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                                 int oldstate = pp->state;
 286                                 condlog(2, "%s: mark as failed", pp->dev_t);
 287                                 mpp->stat_path_failures++;
 288                                 pp->state = PATH_DOWN;
 289                                 if (oldstate == PATH_UP ||
 290                                     oldstate == PATH_GHOST)
 291                                         update_queue_mode_del_path(mpp);
 292
 293                                 /*
 294                                  * if opportune,
 295                                  * schedule the next check earlier
 296                                  */
 297                                 if (pp->tick > conf->checkint)
 298                                         pp->tick = conf->checkint;
 299                         }
 300                 }
 301         }
 302         r = 0;
 303 out:
 304         if (r)
 305                 condlog(0, "failed to update multipath");
 306
 307         return r;
 308 }
 309
 310 static sigset_t unblock_signals(void)
 311 {
 312         sigset_t set, old;
 313
 314         sigemptyset(&set);
 315         sigaddset(&set, SIGHUP);
 316         sigaddset(&set, SIGUSR1);
 317         pthread_sigmask(SIG_UNBLOCK, &set, &old);
 318         return old;
 319 }
 320
 321 /*
 322  * returns the reschedule delay
 323  * negative means *stop*
 324  */
 325 static int
 326 waiteventloop (struct event_thread * waiter)
 327 {
 328         sigset_t set;
 329         int event_nr;
 330         int r;
 331
 332         if (!waiter->event_nr)
 333                 waiter->event_nr = dm_geteventnr(waiter->mapname);
 334
 335         if (!(waiter->dmt = dm_task_create(DM_DEVICE_WAITEVENT))) {
 336                 condlog(0, "%s: devmap event #%i dm_task_create error",
 337                                 waiter->mapname, waiter->event_nr);
 338                 return 1;
 339         }
 340
 341         if (!dm_task_set_name(waiter->dmt, waiter->mapname)) {
 342                 condlog(0, "%s: devmap event #%i dm_task_set_name error",
 343                                 waiter->mapname, waiter->event_nr);
 344                 dm_task_destroy(waiter->dmt);
 345                 return 1;
 346         }
 347
 348         if (waiter->event_nr && !dm_task_set_event_nr(waiter->dmt,
 349                                                       waiter->event_nr)) {
 350                 condlog(0, "%s: devmap event #%i dm_task_set_event_nr error",
 351                                 waiter->mapname, waiter->event_nr);
 352                 dm_task_destroy(waiter->dmt);
 353                 return 1;
 354         }
 355
 356         dm_task_no_open_count(waiter->dmt);
 357
 358         /* accept wait interruption */
 359         set = unblock_signals();
 360
 361         /* interruption spits messages */
 362         dm_shut_log();
 363
 364         /* wait */
 365         r = dm_task_run(waiter->dmt);
 366
 367         /* wait is over : event or interrupt */
 368         pthread_sigmask(SIG_SETMASK, &set, NULL);
 369         //dm_restore_log();
 370
 371         if (!r) /* wait interrupted by signal */
 372                 return -1;
 373
 374         dm_task_destroy(waiter->dmt);
 375         waiter->dmt = NULL;
 376         waiter->event_nr++;
 377
 378         /*
 379          * upon event ...
 380          */
 381         while (1) {
 382                 condlog(3, "%s: devmap event #%i",
 383                                 waiter->mapname, waiter->event_nr);
 384
 385                 /*
 386                  * event might be :
 387                  *
 388                  * 1) a table reload, which means our mpp structure is
 389                  *    obsolete : refresh it through update_multipath()
 390                  * 2) a path failed by DM : mark as such through
 391                  *    update_multipath()
 392                  * 3) map has gone away : stop the thread.
 393                  * 4) a path reinstate : nothing to do
 394                  * 5) a switch group : nothing to do
 395                  */
 396                 pthread_cleanup_push(cleanup_lock, waiter->vecs->lock);
 397                 lock(waiter->vecs->lock);
 398                 r = update_multipath(waiter->vecs, waiter->mapname);
 399                 lock_cleanup_pop(waiter->vecs->lock);
 400
 401                 if (r)
 402                         return -1; /* stop the thread */
 403
 404                 event_nr = dm_geteventnr(waiter->mapname);
 405
 406                 if (waiter->event_nr == event_nr)
 407                         return 1; /* upon problem reschedule 1s later */
 408
 409                 waiter->event_nr = event_nr;
 410         }
 411         return -1; /* never reach there */
 412 }
 413
 414 static void *
 415 waitevent (void * et)
 416 {
 417         int r;
 418         struct event_thread *waiter;
 419
 420         mlockall(MCL_CURRENT | MCL_FUTURE);
 421
 422         waiter = (struct event_thread *)et;
 423         pthread_cleanup_push(free_waiter, et);
 424
 425         while (1) {
 426                 r = waiteventloop(waiter);
 427
 428                 if (r < 0)
 429                         break;
 430
 431                 sleep(r);
 432         }
 433
 434         pthread_cleanup_pop(1);
 435         return NULL;
 436 }
 437
 438 static int
 439 start_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 440 {
 441         pthread_attr_t attr;
 442         struct event_thread * wp;
 443
 444         if (!mpp)
 445                 return 0;
 446
 447         if (pthread_attr_init(&attr))
 448                 goto out;
 449
 450         pthread_attr_setstacksize(&attr, 32 * 1024);
 451         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 452
 453         wp = alloc_waiter();
 454
 455         if (!wp)
 456                 goto out;
 457
 458         mpp->waiter = (void *)wp;
 459         strncpy(wp->mapname, mpp->alias, WWID_SIZE);
 460         wp->vecs = vecs;
 461         wp->mpp = mpp;
 462
 463         if (pthread_create(&wp->thread, &attr, waitevent, wp)) {
 464                 condlog(0, "%s: cannot create event checker", wp->mapname);
 465                 goto out1;
 466         }
 467         condlog(2, "%s: event checker started", wp->mapname);
 468
 469         return 0;
 470 out1:
 471         free_waiter(wp);
 472         mpp->waiter = NULL;
 473 out:
 474         condlog(0, "failed to start waiter thread");
 475         return 1;
 476 }
 477
 478 static int
 479 flush_map(struct multipath * mpp, struct vectors * vecs)
 480 {
 481         /*
 482          * clear references to this map before flushing so we can ignore
 483          * the spurious uevent we may generate with the dm_flush_map call below
 484          */
 485         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 486                 /*
 487                  * May not really be an error -- if the map was already flushed
 488                  * from the device mapper by dmsetup(8) for instance.
