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 #include <sys/time.h>
  16 #include <sys/resource.h>
  17 #include <limits.h>
  18
  19 /*
  20  * libcheckers
  21  */
  22 #include <checkers.h>
  23
  24 /*
  25  * libmultipath
  26  */
  27 #include <parser.h>
  28 #include <vector.h>
  29 #include <memory.h>
  30 #include <config.h>
  31 #include <util.h>
  32 #include <hwtable.h>
  33 #include <defaults.h>
  34 #include <structs.h>
  35 #include <callout.h>
  36 #include <blacklist.h>
  37 #include <structs_vec.h>
  38 #include <dmparser.h>
  39 #include <devmapper.h>
  40 #include <sysfs.h>
  41 #include <dict.h>
  42 #include <discovery.h>
  43 #include <debug.h>
  44 #include <propsel.h>
  45 #include <uevent.h>
  46 #include <switchgroup.h>
  47 #include <print.h>
  48 #include <configure.h>
  49 #include <prio.h>
  50 #include <pgpolicies.h>
  51 #include <uevent.h>
  52
  53 #include "main.h"
  54 #include "pidfile.h"
  55 #include "uxlsnr.h"
  56 #include "uxclnt.h"
  57 #include "cli.h"
  58 #include "cli_handlers.h"
  59 #include "lock.h"
  60 #include "waiter.h"
  61
  62 #define FILE_NAME_SIZE 256
  63 #define CMDSIZE 160
  64
  65 #define LOG_MSG(a,b) \
  66         if (strlen(b)) condlog(a, "%s: %s - %s", pp->mpp->alias, pp->dev, b);
  67
  68 pthread_cond_t exit_cond = PTHREAD_COND_INITIALIZER;
  69 pthread_mutex_t exit_mutex = PTHREAD_MUTEX_INITIALIZER;
  70
  71 int logsink;
  72 enum daemon_status running_state;
  73
  74 /*
  75  * global copy of vecs for use in sig handlers
  76  */
  77 struct vectors * gvecs;
  78
  79 static int
  80 need_switch_pathgroup (struct multipath * mpp, int refresh)
  81 {
  82         struct pathgroup * pgp;
  83         struct path * pp;
  84         unsigned int i, j;
  85
  86         if (!mpp || mpp->pgfailback == -FAILBACK_MANUAL)
  87                 return 0;
  88
  89         /*
  90          * Refresh path priority values
  91          */
  92         if (refresh)
  93                 vector_foreach_slot (mpp->pg, pgp, i)
  94                         vector_foreach_slot (pgp->paths, pp, j)
  95                                 pathinfo(pp, conf->hwtable, DI_PRIO);
  96
  97         mpp->bestpg = select_path_group(mpp);
  98
  99         if (mpp->bestpg != mpp->nextpg)
 100                 return 1;
 101
 102         return 0;
 103 }
 104
 105 static void
 106 switch_pathgroup (struct multipath * mpp)
 107 {
 108         mpp->stat_switchgroup++;
 109         dm_switchgroup(mpp->alias, mpp->bestpg);
 110         condlog(2, "%s: switch to path group #%i",
 111                  mpp->alias, mpp->bestpg);
 112 }
 113
 114 static int
 115 coalesce_maps(struct vectors *vecs, vector nmpv)
 116 {
 117         struct multipath * ompp;
 118         vector ompv = vecs->mpvec;
 119         unsigned int i;
 120         int j;
 121
 122         vector_foreach_slot (ompv, ompp, i) {
 123                 if (!find_mp_by_wwid(nmpv, ompp->wwid)) {
 124                         /*
 125                          * remove all current maps not allowed by the
 126                          * current configuration
 127                          */
 128                         if (dm_flush_map(ompp->alias)) {
 129                                 condlog(0, "%s: unable to flush devmap",
 130                                         ompp->alias);
 131                                 /*
 132                                  * may be just because the device is open
 133                                  */
 134                                 if (!vector_alloc_slot(nmpv))
 135                                         return 1;
 136
 137                                 vector_set_slot(nmpv, ompp);
 138                                 setup_multipath(vecs, ompp);
 139
 140                                 if ((j = find_slot(ompv, (void *)ompp)) != -1)
 141                                         vector_del_slot(ompv, j);
 142
 143                                 continue;
 144                         }
 145                         else {
 146                                 dm_lib_release();
 147                                 condlog(2, "%s devmap removed", ompp->alias);
 148                         }
 149                 }
 150         }
 151         return 0;
 152 }
 153
 154 void
 155 sync_map_state(struct multipath *mpp)
 156 {
 157         struct pathgroup *pgp;
 158         struct path *pp;
 159         unsigned int i, j;
 160
 161         if (!mpp->pg)
 162                 return;
 163
 164         vector_foreach_slot (mpp->pg, pgp, i){
 165                 vector_foreach_slot (pgp->paths, pp, j){
 166                         if (pp->state == PATH_UNCHECKED ||
 167                             pp->state == PATH_WILD)
 168                                 continue;
 169                         if ((pp->dmstate == PSTATE_FAILED ||
 170                              pp->dmstate == PSTATE_UNDEF) &&
 171                             (pp->state == PATH_UP || pp->state == PATH_GHOST))
 172                                 dm_reinstate_path(mpp->alias, pp->dev_t);
 173                         else if ((pp->dmstate == PSTATE_ACTIVE ||
 174                                   pp->dmstate == PSTATE_UNDEF) &&
 175                                  (pp->state == PATH_DOWN ||
 176                                   pp->state == PATH_SHAKY))
 177                                 dm_fail_path(mpp->alias, pp->dev_t);
 178                 }
 179         }
 180 }
 181
 182 static void
 183 sync_maps_state(vector mpvec)
 184 {
 185         unsigned int i;
 186         struct multipath *mpp;
 187
 188         vector_foreach_slot (mpvec, mpp, i)
 189                 sync_map_state(mpp);
 190 }
 191
 192 static int
 193 flush_map(struct multipath * mpp, struct vectors * vecs)
 194 {
 195         /*
 196          * clear references to this map before flushing so we can ignore
 197          * the spurious uevent we may generate with the dm_flush_map call below
 198          */
 199         if (dm_flush_map(mpp->alias)) {
 200                 /*
 201                  * May not really be an error -- if the map was already flushed
 202                  * from the device mapper by dmsetup(8) for instance.
