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
  27 /*
  28  * libmultipath
  29  */
  30 #include <parser.h>
  31 #include <vector.h>
  32 #include <memory.h>
  33 #include <config.h>
  34 #include <util.h>
  35 #include <hwtable.h>
  36 #include <defaults.h>
  37 #include <structs.h>
  38 #include <callout.h>
  39 #include <blacklist.h>
  40 #include <structs_vec.h>
  41 #include <dmparser.h>
  42 #include <devmapper.h>
  43 #include <dict.h>
  44 #include <discovery.h>
  45 #include <debug.h>
  46 #include <propsel.h>
  47 #include <uevent.h>
  48 #include <switchgroup.h>
  49 #include <print.h>
  50 #include <configure.h>
  51
  52 #include "main.h"
  53 #include "pidfile.h"
  54 #include "uxlsnr.h"
  55 #include "uxclnt.h"
  56 #include "cli.h"
  57 #include "cli_handlers.h"
  58
  59 #define FILE_NAME_SIZE 256
  60 #define CMDSIZE 160
  61
  62 #define LOG_MSG(a,b) \
  63         if (strlen(b)) condlog(a, "%s: %s", pp->dev_t, b);
  64
  65 #ifdef LCKDBG
  66 #define lock(a) \
  67         fprintf(stderr, "%s:%s(%i) lock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  68         pthread_mutex_lock(a)
  69 #define unlock(a) \
  70         fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  71         pthread_mutex_unlock(a)
  72 #define lock_cleanup_pop(a) \
  73         fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  74         pthread_cleanup_pop(1);
  75 #else
  76 #define lock(a) pthread_mutex_lock(a)
  77 #define unlock(a) pthread_mutex_unlock(a)
  78 #define lock_cleanup_pop(a) pthread_cleanup_pop(1);
  79 #endif
  80
  81 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  82 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  83
  84 /*
  85  * structs
  86  */
  87 struct vectors * gvecs; /* global copy of vecs for use in sig handlers */
  88
  89 static struct event_thread *
  90 alloc_waiter (void)
  91 {
  92
  93         struct event_thread * wp;
  94
  95         wp = (struct event_thread *)MALLOC(sizeof(struct event_thread));
  96
  97         return wp;
  98 }
  99
 100 static void
 101 free_waiter (void * data)
 102 {
 103         struct event_thread * wp = (struct event_thread *)data;
 104
 105         /*
 106          * indicate in mpp that the wp is already freed storage
 107          */
 108         lock(wp->vecs->lock);
 109
 110         if (wp->mpp)
 111                 /*
 112                  * be careful, mpp may already be freed -- null if so
 113                  */
 114                 wp->mpp->waiter = NULL;
 115         else
 116                 condlog(3, "free_waiter, mpp freed before wp=%p,", wp);
 117
 118         unlock(wp->vecs->lock);
 119
 120         if (wp->dmt)
 121                 dm_task_destroy(wp->dmt);
 122
 123         FREE(wp);
 124 }
 125
 126 static void
 127 stop_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 128 {
 129         struct event_thread * wp = (struct event_thread *)mpp->waiter;
 130
 131         if (!wp) {
 132                 condlog(3, "%s: no waiter thread", mpp->alias);
 133                 return;
 134         }
 135         condlog(2, "%s: stop event checker thread", wp->mapname);
 136         pthread_kill((pthread_t)wp->thread, SIGUSR1);
 137 }
 138
 139 static void
 140 cleanup_lock (void * data)
 141 {
 142         unlock((pthread_mutex_t *)data);
 143 }
 144
 145 /*
 146  * mpp->no_path_retry:
 147  *   -2 (QUEUE) : queue_if_no_path enabled, never turned off
 148  *   -1 (FAIL)  : fail_if_no_path
 149  *    0 (UNDEF) : nothing
 150  *   >0         : queue_if_no_path enabled, turned off after polling n times
 151  */
 152 static void
 153 update_queue_mode_del_path(struct multipath *mpp)
 154 {
 155         if (--mpp->nr_active == 0 && mpp->no_path_retry > 0) {
 156                 /*
 157                  * Enter retry mode.
 158                  * meaning of +1: retry_tick may be decremented in
 159                  *                checkerloop before starting retry.