 489                  */
 490                 condlog(0, "%s: can't flush", mpp->alias);
 491                 return 1;
 492         }
 493         else {
 494                 dm_lib_release();
 495                 condlog(3, "%s: devmap removed", mpp->alias);
 496         }
 497
 498         orphan_paths(vecs->pathvec, mpp);
 499         remove_map(mpp, vecs, stop_waiter_thread, 1);
 500
 501         return 0;
 502 }
 503
 504 static int
 505 uev_add_map (char * devname, struct vectors * vecs)
 506 {
 507         condlog(2, "%s: add map (uevent)", devname);
 508         return ev_add_map(devname, vecs);
 509 }
 510
 511 int
 512 ev_add_map (char * devname, struct vectors * vecs)
 513 {
 514         int major, minor;
 515         char dev_t[BLK_DEV_SIZE];
 516         char * alias;
 517         char * refwwid;
 518         struct multipath * mpp;
 519         int map_present;
 520         int r = 1;
 521
 522         if (sscanf(devname, "dm-%d", &minor) == 1 &&
 523             !sysfs_get_dev(sysfs_path, devname, dev_t, BLK_DEV_SIZE) &&
 524             sscanf(dev_t, "%d:%d", &major, &minor) == 2)
 525                 alias = dm_mapname(major, minor);
 526         else
 527                 alias = STRDUP(devname);
 528
 529         if (!alias)
 530                 return 1;
 531
 532         map_present = dm_map_present(alias);
 533
 534         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 535                 condlog(4, "%s: not a multipath map", alias);
 536                 FREE(alias);
 537                 return 0;
 538         }
 539
 540         mpp = find_mp_by_alias(vecs->mpvec, alias);
 541
 542         if (mpp) {
 543                 /*
 544                  * Not really an error -- we generate our own uevent
 545                  * if we create a multipath mapped device as a result
 546                  * of uev_add_path
 547                  */
 548                 condlog(0, "%s: devmap already registered",
 549                         devname);
 550                 FREE(alias);
 551                 return 0;
 552         }
 553
 554         /*
 555          * now we can register the map
 556          */
 557         if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
 558                                         start_waiter_thread))) {
 559                 condlog(3, "%s: devmap %s added", alias, devname);
 560                 return 0;
 561         }
 562         refwwid = get_refwwid(devname, DEV_DEVMAP, vecs->pathvec);
 563
 564         if (refwwid) {
 565                 r = coalesce_paths(vecs, NULL, refwwid);
 566                 dm_lib_release();
 567         }
 568
 569         if (!r)
 570                 condlog(3, "%s: devmap %s added", alias, devname);
 571         else
 572                 condlog(0, "%s: uev_add_map %s failed", alias, devname);
 573
 574         FREE(refwwid);
 575         FREE(alias);
 576         return r;
 577 }
 578
 579 static int
 580 uev_remove_map (char * devname, struct vectors * vecs)
 581 {
 582         condlog(2, "%s: remove map (uevent)", devname);
 583         return ev_remove_map(devname, vecs);
 584 }
 585
 586 int
 587 ev_remove_map (char * devname, struct vectors * vecs)
 588 {
 589         struct multipath * mpp;
 590
 591         mpp = find_mp_by_str(vecs->mpvec, devname);
 592
 593         if (!mpp) {
 594                 condlog(3, "%s: devmap not registered, can't remove",
 595                         devname);
 596                 return 0;
 597         }
 598         flush_map(mpp, vecs);
 599
 600         return 0;
 601 }
 602
 603 static int
 604 uev_umount_map (char * devname, struct vectors * vecs)
 605 {
 606         struct multipath * mpp;
 607
 608         condlog(2, "%s: umount map (uevent)", devname);
 609
 610         mpp = find_mp_by_str(vecs->mpvec, devname);
 611
 612         if (!mpp)
 613                 return 0;
 614
 615         update_mpp_paths(mpp, vecs->pathvec);
 616         verify_paths(mpp, vecs, NULL);
 617
 618         if (!VECTOR_SIZE(mpp->paths))
 619                 flush_map(mpp, vecs);
 620
 621         return 0;
 622 }
 623
 624 static int
 625 uev_add_path (char * devname, struct vectors * vecs)
 626 {
 627         condlog(2, "%s: add path (uevent)", devname);
 628         return (ev_add_path(devname, vecs) != 1)? 0 : 1;
 629 }
 630
 631
 632 /*
 633  * returns:
 634  * 0: added
 635  * 1: error
 636  * 2: blacklisted
 637  */
 638 int
 639 ev_add_path (char * devname, struct vectors * vecs)
 640 {
 641         struct multipath * mpp;
 642         struct path * pp;
 643         char empty_buff[WWID_SIZE] = {0};
 644
 645         pp = find_path_by_dev(vecs->pathvec, devname);
 646
 647         if (pp) {
 648                 condlog(0, "%s: spurious uevent, path already in pathvec, %p",
 649                         devname, pp->mpp);
 650                 /*
 651                  * allow reconfig of orphaned path here
 652                  */
 653                 if (pp->mpp) return 1;
 654         }
 655         else {
 656                 /*
 657                  * get path vital state
 658                  */
 659                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 660                       devname, DI_ALL))) {
 661                         condlog(0, "%s: failed to store path info", devname);
 662                         return 1;
 663                 }
 664                 pp->checkint = conf->checkint;
 665         }
 666
 667         /*
 668          * need path UID to go any further
 669          */
 670         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 671                 condlog(0, "%s: failed to get path uid", devname);
 672                 return 1; /* leave path added to pathvec */
 673         }
 674         if (blacklist_path(conf, pp)){
 675                 int i = find_slot(vecs->pathvec, (void *)pp);