 203                  */
 204                 condlog(0, "%s: can't flush", mpp->alias);
 205                 return 1;
 206         }
 207         else {
 208                 dm_lib_release();
 209                 condlog(2, "%s: devmap removed", mpp->alias);
 210         }
 211
 212         orphan_paths(vecs->pathvec, mpp);
 213         remove_map_and_stop_waiter(mpp, vecs, 1);
 214
 215         return 0;
 216 }
 217
 218 static int
 219 uev_add_map (struct uevent * uev, struct vectors * vecs)
 220 {
 221         char *alias;
 222         int major = -1, minor = -1, rc;
 223
 224         condlog(2, "%s: add map (uevent)", uev->kernel);
 225         alias = uevent_get_dm_name(uev);
 226         if (!alias) {
 227                 condlog(3, "%s: No DM_NAME in uevent", uev->kernel);
 228                 major = uevent_get_major(uev);
 229                 minor = uevent_get_minor(uev);
 230                 alias = dm_mapname(major, minor);
 231                 if (!alias) {
 232                         condlog(2, "%s: mapname not found for %d:%d",
 233                                 uev->kernel, major, minor);
 234                         return 1;
 235                 }
 236         }
 237         rc = ev_add_map(uev->kernel, alias, vecs);
 238         FREE(alias);
 239         return rc;
 240 }
 241
 242 int
 243 ev_add_map (char * dev, char * alias, struct vectors * vecs)
 244 {
 245         char * refwwid;
 246         struct multipath * mpp;
 247         int map_present;
 248         int r = 1;
 249
 250         map_present = dm_map_present(alias);
 251
 252         if (map_present && dm_type(alias, TGT_MPATH) <= 0) {
 253                 condlog(4, "%s: not a multipath map", alias);
 254                 return 0;
 255         }
 256
 257         mpp = find_mp_by_alias(vecs->mpvec, alias);
 258
 259         if (mpp) {
 260                 /*
 261                  * Not really an error -- we generate our own uevent
 262                  * if we create a multipath mapped device as a result
 263                  * of uev_add_path
 264                  */
 265                 condlog(0, "%s: devmap already registered", dev);
 266                 return 0;
 267         }
 268
 269         /*
 270          * now we can register the map
 271          */
 272         if (map_present && (mpp = add_map_without_path(vecs, alias))) {
 273                 sync_map_state(mpp);
 274                 condlog(2, "%s: devmap %s registered", alias, dev);
 275                 return 0;
 276         }
 277         refwwid = get_refwwid(dev, DEV_DEVMAP, vecs->pathvec);
 278
 279         if (refwwid) {
 280                 r = coalesce_paths(vecs, NULL, refwwid, 0);
 281                 dm_lib_release();
 282         }
 283
 284         if (!r)
 285                 condlog(2, "%s: devmap %s added", alias, dev);
 286         else
 287                 condlog(0, "%s: uev_add_map %s failed", alias, dev);
 288
 289         FREE(refwwid);
 290         return r;
 291 }
 292
 293 static int
 294 uev_remove_map (struct uevent * uev, struct vectors * vecs)
 295 {
 296         char *alias;
 297         int minor, rc;
 298
 299         condlog(2, "%s: remove map (uevent)", uev->kernel);
 300         alias = uevent_get_dm_name(uev);
 301         if (!alias) {
 302                 condlog(3, "%s: No DM_NAME in uevent, ignoring", uev->kernel);
 303                 return 0;
 304         }
 305         minor = uevent_get_minor(uev);
 306         rc = ev_remove_map(uev->kernel, alias, minor, vecs);
 307         FREE(alias);
 308         return rc;
 309 }
 310
 311 int
 312 ev_remove_map (char * devname, char * alias, int minor, struct vectors * vecs)
 313 {
 314         struct multipath * mpp;
 315
 316         mpp = find_mp_by_minor(vecs->mpvec, minor);
 317
 318         if (!mpp) {
 319                 condlog(2, "%s: devmap not registered, can't remove",
 320                         devname);
 321                 return 0;
 322         }
 323         if (strcmp(mpp->alias, alias)) {
 324                 condlog(2, "%s: minor number mismatch (map %d, event %d)",
 325                         mpp->alias, mpp->dmi->minor, minor);
 326                 return 0;
 327         }
 328         return flush_map(mpp, vecs);
 329 }
 330
 331 static int
 332 uev_add_path (struct uevent *uev, struct vectors * vecs)
 333 {
 334         struct sysfs_device * dev;
 335
 336         dev = sysfs_device_get(uev->devpath);
 337         if (!dev) {
 338                 condlog(2, "%s: not found in sysfs", uev->devpath);
 339                 return 1;
 340         }
 341         condlog(2, "%s: add path (uevent)", dev->kernel);
 342         return (ev_add_path(dev->kernel, vecs) != 1)? 0 : 1;
 343 }
 344
 345 /*
 346  * returns:
 347  * 0: added
 348  * 1: error
 349  * 2: blacklisted
 350  */
 351 int
 352 ev_add_path (char * devname, struct vectors * vecs)
 353 {
 354         struct multipath * mpp;
 355         struct path * pp;
 356         char empty_buff[WWID_SIZE] = {0};
 357         char params[PARAMS_SIZE] = {0};
 358         int start_waiter = 0;
 359
 360         if (strstr(devname, "..") != NULL) {
 361                 /*
 362                  * Don't allow relative device names in the pathvec
 363                  */
 364                 condlog(0, "%s: path name is invalid", devname);
 365                 return 1;
 366         }
 367
 368         pp = find_path_by_dev(vecs->pathvec, devname);
 369
 370         if (pp) {
 371                 condlog(0, "%s: spurious uevent, path already in pathvec",
 372                         devname);
 373                 if (pp->mpp)
 374                         return 0;
 375         }
 376         else {
 377                 /*
 378                  * get path vital state
 379                  */
 380                 if (!(pp = store_pathinfo(vecs->pathvec, conf->hwtable,
 381                       devname, DI_ALL))) {
 382                         condlog(0, "%s: failed to store path info", devname);
 383                         return 1;
 384                 }
 385                 pp->checkint = conf->checkint;
 386         }
 387
 388         /*
 389          * need path UID to go any further
 390          */
 391         if (memcmp(empty_buff, pp->wwid, WWID_SIZE) == 0) {
 392                 condlog(0, "%s: failed to get path uid", devname);
 393                 goto fail; /* leave path added to pathvec */
 394         }
 395         if (filter_path(conf, pp) > 0){
 396                 int i = find_slot(vecs->pathvec, (void *)pp);
 397                 if (i != -1)
 398                         vector_del_slot(vecs->pathvec, i);
 399                 free_path(pp);
 400                 return 2;
 401         }
 402         mpp = pp->mpp = find_mp_by_wwid(vecs->mpvec, pp->wwid);
 403 rescan:
 404         if (mpp) {
 405                 if ((!pp->size) || (mpp->size != pp->size)) {
 406                         if (!pp->size)
 407                                 condlog(0, "%s: failed to add new path %s, "
 408                                         "device size is 0",
 409                                         devname, pp->dev);
 410                         else
 411                                 condlog(0, "%s: failed to add new path %s, "
 412                                         "device size mismatch",
 413                                         devname, pp->dev);
 414                         int i = find_slot(vecs->pathvec, (void *)pp);
 415                         if (i != -1)
 416                                 vector_del_slot(vecs->pathvec, i);
 417                         free_path(pp);
 418                         return 1;
 419                 }
 420
 421                 condlog(4,"%s: adopting all paths for path %s",
 422                         mpp->alias, pp->dev);
 423                 if (adopt_paths(vecs->pathvec, mpp, 1))
 424                         goto fail; /* leave path added to pathvec */
 425
 426                 verify_paths(mpp, vecs, NULL);
 427                 mpp->flush_on_last_del = FLUSH_UNDEF;
 428                 mpp->action = ACT_RELOAD;
 429         }
 430         else {
 431                 if (!pp->size) {
 432                         condlog(0, "%s: failed to create new map,"
 433                                 " %s device size is 0 ", devname, pp->dev);
 434                         int i = find_slot(vecs->pathvec, (void *)pp);
 435                         if (i != -1)
 436                                 vector_del_slot(vecs->pathvec, i);
 437                         free_path(pp);
 438                         return 1;
 439                 }
 440
 441                 condlog(4,"%s: creating new map", pp->dev);
 442                 if ((mpp = add_map_with_path(vecs, pp, 1))) {
 443                         mpp->action = ACT_CREATE;
 444                         /*
 445                          * We don't depend on ACT_CREATE, as domap will
 446                          * set it to ACT_NOTHING when complete.
