Monitor.c

   1 /*
   2  * mdadm - manage Linux "md" devices aka RAID arrays.
   3  *
   4  * Copyright (C) 2001-2002 Neil Brown <neilb@cse.unsw.edu.au>
   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 *cmd,
  39                   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
 117         if (!mailaddr) {
 118                 mailaddr = conf_get_mailaddr(config);
 119                 if (mailaddr && ! scan)
 120                         fprintf(stderr, Name ": Monitor using email address \"%s\" from config file\n",
 121                                mailaddr);
 122         }
 123         if (!alert_cmd) {
 124                 alert_cmd = conf_get_program(config);
 125                 if (alert_cmd && ! scan)
 126                         fprintf(stderr, Name ": Monitor using program \"%s\" from config file\n",
 127                                alert_cmd);
 128         }
 129         if (scan && !mailaddr && !alert_cmd) {
 130                 fprintf(stderr, Name ": No mail address or alert command - not monitoring.\n");
 131                 return 1;
 132         }
 133
 134         if (daemonise) {
 135                 int pid = fork();
 136                 if (pid > 0) {
 137                         if (!pidfile)
 138                                 printf("%d\n", pid);
 139                         else {
 140                                 FILE *pid_file;
 141                                 pid_file=fopen(pidfile, "w");
 142                                 if (!pid_file)
 143                                         perror("cannot create pid file");
 144                                 else {
 145                                         fprintf(pid_file,"%d\n", pid);
 146                                         fclose(pid_file);
 147                                 }
 148                         }
 149                         return 0;
 150                 }
 151                 if (pid < 0) {
 152                         perror("daemonise");
 153                         return 1;
 154                 }
 155                 close(0);
 156                 open("/dev/null", 3);
 157                 dup2(0,1);
 158                 dup2(0,2);
 159                 setsid();
 160         }
 161
 162         if (devlist == NULL) {
 163                 mddev_ident_t mdlist = conf_get_ident(config, NULL);
 164                 for (; mdlist; mdlist=mdlist->next) {
 165                         struct state *st = malloc(sizeof *st);
 166                         if (st == NULL)
 167                                 continue;
 168                         st->devname = strdup(mdlist->devname);
 169                         st->utime = 0;
 170                         st->next = statelist;
 171                         st->err = 0;
 172                         st->devnum = MAXINT;
 173                         st->percent = -2;
 174                         st->expected_spares = mdlist->spare_disks;
 175                         if (mdlist->spare_group)
 176                                 st->spare_group = strdup(mdlist->spare_group);
 177                         else
 178                                 st->spare_group = NULL;
 179                         statelist = st;
 180                 }
 181         } else {
 182                 mddev_dev_t dv;
 183                 for (dv=devlist ; dv; dv=dv->next) {
 184                         mddev_ident_t mdlist = conf_get_ident(config, dv->devname);
 185                         struct state *st = malloc(sizeof *st);
 186                         if (st == NULL)
 187                                 continue;
 188                         st->devname = strdup(dv->devname);
 189                         st->utime = 0;
 190                         st->next = statelist;
 191                         st->err = 0;
 192                         st->devnum = MAXINT;
 193                         st->percent = -2;
 194                         st->expected_spares = -1;
 195                         st->spare_group = NULL;
 196                         if (mdlist) {
 197                                 st->expected_spares = mdlist->spare_disks;
 198                                 if (mdlist->spare_group)
 199                                         st->spare_group = strdup(mdlist->spare_group);
 200                         }
 201                         statelist = st;
 202                 }
 203         }
 204
 205
 206         while (! finished) {
 207                 int new_found = 0;
 208                 struct state *st;
 209
 210                 if (mdstat)
 211                         free_mdstat(mdstat);
 212                 mdstat = mdstat_read(oneshot?0:1, 0);
 213
 214                 for (st=statelist; st; st=st->next) {
 215                         mdu_array_info_t array;
 216                         struct mdstat_ent *mse = NULL, *mse2;
 217                         char *dev = st->devname;
 218                         int fd;
 219                         unsigned int i;
 220
 221                         if (test)
 222                                 alert("TestMessage", dev, NULL, mailaddr, alert_cmd, dosyslog);
 223                         fd = open(dev, O_RDONLY);
 224                         if (fd < 0) {
 225                                 if (!st->err)
 226                                         alert("DeviceDisappeared", dev, NULL,
 227                                               mailaddr, alert_cmd, dosyslog);
 228 /*                                      fprintf(stderr, Name ": cannot open %s: %s\n",
 229                                                 dev, strerror(errno));
 230 */                              st->err=1;
 231                                 continue;
 232                         }
 233                         if (ioctl(fd, GET_ARRAY_INFO, &array)<0) {
 234                                 if (!st->err)
 235                                         alert("DeviceDisappeared", dev, NULL,
