multipathd/main.c

   1 #include <unistd.h>
   2 #include <sys/stat.h>
   3 #include <libdevmapper.h>
   4 #include <wait.h>
   5 #include <sys/mman.h>
   6
   7 /*
   8  * libsysfs
   9  */
  10 #include <sysfs/libsysfs.h>
  11 #include <sysfs/dlist.h>
  12
  13 /*
  14  * libcheckers
  15  */
  16 #include <checkers.h>
  17 #include <path_state.h>
  18
  19 /*
  20  * libmultipath
  21  */
  22 #include <parser.h>
  23 #include <vector.h>
  24 #include <memory.h>
  25 #include <config.h>
  26 #include <callout.h>
  27 #include <util.h>
  28 #include <blacklist.h>
  29 #include <hwtable.h>
  30 #include <defaults.h>
  31 #include <structs.h>
  32 #include <dmparser.h>
  33 #include <devmapper.h>
  34 #include <dict.h>
  35 #include <discovery.h>
  36 #include <debug.h>
  37 #include <propsel.h>
  38 #include <uevent.h>
  39 #include <switchgroup.h>
  40 #include <path_state.h>
  41 #include <print.h>
  42
  43 #include "main.h"
  44 #include "pidfile.h"
  45 #include "uxlsnr.h"
  46 #include "uxclnt.h"
  47 #include "cli.h"
  48 #include "cli_handlers.h"
  49
  50 #define FILE_NAME_SIZE 256
  51 #define CMDSIZE 160
  52
  53 #define LOG_MSG(a,b) \
  54         if (strlen(b)) { \
  55                 condlog(a, "%s: %s", pp->dev_t, b); \
  56                 memset(b, 0, MAX_CHECKER_MSG_SIZE); \
  57         }
  58
  59 #ifdef LCKDBG
  60 #define lock(a) \
  61         fprintf(stderr, "%s:%s(%i) lock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  62         pthread_mutex_lock(a)
  63 #define unlock(a) \
  64         fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  65         pthread_mutex_unlock(a)
  66 #define lock_cleanup_pop(a) \
  67         fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  68         pthread_cleanup_pop(1);
  69 #else
  70 #define lock(a) pthread_mutex_lock(a)
  71 #define unlock(a) pthread_mutex_unlock(a)
  72 #define lock_cleanup_pop(a) pthread_cleanup_pop(1);
  73 #endif
  74
  75 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  76 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  77
  78 /*
  79  * structs
  80  */
  81 struct event_thread {
  82         struct dm_task *dmt;
  83         pthread_t thread;
  84         int event_nr;
  85         char mapname[WWID_SIZE];
  86         struct vectors *vecs;
  87 };
  88
  89 static struct event_thread *
  90 alloc_waiter (void)
  91 {
  92
  93         struct event_thread * wp;
  94
  95         wp = (struct event_thread *)MALLOC(sizeof(struct event_thread));
  96
  97         return wp;
  98 }
  99
 100 static void
 101 cleanup_lock (void * data)
 102 {
 103         pthread_mutex_unlock((pthread_mutex_t *)data);
 104 }
 105
 106 static void
 107 adopt_paths (struct vectors * vecs, struct multipath * mpp)
 108 {
 109         int i;
 110         struct path * pp;
 111
 112         if (!mpp)
 113                 return;
 114
 115         vector_foreach_slot (vecs->pathvec, pp, i) {
 116                 if (!strncmp(mpp->wwid, pp->wwid, WWID_SIZE)) {
 117                         condlog(4, "%s ownership set", pp->dev_t);
 118                         pp->mpp = mpp;
 119                 }
 120         }
 121 }
 122
 123 static void
 124 orphan_path (struct path * pp)
 125 {
 126         pp->mpp = NULL;
 127         pp->checkfn = NULL;
 128         pp->dmstate = PSTATE_UNDEF;
 129         pp->checker_context = NULL;
 130         pp->getuid = NULL;
 131         pp->getprio = NULL;
 132
 133         if (pp->fd >= 0)
 134                 close(pp->fd);
 135
 136         pp->fd = -1;
 137 }
 138
 139 static void
 140 orphan_paths (struct vectors * vecs, struct multipath * mpp)
 141 {
 142         int i;
 143         struct path * pp;
 144
 145         vector_foreach_slot (vecs->pathvec, pp, i) {
 146                 if (pp->mpp == mpp) {
 147                         condlog(4, "%s is orphaned", pp->dev_t);
 148                         orphan_path(pp);
 149                 }
 150         }
 151 }
 152
 153 static int
 154 update_multipath_table (struct multipath *mpp, vector pathvec)
 155 {
 156         if (!mpp)
 157                 return 1;
 158
 159         if (dm_get_map(mpp->alias, &mpp->size, mpp->params))
 160                 return 1;
 161
 162         if (disassemble_map(pathvec, mpp->params, mpp))
 163                 return 1;
 164
 165         return 0;
 166 }
 167
 168 static int
 169 update_multipath_status (struct multipath *mpp)
 170 {
 171         if (!mpp)
 172                 return 1;
 173
 174         if(dm_get_status(mpp->alias, mpp->status))
 175                 return 1;
 176
 177         if (disassemble_status(mpp->status, mpp))
 178                 return 1;
 179
 180         return 0;
 181 }
 182
 183 static int
 184 update_multipath_strings (struct multipath *mpp, vector pathvec)
 185 {
 186         if (mpp->selector) {
 187                 FREE(mpp->selector);
 188                 mpp->selector = NULL;
 189         }
 190
 191         if (mpp->features) {
 192                 FREE(mpp->features);
 193                 mpp->features = NULL;
 194         }
 195
 196         if (mpp->hwhandler) {
 197                 FREE(mpp->hwhandler);
 198                 mpp->hwhandler = NULL;
 199         }
 200
 201         free_pgvec(mpp->pg, KEEP_PATHS);
 202         mpp->pg = NULL;
 203
 204         if (update_multipath_table(mpp, pathvec))
 205                 return 1;
 206
 207         if (update_multipath_status(mpp))
 208                 return 1;
 209
 210         return 0;
 211 }
 212
 213 static void
 214 set_multipath_wwid (struct multipath * mpp)
 215 {
 216         if (mpp->wwid)
 217                 return;
 218
 219         dm_get_uuid(mpp->alias, mpp->wwid);
 220 }
 221
 222 static int
 223 pathcount (struct multipath *mpp, int state)
 224 {
 225         struct pathgroup *pgp;
 226         struct path *pp;
 227         int i, j;
 228         int count = 0;
 229
 230         vector_foreach_slot (mpp->pg, pgp, i)
 231                 vector_foreach_slot (pgp->paths, pp, j)
 232                         if (pp->state == state)
 233                                 count++;
 234         return count;
 235 }
 236
 237 /*
 238  * mpp->no_path_retry:
 239  *   -2 (QUEUE) : queue_if_no_path enabled, never turned off
 240  *   -1 (FAIL)  : fail_if_no_path
 241  *    0 (UNDEF) : nothing
 242  *   >0         : queue_if_no_path enabled, turned off after polling n times
 243  */
 244 static void
 245 update_queue_mode_del_path(struct multipath *mpp)
 246 {
 247         if (--mpp->nr_active == 0 && mpp->no_path_retry > 0) {
 248                 /*
 249                  * Enter retry mode.