 447                          */
 448                         start_waiter = 1;
 449                 }
 450                 else
 451                         goto fail; /* leave path added to pathvec */
 452         }
 453
 454         /*
 455          * push the map to the device-mapper
 456          */
 457         if (setup_map(mpp, params, PARAMS_SIZE)) {
 458                 condlog(0, "%s: failed to setup map for addition of new "
 459                         "path %s", mpp->alias, devname);
 460                 goto fail_map;
 461         }
 462         /*
 463          * reload the map for the multipath mapped device
 464          */
 465         if (domap(mpp, params) <= 0) {
 466                 condlog(0, "%s: failed in domap for addition of new "
 467                         "path %s", mpp->alias, devname);
 468                 /*
 469                  * deal with asynchronous uevents :((
 470                  */
 471                 if (mpp->action == ACT_RELOAD) {
 472                         condlog(0, "%s: uev_add_path sleep", mpp->alias);
 473                         sleep(1);
 474                         update_mpp_paths(mpp, vecs->pathvec);
 475                         goto rescan;
 476                 }
 477                 else
 478                         goto fail_map;
 479         }
 480         dm_lib_release();
 481
 482         /*
 483          * update our state from kernel regardless of create or reload
 484          */
 485         if (setup_multipath(vecs, mpp))
 486                 goto fail_map;
 487
 488         sync_map_state(mpp);
 489
 490         if ((mpp->action == ACT_CREATE ||
 491              (mpp->action == ACT_NOTHING && start_waiter && !mpp->waiter)) &&
 492             start_waiter_thread(mpp, vecs))
 493                         goto fail_map;
 494
 495         condlog(2, "%s path added to devmap %s", devname, mpp->alias);
 496         return 0;
 497
 498 fail_map:
 499         remove_map(mpp, vecs, 1);
 500 fail:
 501         orphan_path(pp);
 502         return 1;
 503 }
 504
 505 static int
 506 uev_remove_path (struct uevent *uev, struct vectors * vecs)
 507 {
 508         struct sysfs_device * dev;
 509         int retval;
 510
 511         dev = sysfs_device_get(uev->devpath);
 512         if (!dev) {
 513                 condlog(2, "%s: not found in sysfs", uev->devpath);
 514                 return 1;
 515         }
 516         condlog(2, "%s: remove path (uevent)", uev->kernel);
 517         retval = ev_remove_path(uev->kernel, vecs);
 518
 519         if (!retval)
 520                 sysfs_device_put(dev);
 521
 522         return retval;
 523 }
 524
 525 int
 526 ev_remove_path (char * devname, struct vectors * vecs)
 527 {
 528         struct multipath * mpp;
 529         struct path * pp;
 530         int i, retval = 0;
 531         char params[PARAMS_SIZE] = {0};
 532
 533         pp = find_path_by_dev(vecs->pathvec, devname);
 534
 535         if (!pp) {
 536                 /* Not an error; path might have been purged earlier */
 537                 condlog(0, "%s: path already removed", devname);
 538                 return 0;
 539         }
 540
 541         /*
 542          * avoid referring to the map of an orphaned path
 543          */
 544         if ((mpp = pp->mpp)) {
 545                 /*
 546                  * transform the mp->pg vector of vectors of paths
 547                  * into a mp->params string to feed the device-mapper
 548                  */
 549                 if (update_mpp_paths(mpp, vecs->pathvec)) {
 550                         condlog(0, "%s: failed to update paths",
 551                                 mpp->alias);
 552                         goto fail;
 553                 }
 554                 if ((i = find_slot(mpp->paths, (void *)pp)) != -1)
 555                         vector_del_slot(mpp->paths, i);
 556
 557                 /*
 558                  * remove the map IFF removing the last path
 559                  */
 560                 if (VECTOR_SIZE(mpp->paths) == 0) {
 561                         char alias[WWID_SIZE];
 562
 563                         /*
 564                          * flush_map will fail if the device is open
 565                          */
 566                         strncpy(alias, mpp->alias, WWID_SIZE);
 567                         if (mpp->flush_on_last_del == FLUSH_ENABLED) {
 568                                 condlog(2, "%s Last path deleted, disabling queueing", mpp->alias);
 569                                 mpp->retry_tick = 0;
 570                                 mpp->no_path_retry = NO_PATH_RETRY_FAIL;
 571                                 mpp->flush_on_last_del = FLUSH_IN_PROGRESS;
 572                                 dm_queue_if_no_path(mpp->alias, 0);
 573                         }
 574                         if (!flush_map(mpp, vecs)) {
 575                                 condlog(2, "%s: removed map after"
 576                                         " removing all paths",
 577                                         alias);
 578                                 retval = 0;
 579                                 goto out;
 580                         }
 581                         /*
 582                          * Not an error, continue
 583                          */
 584                 }
 585
 586                 if (setup_map(mpp, params, PARAMS_SIZE)) {
 587                         condlog(0, "%s: failed to setup map for"
 588                                 " removal of path %s", mpp->alias,
 589                                 devname);
 590                         goto fail;
 591                 }
 592                 /*
 593                  * reload the map
 594                  */
 595                 mpp->action = ACT_RELOAD;
 596                 if (domap(mpp, params) <= 0) {
 597                         condlog(0, "%s: failed in domap for "
 598                                 "removal of path %s",
 599                                 mpp->alias, devname);
 600                         retval = 1;
 601                 } else {
 602                         /*
 603                          * update our state from kernel
 604                          */
 605                         if (setup_multipath(vecs, mpp)) {
 606                                 goto fail;
 607                         }
 608                         sync_map_state(mpp);
 609
 610                         condlog(2, "%s: path removed from map %s",
 611                                 devname, mpp->alias);
 612                 }
 613         }
 614
 615 out:
 616         if ((i = find_slot(vecs->pathvec, (void *)pp)) != -1)
 617                 vector_del_slot(vecs->pathvec, i);
 618
 619         free_path(pp);
 620
 621         return retval;
 622
 623 fail:
 624         remove_map_and_stop_waiter(mpp, vecs, 1);
 625         return 1;
 626 }
 627
 628 static int
 629 map_discovery (struct vectors * vecs)
 630 {
 631         struct multipath * mpp;
 632         unsigned int i;
 633
 634         if (dm_get_maps(vecs->mpvec))
 635                 return 1;
 636
 637         vector_foreach_slot (vecs->mpvec, mpp, i)
 638                 if (setup_multipath(vecs, mpp))
 639                         return 1;
 640
 641         return 0;
 642 }
 643
 644 int
 645 uxsock_trigger (char * str, char ** reply, int * len, void * trigger_data)
 646 {
 647         struct vectors * vecs;
 648         int r;
 649
 650         *reply = NULL;
 651         *len = 0;
 652         vecs = (struct vectors *)trigger_data;
 653
 654         pthread_cleanup_push(cleanup_lock, &vecs->lock);
 655         lock(vecs->lock);
 656
 657         r = parse_cmd(str, reply, len, vecs);
 658
 659         if (r > 0) {
 660                 *reply = STRDUP("fail\n");
 661                 *len = strlen(*reply) + 1;
 662                 r = 1;
 663         }
 664         else if (!r && *len == 0) {
 665                 *reply = STRDUP("ok\n");
 666                 *len = strlen(*reply) + 1;
 667                 r = 0;
 668         }
 669         /* else if (r < 0) leave *reply alone */
 670
 671         lock_cleanup_pop(vecs->lock);
 672         return r;
 673 }
 674
 675 static int
 676 uev_discard(char * devpath)
 677 {
 678         char *tmp;
 679         char a[11], b[11];
 680
 681         /*
 682          * keep only block devices, discard partitions
 683          */
 684         tmp = strstr(devpath, "/block/");
 685         if (tmp == NULL){
 686                 condlog(4, "no /block/ in '%s'", devpath);
 687                 return 1;
 688         }
 689         if (sscanf(tmp, "/block/%10s", a) != 1 ||
 690             sscanf(tmp, "/block/%10[^/]/%10s", a, b) == 2) {
 691                 condlog(4, "discard event on %s", devpath);
 692                 return 1;
 693         }
 694         return 0;
 695 }
 696
 697 int
 698 uev_trigger (struct uevent * uev, void * trigger_data)
 699 {
 700         int r = 0;
 701         struct vectors * vecs;
 702
 703         vecs = (struct vectors *)trigger_data;
 704
 705         if (uev_discard(uev->devpath))
 706                 return 0;
 707
 708         lock(vecs->lock);
 709
 710         /*
 711          * device map event
 712          * Add events are ignored here as the tables
 713          * are not fully initialised then.
