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         int finished = 0;
 128         struct mdstat_ent *mdstat = NULL;
 129         char *mailfrom = NULL;
 130         struct alert_info info;
 131
 132         if (!mailaddr) {
 133                 mailaddr = conf_get_mailaddr();
 134                 if (mailaddr && ! scan)
 135                         fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n",
 136                                mailaddr);
 137         }
 138         mailfrom = conf_get_mailfrom();
 139
 140         if (!alert_cmd) {
 141                 alert_cmd = conf_get_program();
 142                 if (alert_cmd && ! scan)
 143                         fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n",
 144                                alert_cmd);
 145         }
 146         if (scan && !mailaddr && !alert_cmd && !dosyslog) {
 147                 fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n");
 148                 return 1;
 149         }
 150         info.alert_cmd = alert_cmd;
 151         info.mailaddr = mailaddr;
 152         info.mailfrom = mailfrom;
 153         info.dosyslog = dosyslog;
 154
 155         if (daemonise) {
 156                 int rv = make_daemon(pidfile);
 157                 if (rv >= 0)
 158                         return rv;
 159         }
 160
 161         if (share)
 162                 if (check_one_sharer(scan))
 163                         return 1;
 164
 165         if (devlist == NULL) {
 166                 struct mddev_ident *mdlist = conf_get_ident(NULL);
 167                 for (; mdlist; mdlist=mdlist->next) {
 168                         struct state *st;
 169                         if (mdlist->devname == NULL)
 170                                 continue;
 171                         if (strcasecmp(mdlist->devname, "<ignore>") == 0)
 172                                 continue;
 173                         st = calloc(1, sizeof *st);
 174                         if (st == NULL)
 175                                 continue;
 176                         if (mdlist->devname[0] == '/')
 177                                 st->devname = strdup(mdlist->devname);
 178                         else {
 179                                 st->devname = malloc(8+strlen(mdlist->devname)+1);
 180                                 strcpy(strcpy(st->devname, "/dev/md/"),
 181                                        mdlist->devname);
 182                         }
 183                         st->next = statelist;
 184                         st->devnum = INT_MAX;
 185                         st->percent = -2;
 186                         st->expected_spares = mdlist->spare_disks;
 187                         if (mdlist->spare_group)
 188                                 st->spare_group = strdup(mdlist->spare_group);
 189                         statelist = st;
 190                 }
 191         } else {
 192                 struct mddev_dev *dv;
 193                 for (dv=devlist ; dv; dv=dv->next) {
 194                         struct mddev_ident *mdlist = conf_get_ident(dv->devname);
 195                         struct state *st = calloc(1, sizeof *st);
 196                         if (st == NULL)
 197                                 continue;
 198                         st->devname = strdup(dv->devname);
 199                         st->next = statelist;
 200                         st->devnum = INT_MAX;
 201                         st->percent = -2;
 202                         st->expected_spares = -1;
 203                         if (mdlist) {
 204                                 st->expected_spares = mdlist->spare_disks;
 205                                 if (mdlist->spare_group)
 206                                         st->spare_group = strdup(mdlist->spare_group);
 207                         }
 208                         statelist = st;
 209                 }
 210         }
 211
 212
 213         while (! finished) {
 214                 int new_found = 0;
 215                 struct state *st;
 216                 int anydegraded = 0;
 217
 218                 if (mdstat)
 219                         free_mdstat(mdstat);
 220                 mdstat = mdstat_read(oneshot?0:1, 0);
 221
 222                 for (st=statelist; st; st=st->next)
 223                         if (check_array(st, mdstat, test, &info, increments))
 224                                 anydegraded = 1;
 225
 226                 /* now check if there are any new devices found in mdstat */
 227                 if (scan)
 228                         new_found = add_new_arrays(mdstat, &statelist, test,
 229                                                    &info);
 230
 231                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 232                  * Look for another array with spare > 0 and active == raid and same spare_group
 233                  *  if found, choose a device and hotremove/hotadd
 234                  */
 235                 if (share && anydegraded)
 236                         try_spare_migration(statelist, &info);
 237                 if (!new_found) {
 238                         if (oneshot)
 239                                 break;
 240                         else
 241                                 mdstat_wait(period);
 242                 }
 243                 test = 0;
 244         }
 245         if (pidfile)
 246                 unlink(pidfile);
 247         return 0;
 248 }
 249
 250 static int make_daemon(char *pidfile)
 251 {
 252         /* Return:
 253          * -1 in the forked daemon
 254          *  0 in the parent
 255          *  1 on error
 256          * so a none-negative becomes the exit code.
