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 void
 479 sync_map_state(struct multipath *mpp)
 480 {
 481         int i, j;
 482         struct pathgroup *pgp;
 483         struct path *pp;
 484
 485         vector_foreach_slot (mpp->pg, pgp, i){
 486                 vector_foreach_slot (pgp->paths, pp, j){
 487                         if (pp->state <= PATH_UNCHECKED)
 488                                 continue;
 489                         if ((pp->dmstate == PSTATE_FAILED ||
 490                              pp->dmstate == PSTATE_UNDEF) &&
 491                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 492                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 493                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 494                                   pp->dmstate == PSTATE_UNDEF) &&
 495                                  (pp->state == PATH_DOWN ||
 496                                   pp->state == PATH_SHAKY))
 497                                 dm_fail_path(mpp->alias, pp->dev_t);
 498                 }
 499         }
 500 }
 501
 502 static void
 503 sync_maps_state(vector mpvec)
 504 {
 505         int i;
 506         struct multipath *mpp;
 507
 508         vector_foreach_slot (mpvec, mpp, i)
 509                 sync_map_state(mpp);
 510 }
 511
 512 static int
 513 flush_map(struct multipath * mpp, struct vectors * vecs)
 514 {
 515         /*
 516          * clear references to this map before flushing so we can ignore
 517          * the spurious uevent we may generate with the dm_flush_map call below
 518          */
 519         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 520                 /*
 521                  * May not really be an error -- if the map was already flushed
 522                  * from the device mapper by dmsetup(8) for instance.
 523                  */
 524                 condlog(0, "%s: can't flush", mpp->alias);
 525                 return 1;
 526         }
 527         else {
 528                 dm_lib_release();
 529                 condlog(3, "%s: devmap removed", mpp->alias);
 530         }
 531
 532         orphan_paths(vecs->pathvec, mpp);
 533         remove_map(mpp, vecs, stop_waiter_thread, 1);
 534
 535         return 0;
 536 }
 537
 538 static int
 539 uev_add_map (char * devname, struct vectors * vecs)
 540 {
 541         condlog(2, "%s: add map (uevent)", devname);
 542         return ev_add_map(devname, vecs);
 543 }
 544
 545 int
 546 ev_add_map (char * devname, struct vectors * vecs)
 547 {
 548         int major, minor;
 549         char dev_t[BLK_DEV_SIZE];
 550         char * alias;
 551         char * refwwid;
 552         struct multipath * mpp;
 553         int map_present;
 554         int r = 1;
 555
 556         if (sscanf(devname, "dm-%d", &minor) == 1 &&
 557             !sysfs_get_dev(sysfs_path, devname, dev_t, BLK_DEV_SIZE) &&
 558             sscanf(dev_t, "%d:%d", &major, &minor) == 2)
 559                 alias = dm_mapname(major, minor);
 560         else
 561                 alias = STRDUP(devname);
 562
 563         if (!alias)
 564                 return 1;
 565
 566         map_present = dm_map_present(alias);
 567
 568         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 569                 condlog(4, "%s: not a multipath map", alias);
 570                 FREE(alias);
 571                 return 0;
 572         }
 573
 574         mpp = find_mp_by_alias(vecs->mpvec, alias);
 575
 576         if (mpp) {
 577                 /*
 578                  * Not really an error -- we generate our own uevent
 579                  * if we create a multipath mapped device as a result
 580                  * of uev_add_path
 581                  */
 582                 condlog(0, "%s: devmap already registered",
 583                         devname);
 584                 FREE(alias);
 585                 return 0;
 586         }
 587
 588         /*
 589          * now we can register the map
 590          */
 591         if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
 592                                         start_waiter_thread))) {
 593                 sync_map_state(mpp);
 594                 condlog(3, "%s: devmap %s added", alias, devname);
 595                 return 0;
 596         }
 597         refwwid = get_refwwid(devname, DEV_DEVMAP, vecs->pathvec);
 598
 599         if (refwwid) {
 600                 r = coalesce_paths(vecs, NULL, refwwid);
 601                 dm_lib_release();
 602         }
 603
 604         if (!r)
 605                 condlog(3, "%s: devmap %s added", alias, devname);
 606         else
 607                 condlog(0, "%s: uev_add_map %s failed", alias, devname);
 608
 609         FREE(refwwid);
 610         FREE(alias);
 611         return r;
 612 }
 613
 614 static int
 615 uev_remove_map (char * devname, struct vectors * vecs)
 616 {
 617         condlog(2, "%s: remove map (uevent)", devname);
 618         return ev_remove_map(devname, vecs);
 619 }
 620
 621 int
 622 ev_remove_map (char * devname, struct vectors * vecs)
 623 {
 624         struct multipath * mpp;
 625
 626         mpp = find_mp_by_str(vecs->mpvec, devname);
 627
 628         if (!mpp) {
 629                 condlog(3, "%s: devmap not registered, can't remove",
 630                         devname);
 631                 return 0;
 632         }
 633         flush_map(mpp, vecs);
 634
 635         return 0;
 636 }
 637
 638 static int
