Monitor.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2006 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@cse.unsw.edu.au>
  23  *    Paper: Neil Brown
  24  *           School of Computer Science and Engineering
  25  *           The University of New South Wales
  26  *           Sydney, 2052
  27  *           Australia
  28  */
  29
  30 #include        "mdadm.h"
  31 #include        "md_p.h"
  32 #include        "md_u.h"
  33 #include        <sys/wait.h>
  34 #include        <sys/signal.h>
  35 #include        <values.h>
  36 #include        <syslog.h>
  37
  38 static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom,
  39                   char *cmd, int dosyslog);
  40
  41 static char *percentalerts[] = {
  42         "RebuildStarted",
  43         "Rebuild20",
  44         "Rebuild40",
  45         "Rebuild60",
  46         "Rebuild80",
  47 };
  48
  49 /* The largest number of disks current arrays can manage is 384
  50  * This really should be dynamically, but that will have to wait
  51  * At least it isn't MD_SB_DISKS.
  52  */
  53 #define MaxDisks 384
  54 int Monitor(mddev_dev_t devlist,
  55             char *mailaddr, char *alert_cmd,
  56             int period, int daemonise, int scan, int oneshot,
  57             int dosyslog, char *config, int test, char* pidfile)
  58 {
  59         /*
  60          * Every few seconds, scan every md device looking for changes
  61          * When a change is found, log it, possibly run the alert command,
  62          * and possibly send Email
  63          *
  64          * For each array, we record:
  65          *   Update time
  66          *   active/working/failed/spare drives
  67          *   State of each device.
  68          *   %rebuilt if rebuilding
  69          *
  70          * If the update time changes, check out all the data again
  71          * It is possible that we cannot get the state of each device
  72          * due to bugs in the md kernel module.
  73          * We also read /proc/mdstat to get rebuild percent,
  74          * and to get state on all active devices incase of kernel bug.
  75          *
  76          * Events are:
  77          *    Fail
  78          *      An active device had Faulty set or Active/Sync removed
  79          *    FailSpare
  80          *      A spare device had Faulty set
  81          *    SpareActive
  82          *      An active device had a reverse transition
  83          *    RebuildStarted
  84          *      percent went from -1 to +ve
  85          *    Rebuild20 Rebuild40 Rebuild60 Rebuild80
  86          *      percent went from below to not-below that number
  87          *    DeviceDisappeared
  88          *      Couldn't access a device which was previously visible
  89          *
  90          * if we detect an array with active<raid and spare==0
  91          * we look at other arrays that have same spare-group
  92          * If we find one with active==raid and spare>0,
  93          *  and if we can get_disk_info and find a name
  94          *  Then we hot-remove and hot-add to the other array
  95          *
  96          * If devlist is NULL, then we can monitor everything because --scan
  97          * was given.  We get an initial list from config file and add anything
  98          * that appears in /proc/mdstat
  99          */
 100
 101         struct state {
 102                 char *devname;
 103                 int devnum;     /* to sync with mdstat info */
 104                 long utime;
 105                 int err;
 106                 char *spare_group;
 107                 int active, working, failed, spare, raid;
 108                 int expected_spares;
 109                 int devstate[MaxDisks];
 110                 int devid[MaxDisks];
 111                 int percent;
 112                 struct state *next;
 113         } *statelist = NULL;
 114         int finished = 0;
 115         struct mdstat_ent *mdstat = NULL;
 116         char *mailfrom = NULL;
 117
 118         if (!mailaddr) {
 119                 mailaddr = conf_get_mailaddr(config);
 120                 if (mailaddr && ! scan)
 121                         fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n",
 122                                mailaddr);
 123         }
 124         mailfrom = conf_get_mailfrom(config);
 125
 126         if (!alert_cmd) {
 127                 alert_cmd = conf_get_program(config);
 128                 if (alert_cmd && ! scan)
 129                         fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n",
 130                                alert_cmd);
 131         }
 132         if (scan && !mailaddr && !alert_cmd) {
 133                 fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n");
 134                 return 1;
 135         }
 136
 137         if (daemonise) {
 138                 int pid = fork();
 139                 if (pid > 0) {
 140                         if (!pidfile)
 141                                 printf("%d\n", pid);
 142                         else {
 143                                 FILE *pid_file;
 144                                 pid_file=fopen(pidfile, "w");
 145                                 if (!pid_file)
 146                                         perror("cannot create pid file");
 147                                 else {
 148                                         fprintf(pid_file,"%d\n", pid);
 149                                         fclose(pid_file);
 150                                 }
 151                         }
 152                         return 0;
 153                 }
 154                 if (pid < 0) {
 155                         perror("daemonise");
 156                         return 1;