 257          */
 258         int pid = fork();
 259         if (pid > 0) {
 260                 if (!pidfile)
 261                         printf("%d\n", pid);
 262                 else {
 263                         FILE *pid_file;
 264                         pid_file=fopen(pidfile, "w");
 265                         if (!pid_file)
 266                                 perror("cannot create pid file");
 267                         else {
 268                                 fprintf(pid_file,"%d\n", pid);
 269                                 fclose(pid_file);
 270                         }
 271                 }
 272                 return 0;
 273         }
 274         if (pid < 0) {
 275                 perror("daemonise");
 276                 return 1;
 277         }
 278         close(0);
 279         open("/dev/null", O_RDWR);
 280         dup2(0,1);
 281         dup2(0,2);
 282         setsid();
 283         return -1;
 284 }
 285
 286 static int check_one_sharer(int scan)
 287 {
 288         int pid, rv;
 289         FILE *fp;
 290         char dir[20];
 291         struct stat buf;
 292         fp = fopen("/var/run/mdadm/autorebuild.pid", "r");
 293         if (fp) {
 294                 if (fscanf(fp, "%d", &pid) != 1)
 295                         pid = -1;
 296                 sprintf(dir, "/proc/%d", pid);
 297                 rv = stat(dir, &buf);
 298                 if (rv != -1) {
 299                         if (scan) {
 300                                 fprintf(stderr, Name ": Only one "
 301                                         "autorebuild process allowed"
 302                                         " in scan mode, aborting\n");
 303                                 fclose(fp);
 304                                 return 1;
 305                         } else {
 306                                 fprintf(stderr, Name ": Warning: One"
 307                                         " autorebuild process already"
 308                                         " running.\n");
 309                         }
 310                 }
 311                 fclose(fp);
 312         }
 313         if (scan) {
 314                 if (mkdir("/var/run/mdadm", S_IRWXU) < 0 &&
 315                     errno != EEXIST) {
 316                         fprintf(stderr, Name ": Can't create "
 317                                 "autorebuild.pid file\n");
 318                 } else {
 319                         fp = fopen("/var/run/mdadm/autorebuild.pid", "w");
 320                         if (!fp)
 321                                 fprintf(stderr, Name ": Cannot create"
 322                                         " autorebuild.pid"
 323                                         "file\n");
 324                         else {
 325                                 pid = getpid();
 326                                 fprintf(fp, "%d\n", pid);
 327                                 fclose(fp);
 328                         }
 329                 }
 330         }
 331         return 0;
 332 }
 333
 334 static void alert(char *event, char *dev, char *disc, struct alert_info *info)
 335 {
 336         int priority;
 337
 338         if (!info->alert_cmd && !info->mailaddr && !info->dosyslog) {
 339                 time_t now = time(0);
 340
 341                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 342         }
 343         if (info->alert_cmd) {
 344                 int pid = fork();
 345                 switch(pid) {
 346                 default:
 347                         waitpid(pid, NULL, 0);
 348                         break;
 349                 case -1:
 350                         break;
 351                 case 0:
 352                         execl(info->alert_cmd, info->alert_cmd,
 353                               event, dev, disc, NULL);
 354                         exit(2);
 355                 }
 356         }
 357         if (info->mailaddr &&
 358             (strncmp(event, "Fail", 4)==0 ||
 359              strncmp(event, "Test", 4)==0 ||
 360              strncmp(event, "Spares", 6)==0 ||
 361              strncmp(event, "Degrade", 7)==0)) {
 362                 FILE *mp = popen(Sendmail, "w");
 363                 if (mp) {
 364                         FILE *mdstat;
 365                         char hname[256];
 366                         gethostname(hname, sizeof(hname));
 367                         signal(SIGPIPE, SIG_IGN);
 368                         if (info->mailfrom)
 369                                 fprintf(mp, "From: %s\n", info->mailfrom);
 370                         else
 371                                 fprintf(mp, "From: " Name " monitoring <root>\n");
 372                         fprintf(mp, "To: %s\n", info->mailaddr);
 373                         fprintf(mp, "Subject: %s event on %s:%s\n\n",
 374                                 event, dev, hname);
 375
 376                         fprintf(mp,
 377                                 "This is an automatically generated"
 378                                 " mail message from " Name "\n");
 379                         fprintf(mp, "running on %s\n\n", hname);
 380
 381                         fprintf(mp,
 382                                 "A %s event had been detected on"
 383                                 " md device %s.\n\n", event, dev);
 384
 385                         if (disc && disc[0] != ' ')
 386                                 fprintf(mp,
 387                                         "It could be related to"
 388                                         " component device %s.\n\n", disc);
 389                         if (disc && disc[0] == ' ')
 390                                 fprintf(mp, "Extra information:%s.\n\n", disc);
 391
 392                         fprintf(mp, "Faithfully yours, etc.\n");
 393
 394                         mdstat = fopen("/proc/mdstat", "r");
 395                         if (mdstat) {
 396                                 char buf[8192];
 397                                 int n;
 398                                 fprintf(mp,
 399                                         "\nP.S. The /proc/mdstat file"
 400                                         " currently contains the following:\n\n");
 401                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 402                                         n=fwrite(buf, 1, n, mp);
 403                                 fclose(mdstat);
 404                         }
 405                         pclose(mp);
 406                 }
 407         }
 408
 409         /* log the event to syslog maybe */
 410         if (info->dosyslog) {
 411                 /* Log at a different severity depending on the event.
