multipathd/main.c

   1 /*
   2  * Copyright (c) 2004, 2005 Christophe Varoqui
   3  * Copyright (c) 2005 Kiyoshi Ueda, NEC
   4  * Copyright (c) 2005 Benjamin Marzinski, Redhat
   5  * Copyright (c) 2005 Edward Goggin, EMC
   6  */
   7 #include <unistd.h>
   8 #include <sys/stat.h>
   9 #include <libdevmapper.h>
  10 #include <wait.h>
  11 #include <sys/mman.h>
  12 #include <sys/types.h>
  13 #include <fcntl.h>
  14 #include <errno.h>
  15
  16 /*
  17  * libsysfs
  18  */
  19 #include <sysfs/libsysfs.h>
  20 #include <sysfs/dlist.h>
  21
  22 /*
  23  * libcheckers
  24  */
  25 #include <checkers.h>
  26 #include <path_state.h>
  27
  28 /*
  29  * libmultipath
  30  */
  31 #include <parser.h>
  32 #include <vector.h>
  33 #include <memory.h>
  34 #include <config.h>
  35 #include <callout.h>
  36 #include <util.h>
  37 #include <blacklist.h>
  38 #include <hwtable.h>
  39 #include <defaults.h>
  40 #include <structs.h>
  41 #include <structs_vec.h>
  42 #include <dmparser.h>
  43 #include <devmapper.h>
  44 #include <dict.h>
  45 #include <discovery.h>
  46 #include <debug.h>
  47 #include <propsel.h>
  48 #include <uevent.h>
  49 #include <switchgroup.h>
  50 #include <path_state.h>
  51 #include <print.h>
  52 #include <configure.h>
  53
  54 #include "main.h"
  55 #include "pidfile.h"
  56 #include "uxlsnr.h"
  57 #include "uxclnt.h"
  58 #include "cli.h"
  59 #include "cli_handlers.h"
  60
  61 #define FILE_NAME_SIZE 256
  62 #define CMDSIZE 160
  63
  64 #define LOG_MSG(a,b) \
  65         if (strlen(b)) { \
  66                 condlog(a, "%s: %s", pp->dev_t, b); \
  67                 memset(b, 0, MAX_CHECKER_MSG_SIZE); \
  68         }
  69
  70 #ifdef LCKDBG
  71 #define lock(a) \
  72         fprintf(stderr, "%s:%s(%i) lock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  73         pthread_mutex_lock(a)
  74 #define unlock(a) \
  75         fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  76         pthread_mutex_unlock(a)
  77 #define lock_cleanup_pop(a) \
  78         fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  79         pthread_cleanup_pop(1);
  80 #else
  81 #define lock(a) pthread_mutex_lock(a)
  82 #define unlock(a) pthread_mutex_unlock(a)
  83 #define lock_cleanup_pop(a) pthread_cleanup_pop(1);
  84 #endif
  85
  86 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  87 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  88
  89 /*
  90  * structs
  91  */
  92 struct vectors * gvecs; /* global copy of vecs for use in sig handlers */
  93
  94 static struct event_thread *
  95 alloc_waiter (void)
  96 {
  97
  98         struct event_thread * wp;
  99
 100         wp = (struct event_thread *)MALLOC(sizeof(struct event_thread));
 101
 102         return wp;
 103 }
 104
 105 static void
 106 free_waiter (void * data)
 107 {
 108         struct event_thread * wp = (struct event_thread *)data;
 109
 110         /*
 111          * indicate in mpp that the wp is already freed storage
 112          */
 113         lock(wp->vecs->lock);
 114
 115         if (wp->mpp)
 116                 /*
 117                  * be careful, mpp may already be freed -- null if so
 118                  */
 119                 wp->mpp->waiter = NULL;
 120         else
 121                 condlog(3, "free_waiter, mpp freed before wp=%p,", wp);
 122
 123         unlock(wp->vecs->lock);
 124
 125         if (wp->dmt)
 126                 dm_task_destroy(wp->dmt);
 127
 128         FREE(wp);
 129 }
 130
 131 static void
 132 stop_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 133 {
 134         struct event_thread * wp = (struct event_thread *)mpp->waiter;
 135
 136         if (!wp) {
 137                 condlog(3, "%s: no waiter thread", mpp->alias);
 138                 return;
 139         }
 140         condlog(2, "%s: stop event checker thread", wp->mapname);
 141         pthread_kill((pthread_t)wp->thread, SIGUSR1);
 142 }
 143
 144 static void
 145 cleanup_lock (void * data)
 146 {
 147         unlock((pthread_mutex_t *)data);
 148 }
 149
 150 /*
 151  * mpp->no_path_retry:
 152  *   -2 (QUEUE) : queue_if_no_path enabled, never turned off
 153  *   -1 (FAIL)  : fail_if_no_path
 154  *    0 (UNDEF) : nothing
 155  *   >0         : queue_if_no_path enabled, turned off after polling n times
 156  */
 157 static void
 158 update_queue_mode_del_path(struct multipath *mpp)
 159 {
 160         if (--mpp->nr_active == 0 && mpp->no_path_retry > 0) {
 161                 /*
 162                  * Enter retry mode.
 163                  * meaning of +1: retry_tick may be decremented in
 164                  *                checkerloop before starting retry.