 250                  * meaning of +1: retry_tick may be decremented in
 251                  *                checkerloop before starting retry.
 252                  */
 253                 mpp->retry_tick = mpp->no_path_retry * conf->checkint + 1;
 254                 condlog(1, "%s: Entering recovery mode: max_retries=%d",
 255                         mpp->alias, mpp->no_path_retry);
 256         }
 257         condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
 258 }
 259
 260 static void
 261 update_queue_mode_add_path(struct multipath *mpp)
 262 {
 263         if (mpp->nr_active++ == 0 && mpp->no_path_retry > 0) {
 264                 /* come back to normal mode from retry mode */
 265                 mpp->retry_tick = 0;
 266                 dm_queue_if_no_path(mpp->alias, 1);
 267                 condlog(2, "%s: queue_if_no_path enabled", mpp->alias);
 268                 condlog(1, "%s: Recovered to normal mode", mpp->alias);
 269         }
 270         condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
 271 }
 272
 273 static void
 274 set_no_path_retry(struct multipath *mpp)
 275 {
 276         mpp->retry_tick = 0;
 277         mpp->nr_active = pathcount(mpp, PATH_UP);
 278         select_no_path_retry(mpp);
 279
 280         switch (mpp->no_path_retry) {
 281         case NO_PATH_RETRY_UNDEF:
 282                 break;
 283         case NO_PATH_RETRY_FAIL:
 284                 dm_queue_if_no_path(mpp->alias, 0);
 285                 break;
 286         case NO_PATH_RETRY_QUEUE:
 287                 dm_queue_if_no_path(mpp->alias, 1);
 288                 break;
 289         default:
 290                 dm_queue_if_no_path(mpp->alias, 1);
 291                 if (mpp->nr_active == 0) {
 292                         /* Enter retry mode */
 293                         mpp->retry_tick = mpp->no_path_retry * conf->checkint;
 294                         condlog(1, "%s: Entering recovery mode: max_retries=%d",
 295                                 mpp->alias, mpp->no_path_retry);
 296                 }
 297                 break;
 298         }
 299 }
 300
 301 static struct hwentry *
 302 extract_hwe_from_path(struct multipath * mpp)
 303 {
 304         struct path * pp;
 305         struct pathgroup * pgp;
 306
 307         pgp = VECTOR_SLOT(mpp->pg, 0);
 308         pp = VECTOR_SLOT(pgp->paths, 0);
 309
 310         return pp->hwe;
 311 }
 312
 313 static int
 314 setup_multipath (struct vectors * vecs, struct multipath * mpp)
 315 {
 316         int i;
 317
 318         set_multipath_wwid(mpp);
 319         mpp->mpe = find_mpe(mpp->wwid);
 320         condlog(4, "discovered map %s", mpp->alias);
 321
 322         if (update_multipath_strings(mpp, vecs->pathvec))
 323                 goto out;
 324
 325         adopt_paths(vecs, mpp);
 326         select_pgfailback(mpp);
 327         mpp->hwe = extract_hwe_from_path(mpp);
 328         set_no_path_retry(mpp);
 329
 330         return 0;
 331 out:
 332         /*
 333          * purge the multipath vector
 334          */
 335         if ((i = find_slot(vecs->mpvec, (void *)mpp)) != -1)
 336                 vector_del_slot(vecs->mpvec, i);
 337
 338         free_multipath(mpp, KEEP_PATHS);
 339         condlog(0, "failed to setup multipath");
 340         return 1;
 341 }
 342
 343 static int
 344 need_switch_pathgroup (struct multipath * mpp, int refresh)
 345 {
 346         struct pathgroup * pgp;
 347         struct path * pp;
 348         int i, j;
 349
 350         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
 351                 return 0;
 352
 353         /*
 354          * Refresh path priority values
 355          */
 356         if (refresh)
 357                 vector_foreach_slot (mpp->pg, pgp, i)
 358                         vector_foreach_slot (pgp->paths, pp, j)
 359                                 pathinfo(pp, conf->hwtable, DI_PRIO);
 360
 361         select_path_group(mpp); /* sets mpp->nextpg */
 362         pgp = VECTOR_SLOT(mpp->pg, mpp->nextpg - 1);
 363
 364         if (pgp && pgp->status != PGSTATE_ACTIVE)
 365                 return 1;
 366
 367         return 0;
 368 }
 369
 370 static void
 371 switch_pathgroup (struct multipath * mpp)
 372 {
 373         struct pathgroup * pgp;
 374
 375         pgp = VECTOR_SLOT(mpp->pg, mpp->nextpg - 1);
 376
 377         if (pgp && pgp->status != PGSTATE_ACTIVE) {
 378                 dm_switchgroup(mpp->alias, mpp->nextpg);
 379                 condlog(2, "%s: switch to path group #%i",
 380                          mpp->alias, mpp->nextpg);
 381         }
 382 }
 383
 384 static int
 385 update_multipath (struct vectors *vecs, char *mapname)
 386 {
 387         struct multipath *mpp;
 388         struct pathgroup  *pgp;
 389         struct path *pp;
 390         int i, j;
 391         int r = 1;
 392
 393         mpp = find_mp(vecs->mpvec, mapname);
 394
 395         if (!mpp)
 396                 goto out;
 397
 398         free_pgvec(mpp->pg, KEEP_PATHS);
 399         mpp->pg = NULL;
 400
 401         if (setup_multipath(vecs, mpp))
 402                 goto out; /* mpp freed in setup_multipath */
 403
 404         /*
 405          * compare checkers states with DM states
 406          */
 407         vector_foreach_slot (mpp->pg, pgp, i) {
 408                 vector_foreach_slot (pgp->paths, pp, j) {
 409                         if (pp->dmstate != PSTATE_FAILED)
