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         struct multipath * mpp;
 612
 613         condlog(2, "%s: umount map (uevent)", devname);
 614
 615         mpp = find_mp_by_str(vecs->mpvec, devname);
 616
 617         if (!mpp)
 618                 return 0;
 619
 620         update_mpp_paths(mpp, vecs->pathvec);
 621         verify_paths(mpp, vecs, NULL);
 622
 623         if (!VECTOR_SIZE(mpp->paths))
 624                 flush_map(mpp, vecs);
 625
 626         return 0;
 627 }
 628
 629 static int
 630 uev_add_path (char * devname, struct vectors * vecs)
 631 {
 632         condlog(2, "%s: add path (uevent)", devname);
 633         return ev_add_path(devname, vecs);
 634 }
 635
 636 int
 637 ev_add_path (char * devname, struct vectors * vecs)
 638 {
 639         struct multipath * mpp;
 640         struct path * pp;
 641         char empty_buff[WWID_SIZE] = {0};
 642
 643         pp = find_path_by_dev(vecs->pathvec, devname);
 644
 645         if (pp) {
 646                 condlog(0, "%s: spurious uevent, path already in pathvec, %p",
 647                         devname, pp->mpp);
 648                 /*
 649                  * allow reconfig of orphaned path here
 650                  */
 651                 if (pp->mpp) return 1;
 652         }
 653         else {
 654                 /*
 655                  * get path vital state
 656                  */
 657                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 658                       devname, DI_ALL))) {
 659                         condlog(0, "%s: failed to store path info", devname);
 660                         return 1;
 661                 }
 662                 pp->checkint = conf->checkint;
 663         }
 664
 665         /*
 666          * need path UID to go any further
 667          */
 668         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 669                 condlog(0, "%s: failed to get path uid", devname);
 670                 return 1; /* leave path added to pathvec */
 671         }
 672
 673         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 674 rescan:
 675         if (mpp) {
 676                 if (adopt_paths(vecs->pathvec, mpp))
 677                         return 1; /* leave path added to pathvec */
 678
 679                 verify_paths(mpp, vecs, NULL);
 680                 condlog(0, "%s: ownership set to %s", pp->dev_t, mpp->alias);
 681                 mpp->action = ACT_RELOAD;
 682         }
 683         else {
 684                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 685                         mpp->action = ACT_CREATE;
 686                 else
 687                         return 1; /* leave path added to pathvec */
 688         }
 689
 690         /*
 691          * push the map to the device-mapper
 692          */
 693         if (setup_map(mpp)) {
 694                 condlog(0, "%s: failed to setup map for addition of new "
 695                         "path %s", mpp->alias, devname);
 696                 goto out;
 697         }
 698         /*
 699          * reload the map for the multipath mapped device
 700          */
 701         if (domap(mpp) <= 0) {
 702                 condlog(0, "%s: failed in domap for addition of new "
 703                         "path %s", mpp->alias, devname);
 704                 /*
 705                  * deal with asynchronous uevents :((
 706                  */
 707                 if (mpp->action == ACT_RELOAD) {
 708                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 709                         sleep(1);
 710                         update_mpp_paths(mpp, vecs->pathvec);
 711                         goto rescan;
 712                 }
 713                 else
 714                         goto out;
 715         }
 716         dm_lib_release();
 717
 718         /*
 719          * update our state from kernel regardless of create or reload
 720          */
 721         if (setup_multipath(vecs, mpp))
 722                 goto out;
 723
 724         if (mpp->action == ACT_CREATE &&
 725             start_waiter_thread(mpp, vecs))
 726                         goto out;
 727
 728         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 729         return 0;
 730
 731 out:
 732         remove_map(mpp, vecs, NULL, 1);
 733         return 1;
 734 }
 735
 736 static int
 737 uev_remove_path (char * devname, struct vectors * vecs)
 738 {
 739         condlog(2, "%s: remove path (uevent)", devname);
 740         return ev_remove_path(devname, vecs);
 741 }
 742
 743 int
 744 ev_remove_path (char * devname, struct vectors * vecs)
 745 {
 746         struct multipath * mpp;
 747         struct path * pp;