 236                                               mailaddr, alert_cmd, dosyslog);
 237 /*                                      fprintf(stderr, Name ": cannot get array info for %s: %s\n",
 238                                                 dev, strerror(errno));
 239 */                              st->err=1;
 240                                 close(fd);
 241                                 continue;
 242                         }
 243                         if (array.level != 1 && array.level != 5 && array.level != -4 &&
 244                                 array.level != 6 && array.level != 10) {
 245                                 if (!st->err)
 246                                         alert("DeviceDisappeared", dev, "Wrong-Level",
 247                                               mailaddr, alert_cmd, dosyslog);
 248                                 st->err = 1;
 249                                 close(fd);
 250                                 continue;
 251                         }
 252                         if (st->devnum == MAXINT) {
 253                                 struct stat stb;
 254                                 if (fstat(fd, &stb) == 0 &&
 255                                     (S_IFMT&stb.st_mode)==S_IFBLK) {
 256                                         if (major(stb.st_rdev) == MD_MAJOR)
 257                                                 st->devnum = minor(stb.st_rdev);
 258                                         else
 259                                                 st->devnum = -1- (minor(stb.st_rdev)>>6);
 260                                 }
 261                         }
 262
 263                         for (mse2 = mdstat ; mse2 ; mse2=mse2->next)
 264                                 if (mse2->devnum == st->devnum) {
 265                                         mse2->devnum = MAXINT; /* flag it as "used" */
 266                                         mse = mse2;
 267                                 }
 268
 269                         if (st->utime == array.utime &&
 270                             st->failed == array.failed_disks &&
 271                             st->working == array.working_disks &&
 272                             st->spare == array.spare_disks &&
 273                             (mse == NULL  || (
 274                                     mse->percent == st->percent
 275                                     ))) {
 276                                 close(fd);
 277                                 st->err = 0;
 278                                 continue;
 279                         }
 280                         if (st->utime == 0 && /* new array */
 281                             mse &&      /* is in /proc/mdstat */
 282                             mse->pattern && strchr(mse->pattern, '_') /* degraded */
 283                                 )
 284                                 alert("DegradedArray", dev, NULL, mailaddr, alert_cmd, dosyslog);
 285
 286                         if (st->utime == 0 && /* new array */
 287                             st->expected_spares > 0 &&
 288                             array.spare_disks < st->expected_spares)
 289                                 alert("SparesMissing", dev, NULL, mailaddr, alert_cmd, dosyslog);
 290                         if (mse &&
 291                             st->percent == -1 &&
 292                             mse->percent >= 0)
 293                                 alert("RebuildStarted", dev, NULL, mailaddr, alert_cmd, dosyslog);
 294                         if (mse &&
 295                             st->percent >= 0 &&
 296                             mse->percent >= 0 &&
 297                             (mse->percent / 20) > (st->percent / 20))
 298                                 alert(percentalerts[mse->percent/20],
 299                                       dev, NULL, mailaddr, alert_cmd, dosyslog);
 300
 301                         if (mse &&
 302                             mse->percent == -1 &&
 303                             st->percent >= 0)
 304                                 alert("RebuildFinished", dev, NULL, mailaddr, alert_cmd, dosyslog);
 305
 306                         if (mse)
 307                                 st->percent = mse->percent;
 308
 309                         for (i=0; i<MaxDisks; i++) {
 310                                 mdu_disk_info_t disc;
 311                                 int newstate=0;
 312                                 int change;
 313                                 char *dv = NULL;
 314                                 disc.number = i;
 315                                 if (i > array.raid_disks + array.nr_disks) {
 316                                         newstate = 0;
 317                                         disc.major = disc.minor = 0;
 318                                 } else if (ioctl(fd, GET_DISK_INFO, &disc)>= 0) {
 319                                         newstate = disc.state;
 320                                         dv = map_dev(disc.major, disc.minor, 1);
 321                                 } else if (mse &&  mse->pattern && i < strlen(mse->pattern)) {
 322                                         switch(mse->pattern[i]) {
 323                                         case 'U': newstate = 6 /* ACTIVE/SYNC */; break;
 324                                         case '_': newstate = 0; break;
 325                                         }
 326                                         disc.major = disc.minor = 0;
 327                                 }
 328                                 if (dv == NULL && st->devid[i])
 329                                         dv = map_dev(major(st->devid[i]),
 330                                                      minor(st->devid[i]), 1);
 331                                 change = newstate ^ st->devstate[i];
 332                                 if (st->utime && change && !st->err) {
 333                                         if (i < (unsigned)array.raid_disks &&