 410                                 continue;
 411
 412                         if (pp->state != PATH_DOWN) {
 413                                 condlog(2, "%s: mark as failed", pp->dev_t);
 414                                 pp->state = PATH_DOWN;
 415                                 update_queue_mode_del_path(mpp);
 416
 417                                 /*
 418                                  * if opportune,
 419                                  * schedule the next check earlier
 420                                  */
 421                                 if (pp->tick > conf->checkint)
 422                                         pp->tick = conf->checkint;
 423                         }
 424                 }
 425         }
 426         r = 0;
 427 out:
 428         if (r)
 429                 condlog(0, "failed to update multipath");
 430
 431         return r;
 432 }
 433
 434 static void
 435 free_waiter (void * data)
 436 {
 437         struct event_thread * wp = (struct event_thread *)data;
 438
 439         if (wp->dmt)
 440                 dm_task_destroy(wp->dmt);
 441         FREE(wp);
 442 }
 443
 444 static sigset_t unblock_sighup(void)
 445 {
 446         sigset_t set, old;
 447
 448         sigemptyset(&set);
 449         sigaddset(&set, SIGHUP);
 450         pthread_sigmask(SIG_UNBLOCK, &set, &old);
 451         return old;
 452 }
 453
 454 /*
 455  * returns the reschedule delay
 456  * negative means *stop*
 457  */
 458 static int
 459 waiteventloop (struct event_thread * waiter)
 460 {
 461         sigset_t set;
 462         int event_nr;
 463         int r;
 464
 465         if (!waiter->event_nr)
 466                 waiter->event_nr = dm_geteventnr(waiter->mapname);
 467
 468         if (!(waiter->dmt = dm_task_create(DM_DEVICE_WAITEVENT)))
 469                 return 1;
 470
 471         if (!dm_task_set_name(waiter->dmt, waiter->mapname))
 472                 return 1;
 473
 474         if (waiter->event_nr && !dm_task_set_event_nr(waiter->dmt,
 475                                                       waiter->event_nr))
 476                 return 1;
 477
 478         dm_task_no_open_count(waiter->dmt);
 479
 480         set = unblock_sighup();
 481         dm_task_run(waiter->dmt);
 482         pthread_sigmask(SIG_SETMASK, &set, NULL);
 483         pthread_testcancel();
 484         dm_task_destroy(waiter->dmt);
 485         waiter->dmt = NULL;
 486
 487         waiter->event_nr++;
 488
 489         /*
 490          * upon event ...
 491          */
 492         while (1) {
 493                 condlog(3, "%s: devmap event #%i",
 494                                 waiter->mapname, waiter->event_nr);
 495
 496                 /*
 497                  * event might be :
 498                  *
 499                  * 1) a table reload, which means our mpp structure is
 500                  *    obsolete : refresh it through update_multipath()
 501                  * 2) a path failed by DM : mark as such through
 502                  *    update_multipath()
 503                  * 3) map has gone away : stop the thread.
 504                  * 4) a path reinstate : nothing to do
 505                  * 5) a switch group : nothing to do
 506                  */
 507                 pthread_cleanup_push(cleanup_lock, waiter->vecs->lock);
 508                 lock(waiter->vecs->lock);
 509                 r = update_multipath(waiter->vecs, waiter->mapname);
 510                 lock_cleanup_pop(waiter->vecs->lock);
 511
 512                 if (r)
 513                         return -1; /* stop the thread */
 514
 515                 event_nr = dm_geteventnr(waiter->mapname);
 516
 517                 if (waiter->event_nr == event_nr)
 518                         return 1; /* upon problem reschedule 1s later */
 519
 520                 waiter->event_nr = event_nr;
 521         }
 522         return -1; /* never reach there */
 523 }
 524
 525 static void *
 526 waitevent (void * et)
 527 {
 528         int r;
 529         struct event_thread *waiter;
 530
 531         mlockall(MCL_CURRENT | MCL_FUTURE);
 532
 533         waiter = (struct event_thread *)et;
 534         pthread_cleanup_push(free_waiter, et);
 535
 536         while (1) {
 537                 r = waiteventloop(waiter);
 538
 539                 if (r < 0)
 540                         break;
 541
 542                 pthread_testcancel();
 543                 sleep(r);
 544                 pthread_testcancel();
 545         }
 546
 547         pthread_cleanup_pop(1);
 548         return NULL;
 549 }
 550
 551 static int
 552 stop_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 553 {
 554         struct event_thread * wp = (struct event_thread *)mpp->waiter;
 555         pthread_t thread = wp->thread;
 556         int r;
 557
 558         if (!wp)
 559                 return 1;
 560
 561         condlog(2, "%s: stop event checker thread", wp->mapname);
 562
 563         if ((r = pthread_cancel(thread)))
 564                 return r;
 565
 566         condlog(2, "%s: couldn't stop event checker thread gracefully, kill",
 567                 wp->mapname);
 568         pthread_kill(thread, SIGHUP);
 569         return 0;
 570 }
 571
 572 static int
 573 start_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 574 {
 575         pthread_attr_t attr;
