multipathd/main.c

   1 /*
   2  * Copyright (c) 2004, 2005 Christophe Varoqui
   3  * Copyright (c) 2005 Kiyoshi Ueda, NEC
   4  * Copyright (c) 2005 Benjamin Marzinski, Redhat
   5  * Copyright (c) 2005 Edward Goggin, EMC
   6  */
   7 #include <unistd.h>
   8 #include <sys/stat.h>
   9 #include <libdevmapper.h>
  10 #include <wait.h>
  11 #include <sys/mman.h>
  12 #include <sys/types.h>
  13 #include <fcntl.h>
  14 #include <errno.h>
  15
  16 /*
  17  * libsysfs
  18  */
  19 #include <sysfs/libsysfs.h>
  20 #include <sysfs/dlist.h>
  21
  22 /*
  23  * libcheckers
  24  */
  25 #include <checkers.h>
  26 #include <path_state.h>
  27
  28 /*
  29  * libmultipath
  30  */
  31 #include <parser.h>
  32 #include <vector.h>
  33 #include <memory.h>
  34 #include <config.h>
  35 #include <callout.h>
  36 #include <util.h>
  37 #include <blacklist.h>
  38 #include <hwtable.h>
  39 #include <defaults.h>
  40 #include <structs.h>
  41 #include <structs_vec.h>
  42 #include <dmparser.h>
  43 #include <devmapper.h>
  44 #include <dict.h>
  45 #include <discovery.h>
  46 #include <debug.h>
  47 #include <propsel.h>
  48 #include <uevent.h>
  49 #include <switchgroup.h>
  50 #include <path_state.h>
  51 #include <print.h>
  52 #include <configure.h>
  53
  54 #include "main.h"
  55 #include "pidfile.h"
  56 #include "uxlsnr.h"
  57 #include "uxclnt.h"
  58 #include "cli.h"
  59 #include "cli_handlers.h"
  60
  61 #define FILE_NAME_SIZE 256
  62 #define CMDSIZE 160
  63
  64 #define LOG_MSG(a,b) \
  65         if (strlen(b)) { \
  66                 condlog(a, "%s: %s", pp->dev_t, b); \
  67                 memset(b, 0, MAX_CHECKER_MSG_SIZE); \
  68         }
  69
  70 #ifdef LCKDBG
  71 #define lock(a) \
  72         fprintf(stderr, "%s:%s(%i) lock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  73         pthread_mutex_lock(a)
  74 #define unlock(a) \
  75         fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  76         pthread_mutex_unlock(a)
  77 #define lock_cleanup_pop(a) \
  78         fprintf(stderr, "%s:%s(%i) unlock %p\n", __FILE__, __FUNCTION__, __LINE__, a); \
  79         pthread_cleanup_pop(1);
  80 #else
  81 #define lock(a) pthread_mutex_lock(a)
  82 #define unlock(a) pthread_mutex_unlock(a)
  83 #define lock_cleanup_pop(a) pthread_cleanup_pop(1);
  84 #endif
  85
  86 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  87 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  88
  89 /*
  90  * structs
  91  */
  92 struct vectors * gvecs; /* global copy of vecs for use in sig handlers */
  93
  94 static struct event_thread *
  95 alloc_waiter (void)
  96 {
  97
  98         struct event_thread * wp;
  99
 100         wp = (struct event_thread *)MALLOC(sizeof(struct event_thread));
 101
 102         return wp;
 103 }
 104
 105 static void
 106 free_waiter (void * data)
 107 {
 108         struct event_thread * wp = (struct event_thread *)data;
 109
 110         /*
 111          * indicate in mpp that the wp is already freed storage
 112          */
 113         lock(wp->vecs->lock);
 114
 115         if (wp->mpp)
 116                 /*
 117                  * be careful, mpp may already be freed -- null if so
 118                  */
 119                 wp->mpp->waiter = NULL;
 120         else
 121                 condlog(3, "free_waiter, mpp freed before wp=%p,", wp);
 122
 123         unlock(wp->vecs->lock);
 124
 125         if (wp->dmt)
 126                 dm_task_destroy(wp->dmt);
 127
 128         FREE(wp);
 129 }
 130
 131 static void
 132 stop_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 133 {
 134         struct event_thread * wp = (struct event_thread *)mpp->waiter;
 135
 136         if (!wp) {
 137                 condlog(3, "%s: no waiter thread", mpp->alias);
 138                 return;
 139         }
 140         condlog(2, "%s: stop event checker thread", wp->mapname);
 141         pthread_kill((pthread_t)wp->thread, SIGUSR1);
 142 }
 143
 144 static void
 145 cleanup_lock (void * data)
 146 {
 147         unlock((pthread_mutex_t *)data);
 148 }
 149
 150 /*
 151  * mpp->no_path_retry:
 152  *   -2 (QUEUE) : queue_if_no_path enabled, never turned off
 153  *   -1 (FAIL)  : fail_if_no_path
 154  *    0 (UNDEF) : nothing
 155  *   >0         : queue_if_no_path enabled, turned off after polling n times
 156  */
 157 static void
 158 update_queue_mode_del_path(struct multipath *mpp)
 159 {
 160         if (--mpp->nr_active == 0 && mpp->no_path_retry > 0) {
 161                 /*
 162                  * Enter retry mode.
 163                  * meaning of +1: retry_tick may be decremented in
 164                  *                checkerloop before starting retry.