 748         int i;
 749         int rm_path = 1;
 750
 751         pp = find_path_by_dev(vecs->pathvec, devname);
 752
 753         if (!pp) {
 754                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 755                 return 1;
 756         }
 757
 758         /*
 759          * avoid referring to the map of an orphanned path
 760          */
 761         if ((mpp = pp->mpp)) {
 762
 763                 /*
 764                  * remove the map IFF removing the last path
 765                  */
 766                 if (pathcount(mpp, PATH_WILD) > 1) {
 767                         vector rpvec = vector_alloc();
 768
 769                         /*
 770                          * transform the mp->pg vector of vectors of paths
 771                          * into a mp->params string to feed the device-mapper
 772                          */
 773                         update_mpp_paths(mpp, vecs->pathvec);
 774                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 775                                 vector_del_slot(mpp->paths, i);
 776
 777                         if (VECTOR_SIZE(mpp->paths) == 0) {
 778                                 char alias[WWID_SIZE];
 779
 780                                 /*
 781                                  * flush_map will fail if the device is open
 782                                  */
 783                                 strncpy(alias, mpp->alias, WWID_SIZE);
 784                                 if (flush_map(mpp, vecs))
 785                                         rm_path = 0;
 786                                 else
 787                                         condlog(3, "%s: removed map after removing"
 788                                                 " multiple paths", alias);
 789                         }
 790                         else {
 791                                 if (setup_map(mpp)) {
 792                                         condlog(0, "%s: failed to setup map for"
 793                                                 " removal of path %s", mpp->alias, devname);
 794                                         free_pathvec(rpvec, KEEP_PATHS);
 795                                         goto out;
 796                                 }
 797                                 /*
 798                                  * reload the map
 799                                  */
 800                                 mpp->action = ACT_RELOAD;
 801                                 if (domap(mpp) <= 0) {
 802                                         condlog(0, "%s: failed in domap for "
 803                                                 "removal of path %s",
 804                                                 mpp->alias, devname);
 805                                         /*
 806                                          * Delete path from pathvec so that
 807                                          * update_mpp_paths wont find it later
 808                                          * when/if another path is removed.
 809                                          */
 810                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 811                                                 vector_del_slot(vecs->pathvec, i);
 812                                         free_path(pp);
 813                                         return 1;
 814                                 }
 815                                 /*
 816                                  * update our state from kernel
 817                                  */
 818                                 if (setup_multipath(vecs, mpp)) {
 819                                         free_pathvec(rpvec, KEEP_PATHS);
 820                                         goto out;
 821                                 }
 822
 823                                 condlog(3, "%s path removed from devmap %s",
 824                                         devname, mpp->alias);
 825                         }
 826                         free_pathvec(rpvec, KEEP_PATHS);
 827                 }
 828                 else {
 829                         char alias[WWID_SIZE];
 830
 831                         /*
 832                          * flush_map will fail if the device is open
 833                          */
 834                         strncpy(alias, mpp->alias, WWID_SIZE);
 835                         if (flush_map(mpp, vecs))
 836                                 rm_path = 0;
 837                         else
 838                                 condlog(3, "%s: removed map", alias);
 839                 }
 840         }
 841
 842         if (rm_path) {