 576         struct event_thread * wp;
 577
 578         if (!mpp)
 579                 return 0;
 580
 581         if (pthread_attr_init(&attr))
 582                 goto out;
 583
 584         pthread_attr_setstacksize(&attr, 32 * 1024);
 585         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 586
 587         wp = alloc_waiter();
 588
 589         if (!wp)
 590                 goto out;
 591
 592         mpp->waiter = (void *)wp;
 593         strncpy(wp->mapname, mpp->alias, WWID_SIZE);
 594         wp->vecs = vecs;
 595
 596         if (pthread_create(&wp->thread, &attr, waitevent, wp)) {
 597                 condlog(0, "%s: cannot create event checker", wp->mapname);
 598                 goto out1;
 599         }
 600         condlog(2, "%s: event checker started", wp->mapname);
 601
 602         return 0;
 603 out1:
 604         free_waiter(wp);
 605         mpp->waiter = NULL;
 606 out:
 607         condlog(0, "failed to start waiter thread");
 608         return 1;
 609 }
 610
 611 static void
 612 remove_map (struct multipath * mpp, struct vectors * vecs)
 613 {
 614         int i;
 615
 616         /*
 617          * stop the DM event waiter thread
 618          */
 619         if (stop_waiter_thread(mpp, vecs)) {
 620                 condlog(0, "%s: error canceling waiter thread", mpp->alias);
 621                 /*
 622                  * warrior mode
 623                  */
 624                 free_waiter(mpp->waiter);
 625         }
 626
 627         /*
 628          * clear references to this map
 629          */
 630         orphan_paths(vecs, mpp);
 631
 632         /*
 633          * purge the multipath vector
 634          */
 635         i = find_slot(vecs->mpvec, (void *)mpp);
 636         vector_del_slot(vecs->mpvec, i);
 637
 638         /*
 639          * final free
 640          */
 641         free_multipath(mpp, KEEP_PATHS);
 642         mpp = NULL;
 643 }
 644
 645 static void
 646 remove_maps (struct vectors * vecs)
 647 {
 648         int i;
 649         struct multipath * mpp;
 650
 651         vector_foreach_slot (vecs->mpvec, mpp, i) {
 652                 remove_map(mpp, vecs);
 653                 i--;
 654         }
 655
 656         vector_free(vecs->mpvec);
 657         vecs->mpvec = NULL;
 658 }
 659
 660 int
 661 uev_add_map (char * devname, struct vectors * vecs)
 662 {
 663         int major, minor, i;
 664         char dev_t[BLK_DEV_SIZE];
 665         char * alias;
 666         struct multipath * mpp;
 667
 668         if (sysfs_get_dev(sysfs_path, devname, dev_t, BLK_DEV_SIZE))
 669                 return 1;
 670
 671         if (sscanf(dev_t, "%d:%d", &major, &minor) != 2)
 672                 return 1;
 673
 674         alias = dm_mapname(major, minor);
 675
 676         if (!alias)
 677                 return 1;
 678
 679         if (!dm_type(alias, DEFAULT_TARGET)) {
 680                 condlog(4, "%s: not a multipath map", alias);
 681                 FREE(alias);
 682                 return 0;
 683         }
 684
 685         mpp = find_mp(vecs->mpvec, alias);
 686
 687         if (mpp) {
 688                 /*
 689                  * this should not happen,
 690                  * we missed a remove map event (not sent ?)
 691                  */
 692                 condlog(2, "%s: already registered", alias);
 693                 remove_map(mpp, vecs);
 694         }
 695
 696         /*
 697          * now we can allocate
 698          */
 699         mpp = alloc_multipath();
 700
 701         if (!mpp)
 702                 return 1;
 703
 704         mpp->minor = minor;
 705         mpp->alias = alias;
 706
 707         if (setup_multipath(vecs, mpp))
 708                 return 1; /* mpp freed in setup_multipath */
 709
 710         if (!vector_alloc_slot(vecs->mpvec))
 711                 goto out;
 712
 713         vector_set_slot(vecs->mpvec, mpp);
 714         adopt_paths(vecs, mpp);
 715
 716         if (start_waiter_thread(mpp, vecs))
 717                 goto out;
 718
 719         return 0;
 720 out:
 721         condlog(2, "%s: add devmap failed", mpp->alias);
 722         /*
 723          * purge the multipath vector
 724          */
 725         if ((i = find_slot(vecs->mpvec, (void *)mpp)) != -1)
 726                 vector_del_slot(vecs->mpvec, i);
 727
 728         free_multipath(mpp, KEEP_PATHS);
 729         return 1;
 730 }
 731
 732 int
 733 uev_remove_map (char * devname, struct vectors * vecs)
 734 {
 735         int minor;
 736         struct multipath * mpp;
 737
 738         if (sscanf(devname, "dm-%d", &minor) != 1)
 739                 return 1;
 740
 741         mpp = find_mp_by_minor(vecs->mpvec, minor);
 742
 743         if (!mpp) {
 744                 condlog(3, "%s: devmap not registered, can't remove",
 745                         devname);
 746                 return 1;
 747         }
 748
 749         condlog(2, "remove %s devmap", mpp->alias);
 750         remove_map(mpp, vecs);
 751
 752         return 0;
 753 }
 754
 755 int
 756 uev_add_path (char * devname, struct vectors * vecs)
 757 {
 758         struct path * pp;
 759
 760         pp = find_path_by_dev(vecs->pathvec, devname);
 761
 762         if (pp) {
 763                 condlog(3, "%s: already in pathvec");
 764                 return 1;
 765         }
 766         pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 767                        devname, DI_SYSFS | DI_WWID);