 639 uev_umount_map (char * devname, struct vectors * vecs)
 640 {
 641         struct multipath * mpp;
 642
 643         condlog(2, "%s: umount map (uevent)", devname);
 644
 645         mpp = find_mp_by_str(vecs->mpvec, devname);
 646
 647         if (!mpp)
 648                 return 0;
 649
 650         update_mpp_paths(mpp, vecs->pathvec);
 651         verify_paths(mpp, vecs, NULL);
 652
 653         if (!VECTOR_SIZE(mpp->paths))
 654                 flush_map(mpp, vecs);
 655
 656         return 0;
 657 }
 658
 659 static int
 660 uev_add_path (char * devname, struct vectors * vecs)
 661 {
 662         condlog(2, "%s: add path (uevent)", devname);
 663         return (ev_add_path(devname, vecs) != 1)? 0 : 1;
 664 }
 665
 666
 667 /*
 668  * returns:
 669  * 0: added
 670  * 1: error
 671  * 2: blacklisted
 672  */
 673 int
 674 ev_add_path (char * devname, struct vectors * vecs)
 675 {
 676         struct multipath * mpp;
 677         struct path * pp;
 678         char empty_buff[WWID_SIZE] = {0};
 679
 680         pp = find_path_by_dev(vecs->pathvec, devname);
 681
 682         if (pp) {
 683                 condlog(0, "%s: spurious uevent, path already in pathvec",
 684                         devname);
 685                 if (pp->mpp)
 686                         return 0;
 687         }
 688         else {
 689                 /*
 690                  * get path vital state
 691                  */
 692                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 693                       devname, DI_ALL))) {
 694                         condlog(0, "%s: failed to store path info", devname);
 695                         return 1;
 696                 }
 697                 pp->checkint = conf->checkint;
 698         }
 699
 700         /*
 701          * need path UID to go any further
 702          */
 703         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 704                 condlog(0, "%s: failed to get path uid", devname);
 705                 return 1; /* leave path added to pathvec */
 706         }
 707         if (blacklist_path(conf, pp)){
 708                 int i = find_slot(vecs->pathvec, (void *)pp);
 709                 if (i != -1)
 710                         vector_del_slot(vecs->pathvec, i);
 711                 free_path(pp);
 712                 return 2;
 713         }
 714         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 715 rescan:
 716         if (mpp) {
 717                 if (adopt_paths(vecs->pathvec, mpp))
 718                         return 1; /* leave path added to pathvec */
 719
 720                 verify_paths(mpp, vecs, NULL);
 721                 mpp->action = ACT_RELOAD;
 722         }
 723         else {
 724                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 725                         mpp->action = ACT_CREATE;
 726                 else
 727                         return 1; /* leave path added to pathvec */
 728         }
 729
 730         /*
 731          * push the map to the device-mapper
 732          */
 733         if (setup_map(mpp)) {
 734                 condlog(0, "%s: failed to setup map for addition of new "
 735                         "path %s", mpp->alias, devname);
 736                 goto out;
 737         }
 738         /*
 739          * reload the map for the multipath mapped device
 740          */
 741         if (domap(mpp) <= 0) {
 742                 condlog(0, "%s: failed in domap for addition of new "
 743                         "path %s", mpp->alias, devname);
 744                 /*
 745                  * deal with asynchronous uevents :((
 746                  */
 747                 if (mpp->action == ACT_RELOAD) {
 748                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 749                         sleep(1);
 750                         update_mpp_paths(mpp, vecs->pathvec);
 751                         goto rescan;
 752                 }
 753                 else
 754                         goto out;
 755         }
 756         dm_lib_release();
 757
 758         /*
 759          * update our state from kernel regardless of create or reload
 760          */
 761         if (setup_multipath(vecs, mpp))
 762                 goto out;
 763
 764         sync_map_state(mpp);
 765
 766         if (mpp->action == ACT_CREATE &&
 767             start_waiter_thread(mpp, vecs))
 768                         goto out;
 769
 770         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 771         return 0;
 772
 773 out:
 774         remove_map(mpp, vecs, NULL, 1);
 775         return 1;
 776 }
 777
 778 static int
 779 uev_remove_path (char * devname, struct vectors * vecs)
 780 {
 781         condlog(2, "%s: remove path (uevent)", devname);
 782         return ev_remove_path(devname, vecs);
 783 }
 784
 785 int
 786 ev_remove_path (char * devname, struct vectors * vecs)
 787 {
 788         struct multipath * mpp;
 789         struct path * pp;
 790         int i;
 791         int rm_path = 1;
 792
 793         pp = find_path_by_dev(vecs->pathvec, devname);
 794
 795         if (!pp) {
 796                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 797                 return 1;
 798         }
 799
 800         /*
 801          * avoid referring to the map of an orphanned path
 802          */
 803         if ((mpp = pp->mpp)) {
 804
 805                 /*
 806                  * remove the map IFF removing the last path
 807                  */
 808                 if (pathcount(mpp, PATH_WILD) > 1) {