 165                  */
 166                 mpp->retry_tick = mpp->no_path_retry * conf->checkint + 1;
 167                 condlog(1, "%s: Entering recovery mode: max_retries=%d",
 168                         mpp->alias, mpp->no_path_retry);
 169         }
 170         condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
 171 }
 172
 173 static void
 174 update_queue_mode_add_path(struct multipath *mpp)
 175 {
 176         if (mpp->nr_active++ == 0 && mpp->no_path_retry > 0) {
 177                 /* come back to normal mode from retry mode */
 178                 mpp->retry_tick = 0;
 179                 dm_queue_if_no_path(mpp->alias, 1);
 180                 condlog(2, "%s: queue_if_no_path enabled", mpp->alias);
 181                 condlog(1, "%s: Recovered to normal mode", mpp->alias);
 182         }
 183         condlog(2, "%s: remaining active paths: %d", mpp->alias, mpp->nr_active);
 184 }
 185
 186 static int
 187 need_switch_pathgroup (struct multipath * mpp, int refresh)
 188 {
 189         struct pathgroup * pgp;
 190         struct path * pp;
 191         int i, j;
 192
 193         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
 194                 return 0;
 195
 196         /*
 197          * Refresh path priority values
 198          */
 199         if (refresh)
 200                 vector_foreach_slot (mpp->pg, pgp, i)
 201                         vector_foreach_slot (pgp->paths, pp, j)
 202                                 pathinfo(pp, conf->hwtable, DI_PRIO);
 203
 204         mpp->bestpg = select_path_group(mpp);
 205
 206         if (mpp->bestpg != mpp->nextpg)
 207                 return 1;
 208
 209         return 0;
 210 }
 211
 212 static void
 213 switch_pathgroup (struct multipath * mpp)
 214 {
 215         dm_switchgroup(mpp->alias, mpp->bestpg);
 216         condlog(2, "%s: switch to path group #%i",
 217                  mpp->alias, mpp->bestpg);
 218 }
 219
 220 static int
 221 coalesce_maps(struct vectors *vecs, vector nmpv)
 222 {
 223         struct multipath * ompp;
 224         vector ompv = vecs->mpvec;
 225         int i, j;
 226
 227         condlog(3, "coalesce_maps vs = %u", VECTOR_SIZE(ompv));
 228         condlog(3, "coalesce_maps vs = %u", VECTOR_SIZE(nmpv));
 229
 230         vector_foreach_slot (ompv, ompp, i) {
 231                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 232                         /*
 233                          * remove all current maps not allowed by the
 234                          * current configuration
 235                          */
 236                         if (dm_flush_map(ompp->alias, DEFAULT_TARGET)) {
 237                                 condlog(0, "%s: unable to flush devmap",
 238                                         ompp->alias);
 239                                 /*
 240                                  * may be just because the device is open
 241                                  */
 242                                 if (!vector_alloc_slot(nmpv))
 243                                         return 1;
 244
 245                                 vector_set_slot(nmpv, ompp);
 246                                 setup_multipath(vecs, ompp);
 247
 248                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 249                                         vector_del_slot(ompv, j);
 250
 251                                 continue;
 252                         }
 253                         else {
 254                                 dm_lib_release();
 255                                 condlog(3, "%s devmap removed", ompp->alias);
 256                         }
 257                 }
 258         }
 259         return 0;
 260 }
 261
 262 static int
 263 update_multipath (struct vectors *vecs, char *mapname)
 264 {
 265         struct multipath *mpp;
 266         struct pathgroup  *pgp;
 267         struct path *pp;
 268         int i, j;
 269         int r = 1;
 270
 271         mpp = find_mp_by_alias(vecs->mpvec, mapname);
 272
 273         if (!mpp)
 274                 goto out;
 275
 276         free_pgvec(mpp->pg, KEEP_PATHS);
 277         mpp->pg = NULL;
 278
 279         if (setup_multipath(vecs, mpp))
 280                 goto out; /* mpp freed in setup_multipath */
 281
 282         /*
 283          * compare checkers states with DM states
 284          */
 285         vector_foreach_slot (mpp->pg, pgp, i) {
 286                 vector_foreach_slot (pgp->paths, pp, j) {
 287                         if (pp->dmstate != PSTATE_FAILED)
 288                                 continue;
 289
 290                         if (pp->state != PATH_DOWN) {
 291                                 condlog(2, "%s: mark as failed", pp->dev_t);
 292                                 pp->state = PATH_DOWN;
 293                                 update_queue_mode_del_path(mpp);
 294
 295                                 /*
 296                                  * if opportune,
 297                                  * schedule the next check earlier
 298                                  */
 299                                 if (pp->tick > conf->checkint)
 300                                         pp->tick = conf->checkint;
 301                         }
 302                 }
 303         }
 304         r = 0;
 305 out:
 306         if (r)
 307                 condlog(0, "failed to update multipath");
 308
 309         return r;
 310 }
 311
 312 static sigset_t unblock_signals(void)
 313 {
 314         sigset_t set, old;
 315
 316         sigemptyset(&set);
 317         sigaddset(&set, SIGHUP);
 318         sigaddset(&set, SIGUSR1);
 319         pthread_sigmask(SIG_UNBLOCK, &set, &old);
 320         return old;
 321 }
 322
 323 /*
 324  * returns the reschedule delay
 325  * negative means *stop*
 326  */
 327 static int
 328 waiteventloop (struct event_thread * waiter)
 329 {
 330         sigset_t set;
 331         int event_nr;
 332         int r;
 333
 334         if (!waiter->event_nr)
 335                 waiter->event_nr = dm_geteventnr(waiter->mapname);
 336
 337         if (!(waiter->dmt = dm_task_create(DM_DEVICE_WAITEVENT))) {
 338                 condlog(0, "%s: devmap event #%i dm_task_create error",
 339                                 waiter->mapname, waiter->event_nr);
 340                 return 1;
 341         }
 342
 343         if (!dm_task_set_name(waiter->dmt, waiter->mapname)) {
 344                 condlog(0, "%s: devmap event #%i dm_task_set_name error",
 345                                 waiter->mapname, waiter->event_nr);
 346                 dm_task_destroy(waiter->dmt);
 347                 return 1;
 348         }
 349
 350         if (waiter->event_nr && !dm_task_set_event_nr(waiter->dmt,
 351                                                       waiter->event_nr)) {
 352                 condlog(0, "%s: devmap event #%i dm_task_set_event_nr error",
 353                                 waiter->mapname, waiter->event_nr);
 354                 dm_task_destroy(waiter->dmt);
 355                 return 1;
 356         }
 357
 358         dm_task_no_open_count(waiter->dmt);
 359
 360         /* accept wait interruption */
 361         set = unblock_signals();
 362
 363         /* interruption spits messages */
 364         dm_shut_log();
 365
 366         /* wait */
 367         r = dm_task_run(waiter->dmt);
 368
 369         /* wait is over : event or interrupt */
 370         pthread_sigmask(SIG_SETMASK, &set, NULL);
 371         //dm_restore_log();
 372
 373         if (!r) /* wait interrupted by signal */
 374                 return -1;
 375
 376         dm_task_destroy(waiter->dmt);
 377         waiter->dmt = NULL;
 378         waiter->event_nr++;
 379
 380         /*
 381          * upon event ...
 382          */
 383         while (1) {
 384                 condlog(3, "%s: devmap event #%i",
 385                                 waiter->mapname, waiter->event_nr);
 386
 387                 /*
 388                  * event might be :
 389                  *
 390                  * 1) a table reload, which means our mpp structure is
 391                  *    obsolete : refresh it through update_multipath()
 392                  * 2) a path failed by DM : mark as such through
 393                  *    update_multipath()
 394                  * 3) map has gone away : stop the thread.