 768
 769         if (!pp) {
 770                 condlog(0, "%s: failed to store path info", devname);
 771                 return 1;
 772         }
 773
 774         condlog(2, "%s: path checker registered", devname);
 775         pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 776
 777         if (pp->mpp) {
 778                 condlog(4, "%s: ownership set to %s",
 779                                 pp->dev_t, pp->mpp->alias);
 780         } else {
 781                 condlog(4, "%s: orphaned", pp->dev_t);
 782                 orphan_path(pp);
 783         }
 784
 785         return 0;
 786 }
 787
 788 int
 789 uev_remove_path (char * devname, struct vectors * vecs)
 790 {
 791         int i;
 792         struct path * pp;
 793
 794         pp = find_path_by_dev(vecs->pathvec, devname);
 795
 796         if (!pp) {
 797                 condlog(3, "%s: not in pathvec");
 798                 return 1;
 799         }
 800
 801         if (pp->mpp && pp->state == PATH_UP)
 802                 update_queue_mode_del_path(pp->mpp);
 803
 804         condlog(2, "remove %s path checker", devname);
 805         i = find_slot(vecs->pathvec, (void *)pp);
 806         vector_del_slot(vecs->pathvec, i);
 807         free_path(pp);
 808
 809         return 0;
 810 }
 811
 812 int
 813 show_paths (char ** r, int * len, struct vectors * vecs)
 814 {
 815         int i;
 816         struct path * pp;
 817         char * c;
 818         char * reply;
 819         struct path_layout pl;
 820
 821         get_path_layout(&pl, vecs->pathvec);
 822         reply = MALLOC(MAX_REPLY_LEN);
 823
 824         if (!reply)
 825                 return 1;
 826
 827         c = reply;
 828         c += snprint_path_header(c, reply + MAX_REPLY_LEN - c,
 829                                  PRINT_PATH_CHECKER, &pl);
 830
 831         vector_foreach_slot(vecs->pathvec, pp, i)
 832                 c += snprint_path(c, reply + MAX_REPLY_LEN - c,
 833                                   PRINT_PATH_CHECKER, pp, &pl);
 834
 835         *r = reply;
 836         *len = (int)(c - reply + 1);
 837         return 0;
 838 }
 839
 840 int
 841 show_maps (char ** r, int *len, struct vectors * vecs)
 842 {
 843         int i;
 844         struct multipath * mpp;
 845         char * c;
 846         char * reply;
 847         struct map_layout ml;
 848
 849         get_map_layout(&ml, vecs->mpvec);
 850         reply = MALLOC(MAX_REPLY_LEN);
 851
 852         if (!reply)
 853                 return 1;
 854
 855         c = reply;
 856         c += snprint_map_header(c, reply + MAX_REPLY_LEN - c,
 857                                 PRINT_MAP_FAILBACK, &ml);
 858
 859         vector_foreach_slot(vecs->mpvec, mpp, i)
 860                 c += snprint_map(c, reply + MAX_REPLY_LEN - c,
 861                                  PRINT_MAP_FAILBACK, mpp, &ml);
 862
 863         *r = reply;
 864         *len = (int)(c - reply + 1);
 865         return 0;
 866 }
 867
 868 int
 869 dump_pathvec (char ** r, int * len, struct vectors * vecs)
 870 {
 871         int i;
 872         struct path * pp;
 873         char * reply;
 874         char * p;
 875
 876         *len = VECTOR_SIZE(vecs->pathvec) * sizeof(struct path);
 877         reply = (char *)MALLOC(*len);
 878         *r = reply;
 879
 880         if (!reply)
 881                 return 1;
 882
 883         p = reply;
 884
 885         vector_foreach_slot (vecs->pathvec, pp, i) {
 886                 memcpy((void *)p, pp, sizeof(struct path));
 887                 p += sizeof(struct path);
 888         }
 889
 890         /* return negative to hint caller not to add "ok" to the dump */
 891         return -1;
 892 }
 893
 894 static int
 895 map_discovery (struct vectors * vecs)
 896 {
 897         int i;
 898         struct multipath * mpp;
 899
 900         if (dm_get_maps(vecs->mpvec, "multipath"))
 901                 return 1;
 902
 903         vector_foreach_slot (vecs->mpvec, mpp, i) {
 904                 if (setup_multipath(vecs, mpp))
 905                         return 1;
 906                 mpp->minor = dm_get_minor(mpp->alias);
 907                 start_waiter_thread(mpp, vecs);
 908         }
 909
 910         return 0;
 911 }
 912
 913 int
 914 reconfigure (struct vectors * vecs)
 915 {
 916         struct config * old = conf;
 917         struct multipath * mpp;
 918         struct path * pp;
 919         int i;
 920
 921         conf = NULL;
 922
 923         if (load_config(DEFAULT_CONFIGFILE)) {
 924                 conf = old;
 925                 condlog(2, "reconfigure failed, continue with old config");
 926                 return 1;
 927         }
 928         conf->verbosity = old->verbosity;
 929         free_config(old);
 930
 931         vector_foreach_slot (vecs->mpvec, mpp, i) {
 932                 mpp->mpe = find_mpe(mpp->wwid);
 933                 mpp->hwe = extract_hwe_from_path(mpp);
 934                 adopt_paths(vecs, mpp);
 935                 set_no_path_retry(mpp);
 936         }
 937         vector_foreach_slot (vecs->pathvec, pp, i) {
 938                 select_checkfn(pp);
 939                 select_getuid(pp);
 940                 select_getprio(pp);
 941         }
 942         condlog(2, "reconfigured");
 943         return 0;
 944 }
 945
 946 int
 947 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 948 {