 809                         vector rpvec = vector_alloc();
 810
 811                         /*
 812                          * transform the mp->pg vector of vectors of paths
 813                          * into a mp->params string to feed the device-mapper
 814                          */
 815                         update_mpp_paths(mpp, vecs->pathvec);
 816                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 817                                 vector_del_slot(mpp->paths, i);
 818
 819                         if (VECTOR_SIZE(mpp->paths) == 0) {
 820                                 char alias[WWID_SIZE];
 821
 822                                 /*
 823                                  * flush_map will fail if the device is open
 824                                  */
 825                                 strncpy(alias, mpp->alias, WWID_SIZE);
 826                                 if (flush_map(mpp, vecs))
 827                                         rm_path = 0;
 828                                 else
 829                                         condlog(3, "%s: removed map after removing"
 830                                                 " multiple paths", alias);
 831                         }
 832                         else {
 833                                 if (setup_map(mpp)) {
 834                                         condlog(0, "%s: failed to setup map for"
 835                                                 " removal of path %s", mpp->alias, devname);
 836                                         free_pathvec(rpvec, KEEP_PATHS);
 837                                         goto out;
 838                                 }
 839                                 /*
 840                                  * reload the map
 841                                  */
 842                                 mpp->action = ACT_RELOAD;
 843                                 if (domap(mpp) <= 0) {
 844                                         condlog(0, "%s: failed in domap for "
 845                                                 "removal of path %s",
 846                                                 mpp->alias, devname);
 847                                         /*
 848                                          * Delete path from pathvec so that
 849                                          * update_mpp_paths wont find it later
 850                                          * when/if another path is removed.
 851                                          */
 852                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 853                                                 vector_del_slot(vecs->pathvec, i);
 854                                         free_path(pp);
 855                                         return 1;
 856                                 }
 857                                 /*
 858                                  * update our state from kernel
 859                                  */
 860                                 if (setup_multipath(vecs, mpp)) {
 861                                         free_pathvec(rpvec, KEEP_PATHS);
 862                                         goto out;
 863                                 }
 864                                 sync_map_state(mpp);
 865
 866                                 condlog(3, "%s path removed from devmap %s",
 867                                         devname, mpp->alias);
 868                         }
 869                         free_pathvec(rpvec, KEEP_PATHS);
 870                 }
 871                 else {
 872                         char alias[WWID_SIZE];
 873
 874                         /*
 875                          * flush_map will fail if the device is open
 876                          */
 877                         strncpy(alias, mpp->alias, WWID_SIZE);
 878                         if (flush_map(mpp, vecs))
 879                                 rm_path = 0;
 880                         else
 881                                 condlog(3, "%s: removed map", alias);
 882                 }
 883         }
 884
 885         if (rm_path) {
 886                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 887                         vector_del_slot(vecs->pathvec, i);
 888                 free_path(pp);
 889         }
 890
 891         return 0;
 892
 893 out:
 894         remove_map(mpp, vecs, stop_waiter_thread, 1);
 895         return 1;
 896 }
 897
 898 static int
 899 map_discovery (struct vectors * vecs)
 900 {
 901         int i;
 902         struct multipath * mpp;
 903
 904         if (dm_get_maps(vecs->mpvec, "multipath"))
 905                 return 1;
 906
 907         vector_foreach_slot (vecs->mpvec, mpp, i)
 908                 if (setup_multipath(vecs, mpp))
 909                         return 1;
 910
 911         return 0;
 912 }
 913
 914 int
 915 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 916 {
 917         struct vectors * vecs;
 918         int r;
 919
 920         *reply = NULL;
 921         *len = 0;
 922         vecs = (struct vectors *)trigger_data;
 923
 924         pthread_cleanup_push(cleanup_lock, vecs->lock);
 925         lock(vecs->lock);
 926
 927         r = parse_cmd(str, reply, len, vecs);
 928
 929         if (r > 0) {
 930                 *reply = STRDUP("fail\n");
 931                 *len = strlen(*reply) + 1;
 932                 r = 1;
 933         }
 934         else if (!r && *len == 0) {
 935                 *reply = STRDUP("ok\n");
