Monitor.c

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