 843                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 844                         vector_del_slot(vecs->pathvec, i);
 845                 free_path(pp);
 846         }
 847
 848         return 0;
 849
 850 out:
 851         remove_map(mpp, vecs, stop_waiter_thread, 1);
 852         return 1;
 853 }
 854
 855 static int
 856 map_discovery (struct vectors * vecs)
 857 {
 858         int i;
 859         struct multipath * mpp;
 860
 861         if (dm_get_maps(vecs->mpvec, "multipath"))
 862                 return 1;
 863
 864         vector_foreach_slot (vecs->mpvec, mpp, i)
 865                 if (setup_multipath(vecs, mpp))
 866                         return 1;
 867
 868         return 0;
 869 }
 870
 871 int
 872 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 873 {
 874         struct vectors * vecs;
 875         int r;
 876
 877         *reply = NULL;
 878         *len = 0;
 879         vecs = (struct vectors *)trigger_data;
 880
 881         pthread_cleanup_push(cleanup_lock, vecs->lock);
 882         lock(vecs->lock);
 883
 884         r = parse_cmd(str, reply, len, vecs);
 885
 886         if (r > 0) {
 887                 *reply = STRDUP("fail\n");
 888                 *len = strlen(*reply) + 1;
 889                 r = 1;
 890         }
 891         else if (!r && *len == 0) {
 892                 *reply = STRDUP("ok\n");
 893                 *len = strlen(*reply) + 1;
 894                 r = 0;
 895         }
 896         /* else if (r < 0) leave *reply alone */
 897
 898         lock_cleanup_pop(vecs->lock);
 899         return r;
 900 }
 901
 902 static int
 903 uev_discard(char * devpath)
 904 {
 905         char a[10], b[10];
 906
 907         /*
 908          * keep only block devices, discard partitions
 909          */
 910         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 911             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 912                 condlog(4, "discard event on %s", devpath);
 913                 return 1;
 914         }
 915         return 0;
 916 }
 917
 918 int
 919 uev_trigger (struct uevent * uev, void * trigger_data)
 920 {
 921         int r = 0;
 922         char devname[32];
 923         struct vectors * vecs;
 924
 925         vecs = (struct vectors *)trigger_data;
 926
 927         if (uev_discard(uev->devpath))
 928                 return 0;
 929
 930         basename(uev->devpath, devname);
 931         lock(vecs->lock);
 932
 933         /*
 934          * device map add/remove event
 935          */
 936         if (!strncmp(devname, "dm-", 3)) {
 937                 if (!strncmp(uev->action, "add", 3)) {
 938                         r = uev_add_map(devname, vecs);
 939                         goto out;
 940                 }
 941                 if (!strncmp(uev->action, "remove", 6)) {
 942                         r = uev_remove_map(devname, vecs);
 943                         goto out;
 944                 }
 945                 if (!strncmp(uev->action, "umount", 6)) {
 946                         r = uev_umount_map(devname, vecs);
 947                         goto out;
 948                 }
 949                 goto out;
 950         }
 951
 952         /*
 953          * path add/remove event
 954          */
 955         if (blacklist(conf->blist, devname))
 956                 goto out;
 957
 958         if (!strncmp(uev->action, "add", 3)) {
 959                 r = uev_add_path(devname, vecs);
 960                 goto out;
 961         }
 962         if (!strncmp(uev->action, "remove", 6)) {
 963                 r = uev_remove_path(devname, vecs);
 964                 goto out;
 965         }
 966
 967 out:
 968         unlock(vecs->lock);
 969         return r;
 970 }
 971
 972 static void *
 973 ueventloop (void * ap)
 974 {
 975         if (uevent_listen(&uev_trigger, ap))
 976                 fprintf(stderr, "error starting uevent listener");
 977
 978         return NULL;
 979 }
 980
 981 static void *
 982 uxlsnrloop (void * ap)
 983 {
 984         if (load_keys())
 985                 return NULL;
 986
 987         if (alloc_handlers())
 988                 return NULL;
 989
 990         add_handler(LIST+PATHS, cli_list_paths);
 991         add_handler(LIST+MAPS, cli_list_maps);
 992         add_handler(LIST+MAPS+STATUS, cli_list_maps_status);
 993         add_handler(LIST+MAPS+STATS, cli_list_maps_stats);
 994         add_handler(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 995         add_handler(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 996         add_handler(ADD+PATH, cli_add_path);
 997         add_handler(DEL+PATH, cli_del_path);
 998         add_handler(ADD+MAP, cli_add_map);