 165                  */
 166                 mpp->stat_queueing_timeouts++;
 167                 mpp->retry_tick = mpp->no_path_retry * conf->checkint + 1;
 168                 condlog(1, "%s: Entering recovery mode: max_retries=%d",
 169                         mpp->alias, mpp->no_path_retry);
 170         }
 171         condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
 172 }
 173
 174 static void
 175 update_queue_mode_add_path(struct multipath *mpp)
 176 {
 177         if (mpp->nr_active++ == 0 && mpp->no_path_retry > 0) {
 178                 /* come back to normal mode from retry mode */
 179                 mpp->retry_tick = 0;
 180                 dm_queue_if_no_path(mpp->alias, 1);
 181                 condlog(2, "%s: queue_if_no_path enabled", mpp->alias);
 182                 condlog(1, "%s: Recovered to normal mode", mpp->alias);
 183         }
 184         condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
 185 }
 186
 187 static int
 188 need_switch_pathgroup (struct multipath * mpp, int refresh)
 189 {
 190         struct pathgroup * pgp;
 191         struct path * pp;
 192         int i, j;
 193
 194         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
 195                 return 0;
 196
 197         /*
 198          * Refresh path priority values
 199          */
 200         if (refresh)
 201                 vector_foreach_slot (mpp->pg, pgp, i)
 202                         vector_foreach_slot (pgp->paths, pp, j)
 203                                 pathinfo(pp, conf->hwtable, DI_PRIO);
 204
 205         mpp->bestpg = select_path_group(mpp);
 206
 207         if (mpp->bestpg != mpp->nextpg)
 208                 return 1;
 209
 210         return 0;
 211 }
 212
 213 static void
 214 switch_pathgroup (struct multipath * mpp)
 215 {
 216         mpp->stat_switchgroup++;
 217         dm_switchgroup(mpp->alias, mpp->bestpg);
 218         condlog(2, "%s: switch to path group #%i",
 219                  mpp->alias, mpp->bestpg);
 220 }
 221
 222 static int
 223 coalesce_maps(struct vectors *vecs, vector nmpv)
 224 {
 225         struct multipath * ompp;
 226         vector ompv = vecs->mpvec;
 227         int i, j;
 228
 229         condlog(3, "coalesce_maps vs = %u", VECTOR_SIZE(ompv));
 230         condlog(3, "coalesce_maps vs = %u", VECTOR_SIZE(nmpv));
 231
 232         vector_foreach_slot (ompv, ompp, i) {
 233                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 234                         /*
 235                          * remove all current maps not allowed by the
 236                          * current configuration
 237                          */
 238                         if (dm_flush_map(ompp->alias, DEFAULT_TARGET)) {
 239                                 condlog(0, "%s: unable to flush devmap",
 240                                         ompp->alias);
 241                                 /*
 242                                  * may be just because the device is open
 243                                  */
 244                                 if (!vector_alloc_slot(nmpv))
 245                                         return 1;
 246
 247                                 vector_set_slot(nmpv, ompp);
 248                                 setup_multipath(vecs, ompp);
 249
 250                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 251                                         vector_del_slot(ompv, j);
 252
 253                                 continue;
 254                         }
 255                         else {
 256                                 dm_lib_release();
 257                                 condlog(3, "%s devmap removed", ompp->alias);
 258                         }
 259                 }
 260         }
 261         return 0;
 262 }
 263
 264 static int
 265 update_multipath (struct vectors *vecs, char *mapname)
 266 {
 267         struct multipath *mpp;
 268         struct pathgroup  *pgp;
 269         struct path *pp;
 270         int i, j;
 271         int r = 1;
 272
 273         mpp = find_mp_by_alias(vecs->mpvec, mapname);
 274
 275         if (!mpp)
 276                 goto out;
 277
 278         free_pgvec(mpp->pg, KEEP_PATHS);
 279         mpp->pg = NULL;
 280
 281         if (setup_multipath(vecs, mpp))
 282                 goto out; /* mpp freed in setup_multipath */
 283
 284         /*
 285          * compare checkers states with DM states
 286          */
 287         vector_foreach_slot (mpp->pg, pgp, i) {
 288                 vector_foreach_slot (pgp->paths, pp, j) {
 289                         if (pp->dmstate != PSTATE_FAILED)
 290                                 continue;
 291
 292                         if (pp->state != PATH_DOWN) {
 293                                 condlog(2, "%s: mark as failed", pp->dev_t);
 294                                 mpp->stat_path_failures++;
 295                                 pp->state = PATH_DOWN;
 296                                 update_queue_mode_del_path(mpp);
 297
 298                                 /*
 299                                  * if opportune,
 300                                  * schedule the next check earlier
 301                                  */
 302                                 if (pp->tick > conf->checkint)
 303                                         pp->tick = conf->checkint;
 304                         }
 305                 }
 306         }
 307         r = 0;
 308 out:
 309         if (r)
 310                 condlog(0, "failed to update multipath");
 311
 312         return r;
 313 }
 314
 315 static sigset_t unblock_signals(void)
 316 {
 317         sigset_t set, old;
 318
 319         sigemptyset(&set);
 320         sigaddset(&set, SIGHUP);
 321         sigaddset(&set, SIGUSR1);
 322         pthread_sigmask(SIG_UNBLOCK, &set, &old);
 323         return old;
 324 }
 325
 326 /*
 327  * returns the reschedule delay
 328  * negative means *stop*
 329  */
 330 static int
 331 waiteventloop (struct event_thread * waiter)
 332 {
 333         sigset_t set;
 334         int event_nr;
 335         int r;
 336
 337         if (!waiter->event_nr)
 338                 waiter->event_nr = dm_geteventnr(waiter->mapname);
 339
 340         if (!(waiter->dmt = dm_task_create(DM_DEVICE_WAITEVENT))) {
 341                 condlog(0, "%s: devmap event #%i dm_task_create error",
 342                                 waiter->mapname, waiter->event_nr);
 343                 return 1;
 344         }
 345
 346         if (!dm_task_set_name(waiter->dmt, waiter->mapname)) {
 347                 condlog(0, "%s: devmap event #%i dm_task_set_name error",
 348                                 waiter->mapname, waiter->event_nr);
 349                 dm_task_destroy(waiter->dmt);
 350                 return 1;
 351         }
 352
 353         if (waiter->event_nr && !dm_task_set_event_nr(waiter->dmt,
 354                                                       waiter->event_nr)) {
 355                 condlog(0, "%s: devmap event #%i dm_task_set_event_nr error",
 356                                 waiter->mapname, waiter->event_nr);
 357                 dm_task_destroy(waiter->dmt);
 358                 return 1;
 359         }
 360
 361         dm_task_no_open_count(waiter->dmt);
 362
 363         /* accept wait interruption */
 364         set = unblock_signals();
 365
 366         /* interruption spits messages */
 367         dm_shut_log();
 368
 369         /* wait */
 370         r = dm_task_run(waiter->dmt);
 371
 372         /* wait is over : event or interrupt */
 373         pthread_sigmask(SIG_SETMASK, &set, NULL);
 374         //dm_restore_log();
 375
 376         if (!r) /* wait interrupted by signal */
 377                 return -1;
 378
 379         dm_task_destroy(waiter->dmt);
 380         waiter->dmt = NULL;
 381         waiter->event_nr++;
 382
 383         /*
 384          * upon event ...
 385          */
 386         while (1) {
 387                 condlog(3, "%s: devmap event #%i",
 388                                 waiter->mapname, waiter->event_nr);
 389
 390                 /*
 391                  * event might be :
 392                  *
 393                  * 1) a table reload, which means our mpp structure is
 394                  *    obsolete : refresh it through update_multipath()
 395                  * 2) a path failed by DM : mark as such through
 396                  *    update_multipath()
 397                  * 3) map has gone away : stop the thread.