 949         struct vectors * vecs;
 950         int r;
 951
 952         *reply = NULL;
 953         *len = 0;
 954         vecs = (struct vectors *)trigger_data;
 955
 956         pthread_cleanup_push(cleanup_lock, vecs->lock);
 957         lock(vecs->lock);
 958
 959         r = parse_cmd(str, reply, len, vecs);
 960
 961         if (r > 0) {
 962                 *reply = STRDUP("fail\n");
 963                 *len = strlen(*reply) + 1;
 964                 r = 1;
 965         }
 966         else if (!r && *len == 0) {
 967                 *reply = STRDUP("ok\n");
 968                 *len = strlen(*reply) + 1;
 969                 r = 0;
 970         }
 971         /* else if (r < 0) leave *reply alone */
 972
 973         lock_cleanup_pop(vecs->lock);
 974         return r;
 975 }
 976
 977 static int
 978 uev_discard(char * devpath)
 979 {
 980         char a[10], b[10];
 981
 982         /*
 983          * keep only block devices, discard partitions
 984          */
 985         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 986             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 987                 condlog(4, "discard event on %s", devpath);
 988                 return 1;
 989         }
 990         return 0;
 991 }
 992
 993 int
 994 uev_trigger (struct uevent * uev, void * trigger_data)
 995 {
 996         int r = 0;
 997         char devname[32];
 998         struct vectors * vecs;
 999
1000         vecs = (struct vectors *)trigger_data;
1001
1002         if (uev_discard(uev->devpath))
1003                 goto out;
1004
1005         basename(uev->devpath, devname);
1006         lock(vecs->lock);
1007
1008         /*
1009          * device map add/remove event
1010          */
1011         if (!strncmp(devname, "dm-", 3)) {
1012                 if (!strncmp(uev->action, "add", 3)) {
1013                         r = uev_add_map(devname, vecs);
1014                         goto out;
1015                 }
1016 #if 0
1017                 if (!strncmp(uev->action, "remove", 6)) {
1018                         r = uev_remove_map(devname, vecs);
1019                         goto out;
1020                 }
1021 #endif
1022                 goto out;
1023         }
1024
1025         /*
1026          * path add/remove event
1027          */
1028         if (blacklist(conf->blist, devname))
1029                 goto out;
1030
1031         if (!strncmp(uev->action, "add", 3)) {
1032                 r = uev_add_path(devname, vecs);
1033                 goto out;
1034         }
1035         if (!strncmp(uev->action, "remove", 6)) {
1036                 r = uev_remove_path(devname, vecs);
1037                 goto out;
1038         }
1039
1040 out:
1041         unlock(vecs->lock);
1042         return r;
1043 }
1044
1045 static void *
1046 ueventloop (void * ap)
1047 {
1048         uevent_listen(&uev_trigger, ap);
1049
1050         return NULL;
1051 }
1052
1053 static void *
1054 uxlsnrloop (void * ap)
1055 {
1056         if (load_keys())
1057                 return NULL;
1058
1059         if (alloc_handlers())
1060                 return NULL;
1061
1062         add_handler(LIST+PATHS, cli_list_paths);
1063         add_handler(LIST+MAPS, cli_list_maps);
1064         add_handler(ADD+PATH, cli_add_path);
1065         add_handler(DEL+PATH, cli_del_path);
1066         add_handler(ADD+MAP, cli_add_map);
1067         add_handler(DEL+MAP, cli_del_map);
1068         add_handler(SWITCH+MAP+GROUP, cli_switch_group);
1069         add_handler(DUMP+PATHVEC, cli_dump_pathvec);
1070         add_handler(RECONFIGURE, cli_reconfigure);
1071
1072         uxsock_listen(&uxsock_trigger, ap);
1073
1074         return NULL;
1075 }
1076
1077 static int
1078 exit_daemon (int status)
1079 {
1080         if (status != 0)
1081                 fprintf(stderr, "bad exit status. see daemon.log\n");
1082
1083         condlog(3, "unlink pidfile");
1084         unlink(DEFAULT_PIDFILE);
1085
1086         lock(&exit_mutex);
1087         pthread_cond_signal(&exit_cond);
1088         unlock(&exit_mutex);
1089
1090         return status;
1091 }
1092
1093 static void
1094 fail_path (struct path * pp)
1095 {
1096         if (!pp->mpp)
1097                 return;
1098
1099         condlog(2, "checker failed path %s in map %s",
1100                  pp->dev_t, pp->mpp->alias);
1101
1102         dm_fail_path(pp->mpp->alias, pp->dev_t);
1103         update_queue_mode_del_path(pp->mpp);
1104 }
1105
1106 /*
1107  * caller must have locked the path list before calling that function
1108  */
1109 static void
1110 reinstate_path (struct path * pp)
1111 {
1112         if (!pp->mpp)
1113                 return;
1114
1115         if (dm_reinstate(pp->mpp->alias, pp->dev_t))
1116                 condlog(0, "%s: reinstate failed", pp->dev_t);
1117         else {
1118                 condlog(2, "%s: reinstated", pp->dev_t);
1119                 update_queue_mode_add_path(pp->mpp);
1120         }
1121 }
1122
1123 static void
1124 enable_group(struct path * pp)
1125 {
1126         struct pathgroup * pgp;
1127
1128         /*
1129          * if path is added through uev_add_path, pgindex can be unset.
1130          * next update_strings() will set it, upon map reload event.
1131          *
1132          * we can safely return here, because upon map reload, all
1133          * PG will be enabled.