 412                  *
 413                  * These are the critical events:  */
 414                 if (strncmp(event, "Fail", 4)==0 ||
 415                     strncmp(event, "Degrade", 7)==0 ||
 416                     strncmp(event, "DeviceDisappeared", 17)==0)
 417                         priority = LOG_CRIT;
 418                 /* Good to know about, but are not failures: */
 419                 else if (strncmp(event, "Rebuild", 7)==0 ||
 420                          strncmp(event, "MoveSpare", 9)==0 ||
 421                          strncmp(event, "Spares", 6) != 0)
 422                         priority = LOG_WARNING;
 423                 /* Everything else: */
 424                 else
 425                         priority = LOG_INFO;
 426
 427                 if (disc)
 428                         syslog(priority,
 429                                "%s event detected on md device %s,"
 430                                " component device %s", event, dev, disc);
 431                 else
 432                         syslog(priority,
 433                                "%s event detected on md device %s",
 434                                event, dev);
 435         }
 436 }
 437
 438 static int check_array(struct state *st, struct mdstat_ent *mdstat,
 439                        int test, struct alert_info *ainfo,
 440                        int increments)
 441 {
 442         /* Update the state 'st' to reflect any changes shown in mdstat,
 443          * or found by directly examining the array, and return
 444          * '1' if the array is degraded, or '0' if it is optimal (or dead).
 445          */
 446         struct { int state, major, minor; } info[MaxDisks];
 447         mdu_array_info_t array;
 448         struct mdstat_ent *mse = NULL, *mse2;
 449         char *dev = st->devname;
 450         int fd;
 451         int i;
 452
 453         if (test)
 454                 alert("TestMessage", dev, NULL, ainfo);
 455         fd = open(dev, O_RDONLY);
 456         if (fd < 0) {
 457                 if (!st->err)
 458                         alert("DeviceDisappeared", dev, NULL, ainfo);
 459                 st->err=1;
 460                 return 0;
 461         }
 462         fcntl(fd, F_SETFD, FD_CLOEXEC);
 463         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 464                 if (!st->err)
 465                         alert("DeviceDisappeared", dev, NULL, ainfo);
 466                 st->err=1;
 467                 close(fd);
 468                 return 0;
 469         }
 470         /* It's much easier to list what array levels can't
 471          * have a device disappear than all of them that can
 472          */
 473         if (array.level == 0 || array.level == -1) {
 474                 if (!st->err)
 475                         alert("DeviceDisappeared", dev, "Wrong-Level", ainfo);
 476                 st->err = 1;
 477                 close(fd);
 478                 return 0;
 479         }
 480         if (st->devnum == INT_MAX) {
 481                 struct stat stb;
 482                 if (fstat(fd, &stb) == 0 &&
 483                     (S_IFMT&stb.st_mode)==S_IFBLK) {
 484                         if (major(stb.st_rdev) == MD_MAJOR)
 485                                 st->devnum = minor(stb.st_rdev);
 486                         else
 487                                 st->devnum = -1- (minor(stb.st_rdev)>>6);
 488                 }
 489         }
 490
 491         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 492                 if (mse2->devnum == st->devnum) {
 493                         mse2->devnum = INT_MAX; /* flag it as "used" */
 494                         mse = mse2;
 495                 }
 496
 497         if (!mse) {
 498                 /* duplicated array in statelist
 499                  * or re-created after reading mdstat*/
 500                 st->err = 1;
 501                 close(fd);
 502                 return 0;
 503         }
 504         /* this array is in /proc/mdstat */
 505         if (array.utime == 0)
 506                 /* external arrays don't update utime, so
 507                  * just make sure it is always different. */
 508                 array.utime = st->utime + 1;;
 509
 510         if (st->utime == array.utime &&
 511             st->failed == array.failed_disks &&
 512             st->working == array.working_disks &&
 513             st->spare == array.spare_disks &&
 514             (mse == NULL  || (
 515                     mse->percent == st->percent
 516                     ))) {
 517                 close(fd);
 518                 st->err = 0;
 519                 if ((st->active < st->raid) && st->spare == 0)
 520                         return 1;
 521                 else