 999         add_handler(DEL+MAP, cli_del_map);
1000         add_handler(SWITCH+MAP+GROUP, cli_switch_group);
1001         add_handler(RECONFIGURE, cli_reconfigure);
1002         add_handler(SUSPEND+MAP, cli_suspend);
1003         add_handler(RESUME+MAP, cli_resume);
1004         add_handler(REINSTATE+PATH, cli_reinstate);
1005         add_handler(FAIL+PATH, cli_fail);
1006
1007         uxsock_listen(&uxsock_trigger, ap);
1008
1009         return NULL;
1010 }
1011
1012 static int
1013 exit_daemon (int status)
1014 {
1015         if (status != 0)
1016                 fprintf(stderr, "bad exit status. see daemon.log\n");
1017
1018         condlog(3, "unlink pidfile");
1019         unlink(DEFAULT_PIDFILE);
1020
1021         lock(&exit_mutex);
1022         pthread_cond_signal(&exit_cond);
1023         unlock(&exit_mutex);
1024
1025         return status;
1026 }
1027
1028 static void
1029 fail_path (struct path * pp)
1030 {
1031         if (!pp->mpp)
1032                 return;
1033
1034         condlog(2, "checker failed path %s in map %s",
1035                  pp->dev_t, pp->mpp->alias);
1036
1037         dm_fail_path(pp->mpp->alias, pp->dev_t);
1038         update_queue_mode_del_path(pp->mpp);
1039 }
1040
1041 /*
1042  * caller must have locked the path list before calling that function
1043  */
1044 static void
1045 reinstate_path (struct path * pp)
1046 {
1047         if (!pp->mpp)
1048                 return;
1049
1050         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
1051                 condlog(0, "%s: reinstate failed", pp->dev_t);
1052         else {
1053                 condlog(2, "%s: reinstated", pp->dev_t);
1054                 update_queue_mode_add_path(pp->mpp);
1055         }
1056 }
1057
1058 static void
1059 enable_group(struct path * pp)
1060 {
1061         struct pathgroup * pgp;
1062
1063         /*
1064          * if path is added through uev_add_path, pgindex can be unset.
1065          * next update_strings() will set it, upon map reload event.
1066          *
1067          * we can safely return here, because upon map reload, all
1068          * PG will be enabled.
1069          */
1070         if (!pp->mpp->pg || !pp->pgindex)
1071                 return;
1072
1073         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1074
1075         if (pgp->status == PGSTATE_DISABLED) {
1076                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1077                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1078         }
1079 }
1080
1081 static void
1082 mpvec_garbage_collector (struct vectors * vecs)
1083 {
1084         struct multipath * mpp;
1085         int i;
1086
1087         vector_foreach_slot (vecs->mpvec, mpp, i) {
1088                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1089                         condlog(2, "%s: remove dead map", mpp->alias);
1090                         remove_map(mpp, vecs, stop_waiter_thread, 1);
1091                         i--;
1092                 }
1093         }
1094 }
1095
1096 static void
1097 defered_failback_tick (vector mpvec)
1098 {
1099         struct multipath * mpp;
1100         int i;
1101
1102         vector_foreach_slot (mpvec, mpp, i) {
1103                 /*
1104                  * defered failback getting sooner
1105                  */
1106                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1107                         mpp->failback_tick--;
1108
1109                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1110                                 switch_pathgroup(mpp);
1111                 }
1112         }
1113 }
1114
1115 static void
1116 retry_count_tick(vector mpvec)
1117 {
1118         struct multipath *mpp;
1119         int i;
1120
1121         vector_foreach_slot (mpvec, mpp, i) {
1122                 if (mpp->retry_tick) {
1123                         mpp->stat_total_queueing_time++;
1124                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1125                         if(--mpp->retry_tick == 0) {
1126                                 dm_queue_if_no_path(mpp->alias, 0);
1127                                 condlog(2, "%s: Disable queueing", mpp->alias);
1128                         }
1129                 }
1130         }
1131 }
1132
1133 static void *
1134 checkerloop (void *ap)
1135 {
1136         struct vectors *vecs;
1137         struct path *pp;
1138         int i, count = 0;
1139         int newstate;
1140         char checker_msg[MAX_CHECKER_MSG_SIZE];
1141
1142         mlockall(MCL_CURRENT | MCL_FUTURE);
1143
1144         memset(checker_msg, 0, MAX_CHECKER_MSG_SIZE);
1145         vecs = (struct vectors *)ap;
1146
1147         condlog(2, "path checkers start up");