 676                 if (i != -1)
 677                         vector_del_slot(vecs->pathvec, i);
 678                 free_path(pp);
 679                 return 2;
 680         }
 681         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 682 rescan:
 683         if (mpp) {
 684                 if (adopt_paths(vecs->pathvec, mpp))
 685                         return 1; /* leave path added to pathvec */
 686
 687                 verify_paths(mpp, vecs, NULL);
 688                 mpp->action = ACT_RELOAD;
 689         }
 690         else {
 691                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 692                         mpp->action = ACT_CREATE;
 693                 else
 694                         return 1; /* leave path added to pathvec */
 695         }
 696
 697         /*
 698          * push the map to the device-mapper
 699          */
 700         if (setup_map(mpp)) {
 701                 condlog(0, "%s: failed to setup map for addition of new "
 702                         "path %s", mpp->alias, devname);
 703                 goto out;
 704         }
 705         /*
 706          * reload the map for the multipath mapped device
 707          */
 708         if (domap(mpp) <= 0) {
 709                 condlog(0, "%s: failed in domap for addition of new "
 710                         "path %s", mpp->alias, devname);
 711                 /*
 712                  * deal with asynchronous uevents :((
 713                  */
 714                 if (mpp->action == ACT_RELOAD) {
 715                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 716                         sleep(1);
 717                         update_mpp_paths(mpp, vecs->pathvec);
 718                         goto rescan;
 719                 }
 720                 else
 721                         goto out;
 722         }
 723         dm_lib_release();
 724
 725         /*
 726          * update our state from kernel regardless of create or reload
 727          */
 728         if (setup_multipath(vecs, mpp))
 729                 goto out;
 730
 731         if (mpp->action == ACT_CREATE &&
 732             start_waiter_thread(mpp, vecs))
 733                         goto out;
 734
 735         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 736         return 0;
 737
 738 out:
 739         remove_map(mpp, vecs, NULL, 1);
 740         return 1;
 741 }
 742
 743 static int
 744 uev_remove_path (char * devname, struct vectors * vecs)
 745 {
 746         condlog(2, "%s: remove path (uevent)", devname);
 747         return ev_remove_path(devname, vecs);
 748 }
 749
 750 int
 751 ev_remove_path (char * devname, struct vectors * vecs)
 752 {
 753         struct multipath * mpp;
 754         struct path * pp;
 755         int i;
 756         int rm_path = 1;
 757
 758         pp = find_path_by_dev(vecs->pathvec, devname);
 759
 760         if (!pp) {
 761                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 762                 return 1;
 763         }
 764
 765         /*
 766          * avoid referring to the map of an orphanned path
 767          */
 768         if ((mpp = pp->mpp)) {
 769
 770                 /*
 771                  * remove the map IFF removing the last path
 772                  */
 773                 if (pathcount(mpp, PATH_WILD) > 1) {
 774                         vector rpvec = vector_alloc();
 775
 776                         /*
 777                          * transform the mp->pg vector of vectors of paths
 778                          * into a mp->params string to feed the device-mapper
 779                          */
 780                         update_mpp_paths(mpp, vecs->pathvec);
 781                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 782                                 vector_del_slot(mpp->paths, i);
 783
 784                         if (VECTOR_SIZE(mpp->paths) == 0) {
 785                                 char alias[WWID_SIZE];
 786
 787                                 /*
 788                                  * flush_map will fail if the device is open
 789                                  */
 790                                 strncpy(alias, mpp->alias, WWID_SIZE);
 791                                 if (flush_map(mpp, vecs))
 792                                         rm_path = 0;
 793                                 else
 794                                         condlog(3, "%s: removed map after removing"
 795                                                 " multiple paths", alias);
 796                         }
 797                         else {
 798                                 if (setup_map(mpp)) {
 799                                         condlog(0, "%s: failed to setup map for"
 800                                                 " removal of path %s", mpp->alias, devname);
 801                                         free_pathvec(rpvec, KEEP_PATHS);
 802                                         goto out;
 803                                 }
 804                                 /*
 805                                  * reload the map
 806                                  */
 807                                 mpp->action = ACT_RELOAD;
 808                                 if (domap(mpp) <= 0) {
 809                                         condlog(0, "%s: failed in domap for "
 810                                                 "removal of path %s",
 811                                                 mpp->alias, devname);
 812                                         /*
 813                                          * Delete path from pathvec so that
 814                                          * update_mpp_paths wont find it later
 815                                          * when/if another path is removed.