 157                 }
 158                 close(0);
 159                 open("/dev/null", 3);
 160                 dup2(0,1);
 161                 dup2(0,2);
 162                 setsid();
 163         }
 164
 165         if (devlist == NULL) {
 166                 mddev_ident_t mdlist = conf_get_ident(config, NULL);
 167                 for (; mdlist; mdlist=mdlist->next) {
 168                         struct state *st = malloc(sizeof *st);
 169                         if (st == NULL)
 170                                 continue;
 171                         st->devname = strdup(mdlist->devname);
 172                         st->utime = 0;
 173                         st->next = statelist;
 174                         st->err = 0;
 175                         st->devnum = MAXINT;
 176                         st->percent = -2;
 177                         st->expected_spares = mdlist->spare_disks;
 178                         if (mdlist->spare_group)
 179                                 st->spare_group = strdup(mdlist->spare_group);
 180                         else
 181                                 st->spare_group = NULL;
 182                         statelist = st;
 183                 }
 184         } else {
 185                 mddev_dev_t dv;
 186                 for (dv=devlist ; dv; dv=dv->next) {
 187                         mddev_ident_t mdlist = conf_get_ident(config, dv->devname);
 188                         struct state *st = malloc(sizeof *st);
 189                         if (st == NULL)
 190                                 continue;
 191                         st->devname = strdup(dv->devname);
 192                         st->utime = 0;
 193                         st->next = statelist;
 194                         st->err = 0;
 195                         st->devnum = MAXINT;
 196                         st->percent = -2;
 197                         st->expected_spares = -1;
 198                         st->spare_group = NULL;
 199                         if (mdlist) {
 200                                 st->expected_spares = mdlist->spare_disks;
 201                                 if (mdlist->spare_group)
 202                                         st->spare_group = strdup(mdlist->spare_group);
 203                         }
 204                         statelist = st;
 205                 }
 206         }
 207
 208
 209         while (! finished) {
 210                 int new_found = 0;
 211                 struct state *st;
 212
 213                 if (mdstat)
 214                         free_mdstat(mdstat);
 215                 mdstat = mdstat_read(oneshot?0:1, 0);
 216
 217                 for (st=statelist; st; st=st->next) {
 218                         mdu_array_info_t array;
 219                         struct mdstat_ent *mse = NULL, *mse2;
 220                         char *dev = st->devname;
 221                         int fd;
 222                         unsigned int i;
 223
 224                         if (test)
 225                                 alert("TestMessage", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
 226                         fd = open(dev, O_RDONLY);
 227                         if (fd < 0) {
 228                                 if (!st->err)
 229                                         alert("DeviceDisappeared", dev, NULL,
 230                                               mailaddr, mailfrom, alert_cmd, dosyslog);
 231 /*                                      fprintf(stderr, Name ": cannot open %s: %s\n",
 232                                                 dev, strerror(errno));
 233 */                              st->err=1;
 234                                 continue;
 235                         }
 236                         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 237                                 if (!st->err)
 238                                         alert("DeviceDisappeared", dev, NULL,
 239                                               mailaddr, mailfrom, alert_cmd, dosyslog);
 240 /*                                      fprintf(stderr, Name ": cannot get array info for %s: %s\n",
 241                                                 dev, strerror(errno));
 242 */                              st->err=1;
 243                                 close(fd);
 244                                 continue;
 245                         }
 246                         if (array.level != 1 && array.level != 5 && array.level != -4 &&
 247                                 array.level != 6 && array.level != 10) {
 248                                 if (!st->err)
 249                                         alert("DeviceDisappeared", dev, "Wrong-Level",
 250                                               mailaddr, mailfrom, alert_cmd, dosyslog);
 251                                 st->err = 1;
 252                                 close(fd);
 253                                 continue;
 254                         }
 255                         if (st->devnum == MAXINT) {
 256                                 struct stat stb;
 257                                 if (fstat(fd, &stb) == 0 &&
 258                                     (S_IFMT&stb.st_mode)==S_IFBLK) {
 259                                         if (major(stb.st_rdev) == MD_MAJOR)
 260                                                 st->devnum = minor(stb.st_rdev);
 261                                         else
 262                                                 st->devnum = -1- (minor(stb.st_rdev)>>6);
 263                                 }
 264                         }
 265
 266                         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 267                                 if (mse2->devnum == st->devnum) {