 398                  * 4) a path reinstate : nothing to do
 399                  * 5) a switch group : nothing to do
 400                  */
 401                 pthread_cleanup_push(cleanup_lock, waiter->vecs->lock);
 402                 lock(waiter->vecs->lock);
 403                 r = update_multipath(waiter->vecs, waiter->mapname);
 404                 lock_cleanup_pop(waiter->vecs->lock);
 405
 406                 if (r)
 407                         return -1; /* stop the thread */
 408
 409                 event_nr = dm_geteventnr(waiter->mapname);
 410
 411                 if (waiter->event_nr == event_nr)
 412                         return 1; /* upon problem reschedule 1s later */
 413
 414                 waiter->event_nr = event_nr;
 415         }
 416         return -1; /* never reach there */
 417 }
 418
 419 static void *
 420 waitevent (void * et)
 421 {
 422         int r;
 423         struct event_thread *waiter;
 424
 425         mlockall(MCL_CURRENT | MCL_FUTURE);
 426
 427         waiter = (struct event_thread *)et;
 428         pthread_cleanup_push(free_waiter, et);
 429
 430         while (1) {
 431                 r = waiteventloop(waiter);
 432
 433                 if (r < 0)
 434                         break;
 435
 436                 sleep(r);
 437         }
 438
 439         pthread_cleanup_pop(1);
 440         return NULL;
 441 }
 442
 443 static int
 444 start_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 445 {
 446         pthread_attr_t attr;
 447         struct event_thread * wp;
 448
 449         if (!mpp)
 450                 return 0;
 451
 452         if (pthread_attr_init(&attr))
 453                 goto out;
 454
 455         pthread_attr_setstacksize(&attr, 32 * 1024);
 456         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 457
 458         wp = alloc_waiter();
 459
 460         if (!wp)
 461                 goto out;
 462
 463         mpp->waiter = (void *)wp;
 464         strncpy(wp->mapname, mpp->alias, WWID_SIZE);
 465         wp->vecs = vecs;
 466         wp->mpp = mpp;
 467
 468         if (pthread_create(&wp->thread, &attr, waitevent, wp)) {
 469                 condlog(0, "%s: cannot create event checker", wp->mapname);
 470                 goto out1;
 471         }
 472         condlog(2, "%s: event checker started", wp->mapname);
 473
 474         return 0;
 475 out1:
 476         free_waiter(wp);
 477         mpp->waiter = NULL;
 478 out:
 479         condlog(0, "failed to start waiter thread");
 480         return 1;
 481 }
 482
 483 static int
 484 flush_map(struct multipath * mpp, struct vectors * vecs)
 485 {
 486         /*
 487          * clear references to this map before flushing so we can ignore
 488          * the spurious uevent we may generate with the dm_flush_map call below
 489          */
 490         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 491                 /*
 492                  * May not really be an error -- if the map was already flushed
 493                  * from the device mapper by dmsetup(8) for instance.
 494                  */
 495                 condlog(0, "%s: can't flush", mpp->alias);
 496                 return 1;
 497         }
 498         else {
 499                 dm_lib_release();
 500                 condlog(3, "%s: devmap removed", mpp->alias);
 501         }
 502
 503         orphan_paths(vecs->pathvec, mpp);
 504         remove_map(mpp, vecs, stop_waiter_thread, 1);
 505
 506         return 0;
 507 }
 508
 509 static int
 510 uev_add_map (char * devname, struct vectors * vecs)
 511 {
 512         condlog(2, "%s: add map (uevent)", devname);
 513         return ev_add_map(devname, vecs);
 514 }
 515
 516 int
 517 ev_add_map (char * devname, struct vectors * vecs)
 518 {
 519         int major, minor;
 520         char dev_t[BLK_DEV_SIZE];
 521         char * alias;
 522         char * refwwid;
 523         struct multipath * mpp;
 524         int map_present;
 525         int r = 1;
 526
 527         if (sscanf(devname, "dm-%d", &minor) == 1 &&
 528             !sysfs_get_dev(sysfs_path, devname, dev_t, BLK_DEV_SIZE) &&
 529             sscanf(dev_t, "%d:%d", &major, &minor) == 2)
 530                 alias = dm_mapname(major, minor);
 531         else
 532                 alias = STRDUP(devname);
 533
 534         if (!alias)
 535                 return 1;
 536
 537         map_present = dm_map_present(alias);
 538
 539         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 540                 condlog(4, "%s: not a multipath map", alias);
 541                 FREE(alias);
 542                 return 0;
 543         }
 544
 545         mpp = find_mp_by_alias(vecs->mpvec, alias);
 546
 547         if (mpp) {
 548                 /*
 549                  * Not really an error -- we generate our own uevent
 550                  * if we create a multipath mapped device as a result
 551                  * of uev_add_path
 552                  */
 553                 condlog(0, "%s: devmap already registered",
 554                         devname);
 555                 FREE(alias);
 556                 return 0;
 557         }
 558
 559         /*
 560          * now we can register the map
 561          */
 562         if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
 563                                         start_waiter_thread))) {
 564                 condlog(3, "%s: devmap %s added", alias, devname);
 565                 return 0;
 566         }
 567         refwwid = get_refwwid(devname, DEV_DEVMAP, vecs->pathvec);
 568
 569         if (refwwid) {
 570                 r = coalesce_paths(vecs, NULL, refwwid);
 571                 dm_lib_release();
 572         }
 573
 574         if (!r)
 575                 condlog(3, "%s: devmap %s added", alias, devname);
 576         else
 577                 condlog(0, "%s: uev_add_map %s failed", alias, devname);
 578
 579         FREE(refwwid);
 580         FREE(alias);
 581         return r;
 582 }
 583
 584 static int
 585 uev_remove_map (char * devname, struct vectors * vecs)
 586 {
 587         condlog(2, "%s: remove map (uevent)", devname);
 588         return ev_remove_map(devname, vecs);
 589 }
 590
 591 int
 592 ev_remove_map (char * devname, struct vectors * vecs)
 593 {
 594         struct multipath * mpp;
 595
 596         mpp = find_mp_by_str(vecs->mpvec, devname);
 597
 598         if (!mpp) {
 599                 condlog(3, "%s: devmap not registered, can't remove",
 600                         devname);
 601                 return 0;
 602         }
 603         flush_map(mpp, vecs);
 604
 605         return 0;
 606 }
 607
 608 static int
 609 uev_umount_map (char * devname, struct vectors * vecs)
 610 {
 611         int minor;
 612         struct multipath * mpp;
 613
 614         condlog(2, "%s: umount map (uevent)", devname);
 615
 616         if (sscanf(devname, "dm-%d", &minor) == 1)
 617                 mpp = find_mp_by_minor(vecs->mpvec, minor);
 618         else
 619                 mpp = find_mp_by_alias(vecs->mpvec, devname);
 620
 621         if (!mpp)
 622                 return 0;
 623
 624         update_mpp_paths(mpp, vecs->pathvec);
 625         verify_paths(mpp, vecs, NULL);
 626
 627         if (!VECTOR_SIZE(mpp->paths))
 628                 flush_map(mpp, vecs);