 395                  * 4) a path reinstate : nothing to do
 396                  * 5) a switch group : nothing to do
 397                  */
 398                 pthread_cleanup_push(cleanup_lock, waiter->vecs->lock);
 399                 lock(waiter->vecs->lock);
 400                 r = update_multipath(waiter->vecs, waiter->mapname);
 401                 lock_cleanup_pop(waiter->vecs->lock);
 402
 403                 if (r)
 404                         return -1; /* stop the thread */
 405
 406                 event_nr = dm_geteventnr(waiter->mapname);
 407
 408                 if (waiter->event_nr == event_nr)
 409                         return 1; /* upon problem reschedule 1s later */
 410
 411                 waiter->event_nr = event_nr;
 412         }
 413         return -1; /* never reach there */
 414 }
 415
 416 static void *
 417 waitevent (void * et)
 418 {
 419         int r;
 420         struct event_thread *waiter;
 421
 422         mlockall(MCL_CURRENT | MCL_FUTURE);
 423
 424         waiter = (struct event_thread *)et;
 425         pthread_cleanup_push(free_waiter, et);
 426
 427         while (1) {
 428                 r = waiteventloop(waiter);
 429
 430                 if (r < 0)
 431                         break;
 432
 433                 sleep(r);
 434         }
 435
 436         pthread_cleanup_pop(1);
 437         return NULL;
 438 }
 439
 440 static int
 441 start_waiter_thread (struct multipath * mpp, struct vectors * vecs)
 442 {
 443         pthread_attr_t attr;
 444         struct event_thread * wp;
 445
 446         if (!mpp)
 447                 return 0;
 448
 449         if (pthread_attr_init(&attr))
 450                 goto out;
 451
 452         pthread_attr_setstacksize(&attr, 32 * 1024);
 453         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
 454
 455         wp = alloc_waiter();
 456
 457         if (!wp)
 458                 goto out;
 459
 460         mpp->waiter = (void *)wp;
 461         strncpy(wp->mapname, mpp->alias, WWID_SIZE);
 462         wp->vecs = vecs;
 463         wp->mpp = mpp;
 464
 465         if (pthread_create(&wp->thread, &attr, waitevent, wp)) {
 466                 condlog(0, "%s: cannot create event checker", wp->mapname);
 467                 goto out1;
 468         }
 469         condlog(2, "%s: event checker started", wp->mapname);
 470
 471         return 0;
 472 out1:
 473         free_waiter(wp);
 474         mpp->waiter = NULL;
 475 out:
 476         condlog(0, "failed to start waiter thread");
 477         return 1;
 478 }
 479
 480 static int
 481 flush_map(struct multipath * mpp, struct vectors * vecs)
 482 {
 483         /*
 484          * clear references to this map before flushing so we can ignore
 485          * the spurious uevent we may generate with the dm_flush_map call below
 486          */
 487         if (dm_flush_map(mpp->alias, DEFAULT_TARGET)) {
 488                 /*
 489                  * May not really be an error -- if the map was already flushed
 490                  * from the device mapper by dmsetup(8) for instance.
 491                  */
 492                 condlog(0, "%s: can't flush", mpp->alias);
 493                 return 1;
 494         }
 495         else {
 496                 dm_lib_release();
 497                 condlog(3, "%s: devmap removed", mpp->alias);
 498         }
 499
 500         orphan_paths(vecs->pathvec, mpp);
 501         remove_map(mpp, vecs, stop_waiter_thread, 1);
 502
 503         return 0;
 504 }
 505
 506 int
 507 uev_add_map (char * devname, struct vectors * vecs)
 508 {
 509         int major, minor;
 510         char dev_t[BLK_DEV_SIZE];
 511         char * alias;
 512         char * refwwid;
 513         struct multipath * mpp;
 514         int map_present;
 515         int r = 1;
 516
 517         condlog(3, "%s: uev_add_map", devname);
 518
 519         if (sscanf(devname, "dm-%d", &minor) == 1 &&
 520             !sysfs_get_dev(sysfs_path, devname, dev_t, BLK_DEV_SIZE) &&
 521             sscanf(dev_t, "%d:%d", &major, &minor) == 2)
 522                 alias = dm_mapname(major, minor);
 523         else
 524                 alias = STRDUP(devname);
 525
 526         if (!alias)
 527                 return 1;
 528
 529         map_present = dm_map_present(alias);
 530
 531         if (map_present && dm_type(alias, DEFAULT_TARGET) <= 0) {
 532                 condlog(4, "%s: not a multipath map", alias);
 533                 FREE(alias);
 534                 return 0;
 535         }
 536
 537         mpp = find_mp_by_alias(vecs->mpvec, alias);
 538
 539         if (mpp) {
 540                 /*
 541                  * Not really an error -- we generate our own uevent
 542                  * if we create a multipath mapped device as a result
 543                  * of uev_add_path
 544                  */
 545                 condlog(0, "%s: spurious uevent, devmap already registered",
 546                         devname);
 547                 FREE(alias);
 548                 return 0;
 549         }
 550
 551         /*
 552          * now we can register the map
 553          */
 554         if (map_present && (mpp = add_map_without_path(vecs, minor, alias,
 555                                         start_waiter_thread))) {
 556                 condlog(3, "%s: devmap %s added", alias, devname);
 557                 return 0;
 558         }
 559         refwwid = get_refwwid(devname, DEV_DEVMAP, vecs->pathvec);
 560
 561         if (refwwid) {
 562                 r = coalesce_paths(vecs, NULL, refwwid);
 563                 dm_lib_release();
 564         }
 565
 566         if (!r)
 567                 condlog(3, "%s devmap %s added", devname, alias);
 568         else
 569                 condlog(0, "%s: uev_add_map failed", alias);
 570
 571         FREE(refwwid);
 572         FREE(alias);
 573         return r;
 574 }
 575
 576 int
 577 uev_remove_map (char * devname, struct vectors * vecs)
 578 {
 579         int minor;
 580         struct multipath * mpp;
 581
 582         condlog(3, "%s: uev_remove_map", devname);
 583
 584         if (sscanf(devname, "dm-%d", &minor) == 1)
 585                 mpp = find_mp_by_minor(vecs->mpvec, minor);
 586         else
 587                 mpp = find_mp_by_alias(vecs->mpvec, devname);
 588
 589         if (!mpp) {
 590                 condlog(3, "%s: devmap not registered, can't remove",
 591                         devname);
 592                 return 0;
 593         }
 594
 595         condlog(2, "%s: remove devmap", mpp->alias);
 596         flush_map(mpp, vecs);
 597
 598         return 0;
 599 }
 600
 601 int
 602 uev_umount_map (char * devname, struct vectors * vecs)
 603 {
 604         int minor;
 605         struct multipath * mpp;
 606
 607         condlog(3, "%s: uev_umount_map", devname);
 608
 609         if (sscanf(devname, "dm-%d", &minor) == 1)
 610                 mpp = find_mp_by_minor(vecs->mpvec, minor);
 611         else
 612                 mpp = find_mp_by_alias(vecs->mpvec, devname);
 613
 614         if (!mpp)
 615                 return 0;
 616
 617         update_mpp_paths(mpp, vecs->pathvec);