 334                                             (((newstate&change)&(1<<MD_DISK_FAULTY)) ||
 335                                              ((st->devstate[i]&change)&(1<<MD_DISK_ACTIVE)) ||
 336                                              ((st->devstate[i]&change)&(1<<MD_DISK_SYNC)))
 337                                                 )
 338                                                 alert("Fail", dev, dv, mailaddr, alert_cmd, dosyslog);
 339                                         else if (i >= (unsigned)array.raid_disks &&
 340                                                  (disc.major || disc.minor) &&
 341                                                  st->devid[i] == makedev(disc.major, disc.minor) &&
 342                                                  ((newstate&change)&(1<<MD_DISK_FAULTY))
 343                                                 )
 344                                                 alert("FailSpare", dev, dv, mailaddr, alert_cmd, dosyslog);
 345                                         else if (i < (unsigned)array.raid_disks &&
 346                                                  (((st->devstate[i]&change)&(1<<MD_DISK_FAULTY)) ||
 347                                                   ((newstate&change)&(1<<MD_DISK_ACTIVE)) ||
 348                                                   ((newstate&change)&(1<<MD_DISK_SYNC)))
 349                                                 )
 350                                                 alert("SpareActive", dev, dv, mailaddr, alert_cmd, dosyslog);
 351                                 }
 352                                 st->devstate[i] = disc.state;
 353                                 st->devid[i] = makedev(disc.major, disc.minor);
 354                         }
 355                         close(fd);
 356                         st->active = array.active_disks;
 357                         st->working = array.working_disks;
 358                         st->spare = array.spare_disks;
 359                         st->failed = array.failed_disks;
 360                         st->utime = array.utime;
 361                         st->raid = array.raid_disks;
 362                         st->err = 0;
 363                 }
 364                 /* now check if there are any new devices found in mdstat */
 365                 if (scan) {
 366                         struct mdstat_ent *mse;
 367                         for (mse=mdstat; mse; mse=mse->next)
 368                                 if (mse->devnum != MAXINT &&
 369                                     (strcmp(mse->level, "raid1")==0 ||
 370                                      strcmp(mse->level, "raid5")==0 ||
 371                                      strcmp(mse->level, "multipath")==0)
 372                                         ) {
 373                                         struct state *st = malloc(sizeof *st);
 374                                         mdu_array_info_t array;
 375                                         int fd;
 376                                         if (st == NULL)
 377                                                 continue;
 378                                         st->devname = strdup(get_md_name(mse->devnum));
 379                                         if ((fd = open(st->devname, O_RDONLY)) < 0 ||
 380                                             ioctl(fd, GET_ARRAY_INFO, &array)< 0) {
 381                                                 /* no such array */
 382                                                 if (fd >=0) close(fd);
 383                                                 free(st->devname);
 384                                                 free(st);
 385                                                 continue;
 386                                         }
 387                                         close(fd);
 388                                         st->utime = 0;
 389                                         st->next = statelist;
 390                                         st->err = 1;
 391                                         st->devnum = mse->devnum;
 392                                         st->percent = -2;
 393                                         st->spare_group = NULL;
 394                                         st->expected_spares = -1;
 395                                         statelist = st;
 396                                         alert("NewArray", st->devname, NULL, mailaddr, alert_cmd, dosyslog);
 397                                         new_found = 1;
 398                                 }
 399                 }
 400                 /* If an array has active < raid && spare == 0 && spare_group != NULL
 401                  * Look for another array with spare > 0 and active == raid and same spare_group
 402                  *  if found, choose a device and hotremove/hotadd
 403                  */
 404                 for (st = statelist; st; st=st->next)
 405                         if (st->active < st->raid &&
 406                             st->spare == 0 &&
 407                             st->spare_group != NULL) {
 408                                 struct state *st2;
 409                                 for (st2=statelist ; st2 ; st2=st2->next)
 410                                         if (st2 != st &&
 411                                             st2->spare > 0 &&
 412                                             st2->active == st2->raid &&
 413                                             st2->spare_group != NULL &&
 414                                             strcmp(st->spare_group, st2->spare_group) == 0) {
 415                                                 /* try to remove and add */
 416                                                 int fd1 = open(st->devname, O_RDONLY);
 417                                                 int fd2 = open(st2->devname, O_RDONLY);
 418                                                 int dev = -1;
 419                                                 int d;