 160                  */
 161                 mpp->stat_queueing_timeouts++;
 162                 mpp->retry_tick = mpp->no_path_retry * conf->checkint + 1;
 163                 condlog(1, "%s: Entering recovery mode: max_retries=%d",
 164                         mpp->alias, mpp->no_path_retry);
 165         }
 166         condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
 167 }
 168
 169 static void
 170 update_queue_mode_add_path(struct multipath *mpp)
 171 {
 172         if (mpp->nr_active++ == 0 && mpp->no_path_retry > 0) {
 173                 /* come back to normal mode from retry mode */
 174                 mpp->retry_tick = 0;
 175                 dm_queue_if_no_path(mpp->alias, 1);
 176                 condlog(2, "%s: queue_if_no_path enabled", mpp->alias);
 177                 condlog(1, "%s: Recovered to normal mode", mpp->alias);
 178         }
 179         condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
 180 }
 181
 182 static int
 183 need_switch_pathgroup (struct multipath * mpp, int refresh)
 184 {
 185         struct pathgroup * pgp;
 186         struct path * pp;
 187         int i, j;
 188
 189         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
 190                 return 0;
 191
 192         /*
 193          * Refresh path priority values
 194          */
 195         if (refresh)
 196                 vector_foreach_slot (mpp->pg, pgp, i)
 197                         vector_foreach_slot (pgp->paths, pp, j)
 198                                 pathinfo(pp, conf->hwtable, DI_PRIO);
 199
 200         mpp->bestpg = select_path_group(mpp);
 201
 202         if (mpp->bestpg != mpp->nextpg)
 203                 return 1;
 204
 205         return 0;
 206 }
 207
 208 static void
 209 switch_pathgroup (struct multipath * mpp)
 210 {
 211         mpp->stat_switchgroup++;
 212         dm_switchgroup(mpp->alias, mpp->bestpg);
 213         condlog(2, "%s: switch to path group #%i",
 214                  mpp->alias, mpp->bestpg);
 215 }
 216
 217 static int
 218 coalesce_maps(struct vectors *vecs, vector nmpv)
 219 {
 220         struct multipath * ompp;
 221         vector ompv = vecs->mpvec;
 222         int i, j;
 223
 224         vector_foreach_slot (ompv, ompp, i) {
 225                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 226                         /*
 227                          * remove all current maps not allowed by the
 228                          * current configuration
 229                          */
 230                         if (dm_flush_map(ompp->alias, DEFAULT_TARGET)) {
 231                                 condlog(0, "%s: unable to flush devmap",
 232                                         ompp->alias);
 233                                 /*
 234                                  * may be just because the device is open
 235                                  */
 236                                 if (!vector_alloc_slot(nmpv))
 237                                         return 1;
 238
 239                                 vector_set_slot(nmpv, ompp);
 240                                 setup_multipath(vecs, ompp);
 241
 242                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 243                                         vector_del_slot(ompv, j);
 244
 245                                 continue;
 246                         }
 247                         else {
 248                                 dm_lib_release();
 249                                 condlog(3, "%s devmap removed", ompp->alias);
 250                         }
 251                 }
 252         }
 253         return 0;
 254 }
 255
 256 static int
 257 update_multipath (struct vectors *vecs, char *mapname)
 258 {
 259         struct multipath *mpp;
 260         struct pathgroup  *pgp;
 261         struct path *pp;
 262         int i, j;
 263         int r = 1;
 264
 265         mpp = find_mp_by_alias(vecs->mpvec, mapname);
 266
 267         if (!mpp)
 268                 goto out;
 269
 270         free_pgvec(mpp->pg, KEEP_PATHS);
 271         mpp->pg = NULL;
 272
 273         if (setup_multipath(vecs, mpp))
 274                 goto out; /* mpp freed in setup_multipath */
 275
 276         /*
 277          * compare checkers states with DM states
 278          */
 279         vector_foreach_slot (mpp->pg, pgp, i) {
 280                 vector_foreach_slot (pgp->paths, pp, j) {
 281                         if (pp->dmstate != PSTATE_FAILED)
 282                                 continue;
 283
 284                         if (pp->state != PATH_DOWN) {
 285                                 condlog(2, "%s: mark as failed", pp->dev_t);
 286                                 mpp->stat_path_failures++;
 287                                 pp->state = PATH_DOWN;
 288                                 update_queue_mode_del_path(mpp);
 289
 290                                 /*
 291                                  * if opportune,
 292                                  * schedule the next check earlier
 293                                  */
 294                                 if (pp->tick > conf->checkint)
 295                                         pp->tick = conf->checkint;
 296                         }
 297                 }
 298         }
 299         r = 0;
 300 out:
 301         if (r)
 302                 condlog(0, "failed to update multipath");
 303
 304         return r;
 305 }
 306
 307 static sigset_t unblock_signals(void)
 308 {
 309         sigset_t set, old;
 310
 311         sigemptyset(&set);
 312         sigaddset(&set, SIGHUP);
 313         sigaddset(&set, SIGUSR1);
 314         pthread_sigmask(SIG_UNBLOCK, &set, &old);
 315         return old;
 316 }
 317
 318 /*
 319  * returns the reschedule delay
 320  * negative means *stop*
 321  */
 322 static int
 323 waiteventloop (struct event_thread * waiter)
 324 {
 325         sigset_t set;
 326         int event_nr;
 327         int r;
 328
 329         if (!waiter->event_nr)
 330                 waiter->event_nr = dm_geteventnr(waiter->mapname);
 331
 332         if (!(waiter->dmt = dm_task_create(DM_DEVICE_WAITEVENT))) {
 333                 condlog(0, "%s: devmap event #%i dm_task_create error",
 334                                 waiter->mapname, waiter->event_nr);
 335                 return 1;
 336         }
 337
 338         if (!dm_task_set_name(waiter->dmt, waiter->mapname)) {
 339                 condlog(0, "%s: devmap event #%i dm_task_set_name error",
 340                                 waiter->mapname, waiter->event_nr);
 341                 dm_task_destroy(waiter->dmt);
 342                 return 1;
 343         }
 344
 345         if (waiter->event_nr && !dm_task_set_event_nr(waiter->dmt,
 346                                                       waiter->event_nr)) {
 347                 condlog(0, "%s: devmap event #%i dm_task_set_event_nr error",
 348                                 waiter->mapname, waiter->event_nr);
 349                 dm_task_destroy(waiter->dmt);
 350                 return 1;
 351         }
 352
 353         dm_task_no_open_count(waiter->dmt);
 354
 355         /* accept wait interruption */
 356         set = unblock_signals();
 357
 358         /* interruption spits messages */
 359         dm_shut_log();
 360
 361         /* wait */
 362         r = dm_task_run(waiter->dmt);
 363
 364         /* wait is over : event or interrupt */
 365         pthread_sigmask(SIG_SETMASK, &set, NULL);
 366         //dm_restore_log();
 367
 368         if (!r) /* wait interrupted by signal */
 369                 return -1;
 370
 371         dm_task_destroy(waiter->dmt);
 372         waiter->dmt = NULL;
 373         waiter->event_nr++;
 374
 375         /*
 376          * upon event ...