 816                                          */
 817                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 818                                                 vector_del_slot(vecs->pathvec, i);
 819                                         free_path(pp);
 820                                         return 1;
 821                                 }
 822                                 /*
 823                                  * update our state from kernel
 824                                  */
 825                                 if (setup_multipath(vecs, mpp)) {
 826                                         free_pathvec(rpvec, KEEP_PATHS);
 827                                         goto out;
 828                                 }
 829
 830                                 condlog(3, "%s path removed from devmap %s",
 831                                         devname, mpp->alias);
 832                         }
 833                         free_pathvec(rpvec, KEEP_PATHS);
 834                 }
 835                 else {
 836                         char alias[WWID_SIZE];
 837
 838                         /*
 839                          * flush_map will fail if the device is open
 840                          */
 841                         strncpy(alias, mpp->alias, WWID_SIZE);
 842                         if (flush_map(mpp, vecs))
 843                                 rm_path = 0;
 844                         else
 845                                 condlog(3, "%s: removed map", alias);
 846                 }
 847         }
 848
 849         if (rm_path) {
 850                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 851                         vector_del_slot(vecs->pathvec, i);
 852                 free_path(pp);
 853         }
 854
 855         return 0;
 856
 857 out:
 858         remove_map(mpp, vecs, stop_waiter_thread, 1);
 859         return 1;
 860 }
 861
 862 static int
 863 map_discovery (struct vectors * vecs)
 864 {
 865         int i;
 866         struct multipath * mpp;
 867
 868         if (dm_get_maps(vecs->mpvec, "multipath"))
 869                 return 1;
 870
 871         vector_foreach_slot (vecs->mpvec, mpp, i)
 872                 if (setup_multipath(vecs, mpp))
 873                         return 1;
 874
 875         return 0;
 876 }
 877
 878 int
 879 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 880 {
 881         struct vectors * vecs;
 882         int r;
 883
 884         *reply = NULL;
 885         *len = 0;
 886         vecs = (struct vectors *)trigger_data;
 887
 888         pthread_cleanup_push(cleanup_lock, vecs->lock);
 889         lock(vecs->lock);
 890
 891         r = parse_cmd(str, reply, len, vecs);
 892
 893         if (r > 0) {
 894                 *reply = STRDUP("fail\n");
 895                 *len = strlen(*reply) + 1;
 896                 r = 1;
 897         }
 898         else if (!r && *len == 0) {
 899                 *reply = STRDUP("ok\n");
 900                 *len = strlen(*reply) + 1;
 901                 r = 0;
 902         }
 903         /* else if (r < 0) leave *reply alone */
 904
 905         lock_cleanup_pop(vecs->lock);
 906         return r;
 907 }
 908
 909 static int
 910 uev_discard(char * devpath)
 911 {
 912         char a[10], b[10];
 913
 914         /*
 915          * keep only block devices, discard partitions
 916          */
 917         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 918             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 919                 condlog(4, "discard event on %s", devpath);
 920                 return 1;
 921         }
 922         return 0;
 923 }
 924
 925 int
 926 uev_trigger (struct uevent * uev, void * trigger_data)
 927 {
 928         int r = 0;
 929         char devname[32];
 930         struct vectors * vecs;
 931
 932         vecs = (struct vectors *)trigger_data;
 933
 934         if (uev_discard(uev->devpath))
 935                 return 0;
 936
 937         basename(uev->devpath, devname);
 938         lock(vecs->lock);
 939
 940         /*
 941          * device map add/remove event
 942          */
 943         if (!strncmp(devname, "dm-", 3)) {
 944                 if (!strncmp(uev->action, "add", 3)) {
 945                         r = uev_add_map(devname, vecs);
 946                         goto out;
 947                 }
 948                 if (!strncmp(uev->action, "remove", 6)) {
 949                         r = uev_remove_map(devname, vecs);
 950                         goto out;
 951                 }
 952                 if (!strncmp(uev->action, "umount", 6)) {
 953                         r = uev_umount_map(devname, vecs);
 954                         goto out;
 955                 }
 956                 goto out;
 957         }
 958
 959         /*
 960          * path add/remove event
 961          */
 962         if (blacklist(conf->blist_devnode, devname))
 963                 goto out;
 964
 965         if (!strncmp(uev->action, "add", 3)) {
 966                 r = uev_add_path(devname, vecs);
 967                 goto out;
 968         }
 969         if (!strncmp(uev->action, "remove", 6)) {
 970                 r = uev_remove_path(devname, vecs);
 971                 goto out;
 972         }
 973
 974 out:
 975         unlock(vecs->lock);
 976         return r;
 977 }
 978
 979 static void *
 980 ueventloop (void * ap)
 981 {
 982         if (uevent_listen(&uev_trigger, ap))
 983                 fprintf(stderr, "error starting uevent listener");
 984
 985         return NULL;
 986 }
 987
 988 static void *
 989 uxlsnrloop (void * ap)
 990 {
 991         if (load_keys())
 992                 return NULL;
 993
 994         if (alloc_handlers())
 995                 return NULL;
 996
 997         add_handler(LIST+PATHS, cli_list_paths);
 998         add_handler(LIST+MAPS, cli_list_maps);
 999         add_handler(LIST+MAPS+STATUS, cli_list_maps_status);
1000         add_handler(LIST+MAPS+STATS, cli_list_maps_stats);
1001         add_handler(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
1002         add_handler(LIST+TOPOLOGY, cli_list_maps_topology);