1134          */
1135         if (!pp->pgindex)
1136                 return;
1137
1138         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1139
1140         if (pgp->status == PGSTATE_DISABLED) {
1141                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1142                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1143         }
1144 }
1145
1146 static void
1147 mpvec_garbage_collector (struct vectors * vecs)
1148 {
1149         struct multipath * mpp;
1150         int i;
1151
1152         vector_foreach_slot (vecs->mpvec, mpp, i) {
1153                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1154                         condlog(2, "%s: remove dead map", mpp->alias);
1155                         remove_map(mpp, vecs);
1156                         i--;
1157                 }
1158         }
1159 }
1160
1161 static void
1162 defered_failback_tick (vector mpvec)
1163 {
1164         struct multipath * mpp;
1165         int i;
1166
1167         vector_foreach_slot (mpvec, mpp, i) {
1168                 /*
1169                  * defered failback getting sooner
1170                  */
1171                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1172                         mpp->failback_tick--;
1173
1174                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1175                                 switch_pathgroup(mpp);
1176                 }
1177         }
1178 }
1179
1180 static void
1181 retry_count_tick(vector mpvec)
1182 {
1183         struct multipath *mpp;
1184         int i;
1185
1186         vector_foreach_slot (mpvec, mpp, i) {
1187                 if (mpp->retry_tick) {
1188                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1189                         if(--mpp->retry_tick == 0) {
1190                                 dm_queue_if_no_path(mpp->alias, 0);
1191                                 condlog(2, "%s: Disable queueing", mpp->alias);
1192                         }
1193                 }
1194         }
1195 }
1196
1197 static void *
1198 checkerloop (void *ap)
1199 {
1200         struct vectors *vecs;
1201         struct path *pp;
1202         int i, count = 0;
1203         int newstate;
1204         char checker_msg[MAX_CHECKER_MSG_SIZE];
1205
1206         mlockall(MCL_CURRENT | MCL_FUTURE);
1207
1208         memset(checker_msg, 0, MAX_CHECKER_MSG_SIZE);
1209         vecs = (struct vectors *)ap;
1210
1211         condlog(2, "path checkers start up");
1212
1213         /*
1214          * init the path check interval
1215          */
1216         vector_foreach_slot (vecs->pathvec, pp, i) {
1217                 pp->checkint = conf->checkint;
1218         }
1219
1220         while (1) {
1221                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1222                 lock(vecs->lock);
1223                 condlog(4, "tick");
1224
1225                 vector_foreach_slot (vecs->pathvec, pp, i) {
1226                         if (!pp->mpp)
1227                                 continue;
1228
1229                         if (pp->tick && --pp->tick)
1230                                 continue; /* don't check this path yet */
1231
1232                         /*
1233                          * provision a next check soonest,
1234                          * in case we exit abnormaly from here
1235                          */
1236                         pp->tick = conf->checkint;
1237
1238                         if (!pp->checkfn) {
1239                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
1240                                 select_checkfn(pp);
1241                         }
1242
1243                         if (!pp->checkfn) {
1244                                 condlog(0, "%s: checkfn is void", pp->dev);
1245                                 continue;
1246                         }
1247                         newstate = pp->checkfn(pp->fd, checker_msg,
1248                                                &pp->checker_context);
1249
1250                         if (newstate != pp->state) {
1251                                 pp->state = newstate;
1252                                 LOG_MSG(1, checker_msg);
1253
1254                                 /*
1255                                  * upon state change, reset the checkint
1256                                  * to the shortest delay
1257                                  */
1258                                 pp->checkint = conf->checkint;
1259
1260                                 if (newstate == PATH_DOWN ||
1261                                     newstate == PATH_SHAKY) {
1262                                         /*
1263                                          * proactively fail path in the DM
1264                                          */
1265                                         fail_path(pp);
1266
1267                                         /*
1268                                          * cancel scheduled failback
1269                                          */
1270                                         pp->mpp->failback_tick = 0;
1271
1272                                         continue;
1273                                 }
1274
1275                                 /*
1276                                  * reinstate this path
1277                                  */
1278                                 reinstate_path(pp);
1279
1280                                 /*
1281                                  * need to switch group ?
1282                                  */
1283                                 update_multipath_strings(pp->mpp,
1284                                                          vecs->pathvec);
1285
1286                                 /*
1287                                  * schedule defered failback
1288                                  */
1289                                 if (pp->mpp->pgfailback > 0)
1290                                         pp->mpp->failback_tick =
1291                                                 pp->mpp->pgfailback + 1;
1292                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
1293                                     need_switch_pathgroup(pp->mpp, 1))
1294                                         switch_pathgroup(pp->mpp);
1295
1296                                 /*
1297                                  * if at least one path is up in a group, and
1298                                  * the group is disabled, re-enable it
1299                                  */
1300                                 if (newstate == PATH_UP)
1301                                         enable_group(pp);
1302                         }
1303                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1304                                 LOG_MSG(4, checker_msg);
1305                                 /*
1306                                  * double the next check delay.
1307                                  * max at conf->max_checkint
1308                                  */
1309                                 if (pp->checkint < (conf->max_checkint / 2))
1310                                         pp->checkint = 2 * pp->checkint;
1311                                 else
1312                                         pp->checkint = conf->max_checkint;
1313
1314                                 pp->tick = pp->checkint;
1315                                 condlog(4, "%s: delay next check %is",
1316                                                 pp->dev_t, pp->tick);
1317
1318                         }
1319                         pp->state = newstate;
1320
1321                         /*
1322                          * path prio refreshing
1323                          */
1324                         condlog(4, "path prio refresh");
1325                         pathinfo(pp, conf->hwtable, DI_PRIO);
1326
1327                         if (need_switch_pathgroup(pp->mpp, 0)) {
1328                                 if (pp->mpp->pgfailback > 0)
1329                                         pp->mpp->failback_tick =
1330                                                 pp->mpp->pgfailback + 1;