 268                                         mse2->devnum = MAXINT; /* flag it as "used" */
 269                                         mse = mse2;
 270                                 }
 271
 272                         if (st->utime == array.utime &&
 273                             st->failed == array.failed_disks &&
 274                             st->working == array.working_disks &&
 275                             st->spare == array.spare_disks &&
 276                             (mse == NULL  || (
 277                                     mse->percent == st->percent
 278                                     ))) {
 279                                 close(fd);
 280                                 st->err = 0;
 281                                 continue;
 282                         }
 283                         if (st->utime == 0 && /* new array */
 284                             mse &&      /* is in /proc/mdstat */
 285                             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 286                                 )
 287                                 alert("DegradedArray", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
 288
 289                         if (st->utime == 0 && /* new array */
 290                             st->expected_spares > 0 &&
 291                             array.spare_disks < st->expected_spares)
 292                                 alert("SparesMissing", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
 293                         if (mse &&
 294                             st->percent == -1 &&
 295                             mse->percent >= 0)
 296                                 alert("RebuildStarted", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
 297                         if (mse &&
 298                             st->percent >= 0 &&
 299                             mse->percent >= 0 &&
 300                             (mse->percent / 20) > (st->percent / 20))
 301                                 alert(percentalerts[mse->percent/20],
 302                                       dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
 303
 304                         if (mse &&
 305                             mse->percent == -1 &&
 306                             st->percent >= 0)
 307                                 alert("RebuildFinished", dev, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
 308
 309                         if (mse)
 310                                 st->percent = mse->percent;
 311
 312                         for (i=0; i<MaxDisks; i++) {
 313                                 mdu_disk_info_t disc;
 314                                 int newstate=0;
 315                                 int change;
 316                                 char *dv = NULL;
 317                                 disc.number = i;
 318                                 if (i > array.raid_disks + array.nr_disks) {
 319                                         newstate = 0;
 320                                         disc.major = disc.minor = 0;
 321                                 } else if (ioctl(fd, GET_DISK_INFO, &disc)>= 0) {
 322                                         newstate = disc.state;
 323                                         dv = map_dev(disc.major, disc.minor, 1);
 324                                 } else if (mse &&  mse->pattern && i < strlen(mse->pattern)) {
 325                                         switch(mse->pattern[i]) {
 326                                         case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
 327                                         case '_': newstate = 0; break;
 328                                         }
 329                                         disc.major = disc.minor = 0;
 330                                 }
 331                                 if (dv == NULL && st->devid[i])
 332                                         dv = map_dev(major(st->devid[i]),
 333                                                      minor(st->devid[i]), 1);
 334                                 change = newstate ^ st->devstate[i];
 335                                 if (st->utime && change && !st->err) {
 336                                         if (i < (unsigned)array.raid_disks &&
 337                                             (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
 338                                              ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
 339                                              ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
 340                                                 )
 341                                                 alert("Fail", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog);
 342                                         else if (i >= (unsigned)array.raid_disks &&
 343                                                  (disc.major || disc.minor) &&
 344                                                  st->devid[i] == makedev(disc.major, disc.minor) &&
 345                                                  ((newstate&change)&(1<<MD_DISK_FAULTY))
 346                                                 )
 347                                                 alert("FailSpare", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog);
 348                                         else if (i < (unsigned)array.raid_disks &&
 349                                                  (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
 350                                                   ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
 351                                                   ((newstate&change)&(1<<MD_DISK_SYNC)))