 936                 *len = strlen(*reply) + 1;
 937                 r = 0;
 938         }
 939         /* else if (r < 0) leave *reply alone */
 940
 941         lock_cleanup_pop(vecs->lock);
 942         return r;
 943 }
 944
 945 static int
 946 uev_discard(char * devpath)
 947 {
 948         char a[10], b[10];
 949
 950         /*
 951          * keep only block devices, discard partitions
 952          */
 953         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 954             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 955                 condlog(4, "discard event on %s", devpath);
 956                 return 1;
 957         }
 958         return 0;
 959 }
 960
 961 int
 962 uev_trigger (struct uevent * uev, void * trigger_data)
 963 {
 964         int r = 0;
 965         char devname[32];
 966         struct vectors * vecs;
 967
 968         vecs = (struct vectors *)trigger_data;
 969
 970         if (uev_discard(uev->devpath))
 971                 return 0;
 972
 973         basename(uev->devpath, devname);
 974         lock(vecs->lock);
 975
 976         /*
 977          * device map add/remove event
 978          */
 979         if (!strncmp(devname, "dm-", 3)) {
 980                 if (!strncmp(uev->action, "add", 3)) {
 981                         r = uev_add_map(devname, vecs);
 982                         goto out;
 983                 }
 984                 if (!strncmp(uev->action, "remove", 6)) {
 985                         r = uev_remove_map(devname, vecs);
 986                         goto out;
 987                 }
 988                 if (!strncmp(uev->action, "umount", 6)) {
 989                         r = uev_umount_map(devname, vecs);
 990                         goto out;
 991                 }
 992                 goto out;
 993         }
 994
 995         /*
 996          * path add/remove event
 997          */
 998         if (blacklist(conf->blist_devnode, devname))
 999                 goto out;
1000
1001         if (!strncmp(uev->action, "add", 3)) {
1002                 r = uev_add_path(devname, vecs);
1003                 goto out;
1004         }
1005         if (!strncmp(uev->action, "remove", 6)) {
1006                 r = uev_remove_path(devname, vecs);
1007                 goto out;
1008         }
1009
1010 out:
1011         unlock(vecs->lock);
1012         return r;
1013 }
1014
1015 static void *
1016 ueventloop (void * ap)
1017 {
1018         if (uevent_listen(&uev_trigger, ap))
1019                 fprintf(stderr, "error starting uevent listener");
1020
1021         return NULL;
1022 }
1023
1024 static void *
1025 uxlsnrloop (void * ap)
1026 {
1027         if (load_keys())
1028                 return NULL;
1029
1030         if (alloc_handlers())
1031                 return NULL;
1032
1033         add_handler(LIST+PATHS, cli_list_paths);
1034         add_handler(LIST+MAPS, cli_list_maps);
1035         add_handler(LIST+MAPS+STATUS, cli_list_maps_status);
1036         add_handler(LIST+MAPS+STATS, cli_list_maps_stats);
1037         add_handler(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
1038         add_handler(LIST+TOPOLOGY, cli_list_maps_topology);
1039         add_handler(LIST+MAP+TOPOLOGY, cli_list_map_topology);
1040         add_handler(LIST+CONFIG, cli_list_config);
1041         add_handler(ADD+PATH, cli_add_path);
1042         add_handler(DEL+PATH, cli_del_path);
1043         add_handler(ADD+MAP, cli_add_map);
1044         add_handler(DEL+MAP, cli_del_map);
1045         add_handler(SWITCH+MAP+GROUP, cli_switch_group);
1046         add_handler(RECONFIGURE, cli_reconfigure);
1047         add_handler(SUSPEND+MAP, cli_suspend);
1048         add_handler(RESUME+MAP, cli_resume);
1049         add_handler(REINSTATE+PATH, cli_reinstate);
1050         add_handler(FAIL+PATH, cli_fail);
1051
1052         uxsock_listen(&uxsock_trigger, ap);
1053
1054         return NULL;
1055 }
1056
1057 static int
1058 exit_daemon (int status)
1059 {
1060         if (status != 0)
1061                 fprintf(stderr, "bad exit status. see daemon.log\n");
1062
1063         condlog(3, "unlink pidfile");
1064         unlink(DEFAULT_PIDFILE);
1065
1066         lock(&exit_mutex);
1067         pthread_cond_signal(&exit_cond);
1068         unlock(&exit_mutex);
1069
1070         return status;
1071 }
1072
1073 static void
1074 fail_path (struct path * pp, int del_active)
1075 {
1076         if (!pp->mpp)
1077                 return;
1078
1079         condlog(2, "checker failed path %s in map %s",
1080                  pp->dev_t, pp->mpp->alias);
1081
1082         dm_fail_path(pp->mpp->alias, pp->dev_t);
1083         if (del_active)
1084                 update_queue_mode_del_path(pp->mpp);
1085 }
1086
1087 /*
1088  * caller must have locked the path list before calling that function
1089  */
1090 static void
1091 reinstate_path (struct path * pp, int add_active)
1092 {
1093         if (!pp->mpp)
1094                 return;
1095
1096         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
1097                 condlog(0, "%s: reinstate failed", pp->dev_t);
1098         else {
1099                 condlog(2, "%s: reinstated", pp->dev_t);
1100                 if (add_active)
1101                         update_queue_mode_add_path(pp->mpp);
1102         }
1103 }
1104
1105 static void
1106 enable_group(struct path * pp)
1107 {
1108         struct pathgroup * pgp;
1109
1110         /*
1111          * if path is added through uev_add_path, pgindex can be unset.