 629
 630         return 0;
 631 }
 632
 633 static int
 634 uev_add_path (char * devname, struct vectors * vecs)
 635 {
 636         condlog(2, "%s: add path (uevent)", devname);
 637         return ev_add_path(devname, vecs);
 638 }
 639
 640 int
 641 ev_add_path (char * devname, struct vectors * vecs)
 642 {
 643         struct multipath * mpp;
 644         struct path * pp;
 645         char empty_buff[WWID_SIZE] = {0};
 646
 647         pp = find_path_by_dev(vecs->pathvec, devname);
 648
 649         if (pp) {
 650                 condlog(0, "%s: spurious uevent, path already in pathvec, %p",
 651                         devname, pp->mpp);
 652                 /*
 653                  * allow reconfig of orphaned path here
 654                  */
 655                 if (pp->mpp) return 1;
 656         }
 657         else {
 658                 /*
 659                  * get path vital state
 660                  */
 661                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 662                       devname, DI_ALL))) {
 663                         condlog(0, "%s: failed to store path info", devname);
 664                         return 1;
 665                 }
 666                 pp->checkint = conf->checkint;
 667         }
 668
 669         /*
 670          * need path UID to go any further
 671          */
 672         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 673                 condlog(0, "%s: failed to get path uid", devname);
 674                 return 1; /* leave path added to pathvec */
 675         }
 676
 677         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 678 rescan:
 679         if (mpp) {
 680                 if (adopt_paths(vecs->pathvec, mpp))
 681                         return 1; /* leave path added to pathvec */
 682
 683                 verify_paths(mpp, vecs, NULL);
 684                 condlog(0, "%s: ownership set to %s", pp->dev_t, mpp->alias);
 685                 mpp->action = ACT_RELOAD;
 686         }
 687         else {
 688                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 689                         mpp->action = ACT_CREATE;
 690                 else
 691                         return 1; /* leave path added to pathvec */
 692         }
 693
 694         /*
 695          * push the map to the device-mapper
 696          */
 697         if (setup_map(mpp)) {
 698                 condlog(0, "%s: failed to setup map for addition of new "
 699                         "path %s", mpp->alias, devname);
 700                 goto out;
 701         }
 702         /*
 703          * reload the map for the multipath mapped device
 704          */
 705         if (domap(mpp) <= 0) {
 706                 condlog(0, "%s: failed in domap for addition of new "
 707                         "path %s", mpp->alias, devname);
 708                 /*
 709                  * deal with asynchronous uevents :((
 710                  */
 711                 if (mpp->action == ACT_RELOAD) {
 712                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 713                         sleep(1);
 714                         update_mpp_paths(mpp, vecs->pathvec);
 715                         goto rescan;
 716                 }
 717                 else
 718                         goto out;
 719         }
 720         dm_lib_release();
 721
 722         /*
 723          * update our state from kernel regardless of create or reload
 724          */
 725         if (setup_multipath(vecs, mpp))
 726                 goto out;
 727
 728         if (mpp->action == ACT_CREATE &&
 729             start_waiter_thread(mpp, vecs))
 730                         goto out;
 731
 732         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 733         return 0;
 734
 735 out:
 736         remove_map(mpp, vecs, NULL, 1);
 737         return 1;
 738 }
 739
 740 static int
 741 uev_remove_path (char * devname, struct vectors * vecs)
 742 {
 743         condlog(2, "%s: remove path (uevent)", devname);
 744         return ev_remove_path(devname, vecs);
 745 }
 746
 747 int
 748 ev_remove_path (char * devname, struct vectors * vecs)
 749 {
 750         struct multipath * mpp;
 751         struct path * pp;
 752         int i;
 753         int rm_path = 1;
 754
 755         pp = find_path_by_dev(vecs->pathvec, devname);
 756
 757         if (!pp) {
 758                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 759                 return 1;
 760         }
 761
 762         /*
 763          * avoid referring to the map of an orphanned path
 764          */
 765         if ((mpp = pp->mpp)) {
 766
 767                 /*
 768                  * remove the map IFF removing the last path
 769                  */
 770                 if (pathcount(mpp, PATH_WILD) > 1) {
 771                         vector rpvec = vector_alloc();
 772
 773                         /*
 774                          * transform the mp->pg vector of vectors of paths
 775                          * into a mp->params string to feed the device-mapper
 776                          */
 777                         update_mpp_paths(mpp, vecs->pathvec);
 778                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 779                                 vector_del_slot(mpp->paths, i);
 780
 781                         if (VECTOR_SIZE(mpp->paths) == 0) {
 782                                 char alias[WWID_SIZE];
 783
 784                                 /*
 785                                  * flush_map will fail if the device is open
 786                                  */
 787                                 strncpy(alias, mpp->alias, WWID_SIZE);
 788                                 if (flush_map(mpp, vecs))
 789                                         rm_path = 0;
 790                                 else
 791                                         condlog(3, "%s: removed map after removing"
 792                                                 " multiple paths", alias);
 793                         }
 794                         else {
 795                                 if (setup_map(mpp)) {
 796                                         condlog(0, "%s: failed to setup map for"
 797                                                 " removal of path %s", mpp->alias, devname);
 798                                         free_pathvec(rpvec, KEEP_PATHS);
 799                                         goto out;
 800                                 }
 801                                 /*
 802                                  * reload the map
 803                                  */
 804                                 mpp->action = ACT_RELOAD;
 805                                 if (domap(mpp) <= 0) {
 806                                         condlog(0, "%s: failed in domap for "
 807                                                 "removal of path %s",
 808                                                 mpp->alias, devname);
 809                                         /*
 810                                          * Delete path from pathvec so that
 811                                          * update_mpp_paths wont find it later
 812                                          * when/if another path is removed.