 377          */
 378         while (1) {
 379                 condlog(3, "%s: devmap event #%i",
 380                                 waiter->mapname, waiter->event_nr);
 381
 382                 /*
 383                  * event might be :
 384                  *
 385                  * 1) a table reload, which means our mpp structure is
 386                  *    obsolete : refresh it through update_multipath()
 387                  * 2) a path failed by DM : mark as such through
 388                  *    update_multipath()
 389                  * 3) map has gone away : stop the thread.
 390                  * 4) a path reinstate : nothing to do
 391                  * 5) a switch group : nothing to do
 392                  */
 393                 pthread_cleanup_push(cleanup_lock, waiter->vecs->lock);
 394                 lock(waiter->vecs->lock);
 395                 r = update_multipath(waiter->vecs, waiter->mapname);
 396                 lock_cleanup_pop(waiter->vecs->lock);
 397
 398                 if (r)
 399                         return -1; /* stop the thread */
 400
 401                 event_nr = dm_geteventnr(waiter->mapname);
 402
 403                 if (waiter->event_nr == event_nr)
 404                         return 1; /* upon problem reschedule 1s later */
 405
 406                 waiter->event_nr = event_nr;
 407         }
 408         return -1; /* never reach there */
 409 }
 410
 411 static void *
 412 waitevent (void * et)
 413 {
 414         int r;
 415         struct event_thread *waiter;
 416
 417         mlockall(MCL_CURRENT | MCL_FUTURE);
 418
 419         waiter = (struct event_thread *)et;
 420         pthread_cleanup_push(free_waiter, et);
 421
 422         while (1) {
 423                 r = waiteventloop(waiter);
 424
 425                 if (r < 0)
 426                         break;
 427
 428                 sleep(r);
 429         }
 430
 431         pthread_cleanup_pop(1);
 432         return NULL;
 433 }
 434
 435 static int
 436 start_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 437 {
 438         pthread_attr_t attr;
 439         struct event_thread * wp;
 440
 441         if (!mpp)
 442                 return 0;
 443
 444         if (pthread_attr_init(&attr))
 445                 goto out;
 446
 447         pthread_attr_setstacksize(&attr, 32 * 1024);
 448         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 449
 450         wp = alloc_waiter();
 451
 452         if (!wp)
 453                 goto out;
 454
 455         mpp->waiter = (void *)wp;
 456         strncpy(wp->mapname, mpp->alias, WWID_SIZE);
 457         wp->vecs = vecs;
 458         wp->mpp = mpp;
 459
 460         if (pthread_create(&wp->thread, &attr, waitevent, wp)) {
 461                 condlog(0, "%s: cannot create event checker", wp->mapname);
 462                 goto out1;
 463         }
 464         condlog(2, "%s: event checker started", wp->mapname);
 465
 466         return 0;
 467 out1:
 468         free_waiter(wp);
 469         mpp->waiter = NULL;
 470 out:
 471         condlog(0, "failed to start waiter thread");
 472         return 1;
 473 }
 474
 475 static int
 476 flush_map(struct multipath * mpp, struct vectors * vecs)
 477 {
 478         /*
 479          * clear references to this map before flushing so we can ignore
 480          * the spurious uevent we may generate with the dm_flush_map call below
 481          */
 482         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 483                 /*
 484                  * May not really be an error -- if the map was already flushed
 485                  * from the device mapper by dmsetup(8) for instance.
