e || e->percent < 0) {
 991                         if (e && e->metadata_version &&
 992                             strncmp(e->metadata_version, "external:", 9) == 0) {
 993                                 if (is_subarray(&e->metadata_version[9]))
 994                                         ping_monitor(&e->metadata_version[9]);
 995                                 else
 996                                         ping_monitor_by_id(devnum);
 997                         }
 998                         free_mdstat(ms);
 999                         return rv;
1000                 }
1001                 free_mdstat(ms);
1002                 rv = 0;
1003                 mdstat_wait(5);
1004         }
1005 }
1006
1007 #ifndef MDASSEMBLE
1008
1009 static char *clean_states[] = {
1010         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
1011
1012 int WaitClean(char *dev, int sock, int verbose)
1013 {
1014         int fd;
1015         struct mdinfo *mdi;
1016         int rv = 1;
1017         int devnum;
1018
1019         fd = open(dev, O_RDONLY);
1020         if (fd < 0) {
1021                 if (verbose)
1022                         fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
1023                 return 1;
1024         }
1025
1026         devnum = fd2devnum(fd);
1027         mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1028         if (!mdi) {
1029                 if (verbose)
1030                         fprintf(stderr, Name ": Failed to read sysfs attributes for "
1031                                 "%s\n", dev);
1032                 close(fd);
1033                 return 0;
1034         }
1035
1036         switch(mdi->array.level) {
1037         case LEVEL_LINEAR:
1038         case LEVEL_MULTIPATH:
1039         case 0:
1040                 /* safemode delay is irrelevant for these levels */
1041                 rv = 0;
1042         }
1043
1044         /* for internal metadata the kernel handles the final clean
1045          * transition, containers can never be dirty
1046          */
1047         if (!is_subarray(mdi->text_version))
1048                 rv = 0;
1049
1050         /* safemode disabled ? */
1051         if (mdi->safe_mode_delay == 0)
1052                 rv = 0;
1053
1054         if (rv) {
1055                 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
1056                 char buf[20];
1057                 fd_set fds;
1058                 struct timeval tm;
1059
1060                 /* minimize the safe_mode_delay and prepare to wait up to 5s
1061                  * for writes to quiesce
1062                  */
1063                 sysfs_set_safemode(mdi, 1);
1064                 tm.tv_sec = 5;
1065                 tm.tv_usec = 0;
1066
1067                 FD_ZERO(&fds);
1068
1069                 /* wait for array_state to be clean */
1070                 while (1) {
1071                         rv = read(state_fd, buf, sizeof(buf));
1072                         if (rv < 0)
1073                                 break;
1074                         if (sysfs_match_word(buf, clean_states) <= 4)
1075                                 break;
1076                         FD_SET(state_fd, &fds);
1077                         rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
1078                         if (rv < 0 && errno != EINTR)
1079                                 break;
1080                         lseek(state_fd, 0, SEEK_SET);
1081                 }
1082                 if (rv < 0)
1083                         rv = 1;
1084                 else if (fping_monitor(sock) == 0 ||
1085                          ping_monitor(mdi->text_version) == 0) {
1086                         /* we need to ping to close the window between array
1087                          * state transitioning to clean and the metadata being
1088                          * marked clean
1089                          */
1090                         rv = 0;
1091                 } else
1092                         rv = 1;
1093                 if (rv && verbose)
1094                         fprintf(stderr, Name ": Error waiting for %s to be clean\n",
1095                                 dev);
1096
1097                 /* restore the original safe_mode_delay */
1098                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1099                 close(state_fd);
1100         }
1101
1102         sysfs_free(mdi);
1103         close(fd);
1104
1105         return rv;
1106 }
1107 #endif /* MDASSEMBLE */