 714          */
 715         if (!strncmp(uev->kernel, "dm-", 3)) {
 716                 if (!strncmp(uev->action, "change", 6)) {
 717                         r = uev_add_map(uev, vecs);
 718                         goto out;
 719                 }
 720                 if (!strncmp(uev->action, "remove", 6)) {
 721                         r = uev_remove_map(uev, vecs);
 722                         goto out;
 723                 }
 724                 goto out;
 725         }
 726
 727         /*
 728          * path add/remove event
 729          */
 730         if (filter_devnode(conf->blist_devnode, conf->elist_devnode,
 731                            uev->kernel) > 0)
 732                 goto out;
 733
 734         if (!strncmp(uev->action, "add", 3)) {
 735                 r = uev_add_path(uev, vecs);
 736                 goto out;
 737         }
 738         if (!strncmp(uev->action, "remove", 6)) {
 739                 r = uev_remove_path(uev, vecs);
 740                 goto out;
 741         }
 742
 743 out:
 744         unlock(vecs->lock);
 745         return r;
 746 }
 747
 748 static void *
 749 ueventloop (void * ap)
 750 {
 751         block_signal(SIGUSR1, NULL);
 752         block_signal(SIGHUP, NULL);
 753
 754         if (uevent_listen())
 755                 condlog(0, "error starting uevent listener");
 756
 757         return NULL;
 758 }
 759
 760 static void *
 761 uevqloop (void * ap)
 762 {
 763         block_signal(SIGUSR1, NULL);
 764         block_signal(SIGHUP, NULL);
 765
 766         if (uevent_dispatch(&uev_trigger, ap))
 767                 condlog(0, "error starting uevent dispatcher");
 768
 769         return NULL;
 770 }
 771 static void *
 772 uxlsnrloop (void * ap)
 773 {
 774         block_signal(SIGUSR1, NULL);
 775         block_signal(SIGHUP, NULL);
 776
 777         if (cli_init())
 778                 return NULL;
 779
 780         set_handler_callback(LIST+PATHS, cli_list_paths);
 781         set_handler_callback(LIST+PATHS+FMT, cli_list_paths_fmt);
 782         set_handler_callback(LIST+MAPS, cli_list_maps);
 783         set_handler_callback(LIST+STATUS, cli_list_status);
 784         set_handler_callback(LIST+DAEMON, cli_list_daemon);
 785         set_handler_callback(LIST+MAPS+STATUS, cli_list_maps_status);
 786         set_handler_callback(LIST+MAPS+STATS, cli_list_maps_stats);
 787         set_handler_callback(LIST+MAPS+FMT, cli_list_maps_fmt);
 788         set_handler_callback(LIST+MAPS+TOPOLOGY, cli_list_maps_topology);
 789         set_handler_callback(LIST+TOPOLOGY, cli_list_maps_topology);
 790         set_handler_callback(LIST+MAP+TOPOLOGY, cli_list_map_topology);
 791         set_handler_callback(LIST+CONFIG, cli_list_config);
 792         set_handler_callback(LIST+BLACKLIST, cli_list_blacklist);
 793         set_handler_callback(LIST+DEVICES, cli_list_devices);
 794         set_handler_callback(LIST+WILDCARDS, cli_list_wildcards);
 795         set_handler_callback(ADD+PATH, cli_add_path);
 796         set_handler_callback(DEL+PATH, cli_del_path);
 797         set_handler_callback(ADD+MAP, cli_add_map);
 798         set_handler_callback(DEL+MAP, cli_del_map);
 799         set_handler_callback(SWITCH+MAP+GROUP, cli_switch_group);
 800         set_handler_callback(RECONFIGURE, cli_reconfigure);
 801         set_handler_callback(SUSPEND+MAP, cli_suspend);
 802         set_handler_callback(RESUME+MAP, cli_resume);
 803         set_handler_callback(RESIZE+MAP, cli_resize);
 804         set_handler_callback(REINSTATE+PATH, cli_reinstate);
 805         set_handler_callback(FAIL+PATH, cli_fail);
 806         set_handler_callback(DISABLEQ+MAP, cli_disable_queueing);
 807         set_handler_callback(RESTOREQ+MAP, cli_restore_queueing);
 808         set_handler_callback(DISABLEQ+MAPS, cli_disable_all_queueing);
 809         set_handler_callback(RESTOREQ+MAPS, cli_restore_all_queueing);
 810         set_handler_callback(QUIT, cli_quit);
 811         set_handler_callback(SHUTDOWN, cli_shutdown);
 812
 813         umask(077);
 814         uxsock_listen(&uxsock_trigger, ap);
 815
 816         return NULL;
 817 }
 818
 819 int
 820 exit_daemon (int status)
 821 {
 822         if (status != 0)
 823                 fprintf(stderr, "bad exit status. see daemon.log\n");
 824
 825         condlog(3, "unlink pidfile");
 826         unlink(DEFAULT_PIDFILE);
 827
 828         pthread_mutex_lock(&exit_mutex);
 829         pthread_cond_signal(&exit_cond);
 830         pthread_mutex_unlock(&exit_mutex);
 831
 832         return status;
 833 }
 834
 835 const char *
 836 daemon_status(void)
 837 {
 838         switch (running_state) {
 839         case DAEMON_INIT:
 840                 return "init";
 841         case DAEMON_START:
 842                 return "startup";
 843         case DAEMON_CONFIGURE:
 844                 return "configure";
 845         case DAEMON_RUNNING:
 846                 return "running";
 847         case DAEMON_SHUTDOWN:
 848                 return "shutdown";
 849         }
 850         return NULL;
 851 }
 852
 853 static void
 854 fail_path (struct path * pp, int del_active)
 855 {
 856         if (!pp->mpp)
 857                 return;
 858
 859         condlog(2, "checker failed path %s in map %s",
 860                  pp->dev_t, pp->mpp->alias);
 861
 862         dm_fail_path(pp->mpp->alias, pp->dev_t);
 863         if (del_active)
 864                 update_queue_mode_del_path(pp->mpp);
 865 }
 866
 867 /*
 868  * caller must have locked the path list before calling that function
 869  */
 870 static void
 871 reinstate_path (struct path * pp, int add_active)
 872 {
 873         if (!pp->mpp)
 874                 return;
 875
 876         if (dm_reinstate_path(pp->mpp->alias, pp->dev_t))
 877                 condlog(0, "%s: reinstate failed", pp->dev_t);
 878         else {
 879                 condlog(2, "%s: reinstated", pp->dev_t);
 880                 if (add_active)
 881                         update_queue_mode_add_path(pp->mpp);
 882         }
 883 }
 884
 885 static void
 886 enable_group(struct path * pp)
 887 {
 888         struct pathgroup * pgp;
 889
 890         /*
 891          * if path is added through uev_add_path, pgindex can be unset.
