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