 420                                                 if (fd1 < 0 || fd2 < 0) {
 421                                                         if (fd1>=0) close(fd1);
 422                                                         if (fd2>=0) close(fd2);
 423                                                         continue;
 424                                                 }
 425                                                 for (d=st2->raid; d < MaxDisks; d++) {
 426                                                         if (st2->devid[d] > 0 &&
 427                                                             st2->devstate[d] == 0) {
 428                                                                 dev = st2->devid[d];
 429                                                                 break;
 430                                                         }
 431                                                 }
 432                                                 if (dev > 0) {
 433                                                         if (ioctl(fd2, HOT_REMOVE_DISK,
 434                                                                   (unsigned long)dev) == 0) {
 435                                                                 if (ioctl(fd1, HOT_ADD_DISK,
 436                                                                           (unsigned long)dev) == 0) {
 437                                                                         alert("MoveSpare", st->devname, st2->devname, mailaddr, alert_cmd, dosyslog);
 438                                                                         close(fd1);
 439                                                                         close(fd2);
 440                                                                         break;
 441                                                                 }
 442                                                                 else ioctl(fd2, HOT_ADD_DISK, (unsigned long) dev);
 443                                                         }
 444                                                 }
 445                                                 close(fd1);
 446                                                 close(fd2);
 447                                         }
 448                         }
 449                 if (!new_found) {
 450                         if (oneshot)
 451                                 break;
 452                         else
 453                                 mdstat_wait(period);
 454                 }
 455                 test = 0;
 456         }
 457         if (pidfile)
 458                 unlink(pidfile);
 459         return 0;
 460 }
 461
 462
 463 static void alert(char *event, char *dev, char *disc, char *mailaddr, char *cmd,
 464                   int dosyslog)
 465 {
 466         int priority;
 467
 468         if (!cmd && !mailaddr) {
 469                 time_t now = time(0);
 470
 471                 printf("%1.15s: %s on %s %s\n", ctime(&now)+4, event, dev, disc?disc:"unknown device");
 472         }
 473         if (cmd) {
 474                 int pid = fork();
 475                 switch(pid) {
 476                 default:
 477                         waitpid(pid, NULL, 0);
 478                         break;
 479                 case -1:
 480                         break;
 481                 case 0:
 482                         execl(cmd, cmd, event, dev, disc, NULL);
 483                         exit(2);
 484                 }
 485         }
 486         if (mailaddr &&
 487             (strncmp(event, "Fail", 4)==0 ||
 488              strncmp(event, "Test", 4)==0 ||
 489              strncmp(event, "Degrade", 7)==0)) {
 490                 FILE *mp = popen(Sendmail, "w");
 491                 if (mp) {
 492                         char hname[256];
 493                         gethostname(hname, sizeof(hname));
 494                         signal(SIGPIPE, SIG_IGN);
 495                         fprintf(mp, "From: " Name " monitoring <root>\n");
 496                         fprintf(mp, "To: %s\n", mailaddr);
 497                         fprintf(mp, "Subject: %s event on %s:%s\n\n", event, dev, hname);
 498
 499                         fprintf(mp, "This is an automatically generated mail message from " Name "\n");
 500                         fprintf(mp, "running on %s\n\n", hname);
 501
 502                         fprintf(mp, "A %s event had been detected on md device %s.\n\n", event, dev);
 503
 504                         if (disc)
 505                                 fprintf(mp, "It could be related to component device %s.\n\n", disc);
 506
 507                         fprintf(mp, "Faithfully yours, etc.\n");
 508                         fclose(mp);
 509                 }
 510
 511         }
 512
 513         /* log the event to syslog maybe */
 514         if (dosyslog) {
 515                 /* Log at a different severity depending on the event.
 516                  *
 517                  * These are the critical events:  */
 518                 if (strncmp(event, "Fail", 4)==0 ||
 519                     strncmp(event, "Degrade", 7)==0 ||
 520                     strncmp(event, "DeviceDisappeared", 17)==0)
 521                         priority = LOG_CRIT;
 522                 /* Good to know about, but are not failures: */
 523                 else if (strncmp(event, "Rebuild", 7)==0 ||
 524                          strncmp(event, "MoveSpare", 9)==0)
 525                         priority = LOG_WARNING;
 526                 /* Everything else: */
 527                 else
 528                         priority = LOG_INFO;
 529
 530                 if (disc)
 531                         syslog(priority, "%s event detected on md device %s, component device %s", event, dev, disc);
 532                 else
 533                         syslog(priority, "%s event detected on md device %s", event, dev);
 534         }
 535 }