 618         verify_paths(mpp, vecs, NULL);
 619
 620         if (!VECTOR_SIZE(mpp->paths))
 621                 flush_map(mpp, vecs);
 622
 623         return 0;
 624 }
 625
 626 int
 627 uev_add_path (char * devname, struct vectors * vecs)
 628 {
 629         struct multipath * mpp;
 630         struct path * pp;
 631         char empty_buff[WWID_SIZE] = {0};
 632
 633         condlog(3, "%s: uev_add_path", devname);
 634
 635         pp = find_path_by_dev(vecs->pathvec, devname);
 636         if (pp) {
 637                 condlog(0, "%s: spurious uevent, path already in pathvec, %p",
 638                         devname, pp->mpp);
 639                 /*
 640                  * allow reconfig of orphaned path here
 641                  */
 642                 if (pp->mpp) return 1;
 643         }
 644         else {
 645                 /*
 646                  * get path vital state
 647                  */
 648                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 649                       devname, DI_ALL))) {
 650                         condlog(0, "%s: failed to store path info", devname);
 651                         return 1;
 652                 }
 653                 pp->checkint = conf->checkint;
 654         }
 655
 656         /*
 657          * need path UID to go any further
 658          */
 659         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 660                 condlog(0, "%s: failed to get path uid", devname);
 661                 return 1; /* leave path added to pathvec */
 662         }
 663
 664         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 665 rescan:
 666         if (mpp) {
 667                 if (adopt_paths(vecs->pathvec, mpp))
 668                         return 1; /* leave path added to pathvec */
 669
 670                 verify_paths(mpp, vecs, NULL);
 671                 condlog(0, "%s: ownership set to %s", pp->dev_t, mpp->alias);
 672                 mpp->action = ACT_RELOAD;
 673         }
 674         else {
 675                 if ((mpp = add_map_with_path(vecs, pp, 1)))
 676                         mpp->action = ACT_CREATE;
 677                 else
 678                         return 1; /* leave path added to pathvec */
 679         }
 680
 681         /*
 682          * push the map to the device-mapper
 683          */
 684         if (setup_map(mpp)) {
 685                 condlog(0, "%s: failed to setup map for addition of new "
 686                         "path %s", mpp->alias, devname);
 687                 goto out;
 688         }
 689         /*
 690          * reload the map for the multipath mapped device
 691          */
 692         if (domap(mpp) <= 0) {
 693                 condlog(0, "%s: failed in domap for addition of new "
 694                         "path %s", mpp->alias, devname);
 695                 /*
 696                  * deal with asynchronous uevents :((
 697                  */
 698                 if (mpp->action == ACT_RELOAD) {
 699                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 700                         sleep(1);
 701                         update_mpp_paths(mpp, vecs->pathvec);
 702                         goto rescan;
 703                 }
 704                 else
 705                         goto out;
 706         }
 707         dm_lib_release();
 708
 709         /*
 710          * update our state from kernel regardless of create or reload
 711          */
 712         if (setup_multipath(vecs, mpp))
 713                 goto out;
 714
 715         if (mpp->action == ACT_CREATE &&
 716             start_waiter_thread(mpp, vecs))
 717                         goto out;
 718
 719         condlog(3, "%s path added to devmap %s", devname, mpp->alias);
 720         return 0;
 721
 722 out:
 723         remove_map(mpp, vecs, NULL, 1);
 724         return 1;
 725 }
 726
 727 int
 728 uev_remove_path (char * devname, struct vectors * vecs)
 729 {
 730         struct multipath * mpp;
 731         struct path * pp;
 732         int i;
 733         int rm_path = 1;
 734
 735         condlog(3, "%s: uev_remove_path", devname);
 736
 737         pp = find_path_by_dev(vecs->pathvec, devname);
 738         if (!pp) {
 739                 condlog(0, "%s: spurious uevent, path not in pathvec", devname);
 740                 return 1;
 741         }
 742
 743         /*
 744          * avoid referring to the map of an orphanned path
 745          */
 746         if ((mpp = pp->mpp)) {
 747
 748                 /*
 749                  * remove the map IFF removing the last path
 750                  */
 751                 if (pathcount(mpp, PATH_WILD) > 1) {
 752                         vector rpvec = vector_alloc();
 753
 754                         /*
 755                          * transform the mp->pg vector of vectors of paths
 756                          * into a mp->params string to feed the device-mapper
 757                          */
 758                         update_mpp_paths(mpp, vecs->pathvec);
 759                         if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 760                                 vector_del_slot(mpp->paths, i);
 761
 762                         if (VECTOR_SIZE(mpp->paths) == 0) {
 763                                 char alias[WWID_SIZE];
 764
 765                                 /*
 766                                  * flush_map will fail if the device is open
 767                                  */
 768                                 strncpy(alias, mpp->alias, WWID_SIZE);
 769                                 if (flush_map(mpp, vecs))
 770                                         rm_path = 0;
 771                                 else
 772                                         condlog(3, "%s: removed map after removing"
 773                                                 " multiple paths", alias);
 774                         }
 775                         else {
 776                                 if (setup_map(mpp)) {
 777                                         condlog(0, "%s: failed to setup map for"
 778                                                 " removal of path %s", mpp->alias, devname);
 779                                         free_pathvec(rpvec, KEEP_PATHS);
 780                                         goto out;
 781                                 }
 782                                 /*
 783                                  * reload the map
 784                                  */
 785                                 mpp->action = ACT_RELOAD;
 786                                 if (domap(mpp) <= 0) {
 787                                         condlog(0, "%s: failed in domap for "
 788                                                 "removal of path %s",
 789                                                 mpp->alias, devname);
 790                                         /*
 791                                          * Delete path from pathvec so that
 792                                          * update_mpp_paths wont find it later
 793                                          * when/if another path is removed.