 892          * next update_strings() will set it, upon map reload event.
 893          *
 894          * we can safely return here, because upon map reload, all
 895          * PG will be enabled.
 896          */
 897         if (!pp->mpp->pg || !pp->pgindex)
 898                 return;
 899
 900         pgp = VECTOR_SLOT(pp->mpp->pg, pp->pgindex - 1);
 901
 902         if (pgp->status == PGSTATE_DISABLED) {
 903                 condlog(2, "%s: enable group #%i", pp->mpp->alias, pp->pgindex);
 904                 dm_enablegroup(pp->mpp->alias, pp->pgindex);
 905         }
 906 }
 907
 908 static void
 909 mpvec_garbage_collector (struct vectors * vecs)
 910 {
 911         struct multipath * mpp;
 912         unsigned int i;
 913
 914         if (!vecs->mpvec)
 915                 return;
 916
 917         vector_foreach_slot (vecs->mpvec, mpp, i) {
 918                 if (mpp && mpp->alias && !dm_map_present(mpp->alias)) {
 919                         condlog(2, "%s: remove dead map", mpp->alias);
 920                         remove_map_and_stop_waiter(mpp, vecs, 1);
 921                         i--;
 922                 }
 923         }
 924 }
 925
 926 static void
 927 defered_failback_tick (vector mpvec)
 928 {
 929         struct multipath * mpp;
 930         unsigned int i;
 931
 932         vector_foreach_slot (mpvec, mpp, i) {
 933                 /*
 934                  * defered failback getting sooner
 935                  */
 936                 if (mpp->pgfailback > 0 && mpp->failback_tick > 0) {
 937                         mpp->failback_tick--;
 938
 939                         if (!mpp->failback_tick && need_switch_pathgroup(mpp, 1))
 940                                 switch_pathgroup(mpp);
 941                 }
 942         }
 943 }
 944
 945 static void
 946 retry_count_tick(vector mpvec)
 947 {
 948         struct multipath *mpp;
 949         unsigned int i;
 950
 951         vector_foreach_slot (mpvec, mpp, i) {
 952                 if (mpp->retry_tick) {
 953                         mpp->stat_total_queueing_time++;
 954                         condlog(4, "%s: Retrying.. No active path", mpp->alias);
 955                         if(--mpp->retry_tick == 0) {
 956                                 dm_queue_if_no_path(mpp->alias, 0);
 957                                 condlog(2, "%s: Disable queueing", mpp->alias);
 958                         }
 959                 }
 960         }
 961 }
 962
 963 int update_prio(struct path *pp, int refresh_all)
 964 {
 965         int oldpriority;
 966         struct path *pp1;
 967         struct pathgroup * pgp;
 968         int i, j, changed = 0;
 969
 970         if (refresh_all) {
 971                 vector_foreach_slot (pp->mpp->pg, pgp, i) {
 972                         vector_foreach_slot (pgp->paths, pp1, j) {
 973                                 oldpriority = pp1->priority;
 974                                 pathinfo(pp1, conf->hwtable, DI_PRIO);
 975                                 if (pp1->priority != oldpriority)
 976                                         changed = 1;
 977                         }
 978                 }
 979                 return changed;
 980         }
 981         oldpriority = pp->priority;
 982         pathinfo(pp, conf->hwtable, DI_PRIO);
 983
 984         if (pp->priority == oldpriority)
 985                 return 0;
 986         return 1;
 987 }
 988
 989 int update_path_groups(struct multipath *mpp, struct vectors *vecs, int refresh)
 990 {
 991         int i;
 992         struct path * pp;
 993         char params[PARAMS_SIZE];
 994
 995         update_mpp_paths(mpp, vecs->pathvec);
 996         if (refresh) {
 997                 vector_foreach_slot (mpp->paths, pp, i)
 998                         pathinfo(pp, conf->hwtable, DI_PRIO);
 999         }
1000         params[0] = '\0';
1001         if (setup_map(mpp, params, PARAMS_SIZE))
1002                 return 1;
1003
1004         mpp->action = ACT_RELOAD;
1005         if (domap(mpp, params) <= 0) {
1006                 condlog(0, "%s: failed to update map : %s", mpp->alias,
1007                         strerror(errno));
1008                 return 1;
1009         }
1010         dm_lib_release();
1011         setup_multipath(vecs, mpp);
1012         sync_map_state(mpp);
1013
1014         return 0;
1015 }
1016
1017 void
1018 check_path (struct vectors * vecs, struct path * pp)
1019 {
1020         int newstate;
1021         int new_path_up = 0;
1022
1023         if (!pp->mpp)
1024                 return;
1025
1026         if (pp->tick && --pp->tick)
1027                 return; /* don't check this path yet */
1028
1029         /*
1030          * provision a next check soonest,
1031          * in case we exit abnormaly from here
1032          */
1033         pp->tick = conf->checkint;
1034
1035         newstate = path_offline(pp);
1036         if (newstate == PATH_UP)
1037                 newstate = get_state(pp, 1);
1038
1039         if (newstate == PATH_WILD || newstate == PATH_UNCHECKED) {
1040                 condlog(2, "%s: unusable path", pp->dev);
1041                 pathinfo(pp, conf->hwtable, 0);
1042                 return;
1043         }
1044         /*
1045          * Async IO in flight. Keep the previous path state
1046          * and reschedule as soon as possible
1047          */
1048         if (newstate == PATH_PENDING) {
1049                 pp->tick = 1;
1050                 return;
1051         }
1052         /*
1053          * Synchronize with kernel state
1054          */
1055         if (update_multipath_strings(pp->mpp, vecs->pathvec)) {
1056                 condlog(1, "%s: Could not synchronize with kernel state\n",
1057                         pp->dev);
1058                 pp->dmstate = PSTATE_UNDEF;
1059         }
1060         if (newstate != pp->state) {
1061                 int oldstate = pp->state;
1062                 pp->state = newstate;
1063                 LOG_MSG(1, checker_message(&pp->checker));
1064
1065                 /*
1066                  * upon state change, reset the checkint
1067                  * to the shortest delay
1068                  */
1069                 pp->checkint = conf->checkint;
1070
1071                 if (newstate == PATH_DOWN || newstate == PATH_SHAKY) {
1072                         /*
1073                          * proactively fail path in the DM
1074                          */
1075                         if (oldstate == PATH_UP ||
1076                             oldstate == PATH_GHOST)
1077                                 fail_path(pp, 1);
1078                         else
1079                                 fail_path(pp, 0);
1080
1081                         /*
1082                          * cancel scheduled failback
1083                          */
1084                         pp->mpp->failback_tick = 0;
1085
1086                         pp->mpp->stat_path_failures++;
1087                         return;
1088                 }
1089
1090                 /*
1091                  * reinstate this path
1092                  */
1093                 if (oldstate != PATH_UP &&
1094                     oldstate != PATH_GHOST)
1095                         reinstate_path(pp, 1);
1096                 else
1097                         reinstate_path(pp, 0);
1098
1099                 new_path_up = 1;
1100
1101                 /*
1102                  * if at least one path is up in a group, and
1103                  * the group is disabled, re-enable it
1104                  */
1105                 if (newstate == PATH_UP)
1106                         enable_group(pp);
1107         }
1108         else if (newstate == PATH_UP || newstate == PATH_GHOST) {
1109                 if (pp->dmstate == PSTATE_FAILED ||
1110                     pp->dmstate == PSTATE_UNDEF) {
1111                         /* Clear IO errors */
1112                         reinstate_path(pp, 0);
1113                 } else {
1114                         LOG_MSG(4, checker_message(&pp->checker));
1115                         /*
1116                          * double the next check delay.