1112          * next update_strings() will set it, upon map reload event.
1113          *
1114          * we can safely return here, because upon map reload, all
1115          * PG will be enabled.
1116          */
1117         if (!pp->mpp->pg || !pp->pgindex)
1118                 return;
1119
1120         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1121
1122         if (pgp->status == PGSTATE_DISABLED) {
1123                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1124                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1125         }
1126 }
1127
1128 static void
1129 mpvec_garbage_collector (struct vectors * vecs)
1130 {
1131         struct multipath * mpp;
1132         int i;
1133
1134         vector_foreach_slot (vecs->mpvec, mpp, i) {
1135                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1136                         condlog(2, "%s: remove dead map", mpp->alias);
1137                         remove_map(mpp, vecs, stop_waiter_thread, 1);
1138                         i--;
1139                 }
1140         }
1141 }
1142
1143 static void
1144 defered_failback_tick (vector mpvec)
1145 {
1146         struct multipath * mpp;
1147         int i;
1148
1149         vector_foreach_slot (mpvec, mpp, i) {
1150                 /*
1151                  * defered failback getting sooner
1152                  */
1153                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1154                         mpp->failback_tick--;
1155
1156                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1157                                 switch_pathgroup(mpp);
1158                 }
1159         }
1160 }
1161
1162 static void
1163 retry_count_tick(vector mpvec)
1164 {
1165         struct multipath *mpp;
1166         int i;
1167
1168         vector_foreach_slot (mpvec, mpp, i) {
1169                 if (mpp->retry_tick) {
1170                         mpp->stat_total_queueing_time++;
1171                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1172                         if(--mpp->retry_tick == 0) {
1173                                 dm_queue_if_no_path(mpp->alias, 0);
1174                                 condlog(2, "%s: Disable queueing", mpp->alias);
1175                         }
1176                 }
1177         }
1178 }
1179
1180 static void *
1181 checkerloop (void *ap)
1182 {
1183         struct vectors *vecs;
1184         struct path *pp;
1185         int i, count = 0;
1186         int newstate;
1187
1188         mlockall(MCL_CURRENT | MCL_FUTURE);
1189         vecs = (struct vectors *)ap;
1190         condlog(2, "path checkers start up");
1191
1192         /*
1193          * init the path check interval
1194          */
1195         vector_foreach_slot (vecs->pathvec, pp, i) {
1196                 pp->checkint = conf->checkint;
1197         }
1198
1199         while (1) {
1200                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1201                 lock(vecs->lock);
1202                 condlog(4, "tick");
1203
1204                 vector_foreach_slot (vecs->pathvec, pp, i) {
1205                         if (!pp->mpp)
1206                                 continue;
1207
1208                         if (pp->tick && --pp->tick)
1209                                 continue; /* don't check this path yet */
1210
1211                         /*
1212                          * provision a next check soonest,
1213                          * in case we exit abnormaly from here
1214                          */
1215                         pp->tick = conf->checkint;
1216
1217                         if (!checker_selected(&pp->checker)) {
1218                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
1219                                 select_checker(pp);
1220                         }
1221
1222                         if (!checker_selected(&pp->checker)) {
1223                                 condlog(0, "%s: checker is not set", pp->dev);
1224                                 continue;
1225                         }
1226                         newstate = checker_check(&pp->checker);
1227
1228                         if (newstate < 0) {
1229                                 condlog(2, "%s: unusable path", pp->dev);
1230                                 pathinfo(pp, conf->hwtable, 0);
1231                                 continue;
1232                         }
1233
1234                         if (newstate != pp->state) {
1235                                 int oldstate = pp->state;
1236                                 pp->state = newstate;
1237                                 LOG_MSG(1, checker_message(&pp->checker));
1238
1239                                 /*
1240                                  * upon state change, reset the checkint
1241                                  * to the shortest delay
1242                                  */
1243                                 pp->checkint = conf->checkint;
1244
1245                                 if (newstate == PATH_DOWN ||
1246                                     newstate == PATH_SHAKY ||
1247                                     update_multipath_strings(pp->mpp,
1248                                                              vecs->pathvec)) {
1249                                         /*
1250                                          * proactively fail path in the DM
1251                                          */
1252                                         if (oldstate == PATH_UP ||
1253                                             oldstate == PATH_GHOST)
1254                                                 fail_path(pp, 1);
1255                                         else
1256                                                 fail_path(pp, 0);
1257
1258                                         /*
1259                                          * cancel scheduled failback
1260                                          */
1261                                         pp->mpp->failback_tick = 0;
1262
1263                                         pp->mpp->stat_path_failures++;
1264                                         continue;
1265                                 }
1266
1267                                 /*
1268                                  * reinstate this path
1269                                  */
1270                                 if (oldstate != PATH_UP &&
1271                                     oldstate != PATH_GHOST)
1272                                         reinstate_path(pp, 1);
1273                                 else
1274                                         reinstate_path(pp, 0);
1275
1276                                 /*
1277                                  * schedule [defered] failback
1278                                  */
1279                                 if (pp->mpp->pgfailback > 0)
1280                                         pp->mpp->failback_tick =
1281                                                 pp->mpp->pgfailback + 1;
1282                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
1283                                     need_switch_pathgroup(pp->mpp, 1))
1284                                         switch_pathgroup(pp->mpp);
1285
1286                                 /*
1287                                  * if at least one path is up in a group, and
1288                                  * the group is disabled, re-enable it
1289                                  */
1290                                 if (newstate == PATH_UP)
1291                                         enable_group(pp);
1292                         }
1293                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1294                                 LOG_MSG(4, checker_message(&pp->checker));
1295                                 /*
1296                                  * double the next check delay.