 813                                          */
 814                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 815                                                 vector_del_slot(vecs->pathvec, i);
 816                                         free_path(pp);
 817                                         return 1;
 818                                 }
 819                                 /*
 820                                  * update our state from kernel
 821                                  */
 822                                 if (setup_multipath(vecs, mpp)) {
 823                                         free_pathvec(rpvec, KEEP_PATHS);
 824                                         goto out;
 825                                 }
 826
 827                                 condlog(3, "%s path removed from devmap %s",
 828                                         devname, mpp->alias);
 829                         }
 830                         free_pathvec(rpvec, KEEP_PATHS);
 831                 }
 832                 else {
 833                         char alias[WWID_SIZE];
 834
 835                         /*
 836                          * flush_map will fail if the device is open
 837                          */
 838                         strncpy(alias, mpp->alias, WWID_SIZE);
 839                         if (flush_map(mpp, vecs))
 840                                 rm_path = 0;
 841                         else
 842                                 condlog(3, "%s: removed map", alias);
 843                 }
 844         }
 845
 846         if (rm_path) {
 847                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 848                         vector_del_slot(vecs->pathvec, i);
 849                 free_path(pp);
 850         }
 851
 852         return 0;
 853
 854 out:
 855         remove_map(mpp, vecs, stop_waiter_thread, 1);
 856         return 1;
 857 }
 858
 859 static int
 860 map_discovery (struct vectors * vecs)
 861 {
 862         int i;
 863         struct multipath * mpp;
 864
 865         if (dm_get_maps(vecs->mpvec, "multipath"))
 866                 return 1;
 867
 868         vector_foreach_slot (vecs->mpvec, mpp, i)
 869                 if (setup_multipath(vecs, mpp))
 870                         return 1;
 871
 872         return 0;
 873 }
 874
 875 int
 876 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 877 {
 878         struct vectors * vecs;
 879         int r;
 880
 881         *reply = NULL;
 882         *len = 0;
 883         vecs = (struct vectors *)trigger_data;
 884
 885         pthread_cleanup_push(cleanup_lock, vecs->lock);
 886         lock(vecs->lock);
 887
 888         r = parse_cmd(str, reply, len, vecs);
 889
 890         if (r > 0) {
 891                 *reply = STRDUP("fail\n");
 892                 *len = strlen(*reply) + 1;
 893                 r = 1;
 894         }
 895         else if (!r && *len == 0) {
 896                 *reply = STRDUP("ok\n");
 897                 *len = strlen(*reply) + 1;
 898                 r = 0;
 899         }
 900         /* else if (r < 0) leave *reply alone */
 901
 902         lock_cleanup_pop(vecs->lock);
 903         return r;
 904 }
 905
 906 static int
 907 uev_discard(char * devpath)
 908 {
 909         char a[10], b[10];
 910
 911         /*
 912          * keep only block devices, discard partitions
 913          */
 914         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 915             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 916                 condlog(4, "discard event on %s", devpath);
 917                 return 1;
 918         }
 919         return 0;
 920 }
 921
 922 int
 923 uev_trigger (struct uevent * uev, void * trigger_data)
 924 {
 925         int r = 0;
 926         char devname[32];
 927         struct vectors * vecs;
 928
 929         vecs = (struct vectors *)trigger_data;
 930
 931         if (uev_discard(uev->devpath))
 932                 return 0;
 933
 934         basename(uev->devpath, devname);
 935         lock(vecs->lock);
 936
 937         /*
 938          * device map add/remove event
 939          */
 940         if (!strncmp(devname, "dm-", 3)) {
 941                 if (!strncmp(uev->action, "add", 3)) {
 942                         r = uev_add_map(devname, vecs);
 943                         goto out;
 944                 }
 945                 if (!strncmp(uev->action, "remove", 6)) {
 946                         r = uev_remove_map(devname, vecs);
 947                         goto out;
 948                 }
 949                 if (!strncmp(uev->action, "umount", 6)) {
 950                         r = uev_umount_map(devname, vecs);
 951                         goto out;
 952                 }
 953                 goto out;
 954         }
 955
 956         /*
 957          * path add/remove event
 958          */
 959         if (blacklist(conf->blist, devname))
 960                 goto out;
 961
 962         if (!strncmp(uev->action, "add", 3)) {
 963                 r = uev_add_path(devname, vecs);
 964                 goto out;
 965         }
 966         if (!strncmp(uev->action, "remove", 6)) {
 967                 r = uev_remove_path(devname, vecs);
 968                 goto out;
 969         }
 970
 971 out:
 972         unlock(vecs->lock);
 973         return r;
 974 }
 975
 976 static void *
 977 ueventloop (void * ap)
 978 {
 979         if (uevent_listen(&uev_trigger, ap))
 980                 fprintf(stderr, "error starting uevent listener");
 981
 982         return NULL;
 983 }
 984
 985 static void *
 986 uxlsnrloop (void * ap)
 987 {
 988         if (load_keys())
 989                 return NULL;
 990
 991         if (alloc_handlers())
 992                 return NULL;
 993
 994         add_handler(LIST+PATHS, cli_list_paths);
 995         add_handler(LIST+MAPS, cli_list_maps);
 996         add_handler(LIST+MAPS+STATS, cli_list_maps_stats);
 997         add_handler(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 998         add_handler(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 999         add_handler(ADD+PATH, cli_add_path);
1000         add_handler(DEL+PATH, cli_del_path);
1001         add_handler(ADD+MAP, cli_add_map);
1002         add_handler(DEL+MAP, cli_del_map);
1003         add_handler(SWITCH+MAP+GROUP, cli_switch_group);
1004         add_handler(RECONFIGURE, cli_reconfigure);
1005         add_handler(SUSPEND+MAP, cli_suspend);
1006         add_handler(RESUME+MAP, cli_resume);
1007         add_handler(REINSTATE+PATH, cli_reinstate);
1008         add_handler(FAIL+PATH, cli_fail);
1009
1010         uxsock_listen(&uxsock_trigger, ap);
1011
1012         return NULL;
1013 }
1014
1015 static int
1016 exit_daemon (int status)
1017 {
1018         if (status != 0)
1019                 fprintf(stderr, "bad exit status. see daemon.log\n");
1020
1021         condlog(3, "unlink pidfile");
1022         unlink(DEFAULT_PIDFILE);
1023
1024         lock(&exit_mutex);
1025         pthread_cond_signal(&exit_cond);
1026         unlock(&exit_mutex);
1027
1028         return status;
1029 }
1030
1031 static void
1032 fail_path (struct path * pp)
1033 {
1034         if (!pp->mpp)
1035                 return;
1036
1037         condlog(2, "checker failed path %s in map %s",
1038                  pp->dev_t, pp->mpp->alias);
1039
1040         dm_fail_path(pp->mpp->alias, pp->dev_t);
1041         update_queue_mode_del_path(pp->mpp);
1042 }
1043
1044 /*
1045  * caller must have locked the path list before calling that function
1046  */
1047 static void
1048 reinstate_path (struct path * pp)
1049 {
1050         if (!pp->mpp)
1051                 return;
1052
1053         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
1054                 condlog(0, "%s: reinstate failed", pp->dev_t);
1055         else {
1056                 condlog(2, "%s: reinstated", pp->dev_t);
1057                 update_queue_mode_add_path(pp->mpp);
1058         }
1059 }
1060
1061 static void
1062 enable_group(struct path * pp)
1063 {
1064         struct pathgroup * pgp;
1065
1066         /*
1067          * if path is added through uev_add_path, pgindex can be unset.