1117                          * max at conf->max_checkint
1118                          */
1119                         if (pp->checkint < (conf->max_checkint / 2))
1120                             pp->checkint = 2 * pp->checkint;
1121                         else
1122                             pp->checkint = conf->max_checkint;
1123
1124                         pp->tick = pp->checkint;
1125                         condlog(4, "%s: delay next check %is",
1126                                 pp->dev_t, pp->tick);
1127                 }
1128         }
1129         else if (newstate == PATH_DOWN)
1130                 LOG_MSG(2, checker_message(&pp->checker));
1131
1132         pp->state = newstate;
1133
1134         /*
1135          * path prio refreshing
1136          */
1137         condlog(4, "path prio refresh");
1138
1139         if (update_prio(pp, new_path_up) &&
1140             pp->mpp->pgpolicyfn == (pgpolicyfn *)group_by_prio)
1141                 update_path_groups(pp->mpp, vecs, !new_path_up);
1142         else if (need_switch_pathgroup(pp->mpp, 0)) {
1143                 if (pp->mpp->pgfailback > 0 &&
1144                     (new_path_up || pp->mpp->failback_tick <= 0))
1145                         pp->mpp->failback_tick =
1146                                 pp->mpp->pgfailback + 1;
1147                 else if (pp->mpp->pgfailback == -FAILBACK_IMMEDIATE)
1148                         switch_pathgroup(pp->mpp);
1149         }
1150 }
1151
1152 static void *
1153 checkerloop (void *ap)
1154 {
1155         struct vectors *vecs;
1156         struct path *pp;
1157         int count = 0;
1158         unsigned int i;
1159         sigset_t old;
1160
1161         mlockall(MCL_CURRENT | MCL_FUTURE);
1162         vecs = (struct vectors *)ap;
1163         condlog(2, "path checkers start up");
1164
1165         /*
1166          * init the path check interval
1167          */
1168         vector_foreach_slot (vecs->pathvec, pp, i) {
1169                 pp->checkint = conf->checkint;
1170         }
1171
1172         while (1) {
1173                 block_signal(SIGHUP, &old);
1174                 pthread_cleanup_push(cleanup_lock, &vecs->lock);
1175                 lock(vecs->lock);
1176                 condlog(4, "tick");
1177
1178                 if (vecs->pathvec) {
1179                         vector_foreach_slot (vecs->pathvec, pp, i) {
1180                                 check_path(vecs, pp);
1181                         }
1182                 }
1183                 if (vecs->mpvec) {
1184                         defered_failback_tick(vecs->mpvec);
1185                         retry_count_tick(vecs->mpvec);
1186                 }
1187                 if (count)
1188                         count--;
1189                 else {
1190                         condlog(4, "map garbage collection");
1191                         mpvec_garbage_collector(vecs);
1192                         count = MAPGCINT;
1193                 }
1194
1195                 lock_cleanup_pop(vecs->lock);
1196                 pthread_sigmask(SIG_SETMASK, &old, NULL);
1197                 sleep(1);
1198         }
1199         return NULL;
1200 }
1201
1202 int
1203 configure (struct vectors * vecs, int start_waiters)
1204 {
1205         struct multipath * mpp;
1206         struct path * pp;
1207         vector mpvec;
1208         int i;
1209
1210         if (!vecs->pathvec && !(vecs->pathvec = vector_alloc()))
1211                 return 1;
1212
1213         if (!vecs->mpvec && !(vecs->mpvec = vector_alloc()))
1214                 return 1;
1215
1216         if (!(mpvec = vector_alloc()))
1217                 return 1;
1218
1219         /*
1220          * probe for current path (from sysfs) and map (from dm) sets
1221          */
1222         path_discovery(vecs->pathvec, conf, DI_ALL);
1223
1224         vector_foreach_slot (vecs->pathvec, pp, i){
1225                 if (filter_path(conf, pp) > 0){
1226                         vector_del_slot(vecs->pathvec, i);
1227                         free_path(pp);
1228                         i--;
1229                 }
1230                 else
1231                         pp->checkint = conf->checkint;
1232         }
1233         if (map_discovery(vecs))
1234                 return 1;
1235
1236         /*
1237          * create new set of maps & push changed ones into dm
1238          */
1239         if (coalesce_paths(vecs, mpvec, NULL, 1))
1240                 return 1;
1241
1242         /*
1243          * may need to remove some maps which are no longer relevant
1244          * e.g., due to blacklist changes in conf file
1245          */
1246         if (coalesce_maps(vecs, mpvec))
1247                 return 1;
1248
1249         dm_lib_release();
1250
1251         sync_maps_state(mpvec);
1252
1253         /*
1254          * purge dm of old maps
1255          */
1256         remove_maps(vecs);
1257
1258         /*
1259          * save new set of maps formed by considering current path state
1260          */
1261         vector_free(vecs->mpvec);
1262         vecs->mpvec = mpvec;
1263
1264         /*
1265          * start dm event waiter threads for these new maps
1266          */
1267         vector_foreach_slot(vecs->mpvec, mpp, i) {
1268                 if (setup_multipath(vecs, mpp))
1269                         return 1;
1270                 if (start_waiters)
1271                         if (start_waiter_thread(mpp, vecs))
1272                                 return 1;
1273         }
1274         return 0;
1275 }
1276
1277 int
1278 reconfigure (struct vectors * vecs)
1279 {
1280         struct config * old = conf;
1281         int retval = 1;
1282
1283         lock(vecs->lock);
1284         /*
1285          * free old map and path vectors ... they use old conf state
1286          */
1287         if (VECTOR_SIZE(vecs->mpvec))
1288                 remove_maps_and_stop_waiters(vecs);
1289
1290         if (VECTOR_SIZE(vecs->pathvec))
1291                 free_pathvec(vecs->pathvec, FREE_PATHS);
1292
1293         vecs->pathvec = NULL;
1294         conf = NULL;
1295
1296         if (!load_config(DEFAULT_CONFIGFILE)) {
1297                 conf->verbosity = old->verbosity;
1298                 conf->daemon = 1;
1299                 configure(vecs, 1);
1300                 free_config(old);
1301                 retval = 0;
1302         }
1303
1304         unlock(vecs->lock);
1305         return retval;
1306 }
1307
1308 static struct vectors *
1309 init_vecs (void)
1310 {
1311         struct vectors * vecs;
1312
1313         vecs = (struct vectors *)MALLOC(sizeof(struct vectors));
1314
1315         if (!vecs)
1316                 return NULL;
1317
1318         vecs->lock.mutex =