1297                                  * max at conf->max_checkint
1298                                  */
1299                                 if (pp->checkint < (conf->max_checkint / 2))
1300                                         pp->checkint = 2 * pp->checkint;
1301                                 else
1302                                         pp->checkint = conf->max_checkint;
1303
1304                                 pp->tick = pp->checkint;
1305                                 condlog(4, "%s: delay next check %is",
1306                                                 pp->dev_t, pp->tick);
1307
1308                         }
1309                         pp->state = newstate;
1310
1311                         /*
1312                          * path prio refreshing
1313                          */
1314                         condlog(4, "path prio refresh");
1315                         pathinfo(pp, conf->hwtable, DI_PRIO);
1316
1317                         if (need_switch_pathgroup(pp->mpp, 0)) {
1318                                 if (pp->mpp->pgfailback > 0 &&
1319                                     pp->mpp->failback_tick <= 0)
1320                                         pp->mpp->failback_tick =
1321                                                 pp->mpp->pgfailback + 1;
1322                                 else if (pp->mpp->pgfailback ==
1323                                                 -FAILBACK_IMMEDIATE)
1324                                         switch_pathgroup(pp->mpp);
1325                         }
1326                 }
1327                 defered_failback_tick(vecs->mpvec);
1328                 retry_count_tick(vecs->mpvec);
1329
1330                 if (count)
1331                         count--;
1332                 else {
1333                         condlog(4, "map garbage collection");
1334                         mpvec_garbage_collector(vecs);
1335                         count = MAPGCINT;
1336                 }
1337
1338                 lock_cleanup_pop(vecs->lock);
1339                 sleep(1);
1340         }
1341         return NULL;
1342 }
1343
1344 int
1345 configure (struct vectors * vecs, int start_waiters)
1346 {
1347         struct multipath * mpp;
1348         struct path * pp;
1349         vector mpvec;
1350         int i;
1351
1352         if (!(vecs->pathvec = vector_alloc()))
1353                 return 1;
1354
1355         if (!(vecs->mpvec = vector_alloc()))
1356                 return 1;
1357
1358         if (!(mpvec = vector_alloc()))
1359                 return 1;
1360
1361         /*
1362          * probe for current path (from sysfs) and map (from dm) sets
1363          */
1364         path_discovery(vecs->pathvec, conf, DI_ALL);
1365
1366         vector_foreach_slot (vecs->pathvec, pp, i){
1367                 if (blacklist_path(conf, pp)){
1368                         vector_del_slot(vecs->pathvec, i);
1369                         free_path(pp);
1370                         i--;
1371                 }
1372                 else
1373                         pp->checkint = conf->checkint;
1374         }
1375         if (map_discovery(vecs))
1376                 return 1;
1377
1378         /*
1379          * create new set of maps & push changed ones into dm
1380          */
1381         if (coalesce_paths(vecs, mpvec, NULL))
1382                 return 1;
1383
1384         /*
1385          * may need to remove some maps which are no longer relevant
1386          * e.g., due to blacklist changes in conf file
1387          */
1388         if (coalesce_maps(vecs, mpvec))
1389                 return 1;
1390
1391         dm_lib_release();
1392
1393         sync_maps_state(mpvec);
1394
1395         if (conf->verbosity > 2)
1396                 vector_foreach_slot(mpvec, mpp, i)
1397                         print_map(mpp);
1398
1399         /*
1400          * purge dm of old maps
1401          */
1402         remove_maps(vecs, NULL);
1403
1404         /*
1405          * save new set of maps formed by considering current path state
1406          */
1407         vecs->mpvec = mpvec;
1408
1409         /*
1410          * start dm event waiter threads for these new maps
1411          */
1412         vector_foreach_slot(vecs->mpvec, mpp, i) {
1413                 if (setup_multipath(vecs, mpp))
1414                         return 1;
1415                 if (start_waiters)
1416                         if (start_waiter_thread(mpp, vecs))
1417                                 return 1;
1418         }
1419         return 0;
1420 }
1421
1422 int
1423 reconfigure (struct vectors * vecs)
1424 {
1425         struct config * old = conf;
1426
1427         /*
1428          * free old map and path vectors ... they use old conf state
1429          */
1430         if (VECTOR_SIZE(vecs->mpvec))
1431                 remove_maps(vecs, stop_waiter_thread);