 522                         return 0;
 523         }
 524         if (st->utime == 0 && /* new array */
 525             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 526                 )
 527                 alert("DegradedArray", dev, NULL, ainfo);
 528
 529         if (st->utime == 0 && /* new array */
 530             st->expected_spares > 0 &&
 531             array.spare_disks < st->expected_spares)
 532                 alert("SparesMissing", dev, NULL, ainfo);
 533         if (st->percent == -1 &&
 534             mse->percent >= 0)
 535                 alert("RebuildStarted", dev, NULL, ainfo);
 536         if (st->percent >= 0 &&
 537             mse->percent >= 0 &&
 538             (mse->percent / increments) > (st->percent / increments)) {
 539                 char percentalert[15]; // "RebuildNN" (10 chars) or "RebuildStarted" (15 chars)
 540
 541                 if((mse->percent / increments) == 0)
 542                         snprintf(percentalert, sizeof(percentalert), "RebuildStarted");
 543                 else
 544                         snprintf(percentalert, sizeof(percentalert), "Rebuild%02d", mse->percent);
 545
 546                 alert(percentalert, dev, NULL, ainfo);
 547         }
 548
 549         if (mse->percent == -1 &&
 550             st->percent >= 0) {
 551                 /* Rebuild/sync/whatever just finished.
 552                  * If there is a number in /mismatch_cnt,
 553                  * we should report that.
 554                  */
 555                 struct mdinfo *sra =
 556                         sysfs_read(-1, st->devnum, GET_MISMATCH);
 557                 if (sra && sra->mismatch_cnt > 0) {
 558                         char cnt[40];
 559                         sprintf(cnt, " mismatches found: %d", sra->mismatch_cnt);
 560                         alert("RebuildFinished", dev, cnt, ainfo);
 561                 } else
 562                         alert("RebuildFinished", dev, NULL, ainfo);
 563                 if (sra)
 564                         free(sra);
 565         }
 566         st->percent = mse->percent;
 567
 568         for (i=0; i<MaxDisks && i <= array.raid_disks + array.nr_disks;
 569              i++) {
 570                 mdu_disk_info_t disc;
 571                 disc.number = i;
 572                 if (ioctl(fd, GET_DISK_INFO, &disc) >= 0) {
 573                         info[i].state = disc.state;
 574                         info[i].major = disc.major;
 575                         info[i].minor = disc.minor;
 576                 } else
 577                         info[i].major = info[i].minor = 0;
 578         }
 579
 580         if (mse->metadata_version &&
 581             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 582             is_subarray(mse->metadata_version+9))
 583                 st->parent_dev =
 584                         devname2devnum(mse->metadata_version+10);
 585         else
 586                 st->parent_dev = NoMdDev;
 587         if (st->metadata == NULL &&
 588             st->parent_dev == NoMdDev)
 589                 st->metadata = super_by_fd(fd, NULL);
 590
 591         close(fd);
 592
 593         for (i=0; i<MaxDisks; i++) {
 594                 mdu_disk_info_t disc = {0,0,0,0,0};
 595                 int newstate=0;
 596                 int change;
 597                 char *dv = NULL;
 598                 disc.number = i;
 599                 if (i > array.raid_disks + array.nr_disks) {
 600                         newstate = 0;
 601                         disc.major = disc.minor = 0;
 602                 } else if (info[i].major || info[i].minor) {
 603                         newstate = info[i].state;
 604                         dv = map_dev(info[i].major, info[i].minor, 1);
 605                         disc.state = newstate;
 606                         disc.major = info[i].major;
 607                         disc.minor = info[i].minor;
 608                 } else if (mse &&  mse->pattern && i < (int)strlen(mse->pattern)) {
 609                         switch(mse->pattern[i]) {
 610                         case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
 611                         case '_': newstate = 8 /* REMOVED */; break;
 612                         }
 613                         disc.major = disc.minor = 0;
 614                 }
 615                 if (dv == NULL && st->devid[i])
 616                         dv = map_dev(major(st->devid[i]),
 617                                      minor(st->devid[i]), 1);
 618                 change = newstate ^ st->devstate[i];
 619                 if (st->utime && change && !st->err) {
 620                         if (i < array.raid_disks &&
 621                             (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