 486                  */
 487                 condlog(0, "%s: can't flush", mpp->alias);
 488                 return 1;
 489         }
 490         else {
 491                 dm_lib_release();
 492                 condlog(3, "%s: devmap removed", mpp->alias);
 493         }
 494
 495         orphan_paths(vecs->pathvec, mpp);
 496         remove_map(mpp, vecs, stop_waiter_thread, 1);
 497
 498         return 0;
 499 }
 500
 501 static int
 502 uev_add_map (char * devname, struct vectors * vecs)
 503 {
 504         condlog(2, "%s: add map (uevent)", devname);
 505         return ev_add_map(devname, vecs);
 506 }
 507
 508 int
 509 ev_add_map (char * devname, struct vectors * vecs)
 510 {
 511         int major, minor;
 512         char dev_t[BLK_DEV_SIZE];
 513         char * alias;
 514         char * refwwid;
 515         struct multipath * mpp;
 516         int map_present;
 517         int r = 1;
 518
 519         if (sscanf(devname, "dm-%d", &minor) == 1 &&
 520             !sysfs_get_dev(sysfs_path, devname, dev_t, BLK_DEV_SIZE) &&
 521             sscanf(dev_t, "%d:%d", &major, &minor) == 2)
 522                 alias = dm_mapname(major, minor);
 523         else
 524                 alias = STRDUP(devname);
 525
 526         if (!alias)
 527                 return 1;
 528
 529         map_present = dm_map_present(alias);
 530
 531         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 532                 condlog(4, "%s: not a multipath map", alias);
 533                 FREE(alias);
 534                 return 0;
 535         }
 536
 537         mpp = find_mp_by_alias(vecs->mpvec, alias);
 538
 539         if (mpp) {
 540                 /*
 541                  * Not really an error -- we generate our own uevent
 542                  * if we create a multipath mapped device as a result
 543                  * of uev_add_path
 544                  */
 545                 condlog(0, "%s: devmap already registered",
 546                         devname);
 547                 FREE(alias);
 548                 return 0;
 549         }
 550
 551         /*
 552          * now we can register the map
 553          */
 554         if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
 555                                         start_waiter_thread))) {
 556                 condlog(3, "%s: devmap %s added", alias, devname);
 557                 return 0;
 558         }
 559         refwwid = get_refwwid(devname, DEV_DEVMAP, vecs->pathvec);
 560
 561         if (refwwid) {
 562                 r = coalesce_paths(vecs, NULL, refwwid);
 563                 dm_lib_release();
 564         }
 565
 566         if (!r)
 567                 condlog(3, "%s: devmap %s added", alias, devname);
 568         else
 569                 condlog(0, "%s: uev_add_map %s failed", alias, devname);
 570
 571         FREE(refwwid);
 572         FREE(alias);
 573         return r;
 574 }
 575
 576 static int
 577 uev_remove_map (char * devname, struct vectors * vecs)
 578 {
 579         condlog(2, "%s: remove map (uevent)", devname);
 580         return ev_remove_map(devname, vecs);
 581 }
 582
 583 int
 584 ev_remove_map (char * devname, struct vectors * vecs)
 585 {
 586         struct multipath * mpp;
 587
 588         mpp = find_mp_by_str(vecs->mpvec, devname);
 589
 590         if (!mpp) {
 591                 condlog(3, "%s: devmap not registered, can't remove",
 592                         devname);
 593                 return 0;
 594         }
 595         flush_map(mpp, vecs);
 596
 597         return 0;
 598 }
 599
 600 static int
 601 uev_umount_map (char * devname, struct vectors * vecs)
 602 {
 603         struct multipath * mpp;
 604
 605         condlog(2, "%s: umount map (uevent)", devname);
 606
 607         mpp = find_mp_by_str(vecs->mpvec, devname);
 608
 609         if (!mpp)
 610                 return 0;
 611
 612         update_mpp_paths(mpp, vecs->pathvec);
 613         verify_paths(mpp, vecs, NULL);
 614
 615         if (!VECTOR_SIZE(mpp->paths))
 616                 flush_map(mpp, vecs);
 617
 618         return 0;
 619 }
 620
 621 static int
 622 uev_add_path (char * devname, struct vectors * vecs)
 623 {
 624         condlog(2, "%s: add path (uevent)", devname);
 625         return (ev_add_path(devname, vecs) != 1)? 0 : 1;
 626 }
 627
 628
 629 /*
 630  * returns:
 631  * 0: added
 632  * 1: error
 633  * 2: blacklisted
 634  */
 635 int
 636 ev_add_path (char * devname, struct vectors * vecs)
 637 {
 638         struct multipath * mpp;
 639         struct path * pp;
 640         char empty_buff[WWID_SIZE] = {0};
 641
 642         pp = find_path_by_dev(vecs->pathvec, devname);
 643
 644         if (pp) {
 645                 condlog(0, "%s: spurious uevent, path already in pathvec, %p",
 646                         devname, pp->mpp);
 647                 /*
 648                  * allow reconfig of orphaned path here
 649                  */
 650                 if (pp->mpp) return 1;
 651         }
 652         else {
 653                 /*
 654                  * get path vital state
 655                  */
 656                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 657                       devname, DI_ALL))) {
 658                         condlog(0, "%s: failed to store path info", devname);
 659                         return 1;
 660                 }
 661                 pp->checkint = conf->checkint;
 662         }
 663
 664         /*
 665          * need path UID to go any further
 666          */
 667         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 668                 condlog(0, "%s: failed to get path uid", devname);
 669                 return 1; /* leave path added to pathvec */
 670         }
 671         if (blacklist_path(conf, pp)){
 672                 int i = find_slot(vecs->pathvec, (void *)pp);
 673                 if (i != -1)
 674                         vector_del_slot(vecs->pathvec, i);
 675                 free_path(pp);
 676                 return 2;
 677         }
 678         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 679 rescan:
 680         if (mpp) {