1432
1433         if (VECTOR_SIZE(vecs->pathvec))
1434                 free_pathvec(vecs->pathvec, FREE_PATHS);
1435
1436         conf = NULL;
1437
1438         if (load_config(DEFAULT_CONFIGFILE))
1439                 return 1;
1440
1441         conf->verbosity = old->verbosity;
1442
1443         if (!conf->checkint) {
1444                 conf->checkint = DEFAULT_CHECKINT;
1445                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1446         }
1447         configure(vecs, 1);
1448         free_config(old);
1449         return 0;
1450 }
1451
1452 static struct vectors *
1453 init_vecs (void)
1454 {
1455         struct vectors * vecs;
1456
1457         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1458
1459         if (!vecs)
1460                 return NULL;
1461
1462         vecs->lock =
1463                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1464
1465         if (!vecs->lock)
1466                 goto out;
1467
1468         vecs->pathvec = vector_alloc();
1469
1470         if (!vecs->pathvec)
1471                 goto out1;
1472
1473         vecs->mpvec = vector_alloc();
1474
1475         if (!vecs->mpvec)
1476                 goto out2;
1477
1478         pthread_mutex_init(vecs->lock, NULL);
1479
1480         return vecs;
1481
1482 out2:
1483         vector_free(vecs->pathvec);
1484 out1:
1485         FREE(vecs->lock);
1486 out:
1487         FREE(vecs);
1488         condlog(0, "failed to init paths");
1489         return NULL;
1490 }
1491
1492 static void *
1493 signal_set(int signo, void (*func) (int))
1494 {
1495         int r;
1496         struct sigaction sig;
1497         struct sigaction osig;
1498
1499         sig.sa_handler = func;
1500         sigemptyset(&sig.sa_mask);
1501         sig.sa_flags = 0;
1502
1503         r = sigaction(signo, &sig, &osig);
1504
1505         if (r < 0)
1506                 return (SIG_ERR);
1507         else
1508                 return (osig.sa_handler);
1509 }
1510
1511 static void
1512 sighup (int sig)
1513 {
1514         condlog(2, "reconfigure (SIGHUP)");
1515
1516         lock(gvecs->lock);
1517         reconfigure(gvecs);
1518         unlock(gvecs->lock);
1519
1520 #ifdef _DEBUG_
1521         dbg_free_final(NULL);
1522 #endif
1523 }
1524
1525 static void
1526 sigend (int sig)
1527 {
1528         exit_daemon(0);
1529 }
1530
1531 static void
1532 sigusr1 (int sig)
1533 {
1534         condlog(3, "SIGUSR1 received");
1535 }
1536
1537 static void
1538 signal_init(void)
1539 {
1540         signal_set(SIGHUP, sighup);
1541         signal_set(SIGUSR1, sigusr1);
1542         signal_set(SIGINT, sigend);
1543         signal_set(SIGTERM, sigend);
1544         signal_set(SIGKILL, sigend);
1545 }
1546
1547 static void
1548 setscheduler (void)
1549 {
1550         int res;
1551         static struct sched_param sched_param = {
1552                 sched_priority: 99
1553         };
1554
1555         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1556
1557         if (res == -1)
1558                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1559         return;
1560 }
1561
1562 static void
1563 set_oom_adj (int val)
1564 {
1565         FILE *fp;
1566
1567         fp = fopen("/proc/self/oom_adj", "w");
1568
1569         if (!fp)
1570                 return;
1571
1572         fprintf(fp, "%i", val);
1573         fclose(fp);
1574 }
1575
1576 static int
1577 child (void * param)
1578 {
1579         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1580         pthread_attr_t attr;
1581         struct vectors * vecs;
1582
1583         mlockall(MCL_CURRENT | MCL_FUTURE);
1584
1585         if (logsink)
1586                 log_thread_start();
1587
1588         condlog(2, "--------start up--------");
1589         condlog(2, "read " DEFAULT_CONFIGFILE);
1590
1591         if (load_config(DEFAULT_CONFIGFILE))
1592                 exit(1);
1593
1594         setlogmask(LOG_UPTO(conf->verbosity + 3));
1595
1596         /*
1597          * fill the voids left in the config file
1598          */
1599         if (!conf->checkint) {
1600                 conf->checkint = DEFAULT_CHECKINT;
1601                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1602         }
1603
1604         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1605                 if (logsink)
1606                         log_thread_stop();
1607
1608                 exit(1);
1609         }
1610         signal_init();
1611         setscheduler();
1612         set_oom_adj(-16);
1613         vecs = gvecs = init_vecs();
1614
1615         if (!vecs)
1616                 exit(1);
1617
1618         if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1619                 condlog(0, "can not find sysfs mount point");
1620                 exit(1);
1621         }
1622
1623         /*
1624          * fetch and configure both paths and multipaths