 622                              ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
 623                              ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
 624                                 )
 625                                 alert("Fail", dev, dv, ainfo);
 626                         else if (i >= array.raid_disks &&
 627                                  (disc.major || disc.minor) &&
 628                                  st->devid[i] == makedev(disc.major, disc.minor) &&
 629                                  ((newstate&change)&(1<<MD_DISK_FAULTY))
 630                                 )
 631                                 alert("FailSpare", dev, dv, ainfo);
 632                         else if (i < array.raid_disks &&
 633                                  ! (newstate & (1<<MD_DISK_REMOVED)) &&
 634                                  (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
 635                                   ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
 636                                   ((newstate&change)&(1<<MD_DISK_SYNC)))
 637                                 )
 638                                 alert("SpareActive", dev, dv, ainfo);
 639                 }
 640                 st->devstate[i] = newstate;
 641                 st->devid[i] = makedev(disc.major, disc.minor);
 642         }
 643         st->active = array.active_disks;
 644         st->working = array.working_disks;
 645         st->spare = array.spare_disks;
 646         st->failed = array.failed_disks;
 647         st->utime = array.utime;
 648         st->raid = array.raid_disks;
 649         st->err = 0;
 650         if ((st->active < st->raid) && st->spare == 0)
 651                 return 1;
 652         return 0;
 653 }
 654
 655 static int add_new_arrays(struct mdstat_ent *mdstat, struct state **statelist,
 656                           int test, struct alert_info *info)
 657 {
 658         struct mdstat_ent *mse;
 659         int new_found = 0;
 660
 661         for (mse=mdstat; mse; mse=mse->next)
 662                 if (mse->devnum != INT_MAX &&
 663                     (!mse->level  || /* retrieve containers */
 664                      (strcmp(mse->level, "raid0") != 0 &&
 665                       strcmp(mse->level, "linear") != 0))
 666                         ) {
 667                         struct state *st = calloc(1, sizeof *st);
 668                         mdu_array_info_t array;
 669                         int fd;
 670                         if (st == NULL)
 671                                 continue;
 672                         st->devname = strdup(get_md_name(mse->devnum));
 673                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 674                             ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
 675                                 /* no such array */
 676                                 if (fd >=0) close(fd);
 677                                 put_md_name(st->devname);
 678                                 free(st->devname);
 679                                 if (st->metadata) {
 680                                         st->metadata->ss->free_super(st->metadata);
 681                                         free(st->metadata);
 682                                 }
 683                                 free(st);
 684                                 continue;
 685                         }
 686                         close(fd);
 687                         st->next = *statelist;
 688                         st->err = 1;
 689                         st->devnum = mse->devnum;
 690                         st->percent = -2;
 691                         st->expected_spares = -1;
 692                         if (mse->metadata_version &&
 693                             strncmp(mse->metadata_version, "external:", 9) == 0 &&
 694                             is_subarray(mse->metadata_version+9))
 695                                 st->parent_dev =
 696                                         devname2devnum(mse->metadata_version+10);
 697                         else
 698                                 st->parent_dev = NoMdDev;
 699                         *statelist = st;
 700                         if (test)
 701                                 alert("TestMessage", st->devname, NULL, info);
 702                         alert("NewArray", st->devname, NULL, info);
 703                         new_found = 1;
 704                 }
 705         return new_found;
 706 }
 707
 708 static int get_min_spare_size_required(struct state *st, unsigned long long *sizep)
 709 {
 710         int fd;
 711
 712         if (!st->metadata ||
 713             !st->metadata->ss->min_acceptable_spare_size) {
 714                 *sizep = 0;