1003         add_handler(LIST+MAP+TOPOLOGY, cli_list_map_topology);
1004         add_handler(LIST+CONFIG, cli_list_config);
1005         add_handler(ADD+PATH, cli_add_path);
1006         add_handler(DEL+PATH, cli_del_path);
1007         add_handler(ADD+MAP, cli_add_map);
1008         add_handler(DEL+MAP, cli_del_map);
1009         add_handler(SWITCH+MAP+GROUP, cli_switch_group);
1010         add_handler(RECONFIGURE, cli_reconfigure);
1011         add_handler(SUSPEND+MAP, cli_suspend);
1012         add_handler(RESUME+MAP, cli_resume);
1013         add_handler(REINSTATE+PATH, cli_reinstate);
1014         add_handler(FAIL+PATH, cli_fail);
1015
1016         uxsock_listen(&uxsock_trigger, ap);
1017
1018         return NULL;
1019 }
1020
1021 static int
1022 exit_daemon (int status)
1023 {
1024         if (status != 0)
1025                 fprintf(stderr, "bad exit status. see daemon.log\n");
1026
1027         condlog(3, "unlink pidfile");
1028         unlink(DEFAULT_PIDFILE);
1029
1030         lock(&exit_mutex);
1031         pthread_cond_signal(&exit_cond);
1032         unlock(&exit_mutex);
1033
1034         return status;
1035 }
1036
1037 static void
1038 fail_path (struct path * pp, int del_active)
1039 {
1040         if (!pp->mpp)
1041                 return;
1042
1043         condlog(2, "checker failed path %s in map %s",
1044                  pp->dev_t, pp->mpp->alias);
1045
1046         dm_fail_path(pp->mpp->alias, pp->dev_t);
1047         if (del_active)
1048                 update_queue_mode_del_path(pp->mpp);
1049 }
1050
1051 /*
1052  * caller must have locked the path list before calling that function
1053  */
1054 static void
1055 reinstate_path (struct path * pp, int add_active)
1056 {
1057         if (!pp->mpp)
1058                 return;
1059
1060         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
1061                 condlog(0, "%s: reinstate failed", pp->dev_t);
1062         else {
1063                 condlog(2, "%s: reinstated", pp->dev_t);
1064                 if (add_active)
1065                         update_queue_mode_add_path(pp->mpp);
1066         }
1067 }
1068
1069 static void
1070 enable_group(struct path * pp)
1071 {
1072         struct pathgroup * pgp;
1073
1074         /*
1075          * if path is added through uev_add_path, pgindex can be unset.
1076          * next update_strings() will set it, upon map reload event.
1077          *
1078          * we can safely return here, because upon map reload, all
1079          * PG will be enabled.
1080          */
1081         if (!pp->mpp->pg || !pp->pgindex)
1082                 return;
1083
1084         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1085
1086         if (pgp->status == PGSTATE_DISABLED) {
1087                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1088                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1089         }
1090 }
1091
1092 static void
1093 mpvec_garbage_collector (struct vectors * vecs)
1094 {
1095         struct multipath * mpp;
1096         int i;
1097
1098         vector_foreach_slot (vecs->mpvec, mpp, i) {
1099                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1100                         condlog(2, "%s: remove dead map", mpp->alias);
1101                         remove_map(mpp, vecs, stop_waiter_thread, 1);
1102                         i--;
1103                 }
1104         }
1105 }
1106
1107 static void
1108 defered_failback_tick (vector mpvec)
1109 {
1110         struct multipath * mpp;
1111         int i;
1112
1113         vector_foreach_slot (mpvec, mpp, i) {
1114                 /*
1115                  * defered failback getting sooner
1116                  */
1117                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1118                         mpp->failback_tick--;
1119
1120                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1121                                 switch_pathgroup(mpp);
1122                 }
1123         }
1124 }
1125
1126 static void
1127 retry_count_tick(vector mpvec)
1128 {
1129         struct multipath *mpp;
1130         int i;
1131
1132         vector_foreach_slot (mpvec, mpp, i) {
1133                 if (mpp->retry_tick) {
1134                         mpp->stat_total_queueing_time++;
1135                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1136                         if(--mpp->retry_tick == 0) {
1137                                 dm_queue_if_no_path(mpp->alias, 0);
1138                                 condlog(2, "%s: Disable queueing", mpp->alias);
1139                         }
1140                 }
1141         }
1142 }
1143
1144 static void *
1145 checkerloop (void *ap)
1146 {
1147         struct vectors *vecs;
1148         struct path *pp;
1149         int i, count = 0;
1150         int newstate;
1151
1152         mlockall(MCL_CURRENT | MCL_FUTURE);
1153         vecs = (struct vectors *)ap;
1154         condlog(2, "path checkers start up");
1155
1156         /*
1157          * init the path check interval
1158          */
1159         vector_foreach_slot (vecs->pathvec, pp, i) {
1160                 pp->checkint = conf->checkint;
1161         }
1162
1163         while (1) {
1164                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1165                 lock(vecs->lock);
1166                 condlog(4, "tick");
1167
1168                 vector_foreach_slot (vecs->pathvec, pp, i) {
1169                         if (!pp->mpp)
1170                                 continue;
1171
1172                         if (pp->tick && --pp->tick)
1173                                 continue; /* don't check this path yet */
1174
1175                         /*
1176                          * provision a next check soonest,
1177                          * in case we exit abnormaly from here
1178                          */
1179                         pp->tick = conf->checkint;
1180
1181                         if (!checker_selected(&pp->checker)) {
1182                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
1183                                 select_checker(pp);
1184                         }
1185
1186                         if (!checker_selected(&pp->checker)) {