1319                 (pthread_mutex_t *)MALLOC(sizeof(pthread_mutex_t));
1320
1321         if (!vecs->lock.mutex)
1322                 goto out;
1323
1324         pthread_mutex_init(vecs->lock.mutex, NULL);
1325         vecs->lock.depth = 0;
1326
1327         return vecs;
1328
1329 out:
1330         FREE(vecs);
1331         condlog(0, "failed to init paths");
1332         return NULL;
1333 }
1334
1335 static void *
1336 signal_set(int signo, void (*func) (int))
1337 {
1338         int r;
1339         struct sigaction sig;
1340         struct sigaction osig;
1341
1342         sig.sa_handler = func;
1343         sigemptyset(&sig.sa_mask);
1344         sig.sa_flags = 0;
1345
1346         r = sigaction(signo, &sig, &osig);
1347
1348         if (r < 0)
1349                 return (SIG_ERR);
1350         else
1351                 return (osig.sa_handler);
1352 }
1353
1354 static void
1355 sighup (int sig)
1356 {
1357         condlog(2, "reconfigure (SIGHUP)");
1358
1359         if (running_state != DAEMON_RUNNING)
1360                 return;
1361
1362         reconfigure(gvecs);
1363
1364 #ifdef _DEBUG_
1365         dbg_free_final(NULL);
1366 #endif
1367 }
1368
1369 static void
1370 sigend (int sig)
1371 {
1372         exit_daemon(0);
1373 }
1374
1375 static void
1376 sigusr1 (int sig)
1377 {
1378         condlog(3, "SIGUSR1 received");
1379 }
1380
1381 static void
1382 signal_init(void)
1383 {
1384         signal_set(SIGHUP, sighup);
1385         signal_set(SIGUSR1, sigusr1);
1386         signal_set(SIGINT, sigend);
1387         signal_set(SIGTERM, sigend);
1388         signal(SIGPIPE, SIG_IGN);
1389 }
1390
1391 static void
1392 setscheduler (void)
1393 {
1394         int res;
1395         static struct sched_param sched_param = {
1396                 .sched_priority = 99
1397         };
1398
1399         res = sched_setscheduler (0, SCHED_RR, &sched_param);
1400
1401         if (res == -1)
1402                 condlog(LOG_WARNING, "Could not set SCHED_RR at priority 99");
1403         return;
1404 }
1405
1406 static void
1407 set_oom_adj (int val)
1408 {
1409         FILE *fp;
1410
1411         fp = fopen("/proc/self/oom_adj", "w");
1412
1413         if (!fp)
1414                 return;
1415
1416         fprintf(fp, "%i", val);
1417         fclose(fp);
1418 }
1419
1420 static int
1421 child (void * param)
1422 {
1423         pthread_t check_thr, uevent_thr, uxlsnr_thr, uevq_thr;
1424         pthread_attr_t log_attr, misc_attr;
1425         struct vectors * vecs;
1426         struct multipath * mpp;
1427         int i;
1428         int rc;
1429
1430         mlockall(MCL_CURRENT | MCL_FUTURE);
1431
1432         setup_thread_attr(&misc_attr, 64 * 1024, 1);
1433         setup_thread_attr(&waiter_attr, 32 * 1024, 1);
1434
1435         if (logsink) {
1436                 setup_thread_attr(&log_attr, 64 * 1024, 0);
1437                 log_thread_start(&log_attr);
1438                 pthread_attr_destroy(&log_attr);
1439         }
1440
1441         running_state = DAEMON_START;
1442
1443         condlog(2, "--------start up--------");
1444         condlog(2, "read " DEFAULT_CONFIGFILE);
1445
1446         if (load_config(DEFAULT_CONFIGFILE))
1447                 exit(1);
1448
1449         if (init_checkers()) {
1450                 condlog(0, "failed to initialize checkers");
1451                 exit(1);
1452         }
1453         if (init_prio()) {
1454                 condlog(0, "failed to initialize prioritizers");
1455                 exit(1);
1456         }
1457
1458         setlogmask(LOG_UPTO(conf->verbosity + 3));
1459
1460         /*
1461          * fill the voids left in the config file
1462          */
1463         if (!conf->checkint) {
1464                 conf->checkint = DEFAULT_CHECKINT;
1465                 conf->max_checkint = MAX_CHECKINT(conf->checkint);
1466         }
1467
1468         if (conf->max_fds) {
1469                 struct rlimit fd_limit;
1470
1471                 if (getrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1472                         condlog(0, "can't get open fds limit: %s\n",
1473                                 strerror(errno));
1474                         fd_limit.rlim_cur = 0;
1475                         fd_limit.rlim_max = 0;
1476                 }
1477                 if (fd_limit.rlim_cur < conf->max_fds) {
1478                         fd_limit.rlim_cur = conf->max_fds;
1479                         if (fd_limit.rlim_max < conf->max_fds)
1480                                 fd_limit.rlim_max = conf->max_fds;
1481                         if (setrlimit(RLIMIT_NOFILE, &fd_limit) < 0) {
1482                                 condlog(0, "can't set open fds limit to "
1483                                         "%lu/%lu : %s\n",
1484                                         fd_limit.rlim_cur, fd_limit.rlim_max,
1485                                         strerror(errno));
1486                         } else {
1487                                 condlog(3, "set open fds limit to %lu/%lu\n",
1488                                         fd_limit.rlim_cur, fd_limit.rlim_max);
1489                         }
1490                 }
1491
1492         }
1493
1494         signal_init();
1495         setscheduler();
1496         set_oom_adj(-16);
1497         vecs = gvecs = init_vecs();
1498
1499         if (!vecs)
1500                 exit(1);
1501
1502         if (sysfs_init(conf->sysfs_dir, FILE_NAME_SIZE)) {
1503                 condlog(0, "can not find sysfs mount point");
1504                 exit(1);
1505         }
1506         conf->daemon = 1;
1507         udev_set_sync_support(0);
1508         /*
1509          * Start uevent listener early to catch events
1510          */
1511         if ((rc = pthread_create(&uevent_thr, &misc_attr, ueventloop, vecs))) {
1512                 condlog(0, "failed to create uevent thread: %d", rc);
1513                 exit(1);
1514         }
1515         if ((rc = pthread_create(&uxlsnr_thr, &misc_attr, uxlsnrloop, vecs))) {
1516                 condlog(0, "failed to create cli listener: %d", rc);
1517                 exit(1);
1518         }
1519         /*
1520          * fetch and configure both paths and multipaths
1521          */
1522         lock(vecs->lock);
1523         running_state = DAEMON_CONFIGURE;
1524
1525         if (configure(vecs, 1)) {
1526                 unlock(vecs->lock);
1527                 condlog(0, "failure during configuration");
1528                 exit(1);
1529         }
1530         unlock(vecs->lock);
1531
1532         /*
1533          * start threads
1534          */
1535         if ((rc = pthread_create(&check_thr, &misc_attr, checkerloop, vecs))) {
1536                 condlog(0,"failed to create checker loop thread: %d", rc);
1537                 exit(1);