 715                 return 0;
 716         }
 717
 718         fd = open(st->devname, O_RDONLY);
 719         if (fd < 0)
 720                 return 1;
 721         if (st->metadata->ss->external)
 722                 st->metadata->ss->load_container(st->metadata, fd, st->devname);
 723         else
 724                 st->metadata->ss->load_super(st->metadata, fd, st->devname);
 725         close(fd);
 726         if (!st->metadata->sb)
 727                 return 1;
 728         *sizep = st->metadata->ss->min_acceptable_spare_size(st->metadata);
 729         st->metadata->ss->free_super(st->metadata);
 730
 731         return 0;
 732 }
 733
 734 static int check_donor(struct state *from, struct state *to)
 735 {
 736         struct state *sub;
 737
 738         if (from == to)
 739                 return 0;
 740         if (from->parent)
 741                 /* Cannot move from a member */
 742                 return 0;
 743         if (from->err)
 744                 return 0;
 745         for (sub = from->subarray; sub; sub = sub->subarray)
 746                 /* If source array has degraded subarrays, don't
 747                  * remove anything
 748                  */
 749                 if (sub->active < sub->raid)
 750                         return 0;
 751         if (from->metadata->ss->external == 0)
 752                 if (from->active < from->raid)
 753                         return 0;
 754         if (from->spare <= 0)
 755                 return 0;
 756         return 1;
 757 }
 758
 759 static dev_t choose_spare(struct state *from, struct state *to,
 760                         struct domainlist *domlist, unsigned long long min_size)
 761 {
 762         int d;
 763         dev_t dev = 0;
 764
 765         for (d = from->raid; !dev && d < MaxDisks; d++) {
 766                 if (from->devid[d] > 0 &&
 767                     from->devstate[d] == 0) {
 768                         struct dev_policy *pol;
 769                         unsigned long long dev_size;
 770
 771                         if (to->metadata->ss->external &&
 772                             test_partition_from_id(from->devid[d]))
 773                                 continue;
 774
 775                         if (min_size &&
 776                             dev_size_from_id(from->devid[d], &dev_size) &&
 777                             dev_size < min_size)
 778                                 continue;
 779
 780                         pol = devnum_policy(from->devid[d]);
 781                         if (from->spare_group)
 782                                 pol_add(&pol, pol_domain,
 783                                         from->spare_group, NULL);
 784                         if (domain_test(domlist, pol, to->metadata->ss->name) == 1)
 785                             dev = from->devid[d];
 786                         dev_policy_free(pol);
 787                 }
 788         }
 789         return dev;
 790 }
 791
 792 static dev_t container_choose_spare(struct state *from, struct state *to,
 793                                     struct domainlist *domlist,
 794                                     unsigned long long min_size, int active)
 795 {
 796         /* This is similar to choose_spare, but we cannot trust devstate,
 797          * so we need to read the metadata instead
 798          */
 799         struct mdinfo *list;
 800         struct supertype *st = from->metadata;
 801         int fd = open(from->devname, O_RDONLY);
 802         int err;
 803         dev_t dev = 0;
 804
 805         if (fd < 0)
 806                 return 0;
 807         if (!st->ss->getinfo_super_disks) {
 808                 close(fd);
 809                 return 0;
 810         }
 811
 812         err = st->ss->load_container(st, fd, NULL);
 813         close(fd);
 814         if (err)
 815                 return 0;
 816
 817         if (from == to) {
 818                 /* We must check if number of active disks has not increased
 819                  * since ioctl in main loop. mdmon may have added spare
 820                  * to subarray. If so we do not need to look for more spares
 821                  * so return non zero value */
 822                 int active_cnt = 0;
 823                 struct mdinfo *dp;
 824                 list = st->ss->getinfo_super_disks(st);
 825                 if (!list) {
 826                         st->ss->free_super(st);
 827                         return 1;
 828                 }
 829                 dp = list->devs;
 830                 while (dp) {
 831                         if (dp->disk.state & (1<<MD_DISK_SYNC) &&
 832                             !(dp->disk.state & (1<<MD_DISK_FAULTY)))
 833                                 active_cnt++;