 794                                          */
 795                                         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 796                                                 vector_del_slot(vecs->pathvec, i);
 797                                         free_path(pp);
 798                                         return 1;
 799                                 }
 800                                 /*
 801                                  * update our state from kernel
 802                                  */
 803                                 if (setup_multipath(vecs, mpp)) {
 804                                         free_pathvec(rpvec, KEEP_PATHS);
 805                                         goto out;
 806                                 }
 807
 808                                 condlog(3, "%s path removed from devmap %s",
 809                                         devname, mpp->alias);
 810                         }
 811                         free_pathvec(rpvec, KEEP_PATHS);
 812                 }
 813                 else {
 814                         char alias[WWID_SIZE];
 815
 816                         /*
 817                          * flush_map will fail if the device is open
 818                          */
 819                         strncpy(alias, mpp->alias, WWID_SIZE);
 820                         if (flush_map(mpp, vecs))
 821                                 rm_path = 0;
 822                         else
 823                                 condlog(3, "%s: removed map", alias);
 824                 }
 825         }
 826
 827         if (rm_path) {
 828                 if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 829                         vector_del_slot(vecs->pathvec, i);
 830                 free_path(pp);
 831         }
 832
 833         return 0;
 834
 835 out:
 836         remove_map(mpp, vecs, stop_waiter_thread, 1);
 837         return 1;
 838 }
 839
 840 int
 841 show_paths (char ** r, int * len, struct vectors * vecs)
 842 {
 843         int i;
 844         struct path * pp;
 845         char * c;
 846         char * reply;
 847         int maxlen = INITIAL_REPLY_LEN;
 848         int again = 1;
 849
 850         get_path_layout(vecs->pathvec);
 851         reply = MALLOC(maxlen);
 852
 853         while (again) {
 854                 if (!reply)
 855                         return 1;
 856
 857                 c = reply;
 858
 859                 if (VECTOR_SIZE(vecs->pathvec) > 0)
 860                         c += snprint_path_header(c, reply + maxlen - c,
 861                                                  PRINT_PATH_CHECKER);
 862
 863                 vector_foreach_slot(vecs->pathvec, pp, i)
 864                         c += snprint_path(c, reply + maxlen - c,
 865                                           PRINT_PATH_CHECKER, pp);
 866
 867                 again = ((c - reply) == (maxlen - 1));
 868
 869                 if (again)
 870                         reply = REALLOC(reply, maxlen *= 2);
 871
 872         }
 873         *r = reply;
 874         *len = (int)(c - reply + 1);
 875         return 0;
 876 }
 877
 878 int
 879 show_maps (char ** r, int *len, struct vectors * vecs)
 880 {
 881         int i;
 882         struct multipath * mpp;
 883         char * c;
 884         char * reply;
 885         int maxlen = INITIAL_REPLY_LEN;
 886         int again = 1;
 887
 888         get_map_layout(vecs->mpvec);
 889         reply = MALLOC(maxlen);
 890
 891         while (again) {
 892                 if (!reply)
 893                         return 1;
 894
 895                 c = reply;
 896                 if (VECTOR_SIZE(vecs->mpvec) > 0)
 897                         c += snprint_map_header(c, reply + maxlen - c,
 898                                                 PRINT_MAP_FAILBACK);
 899
 900                 vector_foreach_slot(vecs->mpvec, mpp, i)
 901                         c += snprint_map(c, reply + maxlen - c,
 902                                          PRINT_MAP_FAILBACK, mpp);
 903
 904                 again = ((c - reply) == (maxlen - 1));
 905
 906                 if (again)
 907                         reply = REALLOC(reply, maxlen *= 2);
 908         }
 909         *r = reply;
 910         *len = (int)(c - reply + 1);
 911         return 0;
 912 }
 913
 914 int
 915 dump_pathvec (char ** r, int * len, struct vectors * vecs)
 916 {
 917         int i;
 918         struct path * pp;
 919         char * reply;
 920         char * p;
 921
 922         *len = VECTOR_SIZE(vecs->pathvec) * sizeof(struct path);
 923         reply = (char *)MALLOC(*len);
 924         *r = reply;
 925
 926         if (!reply)
 927                 return 1;
 928
 929         p = reply;
 930
 931         vector_foreach_slot (vecs->pathvec, pp, i) {
 932                 memcpy((void *)p, pp, sizeof(struct path));
 933                 p += sizeof(struct path);
 934         }
 935
 936         /* return negative to hint caller not to add "ok" to the dump */
 937         return -1;
 938 }
 939
 940 static int
 941 map_discovery (struct vectors * vecs)
 942 {
 943         int i;
 944         struct multipath * mpp;
 945
 946         if (dm_get_maps(vecs->mpvec, "multipath"))
 947                 return 1;
 948
 949         vector_foreach_slot (vecs->mpvec, mpp, i)
 950                 if (setup_multipath(vecs, mpp))
 951                         return 1;
 952
 953         return 0;
 954 }
 955
 956 int
 957 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 958 {
 959         struct vectors * vecs;
 960         int r;
 961
 962         *reply = NULL;
 963         *len = 0;
 964         vecs = (struct vectors *)trigger_data;
 965
 966         pthread_cleanup_push(cleanup_lock, vecs->lock);
 967         lock(vecs->lock);
 968
 969         r = parse_cmd(str, reply, len, vecs);
 970
 971         if (r > 0) {
 972                 *reply = STRDUP("fail\n");
 973                 *len = strlen(*reply) + 1;
 974                 r = 1;
 975         }
 976         else if (!r && *len == 0) {
 977                 *reply = STRDUP("ok\n");
 978                 *len = strlen(*reply) + 1;
 979                 r = 0;
 980         }
 981         /* else if (r < 0) leave *reply alone */
 982
 983         lock_cleanup_pop(vecs->lock);
 984         return r;
 985 }
 986
 987 static int
 988 uev_discard(char * devpath)
 989 {
 990         char a[10], b[10];
 991
 992         /*
 993          * keep only block devices, discard partitions
 994          */
 995         if (sscanf(devpath, "/block/%10s", a) != 1 ||
 996             sscanf(devpath, "/block/%10[^/]/%10s", a, b) == 2) {
 997                 condlog(4, "discard event on %s", devpath);
 998                 return 1;
 999         }
1000         return 0;
1001 }
1002
1003 int
1004 uev_trigger (struct uevent * uev, void * trigger_data)
1005 {
1006         int r = 0;
1007         char devname[32];
1008         struct vectors * vecs;
1009
1010         vecs = (struct vectors *)trigger_data;
1011
1012         if (uev_discard(uev->devpath))
1013                 return 0;
1014
1015         basename(uev->devpath, devname);
1016         lock(vecs->lock);
1017
1018         /*
1019          * device map add/remove event
1020          */
1021         if (!strncmp(devname, "dm-", 3)) {
1022                 if (!strncmp(uev->action, "add", 3)) {
1023                         r = uev_add_map(devname, vecs);
1024                         goto out;
1025                 }
1026                 if (!strncmp(uev->action, "remove", 6)) {
1027                         r = uev_remove_map(devname, vecs);
1028                         goto out;
1029                 }
1030                 if (!strncmp(uev->action, "umount", 6)) {
1031                         r = uev_umount_map(devname, vecs);
1032                         goto out;
1033                 }
1034                 goto out;
1035         }
1036
1037         /*
1038          * path add/remove event
1039          */
1040         if (blacklist(conf->blist, devname))
1041                 goto out;
1042
1043         if (!strncmp(uev->action, "add", 3)) {
1044                 r = uev_add_path(devname, vecs);
1045                 goto out;
1046         }
1047         if (!strncmp(uev->action, "remove", 6)) {
1048                 r = uev_remove_path(devname, vecs);
1049                 goto out;
1050         }
1051
1052 out:
1053         unlock(vecs->lock);
1054         return r;
1055 }
1056
1057 static void *
1058 ueventloop (void * ap)
1059 {
1060         if (uevent_listen(&uev_trigger, ap))
1061                 fprintf(stderr, "error starting uevent listener");
1062
1063         return NULL;
1064 }
1065
1066 static void *
1067 uxlsnrloop (void * ap)
1068 {
1069         if (load_keys())
1070                 return NULL;
1071
1072         if (alloc_handlers())
1073                 return NULL;
1074
1075         add_handler(LIST+PATHS, cli_list_paths);
1076         add_handler(LIST+MAPS, cli_list_maps);
1077         add_handler(ADD+PATH, cli_add_path);
1078         add_handler(DEL+PATH, cli_del_path);
1079         add_handler(ADD+MAP, cli_add_map);
1080         add_handler(DEL+MAP, cli_del_map);
1081         add_handler(SWITCH+MAP+GROUP, cli_switch_group);
1082         add_handler(DUMP+PATHVEC, cli_dump_pathvec);
1083         add_handler(RECONFIGURE, cli_reconfigure);
1084         add_handler(SUSPEND+MAP, cli_suspend);
1085         add_handler(RESUME+MAP, cli_resume);
1086         add_handler(REINSTATE+PATH, cli_reinstate);
1087         add_handler(FAIL+PATH, cli_fail);
1088
1089         uxsock_listen(&uxsock_trigger, ap);
1090
1091         return NULL;
1092 }
1093
1094 static int
1095 exit_daemon (int status)
1096 {
1097         if (status != 0)
1098                 fprintf(stderr, "bad exit status. see daemon.log\n");
1099
1100         condlog(3, "unlink pidfile");
1101         unlink(DEFAULT_PIDFILE);
1102
1103         lock(&exit_mutex);
1104         pthread_cond_signal(&exit_cond);
1105         unlock(&exit_mutex);
1106
1107         return status;
1108 }
1109
1110 static void
1111 fail_path (struct path * pp)
1112 {
1113         if (!pp->mpp)
1114                 return;
1115
1116         condlog(2, "checker failed path %s in map %s",
1117                  pp->dev_t, pp->mpp->alias);
1118
1119         dm_fail_path(pp->mpp->alias, pp->dev_t);
1120         update_queue_mode_del_path(pp->mpp);
1121 }
1122
1123 /*
1124  * caller must have locked the path list before calling that function
1125  */
1126 static void
1127 reinstate_path (struct path * pp)
1128 {
1129         if (!pp->mpp)
1130                 return;
1131
1132         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
1133                 condlog(0, "%s: reinstate failed", pp->dev_t);
1134         else {
1135                 condlog(2, "%s: reinstated", pp->dev_t);
1136                 update_queue_mode_add_path(pp->mpp);
1137         }
1138 }
1139
1140 static void
1141 enable_group(struct path * pp)
1142 {
1143         struct pathgroup * pgp;
1144
1145         /*
1146          * if path is added through uev_add_path, pgindex can be unset.