1331                                 else if (pp->mpp->pgfailback ==
1332                                                 -FAILBACK_IMMEDIATE)
1333                                         switch_pathgroup(pp->mpp);
1334                         }
1335                 }
1336                 defered_failback_tick(vecs->mpvec);
1337                 retry_count_tick(vecs->mpvec);
1338
1339                 if (count)
1340                         count--;
1341                 else {
1342                         condlog(4, "map garbage collection");
1343                         mpvec_garbage_collector(vecs);
1344                         count = MAPGCINT;
1345                 }
1346
1347                 lock_cleanup_pop(vecs->lock);
1348                 sleep(1);
1349         }
1350         return NULL;
1351 }
1352
1353 static struct vectors *
1354 init_paths (void)
1355 {
1356         struct vectors * vecs;
1357
1358         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1359
1360         if (!vecs)
1361                 return NULL;
1362
1363         vecs->lock =
1364                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1365
1366         if (!vecs->lock)
1367                 goto out;
1368
1369         vecs->pathvec = vector_alloc();
1370
1371         if (!vecs->pathvec)
1372                 goto out1;
1373
1374         vecs->mpvec = vector_alloc();
1375
1376         if (!vecs->mpvec)
1377                 goto out2;
1378
1379         pthread_mutex_init(vecs->lock, NULL);
1380
1381         return vecs;
1382
1383 out2:
1384         vector_free(vecs->pathvec);
1385 out1:
1386         FREE(vecs->lock);
1387 out:
1388         FREE(vecs);
1389         condlog(0, "failed to init paths");
1390         return NULL;
1391 }
1392
1393 static void *
1394 signal_set(int signo, void (*func) (int))
1395 {
1396         int r;
1397         struct sigaction sig;
1398         struct sigaction osig;
1399
1400         sig.sa_handler = func;
1401         sigemptyset(&sig.sa_mask);
1402         sig.sa_flags = 0;
1403
1404         r = sigaction(signo, &sig, &osig);
1405
1406         if (r < 0)
1407                 return (SIG_ERR);
1408         else
1409                 return (osig.sa_handler);
1410 }
1411
1412 static void
1413 sighup (int sig)
1414 {
1415         condlog(3, "SIGHUP received");
1416
1417 #ifdef _DEBUG_
1418         dbg_free_final(NULL);
1419 #endif
1420 }
1421
1422 static void
1423 sigend (int sig)
1424 {
1425         exit_daemon(0);
1426 }
1427
1428 static void
1429 signal_init(void)
1430 {
1431         signal_set(SIGHUP, sighup);
1432         signal_set(SIGINT, sigend);
1433         signal_set(SIGTERM, sigend);
1434         signal_set(SIGKILL, sigend);
1435 }
1436
1437 static void
1438 setscheduler (void)
1439 {
1440         int res;
1441         static struct sched_param sched_param = {
1442                 sched_priority: 99
1443         };
1444
1445         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1446
1447         if (res == -1)
1448                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1449         return;
1450 }
1451
1452 static void
1453 set_oom_adj (int val)
1454 {
1455         FILE *fp;
1456
1457         fp = fopen("/proc/self/oom_adj", "w");
1458
1459         if (!fp)
1460                 return;
1461
1462         fprintf(fp, "%i", val);
1463         fclose(fp);
1464 }
1465
1466 static int
1467 child (void * param)
1468 {
1469         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1470         pthread_attr_t attr;
1471         struct vectors * vecs;
1472
1473         mlockall(MCL_CURRENT | MCL_FUTURE);
1474
1475         if (logsink)
1476                 log_thread_start();
1477
1478         condlog(2, "--------start up--------");
1479         condlog(2, "read " DEFAULT_CONFIGFILE);
1480
1481         if (load_config(DEFAULT_CONFIGFILE))
1482                 exit(1);
1483
1484         setlogmask(LOG_UPTO(conf->verbosity + 3));
1485
1486         /*
1487          * fill the voids left in the config file
1488          */
1489         if (!conf->checkint) {
1490                 conf->checkint = CHECKINT;
1491                 conf->max_checkint = MAX_CHECKINT;
1492         }
1493
1494         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1495                 if (logsink)
1496                         log_thread_stop();
1497
1498                 exit(1);
1499         }
1500         signal_init();
1501         setscheduler();
1502         set_oom_adj(-17);
1503         vecs = init_paths();
1504
1505         if (!vecs)
1506                 exit(1);
1507
1508         if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1509                 condlog(0, "can not find sysfs mount point");
1510                 exit(1);
1511         }
1512
1513         /*
1514          * fetch paths and multipaths lists
1515          * no paths and/or no multipaths are valid scenarii
1516          * vectors maintenance will be driven by events
1517          */
1518         path_discovery(vecs->pathvec, conf, DI_SYSFS | DI_WWID | DI_CHECKER);
1519         map_discovery(vecs);
1520
1521         /*
1522          * start threads
1523          */
1524         pthread_attr_init(&attr);
1525         pthread_attr_setstacksize(&attr, 64 * 1024);
1526         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1527
1528         pthread_create(&check_thr, &attr, checkerloop, vecs);
1529         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1530         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1531
1532         pthread_cond_wait(&exit_cond, &exit_mutex);
1533
1534         /*
1535          * exit path
1536          */
1537         lock(vecs->lock);
1538         remove_maps(vecs);
1539         free_pathvec(vecs->pathvec, FREE_PATHS);
1540
1541         pthread_cancel(check_thr);
1542         pthread_cancel(uevent_thr);
1543         pthread_cancel(uxlsnr_thr);
1544
1545         free_keys(keys);
1546         free_handlers(handlers);
1547         free_polls();
1548
1549         unlock(vecs->lock);
1550         pthread_mutex_destroy(vecs->lock);
1551         FREE(vecs->lock);
1552         FREE(vecs);
1553         free_config(conf);
1554
1555         condlog(2, "--------shut down-------");
1556
1557         if (logsink)
1558                 log_thread_stop();
1559
1560 #ifdef _DEBUG_
1561         dbg_free_final(NULL);
1562 #endif
1563
1564         exit(0);
1565 }
1566
1567 int
1568 main (int argc, char *argv[])
1569 {
1570         extern char *optarg;
1571         extern int optind;
1572         int arg;
1573         int err;
1574
1575         logsink = 1;
1576
1577         if (getuid() != 0) {
1578                 fprintf(stderr, "need to be root\n");
1579                 exit(1);
1580         }
1581
1582         /* make sure we don't lock any path */
1583         chdir("/");
1584         umask(umask(077) | 022);
1585
1586         conf = alloc_config();
1587
1588         if (!conf)
1589                 exit(1);
1590
1591         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1592         switch(arg) {
1593                 case 'd':
1594                         logsink = 0;
1595                         //debug=1; /* ### comment me out ### */
1596                         break;
1597                 case 'v':
1598                         if (sizeof(optarg) > sizeof(char *) ||
1599                             !isdigit(optarg[0]))
1600                                 exit(1);
1601
1602                         conf->verbosity = atoi(optarg);
1603                         break;
1604                 case 'k':
1605                         uxclnt(optarg);
1606                         exit(0);
1607                 default:
1608                         ;
1609                 }
1610         }
1611
1612         if (!logsink)
1613                 err = 0;
1614         else
1615                 err = fork();
1616
1617         if (err < 0)
1618                 /* error */
1619                 exit(1);
1620         else if (err > 0)
1621                 /* parent dies */
1622                 exit(0);
1623         else
1624                 /* child lives */
1625                 return (child(NULL));
1626 }