1625          */
1626         if (configure(vecs, 1)) {
1627                 condlog(0, "failure during configuration");
1628                 exit(1);
1629         }
1630
1631         /*
1632          * start threads
1633          */
1634         pthread_attr_init(&attr);
1635         pthread_attr_setstacksize(&attr, 64 * 1024);
1636         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1637
1638         pthread_create(&check_thr, &attr, checkerloop, vecs);
1639         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1640         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1641
1642         pthread_cond_wait(&exit_cond, &exit_mutex);
1643
1644         /*
1645          * exit path
1646          */
1647         lock(vecs->lock);
1648         remove_maps(vecs, stop_waiter_thread);
1649         free_pathvec(vecs->pathvec, FREE_PATHS);
1650
1651         pthread_cancel(check_thr);
1652         pthread_cancel(uevent_thr);
1653         pthread_cancel(uxlsnr_thr);
1654
1655         free_keys(keys);
1656         keys = NULL;
1657         free_handlers(handlers);
1658         handlers = NULL;
1659         free_polls();
1660
1661         unlock(vecs->lock);
1662         pthread_mutex_destroy(vecs->lock);
1663         FREE(vecs->lock);
1664         vecs->lock = NULL;
1665         FREE(vecs);
1666         vecs = NULL;
1667         free_config(conf);
1668         conf = NULL;
1669
1670         condlog(2, "--------shut down-------");
1671
1672         if (logsink)
1673                 log_thread_stop();
1674
1675         dm_lib_release();
1676         dm_lib_exit();
1677
1678 #ifdef _DEBUG_
1679         dbg_free_final(NULL);
1680 #endif
1681
1682         exit(0);
1683 }
1684
1685 static int
1686 daemonize(void)
1687 {
1688         int pid;
1689         int in_fd, out_fd;
1690
1691         if( (pid = fork()) < 0){
1692                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1693                 return -1;
1694         }
1695         else if (pid != 0)
1696                 return pid;
1697
1698         setsid();
1699
1700         if ( (pid = fork()) < 0)
1701                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1702         else if (pid != 0)
1703                 _exit(0);
1704
1705         in_fd = open("/dev/null", O_RDONLY);
1706         if (in_fd < 0){
1707                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1708                         strerror(errno));
1709                 _exit(0);
1710         }
1711         out_fd = open("/dev/console", O_WRONLY);
1712         if (out_fd < 0){
1713                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1714                         strerror(errno));
1715                 _exit(0);
1716         }
1717
1718         close(STDIN_FILENO);
1719         dup(in_fd);
1720         close(STDOUT_FILENO);
1721         dup(out_fd);
1722         close(STDERR_FILENO);
1723         dup(out_fd);
1724
1725         close(in_fd);
1726         close(out_fd);
1727         chdir("/");
1728         umask(0);
1729         return 0;
1730 }
1731
1732 int
1733 main (int argc, char *argv[])
1734 {
1735         extern char *optarg;
1736         extern int optind;
1737         int arg;
1738         int err;
1739
1740         logsink = 1;
1741
1742         if (getuid() != 0) {
1743                 fprintf(stderr, "need to be root\n");
1744                 exit(1);
1745         }
1746
1747         /* make sure we don't lock any path */
1748         chdir("/");
1749         umask(umask(077) | 022);
1750
1751         conf = alloc_config();
1752
1753         if (!conf)
1754                 exit(1);
1755
1756         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1757         switch(arg) {
1758                 case 'd':
1759                         logsink = 0;
1760                         //debug=1; /* ### comment me out ### */
1761                         break;
1762                 case 'v':
1763                         if (sizeof(optarg) > sizeof(char *) ||
1764                             !isdigit(optarg[0]))
1765                                 exit(1);
1766
1767                         conf->verbosity = atoi(optarg);
1768                         break;
1769                 case 'k':
1770                         uxclnt(optarg);
1771                         exit(0);
1772                 default:
1773                         ;
1774                 }
1775         }
1776
1777         if (!logsink)
1778                 err = 0;
1779         else
1780                 err = daemonize();
1781
1782         if (err < 0)
1783                 /* error */
1784                 exit(1);
1785         else if (err > 0)
1786                 /* parent dies */
1787                 exit(0);
1788         else
1789                 /* child lives */
1790                 return (child(NULL));
1791 }
1792