 681                 if (adopt_paths(vecs->pathvec, mpp))
 682                         return 1; /* leave path added to pathvec */
 683
 684                 verify_paths(mpp, vecs, NULL);
 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_devnode, 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+STATUS, cli_list_maps_status);
 997         add_handler(LIST+MAPS+STATS, cli_list_maps_stats);
 998         add_handler(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 999         add_handler(LIST+TOPOLOGY, cli_list_maps_topology);
1000         add_handler(LIST+MAP+TOPOLOGY, cli_list_map_topology);
1001         add_handler(LIST+CONFIG, cli_list_config);
1002         add_handler(ADD+PATH, cli_add_path);
1003         add_handler(DEL+PATH, cli_del_path);
1004         add_handler(ADD+MAP, cli_add_map);
1005         add_handler(DEL+MAP, cli_del_map);
1006         add_handler(SWITCH+MAP+GROUP, cli_switch_group);
1007         add_handler(RECONFIGURE, cli_reconfigure);
1008         add_handler(SUSPEND+MAP, cli_suspend);
1009         add_handler(RESUME+MAP, cli_resume);
1010         add_handler(REINSTATE+PATH, cli_reinstate);
1011         add_handler(FAIL+PATH, cli_fail);
1012
1013         uxsock_listen(&uxsock_trigger, ap);
1014
1015         return NULL;
1016 }
1017
1018 static int
1019 exit_daemon (int status)
1020 {
1021         if (status != 0)
1022                 fprintf(stderr, "bad exit status. see daemon.log\n");
1023
1024         condlog(3, "unlink pidfile");
1025         unlink(DEFAULT_PIDFILE);
1026
1027         lock(&exit_mutex);
1028         pthread_cond_signal(&exit_cond);
1029         unlock(&exit_mutex);
1030
1031         return status;
1032 }
1033
1034 static void
1035 fail_path (struct path * pp)
1036 {
1037         if (!pp->mpp)
1038                 return;
1039
1040         condlog(2, "checker failed path %s in map %s",
1041                  pp->dev_t, pp->mpp->alias);
1042
1043         dm_fail_path(pp->mpp->alias, pp->dev_t);
1044         update_queue_mode_del_path(pp->mpp);
1045 }
1046
1047 /*
1048  * caller must have locked the path list before calling that function
1049  */
1050 static void
1051 reinstate_path (struct path * pp)
1052 {
1053         if (!pp->mpp)
1054                 return;
1055
1056         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
1057                 condlog(0, "%s: reinstate failed", pp->dev_t);
1058         else {
1059                 condlog(2, "%s: reinstated", pp->dev_t);
1060                 update_queue_mode_add_path(pp->mpp);
1061         }
1062 }
1063
1064 static void
1065 enable_group(struct path * pp)
1066 {
1067         struct pathgroup * pgp;
1068
1069         /*
1070          * if path is added through uev_add_path, pgindex can be unset.
1071          * next update_strings() will set it, upon map reload event.
1072          *
1073          * we can safely return here, because upon map reload, all
1074          * PG will be enabled.
1075          */
1076         if (!pp->mpp->pg || !pp->pgindex)
1077                 return;
1078
1079         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1080
1081         if (pgp->status == PGSTATE_DISABLED) {
1082                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1083                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1084         }
1085 }
1086
1087 static void
1088 mpvec_garbage_collector (struct vectors * vecs)
1089 {
1090         struct multipath * mpp;
1091         int i;
1092
1093         vector_foreach_slot (vecs->mpvec, mpp, i) {
1094                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1095                         condlog(2, "%s: remove dead map", mpp->alias);
1096                         remove_map(mpp, vecs, stop_waiter_thread, 1);
1097                         i--;
1098                 }
1099         }
1100 }
1101
1102 static void
1103 defered_failback_tick (vector mpvec)
1104 {
1105         struct multipath * mpp;
1106         int i;
1107
1108         vector_foreach_slot (mpvec, mpp, i) {
1109                 /*
1110                  * defered failback getting sooner
1111                  */
1112                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1113                         mpp->failback_tick--;
1114
1115                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1116                                 switch_pathgroup(mpp);
1117                 }
1118         }
1119 }
1120
1121 static void
1122 retry_count_tick(vector mpvec)
1123 {
1124         struct multipath *mpp;
1125         int i;
1126
1127         vector_foreach_slot (mpvec, mpp, i) {
1128                 if (mpp->retry_tick) {
1129                         mpp->stat_total_queueing_time++;
1130                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1131                         if(--mpp->retry_tick == 0) {
1132                                 dm_queue_if_no_path(mpp->alias, 0);
1133                                 condlog(2, "%s: Disable queueing", mpp->alias);
1134                         }
1135                 }
1136         }
1137 }
1138
1139 static void *
1140 checkerloop (void *ap)
1141 {
1142         struct vectors *vecs;
1143         struct path *pp;
1144         int i, count = 0;
1145         int newstate;
1146
1147         mlockall(MCL_CURRENT | MCL_FUTURE);
1148         vecs = (struct vectors *)ap;
1149         condlog(2, "path checkers start up");
1150
1151         /*
1152          * init the path check interval
1153          */
1154         vector_foreach_slot (vecs->pathvec, pp, i) {
1155                 pp->checkint = conf->checkint;
1156         }
1157
1158         while (1) {
1159                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1160                 lock(vecs->lock);
1161                 condlog(4, "tick");
1162
1163                 vector_foreach_slot (vecs->pathvec, pp, i) {
1164                         if (!pp->mpp)
1165                                 continue;
1166
1167                         if (pp->tick && --pp->tick)
1168                                 continue; /* don't check this path yet */
1169
1170                         /*