1147          * next update_strings() will set it, upon map reload event.
1148          *
1149          * we can safely return here, because upon map reload, all
1150          * PG will be enabled.
1151          */
1152         if (!pp->mpp->pg || !pp->pgindex)
1153                 return;
1154
1155         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
1156
1157         if (pgp->status == PGSTATE_DISABLED) {
1158                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
1159                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
1160         }
1161 }
1162
1163 static void
1164 mpvec_garbage_collector (struct vectors * vecs)
1165 {
1166         struct multipath * mpp;
1167         int i;
1168
1169         vector_foreach_slot (vecs->mpvec, mpp, i) {
1170                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
1171                         condlog(2, "%s: remove dead map", mpp->alias);
1172                         remove_map(mpp, vecs, stop_waiter_thread, 1);
1173                         i--;
1174                 }
1175         }
1176 }
1177
1178 static void
1179 defered_failback_tick (vector mpvec)
1180 {
1181         struct multipath * mpp;
1182         int i;
1183
1184         vector_foreach_slot (mpvec, mpp, i) {
1185                 /*
1186                  * defered failback getting sooner
1187                  */
1188                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
1189                         mpp->failback_tick--;
1190
1191                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
1192                                 switch_pathgroup(mpp);
1193                 }
1194         }
1195 }
1196
1197 static void
1198 retry_count_tick(vector mpvec)
1199 {
1200         struct multipath *mpp;
1201         int i;
1202
1203         vector_foreach_slot (mpvec, mpp, i) {
1204                 if (mpp->retry_tick) {
1205                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
1206                         if(--mpp->retry_tick == 0) {
1207                                 dm_queue_if_no_path(mpp->alias, 0);
1208                                 condlog(2, "%s: Disable queueing", mpp->alias);
1209                         }
1210                 }
1211         }
1212 }
1213
1214 static void *
1215 checkerloop (void *ap)
1216 {
1217         struct vectors *vecs;
1218         struct path *pp;
1219         int i, count = 0;
1220         int newstate;
1221         char checker_msg[MAX_CHECKER_MSG_SIZE];
1222
1223         mlockall(MCL_CURRENT | MCL_FUTURE);
1224
1225         memset(checker_msg, 0, MAX_CHECKER_MSG_SIZE);
1226         vecs = (struct vectors *)ap;
1227
1228         condlog(2, "path checkers start up");
1229
1230         /*
1231          * init the path check interval
1232          */
1233         vector_foreach_slot (vecs->pathvec, pp, i) {
1234                 pp->checkint = conf->checkint;
1235         }
1236
1237         while (1) {
1238                 pthread_cleanup_push(cleanup_lock, vecs->lock);
1239                 lock(vecs->lock);
1240                 condlog(4, "tick");
1241
1242                 vector_foreach_slot (vecs->pathvec, pp, i) {
1243                         if (!pp->mpp)
1244                                 continue;
1245
1246                         if (pp->tick && --pp->tick)
1247                                 continue; /* don't check this path yet */
1248
1249                         /*
1250                          * provision a next check soonest,
1251                          * in case we exit abnormaly from here
1252                          */
1253                         pp->tick = conf->checkint;
1254
1255                         if (!pp->checkfn) {
1256                                 pathinfo(pp, conf->hwtable, DI_SYSFS);
1257                                 select_checkfn(pp);
1258                         }
1259
1260                         if (!pp->checkfn) {
1261                                 condlog(0, "%s: checkfn is void", pp->dev);
1262                                 continue;
1263                         }
1264                         newstate = pp->checkfn(pp->fd, checker_msg,
1265                                                &pp->checker_context);
1266
1267                         if (newstate < 0) {
1268                                 condlog(2, "%s: unusable path", pp->dev);
1269                                 pathinfo(pp, conf->hwtable, 0);
1270                                 continue;
1271                         }
1272
1273                         if (newstate != pp->state) {
1274                                 pp->state = newstate;
1275                                 LOG_MSG(1, checker_msg);
1276
1277                                 /*
1278                                  * upon state change, reset the checkint
1279                                  * to the shortest delay
1280                                  */
1281                                 pp->checkint = conf->checkint;
1282
1283                                 if (newstate == PATH_DOWN ||
1284                                     newstate == PATH_SHAKY ||
1285                                     update_multipath_strings(pp->mpp,
1286                                                              vecs->pathvec)) {
1287                                         /*
1288                                          * proactively fail path in the DM
1289                                          */
1290                                         fail_path(pp);
1291
1292                                         /*
1293                                          * cancel scheduled failback
1294                                          */
1295                                         pp->mpp->failback_tick = 0;
1296
1297                                         continue;
1298                                 }
1299
1300                                 /*
1301                                  * reinstate this path
1302                                  */
1303                                 reinstate_path(pp);
1304
1305                                 /*
1306                                  * schedule [defered] failback
1307                                  */
1308                                 if (pp->mpp->pgfailback > 0)
1309                                         pp->mpp->failback_tick =
1310                                                 pp->mpp->pgfailback + 1;
1311                                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE &&
1312                                     need_switch_pathgroup(pp->mpp, 1))
1313                                         switch_pathgroup(pp->mpp);
1314
1315                                 /*
1316                                  * if at least one path is up in a group, and
1317                                  * the group is disabled, re-enable it
1318                                  */
1319                                 if (newstate == PATH_UP)
1320                                         enable_group(pp);
1321                         }
1322                         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1323                                 LOG_MSG(4, checker_msg);
1324                                 /*
1325                                  * double the next check delay.