1068          * next update_strings() will set it, upon map reload event.
1069          *
1070          * we can safely return here, because upon map reload, all
1071          * PG will be enabled.
1072          */
1073         if (!pp->mpp->pg || !pp->pgindex)
1074                 return;
1075
1076         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1077
1078         if (pgp->status == PGSTATE_DISABLED) {
1079                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1080                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1081         }
1082 }
1083
1084 static void
1085 mpvec_garbage_collector (struct vectors * vecs)
1086 {
1087         struct multipath * mpp;
1088         int i;
1089
1090         vector_foreach_slot (vecs->mpvec, mpp, i) {
1091                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1092                         condlog(2, "%s: remove dead map", mpp->alias);
1093                         remove_map(mpp, vecs, stop_waiter_thread, 1);
1094                         i--;
1095                 }
1096         }
1097 }
1098
1099 static void
1100 defered_failback_tick (vector mpvec)
1101 {
1102         struct multipath * mpp;
1103         int i;
1104
1105         vector_foreach_slot (mpvec, mpp, i) {
1106                 /*
1107                  * defered failback getting sooner
1108                  */
1109                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1110                         mpp->failback_tick--;
1111
1112                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1113                                 switch_pathgroup(mpp);
1114                 }
1115         }
1116 }
1117
1118 static void
1119 retry_count_tick(vector mpvec)
1120 {
1121         struct multipath *mpp;
1122         int i;
1123
1124         vector_foreach_slot (mpvec, mpp, i) {
1125                 if (mpp->retry_tick) {
1126                         mpp->stat_total_queueing_time++;
1127                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1128                         if(--mpp->retry_tick == 0) {
1129                                 dm_queue_if_no_path(mpp->alias, 0);
1130                                 condlog(2, "%s: Disable queueing", mpp->alias);
1131                         }
1132                 }
1133         }
1134 }
1135
1136 static void *
1137 checkerloop (void *ap)
1138 {
1139         struct vectors *vecs;
1140         struct path *pp;
1141         int i, count = 0;
1142         int newstate;
1143         char checker_msg[MAX_CHECKER_MSG_SIZE];
1144
1145         mlockall(MCL_CURRENT | MCL_FUTURE);
1146
1147         memset(checker_msg, 0, MAX_CHECKER_MSG_SIZE);
1148         vecs = (struct vectors *)ap;
1149
1150         condlog(2, "path checkers start up");
1151
1152         /*
1153          * init the path check interval
1154          */
1155         vector_foreach_slot (vecs->pathvec, pp, i) {
1156                 pp->checkint = conf->checkint;
1157         }
1158
1159         while (1) {
1160                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1161                 lock(vecs->lock);
1162                 condlog(4, "tick");
1163
1164                 vector_foreach_slot (vecs->pathvec, pp, i) {
1165                         if (!pp->mpp)
1166                                 continue;
1167
1168                         if (pp->tick && --pp->tick)
1169                                 continue; /* don't check this path yet */
1170
1171                         /*
1172                          * provision a next check soonest,
1173                          * in case we exit abnormaly from here
1174                          */
1175                         pp->tick = conf->checkint;
1176
1177                         if (!pp->checkfn) {
1178                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
1179                                 select_checkfn(pp);
1180                         }
1181
1182                         if (!pp->checkfn) {
1183                                 condlog(0, "%s: checkfn is void", pp->dev);
1184                                 continue;
1185                         }
1186                         newstate = pp->checkfn(pp->fd, checker_msg,
1187                                                &pp->checker_context);
1188
1189                         if (newstate < 0) {
1190                                 condlog(2, "%s: unusable path", pp->dev);
1191                                 pathinfo(pp, conf->hwtable, 0);
1192                                 continue;
1193                         }
1194
1195                         if (newstate != pp->state) {
1196                                 pp->state = newstate;
1197                                 LOG_MSG(1, checker_msg);
1198
1199                                 /*
1200                                  * upon state change, reset the checkint
1201                                  * to the shortest delay
1202                                  */
1203                                 pp->checkint = conf->checkint;
1204
1205                                 if (newstate == PATH_DOWN ||
1206                                     newstate == PATH_SHAKY ||
1207                                     update_multipath_strings(pp->mpp,
1208                                                              vecs->pathvec)) {
1209                                         /*
1210                                          * proactively fail path in the DM
1211                                          */
1212                                         fail_path(pp);
1213
1214                                         /*
1215                                          * cancel scheduled failback
1216                                          */
1217                                         pp->mpp->failback_tick = 0;
1218
1219                                         pp->mpp->stat_path_failures++;
1220                                         continue;
1221                                 }
1222
1223                                 /*
1224                                  * reinstate this path
1225                                  */
1226                                 reinstate_path(pp);
1227
1228                                 /*
1229                                  * schedule [defered] failback
1230                                  */
1231                                 if (pp->mpp->pgfailback > 0)
1232                                         pp->mpp->failback_tick =
1233                                                 pp->mpp->pgfailback + 1;
1234                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
1235                                     need_switch_pathgroup(pp->mpp, 1))
1236                                         switch_pathgroup(pp->mpp);
1237
1238                                 /*
1239                                  * if at least one path is up in a group, and
1240                                  * the group is disabled, re-enable it
1241                                  */
1242                                 if (newstate == PATH_UP)
1243                                         enable_group(pp);
1244                         }
1245                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1246                                 LOG_MSG(4, checker_msg);
1247                                 /*
1248                                  * double the next check delay.