1171                          * provision a next check soonest,
1172                          * in case we exit abnormaly from here
1173                          */
1174                         pp->tick = conf->checkint;
1175
1176                         if (!checker_selected(&pp->checker)) {
1177                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
1178                                 select_checker(pp);
1179                         }
1180
1181                         if (!checker_selected(&pp->checker)) {
1182                                 condlog(0, "%s: checker is not set", pp->dev);
1183                                 continue;
1184                         }
1185                         newstate = checker_check(&pp->checker);
1186
1187                         if (newstate < 0) {
1188                                 condlog(2, "%s: unusable path", pp->dev);
1189                                 pathinfo(pp, conf->hwtable, 0);
1190                                 continue;
1191                         }
1192
1193                         if (newstate != pp->state) {
1194                                 pp->state = newstate;
1195                                 LOG_MSG(1, checker_message(&pp->checker));
1196
1197                                 /*
1198                                  * upon state change, reset the checkint
1199                                  * to the shortest delay
1200                                  */
1201                                 pp->checkint = conf->checkint;
1202
1203                                 if (newstate == PATH_DOWN ||
1204                                     newstate == PATH_SHAKY ||
1205                                     update_multipath_strings(pp->mpp,
1206                                                              vecs->pathvec)) {
1207                                         /*
1208                                          * proactively fail path in the DM
1209                                          */
1210                                         fail_path(pp);
1211
1212                                         /*
1213                                          * cancel scheduled failback
1214                                          */
1215                                         pp->mpp->failback_tick = 0;
1216
1217                                         pp->mpp->stat_path_failures++;
1218                                         continue;
1219                                 }
1220
1221                                 /*
1222                                  * reinstate this path
1223                                  */
1224                                 reinstate_path(pp);
1225
1226                                 /*
1227                                  * schedule [defered] failback
1228                                  */
1229                                 if (pp->mpp->pgfailback > 0)
1230                                         pp->mpp->failback_tick =
1231                                                 pp->mpp->pgfailback + 1;
1232                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
1233                                     need_switch_pathgroup(pp->mpp, 1))
1234                                         switch_pathgroup(pp->mpp);
1235
1236                                 /*
1237                                  * if at least one path is up in a group, and
1238                                  * the group is disabled, re-enable it
1239                                  */
1240                                 if (newstate == PATH_UP)
1241                                         enable_group(pp);
1242                         }
1243                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1244                                 LOG_MSG(4, checker_message(&pp->checker));
1245                                 /*
1246                                  * double the next check delay.
1247                                  * max at conf->max_checkint
1248                                  */
1249                                 if (pp->checkint < (conf->max_checkint / 2))
1250                                         pp->checkint = 2 * pp->checkint;
1251                                 else
1252                                         pp->checkint = conf->max_checkint;
1253
1254                                 pp->tick = pp->checkint;
1255                                 condlog(4, "%s: delay next check %is",
1256                                                 pp->dev_t, pp->tick);
1257
1258                         }
1259                         pp->state = newstate;
1260
1261                         /*
1262                          * path prio refreshing
1263                          */
1264                         condlog(4, "path prio refresh");
1265                         pathinfo(pp, conf->hwtable, DI_PRIO);
1266
1267                         if (need_switch_pathgroup(pp->mpp, 0)) {
1268                                 if (pp->mpp->pgfailback > 0)
1269                                         pp->mpp->failback_tick =
1270                                                 pp->mpp->pgfailback + 1;
1271                                 else if (pp->mpp->pgfailback ==
1272                                                 -FAILBACK_IMMEDIATE)
1273                                         switch_pathgroup(pp->mpp);
1274                         }
1275                 }
1276                 defered_failback_tick(vecs->mpvec);
1277                 retry_count_tick(vecs->mpvec);
1278
1279                 if (count)
1280                         count--;
1281                 else {
1282                         condlog(4, "map garbage collection");
1283                         mpvec_garbage_collector(vecs);
1284                         count = MAPGCINT;
1285                 }
1286
1287                 lock_cleanup_pop(vecs->lock);
1288                 sleep(1);
1289         }
1290         return NULL;
1291 }
1292
1293 int
1294 configure (struct vectors * vecs, int start_waiters)
1295 {
1296         struct multipath * mpp;
1297         struct path * pp;