1326                                  * max at conf->max_checkint
1327                                  */
1328                                 if (pp->checkint < (conf->max_checkint / 2))
1329                                         pp->checkint = 2 * pp->checkint;
1330                                 else
1331                                         pp->checkint = conf->max_checkint;
1332
1333                                 pp->tick = pp->checkint;
1334                                 condlog(4, "%s: delay next check %is",
1335                                                 pp->dev_t, pp->tick);
1336
1337                         }
1338                         pp->state = newstate;
1339
1340                         /*
1341                          * path prio refreshing
1342                          */
1343                         condlog(4, "path prio refresh");
1344                         pathinfo(pp, conf->hwtable, DI_PRIO);
1345
1346                         if (need_switch_pathgroup(pp->mpp, 0)) {
1347                                 if (pp->mpp->pgfailback > 0)
1348                                         pp->mpp->failback_tick =
1349                                                 pp->mpp->pgfailback + 1;
1350                                 else if (pp->mpp->pgfailback ==
1351                                                 -FAILBACK_IMMEDIATE)
1352                                         switch_pathgroup(pp->mpp);
1353                         }
1354                 }
1355                 defered_failback_tick(vecs->mpvec);
1356                 retry_count_tick(vecs->mpvec);
1357
1358                 if (count)
1359                         count--;
1360                 else {
1361                         condlog(4, "map garbage collection");
1362                         mpvec_garbage_collector(vecs);
1363                         count = MAPGCINT;
1364                 }
1365
1366                 lock_cleanup_pop(vecs->lock);
1367                 sleep(1);
1368         }
1369         return NULL;
1370 }
1371
1372 int
1373 configure (struct vectors * vecs, int start_waiters)
1374 {
1375         struct multipath * mpp;
1376         struct path * pp;
1377         vector mpvec;
1378         int i;
1379
1380         if (!(vecs->pathvec = vector_alloc()))
1381                 return 1;
1382
1383         if (!(vecs->mpvec = vector_alloc()))
1384                 return 1;
1385
1386         if (!(mpvec = vector_alloc()))
1387                 return 1;
1388
1389         /*
1390          * probe for current path (from sysfs) and map (from dm) sets
1391          */
1392         path_discovery(vecs->pathvec, conf, DI_ALL);
1393
1394         vector_foreach_slot (vecs->pathvec, pp, i)
1395                 pp->checkint = conf->checkint;
1396
1397         if (map_discovery(vecs))
1398                 return 1;
1399
1400         /*
1401          * create new set of maps & push changed ones into dm
1402          */
1403         if (coalesce_paths(vecs, mpvec, NULL))
1404                 return 1;
1405
1406         /*
1407          * may need to remove some maps which are no longer relevant
1408          * e.g., due to blacklist changes in conf file
1409          */
1410         if (coalesce_maps(vecs, mpvec))
1411                 return 1;
1412
1413         dm_lib_release();
1414
1415         if (conf->verbosity > 2)
1416                 vector_foreach_slot(mpvec, mpp, i)
1417                         print_map(mpp);
1418
1419         /*
1420          * purge dm of old maps
1421          */
1422         remove_maps(vecs, NULL);
1423
1424         /*
1425          * save new set of maps formed by considering current path state
1426          */
1427         vecs->mpvec = mpvec;
1428
1429         /*
1430          * start dm event waiter threads for these new maps
1431          */
1432         vector_foreach_slot(vecs->mpvec, mpp, i) {
1433                 if (setup_multipath(vecs, mpp))
1434                         return 1;
1435                 if (start_waiters)
1436                         if (start_waiter_thread(mpp, vecs))
1437                                 return 1;
1438         }
1439         return 0;
1440 }
1441
1442 int
1443 reconfigure (struct vectors * vecs)
1444 {
1445         struct config * old = conf;
1446
1447         condlog(0, "reconfigure");
1448
1449         /*
1450          * free old map and path vectors ... they use old conf state
1451          */
1452         if (VECTOR_SIZE(vecs->mpvec))
1453                 remove_maps(vecs, stop_waiter_thread);
1454
1455         if (VECTOR_SIZE(vecs->pathvec))
1456                 free_pathvec(vecs->pathvec, FREE_PATHS);
1457
1458         conf = NULL;
1459
1460         if (load_config(DEFAULT_CONFIGFILE))
1461                 return 1;
1462
1463         conf->verbosity = old->verbosity;
1464
1465         if (!conf->checkint) {
1466                 conf->checkint = CHECKINT;
1467                 conf->max_checkint = MAX_CHECKINT;
1468         }
1469         configure(vecs, 1);
1470         free_config(old);
1471         return 0;
1472 }
1473
1474 static struct vectors *
1475 init_vecs (void)
1476 {
1477         struct vectors * vecs;
1478
1479         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1480
1481         if (!vecs)
1482                 return NULL;
1483
1484         vecs->lock =
1485                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1486
1487         if (!vecs->lock)
1488                 goto out;
1489
1490         vecs->pathvec = vector_alloc();
1491
1492         if (!vecs->pathvec)
1493                 goto out1;
1494
1495         vecs->mpvec = vector_alloc();
1496
1497         if (!vecs->mpvec)
1498                 goto out2;
1499
1500         pthread_mutex_init(vecs->lock, NULL);
1501
1502         return vecs;
1503
1504 out2:
1505         vector_free(vecs->pathvec);
1506 out1:
1507         FREE(vecs->lock);
1508 out:
1509         FREE(vecs);
1510         condlog(0, "failed to init paths");
1511         return NULL;
1512 }
1513
1514 static void *
1515 signal_set(int signo, void (*func) (int))
1516 {
1517         int r;
1518         struct sigaction sig;
1519         struct sigaction osig;
1520
1521         sig.sa_handler = func;
1522         sigemptyset(&sig.sa_mask);
1523         sig.sa_flags = 0;
1524
1525         r = sigaction(signo, &sig, &osig);
1526
1527         if (r < 0)
1528                 return (SIG_ERR);
1529         else
1530                 return (osig.sa_handler);
1531 }
1532
1533 static void
1534 sighup (int sig)
1535 {
1536         condlog(3, "SIGHUP received");
1537
1538         lock(gvecs->lock);
1539         reconfigure(gvecs);
1540         unlock(gvecs->lock);
1541
1542 #ifdef _DEBUG_
1543         dbg_free_final(NULL);
1544 #endif
1545 }
1546
1547 static void
1548 sigend (int sig)
1549 {
1550         exit_daemon(0);
1551 }
1552
1553 static void
1554 sigusr1 (int sig)
1555 {
1556         condlog(3, "SIGUSR1 received");
1557 }
1558
1559 static void
1560 signal_init(void)
1561 {
1562         signal_set(SIGHUP, sighup);
1563         signal_set(SIGUSR1, sigusr1);