 352                                                 )
 353                                                 alert("SpareActive", dev, dv, mailaddr, mailfrom, alert_cmd, dosyslog);
 354                                 }
 355                                 st->devstate[i] = disc.state;
 356                                 st->devid[i] = makedev(disc.major, disc.minor);
 357                         }
 358                         close(fd);
 359                         st->active = array.active_disks;
 360                         st->working = array.working_disks;
 361                         st->spare = array.spare_disks;
 362                         st->failed = array.failed_disks;
 363                         st->utime = array.utime;
 364                         st->raid = array.raid_disks;
 365                         st->err = 0;
 366                 }
 367                 /* now check if there are any new devices found in mdstat */
 368                 if (scan) {
 369                         struct mdstat_ent *mse;
 370                         for (mse=mdstat; mse; mse=mse->next)
 371                                 if (mse->devnum != MAXINT &&
 372                                     (strcmp(mse->level, "raid1")==0 ||
 373                                      strcmp(mse->level, "raid5")==0 ||
 374                                      strcmp(mse->level, "multipath")==0)
 375                                         ) {
 376                                         struct state *st = malloc(sizeof *st);
 377                                         mdu_array_info_t array;
 378                                         int fd;
 379                                         if (st == NULL)
 380                                                 continue;
 381                                         st->devname = strdup(get_md_name(mse->devnum));
 382                                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 383                                             ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
 384                                                 /* no such array */
 385                                                 if (fd >=0) close(fd);
 386                                                 free(st->devname);
 387                                                 free(st);
 388                                                 continue;
 389                                         }
 390                                         close(fd);
 391                                         st->utime = 0;
 392                                         st->next = statelist;
 393                                         st->err = 1;
 394                                         st->devnum = mse->devnum;
 395                                         st->percent = -2;
 396                                         st->spare_group = NULL;
 397                                         st->expected_spares = -1;
 398                                         statelist = st;
 399                                         alert("NewArray", st->devname, NULL, mailaddr, mailfrom, alert_cmd, dosyslog);
 400                                         new_found = 1;
 401                                 }
 402                 }
 403                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 404                  * Look for another array with spare > 0 and active == raid and same spare_group
 405                  *  if found, choose a device and hotremove/hotadd
 406                  */
 407                 for (st = statelist; st; st=st->next)
 408                         if (st->active < st->raid &&
 409                             st->spare == 0 &&
 410                             st->spare_group != NULL) {
 411                                 struct state *st2;
 412                                 for (st2=statelist ; st2 ; st2=st2->next)
 413                                         if (st2 != st &&
 414                                             st2->spare > 0 &&
 415                                             st2->active == st2->raid &&
 416                                             st2->spare_group != NULL &&
 417                                             strcmp(st->spare_group, st2->spare_group) == 0) {
 418                                                 /* try to remove and add */
 419                                                 int fd1 = open(st->devname, O_RDONLY);
 420                                                 int fd2 = open(st2->devname, O_RDONLY);
 421                                                 int dev = -1;
 422                                                 int d;
 423                                                 if (fd1 < 0 || fd2 < 0) {
 424                                                         if (fd1>=0) close(fd1);
 425                                                         if (fd2>=0) close(fd2);
 426                                                         continue;
 427                                                 }
 428                                                 for (d=st2->raid; d < MaxDisks; d++) {
 429                                                         if (st2->devid[d] > 0 &&
 430                                                             st2->devstate[d] == 0) {
 431                                                                 dev = st2->devid[d];
 432                                                                 break;
 433                                                         }
 434                                                 }
 435                                                 if (dev > 0) {
 436                                                         if (ioctl(fd2, HOT_REMOVE_DISK,