1148
1149         /*
1150          * init the path check interval
1151          */
1152         vector_foreach_slot (vecs->pathvec, pp, i) {
1153                 pp->checkint = conf->checkint;
1154         }
1155
1156         while (1) {
1157                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1158                 lock(vecs->lock);
1159                 condlog(4, "tick");
1160
1161                 vector_foreach_slot (vecs->pathvec, pp, i) {
1162                         if (!pp->mpp)
1163                                 continue;
1164
1165                         if (pp->tick && --pp->tick)
1166                                 continue; /* don't check this path yet */
1167
1168                         /*
1169                          * provision a next check soonest,
1170                          * in case we exit abnormaly from here
1171                          */
1172                         pp->tick = conf->checkint;
1173
1174                         if (!pp->checkfn) {
1175                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
1176                                 select_checkfn(pp);
1177                         }
1178
1179                         if (!pp->checkfn) {
1180                                 condlog(0, "%s: checkfn is void", pp->dev);
1181                                 continue;
1182                         }
1183                         newstate = pp->checkfn(pp->fd, checker_msg,
1184                                                &pp->checker_context);
1185
1186                         if (newstate < 0) {
1187                                 condlog(2, "%s: unusable path", pp->dev);
1188                                 pathinfo(pp, conf->hwtable, 0);
1189                                 continue;
1190                         }
1191
1192                         if (newstate != pp->state) {
1193                                 pp->state = newstate;
1194                                 LOG_MSG(1, checker_msg);
1195
1196                                 /*
1197                                  * upon state change, reset the checkint
1198                                  * to the shortest delay
1199                                  */
1200                                 pp->checkint = conf->checkint;
1201
1202                                 if (newstate == PATH_DOWN ||
1203                                     newstate == PATH_SHAKY ||
1204                                     update_multipath_strings(pp->mpp,
1205                                                              vecs->pathvec)) {
1206                                         /*
1207                                          * proactively fail path in the DM
1208                                          */
1209                                         fail_path(pp);
1210
1211                                         /*
1212                                          * cancel scheduled failback
1213                                          */
1214                                         pp->mpp->failback_tick = 0;
1215
1216                                         pp->mpp->stat_path_failures++;
1217                                         continue;
1218                                 }
1219
1220                                 /*
1221                                  * reinstate this path
1222                                  */
1223                                 reinstate_path(pp);
1224
1225                                 /*
1226                                  * schedule [defered] failback
1227                                  */
1228                                 if (pp->mpp->pgfailback > 0)
1229                                         pp->mpp->failback_tick =
1230                                                 pp->mpp->pgfailback + 1;
1231                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
1232                                     need_switch_pathgroup(pp->mpp, 1))
1233                                         switch_pathgroup(pp->mpp);
1234
1235                                 /*
1236                                  * if at least one path is up in a group, and
1237                                  * the group is disabled, re-enable it
1238                                  */
1239                                 if (newstate == PATH_UP)
1240                                         enable_group(pp);
1241                         }
1242                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1243                                 LOG_MSG(4, checker_msg);
1244                                 /*
1245                                  * double the next check delay.