 834                         dp = dp->next;
 835                 }
 836                 sysfs_free(list);
 837                 if (active < active_cnt) {
 838                         /* Spare just activated.*/
 839                         st->ss->free_super(st);
 840                         return 1;
 841                 }
 842         }
 843
 844         /* We only need one spare so full list not needed */
 845         list = container_choose_spares(st, min_size, domlist, from->spare_group,
 846                                        to->metadata->ss->name, 1);
 847         if (list) {
 848                 struct mdinfo *disks = list->devs;
 849                 if (disks)
 850                         dev = makedev(disks->disk.major, disks->disk.minor);
 851                 sysfs_free(list);
 852         }
 853         st->ss->free_super(st);
 854         return dev;
 855 }
 856
 857
 858 static void try_spare_migration(struct state *statelist, struct alert_info *info)
 859 {
 860         struct state *from;
 861         struct state *st;
 862
 863         link_containers_with_subarrays(statelist);
 864         for (st = statelist; st; st = st->next)
 865                 if (st->active < st->raid &&
 866                     st->spare == 0 && !st->err) {
 867                         struct domainlist *domlist = NULL;
 868                         int d;
 869                         struct state *to = st;
 870                         unsigned long long min_size;
 871
 872                         if (to->parent_dev != NoMdDev && !to->parent)
 873                                 /* subarray monitored without parent container
 874                                  * we can't move spares here */
 875                                 continue;
 876
 877                         if (to->parent)
 878                                 /* member of a container */
 879                                 to = to->parent;
 880
 881                         if (get_min_spare_size_required(to, &min_size))
 882                                 continue;
 883                         if (to->metadata->ss->external) {
 884                                 /* We must make sure there is
 885                                  * no suitable spare in container already.
 886                                  * If there is we don't add more */
 887                                 dev_t devid = container_choose_spare(
 888                                         to, to, NULL, min_size, st->active);
 889                                 if (devid > 0)
 890                                         continue;
 891                         }
 892                         for (d = 0; d < MaxDisks; d++)
 893                                 if (to->devid[d])
 894                                         domainlist_add_dev(&domlist,
 895                                                            to->devid[d],
 896                                                            to->metadata->ss->name);
 897                         if (to->spare_group)
 898                                 domain_add(&domlist, to->spare_group);
 899                         /*
 900                          * No spare migration if the destination
 901                          * has no domain. Skip this array.
 902                          */
 903                         if (!domlist)
 904                                 continue;
 905                         for (from=statelist ; from ; from=from->next) {
 906                                 dev_t devid;
 907                                 if (!check_donor(from, to))
 908                                         continue;
 909                                 if (from->metadata->ss->external)
 910                                         devid = container_choose_spare(
 911                                                 from, to, domlist, min_size, 0);
 912                                 else
 913                                         devid = choose_spare(from, to, domlist,
 914                                                              min_size);
 915                                 if (devid > 0
 916                                     && move_spare(from->devname, to->devname, devid)) {
 917                                         alert("MoveSpare", to->devname, from->devname, info);
 918                                         break;
 919                                 }
 920                         }
 921                         domain_free(domlist);
 922                 }
 923 }
 924
 925 /* search the statelist to connect external
 926  * metadata subarrays with their containers
 927  * We always completely rebuild the tree from scratch as
 928  * that is safest considering the possibility of entries
 929  * disappearing or changing.