1249                                  * max at conf->max_checkint
1250                                  */
1251                                 if (pp->checkint < (conf->max_checkint / 2))
1252                                         pp->checkint = 2 * pp->checkint;
1253                                 else
1254                                         pp->checkint = conf->max_checkint;
1255
1256                                 pp->tick = pp->checkint;
1257                                 condlog(4, "%s: delay next check %is",
1258                                                 pp->dev_t, pp->tick);
1259
1260                         }
1261                         pp->state = newstate;
1262
1263                         /*
1264                          * path prio refreshing
1265                          */
1266                         condlog(4, "path prio refresh");
1267                         pathinfo(pp, conf->hwtable, DI_PRIO);
1268
1269                         if (need_switch_pathgroup(pp->mpp, 0)) {
1270                                 if (pp->mpp->pgfailback > 0)
1271                                         pp->mpp->failback_tick =
1272                                                 pp->mpp->pgfailback + 1;
1273                                 else if (pp->mpp->pgfailback ==
1274                                                 -FAILBACK_IMMEDIATE)
1275                                         switch_pathgroup(pp->mpp);
1276                         }
1277                 }
1278                 defered_failback_tick(vecs->mpvec);
1279                 retry_count_tick(vecs->mpvec);
1280
1281                 if (count)
1282                         count--;
1283                 else {
1284                         condlog(4, "map garbage collection");
1285                         mpvec_garbage_collector(vecs);
1286                         count = MAPGCINT;
1287                 }
1288
1289                 lock_cleanup_pop(vecs->lock);
1290                 sleep(1);
1291         }
1292         return NULL;
1293 }
1294
1295 int
1296 configure (struct vectors * vecs, int start_waiters)
1297 {
1298         struct multipath * mpp;
1299         struct path * pp;
1300         vector mpvec;
1301         int i;
1302
1303         if (!(vecs->pathvec = vector_alloc()))
1304                 return 1;
1305
1306         if (!(vecs->mpvec = vector_alloc()))
1307                 return 1;
1308
1309         if (!(mpvec = vector_alloc()))
1310                 return 1;
1311
1312         /*
1313          * probe for current path (from sysfs) and map (from dm) sets
1314          */
1315         path_discovery(vecs->pathvec, conf, DI_ALL);
1316
1317         vector_foreach_slot (vecs->pathvec, pp, i)
1318                 pp->checkint = conf->checkint;
1319
1320         if (map_discovery(vecs))
1321                 return 1;
1322
1323         /*
1324          * create new set of maps & push changed ones into dm
1325          */
1326         if (coalesce_paths(vecs, mpvec, NULL))
1327                 return 1;
1328
1329         /*
1330          * may need to remove some maps which are no longer relevant
1331          * e.g., due to blacklist changes in conf file
1332          */
1333         if (coalesce_maps(vecs, mpvec))
1334                 return 1;
1335
1336         dm_lib_release();
1337
1338         if (conf->verbosity > 2)
1339                 vector_foreach_slot(mpvec, mpp, i)
1340                         print_map(mpp);
1341
1342         /*
1343          * purge dm of old maps
1344          */
1345         remove_maps(vecs, NULL);
1346
1347         /*
1348          * save new set of maps formed by considering current path state
1349          */
1350         vecs->mpvec = mpvec;
1351
1352         /*
1353          * start dm event waiter threads for these new maps
1354          */
1355         vector_foreach_slot(vecs->mpvec, mpp, i) {
1356                 if (setup_multipath(vecs, mpp))
1357                         return 1;
1358                 if (start_waiters)
1359                         if (start_waiter_thread(mpp, vecs))
1360                                 return 1;
1361         }
1362         return 0;
1363 }
1364
1365 int
1366 reconfigure (struct vectors * vecs)
1367 {
1368         struct config * old = conf;
1369
1370         /*
1371          * free old map and path vectors ... they use old conf state
1372          */
1373         if (VECTOR_SIZE(vecs->mpvec))
1374                 remove_maps(vecs, stop_waiter_thread);
1375
1376         if (VECTOR_SIZE(vecs->pathvec))
1377                 free_pathvec(vecs->pathvec, FREE_PATHS);
1378
1379         conf = NULL;
1380
1381         if (load_config(DEFAULT_CONFIGFILE))
1382                 return 1;
1383
1384         conf->verbosity = old->verbosity;
1385
1386         if (!conf->checkint) {
1387                 conf->checkint = CHECKINT;
1388                 conf->max_checkint = MAX_CHECKINT;
1389         }
1390         configure(vecs, 1);
1391         free_config(old);
1392         return 0;
1393 }
1394
1395 static struct vectors *
1396 init_vecs (void)
1397 {
1398         struct vectors * vecs;
1399
1400         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1401
1402         if (!vecs)
1403                 return NULL;
1404
1405         vecs->lock =
1406                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1407
1408         if (!vecs->lock)
1409                 goto out;
1410
1411         vecs->pathvec = vector_alloc();
1412
1413         if (!vecs->pathvec)
1414                 goto out1;
1415
1416         vecs->mpvec = vector_alloc();
1417
1418         if (!vecs->mpvec)
1419                 goto out2;
1420
1421         pthread_mutex_init(vecs->lock, NULL);
1422
1423         return vecs;
1424
1425 out2:
1426         vector_free(vecs->pathvec);
1427 out1:
1428         FREE(vecs->lock);
1429 out:
1430         FREE(vecs);
1431         condlog(0, "failed to init paths");
1432         return NULL;
1433 }
1434
1435 static void *
1436 signal_set(int signo, void (*func) (int))
1437 {
1438         int r;
1439         struct sigaction sig;
1440         struct sigaction osig;
1441
1442         sig.sa_handler = func;
1443         sigemptyset(&sig.sa_mask);
1444         sig.sa_flags = 0;
1445
1446         r = sigaction(signo, &sig, &osig);
1447
1448         if (r < 0)
1449                 return (SIG_ERR);
1450         else
1451                 return (osig.sa_handler);
1452 }
1453
1454 static void
1455 sighup (int sig)
1456 {
1457         condlog(2, "reconfigure (SIGHUP)");
1458
1459         lock(gvecs->lock);
1460         reconfigure(gvecs);
1461         unlock(gvecs->lock);
1462
1463 #ifdef _DEBUG_
1464         dbg_free_final(NULL);
1465 #endif
1466 }
1467
1468 static void
1469 sigend (int sig)
1470 {
1471         exit_daemon(0);
1472 }
1473
1474 static void
1475 sigusr1 (int sig)
1476 {
1477         condlog(3, "SIGUSR1 received");
1478 }
1479
1480 static void
1481 signal_init(void)
1482 {
1483         signal_set(SIGHUP, sighup);
1484         signal_set(SIGUSR1, sigusr1);
1485         signal_set(SIGINT, sigend);