1538         }
1539         if ((rc = pthread_create(&uevq_thr, &misc_attr, uevqloop, vecs))) {
1540                 condlog(0, "failed to create uevent dispatcher: %d", rc);
1541                 exit(1);
1542         }
1543         pthread_attr_destroy(&misc_attr);
1544
1545         pthread_mutex_lock(&exit_mutex);
1546         /* Startup complete, create logfile */
1547         if (pidfile_create(DEFAULT_PIDFILE, getpid()))
1548                 /* Ignore errors, we can live without */
1549                 condlog(1, "failed to create pidfile");
1550
1551         running_state = DAEMON_RUNNING;
1552         pthread_cond_wait(&exit_cond, &exit_mutex);
1553
1554         /*
1555          * exit path
1556          */
1557         running_state = DAEMON_SHUTDOWN;
1558         block_signal(SIGHUP, NULL);
1559         lock(vecs->lock);
1560         if (conf->queue_without_daemon == QUE_NO_DAEMON_OFF)
1561                 vector_foreach_slot(vecs->mpvec, mpp, i)
1562                         dm_queue_if_no_path(mpp->alias, 0);
1563         remove_maps_and_stop_waiters(vecs);
1564         unlock(vecs->lock);
1565
1566         pthread_cancel(check_thr);
1567         pthread_cancel(uevent_thr);
1568         pthread_cancel(uxlsnr_thr);
1569         pthread_cancel(uevq_thr);
1570
1571         sysfs_cleanup();
1572
1573         lock(vecs->lock);
1574         free_pathvec(vecs->pathvec, FREE_PATHS);
1575         vecs->pathvec = NULL;
1576         unlock(vecs->lock);
1577         /* Now all the waitevent threads will start rushing in. */
1578         while (vecs->lock.depth > 0) {
1579                 sleep (1); /* This is weak. */
1580                 condlog(3,"Have %d wait event checkers threads to de-alloc, waiting..\n", vecs->lock.depth);
1581         }
1582         pthread_mutex_destroy(vecs->lock.mutex);
1583         FREE(vecs->lock.mutex);
1584         vecs->lock.depth = 0;
1585         vecs->lock.mutex = NULL;
1586         FREE(vecs);
1587         vecs = NULL;
1588
1589         cleanup_checkers();
1590         cleanup_prio();
1591
1592         dm_lib_release();
1593         dm_lib_exit();
1594
1595         /* We're done here */
1596         condlog(3, "unlink pidfile");
1597         unlink(DEFAULT_PIDFILE);
1598
1599         condlog(2, "--------shut down-------");
1600
1601         if (logsink)
1602                 log_thread_stop();
1603
1604         /*
1605          * Freeing config must be done after condlog() and dm_lib_exit(),
1606          * because logging functions like dlog() and dm_write_log()
1607          * reference the config.
1608          */
1609         free_config(conf);
1610         conf = NULL;
1611
1612 #ifdef _DEBUG_
1613         dbg_free_final(NULL);
1614 #endif
1615
1616         exit(0);
1617 }
1618
1619 static int
1620 daemonize(void)
1621 {
1622         int pid;
1623         int in_fd, out_fd;
1624
1625         if( (pid = fork()) < 0){
1626                 fprintf(stderr, "Failed first fork : %s\n", strerror(errno));
1627                 return -1;
1628         }
1629         else if (pid != 0)
1630                 return pid;
1631
1632         setsid();
1633
1634         if ( (pid = fork()) < 0)
1635                 fprintf(stderr, "Failed second fork : %s\n", strerror(errno));
1636         else if (pid != 0)
1637                 _exit(0);
1638
1639         in_fd = open("/dev/null", O_RDONLY);
1640         if (in_fd < 0){
1641                 fprintf(stderr, "cannot open /dev/null for input : %s\n",
1642                         strerror(errno));
1643                 _exit(0);
1644         }
1645         out_fd = open("/dev/console", O_WRONLY);
1646         if (out_fd < 0){
1647                 fprintf(stderr, "cannot open /dev/console for output : %s\n",
1648                         strerror(errno));
1649                 _exit(0);
1650         }
1651
1652         close(STDIN_FILENO);
1653         dup(in_fd);
1654         close(STDOUT_FILENO);
1655         dup(out_fd);
1656         close(STDERR_FILENO);
1657         dup(out_fd);
1658
1659         close(in_fd);
1660         close(out_fd);
1661         if (chdir("/") < 0)
1662                 fprintf(stderr, "cannot chdir to '/', continuing\n");
1663
1664         return 0;
1665 }
1666
1667 int
1668 main (int argc, char *argv[])
1669 {
1670         extern char *optarg;
1671         extern int optind;
1672         int arg;
1673         int err;
1674
1675         logsink = 1;
1676         running_state = DAEMON_INIT;
1677         dm_init();
1678
1679         if (getuid() != 0) {
1680                 fprintf(stderr, "need to be root\n");
1681                 exit(1);
1682         }
1683
1684         /* make sure we don't lock any path */
1685         chdir("/");
1686         umask(umask(077) | 022);
1687
1688         conf = alloc_config();
1689
1690         if (!conf)
1691                 exit(1);
1692
1693         while ((arg = getopt(argc, argv, ":dv:k::")) != EOF ) {
1694         switch(arg) {
1695                 case 'd':
1696                         logsink = 0;
1697                         //debug=1; /* ### comment me out ### */
1698                         break;
1699                 case 'v':
1700                         if (sizeof(optarg) > sizeof(char *) ||
1701                             !isdigit(optarg[0]))
1702                                 exit(1);
1703
1704                         conf->verbosity = atoi(optarg);
1705                         break;
1706                 case 'k':
1707                         uxclnt(optarg);
1708                         exit(0);
1709                 default:
1710                         ;
1711                 }
1712         }
1713         if (optind < argc) {
1714                 char cmd[CMDSIZE];
1715                 char * s = cmd;
1716                 char * c = s;
1717
1718                 while (optind < argc) {
1719                         if (strchr(argv[optind], ' '))
1720                                 c += snprintf(c, s + CMDSIZE - c, "\"%s\" ", argv[optind]);
1721                         else
1722                                 c += snprintf(c, s + CMDSIZE - c, "%s ", argv[optind]);
1723                         optind++;
1724                 }
1725                 c += snprintf(c, s + CMDSIZE - c, "\n");
1726                 uxclnt(s);
1727                 exit(0);
1728         }
1729
1730         if (!logsink)
1731                 err = 0;
1732         else
1733                 err = daemonize();
1734
1735         if (err < 0)
1736                 /* error */
1737                 exit(1);
1738         else if (err > 0)
1739                 /* parent dies */
1740                 exit(0);
1741         else
1742                 /* child lives */
1743                 return (child(NULL));
1744 }
1745