1298         vector mpvec;
1299         int i;
1300
1301         if (!(vecs->pathvec = vector_alloc()))
1302                 return 1;
1303
1304         if (!(vecs->mpvec = vector_alloc()))
1305                 return 1;
1306
1307         if (!(mpvec = vector_alloc()))
1308                 return 1;
1309
1310         /*
1311          * probe for current path (from sysfs) and map (from dm) sets
1312          */
1313         path_discovery(vecs->pathvec, conf, DI_ALL);
1314
1315         vector_foreach_slot (vecs->pathvec, pp, i){
1316                 if (blacklist_path(conf, pp)){
1317                         vector_del_slot(vecs->pathvec, i);
1318                         free_path(pp);
1319                         i--;
1320                 }
1321                 else
1322                         pp->checkint = conf->checkint;
1323         }
1324         if (map_discovery(vecs))
1325                 return 1;
1326
1327         /*
1328          * create new set of maps & push changed ones into dm
1329          */
1330         if (coalesce_paths(vecs, mpvec, NULL))
1331                 return 1;
1332
1333         /*
1334          * may need to remove some maps which are no longer relevant
1335          * e.g., due to blacklist changes in conf file
1336          */
1337         if (coalesce_maps(vecs, mpvec))
1338                 return 1;
1339
1340         dm_lib_release();
1341
1342         if (conf->verbosity > 2)
1343                 vector_foreach_slot(mpvec, mpp, i)
1344                         print_map(mpp);
1345
1346         /*
1347          * purge dm of old maps
1348          */
1349         remove_maps(vecs, NULL);
1350
1351         /*
1352          * save new set of maps formed by considering current path state
1353          */
1354         vecs->mpvec = mpvec;
1355
1356         /*
1357          * start dm event waiter threads for these new maps
1358          */
1359         vector_foreach_slot(vecs->mpvec, mpp, i) {
1360                 if (setup_multipath(vecs, mpp))
1361                         return 1;
1362                 if (start_waiters)
1363                         if (start_waiter_thread(mpp, vecs))
1364                                 return 1;
1365         }
1366         return 0;
1367 }
1368
1369 int
1370 reconfigure (struct vectors * vecs)
1371 {
1372         struct config * old = conf;
1373
1374         /*
1375          * free old map and path vectors ... they use old conf state
1376          */
1377         if (VECTOR_SIZE(vecs->mpvec))
1378                 remove_maps(vecs, stop_waiter_thread);
1379
1380         if (VECTOR_SIZE(vecs->pathvec))
1381                 free_pathvec(vecs->pathvec, FREE_PATHS);
1382
1383         conf = NULL;
1384
1385         if (load_config(DEFAULT_CONFIGFILE))
1386                 return 1;
1387
1388         conf->verbosity = old->verbosity;
1389
1390         if (!conf->checkint) {
1391                 conf->checkint = DEFAULT_CHECKINT;
1392                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1393         }
1394         configure(vecs, 1);
1395         free_config(old);
1396         return 0;
1397 }
1398
1399 static struct vectors *
1400 init_vecs (void)
1401 {
1402         struct vectors * vecs;
1403
1404         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1405
1406         if (!vecs)
1407                 return NULL;
1408
1409         vecs->lock =
1410                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1411
1412         if (!vecs->lock)
1413                 goto out;
1414
1415         vecs->pathvec = vector_alloc();
1416
1417         if (!vecs->pathvec)
1418                 goto out1;
1419
1420         vecs->mpvec = vector_alloc();
1421
1422         if (!vecs->mpvec)
1423                 goto out2;
1424
1425         pthread_mutex_init(vecs->lock, NULL);
1426
1427         return vecs;
1428
1429 out2:
1430         vector_free(vecs->pathvec);
1431 out1:
1432         FREE(vecs->lock);
1433 out:
1434         FREE(vecs);
1435         condlog(0, "failed to init paths");
1436         return NULL;
1437 }
1438
1439 static void *
1440 signal_set(int signo, void (*func) (int))
1441 {
1442         int r;
1443         struct sigaction sig;
1444         struct sigaction osig;
1445
1446         sig.sa_handler = func;
1447         sigemptyset(&sig.sa_mask);
1448         sig.sa_flags = 0;
1449
1450         r = sigaction(signo, &sig, &osig);
1451
1452         if (r < 0)
1453                 return (SIG_ERR);
1454         else
1455                 return (osig.sa_handler);
1456 }
1457
1458 static void
1459 sighup (int sig)
1460 {
1461         condlog(2, "reconfigure (SIGHUP)");
1462
1463         lock(gvecs->lock);
1464         reconfigure(gvecs);
1465         unlock(gvecs->lock);
1466
1467 #ifdef _DEBUG_
1468         dbg_free_final(NULL);
1469 #endif
1470 }
1471
1472 static void
1473 sigend (int sig)
1474 {
1475         exit_daemon(0);
1476 }
1477
1478 static void
1479 sigusr1 (int sig)
1480 {
1481         condlog(3, "SIGUSR1 received");
1482 }
1483
1484 static void
1485 signal_init(void)
1486 {
1487         signal_set(SIGHUP, sighup);
1488         signal_set(SIGUSR1, sigusr1);
1489         signal_set(SIGINT, sigend);
1490         signal_set(SIGTERM, sigend);
1491         signal_set(SIGKILL, sigend);
1492 }
1493
1494 static void
1495 setscheduler (void)
1496 {
1497         int res;
1498         static struct sched_param sched_param = {
1499                 sched_priority: 99
1500         };
1501
1502         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1503
1504         if (res == -1)
1505                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1506         return;
1507 }
1508
1509 static void
1510 set_oom_adj (int val)
1511 {
1512         FILE *fp;
1513
1514         fp = fopen("/proc/self/oom_adj", "w");
1515
1516         if (!fp)
1517                 return;
1518
1519         fprintf(fp, "%i", val);
1520         fclose(fp);
1521 }
1522
1523 static int
1524 child (void * param)
1525 {
1526         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1527         pthread_attr_t attr;
1528         struct vectors * vecs;
1529