 437                                                                   (unsigned long)dev) == 0) {
 438                                                                 if (ioctl(fd1, HOT_ADD_DISK,
 439                                                                           (unsigned long)dev) == 0) {
 440                                                                         alert("MoveSpare", st->devname, st2->devname, mailaddr, mailfrom, alert_cmd, dosyslog);
 441                                                                         close(fd1);
 442                                                                         close(fd2);
 443                                                                         break;
 444                                                                 }
 445                                                                 else ioctl(fd2, HOT_ADD_DISK, (unsigned long) dev);
 446                                                         }
 447                                                 }
 448                                                 close(fd1);
 449                                                 close(fd2);
 450                                         }
 451                         }
 452                 if (!new_found) {
 453                         if (oneshot)
 454                                 break;
 455                         else
 456                                 mdstat_wait(period);
 457                 }
 458                 test = 0;
 459         }
 460         if (pidfile)
 461                 unlink(pidfile);
 462         return 0;
 463 }
 464
 465
 466 static void alert(char *event, char *dev, char *disc, char *mailaddr, char *mailfrom, char *cmd,
 467                   int dosyslog)
 468 {
 469         int priority;
 470
 471         if (!cmd && !mailaddr) {
 472                 time_t now = time(0);
 473
 474                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 475         }
 476         if (cmd) {
 477                 int pid = fork();
 478                 switch(pid) {
 479                 default:
 480                         waitpid(pid, NULL, 0);
 481                         break;
 482                 case -1:
 483                         break;
 484                 case 0:
 485                         execl(cmd, cmd, event, dev, disc, NULL);
 486                         exit(2);
 487                 }
 488         }
 489         if (mailaddr &&
 490             (strncmp(event, "Fail", 4)==0 ||
 491              strncmp(event, "Test", 4)==0 ||
 492              strncmp(event, "Spares", 6)==0 ||
 493              strncmp(event, "Degrade", 7)==0)) {
 494                 FILE *mp = popen(Sendmail, "w");
 495                 if (mp) {
 496                         FILE *mdstat;
 497                         char hname[256];
 498                         gethostname(hname, sizeof(hname));
 499                         signal(SIGPIPE, SIG_IGN);
 500                         if (mailfrom)
 501                                 fprintf(mp, "From: %s\n", mailfrom);
 502                         else
 503                                 fprintf(mp, "From: " Name " monitoring <root>\n");
 504                         fprintf(mp, "To: %s\n", mailaddr);
 505                         fprintf(mp, "Subject: %s event on %s:%s\n\n", event, dev, hname);
 506
 507                         fprintf(mp, "This is an automatically generated mail message from " Name "\n");
 508                         fprintf(mp, "running on %s\n\n", hname);
 509
 510                         fprintf(mp, "A %s event had been detected on md device %s.\n\n", event, dev);
 511
 512                         if (disc)
 513                                 fprintf(mp, "It could be related to component device %s.\n\n", disc);
 514
 515                         fprintf(mp, "Faithfully yours, etc.\n");
 516
 517                         mdstat = fopen("/proc/mdstat", "r");
 518                         if (mdstat) {
 519                                 char buf[8192];
 520                                 int n;
 521                                 fprintf(mp, "\nP.S. The /proc/mdstat file current contains the following:\n\n");
 522                                 while ( (n=fread(buf, 1, sizeof(buf), mdstat)) > 0)
 523                                         fwrite(buf, 1, n, mp);
 524                                 fclose(mdstat);
 525                         }
 526                         fclose(mp);
 527                 }
 528
 529         }
 530
 531         /* log the event to syslog maybe */
 532         if (dosyslog) {
 533                 /* Log at a different severity depending on the event.
 534                  *
 535                  * These are the critical events:  */
 536                 if (strncmp(event, "Fail", 4)==0 ||
 537                     strncmp(event, "Degrade", 7)==0 ||
 538                     strncmp(event, "DeviceDisappeared", 17)==0)
 539                         priority = LOG_CRIT;
 540                 /* Good to know about, but are not failures: */
 541                 else if (strncmp(event, "Rebuild", 7)==0 ||
 542                          strncmp(event, "MoveSpare", 9)==0 ||
 543                          strncmp(event, "Spares", 6) != 0)
 544                         priority = LOG_WARNING;
 545                 /* Everything else: */
 546                 else
 547                         priority = LOG_INFO;
 548
 549                 if (disc)
 550                         syslog(priority, "%s event detected on md device %s, component device %s", event, dev, disc);
 551                 else
 552                         syslog(priority, "%s event detected on md device %s", event, dev);
 553         }
 554 }