1187                                 condlog(0, "%s: checker is not set", pp->dev);
1188                                 continue;
1189                         }
1190                         newstate = checker_check(&pp->checker);
1191
1192                         if (newstate < 0) {
1193                                 condlog(2, "%s: unusable path", pp->dev);
1194                                 pathinfo(pp, conf->hwtable, 0);
1195                                 continue;
1196                         }
1197
1198                         if (newstate != pp->state) {
1199                                 int oldstate = pp->state;
1200                                 pp->state = newstate;
1201                                 LOG_MSG(1, checker_message(&pp->checker));
1202
1203                                 /*
1204                                  * upon state change, reset the checkint
1205                                  * to the shortest delay
1206                                  */
1207                                 pp->checkint = conf->checkint;
1208
1209                                 if (newstate == PATH_DOWN ||
1210                                     newstate == PATH_SHAKY ||
1211                                     update_multipath_strings(pp->mpp,
1212                                                              vecs->pathvec)) {
1213                                         /*
1214                                          * proactively fail path in the DM
1215                                          */
1216                                         if (oldstate == PATH_UP ||
1217                                             oldstate == PATH_GHOST)
1218                                                 fail_path(pp, 1);
1219                                         else
1220                                                 fail_path(pp, 0);
1221
1222                                         /*
1223                                          * cancel scheduled failback
1224                                          */
1225                                         pp->mpp->failback_tick = 0;
1226
1227                                         pp->mpp->stat_path_failures++;
1228                                         continue;
1229                                 }
1230
1231                                 /*
1232                                  * reinstate this path
1233                                  */
1234                                 if (oldstate != PATH_UP &&
1235                                     oldstate != PATH_GHOST)
1236                                         reinstate_path(pp, 1);
1237                                 else
1238                                         reinstate_path(pp, 0);
1239
1240                                 /*
1241                                  * schedule [defered] failback
1242                                  */
1243                                 if (pp->mpp->pgfailback > 0)
1244                                         pp->mpp->failback_tick =
1245                                                 pp->mpp->pgfailback + 1;
1246                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
1247                                     need_switch_pathgroup(pp->mpp, 1))
1248                                         switch_pathgroup(pp->mpp);
1249
1250                                 /*
1251                                  * if at least one path is up in a group, and
1252                                  * the group is disabled, re-enable it
1253                                  */
1254                                 if (newstate == PATH_UP)
1255                                         enable_group(pp);
1256                         }
1257                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1258                                 LOG_MSG(4, checker_message(&pp->checker));
1259                                 /*
1260                                  * double the next check delay.
1261                                  * max at conf->max_checkint
1262                                  */
1263                                 if (pp->checkint < (conf->max_checkint / 2))
1264                                         pp->checkint = 2 * pp->checkint;
1265                                 else
1266                                         pp->checkint = conf->max_checkint;
1267
1268                                 pp->tick = pp->checkint;
1269                                 condlog(4, "%s: delay next check %is",
1270                                                 pp->dev_t, pp->tick);
1271
1272                         }
1273                         pp->state = newstate;
1274
1275                         /*
1276                          * path prio refreshing
1277                          */
1278                         condlog(4, "path prio refresh");
1279                         pathinfo(pp, conf->hwtable, DI_PRIO);
1280
1281                         if (need_switch_pathgroup(pp->mpp, 0)) {
1282                                 if (pp->mpp->pgfailback > 0 &&
1283                                     pp->mpp->failback_tick <= 0)
1284                                         pp->mpp->failback_tick =
1285                                                 pp->mpp->pgfailback + 1;
1286                                 else if (pp->mpp->pgfailback ==
1287                                                 -FAILBACK_IMMEDIATE)
1288                                         switch_pathgroup(pp->mpp);
1289                         }
1290                 }
1291                 defered_failback_tick(vecs->mpvec);
1292                 retry_count_tick(vecs->mpvec);
1293
1294                 if (count)
1295                         count--;
1296                 else {
1297                         condlog(4, "map garbage collection");
1298                         mpvec_garbage_collector(vecs);
1299                         count = MAPGCINT;
1300                 }
1301
1302                 lock_cleanup_pop(vecs->lock);
1303                 sleep(1);
1304         }
1305         return NULL;
1306 }
1307
1308 int
1309 configure (struct vectors * vecs, int start_waiters)
