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