1564         signal_set(SIGINT, sigend);
1565         signal_set(SIGTERM, sigend);
1566         signal_set(SIGKILL, sigend);
1567 }
1568
1569 static void
1570 setscheduler (void)
1571 {
1572         int res;
1573         static struct sched_param sched_param = {
1574                 sched_priority: 99
1575         };
1576
1577         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1578
1579         if (res == -1)
1580                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1581         return;
1582 }
1583
1584 static void
1585 set_oom_adj (int val)
1586 {
1587         FILE *fp;
1588
1589         fp = fopen("/proc/self/oom_adj", "w");
1590
1591         if (!fp)
1592                 return;
1593
1594         fprintf(fp, "%i", val);
1595         fclose(fp);
1596 }
1597
1598 static int
1599 child (void * param)
1600 {
1601         pthread_t check_thr, uevent_thr, uxlsnr_thr;
1602         pthread_attr_t attr;
1603         struct vectors * vecs;
1604
1605         mlockall(MCL_CURRENT | MCL_FUTURE);
1606
1607         if (logsink)
1608                 log_thread_start();
1609
1610         condlog(2, "--------start up--------");
1611         condlog(2, "read " DEFAULT_CONFIGFILE);
1612
1613         if (load_config(DEFAULT_CONFIGFILE))
1614                 exit(1);
1615
1616         setlogmask(LOG_UPTO(conf->verbosity + 3));
1617
1618         /*
1619          * fill the voids left in the config file
1620          */
1621         if (!conf->checkint) {
1622                 conf->checkint = CHECKINT;
1623                 conf->max_checkint = MAX_CHECKINT;
1624         }
1625
1626         if (pidfile_create(DEFAULT_PIDFILE, getpid())) {
1627                 if (logsink)
1628                         log_thread_stop();
1629
1630                 exit(1);
1631         }
1632         signal_init();
1633         setscheduler();
1634         set_oom_adj(-17);
1635         vecs = gvecs = init_vecs();
1636
1637         if (!vecs)
1638                 exit(1);
1639
1640         if (sysfs_get_mnt_path(sysfs_path, FILE_NAME_SIZE)) {
1641                 condlog(0, "can not find sysfs mount point");
1642                 exit(1);
1643         }
1644
1645         /*
1646          * fetch and configure both paths and multipaths
1647          */
1648         if (configure(vecs, 1)) {
1649                 condlog(0, "failure during configuration");
1650                 exit(1);
1651         }
1652
1653         /*
1654          * start threads
1655          */
1656         pthread_attr_init(&attr);
1657         pthread_attr_setstacksize(&attr, 64 * 1024);
1658         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1659
1660         pthread_create(&check_thr, &attr, checkerloop, vecs);
1661         pthread_create(&uevent_thr, &attr, ueventloop, vecs);
1662         pthread_create(&uxlsnr_thr, &attr, uxlsnrloop, vecs);
1663
1664         pthread_cond_wait(&exit_cond, &exit_mutex);
1665
1666         /*
1667          * exit path
1668          */
1669         lock(vecs->lock);
1670         remove_maps(vecs, stop_waiter_thread);
1671         free_pathvec(vecs->pathvec, FREE_PATHS);
1672
1673         pthread_cancel(check_thr);
1674         pthread_cancel(uevent_thr);
1675         pthread_cancel(uxlsnr_thr);
1676
1677         free_keys(keys);
1678         keys = NULL;
1679         free_handlers(handlers);
1680         handlers = NULL;
1681         free_polls();
1682
1683         unlock(vecs->lock);
1684         pthread_mutex_destroy(vecs->lock);
1685         FREE(vecs->lock);
1686         vecs->lock = NULL;
1687         FREE(vecs);
1688         vecs = NULL;
1689         free_config(conf);
1690         conf = NULL;
1691
1692         condlog(2, "--------shut down-------");
1693
1694         if (logsink)
1695                 log_thread_stop();
1696
1697         dm_lib_release();
1698         dm_lib_exit();
1699
1700 #ifdef _DEBUG_
1701         dbg_free_final(NULL);
1702 #endif
1703
1704         exit(0);
1705 }
1706
1707 static int
1708 daemonize(void)
1709 {
1710         int pid;
1711         int in_fd, out_fd;
1712
1713         if( (pid = fork()) < 0){
1714                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1715                 return -1;
1716         }
1717         else if (pid != 0)
1718                 return pid;
1719
1720         setsid();
1721
1722         if ( (pid = fork()) < 0)
1723                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1724         else if (pid != 0)
1725                 _exit(0);
1726
1727         in_fd = open("/dev/null", O_RDONLY);
1728         if (in_fd < 0){
1729                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1730                         strerror(errno));
1731                 _exit(0);
1732         }
1733         out_fd = open("/dev/console", O_WRONLY);
1734         if (out_fd < 0){
1735                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1736                         strerror(errno));
1737                 _exit(0);
1738         }
1739
1740         close(STDIN_FILENO);
1741         dup(in_fd);
1742         close(STDOUT_FILENO);
1743         dup(out_fd);
1744         close(STDERR_FILENO);
1745         dup(out_fd);
1746
1747         close(in_fd);
1748         close(out_fd);
1749         chdir("/");
1750         umask(0);
1751         return 0;
1752 }
1753
1754 int
1755 main (int argc, char *argv[])
1756 {
1757         extern char *optarg;
1758         extern int optind;
1759         int arg;
1760         int err;
1761
1762         logsink = 1;
1763
1764         if (getuid() != 0) {
1765                 fprintf(stderr, "need to be root\n");
1766                 exit(1);
1767         }
1768
1769         /* make sure we don't lock any path */
1770         chdir("/");
1771         umask(umask(077) | 022);
1772
1773         conf = alloc_config();
1774
1775         if (!conf)
1776                 exit(1);
1777
1778         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1779         switch(arg) {
1780                 case 'd':
1781                         logsink = 0;
1782                         //debug=1; /* ### comment me out ### */
1783                         break;
1784                 case 'v':
1785                         if (sizeof(optarg) > sizeof(char *) ||
1786                             !isdigit(optarg[0]))
1787                                 exit(1);
1788
1789                         conf->verbosity = atoi(optarg);
1790                         break;
1791                 case 'k':
1792                         uxclnt(optarg);
1793                         exit(0);
1794                 default:
1795                         ;
1796                 }
1797         }
1798
1799         if (!logsink)
1800                 err = 0;
1801         else
1802                 err = daemonize();
1803
1804         if (err < 0)
1805                 /* error */
1806                 exit(1);
1807         else if (err > 0)
1808                 /* parent dies */
1809                 exit(0);
1810         else
1811                 /* child lives */
1812                 return (child(NULL));
1813 }
1814