1530         mlockall(MCL_CURRENT | MCL_FUTURE);
1531
1532         if (logsink)
1533                 log_thread_start();
1534
1535         condlog(2, "--------start up--------");
1536         condlog(2, "read " DEFAULT_CONFIGFILE);
1537
1538         if (load_config(DEFAULT_CONFIGFILE))
1539                 exit(1);
1540
1541         setlogmask(LOG_UPTO(conf->verbosity + 3));
1542
1543         /*
1544          * fill the voids left in the config file
1545          */
1546         if (!conf->checkint) {
1547                 conf->checkint = DEFAULT_CHECKINT;
1548                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1549         }
1550
1551         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1552                 if (logsink)
1553                         log_thread_stop();
1554
1555                 exit(1);
1556         }
1557         signal_init();
1558         setscheduler();
1559         set_oom_adj(-16);
1560         vecs = gvecs = init_vecs();
1561
1562         if (!vecs)
1563                 exit(1);
1564
1565         if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1566                 condlog(0, "can not find sysfs mount point");
1567                 exit(1);
1568         }
1569
1570         /*
1571          * fetch and configure both paths and multipaths
1572          */
1573         if (configure(vecs, 1)) {
1574                 condlog(0, "failure during configuration");
1575                 exit(1);
1576         }
1577
1578         /*
1579          * start threads
1580          */
1581         pthread_attr_init(&attr);
1582         pthread_attr_setstacksize(&attr, 64 * 1024);
1583         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1584
1585         pthread_create(&check_thr, &attr, checkerloop, vecs);
1586         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1587         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1588
1589         pthread_cond_wait(&exit_cond, &exit_mutex);
1590
1591         /*
1592          * exit path
1593          */
1594         lock(vecs->lock);
1595         remove_maps(vecs, stop_waiter_thread);
1596         free_pathvec(vecs->pathvec, FREE_PATHS);
1597
1598         pthread_cancel(check_thr);
1599         pthread_cancel(uevent_thr);
1600         pthread_cancel(uxlsnr_thr);
1601
1602         free_keys(keys);
1603         keys = NULL;
1604         free_handlers(handlers);
1605         handlers = NULL;
1606         free_polls();
1607
1608         unlock(vecs->lock);
1609         pthread_mutex_destroy(vecs->lock);
1610         FREE(vecs->lock);
1611         vecs->lock = NULL;
1612         FREE(vecs);
1613         vecs = NULL;
1614         free_config(conf);
1615         conf = NULL;
1616
1617         condlog(2, "--------shut down-------");
1618
1619         if (logsink)
1620                 log_thread_stop();
1621
1622         dm_lib_release();
1623         dm_lib_exit();
1624
1625 #ifdef _DEBUG_
1626         dbg_free_final(NULL);
1627 #endif
1628
1629         exit(0);
1630 }
1631
1632 static int
1633 daemonize(void)
1634 {
1635         int pid;
1636         int in_fd, out_fd;
1637
1638         if( (pid = fork()) < 0){
1639                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1640                 return -1;
1641         }
1642         else if (pid != 0)
1643                 return pid;
1644
1645         setsid();
1646
1647         if ( (pid = fork()) < 0)
1648                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1649         else if (pid != 0)
1650                 _exit(0);
1651
1652         in_fd = open("/dev/null", O_RDONLY);
1653         if (in_fd < 0){
1654                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1655                         strerror(errno));
1656                 _exit(0);
1657         }
1658         out_fd = open("/dev/console", O_WRONLY);
1659         if (out_fd < 0){
1660                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1661                         strerror(errno));
1662                 _exit(0);
1663         }
1664
1665         close(STDIN_FILENO);
1666         dup(in_fd);
1667         close(STDOUT_FILENO);
1668         dup(out_fd);
1669         close(STDERR_FILENO);
1670         dup(out_fd);
1671
1672         close(in_fd);
1673         close(out_fd);
1674         chdir("/");
1675         umask(0);
1676         return 0;
1677 }
1678
1679 int
1680 main (int argc, char *argv[])
1681 {
1682         extern char *optarg;
1683         extern int optind;
1684         int arg;
1685         int err;
1686
1687         logsink = 1;
1688
1689         if (getuid() != 0) {
1690                 fprintf(stderr, "need to be root\n");
1691                 exit(1);
1692         }
1693
1694         /* make sure we don't lock any path */
1695         chdir("/");
1696         umask(umask(077) | 022);
1697
1698         conf = alloc_config();
1699
1700         if (!conf)
1701                 exit(1);
1702
1703         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1704         switch(arg) {
1705                 case 'd':
1706                         logsink = 0;
1707                         //debug=1; /* ### comment me out ### */
1708                         break;
1709                 case 'v':
1710                         if (sizeof(optarg) > sizeof(char *) ||
1711                             !isdigit(optarg[0]))
1712                                 exit(1);
1713
1714                         conf->verbosity = atoi(optarg);
1715                         break;
1716                 case 'k':
1717                         uxclnt(optarg);
1718                         exit(0);
1719                 default:
1720                         ;
1721                 }
1722         }
1723
1724         if (!logsink)
1725                 err = 0;
1726         else
1727                 err = daemonize();
1728
1729         if (err < 0)
1730                 /* error */
1731                 exit(1);
1732         else if (err > 0)
1733                 /* parent dies */
1734                 exit(0);
1735         else
1736                 /* child lives */
1737                 return (child(NULL));
1738 }
1739