1246                                  * max at conf->max_checkint
1247                                  */
1248                                 if (pp->checkint < (conf->max_checkint / 2))
1249                                         pp->checkint = 2 * pp->checkint;
1250                                 else
1251                                         pp->checkint = conf->max_checkint;
1252
1253                                 pp->tick = pp->checkint;
1254                                 condlog(4, "%s: delay next check %is",
1255                                                 pp->dev_t, pp->tick);
1256
1257                         }
1258                         pp->state = newstate;
1259
1260                         /*
1261                          * path prio refreshing
1262                          */
1263                         condlog(4, "path prio refresh");
1264                         pathinfo(pp, conf->hwtable, DI_PRIO);
1265
1266                         if (need_switch_pathgroup(pp->mpp, 0)) {
1267                                 if (pp->mpp->pgfailback > 0)
1268                                         pp->mpp->failback_tick =
1269                                                 pp->mpp->pgfailback + 1;
1270                                 else if (pp->mpp->pgfailback ==
1271                                                 -FAILBACK_IMMEDIATE)
1272                                         switch_pathgroup(pp->mpp);
1273                         }
1274                 }
1275                 defered_failback_tick(vecs->mpvec);
1276                 retry_count_tick(vecs->mpvec);
1277
1278                 if (count)
1279                         count--;
1280                 else {
1281                         condlog(4, "map garbage collection");
1282                         mpvec_garbage_collector(vecs);
1283                         count = MAPGCINT;
1284                 }
1285
1286                 lock_cleanup_pop(vecs->lock);
1287                 sleep(1);
1288         }
1289         return NULL;
1290 }
1291
1292 int
1293 configure (struct vectors * vecs, int start_waiters)
1294 {
1295         struct multipath * mpp;
1296         struct path * pp;
1297         vector mpvec;
1298         int i;
1299
1300         if (!(vecs->pathvec = vector_alloc()))
1301                 return 1;
1302
1303         if (!(vecs->mpvec = vector_alloc()))
1304                 return 1;
1305
1306         if (!(mpvec = vector_alloc()))
1307                 return 1;
1308
1309         /*
1310          * probe for current path (from sysfs) and map (from dm) sets
1311          */
1312         path_discovery(vecs->pathvec, conf, DI_ALL);
1313
1314         vector_foreach_slot (vecs->pathvec, pp, i)
1315                 pp->checkint = conf->checkint;
1316
1317         if (map_discovery(vecs))
1318                 return 1;
1319
1320         /*
1321          * create new set of maps & push changed ones into dm
1322          */
1323         if (coalesce_paths(vecs, mpvec, NULL))
1324                 return 1;
1325
1326         /*
1327          * may need to remove some maps which are no longer relevant
1328          * e.g., due to blacklist changes in conf file
1329          */
1330         if (coalesce_maps(vecs, mpvec))
1331                 return 1;
1332
1333         dm_lib_release();
1334
1335         if (conf->verbosity > 2)
1336                 vector_foreach_slot(mpvec, mpp, i)
1337                         print_map(mpp);
1338
1339         /*
1340          * purge dm of old maps
1341          */
1342         remove_maps(vecs, NULL);
1343
1344         /*
1345          * save new set of maps formed by considering current path state
1346          */
1347         vecs->mpvec = mpvec;
1348
1349         /*
1350          * start dm event waiter threads for these new maps
1351          */
1352         vector_foreach_slot(vecs->mpvec, mpp, i) {
1353                 if (setup_multipath(vecs, mpp))
1354                         return 1;
1355                 if (start_waiters)
1356                         if (start_waiter_thread(mpp, vecs))
1357                                 return 1;
1358         }
1359         return 0;
1360 }
1361
1362 int
1363 reconfigure (struct vectors * vecs)
1364 {
1365         struct config * old = conf;
1366
1367         /*
1368          * free old map and path vectors ... they use old conf state
1369          */
1370         if (VECTOR_SIZE(vecs->mpvec))
1371                 remove_maps(vecs, stop_waiter_thread);
1372
1373         if (VECTOR_SIZE(vecs->pathvec))
1374                 free_pathvec(vecs->pathvec, FREE_PATHS);
1375
1376         conf = NULL;
1377
1378         if (load_config(DEFAULT_CONFIGFILE))
1379                 return 1;
1380