1310 {
1311         struct multipath * mpp;
1312         struct path * pp;
1313         vector mpvec;
1314         int i;
1315
1316         if (!(vecs->pathvec = vector_alloc()))
1317                 return 1;
1318
1319         if (!(vecs->mpvec = vector_alloc()))
1320                 return 1;
1321
1322         if (!(mpvec = vector_alloc()))
1323                 return 1;
1324
1325         /*
1326          * probe for current path (from sysfs) and map (from dm) sets
1327          */
1328         path_discovery(vecs->pathvec, conf, DI_ALL);
1329
1330         vector_foreach_slot (vecs->pathvec, pp, i){
1331                 if (blacklist_path(conf, pp)){
1332                         vector_del_slot(vecs->pathvec, i);
1333                         free_path(pp);
1334                         i--;
1335                 }
1336                 else
1337                         pp->checkint = conf->checkint;
1338         }
1339         if (map_discovery(vecs))
1340                 return 1;
1341
1342         /*
1343          * create new set of maps & push changed ones into dm
1344          */
1345         if (coalesce_paths(vecs, mpvec, NULL))
1346                 return 1;
1347
1348         /*
1349          * may need to remove some maps which are no longer relevant
1350          * e.g., due to blacklist changes in conf file
1351          */
1352         if (coalesce_maps(vecs, mpvec))
1353                 return 1;
1354
1355         dm_lib_release();
1356
1357         if (conf->verbosity > 2)
1358                 vector_foreach_slot(mpvec, mpp, i)
1359                         print_map(mpp);
1360
1361         /*
1362          * purge dm of old maps
1363          */
1364         remove_maps(vecs, NULL);
1365
1366         /*
1367          * save new set of maps formed by considering current path state
1368          */
1369         vecs->mpvec = mpvec;
1370
1371         /*
1372          * start dm event waiter threads for these new maps
1373          */
1374         vector_foreach_slot(vecs->mpvec, mpp, i) {
1375                 if (setup_multipath(vecs, mpp))
1376                         return 1;
1377                 if (start_waiters)
1378                         if (start_waiter_thread(mpp, vecs))
1379                                 return 1;
1380         }
1381         return 0;
1382 }
1383
1384 int
1385 reconfigure (struct vectors * vecs)
1386 {
1387         struct config * old = conf;
1388
1389         /*
1390          * free old map and path vectors ... they use old conf state
1391          */
1392         if (VECTOR_SIZE(vecs->mpvec))
1393                 remove_maps(vecs, stop_waiter_thread);
1394
1395         if (VECTOR_SIZE(vecs->pathvec))
1396                 free_pathvec(vecs->pathvec, FREE_PATHS);
1397
1398         conf = NULL;
1399
1400         if (load_config(DEFAULT_CONFIGFILE))
1401                 return 1;
1402
1403         conf->verbosity = old->verbosity;
1404
1405         if (!conf->checkint) {
1406                 conf->checkint = DEFAULT_CHECKINT;
1407                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1408         }
1409         configure(vecs, 1);
1410         free_config(old);
1411         return 0;
1412 }
1413
1414 static struct vectors *
1415 init_vecs (void)
1416 {
1417         struct vectors * vecs;
1418
1419         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1420
1421         if (!vecs)
1422                 return NULL;
1423
1424         vecs->lock =
1425                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1426
1427         if (!vecs->lock)
1428                 goto out;
1429
1430         vecs->pathvec = vector_alloc();
1431
1432         if (!vecs->pathvec)
1433                 goto out1;
1434
1435         vecs->mpvec = vector_alloc();
1436
1437         if (!vecs->mpvec)
1438                 goto out2;
1439
1440         pthread_mutex_init(vecs->lock, NULL);
1441
1442         return vecs;
1443
1444 out2:
1445         vector_free(vecs->pathvec);
1446 out1:
1447         FREE(vecs->lock);
1448 out:
1449         FREE(vecs);
1450         condlog(0, "failed to init paths");
1451         return NULL;
1452 }
1453
1454 static void *
1455 signal_set(int signo, void (*func) (int))
1456 {
1457         int r;
1458         struct sigaction sig;
1459         struct sigaction osig;
1460
1461         sig.sa_handler = func;
1462         sigemptyset(&sig.sa_mask);
1463         sig.sa_flags = 0;
1464
1465         r = sigaction(signo, &sig, &osig);
1466
1467         if (r < 0)
1468                 return (SIG_ERR);
1469         else
1470                 return (osig.sa_handler);
1471 }
1472
1473 static void
1474 sighup (int sig)
1475 {
1476         condlog(2, "reconfigure (SIGHUP)");
1477
1478         lock(gvecs->lock);
1479         reconfigure(gvecs);
1480         unlock(gvecs->lock);
1481
1482 #ifdef _DEBUG_
1483         dbg_free_final(NULL);
1484 #endif
1485 }
1486
1487 static void
1488 sigend (int sig)
1489 {
1490         exit_daemon(0);
1491 }
1492
1493 static void
1494 sigusr1 (int sig)
1495 {
1496         condlog(3, "SIGUSR1 received");
1497 }
1498
1499 static void
1500 signal_init(void)
1501 {
1502         signal_set(SIGHUP, sighup);
1503         signal_set(SIGUSR1, sigusr1);
1504         signal_set(SIGINT, sigend);
1505         signal_set(SIGTERM, sigend);
1506         signal_set(SIGKILL, sigend);
1507 }
1508
1509 static void
1510 setscheduler (void)
1511 {
1512         int res;
1513         static struct sched_param sched_param = {
1514                 sched_priority: 99
1515         };
1516
1517         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1518
1519         if (res == -1)
1520                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1521         return;
1522 }
1523
1524 static void
1525 set_oom_adj (int val)
1526 {
1527         FILE *fp;
1528
1529         fp = fopen("/proc/self/oom_adj", "w");
1530
1531         if (!fp)
1532                 return;
1533
1534         fprintf(fp, "%i", val);
1535         fclose(fp);
1536 }
1537
1538 static int
1539 child (void * param)
1540 {
1541         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1542         pthread_attr_t attr;