1486         signal_set(SIGTERM, sigend);
1487         signal_set(SIGKILL, sigend);
1488 }
1489
1490 static void
1491 setscheduler (void)
1492 {
1493         int res;
1494         static struct sched_param sched_param = {
1495                 sched_priority: 99
1496         };
1497
1498         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1499
1500         if (res == -1)
1501                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1502         return;
1503 }
1504
1505 static void
1506 set_oom_adj (int val)
1507 {
1508         FILE *fp;
1509
1510         fp = fopen("/proc/self/oom_adj", "w");
1511
1512         if (!fp)
1513                 return;
1514
1515         fprintf(fp, "%i", val);
1516         fclose(fp);
1517 }
1518
1519 static int
1520 child (void * param)
1521 {
1522         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1523         pthread_attr_t attr;
1524         struct vectors * vecs;
1525
1526         mlockall(MCL_CURRENT | MCL_FUTURE);
1527
1528         if (logsink)
1529                 log_thread_start();
1530
1531         condlog(2, "--------start up--------");
1532         condlog(2, "read " DEFAULT_CONFIGFILE);
1533
1534         if (load_config(DEFAULT_CONFIGFILE))
1535                 exit(1);
1536
1537         setlogmask(LOG_UPTO(conf->verbosity + 3));
1538
1539         /*
1540          * fill the voids left in the config file
1541          */
1542         if (!conf->checkint) {
1543                 conf->checkint = CHECKINT;
1544                 conf->max_checkint = MAX_CHECKINT;
1545         }
1546
1547         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1548                 if (logsink)
1549                         log_thread_stop();
1550
1551                 exit(1);
1552         }
1553         signal_init();
1554         setscheduler();
1555         set_oom_adj(-16);
1556         vecs = gvecs = init_vecs();
1557
1558         if (!vecs)
1559                 exit(1);
1560
1561         if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1562                 condlog(0, "can not find sysfs mount point");
1563                 exit(1);
1564         }
1565
1566         /*
1567          * fetch and configure both paths and multipaths
1568          */
1569         if (configure(vecs, 1)) {
1570                 condlog(0, "failure during configuration");
1571                 exit(1);
1572         }
1573
1574         /*
1575          * start threads
1576          */
1577         pthread_attr_init(&attr);
1578         pthread_attr_setstacksize(&attr, 64 * 1024);
1579         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1580
1581         pthread_create(&check_thr, &attr, checkerloop, vecs);
1582         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1583         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1584
1585         pthread_cond_wait(&exit_cond, &exit_mutex);
1586
1587         /*
1588          * exit path
1589          */
1590         lock(vecs->lock);
1591         remove_maps(vecs, stop_waiter_thread);
1592         free_pathvec(vecs->pathvec, FREE_PATHS);
1593
1594         pthread_cancel(check_thr);
1595         pthread_cancel(uevent_thr);
1596         pthread_cancel(uxlsnr_thr);
1597
1598         free_keys(keys);
1599         keys = NULL;
1600         free_handlers(handlers);
1601         handlers = NULL;
1602         free_polls();
1603
1604         unlock(vecs->lock);
1605         pthread_mutex_destroy(vecs->lock);
1606         FREE(vecs->lock);
1607         vecs->lock = NULL;
1608         FREE(vecs);
1609         vecs = NULL;
1610         free_config(conf);
1611         conf = NULL;
1612
1613         condlog(2, "--------shut down-------");
1614
1615         if (logsink)
1616                 log_thread_stop();
1617
1618         dm_lib_release();
1619         dm_lib_exit();
1620
1621 #ifdef _DEBUG_
1622         dbg_free_final(NULL);
1623 #endif
1624
1625         exit(0);
1626 }
1627
1628 static int
1629 daemonize(void)
1630 {
1631         int pid;
1632         int in_fd, out_fd;
1633
1634         if( (pid = fork()) < 0){
1635                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1636                 return -1;
1637         }
1638         else if (pid != 0)
1639                 return pid;
1640
1641         setsid();
1642
1643         if ( (pid = fork()) < 0)
1644                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1645         else if (pid != 0)
1646                 _exit(0);
1647
1648         in_fd = open("/dev/null", O_RDONLY);
1649         if (in_fd < 0){
1650                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1651                         strerror(errno));
1652                 _exit(0);
1653         }
1654         out_fd = open("/dev/console", O_WRONLY);
1655         if (out_fd < 0){
1656                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1657                         strerror(errno));
1658                 _exit(0);
1659         }
1660
1661         close(STDIN_FILENO);
1662         dup(in_fd);
1663         close(STDOUT_FILENO);
1664         dup(out_fd);
1665         close(STDERR_FILENO);
1666         dup(out_fd);
1667
1668         close(in_fd);
1669         close(out_fd);
1670         chdir("/");
1671         umask(0);
1672         return 0;
1673 }
1674
1675 int
1676 main (int argc, char *argv[])
1677 {
1678         extern char *optarg;
1679         extern int optind;
1680         int arg;
1681         int err;
1682
1683         logsink = 1;
1684
1685         if (getuid() != 0) {
1686                 fprintf(stderr, "need to be root\n");
1687                 exit(1);
1688         }
1689
1690         /* make sure we don't lock any path */
1691         chdir("/");
1692         umask(umask(077) | 022);
1693
1694         conf = alloc_config();
1695
1696         if (!conf)
1697                 exit(1);
1698
1699         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1700         switch(arg) {
1701                 case 'd':
1702                         logsink = 0;
1703                         //debug=1; /* ### comment me out ### */
1704                         break;
1705                 case 'v':
1706                         if (sizeof(optarg) > sizeof(char *) ||
1707                             !isdigit(optarg[0]))
1708                                 exit(1);
1709
1710                         conf->verbosity = atoi(optarg);
1711                         break;
1712                 case 'k':
1713                         uxclnt(optarg);
1714                         exit(0);
1715                 default:
1716                         ;
1717                 }
1718         }
1719
1720         if (!logsink)
1721                 err = 0;
1722         else
1723                 err = daemonize();
1724
1725         if (err < 0)
1726                 /* error */
1727                 exit(1);
1728         else if (err > 0)
1729                 /* parent dies */
1730                 exit(0);
1731         else
1732                 /* child lives */
1733                 return (child(NULL));
1734 }
1735