1381         conf->verbosity = old->verbosity;
1382
1383         if (!conf->checkint) {
1384                 conf->checkint = DEFAULT_CHECKINT;
1385                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1386         }
1387         configure(vecs, 1);
1388         free_config(old);
1389         return 0;
1390 }
1391
1392 static struct vectors *
1393 init_vecs (void)
1394 {
1395         struct vectors * vecs;
1396
1397         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1398
1399         if (!vecs)
1400                 return NULL;
1401
1402         vecs->lock =
1403                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1404
1405         if (!vecs->lock)
1406                 goto out;
1407
1408         vecs->pathvec = vector_alloc();
1409
1410         if (!vecs->pathvec)
1411                 goto out1;
1412
1413         vecs->mpvec = vector_alloc();
1414
1415         if (!vecs->mpvec)
1416                 goto out2;
1417
1418         pthread_mutex_init(vecs->lock, NULL);
1419
1420         return vecs;
1421
1422 out2:
1423         vector_free(vecs->pathvec);
1424 out1:
1425         FREE(vecs->lock);
1426 out:
1427         FREE(vecs);
1428         condlog(0, "failed to init paths");
1429         return NULL;
1430 }
1431
1432 static void *
1433 signal_set(int signo, void (*func) (int))
1434 {
1435         int r;
1436         struct sigaction sig;
1437         struct sigaction osig;
1438
1439         sig.sa_handler = func;
1440         sigemptyset(&sig.sa_mask);
1441         sig.sa_flags = 0;
1442
1443         r = sigaction(signo, &sig, &osig);
1444
1445         if (r < 0)
1446                 return (SIG_ERR);
1447         else
1448                 return (osig.sa_handler);
1449 }
1450
1451 static void
1452 sighup (int sig)
1453 {
1454         condlog(2, "reconfigure (SIGHUP)");
1455
1456         lock(gvecs->lock);
1457         reconfigure(gvecs);
1458         unlock(gvecs->lock);
1459
1460 #ifdef _DEBUG_
1461         dbg_free_final(NULL);
1462 #endif
1463 }
1464
1465 static void
1466 sigend (int sig)
1467 {
1468         exit_daemon(0);
1469 }
1470
1471 static void
1472 sigusr1 (int sig)
1473 {
1474         condlog(3, "SIGUSR1 received");
1475 }
1476
1477 static void
1478 signal_init(void)
1479 {
1480         signal_set(SIGHUP, sighup);
1481         signal_set(SIGUSR1, sigusr1);
1482         signal_set(SIGINT, sigend);
1483         signal_set(SIGTERM, sigend);
1484         signal_set(SIGKILL, sigend);
1485 }
1486
1487 static void
1488 setscheduler (void)
1489 {
1490         int res;
1491         static struct sched_param sched_param = {
1492                 sched_priority: 99
1493         };
1494
1495         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1496
1497         if (res == -1)
1498                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1499         return;
1500 }
1501
1502 static void
1503 set_oom_adj (int val)
1504 {
1505         FILE *fp;
1506
1507         fp = fopen("/proc/self/oom_adj", "w");
1508
1509         if (!fp)
1510                 return;
1511
1512         fprintf(fp, "%i", val);
1513         fclose(fp);
1514 }
1515
1516 static int
1517 child (void * param)
1518 {
1519         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1520         pthread_attr_t attr;
1521         struct vectors * vecs;
1522
1523         mlockall(MCL_CURRENT | MCL_FUTURE);
1524
1525         if (logsink)
1526                 log_thread_start();
1527
1528         condlog(2, "--------start up--------");
1529         condlog(2, "read " DEFAULT_CONFIGFILE);
1530
1531         if (load_config(DEFAULT_CONFIGFILE))
1532                 exit(1);
1533
1534         setlogmask(LOG_UPTO(conf->verbosity + 3));
1535
1536         /*
1537          * fill the voids left in the config file
1538          */
1539         if (!conf->checkint) {
1540                 conf->checkint = DEFAULT_CHECKINT;
1541                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1542         }
1543
1544         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1545                 if (logsink)
1546                         log_thread_stop();
1547
1548                 exit(1);
1549         }
1550         signal_init();
1551         setscheduler();
1552         set_oom_adj(-16);
1553         vecs = gvecs = init_vecs();
1554
1555         if (!vecs)
1556                 exit(1);
1557
1558         if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1559                 condlog(0, "can not find sysfs mount point");
1560                 exit(1);
1561         }
1562
1563         /*
1564          * fetch and configure both paths and multipaths
1565          */
1566         if (configure(vecs, 1)) {
1567                 condlog(0, "failure during configuration");