 930  */
 931 static void link_containers_with_subarrays(struct state *list)
 932 {
 933         struct state *st;
 934         struct state *cont;
 935         for (st = list; st; st = st->next) {
 936                 st->parent = NULL;
 937                 st->subarray = NULL;
 938         }
 939         for (st = list; st; st = st->next)
 940                 if (st->parent_dev != NoMdDev)
 941                         for (cont = list; cont; cont = cont->next)
 942                                 if (!cont->err &&
 943                                     cont->parent_dev == NoMdDev &&
 944                                     cont->devnum == st->parent_dev) {
 945                                         st->parent = cont;
 946                                         st->subarray = cont->subarray;
 947                                         cont->subarray = st;
 948                                         break;
 949                                 }
 950 }
 951
 952 /* Not really Monitor but ... */
 953 int Wait(char *dev)
 954 {
 955         struct stat stb;
 956         int devnum;
 957         int rv = 1;
 958
 959         if (stat(dev, &stb) != 0) {
 960                 fprintf(stderr, Name ": Cannot find %s: %s\n", dev,
 961                         strerror(errno));
 962                 return 2;
 963         }
 964         devnum = stat2devnum(&stb);
 965
 966         while(1) {
 967                 struct mdstat_ent *ms = mdstat_read(1, 0);
 968                 struct mdstat_ent *e;
 969
 970                 for (e=ms ; e; e=e->next)
 971                         if (e->devnum == devnum)
 972                                 break;
 973
 974                 if (!e || e->percent < 0) {
 975                         if (e && e->metadata_version &&
 976                             strncmp(e->metadata_version, "external:", 9) == 0) {
 977                                 if (is_subarray(&e->metadata_version[9]))
 978                                         ping_monitor(&e->metadata_version[9]);
 979                                 else
 980                                         ping_monitor_by_id(devnum);
 981                         }
 982                         free_mdstat(ms);
 983                         return rv;
 984                 }
 985                 free_mdstat(ms);
 986                 rv = 0;
 987                 mdstat_wait(5);
 988         }
 989 }
 990
 991 #ifndef MDASSEMBLE
 992
 993 static char *clean_states[] = {
 994         "clear", "inactive", "readonly", "read-auto", "clean", NULL };
 995
 996 int WaitClean(char *dev, int sock, int verbose)
 997 {
 998         int fd;
 999         struct mdinfo *mdi;
1000         int rv = 1;
1001         int devnum;
1002
1003         fd = open(dev, O_RDONLY);
1004         if (fd < 0) {
1005                 if (verbose)
1006                         fprintf(stderr, Name ": Couldn't open %s: %s\n", dev, strerror(errno));
1007                 return 1;
1008         }
1009
1010         devnum = fd2devnum(fd);
1011         mdi = sysfs_read(fd, devnum, GET_VERSION|GET_LEVEL|GET_SAFEMODE);
1012         if (!mdi) {
1013                 if (verbose)
1014                         fprintf(stderr, Name ": Failed to read sysfs attributes for "
1015                                 "%s\n", dev);
1016                 close(fd);
1017                 return 0;
1018         }
1019
1020         switch(mdi->array.level) {
1021         case LEVEL_LINEAR:
1022         case LEVEL_MULTIPATH:
1023         case 0:
1024                 /* safemode delay is irrelevant for these levels */
1025                 rv = 0;
1026         }
1027
1028         /* for internal metadata the kernel handles the final clean
1029          * transition, containers can never be dirty
1030          */
1031         if (!is_subarray(mdi->text_version))
1032                 rv = 0;
1033
1034         /* safemode disabled ? */
1035         if (mdi->safe_mode_delay == 0)
1036                 rv = 0;
1037
1038         if (rv) {
1039                 int state_fd = sysfs_open(fd2devnum(fd), NULL, "array_state");
1040                 char buf[20];
1041                 fd_set fds;
1042                 struct timeval tm;
1043
1044                 /* minimize the safe_mode_delay and prepare to wait up to 5s
1045                  * for writes to quiesce
1046                  */
1047                 sysfs_set_safemode(mdi, 1);
1048                 tm.tv_sec = 5;
1049                 tm.tv_usec = 0;
1050
1051                 FD_ZERO(&fds);
1052
1053                 /* wait for array_state to be clean */
1054                 while (1) {
1055                         rv = read(state_fd, buf, sizeof(buf));
1056                         if (rv < 0)
1057                                 break;
1058                         if (sysfs_match_word(buf, clean_states) <= 4)
1059                                 break;
1060                         FD_SET(state_fd, &fds);
1061                         rv = select(state_fd + 1, NULL, NULL, &fds, &tm);
1062                         if (rv < 0 && errno != EINTR)
1063                                 break;
1064                         lseek(state_fd, 0, SEEK_SET);
1065                 }
1066                 if (rv < 0)
1067                         rv = 1;
1068                 else if (fping_monitor(sock) == 0 ||
1069                          ping_monitor(mdi->text_version) == 0) {
1070                         /* we need to ping to close the window between array
1071                          * state transitioning to clean and the metadata being
1072                          * marked clean
1073                          */
1074                         rv = 0;
1075                 } else
1076                         rv = 1;
1077                 if (rv && verbose)
1078                         fprintf(stderr, Name ": Error waiting for %s to be clean\n",
1079                                 dev);
1080
1081                 /* restore the original safe_mode_delay */
1082                 sysfs_set_safemode(mdi, mdi->safe_mode_delay);
1083                 close(state_fd);
1084         }
1085
1086         sysfs_free(mdi);
1087         close(fd);
1088
1089         return rv;
1090 }
1091 #endif /* MDASSEMBLE */