1543         struct vectors * vecs;
1544
1545         mlockall(MCL_CURRENT | MCL_FUTURE);
1546
1547         if (logsink)
1548                 log_thread_start();
1549
1550         condlog(2, "--------start up--------");
1551         condlog(2, "read " DEFAULT_CONFIGFILE);
1552
1553         if (load_config(DEFAULT_CONFIGFILE))
1554                 exit(1);
1555
1556         setlogmask(LOG_UPTO(conf->verbosity + 3));
1557
1558         /*
1559          * fill the voids left in the config file
1560          */
1561         if (!conf->checkint) {
1562                 conf->checkint = DEFAULT_CHECKINT;
1563                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1564         }
1565
1566         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1567                 if (logsink)
1568                         log_thread_stop();
1569
1570                 exit(1);
1571         }
1572         signal_init();
1573         setscheduler();
1574         set_oom_adj(-16);
1575         vecs = gvecs = init_vecs();
1576
1577         if (!vecs)
1578                 exit(1);
1579
1580         if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1581                 condlog(0, "can not find sysfs mount point");
1582                 exit(1);
1583         }
1584
1585         /*
1586          * fetch and configure both paths and multipaths
1587          */
1588         if (configure(vecs, 1)) {
1589                 condlog(0, "failure during configuration");
1590                 exit(1);
1591         }
1592
1593         /*
1594          * start threads
1595          */
1596         pthread_attr_init(&attr);
1597         pthread_attr_setstacksize(&attr, 64 * 1024);
1598         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1599
1600         pthread_create(&check_thr, &attr, checkerloop, vecs);
1601         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1602         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1603
1604         pthread_cond_wait(&exit_cond, &exit_mutex);
1605
1606         /*
1607          * exit path
1608          */
1609         lock(vecs->lock);
1610         remove_maps(vecs, stop_waiter_thread);
1611         free_pathvec(vecs->pathvec, FREE_PATHS);
1612
1613         pthread_cancel(check_thr);
1614         pthread_cancel(uevent_thr);
1615         pthread_cancel(uxlsnr_thr);
1616
1617         free_keys(keys);
1618         keys = NULL;
1619         free_handlers(handlers);
1620         handlers = NULL;
1621         free_polls();
1622
1623         unlock(vecs->lock);
1624         pthread_mutex_destroy(vecs->lock);
1625         FREE(vecs->lock);
1626         vecs->lock = NULL;
1627         FREE(vecs);
1628         vecs = NULL;
1629         free_config(conf);
1630         conf = NULL;
1631
1632         condlog(2, "--------shut down-------");
1633
1634         if (logsink)
1635                 log_thread_stop();
1636
1637         dm_lib_release();
1638         dm_lib_exit();
1639
1640 #ifdef _DEBUG_
1641         dbg_free_final(NULL);
1642 #endif
1643
1644         exit(0);
1645 }
1646
1647 static int
1648 daemonize(void)
1649 {
1650         int pid;
1651         int in_fd, out_fd;
1652
1653         if( (pid = fork()) < 0){
1654                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1655                 return -1;
1656         }
1657         else if (pid != 0)
1658                 return pid;
1659
1660         setsid();
1661
1662         if ( (pid = fork()) < 0)
1663                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1664         else if (pid != 0)
1665                 _exit(0);
1666
1667         in_fd = open("/dev/null", O_RDONLY);
1668         if (in_fd < 0){
1669                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1670                         strerror(errno));
1671                 _exit(0);
1672         }
1673         out_fd = open("/dev/console", O_WRONLY);
1674         if (out_fd < 0){
1675                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1676                         strerror(errno));
1677                 _exit(0);
1678         }
1679
1680         close(STDIN_FILENO);
1681         dup(in_fd);
1682         close(STDOUT_FILENO);
1683         dup(out_fd);
1684         close(STDERR_FILENO);
1685         dup(out_fd);
1686
1687         close(in_fd);
1688         close(out_fd);
1689         chdir("/");
1690         umask(0);
1691         return 0;
1692 }
1693
1694 int
1695 main (int argc, char *argv[])
1696 {
1697         extern char *optarg;
1698         extern int optind;
1699         int arg;
1700         int err;
1701
1702         logsink = 1;
1703
1704         if (getuid() != 0) {
1705                 fprintf(stderr, "need to be root\n");
1706                 exit(1);
1707         }
1708
1709         /* make sure we don't lock any path */
1710         chdir("/");
1711         umask(umask(077) | 022);
1712
1713         conf = alloc_config();
1714
1715         if (!conf)
1716                 exit(1);
1717
1718         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1719         switch(arg) {
1720                 case 'd':
1721                         logsink = 0;
1722                         //debug=1; /* ### comment me out ### */
1723                         break;
1724                 case 'v':
1725                         if (sizeof(optarg) > sizeof(char *) ||
1726                             !isdigit(optarg[0]))
1727                                 exit(1);
1728
1729                         conf->verbosity = atoi(optarg);
1730                         break;
1731                 case 'k':
1732                         uxclnt(optarg);
1733                         exit(0);
1734                 default:
1735                         ;
1736                 }
1737         }
1738
1739         if (!logsink)
1740                 err = 0;
1741         else
1742                 err = daemonize();
1743
1744         if (err < 0)
1745                 /* error */
1746                 exit(1);
1747         else if (err > 0)
1748                 /* parent dies */
1749                 exit(0);
1750         else
1751                 /* child lives */
1752                 return (child(NULL));
1753 }
1754