1568                 exit(1);
1569         }
1570
1571         /*
1572          * start threads
1573          */
1574         pthread_attr_init(&attr);
1575         pthread_attr_setstacksize(&attr, 64 * 1024);
1576         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1577
1578         pthread_create(&check_thr, &attr, checkerloop, vecs);
1579         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1580         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1581
1582         pthread_cond_wait(&exit_cond, &exit_mutex);
1583
1584         /*
1585          * exit path
1586          */
1587         lock(vecs->lock);
1588         remove_maps(vecs, stop_waiter_thread);
1589         free_pathvec(vecs->pathvec, FREE_PATHS);
1590
1591         pthread_cancel(check_thr);
1592         pthread_cancel(uevent_thr);
1593         pthread_cancel(uxlsnr_thr);
1594
1595         free_keys(keys);
1596         keys = NULL;
1597         free_handlers(handlers);
1598         handlers = NULL;
1599         free_polls();
1600
1601         unlock(vecs->lock);
1602         pthread_mutex_destroy(vecs->lock);
1603         FREE(vecs->lock);
1604         vecs->lock = NULL;
1605         FREE(vecs);
1606         vecs = NULL;
1607         free_config(conf);
1608         conf = NULL;
1609
1610         condlog(2, "--------shut down-------");
1611
1612         if (logsink)
1613                 log_thread_stop();
1614
1615         dm_lib_release();
1616         dm_lib_exit();
1617
1618 #ifdef _DEBUG_
1619         dbg_free_final(NULL);
1620 #endif
1621
1622         exit(0);
1623 }
1624
1625 static int
1626 daemonize(void)
1627 {
1628         int pid;
1629         int in_fd, out_fd;
1630
1631         if( (pid = fork()) < 0){
1632                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1633                 return -1;
1634         }
1635         else if (pid != 0)
1636                 return pid;
1637
1638         setsid();
1639
1640         if ( (pid = fork()) < 0)
1641                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1642         else if (pid != 0)
1643                 _exit(0);
1644
1645         in_fd = open("/dev/null", O_RDONLY);
1646         if (in_fd < 0){
1647                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1648                         strerror(errno));
1649                 _exit(0);
1650         }
1651         out_fd = open("/dev/console", O_WRONLY);
1652         if (out_fd < 0){
1653                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1654                         strerror(errno));
1655                 _exit(0);
1656         }
1657
1658         close(STDIN_FILENO);
1659         dup(in_fd);
1660         close(STDOUT_FILENO);
1661         dup(out_fd);
1662         close(STDERR_FILENO);
1663         dup(out_fd);
1664
1665         close(in_fd);
1666         close(out_fd);
1667         chdir("/");
1668         umask(0);
1669         return 0;
1670 }
1671
1672 int
1673 main (int argc, char *argv[])
1674 {
1675         extern char *optarg;
1676         extern int optind;
1677         int arg;
1678         int err;
1679
1680         logsink = 1;
1681
1682         if (getuid() != 0) {
1683                 fprintf(stderr, "need to be root\n");
1684                 exit(1);
1685         }
1686
1687         /* make sure we don't lock any path */
1688         chdir("/");
1689         umask(umask(077) | 022);
1690
1691         conf = alloc_config();
1692
1693         if (!conf)
1694                 exit(1);
1695
1696         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1697         switch(arg) {
1698                 case 'd':
1699                         logsink = 0;
1700                         //debug=1; /* ### comment me out ### */
1701                         break;
1702                 case 'v':
1703                         if (sizeof(optarg) > sizeof(char *) ||
1704                             !isdigit(optarg[0]))
1705                                 exit(1);
1706
1707                         conf->verbosity = atoi(optarg);
1708                         break;
1709                 case 'k':
1710                         uxclnt(optarg);
1711                         exit(0);
1712                 default:
1713                         ;
1714                 }
1715         }
1716
1717         if (!logsink)
1718                 err = 0;
1719         else
1720                 err = daemonize();
1721
1722         if (err < 0)
1723                 /* error */
1724                 exit(1);
1725         else if (err > 0)
1726                 /* parent dies */
1727                 exit(0);
1728         else
1729                 /* child lives */
1730                 return (child(NULL));
1731 }
1732