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